Palacios Public Git Repository

To checkout Palacios execute

  git clone http://v3vee.org/palacios/palacios.web/palacios.git
This will give you the master branch. You probably want the devel branch or one of the release branches. To switch to the devel branch, simply execute
  cd palacios
  git checkout --track -b devel origin/devel
The other branches are similar.


VMX updates to fix multicore and unrestricted guest issues
[palacios.git] / palacios / src / palacios / svm.c
1 /* 
2  * This file is part of the Palacios Virtual Machine Monitor developed
3  * by the V3VEE Project with funding from the United States National 
4  * Science Foundation and the Department of Energy.  
5  *
6  * The V3VEE Project is a joint project between Northwestern University
7  * and the University of New Mexico.  You can find out more at 
8  * http://www.v3vee.org
9  *
10  * Copyright (c) 2008, Jack Lange <jarusl@cs.northwestern.edu> 
11  * Copyright (c) 2008, The V3VEE Project <http://www.v3vee.org> 
12  * All rights reserved.
13  *
14  * Author: Jack Lange <jarusl@cs.northwestern.edu>
15  *
16  * This is free software.  You are permitted to use,
17  * redistribute, and modify it as specified in the file "V3VEE_LICENSE".
18  */
19
20
21 #include <palacios/svm.h>
22 #include <palacios/vmm.h>
23
24 #include <palacios/vmcb.h>
25 #include <palacios/vmm_mem.h>
26 #include <palacios/vmm_paging.h>
27 #include <palacios/svm_handler.h>
28
29 #include <palacios/vmm_debug.h>
30 #include <palacios/vm_guest_mem.h>
31
32 #include <palacios/vmm_decoder.h>
33 #include <palacios/vmm_string.h>
34 #include <palacios/vmm_lowlevel.h>
35 #include <palacios/svm_msr.h>
36
37 #include <palacios/vmm_rbtree.h>
38 #include <palacios/vmm_barrier.h>
39
40 #ifdef V3_CONFIG_CHECKPOINT
41 #include <palacios/vmm_checkpoint.h>
42 #endif
43
44 #include <palacios/vmm_direct_paging.h>
45
46 #include <palacios/vmm_ctrl_regs.h>
47 #include <palacios/svm_io.h>
48
49 #include <palacios/vmm_sprintf.h>
50
51
52 #ifndef V3_CONFIG_DEBUG_SVM
53 #undef PrintDebug
54 #define PrintDebug(fmt, args...)
55 #endif
56
57
58 uint32_t v3_last_exit;
59
60 // This is a global pointer to the host's VMCB
61 static addr_t host_vmcbs[V3_CONFIG_MAX_CPUS] = { [0 ... V3_CONFIG_MAX_CPUS - 1] = 0};
62
63
64
65 extern void v3_stgi();
66 extern void v3_clgi();
67 //extern int v3_svm_launch(vmcb_t * vmcb, struct v3_gprs * vm_regs, uint64_t * fs, uint64_t * gs);
68 extern int v3_svm_launch(vmcb_t * vmcb, struct v3_gprs * vm_regs, vmcb_t * host_vmcb);
69
70
71 static vmcb_t * Allocate_VMCB() {
72     vmcb_t * vmcb_page = NULL;
73     addr_t vmcb_pa = (addr_t)V3_AllocPages(1);
74
75     if ((void *)vmcb_pa == NULL) {
76         PrintError("Error allocating VMCB\n");
77         return NULL;
78     }
79
80     vmcb_page = (vmcb_t *)V3_VAddr((void *)vmcb_pa);
81
82     memset(vmcb_page, 0, 4096);
83
84     return vmcb_page;
85 }
86
87
88 static int v3_svm_handle_efer_write(struct guest_info * core, uint_t msr, struct v3_msr src, void * priv_data)
89 {
90     int status;
91
92     // Call arch-independent handler
93     if ((status = v3_handle_efer_write(core, msr, src, priv_data)) != 0) {
94         return status;
95     }
96
97     // SVM-specific code
98     {
99         // Ensure that hardware visible EFER.SVME bit is set (SVM Enable)
100         struct efer_64 * hw_efer = (struct efer_64 *)&(core->ctrl_regs.efer);
101         hw_efer->svme = 1;
102     }
103
104     return 0;
105 }
106
107
108 static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info * core) {
109     vmcb_ctrl_t * ctrl_area = GET_VMCB_CTRL_AREA(vmcb);
110     vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA(vmcb);
111     uint_t i;
112
113
114     //
115     ctrl_area->svm_instrs.VMRUN = 1;
116     ctrl_area->svm_instrs.VMMCALL = 1;
117     ctrl_area->svm_instrs.VMLOAD = 1;
118     ctrl_area->svm_instrs.VMSAVE = 1;
119     ctrl_area->svm_instrs.STGI = 1;
120     ctrl_area->svm_instrs.CLGI = 1;
121     ctrl_area->svm_instrs.SKINIT = 1;
122     ctrl_area->svm_instrs.ICEBP = 1;
123     ctrl_area->svm_instrs.WBINVD = 1;
124     ctrl_area->svm_instrs.MONITOR = 1;
125     ctrl_area->svm_instrs.MWAIT_always = 1;
126     ctrl_area->svm_instrs.MWAIT_if_armed = 1;
127     ctrl_area->instrs.INVLPGA = 1;
128     ctrl_area->instrs.CPUID = 1;
129
130     ctrl_area->instrs.HLT = 1;
131
132     /* Set at VMM launch as needed */
133     ctrl_area->instrs.RDTSC = 0;
134     ctrl_area->svm_instrs.RDTSCP = 0;
135
136     // guest_state->cr0 = 0x00000001;    // PE 
137   
138     /*
139       ctrl_area->exceptions.de = 1;
140       ctrl_area->exceptions.df = 1;
141       
142       ctrl_area->exceptions.ts = 1;
143       ctrl_area->exceptions.ss = 1;
144       ctrl_area->exceptions.ac = 1;
145       ctrl_area->exceptions.mc = 1;
146       ctrl_area->exceptions.gp = 1;
147       ctrl_area->exceptions.ud = 1;
148       ctrl_area->exceptions.np = 1;
149       ctrl_area->exceptions.of = 1;
150       
151       ctrl_area->exceptions.nmi = 1;
152     */
153     
154
155     ctrl_area->instrs.NMI = 1;
156     ctrl_area->instrs.SMI = 0; // allow SMIs to run in guest
157     ctrl_area->instrs.INIT = 1;
158     //    ctrl_area->instrs.PAUSE = 1;
159     ctrl_area->instrs.shutdown_evts = 1;
160
161
162     /* DEBUG FOR RETURN CODE */
163     ctrl_area->exit_code = 1;
164
165
166     /* Setup Guest Machine state */
167
168     core->vm_regs.rsp = 0x00;
169     core->rip = 0xfff0;
170
171     core->vm_regs.rdx = 0x00000f00;
