]>
Commit | Line | Data |
---|---|---|
05330448 AL |
1 | /* |
2 | * QEMU KVM support | |
3 | * | |
4 | * Copyright IBM, Corp. 2008 | |
5832d1f2 | 5 | * Red Hat, Inc. 2008 |
05330448 AL |
6 | * |
7 | * Authors: | |
8 | * Anthony Liguori <aliguori@us.ibm.com> | |
5832d1f2 | 9 | * Glauber Costa <gcosta@redhat.com> |
05330448 AL |
10 | * |
11 | * This work is licensed under the terms of the GNU GPL, version 2 or later. | |
12 | * See the COPYING file in the top-level directory. | |
13 | * | |
14 | */ | |
15 | ||
d38ea87a | 16 | #include "qemu/osdep.h" |
05330448 | 17 | #include <sys/ioctl.h> |
05330448 AL |
18 | |
19 | #include <linux/kvm.h> | |
20 | ||
1de7afc9 PB |
21 | #include "qemu/atomic.h" |
22 | #include "qemu/option.h" | |
23 | #include "qemu/config-file.h" | |
4b3cfe72 | 24 | #include "qemu/error-report.h" |
556969e9 | 25 | #include "qapi/error.h" |
a2cb15b0 | 26 | #include "hw/pci/msi.h" |
d1f6af6a | 27 | #include "hw/pci/msix.h" |
d426d9fb | 28 | #include "hw/s390x/adapter.h" |
022c62cb | 29 | #include "exec/gdbstub.h" |
8571ed35 | 30 | #include "sysemu/kvm_int.h" |
54d31236 | 31 | #include "sysemu/runstate.h" |
d2528bdc | 32 | #include "sysemu/cpus.h" |
46517dd4 | 33 | #include "sysemu/sysemu.h" |
1de7afc9 | 34 | #include "qemu/bswap.h" |
022c62cb | 35 | #include "exec/memory.h" |
747afd5b | 36 | #include "exec/ram_addr.h" |
022c62cb | 37 | #include "exec/address-spaces.h" |
1de7afc9 | 38 | #include "qemu/event_notifier.h" |
db725815 | 39 | #include "qemu/main-loop.h" |
92229a57 | 40 | #include "trace.h" |
197e3524 | 41 | #include "hw/irq.h" |
b20e3780 | 42 | #include "sysemu/sev.h" |
f5948942 | 43 | #include "sysemu/balloon.h" |
05330448 | 44 | |
135a129a AK |
45 | #include "hw/boards.h" |
46 | ||
d2f2b8a7 SH |
47 | /* This check must be after config-host.h is included */ |
48 | #ifdef CONFIG_EVENTFD | |
49 | #include <sys/eventfd.h> | |
50 | #endif | |
51 | ||
bc92e4e9 AJ |
52 | /* KVM uses PAGE_SIZE in its definition of KVM_COALESCED_MMIO_MAX. We |
53 | * need to use the real host PAGE_SIZE, as that's what KVM will use. | |
54 | */ | |
55 | #define PAGE_SIZE getpagesize() | |
f65ed4c1 | 56 | |
05330448 AL |
57 | //#define DEBUG_KVM |
58 | ||
59 | #ifdef DEBUG_KVM | |
8c0d577e | 60 | #define DPRINTF(fmt, ...) \ |
05330448 AL |
61 | do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0) |
62 | #else | |
8c0d577e | 63 | #define DPRINTF(fmt, ...) \ |
05330448 AL |
64 | do { } while (0) |
65 | #endif | |
66 | ||
04fa27f5 JK |
67 | #define KVM_MSI_HASHTAB_SIZE 256 |
68 | ||
4c055ab5 GZ |
69 | struct KVMParkedVcpu { |
70 | unsigned long vcpu_id; | |
71 | int kvm_fd; | |
72 | QLIST_ENTRY(KVMParkedVcpu) node; | |
73 | }; | |
74 | ||
9d1c35df | 75 | struct KVMState |
05330448 | 76 | { |
fc02086b EH |
77 | AccelState parent_obj; |
78 | ||
fb541ca5 | 79 | int nr_slots; |
05330448 AL |
80 | int fd; |
81 | int vmfd; | |
f65ed4c1 | 82 | int coalesced_mmio; |
e6d34aee | 83 | int coalesced_pio; |
62a2744c | 84 | struct kvm_coalesced_mmio_ring *coalesced_mmio_ring; |
1cae88b9 | 85 | bool coalesced_flush_in_progress; |
a0fb002c | 86 | int vcpu_events; |
b0b1d690 | 87 | int robust_singlestep; |
ff44f1a3 | 88 | int debugregs; |
e22a25c9 | 89 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
b58deb34 | 90 | QTAILQ_HEAD(, kvm_sw_breakpoint) kvm_sw_breakpoints; |
e22a25c9 | 91 | #endif |
ebbfef2f | 92 | int max_nested_state_len; |
d2f2b8a7 | 93 | int many_ioeventfds; |
3ab73842 | 94 | int intx_set_mask; |
62dd4eda | 95 | bool sync_mmu; |
ff4aa114 | 96 | bool manual_dirty_log_protect; |
92e4b519 DG |
97 | /* The man page (and posix) say ioctl numbers are signed int, but |
98 | * they're not. Linux, glibc and *BSD all treat ioctl numbers as | |
99 | * unsigned, and treating them as signed here can break things */ | |
e333cd69 | 100 | unsigned irq_set_ioctl; |
aed6efb9 | 101 | unsigned int sigmask_len; |
197e3524 | 102 | GHashTable *gsimap; |
84b058d7 JK |
103 | #ifdef KVM_CAP_IRQ_ROUTING |
104 | struct kvm_irq_routing *irq_routes; | |
105 | int nr_allocated_irq_routes; | |
8269fb70 | 106 | unsigned long *used_gsi_bitmap; |
4e2e4e63 | 107 | unsigned int gsi_count; |
b58deb34 | 108 | QTAILQ_HEAD(, KVMMSIRoute) msi_hashtab[KVM_MSI_HASHTAB_SIZE]; |
84b058d7 | 109 | #endif |
7bbda04c | 110 | KVMMemoryListener memory_listener; |
4c055ab5 | 111 | QLIST_HEAD(, KVMParkedVcpu) kvm_parked_vcpus; |
b20e3780 BS |
112 | |
113 | /* memory encryption */ | |
114 | void *memcrypt_handle; | |
54e89539 | 115 | int (*memcrypt_encrypt_data)(void *handle, uint8_t *ptr, uint64_t len); |
8072aae3 AK |
116 | |
117 | /* For "info mtree -f" to tell if an MR is registered in KVM */ | |
118 | int nr_as; | |
119 | struct KVMAs { | |
120 | KVMMemoryListener *ml; | |
121 | AddressSpace *as; | |
122 | } *as; | |
9d1c35df | 123 | }; |
05330448 | 124 | |
6a7af8cb | 125 | KVMState *kvm_state; |
3d4b2649 | 126 | bool kvm_kernel_irqchip; |
15eafc2e | 127 | bool kvm_split_irqchip; |
7ae26bd4 | 128 | bool kvm_async_interrupts_allowed; |
215e79c0 | 129 | bool kvm_halt_in_kernel_allowed; |
69e03ae6 | 130 | bool kvm_eventfds_allowed; |
cc7e0ddf | 131 | bool kvm_irqfds_allowed; |
f41389ae | 132 | bool kvm_resamplefds_allowed; |
614e41bc | 133 | bool kvm_msi_via_irqfd_allowed; |
f3e1bed8 | 134 | bool kvm_gsi_routing_allowed; |
76fe21de | 135 | bool kvm_gsi_direct_mapping; |
13eed94e | 136 | bool kvm_allowed; |
df9c8b75 | 137 | bool kvm_readonly_mem_allowed; |
d0a073a1 | 138 | bool kvm_vm_attributes_allowed; |
50bf31b9 | 139 | bool kvm_direct_msi_allowed; |
35108223 | 140 | bool kvm_ioeventfd_any_length_allowed; |
767a554a | 141 | bool kvm_msi_use_devid; |
cf0f7cf9 | 142 | static bool kvm_immediate_exit; |
05330448 | 143 | |
94a8d39a JK |
144 | static const KVMCapabilityInfo kvm_required_capabilites[] = { |
145 | KVM_CAP_INFO(USER_MEMORY), | |
146 | KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS), | |
89de4b91 | 147 | KVM_CAP_INFO(JOIN_MEMORY_REGIONS_WORKS), |
94a8d39a JK |
148 | KVM_CAP_LAST_INFO |
149 | }; | |
150 | ||
36adac49 PX |
151 | #define kvm_slots_lock(kml) qemu_mutex_lock(&(kml)->slots_lock) |
152 | #define kvm_slots_unlock(kml) qemu_mutex_unlock(&(kml)->slots_lock) | |
153 | ||
44f2e6c1 BR |
154 | int kvm_get_max_memslots(void) |
155 | { | |
156 | KVMState *s = KVM_STATE(current_machine->accelerator); | |
157 | ||
158 | return s->nr_slots; | |
159 | } | |
160 | ||
b20e3780 BS |
161 | bool kvm_memcrypt_enabled(void) |
162 | { | |
163 | if (kvm_state && kvm_state->memcrypt_handle) { | |
164 | return true; | |
165 | } | |
166 | ||
167 | return false; | |
168 | } | |
169 | ||
54e89539 BS |
170 | int kvm_memcrypt_encrypt_data(uint8_t *ptr, uint64_t len) |
171 | { | |
172 | if (kvm_state->memcrypt_handle && | |
173 | kvm_state->memcrypt_encrypt_data) { | |
174 | return kvm_state->memcrypt_encrypt_data(kvm_state->memcrypt_handle, | |
175 | ptr, len); | |
176 | } | |
177 | ||
178 | return 1; | |
179 | } | |
180 | ||
36adac49 | 181 | /* Called with KVMMemoryListener.slots_lock held */ |
7bbda04c | 182 | static KVMSlot *kvm_get_free_slot(KVMMemoryListener *kml) |
05330448 | 183 | { |
7bbda04c | 184 | KVMState *s = kvm_state; |
05330448 AL |
185 | int i; |
186 | ||
fb541ca5 | 187 | for (i = 0; i < s->nr_slots; i++) { |
7bbda04c PB |
188 | if (kml->slots[i].memory_size == 0) { |
189 | return &kml->slots[i]; | |
a426e122 | 190 | } |
05330448 AL |
191 | } |
192 | ||
b8865591 IM |
193 | return NULL; |
194 | } | |
195 | ||
196 | bool kvm_has_free_slot(MachineState *ms) | |
197 | { | |
7bbda04c | 198 | KVMState *s = KVM_STATE(ms->accelerator); |
36adac49 PX |
199 | bool result; |
200 | KVMMemoryListener *kml = &s->memory_listener; | |
201 | ||
202 | kvm_slots_lock(kml); | |
203 | result = !!kvm_get_free_slot(kml); | |
204 | kvm_slots_unlock(kml); | |
7bbda04c | 205 | |
36adac49 | 206 | return result; |
b8865591 IM |
207 | } |
208 | ||
36adac49 | 209 | /* Called with KVMMemoryListener.slots_lock held */ |
7bbda04c | 210 | static KVMSlot *kvm_alloc_slot(KVMMemoryListener *kml) |
b8865591 | 211 | { |
7bbda04c | 212 | KVMSlot *slot = kvm_get_free_slot(kml); |
b8865591 IM |
213 | |
214 | if (slot) { | |
215 | return slot; | |
216 | } | |
217 | ||
d3f8d37f AL |
218 | fprintf(stderr, "%s: no free slot available\n", __func__); |
219 | abort(); | |
220 | } | |
221 | ||
7bbda04c | 222 | static KVMSlot *kvm_lookup_matching_slot(KVMMemoryListener *kml, |
a8170e5e | 223 | hwaddr start_addr, |
2747e716 | 224 | hwaddr size) |
d3f8d37f | 225 | { |
7bbda04c | 226 | KVMState *s = kvm_state; |
d3f8d37f AL |
227 | int i; |
228 | ||
fb541ca5 | 229 | for (i = 0; i < s->nr_slots; i++) { |
7bbda04c | 230 | KVMSlot *mem = &kml->slots[i]; |
d3f8d37f | 231 | |
2747e716 | 232 | if (start_addr == mem->start_addr && size == mem->memory_size) { |
d3f8d37f AL |
233 | return mem; |
234 | } | |
235 | } | |
236 | ||
05330448 AL |
237 | return NULL; |
238 | } | |
239 | ||
5ea69c2e DH |
240 | /* |
241 | * Calculate and align the start address and the size of the section. | |
242 | * Return the size. If the size is 0, the aligned section is empty. | |
243 | */ | |
244 | static hwaddr kvm_align_section(MemoryRegionSection *section, | |
245 | hwaddr *start) | |
246 | { | |
247 | hwaddr size = int128_get64(section->size); | |
a6ffc423 | 248 | hwaddr delta, aligned; |
5ea69c2e DH |
249 | |
250 | /* kvm works in page size chunks, but the function may be called | |
251 | with sub-page size and unaligned start address. Pad the start | |
252 | address to next and truncate size to previous page boundary. */ | |
a6ffc423 DH |
253 | aligned = ROUND_UP(section->offset_within_address_space, |
254 | qemu_real_host_page_size); | |
255 | delta = aligned - section->offset_within_address_space; | |
256 | *start = aligned; | |
5ea69c2e DH |
257 | if (delta > size) { |
258 | return 0; | |
259 | } | |
5ea69c2e | 260 | |
a6ffc423 | 261 | return (size - delta) & qemu_real_host_page_mask; |
5ea69c2e DH |
262 | } |
263 | ||
9f213ed9 | 264 | int kvm_physical_memory_addr_from_host(KVMState *s, void *ram, |
a8170e5e | 265 | hwaddr *phys_addr) |
983dfc3b | 266 | { |
7bbda04c | 267 | KVMMemoryListener *kml = &s->memory_listener; |
36adac49 | 268 | int i, ret = 0; |
983dfc3b | 269 | |
36adac49 | 270 | kvm_slots_lock(kml); |
fb541ca5 | 271 | for (i = 0; i < s->nr_slots; i++) { |
7bbda04c | 272 | KVMSlot *mem = &kml->slots[i]; |
983dfc3b | 273 | |
9f213ed9 AK |
274 | if (ram >= mem->ram && ram < mem->ram + mem->memory_size) { |
275 | *phys_addr = mem->start_addr + (ram - mem->ram); | |
36adac49 PX |
276 | ret = 1; |
277 | break; | |
983dfc3b HY |
