X-Git-Url: http://v3vee.org/palacios/gitweb/gitweb.cgi?a=blobdiff_plain;f=palacios%2Fsrc%2Fpalacios%2Fsvm.c;h=89e33d35453311429700c820789c200d90e47a8c;hb=e02cc15e36e8f5d7ee66390986351748dd9a16a1;hp=0a9dc556aed4132f2245e8128a098a34063ee44f;hpb=e3a2ad0c26462c45cf0a7f1ccc8f0042abbbd776;p=palacios.git diff --git a/palacios/src/palacios/svm.c b/palacios/src/palacios/svm.c index 0a9dc55..89e33d3 100644 --- a/palacios/src/palacios/svm.c +++ b/palacios/src/palacios/svm.c @@ -36,22 +36,43 @@ #include -#include - #include #include -#include +#include + +#include + + +#ifndef CONFIG_DEBUG_SVM +#undef PrintDebug +#define PrintDebug(fmt, args...) +#endif + + +uint32_t v3_last_exit; + +// 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); +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 *)V3_VAddr(V3_AllocPages(1)); + vmcb_t * vmcb_page = NULL; + addr_t vmcb_pa = (addr_t)V3_AllocPages(1); + + if ((void *)vmcb_pa == NULL) { + PrintError("Error allocating VMCB\n"); + return NULL; + } + + vmcb_page = (vmcb_t *)V3_VAddr((void *)vmcb_pa); memset(vmcb_page, 0, 4096); @@ -60,22 +81,13 @@ static vmcb_t * Allocate_VMCB() { -static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info *vm_info) { +static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info * core) { 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; - - 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->svm_instrs.VMMCALL = 1; ctrl_area->svm_instrs.VMLOAD = 1; @@ -83,16 +95,21 @@ static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info *vm_info) { 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; + +#ifdef CONFIG_TIME_VIRTUALIZE_TSC + ctrl_area->instrs.RDTSC = 1; + ctrl_area->svm_instrs.RDTSCP = 1; +#endif + // guest_state->cr0 = 0x00000001; // PE /* @@ -113,88 +130,87 @@ static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info *vm_info) { ctrl_area->instrs.NMI = 1; - ctrl_area->instrs.SMI = 1; + ctrl_area->instrs.SMI = 0; // allow SMIs to run in guest ctrl_area->instrs.INIT = 1; ctrl_area->instrs.PAUSE = 1; ctrl_area->instrs.shutdown_evts = 1; - vm_info->vm_regs.rdx = 0x00000f00; - guest_state->cr0 = 0x60000010; + /* DEBUG FOR RETURN CODE */ + ctrl_area->exit_code = 1; - guest_state->cs.selector = 0xf000; - guest_state->cs.limit = 0xffff; - guest_state->cs.base = 0x0000000f0000LL; - guest_state->cs.attrib.raw = 0xf3; + /* Setup Guest Machine state */ + core->vm_regs.rsp = 0x00; + core->rip = 0xfff0; - /* DEBUG FOR RETURN CODE */ - ctrl_area->exit_code = 1; + core->vm_regs.rdx = 0x00000f00; - struct vmcb_selector *segregs [] = {&(guest_state->ss), &(guest_state->ds), - &(guest_state->es), &(guest_state->fs), - &(guest_state->gs), NULL}; + core->cpl = 0; - for ( i = 0; segregs[i] != NULL; i++) { - struct vmcb_selector * seg = segregs[i]; - - seg->selector = 0x0000; - // seg->base = seg->selector << 4; - seg->base = 0x00000000; - seg->attrib.raw = 0xf3; - seg->limit = ~0u; - } + core->ctrl_regs.rflags = 0x00000002; // The reserved bit is always 1 + core->ctrl_regs.cr0 = 0x60010010; // Set the WP flag so the memory hooks work in real-mode + core->ctrl_regs.efer |= EFER_MSR_svm_enable; - 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; - guest_state->dr6 = 0x00000000ffff0ff0LL; - guest_state->dr7 = 0x0000000000000400LL; + core->segments.cs.selector = 0xf000; + core->segments.cs.limit = 0xffff; + core->segments.cs.base = 0x0000000f0000LL; + + // (raw attributes = 0xf3) + core->segments.cs.type = 0x3; + core->segments.cs.system = 0x1; + core->segments.cs.dpl = 0x3; + core->segments.cs.present = 1; - if ( !RB_EMPTY_ROOT(&(vm_info->io_map)) ) { - struct v3_io_hook * iter; - struct rb_node * io_node = v3_rb_first(&(vm_info->io_map)); - addr_t io_port_bitmap; - int i = 0; - - io_port_bitmap = (addr_t)V3_VAddr(V3_AllocPages(3)); - memset((uchar_t*)io_port_bitmap, 0, PAGE_SIZE * 3); - ctrl_area->IOPM_BASE_PA = (addr_t)V3_PAddr((void *)io_port_bitmap); - //PrintDebug("Setting up IO Map at 0x%x\n", io_port_bitmap); + struct v3_segment * segregs [] = {&(core->segments.ss), &(core->segments.ds), + &(core->segments.es), &(core->segments.fs), + &(core->segments.gs), NULL}; - do { - iter = rb_entry(io_node, struct v3_io_hook, tree_node); + 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; + } - ushort_t port = iter->port; - uchar_t * bitmap = (uchar_t *)io_port_bitmap; - //PrintDebug("%d: Hooking Port %d\n", i, port); + core->segments.gdtr.limit = 0x0000ffff; + core->segments.gdtr.base = 0x0000000000000000LL; + core->segments.idtr.limit = 0x0000ffff; + core->segments.idtr.base = 0x0000000000000000LL; - bitmap += (port / 8); - // PrintDebug("Setting Bit for port 0x%x\n", port); - *bitmap |= 1 << (port % 8); + core->segments.ldtr.selector = 0x0000; + core->segments.ldtr.limit = 0x0000ffff; + core->segments.ldtr.base = 0x0000000000000000LL; + core->segments.tr.selector = 0x0000; + core->segments.tr.limit = 0x0000ffff; + core->segments.tr.base = 0x0000000000000000LL; - i++; - } while ((io_node = v3_rb_next(io_node))); - //PrintDebugMemDump((uchar_t*)io_port_bitmap, PAGE_SIZE *2); + core->dbg_regs.dr6 = 0x00000000ffff0ff0LL; + core->dbg_regs.dr7 = 0x0000000000000400LL; - ctrl_area->instrs.IOIO_PROT = 1; - } + + ctrl_area->IOPM_BASE_PA = (addr_t)V3_PAddr(core->vm_info->io_map.arch_data); + ctrl_area->instrs.IOIO_PROT = 1; + + ctrl_area->MSRPM_BASE_PA = (addr_t)V3_PAddr(core->vm_info->msr_map.arch_data); + ctrl_area->instrs.MSR_PROT = 1; PrintDebug("Exiting on interrupts\n"); @@ -202,22 +218,31 @@ static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info *vm_info) { ctrl_area->instrs.INTR = 1; - if (vm_info->shdw_pg_mode == SHADOW_PAGING) { + v3_hook_msr(core->vm_info, EFER_MSR, + &v3_handle_efer_read, + &v3_handle_efer_write, + core); + + if (core->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) { + if (v3_init_passthrough_pts(core) == -1) { PrintError("Could not initialize passthrough page tables\n"); return ; } - vm_info->shdw_pg_state.guest_cr0 = 0x0000000000000010LL; + core->shdw_pg_state.guest_cr0 = 0x0000000000000010LL; PrintDebug("Created\n"); - guest_state->cr3 = vm_info->direct_map_pt; + core->ctrl_regs.cr0 |= 0x80000000; + core->ctrl_regs.cr3 = core->direct_map_pt; ctrl_area->cr_reads.cr0 = 1; ctrl_area->cr_writes.cr0 = 1; @@ -226,26 +251,17 @@ static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info *vm_info) { ctrl_area->cr_reads.cr3 = 1; ctrl_area->cr_writes.cr3 = 1; - vm_info->guest_efer.value = 0x0LL; - 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; - /* JRL: This is a performance killer, and a simplistic solution */ - /* We need to fix this */ - ctrl_area->TLB_CONTROL = 1; - guest_state->g_pat = 0x7040600070406ULL; - guest_state->cr0 |= 0x80000000; - } else if (vm_info->shdw_pg_mode == NESTED_PAGING) { + + } else if (core->shdw_pg_mode == NESTED_PAGING) { // Flush the TLB on entries/exits ctrl_area->TLB_CONTROL = 1; ctrl_area->guest_ASID = 1; @@ -256,197 +272,408 @@ static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info *vm_info) { PrintDebug("NP_Enable at 0x%p\n", (void *)&(ctrl_area->NP_ENABLE)); // Set the Nested Page Table pointer - if (v3_init_passthrough_pts(vm_info) == -1) { + if (v3_init_passthrough_pts(core) == -1) { PrintError("Could not initialize Nested page tables\n"); return ; } - ctrl_area->N_CR3 = vm_info->direct_map_pt; + ctrl_area->N_CR3 = core->direct_map_pt; guest_state->g_pat = 0x7040600070406ULL; } + + /* tell the guest that we don't support SVM */ + v3_hook_msr(core->vm_info, SVM_VM_CR_MSR, + &v3_handle_vm_cr_read, + &v3_handle_vm_cr_write, + core); +} - if (vm_info->msr_map.num_hooks > 0) { - PrintDebug("Hooking %d msrs\n", vm_info->msr_map.num_hooks); - ctrl_area->MSRPM_BASE_PA = v3_init_svm_msr_map(vm_info); - ctrl_area->instrs.MSR_PROT = 1; + +int v3_init_svm_vmcb(struct guest_info * core, v3_vm_class_t vm_class) { + + PrintDebug("Allocating VMCB\n"); + core->vmm_data = (void *)Allocate_VMCB(); + + if (core->vmm_data == NULL) { + PrintError("Could not allocate VMCB, Exiting...\n"); + return -1; } - /* Safety locations for fs/gs */ - // vm_info->fs = 0; - // vm_info->gs = 0; + if (vm_class == V3_PC_VM) { + PrintDebug("Initializing VMCB (addr=%p)\n", (void *)core->vmm_data); + Init_VMCB_BIOS((vmcb_t*)(core->vmm_data), core); + } else { + PrintError("Invalid VM class\n"); + return -1; + } + + return 0; } -static int init_svm_guest(struct guest_info *info, struct v3_vm_config * config_ptr) { - v3_config_guest(info, config_ptr); +int v3_deinit_svm_vmcb(struct guest_info * core) { + V3_FreePages(V3_PAddr(core->vmm_data), 1); + return 0; +} - PrintDebug("Allocating VMCB\n"); - info->vmm_data = (void*)Allocate_VMCB(); - v3_config_devices(info, config_ptr); +static int update_irq_exit_state(struct guest_info * info) { + vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data)); - PrintDebug("Initializing VMCB (addr=%p)\n", (void *)info->vmm_data); - Init_VMCB_BIOS((vmcb_t*)(info->vmm_data), info); - + // Fix for QEMU bug using EVENTINJ as an internal cache + guest_ctrl->EVENTINJ.valid = 0; - - info->run_state = VM_STOPPED; + if ((info->intr_core_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_core_state.irq_vector); +#endif + + info->intr_core_state.irq_started = 1; + info->intr_core_state.irq_pending = 0; + + v3_injecting_intr(info, info->intr_core_state.irq_vector, V3_EXTERNAL_IRQ); + } + + if ((info->intr_core_state.irq_started == 1) && (guest_ctrl->exit_int_info.valid == 0)) { +#ifdef CONFIG_DEBUG_INTERRUPTS + PrintDebug("Interrupt %d taken by guest\n", info->intr_core_state.irq_vector); +#endif + + // Interrupt was taken fully vectored + info->intr_core_state.irq_started = 0; + + } else if ((info->intr_core_state.irq_started == 1) && (guest_ctrl->exit_int_info.valid == 1)) { +#ifdef CONFIG_DEBUG_INTERRUPTS + PrintDebug("EXIT INT INFO is set (vec=%d)\n", guest_ctrl->exit_int_info.vector); +#endif + } - // info->rip = 0; - - 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; } +static int update_irq_entry_state(struct guest_info * info) { + vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data)); -// 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)); - uint_t num_exits = 0; + if (info->intr_core_state.irq_pending == 0) { + guest_ctrl->guest_ctrl.V_IRQ = 0; + guest_ctrl->guest_ctrl.V_INTR_VECTOR = 0; + } + + 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 + v3_injecting_excp(info, excp); + } else if (info->intr_core_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_core_state.irq_vector; + guest_ctrl->guest_ctrl.V_IGN_TPR = 1; + guest_ctrl->guest_ctrl.V_INTR_PRIO = 0xf; - PrintDebug("Launching SVM VM (vmcb=%p)\n", (void *)info->vmm_data); - //PrintDebugVMCB((vmcb_t*)(info->vmm_data)); + } 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_core_state.irq_pending = 1; + info->intr_core_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; + } + + } + + return 0; +} + + +/* + * 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)); + addr_t exit_code = 0, exit_info1 = 0, exit_info2 = 0; + + // Conditionally yield the CPU if the timeslice has expired + v3_yield_cond(info); + + // Perform any additional yielding needed for time adjustment + v3_adjust_time(info); + + // Update timer devices prior to entering VM. + v3_update_timers(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; - info->run_state = VM_RUNNING; + guest_state->cpl = info->cpl; + + v3_set_vmcb_segments((vmcb_t*)(info->vmm_data), &(info->segments)); + + guest_state->rax = info->vm_regs.rax; + guest_state->rip = info->rip; + guest_state->rsp = info->vm_regs.rsp; + +#ifdef CONFIG_SYMCALL + if (info->sym_core_state.symcall_state.sym_call_active == 0) { + update_irq_entry_state(info); + } +#else + update_irq_entry_state(info); +#endif + + + /* ** */ + + /* + PrintDebug("SVM Entry to CS=%p rip=%p...\n", + (void *)(addr_t)info->segments.cs.base, + (void *)(addr_t)info->rip); + */ + +#ifdef CONFIG_SYMCALL + if (info->sym_core_state.symcall_state.sym_call_active == 1) { + if (guest_ctrl->guest_ctrl.V_IRQ == 1) { + V3_Print("!!! Injecting Interrupt during Sym call !!!\n"); + } + } +#endif + + v3_time_enter_vm(info); + guest_ctrl->TSC_OFFSET = v3_tsc_host_offset(&info->time_state); + + //V3_Print("Calling v3_svm_launch\n"); + + v3_svm_launch((vmcb_t *)V3_PAddr(info->vmm_data), &(info->vm_regs), (vmcb_t *)host_vmcbs[info->host_cpu_id]); + + //V3_Print("SVM Returned: Exit Code: %x, guest_rip=%lx\n", (uint32_t)(guest_ctrl->exit_code), (unsigned long)guest_state->rip); + + v3_last_exit = (uint32_t)(guest_ctrl->exit_code); + + // Immediate exit from VM time bookkeeping + v3_time_exit_vm(info); + + info->num_exits++; + + // 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; + + 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; - while (1) { - ullong_t tmp_tsc; - + 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); + /* ** */ -#define MSR_STAR 0xc0000081 -#define MSR_LSTAR 0xc0000082 -#define MSR_CSTAR 0xc0000083 -#define MSR_SF_MASK 0xc0000084 -#define MSR_GS_BASE 0xc0000101 -#define MSR_KERNGS_BASE 0xc0000102 + // save exit info here + exit_code = guest_ctrl->exit_code; + exit_info1 = guest_ctrl->exit_info1; + exit_info2 = guest_ctrl->exit_info2; - struct v3_msr host_cstar; - struct v3_msr host_star; - struct v3_msr host_lstar; - struct v3_msr host_syscall_mask; - struct v3_msr host_gs_base; - struct v3_msr host_kerngs_base; +#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 -/* v3_enable_ints(); */ -/* v3_clgi(); */ + // reenable global interrupts after vm exit + v3_stgi(); - /* - PrintDebug("SVM Entry to CS=%p rip=%p...