172
173
174     core->cpl = 0;
175
176     core->ctrl_regs.rflags = 0x00000002; // The reserved bit is always 1
177     core->ctrl_regs.cr0 = 0x60010010; // Set the WP flag so the memory hooks work in real-mode
178     core->ctrl_regs.efer |= EFER_MSR_svm_enable;
179
180
181
182
183
184     core->segments.cs.selector = 0xf000;
185     core->segments.cs.limit = 0xffff;
186     core->segments.cs.base = 0x0000000f0000LL;
187
188     // (raw attributes = 0xf3)
189     core->segments.cs.type = 0x3;
190     core->segments.cs.system = 0x1;
191     core->segments.cs.dpl = 0x3;
192     core->segments.cs.present = 1;
193
194
195
196     struct v3_segment * segregs [] = {&(core->segments.ss), &(core->segments.ds), 
197                                       &(core->segments.es), &(core->segments.fs), 
198                                       &(core->segments.gs), NULL};
199
200     for ( i = 0; segregs[i] != NULL; i++) {
201         struct v3_segment * seg = segregs[i];
202         
203         seg->selector = 0x0000;
204         //    seg->base = seg->selector << 4;
205         seg->base = 0x00000000;
206         seg->limit = ~0u;
207
208         // (raw attributes = 0xf3)
209         seg->type = 0x3;
210         seg->system = 0x1;
211         seg->dpl = 0x3;
212         seg->present = 1;
213     }
214
215     core->segments.gdtr.limit = 0x0000ffff;
216     core->segments.gdtr.base = 0x0000000000000000LL;
217     core->segments.idtr.limit = 0x0000ffff;
218     core->segments.idtr.base = 0x0000000000000000LL;
219
220     core->segments.ldtr.selector = 0x0000;
221     core->segments.ldtr.limit = 0x0000ffff;
222     core->segments.ldtr.base = 0x0000000000000000LL;
223     core->segments.tr.selector = 0x0000;
224     core->segments.tr.limit = 0x0000ffff;
225     core->segments.tr.base = 0x0000000000000000LL;
226
227
228     core->dbg_regs.dr6 = 0x00000000ffff0ff0LL;
229     core->dbg_regs.dr7 = 0x0000000000000400LL;
230
231
232     ctrl_area->IOPM_BASE_PA = (addr_t)V3_PAddr(core->vm_info->io_map.arch_data);
233     ctrl_area->instrs.IOIO_PROT = 1;
234             
235     ctrl_area->MSRPM_BASE_PA = (addr_t)V3_PAddr(core->vm_info->msr_map.arch_data);
236     ctrl_area->instrs.MSR_PROT = 1;   
237
238
239     PrintDebug("Exiting on interrupts\n");
240     ctrl_area->guest_ctrl.V_INTR_MASKING = 1;
241     ctrl_area->instrs.INTR = 1;
242
243
244     v3_hook_msr(core->vm_info, EFER_MSR, 
245                 &v3_handle_efer_read,
246                 &v3_svm_handle_efer_write, 
247                 core);
248
249     if (core->shdw_pg_mode == SHADOW_PAGING) {
250         PrintDebug("Creating initial shadow page table\n");
251         
252         /* JRL: This is a performance killer, and a simplistic solution */
253         /* We need to fix this */
254         ctrl_area->TLB_CONTROL = 1;
255         ctrl_area->guest_ASID = 1;
256         
257         
258         if (v3_init_passthrough_pts(core) == -1) {
259             PrintError("Could not initialize passthrough page tables\n");
260             return ;
261         }
262
263
264         core->shdw_pg_state.guest_cr0 = 0x0000000000000010LL;
265         PrintDebug("Created\n");
266         
267         core->ctrl_regs.cr0 |= 0x80000000;
268         core->ctrl_regs.cr3 = core->direct_map_pt;
269
270         ctrl_area->cr_reads.cr0 = 1;
271         ctrl_area->cr_writes.cr0 = 1;
272         //ctrl_area->cr_reads.cr4 = 1;
273         ctrl_area->cr_writes.cr4 = 1;
274         ctrl_area->cr_reads.cr3 = 1;
275         ctrl_area->cr_writes.cr3 = 1;
276
277
278
279         ctrl_area->instrs.INVLPG = 1;
280
281         ctrl_area->exceptions.pf = 1;
282
283         guest_state->g_pat = 0x7040600070406ULL;
284
285
286
287     } else if (core->shdw_pg_mode == NESTED_PAGING) {
288         // Flush the TLB on entries/exits
289         ctrl_area->TLB_CONTROL = 1;
290         ctrl_area->guest_ASID = 1;
291
292         // Enable Nested Paging
293         ctrl_area->NP_ENABLE = 1;
294
295         PrintDebug("NP_Enable at 0x%p\n", (void *)&(ctrl_area->NP_ENABLE));
296
297         // Set the Nested Page Table pointer
298         if (v3_init_passthrough_pts(core) == -1) {
299             PrintError("Could not initialize Nested page tables\n");
300             return ;
301         }
302
303         ctrl_area->N_CR3 = core->direct_map_pt;
304
305         guest_state->g_pat = 0x7040600070406ULL;
306     }
307     
308     /* tell the guest that we don't support SVM */
309     v3_hook_msr(core->vm_info, SVM_VM_CR_MSR, 
310         &v3_handle_vm_cr_read,
311         &v3_handle_vm_cr_write, 
312         core);
313
314
315     {
316 #define INT_PENDING_AMD_MSR             0xc0010055
317
318         v3_hook_msr(core->vm_info, IA32_STAR_MSR, NULL, NULL, NULL);
319         v3_hook_msr(core->vm_info, IA32_LSTAR_MSR, NULL, NULL, NULL);
320         v3_hook_msr(core->vm_info, IA32_FMASK_MSR, NULL, NULL, NULL);
321         v3_hook_msr(core->vm_info, IA32_KERN_GS_BASE_MSR, NULL, NULL, NULL);
322         v3_hook_msr(core->vm_info, IA32_CSTAR_MSR, NULL, NULL, NULL);
323
324         v3_hook_msr(core->vm_info, SYSENTER_CS_MSR, NULL, NULL, NULL);
325         v3_hook_msr(core->vm_info, SYSENTER_ESP_MSR, NULL, NULL, NULL);
326         v3_hook_msr(core->vm_info, SYSENTER_EIP_MSR, NULL, NULL, NULL);
327
328
329         v3_hook_msr(core->vm_info, FS_BASE_MSR, NULL, NULL, NULL);
330         v3_hook_msr(core->vm_info, GS_BASE_MSR, NULL, NULL, NULL);
331
332         // Passthrough read operations are ok.
