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>
95 #include "kernemu_dev.h"
102 #include "smbiostbl.h"
103 #ifdef BHYVE_SNAPSHOT
104 #include "snapshot.h"
107 #include "spinup_ap.h"
111 #define GUEST_NIO_PORT 0x488 /* guest upcalls via i/o port */
113 #define MB (1024UL * 1024)
114 #define GB (1024UL * MB)
116 static const char * const vmx_exit_reason_desc[] = {
117 [EXIT_REASON_EXCEPTION] = "Exception or non-maskable interrupt (NMI)",
118 [EXIT_REASON_EXT_INTR] = "External interrupt",
119 [EXIT_REASON_TRIPLE_FAULT] = "Triple fault",
120 [EXIT_REASON_INIT] = "INIT signal",
121 [EXIT_REASON_SIPI] = "Start-up IPI (SIPI)",
122 [EXIT_REASON_IO_SMI] = "I/O system-management interrupt (SMI)",
123 [EXIT_REASON_SMI] = "Other SMI",
124 [EXIT_REASON_INTR_WINDOW] = "Interrupt window",
125 [EXIT_REASON_NMI_WINDOW] = "NMI window",
126 [EXIT_REASON_TASK_SWITCH] = "Task switch",
127 [EXIT_REASON_CPUID] = "CPUID",
128 [EXIT_REASON_GETSEC] = "GETSEC",
129 [EXIT_REASON_HLT] = "HLT",
130 [EXIT_REASON_INVD] = "INVD",
131 [EXIT_REASON_INVLPG] = "INVLPG",
132 [EXIT_REASON_RDPMC] = "RDPMC",
133 [EXIT_REASON_RDTSC] = "RDTSC",
134 [EXIT_REASON_RSM] = "RSM",
135 [EXIT_REASON_VMCALL] = "VMCALL",
136 [EXIT_REASON_VMCLEAR] = "VMCLEAR",
137 [EXIT_REASON_VMLAUNCH] = "VMLAUNCH",
138 [EXIT_REASON_VMPTRLD] = "VMPTRLD",
139 [EXIT_REASON_VMPTRST] = "VMPTRST",
140 [EXIT_REASON_VMREAD] = "VMREAD",
141 [EXIT_REASON_VMRESUME] = "VMRESUME",
142 [EXIT_REASON_VMWRITE] = "VMWRITE",
143 [EXIT_REASON_VMXOFF] = "VMXOFF",
144 [EXIT_REASON_VMXON] = "VMXON",
145 [EXIT_REASON_CR_ACCESS] = "Control-register accesses",
146 [EXIT_REASON_DR_ACCESS] = "MOV DR",
147 [EXIT_REASON_INOUT] = "I/O instruction",
148 [EXIT_REASON_RDMSR] = "RDMSR",
149 [EXIT_REASON_WRMSR] = "WRMSR",
150 [EXIT_REASON_INVAL_VMCS] =
151 "VM-entry failure due to invalid guest state",
152 [EXIT_REASON_INVAL_MSR] = "VM-entry failure due to MSR loading",
153 [EXIT_REASON_MWAIT] = "MWAIT",
154 [EXIT_REASON_MTF] = "Monitor trap flag",
155 [EXIT_REASON_MONITOR] = "MONITOR",
156 [EXIT_REASON_PAUSE] = "PAUSE",
157 [EXIT_REASON_MCE_DURING_ENTRY] =
158 "VM-entry failure due to machine-check event",
159 [EXIT_REASON_TPR] = "TPR below threshold",
160 [EXIT_REASON_APIC_ACCESS] = "APIC access",
161 [EXIT_REASON_VIRTUALIZED_EOI] = "Virtualized EOI",
162 [EXIT_REASON_GDTR_IDTR] = "Access to GDTR or IDTR",
163 [EXIT_REASON_LDTR_TR] = "Access to LDTR or TR",
164 [EXIT_REASON_EPT_FAULT] = "EPT violation",
165 [EXIT_REASON_EPT_MISCONFIG] = "EPT misconfiguration",
166 [EXIT_REASON_INVEPT] = "INVEPT",
167 [EXIT_REASON_RDTSCP] = "RDTSCP",
168 [EXIT_REASON_VMX_PREEMPT] = "VMX-preemption timer expired",
169 [EXIT_REASON_INVVPID] = "INVVPID",
170 [EXIT_REASON_WBINVD] = "WBINVD",
171 [EXIT_REASON_XSETBV] = "XSETBV",
172 [EXIT_REASON_APIC_WRITE] = "APIC write",
173 [EXIT_REASON_RDRAND] = "RDRAND",
