xref: /openbmc/qemu/target/i386/tcg/sysemu/svm_helper.c (revision e1723999)
1 /*
2  *  x86 SVM helpers (sysemu only)
3  *
4  *  Copyright (c) 2003 Fabrice Bellard
5  *
6  * This library is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU Lesser General Public
8  * License as published by the Free Software Foundation; either
9  * version 2.1 of the License, or (at your option) any later version.
10  *
11  * This library is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14  * Lesser General Public License for more details.
15  *
16  * You should have received a copy of the GNU Lesser General Public
17  * License along with this library; if not, see <http://www.gnu.org/licenses/>.
18  */
19 
20 #include "qemu/osdep.h"
21 #include "cpu.h"
22 #include "exec/helper-proto.h"
23 #include "exec/exec-all.h"
24 #include "exec/cpu_ldst.h"
25 #include "tcg/helper-tcg.h"
26 
27 /* Secure Virtual Machine helpers */
28 
29 static inline void svm_save_seg(CPUX86State *env, hwaddr addr,
30                                 const SegmentCache *sc)
31 {
32     CPUState *cs = env_cpu(env);
33 
34     x86_stw_phys(cs, addr + offsetof(struct vmcb_seg, selector),
35              sc->selector);
36     x86_stq_phys(cs, addr + offsetof(struct vmcb_seg, base),
37              sc->base);
38     x86_stl_phys(cs, addr + offsetof(struct vmcb_seg, limit),
39              sc->limit);
40     x86_stw_phys(cs, addr + offsetof(struct vmcb_seg, attrib),
41              ((sc->flags >> 8) & 0xff) | ((sc->flags >> 12) & 0x0f00));
42 }
43 
44 /*
45  * VMRUN and VMLOAD canonicalizes (i.e., sign-extend to bit 63) all base
46  * addresses in the segment registers that have been loaded.
47  */
48 static inline void svm_canonicalization(CPUX86State *env, target_ulong *seg_base)
49 {
50     uint16_t shift_amt = 64 - cpu_x86_virtual_addr_width(env);
51     *seg_base = ((((long) *seg_base) << shift_amt) >> shift_amt);
52 }
53 
54 static inline void svm_load_seg(CPUX86State *env, hwaddr addr,
55                                 SegmentCache *sc)
56 {
57     CPUState *cs = env_cpu(env);
58     unsigned int flags;
59 
60     sc->selector = x86_lduw_phys(cs,
61                              addr + offsetof(struct vmcb_seg, selector));
62     sc->base = x86_ldq_phys(cs, addr + offsetof(struct vmcb_seg, base));
63     sc->limit = x86_ldl_phys(cs, addr + offsetof(struct vmcb_seg, limit));
64     flags = x86_lduw_phys(cs, addr + offsetof(struct vmcb_seg, attrib));
65     sc->flags = ((flags & 0xff) << 8) | ((flags & 0x0f00) << 12);
66     svm_canonicalization(env, &sc->base);
67 }
68 
69 static inline void svm_load_seg_cache(CPUX86State *env, hwaddr addr,
70                                       int seg_reg)
71 {
72     SegmentCache sc1, *sc = &sc1;
73 
74     svm_load_seg(env, addr, sc);
75     cpu_x86_load_seg_cache(env, seg_reg, sc->selector,
76                            sc->base, sc->limit, sc->flags);
77 }
78 
79 static inline bool is_efer_invalid_state (CPUX86State *env)
80 {
81     if (!(env->efer & MSR_EFER_SVME)) {
82         return true;
83     }
84 
85     if (env->efer & MSR_EFER_RESERVED) {
86         return true;
87     }
88 
89     if ((env->efer & (MSR_EFER_LMA | MSR_EFER_LME)) &&
90             !(env->features[FEAT_8000_0001_EDX] & CPUID_EXT2_LM)) {
91         return true;
92     }
93 
94     if ((env->efer & MSR_EFER_LME) && (env->cr[0] & CR0_PG_MASK)
95                                 && !(env->cr[4] & CR4_PAE_MASK)) {
96         return true;
97     }
98 
99     if ((env->efer & MSR_EFER_LME) && (env->cr[0] & CR0_PG_MASK)
100                                 && !(env->cr[0] & CR0_PE_MASK)) {
101         return true;
102     }
103 
104     if ((env->efer & MSR_EFER_LME) && (env->cr[0] & CR0_PG_MASK)
105                                 && (env->cr[4] & CR4_PAE_MASK)
106                                 && (env->segs[R_CS].flags & DESC_L_MASK)
107                                 && (env->segs[R_CS].flags & DESC_B_MASK)) {
108         return true;
109     }
110 
111     return false;
112 }
113 
114 static inline bool virtual_gif_enabled(CPUX86State *env)
115 {
116     if (likely(env->hflags & HF_GUEST_MASK)) {
117         return (env->features[FEAT_SVM] & CPUID_SVM_VGIF)
118                     && (env->int_ctl & V_GIF_ENABLED_MASK);
119     }
120     return false;
121 }
122 
123 static inline bool virtual_vm_load_save_enabled(CPUX86State *env, uint32_t exit_code, uintptr_t retaddr)
124 {
125     uint64_t lbr_ctl;
126 
127     if (likely(env->hflags & HF_GUEST_MASK)) {
