Index: stable/10/lib/libvmmapi/vmmapi.c =================================================================== --- stable/10/lib/libvmmapi/vmmapi.c (revision 270073) +++ stable/10/lib/libvmmapi/vmmapi.c (revision 270074) @@ -1,1094 +1,1108 @@ /*- * Copyright (c) 2011 NetApp, Inc. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "vmmapi.h" #define MB (1024 * 1024UL) #define GB (1024 * 1024 * 1024UL) struct vmctx { int fd; uint32_t lowmem_limit; enum vm_mmap_style vms; int memflags; size_t lowmem; char *lowmem_addr; size_t highmem; char *highmem_addr; char *name; }; #define CREATE(x) sysctlbyname("hw.vmm.create", NULL, NULL, (x), strlen((x))) #define DESTROY(x) sysctlbyname("hw.vmm.destroy", NULL, NULL, (x), strlen((x))) static int vm_device_open(const char *name) { int fd, len; char *vmfile; len = strlen("/dev/vmm/") + strlen(name) + 1; vmfile = malloc(len); assert(vmfile != NULL); snprintf(vmfile, len, "/dev/vmm/%s", name); /* Open the device file */ fd = open(vmfile, O_RDWR, 0); free(vmfile); return (fd); } int vm_create(const char *name) { return (CREATE((char *)name)); } struct vmctx * vm_open(const char *name) { struct vmctx *vm; vm = malloc(sizeof(struct vmctx) + strlen(name) + 1); assert(vm != NULL); vm->fd = -1; vm->memflags = 0; vm->lowmem_limit = 3 * GB; vm->name = (char *)(vm + 1); strcpy(vm->name, name); if ((vm->fd = vm_device_open(vm->name)) < 0) goto err; return (vm); err: vm_destroy(vm); return (NULL); } void vm_destroy(struct vmctx *vm) { assert(vm != NULL); if (vm->fd >= 0) close(vm->fd); DESTROY(vm->name); free(vm); } int vm_parse_memsize(const char *optarg, size_t *ret_memsize) { char *endptr; size_t optval; int error; optval = strtoul(optarg, &endptr, 0); if (*optarg != '\0' && *endptr == '\0') { /* * For the sake of backward compatibility if the memory size * specified on the command line is less than a megabyte then * it is interpreted as being in units of MB. */ if (optval < MB) optval *= MB; *ret_memsize = optval; error = 0; } else error = expand_number(optarg, ret_memsize); return (error); } int vm_get_memory_seg(struct vmctx *ctx, vm_paddr_t gpa, size_t *ret_len, int *wired) { int error; struct vm_memory_segment seg; bzero(&seg, sizeof(seg)); seg.gpa = gpa; error = ioctl(ctx->fd, VM_GET_MEMORY_SEG, &seg); *ret_len = seg.len; if (wired != NULL) *wired = seg.wired; return (error); } uint32_t vm_get_lowmem_limit(struct vmctx *ctx) { return (ctx->lowmem_limit); } void vm_set_lowmem_limit(struct vmctx *ctx, uint32_t limit) { ctx->lowmem_limit = limit; } void vm_set_memflags(struct vmctx *ctx, int flags) { ctx->memflags = flags; } static int setup_memory_segment(struct vmctx *ctx, vm_paddr_t gpa, size_t len, char **addr) { int error, mmap_flags; struct vm_memory_segment seg; /* * Create and optionally map 'len' bytes of memory at guest * physical address 'gpa' */ bzero(&seg, sizeof(seg)); seg.gpa = gpa; seg.len = len; error = ioctl(ctx->fd, VM_MAP_MEMORY, &seg); if (error == 0 && addr != NULL) { mmap_flags = MAP_SHARED; if ((ctx->memflags & VM_MEM_F_INCORE) == 0) mmap_flags |= MAP_NOCORE; *addr = mmap(NULL, len, PROT_READ | PROT_WRITE, mmap_flags, ctx->fd, gpa); } return (error); } int vm_setup_memory(struct vmctx *ctx, size_t memsize, enum vm_mmap_style vms) { char **addr; int error; /* XXX VM_MMAP_SPARSE not implemented yet */ assert(vms == VM_MMAP_NONE || vms == VM_MMAP_ALL); ctx->vms = vms; /* * If 'memsize' cannot fit entirely in the 'lowmem' segment then * create another 'highmem' segment above 4GB for the remainder. */ if (memsize > ctx->lowmem_limit) { ctx->lowmem = ctx->lowmem_limit; ctx->highmem = memsize - ctx->lowmem; } else { ctx->lowmem = memsize; ctx->highmem = 0; } if (ctx->lowmem > 0) { addr = (vms == VM_MMAP_ALL) ? &ctx->lowmem_addr : NULL; error = setup_memory_segment(ctx, 0, ctx->lowmem, addr); if (error) return (error); } if (ctx->highmem > 0) { addr = (vms == VM_MMAP_ALL) ? &ctx->highmem_addr : NULL; error = setup_memory_segment(ctx, 4*GB, ctx->highmem, addr); if (error) return (error); } return (0); } void * vm_map_gpa(struct vmctx *ctx, vm_paddr_t gaddr, size_t len) { /* XXX VM_MMAP_SPARSE not implemented yet */ assert(ctx->vms == VM_MMAP_ALL); if (gaddr < ctx->lowmem && gaddr + len <= ctx->lowmem) return ((void *)(ctx->lowmem_addr + gaddr)); if (gaddr >= 4*GB) { gaddr -= 4*GB; if (gaddr < ctx->highmem && gaddr + len <= ctx->highmem) return ((void *)(ctx->highmem_addr + gaddr)); } return (NULL); } +size_t +vm_get_lowmem_size(struct vmctx *ctx) +{ + + return (ctx->lowmem); +} + +size_t +vm_get_highmem_size(struct vmctx *ctx) +{ + + return (ctx->highmem); +} + int vm_set_desc(struct vmctx *ctx, int vcpu, int reg, uint64_t base, uint32_t limit, uint32_t access) { int error; struct vm_seg_desc vmsegdesc; bzero(&vmsegdesc, sizeof(vmsegdesc)); vmsegdesc.cpuid = vcpu; vmsegdesc.regnum = reg; vmsegdesc.desc.base = base; vmsegdesc.desc.limit = limit; vmsegdesc.desc.access = access; error = ioctl(ctx->fd, VM_SET_SEGMENT_DESCRIPTOR, &vmsegdesc); return (error); } int vm_get_desc(struct vmctx *ctx, int vcpu, int reg, uint64_t *base, uint32_t *limit, uint32_t *access) { int error; struct vm_seg_desc vmsegdesc; bzero(&vmsegdesc, sizeof(vmsegdesc)); vmsegdesc.cpuid = vcpu; vmsegdesc.regnum = reg; error = ioctl(ctx->fd, VM_GET_SEGMENT_DESCRIPTOR, &vmsegdesc); if (error == 0) { *base = vmsegdesc.desc.base; *limit = vmsegdesc.desc.limit; *access = vmsegdesc.desc.access; } return (error); } int vm_set_register(struct vmctx *ctx, int vcpu, int reg, uint64_t val) { int error; struct vm_register vmreg; bzero(&vmreg, sizeof(vmreg)); vmreg.cpuid = vcpu; vmreg.regnum = reg; vmreg.regval = val; error = ioctl(ctx->fd, VM_SET_REGISTER, &vmreg); return (error); } int vm_get_register(struct vmctx *ctx, int vcpu, int reg, uint64_t *ret_val) { int error; struct vm_register vmreg; bzero(&vmreg, sizeof(vmreg)); vmreg.cpuid = vcpu; vmreg.regnum = reg; error = ioctl(ctx->fd, VM_GET_REGISTER, &vmreg); *ret_val = vmreg.regval; return (error); } int vm_run(struct vmctx *ctx, int vcpu, uint64_t rip, struct vm_exit *vmexit) { int error; struct vm_run vmrun; bzero(&vmrun, sizeof(vmrun)); vmrun.cpuid = vcpu; vmrun.rip = rip; error = ioctl(ctx->fd, VM_RUN, &vmrun); bcopy(&vmrun.vm_exit, vmexit, sizeof(struct vm_exit)); return (error); } int vm_suspend(struct vmctx *ctx, enum vm_suspend_how how) { struct vm_suspend vmsuspend; bzero(&vmsuspend, sizeof(vmsuspend)); vmsuspend.how = how; return (ioctl(ctx->fd, VM_SUSPEND, &vmsuspend)); } int vm_reinit(struct vmctx *ctx) { return (ioctl(ctx->fd, VM_REINIT, 0)); } static int vm_inject_exception_real(struct vmctx *ctx, int vcpu, int vector, int error_code, int error_code_valid) { struct vm_exception exc; bzero(&exc, sizeof(exc)); exc.cpuid = vcpu; exc.vector = vector; exc.error_code = error_code; exc.error_code_valid = error_code_valid; return (ioctl(ctx->fd, VM_INJECT_EXCEPTION, &exc)); } int vm_inject_exception(struct vmctx *ctx, int vcpu, int vector) { return (vm_inject_exception_real(ctx, vcpu, vector, 0, 0)); } int vm_inject_exception2(struct vmctx *ctx, int vcpu, int vector, int errcode) { return (vm_inject_exception_real(ctx, vcpu, vector, errcode, 1)); } int vm_apicid2vcpu(struct vmctx *ctx, int apicid) { /* * The apic id associated with the 'vcpu' has the same numerical value * as the 'vcpu' itself. */ return (apicid); } int vm_lapic_irq(struct vmctx *ctx, int vcpu, int vector) { struct vm_lapic_irq vmirq; bzero(&vmirq, sizeof(vmirq)); vmirq.cpuid = vcpu; vmirq.vector = vector; return (ioctl(ctx->fd, VM_LAPIC_IRQ, &vmirq)); } int vm_lapic_local_irq(struct vmctx *ctx, int vcpu, int vector) { struct vm_lapic_irq vmirq; bzero(&vmirq, sizeof(vmirq)); vmirq.cpuid = vcpu; vmirq.vector = vector; return (ioctl(ctx->fd, VM_LAPIC_LOCAL_IRQ, &vmirq)); } int vm_lapic_msi(struct vmctx *ctx, uint64_t addr, uint64_t msg) { struct vm_lapic_msi vmmsi; bzero(&vmmsi, sizeof(vmmsi)); vmmsi.addr = addr; vmmsi.msg = msg; return (ioctl(ctx->fd, VM_LAPIC_MSI, &vmmsi)); } int vm_ioapic_assert_irq(struct vmctx *ctx, int irq) { struct vm_ioapic_irq ioapic_irq; bzero(&ioapic_irq, sizeof(struct vm_ioapic_irq)); ioapic_irq.irq = irq; return (ioctl(ctx->fd, VM_IOAPIC_ASSERT_IRQ, &ioapic_irq)); } int vm_ioapic_deassert_irq(struct vmctx *ctx, int irq) { struct vm_ioapic_irq ioapic_irq; bzero(&ioapic_irq, sizeof(struct vm_ioapic_irq)); ioapic_irq.irq = irq; return (ioctl(ctx->fd, VM_IOAPIC_DEASSERT_IRQ, &ioapic_irq)); } int vm_ioapic_pulse_irq(struct vmctx *ctx, int irq) { struct vm_ioapic_irq ioapic_irq; bzero(&ioapic_irq, sizeof(struct vm_ioapic_irq)); ioapic_irq.irq = irq; return (ioctl(ctx->fd, VM_IOAPIC_PULSE_IRQ, &ioapic_irq)); } int vm_ioapic_pincount(struct vmctx *ctx, int *pincount) { return (ioctl(ctx->fd, VM_IOAPIC_PINCOUNT, pincount)); } int vm_isa_assert_irq(struct vmctx *ctx, int atpic_irq, int ioapic_irq) { struct vm_isa_irq isa_irq; bzero(&isa_irq, sizeof(struct vm_isa_irq)); isa_irq.atpic_irq = atpic_irq; isa_irq.ioapic_irq = ioapic_irq; return (ioctl(ctx->fd, VM_ISA_ASSERT_IRQ, &isa_irq)); } int vm_isa_deassert_irq(struct vmctx *ctx, int atpic_irq, int ioapic_irq) { struct vm_isa_irq isa_irq; bzero(&isa_irq, sizeof(struct vm_isa_irq)); isa_irq.atpic_irq = atpic_irq; isa_irq.ioapic_irq = ioapic_irq; return (ioctl(ctx->fd, VM_ISA_DEASSERT_IRQ, &isa_irq)); } int vm_isa_pulse_irq(struct vmctx *ctx, int atpic_irq, int ioapic_irq) { struct vm_isa_irq isa_irq; bzero(&isa_irq, sizeof(struct vm_isa_irq)); isa_irq.atpic_irq = atpic_irq; isa_irq.ioapic_irq = ioapic_irq; return (ioctl(ctx->fd, VM_ISA_PULSE_IRQ, &isa_irq)); } int vm_isa_set_irq_trigger(struct vmctx *ctx, int atpic_irq, enum vm_intr_trigger trigger) { struct vm_isa_irq_trigger isa_irq_trigger; bzero(&isa_irq_trigger, sizeof(struct vm_isa_irq_trigger)); isa_irq_trigger.atpic_irq = atpic_irq; isa_irq_trigger.trigger = trigger; return (ioctl(ctx->fd, VM_ISA_SET_IRQ_TRIGGER, &isa_irq_trigger)); } int vm_inject_nmi(struct vmctx *ctx, int vcpu) { struct vm_nmi vmnmi; bzero(&vmnmi, sizeof(vmnmi)); vmnmi.cpuid = vcpu; return (ioctl(ctx->fd, VM_INJECT_NMI, &vmnmi)); } static struct { const char *name; int type; } capstrmap[] = { { "hlt_exit", VM_CAP_HALT_EXIT }, { "mtrap_exit", VM_CAP_MTRAP_EXIT }, { "pause_exit", VM_CAP_PAUSE_EXIT }, { "unrestricted_guest", VM_CAP_UNRESTRICTED_GUEST }, { "enable_invpcid", VM_CAP_ENABLE_INVPCID }, { 0 } }; int vm_capability_name2type(const char *capname) { int i; for (i = 0; capstrmap[i].name != NULL && capname != NULL; i++) { if (strcmp(capstrmap[i].name, capname) == 0) return (capstrmap[i].type); } return (-1); } const char * vm_capability_type2name(int type) { int i; for (i = 0; capstrmap[i].name != NULL; i++) { if (capstrmap[i].type == type) return (capstrmap[i].name); } return (NULL); } int vm_get_capability(struct vmctx *ctx, int vcpu, enum vm_cap_type cap, int *retval) { int error; struct vm_capability vmcap; bzero(&vmcap, sizeof(vmcap)); vmcap.cpuid = vcpu; vmcap.captype = cap; error = ioctl(ctx->fd, VM_GET_CAPABILITY, &vmcap); *retval = vmcap.capval; return (error); } int vm_set_capability(struct vmctx *ctx, int vcpu, enum vm_cap_type cap, int val) { struct vm_capability vmcap; bzero(&vmcap, sizeof(vmcap)); vmcap.cpuid = vcpu; vmcap.captype = cap; vmcap.capval = val; return (ioctl(ctx->fd, VM_SET_CAPABILITY, &vmcap)); } int vm_assign_pptdev(struct vmctx *ctx, int bus, int slot, int func) { struct vm_pptdev pptdev; bzero(&pptdev, sizeof(pptdev)); pptdev.bus = bus; pptdev.slot = slot; pptdev.func = func; return (ioctl(ctx->fd, VM_BIND_PPTDEV, &pptdev)); } int vm_unassign_pptdev(struct vmctx *ctx, int bus, int slot, int func) { struct vm_pptdev pptdev; bzero(&pptdev, sizeof(pptdev)); pptdev.bus = bus; pptdev.slot = slot; pptdev.func = func; return (ioctl(ctx->fd, VM_UNBIND_PPTDEV, &pptdev)); } int vm_map_pptdev_mmio(struct vmctx *ctx, int bus, int slot, int func, vm_paddr_t gpa, size_t len, vm_paddr_t hpa) { struct vm_pptdev_mmio pptmmio; bzero(&pptmmio, sizeof(pptmmio)); pptmmio.bus = bus; pptmmio.slot = slot; pptmmio.func = func; pptmmio.gpa = gpa; pptmmio.len = len; pptmmio.hpa = hpa; return (ioctl(ctx->fd, VM_MAP_PPTDEV_MMIO, &pptmmio)); } int vm_setup_pptdev_msi(struct vmctx *ctx, int vcpu, int bus, int slot, int func, uint64_t addr, uint64_t msg, int numvec) { struct vm_pptdev_msi pptmsi; bzero(&pptmsi, sizeof(pptmsi)); pptmsi.vcpu = vcpu; pptmsi.bus = bus; pptmsi.slot = slot; pptmsi.func = func; pptmsi.msg = msg; pptmsi.addr = addr; pptmsi.numvec = numvec; return (ioctl(ctx->fd, VM_PPTDEV_MSI, &pptmsi)); } int vm_setup_pptdev_msix(struct vmctx *ctx, int vcpu, int bus, int slot, int func, int idx, uint64_t addr, uint64_t msg, uint32_t vector_control) { struct vm_pptdev_msix pptmsix; bzero(&pptmsix, sizeof(pptmsix)); pptmsix.vcpu = vcpu; pptmsix.bus = bus; pptmsix.slot = slot; pptmsix.func = func; pptmsix.idx = idx; pptmsix.msg = msg; pptmsix.addr = addr; pptmsix.vector_control = vector_control; return ioctl(ctx->fd, VM_PPTDEV_MSIX, &pptmsix); } uint64_t * vm_get_stats(struct vmctx *ctx, int vcpu, struct timeval *ret_tv, int *ret_entries) { int error; static struct vm_stats vmstats; vmstats.cpuid = vcpu; error = ioctl(ctx->fd, VM_STATS, &vmstats); if (error == 0) { if (ret_entries) *ret_entries = vmstats.num_entries; if (ret_tv) *ret_tv = vmstats.tv; return (vmstats.statbuf); } else return (NULL); } const char * vm_get_stat_desc(struct vmctx *ctx, int index) { static struct vm_stat_desc statdesc; statdesc.index = index; if (ioctl(ctx->fd, VM_STAT_DESC, &statdesc) == 0) return (statdesc.desc); else return (NULL); } int vm_get_x2apic_state(struct vmctx *ctx, int vcpu, enum x2apic_state *state) { int error; struct vm_x2apic x2apic; bzero(&x2apic, sizeof(x2apic)); x2apic.cpuid = vcpu; error = ioctl(ctx->fd, VM_GET_X2APIC_STATE, &x2apic); *state = x2apic.state; return (error); } int vm_set_x2apic_state(struct vmctx *ctx, int vcpu, enum x2apic_state state) { int error; struct vm_x2apic x2apic; bzero(&x2apic, sizeof(x2apic)); x2apic.cpuid = vcpu; x2apic.state = state; error = ioctl(ctx->fd, VM_SET_X2APIC_STATE, &x2apic); return (error); } /* * From Intel Vol 3a: * Table 9-1. IA-32 Processor States Following Power-up, Reset or INIT */ int vcpu_reset(struct vmctx *vmctx, int vcpu) { int error; uint64_t rflags, rip, cr0, cr4, zero, desc_base, rdx; uint32_t desc_access, desc_limit; uint16_t sel; zero = 0; rflags = 0x2; error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RFLAGS, rflags); if (error) goto done; rip = 0xfff0; if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RIP, rip)) != 0) goto done; cr0 = CR0_NE; if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_CR0, cr0)) != 0) goto done; if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_CR3, zero)) != 0) goto done; cr4 = 0; if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_CR4, cr4)) != 0) goto done; /* * CS: present, r/w, accessed, 16-bit, byte granularity, usable */ desc_base = 0xffff0000; desc_limit = 0xffff; desc_access = 0x0093; error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_CS, desc_base, desc_limit, desc_access); if (error) goto done; sel = 0xf000; if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_CS, sel)) != 0) goto done; /* * SS,DS,ES,FS,GS: present, r/w, accessed, 16-bit, byte granularity */ desc_base = 0; desc_limit = 0xffff; desc_access = 0x0093; error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_SS, desc_base, desc_limit, desc_access); if (error) goto done; error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_DS, desc_base, desc_limit, desc_access); if (error) goto done; error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_ES, desc_base, desc_limit, desc_access); if (error) goto done; error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_FS, desc_base, desc_limit, desc_access); if (error) goto done; error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_GS, desc_base, desc_limit, desc_access); if (error) goto done; sel = 0; if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_SS, sel)) != 0) goto done; if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_DS, sel)) != 0) goto done; if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_ES, sel)) != 0) goto done; if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_FS, sel)) != 0) goto done; if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_GS, sel)) != 0) goto done; /* General purpose registers */ rdx = 0xf00; if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RAX, zero)) != 0) goto done; if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RBX, zero)) != 0) goto done; if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RCX, zero)) != 0) goto done; if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RDX, rdx)) != 0) goto done; if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RSI, zero)) != 0) goto done; if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RDI, zero)) != 0) goto done; if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RBP, zero)) != 0) goto done; if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RSP, zero)) != 0) goto done; /* GDTR, IDTR */ desc_base = 0; desc_limit = 0xffff; desc_access = 0; error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_GDTR, desc_base, desc_limit, desc_access); if (error != 0) goto done; error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_IDTR, desc_base, desc_limit, desc_access); if (error != 0) goto done; /* TR */ desc_base = 0; desc_limit = 0xffff; desc_access = 0x0000008b; error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_TR, 0, 0, desc_access); if (error) goto done; sel = 0; if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_TR, sel)) != 0) goto done; /* LDTR */ desc_base = 0; desc_limit = 0xffff; desc_access = 0x00000082; error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_LDTR, desc_base, desc_limit, desc_access); if (error) goto done; sel = 0; if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_LDTR, 0)) != 0) goto done; /* XXX cr2, debug registers */ error = 0; done: return (error); } int vm_get_gpa_pmap(struct vmctx *ctx, uint64_t gpa, uint64_t *pte, int *num) { int error, i; struct vm_gpa_pte gpapte; bzero(&gpapte, sizeof(gpapte)); gpapte.gpa = gpa; error = ioctl(ctx->fd, VM_GET_GPA_PMAP, &gpapte); if (error == 0) { *num = gpapte.ptenum; for (i = 0; i < gpapte.ptenum; i++) pte[i] = gpapte.pte[i]; } return (error); } int vm_get_hpet_capabilities(struct vmctx *ctx, uint32_t *capabilities) { int error; struct vm_hpet_cap cap; bzero(&cap, sizeof(struct vm_hpet_cap)); error = ioctl(ctx->fd, VM_GET_HPET_CAPABILITIES, &cap); if (capabilities != NULL) *capabilities = cap.capabilities; return (error); } static int gla2gpa(struct vmctx *ctx, int vcpu, struct vm_guest_paging *paging, uint64_t gla, int prot, int *fault, uint64_t *gpa) { struct vm_gla2gpa gg; int error; bzero(&gg, sizeof(struct vm_gla2gpa)); gg.vcpuid = vcpu; gg.prot = prot; gg.gla = gla; gg.paging = *paging; error = ioctl(ctx->fd, VM_GLA2GPA, &gg); if (error == 0) { *fault = gg.fault; *gpa = gg.gpa; } return (error); } #ifndef min #define min(a,b) (((a) < (b)) ? (a) : (b)) #endif int vm_gla2gpa(struct vmctx *ctx, int vcpu, struct vm_guest_paging *paging, uint64_t gla, size_t len, int prot, struct iovec *iov, int iovcnt) { uint64_t gpa; int error, fault, i, n, off; for (i = 0; i < iovcnt; i++) { iov[i].iov_base = 0; iov[i].iov_len = 0; } while (len) { assert(iovcnt > 0); error = gla2gpa(ctx, vcpu, paging, gla, prot, &fault, &gpa); if (error) return (-1); if (fault) return (1); off = gpa & PAGE_MASK; n = min(len, PAGE_SIZE - off); iov->iov_base = (void *)gpa; iov->iov_len = n; iov++; iovcnt--; gla += n; len -= n; } return (0); } void vm_copyin(struct vmctx *ctx, int vcpu, struct iovec *iov, void *vp, size_t len) { const char *src; char *dst; uint64_t gpa; size_t n; dst = vp; while (len) { assert(iov->iov_len); gpa = (uint64_t)iov->iov_base; n = min(len, iov->iov_len); src = vm_map_gpa(ctx, gpa, n); bcopy(src, dst, n); iov++; dst += n; len -= n; } } void vm_copyout(struct vmctx *ctx, int vcpu, const void *vp, struct iovec *iov, size_t len) { const char *src; char *dst; uint64_t gpa; size_t n; src = vp; while (len) { assert(iov->iov_len); gpa = (uint64_t)iov->iov_base; n = min(len, iov->iov_len); dst = vm_map_gpa(ctx, gpa, n); bcopy(src, dst, n); iov++; src += n; len -= n; } } static int vm_get_cpus(struct vmctx *ctx, int which, cpuset_t *cpus) { struct vm_cpuset vm_cpuset; int error; bzero(&vm_cpuset, sizeof(struct vm_cpuset)); vm_cpuset.which = which; vm_cpuset.cpusetsize = sizeof(cpuset_t); vm_cpuset.cpus = cpus; error = ioctl(ctx->fd, VM_GET_CPUS, &vm_cpuset); return (error); } int vm_active_cpus(struct vmctx *ctx, cpuset_t *cpus) { return (vm_get_cpus(ctx, VM_ACTIVE_CPUS, cpus)); } int vm_suspended_cpus(struct vmctx *ctx, cpuset_t *cpus) { return (vm_get_cpus(ctx, VM_SUSPENDED_CPUS, cpus)); } int vm_activate_cpu(struct vmctx *ctx, int vcpu) { struct vm_activate_cpu ac; int error; bzero(&ac, sizeof(struct vm_activate_cpu)); ac.vcpuid = vcpu; error = ioctl(ctx->fd, VM_ACTIVATE_CPU, &ac); return (error); } Index: stable/10/lib/libvmmapi/vmmapi.h =================================================================== --- stable/10/lib/libvmmapi/vmmapi.h (revision 270073) +++ stable/10/lib/libvmmapi/vmmapi.h (revision 270074) @@ -1,146 +1,148 @@ /*- * Copyright (c) 2011 NetApp, Inc. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #ifndef _VMMAPI_H_ #define _VMMAPI_H_ #include #include struct iovec; struct vmctx; enum x2apic_state; /* * Different styles of mapping the memory assigned to a VM into the address * space of the controlling process. */ enum vm_mmap_style { VM_MMAP_NONE, /* no mapping */ VM_MMAP_ALL, /* fully and statically mapped */ VM_MMAP_SPARSE, /* mappings created on-demand */ }; #define VM_MEM_F_INCORE 0x01 /* include guest memory in core file */ int vm_create(const char *name); struct vmctx *vm_open(const char *name); void vm_destroy(struct vmctx *ctx); int vm_parse_memsize(const char *optarg, size_t *memsize); int vm_get_memory_seg(struct vmctx *ctx, vm_paddr_t gpa, size_t *ret_len, int *wired); int vm_setup_memory(struct vmctx *ctx, size_t len, enum vm_mmap_style s); void *vm_map_gpa(struct vmctx *ctx, vm_paddr_t gaddr, size_t len); int vm_get_gpa_pmap(struct vmctx *, uint64_t gpa, uint64_t *pte, int *num); uint32_t vm_get_lowmem_limit(struct vmctx *ctx); void vm_set_lowmem_limit(struct vmctx *ctx, uint32_t limit); void vm_set_memflags(struct vmctx *ctx, int flags); +size_t vm_get_lowmem_size(struct vmctx *ctx); +size_t vm_get_highmem_size(struct vmctx *ctx); int vm_set_desc(struct vmctx *ctx, int vcpu, int reg, uint64_t base, uint32_t limit, uint32_t access); int vm_get_desc(struct vmctx *ctx, int vcpu, int reg, uint64_t *base, uint32_t *limit, uint32_t *access); int vm_set_register(struct vmctx *ctx, int vcpu, int reg, uint64_t val); int vm_get_register(struct vmctx *ctx, int vcpu, int reg, uint64_t *retval); int vm_run(struct vmctx *ctx, int vcpu, uint64_t rip, struct vm_exit *ret_vmexit); int vm_suspend(struct vmctx *ctx, enum vm_suspend_how how); int vm_reinit(struct vmctx *ctx); int vm_apicid2vcpu(struct vmctx *ctx, int apicid); int vm_inject_exception(struct vmctx *ctx, int vcpu, int vec); int vm_inject_exception2(struct vmctx *ctx, int vcpu, int vec, int errcode); int vm_lapic_irq(struct vmctx *ctx, int vcpu, int vector); int vm_lapic_local_irq(struct vmctx *ctx, int vcpu, int vector); int vm_lapic_msi(struct vmctx *ctx, uint64_t addr, uint64_t msg); int vm_ioapic_assert_irq(struct vmctx *ctx, int irq); int vm_ioapic_deassert_irq(struct vmctx *ctx, int irq); int vm_ioapic_pulse_irq(struct vmctx *ctx, int irq); int vm_ioapic_pincount(struct vmctx *ctx, int *pincount); int vm_isa_assert_irq(struct vmctx *ctx, int atpic_irq, int ioapic_irq); int vm_isa_deassert_irq(struct vmctx *ctx, int atpic_irq, int ioapic_irq); int vm_isa_pulse_irq(struct vmctx *ctx, int atpic_irq, int ioapic_irq); int vm_isa_set_irq_trigger(struct vmctx *ctx, int atpic_irq, enum vm_intr_trigger trigger); int vm_inject_nmi(struct vmctx *ctx, int vcpu); int vm_capability_name2type(const char *capname); const char *vm_capability_type2name(int type); int vm_get_capability(struct vmctx *ctx, int vcpu, enum vm_cap_type cap, int *retval); int vm_set_capability(struct vmctx *ctx, int vcpu, enum vm_cap_type cap, int val); int vm_assign_pptdev(struct vmctx *ctx, int bus, int slot, int func); int vm_unassign_pptdev(struct vmctx *ctx, int bus, int slot, int func); int vm_map_pptdev_mmio(struct vmctx *ctx, int bus, int slot, int func, vm_paddr_t gpa, size_t len, vm_paddr_t hpa); int vm_setup_pptdev_msi(struct vmctx *ctx, int vcpu, int bus, int slot, int func, uint64_t addr, uint64_t msg, int numvec); int vm_setup_pptdev_msix(struct vmctx *ctx, int vcpu, int bus, int slot, int func, int idx, uint64_t addr, uint64_t msg, uint32_t vector_control); /* * Return a pointer to the statistics buffer. Note that this is not MT-safe. */ uint64_t *vm_get_stats(struct vmctx *ctx, int vcpu, struct timeval *ret_tv, int *ret_entries); const char *vm_get_stat_desc(struct vmctx *ctx, int index); int vm_get_x2apic_state(struct vmctx *ctx, int vcpu, enum x2apic_state *s); int vm_set_x2apic_state(struct vmctx *ctx, int vcpu, enum x2apic_state s); int vm_get_hpet_capabilities(struct vmctx *ctx, uint32_t *capabilities); /* * Translate the GLA range [gla,gla+len) into GPA segments in 'iov'. * The 'iovcnt' should be big enough to accomodate all GPA segments. * Returns 0 on success, 1 on a guest fault condition and -1 otherwise. */ int vm_gla2gpa(struct vmctx *ctx, int vcpu, struct vm_guest_paging *paging, uint64_t gla, size_t len, int prot, struct iovec *iov, int iovcnt); void vm_copyin(struct vmctx *ctx, int vcpu, struct iovec *guest_iov, void *host_dst, size_t len); void vm_copyout(struct vmctx *ctx, int vcpu, const void *host_src, struct iovec *guest_iov, size_t len); /* Reset vcpu register state */ int vcpu_reset(struct vmctx *ctx, int vcpu); int vm_active_cpus(struct vmctx *ctx, cpuset_t *cpus); int vm_suspended_cpus(struct vmctx *ctx, cpuset_t *cpus); int vm_activate_cpu(struct vmctx *ctx, int vcpu); /* * FreeBSD specific APIs */ int vm_setup_freebsd_registers(struct vmctx *ctx, int vcpu, uint64_t rip, uint64_t cr3, uint64_t gdtbase, uint64_t rsp); int vm_setup_freebsd_registers_i386(struct vmctx *vmctx, int vcpu, uint32_t eip, uint32_t gdtbase, uint32_t esp); void vm_setup_freebsd_gdt(uint64_t *gdtr); #endif /* _VMMAPI_H_ */ Index: stable/10/sys/amd64/include/vmm.h =================================================================== --- stable/10/sys/amd64/include/vmm.h (revision 270073) +++ stable/10/sys/amd64/include/vmm.h (revision 270074) @@ -1,494 +1,496 @@ /*- * Copyright (c) 2011 NetApp, Inc. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #ifndef _VMM_H_ #define _VMM_H_ enum vm_suspend_how { VM_SUSPEND_NONE, VM_SUSPEND_RESET, VM_SUSPEND_POWEROFF, VM_SUSPEND_HALT, VM_SUSPEND_LAST }; #ifdef _KERNEL #define VM_MAX_NAMELEN 32 struct vm; struct vm_exception; struct vm_memory_segment; struct seg_desc; struct vm_exit; struct vm_run; struct vhpet; struct vioapic; struct vlapic; struct vmspace; struct vm_object; struct pmap; enum vm_reg_name; enum x2apic_state; typedef int (*vmm_init_func_t)(int ipinum); typedef int (*vmm_cleanup_func_t)(void); typedef void (*vmm_resume_func_t)(void); typedef void * (*vmi_init_func_t)(struct vm *vm, struct pmap *pmap); typedef int (*vmi_run_func_t)(void *vmi, int vcpu, register_t rip, struct pmap *pmap, void *rendezvous_cookie, void *suspend_cookie); typedef void (*vmi_cleanup_func_t)(void *vmi); typedef int (*vmi_get_register_t)(void *vmi, int vcpu, int num, uint64_t *retval); typedef int (*vmi_set_register_t)(void *vmi, int vcpu, int num, uint64_t val); typedef int (*vmi_get_desc_t)(void *vmi, int vcpu, int num, struct seg_desc *desc); typedef int (*vmi_set_desc_t)(void *vmi, int vcpu, int num, struct seg_desc *desc); typedef int (*vmi_get_cap_t)(void *vmi, int vcpu, int num, int *retval); typedef int (*vmi_set_cap_t)(void *vmi, int vcpu, int num, int val); typedef struct vmspace * (*vmi_vmspace_alloc)(vm_offset_t min, vm_offset_t max); typedef void (*vmi_vmspace_free)(struct vmspace *vmspace); typedef struct vlapic * (*vmi_vlapic_init)(void *vmi, int vcpu); typedef void (*vmi_vlapic_cleanup)(void *vmi, struct vlapic *vlapic); struct vmm_ops { vmm_init_func_t init; /* module wide initialization */ vmm_cleanup_func_t cleanup; vmm_resume_func_t resume; vmi_init_func_t vminit; /* vm-specific initialization */ vmi_run_func_t vmrun; vmi_cleanup_func_t vmcleanup; vmi_get_register_t vmgetreg; vmi_set_register_t vmsetreg; vmi_get_desc_t vmgetdesc; vmi_set_desc_t vmsetdesc; vmi_get_cap_t vmgetcap; vmi_set_cap_t vmsetcap; vmi_vmspace_alloc vmspace_alloc; vmi_vmspace_free vmspace_free; vmi_vlapic_init vlapic_init; vmi_vlapic_cleanup vlapic_cleanup; }; extern struct vmm_ops vmm_ops_intel; extern struct vmm_ops vmm_ops_amd; int vm_create(const char *name, struct vm **retvm); void vm_destroy(struct vm *vm); int vm_reinit(struct vm *vm); const char *vm_name(struct vm *vm); int vm_malloc(struct vm *vm, vm_paddr_t gpa, size_t len); int vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa); int vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len); void *vm_gpa_hold(struct vm *, vm_paddr_t gpa, size_t len, int prot, void **cookie); void vm_gpa_release(void *cookie); int vm_gpabase2memseg(struct vm *vm, vm_paddr_t gpabase, struct vm_memory_segment *seg); int vm_get_memobj(struct vm *vm, vm_paddr_t gpa, size_t len, vm_offset_t *offset, struct vm_object **object); boolean_t vm_mem_allocated(struct vm *vm, vm_paddr_t gpa); int vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval); int vm_set_register(struct vm *vm, int vcpu, int reg, uint64_t val); int vm_get_seg_desc(struct vm *vm, int vcpu, int reg, struct seg_desc *ret_desc); int vm_set_seg_desc(struct vm *vm, int vcpu, int reg, struct seg_desc *desc); int vm_run(struct vm *vm, struct vm_run *vmrun); int vm_suspend(struct vm *vm, enum vm_suspend_how how); int vm_inject_nmi(struct vm *vm, int vcpu); int vm_nmi_pending(struct vm *vm, int vcpuid); void vm_nmi_clear(struct vm *vm, int vcpuid); int vm_inject_extint(struct vm *vm, int vcpu); int vm_extint_pending(struct vm *vm, int vcpuid); void vm_extint_clear(struct vm *vm, int vcpuid); uint64_t *vm_guest_msrs(struct vm *vm, int cpu); struct vlapic *vm_lapic(struct vm *vm, int cpu); struct vioapic *vm_ioapic(struct vm *vm); struct vhpet *vm_hpet(struct vm *vm); int vm_get_capability(struct vm *vm, int vcpu, int type, int *val); int vm_set_capability(struct vm *vm, int vcpu, int type, int val); int vm_get_x2apic_state(struct vm *vm, int vcpu, enum x2apic_state *state); int vm_set_x2apic_state(struct vm *vm, int vcpu, enum x2apic_state state); int vm_apicid2vcpuid(struct vm *vm, int apicid); int vm_activate_cpu(struct vm *vm, int vcpu); cpuset_t vm_active_cpus(struct vm *vm); cpuset_t vm_suspended_cpus(struct vm *vm); struct vm_exit *vm_exitinfo(struct vm *vm, int vcpuid); void vm_exit_suspended(struct vm *vm, int vcpuid, uint64_t rip); +void vm_exit_rendezvous(struct vm *vm, int vcpuid, uint64_t rip); +void vm_exit_astpending(struct vm *vm, int vcpuid, uint64_t rip); /* * Rendezvous all vcpus specified in 'dest' and execute 'func(arg)'. * The rendezvous 'func(arg)' is not allowed to do anything that will * cause the thread to be put to sleep. * * If the rendezvous is being initiated from a vcpu context then the * 'vcpuid' must refer to that vcpu, otherwise it should be set to -1. * * The caller cannot hold any locks when initiating the rendezvous. * * The implementation of this API may cause vcpus other than those specified * by 'dest' to be stalled. The caller should not rely on any vcpus making * forward progress when the rendezvous is in progress. */ typedef void (*vm_rendezvous_func_t)(struct vm *vm, int vcpuid, void *arg); void vm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest, vm_rendezvous_func_t func, void *arg); static __inline int vcpu_rendezvous_pending(void *rendezvous_cookie) { return (*(uintptr_t *)rendezvous_cookie != 0); } static __inline int vcpu_suspended(void *suspend_cookie) { return (*(int *)suspend_cookie); } /* * Return 1 if device indicated by bus/slot/func is supposed to be a * pci passthrough device. * * Return 0 otherwise. */ int vmm_is_pptdev(int bus, int slot, int func); void *vm_iommu_domain(struct vm *vm); enum vcpu_state { VCPU_IDLE, VCPU_FROZEN, VCPU_RUNNING, VCPU_SLEEPING, }; int vcpu_set_state(struct vm *vm, int vcpu, enum vcpu_state state, bool from_idle); enum vcpu_state vcpu_get_state(struct vm *vm, int vcpu, int *hostcpu); static int __inline vcpu_is_running(struct vm *vm, int vcpu, int *hostcpu) { return (vcpu_get_state(vm, vcpu, hostcpu) == VCPU_RUNNING); } void *vcpu_stats(struct vm *vm, int vcpu); void vcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr); struct vmspace *vm_get_vmspace(struct vm *vm); int vm_assign_pptdev(struct vm *vm, int bus, int slot, int func); int vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func); struct vatpic *vm_atpic(struct vm *vm); struct vatpit *vm_atpit(struct vm *vm); /* * Inject exception 'vme' into the guest vcpu. This function returns 0 on * success and non-zero on failure. * * Wrapper functions like 'vm_inject_gp()' should be preferred to calling * this function directly because they enforce the trap-like or fault-like * behavior of an exception. * * This function should only be called in the context of the thread that is * executing this vcpu. */ int vm_inject_exception(struct vm *vm, int vcpuid, struct vm_exception *vme); /* * Returns 0 if there is no exception pending for this vcpu. Returns 1 if an * exception is pending and also updates 'vme'. The pending exception is * cleared when this function returns. * * This function should only be called in the context of the thread that is * executing this vcpu. */ int vm_exception_pending(struct vm *vm, int vcpuid, struct vm_exception *vme); void vm_inject_gp(struct vm *vm, int vcpuid); /* general protection fault */ void vm_inject_ud(struct vm *vm, int vcpuid); /* undefined instruction fault */ void vm_inject_pf(struct vm *vm, int vcpuid, int error_code, uint64_t cr2); enum vm_reg_name vm_segment_name(int seg_encoding); #endif /* KERNEL */ #define VM_MAXCPU 16 /* maximum virtual cpus */ /* * Identifiers for architecturally defined registers. */ enum vm_reg_name { VM_REG_GUEST_RAX, VM_REG_GUEST_RBX, VM_REG_GUEST_RCX, VM_REG_GUEST_RDX, VM_REG_GUEST_RSI, VM_REG_GUEST_RDI, VM_REG_GUEST_RBP, VM_REG_GUEST_R8, VM_REG_GUEST_R9, VM_REG_GUEST_R10, VM_REG_GUEST_R11, VM_REG_GUEST_R12, VM_REG_GUEST_R13, VM_REG_GUEST_R14, VM_REG_GUEST_R15, VM_REG_GUEST_CR0, VM_REG_GUEST_CR3, VM_REG_GUEST_CR4, VM_REG_GUEST_DR7, VM_REG_GUEST_RSP, VM_REG_GUEST_RIP, VM_REG_GUEST_RFLAGS, VM_REG_GUEST_ES, VM_REG_GUEST_CS, VM_REG_GUEST_SS, VM_REG_GUEST_DS, VM_REG_GUEST_FS, VM_REG_GUEST_GS, VM_REG_GUEST_LDTR, VM_REG_GUEST_TR, VM_REG_GUEST_IDTR, VM_REG_GUEST_GDTR, VM_REG_GUEST_EFER, VM_REG_GUEST_CR2, VM_REG_LAST }; /* * Identifiers for optional vmm capabilities */ enum vm_cap_type { VM_CAP_HALT_EXIT, VM_CAP_MTRAP_EXIT, VM_CAP_PAUSE_EXIT, VM_CAP_UNRESTRICTED_GUEST, VM_CAP_ENABLE_INVPCID, VM_CAP_MAX }; enum x2apic_state { X2APIC_DISABLED, X2APIC_ENABLED, X2APIC_STATE_LAST }; enum vm_intr_trigger { EDGE_TRIGGER, LEVEL_TRIGGER }; /* * The 'access' field has the format specified in Table 21-2 of the Intel * Architecture Manual vol 3b. * * XXX The contents of the 'access' field are architecturally defined except * bit 16 - Segment Unusable. */ struct seg_desc { uint64_t base; uint32_t limit; uint32_t access; }; #define SEG_DESC_TYPE(desc) ((desc)->access & 0x001f) #define SEG_DESC_PRESENT(desc) ((desc)->access & 0x0080) #define SEG_DESC_DEF32(desc) ((desc)->access & 0x4000) #define SEG_DESC_GRANULARITY(desc) ((desc)->access & 0x8000) #define SEG_DESC_UNUSABLE(desc) ((desc)->access & 0x10000) enum vm_cpu_mode { CPU_MODE_COMPATIBILITY, /* IA-32E mode (CS.L = 0) */ CPU_MODE_64BIT, /* IA-32E mode (CS.L = 1) */ }; enum vm_paging_mode { PAGING_MODE_FLAT, PAGING_MODE_32, PAGING_MODE_PAE, PAGING_MODE_64, }; struct vm_guest_paging { uint64_t cr3; int cpl; enum vm_cpu_mode cpu_mode; enum vm_paging_mode paging_mode; }; /* * The data structures 'vie' and 'vie_op' are meant to be opaque to the * consumers of instruction decoding. The only reason why their contents * need to be exposed is because they are part of the 'vm_exit' structure. */ struct vie_op { uint8_t op_byte; /* actual opcode byte */ uint8_t op_type; /* type of operation (e.g. MOV) */ uint16_t op_flags; }; #define VIE_INST_SIZE 15 struct vie { uint8_t inst[VIE_INST_SIZE]; /* instruction bytes */ uint8_t num_valid; /* size of the instruction */ uint8_t num_processed; uint8_t rex_w:1, /* REX prefix */ rex_r:1, rex_x:1, rex_b:1, rex_present:1; uint8_t mod:2, /* ModRM byte */ reg:4, rm:4; uint8_t ss:2, /* SIB byte */ index:4, base:4; uint8_t disp_bytes; uint8_t imm_bytes; uint8_t scale; int base_register; /* VM_REG_GUEST_xyz */ int index_register; /* VM_REG_GUEST_xyz */ int64_t displacement; /* optional addr displacement */ int64_t immediate; /* optional immediate operand */ uint8_t decoded; /* set to 1 if successfully decoded */ struct vie_op op; /* opcode description */ }; enum vm_exitcode { VM_EXITCODE_INOUT, VM_EXITCODE_VMX, VM_EXITCODE_BOGUS, VM_EXITCODE_RDMSR, VM_EXITCODE_WRMSR, VM_EXITCODE_HLT, VM_EXITCODE_MTRAP, VM_EXITCODE_PAUSE, VM_EXITCODE_PAGING, VM_EXITCODE_INST_EMUL, VM_EXITCODE_SPINUP_AP, VM_EXITCODE_DEPRECATED1, /* used to be SPINDOWN_CPU */ VM_EXITCODE_RENDEZVOUS, VM_EXITCODE_IOAPIC_EOI, VM_EXITCODE_SUSPENDED, VM_EXITCODE_INOUT_STR, VM_EXITCODE_MAX }; struct vm_inout { uint16_t bytes:3; /* 1 or 2 or 4 */ uint16_t in:1; uint16_t string:1; uint16_t rep:1; uint16_t port; uint32_t eax; /* valid for out */ }; struct vm_inout_str { struct vm_inout inout; /* must be the first element */ struct vm_guest_paging paging; uint64_t rflags; uint64_t cr0; uint64_t index; uint64_t count; /* rep=1 (%rcx), rep=0 (1) */ int addrsize; enum vm_reg_name seg_name; struct seg_desc seg_desc; }; struct vm_exit { enum vm_exitcode exitcode; int inst_length; /* 0 means unknown */ uint64_t rip; union { struct vm_inout inout; struct vm_inout_str inout_str; struct { uint64_t gpa; int fault_type; } paging; struct { uint64_t gpa; uint64_t gla; struct vm_guest_paging paging; struct vie vie; } inst_emul; /* * VMX specific payload. Used when there is no "better" * exitcode to represent the VM-exit. */ struct { int status; /* vmx inst status */ /* * 'exit_reason' and 'exit_qualification' are valid * only if 'status' is zero. */ uint32_t exit_reason; uint64_t exit_qualification; /* * 'inst_error' and 'inst_type' are valid * only if 'status' is non-zero. */ int inst_type; int inst_error; } vmx; struct { uint32_t code; /* ecx value */ uint64_t wval; } msr; struct { int vcpu; uint64_t rip; } spinup_ap; struct { uint64_t rflags; } hlt; struct { int vector; } ioapic_eoi; struct { enum vm_suspend_how how; } suspended; } u; }; #endif /* _VMM_H_ */ Index: stable/10/sys/amd64/vmm/intel/vmx.c =================================================================== --- stable/10/sys/amd64/vmm/intel/vmx.c (revision 270073) +++ stable/10/sys/amd64/vmm/intel/vmx.c (revision 270074) @@ -1,3146 +1,3126 @@ /*- * Copyright (c) 2011 NetApp, Inc. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "vmm_host.h" #include "vmm_ioport.h" #include "vmm_ipi.h" #include "vmm_msr.h" #include "vmm_ktr.h" #include "vmm_stat.h" #include "vatpic.h" #include "vlapic.h" #include "vlapic_priv.h" #include "vmx_msr.h" #include "ept.h" #include "vmx_cpufunc.h" #include "vmx.h" #include "x86.h" #include "vmx_controls.h" #define PINBASED_CTLS_ONE_SETTING \ (PINBASED_EXTINT_EXITING | \ PINBASED_NMI_EXITING | \ PINBASED_VIRTUAL_NMI) #define PINBASED_CTLS_ZERO_SETTING 0 #define PROCBASED_CTLS_WINDOW_SETTING \ (PROCBASED_INT_WINDOW_EXITING | \ PROCBASED_NMI_WINDOW_EXITING) #define PROCBASED_CTLS_ONE_SETTING \ (PROCBASED_SECONDARY_CONTROLS | \ PROCBASED_IO_EXITING | \ PROCBASED_MSR_BITMAPS | \ PROCBASED_CTLS_WINDOW_SETTING | \ PROCBASED_CR8_LOAD_EXITING | \ PROCBASED_CR8_STORE_EXITING) #define PROCBASED_CTLS_ZERO_SETTING \ (PROCBASED_CR3_LOAD_EXITING | \ PROCBASED_CR3_STORE_EXITING | \ PROCBASED_IO_BITMAPS) #define PROCBASED_CTLS2_ONE_SETTING PROCBASED2_ENABLE_EPT #define PROCBASED_CTLS2_ZERO_SETTING 0 #define VM_EXIT_CTLS_ONE_SETTING_NO_PAT \ (VM_EXIT_HOST_LMA | \ VM_EXIT_SAVE_EFER | \ VM_EXIT_LOAD_EFER) #define VM_EXIT_CTLS_ONE_SETTING \ (VM_EXIT_CTLS_ONE_SETTING_NO_PAT | \ VM_EXIT_ACKNOWLEDGE_INTERRUPT | \ VM_EXIT_SAVE_PAT | \ VM_EXIT_LOAD_PAT) #define VM_EXIT_CTLS_ZERO_SETTING VM_EXIT_SAVE_DEBUG_CONTROLS #define VM_ENTRY_CTLS_ONE_SETTING_NO_PAT VM_ENTRY_LOAD_EFER #define VM_ENTRY_CTLS_ONE_SETTING \ (VM_ENTRY_CTLS_ONE_SETTING_NO_PAT | \ VM_ENTRY_LOAD_PAT) #define VM_ENTRY_CTLS_ZERO_SETTING \ (VM_ENTRY_LOAD_DEBUG_CONTROLS | \ VM_ENTRY_INTO_SMM | \ VM_ENTRY_DEACTIVATE_DUAL_MONITOR) #define guest_msr_rw(vmx, msr) \ msr_bitmap_change_access((vmx)->msr_bitmap, (msr), MSR_BITMAP_ACCESS_RW) #define guest_msr_ro(vmx, msr) \ msr_bitmap_change_access((vmx)->msr_bitmap, (msr), MSR_BITMAP_ACCESS_READ) #define HANDLED 1 #define UNHANDLED 0 static MALLOC_DEFINE(M_VMX, "vmx", "vmx"); static MALLOC_DEFINE(M_VLAPIC, "vlapic", "vlapic"); SYSCTL_DECL(_hw_vmm); SYSCTL_NODE(_hw_vmm, OID_AUTO, vmx, CTLFLAG_RW, NULL, NULL); int vmxon_enabled[MAXCPU]; static char vmxon_region[MAXCPU][PAGE_SIZE] __aligned(PAGE_SIZE); static uint32_t pinbased_ctls, procbased_ctls, procbased_ctls2; static uint32_t exit_ctls, entry_ctls; static uint64_t cr0_ones_mask, cr0_zeros_mask; SYSCTL_ULONG(_hw_vmm_vmx, OID_AUTO, cr0_ones_mask, CTLFLAG_RD, &cr0_ones_mask, 0, NULL); SYSCTL_ULONG(_hw_vmm_vmx, OID_AUTO, cr0_zeros_mask, CTLFLAG_RD, &cr0_zeros_mask, 0, NULL); static uint64_t cr4_ones_mask, cr4_zeros_mask; SYSCTL_ULONG(_hw_vmm_vmx, OID_AUTO, cr4_ones_mask, CTLFLAG_RD, &cr4_ones_mask, 0, NULL); SYSCTL_ULONG(_hw_vmm_vmx, OID_AUTO, cr4_zeros_mask, CTLFLAG_RD, &cr4_zeros_mask, 0, NULL); static int vmx_no_patmsr; static int vmx_initialized; SYSCTL_INT(_hw_vmm_vmx, OID_AUTO, initialized, CTLFLAG_RD, &vmx_initialized, 0, "Intel VMX initialized"); /* * Optional capabilities */ static int cap_halt_exit; static int cap_pause_exit; static int cap_unrestricted_guest; static int cap_monitor_trap; static int cap_invpcid; static int virtual_interrupt_delivery; SYSCTL_INT(_hw_vmm_vmx, OID_AUTO, virtual_interrupt_delivery, CTLFLAG_RD, &virtual_interrupt_delivery, 0, "APICv virtual interrupt delivery support"); static int posted_interrupts; SYSCTL_INT(_hw_vmm_vmx, OID_AUTO, posted_interrupts, CTLFLAG_RD, &posted_interrupts, 0, "APICv posted interrupt support"); static int pirvec; SYSCTL_INT(_hw_vmm_vmx, OID_AUTO, posted_interrupt_vector, CTLFLAG_RD, &pirvec, 0, "APICv posted interrupt vector"); static struct unrhdr *vpid_unr; static u_int vpid_alloc_failed; SYSCTL_UINT(_hw_vmm_vmx, OID_AUTO, vpid_alloc_failed, CTLFLAG_RD, &vpid_alloc_failed, 0, NULL); /* * Use the last page below 4GB as the APIC access address. This address is * occupied by the boot firmware so it is guaranteed that it will not conflict * with a page in system memory. */ #define APIC_ACCESS_ADDRESS 0xFFFFF000 static int vmx_getdesc(void *arg, int vcpu, int reg, struct seg_desc *desc); static int vmx_getreg(void *arg, int vcpu, int reg, uint64_t *retval); static void vmx_inject_pir(struct vlapic *vlapic); #ifdef KTR static const char * exit_reason_to_str(int reason) { static char reasonbuf[32]; switch (reason) { case EXIT_REASON_EXCEPTION: return "exception"; case EXIT_REASON_EXT_INTR: return "extint"; case EXIT_REASON_TRIPLE_FAULT: return "triplefault"; case EXIT_REASON_INIT: return "init"; case EXIT_REASON_SIPI: return "sipi"; case EXIT_REASON_IO_SMI: return "iosmi"; case EXIT_REASON_SMI: return "smi"; case EXIT_REASON_INTR_WINDOW: return "intrwindow"; case EXIT_REASON_NMI_WINDOW: return "nmiwindow"; case EXIT_REASON_TASK_SWITCH: return "taskswitch"; case EXIT_REASON_CPUID: return "cpuid"; case EXIT_REASON_GETSEC: return "getsec"; case EXIT_REASON_HLT: return "hlt"; case EXIT_REASON_INVD: return "invd"; case EXIT_REASON_INVLPG: return "invlpg"; case EXIT_REASON_RDPMC: return "rdpmc"; case EXIT_REASON_RDTSC: return "rdtsc"; case EXIT_REASON_RSM: return "rsm"; case EXIT_REASON_VMCALL: return "vmcall"; case EXIT_REASON_VMCLEAR: return "vmclear"; case EXIT_REASON_VMLAUNCH: return "vmlaunch"; case EXIT_REASON_VMPTRLD: return "vmptrld"; case EXIT_REASON_VMPTRST: return "vmptrst"; case EXIT_REASON_VMREAD: return "vmread"; case EXIT_REASON_VMRESUME: return "vmresume"; case EXIT_REASON_VMWRITE: return "vmwrite"; case EXIT_REASON_VMXOFF: return "vmxoff"; case EXIT_REASON_VMXON: return "vmxon"; case EXIT_REASON_CR_ACCESS: return "craccess"; case EXIT_REASON_DR_ACCESS: return "draccess"; case EXIT_REASON_INOUT: return "inout"; case EXIT_REASON_RDMSR: return "rdmsr"; case EXIT_REASON_WRMSR: return "wrmsr"; case EXIT_REASON_INVAL_VMCS: return "invalvmcs"; case EXIT_REASON_INVAL_MSR: return "invalmsr"; case EXIT_REASON_MWAIT: return "mwait"; case EXIT_REASON_MTF: return "mtf"; case EXIT_REASON_MONITOR: return "monitor"; case EXIT_REASON_PAUSE: return "pause"; case EXIT_REASON_MCE: return "mce"; case EXIT_REASON_TPR: return "tpr"; case EXIT_REASON_APIC_ACCESS: return "apic-access"; case EXIT_REASON_GDTR_IDTR: return "gdtridtr"; case EXIT_REASON_LDTR_TR: return "ldtrtr"; case EXIT_REASON_EPT_FAULT: return "eptfault"; case EXIT_REASON_EPT_MISCONFIG: return "eptmisconfig"; case EXIT_REASON_INVEPT: return "invept"; case EXIT_REASON_RDTSCP: return "rdtscp"; case EXIT_REASON_VMX_PREEMPT: return "vmxpreempt"; case EXIT_REASON_INVVPID: return "invvpid"; case EXIT_REASON_WBINVD: return "wbinvd"; case EXIT_REASON_XSETBV: return "xsetbv"; case EXIT_REASON_APIC_WRITE: return "apic-write"; default: snprintf(reasonbuf, sizeof(reasonbuf), "%d", reason); return (reasonbuf); } } #endif /* KTR */ static int vmx_allow_x2apic_msrs(struct vmx *vmx) { int i, error; error = 0; /* * Allow readonly access to the following x2APIC MSRs from the guest. */ error += guest_msr_ro(vmx, MSR_APIC_ID); error += guest_msr_ro(vmx, MSR_APIC_VERSION); error += guest_msr_ro(vmx, MSR_APIC_LDR); error += guest_msr_ro(vmx, MSR_APIC_SVR); for (i = 0; i < 8; i++) error += guest_msr_ro(vmx, MSR_APIC_ISR0 + i); for (i = 0; i < 8; i++) error += guest_msr_ro(vmx, MSR_APIC_TMR0 + i); for (i = 0; i < 8; i++) error += guest_msr_ro(vmx, MSR_APIC_IRR0 + i); error += guest_msr_ro(vmx, MSR_APIC_ESR); error += guest_msr_ro(vmx, MSR_APIC_LVT_TIMER); error += guest_msr_ro(vmx, MSR_APIC_LVT_THERMAL); error += guest_msr_ro(vmx, MSR_APIC_LVT_PCINT); error += guest_msr_ro(vmx, MSR_APIC_LVT_LINT0); error += guest_msr_ro(vmx, MSR_APIC_LVT_LINT1); error += guest_msr_ro(vmx, MSR_APIC_LVT_ERROR); error += guest_msr_ro(vmx, MSR_APIC_ICR_TIMER); error += guest_msr_ro(vmx, MSR_APIC_DCR_TIMER); error += guest_msr_ro(vmx, MSR_APIC_ICR); /* * Allow TPR, EOI and SELF_IPI MSRs to be read and written by the guest. * * These registers get special treatment described in the section * "Virtualizing MSR-Based APIC Accesses". */ error += guest_msr_rw(vmx, MSR_APIC_TPR); error += guest_msr_rw(vmx, MSR_APIC_EOI); error += guest_msr_rw(vmx, MSR_APIC_SELF_IPI); return (error); } u_long vmx_fix_cr0(u_long cr0) { return ((cr0 | cr0_ones_mask) & ~cr0_zeros_mask); } u_long vmx_fix_cr4(u_long cr4) { return ((cr4 | cr4_ones_mask) & ~cr4_zeros_mask); } static void vpid_free(int vpid) { if (vpid < 0 || vpid > 0xffff) panic("vpid_free: invalid vpid %d", vpid); /* * VPIDs [0,VM_MAXCPU] are special and are not allocated from * the unit number allocator. */ if (vpid > VM_MAXCPU) free_unr(vpid_unr, vpid); } static void vpid_alloc(uint16_t *vpid, int num) { int i, x; if (num <= 0 || num > VM_MAXCPU) panic("invalid number of vpids requested: %d", num); /* * If the "enable vpid" execution control is not enabled then the * VPID is required to be 0 for all vcpus. */ if ((procbased_ctls2 & PROCBASED2_ENABLE_VPID) == 0) { for (i = 0; i < num; i++) vpid[i] = 0; return; } /* * Allocate a unique VPID for each vcpu from the unit number allocator. */ for (i = 0; i < num; i++) { x = alloc_unr(vpid_unr); if (x == -1) break; else vpid[i] = x; } if (i < num) { atomic_add_int(&vpid_alloc_failed, 1); /* * If the unit number allocator does not have enough unique * VPIDs then we need to allocate from the [1,VM_MAXCPU] range. * * These VPIDs are not be unique across VMs but this does not * affect correctness because the combined mappings are also * tagged with the EP4TA which is unique for each VM. * * It is still sub-optimal because the invvpid will invalidate * combined mappings for a particular VPID across all EP4TAs. */ while (i-- > 0) vpid_free(vpid[i]); for (i = 0; i < num; i++) vpid[i] = i + 1; } } static void vpid_init(void) { /* * VPID 0 is required when the "enable VPID" execution control is * disabled. * * VPIDs [1,VM_MAXCPU] are used as the "overflow namespace" when the * unit number allocator does not have sufficient unique VPIDs to * satisfy the allocation. * * The remaining VPIDs are managed by the unit number allocator. */ vpid_unr = new_unrhdr(VM_MAXCPU + 1, 0xffff, NULL); } static void msr_save_area_init(struct msr_entry *g_area, int *g_count) { int cnt; static struct msr_entry guest_msrs[] = { { MSR_KGSBASE, 0, 0 }, }; cnt = sizeof(guest_msrs) / sizeof(guest_msrs[0]); if (cnt > GUEST_MSR_MAX_ENTRIES) panic("guest msr save area overrun"); bcopy(guest_msrs, g_area, sizeof(guest_msrs)); *g_count = cnt; } static void vmx_disable(void *arg __unused) { struct invvpid_desc invvpid_desc = { 0 }; struct invept_desc invept_desc = { 0 }; if (vmxon_enabled[curcpu]) { /* * See sections 25.3.3.3 and 25.3.3.4 in Intel Vol 3b. * * VMXON or VMXOFF are not required to invalidate any TLB * caching structures. This prevents potential retention of * cached information in the TLB between distinct VMX episodes. */ invvpid(INVVPID_TYPE_ALL_CONTEXTS, invvpid_desc); invept(INVEPT_TYPE_ALL_CONTEXTS, invept_desc); vmxoff(); } load_cr4(rcr4() & ~CR4_VMXE); } static int vmx_cleanup(void) { if (pirvec != 0) vmm_ipi_free(pirvec); if (vpid_unr != NULL) { delete_unrhdr(vpid_unr); vpid_unr = NULL; } smp_rendezvous(NULL, vmx_disable, NULL, NULL); return (0); } static void vmx_enable(void *arg __unused) { int error; uint64_t feature_control; feature_control = rdmsr(MSR_IA32_FEATURE_CONTROL); if ((feature_control & IA32_FEATURE_CONTROL_LOCK) == 0 || (feature_control & IA32_FEATURE_CONTROL_VMX_EN) == 0) { wrmsr(MSR_IA32_FEATURE_CONTROL, feature_control | IA32_FEATURE_CONTROL_VMX_EN | IA32_FEATURE_CONTROL_LOCK); } load_cr4(rcr4() | CR4_VMXE); *(uint32_t *)vmxon_region[curcpu] = vmx_revision(); error = vmxon(vmxon_region[curcpu]); if (error == 0) vmxon_enabled[curcpu] = 1; } static void vmx_restore(void) { if (vmxon_enabled[curcpu]) vmxon(vmxon_region[curcpu]); } static int vmx_init(int ipinum) { int error, use_tpr_shadow; uint64_t basic, fixed0, fixed1, feature_control; uint32_t tmp, procbased2_vid_bits; /* CPUID.1:ECX[bit 5] must be 1 for processor to support VMX */ if (!(cpu_feature2 & CPUID2_VMX)) { printf("vmx_init: processor does not support VMX operation\n"); return (ENXIO); } /* * Verify that MSR_IA32_FEATURE_CONTROL lock and VMXON enable bits * are set (bits 0 and 2 respectively). */ feature_control = rdmsr(MSR_IA32_FEATURE_CONTROL); if ((feature_control & IA32_FEATURE_CONTROL_LOCK) == 1 && (feature_control & IA32_FEATURE_CONTROL_VMX_EN) == 0) { printf("vmx_init: VMX operation disabled by BIOS\n"); return (ENXIO); } /* * Verify capabilities MSR_VMX_BASIC: * - bit 54 indicates support for INS/OUTS decoding */ basic = rdmsr(MSR_VMX_BASIC); if ((basic & (1UL << 54)) == 0) { printf("vmx_init: processor does not support desired basic " "capabilities\n"); return (EINVAL); } /* Check support for primary processor-based VM-execution controls */ error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS, MSR_VMX_TRUE_PROCBASED_CTLS, PROCBASED_CTLS_ONE_SETTING, PROCBASED_CTLS_ZERO_SETTING, &procbased_ctls); if (error) { printf("vmx_init: processor does not support desired primary " "processor-based controls\n"); return (error); } /* Clear the processor-based ctl bits that are set on demand */ procbased_ctls &= ~PROCBASED_CTLS_WINDOW_SETTING; /* Check support for secondary processor-based VM-execution controls */ error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2, MSR_VMX_PROCBASED_CTLS2, PROCBASED_CTLS2_ONE_SETTING, PROCBASED_CTLS2_ZERO_SETTING, &procbased_ctls2); if (error) { printf("vmx_init: processor does not support desired secondary " "processor-based controls\n"); return (error); } /* Check support for VPID */ error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2, MSR_VMX_PROCBASED_CTLS2, PROCBASED2_ENABLE_VPID, 0, &tmp); if (error == 0) procbased_ctls2 |= PROCBASED2_ENABLE_VPID; /* Check support for pin-based VM-execution controls */ error = vmx_set_ctlreg(MSR_VMX_PINBASED_CTLS, MSR_VMX_TRUE_PINBASED_CTLS, PINBASED_CTLS_ONE_SETTING, PINBASED_CTLS_ZERO_SETTING, &pinbased_ctls); if (error) { printf("vmx_init: processor does not support desired " "pin-based controls\n"); return (error); } /* Check support for VM-exit controls */ error = vmx_set_ctlreg(MSR_VMX_EXIT_CTLS, MSR_VMX_TRUE_EXIT_CTLS, VM_EXIT_CTLS_ONE_SETTING, VM_EXIT_CTLS_ZERO_SETTING, &exit_ctls); if (error) { /* Try again without the PAT MSR bits */ error = vmx_set_ctlreg(MSR_VMX_EXIT_CTLS, MSR_VMX_TRUE_EXIT_CTLS, VM_EXIT_CTLS_ONE_SETTING_NO_PAT, VM_EXIT_CTLS_ZERO_SETTING, &exit_ctls); if (error) { printf("vmx_init: processor does not support desired " "exit controls\n"); return (error); } else { if (bootverbose) printf("vmm: PAT MSR access not supported\n"); guest_msr_valid(MSR_PAT); vmx_no_patmsr = 1; } } /* Check support for VM-entry controls */ if (!vmx_no_patmsr) { error = vmx_set_ctlreg(MSR_VMX_ENTRY_CTLS, MSR_VMX_TRUE_ENTRY_CTLS, VM_ENTRY_CTLS_ONE_SETTING, VM_ENTRY_CTLS_ZERO_SETTING, &entry_ctls); } else { error = vmx_set_ctlreg(MSR_VMX_ENTRY_CTLS, MSR_VMX_TRUE_ENTRY_CTLS, VM_ENTRY_CTLS_ONE_SETTING_NO_PAT, VM_ENTRY_CTLS_ZERO_SETTING, &entry_ctls); } if (error) { printf("vmx_init: processor does not support desired " "entry controls\n"); return (error); } /* * Check support for optional features by testing them * as individual bits */ cap_halt_exit = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS, MSR_VMX_TRUE_PROCBASED_CTLS, PROCBASED_HLT_EXITING, 0, &tmp) == 0); cap_monitor_trap = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS, MSR_VMX_PROCBASED_CTLS, PROCBASED_MTF, 0, &tmp) == 0); cap_pause_exit = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS, MSR_VMX_TRUE_PROCBASED_CTLS, PROCBASED_PAUSE_EXITING, 0, &tmp) == 0); cap_unrestricted_guest = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2, MSR_VMX_PROCBASED_CTLS2, PROCBASED2_UNRESTRICTED_GUEST, 0, &tmp) == 0); cap_invpcid = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2, MSR_VMX_PROCBASED_CTLS2, PROCBASED2_ENABLE_INVPCID, 0, &tmp) == 0); /* * Check support for virtual interrupt delivery. */ procbased2_vid_bits = (PROCBASED2_VIRTUALIZE_APIC_ACCESSES | PROCBASED2_VIRTUALIZE_X2APIC_MODE | PROCBASED2_APIC_REGISTER_VIRTUALIZATION | PROCBASED2_VIRTUAL_INTERRUPT_DELIVERY); use_tpr_shadow = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS, MSR_VMX_TRUE_PROCBASED_CTLS, PROCBASED_USE_TPR_SHADOW, 0, &tmp) == 0); error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2, MSR_VMX_PROCBASED_CTLS2, procbased2_vid_bits, 0, &tmp); if (error == 0 && use_tpr_shadow) { virtual_interrupt_delivery = 1; TUNABLE_INT_FETCH("hw.vmm.vmx.use_apic_vid", &virtual_interrupt_delivery); } if (virtual_interrupt_delivery) { procbased_ctls |= PROCBASED_USE_TPR_SHADOW; procbased_ctls2 |= procbased2_vid_bits; procbased_ctls2 &= ~PROCBASED2_VIRTUALIZE_X2APIC_MODE; /* * No need to emulate accesses to %CR8 if virtual * interrupt delivery is enabled. */ procbased_ctls &= ~PROCBASED_CR8_LOAD_EXITING; procbased_ctls &= ~PROCBASED_CR8_STORE_EXITING; /* * Check for Posted Interrupts only if Virtual Interrupt * Delivery is enabled. */ error = vmx_set_ctlreg(MSR_VMX_PINBASED_CTLS, MSR_VMX_TRUE_PINBASED_CTLS, PINBASED_POSTED_INTERRUPT, 0, &tmp); if (error == 0) { pirvec = vmm_ipi_alloc(); if (pirvec == 0) { if (bootverbose) { printf("vmx_init: unable to allocate " "posted interrupt vector\n"); } } else { posted_interrupts = 1; TUNABLE_INT_FETCH("hw.vmm.vmx.use_apic_pir", &posted_interrupts); } } } if (posted_interrupts) pinbased_ctls |= PINBASED_POSTED_INTERRUPT; /* Initialize EPT */ error = ept_init(ipinum); if (error) { printf("vmx_init: ept initialization failed (%d)\n", error); return (error); } /* * Stash the cr0 and cr4 bits that must be fixed to 0 or 1 */ fixed0 = rdmsr(MSR_VMX_CR0_FIXED0); fixed1 = rdmsr(MSR_VMX_CR0_FIXED1); cr0_ones_mask = fixed0 & fixed1; cr0_zeros_mask = ~fixed0 & ~fixed1; /* * CR0_PE and CR0_PG can be set to zero in VMX non-root operation * if unrestricted guest execution is allowed. */ if (cap_unrestricted_guest) cr0_ones_mask &= ~(CR0_PG | CR0_PE); /* * Do not allow the guest to set CR0_NW or CR0_CD. */ cr0_zeros_mask |= (CR0_NW | CR0_CD); fixed0 = rdmsr(MSR_VMX_CR4_FIXED0); fixed1 = rdmsr(MSR_VMX_CR4_FIXED1); cr4_ones_mask = fixed0 & fixed1; cr4_zeros_mask = ~fixed0 & ~fixed1; vpid_init(); /* enable VMX operation */ smp_rendezvous(NULL, vmx_enable, NULL, NULL); vmx_initialized = 1; return (0); } static void vmx_trigger_hostintr(int vector) { uintptr_t func; struct gate_descriptor *gd; gd = &idt[vector]; KASSERT(vector >= 32 && vector <= 255, ("vmx_trigger_hostintr: " "invalid vector %d", vector)); KASSERT(gd->gd_p == 1, ("gate descriptor for vector %d not present", vector)); KASSERT(gd->gd_type == SDT_SYSIGT, ("gate descriptor for vector %d " "has invalid type %d", vector, gd->gd_type)); KASSERT(gd->gd_dpl == SEL_KPL, ("gate descriptor for vector %d " "has invalid dpl %d", vector, gd->gd_dpl)); KASSERT(gd->gd_selector == GSEL(GCODE_SEL, SEL_KPL), ("gate descriptor " "for vector %d has invalid selector %d", vector, gd->gd_selector)); KASSERT(gd->gd_ist == 0, ("gate descriptor for vector %d has invalid " "IST %d", vector, gd->gd_ist)); func = ((long)gd->gd_hioffset << 16 | gd->gd_looffset); vmx_call_isr(func); } static int vmx_setup_cr_shadow(int which, struct vmcs *vmcs, uint32_t initial) { int error, mask_ident, shadow_ident; uint64_t mask_value; if (which != 0 && which != 4) panic("vmx_setup_cr_shadow: unknown cr%d", which); if (which == 0) { mask_ident = VMCS_CR0_MASK; mask_value = cr0_ones_mask | cr0_zeros_mask; shadow_ident = VMCS_CR0_SHADOW; } else { mask_ident = VMCS_CR4_MASK; mask_value = cr4_ones_mask | cr4_zeros_mask; shadow_ident = VMCS_CR4_SHADOW; } error = vmcs_setreg(vmcs, 0, VMCS_IDENT(mask_ident), mask_value); if (error) return (error); error = vmcs_setreg(vmcs, 0, VMCS_IDENT(shadow_ident), initial); if (error) return (error); return (0); } #define vmx_setup_cr0_shadow(vmcs,init) vmx_setup_cr_shadow(0, (vmcs), (init)) #define vmx_setup_cr4_shadow(vmcs,init) vmx_setup_cr_shadow(4, (vmcs), (init)) static void * vmx_vminit(struct vm *vm, pmap_t pmap) { uint16_t vpid[VM_MAXCPU]; int i, error, guest_msr_count; struct vmx *vmx; struct vmcs *vmcs; vmx = malloc(sizeof(struct vmx), M_VMX, M_WAITOK | M_ZERO); if ((uintptr_t)vmx & PAGE_MASK) { panic("malloc of struct vmx not aligned on %d byte boundary", PAGE_SIZE); } vmx->vm = vm; vmx->eptp = eptp(vtophys((vm_offset_t)pmap->pm_pml4)); /* * Clean up EPTP-tagged guest physical and combined mappings * * VMX transitions are not required to invalidate any guest physical * mappings. So, it may be possible for stale guest physical mappings * to be present in the processor TLBs. * * Combined mappings for this EP4TA are also invalidated for all VPIDs. */ ept_invalidate_mappings(vmx->eptp); msr_bitmap_initialize(vmx->msr_bitmap); /* * It is safe to allow direct access to MSR_GSBASE and MSR_FSBASE. * The guest FSBASE and GSBASE are saved and restored during * vm-exit and vm-entry respectively. The host FSBASE and GSBASE are * always restored from the vmcs host state area on vm-exit. * * The SYSENTER_CS/ESP/EIP MSRs are identical to FS/GSBASE in * how they are saved/restored so can be directly accessed by the * guest. * * Guest KGSBASE is saved and restored in the guest MSR save area. * Host KGSBASE is restored before returning to userland from the pcb. * There will be a window of time when we are executing in the host * kernel context with a value of KGSBASE from the guest. This is ok * because the value of KGSBASE is inconsequential in kernel context. * * MSR_EFER is saved and restored in the guest VMCS area on a * VM exit and entry respectively. It is also restored from the * host VMCS area on a VM exit. * * The TSC MSR is exposed read-only. Writes are disallowed as that * will impact the host TSC. * XXX Writes would be implemented with a wrmsr trap, and * then modifying the TSC offset in the VMCS. */ if (guest_msr_rw(vmx, MSR_GSBASE) || guest_msr_rw(vmx, MSR_FSBASE) || guest_msr_rw(vmx, MSR_SYSENTER_CS_MSR) || guest_msr_rw(vmx, MSR_SYSENTER_ESP_MSR) || guest_msr_rw(vmx, MSR_SYSENTER_EIP_MSR) || guest_msr_rw(vmx, MSR_KGSBASE) || guest_msr_rw(vmx, MSR_EFER) || guest_msr_ro(vmx, MSR_TSC)) panic("vmx_vminit: error setting guest msr access"); /* * MSR_PAT is saved and restored in the guest VMCS are on a VM exit * and entry respectively. It is also restored from the host VMCS * area on a VM exit. However, if running on a system with no * MSR_PAT save/restore support, leave access disabled so accesses * will be trapped. */ if (!vmx_no_patmsr && guest_msr_rw(vmx, MSR_PAT)) panic("vmx_vminit: error setting guest pat msr access"); vpid_alloc(vpid, VM_MAXCPU); if (virtual_interrupt_delivery) { error = vm_map_mmio(vm, DEFAULT_APIC_BASE, PAGE_SIZE, APIC_ACCESS_ADDRESS); /* XXX this should really return an error to the caller */ KASSERT(error == 0, ("vm_map_mmio(apicbase) error %d", error)); } for (i = 0; i < VM_MAXCPU; i++) { vmcs = &vmx->vmcs[i]; vmcs->identifier = vmx_revision(); error = vmclear(vmcs); if (error != 0) { panic("vmx_vminit: vmclear error %d on vcpu %d\n", error, i); } error = vmcs_init(vmcs); KASSERT(error == 0, ("vmcs_init error %d", error)); VMPTRLD(vmcs); error = 0; error += vmwrite(VMCS_HOST_RSP, (u_long)&vmx->ctx[i]); error += vmwrite(VMCS_EPTP, vmx->eptp); error += vmwrite(VMCS_PIN_BASED_CTLS, pinbased_ctls); error += vmwrite(VMCS_PRI_PROC_BASED_CTLS, procbased_ctls); error += vmwrite(VMCS_SEC_PROC_BASED_CTLS, procbased_ctls2); error += vmwrite(VMCS_EXIT_CTLS, exit_ctls); error += vmwrite(VMCS_ENTRY_CTLS, entry_ctls); error += vmwrite(VMCS_MSR_BITMAP, vtophys(vmx->msr_bitmap)); error += vmwrite(VMCS_VPID, vpid[i]); if (virtual_interrupt_delivery) { error += vmwrite(VMCS_APIC_ACCESS, APIC_ACCESS_ADDRESS); error += vmwrite(VMCS_VIRTUAL_APIC, vtophys(&vmx->apic_page[i])); error += vmwrite(VMCS_EOI_EXIT0, 0); error += vmwrite(VMCS_EOI_EXIT1, 0); error += vmwrite(VMCS_EOI_EXIT2, 0); error += vmwrite(VMCS_EOI_EXIT3, 0); } if (posted_interrupts) { error += vmwrite(VMCS_PIR_VECTOR, pirvec); error += vmwrite(VMCS_PIR_DESC, vtophys(&vmx->pir_desc[i])); } VMCLEAR(vmcs); KASSERT(error == 0, ("vmx_vminit: error customizing the vmcs")); vmx->cap[i].set = 0; vmx->cap[i].proc_ctls = procbased_ctls; vmx->cap[i].proc_ctls2 = procbased_ctls2; vmx->state[i].lastcpu = -1; vmx->state[i].vpid = vpid[i]; msr_save_area_init(vmx->guest_msrs[i], &guest_msr_count); error = vmcs_set_msr_save(vmcs, vtophys(vmx->guest_msrs[i]), guest_msr_count); if (error != 0) panic("vmcs_set_msr_save error %d", error); /* * Set up the CR0/4 shadows, and init the read shadow * to the power-on register value from the Intel Sys Arch. * CR0 - 0x60000010 * CR4 - 0 */ error = vmx_setup_cr0_shadow(vmcs, 0x60000010); if (error != 0) panic("vmx_setup_cr0_shadow %d", error); error = vmx_setup_cr4_shadow(vmcs, 0); if (error != 0) panic("vmx_setup_cr4_shadow %d", error); vmx->ctx[i].pmap = pmap; } return (vmx); } static int vmx_handle_cpuid(struct vm *vm, int vcpu, struct vmxctx *vmxctx) { int handled, func; func = vmxctx->guest_rax; handled = x86_emulate_cpuid(vm, vcpu, (uint32_t*)(&vmxctx->guest_rax), (uint32_t*)(&vmxctx->guest_rbx), (uint32_t*)(&vmxctx->guest_rcx), (uint32_t*)(&vmxctx->guest_rdx)); return (handled); } static __inline void vmx_run_trace(struct vmx *vmx, int vcpu) { #ifdef KTR VCPU_CTR1(vmx->vm, vcpu, "Resume execution at %#lx", vmcs_guest_rip()); #endif } static __inline void vmx_exit_trace(struct vmx *vmx, int vcpu, uint64_t rip, uint32_t exit_reason, int handled) { #ifdef KTR VCPU_CTR3(vmx->vm, vcpu, "%s %s vmexit at 0x%0lx", handled ? "handled" : "unhandled", exit_reason_to_str(exit_reason), rip); #endif } static __inline void vmx_astpending_trace(struct vmx *vmx, int vcpu, uint64_t rip) { #ifdef KTR VCPU_CTR1(vmx->vm, vcpu, "astpending vmexit at 0x%0lx", rip); #endif } static VMM_STAT_INTEL(VCPU_INVVPID_SAVED, "Number of vpid invalidations saved"); static void vmx_set_pcpu_defaults(struct vmx *vmx, int vcpu, pmap_t pmap) { struct vmxstate *vmxstate; struct invvpid_desc invvpid_desc; vmxstate = &vmx->state[vcpu]; if (vmxstate->lastcpu == curcpu) return; vmxstate->lastcpu = curcpu; vmm_stat_incr(vmx->vm, vcpu, VCPU_MIGRATIONS, 1); vmcs_write(VMCS_HOST_TR_BASE, vmm_get_host_trbase()); vmcs_write(VMCS_HOST_GDTR_BASE, vmm_get_host_gdtrbase()); vmcs_write(VMCS_HOST_GS_BASE, vmm_get_host_gsbase()); /* * If we are using VPIDs then invalidate all mappings tagged with 'vpid' * * We do this because this vcpu was executing on a different host * cpu when it last ran. We do not track whether it invalidated * mappings associated with its 'vpid' during that run. So we must * assume that the mappings associated with 'vpid' on 'curcpu' are * stale and invalidate them. * * Note that we incur this penalty only when the scheduler chooses to * move the thread associated with this vcpu between host cpus. * * Note also that this will invalidate mappings tagged with 'vpid' * for "all" EP4TAs. */ if (vmxstate->vpid != 0) { if (pmap->pm_eptgen == vmx->eptgen[curcpu]) { invvpid_desc._res1 = 0; invvpid_desc._res2 = 0; invvpid_desc.vpid = vmxstate->vpid; invvpid_desc.linear_addr = 0; invvpid(INVVPID_TYPE_SINGLE_CONTEXT, invvpid_desc); } else { /* * The invvpid can be skipped if an invept is going to * be performed before entering the guest. The invept * will invalidate combined mappings tagged with * 'vmx->eptp' for all vpids. */ vmm_stat_incr(vmx->vm, vcpu, VCPU_INVVPID_SAVED, 1); } } } /* * We depend on 'procbased_ctls' to have the Interrupt Window Exiting bit set. */ CTASSERT((PROCBASED_CTLS_ONE_SETTING & PROCBASED_INT_WINDOW_EXITING) != 0); static void __inline vmx_set_int_window_exiting(struct vmx *vmx, int vcpu) { if ((vmx->cap[vcpu].proc_ctls & PROCBASED_INT_WINDOW_EXITING) == 0) { vmx->cap[vcpu].proc_ctls |= PROCBASED_INT_WINDOW_EXITING; vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vmx->cap[vcpu].proc_ctls); VCPU_CTR0(vmx->vm, vcpu, "Enabling interrupt window exiting"); } } static void __inline vmx_clear_int_window_exiting(struct vmx *vmx, int vcpu) { KASSERT((vmx->cap[vcpu].proc_ctls & PROCBASED_INT_WINDOW_EXITING) != 0, ("intr_window_exiting not set: %#x", vmx->cap[vcpu].proc_ctls)); vmx->cap[vcpu].proc_ctls &= ~PROCBASED_INT_WINDOW_EXITING; vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vmx->cap[vcpu].proc_ctls); VCPU_CTR0(vmx->vm, vcpu, "Disabling interrupt window exiting"); } static void __inline vmx_set_nmi_window_exiting(struct vmx *vmx, int vcpu) { if ((vmx->cap[vcpu].proc_ctls & PROCBASED_NMI_WINDOW_EXITING) == 0) { vmx->cap[vcpu].proc_ctls |= PROCBASED_NMI_WINDOW_EXITING; vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vmx->cap[vcpu].proc_ctls); VCPU_CTR0(vmx->vm, vcpu, "Enabling NMI window exiting"); } } static void __inline vmx_clear_nmi_window_exiting(struct vmx *vmx, int vcpu) { KASSERT((vmx->cap[vcpu].proc_ctls & PROCBASED_NMI_WINDOW_EXITING) != 0, ("nmi_window_exiting not set %#x", vmx->cap[vcpu].proc_ctls)); vmx->cap[vcpu].proc_ctls &= ~PROCBASED_NMI_WINDOW_EXITING; vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vmx->cap[vcpu].proc_ctls); VCPU_CTR0(vmx->vm, vcpu, "Disabling NMI window exiting"); } #define NMI_BLOCKING (VMCS_INTERRUPTIBILITY_NMI_BLOCKING | \ VMCS_INTERRUPTIBILITY_MOVSS_BLOCKING) #define HWINTR_BLOCKING (VMCS_INTERRUPTIBILITY_STI_BLOCKING | \ VMCS_INTERRUPTIBILITY_MOVSS_BLOCKING) static void vmx_inject_nmi(struct vmx *vmx, int vcpu) { uint32_t gi, info; gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY); KASSERT((gi & NMI_BLOCKING) == 0, ("vmx_inject_nmi: invalid guest " "interruptibility-state %#x", gi)); info = vmcs_read(VMCS_ENTRY_INTR_INFO); KASSERT((info & VMCS_INTR_VALID) == 0, ("vmx_inject_nmi: invalid " "VM-entry interruption information %#x", info)); /* * Inject the virtual NMI. The vector must be the NMI IDT entry * or the VMCS entry check will fail. */ info = IDT_NMI | VMCS_INTR_T_NMI | VMCS_INTR_VALID; vmcs_write(VMCS_ENTRY_INTR_INFO, info); VCPU_CTR0(vmx->vm, vcpu, "Injecting vNMI"); /* Clear the request */ vm_nmi_clear(vmx->vm, vcpu); } static void vmx_inject_interrupts(struct vmx *vmx, int vcpu, struct vlapic *vlapic) { struct vm_exception exc; int vector, need_nmi_exiting, extint_pending; uint64_t rflags; uint32_t gi, info; if (vm_exception_pending(vmx->vm, vcpu, &exc)) { KASSERT(exc.vector >= 0 && exc.vector < 32, ("%s: invalid exception vector %d", __func__, exc.vector)); info = vmcs_read(VMCS_ENTRY_INTR_INFO); KASSERT((info & VMCS_INTR_VALID) == 0, ("%s: cannot inject " "pending exception %d: %#x", __func__, exc.vector, info)); info = exc.vector | VMCS_INTR_T_HWEXCEPTION | VMCS_INTR_VALID; if (exc.error_code_valid) { info |= VMCS_INTR_DEL_ERRCODE; vmcs_write(VMCS_ENTRY_EXCEPTION_ERROR, exc.error_code); } vmcs_write(VMCS_ENTRY_INTR_INFO, info); } if (vm_nmi_pending(vmx->vm, vcpu)) { /* * If there are no conditions blocking NMI injection then * inject it directly here otherwise enable "NMI window * exiting" to inject it as soon as we can. * * We also check for STI_BLOCKING because some implementations * don't allow NMI injection in this case. If we are running * on a processor that doesn't have this restriction it will * immediately exit and the NMI will be injected in the * "NMI window exiting" handler. */ need_nmi_exiting = 1; gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY); if ((gi & (HWINTR_BLOCKING | NMI_BLOCKING)) == 0) { info = vmcs_read(VMCS_ENTRY_INTR_INFO); if ((info & VMCS_INTR_VALID) == 0) { vmx_inject_nmi(vmx, vcpu); need_nmi_exiting = 0; } else { VCPU_CTR1(vmx->vm, vcpu, "Cannot inject NMI " "due to VM-entry intr info %#x", info); } } else { VCPU_CTR1(vmx->vm, vcpu, "Cannot inject NMI due to " "Guest Interruptibility-state %#x", gi); } if (need_nmi_exiting) vmx_set_nmi_window_exiting(vmx, vcpu); } extint_pending = vm_extint_pending(vmx->vm, vcpu); if (!extint_pending && virtual_interrupt_delivery) { vmx_inject_pir(vlapic); return; } /* * If interrupt-window exiting is already in effect then don't bother * checking for pending interrupts. This is just an optimization and * not needed for correctness. */ if ((vmx->cap[vcpu].proc_ctls & PROCBASED_INT_WINDOW_EXITING) != 0) { VCPU_CTR0(vmx->vm, vcpu, "Skip interrupt injection due to " "pending int_window_exiting"); return; } if (!extint_pending) { /* Ask the local apic for a vector to inject */ if (!vlapic_pending_intr(vlapic, &vector)) return; /* * From the Intel SDM, Volume 3, Section "Maskable * Hardware Interrupts": * - maskable interrupt vectors [16,255] can be delivered * through the local APIC. */ KASSERT(vector >= 16 && vector <= 255, ("invalid vector %d from local APIC", vector)); } else { /* Ask the legacy pic for a vector to inject */ vatpic_pending_intr(vmx->vm, &vector); /* * From the Intel SDM, Volume 3, Section "Maskable * Hardware Interrupts": * - maskable interrupt vectors [0,255] can be delivered * through the INTR pin. */ KASSERT(vector >= 0 && vector <= 255, ("invalid vector %d from INTR", vector)); } /* Check RFLAGS.IF and the interruptibility state of the guest */ rflags = vmcs_read(VMCS_GUEST_RFLAGS); if ((rflags & PSL_I) == 0) { VCPU_CTR2(vmx->vm, vcpu, "Cannot inject vector %d due to " "rflags %#lx", vector, rflags); goto cantinject; } gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY); if (gi & HWINTR_BLOCKING) { VCPU_CTR2(vmx->vm, vcpu, "Cannot inject vector %d due to " "Guest Interruptibility-state %#x", vector, gi); goto cantinject; } info = vmcs_read(VMCS_ENTRY_INTR_INFO); if (info & VMCS_INTR_VALID) { /* * This is expected and could happen for multiple reasons: * - A vectoring VM-entry was aborted due to astpending * - A VM-exit happened during event injection. * - An exception was injected above. * - An NMI was injected above or after "NMI window exiting" */ VCPU_CTR2(vmx->vm, vcpu, "Cannot inject vector %d due to " "VM-entry intr info %#x", vector, info); goto cantinject; } /* Inject the interrupt */ info = VMCS_INTR_T_HWINTR | VMCS_INTR_VALID; info |= vector; vmcs_write(VMCS_ENTRY_INTR_INFO, info); if (!extint_pending) { /* Update the Local APIC ISR */ vlapic_intr_accepted(vlapic, vector); } else { vm_extint_clear(vmx->vm, vcpu); vatpic_intr_accepted(vmx->vm, vector); /* * After we accepted the current ExtINT the PIC may * have posted another one. If that is the case, set * the Interrupt Window Exiting execution control so * we can inject that one too. + * + * Also, interrupt window exiting allows us to inject any + * pending APIC vector that was preempted by the ExtINT + * as soon as possible. This applies both for the software + * emulated vlapic and the hardware assisted virtual APIC. */ - if (vm_extint_pending(vmx->vm, vcpu)) - vmx_set_int_window_exiting(vmx, vcpu); + vmx_set_int_window_exiting(vmx, vcpu); } VCPU_CTR1(vmx->vm, vcpu, "Injecting hwintr at vector %d", vector); return; cantinject: /* * Set the Interrupt Window Exiting execution control so we can inject * the interrupt as soon as blocking condition goes away. */ vmx_set_int_window_exiting(vmx, vcpu); } /* * If the Virtual NMIs execution control is '1' then the logical processor * tracks virtual-NMI blocking in the Guest Interruptibility-state field of * the VMCS. An IRET instruction in VMX non-root operation will remove any * virtual-NMI blocking. * * This unblocking occurs even if the IRET causes a fault. In this case the * hypervisor needs to restore virtual-NMI blocking before resuming the guest. */ static void vmx_restore_nmi_blocking(struct vmx *vmx, int vcpuid) { uint32_t gi; VCPU_CTR0(vmx->vm, vcpuid, "Restore Virtual-NMI blocking"); gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY); gi |= VMCS_INTERRUPTIBILITY_NMI_BLOCKING; vmcs_write(VMCS_GUEST_INTERRUPTIBILITY, gi); } static void vmx_clear_nmi_blocking(struct vmx *vmx, int vcpuid) { uint32_t gi; VCPU_CTR0(vmx->vm, vcpuid, "Clear Virtual-NMI blocking"); gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY); gi &= ~VMCS_INTERRUPTIBILITY_NMI_BLOCKING; vmcs_write(VMCS_GUEST_INTERRUPTIBILITY, gi); } static int vmx_emulate_xsetbv(struct vmx *vmx, int vcpu, struct vm_exit *vmexit) { struct vmxctx *vmxctx; uint64_t xcrval; const struct xsave_limits *limits; vmxctx = &vmx->ctx[vcpu]; limits = vmm_get_xsave_limits(); /* * Note that the processor raises a GP# fault on its own if * xsetbv is executed for CPL != 0, so we do not have to * emulate that fault here. */ /* Only xcr0 is supported. */ if (vmxctx->guest_rcx != 0) { vm_inject_gp(vmx->vm, vcpu); return (HANDLED); } /* We only handle xcr0 if both the host and guest have XSAVE enabled. */ if (!limits->xsave_enabled || !(vmcs_read(VMCS_GUEST_CR4) & CR4_XSAVE)) { vm_inject_ud(vmx->vm, vcpu); return (HANDLED); } xcrval = vmxctx->guest_rdx << 32 | (vmxctx->guest_rax & 0xffffffff); if ((xcrval & ~limits->xcr0_allowed) != 0) { vm_inject_gp(vmx->vm, vcpu); return (HANDLED); } if (!(xcrval & XFEATURE_ENABLED_X87)) { vm_inject_gp(vmx->vm, vcpu); return (HANDLED); } /* AVX (YMM_Hi128) requires SSE. */ if (xcrval & XFEATURE_ENABLED_AVX && (xcrval & XFEATURE_AVX) != XFEATURE_AVX) { vm_inject_gp(vmx->vm, vcpu); return (HANDLED); } /* * AVX512 requires base AVX (YMM_Hi128) as well as OpMask, * ZMM_Hi256, and Hi16_ZMM. */ if (xcrval & XFEATURE_AVX512 && (xcrval & (XFEATURE_AVX512 | XFEATURE_AVX)) != (XFEATURE_AVX512 | XFEATURE_AVX)) { vm_inject_gp(vmx->vm, vcpu); return (HANDLED); } /* * Intel MPX requires both bound register state flags to be * set. */ if (((xcrval & XFEATURE_ENABLED_BNDREGS) != 0) != ((xcrval & XFEATURE_ENABLED_BNDCSR) != 0)) { vm_inject_gp(vmx->vm, vcpu); return (HANDLED); } /* * This runs "inside" vmrun() with the guest's FPU state, so * modifying xcr0 directly modifies the guest's xcr0, not the * host's. */ load_xcr(0, xcrval); return (HANDLED); } static uint64_t vmx_get_guest_reg(struct vmx *vmx, int vcpu, int ident) { const struct vmxctx *vmxctx; vmxctx = &vmx->ctx[vcpu]; switch (ident) { case 0: return (vmxctx->guest_rax); case 1: return (vmxctx->guest_rcx); case 2: return (vmxctx->guest_rdx); case 3: return (vmxctx->guest_rbx); case 4: return (vmcs_read(VMCS_GUEST_RSP)); case 5: return (vmxctx->guest_rbp); case 6: return (vmxctx->guest_rsi); case 7: return (vmxctx->guest_rdi); case 8: return (vmxctx->guest_r8); case 9: return (vmxctx->guest_r9); case 10: return (vmxctx->guest_r10); case 11: return (vmxctx->guest_r11); case 12: return (vmxctx->guest_r12); case 13: return (vmxctx->guest_r13); case 14: return (vmxctx->guest_r14); case 15: return (vmxctx->guest_r15); default: panic("invalid vmx register %d", ident); } } static void vmx_set_guest_reg(struct vmx *vmx, int vcpu, int ident, uint64_t regval) { struct vmxctx *vmxctx; vmxctx = &vmx->ctx[vcpu]; switch (ident) { case 0: vmxctx->guest_rax = regval; break; case 1: vmxctx->guest_rcx = regval; break; case 2: vmxctx->guest_rdx = regval; break; case 3: vmxctx->guest_rbx = regval; break; case 4: vmcs_write(VMCS_GUEST_RSP, regval); break; case 5: vmxctx->guest_rbp = regval; break; case 6: vmxctx->guest_rsi = regval; break; case 7: vmxctx->guest_rdi = regval; break; case 8: vmxctx->guest_r8 = regval; break; case 9: vmxctx->guest_r9 = regval; break; case 10: vmxctx->guest_r10 = regval; break; case 11: vmxctx->guest_r11 = regval; break; case 12: vmxctx->guest_r12 = regval; break; case 13: vmxctx->guest_r13 = regval; break; case 14: vmxctx->guest_r14 = regval; break; case 15: vmxctx->guest_r15 = regval; break; default: panic("invalid vmx register %d", ident); } } static int vmx_emulate_cr0_access(struct vmx *vmx, int vcpu, uint64_t exitqual) { uint64_t crval, regval; /* We only handle mov to %cr0 at this time */ if ((exitqual & 0xf0) != 0x00) return (UNHANDLED); regval = vmx_get_guest_reg(vmx, vcpu, (exitqual >> 8) & 0xf); vmcs_write(VMCS_CR0_SHADOW, regval); crval = regval | cr0_ones_mask; crval &= ~cr0_zeros_mask; vmcs_write(VMCS_GUEST_CR0, crval); if (regval & CR0_PG) { uint64_t efer, entry_ctls; /* * If CR0.PG is 1 and EFER.LME is 1 then EFER.LMA and * the "IA-32e mode guest" bit in VM-entry control must be * equal. */ efer = vmcs_read(VMCS_GUEST_IA32_EFER); if (efer & EFER_LME) { efer |= EFER_LMA; vmcs_write(VMCS_GUEST_IA32_EFER, efer); entry_ctls = vmcs_read(VMCS_ENTRY_CTLS); entry_ctls |= VM_ENTRY_GUEST_LMA; vmcs_write(VMCS_ENTRY_CTLS, entry_ctls); } } return (HANDLED); } static int vmx_emulate_cr4_access(struct vmx *vmx, int vcpu, uint64_t exitqual) { uint64_t crval, regval; /* We only handle mov to %cr4 at this time */ if ((exitqual & 0xf0) != 0x00) return (UNHANDLED); regval = vmx_get_guest_reg(vmx, vcpu, (exitqual >> 8) & 0xf); vmcs_write(VMCS_CR4_SHADOW, regval); crval = regval | cr4_ones_mask; crval &= ~cr4_zeros_mask; vmcs_write(VMCS_GUEST_CR4, crval); return (HANDLED); } static int vmx_emulate_cr8_access(struct vmx *vmx, int vcpu, uint64_t exitqual) { struct vlapic *vlapic; uint64_t cr8; int regnum; /* We only handle mov %cr8 to/from a register at this time. */ if ((exitqual & 0xe0) != 0x00) { return (UNHANDLED); } vlapic = vm_lapic(vmx->vm, vcpu); regnum = (exitqual >> 8) & 0xf; if (exitqual & 0x10) { cr8 = vlapic_get_cr8(vlapic); vmx_set_guest_reg(vmx, vcpu, regnum, cr8); } else { cr8 = vmx_get_guest_reg(vmx, vcpu, regnum); vlapic_set_cr8(vlapic, cr8); } return (HANDLED); } /* * From section "Guest Register State" in the Intel SDM: CPL = SS.DPL */ static int vmx_cpl(void) { uint32_t ssar; ssar = vmcs_read(VMCS_GUEST_SS_ACCESS_RIGHTS); return ((ssar >> 5) & 0x3); } static enum vm_cpu_mode vmx_cpu_mode(void) { if (vmcs_read(VMCS_GUEST_IA32_EFER) & EFER_LMA) return (CPU_MODE_64BIT); else return (CPU_MODE_COMPATIBILITY); } static enum vm_paging_mode vmx_paging_mode(void) { if (!(vmcs_read(VMCS_GUEST_CR0) & CR0_PG)) return (PAGING_MODE_FLAT); if (!(vmcs_read(VMCS_GUEST_CR4) & CR4_PAE)) return (PAGING_MODE_32); if (vmcs_read(VMCS_GUEST_IA32_EFER) & EFER_LME) return (PAGING_MODE_64); else return (PAGING_MODE_PAE); } static uint64_t inout_str_index(struct vmx *vmx, int vcpuid, int in) { uint64_t val; int error; enum vm_reg_name reg; reg = in ? VM_REG_GUEST_RDI : VM_REG_GUEST_RSI; error = vmx_getreg(vmx, vcpuid, reg, &val); KASSERT(error == 0, ("%s: vmx_getreg error %d", __func__, error)); return (val); } static uint64_t inout_str_count(struct vmx *vmx, int vcpuid, int rep) { uint64_t val; int error; if (rep) { error = vmx_getreg(vmx, vcpuid, VM_REG_GUEST_RCX, &val); KASSERT(!error, ("%s: vmx_getreg error %d", __func__, error)); } else { val = 1; } return (val); } static int inout_str_addrsize(uint32_t inst_info) { uint32_t size; size = (inst_info >> 7) & 0x7; switch (size) { case 0: return (2); /* 16 bit */ case 1: return (4); /* 32 bit */ case 2: return (8); /* 64 bit */ default: panic("%s: invalid size encoding %d", __func__, size); } } static void inout_str_seginfo(struct vmx *vmx, int vcpuid, uint32_t inst_info, int in, struct vm_inout_str *vis) { int error, s; if (in) { vis->seg_name = VM_REG_GUEST_ES; } else { s = (inst_info >> 15) & 0x7; vis->seg_name = vm_segment_name(s); } error = vmx_getdesc(vmx, vcpuid, vis->seg_name, &vis->seg_desc); KASSERT(error == 0, ("%s: vmx_getdesc error %d", __func__, error)); /* XXX modify svm.c to update bit 16 of seg_desc.access (unusable) */ } static void vmx_paging_info(struct vm_guest_paging *paging) { paging->cr3 = vmcs_guest_cr3(); paging->cpl = vmx_cpl(); paging->cpu_mode = vmx_cpu_mode(); paging->paging_mode = vmx_paging_mode(); } static void vmexit_inst_emul(struct vm_exit *vmexit, uint64_t gpa, uint64_t gla) { vmexit->exitcode = VM_EXITCODE_INST_EMUL; vmexit->u.inst_emul.gpa = gpa; vmexit->u.inst_emul.gla = gla; vmx_paging_info(&vmexit->u.inst_emul.paging); } static int ept_fault_type(uint64_t ept_qual) { int fault_type; if (ept_qual & EPT_VIOLATION_DATA_WRITE) fault_type = VM_PROT_WRITE; else if (ept_qual & EPT_VIOLATION_INST_FETCH) fault_type = VM_PROT_EXECUTE; else fault_type= VM_PROT_READ; return (fault_type); } static boolean_t ept_emulation_fault(uint64_t ept_qual) { int read, write; /* EPT fault on an instruction fetch doesn't make sense here */ if (ept_qual & EPT_VIOLATION_INST_FETCH) return (FALSE); /* EPT fault must be a read fault or a write fault */ read = ept_qual & EPT_VIOLATION_DATA_READ ? 1 : 0; write = ept_qual & EPT_VIOLATION_DATA_WRITE ? 1 : 0; if ((read | write) == 0) return (FALSE); /* * The EPT violation must have been caused by accessing a * guest-physical address that is a translation of a guest-linear * address. */ if ((ept_qual & EPT_VIOLATION_GLA_VALID) == 0 || (ept_qual & EPT_VIOLATION_XLAT_VALID) == 0) { return (FALSE); } return (TRUE); } static __inline int apic_access_virtualization(struct vmx *vmx, int vcpuid) { uint32_t proc_ctls2; proc_ctls2 = vmx->cap[vcpuid].proc_ctls2; return ((proc_ctls2 & PROCBASED2_VIRTUALIZE_APIC_ACCESSES) ? 1 : 0); } static __inline int x2apic_virtualization(struct vmx *vmx, int vcpuid) { uint32_t proc_ctls2; proc_ctls2 = vmx->cap[vcpuid].proc_ctls2; return ((proc_ctls2 & PROCBASED2_VIRTUALIZE_X2APIC_MODE) ? 1 : 0); } static int vmx_handle_apic_write(struct vmx *vmx, int vcpuid, struct vlapic *vlapic, uint64_t qual) { int error, handled, offset; uint32_t *apic_regs, vector; bool retu; handled = HANDLED; offset = APIC_WRITE_OFFSET(qual); if (!apic_access_virtualization(vmx, vcpuid)) { /* * In general there should not be any APIC write VM-exits * unless APIC-access virtualization is enabled. * * However self-IPI virtualization can legitimately trigger * an APIC-write VM-exit so treat it specially. */ if (x2apic_virtualization(vmx, vcpuid) && offset == APIC_OFFSET_SELF_IPI) { apic_regs = (uint32_t *)(vlapic->apic_page); vector = apic_regs[APIC_OFFSET_SELF_IPI / 4]; vlapic_self_ipi_handler(vlapic, vector); return (HANDLED); } else return (UNHANDLED); } switch (offset) { case APIC_OFFSET_ID: vlapic_id_write_handler(vlapic); break; case APIC_OFFSET_LDR: vlapic_ldr_write_handler(vlapic); break; case APIC_OFFSET_DFR: vlapic_dfr_write_handler(vlapic); break; case APIC_OFFSET_SVR: vlapic_svr_write_handler(vlapic); break; case APIC_OFFSET_ESR: vlapic_esr_write_handler(vlapic); break; case APIC_OFFSET_ICR_LOW: retu = false; error = vlapic_icrlo_write_handler(vlapic, &retu); if (error != 0 || retu) handled = UNHANDLED; break; case APIC_OFFSET_CMCI_LVT: case APIC_OFFSET_TIMER_LVT ... APIC_OFFSET_ERROR_LVT: vlapic_lvt_write_handler(vlapic, offset); break; case APIC_OFFSET_TIMER_ICR: vlapic_icrtmr_write_handler(vlapic); break; case APIC_OFFSET_TIMER_DCR: vlapic_dcr_write_handler(vlapic); break; default: handled = UNHANDLED; break; } return (handled); } static bool apic_access_fault(struct vmx *vmx, int vcpuid, uint64_t gpa) { if (apic_access_virtualization(vmx, vcpuid) && (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE)) return (true); else return (false); } static int vmx_handle_apic_access(struct vmx *vmx, int vcpuid, struct vm_exit *vmexit) { uint64_t qual; int access_type, offset, allowed; if (!apic_access_virtualization(vmx, vcpuid)) return (UNHANDLED); qual = vmexit->u.vmx.exit_qualification; access_type = APIC_ACCESS_TYPE(qual); offset = APIC_ACCESS_OFFSET(qual); allowed = 0; if (access_type == 0) { /* * Read data access to the following registers is expected. */ switch (offset) { case APIC_OFFSET_APR: case APIC_OFFSET_PPR: case APIC_OFFSET_RRR: case APIC_OFFSET_CMCI_LVT: case APIC_OFFSET_TIMER_CCR: allowed = 1; break; default: break; } } else if (access_type == 1) { /* * Write data access to the following registers is expected. */ switch (offset) { case APIC_OFFSET_VER: case APIC_OFFSET_APR: case APIC_OFFSET_PPR: case APIC_OFFSET_RRR: case APIC_OFFSET_ISR0 ... APIC_OFFSET_ISR7: case APIC_OFFSET_TMR0 ... APIC_OFFSET_TMR7: case APIC_OFFSET_IRR0 ... APIC_OFFSET_IRR7: case APIC_OFFSET_CMCI_LVT: case APIC_OFFSET_TIMER_CCR: allowed = 1; break; default: break; } } if (allowed) { vmexit_inst_emul(vmexit, DEFAULT_APIC_BASE + offset, VIE_INVALID_GLA); } /* * Regardless of whether the APIC-access is allowed this handler * always returns UNHANDLED: * - if the access is allowed then it is handled by emulating the * instruction that caused the VM-exit (outside the critical section) * - if the access is not allowed then it will be converted to an * exitcode of VM_EXITCODE_VMX and will be dealt with in userland. */ return (UNHANDLED); } static int vmx_exit_process(struct vmx *vmx, int vcpu, struct vm_exit *vmexit) { int error, handled, in; struct vmxctx *vmxctx; struct vlapic *vlapic; struct vm_inout_str *vis; uint32_t eax, ecx, edx, idtvec_info, idtvec_err, intr_info, inst_info; uint32_t reason; uint64_t qual, gpa; bool retu; CTASSERT((PINBASED_CTLS_ONE_SETTING & PINBASED_VIRTUAL_NMI) != 0); CTASSERT((PINBASED_CTLS_ONE_SETTING & PINBASED_NMI_EXITING) != 0); handled = UNHANDLED; vmxctx = &vmx->ctx[vcpu]; qual = vmexit->u.vmx.exit_qualification; reason = vmexit->u.vmx.exit_reason; vmexit->exitcode = VM_EXITCODE_BOGUS; vmm_stat_incr(vmx->vm, vcpu, VMEXIT_COUNT, 1); /* * VM exits that could be triggered during event injection on the * previous VM entry need to be handled specially by re-injecting * the event. * * See "Information for VM Exits During Event Delivery" in Intel SDM * for details. */ switch (reason) { case EXIT_REASON_EPT_FAULT: case EXIT_REASON_EPT_MISCONFIG: case EXIT_REASON_APIC_ACCESS: case EXIT_REASON_TASK_SWITCH: case EXIT_REASON_EXCEPTION: idtvec_info = vmcs_idt_vectoring_info(); if (idtvec_info & VMCS_IDT_VEC_VALID) { idtvec_info &= ~(1 << 12); /* clear undefined bit */ vmcs_write(VMCS_ENTRY_INTR_INFO, idtvec_info); if (idtvec_info & VMCS_IDT_VEC_ERRCODE_VALID) { idtvec_err = vmcs_idt_vectoring_err(); vmcs_write(VMCS_ENTRY_EXCEPTION_ERROR, idtvec_err); } /* * If 'virtual NMIs' are being used and the VM-exit * happened while injecting an NMI during the previous * VM-entry, then clear "blocking by NMI" in the Guest * Interruptibility-state. */ if ((idtvec_info & VMCS_INTR_T_MASK) == VMCS_INTR_T_NMI) { vmx_clear_nmi_blocking(vmx, vcpu); } vmcs_write(VMCS_ENTRY_INST_LENGTH, vmexit->inst_length); } default: idtvec_info = 0; break; } switch (reason) { case EXIT_REASON_CR_ACCESS: vmm_stat_incr(vmx->vm, vcpu, VMEXIT_CR_ACCESS, 1); switch (qual & 0xf) { case 0: handled = vmx_emulate_cr0_access(vmx, vcpu, qual); break; case 4: handled = vmx_emulate_cr4_access(vmx, vcpu, qual); break; case 8: handled = vmx_emulate_cr8_access(vmx, vcpu, qual); break; } break; case EXIT_REASON_RDMSR: vmm_stat_incr(vmx->vm, vcpu, VMEXIT_RDMSR, 1); retu = false; ecx = vmxctx->guest_rcx; VCPU_CTR1(vmx->vm, vcpu, "rdmsr 0x%08x", ecx); error = emulate_rdmsr(vmx->vm, vcpu, ecx, &retu); if (error) { vmexit->exitcode = VM_EXITCODE_RDMSR; vmexit->u.msr.code = ecx; } else if (!retu) { handled = HANDLED; } else { /* Return to userspace with a valid exitcode */ KASSERT(vmexit->exitcode != VM_EXITCODE_BOGUS, ("emulate_wrmsr retu with bogus exitcode")); } break; case EXIT_REASON_WRMSR: vmm_stat_incr(vmx->vm, vcpu, VMEXIT_WRMSR, 1); retu = false; eax = vmxctx->guest_rax; ecx = vmxctx->guest_rcx; edx = vmxctx->guest_rdx; VCPU_CTR2(vmx->vm, vcpu, "wrmsr 0x%08x value 0x%016lx", ecx, (uint64_t)edx << 32 | eax); error = emulate_wrmsr(vmx->vm, vcpu, ecx, (uint64_t)edx << 32 | eax, &retu); if (error) { vmexit->exitcode = VM_EXITCODE_WRMSR; vmexit->u.msr.code = ecx; vmexit->u.msr.wval = (uint64_t)edx << 32 | eax; } else if (!retu) { handled = HANDLED; } else { /* Return to userspace with a valid exitcode */ KASSERT(vmexit->exitcode != VM_EXITCODE_BOGUS, ("emulate_wrmsr retu with bogus exitcode")); } break; case EXIT_REASON_HLT: vmm_stat_incr(vmx->vm, vcpu, VMEXIT_HLT, 1); vmexit->exitcode = VM_EXITCODE_HLT; vmexit->u.hlt.rflags = vmcs_read(VMCS_GUEST_RFLAGS); break; case EXIT_REASON_MTF: vmm_stat_incr(vmx->vm, vcpu, VMEXIT_MTRAP, 1); vmexit->exitcode = VM_EXITCODE_MTRAP; break; case EXIT_REASON_PAUSE: vmm_stat_incr(vmx->vm, vcpu, VMEXIT_PAUSE, 1); vmexit->exitcode = VM_EXITCODE_PAUSE; break; case EXIT_REASON_INTR_WINDOW: vmm_stat_incr(vmx->vm, vcpu, VMEXIT_INTR_WINDOW, 1); vmx_clear_int_window_exiting(vmx, vcpu); return (1); case EXIT_REASON_EXT_INTR: /* * External interrupts serve only to cause VM exits and allow * the host interrupt handler to run. * * If this external interrupt triggers a virtual interrupt * to a VM, then that state will be recorded by the * host interrupt handler in the VM's softc. We will inject * this virtual interrupt during the subsequent VM enter. */ intr_info = vmcs_read(VMCS_EXIT_INTR_INFO); /* * XXX: Ignore this exit if VMCS_INTR_VALID is not set. * This appears to be a bug in VMware Fusion? */ if (!(intr_info & VMCS_INTR_VALID)) return (1); KASSERT((intr_info & VMCS_INTR_VALID) != 0 && (intr_info & VMCS_INTR_T_MASK) == VMCS_INTR_T_HWINTR, ("VM exit interruption info invalid: %#x", intr_info)); vmx_trigger_hostintr(intr_info & 0xff); /* * This is special. We want to treat this as an 'handled' * VM-exit but not increment the instruction pointer. */ vmm_stat_incr(vmx->vm, vcpu, VMEXIT_EXTINT, 1); return (1); case EXIT_REASON_NMI_WINDOW: /* Exit to allow the pending virtual NMI to be injected */ if (vm_nmi_pending(vmx->vm, vcpu)) vmx_inject_nmi(vmx, vcpu); vmx_clear_nmi_window_exiting(vmx, vcpu); vmm_stat_incr(vmx->vm, vcpu, VMEXIT_NMI_WINDOW, 1); return (1); case EXIT_REASON_INOUT: vmm_stat_incr(vmx->vm, vcpu, VMEXIT_INOUT, 1); vmexit->exitcode = VM_EXITCODE_INOUT; vmexit->u.inout.bytes = (qual & 0x7) + 1; vmexit->u.inout.in = in = (qual & 0x8) ? 1 : 0; vmexit->u.inout.string = (qual & 0x10) ? 1 : 0; vmexit->u.inout.rep = (qual & 0x20) ? 