333         v3_hook_msr(core->vm_info, INT_PENDING_AMD_MSR, NULL, v3_msr_unhandled_write, NULL);
334     }
335 }
336
337
338 int v3_init_svm_vmcb(struct guest_info * core, v3_vm_class_t vm_class) {
339
340     PrintDebug("Allocating VMCB\n");
341     core->vmm_data = (void *)Allocate_VMCB();
342     
343     if (core->vmm_data == NULL) {
344         PrintError("Could not allocate VMCB, Exiting...\n");
345         return -1;
346     }
347
348     if (vm_class == V3_PC_VM) {
349         PrintDebug("Initializing VMCB (addr=%p)\n", (void *)core->vmm_data);
350         Init_VMCB_BIOS((vmcb_t*)(core->vmm_data), core);
351     } else {
352         PrintError("Invalid VM class\n");
353         return -1;
354     }
355
356     core->core_run_state = CORE_STOPPED;
357
358     return 0;
359 }
360
361
362 int v3_deinit_svm_vmcb(struct guest_info * core) {
363     V3_FreePages(V3_PAddr(core->vmm_data), 1);
364     return 0;
365 }
366
367
368 #ifdef V3_CONFIG_CHECKPOINT
369 int v3_svm_save_core(struct guest_info * core, void * ctx){
370
371     v3_chkpt_save_8(ctx, "cpl", &(core->cpl));
372     v3_chkpt_save(ctx, "vmcb_data", PAGE_SIZE, core->vmm_data);
373
374     return 0;
375 }
376
377 int v3_svm_load_core(struct guest_info * core, void * ctx){
378     
379     v3_chkpt_load_8(ctx, "cpl", &(core->cpl));
380
381     if (v3_chkpt_load(ctx, "vmcb_data", PAGE_SIZE, core->vmm_data) == -1) {
382         return -1;
383     }
384
385     return 0;
386 }
387 #endif
388
389 static int update_irq_exit_state(struct guest_info * info) {
390     vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
391
392     // Fix for QEMU bug using EVENTINJ as an internal cache
393     guest_ctrl->EVENTINJ.valid = 0;
394
395     if ((info->intr_core_state.irq_pending == 1) && (guest_ctrl->guest_ctrl.V_IRQ == 0)) {
396         
397 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
398         PrintDebug("INTAK cycle completed for irq %d\n", info->intr_core_state.irq_vector);
399 #endif
400
401         info->intr_core_state.irq_started = 1;
402         info->intr_core_state.irq_pending = 0;
403
404         v3_injecting_intr(info, info->intr_core_state.irq_vector, V3_EXTERNAL_IRQ);
405     }
406
407     if ((info->intr_core_state.irq_started == 1) && (guest_ctrl->exit_int_info.valid == 0)) {
408 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
409         PrintDebug("Interrupt %d taken by guest\n", info->intr_core_state.irq_vector);
410 #endif
411
412         // Interrupt was taken fully vectored
413         info->intr_core_state.irq_started = 0;
414
415     } else if ((info->intr_core_state.irq_started == 1) && (guest_ctrl->exit_int_info.valid == 1)) {
416 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
417         PrintDebug("EXIT INT INFO is set (vec=%d)\n", guest_ctrl->exit_int_info.vector);
418 #endif
419     }
420
421     return 0;
422 }
423
424
425 static int update_irq_entry_state(struct guest_info * info) {
426     vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
427
428
429     if (info->intr_core_state.irq_pending == 0) {
430         guest_ctrl->guest_ctrl.V_IRQ = 0;
431         guest_ctrl->guest_ctrl.V_INTR_VECTOR = 0;
432     }
433     
434     if (v3_excp_pending(info)) {
435         uint_t excp = v3_get_excp_number(info);
436         
437         guest_ctrl->EVENTINJ.type = SVM_INJECTION_EXCEPTION;
438         
439         if (info->excp_state.excp_error_code_valid) {
440             guest_ctrl->EVENTINJ.error_code = info->excp_state.excp_error_code;
441             guest_ctrl->EVENTINJ.ev = 1;
442 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
443             PrintDebug("Injecting exception %d with error code %x\n", excp, guest_ctrl->EVENTINJ.error_code);
444 #endif
445         }
446         
447         guest_ctrl->EVENTINJ.vector = excp;
448         
449         guest_ctrl->EVENTINJ.valid = 1;
450
451 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
452         PrintDebug("<%d> Injecting Exception %d (CR2=%p) (EIP=%p)\n", 
453                    (int)info->num_exits, 
454                    guest_ctrl->EVENTINJ.vector, 
455                    (void *)(addr_t)info->ctrl_regs.cr2,