174 [EXIT_REASON_INVPCID] = "INVPCID",
175 [EXIT_REASON_VMFUNC] = "VMFUNC",
176 [EXIT_REASON_ENCLS] = "ENCLS",
177 [EXIT_REASON_RDSEED] = "RDSEED",
178 [EXIT_REASON_PM_LOG_FULL] = "Page-modification log full",
179 [EXIT_REASON_XSAVES] = "XSAVES",
180 [EXIT_REASON_XRSTORS] = "XRSTORS"
183 typedef int (*vmexit_handler_t)(struct vmctx *, struct vm_exit *, int *vcpu);
184 extern int vmexit_task_switch(struct vmctx *, struct vm_exit *, int *vcpu);
189 uint16_t cores, maxcpus, sockets, threads;
191 char *guest_uuid_str;
195 static int gdb_port = 0;
196 static int guest_vmexit_on_hlt, guest_vmexit_on_pause;
197 static int virtio_msix = 1;
198 static int x2apic_mode = 0; /* default is xAPIC */
201 static int strictmsr = 1;
205 static char *progname;
206 static const int BSP = 0;
208 static cpuset_t cpumask;
210 static void vm_loop(struct vmctx *ctx, int vcpu, uint64_t rip);
212 static struct vm_exit vmexit[VM_MAXCPU];
215 uint64_t vmexit_bogus;
216 uint64_t vmexit_reqidle;
218 uint64_t vmexit_pause;
219 uint64_t vmexit_mtrap;
220 uint64_t vmexit_inst_emul;
221 uint64_t cpu_switch_rotate;
222 uint64_t cpu_switch_direct;
227 struct vmctx *mt_ctx;
229 } mt_vmm_info[VM_MAXCPU];
231 static cpuset_t *vcpumap[VM_MAXCPU] = { NULL };
238 "Usage: %s [-abehuwxACHPSWY]\n"
239 " %*s [-c [[cpus=]numcpus][,sockets=n][,cores=n][,threads=n]]\n"
240 " %*s [-g <gdb port>] [-l <lpc>]\n"
241 " %*s [-m mem] [-p vcpu:hostcpu] [-s <pci>] [-U uuid] <vm>\n"
242 " -a: local apic is in xAPIC mode (deprecated)\n"
243 " -A: create ACPI tables\n"
244 " -c: number of cpus and/or topology specification\n"
245 " -C: include guest memory in core file\n"
246 " -e: exit on unhandled I/O access\n"
249 " -H: vmexit from the guest on hlt\n"
250 " -l: LPC device configuration\n"
251 " -m: memory size in MB\n"
252 #ifdef BHYVE_SNAPSHOT
253 " -r: path to checkpoint file\n"
255 " -p: pin 'vcpu' to 'hostcpu'\n"
256 " -P: vmexit from the guest on pause\n"
257 " -s: <slot,driver,configinfo> PCI slot config\n"
258 " -S: guest memory cannot be swapped\n"
259 " -u: RTC keeps UTC time\n"
261 " -w: ignore unimplemented MSRs\n"
262 " -W: force virtio to use single-vector MSI\n"
263 " -x: local apic is in x2APIC mode\n"
264 " -Y: disable MPtable generation\n",
265 progname, (int)strlen(progname), "", (int)strlen(progname), "",
266 (int)strlen(progname), "");
272 * XXX This parser is known to have the following issues:
273 * 1. It accepts null key=value tokens ",,".
274 * 2. It accepts whitespace after = and before value.
275 * 3. Values out of range of INT are silently wrapped.
276 * 4. It doesn't check non-final values.
277 * 5. The apparently bogus limits of UINT16_MAX are for future expansion.
279 * The acceptance of a null specification ('-c ""') is by design to match the
280 * manual page syntax specification, this results in a topology of 1 vCPU.
