• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 
kvm_xen_shared_info_init(struct kvm * kvm,gfn_t gfn)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 
kvm_xen_update_runstate(struct kvm_vcpu * v,int state)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 
kvm_xen_update_runstate_guest(struct kvm_vcpu * v,int state)93 void kvm_xen_update_runstate_guest(struct kvm_vcpu *v, int state)
94 {
95 	struct kvm_vcpu_xen *vx = &v->arch.xen;
96 	struct gfn_to_hva_cache *ghc = &vx->runstate_cache;
97 	struct kvm_memslots *slots = kvm_memslots(v->kvm);
98 	bool atomic = (state == RUNSTATE_runnable);
99 	uint64_t state_entry_time;
100 	int __user *user_state;
101 	uint64_t __user *user_times;
102 
103 	kvm_xen_update_runstate(v, state);
104 
105 	if (!vx->runstate_set)
106 		return;
107 
108 	if (unlikely(slots->generation != ghc->generation || kvm_is_error_hva(ghc->hva)) &&
109 	    kvm_gfn_to_hva_cache_init(v->kvm, ghc, ghc->gpa, ghc->len))
110 		return;
111 
112 	/* We made sure it fits in a single page */
113 	BUG_ON(!ghc->memslot);
114 
115 	if (atomic)
116 		pagefault_disable();
117 
118 	/*
119 	 * The only difference between 32-bit and 64-bit versions of the
120 	 * runstate struct us the alignment of uint64_t in 32-bit, which
121 	 * means that the 64-bit version has an additional 4 bytes of
122 	 * padding after the first field 'state'.
123 	 *
124 	 * So we use 'int __user *user_state' to point to the state field,
125 	 * and 'uint64_t __user *user_times' for runstate_entry_time. So
126 	 * the actual array of time[] in each state starts at user_times[1].
127 	 */
128 	BUILD_BUG_ON(offsetof(struct vcpu_runstate_info, state) != 0);
129 	BUILD_BUG_ON(offsetof(struct compat_vcpu_runstate_info, state) != 0);
130 	user_state = (int __user *)ghc->hva;
131 
132 	BUILD_BUG_ON(sizeof(struct compat_vcpu_runstate_info) != 0x2c);
133 
134 	user_times = (uint64_t __user *)(ghc->hva +
135 					 offsetof(struct compat_vcpu_runstate_info,
136 						  state_entry_time));
137 #ifdef CONFIG_X86_64
138 	BUILD_BUG_ON(offsetof(struct vcpu_runstate_info, state_entry_time) !=
139 		     offsetof(struct compat_vcpu_runstate_info, state_entry_time) + 4);
140 	BUILD_BUG_ON(offsetof(struct vcpu_runstate_info, time) !=
141 		     offsetof(struct compat_vcpu_runstate_info, time) + 4);
142 
143 	if (v->kvm->arch.xen.long_mode)
144 		user_times = (uint64_t __user *)(ghc->hva +
145 						 offsetof(struct vcpu_runstate_info,
146 							  state_entry_time));
147 #endif
148 	/*
149 	 * First write the updated state_entry_time at the appropriate
150 	 * location determined by 'offset'.
151 	 */
152 	state_entry_time = vx->runstate_entry_time;
153 	state_entry_time |= XEN_RUNSTATE_UPDATE;
154 
155 	BUILD_BUG_ON(sizeof(((struct vcpu_runstate_info *)0)->state_entry_time) !=
156 		     sizeof(state_entry_time));
157 	BUILD_BUG_ON(sizeof(((struct compat_vcpu_runstate_info *)0)->state_entry_time) !=
158 		     sizeof(state_entry_time));
159 
160 	if (__put_user(state_entry_time, user_times))
161 		goto out;
162 	smp_wmb();
163 
164 	/*
165 	 * Next, write the new runstate. This is in the *same* place
166 	 * for 32-bit and 64-bit guests, asserted here for paranoia.
167 	 */
168 	BUILD_BUG_ON(offsetof(struct vcpu_runstate_info, state) !=
169 		     offsetof(struct compat_vcpu_runstate_info, state));
170 	BUILD_BUG_ON(sizeof(((struct vcpu_runstate_info *)0)->state) !=
171 		     sizeof(vx->current_runstate));
172 	BUILD_BUG_ON(sizeof(((struct compat_vcpu_runstate_info *)0)->state) !=
173 		     sizeof(vx->current_runstate));
174 
175 	if (__put_user(vx->current_runstate, user_state))
176 		goto out;
177 
178 	/*
179 	 * Write the actual runstate times immediately after the
180 	 * runstate_entry_time.
