+/*
+ * This file is part of the Palacios Virtual Machine Monitor developed
+ * by the V3VEE Project with funding from the United States National
+ * Science Foundation and the Department of Energy.
+ *
+ * The V3VEE Project is a joint project between Northwestern University
+ * and the University of New Mexico. You can find out more at
+ * http://www.v3vee.org
+ *
+ * Copyright (c) 2008, Jack Lange <jarusl@cs.northwestern.edu>
+ * Copyright (c) 2008, The V3VEE Project <http://www.v3vee.org>
+ * All rights reserved.
+ *
+ * Author: Jack Lange <jarusl@cs.northwestern.edu>
+ *
+ * This is free software. You are permitted to use,
+ * redistribute, and modify it as specified in the file "V3VEE_LICENSE".
+ */
+
+
#include <palacios/svm.h>
#include <palacios/vmm.h>
#include <palacios/vm_guest_mem.h>
#include <palacios/vmm_decoder.h>
+#include <palacios/vmm_string.h>
+#include <palacios/vmm_lowlevel.h>
+#include <palacios/svm_msr.h>
+#include <palacios/vmm_rbtree.h>
-extern struct vmm_os_hooks * os_hooks;
-
-extern uint_t cpuid_ecx(uint_t op);
-extern uint_t cpuid_edx(uint_t op);
-extern void Get_MSR(uint_t MSR, uint_t * high_byte, uint_t * low_byte);
-extern void Set_MSR(uint_t MSR, uint_t high_byte, uint_t low_byte);
-extern uint_t launch_svm(vmcb_t * vmcb_addr);
-extern void safe_svm_launch(vmcb_t * vmcb_addr, struct v3_gprs * gprs);
+#include <palacios/vmm_direct_paging.h>
-extern void STGI();
-extern void CLGI();
+#include <palacios/vmm_ctrl_regs.h>
+#include <palacios/svm_io.h>
-extern uint_t Get_CR3();
-
-
-extern void DisableInts();
+#include <palacios/vmm_sprintf.h>
+// This is a global pointer to the host's VMCB
+static addr_t host_vmcbs[CONFIG_MAX_CPUS] = { [0 ... CONFIG_MAX_CPUS - 1] = 0};
+extern void v3_stgi();
+extern void v3_clgi();
+//extern int v3_svm_launch(vmcb_t * vmcb, struct v3_gprs * vm_regs, uint64_t * fs, uint64_t * gs);
+extern int v3_svm_launch(vmcb_t * vmcb, struct v3_gprs * vm_regs, vmcb_t * host_vmcb);
static vmcb_t * Allocate_VMCB() {
- vmcb_t * vmcb_page = (vmcb_t*)os_hooks->allocate_pages(1);
-
+ vmcb_t * vmcb_page = (vmcb_t *)V3_VAddr(V3_AllocPages(1));
- memset(vmcb_page, 0, 4096);
+ memset(vmcb_page, 0, 4096);
- return vmcb_page;
+ return vmcb_page;
}
+static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info * vm_info) {
+ vmcb_ctrl_t * ctrl_area = GET_VMCB_CTRL_AREA(vmcb);
+ vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA(vmcb);
+ uint_t i;
-static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info vm_info) {
- vmcb_ctrl_t * ctrl_area = GET_VMCB_CTRL_AREA(vmcb);
- vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA(vmcb);
- uint_t i;
-
-
- guest_state->rsp = vm_info.vm_regs.rsp;
- // guest_state->rip = vm_info.rip;
- guest_state->rip = 0xfff0;
-
- guest_state->cpl = 0;
-
- //ctrl_area->instrs.instrs.CR0 = 1;
- ctrl_area->cr_reads.cr0 = 1;
- ctrl_area->cr_writes.cr0 = 1;
-
- guest_state->efer |= EFER_MSR_svm_enable;
- guest_state->rflags = 0x00000002; // The reserved bit is always 1
- ctrl_area->svm_instrs.VMRUN = 1;
- ctrl_area->instrs.HLT = 1;
- // guest_state->cr0 = 0x00000001; // PE
- ctrl_area->guest_ASID = 1;
-
- ctrl_area->exceptions.de = 1;
- ctrl_area->exceptions.df = 1;
- ctrl_area->exceptions.pf = 1;
- ctrl_area->exceptions.ts = 1;
- ctrl_area->exceptions.ss = 1;
- ctrl_area->exceptions.ac = 1;
- ctrl_area->exceptions.mc = 1;
- ctrl_area->exceptions.gp = 1;
- ctrl_area->exceptions.ud = 1;
- ctrl_area->exceptions.np = 1;
- ctrl_area->exceptions.of = 1;
- ctrl_area->exceptions.nmi = 1;
-
- vm_info.vm_regs.rdx = 0x00000f00;
+ //
- guest_state->cr0 = 0x60000010;
- guest_state->cs.selector = 0xf000;
- guest_state->cs.limit=0xffff;
- guest_state->cs.base = 0x0000000f0000LL;
- guest_state->cs.attrib.raw = 0xf3;
+ ctrl_area->svm_instrs.VMRUN = 1;
+ ctrl_area->svm_instrs.VMMCALL = 1;
+ ctrl_area->svm_instrs.VMLOAD = 1;
+ ctrl_area->svm_instrs.VMSAVE = 1;
+ ctrl_area->svm_instrs.STGI = 1;
+ ctrl_area->svm_instrs.CLGI = 1;
+ ctrl_area->svm_instrs.SKINIT = 1;
+ ctrl_area->svm_instrs.RDTSCP = 1;
+ ctrl_area->svm_instrs.ICEBP = 1;