278 | } |
279 | } | |
36adac49 | 280 | kvm_slots_unlock(kml); |
983dfc3b | 281 | |
36adac49 | 282 | return ret; |
983dfc3b HY |
283 | } |
284 | ||
6c090d4a | 285 | static int kvm_set_user_memory_region(KVMMemoryListener *kml, KVMSlot *slot, bool new) |
5832d1f2 | 286 | { |
7bbda04c | 287 | KVMState *s = kvm_state; |
5832d1f2 | 288 | struct kvm_userspace_memory_region mem; |
fe29141b | 289 | int ret; |
5832d1f2 | 290 | |
38bfe691 | 291 | mem.slot = slot->slot | (kml->as_id << 16); |
5832d1f2 | 292 | mem.guest_phys_addr = slot->start_addr; |
9f213ed9 | 293 | mem.userspace_addr = (unsigned long)slot->ram; |
5832d1f2 | 294 | mem.flags = slot->flags; |
651eb0f4 | 295 | |
6c090d4a | 296 | if (slot->memory_size && !new && (mem.flags ^ slot->old_flags) & KVM_MEM_READONLY) { |
235e8982 JJ |
297 | /* Set the slot size to 0 before setting the slot to the desired |
298 | * value. This is needed based on KVM commit 75d61fbc. */ | |
299 | mem.memory_size = 0; | |
300 | kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem); | |
301 | } | |
302 | mem.memory_size = slot->memory_size; | |
fe29141b | 303 | ret = kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem); |
6c090d4a | 304 | slot->old_flags = mem.flags; |
fe29141b AK |
305 | trace_kvm_set_user_memory(mem.slot, mem.flags, mem.guest_phys_addr, |
306 | mem.memory_size, mem.userspace_addr, ret); | |
307 | return ret; | |
5832d1f2 AL |
308 | } |
309 | ||
4c055ab5 GZ |
310 | int kvm_destroy_vcpu(CPUState *cpu) |
311 | { | |
312 | KVMState *s = kvm_state; | |
313 | long mmap_size; | |
314 | struct KVMParkedVcpu *vcpu = NULL; | |
315 | int ret = 0; | |
316 | ||
317 | DPRINTF("kvm_destroy_vcpu\n"); | |
318 | ||
b1115c99 LA |
319 | ret = kvm_arch_destroy_vcpu(cpu); |
320 | if (ret < 0) { | |
321 | goto err; | |
322 | } | |
323 | ||
4c055ab5 GZ |
324 | mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0); |
325 | if (mmap_size < 0) { | |
326 | ret = mmap_size; | |
327 | DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n"); | |
328 | goto err; | |
329 | } | |
330 | ||
331 | ret = munmap(cpu->kvm_run, mmap_size); | |
332 | if (ret < 0) { | |
333 | goto err; | |
334 | } | |
335 | ||
336 | vcpu = g_malloc0(sizeof(*vcpu)); | |
337 | vcpu->vcpu_id = kvm_arch_vcpu_id(cpu); | |
338 | vcpu->kvm_fd = cpu->kvm_fd; | |
339 | QLIST_INSERT_HEAD(&kvm_state->kvm_parked_vcpus, vcpu, node); | |
340 | err: | |
341 | return ret; | |
342 | } | |
343 | ||
344 | static int kvm_get_vcpu(KVMState *s, unsigned long vcpu_id) | |
345 | { | |
346 | struct KVMParkedVcpu *cpu; | |
347 | ||
348 | QLIST_FOREACH(cpu, &s->kvm_parked_vcpus, node) { | |
349 | if (cpu->vcpu_id == vcpu_id) { | |
350 | int kvm_fd; | |
351 | ||
352 | QLIST_REMOVE(cpu, node); | |
353 | kvm_fd = cpu->kvm_fd; | |
354 | g_free(cpu); | |
355 | return kvm_fd; | |
356 | } | |
357 | } | |
358 | ||
359 | return kvm_vm_ioctl(s, KVM_CREATE_VCPU, (void *)vcpu_id); | |
360 | } | |
361 | ||
504134d2 | 362 | int kvm_init_vcpu(CPUState *cpu) |
05330448 AL |
363 | { |
364 | KVMState *s = kvm_state; | |
365 | long mmap_size; | |
366 | int ret; | |
367 | ||
8c0d577e | 368 | DPRINTF("kvm_init_vcpu\n"); |
05330448 | 369 | |
4c055ab5 | 370 | ret = kvm_get_vcpu(s, kvm_arch_vcpu_id(cpu)); |
05330448 | 371 | if (ret < 0) { |
8c0d577e | 372 | DPRINTF("kvm_create_vcpu failed\n"); |
05330448 AL |
373 | goto err; |
374 | } | |
375 | ||
8737c51c | 376 | cpu->kvm_fd = ret; |
a60f24b5 | 377 | cpu->kvm_state = s; |
99f31832 | 378 | cpu->vcpu_dirty = true; |
05330448 AL |
379 | |
380 | mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0); | |
381 | if (mmap_size < 0) { | |
748a680b | 382 | ret = mmap_size; |
8c0d577e | 383 | DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n"); |
05330448 AL |
384 | goto err; |
385 | } | |
386 | ||
f7575c96 | 387 | cpu->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED, |
8737c51c | 388 | cpu->kvm_fd, 0); |
f7575c96 | 389 | if (cpu->kvm_run == MAP_FAILED) { |
05330448 | 390 | ret = -errno; |
8c0d577e | 391 | DPRINTF("mmap'ing vcpu state failed\n"); |
05330448 AL |
392 | goto err; |
393 | } | |
394 | ||
a426e122 JK |
395 | if (s->coalesced_mmio && !s->coalesced_mmio_ring) { |
396 | s->coalesced_mmio_ring = | |
f7575c96 | 397 | (void *)cpu->kvm_run + s->coalesced_mmio * PAGE_SIZE; |
a426e122 | 398 | } |
62a2744c | 399 | |
20d695a9 | 400 | ret = kvm_arch_init_vcpu(cpu); |
05330448 AL |
401 | err: |
402 | return ret; | |
403 | } | |
404 | ||
5832d1f2 AL |
405 | /* |
406 | * dirty pages logging control | |
407 | */ | |
25254bbc | 408 | |
d6ff5cbc | 409 | static int kvm_mem_flags(MemoryRegion *mr) |
25254bbc | 410 | { |
d6ff5cbc | 411 | bool readonly = mr->readonly || memory_region_is_romd(mr); |
235e8982 | 412 | int flags = 0; |
d6ff5cbc AJ |
413 | |
414 | if (memory_region_get_dirty_log_mask(mr) != 0) { | |
415 | flags |= KVM_MEM_LOG_DIRTY_PAGES; | |
416 | } | |
235e8982 JJ |
417 | if (readonly && kvm_readonly_mem_allowed) { |
418 | flags |= KVM_MEM_READONLY; | |
419 | } | |
420 | return flags; | |
25254bbc MT |
421 | } |
422 | ||
36adac49 | 423 | /* Called with KVMMemoryListener.slots_lock held */ |
7bbda04c PB |
424 | static int kvm_slot_update_flags(KVMMemoryListener *kml, KVMSlot *mem, |
425 | MemoryRegion *mr) | |
5832d1f2 | 426 | { |
d6ff5cbc | 427 | mem->flags = kvm_mem_flags(mr); |
5832d1f2 | 428 | |
4495d6a7 | 429 | /* If nothing changed effectively, no need to issue ioctl */ |
6c090d4a | 430 | if (mem->flags == mem->old_flags) { |
25254bbc | 431 | return 0; |
4495d6a7 JK |
432 | } |
433 | ||
6c090d4a | 434 | return kvm_set_user_memory_region(kml, mem, false); |
5832d1f2 AL |
435 | } |
436 | ||
7bbda04c PB |
437 | static int kvm_section_update_flags(KVMMemoryListener *kml, |
438 | MemoryRegionSection *section) | |
25254bbc | 439 | { |
343562e8 DH |
440 | hwaddr start_addr, size; |
441 | KVMSlot *mem; | |
36adac49 | 442 | int ret = 0; |
25254bbc | 443 | |
343562e8 DH |
444 | size = kvm_align_section(section, &start_addr); |
445 | if (!size) { | |
ea8cb1a8 | 446 | return 0; |
25254bbc | 447 | } |
343562e8 | 448 | |
36adac49 PX |
449 | kvm_slots_lock(kml); |
450 | ||
343562e8 DH |
451 | mem = kvm_lookup_matching_slot(kml, start_addr, size); |
452 | if (!mem) { | |
e377e87c | 453 | /* We don't have a slot if we want to trap every access. */ |
36adac49 | 454 | goto out; |
343562e8 DH |
455 | } |
456 | ||
36adac49 PX |
457 | ret = kvm_slot_update_flags(kml, mem, section->mr); |
458 | ||
459 | out: | |
460 | kvm_slots_unlock(kml); | |
461 | return ret; | |
25254bbc MT |
462 | } |
463 | ||
a01672d3 | 464 | static void kvm_log_start(MemoryListener *listener, |
b2dfd71c PB |
465 | MemoryRegionSection *section, |
466 | int old, int new) | |
5832d1f2 | 467 | { |
7bbda04c | 468 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); |
a01672d3 AK |
469 | int r; |
470 | ||
b2dfd71c PB |
471 | if (old != 0) { |
472 | return; | |
473 | } | |
474 | ||
7bbda04c | 475 | r = kvm_section_update_flags(kml, section); |
a01672d3 AK |
476 | if (r < 0) { |
477 | abort(); | |
478 | } | |
5832d1f2 AL |
479 | } |
480 | ||
a01672d3 | 481 | static void kvm_log_stop(MemoryListener *listener, |
b2dfd71c PB |
482 | MemoryRegionSection *section, |
483 | int old, int new) | |
5832d1f2 | 484 | { |
7bbda04c | 485 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); |
a01672d3 AK |
486 | int r; |
487 | ||
b2dfd71c PB |
488 | if (new != 0) { |
489 | return; | |
490 | } | |
491 | ||
7bbda04c | 492 | r = kvm_section_update_flags(kml, section); |
a01672d3 AK |
493 | if (r < 0) { |
494 | abort(); | |
495 | } | |
5832d1f2 AL |
496 | } |
497 | ||
8369e01c | 498 | /* get kvm's dirty pages bitmap and update qemu's */ |
ffcde12f AK |
499 | static int kvm_get_dirty_pages_log_range(MemoryRegionSection *section, |
500 | unsigned long *bitmap) | |
96c1606b | 501 | { |
8e41fb63 FZ |
502 | ram_addr_t start = section->offset_within_region + |
503 | memory_region_get_ram_addr(section->mr); | |
5ff7fb77 JQ |
504 | ram_addr_t pages = int128_get64(section->size) / getpagesize(); |
505 | ||
506 | cpu_physical_memory_set_dirty_lebitmap(bitmap, start, pages); | |
8369e01c | 507 | return 0; |
96c1606b AG |
508 | } |
509 | ||
8369e01c MT |
510 | #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1)) |
511 | ||
5832d1f2 | 512 | /** |
4a12a11a | 513 | * kvm_physical_sync_dirty_bitmap - Sync dirty bitmap from kernel space |
5832d1f2 | 514 | * |
4a12a11a PX |
515 | * This function will first try to fetch dirty bitmap from the kernel, |
516 | * and then updates qemu's dirty bitmap. | |
517 | * | |
36adac49 PX |
518 | * NOTE: caller must be with kml->slots_lock held. |
519 | * | |
4a12a11a PX |
520 | * @kml: the KVM memory listener object |
521 | * @section: the memory section to sync the dirty bitmap with | |
5832d1f2 | 522 | */ |
7bbda04c PB |
523 | static int kvm_physical_sync_dirty_bitmap(KVMMemoryListener *kml, |
524 | MemoryRegionSection *section) | |
5832d1f2 AL |
525 | { |
526 | KVMState *s = kvm_state; | |
714f78c5 | 527 | struct kvm_dirty_log d = {}; |
151f7749 | 528 | KVMSlot *mem; |
67548f09 | 529 | hwaddr start_addr, size; |
36adac49 | 530 | int ret = 0; |
67548f09 DH |
531 | |
532 | size = kvm_align_section(section, &start_addr); | |
533 | if (size) { | |
534 | mem = kvm_lookup_matching_slot(kml, start_addr, size); | |
535 | if (!mem) { | |
e377e87c | 536 | /* We don't have a slot if we want to trap every access. */ |
36adac49 | 537 | goto out; |
151f7749 | 538 | } |
5832d1f2 | 539 | |
51b0c606 MT |
540 | /* XXX bad kernel interface alert |
541 | * For dirty bitmap, kernel allocates array of size aligned to | |
542 | * bits-per-long. But for case when the kernel is 64bits and | |
543 | * the userspace is 32bits, userspace can't align to the same | |
544 | * bits-per-long, since sizeof(long) is different between kernel | |
545 | * and user space. This way, userspace will provide buffer which | |
546 | * may be 4 bytes less than the kernel will use, resulting in | |
547 | * userspace memory corruption (which is not detectable by valgrind | |
548 | * too, in most cases). | |
549 | * So for now, let's align to 64 instead of HOST_LONG_BITS here, in | |
cb8d4c8f | 550 | * a hope that sizeof(long) won't become >8 any time soon. |
51b0c606 MT |
551 | */ |
552 | size = ALIGN(((mem->memory_size) >> TARGET_PAGE_BITS), | |
553 | /*HOST_LONG_BITS*/ 64) / 8; | |
9f4bf4ba PX |
554 | if (!mem->dirty_bmap) { |
555 | /* Allocate on the first log_sync, once and for all */ | |
556 | mem->dirty_bmap = g_malloc0(size); | |
557 | } | |
5832d1f2 | 558 | |
9f4bf4ba | 559 | d.dirty_bitmap = mem->dirty_bmap; |