128         if (likely(!(env->hflags2 & HF2_NPT_MASK)) || !(env->efer & MSR_EFER_LMA)) {
129             cpu_vmexit(env, exit_code, 0, retaddr);
130         }
131 
132         lbr_ctl = x86_ldl_phys(env_cpu(env), env->vm_vmcb + offsetof(struct vmcb,
133                                                   control.lbr_ctl));
134         return (env->features[FEAT_SVM] & CPUID_SVM_V_VMSAVE_VMLOAD)
135                 && (lbr_ctl & V_VMLOAD_VMSAVE_ENABLED_MASK);
136 
137     }
138 
139     return false;
140 }
141 
142 static inline bool virtual_gif_set(CPUX86State *env)
143 {
144     return !virtual_gif_enabled(env) || (env->int_ctl & V_GIF_MASK);
145 }
146 
147 void helper_vmrun(CPUX86State *env, int aflag, int next_eip_addend)
148 {
149     CPUState *cs = env_cpu(env);
150     X86CPU *cpu = env_archcpu(env);
151     target_ulong addr;
152     uint64_t nested_ctl;
153     uint32_t event_inj;
154     uint32_t asid;
155     uint64_t new_cr0;
156     uint64_t new_cr3;
157     uint64_t new_cr4;
158 
159     cpu_svm_check_intercept_param(env, SVM_EXIT_VMRUN, 0, GETPC());
160 
161     if (aflag == 2) {
162         addr = env->regs[R_EAX];
163     } else {
164         addr = (uint32_t)env->regs[R_EAX];
165     }
166 
167     qemu_log_mask(CPU_LOG_TB_IN_ASM, "vmrun! " TARGET_FMT_lx "\n", addr);
168 
169     env->vm_vmcb = addr;
170 
171     /* save the current CPU state in the hsave page */
172     x86_stq_phys(cs, env->vm_hsave + offsetof(struct vmcb, save.gdtr.base),
173              env->gdt.base);
174     x86_stl_phys(cs, env->vm_hsave + offsetof(struct vmcb, save.gdtr.limit),
175              env->gdt.limit);
176 
177     x86_stq_phys(cs, env->vm_hsave + offsetof(struct vmcb, save.idtr.base),
178              env->idt.base);
179     x86_stl_phys(cs, env->vm_hsave + offsetof(struct vmcb, save.idtr.limit),
180              env->idt.limit);
181 
182     x86_stq_phys(cs,
183              env->vm_hsave + offsetof(struct vmcb, save.cr0), env->cr[0]);
184     x86_stq_phys(cs,
185              env->vm_hsave + offsetof(struct vmcb, save.cr2), env->cr[2]);
186     x86_stq_phys(cs,
187              env->vm_hsave + offsetof(struct vmcb, save.cr3), env->cr[3]);
188     x86_stq_phys(cs,
189              env->vm_hsave + offsetof(struct vmcb, save.cr4), env->cr[4]);
190     x86_stq_phys(cs,
191              env->vm_hsave + offsetof(struct vmcb, save.dr6), env->dr[6]);
192     x86_stq_phys(cs,
193              env->vm_hsave + offsetof(struct vmcb, save.dr7), env->dr[7]);
194 
195     x86_stq_phys(cs,
196              env->vm_hsave + offsetof(struct vmcb, save.efer), env->efer);
197     x86_stq_phys(cs,
198              env->vm_hsave + offsetof(struct vmcb, save.rflags),
199              cpu_compute_eflags(env));
200 
201     svm_save_seg(env, env->vm_hsave + offsetof(struct vmcb, save.es),
202                  &env->segs[R_ES]);
203     svm_save_seg(env, env->vm_hsave + offsetof(struct vmcb, save.cs),
204                  &env->segs[R_CS]);
205     svm_save_seg(env, env->vm_hsave + offsetof(struct vmcb, save.ss),
206                  &env->segs[R_SS]);
207     svm_save_seg(env, env->vm_hsave + offsetof(struct vmcb, save.ds),
208                  &env->segs[R_DS]);
209 
210     x86_stq_phys(cs, env->vm_hsave + offsetof(struct vmcb, save.rip),
211              env->eip + next_eip_addend);
212     x86_stq_phys(cs,
213              env->vm_hsave + offsetof(struct vmcb, save.rsp), env->regs[R_ESP]);
214     x86_stq_phys(cs,
215              env->vm_hsave + offsetof(struct vmcb, save.rax), env->regs[R_EAX]);
216 
217     /* load the interception bitmaps so we do not need to access the
218        vmcb in svm mode */
219     env->intercept = x86_ldq_phys(cs, env->vm_vmcb + offsetof(struct vmcb,
220                                                       control.intercept));
221     env->intercept_cr_read = x86_lduw_phys(cs, env->vm_vmcb +
222                                        offsetof(struct vmcb,
223                                                 control.intercept_cr_read));
224     env->intercept_cr_write = x86_lduw_phys(cs, env->vm_vmcb +
225                                         offsetof(struct vmcb,
226                                                  control.intercept_cr_write));
227     env->intercept_dr_read = x86_lduw_phys(cs, env->vm_vmcb +
228                                        offsetof(struct vmcb,
229                                                 control.intercept_dr_read));
230     env->intercept_dr_write = x86_lduw_phys(cs, env->vm_vmcb +