+ ctrl_area->svm_instrs.WBINVD = 1;
+ ctrl_area->svm_instrs.MONITOR = 1;
+ ctrl_area->svm_instrs.MWAIT_always = 1;
+ ctrl_area->svm_instrs.MWAIT_if_armed = 1;
+ ctrl_area->instrs.INVLPGA = 1;
+ ctrl_area->instrs.CPUID = 1;
+ ctrl_area->instrs.HLT = 1;
+ // guest_state->cr0 = 0x00000001; // PE
- struct vmcb_selector *segregs [] = {&(guest_state->ss), &(guest_state->ds), &(guest_state->es), &(guest_state->fs), &(guest_state->gs), NULL};
- for ( i = 0; segregs[i] != NULL; i++) {
- struct vmcb_selector * seg = segregs[i];
+ /*
+ ctrl_area->exceptions.de = 1;
+ ctrl_area->exceptions.df = 1;
+
+ ctrl_area->exceptions.ts = 1;
+ ctrl_area->exceptions.ss = 1;
+ ctrl_area->exceptions.ac = 1;
+ ctrl_area->exceptions.mc = 1;
+ ctrl_area->exceptions.gp = 1;
+ ctrl_area->exceptions.ud = 1;
+ ctrl_area->exceptions.np = 1;
+ ctrl_area->exceptions.of = 1;
+
+ ctrl_area->exceptions.nmi = 1;
+ */
- seg->selector = 0x0000;
- // seg->base = seg->selector << 4;
- seg->base = 0x00000000;
- seg->attrib.raw = 0xf3;
- seg->limit = ~0u;
- }
-
- guest_state->gdtr.limit = 0x0000ffff;
- guest_state->gdtr.base = 0x0000000000000000LL;
- guest_state->idtr.limit = 0x0000ffff;
- guest_state->idtr.base = 0x0000000000000000LL;
-
- guest_state->ldtr.selector = 0x0000;
- guest_state->ldtr.limit = 0x0000ffff;
- guest_state->ldtr.base = 0x0000000000000000LL;
- guest_state->tr.selector = 0x0000;
- guest_state->tr.limit = 0x0000ffff;
- guest_state->tr.base = 0x0000000000000000LL;
+ ctrl_area->instrs.NMI = 1;
+ ctrl_area->instrs.SMI = 1;
+ ctrl_area->instrs.INIT = 1;
+ ctrl_area->instrs.PAUSE = 1;
+ ctrl_area->instrs.shutdown_evts = 1;
- guest_state->dr6 = 0x00000000ffff0ff0LL;
- guest_state->dr7 = 0x0000000000000400LL;
- if (vm_info.io_map.num_ports > 0) {
- vmm_io_hook_t * iter;
- addr_t io_port_bitmap;
-
- io_port_bitmap = (addr_t)os_hooks->allocate_pages(3);
- memset((uchar_t*)io_port_bitmap, 0, PAGE_SIZE * 3);
-
- ctrl_area->IOPM_BASE_PA = io_port_bitmap;
+ /* DEBUG FOR RETURN CODE */
+ ctrl_area->exit_code = 1;
- //PrintDebug("Setting up IO Map at 0x%x\n", io_port_bitmap);
- FOREACH_IO_HOOK(vm_info.io_map, iter) {
- ushort_t port = iter->port;
- uchar_t * bitmap = (uchar_t *)io_port_bitmap;
+ /* Setup Guest Machine state */
- bitmap += (port / 8);
- PrintDebug("Setting Bit for port 0x%x\n", port);
- *bitmap |= 1 << (port % 8);
- }
+ vm_info->vm_regs.rsp = 0x00;
+ vm_info->rip = 0xfff0;
+ vm_info->vm_regs.rdx = 0x00000f00;
- //PrintDebugMemDump((uchar_t*)io_port_bitmap, PAGE_SIZE *2);
- ctrl_area->instrs.IOIO_PROT = 1;
- }
+ vm_info->cpl = 0;
+ vm_info->ctrl_regs.rflags = 0x00000002; // The reserved bit is always 1
+ vm_info->ctrl_regs.cr0 = 0x60010010; // Set the WP flag so the memory hooks work in real-mode
+ vm_info->ctrl_regs.efer |= EFER_MSR_svm_enable;
- PrintDebug("Exiting on interrupts\n");
- ctrl_area->guest_ctrl.V_INTR_MASKING = 1;
- ctrl_area->instrs.INTR = 1;
- if (vm_info.shdw_pg_mode == SHADOW_PAGING) {
- PrintDebug("Creating initial shadow page table\n");
- vm_info.shdw_pg_state.shadow_cr3 |= ((addr_t)create_passthrough_pde32_pts(&vm_info) & ~0xfff);
- PrintDebug("Created\n");
- guest_state->cr3 = vm_info.shdw_pg_state.shadow_cr3;
+ vm_info->segments.cs.selector = 0xf000;
+ vm_info->segments.cs.limit = 0xffff;
+ vm_info->segments.cs.base = 0x0000000f0000LL;
- //PrintDebugPageTables((pde32_t*)(vm_info.shdw_pg_state.shadow_cr3.e_reg.low));
+ // (raw attributes = 0xf3)
+ vm_info->segments.cs.type = 0x3;
+ vm_info->segments.cs.system = 0x1;
+ vm_info->segments.cs.dpl = 0x3;
+ vm_info->segments.cs.present = 1;
- ctrl_area->cr_reads.cr3 = 1;
- ctrl_area->cr_writes.cr3 = 1;
- ctrl_area->instrs.INVLPG = 1;
- ctrl_area->instrs.INVLPGA = 1;
+ struct v3_segment * segregs [] = {&(vm_info->segments.ss), &(vm_info->segments.ds),
+ &(vm_info->segments.es), &(vm_info->segments.fs),
+ &(vm_info->segments.gs), NULL};
- /* JRL: This is a performance killer, and a simplistic solution */
- /* We need to fix this */
- ctrl_area->TLB_CONTROL = 1;
-
+ for ( i = 0; segregs[i] != NULL; i++) {
+ struct v3_segment * seg = segregs[i];
+
+ seg->selector = 0x0000;
