xref: /titanic_52/usr/src/uts/i86xpv/os/xen_machdep.c (revision ad09f8b827db90c9a0093f0b6382803fa64a5fd1)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 /* derived from netbsd's xen_machdep.c 1.1.2.1 */
28 
29 /*
30  *
31  * Copyright (c) 2004 Christian Limpach.
32  * All rights reserved.
33  *
34  * Redistribution and use in source and binary forms, with or without
35  * modification, are permitted provided that the following conditions
36  * are met:
37  * 1. Redistributions of source code must retain the above copyright
38  *    notice, this list of conditions and the following disclaimer.
39  * 2. Redistributions in binary form must reproduce the above copyright
40  *    notice, this list of conditions and the following disclaimer in the
41  *    documentation and/or other materials provided with the distribution.
42  * 3. This section intentionally left blank.
43  * 4. The name of the author may not be used to endorse or promote products
44  *    derived from this software without specific prior written permission.
45  *
46  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
47  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
48  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
49  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
50  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
51  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
52  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
53  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
54  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
55  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
56  */
57 /*
58  * Section 3 of the above license was updated in response to bug 6379571.
59  */
60 
61 #include <sys/xpv_user.h>
62 
63 /* XXX 3.3. TODO remove this include */
64 #include <xen/public/arch-x86/xen-mca.h>
65 
66 #include <sys/ctype.h>
67 #include <sys/types.h>
68 #include <sys/cmn_err.h>
69 #include <sys/trap.h>
70 #include <sys/segments.h>
71 #include <sys/hypervisor.h>
72 #include <sys/xen_mmu.h>
73 #include <sys/machsystm.h>
74 #include <sys/promif.h>
75 #include <sys/bootconf.h>
76 #include <sys/bootinfo.h>
77 #include <sys/cpr.h>
78 #include <sys/taskq.h>
79 #include <sys/uadmin.h>
80 #include <sys/evtchn_impl.h>
81 #include <sys/archsystm.h>
82 #include <xen/sys/xenbus_impl.h>
83 #include <sys/mach_mmu.h>
84 #include <vm/hat_i86.h>
85 #include <sys/gnttab.h>
86 #include <sys/reboot.h>
87 #include <sys/stack.h>
88 #include <sys/clock.h>
89 #include <sys/bitmap.h>
90 #include <sys/processor.h>
91 #include <sys/xen_errno.h>
92 #include <sys/xpv_panic.h>
93 #include <sys/smp_impldefs.h>
94 #include <sys/cpu.h>
95 #include <sys/balloon_impl.h>
96 #include <sys/ddi.h>
97 
98 #ifdef DEBUG
99 #define	SUSPEND_DEBUG if (xen_suspend_debug) xen_printf
100 #else
101 #define	SUSPEND_DEBUG(...)
102 #endif
103 
104 int cpr_debug;
105 cpuset_t cpu_suspend_lost_set;
106 static int xen_suspend_debug;
107 
108 uint_t xen_phys_ncpus;
109 xen_mc_logical_cpu_t *xen_phys_cpus;
110 int xen_physinfo_debug = 0;
111 
112 /*
113  * Determine helpful version information.
114  *
115  * (And leave copies in the data segment so we can look at them later
116  * with e.g. kmdb.)
117  */
118 
119 typedef enum xen_version {
120 	XENVER_BOOT_IDX,
121 	XENVER_CURRENT_IDX
122 } xen_version_t;
123 
124 struct xenver {
125 	ulong_t xv_major;
126 	ulong_t xv_minor;
127 	ulong_t xv_revision;
128 	xen_extraversion_t xv_ver;
129 	ulong_t xv_is_xvm;
130 	xen_changeset_info_t xv_chgset;
131 	xen_compile_info_t xv_build;
132 	xen_capabilities_info_t xv_caps;
133 } xenver[2];
134 
135 #define	XENVER_BOOT(m)	(xenver[XENVER_BOOT_IDX].m)
136 #define	XENVER_CURRENT(m)	(xenver[XENVER_CURRENT_IDX].m)
137 
138 /*
139  * Update the xenver data. We maintain two copies, boot and
140  * current. If we are setting the boot, then also set current.
141  */
142 static void
143 xen_set_version(xen_version_t idx)
144 {
145 	ulong_t ver;
146 
147 	bzero(&xenver[idx], sizeof (xenver[idx]));
148 
149 	ver = HYPERVISOR_xen_version(XENVER_version, 0);
150 
151 	xenver[idx].xv_major = BITX(ver, 31, 16);
152 	xenver[idx].xv_minor = BITX(ver, 15, 0);
153 
154 	(void) HYPERVISOR_xen_version(XENVER_extraversion, &xenver[idx].xv_ver);
155 
156 	/*
157 	 * The revision is buried in the extraversion information that is
158 	 * maintained by the hypervisor. For our purposes we expect that
159 	 * the revision number is:
160 	 * 	- the second character in the extraversion information
161 	 *	- one character long
162 	 *	- numeric digit
163 	 * If it isn't then we can't extract the revision and we leave it
164 	 * set to 0.
165 	 */
166 	if (strlen(xenver[idx].xv_ver) > 1 && isdigit(xenver[idx].xv_ver[1]))
167 		xenver[idx].xv_revision = xenver[idx].xv_ver[1] - '0';
168 	else
169 		cmn_err(CE_WARN, "Cannot extract revision on this hypervisor "
170 		    "version: v%s, unexpected version format",
171 		    xenver[idx].xv_ver);
172 
173 	xenver[idx].xv_is_xvm = 0;
174 
175 	if (strstr(xenver[idx].xv_ver, "-xvm") != NULL)
176 		xenver[idx].xv_is_xvm = 1;
177 
178 	(void) HYPERVISOR_xen_version(XENVER_changeset,
179 	    &xenver[idx].xv_chgset);
180 
181 	(void) HYPERVISOR_xen_version(XENVER_compile_info,
182 	    &xenver[idx].xv_build);
183 	/*
184 	 * Capabilities are a set of space separated ascii strings
185 	 * e.g. 'xen-3.1-x86_32p' or 'hvm-3.2-x86_64'
186 	 */
187 	(void) HYPERVISOR_xen_version(XENVER_capabilities,
188 	    &xenver[idx].xv_caps);
189 
190 	cmn_err(CE_CONT, "?v%lu.%lu%s chgset '%s'\n", xenver[idx].xv_major,
191 	    xenver[idx].xv_minor, xenver[idx].xv_ver, xenver[idx].xv_chgset);
192 
193 	if (idx == XENVER_BOOT_IDX)
194 		bcopy(&xenver[XENVER_BOOT_IDX], &xenver[XENVER_CURRENT_IDX],
195 		    sizeof (xenver[XENVER_BOOT_IDX]));
196 }
197 
198 typedef enum xen_hypervisor_check {
199 	XEN_RUN_CHECK,
200 	XEN_SUSPEND_CHECK
201 } xen_hypervisor_check_t;
202 
203 /*
204  * To run the hypervisor must be 3.0.4 or better. To suspend/resume
205  * we need 3.0.4 or better and if it is 3.0.4. then it must be provided
206  * by the Solaris xVM project.
