#include <palacios/vmx_io.h>
#include <palacios/vmx_msr.h>
+#include <palacios/vmx_hw_info.h>
#ifndef CONFIG_DEBUG_VMX
#undef PrintDebug
#endif
-static addr_t host_vmcs_ptrs[CONFIG_MAX_CPUS] = { [0 ... CONFIG_MAX_CPUS - 1] = 0};
+/* These fields contain the hardware feature sets supported by the local CPU */
+static struct vmx_hw_info hw_info;
+static addr_t active_vmcs_ptrs[CONFIG_MAX_CPUS] = { [0 ... CONFIG_MAX_CPUS - 1] = 0};
+static addr_t host_vmcs_ptrs[CONFIG_MAX_CPUS] = { [0 ... CONFIG_MAX_CPUS - 1] = 0};
extern int v3_vmx_launch(struct v3_gprs * vm_regs, struct guest_info * info, struct v3_ctrl_regs * ctrl_regs);
extern int v3_vmx_resume(struct v3_gprs * vm_regs, struct guest_info * info, struct v3_ctrl_regs * ctrl_regs);
static int inline check_vmcs_write(vmcs_field_t field, addr_t val) {
int ret = 0;
- ret = vmcs_write(field,val);
+ ret = vmcs_write(field, val);
if (ret != VMX_SUCCESS) {
PrintError("VMWRITE error on %s!: %d\n", v3_vmcs_field_to_str(field), ret);
return ret;
}
-#if 0
-// For the 32 bit reserved bit fields
-// MB1s are in the low 32 bits, MBZs are in the high 32 bits of the MSR
-static uint32_t sanitize_bits1(uint32_t msr_num, uint32_t val) {
- v3_msr_t mask_msr;
-
- PrintDebug("sanitize_bits1 (MSR:%x)\n", msr_num);
-
- v3_get_msr(msr_num, &mask_msr.hi, &mask_msr.lo);
-
- PrintDebug("MSR %x = %x : %x \n", msr_num, mask_msr.hi, mask_msr.lo);
-
- val |= mask_msr.lo;
- val |= mask_msr.hi;
-
- return val;
-}
-
-
-
-static addr_t sanitize_bits2(uint32_t msr_num0, uint32_t msr_num1, addr_t val) {
- v3_msr_t msr0, msr1;
- addr_t msr0_val, msr1_val;
-
- PrintDebug("sanitize_bits2 (MSR0=%x, MSR1=%x)\n", msr_num0, msr_num1);
-
- v3_get_msr(msr_num0, &msr0.hi, &msr0.lo);
- v3_get_msr(msr_num1, &msr1.hi, &msr1.lo);
-
- // This generates a mask that is the natural bit width of the CPU
- msr0_val = msr0.value;
- msr1_val = msr1.value;
-
- PrintDebug("MSR %x = %p, %x = %p \n", msr_num0, (void*)msr0_val, msr_num1, (void*)msr1_val);
-
- val |= msr0_val;
- val |= msr1_val;
-
- return val;
-}
-
-
-#endif
static addr_t allocate_vmcs() {
- reg_ex_t msr;
struct vmcs_data * vmcs_page = NULL;
PrintDebug("Allocating page\n");
vmcs_page = (struct vmcs_data *)V3_VAddr(V3_AllocPages(1));
memset(vmcs_page, 0, 4096);
- v3_get_msr(VMX_BASIC_MSR, &(msr.e_reg.high), &(msr.e_reg.low));
-
- vmcs_page->revision = ((struct vmx_basic_msr*)&msr)->revision;
- PrintDebug("VMX Revision: 0x%x\n",vmcs_page->revision);
+ vmcs_page->revision = hw_info.basic_info.revision;
+ PrintDebug("VMX Revision: 0x%x\n", vmcs_page->revision);
return (addr_t)V3_PAddr((void *)vmcs_page);
}
static int init_vmcs_bios(struct guest_info * info, struct vmx_data * vmx_state) {
int vmx_ret = 0;
+ // disable global interrupts for vm state initialization
+ v3_disable_ints();
+
PrintDebug("Loading VMCS\n");
vmx_ret = vmcs_load(vmx_state->vmcs_ptr_phys);