+ // seg->base = seg->selector << 4;
+ seg->base = 0x00000000;
+ seg->limit = ~0u;
+ // (raw attributes = 0xf3)
+ seg->type = 0x3;
+ seg->system = 0x1;
+ seg->dpl = 0x3;
+ seg->present = 1;
+ }
- guest_state->g_pat = 0x7040600070406ULL;
+ vm_info->segments.gdtr.limit = 0x0000ffff;
+ vm_info->segments.gdtr.base = 0x0000000000000000LL;
+ vm_info->segments.idtr.limit = 0x0000ffff;
+ vm_info->segments.idtr.base = 0x0000000000000000LL;
- guest_state->cr0 |= 0x80000000;
+ vm_info->segments.ldtr.selector = 0x0000;
+ vm_info->segments.ldtr.limit = 0x0000ffff;
+ vm_info->segments.ldtr.base = 0x0000000000000000LL;
+ vm_info->segments.tr.selector = 0x0000;
+ vm_info->segments.tr.limit = 0x0000ffff;
+ vm_info->segments.tr.base = 0x0000000000000000LL;
- } else if (vm_info.shdw_pg_mode == NESTED_PAGING) {
- // Flush the TLB on entries/exits
+ vm_info->dbg_regs.dr6 = 0x00000000ffff0ff0LL;
+ vm_info->dbg_regs.dr7 = 0x0000000000000400LL;
- // Enable Nested Paging
- //ctrl_area->NP_ENABLE = 1;
- //PrintDebug("NP_Enable at 0x%x\n", &(ctrl_area->NP_ENABLE));
+ v3_init_svm_io_map(vm_info);
+ ctrl_area->IOPM_BASE_PA = (addr_t)V3_PAddr(vm_info->io_map.arch_data);
+ ctrl_area->instrs.IOIO_PROT = 1;
- // Set the Nested Page Table pointer
- // ctrl_area->N_CR3 = ((addr_t)vm_info.page_tables);
- // ctrl_area->N_CR3 = (addr_t)(vm_info.page_tables);
- // ctrl_area->N_CR3 = Get_CR3();
- // guest_state->cr3 |= (Get_CR3() & 0xfffff000);
+ v3_init_svm_msr_map(vm_info);
+ ctrl_area->MSRPM_BASE_PA = (addr_t)V3_PAddr(vm_info->msr_map.arch_data);
+ ctrl_area->instrs.MSR_PROT = 1;
- // guest_state->g_pat = 0x7040600070406ULL;
- }
+ PrintDebug("Exiting on interrupts\n");
+ ctrl_area->guest_ctrl.V_INTR_MASKING = 1;
+ ctrl_area->instrs.INTR = 1;
-}
+ if (vm_info->shdw_pg_mode == SHADOW_PAGING) {
+ PrintDebug("Creating initial shadow page table\n");
+
+ /* JRL: This is a performance killer, and a simplistic solution */
+ /* We need to fix this */
+ ctrl_area->TLB_CONTROL = 1;
+ ctrl_area->guest_ASID = 1;
+
+
+ if (v3_init_passthrough_pts(vm_info) == -1) {
+ PrintError("Could not initialize passthrough page tables\n");
+ return ;
+ }
+ vm_info->shdw_pg_state.guest_cr0 = 0x0000000000000010LL;
+ PrintDebug("Created\n");
+
+ vm_info->ctrl_regs.cr0 |= 0x80000000;
+ vm_info->ctrl_regs.cr3 = vm_info->direct_map_pt;
+ ctrl_area->cr_reads.cr0 = 1;
+ ctrl_area->cr_writes.cr0 = 1;
+ //ctrl_area->cr_reads.cr4 = 1;
+ ctrl_area->cr_writes.cr4 = 1;
+ ctrl_area->cr_reads.cr3 = 1;
+ ctrl_area->cr_writes.cr3 = 1;
+ v3_hook_msr(vm_info, EFER_MSR,
+ &v3_handle_efer_read,
+ &v3_handle_efer_write,
+ vm_info);
+ ctrl_area->instrs.INVLPG = 1;
+ ctrl_area->exceptions.pf = 1;
+ guest_state->g_pat = 0x7040600070406ULL;
-static int init_svm_guest(struct guest_info *info) {
-
- PrintDebug("Allocating VMCB\n");
- info->vmm_data = (void*)Allocate_VMCB();
+ } else if (vm_info->shdw_pg_mode == NESTED_PAGING) {
+ // Flush the TLB on entries/exits
+ ctrl_area->TLB_CONTROL = 1;
+ ctrl_area->guest_ASID = 1;
- //PrintDebug("Generating Guest nested page tables\n");
- // info->page_tables = NULL;
- //info->page_tables = generate_guest_page_tables_64(&(info->mem_layout), &(info->mem_list));
- //info->page_tables = generate_guest_page_tables(&(info->mem_layout), &(info->mem_list));
- // PrintDebugPageTables(info->page_tables);
+ // Enable Nested Paging
+ ctrl_area->NP_ENABLE = 1;
+ PrintDebug("NP_Enable at 0x%p\n", (void *)&(ctrl_area->NP_ENABLE));
- PrintDebug("Initializing VMCB (addr=%x)\n", info->vmm_data);
- Init_VMCB_BIOS((vmcb_t*)(info->vmm_data), *info);
-
+ // Set the Nested Page Table pointer
+ if (v3_init_passthrough_pts(vm_info) == -1) {
+ PrintError("Could not initialize Nested page tables\n");
+ return ;
+ }
- // info->rip = 0;
+ ctrl_area->N_CR3 = vm_info->direct_map_pt;
- info->vm_regs.rdi = 0;
- info->vm_regs.rsi = 0;
- info->vm_regs.rbp = 0;
- info->vm_regs.rsp = 0;
- info->vm_regs.rbx = 0;
- info->vm_regs.rdx = 0;
- info->vm_regs.rcx = 0;
- info->vm_regs.rax = 0;
-
- return 0;
+ guest_state->g_pat = 0x7040600070406ULL;
+ }
}
-// can we start a kernel thread here...