207  * Checking can be disabled for testing purposes by setting the
208  * xen_suspend_debug variable.
209  */
210 static int
211 xen_hypervisor_supports_solaris(xen_hypervisor_check_t check)
212 {
213 	if (xen_suspend_debug == 1)
214 		return (1);
215 	if (XENVER_CURRENT(xv_major) < 3)
216 		return (0);
217 	if (XENVER_CURRENT(xv_major) > 3)
218 		return (1);
219 	if (XENVER_CURRENT(xv_minor) > 0)
220 		return (1);
221 	if (XENVER_CURRENT(xv_revision) < 4)
222 		return (0);
223 	if (check == XEN_SUSPEND_CHECK && XENVER_CURRENT(xv_revision) == 4 &&
224 	    !XENVER_CURRENT(xv_is_xvm))
225 		return (0);
226 
227 	return (1);
228 }
229 
230 /*
231  * If the hypervisor is -xvm, or 3.1.2 or higher, we don't need the
232  * workaround.
233  */
234 static void
235 xen_pte_workaround(void)
236 {
237 #if defined(__amd64)
238 	extern int pt_kern;
239 
240 	if (XENVER_CURRENT(xv_major) != 3)
241 		return;
242 	if (XENVER_CURRENT(xv_minor) > 1)
243 		return;
244 	if (XENVER_CURRENT(xv_minor) == 1 &&
245 	    XENVER_CURRENT(xv_revision) > 1)
246 		return;
247 	if (XENVER_CURRENT(xv_is_xvm))
248 		return;
249 
250 	pt_kern = PT_USER;
251 #endif
252 }
253 
254 void
255 xen_set_callback(void (*func)(void), uint_t type, uint_t flags)
256 {
257 	struct callback_register cb;
258 
259 	bzero(&cb, sizeof (cb));
260 #if defined(__amd64)
261 	cb.address = (ulong_t)func;
262 #elif defined(__i386)
263 	cb.address.cs = KCS_SEL;
264 	cb.address.eip = (ulong_t)func;
265 #endif
266 	cb.type = type;
267 	cb.flags = flags;
268 
269 	/*
270 	 * XXPV always ignore return value for NMI
271 	 */
272 	if (HYPERVISOR_callback_op(CALLBACKOP_register, &cb) != 0 &&
273 	    type != CALLBACKTYPE_nmi)
274 		panic("HYPERVISOR_callback_op failed");
275 }
276 
277 void
278 xen_init_callbacks(void)
279 {
280 	/*
281 	 * register event (interrupt) handler.
282 	 */
283 	xen_set_callback(xen_callback, CALLBACKTYPE_event, 0);
284 
285 	/*
286 	 * failsafe handler.
287 	 */
288 	xen_set_callback(xen_failsafe_callback, CALLBACKTYPE_failsafe,
289 	    CALLBACKF_mask_events);
290 
291 	/*
292 	 * NMI handler.
293 	 */
294 	xen_set_callback(nmiint, CALLBACKTYPE_nmi, 0);
295 
296 	/*
297 	 * system call handler
298 	 * XXPV move to init_cpu_syscall?
299 	 */
300 #if defined(__amd64)
301 	xen_set_callback(sys_syscall, CALLBACKTYPE_syscall,
302 	    CALLBACKF_mask_events);
303 #endif	/* __amd64 */
304 }
305 
306 
307 /*
308  * cmn_err() followed by a 1/4 second delay; this gives the
309  * logging service a chance to flush messages and helps avoid
310  * intermixing output from prom_printf().
311  * XXPV: doesn't exactly help us on UP though.
312  */
313 /*PRINTFLIKE2*/
314 void
315 cpr_err(int ce, const char *fmt, ...)
316 {
317 	va_list adx;
318 
319 	va_start(adx, fmt);
320 	vcmn_err(ce, fmt, adx);
321 	va_end(adx);
322 	drv_usecwait(MICROSEC >> 2);
323 }
324 
325 void
326 xen_suspend_devices(void)
327 {
328 	int rc;
329 
330 	SUSPEND_DEBUG("xen_suspend_devices\n");
331 
332 	if ((rc = cpr_suspend_devices(ddi_root_node())) != 0)
333 		panic("failed to suspend devices: %d", rc);
334 }
335 
336 void
337 xen_resume_devices(void)
338 {
339 	int rc;
340 
341 	SUSPEND_DEBUG("xen_resume_devices\n");
342 
343 	if ((rc = cpr_resume_devices(ddi_root_node(), 0)) != 0)
344 		panic("failed to resume devices: %d", rc);
345 }
346 
347 /*
348  * The list of mfn pages is out of date.  Recompute it.