\n", - (void *)(addr_t)info->segments.cs.base, - (void *)(addr_t)info->rip); - */ + + // Conditionally yield the CPU if the timeslice has expired + v3_yield_cond(info); - v3_get_msr(MSR_STAR, &(host_star.hi), &(host_star.lo)); - v3_get_msr(MSR_LSTAR, &(host_lstar.hi), &(host_lstar.lo)); - v3_get_msr(MSR_CSTAR, &(host_cstar.hi), &(host_cstar.lo)); - v3_get_msr(MSR_SF_MASK, &(host_syscall_mask.hi), &(host_syscall_mask.lo)); - v3_get_msr(MSR_GS_BASE, &(host_gs_base.hi), &(host_gs_base.lo)); - v3_get_msr(MSR_KERNGS_BASE, &(host_kerngs_base.hi), &(host_kerngs_base.lo)); + if (v3_handle_svm_exit(info, exit_code, exit_info1, exit_info2) != 0) { + PrintError("Error in SVM exit handler\n"); + 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), &(info->fs), &(info->gs)); - v3_svm_launch((vmcb_t*)V3_PAddr(info->vmm_data), &(info->vm_regs)); - - rdtscll(tmp_tsc); - - v3_set_msr(MSR_STAR, host_star.hi, host_star.lo); - v3_set_msr(MSR_LSTAR, host_lstar.hi, host_lstar.lo); - v3_set_msr(MSR_CSTAR, host_cstar.hi, host_cstar.lo); - v3_set_msr(MSR_SF_MASK, host_syscall_mask.hi, host_syscall_mask.lo); - v3_set_msr(MSR_GS_BASE, host_gs_base.hi, host_gs_base.lo); - v3_set_msr(MSR_KERNGS_BASE, host_kerngs_base.hi, host_kerngs_base.lo); - - //PrintDebug("SVM Returned\n"); + + 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)); - v3_update_time(info, tmp_tsc - info->time_state.cached_host_tsc); - num_exits++; + PrintDebug("Starting SVM core %u\n", info->cpu_id); - //PrintDebug("Turning on global interrupts\n"); - v3_stgi(); - v3_clgi(); - - if ((num_exits % 5000) == 0) { - PrintDebug("SVM Exit number %d\n", num_exits); + if (info->cpu_id == 0) { + info->core_run_state = CORE_RUNNING; + info->vm_info->run_state = VM_RUNNING; + } else { + PrintDebug("SVM core %u: Waiting for core initialization\n", info->cpu_id); - if (info->enable_profiler) { - v3_print_profile(info); - } + while (info->core_run_state == CORE_STOPPED) { + v3_yield(info); + //PrintDebug("SVM core %u: still waiting for INIT\n",info->cpu_id); } + PrintDebug("SVM core %u initialized\n", info->cpu_id); + } - - if (v3_handle_svm_exit(info) != 0) { - 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; - - PrintDebug("SVM ERROR!!\n"); - - PrintDebug("RIP: %p\n", (void *)(addr_t)(guest_state->rip)); + PrintDebug("SVM 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)); - linear_addr = get_addr_linear(info, guest_state->rip, &(info->segments.cs)); + PrintDebug("SVM core %u: Launching SVM VM (vmcb=%p)\n", info->cpu_id, (void *)info->vmm_data); + //PrintDebugVMCB((vmcb_t*)(info->vmm_data)); + + v3_start_time(info); - PrintDebug("RIP Linear: %p\n", (void *)linear_addr); - v3_print_segments(info); - v3_print_ctrl_regs(info); - if (info->shdw_pg_mode == SHADOW_PAGING) { - PrintDebug("Shadow Paging Guest Registers:\n"); - PrintDebug("\tGuest CR0=%p\n", (void *)(addr_t)(info->shdw_pg_state.guest_cr0)); - PrintDebug("\tGuest CR3=%p\n", (void *)(addr_t)(info->shdw_pg_state.guest_cr3)); - // efer - // CR4 - } - v3_print_GPRs(info); + while (1) { - PrintDebug("SVM Exit Code: %p\n", (void *)(addr_t)guest_ctrl->exit_code); - - PrintDebug("exit_info1 low = 0x%.8x\n", *(uint_t*)&(guest_ctrl->exit_info1)); - PrintDebug("exit_info1 high = 0x%.8x\n", *(uint_t *)(((uchar_t *)&(guest_ctrl->exit_info1)) + 4)); - - PrintDebug("exit_info2 low = 0x%.8x\n", *(uint_t*)&(guest_ctrl->exit_info2)); - PrintDebug("exit_info2 high = 0x%.8x\n", *(uint_t *)(((uchar_t *)&(guest_ctrl->exit_info2)) + 4)); - + if (info->vm_info->run_state == VM_STOPPED) { + info->core_run_state = CORE_STOPPED; + break; + } + + 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->vm_info->run_state = VM_ERROR; + + V3_Print("SVM core %u: SVM ERROR!!