283 topology_parse(const char *opt)
286 int c, chk, n, s, t, tmp;
290 c = 1, n = 1, s = 1, t = 1;
291 ns = false, scts = false;
296 while ((cp = strsep(&str, ",")) != NULL) {
297 if (sscanf(cp, "%i%n", &tmp, &chk) == 1) {
300 } else if (sscanf(cp, "cpus=%i%n", &tmp, &chk) == 1) {
303 } else if (sscanf(cp, "sockets=%i%n", &tmp, &chk) == 1) {
306 } else if (sscanf(cp, "cores=%i%n", &tmp, &chk) == 1) {
309 } else if (sscanf(cp, "threads=%i%n", &tmp, &chk) == 1) {
312 #ifdef notyet /* Do not expose this until vmm.ko implements it */
313 } else if (sscanf(cp, "maxcpus=%i%n", &tmp, &chk) == 1) {
316 /* Skip the empty argument case from -c "" */
317 } else if (cp[0] == '\0')
321 /* Any trailing garbage causes an error */
329 * Range check 1 <= n <= UINT16_MAX all values
331 if (n < 1 || s < 1 || c < 1 || t < 1 ||
332 n > UINT16_MAX || s > UINT16_MAX || c > UINT16_MAX ||
336 /* If only the cpus was specified, use that as sockets */
340 * Compute sockets * cores * threads avoiding overflow
341 * The range check above insures these are 16 bit values
342 * If n was specified check it against computed ncpus
344 ncpus = (uint64_t)s * c * t;
345 if (ncpus > UINT16_MAX || (ns && n != ncpus))
360 pincpu_parse(const char *opt)
364 if (sscanf(opt, "%d:%d", &vcpu, &pcpu) != 2) {
365 fprintf(stderr, "invalid format: %s\n", opt);
369 if (vcpu < 0 || vcpu >= VM_MAXCPU) {
370 fprintf(stderr, "vcpu '%d' outside valid range from 0 to %d\n",
371 vcpu, VM_MAXCPU - 1);
375 if (pcpu < 0 || pcpu >= CPU_SETSIZE) {
376 fprintf(stderr, "hostcpu '%d' outside valid range from "
377 "0 to %d\n", pcpu, CPU_SETSIZE - 1);
381 if (vcpumap[vcpu] == NULL) {
382 if ((vcpumap[vcpu] = malloc(sizeof(cpuset_t))) == NULL) {
386 CPU_ZERO(vcpumap[vcpu]);
388 CPU_SET(pcpu, vcpumap[vcpu]);
393 vm_inject_fault(void *arg, int vcpu, int vector, int errcode_valid,
397 int error, restart_instruction;
400 restart_instruction = 1;
402 error = vm_inject_exception(ctx, vcpu, vector, errcode_valid, errcode,
403 restart_instruction);
408 paddr_guest2host(struct vmctx *ctx, uintptr_t gaddr, size_t len)
411 return (vm_map_gpa(ctx, gaddr, len));
414 #ifdef BHYVE_SNAPSHOT
416 paddr_host2guest(struct vmctx *ctx, void *addr)
418 return (vm_rev_map_gpa(ctx, addr));
423 fbsdrun_vmexit_on_pause(void)
426 return (guest_vmexit_on_pause);
430 fbsdrun_vmexit_on_hlt(void)
433 return (guest_vmexit_on_hlt);
437 fbsdrun_virtio_msix(void)
440 return (virtio_msix);
444 fbsdrun_start_thread(void *param)
446 char tname[MAXCOMLEN + 1];
447 struct mt_vmm_info *mtp;
453 snprintf(tname, sizeof(tname), "vcpu %d", vcpu);
454 pthread_set_name_np(mtp->mt_thr, tname);
456 #ifdef BHYVE_SNAPSHOT
457 checkpoint_cpu_add(vcpu);
462 vm_loop(mtp->mt_ctx, vcpu, vmexit[vcpu].rip);
470 fbsdrun_addcpu(struct vmctx *ctx, int fromcpu, int newcpu, uint64_t rip)
474 assert(fromcpu == BSP);
477 * The 'newcpu' must be activated in the context of 'fromcpu'. If
478 * vm_activate_cpu() is delayed until newcpu's pthread starts running
479 * then vmm.ko is out-of-sync with bhyve and this can create a race
482 error = vm_activate_cpu(ctx, newcpu);
484 err(EX_OSERR, "could not activate CPU %d", newcpu);
486 CPU_SET_ATOMIC(newcpu, &cpumask);
489 * Set up the vmexit struct to allow execution to start
492 vmexit[newcpu].rip = rip;
493 vmexit[newcpu].inst_length = 0;
495 mt_vmm_info[newcpu].mt_ctx = ctx;
