xref: /illumos-gate/usr/src/uts/i86xpv/os/xen_machdep.c (revision 5a469116729183a46e77dc0620955bbde58d93f7)
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
xen_set_version(xen_version_t idx)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
xen_hypervisor_supports_solaris(xen_hypervisor_check_t check)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
xen_pte_workaround(void)235 xen_pte_workaround(void)
236 {
237 	extern int pt_kern;
238 
239 	if (XENVER_CURRENT(xv_major) != 3)
240 		return;
241 	if (XENVER_CURRENT(xv_minor) > 1)
242 		return;
243 	if (XENVER_CURRENT(xv_minor) == 1 &&
244 	    XENVER_CURRENT(xv_revision) > 1)
245 		return;
246 	if (XENVER_CURRENT(xv_is_xvm))
247 		return;
248 
249 	pt_kern = PT_USER;
250 }
251 
252 void
xen_set_callback(void (* func)(void),uint_t type,uint_t flags)253 xen_set_callback(void (*func)(void), uint_t type, uint_t flags)
254 {
255 	struct callback_register cb;
256 
257 	bzero(&cb, sizeof (cb));
258 	cb.address = (ulong_t)func;
259 	cb.type = type;
260 	cb.flags = flags;
261 
262 	/*
263 	 * XXPV always ignore return value for NMI
264 	 */
265 	if (HYPERVISOR_callback_op(CALLBACKOP_register, &cb) != 0 &&
266 	    type != CALLBACKTYPE_nmi)
267 		panic("HYPERVISOR_callback_op failed");
268 }
269 
270 void
xen_init_callbacks(void)271 xen_init_callbacks(void)
272 {
273 	/*
274 	 * register event (interrupt) handler.
275 	 */
276 	xen_set_callback(xen_callback, CALLBACKTYPE_event, 0);
277 
278 	/*
279 	 * failsafe handler.
280 	 */
281 	xen_set_callback(xen_failsafe_callback, CALLBACKTYPE_failsafe,
282 	    CALLBACKF_mask_events);
283 
284 	/*
285 	 * NMI handler.
286 	 */
287 	xen_set_callback(nmiint, CALLBACKTYPE_nmi, 0);
288 
289 	/*
290 	 * system call handler
291 	 * XXPV move to init_cpu_syscall?
292 	 */
293 	xen_set_callback(sys_syscall, CALLBACKTYPE_syscall,
294 	    CALLBACKF_mask_events);
295 }
296 
297 
298 /*
299  * cmn_err() followed by a 1/4 second delay; this gives the
300  * logging service a chance to flush messages and helps avoid
301  * intermixing output from prom_printf().
302  * XXPV: doesn't exactly help us on UP though.
303  */
304 /*PRINTFLIKE2*/
305 void
cpr_err(int ce,const char * fmt,...)306 cpr_err(int ce, const char *fmt, ...)
307 {
308 	va_list adx;
309 
310 	va_start(adx, fmt);
311 	vcmn_err(ce, fmt, adx);
312 	va_end(adx);
313 	drv_usecwait(MICROSEC >> 2);
314 }
315 
316 void
xen_suspend_devices(void)317 xen_suspend_devices(void)
318 {
319 	int rc;
320 
321 	SUSPEND_DEBUG("xen_suspend_devices\n");
322 
323 	if ((rc = cpr_suspend_devices(ddi_root_node())) != 0)
324 		panic("failed to suspend devices: %d", rc);
325 }
326 
327 void
xen_resume_devices(void)328 xen_resume_devices(void)
329 {
330 	int rc;
331 
332 	SUSPEND_DEBUG("xen_resume_devices\n");
333 
334 	if ((rc = cpr_resume_devices(ddi_root_node(), 0)) != 0)
335 		panic("failed to resume devices: %d", rc);
336 }
337 
338 /*
339  * The list of mfn pages is out of date.  Recompute it.
340  */
341 static void
rebuild_mfn_list(void)342 rebuild_mfn_list(void)
343 {
344 	int i = 0;
345 	size_t sz;
346 	size_t off;
347 	pfn_t pfn;
348 
349 	SUSPEND_DEBUG("rebuild_mfn_list\n");
350 
351 	sz = ((mfn_count * sizeof (mfn_t)) + MMU_PAGEOFFSET) & MMU_PAGEMASK;
352 
353 	for (off = 0; off < sz; off += MMU_PAGESIZE) {
354 		size_t j = mmu_btop(off);
355 		if (((j * sizeof (mfn_t)) & MMU_PAGEOFFSET) == 0) {
356 			pfn = hat_getpfnum(kas.a_hat,
357 			    (caddr_t)&mfn_list_pages[j]);
358 			mfn_list_pages_page[i++] = pfn_to_mfn(pfn);
359 		}
360 
361 		pfn = hat_getpfnum(kas.a_hat, (caddr_t)mfn_list + off);
362 		mfn_list_pages[j] = pfn_to_mfn(pfn);
363 	}
364 
365 	pfn = hat_getpfnum(kas.a_hat, (caddr_t)mfn_list_pages_page);
366 	HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list
367 	    = pfn_to_mfn(pfn);
368 }
369 
370 static void
suspend_cpus(void)371 suspend_cpus(void)
372 {
373 	int i;
374 
375 	SUSPEND_DEBUG("suspend_cpus\n");
376 
377 	mp_enter_barrier();
378 
379 	for (i = 1; i < ncpus; i++) {
380 		if (!CPU_IN_SET(cpu_suspend_lost_set, i)) {
381 			SUSPEND_DEBUG("xen_vcpu_down %d\n", i);
382 			(void) xen_vcpu_down(i);
383 		}
384 
385 		mach_cpucontext_reset(cpu[i]);
386 	}
387 }
388 
389 static void
resume_cpus(void)390 resume_cpus(void)
391 {
392 	int i;
393 
394 	for (i = 1; i < ncpus; i++) {
395 		if (cpu[i] == NULL)
396 			continue;
397 
398 		if (!CPU_IN_SET(cpu_suspend_lost_set, i)) {
399 			SUSPEND_DEBUG("xen_vcpu_up %d\n", i);
400 			mach_cpucontext_restore(cpu[i]);
401 			(void) xen_vcpu_up(i);
402 		}
403 	}
404 
405 	mp_leave_barrier();
406 }
407 
408 /*
409  * Top level routine to direct suspend/resume of a domain.
