Loading...
Note: File does not exist in v5.4.
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#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9
10#include "x86.h"
11#include "xen.h"
12#include "hyperv.h"
13#include "lapic.h"
14
15#include <linux/eventfd.h>
16#include <linux/kvm_host.h>
17#include <linux/sched/stat.h>
18
19#include <trace/events/kvm.h>
20#include <xen/interface/xen.h>
21#include <xen/interface/vcpu.h>
22#include <xen/interface/version.h>
23#include <xen/interface/event_channel.h>
24#include <xen/interface/sched.h>
25
26#include <asm/xen/cpuid.h>
27
28#include "cpuid.h"
29#include "trace.h"
30
31static int kvm_xen_set_evtchn(struct kvm_xen_evtchn *xe, struct kvm *kvm);
32static int kvm_xen_setattr_evtchn(struct kvm *kvm, struct kvm_xen_hvm_attr *data);
33static bool kvm_xen_hcall_evtchn_send(struct kvm_vcpu *vcpu, u64 param, u64 *r);
34
35DEFINE_STATIC_KEY_DEFERRED_FALSE(kvm_xen_enabled, HZ);
36
37static int kvm_xen_shared_info_init(struct kvm *kvm, gfn_t gfn)
38{
39 struct gfn_to_pfn_cache *gpc = &kvm->arch.xen.shinfo_cache;
40 struct pvclock_wall_clock *wc;
41 gpa_t gpa = gfn_to_gpa(gfn);
42 u32 *wc_sec_hi;
43 u32 wc_version;
44 u64 wall_nsec;
45 int ret = 0;
46 int idx = srcu_read_lock(&kvm->srcu);
47
48 if (gfn == KVM_XEN_INVALID_GFN) {
49 kvm_gpc_deactivate(gpc);
50 goto out;
51 }
52
53 do {
54 ret = kvm_gpc_activate(gpc, gpa, PAGE_SIZE);
55 if (ret)
56 goto out;
57
58 /*
59 * This code mirrors kvm_write_wall_clock() except that it writes
60 * directly through the pfn cache and doesn't mark the page dirty.
61 */
62 wall_nsec = kvm_get_wall_clock_epoch(kvm);
63
64 /* It could be invalid again already, so we need to check */
65 read_lock_irq(&gpc->lock);
66
67 if (gpc->valid)
68 break;
69
70 read_unlock_irq(&gpc->lock);
71 } while (1);
72
73 /* Paranoia checks on the 32-bit struct layout */
74 BUILD_BUG_ON(offsetof(struct compat_shared_info, wc) != 0x900);
75 BUILD_BUG_ON(offsetof(struct compat_shared_info, arch.wc_sec_hi) != 0x924);
76 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
77
78#ifdef CONFIG_X86_64
79 /* Paranoia checks on the 64-bit struct layout */
80 BUILD_BUG_ON(offsetof(struct shared_info, wc) != 0xc00);
81 BUILD_BUG_ON(offsetof(struct shared_info, wc_sec_hi) != 0xc0c);
82
83 if (IS_ENABLED(CONFIG_64BIT) && kvm->arch.xen.long_mode) {
84 struct shared_info *shinfo = gpc->khva;
85
86 wc_sec_hi = &shinfo->wc_sec_hi;
87 wc = &shinfo->wc;
88 } else
89#endif
90 {
91 struct compat_shared_info *shinfo = gpc->khva;
92
93 wc_sec_hi = &shinfo->arch.wc_sec_hi;
94 wc = &shinfo->wc;
95 }
96
97 /* Increment and ensure an odd value */
98 wc_version = wc->version = (wc->version + 1) | 1;
99 smp_wmb();
100
101 wc->nsec = do_div(wall_nsec, NSEC_PER_SEC);
102 wc->sec = (u32)wall_nsec;
103 *wc_sec_hi = wall_nsec >> 32;
104 smp_wmb();
105
106 wc->version = wc_version + 1;
107 read_unlock_irq(&gpc->lock);
108
109 kvm_make_all_cpus_request(kvm, KVM_REQ_MASTERCLOCK_UPDATE);
110
111out:
112 srcu_read_unlock(&kvm->srcu, idx);
113 return ret;
114}
115
116void kvm_xen_inject_timer_irqs(struct kvm_vcpu *vcpu)
117{
118 if (atomic_read(&vcpu->arch.xen.timer_pending) > 0) {
119 struct kvm_xen_evtchn e;
120
121 e.vcpu_id = vcpu->vcpu_id;
122 e.vcpu_idx = vcpu->vcpu_idx;
123 e.port = vcpu->arch.xen.timer_virq;
124 e.priority = KVM_IRQ_ROUTING_XEN_EVTCHN_PRIO_2LEVEL;
125
126 kvm_xen_set_evtchn(&e, vcpu->kvm);
127
128 vcpu->arch.xen.timer_expires = 0;
129 atomic_set(&vcpu->arch.xen.timer_pending, 0);
130 }
131}
132
133static enum hrtimer_restart xen_timer_callback(struct hrtimer *timer)
134{
135 struct kvm_vcpu *vcpu = container_of(timer, struct kvm_vcpu,
136 arch.xen.timer);
137 struct kvm_xen_evtchn e;
138 int rc;
139
140 if (atomic_read(&vcpu->arch.xen.timer_pending))
141 return HRTIMER_NORESTART;
142
143 e.vcpu_id = vcpu->vcpu_id;
144 e.vcpu_idx = vcpu->vcpu_idx;
145 e.port = vcpu->arch.xen.timer_virq;
146 e.priority = KVM_IRQ_ROUTING_XEN_EVTCHN_PRIO_2LEVEL;
147
148 rc = kvm_xen_set_evtchn_fast(&e, vcpu->kvm);
149 if (rc != -EWOULDBLOCK) {
150 vcpu->arch.xen.timer_expires = 0;
151 return HRTIMER_NORESTART;
152 }
153
154 atomic_inc(&vcpu->arch.xen.timer_pending);
155 kvm_make_request(KVM_REQ_UNBLOCK, vcpu);
156 kvm_vcpu_kick(vcpu);
157
158 return HRTIMER_NORESTART;
159}
160
161static void kvm_xen_start_timer(struct kvm_vcpu *vcpu, u64 guest_abs, s64 delta_ns)
162{
163 /*
164 * Avoid races with the old timer firing. Checking timer_expires
165 * to avoid calling hrtimer_cancel() will only have false positives
166 * so is fine.
167 */
168 if (vcpu->arch.xen.timer_expires)
169 hrtimer_cancel(&vcpu->arch.xen.timer);
170
171 atomic_set(&vcpu->arch.xen.timer_pending, 0);
172 vcpu->arch.xen.timer_expires = guest_abs;
173
174 if (delta_ns <= 0) {
175 xen_timer_callback(&vcpu->arch.xen.timer);
176 } else {
177 ktime_t ktime_now = ktime_get();
178 hrtimer_start(&vcpu->arch.xen.timer,
179 ktime_add_ns(ktime_now, delta_ns),
180 HRTIMER_MODE_ABS_HARD);
181 }
182}
183
184static void kvm_xen_stop_timer(struct kvm_vcpu *vcpu)
185{
186 hrtimer_cancel(&vcpu->arch.xen.timer);
187 vcpu->arch.xen.timer_expires = 0;
188 atomic_set(&vcpu->arch.xen.timer_pending, 0);
189}
190
191static void kvm_xen_init_timer(struct kvm_vcpu *vcpu)
192{
193 hrtimer_init(&vcpu->arch.xen.timer, CLOCK_MONOTONIC,
194 HRTIMER_MODE_ABS_HARD);
195 vcpu->arch.xen.timer.function = xen_timer_callback;
196}
197
198static void kvm_xen_update_runstate_guest(struct kvm_vcpu *v, bool atomic)
199{
200 struct kvm_vcpu_xen *vx = &v->arch.xen;
201 struct gfn_to_pfn_cache *gpc1 = &vx->runstate_cache;
202 struct gfn_to_pfn_cache *gpc2 = &vx->runstate2_cache;
203 size_t user_len, user_len1, user_len2;
204 struct vcpu_runstate_info rs;
205 unsigned long flags;
206 size_t times_ofs;
207 uint8_t *update_bit = NULL;
208 uint64_t entry_time;
209 uint64_t *rs_times;
210 int *rs_state;
211
212 /*
213 * The only difference between 32-bit and 64-bit versions of the
214 * runstate struct is the alignment of uint64_t in 32-bit, which
215 * means that the 64-bit version has an additional 4 bytes of
216 * padding after the first field 'state'. Let's be really really
217 * paranoid about that, and matching it with our internal data
218 * structures that we memcpy into it...
219 */
220 BUILD_BUG_ON(offsetof(struct vcpu_runstate_info, state) != 0);
221 BUILD_BUG_ON(offsetof(struct compat_vcpu_runstate_info, state) != 0);
222 BUILD_BUG_ON(sizeof(struct compat_vcpu_runstate_info) != 0x2c);
223#ifdef CONFIG_X86_64
224 /*
225 * The 64-bit structure has 4 bytes of padding before 'state_entry_time'
226 * so each subsequent field is shifted by 4, and it's 4 bytes longer.
227 */
228 BUILD_BUG_ON(offsetof(struct vcpu_runstate_info, state_entry_time) !=
229 offsetof(struct compat_vcpu_runstate_info, state_entry_time) + 4);
230 BUILD_BUG_ON(offsetof(struct vcpu_runstate_info, time) !=
231 offsetof(struct compat_vcpu_runstate_info, time) + 4);
232 BUILD_BUG_ON(sizeof(struct vcpu_runstate_info) != 0x2c + 4);
233#endif
234 /*
235 * The state field is in the same place at the start of both structs,
236 * and is the same size (int) as vx->current_runstate.
237 */
238 BUILD_BUG_ON(offsetof(struct vcpu_runstate_info, state) !=
239 offsetof(struct compat_vcpu_runstate_info, state));
240 BUILD_BUG_ON(sizeof_field(struct vcpu_runstate_info, state) !=
241 sizeof(vx->current_runstate));
242 BUILD_BUG_ON(sizeof_field(struct compat_vcpu_runstate_info, state) !=
243 sizeof(vx->current_runstate));
244
245 /*
246 * The state_entry_time field is 64 bits in both versions, and the
247 * XEN_RUNSTATE_UPDATE flag is in the top bit, which given that x86
248 * is little-endian means that it's in the last *byte* of the word.
249 * That detail is important later.
250 */
251 BUILD_BUG_ON(sizeof_field(struct vcpu_runstate_info, state_entry_time) !=
252 sizeof(uint64_t));
253 BUILD_BUG_ON(sizeof_field(struct compat_vcpu_runstate_info, state_entry_time) !=
254 sizeof(uint64_t));
255 BUILD_BUG_ON((XEN_RUNSTATE_UPDATE >> 56) != 0x80);
256
257 /*
258 * The time array is four 64-bit quantities in both versions, matching
259 * the vx->runstate_times and immediately following state_entry_time.