38bfe691 | 560 | d.slot = mem->slot | (kml->as_id << 16); |
50212d63 | 561 | if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) { |
8c0d577e | 562 | DPRINTF("ioctl failed %d\n", errno); |
36adac49 PX |
563 | ret = -1; |
564 | goto out; | |
151f7749 | 565 | } |
5832d1f2 | 566 | |
ffcde12f | 567 | kvm_get_dirty_pages_log_range(section, d.dirty_bitmap); |
5832d1f2 | 568 | } |
36adac49 PX |
569 | out: |
570 | return ret; | |
5832d1f2 AL |
571 | } |
572 | ||
ff4aa114 PX |
573 | /* Alignment requirement for KVM_CLEAR_DIRTY_LOG - 64 pages */ |
574 | #define KVM_CLEAR_LOG_SHIFT 6 | |
575 | #define KVM_CLEAR_LOG_ALIGN (qemu_real_host_page_size << KVM_CLEAR_LOG_SHIFT) | |
576 | #define KVM_CLEAR_LOG_MASK (-KVM_CLEAR_LOG_ALIGN) | |
577 | ||
4222147d PB |
578 | static int kvm_log_clear_one_slot(KVMSlot *mem, int as_id, uint64_t start, |
579 | uint64_t size) | |
ff4aa114 PX |
580 | { |
581 | KVMState *s = kvm_state; | |
4222147d | 582 | uint64_t end, bmap_start, start_delta, bmap_npages; |
ff4aa114 | 583 | struct kvm_clear_dirty_log d; |
ff4aa114 | 584 | unsigned long *bmap_clear = NULL, psize = qemu_real_host_page_size; |
4222147d | 585 | int ret; |
ff4aa114 PX |
586 | |
587 | /* | |
588 | * We need to extend either the start or the size or both to | |
589 | * satisfy the KVM interface requirement. Firstly, do the start | |
590 | * page alignment on 64 host pages | |
591 | */ | |
592 | bmap_start = (start - mem->start_addr) & KVM_CLEAR_LOG_MASK; | |
593 | start_delta = start - mem->start_addr - bmap_start; | |
594 | bmap_start /= psize; | |
595 | ||
596 | /* | |
597 | * The kernel interface has restriction on the size too, that either: | |
598 | * | |
599 | * (1) the size is 64 host pages aligned (just like the start), or | |
600 | * (2) the size fills up until the end of the KVM memslot. | |
601 | */ | |
602 | bmap_npages = DIV_ROUND_UP(size + start_delta, KVM_CLEAR_LOG_ALIGN) | |
603 | << KVM_CLEAR_LOG_SHIFT; | |
604 | end = mem->memory_size / psize; | |
605 | if (bmap_npages > end - bmap_start) { | |
606 | bmap_npages = end - bmap_start; | |
607 | } | |
608 | start_delta /= psize; | |
609 | ||
610 | /* | |
611 | * Prepare the bitmap to clear dirty bits. Here we must guarantee | |
612 | * that we won't clear any unknown dirty bits otherwise we might | |
613 | * accidentally clear some set bits which are not yet synced from | |
614 | * the kernel into QEMU's bitmap, then we'll lose track of the | |
615 | * guest modifications upon those pages (which can directly lead | |
616 | * to guest data loss or panic after migration). | |
617 | * | |
618 | * Layout of the KVMSlot.dirty_bmap: | |
619 | * | |
620 | * |<-------- bmap_npages -----------..>| | |
621 | * [1] | |
622 | * start_delta size | |
623 | * |----------------|-------------|------------------|------------| | |
624 | * ^ ^ ^ ^ | |
625 | * | | | | | |
626 | * start bmap_start (start) end | |
627 | * of memslot of memslot | |
628 | * | |
629 | * [1] bmap_npages can be aligned to either 64 pages or the end of slot | |
630 | */ | |
631 | ||
632 | assert(bmap_start % BITS_PER_LONG == 0); | |
633 | /* We should never do log_clear before log_sync */ | |
634 | assert(mem->dirty_bmap); | |
635 | if (start_delta) { | |
636 | /* Slow path - we need to manipulate a temp bitmap */ | |
637 | bmap_clear = bitmap_new(bmap_npages); | |
638 | bitmap_copy_with_src_offset(bmap_clear, mem->dirty_bmap, | |
639 | bmap_start, start_delta + size / psize); | |
640 | /* | |
641 | * We need to fill the holes at start because that was not | |
642 | * specified by the caller and we extended the bitmap only for | |
643 | * 64 pages alignment | |
644 | */ | |
645 | bitmap_clear(bmap_clear, 0, start_delta); | |
646 | d.dirty_bitmap = bmap_clear; | |
647 | } else { | |
648 | /* Fast path - start address aligns well with BITS_PER_LONG */ | |
649 | d.dirty_bitmap = mem->dirty_bmap + BIT_WORD(bmap_start); | |
650 | } | |
651 | ||
652 | d.first_page = bmap_start; | |
653 | /* It should never overflow. If it happens, say something */ | |
654 | assert(bmap_npages <= UINT32_MAX); | |
655 | d.num_pages = bmap_npages; | |
4222147d | 656 | d.slot = mem->slot | (as_id << 16); |
ff4aa114 PX |
657 | |
658 | if (kvm_vm_ioctl(s, KVM_CLEAR_DIRTY_LOG, &d) == -1) { | |
659 | ret = -errno; | |
660 | error_report("%s: KVM_CLEAR_DIRTY_LOG failed, slot=%d, " | |
661 | "start=0x%"PRIx64", size=0x%"PRIx32", errno=%d", | |
662 | __func__, d.slot, (uint64_t)d.first_page, | |
663 | (uint32_t)d.num_pages, ret); | |
664 | } else { | |
665 | ret = 0; | |
666 | trace_kvm_clear_dirty_log(d.slot, d.first_page, d.num_pages); | |
667 | } | |
668 | ||
669 | /* | |
670 | * After we have updated the remote dirty bitmap, we update the | |
671 | * cached bitmap as well for the memslot, then if another user | |
672 | * clears the same region we know we shouldn't clear it again on | |
673 | * the remote otherwise it's data loss as well. | |
674 | */ | |
675 | bitmap_clear(mem->dirty_bmap, bmap_start + start_delta, | |
676 | size / psize); | |
677 | /* This handles the NULL case well */ | |
678 | g_free(bmap_clear); | |
4222147d PB |
679 | return ret; |
680 | } | |
681 | ||
682 | ||
683 | /** | |
684 | * kvm_physical_log_clear - Clear the kernel's dirty bitmap for range | |
685 | * | |
686 | * NOTE: this will be a no-op if we haven't enabled manual dirty log | |
687 | * protection in the host kernel because in that case this operation | |
688 | * will be done within log_sync(). | |
689 | * | |
690 | * @kml: the kvm memory listener | |
691 | * @section: the memory range to clear dirty bitmap | |
692 | */ | |
693 | static int kvm_physical_log_clear(KVMMemoryListener *kml, | |
694 | MemoryRegionSection *section) | |
695 | { | |
696 | KVMState *s = kvm_state; | |
697 | uint64_t start, size; | |
698 | KVMSlot *mem = NULL; | |
699 | int ret, i; | |
700 | ||
701 | if (!s->manual_dirty_log_protect) { | |
702 | /* No need to do explicit clear */ | |
703 | return 0; | |
704 | } | |
705 | ||
706 | start = section->offset_within_address_space; | |
707 | size = int128_get64(section->size); | |
708 | ||
709 | if (!size) { | |
710 | /* Nothing more we can do... */ | |
711 | return 0; | |
712 | } | |
713 | ||
714 | kvm_slots_lock(kml); | |
715 | ||
716 | /* Find any possible slot that covers the section */ | |
717 | for (i = 0; i < s->nr_slots; i++) { | |
718 | mem = &kml->slots[i]; | |
719 | if (mem->start_addr <= start && | |
720 | start + size <= mem->start_addr + mem->memory_size) { | |
721 | break; | |
722 | } | |
723 | } | |
724 | ||
725 | /* | |
726 | * We should always find one memslot until this point, otherwise | |
727 | * there could be something wrong from the upper layer | |
728 | */ | |
729 | assert(mem && i != s->nr_slots); | |
730 | ret = kvm_log_clear_one_slot(mem, kml->as_id, start, size); | |
ff4aa114 PX |
731 | |
732 | kvm_slots_unlock(kml); | |
733 | ||
734 | return ret; | |
735 | } | |
736 | ||
95d2994a AK |
737 | static void kvm_coalesce_mmio_region(MemoryListener *listener, |
738 | MemoryRegionSection *secion, | |
a8170e5e | 739 | hwaddr start, hwaddr size) |
f65ed4c1 | 740 | { |
f65ed4c1 AL |
741 | KVMState *s = kvm_state; |
742 | ||
743 | if (s->coalesced_mmio) { | |
744 | struct kvm_coalesced_mmio_zone zone; | |
745 | ||
746 | zone.addr = start; | |
747 | zone.size = size; | |
7e680753 | 748 | zone.pad = 0; |
f65ed4c1 | 749 | |
95d2994a | 750 | (void)kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone); |
f65ed4c1 | 751 | } |
f65ed4c1 AL |
752 | } |
753 | ||
95d2994a AK |
754 | static void kvm_uncoalesce_mmio_region(MemoryListener *listener, |
755 | MemoryRegionSection *secion, | |
a8170e5e | 756 | hwaddr start, hwaddr size) |
f65ed4c1 | 757 | { |
f65ed4c1 AL |
758 | KVMState *s = kvm_state; |
759 | ||
760 | if (s->coalesced_mmio) { | |
761 | struct kvm_coalesced_mmio_zone zone; | |
762 | ||
763 | zone.addr = start; | |
764 | zone.size = size; | |
7e680753 | 765 | zone.pad = 0; |
f65ed4c1 | 766 | |
95d2994a | 767 | (void)kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone); |
f65ed4c1 | 768 | } |
f65ed4c1 AL |
769 | } |
770 | ||
e6d34aee PH |
771 | static void kvm_coalesce_pio_add(MemoryListener *listener, |
772 | MemoryRegionSection *section, | |
773 | hwaddr start, hwaddr size) | |
774 | { | |
775 | KVMState *s = kvm_state; | |
776 | ||
777 | if (s->coalesced_pio) { | |
778 | struct kvm_coalesced_mmio_zone zone; | |
779 | ||
780 | zone.addr = start; | |
781 | zone.size = size; | |
782 | zone.pio = 1; | |
783 | ||
784 | (void)kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone); | |
785 | } | |
786 | } | |
787 | ||
788 | static void kvm_coalesce_pio_del(MemoryListener *listener, | |
789 | MemoryRegionSection *section, | |
790 | hwaddr start, hwaddr size) | |
791 | { | |
792 | KVMState *s = kvm_state; | |
793 | ||
794 | if (s->coalesced_pio) { | |
795 | struct kvm_coalesced_mmio_zone zone; | |
796 | ||
797 | zone.addr = start; | |
798 | zone.size = size; | |
799 | zone.pio = 1; | |
800 | ||
801 | (void)kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone); | |
802 | } | |
803 | } | |
804 | ||
805 | static MemoryListener kvm_coalesced_pio_listener = { | |
806 | .coalesced_io_add = kvm_coalesce_pio_add, | |
807 | .coalesced_io_del = kvm_coalesce_pio_del, | |
808 | }; | |
809 | ||
ad7b8b33 AL |
810 | int kvm_check_extension(KVMState *s, unsigned int extension) |
811 | { | |
812 | int ret; | |
813 | ||
814 | ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension); | |
815 | if (ret < 0) { | |
816 | ret = 0; | |
817 | } | |
818 | ||
819 | return ret; | |
820 | } | |
821 | ||
7d0a07fa AG |
822 | int kvm_vm_check_extension(KVMState *s, unsigned int extension) |
823 | { | |
824 | int ret; | |
825 | ||
826 | ret = kvm_vm_ioctl(s, KVM_CHECK_EXTENSION, extension); | |
827 | if (ret < 0) { | |
828 | /* VM wide version not implemented, use global one instead */ | |
829 | ret = kvm_check_extension(s, extension); | |
830 | } | |
831 | ||
832 | return ret; | |
833 | } | |
834 | ||
b680c5ba GK |
835 | static uint32_t adjust_ioeventfd_endianness(uint32_t val, uint32_t size) |
836 | { | |
837 | #if defined(HOST_WORDS_BIGENDIAN) != defined(TARGET_WORDS_BIGENDIAN) | |
838 | /* The kernel expects ioeventfd values in HOST_WORDS_BIGENDIAN | |
839 | * endianness, but the memory core hands them in target endianness. | |
840 | * For example, PPC is always treated as big-endian even if running | |
841 | * on KVM and on PPC64LE. Correct here. | |
842 | */ | |
843 | switch (size) { | |
844 | case 2: | |
