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;
29 int idx = srcu_read_lock(&kvm->srcu);
30
31 ret = kvm_gfn_to_hva_cache_init(kvm, &kvm->arch.xen.shinfo_cache,
32 gpa, PAGE_SIZE);
33 if (ret)
34 goto out;
35
36 kvm->arch.xen.shinfo_set = true;
37
38 /* Paranoia checks on the 32-bit struct layout */
39 BUILD_BUG_ON(offsetof(struct compat_shared_info, wc) != 0x900);
40 BUILD_BUG_ON(offsetof(struct compat_shared_info, arch.wc_sec_hi) != 0x924);
41 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
42
43 /* 32-bit location by default */
44 wc_ofs = offsetof(struct compat_shared_info, wc);
45 sec_hi_ofs = offsetof(struct compat_shared_info, arch.wc_sec_hi);
46
47 #ifdef CONFIG_X86_64
48 /* Paranoia checks on the 64-bit struct layout */
49 BUILD_BUG_ON(offsetof(struct shared_info, wc) != 0xc00);
50 BUILD_BUG_ON(offsetof(struct shared_info, wc_sec_hi) != 0xc0c);
51
52 if (kvm->arch.xen.long_mode) {
53 wc_ofs = offsetof(struct shared_info, wc);
54 sec_hi_ofs = offsetof(struct shared_info, wc_sec_hi);
55 }
56 #endif
57
58 kvm_write_wall_clock(kvm, gpa + wc_ofs, sec_hi_ofs - wc_ofs);
59 kvm_make_all_cpus_request(kvm, KVM_REQ_MASTERCLOCK_UPDATE);
60
61 out:
62 srcu_read_unlock(&kvm->srcu, idx);
63 return ret;
64 }
65
66 static void kvm_xen_update_runstate(struct kvm_vcpu *v, int state)
67 {
68 struct kvm_vcpu_xen *vx = &v->arch.xen;
69 u64 now = get_kvmclock_ns(v->kvm);
70 u64 delta_ns = now - vx->runstate_entry_time;
71 u64 run_delay = current->sched_info.run_delay;
72
73 if (unlikely(!vx->runstate_entry_time))
74 vx->current_runstate = RUNSTATE_offline;
75
76 /*
77 * Time waiting for the scheduler isn't "stolen" if the
78 * vCPU wasn't running anyway.
79 */
80 if (vx->current_runstate == RUNSTATE_running) {
81 u64 steal_ns = run_delay - vx->last_steal;
82
83 delta_ns -= steal_ns;
84
85 vx->runstate_times[RUNSTATE_runnable] += steal_ns;
86 }
87 vx->last_steal = run_delay;
88
89 vx->runstate_times[vx->current_runstate] += delta_ns;
90 vx->current_runstate = state;
91 vx->runstate_entry_time = now;
92 }
93
94 void kvm_xen_update_runstate_guest(struct kvm_vcpu *v, int state)
95 {
96 struct kvm_vcpu_xen *vx = &v->arch.xen;
97 uint64_t state_entry_time;
98 unsigned int offset;
99
100 kvm_xen_update_runstate(v, state);
101
102 if (!vx->runstate_set)
103 return;
104
105 BUILD_BUG_ON(sizeof(struct compat_vcpu_runstate_info) != 0x2c);
106
107 offset = offsetof(struct compat_vcpu_runstate_info, state_entry_time);
108 #ifdef CONFIG_X86_64
109 /*
110 * The only difference is alignment of uint64_t in 32-bit.
111 * So the first field 'state' is accessed directly using
112 * offsetof() (where its offset happens to be zero), while the
113 * remaining fields which are all uint64_t, start at 'offset'
114 * which we tweak here by adding 4.