+ active_vmcs_ptrs[V3_Get_CPU()] = vmx_info->vmcs_ptr_phys;
+ vmx_state->state = VMX_UNLAUNCHED;
if (vmx_ret != VMX_SUCCESS) {
PrintError("VMPTRLD failed\n");
}
+ /*** Setup default state from HW ***/
+
+ vmx_state->pin_ctrls.value = hw_info.pin_ctrls.def_val;
+ vmx_state->pri_proc_ctrls.value = hw_info.proc_ctrls.def_val;
+ vmx_state->exit_ctrls.value = hw_info.exit_ctrls.def_val;
+ vmx_state->entry_ctrls.value = hw_info.entry_ctrls.def_val;;
+
+ /* Print Control MSRs */
+ PrintDebug("CR0 MSR: %p\n", (void *)(addr_t)hw_info.cr0.value);
+ PrintDebug("CR4 MSR: %p\n", (void *)(addr_t)hw_info.cr4.value);
+
+
/******* Setup Host State **********/
/********** Setup and VMX Control Fields from MSR ***********/
- /* Setup IO map */
- struct v3_msr tmp_msr;
-
- v3_get_msr(VMX_PINBASED_CTLS_MSR, &(tmp_msr.hi), &(tmp_msr.lo));
-
/* Add external interrupts, NMI exiting, and virtual NMI */
- vmx_state->pin_ctrls.value = tmp_msr.lo;
vmx_state->pin_ctrls.nmi_exit = 1;
vmx_state->pin_ctrls.ext_int_exit = 1;
- v3_get_msr(VMX_PROCBASED_CTLS_MSR, &(tmp_msr.hi), &(tmp_msr.lo));
-
- vmx_state->pri_proc_ctrls.value = tmp_msr.lo;
vmx_state->pri_proc_ctrls.use_io_bitmap = 1;
vmx_state->pri_proc_ctrls.hlt_exit = 1;
vmx_state->pri_proc_ctrls.invlpg_exit = 1;
vmx_state->pri_proc_ctrls.rdtsc_exit = 1;
#endif
+ /* Setup IO map */
vmx_ret |= check_vmcs_write(VMCS_IO_BITMAP_A_ADDR, (addr_t)V3_PAddr(info->vm_info->io_map.arch_data));
vmx_ret |= check_vmcs_write(VMCS_IO_BITMAP_B_ADDR,
(addr_t)V3_PAddr(info->vm_info->io_map.arch_data) + PAGE_SIZE_4KB);
vmx_ret |= check_vmcs_write(VMCS_MSR_BITMAP, (addr_t)V3_PAddr(info->vm_info->msr_map.arch_data));
- v3_get_msr(VMX_EXIT_CTLS_MSR, &(tmp_msr.hi), &(tmp_msr.lo));
- vmx_state->exit_ctrls.value = tmp_msr.lo;
+
vmx_state->exit_ctrls.host_64_on = 1;
if ((vmx_state->exit_ctrls.save_efer == 1) || (vmx_state->exit_ctrls.ld_efer == 1)) {
vmx_state->ia32e_avail = 1;
}
- v3_get_msr(VMX_ENTRY_CTLS_MSR, &(tmp_msr.hi), &(tmp_msr.lo));
- vmx_state->entry_ctrls.value = tmp_msr.lo;
- {
- struct vmx_exception_bitmap excp_bmap;
- excp_bmap.value = 0;
-
- excp_bmap.pf = 1;
-
- vmx_ret |= check_vmcs_write(VMCS_EXCP_BITMAP, excp_bmap.value);
- }
/******* Setup VMXAssist guest state ***********/
info->rip = 0xd0000;
info->vm_regs.rsp = 0x80000;
-
- struct rflags * flags = (struct rflags *)&(info->ctrl_regs.rflags);
- flags->rsvd1 = 1;
-
- /* Print Control MSRs */
- v3_get_msr(VMX_CR0_FIXED0_MSR, &(tmp_msr.hi), &(tmp_msr.lo));
- PrintDebug("CR0 MSR: %p\n", (void *)(addr_t)tmp_msr.value);
-
- v3_get_msr(VMX_CR4_FIXED0_MSR, &(tmp_msr.hi), &(tmp_msr.lo));
- PrintDebug("CR4 MSR: %p\n", (void *)(addr_t)tmp_msr.value);
-
+ info->ctrl_regs.rflags->rsvd1 = 1;
#define GUEST_CR0 0x80000031
#define GUEST_CR4 0x00002000
/* Add CR exits */
vmx_state->pri_proc_ctrls.cr3_ld_exit = 1;
vmx_state->pri_proc_ctrls.cr3_str_exit = 1;