-static int start_svm_guest(struct guest_info *info) {
- vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA((vmcb_t*)(info->vmm_data));
- vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
+int v3_init_svm_vmcb(struct guest_info * info, v3_vm_class_t vm_class) {
- PrintDebug("Launching SVM VM (vmcb=%x)\n", info->vmm_data);
- //PrintDebugVMCB((vmcb_t*)(info->vmm_data));
+ PrintDebug("Allocating VMCB\n");
+ info->vmm_data = (void*)Allocate_VMCB();
+
+ if (vm_class == V3_PC_VM) {
+ PrintDebug("Initializing VMCB (addr=%p)\n", (void *)info->vmm_data);
+ Init_VMCB_BIOS((vmcb_t*)(info->vmm_data), info);
+ } else {
+ PrintError("Invalid VM class\n");
+ return -1;
+ }
- while (1) {
- ullong_t tmp_tsc;
+ return 0;
+}
- CLGI();
- PrintDebug("SVM Entry to rip=%x...\n", info->rip);
+static int update_irq_exit_state(struct guest_info * info) {
+ vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
- rdtscll(info->time_state.cached_host_tsc);
- guest_ctrl->TSC_OFFSET = info->time_state.guest_tsc - info->time_state.cached_host_tsc;
+ if ((info->intr_state.irq_pending == 1) && (guest_ctrl->guest_ctrl.V_IRQ == 0)) {
+
+#ifdef CONFIG_DEBUG_INTERRUPTS
+ PrintDebug("INTAK cycle completed for irq %d\n", info->intr_state.irq_vector);
+#endif
- safe_svm_launch((vmcb_t*)(info->vmm_data), &(info->vm_regs));
+ info->intr_state.irq_started = 1;
+ info->intr_state.irq_pending = 0;
- rdtscll(tmp_tsc);
- //PrintDebug("SVM Returned\n");
+ v3_injecting_intr(info, info->intr_state.irq_vector, V3_EXTERNAL_IRQ);
+ }
+ if ((info->intr_state.irq_started == 1) && (guest_ctrl->exit_int_info.valid == 0)) {
+#ifdef CONFIG_DEBUG_INTERRUPTS
+ PrintDebug("Interrupt %d taken by guest\n", info->intr_state.irq_vector);
+#endif
- v3_update_time(info, tmp_tsc - info->time_state.cached_host_tsc);
+ // Interrupt was taken fully vectored
+ info->intr_state.irq_started = 0;
- STGI();
+ } else {
+#ifdef CONFIG_DEBUG_INTERRUPTS
+ PrintDebug("EXIT INT INFO is set (vec=%d)\n", guest_ctrl->exit_int_info.vector);
+#endif
+ }
-
- if (handle_svm_exit(info) != 0) {
+ return 0;
+}
- addr_t host_addr;
- addr_t linear_addr = 0;
- PrintDebug("SVM ERROR!!\n");
-
- PrintDebug("RIP: %x\n", guest_state->rip);
+static int update_irq_entry_state(struct guest_info * info) {
+ vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
+ if (v3_excp_pending(info)) {
+ uint_t excp = v3_get_excp_number(info);
+
+ guest_ctrl->EVENTINJ.type = SVM_INJECTION_EXCEPTION;
+
+ if (info->excp_state.excp_error_code_valid) {
+ guest_ctrl->EVENTINJ.error_code = info->excp_state.excp_error_code;
+ guest_ctrl->EVENTINJ.ev = 1;
+#ifdef CONFIG_DEBUG_INTERRUPTS
+ PrintDebug("Injecting exception %d with error code %x\n", excp, guest_ctrl->EVENTINJ.error_code);
+#endif
+ }
+
+ guest_ctrl->EVENTINJ.vector = excp;
+
+ guest_ctrl->EVENTINJ.valid = 1;
+
+#ifdef CONFIG_DEBUG_INTERRUPTS
+ PrintDebug("<%d> Injecting Exception %d (CR2=%p) (EIP=%p)\n",
+ (int)info->num_exits,
+ guest_ctrl->EVENTINJ.vector,
+ (void *)(addr_t)info->ctrl_regs.cr2,
+ (void *)(addr_t)info->rip);
+#endif
- linear_addr = get_addr_linear(info, guest_state->rip, &(info->segments.cs));
+ v3_injecting_excp(info, excp);
+ } else if (info->intr_state.irq_started == 1) {
+#ifdef CONFIG_DEBUG_INTERRUPTS
+ PrintDebug("IRQ pending from previous injection\n");
+#endif
+ guest_ctrl->guest_ctrl.V_IRQ = 1;
+ guest_ctrl->guest_ctrl.V_INTR_VECTOR = info->intr_state.irq_vector;
+ guest_ctrl->guest_ctrl.V_IGN_TPR = 1;
+ guest_ctrl->guest_ctrl.V_INTR_PRIO = 0xf;
+
+ } else {
+ switch (v3_intr_pending(info)) {
+ case V3_EXTERNAL_IRQ: {
+ uint32_t irq = v3_get_intr(info);
+
+ guest_ctrl->guest_ctrl.V_IRQ = 1;
+ guest_ctrl->guest_ctrl.V_INTR_VECTOR = irq;
+ guest_ctrl->guest_ctrl.V_IGN_TPR = 1;
