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);
190 uint16_t cores, maxcpus, sockets, threads;
192 char *guest_uuid_str;
196 static int gdb_port = 0;
197 static int guest_vmexit_on_hlt, guest_vmexit_on_pause;
198 static int virtio_msix = 1;
199 static int x2apic_mode = 0; /* default is xAPIC */
200 static int destroy_on_poweroff = 0;
203 static int strictmsr = 1;
207 static char *progname;
208 static const int BSP = 0;
210 static cpuset_t cpumask;
212 static void vm_loop(struct vmctx *ctx, int vcpu, uint64_t rip);
214 static struct vm_exit vmexit[VM_MAXCPU];
217 uint64_t vmexit_bogus;
218 uint64_t vmexit_reqidle;
220 uint64_t vmexit_pause;
221 uint64_t vmexit_mtrap;
222 uint64_t vmexit_inst_emul;
223 uint64_t cpu_switch_rotate;
224 uint64_t cpu_switch_direct;
229 struct vmctx *mt_ctx;
231 } mt_vmm_info[VM_MAXCPU];
233 static cpuset_t *vcpumap[VM_MAXCPU] = { NULL };
240 "Usage: %s [-abehuwxACDHPSWY]\n"
241 " %*s [-c [[cpus=]numcpus][,sockets=n][,cores=n][,threads=n]]\n"
242 " %*s [-g <gdb port>] [-l <lpc>]\n"
243 " %*s [-m mem] [-p vcpu:hostcpu] [-s <pci>] [-U uuid] <vm>\n"
244 " -a: local apic is in xAPIC mode (deprecated)\n"
245 " -A: create ACPI tables\n"
246 " -c: number of cpus and/or topology specification\n"
247 " -C: include guest memory in core file\n"
248 " -D: destroy on power-off\n"
249 " -e: exit on unhandled I/O access\n"
252 " -H: vmexit from the guest on hlt\n"
253 " -l: LPC device configuration\n"
254 " -m: memory size in MB\n"
255 #ifdef BHYVE_SNAPSHOT
256 " -r: path to checkpoint file\n"
258 " -p: pin 'vcpu' to 'hostcpu'\n"
259 " -P: vmexit from the guest on pause\n"
260 " -s: <slot,driver,configinfo> PCI slot config\n"
261 " -S: guest memory cannot be swapped\n"
262 " -u: RTC keeps UTC time\n"
264 " -w: ignore unimplemented MSRs\n"
265 " -W: force virtio to use single-vector MSI\n"
266 " -x: local apic is in x2APIC mode\n"
267 " -Y: disable MPtable generation\n",
268 progname, (int)strlen(progname), "", (int)strlen(progname), "",
269 (int)strlen(progname), "");
275 * XXX This parser is known to have the following issues:
276 * 1. It accepts null key=value tokens ",,".
277 * 2. It accepts whitespace after = and before value.
278 * 3. Values out of range of INT are silently wrapped.
279 * 4. It doesn't check non-final values.
280 * 5. The apparently bogus limits of UINT16_MAX are for future expansion.
282 * The acceptance of a null specification ('-c ""') is by design to match the
283 * manual page syntax specification, this results in a topology of 1 vCPU.
