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;
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;
218 static cpuset_t **vcpumap;
225 "Usage: %s [-AaCDeHhPSuWwxY]\n"
226 " %*s [-c [[cpus=]numcpus][,sockets=n][,cores=n][,threads=n]]\n"
227 " %*s [-G port] [-k config_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 " -K: PS2 keyboard layout\n"
240 " -l: LPC device configuration\n"
242 " -o: set config 'var' to 'value'\n"
243 " -P: vmexit from the guest on pause\n"
244 " -p: pin 'vcpu' to 'hostcpu'\n"
245 #ifdef BHYVE_SNAPSHOT
246 " -r: path to checkpoint file\n"
248 " -S: guest memory cannot be swapped\n"
249 " -s: <slot,driver,configinfo> PCI slot config\n"
251 " -u: RTC keeps UTC time\n"
252 " -W: force virtio to use single-vector MSI\n"
253 " -w: ignore unimplemented MSRs\n"
254 " -x: local APIC is in x2APIC mode\n"
255 " -Y: disable MPtable generation\n",
256 progname, (int)strlen(progname), "", (int)strlen(progname), "",
257 (int)strlen(progname), "");
263 * XXX This parser is known to have the following issues:
264 * 1. It accepts null key=value tokens ",," as setting "cpus" to an
267 * The acceptance of a null specification ('-c ""') is by design to match the
268 * manual page syntax specification, this results in a topology of 1 vCPU.
271 topology_parse(const char *opt)
273 char *cp, *str, *tofree;
276 set_config_value("sockets", "1");
277 set_config_value("cores", "1");
278 set_config_value("threads", "1");
279 set_config_value("cpus", "1");
283 tofree = str = strdup(opt);
285 errx(4, "Failed to allocate memory");
287 while ((cp = strsep(&str, ",")) != NULL) {
288 if (strncmp(cp, "cpus=", strlen("cpus=")) == 0)
289 set_config_value("cpus", cp + strlen("cpus="));
290 else if (strncmp(cp, "sockets=", strlen("sockets=")) == 0)
291 set_config_value("sockets", cp + strlen("sockets="));
292 else if (strncmp(cp, "cores=", strlen("cores=")) == 0)
293 set_config_value("cores", cp + strlen("cores="));
294 else if (strncmp(cp, "threads=", strlen("threads=")) == 0)
295 set_config_value("threads", cp + strlen("threads="));
296 #ifdef notyet /* Do not expose this until vmm.ko implements it */
297 else if (strncmp(cp, "maxcpus=", strlen("maxcpus=")) == 0)
298 set_config_value("maxcpus", cp + strlen("maxcpus="));
300 else if (strchr(cp, '=') != NULL)
303 set_config_value("cpus", cp);
314 parse_int_value(const char *key, const char *value, int minval, int maxval)
320 lval = strtol(value, &cp, 0);
321 if (errno != 0 || *cp != '\0' || cp == value || lval < minval ||
323 errx(4, "Invalid value for %s: '%s'", key, value);
328 * Set the sockets, cores, threads, and guest_cpus variables based on
329 * the configured topology.
331 * The limits of UINT16_MAX are due to the types passed to
332 * vm_set_topology(). vmm.ko may enforce tighter limits.
341 value = get_config_value("cpus");
343 guest_ncpus = parse_int_value("cpus", value, 1, UINT16_MAX);
344 explicit_cpus = true;
347 explicit_cpus = false;
349 value = get_config_value("cores");
351 cores = parse_int_value("cores", value, 1, UINT16_MAX);
354 value = get_config_value("threads");
356 threads = parse_int_value("threads", value, 1, UINT16_MAX);
359 value = get_config_value("sockets");
361 sockets = parse_int_value("sockets", value, 1, UINT16_MAX);
363 sockets = guest_ncpus;
366 * Compute sockets * cores * threads avoiding overflow. The
367 * range check above insures these are 16 bit values.