115 */
116 BUILD_BUG_ON(offsetof(struct vcpu_runstate_info, state_entry_time) !=
117 offsetof(struct compat_vcpu_runstate_info, state_entry_time) + 4);
118 BUILD_BUG_ON(offsetof(struct vcpu_runstate_info, time) !=
119 offsetof(struct compat_vcpu_runstate_info, time) + 4);
120
121 if (v->kvm->arch.xen.long_mode)
122 offset = offsetof(struct vcpu_runstate_info, state_entry_time);
123 #endif
124 /*
125 * First write the updated state_entry_time at the appropriate
126 * location determined by 'offset'.
127 */
128 state_entry_time = vx->runstate_entry_time;
129 state_entry_time |= XEN_RUNSTATE_UPDATE;
130
131 BUILD_BUG_ON(sizeof(((struct vcpu_runstate_info *)0)->state_entry_time) !=
132 sizeof(state_entry_time));
133 BUILD_BUG_ON(sizeof(((struct compat_vcpu_runstate_info *)0)->state_entry_time) !=
134 sizeof(state_entry_time));
135
136 if (kvm_write_guest_offset_cached(v->kvm, &v->arch.xen.runstate_cache,
137 &state_entry_time, offset,
138 sizeof(state_entry_time)))
139 return;
140 smp_wmb();
141
142 /*
143 * Next, write the new runstate. This is in the *same* place
144 * for 32-bit and 64-bit guests, asserted here for paranoia.
145 */
146 BUILD_BUG_ON(offsetof(struct vcpu_runstate_info, state) !=
147 offsetof(struct compat_vcpu_runstate_info, state));
148 BUILD_BUG_ON(sizeof(((struct vcpu_runstate_info *)0)->state) !=
149 sizeof(vx->current_runstate));
150 BUILD_BUG_ON(sizeof(((struct compat_vcpu_runstate_info *)0)->state) !=
151 sizeof(vx->current_runstate));
152
153 if (kvm_write_guest_offset_cached(v->kvm, &v->arch.xen.runstate_cache,
154 &vx->current_runstate,
155 offsetof(struct vcpu_runstate_info, state),
156 sizeof(vx->current_runstate)))
157 return;
158
159 /*
160 * Write the actual runstate times immediately after the
161 * runstate_entry_time.
162 */
163 BUILD_BUG_ON(offsetof(struct vcpu_runstate_info, state_entry_time) !=
164 offsetof(struct vcpu_runstate_info, time) - sizeof(u64));
165 BUILD_BUG_ON(offsetof(struct compat_vcpu_runstate_info, state_entry_time) !=
166 offsetof(struct compat_vcpu_runstate_info, time) - sizeof(u64));
167 BUILD_BUG_ON(sizeof(((struct vcpu_runstate_info *)0)->time) !=
168 sizeof(((struct compat_vcpu_runstate_info *)0)->time));
169 BUILD_BUG_ON(sizeof(((struct vcpu_runstate_info *)0)->time) !=
170 sizeof(vx->runstate_times));
171
172 if (kvm_write_guest_offset_cached(v->kvm, &v->arch.xen.runstate_cache,
173 &vx->runstate_times[0],
174 offset + sizeof(u64),
175 sizeof(vx->runstate_times)))
176 return;
177
178 smp_wmb();
179
180 /*
181 * Finally, clear the XEN_RUNSTATE_UPDATE bit in the guest's
182 * runstate_entry_time field.
183 */
184
185 state_entry_time &= ~XEN_RUNSTATE_UPDATE;
186 if (kvm_write_guest_offset_cached(v->kvm, &v->arch.xen.runstate_cache,
187 &state_entry_time, offset,
188 sizeof(state_entry_time)))
189 return;
190 }
191
192 int __kvm_xen_has_interrupt(struct kvm_vcpu *v)
193 {
194 int err;
195 u8 rc = 0;
196
197 /*
198 * If the global upcall vector (HVMIRQ_callback_vector) is set and
199 * the vCPU's evtchn_upcall_pending flag is set, the IRQ is pending.
