| File: | arch/amd64/amd64/vmm.c |
| Warning: | line 4465, column 2 Value stored to 'level' is never read |
Press '?' to see keyboard shortcuts
Keyboard shortcuts:
| 1 | /* $OpenBSD: vmm.c,v 1.301 2022/01/11 20:34:22 tobhe Exp $ */ |
| 2 | /* |
| 3 | * Copyright (c) 2014 Mike Larkin <mlarkin@openbsd.org> |
| 4 | * |
| 5 | * Permission to use, copy, modify, and distribute this software for any |
| 6 | * purpose with or without fee is hereby granted, provided that the above |
| 7 | * copyright notice and this permission notice appear in all copies. |
| 8 | * |
| 9 | * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES |
| 10 | * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF |
| 11 | * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR |
| 12 | * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES |
| 13 | * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN |
| 14 | * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF |
| 15 | * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. |
| 16 | */ |
| 17 | |
| 18 | #include <sys/param.h> |
| 19 | #include <sys/systm.h> |
| 20 | #include <sys/signalvar.h> |
| 21 | #include <sys/malloc.h> |
| 22 | #include <sys/device.h> |
| 23 | #include <sys/pool.h> |
| 24 | #include <sys/proc.h> |
| 25 | #include <sys/user.h> |
| 26 | #include <sys/ioctl.h> |
| 27 | #include <sys/queue.h> |
| 28 | #include <sys/rwlock.h> |
| 29 | #include <sys/pledge.h> |
| 30 | #include <sys/memrange.h> |
| 31 | #include <sys/tracepoint.h> |
| 32 | |
| 33 | #include <uvm/uvm_extern.h> |
| 34 | |
| 35 | #include <machine/fpu.h> |
| 36 | #include <machine/pmap.h> |
| 37 | #include <machine/biosvar.h> |
| 38 | #include <machine/segments.h> |
| 39 | #include <machine/cpufunc.h> |
| 40 | #include <machine/vmmvar.h> |
| 41 | |
| 42 | #include <dev/isa/isareg.h> |
| 43 | #include <dev/pv/pvreg.h> |
| 44 | |
| 45 | /* #define VMM_DEBUG */ |
| 46 | |
| 47 | void *l1tf_flush_region; |
| 48 | |
| 49 | #ifdef VMM_DEBUG |
| 50 | #define DPRINTF(x...) do { printf(x); } while(0) |
| 51 | #else |
| 52 | #define DPRINTF(x...) |
| 53 | #endif /* VMM_DEBUG */ |
| 54 | |
| 55 | #define DEVNAME(s)((s)->sc_dev.dv_xname) ((s)->sc_dev.dv_xname) |
| 56 | |
| 57 | #define CTRL_DUMP(x,y,z)printf(" %s: Can set:%s Can clear:%s\n", "z" , vcpu_vmx_check_cap (x, IA32_VMX_y_CTLS, IA32_VMX_z, 1) ? "Yes" : "No", vcpu_vmx_check_cap (x, IA32_VMX_y_CTLS, IA32_VMX_z, 0) ? "Yes" : "No"); printf(" %s: Can set:%s Can clear:%s\n", #z , \ |
| 58 | vcpu_vmx_check_cap(x, IA32_VMX_##y ##_CTLS, \ |
| 59 | IA32_VMX_##z, 1) ? "Yes" : "No", \ |
| 60 | vcpu_vmx_check_cap(x, IA32_VMX_##y ##_CTLS, \ |
| 61 | IA32_VMX_##z, 0) ? "Yes" : "No"); |
| 62 | |
| 63 | #define VMX_EXIT_INFO_HAVE_RIP0x1 0x1 |
| 64 | #define VMX_EXIT_INFO_HAVE_REASON0x2 0x2 |
| 65 | #define VMX_EXIT_INFO_COMPLETE(0x1 | 0x2) \ |
| 66 | (VMX_EXIT_INFO_HAVE_RIP0x1 | VMX_EXIT_INFO_HAVE_REASON0x2) |
| 67 | |
| 68 | struct vm { |
| 69 | struct vmspace *vm_vmspace; |
| 70 | vm_map_t vm_map; |
| 71 | uint32_t vm_id; |
| 72 | pid_t vm_creator_pid; |
| 73 | size_t vm_nmemranges; |
| 74 | size_t vm_memory_size; |
| 75 | char vm_name[VMM_MAX_NAME_LEN64]; |
| 76 | struct vm_mem_range vm_memranges[VMM_MAX_MEM_RANGES16]; |
| 77 | |
| 78 | struct vcpu_head vm_vcpu_list; |
| 79 | uint32_t vm_vcpu_ct; |
| 80 | u_int vm_vcpus_running; |
| 81 | struct rwlock vm_vcpu_lock; |
| 82 | |
| 83 | SLIST_ENTRY(vm)struct { struct vm *sle_next; } vm_link; |
| 84 | }; |
| 85 | |
| 86 | SLIST_HEAD(vmlist_head, vm)struct vmlist_head { struct vm *slh_first; }; |
| 87 | |
| 88 | struct vmm_softc { |
| 89 | struct device sc_dev; |
| 90 | |
| 91 | /* Capabilities */ |
| 92 | uint32_t nr_vmx_cpus; |
| 93 | uint32_t nr_svm_cpus; |
| 94 | uint32_t nr_rvi_cpus; |
| 95 | uint32_t nr_ept_cpus; |
| 96 | |
| 97 | /* Managed VMs */ |
| 98 | struct vmlist_head vm_list; |
| 99 | |
| 100 | int mode; |
| 101 | |
| 102 | size_t vcpu_ct; |
| 103 | size_t vcpu_max; |
| 104 | |
| 105 | struct rwlock vm_lock; |
| 106 | size_t vm_ct; /* number of in-memory VMs */ |
| 107 | size_t vm_idx; /* next unique VM index */ |
| 108 | |
| 109 | struct rwlock vpid_lock; |
| 110 | uint16_t max_vpid; |
| 111 | uint8_t vpids[512]; /* bitmap of used VPID/ASIDs */ |
| 112 | }; |
| 113 | |
| 114 | void vmx_dump_vmcs_field(uint16_t, const char *); |
| 115 | int vmm_enabled(void); |
| 116 | int vmm_probe(struct device *, void *, void *); |
| 117 | void vmm_attach(struct device *, struct device *, void *); |
| 118 | int vmmopen(dev_t, int, int, struct proc *); |
| 119 | int vmmioctl(dev_t, u_long, caddr_t, int, struct proc *); |
| 120 | int vmmclose(dev_t, int, int, struct proc *); |
| 121 | int vmm_start(void); |
| 122 | int vmm_stop(void); |
| 123 | size_t vm_create_check_mem_ranges(struct vm_create_params *); |
| 124 | int vm_create(struct vm_create_params *, struct proc *); |
| 125 | int vm_run(struct vm_run_params *); |
| 126 | int vm_terminate(struct vm_terminate_params *); |
| 127 | int vm_get_info(struct vm_info_params *); |
| 128 | int vm_resetcpu(struct vm_resetcpu_params *); |
| 129 | int vm_intr_pending(struct vm_intr_params *); |
| 130 | int vm_rwregs(struct vm_rwregs_params *, int); |
| 131 | int vm_mprotect_ept(struct vm_mprotect_ept_params *); |
| 132 | int vm_rwvmparams(struct vm_rwvmparams_params *, int); |
| 133 | int vm_find(uint32_t, struct vm **); |
| 134 | int vcpu_readregs_vmx(struct vcpu *, uint64_t, struct vcpu_reg_state *); |
| 135 | int vcpu_readregs_svm(struct vcpu *, uint64_t, struct vcpu_reg_state *); |
| 136 | int vcpu_writeregs_vmx(struct vcpu *, uint64_t, int, struct vcpu_reg_state *); |
| 137 | int vcpu_writeregs_svm(struct vcpu *, uint64_t, struct vcpu_reg_state *); |
| 138 | int vcpu_reset_regs(struct vcpu *, struct vcpu_reg_state *); |
| 139 | int vcpu_reset_regs_vmx(struct vcpu *, struct vcpu_reg_state *); |
| 140 | int vcpu_reset_regs_svm(struct vcpu *, struct vcpu_reg_state *); |
| 141 | int vcpu_reload_vmcs_vmx(struct vcpu *); |
| 142 | int vcpu_init(struct vcpu *); |
| 143 | int vcpu_init_vmx(struct vcpu *); |
| 144 | int vcpu_init_svm(struct vcpu *); |
| 145 | int vcpu_must_stop(struct vcpu *); |
| 146 | int vcpu_run_vmx(struct vcpu *, struct vm_run_params *); |
| 147 | int vcpu_run_svm(struct vcpu *, struct vm_run_params *); |
| 148 | void vcpu_deinit(struct vcpu *); |
| 149 | void vcpu_deinit_vmx(struct vcpu *); |
| 150 | void vcpu_deinit_svm(struct vcpu *); |
| 151 | int vm_impl_init(struct vm *, struct proc *); |
| 152 | int vm_impl_init_vmx(struct vm *, struct proc *); |
| 153 | int vm_impl_init_svm(struct vm *, struct proc *); |
| 154 | void vm_impl_deinit(struct vm *); |
| 155 | void vm_impl_deinit_vmx(struct vm *); |
| 156 | void vm_impl_deinit_svm(struct vm *); |
| 157 | void vm_teardown(struct vm *); |
| 158 | int vcpu_vmx_check_cap(struct vcpu *, uint32_t, uint32_t, int); |
| 159 | int vcpu_vmx_compute_ctrl(uint64_t, uint16_t, uint32_t, uint32_t, uint32_t *); |
| 160 | int vmx_get_exit_info(uint64_t *, uint64_t *); |
| 161 | int vmx_load_pdptes(struct vcpu *); |
| 162 | int vmx_handle_exit(struct vcpu *); |
| 163 | int svm_handle_exit(struct vcpu *); |
| 164 | int svm_handle_msr(struct vcpu *); |
| 165 | int vmm_handle_xsetbv(struct vcpu *, uint64_t *); |
| 166 | int vmx_handle_xsetbv(struct vcpu *); |
| 167 | int svm_handle_xsetbv(struct vcpu *); |
| 168 | int vmm_handle_cpuid(struct vcpu *); |
| 169 | int vmx_handle_rdmsr(struct vcpu *); |
| 170 | int vmx_handle_wrmsr(struct vcpu *); |
| 171 | int vmx_handle_cr0_write(struct vcpu *, uint64_t); |
| 172 | int vmx_handle_cr4_write(struct vcpu *, uint64_t); |
| 173 | int vmx_handle_cr(struct vcpu *); |
| 174 | int svm_handle_inout(struct vcpu *); |
| 175 | int vmx_handle_inout(struct vcpu *); |
| 176 | int svm_handle_hlt(struct vcpu *); |
| 177 | int vmx_handle_hlt(struct vcpu *); |
| 178 | int vmm_inject_ud(struct vcpu *); |
| 179 | int vmm_inject_gp(struct vcpu *); |
| 180 | int vmm_inject_db(struct vcpu *); |
| 181 | void vmx_handle_intr(struct vcpu *); |
| 182 | void vmx_handle_intwin(struct vcpu *); |
| 183 | void vmx_handle_misc_enable_msr(struct vcpu *); |
| 184 | int vmm_get_guest_memtype(struct vm *, paddr_t); |
| 185 | int vmx_get_guest_faulttype(void); |
| 186 | int svm_get_guest_faulttype(struct vmcb *); |
| 187 | int vmx_get_exit_qualification(uint64_t *); |
| 188 | int vmm_get_guest_cpu_cpl(struct vcpu *); |
| 189 | int vmm_get_guest_cpu_mode(struct vcpu *); |
| 190 | int svm_fault_page(struct vcpu *, paddr_t); |
| 191 | int vmx_fault_page(struct vcpu *, paddr_t); |
| 192 | int vmx_handle_np_fault(struct vcpu *); |
| 193 | int svm_handle_np_fault(struct vcpu *); |
| 194 | int vmx_mprotect_ept(vm_map_t, paddr_t, paddr_t, int); |
| 195 | pt_entry_t *vmx_pmap_find_pte_ept(pmap_t, paddr_t); |
| 196 | int vmm_alloc_vpid(uint16_t *); |
| 197 | void vmm_free_vpid(uint16_t); |
| 198 | const char *vcpu_state_decode(u_int); |
| 199 | const char *vmx_exit_reason_decode(uint32_t); |
| 200 | const char *svm_exit_reason_decode(uint32_t); |
| 201 | const char *vmx_instruction_error_decode(uint32_t); |
| 202 | void svm_setmsrbr(struct vcpu *, uint32_t); |
| 203 | void svm_setmsrbw(struct vcpu *, uint32_t); |
| 204 | void svm_setmsrbrw(struct vcpu *, uint32_t); |
| 205 | void vmx_setmsrbr(struct vcpu *, uint32_t); |
| 206 | void vmx_setmsrbw(struct vcpu *, uint32_t); |
| 207 | void vmx_setmsrbrw(struct vcpu *, uint32_t); |
| 208 | void svm_set_clean(struct vcpu *, uint32_t); |
| 209 | void svm_set_dirty(struct vcpu *, uint32_t); |
| 210 | |
| 211 | int vmm_gpa_is_valid(struct vcpu *vcpu, paddr_t gpa, size_t obj_size); |
| 212 | void vmm_init_pvclock(struct vcpu *, paddr_t); |
| 213 | int vmm_update_pvclock(struct vcpu *); |
| 214 | int vmm_pat_is_valid(uint64_t); |
| 215 | |
| 216 | #ifdef MULTIPROCESSOR1 |
| 217 | static int vmx_remote_vmclear(struct cpu_info*, struct vcpu *); |
| 218 | #endif |
| 219 | |
| 220 | #ifdef VMM_DEBUG |
| 221 | void dump_vcpu(struct vcpu *); |
| 222 | void vmx_vcpu_dump_regs(struct vcpu *); |
| 223 | void vmx_dump_vmcs(struct vcpu *); |
| 224 | const char *msr_name_decode(uint32_t); |
| 225 | void vmm_segment_desc_decode(uint64_t); |
| 226 | void vmm_decode_cr0(uint64_t); |
| 227 | void vmm_decode_cr3(uint64_t); |
| 228 | void vmm_decode_cr4(uint64_t); |
| 229 | void vmm_decode_msr_value(uint64_t, uint64_t); |
| 230 | void vmm_decode_apicbase_msr_value(uint64_t); |
| 231 | void vmm_decode_ia32_fc_value(uint64_t); |
| 232 | void vmm_decode_mtrrcap_value(uint64_t); |
| 233 | void vmm_decode_perf_status_value(uint64_t); |
| 234 | void vmm_decode_perf_ctl_value(uint64_t); |
| 235 | void vmm_decode_mtrrdeftype_value(uint64_t); |
| 236 | void vmm_decode_efer_value(uint64_t); |
| 237 | void vmm_decode_rflags(uint64_t); |
| 238 | void vmm_decode_misc_enable_value(uint64_t); |
| 239 | const char *vmm_decode_cpu_mode(struct vcpu *); |
| 240 | |
| 241 | extern int mtrr2mrt(int); |
| 242 | |
| 243 | struct vmm_reg_debug_info { |
| 244 | uint64_t vrdi_bit; |
| 245 | const char *vrdi_present; |
| 246 | const char *vrdi_absent; |
| 247 | }; |
| 248 | #endif /* VMM_DEBUG */ |
| 249 | |
| 250 | extern uint64_t tsc_frequency; |
| 251 | extern int tsc_is_invariant; |
| 252 | |
| 253 | const char *vmm_hv_signature = VMM_HV_SIGNATURE"OpenBSDVMM58"; |
| 254 | |
| 255 | const struct kmem_pa_mode vmm_kp_contig = { |
| 256 | .kp_constraint = &no_constraint, |
| 257 | .kp_maxseg = 1, |
| 258 | .kp_align = 4096, |
| 259 | .kp_zero = 1, |
| 260 | }; |
| 261 | |
| 262 | struct cfdriver vmm_cd = { |
| 263 | NULL((void *)0), "vmm", DV_DULL, CD_SKIPHIBERNATE2 |
| 264 | }; |
| 265 | |
| 266 | const struct cfattach vmm_ca = { |
| 267 | sizeof(struct vmm_softc), vmm_probe, vmm_attach, NULL((void *)0), NULL((void *)0) |
| 268 | }; |
| 269 | |
| 270 | /* |
| 271 | * Helper struct to easily get the VMCS field IDs needed in vmread/vmwrite |
| 272 | * to access the individual fields of the guest segment registers. This |
| 273 | * struct is indexed by VCPU_REGS_* id. |
| 274 | */ |
| 275 | const struct { |
| 276 | uint64_t selid; |
| 277 | uint64_t limitid; |
| 278 | uint64_t arid; |
| 279 | uint64_t baseid; |
| 280 | } vmm_vmx_sreg_vmcs_fields[] = { |
| 281 | { VMCS_GUEST_IA32_CS_SEL0x0802, VMCS_GUEST_IA32_CS_LIMIT0x4802, |
| 282 | VMCS_GUEST_IA32_CS_AR0x4816, VMCS_GUEST_IA32_CS_BASE0x6808 }, |
| 283 | { VMCS_GUEST_IA32_DS_SEL0x0806, VMCS_GUEST_IA32_DS_LIMIT0x4806, |
| 284 | VMCS_GUEST_IA32_DS_AR0x481A, VMCS_GUEST_IA32_DS_BASE0x680C }, |
| 285 | { VMCS_GUEST_IA32_ES_SEL0x0800, VMCS_GUEST_IA32_ES_LIMIT0x4800, |
| 286 | VMCS_GUEST_IA32_ES_AR0x4814, VMCS_GUEST_IA32_ES_BASE0x6806 }, |
| 287 | { VMCS_GUEST_IA32_FS_SEL0x0808, VMCS_GUEST_IA32_FS_LIMIT0x4808, |
| 288 | VMCS_GUEST_IA32_FS_AR0x481C, VMCS_GUEST_IA32_FS_BASE0x680E }, |
| 289 | { VMCS_GUEST_IA32_GS_SEL0x080A, VMCS_GUEST_IA32_GS_LIMIT0x480A, |
| 290 | VMCS_GUEST_IA32_GS_AR0x481E, VMCS_GUEST_IA32_GS_BASE0x6810 }, |
| 291 | { VMCS_GUEST_IA32_SS_SEL0x0804, VMCS_GUEST_IA32_SS_LIMIT0x4804, |
| 292 | VMCS_GUEST_IA32_SS_AR0x4818, VMCS_GUEST_IA32_SS_BASE0x680A }, |
| 293 | { VMCS_GUEST_IA32_LDTR_SEL0x080C, VMCS_GUEST_IA32_LDTR_LIMIT0x480C, |
| 294 | VMCS_GUEST_IA32_LDTR_AR0x4820, VMCS_GUEST_IA32_LDTR_BASE0x6812 }, |
| 295 | { VMCS_GUEST_IA32_TR_SEL0x080E, VMCS_GUEST_IA32_TR_LIMIT0x480E, |
| 296 | VMCS_GUEST_IA32_TR_AR0x4822, VMCS_GUEST_IA32_TR_BASE0x6814 } |
| 297 | }; |
| 298 | |
| 299 | /* Pools for VMs and VCPUs */ |
| 300 | struct pool vm_pool; |
| 301 | struct pool vcpu_pool; |
| 302 | |
| 303 | struct vmm_softc *vmm_softc; |
| 304 | |
| 305 | /* IDT information used when populating host state area */ |
| 306 | extern vaddr_t idt_vaddr; |
| 307 | extern struct gate_descriptor *idt; |
| 308 | |
| 309 | /* Constants used in "CR access exit" */ |
| 310 | #define CR_WRITE0 0 |
| 311 | #define CR_READ1 1 |
| 312 | #define CR_CLTS2 2 |
| 313 | #define CR_LMSW3 3 |
| 314 | |
| 315 | /* |
| 316 | * vmm_enabled |
| 317 | * |
| 318 | * Checks if we have at least one CPU with either VMX or SVM. |
| 319 | * Returns 1 if we have at least one of either type, but not both, 0 otherwise. |
| 320 | */ |
| 321 | int |
| 322 | vmm_enabled(void) |
| 323 | { |
| 324 | struct cpu_info *ci; |
| 325 | CPU_INFO_ITERATORint cii; |
| 326 | int found_vmx = 0, found_svm = 0; |
| 327 | |
| 328 | /* Check if we have at least one CPU with either VMX or SVM */ |
| 329 | CPU_INFO_FOREACH(cii, ci)for (cii = 0, ci = cpu_info_list; ci != ((void *)0); ci = ci-> ci_next) { |
| 330 | if (ci->ci_vmm_flags & CI_VMM_VMX(1 << 0)) |
| 331 | found_vmx = 1; |
| 332 | if (ci->ci_vmm_flags & CI_VMM_SVM(1 << 1)) |
| 333 | found_svm = 1; |
| 334 | } |
| 335 | |
| 336 | /* Don't support both SVM and VMX at the same time */ |
| 337 | if (found_vmx && found_svm) |
| 338 | return (0); |
| 339 | |
| 340 | if (found_vmx || found_svm) |
| 341 | return 1; |
| 342 | |
| 343 | return 0; |
| 344 | } |
| 345 | |
| 346 | int |
| 347 | vmm_probe(struct device *parent, void *match, void *aux) |
| 348 | { |
| 349 | const char **busname = (const char **)aux; |
| 350 | |
| 351 | if (strcmp(*busname, vmm_cd.cd_name) != 0) |
| 352 | return (0); |
| 353 | return (1); |
| 354 | } |
| 355 | |
| 356 | /* |
| 357 | * vmm_attach |
| 358 | * |
| 359 | * Calculates how many of each type of CPU we have, prints this into dmesg |
| 360 | * during attach. Initializes various locks, pools, and list structures for the |
| 361 | * VMM. |
| 362 | */ |
| 363 | void |
| 364 | vmm_attach(struct device *parent, struct device *self, void *aux) |
| 365 | { |
| 366 | struct vmm_softc *sc = (struct vmm_softc *)self; |
| 367 | struct cpu_info *ci; |
| 368 | CPU_INFO_ITERATORint cii; |
| 369 | |
| 370 | sc->nr_vmx_cpus = 0; |
| 371 | sc->nr_svm_cpus = 0; |
| 372 | sc->nr_rvi_cpus = 0; |
| 373 | sc->nr_ept_cpus = 0; |
| 374 | sc->vcpu_ct = 0; |
| 375 | sc->vm_ct = 0; |
| 376 | sc->vm_idx = 0; |
| 377 | |
| 378 | /* Calculate CPU features */ |
| 379 | CPU_INFO_FOREACH(cii, ci)for (cii = 0, ci = cpu_info_list; ci != ((void *)0); ci = ci-> ci_next) { |
| 380 | if (ci->ci_vmm_flags & CI_VMM_VMX(1 << 0)) |
| 381 | sc->nr_vmx_cpus++; |
| 382 | if (ci->ci_vmm_flags & CI_VMM_SVM(1 << 1)) |
| 383 | sc->nr_svm_cpus++; |
| 384 | if (ci->ci_vmm_flags & CI_VMM_RVI(1 << 2)) |
| 385 | sc->nr_rvi_cpus++; |
| 386 | if (ci->ci_vmm_flags & CI_VMM_EPT(1 << 3)) |
| 387 | sc->nr_ept_cpus++; |
| 388 | } |
| 389 | |
| 390 | SLIST_INIT(&sc->vm_list){ ((&sc->vm_list)->slh_first) = ((void *)0); }; |
| 391 | rw_init(&sc->vm_lock, "vm_list")_rw_init_flags(&sc->vm_lock, "vm_list", 0, ((void *)0) ); |
| 392 | |
| 393 | if (sc->nr_ept_cpus) { |
| 394 | printf(": VMX/EPT"); |
| 395 | sc->mode = VMM_MODE_EPT; |
| 396 | } else if (sc->nr_vmx_cpus) { |
| 397 | printf(": VMX"); |
| 398 | sc->mode = VMM_MODE_VMX; |
| 399 | } else if (sc->nr_rvi_cpus) { |
| 400 | printf(": SVM/RVI"); |
| 401 | sc->mode = VMM_MODE_RVI; |
| 402 | } else if (sc->nr_svm_cpus) { |
| 403 | printf(": SVM"); |
| 404 | sc->mode = VMM_MODE_SVM; |
| 405 | } else { |
| 406 | printf(": unknown"); |
| 407 | sc->mode = VMM_MODE_UNKNOWN; |
| 408 | } |
| 409 | |
| 410 | if (sc->mode == VMM_MODE_EPT || sc->mode == VMM_MODE_VMX) { |
| 411 | if (!(curcpu()({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_vmm_cap.vcc_vmx.vmx_has_l1_flush_msr)) { |
| 412 | l1tf_flush_region = km_alloc(VMX_L1D_FLUSH_SIZE(64 * 1024), |
| 413 | &kv_any, &vmm_kp_contig, &kd_waitok); |
| 414 | if (!l1tf_flush_region) { |
| 415 | printf(" (failing, no memory)"); |
| 416 | sc->mode = VMM_MODE_UNKNOWN; |
| 417 | } else { |
| 418 | printf(" (using slow L1TF mitigation)"); |
| 419 | memset(l1tf_flush_region, 0xcc,__builtin_memset((l1tf_flush_region), (0xcc), ((64 * 1024))) |
| 420 | VMX_L1D_FLUSH_SIZE)__builtin_memset((l1tf_flush_region), (0xcc), ((64 * 1024))); |
| 421 | } |
| 422 | } |
| 423 | } |
| 424 | printf("\n"); |
| 425 | |
| 426 | if (sc->mode == VMM_MODE_SVM || sc->mode == VMM_MODE_RVI) { |
| 427 | sc->max_vpid = curcpu()({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_vmm_cap.vcc_svm.svm_max_asid; |
| 428 | } else { |
| 429 | sc->max_vpid = 0xFFF; |
| 430 | } |
| 431 | |
| 432 | bzero(&sc->vpids, sizeof(sc->vpids))__builtin_bzero((&sc->vpids), (sizeof(sc->vpids))); |
| 433 | rw_init(&sc->vpid_lock, "vpid")_rw_init_flags(&sc->vpid_lock, "vpid", 0, ((void *)0)); |
| 434 | |
| 435 | pool_init(&vm_pool, sizeof(struct vm), 0, IPL_MPFLOOR0x9, PR_WAITOK0x0001, |
| 436 | "vmpool", NULL((void *)0)); |
| 437 | pool_init(&vcpu_pool, sizeof(struct vcpu), 64, IPL_MPFLOOR0x9, PR_WAITOK0x0001, |
| 438 | "vcpupl", NULL((void *)0)); |
| 439 | |
| 440 | vmm_softc = sc; |
| 441 | } |
| 442 | |
| 443 | /* |
| 444 | * vmmopen |
| 445 | * |
| 446 | * Called during open of /dev/vmm. |
| 447 | * |
| 448 | * Parameters: |
| 449 | * dev, flag, mode, p: These come from the character device and are |
| 450 | * all unused for this function |
| 451 | * |
| 452 | * Return values: |
| 453 | * ENODEV: if vmm(4) didn't attach or no supported CPUs detected |
| 454 | * 0: successful open |
| 455 | */ |
| 456 | int |
| 457 | vmmopen(dev_t dev, int flag, int mode, struct proc *p) |
| 458 | { |
| 459 | /* Don't allow open if we didn't attach */ |
| 460 | if (vmm_softc == NULL((void *)0)) |
| 461 | return (ENODEV19); |
| 462 | |
| 463 | /* Don't allow open if we didn't detect any supported CPUs */ |
| 464 | if (vmm_softc->mode != VMM_MODE_EPT && vmm_softc->mode != VMM_MODE_RVI) |
| 465 | return (ENODEV19); |
| 466 | |
| 467 | return 0; |
| 468 | } |
| 469 | |
| 470 | /* |
| 471 | * vmmioctl |
| 472 | * |
| 473 | * Main ioctl dispatch routine for /dev/vmm. Parses ioctl type and calls |
| 474 | * appropriate lower level handler routine. Returns result to ioctl caller. |
| 475 | */ |
| 476 | int |
| 477 | vmmioctl(dev_t dev, u_long cmd, caddr_t data, int flag, struct proc *p) |
| 478 | { |
| 479 | int ret; |
| 480 | |
| 481 | KERNEL_UNLOCK()_kernel_unlock(); |
| 482 | |
| 483 | switch (cmd) { |
| 484 | case VMM_IOC_CREATE(((unsigned long)0x80000000|(unsigned long)0x40000000) | ((sizeof (struct vm_create_params) & 0x1fff) << 16) | ((('V' )) << 8) | ((1))): |
| 485 | if ((ret = vmm_start()) != 0) { |
| 486 | vmm_stop(); |
| 487 | break; |
| 488 | } |
| 489 | ret = vm_create((struct vm_create_params *)data, p); |
| 490 | break; |
| 491 | case VMM_IOC_RUN(((unsigned long)0x80000000|(unsigned long)0x40000000) | ((sizeof (struct vm_run_params) & 0x1fff) << 16) | ((('V')) << 8) | ((2))): |
| 492 | ret = vm_run((struct vm_run_params *)data); |
| 493 | break; |
| 494 | case VMM_IOC_INFO(((unsigned long)0x80000000|(unsigned long)0x40000000) | ((sizeof (struct vm_info_params) & 0x1fff) << 16) | ((('V')) << 8) | ((3))): |
| 495 | ret = vm_get_info((struct vm_info_params *)data); |
| 496 | break; |
| 497 | case VMM_IOC_TERM((unsigned long)0x80000000 | ((sizeof(struct vm_terminate_params ) & 0x1fff) << 16) | ((('V')) << 8) | ((4))): |
| 498 | ret = vm_terminate((struct vm_terminate_params *)data); |
| 499 | break; |
| 500 | case VMM_IOC_RESETCPU((unsigned long)0x80000000 | ((sizeof(struct vm_resetcpu_params ) & 0x1fff) << 16) | ((('V')) << 8) | ((5))): |
| 501 | ret = vm_resetcpu((struct vm_resetcpu_params *)data); |
| 502 | break; |
| 503 | case VMM_IOC_INTR((unsigned long)0x80000000 | ((sizeof(struct vm_intr_params) & 0x1fff) << 16) | ((('V')) << 8) | ((6))): |
| 504 | ret = vm_intr_pending((struct vm_intr_params *)data); |
| 505 | break; |
| 506 | case VMM_IOC_READREGS(((unsigned long)0x80000000|(unsigned long)0x40000000) | ((sizeof (struct vm_rwregs_params) & 0x1fff) << 16) | ((('V' )) << 8) | ((7))): |
| 507 | ret = vm_rwregs((struct vm_rwregs_params *)data, 0); |
| 508 | break; |
| 509 | case VMM_IOC_WRITEREGS((unsigned long)0x80000000 | ((sizeof(struct vm_rwregs_params ) & 0x1fff) << 16) | ((('V')) << 8) | ((8))): |
| 510 | ret = vm_rwregs((struct vm_rwregs_params *)data, 1); |
| 511 | break; |
| 512 | case VMM_IOC_MPROTECT_EPT((unsigned long)0x80000000 | ((sizeof(struct vm_mprotect_ept_params ) & 0x1fff) << 16) | ((('V')) << 8) | ((11))): |
| 513 | ret = vm_mprotect_ept((struct vm_mprotect_ept_params *)data); |
| 514 | break; |
| 515 | case VMM_IOC_READVMPARAMS(((unsigned long)0x80000000|(unsigned long)0x40000000) | ((sizeof (struct vm_rwvmparams_params) & 0x1fff) << 16) | (( ('V')) << 8) | ((9))): |
| 516 | ret = vm_rwvmparams((struct vm_rwvmparams_params *)data, 0); |
| 517 | break; |
| 518 | case VMM_IOC_WRITEVMPARAMS((unsigned long)0x80000000 | ((sizeof(struct vm_rwvmparams_params ) & 0x1fff) << 16) | ((('V')) << 8) | ((10))): |
| 519 | ret = vm_rwvmparams((struct vm_rwvmparams_params *)data, 1); |
| 520 | break; |
| 521 | |
| 522 | default: |
| 523 | DPRINTF("%s: unknown ioctl code 0x%lx\n", __func__, cmd); |
| 524 | ret = ENOTTY25; |
| 525 | } |
| 526 | |
| 527 | KERNEL_LOCK()_kernel_lock(); |
| 528 | |
| 529 | return (ret); |
| 530 | } |
| 531 | |
| 532 | /* |
| 533 | * pledge_ioctl_vmm |
| 534 | * |
| 535 | * Restrict the allowed ioctls in a pledged process context. |
| 536 | * Is called from pledge_ioctl(). |
| 537 | */ |
| 538 | int |
| 539 | pledge_ioctl_vmm(struct proc *p, long com) |
| 540 | { |
| 541 | switch (com) { |
| 542 | case VMM_IOC_CREATE(((unsigned long)0x80000000|(unsigned long)0x40000000) | ((sizeof (struct vm_create_params) & 0x1fff) << 16) | ((('V' )) << 8) | ((1))): |
| 543 | case VMM_IOC_INFO(((unsigned long)0x80000000|(unsigned long)0x40000000) | ((sizeof (struct vm_info_params) & 0x1fff) << 16) | ((('V')) << 8) | ((3))): |
| 544 | /* The "parent" process in vmd forks and manages VMs */ |
| 545 | if (p->p_p->ps_pledge & PLEDGE_PROC0x0000000000001000ULL) |
| 546 | return (0); |
| 547 | break; |
| 548 | case VMM_IOC_TERM((unsigned long)0x80000000 | ((sizeof(struct vm_terminate_params ) & 0x1fff) << 16) | ((('V')) << 8) | ((4))): |
| 549 | /* XXX VM processes should only terminate themselves */ |
| 550 | case VMM_IOC_RUN(((unsigned long)0x80000000|(unsigned long)0x40000000) | ((sizeof (struct vm_run_params) & 0x1fff) << 16) | ((('V')) << 8) | ((2))): |
| 551 | case VMM_IOC_RESETCPU((unsigned long)0x80000000 | ((sizeof(struct vm_resetcpu_params ) & 0x1fff) << 16) | ((('V')) << 8) | ((5))): |
| 552 | case VMM_IOC_INTR((unsigned long)0x80000000 | ((sizeof(struct vm_intr_params) & 0x1fff) << 16) | ((('V')) << 8) | ((6))): |
| 553 | case VMM_IOC_READREGS(((unsigned long)0x80000000|(unsigned long)0x40000000) | ((sizeof (struct vm_rwregs_params) & 0x1fff) << 16) | ((('V' )) << 8) | ((7))): |
| 554 | case VMM_IOC_WRITEREGS((unsigned long)0x80000000 | ((sizeof(struct vm_rwregs_params ) & 0x1fff) << 16) | ((('V')) << 8) | ((8))): |
| 555 | case VMM_IOC_MPROTECT_EPT((unsigned long)0x80000000 | ((sizeof(struct vm_mprotect_ept_params ) & 0x1fff) << 16) | ((('V')) << 8) | ((11))): |
| 556 | case VMM_IOC_READVMPARAMS(((unsigned long)0x80000000|(unsigned long)0x40000000) | ((sizeof (struct vm_rwvmparams_params) & 0x1fff) << 16) | (( ('V')) << 8) | ((9))): |
| 557 | case VMM_IOC_WRITEVMPARAMS((unsigned long)0x80000000 | ((sizeof(struct vm_rwvmparams_params ) & 0x1fff) << 16) | ((('V')) << 8) | ((10))): |
| 558 | return (0); |
| 559 | } |
| 560 | |
| 561 | return (EPERM1); |
| 562 | } |
| 563 | |
| 564 | /* |
| 565 | * vmmclose |
| 566 | * |
| 567 | * Called when /dev/vmm is closed. Presently unused. |
| 568 | */ |
| 569 | int |
| 570 | vmmclose(dev_t dev, int flag, int mode, struct proc *p) |
| 571 | { |
| 572 | return 0; |
| 573 | } |
| 574 | |
| 575 | /* |
| 576 | * vm_find_vcpu |
| 577 | * |
| 578 | * Lookup VMM VCPU by ID number |
| 579 | * |
| 580 | * Parameters: |
| 581 | * vm: vm structure |
| 582 | * id: index id of vcpu |
| 583 | * |
| 584 | * Returns pointer to vcpu structure if successful, NULL otherwise |
| 585 | */ |
| 586 | static struct vcpu * |
| 587 | vm_find_vcpu(struct vm *vm, uint32_t id) |
| 588 | { |
| 589 | struct vcpu *vcpu; |
| 590 | |
| 591 | if (vm == NULL((void *)0)) |
| 592 | return NULL((void *)0); |
| 593 | rw_enter_read(&vm->vm_vcpu_lock); |
| 594 | SLIST_FOREACH(vcpu, &vm->vm_vcpu_list, vc_vcpu_link)for((vcpu) = ((&vm->vm_vcpu_list)->slh_first); (vcpu ) != ((void *)0); (vcpu) = ((vcpu)->vc_vcpu_link.sle_next) ) { |
| 595 | if (vcpu->vc_id == id) |
| 596 | break; |
| 597 | } |
| 598 | rw_exit_read(&vm->vm_vcpu_lock); |
| 599 | return vcpu; |
| 600 | } |
| 601 | |
| 602 | |
| 603 | /* |
| 604 | * vm_resetcpu |
| 605 | * |
| 606 | * Resets the vcpu defined in 'vrp' to power-on-init register state |
| 607 | * |
| 608 | * Parameters: |
| 609 | * vrp: ioctl structure defining the vcpu to reset (see vmmvar.h) |
| 610 | * |
| 611 | * Returns 0 if successful, or various error codes on failure: |
| 612 | * ENOENT if the VM id contained in 'vrp' refers to an unknown VM or |
| 613 | * if vrp describes an unknown vcpu for this VM |
| 614 | * EBUSY if the indicated VCPU is not stopped |
| 615 | * EIO if the indicated VCPU failed to reset |
| 616 | */ |
| 617 | int |
| 618 | vm_resetcpu(struct vm_resetcpu_params *vrp) |
| 619 | { |
| 620 | struct vm *vm; |
| 621 | struct vcpu *vcpu; |
| 622 | int error; |
| 623 | |
| 624 | /* Find the desired VM */ |
| 625 | rw_enter_read(&vmm_softc->vm_lock); |
| 626 | error = vm_find(vrp->vrp_vm_id, &vm); |
| 627 | rw_exit_read(&vmm_softc->vm_lock); |
| 628 | |
| 629 | /* Not found? exit. */ |
| 630 | if (error != 0) { |
| 631 | DPRINTF("%s: vm id %u not found\n", __func__, |
| 632 | vrp->vrp_vm_id); |
| 633 | return (error); |
| 634 | } |
| 635 | |
| 636 | vcpu = vm_find_vcpu(vm, vrp->vrp_vcpu_id); |
| 637 | |
| 638 | if (vcpu == NULL((void *)0)) { |
| 639 | DPRINTF("%s: vcpu id %u of vm %u not found\n", __func__, |
| 640 | vrp->vrp_vcpu_id, vrp->vrp_vm_id); |
| 641 | return (ENOENT2); |
| 642 | } |
| 643 | |
| 644 | rw_enter_write(&vcpu->vc_lock); |
| 645 | |
| 646 | if (vcpu->vc_state != VCPU_STATE_STOPPED) { |
| 647 | DPRINTF("%s: reset of vcpu %u on vm %u attempted " |
| 648 | "while vcpu was in state %u (%s)\n", __func__, |
| 649 | vrp->vrp_vcpu_id, vrp->vrp_vm_id, vcpu->vc_state, |
| 650 | vcpu_state_decode(vcpu->vc_state)); |
| 651 | |
| 652 | rw_exit_write(&vcpu->vc_lock); |
| 653 | return (EBUSY16); |
| 654 | } |
| 655 | |
| 656 | DPRINTF("%s: resetting vm %d vcpu %d to power on defaults\n", __func__, |
| 657 | vm->vm_id, vcpu->vc_id); |
| 658 | |
| 659 | if (vcpu_reset_regs(vcpu, &vrp->vrp_init_state)) { |
| 660 | printf("%s: failed\n", __func__); |
| 661 | #ifdef VMM_DEBUG |
| 662 | dump_vcpu(vcpu); |
| 663 | #endif /* VMM_DEBUG */ |
| 664 | rw_exit_write(&vcpu->vc_lock); |
| 665 | return (EIO5); |
| 666 | } |
| 667 | |
| 668 | rw_exit_write(&vcpu->vc_lock); |
| 669 | return (0); |
| 670 | } |
| 671 | |
| 672 | /* |
| 673 | * vm_intr_pending |
| 674 | * |
| 675 | * IOCTL handler routine for VMM_IOC_INTR messages, sent from vmd when an |
| 676 | * interrupt is pending and needs acknowledgment |
| 677 | * |
| 678 | * Parameters: |
| 679 | * vip: Describes the vm/vcpu for which the interrupt is pending |
| 680 | * |
| 681 | * Return values: |
| 682 | * 0: if successful |
| 683 | * ENOENT: if the VM/VCPU defined by 'vip' cannot be found |