456                    (void *)(addr_t)info->rip);
457 #endif
458
459         v3_injecting_excp(info, excp);
460     } else if (info->intr_core_state.irq_started == 1) {
461 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
462         PrintDebug("IRQ pending from previous injection\n");
463 #endif
464         guest_ctrl->guest_ctrl.V_IRQ = 1;
465         guest_ctrl->guest_ctrl.V_INTR_VECTOR = info->intr_core_state.irq_vector;
466         guest_ctrl->guest_ctrl.V_IGN_TPR = 1;
467         guest_ctrl->guest_ctrl.V_INTR_PRIO = 0xf;
468
469     } else {
470         switch (v3_intr_pending(info)) {
471             case V3_EXTERNAL_IRQ: {
472                 uint32_t irq = v3_get_intr(info);
473
474                 guest_ctrl->guest_ctrl.V_IRQ = 1;
475                 guest_ctrl->guest_ctrl.V_INTR_VECTOR = irq;
476                 guest_ctrl->guest_ctrl.V_IGN_TPR = 1;
477                 guest_ctrl->guest_ctrl.V_INTR_PRIO = 0xf;
478
479 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
480                 PrintDebug("Injecting Interrupt %d (EIP=%p)\n", 
481                            guest_ctrl->guest_ctrl.V_INTR_VECTOR, 
482                            (void *)(addr_t)info->rip);
483 #endif
484
485                 info->intr_core_state.irq_pending = 1;
486                 info->intr_core_state.irq_vector = irq;
487                 
488                 break;
489             }
490             case V3_NMI:
491                 guest_ctrl->EVENTINJ.type = SVM_INJECTION_NMI;
492                 break;
493             case V3_SOFTWARE_INTR:
494                 guest_ctrl->EVENTINJ.type = SVM_INJECTION_SOFT_INTR;
495
496 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
497                 PrintDebug("Injecting software interrupt --  type: %d, vector: %d\n", 
498                            SVM_INJECTION_SOFT_INTR, info->intr_core_state.swintr_vector);
499 #endif
500                 guest_ctrl->EVENTINJ.vector = info->intr_core_state.swintr_vector;
501                 guest_ctrl->EVENTINJ.valid = 1;
502             
503                 /* reset swintr state */
504                 info->intr_core_state.swintr_posted = 0;
505                 info->intr_core_state.swintr_vector = 0;
506                 
507                 break;
508             case V3_VIRTUAL_IRQ:
509                 guest_ctrl->EVENTINJ.type = SVM_INJECTION_IRQ;
510                 break;
511
512             case V3_INVALID_INTR:
513             default:
514                 break;
515         }
516         
517     }
518
519     return 0;
520 }
521
522 int 
523 v3_svm_config_tsc_virtualization(struct guest_info * info) {
524     vmcb_ctrl_t * ctrl_area = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
525
526     if (info->time_state.time_flags & V3_TIME_TRAP_RDTSC) {
527         ctrl_area->instrs.RDTSC = 1;
528         ctrl_area->svm_instrs.RDTSCP = 1;
529     } else {
530         ctrl_area->instrs.RDTSC = 0;
531         ctrl_area->svm_instrs.RDTSCP = 0;
532         ctrl_area->TSC_OFFSET = v3_tsc_host_offset(&info->time_state);
533     }
534     return 0;
535 }
536
537 /* 
538  * CAUTION and DANGER!!! 
539  * 
540  * The VMCB CANNOT(!!) be accessed outside of the clgi/stgi calls inside this function
541  * When exectuing a symbiotic call, the VMCB WILL be overwritten, so any dependencies 
542  * on its contents will cause things to break. The contents at the time of the exit WILL 
543  * change before the exit handler is executed.
544  */
545 int v3_svm_enter(struct guest_info * info) {
546     vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
547     vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA((vmcb_t*)(info->vmm_data)); 
548     addr_t exit_code = 0, exit_info1 = 0, exit_info2 = 0;
549     uint64_t guest_cycles = 0;
550
551     // Conditionally yield the CPU if the timeslice has expired
552     v3_yield_cond(info);
553
554     // disable global interrupts for vm state transition
555     v3_clgi();
556
557     // Update timer devices after being in the VM, with interupts
558     // disabled, but before doing IRQ updates, so that any interrupts they 
559     //raise get seen immediately.