410  */
411 void
xen_suspend_domain(void)412 xen_suspend_domain(void)
413 {
414 	extern void rtcsync(void);
415 	extern hrtime_t hres_last_tick;
416 	mfn_t start_info_mfn;
417 	ulong_t flags;
418 	pfn_t pfn;
419 	int i;
420 
421 	/*
422 	 * Check that we are happy to suspend on this hypervisor.
423 	 */
424 	if (xen_hypervisor_supports_solaris(XEN_SUSPEND_CHECK) == 0) {
425 		cpr_err(CE_WARN, "Cannot suspend on this hypervisor "
426 		    "version: v%lu.%lu%s, need at least version v3.0.4 or "
427 		    "-xvm based hypervisor", XENVER_CURRENT(xv_major),
428 		    XENVER_CURRENT(xv_minor), XENVER_CURRENT(xv_ver));
429 		return;
430 	}
431 
432 	/*
433 	 * XXPV - Are we definitely OK to suspend by the time we've connected
434 	 * the handler?
435 	 */
436 
437 	cpr_err(CE_NOTE, "Domain suspending for save/migrate");
438 
439 	SUSPEND_DEBUG("xen_suspend_domain\n");
440 
441 	/*
442 	 * suspend interrupts and devices
443 	 * XXPV - we use suspend/resume for both save/restore domains (like sun
444 	 * cpr) and for migration.  Would be nice to know the difference if
445 	 * possible.  For save/restore where down time may be a long time, we
446 	 * may want to do more of the things that cpr does.  (i.e. notify user
447 	 * processes, shrink memory footprint for faster restore, etc.)
448 	 */
449 	xen_suspend_devices();
450 	SUSPEND_DEBUG("xenbus_suspend\n");
451 	xenbus_suspend();
452 
453 	pfn = hat_getpfnum(kas.a_hat, (caddr_t)xen_info);
454 	start_info_mfn = pfn_to_mfn(pfn);
455 
456 	/*
457 	 * XXPV: cpu hotplug can hold this under a xenbus watch. Are we safe
458 	 * wrt xenbus being suspended here?
459 	 */
460 	mutex_enter(&cpu_lock);
461 
462 	/*
463 	 * Suspend must be done on vcpu 0, as no context for other CPUs is
464 	 * saved.
465 	 *
466 	 * XXPV - add to taskq API ?
467 	 */
468 	thread_affinity_set(curthread, 0);
469 	kpreempt_disable();
470 
471 	SUSPEND_DEBUG("xen_start_migrate\n");
472 	xen_start_migrate();
473 	if (ncpus > 1)
474 		suspend_cpus();
475 
476 	/*
477 	 * We can grab the ec_lock as it's a spinlock with a high SPL. Hence
478 	 * any holder would have dropped it to get through suspend_cpus().
479 	 */
480 	mutex_enter(&ec_lock);
481 
482 	/*
483 	 * From here on in, we can't take locks.
484 	 */
485 	SUSPEND_DEBUG("ec_suspend\n");
486 	ec_suspend();
487 	SUSPEND_DEBUG("gnttab_suspend\n");
488 	gnttab_suspend();
489 
490 	flags = intr_clear();
491 
492 	xpv_time_suspend();
493 
494 	/*
495 	 * Currently, the hypervisor incorrectly fails to bring back
496 	 * powered-down VCPUs.  Thus we need to record any powered-down VCPUs
497 	 * to prevent any attempts to operate on them.  But we have to do this
498 	 * *after* the very first time we do ec_suspend().
499 	 */
500 	for (i = 1; i < ncpus; i++) {
501 		if (cpu[i] == NULL)
502 			continue;
503 
504 		if (cpu_get_state(cpu[i]) == P_POWEROFF)
505 			CPUSET_ATOMIC_ADD(cpu_suspend_lost_set, i);
506 	}
507 
508 	/*
509 	 * The dom0 save/migrate code doesn't automatically translate
510 	 * these into PFNs, but expects them to be, so we do it here.
511 	 * We don't use mfn_to_pfn() because so many OS services have
512 	 * been disabled at this point.
513 	 */
514 	xen_info->store_mfn = mfn_to_pfn_mapping[xen_info->store_mfn];
515 	xen_info->console.domU.mfn =
516 	    mfn_to_pfn_mapping[xen_info->console.domU.mfn];
517 
518 	if (CPU->cpu_m.mcpu_vcpu_info->evtchn_upcall_mask == 0) {
519 		prom_printf("xen_suspend_domain(): "
520 		    "CPU->cpu_m.mcpu_vcpu_info->evtchn_upcall_mask not set\n");
521 		(void) HYPERVISOR_shutdown(SHUTDOWN_crash);
522 	}
523 
524 	if (HYPERVISOR_update_va_mapping((uintptr_t)HYPERVISOR_shared_info,
525 	    0, UVMF_INVLPG)) {
526 		prom_printf("xen_suspend_domain(): "
527 		    "HYPERVISOR_update_va_mapping() failed\n");
528 		(void) HYPERVISOR_shutdown(SHUTDOWN_crash);
529 	}
530 
531 	SUSPEND_DEBUG("HYPERVISOR_suspend\n");
532 
533 	/*
534 	 * At this point we suspend and sometime later resume.