349  */
350 static void
351 rebuild_mfn_list(void)
352 {
353 	int i = 0;
354 	size_t sz;
355 	size_t off;
356 	pfn_t pfn;
357 
358 	SUSPEND_DEBUG("rebuild_mfn_list\n");
359 
360 	sz = ((mfn_count * sizeof (mfn_t)) + MMU_PAGEOFFSET) & MMU_PAGEMASK;
361 
362 	for (off = 0; off < sz; off += MMU_PAGESIZE) {
363 		size_t j = mmu_btop(off);
364 		if (((j * sizeof (mfn_t)) & MMU_PAGEOFFSET) == 0) {
365 			pfn = hat_getpfnum(kas.a_hat,
366 			    (caddr_t)&mfn_list_pages[j]);
367 			mfn_list_pages_page[i++] = pfn_to_mfn(pfn);
368 		}
369 
370 		pfn = hat_getpfnum(kas.a_hat, (caddr_t)mfn_list + off);
371 		mfn_list_pages[j] = pfn_to_mfn(pfn);
372 	}
373 
374 	pfn = hat_getpfnum(kas.a_hat, (caddr_t)mfn_list_pages_page);
375 	HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list
376 	    = pfn_to_mfn(pfn);
377 }
378 
379 static void
380 suspend_cpus(void)
381 {
382 	int i;
383 
384 	SUSPEND_DEBUG("suspend_cpus\n");
385 
386 	mp_enter_barrier();
387 
388 	for (i = 1; i < ncpus; i++) {
389 		if (!CPU_IN_SET(cpu_suspend_lost_set, i)) {
390 			SUSPEND_DEBUG("xen_vcpu_down %d\n", i);
391 			(void) xen_vcpu_down(i);
392 		}
393 
394 		mach_cpucontext_reset(cpu[i]);
395 	}
396 }
397 
398 static void
399 resume_cpus(void)
400 {
401 	int i;
402 
403 	for (i = 1; i < ncpus; i++) {
404 		if (cpu[i] == NULL)
405 			continue;
406 
407 		if (!CPU_IN_SET(cpu_suspend_lost_set, i)) {
408 			SUSPEND_DEBUG("xen_vcpu_up %d\n", i);
409 			mach_cpucontext_restore(cpu[i]);
410 			(void) xen_vcpu_up(i);
411 		}
412 	}
413 
414 	mp_leave_barrier();
415 }
416 
417 /*
418  * Top level routine to direct suspend/resume of a domain.
419  */
420 void
421 xen_suspend_domain(void)
422 {
423 	extern void rtcsync(void);
424 	extern hrtime_t hres_last_tick;
425 	mfn_t start_info_mfn;
426 	ulong_t flags;
427 	pfn_t pfn;
428 	int i;
429 
430 	/*
431 	 * Check that we are happy to suspend on this hypervisor.
432 	 */
433 	if (xen_hypervisor_supports_solaris(XEN_SUSPEND_CHECK) == 0) {
434 		cpr_err(CE_WARN, "Cannot suspend on this hypervisor "
435 		    "version: v%lu.%lu%s, need at least version v3.0.4 or "
436 		    "-xvm based hypervisor", XENVER_CURRENT(xv_major),
437 		    XENVER_CURRENT(xv_minor), XENVER_CURRENT(xv_ver));
438 		return;
439 	}
440 
441 	/*
442 	 * XXPV - Are we definitely OK to suspend by the time we've connected
443 	 * the handler?
444 	 */
445 
446 	cpr_err(CE_NOTE, "Domain suspending for save/migrate");
447 
448 	SUSPEND_DEBUG("xen_suspend_domain\n");
449 
450 	/*
451 	 * suspend interrupts and devices
452 	 * XXPV - we use suspend/resume for both save/restore domains (like sun
453 	 * cpr) and for migration.  Would be nice to know the difference if
454 	 * possible.  For save/restore where down time may be a long time, we
455 	 * may want to do more of the things that cpr does.  (i.e. notify user
456 	 * processes, shrink memory footprint for faster restore, etc.)
457 	 */
458 	xen_suspend_devices();
459 	SUSPEND_DEBUG("xenbus_suspend\n");
460 	xenbus_suspend();
461 
462 	pfn = hat_getpfnum(kas.a_hat, (caddr_t)xen_info);
463 	start_info_mfn = pfn_to_mfn(pfn);
464 
465 	/*
466 	 * XXPV: cpu hotplug can hold this under a xenbus watch. Are we safe
467 	 * wrt xenbus being suspended here?
468 	 */
469 	mutex_enter(&cpu_lock);
470 
471 	/*
472 	 * Suspend must be done on vcpu 0, as no context for other CPUs is
473 	 * saved.
474 	 *
475 	 * XXPV - add to taskq API ?
476 	 */
477 	thread_affinity_set(curthread, 0);
478 	kpreempt_disable();
479 
480 	SUSPEND_DEBUG("xen_start_migrate\n");
481 	xen_start_migrate();
482 	if (ncpus > 1)
483 		suspend_cpus();
484 
485 	/*
486 	 * We can grab the ec_lock as it's a spinlock with a high SPL. Hence
487 	 * any holder would have dropped it to get through suspend_cpus().
488 	 */
489 	mutex_enter(&ec_lock);
490 
491 	/*
492 	 * From here on in, we can't take locks.
493 	 */
494 	SUSPEND_DEBUG("ec_suspend\n");
495 	ec_suspend();
496 	SUSPEND_DEBUG("gnttab_suspend\n");
497 	gnttab_suspend();
498 
499 	flags = intr_clear();
500 
501 	xpv_time_suspend();
502 
503 	/*
504 	 * Currently, the hypervisor incorrectly fails to bring back
505 	 * powered-down VCPUs.  Thus we need to record any powered-down VCPUs
506 	 * to prevent any attempts to operate on them.  But we have to do this
507 	 * *after* the very first time we do ec_suspend().
508 	 */
509 	for (i = 1; i < ncpus; i++) {
510 		if (cpu[i] == NULL)
511 			continue;
512 
513 		if (cpu_get_state(cpu[i]) == P_POWEROFF)
514 			CPUSET_ATOMIC_ADD(cpu_suspend_lost_set, i);
515 	}
516 
517 	/*
518 	 * The dom0 save/migrate code doesn't automatically translate
519 	 * these into PFNs, but expects them to be, so we do it here.
520 	 * We don't use mfn_to_pfn() because so many OS services have
521 	 * been disabled at this point.
522 	 */
523 	xen_info->store_mfn = mfn_to_pfn_mapping[xen_info->store_mfn];
524 	xen_info->console.domU.mfn =
525 	    mfn_to_pfn_mapping[xen_info->console.domU.mfn];
526 
527 	if (CPU->cpu_m.mcpu_vcpu_info->evtchn_upcall_mask == 0) {
528 		prom_printf("xen_suspend_domain(): "
529 		    "CPU->cpu_m.mcpu_vcpu_info->evtchn_upcall_mask not set\n");
530 		(void) HYPERVISOR_shutdown(SHUTDOWN_crash);
531 	}
532 
533 	if (HYPERVISOR_update_va_mapping((uintptr_t)HYPERVISOR_shared_info,
534 	    0, UVMF_INVLPG)) {
535 		prom_printf("xen_suspend_domain(): "
536 		    "HYPERVISOR_update_va_mapping() failed\n");
537 		(void) HYPERVISOR_shutdown(SHUTDOWN_crash);
538 	}
539 
540 	SUSPEND_DEBUG("HYPERVISOR_suspend\n");
541 
542 	/*
543 	 * At this point we suspend and sometime later resume.
