#include <palacios/vmx.h>
-#include <palacios/vmcs.h>
#include <palacios/vmm.h>
+#include <palacios/vmx_handler.h>
+#include <palacios/vmcs.h>
#include <palacios/vmx_lowlevel.h>
#include <palacios/vmm_lowlevel.h>
#include <palacios/vmm_ctrl_regs.h>
#include <palacios/vmm_config.h>
+#include <palacios/vmm_time.h>
#include <palacios/vm_guest_mem.h>
#include <palacios/vmm_direct_paging.h>
#include <palacios/vmx_io.h>
#include <palacios/vmx_msr.h>
-static addr_t vmxon_ptr_phys;
-extern int v3_vmx_exit_handler();
-extern int v3_vmx_vmlaunch(struct v3_gprs * vm_regs, struct guest_info * info);
+#include <palacios/vmx_hw_info.h>
+
+#ifndef CONFIG_DEBUG_VMX
+#undef PrintDebug
+#define PrintDebug(fmt, args...)
+#endif
+
-static int inline check_vmcs_write(vmcs_field_t field, addr_t val)
-{
+/* 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 0;
}
-static void inline translate_segment_access(struct v3_segment * v3_seg,
- struct vmcs_segment_access * access)
-{
- access->type = v3_seg->type;
- access->desc_type = v3_seg->system;
- access->dpl = v3_seg->dpl;
- access->present = v3_seg->present;
- access->avail = v3_seg->avail;
- access->long_mode = v3_seg->long_mode;
- access->db = v3_seg->db;
- access->granularity = v3_seg->granularity;
-}
-
-int v3_update_vmcs_ctrl_fields(struct guest_info * info) {
- int vmx_ret = 0;
- struct vmx_data * arch_data = (struct vmx_data *)(info->vmm_data);
+static int inline check_vmcs_read(vmcs_field_t field, void * val) {
+ int ret = 0;
- vmx_ret |= check_vmcs_write(VMCS_PIN_CTRLS, arch_data->pinbased_ctrls);
- vmx_ret |= check_vmcs_write(VMCS_PROC_CTRLS, arch_data->pri_procbased_ctrls);
+ ret = vmcs_read(field, val);
- if(arch_data->pri_procbased_ctrls & ACTIVE_SEC_CTRLS) {
- vmx_ret |= check_vmcs_write(VMCS_SEC_PROC_CTRLS, arch_data->sec_procbased_ctrls);
+ if (ret != VMX_SUCCESS) {
+ PrintError("VMREAD error on %s!: %d\n", v3_vmcs_field_to_str(field), ret);
}
- vmx_ret |= check_vmcs_write(VMCS_EXIT_CTRLS, arch_data->exit_ctrls);
- vmx_ret |= check_vmcs_write(VMCS_ENTRY_CTRLS, arch_data->entry_ctrls);
-
- return vmx_ret;
+ return ret;
}
-int v3_update_vmcs_host_state(struct guest_info * info) {
- int vmx_ret = 0;
- addr_t tmp;
- struct vmx_data * arch_data = (struct vmx_data *)(info->vmm_data);
- struct v3_msr tmp_msr;
- __asm__ __volatile__ ( "movq %%cr0, %0; "
- : "=q"(tmp)
- :
- );
- vmx_ret |= check_vmcs_write(VMCS_HOST_CR0, tmp);
- __asm__ __volatile__ ( "movq %%cr3, %0; "
- : "=q"(tmp)
- :
- );
- vmx_ret |= check_vmcs_write(VMCS_HOST_CR3, tmp);
+static addr_t allocate_vmcs() {
+ struct vmcs_data * vmcs_page = NULL;
+ PrintDebug("Allocating page\n");
- __asm__ __volatile__ ( "movq %%cr4, %0; "
- : "=q"(tmp)
- :
- );
- vmx_ret |= check_vmcs_write(VMCS_HOST_CR4, tmp);
+ vmcs_page = (struct vmcs_data *)V3_VAddr(V3_AllocPages(1));
+ memset(vmcs_page, 0, 4096);
+ 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);
+}
- vmx_ret |= check_vmcs_write(VMCS_HOST_GDTR_BASE, arch_data->host_state.gdtr.base);
- vmx_ret |= check_vmcs_write(VMCS_HOST_IDTR_BASE, arch_data->host_state.idtr.base);
- vmx_ret |= check_vmcs_write(VMCS_HOST_TR_BASE, arch_data->host_state.tr.base);
-#define FS_BASE_MSR 0xc0000100
-#define GS_BASE_MSR 0xc0000101
- // FS.BASE MSR
- v3_get_msr(FS_BASE_MSR, &(tmp_msr.hi), &(tmp_msr.lo));
- vmx_ret |= check_vmcs_write(VMCS_HOST_FS_BASE, tmp_msr.value);
- // GS.BASE MSR
- v3_get_msr(GS_BASE_MSR, &(tmp_msr.hi), &(tmp_msr.lo));
- vmx_ret |= check_vmcs_write(VMCS_HOST_GS_BASE, tmp_msr.value);
+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;
- __asm__ __volatile__ ( "movq %%cs, %0; "
- : "=q"(tmp)
- :
- );
- vmx_ret |= check_vmcs_write(VMCS_HOST_CS_SELECTOR, tmp);
+ if (vmx_ret != VMX_SUCCESS) {
+ PrintError("VMPTRLD failed\n");
+ return -1;
+ }
- __asm__ __volatile__ ( "movq %%ss, %0; "
- : "=q"(tmp)
- :
- );
- vmx_ret |= check_vmcs_write(VMCS_HOST_SS_SELECTOR, tmp);
- __asm__ __volatile__ ( "movq %%ds, %0; "
- : "=q"(tmp)
- :
- );
- vmx_ret |= check_vmcs_write(VMCS_HOST_DS_SELECTOR, tmp);
+ /*** Setup default state from HW ***/
- __asm__ __volatile__ ( "movq %%es, %0; "
- : "=q"(tmp)
- :
- );
- vmx_ret |= check_vmcs_write(VMCS_HOST_ES_SELECTOR, tmp);
+ 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;;
- __asm__ __volatile__ ( "movq %%fs, %0; "
- : "=q"(tmp)
- :
- );
- vmx_ret |= check_vmcs_write(VMCS_HOST_FS_SELECTOR, tmp);
+ /* 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);
- __asm__ __volatile__ ( "movq %%gs, %0; "
- : "=q"(tmp)
- :
- );
- vmx_ret |= check_vmcs_write(VMCS_HOST_GS_SELECTOR, tmp);