845 | val = bswap16(val); | |
846 | break; | |
847 | case 4: | |
848 | val = bswap32(val); | |
849 | break; | |
850 | } | |
851 | #endif | |
852 | return val; | |
853 | } | |
854 | ||
584f2be7 | 855 | static int kvm_set_ioeventfd_mmio(int fd, hwaddr addr, uint32_t val, |
41cb62c2 | 856 | bool assign, uint32_t size, bool datamatch) |
500ffd4a MT |
857 | { |
858 | int ret; | |
03a96b83 TH |
859 | struct kvm_ioeventfd iofd = { |
860 | .datamatch = datamatch ? adjust_ioeventfd_endianness(val, size) : 0, | |
861 | .addr = addr, | |
862 | .len = size, | |
863 | .flags = 0, | |
864 | .fd = fd, | |
865 | }; | |
500ffd4a | 866 | |
876d16cd DDAG |
867 | trace_kvm_set_ioeventfd_mmio(fd, (uint64_t)addr, val, assign, size, |
868 | datamatch); | |
500ffd4a MT |
869 | if (!kvm_enabled()) { |
870 | return -ENOSYS; | |
871 | } | |
872 | ||
41cb62c2 MT |
873 | if (datamatch) { |
874 | iofd.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH; | |
875 | } | |
500ffd4a MT |
876 | if (!assign) { |
877 | iofd.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; | |
878 | } | |
879 | ||
880 | ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd); | |
881 | ||
882 | if (ret < 0) { | |
883 | return -errno; | |
884 | } | |
885 | ||
886 | return 0; | |
887 | } | |
888 | ||
44c3f8f7 | 889 | static int kvm_set_ioeventfd_pio(int fd, uint16_t addr, uint16_t val, |
41cb62c2 | 890 | bool assign, uint32_t size, bool datamatch) |
500ffd4a MT |
891 | { |
892 | struct kvm_ioeventfd kick = { | |
b680c5ba | 893 | .datamatch = datamatch ? adjust_ioeventfd_endianness(val, size) : 0, |
500ffd4a | 894 | .addr = addr, |
41cb62c2 | 895 | .flags = KVM_IOEVENTFD_FLAG_PIO, |
44c3f8f7 | 896 | .len = size, |
500ffd4a MT |
897 | .fd = fd, |
898 | }; | |
899 | int r; | |
876d16cd | 900 | trace_kvm_set_ioeventfd_pio(fd, addr, val, assign, size, datamatch); |
500ffd4a MT |
901 | if (!kvm_enabled()) { |
902 | return -ENOSYS; | |
903 | } | |
41cb62c2 MT |
904 | if (datamatch) { |
905 | kick.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH; | |
906 | } | |
500ffd4a MT |
907 | if (!assign) { |
908 | kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; | |
909 | } | |
910 | r = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick); | |
911 | if (r < 0) { | |
912 | return r; | |
913 | } | |
914 | return 0; | |
915 | } | |
916 | ||
917 | ||
d2f2b8a7 SH |
918 | static int kvm_check_many_ioeventfds(void) |
919 | { | |
d0dcac83 SH |
920 | /* Userspace can use ioeventfd for io notification. This requires a host |
921 | * that supports eventfd(2) and an I/O thread; since eventfd does not | |
922 | * support SIGIO it cannot interrupt the vcpu. | |
923 | * | |
924 | * Older kernels have a 6 device limit on the KVM io bus. Find out so we | |
d2f2b8a7 SH |
925 | * can avoid creating too many ioeventfds. |
926 | */ | |
12d4536f | 927 | #if defined(CONFIG_EVENTFD) |
d2f2b8a7 SH |
928 | int ioeventfds[7]; |
929 | int i, ret = 0; | |
930 | for (i = 0; i < ARRAY_SIZE(ioeventfds); i++) { | |
931 | ioeventfds[i] = eventfd(0, EFD_CLOEXEC); | |
932 | if (ioeventfds[i] < 0) { | |
933 | break; | |
934 | } | |
41cb62c2 | 935 | ret = kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, true, 2, true); |
d2f2b8a7 SH |
936 | if (ret < 0) { |
937 | close(ioeventfds[i]); | |
938 | break; | |
939 | } | |
940 | } | |
941 | ||
942 | /* Decide whether many devices are supported or not */ | |
943 | ret = i == ARRAY_SIZE(ioeventfds); | |
944 | ||
945 | while (i-- > 0) { | |
41cb62c2 | 946 | kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, false, 2, true); |
d2f2b8a7 SH |
947 | close(ioeventfds[i]); |
948 | } | |
949 | return ret; | |
950 | #else | |
951 | return 0; | |
952 | #endif | |
953 | } | |
954 | ||
94a8d39a JK |
955 | static const KVMCapabilityInfo * |
956 | kvm_check_extension_list(KVMState *s, const KVMCapabilityInfo *list) | |
957 | { | |
958 | while (list->name) { | |
959 | if (!kvm_check_extension(s, list->value)) { | |
960 | return list; | |
961 | } | |
962 | list++; | |
963 | } | |
964 | return NULL; | |
965 | } | |
966 | ||
7bbda04c PB |
967 | static void kvm_set_phys_mem(KVMMemoryListener *kml, |
968 | MemoryRegionSection *section, bool add) | |
46dbef6a | 969 | { |
f357f564 | 970 | KVMSlot *mem; |
46dbef6a | 971 | int err; |
a01672d3 | 972 | MemoryRegion *mr = section->mr; |
235e8982 | 973 | bool writeable = !mr->readonly && !mr->rom_device; |
5ea69c2e DH |
974 | hwaddr start_addr, size; |
975 | void *ram; | |
46dbef6a | 976 | |
a01672d3 | 977 | if (!memory_region_is_ram(mr)) { |
235e8982 JJ |
978 | if (writeable || !kvm_readonly_mem_allowed) { |
979 | return; | |
980 | } else if (!mr->romd_mode) { | |
981 | /* If the memory device is not in romd_mode, then we actually want | |
982 | * to remove the kvm memory slot so all accesses will trap. */ | |
983 | add = false; | |
984 | } | |
9f213ed9 AK |
985 | } |
986 | ||
5ea69c2e DH |
987 | size = kvm_align_section(section, &start_addr); |
988 | if (!size) { | |
989 | return; | |
990 | } | |
991 | ||
bbfd3017 | 992 | /* use aligned delta to align the ram address */ |
5ea69c2e | 993 | ram = memory_region_get_ram_ptr(mr) + section->offset_within_region + |
bbfd3017 | 994 | (start_addr - section->offset_within_address_space); |
a01672d3 | 995 | |
36adac49 PX |
996 | kvm_slots_lock(kml); |
997 | ||
f357f564 | 998 | if (!add) { |
90ed4bcc | 999 | mem = kvm_lookup_matching_slot(kml, start_addr, size); |
46dbef6a | 1000 | if (!mem) { |
36adac49 | 1001 | goto out; |
46dbef6a | 1002 | } |
1bfbac4e | 1003 | if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) { |
7bbda04c | 1004 | kvm_physical_sync_dirty_bitmap(kml, section); |
3fbffb62 AK |
1005 | } |
1006 | ||
f357f564 | 1007 | /* unregister the slot */ |
9f4bf4ba PX |
1008 | g_free(mem->dirty_bmap); |
1009 | mem->dirty_bmap = NULL; | |
46dbef6a | 1010 | mem->memory_size = 0; |
6c090d4a SZ |
1011 | mem->flags = 0; |
1012 | err = kvm_set_user_memory_region(kml, mem, false); | |
46dbef6a | 1013 | if (err) { |
1c4fdaba | 1014 | fprintf(stderr, "%s: error unregistering slot: %s\n", |
46dbef6a MT |
1015 | __func__, strerror(-err)); |
1016 | abort(); | |
1017 | } | |
36adac49 | 1018 | goto out; |
46dbef6a MT |
1019 | } |
1020 | ||
f357f564 | 1021 | /* register the new slot */ |
7bbda04c | 1022 | mem = kvm_alloc_slot(kml); |
46dbef6a MT |
1023 | mem->memory_size = size; |
1024 | mem->start_addr = start_addr; | |
9f213ed9 | 1025 | mem->ram = ram; |
d6ff5cbc | 1026 | mem->flags = kvm_mem_flags(mr); |
46dbef6a | 1027 | |
6c090d4a | 1028 | err = kvm_set_user_memory_region(kml, mem, true); |
46dbef6a MT |
1029 | if (err) { |
1030 | fprintf(stderr, "%s: error registering slot: %s\n", __func__, | |
1031 | strerror(-err)); | |
1032 | abort(); | |
1033 | } | |
36adac49 PX |
1034 | |
1035 | out: | |
1036 | kvm_slots_unlock(kml); | |
46dbef6a MT |
1037 | } |
1038 | ||
a01672d3 AK |
1039 | static void kvm_region_add(MemoryListener *listener, |
1040 | MemoryRegionSection *section) | |
1041 | { | |
7bbda04c PB |
1042 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); |
1043 | ||
dfde4e6e | 1044 | memory_region_ref(section->mr); |
7bbda04c | 1045 | kvm_set_phys_mem(kml, section, true); |
a01672d3 AK |
1046 | } |
1047 | ||
1048 | static void kvm_region_del(MemoryListener *listener, | |
1049 | MemoryRegionSection *section) | |
1050 | { | |
7bbda04c PB |
1051 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); |
1052 | ||
1053 | kvm_set_phys_mem(kml, section, false); | |
dfde4e6e | 1054 | memory_region_unref(section->mr); |
a01672d3 AK |
1055 | } |
1056 | ||
1057 | static void kvm_log_sync(MemoryListener *listener, | |
1058 | MemoryRegionSection *section) | |
7b8f3b78 | 1059 | { |
7bbda04c | 1060 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); |
a01672d3 AK |
1061 | int r; |
1062 | ||
36adac49 | 1063 | kvm_slots_lock(kml); |
7bbda04c | 1064 | r = kvm_physical_sync_dirty_bitmap(kml, section); |
36adac49 | 1065 | kvm_slots_unlock(kml); |
a01672d3 AK |
1066 | if (r < 0) { |
1067 | abort(); | |
1068 | } | |
7b8f3b78 MT |
1069 | } |
1070 | ||
ff4aa114 PX |
1071 | static void kvm_log_clear(MemoryListener *listener, |
1072 | MemoryRegionSection *section) | |
1073 | { | |
1074 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); | |
1075 | int r; | |
1076 | ||
1077 | r = kvm_physical_log_clear(kml, section); | |
1078 | if (r < 0) { | |
1079 | error_report_once("%s: kvm log clear failed: mr=%s " | |
1080 | "offset=%"HWADDR_PRIx" size=%"PRIx64, __func__, | |
1081 | section->mr->name, section->offset_within_region, | |
1082 | int128_get64(section->size)); | |
1083 | abort(); | |
1084 | } | |
1085 | } | |
1086 | ||
d22b096e AK |
1087 | static void kvm_mem_ioeventfd_add(MemoryListener *listener, |
1088 | MemoryRegionSection *section, | |
1089 | bool match_data, uint64_t data, | |
1090 | EventNotifier *e) | |
1091 | { | |
1092 | int fd = event_notifier_get_fd(e); | |
80a1ea37 AK |
1093 | int r; |
1094 | ||
4b8f1c88 | 1095 | r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space, |
052e87b0 PB |
1096 | data, true, int128_get64(section->size), |
1097 | match_data); | |
80a1ea37 | 1098 | if (r < 0) { |
e346bcbf YK |
1099 | fprintf(stderr, "%s: error adding ioeventfd: %s (%d)\n", |
1100 | __func__, strerror(-r), -r); | |
80a1ea37 AK |
1101 | abort(); |
1102 | } | |
1103 | } | |
1104 | ||
d22b096e AK |
1105 | static void kvm_mem_ioeventfd_del(MemoryListener *listener, |
1106 | MemoryRegionSection *section, | |
1107 | bool match_data, uint64_t data, | |
1108 | EventNotifier *e) | |
80a1ea37 | 1109 | { |
d22b096e | 1110 | int fd = event_notifier_get_fd(e); |
80a1ea37 AK |
1111 | int r; |
1112 | ||
4b8f1c88 | 1113 | r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space, |
052e87b0 PB |
1114 | data, false, int128_get64(section->size), |
1115 | match_data); | |
80a1ea37 | 1116 | if (r < 0) { |
e346bcbf YK |
1117 | fprintf(stderr, "%s: error deleting ioeventfd: %s (%d)\n", |
1118 | __func__, strerror(-r), -r); | |
80a1ea37 AK |
1119 | abort(); |
1120 | } | |
1121 | } | |
1122 | ||
d22b096e AK |
1123 | static void kvm_io_ioeventfd_add(MemoryListener *listener, |
1124 | MemoryRegionSection *section, | |
1125 | bool match_data, uint64_t data, | |
1126 | EventNotifier *e) | |
80a1ea37 | 1127 | { |
d22b096e | 1128 | int fd = event_notifier_get_fd(e); |
80a1ea37 AK |
1129 | int r; |
1130 | ||
44c3f8f7 | 1131 | r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space, |
052e87b0 PB |
1132 | data, true, int128_get64(section->size), |
1133 | match_data); | |
80a1ea37 | 1134 | if (r < 0) { |
e346bcbf YK |
