2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2011 NetApp, Inc.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
16 * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
34 #include <sys/types.h>
35 #ifndef WITHOUT_CAPSICUM
36 #include <sys/capsicum.h>
40 #include <sys/socket.h>
48 #include <amd64/vmm/intel/vmcs.h>
50 #include <machine/atomic.h>
51 #include <machine/segments.h>
53 #ifndef WITHOUT_CAPSICUM
54 #include <capsicum_helpers.h>
68 #include <pthread_np.h>
79 #include <machine/vmm.h>
80 #ifndef WITHOUT_CAPSICUM
81 #include <machine/vmm_dev.h>
83 #include <machine/vmm_instruction_emul.h>
96 #include "kernemu_dev.h"
100 #include "pci_emul.h"
103 #include "smbiostbl.h"
104 #ifdef BHYVE_SNAPSHOT
105 #include "snapshot.h"
108 #include "spinup_ap.h"
112 #define GUEST_NIO_PORT 0x488 /* guest upcalls via i/o port */
114 #define MB (1024UL * 1024)
115 #define GB (1024UL * MB)
117 static const char * const vmx_exit_reason_desc[] = {
118 [EXIT_REASON_EXCEPTION] = "Exception or non-maskable interrupt (NMI)",
119 [EXIT_REASON_EXT_INTR] = "External interrupt",
120 [EXIT_REASON_TRIPLE_FAULT] = "Triple fault",
121 [EXIT_REASON_INIT] = "INIT signal",
122 [EXIT_REASON_SIPI] = "Start-up IPI (SIPI)",
123 [EXIT_REASON_IO_SMI] = "I/O system-management interrupt (SMI)",
124 [EXIT_REASON_SMI] = "Other SMI",
125 [EXIT_REASON_INTR_WINDOW] = "Interrupt window",
126 [EXIT_REASON_NMI_WINDOW] = "NMI window",
127 [EXIT_REASON_TASK_SWITCH] = "Task switch",
128 [EXIT_REASON_CPUID] = "CPUID",
129 [EXIT_REASON_GETSEC] = "GETSEC",
130 [EXIT_REASON_HLT] = "HLT",
131 [EXIT_REASON_INVD] = "INVD",
132 [EXIT_REASON_INVLPG] = "INVLPG",
133 [EXIT_REASON_RDPMC] = "RDPMC",
134 [EXIT_REASON_RDTSC] = "RDTSC",
135 [EXIT_REASON_RSM] = "RSM",
136 [EXIT_REASON_VMCALL] = "VMCALL",
137 [EXIT_REASON_VMCLEAR] = "VMCLEAR",
138 [EXIT_REASON_VMLAUNCH] = "VMLAUNCH",
139 [EXIT_REASON_VMPTRLD] = "VMPTRLD",
140 [EXIT_REASON_VMPTRST] = "VMPTRST",
141 [EXIT_REASON_VMREAD] = "VMREAD",
142 [EXIT_REASON_VMRESUME] = "VMRESUME",
143 [EXIT_REASON_VMWRITE] = "VMWRITE",
144 [EXIT_REASON_VMXOFF] = "VMXOFF",
145 [EXIT_REASON_VMXON] = "VMXON",
146 [EXIT_REASON_CR_ACCESS] = "Control-register accesses",
147 [EXIT_REASON_DR_ACCESS] = "MOV DR",
148 [EXIT_REASON_INOUT] = "I/O instruction",
149 [EXIT_REASON_RDMSR] = "RDMSR",
150 [EXIT_REASON_WRMSR] = "WRMSR",
151 [EXIT_REASON_INVAL_VMCS] =
152 "VM-entry failure due to invalid guest state",
153 [EXIT_REASON_INVAL_MSR] = "VM-entry failure due to MSR loading",
154 [EXIT_REASON_MWAIT] = "MWAIT",
155 [EXIT_REASON_MTF] = "Monitor trap flag",
156 [EXIT_REASON_MONITOR] = "MONITOR",
157 [EXIT_REASON_PAUSE] = "PAUSE",
158 [EXIT_REASON_MCE_DURING_ENTRY] =
159 "VM-entry failure due to machine-check event",
160 [EXIT_REASON_TPR] = "TPR below threshold",
161 [EXIT_REASON_APIC_ACCESS] = "APIC access",
162 [EXIT_REASON_VIRTUALIZED_EOI] = "Virtualized EOI",
163 [EXIT_REASON_GDTR_IDTR] = "Access to GDTR or IDTR",
164 [EXIT_REASON_LDTR_TR] = "Access to LDTR or TR",
165 [EXIT_REASON_EPT_FAULT] = "EPT violation",
166 [EXIT_REASON_EPT_MISCONFIG] = "EPT misconfiguration",
167 [EXIT_REASON_INVEPT] = "INVEPT",
168 [EXIT_REASON_RDTSCP] = "RDTSCP",
169 [EXIT_REASON_VMX_PREEMPT] = "VMX-preemption timer expired",
170 [EXIT_REASON_INVVPID] = "INVVPID",
171 [EXIT_REASON_WBINVD] = "WBINVD",
172 [EXIT_REASON_XSETBV] = "XSETBV",
173 [EXIT_REASON_APIC_WRITE] = "APIC write",
174 [EXIT_REASON_RDRAND] = "RDRAND",
175 [EXIT_REASON_INVPCID] = "INVPCID",
176 [EXIT_REASON_VMFUNC] = "VMFUNC",
177 [EXIT_REASON_ENCLS] = "ENCLS",
178 [EXIT_REASON_RDSEED] = "RDSEED",
179 [EXIT_REASON_PM_LOG_FULL] = "Page-modification log full",
180 [EXIT_REASON_XSAVES] = "XSAVES",
181 [EXIT_REASON_XRSTORS] = "XRSTORS"
