xref: /openbmc/linux/arch/x86/kvm/xen.c (revision 67ff4a72)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright © 2019 Oracle and/or its affiliates. All rights reserved.
4  * Copyright © 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved.
5  *
6  * KVM Xen emulation
7  */
8 
9 #include "x86.h"
10 #include "xen.h"
11 #include "hyperv.h"
12 
13 #include <linux/kvm_host.h>
14 #include <linux/sched/stat.h>
15 
16 #include <trace/events/kvm.h>
17 #include <xen/interface/xen.h>
18 #include <xen/interface/vcpu.h>
19 
20 #include "trace.h"
21 
22 DEFINE_STATIC_KEY_DEFERRED_FALSE(kvm_xen_enabled, HZ);
23 
24 static int kvm_xen_shared_info_init(struct kvm *kvm, gfn_t gfn)
25 {
26 	gpa_t gpa = gfn_to_gpa(gfn);
27 	int wc_ofs, sec_hi_ofs;
28 	int ret = 0;
29 	int idx = srcu_read_lock(&kvm->srcu);
30 
31 	if (kvm_is_error_hva(gfn_to_hva(kvm, gfn))) {
32 		ret = -EFAULT;
33 		goto out;
34 	}
35 	kvm->arch.xen.shinfo_gfn = gfn;
36 
37 	/* Paranoia checks on the 32-bit struct layout */
38 	BUILD_BUG_ON(offsetof(struct compat_shared_info, wc) != 0x900);
39 	BUILD_BUG_ON(offsetof(struct compat_shared_info, arch.wc_sec_hi) != 0x924);
40 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
41 
42 	/* 32-bit location by default */
43 	wc_ofs = offsetof(struct compat_shared_info, wc);
44 	sec_hi_ofs = offsetof(struct compat_shared_info, arch.wc_sec_hi);
45 
46 #ifdef CONFIG_X86_64
47 	/* Paranoia checks on the 64-bit struct layout */
48 	BUILD_BUG_ON(offsetof(struct shared_info, wc) != 0xc00);
49 	BUILD_BUG_ON(offsetof(struct shared_info, wc_sec_hi) != 0xc0c);
50 
51 	if (kvm->arch.xen.long_mode) {
52 		wc_ofs = offsetof(struct shared_info, wc);
53 		sec_hi_ofs = offsetof(struct shared_info, wc_sec_hi);
54 	}
55 #endif
56 
57 	kvm_write_wall_clock(kvm, gpa + wc_ofs, sec_hi_ofs - wc_ofs);
58 	kvm_make_all_cpus_request(kvm, KVM_REQ_MASTERCLOCK_UPDATE);
59 
60 out:
61 	srcu_read_unlock(&kvm->srcu, idx);
62 	return ret;
63 }
64 
65 static void kvm_xen_update_runstate(struct kvm_vcpu *v, int state)
66 {
67 	struct kvm_vcpu_xen *vx = &v->arch.xen;
68 	u64 now = get_kvmclock_ns(v->kvm);
69 	u64 delta_ns = now - vx->runstate_entry_time;
70 	u64 run_delay = current->sched_info.run_delay;
71 
72 	if (unlikely(!vx->runstate_entry_time))
73 		vx->current_runstate = RUNSTATE_offline;
74 
75 	/*
76 	 * Time waiting for the scheduler isn't "stolen" if the
77 	 * vCPU wasn't running anyway.
78 	 */
79 	if (vx->current_runstate == RUNSTATE_running) {
80 		u64 steal_ns = run_delay - vx->last_steal;
81 
82 		delta_ns -= steal_ns;
83 
84 		vx->runstate_times[RUNSTATE_runnable] += steal_ns;
85 	}
86 	vx->last_steal = run_delay;
87 
88 	vx->runstate_times[vx->current_runstate] += delta_ns;
89 	vx->current_runstate = state;
90 	vx->runstate_entry_time = now;
91 }
92 
93 void kvm_xen_update_runstate_guest(struct kvm_vcpu *v, int state)
94 {
95 	struct kvm_vcpu_xen *vx = &v->arch.xen;
96 	uint64_t state_entry_time;
97 	unsigned int offset;
98 
99 	kvm_xen_update_runstate(v, state);
100 
101 	if (!vx->runstate_set)
102 		return;
103 
104 	BUILD_BUG_ON(sizeof(struct compat_vcpu_runstate_info) != 0x2c);
105 
106 	offset = offsetof(struct compat_vcpu_runstate_info, state_entry_time);
107 #ifdef CONFIG_X86_64
108 	/*
109 	 * The only difference is alignment of uint64_t in 32-bit.