260 */
261 BUILD_BUG_ON(offsetof(struct vcpu_runstate_info, state_entry_time) !=
262 offsetof(struct vcpu_runstate_info, time) - sizeof(uint64_t));
263 BUILD_BUG_ON(offsetof(struct compat_vcpu_runstate_info, state_entry_time) !=
264 offsetof(struct compat_vcpu_runstate_info, time) - sizeof(uint64_t));
265 BUILD_BUG_ON(sizeof_field(struct vcpu_runstate_info, time) !=
266 sizeof_field(struct compat_vcpu_runstate_info, time));
267 BUILD_BUG_ON(sizeof_field(struct vcpu_runstate_info, time) !=
268 sizeof(vx->runstate_times));
269
270 if (IS_ENABLED(CONFIG_64BIT) && v->kvm->arch.xen.long_mode) {
271 user_len = sizeof(struct vcpu_runstate_info);
272 times_ofs = offsetof(struct vcpu_runstate_info,
273 state_entry_time);
274 } else {
275 user_len = sizeof(struct compat_vcpu_runstate_info);
276 times_ofs = offsetof(struct compat_vcpu_runstate_info,
277 state_entry_time);
278 }
279
280 /*
281 * There are basically no alignment constraints. The guest can set it
282 * up so it crosses from one page to the next, and at arbitrary byte
283 * alignment (and the 32-bit ABI doesn't align the 64-bit integers
284 * anyway, even if the overall struct had been 64-bit aligned).
285 */
286 if ((gpc1->gpa & ~PAGE_MASK) + user_len >= PAGE_SIZE) {
287 user_len1 = PAGE_SIZE - (gpc1->gpa & ~PAGE_MASK);
288 user_len2 = user_len - user_len1;
289 } else {
290 user_len1 = user_len;
291 user_len2 = 0;
292 }
293 BUG_ON(user_len1 + user_len2 != user_len);
294
295 retry:
296 /*
297 * Attempt to obtain the GPC lock on *both* (if there are two)
298 * gfn_to_pfn caches that cover the region.
299 */
300 if (atomic) {
301 local_irq_save(flags);
302 if (!read_trylock(&gpc1->lock)) {
303 local_irq_restore(flags);
304 return;
305 }
306 } else {
307 read_lock_irqsave(&gpc1->lock, flags);
308 }
309 while (!kvm_gpc_check(gpc1, user_len1)) {
310 read_unlock_irqrestore(&gpc1->lock, flags);
311
312 /* When invoked from kvm_sched_out() we cannot sleep */
313 if (atomic)
314 return;
315
316 if (kvm_gpc_refresh(gpc1, user_len1))
317 return;
318
319 read_lock_irqsave(&gpc1->lock, flags);
320 }
321
322 if (likely(!user_len2)) {
323 /*
324 * Set up three pointers directly to the runstate_info
325 * struct in the guest (via the GPC).
326 *
327 * • @rs_state → state field
328 * • @rs_times → state_entry_time field.
329 * • @update_bit → last byte of state_entry_time, which
330 * contains the XEN_RUNSTATE_UPDATE bit.
331 */
332 rs_state = gpc1->khva;
333 rs_times = gpc1->khva + times_ofs;
334 if (v->kvm->arch.xen.runstate_update_flag)
335 update_bit = ((void *)(&rs_times[1])) - 1;
336 } else {
337 /*
338 * The guest's runstate_info is split across two pages and we
339 * need to hold and validate both GPCs simultaneously. We can
340 * declare a lock ordering GPC1 > GPC2 because nothing else
341 * takes them more than one at a time. Set a subclass on the
342 * gpc1 lock to make lockdep shut up about it.
343 */
344 lock_set_subclass(&gpc1->lock.dep_map, 1, _THIS_IP_);
345 if (atomic) {
346 if (!read_trylock(&gpc2->lock)) {
347 read_unlock_irqrestore(&gpc1->lock, flags);
348 return;
349 }
350 } else {
351 read_lock(&gpc2->lock);
352 }
353
354 if (!kvm_gpc_check(gpc2, user_len2)) {
355 read_unlock(&gpc2->lock);
356 read_unlock_irqrestore(&gpc1->lock, flags);
357
358 /* When invoked from kvm_sched_out() we cannot sleep */
359 if (atomic)
360 return;
361
362 /*
363 * Use kvm_gpc_activate() here because if the runstate
364 * area was configured in 32-bit mode and only extends
365 * to the second page now because the guest changed to
366 * 64-bit mode, the second GPC won't have been set up.
367 */
368 if (kvm_gpc_activate(gpc2, gpc1->gpa + user_len1,
369 user_len2))
370 return;
371
372 /*
373 * We dropped the lock on GPC1 so we have to go all the
374 * way back and revalidate that too.
375 */
376 goto retry;
377 }
378
379 /*
380 * In this case, the runstate_info struct will be assembled on
381 * the kernel stack (compat or not as appropriate) and will
382 * be copied to GPC1/GPC2 with a dual memcpy. Set up the three
383 * rs pointers accordingly.
384 */
385 rs_times = &rs.state_entry_time;
386
387 /*
388 * The rs_state pointer points to the start of what we'll
389 * copy to the guest, which in the case of a compat guest
390 * is the 32-bit field that the compiler thinks is padding.
391 */
392 rs_state = ((void *)rs_times) - times_ofs;
393
394 /*
395 * The update_bit is still directly in the guest memory,
396 * via one GPC or the other.
397 */
398 if (v->kvm->arch.xen.runstate_update_flag) {
399 if (user_len1 >= times_ofs + sizeof(uint64_t))
400 update_bit = gpc1->khva + times_ofs +
401 sizeof(uint64_t) - 1;
402 else
403 update_bit = gpc2->khva + times_ofs +
404 sizeof(uint64_t) - 1 - user_len1;
405 }
406
407#ifdef CONFIG_X86_64
408 /*
409 * Don't leak kernel memory through the padding in the 64-bit
410 * version of the struct.
411 */
412 memset(&rs, 0, offsetof(struct vcpu_runstate_info, state_entry_time));
413#endif
414 }
415
416 /*
417 * First, set the XEN_RUNSTATE_UPDATE bit in the top bit of the
418 * state_entry_time field, directly in the guest. We need to set
419 * that (and write-barrier) before writing to the rest of the
420 * structure, and clear it last. Just as Xen does, we address the
421 * single *byte* in which it resides because it might be in a
422 * different cache line to the rest of the 64-bit word, due to
423 * the (lack of) alignment constraints.
424 */
425 entry_time = vx->runstate_entry_time;
426 if (update_bit) {
427 entry_time |= XEN_RUNSTATE_UPDATE;
428 *update_bit = (vx->runstate_entry_time | XEN_RUNSTATE_UPDATE) >> 56;
429 smp_wmb();
430 }
431
432 /*
433 * Now assemble the actual structure, either on our kernel stack
434 * or directly in the guest according to how the rs_state and
435 * rs_times pointers were set up above.
436 */
437 *rs_state = vx->current_runstate;
438 rs_times[0] = entry_time;
439 memcpy(rs_times + 1, vx->runstate_times, sizeof(vx->runstate_times));
440
441 /* For the split case, we have to then copy it to the guest. */
442 if (user_len2) {
443 memcpy(gpc1->khva, rs_state, user_len1);
444 memcpy(gpc2->khva, ((void *)rs_state) + user_len1, user_len2);
445 }
446 smp_wmb();
447
448 /* Finally, clear the XEN_RUNSTATE_UPDATE bit. */
449 if (update_bit) {
450 entry_time &= ~XEN_RUNSTATE_UPDATE;
451 *update_bit = entry_time >> 56;
452 smp_wmb();
453 }
454
455 if (user_len2)
456 read_unlock(&gpc2->lock);
457
458 read_unlock_irqrestore(&gpc1->lock, flags);
459
460 mark_page_dirty_in_slot(v->kvm, gpc1->memslot, gpc1->gpa >> PAGE_SHIFT);
461 if (user_len2)
462 mark_page_dirty_in_slot(v->kvm, gpc2->memslot, gpc2->gpa >> PAGE_SHIFT);
463}
464
465void kvm_xen_update_runstate(struct kvm_vcpu *v, int state)
466{
467 struct kvm_vcpu_xen *vx = &v->arch.xen;
468 u64 now = get_kvmclock_ns(v->kvm);
469 u64 delta_ns = now - vx->runstate_entry_time;
470 u64 run_delay = current->sched_info.run_delay;
471
472 if (unlikely(!vx->runstate_entry_time))
473 vx->current_runstate = RUNSTATE_offline;
474
475 /*
476 * Time waiting for the scheduler isn't "stolen" if the
477 * vCPU wasn't running anyway.
478 */
479 if (vx->current_runstate == RUNSTATE_running) {
480 u64 steal_ns = run_delay - vx->last_steal;
481
482 delta_ns -= steal_ns;
483
484 vx->runstate_times[RUNSTATE_runnable] += steal_ns;
485 }
486 vx->last_steal = run_delay;
487
488 vx->runstate_times[vx->current_runstate] += delta_ns;
489 vx->current_runstate = state;
490 vx->runstate_entry_time = now;
491
492 if (vx->runstate_cache.active)
493 kvm_xen_update_runstate_guest(v, state == RUNSTATE_runnable);
494}
495
496static void kvm_xen_inject_vcpu_vector(struct kvm_vcpu *v)
497{
498 struct kvm_lapic_irq irq = { };
499 int r;
500
501 irq.dest_id = v->vcpu_id;
502 irq.vector = v->arch.xen.upcall_vector;
503 irq.dest_mode = APIC_DEST_PHYSICAL;
504 irq.shorthand = APIC_DEST_NOSHORT;
505 irq.delivery_mode = APIC_DM_FIXED;
506 irq.level = 1;
507
508 /* The fast version will always work for physical unicast */
509 WARN_ON_ONCE(!kvm_irq_delivery_to_apic_fast(v->kvm, NULL, &irq, &r, NULL));
510}
511
512/*
513 * On event channel delivery, the vcpu_info may not have been accessible.
514 * In that case, there are bits in vcpu->arch.xen.evtchn_pending_sel which
515 * need to be marked into the vcpu_info (and evtchn_upcall_pending set).
516 * Do so now that we can sleep in the context of the vCPU to bring the
517 * page in, and refresh the pfn cache for it.
518 */
519void kvm_xen_inject_pending_events(struct kvm_vcpu *v)
520{
521 unsigned long evtchn_pending_sel = READ_ONCE(v->arch.xen.evtchn_pending_sel);
522 struct gfn_to_pfn_cache *gpc = &v->arch.xen.vcpu_info_cache;
523 unsigned long flags;
524
525 if (!evtchn_pending_sel)
526 return;
527
528 /*
529 * Yes, this is an open-coded loop. But that's just what put_user()
530 * does anyway. Page it in and retry the instruction. We're just a
531 * little more honest about it.