286 topology_parse(const char *opt)
289 int c, chk, n, s, t, tmp;
293 c = 1, n = 1, s = 1, t = 1;
294 ns = false, scts = false;
299 while ((cp = strsep(&str, ",")) != NULL) {
300 if (sscanf(cp, "%i%n", &tmp, &chk) == 1) {
303 } else if (sscanf(cp, "cpus=%i%n", &tmp, &chk) == 1) {
306 } else if (sscanf(cp, "sockets=%i%n", &tmp, &chk) == 1) {
309 } else if (sscanf(cp, "cores=%i%n", &tmp, &chk) == 1) {
312 } else if (sscanf(cp, "threads=%i%n", &tmp, &chk) == 1) {
315 #ifdef notyet /* Do not expose this until vmm.ko implements it */
316 } else if (sscanf(cp, "maxcpus=%i%n", &tmp, &chk) == 1) {
319 /* Skip the empty argument case from -c "" */
320 } else if (cp[0] == '\0')
324 /* Any trailing garbage causes an error */
332 * Range check 1 <= n <= UINT16_MAX all values
334 if (n < 1 || s < 1 || c < 1 || t < 1 ||
335 n > UINT16_MAX || s > UINT16_MAX || c > UINT16_MAX ||
339 /* If only the cpus was specified, use that as sockets */
343 * Compute sockets * cores * threads avoiding overflow
344 * The range check above insures these are 16 bit values
345 * If n was specified check it against computed ncpus
347 ncpus = (uint64_t)s * c * t;
348 if (ncpus > UINT16_MAX || (ns && n != ncpus))
363 pincpu_parse(const char *opt)
367 if (sscanf(opt, "%d:%d", &vcpu, &pcpu) != 2) {
368 fprintf(stderr, "invalid format: %s\n", opt);
372 if (vcpu < 0 || vcpu >= VM_MAXCPU) {
373 fprintf(stderr, "vcpu '%d' outside valid range from 0 to %d\n",
374 vcpu, VM_MAXCPU - 1);
378 if (pcpu < 0 || pcpu >= CPU_SETSIZE) {
379 fprintf(stderr, "hostcpu '%d' outside valid range from "
380 "0 to %d\n", pcpu, CPU_SETSIZE - 1);
384 if (vcpumap[vcpu] == NULL) {
385 if ((vcpumap[vcpu] = malloc(sizeof(cpuset_t))) == NULL) {
389 CPU_ZERO(vcpumap[vcpu]);
391 CPU_SET(pcpu, vcpumap[vcpu]);
396 vm_inject_fault(void *arg, int vcpu, int vector, int errcode_valid,
400 int error, restart_instruction;
403 restart_instruction = 1;
405 error = vm_inject_exception(ctx, vcpu, vector, errcode_valid, errcode,
406 restart_instruction);
411 paddr_guest2host(struct vmctx *ctx, uintptr_t gaddr, size_t len)
414 return (vm_map_gpa(ctx, gaddr, len));
417 #ifdef BHYVE_SNAPSHOT
419 paddr_host2guest(struct vmctx *ctx, void *addr)
421 return (vm_rev_map_gpa(ctx, addr));
426 fbsdrun_vmexit_on_pause(void)
429 return (guest_vmexit_on_pause);
433 fbsdrun_vmexit_on_hlt(void)
436 return (guest_vmexit_on_hlt);
440 fbsdrun_virtio_msix(void)
443 return (virtio_msix);
447 fbsdrun_start_thread(void *param)
449 char tname[MAXCOMLEN + 1];
450 struct mt_vmm_info *mtp;
456 snprintf(tname, sizeof(tname), "vcpu %d", vcpu);
457 pthread_set_name_np(mtp->mt_thr, tname);
459 #ifdef BHYVE_SNAPSHOT
460 checkpoint_cpu_add(vcpu);
465 vm_loop(mtp->mt_ctx, vcpu, vmexit[vcpu].rip);
473 fbsdrun_addcpu(struct vmctx *ctx, int fromcpu, int newcpu, uint64_t rip)
477 assert(fromcpu == BSP);
480 * The 'newcpu' must be activated in the context of 'fromcpu'. If
481 * vm_activate_cpu() is delayed until newcpu's pthread starts running
482 * then vmm.ko is out-of-sync with bhyve and this can create a race
485 error = vm_activate_cpu(ctx, newcpu);
487 err(EX_OSERR, "could not activate CPU %d", newcpu);
489 CPU_SET_ATOMIC(newcpu, &cpumask);