200 */
201 struct gfn_to_hva_cache *ghc = &v->arch.xen.vcpu_info_cache;
202 struct kvm_memslots *slots = kvm_memslots(v->kvm);
203 unsigned int offset = offsetof(struct vcpu_info, evtchn_upcall_pending);
204
205 /* No need for compat handling here */
206 BUILD_BUG_ON(offsetof(struct vcpu_info, evtchn_upcall_pending) !=
207 offsetof(struct compat_vcpu_info, evtchn_upcall_pending));
208 BUILD_BUG_ON(sizeof(rc) !=
209 sizeof(((struct vcpu_info *)0)->evtchn_upcall_pending));
210 BUILD_BUG_ON(sizeof(rc) !=
211 sizeof(((struct compat_vcpu_info *)0)->evtchn_upcall_pending));
212
213 /*
214 * For efficiency, this mirrors the checks for using the valid
215 * cache in kvm_read_guest_offset_cached(), but just uses
216 * __get_user() instead. And falls back to the slow path.
217 */
218 if (likely(slots->generation == ghc->generation &&
219 !kvm_is_error_hva(ghc->hva) && ghc->memslot)) {
220 /* Fast path */
221 pagefault_disable();
222 err = __get_user(rc, (u8 __user *)ghc->hva + offset);
223 pagefault_enable();
224 if (!err)
225 return rc;
226 }
227
228 /* Slow path */
229
230 /*
231 * This function gets called from kvm_vcpu_block() after setting the
232 * task to TASK_INTERRUPTIBLE, to see if it needs to wake immediately
233 * from a HLT. So we really mustn't sleep. If the page ended up absent
234 * at that point, just return 1 in order to trigger an immediate wake,
235 * and we'll end up getting called again from a context where we *can*
236 * fault in the page and wait for it.
237 */
238 if (in_atomic() || !task_is_running(current))
239 return 1;
240
241 kvm_read_guest_offset_cached(v->kvm, ghc, &rc, offset,
242 sizeof(rc));
243
244 return rc;
245 }
246
247 int kvm_xen_hvm_set_attr(struct kvm *kvm, struct kvm_xen_hvm_attr *data)
248 {
249 int r = -ENOENT;
250
251 mutex_lock(&kvm->lock);
252
253 switch (data->type) {
254 case KVM_XEN_ATTR_TYPE_LONG_MODE:
255 if (!IS_ENABLED(CONFIG_64BIT) && data->u.long_mode) {
256 r = -EINVAL;
257 } else {
258 kvm->arch.xen.long_mode = !!data->u.long_mode;
259 r = 0;
260 }
261 break;
262
263 case KVM_XEN_ATTR_TYPE_SHARED_INFO:
264 if (data->u.shared_info.gfn == GPA_INVALID) {
265 kvm->arch.xen.shinfo_set = false;
266 r = 0;
267 break;
268 }
269 r = kvm_xen_shared_info_init(kvm, data->u.shared_info.gfn);
270 break;
271
272
273 case KVM_XEN_ATTR_TYPE_UPCALL_VECTOR:
274 if (data->u.vector && data->u.vector < 0x10)
275 r = -EINVAL;
276 else {
277 kvm->arch.xen.upcall_vector = data->u.vector;
278 r = 0;
279 }
280 break;
281
282 default:
283 break;
284 }
285
286 mutex_unlock(&kvm->lock);
287 return r;
288 }
289
290 int kvm_xen_hvm_get_attr(struct kvm *kvm, struct kvm_xen_hvm_attr *data)
291 {
292 int r = -ENOENT;
293
294 mutex_lock(&kvm->lock);
295
296 switch (data->type) {
297 case KVM_XEN_ATTR_TYPE_LONG_MODE:
298 data->u.long_mode = kvm->arch.xen.long_mode;
299 r = 0;
300 break;
301
302 case KVM_XEN_ATTR_TYPE_SHARED_INFO:
303 if (kvm->arch.xen.shinfo_set)
304 data->u.shared_info.gfn = gpa_to_gfn(kvm->arch.xen.shinfo_cache.gpa);
305 else