181 	 */
182 	BUILD_BUG_ON(offsetof(struct vcpu_runstate_info, state_entry_time) !=
183 		     offsetof(struct vcpu_runstate_info, time) - sizeof(u64));
184 	BUILD_BUG_ON(offsetof(struct compat_vcpu_runstate_info, state_entry_time) !=
185 		     offsetof(struct compat_vcpu_runstate_info, time) - sizeof(u64));
186 	BUILD_BUG_ON(sizeof(((struct vcpu_runstate_info *)0)->time) !=
187 		     sizeof(((struct compat_vcpu_runstate_info *)0)->time));
188 	BUILD_BUG_ON(sizeof(((struct vcpu_runstate_info *)0)->time) !=
189 		     sizeof(vx->runstate_times));
190 
191 	if (__copy_to_user(user_times + 1, vx->runstate_times, sizeof(vx->runstate_times)))
192 		goto out;
193 	smp_wmb();
194 
195 	/*
196 	 * Finally, clear the XEN_RUNSTATE_UPDATE bit in the guest's
197 	 * runstate_entry_time field.
198 	 */
199 	state_entry_time &= ~XEN_RUNSTATE_UPDATE;
200 	__put_user(state_entry_time, user_times);
201 	smp_wmb();
202 
203  out:
204 	mark_page_dirty_in_slot(v->kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
205 
206 	if (atomic)
207 		pagefault_enable();
208 }
209 
__kvm_xen_has_interrupt(struct kvm_vcpu * v)210 int __kvm_xen_has_interrupt(struct kvm_vcpu *v)
211 {
212 	int err;
213 	u8 rc = 0;
214 
215 	/*
216 	 * If the global upcall vector (HVMIRQ_callback_vector) is set and
217 	 * the vCPU's evtchn_upcall_pending flag is set, the IRQ is pending.
218 	 */
219 	struct gfn_to_hva_cache *ghc = &v->arch.xen.vcpu_info_cache;
220 	struct kvm_memslots *slots = kvm_memslots(v->kvm);
221 	unsigned int offset = offsetof(struct vcpu_info, evtchn_upcall_pending);
222 
223 	/* No need for compat handling here */
224 	BUILD_BUG_ON(offsetof(struct vcpu_info, evtchn_upcall_pending) !=
225 		     offsetof(struct compat_vcpu_info, evtchn_upcall_pending));
226 	BUILD_BUG_ON(sizeof(rc) !=
227 		     sizeof(((struct vcpu_info *)0)->evtchn_upcall_pending));
228 	BUILD_BUG_ON(sizeof(rc) !=
229 		     sizeof(((struct compat_vcpu_info *)0)->evtchn_upcall_pending));
230 
231 	/*
232 	 * For efficiency, this mirrors the checks for using the valid
233 	 * cache in kvm_read_guest_offset_cached(), but just uses
234 	 * __get_user() instead. And falls back to the slow path.
235 	 */
236 	if (likely(slots->generation == ghc->generation &&
237 		   !kvm_is_error_hva(ghc->hva) && ghc->memslot)) {
238 		/* Fast path */
239 		pagefault_disable();
240 		err = __get_user(rc, (u8 __user *)ghc->hva + offset);
241 		pagefault_enable();
242 		if (!err)
243 			return rc;
244 	}
245 
246 	/* Slow path */
247 
248 	/*
249 	 * This function gets called from kvm_vcpu_block() after setting the
250 	 * task to TASK_INTERRUPTIBLE, to see if it needs to wake immediately
251 	 * from a HLT. So we really mustn't sleep. If the page ended up absent
252 	 * at that point, just return 1 in order to trigger an immediate wake,
253 	 * and we'll end up getting called again from a context where we *can*
254 	 * fault in the page and wait for it.