184 typedef int (*vmexit_handler_t)(struct vmctx *, struct vm_exit *, int *vcpu);
185 extern int vmexit_task_switch(struct vmctx *, struct vm_exit *, int *vcpu);
188 uint16_t cores, maxcpus, sockets, threads;
192 static char *progname;
193 static const int BSP = 0;
195 static cpuset_t cpumask;
197 static void vm_loop(struct vmctx *ctx, int vcpu, uint64_t rip);
199 static struct vm_exit vmexit[VM_MAXCPU];
202 uint64_t vmexit_bogus;
203 uint64_t vmexit_reqidle;
205 uint64_t vmexit_pause;
206 uint64_t vmexit_mtrap;
207 uint64_t vmexit_inst_emul;
208 uint64_t cpu_switch_rotate;
209 uint64_t cpu_switch_direct;
214 struct vmctx *mt_ctx;
216 } mt_vmm_info[VM_MAXCPU];
218 static cpuset_t *vcpumap[VM_MAXCPU] = { NULL };
225 "Usage: %s [-AaCDeHhPSuWwxY]\n"
226 " %*s [-c [[cpus=]numcpus][,sockets=n][,cores=n][,threads=n]]\n"
227 " %*s [-G port] [-k file] [-l lpc] [-m mem] [-o var=value]\n"
228 " %*s [-p vcpu:hostcpu] [-r file] [-s pci] [-U uuid] vmname\n"
229 " -A: create ACPI tables\n"
230 " -a: local apic is in xAPIC mode (deprecated)\n"
231 " -C: include guest memory in core file\n"
232 " -c: number of CPUs and/or topology specification\n"
233 " -D: destroy on power-off\n"
234 " -e: exit on unhandled I/O access\n"
235 " -G: start a debug server\n"
236 " -H: vmexit from the guest on HLT\n"
238 " -k: key=value flat config file\n"
239 " -l: LPC device configuration\n"
240 " -m: memory size in MB\n"
241 " -o: set config 'var' to 'value'\n"
242 " -P: vmexit from the guest on pause\n"
243 " -p: pin 'vcpu' to 'hostcpu'\n"
244 #ifdef BHYVE_SNAPSHOT
245 " -r: path to checkpoint file\n"
247 " -S: guest memory cannot be swapped\n"
248 " -s: <slot,driver,configinfo> PCI slot config\n"
250 " -u: RTC keeps UTC time\n"
251 " -W: force virtio to use single-vector MSI\n"
252 " -w: ignore unimplemented MSRs\n"
253 " -x: local APIC is in x2APIC mode\n"
254 " -Y: disable MPtable generation\n",
255 progname, (int)strlen(progname), "", (int)strlen(progname), "",
256 (int)strlen(progname), "");
262 * XXX This parser is known to have the following issues:
263 * 1. It accepts null key=value tokens ",," as setting "cpus" to an
266 * The acceptance of a null specification ('-c ""') is by design to match the
267 * manual page syntax specification, this results in a topology of 1 vCPU.
270 topology_parse(const char *opt)
275 set_config_value("sockets", "1");
276 set_config_value("cores", "1");
277 set_config_value("threads", "1");
278 set_config_value("cpus", "1");
284 errx(4, "Failed to allocate memory");
286 while ((cp = strsep(&str, ",")) != NULL) {
287 if (strncmp(cp, "cpus=", strlen("cpus=")) == 0)
288 set_config_value("cpus", cp + strlen("cpus="));
289 else if (strncmp(cp, "sockets=", strlen("sockets=")) == 0)
290 set_config_value("sockets", cp + strlen("sockets="));
291 else if (strncmp(cp, "cores=", strlen("cores=")) == 0)
292 set_config_value("cores", cp + strlen("cores="));
293 else if (strncmp(cp, "threads=", strlen("threads=")) == 0)
294 set_config_value("threads", cp + strlen("threads="));
295 #ifdef notyet /* Do not expose this until vmm.ko implements it */
296 else if (strncmp(cp, "maxcpus=", strlen("maxcpus=")) == 0)
297 set_config_value("maxcpus", cp + strlen("maxcpus="));
299 else if (strchr(cp, '=') != NULL)
302 set_config_value("cpus", cp);
313 parse_int_value(const char *key, const char *value, int minval, int maxval)
319 lval = strtol(value, &cp, 0);
320 if (errno != 0 || *cp != '\0' || cp == value || lval < minval ||
322 errx(4, "Invalid value for %s: '%s'", key, value);
327 * Set the sockets, cores, threads, and guest_cpus variables based on
328 * the configured topology.
330 * The limits of UINT16_MAX are due to the types passed to
331 * vm_set_topology(). vmm.ko may enforce tighter limits.