| 684 | */ |
| 685 | int |
| 686 | vm_intr_pending(struct vm_intr_params *vip) |
| 687 | { |
| 688 | struct vm *vm; |
| 689 | struct vcpu *vcpu; |
| 690 | int error; |
| 691 | |
| 692 | /* Find the desired VM */ |
| 693 | rw_enter_read(&vmm_softc->vm_lock); |
| 694 | error = vm_find(vip->vip_vm_id, &vm); |
| 695 | |
| 696 | /* Not found? exit. */ |
| 697 | if (error != 0) { |
| 698 | rw_exit_read(&vmm_softc->vm_lock); |
| 699 | return (error); |
| 700 | } |
| 701 | |
| 702 | vcpu = vm_find_vcpu(vm, vip->vip_vcpu_id); |
| 703 | rw_exit_read(&vmm_softc->vm_lock); |
| 704 | |
| 705 | if (vcpu == NULL((void *)0)) |
| 706 | return (ENOENT2); |
| 707 | |
| 708 | rw_enter_write(&vcpu->vc_lock); |
| 709 | vcpu->vc_intr = vip->vip_intr; |
| 710 | rw_exit_write(&vcpu->vc_lock); |
| 711 | |
| 712 | return (0); |
| 713 | } |
| 714 | |
| 715 | /* |
| 716 | * vm_rwvmparams |
| 717 | * |
| 718 | * IOCTL handler to read/write the current vmm params like pvclock gpa, pvclock |
| 719 | * version, etc. |
| 720 | * |
| 721 | * Parameters: |
| 722 | * vrwp: Describes the VM and VCPU to get/set the params from |
| 723 | * dir: 0 for reading, 1 for writing |
| 724 | * |
| 725 | * Return values: |
| 726 | * 0: if successful |
| 727 | * ENOENT: if the VM/VCPU defined by 'vpp' cannot be found |
| 728 | * EINVAL: if an error occurred reading the registers of the guest |
| 729 | */ |
| 730 | int |
| 731 | vm_rwvmparams(struct vm_rwvmparams_params *vpp, int dir) { |
| 732 | struct vm *vm; |
| 733 | struct vcpu *vcpu; |
| 734 | int error; |
| 735 | |
| 736 | /* Find the desired VM */ |
| 737 | rw_enter_read(&vmm_softc->vm_lock); |
| 738 | error = vm_find(vpp->vpp_vm_id, &vm); |
| 739 | |
| 740 | /* Not found? exit. */ |
| 741 | if (error != 0) { |
| 742 | rw_exit_read(&vmm_softc->vm_lock); |
| 743 | return (error); |
| 744 | } |
| 745 | |
| 746 | vcpu = vm_find_vcpu(vm, vpp->vpp_vcpu_id); |
| 747 | rw_exit_read(&vmm_softc->vm_lock); |
| 748 | |
| 749 | if (vcpu == NULL((void *)0)) |
| 750 | return (ENOENT2); |
| 751 | |
| 752 | if (dir == 0) { |
| 753 | if (vpp->vpp_mask & VM_RWVMPARAMS_PVCLOCK_VERSION0x2) |
| 754 | vpp->vpp_pvclock_version = vcpu->vc_pvclock_version; |
| 755 | if (vpp->vpp_mask & VM_RWVMPARAMS_PVCLOCK_SYSTEM_GPA0x1) |
| 756 | vpp->vpp_pvclock_system_gpa = \ |
| 757 | vcpu->vc_pvclock_system_gpa; |
| 758 | return (0); |
| 759 | } |
| 760 | |
| 761 | if (vpp->vpp_mask & VM_RWVMPARAMS_PVCLOCK_VERSION0x2) |
| 762 | vcpu->vc_pvclock_version = vpp->vpp_pvclock_version; |
| 763 | if (vpp->vpp_mask & VM_RWVMPARAMS_PVCLOCK_SYSTEM_GPA0x1) { |
| 764 | vmm_init_pvclock(vcpu, vpp->vpp_pvclock_system_gpa); |
| 765 | } |
| 766 | return (0); |
| 767 | |
| 768 | } |
| 769 | |
| 770 | /* |
| 771 | * vm_readregs |
| 772 | * |
| 773 | * IOCTL handler to read/write the current register values of a guest VCPU. |
| 774 | * The VCPU must not be running. |
| 775 | * |
| 776 | * Parameters: |
| 777 | * vrwp: Describes the VM and VCPU to get/set the registers from. The |
| 778 | * register values are returned here as well. |
| 779 | * dir: 0 for reading, 1 for writing |
| 780 | * |
| 781 | * Return values: |
| 782 | * 0: if successful |
| 783 | * ENOENT: if the VM/VCPU defined by 'vrwp' cannot be found |
| 784 | * EINVAL: if an error occurred accessing the registers of the guest |
| 785 | * EPERM: if the vm cannot be accessed from the calling process |
| 786 | */ |
| 787 | int |
| 788 | vm_rwregs(struct vm_rwregs_params *vrwp, int dir) |
| 789 | { |
| 790 | struct vm *vm; |
| 791 | struct vcpu *vcpu; |
| 792 | struct vcpu_reg_state *vrs = &vrwp->vrwp_regs; |
| 793 | int error, ret; |
| 794 | |
| 795 | /* Find the desired VM */ |
| 796 | rw_enter_read(&vmm_softc->vm_lock); |
| 797 | error = vm_find(vrwp->vrwp_vm_id, &vm); |
| 798 | |
| 799 | /* Not found? exit. */ |
| 800 | if (error != 0) { |
| 801 | rw_exit_read(&vmm_softc->vm_lock); |
| 802 | return (error); |
| 803 | } |
| 804 | |
| 805 | vcpu = vm_find_vcpu(vm, vrwp->vrwp_vcpu_id); |
| 806 | rw_exit_read(&vmm_softc->vm_lock); |
| 807 | |
| 808 | if (vcpu == NULL((void *)0)) |
| 809 | return (ENOENT2); |
| 810 | |
| 811 | rw_enter_write(&vcpu->vc_lock); |
| 812 | if (vmm_softc->mode == VMM_MODE_VMX || |
| 813 | vmm_softc->mode == VMM_MODE_EPT) |
| 814 | ret = (dir == 0) ? |
| 815 | vcpu_readregs_vmx(vcpu, vrwp->vrwp_mask, vrs) : |
| 816 | vcpu_writeregs_vmx(vcpu, vrwp->vrwp_mask, 1, vrs); |
| 817 | else if (vmm_softc->mode == VMM_MODE_SVM || |
| 818 | vmm_softc->mode == VMM_MODE_RVI) |
| 819 | ret = (dir == 0) ? |
| 820 | vcpu_readregs_svm(vcpu, vrwp->vrwp_mask, vrs) : |
| 821 | vcpu_writeregs_svm(vcpu, vrwp->vrwp_mask, vrs); |
| 822 | else { |
| 823 | DPRINTF("%s: unknown vmm mode", __func__); |
| 824 | ret = EINVAL22; |
| 825 | } |
| 826 | rw_exit_write(&vcpu->vc_lock); |
| 827 | |
| 828 | return (ret); |
| 829 | } |
| 830 | |
| 831 | /* |
| 832 | * vm_mprotect_ept |
| 833 | * |
| 834 | * IOCTL handler to sets the access protections of the ept |
| 835 | * |
| 836 | * Parameters: |
| 837 | * vmep: describes the memory for which the protect will be applied.. |
| 838 | * |
| 839 | * Return values: |
| 840 | * 0: if successful |
| 841 | * ENOENT: if the VM defined by 'vmep' cannot be found |
| 842 | * EINVAL: if the sgpa or size is not page aligned, the prot is invalid, |
| 843 | * size is too large (512GB), there is wraparound |
| 844 | * (like start = 512GB-1 and end = 512GB-2), |
| 845 | * the address specified is not within the vm's mem range |
| 846 | * or the address lies inside reserved (MMIO) memory |
| 847 | */ |
| 848 | int |
| 849 | vm_mprotect_ept(struct vm_mprotect_ept_params *vmep) |
| 850 | { |
| 851 | struct vm *vm; |
| 852 | struct vcpu *vcpu; |
| 853 | vaddr_t sgpa; |
| 854 | size_t size; |
| 855 | vm_prot_t prot; |
| 856 | uint64_t msr; |
| 857 | int ret, memtype; |
| 858 | |
| 859 | /* If not EPT or RVI, nothing to do here */ |
| 860 | if (!(vmm_softc->mode == VMM_MODE_EPT |
| 861 | || vmm_softc->mode == VMM_MODE_RVI)) |
| 862 | return (0); |
| 863 | |
| 864 | /* Find the desired VM */ |
| 865 | rw_enter_read(&vmm_softc->vm_lock); |
| 866 | ret = vm_find(vmep->vmep_vm_id, &vm); |
| 867 | rw_exit_read(&vmm_softc->vm_lock); |
| 868 | |
| 869 | /* Not found? exit. */ |
| 870 | if (ret != 0) { |
| 871 | DPRINTF("%s: vm id %u not found\n", __func__, |
| 872 | vmep->vmep_vm_id); |
| 873 | return (ret); |
| 874 | } |
| 875 | |
| 876 | vcpu = vm_find_vcpu(vm, vmep->vmep_vcpu_id); |
| 877 | |
| 878 | if (vcpu == NULL((void *)0)) { |
| 879 | DPRINTF("%s: vcpu id %u of vm %u not found\n", __func__, |
| 880 | vmep->vmep_vcpu_id, vmep->vmep_vm_id); |
| 881 | return (ENOENT2); |
| 882 | } |
| 883 | |
| 884 | if (vcpu->vc_state != VCPU_STATE_STOPPED) { |
| 885 | DPRINTF("%s: mprotect_ept %u on vm %u attempted " |
| 886 | "while vcpu was in state %u (%s)\n", __func__, |
| 887 | vmep->vmep_vcpu_id, vmep->vmep_vm_id, vcpu->vc_state, |
| 888 | vcpu_state_decode(vcpu->vc_state)); |
| 889 | |
| 890 | return (EBUSY16); |
| 891 | } |
| 892 | |
| 893 | /* Only proceed if the pmap is in the correct mode */ |
| 894 | KASSERT((vmm_softc->mode == VMM_MODE_EPT &&(((vmm_softc->mode == VMM_MODE_EPT && vm->vm_map ->pmap->pm_type == 2) || (vmm_softc->mode == VMM_MODE_RVI && vm->vm_map->pmap->pm_type == 3)) ? (void )0 : __assert("diagnostic ", "/usr/src/sys/arch/amd64/amd64/vmm.c" , 897, "(vmm_softc->mode == VMM_MODE_EPT && vm->vm_map->pmap->pm_type == PMAP_TYPE_EPT) || (vmm_softc->mode == VMM_MODE_RVI && vm->vm_map->pmap->pm_type == PMAP_TYPE_RVI)" )) |
| 895 | vm->vm_map->pmap->pm_type == PMAP_TYPE_EPT) ||(((vmm_softc->mode == VMM_MODE_EPT && vm->vm_map ->pmap->pm_type == 2) || (vmm_softc->mode == VMM_MODE_RVI && vm->vm_map->pmap->pm_type == 3)) ? (void )0 : __assert("diagnostic ", "/usr/src/sys/arch/amd64/amd64/vmm.c" , 897, "(vmm_softc->mode == VMM_MODE_EPT && vm->vm_map->pmap->pm_type == PMAP_TYPE_EPT) || (vmm_softc->mode == VMM_MODE_RVI && vm->vm_map->pmap->pm_type == PMAP_TYPE_RVI)" )) |
| 896 | (vmm_softc->mode == VMM_MODE_RVI &&(((vmm_softc->mode == VMM_MODE_EPT && vm->vm_map ->pmap->pm_type == 2) || (vmm_softc->mode == VMM_MODE_RVI && vm->vm_map->pmap->pm_type == 3)) ? (void )0 : __assert("diagnostic ", "/usr/src/sys/arch/amd64/amd64/vmm.c" , 897, "(vmm_softc->mode == VMM_MODE_EPT && vm->vm_map->pmap->pm_type == PMAP_TYPE_EPT) || (vmm_softc->mode == VMM_MODE_RVI && vm->vm_map->pmap->pm_type == PMAP_TYPE_RVI)" )) |
| 897 | vm->vm_map->pmap->pm_type == PMAP_TYPE_RVI))(((vmm_softc->mode == VMM_MODE_EPT && vm->vm_map ->pmap->pm_type == 2) || (vmm_softc->mode == VMM_MODE_RVI && vm->vm_map->pmap->pm_type == 3)) ? (void )0 : __assert("diagnostic ", "/usr/src/sys/arch/amd64/amd64/vmm.c" , 897, "(vmm_softc->mode == VMM_MODE_EPT && vm->vm_map->pmap->pm_type == PMAP_TYPE_EPT) || (vmm_softc->mode == VMM_MODE_RVI && vm->vm_map->pmap->pm_type == PMAP_TYPE_RVI)" )); |
| 898 | |
| 899 | sgpa = vmep->vmep_sgpa; |
| 900 | size = vmep->vmep_size; |
| 901 | prot = vmep->vmep_prot; |
| 902 | |
| 903 | /* No W^X permissions */ |
| 904 | if ((prot & PROT_MASK(0x01 | 0x02 | 0x04)) != prot && |
| 905 | (prot & (PROT_WRITE0x02 | PROT_EXEC0x04)) == (PROT_WRITE0x02 | PROT_EXEC0x04)) { |
| 906 | DPRINTF("%s: W+X permission requested\n", __func__); |
| 907 | return (EINVAL22); |
| 908 | } |
| 909 | |
| 910 | /* No Write only permissions */ |
| 911 | if ((prot & (PROT_READ0x01 | PROT_WRITE0x02 | PROT_EXEC0x04)) == PROT_WRITE0x02) { |
| 912 | DPRINTF("%s: No Write only permissions\n", __func__); |
| 913 | return (EINVAL22); |
| 914 | } |
| 915 | |
| 916 | /* No empty permissions */ |
| 917 | if (prot == 0) { |
| 918 | DPRINTF("%s: No empty permissions\n", __func__); |
| 919 | return (EINVAL22); |
| 920 | } |
| 921 | |
| 922 | /* No execute only on EPT CPUs that don't have that capability */ |
| 923 | if (vmm_softc->mode == VMM_MODE_EPT) { |
| 924 | msr = rdmsr(IA32_VMX_EPT_VPID_CAP0x48C); |
| 925 | if (prot == PROT_EXEC0x04 && |
| 926 | (msr & IA32_EPT_VPID_CAP_XO_TRANSLATIONS(1ULL << 0)) == 0) { |
| 927 | DPRINTF("%s: Execute only permissions unsupported," |
| 928 | " adding read permission\n", __func__); |
| 929 | |
| 930 | prot |= PROT_READ0x01; |
| 931 | } |
| 932 | } |
| 933 | |
| 934 | /* Must be page aligned */ |
| 935 | if ((sgpa & PAGE_MASK((1 << 12) - 1)) || (size & PAGE_MASK((1 << 12) - 1)) || size == 0) |
| 936 | return (EINVAL22); |
| 937 | |
| 938 | /* size must be less then 512GB */ |
| 939 | if (size >= NBPD_L4(1ULL << 39)) |
| 940 | return (EINVAL22); |
| 941 | |
| 942 | /* no wraparound */ |
| 943 | if (sgpa + size < sgpa) |
| 944 | return (EINVAL22); |
| 945 | |
| 946 | /* |
| 947 | * Specifying addresses within the PCI MMIO space is forbidden. |
| 948 | * Disallow addresses that start inside the MMIO space: |
| 949 | * [VMM_PCI_MMIO_BAR_BASE .. VMM_PCI_MMIO_BAR_END] |
| 950 | */ |
| 951 | if (sgpa >= VMM_PCI_MMIO_BAR_BASE0xF0000000ULL && sgpa <= VMM_PCI_MMIO_BAR_END0xFFFFFFFFULL) |
| 952 | return (EINVAL22); |
| 953 | |
| 954 | /* |
| 955 | * ... and disallow addresses that end inside the MMIO space: |
| 956 | * (VMM_PCI_MMIO_BAR_BASE .. VMM_PCI_MMIO_BAR_END] |
| 957 | */ |
| 958 | if (sgpa + size > VMM_PCI_MMIO_BAR_BASE0xF0000000ULL && |
| 959 | sgpa + size <= VMM_PCI_MMIO_BAR_END0xFFFFFFFFULL) |
| 960 | return (EINVAL22); |
| 961 | |
| 962 | memtype = vmm_get_guest_memtype(vm, sgpa); |
| 963 | if (memtype == VMM_MEM_TYPE_UNKNOWN) |
| 964 | return (EINVAL22); |
| 965 | |
| 966 | if (vmm_softc->mode == VMM_MODE_EPT) |
| 967 | ret = vmx_mprotect_ept(vm->vm_map, sgpa, sgpa + size, prot); |
| 968 | else if (vmm_softc->mode == VMM_MODE_RVI) { |
| 969 | pmap_write_protect(vm->vm_map->pmap, sgpa, sgpa + size, prot); |
| 970 | /* XXX requires a invlpga */ |
| 971 | ret = 0; |
| 972 | } else |
| 973 | return (EINVAL22); |
| 974 | |
| 975 | return (ret); |
| 976 | } |
| 977 | |
| 978 | /* |
| 979 | * vmx_mprotect_ept |
| 980 | * |
| 981 | * apply the ept protections to the requested pages, faulting in the page if |
| 982 | * required. |
| 983 | */ |
| 984 | int |
| 985 | vmx_mprotect_ept(vm_map_t vm_map, paddr_t sgpa, paddr_t egpa, int prot) |
| 986 | { |
| 987 | struct vmx_invept_descriptor vid; |
| 988 | pmap_t pmap; |
| 989 | pt_entry_t *pte; |
| 990 | paddr_t addr; |
| 991 | int ret = 0; |
| 992 | |
| 993 | pmap = vm_map->pmap; |
| 994 | |
| 995 | KERNEL_LOCK()_kernel_lock(); |
| 996 | |
| 997 | for (addr = sgpa; addr < egpa; addr += PAGE_SIZE(1 << 12)) { |
| 998 | pte = vmx_pmap_find_pte_ept(pmap, addr); |
| 999 | if (pte == NULL((void *)0)) { |
| 1000 | ret = uvm_fault(vm_map, addr, VM_FAULT_WIRE((vm_fault_t) 0x2), |
| 1001 | PROT_READ0x01 | PROT_WRITE0x02 | PROT_EXEC0x04); |
| 1002 | if (ret) |
| 1003 | printf("%s: uvm_fault returns %d, GPA=0x%llx\n", |
| 1004 | __func__, ret, (uint64_t)addr); |
| 1005 | |
| 1006 | pte = vmx_pmap_find_pte_ept(pmap, addr); |
| 1007 | if (pte == NULL((void *)0)) { |
| 1008 | KERNEL_UNLOCK()_kernel_unlock(); |
| 1009 | return EFAULT14; |
| 1010 | } |
| 1011 | } |
| 1012 | |
| 1013 | if (prot & PROT_READ0x01) |
| 1014 | *pte |= EPT_R(1ULL << 0); |
| 1015 | else |
| 1016 | *pte &= ~EPT_R(1ULL << 0); |
| 1017 | |
| 1018 | if (prot & PROT_WRITE0x02) |
| 1019 | *pte |= EPT_W(1ULL << 1); |
| 1020 | else |
| 1021 | *pte &= ~EPT_W(1ULL << 1); |
| 1022 | |
| 1023 | if (prot & PROT_EXEC0x04) |
| 1024 | *pte |= EPT_X(1ULL << 2); |
| 1025 | else |
| 1026 | *pte &= ~EPT_X(1ULL << 2); |
| 1027 | } |
| 1028 | |
| 1029 | /* |
| 1030 | * SDM 3C: 28.3.3.4 Guidelines for Use of the INVEPT Instruction |
| 1031 | * the first bullet point seems to say we should call invept. |
| 1032 | * |
| 1033 | * Software should use the INVEPT instruction with the “single-context” |
| 1034 | * INVEPT type after making any of the following changes to an EPT |
| 1035 | * paging-structure entry (the INVEPT descriptor should contain an |
| 1036 | * EPTP value that references — directly or indirectly |
| 1037 | * — the modified EPT paging structure): |
| 1038 | * — Changing any of the privilege bits 2:0 from 1 to 0. |
| 1039 | * */ |
| 1040 | if (pmap->eptp != 0) { |
| 1041 | memset(&vid, 0, sizeof(vid))__builtin_memset((&vid), (0), (sizeof(vid))); |
| 1042 | vid.vid_eptp = pmap->eptp; |
| 1043 | DPRINTF("%s: flushing EPT TLB for EPTP 0x%llx\n", __func__, |
| 1044 | vid.vid_eptp); |
| 1045 | invept(IA32_VMX_INVEPT_SINGLE_CTX0x1, &vid); |
| 1046 | } |
| 1047 | |
| 1048 | KERNEL_UNLOCK()_kernel_unlock(); |
| 1049 | |
| 1050 | return ret; |
| 1051 | } |
| 1052 | |
| 1053 | /* |
| 1054 | * vmx_pmap_find_pte_ept |
| 1055 | * |
| 1056 | * find the page table entry specified by addr in the pmap supplied. |
| 1057 | */ |
| 1058 | pt_entry_t * |
| 1059 | vmx_pmap_find_pte_ept(pmap_t pmap, paddr_t addr) |
| 1060 | { |
| 1061 | int l4idx, l3idx, l2idx, l1idx; |
| 1062 | pd_entry_t *pd; |
| 1063 | paddr_t pdppa; |
| 1064 | pt_entry_t *ptes, *pte; |
| 1065 | |
| 1066 | l4idx = (addr & L4_MASK0x0000ff8000000000UL) >> L4_SHIFT39; /* PML4E idx */ |
| 1067 | l3idx = (addr & L3_MASK0x0000007fc0000000UL) >> L3_SHIFT30; /* PDPTE idx */ |
| 1068 | l2idx = (addr & L2_MASK0x000000003fe00000UL) >> L2_SHIFT21; /* PDE idx */ |
| 1069 | l1idx = (addr & L1_MASK0x00000000001ff000UL) >> L1_SHIFT12; /* PTE idx */ |
| 1070 | |
| 1071 | pd = (pd_entry_t *)pmap->pm_pdir; |
| 1072 | if (pd == NULL((void *)0)) |
| 1073 | return NULL((void *)0); |
| 1074 | |
| 1075 | /* |
| 1076 | * l4idx should always be 0 since we don't support more than 512GB |
| 1077 | * guest physical memory. |
| 1078 | */ |
| 1079 | if (l4idx > 0) |
| 1080 | return NULL((void *)0); |
| 1081 | |
| 1082 | /* |
| 1083 | * l3idx should always be < MAXDSIZ/1GB because we don't support more |
| 1084 | * than MAXDSIZ guest phys mem. |
| 1085 | */ |
| 1086 | if (l3idx >= MAXDSIZ((paddr_t)32*1024*1024*1024) / ((paddr_t)1024 * 1024 * 1024)) |
| 1087 | return NULL((void *)0); |
| 1088 | |
| 1089 | pdppa = pd[l4idx] & PG_FRAME0x000ffffffffff000UL; |
| 1090 | if (pdppa == 0) |
| 1091 | return NULL((void *)0); |
| 1092 | |
| 1093 | ptes = (pt_entry_t *)PMAP_DIRECT_MAP(pdppa)((vaddr_t)(((((511 - 4) * (1ULL << 39))) | 0xffff000000000000 )) + (pdppa)); |
| 1094 | |
| 1095 | pdppa = ptes[l3idx] & PG_FRAME0x000ffffffffff000UL; |
| 1096 | if (pdppa == 0) |
| 1097 | return NULL((void *)0); |
| 1098 | |
| 1099 | ptes = (pt_entry_t *)PMAP_DIRECT_MAP(pdppa)((vaddr_t)(((((511 - 4) * (1ULL << 39))) | 0xffff000000000000 )) + (pdppa)); |
| 1100 | |
| 1101 | pdppa = ptes[l2idx] & PG_FRAME0x000ffffffffff000UL; |
| 1102 | if (pdppa == 0) |
| 1103 | return NULL((void *)0); |
| 1104 | |
| 1105 | ptes = (pt_entry_t *)PMAP_DIRECT_MAP(pdppa)((vaddr_t)(((((511 - 4) * (1ULL << 39))) | 0xffff000000000000 )) + (pdppa)); |
| 1106 | |
| 1107 | pte = &ptes[l1idx]; |
| 1108 | if (*pte == 0) |
| 1109 | return NULL((void *)0); |
| 1110 | |
| 1111 | return pte; |
| 1112 | } |
| 1113 | |
| 1114 | /* |
| 1115 | * vm_find |
| 1116 | * |
| 1117 | * Function to find an existing VM by its identifier. |
| 1118 | * Must be called under the global vm_lock. |
| 1119 | * |
| 1120 | * Parameters: |
| 1121 | * id: The VM identifier. |
| 1122 | * *res: A pointer to the VM or NULL if not found |
| 1123 | * |
| 1124 | * Return values: |
| 1125 | * 0: if successful |
| 1126 | * ENOENT: if the VM defined by 'id' cannot be found |
| 1127 | * EPERM: if the VM cannot be accessed by the current process |
| 1128 | */ |
| 1129 | int |
| 1130 | vm_find(uint32_t id, struct vm **res) |
| 1131 | { |
| 1132 | struct proc *p = curproc({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc; |
| 1133 | struct vm *vm; |
| 1134 | |
| 1135 | *res = NULL((void *)0); |
| 1136 | SLIST_FOREACH(vm, &vmm_softc->vm_list, vm_link)for((vm) = ((&vmm_softc->vm_list)->slh_first); (vm) != ((void *)0); (vm) = ((vm)->vm_link.sle_next)) { |
| 1137 | if (vm->vm_id == id) { |
| 1138 | /* |
| 1139 | * In the pledged VM process, only allow to find |
| 1140 | * the VM that is running in the current process. |
| 1141 | * The managing vmm parent process can lookup all |
| 1142 | * all VMs and is indicated by PLEDGE_PROC. |
| 1143 | */ |
| 1144 | if (((p->p_p->ps_pledge & |
| 1145 | (PLEDGE_VMM0x0000000040000000ULL | PLEDGE_PROC0x0000000000001000ULL)) == PLEDGE_VMM0x0000000040000000ULL) && |
| 1146 | (vm->vm_creator_pid != p->p_p->ps_pid)) |
| 1147 | return (pledge_fail(p, EPERM1, PLEDGE_VMM0x0000000040000000ULL)); |
| 1148 | *res = vm; |
| 1149 | return (0); |
| 1150 | } |
| 1151 | } |
| 1152 | |
| 1153 | return (ENOENT2); |
| 1154 | } |
| 1155 | |
| 1156 | /* |
| 1157 | * vmm_start |
| 1158 | * |
| 1159 | * Starts VMM mode on the system |
| 1160 | */ |
| 1161 | int |
| 1162 | vmm_start(void) |
| 1163 | { |
| 1164 | struct cpu_info *self = curcpu()({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;}); |
| 1165 | int ret = 0; |
| 1166 | #ifdef MULTIPROCESSOR1 |
| 1167 | struct cpu_info *ci; |
| 1168 | CPU_INFO_ITERATORint cii; |
| 1169 | int i; |
| 1170 | #endif |
| 1171 | |
| 1172 | /* VMM is already running */ |
| 1173 | if (self->ci_flags & CPUF_VMM0x20000) |
| 1174 | return (0); |
| 1175 | |
| 1176 | #ifdef MULTIPROCESSOR1 |
| 1177 | /* Broadcast start VMM IPI */ |
| 1178 | x86_broadcast_ipi(X86_IPI_START_VMM0x00000100); |
| 1179 | |
| 1180 | CPU_INFO_FOREACH(cii, ci)for (cii = 0, ci = cpu_info_list; ci != ((void *)0); ci = ci-> ci_next) { |
| 1181 | if (ci == self) |
| 1182 | continue; |
| 1183 | for (i = 100000; (!(ci->ci_flags & CPUF_VMM0x20000)) && i>0;i--) |
| 1184 | delay(10)(*delay_func)(10); |
| 1185 | if (!(ci->ci_flags & CPUF_VMM0x20000)) { |
| 1186 | printf("%s: failed to enter VMM mode\n", |
| 1187 | ci->ci_dev->dv_xname); |
| 1188 | ret = EIO5; |
| 1189 | } |
| 1190 | } |
| 1191 | #endif /* MULTIPROCESSOR */ |
| 1192 | |
| 1193 | /* Start VMM on this CPU */ |
| 1194 | start_vmm_on_cpu(self); |
| 1195 | if (!(self->ci_flags & CPUF_VMM0x20000)) { |
| 1196 | printf("%s: failed to enter VMM mode\n", |
| 1197 | self->ci_dev->dv_xname); |
| 1198 | ret = EIO5; |
| 1199 | } |
| 1200 | |
| 1201 | return (ret); |
| 1202 | } |
| 1203 | |
| 1204 | /* |
| 1205 | * vmm_stop |
| 1206 | * |
| 1207 | * Stops VMM mode on the system |
| 1208 | */ |
| 1209 | int |
| 1210 | vmm_stop(void) |
| 1211 | { |
| 1212 | struct cpu_info *self = curcpu()({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;}); |
| 1213 | int ret = 0; |
| 1214 | #ifdef MULTIPROCESSOR1 |
| 1215 | struct cpu_info *ci; |
| 1216 | CPU_INFO_ITERATORint cii; |
| 1217 | int i; |
| 1218 | #endif |
| 1219 | |
| 1220 | /* VMM is not running */ |
| 1221 | if (!(self->ci_flags & CPUF_VMM0x20000)) |
| 1222 | return (0); |
| 1223 | |
| 1224 | #ifdef MULTIPROCESSOR1 |
| 1225 | /* Stop VMM on other CPUs */ |
| 1226 | x86_broadcast_ipi(X86_IPI_STOP_VMM0x00000200); |
| 1227 | |
| 1228 | CPU_INFO_FOREACH(cii, ci)for (cii = 0, ci = cpu_info_list; ci != ((void *)0); ci = ci-> ci_next) { |
| 1229 | if (ci == self) |
| 1230 | continue; |
| 1231 | for (i = 100000; (ci->ci_flags & CPUF_VMM0x20000) && i>0 ;i--) |
| 1232 | delay(10)(*delay_func)(10); |
| 1233 | if (ci->ci_flags & CPUF_VMM0x20000) { |
| 1234 | printf("%s: failed to exit VMM mode\n", |
| 1235 | ci->ci_dev->dv_xname); |
| 1236 | ret = EIO5; |
| 1237 | } |
| 1238 | } |
| 1239 | #endif /* MULTIPROCESSOR */ |
| 1240 | |
| 1241 | /* Stop VMM on this CPU */ |
| 1242 | stop_vmm_on_cpu(self); |
| 1243 | if (self->ci_flags & CPUF_VMM0x20000) { |
| 1244 | printf("%s: failed to exit VMM mode\n", |
| 1245 | self->ci_dev->dv_xname); |
| 1246 | ret = EIO5; |
| 1247 | } |
| 1248 | |
| 1249 | return (ret); |
| 1250 | } |
| 1251 | |
| 1252 | /* |
| 1253 | * start_vmm_on_cpu |
| 1254 | * |
| 1255 | * Starts VMM mode on 'ci' by executing the appropriate CPU-specific insn |
| 1256 | * sequence to enter VMM mode (eg, VMXON) |
| 1257 | */ |
| 1258 | void |
| 1259 | start_vmm_on_cpu(struct cpu_info *ci) |
| 1260 | { |
| 1261 | uint64_t msr; |
| 1262 | uint32_t cr4; |
| 1263 | |
| 1264 | /* No VMM mode? exit. */ |
| 1265 | if ((ci->ci_vmm_flags & CI_VMM_VMX(1 << 0)) == 0 && |
| 1266 | (ci->ci_vmm_flags & CI_VMM_SVM(1 << 1)) == 0) |
| 1267 | return; |
| 1268 | |
| 1269 | /* |
| 1270 | * AMD SVM |
| 1271 | */ |
| 1272 | if (ci->ci_vmm_flags & CI_VMM_SVM(1 << 1)) { |
| 1273 | msr = rdmsr(MSR_EFER0xc0000080); |
| 1274 | msr |= EFER_SVME0x00001000; |
| 1275 | wrmsr(MSR_EFER0xc0000080, msr); |
| 1276 | } |
| 1277 | |
| 1278 | /* |
| 1279 | * Intel VMX |
| 1280 | */ |
| 1281 | if (ci->ci_vmm_flags & CI_VMM_VMX(1 << 0)) { |
| 1282 | if (ci->ci_vmxon_region == 0) |
| 1283 | return; |
| 1284 | else { |
| 1285 | bzero(ci->ci_vmxon_region, PAGE_SIZE)__builtin_bzero((ci->ci_vmxon_region), ((1 << 12))); |
| 1286 | ci->ci_vmxon_region->vr_revision = |
| 1287 | ci->ci_vmm_cap.vcc_vmx.vmx_vmxon_revision; |
| 1288 | |
| 1289 | /* Set CR4.VMXE */ |
| 1290 | cr4 = rcr4(); |
| 1291 | cr4 |= CR4_VMXE0x00002000; |
| 1292 | lcr4(cr4); |
| 1293 | |
| 1294 | /* Enable VMX */ |
| 1295 | msr = rdmsr(MSR_IA32_FEATURE_CONTROL0x03a); |
| 1296 | if (msr & IA32_FEATURE_CONTROL_LOCK0x01) { |
| 1297 | if (!(msr & IA32_FEATURE_CONTROL_VMX_EN0x04)) |
| 1298 | return; |
| 1299 | } else { |
| 1300 | msr |= IA32_FEATURE_CONTROL_VMX_EN0x04 | |
| 1301 | IA32_FEATURE_CONTROL_LOCK0x01; |
| 1302 | wrmsr(MSR_IA32_FEATURE_CONTROL0x03a, msr); |
| 1303 | } |
| 1304 | |
| 1305 | /* Enter VMX mode */ |
| 1306 | if (vmxon((uint64_t *)&ci->ci_vmxon_region_pa)) |
| 1307 | return; |
| 1308 | } |
| 1309 | } |
| 1310 | |
| 1311 | ci->ci_flags |= CPUF_VMM0x20000; |
| 1312 | } |
| 1313 | |
| 1314 | /* |
| 1315 | * stop_vmm_on_cpu |
| 1316 | * |
| 1317 | * Stops VMM mode on 'ci' by executing the appropriate CPU-specific insn |
| 1318 | * sequence to exit VMM mode (eg, VMXOFF) |
| 1319 | */ |
| 1320 | void |
| 1321 | stop_vmm_on_cpu(struct cpu_info *ci) |
| 1322 | { |
| 1323 | uint64_t msr; |
| 1324 | uint32_t cr4; |
| 1325 | |
| 1326 | if (!(ci->ci_flags & CPUF_VMM0x20000)) |
| 1327 | return; |
| 1328 | |
| 1329 | /* |
| 1330 | * AMD SVM |
| 1331 | */ |
| 1332 | if (ci->ci_vmm_flags & CI_VMM_SVM(1 << 1)) { |
| 1333 | msr = rdmsr(MSR_EFER0xc0000080); |
| 1334 | msr &= ~EFER_SVME0x00001000; |
| 1335 | wrmsr(MSR_EFER0xc0000080, msr); |
| 1336 | } |
| 1337 | |
| 1338 | /* |
| 1339 | * Intel VMX |
| 1340 | */ |
| 1341 | if (ci->ci_vmm_flags & CI_VMM_VMX(1 << 0)) { |
| 1342 | if (vmxoff()) |
| 1343 | panic("VMXOFF failed"); |
| 1344 | |
| 1345 | cr4 = rcr4(); |
| 1346 | cr4 &= ~CR4_VMXE0x00002000; |
| 1347 | lcr4(cr4); |
| 1348 | } |
| 1349 | |
| 1350 | ci->ci_flags &= ~CPUF_VMM0x20000; |
| 1351 | } |
| 1352 | |
| 1353 | /* |
| 1354 | * vmclear_on_cpu |
| 1355 | * |
| 1356 | * Flush and clear VMCS on 'ci' by executing vmclear. |
| 1357 | * |
| 1358 | */ |
| 1359 | void |
| 1360 | vmclear_on_cpu(struct cpu_info *ci) |
| 1361 | { |
| 1362 | if ((ci->ci_flags & CPUF_VMM0x20000) && (ci->ci_vmm_flags & CI_VMM_VMX(1 << 0))) { |
| 1363 | if (vmclear(&ci->ci_vmcs_pa)) |
| 1364 | panic("VMCLEAR ipi failed"); |
| 1365 | atomic_swap_ulong(&ci->ci_vmcs_pa, VMX_VMCS_PA_CLEAR)_atomic_swap_ulong((&ci->ci_vmcs_pa), (0xFFFFFFFFFFFFFFFFUL )); |
| 1366 | } |
| 1367 | } |
| 1368 | |
| 1369 | #ifdef MULTIPROCESSOR1 |
| 1370 | static int |
| 1371 | vmx_remote_vmclear(struct cpu_info *ci, struct vcpu *vcpu) |
| 1372 | { |
| 1373 | int ret = 0, nticks = 200000000; |
| 1374 | |
| 1375 | rw_enter_write(&ci->ci_vmcs_lock); |
| 1376 | atomic_swap_ulong(&ci->ci_vmcs_pa, vcpu->vc_control_pa)_atomic_swap_ulong((&ci->ci_vmcs_pa), (vcpu->vc_control_pa )); |
| 1377 | x86_send_ipi(ci, X86_IPI_VMCLEAR_VMM0x00000004); |
| 1378 | |
| 1379 | while (ci->ci_vmcs_pa != VMX_VMCS_PA_CLEAR0xFFFFFFFFFFFFFFFFUL) { |
| 1380 | CPU_BUSY_CYCLE()__asm volatile("pause": : : "memory"); |
| 1381 | if (--nticks <= 0) { |
| 1382 | printf("%s: spun out\n", __func__); |
| 1383 | ret = 1; |
| 1384 | break; |
| 1385 | } |
| 1386 | } |
| 1387 | atomic_swap_uint(&vcpu->vc_vmx_vmcs_state, VMCS_CLEARED)_atomic_swap_uint((&vcpu->vc_vmx_vmcs_state), (0)); |
| 1388 | rw_exit_write(&ci->ci_vmcs_lock); |
| 1389 | |
| 1390 | return (ret); |
| 1391 | } |
| 1392 | #endif /* MULTIPROCESSOR */ |
| 1393 | |
| 1394 | /* |
| 1395 | * vm_create_check_mem_ranges |
| 1396 | * |
| 1397 | * Make sure that the guest physical memory ranges given by the user process |
| 1398 | * do not overlap and are in ascending order. |
| 1399 | * |
| 1400 | * The last physical address may not exceed VMM_MAX_VM_MEM_SIZE. |
| 1401 | * |
| 1402 | * Return Values: |
| 1403 | * The total memory size in MB if the checks were successful |
| 1404 | * 0: One of the memory ranges was invalid, or VMM_MAX_VM_MEM_SIZE was |
| 1405 | * exceeded |
| 1406 | */ |
| 1407 | size_t |
| 1408 | vm_create_check_mem_ranges(struct vm_create_params *vcp) |
| 1409 | { |
| 1410 | size_t i, memsize = 0; |
| 1411 | struct vm_mem_range *vmr, *pvmr; |
| 1412 | const paddr_t maxgpa = (uint64_t)VMM_MAX_VM_MEM_SIZE32768 * 1024 * 1024; |
| 1413 | |
| 1414 | if (vcp->vcp_nmemranges == 0 || |
| 1415 | vcp->vcp_nmemranges > VMM_MAX_MEM_RANGES16) |
| 1416 | return (0); |
| 1417 | |
| 1418 | for (i = 0; i < vcp->vcp_nmemranges; i++) { |
| 1419 | vmr = &vcp->vcp_memranges[i]; |
| 1420 | |
| 1421 | /* Only page-aligned addresses and sizes are permitted */ |
| 1422 | if ((vmr->vmr_gpa & PAGE_MASK((1 << 12) - 1)) || (vmr->vmr_va & PAGE_MASK((1 << 12) - 1)) || |
| 1423 | (vmr->vmr_size & PAGE_MASK((1 << 12) - 1)) || vmr->vmr_size == 0) |
| 1424 | return (0); |
| 1425 | |
| 1426 | /* Make sure that VMM_MAX_VM_MEM_SIZE is not exceeded */ |
| 1427 | if (vmr->vmr_gpa >= maxgpa || |
| 1428 | vmr->vmr_size > maxgpa - vmr->vmr_gpa) |
| 1429 | return (0); |
| 1430 | |
| 1431 | /* |
| 1432 | * Make sure that all virtual addresses are within the address |
| 1433 | * space of the process and that they do not wrap around. |
| 1434 | * Calling uvm_share() when creating the VM will take care of |
| 1435 | * further checks. |
| 1436 | */ |
| 1437 | if (vmr->vmr_va < VM_MIN_ADDRESS(1 << 12) || |
| 1438 | vmr->vmr_va >= VM_MAXUSER_ADDRESS0x00007f7fffffc000 || |
| 1439 | vmr->vmr_size >= VM_MAXUSER_ADDRESS0x00007f7fffffc000 - vmr->vmr_va) |
| 1440 | return (0); |
| 1441 | |
| 1442 | /* |
| 1443 | * Specifying ranges within the PCI MMIO space is forbidden. |
| 1444 | * Disallow ranges that start inside the MMIO space: |
| 1445 | * [VMM_PCI_MMIO_BAR_BASE .. VMM_PCI_MMIO_BAR_END] |
| 1446 | */ |
| 1447 | if (vmr->vmr_gpa >= VMM_PCI_MMIO_BAR_BASE0xF0000000ULL && |
| 1448 | vmr->vmr_gpa <= VMM_PCI_MMIO_BAR_END0xFFFFFFFFULL) |
| 1449 | return (0); |
| 1450 | |
| 1451 | /* |
| 1452 | * ... and disallow ranges that end inside the MMIO space: |
| 1453 | * (VMM_PCI_MMIO_BAR_BASE .. VMM_PCI_MMIO_BAR_END] |
| 1454 | */ |
| 1455 | if (vmr->vmr_gpa + vmr->vmr_size > VMM_PCI_MMIO_BAR_BASE0xF0000000ULL && |
| 1456 | vmr->vmr_gpa + vmr->vmr_size <= VMM_PCI_MMIO_BAR_END0xFFFFFFFFULL) |
| 1457 | return (0); |
| 1458 | |
| 1459 | /* |
| 1460 | * Make sure that guest physical memory ranges do not overlap |
| 1461 | * and that they are ascending. |
| 1462 | */ |