560     v3_advance_time(info);
561     v3_update_timers(info);
562
563     // Synchronize the guest state to the VMCB
564     guest_state->cr0 = info->ctrl_regs.cr0;
565     guest_state->cr2 = info->ctrl_regs.cr2;
566     guest_state->cr3 = info->ctrl_regs.cr3;
567     guest_state->cr4 = info->ctrl_regs.cr4;
568     guest_state->dr6 = info->dbg_regs.dr6;
569     guest_state->dr7 = info->dbg_regs.dr7;
570     guest_ctrl->guest_ctrl.V_TPR = info->ctrl_regs.cr8 & 0xff;
571     guest_state->rflags = info->ctrl_regs.rflags;
572     guest_state->efer = info->ctrl_regs.efer;
573     
574     /* Synchronize MSRs */
575     guest_state->star = info->msrs.star;
576     guest_state->lstar = info->msrs.lstar;
577     guest_state->sfmask = info->msrs.sfmask;
578     guest_state->KernelGsBase = info->msrs.kern_gs_base;
579
580     guest_state->cpl = info->cpl;
581
582     v3_set_vmcb_segments((vmcb_t*)(info->vmm_data), &(info->segments));
583
584     guest_state->rax = info->vm_regs.rax;
585     guest_state->rip = info->rip;
586     guest_state->rsp = info->vm_regs.rsp;
587
588 #ifdef V3_CONFIG_SYMCALL
589     if (info->sym_core_state.symcall_state.sym_call_active == 0) {
590         update_irq_entry_state(info);
591     }
592 #else 
593     update_irq_entry_state(info);
594 #endif
595
596
597     /* ** */
598
599     /*
600       PrintDebug("SVM Entry to CS=%p  rip=%p...\n", 
601       (void *)(addr_t)info->segments.cs.base, 
602       (void *)(addr_t)info->rip);
603     */
604
605 #ifdef V3_CONFIG_SYMCALL
606     if (info->sym_core_state.symcall_state.sym_call_active == 1) {
607         if (guest_ctrl->guest_ctrl.V_IRQ == 1) {
608             V3_Print("!!! Injecting Interrupt during Sym call !!!\n");
609         }
610     }
611 #endif
612
613     v3_time_enter_vm(info);
614     v3_svm_config_tsc_virtualization(info);
615
616     //V3_Print("Calling v3_svm_launch\n");
617     {   
618         uint64_t entry_tsc = 0;
619         uint64_t exit_tsc = 0;
620         
621         rdtscll(entry_tsc);
622
623         v3_svm_launch((vmcb_t *)V3_PAddr(info->vmm_data), &(info->vm_regs), (vmcb_t *)host_vmcbs[V3_Get_CPU()]);
624
625         rdtscll(exit_tsc);
626
627         guest_cycles = exit_tsc - entry_tsc;
628     }
629
630
631     //V3_Print("SVM Returned: Exit Code: %x, guest_rip=%lx\n", (uint32_t)(guest_ctrl->exit_code), (unsigned long)guest_state->rip);
632
633     v3_last_exit = (uint32_t)(guest_ctrl->exit_code);
634
635     // Immediate exit from VM time bookkeeping
636     v3_time_exit_vm(info, &guest_cycles);
637
638     info->num_exits++;
639
640     // Save Guest state from VMCB
641     info->rip = guest_state->rip;
642     info->vm_regs.rsp = guest_state->rsp;
643     info->vm_regs.rax = guest_state->rax;
644
645     info->cpl = guest_state->cpl;
646
647     info->ctrl_regs.cr0 = guest_state->cr0;
648     info->ctrl_regs.cr2 = guest_state->cr2;
649     info->ctrl_regs.cr3 = guest_state->cr3;
650     info->ctrl_regs.cr4 = guest_state->cr4;
651     info->dbg_regs.dr6 = guest_state->dr6;
652     info->dbg_regs.dr7 = guest_state->dr7;
653     info->ctrl_regs.cr8 = guest_ctrl->guest_ctrl.V_TPR;
654     info->ctrl_regs.rflags = guest_state->rflags;
655     info->ctrl_regs.efer = guest_state->efer;
656     
657     /* Synchronize MSRs */
658     info->msrs.star =  guest_state->star;
659     info->msrs.lstar = guest_state->lstar;
660     info->msrs.sfmask = guest_state->sfmask;
661     info->msrs.kern_gs_base = guest_state->KernelGsBase;
662
663     v3_get_vmcb_segments((vmcb_t*)(info->vmm_data), &(info->segments));
664     info->cpu_mode = v3_get_vm_cpu_mode(info);
665     info->mem_mode = v3_get_vm_mem_mode(info);
666     /* ** */
667
668     // save exit info here
669     exit_code = guest_ctrl->exit_code;
670     exit_info1 = guest_ctrl->exit_info1;
671     exit_info2 = guest_ctrl->exit_info2;
672
673 #ifdef V3_CONFIG_SYMCALL
674     if (info->sym_core_state.symcall_state.sym_call_active == 0) {
675         update_irq_exit_state(info);
676     }
677 #else
678     update_irq_exit_state(info);
679 #endif
680
681     // reenable global interrupts after vm exit
682     v3_stgi();
683  
684     // Conditionally yield the CPU if the timeslice has expired
685     v3_yield_cond(info);
686
687     {
688         int ret = v3_handle_svm_exit(info, exit_code, exit_info1, exit_info2);
689         
690         if (ret != 0) {
691             PrintError("Error in SVM exit handler (ret=%d)\n", ret);
692             PrintError("  last Exit was %d (exit code=0x%llx)\n", v3_last_exit, (uint64_t) exit_code);
693             return -1;
694         }
695     }
696
697     if (info->timeouts.timeout_active) {
698         /* Check to see if any timeouts have expired */
699         v3_handle_timeouts(info, guest_cycles);
700     }
701
702
703     return 0;
704 }
705
706
707 int v3_start_svm_guest(struct guest_info * info) {
708     //    vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA((vmcb_t*)(info->vmm_data));
709     //  vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
710
711     PrintDebug("Starting SVM core %u (on logical core %u)\n", info->vcpu_id, info->pcpu_id);
712
713     if (info->vcpu_id == 0) {
714         info->core_run_state = CORE_RUNNING;
715     } else  { 
716         PrintDebug("SVM core %u (on %u): Waiting for core initialization\n", info->vcpu_id, info->pcpu_id);
717
718         while (info->core_run_state == CORE_STOPPED) {
719             
720             if (info->vm_info->run_state == VM_STOPPED) {
721                 // The VM was stopped before this core was initialized. 