369 ncpus = (uint64_t)sockets * cores * threads;
370 if (ncpus > UINT16_MAX)
371 errx(4, "Computed number of vCPUs too high: %ju",
375 if (guest_ncpus != ncpus)
376 errx(4, "Topology (%d sockets, %d cores, %d threads) "
377 "does not match %d vCPUs", sockets, cores, threads,
384 pincpu_parse(const char *opt)
391 if (sscanf(opt, "%d:%d", &vcpu, &pcpu) != 2) {
392 fprintf(stderr, "invalid format: %s\n", opt);
397 fprintf(stderr, "invalid vcpu '%d'\n", vcpu);
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 vcpumap = calloc(guest_ncpus, sizeof(*vcpumap));
475 for (vcpu = 0; vcpu < guest_ncpus; vcpu++) {
476 snprintf(key, sizeof(key), "vcpu.%d.cpuset", vcpu);
477 value = get_config_value(key);
480 vcpumap[vcpu] = malloc(sizeof(cpuset_t));
481 if (vcpumap[vcpu] == NULL)
482 err(4, "Failed to allocate cpuset for vcpu %d", vcpu);
483 parse_cpuset(vcpu, value, vcpumap[vcpu]);
488 vm_inject_fault(void *arg, int vcpu, int vector, int errcode_valid,
492 int error, restart_instruction;
495 restart_instruction = 1;
497 error = vm_inject_exception(ctx, vcpu, vector, errcode_valid, errcode,
498 restart_instruction);
503 paddr_guest2host(struct vmctx *ctx, uintptr_t gaddr, size_t len)
506 return (vm_map_gpa(ctx, gaddr, len));
509 #ifdef BHYVE_SNAPSHOT
511 paddr_host2guest(struct vmctx *ctx, void *addr)
513 return (vm_rev_map_gpa(ctx, addr));
518 fbsdrun_virtio_msix(void)
521 return (get_config_bool_default("virtio_msix", true));
525 fbsdrun_start_thread(void *param)
527 char tname[MAXCOMLEN + 1];
528 struct mt_vmm_info *mtp;
534 snprintf(tname, sizeof(tname), "vcpu %d", vcpu);
535 pthread_set_name_np(mtp->mt_thr, tname);
537 #ifdef BHYVE_SNAPSHOT
538 checkpoint_cpu_add(vcpu);
542 vm_loop(mtp->mt_ctx, vcpu, vmexit[vcpu].rip);
550 fbsdrun_addcpu(struct vmctx *ctx, int fromcpu, int newcpu, uint64_t rip)
554 assert(fromcpu == BSP);
557 * The 'newcpu' must be activated in the context of 'fromcpu'. If
558 * vm_activate_cpu() is delayed until newcpu's pthread starts running
559 * then vmm.ko is out-of-sync with bhyve and this can create a race
562 error = vm_activate_cpu(ctx, newcpu);
564 err(EX_OSERR, "could not activate CPU %d", newcpu);
566 CPU_SET_ATOMIC(newcpu, &cpumask);
569 * Set up the vmexit struct to allow execution to start
572 vmexit[newcpu].rip = rip;
573 vmexit[newcpu].inst_length = 0;
575 mt_vmm_info[newcpu].mt_ctx = ctx;
576 mt_vmm_info[newcpu].mt_vcpu = newcpu;
578 error = pthread_create(&mt_vmm_info[newcpu].mt_thr, NULL,
579 fbsdrun_start_thread, &mt_vmm_info[newcpu]);
584 fbsdrun_deletecpu(struct vmctx *ctx, int vcpu)
587 if (!CPU_ISSET(vcpu, &cpumask)) {
588 fprintf(stderr, "Attempting to delete unknown cpu %d\n", vcpu);
592 CPU_CLR_ATOMIC(vcpu, &cpumask);
593 return (CPU_EMPTY(&cpumask));
597 vmexit_handle_notify(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu,
602 * put guest-driven debug here
605 return (VMEXIT_CONTINUE);
609 vmexit_inout(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu)