| 1463 | if (i > 0 && pvmr->vmr_gpa + pvmr->vmr_size > vmr->vmr_gpa) |
| 1464 | return (0); |
| 1465 | |
| 1466 | memsize += vmr->vmr_size; |
| 1467 | pvmr = vmr; |
| 1468 | } |
| 1469 | |
| 1470 | if (memsize % (1024 * 1024) != 0) |
| 1471 | return (0); |
| 1472 | memsize /= 1024 * 1024; |
| 1473 | return (memsize); |
| 1474 | } |
| 1475 | |
| 1476 | /* |
| 1477 | * vm_create |
| 1478 | * |
| 1479 | * Creates the in-memory VMM structures for the VM defined by 'vcp'. The |
| 1480 | * parent of this VM shall be the process defined by 'p'. |
| 1481 | * This function does not start the VCPU(s) - see vm_start. |
| 1482 | * |
| 1483 | * Return Values: |
| 1484 | * 0: the create operation was successful |
| 1485 | * ENOMEM: out of memory |
| 1486 | * various other errors from vcpu_init/vm_impl_init |
| 1487 | */ |
| 1488 | int |
| 1489 | vm_create(struct vm_create_params *vcp, struct proc *p) |
| 1490 | { |
| 1491 | int i, ret; |
| 1492 | size_t memsize; |
| 1493 | struct vm *vm; |
| 1494 | struct vcpu *vcpu; |
| 1495 | |
| 1496 | if (!(curcpu()({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_flags & CPUF_VMM0x20000)) |
| 1497 | return (EINVAL22); |
| 1498 | |
| 1499 | memsize = vm_create_check_mem_ranges(vcp); |
| 1500 | if (memsize == 0) |
| 1501 | return (EINVAL22); |
| 1502 | |
| 1503 | /* XXX - support UP only (for now) */ |
| 1504 | if (vcp->vcp_ncpus != 1) |
| 1505 | return (EINVAL22); |
| 1506 | |
| 1507 | rw_enter_write(&vmm_softc->vm_lock); |
| 1508 | if (vmm_softc->vcpu_ct + vcp->vcp_ncpus > VMM_MAX_VCPUS512) { |
| 1509 | DPRINTF("%s: maximum vcpus (%lu) reached\n", __func__, |
| 1510 | vmm_softc->vcpu_max); |
| 1511 | rw_exit_write(&vmm_softc->vm_lock); |
| 1512 | return (ENOMEM12); |
| 1513 | } |
| 1514 | vmm_softc->vcpu_ct += vcp->vcp_ncpus; |
| 1515 | |
| 1516 | vm = pool_get(&vm_pool, PR_WAITOK0x0001 | PR_ZERO0x0008); |
| 1517 | SLIST_INIT(&vm->vm_vcpu_list){ ((&vm->vm_vcpu_list)->slh_first) = ((void *)0); }; |
| 1518 | rw_init(&vm->vm_vcpu_lock, "vcpu_list")_rw_init_flags(&vm->vm_vcpu_lock, "vcpu_list", 0, ((void *)0)); |
| 1519 | |
| 1520 | vm->vm_creator_pid = p->p_p->ps_pid; |
| 1521 | vm->vm_nmemranges = vcp->vcp_nmemranges; |
| 1522 | memcpy(vm->vm_memranges, vcp->vcp_memranges,__builtin_memcpy((vm->vm_memranges), (vcp->vcp_memranges ), (vm->vm_nmemranges * sizeof(vm->vm_memranges[0]))) |
| 1523 | vm->vm_nmemranges * sizeof(vm->vm_memranges[0]))__builtin_memcpy((vm->vm_memranges), (vcp->vcp_memranges ), (vm->vm_nmemranges * sizeof(vm->vm_memranges[0]))); |
| 1524 | vm->vm_memory_size = memsize; |
| 1525 | strncpy(vm->vm_name, vcp->vcp_name, VMM_MAX_NAME_LEN64 - 1); |
| 1526 | |
| 1527 | if (vm_impl_init(vm, p)) { |
| 1528 | printf("failed to init arch-specific features for vm %p\n", vm); |
| 1529 | vm_teardown(vm); |
| 1530 | rw_exit_write(&vmm_softc->vm_lock); |
| 1531 | return (ENOMEM12); |
| 1532 | } |
| 1533 | |
| 1534 | vmm_softc->vm_ct++; |
| 1535 | vmm_softc->vm_idx++; |
| 1536 | |
| 1537 | vm->vm_id = vmm_softc->vm_idx; |
| 1538 | vm->vm_vcpu_ct = 0; |
| 1539 | vm->vm_vcpus_running = 0; |
| 1540 | |
| 1541 | /* Initialize each VCPU defined in 'vcp' */ |
| 1542 | for (i = 0; i < vcp->vcp_ncpus; i++) { |
| 1543 | vcpu = pool_get(&vcpu_pool, PR_WAITOK0x0001 | PR_ZERO0x0008); |
| 1544 | vcpu->vc_parent = vm; |
| 1545 | if ((ret = vcpu_init(vcpu)) != 0) { |
| 1546 | printf("failed to init vcpu %d for vm %p\n", i, vm); |
| 1547 | vm_teardown(vm); |
| 1548 | vmm_softc->vm_idx--; |
| 1549 | rw_exit_write(&vmm_softc->vm_lock); |
| 1550 | return (ret); |
| 1551 | } |
| 1552 | rw_enter_write(&vm->vm_vcpu_lock); |
| 1553 | vcpu->vc_id = vm->vm_vcpu_ct; |
| 1554 | vm->vm_vcpu_ct++; |
| 1555 | SLIST_INSERT_HEAD(&vm->vm_vcpu_list, vcpu, vc_vcpu_link)do { (vcpu)->vc_vcpu_link.sle_next = (&vm->vm_vcpu_list )->slh_first; (&vm->vm_vcpu_list)->slh_first = ( vcpu); } while (0); |
| 1556 | rw_exit_write(&vm->vm_vcpu_lock); |
| 1557 | } |
| 1558 | |
| 1559 | /* XXX init various other hardware parts (vlapic, vioapic, etc) */ |
| 1560 | |
| 1561 | SLIST_INSERT_HEAD(&vmm_softc->vm_list, vm, vm_link)do { (vm)->vm_link.sle_next = (&vmm_softc->vm_list) ->slh_first; (&vmm_softc->vm_list)->slh_first = ( vm); } while (0); |
| 1562 | rw_exit_write(&vmm_softc->vm_lock); |
| 1563 | |
| 1564 | vcp->vcp_id = vm->vm_id; |
| 1565 | |
| 1566 | return (0); |
| 1567 | } |
| 1568 | |
| 1569 | /* |
| 1570 | * vm_impl_init_vmx |
| 1571 | * |
| 1572 | * Intel VMX specific VM initialization routine |
| 1573 | * |
| 1574 | * Parameters: |
| 1575 | * vm: the VM being initialized |
| 1576 | * p: vmd process owning the VM |
| 1577 | * |
| 1578 | * Return values: |
| 1579 | * 0: the initialization was successful |
| 1580 | * ENOMEM: the initialization failed (lack of resources) |
| 1581 | */ |
| 1582 | int |
| 1583 | vm_impl_init_vmx(struct vm *vm, struct proc *p) |
| 1584 | { |
| 1585 | int i, ret; |
| 1586 | vaddr_t mingpa, maxgpa; |
| 1587 | struct vm_mem_range *vmr; |
| 1588 | |
| 1589 | /* If not EPT, nothing to do here */ |
| 1590 | if (vmm_softc->mode != VMM_MODE_EPT) |
| 1591 | return (0); |
| 1592 | |
| 1593 | vmr = &vm->vm_memranges[0]; |
| 1594 | mingpa = vmr->vmr_gpa; |
| 1595 | vmr = &vm->vm_memranges[vm->vm_nmemranges - 1]; |
| 1596 | maxgpa = vmr->vmr_gpa + vmr->vmr_size; |
| 1597 | |
| 1598 | /* |
| 1599 | * uvmspace_alloc (currently) always returns a valid vmspace |
| 1600 | */ |
| 1601 | vm->vm_vmspace = uvmspace_alloc(mingpa, maxgpa, TRUE1, FALSE0); |
| 1602 | vm->vm_map = &vm->vm_vmspace->vm_map; |
| 1603 | |
| 1604 | /* Map the new map with an anon */ |
| 1605 | DPRINTF("%s: created vm_map @ %p\n", __func__, vm->vm_map); |
| 1606 | for (i = 0; i < vm->vm_nmemranges; i++) { |
| 1607 | vmr = &vm->vm_memranges[i]; |
| 1608 | ret = uvm_share(vm->vm_map, vmr->vmr_gpa, |
| 1609 | PROT_READ0x01 | PROT_WRITE0x02 | PROT_EXEC0x04, |
| 1610 | &p->p_vmspace->vm_map, vmr->vmr_va, vmr->vmr_size); |
| 1611 | if (ret) { |
| 1612 | printf("%s: uvm_share failed (%d)\n", __func__, ret); |
| 1613 | /* uvmspace_free calls pmap_destroy for us */ |
| 1614 | uvmspace_free(vm->vm_vmspace); |
| 1615 | vm->vm_vmspace = NULL((void *)0); |
| 1616 | return (ENOMEM12); |
| 1617 | } |
| 1618 | } |
| 1619 | |
| 1620 | ret = pmap_convert(vm->vm_map->pmap, PMAP_TYPE_EPT2); |
| 1621 | if (ret) { |
| 1622 | printf("%s: pmap_convert failed\n", __func__); |
| 1623 | /* uvmspace_free calls pmap_destroy for us */ |
| 1624 | uvmspace_free(vm->vm_vmspace); |
| 1625 | vm->vm_vmspace = NULL((void *)0); |
| 1626 | return (ENOMEM12); |
| 1627 | } |
| 1628 | |
| 1629 | return (0); |
| 1630 | } |
| 1631 | |
| 1632 | /* |
| 1633 | * vm_impl_init_svm |
| 1634 | * |
| 1635 | * AMD SVM specific VM initialization routine |
| 1636 | * |
| 1637 | * Parameters: |
| 1638 | * vm: the VM being initialized |
| 1639 | * p: vmd process owning the VM |
| 1640 | * |
| 1641 | * Return values: |
| 1642 | * 0: the initialization was successful |
| 1643 | * ENOMEM: the initialization failed (lack of resources) |
| 1644 | */ |
| 1645 | int |
| 1646 | vm_impl_init_svm(struct vm *vm, struct proc *p) |
| 1647 | { |
| 1648 | int i, ret; |
| 1649 | vaddr_t mingpa, maxgpa; |
| 1650 | struct vm_mem_range *vmr; |
| 1651 | |
| 1652 | /* If not RVI, nothing to do here */ |
| 1653 | if (vmm_softc->mode != VMM_MODE_RVI) |
| 1654 | return (0); |
| 1655 | |
| 1656 | vmr = &vm->vm_memranges[0]; |
| 1657 | mingpa = vmr->vmr_gpa; |
| 1658 | vmr = &vm->vm_memranges[vm->vm_nmemranges - 1]; |
| 1659 | maxgpa = vmr->vmr_gpa + vmr->vmr_size; |
| 1660 | |
| 1661 | /* |
| 1662 | * uvmspace_alloc (currently) always returns a valid vmspace |
| 1663 | */ |
| 1664 | vm->vm_vmspace = uvmspace_alloc(mingpa, maxgpa, TRUE1, FALSE0); |
| 1665 | vm->vm_map = &vm->vm_vmspace->vm_map; |
| 1666 | |
| 1667 | /* Map the new map with an anon */ |
| 1668 | DPRINTF("%s: created vm_map @ %p\n", __func__, vm->vm_map); |
| 1669 | for (i = 0; i < vm->vm_nmemranges; i++) { |
| 1670 | vmr = &vm->vm_memranges[i]; |
| 1671 | ret = uvm_share(vm->vm_map, vmr->vmr_gpa, |
| 1672 | PROT_READ0x01 | PROT_WRITE0x02 | PROT_EXEC0x04, |
| 1673 | &p->p_vmspace->vm_map, vmr->vmr_va, vmr->vmr_size); |
| 1674 | if (ret) { |
| 1675 | printf("%s: uvm_share failed (%d)\n", __func__, ret); |
| 1676 | /* uvmspace_free calls pmap_destroy for us */ |
| 1677 | uvmspace_free(vm->vm_vmspace); |
| 1678 | vm->vm_vmspace = NULL((void *)0); |
| 1679 | return (ENOMEM12); |
| 1680 | } |
| 1681 | } |
| 1682 | |
| 1683 | /* Convert pmap to RVI */ |
| 1684 | ret = pmap_convert(vm->vm_map->pmap, PMAP_TYPE_RVI3); |
| 1685 | |
| 1686 | return (ret); |
| 1687 | } |
| 1688 | |
| 1689 | /* |
| 1690 | * vm_impl_init |
| 1691 | * |
| 1692 | * Calls the architecture-specific VM init routine |
| 1693 | * |
| 1694 | * Parameters: |
| 1695 | * vm: the VM being initialized |
| 1696 | * p: vmd process owning the VM |
| 1697 | * |
| 1698 | * Return values (from architecture-specific init routines): |
| 1699 | * 0: the initialization was successful |
| 1700 | * ENOMEM: the initialization failed (lack of resources) |
| 1701 | */ |
| 1702 | int |
| 1703 | vm_impl_init(struct vm *vm, struct proc *p) |
| 1704 | { |
| 1705 | int ret; |
| 1706 | |
| 1707 | KERNEL_LOCK()_kernel_lock(); |
| 1708 | if (vmm_softc->mode == VMM_MODE_VMX || |
| 1709 | vmm_softc->mode == VMM_MODE_EPT) |
| 1710 | ret = vm_impl_init_vmx(vm, p); |
| 1711 | else if (vmm_softc->mode == VMM_MODE_SVM || |
| 1712 | vmm_softc->mode == VMM_MODE_RVI) |
| 1713 | ret = vm_impl_init_svm(vm, p); |
| 1714 | else |
| 1715 | panic("%s: unknown vmm mode: %d", __func__, vmm_softc->mode); |
| 1716 | KERNEL_UNLOCK()_kernel_unlock(); |
| 1717 | |
| 1718 | return (ret); |
| 1719 | } |
| 1720 | |
| 1721 | /* |
| 1722 | * vm_impl_deinit_vmx |
| 1723 | * |
| 1724 | * Intel VMX specific VM deinitialization routine |
| 1725 | * |
| 1726 | * Parameters: |
| 1727 | * vm: VM to deinit |
| 1728 | */ |
| 1729 | void |
| 1730 | vm_impl_deinit_vmx(struct vm *vm) |
| 1731 | { |
| 1732 | /* Unused */ |
| 1733 | } |
| 1734 | |
| 1735 | /* |
| 1736 | * vm_impl_deinit_svm |
| 1737 | * |
| 1738 | * AMD SVM specific VM deinitialization routine |
| 1739 | * |
| 1740 | * Parameters: |
| 1741 | * vm: VM to deinit |
| 1742 | */ |
| 1743 | void |
| 1744 | vm_impl_deinit_svm(struct vm *vm) |
| 1745 | { |
| 1746 | /* Unused */ |
| 1747 | } |
| 1748 | |
| 1749 | /* |
| 1750 | * vm_impl_deinit |
| 1751 | * |
| 1752 | * Calls the architecture-specific VM init routine |
| 1753 | * |
| 1754 | * Parameters: |
| 1755 | * vm: VM to deinit |
| 1756 | */ |
| 1757 | void |
| 1758 | vm_impl_deinit(struct vm *vm) |
| 1759 | { |
| 1760 | if (vmm_softc->mode == VMM_MODE_VMX || |
| 1761 | vmm_softc->mode == VMM_MODE_EPT) |
| 1762 | vm_impl_deinit_vmx(vm); |
| 1763 | else if (vmm_softc->mode == VMM_MODE_SVM || |
| 1764 | vmm_softc->mode == VMM_MODE_RVI) |
| 1765 | vm_impl_deinit_svm(vm); |
| 1766 | else |
| 1767 | panic("%s: unknown vmm mode: %d", __func__, vmm_softc->mode); |
| 1768 | } |
| 1769 | |
| 1770 | /* |
| 1771 | * vcpu_reload_vmcs_vmx |
| 1772 | * |
| 1773 | * (Re)load the VMCS on the current cpu. Must be called with the VMCS write |
| 1774 | * lock acquired. If the VMCS is determined to be loaded on a remote cpu, an |
| 1775 | * ipi will be used to remotely flush it before loading the VMCS locally. |
| 1776 | * |
| 1777 | * Parameters: |
| 1778 | * vcpu: Pointer to the vcpu needing its VMCS |
| 1779 | * |
| 1780 | * Return values: |
| 1781 | * 0: if successful |
| 1782 | * EINVAL: an error occurred during flush or reload |
| 1783 | */ |
| 1784 | int |
| 1785 | vcpu_reload_vmcs_vmx(struct vcpu *vcpu) |
| 1786 | { |
| 1787 | struct cpu_info *ci, *last_ci; |
| 1788 | |
| 1789 | rw_assert_wrlock(&vcpu->vc_lock); |
| 1790 | |
| 1791 | ci = curcpu()({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;}); |
| 1792 | last_ci = vcpu->vc_last_pcpu; |
| 1793 | |
| 1794 | if (last_ci == NULL((void *)0)) { |
| 1795 | /* First launch */ |
| 1796 | if (vmclear(&vcpu->vc_control_pa)) |
| 1797 | return (EINVAL22); |
| 1798 | atomic_swap_uint(&vcpu->vc_vmx_vmcs_state, VMCS_CLEARED)_atomic_swap_uint((&vcpu->vc_vmx_vmcs_state), (0)); |
| 1799 | #ifdef MULTIPROCESSOR1 |
| 1800 | } else if (last_ci != ci) { |
| 1801 | /* We've moved CPUs at some point, so remote VMCLEAR */ |
| 1802 | if (vmx_remote_vmclear(last_ci, vcpu)) |
| 1803 | return (EINVAL22); |
| 1804 | KASSERT(vcpu->vc_vmx_vmcs_state == VMCS_CLEARED)((vcpu->vc_vmx_vmcs_state == 0) ? (void)0 : __assert("diagnostic " , "/usr/src/sys/arch/amd64/amd64/vmm.c", 1804, "vcpu->vc_vmx_vmcs_state == VMCS_CLEARED" )); |
| 1805 | #endif /* MULTIPROCESSOR */ |
| 1806 | } |
| 1807 | |
| 1808 | if (vmptrld(&vcpu->vc_control_pa)) { |
| 1809 | printf("%s: vmptrld\n", __func__); |
| 1810 | return (EINVAL22); |
| 1811 | } |
| 1812 | |
| 1813 | return (0); |
| 1814 | } |
| 1815 | |
| 1816 | /* |
| 1817 | * vcpu_readregs_vmx |
| 1818 | * |
| 1819 | * Reads 'vcpu's registers |
| 1820 | * |
| 1821 | * Parameters: |
| 1822 | * vcpu: the vcpu to read register values from |
| 1823 | * regmask: the types of registers to read |
| 1824 | * vrs: output parameter where register values are stored |
| 1825 | * |
| 1826 | * Return values: |
| 1827 | * 0: if successful |
| 1828 | * EINVAL: an error reading registers occurred |
| 1829 | */ |
| 1830 | int |
| 1831 | vcpu_readregs_vmx(struct vcpu *vcpu, uint64_t regmask, |
| 1832 | struct vcpu_reg_state *vrs) |
| 1833 | { |
| 1834 | int i, ret = 0; |
| 1835 | uint64_t sel, limit, ar; |
| 1836 | uint64_t *gprs = vrs->vrs_gprs; |
| 1837 | uint64_t *crs = vrs->vrs_crs; |
| 1838 | uint64_t *msrs = vrs->vrs_msrs; |
| 1839 | uint64_t *drs = vrs->vrs_drs; |
| 1840 | struct vcpu_segment_info *sregs = vrs->vrs_sregs; |
| 1841 | struct vmx_msr_store *msr_store; |
| 1842 | |
| 1843 | #ifdef VMM_DEBUG |
| 1844 | /* VMCS should be loaded... */ |
| 1845 | paddr_t pa = 0ULL; |
| 1846 | if (vmptrst(&pa)) |
| 1847 | panic("%s: vmptrst", __func__); |
| 1848 | KASSERT(pa == vcpu->vc_control_pa)((pa == vcpu->vc_control_pa) ? (void)0 : __assert("diagnostic " , "/usr/src/sys/arch/amd64/amd64/vmm.c", 1848, "pa == vcpu->vc_control_pa" )); |
| 1849 | #endif /* VMM_DEBUG */ |
| 1850 | |
| 1851 | if (regmask & VM_RWREGS_GPRS0x1) { |
| 1852 | gprs[VCPU_REGS_RAX0] = vcpu->vc_gueststate.vg_rax; |
| 1853 | gprs[VCPU_REGS_RBX1] = vcpu->vc_gueststate.vg_rbx; |
| 1854 | gprs[VCPU_REGS_RCX2] = vcpu->vc_gueststate.vg_rcx; |
| 1855 | gprs[VCPU_REGS_RDX3] = vcpu->vc_gueststate.vg_rdx; |
| 1856 | gprs[VCPU_REGS_RSI4] = vcpu->vc_gueststate.vg_rsi; |
| 1857 | gprs[VCPU_REGS_RDI5] = vcpu->vc_gueststate.vg_rdi; |
| 1858 | gprs[VCPU_REGS_R86] = vcpu->vc_gueststate.vg_r8; |
| 1859 | gprs[VCPU_REGS_R97] = vcpu->vc_gueststate.vg_r9; |
| 1860 | gprs[VCPU_REGS_R108] = vcpu->vc_gueststate.vg_r10; |
| 1861 | gprs[VCPU_REGS_R119] = vcpu->vc_gueststate.vg_r11; |
| 1862 | gprs[VCPU_REGS_R1210] = vcpu->vc_gueststate.vg_r12; |
| 1863 | gprs[VCPU_REGS_R1311] = vcpu->vc_gueststate.vg_r13; |
| 1864 | gprs[VCPU_REGS_R1412] = vcpu->vc_gueststate.vg_r14; |
| 1865 | gprs[VCPU_REGS_R1513] = vcpu->vc_gueststate.vg_r15; |
| 1866 | gprs[VCPU_REGS_RBP15] = vcpu->vc_gueststate.vg_rbp; |
| 1867 | gprs[VCPU_REGS_RIP16] = vcpu->vc_gueststate.vg_rip; |
| 1868 | if (vmread(VMCS_GUEST_IA32_RSP0x681C, &gprs[VCPU_REGS_RSP14])) |
| 1869 | goto errout; |
| 1870 | if (vmread(VMCS_GUEST_IA32_RFLAGS0x6820, &gprs[VCPU_REGS_RFLAGS17])) |
| 1871 | goto errout; |
| 1872 | } |
| 1873 | |
| 1874 | if (regmask & VM_RWREGS_SREGS0x2) { |
| 1875 | for (i = 0; i < nitems(vmm_vmx_sreg_vmcs_fields)(sizeof((vmm_vmx_sreg_vmcs_fields)) / sizeof((vmm_vmx_sreg_vmcs_fields )[0])); i++) { |
| 1876 | if (vmread(vmm_vmx_sreg_vmcs_fields[i].selid, &sel)) |
| 1877 | goto errout; |
| 1878 | if (vmread(vmm_vmx_sreg_vmcs_fields[i].limitid, &limit)) |
| 1879 | goto errout; |
| 1880 | if (vmread(vmm_vmx_sreg_vmcs_fields[i].arid, &ar)) |
| 1881 | goto errout; |
| 1882 | if (vmread(vmm_vmx_sreg_vmcs_fields[i].baseid, |
| 1883 | &sregs[i].vsi_base)) |
| 1884 | goto errout; |
| 1885 | |
| 1886 | sregs[i].vsi_sel = sel; |
| 1887 | sregs[i].vsi_limit = limit; |
| 1888 | sregs[i].vsi_ar = ar; |
| 1889 | } |
| 1890 | |
| 1891 | if (vmread(VMCS_GUEST_IA32_GDTR_LIMIT0x4810, &limit)) |
| 1892 | goto errout; |
| 1893 | if (vmread(VMCS_GUEST_IA32_GDTR_BASE0x6816, |
| 1894 | &vrs->vrs_gdtr.vsi_base)) |
| 1895 | goto errout; |
| 1896 | vrs->vrs_gdtr.vsi_limit = limit; |
| 1897 | |
| 1898 | if (vmread(VMCS_GUEST_IA32_IDTR_LIMIT0x4812, &limit)) |
| 1899 | goto errout; |
| 1900 | if (vmread(VMCS_GUEST_IA32_IDTR_BASE0x6818, |
| 1901 | &vrs->vrs_idtr.vsi_base)) |
| 1902 | goto errout; |
| 1903 | vrs->vrs_idtr.vsi_limit = limit; |
| 1904 | } |
| 1905 | |
| 1906 | if (regmask & VM_RWREGS_CRS0x4) { |
| 1907 | crs[VCPU_REGS_CR21] = vcpu->vc_gueststate.vg_cr2; |
| 1908 | crs[VCPU_REGS_XCR05] = vcpu->vc_gueststate.vg_xcr0; |
| 1909 | if (vmread(VMCS_GUEST_IA32_CR00x6800, &crs[VCPU_REGS_CR00])) |
| 1910 | goto errout; |
| 1911 | if (vmread(VMCS_GUEST_IA32_CR30x6802, &crs[VCPU_REGS_CR32])) |
| 1912 | goto errout; |
| 1913 | if (vmread(VMCS_GUEST_IA32_CR40x6804, &crs[VCPU_REGS_CR43])) |
| 1914 | goto errout; |
| 1915 | if (vmread(VMCS_GUEST_PDPTE00x280A, &crs[VCPU_REGS_PDPTE06])) |
| 1916 | goto errout; |
| 1917 | if (vmread(VMCS_GUEST_PDPTE10x280C, &crs[VCPU_REGS_PDPTE17])) |
| 1918 | goto errout; |
| 1919 | if (vmread(VMCS_GUEST_PDPTE20x280E, &crs[VCPU_REGS_PDPTE28])) |
| 1920 | goto errout; |
| 1921 | if (vmread(VMCS_GUEST_PDPTE30x2810, &crs[VCPU_REGS_PDPTE39])) |
| 1922 | goto errout; |
| 1923 | } |
| 1924 | |
| 1925 | msr_store = (struct vmx_msr_store *)vcpu->vc_vmx_msr_exit_save_va; |
| 1926 | |
| 1927 | if (regmask & VM_RWREGS_MSRS0x8) { |
| 1928 | for (i = 0; i < VCPU_REGS_NMSRS(6 + 1); i++) { |
| 1929 | msrs[i] = msr_store[i].vms_data; |
| 1930 | } |
| 1931 | } |
| 1932 | |
| 1933 | if (regmask & VM_RWREGS_DRS0x10) { |
| 1934 | drs[VCPU_REGS_DR00] = vcpu->vc_gueststate.vg_dr0; |
| 1935 | drs[VCPU_REGS_DR11] = vcpu->vc_gueststate.vg_dr1; |
| 1936 | drs[VCPU_REGS_DR22] = vcpu->vc_gueststate.vg_dr2; |
| 1937 | drs[VCPU_REGS_DR33] = vcpu->vc_gueststate.vg_dr3; |
| 1938 | drs[VCPU_REGS_DR64] = vcpu->vc_gueststate.vg_dr6; |
| 1939 | if (vmread(VMCS_GUEST_IA32_DR70x681A, &drs[VCPU_REGS_DR75])) |
| 1940 | goto errout; |
| 1941 | } |
| 1942 | |
| 1943 | goto out; |
| 1944 | |
| 1945 | errout: |
| 1946 | ret = EINVAL22; |
| 1947 | out: |
| 1948 | return (ret); |
| 1949 | } |
| 1950 | |
| 1951 | /* |
| 1952 | * vcpu_readregs_svm |
| 1953 | * |
| 1954 | * Reads 'vcpu's registers |
| 1955 | * |
| 1956 | * Parameters: |
| 1957 | * vcpu: the vcpu to read register values from |
| 1958 | * regmask: the types of registers to read |
| 1959 | * vrs: output parameter where register values are stored |
| 1960 | * |
| 1961 | * Return values: |
| 1962 | * 0: if successful |
| 1963 | */ |
| 1964 | int |
| 1965 | vcpu_readregs_svm(struct vcpu *vcpu, uint64_t regmask, |
| 1966 | struct vcpu_reg_state *vrs) |
| 1967 | { |
| 1968 | uint64_t *gprs = vrs->vrs_gprs; |
| 1969 | uint64_t *crs = vrs->vrs_crs; |
| 1970 | uint64_t *msrs = vrs->vrs_msrs; |
| 1971 | uint64_t *drs = vrs->vrs_drs; |
| 1972 | uint32_t attr; |
| 1973 | struct vcpu_segment_info *sregs = vrs->vrs_sregs; |
| 1974 | struct vmcb *vmcb = (struct vmcb *)vcpu->vc_control_va; |
| 1975 | |
| 1976 | if (regmask & VM_RWREGS_GPRS0x1) { |
| 1977 | gprs[VCPU_REGS_RAX0] = vcpu->vc_gueststate.vg_rax; |
| 1978 | gprs[VCPU_REGS_RBX1] = vcpu->vc_gueststate.vg_rbx; |
| 1979 | gprs[VCPU_REGS_RCX2] = vcpu->vc_gueststate.vg_rcx; |
| 1980 | gprs[VCPU_REGS_RDX3] = vcpu->vc_gueststate.vg_rdx; |
| 1981 | gprs[VCPU_REGS_RSI4] = vcpu->vc_gueststate.vg_rsi; |
| 1982 | gprs[VCPU_REGS_RDI5] = vcpu->vc_gueststate.vg_rdi; |
| 1983 | gprs[VCPU_REGS_R86] = vcpu->vc_gueststate.vg_r8; |
| 1984 | gprs[VCPU_REGS_R97] = vcpu->vc_gueststate.vg_r9; |
| 1985 | gprs[VCPU_REGS_R108] = vcpu->vc_gueststate.vg_r10; |
| 1986 | gprs[VCPU_REGS_R119] = vcpu->vc_gueststate.vg_r11; |
| 1987 | gprs[VCPU_REGS_R1210] = vcpu->vc_gueststate.vg_r12; |
| 1988 | gprs[VCPU_REGS_R1311] = vcpu->vc_gueststate.vg_r13; |
| 1989 | gprs[VCPU_REGS_R1412] = vcpu->vc_gueststate.vg_r14; |
| 1990 | gprs[VCPU_REGS_R1513] = vcpu->vc_gueststate.vg_r15; |
| 1991 | gprs[VCPU_REGS_RBP15] = vcpu->vc_gueststate.vg_rbp; |
| 1992 | gprs[VCPU_REGS_RIP16] = vmcb->v_rip; |
| 1993 | gprs[VCPU_REGS_RSP14] = vmcb->v_rsp; |
| 1994 | gprs[VCPU_REGS_RFLAGS17] = vmcb->v_rflags; |
| 1995 | } |
| 1996 | |
| 1997 | if (regmask & VM_RWREGS_SREGS0x2) { |
| 1998 | sregs[VCPU_REGS_CS0].vsi_sel = vmcb->v_cs.vs_sel; |
| 1999 | sregs[VCPU_REGS_CS0].vsi_limit = vmcb->v_cs.vs_lim; |
| 2000 | attr = vmcb->v_cs.vs_attr; |
| 2001 | sregs[VCPU_REGS_CS0].vsi_ar = (attr & 0xff) | ((attr << 4) & |
| 2002 | 0xf000); |
| 2003 | sregs[VCPU_REGS_CS0].vsi_base = vmcb->v_cs.vs_base; |
| 2004 | |
| 2005 | sregs[VCPU_REGS_DS1].vsi_sel = vmcb->v_ds.vs_sel; |
| 2006 | sregs[VCPU_REGS_DS1].vsi_limit = vmcb->v_ds.vs_lim; |
| 2007 | attr = vmcb->v_ds.vs_attr; |
| 2008 | sregs[VCPU_REGS_DS1].vsi_ar = (attr & 0xff) | ((attr << 4) & |
| 2009 | 0xf000); |
| 2010 | sregs[VCPU_REGS_DS1].vsi_base = vmcb->v_ds.vs_base; |
| 2011 | |
| 2012 | sregs[VCPU_REGS_ES2].vsi_sel = vmcb->v_es.vs_sel; |
| 2013 | sregs[VCPU_REGS_ES2].vsi_limit = vmcb->v_es.vs_lim; |
| 2014 | attr = vmcb->v_es.vs_attr; |
| 2015 | sregs[VCPU_REGS_ES2].vsi_ar = (attr & 0xff) | ((attr << 4) & |
| 2016 | 0xf000); |
| 2017 | sregs[VCPU_REGS_ES2].vsi_base = vmcb->v_es.vs_base; |
| 2018 | |
| 2019 | sregs[VCPU_REGS_FS3].vsi_sel = vmcb->v_fs.vs_sel; |
| 2020 | sregs[VCPU_REGS_FS3].vsi_limit = vmcb->v_fs.vs_lim; |
| 2021 | attr = vmcb->v_fs.vs_attr; |
| 2022 | sregs[VCPU_REGS_FS3].vsi_ar = (attr & 0xff) | ((attr << 4) & |
| 2023 | 0xf000); |
| 2024 | sregs[VCPU_REGS_FS3].vsi_base = vmcb->v_fs.vs_base; |
| 2025 | |
| 2026 | sregs[VCPU_REGS_GS4].vsi_sel = vmcb->v_gs.vs_sel; |
| 2027 | sregs[VCPU_REGS_GS4].vsi_limit = vmcb->v_gs.vs_lim; |
| 2028 | attr = vmcb->v_gs.vs_attr; |
| 2029 | sregs[VCPU_REGS_GS4].vsi_ar = (attr & 0xff) | ((attr << 4) & |
| 2030 | 0xf000); |
| 2031 | sregs[VCPU_REGS_GS4].vsi_base = vmcb->v_gs.vs_base; |
| 2032 | |
| 2033 | sregs[VCPU_REGS_SS5].vsi_sel = vmcb->v_ss.vs_sel; |
| 2034 | sregs[VCPU_REGS_SS5].vsi_limit = vmcb->v_ss.vs_lim; |
| 2035 | attr = vmcb->v_ss.vs_attr; |
| 2036 | sregs[VCPU_REGS_SS5].vsi_ar = (attr & 0xff) | ((attr << 4) & |
| 2037 | 0xf000); |
| 2038 | sregs[VCPU_REGS_SS5].vsi_base = vmcb->v_ss.vs_base; |
| 2039 | |
| 2040 | sregs[VCPU_REGS_LDTR6].vsi_sel = vmcb->v_ldtr.vs_sel; |
| 2041 | sregs[VCPU_REGS_LDTR6].vsi_limit = vmcb->v_ldtr.vs_lim; |
| 2042 | attr = vmcb->v_ldtr.vs_attr; |
| 2043 | sregs[VCPU_REGS_LDTR6].vsi_ar = (attr & 0xff) | ((attr << 4) |
| 2044 | & 0xf000); |
| 2045 | sregs[VCPU_REGS_LDTR6].vsi_base = vmcb->v_ldtr.vs_base; |
| 2046 | |
| 2047 | sregs[VCPU_REGS_TR7].vsi_sel = vmcb->v_tr.vs_sel; |
| 2048 | sregs[VCPU_REGS_TR7].vsi_limit = vmcb->v_tr.vs_lim; |
| 2049 | attr = vmcb->v_tr.vs_attr; |
| 2050 | sregs[VCPU_REGS_TR7].vsi_ar = (attr & 0xff) | ((attr << 4) & |
| 2051 | 0xf000); |
| 2052 | sregs[VCPU_REGS_TR7].vsi_base = vmcb->v_tr.vs_base; |
| 2053 | |
| 2054 | vrs->vrs_gdtr.vsi_limit = vmcb->v_gdtr.vs_lim; |
| 2055 | vrs->vrs_gdtr.vsi_base = vmcb->v_gdtr.vs_base; |
| 2056 | vrs->vrs_idtr.vsi_limit = vmcb->v_idtr.vs_lim; |
| 2057 | vrs->vrs_idtr.vsi_base = vmcb->v_idtr.vs_base; |
| 2058 | } |
| 2059 | |
| 2060 | if (regmask & VM_RWREGS_CRS0x4) { |
| 2061 | crs[VCPU_REGS_CR00] = vmcb->v_cr0; |
| 2062 | crs[VCPU_REGS_CR32] = vmcb->v_cr3; |
| 2063 | crs[VCPU_REGS_CR43] = vmcb->v_cr4; |
| 2064 | crs[VCPU_REGS_CR21] = vcpu->vc_gueststate.vg_cr2; |
| 2065 | crs[VCPU_REGS_XCR05] = vcpu->vc_gueststate.vg_xcr0; |
| 2066 | } |
| 2067 | |
| 2068 | if (regmask & VM_RWREGS_MSRS0x8) { |
| 2069 | msrs[VCPU_REGS_EFER0] = vmcb->v_efer; |
| 2070 | msrs[VCPU_REGS_STAR1] = vmcb->v_star; |
| 2071 | msrs[VCPU_REGS_LSTAR2] = vmcb->v_lstar; |
| 2072 | msrs[VCPU_REGS_CSTAR3] = vmcb->v_cstar; |
| 2073 | msrs[VCPU_REGS_SFMASK4] = vmcb->v_sfmask; |
| 2074 | msrs[VCPU_REGS_KGSBASE5] = vmcb->v_kgsbase; |
| 2075 | } |
| 2076 | |
| 2077 | if (regmask & VM_RWREGS_DRS0x10) { |
| 2078 | drs[VCPU_REGS_DR00] = vcpu->vc_gueststate.vg_dr0; |
| 2079 | drs[VCPU_REGS_DR11] = vcpu->vc_gueststate.vg_dr1; |
| 2080 | drs[VCPU_REGS_DR22] = vcpu->vc_gueststate.vg_dr2; |
| 2081 | drs[VCPU_REGS_DR33] = vcpu->vc_gueststate.vg_dr3; |
| 2082 | drs[VCPU_REGS_DR64] = vmcb->v_dr6; |
| 2083 | drs[VCPU_REGS_DR75] = vmcb->v_dr7; |
| 2084 | } |
| 2085 | |
| 2086 | return (0); |
| 2087 | } |
| 2088 | |
| 2089 | /* |
| 2090 | * vcpu_writeregs_vmx |
| 2091 | * |
| 2092 | * Writes VCPU registers |
| 2093 | * |
| 2094 | * Parameters: |
| 2095 | * vcpu: the vcpu that has to get its registers written to |
| 2096 | * regmask: the types of registers to write |
| 2097 | * loadvmcs: bit to indicate whether the VMCS has to be loaded first |
| 2098 | * vrs: the register values to write |
| 2099 | * |
| 2100 | * Return values: |
| 2101 | * 0: if successful |
| 2102 | * EINVAL an error writing registers occurred |
| 2103 | */ |
| 2104 | int |
| 2105 | vcpu_writeregs_vmx(struct vcpu *vcpu, uint64_t regmask, int loadvmcs, |
| 2106 | struct vcpu_reg_state *vrs) |
| 2107 | { |
| 2108 | int i, ret = 0; |
| 2109 | uint16_t sel; |
| 2110 | uint64_t limit, ar; |
| 2111 | uint64_t *gprs = vrs->vrs_gprs; |
| 2112 | uint64_t *crs = vrs->vrs_crs; |
| 2113 | uint64_t *msrs = vrs->vrs_msrs; |
| 2114 | uint64_t *drs = vrs->vrs_drs; |
| 2115 | struct vcpu_segment_info *sregs = vrs->vrs_sregs; |
| 2116 | struct vmx_msr_store *msr_store; |
| 2117 | |
| 2118 | if (loadvmcs) { |
| 2119 | if (vcpu_reload_vmcs_vmx(vcpu)) |
| 2120 | return (EINVAL22); |
| 2121 | } |
| 2122 | |
| 2123 | #ifdef VMM_DEBUG |
| 2124 | /* VMCS should be loaded... */ |
| 2125 | paddr_t pa = 0ULL; |
| 2126 | if (vmptrst(&pa)) |
| 2127 | panic("%s: vmptrst", __func__); |
| 2128 | KASSERT(pa == vcpu->vc_control_pa)((pa == vcpu->vc_control_pa) ? (void)0 : __assert("diagnostic " , "/usr/src/sys/arch/amd64/amd64/vmm.c", 2128, "pa == vcpu->vc_control_pa" )); |
| 2129 | #endif /* VMM_DEBUG */ |
| 2130 | |
| 2131 | if (regmask & VM_RWREGS_GPRS0x1) { |
| 2132 | vcpu->vc_gueststate.vg_rax = gprs[VCPU_REGS_RAX0]; |
| 2133 | vcpu->vc_gueststate.vg_rbx = gprs[VCPU_REGS_RBX1]; |
| 2134 | vcpu->vc_gueststate.vg_rcx = gprs[VCPU_REGS_RCX2]; |
| 2135 | vcpu->vc_gueststate.vg_rdx = gprs[VCPU_REGS_RDX3]; |
| 2136 | vcpu->vc_gueststate.vg_rsi = gprs[VCPU_REGS_RSI4]; |
| 2137 | vcpu->vc_gueststate.vg_rdi = gprs[VCPU_REGS_RDI5]; |
| 2138 | vcpu->vc_gueststate.vg_r8 = gprs[VCPU_REGS_R86]; |
| 2139 | vcpu->vc_gueststate.vg_r9 = gprs[VCPU_REGS_R97]; |
| 2140 | vcpu->vc_gueststate.vg_r10 = gprs[VCPU_REGS_R108]; |
| 2141 | vcpu->vc_gueststate.vg_r11 = gprs[VCPU_REGS_R119]; |
| 2142 | vcpu->vc_gueststate.vg_r12 = gprs[VCPU_REGS_R1210]; |
| 2143 | vcpu->vc_gueststate.vg_r13 = gprs[VCPU_REGS_R1311]; |
| 2144 | vcpu->vc_gueststate.vg_r14 = gprs[VCPU_REGS_R1412]; |
| 2145 | vcpu->vc_gueststate.vg_r15 = gprs[VCPU_REGS_R1513]; |
| 2146 | vcpu->vc_gueststate.vg_rbp = gprs[VCPU_REGS_RBP15]; |
| 2147 | vcpu->vc_gueststate.vg_rip = gprs[VCPU_REGS_RIP16]; |
| 2148 | if (vmwrite(VMCS_GUEST_IA32_RIP0x681E, gprs[VCPU_REGS_RIP16])) |
| 2149 | goto errout; |
| 2150 | if (vmwrite(VMCS_GUEST_IA32_RSP0x681C, gprs[VCPU_REGS_RSP14])) |
| 2151 | goto errout; |
| 2152 | if (vmwrite(VMCS_GUEST_IA32_RFLAGS0x6820, gprs[VCPU_REGS_RFLAGS17])) |
| 2153 | goto errout; |
| 2154 | } |
| 2155 | |
| 2156 | if (regmask & VM_RWREGS_SREGS0x2) { |
| 2157 | for (i = 0; i < nitems(vmm_vmx_sreg_vmcs_fields)(sizeof((vmm_vmx_sreg_vmcs_fields)) / sizeof((vmm_vmx_sreg_vmcs_fields )[0])); i++) { |
| 2158 | sel = sregs[i].vsi_sel; |
| 2159 | limit = sregs[i].vsi_limit; |
| 2160 | ar = sregs[i].vsi_ar; |
| 2161 | |
| 2162 | if (vmwrite(vmm_vmx_sreg_vmcs_fields[i].selid, sel)) |
| 2163 | goto errout; |
| 2164 | if (vmwrite(vmm_vmx_sreg_vmcs_fields[i].limitid, limit)) |
| 2165 | goto errout; |
| 2166 | if (vmwrite(vmm_vmx_sreg_vmcs_fields[i].arid, ar)) |
| 2167 | goto errout; |
| 2168 | if (vmwrite(vmm_vmx_sreg_vmcs_fields[i].baseid, |
| 2169 | sregs[i].vsi_base)) |
| 2170 | goto errout; |
| 2171 | } |
| 2172 | |
| 2173 | if (vmwrite(VMCS_GUEST_IA32_GDTR_LIMIT0x4810, |
| 2174 | vrs->vrs_gdtr.vsi_limit)) |
| 2175 | goto errout; |
| 2176 | if (vmwrite(VMCS_GUEST_IA32_GDTR_BASE0x6816, |
| 2177 | vrs->vrs_gdtr.vsi_base)) |
| 2178 | goto errout; |
| 2179 | if (vmwrite(VMCS_GUEST_IA32_IDTR_LIMIT0x4812, |
| 2180 | vrs->vrs_idtr.vsi_limit)) |
| 2181 | goto errout; |
| 2182 | if (vmwrite(VMCS_GUEST_IA32_IDTR_BASE0x6818, |
| 2183 | vrs->vrs_idtr.vsi_base)) |
| 2184 | goto errout; |
| 2185 | } |
| 2186 | |
| 2187 | if (regmask & VM_RWREGS_CRS0x4) { |
| 2188 | vcpu->vc_gueststate.vg_xcr0 = crs[VCPU_REGS_XCR05]; |
| 2189 | if (vmwrite(VMCS_GUEST_IA32_CR00x6800, crs[VCPU_REGS_CR00])) |
| 2190 | goto errout; |
| 2191 | if (vmwrite(VMCS_GUEST_IA32_CR30x6802, crs[VCPU_REGS_CR32])) |
| 2192 | goto errout; |