+ guest_ctrl->guest_ctrl.V_INTR_PRIO = 0xf;
+
+#ifdef CONFIG_DEBUG_INTERRUPTS
+ PrintDebug("Injecting Interrupt %d (EIP=%p)\n",
+ guest_ctrl->guest_ctrl.V_INTR_VECTOR,
+ (void *)(addr_t)info->rip);
+#endif
+ info->intr_state.irq_pending = 1;
+ info->intr_state.irq_vector = irq;
+
+ break;
+ }
+ case V3_NMI:
+ guest_ctrl->EVENTINJ.type = SVM_INJECTION_NMI;
+ break;
+ case V3_SOFTWARE_INTR:
+ guest_ctrl->EVENTINJ.type = SVM_INJECTION_SOFT_INTR;
+ break;
+ case V3_VIRTUAL_IRQ:
+ guest_ctrl->EVENTINJ.type = SVM_INJECTION_IRQ;
+ break;
+
+ case V3_INVALID_INTR:
+ default:
+ break;
+ }
+
+ }
- PrintDebug("RIP Linear: %x\n", linear_addr);
+ return 0;
+}
-
- if (info->mem_mode == PHYSICAL_MEM) {
- guest_pa_to_host_pa(info, linear_addr, &host_addr);
- } else if (info->mem_mode == VIRTUAL_MEM) {
- guest_va_to_host_pa(info, linear_addr, &host_addr);
- }
+/*
+ * CAUTION and DANGER!!!
+ *
+ * The VMCB CANNOT(!!) be accessed outside of the clgi/stgi calls inside this function
+ * When exectuing a symbiotic call, the VMCB WILL be overwritten, so any dependencies
+ * on its contents will cause things to break. The contents at the time of the exit WILL
+ * change before the exit handler is executed.
+ */
+int v3_svm_enter(struct guest_info * info) {
+ vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
+ vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA((vmcb_t*)(info->vmm_data));
+ ullong_t tmp_tsc;
+ addr_t exit_code = 0, exit_info1 = 0, exit_info2 = 0;
+
+ // Conditionally yield the CPU if the timeslice has expired
+ v3_yield_cond(info);
+
+ // disable global interrupts for vm state transition
+ v3_clgi();
+
+ // Synchronize the guest state to the VMCB
+ guest_state->cr0 = info->ctrl_regs.cr0;
+ guest_state->cr2 = info->ctrl_regs.cr2;
+ guest_state->cr3 = info->ctrl_regs.cr3;
+ guest_state->cr4 = info->ctrl_regs.cr4;
+ guest_state->dr6 = info->dbg_regs.dr6;
+ guest_state->dr7 = info->dbg_regs.dr7;
+ guest_ctrl->guest_ctrl.V_TPR = info->ctrl_regs.cr8 & 0xff;
+ guest_state->rflags = info->ctrl_regs.rflags;
+ guest_state->efer = info->ctrl_regs.efer;
+
+ guest_state->cpl = info->cpl;
- PrintDebug("Host Address of rip = 0x%x\n", host_addr);
+ v3_set_vmcb_segments((vmcb_t*)(info->vmm_data), &(info->segments));
- PrintDebug("Instr (15 bytes) at %x:\n", host_addr);
- PrintTraceMemDump((char*)host_addr, 15);
+ guest_state->rax = info->vm_regs.rax;
+ guest_state->rip = info->rip;
+ guest_state->rsp = info->vm_regs.rsp;
- break;
+#ifdef CONFIG_SYMBIOTIC
+ if (info->sym_state.sym_call_active == 0) {
+ update_irq_entry_state(info);
}
- }
- return 0;
-}
-
-
-
-
-/* Checks machine SVM capability */
-/* Implemented from: AMD Arch Manual 3, sect 15.4 */
-int is_svm_capable() {
- uint_t ret = cpuid_ecx(CPUID_FEATURE_IDS);
- uint_t vm_cr_low = 0, vm_cr_high = 0;
+#else
+ update_irq_entry_state(info);
+#endif
- if ((ret & CPUID_FEATURE_IDS_ecx_svm_avail) == 0) {
- PrintDebug("SVM Not Available\n");
- return 0;
- }
+ /* ** */
- Get_MSR(SVM_VM_CR_MSR, &vm_cr_high, &vm_cr_low);
+ /*
+ PrintDebug("SVM Entry to CS=%p rip=%p...\n",
+ (void *)(addr_t)info->segments.cs.base,
+ (void *)(addr_t)info->rip);
+ */
+
+#ifdef CONFIG_SYMBIOTIC
+ if (info->sym_state.sym_call_active == 1) {
+ if (guest_ctrl->guest_ctrl.V_IRQ == 1) {
+ V3_Print("!!! Injecting Interrupt during Sym call !!!\n");
+ }
+ }
+#endif
- if ((ret & CPUID_SVM_REV_AND_FEATURE_IDS_edx_np) == 1) {
- PrintDebug("Nested Paging not supported\n");
- }
- if ((vm_cr_low & SVM_VM_CR_MSR_svmdis) == 0) {
- return 1;
- }
+ rdtscll(info->time_state.cached_host_tsc);
+ // guest_ctrl->TSC_OFFSET = info->time_state.guest_tsc - info->time_state.cached_host_tsc;
+
+ v3_svm_launch((vmcb_t *)V3_PAddr(info->vmm_data), &(info->vm_regs), (vmcb_t *)host_vmcbs[info->cpu_id]);
+