- vmx_ret |= check_vmcs_write(VMCS_HOST_TR_SELECTOR, arch_data->host_state.tr.selector);
+ /******* Setup Host State **********/
-#define SYSENTER_CS_MSR 0x00000174
-#define SYSENTER_ESP_MSR 0x00000175
-#define SYSENTER_EIP_MSR 0x00000176
+ /* Cache GDTR, IDTR, and TR in host struct */
+ addr_t gdtr_base;
+ struct {
+ uint16_t selector;
+ addr_t base;
+ } __attribute__((packed)) tmp_seg;
+
- // SYSENTER CS MSR
- v3_get_msr(SYSENTER_CS_MSR, &(tmp_msr.hi), &(tmp_msr.lo));
- vmx_ret |= check_vmcs_write(VMCS_HOST_SYSENTER_CS, tmp_msr.lo);
+ __asm__ __volatile__(
+ "sgdt (%0);"
+ :
+ : "q"(&tmp_seg)
+ : "memory"
+ );
+ gdtr_base = tmp_seg.base;
+ vmx_state->host_state.gdtr.base = gdtr_base;
- // SYSENTER_ESP MSR
- v3_get_msr(SYSENTER_ESP_MSR, &(tmp_msr.hi), &(tmp_msr.lo));
- vmx_ret |= check_vmcs_write(VMCS_HOST_SYSENTER_ESP, tmp_msr.value);
+ __asm__ __volatile__(
+ "sidt (%0);"
+ :
+ : "q"(&tmp_seg)
+ : "memory"
+ );
+ vmx_state->host_state.idtr.base = tmp_seg.base;
- // SYSENTER_EIP MSR
- v3_get_msr(SYSENTER_EIP_MSR, &(tmp_msr.hi), &(tmp_msr.lo));
- vmx_ret |= check_vmcs_write(VMCS_HOST_SYSENTER_EIP, tmp_msr.value);
+ __asm__ __volatile__(
+ "str (%0);"
+ :
+ : "q"(&tmp_seg)
+ : "memory"
+ );
+ vmx_state->host_state.tr.selector = tmp_seg.selector;
- return vmx_ret;
-}
+ /* The GDTR *index* is bits 3-15 of the selector. */
+ struct tss_descriptor * desc = NULL;
+ desc = (struct tss_descriptor *)(gdtr_base + (8 * (tmp_seg.selector >> 3)));
+ tmp_seg.base = ((desc->base1) |
+ (desc->base2 << 16) |
+ (desc->base3 << 24) |
+#ifdef __V3_64BIT__
+ ((uint64_t)desc->base4 << 32)
+#else
+ (0)
+#endif
+ );
-int v3_update_vmcs_guest_state(struct guest_info * info)
-{
- int vmx_ret = 0;
+ vmx_state->host_state.tr.base = tmp_seg.base;
- vmx_ret |= check_vmcs_write(VMCS_GUEST_RIP, info->rip);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_RSP, info->vm_regs.rsp);
-
+
- vmx_ret |= check_vmcs_write(VMCS_GUEST_CR0, info->ctrl_regs.cr0);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_CR3, info->ctrl_regs.cr3);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_CR4, info->ctrl_regs.cr4);
+ /********** Setup and VMX Control Fields from MSR ***********/
- vmx_ret |= check_vmcs_write(VMCS_GUEST_RFLAGS, info->ctrl_regs.rflags);
+ /* Add external interrupts, NMI exiting, and virtual NMI */
+ vmx_state->pin_ctrls.nmi_exit = 1;
+ vmx_state->pin_ctrls.ext_int_exit = 1;
+ 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.use_msr_bitmap = 1;
+ vmx_state->pri_proc_ctrls.pause_exit = 1;
+ vmx_state->pri_proc_ctrls.tsc_offset = 1;
+#ifdef CONFIG_TIME_VIRTUALIZE_TSC
+ vmx_state->pri_proc_ctrls.rdtsc_exit = 1;
+#endif
- /*** Write VMCS Segments ***/
- struct vmcs_segment_access access;
+ /* 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);
- memset(&access, 0, sizeof(access));
- /* CS Segment */
- translate_segment_access(&(info->segments.cs), &access);
+ vmx_ret |= check_vmcs_write(VMCS_MSR_BITMAP, (addr_t)V3_PAddr(info->vm_info->msr_map.arch_data));
- vmx_ret |= check_vmcs_write(VMCS_GUEST_CS_BASE, info->segments.cs.base);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_CS_SELECTOR, info->segments.cs.selector);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_CS_LIMIT, info->segments.cs.limit);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_CS_ACCESS, access.value);
- /* SS Segment */
- memset(&access, 0, sizeof(access));
- translate_segment_access(&(info->segments.ss), &access);
+ vmx_state->exit_ctrls.host_64_on = 1;
- vmx_ret |= check_vmcs_write(VMCS_GUEST_SS_BASE, info->segments.ss.base);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_SS_SELECTOR, info->segments.ss.selector);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_SS_LIMIT, info->segments.ss.limit);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_SS_ACCESS, access.value);
+ if ((vmx_state->exit_ctrls.save_efer == 1) || (vmx_state->exit_ctrls.ld_efer == 1)) {
+ vmx_state->ia32e_avail = 1;
+ }
- /* DS Segment */
- memset(&access, 0, sizeof(access));
- translate_segment_access(&(info->segments.ds), &access);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_DS_BASE, info->segments.ds.base);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_DS_SELECTOR, info->segments.ds.selector);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_DS_LIMIT, info->segments.ds.limit);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_DS_ACCESS, access.value);
+ /******* Setup VMXAssist guest state ***********/
+ info->rip = 0xd0000;
+ info->vm_regs.rsp = 0x80000;
+ info->ctrl_regs.rflags->rsvd1 = 1;