535 	 */
536 	if (HYPERVISOR_suspend(start_info_mfn)) {
537 		prom_printf("xen_suspend_domain(): "
538 		    "HYPERVISOR_suspend() failed\n");
539 		(void) HYPERVISOR_shutdown(SHUTDOWN_crash);
540 	}
541 
542 	/*
543 	 * Point HYPERVISOR_shared_info to its new value.
544 	 */
545 	if (HYPERVISOR_update_va_mapping((uintptr_t)HYPERVISOR_shared_info,
546 	    xen_info->shared_info | PT_NOCONSIST | PT_VALID | PT_WRITABLE,
547 	    UVMF_INVLPG))
548 		(void) HYPERVISOR_shutdown(SHUTDOWN_crash);
549 
550 	if (xen_info->nr_pages != mfn_count) {
551 		prom_printf("xen_suspend_domain(): number of pages"
552 		    " changed, was 0x%lx, now 0x%lx\n", mfn_count,
553 		    xen_info->nr_pages);
554 		(void) HYPERVISOR_shutdown(SHUTDOWN_crash);
555 	}
556 
557 	xpv_time_resume();
558 
559 	cached_max_mfn = 0;
560 
561 	SUSPEND_DEBUG("gnttab_resume\n");
562 	gnttab_resume();
563 
564 	/* XXPV: add a note that this must be lockless. */
565 	SUSPEND_DEBUG("ec_resume\n");
566 	ec_resume();
567 
568 	intr_restore(flags);
569 
570 	if (ncpus > 1)
571 		resume_cpus();
572 
573 	mutex_exit(&ec_lock);
574 	xen_end_migrate();
575 	mutex_exit(&cpu_lock);
576 
577 	/*
578 	 * Now we can take locks again.
579 	 */
580 
581 	/*
582 	 * Force the tick value used for tv_nsec in hres_tick() to be up to
583 	 * date. rtcsync() will reset the hrestime value appropriately.
584 	 */
585 	hres_last_tick = xpv_gethrtime();
586 
587 	/*
588 	 * XXPV: we need to have resumed the CPUs since this takes locks, but
589 	 * can remote CPUs see bad state? Presumably yes. Should probably nest
590 	 * taking of todlock inside of cpu_lock, or vice versa, then provide an
591 	 * unlocked version.  Probably need to call clkinitf to reset cpu freq
592 	 * and re-calibrate if we migrated to a different speed cpu.  Also need
593 	 * to make a (re)init_cpu_info call to update processor info structs
594 	 * and device tree info.  That remains to be written at the moment.
595 	 */
596 	rtcsync();
597 
598 	rebuild_mfn_list();
599 
600 	SUSPEND_DEBUG("xenbus_resume\n");
601 	xenbus_resume();
602 	SUSPEND_DEBUG("xenbus_resume_devices\n");
603 	xen_resume_devices();
604 
605 	thread_affinity_clear(curthread);
606 	kpreempt_enable();
607 
608 	SUSPEND_DEBUG("finished xen_suspend_domain\n");
609 
610 	/*
611 	 * We have restarted our suspended domain, update the hypervisor
612 	 * details. NB: This must be done at the end of this function,
613 	 * since we need the domain to be completely resumed before
614 	 * these functions will work correctly.
615 	 */
616 	xen_set_version(XENVER_CURRENT_IDX);
617 
618 	/*
619 	 * We can check and report a warning, but we don't stop the
620 	 * process.
621 	 */
622 	if (xen_hypervisor_supports_solaris(XEN_SUSPEND_CHECK) == 0)
623 		cmn_err(CE_WARN, "Found hypervisor version: v%lu.%lu%s "
624 		    "but need at least version v3.0.4",
625 		    XENVER_CURRENT(xv_major), XENVER_CURRENT(xv_minor),
626 		    XENVER_CURRENT(xv_ver));
627 
628 	cmn_err(CE_NOTE, "domain restore/migrate completed");
629 }
630 
631 uint_t
xen_debug_handler(caddr_t arg __unused,caddr_t arg1 __unused)632 xen_debug_handler(caddr_t arg __unused, caddr_t arg1 __unused)
633 {
634 	debug_enter("External debug event received");
635 
636 	/*
637 	 * If we've not got KMDB loaded, output some stuff difficult to capture
638 	 * from a domain core.
639 	 */
640 	if (!(boothowto & RB_DEBUG)) {
641 		shared_info_t *si = HYPERVISOR_shared_info;
642 		int i;
643 
644 		prom_printf("evtchn_pending [ ");
645 		for (i = 0; i < 8; i++)
646 			prom_printf("%lx ", si->evtchn_pending[i]);
647 		prom_printf("]\nevtchn_mask [ ");
648 		for (i = 0; i < 8; i++)
649 			prom_printf("%lx ", si->evtchn_mask[i]);
650 		prom_printf("]\n");
651 
652 		for (i = 0; i < ncpus; i++) {
653 			vcpu_info_t *vcpu = &si->vcpu_info[i];
654 			if (cpu[i] == NULL)
655 				continue;
656 			prom_printf("CPU%d pending %d mask %d sel %lx\n",
657 			    i, vcpu->evtchn_upcall_pending,
658 			    vcpu->evtchn_upcall_mask,
659 			    vcpu->evtchn_pending_sel);
660 		}
661 	}
662 
663 	return (0);
664 }
665 
666 /*ARGSUSED*/
667 static void
xen_sysrq_handler(struct xenbus_watch * watch,const char ** vec,unsigned int len)668 xen_sysrq_handler(struct xenbus_watch *watch, const char **vec,
669     unsigned int len)
670 {
671 	xenbus_transaction_t xbt;
672 	char key = '\0';
673 	int ret;
674 
675 retry:
676 	if (xenbus_transaction_start(&xbt)) {
677 		cmn_err(CE_WARN, "failed to start sysrq transaction");
678 		return;
679 	}
680 
681 	if ((ret = xenbus_scanf(xbt, "control", "sysrq", "%c", &key)) != 0) {
682 		/*
683 		 * ENOENT happens in response to our own xenbus_rm.