532 */
533 read_lock_irqsave(&gpc->lock, flags);
534 while (!kvm_gpc_check(gpc, sizeof(struct vcpu_info))) {
535 read_unlock_irqrestore(&gpc->lock, flags);
536
537 if (kvm_gpc_refresh(gpc, sizeof(struct vcpu_info)))
538 return;
539
540 read_lock_irqsave(&gpc->lock, flags);
541 }
542
543 /* Now gpc->khva is a valid kernel address for the vcpu_info */
544 if (IS_ENABLED(CONFIG_64BIT) && v->kvm->arch.xen.long_mode) {
545 struct vcpu_info *vi = gpc->khva;
546
547 asm volatile(LOCK_PREFIX "orq %0, %1\n"
548 "notq %0\n"
549 LOCK_PREFIX "andq %0, %2\n"
550 : "=r" (evtchn_pending_sel),
551 "+m" (vi->evtchn_pending_sel),
552 "+m" (v->arch.xen.evtchn_pending_sel)
553 : "0" (evtchn_pending_sel));
554 WRITE_ONCE(vi->evtchn_upcall_pending, 1);
555 } else {
556 u32 evtchn_pending_sel32 = evtchn_pending_sel;
557 struct compat_vcpu_info *vi = gpc->khva;
558
559 asm volatile(LOCK_PREFIX "orl %0, %1\n"
560 "notl %0\n"
561 LOCK_PREFIX "andl %0, %2\n"
562 : "=r" (evtchn_pending_sel32),
563 "+m" (vi->evtchn_pending_sel),
564 "+m" (v->arch.xen.evtchn_pending_sel)
565 : "0" (evtchn_pending_sel32));
566 WRITE_ONCE(vi->evtchn_upcall_pending, 1);
567 }
568 read_unlock_irqrestore(&gpc->lock, flags);
569
570 /* For the per-vCPU lapic vector, deliver it as MSI. */
571 if (v->arch.xen.upcall_vector)
572 kvm_xen_inject_vcpu_vector(v);
573
574 mark_page_dirty_in_slot(v->kvm, gpc->memslot, gpc->gpa >> PAGE_SHIFT);
575}
576
577int __kvm_xen_has_interrupt(struct kvm_vcpu *v)
578{
579 struct gfn_to_pfn_cache *gpc = &v->arch.xen.vcpu_info_cache;
580 unsigned long flags;
581 u8 rc = 0;
582
583 /*
584 * If the global upcall vector (HVMIRQ_callback_vector) is set and
585 * the vCPU's evtchn_upcall_pending flag is set, the IRQ is pending.
586 */
587
588 /* No need for compat handling here */
589 BUILD_BUG_ON(offsetof(struct vcpu_info, evtchn_upcall_pending) !=
590 offsetof(struct compat_vcpu_info, evtchn_upcall_pending));
591 BUILD_BUG_ON(sizeof(rc) !=
592 sizeof_field(struct vcpu_info, evtchn_upcall_pending));
593 BUILD_BUG_ON(sizeof(rc) !=
594 sizeof_field(struct compat_vcpu_info, evtchn_upcall_pending));
595
596 read_lock_irqsave(&gpc->lock, flags);
597 while (!kvm_gpc_check(gpc, sizeof(struct vcpu_info))) {
598 read_unlock_irqrestore(&gpc->lock, flags);
599
600 /*
601 * This function gets called from kvm_vcpu_block() after setting the
602 * task to TASK_INTERRUPTIBLE, to see if it needs to wake immediately
603 * from a HLT. So we really mustn't sleep. If the page ended up absent
604 * at that point, just return 1 in order to trigger an immediate wake,
605 * and we'll end up getting called again from a context where we *can*
606 * fault in the page and wait for it.
607 */
608 if (in_atomic() || !task_is_running(current))
609 return 1;
610
611 if (kvm_gpc_refresh(gpc, sizeof(struct vcpu_info))) {
612 /*
613 * If this failed, userspace has screwed up the
614 * vcpu_info mapping. No interrupts for you.
615 */
616 return 0;
617 }
618 read_lock_irqsave(&gpc->lock, flags);
619 }
620
621 rc = ((struct vcpu_info *)gpc->khva)->evtchn_upcall_pending;
622 read_unlock_irqrestore(&gpc->lock, flags);
623 return rc;
624}
625
626int kvm_xen_hvm_set_attr(struct kvm *kvm, struct kvm_xen_hvm_attr *data)
627{
628 int r = -ENOENT;
629
630
631 switch (data->type) {
632 case KVM_XEN_ATTR_TYPE_LONG_MODE:
633 if (!IS_ENABLED(CONFIG_64BIT) && data->u.long_mode) {
634 r = -EINVAL;
635 } else {
636 mutex_lock(&kvm->arch.xen.xen_lock);
637 kvm->arch.xen.long_mode = !!data->u.long_mode;
638 mutex_unlock(&kvm->arch.xen.xen_lock);
639 r = 0;
640 }
641 break;
642
643 case KVM_XEN_ATTR_TYPE_SHARED_INFO:
644 mutex_lock(&kvm->arch.xen.xen_lock);
645 r = kvm_xen_shared_info_init(kvm, data->u.shared_info.gfn);
646 mutex_unlock(&kvm->arch.xen.xen_lock);
647 break;
648
649 case KVM_XEN_ATTR_TYPE_UPCALL_VECTOR:
650 if (data->u.vector && data->u.vector < 0x10)
651 r = -EINVAL;
652 else {
653 mutex_lock(&kvm->arch.xen.xen_lock);
654 kvm->arch.xen.upcall_vector = data->u.vector;
655 mutex_unlock(&kvm->arch.xen.xen_lock);
656 r = 0;
657 }
658 break;
659
660 case KVM_XEN_ATTR_TYPE_EVTCHN:
661 r = kvm_xen_setattr_evtchn(kvm, data);
662 break;
663
664 case KVM_XEN_ATTR_TYPE_XEN_VERSION:
665 mutex_lock(&kvm->arch.xen.xen_lock);
666 kvm->arch.xen.xen_version = data->u.xen_version;
667 mutex_unlock(&kvm->arch.xen.xen_lock);
668 r = 0;
669 break;
670
671 case KVM_XEN_ATTR_TYPE_RUNSTATE_UPDATE_FLAG:
672 if (!sched_info_on()) {
673 r = -EOPNOTSUPP;
674 break;
675 }
676 mutex_lock(&kvm->arch.xen.xen_lock);
677 kvm->arch.xen.runstate_update_flag = !!data->u.runstate_update_flag;
678 mutex_unlock(&kvm->arch.xen.xen_lock);
679 r = 0;
680 break;
681
682 default:
683 break;
684 }
685
686 return r;
687}
688
689int kvm_xen_hvm_get_attr(struct kvm *kvm, struct kvm_xen_hvm_attr *data)
690{
691 int r = -ENOENT;
692
693 mutex_lock(&kvm->arch.xen.xen_lock);
694
695 switch (data->type) {
696 case KVM_XEN_ATTR_TYPE_LONG_MODE:
697 data->u.long_mode = kvm->arch.xen.long_mode;
698 r = 0;
699 break;
700
701 case KVM_XEN_ATTR_TYPE_SHARED_INFO:
702 if (kvm->arch.xen.shinfo_cache.active)
703 data->u.shared_info.gfn = gpa_to_gfn(kvm->arch.xen.shinfo_cache.gpa);
704 else
705 data->u.shared_info.gfn = KVM_XEN_INVALID_GFN;
706 r = 0;
707 break;
708
709 case KVM_XEN_ATTR_TYPE_UPCALL_VECTOR:
710 data->u.vector = kvm->arch.xen.upcall_vector;
711 r = 0;
712 break;
713
714 case KVM_XEN_ATTR_TYPE_XEN_VERSION:
715 data->u.xen_version = kvm->arch.xen.xen_version;
716 r = 0;
717 break;
718
719 case KVM_XEN_ATTR_TYPE_RUNSTATE_UPDATE_FLAG:
720 if (!sched_info_on()) {
721 r = -EOPNOTSUPP;
722 break;
723 }
724 data->u.runstate_update_flag = kvm->arch.xen.runstate_update_flag;
725 r = 0;
726 break;
727
728 default:
729 break;
730 }
731
732 mutex_unlock(&kvm->arch.xen.xen_lock);
733 return r;
734}
735
736int kvm_xen_vcpu_set_attr(struct kvm_vcpu *vcpu, struct kvm_xen_vcpu_attr *data)
737{
738 int idx, r = -ENOENT;
739
740 mutex_lock(&vcpu->kvm->arch.xen.xen_lock);
741 idx = srcu_read_lock(&vcpu->kvm->srcu);
742
743 switch (data->type) {
744 case KVM_XEN_VCPU_ATTR_TYPE_VCPU_INFO:
745 /* No compat necessary here. */
746 BUILD_BUG_ON(sizeof(struct vcpu_info) !=
747 sizeof(struct compat_vcpu_info));
748 BUILD_BUG_ON(offsetof(struct vcpu_info, time) !=
749 offsetof(struct compat_vcpu_info, time));
750
751 if (data->u.gpa == KVM_XEN_INVALID_GPA) {
752 kvm_gpc_deactivate(&vcpu->arch.xen.vcpu_info_cache);
753 r = 0;
754 break;
755 }
756
757 r = kvm_gpc_activate(&vcpu->arch.xen.vcpu_info_cache,
758 data->u.gpa, sizeof(struct vcpu_info));
759 if (!r)
760 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
761
762 break;
763
764 case KVM_XEN_VCPU_ATTR_TYPE_VCPU_TIME_INFO:
765 if (data->u.gpa == KVM_XEN_INVALID_GPA) {
766 kvm_gpc_deactivate(&vcpu->arch.xen.vcpu_time_info_cache);
767 r = 0;
768 break;
769 }
770
771 r = kvm_gpc_activate(&vcpu->arch.xen.vcpu_time_info_cache,
772 data->u.gpa,
773 sizeof(struct pvclock_vcpu_time_info));
774 if (!r)
775 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
776 break;
777
778 case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_ADDR: {
779 size_t sz, sz1, sz2;
780
781 if (!sched_info_on()) {
782 r = -EOPNOTSUPP;
783 break;
784 }
785 if (data->u.gpa == KVM_XEN_INVALID_GPA) {
786 r = 0;
787 deactivate_out:
788 kvm_gpc_deactivate(&vcpu->arch.xen.runstate_cache);
789 kvm_gpc_deactivate(&vcpu->arch.xen.runstate2_cache);
790 break;
791 }
792
793 /*
794 * If the guest switches to 64-bit mode after setting the runstate
795 * address, that's actually OK. kvm_xen_update_runstate_guest()
796 * will cope.