492 * Set up the vmexit struct to allow execution to start
495 vmexit[newcpu].rip = rip;
496 vmexit[newcpu].inst_length = 0;
498 mt_vmm_info[newcpu].mt_ctx = ctx;
499 mt_vmm_info[newcpu].mt_vcpu = newcpu;
501 error = pthread_create(&mt_vmm_info[newcpu].mt_thr, NULL,
502 fbsdrun_start_thread, &mt_vmm_info[newcpu]);
507 fbsdrun_deletecpu(struct vmctx *ctx, int vcpu)
510 if (!CPU_ISSET(vcpu, &cpumask)) {
511 fprintf(stderr, "Attempting to delete unknown cpu %d\n", vcpu);
515 CPU_CLR_ATOMIC(vcpu, &cpumask);
516 return (CPU_EMPTY(&cpumask));
520 vmexit_handle_notify(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu,
525 * put guest-driven debug here
528 return (VMEXIT_CONTINUE);
532 vmexit_inout(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu)
535 int bytes, port, in, out;
540 port = vme->u.inout.port;
541 bytes = vme->u.inout.bytes;
542 in = vme->u.inout.in;
545 /* Extra-special case of host notifications */
546 if (out && port == GUEST_NIO_PORT) {
547 error = vmexit_handle_notify(ctx, vme, pvcpu, vme->u.inout.eax);
551 error = emulate_inout(ctx, vcpu, vme, strictio);
553 fprintf(stderr, "Unhandled %s%c 0x%04x at 0x%lx\n",
555 bytes == 1 ? 'b' : (bytes == 2 ? 'w' : 'l'),
557 return (VMEXIT_ABORT);
559 return (VMEXIT_CONTINUE);
564 vmexit_rdmsr(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu)
571 error = emulate_rdmsr(ctx, *pvcpu, vme->u.msr.code, &val);
573 fprintf(stderr, "rdmsr to register %#x on vcpu %d\n",
574 vme->u.msr.code, *pvcpu);
576 vm_inject_gp(ctx, *pvcpu);
577 return (VMEXIT_CONTINUE);
582 error = vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RAX, eax);
586 error = vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RDX, edx);
589 return (VMEXIT_CONTINUE);
593 vmexit_wrmsr(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu)
597 error = emulate_wrmsr(ctx, *pvcpu, vme->u.msr.code, vme->u.msr.wval);
599 fprintf(stderr, "wrmsr to register %#x(%#lx) on vcpu %d\n",
600 vme->u.msr.code, vme->u.msr.wval, *pvcpu);
602 vm_inject_gp(ctx, *pvcpu);
603 return (VMEXIT_CONTINUE);
606 return (VMEXIT_CONTINUE);
610 vmexit_spinup_ap(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu)
613 (void)spinup_ap(ctx, *pvcpu,
614 vme->u.spinup_ap.vcpu, vme->u.spinup_ap.rip);
616 return (VMEXIT_CONTINUE);
619 #define DEBUG_EPT_MISCONFIG
620 #ifdef DEBUG_EPT_MISCONFIG
621 #define VMCS_GUEST_PHYSICAL_ADDRESS 0x00002400
623 static uint64_t ept_misconfig_gpa, ept_misconfig_pte[4];
624 static int ept_misconfig_ptenum;
628 vmexit_vmx_desc(uint32_t exit_reason)
631 if (exit_reason >= nitems(vmx_exit_reason_desc) ||
632 vmx_exit_reason_desc[exit_reason] == NULL)
634 return (vmx_exit_reason_desc[exit_reason]);
638 vmexit_vmx(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
641 fprintf(stderr, "vm exit[%d]\n", *pvcpu);
642 fprintf(stderr, "\treason\t\tVMX\n");
643 fprintf(stderr, "\trip\t\t0x%016lx\n", vmexit->rip);
644 fprintf(stderr, "\tinst_length\t%d\n", vmexit->inst_length);
645 fprintf(stderr, "\tstatus\t\t%d\n", vmexit->u.vmx.status);
646 fprintf(stderr, "\texit_reason\t%u (%s)\n", vmexit->u.vmx.exit_reason,
647 vmexit_vmx_desc(vmexit->u.vmx.exit_reason));
648 fprintf(stderr, "\tqualification\t0x%016lx\n",