+
+ /* Add page fault exits */
+ vmx_state->excp_bmap.pf = 1;
}
// Setup segment registers
}
memcpy((void *)vmxassist_dst, v3_vmxassist_start, v3_vmxassist_end - v3_vmxassist_start);
+
+
+ vmx_state->assist_state = VMXASSIST_DISABLED;
}
- /*** Write all the info to the VMCS ***/
+
+
+ /* Sanity check ctrl/reg fields against hw_defaults */
+
+
+
+ /*** Write all the info to the VMCS ***/
+
+ {
#define DEBUGCTL_MSR 0x1d9
- v3_get_msr(DEBUGCTL_MSR, &(tmp_msr.hi), &(tmp_msr.lo));
- vmx_ret |= check_vmcs_write(VMCS_GUEST_DBG_CTL, tmp_msr.value);
+ struct v3_msr tmp_msr;
+ v3_get_msr(DEBUGCTL_MSR, &(tmp_msr.hi), &(tmp_msr.lo));
+ vmx_ret |= check_vmcs_write(VMCS_GUEST_DBG_CTL, tmp_msr.value);
+ info->dbg_regs.dr7 = 0x400;
+ }
- info->dbg_regs.dr7 = 0x400;
#ifdef __V3_64BIT__
vmx_ret |= check_vmcs_write(VMCS_LINK_PTR, (addr_t)0xffffffffffffffffULL);
vmx_ret |= check_vmcs_write(VMCS_LINK_PTR_HIGH, (addr_t)0xffffffffUL);
#endif
+
+
+
if (v3_update_vmcs_ctrl_fields(info)) {
PrintError("Could not write control fields!\n");
return -1;
}
- vmx_state->state = VMXASSIST_DISABLED;
+
+ // reenable global interrupts for vm state initialization now
+ // that the vm state is initialized. If another VM kicks us off,
+ // it'll update our vmx state so that we know to reload ourself
+ v3_enable_ints();
return 0;
}
int vmx_ret = 0;
vmx_state = (struct vmx_data *)V3_Malloc(sizeof(struct vmx_data));
+ memset(vmx_state, 0, sizeof(struct vmx_data));
PrintDebug("vmx_data pointer: %p\n", (void *)vmx_state);
PrintDebug("VMCS pointer: %p\n", (void *)(vmx_state->vmcs_ptr_phys));
info->vmm_data = vmx_state;
+ vmx_state->state = VMX_UNLAUNCHED;
PrintDebug("Initializing VMCS (addr=%p)\n", info->vmm_data);
int ret = 0;
uint32_t tsc_offset_low, tsc_offset_high;
struct vmx_exit_info exit_info;
+ struct vmx_data * vmx_info = (struct vmx_data *)(info->vmm_data);
// Conditionally yield the CPU if the timeslice has expired
v3_yield_cond(info);
// disable global interrupts for vm state transition
v3_disable_ints();
+
+ if (active_vmcs_ptrs[V3_Get_CPU()] != vmx_info->vmcs_ptr_phys) {
+ vmcs_load(vmx_info->vmcs_ptr_phys);
+ active_vmcs_ptrs[V3_Get_CPU()] = vmx_info->vmcs_ptr_phys;
+ }
+
+
v3_vmx_restore_vmcs(info);
check_vmcs_write(VMCS_TSC_OFFSET_HIGH, tsc_offset_high);
check_vmcs_write(VMCS_TSC_OFFSET, tsc_offset_low);
- if (info->vm_info->run_state == VM_STOPPED) {
+
+ if (vmx_info->state == VMX_UNLAUNCHED) {
+ vmx_info->state = VMX_LAUNCHED;
info->vm_info->run_state = VM_RUNNING;
ret = v3_vmx_launch(&(info->vm_regs), info, &(info->ctrl_regs));
} else {
+ V3_ASSERT(vmx_info->state != VMX_UNLAUNCHED);
ret = v3_vmx_resume(&(info->vm_regs), info, &(info->ctrl_regs));
}
-
+
// PrintDebug("VMX Exit: ret=%d\n", ret);
if (ret != VMX_SUCCESS) {
update_irq_exit_state(info);
#endif
+ if (exit_info.exit_reason == VMEXIT_INTR_WINDOW) {