231                                         offsetof(struct vmcb,
232                                                  control.intercept_dr_write));
233     env->intercept_exceptions = x86_ldl_phys(cs, env->vm_vmcb +
234                                          offsetof(struct vmcb,
235                                                   control.intercept_exceptions
236                                                   ));
237 
238     nested_ctl = x86_ldq_phys(cs, env->vm_vmcb + offsetof(struct vmcb,
239                                                           control.nested_ctl));
240     asid = x86_ldq_phys(cs, env->vm_vmcb + offsetof(struct vmcb,
241                                                           control.asid));
242 
243     uint64_t msrpm_base_pa = x86_ldq_phys(cs, env->vm_vmcb +
244                                     offsetof(struct vmcb,
245                                             control.msrpm_base_pa));
246     uint64_t iopm_base_pa = x86_ldq_phys(cs, env->vm_vmcb +
247                                  offsetof(struct vmcb, control.iopm_base_pa));
248 
249     if ((msrpm_base_pa & ~0xfff) >= (1ull << cpu->phys_bits) - SVM_MSRPM_SIZE) {
250         cpu_vmexit(env, SVM_EXIT_ERR, 0, GETPC());
251     }
252 
253     if ((iopm_base_pa & ~0xfff) >= (1ull << cpu->phys_bits) - SVM_IOPM_SIZE) {
254         cpu_vmexit(env, SVM_EXIT_ERR, 0, GETPC());
255     }
256 
257     env->nested_pg_mode = 0;
258 
259     if (!cpu_svm_has_intercept(env, SVM_EXIT_VMRUN)) {
260         cpu_vmexit(env, SVM_EXIT_ERR, 0, GETPC());
261     }
262     if (asid == 0) {
263         cpu_vmexit(env, SVM_EXIT_ERR, 0, GETPC());
264     }
265 
266     if (nested_ctl & SVM_NPT_ENABLED) {
267         env->nested_cr3 = x86_ldq_phys(cs,
268                                 env->vm_vmcb + offsetof(struct vmcb,
269                                                         control.nested_cr3));
270         env->hflags2 |= HF2_NPT_MASK;
271 
272         env->nested_pg_mode = get_pg_mode(env) & PG_MODE_SVM_MASK;
273     }
274 
275     /* enable intercepts */
276     env->hflags |= HF_GUEST_MASK;
277 
278     env->tsc_offset = x86_ldq_phys(cs, env->vm_vmcb +
279                                offsetof(struct vmcb, control.tsc_offset));
280 
281     new_cr0 = x86_ldq_phys(cs, env->vm_vmcb + offsetof(struct vmcb, save.cr0));
282     if (new_cr0 & SVM_CR0_RESERVED_MASK) {
283         cpu_vmexit(env, SVM_EXIT_ERR, 0, GETPC());
284     }
285     if ((new_cr0 & CR0_NW_MASK) && !(new_cr0 & CR0_CD_MASK)) {
286         cpu_vmexit(env, SVM_EXIT_ERR, 0, GETPC());
287     }
288     new_cr3 = x86_ldq_phys(cs, env->vm_vmcb + offsetof(struct vmcb, save.cr3));
289     if ((env->efer & MSR_EFER_LMA) &&
290             (new_cr3 & ((~0ULL) << cpu->phys_bits))) {
291         cpu_vmexit(env, SVM_EXIT_ERR, 0, GETPC());
292     }
293     new_cr4 = x86_ldq_phys(cs, env->vm_vmcb + offsetof(struct vmcb, save.cr4));
294     if (new_cr4 & cr4_reserved_bits(env)) {
295         cpu_vmexit(env, SVM_EXIT_ERR, 0, GETPC());
296     }
297     /* clear exit_info_2 so we behave like the real hardware */
298     x86_stq_phys(cs,
299              env->vm_vmcb + offsetof(struct vmcb, control.exit_info_2), 0);
300 
301     cpu_x86_update_cr0(env, new_cr0);
302     cpu_x86_update_cr4(env, new_cr4);
303     cpu_x86_update_cr3(env, new_cr3);
304     env->cr[2] = x86_ldq_phys(cs,
305                           env->vm_vmcb + offsetof(struct vmcb, save.cr2));
306     env->int_ctl = x86_ldl_phys(cs,
307                        env->vm_vmcb + offsetof(struct vmcb, control.int_ctl));
308     env->hflags2 &= ~(HF2_HIF_MASK | HF2_VINTR_MASK);
309     if (env->int_ctl & V_INTR_MASKING_MASK) {
310         env->hflags2 |= HF2_VINTR_MASK;
311         if (env->eflags & IF_MASK) {
312             env->hflags2 |= HF2_HIF_MASK;
313         }
314     }
315 
316     cpu_load_efer(env,
317                   x86_ldq_phys(cs,
318                            env->vm_vmcb + offsetof(struct vmcb, save.efer)));
319     env->eflags = 0;
320     cpu_load_eflags(env, x86_ldq_phys(cs,
321                                   env->vm_vmcb + offsetof(struct vmcb,
322                                                           save.rflags)),
323                     ~(CC_O | CC_S | CC_Z | CC_A | CC_P | CC_C | DF_MASK));
324 
325     svm_load_seg_cache(env, env->vm_vmcb + offsetof(struct vmcb, save.es),
326                        R_ES);
327     svm_load_seg_cache(env, env->vm_vmcb + offsetof(struct vmcb, save.cs),
328                        R_CS);