1 : 0; vmexit->u.inout.port = (uint16_t)(qual >> 16); vmexit->u.inout.eax = (uint32_t)(vmxctx->guest_rax); if (vmexit->u.inout.string) { inst_info = vmcs_read(VMCS_EXIT_INSTRUCTION_INFO); vmexit->exitcode = VM_EXITCODE_INOUT_STR; vis = &vmexit->u.inout_str; vmx_paging_info(&vis->paging); vis->rflags = vmcs_read(VMCS_GUEST_RFLAGS); vis->cr0 = vmcs_read(VMCS_GUEST_CR0); vis->index = inout_str_index(vmx, vcpu, in); vis->count = inout_str_count(vmx, vcpu, vis->inout.rep); vis->addrsize = inout_str_addrsize(inst_info); inout_str_seginfo(vmx, vcpu, inst_info, in, vis); } break; case EXIT_REASON_CPUID: vmm_stat_incr(vmx->vm, vcpu, VMEXIT_CPUID, 1); handled = vmx_handle_cpuid(vmx->vm, vcpu, vmxctx); break; case EXIT_REASON_EXCEPTION: vmm_stat_incr(vmx->vm, vcpu, VMEXIT_EXCEPTION, 1); intr_info = vmcs_read(VMCS_EXIT_INTR_INFO); KASSERT((intr_info & VMCS_INTR_VALID) != 0, ("VM exit interruption info invalid: %#x", intr_info)); /* * If Virtual NMIs control is 1 and the VM-exit is due to a * fault encountered during the execution of IRET then we must * restore the state of "virtual-NMI blocking" before resuming * the guest. * * See "Resuming Guest Software after Handling an Exception". */ if ((idtvec_info & VMCS_IDT_VEC_VALID) == 0 && (intr_info & 0xff) != IDT_DF && (intr_info & EXIT_QUAL_NMIUDTI) != 0) vmx_restore_nmi_blocking(vmx, vcpu); /* * The NMI has already been handled in vmx_exit_handle_nmi(). */ if ((intr_info & VMCS_INTR_T_MASK) == VMCS_INTR_T_NMI) return (1); break; case EXIT_REASON_EPT_FAULT: /* * If 'gpa' lies within the address space allocated to * memory then this must be a nested page fault otherwise * this must be an instruction that accesses MMIO space. */ gpa = vmcs_gpa(); if (vm_mem_allocated(vmx->vm, gpa) || apic_access_fault(vmx, vcpu, gpa)) { vmexit->exitcode = VM_EXITCODE_PAGING; vmexit->u.paging.gpa = gpa; vmexit->u.paging.fault_type = ept_fault_type(qual); vmm_stat_incr(vmx->vm, vcpu, VMEXIT_NESTED_FAULT, 1); } else if (ept_emulation_fault(qual)) { vmexit_inst_emul(vmexit, gpa, vmcs_gla()); vmm_stat_incr(vmx->vm, vcpu, VMEXIT_INST_EMUL, 1); } /* * If Virtual NMIs control is 1 and the VM-exit is due to an * EPT fault during the execution of IRET then we must restore * the state of "virtual-NMI blocking" before resuming. * * See description of "NMI unblocking due to IRET" in * "Exit Qualification for EPT Violations". */ if ((idtvec_info & VMCS_IDT_VEC_VALID) == 0 && (qual & EXIT_QUAL_NMIUDTI) != 0) vmx_restore_nmi_blocking(vmx, vcpu); break; case EXIT_REASON_VIRTUALIZED_EOI: vmexit->exitcode = VM_EXITCODE_IOAPIC_EOI; vmexit->u.ioapic_eoi.vector = qual & 0xFF; vmexit->inst_length = 0; /* trap-like */ break; case EXIT_REASON_APIC_ACCESS: handled = vmx_handle_apic_access(vmx, vcpu, vmexit); break; case EXIT_REASON_APIC_WRITE: /* * APIC-write VM exit is trap-like so the %rip is already * pointing to the next instruction. */ vmexit->inst_length = 0; vlapic = vm_lapic(vmx->vm, vcpu); handled = vmx_handle_apic_write(vmx, vcpu, vlapic, qual); break; case EXIT_REASON_XSETBV: handled = vmx_emulate_xsetbv(vmx, vcpu, vmexit); break; default: vmm_stat_incr(vmx->vm, vcpu, VMEXIT_UNKNOWN, 1); break; } if (handled) { /* * It is possible that control is returned to userland * even though we were able to handle the VM exit in the * kernel. * * In such a case we want to make sure that the userland * restarts guest execution at the instruction *after* * the one we just processed. Therefore we update the * guest rip in the VMCS and in 'vmexit'. */ vmexit->rip += vmexit->inst_length; vmexit->inst_length = 0; vmcs_write(VMCS_GUEST_RIP, vmexit->rip); } else { if (vmexit->exitcode == VM_EXITCODE_BOGUS) { /* * If this VM exit was not claimed by anybody then * treat it as a generic VMX exit. */ vmexit->exitcode = VM_EXITCODE_VMX; vmexit->u.vmx.status = VM_SUCCESS; vmexit->u.vmx.inst_type = 0; vmexit->u.vmx.inst_error = 0; } else { /* * The exitcode and collateral have been populated. * The VM exit will be processed further in userland. */ } } return (handled); } -static __inline int -vmx_exit_astpending(struct vmx *vmx, int vcpu, struct vm_exit *vmexit) -{ - - vmexit->rip = vmcs_guest_rip(); - vmexit->inst_length = 0; - vmexit->exitcode = VM_EXITCODE_BOGUS; - vmx_astpending_trace(vmx, vcpu, vmexit->rip); - vmm_stat_incr(vmx->vm, vcpu, VMEXIT_ASTPENDING, 1); - - return (HANDLED); -} - -static __inline int -vmx_exit_rendezvous(struct vmx *vmx, int vcpu, struct vm_exit *vmexit) -{ - - vmexit->rip = vmcs_guest_rip(); - vmexit->inst_length = 0; - vmexit->exitcode = VM_EXITCODE_RENDEZVOUS; - vmm_stat_incr(vmx->vm, vcpu, VMEXIT_RENDEZVOUS, 1); - - return (UNHANDLED); -} - -static __inline int +static __inline void vmx_exit_inst_error(struct vmxctx *vmxctx, int rc, struct vm_exit *vmexit) { KASSERT(vmxctx->inst_fail_status != VM_SUCCESS, ("vmx_exit_inst_error: invalid inst_fail_status %d", vmxctx->inst_fail_status)); vmexit->inst_length = 0; vmexit->exitcode = VM_EXITCODE_VMX; vmexit->u.vmx.status = vmxctx->inst_fail_status; vmexit->u.vmx.inst_error = vmcs_instruction_error(); vmexit->u.vmx.exit_reason = ~0; vmexit->u.vmx.exit_qualification = ~0; switch (rc) { case VMX_VMRESUME_ERROR: case VMX_VMLAUNCH_ERROR: case VMX_INVEPT_ERROR: vmexit->u.vmx.inst_type = rc; break; default: panic("vm_exit_inst_error: vmx_enter_guest returned %d", rc); } - - return (UNHANDLED); } /* * If the NMI-exiting VM execution control is set to '1' then an NMI in * non-root operation causes a VM-exit. NMI blocking is in effect so it is * sufficient to simply vector to the NMI handler via a software interrupt. * However, this must be done before maskable interrupts are enabled * otherwise the "iret" issued by an interrupt handler will incorrectly * clear NMI blocking. */ static __inline void vmx_exit_handle_nmi(struct vmx *vmx, int vcpuid, struct vm_exit *vmexit) { uint32_t intr_info; KASSERT((read_rflags() & PSL_I) == 0, ("interrupts enabled")); if (vmexit->u.vmx.exit_reason != EXIT_REASON_EXCEPTION) return; intr_info = vmcs_read(VMCS_EXIT_INTR_INFO); KASSERT((intr_info & VMCS_INTR_VALID) != 0, ("VM exit interruption info invalid: %#x", intr_info)); if ((intr_info & VMCS_INTR_T_MASK) == VMCS_INTR_T_NMI) { KASSERT((intr_info & 0xff) == IDT_NMI, ("VM exit due " "to NMI has invalid vector: %#x", intr_info)); VCPU_CTR0(vmx->vm, vcpuid, "Vectoring to NMI handler"); __asm __volatile("int $2"); } } static int vmx_run(void *arg, int vcpu, register_t startrip, pmap_t pmap, void *rendezvous_cookie, void *suspend_cookie) { int rc, handled, launched; struct vmx *vmx; struct vm *vm; struct vmxctx *vmxctx; struct vmcs *vmcs; struct vm_exit *vmexit; struct vlapic *vlapic; uint64_t rip; uint32_t exit_reason; vmx = arg; vm = vmx->vm; vmcs = &vmx->vmcs[vcpu]; vmxctx = &vmx->ctx[vcpu]; vlapic = vm_lapic(vm, vcpu); vmexit = vm_exitinfo(vm, vcpu); launched = 0; KASSERT(vmxctx->pmap == pmap, ("pmap %p different than ctx pmap %p", pmap, vmxctx->pmap)); VMPTRLD(vmcs); /* * XXX * We do this every time because we may setup the virtual machine * from a different process than the one that actually runs it. * * If the life of a virtual machine was spent entirely in the context * of a single process we could do this once in vmx_vminit(). */ vmcs_write(VMCS_HOST_CR3, rcr3()); vmcs_write(VMCS_GUEST_RIP, startrip); vmx_set_pcpu_defaults(vmx, vcpu, pmap); do { + handled = UNHANDLED; + /* * Interrupts are disabled from this point on until the * guest starts executing. This is done for the following * reasons: * * If an AST is asserted on this thread after the check below, * then the IPI_AST notification will not be lost, because it * will cause a VM exit due to external interrupt as soon as * the guest state is loaded. * * A posted interrupt after 'vmx_inject_interrupts()' will * not be "lost" because it will be held pending in the host * APIC because interrupts are disabled. The pending interrupt * will be recognized as soon as the guest state is loaded. * * The same reasoning applies to the IPI generated by * pmap_invalidate_ept(). */ disable_intr(); if (vcpu_suspended(suspend_cookie)) { enable_intr(); vm_exit_suspended(vmx->vm, vcpu, vmcs_guest_rip()); - handled = UNHANDLED; break; } if (vcpu_rendezvous_pending(rendezvous_cookie)) { enable_intr(); - handled = vmx_exit_rendezvous(vmx, vcpu, vmexit); + vm_exit_rendezvous(vmx->vm, vcpu, vmcs_guest_rip()); break; } if (curthread->td_flags & (TDF_ASTPENDING | TDF_NEEDRESCHED)) { enable_intr(); - handled = vmx_exit_astpending(vmx, vcpu, vmexit); + vm_exit_astpending(vmx->vm, vcpu, vmcs_guest_rip()); + vmx_astpending_trace(vmx, vcpu, vmexit->rip); + handled = HANDLED; break; } vmx_inject_interrupts(vmx, vcpu, vlapic); vmx_run_trace(vmx, vcpu); rc = vmx_enter_guest(vmxctx, vmx, launched); /* Collect some information for VM exit processing */ vmexit->rip = rip = vmcs_guest_rip(); vmexit->inst_length = vmexit_instruction_length(); vmexit->u.vmx.exit_reason = exit_reason = vmcs_exit_reason(); vmexit->u.vmx.exit_qualification = vmcs_exit_qualification(); if (rc == VMX_GUEST_VMEXIT) { vmx_exit_handle_nmi(vmx, vcpu, vmexit); enable_intr(); handled = vmx_exit_process(vmx, vcpu, vmexit); } else { enable_intr(); - handled = vmx_exit_inst_error(vmxctx, rc, vmexit); + vmx_exit_inst_error(vmxctx, rc, vmexit); } launched = 1; vmx_exit_trace(vmx, vcpu, rip, exit_reason, handled); } while (handled); /* * If a VM exit has been handled then the exitcode must be BOGUS * If a VM exit is not handled then the exitcode must not be BOGUS */ if ((handled && vmexit->exitcode != VM_EXITCODE_BOGUS) || (!handled && vmexit->exitcode == VM_EXITCODE_BOGUS)) { panic("Mismatch between handled (%d) and exitcode (%d)", handled, vmexit->exitcode); } if (!handled) vmm_stat_incr(vm, vcpu, VMEXIT_USERSPACE, 1); VCPU_CTR1(vm, vcpu, "returning from vmx_run: exitcode %d", vmexit->exitcode); VMCLEAR(vmcs); return (0); } static void vmx_vmcleanup(void *arg) { int i; struct vmx *vmx = arg; if (apic_access_virtualization(vmx, 0)) vm_unmap_mmio(vmx->vm, DEFAULT_APIC_BASE, PAGE_SIZE); for (i = 0; i < VM_MAXCPU; i++) vpid_free(vmx->state[i].vpid); free(vmx, M_VMX); return; } static register_t * vmxctx_regptr(struct vmxctx *vmxctx, int reg) { switch (reg) { case VM_REG_GUEST_RAX: return (&vmxctx->guest_rax); case VM_REG_GUEST_RBX: return (&vmxctx->guest_rbx); case VM_REG_GUEST_RCX: return (&vmxctx->guest_rcx); case VM_REG_GUEST_RDX: return (&vmxctx->guest_rdx); case VM_REG_GUEST_RSI: return (&vmxctx->guest_rsi); case VM_REG_GUEST_RDI: return (&vmxctx->guest_rdi); case VM_REG_GUEST_RBP: return (&vmxctx->guest_rbp); case VM_REG_GUEST_R8: return (&vmxctx->guest_r8); case VM_REG_GUEST_R9: return (&vmxctx->guest_r9); case VM_REG_GUEST_R10: return (&vmxctx->guest_r10); case VM_REG_GUEST_R11: return (&vmxctx->guest_r11); case VM_REG_GUEST_R12: return (&vmxctx->guest_r12); case VM_REG_GUEST_R13: return (&vmxctx->guest_r13); case VM_REG_GUEST_R14: return (&vmxctx->guest_r14); case VM_REG_GUEST_R15: return (&vmxctx->guest_r15); case VM_REG_GUEST_CR2: return (&vmxctx->guest_cr2); default: break; } return (NULL); } static int vmxctx_getreg(struct vmxctx *vmxctx, int reg, uint64_t *retval) { register_t *regp; if ((regp = vmxctx_regptr(vmxctx, reg)) != NULL) { *retval = *regp; return (0); } else return (EINVAL); } static int vmxctx_setreg(struct vmxctx *vmxctx, int reg, uint64_t val) { register_t *regp; if ((regp = vmxctx_regptr(vmxctx, reg)) != NULL) { *regp = val; return (0); } else return (EINVAL); } static int vmx_shadow_reg(int reg) { int shreg; shreg = -1; switch (reg) { case VM_REG_GUEST_CR0: shreg = VMCS_CR0_SHADOW; break; case VM_REG_GUEST_CR4: shreg = VMCS_CR4_SHADOW; break; default: break; } return (shreg); } static int vmx_getreg(void *arg, int vcpu, int reg, uint64_t *retval) { int running, hostcpu; struct vmx *vmx = arg; running = vcpu_is_running(vmx->vm, vcpu, &hostcpu); if (running && hostcpu != curcpu) panic("vmx_getreg: %s%d is running", vm_name(vmx->vm), vcpu); if (vmxctx_getreg(&vmx->ctx[vcpu], reg, retval) == 0) return (0); return (vmcs_getreg(&vmx->vmcs[vcpu], running, reg, retval)); } static int vmx_setreg(void *arg, int vcpu, int reg, uint64_t val) { int error, hostcpu, running, shadow; uint64_t ctls; struct vmx *vmx = arg; running = vcpu_is_running(vmx->vm, vcpu, &hostcpu); if (running && hostcpu != curcpu) panic("vmx_setreg: %s%d is running", vm_name(vmx->vm), vcpu); if (vmxctx_setreg(&vmx->ctx[vcpu], reg, val) == 0) return (0); error = vmcs_setreg(&vmx->vmcs[vcpu], running, reg, val); if (error == 0) { /* * If the "load EFER" VM-entry control is 1 then the * value of EFER.LMA must be identical to "IA-32e mode guest" * bit in the VM-entry control. */ if ((entry_ctls & VM_ENTRY_LOAD_EFER) != 0 && (reg == VM_REG_GUEST_EFER)) { vmcs_getreg(&vmx->vmcs[vcpu], running, VMCS_IDENT(VMCS_ENTRY_CTLS), &ctls); if (val & EFER_LMA) ctls |= VM_ENTRY_GUEST_LMA; else ctls &= ~VM_ENTRY_GUEST_LMA; vmcs_setreg(&vmx->vmcs[vcpu], running, VMCS_IDENT(VMCS_ENTRY_CTLS), ctls); } shadow = vmx_shadow_reg(reg); if (shadow > 0) { /* * Store the unmodified value in the shadow */ error = vmcs_setreg(&vmx->vmcs[vcpu], running, VMCS_IDENT(shadow), val); } } return (error); } static int vmx_getdesc(void *arg, int vcpu, int reg, struct seg_desc *desc) { int hostcpu, running; struct vmx *vmx = arg; running = vcpu_is_running(vmx->vm, vcpu, &hostcpu); if (running && hostcpu != curcpu) panic("vmx_getdesc: %s%d is running", vm_name(vmx->vm), vcpu); return (vmcs_getdesc(&vmx->vmcs[vcpu], running, reg, desc)); } static int vmx_setdesc(void *arg, int vcpu, int reg, struct seg_desc *desc) { int hostcpu, running; struct vmx *vmx = arg; running = vcpu_is_running(vmx->vm, vcpu, &hostcpu); if (running && hostcpu != curcpu) panic("vmx_setdesc: %s%d is running", vm_name(vmx->vm), vcpu); return (vmcs_setdesc(&vmx->vmcs[vcpu], running, reg, desc)); } static int vmx_getcap(void *arg, int vcpu, int type, int *retval) { struct vmx *vmx = arg; int vcap; int ret; ret = ENOENT; vcap = vmx->cap[vcpu].set; switch (type) { case VM_CAP_HALT_EXIT: if (cap_halt_exit) ret = 0; break; case VM_CAP_PAUSE_EXIT: if (cap_pause_exit) ret = 0; break; case VM_CAP_MTRAP_EXIT: if (cap_monitor_trap) ret = 0; break; case VM_CAP_UNRESTRICTED_GUEST: if (cap_unrestricted_guest) ret = 0; break; case VM_CAP_ENABLE_INVPCID: if (cap_invpcid) ret = 0; break; default: break; } if (ret == 0) *retval = (vcap & (1 << type)) ? 1 : 0; return (ret); } static int vmx_setcap(void *arg, int vcpu, int type, int val) { struct vmx *vmx = arg; struct vmcs *vmcs = &vmx->vmcs[vcpu]; uint32_t baseval; uint32_t *pptr; int error; int flag; int reg; int retval; retval = ENOENT; pptr = NULL; switch (type) { case VM_CAP_HALT_EXIT: if (cap_halt_exit) { retval = 0; pptr = &vmx->cap[vcpu].proc_ctls; baseval = *pptr; flag = PROCBASED_HLT_EXITING; reg = VMCS_PRI_PROC_BASED_CTLS; } break; case VM_CAP_MTRAP_EXIT: if (cap_monitor_trap) { retval = 0; pptr = &vmx->cap[vcpu].proc_ctls; baseval = *pptr; flag = PROCBASED_MTF; reg = VMCS_PRI_PROC_BASED_CTLS; } break; case VM_CAP_PAUSE_EXIT: if (cap_pause_exit) { retval = 0; pptr = &vmx->cap[vcpu].proc_ctls; baseval = *pptr; flag = PROCBASED_PAUSE_EXITING; reg = VMCS_PRI_PROC_BASED_CTLS; } break; case VM_CAP_UNRESTRICTED_GUEST: if (cap_unrestricted_guest) { retval = 0; pptr = &vmx->cap[vcpu].proc_ctls2; baseval = *pptr; flag = PROCBASED2_UNRESTRICTED_GUEST; reg = VMCS_SEC_PROC_BASED_CTLS; } break; case VM_CAP_ENABLE_INVPCID: if (cap_invpcid) { retval = 0; pptr = &vmx->cap[vcpu].proc_ctls2; baseval = *pptr; flag = PROCBASED2_ENABLE_INVPCID; reg = VMCS_SEC_PROC_BASED_CTLS; } break; default: break; } if (retval == 0) { if (val) { baseval |= flag; } else { baseval &= ~flag; } VMPTRLD(vmcs); error = vmwrite(reg, baseval); VMCLEAR(vmcs); if (error) { retval = error; } else { /* * Update optional stored flags, and record * setting */ if (pptr != NULL) { *pptr = baseval; } if (val) { vmx->cap[vcpu].set |= (1 << type); } else { vmx->cap[vcpu].set &= ~(1 << type); } } } return (retval); } struct vlapic_vtx { struct vlapic vlapic; struct pir_desc *pir_desc; struct vmx *vmx; }; #define VMX_CTR_PIR(vm, vcpuid, pir_desc, notify, vector, level, msg) \ do { \ VCPU_CTR2(vm, vcpuid, msg " assert %s-triggered vector %d", \ level ? "level" : "edge", vector); \ VCPU_CTR1(vm, vcpuid, msg " pir0 0x%016lx", pir_desc->pir[0]); \ VCPU_CTR1(vm, vcpuid, msg " pir1 0x%016lx", pir_desc->pir[1]); \ VCPU_CTR1(vm, vcpuid, msg " pir2 0x%016lx", pir_desc->pir[2]); \ VCPU_CTR1(vm, vcpuid, msg " pir3 0x%016lx", pir_desc->pir[3]); \ VCPU_CTR1(vm, vcpuid, msg " notify: %s", notify ? "yes" : "no");\ } while (0) /* * vlapic->ops handlers that utilize the APICv hardware assist described in * Chapter 29 of the Intel SDM. */ static int vmx_set_intr_ready(struct vlapic *vlapic, int vector, bool level) { struct vlapic_vtx *vlapic_vtx; struct pir_desc *pir_desc; uint64_t mask; int idx, notify; vlapic_vtx = (struct vlapic_vtx *)vlapic; pir_desc = vlapic_vtx->pir_desc; /* * Keep track of interrupt requests in the PIR descriptor. This is * because the virtual APIC page pointed to by the VMCS cannot be * modified if the vcpu is running. */ idx = vector / 64; mask = 1UL << (vector % 64); atomic_set_long(&pir_desc->pir[idx], mask); notify = atomic_cmpset_long(&pir_desc->pending, 0, 1); VMX_CTR_PIR(vlapic->vm, vlapic->vcpuid, pir_desc, notify, vector, level, "vmx_set_intr_ready"); return (notify); } static int vmx_pending_intr(struct vlapic *vlapic, int *vecptr) { struct vlapic_vtx *vlapic_vtx; struct pir_desc *pir_desc; struct LAPIC *lapic; uint64_t pending, pirval; uint32_t ppr, vpr; int i; /* * This function is only expected to be called from the 'HLT' exit * handler which does not care about the vector that is pending. */ KASSERT(vecptr == NULL, ("vmx_pending_intr: vecptr must be NULL")); vlapic_vtx = (struct vlapic_vtx *)vlapic; pir_desc = vlapic_vtx->pir_desc; pending = atomic_load_acq_long(&pir_desc->pending); if (!pending) return (0); /* common case */ /* * If there is an interrupt pending then it will be recognized only * if its priority is greater than the processor priority. * * Special case: if the processor priority is zero then any pending * interrupt will be recognized. */ lapic = vlapic->apic_page; ppr = lapic->ppr & 0xf0; if (ppr == 0) return (1); VCPU_CTR1(vlapic->vm, vlapic->vcpuid, "HLT with non-zero PPR %d", lapic->ppr); for (i = 3; i >= 0; i--) { pirval = pir_desc->pir[i]; if (pirval != 0) { vpr = (i * 64 + flsl(pirval) - 1) & 0xf0; return (vpr > ppr); } } return (0); } static void vmx_intr_accepted(struct vlapic *vlapic, int vector) { panic("vmx_intr_accepted: not expected to be called"); } static void vmx_set_tmr(struct vlapic *vlapic, int vector, bool level) { struct vlapic_vtx *vlapic_vtx; struct vmx *vmx; struct vmcs *vmcs; uint64_t mask, val; KASSERT(vector >= 0 && vector <= 255, ("invalid vector %d", vector)); KASSERT(!vcpu_is_running(vlapic->vm, vlapic->vcpuid, NULL), ("vmx_set_tmr: vcpu cannot be running")); vlapic_vtx = (struct vlapic_vtx *)vlapic; vmx = vlapic_vtx->vmx; vmcs = &vmx->vmcs[vlapic->vcpuid]; mask = 1UL << (vector % 64); VMPTRLD(vmcs); val = vmcs_read(VMCS_EOI_EXIT(vector)); if (level) val |= mask; else val &= ~mask; vmcs_write(VMCS_EOI_EXIT(vector), val); VMCLEAR(vmcs); } static void vmx_enable_x2apic_mode(struct vlapic *vlapic) { struct vmx *vmx; struct vmcs *vmcs; uint32_t proc_ctls2; int vcpuid, error; vcpuid = vlapic->vcpuid; vmx = ((struct vlapic_vtx *)vlapic)->vmx; vmcs = &vmx->vmcs[vcpuid]; proc_ctls2 = vmx->cap[vcpuid].proc_ctls2; KASSERT((proc_ctls2 & PROCBASED2_VIRTUALIZE_APIC_ACCESSES) != 0, ("%s: invalid proc_ctls2 %#x", __func__, proc_ctls2)); proc_ctls2 &= ~PROCBASED2_VIRTUALIZE_APIC_ACCESSES; proc_ctls2 |= PROCBASED2_VIRTUALIZE_X2APIC_MODE; vmx->cap[vcpuid].proc_ctls2 = proc_ctls2; VMPTRLD(vmcs); vmcs_write(VMCS_SEC_PROC_BASED_CTLS, proc_ctls2); VMCLEAR(vmcs); if (vlapic->vcpuid == 0) { /* * The nested page table mappings are shared by all vcpus * so unmap the APIC access page just once. */ error = vm_unmap_mmio(vmx->vm, DEFAULT_APIC_BASE, PAGE_SIZE); KASSERT(error == 0, ("%s: vm_unmap_mmio error %d", __func__, error)); /* * The MSR bitmap is shared by all vcpus so modify it only * once in the context of vcpu 0. */ error = vmx_allow_x2apic_msrs(vmx); KASSERT(error == 0, ("%s: vmx_allow_x2apic_msrs error %d", __func__, error)); } } static void vmx_post_intr(struct vlapic *vlapic, int hostcpu) { ipi_cpu(hostcpu, pirvec); } /* * Transfer the pending interrupts in the PIR descriptor to the IRR * in the virtual APIC page. */ static void vmx_inject_pir(struct vlapic *vlapic) { struct vlapic_vtx *vlapic_vtx; struct pir_desc *pir_desc; struct LAPIC *lapic; uint64_t val, pirval; int rvi, pirbase = -1; uint16_t intr_status_old, intr_status_new; vlapic_vtx = (struct vlapic_vtx *)vlapic; pir_desc = vlapic_vtx->pir_desc; if (atomic_cmpset_long(&pir_desc->pending, 1, 0) == 0) { VCPU_CTR0(vlapic->vm, vlapic->vcpuid, "vmx_inject_pir: " "no posted interrupt pending"); return; } pirval = 0; pirbase = -1; lapic = vlapic->apic_page; val = atomic_readandclear_long(&pir_desc->pir[0]); if (val != 0) { lapic->irr0 |= val; lapic->irr1 |= val >> 32; pirbase = 0; pirval = val; } val = atomic_readandclear_long(&pir_desc->pir[1]); if (val != 0) { lapic->irr2 |= val; lapic->irr3 |= val >> 32; pirbase = 64; pirval = val; } val = atomic_readandclear_long(&pir_desc->pir[2]); if (val != 0) { lapic->irr4 |= val; lapic->irr5 |= val >> 32; pirbase = 128; pirval = val; } val = atomic_readandclear_long(&pir_desc->pir[3]); if (val != 0) { lapic->irr6 |= val; lapic->irr7 |= val >> 32; pirbase = 192; pirval = val; } VLAPIC_CTR_IRR(vlapic, "vmx_inject_pir"); /* * Update RVI so the processor can evaluate pending virtual * interrupts on VM-entry. * * It is possible for pirval to be 0 here, even though the * pending bit has been set. The scenario is: * CPU-Y is sending a posted interrupt to CPU-X, which * is running a guest and processing posted interrupts in h/w. * CPU-X will eventually exit and the state seen in s/w is * the pending bit set, but no PIR bits set. * * CPU-X CPU-Y * (vm running) (host running) * rx posted interrupt * CLEAR pending bit * SET PIR bit * READ/CLEAR PIR bits * SET pending bit * (vm exit) * pending bit set, PIR 0 */ if (pirval != 0) { rvi = pirbase + flsl(pirval) - 1; intr_status_old = vmcs_read(VMCS_GUEST_INTR_STATUS); intr_status_new = (intr_status_old & 0xFF00) | rvi; if (intr_status_new > intr_status_old) { vmcs_write(VMCS_GUEST_INTR_STATUS, intr_status_new); VCPU_CTR2(vlapic->vm, vlapic->vcpuid, "vmx_inject_pir: " "guest_intr_status changed from 0x%04x to 0x%04x", intr_status_old, intr_status_new); } } } static struct vlapic * vmx_vlapic_init(void *arg, int vcpuid) { struct vmx *vmx; struct vlapic *vlapic; struct vlapic_vtx *vlapic_vtx; vmx = arg; vlapic = malloc(sizeof(struct vlapic_vtx), M_VLAPIC, M_WAITOK | M_ZERO); vlapic->vm = vmx->vm; vlapic->vcpuid = vcpuid; vlapic->apic_page = (struct LAPIC *)&vmx->apic_page[vcpuid]; vlapic_vtx = (struct vlapic_vtx *)vlapic; vlapic_vtx->pir_desc = &vmx->pir_desc[vcpuid]; vlapic_vtx->vmx = vmx; if (virtual_interrupt_delivery) { vlapic->ops.set_intr_ready = vmx_set_intr_ready; vlapic->ops.pending_intr = vmx_pending_intr; vlapic->ops.intr_accepted = vmx_intr_accepted; vlapic->ops.set_tmr = vmx_set_tmr; vlapic->ops.enable_x2apic_mode = vmx_enable_x2apic_mode; } if (posted_interrupts) vlapic->ops.post_intr = vmx_post_intr; vlapic_init(vlapic); return (vlapic); } static void vmx_vlapic_cleanup(void *arg, struct vlapic *vlapic) { vlapic_cleanup(vlapic); free(vlapic, M_VLAPIC); } struct vmm_ops vmm_ops_intel = { vmx_init, vmx_cleanup, vmx_restore, vmx_vminit, vmx_run, vmx_vmcleanup, vmx_getreg, vmx_setreg, vmx_getdesc, vmx_setdesc, vmx_getcap, vmx_setcap, ept_vmspace_alloc, ept_vmspace_free, vmx_vlapic_init, vmx_vlapic_cleanup, }; Index: stable/10/sys/amd64/vmm/vmm.c =================================================================== --- stable/10/sys/amd64/vmm/vmm.c (revision 270073) +++ stable/10/sys/amd64/vmm/vmm.c (revision 270074) @@ -1,1968 +1,2025 @@ /*- * Copyright (c) 2011 NetApp, Inc. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "vmm_ioport.h" #include "vmm_ktr.h" #include "vmm_host.h" #include "vmm_mem.h" #include "vmm_util.h" #include "vatpic.h" #include "vatpit.h" #include "vhpet.h" #include "vioapic.h" #include "vlapic.h" #include "vmm_msr.h" #include "vmm_ipi.h" #include "vmm_stat.h" #include "vmm_lapic.h" #include "io/ppt.h" #include "io/iommu.h" struct vlapic; /* * Initialization: * (a) allocated when vcpu is created * (i) initialized when vcpu is created and when it is reinitialized * (o) initialized the first time the vcpu is created * (x) initialized before use */ struct vcpu { struct mtx mtx; /* (o) protects 'state' and 'hostcpu' */ enum vcpu_state state; /* (o) vcpu state */ int hostcpu; /* (o) vcpu's host cpu */ struct vlapic *vlapic; /* (i) APIC device model */ enum x2apic_state x2apic_state; /* (i) APIC mode */ int nmi_pending; /* (i) NMI pending */ int extint_pending; /* (i) INTR pending */ struct vm_exception exception; /* (x) exception collateral */ int exception_pending; /* (i) exception pending */ struct savefpu *guestfpu; /* (a,i) guest fpu state */ uint64_t guest_xcr0; /* (i) guest %xcr0 register */ void *stats; /* (a,i) statistics */ uint64_t guest_msrs[VMM_MSR_NUM]; /* (i) emulated MSRs */ struct vm_exit exitinfo; /* (x) exit reason and collateral */ }; #define vcpu_lock_initialized(v) mtx_initialized(&((v)->mtx)) #define vcpu_lock_init(v) mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN) #define vcpu_lock(v) mtx_lock_spin(&((v)->mtx)) #define vcpu_unlock(v) mtx_unlock_spin(&((v)->mtx)) #define vcpu_assert_locked(v) mtx_assert(&((v)->mtx), MA_OWNED) struct mem_seg { vm_paddr_t gpa; size_t len; boolean_t wired; vm_object_t object; }; #define VM_MAX_MEMORY_SEGMENTS 2 /* * Initialization: * (o) initialized the first time the VM is created * (i) initialized when VM is created and when it is reinitialized * (x) initialized before use */ struct vm { void *cookie; /* (i) cpu-specific data */ void *iommu; /* (x) iommu-specific data */ struct vhpet *vhpet; /* (i) virtual HPET */ struct vioapic *vioapic; /* (i) virtual ioapic */ struct vatpic *vatpic; /* (i) virtual atpic */ struct vatpit *vatpit; /* (i) virtual atpit */ volatile cpuset_t active_cpus; /* (i) active vcpus */ int suspend; /* (i) stop VM execution */ volatile cpuset_t suspended_cpus; /* (i) suspended vcpus */ volatile cpuset_t halted_cpus; /* (x) cpus in a hard halt */ cpuset_t rendezvous_req_cpus; /* (x) rendezvous requested */ cpuset_t rendezvous_done_cpus; /* (x) rendezvous finished */ void *rendezvous_arg; /* (x) rendezvous func/arg */ vm_rendezvous_func_t rendezvous_func; struct mtx rendezvous_mtx; /* (o) rendezvous lock */ int num_mem_segs; /* (o) guest memory segments */ struct mem_seg mem_segs[VM_MAX_MEMORY_SEGMENTS]; struct vmspace *vmspace; /* (o) guest's address space */ char name[VM_MAX_NAMELEN]; /* (o) virtual machine name */ struct vcpu vcpu[VM_MAXCPU]; /* (i) guest vcpus */ }; static int vmm_initialized; static struct vmm_ops *ops; #define VMM_INIT(num) (ops != NULL ? (*ops->init)(num) : 0) #define VMM_CLEANUP() (ops != NULL ? (*ops->cleanup)() : 0) #define VMM_RESUME() (ops != NULL ? (*ops->resume)() : 0) #define VMINIT(vm, pmap) (ops != NULL ? (*ops->vminit)(vm, pmap): NULL) #define VMRUN(vmi, vcpu, rip, pmap, rptr, sptr) \ (ops != NULL ? (*ops->vmrun)(vmi, vcpu, rip, pmap, rptr, sptr) : ENXIO) #define VMCLEANUP(vmi) (ops != NULL ? (*ops->vmcleanup)(vmi) : NULL) #define VMSPACE_ALLOC(min, max) \ (ops != NULL ? (*ops->vmspace_alloc)(min, max) : NULL) #define VMSPACE_FREE(vmspace) \ (ops != NULL ? (*ops->vmspace_free)(vmspace) : ENXIO) #define VMGETREG(vmi, vcpu, num, retval) \ (ops != NULL ? (*ops->vmgetreg)(vmi, vcpu, num, retval) : ENXIO) #define VMSETREG(vmi, vcpu, num, val) \ (ops != NULL ? (*ops->vmsetreg)(vmi, vcpu, num, val) : ENXIO) #define VMGETDESC(vmi, vcpu, num, desc) \ (ops != NULL ? (*ops->vmgetdesc)(vmi, vcpu, num, desc) : ENXIO) #define VMSETDESC(vmi, vcpu, num, desc) \ (ops != NULL ? (*ops->vmsetdesc)(vmi, vcpu, num, desc) : ENXIO) #define VMGETCAP(vmi, vcpu, num, retval) \ (ops != NULL ? (*ops->vmgetcap)(vmi, vcpu, num, retval) : ENXIO) #define VMSETCAP(vmi, vcpu, num, val) \ (ops != NULL ? (*ops->vmsetcap)(vmi, vcpu, num, val) : ENXIO) #define VLAPIC_INIT(vmi, vcpu) \ (ops != NULL ? (*ops->vlapic_init)(vmi, vcpu) : NULL) #define VLAPIC_CLEANUP(vmi, vlapic) \ (ops != NULL ? (*ops->vlapic_cleanup)(vmi, vlapic) : NULL) #define fpu_start_emulating() load_cr0(rcr0() | CR0_TS) #define fpu_stop_emulating() clts() static MALLOC_DEFINE(M_VM, "vm", "vm"); CTASSERT(VMM_MSR_NUM <= 64); /* msr_mask can keep track of up to 64 msrs */ /* statistics */ static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime"); SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL); /* * Halt the guest if all vcpus are executing a HLT instruction with * interrupts disabled. */ static int halt_detection_enabled = 1; TUNABLE_INT("hw.vmm.halt_detection", &halt_detection_enabled); SYSCTL_INT(_hw_vmm, OID_AUTO, halt_detection, CTLFLAG_RDTUN, &halt_detection_enabled, 0, "Halt VM if all vcpus execute HLT with interrupts disabled"); static int vmm_ipinum; SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0, "IPI vector used for vcpu notifications"); static void vcpu_cleanup(struct vm *vm, int i, bool destroy) { struct vcpu *vcpu = &vm->vcpu[i]; VLAPIC_CLEANUP(vm->cookie, vcpu->vlapic); if (destroy) { vmm_stat_free(vcpu->stats); fpu_save_area_free(vcpu->guestfpu); } } static void vcpu_init(struct vm *vm, int vcpu_id, bool create) { struct vcpu *vcpu; KASSERT(vcpu_id >= 0 && vcpu_id < VM_MAXCPU, ("vcpu_init: invalid vcpu %d", vcpu_id)); vcpu = &vm->vcpu[vcpu_id]; if (create) { KASSERT(!vcpu_lock_initialized(vcpu), ("vcpu %d already " "initialized", vcpu_id)); vcpu_lock_init(vcpu); vcpu->state = VCPU_IDLE; vcpu->hostcpu = NOCPU; vcpu->guestfpu = fpu_save_area_alloc(); vcpu->stats = vmm_stat_alloc(); } vcpu->vlapic = VLAPIC_INIT(vm->cookie, vcpu_id); vm_set_x2apic_state(vm, vcpu_id, X2APIC_DISABLED); vcpu->nmi_pending = 0; vcpu->extint_pending = 0; vcpu->exception_pending = 0; vcpu->guest_xcr0 = XFEATURE_ENABLED_X87; fpu_save_area_reset(vcpu->guestfpu); vmm_stat_init(vcpu->stats); guest_msrs_init(vm, vcpu_id); } struct vm_exit * vm_exitinfo(struct vm *vm, int cpuid) { struct vcpu *vcpu; if (cpuid < 0 || cpuid >= VM_MAXCPU) panic("vm_exitinfo: invalid cpuid %d", cpuid); vcpu = &vm->vcpu[cpuid]; return (&vcpu->exitinfo); } static void vmm_resume(void) { VMM_RESUME(); } static int vmm_init(void) { int error; vmm_host_state_init(); vmm_ipinum = vmm_ipi_alloc(); if (vmm_ipinum == 0) vmm_ipinum = IPI_AST; error = vmm_mem_init(); if (error) return (error); if (vmm_is_intel()) ops = &vmm_ops_intel; else if (vmm_is_amd()) ops = &vmm_ops_amd; else return (ENXIO); vmm_msr_init(); vmm_resume_p = vmm_resume; return (VMM_INIT(vmm_ipinum)); } static int vmm_handler(module_t mod, int what, void *arg) { int error; switch (what) { case MOD_LOAD: vmmdev_init(); if (ppt_avail_devices() > 0) iommu_init(); error = vmm_init(); if (error == 0) vmm_initialized = 1; break; case MOD_UNLOAD: error = vmmdev_cleanup(); if (error == 0) { vmm_resume_p = NULL; iommu_cleanup(); if (vmm_ipinum != IPI_AST) vmm_ipi_free(vmm_ipinum); error = VMM_CLEANUP(); /* * Something bad happened - prevent new * VMs from being created */ if (error) vmm_initialized = 0; } break; default: error = 0; break; } return (error); } static moduledata_t vmm_kmod = { "vmm", vmm_handler, NULL }; /* * vmm initialization has the following dependencies: * * - iommu initialization must happen after the pci passthru driver has had * a chance to attach to any passthru devices (after SI_SUB_CONFIGURE). * * - VT-x initialization requires smp_rendezvous() and therefore must happen * after SMP is fully functional (after SI_SUB_SMP). */ DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY); MODULE_VERSION(vmm, 1); static void vm_init(struct vm *vm, bool create) { int i; vm->cookie = VMINIT(vm, vmspace_pmap(vm->vmspace)); vm->iommu = NULL; vm->vioapic = vioapic_init(vm); vm->vhpet = vhpet_init(vm); vm->vatpic = vatpic_init(vm); vm->vatpit = vatpit_init(vm); CPU_ZERO(&vm->active_cpus); vm->suspend = 0; CPU_ZERO(&vm->suspended_cpus); for (i = 0; i < VM_MAXCPU; i++) vcpu_init(vm, i, create); } int vm_create(const char *name, struct vm **retvm) { struct vm *vm; struct vmspace *vmspace; /* * If vmm.ko could not be successfully initialized then don't attempt * to create the virtual machine. */ if (!vmm_initialized) return (ENXIO); if (name == NULL || strlen(name) >= VM_MAX_NAMELEN) return (EINVAL); vmspace = VMSPACE_ALLOC(VM_MIN_ADDRESS, VM_MAXUSER_ADDRESS); if (vmspace == NULL) return (ENOMEM); vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO); strcpy(vm->name, name); vm->num_mem_segs = 0; vm->vmspace = vmspace; mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF); vm_init(vm, true); *retvm = vm; return (0); } static void vm_free_mem_seg(struct vm *vm, struct mem_seg *seg) { if (seg->object != NULL) vmm_mem_free(vm->vmspace, seg->gpa, seg->len); bzero(seg, sizeof(*seg)); } static void vm_cleanup(struct vm *vm, bool destroy) { int i; ppt_unassign_all(vm); if (vm->iommu != NULL) iommu_destroy_domain(vm->iommu); vatpit_cleanup(vm->vatpit); vhpet_cleanup(vm->vhpet); vatpic_cleanup(vm->vatpic); vioapic_cleanup(vm->vioapic); for (i = 0; i < VM_MAXCPU; i++) vcpu_cleanup(vm, i, destroy); VMCLEANUP(vm->cookie); if (destroy) { for (i = 0; i < vm->num_mem_segs; i++) vm_free_mem_seg(vm, &vm->mem_segs[i]); vm->num_mem_segs = 0; VMSPACE_FREE(vm->vmspace); vm->vmspace = NULL; } } void vm_destroy(struct vm *vm) { vm_cleanup(vm, true); free(vm, M_VM); } int vm_reinit(struct vm *vm) { int error; /* * A virtual machine can be reset only if all vcpus are suspended. */ if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { vm_cleanup(vm, false); vm_init(vm, false); error = 0; } else { error = EBUSY; } return (error); } const char * vm_name(struct vm *vm) { return (vm->name); } int vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa) { vm_object_t obj; if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL) return (ENOMEM); else return (0); } int vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len) { vmm_mmio_free(vm->vmspace, gpa, len); return (0); } boolean_t vm_mem_allocated(struct vm *vm, vm_paddr_t gpa) { int i; vm_paddr_t gpabase, gpalimit; for (i = 0; i < vm->num_mem_segs; i++) { gpabase = vm->mem_segs[i].gpa; gpalimit = gpabase + vm->mem_segs[i].len; if (gpa >= gpabase && gpa < gpalimit) return (TRUE); /* 'gpa' is regular memory */ } if (ppt_is_mmio(vm, gpa)) return (TRUE); /* 'gpa' is pci passthru mmio */ return (FALSE); } int vm_malloc(struct vm *vm, vm_paddr_t gpa, size_t len) { int available, allocated; struct mem_seg *seg; vm_object_t object; vm_paddr_t g; if ((gpa & PAGE_MASK) || (len & PAGE_MASK) || len == 0) return (EINVAL); available = allocated = 0; g = gpa; while (g < gpa + len) { if (vm_mem_allocated(vm, g)) allocated++; else available++; g += PAGE_SIZE; } /* * If there are some allocated and some available pages in the address * range then it is an error. */ if (allocated && available) return (EINVAL); /* * If the entire address range being requested has already been * allocated then there isn't anything more to do. */ if (allocated && available == 0) return (0); if (vm->num_mem_segs >= VM_MAX_MEMORY_SEGMENTS) return (E2BIG); seg = &vm->mem_segs[vm->num_mem_segs]; if ((object = vmm_mem_alloc(vm->vmspace, gpa, len)) == NULL) return (ENOMEM); seg->gpa = gpa; seg->len = len; seg->object = object; seg->wired = FALSE; vm->num_mem_segs++; return (0); } static void vm_gpa_unwire(struct vm *vm) { int i, rv; struct mem_seg *seg; for (i = 0; i < vm->num_mem_segs; i++) { seg = &vm->mem_segs[i]; if (!seg->wired) continue; rv = vm_map_unwire(&vm->vmspace->vm_map, seg->gpa, seg->gpa + seg->len, VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); KASSERT(rv == KERN_SUCCESS, ("vm(%s) memory segment " "%#lx/%ld could not be unwired: %d", vm_name(vm), seg->gpa, seg->len, rv)); seg->wired = FALSE; } } static int vm_gpa_wire(struct vm *vm) { int i, rv; struct mem_seg *seg; for (i = 0; i < vm->num_mem_segs; i++) { seg = &vm->mem_segs[i]; if (seg->wired) continue; /* XXX rlimits? */ rv = vm_map_wire(&vm->vmspace->vm_map, seg->gpa, seg->gpa + seg->len, VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); if (rv != KERN_SUCCESS) break; seg->wired = TRUE; } if (i < vm->num_mem_segs) { /* * Undo the wiring before returning an error. */ vm_gpa_unwire(vm); return (EAGAIN); } return (0); } static void vm_iommu_modify(struct vm *vm, boolean_t map) { int i, sz; vm_paddr_t gpa, hpa; struct mem_seg *seg; void *vp, *cookie, *host_domain; sz = PAGE_SIZE; host_domain = iommu_host_domain(); for (i = 0; i < vm->num_mem_segs; i++) { seg = &vm->mem_segs[i]; KASSERT(seg->wired, ("vm(%s) memory segment %#lx/%ld not wired", vm_name(vm), seg->gpa, seg->len)); gpa = seg->gpa; while (gpa < seg->gpa + seg->len) { vp = vm_gpa_hold(vm, gpa, PAGE_SIZE, VM_PROT_WRITE, &cookie); KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx", vm_name(vm), gpa)); vm_gpa_release(cookie); hpa = DMAP_TO_PHYS((uintptr_t)vp); if (map) { iommu_create_mapping(vm->iommu, gpa, hpa, sz); iommu_remove_mapping(host_domain, hpa, sz); } else { iommu_remove_mapping(vm->iommu, gpa, sz); iommu_create_mapping(host_domain, hpa, hpa, sz); } gpa += PAGE_SIZE; } } /* * Invalidate the cached translations associated with the domain * from which pages were removed. */ if (map) iommu_invalidate_tlb(host_domain); else iommu_invalidate_tlb(vm->iommu); } #define vm_iommu_unmap(vm) vm_iommu_modify((vm), FALSE) #define vm_iommu_map(vm) vm_iommu_modify((vm), TRUE) int vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func) { int error; error = ppt_unassign_device(vm, bus, slot, func); if (error) return (error); if (ppt_assigned_devices(vm) == 0) { vm_iommu_unmap(vm); vm_gpa_unwire(vm); } return (0); } int vm_assign_pptdev(struct vm *vm, int bus, int slot, int func) { int error; vm_paddr_t maxaddr; /* * Virtual machines with pci passthru devices get special treatment: * - the guest physical memory is wired * - the iommu is programmed to do the 'gpa' to 'hpa' translation * * We need to do this before the first pci passthru device is attached. */ if (ppt_assigned_devices(vm) == 0) { KASSERT(vm->iommu == NULL, ("vm_assign_pptdev: iommu must be NULL")); maxaddr = vmm_mem_maxaddr(); vm->iommu = iommu_create_domain(maxaddr); error = vm_gpa_wire(vm); if (error) return (error); vm_iommu_map(vm); } error = ppt_assign_device(vm, bus, slot, func); return (error); } void * vm_gpa_hold(struct vm *vm, vm_paddr_t gpa, size_t len, int reqprot, void **cookie) { int count, pageoff; vm_page_t m; pageoff = gpa & PAGE_MASK; if (len > PAGE_SIZE - pageoff) panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len); count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map, trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1); if (count == 1) { *cookie = m; return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff)); } else { *cookie = NULL; return (NULL); } } void vm_gpa_release(void *cookie) { vm_page_t m = cookie; vm_page_lock(m); vm_page_unhold(m); vm_page_unlock(m); } int vm_gpabase2memseg(struct vm *vm, vm_paddr_t gpabase, struct vm_memory_segment *seg) { int i; for (i = 0; i < vm->num_mem_segs; i++) { if (gpabase == vm->mem_segs[i].gpa) { seg->gpa = vm->mem_segs[i].gpa; seg->len = vm->mem_segs[i].len; seg->wired = vm->mem_segs[i].wired; return (0); } } return (-1); } int vm_get_memobj(struct vm *vm, vm_paddr_t gpa, size_t len, vm_offset_t *offset, struct vm_object **object) { int i; size_t seg_len; vm_paddr_t seg_gpa; vm_object_t seg_obj; for (i = 0; i < vm->num_mem_segs; i++) { if ((seg_obj = vm->mem_segs[i].object) == NULL) continue; seg_gpa = vm->mem_segs[i].gpa; seg_len = vm->mem_segs[i].len; if (gpa >= seg_gpa && gpa < seg_gpa + seg_len) { *offset = gpa - seg_gpa; *object = seg_obj; vm_object_reference(seg_obj); return (0); } } return (EINVAL); } int vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval) { if (vcpu < 0 || vcpu >= VM_MAXCPU) return (EINVAL); if (reg >= VM_REG_LAST) return (EINVAL); return (VMGETREG(vm->cookie, vcpu, reg, retval)); } int vm_set_register(struct vm *vm, int vcpu, int reg, uint64_t val) { if (vcpu < 0 || vcpu >= VM_MAXCPU) return (EINVAL); if (reg >= VM_REG_LAST) return (EINVAL); return (VMSETREG(vm->cookie, vcpu, reg, val)); } static boolean_t is_descriptor_table(int reg) { switch (reg) { case VM_REG_GUEST_IDTR: case VM_REG_GUEST_GDTR: return (TRUE); default: return (FALSE); } } static boolean_t is_segment_register(int reg) { switch (reg) { case VM_REG_GUEST_ES: case VM_REG_GUEST_CS: case VM_REG_GUEST_SS: case VM_REG_GUEST_DS: case VM_REG_GUEST_FS: case VM_REG_GUEST_GS: case VM_REG_GUEST_TR: case VM_REG_GUEST_LDTR: return (TRUE); default: return (FALSE); } } int vm_get_seg_desc(struct vm *vm, int vcpu, int reg, struct seg_desc *desc) { if (vcpu < 0 || vcpu >= VM_MAXCPU) return (EINVAL); if (!is_segment_register(reg) && !is_descriptor_table(reg)) return (EINVAL); return (VMGETDESC(vm->cookie, vcpu, reg, desc)); } int vm_set_seg_desc(struct vm *vm, int vcpu, int reg, struct seg_desc *desc) { if (vcpu < 0 || vcpu >= VM_MAXCPU) return (EINVAL); if (!is_segment_register(reg) && !is_descriptor_table(reg)) return (EINVAL); return (VMSETDESC(vm->cookie, vcpu, reg, desc)); } static void restore_guest_fpustate(struct vcpu *vcpu) { /* flush host state to the pcb */ fpuexit(curthread); /* restore guest FPU state */ fpu_stop_emulating(); fpurestore(vcpu->guestfpu); /* restore guest XCR0 if XSAVE is enabled in the host */ if (rcr4() & CR4_XSAVE) load_xcr(0, vcpu->guest_xcr0); /* * The FPU is now "dirty" with the guest's state so turn on emulation * to trap any access to the FPU by the host. */ fpu_start_emulating(); } static void save_guest_fpustate(struct vcpu *vcpu) { if ((rcr0() & CR0_TS) == 0) panic("fpu emulation not enabled in host!"); /* save guest XCR0 and restore host XCR0 */ if (rcr4() & CR4_XSAVE) { vcpu->guest_xcr0 = rxcr(0); load_xcr(0, vmm_get_host_xcr0()); } /* save guest FPU state */ fpu_stop_emulating(); fpusave(vcpu->guestfpu); fpu_start_emulating(); } static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle"); static int vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate, bool from_idle) { int error; vcpu_assert_locked(vcpu); /* * State transitions from the vmmdev_ioctl() must always begin from * the VCPU_IDLE state. This guarantees that there is only a single * ioctl() operating on a vcpu at any point. */ if (from_idle) { while (vcpu->state != VCPU_IDLE) msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz); } else { KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from " "vcpu idle state")); } if (vcpu->state == VCPU_RUNNING) { KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d " "mismatch for running vcpu", curcpu, vcpu->hostcpu)); } else { KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a " "vcpu that is not running", vcpu->hostcpu)); } /* * The following state transitions are allowed: * IDLE -> FROZEN -> IDLE * FROZEN -> RUNNING -> FROZEN * FROZEN -> SLEEPING -> FROZEN */ switch (vcpu->state) { case VCPU_IDLE: case VCPU_RUNNING: case VCPU_SLEEPING: error = (newstate != VCPU_FROZEN); break; case VCPU_FROZEN: error = (newstate == VCPU_FROZEN); break; default: error = 1; break; } if (error) return (EBUSY); vcpu->state = newstate; if (newstate == VCPU_RUNNING) vcpu->hostcpu = curcpu; else vcpu->hostcpu = NOCPU; if (newstate == VCPU_IDLE) wakeup(&vcpu->state); return (0); } static void vcpu_require_state(struct vm *vm, int vcpuid, enum vcpu_state newstate) { int error; if ((error = vcpu_set_state(vm, vcpuid, newstate, false)) != 0) panic("Error %d setting state to %d\n", error, newstate); } static void vcpu_require_state_locked(struct vcpu *vcpu, enum vcpu_state newstate) { int error; if ((error = vcpu_set_state_locked(vcpu, newstate, false)) != 0) panic("Error %d setting state to %d", error, newstate); } static void vm_set_rendezvous_func(struct vm *vm, vm_rendezvous_func_t func) { KASSERT(mtx_owned(&vm->rendezvous_mtx), ("rendezvous_mtx not locked")); /* * Update 'rendezvous_func' and execute a write memory barrier to * ensure that it is visible across all host cpus. This is not needed * for correctness but it does ensure that all the vcpus will notice * that the rendezvous is requested immediately. */ vm->rendezvous_func = func; wmb(); } #define RENDEZVOUS_CTR0(vm, vcpuid, fmt) \ do { \ if (vcpuid >= 0) \ VCPU_CTR0(vm, vcpuid, fmt); \ else \ VM_CTR0(vm, fmt); \ } while (0) static void vm_handle_rendezvous(struct vm *vm, int vcpuid) { KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU), ("vm_handle_rendezvous: invalid vcpuid %d", vcpuid)); mtx_lock(&vm->rendezvous_mtx); while (vm->rendezvous_func != NULL) { /* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */ CPU_AND(&vm->rendezvous_req_cpus, &vm->active_cpus); if (vcpuid != -1 && CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) && !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) { VCPU_CTR0(vm, vcpuid, "Calling rendezvous func"); (*vm->rendezvous_func)(vm, vcpuid, vm->rendezvous_arg); CPU_SET(vcpuid, &vm->rendezvous_done_cpus); } if (CPU_CMP(&vm->rendezvous_req_cpus, &vm->rendezvous_done_cpus) == 0) { VCPU_CTR0(vm, vcpuid, "Rendezvous completed"); vm_set_rendezvous_func(vm, NULL); wakeup(&vm->rendezvous_func); break; } RENDEZVOUS_CTR0(vm, vcpuid, "Wait for rendezvous completion"); mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0, "vmrndv", 0); } mtx_unlock(&vm->rendezvous_mtx); } /* * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run. */ static int vm_handle_hlt(struct vm *vm, int vcpuid, bool intr_disabled, bool *retu) { struct vcpu *vcpu; const char *wmesg; int t, vcpu_halted, vm_halted; KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted")); vcpu = &vm->vcpu[vcpuid]; vcpu_halted = 0; vm_halted = 0; vcpu_lock(vcpu); while (1) { /* * Do a final check for pending NMI or interrupts before * really putting this thread to sleep. Also check for * software events that would cause this vcpu to wakeup. * * These interrupts/events could have happened after the * vcpu returned from VMRUN() and before it acquired the * vcpu lock above. */ if (vm->rendezvous_func != NULL || vm->suspend) break; if (vm_nmi_pending(vm, vcpuid)) break; if (!intr_disabled) { if (vm_extint_pending(vm, vcpuid) || vlapic_pending_intr(vcpu->vlapic, NULL)) { break; } } /* * Some Linux guests implement "halt" by having all vcpus * execute HLT with interrupts disabled. 'halted_cpus' keeps * track of the vcpus that have entered this state. When all * vcpus enter the halted state the virtual machine is halted. */ if (intr_disabled) { wmesg = "vmhalt"; VCPU_CTR0(vm, vcpuid, "Halted"); if (!vcpu_halted && halt_detection_enabled) { vcpu_halted = 1; CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus); } if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) { vm_halted = 1; break; } } else { wmesg = "vmidle"; } t = ticks; vcpu_require_state_locked(vcpu, VCPU_SLEEPING); msleep_spin(vcpu, &vcpu->mtx, wmesg, 0); vcpu_require_state_locked(vcpu, VCPU_FROZEN); vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t); } if (vcpu_halted) CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus); vcpu_unlock(vcpu); if (vm_halted) vm_suspend(vm, VM_SUSPEND_HALT); return (0); } static int vm_handle_paging(struct vm *vm, int vcpuid, bool *retu) { int rv, ftype; struct vm_map *map; struct vcpu *vcpu; struct vm_exit *vme; vcpu = &vm->vcpu[vcpuid]; vme = &vcpu->exitinfo; ftype = vme->u.paging.fault_type; KASSERT(ftype == VM_PROT_READ || ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE, ("vm_handle_paging: invalid fault_type %d", ftype)); if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) { rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace), vme->u.paging.gpa, ftype); if (rv == 0) goto done; } map = &vm->vmspace->vm_map; rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL); VCPU_CTR3(vm, vcpuid, "vm_handle_paging rv = %d, gpa = %#lx, " "ftype = %d", rv, vme->u.paging.gpa, ftype); if (rv != KERN_SUCCESS) return (EFAULT); done: /* restart execution at the faulting instruction */ vme->inst_length = 0; return (0); } static int vm_handle_inst_emul(struct vm *vm, int vcpuid, bool *retu) { struct vie *vie; struct vcpu *vcpu; struct vm_exit *vme; uint64_t gla, gpa; struct vm_guest_paging *paging; mem_region_read_t mread; mem_region_write_t mwrite; int error; vcpu = &vm->vcpu[vcpuid]; vme = &vcpu->exitinfo; gla = vme->u.inst_emul.gla; gpa = vme->u.inst_emul.gpa; vie = &vme->u.inst_emul.vie; paging = &vme->u.inst_emul.paging; vie_init(vie); /* Fetch, decode and emulate the faulting instruction */ error = vmm_fetch_instruction(vm, vcpuid, paging, vme->rip, vme->inst_length, vie); if (error == 1) return (0); /* Resume guest to handle page fault */ else if (error == -1) return (EFAULT); else if (error != 0) panic("%s: vmm_fetch_instruction error %d", __func__, error); if (vmm_decode_instruction(vm, vcpuid, gla, paging->cpu_mode, vie) != 0) return (EFAULT); /* return to userland unless this is an in-kernel emulated device */ if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) { mread = lapic_mmio_read; mwrite = lapic_mmio_write; } else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) { mread = vioapic_mmio_read; mwrite = vioapic_mmio_write; } else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) { mread = vhpet_mmio_read; mwrite = vhpet_mmio_write; } else { *retu = true; return (0); } error = vmm_emulate_instruction(vm, vcpuid, gpa, vie, mread, mwrite, retu); return (error); } static int vm_handle_suspend(struct vm *vm, int vcpuid, bool *retu) { int i, done; struct vcpu *vcpu; done = 0; vcpu = &vm->vcpu[vcpuid]; CPU_SET_ATOMIC(vcpuid, &vm->suspended_cpus); /* * Wait until all 'active_cpus' have suspended themselves. * * Since a VM may be suspended at any time including when one or * more vcpus are doing a rendezvous we need to call the rendezvous * handler while we are waiting to prevent a deadlock. */ vcpu_lock(vcpu); while (1) { if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { VCPU_CTR0(vm, vcpuid, "All vcpus suspended"); break; } if (vm->rendezvous_func == NULL) { VCPU_CTR0(vm, vcpuid, "Sleeping during suspend"); vcpu_require_state_locked(vcpu, VCPU_SLEEPING); msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz); vcpu_require_state_locked(vcpu, VCPU_FROZEN); } else { VCPU_CTR0(vm, vcpuid, "Rendezvous during suspend"); vcpu_unlock(vcpu); vm_handle_rendezvous(vm, vcpuid); vcpu_lock(vcpu); } } vcpu_unlock(vcpu); /* * Wakeup the other sleeping vcpus and return to userspace. */ for (i = 0; i < VM_MAXCPU; i++) { if (CPU_ISSET(i, &vm->suspended_cpus)) { vcpu_notify_event(vm, i, false); } } *retu = true; return (0); } int vm_suspend(struct vm *vm, enum vm_suspend_how how) { int i; if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST) return (EINVAL); if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) { VM_CTR2(vm, "virtual machine already suspended %d/%d", vm->suspend, how); return (EALREADY); } VM_CTR1(vm, "virtual machine successfully suspended %d", how); /* * Notify all active vcpus that they are now suspended. */ for (i = 0; i < VM_MAXCPU; i++) { if (CPU_ISSET(i, &vm->active_cpus)) vcpu_notify_event(vm, i, false); } return (0); } void vm_exit_suspended(struct vm *vm, int vcpuid, uint64_t rip) { struct vm_exit *vmexit; KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST, ("vm_exit_suspended: invalid suspend type %d", vm->suspend)); vmexit = vm_exitinfo(vm, vcpuid); vmexit->rip = rip; vmexit->inst_length = 0; vmexit->exitcode = VM_EXITCODE_SUSPENDED; vmexit->u.suspended.how = vm->suspend; } +void +vm_exit_rendezvous(struct vm *vm, int vcpuid, uint64_t rip) +{ + struct vm_exit *vmexit; + + KASSERT(vm->rendezvous_func != NULL, ("rendezvous not in progress")); + + vmexit = vm_exitinfo(vm, vcpuid); + vmexit->rip = rip; + vmexit->inst_length = 0; + vmexit->exitcode = VM_EXITCODE_RENDEZVOUS; + vmm_stat_incr(vm, vcpuid, VMEXIT_RENDEZVOUS, 1); +} + +void +vm_exit_astpending(struct vm *vm, int vcpuid, uint64_t rip) +{ + struct vm_exit *vmexit; + + vmexit = vm_exitinfo(vm, vcpuid); + vmexit->rip = rip; + vmexit->inst_length = 0; + vmexit->exitcode = VM_EXITCODE_BOGUS; + vmm_stat_incr(vm, vcpuid, VMEXIT_ASTPENDING, 1); +} + int vm_run(struct vm *vm, struct vm_run *vmrun) { int error, vcpuid; struct vcpu *vcpu; struct pcb *pcb; uint64_t tscval, rip; struct vm_exit *vme; bool retu, intr_disabled; pmap_t pmap; void *rptr, *sptr; vcpuid = vmrun->cpuid; if (vcpuid < 0 || vcpuid >= VM_MAXCPU) return (EINVAL); if (!CPU_ISSET(vcpuid, &vm->active_cpus)) return (EINVAL); if (CPU_ISSET(vcpuid, &vm->suspended_cpus)) return (EINVAL); rptr = &vm->rendezvous_func; sptr = &vm->suspend; pmap = vmspace_pmap(vm->vmspace); vcpu = &vm->vcpu[vcpuid]; vme = &vcpu->exitinfo; rip = vmrun->rip; restart: critical_enter(); KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active), ("vm_run: absurd pm_active")); tscval = rdtsc(); pcb = PCPU_GET(curpcb); set_pcb_flags(pcb, PCB_FULL_IRET); restore_guest_msrs(vm, vcpuid); restore_guest_fpustate(vcpu); vcpu_require_state(vm, vcpuid, VCPU_RUNNING); error = VMRUN(vm->cookie, vcpuid, rip, pmap, rptr, sptr); vcpu_require_state(vm, vcpuid, VCPU_FROZEN); save_guest_fpustate(vcpu); restore_host_msrs(vm, vcpuid); vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval); critical_exit(); if (error == 0) { retu = false; switch (vme->exitcode) { case VM_EXITCODE_SUSPENDED: error = vm_handle_suspend(vm, vcpuid, &retu); break; case VM_EXITCODE_IOAPIC_EOI: vioapic_process_eoi(vm, vcpuid, vme->u.ioapic_eoi.vector); break; case VM_EXITCODE_RENDEZVOUS: vm_handle_rendezvous(vm, vcpuid); error = 0; break; case VM_EXITCODE_HLT: intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0); error = vm_handle_hlt(vm, vcpuid, intr_disabled, &retu); break; case VM_EXITCODE_PAGING: error = vm_handle_paging(vm, vcpuid, &retu); break; case VM_EXITCODE_INST_EMUL: error = vm_handle_inst_emul(vm, vcpuid, &retu); break; case VM_EXITCODE_INOUT: case VM_EXITCODE_INOUT_STR: error = vm_handle_inout(vm, vcpuid, vme, &retu); break; default: retu = true; /* handled in userland */ break; } } if (error == 0 && retu == false) { rip = vme->rip + vme->inst_length; goto restart; } /* copy the exit information */ bcopy(vme, &vmrun->vm_exit, sizeof(struct vm_exit)); return (error); } int vm_inject_exception(struct vm *vm, int vcpuid, struct vm_exception *exception) { struct vcpu *vcpu; if (vcpuid < 0 || vcpuid >= VM_MAXCPU) return (EINVAL); if (exception->vector < 0 || exception->vector >= 32) return (EINVAL); vcpu = &vm->vcpu[vcpuid]; if (vcpu->exception_pending) { VCPU_CTR2(vm, vcpuid, "Unable to inject exception %d due to " "pending exception %d", exception->vector, vcpu->exception.vector); return (EBUSY); } vcpu->exception_pending = 1; vcpu->exception = *exception; VCPU_CTR1(vm, vcpuid, "Exception %d pending", exception->vector); return (0); } int vm_exception_pending(struct vm *vm, int vcpuid, struct vm_exception *exception) { struct vcpu *vcpu; int pending; KASSERT(vcpuid >= 0 && vcpuid < VM_MAXCPU, ("invalid vcpu %d", vcpuid)); vcpu = &vm->vcpu[vcpuid]; pending = vcpu->exception_pending; if (pending) { vcpu->exception_pending = 0; *exception = vcpu->exception; VCPU_CTR1(vm, vcpuid, "Exception %d delivered", exception->vector); } return (pending); } static void vm_inject_fault(struct vm *vm, int vcpuid, struct vm_exception *exception) { struct vm_exit *vmexit; int error; error = vm_inject_exception(vm, vcpuid, exception); KASSERT(error == 0, ("vm_inject_exception error %d", error)); /* * A fault-like exception allows the instruction to be restarted * after the exception handler returns. * * By setting the inst_length to 0 we ensure that the instruction * pointer remains at the faulting instruction. */ vmexit = vm_exitinfo(vm, vcpuid); vmexit->inst_length = 0; } void vm_inject_pf(struct vm *vm, int vcpuid, int error_code, uint64_t cr2) { struct vm_exception pf = { .vector = IDT_PF, .error_code_valid = 1, .error_code = error_code }; int error; VCPU_CTR2(vm, vcpuid, "Injecting page fault: error_code %#x, cr2 %#lx", error_code, cr2); error = vm_set_register(vm, vcpuid, VM_REG_GUEST_CR2, cr2); KASSERT(error == 0, ("vm_set_register(cr2) error %d", error)); vm_inject_fault(vm, vcpuid, &pf); } void vm_inject_gp(struct vm *vm, int vcpuid) { struct vm_exception gpf = { .vector = IDT_GP, .error_code_valid = 1, .error_code = 0 }; vm_inject_fault(vm, vcpuid, &gpf); } void vm_inject_ud(struct vm *vm, int vcpuid) { struct vm_exception udf = { .vector = IDT_UD, .error_code_valid = 0 }; vm_inject_fault(vm, vcpuid, &udf); } static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu"); int vm_inject_nmi(struct vm *vm, int vcpuid) { struct vcpu *vcpu; if (vcpuid < 0 || vcpuid >= VM_MAXCPU) return (EINVAL); vcpu = &vm->vcpu[vcpuid]; vcpu->nmi_pending = 1; vcpu_notify_event(vm, vcpuid, false); return (0); } int vm_nmi_pending(struct vm *vm, int vcpuid) { struct vcpu *vcpu; if (vcpuid < 0 || vcpuid >= VM_MAXCPU) panic("vm_nmi_pending: invalid vcpuid %d", vcpuid); vcpu = &vm->vcpu[vcpuid]; return (vcpu->nmi_pending); } void vm_nmi_clear(struct vm *vm, int vcpuid) { struct vcpu *vcpu; if (vcpuid < 0 || vcpuid >= VM_MAXCPU) panic("vm_nmi_pending: invalid vcpuid %d", vcpuid); vcpu = &vm->vcpu[vcpuid]; if (vcpu->nmi_pending == 0) panic("vm_nmi_clear: inconsistent nmi_pending state"); vcpu->nmi_pending = 0; vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1); } static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu"); int vm_inject_extint(struct vm *vm, int vcpuid) { struct vcpu *vcpu; if (vcpuid < 0 || vcpuid >= VM_MAXCPU) return (EINVAL); vcpu = &vm->vcpu[vcpuid]; vcpu->extint_pending = 1; vcpu_notify_event(vm, vcpuid, false); return (0); } int vm_extint_pending(struct vm *vm, int vcpuid) { struct vcpu *vcpu; if (vcpuid < 0 || vcpuid >= VM_MAXCPU) panic("vm_extint_pending: invalid vcpuid %d", vcpuid); vcpu = &vm->vcpu[vcpuid]; return (vcpu->extint_pending); } void vm_extint_clear(struct vm *vm, int vcpuid) { struct vcpu *vcpu; if (vcpuid < 0 || vcpuid >= VM_MAXCPU) panic("vm_extint_pending: invalid vcpuid %d", vcpuid); vcpu = &vm->vcpu[vcpuid]; if (vcpu->extint_pending == 0) panic("vm_extint_clear: inconsistent extint_pending state"); vcpu->extint_pending = 0; vmm_stat_incr(vm, vcpuid, VCPU_EXTINT_COUNT, 1); } int vm_get_capability(struct vm *vm, int vcpu, int type, int *retval) { if (vcpu < 0 || vcpu >= VM_MAXCPU) return (EINVAL); if (type < 0 || type >= VM_CAP_MAX) return (EINVAL); return (VMGETCAP(vm->cookie, vcpu, type, retval)); } int vm_set_capability(struct vm *vm, int vcpu, int type, int val) { if (vcpu < 0 || vcpu >= VM_MAXCPU) return (EINVAL); if (type < 0 || type >= VM_CAP_MAX) return (EINVAL); return (VMSETCAP(vm->cookie, vcpu, type, val)); } uint64_t * vm_guest_msrs(struct vm *vm, int cpu) { return (vm->vcpu[cpu].guest_msrs); } struct vlapic * vm_lapic(struct vm *vm, int cpu) { return (vm->vcpu[cpu].vlapic); } struct vioapic * vm_ioapic(struct vm *vm) { return (vm->vioapic); } struct vhpet * vm_hpet(struct vm *vm) { return (vm->vhpet); } boolean_t vmm_is_pptdev(int bus, int slot, int func) { int found, i, n; int b, s, f; char *val, *cp, *cp2; /* * XXX * The length of an environment variable is limited to 128 bytes which * puts an upper limit on the number of passthru devices that may be * specified using a single environment variable. * * Work around this by scanning multiple environment variable * names instead of a single one - yuck! */ const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL }; /* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */ found = 0; for (i = 0; names[i] != NULL && !found; i++) { cp = val = getenv(names[i]); while (cp != NULL && *cp != '\0') { if ((cp2 = strchr(cp, ' ')) != NULL) *cp2 = '\0'; n = sscanf(cp, "%d/%d/%d", &b, &s, &f); if (n == 3 && bus == b && slot == s && func == f) { found = 1; break; } if (cp2 != NULL) *cp2++ = ' '; cp = cp2; } freeenv(val); } return (found); } void * vm_iommu_domain(struct vm *vm) { return (vm->iommu); } int vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state newstate, bool from_idle) { int error; struct vcpu *vcpu; if (vcpuid < 0 || vcpuid >= VM_MAXCPU) panic("vm_set_run_state: invalid vcpuid %d", vcpuid); vcpu = &vm->vcpu[vcpuid]; vcpu_lock(vcpu); error = vcpu_set_state_locked(vcpu, newstate, from_idle); vcpu_unlock(vcpu); return (error); } enum vcpu_state vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu) { struct vcpu *vcpu; enum vcpu_state state; if (vcpuid < 0 || vcpuid >= VM_MAXCPU) panic("vm_get_run_state: invalid vcpuid %d", vcpuid); vcpu = &vm->vcpu[vcpuid]; vcpu_lock(vcpu); state = vcpu->state; if (hostcpu != NULL) *hostcpu = vcpu->hostcpu; vcpu_unlock(vcpu); return (state); } int vm_activate_cpu(struct vm *vm, int vcpuid) { if (vcpuid < 0 || vcpuid >= VM_MAXCPU) return (EINVAL); if (CPU_ISSET(vcpuid, &vm->active_cpus)) return (EBUSY); VCPU_CTR0(vm, vcpuid, "activated"); CPU_SET_ATOMIC(vcpuid, &vm->active_cpus); return (0); } cpuset_t vm_active_cpus(struct vm *vm) { return (vm->active_cpus); } cpuset_t vm_suspended_cpus(struct vm *vm) { return (vm->suspended_cpus); } void * vcpu_stats(struct vm *vm, int vcpuid) { return (vm->vcpu[vcpuid].stats); } int vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state) { if (vcpuid < 0 || vcpuid >= VM_MAXCPU) return (EINVAL); *state = vm->vcpu[vcpuid].x2apic_state; return (0); } int vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state) { if (vcpuid < 0 || vcpuid >= VM_MAXCPU) return (EINVAL); if (state >= X2APIC_STATE_LAST) return (EINVAL); vm->vcpu[vcpuid].x2apic_state = state; vlapic_set_x2apic_state(vm, vcpuid, state); return (0); } /* * This function is called to ensure that a vcpu "sees" a pending event * as soon as possible: * - If the vcpu thread is sleeping then it is woken up. * - If the vcpu is running on a different host_cpu then an IPI will be directed * to the host_cpu to cause the vcpu to trap into the hypervisor. */ void vcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr) { int hostcpu; struct vcpu *vcpu; vcpu = &vm->vcpu[vcpuid]; vcpu_lock(vcpu); hostcpu = vcpu->hostcpu; if (vcpu->state == VCPU_RUNNING) { KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu")); if (hostcpu != curcpu) { if (lapic_intr) { vlapic_post_intr(vcpu->vlapic, hostcpu, vmm_ipinum); } else { ipi_cpu(hostcpu, vmm_ipinum); } } else { /* * If the 'vcpu' is running on 'curcpu' then it must * be sending a notification to itself (e.g. SELF_IPI). * The pending event will be picked up when the vcpu * transitions back to guest context. */ } } else { KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent " "with hostcpu %d", vcpu->state, hostcpu)); if (vcpu->state == VCPU_SLEEPING) wakeup_one(vcpu); } vcpu_unlock(vcpu); } struct vmspace * vm_get_vmspace(struct vm *vm) { return (vm->vmspace); } int vm_apicid2vcpuid(struct vm *vm, int apicid) { /* * XXX apic id is assumed to be numerically identical to vcpu id */ return (apicid); } void vm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest, vm_rendezvous_func_t func, void *arg) { int i; /* * Enforce that this function is called without any locks */ WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous"); KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU), ("vm_smp_rendezvous: invalid vcpuid %d", vcpuid)); restart: mtx_lock(&vm->rendezvous_mtx); if (vm->rendezvous_func != NULL) { /* * If a rendezvous is already in progress then we need to * call the rendezvous handler in case this 'vcpuid' is one * of the targets of the rendezvous. */ RENDEZVOUS_CTR0(vm, vcpuid, "Rendezvous already in progress"); mtx_unlock(&vm->rendezvous_mtx); vm_handle_rendezvous(vm, vcpuid); goto restart; } KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous " "rendezvous is still in progress")); RENDEZVOUS_CTR0(vm, vcpuid, "Initiating rendezvous"); vm->rendezvous_req_cpus = dest; CPU_ZERO(&vm->rendezvous_done_cpus); vm->rendezvous_arg = arg; vm_set_rendezvous_func(vm, func); mtx_unlock(&vm->rendezvous_mtx); /* * Wake up any sleeping vcpus and trigger a VM-exit in any running * vcpus so they handle the rendezvous as soon as possible. */ for (i = 0; i < VM_MAXCPU; i++) { if (CPU_ISSET(i, &dest)) vcpu_notify_event(vm, i, false); } vm_handle_rendezvous(vm, vcpuid); } struct vatpic * vm_atpic(struct vm *vm) { return (vm->vatpic); } struct vatpit * vm_atpit(struct vm *vm) { return (vm->vatpit); } enum vm_reg_name vm_segment_name(int seg) { static enum vm_reg_name seg_names[] = { VM_REG_GUEST_ES, VM_REG_GUEST_CS, VM_REG_GUEST_SS, VM_REG_GUEST_DS, VM_REG_GUEST_FS, VM_REG_GUEST_GS }; KASSERT(seg >= 0 && seg < nitems(seg_names), ("%s: invalid segment encoding %d", __func__, seg)); return (seg_names[seg]); } + + +/* + * Return the amount of in-use and wired memory for the VM. Since + * these are global stats, only return the values with for vCPU 0 + */ +VMM_STAT_DECLARE(VMM_MEM_RESIDENT); +VMM_STAT_DECLARE(VMM_MEM_WIRED); + +static void +vm_get_rescnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat) +{ + + if (vcpu == 0) { + vmm_stat_set(vm, vcpu, VMM_MEM_RESIDENT, + PAGE_SIZE * vmspace_resident_count(vm->vmspace)); + } +} + +static void +vm_get_wiredcnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat) +{ + + if (vcpu == 0) { + vmm_stat_set(vm, vcpu, VMM_MEM_WIRED, + PAGE_SIZE * pmap_wired_count(vmspace_pmap(vm->vmspace))); + } +} + +VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt); +VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt); Index: stable/10/sys/amd64/vmm/vmm_stat.c =================================================================== --- stable/10/sys/amd64/vmm/vmm_stat.c (revision 270073) +++ stable/10/sys/amd64/vmm/vmm_stat.c (revision 270074) @@ -1,161 +1,170 @@ /*- * Copyright (c) 2011 NetApp, Inc. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include "vmm_util.h" #include "vmm_stat.h" /* * 'vst_num_elems' is the total number of addressable statistic elements * 'vst_num_types' is the number of unique statistic types * * It is always true that 'vst_num_elems' is greater than or equal to * 'vst_num_types'. This is because a stat type may represent more than * one element (for e.g. VMM_STAT_ARRAY). */ static int vst_num_elems, vst_num_types; static struct vmm_stat_type *vsttab[MAX_VMM_STAT_ELEMS]; static MALLOC_DEFINE(M_VMM_STAT, "vmm stat", "vmm stat"); #define vst_size ((size_t)vst_num_elems * sizeof(uint64_t)) void vmm_stat_register(void *arg) { struct vmm_stat_type *vst = arg; /* We require all stats to identify themselves with a description */ if (vst->desc == NULL) return; if (vst->scope == VMM_STAT_SCOPE_INTEL && !vmm_is_intel()) return; if (vst->scope == VMM_STAT_SCOPE_AMD && !vmm_is_amd()) return; if (vst_num_elems + vst->nelems >= MAX_VMM_STAT_ELEMS) { printf("Cannot accomodate vmm stat type \"%s\"!\n", vst->desc); return; } vst->index = vst_num_elems; vst_num_elems += vst->nelems; vsttab[vst_num_types++] = vst; } int vmm_stat_copy(struct vm *vm, int vcpu, int *num_stats, uint64_t *buf) { - int i; + struct vmm_stat_type *vst; uint64_t *stats; + int i; if (vcpu < 0 || vcpu >= VM_MAXCPU) return (EINVAL); - + + /* Let stats functions update their counters */ + for (i = 0; i < vst_num_types; i++) { + vst = vsttab[i]; + if (vst->func != NULL) + (*vst->func)(vm, vcpu, vst); + } + + /* Copy over the stats */ stats = vcpu_stats(vm, vcpu); for (i = 0; i < vst_num_elems; i++) buf[i] = stats[i]; *num_stats = vst_num_elems; return (0); } void * vmm_stat_alloc(void) { return (malloc(vst_size, M_VMM_STAT, M_WAITOK)); } void vmm_stat_init(void *vp) { bzero(vp, vst_size); } void vmm_stat_free(void *vp) { free(vp, M_VMM_STAT); } int vmm_stat_desc_copy(int index, char *buf, int bufsize) { int i; struct vmm_stat_type *vst; for (i = 0; i < vst_num_types; i++) { vst = vsttab[i]; if (index >= vst->index && index < vst->index + vst->nelems) { if (vst->nelems > 1) { snprintf(buf, bufsize, "%s[%d]", vst->desc, index - vst->index); } else { strlcpy(buf, vst->desc, bufsize); } return (0); /* found it */ } } return (EINVAL); } /* global statistics */ VMM_STAT(VCPU_MIGRATIONS, "vcpu migration across host cpus"); VMM_STAT(VMEXIT_COUNT, "total number of vm exits"); VMM_STAT(VMEXIT_EXTINT, "vm exits due to external interrupt"); VMM_STAT(VMEXIT_HLT, "number of times hlt was intercepted"); VMM_STAT(VMEXIT_CR_ACCESS, "number of times %cr access was intercepted"); VMM_STAT(VMEXIT_RDMSR, "number of times rdmsr was intercepted"); VMM_STAT(VMEXIT_WRMSR, "number of times wrmsr was intercepted"); VMM_STAT(VMEXIT_MTRAP, "number of monitor trap exits"); VMM_STAT(VMEXIT_PAUSE, "number of times pause was intercepted"); VMM_STAT(VMEXIT_INTR_WINDOW, "vm exits due to interrupt window opening"); VMM_STAT(VMEXIT_NMI_WINDOW, "vm exits due to nmi window opening"); VMM_STAT(VMEXIT_INOUT, "number of times in/out was intercepted"); VMM_STAT(VMEXIT_CPUID, "number of times cpuid was intercepted"); VMM_STAT(VMEXIT_NESTED_FAULT, "vm exits due to nested page fault"); VMM_STAT(VMEXIT_INST_EMUL, "vm exits for instruction emulation"); VMM_STAT(VMEXIT_UNKNOWN, "number of vm exits for unknown reason"); VMM_STAT(VMEXIT_ASTPENDING, "number of times astpending at exit"); VMM_STAT(VMEXIT_USERSPACE, "number of vm exits handled in userspace"); VMM_STAT(VMEXIT_RENDEZVOUS, "number of times rendezvous pending at exit"); VMM_STAT(VMEXIT_EXCEPTION, "number of vm exits due to exceptions"); Index: stable/10/sys/amd64/vmm/vmm_stat.h =================================================================== --- stable/10/sys/amd64/vmm/vmm_stat.h (revision 270073) +++ stable/10/sys/amd64/vmm/vmm_stat.h (revision 270074) @@ -1,127 +1,160 @@ /*- * Copyright (c) 2011 NetApp, Inc. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #ifndef _VMM_STAT_H_ #define _VMM_STAT_H_ struct vm; #define MAX_VMM_STAT_ELEMS 64 /* arbitrary */ enum vmm_stat_scope { VMM_STAT_SCOPE_ANY, VMM_STAT_SCOPE_INTEL, /* Intel VMX specific statistic */ VMM_STAT_SCOPE_AMD, /* AMD SVM specific statistic */ }; +struct vmm_stat_type; +typedef void (*vmm_stat_func_t)(struct vm *vm, int vcpu, + struct vmm_stat_type *stat); + struct vmm_stat_type { int index; /* position in the stats buffer */ int nelems; /* standalone or array */ const char *desc; /* description of statistic */ + vmm_stat_func_t func; enum vmm_stat_scope scope; }; void vmm_stat_register(void *arg); -#define VMM_STAT_DEFINE(type, nelems, desc, scope) \ +#define VMM_STAT_FDEFINE(type, nelems, desc, func, scope) \ struct vmm_stat_type type[1] = { \ - { -1, nelems, desc, scope } \ + { -1, nelems, desc, func, scope } \ }; \ SYSINIT(type##_stat, SI_SUB_KLD, SI_ORDER_ANY, vmm_stat_register, type) +#define VMM_STAT_DEFINE(type, nelems, desc, scope) \ + VMM_STAT_FDEFINE(type, nelems, desc, NULL, scope) + #define VMM_STAT_DECLARE(type) \ extern struct vmm_stat_type type[1] #define VMM_STAT(type, desc) \ VMM_STAT_DEFINE(type, 1, desc, VMM_STAT_SCOPE_ANY) #define VMM_STAT_INTEL(type, desc) \ VMM_STAT_DEFINE(type, 1, desc, VMM_STAT_SCOPE_INTEL) #define VMM_STAT_AMD(type, desc) \ VMM_STAT_DEFINE(type, 1, desc, VMM_STAT_SCOPE_AMD) +#define VMM_STAT_FUNC(type, desc, func) \ + VMM_STAT_FDEFINE(type, 1, desc, func, VMM_STAT_SCOPE_ANY) + #define VMM_STAT_ARRAY(type, nelems, desc) \ VMM_STAT_DEFINE(type, nelems, desc, VMM_STAT_SCOPE_ANY) void *vmm_stat_alloc(void); void vmm_stat_init(void *vp); void vmm_stat_free(void *vp); /* * 'buf' should be at least fit 'MAX_VMM_STAT_TYPES' entries */ int vmm_stat_copy(struct vm *vm, int vcpu, int *num_stats, uint64_t *buf); int vmm_stat_desc_copy(int index, char *buf, int buflen); static void __inline vmm_stat_array_incr(struct vm *vm, int vcpu, struct vmm_stat_type *vst, int statidx, uint64_t x) { #ifdef VMM_KEEP_STATS uint64_t *stats; stats = vcpu_stats(vm, vcpu); if (vst->index >= 0 && statidx < vst->nelems) stats[vst->index + statidx] += x; #endif } - static void __inline +vmm_stat_array_set(struct vm *vm, int vcpu, struct vmm_stat_type *vst, + int statidx, uint64_t val) +{ +#ifdef VMM_KEEP_STATS + uint64_t *stats; + + stats = vcpu_stats(vm, vcpu); + + if (vst->index >= 0 && statidx < vst->nelems) + stats[vst->index + statidx] = val; +#endif +} + +static void __inline vmm_stat_incr(struct vm *vm, int vcpu, struct vmm_stat_type *vst, uint64_t x) { #ifdef VMM_KEEP_STATS vmm_stat_array_incr(vm, vcpu, vst, 0, x); +#endif +} + +static void __inline +vmm_stat_set(struct vm *vm, int vcpu, struct vmm_stat_type *vst, uint64_t val) +{ + +#ifdef VMM_KEEP_STATS + vmm_stat_array_set(vm, vcpu, vst, 0, val); #endif } VMM_STAT_DECLARE(VCPU_MIGRATIONS); VMM_STAT_DECLARE(VMEXIT_COUNT); VMM_STAT_DECLARE(VMEXIT_EXTINT); VMM_STAT_DECLARE(VMEXIT_HLT); VMM_STAT_DECLARE(VMEXIT_CR_ACCESS); VMM_STAT_DECLARE(VMEXIT_RDMSR); VMM_STAT_DECLARE(VMEXIT_WRMSR); VMM_STAT_DECLARE(VMEXIT_MTRAP); VMM_STAT_DECLARE(VMEXIT_PAUSE); VMM_STAT_DECLARE(VMEXIT_INTR_WINDOW); VMM_STAT_DECLARE(VMEXIT_NMI_WINDOW); VMM_STAT_DECLARE(VMEXIT_INOUT); VMM_STAT_DECLARE(VMEXIT_CPUID); VMM_STAT_DECLARE(VMEXIT_NESTED_FAULT); VMM_STAT_DECLARE(VMEXIT_INST_EMUL); VMM_STAT_DECLARE(VMEXIT_UNKNOWN); VMM_STAT_DECLARE(VMEXIT_ASTPENDING); VMM_STAT_DECLARE(VMEXIT_USERSPACE); VMM_STAT_DECLARE(VMEXIT_RENDEZVOUS); VMM_STAT_DECLARE(VMEXIT_EXCEPTION); #endif Index: stable/10/usr.sbin/bhyve/pci_emul.c =================================================================== --- stable/10/usr.sbin/bhyve/pci_emul.c (revision 270073) +++ stable/10/usr.sbin/bhyve/pci_emul.c (revision 270074) @@ -1,2024 +1,2023 @@ /*- * Copyright (c) 2011 NetApp, Inc. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include "acpi.h" #include "bhyverun.h" #include "inout.h" #include "ioapic.h" #include "mem.h" #include "pci_emul.h" #include "pci_irq.h" #include "pci_lpc.h" #define CONF1_ADDR_PORT 0x0cf8 #define CONF1_DATA_PORT 0x0cfc #define CONF1_ENABLE 0x80000000ul #define CFGWRITE(pi,off,val,b) \ do { \ if ((b) == 1) { \ pci_set_cfgdata8((pi),(off),(val)); \ } else if ((b) == 2) { \ pci_set_cfgdata16((pi),(off),(val)); \ } else { \ pci_set_cfgdata32((pi),(off),(val)); \ } \ } while (0) #define MAXBUSES (PCI_BUSMAX + 1) #define MAXSLOTS (PCI_SLOTMAX + 1) #define MAXFUNCS (PCI_FUNCMAX + 1) struct funcinfo { char *fi_name; char *fi_param; struct pci_devinst *fi_devi; }; struct intxinfo { int ii_count; int ii_pirq_pin; int ii_ioapic_irq; }; struct slotinfo { struct intxinfo si_intpins[4]; struct funcinfo si_funcs[MAXFUNCS]; }; struct businfo { uint16_t iobase, iolimit; /* I/O window */ uint32_t membase32, memlimit32; /* mmio window below 4GB */ uint64_t membase64, memlimit64; /* mmio window above 4GB */ struct slotinfo slotinfo[MAXSLOTS]; }; static struct businfo *pci_businfo[MAXBUSES]; SET_DECLARE(pci_devemu_set, struct pci_devemu); static uint64_t pci_emul_iobase; static uint64_t pci_emul_membase32; static uint64_t pci_emul_membase64; #define PCI_EMUL_IOBASE 0x2000 #define PCI_EMUL_IOLIMIT 0x10000 #define PCI_EMUL_MEMLIMIT32 0xE0000000 /* 3.5GB */ #define PCI_EMUL_MEMBASE64 0xD000000000UL #define PCI_EMUL_MEMLIMIT64 0xFD00000000UL static struct pci_devemu *pci_emul_finddev(char *name); static void pci_lintr_route(struct pci_devinst *pi); static void pci_lintr_update(struct pci_devinst *pi); static struct mem_range pci_mem_hole; /* * I/O access */ /* * Slot options are in the form: * * ::,[,] * [:],[,] * * slot is 0..31 * func is 0..7 * emul is a string describing the type of PCI device e.g. virtio-net * config is an optional string, depending on the device, that can be * used for configuration. * Examples are: * 1,virtio-net,tap0 * 3:0,dummy */ static void pci_parse_slot_usage(char *aopt) { fprintf(stderr, "Invalid PCI slot info field \"%s\"\n", aopt); } int pci_parse_slot(char *opt) { struct businfo *bi; struct slotinfo *si; char *emul, *config, *str, *cp; int error, bnum, snum, fnum; error = -1; str = strdup(opt); emul = config = NULL; if ((cp = strchr(str, ',')) != NULL) { *cp = '\0'; emul = cp + 1; if ((cp = strchr(emul, ',')) != NULL) { *cp = '\0'; config = cp + 1; } } else { pci_parse_slot_usage(opt); goto done; } /* :: */ if (sscanf(str, "%d:%d:%d", &bnum, &snum, &fnum) != 3) { bnum = 0; /* : */ if (sscanf(str, "%d:%d", &snum, &fnum) != 2) { fnum = 0; /* */ if (sscanf(str, "%d", &snum) != 1) { snum = -1; } } } if (bnum < 0 || bnum >= MAXBUSES || snum < 0 || snum >= MAXSLOTS || fnum < 0 || fnum >= MAXFUNCS) { pci_parse_slot_usage(opt); goto done; } if (pci_businfo[bnum] == NULL) pci_businfo[bnum] = calloc(1, sizeof(struct businfo)); bi = pci_businfo[bnum]; si = &bi->slotinfo[snum]; if (si->si_funcs[fnum].fi_name != NULL) { fprintf(stderr, "pci slot %d:%d already occupied!\n", snum, fnum); goto done; } if (pci_emul_finddev(emul) == NULL) { fprintf(stderr, "pci slot %d:%d: unknown device \"%s\"\n", snum, fnum, emul); goto done; } error = 0; si->si_funcs[fnum].fi_name = emul; si->si_funcs[fnum].fi_param = config; done: if (error) free(str); return (error); } static int pci_valid_pba_offset(struct pci_devinst *pi, uint64_t offset) { if (offset < pi->pi_msix.pba_offset) return (0); if (offset >= pi->pi_msix.pba_offset + pi->pi_msix.pba_size) { return (0); } return (1); } int pci_emul_msix_twrite(struct pci_devinst *pi, uint64_t offset, int size, uint64_t value) { int msix_entry_offset; int tab_index; char *dest; /* support only 4 or 8 byte writes */ if (size != 4 && size != 8) return (-1); /* * Return if table index is beyond what device supports */ tab_index = offset / MSIX_TABLE_ENTRY_SIZE; if (tab_index >= pi->pi_msix.table_count) return (-1); msix_entry_offset = offset % MSIX_TABLE_ENTRY_SIZE; /* support only aligned writes */ if ((msix_entry_offset % size) != 0) return (-1); dest = (char *)(pi->pi_msix.table + tab_index); dest += msix_entry_offset; if (size == 4) *((uint32_t *)dest) = value; else *((uint64_t *)dest) = value; return (0); } uint64_t pci_emul_msix_tread(struct pci_devinst *pi, uint64_t offset, int size) { char *dest; int msix_entry_offset; int tab_index; uint64_t retval = ~0; /* * The PCI standard only allows 4 and 8 byte accesses to the MSI-X * table but we also allow 1 byte access to accomodate reads from * ddb. */ if (size != 1 && size != 4 && size != 8) return (retval); msix_entry_offset = offset % MSIX_TABLE_ENTRY_SIZE; /* support only aligned reads */ if ((msix_entry_offset % size) != 0) { return (retval); } tab_index = offset / MSIX_TABLE_ENTRY_SIZE; if (tab_index < pi->pi_msix.table_count) { /* valid MSI-X Table access */ dest = (char *)(pi->pi_msix.table + tab_index); dest += msix_entry_offset; if (size == 1) retval = *((uint8_t *)dest); else if (size == 4) retval = *((uint32_t *)dest); else retval = *((uint64_t *)dest); } else if (pci_valid_pba_offset(pi, offset)) { /* return 0 for PBA access */ retval = 0; } return (retval); } int pci_msix_table_bar(struct pci_devinst *pi) { if (pi->pi_msix.table != NULL) return (pi->pi_msix.table_bar); else return (-1); } int pci_msix_pba_bar(struct pci_devinst *pi) { if (pi->pi_msix.table != NULL) return (pi->pi_msix.pba_bar); else return (-1); } static int pci_emul_io_handler(struct vmctx *ctx, int vcpu, int in, int port, int bytes, uint32_t *eax, void *arg) { struct pci_devinst *pdi = arg; struct pci_devemu *pe = pdi->pi_d; uint64_t offset; int i; for (i = 0; i <= PCI_BARMAX; i++) { if (pdi->pi_bar[i].type == PCIBAR_IO && port >= pdi->pi_bar[i].addr && port + bytes <= pdi->pi_bar[i].addr + pdi->pi_bar[i].size) { offset = port - pdi->pi_bar[i].addr; if (in) *eax = (*pe->pe_barread)(ctx, vcpu, pdi, i, offset, bytes); else (*pe->pe_barwrite)(ctx, vcpu, pdi, i, offset, bytes, *eax); return (0); } } return (-1); } static int pci_emul_mem_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr, int size, uint64_t *val, void *arg1, long arg2) { struct pci_devinst *pdi = arg1; struct pci_devemu *pe = pdi->pi_d; uint64_t offset; int bidx = (int) arg2; assert(bidx <= PCI_BARMAX); assert(pdi->pi_bar[bidx].type == PCIBAR_MEM32 || pdi->pi_bar[bidx].type == PCIBAR_MEM64); assert(addr >= pdi->pi_bar[bidx].addr && addr + size <= pdi->pi_bar[bidx].addr + pdi->pi_bar[bidx].size); offset = addr - pdi->pi_bar[bidx].addr; if (dir == MEM_F_WRITE) { if (size == 8) { (*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset, 4, *val & 0xffffffff); (*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset + 4, 4, *val >> 32); } else { (*pe->pe_barwrite)(ctx, vcpu, pdi, bidx, offset, size, *val); } } else { if (size == 8) { *val = (*pe->pe_barread)(ctx, vcpu, pdi, bidx, offset, 4); *val |= (*pe->pe_barread)(ctx, vcpu, pdi, bidx, offset + 4, 4) << 32; } else { *val = (*pe->pe_barread)(ctx, vcpu, pdi, bidx, offset, size); } } return (0); } static int pci_emul_alloc_resource(uint64_t *baseptr, uint64_t limit, uint64_t size, uint64_t *addr) { uint64_t base; assert((size & (size - 1)) == 0); /* must be a power of 2 */ base = roundup2(*baseptr, size); if (base + size <= limit) { *addr = base; *baseptr = base + size; return (0); } else return (-1); } int pci_emul_alloc_bar(struct pci_devinst *pdi, int idx, enum pcibar_type type, uint64_t size) { return (pci_emul_alloc_pbar(pdi, idx, 0, type, size)); } /* * Register (or unregister) the MMIO or I/O region associated with the BAR * register 'idx' of an emulated pci device. */ static void modify_bar_registration(struct pci_devinst *pi, int idx, int registration) { int error; struct inout_port iop; struct mem_range mr; switch (pi->pi_bar[idx].type) { case PCIBAR_IO: bzero(&iop, sizeof(struct inout_port)); iop.name = pi->pi_name; iop.port = pi->pi_bar[idx].addr; iop.size = pi->pi_bar[idx].size; if (registration) { iop.flags = IOPORT_F_INOUT; iop.handler = pci_emul_io_handler; iop.arg = pi; error = register_inout(&iop); } else error = unregister_inout(&iop); break; case PCIBAR_MEM32: case PCIBAR_MEM64: bzero(&mr, sizeof(struct mem_range)); mr.name = pi->pi_name; mr.base = pi->pi_bar[idx].addr; mr.size = pi->pi_bar[idx].size; if (registration) { mr.flags = MEM_F_RW; mr.handler = pci_emul_mem_handler; mr.arg1 = pi; mr.arg2 = idx; error = register_mem(&mr); } else error = unregister_mem(&mr); break; default: error = EINVAL; break; } assert(error == 0); } static void unregister_bar(struct pci_devinst *pi, int idx) { modify_bar_registration(pi, idx, 0); } static void register_bar(struct pci_devinst *pi, int idx) { modify_bar_registration(pi, idx, 1); } /* Are we decoding i/o port accesses for the emulated pci device? */ static int porten(struct pci_devinst *pi) { uint16_t cmd; cmd = pci_get_cfgdata16(pi, PCIR_COMMAND); return (cmd & PCIM_CMD_PORTEN); } /* Are we decoding memory accesses for the emulated pci device? */ static int memen(struct pci_devinst *pi) { uint16_t cmd; cmd = pci_get_cfgdata16(pi, PCIR_COMMAND); return (cmd & PCIM_CMD_MEMEN); } /* * Update the MMIO or I/O address that is decoded by the BAR register. * * If the pci device has enabled the address space decoding then intercept * the address range decoded by the BAR register. */ static void update_bar_address(struct pci_devinst *pi, uint64_t addr, int idx, int type) { int decode; if (pi->pi_bar[idx].type == PCIBAR_IO) decode = porten(pi); else decode = memen(pi); if (decode) unregister_bar(pi, idx); switch (type) { case PCIBAR_IO: case PCIBAR_MEM32: pi->pi_bar[idx].addr = addr; break; case PCIBAR_MEM64: pi->pi_bar[idx].addr &= ~0xffffffffUL; pi->pi_bar[idx].addr |= addr; break; case PCIBAR_MEMHI64: pi->pi_bar[idx].addr &= 0xffffffff; pi->pi_bar[idx].addr |= addr; break; default: assert(0); } if (decode) register_bar(pi, idx); } int pci_emul_alloc_pbar(struct pci_devinst *pdi, int idx, uint64_t hostbase, enum pcibar_type type, uint64_t size) { int error; uint64_t *baseptr, limit, addr, mask, lobits, bar; assert(idx >= 0 && idx <= PCI_BARMAX); if ((size & (size - 1)) != 0) size = 1UL << flsl(size); /* round up to a power of 2 */ /* Enforce minimum BAR sizes required by the PCI standard */ if (type == PCIBAR_IO) { if (size < 4) size = 4; } else { if (size < 16) size = 16; } switch (type) { case PCIBAR_NONE: baseptr = NULL; addr = mask = lobits = 0; break; case PCIBAR_IO: baseptr = &pci_emul_iobase; limit = PCI_EMUL_IOLIMIT; mask = PCIM_BAR_IO_BASE; lobits = PCIM_BAR_IO_SPACE; break; case PCIBAR_MEM64: /* * XXX * Some drivers do not work well if the 64-bit BAR is allocated * above 4GB. Allow for this by allocating small requests under * 4GB unless then allocation size is larger than some arbitrary * number (32MB currently). */ if (size > 32 * 1024 * 1024) { /* * XXX special case for device requiring peer-peer DMA */ if (size == 0x100000000UL) baseptr = &hostbase; else baseptr = &pci_emul_membase64; limit = PCI_EMUL_MEMLIMIT64; mask = PCIM_BAR_MEM_BASE; lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64 | PCIM_BAR_MEM_PREFETCH; break; } else { baseptr = &pci_emul_membase32; limit = PCI_EMUL_MEMLIMIT32; mask = PCIM_BAR_MEM_BASE; lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64; } break; case PCIBAR_MEM32: baseptr = &pci_emul_membase32; limit = PCI_EMUL_MEMLIMIT32; mask = PCIM_BAR_MEM_BASE; lobits = PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_32; break; default: printf("pci_emul_alloc_base: invalid bar type %d\n", type); assert(0); } if (baseptr != NULL) { error = pci_emul_alloc_resource(baseptr, limit, size, &addr); if (error != 0) return (error); } pdi->pi_bar[idx].type = type; pdi->pi_bar[idx].addr = addr; pdi->pi_bar[idx].size = size; /* Initialize the BAR register in config space */ bar = (addr & mask) | lobits; pci_set_cfgdata32(pdi, PCIR_BAR(idx), bar); if (type == PCIBAR_MEM64) { assert(idx + 1 <= PCI_BARMAX); pdi->pi_bar[idx + 1].type = PCIBAR_MEMHI64; pci_set_cfgdata32(pdi, PCIR_BAR(idx + 1), bar >> 32); } register_bar(pdi, idx); return (0); } #define CAP_START_OFFSET 0x40 static int pci_emul_add_capability(struct pci_devinst *pi, u_char *capdata, int caplen) { int i, capoff, reallen; uint16_t sts; assert(caplen > 0); reallen = roundup2(caplen, 4); /* dword aligned */ sts = pci_get_cfgdata16(pi, PCIR_STATUS); if ((sts & PCIM_STATUS_CAPPRESENT) == 0) capoff = CAP_START_OFFSET; else capoff = pi->pi_capend + 1; /* Check if we have enough space */ if (capoff + reallen > PCI_REGMAX + 1) return (-1); /* Set the previous capability pointer */ if ((sts & PCIM_STATUS_CAPPRESENT) == 0) { pci_set_cfgdata8(pi, PCIR_CAP_PTR, capoff); pci_set_cfgdata16(pi, PCIR_STATUS, sts|PCIM_STATUS_CAPPRESENT); } else pci_set_cfgdata8(pi, pi->pi_prevcap + 1, capoff); /* Copy the capability */ for (i = 0; i < caplen; i++) pci_set_cfgdata8(pi, capoff + i, capdata[i]); /* Set the next capability pointer */ pci_set_cfgdata8(pi, capoff + 1, 0); pi->pi_prevcap = capoff; pi->pi_capend = capoff + reallen - 1; return (0); } static struct pci_devemu * pci_emul_finddev(char *name) { struct pci_devemu **pdpp, *pdp; SET_FOREACH(pdpp, pci_devemu_set) { pdp = *pdpp; if (!strcmp(pdp->pe_emu, name)) { return (pdp); } } return (NULL); } static int pci_emul_init(struct vmctx *ctx, struct pci_devemu *pde, int bus, int slot, int func, struct funcinfo *fi) { struct pci_devinst *pdi; int err; pdi = calloc(1, sizeof(struct pci_devinst)); pdi->pi_vmctx = ctx; pdi->pi_bus = bus; pdi->pi_slot = slot; pdi->pi_func = func; pthread_mutex_init(&pdi->pi_lintr.lock, NULL); pdi->pi_lintr.pin = 0; pdi->pi_lintr.state = IDLE; pdi->pi_lintr.pirq_pin = 0; pdi->pi_lintr.ioapic_irq = 0; pdi->pi_d = pde; snprintf(pdi->pi_name, PI_NAMESZ, "%s-pci-%d", pde->pe_emu, slot); /* Disable legacy interrupts */ pci_set_cfgdata8(pdi, PCIR_INTLINE, 255); pci_set_cfgdata8(pdi, PCIR_INTPIN, 0); pci_set_cfgdata8(pdi, PCIR_COMMAND, PCIM_CMD_PORTEN | PCIM_CMD_MEMEN | PCIM_CMD_BUSMASTEREN); err = (*pde->pe_init)(ctx, pdi, fi->fi_param); if (err == 0) fi->fi_devi = pdi; else free(pdi); return (err); } void pci_populate_msicap(struct msicap *msicap, int msgnum, int nextptr) { int mmc; CTASSERT(sizeof(struct msicap) == 14); /* Number of msi messages must be a power of 2 between 1 and 32 */ assert((msgnum & (msgnum - 1)) == 0 && msgnum >= 1 && msgnum <= 32); mmc = ffs(msgnum) - 1; bzero(msicap, sizeof(struct msicap)); msicap->capid = PCIY_MSI; msicap->nextptr = nextptr; msicap->msgctrl = PCIM_MSICTRL_64BIT | (mmc << 1); } int pci_emul_add_msicap(struct pci_devinst *pi, int msgnum) { struct msicap msicap; pci_populate_msicap(&msicap, msgnum, 0); return (pci_emul_add_capability(pi, (u_char *)&msicap, sizeof(msicap))); } static void pci_populate_msixcap(struct msixcap *msixcap, int msgnum, int barnum, uint32_t msix_tab_size) { CTASSERT(sizeof(struct msixcap) == 12); assert(msix_tab_size % 4096 == 0); bzero(msixcap, sizeof(struct msixcap)); msixcap->capid = PCIY_MSIX; /* * Message Control Register, all fields set to * zero except for the Table Size. * Note: Table size N is encoded as N-1 */ msixcap->msgctrl = msgnum - 1; /* * MSI-X BAR setup: * - MSI-X table start at offset 0 * - PBA table starts at a 4K aligned offset after the MSI-X table */ msixcap->table_info = barnum & PCIM_MSIX_BIR_MASK; msixcap->pba_info = msix_tab_size | (barnum & PCIM_MSIX_BIR_MASK); } static void pci_msix_table_init(struct pci_devinst *pi, int table_entries) { int i, table_size; assert(table_entries > 0); assert(table_entries <= MAX_MSIX_TABLE_ENTRIES); table_size = table_entries * MSIX_TABLE_ENTRY_SIZE; pi->pi_msix.table = calloc(1, table_size); /* set mask bit of vector control register */ for (i = 0; i < table_entries; i++) pi->pi_msix.table[i].vector_control |= PCIM_MSIX_VCTRL_MASK; } int pci_emul_add_msixcap(struct pci_devinst *pi, int msgnum, int barnum) { uint32_t tab_size; struct msixcap msixcap; assert(msgnum >= 1 && msgnum <= MAX_MSIX_TABLE_ENTRIES); assert(barnum >= 0 && barnum <= PCIR_MAX_BAR_0); tab_size = msgnum * MSIX_TABLE_ENTRY_SIZE; /* Align table size to nearest 4K */ tab_size = roundup2(tab_size, 4096); pi->pi_msix.table_bar = barnum; pi->pi_msix.pba_bar = barnum; pi->pi_msix.table_offset = 0; pi->pi_msix.table_count = msgnum; pi->pi_msix.pba_offset = tab_size; pi->pi_msix.pba_size = PBA_SIZE(msgnum); pci_msix_table_init(pi, msgnum); pci_populate_msixcap(&msixcap, msgnum, barnum, tab_size); /* allocate memory for MSI-X Table and PBA */ pci_emul_alloc_bar(pi, barnum, PCIBAR_MEM32, tab_size + pi->pi_msix.pba_size); return (pci_emul_add_capability(pi, (u_char *)&msixcap, sizeof(msixcap))); } void msixcap_cfgwrite(struct pci_devinst *pi, int capoff, int offset, int bytes, uint32_t val) { uint16_t msgctrl, rwmask; int off, table_bar; off = offset - capoff; table_bar = pi->pi_msix.table_bar; /* Message Control Register */ if (off == 2 && bytes == 2) { rwmask = PCIM_MSIXCTRL_MSIX_ENABLE | PCIM_MSIXCTRL_FUNCTION_MASK; msgctrl = pci_get_cfgdata16(pi, offset); msgctrl &= ~rwmask; msgctrl |= val & rwmask; val = msgctrl; pi->pi_msix.enabled = val & PCIM_MSIXCTRL_MSIX_ENABLE; pi->pi_msix.function_mask = val & PCIM_MSIXCTRL_FUNCTION_MASK; pci_lintr_update(pi); } CFGWRITE(pi, offset, val, bytes); } void msicap_cfgwrite(struct pci_devinst *pi, int capoff, int offset, int bytes, uint32_t val) { uint16_t msgctrl, rwmask, msgdata, mme; uint32_t addrlo; /* * If guest is writing to the message control register make sure * we do not overwrite read-only fields. */ if ((offset - capoff) == 2 && bytes == 2) { rwmask = PCIM_MSICTRL_MME_MASK | PCIM_MSICTRL_MSI_ENABLE; msgctrl = pci_get_cfgdata16(pi, offset); msgctrl &= ~rwmask; msgctrl |= val & rwmask; val = msgctrl; addrlo = pci_get_cfgdata32(pi, capoff + 4); if (msgctrl & PCIM_MSICTRL_64BIT) msgdata = pci_get_cfgdata16(pi, capoff + 12); else msgdata = pci_get_cfgdata16(pi, capoff + 8); mme = msgctrl & PCIM_MSICTRL_MME_MASK; pi->pi_msi.enabled = msgctrl & PCIM_MSICTRL_MSI_ENABLE ? 