+ // This is a special case whose only job is to inject an interrupt
+ vmcs_read(VMCS_PROC_CTRLS, &(vmx_info->pri_proc_ctrls.value));
+ vmx_info->pri_proc_ctrls.int_wndw_exit = 0;
+ vmcs_write(VMCS_PROC_CTRLS, vmx_info->pri_proc_ctrls.value);
+
+#ifdef CONFIG_DEBUG_INTERRUPTS
+ PrintDebug("Interrupts available again! (RIP=%llx)\n", info->rip);
+#endif
+ }
+
// reenable global interrupts after vm exit
v3_enable_ints();
}
+
+
+#define VMX_FEATURE_CONTROL_MSR 0x0000003a
+#define CPUID_VMX_FEATURES 0x00000005 /* LOCK and VMXON */
+#define CPUID_1_ECX_VTXFLAG 0x00000020
+
int v3_is_vmx_capable() {
v3_msr_t feature_msr;
uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0;
PrintDebug("MSRREGlow: 0x%.8x\n", feature_msr.lo);
- if ((feature_msr.lo & FEATURE_CONTROL_VALID) != FEATURE_CONTROL_VALID) {
+ if ((feature_msr.lo & CPUID_VMX_FEATURES) != CPUID_VMX_FEATURES) {
PrintDebug("VMX is locked -- enable in the BIOS\n");
return 0;
}
return 1;
}
-static int has_vmx_nested_paging() {
- return 0;
-}
-void v3_init_vmx_cpu(int cpu_id) {
- extern v3_cpu_arch_t v3_cpu_types[];
- struct v3_msr tmp_msr;
- uint64_t ret = 0;
- v3_get_msr(VMX_CR4_FIXED0_MSR, &(tmp_msr.hi), &(tmp_msr.lo));
-#ifdef __V3_64BIT__
- __asm__ __volatile__ (
- "movq %%cr4, %%rbx;"
- "orq $0x00002000, %%rbx;"
- "movq %%rbx, %0;"
- : "=m"(ret)
- :
- : "%rbx"
- );
-
- if ((~ret & tmp_msr.value) == 0) {
- __asm__ __volatile__ (
- "movq %0, %%cr4;"
- :
- : "q"(ret)
- );
- } else {
- PrintError("Invalid CR4 Settings!\n");
- return;
- }
+void v3_init_vmx_cpu(int cpu_id) {
+ extern v3_cpu_arch_t v3_cpu_types[];
- __asm__ __volatile__ (
- "movq %%cr0, %%rbx; "
- "orq $0x00000020,%%rbx; "
- "movq %%rbx, %%cr0;"
- :
- :
- : "%rbx"
- );
-#elif __V3_32BIT__
- __asm__ __volatile__ (
- "movl %%cr4, %%ecx;"
- "orl $0x00002000, %%ecx;"
- "movl %%ecx, %0;"
- : "=m"(ret)
- :
- : "%ecx"
- );
-
- if ((~ret & tmp_msr.value) == 0) {
- __asm__ __volatile__ (
- "movl %0, %%cr4;"
- :
- : "q"(ret)
- );
- } else {
- PrintError("Invalid CR4 Settings!\n");
- return;
+ if (cpu_id == 0) {
+ if (v3_init_vmx_hw(&hw_info) == -1) {
+ PrintError("Could not initialize VMX hardware features on cpu %d\n", cpu_id);
+ return;
+ }
}
- __asm__ __volatile__ (
- "movl %%cr0, %%ecx; "
- "orl $0x00000020,%%ecx; "
- "movl %%ecx, %%cr0;"
- :
- :
- : "%ecx"
- );
-
-#endif
- //
- // Should check and return Error here....
+ enable_vmx();
// Setup VMXON Region
PrintDebug("VMXON pointer: 0x%p\n", (void *)host_vmcs_ptrs[cpu_id]);
- if (v3_enable_vmx(host_vmcs_ptrs[cpu_id]) == VMX_SUCCESS) {
+ if (vmx_on(host_vmcs_ptrs[cpu_id]) == VMX_SUCCESS) {
PrintDebug("VMX Enabled\n");
} else {
PrintError("VMX initialization failure\n");
}
- if (has_vmx_nested_paging() == 1) {
- v3_cpu_types[cpu_id] = V3_VMX_EPT_CPU;
- } else {
- v3_cpu_types[cpu_id] = V3_VMX_CPU;
- }
+ v3_cpu_types[cpu_id] = V3_VMX_CPU;
+
}