bhyverun.c revision 242131 
1/*-
2 * Copyright (c) 2011 NetApp, Inc.
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in the
12 *    documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED.  IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 *
26 * $FreeBSD$
27 */
28
29#include <sys/cdefs.h>
30__FBSDID("$FreeBSD$");
31
32#include <sys/types.h>
33#include <sys/mman.h>
34#include <sys/time.h>
35
36#include <machine/segments.h>
37
38#include <stdio.h>
39#include <stdlib.h>
40#include <libgen.h>
41#include <unistd.h>
42#include <assert.h>
43#include <errno.h>
44#include <signal.h>
45#include <pthread.h>
46
47#include <machine/vmm.h>
48#include <vmmapi.h>
49
50#include "fbsdrun.h"
51#include "inout.h"
52#include "dbgport.h"
53#include "mem.h"
54#include "mevent.h"
55#include "mptbl.h"
56#include "pci_emul.h"
57#include "xmsr.h"
58#include "instruction_emul.h"
59#include "ioapic.h"
60#include "spinup_ap.h"
61
62#define	DEFAULT_GUEST_HZ	100
63#define	DEFAULT_GUEST_TSLICE	200
64
65#define GUEST_NIO_PORT		0x488	/* guest upcalls via i/o port */
66
67#define	VMEXIT_SWITCH		0	/* force vcpu switch in mux mode */
68#define	VMEXIT_CONTINUE		1	/* continue from next instruction */
69#define	VMEXIT_RESTART		2	/* restart current instruction */
70#define	VMEXIT_ABORT		3	/* abort the vm run loop */
71#define	VMEXIT_RESET		4	/* guest machine has reset */
72
73#define MB		(1024UL * 1024)
74#define GB		(1024UL * MB)
75
76typedef int (*vmexit_handler_t)(struct vmctx *, struct vm_exit *, int *vcpu);
77
78int guest_tslice = DEFAULT_GUEST_TSLICE;
79int guest_hz = DEFAULT_GUEST_HZ;
80char *vmname;
81
82u_long lomem_sz;
83u_long himem_sz;
84
85int guest_ncpus;
86
87static int pincpu = -1;
88static int guest_vcpu_mux;
89static int guest_vmexit_on_hlt, guest_vmexit_on_pause, disable_x2apic;
90
91static int foundcpus;
92
93static int strictio;
94
95static char *lomem_addr;
96static char *himem_addr;
97
98static char *progname;
99static const int BSP = 0;
100
101static int cpumask;
102
103static void vm_loop(struct vmctx *ctx, int vcpu, uint64_t rip);
104
105struct vm_exit vmexit[VM_MAXCPU];
106
107struct fbsdstats {
108        uint64_t        vmexit_bogus;
109        uint64_t        vmexit_bogus_switch;
110        uint64_t        vmexit_hlt;
111        uint64_t        vmexit_pause;
112        uint64_t        vmexit_mtrap;
113        uint64_t        vmexit_paging;
114        uint64_t        cpu_switch_rotate;
115        uint64_t        cpu_switch_direct;
116        int             io_reset;
117} stats;
118
119struct mt_vmm_info {
120	pthread_t	mt_thr;
121	struct vmctx	*mt_ctx;
122	int		mt_vcpu;
123} mt_vmm_info[VM_MAXCPU];
124
125static void
126usage(int code)
127{
128
129        fprintf(stderr,
130                "Usage: %s [-aehBHIP][-g <gdb port>][-z <hz>][-s <pci>]"
131		"[-S <pci>][-p pincpu][-n <pci>][-m lowmem][-M highmem] <vm>\n"
132		"       -a: local apic is in XAPIC mode (default is X2APIC)\n"
133		"       -g: gdb port (default is %d and 0 means don't open)\n"
134		"       -c: # cpus (default 1)\n"
135		"       -p: pin vcpu 'n' to host cpu 'pincpu + n'\n"
136		"       -B: inject breakpoint exception on vm entry\n"
137		"       -H: vmexit from the guest on hlt\n"
138		"       -I: present an ioapic to the guest\n"
139		"       -P: vmexit from the guest on pause\n"
140		"	-e: exit on unhandled i/o access\n"
141		"       -h: help\n"
142		"       -z: guest hz (default is %d)\n"
143		"       -s: <slot,driver,configinfo> PCI slot config\n"
144		"       -S: <slot,driver,configinfo> legacy PCI slot config\n"
145		"       -m: lowmem in MB\n"
146		"       -M: highmem in MB\n"
147		"       -x: mux vcpus to 1 hcpu\n"
148		"       -t: mux vcpu timeslice hz (default %d)\n",