797 */
798 if (IS_ENABLED(CONFIG_64BIT) && vcpu->kvm->arch.xen.long_mode)
799 sz = sizeof(struct vcpu_runstate_info);
800 else
801 sz = sizeof(struct compat_vcpu_runstate_info);
802
803 /* How much fits in the (first) page? */
804 sz1 = PAGE_SIZE - (data->u.gpa & ~PAGE_MASK);
805 r = kvm_gpc_activate(&vcpu->arch.xen.runstate_cache,
806 data->u.gpa, sz1);
807 if (r)
808 goto deactivate_out;
809
810 /* Either map the second page, or deactivate the second GPC */
811 if (sz1 >= sz) {
812 kvm_gpc_deactivate(&vcpu->arch.xen.runstate2_cache);
813 } else {
814 sz2 = sz - sz1;
815 BUG_ON((data->u.gpa + sz1) & ~PAGE_MASK);
816 r = kvm_gpc_activate(&vcpu->arch.xen.runstate2_cache,
817 data->u.gpa + sz1, sz2);
818 if (r)
819 goto deactivate_out;
820 }
821
822 kvm_xen_update_runstate_guest(vcpu, false);
823 break;
824 }
825 case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_CURRENT:
826 if (!sched_info_on()) {
827 r = -EOPNOTSUPP;
828 break;
829 }
830 if (data->u.runstate.state > RUNSTATE_offline) {
831 r = -EINVAL;
832 break;
833 }
834
835 kvm_xen_update_runstate(vcpu, data->u.runstate.state);
836 r = 0;
837 break;
838
839 case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_DATA:
840 if (!sched_info_on()) {
841 r = -EOPNOTSUPP;
842 break;
843 }
844 if (data->u.runstate.state > RUNSTATE_offline) {
845 r = -EINVAL;
846 break;
847 }
848 if (data->u.runstate.state_entry_time !=
849 (data->u.runstate.time_running +
850 data->u.runstate.time_runnable +
851 data->u.runstate.time_blocked +
852 data->u.runstate.time_offline)) {
853 r = -EINVAL;
854 break;
855 }
856 if (get_kvmclock_ns(vcpu->kvm) <
857 data->u.runstate.state_entry_time) {
858 r = -EINVAL;
859 break;
860 }
861
862 vcpu->arch.xen.current_runstate = data->u.runstate.state;
863 vcpu->arch.xen.runstate_entry_time =
864 data->u.runstate.state_entry_time;
865 vcpu->arch.xen.runstate_times[RUNSTATE_running] =
866 data->u.runstate.time_running;
867 vcpu->arch.xen.runstate_times[RUNSTATE_runnable] =
868 data->u.runstate.time_runnable;
869 vcpu->arch.xen.runstate_times[RUNSTATE_blocked] =
870 data->u.runstate.time_blocked;
871 vcpu->arch.xen.runstate_times[RUNSTATE_offline] =
872 data->u.runstate.time_offline;
873 vcpu->arch.xen.last_steal = current->sched_info.run_delay;
874 r = 0;
875 break;
876
877 case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_ADJUST:
878 if (!sched_info_on()) {
879 r = -EOPNOTSUPP;
880 break;
881 }
882 if (data->u.runstate.state > RUNSTATE_offline &&
883 data->u.runstate.state != (u64)-1) {
884 r = -EINVAL;
885 break;
886 }
887 /* The adjustment must add up */
888 if (data->u.runstate.state_entry_time !=
889 (data->u.runstate.time_running +
890 data->u.runstate.time_runnable +
891 data->u.runstate.time_blocked +
892 data->u.runstate.time_offline)) {
893 r = -EINVAL;
894 break;
895 }
896
897 if (get_kvmclock_ns(vcpu->kvm) <
898 (vcpu->arch.xen.runstate_entry_time +
899 data->u.runstate.state_entry_time)) {
900 r = -EINVAL;
901 break;
902 }
903
904 vcpu->arch.xen.runstate_entry_time +=
905 data->u.runstate.state_entry_time;
906 vcpu->arch.xen.runstate_times[RUNSTATE_running] +=
907 data->u.runstate.time_running;
908 vcpu->arch.xen.runstate_times[RUNSTATE_runnable] +=
909 data->u.runstate.time_runnable;
910 vcpu->arch.xen.runstate_times[RUNSTATE_blocked] +=
911 data->u.runstate.time_blocked;
912 vcpu->arch.xen.runstate_times[RUNSTATE_offline] +=
913 data->u.runstate.time_offline;
914
915 if (data->u.runstate.state <= RUNSTATE_offline)
916 kvm_xen_update_runstate(vcpu, data->u.runstate.state);
917 else if (vcpu->arch.xen.runstate_cache.active)
918 kvm_xen_update_runstate_guest(vcpu, false);
919 r = 0;
920 break;
921
922 case KVM_XEN_VCPU_ATTR_TYPE_VCPU_ID:
923 if (data->u.vcpu_id >= KVM_MAX_VCPUS)
924 r = -EINVAL;
925 else {
926 vcpu->arch.xen.vcpu_id = data->u.vcpu_id;
927 r = 0;
928 }
929 break;
930
931 case KVM_XEN_VCPU_ATTR_TYPE_TIMER:
932 if (data->u.timer.port &&
933 data->u.timer.priority != KVM_IRQ_ROUTING_XEN_EVTCHN_PRIO_2LEVEL) {
934 r = -EINVAL;
935 break;
936 }
937
938 if (!vcpu->arch.xen.timer.function)
939 kvm_xen_init_timer(vcpu);
940
941 /* Stop the timer (if it's running) before changing the vector */
942 kvm_xen_stop_timer(vcpu);
943 vcpu->arch.xen.timer_virq = data->u.timer.port;
944
945 /* Start the timer if the new value has a valid vector+expiry. */
946 if (data->u.timer.port && data->u.timer.expires_ns)
947 kvm_xen_start_timer(vcpu, data->u.timer.expires_ns,
948 data->u.timer.expires_ns -
949 get_kvmclock_ns(vcpu->kvm));
950
951 r = 0;
952 break;
953
954 case KVM_XEN_VCPU_ATTR_TYPE_UPCALL_VECTOR:
955 if (data->u.vector && data->u.vector < 0x10)
956 r = -EINVAL;
957 else {
958 vcpu->arch.xen.upcall_vector = data->u.vector;
959 r = 0;
960 }
961 break;
962
963 default:
964 break;
965 }
966
967 srcu_read_unlock(&vcpu->kvm->srcu, idx);
968 mutex_unlock(&vcpu->kvm->arch.xen.xen_lock);
969 return r;
970}
971
972int kvm_xen_vcpu_get_attr(struct kvm_vcpu *vcpu, struct kvm_xen_vcpu_attr *data)
973{
974 int r = -ENOENT;
975
976 mutex_lock(&vcpu->kvm->arch.xen.xen_lock);
977
978 switch (data->type) {
979 case KVM_XEN_VCPU_ATTR_TYPE_VCPU_INFO:
980 if (vcpu->arch.xen.vcpu_info_cache.active)
981 data->u.gpa = vcpu->arch.xen.vcpu_info_cache.gpa;
982 else
983 data->u.gpa = KVM_XEN_INVALID_GPA;
984 r = 0;
985 break;
986
987 case KVM_XEN_VCPU_ATTR_TYPE_VCPU_TIME_INFO:
988 if (vcpu->arch.xen.vcpu_time_info_cache.active)
989 data->u.gpa = vcpu->arch.xen.vcpu_time_info_cache.gpa;
990 else
991 data->u.gpa = KVM_XEN_INVALID_GPA;
992 r = 0;
993 break;
994
995 case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_ADDR:
996 if (!sched_info_on()) {
997 r = -EOPNOTSUPP;
998 break;
999 }
1000 if (vcpu->arch.xen.runstate_cache.active) {
1001 data->u.gpa = vcpu->arch.xen.runstate_cache.gpa;
1002 r = 0;
1003 }
1004 break;
1005
1006 case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_CURRENT:
1007 if (!sched_info_on()) {
1008 r = -EOPNOTSUPP;
1009 break;
1010 }
1011 data->u.runstate.state = vcpu->arch.xen.current_runstate;
1012 r = 0;
1013 break;
1014
1015 case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_DATA:
1016 if (!sched_info_on()) {
1017 r = -EOPNOTSUPP;
1018 break;
1019 }
1020 data->u.runstate.state = vcpu->arch.xen.current_runstate;
1021 data->u.runstate.state_entry_time =
1022 vcpu->arch.xen.runstate_entry_time;
1023 data->u.runstate.time_running =
1024 vcpu->arch.xen.runstate_times[RUNSTATE_running];
1025 data->u.runstate.time_runnable =
1026 vcpu->arch.xen.runstate_times[RUNSTATE_runnable];
1027 data->u.runstate.time_blocked =
1028 vcpu->arch.xen.runstate_times[RUNSTATE_blocked];
1029 data->u.runstate.time_offline =
1030 vcpu->arch.xen.runstate_times[RUNSTATE_offline];
1031 r = 0;
1032 break;
1033
1034 case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_ADJUST:
1035 r = -EINVAL;
1036 break;
1037
1038 case KVM_XEN_VCPU_ATTR_TYPE_VCPU_ID:
1039 data->u.vcpu_id = vcpu->arch.xen.vcpu_id;
1040 r = 0;
1041 break;
1042
1043 case KVM_XEN_VCPU_ATTR_TYPE_TIMER:
1044 /*
1045 * Ensure a consistent snapshot of state is captured, with a
1046 * timer either being pending, or the event channel delivered
1047 * to the corresponding bit in the shared_info. Not still
1048 * lurking in the timer_pending flag for deferred delivery.
1049 * Purely as an optimisation, if the timer_expires field is
1050 * zero, that means the timer isn't active (or even in the
1051 * timer_pending flag) and there is no need to cancel it.
1052 */
1053 if (vcpu->arch.xen.timer_expires) {
1054 hrtimer_cancel(&vcpu->arch.xen.timer);
1055 kvm_xen_inject_timer_irqs(vcpu);
1056 }
1057
1058 data->u.timer.port = vcpu->arch.xen.timer_virq;
1059 data->u.timer.priority = KVM_IRQ_ROUTING_XEN_EVTCHN_PRIO_2LEVEL;
1060 data->u.timer.expires_ns = vcpu->arch.xen.timer_expires;
1061
1062 /*
1063 * The hrtimer may trigger and raise the IRQ immediately,
1064 * while the returned state causes it to be set up and
1065 * raised again on the destination system after migration.
1066 * That's fine, as the guest won't even have had a chance
1067 * to run and handle the interrupt. Asserting an already
1068 * pending event channel is idempotent.
1069 */
1070 if (vcpu->arch.xen.timer_expires)
1071 hrtimer_start_expires(&vcpu->arch.xen.timer,
1072 HRTIMER_MODE_ABS_HARD);
1073
1074 r = 0;
1075 break;
1076
1077 case KVM_XEN_VCPU_ATTR_TYPE_UPCALL_VECTOR:
1078 data->u.vector = vcpu->arch.xen.upcall_vector;
1079 r = 0;
1080 break;
1081
1082 default:
1083 break;
1084 }
1085
1086 mutex_unlock(&vcpu->kvm->arch.xen.xen_lock);
1087 return r;
1088}
1089
1090int kvm_xen_write_hypercall_page(struct kvm_vcpu *vcpu, u64 data)
1091{
1092 struct kvm *kvm = vcpu->kvm;
1093 u32 page_num = data & ~PAGE_MASK;
1094 u64 page_addr = data & PAGE_MASK;
1095 bool lm = is_long_mode(vcpu);
1096
1097 /* Latch long_mode for shared_info pages etc. */
1098 vcpu->kvm->arch.xen.long_mode = lm;
1099
1100 /*
1101 * If Xen hypercall intercept is enabled, fill the hypercall
1102 * page with VMCALL/VMMCALL instructions since that's what
1103 * we catch. Else the VMM has provided the hypercall pages
1104 * with instructions of its own choosing, so use those.