612 int bytes, port, in, out;
617 port = vme->u.inout.port;
618 bytes = vme->u.inout.bytes;
619 in = vme->u.inout.in;
622 /* Extra-special case of host notifications */
623 if (out && port == GUEST_NIO_PORT) {
624 error = vmexit_handle_notify(ctx, vme, pvcpu, vme->u.inout.eax);
628 error = emulate_inout(ctx, vcpu, vme);
630 fprintf(stderr, "Unhandled %s%c 0x%04x at 0x%lx\n",
632 bytes == 1 ? 'b' : (bytes == 2 ? 'w' : 'l'),
634 return (VMEXIT_ABORT);
636 return (VMEXIT_CONTINUE);
641 vmexit_rdmsr(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu)
648 error = emulate_rdmsr(ctx, *pvcpu, vme->u.msr.code, &val);
650 fprintf(stderr, "rdmsr to register %#x on vcpu %d\n",
651 vme->u.msr.code, *pvcpu);
652 if (get_config_bool("x86.strictmsr")) {
653 vm_inject_gp(ctx, *pvcpu);
654 return (VMEXIT_CONTINUE);
659 error = vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RAX, eax);
663 error = vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RDX, edx);
666 return (VMEXIT_CONTINUE);
670 vmexit_wrmsr(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu)
674 error = emulate_wrmsr(ctx, *pvcpu, vme->u.msr.code, vme->u.msr.wval);
676 fprintf(stderr, "wrmsr to register %#x(%#lx) on vcpu %d\n",
677 vme->u.msr.code, vme->u.msr.wval, *pvcpu);
678 if (get_config_bool("x86.strictmsr")) {
679 vm_inject_gp(ctx, *pvcpu);
680 return (VMEXIT_CONTINUE);
683 return (VMEXIT_CONTINUE);
687 vmexit_spinup_ap(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu)
690 (void)spinup_ap(ctx, *pvcpu,
691 vme->u.spinup_ap.vcpu, vme->u.spinup_ap.rip);
693 return (VMEXIT_CONTINUE);
696 #define DEBUG_EPT_MISCONFIG
697 #ifdef DEBUG_EPT_MISCONFIG
698 #define VMCS_GUEST_PHYSICAL_ADDRESS 0x00002400
700 static uint64_t ept_misconfig_gpa, ept_misconfig_pte[4];
701 static int ept_misconfig_ptenum;
705 vmexit_vmx_desc(uint32_t exit_reason)
708 if (exit_reason >= nitems(vmx_exit_reason_desc) ||
709 vmx_exit_reason_desc[exit_reason] == NULL)
711 return (vmx_exit_reason_desc[exit_reason]);
715 vmexit_vmx(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
718 fprintf(stderr, "vm exit[%d]\n", *pvcpu);
719 fprintf(stderr, "\treason\t\tVMX\n");
720 fprintf(stderr, "\trip\t\t0x%016lx\n", vmexit->rip);
721 fprintf(stderr, "\tinst_length\t%d\n", vmexit->inst_length);
722 fprintf(stderr, "\tstatus\t\t%d\n", vmexit->u.vmx.status);
723 fprintf(stderr, "\texit_reason\t%u (%s)\n", vmexit->u.vmx.exit_reason,
724 vmexit_vmx_desc(vmexit->u.vmx.exit_reason));
725 fprintf(stderr, "\tqualification\t0x%016lx\n",
726 vmexit->u.vmx.exit_qualification);
727 fprintf(stderr, "\tinst_type\t\t%d\n", vmexit->u.vmx.inst_type);
728 fprintf(stderr, "\tinst_error\t\t%d\n", vmexit->u.vmx.inst_error);
729 #ifdef DEBUG_EPT_MISCONFIG
730 if (vmexit->u.vmx.exit_reason == EXIT_REASON_EPT_MISCONFIG) {
731 vm_get_register(ctx, *pvcpu,
732 VMCS_IDENT(VMCS_GUEST_PHYSICAL_ADDRESS),
734 vm_get_gpa_pmap(ctx, ept_misconfig_gpa, ept_misconfig_pte,
