X-Git-Url: http://v3vee.org/palacios/gitweb/gitweb.cgi?a=blobdiff_plain;f=palacios%2Fsrc%2Fpalacios%2Fvmx.c;h=c341982e37cbf584e6212fcab4f3c0b453a41f91;hb=a7bd09077efcd826376f9c49ab97f9f9a1fbbefe;hp=2eb4ae19a57568827eda4a696fa48f921652fc8a;hpb=c445b653bb1fe979e7146fb6b15923bc99277784;p=palacios.git diff --git a/palacios/src/palacios/vmx.c b/palacios/src/palacios/vmx.c index 2eb4ae1..c341982 100644 --- a/palacios/src/palacios/vmx.c +++ b/palacios/src/palacios/vmx.c @@ -34,6 +34,7 @@ #include #include #include +#include #ifdef V3_CONFIG_CHECKPOINT #include @@ -52,7 +53,7 @@ /* These fields contain the hardware feature sets supported by the local CPU */ static struct vmx_hw_info hw_info; -extern v3_cpu_arch_t v3_cpu_types[]; +extern v3_cpu_arch_t v3_mach_type; static addr_t host_vmcs_ptrs[V3_CONFIG_MAX_CPUS] = { [0 ... V3_CONFIG_MAX_CPUS - 1] = 0}; @@ -91,11 +92,17 @@ static int inline check_vmcs_read(vmcs_field_t field, void * val) { static addr_t allocate_vmcs() { + void *temp; struct vmcs_data * vmcs_page = NULL; PrintDebug("Allocating page\n"); - vmcs_page = (struct vmcs_data *)V3_VAddr(V3_AllocPages(1)); + temp = V3_AllocPages(1); + if (!temp) { + PrintError("Cannot allocate VMCS\n"); + return -1; + } + vmcs_page = (struct vmcs_data *)V3_VAddr(temp); memset(vmcs_page, 0, 4096); vmcs_page->revision = hw_info.basic_info.revision; @@ -104,13 +111,15 @@ static addr_t allocate_vmcs() { return (addr_t)V3_PAddr((void *)vmcs_page); } -/* +#if 0 static int debug_efer_read(struct guest_info * core, uint_t msr, struct v3_msr * src, void * priv_data) { struct v3_msr * efer = (struct v3_msr *)&(core->ctrl_regs.efer); - V3_Print("\n\nEFER READ\n"); + V3_Print("\n\nEFER READ (val = %p)\n", (void *)efer->value); v3_print_guest_state(core); + v3_print_vmcs(); + src->value = efer->value; return 0; @@ -118,25 +127,16 @@ static int debug_efer_read(struct guest_info * core, uint_t msr, struct v3_msr * static int debug_efer_write(struct guest_info * core, uint_t msr, struct v3_msr src, void * priv_data) { struct v3_msr * efer = (struct v3_msr *)&(core->ctrl_regs.efer); - V3_Print("\n\nEFER WRITE\n"); + V3_Print("\n\nEFER WRITE (old_val = %p) (new_val = %p)\n", (void *)efer->value, (void *)src.value); v3_print_guest_state(core); + v3_print_vmcs(); efer->value = src.value; - { - struct vmx_data * vmx_state = core->vmm_data; - - V3_Print("Trapping page faults and GPFs\n"); - vmx_state->excp_bmap.pf = 1; - vmx_state->excp_bmap.gp = 1; - - check_vmcs_write(VMCS_EXCP_BITMAP, vmx_state->excp_bmap.value); - } - return 0; } -*/ +#endif static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state) { @@ -170,8 +170,8 @@ static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state) vmx_state->sec_proc_ctrls.value = hw_info.sec_proc_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); + V3_Print("CR0 MSR: req_val=%p, req_mask=%p\n", (void *)(addr_t)hw_info.cr0.req_val, (void *)(addr_t)hw_info.cr0.req_mask); + V3_Print("CR4 MSR: req_val=%p, req_mask=%p\n", (void *)(addr_t)hw_info.cr4.req_val, (void *)(addr_t)hw_info.cr4.req_mask); @@ -184,9 +184,11 @@ static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state) /* Add