1105 */
1106 if (kvm_xen_hypercall_enabled(kvm)) {
1107 u8 instructions[32];
1108 int i;
1109
1110 if (page_num)
1111 return 1;
1112
1113 /* mov imm32, %eax */
1114 instructions[0] = 0xb8;
1115
1116 /* vmcall / vmmcall */
1117 static_call(kvm_x86_patch_hypercall)(vcpu, instructions + 5);
1118
1119 /* ret */
1120 instructions[8] = 0xc3;
1121
1122 /* int3 to pad */
1123 memset(instructions + 9, 0xcc, sizeof(instructions) - 9);
1124
1125 for (i = 0; i < PAGE_SIZE / sizeof(instructions); i++) {
1126 *(u32 *)&instructions[1] = i;
1127 if (kvm_vcpu_write_guest(vcpu,
1128 page_addr + (i * sizeof(instructions)),
1129 instructions, sizeof(instructions)))
1130 return 1;
1131 }
1132 } else {
1133 /*
1134 * Note, truncation is a non-issue as 'lm' is guaranteed to be
1135 * false for a 32-bit kernel, i.e. when hva_t is only 4 bytes.
1136 */
1137 hva_t blob_addr = lm ? kvm->arch.xen_hvm_config.blob_addr_64
1138 : kvm->arch.xen_hvm_config.blob_addr_32;
1139 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1140 : kvm->arch.xen_hvm_config.blob_size_32;
1141 u8 *page;
1142 int ret;
1143
1144 if (page_num >= blob_size)
1145 return 1;
1146
1147 blob_addr += page_num * PAGE_SIZE;
1148
1149 page = memdup_user((u8 __user *)blob_addr, PAGE_SIZE);
1150 if (IS_ERR(page))
1151 return PTR_ERR(page);
1152
1153 ret = kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE);
1154 kfree(page);
1155 if (ret)
1156 return 1;
1157 }
1158 return 0;
1159}
1160
1161int kvm_xen_hvm_config(struct kvm *kvm, struct kvm_xen_hvm_config *xhc)
1162{
1163 /* Only some feature flags need to be *enabled* by userspace */
1164 u32 permitted_flags = KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL |
1165 KVM_XEN_HVM_CONFIG_EVTCHN_SEND |
1166 KVM_XEN_HVM_CONFIG_PVCLOCK_TSC_UNSTABLE;
1167 u32 old_flags;
1168
1169 if (xhc->flags & ~permitted_flags)
1170 return -EINVAL;
1171
1172 /*
1173 * With hypercall interception the kernel generates its own
1174 * hypercall page so it must not be provided.
1175 */
1176 if ((xhc->flags & KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL) &&
1177 (xhc->blob_addr_32 || xhc->blob_addr_64 ||
1178 xhc->blob_size_32 || xhc->blob_size_64))
1179 return -EINVAL;
1180
1181 mutex_lock(&kvm->arch.xen.xen_lock);
1182
1183 if (xhc->msr && !kvm->arch.xen_hvm_config.msr)
1184 static_branch_inc(&kvm_xen_enabled.key);
1185 else if (!xhc->msr && kvm->arch.xen_hvm_config.msr)
1186 static_branch_slow_dec_deferred(&kvm_xen_enabled);
1187
1188 old_flags = kvm->arch.xen_hvm_config.flags;
1189 memcpy(&kvm->arch.xen_hvm_config, xhc, sizeof(*xhc));
1190
1191 mutex_unlock(&kvm->arch.xen.xen_lock);
1192
1193 if ((old_flags ^ xhc->flags) & KVM_XEN_HVM_CONFIG_PVCLOCK_TSC_UNSTABLE)
1194 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
1195
1196 return 0;
1197}
1198
1199static int kvm_xen_hypercall_set_result(struct kvm_vcpu *vcpu, u64 result)
1200{
1201 kvm_rax_write(vcpu, result);
1202 return kvm_skip_emulated_instruction(vcpu);
1203}
1204
1205static int kvm_xen_hypercall_complete_userspace(struct kvm_vcpu *vcpu)
1206{
1207 struct kvm_run *run = vcpu->run;
1208
1209 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.xen.hypercall_rip)))
1210 return 1;
1211
1212 return kvm_xen_hypercall_set_result(vcpu, run->xen.u.hcall.result);
1213}
1214
1215static inline int max_evtchn_port(struct kvm *kvm)
1216{
1217 if (IS_ENABLED(CONFIG_64BIT) && kvm->arch.xen.long_mode)
1218 return EVTCHN_2L_NR_CHANNELS;
1219 else
1220 return COMPAT_EVTCHN_2L_NR_CHANNELS;
1221}
1222
1223static bool wait_pending_event(struct kvm_vcpu *vcpu, int nr_ports,
1224 evtchn_port_t *ports)
1225{
1226 struct kvm *kvm = vcpu->kvm;
1227 struct gfn_to_pfn_cache *gpc = &kvm->arch.xen.shinfo_cache;
1228 unsigned long *pending_bits;
1229 unsigned long flags;
1230 bool ret = true;
1231 int idx, i;
1232
1233 idx = srcu_read_lock(&kvm->srcu);
1234 read_lock_irqsave(&gpc->lock, flags);
1235 if (!kvm_gpc_check(gpc, PAGE_SIZE))
1236 goto out_rcu;
1237
1238 ret = false;
1239 if (IS_ENABLED(CONFIG_64BIT) && kvm->arch.xen.long_mode) {
1240 struct shared_info *shinfo = gpc->khva;
1241 pending_bits = (unsigned long *)&shinfo->evtchn_pending;
1242 } else {
1243 struct compat_shared_info *shinfo = gpc->khva;
1244 pending_bits = (unsigned long *)&shinfo->evtchn_pending;
1245 }
1246
1247 for (i = 0; i < nr_ports; i++) {
1248 if (test_bit(ports[i], pending_bits)) {
1249 ret = true;
1250 break;
1251 }
1252 }
1253
1254 out_rcu:
1255 read_unlock_irqrestore(&gpc->lock, flags);
1256 srcu_read_unlock(&kvm->srcu, idx);
1257
1258 return ret;
1259}
1260
1261static bool kvm_xen_schedop_poll(struct kvm_vcpu *vcpu, bool longmode,
1262 u64 param, u64 *r)
1263{
1264 struct sched_poll sched_poll;
1265 evtchn_port_t port, *ports;
1266 struct x86_exception e;
1267 int i;
1268
1269 if (!lapic_in_kernel(vcpu) ||
1270 !(vcpu->kvm->arch.xen_hvm_config.flags & KVM_XEN_HVM_CONFIG_EVTCHN_SEND))
1271 return false;
1272
1273 if (IS_ENABLED(CONFIG_64BIT) && !longmode) {
1274 struct compat_sched_poll sp32;
1275
1276 /* Sanity check that the compat struct definition is correct */
1277 BUILD_BUG_ON(sizeof(sp32) != 16);
1278
1279 if (kvm_read_guest_virt(vcpu, param, &sp32, sizeof(sp32), &e)) {
1280 *r = -EFAULT;
1281 return true;
1282 }
1283
1284 /*
1285 * This is a 32-bit pointer to an array of evtchn_port_t which
1286 * are uint32_t, so once it's converted no further compat
1287 * handling is needed.
1288 */
1289 sched_poll.ports = (void *)(unsigned long)(sp32.ports);
1290 sched_poll.nr_ports = sp32.nr_ports;
1291 sched_poll.timeout = sp32.timeout;
1292 } else {
1293 if (kvm_read_guest_virt(vcpu, param, &sched_poll,
1294 sizeof(sched_poll), &e)) {
1295 *r = -EFAULT;
1296 return true;
1297 }
1298 }
1299
1300 if (unlikely(sched_poll.nr_ports > 1)) {
1301 /* Xen (unofficially) limits number of pollers to 128 */
1302 if (sched_poll.nr_ports > 128) {
1303 *r = -EINVAL;
1304 return true;
1305 }
1306
1307 ports = kmalloc_array(sched_poll.nr_ports,
1308 sizeof(*ports), GFP_KERNEL);
1309 if (!ports) {
1310 *r = -ENOMEM;
1311 return true;
1312 }
1313 } else
1314 ports = &port;
1315
1316 if (kvm_read_guest_virt(vcpu, (gva_t)sched_poll.ports, ports,
1317 sched_poll.nr_ports * sizeof(*ports), &e)) {
1318 *r = -EFAULT;
1319 return true;
1320 }
1321
1322 for (i = 0; i < sched_poll.nr_ports; i++) {
1323 if (ports[i] >= max_evtchn_port(vcpu->kvm)) {
1324 *r = -EINVAL;
1325 goto out;
1326 }
1327 }
1328
1329 if (sched_poll.nr_ports == 1)
1330 vcpu->arch.xen.poll_evtchn = port;
1331 else
1332 vcpu->arch.xen.poll_evtchn = -1;
1333
1334 set_bit(vcpu->vcpu_idx, vcpu->kvm->arch.xen.poll_mask);
1335
1336 if (!wait_pending_event(vcpu, sched_poll.nr_ports, ports)) {
1337 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
1338
1339 if (sched_poll.timeout)
1340 mod_timer(&vcpu->arch.xen.poll_timer,
1341 jiffies + nsecs_to_jiffies(sched_poll.timeout));
1342
1343 kvm_vcpu_halt(vcpu);
1344
1345 if (sched_poll.timeout)
1346 del_timer(&vcpu->arch.xen.poll_timer);
1347
1348 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1349 }
1350
1351 vcpu->arch.xen.poll_evtchn = 0;
1352 *r = 0;
1353out:
1354 /* Really, this is only needed in case of timeout */
1355 clear_bit(vcpu->vcpu_idx, vcpu->kvm->arch.xen.poll_mask);
1356
1357 if (unlikely(sched_poll.nr_ports > 1))
1358 kfree(ports);
1359 return true;
1360}
1361
1362static void cancel_evtchn_poll(struct timer_list *t)
1363{
1364 struct kvm_vcpu *vcpu = from_timer(vcpu, t, arch.xen.poll_timer);
1365
1366 kvm_make_request(KVM_REQ_UNBLOCK, vcpu);
1367 kvm_vcpu_kick(vcpu);
1368}
1369
1370static bool kvm_xen_hcall_sched_op(struct kvm_vcpu *vcpu, bool longmode,
1371 int cmd, u64 param, u64 *r)
1372{
1373 switch (cmd) {
1374 case SCHEDOP_poll:
1375 if (kvm_xen_schedop_poll(vcpu, longmode, param, r))
1376 return true;
1377 fallthrough;
1378 case SCHEDOP_yield:
1379 kvm_vcpu_on_spin(vcpu, true);
1380 *r = 0;
1381 return true;
1382 default:
1383 break;
1384 }
1385
1386 return false;
1387}
1388
1389struct compat_vcpu_set_singleshot_timer {
1390 uint64_t timeout_abs_ns;
1391 uint32_t flags;
1392} __attribute__((packed));
1393
1394static bool kvm_xen_hcall_vcpu_op(struct kvm_vcpu *vcpu, bool longmode, int cmd,
1395 int vcpu_id, u64 param, u64 *r)
1396{
1397 struct vcpu_set_singleshot_timer oneshot;
1398 struct x86_exception e;
1399 s64 delta;
1400
1401 if (!kvm_xen_timer_enabled(vcpu))
1402 return false;
1403
1404 switch (cmd) {
1405 case VCPUOP_set_singleshot_timer:
1406 if (vcpu->arch.xen.vcpu_id != vcpu_id) {
1407 *r = -EINVAL;
1408 return true;
1409 }
1410
1411 /*
1412 * The only difference for 32-bit compat is the 4 bytes of
1413 * padding after the interesting part of the structure. So
1414 * for a faithful emulation of Xen we have to *try* to copy
1415 * the padding and return -EFAULT if we can't. Otherwise we
1416 * might as well just have copied the 12-byte 32-bit struct.