735 &ept_misconfig_ptenum);
736 fprintf(stderr, "\tEPT misconfiguration:\n");
737 fprintf(stderr, "\t\tGPA: %#lx\n", ept_misconfig_gpa);
738 fprintf(stderr, "\t\tPTE(%d): %#lx %#lx %#lx %#lx\n",
739 ept_misconfig_ptenum, ept_misconfig_pte[0],
740 ept_misconfig_pte[1], ept_misconfig_pte[2],
741 ept_misconfig_pte[3]);
743 #endif /* DEBUG_EPT_MISCONFIG */
744 return (VMEXIT_ABORT);
748 vmexit_svm(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
751 fprintf(stderr, "vm exit[%d]\n", *pvcpu);
752 fprintf(stderr, "\treason\t\tSVM\n");
753 fprintf(stderr, "\trip\t\t0x%016lx\n", vmexit->rip);
754 fprintf(stderr, "\tinst_length\t%d\n", vmexit->inst_length);
755 fprintf(stderr, "\texitcode\t%#lx\n", vmexit->u.svm.exitcode);
756 fprintf(stderr, "\texitinfo1\t%#lx\n", vmexit->u.svm.exitinfo1);
757 fprintf(stderr, "\texitinfo2\t%#lx\n", vmexit->u.svm.exitinfo2);
758 return (VMEXIT_ABORT);
762 vmexit_bogus(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
765 assert(vmexit->inst_length == 0);
767 stats.vmexit_bogus++;
769 return (VMEXIT_CONTINUE);
773 vmexit_reqidle(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
776 assert(vmexit->inst_length == 0);
778 stats.vmexit_reqidle++;
780 return (VMEXIT_CONTINUE);
784 vmexit_hlt(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
790 * Just continue execution with the next instruction. We use
791 * the HLT VM exit as a way to be friendly with the host
794 return (VMEXIT_CONTINUE);
798 vmexit_pause(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
801 stats.vmexit_pause++;
803 return (VMEXIT_CONTINUE);
807 vmexit_mtrap(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
810 assert(vmexit->inst_length == 0);
812 stats.vmexit_mtrap++;
814 #ifdef BHYVE_SNAPSHOT
815 checkpoint_cpu_suspend(*pvcpu);
817 gdb_cpu_mtrap(*pvcpu);
818 #ifdef BHYVE_SNAPSHOT
819 checkpoint_cpu_resume(*pvcpu);
822 return (VMEXIT_CONTINUE);
826 vmexit_inst_emul(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
830 enum vm_cpu_mode mode;
832 stats.vmexit_inst_emul++;
834 vie = &vmexit->u.inst_emul.vie;
837 * Attempt to decode in userspace as a fallback. This allows
838 * updating instruction decode in bhyve without rebooting the
839 * kernel (rapid prototyping), albeit with much slower
843 mode = vmexit->u.inst_emul.paging.cpu_mode;
844 cs_d = vmexit->u.inst_emul.cs_d;
845 if (vmm_decode_instruction(mode, cs_d, vie) != 0)
847 if (vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RIP,
848 vmexit->rip + vie->num_processed) != 0)
852 err = emulate_mem(ctx, *pvcpu, vmexit->u.inst_emul.gpa,
853 vie, &vmexit->u.inst_emul.paging);
857 EPRINTLN("Unhandled memory access to 0x%lx\n",
858 vmexit->u.inst_emul.gpa);
863 return (VMEXIT_CONTINUE);
866 fprintf(stderr, "Failed to emulate instruction sequence [ ");
867 for (i = 0; i < vie->num_valid; i++)
868 fprintf(stderr, "%02x", vie->inst[i]);
869 FPRINTLN(stderr, " ] at 0x%lx", vmexit->rip);
870 return (VMEXIT_ABORT);