340 value = get_config_value("cpus");
342 guest_ncpus = parse_int_value("cpus", value, 1, UINT16_MAX);
343 explicit_cpus = true;
346 explicit_cpus = false;
348 value = get_config_value("cores");
350 cores = parse_int_value("cores", value, 1, UINT16_MAX);
353 value = get_config_value("threads");
355 threads = parse_int_value("threads", value, 1, UINT16_MAX);
358 value = get_config_value("sockets");
360 sockets = parse_int_value("sockets", value, 1, UINT16_MAX);
362 sockets = guest_ncpus;
365 * Compute sockets * cores * threads avoiding overflow. The
366 * range check above insures these are 16 bit values.
368 ncpus = (uint64_t)sockets * cores * threads;
369 if (ncpus > UINT16_MAX)
370 errx(4, "Computed number of vCPUs too high: %ju",
374 if (guest_ncpus != ncpus)
375 errx(4, "Topology (%d sockets, %d cores, %d threads) "
376 "does not match %d vCPUs", sockets, cores, threads,
383 pincpu_parse(const char *opt)
390 if (sscanf(opt, "%d:%d", &vcpu, &pcpu) != 2) {
391 fprintf(stderr, "invalid format: %s\n", opt);
395 if (vcpu < 0 || vcpu >= VM_MAXCPU) {
396 fprintf(stderr, "vcpu '%d' outside valid range from 0 to %d\n",
397 vcpu, VM_MAXCPU - 1);
401 if (pcpu < 0 || pcpu >= CPU_SETSIZE) {
402 fprintf(stderr, "hostcpu '%d' outside valid range from "
403 "0 to %d\n", pcpu, CPU_SETSIZE - 1);
407 snprintf(key, sizeof(key), "vcpu.%d.cpuset", vcpu);
408 value = get_config_value(key);
410 if (asprintf(&newval, "%s%s%d", value != NULL ? value : "",
411 value != NULL ? "," : "", pcpu) == -1) {
412 perror("failed to build new cpuset string");
416 set_config_value(key, newval);
422 parse_cpuset(int vcpu, const char *list, cpuset_t *set)
429 token = __DECONST(char *, list);
431 pcpu = strtoul(token, &cp, 0);
433 errx(4, "invalid cpuset for vcpu %d: '%s'", vcpu, list);
434 if (pcpu < 0 || pcpu >= CPU_SETSIZE)
435 errx(4, "hostcpu '%d' outside valid range from 0 to %d",
436 pcpu, CPU_SETSIZE - 1);
442 errx(4, "Invalid hostcpu range %d-%d",
444 while (start < pcpu) {
445 CPU_SET(start, vcpumap[vcpu]);
450 CPU_SET(pcpu, vcpumap[vcpu]);
454 errx(4, "invalid cpuset for vcpu %d: '%s'",
459 errx(4, "invalid cpuset for vcpu %d: '%s'", vcpu, list);
474 for (vcpu = 0; vcpu < guest_ncpus; vcpu++) {
475 snprintf(key, sizeof(key), "vcpu.%d.cpuset", vcpu);
476 value = get_config_value(key);
479 vcpumap[vcpu] = malloc(sizeof(cpuset_t));
480 if (vcpumap[vcpu] == NULL)
481 err(4, "Failed to allocate cpuset for vcpu %d", vcpu);
482 parse_cpuset(vcpu, value, vcpumap[vcpu]);
487 vm_inject_fault(void *arg, int vcpu, int vector, int errcode_valid,
491 int error, restart_instruction;
494 restart_instruction = 1;
496 error = vm_inject_exception(ctx, vcpu, vector, errcode_valid, errcode,
497 restart_instruction);
502 paddr_guest2host(struct vmctx *ctx, uintptr_t gaddr, size_t len)
505 return (vm_map_gpa(ctx, gaddr, len));
508 #ifdef BHYVE_SNAPSHOT
510 paddr_host2guest(struct vmctx *ctx, void *addr)
512 return (vm_rev_map_gpa(ctx, addr));
517 fbsdrun_virtio_msix(void)
520 return (get_config_bool_default("virtio_msix", true));
524 fbsdrun_start_thread(void *param)
526 char tname[MAXCOMLEN + 1];
527 struct mt_vmm_info *mtp;
533 snprintf(tname, sizeof(tname), "vcpu %d", vcpu);
534 pthread_set_name_np(mtp->mt_thr, tname);
536 #ifdef BHYVE_SNAPSHOT
537 checkpoint_cpu_add(vcpu);
541 vm_loop(mtp->mt_ctx, vcpu, vmexit[vcpu].rip);
549 fbsdrun_addcpu(struct vmctx *ctx, int fromcpu, int newcpu, uint64_t rip)
553 assert(fromcpu == BSP);
556 * The 'newcpu' must be activated in the context of 'fromcpu'. If
557 * vm_activate_cpu() is delayed until newcpu's pthread starts running
558 * then vmm.ko is out-of-sync with bhyve and this can create a race
561 error = vm_activate_cpu(ctx, newcpu);