1417 */
1418 BUILD_BUG_ON(offsetof(struct compat_vcpu_set_singleshot_timer, timeout_abs_ns) !=
1419 offsetof(struct vcpu_set_singleshot_timer, timeout_abs_ns));
1420 BUILD_BUG_ON(sizeof_field(struct compat_vcpu_set_singleshot_timer, timeout_abs_ns) !=
1421 sizeof_field(struct vcpu_set_singleshot_timer, timeout_abs_ns));
1422 BUILD_BUG_ON(offsetof(struct compat_vcpu_set_singleshot_timer, flags) !=
1423 offsetof(struct vcpu_set_singleshot_timer, flags));
1424 BUILD_BUG_ON(sizeof_field(struct compat_vcpu_set_singleshot_timer, flags) !=
1425 sizeof_field(struct vcpu_set_singleshot_timer, flags));
1426
1427 if (kvm_read_guest_virt(vcpu, param, &oneshot, longmode ? sizeof(oneshot) :
1428 sizeof(struct compat_vcpu_set_singleshot_timer), &e)) {
1429 *r = -EFAULT;
1430 return true;
1431 }
1432
1433 /* A delta <= 0 results in an immediate callback, which is what we want */
1434 delta = oneshot.timeout_abs_ns - get_kvmclock_ns(vcpu->kvm);
1435 kvm_xen_start_timer(vcpu, oneshot.timeout_abs_ns, delta);
1436 *r = 0;
1437 return true;
1438
1439 case VCPUOP_stop_singleshot_timer:
1440 if (vcpu->arch.xen.vcpu_id != vcpu_id) {
1441 *r = -EINVAL;
1442 return true;
1443 }
1444 kvm_xen_stop_timer(vcpu);
1445 *r = 0;
1446 return true;
1447 }
1448
1449 return false;
1450}
1451
1452static bool kvm_xen_hcall_set_timer_op(struct kvm_vcpu *vcpu, uint64_t timeout,
1453 u64 *r)
1454{
1455 if (!kvm_xen_timer_enabled(vcpu))
1456 return false;
1457
1458 if (timeout) {
1459 uint64_t guest_now = get_kvmclock_ns(vcpu->kvm);
1460 int64_t delta = timeout - guest_now;
1461
1462 /* Xen has a 'Linux workaround' in do_set_timer_op() which
1463 * checks for negative absolute timeout values (caused by
1464 * integer overflow), and for values about 13 days in the
1465 * future (2^50ns) which would be caused by jiffies
1466 * overflow. For those cases, it sets the timeout 100ms in
1467 * the future (not *too* soon, since if a guest really did
1468 * set a long timeout on purpose we don't want to keep
1469 * churning CPU time by waking it up).
1470 */
1471 if (unlikely((int64_t)timeout < 0 ||
1472 (delta > 0 && (uint32_t) (delta >> 50) != 0))) {
1473 delta = 100 * NSEC_PER_MSEC;
1474 timeout = guest_now + delta;
1475 }
1476
1477 kvm_xen_start_timer(vcpu, timeout, delta);
1478 } else {
1479 kvm_xen_stop_timer(vcpu);
1480 }
1481
1482 *r = 0;
1483 return true;
1484}
1485
1486int kvm_xen_hypercall(struct kvm_vcpu *vcpu)
1487{
1488 bool longmode;
1489 u64 input, params[6], r = -ENOSYS;
1490 bool handled = false;
1491 u8 cpl;
1492
1493 input = (u64)kvm_register_read(vcpu, VCPU_REGS_RAX);
1494
1495 /* Hyper-V hypercalls get bit 31 set in EAX */
1496 if ((input & 0x80000000) &&
1497 kvm_hv_hypercall_enabled(vcpu))
1498 return kvm_hv_hypercall(vcpu);
1499
1500 longmode = is_64_bit_hypercall(vcpu);
1501 if (!longmode) {
1502 params[0] = (u32)kvm_rbx_read(vcpu);
1503 params[1] = (u32)kvm_rcx_read(vcpu);
1504 params[2] = (u32)kvm_rdx_read(vcpu);
1505 params[3] = (u32)kvm_rsi_read(vcpu);
1506 params[4] = (u32)kvm_rdi_read(vcpu);
1507 params[5] = (u32)kvm_rbp_read(vcpu);
1508 }
1509#ifdef CONFIG_X86_64
1510 else {
1511 params[0] = (u64)kvm_rdi_read(vcpu);
1512 params[1] = (u64)kvm_rsi_read(vcpu);
1513 params[2] = (u64)kvm_rdx_read(vcpu);
1514 params[3] = (u64)kvm_r10_read(vcpu);
1515 params[4] = (u64)kvm_r8_read(vcpu);
1516 params[5] = (u64)kvm_r9_read(vcpu);
1517 }
1518#endif
1519 cpl = static_call(kvm_x86_get_cpl)(vcpu);
1520 trace_kvm_xen_hypercall(cpl, input, params[0], params[1], params[2],
1521 params[3], params[4], params[5]);
1522
1523 /*
1524 * Only allow hypercall acceleration for CPL0. The rare hypercalls that
1525 * are permitted in guest userspace can be handled by the VMM.
1526 */
1527 if (unlikely(cpl > 0))
1528 goto handle_in_userspace;
1529
1530 switch (input) {
1531 case __HYPERVISOR_xen_version:
1532 if (params[0] == XENVER_version && vcpu->kvm->arch.xen.xen_version) {
1533 r = vcpu->kvm->arch.xen.xen_version;
1534 handled = true;
1535 }
1536 break;
1537 case __HYPERVISOR_event_channel_op:
1538 if (params[0] == EVTCHNOP_send)
1539 handled = kvm_xen_hcall_evtchn_send(vcpu, params[1], &r);
1540 break;
1541 case __HYPERVISOR_sched_op:
1542 handled = kvm_xen_hcall_sched_op(vcpu, longmode, params[0],
1543 params[1], &r);
1544 break;
1545 case __HYPERVISOR_vcpu_op:
1546 handled = kvm_xen_hcall_vcpu_op(vcpu, longmode, params[0], params[1],
1547 params[2], &r);
1548 break;
1549 case __HYPERVISOR_set_timer_op: {
1550 u64 timeout = params[0];
1551 /* In 32-bit mode, the 64-bit timeout is in two 32-bit params. */
1552 if (!longmode)
1553 timeout |= params[1] << 32;
1554 handled = kvm_xen_hcall_set_timer_op(vcpu, timeout, &r);
1555 break;
1556 }
1557 default:
1558 break;
1559 }
1560
1561 if (handled)
1562 return kvm_xen_hypercall_set_result(vcpu, r);
1563
1564handle_in_userspace:
1565 vcpu->run->exit_reason = KVM_EXIT_XEN;
1566 vcpu->run->xen.type = KVM_EXIT_XEN_HCALL;
1567 vcpu->run->xen.u.hcall.longmode = longmode;
1568 vcpu->run->xen.u.hcall.cpl = cpl;
1569 vcpu->run->xen.u.hcall.input = input;
1570 vcpu->run->xen.u.hcall.params[0] = params[0];
1571 vcpu->run->xen.u.hcall.params[1] = params[1];
1572 vcpu->run->xen.u.hcall.params[2] = params[2];
1573 vcpu->run->xen.u.hcall.params[3] = params[3];
1574 vcpu->run->xen.u.hcall.params[4] = params[4];
1575 vcpu->run->xen.u.hcall.params[5] = params[5];
1576 vcpu->arch.xen.hypercall_rip = kvm_get_linear_rip(vcpu);
1577 vcpu->arch.complete_userspace_io =
1578 kvm_xen_hypercall_complete_userspace;
1579
1580 return 0;
1581}
1582
1583static void kvm_xen_check_poller(struct kvm_vcpu *vcpu, int port)
1584{
1585 int poll_evtchn = vcpu->arch.xen.poll_evtchn;
1586
1587 if ((poll_evtchn == port || poll_evtchn == -1) &&
1588 test_and_clear_bit(vcpu->vcpu_idx, vcpu->kvm->arch.xen.poll_mask)) {
1589 kvm_make_request(KVM_REQ_UNBLOCK, vcpu);
1590 kvm_vcpu_kick(vcpu);
1591 }
1592}
1593
1594/*
1595 * The return value from this function is propagated to kvm_set_irq() API,
1596 * so it returns:
1597 * < 0 Interrupt was ignored (masked or not delivered for other reasons)
1598 * = 0 Interrupt was coalesced (previous irq is still pending)
1599 * > 0 Number of CPUs interrupt was delivered to
1600 *
1601 * It is also called directly from kvm_arch_set_irq_inatomic(), where the
1602 * only check on its return value is a comparison with -EWOULDBLOCK'.
1603 */
1604int kvm_xen_set_evtchn_fast(struct kvm_xen_evtchn *xe, struct kvm *kvm)
1605{
1606 struct gfn_to_pfn_cache *gpc = &kvm->arch.xen.shinfo_cache;
1607 struct kvm_vcpu *vcpu;
1608 unsigned long *pending_bits, *mask_bits;
1609 unsigned long flags;
1610 int port_word_bit;
1611 bool kick_vcpu = false;
1612 int vcpu_idx, idx, rc;
1613
1614 vcpu_idx = READ_ONCE(xe->vcpu_idx);
1615 if (vcpu_idx >= 0)
1616 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1617 else {
1618 vcpu = kvm_get_vcpu_by_id(kvm, xe->vcpu_id);
1619 if (!vcpu)
1620 return -EINVAL;
1621 WRITE_ONCE(xe->vcpu_idx, vcpu->vcpu_idx);
1622 }
1623
1624 if (!vcpu->arch.xen.vcpu_info_cache.active)
1625 return -EINVAL;
1626
1627 if (xe->port >= max_evtchn_port(kvm))
1628 return -EINVAL;
1629
1630 rc = -EWOULDBLOCK;
1631
1632 idx = srcu_read_lock(&kvm->srcu);
1633
1634 read_lock_irqsave(&gpc->lock, flags);
1635 if (!kvm_gpc_check(gpc, PAGE_SIZE))
1636 goto out_rcu;
1637
1638 if (IS_ENABLED(CONFIG_64BIT) && kvm->arch.xen.long_mode) {
1639 struct shared_info *shinfo = gpc->khva;
1640 pending_bits = (unsigned long *)&shinfo->evtchn_pending;
1641 mask_bits = (unsigned long *)&shinfo->evtchn_mask;
1642 port_word_bit = xe->port / 64;
1643 } else {
1644 struct compat_shared_info *shinfo = gpc->khva;
1645 pending_bits = (unsigned long *)&shinfo->evtchn_pending;
1646 mask_bits = (unsigned long *)&shinfo->evtchn_mask;
1647 port_word_bit = xe->port / 32;
1648 }
1649
1650 /*
1651 * If this port wasn't already set, and if it isn't masked, then
1652 * we try to set the corresponding bit in the in-kernel shadow of
1653 * evtchn_pending_sel for the target vCPU. And if *that* wasn't
1654 * already set, then we kick the vCPU in question to write to the
1655 * *real* evtchn_pending_sel in its own guest vcpu_info struct.