\n", info->cpu_id); + + v3_print_guest_state(info); + + V3_Print("SVM core %u: SVM Exit Code: %p\n", info->cpu_id, (void *)(addr_t)guest_ctrl->exit_code); + + V3_Print("SVM core %u: exit_info1 low = 0x%.8x\n", info->cpu_id, *(uint_t*)&(guest_ctrl->exit_info1)); + V3_Print("SVM core %u: exit_info1 high = 0x%.8x\n", info->cpu_id, *(uint_t *)(((uchar_t *)&(guest_ctrl->exit_info1)) + 4)); + + V3_Print("SVM core %u: exit_info2 low = 0x%.8x\n", info->cpu_id, *(uint_t*)&(guest_ctrl->exit_info2)); + V3_Print("SVM core %u: exit_info2 high = 0x%.8x\n", info->cpu_id, *(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); + v3_gpa_to_hva(info, linear_addr, &host_addr); } else if (info->mem_mode == VIRTUAL_MEM) { - guest_va_to_host_va(info, linear_addr, &host_addr); + v3_gva_to_hva(info, linear_addr, &host_addr); } + + V3_Print("SVM core %u: Host Address of rip = 0x%p\n", info->cpu_id, (void *)host_addr); + + V3_Print("SVM core %u: Instr (15 bytes) at %p:\n", info->cpu_id, (void *)host_addr); + v3_dump_mem((uint8_t *)host_addr, 15); + + v3_print_stack(info); + break; + } - PrintDebug("Host Address of rip = 0x%p\n", (void *)host_addr); - - PrintDebug("Instr (15 bytes) at %p:\n", (void *)host_addr); - PrintTraceMemDump((uchar_t *)host_addr, 15); + if (info->vm_info->run_state == VM_STOPPED) { + info->core_run_state = CORE_STOPPED; break; } + + +/* + if ((info->num_exits % 5000) == 0) { + V3_Print("SVM Exit number %d\n", (uint32_t)info->num_exits); + } +*/ + } + + // Need to take down the other cores on error... + return 0; } @@ -457,16 +684,15 @@ static int start_svm_guest(struct guest_info *info) { /* Checks machine SVM capability */ /* Implemented from: AMD Arch Manual 3, sect 15.4 */ int v3_is_svm_capable() { - // Dinda uint_t vm_cr_low = 0, vm_cr_high = 0; - addr_t eax = 0, ebx = 0, ecx = 0, edx = 0; + uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0; - v3_cpuid(CPUID_FEATURE_IDS, &eax, &ebx, &ecx, &edx); + v3_cpuid(CPUID_EXT_FEATURE_IDS, &eax, &ebx, &ecx, &edx); - PrintDebug("CPUID_FEATURE_IDS_ecx=%p\n", (void *)ecx); + PrintDebug("CPUID_EXT_FEATURE_IDS_ecx=0x%x\n", ecx); - if ((ecx & CPUID_FEATURE_IDS_ecx_svm_avail) == 0) { - PrintDebug("SVM Not Available\n"); + 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); @@ -474,34 +700,27 @@ int v3_is_svm_capable() { 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) { - PrintDebug("SVM is available but is disabled.\n"); + V3_Print("SVM is available but is disabled.\n"); v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx); - PrintDebug("CPUID_FEATURE_IDS_edx=%p\n", (void *)edx); + PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_edx=0x%x\n", edx); if ((edx & CPUID_SVM_REV_AND_FEATURE_IDS_edx_svml) == 0) { - PrintDebug("SVM BIOS Disabled, not unlockable\n"); + V3_Print("SVM BIOS Disabled, not unlockable\n"); } else { - PrintDebug("SVM is locked with a key\n"); + V3_Print("SVM is locked with a key\n"); } return 0; } else { - PrintDebug("SVM is available and enabled.\n"); + V3_Print("SVM is available and enabled.