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 = ktime_get_real_ns() - get_kvmclock_ns(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, 1000000000); 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 if (atomic_read(&vcpu->arch.xen.timer_pending)) 138 return HRTIMER_NORESTART; 139 140 atomic_inc(&vcpu->arch.xen.timer_pending); 141 kvm_make_request(KVM_REQ_UNBLOCK, vcpu); 142 kvm_vcpu_kick(vcpu); 143 144 return HRTIMER_NORESTART; 145} 146 147static void kvm_xen_start_timer(struct kvm_vcpu *vcpu, u64 guest_abs, s64 delta_ns) 148{ 149 atomic_set(&vcpu->arch.xen.timer_pending, 0); 150 vcpu->arch.xen.timer_expires = guest_abs; 151 152 if (delta_ns <= 0) { 153 xen_timer_callback(&vcpu->arch.xen.timer); 154 } else { 155 ktime_t ktime_now = ktime_get(); 156 hrtimer_start(&vcpu->arch.xen.timer, 157 ktime_add_ns(ktime_now, delta_ns), 158 HRTIMER_MODE_ABS_HARD); 159 } 160} 161 162static void kvm_xen_stop_timer(struct kvm_vcpu *vcpu) 163{ 164 hrtimer_cancel(&vcpu->arch.xen.timer); 165 vcpu->arch.xen.timer_expires = 0; 166 atomic_set(&vcpu->arch.xen.timer_pending, 0); 167} 168 169static void kvm_xen_init_timer(struct kvm_vcpu *vcpu) 170{ 171 hrtimer_init(&vcpu->arch.xen.timer, CLOCK_MONOTONIC, 172 HRTIMER_MODE_ABS_HARD); 173 vcpu->arch.xen.timer.function = xen_timer_callback; 174} 175 176static void kvm_xen_update_runstate_guest(struct kvm_vcpu *v, bool atomic) 177{ 178 struct kvm_vcpu_xen *vx = &v->arch.xen; 179 struct gfn_to_pfn_cache *gpc1 = &vx->runstate_cache; 180 struct gfn_to_pfn_cache *gpc2 = &vx->runstate2_cache; 181 size_t user_len, user_len1, user_len2; 182 struct vcpu_runstate_info rs; 183 unsigned long flags; 184 size_t times_ofs; 185 uint8_t *update_bit = NULL; 186 uint64_t entry_time; 187 uint64_t *rs_times; 188 int *rs_state; 189 190 /* 191 * The only difference between 32-bit and 64-bit versions of the 192 * runstate struct is the alignment of uint64_t in 32-bit, which 193 * means that the 64-bit version has an additional 4 bytes of 194 * padding after the first field 'state'. Let's be really really 195 * paranoid about that, and matching it with our internal data 196 * structures that we memcpy into it... 197 */ 198 BUILD_BUG_ON(offsetof(struct vcpu_runstate_info, state) != 0); 199 BUILD_BUG_ON(offsetof(struct compat_vcpu_runstate_info, state) != 0); 200 BUILD_BUG_ON(sizeof(struct compat_vcpu_runstate_info) != 0x2c); 201#ifdef CONFIG_X86_64 202 /* 203 * The 64-bit structure has 4 bytes of padding before 'state_entry_time' 204 * so each subsequent field is shifted by 4, and it's 4 bytes longer. 205 */ 206 BUILD_BUG_ON(offsetof(struct vcpu_runstate_info, state_entry_time) != 207 offsetof(struct compat_vcpu_runstate_info, state_entry_time) + 4); 208 BUILD_BUG_ON(offsetof(struct vcpu_runstate_info, time) != 209 offsetof(struct compat_vcpu_runstate_info, time) + 4); 210 BUILD_BUG_ON(sizeof(struct vcpu_runstate_info) != 0x2c + 4); 211#endif 212 /* 213 * The state field is in the same place at the start of both structs, 214 * and is the same size (int) as vx->current_runstate. 215 */ 216 BUILD_BUG_ON(offsetof(struct vcpu_runstate_info, state) != 217 offsetof(struct compat_vcpu_runstate_info, state)); 218 BUILD_BUG_ON(sizeof_field(struct vcpu_runstate_info, state) != 219 sizeof(vx->current_runstate)); 220 BUILD_BUG_ON(sizeof_field(struct compat_vcpu_runstate_info, state) != 221 sizeof(vx->current_runstate)); 222 223 /* 224 * The state_entry_time field is 64 bits in both versions, and the 225 * XEN_RUNSTATE_UPDATE flag is in the top bit, which given that x86 226 * is little-endian means that it's in the last *byte* of the word. 227 * That detail is important later. 228 */ 229 BUILD_BUG_ON(sizeof_field(struct vcpu_runstate_info, state_entry_time) != 230 sizeof(uint64_t)); 231 BUILD_BUG_ON(sizeof_field(struct compat_vcpu_runstate_info, state_entry_time) != 232 sizeof(uint64_t)); 233 BUILD_BUG_ON((XEN_RUNSTATE_UPDATE >> 56) != 0x80); 234 235 /* 236 * The time array is four 64-bit quantities in both versions, matching 237 * the vx->runstate_times and immediately following state_entry_time. 238 */ 239 BUILD_BUG_ON(offsetof(struct vcpu_runstate_info, state_entry_time) != 240 offsetof(struct vcpu_runstate_info, time) - sizeof(uint64_t)); 241 BUILD_BUG_ON(offsetof(struct compat_vcpu_runstate_info, state_entry_time) != 242 offsetof(struct compat_vcpu_runstate_info, time) - sizeof(uint64_t)); 243 BUILD_BUG_ON(sizeof_field(struct vcpu_runstate_info, time) != 244 sizeof_field(struct compat_vcpu_runstate_info, time)); 245 BUILD_BUG_ON(sizeof_field(struct vcpu_runstate_info, time) != 246 sizeof(vx->runstate_times)); 247 248 if (IS_ENABLED(CONFIG_64BIT) && v->kvm->arch.xen.long_mode) { 249 user_len = sizeof(struct vcpu_runstate_info); 250 times_ofs = offsetof(struct vcpu_runstate_info, 251 state_entry_time); 252 } else { 253 user_len = sizeof(struct compat_vcpu_runstate_info); 254 times_ofs = offsetof(struct compat_vcpu_runstate_info, 255 state_entry_time); 256 } 257 258 /* 259 * There are basically no alignment constraints. The guest can set it 260 * up so it crosses from one page to the next, and at arbitrary byte 261 * alignment (and the 32-bit ABI doesn't align the 64-bit integers 262 * anyway, even if the overall struct had been 64-bit aligned). 263 */ 264 if ((gpc1->gpa & ~PAGE_MASK) + user_len >= PAGE_SIZE) { 265 user_len1 = PAGE_SIZE - (gpc1->gpa & ~PAGE_MASK); 266 user_len2 = user_len - user_len1; 267 } else { 268 user_len1 = user_len; 269 user_len2 = 0; 270 } 271 BUG_ON(user_len1 + user_len2 != user_len); 272 273 retry: 274 /* 275 * Attempt to obtain the GPC lock on *both* (if there are two) 276 * gfn_to_pfn caches that cover the region. 277 */ 278 if (atomic) { 279 local_irq_save(flags); 280 if (!read_trylock(&gpc1->lock)) { 281 local_irq_restore(flags); 282 return; 283 } 284 } else { 285 read_lock_irqsave(&gpc1->lock, flags); 286 } 287 while (!kvm_gpc_check(gpc1, user_len1)) { 288 read_unlock_irqrestore(&gpc1->lock, flags); 289 290 /* When invoked from kvm_sched_out() we cannot sleep */ 291 if (atomic) 292 return; 293 294 if (kvm_gpc_refresh(gpc1, user_len1)) 295 return; 296 297 read_lock_irqsave(&gpc1->lock, flags); 298 } 299 300 if (likely(!user_len2)) { 301 /* 302 * Set up three pointers directly to the runstate_info 303 * struct in the guest (via the GPC). 304 * 305 * • @rs_state → state field 306 * • @rs_times → state_entry_time field. 307 * • @update_bit → last byte of state_entry_time, which 308 * contains the XEN_RUNSTATE_UPDATE bit. 309 */ 310 rs_state = gpc1->khva; 311 rs_times = gpc1->khva + times_ofs; 312 if (v->kvm->arch.xen.runstate_update_flag) 313 update_bit = ((void *)(&rs_times[1])) - 1; 314 } else { 315 /* 316 * The guest's runstate_info is split across two pages and we 317 * need to hold and validate both GPCs simultaneously. We can 318 * declare a lock ordering GPC1 > GPC2 because nothing else 319 * takes them more than one at a time. Set a subclass on the 320 * gpc1 lock to make lockdep shut up about it. 321 */ 322 lock_set_subclass(&gpc1->lock.dep_map, 1, _THIS_IP_); 323 if (atomic) { 324 if (!read_trylock(&gpc2->lock)) { 325 read_unlock_irqrestore(&gpc1->lock, flags); 326 return; 327 } 328 } else { 329 read_lock(&gpc2->lock); 330 } 331 332 if (!kvm_gpc_check(gpc2, user_len2)) { 333 read_unlock(&gpc2->lock); 334 read_unlock_irqrestore(&gpc1->lock, flags); 335 336 /* When invoked from kvm_sched_out() we cannot sleep */ 337 if (atomic) 338 return; 339 340 /* 341 * Use kvm_gpc_activate() here because if the runstate 342 * area was configured in 32-bit mode and only extends 343 * to the second page now because the guest changed to 344 * 64-bit mode, the second GPC won't have been set up. 345 */ 346 if (kvm_gpc_activate(gpc2, gpc1->gpa + user_len1, 347 user_len2)) 348 return; 349 350 /* 351 * We dropped the lock on GPC1 so we have to go all the 352 * way back and revalidate that too. 353 */ 354 goto retry; 355 } 356 357 /* 358 * In this case, the runstate_info struct will be assembled on 359 * the kernel stack (compat or not as appropriate) and will 360 * be copied to GPC1/GPC2 with a dual memcpy. Set up the three 361 * rs pointers accordingly. 