649 vmexit->u.vmx.exit_qualification);
650 fprintf(stderr, "\tinst_type\t\t%d\n", vmexit->u.vmx.inst_type);
651 fprintf(stderr, "\tinst_error\t\t%d\n", vmexit->u.vmx.inst_error);
652 #ifdef DEBUG_EPT_MISCONFIG
653 if (vmexit->u.vmx.exit_reason == EXIT_REASON_EPT_MISCONFIG) {
654 vm_get_register(ctx, *pvcpu,
655 VMCS_IDENT(VMCS_GUEST_PHYSICAL_ADDRESS),
657 vm_get_gpa_pmap(ctx, ept_misconfig_gpa, ept_misconfig_pte,
658 &ept_misconfig_ptenum);
659 fprintf(stderr, "\tEPT misconfiguration:\n");
660 fprintf(stderr, "\t\tGPA: %#lx\n", ept_misconfig_gpa);
661 fprintf(stderr, "\t\tPTE(%d): %#lx %#lx %#lx %#lx\n",
662 ept_misconfig_ptenum, ept_misconfig_pte[0],
663 ept_misconfig_pte[1], ept_misconfig_pte[2],
664 ept_misconfig_pte[3]);
666 #endif /* DEBUG_EPT_MISCONFIG */
667 return (VMEXIT_ABORT);
671 vmexit_svm(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
674 fprintf(stderr, "vm exit[%d]\n", *pvcpu);
675 fprintf(stderr, "\treason\t\tSVM\n");
676 fprintf(stderr, "\trip\t\t0x%016lx\n", vmexit->rip);
677 fprintf(stderr, "\tinst_length\t%d\n", vmexit->inst_length);
678 fprintf(stderr, "\texitcode\t%#lx\n", vmexit->u.svm.exitcode);
679 fprintf(stderr, "\texitinfo1\t%#lx\n", vmexit->u.svm.exitinfo1);
680 fprintf(stderr, "\texitinfo2\t%#lx\n", vmexit->u.svm.exitinfo2);
681 return (VMEXIT_ABORT);
685 vmexit_bogus(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
688 assert(vmexit->inst_length == 0);
690 stats.vmexit_bogus++;
692 return (VMEXIT_CONTINUE);
696 vmexit_reqidle(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
699 assert(vmexit->inst_length == 0);
701 stats.vmexit_reqidle++;
703 return (VMEXIT_CONTINUE);
707 vmexit_hlt(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
713 * Just continue execution with the next instruction. We use
714 * the HLT VM exit as a way to be friendly with the host
717 return (VMEXIT_CONTINUE);
721 vmexit_pause(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
724 stats.vmexit_pause++;
726 return (VMEXIT_CONTINUE);
730 vmexit_mtrap(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
733 assert(vmexit->inst_length == 0);
735 stats.vmexit_mtrap++;
737 #ifdef BHYVE_SNAPSHOT
738 checkpoint_cpu_suspend(*pvcpu);
741 gdb_cpu_mtrap(*pvcpu);
742 #ifdef BHYVE_SNAPSHOT
743 checkpoint_cpu_resume(*pvcpu);
746 return (VMEXIT_CONTINUE);
750 vmexit_inst_emul(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
754 enum vm_cpu_mode mode;
756 stats.vmexit_inst_emul++;
758 vie = &vmexit->u.inst_emul.vie;
761 * Attempt to decode in userspace as a fallback. This allows
762 * updating instruction decode in bhyve without rebooting the
763 * kernel (rapid prototyping), albeit with much slower
767 mode = vmexit->u.inst_emul.paging.cpu_mode;
768 cs_d = vmexit->u.inst_emul.cs_d;
769 if (vmm_decode_instruction(mode, cs_d, vie) != 0)
771 if (vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RIP,
772 vmexit->rip + vie->num_processed) != 0)
776 err = emulate_mem(ctx, *pvcpu, vmexit->u.inst_emul.gpa,
777 vie, &vmexit->u.inst_emul.paging);
781 EPRINTLN("Unhandled memory access to 0x%lx\n",
782 vmexit->u.inst_emul.gpa);
787 return (VMEXIT_CONTINUE);
790 fprintf(stderr, "Failed to emulate instruction sequence [ ");