684 		 * XXPV - this happens spuriously on boot?
685 		 */
686 		if (ret != ENOENT)
687 			cmn_err(CE_WARN, "failed to read sysrq: %d", ret);
688 		goto out;
689 	}
690 
691 	if ((ret = xenbus_rm(xbt, "control", "sysrq")) != 0) {
692 		cmn_err(CE_WARN, "failed to reset sysrq: %d", ret);
693 		goto out;
694 	}
695 
696 	if (xenbus_transaction_end(xbt, 0) == EAGAIN)
697 		goto retry;
698 
699 	/*
700 	 * Somewhat arbitrary - on Linux this means 'reboot'. We could just
701 	 * accept any key, but this might increase the risk of sending a
702 	 * harmless sysrq to the wrong domain...
703 	 */
704 	if (key == 'b')
705 		(void) xen_debug_handler(NULL, NULL);
706 	else
707 		cmn_err(CE_WARN, "Ignored sysrq %c", key);
708 	return;
709 
710 out:
711 	(void) xenbus_transaction_end(xbt, 1);
712 }
713 
714 taskq_t *xen_shutdown_tq;
715 
716 #define	SHUTDOWN_INVALID	-1
717 #define	SHUTDOWN_POWEROFF	0
718 #define	SHUTDOWN_REBOOT		1
719 #define	SHUTDOWN_SUSPEND	2
720 #define	SHUTDOWN_HALT		3
721 #define	SHUTDOWN_MAX		4
722 
723 #define	SHUTDOWN_TIMEOUT_SECS (60 * 5)
724 
725 static const char *cmd_strings[SHUTDOWN_MAX] = {
726 	"poweroff",
727 	"reboot",
728 	"suspend",
729 	"halt"
730 };
731 
732 static void
xen_dirty_shutdown(void * arg)733 xen_dirty_shutdown(void *arg)
734 {
735 	int cmd = (uintptr_t)arg;
736 
737 	cmn_err(CE_WARN, "Externally requested shutdown failed or "
738 	    "timed out.\nShutting down.\n");
739 
740 	switch (cmd) {
741 	case SHUTDOWN_HALT:
742 	case SHUTDOWN_POWEROFF:
743 		(void) kadmin(A_SHUTDOWN, AD_POWEROFF, NULL, kcred);
744 		break;
745 	case SHUTDOWN_REBOOT:
746 		(void) kadmin(A_REBOOT, AD_BOOT, NULL, kcred);
747 		break;
748 	}
749 }
750 
751 static void
xen_shutdown(void * arg)752 xen_shutdown(void *arg)
753 {
754 	int cmd = (uintptr_t)arg;
755 	proc_t *initpp;
756 
757 	ASSERT(cmd > SHUTDOWN_INVALID && cmd < SHUTDOWN_MAX);
758 
759 	if (cmd == SHUTDOWN_SUSPEND) {
760 		xen_suspend_domain();
761 		return;
762 	}
763 
764 	switch (cmd) {
765 	case SHUTDOWN_POWEROFF:
766 		force_shutdown_method = AD_POWEROFF;
767 		break;
768 	case SHUTDOWN_HALT:
769 		force_shutdown_method = AD_HALT;
770 		break;
771 	case SHUTDOWN_REBOOT:
772 		force_shutdown_method = AD_BOOT;
773 		break;
774 	}
775 
776 	/*
777 	 * If we're still booting and init(1) isn't set up yet, simply halt.
778 	 */
779 	mutex_enter(&pidlock);
780 	initpp = prfind(P_INITPID);
781 	mutex_exit(&pidlock);
782 	if (initpp == NULL) {
783 		extern void halt(char *);
784 		halt("Power off the System");   /* just in case */
785 	}
786 
787 	/*
788 	 * else, graceful shutdown with inittab and all getting involved
789 	 */
790 	psignal(initpp, SIGPWR);
791 
792 	(void) timeout(xen_dirty_shutdown, arg,
793 	    SHUTDOWN_TIMEOUT_SECS * drv_usectohz(MICROSEC));
794 }
795 
796 /*ARGSUSED*/
797 static void
xen_shutdown_handler(struct xenbus_watch * watch,const char ** vec,unsigned int len)798 xen_shutdown_handler(struct xenbus_watch *watch, const char **vec,
799     unsigned int len)
800 {
801 	char *str;
802 	xenbus_transaction_t xbt;
803 	int err, shutdown_code = SHUTDOWN_INVALID;
804 	unsigned int slen;
805 
806 again:
807 	err = xenbus_transaction_start(&xbt);
808 	if (err)
809 		return;
810 	if (xenbus_read(xbt, "control", "shutdown", (void *)&str, &slen)) {
811 		(void) xenbus_transaction_end(xbt, 1);
812 		return;
813 	}
814 
815 	SUSPEND_DEBUG("%d: xen_shutdown_handler: \"%s\"\n", CPU->cpu_id, str);
816 
817 	/*
818 	 * If this is a watch fired from our write below, check out early to
819 	 * avoid an infinite loop.
820 	 */
821 	if (strcmp(str, "") == 0) {
822 		(void) xenbus_transaction_end(xbt, 0);
823 		kmem_free(str, slen);
824 		return;
825 	} else if (strcmp(str, "poweroff") == 0) {
826 		shutdown_code = SHUTDOWN_POWEROFF;
827 	} else if (strcmp(str, "reboot") == 0) {
828 		shutdown_code = SHUTDOWN_REBOOT;
829 	} else if (strcmp(str, "suspend") == 0) {
830 		shutdown_code = SHUTDOWN_SUSPEND;
831 	} else if (strcmp(str, "halt") == 0) {
832 		shutdown_code = SHUTDOWN_HALT;
833 	} else {
834 		printf("Ignoring shutdown request: %s\n", str);
835 	}
836 
837 	/*
838 	 * XXPV	Should we check the value of xenbus_write() too, or are all
839 	 *	errors automatically folded into xenbus_transaction_end() ??