306 data->u.shared_info.gfn = GPA_INVALID;
307 r = 0;
308 break;
309
310 case KVM_XEN_ATTR_TYPE_UPCALL_VECTOR:
311 data->u.vector = kvm->arch.xen.upcall_vector;
312 r = 0;
313 break;
314
315 default:
316 break;
317 }
318
319 mutex_unlock(&kvm->lock);
320 return r;
321 }
322
323 int kvm_xen_vcpu_set_attr(struct kvm_vcpu *vcpu, struct kvm_xen_vcpu_attr *data)
324 {
325 int idx, r = -ENOENT;
326
327 mutex_lock(&vcpu->kvm->lock);
328 idx = srcu_read_lock(&vcpu->kvm->srcu);
329
330 switch (data->type) {
331 case KVM_XEN_VCPU_ATTR_TYPE_VCPU_INFO:
332 /* No compat necessary here. */
333 BUILD_BUG_ON(sizeof(struct vcpu_info) !=
334 sizeof(struct compat_vcpu_info));
335 BUILD_BUG_ON(offsetof(struct vcpu_info, time) !=
336 offsetof(struct compat_vcpu_info, time));
337
338 if (data->u.gpa == GPA_INVALID) {
339 vcpu->arch.xen.vcpu_info_set = false;
340 r = 0;
341 break;
342 }
343
344 r = kvm_gfn_to_hva_cache_init(vcpu->kvm,
345 &vcpu->arch.xen.vcpu_info_cache,
346 data->u.gpa,
347 sizeof(struct vcpu_info));
348 if (!r) {
349 vcpu->arch.xen.vcpu_info_set = true;
350 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
351 }
352 break;
353
354 case KVM_XEN_VCPU_ATTR_TYPE_VCPU_TIME_INFO:
355 if (data->u.gpa == GPA_INVALID) {
356 vcpu->arch.xen.vcpu_time_info_set = false;
357 r = 0;
358 break;
359 }
360
361 r = kvm_gfn_to_hva_cache_init(vcpu->kvm,
362 &vcpu->arch.xen.vcpu_time_info_cache,
363 data->u.gpa,
364 sizeof(struct pvclock_vcpu_time_info));
365 if (!r) {
366 vcpu->arch.xen.vcpu_time_info_set = true;
367 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
368 }
369 break;
370
371 case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_ADDR:
372 if (!sched_info_on()) {
373 r = -EOPNOTSUPP;
374 break;
375 }
376 if (data->u.gpa == GPA_INVALID) {
377 vcpu->arch.xen.runstate_set = false;
378 r = 0;
379 break;
380 }
381
382 r = kvm_gfn_to_hva_cache_init(vcpu->kvm,
383 &vcpu->arch.xen.runstate_cache,
384 data->u.gpa,
385 sizeof(struct vcpu_runstate_info));
386 if (!r) {
387 vcpu->arch.xen.runstate_set = true;
388 }
389 break;
390
391 case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_CURRENT:
392 if (!sched_info_on()) {
393 r = -EOPNOTSUPP;
394 break;
395 }
396 if (data->u.runstate.state > RUNSTATE_offline) {
397 r = -EINVAL;
398 break;
399 }
400
401 kvm_xen_update_runstate(vcpu, data->u.runstate.state);
402 r = 0;
403 break;
404
405 case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_DATA:
406 if (!sched_info_on()) {
407 r = -EOPNOTSUPP;
408 break;
409 }
410 if (data->u.runstate.state > RUNSTATE_offline) {
411 r = -EINVAL;
412 break;
413 }
414 if (data->u.runstate.state_entry_time !=
415 (data->u.runstate.time_running +
416 data->u.runstate.time_runnable +
417 data->u.runstate.time_blocked +
418 data->u.runstate.time_offline)) {
419 r = -EINVAL;
420 break;
421 }
422 if (get_kvmclock_ns(vcpu->kvm) <
423 data->u.runstate.state_entry_time) {
424 r = -EINVAL;
425 break;
426 }
427
428 vcpu->arch.xen.current_runstate = data->u.runstate.state;
429 vcpu->arch.xen.runstate_entry_time =