110 	 * So the first field 'state' is accessed directly using
111 	 * offsetof() (where its offset happens to be zero), while the
112 	 * remaining fields which are all uint64_t, start at 'offset'
113 	 * which we tweak here by adding 4.
114 	 */
115 	BUILD_BUG_ON(offsetof(struct vcpu_runstate_info, state_entry_time) !=
116 		     offsetof(struct compat_vcpu_runstate_info, state_entry_time) + 4);
117 	BUILD_BUG_ON(offsetof(struct vcpu_runstate_info, time) !=
118 		     offsetof(struct compat_vcpu_runstate_info, time) + 4);
119 
120 	if (v->kvm->arch.xen.long_mode)
121 		offset = offsetof(struct vcpu_runstate_info, state_entry_time);
122 #endif
123 	/*
124 	 * First write the updated state_entry_time at the appropriate
125 	 * location determined by 'offset'.
126 	 */
127 	state_entry_time = vx->runstate_entry_time;
128 	state_entry_time |= XEN_RUNSTATE_UPDATE;
129 
130 	BUILD_BUG_ON(sizeof_field(struct vcpu_runstate_info, state_entry_time) !=
131 		     sizeof(state_entry_time));
132 	BUILD_BUG_ON(sizeof_field(struct compat_vcpu_runstate_info, state_entry_time) !=
133 		     sizeof(state_entry_time));
134 
135 	if (kvm_write_guest_offset_cached(v->kvm, &v->arch.xen.runstate_cache,
136 					  &state_entry_time, offset,
137 					  sizeof(state_entry_time)))
138 		return;
139 	smp_wmb();
140 
141 	/*
142 	 * Next, write the new runstate. This is in the *same* place
143 	 * for 32-bit and 64-bit guests, asserted here for paranoia.
144 	 */
145 	BUILD_BUG_ON(offsetof(struct vcpu_runstate_info, state) !=
146 		     offsetof(struct compat_vcpu_runstate_info, state));
147 	BUILD_BUG_ON(sizeof_field(struct vcpu_runstate_info, state) !=
148 		     sizeof(vx->current_runstate));
149 	BUILD_BUG_ON(sizeof_field(struct compat_vcpu_runstate_info, state) !=
150 		     sizeof(vx->current_runstate));
151 
152 	if (kvm_write_guest_offset_cached(v->kvm, &v->arch.xen.runstate_cache,
153 					  &vx->current_runstate,
154 					  offsetof(struct vcpu_runstate_info, state),
155 					  sizeof(vx->current_runstate)))
156 		return;
157 
158 	/*
159 	 * Write the actual runstate times immediately after the
160 	 * runstate_entry_time.
161 	 */
162 	BUILD_BUG_ON(offsetof(struct vcpu_runstate_info, state_entry_time) !=
163 		     offsetof(struct vcpu_runstate_info, time) - sizeof(u64));
164 	BUILD_BUG_ON(offsetof(struct compat_vcpu_runstate_info, state_entry_time) !=
165 		     offsetof(struct compat_vcpu_runstate_info, time) - sizeof(u64));
166 	BUILD_BUG_ON(sizeof_field(struct vcpu_runstate_info, time) !=
167 		     sizeof_field(struct compat_vcpu_runstate_info, time));
168 	BUILD_BUG_ON(sizeof_field(struct vcpu_runstate_info, time) !=
169 		     sizeof(vx->runstate_times));
170 
171 	if (kvm_write_guest_offset_cached(v->kvm, &v->arch.xen.runstate_cache,
172 					  &vx->runstate_times[0],
173 					  offset + sizeof(u64),
174 					  sizeof(vx->runstate_times)))
175 		return;
176 
177 	smp_wmb();
178 
179 	/*
180 	 * Finally, clear the XEN_RUNSTATE_UPDATE bit in the guest's
181 	 * runstate_entry_time field.