external interrupts, NMI exiting, and virtual NMI */ vmx_state->pin_ctrls.nmi_exit = 1; + vmx_state->pin_ctrls.virt_nmi = 1; vmx_state->pin_ctrls.ext_int_exit = 1; + /* We enable the preemption timer by default to measure accurate guest time */ if (avail_pin_ctrls.active_preempt_timer) { V3_Print("VMX Preemption Timer is available\n"); @@ -194,10 +196,17 @@ static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state) vmx_state->exit_ctrls.save_preempt_timer = 1; } + // we want it to use this when halting vmx_state->pri_proc_ctrls.hlt_exit = 1; + // cpuid tells it that it does not have these instructions + vmx_state->pri_proc_ctrls.monitor_exit = 1; + vmx_state->pri_proc_ctrls.mwait_exit = 1; + // we don't need to handle a pause, although this is where + // we could pull out of a spin lock acquire or schedule to find its partner vmx_state->pri_proc_ctrls.pause_exit = 0; + vmx_state->pri_proc_ctrls.tsc_offset = 1; #ifdef V3_CONFIG_TIME_VIRTUALIZE_TSC vmx_state->pri_proc_ctrls.rdtsc_exit = 1; @@ -234,11 +243,16 @@ static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state) vmx_state->entry_ctrls.ld_pat = 1; /* Temporary GPF trap */ - // vmx_state->excp_bmap.gp = 1; + // vmx_state->excp_bmap.gp = 1; // Setup Guests initial PAT field vmx_ret |= check_vmcs_write(VMCS_GUEST_PAT, 0x0007040600070406LL); + // Capture CR8 mods so that we can keep the apic_tpr correct + vmx_state->pri_proc_ctrls.cr8_ld_exit = 1; + vmx_state->pri_proc_ctrls.cr8_str_exit = 1; + + /* Setup paging */ if (core->shdw_pg_mode == SHADOW_PAGING) { PrintDebug("Creating initial shadow page table\n"); @@ -251,7 +265,8 @@ static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state) #define CR0_PE 0x00000001 #define CR0_PG 0x80000000 #define CR0_WP 0x00010000 // To ensure mem hooks work - vmx_ret |= check_vmcs_write(VMCS_CR0_MASK, (CR0_PE | CR0_PG | CR0_WP)); +#define CR0_NE 0x00000020 + vmx_ret |= check_vmcs_write(VMCS_CR0_MASK, (CR0_PE | CR0_PG | CR0_WP | CR0_NE)); // Cause VM_EXIT whenever CR4.VMXE or CR4.PAE bits are written @@ -280,12 +295,13 @@ static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state) core); } else if ((core->shdw_pg_mode == NESTED_PAGING) && - (v3_cpu_types[core->pcpu_id] == V3_VMX_EPT_CPU)) { + (v3_mach_type == V3_VMX_EPT_CPU)) { #define CR0_PE 0x00000001 #define CR0_PG 0x80000000 #define CR0_WP 0x00010000 // To ensure mem hooks work - vmx_ret |= check_vmcs_write(VMCS_CR0_MASK, (CR0_PE | CR0_PG | CR0_WP)); +#define CR0_NE 0x00000020 + vmx_ret |= check_vmcs_write(VMCS_CR0_MASK, (CR0_PE | CR0_PG | CR0_WP | CR0_NE)); // vmx_state->pinbased_ctrls |= NMI_EXIT; @@ -319,7 +335,7 @@ static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state) v3_hook_msr(core->vm_info, EFER_MSR, NULL, NULL, NULL); } else if ((core->shdw_pg_mode == NESTED_PAGING) && - (v3_cpu_types[core->pcpu_id] == V3_VMX_EPT_UG_CPU)) { + (v3_mach_type == V3_VMX_EPT_UG_CPU)) { int i = 0; // For now we will assume that unrestricted guest mode is assured w/ EPT @@ -328,7 +344,7 @@ static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state) core->rip = 0xfff0; core->vm_regs.rdx = 0x00000f00; core->ctrl_regs.rflags = 0x00000002; // The reserved bit is always 1 - core->ctrl_regs.cr0 = 0x00000030; + core->ctrl_regs.cr0 = 0x60010030; core->ctrl_regs.cr4 = 0x00002010; // Enable VMX and PSE flag @@ -402,7 +418,9 @@ static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state) // Cause VM_EXIT whenever the CR4.VMXE bit is set vmx_ret |= check_vmcs_write(VMCS_CR4_MASK, CR4_VMXE); - +#define CR0_NE 0x00000020 + vmx_ret |= check_vmcs_write(VMCS_CR0_MASK, CR0_NE); + ((struct cr0_32 *)&(core->shdw_pg_state.guest_cr0))->ne = 1; if (v3_init_ept(core, &hw_info) == -1) { PrintError("Error initializing EPT\n"); @@ -410,10 +428,10 @@ static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state) } // Hook all accesses to EFER register - //v3_hook_msr(core->vm_info, EFER_MSR, &debug_efer_read, &debug_efer_write, core); + // v3_hook_msr(core->vm_info, EFER_MSR, &debug_efer_read, &debug_efer_write, core); v3_hook_msr(core->vm_info, EFER_MSR, NULL, NULL, NULL); } else { - PrintError("Invalid Virtual paging mode\n"); + PrintError("Invalid Virtual paging mode (pg_mode=%d) (mach_type=%d)\n", core->shdw_pg_mode, v3_mach_type); return -1; } @@ -520,6 +538,7 @@ static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state) #endif + if (v3_update_vmcs_ctrl_fields(core)) { @@ -542,11 +561,19 @@ static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state) return 0; } -int v3_init_vmx_vmcs(struct guest_info * core, v3_vm_class_t vm_class) { + +static void __init_vmx_vmcs(void * arg) { + struct guest_info * core = arg; struct vmx_data * vmx_state = NULL; int vmx_ret = 0; vmx_state = (struct vmx_data *)V3_Malloc(sizeof(struct vmx_data)); + + if (!vmx_state) { + PrintError("Unable to allocate in initializing vmx vmcs\n"); + return; + } + memset(vmx_state, 0, sizeof(struct vmx_data)); PrintDebug("vmx_data pointer: %p\n", (void *)vmx_state); @@ -568,23 +595,57 @@ int v3_init_vmx_vmcs(struct guest_info * core, v3_vm_class_t vm_class) { if (vmx_ret != VMX_SUCCESS) { PrintError("VMCLEAR failed\n"); - return -1; + return; } - if (vm_class == V3_PC_VM) { + if (core->vm_info->vm_class == V3_PC_VM) { PrintDebug("Initializing VMCS\n"); if (init_vmcs_bios(core, vmx_state) == -1) { PrintError("Error initializing VMCS to BIOS state\n"); - return -1; + return; } } else { PrintError("Invalid VM Class\n"); - return -1; + return; } PrintDebug("Serializing VMCS: %p\n", (void *)vmx_state->vmcs_ptr_phys); vmx_ret = vmcs_clear(vmx_state->vmcs_ptr_phys); + core->core_run_state = CORE_STOPPED; + return; +} + + + +int v3_init_vmx_vmcs(struct guest_info * core, v3_vm_class_t vm_class) { + extern v3_cpu_arch_t v3_cpu_types[]; + + if (v3_cpu_types[V3_Get_CPU()] == V3_INVALID_CPU) { + int i = 0; + + for (i = 0; i < V3_CONFIG_MAX_CPUS; i++) { + if (v3_cpu_types[i] != V3_INVALID_CPU) { + break; + } + } + + if (i == V3_CONFIG_MAX_CPUS) { + PrintError("Could not find VALID CPU for VMX guest initialization\n"); + return -1; + } + + V3_Call_On_CPU(i, __init_vmx_vmcs, core); + + } else { + __init_vmx_vmcs(core); + } + + if (core->core_run_state != CORE_STOPPED) { + PrintError("Error