1135 | fprintf(stderr, "%s: error adding ioeventfd: %s (%d)\n", |
1136 | __func__, strerror(-r), -r); | |
80a1ea37 AK |
1137 | abort(); |
1138 | } | |
1139 | } | |
1140 | ||
d22b096e AK |
1141 | static void kvm_io_ioeventfd_del(MemoryListener *listener, |
1142 | MemoryRegionSection *section, | |
1143 | bool match_data, uint64_t data, | |
1144 | EventNotifier *e) | |
80a1ea37 AK |
1145 | |
1146 | { | |
d22b096e | 1147 | int fd = event_notifier_get_fd(e); |
80a1ea37 AK |
1148 | int r; |
1149 | ||
44c3f8f7 | 1150 | r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space, |
052e87b0 PB |
1151 | data, false, int128_get64(section->size), |
1152 | match_data); | |
80a1ea37 | 1153 | if (r < 0) { |
e346bcbf YK |
1154 | fprintf(stderr, "%s: error deleting ioeventfd: %s (%d)\n", |
1155 | __func__, strerror(-r), -r); | |
80a1ea37 AK |
1156 | abort(); |
1157 | } | |
1158 | } | |
1159 | ||
38bfe691 PB |
1160 | void kvm_memory_listener_register(KVMState *s, KVMMemoryListener *kml, |
1161 | AddressSpace *as, int as_id) | |
7bbda04c PB |
1162 | { |
1163 | int i; | |
1164 | ||
36adac49 | 1165 | qemu_mutex_init(&kml->slots_lock); |
7bbda04c | 1166 | kml->slots = g_malloc0(s->nr_slots * sizeof(KVMSlot)); |
38bfe691 | 1167 | kml->as_id = as_id; |
7bbda04c PB |
1168 | |
1169 | for (i = 0; i < s->nr_slots; i++) { | |
1170 | kml->slots[i].slot = i; | |
1171 | } | |
1172 | ||
1173 | kml->listener.region_add = kvm_region_add; | |
1174 | kml->listener.region_del = kvm_region_del; | |
1175 | kml->listener.log_start = kvm_log_start; | |
1176 | kml->listener.log_stop = kvm_log_stop; | |
1177 | kml->listener.log_sync = kvm_log_sync; | |
ff4aa114 | 1178 | kml->listener.log_clear = kvm_log_clear; |
7bbda04c PB |
1179 | kml->listener.priority = 10; |
1180 | ||
1181 | memory_listener_register(&kml->listener, as); | |
8072aae3 AK |
1182 | |
1183 | for (i = 0; i < s->nr_as; ++i) { | |
1184 | if (!s->as[i].as) { | |
1185 | s->as[i].as = as; | |
1186 | s->as[i].ml = kml; | |
1187 | break; | |
1188 | } | |
1189 | } | |
7bbda04c | 1190 | } |
d22b096e AK |
1191 | |
1192 | static MemoryListener kvm_io_listener = { | |
d22b096e AK |
1193 | .eventfd_add = kvm_io_ioeventfd_add, |
1194 | .eventfd_del = kvm_io_ioeventfd_del, | |
72e22d2f | 1195 | .priority = 10, |
7b8f3b78 MT |
1196 | }; |
1197 | ||
3889c3fa | 1198 | int kvm_set_irq(KVMState *s, int irq, int level) |
84b058d7 JK |
1199 | { |
1200 | struct kvm_irq_level event; | |
1201 | int ret; | |
1202 | ||
7ae26bd4 | 1203 | assert(kvm_async_interrupts_enabled()); |
84b058d7 JK |
1204 | |
1205 | event.level = level; | |
1206 | event.irq = irq; | |
e333cd69 | 1207 | ret = kvm_vm_ioctl(s, s->irq_set_ioctl, &event); |
84b058d7 | 1208 | if (ret < 0) { |
3889c3fa | 1209 | perror("kvm_set_irq"); |
84b058d7 JK |
1210 | abort(); |
1211 | } | |
1212 | ||
e333cd69 | 1213 | return (s->irq_set_ioctl == KVM_IRQ_LINE) ? 1 : event.status; |
84b058d7 JK |
1214 | } |
1215 | ||
1216 | #ifdef KVM_CAP_IRQ_ROUTING | |
d3d3bef0 JK |
1217 | typedef struct KVMMSIRoute { |
1218 | struct kvm_irq_routing_entry kroute; | |
1219 | QTAILQ_ENTRY(KVMMSIRoute) entry; | |
1220 | } KVMMSIRoute; | |
1221 | ||
84b058d7 JK |
1222 | static void set_gsi(KVMState *s, unsigned int gsi) |
1223 | { | |
8269fb70 | 1224 | set_bit(gsi, s->used_gsi_bitmap); |
84b058d7 JK |
1225 | } |
1226 | ||
04fa27f5 JK |
1227 | static void clear_gsi(KVMState *s, unsigned int gsi) |
1228 | { | |
8269fb70 | 1229 | clear_bit(gsi, s->used_gsi_bitmap); |
04fa27f5 JK |
1230 | } |
1231 | ||
7b774593 | 1232 | void kvm_init_irq_routing(KVMState *s) |
84b058d7 | 1233 | { |
04fa27f5 | 1234 | int gsi_count, i; |
84b058d7 | 1235 | |
00008418 | 1236 | gsi_count = kvm_check_extension(s, KVM_CAP_IRQ_ROUTING) - 1; |
84b058d7 | 1237 | if (gsi_count > 0) { |
84b058d7 | 1238 | /* Round up so we can search ints using ffs */ |
8269fb70 | 1239 | s->used_gsi_bitmap = bitmap_new(gsi_count); |
4e2e4e63 | 1240 | s->gsi_count = gsi_count; |
84b058d7 JK |
1241 | } |
1242 | ||
1243 | s->irq_routes = g_malloc0(sizeof(*s->irq_routes)); | |
1244 | s->nr_allocated_irq_routes = 0; | |
1245 | ||
50bf31b9 | 1246 | if (!kvm_direct_msi_allowed) { |
4a3adebb JK |
1247 | for (i = 0; i < KVM_MSI_HASHTAB_SIZE; i++) { |
1248 | QTAILQ_INIT(&s->msi_hashtab[i]); | |
1249 | } | |
04fa27f5 JK |
1250 | } |
1251 | ||
84b058d7 JK |
1252 | kvm_arch_init_irq_routing(s); |
1253 | } | |
1254 | ||
cb925cf9 | 1255 | void kvm_irqchip_commit_routes(KVMState *s) |
e7b20308 JK |
1256 | { |
1257 | int ret; | |
1258 | ||
7005f7f8 PX |
1259 | if (kvm_gsi_direct_mapping()) { |
1260 | return; | |
1261 | } | |
1262 | ||
1263 | if (!kvm_gsi_routing_enabled()) { | |
1264 | return; | |
1265 | } | |
1266 | ||
e7b20308 | 1267 | s->irq_routes->flags = 0; |
54a6c11b | 1268 | trace_kvm_irqchip_commit_routes(); |
e7b20308 JK |
1269 | ret = kvm_vm_ioctl(s, KVM_SET_GSI_ROUTING, s->irq_routes); |
1270 | assert(ret == 0); | |
1271 | } | |
1272 | ||
84b058d7 JK |
1273 | static void kvm_add_routing_entry(KVMState *s, |
1274 | struct kvm_irq_routing_entry *entry) | |
1275 | { | |
1276 | struct kvm_irq_routing_entry *new; | |
1277 | int n, size; | |
1278 | ||
1279 | if (s->irq_routes->nr == s->nr_allocated_irq_routes) { | |
1280 | n = s->nr_allocated_irq_routes * 2; | |
1281 | if (n < 64) { | |
1282 | n = 64; | |
1283 | } | |
1284 | size = sizeof(struct kvm_irq_routing); | |
1285 | size += n * sizeof(*new); | |
1286 | s->irq_routes = g_realloc(s->irq_routes, size); | |
1287 | s->nr_allocated_irq_routes = n; | |
1288 | } | |
1289 | n = s->irq_routes->nr++; | |
1290 | new = &s->irq_routes->entries[n]; | |
0fbc2074 MT |
1291 | |
1292 | *new = *entry; | |
84b058d7 JK |
1293 | |
1294 | set_gsi(s, entry->gsi); | |
1295 | } | |
1296 | ||
cc57407e JK |
1297 | static int kvm_update_routing_entry(KVMState *s, |
1298 | struct kvm_irq_routing_entry *new_entry) | |
1299 | { | |
1300 | struct kvm_irq_routing_entry *entry; | |
1301 | int n; | |
1302 | ||
1303 | for (n = 0; n < s->irq_routes->nr; n++) { | |
1304 | entry = &s->irq_routes->entries[n]; | |
1305 | if (entry->gsi != new_entry->gsi) { | |
1306 | continue; | |
1307 | } | |
1308 | ||
40509f7f MT |
1309 | if(!memcmp(entry, new_entry, sizeof *entry)) { |
1310 | return 0; | |
1311 | } | |
1312 | ||
0fbc2074 | 1313 | *entry = *new_entry; |
cc57407e | 1314 | |
cc57407e JK |
1315 | return 0; |
1316 | } | |
1317 | ||
1318 | return -ESRCH; | |
1319 | } | |
1320 | ||
1df186df | 1321 | void kvm_irqchip_add_irq_route(KVMState *s, int irq, int irqchip, int pin) |
84b058d7 | 1322 | { |
0fbc2074 | 1323 | struct kvm_irq_routing_entry e = {}; |
84b058d7 | 1324 | |
4e2e4e63 JK |
1325 | assert(pin < s->gsi_count); |
1326 | ||
84b058d7 JK |
1327 | e.gsi = irq; |
1328 | e.type = KVM_IRQ_ROUTING_IRQCHIP; | |
1329 | e.flags = 0; | |
1330 | e.u.irqchip.irqchip = irqchip; | |
1331 | e.u.irqchip.pin = pin; | |
1332 | kvm_add_routing_entry(s, &e); | |
1333 | } | |
1334 | ||
1e2aa8be | 1335 | void kvm_irqchip_release_virq(KVMState *s, int virq) |
04fa27f5 JK |
1336 | { |
1337 | struct kvm_irq_routing_entry *e; | |
1338 | int i; | |
1339 | ||
76fe21de AK |
1340 | if (kvm_gsi_direct_mapping()) { |
1341 | return; | |
1342 | } | |
1343 | ||
04fa27f5 JK |
1344 | for (i = 0; i < s->irq_routes->nr; i++) { |
1345 | e = &s->irq_routes->entries[i]; | |
1346 | if (e->gsi == virq) { | |
1347 | s->irq_routes->nr--; | |
1348 | *e = s->irq_routes->entries[s->irq_routes->nr]; | |
1349 | } | |
1350 | } | |
1351 | clear_gsi(s, virq); | |
38d87493 | 1352 | kvm_arch_release_virq_post(virq); |
9ba35d0b | 1353 | trace_kvm_irqchip_release_virq(virq); |
04fa27f5 JK |
1354 | } |
1355 | ||
1356 | static unsigned int kvm_hash_msi(uint32_t data) | |
1357 | { | |
1358 | /* This is optimized for IA32 MSI layout. However, no other arch shall | |
1359 | * repeat the mistake of not providing a direct MSI injection API. */ | |
1360 | return data & 0xff; | |
1361 | } | |
1362 | ||
1363 | static void kvm_flush_dynamic_msi_routes(KVMState *s) | |
1364 | { | |
1365 | KVMMSIRoute *route, *next; | |
1366 | unsigned int hash; | |
1367 | ||
1368 | for (hash = 0; hash < KVM_MSI_HASHTAB_SIZE; hash++) { | |
1369 | QTAILQ_FOREACH_SAFE(route, &s->msi_hashtab[hash], entry, next) { | |
1370 | kvm_irqchip_release_virq(s, route->kroute.gsi); | |
1371 | QTAILQ_REMOVE(&s->msi_hashtab[hash], route, entry); | |
1372 | g_free(route); | |
1373 | } | |
1374 | } | |
1375 | } | |
1376 | ||
1377 | static int kvm_irqchip_get_virq(KVMState *s) | |
1378 | { | |
8269fb70 | 1379 | int next_virq; |
04fa27f5 | 1380 | |
bdf02631 WM |
1381 | /* |
1382 | * PIC and IOAPIC share the first 16 GSI numbers, thus the available | |
1383 | * GSI numbers are more than the number of IRQ route. Allocating a GSI | |
1384 | * number can succeed even though a new route entry cannot be added. | |
1385 | * When this happens, flush dynamic MSI entries to free IRQ route entries. | |
1386 | */ | |
50bf31b9 | 1387 | if (!kvm_direct_msi_allowed && s->irq_routes->nr == s->gsi_count) { |
bdf02631 WM |
1388 | kvm_flush_dynamic_msi_routes(s); |
1389 | } | |
1390 | ||
04fa27f5 | 1391 | /* Return the lowest unused GSI in the bitmap */ |
8269fb70 WY |
1392 | next_virq = find_first_zero_bit(s->used_gsi_bitmap, s->gsi_count); |
1393 | if (next_virq >= s->gsi_count) { | |
1394 | return -ENOSPC; | |
1395 | } else { | |
1396 | return next_virq; | |
04fa27f5 | 1397 | } |
04fa27f5 JK |
1398 | } |
1399 | ||
1400 | static KVMMSIRoute *kvm_lookup_msi_route(KVMState *s, MSIMessage msg) | |
1401 | { | |
1402 | unsigned int hash = kvm_hash_msi(msg.data); | |
1403 | KVMMSIRoute *route; | |
1404 | ||
1405 | QTAILQ_FOREACH(route, &s->msi_hashtab[hash], entry) { | |
1406 | if (route->kroute.u.msi.address_lo == (uint32_t)msg.address && | |
1407 | route->kroute.u.msi.address_hi == (msg.address >> 32) && | |
d07cc1f1 | 1408 | route->kroute.u.msi.data == le32_to_cpu(msg.data)) { |
04fa27f5 JK |
1409 | return route; |
1410 | } | |
1411 | } | |
1412 | return NULL; | |
1413 | } | |
1414 | ||
1415 | int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg) | |
1416 | { | |
4a3adebb | 1417 | struct kvm_msi msi; |
04fa27f5 JK |
1418 | KVMMSIRoute *route; |
1419 | ||
50bf31b9 | 1420 | if (kvm_direct_msi_allowed) { |
4a3adebb JK |
1421 | msi.address_lo = (uint32_t)msg.address; |
1422 | msi.address_hi = msg.address >> 32; | |
d07cc1f1 | 1423 | msi.data = le32_to_cpu(msg.data); |
4a3adebb JK |
1424 | msi.flags = 0; |
1425 | memset(msi.pad, 0, sizeof(msi.pad)); | |
1426 | ||
1427 | return kvm_vm_ioctl(s, KVM_SIGNAL_MSI, &msi); | |
1428 | } | |
1429 | ||
04fa27f5 JK |