149		progname, DEFAULT_GDB_PORT, DEFAULT_GUEST_HZ,
150		DEFAULT_GUEST_TSLICE);
151	exit(code);
152}
153
154void *
155paddr_guest2host(uintptr_t gaddr)
156{
157	if (lomem_sz == 0)
158		return (NULL);
159
160	if (gaddr < lomem_sz) {
161		return ((void *)(lomem_addr + gaddr));
162	} else if (gaddr >= 4*GB && gaddr < (4*GB + himem_sz)) {
163		return ((void *)(himem_addr + gaddr - 4*GB));
164	} else
165		return (NULL);
166}
167
168int
169fbsdrun_disable_x2apic(void)
170{
171
172	return (disable_x2apic);
173}
174
175int
176fbsdrun_vmexit_on_pause(void)
177{
178
179	return (guest_vmexit_on_pause);
180}
181
182int
183fbsdrun_vmexit_on_hlt(void)
184{
185
186	return (guest_vmexit_on_hlt);
187}
188
189int
190fbsdrun_muxed(void)
191{
192
193	return (guest_vcpu_mux);
194}
195
196static void *
197fbsdrun_start_thread(void *param)
198{
199	int vcpu;
200	struct mt_vmm_info *mtp = param;
201
202	vcpu = mtp->mt_vcpu;
203	vm_loop(mtp->mt_ctx, vcpu, vmexit[vcpu].rip);
204
205	/* not reached */
206	exit(1);
207	return (NULL);
208}
209
210void
211fbsdrun_addcpu(struct vmctx *ctx, int vcpu, uint64_t rip)
212{
213	int error;
214
215	if (cpumask & (1 << vcpu)) {
216		printf("addcpu: attempting to add existing cpu %d\n", vcpu);
217		exit(1);
218	}
219
220	cpumask |= 1 << vcpu;
221	foundcpus++;
222
223	/*
224	 * Set up the vmexit struct to allow execution to start
225	 * at the given RIP
226	 */
227	vmexit[vcpu].rip = rip;
228	vmexit[vcpu].inst_length = 0;
229
230	if (vcpu == BSP || !guest_vcpu_mux){
231		mt_vmm_info[vcpu].mt_ctx = ctx;
232		mt_vmm_info[vcpu].mt_vcpu = vcpu;
233
234		error = pthread_create(&mt_vmm_info[vcpu].mt_thr, NULL,
235				fbsdrun_start_thread, &mt_vmm_info[vcpu]);
236		assert(error == 0);
237	}
238}
239
240static int
241fbsdrun_get_next_cpu(int curcpu)
242{
243
244	/*
245	 * Get the next available CPU. Assumes they arrive
246	 * in ascending order with no gaps.
247	 */
248	return ((curcpu + 1) % foundcpus);
249}
250
251static int
252vmexit_catch_reset(void)
253{
254        stats.io_reset++;
255        return (VMEXIT_RESET);
256}
257
258static int
259vmexit_catch_inout(void)
260{
261	return (VMEXIT_ABORT);
262}
263
264static int
265vmexit_handle_notify(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu,
266		     uint32_t eax)
267{
268#if PG_DEBUG /* put all types of debug here */
269        if (eax == 0) {
270		pause_noswitch = 1;
271	} else if (eax == 1) {
272		pause_noswitch = 0;
273	} else {
274		pause_noswitch = 0;
275		if (eax == 5) {
276			vm_set_capability(ctx, *pvcpu, VM_CAP_MTRAP_EXIT, 1);
277		}
278	}
279#endif
280        return (VMEXIT_CONTINUE);
281}
282
283static int
284vmexit_inout(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu)
285{
286	int error;
287	int bytes, port, in, out;
288	uint32_t eax;
289	int vcpu;
290
291	vcpu = *pvcpu;
292
293	port = vme->u.inout.port;
294	bytes = vme->u.inout.bytes;
295	eax = vme->u.inout.eax;
296	in = vme->u.inout.in;
297	out = !in;
298
299	/* We don't deal with these */
300	if (vme->u.inout.string || vme->u.inout.rep)
301		return (VMEXIT_ABORT);
302
303	/* Special case of guest reset */
304	if (out && port == 0x64 && (uint8_t)eax == 0xFE)
305		return (vmexit_catch_reset());
306
307        /* Extra-special case of host notifications */
308        if (out && port == GUEST_NIO_PORT)
309                return (vmexit_handle_notify(ctx, vme, pvcpu, eax));
310
311	error = emulate_inout(ctx, vcpu, in, port, bytes, &eax, strictio);
312	if (error == 0 && in)
313		error = vm_set_register(ctx, vcpu, VM_REG_GUEST_RAX, eax);
314
315	if (error == 0)
316		return (VMEXIT_CONTINUE);
317	else {
318		fprintf(stderr, "Unhandled %s%c 0x%04x\n",
319			in ? "in" : "out",