496 mt_vmm_info[newcpu].mt_vcpu = newcpu;
498 error = pthread_create(&mt_vmm_info[newcpu].mt_thr, NULL,
499 fbsdrun_start_thread, &mt_vmm_info[newcpu]);
504 fbsdrun_deletecpu(struct vmctx *ctx, int vcpu)
507 if (!CPU_ISSET(vcpu, &cpumask)) {
508 fprintf(stderr, "Attempting to delete unknown cpu %d\n", vcpu);
512 CPU_CLR_ATOMIC(vcpu, &cpumask);
513 return (CPU_EMPTY(&cpumask));
517 vmexit_handle_notify(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu,
522 * put guest-driven debug here
525 return (VMEXIT_CONTINUE);
529 vmexit_inout(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu)
532 int bytes, port, in, out;
537 port = vme->u.inout.port;
538 bytes = vme->u.inout.bytes;
539 in = vme->u.inout.in;
542 /* Extra-special case of host notifications */
543 if (out && port == GUEST_NIO_PORT) {
544 error = vmexit_handle_notify(ctx, vme, pvcpu, vme->u.inout.eax);
548 error = emulate_inout(ctx, vcpu, vme, strictio);
550 fprintf(stderr, "Unhandled %s%c 0x%04x at 0x%lx\n",
552 bytes == 1 ? 'b' : (bytes == 2 ? 'w' : 'l'),
554 return (VMEXIT_ABORT);
556 return (VMEXIT_CONTINUE);
561 vmexit_rdmsr(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu)
568 error = emulate_rdmsr(ctx, *pvcpu, vme->u.msr.code, &val);
570 fprintf(stderr, "rdmsr to register %#x on vcpu %d\n",
571 vme->u.msr.code, *pvcpu);
573 vm_inject_gp(ctx, *pvcpu);
574 return (VMEXIT_CONTINUE);
579 error = vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RAX, eax);
583 error = vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RDX, edx);
586 return (VMEXIT_CONTINUE);
590 vmexit_wrmsr(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu)
594 error = emulate_wrmsr(ctx, *pvcpu, vme->u.msr.code, vme->u.msr.wval);
596 fprintf(stderr, "wrmsr to register %#x(%#lx) on vcpu %d\n",
597 vme->u.msr.code, vme->u.msr.wval, *pvcpu);
599 vm_inject_gp(ctx, *pvcpu);
600 return (VMEXIT_CONTINUE);
603 return (VMEXIT_CONTINUE);
607 vmexit_spinup_ap(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu)
610 (void)spinup_ap(ctx, *pvcpu,
611 vme->u.spinup_ap.vcpu, vme->u.spinup_ap.rip);
613 return (VMEXIT_CONTINUE);
616 #define DEBUG_EPT_MISCONFIG
617 #ifdef DEBUG_EPT_MISCONFIG
618 #define VMCS_GUEST_PHYSICAL_ADDRESS 0x00002400
620 static uint64_t ept_misconfig_gpa, ept_misconfig_pte[4];
621 static int ept_misconfig_ptenum;
625 vmexit_vmx_desc(uint32_t exit_reason)
628 if (exit_reason >= nitems(vmx_exit_reason_desc) ||
629 vmx_exit_reason_desc[exit_reason] == NULL)
631 return (vmx_exit_reason_desc[exit_reason]);
635 vmexit_vmx(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
638 fprintf(stderr, "vm exit[%d]\n", *pvcpu);
639 fprintf(stderr, "\treason\t\tVMX\n");
640 fprintf(stderr, "\trip\t\t0x%016lx\n", vmexit->rip);
641 fprintf(stderr, "\tinst_length\t%d\n", vmexit->inst_length);
642 fprintf(stderr, "\tstatus\t\t%d\n", vmexit->u.vmx.status);
643 fprintf(stderr, "\texit_reason\t%u (%s)\n", vmexit->u.vmx.exit_reason,
644 vmexit_vmx_desc(vmexit->u.vmx.exit_reason));
645 fprintf(stderr, "\tqualification\t0x%016lx\n",
646 vmexit->u.vmx.exit_qualification);
647 fprintf(stderr, "\tinst_type\t\t%d\n", vmexit->u.vmx.inst_type);
648 fprintf(stderr, "\tinst_error\t\t%d\n", vmexit->u.vmx.inst_error);
649 #ifdef DEBUG_EPT_MISCONFIG
650 if (vmexit->u.vmx.exit_reason == EXIT_REASON_EPT_MISCONFIG) {
651 vm_get_register(ctx, *pvcpu,