362 */ 363 rs_times = &rs.state_entry_time; 364 365 /* 366 * The rs_state pointer points to the start of what we'll 367 * copy to the guest, which in the case of a compat guest 368 * is the 32-bit field that the compiler thinks is padding. 369 */ 370 rs_state = ((void *)rs_times) - times_ofs; 371 372 /* 373 * The update_bit is still directly in the guest memory, 374 * via one GPC or the other. 375 */ 376 if (v->kvm->arch.xen.runstate_update_flag) { 377 if (user_len1 >= times_ofs + sizeof(uint64_t)) 378 update_bit = gpc1->khva + times_ofs + 379 sizeof(uint64_t) - 1; 380 else 381 update_bit = gpc2->khva + times_ofs + 382 sizeof(uint64_t) - 1 - user_len1; 383 } 384 385#ifdef CONFIG_X86_64 386 /* 387 * Don't leak kernel memory through the padding in the 64-bit 388 * version of the struct. 389 */ 390 memset(&rs, 0, offsetof(struct vcpu_runstate_info, state_entry_time)); 391#endif 392 } 393 394 /* 395 * First, set the XEN_RUNSTATE_UPDATE bit in the top bit of the 396 * state_entry_time field, directly in the guest. We need to set 397 * that (and write-barrier) before writing to the rest of the 398 * structure, and clear it last. Just as Xen does, we address the 399 * single *byte* in which it resides because it might be in a 400 * different cache line to the rest of the 64-bit word, due to 401 * the (lack of) alignment constraints. 402 */ 403 entry_time = vx->runstate_entry_time; 404 if (update_bit) { 405 entry_time |= XEN_RUNSTATE_UPDATE; 406 *update_bit = (vx->runstate_entry_time | XEN_RUNSTATE_UPDATE) >> 56; 407 smp_wmb(); 408 } 409 410 /* 411 * Now assemble the actual structure, either on our kernel stack 412 * or directly in the guest according to how the rs_state and 413 * rs_times pointers were set up above. 414 */ 415 *rs_state = vx->current_runstate; 416 rs_times[0] = entry_time; 417 memcpy(rs_times + 1, vx->runstate_times, sizeof(vx->runstate_times)); 418 419 /* For the split case, we have to then copy it to the guest. */ 420 if (user_len2) { 421 memcpy(gpc1->khva, rs_state, user_len1); 422 memcpy(gpc2->khva, ((void *)rs_state) + user_len1, user_len2); 423 } 424 smp_wmb(); 425 426 /* Finally, clear the XEN_RUNSTATE_UPDATE bit. */ 427 if (update_bit) { 428 entry_time &= ~XEN_RUNSTATE_UPDATE; 429 *update_bit = entry_time >> 56; 430 smp_wmb(); 431 } 432 433 if (user_len2) 434 read_unlock(&gpc2->lock); 435 436 read_unlock_irqrestore(&gpc1->lock, flags); 437 438 mark_page_dirty_in_slot(v->kvm, gpc1->memslot, gpc1->gpa >> PAGE_SHIFT); 439 if (user_len2) 440 mark_page_dirty_in_slot(v->kvm, gpc2->memslot, gpc2->gpa >> PAGE_SHIFT); 441} 442 443void kvm_xen_update_runstate(struct kvm_vcpu *v, int state) 444{ 445 struct kvm_vcpu_xen *vx = &v->arch.xen; 446 u64 now = get_kvmclock_ns(v->kvm); 447 u64 delta_ns = now - vx->runstate_entry_time; 448 u64 run_delay = current->sched_info.run_delay; 449 450 if (unlikely(!vx->runstate_entry_time)) 451 vx->current_runstate = RUNSTATE_offline; 452 453 /* 454 * Time waiting for the scheduler isn't "stolen" if the 455 * vCPU wasn't running anyway. 456 */ 457 if (vx->current_runstate == RUNSTATE_running) { 458 u64 steal_ns = run_delay - vx->last_steal; 459 460 delta_ns -= steal_ns; 461 462 vx->runstate_times[RUNSTATE_runnable] += steal_ns; 463 } 464 vx->last_steal = run_delay; 465 466 vx->runstate_times[vx->current_runstate] += delta_ns; 467 vx->current_runstate = state; 468 vx->runstate_entry_time = now; 469 470 if (vx->runstate_cache.active) 471 kvm_xen_update_runstate_guest(v, state == RUNSTATE_runnable); 472} 473 474void kvm_xen_inject_vcpu_vector(struct kvm_vcpu *v) 475{ 476 struct kvm_lapic_irq irq = { }; 477 int r; 478 479 irq.dest_id = v->vcpu_id; 480 irq.vector = v->arch.xen.upcall_vector; 481 irq.dest_mode = APIC_DEST_PHYSICAL; 482 irq.shorthand = APIC_DEST_NOSHORT; 483 irq.delivery_mode = APIC_DM_FIXED; 484 irq.level = 1; 485 486 /* The fast version will always work for physical unicast */ 487 WARN_ON_ONCE(!kvm_irq_delivery_to_apic_fast(v->kvm, NULL, &irq, &r, NULL)); 488} 489 490/* 491 * On event channel delivery, the vcpu_info may not have been accessible. 492 * In that case, there are bits in vcpu->arch.xen.evtchn_pending_sel which 493 * need to be marked into the vcpu_info (and evtchn_upcall_pending set). 494 * Do so now that we can sleep in the context of the vCPU to bring the 495 * page in, and refresh the pfn cache for it. 496 */ 497void kvm_xen_inject_pending_events(struct kvm_vcpu *v) 498{ 499 unsigned long evtchn_pending_sel = READ_ONCE(v->arch.xen.evtchn_pending_sel); 500 struct gfn_to_pfn_cache *gpc = &v->arch.xen.vcpu_info_cache; 501 unsigned long flags; 502 503 if (!evtchn_pending_sel) 504 return; 505 506 /* 507 * Yes, this is an open-coded loop. But that's just what put_user() 508 * does anyway. Page it in and retry the instruction. We're just a 509 * little more honest about it. 510 */ 511 read_lock_irqsave(&gpc->lock, flags); 512 while (!kvm_gpc_check(gpc, sizeof(struct vcpu_info))) { 513 read_unlock_irqrestore(&gpc->lock, flags); 514 515 if (kvm_gpc_refresh(gpc, sizeof(struct vcpu_info))) 516 return; 517 518 read_lock_irqsave(&gpc->lock, flags); 519 } 520 521 /* Now gpc->khva is a valid kernel address for the vcpu_info */ 522 if (IS_ENABLED(CONFIG_64BIT) && v->kvm->arch.xen.long_mode) { 523 struct vcpu_info *vi = gpc->khva; 524 525 asm volatile(LOCK_PREFIX "orq %0, %1\n" 526 "notq %0\n" 527 LOCK_PREFIX "andq %0, %2\n" 528 : "=r" (evtchn_pending_sel), 529 "+m" (vi->evtchn_pending_sel), 530 "+m" (v->arch.xen.evtchn_pending_sel) 531 : "0" (evtchn_pending_sel)); 532 WRITE_ONCE(vi->evtchn_upcall_pending, 1); 533 } else { 534 u32 evtchn_pending_sel32 = evtchn_pending_sel; 535 struct compat_vcpu_info *vi = gpc->khva; 536 537 asm volatile(LOCK_PREFIX "orl %0, %1\n" 538 "notl %0\n" 539 LOCK_PREFIX "andl %0, %2\n" 540 : "=r" (evtchn_pending_sel32), 541 "+m" (vi->evtchn_pending_sel), 542 "+m" (v->arch.xen.evtchn_pending_sel) 543 : "0" (evtchn_pending_sel32)); 544 WRITE_ONCE(vi->evtchn_upcall_pending, 1); 545 } 546 read_unlock_irqrestore(&gpc->lock, flags); 547 548 /* For the per-vCPU lapic vector, deliver it as MSI. */ 549 if (v->arch.xen.upcall_vector) 550 kvm_xen_inject_vcpu_vector(v); 551 552 mark_page_dirty_in_slot(v->kvm, gpc->memslot, gpc->gpa >> PAGE_SHIFT); 553} 554 555int __kvm_xen_has_interrupt(struct kvm_vcpu *v) 556{ 557 struct gfn_to_pfn_cache *gpc = &v->arch.xen.vcpu_info_cache; 558 unsigned long flags; 559 u8 rc = 0; 560 561 /* 562 * If the global upcall vector (HVMIRQ_callback_vector) is set and 563 * the vCPU's evtchn_upcall_pending flag is set, the IRQ is pending. 