182 	 */
183 
184 	state_entry_time &= ~XEN_RUNSTATE_UPDATE;
185 	if (kvm_write_guest_offset_cached(v->kvm, &v->arch.xen.runstate_cache,
186 					  &state_entry_time, offset,
187 					  sizeof(state_entry_time)))
188 		return;
189 }
190 
191 int __kvm_xen_has_interrupt(struct kvm_vcpu *v)
192 {
193 	int err;
194 	u8 rc = 0;
195 
196 	/*
197 	 * If the global upcall vector (HVMIRQ_callback_vector) is set and
198 	 * the vCPU's evtchn_upcall_pending flag is set, the IRQ is pending.
199 	 */
200 	struct gfn_to_hva_cache *ghc = &v->arch.xen.vcpu_info_cache;
201 	struct kvm_memslots *slots = kvm_memslots(v->kvm);
202 	unsigned int offset = offsetof(struct vcpu_info, evtchn_upcall_pending);
203 
204 	/* No need for compat handling here */
205 	BUILD_BUG_ON(offsetof(struct vcpu_info, evtchn_upcall_pending) !=
206 		     offsetof(struct compat_vcpu_info, evtchn_upcall_pending));
207 	BUILD_BUG_ON(sizeof(rc) !=
208 		     sizeof_field(struct vcpu_info, evtchn_upcall_pending));
209 	BUILD_BUG_ON(sizeof(rc) !=
210 		     sizeof_field(struct compat_vcpu_info, evtchn_upcall_pending));
211 
212 	/*
213 	 * For efficiency, this mirrors the checks for using the valid
214 	 * cache in kvm_read_guest_offset_cached(), but just uses
215 	 * __get_user() instead. And falls back to the slow path.
216 	 */
217 	if (likely(slots->generation == ghc->generation &&
218 		   !kvm_is_error_hva(ghc->hva) && ghc->memslot)) {
219 		/* Fast path */
220 		pagefault_disable();
221 		err = __get_user(rc, (u8 __user *)ghc->hva + offset);
222 		pagefault_enable();
223 		if (!err)
224 			return rc;
225 	}
226 
227 	/* Slow path */
228 
229 	/*
230 	 * This function gets called from kvm_vcpu_block() after setting the
231 	 * task to TASK_INTERRUPTIBLE, to see if it needs to wake immediately
232 	 * from a HLT. So we really mustn't sleep. If the page ended up absent
233 	 * at that point, just return 1 in order to trigger an immediate wake,
234 	 * and we'll end up getting called again from a context where we *can*
235 	 * fault in the page and wait for it.
236 	 */
237 	if (in_atomic() || !task_is_running(current))
238 		return 1;
239 
240 	kvm_read_guest_offset_cached(v->kvm, ghc, &rc, offset,
241 				     sizeof(rc));
242 
243 	return rc;
244 }
245 
246 int kvm_xen_hvm_set_attr(struct kvm *kvm, struct kvm_xen_hvm_attr *data)
247 {
248 	int r = -ENOENT;
249 
250 	mutex_lock(&kvm->lock);
251 
252 	switch (data->type) {
253 	case KVM_XEN_ATTR_TYPE_LONG_MODE:
254 		if (!IS_ENABLED(CONFIG_64BIT) && data->u.long_mode) {
255 			r = -EINVAL;
256 		} else {
257 			kvm->arch.xen.long_mode = !!data->u.long_mode;
258 			r = 0;
259 		}
260 		break;
261 
262 	case KVM_XEN_ATTR_TYPE_SHARED_INFO:
263 		if (data->u.shared_info.gfn == GPA_INVALID) {
264 			kvm->arch.xen.shinfo_gfn = GPA_INVALID;
265 			r = 0;
266 			break;
267 		}
268 		r = kvm_xen_shared_info_init(kvm, data->u.shared_info.gfn);
269 		break;
270 
271 
272 	case KVM_XEN_ATTR_TYPE_UPCALL_VECTOR:
273 		if (data->u.vector && data->u.vector < 0x10)
274 			r = -EINVAL;
275 		else {
276 			kvm->arch.xen.upcall_vector = data->u.vector;
277 			r = 0;
278 		}
279 		break;
280 
281 	default:
282 		break;
283 	}
284 
285 	mutex_unlock(&kvm->lock);
286 	return r;
287 }
288 
289 int kvm_xen_hvm_get_attr(struct kvm *kvm, struct kvm_xen_hvm_attr *data)
290 {
291 	int r = -ENOENT;
292 
293 	mutex_lock(&kvm->lock);
294 