320			bytes == 1 ? 'b' : (bytes == 2 ? 'w' : 'l'), port);
321		return (vmexit_catch_inout());
322	}
323}
324
325static int
326vmexit_rdmsr(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu)
327{
328	printf("vm exit rdmsr 0x%x, cpu %d\n", vme->u.msr.code, *pvcpu);
329	return (VMEXIT_ABORT);
330}
331
332static int
333vmexit_wrmsr(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu)
334{
335	int newcpu;
336	int retval = VMEXIT_CONTINUE;
337
338	newcpu = emulate_wrmsr(ctx, *pvcpu, vme->u.msr.code,vme->u.msr.wval);
339
340	if (guest_vcpu_mux && *pvcpu != newcpu) {
341                retval = VMEXIT_SWITCH;
342                *pvcpu = newcpu;
343        }
344
345        return (retval);
346}
347
348static int
349vmexit_spinup_ap(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu)
350{
351	int newcpu;
352	int retval = VMEXIT_CONTINUE;
353
354	newcpu = spinup_ap(ctx, *pvcpu,
355			   vme->u.spinup_ap.vcpu, vme->u.spinup_ap.rip);
356
357	if (guest_vcpu_mux && *pvcpu != newcpu) {
358		retval = VMEXIT_SWITCH;
359		*pvcpu = newcpu;
360	}
361
362	return (retval);
363}
364
365static int
366vmexit_vmx(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
367{
368
369	printf("vm exit[%d]\n", *pvcpu);
370	printf("\treason\t\tVMX\n");
371	printf("\trip\t\t0x%016lx\n", vmexit->rip);
372	printf("\tinst_length\t%d\n", vmexit->inst_length);
373	printf("\terror\t\t%d\n", vmexit->u.vmx.error);
374	printf("\texit_reason\t%u\n", vmexit->u.vmx.exit_reason);
375	printf("\tqualification\t0x%016lx\n", vmexit->u.vmx.exit_qualification);
376
377	return (VMEXIT_ABORT);
378}
379
380static int bogus_noswitch = 1;
381
382static int
383vmexit_bogus(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
384{
385	stats.vmexit_bogus++;
386
387	if (!guest_vcpu_mux || guest_ncpus == 1 || bogus_noswitch) {
388		return (VMEXIT_RESTART);
389	} else {
390		stats.vmexit_bogus_switch++;
391		vmexit->inst_length = 0;
392		*pvcpu = -1;
393		return (VMEXIT_SWITCH);
394	}
395}
396
397static int
398vmexit_hlt(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
399{
400	stats.vmexit_hlt++;
401	if (fbsdrun_muxed()) {
402		*pvcpu = -1;
403		return (VMEXIT_SWITCH);
404	} else {
405		/*
406		 * Just continue execution with the next instruction. We use
407		 * the HLT VM exit as a way to be friendly with the host
408		 * scheduler.
409		 */
410		return (VMEXIT_CONTINUE);
411	}
412}
413
414static int pause_noswitch;
415
416static int
417vmexit_pause(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
418{
419	stats.vmexit_pause++;
420
421	if (fbsdrun_muxed() && !pause_noswitch) {
422		*pvcpu = -1;
423		return (VMEXIT_SWITCH);
424        } else {
425		return (VMEXIT_CONTINUE);
426	}
427}
428
429static int
430vmexit_mtrap(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
431{
432	stats.vmexit_mtrap++;
433
434	return (VMEXIT_RESTART);
435}
436
437static int
438vmexit_paging(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
439{
440	int err;
441	stats.vmexit_paging++;
442
443	err = emulate_mem(ctx, *pvcpu, vmexit->u.paging.gpa, vmexit->rip,
444			  vmexit->u.paging.cr3, vmexit->u.paging.rwx);
445
446	if (err) {
447		if (err == EINVAL) {
448			printf("Failed to emulate instruction at 0x%lx\n",
449			       vmexit->rip);
450		} else if (err == ESRCH) {
451			printf("Unhandled memory access to 0x%lx\n",
452			       vmexit->u.paging.gpa);
453		}
454
455		return (VMEXIT_ABORT);
456	}
457
458	return (VMEXIT_CONTINUE);
459}
460
461static void
462sigalrm(int sig)
463{
464	return;
465}
466
467static void
468setup_timeslice(void)
469{
470	struct sigaction sa;
471	struct itimerval itv;
472	int error;
473
474	/*
475	 * Setup a realtime timer to generate a SIGALRM at a
476	 * frequency of 'guest_tslice' ticks per second.