722                 return 0;
723             }
724
725             v3_yield(info);
726             //PrintDebug("SVM core %u: still waiting for INIT\n", info->vcpu_id);
727         }
728
729         PrintDebug("SVM core %u(on %u) initialized\n", info->vcpu_id, info->pcpu_id);
730
731         // We'll be paranoid about race conditions here
732         v3_wait_at_barrier(info);
733     } 
734
735     PrintDebug("SVM core %u(on %u): I am starting at CS=0x%x (base=0x%p, limit=0x%x),  RIP=0x%p\n", 
736                info->vcpu_id, info->pcpu_id, 
737                info->segments.cs.selector, (void *)(info->segments.cs.base), 
738                info->segments.cs.limit, (void *)(info->rip));
739
740
741
742     PrintDebug("SVM core %u: Launching SVM VM (vmcb=%p) (on cpu %u)\n", 
743                info->vcpu_id, (void *)info->vmm_data, info->pcpu_id);
744     //PrintDebugVMCB((vmcb_t*)(info->vmm_data));
745     
746     v3_start_time(info);
747
748     while (1) {
749
750         if (info->vm_info->run_state == VM_STOPPED) {
751             info->core_run_state = CORE_STOPPED;
752             break;
753         }
754         
755         if (v3_svm_enter(info) == -1) {
756             vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
757             addr_t host_addr;
758             addr_t linear_addr = 0;
759             
760             info->vm_info->run_state = VM_ERROR;
761             
762             V3_Print("SVM core %u: SVM ERROR!!\n", info->vcpu_id); 
763             
764             v3_print_guest_state(info);
765             
766             V3_Print("SVM core %u: SVM Exit Code: %p\n", info->vcpu_id, (void *)(addr_t)guest_ctrl->exit_code); 
767             
768             V3_Print("SVM core %u: exit_info1 low = 0x%.8x\n", info->vcpu_id, *(uint_t*)&(guest_ctrl->exit_info1));
769             V3_Print("SVM core %u: exit_info1 high = 0x%.8x\n", info->vcpu_id, *(uint_t *)(((uchar_t *)&(guest_ctrl->exit_info1)) + 4));
770             
771             V3_Print("SVM core %u: exit_info2 low = 0x%.8x\n", info->vcpu_id, *(uint_t*)&(guest_ctrl->exit_info2));
772             V3_Print("SVM core %u: exit_info2 high = 0x%.8x\n", info->vcpu_id, *(uint_t *)(((uchar_t *)&(guest_ctrl->exit_info2)) + 4));
773             
774             linear_addr = get_addr_linear(info, info->rip, &(info->segments.cs));
775             
776             if (info->mem_mode == PHYSICAL_MEM) {
777                 v3_gpa_to_hva(info, linear_addr, &host_addr);
778             } else if (info->mem_mode == VIRTUAL_MEM) {
779                 v3_gva_to_hva(info, linear_addr, &host_addr);
780             }
781             
782             V3_Print("SVM core %u: Host Address of rip = 0x%p\n", info->vcpu_id, (void *)host_addr);
783             
784             V3_Print("SVM core %u: Instr (15 bytes) at %p:\n", info->vcpu_id, (void *)host_addr);
785             v3_dump_mem((uint8_t *)host_addr, 15);
786             
787             v3_print_stack(info);
788
789             break;
790         }
791
792         v3_wait_at_barrier(info);
793
794
795         if (info->vm_info->run_state == VM_STOPPED) {
796             info->core_run_state = CORE_STOPPED;
797             break;
798         }
799
800         
801
802 /*
803         if ((info->num_exits % 50000) == 0) {
804             V3_Print("SVM Exit number %d\n", (uint32_t)info->num_exits);
805             v3_print_guest_state(info);
806         }
807 */
808         
809     }
810
811     // Need to take down the other cores on error... 
812
813     return 0;
814 }
815
816
817
818
819 int v3_reset_svm_vm_core(struct guest_info * core, addr_t rip) {
820     // init vmcb_bios
821
822     // Write the RIP, CS, and descriptor
823     // assume the rest is already good to go
824     //
825     // vector VV -> rip at 0
826     //              CS = VV00
827     //  This means we start executing at linear address VV000
828     //
829     // So the selector needs to be VV00
830     // and the base needs to be VV000
831     //
832     core->rip = 0;
833     core->segments.cs.selector = rip << 8;
834     core->segments.cs.limit = 0xffff;
835     core->segments.cs.base = rip << 12;
836
837     return 0;
838 }
839
840
841
842
843
844
845 /* Checks machine SVM capability */
846 /* Implemented from: AMD Arch Manual 3, sect 15.4 */ 
847 int v3_is_svm_capable() {
848     uint_t vm_cr_low = 0, vm_cr_high = 0;
849     uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0;
850
851     v3_cpuid(CPUID_EXT_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
852   
853     PrintDebug("CPUID_EXT_FEATURE_IDS_ecx=0x%x\n", ecx);
854
855     if ((ecx & CPUID_EXT_FEATURE_IDS_ecx_svm_avail) == 0) {
856       V3_Print("SVM Not Available\n");
857       return 0;
858     }  else {
859         v3_get_msr(SVM_VM_CR_MSR, &vm_cr_high, &vm_cr_low);
860         
861         PrintDebug("SVM_VM_CR_MSR = 0x%x 0x%x\n", vm_cr_high, vm_cr_low);
862         
863         if ((vm_cr_low & SVM_VM_CR_MSR_svmdis) == 1) {
864             V3_Print("SVM is available but is disabled.\n");
865             
866             v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