652 VMCS_IDENT(VMCS_GUEST_PHYSICAL_ADDRESS),
654 vm_get_gpa_pmap(ctx, ept_misconfig_gpa, ept_misconfig_pte,
655 &ept_misconfig_ptenum);
656 fprintf(stderr, "\tEPT misconfiguration:\n");
657 fprintf(stderr, "\t\tGPA: %#lx\n", ept_misconfig_gpa);
658 fprintf(stderr, "\t\tPTE(%d): %#lx %#lx %#lx %#lx\n",
659 ept_misconfig_ptenum, ept_misconfig_pte[0],
660 ept_misconfig_pte[1], ept_misconfig_pte[2],
661 ept_misconfig_pte[3]);
663 #endif /* DEBUG_EPT_MISCONFIG */
664 return (VMEXIT_ABORT);
668 vmexit_svm(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
671 fprintf(stderr, "vm exit[%d]\n", *pvcpu);
672 fprintf(stderr, "\treason\t\tSVM\n");
673 fprintf(stderr, "\trip\t\t0x%016lx\n", vmexit->rip);
674 fprintf(stderr, "\tinst_length\t%d\n", vmexit->inst_length);
675 fprintf(stderr, "\texitcode\t%#lx\n", vmexit->u.svm.exitcode);
676 fprintf(stderr, "\texitinfo1\t%#lx\n", vmexit->u.svm.exitinfo1);
677 fprintf(stderr, "\texitinfo2\t%#lx\n", vmexit->u.svm.exitinfo2);
678 return (VMEXIT_ABORT);
682 vmexit_bogus(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
685 assert(vmexit->inst_length == 0);
687 stats.vmexit_bogus++;
689 return (VMEXIT_CONTINUE);
693 vmexit_reqidle(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
696 assert(vmexit->inst_length == 0);
698 stats.vmexit_reqidle++;
700 return (VMEXIT_CONTINUE);
704 vmexit_hlt(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
710 * Just continue execution with the next instruction. We use
711 * the HLT VM exit as a way to be friendly with the host
714 return (VMEXIT_CONTINUE);
718 vmexit_pause(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
721 stats.vmexit_pause++;
723 return (VMEXIT_CONTINUE);
727 vmexit_mtrap(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
730 assert(vmexit->inst_length == 0);
732 stats.vmexit_mtrap++;
734 #ifdef BHYVE_SNAPSHOT
735 checkpoint_cpu_suspend(*pvcpu);
738 gdb_cpu_mtrap(*pvcpu);
739 #ifdef BHYVE_SNAPSHOT
740 checkpoint_cpu_resume(*pvcpu);
743 return (VMEXIT_CONTINUE);
747 vmexit_inst_emul(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
752 stats.vmexit_inst_emul++;
754 vie = &vmexit->u.inst_emul.vie;
755 err = emulate_mem(ctx, *pvcpu, vmexit->u.inst_emul.gpa,
756 vie, &vmexit->u.inst_emul.paging);
760 EPRINTLN("Unhandled memory access to 0x%lx\n",
761 vmexit->u.inst_emul.gpa);
764 fprintf(stderr, "Failed to emulate instruction sequence [ ");
765 for (i = 0; i < vie->num_valid; i++)
766 fprintf(stderr, "%02x", vie->inst[i]);
767 FPRINTLN(stderr, " ] at 0x%lx", vmexit->rip);
768 return (VMEXIT_ABORT);
771 return (VMEXIT_CONTINUE);
774 static pthread_mutex_t resetcpu_mtx = PTHREAD_MUTEX_INITIALIZER;
775 static pthread_cond_t resetcpu_cond = PTHREAD_COND_INITIALIZER;
778 vmexit_suspend(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
780 enum vm_suspend_how how;
782 how = vmexit->u.suspended.how;
784 fbsdrun_deletecpu(ctx, *pvcpu);
787 pthread_mutex_lock(&resetcpu_mtx);
788 pthread_cond_signal(&resetcpu_cond);
789 pthread_mutex_unlock(&resetcpu_mtx);
793 pthread_mutex_lock(&resetcpu_mtx);
794 while (!CPU_EMPTY(&cpumask)) {
795 pthread_cond_wait(&resetcpu_cond, &resetcpu_mtx);
797 pthread_mutex_unlock(&resetcpu_mtx);
800 case VM_SUSPEND_RESET:
802 case VM_SUSPEND_POWEROFF:
804 case VM_SUSPEND_HALT:
806 case VM_SUSPEND_TRIPLEFAULT:
809 fprintf(stderr, "vmexit_suspend: invalid reason %d\n", how);