563 err(EX_OSERR, "could not activate CPU %d", newcpu);
565 CPU_SET_ATOMIC(newcpu, &cpumask);
568 * Set up the vmexit struct to allow execution to start
571 vmexit[newcpu].rip = rip;
572 vmexit[newcpu].inst_length = 0;
574 mt_vmm_info[newcpu].mt_ctx = ctx;
575 mt_vmm_info[newcpu].mt_vcpu = newcpu;
577 error = pthread_create(&mt_vmm_info[newcpu].mt_thr, NULL,
578 fbsdrun_start_thread, &mt_vmm_info[newcpu]);
583 fbsdrun_deletecpu(struct vmctx *ctx, int vcpu)
586 if (!CPU_ISSET(vcpu, &cpumask)) {
587 fprintf(stderr, "Attempting to delete unknown cpu %d\n", vcpu);
591 CPU_CLR_ATOMIC(vcpu, &cpumask);
592 return (CPU_EMPTY(&cpumask));
596 vmexit_handle_notify(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu,
601 * put guest-driven debug here
604 return (VMEXIT_CONTINUE);
608 vmexit_inout(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu)
611 int bytes, port, in, out;
616 port = vme->u.inout.port;
617 bytes = vme->u.inout.bytes;
618 in = vme->u.inout.in;
621 /* Extra-special case of host notifications */
622 if (out && port == GUEST_NIO_PORT) {
623 error = vmexit_handle_notify(ctx, vme, pvcpu, vme->u.inout.eax);
627 error = emulate_inout(ctx, vcpu, vme);
629 fprintf(stderr, "Unhandled %s%c 0x%04x at 0x%lx\n",
631 bytes == 1 ? 'b' : (bytes == 2 ? 'w' : 'l'),
633 return (VMEXIT_ABORT);
635 return (VMEXIT_CONTINUE);
640 vmexit_rdmsr(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu)
647 error = emulate_rdmsr(ctx, *pvcpu, vme->u.msr.code, &val);
649 fprintf(stderr, "rdmsr to register %#x on vcpu %d\n",
650 vme->u.msr.code, *pvcpu);
651 if (get_config_bool("x86.strictmsr")) {
652 vm_inject_gp(ctx, *pvcpu);
653 return (VMEXIT_CONTINUE);
658 error = vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RAX, eax);
662 error = vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RDX, edx);
665 return (VMEXIT_CONTINUE);
669 vmexit_wrmsr(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu)
673 error = emulate_wrmsr(ctx, *pvcpu, vme->u.msr.code, vme->u.msr.wval);
675 fprintf(stderr, "wrmsr to register %#x(%#lx) on vcpu %d\n",
676 vme->u.msr.code, vme->u.msr.wval, *pvcpu);
677 if (get_config_bool("x86.strictmsr")) {
678 vm_inject_gp(ctx, *pvcpu);
679 return (VMEXIT_CONTINUE);
682 return (VMEXIT_CONTINUE);
686 vmexit_spinup_ap(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu)
689 (void)spinup_ap(ctx, *pvcpu,
690 vme->u.spinup_ap.vcpu, vme->u.spinup_ap.rip);
692 return (VMEXIT_CONTINUE);
695 #define DEBUG_EPT_MISCONFIG
696 #ifdef DEBUG_EPT_MISCONFIG
697 #define VMCS_GUEST_PHYSICAL_ADDRESS 0x00002400
699 static uint64_t ept_misconfig_gpa, ept_misconfig_pte[4];
700 static int ept_misconfig_ptenum;
704 vmexit_vmx_desc(uint32_t exit_reason)
707 if (exit_reason >= nitems(vmx_exit_reason_desc) ||
708 vmx_exit_reason_desc[exit_reason] == NULL)
710 return (vmx_exit_reason_desc[exit_reason]);
714 vmexit_vmx(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
717 fprintf(stderr, "vm exit[%d]\n", *pvcpu);
718 fprintf(stderr, "\treason\t\tVMX\n");
719 fprintf(stderr, "\trip\t\t0x%016lx\n", vmexit->rip);
720 fprintf(stderr, "\tinst_length\t%d\n", vmexit->inst_length);
721 fprintf(stderr, "\tstatus\t\t%d\n", vmexit->u.vmx.status);
722 fprintf(stderr, "\texit_reason\t%u (%s)\n", vmexit->u.vmx.exit_reason,
723 vmexit_vmx_desc(vmexit->u.vmx.exit_reason));
724 fprintf(stderr, "\tqualification\t0x%016lx\n",
725 vmexit->u.vmx.exit_qualification);
726 fprintf(stderr, "\tinst_type\t\t%d\n", vmexit->u.vmx.inst_type);
727 fprintf(stderr, "\tinst_error\t\t%d\n", vmexit->u.vmx.inst_error);
728 #ifdef DEBUG_EPT_MISCONFIG
729 if (vmexit->u.vmx.exit_reason == EXIT_REASON_EPT_MISCONFIG) {
730 vm_get_register(ctx, *pvcpu,
731 VMCS_IDENT(VMCS_GUEST_PHYSICAL_ADDRESS),