1 : 0; if (pi->pi_msi.enabled) { pi->pi_msi.addr = addrlo; pi->pi_msi.msg_data = msgdata; pi->pi_msi.maxmsgnum = 1 << (mme >> 4); } else { pi->pi_msi.maxmsgnum = 0; } pci_lintr_update(pi); } CFGWRITE(pi, offset, val, bytes); } void pciecap_cfgwrite(struct pci_devinst *pi, int capoff, int offset, int bytes, uint32_t val) { /* XXX don't write to the readonly parts */ CFGWRITE(pi, offset, val, bytes); } #define PCIECAP_VERSION 0x2 int pci_emul_add_pciecap(struct pci_devinst *pi, int type) { int err; struct pciecap pciecap; CTASSERT(sizeof(struct pciecap) == 60); if (type != PCIEM_TYPE_ROOT_PORT) return (-1); bzero(&pciecap, sizeof(pciecap)); pciecap.capid = PCIY_EXPRESS; pciecap.pcie_capabilities = PCIECAP_VERSION | PCIEM_TYPE_ROOT_PORT; pciecap.link_capabilities = 0x411; /* gen1, x1 */ pciecap.link_status = 0x11; /* gen1, x1 */ err = pci_emul_add_capability(pi, (u_char *)&pciecap, sizeof(pciecap)); return (err); } /* * This function assumes that 'coff' is in the capabilities region of the * config space. */ static void pci_emul_capwrite(struct pci_devinst *pi, int offset, int bytes, uint32_t val) { int capid; uint8_t capoff, nextoff; /* Do not allow un-aligned writes */ if ((offset & (bytes - 1)) != 0) return; /* Find the capability that we want to update */ capoff = CAP_START_OFFSET; while (1) { nextoff = pci_get_cfgdata8(pi, capoff + 1); if (nextoff == 0) break; if (offset >= capoff && offset < nextoff) break; capoff = nextoff; } assert(offset >= capoff); /* * Capability ID and Next Capability Pointer are readonly. * However, some o/s's do 4-byte writes that include these. * For this case, trim the write back to 2 bytes and adjust * the data. */ if (offset == capoff || offset == capoff + 1) { if (offset == capoff && bytes == 4) { bytes = 2; offset += 2; val >>= 16; } else return; } capid = pci_get_cfgdata8(pi, capoff); switch (capid) { case PCIY_MSI: msicap_cfgwrite(pi, capoff, offset, bytes, val); break; case PCIY_MSIX: msixcap_cfgwrite(pi, capoff, offset, bytes, val); break; case PCIY_EXPRESS: pciecap_cfgwrite(pi, capoff, offset, bytes, val); break; default: break; } } static int pci_emul_iscap(struct pci_devinst *pi, int offset) { uint16_t sts; sts = pci_get_cfgdata16(pi, PCIR_STATUS); if ((sts & PCIM_STATUS_CAPPRESENT) != 0) { if (offset >= CAP_START_OFFSET && offset <= pi->pi_capend) return (1); } return (0); } static int pci_emul_fallback_handler(struct vmctx *ctx, int vcpu, int dir, uint64_t addr, int size, uint64_t *val, void *arg1, long arg2) { /* * Ignore writes; return 0xff's for reads. The mem read code * will take care of truncating to the correct size. */ if (dir == MEM_F_READ) { *val = 0xffffffffffffffff; } return (0); } #define BUSIO_ROUNDUP 32 #define BUSMEM_ROUNDUP (1024 * 1024) int init_pci(struct vmctx *ctx) { struct pci_devemu *pde; struct businfo *bi; struct slotinfo *si; struct funcinfo *fi; size_t lowmem; int bus, slot, func; int error; pci_emul_iobase = PCI_EMUL_IOBASE; pci_emul_membase32 = vm_get_lowmem_limit(ctx); pci_emul_membase64 = PCI_EMUL_MEMBASE64; for (bus = 0; bus < MAXBUSES; bus++) { if ((bi = pci_businfo[bus]) == NULL) continue; /* * Keep track of the i/o and memory resources allocated to * this bus. */ bi->iobase = pci_emul_iobase; bi->membase32 = pci_emul_membase32; bi->membase64 = pci_emul_membase64; for (slot = 0; slot < MAXSLOTS; slot++) { si = &bi->slotinfo[slot]; for (func = 0; func < MAXFUNCS; func++) { fi = &si->si_funcs[func]; if (fi->fi_name == NULL) continue; pde = pci_emul_finddev(fi->fi_name); assert(pde != NULL); error = pci_emul_init(ctx, pde, bus, slot, func, fi); if (error) return (error); } } /* * Add some slop to the I/O and memory resources decoded by * this bus to give a guest some flexibility if it wants to * reprogram the BARs. */ pci_emul_iobase += BUSIO_ROUNDUP; pci_emul_iobase = roundup2(pci_emul_iobase, BUSIO_ROUNDUP); bi->iolimit = pci_emul_iobase; pci_emul_membase32 += BUSMEM_ROUNDUP; pci_emul_membase32 = roundup2(pci_emul_membase32, BUSMEM_ROUNDUP); bi->memlimit32 = pci_emul_membase32; pci_emul_membase64 += BUSMEM_ROUNDUP; pci_emul_membase64 = roundup2(pci_emul_membase64, BUSMEM_ROUNDUP); bi->memlimit64 = pci_emul_membase64; } /* * PCI backends are initialized before routing INTx interrupts * so that LPC devices are able to reserve ISA IRQs before * routing PIRQ pins. */ for (bus = 0; bus < MAXBUSES; bus++) { if ((bi = pci_businfo[bus]) == NULL) continue; for (slot = 0; slot < MAXSLOTS; slot++) { si = &bi->slotinfo[slot]; for (func = 0; func < MAXFUNCS; func++) { fi = &si->si_funcs[func]; if (fi->fi_devi == NULL) continue; pci_lintr_route(fi->fi_devi); } } } lpc_pirq_routed(); /* * The guest physical memory map looks like the following: * [0, lowmem) guest system memory * [lowmem, lowmem_limit) memory hole (may be absent) * [lowmem_limit, 4GB) PCI hole (32-bit BAR allocation) * [4GB, 4GB + highmem) * * Accesses to memory addresses that are not allocated to system * memory or PCI devices return 0xff's. */ - error = vm_get_memory_seg(ctx, 0, &lowmem, NULL); - assert(error == 0); + lowmem = vm_get_lowmem_size(ctx); memset(&pci_mem_hole, 0, sizeof(struct mem_range)); pci_mem_hole.name = "PCI hole"; pci_mem_hole.flags = MEM_F_RW; pci_mem_hole.base = lowmem; pci_mem_hole.size = (4ULL * 1024 * 1024 * 1024) - lowmem; pci_mem_hole.handler = pci_emul_fallback_handler; error = register_mem_fallback(&pci_mem_hole); assert(error == 0); return (0); } static void pci_apic_prt_entry(int bus, int slot, int pin, int pirq_pin, int ioapic_irq, void *arg) { dsdt_line(" Package ()"); dsdt_line(" {"); dsdt_line(" 0x%X,", slot << 16 | 0xffff); dsdt_line(" 0x%02X,", pin - 1); dsdt_line(" Zero,"); dsdt_line(" 0x%X", ioapic_irq); dsdt_line(" },"); } static void pci_pirq_prt_entry(int bus, int slot, int pin, int pirq_pin, int ioapic_irq, void *arg) { char *name; name = lpc_pirq_name(pirq_pin); if (name == NULL) return; dsdt_line(" Package ()"); dsdt_line(" {"); dsdt_line(" 0x%X,", slot << 16 | 0xffff); dsdt_line(" 0x%02X,", pin - 1); dsdt_line(" %s,", name); dsdt_line(" 0x00"); dsdt_line(" },"); free(name); } /* * A bhyve virtual machine has a flat PCI hierarchy with a root port * corresponding to each PCI bus. */ static void pci_bus_write_dsdt(int bus) { struct businfo *bi; struct slotinfo *si; struct pci_devinst *pi; int count, func, slot; /* * If there are no devices on this 'bus' then just return. */ if ((bi = pci_businfo[bus]) == NULL) { /* * Bus 0 is special because it decodes the I/O ports used * for PCI config space access even if there are no devices * on it. */ if (bus != 0) return; } dsdt_line(" Device (PC%02X)", bus); dsdt_line(" {"); dsdt_line(" Name (_HID, EisaId (\"PNP0A03\"))"); dsdt_line(" Name (_ADR, Zero)"); dsdt_line(" Method (_BBN, 0, NotSerialized)"); dsdt_line(" {"); dsdt_line(" Return (0x%08X)", bus); dsdt_line(" }"); dsdt_line(" Name (_CRS, ResourceTemplate ()"); dsdt_line(" {"); dsdt_line(" WordBusNumber (ResourceProducer, MinFixed, " "MaxFixed, PosDecode,"); dsdt_line(" 0x0000, // Granularity"); dsdt_line(" 0x%04X, // Range Minimum", bus); dsdt_line(" 0x%04X, // Range Maximum", bus); dsdt_line(" 0x0000, // Translation Offset"); dsdt_line(" 0x0001, // Length"); dsdt_line(" ,, )"); if (bus == 0) { dsdt_indent(3); dsdt_fixed_ioport(0xCF8, 8); dsdt_unindent(3); dsdt_line(" WordIO (ResourceProducer, MinFixed, MaxFixed, " "PosDecode, EntireRange,"); dsdt_line(" 0x0000, // Granularity"); dsdt_line(" 0x0000, // Range Minimum"); dsdt_line(" 0x0CF7, // Range Maximum"); dsdt_line(" 0x0000, // Translation Offset"); dsdt_line(" 0x0CF8, // Length"); dsdt_line(" ,, , TypeStatic)"); dsdt_line(" WordIO (ResourceProducer, MinFixed, MaxFixed, " "PosDecode, EntireRange,"); dsdt_line(" 0x0000, // Granularity"); dsdt_line(" 0x0D00, // Range Minimum"); dsdt_line(" 0x%04X, // Range Maximum", PCI_EMUL_IOBASE - 1); dsdt_line(" 0x0000, // Translation Offset"); dsdt_line(" 0x%04X, // Length", PCI_EMUL_IOBASE - 0x0D00); dsdt_line(" ,, , TypeStatic)"); if (bi == NULL) { dsdt_line(" })"); goto done; } } assert(bi != NULL); /* i/o window */ dsdt_line(" WordIO (ResourceProducer, MinFixed, MaxFixed, " "PosDecode, EntireRange,"); dsdt_line(" 0x0000, // Granularity"); dsdt_line(" 0x%04X, // Range Minimum", bi->iobase); dsdt_line(" 0x%04X, // Range Maximum", bi->iolimit - 1); dsdt_line(" 0x0000, // Translation Offset"); dsdt_line(" 0x%04X, // Length", bi->iolimit - bi->iobase); dsdt_line(" ,, , TypeStatic)"); /* mmio window (32-bit) */ dsdt_line(" DWordMemory (ResourceProducer, PosDecode, " "MinFixed, MaxFixed, NonCacheable, ReadWrite,"); dsdt_line(" 0x00000000, // Granularity"); dsdt_line(" 0x%08X, // Range Minimum\n", bi->membase32); dsdt_line(" 0x%08X, // Range Maximum\n", bi->memlimit32 - 1); dsdt_line(" 0x00000000, // Translation Offset"); dsdt_line(" 0x%08X, // Length\n", bi->memlimit32 - bi->membase32); dsdt_line(" ,, , AddressRangeMemory, TypeStatic)"); /* mmio window (64-bit) */ dsdt_line(" QWordMemory (ResourceProducer, PosDecode, " "MinFixed, MaxFixed, NonCacheable, ReadWrite,"); dsdt_line(" 0x0000000000000000, // Granularity"); dsdt_line(" 0x%016lX, // Range Minimum\n", bi->membase64); dsdt_line(" 0x%016lX, // Range Maximum\n", bi->memlimit64 - 1); dsdt_line(" 0x0000000000000000, // Translation Offset"); dsdt_line(" 0x%016lX, // Length\n", bi->memlimit64 - bi->membase64); dsdt_line(" ,, , AddressRangeMemory, TypeStatic)"); dsdt_line(" })"); count = pci_count_lintr(bus); if (count != 0) { dsdt_indent(2); dsdt_line("Name (PPRT, Package ()"); dsdt_line("{"); pci_walk_lintr(bus, pci_pirq_prt_entry, NULL); dsdt_line("})"); dsdt_line("Name (APRT, Package ()"); dsdt_line("{"); pci_walk_lintr(bus, pci_apic_prt_entry, NULL); dsdt_line("})"); dsdt_line("Method (_PRT, 0, NotSerialized)"); dsdt_line("{"); dsdt_line(" If (PICM)"); dsdt_line(" {"); dsdt_line(" Return (APRT)"); dsdt_line(" }"); dsdt_line(" Else"); dsdt_line(" {"); dsdt_line(" Return (PPRT)"); dsdt_line(" }"); dsdt_line("}"); dsdt_unindent(2); } dsdt_indent(2); for (slot = 0; slot < MAXSLOTS; slot++) { si = &bi->slotinfo[slot]; for (func = 0; func < MAXFUNCS; func++) { pi = si->si_funcs[func].fi_devi; if (pi != NULL && pi->pi_d->pe_write_dsdt != NULL) pi->pi_d->pe_write_dsdt(pi); } } dsdt_unindent(2); done: dsdt_line(" }"); } void pci_write_dsdt(void) { int bus; dsdt_indent(1); dsdt_line("Name (PICM, 0x00)"); dsdt_line("Method (_PIC, 1, NotSerialized)"); dsdt_line("{"); dsdt_line(" Store (Arg0, PICM)"); dsdt_line("}"); dsdt_line(""); dsdt_line("Scope (_SB)"); dsdt_line("{"); for (bus = 0; bus < MAXBUSES; bus++) pci_bus_write_dsdt(bus); dsdt_line("}"); dsdt_unindent(1); } int pci_bus_configured(int bus) { assert(bus >= 0 && bus < MAXBUSES); return (pci_businfo[bus] != NULL); } int pci_msi_enabled(struct pci_devinst *pi) { return (pi->pi_msi.enabled); } int pci_msi_maxmsgnum(struct pci_devinst *pi) { if (pi->pi_msi.enabled) return (pi->pi_msi.maxmsgnum); else return (0); } int pci_msix_enabled(struct pci_devinst *pi) { return (pi->pi_msix.enabled && !pi->pi_msi.enabled); } void pci_generate_msix(struct pci_devinst *pi, int index) { struct msix_table_entry *mte; if (!pci_msix_enabled(pi)) return; if (pi->pi_msix.function_mask) return; if (index >= pi->pi_msix.table_count) return; mte = &pi->pi_msix.table[index]; if ((mte->vector_control & PCIM_MSIX_VCTRL_MASK) == 0) { /* XXX Set PBA bit if interrupt is disabled */ vm_lapic_msi(pi->pi_vmctx, mte->addr, mte->msg_data); } } void pci_generate_msi(struct pci_devinst *pi, int index) { if (pci_msi_enabled(pi) && index < pci_msi_maxmsgnum(pi)) { vm_lapic_msi(pi->pi_vmctx, pi->pi_msi.addr, pi->pi_msi.msg_data + index); } } static bool pci_lintr_permitted(struct pci_devinst *pi) { uint16_t cmd; cmd = pci_get_cfgdata16(pi, PCIR_COMMAND); return (!(pi->pi_msi.enabled || pi->pi_msix.enabled || (cmd & PCIM_CMD_INTxDIS))); } void pci_lintr_request(struct pci_devinst *pi) { struct businfo *bi; struct slotinfo *si; int bestpin, bestcount, pin; bi = pci_businfo[pi->pi_bus]; assert(bi != NULL); /* * Just allocate a pin from our slot. The pin will be * assigned IRQs later when interrupts are routed. */ si = &bi->slotinfo[pi->pi_slot]; bestpin = 0; bestcount = si->si_intpins[0].ii_count; for (pin = 1; pin < 4; pin++) { if (si->si_intpins[pin].ii_count < bestcount) { bestpin = pin; bestcount = si->si_intpins[pin].ii_count; } } si->si_intpins[bestpin].ii_count++; pi->pi_lintr.pin = bestpin + 1; pci_set_cfgdata8(pi, PCIR_INTPIN, bestpin + 1); } static void pci_lintr_route(struct pci_devinst *pi) { struct businfo *bi; struct intxinfo *ii; if (pi->pi_lintr.pin == 0) return; bi = pci_businfo[pi->pi_bus]; assert(bi != NULL); ii = &bi->slotinfo[pi->pi_slot].si_intpins[pi->pi_lintr.pin - 1]; /* * Attempt to allocate an I/O APIC pin for this intpin if one * is not yet assigned. */ if (ii->ii_ioapic_irq == 0) ii->ii_ioapic_irq = ioapic_pci_alloc_irq(); assert(ii->ii_ioapic_irq > 0); /* * Attempt to allocate a PIRQ pin for this intpin if one is * not yet assigned. */ if (ii->ii_pirq_pin == 0) ii->ii_pirq_pin = pirq_alloc_pin(pi->pi_vmctx); assert(ii->ii_pirq_pin > 0); pi->pi_lintr.ioapic_irq = ii->ii_ioapic_irq; pi->pi_lintr.pirq_pin = ii->ii_pirq_pin; pci_set_cfgdata8(pi, PCIR_INTLINE, pirq_irq(ii->ii_pirq_pin)); } void pci_lintr_assert(struct pci_devinst *pi) { assert(pi->pi_lintr.pin > 0); pthread_mutex_lock(&pi->pi_lintr.lock); if (pi->pi_lintr.state == IDLE) { if (pci_lintr_permitted(pi)) { pi->pi_lintr.state = ASSERTED; pci_irq_assert(pi); } else pi->pi_lintr.state = PENDING; } pthread_mutex_unlock(&pi->pi_lintr.lock); } void pci_lintr_deassert(struct pci_devinst *pi) { assert(pi->pi_lintr.pin > 0); pthread_mutex_lock(&pi->pi_lintr.lock); if (pi->pi_lintr.state == ASSERTED) { pi->pi_lintr.state = IDLE; pci_irq_deassert(pi); } else if (pi->pi_lintr.state == PENDING) pi->pi_lintr.state = IDLE; pthread_mutex_unlock(&pi->pi_lintr.lock); } static void pci_lintr_update(struct pci_devinst *pi) { pthread_mutex_lock(&pi->pi_lintr.lock); if (pi->pi_lintr.state == ASSERTED && !pci_lintr_permitted(pi)) { pci_irq_deassert(pi); pi->pi_lintr.state = PENDING; } else if (pi->pi_lintr.state == PENDING && pci_lintr_permitted(pi)) { pi->pi_lintr.state = ASSERTED; pci_irq_assert(pi); } pthread_mutex_unlock(&pi->pi_lintr.lock); } int pci_count_lintr(int bus) { int count, slot, pin; struct slotinfo *slotinfo; count = 0; if (pci_businfo[bus] != NULL) { for (slot = 0; slot < MAXSLOTS; slot++) { slotinfo = &pci_businfo[bus]->slotinfo[slot]; for (pin = 0; pin < 4; pin++) { if (slotinfo->si_intpins[pin].ii_count != 0) count++; } } } return (count); } void pci_walk_lintr(int bus, pci_lintr_cb cb, void *arg) { struct businfo *bi; struct slotinfo *si; struct intxinfo *ii; int slot, pin; if ((bi = pci_businfo[bus]) == NULL) return; for (slot = 0; slot < MAXSLOTS; slot++) { si = &bi->slotinfo[slot]; for (pin = 0; pin < 4; pin++) { ii = &si->si_intpins[pin]; if (ii->ii_count != 0) cb(bus, slot, pin + 1, ii->ii_pirq_pin, ii->ii_ioapic_irq, arg); } } } /* * Return 1 if the emulated device in 'slot' is a multi-function device. * Return 0 otherwise. */ static int pci_emul_is_mfdev(int bus, int slot) { struct businfo *bi; struct slotinfo *si; int f, numfuncs; numfuncs = 0; if ((bi = pci_businfo[bus]) != NULL) { si = &bi->slotinfo[slot]; for (f = 0; f < MAXFUNCS; f++) { if (si->si_funcs[f].fi_devi != NULL) { numfuncs++; } } } return (numfuncs > 1); } /* * Ensure that the PCIM_MFDEV bit is properly set (or unset) depending on * whether or not is a multi-function being emulated in the pci 'slot'. */ static void pci_emul_hdrtype_fixup(int bus, int slot, int off, int bytes, uint32_t *rv) { int mfdev; if (off <= PCIR_HDRTYPE && off + bytes > PCIR_HDRTYPE) { mfdev = pci_emul_is_mfdev(bus, slot); switch (bytes) { case 1: case 2: *rv &= ~PCIM_MFDEV; if (mfdev) { *rv |= PCIM_MFDEV; } break; case 4: *rv &= ~(PCIM_MFDEV << 16); if (mfdev) { *rv |= (PCIM_MFDEV << 16); } break; } } } static int cfgenable, cfgbus, cfgslot, cfgfunc, cfgoff; static int pci_emul_cfgaddr(struct vmctx *ctx, int vcpu, int in, int port, int bytes, uint32_t *eax, void *arg) { uint32_t x; if (bytes != 4) { if (in) *eax = (bytes == 2) ? 0xffff : 0xff; return (0); } if (in) { x = (cfgbus << 16) | (cfgslot << 11) | (cfgfunc << 8) | cfgoff; if (cfgenable) x |= CONF1_ENABLE; *eax = x; } else { x = *eax; cfgenable = (x & CONF1_ENABLE) == CONF1_ENABLE; cfgoff = x & PCI_REGMAX; cfgfunc = (x >> 8) & PCI_FUNCMAX; cfgslot = (x >> 11) & PCI_SLOTMAX; cfgbus = (x >> 16) & PCI_BUSMAX; } return (0); } INOUT_PORT(pci_cfgaddr, CONF1_ADDR_PORT, IOPORT_F_INOUT, pci_emul_cfgaddr); static uint32_t bits_changed(uint32_t old, uint32_t new, uint32_t mask) { return ((old ^ new) & mask); } static void pci_emul_cmdwrite(struct pci_devinst *pi, uint32_t new, int bytes) { int i; uint16_t old; /* * The command register is at an offset of 4 bytes and thus the * guest could write 1, 2 or 4 bytes starting at this offset. */ old = pci_get_cfgdata16(pi, PCIR_COMMAND); /* stash old value */ CFGWRITE(pi, PCIR_COMMAND, new, bytes); /* update config */ new = pci_get_cfgdata16(pi, PCIR_COMMAND); /* get updated value */ /* * If the MMIO or I/O address space decoding has changed then * register/unregister all BARs that decode that address space. */ for (i = 0; i <= PCI_BARMAX; i++) { switch (pi->pi_bar[i].type) { case PCIBAR_NONE: case PCIBAR_MEMHI64: break; case PCIBAR_IO: /* I/O address space decoding changed? */ if (bits_changed(old, new, PCIM_CMD_PORTEN)) { if (porten(pi)) register_bar(pi, i); else unregister_bar(pi, i); } break; case PCIBAR_MEM32: case PCIBAR_MEM64: /* MMIO address space decoding changed? */ if (bits_changed(old, new, PCIM_CMD_MEMEN)) { if (memen(pi)) register_bar(pi, i); else unregister_bar(pi, i); } break; default: assert(0); } } /* * If INTx has been unmasked and is pending, assert the * interrupt. */ pci_lintr_update(pi); } static int pci_emul_cfgdata(struct vmctx *ctx, int vcpu, int in, int port, int bytes, uint32_t *eax, void *arg) { struct businfo *bi; struct slotinfo *si; struct pci_devinst *pi; struct pci_devemu *pe; int coff, idx, needcfg; uint64_t addr, bar, mask; assert(bytes == 1 || bytes == 2 || bytes == 4); if ((bi = pci_businfo[cfgbus]) != NULL) { si = &bi->slotinfo[cfgslot]; pi = si->si_funcs[cfgfunc].fi_devi; } else pi = NULL; coff = cfgoff + (port - CONF1_DATA_PORT); #if 0 printf("pcicfg-%s from 0x%0x of %d bytes (%d/%d/%d)\n\r", in ? "read" : "write", coff, bytes, cfgbus, cfgslot, cfgfunc); #endif /* * Just return if there is no device at this cfgslot:cfgfunc, * if the guest is doing an un-aligned access, or if the config * address word isn't enabled. */ if (!cfgenable || pi == NULL || (coff & (bytes - 1)) != 0) { if (in) *eax = 0xffffffff; return (0); } pe = pi->pi_d; /* * Config read */ if (in) { /* Let the device emulation override the default handler */ if (pe->pe_cfgread != NULL) { needcfg = pe->pe_cfgread(ctx, vcpu, pi, coff, bytes, eax); } else { needcfg = 1; } if (needcfg) { if (bytes == 1) *eax = pci_get_cfgdata8(pi, coff); else if (bytes == 2) *eax = pci_get_cfgdata16(pi, coff); else *eax = pci_get_cfgdata32(pi, coff); } pci_emul_hdrtype_fixup(cfgbus, cfgslot, coff, bytes, eax); } else { /* Let the device emulation override the default handler */ if (pe->pe_cfgwrite != NULL && (*pe->pe_cfgwrite)(ctx, vcpu, pi, coff, bytes, *eax) == 0) return (0); /* * Special handling for write to BAR registers */ if (coff >= PCIR_BAR(0) && coff < PCIR_BAR(PCI_BARMAX + 1)) { /* * Ignore writes to BAR registers that are not * 4-byte aligned. */ if (bytes != 4 || (coff & 0x3) != 0) return (0); idx = (coff - PCIR_BAR(0)) / 4; mask = ~(pi->pi_bar[idx].size - 1); switch (pi->pi_bar[idx].type) { case PCIBAR_NONE: pi->pi_bar[idx].addr = bar = 0; break; case PCIBAR_IO: addr = *eax & mask; addr &= 0xffff; bar = addr | PCIM_BAR_IO_SPACE; /* * Register the new BAR value for interception */ if (addr != pi->pi_bar[idx].addr) { update_bar_address(pi, addr, idx, PCIBAR_IO); } break; case PCIBAR_MEM32: addr = bar = *eax & mask; bar |= PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_32; if (addr != pi->pi_bar[idx].addr) { update_bar_address(pi, addr, idx, PCIBAR_MEM32); } break; case PCIBAR_MEM64: addr = bar = *eax & mask; bar |= PCIM_BAR_MEM_SPACE | PCIM_BAR_MEM_64 | PCIM_BAR_MEM_PREFETCH; if (addr != (uint32_t)pi->pi_bar[idx].addr) { update_bar_address(pi, addr, idx, PCIBAR_MEM64); } break; case PCIBAR_MEMHI64: mask = ~(pi->pi_bar[idx - 1].size - 1); addr = ((uint64_t)*eax << 32) & mask; bar = addr >> 32; if (bar != pi->pi_bar[idx - 1].addr >> 32) { update_bar_address(pi, addr, idx - 1, PCIBAR_MEMHI64); } break; default: assert(0); } pci_set_cfgdata32(pi, coff, bar); } else if (pci_emul_iscap(pi, coff)) { pci_emul_capwrite(pi, coff, bytes, *eax); } else if (coff == PCIR_COMMAND) { pci_emul_cmdwrite(pi, *eax, bytes); } else { CFGWRITE(pi, coff, *eax, bytes); } } return (0); } INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+0, IOPORT_F_INOUT, pci_emul_cfgdata); INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+1, IOPORT_F_INOUT, pci_emul_cfgdata); INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+2, IOPORT_F_INOUT, pci_emul_cfgdata); INOUT_PORT(pci_cfgdata, CONF1_DATA_PORT+3, IOPORT_F_INOUT, pci_emul_cfgdata); #define PCI_EMUL_TEST #ifdef PCI_EMUL_TEST /* * Define a dummy test device */ #define DIOSZ 8 #define DMEMSZ 4096 struct pci_emul_dsoftc { uint8_t ioregs[DIOSZ]; uint8_t memregs[DMEMSZ]; }; #define PCI_EMUL_MSI_MSGS 4 #define PCI_EMUL_MSIX_MSGS 16 static int pci_emul_dinit(struct vmctx *ctx, struct pci_devinst *pi, char *opts) { int error; struct pci_emul_dsoftc *sc; sc = calloc(1, sizeof(struct pci_emul_dsoftc)); pi->pi_arg = sc; pci_set_cfgdata16(pi, PCIR_DEVICE, 0x0001); pci_set_cfgdata16(pi, PCIR_VENDOR, 0x10DD); pci_set_cfgdata8(pi, PCIR_CLASS, 0x02); error = pci_emul_add_msicap(pi, PCI_EMUL_MSI_MSGS); assert(error == 0); error = pci_emul_alloc_bar(pi, 0, PCIBAR_IO, DIOSZ); assert(error == 0); error = pci_emul_alloc_bar(pi, 1, PCIBAR_MEM32, DMEMSZ); assert(error == 0); return (0); } static void pci_emul_diow(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx, uint64_t offset, int size, uint64_t value) { int i; struct pci_emul_dsoftc *sc = pi->pi_arg; if (baridx == 0) { if (offset + size > DIOSZ) { printf("diow: iow too large, offset %ld size %d\n", offset, size); return; } if (size == 1) { sc->ioregs[offset] = value & 0xff; } else if (size == 2) { *(uint16_t *)&sc->ioregs[offset] = value & 0xffff; } else if (size == 4) { *(uint32_t *)&sc->ioregs[offset] = value; } else { printf("diow: iow unknown size %d\n", size); } /* * Special magic value to generate an interrupt */ if (offset == 4 && size == 4 && pci_msi_enabled(pi)) pci_generate_msi(pi, value % pci_msi_maxmsgnum(pi)); if (value == 0xabcdef) { for (i = 0; i < pci_msi_maxmsgnum(pi); i++) pci_generate_msi(pi, i); } } if (baridx == 1) { if (offset + size > DMEMSZ) { printf("diow: memw too large, offset %ld size %d\n", offset, size); return; } if (size == 1) { sc->memregs[offset] = value; } else if (size == 2) { *(uint16_t *)&sc->memregs[offset] = value; } else if (size == 4) { *(uint32_t *)&sc->memregs[offset] = value; } else if (size == 8) { *(uint64_t *)&sc->memregs[offset] = value; } else { printf("diow: memw unknown size %d\n", size); } /* * magic interrupt ?? */ } if (baridx > 1) { printf("diow: unknown bar idx %d\n", baridx); } } static uint64_t pci_emul_dior(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx, uint64_t offset, int size) { struct pci_emul_dsoftc *sc = pi->pi_arg; uint32_t value; if (baridx == 0) { if (offset + size > DIOSZ) { printf("dior: ior too large, offset %ld size %d\n", offset, size); return (0); } if (size == 1) { value = sc->ioregs[offset]; } else if (size == 2) { value = *(uint16_t *) &sc->ioregs[offset]; } else if (size == 4) { value = *(uint32_t *) &sc->ioregs[offset]; } else { printf("dior: ior unknown size %d\n", size); } } if (baridx == 1) { if (offset + size > DMEMSZ) { printf("dior: memr too large, offset %ld size %d\n", offset, size); return (0); } if (size == 1) { value = sc->memregs[offset]; } else if (size == 2) { value = *(uint16_t *) &sc->memregs[offset]; } else if (size == 4) { value = *(uint32_t *) &sc->memregs[offset]; } else if (size == 8) { value = *(uint64_t *) &sc->memregs[offset]; } else { printf("dior: ior unknown size %d\n", size); } } if (baridx > 1) { printf("dior: unknown bar idx %d\n", baridx); return (0); } return (value); } struct pci_devemu pci_dummy = { .pe_emu = "dummy", .pe_init = pci_emul_dinit, .pe_barwrite = pci_emul_diow, .pe_barread = pci_emul_dior }; PCI_EMUL_SET(pci_dummy); #endif /* PCI_EMUL_TEST */ Index: stable/10/usr.sbin/bhyve/rtc.c =================================================================== --- stable/10/usr.sbin/bhyve/rtc.c (revision 270073) +++ stable/10/usr.sbin/bhyve/rtc.c (revision 270074) @@ -1,383 +1,378 @@ /*- * Copyright (c) 2011 NetApp, Inc. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include "acpi.h" #include "inout.h" #include "pci_lpc.h" #include "rtc.h" #define IO_RTC 0x70 #define RTC_SEC 0x00 /* seconds */ #define RTC_SEC_ALARM 0x01 #define RTC_MIN 0x02 #define RTC_MIN_ALARM 0x03 #define RTC_HRS 0x04 #define RTC_HRS_ALARM 0x05 #define RTC_WDAY 0x06 #define RTC_DAY 0x07 #define RTC_MONTH 0x08 #define RTC_YEAR 0x09 #define RTC_CENTURY 0x32 /* current century */ #define RTC_STATUSA 0xA #define RTCSA_TUP 0x80 /* time update, don't look now */ #define RTC_STATUSB 0xB #define RTCSB_DST 0x01 #define RTCSB_24HR 0x02 #define RTCSB_BIN 0x04 /* 0 = BCD, 1 = Binary */ #define RTCSB_PINTR 0x40 /* 1 = enable periodic clock interrupt */ #define RTCSB_HALT 0x80 /* stop clock updates */ #define RTC_INTR 0x0c /* status register C (R) interrupt source */ #define RTC_STATUSD 0x0d /* status register D (R) Lost Power */ #define RTCSD_PWR 0x80 /* clock power OK */ #define RTC_NVRAM_START 0x0e #define RTC_NVRAM_END 0x7f #define RTC_NVRAM_SZ (128 - RTC_NVRAM_START) #define nvoff(x) ((x) - RTC_NVRAM_START) #define RTC_DIAG 0x0e #define RTC_RSTCODE 0x0f #define RTC_EQUIPMENT 0x14 #define RTC_LMEM_LSB 0x34 #define RTC_LMEM_MSB 0x35 #define RTC_HMEM_LSB 0x5b #define RTC_HMEM_SB 0x5c #define RTC_HMEM_MSB 0x5d #define m_64KB (64*1024) #define m_16MB (16*1024*1024) #define m_4GB (4ULL*1024*1024*1024) static int addr; static uint8_t rtc_nvram[RTC_NVRAM_SZ]; /* XXX initialize these to default values as they would be from BIOS */ static uint8_t status_a, status_b; static struct { uint8_t hours; uint8_t mins; uint8_t secs; } rtc_alarm; static u_char const bin2bcd_data[] = { 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08, 0x09, 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18, 0x19, 0x20, 0x21, 0x22, 0x23, 0x24, 0x25, 0x26, 0x27, 0x28, 0x29, 0x30, 0x31, 0x32, 0x33, 0x34, 0x35, 0x36, 0x37, 0x38, 0x39, 0x40, 0x41, 0x42, 0x43, 0x44, 0x45, 0x46, 0x47, 0x48, 0x49, 0x50, 0x51, 0x52, 0x53, 0x54, 0x55, 0x56, 0x57, 0x58, 0x59, 0x60, 0x61, 0x62, 0x63, 0x64, 0x65, 0x66, 0x67, 0x68, 0x69, 0x70, 0x71, 0x72, 0x73, 0x74, 0x75, 0x76, 0x77, 0x78, 0x79, 0x80, 0x81, 0x82, 0x83, 0x84, 0x85, 0x86, 0x87, 0x88, 0x89, 0x90, 0x91, 0x92, 0x93, 0x94, 0x95, 0x96, 0x97, 0x98, 0x99 }; #define bin2bcd(bin) (bin2bcd_data[bin]) #define rtcout(val) ((status_b & RTCSB_BIN) ? (val) : bin2bcd((val))) static void timevalfix(struct timeval *t1) { if (t1->tv_usec < 0) { t1->tv_sec--; t1->tv_usec += 1000000; } if (t1->tv_usec >= 1000000) { t1->tv_sec++; t1->tv_usec -= 1000000; } } static void timevalsub(struct timeval *t1, const struct timeval *t2) { t1->tv_sec -= t2->tv_sec; t1->tv_usec -= t2->tv_usec; timevalfix(t1); } static int rtc_addr_handler(struct vmctx *ctx, int vcpu, int in, int port, int bytes, uint32_t *eax, void *arg) { if (bytes != 1) return (-1); if (in) { /* straight read of this register will return 0xFF */ *eax = 0xff; return (0); } switch (*eax & 0x7f) { case RTC_SEC: case RTC_SEC_ALARM: case RTC_MIN: case RTC_MIN_ALARM: case RTC_HRS: case RTC_HRS_ALARM: case RTC_WDAY: case RTC_DAY: case RTC_MONTH: case RTC_YEAR: case RTC_STATUSA: case RTC_STATUSB: case RTC_INTR: case RTC_STATUSD: case RTC_NVRAM_START ... RTC_NVRAM_END: break; default: return (-1); } addr = *eax & 0x7f; return (0); } static int rtc_data_handler(struct vmctx *ctx, int vcpu, int in, int port, int bytes, uint32_t *eax, void *arg) { int hour; time_t t; struct timeval cur, delta; static struct timeval last; static struct tm tm; if (bytes != 1) return (-1); gettimeofday(&cur, NULL); /* * Increment the cached time only once per second so we can guarantee * that the guest has at least one second to read the hour:min:sec * separately and still get a coherent view of the time. */ delta = cur; timevalsub(&delta, &last); if (delta.tv_sec >= 1 && (status_b & RTCSB_HALT) == 0) { t = cur.tv_sec; localtime_r(&t, &tm); last = cur; } if (in) { switch (addr) { case RTC_SEC_ALARM: *eax = rtc_alarm.secs; break; case RTC_MIN_ALARM: *eax = rtc_alarm.mins; break; case RTC_HRS_ALARM: *eax = rtc_alarm.hours; break; case RTC_SEC: *eax = rtcout(tm.tm_sec); return (0); case RTC_MIN: *eax = rtcout(tm.tm_min); return (0); case RTC_HRS: if (status_b & RTCSB_24HR) hour = tm.tm_hour; else hour = (tm.tm_hour % 12) + 1; *eax = rtcout(hour); /* * If we are representing time in the 12-hour format * then set the MSB to indicate PM. */ if ((status_b & RTCSB_24HR) == 0 && tm.tm_hour >= 12) *eax |= 0x80; return (0); case RTC_WDAY: *eax = rtcout(tm.tm_wday + 1); return (0); case RTC_DAY: *eax = rtcout(tm.tm_mday); return (0); case RTC_MONTH: *eax = rtcout(tm.tm_mon + 1); return (0); case RTC_YEAR: *eax = rtcout(tm.tm_year % 100); return (0); case RTC_STATUSA: *eax = status_a; return (0); case RTC_STATUSB: *eax = status_b; return (0); case RTC_INTR: *eax = 0; return (0); case RTC_STATUSD: *eax = RTCSD_PWR; return (0); case RTC_NVRAM_START ... RTC_NVRAM_END: *eax = rtc_nvram[addr - RTC_NVRAM_START]; return (0); default: return (-1); } } switch (addr) { case RTC_STATUSA: status_a = *eax & ~RTCSA_TUP; break; case RTC_STATUSB: /* XXX not implemented yet XXX */ if (*eax & RTCSB_PINTR) return (-1); status_b = *eax; break; case RTC_STATUSD: /* ignore write */ break; case RTC_SEC_ALARM: rtc_alarm.secs = *eax; break; case RTC_MIN_ALARM: rtc_alarm.mins = *eax; break; case RTC_HRS_ALARM: rtc_alarm.hours = *eax; break; case RTC_SEC: case RTC_MIN: case RTC_HRS: case RTC_WDAY: case RTC_DAY: case RTC_MONTH: case RTC_YEAR: /* * Ignore writes to the time of day registers */ break; case RTC_NVRAM_START ... RTC_NVRAM_END: rtc_nvram[addr - RTC_NVRAM_START] = *eax; break; default: return (-1); } return (0); } void rtc_init(struct vmctx *ctx) { struct timeval cur; struct tm tm; size_t himem; size_t lomem; int err; err = gettimeofday(&cur, NULL); assert(err == 0); (void) localtime_r(&cur.tv_sec, &tm); memset(rtc_nvram, 0, sizeof(rtc_nvram)); rtc_nvram[nvoff(RTC_CENTURY)] = bin2bcd((tm.tm_year + 1900) / 100); /* XXX init diag/reset code/equipment/checksum ? */ /* * Report guest memory size in nvram cells as required by UEFI. * Little-endian encoding. * 0x34/0x35 - 64KB chunks above 16MB, below 4GB * 0x5b/0x5c/0x5d - 64KB chunks above 4GB */ - err = vm_get_memory_seg(ctx, 0, &lomem, NULL); - assert(err == 0); - - lomem = (lomem - m_16MB) / m_64KB; + lomem = (vm_get_lowmem_size(ctx) - m_16MB) / m_64KB; rtc_nvram[nvoff(RTC_LMEM_LSB)] = lomem; rtc_nvram[nvoff(RTC_LMEM_MSB)] = lomem >> 8; - if (vm_get_memory_seg(ctx, m_4GB, &himem, NULL) == 0) { - himem /= m_64KB; - rtc_nvram[nvoff(RTC_HMEM_LSB)] = himem; - rtc_nvram[nvoff(RTC_HMEM_SB)] = himem >> 8; - rtc_nvram[nvoff(RTC_HMEM_MSB)] = himem >> 16; - } + himem = vm_get_highmem_size(ctx) / m_64KB; + rtc_nvram[nvoff(RTC_HMEM_LSB)] = himem; + rtc_nvram[nvoff(RTC_HMEM_SB)] = himem >> 8; + rtc_nvram[nvoff(RTC_HMEM_MSB)] = himem >> 16; } INOUT_PORT(rtc, IO_RTC, IOPORT_F_INOUT, rtc_addr_handler); INOUT_PORT(rtc, IO_RTC + 1, IOPORT_F_INOUT, rtc_data_handler); static void rtc_dsdt(void) { dsdt_line(""); dsdt_line("Device (RTC)"); dsdt_line("{"); dsdt_line(" Name (_HID, EisaId (\"PNP0B00\"))"); dsdt_line(" Name (_CRS, ResourceTemplate ()"); dsdt_line(" {"); dsdt_indent(2); dsdt_fixed_ioport(IO_RTC, 2); dsdt_fixed_irq(8); dsdt_unindent(2); dsdt_line(" })"); dsdt_line("}"); } LPC_DSDT(rtc_dsdt); SYSRES_IO(0x72, 6); Index: stable/10/usr.sbin/bhyve/smbiostbl.c =================================================================== --- stable/10/usr.sbin/bhyve/smbiostbl.c (revision 270073) +++ stable/10/usr.sbin/bhyve/smbiostbl.c (revision 270074) @@ -1,832 +1,827 @@ /*- * Copyright (c) 2014 Tycho Nightingale * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include "bhyverun.h" #include "smbiostbl.h" #define MB (1024*1024) #define GB (1024ULL*1024*1024) #define SMBIOS_BASE 0xF1000 /* BHYVE_ACPI_BASE - SMBIOS_BASE) */ #define SMBIOS_MAX_LENGTH (0xF2400 - 0xF1000) #define SMBIOS_TYPE_BIOS 0 #define SMBIOS_TYPE_SYSTEM 1 #define SMBIOS_TYPE_CHASSIS 3 #define SMBIOS_TYPE_PROCESSOR 4 #define SMBIOS_TYPE_MEMARRAY 16 #define SMBIOS_TYPE_MEMDEVICE 17 #define SMBIOS_TYPE_MEMARRAYMAP 19 #define SMBIOS_TYPE_BOOT 32 #define SMBIOS_TYPE_EOT 127 struct smbios_structure { uint8_t type; uint8_t length; uint16_t handle; } __packed; typedef int (*initializer_func_t)(struct smbios_structure *template_entry, const char **template_strings, char *curaddr, char **endaddr, uint16_t *n, uint16_t *size); struct smbios_template_entry { struct smbios_structure *entry; const char **strings; initializer_func_t initializer; }; /* * SMBIOS Structure Table Entry Point */ #define SMBIOS_ENTRY_EANCHOR "_SM_" #define SMBIOS_ENTRY_EANCHORLEN 4 #define SMBIOS_ENTRY_IANCHOR "_DMI_" #define SMBIOS_ENTRY_IANCHORLEN 5 struct smbios_entry_point { char eanchor[4]; /* anchor tag */ uint8_t echecksum; /* checksum of entry point structure */ uint8_t eplen; /* length in bytes of entry point */ uint8_t major; /* major version of the SMBIOS spec */ uint8_t minor; /* minor version of the SMBIOS spec */ uint16_t maxssize; /* maximum size in bytes of a struct */ uint8_t revision; /* entry point structure revision */ uint8_t format[5]; /* entry point rev-specific data */ char ianchor[5]; /* intermediate anchor tag */ uint8_t ichecksum; /* intermediate checksum */ uint16_t stlen; /* len in bytes of structure