873 static pthread_mutex_t resetcpu_mtx = PTHREAD_MUTEX_INITIALIZER;
874 static pthread_cond_t resetcpu_cond = PTHREAD_COND_INITIALIZER;
877 vmexit_suspend(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
879 enum vm_suspend_how how;
881 how = vmexit->u.suspended.how;
883 fbsdrun_deletecpu(ctx, *pvcpu);
886 pthread_mutex_lock(&resetcpu_mtx);
887 pthread_cond_signal(&resetcpu_cond);
888 pthread_mutex_unlock(&resetcpu_mtx);
892 pthread_mutex_lock(&resetcpu_mtx);
893 while (!CPU_EMPTY(&cpumask)) {
894 pthread_cond_wait(&resetcpu_cond, &resetcpu_mtx);
896 pthread_mutex_unlock(&resetcpu_mtx);
899 case VM_SUSPEND_RESET:
901 case VM_SUSPEND_POWEROFF:
902 if (get_config_bool_default("destroy_on_poweroff", false))
905 case VM_SUSPEND_HALT:
907 case VM_SUSPEND_TRIPLEFAULT:
910 fprintf(stderr, "vmexit_suspend: invalid reason %d\n", how);
913 return (0); /* NOTREACHED */
917 vmexit_debug(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
920 #ifdef BHYVE_SNAPSHOT
921 checkpoint_cpu_suspend(*pvcpu);
923 gdb_cpu_suspend(*pvcpu);
924 #ifdef BHYVE_SNAPSHOT
925 checkpoint_cpu_resume(*pvcpu);
927 return (VMEXIT_CONTINUE);
931 vmexit_breakpoint(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
934 gdb_cpu_breakpoint(*pvcpu, vmexit);
935 return (VMEXIT_CONTINUE);
938 static vmexit_handler_t handler[VM_EXITCODE_MAX] = {
939 [VM_EXITCODE_INOUT] = vmexit_inout,
940 [VM_EXITCODE_INOUT_STR] = vmexit_inout,
941 [VM_EXITCODE_VMX] = vmexit_vmx,
942 [VM_EXITCODE_SVM] = vmexit_svm,
943 [VM_EXITCODE_BOGUS] = vmexit_bogus,
944 [VM_EXITCODE_REQIDLE] = vmexit_reqidle,
945 [VM_EXITCODE_RDMSR] = vmexit_rdmsr,
946 [VM_EXITCODE_WRMSR] = vmexit_wrmsr,
947 [VM_EXITCODE_MTRAP] = vmexit_mtrap,
948 [VM_EXITCODE_INST_EMUL] = vmexit_inst_emul,
949 [VM_EXITCODE_SPINUP_AP] = vmexit_spinup_ap,
950 [VM_EXITCODE_SUSPENDED] = vmexit_suspend,
951 [VM_EXITCODE_TASK_SWITCH] = vmexit_task_switch,
952 [VM_EXITCODE_DEBUG] = vmexit_debug,
953 [VM_EXITCODE_BPT] = vmexit_breakpoint,
957 vm_loop(struct vmctx *ctx, int vcpu, uint64_t startrip)
960 enum vm_exitcode exitcode;
961 cpuset_t active_cpus;
963 if (vcpumap[vcpu] != NULL) {
964 error = pthread_setaffinity_np(pthread_self(),
965 sizeof(cpuset_t), vcpumap[vcpu]);
969 error = vm_active_cpus(ctx, &active_cpus);
970 assert(CPU_ISSET(vcpu, &active_cpus));
972 error = vm_set_register(ctx, vcpu, VM_REG_GUEST_RIP, startrip);
976 error = vm_run(ctx, vcpu, &vmexit[vcpu]);
980 exitcode = vmexit[vcpu].exitcode;
981 if (exitcode >= VM_EXITCODE_MAX || handler[exitcode] == NULL) {
982 fprintf(stderr, "vm_loop: unexpected exitcode 0x%x\n",
987 rc = (*handler[exitcode])(ctx, &vmexit[vcpu], &vcpu);
990 case VMEXIT_CONTINUE:
998 fprintf(stderr, "vm_run error %d, errno %d\n", error, errno);
1002 num_vcpus_allowed(struct vmctx *ctx)
1004 uint16_t sockets, cores, threads, maxcpus;
1008 * The guest is allowed to spinup more than one processor only if the
1009 * UNRESTRICTED_GUEST capability is available.