733 vm_get_gpa_pmap(ctx, ept_misconfig_gpa, ept_misconfig_pte,
734 &ept_misconfig_ptenum);
735 fprintf(stderr, "\tEPT misconfiguration:\n");
736 fprintf(stderr, "\t\tGPA: %#lx\n", ept_misconfig_gpa);
737 fprintf(stderr, "\t\tPTE(%d): %#lx %#lx %#lx %#lx\n",
738 ept_misconfig_ptenum, ept_misconfig_pte[0],
739 ept_misconfig_pte[1], ept_misconfig_pte[2],
740 ept_misconfig_pte[3]);
742 #endif /* DEBUG_EPT_MISCONFIG */
743 return (VMEXIT_ABORT);
747 vmexit_svm(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
750 fprintf(stderr, "vm exit[%d]\n", *pvcpu);
751 fprintf(stderr, "\treason\t\tSVM\n");
752 fprintf(stderr, "\trip\t\t0x%016lx\n", vmexit->rip);
753 fprintf(stderr, "\tinst_length\t%d\n", vmexit->inst_length);
754 fprintf(stderr, "\texitcode\t%#lx\n", vmexit->u.svm.exitcode);
755 fprintf(stderr, "\texitinfo1\t%#lx\n", vmexit->u.svm.exitinfo1);
756 fprintf(stderr, "\texitinfo2\t%#lx\n", vmexit->u.svm.exitinfo2);
757 return (VMEXIT_ABORT);
761 vmexit_bogus(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
764 assert(vmexit->inst_length == 0);
766 stats.vmexit_bogus++;
768 return (VMEXIT_CONTINUE);
772 vmexit_reqidle(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
775 assert(vmexit->inst_length == 0);
777 stats.vmexit_reqidle++;
779 return (VMEXIT_CONTINUE);
783 vmexit_hlt(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
789 * Just continue execution with the next instruction. We use
790 * the HLT VM exit as a way to be friendly with the host
793 return (VMEXIT_CONTINUE);
797 vmexit_pause(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
800 stats.vmexit_pause++;
802 return (VMEXIT_CONTINUE);
806 vmexit_mtrap(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
809 assert(vmexit->inst_length == 0);
811 stats.vmexit_mtrap++;
813 #ifdef BHYVE_SNAPSHOT
814 checkpoint_cpu_suspend(*pvcpu);
816 gdb_cpu_mtrap(*pvcpu);
817 #ifdef BHYVE_SNAPSHOT
818 checkpoint_cpu_resume(*pvcpu);
821 return (VMEXIT_CONTINUE);
825 vmexit_inst_emul(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
829 enum vm_cpu_mode mode;
831 stats.vmexit_inst_emul++;
833 vie = &vmexit->u.inst_emul.vie;
836 * Attempt to decode in userspace as a fallback. This allows
837 * updating instruction decode in bhyve without rebooting the
838 * kernel (rapid prototyping), albeit with much slower
842 mode = vmexit->u.inst_emul.paging.cpu_mode;
843 cs_d = vmexit->u.inst_emul.cs_d;
844 if (vmm_decode_instruction(mode, cs_d, vie) != 0)
846 if (vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RIP,
847 vmexit->rip + vie->num_processed) != 0)
851 err = emulate_mem(ctx, *pvcpu, vmexit->u.inst_emul.gpa,
852 vie, &vmexit->u.inst_emul.paging);
856 EPRINTLN("Unhandled memory access to 0x%lx\n",
857 vmexit->u.inst_emul.gpa);
862 return (VMEXIT_CONTINUE);
865 fprintf(stderr, "Failed to emulate instruction sequence [ ");
866 for (i = 0; i < vie->num_valid; i++)
867 fprintf(stderr, "%02x", vie->inst[i]);
868 FPRINTLN(stderr, " ] at 0x%lx", vmexit->rip);
869 return (VMEXIT_ABORT);
872 static pthread_mutex_t resetcpu_mtx = PTHREAD_MUTEX_INITIALIZER;
873 static pthread_cond_t resetcpu_cond = PTHREAD_COND_INITIALIZER;
876 vmexit_suspend(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
878 enum vm_suspend_how how;
880 how = vmexit->u.suspended.how;
882 fbsdrun_deletecpu(ctx, *pvcpu);
885 pthread_mutex_lock(&resetcpu_mtx);
886 pthread_cond_signal(&resetcpu_cond);
887 pthread_mutex_unlock(&resetcpu_mtx);
891 pthread_mutex_lock(&resetcpu_mtx);
892 while (!CPU_EMPTY(&cpumask)) {
893 pthread_cond_wait(&resetcpu_cond, &resetcpu_mtx);
895 pthread_mutex_unlock(&resetcpu_mtx);
898 case VM_SUSPEND_RESET:
900 case VM_SUSPEND_POWEROFF:
901 if (get_config_bool_default("destroy_on_poweroff", false))