295 	switch (data->type) {
296 	case KVM_XEN_ATTR_TYPE_LONG_MODE:
297 		data->u.long_mode = kvm->arch.xen.long_mode;
298 		r = 0;
299 		break;
300 
301 	case KVM_XEN_ATTR_TYPE_SHARED_INFO:
302 		data->u.shared_info.gfn = kvm->arch.xen.shinfo_gfn;
303 		r = 0;
304 		break;
305 
306 	case KVM_XEN_ATTR_TYPE_UPCALL_VECTOR:
307 		data->u.vector = kvm->arch.xen.upcall_vector;
308 		r = 0;
309 		break;
310 
311 	default:
312 		break;
313 	}
314 
315 	mutex_unlock(&kvm->lock);
316 	return r;
317 }
318 
319 int kvm_xen_vcpu_set_attr(struct kvm_vcpu *vcpu, struct kvm_xen_vcpu_attr *data)
320 {
321 	int idx, r = -ENOENT;
322 
323 	mutex_lock(&vcpu->kvm->lock);
324 	idx = srcu_read_lock(&vcpu->kvm->srcu);
325 
326 	switch (data->type) {
327 	case KVM_XEN_VCPU_ATTR_TYPE_VCPU_INFO:
328 		/* No compat necessary here. */
329 		BUILD_BUG_ON(sizeof(struct vcpu_info) !=
330 			     sizeof(struct compat_vcpu_info));
331 		BUILD_BUG_ON(offsetof(struct vcpu_info, time) !=
332 			     offsetof(struct compat_vcpu_info, time));
333 
334 		if (data->u.gpa == GPA_INVALID) {
335 			vcpu->arch.xen.vcpu_info_set = false;
336 			r = 0;
337 			break;
338 		}
339 
340 		r = kvm_gfn_to_hva_cache_init(vcpu->kvm,
341 					      &vcpu->arch.xen.vcpu_info_cache,
342 					      data->u.gpa,
343 					      sizeof(struct vcpu_info));
344 		if (!r) {
345 			vcpu->arch.xen.vcpu_info_set = true;
346 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
347 		}
348 		break;
349 
350 	case KVM_XEN_VCPU_ATTR_TYPE_VCPU_TIME_INFO:
351 		if (data->u.gpa == GPA_INVALID) {
352 			vcpu->arch.xen.vcpu_time_info_set = false;
353 			r = 0;
354 			break;
355 		}
356 
357 		r = kvm_gfn_to_hva_cache_init(vcpu->kvm,
358 					      &vcpu->arch.xen.vcpu_time_info_cache,
359 					      data->u.gpa,
360 					      sizeof(struct pvclock_vcpu_time_info));
361 		if (!r) {
362 			vcpu->arch.xen.vcpu_time_info_set = true;
363 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
364 		}
365 		break;
366 
367 	case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_ADDR:
368 		if (!sched_info_on()) {
369 			r = -EOPNOTSUPP;
370 			break;
371 		}
372 		if (data->u.gpa == GPA_INVALID) {
373 			vcpu->arch.xen.runstate_set = false;
374 			r = 0;
375 			break;
376 		}
377 
378 		r = kvm_gfn_to_hva_cache_init(vcpu->kvm,
379 					      &vcpu->arch.xen.runstate_cache,
380 					      data->u.gpa,
381 					      sizeof(struct vcpu_runstate_info));
382 		if (!r) {
383 			vcpu->arch.xen.runstate_set = true;
384 		}
385 		break;
386 
387 	case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_CURRENT:
388 		if (!sched_info_on()) {
389 			r = -EOPNOTSUPP;
390 			break;
391 		}
392 		if (data->u.runstate.state > RUNSTATE_offline) {
393 			r = -EINVAL;
394 			break;
395 		}
396 
397 		kvm_xen_update_runstate(vcpu, data->u.runstate.state);
398 		r = 0;
399 		break;
400 
401 	case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_DATA:
402 		if (!sched_info_on()) {
403 			r = -EOPNOTSUPP;
404 			break;
405 		}
406 		if (data->u.runstate.state > RUNSTATE_offline) {
407 			r = -EINVAL;
408 			break;
409 		}
410 		if (data->u.runstate.state_entry_time !=
411 		    (data->u.runstate.time_running +
412 		     data->u.runstate.time_runnable +
413 		     data->u.runstate.time_blocked +
414 		     data->u.runstate.time_offline)) {
415 			r = -EINVAL;
416 			break;
417 		}
418 		if (get_kvmclock_ns(vcpu->kvm) <