+ rdtscll(tmp_tsc);
- ret = cpuid_edx(CPUID_SVM_REV_AND_FEATURE_IDS);
+ //PrintDebug("SVM Returned\n");
+
+ info->num_exits++;
- if ((ret & CPUID_SVM_REV_AND_FEATURE_IDS_edx_svml) == 0) {
- PrintDebug("SVM BIOS Disabled, not unlockable\n");
- } else {
- PrintDebug("SVM is locked with a key\n");
- }
+ v3_update_time(info, tmp_tsc - info->time_state.cached_host_tsc);
- return 0;
-}
+ // Save Guest state from VMCB
+ info->rip = guest_state->rip;
+ info->vm_regs.rsp = guest_state->rsp;
+ info->vm_regs.rax = guest_state->rax;
+ info->cpl = guest_state->cpl;
-void Init_SVM(struct vmm_ctrl_ops * vmm_ops) {
- reg_ex_t msr;
- void * host_state;
+ info->ctrl_regs.cr0 = guest_state->cr0;
+ info->ctrl_regs.cr2 = guest_state->cr2;
+ info->ctrl_regs.cr3 = guest_state->cr3;
+ info->ctrl_regs.cr4 = guest_state->cr4;
+ info->dbg_regs.dr6 = guest_state->dr6;
+ info->dbg_regs.dr7 = guest_state->dr7;
+ info->ctrl_regs.cr8 = guest_ctrl->guest_ctrl.V_TPR;
+ info->ctrl_regs.rflags = guest_state->rflags;
+ info->ctrl_regs.efer = guest_state->efer;
+
+ v3_get_vmcb_segments((vmcb_t*)(info->vmm_data), &(info->segments));
+ info->cpu_mode = v3_get_vm_cpu_mode(info);
+ info->mem_mode = v3_get_vm_mem_mode(info);
+ /* ** */
- // Enable SVM on the CPU
- Get_MSR(EFER_MSR, &(msr.e_reg.high), &(msr.e_reg.low));
- msr.e_reg.low |= EFER_MSR_svm_enable;
- Set_MSR(EFER_MSR, 0, msr.e_reg.low);
-
- PrintDebug("SVM Enabled\n");
+ // save exit info here
+ exit_code = guest_ctrl->exit_code;
+ exit_info1 = guest_ctrl->exit_info1;
+ exit_info2 = guest_ctrl->exit_info2;
- // Setup the host state save area
- host_state = os_hooks->allocate_pages(4);
-
- msr.e_reg.high = 0;
- msr.e_reg.low = (uint_t)host_state;
+#ifdef CONFIG_SYMBIOTIC
+ if (info->sym_state.sym_call_active == 0) {
+ update_irq_exit_state(info);
+ }
+#else
+ update_irq_exit_state(info);
+#endif
- PrintDebug("Host State being saved at %x\n", (uint_t)host_state);
- Set_MSR(SVM_VM_HSAVE_PA_MSR, msr.e_reg.high, msr.e_reg.low);
+ // reenable global interrupts after vm exit
+ v3_stgi();
+
+ // Conditionally yield the CPU if the timeslice has expired
+ v3_yield_cond(info);
- // Setup the SVM specific vmm operations
- vmm_ops->init_guest = &init_svm_guest;
- vmm_ops->start_guest = &start_svm_guest;
+ if (v3_handle_svm_exit(info, exit_code, exit_info1, exit_info2) != 0) {
+ PrintError("Error in SVM exit handler\n");
+ return -1;
+ }
- return;
+ return 0;
}
+int v3_start_svm_guest(struct guest_info *info) {
+ // vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA((vmcb_t*)(info->vmm_data));
+ // vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
+ PrintDebug("Launching SVM VM (vmcb=%p)\n", (void *)info->vmm_data);
+ //PrintDebugVMCB((vmcb_t*)(info->vmm_data));
+
+ info->run_state = VM_RUNNING;
+ rdtscll(info->yield_start_cycle);
+
+
+ while (1) {
+ if (v3_svm_enter(info) == -1) {
+ vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
+ addr_t host_addr;
+ addr_t linear_addr = 0;
+
+ info->run_state = VM_ERROR;
+
+ V3_Print("SVM ERROR!!\n");
+
+ v3_print_guest_state(info);
+
+ V3_Print("SVM Exit Code: %p\n", (void *)(addr_t)guest_ctrl->exit_code);
+
+ V3_Print("exit_info1 low = 0x%.8x\n", *(uint_t*)&(guest_ctrl->exit_info1));
+ V3_Print("exit_info1 high = 0x%.8x\n", *(uint_t *)(((uchar_t *)&(guest_ctrl->exit_info1)) + 4));
+
+ V3_Print("exit_info2 low = 0x%.8x\n", *(uint_t*)&(guest_ctrl->exit_info2));
+ V3_Print("exit_info2 high = 0x%.8x\n", *(uint_t *)(((uchar_t *)&(guest_ctrl->exit_info2)) + 4));
+
+ linear_addr = get_addr_linear(info, info->rip, &(info->segments.cs));
+
+ if (info->mem_mode == PHYSICAL_MEM) {
+ guest_pa_to_host_va(info, linear_addr, &host_addr);
+ } else if (info->mem_mode == VIRTUAL_MEM) {
+ guest_va_to_host_va(info, linear_addr, &host_addr);