1430 | route = kvm_lookup_msi_route(s, msg); |
1431 | if (!route) { | |
e7b20308 | 1432 | int virq; |
04fa27f5 JK |
1433 | |
1434 | virq = kvm_irqchip_get_virq(s); | |
1435 | if (virq < 0) { | |
1436 | return virq; | |
1437 | } | |
1438 | ||
0fbc2074 | 1439 | route = g_malloc0(sizeof(KVMMSIRoute)); |
04fa27f5 JK |
1440 | route->kroute.gsi = virq; |
1441 | route->kroute.type = KVM_IRQ_ROUTING_MSI; | |
1442 | route->kroute.flags = 0; | |
1443 | route->kroute.u.msi.address_lo = (uint32_t)msg.address; | |
1444 | route->kroute.u.msi.address_hi = msg.address >> 32; | |
d07cc1f1 | 1445 | route->kroute.u.msi.data = le32_to_cpu(msg.data); |
04fa27f5 JK |
1446 | |
1447 | kvm_add_routing_entry(s, &route->kroute); | |
cb925cf9 | 1448 | kvm_irqchip_commit_routes(s); |
04fa27f5 JK |
1449 | |
1450 | QTAILQ_INSERT_TAIL(&s->msi_hashtab[kvm_hash_msi(msg.data)], route, | |
1451 | entry); | |
04fa27f5 JK |
1452 | } |
1453 | ||
1454 | assert(route->kroute.type == KVM_IRQ_ROUTING_MSI); | |
1455 | ||
3889c3fa | 1456 | return kvm_set_irq(s, route->kroute.gsi, 1); |
04fa27f5 JK |
1457 | } |
1458 | ||
d1f6af6a | 1459 | int kvm_irqchip_add_msi_route(KVMState *s, int vector, PCIDevice *dev) |
92b4e489 | 1460 | { |
0fbc2074 | 1461 | struct kvm_irq_routing_entry kroute = {}; |
92b4e489 | 1462 | int virq; |
d1f6af6a PX |
1463 | MSIMessage msg = {0, 0}; |
1464 | ||
88c725c7 | 1465 | if (pci_available && dev) { |
e1d4fb2d | 1466 | msg = pci_get_msi_message(dev, vector); |
d1f6af6a | 1467 | } |
92b4e489 | 1468 | |
76fe21de | 1469 | if (kvm_gsi_direct_mapping()) { |
1850b6b7 | 1470 | return kvm_arch_msi_data_to_gsi(msg.data); |
76fe21de AK |
1471 | } |
1472 | ||
f3e1bed8 | 1473 | if (!kvm_gsi_routing_enabled()) { |
92b4e489 JK |
1474 | return -ENOSYS; |
1475 | } | |
1476 | ||
1477 | virq = kvm_irqchip_get_virq(s); | |
1478 | if (virq < 0) { | |
1479 | return virq; | |
1480 | } | |
1481 | ||
1482 | kroute.gsi = virq; | |
1483 | kroute.type = KVM_IRQ_ROUTING_MSI; | |
1484 | kroute.flags = 0; | |
1485 | kroute.u.msi.address_lo = (uint32_t)msg.address; | |
1486 | kroute.u.msi.address_hi = msg.address >> 32; | |
d07cc1f1 | 1487 | kroute.u.msi.data = le32_to_cpu(msg.data); |
88c725c7 | 1488 | if (pci_available && kvm_msi_devid_required()) { |
767a554a PF |
1489 | kroute.flags = KVM_MSI_VALID_DEVID; |
1490 | kroute.u.msi.devid = pci_requester_id(dev); | |
1491 | } | |
dc9f06ca | 1492 | if (kvm_arch_fixup_msi_route(&kroute, msg.address, msg.data, dev)) { |
9e03a040 FB |
1493 | kvm_irqchip_release_virq(s, virq); |
1494 | return -EINVAL; | |
1495 | } | |
92b4e489 | 1496 | |
9ba35d0b PX |
1497 | trace_kvm_irqchip_add_msi_route(dev ? dev->name : (char *)"N/A", |
1498 | vector, virq); | |
54a6c11b | 1499 | |
92b4e489 | 1500 | kvm_add_routing_entry(s, &kroute); |
38d87493 | 1501 | kvm_arch_add_msi_route_post(&kroute, vector, dev); |
cb925cf9 | 1502 | kvm_irqchip_commit_routes(s); |
92b4e489 JK |
1503 | |
1504 | return virq; | |
1505 | } | |
1506 | ||
dc9f06ca PF |
1507 | int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg, |
1508 | PCIDevice *dev) | |
cc57407e | 1509 | { |
0fbc2074 | 1510 | struct kvm_irq_routing_entry kroute = {}; |
cc57407e | 1511 | |
76fe21de AK |
1512 | if (kvm_gsi_direct_mapping()) { |
1513 | return 0; | |
1514 | } | |
1515 | ||
cc57407e JK |
1516 | if (!kvm_irqchip_in_kernel()) { |
1517 | return -ENOSYS; | |
1518 | } | |
1519 | ||
1520 | kroute.gsi = virq; | |
1521 | kroute.type = KVM_IRQ_ROUTING_MSI; | |
1522 | kroute.flags = 0; | |
1523 | kroute.u.msi.address_lo = (uint32_t)msg.address; | |
1524 | kroute.u.msi.address_hi = msg.address >> 32; | |
d07cc1f1 | 1525 | kroute.u.msi.data = le32_to_cpu(msg.data); |
88c725c7 | 1526 | if (pci_available && kvm_msi_devid_required()) { |
767a554a PF |
1527 | kroute.flags = KVM_MSI_VALID_DEVID; |
1528 | kroute.u.msi.devid = pci_requester_id(dev); | |
1529 | } | |
dc9f06ca | 1530 | if (kvm_arch_fixup_msi_route(&kroute, msg.address, msg.data, dev)) { |
9e03a040 FB |
1531 | return -EINVAL; |
1532 | } | |
cc57407e | 1533 | |
54a6c11b PX |
1534 | trace_kvm_irqchip_update_msi_route(virq); |
1535 | ||
cc57407e JK |
1536 | return kvm_update_routing_entry(s, &kroute); |
1537 | } | |
1538 | ||
ca916d37 VM |
1539 | static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int rfd, int virq, |
1540 | bool assign) | |
39853bbc JK |
1541 | { |
1542 | struct kvm_irqfd irqfd = { | |
1543 | .fd = fd, | |
1544 | .gsi = virq, | |
1545 | .flags = assign ? 0 : KVM_IRQFD_FLAG_DEASSIGN, | |
1546 | }; | |
1547 | ||
ca916d37 VM |
1548 | if (rfd != -1) { |
1549 | irqfd.flags |= KVM_IRQFD_FLAG_RESAMPLE; | |
1550 | irqfd.resamplefd = rfd; | |
1551 | } | |
1552 | ||
cc7e0ddf | 1553 | if (!kvm_irqfds_enabled()) { |
39853bbc JK |
1554 | return -ENOSYS; |
1555 | } | |
1556 | ||
1557 | return kvm_vm_ioctl(s, KVM_IRQFD, &irqfd); | |
1558 | } | |
1559 | ||
d426d9fb CH |
1560 | int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter) |
1561 | { | |
e9af2fef | 1562 | struct kvm_irq_routing_entry kroute = {}; |
d426d9fb CH |
1563 | int virq; |
1564 | ||
1565 | if (!kvm_gsi_routing_enabled()) { | |
1566 | return -ENOSYS; | |
1567 | } | |
1568 | ||
1569 | virq = kvm_irqchip_get_virq(s); | |
1570 | if (virq < 0) { | |
1571 | return virq; | |
1572 | } | |
1573 | ||
1574 | kroute.gsi = virq; | |
1575 | kroute.type = KVM_IRQ_ROUTING_S390_ADAPTER; | |
1576 | kroute.flags = 0; | |
1577 | kroute.u.adapter.summary_addr = adapter->summary_addr; | |
1578 | kroute.u.adapter.ind_addr = adapter->ind_addr; | |
1579 | kroute.u.adapter.summary_offset = adapter->summary_offset; | |
1580 | kroute.u.adapter.ind_offset = adapter->ind_offset; | |
1581 | kroute.u.adapter.adapter_id = adapter->adapter_id; | |
1582 | ||
1583 | kvm_add_routing_entry(s, &kroute); | |
d426d9fb CH |
1584 | |
1585 | return virq; | |
1586 | } | |
1587 | ||
977a8d9c AS |
1588 | int kvm_irqchip_add_hv_sint_route(KVMState *s, uint32_t vcpu, uint32_t sint) |
1589 | { | |
1590 | struct kvm_irq_routing_entry kroute = {}; | |
1591 | int virq; | |
1592 | ||
1593 | if (!kvm_gsi_routing_enabled()) { | |
1594 | return -ENOSYS; | |
1595 | } | |
1596 | if (!kvm_check_extension(s, KVM_CAP_HYPERV_SYNIC)) { | |
1597 | return -ENOSYS; | |
1598 | } | |
1599 | virq = kvm_irqchip_get_virq(s); | |
1600 | if (virq < 0) { | |
1601 | return virq; | |
1602 | } | |
1603 | ||
1604 | kroute.gsi = virq; | |
1605 | kroute.type = KVM_IRQ_ROUTING_HV_SINT; | |
1606 | kroute.flags = 0; | |
1607 | kroute.u.hv_sint.vcpu = vcpu; | |
1608 | kroute.u.hv_sint.sint = sint; | |
1609 | ||
1610 | kvm_add_routing_entry(s, &kroute); | |
1611 | kvm_irqchip_commit_routes(s); | |
1612 | ||
1613 | return virq; | |
1614 | } | |
1615 | ||
84b058d7 JK |
1616 | #else /* !KVM_CAP_IRQ_ROUTING */ |
1617 | ||
7b774593 | 1618 | void kvm_init_irq_routing(KVMState *s) |
84b058d7 JK |
1619 | { |
1620 | } | |
04fa27f5 | 1621 | |
d3d3bef0 JK |
1622 | void kvm_irqchip_release_virq(KVMState *s, int virq) |
1623 | { | |
1624 | } | |
1625 | ||
04fa27f5 JK |
1626 | int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg) |
1627 | { | |
1628 | abort(); | |
1629 | } | |
92b4e489 | 1630 | |
d1f6af6a | 1631 | int kvm_irqchip_add_msi_route(KVMState *s, int vector, PCIDevice *dev) |
92b4e489 | 1632 | { |
df410675 | 1633 | return -ENOSYS; |
92b4e489 | 1634 | } |
39853bbc | 1635 | |
d426d9fb CH |
1636 | int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter) |
1637 | { | |
1638 | return -ENOSYS; | |
1639 | } | |
1640 | ||
977a8d9c AS |
1641 | int kvm_irqchip_add_hv_sint_route(KVMState *s, uint32_t vcpu, uint32_t sint) |
1642 | { | |
1643 | return -ENOSYS; | |
1644 | } | |
1645 | ||
39853bbc JK |
1646 | static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int virq, bool assign) |
1647 | { | |
1648 | abort(); | |
1649 | } | |
dabe3143 MT |
1650 | |
1651 | int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg) | |
1652 | { | |
1653 | return -ENOSYS; | |
1654 | } | |
84b058d7 JK |
1655 | #endif /* !KVM_CAP_IRQ_ROUTING */ |
1656 | ||
1c9b71a7 EA |
1657 | int kvm_irqchip_add_irqfd_notifier_gsi(KVMState *s, EventNotifier *n, |
1658 | EventNotifier *rn, int virq) | |
39853bbc | 1659 | { |
ca916d37 VM |
1660 | return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n), |
1661 | rn ? event_notifier_get_fd(rn) : -1, virq, true); | |
39853bbc JK |
1662 | } |
1663 | ||
1c9b71a7 EA |
1664 | int kvm_irqchip_remove_irqfd_notifier_gsi(KVMState *s, EventNotifier *n, |
1665 | int virq) | |
15b2bd18 | 1666 | { |
ca916d37 VM |
1667 | return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n), -1, virq, |
1668 | false); | |
15b2bd18 PB |
1669 | } |
1670 | ||
197e3524 EA |
1671 | int kvm_irqchip_add_irqfd_notifier(KVMState *s, EventNotifier *n, |
1672 | EventNotifier *rn, qemu_irq irq) | |
1673 | { | |
1674 | gpointer key, gsi; | |
1675 | gboolean found = g_hash_table_lookup_extended(s->gsimap, irq, &key, &gsi); | |
1676 | ||
1677 | if (!found) { | |
1678 | return -ENXIO; | |
1679 | } | |
1680 | return kvm_irqchip_add_irqfd_notifier_gsi(s, n, rn, GPOINTER_TO_INT(gsi)); | |
1681 | } | |
1682 | ||
1683 | int kvm_irqchip_remove_irqfd_notifier(KVMState *s, EventNotifier *n, | |
1684 | qemu_irq irq) | |
1685 | { | |
1686 | gpointer key, gsi; | |
1687 | gboolean found = g_hash_table_lookup_extended(s->gsimap, irq, &key, &gsi); | |
1688 | ||
1689 | if (!found) { | |
1690 | return -ENXIO; | |
1691 | } | |
1692 | return kvm_irqchip_remove_irqfd_notifier_gsi(s, n, GPOINTER_TO_INT(gsi)); | |
1693 | } | |
1694 | ||
1695 | void kvm_irqchip_set_qemuirq_gsi(KVMState *s, qemu_irq irq, int gsi) | |
1696 | { | |
1697 | g_hash_table_insert(s->gsimap, irq, GINT_TO_POINTER(gsi)); | |
1698 | } | |
1699 | ||
8db4936b | 1700 | static void kvm_irqchip_create(MachineState *machine, KVMState *s) |
84b058d7 | 1701 | { |
84b058d7 JK |
1702 | int ret; |
1703 | ||
8db4936b PB |
1704 | if (kvm_check_extension(s, KVM_CAP_IRQCHIP)) { |
1705 | ; | |
1706 | } else if (kvm_check_extension(s, KVM_CAP_S390_IRQCHIP)) { | |
1707 | ret = kvm_vm_enable_cap(s, KVM_CAP_S390_IRQCHIP, 0); | |
1708 | if (ret < 0) { | |
1709 | fprintf(stderr, "Enable kernel irqchip failed: %s\n", strerror(-ret)); | |
1710 | exit(1); | |
1711 | } | |
1712 | } else { | |
1713 | return; | |
84b058d7 JK |
1714 | } |
1715 | ||
d6032e06 CD |
1716 | /* First probe and see if there's a arch-specific hook to create the |
1717 | * in-kernel irqchip for us */ | |
15eafc2e | 1718 | ret = kvm_arch_irqchip_create(machine, s); |
8db4936b | 1719 | if (ret == 0) { |
15eafc2e PB |
1720 | if (machine_kernel_irqchip_split(machine)) { |