419 		    data->u.runstate.state_entry_time) {
420 			r = -EINVAL;
421 			break;
422 		}
423 
424 		vcpu->arch.xen.current_runstate = data->u.runstate.state;
425 		vcpu->arch.xen.runstate_entry_time =
426 			data->u.runstate.state_entry_time;
427 		vcpu->arch.xen.runstate_times[RUNSTATE_running] =
428 			data->u.runstate.time_running;
429 		vcpu->arch.xen.runstate_times[RUNSTATE_runnable] =
430 			data->u.runstate.time_runnable;
431 		vcpu->arch.xen.runstate_times[RUNSTATE_blocked] =
432 			data->u.runstate.time_blocked;
433 		vcpu->arch.xen.runstate_times[RUNSTATE_offline] =
434 			data->u.runstate.time_offline;
435 		vcpu->arch.xen.last_steal = current->sched_info.run_delay;
436 		r = 0;
437 		break;
438 
439 	case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_ADJUST:
440 		if (!sched_info_on()) {
441 			r = -EOPNOTSUPP;
442 			break;
443 		}
444 		if (data->u.runstate.state > RUNSTATE_offline &&
445 		    data->u.runstate.state != (u64)-1) {
446 			r = -EINVAL;
447 			break;
448 		}
449 		/* The adjustment must add up */
450 		if (data->u.runstate.state_entry_time !=
451 		    (data->u.runstate.time_running +
452 		     data->u.runstate.time_runnable +
453 		     data->u.runstate.time_blocked +
454 		     data->u.runstate.time_offline)) {
455 			r = -EINVAL;
456 			break;
457 		}
458 
459 		if (get_kvmclock_ns(vcpu->kvm) <
460 		    (vcpu->arch.xen.runstate_entry_time +
461 		     data->u.runstate.state_entry_time)) {
462 			r = -EINVAL;
463 			break;
464 		}
465 
466 		vcpu->arch.xen.runstate_entry_time +=
467 			data->u.runstate.state_entry_time;
468 		vcpu->arch.xen.runstate_times[RUNSTATE_running] +=
469 			data->u.runstate.time_running;
470 		vcpu->arch.xen.runstate_times[RUNSTATE_runnable] +=
471 			data->u.runstate.time_runnable;
472 		vcpu->arch.xen.runstate_times[RUNSTATE_blocked] +=
473 			data->u.runstate.time_blocked;
474 		vcpu->arch.xen.runstate_times[RUNSTATE_offline] +=
475 			data->u.runstate.time_offline;
476 
477 		if (data->u.runstate.state <= RUNSTATE_offline)
478 			kvm_xen_update_runstate(vcpu, data->u.runstate.state);
479 		r = 0;
480 		break;
481 
482 	default:
483 		break;
484 	}
485 
486 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
487 	mutex_unlock(&vcpu->kvm->lock);
488 	return r;
489 }
490 
491 int kvm_xen_vcpu_get_attr(struct kvm_vcpu *vcpu, struct kvm_xen_vcpu_attr *data)
492 {
493 	int r = -ENOENT;
494 
495 	mutex_lock(&vcpu->kvm->lock);
496 
497 	switch (data->type) {
498 	case KVM_XEN_VCPU_ATTR_TYPE_VCPU_INFO:
499 		if (vcpu->arch.xen.vcpu_info_set)
500 			data->u.gpa = vcpu->arch.xen.vcpu_info_cache.gpa;
501 		else
502 			data->u.gpa = GPA_INVALID;
503 		r = 0;
504 		break;
505 
506 	case KVM_XEN_VCPU_ATTR_TYPE_VCPU_TIME_INFO:
507 		if (vcpu->arch.xen.vcpu_time_info_set)
508 			data->u.gpa = vcpu->arch.xen.vcpu_time_info_cache.gpa;
509 		else
510 			data->u.gpa = GPA_INVALID;
511 		r = 0;
512 		break;
513 
514 	case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_ADDR:
515 		if (!sched_info_on()) {
516 			r = -EOPNOTSUPP;
517 			break;
518 		}
519 		if (vcpu->arch.xen.runstate_set) {
520 			data->u.gpa = vcpu->arch.xen.runstate_cache.gpa;
521 			r = 0;
522 		}
523 		break;
524 
525 	case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_CURRENT:
526 		if (!sched_info_on()) {
527 			r = -EOPNOTSUPP;
528 			break;
529 		}