255 	 */
256 	if (in_atomic() || !task_is_running(current))
257 		return 1;
258 
259 	kvm_read_guest_offset_cached(v->kvm, ghc, &rc, offset,
260 				     sizeof(rc));
261 
262 	return rc;
263 }
264 
kvm_xen_hvm_set_attr(struct kvm * kvm,struct kvm_xen_hvm_attr * data)265 int kvm_xen_hvm_set_attr(struct kvm *kvm, struct kvm_xen_hvm_attr *data)
266 {
267 	int r = -ENOENT;
268 
269 	mutex_lock(&kvm->lock);
270 
271 	switch (data->type) {
272 	case KVM_XEN_ATTR_TYPE_LONG_MODE:
273 		if (!IS_ENABLED(CONFIG_64BIT) && data->u.long_mode) {
274 			r = -EINVAL;
275 		} else {
276 			kvm->arch.xen.long_mode = !!data->u.long_mode;
277 			r = 0;
278 		}
279 		break;
280 
281 	case KVM_XEN_ATTR_TYPE_SHARED_INFO:
282 		if (data->u.shared_info.gfn == GPA_INVALID) {
283 			kvm->arch.xen.shinfo_gfn = GPA_INVALID;
284 			r = 0;
285 			break;
286 		}
287 		r = kvm_xen_shared_info_init(kvm, data->u.shared_info.gfn);
288 		break;
289 
290 
291 	case KVM_XEN_ATTR_TYPE_UPCALL_VECTOR:
292 		if (data->u.vector && data->u.vector < 0x10)
293 			r = -EINVAL;
294 		else {
295 			kvm->arch.xen.upcall_vector = data->u.vector;
296 			r = 0;
297 		}
298 		break;
299 
300 	default:
301 		break;
302 	}
303 
304 	mutex_unlock(&kvm->lock);
305 	return r;
306 }
307 
kvm_xen_hvm_get_attr(struct kvm * kvm,struct kvm_xen_hvm_attr * data)308 int kvm_xen_hvm_get_attr(struct kvm *kvm, struct kvm_xen_hvm_attr *data)
309 {
310 	int r = -ENOENT;
311 
312 	mutex_lock(&kvm->lock);
313 
314 	switch (data->type) {
315 	case KVM_XEN_ATTR_TYPE_LONG_MODE:
316 		data->u.long_mode = kvm->arch.xen.long_mode;
317 		r = 0;
318 		break;
319 
320 	case KVM_XEN_ATTR_TYPE_SHARED_INFO:
321 		data->u.shared_info.gfn = kvm->arch.xen.shinfo_gfn;
322 		r = 0;
323 		break;
324 
325 	case KVM_XEN_ATTR_TYPE_UPCALL_VECTOR:
326 		data->u.vector = kvm->arch.xen.upcall_vector;
327 		r = 0;
328 		break;
329 
330 	default:
331 		break;
332 	}
333 
334 	mutex_unlock(&kvm->lock);
335 	return r;
336 }
337 
kvm_xen_vcpu_set_attr(struct kvm_vcpu * vcpu,struct kvm_xen_vcpu_attr * data)338 int kvm_xen_vcpu_set_attr(struct kvm_vcpu *vcpu, struct kvm_xen_vcpu_attr *data)
339 {
340 	int idx, r = -ENOENT;
341 
342 	mutex_lock(&vcpu->kvm->lock);
343 	idx = srcu_read_lock(&vcpu->kvm->srcu);
344 
345 	switch (data->type) {
346 	case KVM_XEN_VCPU_ATTR_TYPE_VCPU_INFO:
347 		/* No compat necessary here. */
348 		BUILD_BUG_ON(sizeof(struct vcpu_info) !=
349 			     sizeof(struct compat_vcpu_info));
350 		BUILD_BUG_ON(offsetof(struct vcpu_info, time) !=
351 			     offsetof(struct compat_vcpu_info, time));
352 
353 		if (data->u.gpa == GPA_INVALID) {
354 			vcpu->arch.xen.vcpu_info_set = false;
355 			r = 0;
356 			break;
357 		}
358 
359 		/* It must fit within a single page */
360 		if ((data->u.gpa & ~PAGE_MASK) + sizeof(struct vcpu_info) > PAGE_SIZE) {
361 			r = -EINVAL;
362 			break;
363 		}
364 
365 		r = kvm_gfn_to_hva_cache_init(vcpu->kvm,
366 					      &vcpu->arch.xen.vcpu_info_cache,
367 					      data->u.gpa,
368 					      sizeof(struct vcpu_info));
369 		if (!r) {
370 			vcpu->arch.xen.vcpu_info_set = true;
371 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
372 		}
373 		break;
374 
375 	case KVM_XEN_VCPU_ATTR_TYPE_VCPU_TIME_INFO:
376 		if (data->u.gpa == GPA_INVALID) {
377 			vcpu->arch.xen.vcpu_time_info_set = false;
378 			r = 0;
379 			break;
380 		}
381 
382 		/* It must fit within a single page */
383 		if ((data->u.gpa & ~PAGE_MASK) + sizeof(struct pvclock_vcpu_time_info) > PAGE_SIZE) {
384 			r = -EINVAL;
385 			break;