- /* ES Segment */
- memset(&access, 0, sizeof(access));
- translate_segment_access(&(info->segments.es), &access);
+#define GUEST_CR0 0x80000031
+#define GUEST_CR4 0x00002000
+ info->ctrl_regs.cr0 = GUEST_CR0;
+ info->ctrl_regs.cr4 = GUEST_CR4;
- vmx_ret |= check_vmcs_write(VMCS_GUEST_ES_BASE, info->segments.es.base);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_ES_SELECTOR, info->segments.es.selector);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_ES_LIMIT, info->segments.es.limit);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_ES_ACCESS, access.value);
+ ((struct cr0_32 *)&(info->shdw_pg_state.guest_cr0))->pe = 1;
+
+ /* Setup paging */
+ if (info->shdw_pg_mode == SHADOW_PAGING) {
+ PrintDebug("Creating initial shadow page table\n");
- /* FS Segment */
- memset(&access, 0, sizeof(access));
- translate_segment_access(&(info->segments.fs), &access);
+ if (v3_init_passthrough_pts(info) == -1) {
+ PrintError("Could not initialize passthrough page tables\n");
+ return -1;
+ }
+
+#define CR0_PE 0x00000001
+#define CR0_PG 0x80000000
- vmx_ret |= check_vmcs_write(VMCS_GUEST_FS_BASE, info->segments.fs.base);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_FS_SELECTOR, info->segments.fs.selector);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_FS_LIMIT, info->segments.fs.limit);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_FS_ACCESS, access.value);
- /* GS Segment */
- memset(&access, 0, sizeof(access));
- translate_segment_access(&(info->segments.gs), &access);
+ vmx_ret |= check_vmcs_write(VMCS_CR0_MASK, (CR0_PE | CR0_PG) );
+ vmx_ret |= check_vmcs_write(VMCS_CR4_MASK, CR4_VMXE);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_GS_BASE, info->segments.gs.base);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_GS_SELECTOR, info->segments.gs.selector);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_GS_LIMIT, info->segments.gs.limit);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_GS_ACCESS, access.value);
+ info->ctrl_regs.cr3 = info->direct_map_pt;
- /* LDTR segment */
- memset(&access, 0, sizeof(access));
- translate_segment_access(&(info->segments.ldtr), &access);
+ // vmx_state->pinbased_ctrls |= NMI_EXIT;
- vmx_ret |= check_vmcs_write(VMCS_GUEST_LDTR_BASE, info->segments.ldtr.base);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_LDTR_SELECTOR, info->segments.ldtr.selector);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_LDTR_LIMIT, info->segments.ldtr.limit);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_LDTR_ACCESS, access.value);
+ /* 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;
+ }
- /* TR Segment */
- memset(&access, 0, sizeof(access));
- translate_segment_access(&(info->segments.tr), &access);
+ // Setup segment registers
+ {
+ struct v3_segment * seg_reg = (struct v3_segment *)&(info->segments);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_TR_BASE, info->segments.tr.base);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_TR_SELECTOR, info->segments.tr.selector);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_TR_LIMIT, info->segments.tr.limit);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_TR_ACCESS, access.value);
+ int i;
- /* GDTR Segment */
+ for (i = 0; i < 10; i++) {
+ seg_reg[i].selector = 3 << 3;
+ seg_reg[i].limit = 0xffff;
+ seg_reg[i].base = 0x0;
+ }
- vmx_ret |= check_vmcs_write(VMCS_GUEST_GDTR_BASE, info->segments.gdtr.base);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_GDTR_LIMIT, info->segments.gdtr.limit);
+ info->segments.cs.selector = 2<<3;
- /* IDTR Segment*/
- vmx_ret |= check_vmcs_write(VMCS_GUEST_IDTR_BASE, info->segments.idtr.base);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_IDTR_LIMIT, info->segments.idtr.limit);
+ /* Set only the segment registers */
+ for (i = 0; i < 6; i++) {
+ seg_reg[i].limit = 0xfffff;
+ seg_reg[i].granularity = 1;
+ seg_reg[i].type = 3;
+ seg_reg[i].system = 1;
+ seg_reg[i].dpl = 0;
+ seg_reg[i].present = 1;
+ seg_reg[i].db = 1;
+ }
- return vmx_ret;
+ info->segments.cs.type = 0xb;
-}
+ info->segments.ldtr.selector = 0x20;
+ info->segments.ldtr.type = 2;
+ info->segments.ldtr.system = 0;
+ info->segments.ldtr.present = 1;
+ info->segments.ldtr.granularity = 0;
+
+ /************* Map in GDT and vmxassist *************/
+ uint64_t gdt[] __attribute__ ((aligned(32))) = {
+ 0x0000000000000000ULL, /* 0x00: reserved */
+ 0x0000830000000000ULL, /* 0x08: 32-bit TSS */
+ //0x0000890000000000ULL, /* 0x08: 32-bit TSS */
+ 0x00CF9b000000FFFFULL, /* 0x10: CS 32-bit */
+ 0x00CF93000000FFFFULL, /* 0x18: DS 32-bit */