904 case VM_SUSPEND_HALT:
906 case VM_SUSPEND_TRIPLEFAULT:
909 fprintf(stderr, "vmexit_suspend: invalid reason %d\n", how);
912 return (0); /* NOTREACHED */
916 vmexit_debug(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
919 #ifdef BHYVE_SNAPSHOT
920 checkpoint_cpu_suspend(*pvcpu);
922 gdb_cpu_suspend(*pvcpu);
923 #ifdef BHYVE_SNAPSHOT
924 checkpoint_cpu_resume(*pvcpu);
926 return (VMEXIT_CONTINUE);
930 vmexit_breakpoint(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
933 gdb_cpu_breakpoint(*pvcpu, vmexit);
934 return (VMEXIT_CONTINUE);
937 static vmexit_handler_t handler[VM_EXITCODE_MAX] = {
938 [VM_EXITCODE_INOUT] = vmexit_inout,
939 [VM_EXITCODE_INOUT_STR] = vmexit_inout,
940 [VM_EXITCODE_VMX] = vmexit_vmx,
941 [VM_EXITCODE_SVM] = vmexit_svm,
942 [VM_EXITCODE_BOGUS] = vmexit_bogus,
943 [VM_EXITCODE_REQIDLE] = vmexit_reqidle,
944 [VM_EXITCODE_RDMSR] = vmexit_rdmsr,
945 [VM_EXITCODE_WRMSR] = vmexit_wrmsr,
946 [VM_EXITCODE_MTRAP] = vmexit_mtrap,
947 [VM_EXITCODE_INST_EMUL] = vmexit_inst_emul,
948 [VM_EXITCODE_SPINUP_AP] = vmexit_spinup_ap,
949 [VM_EXITCODE_SUSPENDED] = vmexit_suspend,
950 [VM_EXITCODE_TASK_SWITCH] = vmexit_task_switch,
951 [VM_EXITCODE_DEBUG] = vmexit_debug,
952 [VM_EXITCODE_BPT] = vmexit_breakpoint,
956 vm_loop(struct vmctx *ctx, int vcpu, uint64_t startrip)
959 enum vm_exitcode exitcode;
960 cpuset_t active_cpus;
962 if (vcpumap[vcpu] != NULL) {
963 error = pthread_setaffinity_np(pthread_self(),
964 sizeof(cpuset_t), vcpumap[vcpu]);
968 error = vm_active_cpus(ctx, &active_cpus);
969 assert(CPU_ISSET(vcpu, &active_cpus));
971 error = vm_set_register(ctx, vcpu, VM_REG_GUEST_RIP, startrip);
975 error = vm_run(ctx, vcpu, &vmexit[vcpu]);
979 exitcode = vmexit[vcpu].exitcode;
980 if (exitcode >= VM_EXITCODE_MAX || handler[exitcode] == NULL) {
981 fprintf(stderr, "vm_loop: unexpected exitcode 0x%x\n",
986 rc = (*handler[exitcode])(ctx, &vmexit[vcpu], &vcpu);
989 case VMEXIT_CONTINUE:
997 fprintf(stderr, "vm_run error %d, errno %d\n", error, errno);
1001 num_vcpus_allowed(struct vmctx *ctx)
1005 error = vm_get_capability(ctx, BSP, VM_CAP_UNRESTRICTED_GUEST, &tmp);
1008 * The guest is allowed to spinup more than one processor only if the
1009 * UNRESTRICTED_GUEST capability is available.
1018 fbsdrun_set_capabilities(struct vmctx *ctx, int cpu)
1022 if (get_config_bool_default("x86.vmexit_on_hlt", false)) {
1023 err = vm_get_capability(ctx, cpu, VM_CAP_HALT_EXIT, &tmp);
1025 fprintf(stderr, "VM exit on HLT not supported\n");
1028 vm_set_capability(ctx, cpu, VM_CAP_HALT_EXIT, 1);
1030 handler[VM_EXITCODE_HLT] = vmexit_hlt;
1033 if (get_config_bool_default("x86.vmexit_on_pause", false)) {
1035 * pause exit support required for this mode
1037 err = vm_get_capability(ctx, cpu, VM_CAP_PAUSE_EXIT, &tmp);
1040 "SMP mux requested, no pause support\n");
1043 vm_set_capability(ctx, cpu, VM_CAP_PAUSE_EXIT, 1);
1045 handler[VM_EXITCODE_PAUSE] = vmexit_pause;
1048 if (get_config_bool_default("x86.x2apic", false))
1049 err = vm_set_x2apic_state(ctx, cpu, X2APIC_ENABLED);
1051 err = vm_set_x2apic_state(ctx, cpu, X2APIC_DISABLED);
1054 fprintf(stderr, "Unable to set x2apic state (%d)\n", err);
1058 vm_set_capability(ctx, cpu, VM_CAP_ENABLE_INVPCID, 1);
1061 static struct vmctx *
1062 do_open(const char *vmname)
1066 bool reinit, romboot;
1067 #ifndef WITHOUT_CAPSICUM
1068 cap_rights_t rights;
1069 const cap_ioctl_t *cmds;
1073 reinit = romboot = false;
1078 error = vm_create(vmname);
1080 if (errno == EEXIST) {
1085 * The virtual machine has been setup by the
1086 * userspace bootloader.
1090 perror("vm_create");
1096 * If the virtual machine was just created then a
1097 * bootrom must be configured to boot it.