329     svm_load_seg_cache(env, env->vm_vmcb + offsetof(struct vmcb, save.ss),
330                        R_SS);
331     svm_load_seg_cache(env, env->vm_vmcb + offsetof(struct vmcb, save.ds),
332                        R_DS);
333     svm_load_seg(env, env->vm_vmcb + offsetof(struct vmcb, save.idtr),
334                        &env->idt);
335     svm_load_seg(env, env->vm_vmcb + offsetof(struct vmcb, save.gdtr),
336                        &env->gdt);
337 
338     env->eip = x86_ldq_phys(cs,
339                         env->vm_vmcb + offsetof(struct vmcb, save.rip));
340 
341     env->regs[R_ESP] = x86_ldq_phys(cs,
342                                 env->vm_vmcb + offsetof(struct vmcb, save.rsp));
343     env->regs[R_EAX] = x86_ldq_phys(cs,
344                                 env->vm_vmcb + offsetof(struct vmcb, save.rax));
345     env->dr[7] = x86_ldq_phys(cs,
346                           env->vm_vmcb + offsetof(struct vmcb, save.dr7));
347     env->dr[6] = x86_ldq_phys(cs,
348                           env->vm_vmcb + offsetof(struct vmcb, save.dr6));
349 
350 #ifdef TARGET_X86_64
351     if (env->dr[6] & DR_RESERVED_MASK) {
352         cpu_vmexit(env, SVM_EXIT_ERR, 0, GETPC());
353     }
354     if (env->dr[7] & DR_RESERVED_MASK) {
355         cpu_vmexit(env, SVM_EXIT_ERR, 0, GETPC());
356     }
357 #endif
358 
359     if (is_efer_invalid_state(env)) {
360         cpu_vmexit(env, SVM_EXIT_ERR, 0, GETPC());
361     }
362 
363     switch (x86_ldub_phys(cs,
364                       env->vm_vmcb + offsetof(struct vmcb, control.tlb_ctl))) {
365     case TLB_CONTROL_DO_NOTHING:
366         break;
367     case TLB_CONTROL_FLUSH_ALL_ASID:
368         /* FIXME: this is not 100% correct but should work for now */
369         tlb_flush(cs);
370         break;
371     }
372 
373     env->hflags2 |= HF2_GIF_MASK;
374 
375     if (ctl_has_irq(env)) {
376         CPUState *cs = env_cpu(env);
377 
378         cs->interrupt_request |= CPU_INTERRUPT_VIRQ;
379     }
380 
381     if (virtual_gif_set(env)) {
382         env->hflags2 |= HF2_VGIF_MASK;
383     }
384 
385     /* maybe we need to inject an event */
386     event_inj = x86_ldl_phys(cs, env->vm_vmcb + offsetof(struct vmcb,
387                                                  control.event_inj));
388     if (event_inj & SVM_EVTINJ_VALID) {
389         uint8_t vector = event_inj & SVM_EVTINJ_VEC_MASK;
390         uint16_t valid_err = event_inj & SVM_EVTINJ_VALID_ERR;
391         uint32_t event_inj_err = x86_ldl_phys(cs, env->vm_vmcb +
392                                           offsetof(struct vmcb,
393                                                    control.event_inj_err));
394 
395         qemu_log_mask(CPU_LOG_TB_IN_ASM, "Injecting(%#hx): ", valid_err);
396         /* FIXME: need to implement valid_err */
397         switch (event_inj & SVM_EVTINJ_TYPE_MASK) {
398         case SVM_EVTINJ_TYPE_INTR:
399             cs->exception_index = vector;
400             env->error_code = event_inj_err;
401             env->exception_is_int = 0;
402             env->exception_next_eip = -1;
403             qemu_log_mask(CPU_LOG_TB_IN_ASM, "INTR");
404             /* XXX: is it always correct? */
405             do_interrupt_x86_hardirq(env, vector, 1);
406             break;
407         case SVM_EVTINJ_TYPE_NMI:
408             cs->exception_index = EXCP02_NMI;
409             env->error_code = event_inj_err;
410             env->exception_is_int = 0;
411             env->exception_next_eip = env->eip;
412             qemu_log_mask(CPU_LOG_TB_IN_ASM, "NMI");
413             cpu_loop_exit(cs);
414             break;
415         case SVM_EVTINJ_TYPE_EXEPT:
416             if (vector == EXCP02_NMI || vector >= 31)  {
417                 cpu_vmexit(env, SVM_EXIT_ERR, 0, GETPC());
418             }
419             cs->exception_index = vector;
420             env->error_code = event_inj_err;
421             env->exception_is_int = 0;
422             env->exception_next_eip = -1;
423             qemu_log_mask(CPU_LOG_TB_IN_ASM, "EXEPT");
424             cpu_loop_exit(cs);
425             break;
426         case SVM_EVTINJ_TYPE_SOFT:
427             cs->exception_index = vector;
428             env->error_code = event_inj_err;
429             env->exception_is_int = 1;
430             env->exception_next_eip = env->eip;
431             qemu_log_mask(CPU_LOG_TB_IN_ASM, "SOFT");
432             cpu_loop_exit(cs);
433             break;
434         default:
435             cpu_vmexit(env, SVM_EXIT_ERR, 0, GETPC());