+ 0x000082000000FFFFULL, /* 0x20: LDTR 32-bit */
+ };
+#define VMXASSIST_GDT 0x10000
+ addr_t vmxassist_gdt = 0;
-#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;
+ if (v3_gpa_to_hva(info, VMXASSIST_GDT, &vmxassist_gdt) == -1) {
+ PrintError("Could not find VMXASSIST GDT destination\n");
+ return -1;
+ }
- PrintDebug("sanitize_bits1 (MSR:%x)\n", msr_num);
+ memcpy((void *)vmxassist_gdt, gdt, sizeof(uint64_t) * 5);
+
+ info->segments.gdtr.base = VMXASSIST_GDT;
- v3_get_msr(msr_num, &mask_msr.hi, &mask_msr.lo);
+#define VMXASSIST_TSS 0x40000
+ uint64_t vmxassist_tss = VMXASSIST_TSS;
+ gdt[0x08 / sizeof(gdt[0])] |=
+ ((vmxassist_tss & 0xFF000000) << (56 - 24)) |
+ ((vmxassist_tss & 0x00FF0000) << (32 - 16)) |
+ ((vmxassist_tss & 0x0000FFFF) << (16)) |
+ (8392 - 1);
+
+ info->segments.tr.selector = 0x08;
+ info->segments.tr.base = vmxassist_tss;
+
+ //info->segments.tr.type = 0x9;
+ info->segments.tr.type = 0x3;
+ info->segments.tr.system = 0;
+ info->segments.tr.present = 1;
+ info->segments.tr.granularity = 0;
+ }
+
+ // setup VMXASSIST
+ {
+#define VMXASSIST_START 0x000d0000
+ extern uint8_t v3_vmxassist_start[];
+ extern uint8_t v3_vmxassist_end[];
+ addr_t vmxassist_dst = 0;
- PrintDebug("MSR %x = %x : %x \n", msr_num, mask_msr.hi, mask_msr.lo);
+ if (v3_gpa_to_hva(info, VMXASSIST_START, &vmxassist_dst) == -1) {
+ PrintError("Could not find VMXASSIST destination\n");
+ return -1;
+ }
- val |= mask_msr.lo;
- val |= mask_msr.hi;
-
- return val;
-}
+ memcpy((void *)vmxassist_dst, v3_vmxassist_start, v3_vmxassist_end - v3_vmxassist_start);
+ vmx_state->assist_state = VMXASSIST_DISABLED;
+ }
-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);
+ /* Sanity check ctrl/reg fields against hw_defaults */
- val |= msr0_val;
- val |= msr1_val;
- return val;
-}
+ /*** Write all the info to the VMCS ***/
+
+ {
+#define DEBUGCTL_MSR 0x1d9
+ 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;
+ }
+#ifdef __V3_64BIT__
+ vmx_ret |= check_vmcs_write(VMCS_LINK_PTR, (addr_t)0xffffffffffffffffULL);
+#else
+ vmx_ret |= check_vmcs_write(VMCS_LINK_PTR, (addr_t)0xffffffffUL);
+ vmx_ret |= check_vmcs_write(VMCS_LINK_PTR_HIGH, (addr_t)0xffffffffUL);
#endif
-static addr_t allocate_vmcs()
-{
- reg_ex_t msr;
- PrintDebug("Allocating page\n");
- struct vmcs_data * vmcs_page = (struct vmcs_data *)V3_VAddr(V3_AllocPages(1));
+
+
+ if (v3_update_vmcs_ctrl_fields(info)) {
+ PrintError("Could not write control fields!\n");
+ return -1;
+ }
+
+ if (v3_update_vmcs_host_state(info)) {
+ PrintError("Could not write host state\n");
+ return -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);
+ // 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 (addr_t)V3_PAddr((void *)vmcs_page);
+ return 0;
}
-#if 0
+int v3_init_vmx_vmcs(struct guest_info * info, v3_vm_class_t vm_class) {
+ struct vmx_data * vmx_state = NULL;
+ int vmx_ret = 0;
+
+ vmx_state = (struct vmx_data *)V3_Malloc(sizeof(struct vmx_data));
+ memset(vmx_state, 0, sizeof(struct vmx_data));
-#endif
+ PrintDebug("vmx_data pointer: %p\n", (void *)vmx_state);
-#if 0
-static int init_vmcs_bios(struct guest_info * vm_info)
-{
-#if 0
+ PrintDebug("Allocating VMCS\n");
+ vmx_state->vmcs_ptr_phys = allocate_vmcs();
- setup_v8086_mode_for_boot(vm_info);
+ PrintDebug("VMCS pointer: %p\n", (void *)(vmx_state->vmcs_ptr_phys));
+ info->vmm_data = vmx_state;
+ vmx_state->state = VMX_UNLAUNCHED;
- // Setup guest state
- // TODO: This is not 32-bit safe!
- vmx_ret |= check_vmcs_write(VMCS_GUEST_RIP, vm_info->rip);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_RSP, vm_info->vm_regs.rsp);
+ PrintDebug("Initializing VMCS (addr=%p)\n", info->vmm_data);
+ // TODO: Fix vmcs fields so they're 32-bit
- vmx_ret |= check_vmcs_write(VMCS_GUEST_CR0, vm_info->ctrl_regs.cr0);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_CR4, vm_info->ctrl_regs.cr4);
+ PrintDebug("Clearing VMCS: %p\n", (void *)vmx_state->vmcs_ptr_phys);
+ vmx_ret = vmcs_clear(vmx_state->vmcs_ptr_phys);
- vmx_ret |= vmcs_write_guest_segments(vm_info);
+ if (vmx_ret != VMX_SUCCESS) {
+ PrintError("VMCLEAR failed\n");
+ return -1;
+ }
- vmx_ret |= check_vmcs_write(VMCS_GUEST_RFLAGS, vm_info->ctrl_regs.rflags);
-#define DEBUGCTL_MSR 0x1d9
+ if (vm_class == V3_PC_VM) {
+ PrintDebug("Initializing VMCS\n");
+ init_vmcs_bios(info, vmx_state);
+ } else {
+ PrintError("Invalid VM Class\n");
+ return -1;
+ }
- v3_get_msr(DEBUGCTL_MSR, &(tmp_msr.hi), &(tmp_msr.lo));
- vmx_ret |= check_vmcs_write(VMCS_GUEST_DBG_CTL, tmp_msr.value);
+ return 0;
+}