+ }
+
+ V3_Print("Host Address of rip = 0x%p\n", (void *)host_addr);
+
+ V3_Print("Instr (15 bytes) at %p:\n", (void *)host_addr);
+ v3_dump_mem((uint8_t *)host_addr, 15);
+
+ v3_print_stack(info);
+
+ break;
+ }
+
+ if ((info->num_exits % 5000) == 0) {
+ V3_Print("SVM Exit number %d\n", (uint32_t)info->num_exits);
+ }
+
+
+
+ }
+ return 0;
+}
+/* Checks machine SVM capability */
+/* Implemented from: AMD Arch Manual 3, sect 15.4 */
+int v3_is_svm_capable() {
+ uint_t vm_cr_low = 0, vm_cr_high = 0;
+ uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0;
+ v3_cpuid(CPUID_EXT_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
+
+ PrintDebug("CPUID_EXT_FEATURE_IDS_ecx=0x%x\n", ecx);
+
+ if ((ecx & CPUID_EXT_FEATURE_IDS_ecx_svm_avail) == 0) {
+ V3_Print("SVM Not Available\n");
+ return 0;
+ } else {
+ v3_get_msr(SVM_VM_CR_MSR, &vm_cr_high, &vm_cr_low);
+
+ PrintDebug("SVM_VM_CR_MSR = 0x%x 0x%x\n", vm_cr_high, vm_cr_low);
+
+ if ((vm_cr_low & SVM_VM_CR_MSR_svmdis) == 1) {
+ V3_Print("SVM is available but is disabled.\n");
+
+ v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
+
+ PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_edx=0x%x\n", edx);
+
+ if ((edx & CPUID_SVM_REV_AND_FEATURE_IDS_edx_svml) == 0) {
+ V3_Print("SVM BIOS Disabled, not unlockable\n");
+ } else {
+ V3_Print("SVM is locked with a key\n");
+ }
+ return 0;
+
+ } else {
+ V3_Print("SVM is available and enabled.\n");
+
+ v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
+ PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_eax=0x%x\n", eax);
+ PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_ebx=0x%x\n", ebx);
+ PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_ecx=0x%x\n", ecx);
+ PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_edx=0x%x\n", edx);
+
+ return 1;
+ }
+ }
+}
+static int has_svm_nested_paging() {
+ uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0;
+ v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
+ //PrintDebug("CPUID_EXT_FEATURE_IDS_edx=0x%x\n", edx);
+ if ((edx & CPUID_SVM_REV_AND_FEATURE_IDS_edx_np) == 0) {
+ V3_Print("SVM Nested Paging not supported\n");
+ return 0;
+ } else {
+ V3_Print("SVM Nested Paging supported\n");
+ return 1;
+ }
+}
+void v3_init_svm_cpu(int cpu_id) {
+ reg_ex_t msr;
+ extern v3_cpu_arch_t v3_cpu_types[];
+ // Enable SVM on the CPU
+ v3_get_msr(EFER_MSR, &(msr.e_reg.high), &(msr.e_reg.low));
+ msr.e_reg.low |= EFER_MSR_svm_enable;
+ v3_set_msr(EFER_MSR, 0, msr.e_reg.low);
+ V3_Print("SVM Enabled\n");
+ // Setup the host state save area
+ host_vmcbs[cpu_id] = (addr_t)V3_AllocPages(4);
+ /* 64-BIT-ISSUE */
+ // msr.e_reg.high = 0;
+ //msr.e_reg.low = (uint_t)host_vmcb;
+ msr.r_reg = host_vmcbs[cpu_id];
+ PrintDebug("Host State being saved at %p\n", (void *)host_vmcbs[cpu_id]);
+ v3_set_msr(SVM_VM_HSAVE_PA_MSR, msr.e_reg.high, msr.e_reg.low);
+ if (has_svm_nested_paging() == 1) {
+ v3_cpu_types[cpu_id] = V3_SVM_REV3_CPU;
+ } else {
+ v3_cpu_types[cpu_id] = V3_SVM_CPU;
+ }
+}
-/*static void Init_VMCB(vmcb_t * vmcb, struct guest_info vm_info) {
- vmcb_ctrl_t * ctrl_area = GET_VMCB_CTRL_AREA(vmcb);
- vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA(vmcb);
- uint_t i;
- guest_state->rsp = vm_info.vm_regs.rsp;
- guest_state->rip = vm_info.rip;
- //ctrl_area->instrs.instrs.CR0 = 1;
- ctrl_area->cr_reads.cr0 = 1;
- ctrl_area->cr_writes.cr0 = 1;
- guest_state->efer |= EFER_MSR_svm_enable;
- guest_state->rflags = 0x00000002; // The reserved bit is always 1
- ctrl_area->svm_instrs.VMRUN = 1;
- // guest_state->cr0 = 0x00000001; // PE
- ctrl_area->guest_ASID = 1;
- ctrl_area->exceptions.de = 1;
- ctrl_area->exceptions.df = 1;
- ctrl_area->exceptions.pf = 1;
- ctrl_area->exceptions.ts = 1;
- ctrl_area->exceptions.ss = 1;