544 	 */
545 	if (HYPERVISOR_suspend(start_info_mfn)) {
546 		prom_printf("xen_suspend_domain(): "
547 		    "HYPERVISOR_suspend() failed\n");
548 		(void) HYPERVISOR_shutdown(SHUTDOWN_crash);
549 	}
550 
551 	/*
552 	 * Point HYPERVISOR_shared_info to its new value.
553 	 */
554 	if (HYPERVISOR_update_va_mapping((uintptr_t)HYPERVISOR_shared_info,
555 	    xen_info->shared_info | PT_NOCONSIST | PT_VALID | PT_WRITABLE,
556 	    UVMF_INVLPG))
557 		(void) HYPERVISOR_shutdown(SHUTDOWN_crash);
558 
559 	if (xen_info->nr_pages != mfn_count) {
560 		prom_printf("xen_suspend_domain(): number of pages"
561 		    " changed, was 0x%lx, now 0x%lx\n", mfn_count,
562 		    xen_info->nr_pages);
563 		(void) HYPERVISOR_shutdown(SHUTDOWN_crash);
564 	}
565 
566 	xpv_time_resume();
567 
568 	cached_max_mfn = 0;
569 
570 	SUSPEND_DEBUG("gnttab_resume\n");
571 	gnttab_resume();
572 
573 	/* XXPV: add a note that this must be lockless. */
574 	SUSPEND_DEBUG("ec_resume\n");
575 	ec_resume();
576 
577 	intr_restore(flags);
578 
579 	if (ncpus > 1)
580 		resume_cpus();
581 
582 	mutex_exit(&ec_lock);
583 	xen_end_migrate();
584 	mutex_exit(&cpu_lock);
585 
586 	/*
587 	 * Now we can take locks again.
588 	 */
589 
590 	/*
591 	 * Force the tick value used for tv_nsec in hres_tick() to be up to
592 	 * date. rtcsync() will reset the hrestime value appropriately.
593 	 */
594 	hres_last_tick = xpv_gethrtime();
595 
596 	/*
597 	 * XXPV: we need to have resumed the CPUs since this takes locks, but
598 	 * can remote CPUs see bad state? Presumably yes. Should probably nest
599 	 * taking of todlock inside of cpu_lock, or vice versa, then provide an
600 	 * unlocked version.  Probably need to call clkinitf to reset cpu freq
601 	 * and re-calibrate if we migrated to a different speed cpu.  Also need
602 	 * to make a (re)init_cpu_info call to update processor info structs
603 	 * and device tree info.  That remains to be written at the moment.
604 	 */
605 	rtcsync();
606 
607 	rebuild_mfn_list();
608 
609 	SUSPEND_DEBUG("xenbus_resume\n");
610 	xenbus_resume();
611 	SUSPEND_DEBUG("xenbus_resume_devices\n");
612 	xen_resume_devices();
613 
614 	thread_affinity_clear(curthread);
615 	kpreempt_enable();
616 
617 	SUSPEND_DEBUG("finished xen_suspend_domain\n");
618 
619 	/*
620 	 * We have restarted our suspended domain, update the hypervisor
621 	 * details. NB: This must be done at the end of this function,
622 	 * since we need the domain to be completely resumed before
623 	 * these functions will work correctly.
624 	 */
625 	xen_set_version(XENVER_CURRENT_IDX);
626 
627 	/*
628 	 * We can check and report a warning, but we don't stop the
629 	 * process.
630 	 */
631 	if (xen_hypervisor_supports_solaris(XEN_SUSPEND_CHECK) == 0)
632 		cmn_err(CE_WARN, "Found hypervisor version: v%lu.%lu%s "
633 		    "but need at least version v3.0.4",
634 		    XENVER_CURRENT(xv_major), XENVER_CURRENT(xv_minor),
635 		    XENVER_CURRENT(xv_ver));
636 
637 	cmn_err(CE_NOTE, "domain restore/migrate completed");
638 }
639 
640 /*ARGSUSED*/
641 int
642 xen_debug_handler(void *arg)
643 {
644 	debug_enter("External debug event received");
645 
646 	/*
647 	 * If we've not got KMDB loaded, output some stuff difficult to capture
648 	 * from a domain core.
649 	 */
650 	if (!(boothowto & RB_DEBUG)) {
651 		shared_info_t *si = HYPERVISOR_shared_info;
652 		int i;
653 
654 		prom_printf("evtchn_pending [ ");
655 		for (i = 0; i < 8; i++)
656 			prom_printf("%lx ", si->evtchn_pending[i]);
657 		prom_printf("]\nevtchn_mask [ ");
658 		for (i = 0; i < 8; i++)
659 			prom_printf("%lx ", si->evtchn_mask[i]);
660 		prom_printf("]\n");
661 
662 		for (i = 0; i < ncpus; i++) {
663 			vcpu_info_t *vcpu = &si->vcpu_info[i];
664 			if (cpu[i] == NULL)
665 				continue;
666 			prom_printf("CPU%d pending %d mask %d sel %lx\n",
667 			    i, vcpu->evtchn_upcall_pending,
668 			    vcpu->evtchn_upcall_mask,
669 			    vcpu->evtchn_pending_sel);
670 		}
671 	}
672 
673 	return (0);
674 }
675 
676 /*ARGSUSED*/
677 static void
678 xen_sysrq_handler(struct xenbus_watch *watch, const char **vec,
679     unsigned int len)
680 {
681 	xenbus_transaction_t xbt;
682 	char key = '\0';
683 	int ret;
684 
685 retry:
686 	if (xenbus_transaction_start(&xbt)) {
687 		cmn_err(CE_WARN, "failed to start sysrq transaction");
688 		return;
689 	}
690 
691 	if ((ret = xenbus_scanf(xbt, "control", "sysrq", "%c", &key)) != 0) {
692 		/*
693 		 * ENOENT happens in response to our own xenbus_rm.
694 		 * XXPV - this happens spuriously on boot?
695 		 */
696 		if (ret != ENOENT)
697 			cmn_err(CE_WARN, "failed to read sysrq: %d", ret);
698 		goto out;
699 	}
700 
701 	if ((ret = xenbus_rm(xbt, "control", "sysrq")) != 0) {
702 		cmn_err(CE_WARN, "failed to reset sysrq: %d", ret);
703 		goto out;
704 	}
705 
706 	if (xenbus_transaction_end(xbt, 0) == EAGAIN)
707 		goto retry;
708 
709 	/*
710 	 * Somewhat arbitrary - on Linux this means 'reboot'. We could just
711 	 * accept any key, but this might increase the risk of sending a
712 	 * harmless sysrq to the wrong domain...