812 return (0); /* NOTREACHED */
816 vmexit_debug(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
819 #ifdef BHYVE_SNAPSHOT
820 checkpoint_cpu_suspend(*pvcpu);
823 gdb_cpu_suspend(*pvcpu);
824 #ifdef BHYVE_SNAPSHOT
825 checkpoint_cpu_resume(*pvcpu);
827 return (VMEXIT_CONTINUE);
831 vmexit_breakpoint(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
835 fprintf(stderr, "vm_loop: unexpected VMEXIT_DEBUG\n");
838 gdb_cpu_breakpoint(*pvcpu, vmexit);
839 return (VMEXIT_CONTINUE);
842 static vmexit_handler_t handler[VM_EXITCODE_MAX] = {
843 [VM_EXITCODE_INOUT] = vmexit_inout,
844 [VM_EXITCODE_INOUT_STR] = vmexit_inout,
845 [VM_EXITCODE_VMX] = vmexit_vmx,
846 [VM_EXITCODE_SVM] = vmexit_svm,
847 [VM_EXITCODE_BOGUS] = vmexit_bogus,
848 [VM_EXITCODE_REQIDLE] = vmexit_reqidle,
849 [VM_EXITCODE_RDMSR] = vmexit_rdmsr,
850 [VM_EXITCODE_WRMSR] = vmexit_wrmsr,
851 [VM_EXITCODE_MTRAP] = vmexit_mtrap,
852 [VM_EXITCODE_INST_EMUL] = vmexit_inst_emul,
853 [VM_EXITCODE_SPINUP_AP] = vmexit_spinup_ap,
854 [VM_EXITCODE_SUSPENDED] = vmexit_suspend,
855 [VM_EXITCODE_TASK_SWITCH] = vmexit_task_switch,
856 [VM_EXITCODE_DEBUG] = vmexit_debug,
857 [VM_EXITCODE_BPT] = vmexit_breakpoint,
861 vm_loop(struct vmctx *ctx, int vcpu, uint64_t startrip)
864 enum vm_exitcode exitcode;
865 cpuset_t active_cpus;
867 if (vcpumap[vcpu] != NULL) {
868 error = pthread_setaffinity_np(pthread_self(),
869 sizeof(cpuset_t), vcpumap[vcpu]);
873 error = vm_active_cpus(ctx, &active_cpus);
874 assert(CPU_ISSET(vcpu, &active_cpus));
876 error = vm_set_register(ctx, vcpu, VM_REG_GUEST_RIP, startrip);
880 error = vm_run(ctx, vcpu, &vmexit[vcpu]);
884 exitcode = vmexit[vcpu].exitcode;
885 if (exitcode >= VM_EXITCODE_MAX || handler[exitcode] == NULL) {
886 fprintf(stderr, "vm_loop: unexpected exitcode 0x%x\n",
891 rc = (*handler[exitcode])(ctx, &vmexit[vcpu], &vcpu);
894 case VMEXIT_CONTINUE:
902 fprintf(stderr, "vm_run error %d, errno %d\n", error, errno);
906 num_vcpus_allowed(struct vmctx *ctx)
910 error = vm_get_capability(ctx, BSP, VM_CAP_UNRESTRICTED_GUEST, &tmp);
913 * The guest is allowed to spinup more than one processor only if the
914 * UNRESTRICTED_GUEST capability is available.
923 fbsdrun_set_capabilities(struct vmctx *ctx, int cpu)
927 if (fbsdrun_vmexit_on_hlt()) {
928 err = vm_get_capability(ctx, cpu, VM_CAP_HALT_EXIT, &tmp);
930 fprintf(stderr, "VM exit on HLT not supported\n");
933 vm_set_capability(ctx, cpu, VM_CAP_HALT_EXIT, 1);
935 handler[VM_EXITCODE_HLT] = vmexit_hlt;
938 if (fbsdrun_vmexit_on_pause()) {
940 * pause exit support required for this mode
942 err = vm_get_capability(ctx, cpu, VM_CAP_PAUSE_EXIT, &tmp);
945 "SMP mux requested, no pause support\n");
948 vm_set_capability(ctx, cpu, VM_CAP_PAUSE_EXIT, 1);
950 handler[VM_EXITCODE_PAUSE] = vmexit_pause;
954 err = vm_set_x2apic_state(ctx, cpu, X2APIC_ENABLED);
956 err = vm_set_x2apic_state(ctx, cpu, X2APIC_DISABLED);
959 fprintf(stderr, "Unable to set x2apic state (%d)\n", err);
963 vm_set_capability(ctx, cpu, VM_CAP_ENABLE_INVPCID, 1);
966 static struct vmctx *
967 do_open(const char *vmname)
971 bool reinit, romboot;
972 #ifndef WITHOUT_CAPSICUM
974 const cap_ioctl_t *cmds;
978 reinit = romboot = false;
983 error = vm_create(vmname);
985 if (errno == EEXIST) {
990 * The virtual machine has been setup by the
991 * userspace bootloader.