- ctrl_area->exceptions.ac = 1;
- ctrl_area->exceptions.mc = 1;
- ctrl_area->exceptions.gp = 1;
- ctrl_area->exceptions.ud = 1;
- ctrl_area->exceptions.np = 1;
- ctrl_area->exceptions.of = 1;
- ctrl_area->exceptions.nmi = 1;
- guest_state->cs.selector = 0x0000;
- guest_state->cs.limit=~0u;
- guest_state->cs.base = guest_state->cs.selector<<4;
- guest_state->cs.attrib.raw = 0xf3;
-
- struct vmcb_selector *segregs [] = {&(guest_state->ss), &(guest_state->ds), &(guest_state->es), &(guest_state->fs), &(guest_state->gs), NULL};
- for ( i = 0; segregs[i] != NULL; i++) {
- struct vmcb_selector * seg = segregs[i];
-
- seg->selector = 0x0000;
- seg->base = seg->selector << 4;
- seg->attrib.raw = 0xf3;
- seg->limit = ~0u;
- }
-
- if (vm_info.io_map.num_ports > 0) {
- vmm_io_hook_t * iter;
- addr_t io_port_bitmap;
-
- io_port_bitmap = (addr_t)os_hooks->allocate_pages(3);
- memset((uchar_t*)io_port_bitmap, 0, PAGE_SIZE * 3);
-
- ctrl_area->IOPM_BASE_PA = io_port_bitmap;
- //PrintDebug("Setting up IO Map at 0x%x\n", io_port_bitmap);
- FOREACH_IO_HOOK(vm_info.io_map, iter) {
- ushort_t port = iter->port;
- uchar_t * bitmap = (uchar_t *)io_port_bitmap;
- bitmap += (port / 8);
- PrintDebug("Setting Bit in block %x\n", bitmap);
- *bitmap |= 1 << (port % 8);
- }
- //PrintDebugMemDump((uchar_t*)io_port_bitmap, PAGE_SIZE *2);
-
- ctrl_area->instrs.IOIO_PROT = 1;
- }
-
- ctrl_area->instrs.INTR = 1;
- if (vm_info.page_mode == SHADOW_PAGING) {
- PrintDebug("Creating initial shadow page table\n");
- vm_info.shdw_pg_state.shadow_cr3 |= ((addr_t)create_passthrough_pde32_pts(&vm_info) & ~0xfff);
- PrintDebug("Created\n");
- guest_state->cr3 = vm_info.shdw_pg_state.shadow_cr3;
- ctrl_area->cr_reads.cr3 = 1;
- ctrl_area->cr_writes.cr3 = 1;
- ctrl_area->instrs.INVLPG = 1;
- ctrl_area->instrs.INVLPGA = 1;
-
- guest_state->g_pat = 0x7040600070406ULL;
- guest_state->cr0 |= 0x80000000;
- } else if (vm_info.page_mode == NESTED_PAGING) {
- // Flush the TLB on entries/exits
- //ctrl_area->TLB_CONTROL = 1;
- // Enable Nested Paging
- //ctrl_area->NP_ENABLE = 1;
- //PrintDebug("NP_Enable at 0x%x\n", &(ctrl_area->NP_ENABLE));
- // Set the Nested Page Table pointer
- // ctrl_area->N_CR3 = ((addr_t)vm_info.page_tables);
- // ctrl_area->N_CR3 = (addr_t)(vm_info.page_tables);
-
- // ctrl_area->N_CR3 = Get_CR3();
- // guest_state->cr3 |= (Get_CR3() & 0xfffff000);
-
- // guest_state->g_pat = 0x7040600070406ULL;
- }
+#if 0
+/*
+ * Test VMSAVE/VMLOAD Latency
+ */
+#define vmsave ".byte 0x0F,0x01,0xDB ; "
+#define vmload ".byte 0x0F,0x01,0xDA ; "
+{
+ uint32_t start_lo, start_hi;
+ uint32_t end_lo, end_hi;
+ uint64_t start, end;
+
+ __asm__ __volatile__ (
+ "rdtsc ; "
+ "movl %%eax, %%esi ; "
+ "movl %%edx, %%edi ; "
+ "movq %%rcx, %%rax ; "
+ vmsave
+ "rdtsc ; "
+ : "=D"(start_hi), "=S"(start_lo), "=a"(end_lo),"=d"(end_hi)
+ : "c"(host_vmcb[cpu_id]), "0"(0), "1"(0), "2"(0), "3"(0)
+ );
+
+ start = start_hi;
+ start <<= 32;
+ start += start_lo;
+
+ end = end_hi;
+ end <<= 32;
+ end += end_lo;
+
+ PrintDebug("VMSave Cycle Latency: %d\n", (uint32_t)(end - start));
+
+ __asm__ __volatile__ (
+ "rdtsc ; "
+ "movl %%eax, %%esi ; "
+ "movl %%edx, %%edi ; "
+ "movq %%rcx, %%rax ; "
+ vmload
+ "rdtsc ; "
+ : "=D"(start_hi), "=S"(start_lo), "=a"(end_lo),"=d"(end_hi)
+ : "c"(host_vmcb[cpu_id]), "0"(0), "1"(0), "2"(0), "3"(0)
+ );
+
+ start = start_hi;
+ start <<= 32;
+ start += start_lo;
+
+ end = end_hi;
+ end <<= 32;
+ end += end_lo;
+
+
+ PrintDebug("VMLoad Cycle Latency: %d\n", (uint32_t)(end - start));
+ }
+ /* End Latency Test */
-}
-*/
+#endif
ctrl_area->instrs.IOIO_PROT = 1;
- ctrl_area->IOPM_BASE_PA = (uint_t)os_hooks->allocate_pages(3);
+ ctrl_area->IOPM_BASE_PA = (uint_t)V3_AllocPages(3);
{
reg_ex_t tmp_reg;