1721 | perror("Split IRQ chip mode not supported."); | |
1722 | exit(1); | |
1723 | } else { | |
1724 | ret = kvm_vm_ioctl(s, KVM_CREATE_IRQCHIP); | |
1725 | } | |
8db4936b PB |
1726 | } |
1727 | if (ret < 0) { | |
1728 | fprintf(stderr, "Create kernel irqchip failed: %s\n", strerror(-ret)); | |
1729 | exit(1); | |
84b058d7 JK |
1730 | } |
1731 | ||
3d4b2649 | 1732 | kvm_kernel_irqchip = true; |
7ae26bd4 PM |
1733 | /* If we have an in-kernel IRQ chip then we must have asynchronous |
1734 | * interrupt delivery (though the reverse is not necessarily true) | |
1735 | */ | |
1736 | kvm_async_interrupts_allowed = true; | |
215e79c0 | 1737 | kvm_halt_in_kernel_allowed = true; |
84b058d7 JK |
1738 | |
1739 | kvm_init_irq_routing(s); | |
1740 | ||
197e3524 | 1741 | s->gsimap = g_hash_table_new(g_direct_hash, g_direct_equal); |
84b058d7 JK |
1742 | } |
1743 | ||
670436ce AJ |
1744 | /* Find number of supported CPUs using the recommended |
1745 | * procedure from the kernel API documentation to cope with | |
1746 | * older kernels that may be missing capabilities. | |
1747 | */ | |
1748 | static int kvm_recommended_vcpus(KVMState *s) | |
3ed444e9 | 1749 | { |
11748ba7 | 1750 | int ret = kvm_vm_check_extension(s, KVM_CAP_NR_VCPUS); |
670436ce AJ |
1751 | return (ret) ? ret : 4; |
1752 | } | |
3ed444e9 | 1753 | |
670436ce AJ |
1754 | static int kvm_max_vcpus(KVMState *s) |
1755 | { | |
1756 | int ret = kvm_check_extension(s, KVM_CAP_MAX_VCPUS); | |
1757 | return (ret) ? ret : kvm_recommended_vcpus(s); | |
3ed444e9 DH |
1758 | } |
1759 | ||
f31e3266 GK |
1760 | static int kvm_max_vcpu_id(KVMState *s) |
1761 | { | |
1762 | int ret = kvm_check_extension(s, KVM_CAP_MAX_VCPU_ID); | |
1763 | return (ret) ? ret : kvm_max_vcpus(s); | |
1764 | } | |
1765 | ||
41264b38 GK |
1766 | bool kvm_vcpu_id_is_valid(int vcpu_id) |
1767 | { | |
1768 | KVMState *s = KVM_STATE(current_machine->accelerator); | |
f31e3266 | 1769 | return vcpu_id >= 0 && vcpu_id < kvm_max_vcpu_id(s); |
41264b38 GK |
1770 | } |
1771 | ||
f6a1ef64 | 1772 | static int kvm_init(MachineState *ms) |
05330448 | 1773 | { |
f6a1ef64 | 1774 | MachineClass *mc = MACHINE_GET_CLASS(ms); |
168ccc11 JK |
1775 | static const char upgrade_note[] = |
1776 | "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n" | |
1777 | "(see http://sourceforge.net/projects/kvm).\n"; | |
670436ce AJ |
1778 | struct { |
1779 | const char *name; | |
1780 | int num; | |
1781 | } num_cpus[] = { | |
5cc8767d LX |
1782 | { "SMP", ms->smp.cpus }, |
1783 | { "hotpluggable", ms->smp.max_cpus }, | |
670436ce AJ |
1784 | { NULL, } |
1785 | }, *nc = num_cpus; | |
1786 | int soft_vcpus_limit, hard_vcpus_limit; | |
05330448 | 1787 | KVMState *s; |
94a8d39a | 1788 | const KVMCapabilityInfo *missing_cap; |
05330448 | 1789 | int ret; |
7bbda04c | 1790 | int type = 0; |
135a129a | 1791 | const char *kvm_type; |
05330448 | 1792 | |
fc02086b | 1793 | s = KVM_STATE(ms->accelerator); |
05330448 | 1794 | |
3145fcb6 DG |
1795 | /* |
1796 | * On systems where the kernel can support different base page | |
1797 | * sizes, host page size may be different from TARGET_PAGE_SIZE, | |
1798 | * even with KVM. TARGET_PAGE_SIZE is assumed to be the minimum | |
1799 | * page size for the system though. | |
1800 | */ | |
1801 | assert(TARGET_PAGE_SIZE <= getpagesize()); | |
1802 | ||
aed6efb9 JH |
1803 | s->sigmask_len = 8; |
1804 | ||
e22a25c9 | 1805 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
72cf2d4f | 1806 | QTAILQ_INIT(&s->kvm_sw_breakpoints); |
e22a25c9 | 1807 | #endif |
4c055ab5 | 1808 | QLIST_INIT(&s->kvm_parked_vcpus); |
05330448 | 1809 | s->vmfd = -1; |
40ff6d7e | 1810 | s->fd = qemu_open("/dev/kvm", O_RDWR); |
05330448 AL |
1811 | if (s->fd == -1) { |
1812 | fprintf(stderr, "Could not access KVM kernel module: %m\n"); | |
1813 | ret = -errno; | |
1814 | goto err; | |
1815 | } | |
1816 | ||
1817 | ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0); | |
1818 | if (ret < KVM_API_VERSION) { | |
0e1dac6c | 1819 | if (ret >= 0) { |
05330448 | 1820 | ret = -EINVAL; |
a426e122 | 1821 | } |
05330448 AL |
1822 | fprintf(stderr, "kvm version too old\n"); |
1823 | goto err; | |
1824 | } | |
1825 | ||
1826 | if (ret > KVM_API_VERSION) { | |
1827 | ret = -EINVAL; | |
1828 | fprintf(stderr, "kvm version not supported\n"); | |
1829 | goto err; | |
1830 | } | |
1831 | ||
cf0f7cf9 | 1832 | kvm_immediate_exit = kvm_check_extension(s, KVM_CAP_IMMEDIATE_EXIT); |
fb541ca5 AW |
1833 | s->nr_slots = kvm_check_extension(s, KVM_CAP_NR_MEMSLOTS); |
1834 | ||
1835 | /* If unspecified, use the default value */ | |
1836 | if (!s->nr_slots) { | |
1837 | s->nr_slots = 32; | |
1838 | } | |
1839 | ||
8072aae3 AK |
1840 | s->nr_as = kvm_check_extension(s, KVM_CAP_MULTI_ADDRESS_SPACE); |
1841 | if (s->nr_as <= 1) { | |
1842 | s->nr_as = 1; | |
1843 | } | |
1844 | s->as = g_new0(struct KVMAs, s->nr_as); | |
1845 | ||
135a129a | 1846 | kvm_type = qemu_opt_get(qemu_get_machine_opts(), "kvm-type"); |
f1e29879 | 1847 | if (mc->kvm_type) { |
dc0ca80e | 1848 | type = mc->kvm_type(ms, kvm_type); |
135a129a | 1849 | } else if (kvm_type) { |
0e1dac6c | 1850 | ret = -EINVAL; |
135a129a AK |
1851 | fprintf(stderr, "Invalid argument kvm-type=%s\n", kvm_type); |
1852 | goto err; | |
1853 | } | |
1854 | ||
94ccff13 | 1855 | do { |
135a129a | 1856 | ret = kvm_ioctl(s, KVM_CREATE_VM, type); |
94ccff13 TK |
1857 | } while (ret == -EINTR); |
1858 | ||
1859 | if (ret < 0) { | |
521f438e | 1860 | fprintf(stderr, "ioctl(KVM_CREATE_VM) failed: %d %s\n", -ret, |
94ccff13 TK |
1861 | strerror(-ret)); |
1862 | ||
0104dcac | 1863 | #ifdef TARGET_S390X |
2c80e996 CH |
1864 | if (ret == -EINVAL) { |
1865 | fprintf(stderr, | |
1866 | "Host kernel setup problem detected. Please verify:\n"); | |
1867 | fprintf(stderr, "- for kernels supporting the switch_amode or" | |
1868 | " user_mode parameters, whether\n"); | |
1869 | fprintf(stderr, | |
1870 | " user space is running in primary address space\n"); | |
1871 | fprintf(stderr, | |
1872 | "- for kernels supporting the vm.allocate_pgste sysctl, " | |
1873 | "whether it is enabled\n"); | |
1874 | } | |
0104dcac | 1875 | #endif |
05330448 | 1876 | goto err; |
0104dcac | 1877 | } |
05330448 | 1878 | |
94ccff13 | 1879 | s->vmfd = ret; |
11748ba7 GK |
1880 | |
1881 | /* check the vcpu limits */ | |
1882 | soft_vcpus_limit = kvm_recommended_vcpus(s); | |
1883 | hard_vcpus_limit = kvm_max_vcpus(s); | |
1884 | ||
1885 | while (nc->name) { | |
1886 | if (nc->num > soft_vcpus_limit) { | |
1887 | warn_report("Number of %s cpus requested (%d) exceeds " | |
1888 | "the recommended cpus supported by KVM (%d)", | |
1889 | nc->name, nc->num, soft_vcpus_limit); | |
1890 | ||
1891 | if (nc->num > hard_vcpus_limit) { | |
1892 | fprintf(stderr, "Number of %s cpus requested (%d) exceeds " | |
1893 | "the maximum cpus supported by KVM (%d)\n", | |
1894 | nc->name, nc->num, hard_vcpus_limit); | |
1895 | exit(1); | |
1896 | } | |
1897 | } | |
1898 | nc++; | |
1899 | } | |
1900 | ||
94a8d39a JK |
1901 | missing_cap = kvm_check_extension_list(s, kvm_required_capabilites); |
1902 | if (!missing_cap) { | |
1903 | missing_cap = | |
1904 | kvm_check_extension_list(s, kvm_arch_required_capabilities); | |
05330448 | 1905 | } |
94a8d39a | 1906 | if (missing_cap) { |
ad7b8b33 | 1907 | ret = -EINVAL; |
94a8d39a JK |
1908 | fprintf(stderr, "kvm does not support %s\n%s", |
1909 | missing_cap->name, upgrade_note); | |
d85dc283 AL |
1910 | goto err; |
1911 | } | |
1912 | ||
ad7b8b33 | 1913 | s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO); |
e6d34aee PH |
1914 | s->coalesced_pio = s->coalesced_mmio && |
1915 | kvm_check_extension(s, KVM_CAP_COALESCED_PIO); | |
f65ed4c1 | 1916 | |
ff4aa114 PX |
1917 | s->manual_dirty_log_protect = |
1918 | kvm_check_extension(s, KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2); | |
1919 | if (s->manual_dirty_log_protect) { | |
1920 | ret = kvm_vm_enable_cap(s, KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2, 0, 1); | |
1921 | if (ret) { | |
1922 | warn_report("Trying to enable KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2 " | |
1923 | "but failed. Falling back to the legacy mode. "); | |
1924 | s->manual_dirty_log_protect = false; | |
1925 | } | |
1926 | } | |
1927 | ||
a0fb002c JK |
1928 | #ifdef KVM_CAP_VCPU_EVENTS |
1929 | s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS); | |
1930 | #endif | |
1931 | ||
b0b1d690 JK |
1932 | s->robust_singlestep = |
1933 | kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP); | |
b0b1d690 | 1934 | |
ff44f1a3 JK |
1935 | #ifdef KVM_CAP_DEBUGREGS |
1936 | s->debugregs = kvm_check_extension(s, KVM_CAP_DEBUGREGS); | |
1937 | #endif | |
1938 | ||
ebbfef2f LA |
1939 | s->max_nested_state_len = kvm_check_extension(s, KVM_CAP_NESTED_STATE); |
1940 | ||
d3d3bef0 | 1941 | #ifdef KVM_CAP_IRQ_ROUTING |
50bf31b9 | 1942 | kvm_direct_msi_allowed = (kvm_check_extension(s, KVM_CAP_SIGNAL_MSI) > 0); |
d3d3bef0 | 1943 | #endif |
4a3adebb | 1944 | |
3ab73842 JK |
1945 | s->intx_set_mask = kvm_check_extension(s, KVM_CAP_PCI_2_3); |
1946 | ||
e333cd69 | 1947 | s->irq_set_ioctl = KVM_IRQ_LINE; |
8732fbd2 | 1948 | if (kvm_check_extension(s, KVM_CAP_IRQ_INJECT_STATUS)) { |
e333cd69 | 1949 | s->irq_set_ioctl = KVM_IRQ_LINE_STATUS; |
8732fbd2 PM |
1950 | } |
1951 | ||
df9c8b75 JJ |
1952 | kvm_readonly_mem_allowed = |
1953 | (kvm_check_extension(s, KVM_CAP_READONLY_MEM) > 0); | |
df9c8b75 | 1954 | |
69e03ae6 NN |
1955 | kvm_eventfds_allowed = |
1956 | (kvm_check_extension(s, KVM_CAP_IOEVENTFD) > 0); | |
1957 | ||
f41389ae EA |
1958 | kvm_irqfds_allowed = |
1959 | (kvm_check_extension(s, KVM_CAP_IRQFD) > 0); | |
1960 | ||
1961 | kvm_resamplefds_allowed = | |
1962 | (kvm_check_extension(s, KVM_CAP_IRQFD_RESAMPLE) > 0); | |
1963 | ||
d0a073a1 DD |
1964 | kvm_vm_attributes_allowed = |
1965 | (kvm_check_extension(s, KVM_CAP_VM_ATTRIBUTES) > 0); | |
1966 | ||
35108223 JW |
1967 | kvm_ioeventfd_any_length_allowed = |
1968 | (kvm_check_extension(s, KVM_CAP_IOEVENTFD_ANY_LENGTH) > 0); | |
1969 | ||
d870cfde GA |
1970 | kvm_state = s; |
1971 | ||
b20e3780 BS |
1972 | /* |
1973 | * if memory encryption object is specified then initialize the memory | |
1974 | * encryption context. | |
1975 | */ | |
1976 | if (ms->memory_encryption) { | |
1977 | kvm_state->memcrypt_handle = sev_guest_init(ms->memory_encryption); | |
1978 | if (!kvm_state->memcrypt_handle) { | |
1979 | ret = -1; | |
1980 | goto err; | |
1981 | } | |
54e89539 BS |
1982 | |
1983 | kvm_state->memcrypt_encrypt_data = sev_encrypt_data; | |
b20e3780 BS |
1984 | } |