564 */ 565 566 /* No need for compat handling here */ 567 BUILD_BUG_ON(offsetof(struct vcpu_info, evtchn_upcall_pending) != 568 offsetof(struct compat_vcpu_info, evtchn_upcall_pending)); 569 BUILD_BUG_ON(sizeof(rc) != 570 sizeof_field(struct vcpu_info, evtchn_upcall_pending)); 571 BUILD_BUG_ON(sizeof(rc) != 572 sizeof_field(struct compat_vcpu_info, evtchn_upcall_pending)); 573 574 read_lock_irqsave(&gpc->lock, flags); 575 while (!kvm_gpc_check(gpc, sizeof(struct vcpu_info))) { 576 read_unlock_irqrestore(&gpc->lock, flags); 577 578 /* 579 * This function gets called from kvm_vcpu_block() after setting the 580 * task to TASK_INTERRUPTIBLE, to see if it needs to wake immediately 581 * from a HLT. So we really mustn't sleep. If the page ended up absent 582 * at that point, just return 1 in order to trigger an immediate wake, 583 * and we'll end up getting called again from a context where we *can* 584 * fault in the page and wait for it. 585 */ 586 if (in_atomic() || !task_is_running(current)) 587 return 1; 588 589 if (kvm_gpc_refresh(gpc, sizeof(struct vcpu_info))) { 590 /* 591 * If this failed, userspace has screwed up the 592 * vcpu_info mapping. No interrupts for you. 593 */ 594 return 0; 595 } 596 read_lock_irqsave(&gpc->lock, flags); 597 } 598 599 rc = ((struct vcpu_info *)gpc->khva)->evtchn_upcall_pending; 600 read_unlock_irqrestore(&gpc->lock, flags); 601 return rc; 602} 603 604int kvm_xen_hvm_set_attr(struct kvm *kvm, struct kvm_xen_hvm_attr *data) 605{ 606 int r = -ENOENT; 607 608 609 switch (data->type) { 610 case KVM_XEN_ATTR_TYPE_LONG_MODE: 611 if (!IS_ENABLED(CONFIG_64BIT) && data->u.long_mode) { 612 r = -EINVAL; 613 } else { 614 mutex_lock(&kvm->arch.xen.xen_lock); 615 kvm->arch.xen.long_mode = !!data->u.long_mode; 616 mutex_unlock(&kvm->arch.xen.xen_lock); 617 r = 0; 618 } 619 break; 620 621 case KVM_XEN_ATTR_TYPE_SHARED_INFO: 622 mutex_lock(&kvm->arch.xen.xen_lock); 623 r = kvm_xen_shared_info_init(kvm, data->u.shared_info.gfn); 624 mutex_unlock(&kvm->arch.xen.xen_lock); 625 break; 626 627 case KVM_XEN_ATTR_TYPE_UPCALL_VECTOR: 628 if (data->u.vector && data->u.vector < 0x10) 629 r = -EINVAL; 630 else { 631 mutex_lock(&kvm->arch.xen.xen_lock); 632 kvm->arch.xen.upcall_vector = data->u.vector; 633 mutex_unlock(&kvm->arch.xen.xen_lock); 634 r = 0; 635 } 636 break; 637 638 case KVM_XEN_ATTR_TYPE_EVTCHN: 639 r = kvm_xen_setattr_evtchn(kvm, data); 640 break; 641 642 case KVM_XEN_ATTR_TYPE_XEN_VERSION: 643 mutex_lock(&kvm->arch.xen.xen_lock); 644 kvm->arch.xen.xen_version = data->u.xen_version; 645 mutex_unlock(&kvm->arch.xen.xen_lock); 646 r = 0; 647 break; 648 649 case KVM_XEN_ATTR_TYPE_RUNSTATE_UPDATE_FLAG: 650 if (!sched_info_on()) { 651 r = -EOPNOTSUPP; 652 break; 653 } 654 mutex_lock(&kvm->arch.xen.xen_lock); 655 kvm->arch.xen.runstate_update_flag = !!data->u.runstate_update_flag; 656 mutex_unlock(&kvm->arch.xen.xen_lock); 657 r = 0; 658 break; 659 660 default: 661 break; 662 } 663 664 return r; 665} 666 667int kvm_xen_hvm_get_attr(struct kvm *kvm, struct kvm_xen_hvm_attr *data) 668{ 669 int r = -ENOENT; 670 671 mutex_lock(&kvm->arch.xen.xen_lock); 672 673 switch (data->type) { 674 case KVM_XEN_ATTR_TYPE_LONG_MODE: 675 data->u.long_mode = kvm->arch.xen.long_mode; 676 r = 0; 677 break; 678 679 case KVM_XEN_ATTR_TYPE_SHARED_INFO: 680 if (kvm->arch.xen.shinfo_cache.active) 681 data->u.shared_info.gfn = gpa_to_gfn(kvm->arch.xen.shinfo_cache.gpa); 682 else 683 data->u.shared_info.gfn = KVM_XEN_INVALID_GFN; 684 r = 0; 685 break; 686 687 case KVM_XEN_ATTR_TYPE_UPCALL_VECTOR: 688 data->u.vector = kvm->arch.xen.upcall_vector; 689 r = 0; 690 break; 691 692 case KVM_XEN_ATTR_TYPE_XEN_VERSION: 693 data->u.xen_version = kvm->arch.xen.xen_version; 694 r = 0; 695 break; 696 697 case KVM_XEN_ATTR_TYPE_RUNSTATE_UPDATE_FLAG: 698 if (!sched_info_on()) { 699 r = -EOPNOTSUPP; 700 break; 701 } 702 data->u.runstate_update_flag = kvm->arch.xen.runstate_update_flag; 703 r = 0; 704 break; 705 706 default: 707 break; 708 } 709 710 mutex_unlock(&kvm->arch.xen.xen_lock); 711 return r; 712} 713 714int kvm_xen_vcpu_set_attr(struct kvm_vcpu *vcpu, struct kvm_xen_vcpu_attr *data) 715{ 716 int idx, r = -ENOENT; 717 718 mutex_lock(&vcpu->kvm->arch.xen.xen_lock); 719 idx = srcu_read_lock(&vcpu->kvm->srcu); 720 721 switch (data->type) { 722 case KVM_XEN_VCPU_ATTR_TYPE_VCPU_INFO: 723 /* No compat necessary here. */ 724 BUILD_BUG_ON(sizeof(struct vcpu_info) != 725 sizeof(struct compat_vcpu_info)); 726 BUILD_BUG_ON(offsetof(struct vcpu_info, time) != 727 offsetof(struct compat_vcpu_info, time)); 728 729 if (data->u.gpa == KVM_XEN_INVALID_GPA) { 730 kvm_gpc_deactivate(&vcpu->arch.xen.vcpu_info_cache); 731 r = 0; 732 break; 733 } 734 735 r = kvm_gpc_activate(&vcpu->arch.xen.vcpu_info_cache, 736 data->u.gpa, sizeof(struct vcpu_info)); 737 if (!r) 738 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); 739 740 break; 741 742 case KVM_XEN_VCPU_ATTR_TYPE_VCPU_TIME_INFO: 743 if (data->u.gpa == KVM_XEN_INVALID_GPA) { 744 kvm_gpc_deactivate(&vcpu->arch.xen.vcpu_time_info_cache); 745 r = 0; 746 break; 747 } 748 749 r = kvm_gpc_activate(&vcpu->arch.xen.vcpu_time_info_cache, 750 data->u.gpa, 751 sizeof(struct pvclock_vcpu_time_info)); 752 if (!r) 753 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); 754 break; 755 756 case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_ADDR: { 757 size_t sz, sz1, sz2; 758 759 if (!sched_info_on()) { 760 r = -EOPNOTSUPP; 761 break; 762 } 763 if (data->u.gpa == KVM_XEN_INVALID_GPA) { 764 r = 0; 765 deactivate_out: 766 kvm_gpc_deactivate(&vcpu->arch.xen.runstate_cache); 767 kvm_gpc_deactivate(&vcpu->arch.xen.runstate2_cache); 768 break; 769 } 770 771 /* 772 * If the guest switches to 64-bit mode after setting the runstate 773 * address, that's actually OK. kvm_xen_update_runstate_guest() 774 * will cope. 775 */ 776 if (IS_ENABLED(CONFIG_64BIT) && vcpu->kvm->arch.xen.long_mode) 777 sz = sizeof(struct vcpu_runstate_info); 778 else 779 sz = sizeof(struct compat_vcpu_runstate_info); 780 781 /* How much fits in the (first) page? */ 782 sz1 = PAGE_SIZE - (data->u.gpa & ~PAGE_MASK); 783 r = kvm_gpc_activate(&vcpu->arch.xen.runstate_cache, 784 data->u.gpa, sz1); 785 if (r) 786 goto deactivate_out; 787 788 /* Either map the second page, or deactivate the second GPC */ 789 if (sz1 >= sz) { 790 kvm_gpc_deactivate(&vcpu->arch.xen.runstate2_cache); 791 } else { 792 sz2 = sz - sz1; 793 BUG_ON((data->u.gpa + sz1) & ~PAGE_MASK); 794 r = kvm_gpc_activate(&vcpu->arch.xen.runstate2_cache, 795 data->u.gpa + sz1, sz2); 796 if (r) 797 goto deactivate_out; 798 } 799 800 kvm_xen_update_runstate_guest(vcpu, false); 801 break; 802 } 803 case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_CURRENT: 804 if (!sched_info_on()) { 805 r = -EOPNOTSUPP; 806 break; 807 } 808 if (data->u.runstate.state > RUNSTATE_offline) { 809 r = -EINVAL; 810 break; 811 } 812 813 kvm_xen_update_runstate(vcpu, data->u.runstate.state); 814 r = 0; 815 break; 816 817 case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_DATA: 818 if (!sched_info_on()) { 819 r = -EOPNOTSUPP; 820 break; 821 } 822 if (data->u.runstate.state > RUNSTATE_offline) { 823 r = -EINVAL; 824 break; 825 } 826 if (data->u.runstate.state_entry_time != 827 (data->u.runstate.time_running + 828 data->u.runstate.time_runnable + 829 data->u.runstate.time_blocked + 830 data->u.runstate.time_offline)) { 831 r = -EINVAL; 832 break; 833 } 834 if (get_kvmclock_ns(vcpu->kvm) < 835 data->u.runstate.state_entry_time) { 836 r = -EINVAL; 837 break; 838 } 839 840 vcpu->arch.xen.current_runstate = data->u.runstate.state; 841 vcpu->arch.xen.runstate_entry_time = 842 data->u.runstate.state_entry_time; 843 vcpu->arch.xen.runstate_times[RUNSTATE_running] = 844 data->u.runstate.time_running; 845 vcpu->arch.xen.runstate_times[RUNSTATE_runnable] = 846 data->u.runstate.time_runnable; 847 vcpu->arch.xen.runstate_times[RUNSTATE_blocked] = 848 data->u.runstate.time_blocked; 849 vcpu->arch.xen.runstate_times[RUNSTATE_offline] = 850 data->u.runstate.time_offline; 851 vcpu->arch.xen.last_steal = current->sched_info.run_delay; 852 r = 0; 853 break; 854 855 case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_ADJUST: 856 if (!sched_info_on()) { 857 r = -EOPNOTSUPP; 858 