1099 fprintf(stderr, "virtual machine cannot be booted\n");
1104 ctx = vm_open(vmname);
1110 #ifndef WITHOUT_CAPSICUM
1111 cap_rights_init(&rights, CAP_IOCTL, CAP_MMAP_RW);
1112 if (caph_rights_limit(vm_get_device_fd(ctx), &rights) == -1)
1113 errx(EX_OSERR, "Unable to apply rights for sandbox");
1114 vm_get_ioctls(&ncmds);
1115 cmds = vm_get_ioctls(NULL);
1117 errx(EX_OSERR, "out of memory");
1118 if (caph_ioctls_limit(vm_get_device_fd(ctx), cmds, ncmds) == -1)
1119 errx(EX_OSERR, "Unable to apply rights for sandbox");
1120 free((cap_ioctl_t *)cmds);
1124 error = vm_reinit(ctx);
1126 perror("vm_reinit");
1130 error = vm_set_topology(ctx, sockets, cores, threads, maxcpus);
1132 errx(EX_OSERR, "vm_set_topology");
1137 spinup_vcpu(struct vmctx *ctx, int vcpu)
1142 error = vm_get_register(ctx, vcpu, VM_REG_GUEST_RIP, &rip);
1145 fbsdrun_set_capabilities(ctx, vcpu);
1146 error = vm_set_capability(ctx, vcpu, VM_CAP_UNRESTRICTED_GUEST, 1);
1149 fbsdrun_addcpu(ctx, BSP, vcpu, rip);
1153 parse_config_option(const char *option)
1158 value = strchr(option, '=');
1159 if (value == NULL || value[1] == '\0')
1161 path = strndup(option, value - option);
1163 err(4, "Failed to allocate memory");
1164 set_config_value(path, value + 1);
1169 parse_simple_config_file(const char *path)
1174 unsigned int lineno;
1176 fp = fopen(path, "r");
1178 err(4, "Failed to open configuration file %s", path);
1182 for (lineno = 1; getline(&line, &linecap, fp) > 0; lineno++) {
1183 if (*line == '#' || *line == '\n')
1185 cp = strchr(line, '\n');
1188 if (!parse_config_option(line))
1189 errx(4, "%s line %u: invalid config option '%s'", path,
1197 parse_gdb_options(char *optarg)
1202 if (optarg[0] == 'w') {
1203 set_config_bool("gdb.wait", true);
1207 colon = strrchr(optarg, ':');
1208 if (colon == NULL) {
1214 set_config_value("gdb.address", optarg);
1217 set_config_value("gdb.port", sport);
1224 set_config_bool("acpi_tables", false);
1225 set_config_value("memory.size", "256M");
1226 set_config_bool("x86.strictmsr", true);
1230 main(int argc, char *argv[])
1233 int max_vcpus, memflags;
1237 const char *value, *vmname;
1239 #ifdef BHYVE_SNAPSHOT
1241 struct restore_state rstate;
1244 restore_file = NULL;
1249 progname = basename(argv[0]);
1251 #ifdef BHYVE_SNAPSHOT
1252 optstr = "aehuwxACDHIPSWYk:o:p:G:c:s:m:l:U:r:";
1254 optstr = "aehuwxACDHIPSWYk:o:p:G:c:s:m:l:U:";
1256 while ((c = getopt(argc, argv, optstr)) != -1) {
1259 set_config_bool("x86.x2apic", false);
1262 set_config_bool("acpi_tables", true);
1265 set_config_bool("destroy_on_poweroff", true);
1268 if (pincpu_parse(optarg) != 0) {
1269 errx(EX_USAGE, "invalid vcpu pinning "
1270 "configuration '%s'", optarg);
1274 if (topology_parse(optarg) != 0) {
1275 errx(EX_USAGE, "invalid cpu topology "
1280 set_config_bool("memory.guest_in_core", true);
1283 parse_gdb_options(optarg);
1286 parse_simple_config_file(optarg);
1289 if (strncmp(optarg, "help", strlen(optarg)) == 0) {
1290 lpc_print_supported_devices();
1292 } else if (lpc_device_parse(optarg) != 0) {
1293 errx(EX_USAGE, "invalid lpc device "
1294 "configuration '%s'", optarg);
1297 #ifdef BHYVE_SNAPSHOT
1299 restore_file = optarg;
1303 if (strncmp(optarg, "help", strlen(optarg)) == 0) {
1304 pci_print_supported_devices();
1306 } else if (pci_parse_slot(optarg) != 0)
1311 set_config_bool("memory.wired", true);
1314 set_config_value("memory.size", optarg);
1317 if (!parse_config_option(optarg))
1318 errx(EX_USAGE, "invalid configuration option '%s'", optarg);
1321 set_config_bool("x86.vmexit_on_hlt", true);
1325 * The "-I" option was used to add an ioapic to the
1328 * An ioapic is now provided unconditionally for each
1329 * virtual machine and this option is now deprecated.