386 		}
387 
388 		r = kvm_gfn_to_hva_cache_init(vcpu->kvm,
389 					      &vcpu->arch.xen.vcpu_time_info_cache,
390 					      data->u.gpa,
391 					      sizeof(struct pvclock_vcpu_time_info));
392 		if (!r) {
393 			vcpu->arch.xen.vcpu_time_info_set = true;
394 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
395 		}
396 		break;
397 
398 	case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_ADDR:
399 		if (!sched_info_on()) {
400 			r = -EOPNOTSUPP;
401 			break;
402 		}
403 		if (data->u.gpa == GPA_INVALID) {
404 			vcpu->arch.xen.runstate_set = false;
405 			r = 0;
406 			break;
407 		}
408 
409 		/* It must fit within a single page */
410 		if ((data->u.gpa & ~PAGE_MASK) + sizeof(struct vcpu_runstate_info) > PAGE_SIZE) {
411 			r = -EINVAL;
412 			break;
413 		}
414 
415 		r = kvm_gfn_to_hva_cache_init(vcpu->kvm,
416 					      &vcpu->arch.xen.runstate_cache,
417 					      data->u.gpa,
418 					      sizeof(struct vcpu_runstate_info));
419 		if (!r) {
420 			vcpu->arch.xen.runstate_set = true;
421 		}
422 		break;
423 
424 	case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_CURRENT:
425 		if (!sched_info_on()) {
426 			r = -EOPNOTSUPP;
427 			break;
428 		}
429 		if (data->u.runstate.state > RUNSTATE_offline) {
430 			r = -EINVAL;
431 			break;
432 		}
433 
434 		kvm_xen_update_runstate(vcpu, data->u.runstate.state);
435 		r = 0;
436 		break;
437 
438 	case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_DATA:
439 		if (!sched_info_on()) {
440 			r = -EOPNOTSUPP;
441 			break;
442 		}
443 		if (data->u.runstate.state > RUNSTATE_offline) {
444 			r = -EINVAL;
445 			break;
446 		}
447 		if (data->u.runstate.state_entry_time !=
448 		    (data->u.runstate.time_running +
449 		     data->u.runstate.time_runnable +
450 		     data->u.runstate.time_blocked +
451 		     data->u.runstate.time_offline)) {
452 			r = -EINVAL;
453 			break;
454 		}
455 		if (get_kvmclock_ns(vcpu->kvm) <
456 		    data->u.runstate.state_entry_time) {
457 			r = -EINVAL;
458 			break;
459 		}
460 
461 		vcpu->arch.xen.current_runstate = data->u.runstate.state;
462 		vcpu->arch.xen.runstate_entry_time =
463 			data->u.runstate.state_entry_time;
464 		vcpu->arch.xen.runstate_times[RUNSTATE_running] =
465 			data->u.runstate.time_running;
466 		vcpu->arch.xen.runstate_times[RUNSTATE_runnable] =
467 			data->u.runstate.time_runnable;
468 		vcpu->arch.xen.runstate_times[RUNSTATE_blocked] =
469 			data->u.runstate.time_blocked;
470 		vcpu->arch.xen.runstate_times[RUNSTATE_offline] =
471 			data->u.runstate.time_offline;
472 		vcpu->arch.xen.last_steal = current->sched_info.run_delay;
473 		r = 0;
474 		break;
475 
476 	case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_ADJUST:
477 		if (!sched_info_on()) {
478 			r = -EOPNOTSUPP;
479 			break;
480 		}
481 		if (data->u.runstate.state > RUNSTATE_offline &&
482 		    data->u.runstate.state != (u64)-1) {
483 			r = -EINVAL;
484 			break;
485 		}
486 		/* The adjustment must add up */
487 		if (data->u.runstate.state_entry_time !=
488 		    (data->u.runstate.time_running +
489 		     data->u.runstate.time_runnable +
490 		     data->u.runstate.time_blocked +
491 		     data->u.runstate.time_offline)) {
492 			r = -EINVAL;
493 			break;
494 		}
495 
496 		if (get_kvmclock_ns(vcpu->kvm) <
497 		    (vcpu->arch.xen.runstate_entry_time +
498 		     data->u.runstate.state_entry_time)) {
499 			r = -EINVAL;
500 			break;
501 		}
502 
503 		vcpu->arch.xen.runstate_entry_time +=
504 			data->u.runstate.state_entry_time;