break; 859 } 860 if (data->u.runstate.state > RUNSTATE_offline && 861 data->u.runstate.state != (u64)-1) { 862 r = -EINVAL; 863 break; 864 } 865 /* The adjustment must add up */ 866 if (data->u.runstate.state_entry_time != 867 (data->u.runstate.time_running + 868 data->u.runstate.time_runnable + 869 data->u.runstate.time_blocked + 870 data->u.runstate.time_offline)) { 871 r = -EINVAL; 872 break; 873 } 874 875 if (get_kvmclock_ns(vcpu->kvm) < 876 (vcpu->arch.xen.runstate_entry_time + 877 data->u.runstate.state_entry_time)) { 878 r = -EINVAL; 879 break; 880 } 881 882 vcpu->arch.xen.runstate_entry_time += 883 data->u.runstate.state_entry_time; 884 vcpu->arch.xen.runstate_times[RUNSTATE_running] += 885 data->u.runstate.time_running; 886 vcpu->arch.xen.runstate_times[RUNSTATE_runnable] += 887 data->u.runstate.time_runnable; 888 vcpu->arch.xen.runstate_times[RUNSTATE_blocked] += 889 data->u.runstate.time_blocked; 890 vcpu->arch.xen.runstate_times[RUNSTATE_offline] += 891 data->u.runstate.time_offline; 892 893 if (data->u.runstate.state <= RUNSTATE_offline) 894 kvm_xen_update_runstate(vcpu, data->u.runstate.state); 895 else if (vcpu->arch.xen.runstate_cache.active) 896 kvm_xen_update_runstate_guest(vcpu, false); 897 r = 0; 898 break; 899 900 case KVM_XEN_VCPU_ATTR_TYPE_VCPU_ID: 901 if (data->u.vcpu_id >= KVM_MAX_VCPUS) 902 r = -EINVAL; 903 else { 904 vcpu->arch.xen.vcpu_id = data->u.vcpu_id; 905 r = 0; 906 } 907 break; 908 909 case KVM_XEN_VCPU_ATTR_TYPE_TIMER: 910 if (data->u.timer.port && 911 data->u.timer.priority != KVM_IRQ_ROUTING_XEN_EVTCHN_PRIO_2LEVEL) { 912 r = -EINVAL; 913 break; 914 } 915 916 if (!vcpu->arch.xen.timer.function) 917 kvm_xen_init_timer(vcpu); 918 919 /* Stop the timer (if it's running) before changing the vector */ 920 kvm_xen_stop_timer(vcpu); 921 vcpu->arch.xen.timer_virq = data->u.timer.port; 922 923 /* Start the timer if the new value has a valid vector+expiry. */ 924 if (data->u.timer.port && data->u.timer.expires_ns) 925 kvm_xen_start_timer(vcpu, data->u.timer.expires_ns, 926 data->u.timer.expires_ns - 927 get_kvmclock_ns(vcpu->kvm)); 928 929 r = 0; 930 break; 931 932 case KVM_XEN_VCPU_ATTR_TYPE_UPCALL_VECTOR: 933 if (data->u.vector && data->u.vector < 0x10) 934 r = -EINVAL; 935 else { 936 vcpu->arch.xen.upcall_vector = data->u.vector; 937 r = 0; 938 } 939 break; 940 941 default: 942 break; 943 } 944 945 srcu_read_unlock(&vcpu->kvm->srcu, idx); 946 mutex_unlock(&vcpu->kvm->arch.xen.xen_lock); 947 return r; 948} 949 950int kvm_xen_vcpu_get_attr(struct kvm_vcpu *vcpu, struct kvm_xen_vcpu_attr *data) 951{ 952 int r = -ENOENT; 953 954 mutex_lock(&vcpu->kvm->arch.xen.xen_lock); 955 956 switch (data->type) { 957 case KVM_XEN_VCPU_ATTR_TYPE_VCPU_INFO: 958 if (vcpu->arch.xen.vcpu_info_cache.active) 959 data->u.gpa = vcpu->arch.xen.vcpu_info_cache.gpa; 960 else 961 data->u.gpa = KVM_XEN_INVALID_GPA; 962 r = 0; 963 break; 964 965 case KVM_XEN_VCPU_ATTR_TYPE_VCPU_TIME_INFO: 966 if (vcpu->arch.xen.vcpu_time_info_cache.active) 967 data->u.gpa = vcpu->arch.xen.vcpu_time_info_cache.gpa; 968 else 969 data->u.gpa = KVM_XEN_INVALID_GPA; 970 r = 0; 971 break; 972 973 case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_ADDR: 974 if (!sched_info_on()) { 975 r = -EOPNOTSUPP; 976 break; 977 } 978 if (vcpu->arch.xen.runstate_cache.active) { 979 data->u.gpa = vcpu->arch.xen.runstate_cache.gpa; 980 r = 0; 981 } 982 break; 983 984 case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_CURRENT: 985 if (!sched_info_on()) { 986 r = -EOPNOTSUPP; 987 break; 988 } 989 data->u.runstate.state = vcpu->arch.xen.current_runstate; 990 r = 0; 991 break; 992 993 case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_DATA: 994 if (!sched_info_on()) { 995 r = -EOPNOTSUPP; 996 break; 997 } 998 data->u.runstate.state = vcpu->arch.xen.current_runstate; 999 data->u.runstate.state_entry_time = 1000 vcpu->arch.xen.runstate_entry_time; 1001 data->u.runstate.time_running = 1002 vcpu->arch.xen.runstate_times[RUNSTATE_running]; 1003 data->u.runstate.time_runnable = 1004 vcpu->arch.xen.runstate_times[RUNSTATE_runnable]; 1005 data->u.runstate.time_blocked = 1006 vcpu->arch.xen.runstate_times[RUNSTATE_blocked]; 1007 data->u.runstate.time_offline = 1008 vcpu->arch.xen.runstate_times[RUNSTATE_offline]; 1009 r = 0; 1010 break; 1011 1012 case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_ADJUST: 1013 r = -EINVAL; 1014 break; 1015 1016 case KVM_XEN_VCPU_ATTR_TYPE_VCPU_ID: 1017 data->u.vcpu_id = vcpu->arch.xen.vcpu_id; 1018 r = 0; 1019 break; 1020 1021 case KVM_XEN_VCPU_ATTR_TYPE_TIMER: 1022 data->u.timer.port = vcpu->arch.xen.timer_virq; 1023 data->u.timer.priority = KVM_IRQ_ROUTING_XEN_EVTCHN_PRIO_2LEVEL; 1024 data->u.timer.expires_ns = vcpu->arch.xen.timer_expires; 1025 r = 0; 1026 break; 1027 1028 case KVM_XEN_VCPU_ATTR_TYPE_UPCALL_VECTOR: 1029 data->u.vector = vcpu->arch.xen.upcall_vector; 1030 r = 0; 1031 break; 1032 1033 default: 1034 break; 1035 } 1036 1037 mutex_unlock(&vcpu->kvm->arch.xen.xen_lock); 1038 return r; 1039} 1040 1041int kvm_xen_write_hypercall_page(struct kvm_vcpu *vcpu, u64 data) 1042{ 1043 struct kvm *kvm = vcpu->kvm; 1044 u32 page_num = data & ~PAGE_MASK; 1045 u64 page_addr = data & PAGE_MASK; 1046 bool lm = is_long_mode(vcpu); 1047 1048 /* Latch long_mode for shared_info pages etc. */ 1049 vcpu->kvm->arch.xen.long_mode = lm; 1050 1051 /* 1052 * If Xen hypercall intercept is enabled, fill the hypercall 1053 * page with VMCALL/VMMCALL instructions since that's what 1054 * we catch. Else the VMM has provided the hypercall pages 1055 * with instructions of its own choosing, so use those. 1056 */ 1057 if (kvm_xen_hypercall_enabled(kvm)) { 1058 u8 instructions[32]; 1059 int i; 1060 1061 if (page_num) 1062 return 1; 1063 1064 /* mov imm32, %eax */ 1065 instructions[0] = 0xb8; 1066 1067 /* vmcall / vmmcall */ 1068 static_call(kvm_x86_patch_hypercall)(vcpu, instructions + 5); 1069 1070 /* ret */ 1071 instructions[8] = 0xc3; 1072 1073 /* int3 to pad */ 1074 memset(instructions + 9, 0xcc, sizeof(instructions) - 9); 1075 1076 for (i = 0; i < PAGE_SIZE / sizeof(instructions); i++) { 1077 *(u32 *)&instructions[1] = i; 1078 if (kvm_vcpu_write_guest(vcpu, 1079 page_addr + (i * sizeof(instructions)), 1080 instructions, sizeof(instructions))) 1081 return 1; 1082 } 1083 } else { 1084 /* 1085 * Note, truncation is a non-issue as 'lm' is guaranteed to be 1086 * false for a 32-bit kernel, i.e. when hva_t is only 4 bytes. 1087 */ 1088 hva_t blob_addr = lm ? kvm->arch.xen_hvm_config.blob_addr_64 1089 : kvm->arch.xen_hvm_config.blob_addr_32; 1090 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64 1091 : kvm->arch.xen_hvm_config.blob_size_32; 1092 u8 *page; 1093 int ret; 1094 1095 if (page_num >= blob_size) 1096 return 1; 1097 1098 blob_addr += page_num * PAGE_SIZE; 1099 1100 page = memdup_user((u8 __user *)blob_addr, PAGE_SIZE); 1101 if (IS_ERR(page)) 1102 return PTR_ERR(page); 1103 1104 ret = kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE); 1105 kfree(page); 1106 if (ret) 1107 return 1; 1108 } 1109 return 0; 1110} 1111 1112int kvm_xen_hvm_config(struct kvm *kvm, struct kvm_xen_hvm_config *xhc) 1113{ 1114 /* Only some feature flags need to be *enabled* by userspace */ 1115 u32 permitted_flags = KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL | 1116 KVM_XEN_HVM_CONFIG_EVTCHN_SEND; 1117 1118 if (xhc->flags & ~permitted_flags) 1119 return -EINVAL; 1120 1121 /* 1122 * With hypercall interception the kernel generates its own 1123 * hypercall page so it must not be provided. 