436             break;
437         }
438         qemu_log_mask(CPU_LOG_TB_IN_ASM, " %#x %#x\n", cs->exception_index,
439                       env->error_code);
440     }
441 }
442 
443 void helper_vmmcall(CPUX86State *env)
444 {
445     cpu_svm_check_intercept_param(env, SVM_EXIT_VMMCALL, 0, GETPC());
446     raise_exception(env, EXCP06_ILLOP);
447 }
448 
449 void helper_vmload(CPUX86State *env, int aflag)
450 {
451     CPUState *cs = env_cpu(env);
452     target_ulong addr;
453     int prot;
454 
455     cpu_svm_check_intercept_param(env, SVM_EXIT_VMLOAD, 0, GETPC());
456 
457     if (aflag == 2) {
458         addr = env->regs[R_EAX];
459     } else {
460         addr = (uint32_t)env->regs[R_EAX];
461     }
462 
463     if (virtual_vm_load_save_enabled(env, SVM_EXIT_VMLOAD, GETPC())) {
464         addr = get_hphys(cs, addr, MMU_DATA_LOAD, &prot);
465     }
466 
467     qemu_log_mask(CPU_LOG_TB_IN_ASM, "vmload! " TARGET_FMT_lx
468                   "\nFS: %016" PRIx64 " | " TARGET_FMT_lx "\n",
469                   addr, x86_ldq_phys(cs, addr + offsetof(struct vmcb,
470                                                           save.fs.base)),
471                   env->segs[R_FS].base);
472 
473     svm_load_seg_cache(env, addr + offsetof(struct vmcb, save.fs), R_FS);
474     svm_load_seg_cache(env, addr + offsetof(struct vmcb, save.gs), R_GS);
475     svm_load_seg(env, addr + offsetof(struct vmcb, save.tr), &env->tr);
476     svm_load_seg(env, addr + offsetof(struct vmcb, save.ldtr), &env->ldt);
477 
478 #ifdef TARGET_X86_64
479     env->kernelgsbase = x86_ldq_phys(cs, addr + offsetof(struct vmcb,
480                                                  save.kernel_gs_base));
481     env->lstar = x86_ldq_phys(cs, addr + offsetof(struct vmcb, save.lstar));
482     env->cstar = x86_ldq_phys(cs, addr + offsetof(struct vmcb, save.cstar));
483     env->fmask = x86_ldq_phys(cs, addr + offsetof(struct vmcb, save.sfmask));
484     svm_canonicalization(env, &env->kernelgsbase);
485 #endif
486     env->star = x86_ldq_phys(cs, addr + offsetof(struct vmcb, save.star));
487     env->sysenter_cs = x86_ldq_phys(cs,
488                                 addr + offsetof(struct vmcb, save.sysenter_cs));
489     env->sysenter_esp = x86_ldq_phys(cs, addr + offsetof(struct vmcb,
490                                                  save.sysenter_esp));
491     env->sysenter_eip = x86_ldq_phys(cs, addr + offsetof(struct vmcb,
492                                                  save.sysenter_eip));
493 
494 }
495 
496 void helper_vmsave(CPUX86State *env, int aflag)
497 {
498     CPUState *cs = env_cpu(env);
499     target_ulong addr;
500     int prot;
501 
502     cpu_svm_check_intercept_param(env, SVM_EXIT_VMSAVE, 0, GETPC());
503 
504     if (aflag == 2) {
505         addr = env->regs[R_EAX];
506     } else {
507         addr = (uint32_t)env->regs[R_EAX];
508     }
509 
510     if (virtual_vm_load_save_enabled(env, SVM_EXIT_VMSAVE, GETPC())) {
511         addr = get_hphys(cs, addr, MMU_DATA_STORE, &prot);
512     }
513 
514     qemu_log_mask(CPU_LOG_TB_IN_ASM, "vmsave! " TARGET_FMT_lx
515                   "\nFS: %016" PRIx64 " | " TARGET_FMT_lx "\n",
516                   addr, x86_ldq_phys(cs,
517                                  addr + offsetof(struct vmcb, save.fs.base)),
518                   env->segs[R_FS].base);
519 
520     svm_save_seg(env, addr + offsetof(struct vmcb, save.fs),
521                  &env->segs[R_FS]);
522     svm_save_seg(env, addr + offsetof(struct vmcb, save.gs),
523                  &env->segs[R_GS]);
524     svm_save_seg(env, addr + offsetof(struct vmcb, save.tr),
525                  &env->tr);
526     svm_save_seg(env, addr + offsetof(struct vmcb, save.ldtr),
527                  &env->ldt);
528 
529 #ifdef TARGET_X86_64
530     x86_stq_phys(cs, addr + offsetof(struct vmcb, save.kernel_gs_base),
531              env->kernelgsbase);
532     x86_stq_phys(cs, addr + offsetof(struct vmcb, save.lstar), env->lstar);
533     x86_stq_phys(cs, addr + offsetof(struct vmcb, save.cstar), env->cstar);
534     x86_stq_phys(cs, addr + offsetof(struct vmcb, save.sfmask), env->fmask);
535 #endif
536     x86_stq_phys(cs, addr + offsetof(struct vmcb, save.star), env->star);
537     x86_stq_phys(cs,
538              addr + offsetof(struct vmcb, save.sysenter_cs), env->sysenter_cs);
539     x86_stq_phys(cs, addr + offsetof(struct vmcb, save.sysenter_esp),
540              env->sysenter_esp);