505 		vcpu->arch.xen.runstate_times[RUNSTATE_running] +=
506 			data->u.runstate.time_running;
507 		vcpu->arch.xen.runstate_times[RUNSTATE_runnable] +=
508 			data->u.runstate.time_runnable;
509 		vcpu->arch.xen.runstate_times[RUNSTATE_blocked] +=
510 			data->u.runstate.time_blocked;
511 		vcpu->arch.xen.runstate_times[RUNSTATE_offline] +=
512 			data->u.runstate.time_offline;
513 
514 		if (data->u.runstate.state <= RUNSTATE_offline)
515 			kvm_xen_update_runstate(vcpu, data->u.runstate.state);
516 		r = 0;
517 		break;
518 
519 	default:
520 		break;
521 	}
522 
523 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
524 	mutex_unlock(&vcpu->kvm->lock);
525 	return r;
526 }
527 
kvm_xen_vcpu_get_attr(struct kvm_vcpu * vcpu,struct kvm_xen_vcpu_attr * data)528 int kvm_xen_vcpu_get_attr(struct kvm_vcpu *vcpu, struct kvm_xen_vcpu_attr *data)
529 {
530 	int r = -ENOENT;
531 
532 	mutex_lock(&vcpu->kvm->lock);
533 
534 	switch (data->type) {
535 	case KVM_XEN_VCPU_ATTR_TYPE_VCPU_INFO:
536 		if (vcpu->arch.xen.vcpu_info_set)
537 			data->u.gpa = vcpu->arch.xen.vcpu_info_cache.gpa;
538 		else
539 			data->u.gpa = GPA_INVALID;
540 		r = 0;
541 		break;
542 
543 	case KVM_XEN_VCPU_ATTR_TYPE_VCPU_TIME_INFO:
544 		if (vcpu->arch.xen.vcpu_time_info_set)
545 			data->u.gpa = vcpu->arch.xen.vcpu_time_info_cache.gpa;
546 		else
547 			data->u.gpa = GPA_INVALID;
548 		r = 0;
549 		break;
550 
551 	case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_ADDR:
552 		if (!sched_info_on()) {
553 			r = -EOPNOTSUPP;
554 			break;
555 		}
556 		if (vcpu->arch.xen.runstate_set) {
557 			data->u.gpa = vcpu->arch.xen.runstate_cache.gpa;
558 			r = 0;
559 		}
560 		break;
561 
562 	case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_CURRENT:
563 		if (!sched_info_on()) {
564 			r = -EOPNOTSUPP;
565 			break;
566 		}
567 		data->u.runstate.state = vcpu->arch.xen.current_runstate;
568 		r = 0;
569 		break;
570 
571 	case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_DATA:
572 		if (!sched_info_on()) {
573 			r = -EOPNOTSUPP;
574 			break;
575 		}
576 		data->u.runstate.state = vcpu->arch.xen.current_runstate;
577 		data->u.runstate.state_entry_time =
578 			vcpu->arch.xen.runstate_entry_time;
579 		data->u.runstate.time_running =
580 			vcpu->arch.xen.runstate_times[RUNSTATE_running];
581 		data->u.runstate.time_runnable =
582 			vcpu->arch.xen.runstate_times[RUNSTATE_runnable];
583 		data->u.runstate.time_blocked =
584 			vcpu->arch.xen.runstate_times[RUNSTATE_blocked];
585 		data->u.runstate.time_offline =
586 			vcpu->arch.xen.runstate_times[RUNSTATE_offline];
587 		r = 0;
588 		break;
589 
590 	case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_ADJUST:
591 		r = -EINVAL;
592 		break;
593 
594 	default:
595 		break;
596 	}
597 
598 	mutex_unlock(&vcpu->kvm->lock);
599 	return r;
600 }
601 
kvm_xen_write_hypercall_page(struct kvm_vcpu * vcpu,u64 data)602 int kvm_xen_write_hypercall_page(struct kvm_vcpu *vcpu, u64 data)
603 {
604 	struct kvm *kvm = vcpu->kvm;
605 	u32 page_num = data & ~PAGE_MASK;
606 	u64 page_addr = data & PAGE_MASK;
607 	bool lm = is_long_mode(vcpu);
608 
609 	/* Latch long_mode for shared_info pages etc. */
610 	vcpu->kvm->arch.xen.long_mode = lm;
611 
612 	/*
613 	 * If Xen hypercall intercept is enabled, fill the hypercall
614 	 * page with VMCALL/VMMCALL instructions since that's what
615 	 * we catch. Else the VMM has provided the hypercall pages
616 	 * with instructions of its own choosing, so use those.