1124 */ 1125 if ((xhc->flags & KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL) && 1126 (xhc->blob_addr_32 || xhc->blob_addr_64 || 1127 xhc->blob_size_32 || xhc->blob_size_64)) 1128 return -EINVAL; 1129 1130 mutex_lock(&kvm->arch.xen.xen_lock); 1131 1132 if (xhc->msr && !kvm->arch.xen_hvm_config.msr) 1133 static_branch_inc(&kvm_xen_enabled.key); 1134 else if (!xhc->msr && kvm->arch.xen_hvm_config.msr) 1135 static_branch_slow_dec_deferred(&kvm_xen_enabled); 1136 1137 memcpy(&kvm->arch.xen_hvm_config, xhc, sizeof(*xhc)); 1138 1139 mutex_unlock(&kvm->arch.xen.xen_lock); 1140 return 0; 1141} 1142 1143static int kvm_xen_hypercall_set_result(struct kvm_vcpu *vcpu, u64 result) 1144{ 1145 kvm_rax_write(vcpu, result); 1146 return kvm_skip_emulated_instruction(vcpu); 1147} 1148 1149static int kvm_xen_hypercall_complete_userspace(struct kvm_vcpu *vcpu) 1150{ 1151 struct kvm_run *run = vcpu->run; 1152 1153 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.xen.hypercall_rip))) 1154 return 1; 1155 1156 return kvm_xen_hypercall_set_result(vcpu, run->xen.u.hcall.result); 1157} 1158 1159static inline int max_evtchn_port(struct kvm *kvm) 1160{ 1161 if (IS_ENABLED(CONFIG_64BIT) && kvm->arch.xen.long_mode) 1162 return EVTCHN_2L_NR_CHANNELS; 1163 else 1164 return COMPAT_EVTCHN_2L_NR_CHANNELS; 1165} 1166 1167static bool wait_pending_event(struct kvm_vcpu *vcpu, int nr_ports, 1168 evtchn_port_t *ports) 1169{ 1170 struct kvm *kvm = vcpu->kvm; 1171 struct gfn_to_pfn_cache *gpc = &kvm->arch.xen.shinfo_cache; 1172 unsigned long *pending_bits; 1173 unsigned long flags; 1174 bool ret = true; 1175 int idx, i; 1176 1177 idx = srcu_read_lock(&kvm->srcu); 1178 read_lock_irqsave(&gpc->lock, flags); 1179 if (!kvm_gpc_check(gpc, PAGE_SIZE)) 1180 goto out_rcu; 1181 1182 ret = false; 1183 if (IS_ENABLED(CONFIG_64BIT) && kvm->arch.xen.long_mode) { 1184 struct shared_info *shinfo = gpc->khva; 1185 pending_bits = (unsigned long *)&shinfo->evtchn_pending; 1186 } else { 1187 struct compat_shared_info *shinfo = gpc->khva; 1188 pending_bits = (unsigned long *)&shinfo->evtchn_pending; 1189 } 1190 1191 for (i = 0; i < nr_ports; i++) { 1192 if (test_bit(ports[i], pending_bits)) { 1193 ret = true; 1194 break; 1195 } 1196 } 1197 1198 out_rcu: 1199 read_unlock_irqrestore(&gpc->lock, flags); 1200 srcu_read_unlock(&kvm->srcu, idx); 1201 1202 return ret; 1203} 1204 1205static bool kvm_xen_schedop_poll(struct kvm_vcpu *vcpu, bool longmode, 1206 u64 param, u64 *r) 1207{ 1208 struct sched_poll sched_poll; 1209 evtchn_port_t port, *ports; 1210 struct x86_exception e; 1211 int i; 1212 1213 if (!lapic_in_kernel(vcpu) || 1214 !(vcpu->kvm->arch.xen_hvm_config.flags & KVM_XEN_HVM_CONFIG_EVTCHN_SEND)) 1215 return false; 1216 1217 if (IS_ENABLED(CONFIG_64BIT) && !longmode) { 1218 struct compat_sched_poll sp32; 1219 1220 /* Sanity check that the compat struct definition is correct */ 1221 BUILD_BUG_ON(sizeof(sp32) != 16); 1222 1223 if (kvm_read_guest_virt(vcpu, param, &sp32, sizeof(sp32), &e)) { 1224 *r = -EFAULT; 1225 return true; 1226 } 1227 1228 /* 1229 * This is a 32-bit pointer to an array of evtchn_port_t which 1230 * are uint32_t, so once it's converted no further compat 1231 * handling is needed. 1232 */ 1233 sched_poll.ports = (void *)(unsigned long)(sp32.ports); 1234 sched_poll.nr_ports = sp32.nr_ports; 1235 sched_poll.timeout = sp32.timeout; 1236 } else { 1237 if (kvm_read_guest_virt(vcpu, param, &sched_poll, 1238 sizeof(sched_poll), &e)) { 1239 *r = -EFAULT; 1240 return true; 1241 } 1242 } 1243 1244 if (unlikely(sched_poll.nr_ports > 1)) { 1245 /* Xen (unofficially) limits number of pollers to 128 */ 1246 if (sched_poll.nr_ports > 128) { 1247 *r = -EINVAL; 1248 return true; 1249 } 1250 1251 ports = kmalloc_array(sched_poll.nr_ports, 1252 sizeof(*ports), GFP_KERNEL); 1253 if (!ports) { 1254 *r = -ENOMEM; 1255 return true; 1256 } 1257 } else 1258 ports = &port; 1259 1260 if (kvm_read_guest_virt(vcpu, (gva_t)sched_poll.ports, ports, 1261 sched_poll.nr_ports * sizeof(*ports), &e)) { 1262 *r = -EFAULT; 1263 return true; 1264 } 1265 1266 for (i = 0; i < sched_poll.nr_ports; i++) { 1267 if (ports[i] >= max_evtchn_port(vcpu->kvm)) { 1268 *r = -EINVAL; 1269 goto out; 1270 } 1271 } 1272 1273 if (sched_poll.nr_ports == 1) 1274 vcpu->arch.xen.poll_evtchn = port; 1275 else 1276 vcpu->arch.xen.poll_evtchn = -1; 1277 1278 set_bit(vcpu->vcpu_idx, vcpu->kvm->arch.xen.poll_mask); 1279 1280 if (!wait_pending_event(vcpu, sched_poll.nr_ports, ports)) { 1281 vcpu->arch.mp_state = KVM_MP_STATE_HALTED; 1282 1283 if (sched_poll.timeout) 1284 mod_timer(&vcpu->arch.xen.poll_timer, 1285 jiffies + nsecs_to_jiffies(sched_poll.timeout)); 1286 1287 kvm_vcpu_halt(vcpu); 1288 1289 if (sched_poll.timeout) 1290 del_timer(&vcpu->arch.xen.poll_timer); 1291 1292 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE; 1293 } 1294 1295 vcpu->arch.xen.poll_evtchn = 0; 1296 *r = 0; 1297out: 1298 /* Really, this is only needed in case of timeout */ 1299 clear_bit(vcpu->vcpu_idx, vcpu->kvm->arch.xen.poll_mask); 1300 1301 if (unlikely(sched_poll.nr_ports > 1)) 1302 kfree(ports); 1303 return true; 1304} 1305 1306static void cancel_evtchn_poll(struct timer_list *t) 1307{ 1308 struct kvm_vcpu *vcpu = from_timer(vcpu, t, arch.xen.poll_timer); 1309 1310 kvm_make_request(KVM_REQ_UNBLOCK, vcpu); 1311 kvm_vcpu_kick(vcpu); 1312} 1313 1314static bool kvm_xen_hcall_sched_op(struct kvm_vcpu *vcpu, bool longmode, 1315 int cmd, u64 param, u64 *r) 1316{ 1317 switch (cmd) { 1318 case SCHEDOP_poll: 1319 if (kvm_xen_schedop_poll(vcpu, longmode, param, r)) 1320 return true; 1321 fallthrough; 1322 case SCHEDOP_yield: 1323 kvm_vcpu_on_spin(vcpu, true); 1324 *r = 0; 1325 return true; 1326 default: 1327 break; 1328 } 1329 1330 return false; 1331} 1332 1333struct compat_vcpu_set_singleshot_timer { 1334 uint64_t timeout_abs_ns; 1335 uint32_t flags; 1336} __attribute__((packed)); 1337 1338static bool kvm_xen_hcall_vcpu_op(struct kvm_vcpu *vcpu, bool longmode, int cmd, 1339 int vcpu_id, u64 param, u64 *r) 1340{ 1341 struct vcpu_set_singleshot_timer oneshot; 1342 struct x86_exception e; 1343 s64 delta; 1344 1345 if (!kvm_xen_timer_enabled(vcpu)) 1346 return false; 1347 1348 switch (cmd) { 1349 case VCPUOP_set_singleshot_timer: 1350 if (vcpu->arch.xen.vcpu_id != vcpu_id) { 1351 *r = -EINVAL; 1352 return true; 1353 } 1354 1355 /* 1356 * The only difference for 32-bit compat is the 4 bytes of 1357 * padding after the interesting part of the structure. So 1358 * for a faithful emulation of Xen we have to *try* to copy 1359 * the padding and return -EFAULT if we can't. Otherwise we 1360 * might as well just have copied the 12-byte 32-bit struct. 1361 */ 1362 BUILD_BUG_ON(offsetof(struct compat_vcpu_set_singleshot_timer, timeout_abs_ns) != 1363 offsetof(struct vcpu_set_singleshot_timer, timeout_abs_ns)); 1364 BUILD_BUG_ON(sizeof_field(struct compat_vcpu_set_singleshot_timer, timeout_abs_ns) != 1365 sizeof_field(struct vcpu_set_singleshot_timer, timeout_abs_ns)); 1366 BUILD_BUG_ON(offsetof(struct compat_vcpu_set_singleshot_timer, flags) != 1367 offsetof(struct vcpu_set_singleshot_timer, flags)); 1368 BUILD_BUG_ON(sizeof_field(struct compat_vcpu_set_singleshot_timer, flags) != 1369 sizeof_field(struct vcpu_set_singleshot_timer, flags)); 1370 1371 if (kvm_read_guest_virt(vcpu, param, &oneshot, longmode ? sizeof(oneshot) : 1372 sizeof(struct compat_vcpu_set_singleshot_timer), &e)) { 1373 *r = -EFAULT; 1374 return true; 1375 } 1376 1377 delta = oneshot.timeout_abs_ns - get_kvmclock_ns(vcpu->kvm); 1378 if ((oneshot.flags & VCPU_SSHOTTMR_future) && delta < 0) { 1379 *r = -ETIME; 1380 return true; 1381 } 1382 1383 kvm_xen_start_timer(vcpu, oneshot.timeout_abs_ns, delta); 1384 *r = 0; 1385 return true; 1386 1387 case VCPUOP_stop_singleshot_timer: 1388 if (vcpu->arch.xen.vcpu_id != vcpu_id) { 1389 *r = -EINVAL; 1390 return true; 1391 } 1392 kvm_xen_stop_timer(vcpu); 1393 *r = 0; 1394 return true; 1395 } 1396 1397 return false; 1398} 1399 1400static bool kvm_xen_hcall_set_timer_op(struct kvm_vcpu *vcpu, uint64_t timeout, 1401 u64 *r) 1402{ 1403 if (!kvm_xen_timer_enabled(vcpu)) 1404 return false; 1405 1406 if (timeout) { 1407 uint64_t guest_now = get_kvmclock_ns(vcpu->kvm); 1408 int64_t delta = timeout - guest_now; 1409 1410 /* Xen has a 'Linux workaround' in do_set_timer_op() which 1411 * checks for negative absolute timeout values (caused by 1412 * integer overflow), and for values about 13 days in the 1413 * future (2^50ns) which would be caused by jiffies 1414 * overflow. For those cases, it sets the timeout 100ms in 1415 * the future (not *too* soon, since if a guest really did 1416 * set a long timeout on purpose we don't want to keep 1417 * churning CPU time by waking it up). 