1011 error = vm_get_capability(ctx, BSP, VM_CAP_UNRESTRICTED_GUEST, &tmp);
1015 error = vm_get_topology(ctx, &sockets, &cores, &threads, &maxcpus);
1023 fbsdrun_set_capabilities(struct vmctx *ctx, int cpu)
1027 if (get_config_bool_default("x86.vmexit_on_hlt", false)) {
1028 err = vm_get_capability(ctx, cpu, VM_CAP_HALT_EXIT, &tmp);
1030 fprintf(stderr, "VM exit on HLT not supported\n");
1033 vm_set_capability(ctx, cpu, VM_CAP_HALT_EXIT, 1);
1035 handler[VM_EXITCODE_HLT] = vmexit_hlt;
1038 if (get_config_bool_default("x86.vmexit_on_pause", false)) {
1040 * pause exit support required for this mode
1042 err = vm_get_capability(ctx, cpu, VM_CAP_PAUSE_EXIT, &tmp);
1045 "SMP mux requested, no pause support\n");
1048 vm_set_capability(ctx, cpu, VM_CAP_PAUSE_EXIT, 1);
1050 handler[VM_EXITCODE_PAUSE] = vmexit_pause;
1053 if (get_config_bool_default("x86.x2apic", false))
1054 err = vm_set_x2apic_state(ctx, cpu, X2APIC_ENABLED);
1056 err = vm_set_x2apic_state(ctx, cpu, X2APIC_DISABLED);
1059 fprintf(stderr, "Unable to set x2apic state (%d)\n", err);
1063 vm_set_capability(ctx, cpu, VM_CAP_ENABLE_INVPCID, 1);
1066 static struct vmctx *
1067 do_open(const char *vmname)
1071 bool reinit, romboot;
1072 #ifndef WITHOUT_CAPSICUM
1073 cap_rights_t rights;
1074 const cap_ioctl_t *cmds;
1078 reinit = romboot = false;
1083 error = vm_create(vmname);
1085 if (errno == EEXIST) {
1090 * The virtual machine has been setup by the
1091 * userspace bootloader.
1095 perror("vm_create");
1101 * If the virtual machine was just created then a
1102 * bootrom must be configured to boot it.
1104 fprintf(stderr, "virtual machine cannot be booted\n");
1109 ctx = vm_open(vmname);
1115 #ifndef WITHOUT_CAPSICUM
1116 cap_rights_init(&rights, CAP_IOCTL, CAP_MMAP_RW);
1117 if (caph_rights_limit(vm_get_device_fd(ctx), &rights) == -1)
1118 errx(EX_OSERR, "Unable to apply rights for sandbox");
1119 vm_get_ioctls(&ncmds);
1120 cmds = vm_get_ioctls(NULL);
1122 errx(EX_OSERR, "out of memory");
1123 if (caph_ioctls_limit(vm_get_device_fd(ctx), cmds, ncmds) == -1)
1124 errx(EX_OSERR, "Unable to apply rights for sandbox");
1125 free((cap_ioctl_t *)cmds);
1129 error = vm_reinit(ctx);
1131 perror("vm_reinit");
1135 error = vm_set_topology(ctx, sockets, cores, threads, maxcpus);
1137 errx(EX_OSERR, "vm_set_topology");
1142 spinup_vcpu(struct vmctx *ctx, int vcpu)
1147 error = vm_get_register(ctx, vcpu, VM_REG_GUEST_RIP, &rip);
1150 fbsdrun_set_capabilities(ctx, vcpu);
1151 error = vm_set_capability(ctx, vcpu, VM_CAP_UNRESTRICTED_GUEST, 1);
1154 fbsdrun_addcpu(ctx, BSP, vcpu, rip);
1158 parse_config_option(const char *option)
1163 value = strchr(option, '=');
1164 if (value == NULL || value[1] == '\0')
1166 path = strndup(option, value - option);
1168 err(4, "Failed to allocate memory");
1169 set_config_value(path, value + 1);
1174 parse_simple_config_file(const char *path)
1179 unsigned int lineno;
1181 fp = fopen(path, "r");
1183 err(4, "Failed to open configuration file %s", path);
1187 for (lineno = 1; getline(&line, &linecap, fp) > 0; lineno++) {
1188 if (*line == '#' || *line == '\n')
1190 cp = strchr(line, '\n');
1193 if (!parse_config_option(line))
1194 errx(4, "%s line %u: invalid config option '%s'", path,
1202 parse_gdb_options(char *optarg)
1207 if (optarg[0] == 'w') {
1208 set_config_bool("gdb.wait", true);
1212 colon = strrchr(optarg, ':');
1213 if (colon == NULL) {