| 2193 | if (vmwrite(VMCS_GUEST_IA32_CR40x6804, crs[VCPU_REGS_CR43])) |
| 2194 | goto errout; |
| 2195 | if (vmwrite(VMCS_GUEST_PDPTE00x280A, crs[VCPU_REGS_PDPTE06])) |
| 2196 | goto errout; |
| 2197 | if (vmwrite(VMCS_GUEST_PDPTE10x280C, crs[VCPU_REGS_PDPTE17])) |
| 2198 | goto errout; |
| 2199 | if (vmwrite(VMCS_GUEST_PDPTE20x280E, crs[VCPU_REGS_PDPTE28])) |
| 2200 | goto errout; |
| 2201 | if (vmwrite(VMCS_GUEST_PDPTE30x2810, crs[VCPU_REGS_PDPTE39])) |
| 2202 | goto errout; |
| 2203 | } |
| 2204 | |
| 2205 | msr_store = (struct vmx_msr_store *)vcpu->vc_vmx_msr_exit_save_va; |
| 2206 | |
| 2207 | if (regmask & VM_RWREGS_MSRS0x8) { |
| 2208 | for (i = 0; i < VCPU_REGS_NMSRS(6 + 1); i++) { |
| 2209 | msr_store[i].vms_data = msrs[i]; |
| 2210 | } |
| 2211 | } |
| 2212 | |
| 2213 | if (regmask & VM_RWREGS_DRS0x10) { |
| 2214 | vcpu->vc_gueststate.vg_dr0 = drs[VCPU_REGS_DR00]; |
| 2215 | vcpu->vc_gueststate.vg_dr1 = drs[VCPU_REGS_DR11]; |
| 2216 | vcpu->vc_gueststate.vg_dr2 = drs[VCPU_REGS_DR22]; |
| 2217 | vcpu->vc_gueststate.vg_dr3 = drs[VCPU_REGS_DR33]; |
| 2218 | vcpu->vc_gueststate.vg_dr6 = drs[VCPU_REGS_DR64]; |
| 2219 | if (vmwrite(VMCS_GUEST_IA32_DR70x681A, drs[VCPU_REGS_DR75])) |
| 2220 | goto errout; |
| 2221 | } |
| 2222 | |
| 2223 | goto out; |
| 2224 | |
| 2225 | errout: |
| 2226 | ret = EINVAL22; |
| 2227 | out: |
| 2228 | if (loadvmcs) { |
| 2229 | if (vmclear(&vcpu->vc_control_pa)) |
| 2230 | ret = EINVAL22; |
| 2231 | } |
| 2232 | return (ret); |
| 2233 | } |
| 2234 | |
| 2235 | /* |
| 2236 | * vcpu_writeregs_svm |
| 2237 | * |
| 2238 | * Writes 'vcpu's registers |
| 2239 | * |
| 2240 | * Parameters: |
| 2241 | * vcpu: the vcpu that has to get its registers written to |
| 2242 | * regmask: the types of registers to write |
| 2243 | * vrs: the register values to write |
| 2244 | * |
| 2245 | * Return values: |
| 2246 | * 0: if successful |
| 2247 | * EINVAL an error writing registers occurred |
| 2248 | */ |
| 2249 | int |
| 2250 | vcpu_writeregs_svm(struct vcpu *vcpu, uint64_t regmask, |
| 2251 | struct vcpu_reg_state *vrs) |
| 2252 | { |
| 2253 | uint64_t *gprs = vrs->vrs_gprs; |
| 2254 | uint64_t *crs = vrs->vrs_crs; |
| 2255 | uint16_t attr; |
| 2256 | uint64_t *msrs = vrs->vrs_msrs; |
| 2257 | uint64_t *drs = vrs->vrs_drs; |
| 2258 | struct vcpu_segment_info *sregs = vrs->vrs_sregs; |
| 2259 | struct vmcb *vmcb = (struct vmcb *)vcpu->vc_control_va; |
| 2260 | |
| 2261 | if (regmask & VM_RWREGS_GPRS0x1) { |
| 2262 | vcpu->vc_gueststate.vg_rax = gprs[VCPU_REGS_RAX0]; |
| 2263 | vcpu->vc_gueststate.vg_rbx = gprs[VCPU_REGS_RBX1]; |
| 2264 | vcpu->vc_gueststate.vg_rcx = gprs[VCPU_REGS_RCX2]; |
| 2265 | vcpu->vc_gueststate.vg_rdx = gprs[VCPU_REGS_RDX3]; |
| 2266 | vcpu->vc_gueststate.vg_rsi = gprs[VCPU_REGS_RSI4]; |
| 2267 | vcpu->vc_gueststate.vg_rdi = gprs[VCPU_REGS_RDI5]; |
| 2268 | vcpu->vc_gueststate.vg_r8 = gprs[VCPU_REGS_R86]; |
| 2269 | vcpu->vc_gueststate.vg_r9 = gprs[VCPU_REGS_R97]; |
| 2270 | vcpu->vc_gueststate.vg_r10 = gprs[VCPU_REGS_R108]; |
| 2271 | vcpu->vc_gueststate.vg_r11 = gprs[VCPU_REGS_R119]; |
| 2272 | vcpu->vc_gueststate.vg_r12 = gprs[VCPU_REGS_R1210]; |
| 2273 | vcpu->vc_gueststate.vg_r13 = gprs[VCPU_REGS_R1311]; |
| 2274 | vcpu->vc_gueststate.vg_r14 = gprs[VCPU_REGS_R1412]; |
| 2275 | vcpu->vc_gueststate.vg_r15 = gprs[VCPU_REGS_R1513]; |
| 2276 | vcpu->vc_gueststate.vg_rbp = gprs[VCPU_REGS_RBP15]; |
| 2277 | vcpu->vc_gueststate.vg_rip = gprs[VCPU_REGS_RIP16]; |
| 2278 | |
| 2279 | vmcb->v_rip = gprs[VCPU_REGS_RIP16]; |
| 2280 | vmcb->v_rsp = gprs[VCPU_REGS_RSP14]; |
| 2281 | vmcb->v_rflags = gprs[VCPU_REGS_RFLAGS17]; |
| 2282 | } |
| 2283 | |
| 2284 | if (regmask & VM_RWREGS_SREGS0x2) { |
| 2285 | vmcb->v_cs.vs_sel = sregs[VCPU_REGS_CS0].vsi_sel; |
| 2286 | vmcb->v_cs.vs_lim = sregs[VCPU_REGS_CS0].vsi_limit; |
| 2287 | attr = sregs[VCPU_REGS_CS0].vsi_ar; |
| 2288 | vmcb->v_cs.vs_attr = (attr & 0xff) | ((attr >> 4) & 0xf00); |
| 2289 | vmcb->v_cs.vs_base = sregs[VCPU_REGS_CS0].vsi_base; |
| 2290 | vmcb->v_ds.vs_sel = sregs[VCPU_REGS_DS1].vsi_sel; |
| 2291 | vmcb->v_ds.vs_lim = sregs[VCPU_REGS_DS1].vsi_limit; |
| 2292 | attr = sregs[VCPU_REGS_DS1].vsi_ar; |
| 2293 | vmcb->v_ds.vs_attr = (attr & 0xff) | ((attr >> 4) & 0xf00); |
| 2294 | vmcb->v_ds.vs_base = sregs[VCPU_REGS_DS1].vsi_base; |
| 2295 | vmcb->v_es.vs_sel = sregs[VCPU_REGS_ES2].vsi_sel; |
| 2296 | vmcb->v_es.vs_lim = sregs[VCPU_REGS_ES2].vsi_limit; |
| 2297 | attr = sregs[VCPU_REGS_ES2].vsi_ar; |
| 2298 | vmcb->v_es.vs_attr = (attr & 0xff) | ((attr >> 4) & 0xf00); |
| 2299 | vmcb->v_es.vs_base = sregs[VCPU_REGS_ES2].vsi_base; |
| 2300 | vmcb->v_fs.vs_sel = sregs[VCPU_REGS_FS3].vsi_sel; |
| 2301 | vmcb->v_fs.vs_lim = sregs[VCPU_REGS_FS3].vsi_limit; |
| 2302 | attr = sregs[VCPU_REGS_FS3].vsi_ar; |
| 2303 | vmcb->v_fs.vs_attr = (attr & 0xff) | ((attr >> 4) & 0xf00); |
| 2304 | vmcb->v_fs.vs_base = sregs[VCPU_REGS_FS3].vsi_base; |
| 2305 | vmcb->v_gs.vs_sel = sregs[VCPU_REGS_GS4].vsi_sel; |
| 2306 | vmcb->v_gs.vs_lim = sregs[VCPU_REGS_GS4].vsi_limit; |
| 2307 | attr = sregs[VCPU_REGS_GS4].vsi_ar; |
| 2308 | vmcb->v_gs.vs_attr = (attr & 0xff) | ((attr >> 4) & 0xf00); |
| 2309 | vmcb->v_gs.vs_base = sregs[VCPU_REGS_GS4].vsi_base; |
| 2310 | vmcb->v_ss.vs_sel = sregs[VCPU_REGS_SS5].vsi_sel; |
| 2311 | vmcb->v_ss.vs_lim = sregs[VCPU_REGS_SS5].vsi_limit; |
| 2312 | attr = sregs[VCPU_REGS_SS5].vsi_ar; |
| 2313 | vmcb->v_ss.vs_attr = (attr & 0xff) | ((attr >> 4) & 0xf00); |
| 2314 | vmcb->v_ss.vs_base = sregs[VCPU_REGS_SS5].vsi_base; |
| 2315 | vmcb->v_ldtr.vs_sel = sregs[VCPU_REGS_LDTR6].vsi_sel; |
| 2316 | vmcb->v_ldtr.vs_lim = sregs[VCPU_REGS_LDTR6].vsi_limit; |
| 2317 | attr = sregs[VCPU_REGS_LDTR6].vsi_ar; |
| 2318 | vmcb->v_ldtr.vs_attr = (attr & 0xff) | ((attr >> 4) & 0xf00); |
| 2319 | vmcb->v_ldtr.vs_base = sregs[VCPU_REGS_LDTR6].vsi_base; |
| 2320 | vmcb->v_tr.vs_sel = sregs[VCPU_REGS_TR7].vsi_sel; |
| 2321 | vmcb->v_tr.vs_lim = sregs[VCPU_REGS_TR7].vsi_limit; |
| 2322 | attr = sregs[VCPU_REGS_TR7].vsi_ar; |
| 2323 | vmcb->v_tr.vs_attr = (attr & 0xff) | ((attr >> 4) & 0xf00); |
| 2324 | vmcb->v_tr.vs_base = sregs[VCPU_REGS_TR7].vsi_base; |
| 2325 | vmcb->v_gdtr.vs_lim = vrs->vrs_gdtr.vsi_limit; |
| 2326 | vmcb->v_gdtr.vs_base = vrs->vrs_gdtr.vsi_base; |
| 2327 | vmcb->v_idtr.vs_lim = vrs->vrs_idtr.vsi_limit; |
| 2328 | vmcb->v_idtr.vs_base = vrs->vrs_idtr.vsi_base; |
| 2329 | } |
| 2330 | |
| 2331 | if (regmask & VM_RWREGS_CRS0x4) { |
| 2332 | vmcb->v_cr0 = crs[VCPU_REGS_CR00]; |
| 2333 | vmcb->v_cr3 = crs[VCPU_REGS_CR32]; |
| 2334 | vmcb->v_cr4 = crs[VCPU_REGS_CR43]; |
| 2335 | vcpu->vc_gueststate.vg_cr2 = crs[VCPU_REGS_CR21]; |
| 2336 | vcpu->vc_gueststate.vg_xcr0 = crs[VCPU_REGS_XCR05]; |
| 2337 | } |
| 2338 | |
| 2339 | if (regmask & VM_RWREGS_MSRS0x8) { |
| 2340 | vmcb->v_efer |= msrs[VCPU_REGS_EFER0]; |
| 2341 | vmcb->v_star = msrs[VCPU_REGS_STAR1]; |
| 2342 | vmcb->v_lstar = msrs[VCPU_REGS_LSTAR2]; |
| 2343 | vmcb->v_cstar = msrs[VCPU_REGS_CSTAR3]; |
| 2344 | vmcb->v_sfmask = msrs[VCPU_REGS_SFMASK4]; |
| 2345 | vmcb->v_kgsbase = msrs[VCPU_REGS_KGSBASE5]; |
| 2346 | } |
| 2347 | |
| 2348 | if (regmask & VM_RWREGS_DRS0x10) { |
| 2349 | vcpu->vc_gueststate.vg_dr0 = drs[VCPU_REGS_DR00]; |
| 2350 | vcpu->vc_gueststate.vg_dr1 = drs[VCPU_REGS_DR11]; |
| 2351 | vcpu->vc_gueststate.vg_dr2 = drs[VCPU_REGS_DR22]; |
| 2352 | vcpu->vc_gueststate.vg_dr3 = drs[VCPU_REGS_DR33]; |
| 2353 | vmcb->v_dr6 = drs[VCPU_REGS_DR64]; |
| 2354 | vmcb->v_dr7 = drs[VCPU_REGS_DR75]; |
| 2355 | } |
| 2356 | |
| 2357 | return (0); |
| 2358 | } |
| 2359 | |
| 2360 | /* |
| 2361 | * vcpu_reset_regs_svm |
| 2362 | * |
| 2363 | * Initializes 'vcpu's registers to supplied state |
| 2364 | * |
| 2365 | * Parameters: |
| 2366 | * vcpu: the vcpu whose register state is to be initialized |
| 2367 | * vrs: the register state to set |
| 2368 | * |
| 2369 | * Return values: |
| 2370 | * 0: registers init'ed successfully |
| 2371 | * EINVAL: an error occurred setting register state |
| 2372 | */ |
| 2373 | int |
| 2374 | vcpu_reset_regs_svm(struct vcpu *vcpu, struct vcpu_reg_state *vrs) |
| 2375 | { |
| 2376 | struct vmcb *vmcb; |
| 2377 | int ret; |
| 2378 | uint16_t asid; |
| 2379 | |
| 2380 | vmcb = (struct vmcb *)vcpu->vc_control_va; |
| 2381 | |
| 2382 | /* |
| 2383 | * Intercept controls |
| 2384 | * |
| 2385 | * External Interrupt exiting (SVM_INTERCEPT_INTR) |
| 2386 | * External NMI exiting (SVM_INTERCEPT_NMI) |
| 2387 | * CPUID instruction (SVM_INTERCEPT_CPUID) |
| 2388 | * HLT instruction (SVM_INTERCEPT_HLT) |
| 2389 | * I/O instructions (SVM_INTERCEPT_INOUT) |
| 2390 | * MSR access (SVM_INTERCEPT_MSR) |
| 2391 | * shutdown events (SVM_INTERCEPT_SHUTDOWN) |
| 2392 | * |
| 2393 | * VMRUN instruction (SVM_INTERCEPT_VMRUN) |
| 2394 | * VMMCALL instruction (SVM_INTERCEPT_VMMCALL) |
| 2395 | * VMLOAD instruction (SVM_INTERCEPT_VMLOAD) |
| 2396 | * VMSAVE instruction (SVM_INTERCEPT_VMSAVE) |
| 2397 | * STGI instruction (SVM_INTERCEPT_STGI) |
| 2398 | * CLGI instruction (SVM_INTERCEPT_CLGI) |
| 2399 | * SKINIT instruction (SVM_INTERCEPT_SKINIT) |
| 2400 | * ICEBP instruction (SVM_INTERCEPT_ICEBP) |
| 2401 | * MWAIT instruction (SVM_INTERCEPT_MWAIT_UNCOND) |
| 2402 | * MWAIT instruction (SVM_INTERCEPT_MWAIT_COND) |
| 2403 | * MONITOR instruction (SVM_INTERCEPT_MONITOR) |
| 2404 | * RDTSCP instruction (SVM_INTERCEPT_RDTSCP) |
| 2405 | * INVLPGA instruction (SVM_INTERCEPT_INVLPGA) |
| 2406 | * XSETBV instruction (SVM_INTERCEPT_XSETBV) (if available) |
| 2407 | */ |
| 2408 | vmcb->v_intercept1 = SVM_INTERCEPT_INTR(1UL << 0) | SVM_INTERCEPT_NMI(1UL << 1) | |
| 2409 | SVM_INTERCEPT_CPUID(1UL << 18) | SVM_INTERCEPT_HLT(1UL << 24) | SVM_INTERCEPT_INOUT(1UL << 27) | |
| 2410 | SVM_INTERCEPT_MSR(1UL << 28) | SVM_INTERCEPT_SHUTDOWN(1UL << 31); |
| 2411 | |
| 2412 | vmcb->v_intercept2 = SVM_INTERCEPT_VMRUN(1UL << 0) | SVM_INTERCEPT_VMMCALL(1UL << 1) | |
| 2413 | SVM_INTERCEPT_VMLOAD(1UL << 2) | SVM_INTERCEPT_VMSAVE(1UL << 3) | SVM_INTERCEPT_STGI(1UL << 4) | |
| 2414 | SVM_INTERCEPT_CLGI(1UL << 5) | SVM_INTERCEPT_SKINIT(1UL << 6) | SVM_INTERCEPT_ICEBP(1UL << 8) | |
| 2415 | SVM_INTERCEPT_MWAIT_UNCOND(1UL << 11) | SVM_INTERCEPT_MONITOR(1UL << 10) | |
| 2416 | SVM_INTERCEPT_MWAIT_COND(1UL << 12) | SVM_INTERCEPT_RDTSCP(1UL << 7) | |
| 2417 | SVM_INTERCEPT_INVLPGA(1UL << 26); |
| 2418 | |
| 2419 | if (xsave_mask) |
| 2420 | vmcb->v_intercept2 |= SVM_INTERCEPT_XSETBV(1UL << 13); |
| 2421 | |
| 2422 | /* Setup I/O bitmap */ |
| 2423 | memset((uint8_t *)vcpu->vc_svm_ioio_va, 0xFF, 3 * PAGE_SIZE)__builtin_memset(((uint8_t *)vcpu->vc_svm_ioio_va), (0xFF) , (3 * (1 << 12))); |
| 2424 | vmcb->v_iopm_pa = (uint64_t)(vcpu->vc_svm_ioio_pa); |
| 2425 | |
| 2426 | /* Setup MSR bitmap */ |
| 2427 | memset((uint8_t *)vcpu->vc_msr_bitmap_va, 0xFF, 2 * PAGE_SIZE)__builtin_memset(((uint8_t *)vcpu->vc_msr_bitmap_va), (0xFF ), (2 * (1 << 12))); |
| 2428 | vmcb->v_msrpm_pa = (uint64_t)(vcpu->vc_msr_bitmap_pa); |
| 2429 | svm_setmsrbrw(vcpu, MSR_IA32_FEATURE_CONTROL0x03a); |
| 2430 | svm_setmsrbrw(vcpu, MSR_SYSENTER_CS0x174); |
| 2431 | svm_setmsrbrw(vcpu, MSR_SYSENTER_ESP0x175); |
| 2432 | svm_setmsrbrw(vcpu, MSR_SYSENTER_EIP0x176); |
| 2433 | svm_setmsrbrw(vcpu, MSR_STAR0xc0000081); |
| 2434 | svm_setmsrbrw(vcpu, MSR_LSTAR0xc0000082); |
| 2435 | svm_setmsrbrw(vcpu, MSR_CSTAR0xc0000083); |
| 2436 | svm_setmsrbrw(vcpu, MSR_SFMASK0xc0000084); |
| 2437 | svm_setmsrbrw(vcpu, MSR_FSBASE0xc0000100); |
| 2438 | svm_setmsrbrw(vcpu, MSR_GSBASE0xc0000101); |
| 2439 | svm_setmsrbrw(vcpu, MSR_KERNELGSBASE0xc0000102); |
| 2440 | |
| 2441 | /* EFER is R/O so we can ensure the guest always has SVME */ |
| 2442 | svm_setmsrbr(vcpu, MSR_EFER0xc0000080); |
| 2443 | |
| 2444 | /* allow reading TSC */ |
| 2445 | svm_setmsrbr(vcpu, MSR_TSC0x010); |
| 2446 | |
| 2447 | /* Guest VCPU ASID */ |
| 2448 | if (vmm_alloc_vpid(&asid)) { |
| 2449 | DPRINTF("%s: could not allocate asid\n", __func__); |
| 2450 | ret = EINVAL22; |
| 2451 | goto exit; |
| 2452 | } |
| 2453 | |
| 2454 | vmcb->v_asid = asid; |
| 2455 | vcpu->vc_vpid = asid; |
| 2456 | |
| 2457 | /* TLB Control - First time in, flush all*/ |
| 2458 | vmcb->v_tlb_control = SVM_TLB_CONTROL_FLUSH_ALL1; |
| 2459 | |
| 2460 | /* INTR masking */ |
| 2461 | vmcb->v_intr_masking = 1; |
| 2462 | |
| 2463 | /* PAT */ |
| 2464 | vmcb->v_g_pat = PATENTRY(0, PAT_WB)(0x6UL << ((0) * 8)) | PATENTRY(1, PAT_WC)(0x1UL << ((1) * 8)) | |
| 2465 | PATENTRY(2, PAT_UCMINUS)(0x7UL << ((2) * 8)) | PATENTRY(3, PAT_UC)(0x0UL << ((3) * 8)) | |
| 2466 | PATENTRY(4, PAT_WB)(0x6UL << ((4) * 8)) | PATENTRY(5, PAT_WC)(0x1UL << ((5) * 8)) | |
| 2467 | PATENTRY(6, PAT_UCMINUS)(0x7UL << ((6) * 8)) | PATENTRY(7, PAT_UC)(0x0UL << ((7) * 8)); |
| 2468 | |
| 2469 | /* NPT */ |
| 2470 | if (vmm_softc->mode == VMM_MODE_RVI) { |
| 2471 | vmcb->v_np_enable = 1; |
| 2472 | vmcb->v_n_cr3 = vcpu->vc_parent->vm_map->pmap->pm_pdirpa; |
| 2473 | } |
| 2474 | |
| 2475 | /* Enable SVME in EFER (must always be set) */ |
| 2476 | vmcb->v_efer |= EFER_SVME0x00001000; |
| 2477 | |
| 2478 | ret = vcpu_writeregs_svm(vcpu, VM_RWREGS_ALL(0x1 | 0x2 | 0x4 | 0x8 | 0x10), vrs); |
| 2479 | |
| 2480 | /* xcr0 power on default sets bit 0 (x87 state) */ |
| 2481 | vcpu->vc_gueststate.vg_xcr0 = XCR0_X870x00000001 & xsave_mask; |
| 2482 | |
| 2483 | vcpu->vc_parent->vm_map->pmap->eptp = 0; |
| 2484 | |
| 2485 | exit: |
| 2486 | return ret; |
| 2487 | } |
| 2488 | |
| 2489 | /* |
| 2490 | * svm_setmsrbr |
| 2491 | * |
| 2492 | * Allow read access to the specified msr on the supplied vcpu. |
| 2493 | * |
| 2494 | * Parameters: |
| 2495 | * vcpu: the VCPU to allow access |
| 2496 | * msr: the MSR number to allow access to |
| 2497 | */ |
| 2498 | void |
| 2499 | svm_setmsrbr(struct vcpu *vcpu, uint32_t msr) |
| 2500 | { |
| 2501 | uint8_t *msrs; |
| 2502 | uint16_t idx; |
| 2503 | |
| 2504 | msrs = (uint8_t *)vcpu->vc_msr_bitmap_va; |
| 2505 | |
| 2506 | /* |
| 2507 | * MSR Read bitmap layout: |
| 2508 | * Pentium MSRs (0x0 - 0x1fff) @ 0x0 |
| 2509 | * Gen6 and Syscall MSRs (0xc0000000 - 0xc0001fff) @ 0x800 |
| 2510 | * Gen7 and Gen8 MSRs (0xc0010000 - 0xc0011fff) @ 0x1000 |
| 2511 | * |
| 2512 | * Read enable bit is low order bit of 2-bit pair |
| 2513 | * per MSR (eg, MSR 0x0 write bit is at bit 0 @ 0x0) |
| 2514 | */ |
| 2515 | if (msr <= 0x1fff) { |
| 2516 | idx = SVM_MSRIDX(msr)((msr) / 4); |
| 2517 | msrs[idx] &= ~(SVM_MSRBIT_R(msr)(1 << (((msr) % 4) * 2))); |
| 2518 | } else if (msr >= 0xc0000000 && msr <= 0xc0001fff) { |
| 2519 | idx = SVM_MSRIDX(msr - 0xc0000000)((msr - 0xc0000000) / 4) + 0x800; |
| 2520 | msrs[idx] &= ~(SVM_MSRBIT_R(msr - 0xc0000000)(1 << (((msr - 0xc0000000) % 4) * 2))); |
| 2521 | } else if (msr >= 0xc0010000 && msr <= 0xc0011fff) { |
| 2522 | idx = SVM_MSRIDX(msr - 0xc0010000)((msr - 0xc0010000) / 4) + 0x1000; |
| 2523 | msrs[idx] &= ~(SVM_MSRBIT_R(msr - 0xc0010000)(1 << (((msr - 0xc0010000) % 4) * 2))); |
| 2524 | } else { |
| 2525 | printf("%s: invalid msr 0x%x\n", __func__, msr); |
| 2526 | return; |
| 2527 | } |
| 2528 | } |
| 2529 | |
| 2530 | /* |
| 2531 | * svm_setmsrbw |
| 2532 | * |
| 2533 | * Allow write access to the specified msr on the supplied vcpu |
| 2534 | * |
| 2535 | * Parameters: |
| 2536 | * vcpu: the VCPU to allow access |
| 2537 | * msr: the MSR number to allow access to |
| 2538 | */ |
| 2539 | void |
| 2540 | svm_setmsrbw(struct vcpu *vcpu, uint32_t msr) |
| 2541 | { |
| 2542 | uint8_t *msrs; |
| 2543 | uint16_t idx; |
| 2544 | |
| 2545 | msrs = (uint8_t *)vcpu->vc_msr_bitmap_va; |
| 2546 | |
| 2547 | /* |
| 2548 | * MSR Write bitmap layout: |
| 2549 | * Pentium MSRs (0x0 - 0x1fff) @ 0x0 |
| 2550 | * Gen6 and Syscall MSRs (0xc0000000 - 0xc0001fff) @ 0x800 |
| 2551 | * Gen7 and Gen8 MSRs (0xc0010000 - 0xc0011fff) @ 0x1000 |
| 2552 | * |
| 2553 | * Write enable bit is high order bit of 2-bit pair |
| 2554 | * per MSR (eg, MSR 0x0 write bit is at bit 1 @ 0x0) |
| 2555 | */ |
| 2556 | if (msr <= 0x1fff) { |
| 2557 | idx = SVM_MSRIDX(msr)((msr) / 4); |
| 2558 | msrs[idx] &= ~(SVM_MSRBIT_W(msr)(1 << (((msr) % 4) * 2 + 1))); |
| 2559 | } else if (msr >= 0xc0000000 && msr <= 0xc0001fff) { |
| 2560 | idx = SVM_MSRIDX(msr - 0xc0000000)((msr - 0xc0000000) / 4) + 0x800; |
| 2561 | msrs[idx] &= ~(SVM_MSRBIT_W(msr - 0xc0000000)(1 << (((msr - 0xc0000000) % 4) * 2 + 1))); |
| 2562 | } else if (msr >= 0xc0010000 && msr <= 0xc0011fff) { |
| 2563 | idx = SVM_MSRIDX(msr - 0xc0010000)((msr - 0xc0010000) / 4) + 0x1000; |
| 2564 | msrs[idx] &= ~(SVM_MSRBIT_W(msr - 0xc0010000)(1 << (((msr - 0xc0010000) % 4) * 2 + 1))); |
| 2565 | } else { |
| 2566 | printf("%s: invalid msr 0x%x\n", __func__, msr); |
| 2567 | return; |
| 2568 | } |
| 2569 | } |
| 2570 | |
| 2571 | /* |
| 2572 | * svm_setmsrbrw |
| 2573 | * |
| 2574 | * Allow read/write access to the specified msr on the supplied vcpu |
| 2575 | * |
| 2576 | * Parameters: |
| 2577 | * vcpu: the VCPU to allow access |
| 2578 | * msr: the MSR number to allow access to |
| 2579 | */ |
| 2580 | void |
| 2581 | svm_setmsrbrw(struct vcpu *vcpu, uint32_t msr) |
| 2582 | { |
| 2583 | svm_setmsrbr(vcpu, msr); |
| 2584 | svm_setmsrbw(vcpu, msr); |
| 2585 | } |
| 2586 | |
| 2587 | /* |
| 2588 | * vmx_setmsrbr |
| 2589 | * |
| 2590 | * Allow read access to the specified msr on the supplied vcpu. |
| 2591 | * |
| 2592 | * Parameters: |
| 2593 | * vcpu: the VCPU to allow access |
| 2594 | * msr: the MSR number to allow access to |
| 2595 | */ |
| 2596 | void |
| 2597 | vmx_setmsrbr(struct vcpu *vcpu, uint32_t msr) |
| 2598 | { |
| 2599 | uint8_t *msrs; |
| 2600 | uint16_t idx; |
| 2601 | |
| 2602 | msrs = (uint8_t *)vcpu->vc_msr_bitmap_va; |
| 2603 | |
| 2604 | /* |
| 2605 | * MSR Read bitmap layout: |
| 2606 | * "Low" MSRs (0x0 - 0x1fff) @ 0x0 |
| 2607 | * "High" MSRs (0xc0000000 - 0xc0001fff) @ 0x400 |
| 2608 | */ |
| 2609 | if (msr <= 0x1fff) { |
| 2610 | idx = VMX_MSRIDX(msr)((msr) / 8); |
| 2611 | msrs[idx] &= ~(VMX_MSRBIT(msr)(1 << (msr) % 8)); |
| 2612 | } else if (msr >= 0xc0000000 && msr <= 0xc0001fff) { |
| 2613 | idx = VMX_MSRIDX(msr - 0xc0000000)((msr - 0xc0000000) / 8) + 0x400; |
| 2614 | msrs[idx] &= ~(VMX_MSRBIT(msr - 0xc0000000)(1 << (msr - 0xc0000000) % 8)); |
| 2615 | } else |
| 2616 | printf("%s: invalid msr 0x%x\n", __func__, msr); |
| 2617 | } |
| 2618 | |
| 2619 | /* |
| 2620 | * vmx_setmsrbw |
| 2621 | * |
| 2622 | * Allow write access to the specified msr on the supplied vcpu |
| 2623 | * |
| 2624 | * Parameters: |
| 2625 | * vcpu: the VCPU to allow access |
| 2626 | * msr: the MSR number to allow access to |
| 2627 | */ |
| 2628 | void |
| 2629 | vmx_setmsrbw(struct vcpu *vcpu, uint32_t msr) |
| 2630 | { |
| 2631 | uint8_t *msrs; |
| 2632 | uint16_t idx; |
| 2633 | |
| 2634 | msrs = (uint8_t *)vcpu->vc_msr_bitmap_va; |
| 2635 | |
| 2636 | /* |
| 2637 | * MSR Write bitmap layout: |
| 2638 | * "Low" MSRs (0x0 - 0x1fff) @ 0x800 |
| 2639 | * "High" MSRs (0xc0000000 - 0xc0001fff) @ 0xc00 |
| 2640 | */ |
| 2641 | if (msr <= 0x1fff) { |
| 2642 | idx = VMX_MSRIDX(msr)((msr) / 8) + 0x800; |
| 2643 | msrs[idx] &= ~(VMX_MSRBIT(msr)(1 << (msr) % 8)); |
| 2644 | } else if (msr >= 0xc0000000 && msr <= 0xc0001fff) { |
| 2645 | idx = VMX_MSRIDX(msr - 0xc0000000)((msr - 0xc0000000) / 8) + 0xc00; |
| 2646 | msrs[idx] &= ~(VMX_MSRBIT(msr - 0xc0000000)(1 << (msr - 0xc0000000) % 8)); |
| 2647 | } else |
| 2648 | printf("%s: invalid msr 0x%x\n", __func__, msr); |
| 2649 | } |
| 2650 | |
| 2651 | /* |
| 2652 | * vmx_setmsrbrw |
| 2653 | * |
| 2654 | * Allow read/write access to the specified msr on the supplied vcpu |
| 2655 | * |
| 2656 | * Parameters: |
| 2657 | * vcpu: the VCPU to allow access |
| 2658 | * msr: the MSR number to allow access to |
| 2659 | */ |
| 2660 | void |
| 2661 | vmx_setmsrbrw(struct vcpu *vcpu, uint32_t msr) |
| 2662 | { |
| 2663 | vmx_setmsrbr(vcpu, msr); |
| 2664 | vmx_setmsrbw(vcpu, msr); |
| 2665 | } |
| 2666 | |
| 2667 | /* |
| 2668 | * svm_set_clean |
| 2669 | * |
| 2670 | * Sets (mark as unmodified) the VMCB clean bit set in 'value'. |
| 2671 | * For example, to set the clean bit for the VMCB intercepts (bit position 0), |
| 2672 | * the caller provides 'SVM_CLEANBITS_I' (0x1) for the 'value' argument. |
| 2673 | * Multiple cleanbits can be provided in 'value' at the same time (eg, |
| 2674 | * "SVM_CLEANBITS_I | SVM_CLEANBITS_TPR"). |
| 2675 | * |
| 2676 | * Note that this function does not clear any bits; to clear bits in the |
| 2677 | * vmcb cleanbits bitfield, use 'svm_set_dirty'. |
| 2678 | * |
| 2679 | * Parameters: |
| 2680 | * vmcs: the VCPU whose VMCB clean value should be set |
| 2681 | * value: the value(s) to enable in the cleanbits mask |
| 2682 | */ |
| 2683 | void |
| 2684 | svm_set_clean(struct vcpu *vcpu, uint32_t value) |
| 2685 | { |
| 2686 | struct vmcb *vmcb; |
| 2687 | |
| 2688 | /* If no cleanbits support, do nothing */ |
| 2689 | if (!curcpu()({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_vmm_cap.vcc_svm.svm_vmcb_clean) |
| 2690 | return; |
| 2691 | |
| 2692 | vmcb = (struct vmcb *)vcpu->vc_control_va; |
| 2693 | |
| 2694 | vmcb->v_vmcb_clean_bits |= value; |
| 2695 | } |
| 2696 | |
| 2697 | /* |
| 2698 | * svm_set_dirty |
| 2699 | * |
| 2700 | * Clears (mark as modified) the VMCB clean bit set in 'value'. |
| 2701 | * For example, to clear the bit for the VMCB intercepts (bit position 0) |
| 2702 | * the caller provides 'SVM_CLEANBITS_I' (0x1) for the 'value' argument. |
| 2703 | * Multiple dirty bits can be provided in 'value' at the same time (eg, |
| 2704 | * "SVM_CLEANBITS_I | SVM_CLEANBITS_TPR"). |
| 2705 | * |
| 2706 | * Parameters: |
| 2707 | * vmcs: the VCPU whose VMCB dirty value should be set |
| 2708 | * value: the value(s) to dirty in the cleanbits mask |
| 2709 | */ |
| 2710 | void |
| 2711 | svm_set_dirty(struct vcpu *vcpu, uint32_t value) |
| 2712 | { |
| 2713 | struct vmcb *vmcb; |
| 2714 | |
| 2715 | /* If no cleanbits support, do nothing */ |
| 2716 | if (!curcpu()({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_vmm_cap.vcc_svm.svm_vmcb_clean) |
| 2717 | return; |
| 2718 | |
| 2719 | vmcb = (struct vmcb *)vcpu->vc_control_va; |
| 2720 | |
| 2721 | vmcb->v_vmcb_clean_bits &= ~value; |
| 2722 | } |
| 2723 | |
| 2724 | /* |
| 2725 | * vcpu_reset_regs_vmx |
| 2726 | * |
| 2727 | * Initializes 'vcpu's registers to supplied state |
| 2728 | * |
| 2729 | * Parameters: |
| 2730 | * vcpu: the vcpu whose register state is to be initialized |
| 2731 | * vrs: the register state to set |
| 2732 | * |
| 2733 | * Return values: |
| 2734 | * 0: registers init'ed successfully |
| 2735 | * EINVAL: an error occurred setting register state |
| 2736 | */ |
| 2737 | int |
| 2738 | vcpu_reset_regs_vmx(struct vcpu *vcpu, struct vcpu_reg_state *vrs) |
| 2739 | { |
| 2740 | int ret = 0, ug = 0; |
| 2741 | uint32_t cr0, cr4; |
| 2742 | uint32_t pinbased, procbased, procbased2, exit, entry; |
| 2743 | uint32_t want1, want0; |
| 2744 | uint64_t msr, ctrlval, eptp, cr3; |
| 2745 | uint16_t ctrl, vpid; |
| 2746 | struct vmx_msr_store *msr_store; |
| 2747 | |
| 2748 | rw_assert_wrlock(&vcpu->vc_lock); |
| 2749 | |
| 2750 | cr0 = vrs->vrs_crs[VCPU_REGS_CR00]; |
| 2751 | |
| 2752 | if (vcpu_reload_vmcs_vmx(vcpu)) { |
| 2753 | DPRINTF("%s: error reloading VMCS\n", __func__); |
| 2754 | ret = EINVAL22; |
| 2755 | goto exit; |
| 2756 | } |
| 2757 | |
| 2758 | #ifdef VMM_DEBUG |
| 2759 | /* VMCS should be loaded... */ |
| 2760 | paddr_t pa = 0ULL; |
| 2761 | if (vmptrst(&pa)) |
| 2762 | panic("%s: vmptrst", __func__); |
| 2763 | KASSERT(pa == vcpu->vc_control_pa)((pa == vcpu->vc_control_pa) ? (void)0 : __assert("diagnostic " , "/usr/src/sys/arch/amd64/amd64/vmm.c", 2763, "pa == vcpu->vc_control_pa" )); |
| 2764 | #endif /* VMM_DEBUG */ |
| 2765 | |
| 2766 | /* Compute Basic Entry / Exit Controls */ |
| 2767 | vcpu->vc_vmx_basic = rdmsr(IA32_VMX_BASIC0x480); |
| 2768 | vcpu->vc_vmx_entry_ctls = rdmsr(IA32_VMX_ENTRY_CTLS0x484); |
| 2769 | vcpu->vc_vmx_exit_ctls = rdmsr(IA32_VMX_EXIT_CTLS0x483); |
| 2770 | vcpu->vc_vmx_pinbased_ctls = rdmsr(IA32_VMX_PINBASED_CTLS0x481); |
| 2771 | vcpu->vc_vmx_procbased_ctls = rdmsr(IA32_VMX_PROCBASED_CTLS0x482); |
| 2772 | |
| 2773 | /* Compute True Entry / Exit Controls (if applicable) */ |
| 2774 | if (vcpu->vc_vmx_basic & IA32_VMX_TRUE_CTLS_AVAIL(1ULL << 55)) { |
| 2775 | vcpu->vc_vmx_true_entry_ctls = rdmsr(IA32_VMX_TRUE_ENTRY_CTLS0x490); |
| 2776 | vcpu->vc_vmx_true_exit_ctls = rdmsr(IA32_VMX_TRUE_EXIT_CTLS0x48F); |
| 2777 | vcpu->vc_vmx_true_pinbased_ctls = |
| 2778 | rdmsr(IA32_VMX_TRUE_PINBASED_CTLS0x48D); |
| 2779 | vcpu->vc_vmx_true_procbased_ctls = |
| 2780 | rdmsr(IA32_VMX_TRUE_PROCBASED_CTLS0x48E); |
| 2781 | } |
| 2782 | |
| 2783 | /* Compute Secondary Procbased Controls (if applicable) */ |
| 2784 | if (vcpu_vmx_check_cap(vcpu, IA32_VMX_PROCBASED_CTLS0x482, |
| 2785 | IA32_VMX_ACTIVATE_SECONDARY_CONTROLS(1ULL << 31), 1)) |
| 2786 | vcpu->vc_vmx_procbased2_ctls = rdmsr(IA32_VMX_PROCBASED2_CTLS0x48B); |
| 2787 | |
| 2788 | /* |
| 2789 | * Pinbased ctrls |
| 2790 | * |
| 2791 | * We must be able to set the following: |
| 2792 | * IA32_VMX_EXTERNAL_INT_EXITING - exit on host interrupt |
| 2793 | * IA32_VMX_NMI_EXITING - exit on host NMI |
| 2794 | */ |
| 2795 | want1 = IA32_VMX_EXTERNAL_INT_EXITING(1ULL << 0) | |
| 2796 | IA32_VMX_NMI_EXITING(1ULL << 3); |
| 2797 | want0 = 0; |
| 2798 | |
| 2799 | if (vcpu->vc_vmx_basic & IA32_VMX_TRUE_CTLS_AVAIL(1ULL << 55)) { |
| 2800 | ctrl = IA32_VMX_TRUE_PINBASED_CTLS0x48D; |
| 2801 | ctrlval = vcpu->vc_vmx_true_pinbased_ctls; |
| 2802 | } else { |
| 2803 | ctrl = IA32_VMX_PINBASED_CTLS0x481; |
| 2804 | ctrlval = vcpu->vc_vmx_pinbased_ctls; |
| 2805 | } |
| 2806 | |
| 2807 | if (vcpu_vmx_compute_ctrl(ctrlval, ctrl, want1, want0, &pinbased)) { |
| 2808 | DPRINTF("%s: error computing pinbased controls\n", __func__); |
| 2809 | ret = EINVAL22; |
| 2810 | goto exit; |
| 2811 | } |
| 2812 | |
| 2813 | if (vmwrite(VMCS_PINBASED_CTLS0x4000, pinbased)) { |
| 2814 | DPRINTF("%s: error setting pinbased controls\n", __func__); |
| 2815 | ret = EINVAL22; |
| 2816 | goto exit; |
| 2817 | } |
| 2818 | |
| 2819 | /* |
| 2820 | * Procbased ctrls |
| 2821 | * |
| 2822 | * We must be able to set the following: |
| 2823 | * IA32_VMX_HLT_EXITING - exit on HLT instruction |
| 2824 | * IA32_VMX_MWAIT_EXITING - exit on MWAIT instruction |
| 2825 | * IA32_VMX_UNCONDITIONAL_IO_EXITING - exit on I/O instructions |
| 2826 | * IA32_VMX_USE_MSR_BITMAPS - exit on various MSR accesses |
| 2827 | * IA32_VMX_CR8_LOAD_EXITING - guest TPR access |
| 2828 | * IA32_VMX_CR8_STORE_EXITING - guest TPR access |
| 2829 | * IA32_VMX_USE_TPR_SHADOW - guest TPR access (shadow) |
| 2830 | * IA32_VMX_MONITOR_EXITING - exit on MONITOR instruction |
| 2831 | * |
| 2832 | * If we have EPT, we must be able to clear the following |
| 2833 | * IA32_VMX_CR3_LOAD_EXITING - don't care about guest CR3 accesses |
| 2834 | * IA32_VMX_CR3_STORE_EXITING - don't care about guest CR3 accesses |
| 2835 | */ |
| 2836 | want1 = IA32_VMX_HLT_EXITING(1ULL << 7) | |
| 2837 | IA32_VMX_MWAIT_EXITING(1ULL << 10) | |
| 2838 | IA32_VMX_UNCONDITIONAL_IO_EXITING(1ULL << 24) | |
| 2839 | IA32_VMX_USE_MSR_BITMAPS(1ULL << 28) | |
| 2840 | IA32_VMX_CR8_LOAD_EXITING(1ULL << 19) | |
| 2841 | IA32_VMX_CR8_STORE_EXITING(1ULL << 20) | |
| 2842 | IA32_VMX_MONITOR_EXITING(1ULL << 29) | |
| 2843 | IA32_VMX_USE_TPR_SHADOW(1ULL << 21); |
| 2844 | want0 = 0; |
| 2845 | |
| 2846 | if (vmm_softc->mode == VMM_MODE_EPT) { |
| 2847 | want1 |= IA32_VMX_ACTIVATE_SECONDARY_CONTROLS(1ULL << 31); |
| 2848 | want0 |= IA32_VMX_CR3_LOAD_EXITING(1ULL << 15) | |
| 2849 | IA32_VMX_CR3_STORE_EXITING(1ULL << 16); |
| 2850 | } |
| 2851 | |
| 2852 | if (vcpu->vc_vmx_basic & IA32_VMX_TRUE_CTLS_AVAIL(1ULL << 55)) { |
| 2853 | ctrl = IA32_VMX_TRUE_PROCBASED_CTLS0x48E; |
| 2854 | ctrlval = vcpu->vc_vmx_true_procbased_ctls; |
| 2855 | } else { |
| 2856 | ctrl = IA32_VMX_PROCBASED_CTLS0x482; |
| 2857 | ctrlval = vcpu->vc_vmx_procbased_ctls; |
| 2858 | } |
| 2859 | |
| 2860 | if (vcpu_vmx_compute_ctrl(ctrlval, ctrl, want1, want0, &procbased)) { |
| 2861 | DPRINTF("%s: error computing procbased controls\n", __func__); |
| 2862 | ret = EINVAL22; |
| 2863 | goto exit; |
| 2864 | } |
| 2865 | |
| 2866 | if (vmwrite(VMCS_PROCBASED_CTLS0x4002, procbased)) { |
| 2867 | DPRINTF("%s: error setting procbased controls\n", __func__); |
| 2868 | ret = EINVAL22; |
| 2869 | goto exit; |
| 2870 | } |
| 2871 | |
| 2872 | /* |
| 2873 | * Secondary Procbased ctrls |
| 2874 | * |
| 2875 | * We want to be able to set the following, if available: |
| 2876 | * IA32_VMX_ENABLE_VPID - use VPIDs where available |
| 2877 | * |
| 2878 | * If we have EPT, we must be able to set the following: |
| 2879 | * IA32_VMX_ENABLE_EPT - enable EPT |
| 2880 | * |
| 2881 | * If we have unrestricted guest capability, we must be able to set |
| 2882 | * the following: |
| 2883 | * IA32_VMX_UNRESTRICTED_GUEST - enable unrestricted guest (if caller |