541     x86_stq_phys(cs, addr + offsetof(struct vmcb, save.sysenter_eip),
542              env->sysenter_eip);
543 }
544 
545 void helper_stgi(CPUX86State *env)
546 {
547     cpu_svm_check_intercept_param(env, SVM_EXIT_STGI, 0, GETPC());
548 
549     if (virtual_gif_enabled(env)) {
550         env->int_ctl |= V_GIF_MASK;
551         env->hflags2 |= HF2_VGIF_MASK;
552     } else {
553         env->hflags2 |= HF2_GIF_MASK;
554     }
555 }
556 
557 void helper_clgi(CPUX86State *env)
558 {
559     cpu_svm_check_intercept_param(env, SVM_EXIT_CLGI, 0, GETPC());
560 
561     if (virtual_gif_enabled(env)) {
562         env->int_ctl &= ~V_GIF_MASK;
563         env->hflags2 &= ~HF2_VGIF_MASK;
564     } else {
565         env->hflags2 &= ~HF2_GIF_MASK;
566     }
567 }
568 
569 bool cpu_svm_has_intercept(CPUX86State *env, uint32_t type)
570 {
571     switch (type) {
572     case SVM_EXIT_READ_CR0 ... SVM_EXIT_READ_CR0 + 8:
573         if (env->intercept_cr_read & (1 << (type - SVM_EXIT_READ_CR0))) {
574             return true;
575         }
576         break;
577     case SVM_EXIT_WRITE_CR0 ... SVM_EXIT_WRITE_CR0 + 8:
578         if (env->intercept_cr_write & (1 << (type - SVM_EXIT_WRITE_CR0))) {
579             return true;
580         }
581         break;
582     case SVM_EXIT_READ_DR0 ... SVM_EXIT_READ_DR0 + 7:
583         if (env->intercept_dr_read & (1 << (type - SVM_EXIT_READ_DR0))) {
584             return true;
585         }
586         break;
587     case SVM_EXIT_WRITE_DR0 ... SVM_EXIT_WRITE_DR0 + 7:
588         if (env->intercept_dr_write & (1 << (type - SVM_EXIT_WRITE_DR0))) {
589             return true;
590         }
591         break;
592     case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 31:
593         if (env->intercept_exceptions & (1 << (type - SVM_EXIT_EXCP_BASE))) {
594             return true;
595         }
596         break;
597     default:
598         if (env->intercept & (1ULL << (type - SVM_EXIT_INTR))) {
599             return true;
600         }
601         break;
602     }
603     return false;
604 }
605 
606 void cpu_svm_check_intercept_param(CPUX86State *env, uint32_t type,
607                                    uint64_t param, uintptr_t retaddr)
608 {
609     CPUState *cs = env_cpu(env);
610 
611     if (likely(!(env->hflags & HF_GUEST_MASK))) {
612         return;
613     }
614 
615     if (!cpu_svm_has_intercept(env, type)) {
616         return;
617     }
618 
619     if (type == SVM_EXIT_MSR) {
620         /* FIXME: this should be read in at vmrun (faster this way?) */
621         uint64_t addr = x86_ldq_phys(cs, env->vm_vmcb +
622                                     offsetof(struct vmcb,
623                                             control.msrpm_base_pa));
624         uint32_t t0, t1;
625 
626         switch ((uint32_t)env->regs[R_ECX]) {
627         case 0 ... 0x1fff:
628             t0 = (env->regs[R_ECX] * 2) % 8;
629             t1 = (env->regs[R_ECX] * 2) / 8;
630             break;
631         case 0xc0000000 ... 0xc0001fff:
632             t0 = (8192 + env->regs[R_ECX] - 0xc0000000) * 2;
633             t1 = (t0 / 8);
634             t0 %= 8;
635             break;
636         case 0xc0010000 ... 0xc0011fff:
637             t0 = (16384 + env->regs[R_ECX] - 0xc0010000) * 2;
638             t1 = (t0 / 8);
639             t0 %= 8;
640             break;
641         default:
642             cpu_vmexit(env, type, param, retaddr);
643             t0 = 0;
644             t1 = 0;
645             break;
646         }
647         if (x86_ldub_phys(cs, addr + t1) & ((1 << param) << t0)) {
648             cpu_vmexit(env, type, param, retaddr);
649         }
650         return;
651     }
652 
653     cpu_vmexit(env, type, param, retaddr);
654 }
655 
656 void helper_svm_check_intercept(CPUX86State *env, uint32_t type)
657 {
658     cpu_svm_check_intercept_param(env, type, 0, GETPC());
659 }
660 
661 void helper_svm_check_io(CPUX86State *env, uint32_t port, uint32_t param,
662                          uint32_t next_eip_addend)
663 {
664     CPUState *cs = env_cpu(env);
665 
666     if (env->intercept & (1ULL << (SVM_EXIT_IOIO - SVM_EXIT_INTR))) {
667         /* FIXME: this should be read in at vmrun (faster this way?) */
668         uint64_t addr = x86_ldq_phys(cs, env->vm_vmcb +
669                                  offsetof(struct vmcb, control.iopm_base_pa));
670         uint16_t mask = (1 << ((param >> 4) & 7)) - 1;
671 
672         if (x86_lduw_phys(cs, addr + port / 8) & (mask << (port & 7))) {
673             /* next env->eip */