791 for (i = 0; i < vie->num_valid; i++)
792 fprintf(stderr, "%02x", vie->inst[i]);
793 FPRINTLN(stderr, " ] at 0x%lx", vmexit->rip);
794 return (VMEXIT_ABORT);
797 static pthread_mutex_t resetcpu_mtx = PTHREAD_MUTEX_INITIALIZER;
798 static pthread_cond_t resetcpu_cond = PTHREAD_COND_INITIALIZER;
801 vmexit_suspend(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
803 enum vm_suspend_how how;
805 how = vmexit->u.suspended.how;
807 fbsdrun_deletecpu(ctx, *pvcpu);
810 pthread_mutex_lock(&resetcpu_mtx);
811 pthread_cond_signal(&resetcpu_cond);
812 pthread_mutex_unlock(&resetcpu_mtx);
816 pthread_mutex_lock(&resetcpu_mtx);
817 while (!CPU_EMPTY(&cpumask)) {
818 pthread_cond_wait(&resetcpu_cond, &resetcpu_mtx);
820 pthread_mutex_unlock(&resetcpu_mtx);
823 case VM_SUSPEND_RESET:
825 case VM_SUSPEND_POWEROFF:
826 if (destroy_on_poweroff)
829 case VM_SUSPEND_HALT:
831 case VM_SUSPEND_TRIPLEFAULT:
834 fprintf(stderr, "vmexit_suspend: invalid reason %d\n", how);
837 return (0); /* NOTREACHED */
841 vmexit_debug(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
844 #ifdef BHYVE_SNAPSHOT
845 checkpoint_cpu_suspend(*pvcpu);
848 gdb_cpu_suspend(*pvcpu);
849 #ifdef BHYVE_SNAPSHOT
850 checkpoint_cpu_resume(*pvcpu);
852 return (VMEXIT_CONTINUE);
856 vmexit_breakpoint(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
860 fprintf(stderr, "vm_loop: unexpected VMEXIT_DEBUG\n");
863 gdb_cpu_breakpoint(*pvcpu, vmexit);
864 return (VMEXIT_CONTINUE);
867 static vmexit_handler_t handler[VM_EXITCODE_MAX] = {
868 [VM_EXITCODE_INOUT] = vmexit_inout,
869 [VM_EXITCODE_INOUT_STR] = vmexit_inout,
870 [VM_EXITCODE_VMX] = vmexit_vmx,
871 [VM_EXITCODE_SVM] = vmexit_svm,
872 [VM_EXITCODE_BOGUS] = vmexit_bogus,
873 [VM_EXITCODE_REQIDLE] = vmexit_reqidle,
874 [VM_EXITCODE_RDMSR] = vmexit_rdmsr,
875 [VM_EXITCODE_WRMSR] = vmexit_wrmsr,
876 [VM_EXITCODE_MTRAP] = vmexit_mtrap,
877 [VM_EXITCODE_INST_EMUL] = vmexit_inst_emul,
878 [VM_EXITCODE_SPINUP_AP] = vmexit_spinup_ap,
879 [VM_EXITCODE_SUSPENDED] = vmexit_suspend,
880 [VM_EXITCODE_TASK_SWITCH] = vmexit_task_switch,
881 [VM_EXITCODE_DEBUG] = vmexit_debug,
882 [VM_EXITCODE_BPT] = vmexit_breakpoint,
886 vm_loop(struct vmctx *ctx, int vcpu, uint64_t startrip)
889 enum vm_exitcode exitcode;
890 cpuset_t active_cpus;
892 if (vcpumap[vcpu] != NULL) {
893 error = pthread_setaffinity_np(pthread_self(),
894 sizeof(cpuset_t), vcpumap[vcpu]);
898 error = vm_active_cpus(ctx, &active_cpus);
899 assert(CPU_ISSET(vcpu, &active_cpus));
901 error = vm_set_register(ctx, vcpu, VM_REG_GUEST_RIP, startrip);
905 error = vm_run(ctx, vcpu, &vmexit[vcpu]);
909 exitcode = vmexit[vcpu].exitcode;
910 if (exitcode >= VM_EXITCODE_MAX || handler[exitcode] == NULL) {
911 fprintf(stderr, "vm_loop: unexpected exitcode 0x%x\n",
916 rc = (*handler[exitcode])(ctx, &vmexit[vcpu], &vcpu);
919 case VMEXIT_CONTINUE:
927 fprintf(stderr, "vm_run error %d, errno %d\n", error, errno);
931 num_vcpus_allowed(struct vmctx *ctx)
935 error = vm_get_capability(ctx, BSP, VM_CAP_UNRESTRICTED_GUEST, &tmp);
938 * The guest is allowed to spinup more than one processor only if the
939 * UNRESTRICTED_GUEST capability is available.