1001 * If the virtual machine was just created then a
1002 * bootrom must be configured to boot it.
1004 fprintf(stderr, "virtual machine cannot be booted\n");
1009 ctx = vm_open(vmname);
1015 #ifndef WITHOUT_CAPSICUM
1016 cap_rights_init(&rights, CAP_IOCTL, CAP_MMAP_RW);
1017 if (caph_rights_limit(vm_get_device_fd(ctx), &rights) == -1)
1018 errx(EX_OSERR, "Unable to apply rights for sandbox");
1019 vm_get_ioctls(&ncmds);
1020 cmds = vm_get_ioctls(NULL);
1022 errx(EX_OSERR, "out of memory");
1023 if (caph_ioctls_limit(vm_get_device_fd(ctx), cmds, ncmds) == -1)
1024 errx(EX_OSERR, "Unable to apply rights for sandbox");
1025 free((cap_ioctl_t *)cmds);
1029 error = vm_reinit(ctx);
1031 perror("vm_reinit");
1035 error = vm_set_topology(ctx, sockets, cores, threads, maxcpus);
1037 errx(EX_OSERR, "vm_set_topology");
1042 spinup_vcpu(struct vmctx *ctx, int vcpu)
1047 error = vm_get_register(ctx, vcpu, VM_REG_GUEST_RIP, &rip);
1050 fbsdrun_set_capabilities(ctx, vcpu);
1051 error = vm_set_capability(ctx, vcpu, VM_CAP_UNRESTRICTED_GUEST, 1);
1054 fbsdrun_addcpu(ctx, BSP, vcpu, rip);
1058 main(int argc, char *argv[])
1060 int c, error, dbg_port, err, bvmcons;
1061 int max_vcpus, mptgen, memflags;
1068 #ifdef BHYVE_SNAPSHOT
1070 struct restore_state rstate;
1073 restore_file = NULL;
1077 progname = basename(argv[0]);
1081 sockets = cores = threads = 1;
1088 #ifdef BHYVE_SNAPSHOT
1089 optstr = "abehuwxACHIPSWYp:g:G:c:s:m:l:U:r:";
1091 optstr = "abehuwxACHIPSWYp:g:G:c:s:m:l:U:";
1093 while ((c = getopt(argc, argv, optstr)) != -1) {
1105 if (pincpu_parse(optarg) != 0) {
1106 errx(EX_USAGE, "invalid vcpu pinning "
1107 "configuration '%s'", optarg);
1111 if (topology_parse(optarg) != 0) {
1112 errx(EX_USAGE, "invalid cpu topology "
1117 memflags |= VM_MEM_F_INCORE;
1120 dbg_port = atoi(optarg);
1123 if (optarg[0] == 'w') {
1127 gdb_port = atoi(optarg);
1130 if (strncmp(optarg, "help", strlen(optarg)) == 0) {
1131 lpc_print_supported_devices();
1133 } else if (lpc_device_parse(optarg) != 0) {
1134 errx(EX_USAGE, "invalid lpc device "
1135 "configuration '%s'", optarg);
1138 #ifdef BHYVE_SNAPSHOT
1140 restore_file = optarg;
1144 if (strncmp(optarg, "help", strlen(optarg)) == 0) {
1145 pci_print_supported_devices();
1147 } else if (pci_parse_slot(optarg) != 0)
1152 memflags |= VM_MEM_F_WIRED;
1155 error = vm_parse_memsize(optarg, &memsize);
1157 errx(EX_USAGE, "invalid memsize '%s'", optarg);
1160 guest_vmexit_on_hlt = 1;
1164 * The "-I" option was used to add an ioapic to the
1167 * An ioapic is now provided unconditionally for each
1168 * virtual machine and this option is now deprecated.