674             x86_stq_phys(cs,
675                      env->vm_vmcb + offsetof(struct vmcb, control.exit_info_2),
676                      env->eip + next_eip_addend);
677             cpu_vmexit(env, SVM_EXIT_IOIO, param | (port << 16), GETPC());
678         }
679     }
680 }
681 
682 void cpu_vmexit(CPUX86State *env, uint32_t exit_code, uint64_t exit_info_1,
683                 uintptr_t retaddr)
684 {
685     CPUState *cs = env_cpu(env);
686 
687     cpu_restore_state(cs, retaddr, true);
688 
689     qemu_log_mask(CPU_LOG_TB_IN_ASM, "vmexit(%08x, %016" PRIx64 ", %016"
690                   PRIx64 ", " TARGET_FMT_lx ")!\n",
691                   exit_code, exit_info_1,
692                   x86_ldq_phys(cs, env->vm_vmcb + offsetof(struct vmcb,
693                                                    control.exit_info_2)),
694                   env->eip);
695 
696     cs->exception_index = EXCP_VMEXIT;
697     x86_stq_phys(cs, env->vm_vmcb + offsetof(struct vmcb, control.exit_code),
698              exit_code);
699 
700     x86_stq_phys(cs, env->vm_vmcb + offsetof(struct vmcb,
701                                              control.exit_info_1), exit_info_1),
702 
703     /* remove any pending exception */
704     env->old_exception = -1;
705     cpu_loop_exit(cs);
706 }
707 
708 void do_vmexit(CPUX86State *env)
709 {
710     CPUState *cs = env_cpu(env);
711 
712     if (env->hflags & HF_INHIBIT_IRQ_MASK) {
713         x86_stl_phys(cs,
714                  env->vm_vmcb + offsetof(struct vmcb, control.int_state),
715                  SVM_INTERRUPT_SHADOW_MASK);
716         env->hflags &= ~HF_INHIBIT_IRQ_MASK;
717     } else {
718         x86_stl_phys(cs,
719                  env->vm_vmcb + offsetof(struct vmcb, control.int_state), 0);
720     }
721     env->hflags2 &= ~HF2_NPT_MASK;
722 
723     /* Save the VM state in the vmcb */
724     svm_save_seg(env, env->vm_vmcb + offsetof(struct vmcb, save.es),
725                  &env->segs[R_ES]);
726     svm_save_seg(env, env->vm_vmcb + offsetof(struct vmcb, save.cs),
727                  &env->segs[R_CS]);
728     svm_save_seg(env, env->vm_vmcb + offsetof(struct vmcb, save.ss),
729                  &env->segs[R_SS]);
730     svm_save_seg(env, env->vm_vmcb + offsetof(struct vmcb, save.ds),
731                  &env->segs[R_DS]);
732 
733     x86_stq_phys(cs, env->vm_vmcb + offsetof(struct vmcb, save.gdtr.base),
734              env->gdt.base);
735     x86_stl_phys(cs, env->vm_vmcb + offsetof(struct vmcb, save.gdtr.limit),
736              env->gdt.limit);
737 
738     x86_stq_phys(cs, env->vm_vmcb + offsetof(struct vmcb, save.idtr.base),
739              env->idt.base);
740     x86_stl_phys(cs, env->vm_vmcb + offsetof(struct vmcb, save.idtr.limit),
741              env->idt.limit);
742 
743     x86_stq_phys(cs,
744              env->vm_vmcb + offsetof(struct vmcb, save.efer), env->efer);
745     x86_stq_phys(cs,
746              env->vm_vmcb + offsetof(struct vmcb, save.cr0), env->cr[0]);
747     x86_stq_phys(cs,
748              env->vm_vmcb + offsetof(struct vmcb, save.cr2), env->cr[2]);
749     x86_stq_phys(cs,
750              env->vm_vmcb + offsetof(struct vmcb, save.cr3), env->cr[3]);
751     x86_stq_phys(cs,
752              env->vm_vmcb + offsetof(struct vmcb, save.cr4), env->cr[4]);
753     x86_stl_phys(cs,
754              env->vm_vmcb + offsetof(struct vmcb, control.int_ctl), env->int_ctl);
755 
756     x86_stq_phys(cs, env->vm_vmcb + offsetof(struct vmcb, save.rflags),
757              cpu_compute_eflags(env));
758     x86_stq_phys(cs, env->vm_vmcb + offsetof(struct vmcb, save.rip),
759              env->eip);
760     x86_stq_phys(cs,
761              env->vm_vmcb + offsetof(struct vmcb, save.rsp), env->regs[R_ESP]);
762     x86_stq_phys(cs,
763              env->vm_vmcb + offsetof(struct vmcb, save.rax), env->regs[R_EAX]);
764     x86_stq_phys(cs,
765              env->vm_vmcb + offsetof(struct vmcb, save.dr7), env->dr[7]);
766     x86_stq_phys(cs,
767              env->vm_vmcb + offsetof(struct vmcb, save.dr6), env->dr[6]);
768     x86_stb_phys(cs, env->vm_vmcb + offsetof(struct vmcb, save.cpl),
769              env->hflags & HF_CPL_MASK);
770 
771     /* Reload the host state from vm_hsave */
772     env->hflags2 &= ~(HF2_HIF_MASK | HF2_VINTR_MASK);
773     env->hflags &= ~HF_GUEST_MASK;
774     env->intercept = 0;
775     env->intercept_exceptions = 0;
776     cs->interrupt_request &= ~CPU_INTERRUPT_VIRQ;
777     env->int_ctl = 0;
778     env->tsc_offset = 0;