948 fbsdrun_set_capabilities(struct vmctx *ctx, int cpu)
952 if (fbsdrun_vmexit_on_hlt()) {
953 err = vm_get_capability(ctx, cpu, VM_CAP_HALT_EXIT, &tmp);
955 fprintf(stderr, "VM exit on HLT not supported\n");
958 vm_set_capability(ctx, cpu, VM_CAP_HALT_EXIT, 1);
960 handler[VM_EXITCODE_HLT] = vmexit_hlt;
963 if (fbsdrun_vmexit_on_pause()) {
965 * pause exit support required for this mode
967 err = vm_get_capability(ctx, cpu, VM_CAP_PAUSE_EXIT, &tmp);
970 "SMP mux requested, no pause support\n");
973 vm_set_capability(ctx, cpu, VM_CAP_PAUSE_EXIT, 1);
975 handler[VM_EXITCODE_PAUSE] = vmexit_pause;
979 err = vm_set_x2apic_state(ctx, cpu, X2APIC_ENABLED);
981 err = vm_set_x2apic_state(ctx, cpu, X2APIC_DISABLED);
984 fprintf(stderr, "Unable to set x2apic state (%d)\n", err);
988 vm_set_capability(ctx, cpu, VM_CAP_ENABLE_INVPCID, 1);
991 static struct vmctx *
992 do_open(const char *vmname)
996 bool reinit, romboot;
997 #ifndef WITHOUT_CAPSICUM
999 const cap_ioctl_t *cmds;
1003 reinit = romboot = false;
1008 error = vm_create(vmname);
1010 if (errno == EEXIST) {
1015 * The virtual machine has been setup by the
1016 * userspace bootloader.
1020 perror("vm_create");
1026 * If the virtual machine was just created then a
1027 * bootrom must be configured to boot it.
1029 fprintf(stderr, "virtual machine cannot be booted\n");
1034 ctx = vm_open(vmname);
1040 #ifndef WITHOUT_CAPSICUM
1041 cap_rights_init(&rights, CAP_IOCTL, CAP_MMAP_RW);
1042 if (caph_rights_limit(vm_get_device_fd(ctx), &rights) == -1)
1043 errx(EX_OSERR, "Unable to apply rights for sandbox");
1044 vm_get_ioctls(&ncmds);
1045 cmds = vm_get_ioctls(NULL);
1047 errx(EX_OSERR, "out of memory");
1048 if (caph_ioctls_limit(vm_get_device_fd(ctx), cmds, ncmds) == -1)
1049 errx(EX_OSERR, "Unable to apply rights for sandbox");
1050 free((cap_ioctl_t *)cmds);
1054 error = vm_reinit(ctx);
1056 perror("vm_reinit");
1060 error = vm_set_topology(ctx, sockets, cores, threads, maxcpus);
1062 errx(EX_OSERR, "vm_set_topology");
1067 spinup_vcpu(struct vmctx *ctx, int vcpu)
1072 error = vm_get_register(ctx, vcpu, VM_REG_GUEST_RIP, &rip);
1075 fbsdrun_set_capabilities(ctx, vcpu);
1076 error = vm_set_capability(ctx, vcpu, VM_CAP_UNRESTRICTED_GUEST, 1);
1079 fbsdrun_addcpu(ctx, BSP, vcpu, rip);
1083 main(int argc, char *argv[])
1085 int c, error, dbg_port, err, bvmcons;
1086 int max_vcpus, mptgen, memflags;
1093 #ifdef BHYVE_SNAPSHOT
1095 struct restore_state rstate;
1098 restore_file = NULL;
1102 progname = basename(argv[0]);
1106 sockets = cores = threads = 1;
1113 #ifdef BHYVE_SNAPSHOT
1114 optstr = "abehuwxACDHIPSWYp:g:G:c:s:m:l:U:r:";
1116 optstr = "abehuwxACDHIPSWYp:g:G:c:s:m:l:U:";