713 	 */
714 	if (key == 'b')
715 		(void) xen_debug_handler(NULL);
716 	else
717 		cmn_err(CE_WARN, "Ignored sysrq %c", key);
718 	return;
719 
720 out:
721 	(void) xenbus_transaction_end(xbt, 1);
722 }
723 
724 taskq_t *xen_shutdown_tq;
725 
726 #define	SHUTDOWN_INVALID	-1
727 #define	SHUTDOWN_POWEROFF	0
728 #define	SHUTDOWN_REBOOT		1
729 #define	SHUTDOWN_SUSPEND	2
730 #define	SHUTDOWN_HALT		3
731 #define	SHUTDOWN_MAX		4
732 
733 #define	SHUTDOWN_TIMEOUT_SECS (60 * 5)
734 
735 static const char *cmd_strings[SHUTDOWN_MAX] = {
736 	"poweroff",
737 	"reboot",
738 	"suspend",
739 	"halt"
740 };
741 
742 static void
743 xen_dirty_shutdown(void *arg)
744 {
745 	int cmd = (uintptr_t)arg;
746 
747 	cmn_err(CE_WARN, "Externally requested shutdown failed or "
748 	    "timed out.\nShutting down.\n");
749 
750 	switch (cmd) {
751 	case SHUTDOWN_HALT:
752 	case SHUTDOWN_POWEROFF:
753 		(void) kadmin(A_SHUTDOWN, AD_POWEROFF, NULL, kcred);
754 		break;
755 	case SHUTDOWN_REBOOT:
756 		(void) kadmin(A_REBOOT, AD_BOOT, NULL, kcred);
757 		break;
758 	}
759 }
760 
761 static void
762 xen_shutdown(void *arg)
763 {
764 	int cmd = (uintptr_t)arg;
765 	proc_t *initpp;
766 
767 	ASSERT(cmd > SHUTDOWN_INVALID && cmd < SHUTDOWN_MAX);
768 
769 	if (cmd == SHUTDOWN_SUSPEND) {
770 		xen_suspend_domain();
771 		return;
772 	}
773 
774 	switch (cmd) {
775 	case SHUTDOWN_POWEROFF:
776 		force_shutdown_method = AD_POWEROFF;
777 		break;
778 	case SHUTDOWN_HALT:
779 		force_shutdown_method = AD_HALT;
780 		break;
781 	case SHUTDOWN_REBOOT:
782 		force_shutdown_method = AD_BOOT;
783 		break;
784 	}
785 
786 	/*
787 	 * If we're still booting and init(1) isn't set up yet, simply halt.
788 	 */
789 	mutex_enter(&pidlock);
790 	initpp = prfind(P_INITPID);
791 	mutex_exit(&pidlock);
792 	if (initpp == NULL) {
793 		extern void halt(char *);
794 		halt("Power off the System");   /* just in case */
795 	}
796 
797 	/*
798 	 * else, graceful shutdown with inittab and all getting involved
799 	 */
800 	psignal(initpp, SIGPWR);
801 
802 	(void) timeout(xen_dirty_shutdown, arg,
803 	    SHUTDOWN_TIMEOUT_SECS * drv_usectohz(MICROSEC));
804 }
805 
806 /*ARGSUSED*/
807 static void
808 xen_shutdown_handler(struct xenbus_watch *watch, const char **vec,
809 	unsigned int len)
810 {
811 	char *str;
812 	xenbus_transaction_t xbt;
813 	int err, shutdown_code = SHUTDOWN_INVALID;
814 	unsigned int slen;
815 
816 again:
817 	err = xenbus_transaction_start(&xbt);
818 	if (err)
819 		return;
820 	if (xenbus_read(xbt, "control", "shutdown", (void *)&str, &slen)) {
821 		(void) xenbus_transaction_end(xbt, 1);
822 		return;
823 	}
824 
825 	SUSPEND_DEBUG("%d: xen_shutdown_handler: \"%s\"\n", CPU->cpu_id, str);
826 
827 	/*
828 	 * If this is a watch fired from our write below, check out early to
829 	 * avoid an infinite loop.
830 	 */
831 	if (strcmp(str, "") == 0) {
832 		(void) xenbus_transaction_end(xbt, 0);
833 		kmem_free(str, slen);
834 		return;
835 	} else if (strcmp(str, "poweroff") == 0) {
836 		shutdown_code = SHUTDOWN_POWEROFF;
837 	} else if (strcmp(str, "reboot") == 0) {
838 		shutdown_code = SHUTDOWN_REBOOT;
839 	} else if (strcmp(str, "suspend") == 0) {
840 		shutdown_code = SHUTDOWN_SUSPEND;
841 	} else if (strcmp(str, "halt") == 0) {
842 		shutdown_code = SHUTDOWN_HALT;
843 	} else {
844 		printf("Ignoring shutdown request: %s\n", str);
845 	}
846 
847 	/*
848 	 * XXPV	Should we check the value of xenbus_write() too, or are all
849 	 *	errors automatically folded into xenbus_transaction_end() ??
850 	 */
851 	(void) xenbus_write(xbt, "control", "shutdown", "");
852 	err = xenbus_transaction_end(xbt, 0);
853 	if (err == EAGAIN) {
854 		SUSPEND_DEBUG("%d: trying again\n", CPU->cpu_id);
855 		kmem_free(str, slen);
856 		goto again;
857 	}
858 
859 	kmem_free(str, slen);
860 	if (shutdown_code != SHUTDOWN_INVALID) {
861 		(void) taskq_dispatch(xen_shutdown_tq, xen_shutdown,
862 		    (void *)(intptr_t)shutdown_code, 0);
863 	}
864 }
865 
866 static struct xenbus_watch shutdown_watch;
867 static struct xenbus_watch sysrq_watch;
868 
869 void
870 xen_late_startup(void)
871 {
872 	if (!DOMAIN_IS_INITDOMAIN(xen_info)) {
873 		xen_shutdown_tq = taskq_create("shutdown_taskq", 1,
874 		    maxclsyspri - 1, 1, 1, TASKQ_PREPOPULATE);
875 		shutdown_watch.node = "control/shutdown";
876 		shutdown_watch.callback = xen_shutdown_handler;
877 		if (register_xenbus_watch(&shutdown_watch))
878 			cmn_err(CE_WARN, "Failed to set shutdown watcher");
879 
880 		sysrq_watch.node = "control/sysrq";
881 		sysrq_watch.callback = xen_sysrq_handler;
882 		if (register_xenbus_watch(&sysrq_watch))
883 			cmn_err(CE_WARN, "Failed to set sysrq watcher");
884 	}
885 	balloon_init(xen_info->nr_pages);
886 }
887 
888 #ifdef DEBUG
889 #define	XEN_PRINTF_BUFSIZE	1024
890 
891 char xen_printf_buffer[XEN_PRINTF_BUFSIZE];
892 
893 /*
894  * Printf function that calls hypervisor directly.  For DomU it only
895  * works when running on a xen hypervisor built with debug on.  Works
896  * always since no I/O ring interaction is needed.