530 		data->u.runstate.state = vcpu->arch.xen.current_runstate;
531 		r = 0;
532 		break;
533 
534 	case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_DATA:
535 		if (!sched_info_on()) {
536 			r = -EOPNOTSUPP;
537 			break;
538 		}
539 		data->u.runstate.state = vcpu->arch.xen.current_runstate;
540 		data->u.runstate.state_entry_time =
541 			vcpu->arch.xen.runstate_entry_time;
542 		data->u.runstate.time_running =
543 			vcpu->arch.xen.runstate_times[RUNSTATE_running];
544 		data->u.runstate.time_runnable =
545 			vcpu->arch.xen.runstate_times[RUNSTATE_runnable];
546 		data->u.runstate.time_blocked =
547 			vcpu->arch.xen.runstate_times[RUNSTATE_blocked];
548 		data->u.runstate.time_offline =
549 			vcpu->arch.xen.runstate_times[RUNSTATE_offline];
550 		r = 0;
551 		break;
552 
553 	case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_ADJUST:
554 		r = -EINVAL;
555 		break;
556 
557 	default:
558 		break;
559 	}
560 
561 	mutex_unlock(&vcpu->kvm->lock);
562 	return r;
563 }
564 
565 int kvm_xen_write_hypercall_page(struct kvm_vcpu *vcpu, u64 data)
566 {
567 	struct kvm *kvm = vcpu->kvm;
568 	u32 page_num = data & ~PAGE_MASK;
569 	u64 page_addr = data & PAGE_MASK;
570 	bool lm = is_long_mode(vcpu);
571 
572 	/* Latch long_mode for shared_info pages etc. */
573 	vcpu->kvm->arch.xen.long_mode = lm;
574 
575 	/*
576 	 * If Xen hypercall intercept is enabled, fill the hypercall
577 	 * page with VMCALL/VMMCALL instructions since that's what
578 	 * we catch. Else the VMM has provided the hypercall pages
579 	 * with instructions of its own choosing, so use those.
580 	 */
581 	if (kvm_xen_hypercall_enabled(kvm)) {
582 		u8 instructions[32];
583 		int i;
584 
585 		if (page_num)
586 			return 1;
587 
588 		/* mov imm32, %eax */
589 		instructions[0] = 0xb8;
590 
591 		/* vmcall / vmmcall */
592 		kvm_x86_ops.patch_hypercall(vcpu, instructions + 5);
593 
594 		/* ret */
595 		instructions[8] = 0xc3;
596 
597 		/* int3 to pad */
598 		memset(instructions + 9, 0xcc, sizeof(instructions) - 9);
599 
600 		for (i = 0; i < PAGE_SIZE / sizeof(instructions); i++) {
601 			*(u32 *)&instructions[1] = i;
602 			if (kvm_vcpu_write_guest(vcpu,
603 						 page_addr + (i * sizeof(instructions)),
604 						 instructions, sizeof(instructions)))
605 				return 1;
606 		}
607 	} else {
608 		/*
609 		 * Note, truncation is a non-issue as 'lm' is guaranteed to be
610 		 * false for a 32-bit kernel, i.e. when hva_t is only 4 bytes.
611 		 */
612 		hva_t blob_addr = lm ? kvm->arch.xen_hvm_config.blob_addr_64
613 				     : kvm->arch.xen_hvm_config.blob_addr_32;
614 		u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
615 				  : kvm->arch.xen_hvm_config.blob_size_32;
616 		u8 *page;
617 
618 		if (page_num >= blob_size)
619 			return 1;
620 
621 		blob_addr += page_num * PAGE_SIZE;
622 
623 		page = memdup_user((u8 __user *)blob_addr, PAGE_SIZE);
624 		if (IS_ERR(page))
625 			return PTR_ERR(page);
626 
627 		if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE)) {
628 			kfree(page);
629 			return 1;
630 		}
631 	}
632 	return 0;
633 }
634 
635 int kvm_xen_hvm_config(struct kvm *kvm, struct kvm_xen_hvm_config *xhc)
636 {
637 	if (xhc->flags & ~KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL)
638 		return -EINVAL;
639 
640 	/*
641 	 * With hypercall interception the kernel generates its own
642 	 * hypercall page so it must not be provided.