477	 */
478	sigemptyset(&sa.sa_mask);
479	sa.sa_flags = 0;
480	sa.sa_handler = sigalrm;
481
482	error = sigaction(SIGALRM, &sa, NULL);
483	assert(error == 0);
484
485	itv.it_interval.tv_sec = 0;
486	itv.it_interval.tv_usec = 1000000 / guest_tslice;
487	itv.it_value.tv_sec = 0;
488	itv.it_value.tv_usec = 1000000 / guest_tslice;
489
490	error = setitimer(ITIMER_REAL, &itv, NULL);
491	assert(error == 0);
492}
493
494static vmexit_handler_t handler[VM_EXITCODE_MAX] = {
495	[VM_EXITCODE_INOUT]  = vmexit_inout,
496	[VM_EXITCODE_VMX]    = vmexit_vmx,
497	[VM_EXITCODE_BOGUS]  = vmexit_bogus,
498	[VM_EXITCODE_RDMSR]  = vmexit_rdmsr,
499	[VM_EXITCODE_WRMSR]  = vmexit_wrmsr,
500	[VM_EXITCODE_MTRAP]  = vmexit_mtrap,
501	[VM_EXITCODE_PAGING] = vmexit_paging,
502	[VM_EXITCODE_SPINUP_AP] = vmexit_spinup_ap,
503};
504
505static void
506vm_loop(struct vmctx *ctx, int vcpu, uint64_t rip)
507{
508	int error, rc, prevcpu;
509
510	if (guest_vcpu_mux)
511		setup_timeslice();
512
513	if (pincpu >= 0) {
514		error = vm_set_pinning(ctx, vcpu, pincpu + vcpu);
515		assert(error == 0);
516	}
517
518	while (1) {
519		error = vm_run(ctx, vcpu, rip, &vmexit[vcpu]);
520		if (error != 0) {
521			/*
522			 * It is possible that 'vmmctl' or some other process
523			 * has transitioned the vcpu to CANNOT_RUN state right
524			 * before we tried to transition it to RUNNING.
525			 *
526			 * This is expected to be temporary so just retry.
527			 */
528			if (errno == EBUSY)
529				continue;
530			else
531				break;
532		}
533
534		prevcpu = vcpu;
535                rc = (*handler[vmexit[vcpu].exitcode])(ctx, &vmexit[vcpu],
536                                                       &vcpu);
537		switch (rc) {
538                case VMEXIT_SWITCH:
539			assert(guest_vcpu_mux);
540			if (vcpu == -1) {
541				stats.cpu_switch_rotate++;
542				vcpu = fbsdrun_get_next_cpu(prevcpu);
543			} else {
544				stats.cpu_switch_direct++;
545			}
546			/* fall through */
547		case VMEXIT_CONTINUE:
548                        rip = vmexit[vcpu].rip + vmexit[vcpu].inst_length;
549			break;
550		case VMEXIT_RESTART:
551                        rip = vmexit[vcpu].rip;
552			break;
553		case VMEXIT_RESET:
554			exit(0);
555		default:
556			exit(1);
557		}
558	}
559	fprintf(stderr, "vm_run error %d, errno %d\n", error, errno);
560}
561
562
563int
564main(int argc, char *argv[])
565{
566	int c, error, gdb_port, inject_bkpt, tmp, err, ioapic;
567	struct vmctx *ctx;
568	uint64_t rip;
569
570	inject_bkpt = 0;
571	progname = basename(argv[0]);
572	gdb_port = DEFAULT_GDB_PORT;
573	guest_ncpus = 1;
574	ioapic = 0;
575
576	while ((c = getopt(argc, argv, "aehBHIPxp:g:c:z:s:S:n:m:M:")) != -1) {
577		switch (c) {
578		case 'a':
579			disable_x2apic = 1;
580			break;
581		case 'B':
582			inject_bkpt = 1;
583			break;
584		case 'x':
585			guest_vcpu_mux = 1;
586			break;
587		case 'p':
588			pincpu = atoi(optarg);
589			break;
590                case 'c':
591			guest_ncpus = atoi(optarg);
592			break;
593		case 'g':
594			gdb_port = atoi(optarg);
595			break;
596		case 'z':