1219 set_config_value("gdb.address", optarg);
1222 set_config_value("gdb.port", sport);
1229 set_config_bool("acpi_tables", false);
1230 set_config_value("memory.size", "256M");
1231 set_config_bool("x86.strictmsr", true);
1235 main(int argc, char *argv[])
1238 int max_vcpus, memflags;
1242 const char *value, *vmname;
1244 #ifdef BHYVE_SNAPSHOT
1246 struct restore_state rstate;
1249 restore_file = NULL;
1254 progname = basename(argv[0]);
1256 #ifdef BHYVE_SNAPSHOT
1257 optstr = "aehuwxACDHIPSWYk:o:p:G:c:s:m:l:K:U:r:";
1259 optstr = "aehuwxACDHIPSWYk:o:p:G:c:s:m:l:K:U:";
1261 while ((c = getopt(argc, argv, optstr)) != -1) {
1264 set_config_bool("x86.x2apic", false);
1267 set_config_bool("acpi_tables", true);
1270 set_config_bool("destroy_on_poweroff", true);
1273 if (pincpu_parse(optarg) != 0) {
1274 errx(EX_USAGE, "invalid vcpu pinning "
1275 "configuration '%s'", optarg);
1279 if (topology_parse(optarg) != 0) {
1280 errx(EX_USAGE, "invalid cpu topology "
1285 set_config_bool("memory.guest_in_core", true);
1288 parse_gdb_options(optarg);
1291 parse_simple_config_file(optarg);
1294 set_config_value("keyboard.layout", optarg);
1297 if (strncmp(optarg, "help", strlen(optarg)) == 0) {
1298 lpc_print_supported_devices();
1300 } else if (lpc_device_parse(optarg) != 0) {
1301 errx(EX_USAGE, "invalid lpc device "
1302 "configuration '%s'", optarg);
1305 #ifdef BHYVE_SNAPSHOT
1307 restore_file = optarg;
1311 if (strncmp(optarg, "help", strlen(optarg)) == 0) {
1312 pci_print_supported_devices();
1314 } else if (pci_parse_slot(optarg) != 0)
1319 set_config_bool("memory.wired", true);
1322 set_config_value("memory.size", optarg);
1325 if (!parse_config_option(optarg))
1326 errx(EX_USAGE, "invalid configuration option '%s'", optarg);
1329 set_config_bool("x86.vmexit_on_hlt", true);
1333 * The "-I" option was used to add an ioapic to the
1336 * An ioapic is now provided unconditionally for each
1337 * virtual machine and this option is now deprecated.
1341 set_config_bool("x86.vmexit_on_pause", true);
1344 set_config_bool("x86.strictio", true);
1347 set_config_bool("rtc.use_localtime", false);
1350 set_config_value("uuid", optarg);
1353 set_config_bool("x86.strictmsr", false);
1356 set_config_bool("virtio_msix", false);
1359 set_config_bool("x86.x2apic", true);
1362 set_config_bool("x86.mptable", false);
1376 #ifdef BHYVE_SNAPSHOT
1377 if (restore_file != NULL) {
1378 error = load_restore_file(restore_file, &rstate);
1380 fprintf(stderr, "Failed to read checkpoint info from "
1381 "file: '%s'.\n", restore_file);
1384 vmname = lookup_vmname(&rstate);
1386 set_config_value("name", vmname);
1391 set_config_value("name", argv[0]);
1393 vmname = get_config_value("name");
1397 if (get_config_bool_default("config.dump", false)) {
1405 value = get_config_value("memory.size");
1406 error = vm_parse_memsize(value, &memsize);
1408 errx(EX_USAGE, "invalid memsize '%s'", value);
1410 ctx = do_open(vmname);
1412 #ifdef BHYVE_SNAPSHOT
1413 if (restore_file != NULL) {
1414 guest_ncpus = lookup_guest_ncpus(&rstate);
1415 memflags = lookup_memflags(&rstate);
1416 memsize = lookup_memsize(&rstate);
1419 if (guest_ncpus < 1) {
1420 fprintf(stderr, "Invalid guest vCPUs (%d)\n", guest_ncpus);
1425 max_vcpus = num_vcpus_allowed(ctx);
1426 if (guest_ncpus > max_vcpus) {
1427 fprintf(stderr, "%d vCPUs requested but only %d available\n",
1428 guest_ncpus, max_vcpus);
1432 fbsdrun_set_capabilities(ctx, BSP);