\n"); v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx); - PrintDebug("CPUID_FEATURE_IDS_eax=%p\n", (void *)eax); - PrintDebug("CPUID_FEATURE_IDS_ebx=%p\n", (void *)ebx); - PrintDebug("CPUID_FEATURE_IDS_ecx=%p\n", (void *)ecx); - PrintDebug("CPUID_FEATURE_IDS_edx=%p\n", (void *)edx); - - - if ((edx & CPUID_SVM_REV_AND_FEATURE_IDS_edx_np) == 0) { - PrintDebug("SVM Nested Paging not supported\n"); - } else { - PrintDebug("SVM Nested Paging supported\n"); - } + 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; } @@ -509,69 +728,75 @@ int v3_is_svm_capable() { } static int has_svm_nested_paging() { - addr_t eax = 0, ebx = 0, ecx = 0, edx = 0; + uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0; v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx); - //PrintDebug("CPUID_FEATURE_IDS_edx=0x%x\n", edx); + //PrintDebug("CPUID_EXT_FEATURE_IDS_edx=0x%x\n", edx); if ((edx & CPUID_SVM_REV_AND_FEATURE_IDS_edx_np) == 0) { - PrintDebug("SVM Nested Paging not supported\n"); + V3_Print("SVM Nested Paging not supported\n"); return 0; } else { - PrintDebug("SVM Nested Paging supported\n"); + V3_Print("SVM Nested Paging supported\n"); return 1; } } - -void v3_init_SVM(struct v3_ctrl_ops * vmm_ops) { +void v3_init_svm_cpu(int cpu_id) { reg_ex_t msr; - void * host_state; - extern v3_cpu_arch_t v3_cpu_type; + 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); - PrintDebug("SVM Enabled\n"); - + V3_Print("SVM Enabled\n"); // Setup the host state save area - host_state = V3_AllocPages(4); - + host_vmcbs[cpu_id] = (addr_t)V3_AllocPages(4); /* 64-BIT-ISSUE */ // msr.e_reg.high = 0; - //msr.e_reg.low = (uint_t)host_state; - msr.r_reg = (addr_t)host_state; + //msr.e_reg.low = (uint_t)host_vmcb; + msr.r_reg = host_vmcbs[cpu_id]; - PrintDebug("Host State being saved at %p\n", (void *)(addr_t)host_state); + 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_type = V3_SVM_REV3_CPU; + v3_cpu_types[cpu_id] = V3_SVM_REV3_CPU; } else { - v3_cpu_type = V3_SVM_CPU; + v3_cpu_types[cpu_id] = V3_SVM_CPU; } - - // Setup the SVM specific vmm operations - vmm_ops->init_guest = &init_svm_guest; - vmm_ops->start_guest = &start_svm_guest; - vmm_ops->has_nested_paging = &has_svm_nested_paging; - - return; } +void v3_deinit_svm_cpu(int cpu_id) { + reg_ex_t msr; + extern v3_cpu_arch_t v3_cpu_types[]; + // reset SVM_VM_HSAVE_PA_MSR + // Does setting it to NULL disable?? + msr.r_reg = 0; + v3_set_msr(SVM_VM_HSAVE_PA_MSR, msr.e_reg.high, msr.e_reg.low); + // Disable SVM? + 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_cpu_types[cpu_id] = V3_INVALID_CPU; + V3_FreePages((void *)host_vmcbs[cpu_id], 4); + V3_Print("Host CPU %d host area freed, and SVM disabled\n", cpu_id); + return; +} @@ -616,126 +841,69 @@ void v3_init_SVM(struct v3_ctrl_ops * vmm_ops) { -/*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]; +#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; - seg->selector = 0x0000; - seg->base = seg->selector << 4; - seg->attrib.raw = 0xf3; - seg->limit = ~0u; - } - - if (vm_info.io_map.num_ports > 0) { - struct vmm_io_hook * iter; - addr_t io_port_bitmap; + __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; - io_port_bitmap = (addr_t)V3_AllocPages(3); - memset((uchar_t*)io_port_bitmap, 0, PAGE_SIZE * 3); + end = end_hi; + end <<= 32; + end += end_lo; - ctrl_area->IOPM_BASE_PA = io_port_bitmap; + 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; - //PrintDebug("Setting up IO Map at 0x%x\n", io_port_bitmap); + end = end_hi; + end <<= 32; + end += end_lo; - 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); + PrintDebug("VMLoad Cycle Latency: %d\n", (uint32_t)(end - start)); } + /* End Latency Test */ - - //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_pts_32(&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; - } - - - -} -*/ +#endif