840 	 */
841 	(void) xenbus_write(xbt, "control", "shutdown", "");
842 	err = xenbus_transaction_end(xbt, 0);
843 	if (err == EAGAIN) {
844 		SUSPEND_DEBUG("%d: trying again\n", CPU->cpu_id);
845 		kmem_free(str, slen);
846 		goto again;
847 	}
848 
849 	kmem_free(str, slen);
850 	if (shutdown_code != SHUTDOWN_INVALID) {
851 		(void) taskq_dispatch(xen_shutdown_tq, xen_shutdown,
852 		    (void *)(intptr_t)shutdown_code, 0);
853 	}
854 }
855 
856 static struct xenbus_watch shutdown_watch;
857 static struct xenbus_watch sysrq_watch;
858 
859 void
xen_late_startup(void)860 xen_late_startup(void)
861 {
862 	if (!DOMAIN_IS_INITDOMAIN(xen_info)) {
863 		xen_shutdown_tq = taskq_create("shutdown_taskq", 1,
864 		    maxclsyspri - 1, 1, 1, TASKQ_PREPOPULATE);
865 		shutdown_watch.node = "control/shutdown";
866 		shutdown_watch.callback = xen_shutdown_handler;
867 		if (register_xenbus_watch(&shutdown_watch))
868 			cmn_err(CE_WARN, "Failed to set shutdown watcher");
869 
870 		sysrq_watch.node = "control/sysrq";
871 		sysrq_watch.callback = xen_sysrq_handler;
872 		if (register_xenbus_watch(&sysrq_watch))
873 			cmn_err(CE_WARN, "Failed to set sysrq watcher");
874 	}
875 	balloon_init(xen_info->nr_pages);
876 }
877 
878 #ifdef DEBUG
879 #define	XEN_PRINTF_BUFSIZE	1024
880 
881 char xen_printf_buffer[XEN_PRINTF_BUFSIZE];
882 
883 /*
884  * Printf function that calls hypervisor directly.  For DomU it only
885  * works when running on a xen hypervisor built with debug on.  Works
886  * always since no I/O ring interaction is needed.
887  */
888 /*PRINTFLIKE1*/
889 void
xen_printf(const char * fmt,...)890 xen_printf(const char *fmt, ...)
891 {
892 	va_list	ap;
893 
894 	va_start(ap, fmt);
895 	(void) vsnprintf(xen_printf_buffer, XEN_PRINTF_BUFSIZE, fmt, ap);
896 	va_end(ap);
897 
898 	(void) HYPERVISOR_console_io(CONSOLEIO_write,
899 	    strlen(xen_printf_buffer), xen_printf_buffer);
900 }
901 #else
902 void
xen_printf(const char * fmt,...)903 xen_printf(const char *fmt, ...)
904 {
905 }
906 #endif	/* DEBUG */
907 
908 void
startup_xen_version(void)909 startup_xen_version(void)
910 {
911 	xen_set_version(XENVER_BOOT_IDX);
912 	if (xen_hypervisor_supports_solaris(XEN_RUN_CHECK) == 0)
913 		cmn_err(CE_WARN, "Found hypervisor version: v%lu.%lu%s "
914 		    "but need at least version v3.0.4",
915 		    XENVER_CURRENT(xv_major), XENVER_CURRENT(xv_minor),
916 		    XENVER_CURRENT(xv_ver));
917 	xen_pte_workaround();
918 }
919 
920 int xen_mca_simulate_mc_physinfo_failure = 0;
921 
922 void
startup_xen_mca(void)923 startup_xen_mca(void)
924 {
925 	if (!DOMAIN_IS_INITDOMAIN(xen_info))
926 		return;
927 
928 	xen_phys_ncpus = 0;
929 	xen_phys_cpus = NULL;
930 
931 	if (xen_mca_simulate_mc_physinfo_failure ||
932 	    xen_get_mc_physcpuinfo(NULL, &xen_phys_ncpus) != 0) {
933 		cmn_err(CE_WARN,
934 		    "%sxen_get_mc_physinfo failure during xen MCA startup: "
935 		    "there will be no machine check support",
936 		    xen_mca_simulate_mc_physinfo_failure ? "(simulated) " : "");
937 		return;
938 	}
939 
940 	xen_phys_cpus = kmem_alloc(xen_phys_ncpus *
941 	    sizeof (xen_mc_logical_cpu_t), KM_NOSLEEP);
942 
943 	if (xen_phys_cpus == NULL) {
944 		cmn_err(CE_WARN,
945 		    "xen_get_mc_physinfo failure: can't allocate CPU array");
946 		return;
947 	}
948 
949 	if (xen_get_mc_physcpuinfo(xen_phys_cpus, &xen_phys_ncpus) != 0) {
950 		cmn_err(CE_WARN, "xen_get_mc_physinfo failure: no "
951 		    "physical CPU info");
952 		kmem_free(xen_phys_cpus,
953 		    xen_phys_ncpus * sizeof (xen_mc_logical_cpu_t));
954 		xen_phys_ncpus = 0;
955 		xen_phys_cpus = NULL;
956 	}
957 
958 	if (xen_physinfo_debug) {
959 		xen_mc_logical_cpu_t *xcp;
960 		unsigned i;
961 
962 		cmn_err(CE_NOTE, "xvm mca: %u physical cpus:\n",
963 		    xen_phys_ncpus);
964 		for (i = 0; i < xen_phys_ncpus; i++) {
965 			xcp = &xen_phys_cpus[i];
966 			cmn_err(CE_NOTE, "cpu%u: (%u, %u, %u) apid %u",
967 			    xcp->mc_cpunr, xcp->mc_chipid, xcp->mc_coreid,
968 			    xcp->mc_threadid, xcp->mc_apicid);
969 		}
970 	}
971 }
972 
973 /*
974  * Miscellaneous hypercall wrappers with slightly more verbose diagnostics.