897  */
898 /*PRINTFLIKE1*/
899 void
900 xen_printf(const char *fmt, ...)
901 {
902 	va_list	ap;
903 
904 	va_start(ap, fmt);
905 	(void) vsnprintf(xen_printf_buffer, XEN_PRINTF_BUFSIZE, fmt, ap);
906 	va_end(ap);
907 
908 	(void) HYPERVISOR_console_io(CONSOLEIO_write,
909 	    strlen(xen_printf_buffer), xen_printf_buffer);
910 }
911 #else
912 void
913 xen_printf(const char *fmt, ...)
914 {
915 }
916 #endif	/* DEBUG */
917 
918 void
919 startup_xen_version(void)
920 {
921 	xen_set_version(XENVER_BOOT_IDX);
922 	if (xen_hypervisor_supports_solaris(XEN_RUN_CHECK) == 0)
923 		cmn_err(CE_WARN, "Found hypervisor version: v%lu.%lu%s "
924 		    "but need at least version v3.0.4",
925 		    XENVER_CURRENT(xv_major), XENVER_CURRENT(xv_minor),
926 		    XENVER_CURRENT(xv_ver));
927 	xen_pte_workaround();
928 }
929 
930 int xen_mca_simulate_mc_physinfo_failure = 0;
931 
932 void
933 startup_xen_mca(void)
934 {
935 	if (!DOMAIN_IS_INITDOMAIN(xen_info))
936 		return;
937 
938 	xen_phys_ncpus = 0;
939 	xen_phys_cpus = NULL;
940 
941 	if (xen_mca_simulate_mc_physinfo_failure ||
942 	    xen_get_mc_physcpuinfo(NULL, &xen_phys_ncpus) != 0) {
943 		cmn_err(CE_WARN,
944 		    "%sxen_get_mc_physinfo failure during xen MCA startup: "
945 		    "there will be no machine check support",
946 		    xen_mca_simulate_mc_physinfo_failure ? "(simulated) " : "");
947 		return;
948 	}
949 
950 	xen_phys_cpus = kmem_alloc(xen_phys_ncpus *
951 	    sizeof (xen_mc_logical_cpu_t), KM_NOSLEEP);
952 
953 	if (xen_phys_cpus == NULL) {
954 		cmn_err(CE_WARN,
955 		    "xen_get_mc_physinfo failure: can't allocate CPU array");
956 		return;
957 	}
958 
959 	if (xen_get_mc_physcpuinfo(xen_phys_cpus, &xen_phys_ncpus) != 0) {
960 		cmn_err(CE_WARN, "xen_get_mc_physinfo failure: no "
961 		    "physical CPU info");
962 		kmem_free(xen_phys_cpus,
963 		    xen_phys_ncpus * sizeof (xen_mc_logical_cpu_t));
964 		xen_phys_ncpus = 0;
965 		xen_phys_cpus = NULL;
966 	}
967 
968 	if (xen_physinfo_debug) {
969 		xen_mc_logical_cpu_t *xcp;
970 		unsigned i;
971 
972 		cmn_err(CE_NOTE, "xvm mca: %u physical cpus:\n",
973 		    xen_phys_ncpus);
974 		for (i = 0; i < xen_phys_ncpus; i++) {
975 			xcp = &xen_phys_cpus[i];
976 			cmn_err(CE_NOTE, "cpu%u: (%u, %u, %u) apid %u",
977 			    xcp->mc_cpunr, xcp->mc_chipid, xcp->mc_coreid,
978 			    xcp->mc_threadid, xcp->mc_apicid);
979 		}
980 	}
981 }
982 
983 /*
984  * Miscellaneous hypercall wrappers with slightly more verbose diagnostics.
985  */
986 
987 void
988 xen_set_gdt(ulong_t *frame_list, int entries)
989 {
990 	int err;
991 	if ((err = HYPERVISOR_set_gdt(frame_list, entries)) != 0) {
992 		/*
993 		 * X_EINVAL:	reserved entry or bad frames
994 		 * X_EFAULT:	bad address
995 		 */
996 		panic("xen_set_gdt(%p, %d): error %d",
997 		    (void *)frame_list, entries, -(int)err);
998 	}
999 }
1000 
1001 void
1002 xen_set_ldt(user_desc_t *ldt, uint_t nsels)
1003 {
1004 	struct mmuext_op	op;
1005 	long			err;
1006 
1007 	op.cmd = MMUEXT_SET_LDT;
1008 	op.arg1.linear_addr = (uintptr_t)ldt;
1009 	op.arg2.nr_ents = nsels;
1010 
1011 	if ((err = HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF)) != 0) {
1012 		panic("xen_set_ldt(%p, %d): error %d",
1013 		    (void *)ldt, nsels, -(int)err);
1014 	}
1015 }
1016 
1017 void
1018 xen_stack_switch(ulong_t ss, ulong_t esp)
1019 {
1020 	long err;
1021 
1022 	if ((err = HYPERVISOR_stack_switch(ss, esp)) != 0) {
1023 		/*
1024 		 * X_EPERM:	bad selector
1025 		 */
1026 		panic("xen_stack_switch(%lx, %lx): error %d", ss, esp,
1027 		    -(int)err);
1028 	}
1029 }
1030 
1031 long
1032 xen_set_trap_table(trap_info_t *table)
1033 {
1034 	long err;
1035 
1036 	if ((err = HYPERVISOR_set_trap_table(table)) != 0) {
1037 		/*
1038 		 * X_EFAULT:	bad address
1039 		 * X_EPERM:	bad selector
1040 		 */
1041 		panic("xen_set_trap_table(%p): error %d", (void *)table,
1042 		    -(int)err);
1043 	}
1044 	return (err);
1045 }
1046 
1047 #if defined(__amd64)
1048 void
1049 xen_set_segment_base(int reg, ulong_t value)
1050 {
1051 	long err;
1052 
1053 	if ((err = HYPERVISOR_set_segment_base(reg, value)) != 0) {
1054 		/*
1055 		 * X_EFAULT:	bad address
1056 		 * X_EINVAL:	bad type
1057 		 */
1058 		panic("xen_set_segment_base(%d, %lx): error %d",
1059 		    reg, value, -(int)err);
1060 	}
1061 }
1062 #endif	/* __amd64 */
1063 
1064 /*
1065  * Translate a hypervisor errcode to a Solaris error code.