1656 */
1657 if (test_and_set_bit(xe->port, pending_bits)) {
1658 rc = 0; /* It was already raised */
1659 } else if (test_bit(xe->port, mask_bits)) {
1660 rc = -ENOTCONN; /* Masked */
1661 kvm_xen_check_poller(vcpu, xe->port);
1662 } else {
1663 rc = 1; /* Delivered to the bitmap in shared_info. */
1664 /* Now switch to the vCPU's vcpu_info to set the index and pending_sel */
1665 read_unlock_irqrestore(&gpc->lock, flags);
1666 gpc = &vcpu->arch.xen.vcpu_info_cache;
1667
1668 read_lock_irqsave(&gpc->lock, flags);
1669 if (!kvm_gpc_check(gpc, sizeof(struct vcpu_info))) {
1670 /*
1671 * Could not access the vcpu_info. Set the bit in-kernel
1672 * and prod the vCPU to deliver it for itself.
1673 */
1674 if (!test_and_set_bit(port_word_bit, &vcpu->arch.xen.evtchn_pending_sel))
1675 kick_vcpu = true;
1676 goto out_rcu;
1677 }
1678
1679 if (IS_ENABLED(CONFIG_64BIT) && kvm->arch.xen.long_mode) {
1680 struct vcpu_info *vcpu_info = gpc->khva;
1681 if (!test_and_set_bit(port_word_bit, &vcpu_info->evtchn_pending_sel)) {
1682 WRITE_ONCE(vcpu_info->evtchn_upcall_pending, 1);
1683 kick_vcpu = true;
1684 }
1685 } else {
1686 struct compat_vcpu_info *vcpu_info = gpc->khva;
1687 if (!test_and_set_bit(port_word_bit,
1688 (unsigned long *)&vcpu_info->evtchn_pending_sel)) {
1689 WRITE_ONCE(vcpu_info->evtchn_upcall_pending, 1);
1690 kick_vcpu = true;
1691 }
1692 }
1693
1694 /* For the per-vCPU lapic vector, deliver it as MSI. */
1695 if (kick_vcpu && vcpu->arch.xen.upcall_vector) {
1696 kvm_xen_inject_vcpu_vector(vcpu);
1697 kick_vcpu = false;
1698 }
1699 }
1700
1701 out_rcu:
1702 read_unlock_irqrestore(&gpc->lock, flags);
1703 srcu_read_unlock(&kvm->srcu, idx);
1704
1705 if (kick_vcpu) {
1706 kvm_make_request(KVM_REQ_UNBLOCK, vcpu);
1707 kvm_vcpu_kick(vcpu);
1708 }
1709
1710 return rc;
1711}
1712
1713static int kvm_xen_set_evtchn(struct kvm_xen_evtchn *xe, struct kvm *kvm)
1714{
1715 bool mm_borrowed = false;
1716 int rc;
1717
1718 rc = kvm_xen_set_evtchn_fast(xe, kvm);
1719 if (rc != -EWOULDBLOCK)
1720 return rc;
1721
1722 if (current->mm != kvm->mm) {
1723 /*
1724 * If not on a thread which already belongs to this KVM,
1725 * we'd better be in the irqfd workqueue.
1726 */
1727 if (WARN_ON_ONCE(current->mm))
1728 return -EINVAL;
1729
1730 kthread_use_mm(kvm->mm);
1731 mm_borrowed = true;
1732 }
1733
1734 mutex_lock(&kvm->arch.xen.xen_lock);
1735
1736 /*
1737 * It is theoretically possible for the page to be unmapped
1738 * and the MMU notifier to invalidate the shared_info before
1739 * we even get to use it. In that case, this looks like an
1740 * infinite loop. It was tempting to do it via the userspace
1741 * HVA instead... but that just *hides* the fact that it's
1742 * an infinite loop, because if a fault occurs and it waits
1743 * for the page to come back, it can *still* immediately
1744 * fault and have to wait again, repeatedly.
1745 *
1746 * Conversely, the page could also have been reinstated by
1747 * another thread before we even obtain the mutex above, so
1748 * check again *first* before remapping it.
1749 */
1750 do {
1751 struct gfn_to_pfn_cache *gpc = &kvm->arch.xen.shinfo_cache;
1752 int idx;
1753
1754 rc = kvm_xen_set_evtchn_fast(xe, kvm);
1755 if (rc != -EWOULDBLOCK)
1756 break;
1757
1758 idx = srcu_read_lock(&kvm->srcu);
1759 rc = kvm_gpc_refresh(gpc, PAGE_SIZE);
1760 srcu_read_unlock(&kvm->srcu, idx);
1761 } while(!rc);
1762
1763 mutex_unlock(&kvm->arch.xen.xen_lock);
1764
1765 if (mm_borrowed)
1766 kthread_unuse_mm(kvm->mm);
1767
1768 return rc;
1769}
1770
1771/* This is the version called from kvm_set_irq() as the .set function */
1772static int evtchn_set_fn(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm,
1773 int irq_source_id, int level, bool line_status)
1774{
1775 if (!level)
1776 return -EINVAL;
1777
1778 return kvm_xen_set_evtchn(&e->xen_evtchn, kvm);
1779}
1780
1781/*
1782 * Set up an event channel interrupt from the KVM IRQ routing table.
1783 * Used for e.g. PIRQ from passed through physical devices.
1784 */
1785int kvm_xen_setup_evtchn(struct kvm *kvm,
1786 struct kvm_kernel_irq_routing_entry *e,
1787 const struct kvm_irq_routing_entry *ue)
1788
1789{
1790 struct kvm_vcpu *vcpu;
1791
1792 if (ue->u.xen_evtchn.port >= max_evtchn_port(kvm))
1793 return -EINVAL;
1794
1795 /* We only support 2 level event channels for now */
1796 if (ue->u.xen_evtchn.priority != KVM_IRQ_ROUTING_XEN_EVTCHN_PRIO_2LEVEL)
1797 return -EINVAL;
1798
1799 /*
1800 * Xen gives us interesting mappings from vCPU index to APIC ID,
1801 * which means kvm_get_vcpu_by_id() has to iterate over all vCPUs
1802 * to find it. Do that once at setup time, instead of every time.
1803 * But beware that on live update / live migration, the routing
1804 * table might be reinstated before the vCPU threads have finished
1805 * recreating their vCPUs.
1806 */
1807 vcpu = kvm_get_vcpu_by_id(kvm, ue->u.xen_evtchn.vcpu);
1808 if (vcpu)
1809 e->xen_evtchn.vcpu_idx = vcpu->vcpu_idx;
1810 else
1811 e->xen_evtchn.vcpu_idx = -1;
1812
1813 e->xen_evtchn.port = ue->u.xen_evtchn.port;
1814 e->xen_evtchn.vcpu_id = ue->u.xen_evtchn.vcpu;
1815 e->xen_evtchn.priority = ue->u.xen_evtchn.priority;
1816 e->set = evtchn_set_fn;
1817
1818 return 0;
1819}
1820
1821/*
1822 * Explicit event sending from userspace with KVM_XEN_HVM_EVTCHN_SEND ioctl.
1823 */
1824int kvm_xen_hvm_evtchn_send(struct kvm *kvm, struct kvm_irq_routing_xen_evtchn *uxe)
1825{
1826 struct kvm_xen_evtchn e;
1827 int ret;
1828
1829 if (!uxe->port || uxe->port >= max_evtchn_port(kvm))
1830 return -EINVAL;
1831
1832 /* We only support 2 level event channels for now */
1833 if (uxe->priority != KVM_IRQ_ROUTING_XEN_EVTCHN_PRIO_2LEVEL)
1834 return -EINVAL;
1835
1836 e.port = uxe->port;
1837 e.vcpu_id = uxe->vcpu;
1838 e.vcpu_idx = -1;
1839 e.priority = uxe->priority;
1840
1841 ret = kvm_xen_set_evtchn(&e, kvm);
1842
1843 /*
1844 * None of that 'return 1 if it actually got delivered' nonsense.
1845 * We don't care if it was masked (-ENOTCONN) either.
1846 */
1847 if (ret > 0 || ret == -ENOTCONN)
1848 ret = 0;
1849
1850 return ret;
1851}
1852
1853/*
1854 * Support for *outbound* event channel events via the EVTCHNOP_send hypercall.
1855 */
1856struct evtchnfd {
1857 u32 send_port;
1858 u32 type;
1859 union {
1860 struct kvm_xen_evtchn port;
1861 struct {
1862 u32 port; /* zero */
1863 struct eventfd_ctx *ctx;
1864 } eventfd;
1865 } deliver;
1866};
1867
1868/*
1869 * Update target vCPU or priority for a registered sending channel.
1870 */
1871static int kvm_xen_eventfd_update(struct kvm *kvm,
1872 struct kvm_xen_hvm_attr *data)
1873{
1874 u32 port = data->u.evtchn.send_port;
1875 struct evtchnfd *evtchnfd;
1876 int ret;
1877
1878 /* Protect writes to evtchnfd as well as the idr lookup. */
1879 mutex_lock(&kvm->arch.xen.xen_lock);
1880 evtchnfd = idr_find(&kvm->arch.xen.evtchn_ports, port);
1881
1882 ret = -ENOENT;
1883 if (!evtchnfd)
1884 goto out_unlock;
1885
1886 /* For an UPDATE, nothing may change except the priority/vcpu */
1887 ret = -EINVAL;
1888 if (evtchnfd->type != data->u.evtchn.type)
1889 goto out_unlock;
1890
1891 /*
1892 * Port cannot change, and if it's zero that was an eventfd
1893 * which can't be changed either.
1894 */
1895 if (!evtchnfd->deliver.port.port ||
1896 evtchnfd->deliver.port.port != data->u.evtchn.deliver.port.port)
1897 goto out_unlock;
1898
1899 /* We only support 2 level event channels for now */
1900 if (data->u.evtchn.deliver.port.priority != KVM_IRQ_ROUTING_XEN_EVTCHN_PRIO_2LEVEL)
1901 goto out_unlock;
1902
1903 evtchnfd->deliver.port.priority = data->u.evtchn.deliver.port.priority;
1904 if (evtchnfd->deliver.port.vcpu_id != data->u.evtchn.deliver.port.vcpu) {
1905 evtchnfd->deliver.port.vcpu_id = data->u.evtchn.deliver.port.vcpu;
1906 evtchnfd->deliver.port.vcpu_idx = -1;
1907 }
1908 ret = 0;
1909out_unlock:
1910 mutex_unlock(&kvm->arch.xen.xen_lock);
1911 return ret;
1912}
1913
1914/*
1915 * Configure the target (eventfd or local port delivery) for sending on
1916 * a given event channel.