1418 */ 1419 if (unlikely((int64_t)timeout < 0 || 1420 (delta > 0 && (uint32_t) (delta >> 50) != 0))) { 1421 delta = 100 * NSEC_PER_MSEC; 1422 timeout = guest_now + delta; 1423 } 1424 1425 kvm_xen_start_timer(vcpu, timeout, delta); 1426 } else { 1427 kvm_xen_stop_timer(vcpu); 1428 } 1429 1430 *r = 0; 1431 return true; 1432} 1433 1434int kvm_xen_hypercall(struct kvm_vcpu *vcpu) 1435{ 1436 bool longmode; 1437 u64 input, params[6], r = -ENOSYS; 1438 bool handled = false; 1439 u8 cpl; 1440 1441 input = (u64)kvm_register_read(vcpu, VCPU_REGS_RAX); 1442 1443 /* Hyper-V hypercalls get bit 31 set in EAX */ 1444 if ((input & 0x80000000) && 1445 kvm_hv_hypercall_enabled(vcpu)) 1446 return kvm_hv_hypercall(vcpu); 1447 1448 longmode = is_64_bit_hypercall(vcpu); 1449 if (!longmode) { 1450 params[0] = (u32)kvm_rbx_read(vcpu); 1451 params[1] = (u32)kvm_rcx_read(vcpu); 1452 params[2] = (u32)kvm_rdx_read(vcpu); 1453 params[3] = (u32)kvm_rsi_read(vcpu); 1454 params[4] = (u32)kvm_rdi_read(vcpu); 1455 params[5] = (u32)kvm_rbp_read(vcpu); 1456 } 1457#ifdef CONFIG_X86_64 1458 else { 1459 params[0] = (u64)kvm_rdi_read(vcpu); 1460 params[1] = (u64)kvm_rsi_read(vcpu); 1461 params[2] = (u64)kvm_rdx_read(vcpu); 1462 params[3] = (u64)kvm_r10_read(vcpu); 1463 params[4] = (u64)kvm_r8_read(vcpu); 1464 params[5] = (u64)kvm_r9_read(vcpu); 1465 } 1466#endif 1467 cpl = static_call(kvm_x86_get_cpl)(vcpu); 1468 trace_kvm_xen_hypercall(cpl, input, params[0], params[1], params[2], 1469 params[3], params[4], params[5]); 1470 1471 /* 1472 * Only allow hypercall acceleration for CPL0. The rare hypercalls that 1473 * are permitted in guest userspace can be handled by the VMM. 1474 */ 1475 if (unlikely(cpl > 0)) 1476 goto handle_in_userspace; 1477 1478 switch (input) { 1479 case __HYPERVISOR_xen_version: 1480 if (params[0] == XENVER_version && vcpu->kvm->arch.xen.xen_version) { 1481 r = vcpu->kvm->arch.xen.xen_version; 1482 handled = true; 1483 } 1484 break; 1485 case __HYPERVISOR_event_channel_op: 1486 if (params[0] == EVTCHNOP_send) 1487 handled = kvm_xen_hcall_evtchn_send(vcpu, params[1], &r); 1488 break; 1489 case __HYPERVISOR_sched_op: 1490 handled = kvm_xen_hcall_sched_op(vcpu, longmode, params[0], 1491 params[1], &r); 1492 break; 1493 case __HYPERVISOR_vcpu_op: 1494 handled = kvm_xen_hcall_vcpu_op(vcpu, longmode, params[0], params[1], 1495 params[2], &r); 1496 break; 1497 case __HYPERVISOR_set_timer_op: { 1498 u64 timeout = params[0]; 1499 /* In 32-bit mode, the 64-bit timeout is in two 32-bit params. */ 1500 if (!longmode) 1501 timeout |= params[1] << 32; 1502 handled = kvm_xen_hcall_set_timer_op(vcpu, timeout, &r); 1503 break; 1504 } 1505 default: 1506 break; 1507 } 1508 1509 if (handled) 1510 return kvm_xen_hypercall_set_result(vcpu, r); 1511 1512handle_in_userspace: 1513 vcpu->run->exit_reason = KVM_EXIT_XEN; 1514 vcpu->run->xen.type = KVM_EXIT_XEN_HCALL; 1515 vcpu->run->xen.u.hcall.longmode = longmode; 1516 vcpu->run->xen.u.hcall.cpl = cpl; 1517 vcpu->run->xen.u.hcall.input = input; 1518 vcpu->run->xen.u.hcall.params[0] = params[0]; 1519 vcpu->run->xen.u.hcall.params[1] = params[1]; 1520 vcpu->run->xen.u.hcall.params[2] = params[2]; 1521 vcpu->run->xen.u.hcall.params[3] = params[3]; 1522 vcpu->run->xen.u.hcall.params[4] = params[4]; 1523 vcpu->run->xen.u.hcall.params[5] = params[5]; 1524 vcpu->arch.xen.hypercall_rip = kvm_get_linear_rip(vcpu); 1525 vcpu->arch.complete_userspace_io = 1526 kvm_xen_hypercall_complete_userspace; 1527 1528 return 0; 1529} 1530 1531static void kvm_xen_check_poller(struct kvm_vcpu *vcpu, int port) 1532{ 1533 int poll_evtchn = vcpu->arch.xen.poll_evtchn; 1534 1535 if ((poll_evtchn == port || poll_evtchn == -1) && 1536 test_and_clear_bit(vcpu->vcpu_idx, vcpu->kvm->arch.xen.poll_mask)) { 1537 kvm_make_request(KVM_REQ_UNBLOCK, vcpu); 1538 kvm_vcpu_kick(vcpu); 1539 } 1540} 1541 1542/* 1543 * The return value from this function is propagated to kvm_set_irq() API, 1544 * so it returns: 1545 * < 0 Interrupt was ignored (masked or not delivered for other reasons) 1546 * = 0 Interrupt was coalesced (previous irq is still pending) 1547 * > 0 Number of CPUs interrupt was delivered to 1548 * 1549 * It is also called directly from kvm_arch_set_irq_inatomic(), where the 1550 * only check on its return value is a comparison with -EWOULDBLOCK'. 1551 */ 1552int kvm_xen_set_evtchn_fast(struct kvm_xen_evtchn *xe, struct kvm *kvm) 1553{ 1554 struct gfn_to_pfn_cache *gpc = &kvm->arch.xen.shinfo_cache; 1555 struct kvm_vcpu *vcpu; 1556 unsigned long *pending_bits, *mask_bits; 1557 unsigned long flags; 1558 int port_word_bit; 1559 bool kick_vcpu = false; 1560 int vcpu_idx, idx, rc; 1561 1562 vcpu_idx = READ_ONCE(xe->vcpu_idx); 1563 if (vcpu_idx >= 0) 1564 vcpu = kvm_get_vcpu(kvm, vcpu_idx); 1565 else { 1566 vcpu = kvm_get_vcpu_by_id(kvm, xe->vcpu_id); 1567 if (!vcpu) 1568 return -EINVAL; 1569 WRITE_ONCE(xe->vcpu_idx, vcpu->vcpu_idx); 1570 } 1571 1572 if (!vcpu->arch.xen.vcpu_info_cache.active) 1573 return -EINVAL; 1574 1575 if (xe->port >= max_evtchn_port(kvm)) 1576 return -EINVAL; 1577 1578 rc = -EWOULDBLOCK; 1579 1580 idx = srcu_read_lock(&kvm->srcu); 1581 1582 read_lock_irqsave(&gpc->lock, flags); 1583 if (!kvm_gpc_check(gpc, PAGE_SIZE)) 1584 goto out_rcu; 1585 1586 if (IS_ENABLED(CONFIG_64BIT) && kvm->arch.xen.long_mode) { 1587 struct shared_info *shinfo = gpc->khva; 1588 pending_bits = (unsigned long *)&shinfo->evtchn_pending; 1589 mask_bits = (unsigned long *)&shinfo->evtchn_mask; 1590 port_word_bit = xe->port / 64; 1591 } else { 1592 struct compat_shared_info *shinfo = gpc->khva; 1593 pending_bits = (unsigned long *)&shinfo->evtchn_pending; 1594 mask_bits = (unsigned long *)&shinfo->evtchn_mask; 1595 port_word_bit = xe->port / 32; 1596 } 1597 1598 /* 1599 * If this port wasn't already set, and if it isn't masked, then 1600 * we try to set the corresponding bit in the in-kernel shadow of 1601 * evtchn_pending_sel for the target vCPU. And if *that* wasn't 1602 * already set, then we kick the vCPU in question to write to the 1603 * *real* evtchn_pending_sel in its own guest vcpu_info struct. 1604 */ 1605 if (test_and_set_bit(xe->port, pending_bits)) { 1606 rc = 0; /* It was already raised */ 1607 } else if (test_bit(xe->port, mask_bits)) { 1608 rc = -ENOTCONN; /* Masked */ 1609 kvm_xen_check_poller(vcpu, xe->port); 1610 } else { 1611 rc = 1; /* Delivered to the bitmap in shared_info. */ 1612 /* Now switch to the vCPU's vcpu_info to set the index and pending_sel */ 1613 read_unlock_irqrestore(&gpc->lock, flags); 1614 gpc = &vcpu->arch.xen.vcpu_info_cache; 1615 1616 read_lock_irqsave(&gpc->lock, flags); 1617 if (!kvm_gpc_check(gpc, sizeof(struct vcpu_info))) { 1618 /* 1619 * Could not access the vcpu_info. Set the bit in-kernel 1620 * and prod the vCPU to deliver it for itself. 