| 2884 | * specified CR0_PG | CR0_PE in %cr0 in the 'vrs' parameter) |
| 2885 | */ |
| 2886 | want1 = 0; |
| 2887 | |
| 2888 | /* XXX checking for 2ndary controls can be combined here */ |
| 2889 | if (vcpu_vmx_check_cap(vcpu, IA32_VMX_PROCBASED_CTLS0x482, |
| 2890 | IA32_VMX_ACTIVATE_SECONDARY_CONTROLS(1ULL << 31), 1)) { |
| 2891 | if (vcpu_vmx_check_cap(vcpu, IA32_VMX_PROCBASED2_CTLS0x48B, |
| 2892 | IA32_VMX_ENABLE_VPID(1ULL << 5), 1)) { |
| 2893 | want1 |= IA32_VMX_ENABLE_VPID(1ULL << 5); |
| 2894 | vcpu->vc_vmx_vpid_enabled = 1; |
| 2895 | } |
| 2896 | } |
| 2897 | |
| 2898 | if (vmm_softc->mode == VMM_MODE_EPT) |
| 2899 | want1 |= IA32_VMX_ENABLE_EPT(1ULL << 1); |
| 2900 | |
| 2901 | if (vcpu_vmx_check_cap(vcpu, IA32_VMX_PROCBASED_CTLS0x482, |
| 2902 | IA32_VMX_ACTIVATE_SECONDARY_CONTROLS(1ULL << 31), 1)) { |
| 2903 | if (vcpu_vmx_check_cap(vcpu, IA32_VMX_PROCBASED2_CTLS0x48B, |
| 2904 | IA32_VMX_UNRESTRICTED_GUEST(1ULL << 7), 1)) { |
| 2905 | if ((cr0 & (CR0_PE0x00000001 | CR0_PG0x80000000)) == 0) { |
| 2906 | want1 |= IA32_VMX_UNRESTRICTED_GUEST(1ULL << 7); |
| 2907 | ug = 1; |
| 2908 | } |
| 2909 | } |
| 2910 | } |
| 2911 | |
| 2912 | want0 = ~want1; |
| 2913 | ctrlval = vcpu->vc_vmx_procbased2_ctls; |
| 2914 | ctrl = IA32_VMX_PROCBASED2_CTLS0x48B; |
| 2915 | |
| 2916 | if (vcpu_vmx_compute_ctrl(ctrlval, ctrl, want1, want0, &procbased2)) { |
| 2917 | DPRINTF("%s: error computing secondary procbased controls\n", |
| 2918 | __func__); |
| 2919 | ret = EINVAL22; |
| 2920 | goto exit; |
| 2921 | } |
| 2922 | |
| 2923 | if (vmwrite(VMCS_PROCBASED2_CTLS0x401E, procbased2)) { |
| 2924 | DPRINTF("%s: error setting secondary procbased controls\n", |
| 2925 | __func__); |
| 2926 | ret = EINVAL22; |
| 2927 | goto exit; |
| 2928 | } |
| 2929 | |
| 2930 | /* |
| 2931 | * Exit ctrls |
| 2932 | * |
| 2933 | * We must be able to set the following: |
| 2934 | * IA32_VMX_SAVE_DEBUG_CONTROLS |
| 2935 | * IA32_VMX_HOST_SPACE_ADDRESS_SIZE - exit to long mode |
| 2936 | * IA32_VMX_ACKNOWLEDGE_INTERRUPT_ON_EXIT - ack interrupt on exit |
| 2937 | */ |
| 2938 | want1 = IA32_VMX_HOST_SPACE_ADDRESS_SIZE(1ULL << 9) | |
| 2939 | IA32_VMX_ACKNOWLEDGE_INTERRUPT_ON_EXIT(1ULL << 15) | |
| 2940 | IA32_VMX_SAVE_DEBUG_CONTROLS(1ULL << 2); |
| 2941 | want0 = 0; |
| 2942 | |
| 2943 | if (vcpu->vc_vmx_basic & IA32_VMX_TRUE_CTLS_AVAIL(1ULL << 55)) { |
| 2944 | ctrl = IA32_VMX_TRUE_EXIT_CTLS0x48F; |
| 2945 | ctrlval = vcpu->vc_vmx_true_exit_ctls; |
| 2946 | } else { |
| 2947 | ctrl = IA32_VMX_EXIT_CTLS0x483; |
| 2948 | ctrlval = vcpu->vc_vmx_exit_ctls; |
| 2949 | } |
| 2950 | |
| 2951 | if (vcpu_vmx_compute_ctrl(ctrlval, ctrl, want1, want0, &exit)) { |
| 2952 | DPRINTF("%s: error computing exit controls\n", __func__); |
| 2953 | ret = EINVAL22; |
| 2954 | goto exit; |
| 2955 | } |
| 2956 | |
| 2957 | if (vmwrite(VMCS_EXIT_CTLS0x400C, exit)) { |
| 2958 | DPRINTF("%s: error setting exit controls\n", __func__); |
| 2959 | ret = EINVAL22; |
| 2960 | goto exit; |
| 2961 | } |
| 2962 | |
| 2963 | /* |
| 2964 | * Entry ctrls |
| 2965 | * |
| 2966 | * We must be able to set the following: |
| 2967 | * IA32_VMX_IA32E_MODE_GUEST (if no unrestricted guest) |
| 2968 | * IA32_VMX_LOAD_DEBUG_CONTROLS |
| 2969 | * We must be able to clear the following: |
| 2970 | * IA32_VMX_ENTRY_TO_SMM - enter to SMM |
| 2971 | * IA32_VMX_DEACTIVATE_DUAL_MONITOR_TREATMENT |
| 2972 | * IA32_VMX_LOAD_IA32_PERF_GLOBAL_CTRL_ON_ENTRY |
| 2973 | */ |
| 2974 | want1 = IA32_VMX_LOAD_DEBUG_CONTROLS(1ULL << 2); |
| 2975 | if (vrs->vrs_msrs[VCPU_REGS_EFER0] & EFER_LMA0x00000400) |
| 2976 | want1 |= IA32_VMX_IA32E_MODE_GUEST(1ULL << 9); |
| 2977 | |
| 2978 | want0 = IA32_VMX_ENTRY_TO_SMM(1ULL << 10) | |
| 2979 | IA32_VMX_DEACTIVATE_DUAL_MONITOR_TREATMENT(1ULL << 11) | |
| 2980 | IA32_VMX_LOAD_IA32_PERF_GLOBAL_CTRL_ON_ENTRY(1ULL << 13); |
| 2981 | |
| 2982 | if (vcpu->vc_vmx_basic & IA32_VMX_TRUE_CTLS_AVAIL(1ULL << 55)) { |
| 2983 | ctrl = IA32_VMX_TRUE_ENTRY_CTLS0x490; |
| 2984 | ctrlval = vcpu->vc_vmx_true_entry_ctls; |
| 2985 | } else { |
| 2986 | ctrl = IA32_VMX_ENTRY_CTLS0x484; |
| 2987 | ctrlval = vcpu->vc_vmx_entry_ctls; |
| 2988 | } |
| 2989 | |
| 2990 | if (vcpu_vmx_compute_ctrl(ctrlval, ctrl, want1, want0, &entry)) { |
| 2991 | ret = EINVAL22; |
| 2992 | goto exit; |
| 2993 | } |
| 2994 | |
| 2995 | if (vmwrite(VMCS_ENTRY_CTLS0x4012, entry)) { |
| 2996 | ret = EINVAL22; |
| 2997 | goto exit; |
| 2998 | } |
| 2999 | |
| 3000 | if (vmm_softc->mode == VMM_MODE_EPT) { |
| 3001 | eptp = vcpu->vc_parent->vm_map->pmap->pm_pdirpa; |
| 3002 | msr = rdmsr(IA32_VMX_EPT_VPID_CAP0x48C); |
| 3003 | if (msr & IA32_EPT_VPID_CAP_PAGE_WALK_4(1ULL << 6)) { |
| 3004 | /* Page walk length 4 supported */ |
| 3005 | eptp |= ((IA32_EPT_PAGE_WALK_LENGTH0x4 - 1) << 3); |
| 3006 | } else { |
| 3007 | DPRINTF("EPT page walk length 4 not supported\n"); |
| 3008 | ret = EINVAL22; |
| 3009 | goto exit; |
| 3010 | } |
| 3011 | |
| 3012 | if (msr & IA32_EPT_VPID_CAP_WB(1ULL << 14)) { |
| 3013 | /* WB cache type supported */ |
| 3014 | eptp |= IA32_EPT_PAGING_CACHE_TYPE_WB0x6; |
| 3015 | } else |
| 3016 | DPRINTF("%s: no WB cache type available, guest VM " |
| 3017 | "will run uncached\n", __func__); |
| 3018 | |
| 3019 | DPRINTF("Guest EPTP = 0x%llx\n", eptp); |
| 3020 | if (vmwrite(VMCS_GUEST_IA32_EPTP0x201A, eptp)) { |
| 3021 | DPRINTF("%s: error setting guest EPTP\n", __func__); |
| 3022 | ret = EINVAL22; |
| 3023 | goto exit; |
| 3024 | } |
| 3025 | |
| 3026 | vcpu->vc_parent->vm_map->pmap->eptp = eptp; |
| 3027 | } |
| 3028 | |
| 3029 | if (vcpu_vmx_check_cap(vcpu, IA32_VMX_PROCBASED_CTLS0x482, |
| 3030 | IA32_VMX_ACTIVATE_SECONDARY_CONTROLS(1ULL << 31), 1)) { |
| 3031 | if (vcpu_vmx_check_cap(vcpu, IA32_VMX_PROCBASED2_CTLS0x48B, |
| 3032 | IA32_VMX_ENABLE_VPID(1ULL << 5), 1)) { |
| 3033 | |
| 3034 | /* We may sleep during allocation, so reload VMCS. */ |
| 3035 | vcpu->vc_last_pcpu = curcpu()({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;}); |
| 3036 | ret = vmm_alloc_vpid(&vpid); |
| 3037 | if (vcpu_reload_vmcs_vmx(vcpu)) { |
| 3038 | printf("%s: failed to reload vmcs\n", __func__); |
| 3039 | ret = EINVAL22; |
| 3040 | goto exit; |
| 3041 | } |
| 3042 | if (ret) { |
| 3043 | DPRINTF("%s: could not allocate VPID\n", |
| 3044 | __func__); |
| 3045 | ret = EINVAL22; |
| 3046 | goto exit; |
| 3047 | } |
| 3048 | |
| 3049 | if (vmwrite(VMCS_GUEST_VPID0x0000, vpid)) { |
| 3050 | DPRINTF("%s: error setting guest VPID\n", |
| 3051 | __func__); |
| 3052 | ret = EINVAL22; |
| 3053 | goto exit; |
| 3054 | } |
| 3055 | |
| 3056 | vcpu->vc_vpid = vpid; |
| 3057 | } |
| 3058 | } |
| 3059 | |
| 3060 | /* |
| 3061 | * Determine which bits in CR0 have to be set to a fixed |
| 3062 | * value as per Intel SDM A.7. |
| 3063 | * CR0 bits in the vrs parameter must match these. |
| 3064 | */ |
| 3065 | want1 = (curcpu()({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_vmm_cap.vcc_vmx.vmx_cr0_fixed0) & |
| 3066 | (curcpu()({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_vmm_cap.vcc_vmx.vmx_cr0_fixed1); |
| 3067 | want0 = ~(curcpu()({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_vmm_cap.vcc_vmx.vmx_cr0_fixed0) & |
| 3068 | ~(curcpu()({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_vmm_cap.vcc_vmx.vmx_cr0_fixed1); |
| 3069 | |
| 3070 | /* |
| 3071 | * CR0_FIXED0 and CR0_FIXED1 may report the CR0_PG and CR0_PE bits as |
| 3072 | * fixed to 1 even if the CPU supports the unrestricted guest |
| 3073 | * feature. Update want1 and want0 accordingly to allow |
| 3074 | * any value for CR0_PG and CR0_PE in vrs->vrs_crs[VCPU_REGS_CR0] if |
| 3075 | * the CPU has the unrestricted guest capability. |
| 3076 | */ |
| 3077 | if (ug) { |
| 3078 | want1 &= ~(CR0_PG0x80000000 | CR0_PE0x00000001); |
| 3079 | want0 &= ~(CR0_PG0x80000000 | CR0_PE0x00000001); |
| 3080 | } |
| 3081 | |
| 3082 | /* |
| 3083 | * VMX may require some bits to be set that userland should not have |
| 3084 | * to care about. Set those here. |
| 3085 | */ |
| 3086 | if (want1 & CR0_NE0x00000020) |
| 3087 | cr0 |= CR0_NE0x00000020; |
| 3088 | |
| 3089 | if ((cr0 & want1) != want1) { |
| 3090 | ret = EINVAL22; |
| 3091 | goto exit; |
| 3092 | } |
| 3093 | |
| 3094 | if ((~cr0 & want0) != want0) { |
| 3095 | ret = EINVAL22; |
| 3096 | goto exit; |
| 3097 | } |
| 3098 | |
| 3099 | vcpu->vc_vmx_cr0_fixed1 = want1; |
| 3100 | vcpu->vc_vmx_cr0_fixed0 = want0; |
| 3101 | /* |
| 3102 | * Determine which bits in CR4 have to be set to a fixed |
| 3103 | * value as per Intel SDM A.8. |
| 3104 | * CR4 bits in the vrs parameter must match these, except |
| 3105 | * CR4_VMXE - we add that here since it must always be set. |
| 3106 | */ |
| 3107 | want1 = (curcpu()({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_vmm_cap.vcc_vmx.vmx_cr4_fixed0) & |
| 3108 | (curcpu()({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_vmm_cap.vcc_vmx.vmx_cr4_fixed1); |
| 3109 | want0 = ~(curcpu()({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_vmm_cap.vcc_vmx.vmx_cr4_fixed0) & |
| 3110 | ~(curcpu()({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_vmm_cap.vcc_vmx.vmx_cr4_fixed1); |
| 3111 | |
| 3112 | cr4 = vrs->vrs_crs[VCPU_REGS_CR43] | CR4_VMXE0x00002000; |
| 3113 | |
| 3114 | if ((cr4 & want1) != want1) { |
| 3115 | ret = EINVAL22; |
| 3116 | goto exit; |
| 3117 | } |
| 3118 | |
| 3119 | if ((~cr4 & want0) != want0) { |
| 3120 | ret = EINVAL22; |
| 3121 | goto exit; |
| 3122 | } |
| 3123 | |
| 3124 | cr3 = vrs->vrs_crs[VCPU_REGS_CR32]; |
| 3125 | |
| 3126 | /* Restore PDPTEs if 32-bit PAE paging is being used */ |
| 3127 | if (cr3 && (cr4 & CR4_PAE0x00000020) && |
| 3128 | !(vrs->vrs_msrs[VCPU_REGS_EFER0] & EFER_LMA0x00000400)) { |
| 3129 | if (vmwrite(VMCS_GUEST_PDPTE00x280A, |
| 3130 | vrs->vrs_crs[VCPU_REGS_PDPTE06])) { |
| 3131 | ret = EINVAL22; |
| 3132 | goto exit; |
| 3133 | } |
| 3134 | |
| 3135 | if (vmwrite(VMCS_GUEST_PDPTE10x280C, |
| 3136 | vrs->vrs_crs[VCPU_REGS_PDPTE17])) { |
| 3137 | ret = EINVAL22; |
| 3138 | goto exit; |
| 3139 | } |
| 3140 | |
| 3141 | if (vmwrite(VMCS_GUEST_PDPTE20x280E, |
| 3142 | vrs->vrs_crs[VCPU_REGS_PDPTE28])) { |
| 3143 | ret = EINVAL22; |
| 3144 | goto exit; |
| 3145 | } |
| 3146 | |
| 3147 | if (vmwrite(VMCS_GUEST_PDPTE30x2810, |
| 3148 | vrs->vrs_crs[VCPU_REGS_PDPTE39])) { |
| 3149 | ret = EINVAL22; |
| 3150 | goto exit; |
| 3151 | } |
| 3152 | } |
| 3153 | |
| 3154 | vrs->vrs_crs[VCPU_REGS_CR00] = cr0; |
| 3155 | vrs->vrs_crs[VCPU_REGS_CR43] = cr4; |
| 3156 | |
| 3157 | /* |
| 3158 | * Select host MSRs to be loaded on exit |
| 3159 | */ |
| 3160 | msr_store = (struct vmx_msr_store *)vcpu->vc_vmx_msr_exit_load_va; |
| 3161 | msr_store[0].vms_index = MSR_EFER0xc0000080; |
| 3162 | msr_store[0].vms_data = rdmsr(MSR_EFER0xc0000080); |
| 3163 | msr_store[1].vms_index = MSR_STAR0xc0000081; |
| 3164 | msr_store[1].vms_data = rdmsr(MSR_STAR0xc0000081); |
| 3165 | msr_store[2].vms_index = MSR_LSTAR0xc0000082; |
| 3166 | msr_store[2].vms_data = rdmsr(MSR_LSTAR0xc0000082); |
| 3167 | msr_store[3].vms_index = MSR_CSTAR0xc0000083; |
| 3168 | msr_store[3].vms_data = rdmsr(MSR_CSTAR0xc0000083); |
| 3169 | msr_store[4].vms_index = MSR_SFMASK0xc0000084; |
| 3170 | msr_store[4].vms_data = rdmsr(MSR_SFMASK0xc0000084); |
| 3171 | msr_store[5].vms_index = MSR_KERNELGSBASE0xc0000102; |
| 3172 | msr_store[5].vms_data = rdmsr(MSR_KERNELGSBASE0xc0000102); |
| 3173 | msr_store[6].vms_index = MSR_MISC_ENABLE0x1a0; |
| 3174 | msr_store[6].vms_data = rdmsr(MSR_MISC_ENABLE0x1a0); |
| 3175 | |
| 3176 | /* |
| 3177 | * Select guest MSRs to be loaded on entry / saved on exit |
| 3178 | */ |
| 3179 | msr_store = (struct vmx_msr_store *)vcpu->vc_vmx_msr_exit_save_va; |
| 3180 | |
| 3181 | msr_store[VCPU_REGS_EFER0].vms_index = MSR_EFER0xc0000080; |
| 3182 | msr_store[VCPU_REGS_STAR1].vms_index = MSR_STAR0xc0000081; |
| 3183 | msr_store[VCPU_REGS_LSTAR2].vms_index = MSR_LSTAR0xc0000082; |
| 3184 | msr_store[VCPU_REGS_CSTAR3].vms_index = MSR_CSTAR0xc0000083; |
| 3185 | msr_store[VCPU_REGS_SFMASK4].vms_index = MSR_SFMASK0xc0000084; |
| 3186 | msr_store[VCPU_REGS_KGSBASE5].vms_index = MSR_KERNELGSBASE0xc0000102; |
| 3187 | msr_store[VCPU_REGS_MISC_ENABLE6].vms_index = MSR_MISC_ENABLE0x1a0; |
| 3188 | |
| 3189 | /* |
| 3190 | * Initialize MSR_MISC_ENABLE as it can't be read and populated from vmd |
| 3191 | * and some of the content is based on the host. |
| 3192 | */ |
| 3193 | msr_store[VCPU_REGS_MISC_ENABLE6].vms_data = rdmsr(MSR_MISC_ENABLE0x1a0); |
| 3194 | msr_store[VCPU_REGS_MISC_ENABLE6].vms_data &= |
| 3195 | ~(MISC_ENABLE_TCC(1 << 3) | MISC_ENABLE_PERF_MON_AVAILABLE(1 << 7) | |
| 3196 | MISC_ENABLE_EIST_ENABLED(1 << 16) | MISC_ENABLE_ENABLE_MONITOR_FSM(1 << 18) | |
| 3197 | MISC_ENABLE_xTPR_MESSAGE_DISABLE(1 << 23)); |
| 3198 | msr_store[VCPU_REGS_MISC_ENABLE6].vms_data |= |
| 3199 | MISC_ENABLE_BTS_UNAVAILABLE(1 << 11) | MISC_ENABLE_PEBS_UNAVAILABLE(1 << 12); |
| 3200 | |
| 3201 | /* |
| 3202 | * Currently we have the same count of entry/exit MSRs loads/stores |
| 3203 | * but this is not an architectural requirement. |
| 3204 | */ |
| 3205 | if (vmwrite(VMCS_EXIT_MSR_STORE_COUNT0x400E, VMX_NUM_MSR_STORE7)) { |
| 3206 | DPRINTF("%s: error setting guest MSR exit store count\n", |
| 3207 | __func__); |
| 3208 | ret = EINVAL22; |
| 3209 | goto exit; |
| 3210 | } |
| 3211 | |
| 3212 | if (vmwrite(VMCS_EXIT_MSR_LOAD_COUNT0x4010, VMX_NUM_MSR_STORE7)) { |
| 3213 | DPRINTF("%s: error setting guest MSR exit load count\n", |
| 3214 | __func__); |
| 3215 | ret = EINVAL22; |
| 3216 | goto exit; |
| 3217 | } |
| 3218 | |
| 3219 | if (vmwrite(VMCS_ENTRY_MSR_LOAD_COUNT0x4014, VMX_NUM_MSR_STORE7)) { |
| 3220 | DPRINTF("%s: error setting guest MSR entry load count\n", |
| 3221 | __func__); |
| 3222 | ret = EINVAL22; |
| 3223 | goto exit; |
| 3224 | } |
| 3225 | |
| 3226 | if (vmwrite(VMCS_EXIT_STORE_MSR_ADDRESS0x2006, |
| 3227 | vcpu->vc_vmx_msr_exit_save_pa)) { |
| 3228 | DPRINTF("%s: error setting guest MSR exit store address\n", |
| 3229 | __func__); |
| 3230 | ret = EINVAL22; |
| 3231 | goto exit; |
| 3232 | } |
| 3233 | |
| 3234 | if (vmwrite(VMCS_EXIT_LOAD_MSR_ADDRESS0x2008, |
| 3235 | vcpu->vc_vmx_msr_exit_load_pa)) { |
| 3236 | DPRINTF("%s: error setting guest MSR exit load address\n", |
| 3237 | __func__); |
| 3238 | ret = EINVAL22; |
| 3239 | goto exit; |
| 3240 | } |
| 3241 | |
| 3242 | if (vmwrite(VMCS_ENTRY_LOAD_MSR_ADDRESS0x200A, |
| 3243 | vcpu->vc_vmx_msr_exit_save_pa)) { |
| 3244 | DPRINTF("%s: error setting guest MSR entry load address\n", |
| 3245 | __func__); |
| 3246 | ret = EINVAL22; |
| 3247 | goto exit; |
| 3248 | } |
| 3249 | |
| 3250 | if (vmwrite(VMCS_MSR_BITMAP_ADDRESS0x2004, |
| 3251 | vcpu->vc_msr_bitmap_pa)) { |
| 3252 | DPRINTF("%s: error setting guest MSR bitmap address\n", |
| 3253 | __func__); |
| 3254 | ret = EINVAL22; |
| 3255 | goto exit; |
| 3256 | } |
| 3257 | |
| 3258 | if (vmwrite(VMCS_CR4_MASK0x6002, CR4_VMXE0x00002000)) { |
| 3259 | DPRINTF("%s: error setting guest CR4 mask\n", __func__); |
| 3260 | ret = EINVAL22; |
| 3261 | goto exit; |
| 3262 | } |
| 3263 | |
| 3264 | if (vmwrite(VMCS_CR0_MASK0x6000, CR0_NE0x00000020)) { |
| 3265 | DPRINTF("%s: error setting guest CR0 mask\n", __func__); |
| 3266 | ret = EINVAL22; |
| 3267 | goto exit; |
| 3268 | } |
| 3269 | |
| 3270 | /* |
| 3271 | * Set up the VMCS for the register state we want during VCPU start. |
| 3272 | * This matches what the CPU state would be after a bootloader |
| 3273 | * transition to 'start'. |
| 3274 | */ |
| 3275 | ret = vcpu_writeregs_vmx(vcpu, VM_RWREGS_ALL(0x1 | 0x2 | 0x4 | 0x8 | 0x10), 0, vrs); |
| 3276 | |
| 3277 | /* |
| 3278 | * Set up the MSR bitmap |
| 3279 | */ |
| 3280 | memset((uint8_t *)vcpu->vc_msr_bitmap_va, 0xFF, PAGE_SIZE)__builtin_memset(((uint8_t *)vcpu->vc_msr_bitmap_va), (0xFF ), ((1 << 12))); |
| 3281 | vmx_setmsrbrw(vcpu, MSR_IA32_FEATURE_CONTROL0x03a); |
| 3282 | vmx_setmsrbrw(vcpu, MSR_SYSENTER_CS0x174); |
| 3283 | vmx_setmsrbrw(vcpu, MSR_SYSENTER_ESP0x175); |
| 3284 | vmx_setmsrbrw(vcpu, MSR_SYSENTER_EIP0x176); |
| 3285 | vmx_setmsrbrw(vcpu, MSR_EFER0xc0000080); |
| 3286 | vmx_setmsrbrw(vcpu, MSR_STAR0xc0000081); |
| 3287 | vmx_setmsrbrw(vcpu, MSR_LSTAR0xc0000082); |
| 3288 | vmx_setmsrbrw(vcpu, MSR_CSTAR0xc0000083); |
| 3289 | vmx_setmsrbrw(vcpu, MSR_SFMASK0xc0000084); |
| 3290 | vmx_setmsrbrw(vcpu, MSR_FSBASE0xc0000100); |
| 3291 | vmx_setmsrbrw(vcpu, MSR_GSBASE0xc0000101); |
| 3292 | vmx_setmsrbrw(vcpu, MSR_KERNELGSBASE0xc0000102); |
| 3293 | vmx_setmsrbr(vcpu, MSR_MISC_ENABLE0x1a0); |
| 3294 | |
| 3295 | /* XXX CR0 shadow */ |
| 3296 | /* XXX CR4 shadow */ |
| 3297 | |
| 3298 | /* xcr0 power on default sets bit 0 (x87 state) */ |
| 3299 | vcpu->vc_gueststate.vg_xcr0 = XCR0_X870x00000001 & xsave_mask; |
| 3300 | |
| 3301 | /* XXX PAT shadow */ |
| 3302 | vcpu->vc_shadow_pat = rdmsr(MSR_CR_PAT0x277); |
| 3303 | |
| 3304 | /* Flush the VMCS */ |
| 3305 | if (vmclear(&vcpu->vc_control_pa)) { |
| 3306 | DPRINTF("%s: vmclear failed\n", __func__); |
| 3307 | ret = EINVAL22; |
| 3308 | } |
| 3309 | atomic_swap_uint(&vcpu->vc_vmx_vmcs_state, VMCS_CLEARED)_atomic_swap_uint((&vcpu->vc_vmx_vmcs_state), (0)); |
| 3310 | |
| 3311 | exit: |
| 3312 | return (ret); |
| 3313 | } |
| 3314 | |
| 3315 | /* |
| 3316 | * vcpu_init_vmx |
| 3317 | * |
| 3318 | * Intel VMX specific VCPU initialization routine. |
| 3319 | * |
| 3320 | * This function allocates various per-VCPU memory regions, sets up initial |
| 3321 | * VCPU VMCS controls, and sets initial register values. |
| 3322 | * |
| 3323 | * Parameters: |
| 3324 | * vcpu: the VCPU structure being initialized |
| 3325 | * |
| 3326 | * Return values: |
| 3327 | * 0: the VCPU was initialized successfully |
| 3328 | * ENOMEM: insufficient resources |
| 3329 | * EINVAL: an error occurred during VCPU initialization |
| 3330 | */ |
| 3331 | int |
| 3332 | vcpu_init_vmx(struct vcpu *vcpu) |
| 3333 | { |
| 3334 | struct vmcs *vmcs; |
| 3335 | uint32_t cr0, cr4; |
| 3336 | int ret = 0; |
| 3337 | |
| 3338 | /* Allocate VMCS VA */ |
| 3339 | vcpu->vc_control_va = (vaddr_t)km_alloc(PAGE_SIZE(1 << 12), &kv_page, &kp_zero, |
| 3340 | &kd_waitok); |
| 3341 | vcpu->vc_vmx_vmcs_state = VMCS_CLEARED0; |
| 3342 | |
| 3343 | if (!vcpu->vc_control_va) |
| 3344 | return (ENOMEM12); |
| 3345 | |
| 3346 | /* Compute VMCS PA */ |
| 3347 | if (!pmap_extract(pmap_kernel()(&kernel_pmap_store), vcpu->vc_control_va, |
| 3348 | (paddr_t *)&vcpu->vc_control_pa)) { |
| 3349 | ret = ENOMEM12; |
| 3350 | goto exit; |
| 3351 | } |
| 3352 | |
| 3353 | /* Allocate MSR bitmap VA */ |
| 3354 | vcpu->vc_msr_bitmap_va = (vaddr_t)km_alloc(PAGE_SIZE(1 << 12), &kv_page, &kp_zero, |
| 3355 | &kd_waitok); |
| 3356 | |
| 3357 | if (!vcpu->vc_msr_bitmap_va) { |
| 3358 | ret = ENOMEM12; |
| 3359 | goto exit; |
| 3360 | } |
| 3361 | |
| 3362 | /* Compute MSR bitmap PA */ |
| 3363 | if (!pmap_extract(pmap_kernel()(&kernel_pmap_store), vcpu->vc_msr_bitmap_va, |
| 3364 | (paddr_t *)&vcpu->vc_msr_bitmap_pa)) { |
| 3365 | ret = ENOMEM12; |
| 3366 | goto exit; |
| 3367 | } |
| 3368 | |
| 3369 | /* Allocate MSR exit load area VA */ |
| 3370 | vcpu->vc_vmx_msr_exit_load_va = (vaddr_t)km_alloc(PAGE_SIZE(1 << 12), &kv_page, |
| 3371 | &kp_zero, &kd_waitok); |
| 3372 | |
| 3373 | if (!vcpu->vc_vmx_msr_exit_load_va) { |
| 3374 | ret = ENOMEM12; |
| 3375 | goto exit; |
| 3376 | } |
| 3377 | |
| 3378 | /* Compute MSR exit load area PA */ |
| 3379 | if (!pmap_extract(pmap_kernel()(&kernel_pmap_store), vcpu->vc_vmx_msr_exit_load_va, |
| 3380 | &vcpu->vc_vmx_msr_exit_load_pa)) { |
| 3381 | ret = ENOMEM12; |
| 3382 | goto exit; |
| 3383 | } |
| 3384 | |
| 3385 | /* Allocate MSR exit save area VA */ |
| 3386 | vcpu->vc_vmx_msr_exit_save_va = (vaddr_t)km_alloc(PAGE_SIZE(1 << 12), &kv_page, |
| 3387 | &kp_zero, &kd_waitok); |
| 3388 | |
| 3389 | if (!vcpu->vc_vmx_msr_exit_save_va) { |
| 3390 | ret = ENOMEM12; |
| 3391 | goto exit; |
| 3392 | } |
| 3393 | |
| 3394 | /* Compute MSR exit save area PA */ |
| 3395 | if (!pmap_extract(pmap_kernel()(&kernel_pmap_store), vcpu->vc_vmx_msr_exit_save_va, |
| 3396 | &vcpu->vc_vmx_msr_exit_save_pa)) { |
| 3397 | ret = ENOMEM12; |
| 3398 | goto exit; |
| 3399 | } |
| 3400 | |
| 3401 | /* Allocate MSR entry load area VA */ |
| 3402 | vcpu->vc_vmx_msr_entry_load_va = (vaddr_t)km_alloc(PAGE_SIZE(1 << 12), &kv_page, |
| 3403 | &kp_zero, &kd_waitok); |
| 3404 | |
| 3405 | if (!vcpu->vc_vmx_msr_entry_load_va) { |
| 3406 | ret = ENOMEM12; |
| 3407 | goto exit; |
| 3408 | } |
| 3409 | |
| 3410 | /* Compute MSR entry load area PA */ |
| 3411 | if (!pmap_extract(pmap_kernel()(&kernel_pmap_store), vcpu->vc_vmx_msr_entry_load_va, |
| 3412 | &vcpu->vc_vmx_msr_entry_load_pa)) { |
| 3413 | ret = ENOMEM12; |
| 3414 | goto exit; |
| 3415 | } |
| 3416 | |
| 3417 | vmcs = (struct vmcs *)vcpu->vc_control_va; |
| 3418 | vmcs->vmcs_revision = curcpu()({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_vmm_cap.vcc_vmx.vmx_vmxon_revision; |
| 3419 | |
| 3420 | /* |
| 3421 | * Load the VMCS onto this PCPU so we can write registers |
| 3422 | */ |
| 3423 | if (vmptrld(&vcpu->vc_control_pa)) { |
| 3424 | ret = EINVAL22; |
| 3425 | goto exit; |
| 3426 | } |
| 3427 | |
| 3428 | /* Host CR0 */ |
| 3429 | cr0 = rcr0() & ~CR0_TS0x00000008; |
| 3430 | if (vmwrite(VMCS_HOST_IA32_CR00x6C00, cr0)) { |
| 3431 | DPRINTF("%s: error writing host CR0\n", __func__); |
| 3432 | ret = EINVAL22; |
| 3433 | goto exit; |
| 3434 | } |
| 3435 | |
| 3436 | /* Host CR4 */ |
| 3437 | cr4 = rcr4(); |
| 3438 | if (vmwrite(VMCS_HOST_IA32_CR40x6C04, cr4)) { |
| 3439 | DPRINTF("%s: error writing host CR4\n", __func__); |
| 3440 | ret = EINVAL22; |
| 3441 | goto exit; |
| 3442 | } |
| 3443 | |
| 3444 | /* Host Segment Selectors */ |
| 3445 | if (vmwrite(VMCS_HOST_IA32_CS_SEL0x0C02, GSEL(GCODE_SEL, SEL_KPL)(((1) << 3) | 0))) { |
| 3446 | DPRINTF("%s: error writing host CS selector\n", __func__); |
| 3447 | ret = EINVAL22; |
| 3448 | goto exit; |
| 3449 | } |
| 3450 | |
| 3451 | if (vmwrite(VMCS_HOST_IA32_DS_SEL0x0C06, GSEL(GDATA_SEL, SEL_KPL)(((2) << 3) | 0))) { |
| 3452 | DPRINTF("%s: error writing host DS selector\n", __func__); |
| 3453 | ret = EINVAL22; |
| 3454 | goto exit; |
| 3455 | } |
| 3456 | |
| 3457 | if (vmwrite(VMCS_HOST_IA32_ES_SEL0x0C00, GSEL(GDATA_SEL, SEL_KPL)(((2) << 3) | 0))) { |
| 3458 | DPRINTF("%s: error writing host ES selector\n", __func__); |
| 3459 | ret = EINVAL22; |
| 3460 | goto exit; |
| 3461 | } |
| 3462 | |
| 3463 | if (vmwrite(VMCS_HOST_IA32_FS_SEL0x0C08, GSEL(GDATA_SEL, SEL_KPL)(((2) << 3) | 0))) { |
| 3464 | DPRINTF("%s: error writing host FS selector\n", __func__); |
| 3465 | ret = EINVAL22; |
| 3466 | goto exit; |
| 3467 | } |
| 3468 | |
| 3469 | if (vmwrite(VMCS_HOST_IA32_GS_SEL0x0C0A, GSEL(GDATA_SEL, SEL_KPL)(((2) << 3) | 0))) { |
| 3470 | DPRINTF("%s: error writing host GS selector\n", __func__); |
| 3471 | ret = EINVAL22; |
| 3472 | goto exit; |
| 3473 | } |
| 3474 | |
| 3475 | if (vmwrite(VMCS_HOST_IA32_SS_SEL0x0C04, GSEL(GDATA_SEL, SEL_KPL)(((2) << 3) | 0))) { |
| 3476 | DPRINTF("%s: error writing host SS selector\n", __func__); |
| 3477 | ret = EINVAL22; |
| 3478 | goto exit; |
| 3479 | } |
| 3480 | |
| 3481 | if (vmwrite(VMCS_HOST_IA32_TR_SEL0x0C0C, GSYSSEL(GPROC0_SEL, SEL_KPL)((((0) << 4) + (6 << 3)) | 0))) { |
| 3482 | DPRINTF("%s: error writing host TR selector\n", __func__); |
| 3483 | ret = EINVAL22; |
| 3484 | goto exit; |
| 3485 | } |
| 3486 | |
| 3487 | /* Host IDTR base */ |
| 3488 | if (vmwrite(VMCS_HOST_IA32_IDTR_BASE0x6C0E, idt_vaddr)) { |
| 3489 | DPRINTF("%s: error writing host IDTR base\n", __func__); |
| 3490 | ret = EINVAL22; |
| 3491 | goto exit; |
| 3492 | } |
| 3493 | |
| 3494 | /* VMCS link */ |
| 3495 | if (vmwrite(VMCS_LINK_POINTER0x2800, VMX_VMCS_PA_CLEAR0xFFFFFFFFFFFFFFFFUL)) { |
| 3496 | DPRINTF("%s: error writing VMCS link pointer\n", __func__); |
| 3497 | ret = EINVAL22; |
| 3498 | goto exit; |
| 3499 | } |
| 3500 | |
| 3501 | /* Flush the initial VMCS */ |
| 3502 | if (vmclear(&vcpu->vc_control_pa)) { |
| 3503 | DPRINTF("%s: vmclear failed\n", __func__); |
| 3504 | ret = EINVAL22; |
| 3505 | } |
| 3506 | |
| 3507 | exit: |
| 3508 | if (ret) { |
| 3509 | if (vcpu->vc_control_va) |
| 3510 | km_free((void *)vcpu->vc_control_va, PAGE_SIZE(1 << 12), |
| 3511 | &kv_page, &kp_zero); |
| 3512 | if (vcpu->vc_msr_bitmap_va) |
| 3513 | km_free((void *)vcpu->vc_msr_bitmap_va, PAGE_SIZE(1 << 12), |
| 3514 | &kv_page, &kp_zero); |
| 3515 | if (vcpu->vc_vmx_msr_exit_save_va) |
| 3516 | km_free((void *)vcpu->vc_vmx_msr_exit_save_va, |
| 3517 | PAGE_SIZE(1 << 12), &kv_page, &kp_zero); |
| 3518 | if (vcpu->vc_vmx_msr_exit_load_va) |
| 3519 | km_free((void *)vcpu->vc_vmx_msr_exit_load_va, |
| 3520 | PAGE_SIZE(1 << 12), &kv_page, &kp_zero); |
| 3521 | if (vcpu->vc_vmx_msr_entry_load_va) |
| 3522 | km_free((void *)vcpu->vc_vmx_msr_entry_load_va, |
| 3523 | PAGE_SIZE(1 << 12), &kv_page, &kp_zero); |
| 3524 | } |
| 3525 | |
| 3526 | return (ret); |
| 3527 | } |
| 3528 | |
| 3529 | /* |
| 3530 | * vcpu_reset_regs |
| 3531 | * |
| 3532 | * Resets a vcpu's registers to the provided state |
| 3533 | * |
| 3534 | * Parameters: |
| 3535 | * vcpu: the vcpu whose registers shall be reset |
| 3536 | * vrs: the desired register state |
| 3537 | * |
| 3538 | * Return values: |
| 3539 | * 0: the vcpu's registers were successfully reset |
| 3540 | * !0: the vcpu's registers could not be reset (see arch-specific reset |
| 3541 | * function for various values that can be returned here) |
| 3542 | */ |
| 3543 | int |
| 3544 | vcpu_reset_regs(struct vcpu *vcpu, struct vcpu_reg_state *vrs) |
| 3545 | { |
| 3546 | int ret; |
| 3547 | |
| 3548 | if (vmm_softc->mode == VMM_MODE_VMX || |
| 3549 | vmm_softc->mode == VMM_MODE_EPT) |
| 3550 | ret = vcpu_reset_regs_vmx(vcpu, vrs); |
| 3551 | else if (vmm_softc->mode == VMM_MODE_SVM || |
| 3552 | vmm_softc->mode == VMM_MODE_RVI) |
| 3553 | ret = vcpu_reset_regs_svm(vcpu, vrs); |
| 3554 | else |
| 3555 | panic("%s: unknown vmm mode: %d", __func__, vmm_softc->mode); |
| 3556 | |
| 3557 | return (ret); |
| 3558 | } |
| 3559 | |
| 3560 | /* |
| 3561 | * vcpu_init_svm |
| 3562 | * |
| 3563 | * AMD SVM specific VCPU initialization routine. |
| 3564 | * |
| 3565 | * This function allocates various per-VCPU memory regions, sets up initial |
| 3566 | * VCPU VMCB controls, and sets initial register values. |
| 3567 | * |
| 3568 | * Parameters: |
| 3569 | * vcpu: the VCPU structure being initialized |
| 3570 | * |
| 3571 | * Return values: |
| 3572 | * 0: the VCPU was initialized successfully |
| 3573 | * ENOMEM: insufficient resources |
| 3574 | * EINVAL: an error occurred during VCPU initialization |
| 3575 | */ |
| 3576 | int |
| 3577 | vcpu_init_svm(struct vcpu *vcpu) |
| 3578 | { |
| 3579 | int ret = 0; |
| 3580 | |
| 3581 | /* Allocate VMCB VA */ |
| 3582 | vcpu->vc_control_va = (vaddr_t)km_alloc(PAGE_SIZE(1 << 12), &kv_page, &kp_zero, |
| 3583 | &kd_waitok); |
| 3584 | |
| 3585 | if (!vcpu->vc_control_va) |
| 3586 | return (ENOMEM12); |
| 3587 | |
| 3588 | /* Compute VMCB PA */ |
| 3589 | if (!pmap_extract(pmap_kernel()(&kernel_pmap_store), vcpu->vc_control_va, |
| 3590 | (paddr_t *)&vcpu->vc_control_pa)) { |
| 3591 | ret = ENOMEM12; |
| 3592 | goto exit; |
| 3593 | } |
| 3594 | |
| 3595 | DPRINTF("%s: VMCB va @ 0x%llx, pa @ 0x%llx\n", __func__, |
| 3596 | (uint64_t)vcpu->vc_control_va, |
| 3597 | (uint64_t)vcpu->vc_control_pa); |
| 3598 | |
| 3599 | |
| 3600 | /* Allocate MSR bitmap VA (2 pages) */ |
| 3601 | vcpu->vc_msr_bitmap_va = (vaddr_t)km_alloc(2 * PAGE_SIZE(1 << 12), &kv_any, |
| 3602 | &vmm_kp_contig, &kd_waitok); |
| 3603 | |
| 3604 | if (!vcpu->vc_msr_bitmap_va) { |
| 3605 | ret = ENOMEM12; |
| 3606 | goto exit; |
| 3607 | } |
| 3608 | |
| 3609 | /* Compute MSR bitmap PA */ |
| 3610 | if (!pmap_extract(pmap_kernel()(&kernel_pmap_store), vcpu->vc_msr_bitmap_va, |
| 3611 | (paddr_t *)&vcpu->vc_msr_bitmap_pa)) { |
| 3612 | ret = ENOMEM12; |
| 3613 | goto exit; |
| 3614 | } |
| 3615 | |
| 3616 | DPRINTF("%s: MSR bitmap va @ 0x%llx, pa @ 0x%llx\n", __func__, |
| 3617 | (uint64_t)vcpu->vc_msr_bitmap_va, |
| 3618 | (uint64_t)vcpu->vc_msr_bitmap_pa); |
| 3619 | |
| 3620 | /* Allocate host state area VA */ |
| 3621 | vcpu->vc_svm_hsa_va = (vaddr_t)km_alloc(PAGE_SIZE(1 << 12), &kv_page, |
| 3622 | &kp_zero, &kd_waitok); |
| 3623 | |
| 3624 | if (!vcpu->vc_svm_hsa_va) { |
| 3625 | ret = ENOMEM12; |
| 3626 | goto exit; |
| 3627 | } |
| 3628 | |
| 3629 | /* Compute host state area PA */ |
| 3630 | if (!pmap_extract(pmap_kernel()(&kernel_pmap_store), vcpu->vc_svm_hsa_va, |
| 3631 | &vcpu->vc_svm_hsa_pa)) { |
| 3632 | ret = ENOMEM12; |
| 3633 | goto exit; |
| 3634 | } |
| 3635 | |
| 3636 | DPRINTF("%s: HSA va @ 0x%llx, pa @ 0x%llx\n", __func__, |
| 3637 | (uint64_t)vcpu->vc_svm_hsa_va, |
| 3638 | (uint64_t)vcpu->vc_svm_hsa_pa); |
| 3639 | |
| 3640 | /* Allocate IOIO area VA (3 pages) */ |
| 3641 | vcpu->vc_svm_ioio_va = (vaddr_t)km_alloc(3 * PAGE_SIZE(1 << 12), &kv_any, |