867             
868             PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_edx=0x%x\n", edx);
869             
870             if ((edx & CPUID_SVM_REV_AND_FEATURE_IDS_edx_svml) == 0) {
871                 V3_Print("SVM BIOS Disabled, not unlockable\n");
872             } else {
873                 V3_Print("SVM is locked with a key\n");
874             }
875             return 0;
876
877         } else {
878             V3_Print("SVM is available and  enabled.\n");
879
880             v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
881             PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_eax=0x%x\n", eax);
882             PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_ebx=0x%x\n", ebx);
883             PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_ecx=0x%x\n", ecx);
884             PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_edx=0x%x\n", edx);
885
886             return 1;
887         }
888     }
889 }
890
891 static int has_svm_nested_paging() {
892     uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0;
893     
894     v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
895     
896     //PrintDebug("CPUID_EXT_FEATURE_IDS_edx=0x%x\n", edx);
897     
898     if ((edx & CPUID_SVM_REV_AND_FEATURE_IDS_edx_np) == 0) {
899         V3_Print("SVM Nested Paging not supported\n");
900         return 0;
901     } else {
902         V3_Print("SVM Nested Paging supported\n");
903         return 1;
904     }
905  }
906  
907
908
909 void v3_init_svm_cpu(int cpu_id) {
910     reg_ex_t msr;
911     extern v3_cpu_arch_t v3_cpu_types[];
912
913     // Enable SVM on the CPU
914     v3_get_msr(EFER_MSR, &(msr.e_reg.high), &(msr.e_reg.low));
915     msr.e_reg.low |= EFER_MSR_svm_enable;
916     v3_set_msr(EFER_MSR, 0, msr.e_reg.low);
917
918     V3_Print("SVM Enabled\n");
919
920     // Setup the host state save area
921     host_vmcbs[cpu_id] = (addr_t)V3_AllocPages(4);
922
923     /* 64-BIT-ISSUE */
924     //  msr.e_reg.high = 0;
925     //msr.e_reg.low = (uint_t)host_vmcb;
926     msr.r_reg = host_vmcbs[cpu_id];
927
928     PrintDebug("Host State being saved at %p\n", (void *)host_vmcbs[cpu_id]);
929     v3_set_msr(SVM_VM_HSAVE_PA_MSR, msr.e_reg.high, msr.e_reg.low);
930
931
932     if (has_svm_nested_paging() == 1) {
933         v3_cpu_types[cpu_id] = V3_SVM_REV3_CPU;
934     } else {
935         v3_cpu_types[cpu_id] = V3_SVM_CPU;
936     }
937 }
938
939
940
941 void v3_deinit_svm_cpu(int cpu_id) {
942     reg_ex_t msr;
943     extern v3_cpu_arch_t v3_cpu_types[];
944
945     // reset SVM_VM_HSAVE_PA_MSR
946     // Does setting it to NULL disable??
947     msr.r_reg = 0;
948     v3_set_msr(SVM_VM_HSAVE_PA_MSR, msr.e_reg.high, msr.e_reg.low);
949
950     // Disable SVM?
951     v3_get_msr(EFER_MSR, &(msr.e_reg.high), &(msr.e_reg.low));
952     msr.e_reg.low &= ~EFER_MSR_svm_enable;
953     v3_set_msr(EFER_MSR, 0, msr.e_reg.low);
954
955     v3_cpu_types[cpu_id] = V3_INVALID_CPU;
956
957     V3_FreePages((void *)host_vmcbs[cpu_id], 4);
958
959     V3_Print("Host CPU %d host area freed, and SVM disabled\n", cpu_id);
960     return;
961 }
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012 #if 0
1013 /* 
1014  * Test VMSAVE/VMLOAD Latency 
1015  */
1016 #define vmsave ".byte 0x0F,0x01,0xDB ; "
1017 #define vmload ".byte 0x0F,0x01,0xDA ; "
1018 {
1019     uint32_t start_lo, start_hi;
1020     uint32_t end_lo, end_hi;
1021     uint64_t start, end;
1022     
1023     __asm__ __volatile__ (
1024                           "rdtsc ; "
1025                           "movl %%eax, %%esi ; "
1026                           "movl %%edx, %%edi ; "
1027                           "movq  %%rcx, %%rax ; "
1028                           vmsave
1029                           "rdtsc ; "
1030                           : "=D"(start_hi), "=S"(start_lo), "=a"(end_lo),"=d"(end_hi)
1031                           : "c"(host_vmcb[cpu_id]), "0"(0), "1"(0), "2"(0), "3"(0)
1032                           );
1033     
1034     start = start_hi;
1035     start <<= 32;
1036     start += start_lo;
1037     
1038     end = end_hi;
1039     end <<= 32;
1040     end += end_lo;
1041     
1042     PrintDebug("VMSave Cycle Latency: %d\n", (uint32_t)(end - start));
1043     
1044     __asm__ __volatile__ (
1045                           "rdtsc ; "
1046                           "movl %%eax, %%esi ; "
1047                           "movl %%edx, %%edi ; "
1048                           "movq  %%rcx, %%rax ; "
1049                           vmload
1050                           "rdtsc ; "
1051                           : "=D"(start_hi), "=S"(start_lo), "=a"(end_lo),"=d"(end_hi)
1052                               : "c"(host_vmcb[cpu_id]), "0"(0), "1"(0), "2"(0), "3"(0)
1053                               );
1054         
1055         start = start_hi;
1056         start <<= 32;
1057         start += start_lo;
1058
1059         end = end_hi;
1060         end <<= 32;
1061         end += end_lo;
1062
1063
1064         PrintDebug("VMLoad Cycle Latency: %d\n", (uint32_t)(end - start));