- vmx_ret |= check_vmcs_write(VMCS_GUEST_DR7, 0x400);
- vmx_ret |= check_vmcs_write(VMCS_LINK_PTR, 0xffffffffffffffff);
+int v3_deinit_vmx_vmcs(struct guest_info * core) {
+ struct vmx_data * vmx_state = core->vmm_data;
- if (vmx_ret != 0) {
- PrintError("Could not initialize VMCS segments\n");
- return -1;
- }
+ V3_FreePages((void *)(vmx_state->vmcs_ptr_phys), 1);
+
+ V3_Free(vmx_state);
-#endif
return 0;
}
-#endif
-static int init_vmx_guest(struct guest_info * info, struct v3_vm_config * config_ptr) {
- v3_pre_config_guest(info, config_ptr);
- struct vmx_data * vmx_data = NULL;
+static int update_irq_exit_state(struct guest_info * info) {
+ struct vmx_exit_idt_vec_info idt_vec_info;
- vmx_data = (struct vmx_data *)V3_Malloc(sizeof(struct vmx_data));
+ check_vmcs_read(VMCS_IDT_VECTOR_INFO, &(idt_vec_info.value));
- PrintDebug("vmx_data pointer: %p\n", (void *)vmx_data);
+ if ((info->intr_core_state.irq_started == 1) && (idt_vec_info.valid == 0)) {
+#ifdef CONFIG_DEBUG_INTERRUPTS
+ PrintDebug("Calling v3_injecting_intr\n");
+#endif
+ info->intr_core_state.irq_started = 0;
+ v3_injecting_intr(info, info->intr_core_state.irq_vector, V3_EXTERNAL_IRQ);
+ }
- PrintDebug("Allocating VMCS\n");
- vmx_data->vmcs_ptr_phys = allocate_vmcs();
+ return 0;
+}
- PrintDebug("VMCS pointer: %p\n", (void *)(vmx_data->vmcs_ptr_phys));
+static int update_irq_entry_state(struct guest_info * info) {
+ struct vmx_exit_idt_vec_info idt_vec_info;
+ struct vmcs_interrupt_state intr_core_state;
+ struct vmx_data * vmx_info = (struct vmx_data *)(info->vmm_data);
- info->vmm_data = vmx_data;
+ check_vmcs_read(VMCS_IDT_VECTOR_INFO, &(idt_vec_info.value));
+ check_vmcs_read(VMCS_GUEST_INT_STATE, &(intr_core_state));
- PrintDebug("Initializing VMCS (addr=%p)\n", info->vmm_data);
-
- // TODO: Fix vmcs fields so they're 32-bit
- int vmx_ret = 0;
+ /* Check for pending exceptions to inject */
+ if (v3_excp_pending(info)) {
+ struct vmx_entry_int_info int_info;
+ int_info.value = 0;
- PrintDebug("Clearing VMCS: %p\n",(void*)vmx_data->vmcs_ptr_phys);
- vmx_ret = vmcs_clear(vmx_data->vmcs_ptr_phys);
+ // In VMX, almost every exception is hardware
+ // Software exceptions are pretty much only for breakpoint or overflow
+ int_info.type = 3;
+ int_info.vector = v3_get_excp_number(info);
- if (vmx_ret != VMX_SUCCESS) {
- PrintError("VMCLEAR failed\n");
- return -1;
- }
-
- PrintDebug("Loading VMCS\n");
- vmx_ret = vmcs_load(vmx_data->vmcs_ptr_phys);
-
- if (vmx_ret != VMX_SUCCESS) {
- PrintError("VMPTRLD failed\n");
- return -1;
- }
+ if (info->excp_state.excp_error_code_valid) {
+ check_vmcs_write(VMCS_ENTRY_EXCP_ERR, info->excp_state.excp_error_code);
+ int_info.error_code = 1;
+#ifdef CONFIG_DEBUG_INTERRUPTS
+ PrintDebug("Injecting exception %d with error code %x\n",
+ int_info.vector, info->excp_state.excp_error_code);
+#endif
+ }
+ int_info.valid = 1;
+#ifdef CONFIG_DEBUG_INTERRUPTS
+ PrintDebug("Injecting exception %d (EIP=%p)\n", int_info.vector, (void *)(addr_t)info->rip);
+#endif
+ check_vmcs_write(VMCS_ENTRY_INT_INFO, int_info.value);
- /******* Setup Host State **********/
+ v3_injecting_excp(info, int_info.vector);
- /* Cache GDTR, IDTR, and TR in host struct */
- addr_t gdtr_base;
- struct {
- uint16_t selector;
- addr_t base;
- } __attribute__((packed)) tmp_seg;
-
+ } else if ((((struct rflags *)&(info->ctrl_regs.rflags))->intr == 1) &&
+ (intr_core_state.val == 0)) {
+
+ if ((info->intr_core_state.irq_started == 1) && (idt_vec_info.valid == 1)) {
- __asm__ __volatile__(
- "sgdt (%0);"
- :
- : "q"(&tmp_seg)
- : "memory"
- );
- gdtr_base = tmp_seg.base;
- vmx_data->host_state.gdtr.base = gdtr_base;
+#ifdef CONFIG_DEBUG_INTERRUPTS
+ PrintDebug("IRQ pending from previous injection\n");
+#endif
- __asm__ __volatile__(
- "sidt (%0);"
- :
- : "q"(&tmp_seg)
- : "memory"
- );
- vmx_data->host_state.idtr.base = tmp_seg.base;
+ // Copy the IDT vectoring info over to reinject the old interrupt
+ if (idt_vec_info.error_code == 1) {
+ uint32_t err_code = 0;
+
+ check_vmcs_read(VMCS_IDT_VECTOR_ERR, &err_code);
+ check_vmcs_write(VMCS_ENTRY_EXCP_ERR, err_code);
+ }
+
+ idt_vec_info.undef = 0;
+ check_vmcs_write(VMCS_ENTRY_INT_INFO, idt_vec_info.value);
+
+ } else {
+ struct vmx_entry_int_info ent_int;
+ ent_int.value = 0;
+
+ switch (v3_intr_pending(info)) {
+ case V3_EXTERNAL_IRQ: {
+ info->intr_core_state.irq_vector = v3_get_intr(info);
+ ent_int.vector = info->intr_core_state.irq_vector;
+ ent_int.type = 0;
+ ent_int.error_code = 0;
+ ent_int.valid = 1;
+
+#ifdef CONFIG_DEBUG_INTERRUPTS
+ PrintDebug("Injecting Interrupt %d at exit %u(EIP=%p)\n",