1621 */ 1622 if (!test_and_set_bit(port_word_bit, &vcpu->arch.xen.evtchn_pending_sel)) 1623 kick_vcpu = true; 1624 goto out_rcu; 1625 } 1626 1627 if (IS_ENABLED(CONFIG_64BIT) && kvm->arch.xen.long_mode) { 1628 struct vcpu_info *vcpu_info = gpc->khva; 1629 if (!test_and_set_bit(port_word_bit, &vcpu_info->evtchn_pending_sel)) { 1630 WRITE_ONCE(vcpu_info->evtchn_upcall_pending, 1); 1631 kick_vcpu = true; 1632 } 1633 } else { 1634 struct compat_vcpu_info *vcpu_info = gpc->khva; 1635 if (!test_and_set_bit(port_word_bit, 1636 (unsigned long *)&vcpu_info->evtchn_pending_sel)) { 1637 WRITE_ONCE(vcpu_info->evtchn_upcall_pending, 1); 1638 kick_vcpu = true; 1639 } 1640 } 1641 1642 /* For the per-vCPU lapic vector, deliver it as MSI. */ 1643 if (kick_vcpu && vcpu->arch.xen.upcall_vector) { 1644 kvm_xen_inject_vcpu_vector(vcpu); 1645 kick_vcpu = false; 1646 } 1647 } 1648 1649 out_rcu: 1650 read_unlock_irqrestore(&gpc->lock, flags); 1651 srcu_read_unlock(&kvm->srcu, idx); 1652 1653 if (kick_vcpu) { 1654 kvm_make_request(KVM_REQ_UNBLOCK, vcpu); 1655 kvm_vcpu_kick(vcpu); 1656 } 1657 1658 return rc; 1659} 1660 1661static int kvm_xen_set_evtchn(struct kvm_xen_evtchn *xe, struct kvm *kvm) 1662{ 1663 bool mm_borrowed = false; 1664 int rc; 1665 1666 rc = kvm_xen_set_evtchn_fast(xe, kvm); 1667 if (rc != -EWOULDBLOCK) 1668 return rc; 1669 1670 if (current->mm != kvm->mm) { 1671 /* 1672 * If not on a thread which already belongs to this KVM, 1673 * we'd better be in the irqfd workqueue. 1674 */ 1675 if (WARN_ON_ONCE(current->mm)) 1676 return -EINVAL; 1677 1678 kthread_use_mm(kvm->mm); 1679 mm_borrowed = true; 1680 } 1681 1682 mutex_lock(&kvm->arch.xen.xen_lock); 1683 1684 /* 1685 * It is theoretically possible for the page to be unmapped 1686 * and the MMU notifier to invalidate the shared_info before 1687 * we even get to use it. In that case, this looks like an 1688 * infinite loop. It was tempting to do it via the userspace 1689 * HVA instead... but that just *hides* the fact that it's 1690 * an infinite loop, because if a fault occurs and it waits 1691 * for the page to come back, it can *still* immediately 1692 * fault and have to wait again, repeatedly. 1693 * 1694 * Conversely, the page could also have been reinstated by 1695 * another thread before we even obtain the mutex above, so 1696 * check again *first* before remapping it. 1697 */ 1698 do { 1699 struct gfn_to_pfn_cache *gpc = &kvm->arch.xen.shinfo_cache; 1700 int idx; 1701 1702 rc = kvm_xen_set_evtchn_fast(xe, kvm); 1703 if (rc != -EWOULDBLOCK) 1704 break; 1705 1706 idx = srcu_read_lock(&kvm->srcu); 1707 rc = kvm_gpc_refresh(gpc, PAGE_SIZE); 1708 srcu_read_unlock(&kvm->srcu, idx); 1709 } while(!rc); 1710 1711 mutex_unlock(&kvm->arch.xen.xen_lock); 1712 1713 if (mm_borrowed) 1714 kthread_unuse_mm(kvm->mm); 1715 1716 return rc; 1717} 1718 1719/* This is the version called from kvm_set_irq() as the .set function */ 1720static int evtchn_set_fn(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm, 1721 int irq_source_id, int level, bool line_status) 1722{ 1723 if (!level) 1724 return -EINVAL; 1725 1726 return kvm_xen_set_evtchn(&e->xen_evtchn, kvm); 1727} 1728 1729/* 1730 * Set up an event channel interrupt from the KVM IRQ routing table. 1731 * Used for e.g. PIRQ from passed through physical devices. 1732 */ 1733int kvm_xen_setup_evtchn(struct kvm *kvm, 1734 struct kvm_kernel_irq_routing_entry *e, 1735 const struct kvm_irq_routing_entry *ue) 1736 1737{ 1738 struct kvm_vcpu *vcpu; 1739 1740 if (ue->u.xen_evtchn.port >= max_evtchn_port(kvm)) 1741 return -EINVAL; 1742 1743 /* We only support 2 level event channels for now */ 1744 if (ue->u.xen_evtchn.priority != KVM_IRQ_ROUTING_XEN_EVTCHN_PRIO_2LEVEL) 1745 return -EINVAL; 1746 1747 /* 1748 * Xen gives us interesting mappings from vCPU index to APIC ID, 1749 * which means kvm_get_vcpu_by_id() has to iterate over all vCPUs 1750 * to find it. Do that once at setup time, instead of every time. 1751 * But beware that on live update / live migration, the routing 1752 * table might be reinstated before the vCPU threads have finished 1753 * recreating their vCPUs. 1754 */ 1755 vcpu = kvm_get_vcpu_by_id(kvm, ue->u.xen_evtchn.vcpu); 1756 if (vcpu) 1757 e->xen_evtchn.vcpu_idx = vcpu->vcpu_idx; 1758 else 1759 e->xen_evtchn.vcpu_idx = -1; 1760 1761 e->xen_evtchn.port = ue->u.xen_evtchn.port; 1762 e->xen_evtchn.vcpu_id = ue->u.xen_evtchn.vcpu; 1763 e->xen_evtchn.priority = ue->u.xen_evtchn.priority; 1764 e->set = evtchn_set_fn; 1765 1766 return 0; 1767} 1768 1769/* 1770 * Explicit event sending from userspace with KVM_XEN_HVM_EVTCHN_SEND ioctl. 1771 */ 1772int kvm_xen_hvm_evtchn_send(struct kvm *kvm, struct kvm_irq_routing_xen_evtchn *uxe) 1773{ 1774 struct kvm_xen_evtchn e; 1775 int ret; 1776 1777 if (!uxe->port || uxe->port >= max_evtchn_port(kvm)) 1778 return -EINVAL; 1779 1780 /* We only support 2 level event channels for now */ 1781 if (uxe->priority != KVM_IRQ_ROUTING_XEN_EVTCHN_PRIO_2LEVEL) 1782 return -EINVAL; 1783 1784 e.port = uxe->port; 1785 e.vcpu_id = uxe->vcpu; 1786 e.vcpu_idx = -1; 1787 e.priority = uxe->priority; 1788 1789 ret = kvm_xen_set_evtchn(&e, kvm); 1790 1791 /* 1792 * None of that 'return 1 if it actually got delivered' nonsense. 1793 * We don't care if it was masked (-ENOTCONN) either. 1794 */ 1795 if (ret > 0 || ret == -ENOTCONN) 1796 ret = 0; 1797 1798 return ret; 1799} 1800 1801/* 1802 * Support for *outbound* event channel events via the EVTCHNOP_send hypercall. 1803 */ 1804struct evtchnfd { 1805 u32 send_port; 1806 u32 type; 1807 union { 1808 struct kvm_xen_evtchn port; 1809 struct { 1810 u32 port; /* zero */ 1811 struct eventfd_ctx *ctx; 1812 } eventfd; 1813 } deliver; 1814}; 1815 1816/* 1817 * Update target vCPU or priority for a registered sending channel. 1818 */ 1819static int kvm_xen_eventfd_update(struct kvm *kvm, 1820 struct kvm_xen_hvm_attr *data) 1821{ 1822 u32 port = data->u.evtchn.send_port; 1823 struct evtchnfd *evtchnfd; 1824 int ret; 1825 1826 /* Protect writes to evtchnfd as well as the idr lookup. */ 1827 mutex_lock(&kvm->arch.xen.xen_lock); 1828 evtchnfd = idr_find(&kvm->arch.xen.evtchn_ports, port); 1829 1830 ret = -ENOENT; 1831 if (!evtchnfd) 1832 goto out_unlock; 1833 1834 /* For an UPDATE, nothing may change except the priority/vcpu */ 1835 ret = -EINVAL; 1836 if (evtchnfd->type != data->u.evtchn.type) 1837 goto out_unlock; 1838 1839 /* 1840 * Port cannot change, and if it's zero that was an eventfd 1841 * which can't be changed either. 1842 */ 1843 if (!evtchnfd->deliver.port.port || 1844 evtchnfd->deliver.port.port != data->u.evtchn.deliver.port.port) 1845 goto out_unlock; 1846 1847 /* We only support 2 level event channels for now */ 1848 if (data->u.evtchn.deliver.port.priority != KVM_IRQ_ROUTING_XEN_EVTCHN_PRIO_2LEVEL) 1849 goto out_unlock; 1850 1851 evtchnfd->deliver.port.priority = data->u.evtchn.deliver.port.priority; 1852 if (evtchnfd->deliver.port.vcpu_id != data->u.evtchn.deliver.port.vcpu) { 1853 evtchnfd->deliver.port.vcpu_id = data->u.evtchn.deliver.port.vcpu; 1854 evtchnfd->deliver.port.vcpu_idx = -1; 1855 } 1856 ret = 0; 1857out_unlock: 1858 mutex_unlock(&kvm->arch.xen.xen_lock); 1859 return ret; 1860} 1861 1862/* 1863 * Configure the target (eventfd or local port delivery) for sending on 1864 * a given event channel. 