1065     }
1066     /* End Latency Test */
1067
1068 #endif
1069
1070
1071
1072
1073
1074
1075
1076 #if 0
1077 void Init_VMCB_pe(vmcb_t *vmcb, struct guest_info vm_info) {
1078   vmcb_ctrl_t * ctrl_area = GET_VMCB_CTRL_AREA(vmcb);
1079   vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA(vmcb);
1080   uint_t i = 0;
1081
1082
1083   guest_state->rsp = vm_info.vm_regs.rsp;
1084   guest_state->rip = vm_info.rip;
1085
1086
1087   /* I pretty much just gutted this from TVMM */
1088   /* Note: That means its probably wrong */
1089
1090   // set the segment registers to mirror ours
1091   guest_state->cs.selector = 1<<3;
1092   guest_state->cs.attrib.fields.type = 0xa; // Code segment+read
1093   guest_state->cs.attrib.fields.S = 1;
1094   guest_state->cs.attrib.fields.P = 1;
1095   guest_state->cs.attrib.fields.db = 1;
1096   guest_state->cs.attrib.fields.G = 1;
1097   guest_state->cs.limit = 0xfffff;
1098   guest_state->cs.base = 0;
1099   
1100   struct vmcb_selector *segregs [] = {&(guest_state->ss), &(guest_state->ds), &(guest_state->es), &(guest_state->fs), &(guest_state->gs), NULL};
1101   for ( i = 0; segregs[i] != NULL; i++) {
1102     struct vmcb_selector * seg = segregs[i];
1103     
1104     seg->selector = 2<<3;
1105     seg->attrib.fields.type = 0x2; // Data Segment+read/write
1106     seg->attrib.fields.S = 1;
1107     seg->attrib.fields.P = 1;
1108     seg->attrib.fields.db = 1;
1109     seg->attrib.fields.G = 1;
1110     seg->limit = 0xfffff;
1111     seg->base = 0;
1112   }
1113
1114
1115   {
1116     /* JRL THIS HAS TO GO */
1117     
1118     //    guest_state->tr.selector = GetTR_Selector();
1119     guest_state->tr.attrib.fields.type = 0x9; 
1120     guest_state->tr.attrib.fields.P = 1;
1121     // guest_state->tr.limit = GetTR_Limit();
1122     //guest_state->tr.base = GetTR_Base();// - 0x2000;
1123     /* ** */
1124   }
1125
1126
1127   /* ** */
1128
1129
1130   guest_state->efer |= EFER_MSR_svm_enable;
1131   guest_state->rflags = 0x00000002; // The reserved bit is always 1
1132   ctrl_area->svm_instrs.VMRUN = 1;
1133   guest_state->cr0 = 0x00000001;    // PE 
1134   ctrl_area->guest_ASID = 1;
1135
1136
1137   //  guest_state->cpl = 0;
1138
1139
1140
1141   // Setup exits
1142
1143   ctrl_area->cr_writes.cr4 = 1;
1144   
1145   ctrl_area->exceptions.de = 1;
1146   ctrl_area->exceptions.df = 1;
1147   ctrl_area->exceptions.pf = 1;
1148   ctrl_area->exceptions.ts = 1;
1149   ctrl_area->exceptions.ss = 1;
1150   ctrl_area->exceptions.ac = 1;
1151   ctrl_area->exceptions.mc = 1;
1152   ctrl_area->exceptions.gp = 1;
1153   ctrl_area->exceptions.ud = 1;
1154   ctrl_area->exceptions.np = 1;
1155   ctrl_area->exceptions.of = 1;
1156   ctrl_area->exceptions.nmi = 1;
1157
1158   
1159
1160   ctrl_area->instrs.IOIO_PROT = 1;
1161   ctrl_area->IOPM_BASE_PA = (uint_t)V3_AllocPages(3);
1162   
1163   {
1164     reg_ex_t tmp_reg;
1165     tmp_reg.r_reg = ctrl_area->IOPM_BASE_PA;
1166     memset((void*)(tmp_reg.e_reg.low), 0xffffffff, PAGE_SIZE * 2);
1167   }
1168
1169   ctrl_area->instrs.INTR = 1;
1170
1171   
1172   {
1173     char gdt_buf[6];
1174     char idt_buf[6];
1175
1176     memset(gdt_buf, 0, 6);
1177     memset(idt_buf, 0, 6);
1178
1179
1180     uint_t gdt_base, idt_base;
1181     ushort_t gdt_limit, idt_limit;
1182     
1183     GetGDTR(gdt_buf);
1184     gdt_base = *(ulong_t*)((uchar_t*)gdt_buf + 2) & 0xffffffff;
1185     gdt_limit = *(ushort_t*)(gdt_buf) & 0xffff;
1186     PrintDebug("GDT: base: %x, limit: %x\n", gdt_base, gdt_limit);
1187
1188     GetIDTR(idt_buf);
1189     idt_base = *(ulong_t*)(idt_buf + 2) & 0xffffffff;
1190     idt_limit = *(ushort_t*)(idt_buf) & 0xffff;
1191     PrintDebug("IDT: base: %x, limit: %x\n",idt_base, idt_limit);
1192
1193
1194     // gdt_base -= 0x2000;
1195     //idt_base -= 0x2000;
1196
1197     guest_state->gdtr.base = gdt_base;
1198     guest_state->gdtr.limit = gdt_limit;
1199     guest_state->idtr.base = idt_base;
1200     guest_state->idtr.limit = idt_limit;
1201
1202
1203   }
1204   
1205   
1206   // also determine if CPU supports nested paging
1207   /*
1208   if (vm_info.page_tables) {
1209     //   if (0) {
1210     // Flush the TLB on entries/exits
1211     ctrl_area->TLB_CONTROL = 1;
1212
1213     // Enable Nested Paging
1214     ctrl_area->NP_ENABLE = 1;
1215
1216     PrintDebug("NP_Enable at 0x%x\n", &(ctrl_area->NP_ENABLE));
1217
1218         // Set the Nested Page Table pointer
1219     ctrl_area->N_CR3 |= ((addr_t)vm_info.page_tables & 0xfffff000);
1220
1221
1222     //   ctrl_area->N_CR3 = Get_CR3();
1223     // guest_state->cr3 |= (Get_CR3() & 0xfffff000);
1224
1225     guest_state->g_pat = 0x7040600070406ULL;
1226
1227     PrintDebug("Set Nested CR3: lo: 0x%x  hi: 0x%x\n", (uint_t)*(&(ctrl_area->N_CR3)), (uint_t)*((unsigned char *)&(ctrl_area->N_CR3) + 4));
1228     PrintDebug("Set Guest CR3: lo: 0x%x  hi: 0x%x\n", (uint_t)*(&(guest_state->cr3)), (uint_t)*((unsigned char *)&(guest_state->cr3) + 4));
1229     // Enable Paging
1230     //    guest_state->cr0 |= 0x80000000;
1231   }
1232   */
1233
1234 }
1235
1236
1237
1238
1239
1240 #endif
1241
1242