+ info->intr_core_state.irq_vector,
+ (uint32_t)info->num_exits,
+ (void *)(addr_t)info->rip);
+#endif
- __asm__ __volatile__(
- "str (%0);"
- :
- : "q"(&tmp_seg)
- : "memory"
- );
- vmx_data->host_state.tr.selector = tmp_seg.selector;
+ check_vmcs_write(VMCS_ENTRY_INT_INFO, ent_int.value);
+ info->intr_core_state.irq_started = 1;
+
+ break;
+ }
+ case V3_NMI:
+ PrintDebug("Injecting NMI\n");
+
+ ent_int.type = 2;
+ ent_int.vector = 2;
+ ent_int.valid = 1;
+ check_vmcs_write(VMCS_ENTRY_INT_INFO, ent_int.value);
+
+ break;
+ case V3_SOFTWARE_INTR:
+ PrintDebug("Injecting software interrupt\n");
+ ent_int.type = 4;
+
+ ent_int.valid = 1;
+ check_vmcs_write(VMCS_ENTRY_INT_INFO, ent_int.value);
+
+ break;
+ case V3_VIRTUAL_IRQ:
+ // Not sure what to do here, Intel doesn't have virtual IRQs
+ // May be the same as external interrupts/IRQs
+
+ break;
+ case V3_INVALID_INTR:
+ default:
+ break;
+ }
+ }
+ } else if ((v3_intr_pending(info)) && (vmx_info->pri_proc_ctrls.int_wndw_exit == 0)) {
+ // Enable INTR window exiting so we know when IF=1
+ uint32_t instr_len;
- /* The GDTR *index* is bits 3-15 of the selector. */
- struct tss_descriptor * desc = (struct tss_descriptor *)
- (gdtr_base + 8*(tmp_seg.selector>>3));
+ check_vmcs_read(VMCS_EXIT_INSTR_LEN, &instr_len);
- tmp_seg.base = (
- (desc->base1) |
- (desc->base2 << 16) |
- (desc->base3 << 24) |
-#ifdef __V3_64BIT__
- ((uint64_t)desc->base4 << 32)
-#else
- (0)
+#ifdef CONFIG_DEBUG_INTERRUPTS
+ PrintDebug("Enabling Interrupt-Window exiting: %d\n", instr_len);
#endif
- );
- vmx_data->host_state.tr.base = tmp_seg.base;
+ vmx_info->pri_proc_ctrls.int_wndw_exit = 1;
+ check_vmcs_write(VMCS_PROC_CTRLS, vmx_info->pri_proc_ctrls.value);
+ }
-
- /********** Setup and VMX Control Fields from MSR ***********/
- struct v3_msr tmp_msr;
+ return 0;
+}
- v3_get_msr(VMX_PINBASED_CTLS_MSR,&(tmp_msr.hi),&(tmp_msr.lo));
- /* Add NMI exiting */
- vmx_data->pinbased_ctrls = tmp_msr.lo | NMI_EXIT;
- v3_get_msr(VMX_PROCBASED_CTLS_MSR, &(tmp_msr.hi), &(tmp_msr.lo));
- vmx_data->pri_procbased_ctrls = tmp_msr.lo;
- v3_get_msr(VMX_EXIT_CTLS_MSR, &(tmp_msr.hi), &(tmp_msr.lo));
- vmx_data->exit_ctrls = tmp_msr.lo ;
+static struct vmx_exit_info exit_log[10];
- v3_get_msr(VMX_ENTRY_CTLS_MSR, &(tmp_msr.hi), &(tmp_msr.lo));
- vmx_data->entry_ctrls = tmp_msr.lo;
+static void print_exit_log(struct guest_info * info) {
+ int cnt = info->num_exits % 10;
+ int i = 0;
+
- struct vmx_exception_bitmap excp_bmap;
- excp_bmap.value = 0xffffffff;
- vmx_ret |= check_vmcs_write(VMCS_EXCP_BITMAP, excp_bmap.value);
+ V3_Print("\nExit Log (%d total exits):\n", (uint32_t)info->num_exits);
+ for (i = 0; i < 10; i++) {
+ struct vmx_exit_info * tmp = &exit_log[cnt];
- /******* Setup VMXAssist guest state ***********/
+ V3_Print("%d:\texit_reason = %p\n", i, (void *)(addr_t)tmp->exit_reason);
+ V3_Print("\texit_qual = %p\n", (void *)tmp->exit_qual);
+ V3_Print("\tint_info = %p\n", (void *)(addr_t)tmp->int_info);
+ V3_Print("\tint_err = %p\n", (void *)(addr_t)tmp->int_err);
+ V3_Print("\tinstr_info = %p\n", (void *)(addr_t)tmp->instr_info);
- info->rip = 0xd0000;
- info->vm_regs.rsp = 0x80000;
+ cnt--;
- struct rflags * flags = (struct rflags *)&(info->ctrl_regs.rflags);
- flags->rsvd1 = 1;
+ if (cnt == -1) {
+ cnt = 9;
+ }
- /* 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);
+ }
+}
-#define GUEST_CR0 0x80000031
-#define GUEST_CR4 0x00002000
- info->ctrl_regs.cr0 = GUEST_CR0;
- info->ctrl_regs.cr4 = GUEST_CR4;
-
- /* Setup paging */
- if(info->shdw_pg_mode == SHADOW_PAGING) {
- PrintDebug("Creating initial shadow page table\n");
+/*
+ * CAUTION and DANGER!!!
+ *
+ * The VMCS CANNOT(!!) be accessed outside of the cli/sti calls inside this function
+ * When exectuing a symbiotic call, the VMCS 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_vmx_enter(struct guest_info * info) {
+ 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);
- if(v3_init_passthrough_pts(info) == -1) {
- PrintError("Could not initialize passthrough page tables\n");
- return -1;
- }
+ // Conditionally yield the CPU if the timeslice has expired
+ v3_yield_cond(info);
- info->shdw_pg_state.guest_cr0 = CR0_PE;
- PrintDebug("Created\n");
+ // Perform any additional yielding needed for time adjustment
+ v3_adjust_time(info);
- vmx_ret |= check_vmcs_write(VMCS_CR0_MASK, (CR0_PE | CR0_PG) );
- vmx_ret |= check_vmcs_write(VMCS_CR0_READ_SHDW, info->shdw_pg_state.guest_cr0);
- vmx_ret |= check_vmcs_write(VMCS_CR4_MASK, CR4_VMXE);
+ // Update timer devices prior to entering VM.