initializing VMX Core\n"); + return -1; + } + return 0; } @@ -607,9 +668,14 @@ int v3_deinit_vmx_vmcs(struct guest_info * core) { * JRL: This is broken */ int v3_vmx_save_core(struct guest_info * core, void * ctx){ - uint64_t vmcs_ptr = vmcs_store(); + struct vmx_data * vmx_info = (struct vmx_data *)(core->vmm_data); - v3_chkpt_save(ctx, "vmcs_data", PAGE_SIZE, (void *)vmcs_ptr); + // note that the vmcs pointer is an HPA, but we need an HVA + if (v3_chkpt_save(ctx, "vmcs_data", PAGE_SIZE_4KB, + V3_VAddr((void*) (vmx_info->vmcs_ptr_phys))) ==-1) { + PrintError("Could not save vmcs data for VMX\n"); + return -1; + } return 0; } @@ -617,12 +683,27 @@ int v3_vmx_save_core(struct guest_info * core, void * ctx){ int v3_vmx_load_core(struct guest_info * core, void * ctx){ struct vmx_data * vmx_info = (struct vmx_data *)(core->vmm_data); struct cr0_32 * shadow_cr0; - char vmcs[PAGE_SIZE_4KB]; + addr_t vmcs_page_paddr; //HPA + + vmcs_page_paddr = (addr_t) V3_AllocPages(1); + + if (!vmcs_page_paddr) { + PrintError("Could not allocate space for a vmcs in VMX\n"); + return -1; + } - v3_chkpt_load(ctx, "vmcs_data", PAGE_SIZE_4KB, vmcs); + if (v3_chkpt_load(ctx, "vmcs_data", PAGE_SIZE_4KB, + V3_VAddr((void *)vmcs_page_paddr)) == -1) { + PrintError("Could not load vmcs data for VMX\n"); + return -1; + } vmcs_clear(vmx_info->vmcs_ptr_phys); - vmcs_load((addr_t)vmcs); + + // Probably need to delete the old one... + V3_FreePages((void*)(vmx_info->vmcs_ptr_phys),1); + + vmcs_load(vmcs_page_paddr); v3_vmx_save_vmcs(core); @@ -835,7 +916,37 @@ static void print_exit_log(struct guest_info * info) { } +int +v3_vmx_config_tsc_virtualization(struct guest_info * info) { + struct vmx_data * vmx_info = (struct vmx_data *)(info->vmm_data); + + if (info->time_state.flags & VM_TIME_TRAP_RDTSC) { + if (!vmx_info->pri_proc_ctrls.rdtsc_exit) { + vmx_info->pri_proc_ctrls.rdtsc_exit = 1; + check_vmcs_write(VMCS_PROC_CTRLS, vmx_info->pri_proc_ctrls.value); + } + } else { + sint64_t tsc_offset; + uint32_t tsc_offset_low, tsc_offset_high; + if (vmx_info->pri_proc_ctrls.rdtsc_exit) { + vmx_info->pri_proc_ctrls.rdtsc_exit = 0; + check_vmcs_write(VMCS_PROC_CTRLS, vmx_info->pri_proc_ctrls.value); + } + + if (info->time_state.flags & VM_TIME_TSC_PASSTHROUGH) { + tsc_offset = 0; + } else { + tsc_offset = v3_tsc_host_offset(&info->time_state); + } + tsc_offset_high = (uint32_t)(( tsc_offset >> 32) & 0xffffffff); + tsc_offset_low = (uint32_t)(tsc_offset & 0xffffffff); + + check_vmcs_write(VMCS_TSC_OFFSET_HIGH, tsc_offset_high); + check_vmcs_write(VMCS_TSC_OFFSET, tsc_offset_low); + } + return 0; +} /* * CAUTION and DANGER!!! @@ -847,27 +958,23 @@ static void print_exit_log(struct guest_info * info) { */ int v3_vmx_enter(struct guest_info * info) { int ret = 0; - sint64_t tsc_offset; - 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); uint64_t guest_cycles = 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); - - // disable global interrupts for vm state transition - v3_disable_ints(); + v3_yield_cond(info,-1); // Update timer devices late after being in the VM so that as much // of the time in the VM is