597			guest_hz = atoi(optarg);
598			break;
599		case 't':
600			guest_tslice = atoi(optarg);
601			break;
602		case 's':
603			pci_parse_slot(optarg, 0);
604			break;
605		case 'S':
606			pci_parse_slot(optarg, 1);
607			break;
608                case 'm':
609			lomem_sz = strtoul(optarg, NULL, 0) * MB;
610			break;
611                case 'M':
612			himem_sz = strtoul(optarg, NULL, 0) * MB;
613			break;
614		case 'H':
615			guest_vmexit_on_hlt = 1;
616			break;
617		case 'I':
618			ioapic = 1;
619			break;
620		case 'P':
621			guest_vmexit_on_pause = 1;
622			break;
623		case 'e':
624			strictio = 1;
625			break;
626		case 'h':
627			usage(0);
628		default:
629			usage(1);
630		}
631	}
632	argc -= optind;
633	argv += optind;
634
635	if (argc != 1)
636		usage(1);
637
638	/* No need to mux if guest is uni-processor */
639	if (guest_ncpus <= 1)
640		guest_vcpu_mux = 0;
641
642	/* vmexit on hlt if guest is muxed */
643	if (guest_vcpu_mux) {
644		guest_vmexit_on_hlt = 1;
645		guest_vmexit_on_pause = 1;
646	}
647
648	vmname = argv[0];
649
650	ctx = vm_open(vmname);
651	if (ctx == NULL) {
652		perror("vm_open");
653		exit(1);
654	}
655
656	if (fbsdrun_vmexit_on_hlt()) {
657		err = vm_get_capability(ctx, BSP, VM_CAP_HALT_EXIT, &tmp);
658		if (err < 0) {
659			printf("VM exit on HLT not supported\n");
660			exit(1);
661		}
662		vm_set_capability(ctx, BSP, VM_CAP_HALT_EXIT, 1);
663		handler[VM_EXITCODE_HLT] = vmexit_hlt;
664	}
665
666        if (fbsdrun_vmexit_on_pause()) {
667		/*
668		 * pause exit support required for this mode
669		 */
670		err = vm_get_capability(ctx, BSP, VM_CAP_PAUSE_EXIT, &tmp);
671		if (err < 0) {
672			printf("SMP mux requested, no pause support\n");
673			exit(1);
674		}
675		vm_set_capability(ctx, BSP, VM_CAP_PAUSE_EXIT, 1);
676		handler[VM_EXITCODE_PAUSE] = vmexit_pause;
677        }
678
679	if (fbsdrun_disable_x2apic())
680		err = vm_set_x2apic_state(ctx, BSP, X2APIC_DISABLED);
681	else
682		err = vm_set_x2apic_state(ctx, BSP, X2APIC_ENABLED);
683
684	if (err) {
685		printf("Unable to set x2apic state (%d)\n", err);
686		exit(1);
687	}
688
689	if (lomem_sz != 0) {
690		lomem_addr = vm_map_memory(ctx, 0, lomem_sz);
691		if (lomem_addr == (char *) MAP_FAILED) {
692			lomem_sz = 0;
693		} else if (himem_sz != 0) {
694			himem_addr = vm_map_memory(ctx, 4*GB, himem_sz);
695			if (himem_addr == (char *) MAP_FAILED) {
696				lomem_sz = 0;
697				himem_sz = 0;
698			}
699		}
700	}
701
702	init_inout();
703	init_pci(ctx);
704	if (ioapic)
705		ioapic_init(0);
706
707	if (gdb_port != 0)
708		init_dbgport(gdb_port);
709
710	error = vm_get_register(ctx, BSP, VM_REG_GUEST_RIP, &rip);
711	assert(error == 0);
712
713	if (inject_bkpt) {
714		error = vm_inject_event(ctx, BSP, VM_HW_EXCEPTION, IDT_BP);
715		assert(error == 0);
716	}
717
718	/*
719	 * build the guest tables, MP etc.
720	 */
721	mptable_build(ctx, guest_ncpus, ioapic);
722
723	/*
724	 * Add CPU 0
725	 */
726	fbsdrun_addcpu(ctx, BSP, rip);
727
728	/*
729	 * Head off to the main event dispatch loop
730	 */
731	mevent_dispatch();
732
733	exit(1);
734}
735