975  */
976 
977 void
xen_set_gdt(ulong_t * frame_list,int entries)978 xen_set_gdt(ulong_t *frame_list, int entries)
979 {
980 	int err;
981 	if ((err = HYPERVISOR_set_gdt(frame_list, entries)) != 0) {
982 		/*
983 		 * X_EINVAL:	reserved entry or bad frames
984 		 * X_EFAULT:	bad address
985 		 */
986 		panic("xen_set_gdt(%p, %d): error %d",
987 		    (void *)frame_list, entries, -(int)err);
988 	}
989 }
990 
991 void
xen_set_ldt(user_desc_t * ldt,uint_t nsels)992 xen_set_ldt(user_desc_t *ldt, uint_t nsels)
993 {
994 	struct mmuext_op	op;
995 	long			err;
996 
997 	op.cmd = MMUEXT_SET_LDT;
998 	op.arg1.linear_addr = (uintptr_t)ldt;
999 	op.arg2.nr_ents = nsels;
1000 
1001 	if ((err = HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF)) != 0) {
1002 		panic("xen_set_ldt(%p, %d): error %d",
1003 		    (void *)ldt, nsels, -(int)err);
1004 	}
1005 }
1006 
1007 void
xen_stack_switch(ulong_t ss,ulong_t esp)1008 xen_stack_switch(ulong_t ss, ulong_t esp)
1009 {
1010 	long err;
1011 
1012 	if ((err = HYPERVISOR_stack_switch(ss, esp)) != 0) {
1013 		/*
1014 		 * X_EPERM:	bad selector
1015 		 */
1016 		panic("xen_stack_switch(%lx, %lx): error %d", ss, esp,
1017 		    -(int)err);
1018 	}
1019 }
1020 
1021 long
xen_set_trap_table(trap_info_t * table)1022 xen_set_trap_table(trap_info_t *table)
1023 {
1024 	long err;
1025 
1026 	if ((err = HYPERVISOR_set_trap_table(table)) != 0) {
1027 		/*
1028 		 * X_EFAULT:	bad address
1029 		 * X_EPERM:	bad selector
1030 		 */
1031 		panic("xen_set_trap_table(%p): error %d", (void *)table,
1032 		    -(int)err);
1033 	}
1034 	return (err);
1035 }
1036 
1037 void
xen_set_segment_base(int reg,ulong_t value)1038 xen_set_segment_base(int reg, ulong_t value)
1039 {
1040 	long err;
1041 
1042 	if ((err = HYPERVISOR_set_segment_base(reg, value)) != 0) {
1043 		/*
1044 		 * X_EFAULT:	bad address
1045 		 * X_EINVAL:	bad type
1046 		 */
1047 		panic("xen_set_segment_base(%d, %lx): error %d",
1048 		    reg, value, -(int)err);
1049 	}
1050 }
1051 
1052 /*
1053  * Translate a hypervisor errcode to a Solaris error code.
1054  */
1055 int
xen_xlate_errcode(int error)1056 xen_xlate_errcode(int error)
1057 {
1058 	switch (-error) {
1059 
1060 	/*
1061 	 * Translate hypervisor errno's into native errno's
1062 	 */
1063 
1064 #define	CASE(num)	case X_##num: error = num; break
1065 
1066 	CASE(EPERM);	CASE(ENOENT);	CASE(ESRCH);
1067 	CASE(EINTR);	CASE(EIO);	CASE(ENXIO);
1068 	CASE(E2BIG);	CASE(ENOMEM);	CASE(EACCES);
1069 	CASE(EFAULT);	CASE(EBUSY);	CASE(EEXIST);
1070 	CASE(ENODEV);	CASE(EISDIR);	CASE(EINVAL);
1071 	CASE(ENOSPC);	CASE(ESPIPE);	CASE(EROFS);
1072 	CASE(ENOSYS);	CASE(ENOTEMPTY); CASE(EISCONN);
1073 	CASE(ENODATA);	CASE(EAGAIN);
1074 
1075 #undef CASE
1076 
1077 	default:
1078 		panic("xen_xlate_errcode: unknown error %d", error);
1079 	}
1080 
1081 	return (error);
1082 }
1083 
1084 /*
1085  * Raise PS_IOPL on current vcpu to user level.
1086  * Caller responsible for preventing kernel preemption.