accounted for as possible. Also do it before // updating IRQ entry state so that any interrupts the timers raise get - // handled on the next VM entry. Must be done with interrupts disabled. + // handled on the next VM entry. + v3_advance_time(info, NULL); v3_update_timers(info); + // disable global interrupts for vm state transition + v3_disable_ints(); + if (vmcs_store() != vmx_info->vmcs_ptr_phys) { vmcs_clear(vmx_info->vmcs_ptr_phys); vmcs_load(vmx_info->vmcs_ptr_phys); @@ -892,17 +999,8 @@ int v3_vmx_enter(struct guest_info * info) { } - // Perform last-minute time bookkeeping prior to entering the VM - v3_time_enter_vm(info); - - tsc_offset = v3_tsc_host_offset(&info->time_state); - tsc_offset_high = (uint32_t)(( tsc_offset >> 32) & 0xffffffff); - tsc_offset_low = (uint32_t)(tsc_offset & 0xffffffff); - - check_vmcs_write(VMCS_TSC_OFFSET_HIGH, tsc_offset_high); - check_vmcs_write(VMCS_TSC_OFFSET, tsc_offset_low); - - + // Perform last-minute time setup prior to entering the VM + v3_vmx_config_tsc_virtualization(info); if (v3_update_vmcs_host_state(info)) { v3_enable_ints(); @@ -970,8 +1068,7 @@ int v3_vmx_enter(struct guest_info * info) { } // Immediate exit from VM time bookkeeping - v3_time_exit_vm(info, &guest_cycles); - + v3_advance_time(info, &guest_cycles); /* Update guest state */ v3_vmx_save_vmcs(info); @@ -1008,7 +1105,7 @@ int v3_vmx_enter(struct guest_info * info) { update_irq_exit_state(info); #endif - if (exit_info.exit_reason == VMEXIT_INTR_WINDOW) { + if (exit_info.exit_reason == VMX_EXIT_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; @@ -1019,11 +1116,25 @@ int v3_vmx_enter(struct guest_info * info) { #endif } + + // Lastly we check for an NMI exit, and reinject if so + { + struct vmx_basic_exit_info * basic_info = (struct vmx_basic_exit_info *)&(exit_info.exit_reason); + + if (basic_info->reason == VMX_EXIT_INFO_EXCEPTION_OR_NMI) { + if ((uint8_t)exit_info.int_info == 2) { + asm("int $2"); + } + } + } + // reenable global interrupts after vm exit v3_enable_ints(); // Conditionally yield the CPU if the timeslice has expired - v3_yield_cond(info); + v3_yield_cond(info,-1); + v3_advance_time(info, NULL); + v3_update_timers(info); if (v3_handle_vmx_exit(info, &exit_info) == -1) { PrintError("Error in VMX exit handler (Exit reason=%x)\n", exit_info.exit_reason); @@ -1056,7 +1167,7 @@ int v3_start_vmx_guest(struct guest_info * info) { return 0; } - v3_yield(info); + v3_yield(info,-1); //PrintDebug("VMX core %u: still waiting for INIT\n",info->vcpu_id); } @@ -1173,7 +1284,7 @@ int v3_reset_vmx_vm_core(struct guest_info * core, addr_t rip) { // init vmcs bios if ((core->shdw_pg_mode == NESTED_PAGING) && - (v3_cpu_types[core->pcpu_id] == V3_VMX_EPT_UG_CPU)) { + (v3_mach_type == V3_VMX_EPT_UG_CPU)) { // easy core->rip = 0; core->segments.cs.selector = rip << 8; @@ -1192,6 +1303,7 @@ int v3_reset_vmx_vm_core(struct guest_info * core, addr_t rip) { void v3_init_vmx_cpu(int cpu_id) { addr_t vmx_on_region = 0; extern v3_cpu_arch_t v3_mach_type; + extern v3_cpu_arch_t v3_cpu_types[]; if (v3_mach_type == V3_INVALID_CPU) { if (v3_init_vmx_hw(&hw_info) == -1) {