1435 if (get_config_bool_default("memory.wired", false))
1436 memflags |= VM_MEM_F_WIRED;
1437 if (get_config_bool_default("memory.guest_in_core", false))
1438 memflags |= VM_MEM_F_INCORE;
1439 vm_set_memflags(ctx, memflags);
1440 err = vm_setup_memory(ctx, memsize, VM_MMAP_ALL);
1442 fprintf(stderr, "Unable to setup memory (%d)\n", errno);
1448 fprintf(stderr, "init_msr error %d", error);
1452 init_mem(guest_ncpus);
1464 * Exit if a device emulation finds an error in its initilization
1466 if (init_pci(ctx) != 0) {
1467 perror("device emulation initialization error");
1472 * Initialize after PCI, to allow a bootrom file to reserve the high
1475 if (get_config_bool("acpi_tables"))
1480 if (lpc_bootrom()) {
1481 if (vm_set_capability(ctx, BSP, VM_CAP_UNRESTRICTED_GUEST, 1)) {
1482 fprintf(stderr, "ROM boot failed: unrestricted guest "
1483 "capability not available\n");
1486 error = vcpu_reset(ctx, BSP);
1490 #ifdef BHYVE_SNAPSHOT
1491 if (restore_file != NULL) {
1492 fprintf(stdout, "Pausing pci devs...\r\n");
1493 if (vm_pause_user_devs(ctx) != 0) {
1494 fprintf(stderr, "Failed to pause PCI device state.\n");
1498 fprintf(stdout, "Restoring vm mem...\r\n");
1499 if (restore_vm_mem(ctx, &rstate) != 0) {
1500 fprintf(stderr, "Failed to restore VM memory.\n");
1504 fprintf(stdout, "Restoring pci devs...\r\n");
1505 if (vm_restore_user_devs(ctx, &rstate) != 0) {
1506 fprintf(stderr, "Failed to restore PCI device state.\n");
1510 fprintf(stdout, "Restoring kernel structs...\r\n");
1511 if (vm_restore_kern_structs(ctx, &rstate) != 0) {
1512 fprintf(stderr, "Failed to restore kernel structs.\n");
1516 fprintf(stdout, "Resuming pci devs...\r\n");
1517 if (vm_resume_user_devs(ctx) != 0) {
1518 fprintf(stderr, "Failed to resume PCI device state.\n");
1524 error = vm_get_register(ctx, BSP, VM_REG_GUEST_RIP, &rip);
1528 * build the guest tables, MP etc.
1530 if (get_config_bool_default("x86.mptable", true)) {
1531 error = mptable_build(ctx, guest_ncpus);
1533 perror("error to build the guest tables");
1538 error = smbios_build(ctx);
1542 if (get_config_bool("acpi_tables")) {
1543 error = acpi_build(ctx, guest_ncpus);
1551 * Change the proc title to include the VM name.
1553 setproctitle("%s", vmname);
1555 #ifndef WITHOUT_CAPSICUM
1556 caph_cache_catpages();
1558 if (caph_limit_stdout() == -1 || caph_limit_stderr() == -1)
1559 errx(EX_OSERR, "Unable to apply rights for sandbox");
1561 if (caph_enter() == -1)
1562 errx(EX_OSERR, "cap_enter() failed");
1565 #ifdef BHYVE_SNAPSHOT
1566 if (restore_file != NULL)
1567 destroy_restore_state(&rstate);
1569 /* initialize mutex/cond variables */
1573 * checkpointing thread for communication with bhyvectl
1575 if (init_checkpoint_thread(ctx) < 0)
1576 printf("Failed to start checkpoint thread!\r\n");
1578 if (restore_file != NULL)
1579 vm_restore_time(ctx);
1582 /* Allocate per-VCPU resources. */
1583 vmexit = calloc(guest_ncpus, sizeof(*vmexit));
1584 mt_vmm_info = calloc(guest_ncpus, sizeof(*mt_vmm_info));
1589 fbsdrun_addcpu(ctx, BSP, BSP, rip);
1591 #ifdef BHYVE_SNAPSHOT
1593 * If we restore a VM, start all vCPUs now (including APs), otherwise,
1594 * let the guest OS to spin them up later via vmexits.
1596 if (restore_file != NULL) {
1597 for (vcpu = 0; vcpu < guest_ncpus; vcpu++) {
1601 fprintf(stdout, "spinning up vcpu no %d...\r\n", vcpu);
1602 spinup_vcpu(ctx, vcpu);
1608 * Head off to the main event dispatch loop