+ v3_update_timers(info);
- info->ctrl_regs.cr3 = info->direct_map_pt;
+ // disable global interrupts for vm state transition
+ v3_disable_ints();
- // vmx_data->pinbased_ctrls |= NMI_EXIT;
- /* Add unconditional I/O and CR exits */
- vmx_data->pri_procbased_ctrls |= UNCOND_IO_EXIT
- | CR3_LOAD_EXIT
- | CR3_STORE_EXIT;
-
- vmx_data->exit_ctrls |= HOST_ADDR_SPACE_SIZE;
+ 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;
}
- struct v3_segment * seg_reg = (struct v3_segment *)&(info->segments);
- int i;
- for(i=0; i < 10; i++)
- {
- seg_reg[i].selector = 3<<3;
- seg_reg[i].limit = 0xffff;
- seg_reg[i].base = 0x0;
+ v3_vmx_restore_vmcs(info);
+
+
+#ifdef CONFIG_SYMCALL
+ if (info->sym_core_state.symcall_state.sym_call_active == 0) {
+ update_irq_entry_state(info);
}
- info->segments.cs.selector = 2<<3;
-
- /* Set only the segment registers */
- for(i=0; i < 6; i++) {
- seg_reg[i].limit = 0xfffff;
- seg_reg[i].granularity = 1;
- seg_reg[i].type = 3;
- seg_reg[i].system = 1;
- seg_reg[i].dpl = 0;
- seg_reg[i].present = 1;
- seg_reg[i].db = 1;
+#else
+ update_irq_entry_state(info);
+#endif
+
+ {
+ addr_t guest_cr3;
+ vmcs_read(VMCS_GUEST_CR3, &guest_cr3);
+ vmcs_write(VMCS_GUEST_CR3, guest_cr3);
}
- info->segments.cs.type = 0xb;
- info->segments.ldtr.selector = 0x20;
- info->segments.ldtr.type = 2;
- info->segments.ldtr.system = 0;
- info->segments.ldtr.present = 1;
- info->segments.ldtr.granularity = 0;
-
- /* Setup IO map */
- (void) v3_init_vmx_io_map(info);
- (void) v3_init_vmx_msr_map(info);
+ // Perform last-minute time bookkeeping prior to entering the VM
+ v3_time_enter_vm(info);
- /************* Map in GDT and vmxassist *************/
+ tsc_offset_high = (uint32_t)((v3_tsc_host_offset(&info->time_state) >> 32) & 0xffffffff);
+ tsc_offset_low = (uint32_t)(v3_tsc_host_offset(&info->time_state) & 0xffffffff);
+ check_vmcs_write(VMCS_TSC_OFFSET_HIGH, tsc_offset_high);
+ check_vmcs_write(VMCS_TSC_OFFSET, tsc_offset_low);
- uint64_t gdt[] __attribute__ ((aligned(32))) = {
- 0x0000000000000000ULL, /* 0x00: reserved */
- 0x0000830000000000ULL, /* 0x08: 32-bit TSS */
- //0x0000890000000000ULL, /* 0x08: 32-bit TSS */
- 0x00CF9b000000FFFFULL, /* 0x10: CS 32-bit */
- 0x00CF93000000FFFFULL, /* 0x18: DS 32-bit */
- 0x000082000000FFFFULL, /* 0x20: LDTR 32-bit */
- };
-#define VMXASSIST_GDT 0x10000
- addr_t vmxassist_gdt = 0;
- if(guest_pa_to_host_va(info, VMXASSIST_GDT, &vmxassist_gdt) == -1) {
- PrintError("Could not find VMXASSIST GDT destination\n");
- return -1;
+ 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));
}
- memcpy((void*)vmxassist_gdt, gdt, sizeof(uint64_t) * 5);
-
- info->segments.gdtr.base = VMXASSIST_GDT;
+
+ // PrintDebug("VMX Exit: ret=%d\n", ret);
-#define VMXASSIST_TSS 0x40000
- uint64_t vmxassist_tss = VMXASSIST_TSS;
- gdt[0x08 / sizeof(gdt[0])] |=
- ((vmxassist_tss & 0xFF000000) << (56 - 24)) |
- ((vmxassist_tss & 0x00FF0000) << (32 - 16)) |
- ((vmxassist_tss & 0x0000FFFF) << (16)) |
- (8392 - 1);
-
- info->segments.tr.selector = 0x08;
- info->segments.tr.base = vmxassist_tss;
-
- //info->segments.tr.type = 0x9;
- info->segments.tr.type = 0x3;
- info->segments.tr.system = 0;
- info->segments.tr.present = 1;
- info->segments.tr.granularity = 0;
+ if (ret != VMX_SUCCESS) {
+ uint32_t error = 0;
-
-#define VMXASSIST_START 0x000d0000
- extern uint8_t v3_vmxassist_start[];
- extern uint8_t v3_vmxassist_end[];
+ vmcs_read(VMCS_INSTR_ERR, &error);
+ PrintError("VMENTRY Error: %d\n", error);
- addr_t vmxassist_dst = 0;
- if(guest_pa_to_host_va(info, VMXASSIST_START, &vmxassist_dst) == -1) {
- PrintError("Could not find VMXASSIST destination\n");
- return -1;
+ return -1;
}
- memcpy((void*)vmxassist_dst, v3_vmxassist_start, v3_vmxassist_end - v3_vmxassist_start);
-
- /*** 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);
+ // Immediate exit from VM time bookkeeping
+ v3_time_exit_vm(info);
- vmx_ret |= check_vmcs_write(VMCS_GUEST_DR7, 0x400);
+ info->num_exits++;
- vmx_ret |= check_vmcs_write(VMCS_LINK_PTR, (addr_t)0xffffffffffffffffULL);
-
- if(v3_update_vmcs_ctrl_fields(info)) {
- PrintError("Could not write control fields!\n");
- return -1;
- }
-
- if(v3_update_vmcs_host_state(info)) {
- PrintError("Could not write host state\n");
- return -1;
+ /* Update guest state */
+ v3_vmx_save_vmcs(info);
+
+ // info->cpl = info->segments.cs.selector & 0x3;
+
+ info->mem_mode = v3_get_vm_mem_mode(info);
+ info->cpu_mode = v3_get_vm_cpu_mode(info);
+
+
+ check_vmcs_read(VMCS_EXIT_INSTR_LEN, &(exit_info.instr_len));
+ check_vmcs_read(VMCS_EXIT_INSTR_INFO, &(exit_info.instr_info));
+ check_vmcs_read(VMCS_EXIT_REASON, &(exit_info.exit_reason));
+ check_vmcs_read(VMCS_EXIT_QUAL, &(exit_info.exit_qual));