430 data->u.runstate.state_entry_time;
431 vcpu->arch.xen.runstate_times[RUNSTATE_running] =
432 data->u.runstate.time_running;
433 vcpu->arch.xen.runstate_times[RUNSTATE_runnable] =
434 data->u.runstate.time_runnable;
435 vcpu->arch.xen.runstate_times[RUNSTATE_blocked] =
436 data->u.runstate.time_blocked;
437 vcpu->arch.xen.runstate_times[RUNSTATE_offline] =
438 data->u.runstate.time_offline;
439 vcpu->arch.xen.last_steal = current->sched_info.run_delay;
440 r = 0;
441 break;
442
443 case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_ADJUST:
444 if (!sched_info_on()) {
445 r = -EOPNOTSUPP;
446 break;
447 }
448 if (data->u.runstate.state > RUNSTATE_offline &&
449 data->u.runstate.state != (u64)-1) {
450 r = -EINVAL;
451 break;
452 }
453 /* The adjustment must add up */
454 if (data->u.runstate.state_entry_time !=
455 (data->u.runstate.time_running +
456 data->u.runstate.time_runnable +
457 data->u.runstate.time_blocked +
458 data->u.runstate.time_offline)) {
459 r = -EINVAL;
460 break;
461 }
462
463 if (get_kvmclock_ns(vcpu->kvm) <
464 (vcpu->arch.xen.runstate_entry_time +
465 data->u.runstate.state_entry_time)) {
466 r = -EINVAL;
467 break;
468 }
469
470 vcpu->arch.xen.runstate_entry_time +=
471 data->u.runstate.state_entry_time;
472 vcpu->arch.xen.runstate_times[RUNSTATE_running] +=
473 data->u.runstate.time_running;
474 vcpu->arch.xen.runstate_times[RUNSTATE_runnable] +=
475 data->u.runstate.time_runnable;
476 vcpu->arch.xen.runstate_times[RUNSTATE_blocked] +=
477 data->u.runstate.time_blocked;
478 vcpu->arch.xen.runstate_times[RUNSTATE_offline] +=
479 data->u.runstate.time_offline;
480
481 if (data->u.runstate.state <= RUNSTATE_offline)
482 kvm_xen_update_runstate(vcpu, data->u.runstate.state);
483 r = 0;
484 break;
485
486 default:
487 break;
488 }
489
490 srcu_read_unlock(&vcpu->kvm->srcu, idx);
491 mutex_unlock(&vcpu->kvm->lock);
492 return r;
493 }
494
495 int kvm_xen_vcpu_get_attr(struct kvm_vcpu *vcpu, struct kvm_xen_vcpu_attr *data)
496 {
497 int r = -ENOENT;
498
499 mutex_lock(&vcpu->kvm->lock);
500
501 switch (data->type) {
502 case KVM_XEN_VCPU_ATTR_TYPE_VCPU_INFO:
503 if (vcpu->arch.xen.vcpu_info_set)
504 data->u.gpa = vcpu->arch.xen.vcpu_info_cache.gpa;
505 else
506 data->u.gpa = GPA_INVALID;
507 r = 0;
508 break;
509
510 case KVM_XEN_VCPU_ATTR_TYPE_VCPU_TIME_INFO:
511 if (vcpu->arch.xen.vcpu_time_info_set)
512 data->u.gpa = vcpu->arch.xen.vcpu_time_info_cache.gpa;
513 else
514 data->u.gpa = GPA_INVALID;
515 r = 0;
516 break;
517
518 case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_ADDR:
519 if (!sched_info_on()) {
520 r = -EOPNOTSUPP;
521 break;
522 }
523 if (vcpu->arch.xen.runstate_set) {
524 data->u.gpa = vcpu->arch.xen.runstate_cache.gpa;
525 r = 0;
526 }
527 break;
528
529 case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_CURRENT:
530 if (!sched_info_on()) {
531 r = -EOPNOTSUPP;
532 break;
533 }
534 data->u.runstate.state = vcpu->arch.xen.current_runstate;
535 r = 0;
536 break;
537
538 case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_DATA:
539 if (!sched_info_on()) {
540 r = -EOPNOTSUPP;
541 break;
542 }
543 data->u.runstate.state = vcpu->arch.xen.current_runstate;
544 data->u.runstate.state_entry_time =
545 vcpu->arch.xen.runstate_entry_time;
546 data->u.runstate.time_running =
547 vcpu->arch.xen.runstate_times[RUNSTATE_running];
548 data->u.runstate.time_runnable =
549 vcpu->arch.xen.runstate_times[RUNSTATE_runnable];
550 data->u.runstate.time_blocked =
551 vcpu->arch.xen.runstate_times[RUNSTATE_blocked];
552 data->u.runstate.time_offline =
553 vcpu->arch.xen.runstate_times[RUNSTATE_offline];
554 r = 0;
555 break;
556
557 case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_ADJUST:
558 r = -EINVAL;
559 break;
560
561 default:
562 break;
563 }
564
565 mutex_unlock(&vcpu->kvm->lock);
566 return r;
567 }
568
569 int kvm_xen_write_hypercall_page(struct kvm_vcpu *vcpu, u64 data)
570 {
571 struct kvm *kvm = vcpu->kvm;
572 u32 page_num = data & ~PAGE_MASK;
573 u64 page_addr = data & PAGE_MASK;
574 bool lm = is_long_mode(vcpu);
575
576 /* Latch long_mode for shared_info pages etc. */
577 vcpu->kvm->arch.xen.long_mode = lm;
578
579 /*
580 * If Xen hypercall intercept is enabled, fill the hypercall
581 * page with VMCALL/VMMCALL instructions since that's what
582 * we catch. Else the VMM has provided the hypercall pages
583 * with instructions of its own choosing, so use those.
584 */
585 if (kvm_xen_hypercall_enabled(kvm)) {
586 u8 instructions[32];
587 int i;
588
589 if (page_num)
590 return 1;
591
592 /* mov imm32, %eax */
593 instructions[0] = 0xb8;
594
595 /* vmcall / vmmcall */
596 kvm_x86_ops.patch_hypercall(vcpu, instructions + 5);
597
598 /* ret */
599 instructions[8] = 0xc3;
600
601 /* int3 to pad */
602 memset(instructions + 9, 0xcc, sizeof(instructions) - 9);
603
604 for (i = 0; i < PAGE_SIZE / sizeof(instructions); i++) {
605 *(u32 *)&instructions[1] = i;
606 if (kvm_vcpu_write_guest(vcpu,
607 page_addr + (i * sizeof(instructions)),
608 instructions, sizeof(instructions)))
609 return 1;
610 }
611 } else {
612 /*
613 * Note, truncation is a non-issue as 'lm' is guaranteed to be
614 * false for a 32-bit kernel, i.e. when hva_t is only 4 bytes.
615 */
616 hva_t blob_addr = lm ? kvm->arch.xen_hvm_config.blob_addr_64
617 : kvm->arch.xen_hvm_config.blob_addr_32;
618 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
619 : kvm->arch.xen_hvm_config.blob_size_32;
620 u8 *page;
621
622 if (page_num >= blob_size)
623 return 1;
624
625 blob_addr += page_num * PAGE_SIZE;
626
627 page = memdup_user((u8 __user *)blob_addr, PAGE_SIZE);
628 if (IS_ERR(page))
629 return PTR_ERR(page);
630
631 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE)) {
632 kfree(page);
633 return 1;
634 }
635 }
636 return 0;
637 }
638
639 int kvm_xen_hvm_config(struct kvm *kvm, struct kvm_xen_hvm_config *xhc)
640 {
641 if (xhc->flags & ~KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL)
642 return -EINVAL;
643
644 /*
645 * With hypercall interception the kernel generates its own
646 * hypercall page so it must not be provided.