1917 */
1918static int kvm_xen_eventfd_assign(struct kvm *kvm,
1919 struct kvm_xen_hvm_attr *data)
1920{
1921 u32 port = data->u.evtchn.send_port;
1922 struct eventfd_ctx *eventfd = NULL;
1923 struct evtchnfd *evtchnfd;
1924 int ret = -EINVAL;
1925
1926 evtchnfd = kzalloc(sizeof(struct evtchnfd), GFP_KERNEL);
1927 if (!evtchnfd)
1928 return -ENOMEM;
1929
1930 switch(data->u.evtchn.type) {
1931 case EVTCHNSTAT_ipi:
1932 /* IPI must map back to the same port# */
1933 if (data->u.evtchn.deliver.port.port != data->u.evtchn.send_port)
1934 goto out_noeventfd; /* -EINVAL */
1935 break;
1936
1937 case EVTCHNSTAT_interdomain:
1938 if (data->u.evtchn.deliver.port.port) {
1939 if (data->u.evtchn.deliver.port.port >= max_evtchn_port(kvm))
1940 goto out_noeventfd; /* -EINVAL */
1941 } else {
1942 eventfd = eventfd_ctx_fdget(data->u.evtchn.deliver.eventfd.fd);
1943 if (IS_ERR(eventfd)) {
1944 ret = PTR_ERR(eventfd);
1945 goto out_noeventfd;
1946 }
1947 }
1948 break;
1949
1950 case EVTCHNSTAT_virq:
1951 case EVTCHNSTAT_closed:
1952 case EVTCHNSTAT_unbound:
1953 case EVTCHNSTAT_pirq:
1954 default: /* Unknown event channel type */
1955 goto out; /* -EINVAL */
1956 }
1957
1958 evtchnfd->send_port = data->u.evtchn.send_port;
1959 evtchnfd->type = data->u.evtchn.type;
1960 if (eventfd) {
1961 evtchnfd->deliver.eventfd.ctx = eventfd;
1962 } else {
1963 /* We only support 2 level event channels for now */
1964 if (data->u.evtchn.deliver.port.priority != KVM_IRQ_ROUTING_XEN_EVTCHN_PRIO_2LEVEL)
1965 goto out; /* -EINVAL; */
1966
1967 evtchnfd->deliver.port.port = data->u.evtchn.deliver.port.port;
1968 evtchnfd->deliver.port.vcpu_id = data->u.evtchn.deliver.port.vcpu;
1969 evtchnfd->deliver.port.vcpu_idx = -1;
1970 evtchnfd->deliver.port.priority = data->u.evtchn.deliver.port.priority;
1971 }
1972
1973 mutex_lock(&kvm->arch.xen.xen_lock);
1974 ret = idr_alloc(&kvm->arch.xen.evtchn_ports, evtchnfd, port, port + 1,
1975 GFP_KERNEL);
1976 mutex_unlock(&kvm->arch.xen.xen_lock);
1977 if (ret >= 0)
1978 return 0;
1979
1980 if (ret == -ENOSPC)
1981 ret = -EEXIST;
1982out:
1983 if (eventfd)
1984 eventfd_ctx_put(eventfd);
1985out_noeventfd:
1986 kfree(evtchnfd);
1987 return ret;
1988}
1989
1990static int kvm_xen_eventfd_deassign(struct kvm *kvm, u32 port)
1991{
1992 struct evtchnfd *evtchnfd;
1993
1994 mutex_lock(&kvm->arch.xen.xen_lock);
1995 evtchnfd = idr_remove(&kvm->arch.xen.evtchn_ports, port);
1996 mutex_unlock(&kvm->arch.xen.xen_lock);
1997
1998 if (!evtchnfd)
1999 return -ENOENT;
2000
2001 synchronize_srcu(&kvm->srcu);
2002 if (!evtchnfd->deliver.port.port)
2003 eventfd_ctx_put(evtchnfd->deliver.eventfd.ctx);
2004 kfree(evtchnfd);
2005 return 0;
2006}
2007
2008static int kvm_xen_eventfd_reset(struct kvm *kvm)
2009{
2010 struct evtchnfd *evtchnfd, **all_evtchnfds;
2011 int i;
2012 int n = 0;
2013
2014 mutex_lock(&kvm->arch.xen.xen_lock);
2015
2016 /*
2017 * Because synchronize_srcu() cannot be called inside the
2018 * critical section, first collect all the evtchnfd objects
2019 * in an array as they are removed from evtchn_ports.
2020 */
2021 idr_for_each_entry(&kvm->arch.xen.evtchn_ports, evtchnfd, i)
2022 n++;
2023
2024 all_evtchnfds = kmalloc_array(n, sizeof(struct evtchnfd *), GFP_KERNEL);
2025 if (!all_evtchnfds) {
2026 mutex_unlock(&kvm->arch.xen.xen_lock);
2027 return -ENOMEM;
2028 }
2029
2030 n = 0;
2031 idr_for_each_entry(&kvm->arch.xen.evtchn_ports, evtchnfd, i) {
2032 all_evtchnfds[n++] = evtchnfd;
2033 idr_remove(&kvm->arch.xen.evtchn_ports, evtchnfd->send_port);
2034 }
2035 mutex_unlock(&kvm->arch.xen.xen_lock);
2036
2037 synchronize_srcu(&kvm->srcu);
2038
2039 while (n--) {
2040 evtchnfd = all_evtchnfds[n];
2041 if (!evtchnfd->deliver.port.port)
2042 eventfd_ctx_put(evtchnfd->deliver.eventfd.ctx);
2043 kfree(evtchnfd);
2044 }
2045 kfree(all_evtchnfds);
2046
2047 return 0;
2048}
2049
2050static int kvm_xen_setattr_evtchn(struct kvm *kvm, struct kvm_xen_hvm_attr *data)
2051{
2052 u32 port = data->u.evtchn.send_port;
2053
2054 if (data->u.evtchn.flags == KVM_XEN_EVTCHN_RESET)
2055 return kvm_xen_eventfd_reset(kvm);
2056
2057 if (!port || port >= max_evtchn_port(kvm))
2058 return -EINVAL;
2059
2060 if (data->u.evtchn.flags == KVM_XEN_EVTCHN_DEASSIGN)
2061 return kvm_xen_eventfd_deassign(kvm, port);
2062 if (data->u.evtchn.flags == KVM_XEN_EVTCHN_UPDATE)
2063 return kvm_xen_eventfd_update(kvm, data);
2064 if (data->u.evtchn.flags)
2065 return -EINVAL;
2066
2067 return kvm_xen_eventfd_assign(kvm, data);
2068}
2069
2070static bool kvm_xen_hcall_evtchn_send(struct kvm_vcpu *vcpu, u64 param, u64 *r)
2071{
2072 struct evtchnfd *evtchnfd;
2073 struct evtchn_send send;
2074 struct x86_exception e;
2075
2076 /* Sanity check: this structure is the same for 32-bit and 64-bit */
2077 BUILD_BUG_ON(sizeof(send) != 4);
2078 if (kvm_read_guest_virt(vcpu, param, &send, sizeof(send), &e)) {
2079 *r = -EFAULT;
2080 return true;
2081 }
2082
2083 /*
2084 * evtchnfd is protected by kvm->srcu; the idr lookup instead
2085 * is protected by RCU.
2086 */
2087 rcu_read_lock();
2088 evtchnfd = idr_find(&vcpu->kvm->arch.xen.evtchn_ports, send.port);
2089 rcu_read_unlock();
2090 if (!evtchnfd)
2091 return false;
2092
2093 if (evtchnfd->deliver.port.port) {
2094 int ret = kvm_xen_set_evtchn(&evtchnfd->deliver.port, vcpu->kvm);
2095 if (ret < 0 && ret != -ENOTCONN)
2096 return false;
2097 } else {
2098 eventfd_signal(evtchnfd->deliver.eventfd.ctx);
2099 }
2100
2101 *r = 0;
2102 return true;
2103}
2104
2105void kvm_xen_init_vcpu(struct kvm_vcpu *vcpu)
2106{
2107 vcpu->arch.xen.vcpu_id = vcpu->vcpu_idx;
2108 vcpu->arch.xen.poll_evtchn = 0;
2109
2110 timer_setup(&vcpu->arch.xen.poll_timer, cancel_evtchn_poll, 0);
2111
2112 kvm_gpc_init(&vcpu->arch.xen.runstate_cache, vcpu->kvm, NULL,
2113 KVM_HOST_USES_PFN);
2114 kvm_gpc_init(&vcpu->arch.xen.runstate2_cache, vcpu->kvm, NULL,
2115 KVM_HOST_USES_PFN);
2116 kvm_gpc_init(&vcpu->arch.xen.vcpu_info_cache, vcpu->kvm, NULL,
2117 KVM_HOST_USES_PFN);
2118 kvm_gpc_init(&vcpu->arch.xen.vcpu_time_info_cache, vcpu->kvm, NULL,
2119 KVM_HOST_USES_PFN);
2120}
2121
2122void kvm_xen_destroy_vcpu(struct kvm_vcpu *vcpu)
2123{
2124 if (kvm_xen_timer_enabled(vcpu))
2125 kvm_xen_stop_timer(vcpu);
2126
2127 kvm_gpc_deactivate(&vcpu->arch.xen.runstate_cache);
2128 kvm_gpc_deactivate(&vcpu->arch.xen.runstate2_cache);
2129 kvm_gpc_deactivate(&vcpu->arch.xen.vcpu_info_cache);
2130 kvm_gpc_deactivate(&vcpu->arch.xen.vcpu_time_info_cache);
2131
2132 del_timer_sync(&vcpu->arch.xen.poll_timer);
2133}
2134
2135void kvm_xen_update_tsc_info(struct kvm_vcpu *vcpu)
2136{
2137 struct kvm_cpuid_entry2 *entry;
2138 u32 function;
2139
2140 if (!vcpu->arch.xen.cpuid.base)
2141 return;
2142
2143 function = vcpu->arch.xen.cpuid.base | XEN_CPUID_LEAF(3);
2144 if (function > vcpu->arch.xen.cpuid.limit)
2145 return;
2146
2147 entry = kvm_find_cpuid_entry_index(vcpu, function, 1);
2148 if (entry) {
2149 entry->ecx = vcpu->arch.hv_clock.tsc_to_system_mul;
2150 entry->edx = vcpu->arch.hv_clock.tsc_shift;
2151 }
2152
2153 entry = kvm_find_cpuid_entry_index(vcpu, function, 2);
2154 if (entry)
2155 entry->eax = vcpu->arch.hw_tsc_khz;
2156}
2157
2158void kvm_xen_init_vm(struct kvm *kvm)
2159{
2160 mutex_init(&kvm->arch.xen.xen_lock);
2161 idr_init(&kvm->arch.xen.evtchn_ports);
2162 kvm_gpc_init(&kvm->arch.xen.shinfo_cache, kvm, NULL, KVM_HOST_USES_PFN);
2163}
2164
2165void kvm_xen_destroy_vm(struct kvm *kvm)
2166{
2167 struct evtchnfd *evtchnfd;
2168 int i;
2169
2170 kvm_gpc_deactivate(&kvm->arch.xen.shinfo_cache);
2171
2172 idr_for_each_entry(&kvm->arch.xen.evtchn_ports, evtchnfd, i) {
2173 if (!evtchnfd->deliver.port.port)
2174 eventfd_ctx_put(evtchnfd->deliver.eventfd.ctx);
2175 kfree(evtchnfd);
2176 }
2177 idr_destroy(&kvm->arch.xen.evtchn_ports);
2178
2179 if (kvm->arch.xen_hvm_config.msr)
2180 static_branch_slow_dec_deferred(&kvm_xen_enabled);
2181}