617 	 */
618 	if (kvm_xen_hypercall_enabled(kvm)) {
619 		u8 instructions[32];
620 		int i;
621 
622 		if (page_num)
623 			return 1;
624 
625 		/* mov imm32, %eax */
626 		instructions[0] = 0xb8;
627 
628 		/* vmcall / vmmcall */
629 		kvm_x86_ops.patch_hypercall(vcpu, instructions + 5);
630 
631 		/* ret */
632 		instructions[8] = 0xc3;
633 
634 		/* int3 to pad */
635 		memset(instructions + 9, 0xcc, sizeof(instructions) - 9);
636 
637 		for (i = 0; i < PAGE_SIZE / sizeof(instructions); i++) {
638 			*(u32 *)&instructions[1] = i;
639 			if (kvm_vcpu_write_guest(vcpu,
640 						 page_addr + (i * sizeof(instructions)),
641 						 instructions, sizeof(instructions)))
642 				return 1;
643 		}
644 	} else {
645 		/*
646 		 * Note, truncation is a non-issue as 'lm' is guaranteed to be
647 		 * false for a 32-bit kernel, i.e. when hva_t is only 4 bytes.
648 		 */
649 		hva_t blob_addr = lm ? kvm->arch.xen_hvm_config.blob_addr_64
650 				     : kvm->arch.xen_hvm_config.blob_addr_32;
651 		u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
652 				  : kvm->arch.xen_hvm_config.blob_size_32;
653 		u8 *page;
654 
655 		if (page_num >= blob_size)
656 			return 1;
657 
658 		blob_addr += page_num * PAGE_SIZE;
659 
660 		page = memdup_user((u8 __user *)blob_addr, PAGE_SIZE);
661 		if (IS_ERR(page))
662 			return PTR_ERR(page);
663 
664 		if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE)) {
665 			kfree(page);
666 			return 1;
667 		}
668 	}
669 	return 0;
670 }
671 
kvm_xen_hvm_config(struct kvm * kvm,struct kvm_xen_hvm_config * xhc)672 int kvm_xen_hvm_config(struct kvm *kvm, struct kvm_xen_hvm_config *xhc)
673 {
674 	if (xhc->flags & ~KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL)
675 		return -EINVAL;
676 
677 	/*
678 	 * With hypercall interception the kernel generates its own
679 	 * hypercall page so it must not be provided.