| 3642 | &vmm_kp_contig, &kd_waitok); |
| 3643 | |
| 3644 | if (!vcpu->vc_svm_ioio_va) { |
| 3645 | ret = ENOMEM12; |
| 3646 | goto exit; |
| 3647 | } |
| 3648 | |
| 3649 | /* Compute IOIO area PA */ |
| 3650 | if (!pmap_extract(pmap_kernel()(&kernel_pmap_store), vcpu->vc_svm_ioio_va, |
| 3651 | &vcpu->vc_svm_ioio_pa)) { |
| 3652 | ret = ENOMEM12; |
| 3653 | goto exit; |
| 3654 | } |
| 3655 | |
| 3656 | DPRINTF("%s: IOIO va @ 0x%llx, pa @ 0x%llx\n", __func__, |
| 3657 | (uint64_t)vcpu->vc_svm_ioio_va, |
| 3658 | (uint64_t)vcpu->vc_svm_ioio_pa); |
| 3659 | |
| 3660 | exit: |
| 3661 | if (ret) { |
| 3662 | if (vcpu->vc_control_va) |
| 3663 | km_free((void *)vcpu->vc_control_va, PAGE_SIZE(1 << 12), |
| 3664 | &kv_page, &kp_zero); |
| 3665 | if (vcpu->vc_msr_bitmap_va) |
| 3666 | km_free((void *)vcpu->vc_msr_bitmap_va, 2 * PAGE_SIZE(1 << 12), |
| 3667 | &kv_any, &vmm_kp_contig); |
| 3668 | if (vcpu->vc_svm_hsa_va) |
| 3669 | km_free((void *)vcpu->vc_svm_hsa_va, PAGE_SIZE(1 << 12), |
| 3670 | &kv_page, &kp_zero); |
| 3671 | if (vcpu->vc_svm_ioio_va) |
| 3672 | km_free((void *)vcpu->vc_svm_ioio_va, |
| 3673 | 3 * PAGE_SIZE(1 << 12), &kv_any, &vmm_kp_contig); |
| 3674 | } |
| 3675 | |
| 3676 | return (ret); |
| 3677 | } |
| 3678 | |
| 3679 | /* |
| 3680 | * vcpu_init |
| 3681 | * |
| 3682 | * Calls the architecture-specific VCPU init routine |
| 3683 | */ |
| 3684 | int |
| 3685 | vcpu_init(struct vcpu *vcpu) |
| 3686 | { |
| 3687 | int ret = 0; |
| 3688 | |
| 3689 | vcpu->vc_virt_mode = vmm_softc->mode; |
| 3690 | vcpu->vc_state = VCPU_STATE_STOPPED; |
| 3691 | vcpu->vc_vpid = 0; |
| 3692 | vcpu->vc_pvclock_system_gpa = 0; |
| 3693 | vcpu->vc_last_pcpu = NULL((void *)0); |
| 3694 | |
| 3695 | rw_init(&vcpu->vc_lock, "vcpu")_rw_init_flags(&vcpu->vc_lock, "vcpu", 0, ((void *)0)); |
| 3696 | |
| 3697 | /* Shadow PAT MSR, starting with host's value. */ |
| 3698 | vcpu->vc_shadow_pat = rdmsr(MSR_CR_PAT0x277); |
| 3699 | |
| 3700 | if (vmm_softc->mode == VMM_MODE_VMX || |
| 3701 | vmm_softc->mode == VMM_MODE_EPT) |
| 3702 | ret = vcpu_init_vmx(vcpu); |
| 3703 | else if (vmm_softc->mode == VMM_MODE_SVM || |
| 3704 | vmm_softc->mode == VMM_MODE_RVI) |
| 3705 | ret = vcpu_init_svm(vcpu); |
| 3706 | else |
| 3707 | panic("%s: unknown vmm mode: %d", __func__, vmm_softc->mode); |
| 3708 | |
| 3709 | return (ret); |
| 3710 | } |
| 3711 | |
| 3712 | /* |
| 3713 | * vcpu_deinit_vmx |
| 3714 | * |
| 3715 | * Deinitializes the vcpu described by 'vcpu' |
| 3716 | * |
| 3717 | * Parameters: |
| 3718 | * vcpu: the vcpu to be deinited |
| 3719 | */ |
| 3720 | void |
| 3721 | vcpu_deinit_vmx(struct vcpu *vcpu) |
| 3722 | { |
| 3723 | if (vcpu->vc_control_va) |
| 3724 | km_free((void *)vcpu->vc_control_va, PAGE_SIZE(1 << 12), |
| 3725 | &kv_page, &kp_zero); |
| 3726 | if (vcpu->vc_vmx_msr_exit_save_va) |
| 3727 | km_free((void *)vcpu->vc_vmx_msr_exit_save_va, |
| 3728 | PAGE_SIZE(1 << 12), &kv_page, &kp_zero); |
| 3729 | if (vcpu->vc_vmx_msr_exit_load_va) |
| 3730 | km_free((void *)vcpu->vc_vmx_msr_exit_load_va, |
| 3731 | PAGE_SIZE(1 << 12), &kv_page, &kp_zero); |
| 3732 | if (vcpu->vc_vmx_msr_entry_load_va) |
| 3733 | km_free((void *)vcpu->vc_vmx_msr_entry_load_va, |
| 3734 | PAGE_SIZE(1 << 12), &kv_page, &kp_zero); |
| 3735 | |
| 3736 | if (vcpu->vc_vmx_vpid_enabled) |
| 3737 | vmm_free_vpid(vcpu->vc_vpid); |
| 3738 | } |
| 3739 | |
| 3740 | /* |
| 3741 | * vcpu_deinit_svm |
| 3742 | * |
| 3743 | * Deinitializes the vcpu described by 'vcpu' |
| 3744 | * |
| 3745 | * Parameters: |
| 3746 | * vcpu: the vcpu to be deinited |
| 3747 | */ |
| 3748 | void |
| 3749 | vcpu_deinit_svm(struct vcpu *vcpu) |
| 3750 | { |
| 3751 | if (vcpu->vc_control_va) |
| 3752 | km_free((void *)vcpu->vc_control_va, PAGE_SIZE(1 << 12), &kv_page, |
| 3753 | &kp_zero); |
| 3754 | if (vcpu->vc_msr_bitmap_va) |
| 3755 | km_free((void *)vcpu->vc_msr_bitmap_va, 2 * PAGE_SIZE(1 << 12), &kv_any, |
| 3756 | &vmm_kp_contig); |
| 3757 | if (vcpu->vc_svm_hsa_va) |
| 3758 | km_free((void *)vcpu->vc_svm_hsa_va, PAGE_SIZE(1 << 12), &kv_page, |
| 3759 | &kp_zero); |
| 3760 | if (vcpu->vc_svm_ioio_va) |
| 3761 | km_free((void *)vcpu->vc_svm_ioio_va, 3 * PAGE_SIZE(1 << 12), &kv_any, |
| 3762 | &vmm_kp_contig); |
| 3763 | |
| 3764 | vmm_free_vpid(vcpu->vc_vpid); |
| 3765 | } |
| 3766 | |
| 3767 | /* |
| 3768 | * vcpu_deinit |
| 3769 | * |
| 3770 | * Calls the architecture-specific VCPU deinit routine |
| 3771 | * |
| 3772 | * Parameters: |
| 3773 | * vcpu: the vcpu to be deinited |
| 3774 | */ |
| 3775 | void |
| 3776 | vcpu_deinit(struct vcpu *vcpu) |
| 3777 | { |
| 3778 | if (vmm_softc->mode == VMM_MODE_VMX || |
| 3779 | vmm_softc->mode == VMM_MODE_EPT) |
| 3780 | vcpu_deinit_vmx(vcpu); |
| 3781 | else if (vmm_softc->mode == VMM_MODE_SVM || |
| 3782 | vmm_softc->mode == VMM_MODE_RVI) |
| 3783 | vcpu_deinit_svm(vcpu); |
| 3784 | else |
| 3785 | panic("%s: unknown vmm mode: %d", __func__, vmm_softc->mode); |
| 3786 | } |
| 3787 | |
| 3788 | /* |
| 3789 | * vm_teardown |
| 3790 | * |
| 3791 | * Tears down (destroys) the vm indicated by 'vm'. |
| 3792 | * |
| 3793 | * Parameters: |
| 3794 | * vm: vm to be torn down |
| 3795 | */ |
| 3796 | void |
| 3797 | vm_teardown(struct vm *vm) |
| 3798 | { |
| 3799 | struct vcpu *vcpu, *tmp; |
| 3800 | |
| 3801 | rw_assert_wrlock(&vmm_softc->vm_lock); |
| 3802 | KERNEL_LOCK()_kernel_lock(); |
| 3803 | |
| 3804 | /* Free VCPUs */ |
| 3805 | rw_enter_write(&vm->vm_vcpu_lock); |
| 3806 | SLIST_FOREACH_SAFE(vcpu, &vm->vm_vcpu_list, vc_vcpu_link, tmp)for ((vcpu) = ((&vm->vm_vcpu_list)->slh_first); (vcpu ) && ((tmp) = ((vcpu)->vc_vcpu_link.sle_next), 1); (vcpu) = (tmp)) { |
| 3807 | SLIST_REMOVE(&vm->vm_vcpu_list, vcpu, vcpu, vc_vcpu_link)do { if ((&vm->vm_vcpu_list)->slh_first == (vcpu)) { do { ((&vm->vm_vcpu_list))->slh_first = ((&vm-> vm_vcpu_list))->slh_first->vc_vcpu_link.sle_next; } while (0); } else { struct vcpu *curelm = (&vm->vm_vcpu_list )->slh_first; while (curelm->vc_vcpu_link.sle_next != ( vcpu)) curelm = curelm->vc_vcpu_link.sle_next; curelm-> vc_vcpu_link.sle_next = curelm->vc_vcpu_link.sle_next-> vc_vcpu_link.sle_next; } ((vcpu)->vc_vcpu_link.sle_next) = ((void *)-1); } while (0); |
| 3808 | vcpu_deinit(vcpu); |
| 3809 | pool_put(&vcpu_pool, vcpu); |
| 3810 | vmm_softc->vcpu_ct--; |
| 3811 | } |
| 3812 | |
| 3813 | vm_impl_deinit(vm); |
| 3814 | |
| 3815 | /* teardown guest vmspace */ |
| 3816 | if (vm->vm_vmspace != NULL((void *)0)) { |
| 3817 | uvmspace_free(vm->vm_vmspace); |
| 3818 | vm->vm_vmspace = NULL((void *)0); |
| 3819 | } |
| 3820 | |
| 3821 | if (vm->vm_id > 0) { |
| 3822 | vmm_softc->vm_ct--; |
| 3823 | if (vmm_softc->vm_ct < 1) |
| 3824 | vmm_stop(); |
| 3825 | } |
| 3826 | pool_put(&vm_pool, vm); |
| 3827 | |
| 3828 | KERNEL_UNLOCK()_kernel_unlock(); |
| 3829 | rw_exit_write(&vm->vm_vcpu_lock); |
| 3830 | } |
| 3831 | |
| 3832 | /* |
| 3833 | * vcpu_vmx_check_cap |
| 3834 | * |
| 3835 | * Checks if the 'cap' bit in the 'msr' MSR can be set or cleared (set = 1 |
| 3836 | * or set = 0, respectively). |
| 3837 | * |
| 3838 | * When considering 'msr', we check to see if true controls are available, |
| 3839 | * and use those if so. |
| 3840 | * |
| 3841 | * Returns 1 of 'cap' can be set/cleared as requested, 0 otherwise. |
| 3842 | */ |
| 3843 | int |
| 3844 | vcpu_vmx_check_cap(struct vcpu *vcpu, uint32_t msr, uint32_t cap, int set) |
| 3845 | { |
| 3846 | uint64_t ctl; |
| 3847 | |
| 3848 | if (vcpu->vc_vmx_basic & IA32_VMX_TRUE_CTLS_AVAIL(1ULL << 55)) { |
| 3849 | switch (msr) { |
| 3850 | case IA32_VMX_PINBASED_CTLS0x481: |
| 3851 | ctl = vcpu->vc_vmx_true_pinbased_ctls; |
| 3852 | break; |
| 3853 | case IA32_VMX_PROCBASED_CTLS0x482: |
| 3854 | ctl = vcpu->vc_vmx_true_procbased_ctls; |
| 3855 | break; |
| 3856 | case IA32_VMX_PROCBASED2_CTLS0x48B: |
| 3857 | ctl = vcpu->vc_vmx_procbased2_ctls; |
| 3858 | break; |
| 3859 | case IA32_VMX_ENTRY_CTLS0x484: |
| 3860 | ctl = vcpu->vc_vmx_true_entry_ctls; |
| 3861 | break; |
| 3862 | case IA32_VMX_EXIT_CTLS0x483: |
| 3863 | ctl = vcpu->vc_vmx_true_exit_ctls; |
| 3864 | break; |
| 3865 | default: |
| 3866 | return (0); |
| 3867 | } |
| 3868 | } else { |
| 3869 | switch (msr) { |
| 3870 | case IA32_VMX_PINBASED_CTLS0x481: |
| 3871 | ctl = vcpu->vc_vmx_pinbased_ctls; |
| 3872 | break; |
| 3873 | case IA32_VMX_PROCBASED_CTLS0x482: |
| 3874 | ctl = vcpu->vc_vmx_procbased_ctls; |
| 3875 | break; |
| 3876 | case IA32_VMX_PROCBASED2_CTLS0x48B: |
| 3877 | ctl = vcpu->vc_vmx_procbased2_ctls; |
| 3878 | break; |
| 3879 | case IA32_VMX_ENTRY_CTLS0x484: |
| 3880 | ctl = vcpu->vc_vmx_entry_ctls; |
| 3881 | break; |
| 3882 | case IA32_VMX_EXIT_CTLS0x483: |
| 3883 | ctl = vcpu->vc_vmx_exit_ctls; |
| 3884 | break; |
| 3885 | default: |
| 3886 | return (0); |
| 3887 | } |
| 3888 | } |
| 3889 | |
| 3890 | if (set) { |
| 3891 | /* Check bit 'cap << 32', must be !0 */ |
| 3892 | return (ctl & ((uint64_t)cap << 32)) != 0; |
| 3893 | } else { |
| 3894 | /* Check bit 'cap', must be 0 */ |
| 3895 | return (ctl & cap) == 0; |
| 3896 | } |
| 3897 | } |
| 3898 | |
| 3899 | /* |
| 3900 | * vcpu_vmx_compute_ctrl |
| 3901 | * |
| 3902 | * Computes the appropriate control value, given the supplied parameters |
| 3903 | * and CPU capabilities. |
| 3904 | * |
| 3905 | * Intel has made somewhat of a mess of this computation - it is described |
| 3906 | * using no fewer than three different approaches, spread across many |
| 3907 | * pages of the SDM. Further compounding the problem is the fact that now |
| 3908 | * we have "true controls" for each type of "control", and each needs to |
| 3909 | * be examined to get the calculation right, but only if "true" controls |
| 3910 | * are present on the CPU we're on. |
| 3911 | * |
| 3912 | * Parameters: |
| 3913 | * ctrlval: the control value, as read from the CPU MSR |
| 3914 | * ctrl: which control is being set (eg, pinbased, procbased, etc) |
| 3915 | * want0: the set of desired 0 bits |
| 3916 | * want1: the set of desired 1 bits |
| 3917 | * out: (out) the correct value to write into the VMCS for this VCPU, |
| 3918 | * for the 'ctrl' desired. |
| 3919 | * |
| 3920 | * Returns 0 if successful, or EINVAL if the supplied parameters define |
| 3921 | * an unworkable control setup. |
| 3922 | */ |
| 3923 | int |
| 3924 | vcpu_vmx_compute_ctrl(uint64_t ctrlval, uint16_t ctrl, uint32_t want1, |
| 3925 | uint32_t want0, uint32_t *out) |
| 3926 | { |
| 3927 | int i, set, clear; |
| 3928 | |
| 3929 | *out = 0; |
| 3930 | |
| 3931 | /* |
| 3932 | * The Intel SDM gives three formulae for determining which bits to |
| 3933 | * set/clear for a given control and desired functionality. Formula |
| 3934 | * 1 is the simplest but disallows use of newer features that are |
| 3935 | * enabled by functionality in later CPUs. |
| 3936 | * |
| 3937 | * Formulas 2 and 3 allow such extra functionality. We use formula |
| 3938 | * 2 - this requires us to know the identity of controls in the |
| 3939 | * "default1" class for each control register, but allows us to not |
| 3940 | * have to pass along and/or query both sets of capability MSRs for |
| 3941 | * each control lookup. This makes the code slightly longer, |
| 3942 | * however. |
| 3943 | */ |
| 3944 | for (i = 0; i < 32; i++) { |
| 3945 | /* Figure out if we can set and / or clear this bit */ |
| 3946 | set = (ctrlval & (1ULL << (i + 32))) != 0; |
| 3947 | clear = ((1ULL << i) & ((uint64_t)ctrlval)) == 0; |
| 3948 | |
| 3949 | /* If the bit can't be set nor cleared, something's wrong */ |
| 3950 | if (!set && !clear) |
| 3951 | return (EINVAL22); |
| 3952 | |
| 3953 | /* |
| 3954 | * Formula 2.c.i - "If the relevant VMX capability MSR |
| 3955 | * reports that a control has a single setting, use that |
| 3956 | * setting." |
| 3957 | */ |
| 3958 | if (set && !clear) { |
| 3959 | if (want0 & (1ULL << i)) |
| 3960 | return (EINVAL22); |
| 3961 | else |
| 3962 | *out |= (1ULL << i); |
| 3963 | } else if (clear && !set) { |
| 3964 | if (want1 & (1ULL << i)) |
| 3965 | return (EINVAL22); |
| 3966 | else |
| 3967 | *out &= ~(1ULL << i); |
| 3968 | } else { |
| 3969 | /* |
| 3970 | * 2.c.ii - "If the relevant VMX capability MSR |
| 3971 | * reports that a control can be set to 0 or 1 |
| 3972 | * and that control's meaning is known to the VMM, |
| 3973 | * set the control based on the functionality desired." |
| 3974 | */ |
| 3975 | if (want1 & (1ULL << i)) |
| 3976 | *out |= (1ULL << i); |
| 3977 | else if (want0 & (1 << i)) |
| 3978 | *out &= ~(1ULL << i); |
| 3979 | else { |
| 3980 | /* |
| 3981 | * ... assuming the control's meaning is not |
| 3982 | * known to the VMM ... |
| 3983 | * |
| 3984 | * 2.c.iii - "If the relevant VMX capability |
| 3985 | * MSR reports that a control can be set to 0 |
| 3986 | * or 1 and the control is not in the default1 |
| 3987 | * class, set the control to 0." |
| 3988 | * |
| 3989 | * 2.c.iv - "If the relevant VMX capability |
| 3990 | * MSR reports that a control can be set to 0 |
| 3991 | * or 1 and the control is in the default1 |
| 3992 | * class, set the control to 1." |
| 3993 | */ |
| 3994 | switch (ctrl) { |
| 3995 | case IA32_VMX_PINBASED_CTLS0x481: |
| 3996 | case IA32_VMX_TRUE_PINBASED_CTLS0x48D: |
| 3997 | /* |
| 3998 | * A.3.1 - default1 class of pinbased |
| 3999 | * controls comprises bits 1,2,4 |
| 4000 | */ |
| 4001 | switch (i) { |
| 4002 | case 1: |
| 4003 | case 2: |
| 4004 | case 4: |
| 4005 | *out |= (1ULL << i); |
| 4006 | break; |
| 4007 | default: |
| 4008 | *out &= ~(1ULL << i); |
| 4009 | break; |
| 4010 | } |
| 4011 | break; |
| 4012 | case IA32_VMX_PROCBASED_CTLS0x482: |
| 4013 | case IA32_VMX_TRUE_PROCBASED_CTLS0x48E: |
| 4014 | /* |
| 4015 | * A.3.2 - default1 class of procbased |
| 4016 | * controls comprises bits 1, 4-6, 8, |
| 4017 | * 13-16, 26 |
| 4018 | */ |
| 4019 | switch (i) { |
| 4020 | case 1: |
| 4021 | case 4 ... 6: |
| 4022 | case 8: |
| 4023 | case 13 ... 16: |
| 4024 | case 26: |
| 4025 | *out |= (1ULL << i); |
| 4026 | break; |
| 4027 | default: |
| 4028 | *out &= ~(1ULL << i); |
| 4029 | break; |
| 4030 | } |
| 4031 | break; |
| 4032 | /* |
| 4033 | * Unknown secondary procbased controls |
| 4034 | * can always be set to 0 |
| 4035 | */ |
| 4036 | case IA32_VMX_PROCBASED2_CTLS0x48B: |
| 4037 | *out &= ~(1ULL << i); |
| 4038 | break; |
| 4039 | case IA32_VMX_EXIT_CTLS0x483: |
| 4040 | case IA32_VMX_TRUE_EXIT_CTLS0x48F: |
| 4041 | /* |
| 4042 | * A.4 - default1 class of exit |
| 4043 | * controls comprises bits 0-8, 10, |
| 4044 | * 11, 13, 14, 16, 17 |
| 4045 | */ |
| 4046 | switch (i) { |
| 4047 | case 0 ... 8: |
| 4048 | case 10 ... 11: |
| 4049 | case 13 ... 14: |
| 4050 | case 16 ... 17: |
| 4051 | *out |= (1ULL << i); |
| 4052 | break; |
| 4053 | default: |
| 4054 | *out &= ~(1ULL << i); |
| 4055 | break; |
| 4056 | } |
| 4057 | break; |
| 4058 | case IA32_VMX_ENTRY_CTLS0x484: |
| 4059 | case IA32_VMX_TRUE_ENTRY_CTLS0x490: |
| 4060 | /* |
| 4061 | * A.5 - default1 class of entry |
| 4062 | * controls comprises bits 0-8, 12 |
| 4063 | */ |
| 4064 | switch (i) { |
| 4065 | case 0 ... 8: |
| 4066 | case 12: |
| 4067 | *out |= (1ULL << i); |
| 4068 | break; |
| 4069 | default: |
| 4070 | *out &= ~(1ULL << i); |
| 4071 | break; |
| 4072 | } |
| 4073 | break; |
| 4074 | } |
| 4075 | } |
| 4076 | } |
| 4077 | } |
| 4078 | |
| 4079 | return (0); |
| 4080 | } |
| 4081 | |
| 4082 | /* |
| 4083 | * vm_get_info |
| 4084 | * |
| 4085 | * Returns information about the VM indicated by 'vip'. The 'vip_size' field |
| 4086 | * in the 'vip' parameter is used to indicate the size of the caller's buffer. |
| 4087 | * If insufficient space exists in that buffer, the required size needed is |
| 4088 | * returned in vip_size and the number of VM information structures returned |
| 4089 | * in vip_info_count is set to 0. The caller should then try the ioctl again |
| 4090 | * after allocating a sufficiently large buffer. |
| 4091 | * |
| 4092 | * Parameters: |
| 4093 | * vip: information structure identifying the VM to query |
| 4094 | * |
| 4095 | * Return values: |
| 4096 | * 0: the operation succeeded |
| 4097 | * ENOMEM: memory allocation error during processing |
| 4098 | * EFAULT: error copying data to user process |
| 4099 | */ |
| 4100 | int |
| 4101 | vm_get_info(struct vm_info_params *vip) |
| 4102 | { |
| 4103 | struct vm_info_result *out; |
| 4104 | struct vm *vm; |
| 4105 | struct vcpu *vcpu; |
| 4106 | int i, j; |
| 4107 | size_t need; |
| 4108 | |
| 4109 | rw_enter_read(&vmm_softc->vm_lock); |
| 4110 | need = vmm_softc->vm_ct * sizeof(struct vm_info_result); |
| 4111 | if (vip->vip_size < need) { |
| 4112 | vip->vip_info_ct = 0; |
| 4113 | vip->vip_size = need; |
| 4114 | rw_exit_read(&vmm_softc->vm_lock); |
| 4115 | return (0); |
| 4116 | } |
| 4117 | |
| 4118 | out = malloc(need, M_DEVBUF2, M_NOWAIT0x0002|M_ZERO0x0008); |
| 4119 | if (out == NULL((void *)0)) { |
| 4120 | vip->vip_info_ct = 0; |
| 4121 | rw_exit_read(&vmm_softc->vm_lock); |
| 4122 | return (ENOMEM12); |
| 4123 | } |
| 4124 | |
| 4125 | i = 0; |
| 4126 | vip->vip_info_ct = vmm_softc->vm_ct; |
| 4127 | SLIST_FOREACH(vm, &vmm_softc->vm_list, vm_link)for((vm) = ((&vmm_softc->vm_list)->slh_first); (vm) != ((void *)0); (vm) = ((vm)->vm_link.sle_next)) { |
| 4128 | out[i].vir_memory_size = vm->vm_memory_size; |
| 4129 | out[i].vir_used_size = |
| 4130 | pmap_resident_count(vm->vm_map->pmap)((vm->vm_map->pmap)->pm_stats.resident_count) * PAGE_SIZE(1 << 12); |
| 4131 | out[i].vir_ncpus = vm->vm_vcpu_ct; |
| 4132 | out[i].vir_id = vm->vm_id; |
| 4133 | out[i].vir_creator_pid = vm->vm_creator_pid; |
| 4134 | strlcpy(out[i].vir_name, vm->vm_name, VMM_MAX_NAME_LEN64); |
| 4135 | rw_enter_read(&vm->vm_vcpu_lock); |
| 4136 | for (j = 0; j < vm->vm_vcpu_ct; j++) { |
| 4137 | out[i].vir_vcpu_state[j] = VCPU_STATE_UNKNOWN; |
| 4138 | SLIST_FOREACH(vcpu, &vm->vm_vcpu_list,for((vcpu) = ((&vm->vm_vcpu_list)->slh_first); (vcpu ) != ((void *)0); (vcpu) = ((vcpu)->vc_vcpu_link.sle_next) ) |
| 4139 | vc_vcpu_link)for((vcpu) = ((&vm->vm_vcpu_list)->slh_first); (vcpu ) != ((void *)0); (vcpu) = ((vcpu)->vc_vcpu_link.sle_next) ) { |
| 4140 | if (vcpu->vc_id == j) |
| 4141 | out[i].vir_vcpu_state[j] = |
| 4142 | vcpu->vc_state; |
| 4143 | } |
| 4144 | } |
| 4145 | rw_exit_read(&vm->vm_vcpu_lock); |
| 4146 | i++; |
| 4147 | } |
| 4148 | rw_exit_read(&vmm_softc->vm_lock); |
| 4149 | if (copyout(out, vip->vip_info, need) == EFAULT14) { |
| 4150 | free(out, M_DEVBUF2, need); |
| 4151 | return (EFAULT14); |
| 4152 | } |
| 4153 | |
| 4154 | free(out, M_DEVBUF2, need); |
| 4155 | return (0); |
| 4156 | } |
| 4157 | |
| 4158 | /* |
| 4159 | * vm_terminate |
| 4160 | * |
| 4161 | * Terminates the VM indicated by 'vtp'. |
| 4162 | * |
| 4163 | * Parameters: |
| 4164 | * vtp: structure defining the VM to terminate |
| 4165 | * |
| 4166 | * Return values: |
| 4167 | * 0: the VM was terminated |
| 4168 | * !0: the VM could not be located |
| 4169 | */ |
| 4170 | int |
| 4171 | vm_terminate(struct vm_terminate_params *vtp) |
| 4172 | { |
| 4173 | struct vm *vm; |
| 4174 | struct vcpu *vcpu; |
| 4175 | u_int old, next; |
| 4176 | int error; |
| 4177 | |
| 4178 | /* |
| 4179 | * Find desired VM |
| 4180 | */ |
| 4181 | rw_enter_write(&vmm_softc->vm_lock); |
| 4182 | error = vm_find(vtp->vtp_vm_id, &vm); |
| 4183 | |
| 4184 | if (error == 0) { |
| 4185 | rw_enter_read(&vm->vm_vcpu_lock); |
| 4186 | SLIST_FOREACH(vcpu, &vm->vm_vcpu_list, vc_vcpu_link)for((vcpu) = ((&vm->vm_vcpu_list)->slh_first); (vcpu ) != ((void *)0); (vcpu) = ((vcpu)->vc_vcpu_link.sle_next) ) { |
| 4187 | do { |
| 4188 | old = vcpu->vc_state; |
| 4189 | if (old == VCPU_STATE_RUNNING) |
| 4190 | next = VCPU_STATE_REQTERM; |
| 4191 | else if (old == VCPU_STATE_STOPPED) |
| 4192 | next = VCPU_STATE_TERMINATED; |
| 4193 | else /* must be REQTERM or TERMINATED */ |
| 4194 | break; |
| 4195 | } while (old != atomic_cas_uint(&vcpu->vc_state,_atomic_cas_uint((&vcpu->vc_state), (old), (next)) |
| 4196 | old, next)_atomic_cas_uint((&vcpu->vc_state), (old), (next))); |
| 4197 | } |
| 4198 | rw_exit_read(&vm->vm_vcpu_lock); |
| 4199 | } else { |
| 4200 | rw_exit_write(&vmm_softc->vm_lock); |
| 4201 | return (error); |
| 4202 | } |
| 4203 | |
| 4204 | SLIST_REMOVE(&vmm_softc->vm_list, vm, vm, vm_link)do { if ((&vmm_softc->vm_list)->slh_first == (vm)) { do { ((&vmm_softc->vm_list))->slh_first = ((&vmm_softc ->vm_list))->slh_first->vm_link.sle_next; } while (0 ); } else { struct vm *curelm = (&vmm_softc->vm_list)-> slh_first; while (curelm->vm_link.sle_next != (vm)) curelm = curelm->vm_link.sle_next; curelm->vm_link.sle_next = curelm->vm_link.sle_next->vm_link.sle_next; } ((vm)-> vm_link.sle_next) = ((void *)-1); } while (0); |
| 4205 | if (vm->vm_vcpus_running == 0) |
| 4206 | vm_teardown(vm); |
| 4207 | |
| 4208 | rw_exit_write(&vmm_softc->vm_lock); |
| 4209 | |
| 4210 | return (0); |
| 4211 | } |
| 4212 | |
| 4213 | /* |
| 4214 | * vm_run |
| 4215 | * |
| 4216 | * Run the vm / vcpu specified by 'vrp' |
| 4217 | * |
| 4218 | * Parameters: |
| 4219 | * vrp: structure defining the VM to run |
| 4220 | * |
| 4221 | * Return value: |
| 4222 | * ENOENT: the VM defined in 'vrp' could not be located |
| 4223 | * EBUSY: the VM defined in 'vrp' is already running |
| 4224 | * EFAULT: error copying data from userspace (vmd) on return from previous |
| 4225 | * exit. |
| 4226 | * EAGAIN: help is needed from vmd(8) (device I/O or exit vmm(4) cannot |
| 4227 | * handle in-kernel.) |
| 4228 | * 0: the run loop exited and no help is needed from vmd(8) |
| 4229 | */ |
| 4230 | int |
| 4231 | vm_run(struct vm_run_params *vrp) |
| 4232 | { |
| 4233 | struct vm *vm; |
| 4234 | struct vcpu *vcpu; |
| 4235 | int ret = 0, error; |
| 4236 | u_int old, next; |
| 4237 | |
| 4238 | /* |
| 4239 | * Find desired VM |
| 4240 | */ |
| 4241 | rw_enter_read(&vmm_softc->vm_lock); |
| 4242 | error = vm_find(vrp->vrp_vm_id, &vm); |
| 4243 | |
| 4244 | /* |
| 4245 | * Attempt to locate the requested VCPU. If found, attempt to |
| 4246 | * to transition from VCPU_STATE_STOPPED -> VCPU_STATE_RUNNING. |
| 4247 | * Failure to make the transition indicates the VCPU is busy. |
| 4248 | */ |
| 4249 | if (error == 0) { |
| 4250 | rw_enter_read(&vm->vm_vcpu_lock); |
| 4251 | SLIST_FOREACH(vcpu, &vm->vm_vcpu_list, vc_vcpu_link)for((vcpu) = ((&vm->vm_vcpu_list)->slh_first); (vcpu ) != ((void *)0); (vcpu) = ((vcpu)->vc_vcpu_link.sle_next) ) { |
| 4252 | if (vcpu->vc_id == vrp->vrp_vcpu_id) |
| 4253 | break; |
| 4254 | } |
| 4255 | |
| 4256 | if (vcpu != NULL((void *)0)) { |
| 4257 | old = VCPU_STATE_STOPPED; |
| 4258 | next = VCPU_STATE_RUNNING; |
| 4259 | |
| 4260 | if (atomic_cas_uint(&vcpu->vc_state, old, next)_atomic_cas_uint((&vcpu->vc_state), (old), (next)) != old) |
| 4261 | ret = EBUSY16; |
| 4262 | else { |
| 4263 | atomic_inc_int(&vm->vm_vcpus_running)_atomic_inc_int(&vm->vm_vcpus_running); |
| 4264 | rw_enter_write(&vcpu->vc_lock); |
| 4265 | } |
| 4266 | } else |
| 4267 | ret = ENOENT2; |
| 4268 | |
| 4269 | rw_exit_read(&vm->vm_vcpu_lock); |
| 4270 | } |
| 4271 | rw_exit_read(&vmm_softc->vm_lock); |
| 4272 | |
| 4273 | if (error != 0) |
| 4274 | ret = error; |
| 4275 | |
| 4276 | /* Bail if errors detected in the previous steps */ |
| 4277 | if (ret) |
| 4278 | return (ret); |
| 4279 | |
| 4280 | /* |
| 4281 | * We may be returning from userland helping us from the last exit. |
| 4282 | * If so (vrp_continue == 1), copy in the exit data from vmd. The |
| 4283 | * exit data will be consumed before the next entry (this typically |
| 4284 | * comprises VCPU register changes as the result of vmd(8)'s actions). |
| 4285 | */ |
| 4286 | if (vrp->vrp_continue) { |
| 4287 | if (copyin(vrp->vrp_exit, &vcpu->vc_exit, |
| 4288 | sizeof(struct vm_exit)) == EFAULT14) { |
| 4289 | rw_exit_write(&vcpu->vc_lock); |
| 4290 | return (EFAULT14); |
| 4291 | } |
| 4292 | } |
| 4293 | |
| 4294 | /* Run the VCPU specified in vrp */ |
| 4295 | if (vcpu->vc_virt_mode == VMM_MODE_VMX || |
| 4296 | vcpu->vc_virt_mode == VMM_MODE_EPT) { |
| 4297 | ret = vcpu_run_vmx(vcpu, vrp); |
| 4298 | } else if (vcpu->vc_virt_mode == VMM_MODE_SVM || |
| 4299 | vcpu->vc_virt_mode == VMM_MODE_RVI) { |
| 4300 | ret = vcpu_run_svm(vcpu, vrp); |
| 4301 | } |
| 4302 | |
| 4303 | /* |
| 4304 | * We can set the VCPU states here without CAS because once |
| 4305 | * a VCPU is in state RUNNING or REQTERM, only the VCPU itself |
| 4306 | * can switch the state. |
| 4307 | */ |
| 4308 | atomic_dec_int(&vm->vm_vcpus_running)_atomic_dec_int(&vm->vm_vcpus_running); |
| 4309 | if (vcpu->vc_state == VCPU_STATE_REQTERM) { |
| 4310 | vrp->vrp_exit_reason = VM_EXIT_TERMINATED0xFFFE; |
| 4311 | vcpu->vc_state = VCPU_STATE_TERMINATED; |
| 4312 | if (vm->vm_vcpus_running == 0) { |
| 4313 | rw_enter_write(&vmm_softc->vm_lock); |
| 4314 | vm_teardown(vm); |
| 4315 | rw_exit_write(&vmm_softc->vm_lock); |
| 4316 | } |
| 4317 | ret = 0; |
| 4318 | } else if (ret == 0 || ret == EAGAIN35) { |
| 4319 | /* If we are exiting, populate exit data so vmd can help. */ |
| 4320 | vrp->vrp_exit_reason = (ret == 0) ? VM_EXIT_NONE0xFFFF |
| 4321 | : vcpu->vc_gueststate.vg_exit_reason; |
| 4322 | vrp->vrp_irqready = vcpu->vc_irqready; |
| 4323 | vcpu->vc_state = VCPU_STATE_STOPPED; |
| 4324 | |
| 4325 | if (copyout(&vcpu->vc_exit, vrp->vrp_exit, |
| 4326 | sizeof(struct vm_exit)) == EFAULT14) { |
| 4327 | ret = EFAULT14; |
| 4328 | } else |
| 4329 | ret = 0; |
| 4330 | } else { |
| 4331 | vrp->vrp_exit_reason = VM_EXIT_TERMINATED0xFFFE; |
| 4332 | vcpu->vc_state = VCPU_STATE_TERMINATED; |
| 4333 | } |
| 4334 | |
| 4335 | rw_exit_write(&vcpu->vc_lock); |
| 4336 | |
| 4337 | return (ret); |
| 4338 | } |
| 4339 | |
| 4340 | /* |
| 4341 | * vcpu_must_stop |
| 4342 | * |
| 4343 | * Check if we need to (temporarily) stop running the VCPU for some reason, |
| 4344 | * such as: |
| 4345 | * - the VM was requested to terminate |
| 4346 | * - the proc running this VCPU has pending signals |
| 4347 | * |
| 4348 | * Parameters: |
| 4349 | * vcpu: the VCPU to check |
| 4350 | * |
| 4351 | * Return values: |
| 4352 | * 1: the VM owning this VCPU should stop |
| 4353 | * 0: no stop is needed |
| 4354 | */ |
| 4355 | int |
| 4356 | vcpu_must_stop(struct vcpu *vcpu) |
| 4357 | { |
| 4358 | struct proc *p = curproc({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc; |
| 4359 | |
| 4360 | if (vcpu->vc_state == VCPU_STATE_REQTERM) |
| 4361 | return (1); |
| 4362 | if (SIGPENDING(p)(((p)->p_siglist | (p)->p_p->ps_siglist) & ~(p)-> p_sigmask) != 0) |
| 4363 | return (1); |
| 4364 | return (0); |
| 4365 | } |
| 4366 | |
| 4367 | /* |
| 4368 | * vmm_fpurestore |
| 4369 | * |
| 4370 | * Restore the guest's FPU state, saving the existing userland thread's |
| 4371 | * FPU context if necessary. Must be called with interrupts disabled. |
| 4372 | */ |
| 4373 | int |
| 4374 | vmm_fpurestore(struct vcpu *vcpu) |
| 4375 | { |
| 4376 | struct cpu_info *ci = curcpu()({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;}); |
| 4377 | |
| 4378 | /* save vmm's FPU state if we haven't already */ |
| 4379 | if (ci->ci_flags & CPUF_USERXSTATE0x0200) { |
| 4380 | ci->ci_flags &= ~CPUF_USERXSTATE0x0200; |
| 4381 | fpusavereset(&curproc({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_addr->u_pcb.pcb_savefpu); |
| 4382 | } |
| 4383 | |
| 4384 | if (vcpu->vc_fpuinited) { |
| 4385 | if (xrstor_user(&vcpu->vc_g_fpu, xsave_mask)) { |
| 4386 | DPRINTF("%s: guest attempted to set invalid %s\n", |
| 4387 | __func__, "xsave/xrstor state"); |
| 4388 | return EINVAL22; |
| 4389 | } |
| 4390 | } |
| 4391 | |
| 4392 | if (xsave_mask) { |
| 4393 | /* Restore guest %xcr0 */ |
| 4394 | if (xsetbv_user(0, vcpu->vc_gueststate.vg_xcr0)) { |
| 4395 | DPRINTF("%s: guest attempted to set invalid bits in " |
| 4396 | "xcr0 (guest %%xcr0=0x%llx, host %%xcr0=0x%llx)\n", |
| 4397 | __func__, vcpu->vc_gueststate.vg_xcr0, xsave_mask); |
| 4398 | return EINVAL22; |
| 4399 | } |
| 4400 | } |
| 4401 | |
| 4402 | return 0; |
| 4403 | } |
| 4404 | |
| 4405 | /* |
| 4406 | * vmm_fpusave |
| 4407 | * |
| 4408 | * Save the guest's FPU state. Must be called with interrupts disabled. |
| 4409 | */ |
| 4410 | void |
| 4411 | vmm_fpusave(struct vcpu *vcpu) |
| 4412 | { |
| 4413 | if (xsave_mask) { |
| 4414 | /* Save guest %xcr0 */ |
| 4415 | vcpu->vc_gueststate.vg_xcr0 = xgetbv(0); |
| 4416 | |
| 4417 | /* Restore host %xcr0 */ |
| 4418 | xsetbv(0, xsave_mask); |
| 4419 | } |
| 4420 | |
| 4421 | /* |
| 4422 | * Save full copy of FPU state - guest content is always |
| 4423 | * a subset of host's save area (see xsetbv exit handler) |
| 4424 | */ |
| 4425 | fpusavereset(&vcpu->vc_g_fpu); |
| 4426 | vcpu->vc_fpuinited = 1; |
| 4427 | } |
| 4428 | |
| 4429 | /* |
| 4430 | * vmm_translate_gva |
| 4431 | * |
| 4432 | * Translates a guest virtual address to a guest physical address by walking |
| 4433 | * the currently active page table (if needed). |
| 4434 | * |
| 4435 | * Note - this function can possibly alter the supplied VCPU state. |