1087  */
1088 void
xen_enable_user_iopl(void * arg __unused)1089 xen_enable_user_iopl(void *arg __unused)
1090 {
1091 	physdev_set_iopl_t set_iopl;
1092 	set_iopl.iopl = 3;		/* user ring 3 */
1093 	(void) HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1094 }
1095 
1096 /*
1097  * Drop PS_IOPL on current vcpu to kernel level
1098  */
1099 void
xen_disable_user_iopl(void * arg __unused)1100 xen_disable_user_iopl(void *arg __unused)
1101 {
1102 	physdev_set_iopl_t set_iopl;
1103 	set_iopl.iopl = 1;		/* kernel pseudo ring 1 */
1104 	(void) HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1105 }
1106 
1107 int
xen_gdt_setprot(cpu_t * cp,uint_t prot)1108 xen_gdt_setprot(cpu_t *cp, uint_t prot)
1109 {
1110 	int err;
1111 	int pt_bits = PT_VALID;
1112 	if (prot & PROT_WRITE)
1113 		pt_bits |= PT_WRITABLE;
1114 
1115 	if ((err = as_setprot(&kas, (caddr_t)cp->cpu_gdt,
1116 	    MMU_PAGESIZE, prot)) != 0)
1117 		goto done;
1118 
1119 	err = xen_kpm_page(mmu_btop(cp->cpu_m.mcpu_gdtpa), pt_bits);
1120 
1121 done:
1122 	if (err) {
1123 		cmn_err(CE_WARN, "cpu%d: xen_gdt_setprot(%s) failed: error %d",
1124 		    cp->cpu_id, (prot & PROT_WRITE) ? "writable" : "read-only",
1125 		    err);
1126 	}
1127 
1128 	return (err);
1129 }
1130 
1131 int
xen_ldt_setprot(user_desc_t * ldt,size_t lsize,uint_t prot)1132 xen_ldt_setprot(user_desc_t *ldt, size_t lsize, uint_t prot)
1133 {
1134 	int err;
1135 	caddr_t	lva = (caddr_t)ldt;
1136 	int pt_bits = PT_VALID;
1137 	pgcnt_t npgs;
1138 	if (prot & PROT_WRITE)
1139 		pt_bits |= PT_WRITABLE;
1140 
1141 	if ((err = as_setprot(&kas, (caddr_t)ldt, lsize, prot)) != 0)
1142 		goto done;
1143 
1144 
1145 	ASSERT(IS_P2ALIGNED(lsize, PAGESIZE));
1146 	npgs = mmu_btop(lsize);
1147 	while (npgs--) {
1148 		if ((err = xen_kpm_page(hat_getpfnum(kas.a_hat, lva),
1149 		    pt_bits)) != 0)
1150 			break;
1151 		lva += PAGESIZE;
1152 	}
1153 
1154 done:
1155 	if (err) {
1156 		cmn_err(CE_WARN, "xen_ldt_setprot(%p, %s) failed: error %d",
1157 		    (void *)lva,
1158 		    (prot & PROT_WRITE) ? "writable" : "read-only", err);
1159 	}
1160 
1161 	return (err);
1162 }
1163 
1164 int
xen_get_mc_physcpuinfo(xen_mc_logical_cpu_t * log_cpus,uint_t * ncpus)1165 xen_get_mc_physcpuinfo(xen_mc_logical_cpu_t *log_cpus, uint_t *ncpus)
1166 {
1167 	xen_mc_t xmc;
1168 	struct xen_mc_physcpuinfo *cpi = &xmc.u.mc_physcpuinfo;
1169 
1170 	cpi->ncpus = *ncpus;
1171 	/*LINTED: constant in conditional context*/
1172 	set_xen_guest_handle(cpi->info, log_cpus);
1173 
1174 	if (HYPERVISOR_mca(XEN_MC_physcpuinfo, &xmc) != 0)
1175 		return (-1);
1176 
1177 	*ncpus = cpi->ncpus;
1178 	return (0);
1179 }
1180 
1181 void
print_panic(const char * str)1182 print_panic(const char *str)
1183 {
1184 	xen_printf(str);
1185 }
1186 
1187 /*
1188  * Interfaces to iterate over real cpu information, but only that info
1189  * which we choose to expose here.  These are of interest to dom0
1190  * only (and the backing hypercall should not work for domu).
1191  */
1192 
1193 xen_mc_lcpu_cookie_t
xen_physcpu_next(xen_mc_lcpu_cookie_t cookie)1194 xen_physcpu_next(xen_mc_lcpu_cookie_t cookie)
1195 {
1196 	xen_mc_logical_cpu_t *xcp = (xen_mc_logical_cpu_t *)cookie;
1197 
1198 	if (!DOMAIN_IS_INITDOMAIN(xen_info))
1199 		return (NULL);
1200 
1201 	if (cookie == NULL)
1202 		return ((xen_mc_lcpu_cookie_t)xen_phys_cpus);
1203 
1204 	if (xcp == xen_phys_cpus + xen_phys_ncpus - 1)
1205 		return (NULL);
1206 	else
1207 		return ((xen_mc_lcpu_cookie_t)++xcp);
1208 }
1209 
1210 #define	COOKIE2XCP(c) ((xen_mc_logical_cpu_t *)(c))
1211 
1212 const char *
xen_physcpu_vendorstr(xen_mc_lcpu_cookie_t cookie)1213 xen_physcpu_vendorstr(xen_mc_lcpu_cookie_t cookie)
1214 {
1215 	xen_mc_logical_cpu_t *xcp = COOKIE2XCP(cookie);
1216 
1217 	return ((const char *)&xcp->mc_vendorid[0]);
1218 }
1219 
1220 int
xen_physcpu_family(xen_mc_lcpu_cookie_t cookie)1221 xen_physcpu_family(xen_mc_lcpu_cookie_t cookie)
1222 {
1223 	return (COOKIE2XCP(cookie)->mc_family);
1224 }
1225 
1226 int
xen_physcpu_model(xen_mc_lcpu_cookie_t cookie)1227 xen_physcpu_model(xen_mc_lcpu_cookie_t cookie)
1228 {
1229 	return (COOKIE2XCP(cookie)->mc_model);