680 	 */
681 	if ((xhc->flags & KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL) &&
682 	    (xhc->blob_addr_32 || xhc->blob_addr_64 ||
683 	     xhc->blob_size_32 || xhc->blob_size_64))
684 		return -EINVAL;
685 
686 	mutex_lock(&kvm->lock);
687 
688 	if (xhc->msr && !kvm->arch.xen_hvm_config.msr)
689 		static_branch_inc(&kvm_xen_enabled.key);
690 	else if (!xhc->msr && kvm->arch.xen_hvm_config.msr)
691 		static_branch_slow_dec_deferred(&kvm_xen_enabled);
692 
693 	memcpy(&kvm->arch.xen_hvm_config, xhc, sizeof(*xhc));
694 
695 	mutex_unlock(&kvm->lock);
696 	return 0;
697 }
698 
kvm_xen_init_vm(struct kvm * kvm)699 void kvm_xen_init_vm(struct kvm *kvm)
700 {
701 	kvm->arch.xen.shinfo_gfn = GPA_INVALID;
702 }
703 
kvm_xen_destroy_vm(struct kvm * kvm)704 void kvm_xen_destroy_vm(struct kvm *kvm)
705 {
706 	if (kvm->arch.xen_hvm_config.msr)
707 		static_branch_slow_dec_deferred(&kvm_xen_enabled);
708 }
709 
kvm_xen_hypercall_set_result(struct kvm_vcpu * vcpu,u64 result)710 static int kvm_xen_hypercall_set_result(struct kvm_vcpu *vcpu, u64 result)
711 {
712 	kvm_rax_write(vcpu, result);
713 	return kvm_skip_emulated_instruction(vcpu);
714 }
715 
kvm_xen_hypercall_complete_userspace(struct kvm_vcpu * vcpu)716 static int kvm_xen_hypercall_complete_userspace(struct kvm_vcpu *vcpu)
717 {
718 	struct kvm_run *run = vcpu->run;
719 
720 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.xen.hypercall_rip)))
721 		return 1;
722 
723 	return kvm_xen_hypercall_set_result(vcpu, run->xen.u.hcall.result);
724 }
725 
kvm_xen_hypercall(struct kvm_vcpu * vcpu)726 int kvm_xen_hypercall(struct kvm_vcpu *vcpu)
727 {
728 	bool longmode;
729 	u64 input, params[6];
730 
731 	input = (u64)kvm_register_read(vcpu, VCPU_REGS_RAX);
732 
733 	/* Hyper-V hypercalls get bit 31 set in EAX */
734 	if ((input & 0x80000000) &&
735 	    kvm_hv_hypercall_enabled(vcpu))
736 		return kvm_hv_hypercall(vcpu);
737 
738 	longmode = is_64_bit_hypercall(vcpu);
739 	if (!longmode) {
740 		params[0] = (u32)kvm_rbx_read(vcpu);
741 		params[1] = (u32)kvm_rcx_read(vcpu);
742 		params[2] = (u32)kvm_rdx_read(vcpu);
743 		params[3] = (u32)kvm_rsi_read(vcpu);
744 		params[4] = (u32)kvm_rdi_read(vcpu);
745 		params[5] = (u32)kvm_rbp_read(vcpu);
746 	}
747 #ifdef CONFIG_X86_64
748 	else {
749 		params[0] = (u64)kvm_rdi_read(vcpu);
750 		params[1] = (u64)kvm_rsi_read(vcpu);
751 		params[2] = (u64)kvm_rdx_read(vcpu);
752 		params[3] = (u64)kvm_r10_read(vcpu);
753 		params[4] = (u64)kvm_r8_read(vcpu);
754 		params[5] = (u64)kvm_r9_read(vcpu);
755 	}
756 #endif
757 	trace_kvm_xen_hypercall(input, params[0], params[1], params[2],
758 				params[3], params[4], params[5]);
759 
760 	vcpu->run->exit_reason = KVM_EXIT_XEN;
761 	vcpu->run->xen.type = KVM_EXIT_XEN_HCALL;
762 	vcpu->run->xen.u.hcall.longmode = longmode;
763 	vcpu->run->xen.u.hcall.cpl = kvm_x86_ops.get_cpl(vcpu);
764 	vcpu->run->xen.u.hcall.input = input;
765 	vcpu->run->xen.u.hcall.params[0] = params[0];
766 	vcpu->run->xen.u.hcall.params[1] = params[1];
767 	vcpu->run->xen.u.hcall.params[2] = params[2];
768 	vcpu->run->xen.u.hcall.params[3] = params[3];
769 	vcpu->run->xen.u.hcall.params[4] = params[4];
770 	vcpu->run->xen.u.hcall.params[5] = params[5];
771 	vcpu->arch.xen.hypercall_rip = kvm_get_linear_rip(vcpu);
772 	vcpu->arch.complete_userspace_io =
773 		kvm_xen_hypercall_complete_userspace;
774 
775 	return 0;
776 }
777