643 	 */
644 	if ((xhc->flags & KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL) &&
645 	    (xhc->blob_addr_32 || xhc->blob_addr_64 ||
646 	     xhc->blob_size_32 || xhc->blob_size_64))
647 		return -EINVAL;
648 
649 	mutex_lock(&kvm->lock);
650 
651 	if (xhc->msr && !kvm->arch.xen_hvm_config.msr)
652 		static_branch_inc(&kvm_xen_enabled.key);
653 	else if (!xhc->msr && kvm->arch.xen_hvm_config.msr)
654 		static_branch_slow_dec_deferred(&kvm_xen_enabled);
655 
656 	memcpy(&kvm->arch.xen_hvm_config, xhc, sizeof(*xhc));
657 
658 	mutex_unlock(&kvm->lock);
659 	return 0;
660 }
661 
662 void kvm_xen_init_vm(struct kvm *kvm)
663 {
664 	kvm->arch.xen.shinfo_gfn = GPA_INVALID;
665 }
666 
667 void kvm_xen_destroy_vm(struct kvm *kvm)
668 {
669 	if (kvm->arch.xen_hvm_config.msr)
670 		static_branch_slow_dec_deferred(&kvm_xen_enabled);
671 }
672 
673 static int kvm_xen_hypercall_set_result(struct kvm_vcpu *vcpu, u64 result)
674 {
675 	kvm_rax_write(vcpu, result);
676 	return kvm_skip_emulated_instruction(vcpu);
677 }
678 
679 static int kvm_xen_hypercall_complete_userspace(struct kvm_vcpu *vcpu)
680 {
681 	struct kvm_run *run = vcpu->run;
682 
683 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.xen.hypercall_rip)))
684 		return 1;
685 
686 	return kvm_xen_hypercall_set_result(vcpu, run->xen.u.hcall.result);
687 }
688 
689 int kvm_xen_hypercall(struct kvm_vcpu *vcpu)
690 {
691 	bool longmode;
692 	u64 input, params[6];
693 
694 	input = (u64)kvm_register_read(vcpu, VCPU_REGS_RAX);
695 
696 	/* Hyper-V hypercalls get bit 31 set in EAX */
697 	if ((input & 0x80000000) &&
698 	    kvm_hv_hypercall_enabled(vcpu))
699 		return kvm_hv_hypercall(vcpu);
700 
701 	longmode = is_64_bit_hypercall(vcpu);
702 	if (!longmode) {
703 		params[0] = (u32)kvm_rbx_read(vcpu);
704 		params[1] = (u32)kvm_rcx_read(vcpu);
705 		params[2] = (u32)kvm_rdx_read(vcpu);
706 		params[3] = (u32)kvm_rsi_read(vcpu);
707 		params[4] = (u32)kvm_rdi_read(vcpu);
708 		params[5] = (u32)kvm_rbp_read(vcpu);
709 	}
710 #ifdef CONFIG_X86_64
711 	else {
712 		params[0] = (u64)kvm_rdi_read(vcpu);
713 		params[1] = (u64)kvm_rsi_read(vcpu);
714 		params[2] = (u64)kvm_rdx_read(vcpu);
715 		params[3] = (u64)kvm_r10_read(vcpu);
716 		params[4] = (u64)kvm_r8_read(vcpu);
717 		params[5] = (u64)kvm_r9_read(vcpu);
718 	}
719 #endif
720 	trace_kvm_xen_hypercall(input, params[0], params[1], params[2],
721 				params[3], params[4], params[5]);
722 
723 	vcpu->run->exit_reason = KVM_EXIT_XEN;
724 	vcpu->run->xen.type = KVM_EXIT_XEN_HCALL;
725 	vcpu->run->xen.u.hcall.longmode = longmode;
726 	vcpu->run->xen.u.hcall.cpl = kvm_x86_ops.get_cpl(vcpu);
727 	vcpu->run->xen.u.hcall.input = input;
728 	vcpu->run->xen.u.hcall.params[0] = params[0];
729 	vcpu->run->xen.u.hcall.params[1] = params[1];
730 	vcpu->run->xen.u.hcall.params[2] = params[2];
731 	vcpu->run->xen.u.hcall.params[3] = params[3];
732 	vcpu->run->xen.u.hcall.params[4] = params[4];
733 	vcpu->run->xen.u.hcall.params[5] = params[5];
734 	vcpu->arch.xen.hypercall_rip = kvm_get_linear_rip(vcpu);
735 	vcpu->arch.complete_userspace_io =
736 		kvm_xen_hypercall_complete_userspace;
737 
738 	return 0;
739 }
740