| 4436 | * Specifically, it may inject exceptions depending on the current VCPU |
| 4437 | * configuration, and may alter %cr2 on #PF. Consequently, this function |
| 4438 | * should only be used as part of instruction emulation. |
| 4439 | * |
| 4440 | * Parameters: |
| 4441 | * vcpu: The VCPU this translation should be performed for (guest MMU settings |
| 4442 | * are gathered from this VCPU) |
| 4443 | * va: virtual address to translate |
| 4444 | * pa: pointer to paddr_t variable that will receive the translated physical |
| 4445 | * address. 'pa' is unchanged on error. |
| 4446 | * mode: one of PROT_READ, PROT_WRITE, PROT_EXEC indicating the mode in which |
| 4447 | * the address should be translated |
| 4448 | * |
| 4449 | * Return values: |
| 4450 | * 0: the address was successfully translated - 'pa' contains the physical |
| 4451 | * address currently mapped by 'va'. |
| 4452 | * EFAULT: the PTE for 'VA' is unmapped. A #PF will be injected in this case |
| 4453 | * and %cr2 set in the vcpu structure. |
| 4454 | * EINVAL: an error occurred reading paging table structures |
| 4455 | */ |
| 4456 | int |
| 4457 | vmm_translate_gva(struct vcpu *vcpu, uint64_t va, uint64_t *pa, int mode) |
| 4458 | { |
| 4459 | int level, shift, pdidx; |
| 4460 | uint64_t pte, pt_paddr, pte_paddr, mask, low_mask, high_mask; |
| 4461 | uint64_t shift_width, pte_size, *hva; |
| 4462 | paddr_t hpa; |
| 4463 | struct vcpu_reg_state vrs; |
| 4464 | |
| 4465 | level = 0; |
Value stored to 'level' is never read | |
| 4466 | |
| 4467 | if (vmm_softc->mode == VMM_MODE_EPT || |
| 4468 | vmm_softc->mode == VMM_MODE_VMX) { |
| 4469 | if (vcpu_readregs_vmx(vcpu, VM_RWREGS_ALL(0x1 | 0x2 | 0x4 | 0x8 | 0x10), &vrs)) |
| 4470 | return (EINVAL22); |
| 4471 | } else if (vmm_softc->mode == VMM_MODE_RVI || |
| 4472 | vmm_softc->mode == VMM_MODE_SVM) { |
| 4473 | if (vcpu_readregs_svm(vcpu, VM_RWREGS_ALL(0x1 | 0x2 | 0x4 | 0x8 | 0x10), &vrs)) |
| 4474 | return (EINVAL22); |
| 4475 | } else { |
| 4476 | printf("%s: unknown vmm mode", __func__); |
| 4477 | return (EINVAL22); |
| 4478 | } |
| 4479 | |
| 4480 | DPRINTF("%s: guest %%cr0=0x%llx, %%cr3=0x%llx\n", __func__, |
| 4481 | vrs.vrs_crs[VCPU_REGS_CR0], vrs.vrs_crs[VCPU_REGS_CR3]); |
| 4482 | |
| 4483 | if (!(vrs.vrs_crs[VCPU_REGS_CR00] & CR0_PG0x80000000)) { |
| 4484 | DPRINTF("%s: unpaged, va=pa=0x%llx\n", __func__, |
| 4485 | va); |
| 4486 | *pa = va; |
| 4487 | return (0); |
| 4488 | } |
| 4489 | |
| 4490 | pt_paddr = vrs.vrs_crs[VCPU_REGS_CR32]; |
| 4491 | |
| 4492 | if (vrs.vrs_crs[VCPU_REGS_CR00] & CR0_PE0x00000001) { |
| 4493 | if (vrs.vrs_crs[VCPU_REGS_CR43] & CR4_PAE0x00000020) { |
| 4494 | pte_size = sizeof(uint64_t); |
| 4495 | shift_width = 9; |
| 4496 | |
| 4497 | if (vrs.vrs_msrs[VCPU_REGS_EFER0] & EFER_LMA0x00000400) { |
| 4498 | level = 4; |
| 4499 | mask = L4_MASK0x0000ff8000000000UL; |
| 4500 | shift = L4_SHIFT39; |
| 4501 | } else { |
| 4502 | level = 3; |
| 4503 | mask = L3_MASK0x0000007fc0000000UL; |
| 4504 | shift = L3_SHIFT30; |
| 4505 | } |
| 4506 | } else { |
| 4507 | level = 2; |
| 4508 | shift_width = 10; |
| 4509 | mask = 0xFFC00000; |
| 4510 | shift = 22; |
| 4511 | pte_size = sizeof(uint32_t); |
| 4512 | } |
| 4513 | } else { |
| 4514 | return (EINVAL22); |
| 4515 | } |
| 4516 | |
| 4517 | DPRINTF("%s: pte size=%lld level=%d mask=0x%llx, shift=%d, " |
| 4518 | "shift_width=%lld\n", __func__, pte_size, level, mask, shift, |
| 4519 | shift_width); |
| 4520 | |
| 4521 | /* XXX: Check for R bit in segment selector and set A bit */ |
| 4522 | |
| 4523 | for (;level > 0; level--) { |
| 4524 | pdidx = (va & mask) >> shift; |
| 4525 | pte_paddr = (pt_paddr) + (pdidx * pte_size); |
| 4526 | |
| 4527 | DPRINTF("%s: read pte level %d @ GPA 0x%llx\n", __func__, |
| 4528 | level, pte_paddr); |
| 4529 | if (!pmap_extract(vcpu->vc_parent->vm_map->pmap, pte_paddr, |
| 4530 | &hpa)) { |
| 4531 | DPRINTF("%s: cannot extract HPA for GPA 0x%llx\n", |
| 4532 | __func__, pte_paddr); |
| 4533 | return (EINVAL22); |
| 4534 | } |
| 4535 | |
| 4536 | hpa = hpa | (pte_paddr & 0xFFF); |
| 4537 | hva = (uint64_t *)PMAP_DIRECT_MAP(hpa)((vaddr_t)(((((511 - 4) * (1ULL << 39))) | 0xffff000000000000 )) + (hpa)); |
| 4538 | DPRINTF("%s: GPA 0x%llx -> HPA 0x%llx -> HVA 0x%llx\n", |
| 4539 | __func__, pte_paddr, (uint64_t)hpa, (uint64_t)hva); |
| 4540 | if (pte_size == 8) |
| 4541 | pte = *hva; |
| 4542 | else |
| 4543 | pte = *(uint32_t *)hva; |
| 4544 | |
| 4545 | DPRINTF("%s: PTE @ 0x%llx = 0x%llx\n", __func__, pte_paddr, |
| 4546 | pte); |
| 4547 | |
| 4548 | /* XXX: Set CR2 */ |
| 4549 | if (!(pte & PG_V0x0000000000000001UL)) |
| 4550 | return (EFAULT14); |
| 4551 | |
| 4552 | /* XXX: Check for SMAP */ |
| 4553 | if ((mode == PROT_WRITE0x02) && !(pte & PG_RW0x0000000000000002UL)) |
| 4554 | return (EPERM1); |
| 4555 | |
| 4556 | if ((vcpu->vc_exit.cpl > 0) && !(pte & PG_u0x0000000000000004UL)) |
| 4557 | return (EPERM1); |
| 4558 | |
| 4559 | pte = pte | PG_U0x0000000000000020UL; |
| 4560 | if (mode == PROT_WRITE0x02) |
| 4561 | pte = pte | PG_M0x0000000000000040UL; |
| 4562 | *hva = pte; |
| 4563 | |
| 4564 | /* XXX: EINVAL if in 32bit and PG_PS is 1 but CR4.PSE is 0 */ |
| 4565 | if (pte & PG_PS0x0000000000000080UL) |
| 4566 | break; |
| 4567 | |
| 4568 | if (level > 1) { |
| 4569 | pt_paddr = pte & PG_FRAME0x000ffffffffff000UL; |
| 4570 | shift -= shift_width; |
| 4571 | mask = mask >> shift_width; |
| 4572 | } |
| 4573 | } |
| 4574 | |
| 4575 | low_mask = ((uint64_t)1ULL << shift) - 1; |
| 4576 | high_mask = (((uint64_t)1ULL << ((pte_size * 8) - 1)) - 1) ^ low_mask; |
| 4577 | *pa = (pte & high_mask) | (va & low_mask); |
| 4578 | |
| 4579 | DPRINTF("%s: final GPA for GVA 0x%llx = 0x%llx\n", __func__, |
| 4580 | va, *pa); |
| 4581 | |
| 4582 | return (0); |
| 4583 | } |
| 4584 | |
| 4585 | |
| 4586 | /* |
| 4587 | * vcpu_run_vmx |
| 4588 | * |
| 4589 | * VMX main loop used to run a VCPU. |
| 4590 | * |
| 4591 | * Parameters: |
| 4592 | * vcpu: The VCPU to run |
| 4593 | * vrp: run parameters |
| 4594 | * |
| 4595 | * Return values: |
| 4596 | * 0: The run loop exited and no help is needed from vmd |
| 4597 | * EAGAIN: The run loop exited and help from vmd is needed |
| 4598 | * EINVAL: an error occurred |
| 4599 | */ |
| 4600 | int |
| 4601 | vcpu_run_vmx(struct vcpu *vcpu, struct vm_run_params *vrp) |
| 4602 | { |
| 4603 | int ret = 0, exitinfo; |
| 4604 | struct region_descriptor gdt; |
| 4605 | struct cpu_info *ci = curcpu()({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;}); |
| 4606 | uint64_t exit_reason, cr3, insn_error; |
| 4607 | struct schedstate_percpu *spc; |
| 4608 | struct vmx_invvpid_descriptor vid; |
| 4609 | uint64_t eii, procbased, int_st; |
| 4610 | uint16_t irq, ldt_sel; |
| 4611 | u_long s; |
| 4612 | struct region_descriptor gdtr, idtr; |
| 4613 | |
| 4614 | rw_assert_wrlock(&vcpu->vc_lock); |
| 4615 | |
| 4616 | if (vcpu_reload_vmcs_vmx(vcpu)) { |
| 4617 | printf("%s: failed (re)loading vmcs\n", __func__); |
| 4618 | return (EINVAL22); |
| 4619 | } |
| 4620 | |
| 4621 | /* |
| 4622 | * If we are returning from userspace (vmd) because we exited |
| 4623 | * last time, fix up any needed vcpu state first. Which state |
| 4624 | * needs to be fixed up depends on what vmd populated in the |
| 4625 | * exit data structure. |
| 4626 | */ |
| 4627 | irq = vrp->vrp_irq; |
| 4628 | |
| 4629 | if (vrp->vrp_continue) { |
| 4630 | switch (vcpu->vc_gueststate.vg_exit_reason) { |
| 4631 | case VMX_EXIT_IO30: |
| 4632 | if (vcpu->vc_exit.vei.vei_dir == VEI_DIR_IN) |
| 4633 | vcpu->vc_gueststate.vg_rax = |
| 4634 | vcpu->vc_exit.vei.vei_data; |
| 4635 | break; |
| 4636 | case VM_EXIT_NONE0xFFFF: |
| 4637 | case VMX_EXIT_HLT12: |
| 4638 | case VMX_EXIT_INT_WINDOW7: |
| 4639 | case VMX_EXIT_EXTINT1: |
| 4640 | case VMX_EXIT_EPT_VIOLATION48: |
| 4641 | case VMX_EXIT_CPUID10: |
| 4642 | case VMX_EXIT_XSETBV55: |
| 4643 | break; |
| 4644 | #ifdef VMM_DEBUG |
| 4645 | case VMX_EXIT_TRIPLE_FAULT2: |
| 4646 | DPRINTF("%s: vm %d vcpu %d triple fault\n", |
| 4647 | __func__, vcpu->vc_parent->vm_id, |
| 4648 | vcpu->vc_id); |
| 4649 | vmx_vcpu_dump_regs(vcpu); |
| 4650 | dump_vcpu(vcpu); |
| 4651 | vmx_dump_vmcs(vcpu); |
| 4652 | break; |
| 4653 | case VMX_EXIT_ENTRY_FAILED_GUEST_STATE33: |
| 4654 | DPRINTF("%s: vm %d vcpu %d failed entry " |
| 4655 | "due to invalid guest state\n", |
| 4656 | __func__, vcpu->vc_parent->vm_id, |
| 4657 | vcpu->vc_id); |
| 4658 | vmx_vcpu_dump_regs(vcpu); |
| 4659 | dump_vcpu(vcpu); |
| 4660 | return (EINVAL22); |
| 4661 | default: |
| 4662 | DPRINTF("%s: unimplemented exit type %d (%s)\n", |
| 4663 | __func__, |
| 4664 | vcpu->vc_gueststate.vg_exit_reason, |
| 4665 | vmx_exit_reason_decode( |
| 4666 | vcpu->vc_gueststate.vg_exit_reason)); |
| 4667 | vmx_vcpu_dump_regs(vcpu); |
| 4668 | dump_vcpu(vcpu); |
| 4669 | break; |
| 4670 | #endif /* VMM_DEBUG */ |
| 4671 | } |
| 4672 | } |
| 4673 | |
| 4674 | setregion(&gdt, ci->ci_gdt, GDT_SIZE((6 << 3) + (1 << 4)) - 1); |
| 4675 | if (gdt.rd_base == 0) { |
| 4676 | printf("%s: setregion\n", __func__); |
| 4677 | return (EINVAL22); |
| 4678 | } |
| 4679 | |
| 4680 | /* Host GDTR base */ |
| 4681 | if (vmwrite(VMCS_HOST_IA32_GDTR_BASE0x6C0C, gdt.rd_base)) { |
| 4682 | printf("%s: vmwrite(0x%04X, 0x%llx)\n", __func__, |
| 4683 | VMCS_HOST_IA32_GDTR_BASE0x6C0C, gdt.rd_base); |
| 4684 | return (EINVAL22); |
| 4685 | } |
| 4686 | |
| 4687 | /* Host TR base */ |
| 4688 | if (vmwrite(VMCS_HOST_IA32_TR_BASE0x6C0A, (uint64_t)ci->ci_tss)) { |
| 4689 | printf("%s: vmwrite(0x%04X, 0x%llx)\n", __func__, |
| 4690 | VMCS_HOST_IA32_TR_BASE0x6C0A, (uint64_t)ci->ci_tss); |
| 4691 | return (EINVAL22); |
| 4692 | } |
| 4693 | |
| 4694 | /* Host CR3 */ |
| 4695 | cr3 = rcr3(); |
| 4696 | if (vmwrite(VMCS_HOST_IA32_CR30x6C02, cr3)) { |
| 4697 | printf("%s: vmwrite(0x%04X, 0x%llx)\n", __func__, |
| 4698 | VMCS_HOST_IA32_CR30x6C02, cr3); |
| 4699 | return (EINVAL22); |
| 4700 | } |
| 4701 | |
| 4702 | /* Handle vmd(8) injected interrupts */ |
| 4703 | /* Is there an interrupt pending injection? */ |
| 4704 | if (irq != 0xFFFF) { |
| 4705 | if (vmread(VMCS_GUEST_INTERRUPTIBILITY_ST0x4824, &int_st)) { |
| 4706 | printf("%s: can't get interruptibility state\n", |
| 4707 | __func__); |
| 4708 | return (EINVAL22); |
| 4709 | } |
| 4710 | |
| 4711 | /* Interruptibility state 0x3 covers NMIs and STI */ |
| 4712 | if (!(int_st & 0x3) && vcpu->vc_irqready) { |
| 4713 | eii = (irq & 0xFF); |
| 4714 | eii |= (1ULL << 31); /* Valid */ |
| 4715 | eii |= (0ULL << 8); /* Hardware Interrupt */ |
| 4716 | if (vmwrite(VMCS_ENTRY_INTERRUPTION_INFO0x4016, eii)) { |
| 4717 | printf("vcpu_run_vmx: can't vector " |
| 4718 | "interrupt to guest\n"); |
| 4719 | return (EINVAL22); |
| 4720 | } |
| 4721 | |
| 4722 | irq = 0xFFFF; |
| 4723 | } |
| 4724 | } else if (!vcpu->vc_intr) { |
| 4725 | /* |
| 4726 | * Disable window exiting |
| 4727 | */ |
| 4728 | if (vmread(VMCS_PROCBASED_CTLS0x4002, &procbased)) { |
| 4729 | printf("%s: can't read procbased ctls on exit\n", |
| 4730 | __func__); |
| 4731 | return (EINVAL22); |
| 4732 | } else { |
| 4733 | procbased &= ~IA32_VMX_INTERRUPT_WINDOW_EXITING(1ULL << 2); |
| 4734 | if (vmwrite(VMCS_PROCBASED_CTLS0x4002, procbased)) { |
| 4735 | printf("%s: can't write procbased ctls " |
| 4736 | "on exit\n", __func__); |
| 4737 | return (EINVAL22); |
| 4738 | } |
| 4739 | } |
| 4740 | } |
| 4741 | |
| 4742 | while (ret == 0) { |
| 4743 | #ifdef VMM_DEBUG |
| 4744 | paddr_t pa = 0ULL; |
| 4745 | vmptrst(&pa); |
| 4746 | KASSERT(pa == vcpu->vc_control_pa)((pa == vcpu->vc_control_pa) ? (void)0 : __assert("diagnostic " , "/usr/src/sys/arch/amd64/amd64/vmm.c", 4746, "pa == vcpu->vc_control_pa" )); |
| 4747 | #endif /* VMM_DEBUG */ |
| 4748 | |
| 4749 | vmm_update_pvclock(vcpu); |
| 4750 | |
| 4751 | /* Inject event if present */ |
| 4752 | if (vcpu->vc_event != 0) { |
| 4753 | eii = (vcpu->vc_event & 0xFF); |
| 4754 | eii |= (1ULL << 31); /* Valid */ |
| 4755 | |
| 4756 | /* Set the "Send error code" flag for certain vectors */ |
| 4757 | switch (vcpu->vc_event & 0xFF) { |
| 4758 | case VMM_EX_DF8: |
| 4759 | case VMM_EX_TS10: |
| 4760 | case VMM_EX_NP11: |
| 4761 | case VMM_EX_SS12: |
| 4762 | case VMM_EX_GP13: |
| 4763 | case VMM_EX_PF14: |
| 4764 | case VMM_EX_AC17: |
| 4765 | eii |= (1ULL << 11); |
| 4766 | } |
| 4767 | |
| 4768 | eii |= (3ULL << 8); /* Hardware Exception */ |
| 4769 | if (vmwrite(VMCS_ENTRY_INTERRUPTION_INFO0x4016, eii)) { |
| 4770 | printf("%s: can't vector event to guest\n", |
| 4771 | __func__); |
| 4772 | ret = EINVAL22; |
| 4773 | break; |
| 4774 | } |
| 4775 | |
| 4776 | if (vmwrite(VMCS_ENTRY_EXCEPTION_ERROR_CODE0x4018, 0)) { |
| 4777 | printf("%s: can't write error code to guest\n", |
| 4778 | __func__); |
| 4779 | ret = EINVAL22; |
| 4780 | break; |
| 4781 | } |
| 4782 | |
| 4783 | vcpu->vc_event = 0; |
| 4784 | } |
| 4785 | |
| 4786 | if (vcpu->vc_vmx_vpid_enabled) { |
| 4787 | /* Invalidate old TLB mappings */ |
| 4788 | vid.vid_vpid = vcpu->vc_parent->vm_id; |
| 4789 | vid.vid_addr = 0; |
| 4790 | invvpid(IA32_VMX_INVVPID_SINGLE_CTX_GLB0x3, &vid); |
| 4791 | } |
| 4792 | |
| 4793 | /* Start / resume the VCPU */ |
| 4794 | |
| 4795 | /* Disable interrupts and save the current host FPU state. */ |
| 4796 | s = intr_disable(); |
| 4797 | if ((ret = vmm_fpurestore(vcpu))) { |
| 4798 | intr_restore(s); |
| 4799 | break; |
| 4800 | } |
| 4801 | |
| 4802 | sgdt(&gdtr); |
| 4803 | sidt(&idtr); |
| 4804 | sldt(&ldt_sel); |
| 4805 | |
| 4806 | TRACEPOINT(vmm, guest_enter, vcpu, vrp)do { extern struct dt_probe (dt_static_vmm_guest_enter); struct dt_probe *dtp = &(dt_static_vmm_guest_enter); struct dt_provider *dtpv = dtp->dtp_prov; if (__builtin_expect(((dt_tracing) != 0), 0) && __builtin_expect(((dtp->dtp_recording ) != 0), 0)) { dtpv->dtpv_enter(dtpv, dtp, vcpu, vrp); } } while (0); |
| 4807 | |
| 4808 | ret = vmx_enter_guest(&vcpu->vc_control_pa, |
| 4809 | &vcpu->vc_gueststate, |
| 4810 | (vcpu->vc_vmx_vmcs_state == VMCS_LAUNCHED1), |
| 4811 | ci->ci_vmm_cap.vcc_vmx.vmx_has_l1_flush_msr); |
| 4812 | |
| 4813 | bare_lgdt(&gdtr); |
| 4814 | lidt(&idtr); |
| 4815 | lldt(ldt_sel); |
| 4816 | |
| 4817 | /* |
| 4818 | * On exit, interrupts are disabled, and we are running with |
| 4819 | * the guest FPU state still possibly on the CPU. Save the FPU |
| 4820 | * state before re-enabling interrupts. |
| 4821 | */ |
| 4822 | vmm_fpusave(vcpu); |
| 4823 | intr_restore(s); |
| 4824 | |
| 4825 | TRACEPOINT(vmm, guest_exit, vcpu, vrp, exit_reason)do { extern struct dt_probe (dt_static_vmm_guest_exit); struct dt_probe *dtp = &(dt_static_vmm_guest_exit); struct dt_provider *dtpv = dtp->dtp_prov; if (__builtin_expect(((dt_tracing) != 0), 0) && __builtin_expect(((dtp->dtp_recording ) != 0), 0)) { dtpv->dtpv_enter(dtpv, dtp, vcpu, vrp, exit_reason ); } } while (0); |
| 4826 | |
| 4827 | atomic_swap_uint(&vcpu->vc_vmx_vmcs_state, VMCS_LAUNCHED)_atomic_swap_uint((&vcpu->vc_vmx_vmcs_state), (1)); |
| 4828 | exit_reason = VM_EXIT_NONE0xFFFF; |
| 4829 | |
| 4830 | /* If we exited successfully ... */ |
| 4831 | if (ret == 0) { |
| 4832 | /* |
| 4833 | * ret == 0 implies we entered the guest, and later |
| 4834 | * exited for some valid reason |
| 4835 | */ |
| 4836 | exitinfo = vmx_get_exit_info( |
| 4837 | &vcpu->vc_gueststate.vg_rip, &exit_reason); |
| 4838 | if (vmread(VMCS_GUEST_IA32_RFLAGS0x6820, |
| 4839 | &vcpu->vc_gueststate.vg_rflags)) { |
| 4840 | printf("%s: can't read guest rflags during " |
| 4841 | "exit\n", __func__); |
| 4842 | ret = EINVAL22; |
| 4843 | break; |
| 4844 | } |
| 4845 | |
| 4846 | /* Update our state */ |
| 4847 | if (!(exitinfo & VMX_EXIT_INFO_HAVE_RIP0x1)) { |
| 4848 | printf("%s: cannot read guest rip\n", __func__); |
| 4849 | ret = EINVAL22; |
| 4850 | break; |
| 4851 | } |
| 4852 | |
| 4853 | if (!(exitinfo & VMX_EXIT_INFO_HAVE_REASON0x2)) { |
| 4854 | printf("%s: cant read exit reason\n", __func__); |
| 4855 | ret = EINVAL22; |
| 4856 | break; |
| 4857 | } |
| 4858 | |
| 4859 | /* |
| 4860 | * Handle the exit. This will alter "ret" to EAGAIN if |
| 4861 | * the exit handler determines help from vmd is needed. |
| 4862 | */ |
| 4863 | vcpu->vc_gueststate.vg_exit_reason = exit_reason; |
| 4864 | ret = vmx_handle_exit(vcpu); |
| 4865 | |
| 4866 | if (vcpu->vc_gueststate.vg_rflags & PSL_I0x00000200) |
| 4867 | vcpu->vc_irqready = 1; |
| 4868 | else |
| 4869 | vcpu->vc_irqready = 0; |
| 4870 | |
| 4871 | /* |
| 4872 | * If not ready for interrupts, but interrupts pending, |
| 4873 | * enable interrupt window exiting. |
| 4874 | */ |
| 4875 | if (vcpu->vc_irqready == 0 && vcpu->vc_intr) { |
| 4876 | if (vmread(VMCS_PROCBASED_CTLS0x4002, &procbased)) { |
| 4877 | printf("%s: can't read procbased ctls " |
| 4878 | "on intwin exit\n", __func__); |
| 4879 | ret = EINVAL22; |
| 4880 | break; |
| 4881 | } |
| 4882 | |
| 4883 | procbased |= IA32_VMX_INTERRUPT_WINDOW_EXITING(1ULL << 2); |
| 4884 | if (vmwrite(VMCS_PROCBASED_CTLS0x4002, procbased)) { |
| 4885 | printf("%s: can't write procbased ctls " |
| 4886 | "on intwin exit\n", __func__); |
| 4887 | ret = EINVAL22; |
| 4888 | break; |
| 4889 | } |
| 4890 | } |
| 4891 | |
| 4892 | /* |
| 4893 | * Exit to vmd if we are terminating, failed to enter, |
| 4894 | * or need help (device I/O) |
| 4895 | */ |
| 4896 | if (ret || vcpu_must_stop(vcpu)) |
| 4897 | break; |
| 4898 | |
| 4899 | if (vcpu->vc_intr && vcpu->vc_irqready) { |
| 4900 | ret = EAGAIN35; |
| 4901 | break; |
| 4902 | } |
| 4903 | |
| 4904 | /* Check if we should yield - don't hog the {p,v}pu */ |
| 4905 | spc = &ci->ci_schedstate; |
| 4906 | if (spc->spc_schedflags & SPCF_SHOULDYIELD0x0002) |
| 4907 | break; |
| 4908 | |
| 4909 | } else { |
| 4910 | /* |
| 4911 | * We failed vmresume or vmlaunch for some reason, |
| 4912 | * typically due to invalid vmcs state or other |
| 4913 | * reasons documented in SDM Vol 3C 30.4. |
| 4914 | */ |
| 4915 | switch (ret) { |
| 4916 | case VMX_FAIL_LAUNCH_INVALID_VMCS2: |
| 4917 | printf("%s: failed %s with invalid vmcs\n", |
| 4918 | __func__, |
| 4919 | (vcpu->vc_vmx_vmcs_state == VMCS_LAUNCHED1 |
| 4920 | ? "vmresume" : "vmlaunch")); |
| 4921 | break; |
| 4922 | case VMX_FAIL_LAUNCH_VALID_VMCS3: |
| 4923 | printf("%s: failed %s with valid vmcs\n", |
| 4924 | __func__, |
| 4925 | (vcpu->vc_vmx_vmcs_state == VMCS_LAUNCHED1 |
| 4926 | ? "vmresume" : "vmlaunch")); |
| 4927 | break; |
| 4928 | default: |
| 4929 | printf("%s: failed %s for unknown reason\n", |
| 4930 | __func__, |
| 4931 | (vcpu->vc_vmx_vmcs_state == VMCS_LAUNCHED1 |
| 4932 | ? "vmresume" : "vmlaunch")); |
| 4933 | } |
| 4934 | |
| 4935 | ret = EINVAL22; |
| 4936 | |
| 4937 | /* Try to translate a vmfail error code, if possible. */ |
| 4938 | if (vmread(VMCS_INSTRUCTION_ERROR0x4400, &insn_error)) { |
| 4939 | printf("%s: can't read insn error field\n", |
| 4940 | __func__); |
| 4941 | } else |
| 4942 | printf("%s: error code = %lld, %s\n", __func__, |
| 4943 | insn_error, |
| 4944 | vmx_instruction_error_decode(insn_error)); |
| 4945 | #ifdef VMM_DEBUG |
| 4946 | vmx_vcpu_dump_regs(vcpu); |
| 4947 | dump_vcpu(vcpu); |
| 4948 | #endif /* VMM_DEBUG */ |
| 4949 | } |
| 4950 | } |
| 4951 | |
| 4952 | vcpu->vc_last_pcpu = curcpu()({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;}); |
| 4953 | |
| 4954 | /* Copy the VCPU register state to the exit structure */ |
| 4955 | if (vcpu_readregs_vmx(vcpu, VM_RWREGS_ALL(0x1 | 0x2 | 0x4 | 0x8 | 0x10), &vcpu->vc_exit.vrs)) |
| 4956 | ret = EINVAL22; |
| 4957 | vcpu->vc_exit.cpl = vmm_get_guest_cpu_cpl(vcpu); |
| 4958 | |
| 4959 | return (ret); |
| 4960 | } |
| 4961 | |
| 4962 | /* |
| 4963 | * vmx_handle_intr |
| 4964 | * |
| 4965 | * Handle host (external) interrupts. We read which interrupt fired by |
| 4966 | * extracting the vector from the VMCS and dispatch the interrupt directly |
| 4967 | * to the host using vmm_dispatch_intr. |
| 4968 | */ |
| 4969 | void |
| 4970 | vmx_handle_intr(struct vcpu *vcpu) |
| 4971 | { |
| 4972 | uint8_t vec; |
| 4973 | uint64_t eii; |
| 4974 | struct gate_descriptor *idte; |
| 4975 | vaddr_t handler; |
| 4976 | |
| 4977 | if (vmread(VMCS_EXIT_INTERRUPTION_INFO0x4404, &eii)) { |
| 4978 | printf("%s: can't obtain intr info\n", __func__); |
| 4979 | return; |
| 4980 | } |
| 4981 | |
| 4982 | vec = eii & 0xFF; |
| 4983 | |
| 4984 | /* XXX check "error valid" code in eii, abort if 0 */ |
| 4985 | idte=&idt[vec]; |
| 4986 | handler = idte->gd_looffset + ((uint64_t)idte->gd_hioffset << 16); |
| 4987 | vmm_dispatch_intr(handler); |
| 4988 | } |
| 4989 | |
| 4990 | /* |
| 4991 | * svm_handle_hlt |
| 4992 | * |
| 4993 | * Handle HLT exits |
| 4994 | * |
| 4995 | * Parameters |
| 4996 | * vcpu: The VCPU that executed the HLT instruction |
| 4997 | * |
| 4998 | * Return Values: |
| 4999 | * EIO: The guest halted with interrupts disabled |
| 5000 | * EAGAIN: Normal return to vmd - vmd should halt scheduling this VCPU |
| 5001 | * until a virtual interrupt is ready to inject |
| 5002 | */ |
| 5003 | int |
| 5004 | svm_handle_hlt(struct vcpu *vcpu) |
| 5005 | { |
| 5006 | struct vmcb *vmcb = (struct vmcb *)vcpu->vc_control_va; |
| 5007 | uint64_t rflags = vmcb->v_rflags; |
| 5008 | |
| 5009 | /* All HLT insns are 1 byte */ |
| 5010 | vcpu->vc_gueststate.vg_rip += 1; |
| 5011 | |
| 5012 | if (!(rflags & PSL_I0x00000200)) { |
| 5013 | DPRINTF("%s: guest halted with interrupts disabled\n", |
| 5014 | __func__); |
| 5015 | return (EIO5); |
| 5016 | } |
| 5017 | |
| 5018 | return (EAGAIN35); |
| 5019 | } |
| 5020 | |
| 5021 | /* |
| 5022 | * vmx_handle_hlt |
| 5023 | * |
| 5024 | * Handle HLT exits. HLTing the CPU with interrupts disabled will terminate |
| 5025 | * the guest (no NMIs handled) by returning EIO to vmd. |
| 5026 | * |
| 5027 | * Parameters: |
| 5028 | * vcpu: The VCPU that executed the HLT instruction |
| 5029 | * |
| 5030 | * Return Values: |
| 5031 | * EINVAL: An error occurred extracting information from the VMCS, or an |
| 5032 | * invalid HLT instruction was encountered |
| 5033 | * EIO: The guest halted with interrupts disabled |
| 5034 | * EAGAIN: Normal return to vmd - vmd should halt scheduling this VCPU |
| 5035 | * until a virtual interrupt is ready to inject |
| 5036 | * |
| 5037 | */ |
| 5038 | int |
| 5039 | vmx_handle_hlt(struct vcpu *vcpu) |
| 5040 | { |
| 5041 | uint64_t insn_length, rflags; |
| 5042 | |
| 5043 | if (vmread(VMCS_INSTRUCTION_LENGTH0x440C, &insn_length)) { |
| 5044 | printf("%s: can't obtain instruction length\n", __func__); |
| 5045 | return (EINVAL22); |
| 5046 | } |
| 5047 | |
| 5048 | if (vmread(VMCS_GUEST_IA32_RFLAGS0x6820, &rflags)) { |
| 5049 | printf("%s: can't obtain guest rflags\n", __func__); |
| 5050 | return (EINVAL22); |
| 5051 | } |
| 5052 | |
| 5053 | if (insn_length != 1) { |
| 5054 | DPRINTF("%s: HLT with instruction length %lld not supported\n", |
| 5055 | __func__, insn_length); |
| 5056 | return (EINVAL22); |
| 5057 | } |
| 5058 | |
| 5059 | if (!(rflags & PSL_I0x00000200)) { |
| 5060 | DPRINTF("%s: guest halted with interrupts disabled\n", |
| 5061 | __func__); |
| 5062 | return (EIO5); |
| 5063 | } |
| 5064 | |
| 5065 | vcpu->vc_gueststate.vg_rip += insn_length; |
| 5066 | return (EAGAIN35); |
| 5067 | } |
| 5068 | |
| 5069 | /* |
| 5070 | * vmx_get_exit_info |
| 5071 | * |
| 5072 | * Returns exit information containing the current guest RIP and exit reason |
| 5073 | * in rip and exit_reason. The return value is a bitmask indicating whether |
| 5074 | * reading the RIP and exit reason was successful. |
| 5075 | */ |
| 5076 | int |
| 5077 | vmx_get_exit_info(uint64_t *rip, uint64_t *exit_reason) |
| 5078 | { |
| 5079 | int rv = 0; |
| 5080 | |
| 5081 | if (vmread(VMCS_GUEST_IA32_RIP0x681E, rip) == 0) { |
| 5082 | rv |= VMX_EXIT_INFO_HAVE_RIP0x1; |
| 5083 | if (vmread(VMCS_EXIT_REASON0x4402, exit_reason) == 0) |
| 5084 | rv |= VMX_EXIT_INFO_HAVE_REASON0x2; |
| 5085 | } |
| 5086 | return (rv); |
| 5087 | } |
| 5088 | |
| 5089 | /* |
| 5090 | * svm_handle_exit |
| 5091 | * |
| 5092 | * Handle exits from the VM by decoding the exit reason and calling various |
| 5093 | * subhandlers as needed. |
| 5094 | */ |
| 5095 | int |
| 5096 | svm_handle_exit(struct vcpu *vcpu) |
| 5097 | { |
| 5098 | uint64_t exit_reason, rflags; |
| 5099 | int update_rip, ret = 0; |
| 5100 | struct vmcb *vmcb = (struct vmcb *)vcpu->vc_control_va; |
| 5101 | |
| 5102 | update_rip = 0; |
| 5103 | exit_reason = vcpu->vc_gueststate.vg_exit_reason; |
| 5104 | rflags = vcpu->vc_gueststate.vg_rflags; |
| 5105 | |
| 5106 | switch (exit_reason) { |
| 5107 | case SVM_VMEXIT_VINTR0x64: |
| 5108 | if (!(rflags & PSL_I0x00000200)) { |
| 5109 | DPRINTF("%s: impossible interrupt window exit " |
| 5110 | "config\n", __func__); |
| 5111 | ret = EINVAL22; |
| 5112 | break; |
| 5113 | } |
| 5114 | |
| 5115 | /* |
| 5116 | * Guest is now ready for interrupts, so disable interrupt |
| 5117 | * window exiting. |
| 5118 | */ |
| 5119 | vmcb->v_irq = 0; |
| 5120 | vmcb->v_intr_vector = 0; |
| 5121 | vmcb->v_intercept1 &= ~SVM_INTERCEPT_VINTR(1UL << 4); |
| 5122 | svm_set_dirty(vcpu, SVM_CLEANBITS_TPR(1 << 3) | SVM_CLEANBITS_I(1 << 0)); |
| 5123 | |
| 5124 | update_rip = 0; |
| 5125 | break; |
| 5126 | case SVM_VMEXIT_INTR0x60: |
| 5127 | update_rip = 0; |
| 5128 | break; |
| 5129 | case SVM_VMEXIT_SHUTDOWN0x7F: |
| 5130 | update_rip = 0; |
| 5131 | ret = EAGAIN35; |
| 5132 | break; |
| 5133 | case SVM_VMEXIT_NPF0x400: |
| 5134 | ret = svm_handle_np_fault(vcpu); |
| 5135 | break; |
| 5136 | case SVM_VMEXIT_CPUID0x72: |
| 5137 | ret = vmm_handle_cpuid(vcpu); |
| 5138 | update_rip = 1; |
| 5139 | break; |
| 5140 | case SVM_VMEXIT_MSR0x7C: |
| 5141 | ret = svm_handle_msr(vcpu); |
| 5142 | update_rip = 1; |
| 5143 | break; |
| 5144 | case SVM_VMEXIT_XSETBV0x8D: |
| 5145 | ret = svm_handle_xsetbv(vcpu); |
| 5146 | update_rip = 1; |
| 5147 | break; |
| 5148 | case SVM_VMEXIT_IOIO0x7B: |
| 5149 | ret = svm_handle_inout(vcpu); |
| 5150 | update_rip = 1; |
| 5151 | break; |
| 5152 | case SVM_VMEXIT_HLT0x78: |
| 5153 | ret = svm_handle_hlt(vcpu); |
| 5154 | update_rip = 1; |
| 5155 | break; |
| 5156 | case SVM_VMEXIT_MWAIT0x8B: |
| 5157 | case SVM_VMEXIT_MWAIT_CONDITIONAL0x8C: |
| 5158 | case SVM_VMEXIT_MONITOR0x8A: |
| 5159 | case SVM_VMEXIT_VMRUN0x80: |
| 5160 | case SVM_VMEXIT_VMMCALL0x81: |
| 5161 | case SVM_VMEXIT_VMLOAD0x82: |
| 5162 | case SVM_VMEXIT_VMSAVE0x83: |
| 5163 | case SVM_VMEXIT_STGI0x84: |
| 5164 | case SVM_VMEXIT_CLGI0x85: |
| 5165 | case SVM_VMEXIT_SKINIT0x86: |
| 5166 | case SVM_VMEXIT_RDTSCP0x87: |
| 5167 | case SVM_VMEXIT_ICEBP0x88: |
| 5168 | case SVM_VMEXIT_INVLPGA0x7A: |
| 5169 | ret = vmm_inject_ud(vcpu); |
| 5170 | update_rip = 0; |
| 5171 | break; |
| 5172 | default: |
| 5173 | DPRINTF("%s: unhandled exit 0x%llx (pa=0x%llx)\n", __func__, |
| 5174 | exit_reason, (uint64_t)vcpu->vc_control_pa); |
| 5175 | return (EINVAL22); |
| 5176 | } |
| 5177 | |
| 5178 | if (update_rip) { |
| 5179 | vmcb->v_rip = vcpu->vc_gueststate.vg_rip; |
| 5180 | |
| 5181 | if (rflags & PSL_T0x00000100) { |
| 5182 | if (vmm_inject_db(vcpu)) { |
| 5183 | printf("%s: can't inject #DB exception to " |
| 5184 | "guest", __func__); |
| 5185 | return (EINVAL22); |
| 5186 | } |
| 5187 | } |
| 5188 | } |
| 5189 | |
| 5190 | /* Enable SVME in EFER (must always be set) */ |
| 5191 | vmcb->v_efer |= EFER_SVME0x00001000; |
| 5192 | svm_set_dirty(vcpu, SVM_CLEANBITS_CR(1 << 5)); |
| 5193 | |
| 5194 | return (ret); |
| 5195 | } |
| 5196 | |
| 5197 | /* |
| 5198 | * vmx_handle_exit |
| 5199 | * |
| 5200 | * Handle exits from the VM by decoding the exit reason and calling various |
| 5201 | * subhandlers as needed. |
| 5202 | */ |
| 5203 | int |
| 5204 | vmx_handle_exit(struct vcpu *vcpu) |
| 5205 | { |
| 5206 | uint64_t exit_reason, rflags, istate; |
| 5207 | int update_rip, ret = 0; |
| 5208 | |
| 5209 | update_rip = 0; |
| 5210 | exit_reason = vcpu->vc_gueststate.vg_exit_reason; |
| 5211 | rflags = vcpu->vc_gueststate.vg_rflags; |
| 5212 | |
| 5213 | switch (exit_reason) { |
| 5214 | case VMX_EXIT_INT_WINDOW7: |
| 5215 | if (!(rflags & PSL_I0x00000200)) { |
| 5216 | DPRINTF("%s: impossible interrupt window exit " |
| 5217 | "config\n", __func__); |
| 5218 | ret = EINVAL22; |
| 5219 | break; |
| 5220 | } |
| 5221 | |
| 5222 | ret = EAGAIN35; |
| 5223 | update_rip = 0; |
| 5224 | break; |
| 5225 | case VMX_EXIT_EPT_VIOLATION48: |
| 5226 | ret = vmx_handle_np_fault(vcpu); |
| 5227 | break; |
| 5228 | case VMX_EXIT_CPUID10: |
| 5229 | ret = vmm_handle_cpuid(vcpu); |
| 5230 | update_rip = 1; |
| 5231 | break; |
| 5232 | case VMX_EXIT_IO30: |
| 5233 | ret = vmx_handle_inout(vcpu); |
| 5234 | update_rip = 1; |
| 5235 | break; |
| 5236 | case VMX_EXIT_EXTINT1: |
| 5237 | vmx_handle_intr(vcpu); |
| 5238 | update_rip = 0; |
| 5239 | break; |
| 5240 | case VMX_EXIT_CR_ACCESS28: |
| 5241 | ret = vmx_handle_cr(vcpu); |
| 5242 | update_rip = 1; |
| 5243 | break; |
| 5244 | case VMX_EXIT_HLT12: |