1118 while ((c = getopt(argc, argv, optstr)) != -1) {
1127 warnx("-b flag is deprecated and will be removed in FreeBSD 13.0");
1131 destroy_on_poweroff = 1;
1134 if (pincpu_parse(optarg) != 0) {
1135 errx(EX_USAGE, "invalid vcpu pinning "
1136 "configuration '%s'", optarg);
1140 if (topology_parse(optarg) != 0) {
1141 errx(EX_USAGE, "invalid cpu topology "
1146 memflags |= VM_MEM_F_INCORE;
1149 warnx("-g flag is deprecated and will be removed in FreeBSD 13.0");
1150 dbg_port = atoi(optarg);
1153 if (optarg[0] == 'w') {
1157 gdb_port = atoi(optarg);
1160 if (strncmp(optarg, "help", strlen(optarg)) == 0) {
1161 lpc_print_supported_devices();
1163 } else if (lpc_device_parse(optarg) != 0) {
1164 errx(EX_USAGE, "invalid lpc device "
1165 "configuration '%s'", optarg);
1168 #ifdef BHYVE_SNAPSHOT
1170 restore_file = optarg;
1174 if (strncmp(optarg, "help", strlen(optarg)) == 0) {
1175 pci_print_supported_devices();
1177 } else if (pci_parse_slot(optarg) != 0)
1182 memflags |= VM_MEM_F_WIRED;
1185 error = vm_parse_memsize(optarg, &memsize);
1187 errx(EX_USAGE, "invalid memsize '%s'", optarg);
1190 guest_vmexit_on_hlt = 1;
1194 * The "-I" option was used to add an ioapic to the
1197 * An ioapic is now provided unconditionally for each
1198 * virtual machine and this option is now deprecated.
1202 guest_vmexit_on_pause = 1;
1211 guest_uuid_str = optarg;
1234 #ifdef BHYVE_SNAPSHOT
1235 if (argc > 1 || (argc == 0 && restore_file == NULL))
1238 if (restore_file != NULL) {
1239 error = load_restore_file(restore_file, &rstate);
1241 fprintf(stderr, "Failed to read checkpoint info from "
1242 "file: '%s'.\n", restore_file);
1250 vmname = lookup_vmname(&rstate);
1251 if (vmname == NULL) {
1252 fprintf(stderr, "Cannot find VM name in restore file. "
1253 "Please specify one.\n");
1263 ctx = do_open(vmname);
1265 #ifdef BHYVE_SNAPSHOT
1266 if (restore_file != NULL) {
1267 guest_ncpus = lookup_guest_ncpus(&rstate);
1268 memflags = lookup_memflags(&rstate);
1269 memsize = lookup_memsize(&rstate);
1272 if (guest_ncpus < 1) {
1273 fprintf(stderr, "Invalid guest vCPUs (%d)\n", guest_ncpus);
1278 max_vcpus = num_vcpus_allowed(ctx);
1279 if (guest_ncpus > max_vcpus) {
1280 fprintf(stderr, "%d vCPUs requested but only %d available\n",
1281 guest_ncpus, max_vcpus);
1285 fbsdrun_set_capabilities(ctx, BSP);
1287 vm_set_memflags(ctx, memflags);
1288 err = vm_setup_memory(ctx, memsize, VM_MMAP_ALL);
1290 fprintf(stderr, "Unable to setup memory (%d)\n", errno);
1296 fprintf(stderr, "init_msr error %d", error);
1308 rtc_init(ctx, rtc_localtime);
1312 * Exit if a device emulation finds an error in its initilization