647 */
648 if ((xhc->flags & KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL) &&
649 (xhc->blob_addr_32 || xhc->blob_addr_64 ||
650 xhc->blob_size_32 || xhc->blob_size_64))
651 return -EINVAL;
652
653 mutex_lock(&kvm->lock);
654
655 if (xhc->msr && !kvm->arch.xen_hvm_config.msr)
656 static_branch_inc(&kvm_xen_enabled.key);
657 else if (!xhc->msr && kvm->arch.xen_hvm_config.msr)
658 static_branch_slow_dec_deferred(&kvm_xen_enabled);
659
660 memcpy(&kvm->arch.xen_hvm_config, xhc, sizeof(*xhc));
661
662 mutex_unlock(&kvm->lock);
663 return 0;
664 }
665
666 void kvm_xen_destroy_vm(struct kvm *kvm)
667 {
668 if (kvm->arch.xen_hvm_config.msr)
669 static_branch_slow_dec_deferred(&kvm_xen_enabled);
670 }
671
672 static int kvm_xen_hypercall_set_result(struct kvm_vcpu *vcpu, u64 result)
673 {
674 kvm_rax_write(vcpu, result);
675 return kvm_skip_emulated_instruction(vcpu);
676 }
677
678 static int kvm_xen_hypercall_complete_userspace(struct kvm_vcpu *vcpu)
679 {
680 struct kvm_run *run = vcpu->run;
681
682 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.xen.hypercall_rip)))
683 return 1;
684
685 return kvm_xen_hypercall_set_result(vcpu, run->xen.u.hcall.result);
686 }
687
688 int kvm_xen_hypercall(struct kvm_vcpu *vcpu)
689 {
690 bool longmode;
691 u64 input, params[6];
692
693 input = (u64)kvm_register_read(vcpu, VCPU_REGS_RAX);
694
695 /* Hyper-V hypercalls get bit 31 set in EAX */
696 if ((input & 0x80000000) &&
697 kvm_hv_hypercall_enabled(vcpu))
698 return kvm_hv_hypercall(vcpu);
699
700 longmode = is_64_bit_mode(vcpu);
701 if (!longmode) {
702 params[0] = (u32)kvm_rbx_read(vcpu);
703 params[1] = (u32)kvm_rcx_read(vcpu);
704 params[2] = (u32)kvm_rdx_read(vcpu);
705 params[3] = (u32)kvm_rsi_read(vcpu);
706 params[4] = (u32)kvm_rdi_read(vcpu);
707 params[5] = (u32)kvm_rbp_read(vcpu);
708 }
709 #ifdef CONFIG_X86_64
710 else {
711 params[0] = (u64)kvm_rdi_read(vcpu);
712 params[1] = (u64)kvm_rsi_read(vcpu);
713 params[2] = (u64)kvm_rdx_read(vcpu);
714 params[3] = (u64)kvm_r10_read(vcpu);
715 params[4] = (u64)kvm_r8_read(vcpu);
716 params[5] = (u64)kvm_r9_read(vcpu);
717 }
718 #endif
719 trace_kvm_xen_hypercall(input, params[0], params[1], params[2],
720 params[3], params[4], params[5]);
721
722 vcpu->run->exit_reason = KVM_EXIT_XEN;
723 vcpu->run->xen.type = KVM_EXIT_XEN_HCALL;
724 vcpu->run->xen.u.hcall.longmode = longmode;
725 vcpu->run->xen.u.hcall.cpl = kvm_x86_ops.get_cpl(vcpu);
726 vcpu->run->xen.u.hcall.input = input;
727 vcpu->run->xen.u.hcall.params[0] = params[0];
728 vcpu->run->xen.u.hcall.params[1] = params[1];
729 vcpu->run->xen.u.hcall.params[2] = params[2];
730 vcpu->run->xen.u.hcall.params[3] = params[3];
731 vcpu->run->xen.u.hcall.params[4] = params[4];
732 vcpu->run->xen.u.hcall.params[5] = params[5];
733 vcpu->arch.xen.hypercall_rip = kvm_get_linear_rip(vcpu);
734 vcpu->arch.complete_userspace_io =
735 kvm_xen_hypercall_complete_userspace;
736
737 return 0;
738 }
739
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.