1333 set_config_bool("x86.vmexit_on_pause", true);
1336 set_config_bool("x86.strictio", true);
1339 set_config_bool("rtc.use_localtime", false);
1342 set_config_value("uuid", optarg);
1345 set_config_bool("x86.strictmsr", false);
1348 set_config_bool("virtio_msix", false);
1351 set_config_bool("x86.x2apic", true);
1354 set_config_bool("x86.mptable", false);
1368 #ifdef BHYVE_SNAPSHOT
1369 if (restore_file != NULL) {
1370 error = load_restore_file(restore_file, &rstate);
1372 fprintf(stderr, "Failed to read checkpoint info from "
1373 "file: '%s'.\n", restore_file);
1376 vmname = lookup_vmname(&rstate);
1378 set_config_value("name", vmname);
1383 set_config_value("name", argv[0]);
1385 vmname = get_config_value("name");
1389 if (get_config_bool_default("config.dump", false)) {
1397 value = get_config_value("memory.size");
1398 error = vm_parse_memsize(value, &memsize);
1400 errx(EX_USAGE, "invalid memsize '%s'", value);
1402 ctx = do_open(vmname);
1404 #ifdef BHYVE_SNAPSHOT
1405 if (restore_file != NULL) {
1406 guest_ncpus = lookup_guest_ncpus(&rstate);
1407 memflags = lookup_memflags(&rstate);
1408 memsize = lookup_memsize(&rstate);
1411 if (guest_ncpus < 1) {
1412 fprintf(stderr, "Invalid guest vCPUs (%d)\n", guest_ncpus);
1417 max_vcpus = num_vcpus_allowed(ctx);
1418 if (guest_ncpus > max_vcpus) {
1419 fprintf(stderr, "%d vCPUs requested but only %d available\n",
1420 guest_ncpus, max_vcpus);
1424 fbsdrun_set_capabilities(ctx, BSP);
1427 if (get_config_bool_default("memory.wired", false))
1428 memflags |= VM_MEM_F_WIRED;
1429 if (get_config_bool_default("memory.guest_in_core", false))
1430 memflags |= VM_MEM_F_INCORE;
1431 vm_set_memflags(ctx, memflags);
1432 err = vm_setup_memory(ctx, memsize, VM_MMAP_ALL);
1434 fprintf(stderr, "Unable to setup memory (%d)\n", errno);
1440 fprintf(stderr, "init_msr error %d", error);
1456 * Exit if a device emulation finds an error in its initilization
1458 if (init_pci(ctx) != 0) {
1459 perror("device emulation initialization error");
1464 * Initialize after PCI, to allow a bootrom file to reserve the high
1467 if (get_config_bool("acpi_tables"))
1472 if (lpc_bootrom()) {
1473 if (vm_set_capability(ctx, BSP, VM_CAP_UNRESTRICTED_GUEST, 1)) {
1474 fprintf(stderr, "ROM boot failed: unrestricted guest "
1475 "capability not available\n");
1478 error = vcpu_reset(ctx, BSP);
1482 #ifdef BHYVE_SNAPSHOT
1483 if (restore_file != NULL) {
1484 fprintf(stdout, "Pausing pci devs...\r\n");
1485 if (vm_pause_user_devs(ctx) != 0) {
1486 fprintf(stderr, "Failed to pause PCI device state.\n");
1490 fprintf(stdout, "Restoring vm mem...\r\n");
1491 if (restore_vm_mem(ctx, &rstate) != 0) {
1492 fprintf(stderr, "Failed to restore VM memory.\n");
1496 fprintf(stdout, "Restoring pci devs...\r\n");
1497 if (vm_restore_user_devs(ctx, &rstate) != 0) {
1498 fprintf(stderr, "Failed to restore PCI device state.\n");
1502 fprintf(stdout, "Restoring kernel structs...\r\n");
1503 if (vm_restore_kern_structs(ctx, &rstate) != 0) {
1504 fprintf(stderr, "Failed to restore kernel structs.\n");
1508 fprintf(stdout, "Resuming pci devs...\r\n");
1509 if (vm_resume_user_devs(ctx) != 0) {
1510 fprintf(stderr, "Failed to resume PCI device state.\n");
1516 error = vm_get_register(ctx, BSP, VM_REG_GUEST_RIP, &rip);
1520 * build the guest tables, MP etc.
1522 if (get_config_bool_default("x86.mptable", true)) {
1523 error = mptable_build(ctx, guest_ncpus);
1525 perror("error to build the guest tables");
1530 error = smbios_build(ctx);
1533 if (get_config_bool("acpi_tables")) {
1534 error = acpi_build(ctx, guest_ncpus);
1542 * Change the proc title to include the VM name.
1544 setproctitle("%s", vmname);
1546 #ifndef WITHOUT_CAPSICUM
1547 caph_cache_catpages();
1549 if (caph_limit_stdout() == -1 || caph_limit_stderr() == -1)
1550 errx(EX_OSERR, "Unable to apply rights for sandbox");
1552 if (caph_enter() == -1)
1553 errx(EX_OSERR, "cap_enter() failed");
1556 #ifdef BHYVE_SNAPSHOT
1557 if (restore_file != NULL)
1558 destroy_restore_state(&rstate);
1560 /* initialize mutex/cond variables */
1564 * checkpointing thread for communication with bhyvectl
1566 if (init_checkpoint_thread(ctx) < 0)
1567 printf("Failed to start checkpoint thread!\r\n");
1569 if (restore_file != NULL)
1570 vm_restore_time(ctx);
1576 fbsdrun_addcpu(ctx, BSP, BSP, rip);
1578 #ifdef BHYVE_SNAPSHOT
1580 * If we restore a VM, start all vCPUs now (including APs), otherwise,
1581 * let the guest OS to spin them up later via vmexits.
1583 if (restore_file != NULL) {
1584 for (vcpu = 0; vcpu < guest_ncpus; vcpu++) {
1588 fprintf(stdout, "spinning up vcpu no %d...\r\n", vcpu);
1589 spinup_vcpu(ctx, vcpu);
1595 * Head off to the main event dispatch loop