table */ uint32_t staddr; /* physical addr of structure table */ uint16_t stnum; /* number of structure table entries */ uint8_t bcdrev; /* BCD value representing DMI ver */ } __packed; /* * BIOS Information */ #define SMBIOS_FL_ISA 0x00000010 /* ISA is supported */ #define SMBIOS_FL_PCI 0x00000080 /* PCI is supported */ #define SMBIOS_FL_SHADOW 0x00001000 /* BIOS shadowing is allowed */ #define SMBIOS_FL_CDBOOT 0x00008000 /* Boot from CD is supported */ #define SMBIOS_FL_SELBOOT 0x00010000 /* Selectable Boot supported */ #define SMBIOS_FL_EDD 0x00080000 /* EDD Spec is supported */ #define SMBIOS_XB1_FL_ACPI 0x00000001 /* ACPI is supported */ #define SMBIOS_XB2_FL_BBS 0x00000001 /* BIOS Boot Specification */ #define SMBIOS_XB2_FL_VM 0x00000010 /* Virtual Machine */ struct smbios_table_type0 { struct smbios_structure header; uint8_t vendor; /* vendor string */ uint8_t version; /* version string */ uint16_t segment; /* address segment location */ uint8_t rel_date; /* release date */ uint8_t size; /* rom size */ uint64_t cflags; /* characteristics */ uint8_t xc_bytes[2]; /* characteristics ext bytes */ uint8_t sb_major_rel; /* system bios version */ uint8_t sb_minor_rele; uint8_t ecfw_major_rel; /* embedded ctrl fw version */ uint8_t ecfw_minor_rel; } __packed; /* * System Information */ #define SMBIOS_WAKEUP_SWITCH 0x06 /* power switch */ struct smbios_table_type1 { struct smbios_structure header; uint8_t manufacturer; /* manufacturer string */ uint8_t product; /* product name string */ uint8_t version; /* version string */ uint8_t serial; /* serial number string */ uint8_t uuid[16]; /* uuid byte array */ uint8_t wakeup; /* wake-up event */ uint8_t sku; /* sku number string */ uint8_t family; /* family name string */ } __packed; /* * System Enclosure or Chassis */ #define SMBIOS_CHT_UNKNOWN 0x02 /* unknown */ #define SMBIOS_CHST_SAFE 0x03 /* safe */ #define SMBIOS_CHSC_NONE 0x03 /* none */ struct smbios_table_type3 { struct smbios_structure header; uint8_t manufacturer; /* manufacturer string */ uint8_t type; /* type */ uint8_t version; /* version string */ uint8_t serial; /* serial number string */ uint8_t asset; /* asset tag string */ uint8_t bustate; /* boot-up state */ uint8_t psstate; /* power supply state */ uint8_t tstate; /* thermal state */ uint8_t security; /* security status */ uint8_t uheight; /* height in 'u's */ uint8_t cords; /* number of power cords */ uint8_t elems; /* number of element records */ uint8_t elemlen; /* length of records */ uint8_t sku; /* sku number string */ } __packed; /* * Processor Information */ #define SMBIOS_PRT_CENTRAL 0x03 /* central processor */ #define SMBIOS_PRF_OTHER 0x01 /* other */ #define SMBIOS_PRS_PRESENT 0x40 /* socket is populated */ #define SMBIOS_PRS_ENABLED 0x1 /* enabled */ #define SMBIOS_PRU_NONE 0x06 /* none */ #define SMBIOS_PFL_64B 0x04 /* 64-bit capable */ struct smbios_table_type4 { struct smbios_structure header; uint8_t socket; /* socket designation string */ uint8_t type; /* processor type */ uint8_t family; /* processor family */ uint8_t manufacturer; /* manufacturer string */ uint64_t cpuid; /* processor cpuid */ uint8_t version; /* version string */ uint8_t voltage; /* voltage */ uint16_t clkspeed; /* ext clock speed in mhz */ uint16_t maxspeed; /* maximum speed in mhz */ uint16_t curspeed; /* current speed in mhz */ uint8_t status; /* status */ uint8_t upgrade; /* upgrade */ uint16_t l1handle; /* l1 cache handle */ uint16_t l2handle; /* l2 cache handle */ uint16_t l3handle; /* l3 cache handle */ uint8_t serial; /* serial number string */ uint8_t asset; /* asset tag string */ uint8_t part; /* part number string */ uint8_t cores; /* cores per socket */ uint8_t ecores; /* enabled cores */ uint8_t threads; /* threads per socket */ uint16_t cflags; /* processor characteristics */ uint16_t family2; /* processor family 2 */ } __packed; /* * Physical Memory Array */ #define SMBIOS_MAL_SYSMB 0x03 /* system board or motherboard */ #define SMBIOS_MAU_SYSTEM 0x03 /* system memory */ #define SMBIOS_MAE_NONE 0x03 /* none */ struct smbios_table_type16 { struct smbios_structure header; uint8_t location; /* physical device location */ uint8_t use; /* device functional purpose */ uint8_t ecc; /* err detect/correct method */ uint32_t size; /* max mem capacity in kb */ uint16_t errhand; /* handle of error (if any) */ uint16_t ndevs; /* num of slots or sockets */ uint64_t xsize; /* max mem capacity in bytes */ } __packed; /* * Memory Device */ #define SMBIOS_MDFF_UNKNOWN 0x02 /* unknown */ #define SMBIOS_MDT_UNKNOWN 0x02 /* unknown */ #define SMBIOS_MDF_UNKNOWN 0x0004 /* unknown */ struct smbios_table_type17 { struct smbios_structure header; uint16_t arrayhand; /* handle of physl mem array */ uint16_t errhand; /* handle of mem error data */ uint16_t twidth; /* total width in bits */ uint16_t dwidth; /* data width in bits */ uint16_t size; /* size in bytes */ uint8_t form; /* form factor */ uint8_t set; /* set */ uint8_t dloc; /* device locator string */ uint8_t bloc; /* phys bank locator string */ uint8_t type; /* memory type */ uint16_t flags; /* memory characteristics */ uint16_t maxspeed; /* maximum speed in mhz */ uint8_t manufacturer; /* manufacturer string */ uint8_t serial; /* serial number string */ uint8_t asset; /* asset tag string */ uint8_t part; /* part number string */ uint8_t attributes; /* attributes */ uint32_t xsize; /* extended size in mbs */ uint16_t curspeed; /* current speed in mhz */ uint16_t minvoltage; /* minimum voltage */ uint16_t maxvoltage; /* maximum voltage */ uint16_t curvoltage; /* configured voltage */ } __packed; /* * Memory Array Mapped Address */ struct smbios_table_type19 { struct smbios_structure header; uint32_t saddr; /* start phys addr in kb */ uint32_t eaddr; /* end phys addr in kb */ uint16_t arrayhand; /* physical mem array handle */ uint8_t width; /* num of dev in row */ uint64_t xsaddr; /* start phys addr in bytes */ uint64_t xeaddr; /* end phys addr in bytes */ } __packed; /* * System Boot Information */ #define SMBIOS_BOOT_NORMAL 0 /* no errors detected */ struct smbios_table_type32 { struct smbios_structure header; uint8_t reserved[6]; uint8_t status; /* boot status */ } __packed; /* * End-of-Table */ struct smbios_table_type127 { struct smbios_structure header; } __packed; struct smbios_table_type0 smbios_type0_template = { { SMBIOS_TYPE_BIOS, sizeof (struct smbios_table_type0), 0 }, 1, /* bios vendor string */ 2, /* bios version string */ 0xF000, /* bios address segment location */ 3, /* bios release date */ 0x0, /* bios size (64k * (n + 1) is the size in bytes) */ SMBIOS_FL_ISA | SMBIOS_FL_PCI | SMBIOS_FL_SHADOW | SMBIOS_FL_CDBOOT | SMBIOS_FL_EDD, { SMBIOS_XB1_FL_ACPI, SMBIOS_XB2_FL_BBS | SMBIOS_XB2_FL_VM }, 0x0, /* bios major release */ 0x0, /* bios minor release */ 0xff, /* embedded controller firmware major release */ 0xff /* embedded controller firmware minor release */ }; const char *smbios_type0_strings[] = { "BHYVE", /* vendor string */ __TIME__, /* bios version string */ __DATE__, /* bios release date string */ NULL }; struct smbios_table_type1 smbios_type1_template = { { SMBIOS_TYPE_SYSTEM, sizeof (struct smbios_table_type1), 0 }, 1, /* manufacturer string */ 2, /* product string */ 3, /* version string */ 4, /* serial number string */ { 0 }, SMBIOS_WAKEUP_SWITCH, 5, /* sku string */ 6 /* family string */ }; static int smbios_type1_initializer(struct smbios_structure *template_entry, const char **template_strings, char *curaddr, char **endaddr, uint16_t *n, uint16_t *size); const char *smbios_type1_strings[] = { " ", /* manufacturer string */ "BHYVE", /* product name string */ "1.0", /* version string */ "None", /* serial number string */ "None", /* sku string */ " ", /* family name string */ NULL }; struct smbios_table_type3 smbios_type3_template = { { SMBIOS_TYPE_CHASSIS, sizeof (struct smbios_table_type3), 0 }, 1, /* manufacturer string */ SMBIOS_CHT_UNKNOWN, 2, /* version string */ 3, /* serial number string */ 4, /* asset tag string */ SMBIOS_CHST_SAFE, SMBIOS_CHST_SAFE, SMBIOS_CHST_SAFE, SMBIOS_CHSC_NONE, 0, /* height in 'u's (0=enclosure height unspecified) */ 0, /* number of power cords (0=number unspecified) */ 0, /* number of contained element records */ 0, /* length of records */ 5 /* sku number string */ }; const char *smbios_type3_strings[] = { " ", /* manufacturer string */ "1.0", /* version string */ "None", /* serial number string */ "None", /* asset tag string */ "None", /* sku number string */ NULL }; struct smbios_table_type4 smbios_type4_template = { { SMBIOS_TYPE_PROCESSOR, sizeof (struct smbios_table_type4), 0 }, 1, /* socket designation string */ SMBIOS_PRT_CENTRAL, SMBIOS_PRF_OTHER, 2, /* manufacturer string */ 0, /* cpuid */ 3, /* version string */ 0, /* voltage */ 0, /* external clock frequency in mhz (0=unknown) */ 0, /* maximum frequency in mhz (0=unknown) */ 0, /* current frequency in mhz (0=unknown) */ SMBIOS_PRS_PRESENT | SMBIOS_PRS_ENABLED, SMBIOS_PRU_NONE, -1, /* l1 cache handle */ -1, /* l2 cache handle */ -1, /* l3 cache handle */ 4, /* serial number string */ 5, /* asset tag string */ 6, /* part number string */ 0, /* cores per socket (0=unknown) */ 0, /* enabled cores per socket (0=unknown) */ 0, /* threads per socket (0=unknown) */ SMBIOS_PFL_64B, SMBIOS_PRF_OTHER }; const char *smbios_type4_strings[] = { " ", /* socket designation string */ " ", /* manufacturer string */ " ", /* version string */ "None", /* serial number string */ "None", /* asset tag string */ "None", /* part number string */ NULL }; static int smbios_type4_initializer(struct smbios_structure *template_entry, const char **template_strings, char *curaddr, char **endaddr, uint16_t *n, uint16_t *size); struct smbios_table_type16 smbios_type16_template = { { SMBIOS_TYPE_MEMARRAY, sizeof (struct smbios_table_type16), 0 }, SMBIOS_MAL_SYSMB, SMBIOS_MAU_SYSTEM, SMBIOS_MAE_NONE, 0x80000000, /* max mem capacity in kb (0x80000000=use extended) */ -1, /* handle of error (if any) */ 0, /* number of slots or sockets (TBD) */ 0 /* extended maximum memory capacity in bytes (TBD) */ }; static int smbios_type16_initializer(struct smbios_structure *template_entry, const char **template_strings, char *curaddr, char **endaddr, uint16_t *n, uint16_t *size); struct smbios_table_type17 smbios_type17_template = { { SMBIOS_TYPE_MEMDEVICE, sizeof (struct smbios_table_type17), 0 }, -1, /* handle of physical memory array */ -1, /* handle of memory error data */ 64, /* total width in bits including ecc */ 64, /* data width in bits */ 0x7fff, /* size in bytes (0x7fff=use extended)*/ SMBIOS_MDFF_UNKNOWN, 0, /* set (0x00=none, 0xff=unknown) */ 1, /* device locator string */ 2, /* physical bank locator string */ SMBIOS_MDT_UNKNOWN, SMBIOS_MDF_UNKNOWN, 0, /* maximum memory speed in mhz (0=unknown) */ 3, /* manufacturer string */ 4, /* serial number string */ 5, /* asset tag string */ 6, /* part number string */ 0, /* attributes (0=unknown rank information) */ 0, /* extended size in mb (TBD) */ 0, /* current speed in mhz (0=unknown) */ 0, /* minimum voltage in mv (0=unknown) */ 0, /* maximum voltage in mv (0=unknown) */ 0 /* configured voltage in mv (0=unknown) */ }; const char *smbios_type17_strings[] = { " ", /* device locator string */ " ", /* physical bank locator string */ " ", /* manufacturer string */ "None", /* serial number string */ "None", /* asset tag string */ "None", /* part number string */ NULL }; static int smbios_type17_initializer(struct smbios_structure *template_entry, const char **template_strings, char *curaddr, char **endaddr, uint16_t *n, uint16_t *size); struct smbios_table_type19 smbios_type19_template = { { SMBIOS_TYPE_MEMARRAYMAP, sizeof (struct smbios_table_type19), 0 }, 0xffffffff, /* starting phys addr in kb (0xffffffff=use ext) */ 0xffffffff, /* ending phys addr in kb (0xffffffff=use ext) */ -1, /* physical memory array handle */ 1, /* number of devices that form a row */ 0, /* extended starting phys addr in bytes (TDB) */ 0 /* extended ending phys addr in bytes (TDB) */ }; static int smbios_type19_initializer(struct smbios_structure *template_entry, const char **template_strings, char *curaddr, char **endaddr, uint16_t *n, uint16_t *size); struct smbios_table_type32 smbios_type32_template = { { SMBIOS_TYPE_BOOT, sizeof (struct smbios_table_type32), 0 }, { 0, 0, 0, 0, 0, 0 }, SMBIOS_BOOT_NORMAL }; struct smbios_table_type127 smbios_type127_template = { { SMBIOS_TYPE_EOT, sizeof (struct smbios_table_type127), 0 } }; static int smbios_generic_initializer(struct smbios_structure *template_entry, const char **template_strings, char *curaddr, char **endaddr, uint16_t *n, uint16_t *size); static struct smbios_template_entry smbios_template[] = { { (struct smbios_structure *)&smbios_type0_template, smbios_type0_strings, smbios_generic_initializer }, { (struct smbios_structure *)&smbios_type1_template, smbios_type1_strings, smbios_type1_initializer }, { (struct smbios_structure *)&smbios_type3_template, smbios_type3_strings, smbios_generic_initializer }, { (struct smbios_structure *)&smbios_type4_template, smbios_type4_strings, smbios_type4_initializer }, { (struct smbios_structure *)&smbios_type16_template, NULL, smbios_type16_initializer }, { (struct smbios_structure *)&smbios_type17_template, smbios_type17_strings, smbios_type17_initializer }, { (struct smbios_structure *)&smbios_type19_template, NULL, smbios_type19_initializer }, { (struct smbios_structure *)&smbios_type32_template, NULL, smbios_generic_initializer }, { (struct smbios_structure *)&smbios_type127_template, NULL, smbios_generic_initializer }, { NULL,NULL, NULL } }; static uint64_t guest_lomem, guest_himem; static uint16_t type16_handle; static int smbios_generic_initializer(struct smbios_structure *template_entry, const char **template_strings, char *curaddr, char **endaddr, uint16_t *n, uint16_t *size) { struct smbios_structure *entry; memcpy(curaddr, template_entry, template_entry->length); entry = (struct smbios_structure *)curaddr; entry->handle = *n + 1; curaddr += entry->length; if (template_strings != NULL) { int i; for (i = 0; template_strings[i] != NULL; i++) { const char *string; int len; string = template_strings[i]; len = strlen(string) + 1; memcpy(curaddr, string, len); curaddr += len; } *curaddr = '\0'; curaddr++; } else { /* Minimum string section is double nul */ *curaddr = '\0'; curaddr++; *curaddr = '\0'; curaddr++; } (*n)++; *endaddr = curaddr; return (0); } static int smbios_type1_initializer(struct smbios_structure *template_entry, const char **template_strings, char *curaddr, char **endaddr, uint16_t *n, uint16_t *size) { struct smbios_table_type1 *type1; smbios_generic_initializer(template_entry, template_strings, curaddr, endaddr, n, size); type1 = (struct smbios_table_type1 *)curaddr; if (guest_uuid_str != NULL) { uuid_t uuid; uint32_t status; uuid_from_string(guest_uuid_str, &uuid, &status); if (status != uuid_s_ok) return (-1); uuid_enc_le(&type1->uuid, &uuid); } else { MD5_CTX mdctx; u_char digest[16]; char hostname[MAXHOSTNAMELEN]; /* * Universally unique and yet reproducible are an * oxymoron, however reproducible is desirable in * this case. */ if (gethostname(hostname, sizeof(hostname))) return (-1); MD5Init(&mdctx); MD5Update(&mdctx, vmname, strlen(vmname)); MD5Update(&mdctx, hostname, sizeof(hostname)); MD5Final(digest, &mdctx); /* * Set the variant and version number. */ digest[6] &= 0x0F; digest[6] |= 0x30; /* version 3 */ digest[8] &= 0x3F; digest[8] |= 0x80; memcpy(&type1->uuid, digest, sizeof (digest)); } return (0); } static int smbios_type4_initializer(struct smbios_structure *template_entry, const char **template_strings, char *curaddr, char **endaddr, uint16_t *n, uint16_t *size) { int i; for (i = 0; i < guest_ncpus; i++) { struct smbios_table_type4 *type4; char *p; int nstrings, len; smbios_generic_initializer(template_entry, template_strings, curaddr, endaddr, n, size); type4 = (struct smbios_table_type4 *)curaddr; p = curaddr + sizeof (struct smbios_table_type4); nstrings = 0; while (p < *endaddr - 1) { if (*p++ == '\0') nstrings++; } len = sprintf(*endaddr - 1, "CPU #%d", i) + 1; *endaddr += len - 1; *(*endaddr) = '\0'; (*endaddr)++; type4->socket = nstrings + 1; curaddr = *endaddr; } return (0); } static int smbios_type16_initializer(struct smbios_structure *template_entry, const char **template_strings, char *curaddr, char **endaddr, uint16_t *n, uint16_t *size) { struct smbios_table_type16 *type16; type16_handle = *n; smbios_generic_initializer(template_entry, template_strings, curaddr, endaddr, n, size); type16 = (struct smbios_table_type16 *)curaddr; type16->xsize = guest_lomem + guest_himem; type16->ndevs = guest_himem > 0 ? 2 : 1; return (0); } static int smbios_type17_initializer(struct smbios_structure *template_entry, const char **template_strings, char *curaddr, char **endaddr, uint16_t *n, uint16_t *size) { struct smbios_table_type17 *type17; smbios_generic_initializer(template_entry, template_strings, curaddr, endaddr, n, size); type17 = (struct smbios_table_type17 *)curaddr; type17->arrayhand = type16_handle; type17->xsize = guest_lomem; if (guest_himem > 0) { curaddr = *endaddr; smbios_generic_initializer(template_entry, template_strings, curaddr, endaddr, n, size); type17 = (struct smbios_table_type17 *)curaddr; type17->arrayhand = type16_handle; type17->xsize = guest_himem; } return (0); } static int smbios_type19_initializer(struct smbios_structure *template_entry, const char **template_strings, char *curaddr, char **endaddr, uint16_t *n, uint16_t *size) { struct smbios_table_type19 *type19; smbios_generic_initializer(template_entry, template_strings, curaddr, endaddr, n, size); type19 = (struct smbios_table_type19 *)curaddr; type19->arrayhand = type16_handle; type19->xsaddr = 0; type19->xeaddr = guest_lomem; if (guest_himem > 0) { curaddr = *endaddr; smbios_generic_initializer(template_entry, template_strings, curaddr, endaddr, n, size); type19 = (struct smbios_table_type19 *)curaddr; type19->arrayhand = type16_handle; type19->xsaddr = 4*GB; type19->xeaddr = guest_himem; } return (0); } static void smbios_ep_initializer(struct smbios_entry_point *smbios_ep, uint32_t staddr) { memset(smbios_ep, 0, sizeof(*smbios_ep)); memcpy(smbios_ep->eanchor, SMBIOS_ENTRY_EANCHOR, SMBIOS_ENTRY_EANCHORLEN); smbios_ep->eplen = 0x1F; assert(sizeof (struct smbios_entry_point) == smbios_ep->eplen); smbios_ep->major = 2; smbios_ep->minor = 4; smbios_ep->revision = 0; memcpy(smbios_ep->ianchor, SMBIOS_ENTRY_IANCHOR, SMBIOS_ENTRY_IANCHORLEN); smbios_ep->staddr = staddr; smbios_ep->bcdrev = 0x24; } static void smbios_ep_finalizer(struct smbios_entry_point *smbios_ep, uint16_t len, uint16_t num, uint16_t maxssize) { uint8_t checksum; int i; smbios_ep->maxssize = maxssize; smbios_ep->stlen = len; smbios_ep->stnum = num; checksum = 0; for (i = 0x10; i < 0x1f; i++) { checksum -= ((uint8_t *)smbios_ep)[i]; } smbios_ep->ichecksum = checksum; checksum = 0; for (i = 0; i < 0x1f; i++) { checksum -= ((uint8_t *)smbios_ep)[i]; } smbios_ep->echecksum = checksum; } int smbios_build(struct vmctx *ctx) { struct smbios_entry_point *smbios_ep; uint16_t n; uint16_t maxssize; char *curaddr, *startaddr, *ststartaddr; int i; int err; - err = vm_get_memory_seg(ctx, 0, &guest_lomem, NULL); - if (err != 0) - return (err); - - err = vm_get_memory_seg(ctx, 4*GB, &guest_himem, NULL); - if (err != 0) - return (err); + guest_lomem = vm_get_lowmem_size(ctx); + guest_himem = vm_get_highmem_size(ctx); startaddr = paddr_guest2host(ctx, SMBIOS_BASE, SMBIOS_MAX_LENGTH); if (startaddr == NULL) { fprintf(stderr, "smbios table requires mapped mem\n"); return (ENOMEM); } curaddr = startaddr; smbios_ep = (struct smbios_entry_point *)curaddr; smbios_ep_initializer(smbios_ep, SMBIOS_BASE + sizeof(struct smbios_entry_point)); curaddr += sizeof(struct smbios_entry_point); ststartaddr = curaddr; n = 0; maxssize = 0; for (i = 0; smbios_template[i].entry != NULL; i++) { struct smbios_structure *entry; const char **strings; initializer_func_t initializer; char *endaddr; uint16_t size; entry = smbios_template[i].entry; strings = smbios_template[i].strings; initializer = smbios_template[i].initializer; err = (*initializer)(entry, strings, curaddr, &endaddr, &n, &size); if (err != 0) return (err); if (size > maxssize) maxssize = size; curaddr = endaddr; } assert(curaddr - startaddr < SMBIOS_MAX_LENGTH); smbios_ep_finalizer(smbios_ep, curaddr - ststartaddr, n, maxssize); return (0); } Index: stable/10/usr.sbin/bhyveload/bhyveload.c =================================================================== --- stable/10/usr.sbin/bhyveload/bhyveload.c (revision 270073) +++ stable/10/usr.sbin/bhyveload/bhyveload.c (revision 270074) @@ -1,746 +1,746 @@ /*- * Copyright (c) 2011 NetApp, Inc. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ /*- * Copyright (c) 2011 Google, Inc. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "userboot.h" #define MB (1024 * 1024UL) #define GB (1024 * 1024 * 1024UL) #define BSP 0 #define NDISKS 32 static char *host_base; static struct termios term, oldterm; static int disk_fd[NDISKS]; static int ndisks; static int consin_fd, consout_fd; static char *vmname, *progname; static struct vmctx *ctx; static uint64_t gdtbase, cr3, rsp; static void cb_exit(void *arg, int v); /* * Console i/o callbacks */ static void cb_putc(void *arg, int ch) { char c = ch; (void) write(consout_fd, &c, 1); } static int cb_getc(void *arg) { char c; if (read(consin_fd, &c, 1) == 1) return (c); return (-1); } static int cb_poll(void *arg) { int n; if (ioctl(consin_fd, FIONREAD, &n) >= 0) return (n > 0); return (0); } /* * Host filesystem i/o callbacks */ struct cb_file { int cf_isdir; size_t cf_size; struct stat cf_stat; union { int fd; DIR *dir; } cf_u; }; static int cb_open(void *arg, const char *filename, void **hp) { struct stat st; struct cb_file *cf; char path[PATH_MAX]; if (!host_base) return (ENOENT); strlcpy(path, host_base, PATH_MAX); if (path[strlen(path) - 1] == '/') path[strlen(path) - 1] = 0; strlcat(path, filename, PATH_MAX); cf = malloc(sizeof(struct cb_file)); if (stat(path, &cf->cf_stat) < 0) { free(cf); return (errno); } cf->cf_size = st.st_size; if (S_ISDIR(cf->cf_stat.st_mode)) { cf->cf_isdir = 1; cf->cf_u.dir = opendir(path); if (!cf->cf_u.dir) goto out; *hp = cf; return (0); } if (S_ISREG(cf->cf_stat.st_mode)) { cf->cf_isdir = 0; cf->cf_u.fd = open(path, O_RDONLY); if (cf->cf_u.fd < 0) goto out; *hp = cf; return (0); } out: free(cf); return (EINVAL); } static int cb_close(void *arg, void *h) { struct cb_file *cf = h; if (cf->cf_isdir) closedir(cf->cf_u.dir); else close(cf->cf_u.fd); free(cf); return (0); } static int cb_isdir(void *arg, void *h) { struct cb_file *cf = h; return (cf->cf_isdir); } static int cb_read(void *arg, void *h, void *buf, size_t size, size_t *resid) { struct cb_file *cf = h; ssize_t sz; if (cf->cf_isdir) return (EINVAL); sz = read(cf->cf_u.fd, buf, size); if (sz < 0) return (EINVAL); *resid = size - sz; return (0); } static int cb_readdir(void *arg, void *h, uint32_t *fileno_return, uint8_t *type_return, size_t *namelen_return, char *name) { struct cb_file *cf = h; struct dirent *dp; if (!cf->cf_isdir) return (EINVAL); dp = readdir(cf->cf_u.dir); if (!dp) return (ENOENT); /* * Note: d_namlen is in the range 0..255 and therefore less * than PATH_MAX so we don't need to test before copying. */ *fileno_return = dp->d_fileno; *type_return = dp->d_type; *namelen_return = dp->d_namlen; memcpy(name, dp->d_name, dp->d_namlen); name[dp->d_namlen] = 0; return (0); } static int cb_seek(void *arg, void *h, uint64_t offset, int whence) { struct cb_file *cf = h; if (cf->cf_isdir) return (EINVAL); if (lseek(cf->cf_u.fd, offset, whence) < 0) return (errno); return (0); } static int cb_stat(void *arg, void *h, int *mode, int *uid, int *gid, uint64_t *size) { struct cb_file *cf = h; *mode = cf->cf_stat.st_mode; *uid = cf->cf_stat.st_uid; *gid = cf->cf_stat.st_gid; *size = cf->cf_stat.st_size; return (0); } /* * Disk image i/o callbacks */ static int cb_diskread(void *arg, int unit, uint64_t from, void *to, size_t size, size_t *resid) { ssize_t n; if (unit < 0 || unit >= ndisks ) return (EIO); n = pread(disk_fd[unit], to, size, from); if (n < 0) return (errno); *resid = size - n; return (0); } static int cb_diskioctl(void *arg, int unit, u_long cmd, void *data) { struct stat sb; if (unit < 0 || unit >= ndisks) return (EBADF); switch (cmd) { case DIOCGSECTORSIZE: *(u_int *)data = 512; break; case DIOCGMEDIASIZE: if (fstat(disk_fd[unit], &sb) == 0) *(off_t *)data = sb.st_size; else return (ENOTTY); break; default: return (ENOTTY); } return (0); } /* * Guest virtual machine i/o callbacks */ static int cb_copyin(void *arg, const void *from, uint64_t to, size_t size) { char *ptr; to &= 0x7fffffff; ptr = vm_map_gpa(ctx, to, size); if (ptr == NULL) return (EFAULT); memcpy(ptr, from, size); return (0); } static int cb_copyout(void *arg, uint64_t from, void *to, size_t size) { char *ptr; from &= 0x7fffffff; ptr = vm_map_gpa(ctx, from, size); if (ptr == NULL) return (EFAULT); memcpy(to, ptr, size); return (0); } static void cb_setreg(void *arg, int r, uint64_t v) { int error; enum vm_reg_name vmreg; vmreg = VM_REG_LAST; switch (r) { case 4: vmreg = VM_REG_GUEST_RSP; rsp = v; break; default: break; } if (vmreg == VM_REG_LAST) { printf("test_setreg(%d): not implemented\n", r); cb_exit(NULL, USERBOOT_EXIT_QUIT); } error = vm_set_register(ctx, BSP, vmreg, v); if (error) { perror("vm_set_register"); cb_exit(NULL, USERBOOT_EXIT_QUIT); } } static void cb_setmsr(void *arg, int r, uint64_t v) { int error; enum vm_reg_name vmreg; vmreg = VM_REG_LAST; switch (r) { case MSR_EFER: vmreg = VM_REG_GUEST_EFER; break; default: break; } if (vmreg == VM_REG_LAST) { printf("test_setmsr(%d): not implemented\n", r); cb_exit(NULL, USERBOOT_EXIT_QUIT); } error = vm_set_register(ctx, BSP, vmreg, v); if (error) { perror("vm_set_msr"); cb_exit(NULL, USERBOOT_EXIT_QUIT); } } static void cb_setcr(void *arg, int r, uint64_t v) { int error; enum vm_reg_name vmreg; vmreg = VM_REG_LAST; switch (r) { case 0: vmreg = VM_REG_GUEST_CR0; break; case 3: vmreg = VM_REG_GUEST_CR3; cr3 = v; break; case 4: vmreg = VM_REG_GUEST_CR4; break; default: break; } if (vmreg == VM_REG_LAST) { printf("test_setcr(%d): not implemented\n", r); cb_exit(NULL, USERBOOT_EXIT_QUIT); } error = vm_set_register(ctx, BSP, vmreg, v); if (error) { perror("vm_set_cr"); cb_exit(NULL, USERBOOT_EXIT_QUIT); } } static void cb_setgdt(void *arg, uint64_t base, size_t size) { int error; error = vm_set_desc(ctx, BSP, VM_REG_GUEST_GDTR, base, size - 1, 0); if (error != 0) { perror("vm_set_desc(gdt)"); cb_exit(NULL, USERBOOT_EXIT_QUIT); } gdtbase = base; } static void cb_exec(void *arg, uint64_t rip) { int error; if (cr3 == 0) error = vm_setup_freebsd_registers_i386(ctx, BSP, rip, gdtbase, rsp); else error = vm_setup_freebsd_registers(ctx, BSP, rip, cr3, gdtbase, rsp); if (error) { perror("vm_setup_freebsd_registers"); cb_exit(NULL, USERBOOT_EXIT_QUIT); } cb_exit(NULL, 0); } /* * Misc */ static void cb_delay(void *arg, int usec) { usleep(usec); } static void cb_exit(void *arg, int v) { tcsetattr(consout_fd, TCSAFLUSH, &oldterm); exit(v); } static void cb_getmem(void *arg, uint64_t *ret_lowmem, uint64_t *ret_highmem) { - vm_get_memory_seg(ctx, 0, ret_lowmem, NULL); - vm_get_memory_seg(ctx, 4 * GB, ret_highmem, NULL); + *ret_lowmem = vm_get_lowmem_size(ctx); + *ret_highmem = vm_get_highmem_size(ctx); } struct env { const char *str; /* name=value */ SLIST_ENTRY(env) next; }; static SLIST_HEAD(envhead, env) envhead; static void addenv(const char *str) { struct env *env; env = malloc(sizeof(struct env)); env->str = str; SLIST_INSERT_HEAD(&envhead, env, next); } static const char * cb_getenv(void *arg, int num) { int i; struct env *env; i = 0; SLIST_FOREACH(env, &envhead, next) { if (i == num) return (env->str); i++; } return (NULL); } static struct loader_callbacks cb = { .getc = cb_getc, .putc = cb_putc, .poll = cb_poll, .open = cb_open, .close = cb_close, .isdir = cb_isdir, .read = cb_read, .readdir = cb_readdir, .seek = cb_seek, .stat = cb_stat, .diskread = cb_diskread, .diskioctl = cb_diskioctl, .copyin = cb_copyin, .copyout = cb_copyout, .setreg = cb_setreg, .setmsr = cb_setmsr, .setcr = cb_setcr, .setgdt = cb_setgdt, .exec = cb_exec, .delay = cb_delay, .exit = cb_exit, .getmem = cb_getmem, .getenv = cb_getenv, }; static int altcons_open(char *path) { struct stat sb; int err; int fd; /* * Allow stdio to be passed in so that the same string * can be used for the bhyveload console and bhyve com-port * parameters */ if (!strcmp(path, "stdio")) return (0); err = stat(path, &sb); if (err == 0) { if (!S_ISCHR(sb.st_mode)) err = ENOTSUP; else { fd = open(path, O_RDWR | O_NONBLOCK); if (fd < 0) err = errno; else consin_fd = consout_fd = fd; } } return (err); } static int disk_open(char *path) { int err, fd; if (ndisks > NDISKS) return (ERANGE); err = 0; fd = open(path, O_RDONLY); if (fd > 0) { disk_fd[ndisks] = fd; ndisks++; } else err = errno; return (err); } static void usage(void) { fprintf(stderr, "usage: %s [-m mem-size] [-d ] [-h ]\n" " %*s [-e ] [-c ] \n", progname, (int)strlen(progname), ""); exit(1); } int main(int argc, char** argv) { void *h; void (*func)(struct loader_callbacks *, void *, int, int); uint64_t mem_size; int opt, error, need_reinit; progname = basename(argv[0]); mem_size = 256 * MB; consin_fd = STDIN_FILENO; consout_fd = STDOUT_FILENO; while ((opt = getopt(argc, argv, "c:d:e:h:m:")) != -1) { switch (opt) { case 'c': error = altcons_open(optarg); if (error != 0) errx(EX_USAGE, "Could not open '%s'", optarg); break; case 'd': error = disk_open(optarg); if (error != 0) errx(EX_USAGE, "Could not open '%s'", optarg); break; case 'e': addenv(optarg); break; case 'h': host_base = optarg; break; case 'm': error = vm_parse_memsize(optarg, &mem_size); if (error != 0) errx(EX_USAGE, "Invalid memsize '%s'", optarg); break; case '?': usage(); } } argc -= optind; argv += optind; if (argc != 1) usage(); vmname = argv[0]; need_reinit = 0; error = vm_create(vmname); if (error) { if (errno != EEXIST) { perror("vm_create"); exit(1); } need_reinit = 1; } ctx = vm_open(vmname); if (ctx == NULL) { perror("vm_open"); exit(1); } if (need_reinit) { error = vm_reinit(ctx); if (error) { perror("vm_reinit"); exit(1); } } error = vm_setup_memory(ctx, mem_size, VM_MMAP_ALL); if (error) { perror("vm_setup_memory"); exit(1); } tcgetattr(consout_fd, &term); oldterm = term; cfmakeraw(&term); term.c_cflag |= CLOCAL; tcsetattr(consout_fd, TCSAFLUSH, &term); h = dlopen("/boot/userboot.so", RTLD_LOCAL); if (!h) { printf("%s\n", dlerror()); return (1); } func = dlsym(h, "loader_main"); if (!func) { printf("%s\n", dlerror()); return (1); } addenv("smbios.bios.vendor=BHYVE"); addenv("boot_serial=1"); func(&cb, NULL, USERBOOT_VERSION_3, ndisks); } Index: stable/10 =================================================================== --- stable/10 (revision 270073) +++ stable/10 (revision 270074) Property changes on: stable/10 ___________________________________________________________________ Modified: svn:mergeinfo ## -0,0 +0,1 ## Merged /head:r267311,267330,267811,267884