1865 */ 1866static int kvm_xen_eventfd_assign(struct kvm *kvm, 1867 struct kvm_xen_hvm_attr *data) 1868{ 1869 u32 port = data->u.evtchn.send_port; 1870 struct eventfd_ctx *eventfd = NULL; 1871 struct evtchnfd *evtchnfd; 1872 int ret = -EINVAL; 1873 1874 evtchnfd = kzalloc(sizeof(struct evtchnfd), GFP_KERNEL); 1875 if (!evtchnfd) 1876 return -ENOMEM; 1877 1878 switch(data->u.evtchn.type) { 1879 case EVTCHNSTAT_ipi: 1880 /* IPI must map back to the same port# */ 1881 if (data->u.evtchn.deliver.port.port != data->u.evtchn.send_port) 1882 goto out_noeventfd; /* -EINVAL */ 1883 break; 1884 1885 case EVTCHNSTAT_interdomain: 1886 if (data->u.evtchn.deliver.port.port) { 1887 if (data->u.evtchn.deliver.port.port >= max_evtchn_port(kvm)) 1888 goto out_noeventfd; /* -EINVAL */ 1889 } else { 1890 eventfd = eventfd_ctx_fdget(data->u.evtchn.deliver.eventfd.fd); 1891 if (IS_ERR(eventfd)) { 1892 ret = PTR_ERR(eventfd); 1893 goto out_noeventfd; 1894 } 1895 } 1896 break; 1897 1898 case EVTCHNSTAT_virq: 1899 case EVTCHNSTAT_closed: 1900 case EVTCHNSTAT_unbound: 1901 case EVTCHNSTAT_pirq: 1902 default: /* Unknown event channel type */ 1903 goto out; /* -EINVAL */ 1904 } 1905 1906 evtchnfd->send_port = data->u.evtchn.send_port; 1907 evtchnfd->type = data->u.evtchn.type; 1908 if (eventfd) { 1909 evtchnfd->deliver.eventfd.ctx = eventfd; 1910 } else { 1911 /* We only support 2 level event channels for now */ 1912 if (data->u.evtchn.deliver.port.priority != KVM_IRQ_ROUTING_XEN_EVTCHN_PRIO_2LEVEL) 1913 goto out; /* -EINVAL; */ 1914 1915 evtchnfd->deliver.port.port = data->u.evtchn.deliver.port.port; 1916 evtchnfd->deliver.port.vcpu_id = data->u.evtchn.deliver.port.vcpu; 1917 evtchnfd->deliver.port.vcpu_idx = -1; 1918 evtchnfd->deliver.port.priority = data->u.evtchn.deliver.port.priority; 1919 } 1920 1921 mutex_lock(&kvm->arch.xen.xen_lock); 1922 ret = idr_alloc(&kvm->arch.xen.evtchn_ports, evtchnfd, port, port + 1, 1923 GFP_KERNEL); 1924 mutex_unlock(&kvm->arch.xen.xen_lock); 1925 if (ret >= 0) 1926 return 0; 1927 1928 if (ret == -ENOSPC) 1929 ret = -EEXIST; 1930out: 1931 if (eventfd) 1932 eventfd_ctx_put(eventfd); 1933out_noeventfd: 1934 kfree(evtchnfd); 1935 return ret; 1936} 1937 1938static int kvm_xen_eventfd_deassign(struct kvm *kvm, u32 port) 1939{ 1940 struct evtchnfd *evtchnfd; 1941 1942 mutex_lock(&kvm->arch.xen.xen_lock); 1943 evtchnfd = idr_remove(&kvm->arch.xen.evtchn_ports, port); 1944 mutex_unlock(&kvm->arch.xen.xen_lock); 1945 1946 if (!evtchnfd) 1947 return -ENOENT; 1948 1949 synchronize_srcu(&kvm->srcu); 1950 if (!evtchnfd->deliver.port.port) 1951 eventfd_ctx_put(evtchnfd->deliver.eventfd.ctx); 1952 kfree(evtchnfd); 1953 return 0; 1954} 1955 1956static int kvm_xen_eventfd_reset(struct kvm *kvm) 1957{ 1958 struct evtchnfd *evtchnfd, **all_evtchnfds; 1959 int i; 1960 int n = 0; 1961 1962 mutex_lock(&kvm->arch.xen.xen_lock); 1963 1964 /* 1965 * Because synchronize_srcu() cannot be called inside the 1966 * critical section, first collect all the evtchnfd objects 1967 * in an array as they are removed from evtchn_ports. 1968 */ 1969 idr_for_each_entry(&kvm->arch.xen.evtchn_ports, evtchnfd, i) 1970 n++; 1971 1972 all_evtchnfds = kmalloc_array(n, sizeof(struct evtchnfd *), GFP_KERNEL); 1973 if (!all_evtchnfds) { 1974 mutex_unlock(&kvm->arch.xen.xen_lock); 1975 return -ENOMEM; 1976 } 1977 1978 n = 0; 1979 idr_for_each_entry(&kvm->arch.xen.evtchn_ports, evtchnfd, i) { 1980 all_evtchnfds[n++] = evtchnfd; 1981 idr_remove(&kvm->arch.xen.evtchn_ports, evtchnfd->send_port); 1982 } 1983 mutex_unlock(&kvm->arch.xen.xen_lock); 1984 1985 synchronize_srcu(&kvm->srcu); 1986 1987 while (n--) { 1988 evtchnfd = all_evtchnfds[n]; 1989 if (!evtchnfd->deliver.port.port) 1990 eventfd_ctx_put(evtchnfd->deliver.eventfd.ctx); 1991 kfree(evtchnfd); 1992 } 1993 kfree(all_evtchnfds); 1994 1995 return 0; 1996} 1997 1998static int kvm_xen_setattr_evtchn(struct kvm *kvm, struct kvm_xen_hvm_attr *data) 1999{ 2000 u32 port = data->u.evtchn.send_port; 2001 2002 if (data->u.evtchn.flags == KVM_XEN_EVTCHN_RESET) 2003 return kvm_xen_eventfd_reset(kvm); 2004 2005 if (!port || port >= max_evtchn_port(kvm)) 2006 return -EINVAL; 2007 2008 if (data->u.evtchn.flags == KVM_XEN_EVTCHN_DEASSIGN) 2009 return kvm_xen_eventfd_deassign(kvm, port); 2010 if (data->u.evtchn.flags == KVM_XEN_EVTCHN_UPDATE) 2011 return kvm_xen_eventfd_update(kvm, data); 2012 if (data->u.evtchn.flags) 2013 return -EINVAL; 2014 2015 return kvm_xen_eventfd_assign(kvm, data); 2016} 2017 2018static bool kvm_xen_hcall_evtchn_send(struct kvm_vcpu *vcpu, u64 param, u64 *r) 2019{ 2020 struct evtchnfd *evtchnfd; 2021 struct evtchn_send send; 2022 struct x86_exception e; 2023 2024 /* Sanity check: this structure is the same for 32-bit and 64-bit */ 2025 BUILD_BUG_ON(sizeof(send) != 4); 2026 if (kvm_read_guest_virt(vcpu, param, &send, sizeof(send), &e)) { 2027 *r = -EFAULT; 2028 return true; 2029 } 2030 2031 /* 2032 * evtchnfd is protected by kvm->srcu; the idr lookup instead 2033 * is protected by RCU. 2034 */ 2035 rcu_read_lock(); 2036 evtchnfd = idr_find(&vcpu->kvm->arch.xen.evtchn_ports, send.port); 2037 rcu_read_unlock(); 2038 if (!evtchnfd) 2039 return false; 2040 2041 if (evtchnfd->deliver.port.port) { 2042 int ret = kvm_xen_set_evtchn(&evtchnfd->deliver.port, vcpu->kvm); 2043 if (ret < 0 && ret != -ENOTCONN) 2044 return false; 2045 } else { 2046 eventfd_signal(evtchnfd->deliver.eventfd.ctx, 1); 2047 } 2048 2049 *r = 0; 2050 return true; 2051} 2052 2053void kvm_xen_init_vcpu(struct kvm_vcpu *vcpu) 2054{ 2055 vcpu->arch.xen.vcpu_id = vcpu->vcpu_idx; 2056 vcpu->arch.xen.poll_evtchn = 0; 2057 2058 timer_setup(&vcpu->arch.xen.poll_timer, cancel_evtchn_poll, 0); 2059 2060 kvm_gpc_init(&vcpu->arch.xen.runstate_cache, vcpu->kvm, NULL, 2061 KVM_HOST_USES_PFN); 2062 kvm_gpc_init(&vcpu->arch.xen.runstate2_cache, vcpu->kvm, NULL, 2063 KVM_HOST_USES_PFN); 2064 kvm_gpc_init(&vcpu->arch.xen.vcpu_info_cache, vcpu->kvm, NULL, 2065 KVM_HOST_USES_PFN); 2066 kvm_gpc_init(&vcpu->arch.xen.vcpu_time_info_cache, vcpu->kvm, NULL, 2067 KVM_HOST_USES_PFN); 2068} 2069 2070void kvm_xen_destroy_vcpu(struct kvm_vcpu *vcpu) 2071{ 2072 if (kvm_xen_timer_enabled(vcpu)) 2073 kvm_xen_stop_timer(vcpu); 2074 2075 kvm_gpc_deactivate(&vcpu->arch.xen.runstate_cache); 2076 kvm_gpc_deactivate(&vcpu->arch.xen.runstate2_cache); 2077 kvm_gpc_deactivate(&vcpu->arch.xen.vcpu_info_cache); 2078 kvm_gpc_deactivate(&vcpu->arch.xen.vcpu_time_info_cache); 2079 2080 del_timer_sync(&vcpu->arch.xen.poll_timer); 2081} 2082 2083void kvm_xen_update_tsc_info(struct kvm_vcpu *vcpu) 2084{ 2085 struct kvm_cpuid_entry2 *entry; 2086 u32 function; 2087 2088 if (!vcpu->arch.xen.cpuid.base) 2089 return; 2090 2091 function = vcpu->arch.xen.cpuid.base | XEN_CPUID_LEAF(3); 2092 if (function > vcpu->arch.xen.cpuid.limit) 2093 return; 2094 2095 entry = kvm_find_cpuid_entry_index(vcpu, function, 1); 2096 if (entry) { 2097 entry->ecx = vcpu->arch.hv_clock.tsc_to_system_mul; 2098 entry->edx = vcpu->arch.hv_clock.tsc_shift; 2099 } 2100 2101 entry = kvm_find_cpuid_entry_index(vcpu, function, 2); 2102 if (entry) 2103 entry->eax = vcpu->arch.hw_tsc_khz; 2104} 2105 2106void kvm_xen_init_vm(struct kvm *kvm) 2107{ 2108 mutex_init(&kvm->arch.xen.xen_lock); 2109 idr_init(&kvm->arch.xen.evtchn_ports); 2110 kvm_gpc_init(&kvm->arch.xen.shinfo_cache, kvm, NULL, KVM_HOST_USES_PFN); 2111} 2112 2113void kvm_xen_destroy_vm(struct kvm *kvm) 2114{ 2115 struct evtchnfd *evtchnfd; 2116 int i; 2117 2118 kvm_gpc_deactivate(&kvm->arch.xen.shinfo_cache); 2119 2120 idr_for_each_entry(&kvm->arch.xen.evtchn_ports, evtchnfd, i) { 2121 if (!evtchnfd->deliver.port.port) 2122 eventfd_ctx_put(evtchnfd->deliver.eventfd.ctx); 2123 kfree(evtchnfd); 2124 } 2125 idr_destroy(&kvm->arch.xen.evtchn_ports); 2126 2127 if (kvm->arch.xen_hvm_config.msr) 2128 static_branch_slow_dec_deferred(&kvm_xen_enabled); 2129} 2130