| 5245 | ret = vmx_handle_hlt(vcpu); |
| 5246 | update_rip = 1; |
| 5247 | break; |
| 5248 | case VMX_EXIT_RDMSR31: |
| 5249 | ret = vmx_handle_rdmsr(vcpu); |
| 5250 | update_rip = 1; |
| 5251 | break; |
| 5252 | case VMX_EXIT_WRMSR32: |
| 5253 | ret = vmx_handle_wrmsr(vcpu); |
| 5254 | update_rip = 1; |
| 5255 | break; |
| 5256 | case VMX_EXIT_XSETBV55: |
| 5257 | ret = vmx_handle_xsetbv(vcpu); |
| 5258 | update_rip = 1; |
| 5259 | break; |
| 5260 | case VMX_EXIT_MWAIT36: |
| 5261 | case VMX_EXIT_MONITOR39: |
| 5262 | case VMX_EXIT_VMXON27: |
| 5263 | case VMX_EXIT_VMWRITE25: |
| 5264 | case VMX_EXIT_VMREAD23: |
| 5265 | case VMX_EXIT_VMLAUNCH20: |
| 5266 | case VMX_EXIT_VMRESUME24: |
| 5267 | case VMX_EXIT_VMPTRLD21: |
| 5268 | case VMX_EXIT_VMPTRST22: |
| 5269 | case VMX_EXIT_VMCLEAR19: |
| 5270 | case VMX_EXIT_VMCALL18: |
| 5271 | case VMX_EXIT_VMFUNC59: |
| 5272 | case VMX_EXIT_VMXOFF26: |
| 5273 | case VMX_EXIT_INVVPID53: |
| 5274 | case VMX_EXIT_INVEPT50: |
| 5275 | ret = vmm_inject_ud(vcpu); |
| 5276 | update_rip = 0; |
| 5277 | break; |
| 5278 | case VMX_EXIT_TRIPLE_FAULT2: |
| 5279 | #ifdef VMM_DEBUG |
| 5280 | DPRINTF("%s: vm %d vcpu %d triple fault\n", __func__, |
| 5281 | vcpu->vc_parent->vm_id, vcpu->vc_id); |
| 5282 | vmx_vcpu_dump_regs(vcpu); |
| 5283 | dump_vcpu(vcpu); |
| 5284 | vmx_dump_vmcs(vcpu); |
| 5285 | #endif /* VMM_DEBUG */ |
| 5286 | ret = EAGAIN35; |
| 5287 | update_rip = 0; |
| 5288 | break; |
| 5289 | default: |
| 5290 | #ifdef VMM_DEBUG |
| 5291 | DPRINTF("%s: unhandled exit 0x%llx (%s)\n", __func__, |
| 5292 | exit_reason, vmx_exit_reason_decode(exit_reason)); |
| 5293 | #endif /* VMM_DEBUG */ |
| 5294 | return (EINVAL22); |
| 5295 | } |
| 5296 | |
| 5297 | if (update_rip) { |
| 5298 | if (vmwrite(VMCS_GUEST_IA32_RIP0x681E, |
| 5299 | vcpu->vc_gueststate.vg_rip)) { |
| 5300 | printf("%s: can't advance rip\n", __func__); |
| 5301 | return (EINVAL22); |
| 5302 | } |
| 5303 | |
| 5304 | if (vmread(VMCS_GUEST_INTERRUPTIBILITY_ST0x4824, |
| 5305 | &istate)) { |
| 5306 | printf("%s: can't read interruptibility state\n", |
| 5307 | __func__); |
| 5308 | return (EINVAL22); |
| 5309 | } |
| 5310 | |
| 5311 | /* Interruptibility state 0x3 covers NMIs and STI */ |
| 5312 | istate &= ~0x3; |
| 5313 | |
| 5314 | if (vmwrite(VMCS_GUEST_INTERRUPTIBILITY_ST0x4824, |
| 5315 | istate)) { |
| 5316 | printf("%s: can't write interruptibility state\n", |
| 5317 | __func__); |
| 5318 | return (EINVAL22); |
| 5319 | } |
| 5320 | |
| 5321 | if (rflags & PSL_T0x00000100) { |
| 5322 | if (vmm_inject_db(vcpu)) { |
| 5323 | printf("%s: can't inject #DB exception to " |
| 5324 | "guest", __func__); |
| 5325 | return (EINVAL22); |
| 5326 | } |
| 5327 | } |
| 5328 | } |
| 5329 | |
| 5330 | return (ret); |
| 5331 | } |
| 5332 | |
| 5333 | /* |
| 5334 | * vmm_inject_gp |
| 5335 | * |
| 5336 | * Injects an #GP exception into the guest VCPU. |
| 5337 | * |
| 5338 | * Parameters: |
| 5339 | * vcpu: vcpu to inject into |
| 5340 | * |
| 5341 | * Return values: |
| 5342 | * Always 0 |
| 5343 | */ |
| 5344 | int |
| 5345 | vmm_inject_gp(struct vcpu *vcpu) |
| 5346 | { |
| 5347 | DPRINTF("%s: injecting #GP at guest %%rip 0x%llx\n", __func__, |
| 5348 | vcpu->vc_gueststate.vg_rip); |
| 5349 | vcpu->vc_event = VMM_EX_GP13; |
| 5350 | |
| 5351 | return (0); |
| 5352 | } |
| 5353 | |
| 5354 | /* |
| 5355 | * vmm_inject_ud |
| 5356 | * |
| 5357 | * Injects an #UD exception into the guest VCPU. |
| 5358 | * |
| 5359 | * Parameters: |
| 5360 | * vcpu: vcpu to inject into |
| 5361 | * |
| 5362 | * Return values: |
| 5363 | * Always 0 |
| 5364 | */ |
| 5365 | int |
| 5366 | vmm_inject_ud(struct vcpu *vcpu) |
| 5367 | { |
| 5368 | DPRINTF("%s: injecting #UD at guest %%rip 0x%llx\n", __func__, |
| 5369 | vcpu->vc_gueststate.vg_rip); |
| 5370 | vcpu->vc_event = VMM_EX_UD6; |
| 5371 | |
| 5372 | return (0); |
| 5373 | } |
| 5374 | |
| 5375 | /* |
| 5376 | * vmm_inject_db |
| 5377 | * |
| 5378 | * Injects a #DB exception into the guest VCPU. |
| 5379 | * |
| 5380 | * Parameters: |
| 5381 | * vcpu: vcpu to inject into |
| 5382 | * |
| 5383 | * Return values: |
| 5384 | * Always 0 |
| 5385 | */ |
| 5386 | int |
| 5387 | vmm_inject_db(struct vcpu *vcpu) |
| 5388 | { |
| 5389 | DPRINTF("%s: injecting #DB at guest %%rip 0x%llx\n", __func__, |
| 5390 | vcpu->vc_gueststate.vg_rip); |
| 5391 | vcpu->vc_event = VMM_EX_DB1; |
| 5392 | |
| 5393 | return (0); |
| 5394 | } |
| 5395 | |
| 5396 | /* |
| 5397 | * vmm_get_guest_memtype |
| 5398 | * |
| 5399 | * Returns the type of memory 'gpa' refers to in the context of vm 'vm' |
| 5400 | */ |
| 5401 | int |
| 5402 | vmm_get_guest_memtype(struct vm *vm, paddr_t gpa) |
| 5403 | { |
| 5404 | int i; |
| 5405 | struct vm_mem_range *vmr; |
| 5406 | |
| 5407 | if (gpa >= VMM_PCI_MMIO_BAR_BASE0xF0000000ULL && gpa <= VMM_PCI_MMIO_BAR_END0xFFFFFFFFULL) { |
| 5408 | DPRINTF("guest mmio access @ 0x%llx\n", (uint64_t)gpa); |
| 5409 | return (VMM_MEM_TYPE_REGULAR); |
| 5410 | } |
| 5411 | |
| 5412 | /* XXX Use binary search? */ |
| 5413 | for (i = 0; i < vm->vm_nmemranges; i++) { |
| 5414 | vmr = &vm->vm_memranges[i]; |
| 5415 | |
| 5416 | /* |
| 5417 | * vm_memranges are ascending. gpa can no longer be in one of |
| 5418 | * the memranges |
| 5419 | */ |
| 5420 | if (gpa < vmr->vmr_gpa) |
| 5421 | break; |
| 5422 | |
| 5423 | if (gpa < vmr->vmr_gpa + vmr->vmr_size) |
| 5424 | return (VMM_MEM_TYPE_REGULAR); |
| 5425 | } |
| 5426 | |
| 5427 | DPRINTF("guest memtype @ 0x%llx unknown\n", (uint64_t)gpa); |
| 5428 | return (VMM_MEM_TYPE_UNKNOWN); |
| 5429 | } |
| 5430 | |
| 5431 | /* |
| 5432 | * vmx_get_exit_qualification |
| 5433 | * |
| 5434 | * Return the current VMCS' exit qualification information |
| 5435 | */ |
| 5436 | int |
| 5437 | vmx_get_exit_qualification(uint64_t *exit_qualification) |
| 5438 | { |
| 5439 | if (vmread(VMCS_GUEST_EXIT_QUALIFICATION0x6400, exit_qualification)) { |
| 5440 | printf("%s: can't extract exit qual\n", __func__); |
| 5441 | return (EINVAL22); |
| 5442 | } |
| 5443 | |
| 5444 | return (0); |
| 5445 | } |
| 5446 | |
| 5447 | /* |
| 5448 | * vmx_get_guest_faulttype |
| 5449 | * |
| 5450 | * Determines the type (R/W/X) of the last fault on the VCPU last run on |
| 5451 | * this PCPU. |
| 5452 | */ |
| 5453 | int |
| 5454 | vmx_get_guest_faulttype(void) |
| 5455 | { |
| 5456 | uint64_t exit_qual; |
| 5457 | uint64_t presentmask = IA32_VMX_EPT_FAULT_WAS_READABLE(1ULL << 3) | |
| 5458 | IA32_VMX_EPT_FAULT_WAS_WRITABLE(1ULL << 4) | IA32_VMX_EPT_FAULT_WAS_EXECABLE(1ULL << 5); |
| 5459 | vm_prot_t prot, was_prot; |
| 5460 | |
| 5461 | if (vmx_get_exit_qualification(&exit_qual)) |
| 5462 | return (-1); |
| 5463 | |
| 5464 | if ((exit_qual & presentmask) == 0) |
| 5465 | return VM_FAULT_INVALID((vm_fault_t) 0x0); |
| 5466 | |
| 5467 | was_prot = 0; |
| 5468 | if (exit_qual & IA32_VMX_EPT_FAULT_WAS_READABLE(1ULL << 3)) |
| 5469 | was_prot |= PROT_READ0x01; |
| 5470 | if (exit_qual & IA32_VMX_EPT_FAULT_WAS_WRITABLE(1ULL << 4)) |
| 5471 | was_prot |= PROT_WRITE0x02; |
| 5472 | if (exit_qual & IA32_VMX_EPT_FAULT_WAS_EXECABLE(1ULL << 5)) |
| 5473 | was_prot |= PROT_EXEC0x04; |
| 5474 | |
| 5475 | prot = 0; |
| 5476 | if (exit_qual & IA32_VMX_EPT_FAULT_READ(1ULL << 0)) |
| 5477 | prot = PROT_READ0x01; |
| 5478 | else if (exit_qual & IA32_VMX_EPT_FAULT_WRITE(1ULL << 1)) |
| 5479 | prot = PROT_WRITE0x02; |
| 5480 | else if (exit_qual & IA32_VMX_EPT_FAULT_EXEC(1ULL << 2)) |
| 5481 | prot = PROT_EXEC0x04; |
| 5482 | |
| 5483 | if ((was_prot & prot) == 0) |
| 5484 | return VM_FAULT_PROTECT((vm_fault_t) 0x1); |
| 5485 | |
| 5486 | return (-1); |
| 5487 | } |
| 5488 | |
| 5489 | /* |
| 5490 | * svm_get_guest_faulttype |
| 5491 | * |
| 5492 | * Determines the type (R/W/X) of the last fault on the VCPU last run on |
| 5493 | * this PCPU. |
| 5494 | */ |
| 5495 | int |
| 5496 | svm_get_guest_faulttype(struct vmcb *vmcb) |
| 5497 | { |
| 5498 | if (!(vmcb->v_exitinfo1 & 0x1)) |
| 5499 | return VM_FAULT_INVALID((vm_fault_t) 0x0); |
| 5500 | return VM_FAULT_PROTECT((vm_fault_t) 0x1); |
| 5501 | } |
| 5502 | |
| 5503 | /* |
| 5504 | * svm_fault_page |
| 5505 | * |
| 5506 | * Request a new page to be faulted into the UVM map of the VM owning 'vcpu' |
| 5507 | * at address 'gpa'. |
| 5508 | */ |
| 5509 | int |
| 5510 | svm_fault_page(struct vcpu *vcpu, paddr_t gpa) |
| 5511 | { |
| 5512 | int ret; |
| 5513 | |
| 5514 | ret = uvm_fault(vcpu->vc_parent->vm_map, gpa, VM_FAULT_WIRE((vm_fault_t) 0x2), |
| 5515 | PROT_READ0x01 | PROT_WRITE0x02 | PROT_EXEC0x04); |
| 5516 | if (ret) |
| 5517 | printf("%s: uvm_fault returns %d, GPA=0x%llx, rip=0x%llx\n", |
| 5518 | __func__, ret, (uint64_t)gpa, vcpu->vc_gueststate.vg_rip); |
| 5519 | |
| 5520 | return (ret); |
| 5521 | } |
| 5522 | |
| 5523 | /* |
| 5524 | * svm_handle_np_fault |
| 5525 | * |
| 5526 | * High level nested paging handler for SVM. Verifies that a fault is for a |
| 5527 | * valid memory region, then faults a page, or aborts otherwise. |
| 5528 | */ |
| 5529 | int |
| 5530 | svm_handle_np_fault(struct vcpu *vcpu) |
| 5531 | { |
| 5532 | uint64_t gpa; |
| 5533 | int gpa_memtype, ret; |
| 5534 | struct vmcb *vmcb = (struct vmcb *)vcpu->vc_control_va; |
| 5535 | |
| 5536 | ret = 0; |
| 5537 | |
| 5538 | gpa = vmcb->v_exitinfo2; |
| 5539 | |
| 5540 | gpa_memtype = vmm_get_guest_memtype(vcpu->vc_parent, gpa); |
| 5541 | switch (gpa_memtype) { |
| 5542 | case VMM_MEM_TYPE_REGULAR: |
| 5543 | ret = svm_fault_page(vcpu, gpa); |
| 5544 | break; |
| 5545 | default: |
| 5546 | printf("unknown memory type %d for GPA 0x%llx\n", |
| 5547 | gpa_memtype, gpa); |
| 5548 | return (EINVAL22); |
| 5549 | } |
| 5550 | |
| 5551 | return (ret); |
| 5552 | } |
| 5553 | |
| 5554 | /* |
| 5555 | * vmx_fault_page |
| 5556 | * |
| 5557 | * Request a new page to be faulted into the UVM map of the VM owning 'vcpu' |
| 5558 | * at address 'gpa'. |
| 5559 | * |
| 5560 | * Parameters: |
| 5561 | * vcpu: guest VCPU requiring the page to be faulted into the UVM map |
| 5562 | * gpa: guest physical address that triggered the fault |
| 5563 | * |
| 5564 | * Return Values: |
| 5565 | * 0: if successful |
| 5566 | * EINVAL: if fault type could not be determined or VMCS reload fails |
| 5567 | * EAGAIN: if a protection fault occurred, ie writing to a read-only page |
| 5568 | * errno: if uvm_fault(9) fails to wire in the page |
| 5569 | */ |
| 5570 | int |
| 5571 | vmx_fault_page(struct vcpu *vcpu, paddr_t gpa) |
| 5572 | { |
| 5573 | int fault_type, ret; |
| 5574 | |
| 5575 | fault_type = vmx_get_guest_faulttype(); |
| 5576 | if (fault_type == -1) { |
| 5577 | printf("%s: invalid fault type\n", __func__); |
| 5578 | return (EINVAL22); |
| 5579 | } |
| 5580 | |
| 5581 | if (fault_type == VM_FAULT_PROTECT((vm_fault_t) 0x1)) { |
| 5582 | vcpu->vc_exit.vee.vee_fault_type = VEE_FAULT_PROTECT; |
| 5583 | return (EAGAIN35); |
| 5584 | } |
| 5585 | |
| 5586 | /* We may sleep during uvm_fault(9), so reload VMCS. */ |
| 5587 | vcpu->vc_last_pcpu = curcpu()({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;}); |
| 5588 | ret = uvm_fault(vcpu->vc_parent->vm_map, gpa, VM_FAULT_WIRE((vm_fault_t) 0x2), |
| 5589 | PROT_READ0x01 | PROT_WRITE0x02 | PROT_EXEC0x04); |
| 5590 | if (vcpu_reload_vmcs_vmx(vcpu)) { |
| 5591 | printf("%s: failed to reload vmcs\n", __func__); |
| 5592 | return (EINVAL22); |
| 5593 | } |
| 5594 | |
| 5595 | if (ret) |
| 5596 | printf("%s: uvm_fault returns %d, GPA=0x%llx, rip=0x%llx\n", |
| 5597 | __func__, ret, (uint64_t)gpa, vcpu->vc_gueststate.vg_rip); |
| 5598 | |
| 5599 | return (ret); |
| 5600 | } |
| 5601 | |
| 5602 | /* |
| 5603 | * vmx_handle_np_fault |
| 5604 | * |
| 5605 | * High level nested paging handler for VMX. Verifies that a fault is for a |
| 5606 | * valid memory region, then faults a page, or aborts otherwise. |
| 5607 | */ |
| 5608 | int |
| 5609 | vmx_handle_np_fault(struct vcpu *vcpu) |
| 5610 | { |
| 5611 | uint64_t gpa; |
| 5612 | int gpa_memtype, ret; |
| 5613 | |
| 5614 | ret = 0; |
| 5615 | if (vmread(VMCS_GUEST_PHYSICAL_ADDRESS0x2400, &gpa)) { |
| 5616 | printf("%s: cannot extract faulting pa\n", __func__); |
| 5617 | return (EINVAL22); |
| 5618 | } |
| 5619 | |
| 5620 | gpa_memtype = vmm_get_guest_memtype(vcpu->vc_parent, gpa); |
| 5621 | switch (gpa_memtype) { |
| 5622 | case VMM_MEM_TYPE_REGULAR: |
| 5623 | ret = vmx_fault_page(vcpu, gpa); |
| 5624 | break; |
| 5625 | default: |
| 5626 | printf("unknown memory type %d for GPA 0x%llx\n", |
| 5627 | gpa_memtype, gpa); |
| 5628 | return (EINVAL22); |
| 5629 | } |
| 5630 | |
| 5631 | return (ret); |
| 5632 | } |
| 5633 | |
| 5634 | /* |
| 5635 | * vmm_get_guest_cpu_cpl |
| 5636 | * |
| 5637 | * Determines current CPL of 'vcpu'. On VMX/Intel, this is gathered from the |
| 5638 | * VMCS field for the DPL of SS (this seems odd, but is documented that way |
| 5639 | * in the SDM). For SVM/AMD, this is gathered directly from the VMCB's 'cpl' |
| 5640 | * field, as per the APM. |
| 5641 | * |
| 5642 | * Parameters: |
| 5643 | * vcpu: guest VCPU for which CPL is to be checked |
| 5644 | * |
| 5645 | * Return Values: |
| 5646 | * -1: the CPL could not be determined |
| 5647 | * 0-3 indicating the current CPL. For real mode operation, 0 is returned. |
| 5648 | */ |
| 5649 | int |
| 5650 | vmm_get_guest_cpu_cpl(struct vcpu *vcpu) |
| 5651 | { |
| 5652 | int mode; |
| 5653 | struct vmcb *vmcb; |
| 5654 | uint64_t ss_ar; |
| 5655 | |
| 5656 | mode = vmm_get_guest_cpu_mode(vcpu); |
| 5657 | |
| 5658 | if (mode == VMM_CPU_MODE_UNKNOWN) |
| 5659 | return (-1); |
| 5660 | |
| 5661 | if (mode == VMM_CPU_MODE_REAL) |
| 5662 | return (0); |
| 5663 | |
| 5664 | if (vmm_softc->mode == VMM_MODE_SVM || |
| 5665 | vmm_softc->mode == VMM_MODE_RVI) { |
| 5666 | vmcb = (struct vmcb *)vcpu->vc_control_va; |
| 5667 | return (vmcb->v_cpl); |
| 5668 | } else if (vmm_softc->mode == VMM_MODE_VMX || |
| 5669 | vmm_softc->mode == VMM_MODE_EPT) { |
| 5670 | if (vmread(VMCS_GUEST_IA32_SS_AR0x4818, &ss_ar)) |
| 5671 | return (-1); |
| 5672 | return ((ss_ar & 0x60) >> 5); |
| 5673 | } else |
| 5674 | return (-1); |
| 5675 | } |
| 5676 | |
| 5677 | /* |
| 5678 | * vmm_get_guest_cpu_mode |
| 5679 | * |
| 5680 | * Determines current CPU mode of 'vcpu'. |
| 5681 | * |
| 5682 | * Parameters: |
| 5683 | * vcpu: guest VCPU for which mode is to be checked |
| 5684 | * |
| 5685 | * Return Values: |
| 5686 | * One of VMM_CPU_MODE_*, or VMM_CPU_MODE_UNKNOWN if the mode could not be |
| 5687 | * ascertained. |
| 5688 | */ |
| 5689 | int |
| 5690 | vmm_get_guest_cpu_mode(struct vcpu *vcpu) |
| 5691 | { |
| 5692 | uint64_t cr0, efer, cs_ar; |
| 5693 | uint8_t l, dib; |
| 5694 | struct vmcb *vmcb; |
| 5695 | struct vmx_msr_store *msr_store; |
| 5696 | |
| 5697 | if (vmm_softc->mode == VMM_MODE_SVM || |
| 5698 | vmm_softc->mode == VMM_MODE_RVI) { |
| 5699 | vmcb = (struct vmcb *)vcpu->vc_control_va; |
| 5700 | cr0 = vmcb->v_cr0; |
| 5701 | efer = vmcb->v_efer; |
| 5702 | cs_ar = vmcb->v_cs.vs_attr; |
| 5703 | cs_ar = (cs_ar & 0xff) | ((cs_ar << 4) & 0xf000); |
| 5704 | } else if (vmm_softc->mode == VMM_MODE_VMX || |
| 5705 | vmm_softc->mode == VMM_MODE_EPT) { |
| 5706 | if (vmread(VMCS_GUEST_IA32_CR00x6800, &cr0)) |
| 5707 | return (VMM_CPU_MODE_UNKNOWN); |
| 5708 | if (vmread(VMCS_GUEST_IA32_CS_AR0x4816, &cs_ar)) |
| 5709 | return (VMM_CPU_MODE_UNKNOWN); |
| 5710 | msr_store = |
| 5711 | (struct vmx_msr_store *)vcpu->vc_vmx_msr_exit_save_va; |
| 5712 | efer = msr_store[VCPU_REGS_EFER0].vms_data; |
| 5713 | } else |
| 5714 | return (VMM_CPU_MODE_UNKNOWN); |
| 5715 | |
| 5716 | l = (cs_ar & 0x2000) >> 13; |
| 5717 | dib = (cs_ar & 0x4000) >> 14; |
| 5718 | |
| 5719 | /* Check CR0.PE */ |
| 5720 | if (!(cr0 & CR0_PE0x00000001)) |
| 5721 | return (VMM_CPU_MODE_REAL); |
| 5722 | |
| 5723 | /* Check EFER */ |
| 5724 | if (efer & EFER_LMA0x00000400) { |
| 5725 | /* Could be compat or long mode, check CS.L */ |
| 5726 | if (l) |
| 5727 | return (VMM_CPU_MODE_LONG); |
| 5728 | else |
| 5729 | return (VMM_CPU_MODE_COMPAT); |
| 5730 | } |
| 5731 | |
| 5732 | /* Check prot vs prot32 */ |
| 5733 | if (dib) |
| 5734 | return (VMM_CPU_MODE_PROT32); |
| 5735 | else |
| 5736 | return (VMM_CPU_MODE_PROT); |
| 5737 | } |
| 5738 | |
| 5739 | /* |
| 5740 | * svm_handle_inout |
| 5741 | * |
| 5742 | * Exit handler for IN/OUT instructions. |
| 5743 | * |
| 5744 | * The vmm can handle certain IN/OUTS without exiting to vmd, but most of these |
| 5745 | * will be passed to vmd for completion. |
| 5746 | * |
| 5747 | * Parameters: |
| 5748 | * vcpu: The VCPU where the IN/OUT instruction occurred |
| 5749 | * |
| 5750 | * Return values: |
| 5751 | * 0: if successful |
| 5752 | * EINVAL: an invalid IN/OUT instruction was encountered |
| 5753 | * EAGAIN: return to vmd - more processing needed in userland |
| 5754 | */ |
| 5755 | int |
| 5756 | svm_handle_inout(struct vcpu *vcpu) |
| 5757 | { |
| 5758 | uint64_t insn_length, exit_qual; |
| 5759 | int ret; |
| 5760 | struct vmcb *vmcb = (struct vmcb *)vcpu->vc_control_va; |
| 5761 | |
| 5762 | insn_length = vmcb->v_exitinfo2 - vmcb->v_rip; |
| 5763 | if (insn_length != 1 && insn_length != 2) { |
| 5764 | DPRINTF("%s: IN/OUT instruction with length %lld not " |
| 5765 | "supported\n", __func__, insn_length); |
| 5766 | return (EINVAL22); |
| 5767 | } |
| 5768 | |
| 5769 | exit_qual = vmcb->v_exitinfo1; |
| 5770 | |
| 5771 | /* Bit 0 - direction */ |
| 5772 | vcpu->vc_exit.vei.vei_dir = (exit_qual & 0x1); |
| 5773 | /* Bit 2 - string instruction? */ |
| 5774 | vcpu->vc_exit.vei.vei_string = (exit_qual & 0x4) >> 2; |
| 5775 | /* Bit 3 - REP prefix? */ |
| 5776 | vcpu->vc_exit.vei.vei_rep = (exit_qual & 0x8) >> 3; |
| 5777 | |
| 5778 | /* Bits 4:6 - size of exit */ |
| 5779 | if (exit_qual & 0x10) |
| 5780 | vcpu->vc_exit.vei.vei_size = 1; |
| 5781 | else if (exit_qual & 0x20) |
| 5782 | vcpu->vc_exit.vei.vei_size = 2; |
| 5783 | else if (exit_qual & 0x40) |
| 5784 | vcpu->vc_exit.vei.vei_size = 4; |
| 5785 | |
| 5786 | /* Bit 16:31 - port */ |
| 5787 | vcpu->vc_exit.vei.vei_port = (exit_qual & 0xFFFF0000) >> 16; |
| 5788 | /* Data */ |
| 5789 | vcpu->vc_exit.vei.vei_data = vmcb->v_rax; |
| 5790 | |
| 5791 | vcpu->vc_gueststate.vg_rip += insn_length; |
| 5792 | |
| 5793 | /* |
| 5794 | * The following ports usually belong to devices owned by vmd. |
| 5795 | * Return EAGAIN to signal help needed from userspace (vmd). |
| 5796 | * Return 0 to indicate we don't care about this port. |
| 5797 | * |
| 5798 | * XXX something better than a hardcoded list here, maybe |
| 5799 | * configure via vmd via the device list in vm create params? |
| 5800 | */ |
| 5801 | switch (vcpu->vc_exit.vei.vei_port) { |
| 5802 | case IO_ICU10x020 ... IO_ICU10x020 + 1: |
| 5803 | case 0x40 ... 0x43: |
| 5804 | case PCKBC_AUX0x61: |
| 5805 | case IO_RTC0x070 ... IO_RTC0x070 + 1: |
| 5806 | case IO_ICU20x0A0 ... IO_ICU20x0A0 + 1: |
| 5807 | case 0x3f8 ... 0x3ff: |
| 5808 | case ELCR00x4D0 ... ELCR10x4D1: |
| 5809 | case 0x500 ... 0x511: |
| 5810 | case 0x514: |
| 5811 | case 0x518: |
| 5812 | case 0xcf8: |
| 5813 | case 0xcfc ... 0xcff: |
| 5814 | case VMM_PCI_IO_BAR_BASE0x1000 ... VMM_PCI_IO_BAR_END0xFFFF: |
| 5815 | ret = EAGAIN35; |
| 5816 | break; |
| 5817 | default: |
| 5818 | /* Read from unsupported ports returns FFs */ |
| 5819 | if (vcpu->vc_exit.vei.vei_dir == 1) { |
| 5820 | switch(vcpu->vc_exit.vei.vei_size) { |
| 5821 | case 1: |
| 5822 | vcpu->vc_gueststate.vg_rax |= 0xFF; |
| 5823 | vmcb->v_rax |= 0xFF; |
| 5824 | break; |
| 5825 | case 2: |
| 5826 | vcpu->vc_gueststate.vg_rax |= 0xFFFF; |
| 5827 | vmcb->v_rax |= 0xFFFF; |
| 5828 | break; |
| 5829 | case 4: |
| 5830 | vcpu->vc_gueststate.vg_rax |= 0xFFFFFFFF; |
| 5831 | vmcb->v_rax |= 0xFFFFFFFF; |
| 5832 | break; |
| 5833 | } |
| 5834 | } |
| 5835 | ret = 0; |
| 5836 | } |
| 5837 | |
| 5838 | return (ret); |
| 5839 | } |
| 5840 | |
| 5841 | /* |
| 5842 | * vmx_handle_inout |
| 5843 | * |
| 5844 | * Exit handler for IN/OUT instructions. |
| 5845 | * |
| 5846 | * The vmm can handle certain IN/OUTS without exiting to vmd, but most of these |
| 5847 | * will be passed to vmd for completion. |
| 5848 | */ |
| 5849 | int |
| 5850 | vmx_handle_inout(struct vcpu *vcpu) |
| 5851 | { |
| 5852 | uint64_t insn_length, exit_qual; |
| 5853 | int ret; |
| 5854 | |
| 5855 | if (vmread(VMCS_INSTRUCTION_LENGTH0x440C, &insn_length)) { |
| 5856 | printf("%s: can't obtain instruction length\n", __func__); |
| 5857 | return (EINVAL22); |
| 5858 | } |
| 5859 | |
| 5860 | if (insn_length != 1 && insn_length != 2) { |
| 5861 | DPRINTF("%s: IN/OUT instruction with length %lld not " |
| 5862 | "supported\n", __func__, insn_length); |
| 5863 | return (EINVAL22); |
| 5864 | } |
| 5865 | |
| 5866 | if (vmx_get_exit_qualification(&exit_qual)) { |
| 5867 | printf("%s: can't get exit qual\n", __func__); |
| 5868 | return (EINVAL22); |
| 5869 | } |
| 5870 | |
| 5871 | /* Bits 0:2 - size of exit */ |
| 5872 | vcpu->vc_exit.vei.vei_size = (exit_qual & 0x7) + 1; |
| 5873 | /* Bit 3 - direction */ |
| 5874 | vcpu->vc_exit.vei.vei_dir = (exit_qual & 0x8) >> 3; |
| 5875 | /* Bit 4 - string instruction? */ |
| 5876 | vcpu->vc_exit.vei.vei_string = (exit_qual & 0x10) >> 4; |
| 5877 | /* Bit 5 - REP prefix? */ |
| 5878 | vcpu->vc_exit.vei.vei_rep = (exit_qual & 0x20) >> 5; |
| 5879 | /* Bit 6 - Operand encoding */ |
| 5880 | vcpu->vc_exit.vei.vei_encoding = (exit_qual & 0x40) >> 6; |
| 5881 | /* Bit 16:31 - port */ |
| 5882 | vcpu->vc_exit.vei.vei_port = (exit_qual & 0xFFFF0000) >> 16; |
| 5883 | /* Data */ |
| 5884 | vcpu->vc_exit.vei.vei_data = (uint32_t)vcpu->vc_gueststate.vg_rax; |
| 5885 | |
| 5886 | vcpu->vc_gueststate.vg_rip += insn_length; |
| 5887 | |
| 5888 | /* |
| 5889 | * The following ports usually belong to devices owned by vmd. |
| 5890 | * Return EAGAIN to signal help needed from userspace (vmd). |
| 5891 | * Return 0 to indicate we don't care about this port. |
| 5892 | * |
| 5893 | * XXX something better than a hardcoded list here, maybe |
| 5894 | * configure via vmd via the device list in vm create params? |
| 5895 | */ |
| 5896 | switch (vcpu->vc_exit.vei.vei_port) { |
| 5897 | case IO_ICU10x020 ... IO_ICU10x020 + 1: |
| 5898 | case 0x40 ... 0x43: |
| 5899 | case PCKBC_AUX0x61: |
| 5900 | case IO_RTC0x070 ... IO_RTC0x070 + 1: |
| 5901 | case IO_ICU20x0A0 ... IO_ICU20x0A0 + 1: |
| 5902 | case 0x3f8 ... 0x3ff: |
| 5903 | case ELCR00x4D0 ... ELCR10x4D1: |
| 5904 | case 0x500 ... 0x511: |
| 5905 | case 0x514: |
| 5906 | case 0x518: |
| 5907 | case 0xcf8: |
| 5908 | case 0xcfc ... 0xcff: |
| 5909 | case VMM_PCI_IO_BAR_BASE0x1000 ... VMM_PCI_IO_BAR_END0xFFFF: |
| 5910 | ret = EAGAIN35; |
| 5911 | break; |
| 5912 | default: |
| 5913 | /* Read from unsupported ports returns FFs */ |
| 5914 | if (vcpu->vc_exit.vei.vei_dir == VEI_DIR_IN) { |
| 5915 | if (vcpu->vc_exit.vei.vei_size == 4) |
| 5916 | vcpu->vc_gueststate.vg_rax |= 0xFFFFFFFF; |
| 5917 | else if (vcpu->vc_exit.vei.vei_size == 2) |
| 5918 | vcpu->vc_gueststate.vg_rax |= 0xFFFF; |
| 5919 | else if (vcpu->vc_exit.vei.vei_size == 1) |
| 5920 | vcpu->vc_gueststate.vg_rax |= 0xFF; |
| 5921 | } |
| 5922 | ret = 0; |
| 5923 | } |
| 5924 | |
| 5925 | return (ret); |
| 5926 | } |
| 5927 | |
| 5928 | /* |
| 5929 | * vmx_load_pdptes |
| 5930 | * |
| 5931 | * Update the PDPTEs in the VMCS with the values currently indicated by the |
| 5932 | * guest CR3. This is used for 32-bit PAE guests when enabling paging. |
| 5933 | * |
| 5934 | * Parameters |
| 5935 | * vcpu: The vcpu whose PDPTEs should be loaded |
| 5936 | * |
| 5937 | * Return values: |
| 5938 | * 0: if successful |
| 5939 | * EINVAL: if the PDPTEs could not be loaded |
| 5940 | * ENOMEM: memory allocation failure |
| 5941 | */ |
| 5942 | int |
| 5943 | vmx_load_pdptes(struct vcpu *vcpu) |
| 5944 | { |
| 5945 | uint64_t cr3, cr3_host_phys; |
| 5946 | vaddr_t cr3_host_virt; |
| 5947 | pd_entry_t *pdptes; |
| 5948 | int ret; |
| 5949 | |
| 5950 | if (vmread(VMCS_GUEST_IA32_CR30x6802, &cr3)) { |
| 5951 | printf("%s: can't read guest cr3\n", __func__); |
| 5952 | return (EINVAL22); |
| 5953 | } |
| 5954 | |
| 5955 | if (!pmap_extract(vcpu->vc_parent->vm_map->pmap, (vaddr_t)cr3, |
| 5956 | (paddr_t *)&cr3_host_phys)) { |
| 5957 | DPRINTF("%s: nonmapped guest CR3, setting PDPTEs to 0\n", |
| 5958 | __func__); |
| 5959 | if (vmwrite(VMCS_GUEST_PDPTE00x280A, 0)) { |
| 5960 | printf("%s: can't write guest PDPTE0\n", __func__); |
| 5961 | return (EINVAL22); |
| 5962 | } |
| 5963 | |
| 5964 | if (vmwrite(VMCS_GUEST_PDPTE10x280C, 0)) { |
| 5965 | printf("%s: can't write guest PDPTE1\n", __func__); |
| 5966 | return (EINVAL22); |
| 5967 | } |
| 5968 | |
| 5969 | if (vmwrite(VMCS_GUEST_PDPTE20x280E, 0)) { |
| 5970 | printf("%s: can't write guest PDPTE2\n", __func__); |
| 5971 | return (EINVAL22); |
| 5972 | } |
| 5973 | |
| 5974 | if (vmwrite(VMCS_GUEST_PDPTE30x2810, 0)) { |
| 5975 | printf("%s: can't write guest PDPTE3\n", __func__); |
| 5976 | return (EINVAL22); |
| 5977 | } |
| 5978 | return (0); |
| 5979 | } |
| 5980 | |
| 5981 | ret = 0; |
| 5982 | |
| 5983 | /* We may sleep during km_alloc(9), so reload VMCS. */ |
| 5984 | vcpu->vc_last_pcpu = curcpu()({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;}); |
| 5985 | cr3_host_virt = (vaddr_t)km_alloc(PAGE_SIZE(1 << 12), &kv_any, &kp_none, |
| 5986 | &kd_waitok); |
| 5987 | if (vcpu_reload_vmcs_vmx(vcpu)) { |
| 5988 | printf("%s: failed to reload vmcs\n", __func__); |
| 5989 | ret = EINVAL22; |
| 5990 | goto exit; |
| 5991 | } |
| 5992 | |
| 5993 | if (!cr3_host_virt) { |
| 5994 | printf("%s: can't allocate address for guest CR3 mapping\n", |
| 5995 | __func__); |
| 5996 | return (ENOMEM12); |
| 5997 | } |
| 5998 | |
| 5999 | pmap_kenter_pa(cr3_host_virt, cr3_host_phys, PROT_READ0x01); |
| 6000 | |
| 6001 | pdptes = (pd_entry_t *)cr3_host_virt; |
| 6002 | if (vmwrite(VMCS_GUEST_PDPTE00x280A, pdptes[0])) { |
| 6003 | printf("%s: can't write guest PDPTE0\n", __func__); |
| 6004 | ret = EINVAL22; |
| 6005 | goto exit; |
| 6006 | } |
| 6007 | |
| 6008 | if (vmwrite(VMCS_GUEST_PDPTE10x280C, pdptes[1])) { |
| 6009 | printf("%s: can't write guest PDPTE1\n", __func__); |
| 6010 | ret = EINVAL22; |
| 6011 | goto exit; |
| 6012 | } |
| 6013 | |
| 6014 | if (vmwrite(VMCS_GUEST_PDPTE20x280E, pdptes[2])) { |
| 6015 | printf("%s: can't write guest PDPTE2\n", __func__); |
| 6016 | ret = EINVAL22; |
| 6017 | goto exit; |
| 6018 | } |
| 6019 | |
| 6020 | if (vmwrite(VMCS_GUEST_PDPTE30x2810, pdptes[3])) { |
| 6021 | printf("%s: can't write guest PDPTE3\n", __func__); |
| 6022 | ret = EINVAL22; |
| 6023 | goto exit; |
| 6024 | } |
| 6025 | |
| 6026 | exit: |
| 6027 | pmap_kremove(cr3_host_virt, PAGE_SIZE(1 << 12)); |
| 6028 | |
| 6029 | /* km_free(9) might sleep, so we need to reload VMCS. */ |
| 6030 | vcpu->vc_last_pcpu = curcpu()({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;}); |
| 6031 | km_free((void *)cr3_host_virt, PAGE_SIZE(1 << 12), &kv_any, &kp_none); |
| 6032 | if (vcpu_reload_vmcs_vmx(vcpu)) { |
| 6033 | printf("%s: failed to reload vmcs after km_free\n", __func__); |
| 6034 | ret = EINVAL22; |
| 6035 | } |
| 6036 | |
| 6037 | return (ret); |
| 6038 | } |
| 6039 | |
| 6040 | /* |
| 6041 | * vmx_handle_cr0_write |
| 6042 | * |
| 6043 | * Write handler for CR0. This function ensures valid values are written into |
| 6044 | * CR0 for the cpu/vmm mode in use (cr0 must-be-0 and must-be-1 bits, etc). |
| 6045 | * |
| 6046 | * Parameters |
| 6047 | * vcpu: The vcpu taking the cr0 write exit |
| 6048 | * r: The guest's desired (incoming) cr0 value |
| 6049 | * |
| 6050 | * Return values: |
| 6051 | * 0: if successful |
| 6052 | * EINVAL: if an error occurred |
| 6053 | */ |
| 6054 | int |
| 6055 | vmx_handle_cr0_write(struct vcpu *vcpu, uint64_t r) |
| 6056 | { |
| 6057 | struct vmx_msr_store *msr_store; |
| 6058 | struct vmx_invvpid_descriptor vid; |
| 6059 | uint64_t ectls, oldcr0, cr4, mask; |
| 6060 | int ret; |
| 6061 | |
| 6062 | /* Check must-be-0 bits */ |
| 6063 | mask = vcpu->vc_vmx_cr0_fixed1; |
| 6064 | if (~r & mask) { |
| 6065 | /* Inject #GP, let the guest handle it */ |
| 6066 | DPRINTF("%s: guest set invalid bits in %%cr0. Zeros " |
| 6067 | "mask=0x%llx, data=0x%llx\n", __func__, |
| 6068 | vcpu->vc_vmx_cr0_fixed1, r); |
| 6069 | vmm_inject_gp(vcpu); |
| 6070 | return (0); |
| 6071 | } |
| 6072 | |
| 6073 | /* Check must-be-1 bits */ |
| 6074 | mask = vcpu->vc_vmx_cr0_fixed0; |
| 6075 | if ((r & mask) != mask) { |
| 6076 | /* Inject #GP, let the guest handle it */ |
| 6077 | DPRINTF("%s: guest set invalid bits in %%cr0. Ones " |
| 6078 | "mask=0x%llx, data=0x%llx\n", __func__, |
| 6079 | vcpu->vc_vmx_cr0_fixed0, r); |
| 6080 | vmm_inject_gp(vcpu); |
| 6081 | return (0); |
| 6082 | } |
| 6083 | |
| 6084 | if (r & 0xFFFFFFFF00000000ULL) { |
| 6085 | DPRINTF("%s: setting bits 63:32 of %%cr0 is invalid," |
| 6086 | " inject #GP, cr0=0x%llx\n", __func__, r); |
| 6087 | vmm_inject_gp(vcpu); |
| 6088 | return (0); |
| 6089 | } |
| 6090 | |
| 6 |