+ check_vmcs_read(VMCS_EXIT_INT_INFO, &(exit_info.int_info));
+ check_vmcs_read(VMCS_EXIT_INT_ERR, &(exit_info.int_err));
+ check_vmcs_read(VMCS_GUEST_LINEAR_ADDR, &(exit_info.guest_linear_addr));
+
+ //PrintDebug("VMX Exit taken, id-qual: %u-%lu\n", exit_info.exit_reason, exit_info.exit_qual);
+
+ exit_log[info->num_exits % 10] = exit_info;
+
+
+#ifdef CONFIG_SYMCALL
+ if (info->sym_core_state.symcall_state.sym_call_active == 0) {
+ update_irq_exit_state(info);
}
+#else
+ 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);
- if(v3_update_vmcs_guest_state(info) != VMX_SUCCESS) {
- PrintError("Writing guest state failed!\n");
- return -1;
+#ifdef CONFIG_DEBUG_INTERRUPTS
+ PrintDebug("Interrupts available again! (RIP=%llx)\n", info->rip);
+#endif
}
- v3_print_vmcs();
+ // reenable global interrupts after vm exit
+ v3_enable_ints();
- vmx_data->state = VMXASSIST_STARTUP;
+ // Conditionally yield the CPU if the timeslice has expired
+ v3_yield_cond(info);
- v3_post_config_guest(info, config_ptr);
+ if (v3_handle_vmx_exit(info, &exit_info) == -1) {
+ PrintError("Error in VMX exit handler\n");
+ return -1;
+ }
return 0;
}
-static int start_vmx_guest(struct guest_info* info) {
- uint32_t error = 0;
- int ret = 0;
+int v3_start_vmx_guest(struct guest_info * info) {
+
+ PrintDebug("Starting VMX core %u\n", info->cpu_id);
+
+ if (info->cpu_id == 0) {
+ info->core_run_state = CORE_RUNNING;
+ info->vm_info->run_state = VM_RUNNING;
+ } else {
+
+ PrintDebug("VMX core %u: Waiting for core initialization\n", info->cpu_id);
+
+ while (info->core_run_state == CORE_STOPPED) {
+ v3_yield(info);
+ //PrintDebug("VMX core %u: still waiting for INIT\n",info->cpu_id);
+ }
+
+ PrintDebug("VMX core %u initialized\n", info->cpu_id);
+ }
+
+
+ PrintDebug("VMX core %u: I am starting at CS=0x%x (base=0x%p, limit=0x%x), RIP=0x%p\n",
+ info->cpu_id, info->segments.cs.selector, (void *)(info->segments.cs.base),
+ info->segments.cs.limit, (void *)(info->rip));
+
+
+ PrintDebug("VMX core %u: Launching VMX VM\n", info->cpu_id);
+
+ v3_start_time(info);
+
+ while (1) {
+
+ if (info->vm_info->run_state == VM_STOPPED) {
+ info->core_run_state = CORE_STOPPED;
+ break;
+ }
+
+ if (v3_vmx_enter(info) == -1) {
+ v3_print_vmcs();
+ print_exit_log(info);
+ return -1;
+ }
- PrintDebug("Attempting VMLAUNCH\n");
- ret = v3_vmx_vmlaunch(&(info->vm_regs), info);
- if (ret != VMX_SUCCESS) {
- vmcs_read(VMCS_INSTR_ERR, &error);
- PrintError("VMLAUNCH failed: %d\n", error);
- v3_print_vmcs();
+ if (info->vm_info->run_state == VM_STOPPED) {
+ info->core_run_state = CORE_STOPPED;
+ break;
+ }
+/*
+ if ((info->num_exits % 5000) == 0) {
+ V3_Print("VMX Exit number %d\n", (uint32_t)info->num_exits);
+ }
+*/
}
- PrintDebug("Returned from VMLAUNCH ret=%d(0x%x)\n", ret, ret);
- return -1;
+ return 0;
}
+
+
+#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;
- addr_t eax = 0, ebx = 0, ecx = 0, edx = 0;
+ uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0;
v3_cpuid(0x1, &eax, &ebx, &ecx, &edx);
- PrintDebug("ECX: %p\n", (void*)ecx);
+ PrintDebug("ECX: 0x%x\n", ecx);
if (ecx & CPUID_1_ECX_VTXFLAG) {
v3_get_msr(VMX_FEATURE_CONTROL_MSR, &(feature_msr.hi), &(feature_msr.lo));
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(struct v3_ctrl_ops * vm_ops) {
- extern v3_cpu_arch_t v3_cpu_type;
- struct v3_msr tmp_msr;
- uint64_t ret=0;
- v3_get_msr(VMX_CR4_FIXED0_MSR,&(tmp_msr.hi),&(tmp_msr.lo));
-
- __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[];
+
+ 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__ (
- "movq %%cr0, %%rbx; "
- "orq $0x00000020,%%rbx; "
- "movq %%rbx, %%cr0;"
- :
- :
- : "%rbx"
- );
- //
- // Should check and return Error here....
+
+
+ enable_vmx();
// Setup VMXON Region
- vmxon_ptr_phys = allocate_vmcs();
- PrintDebug("VMXON pointer: 0x%p\n", (void*)vmxon_ptr_phys);
+ host_vmcs_ptrs[cpu_id] = allocate_vmcs();
+
+ PrintDebug("VMXON pointer: 0x%p\n", (void *)host_vmcs_ptrs[cpu_id]);
- if (v3_enable_vmx(vmxon_ptr_phys) == VMX_SUCCESS) {
+ if (vmx_on(host_vmcs_ptrs[cpu_id]) == VMX_SUCCESS) {
PrintDebug("VMX Enabled\n");
} else {
PrintError("VMX initialization failure\n");
return;
}
-
+
- if (has_vmx_nested_paging() == 1) {
- v3_cpu_type = V3_VMX_EPT_CPU;
- } else {
- v3_cpu_type = V3_VMX_CPU;
- }
+ v3_cpu_types[cpu_id] = V3_VMX_CPU;
- // Setup the VMX specific vmm operations
- vm_ops->init_guest = &init_vmx_guest;
- vm_ops->start_guest = &start_vmx_guest;
- vm_ops->has_nested_paging = &has_vmx_nested_paging;
}
+
+void v3_deinit_vmx_cpu(int cpu_id) {
+ extern v3_cpu_arch_t v3_cpu_types[];
+ v3_cpu_types[cpu_id] = V3_INVALID_CPU;
+ V3_FreePages((void *)host_vmcs_ptrs[cpu_id], 1);
+}