779 
780     env->gdt.base  = x86_ldq_phys(cs, env->vm_hsave + offsetof(struct vmcb,
781                                                        save.gdtr.base));
782     env->gdt.limit = x86_ldl_phys(cs, env->vm_hsave + offsetof(struct vmcb,
783                                                        save.gdtr.limit));
784 
785     env->idt.base  = x86_ldq_phys(cs, env->vm_hsave + offsetof(struct vmcb,
786                                                        save.idtr.base));
787     env->idt.limit = x86_ldl_phys(cs, env->vm_hsave + offsetof(struct vmcb,
788                                                        save.idtr.limit));
789 
790     cpu_x86_update_cr0(env, x86_ldq_phys(cs,
791                                      env->vm_hsave + offsetof(struct vmcb,
792                                                               save.cr0)) |
793                        CR0_PE_MASK);
794     cpu_x86_update_cr4(env, x86_ldq_phys(cs,
795                                      env->vm_hsave + offsetof(struct vmcb,
796                                                               save.cr4)));
797     cpu_x86_update_cr3(env, x86_ldq_phys(cs,
798                                      env->vm_hsave + offsetof(struct vmcb,
799                                                               save.cr3)));
800     /* we need to set the efer after the crs so the hidden flags get
801        set properly */
802     cpu_load_efer(env, x86_ldq_phys(cs, env->vm_hsave + offsetof(struct vmcb,
803                                                          save.efer)));
804     env->eflags = 0;
805     cpu_load_eflags(env, x86_ldq_phys(cs,
806                                   env->vm_hsave + offsetof(struct vmcb,
807                                                            save.rflags)),
808                     ~(CC_O | CC_S | CC_Z | CC_A | CC_P | CC_C | DF_MASK |
809                       VM_MASK));
810 
811     svm_load_seg_cache(env, env->vm_hsave + offsetof(struct vmcb, save.es),
812                        R_ES);
813     svm_load_seg_cache(env, env->vm_hsave + offsetof(struct vmcb, save.cs),
814                        R_CS);
815     svm_load_seg_cache(env, env->vm_hsave + offsetof(struct vmcb, save.ss),
816                        R_SS);
817     svm_load_seg_cache(env, env->vm_hsave + offsetof(struct vmcb, save.ds),
818                        R_DS);
819 
820     env->eip = x86_ldq_phys(cs,
821                         env->vm_hsave + offsetof(struct vmcb, save.rip));
822     env->regs[R_ESP] = x86_ldq_phys(cs, env->vm_hsave +
823                                 offsetof(struct vmcb, save.rsp));
824     env->regs[R_EAX] = x86_ldq_phys(cs, env->vm_hsave +
825                                 offsetof(struct vmcb, save.rax));
826 
827     env->dr[6] = x86_ldq_phys(cs,
828                           env->vm_hsave + offsetof(struct vmcb, save.dr6));
829     env->dr[7] = x86_ldq_phys(cs,
830                           env->vm_hsave + offsetof(struct vmcb, save.dr7));
831 
832     /* other setups */
833     x86_stl_phys(cs,
834              env->vm_vmcb + offsetof(struct vmcb, control.exit_int_info),
835              x86_ldl_phys(cs, env->vm_vmcb + offsetof(struct vmcb,
836                                               control.event_inj)));
837     x86_stl_phys(cs,
838              env->vm_vmcb + offsetof(struct vmcb, control.exit_int_info_err),
839              x86_ldl_phys(cs, env->vm_vmcb + offsetof(struct vmcb,
840                                               control.event_inj_err)));
841     x86_stl_phys(cs,
842              env->vm_vmcb + offsetof(struct vmcb, control.event_inj), 0);
843 
844     env->hflags2 &= ~HF2_GIF_MASK;
845     env->hflags2 &= ~HF2_VGIF_MASK;
846     /* FIXME: Resets the current ASID register to zero (host ASID). */
847 
848     /* Clears the V_IRQ and V_INTR_MASKING bits inside the processor. */
849 
850     /* Clears the TSC_OFFSET inside the processor. */
851 
852     /* If the host is in PAE mode, the processor reloads the host's PDPEs
853        from the page table indicated the host's CR3. If the PDPEs contain
854        illegal state, the processor causes a shutdown. */
855 
856     /* Disables all breakpoints in the host DR7 register. */
857 
858     /* Checks the reloaded host state for consistency. */
859 
860     /* If the host's rIP reloaded by #VMEXIT is outside the limit of the
861        host's code segment or non-canonical (in the case of long mode), a
862        #GP fault is delivered inside the host. */
863 }
864