1230 }
1231 
1232 int
xen_physcpu_stepping(xen_mc_lcpu_cookie_t cookie)1233 xen_physcpu_stepping(xen_mc_lcpu_cookie_t cookie)
1234 {
1235 	return (COOKIE2XCP(cookie)->mc_step);
1236 }
1237 
1238 id_t
xen_physcpu_chipid(xen_mc_lcpu_cookie_t cookie)1239 xen_physcpu_chipid(xen_mc_lcpu_cookie_t cookie)
1240 {
1241 	return (COOKIE2XCP(cookie)->mc_chipid);
1242 }
1243 
1244 id_t
xen_physcpu_coreid(xen_mc_lcpu_cookie_t cookie)1245 xen_physcpu_coreid(xen_mc_lcpu_cookie_t cookie)
1246 {
1247 	return (COOKIE2XCP(cookie)->mc_coreid);
1248 }
1249 
1250 id_t
xen_physcpu_strandid(xen_mc_lcpu_cookie_t cookie)1251 xen_physcpu_strandid(xen_mc_lcpu_cookie_t cookie)
1252 {
1253 	return (COOKIE2XCP(cookie)->mc_threadid);
1254 }
1255 
1256 id_t
xen_physcpu_initial_apicid(xen_mc_lcpu_cookie_t cookie)1257 xen_physcpu_initial_apicid(xen_mc_lcpu_cookie_t cookie)
1258 {
1259 	return (COOKIE2XCP(cookie)->mc_clusterid);
1260 }
1261 
1262 id_t
xen_physcpu_logical_id(xen_mc_lcpu_cookie_t cookie)1263 xen_physcpu_logical_id(xen_mc_lcpu_cookie_t cookie)
1264 {
1265 	return (COOKIE2XCP(cookie)->mc_cpunr);
1266 }
1267 
1268 boolean_t
xen_physcpu_is_cmt(xen_mc_lcpu_cookie_t cookie)1269 xen_physcpu_is_cmt(xen_mc_lcpu_cookie_t cookie)
1270 {
1271 	return (COOKIE2XCP(cookie)->mc_nthreads > 1);
1272 }
1273 
1274 uint64_t
xen_physcpu_mcg_cap(xen_mc_lcpu_cookie_t cookie)1275 xen_physcpu_mcg_cap(xen_mc_lcpu_cookie_t cookie)
1276 {
1277 	xen_mc_logical_cpu_t *xcp = COOKIE2XCP(cookie);
1278 
1279 	/*
1280 	 * Need to #define the indices, or search through the array.
1281 	 */
1282 	return (xcp->mc_msrvalues[0].value);
1283 }
1284 
1285 int
xen_map_gref(uint_t cmd,gnttab_map_grant_ref_t * mapop,uint_t count,boolean_t uvaddr)1286 xen_map_gref(uint_t cmd, gnttab_map_grant_ref_t *mapop, uint_t count,
1287     boolean_t uvaddr)
1288 {
1289 	long rc;
1290 	uint_t i;
1291 
1292 	ASSERT(cmd == GNTTABOP_map_grant_ref);
1293 
1294 #if !defined(_BOOT)
1295 	if (uvaddr == B_FALSE) {
1296 		for (i = 0; i < count; ++i) {
1297 			mapop[i].flags |= (PT_FOREIGN <<_GNTMAP_guest_avail0);
1298 		}
1299 	}
1300 #endif
1301 
1302 	rc = HYPERVISOR_grant_table_op(cmd, mapop, count);
1303 
1304 	return (rc);
1305 }
1306 
1307 static int
xpv_get_physinfo(xen_sysctl_physinfo_t * pi)1308 xpv_get_physinfo(xen_sysctl_physinfo_t *pi)
1309 {
1310 	xen_sysctl_t op;
1311 	struct sp { void *p; } *sp = (struct sp *)&op.u.physinfo.cpu_to_node;
1312 	int ret;
1313 
1314 	bzero(&op, sizeof (op));
1315 	op.cmd = XEN_SYSCTL_physinfo;
1316 	op.interface_version = XEN_SYSCTL_INTERFACE_VERSION;
1317 	/*LINTED: constant in conditional context*/
1318 	set_xen_guest_handle(*sp, NULL);
1319 
1320 	ret = HYPERVISOR_sysctl(&op);
1321 
1322 	if (ret != 0)
1323 		return (xen_xlate_errcode(ret));
1324 
1325 	bcopy(&op.u.physinfo, pi, sizeof (op.u.physinfo));
1326 	return (0);
1327 }
1328 
1329 /*
1330  * On dom0, we can determine the number of physical cpus on the machine.
1331  * This number is important when figuring out what workarounds are
1332  * appropriate, so compute it now.
1333  */
1334 uint_t
xpv_nr_phys_cpus(void)1335 xpv_nr_phys_cpus(void)
1336 {
1337 	static uint_t nphyscpus = 0;
1338 
1339 	ASSERT(DOMAIN_IS_INITDOMAIN(xen_info));
1340 
1341 	if (nphyscpus == 0) {
1342 		xen_sysctl_physinfo_t pi;
1343 		int ret;
1344 
1345 		if ((ret = xpv_get_physinfo(&pi)) != 0)
1346 			panic("xpv_get_physinfo() failed: %d\n", ret);
1347 		nphyscpus = pi.nr_cpus;
1348 	}
1349 	return (nphyscpus);
1350 }
1351 
1352 pgcnt_t
xpv_nr_phys_pages(void)1353 xpv_nr_phys_pages(void)
1354 {
1355 	xen_sysctl_physinfo_t pi;
1356 	int ret;
1357 
1358 	ASSERT(DOMAIN_IS_INITDOMAIN(xen_info));
1359 
1360 	if ((ret = xpv_get_physinfo(&pi)) != 0)
1361 		panic("xpv_get_physinfo() failed: %d\n", ret);
1362 
1363 	return ((pgcnt_t)pi.total_pages);
1364 }
1365 
1366 uint64_t
xpv_cpu_khz(void)1367 xpv_cpu_khz(void)
1368 {
1369 	xen_sysctl_physinfo_t pi;
1370 	int ret;
1371 
1372 	ASSERT(DOMAIN_IS_INITDOMAIN(xen_info));
1373 
1374 	if ((ret = xpv_get_physinfo(&pi)) != 0)
1375 		panic("xpv_get_physinfo() failed: %d\n", ret);
1376 	return ((uint64_t)pi.cpu_khz);
1377 }
1378