1314 if (init_pci(ctx) != 0) {
1315 perror("device emulation initialization error");
1320 * Initialize after PCI, to allow a bootrom file to reserve the high
1327 init_dbgport(dbg_port);
1330 init_gdb(ctx, gdb_port, gdb_stop);
1335 if (lpc_bootrom()) {
1336 if (vm_set_capability(ctx, BSP, VM_CAP_UNRESTRICTED_GUEST, 1)) {
1337 fprintf(stderr, "ROM boot failed: unrestricted guest "
1338 "capability not available\n");
1341 error = vcpu_reset(ctx, BSP);
1345 #ifdef BHYVE_SNAPSHOT
1346 if (restore_file != NULL) {
1347 fprintf(stdout, "Pausing pci devs...\r\n");
1348 if (vm_pause_user_devs(ctx) != 0) {
1349 fprintf(stderr, "Failed to pause PCI device state.\n");
1353 fprintf(stdout, "Restoring vm mem...\r\n");
1354 if (restore_vm_mem(ctx, &rstate) != 0) {
1355 fprintf(stderr, "Failed to restore VM memory.\n");
1359 fprintf(stdout, "Restoring pci devs...\r\n");
1360 if (vm_restore_user_devs(ctx, &rstate) != 0) {
1361 fprintf(stderr, "Failed to restore PCI device state.\n");
1365 fprintf(stdout, "Restoring kernel structs...\r\n");
1366 if (vm_restore_kern_structs(ctx, &rstate) != 0) {
1367 fprintf(stderr, "Failed to restore kernel structs.\n");
1371 fprintf(stdout, "Resuming pci devs...\r\n");
1372 if (vm_resume_user_devs(ctx) != 0) {
1373 fprintf(stderr, "Failed to resume PCI device state.\n");
1379 error = vm_get_register(ctx, BSP, VM_REG_GUEST_RIP, &rip);
1383 * build the guest tables, MP etc.
1386 error = mptable_build(ctx, guest_ncpus);
1388 perror("error to build the guest tables");
1393 error = smbios_build(ctx);
1397 error = acpi_build(ctx, guest_ncpus);
1405 * Change the proc title to include the VM name.
1407 setproctitle("%s", vmname);
1409 #ifndef WITHOUT_CAPSICUM
1410 caph_cache_catpages();
1412 if (caph_limit_stdout() == -1 || caph_limit_stderr() == -1)
1413 errx(EX_OSERR, "Unable to apply rights for sandbox");
1415 if (caph_enter() == -1)
1416 errx(EX_OSERR, "cap_enter() failed");
1419 #ifdef BHYVE_SNAPSHOT
1420 if (restore_file != NULL)
1421 destroy_restore_state(&rstate);
1424 * checkpointing thread for communication with bhyvectl
1426 if (init_checkpoint_thread(ctx) < 0)
1427 printf("Failed to start checkpoint thread!\r\n");
1429 if (restore_file != NULL)
1430 vm_restore_time(ctx);
1436 fbsdrun_addcpu(ctx, BSP, BSP, rip);
1438 #ifdef BHYVE_SNAPSHOT
1440 * If we restore a VM, start all vCPUs now (including APs), otherwise,
1441 * let the guest OS to spin them up later via vmexits.
1443 if (restore_file != NULL) {
1444 for (vcpu = 0; vcpu < guest_ncpus; vcpu++) {
1448 fprintf(stdout, "spinning up vcpu no %d...\r\n", vcpu);
1449 spinup_vcpu(ctx, vcpu);
1455 * Head off to the main event dispatch loop