1066  */
1067 int
1068 xen_xlate_errcode(int error)
1069 {
1070 	switch (-error) {
1071 
1072 	/*
1073 	 * Translate hypervisor errno's into native errno's
1074 	 */
1075 
1076 #define	CASE(num)	case X_##num: error = num; break
1077 
1078 	CASE(EPERM);	CASE(ENOENT);	CASE(ESRCH);
1079 	CASE(EINTR);	CASE(EIO);	CASE(ENXIO);
1080 	CASE(E2BIG);	CASE(ENOMEM);	CASE(EACCES);
1081 	CASE(EFAULT);	CASE(EBUSY);	CASE(EEXIST);
1082 	CASE(ENODEV);	CASE(EISDIR);	CASE(EINVAL);
1083 	CASE(ENOSPC);	CASE(ESPIPE);	CASE(EROFS);
1084 	CASE(ENOSYS);	CASE(ENOTEMPTY); CASE(EISCONN);
1085 	CASE(ENODATA);	CASE(EAGAIN);
1086 
1087 #undef CASE
1088 
1089 	default:
1090 		panic("xen_xlate_errcode: unknown error %d", error);
1091 	}
1092 
1093 	return (error);
1094 }
1095 
1096 /*
1097  * Raise PS_IOPL on current vcpu to user level.
1098  * Caller responsible for preventing kernel preemption.
1099  */
1100 void
1101 xen_enable_user_iopl(void)
1102 {
1103 	physdev_set_iopl_t set_iopl;
1104 	set_iopl.iopl = 3;		/* user ring 3 */
1105 	(void) HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1106 }
1107 
1108 /*
1109  * Drop PS_IOPL on current vcpu to kernel level
1110  */
1111 void
1112 xen_disable_user_iopl(void)
1113 {
1114 	physdev_set_iopl_t set_iopl;
1115 	set_iopl.iopl = 1;		/* kernel pseudo ring 1 */
1116 	(void) HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1117 }
1118 
1119 int
1120 xen_gdt_setprot(cpu_t *cp, uint_t prot)
1121 {
1122 	int err;
1123 #if defined(__amd64)
1124 	int pt_bits = PT_VALID;
1125 	if (prot & PROT_WRITE)
1126 		pt_bits |= PT_WRITABLE;
1127 #endif
1128 
1129 	if ((err = as_setprot(&kas, (caddr_t)cp->cpu_gdt,
1130 	    MMU_PAGESIZE, prot)) != 0)
1131 		goto done;
1132 
1133 #if defined(__amd64)
1134 	err = xen_kpm_page(mmu_btop(cp->cpu_m.mcpu_gdtpa), pt_bits);
1135 #endif
1136 
1137 done:
1138 	if (err) {
1139 		cmn_err(CE_WARN, "cpu%d: xen_gdt_setprot(%s) failed: error %d",
1140 		    cp->cpu_id, (prot & PROT_WRITE) ? "writable" : "read-only",
1141 		    err);
1142 	}
1143 
1144 	return (err);
1145 }
1146 
1147 int
1148 xen_ldt_setprot(user_desc_t *ldt, size_t lsize, uint_t prot)
1149 {
1150 	int err;
1151 	caddr_t	lva = (caddr_t)ldt;
1152 #if defined(__amd64)
1153 	int pt_bits = PT_VALID;
1154 	pgcnt_t npgs;
1155 	if (prot & PROT_WRITE)
1156 		pt_bits |= PT_WRITABLE;
1157 #endif	/* __amd64 */
1158 
1159 	if ((err = as_setprot(&kas, (caddr_t)ldt, lsize, prot)) != 0)
1160 		goto done;
1161 
1162 #if defined(__amd64)
1163 
1164 	ASSERT(IS_P2ALIGNED(lsize, PAGESIZE));
1165 	npgs = mmu_btop(lsize);
1166 	while (npgs--) {
1167 		if ((err = xen_kpm_page(hat_getpfnum(kas.a_hat, lva),
1168 		    pt_bits)) != 0)
1169 			break;
1170 		lva += PAGESIZE;
1171 	}
1172 #endif	/* __amd64 */
1173 
1174 done:
1175 	if (err) {
1176 		cmn_err(CE_WARN, "xen_ldt_setprot(%p, %s) failed: error %d",
1177 		    (void *)lva,
1178 		    (prot & PROT_WRITE) ? "writable" : "read-only", err);
1179 	}
1180 
1181 	return (err);
1182 }
1183 
1184 int
1185 xen_get_mc_physcpuinfo(xen_mc_logical_cpu_t *log_cpus, uint_t *ncpus)
1186 {
1187 	xen_mc_t xmc;
1188 	struct xen_mc_physcpuinfo *cpi = &xmc.u.mc_physcpuinfo;
1189 
1190 	cpi->ncpus = *ncpus;
1191 	/*LINTED: constant in conditional context*/
1192 	set_xen_guest_handle(cpi->info, log_cpus);
1193 
1194 	if (HYPERVISOR_mca(XEN_MC_physcpuinfo, &xmc) != 0)
1195 		return (-1);
1196 
1197 	*ncpus = cpi->ncpus;
1198 	return (0);
1199 }
1200 
1201 void
1202 print_panic(const char *str)
1203 {
1204 	xen_printf(str);
1205 }
1206 
1207 /*
1208  * Interfaces to iterate over real cpu information, but only that info
1209  * which we choose to expose here.  These are of interest to dom0
1210  * only (and the backing hypercall should not work for domu).