1172 guest_vmexit_on_pause = 1;
1181 guest_uuid_str = optarg;
1204 #ifdef BHYVE_SNAPSHOT
1205 if (argc > 1 || (argc == 0 && restore_file == NULL))
1208 if (restore_file != NULL) {
1209 error = load_restore_file(restore_file, &rstate);
1211 fprintf(stderr, "Failed to read checkpoint info from "
1212 "file: '%s'.\n", restore_file);
1220 vmname = lookup_vmname(&rstate);
1221 if (vmname == NULL) {
1222 fprintf(stderr, "Cannot find VM name in restore file. "
1223 "Please specify one.\n");
1233 ctx = do_open(vmname);
1235 #ifdef BHYVE_SNAPSHOT
1236 if (restore_file != NULL) {
1237 guest_ncpus = lookup_guest_ncpus(&rstate);
1238 memflags = lookup_memflags(&rstate);
1239 memsize = lookup_memsize(&rstate);
1242 if (guest_ncpus < 1) {
1243 fprintf(stderr, "Invalid guest vCPUs (%d)\n", guest_ncpus);
1248 max_vcpus = num_vcpus_allowed(ctx);
1249 if (guest_ncpus > max_vcpus) {
1250 fprintf(stderr, "%d vCPUs requested but only %d available\n",
1251 guest_ncpus, max_vcpus);
1255 fbsdrun_set_capabilities(ctx, BSP);
1257 vm_set_memflags(ctx, memflags);
1258 err = vm_setup_memory(ctx, memsize, VM_MMAP_ALL);
1260 fprintf(stderr, "Unable to setup memory (%d)\n", errno);
1266 fprintf(stderr, "init_msr error %d", error);
1278 rtc_init(ctx, rtc_localtime);
1282 * Exit if a device emulation finds an error in its initilization
1284 if (init_pci(ctx) != 0) {
1285 perror("device emulation initialization error");
1290 * Initialize after PCI, to allow a bootrom file to reserve the high
1297 init_dbgport(dbg_port);
1300 init_gdb(ctx, gdb_port, gdb_stop);
1305 if (lpc_bootrom()) {
1306 if (vm_set_capability(ctx, BSP, VM_CAP_UNRESTRICTED_GUEST, 1)) {
1307 fprintf(stderr, "ROM boot failed: unrestricted guest "
1308 "capability not available\n");
1311 error = vcpu_reset(ctx, BSP);
1315 #ifdef BHYVE_SNAPSHOT
1316 if (restore_file != NULL) {
1317 fprintf(stdout, "Pausing pci devs...\r\n");
1318 if (vm_pause_user_devs(ctx) != 0) {
1319 fprintf(stderr, "Failed to pause PCI device state.\n");
1323 fprintf(stdout, "Restoring vm mem...\r\n");
1324 if (restore_vm_mem(ctx, &rstate) != 0) {
1325 fprintf(stderr, "Failed to restore VM memory.\n");
1329 fprintf(stdout, "Restoring pci devs...\r\n");
1330 if (vm_restore_user_devs(ctx, &rstate) != 0) {
1331 fprintf(stderr, "Failed to restore PCI device state.\n");
1335 fprintf(stdout, "Restoring kernel structs...\r\n");
1336 if (vm_restore_kern_structs(ctx, &rstate) != 0) {
1337 fprintf(stderr, "Failed to restore kernel structs.\n");
1341 fprintf(stdout, "Resuming pci devs...\r\n");
1342 if (vm_resume_user_devs(ctx) != 0) {
1343 fprintf(stderr, "Failed to resume PCI device state.\n");
1349 error = vm_get_register(ctx, BSP, VM_REG_GUEST_RIP, &rip);
1353 * build the guest tables, MP etc.
1356 error = mptable_build(ctx, guest_ncpus);
1358 perror("error to build the guest tables");
1363 error = smbios_build(ctx);
1367 error = acpi_build(ctx, guest_ncpus);
1375 * Change the proc title to include the VM name.
1377 setproctitle("%s", vmname);
1379 #ifndef WITHOUT_CAPSICUM
1380 caph_cache_catpages();
1382 if (caph_limit_stdout() == -1 || caph_limit_stderr() == -1)
1383 errx(EX_OSERR, "Unable to apply rights for sandbox");
1385 if (caph_enter() == -1)
1386 errx(EX_OSERR, "cap_enter() failed");
1389 #ifdef BHYVE_SNAPSHOT
1390 if (restore_file != NULL)
1391 destroy_restore_state(&rstate);
1394 * checkpointing thread for communication with bhyvectl
1396 if (init_checkpoint_thread(ctx) < 0)
1397 printf("Failed to start checkpoint thread!\r\n");
1399 if (restore_file != NULL)
1400 vm_restore_time(ctx);
1406 fbsdrun_addcpu(ctx, BSP, BSP, rip);
1408 #ifdef BHYVE_SNAPSHOT
1410 * If we restore a VM, start all vCPUs now (including APs), otherwise,
1411 * let the guest OS to spin them up later via vmexits.
1413 if (restore_file != NULL) {
1414 for (vcpu = 0; vcpu < guest_ncpus; vcpu++) {
1418 fprintf(stdout, "spinning up vcpu no %d...\r\n", vcpu);
1419 spinup_vcpu(ctx, vcpu);
1425 * Head off to the main event dispatch loop