1211  */
1212 
1213 xen_mc_lcpu_cookie_t
1214 xen_physcpu_next(xen_mc_lcpu_cookie_t cookie)
1215 {
1216 	xen_mc_logical_cpu_t *xcp = (xen_mc_logical_cpu_t *)cookie;
1217 
1218 	if (!DOMAIN_IS_INITDOMAIN(xen_info))
1219 		return (NULL);
1220 
1221 	if (cookie == NULL)
1222 		return ((xen_mc_lcpu_cookie_t)xen_phys_cpus);
1223 
1224 	if (xcp == xen_phys_cpus + xen_phys_ncpus - 1)
1225 		return (NULL);
1226 	else
1227 		return ((xen_mc_lcpu_cookie_t)++xcp);
1228 }
1229 
1230 #define	COOKIE2XCP(c) ((xen_mc_logical_cpu_t *)(c))
1231 
1232 const char *
1233 xen_physcpu_vendorstr(xen_mc_lcpu_cookie_t cookie)
1234 {
1235 	xen_mc_logical_cpu_t *xcp = COOKIE2XCP(cookie);
1236 
1237 	return ((const char *)&xcp->mc_vendorid[0]);
1238 }
1239 
1240 int
1241 xen_physcpu_family(xen_mc_lcpu_cookie_t cookie)
1242 {
1243 	return (COOKIE2XCP(cookie)->mc_family);
1244 }
1245 
1246 int
1247 xen_physcpu_model(xen_mc_lcpu_cookie_t cookie)
1248 {
1249 	return (COOKIE2XCP(cookie)->mc_model);
1250 }
1251 
1252 int
1253 xen_physcpu_stepping(xen_mc_lcpu_cookie_t cookie)
1254 {
1255 	return (COOKIE2XCP(cookie)->mc_step);
1256 }
1257 
1258 id_t
1259 xen_physcpu_chipid(xen_mc_lcpu_cookie_t cookie)
1260 {
1261 	return (COOKIE2XCP(cookie)->mc_chipid);
1262 }
1263 
1264 id_t
1265 xen_physcpu_coreid(xen_mc_lcpu_cookie_t cookie)
1266 {
1267 	return (COOKIE2XCP(cookie)->mc_coreid);
1268 }
1269 
1270 id_t
1271 xen_physcpu_strandid(xen_mc_lcpu_cookie_t cookie)
1272 {
1273 	return (COOKIE2XCP(cookie)->mc_threadid);
1274 }
1275 
1276 id_t
1277 xen_physcpu_initial_apicid(xen_mc_lcpu_cookie_t cookie)
1278 {
1279 	return (COOKIE2XCP(cookie)->mc_clusterid);
1280 }
1281 
1282 id_t
1283 xen_physcpu_logical_id(xen_mc_lcpu_cookie_t cookie)
1284 {
1285 	return (COOKIE2XCP(cookie)->mc_cpunr);
1286 }
1287 
1288 boolean_t
1289 xen_physcpu_is_cmt(xen_mc_lcpu_cookie_t cookie)
1290 {
1291 	return (COOKIE2XCP(cookie)->mc_nthreads > 1);
1292 }
1293 
1294 uint64_t
1295 xen_physcpu_mcg_cap(xen_mc_lcpu_cookie_t cookie)
1296 {
1297 	xen_mc_logical_cpu_t *xcp = COOKIE2XCP(cookie);
1298 
1299 	/*
1300 	 * Need to #define the indices, or search through the array.
1301 	 */
1302 	return (xcp->mc_msrvalues[0].value);
1303 }
1304 
1305 int
1306 xen_map_gref(uint_t cmd, gnttab_map_grant_ref_t *mapop, uint_t count,
1307     boolean_t uvaddr)
1308 {
1309 	long rc;
1310 	uint_t i;
1311 
1312 	ASSERT(cmd == GNTTABOP_map_grant_ref);
1313 
1314 #if !defined(_BOOT)
1315 	if (uvaddr == B_FALSE) {
1316 		for (i = 0; i < count; ++i) {
1317 			mapop[i].flags |= (PT_FOREIGN <<_GNTMAP_guest_avail0);
1318 		}
1319 	}
1320 #endif
1321 
1322 	rc = HYPERVISOR_grant_table_op(cmd, mapop, count);
1323 
1324 	return (rc);
1325 }
1326 
1327 static int
1328 xpv_get_physinfo(xen_sysctl_physinfo_t *pi)
1329 {
1330 	xen_sysctl_t op;
1331 	struct sp { void *p; } *sp = (struct sp *)&op.u.physinfo.cpu_to_node;
1332 	int ret;
1333 
1334 	bzero(&op, sizeof (op));
1335 	op.cmd = XEN_SYSCTL_physinfo;
1336 	op.interface_version = XEN_SYSCTL_INTERFACE_VERSION;
1337 	/*LINTED: constant in conditional context*/
1338 	set_xen_guest_handle(*sp, NULL);
1339 
1340 	ret = HYPERVISOR_sysctl(&op);
1341 
1342 	if (ret != 0)
1343 		return (xen_xlate_errcode(ret));
1344 
1345 	bcopy(&op.u.physinfo, pi, sizeof (op.u.physinfo));
1346 	return (0);
1347 }
1348 
1349 /*
1350  * On dom0, we can determine the number of physical cpus on the machine.
1351  * This number is important when figuring out what workarounds are
1352  * appropriate, so compute it now.
1353  */
1354 uint_t
1355 xpv_nr_phys_cpus(void)
1356 {
1357 	static uint_t nphyscpus = 0;
1358 
1359 	ASSERT(DOMAIN_IS_INITDOMAIN(xen_info));
1360 
1361 	if (nphyscpus == 0) {
1362 		xen_sysctl_physinfo_t pi;
1363 		int ret;
1364 
1365 		if ((ret = xpv_get_physinfo(&pi)) != 0)
1366 			panic("xpv_get_physinfo() failed: %d\n", ret);
1367 		nphyscpus = pi.nr_cpus;
1368 	}
1369 	return (nphyscpus);
1370 }
1371 
1372 pgcnt_t
1373 xpv_nr_phys_pages(void)
1374 {
1375 	xen_sysctl_physinfo_t pi;
1376 	int ret;
1377 
1378 	ASSERT(DOMAIN_IS_INITDOMAIN(xen_info));
1379 
1380 	if ((ret = xpv_get_physinfo(&pi)) != 0)
1381 		panic("xpv_get_physinfo() failed: %d\n", ret);
1382 
1383 	return ((pgcnt_t)pi.total_pages);
1384 }
1385 
1386 uint64_t
1387 xpv_cpu_khz(void)
1388 {
1389 	xen_sysctl_physinfo_t pi;
1390 	int ret;
1391 
1392 	ASSERT(DOMAIN_IS_INITDOMAIN(xen_info));
1393 
1394 	if ((ret = xpv_get_physinfo(&pi)) != 0)
1395 		panic("xpv_get_physinfo() failed: %d\n", ret);
1396 	return ((uint64_t)pi.cpu_khz);
1397 }
1398