1985 | ||
b16565b3 | 1986 | ret = kvm_arch_init(ms, s); |
a426e122 | 1987 | if (ret < 0) { |
05330448 | 1988 | goto err; |
a426e122 | 1989 | } |
05330448 | 1990 | |
8db4936b PB |
1991 | if (machine_kernel_irqchip_allowed(ms)) { |
1992 | kvm_irqchip_create(ms, s); | |
84b058d7 JK |
1993 | } |
1994 | ||
8c56c1a5 PF |
1995 | if (kvm_eventfds_allowed) { |
1996 | s->memory_listener.listener.eventfd_add = kvm_mem_ioeventfd_add; | |
1997 | s->memory_listener.listener.eventfd_del = kvm_mem_ioeventfd_del; | |
1998 | } | |
e6d34aee PH |
1999 | s->memory_listener.listener.coalesced_io_add = kvm_coalesce_mmio_region; |
2000 | s->memory_listener.listener.coalesced_io_del = kvm_uncoalesce_mmio_region; | |
7bbda04c PB |
2001 | |
2002 | kvm_memory_listener_register(s, &s->memory_listener, | |
38bfe691 | 2003 | &address_space_memory, 0); |
7bbda04c PB |
2004 | memory_listener_register(&kvm_io_listener, |
2005 | &address_space_io); | |
e6d34aee PH |
2006 | memory_listener_register(&kvm_coalesced_pio_listener, |
2007 | &address_space_io); | |
05330448 | 2008 | |
d2f2b8a7 SH |
2009 | s->many_ioeventfds = kvm_check_many_ioeventfds(); |
2010 | ||
62dd4eda | 2011 | s->sync_mmu = !!kvm_vm_check_extension(kvm_state, KVM_CAP_SYNC_MMU); |
f5948942 AW |
2012 | if (!s->sync_mmu) { |
2013 | qemu_balloon_inhibit(true); | |
2014 | } | |
62dd4eda | 2015 | |
05330448 AL |
2016 | return 0; |
2017 | ||
2018 | err: | |
0e1dac6c | 2019 | assert(ret < 0); |
6d1cc321 SW |
2020 | if (s->vmfd >= 0) { |
2021 | close(s->vmfd); | |
2022 | } | |
2023 | if (s->fd != -1) { | |
2024 | close(s->fd); | |
05330448 | 2025 | } |
7bbda04c | 2026 | g_free(s->memory_listener.slots); |
05330448 AL |
2027 | |
2028 | return ret; | |
2029 | } | |
2030 | ||
aed6efb9 JH |
2031 | void kvm_set_sigmask_len(KVMState *s, unsigned int sigmask_len) |
2032 | { | |
2033 | s->sigmask_len = sigmask_len; | |
2034 | } | |
2035 | ||
4c663752 PB |
2036 | static void kvm_handle_io(uint16_t port, MemTxAttrs attrs, void *data, int direction, |
2037 | int size, uint32_t count) | |
05330448 AL |
2038 | { |
2039 | int i; | |
2040 | uint8_t *ptr = data; | |
2041 | ||
2042 | for (i = 0; i < count; i++) { | |
4c663752 | 2043 | address_space_rw(&address_space_io, port, attrs, |
5c9eb028 | 2044 | ptr, size, |
354678c5 | 2045 | direction == KVM_EXIT_IO_OUT); |
05330448 AL |
2046 | ptr += size; |
2047 | } | |
05330448 AL |
2048 | } |
2049 | ||
5326ab55 | 2050 | static int kvm_handle_internal_error(CPUState *cpu, struct kvm_run *run) |
7c80eef8 | 2051 | { |
977c7b6d RK |
2052 | fprintf(stderr, "KVM internal error. Suberror: %d\n", |
2053 | run->internal.suberror); | |
2054 | ||
7c80eef8 MT |
2055 | if (kvm_check_extension(kvm_state, KVM_CAP_INTERNAL_ERROR_DATA)) { |
2056 | int i; | |
2057 | ||
7c80eef8 MT |
2058 | for (i = 0; i < run->internal.ndata; ++i) { |
2059 | fprintf(stderr, "extra data[%d]: %"PRIx64"\n", | |
2060 | i, (uint64_t)run->internal.data[i]); | |
2061 | } | |
2062 | } | |
7c80eef8 MT |
2063 | if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) { |
2064 | fprintf(stderr, "emulation failure\n"); | |
20d695a9 | 2065 | if (!kvm_arch_stop_on_emulation_error(cpu)) { |
90c84c56 | 2066 | cpu_dump_state(cpu, stderr, CPU_DUMP_CODE); |
d73cd8f4 | 2067 | return EXCP_INTERRUPT; |
a426e122 | 2068 | } |
7c80eef8 MT |
2069 | } |
2070 | /* FIXME: Should trigger a qmp message to let management know | |
2071 | * something went wrong. | |
2072 | */ | |
73aaec4a | 2073 | return -1; |
7c80eef8 | 2074 | } |
7c80eef8 | 2075 | |
62a2744c | 2076 | void kvm_flush_coalesced_mmio_buffer(void) |
f65ed4c1 | 2077 | { |
f65ed4c1 | 2078 | KVMState *s = kvm_state; |
1cae88b9 AK |
2079 | |
2080 | if (s->coalesced_flush_in_progress) { | |
2081 | return; | |
2082 | } | |
2083 | ||
2084 | s->coalesced_flush_in_progress = true; | |
2085 | ||
62a2744c SY |
2086 | if (s->coalesced_mmio_ring) { |
2087 | struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring; | |
f65ed4c1 AL |
2088 | while (ring->first != ring->last) { |
2089 | struct kvm_coalesced_mmio *ent; | |
2090 | ||
2091 | ent = &ring->coalesced_mmio[ring->first]; | |
2092 | ||
e6d34aee PH |
2093 | if (ent->pio == 1) { |
2094 | address_space_rw(&address_space_io, ent->phys_addr, | |
2095 | MEMTXATTRS_UNSPECIFIED, ent->data, | |
2096 | ent->len, true); | |
2097 | } else { | |
2098 | cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len); | |
2099 | } | |
85199474 | 2100 | smp_wmb(); |
f65ed4c1 AL |
2101 | ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX; |
2102 | } | |
2103 | } | |
1cae88b9 AK |
2104 | |
2105 | s->coalesced_flush_in_progress = false; | |
f65ed4c1 AL |
2106 | } |
2107 | ||
14e6fe12 | 2108 | static void do_kvm_cpu_synchronize_state(CPUState *cpu, run_on_cpu_data arg) |
4c0960c0 | 2109 | { |
99f31832 | 2110 | if (!cpu->vcpu_dirty) { |
20d695a9 | 2111 | kvm_arch_get_registers(cpu); |
99f31832 | 2112 | cpu->vcpu_dirty = true; |
4c0960c0 AK |
2113 | } |
2114 | } | |
2115 | ||
dd1750d7 | 2116 | void kvm_cpu_synchronize_state(CPUState *cpu) |
2705d56a | 2117 | { |
99f31832 | 2118 | if (!cpu->vcpu_dirty) { |
14e6fe12 | 2119 | run_on_cpu(cpu, do_kvm_cpu_synchronize_state, RUN_ON_CPU_NULL); |
a426e122 | 2120 | } |
2705d56a JK |
2121 | } |
2122 | ||
14e6fe12 | 2123 | static void do_kvm_cpu_synchronize_post_reset(CPUState *cpu, run_on_cpu_data arg) |
ea375f9a | 2124 | { |
20d695a9 | 2125 | kvm_arch_put_registers(cpu, KVM_PUT_RESET_STATE); |
99f31832 | 2126 | cpu->vcpu_dirty = false; |
ea375f9a JK |
2127 | } |
2128 | ||
c8e2085d DH |
2129 | void kvm_cpu_synchronize_post_reset(CPUState *cpu) |
2130 | { | |
14e6fe12 | 2131 | run_on_cpu(cpu, do_kvm_cpu_synchronize_post_reset, RUN_ON_CPU_NULL); |
c8e2085d DH |
2132 | } |
2133 | ||
14e6fe12 | 2134 | static void do_kvm_cpu_synchronize_post_init(CPUState *cpu, run_on_cpu_data arg) |
ea375f9a | 2135 | { |
20d695a9 | 2136 | kvm_arch_put_registers(cpu, KVM_PUT_FULL_STATE); |
99f31832 | 2137 | cpu->vcpu_dirty = false; |
ea375f9a JK |
2138 | } |
2139 | ||
c8e2085d DH |
2140 | void kvm_cpu_synchronize_post_init(CPUState *cpu) |
2141 | { | |
14e6fe12 | 2142 | run_on_cpu(cpu, do_kvm_cpu_synchronize_post_init, RUN_ON_CPU_NULL); |
c8e2085d DH |
2143 | } |
2144 | ||
75e972da DG |
2145 | static void do_kvm_cpu_synchronize_pre_loadvm(CPUState *cpu, run_on_cpu_data arg) |
2146 | { | |
99f31832 | 2147 | cpu->vcpu_dirty = true; |
75e972da DG |
2148 | } |
2149 | ||
2150 | void kvm_cpu_synchronize_pre_loadvm(CPUState *cpu) | |
2151 | { | |
2152 | run_on_cpu(cpu, do_kvm_cpu_synchronize_pre_loadvm, RUN_ON_CPU_NULL); | |
2153 | } | |
2154 | ||
2ae41db2 PB |
2155 | #ifdef KVM_HAVE_MCE_INJECTION |
2156 | static __thread void *pending_sigbus_addr; | |
2157 | static __thread int pending_sigbus_code; | |
2158 | static __thread bool have_sigbus_pending; | |
2159 | #endif | |
2160 | ||
cf0f7cf9 PB |
2161 | static void kvm_cpu_kick(CPUState *cpu) |
2162 | { | |
2163 | atomic_set(&cpu->kvm_run->immediate_exit, 1); | |
2164 | } | |
2165 | ||
2166 | static void kvm_cpu_kick_self(void) | |
2167 | { | |
2168 | if (kvm_immediate_exit) { | |
2169 | kvm_cpu_kick(current_cpu); | |
2170 | } else { | |
2171 | qemu_cpu_kick_self(); | |
2172 | } | |
2173 | } | |
2174 | ||
18268b60 PB |
2175 | static void kvm_eat_signals(CPUState *cpu) |
2176 | { | |
2177 | struct timespec ts = { 0, 0 }; | |
2178 | siginfo_t siginfo; | |
2179 | sigset_t waitset; | |
2180 | sigset_t chkset; | |
2181 | int r; | |
2182 | ||
cf0f7cf9 PB |
2183 | if (kvm_immediate_exit) { |
2184 | atomic_set(&cpu->kvm_run->immediate_exit, 0); | |
2185 | /* Write kvm_run->immediate_exit before the cpu->exit_request | |
2186 | * write in kvm_cpu_exec. | |
2187 | */ | |
2188 | smp_wmb(); | |
2189 | return; | |
2190 | } | |
2191 | ||
18268b60 PB |
2192 | sigemptyset(&waitset); |
2193 | sigaddset(&waitset, SIG_IPI); | |
2194 | ||
2195 | do { | |
2196 | r = sigtimedwait(&waitset, &siginfo, &ts); | |
2197 | if (r == -1 && !(errno == EAGAIN || errno == EINTR)) { | |
2198 | perror("sigtimedwait"); | |
2199 | exit(1); | |
2200 | } | |
2201 | ||
2202 | r = sigpending(&chkset); | |
2203 | if (r == -1) { | |
2204 | perror("sigpending"); | |
2205 | exit(1); | |
2206 | } | |
2207 | } while (sigismember(&chkset, SIG_IPI)); | |
2208 | } | |
2209 | ||
1458c363 | 2210 | int kvm_cpu_exec(CPUState *cpu) |
05330448 | 2211 | { |
f7575c96 | 2212 | struct kvm_run *run = cpu->kvm_run; |
7cbb533f | 2213 | int ret, run_ret; |
05330448 | 2214 | |
8c0d577e | 2215 | DPRINTF("kvm_cpu_exec()\n"); |
05330448 | 2216 | |
20d695a9 | 2217 | if (kvm_arch_process_async_events(cpu)) { |
c5c6679d | 2218 | atomic_set(&cpu->exit_request, 0); |
6792a57b | 2219 | return EXCP_HLT; |
9ccfac9e | 2220 | } |
0af691d7 | 2221 | |
4b8523ee | 2222 | qemu_mutex_unlock_iothread(); |
1d78a3c3 | 2223 | cpu_exec_start(cpu); |
4b8523ee | 2224 | |
9ccfac9e | 2225 | do { |
4c663752 PB |
2226 | MemTxAttrs attrs; |
2227 | ||
99f31832 | 2228 | if (cpu->vcpu_dirty) { |
20d695a9 | 2229 | kvm_arch_put_registers(cpu, KVM_PUT_RUNTIME_STATE); |
99f31832 | 2230 | cpu->vcpu_dirty = false; |
4c0960c0 AK |
2231 | } |
2232 | ||
20d695a9 | 2233 | kvm_arch_pre_run(cpu, run); |
c5c6679d | 2234 | if (atomic_read(&cpu->exit_request)) { |
9ccfac9e JK |
2235 | DPRINTF("interrupt exit requested\n"); |
2236 | /* | |
2237 | * KVM requires us to reenter the kernel after IO exits to complete | |
2238 | * instruction emulation. This self-signal will ensure that we | |
2239 | * leave ASAP again. | |
2240 | */ | |
cf0f7cf9 | 2241 | kvm_cpu_kick_self(); |
9ccfac9e | 2242 | } |
9ccfac9e | 2243 | |
cf0f7cf9 PB |
2244 | /* Read cpu->exit_request before KVM_RUN reads run->immediate_exit. |
2245 | * Matching barrier in kvm_eat_signals. | |
2246 | */ | |
2247 | smp_rmb(); | |
2248 | ||
1bc22652 | 2249 | run_ret = kvm_vcpu_ioctl(cpu, KVM_RUN, 0); |
9ccfac9e | 2250 | |
4c663752 | 2251 | attrs = kvm_arch_post_run(cpu, run); |
05330448 | 2252 | |
2ae41db2 PB |
2253 | #ifdef KVM_HAVE_MCE_INJECTION |
2254 | if (unlikely(have_sigbus_pending)) { | |
2255 | qemu_mutex_lock_iothread(); | |
2256 | kvm_arch_on_sigbus_vcpu(cpu, pending_sigbus_code, | |
2257 | pending_sigbus_addr); | |
2258 | have_sigbus_pending = false; | |
2259 | qemu_mutex_unlock_iothread(); | |
2260 | } | |
2261 | #endif | |
2262 | ||
7cbb533f | 2263 | if (run_ret < 0) { |
dc77d341 JK |
2264 | if (run_ret == -EINTR || run_ret == -EAGAIN) { |
2265 | DPRINTF("io window exit\n"); | |
18268b60 | 2266 | kvm_eat_signals(cpu); |
d73cd8f4 | 2267 | ret = EXCP_INTERRUPT; |
dc77d341 JK |
2268 | break; |
2269 | } | |
7b011fbc ME |
2270 | fprintf(stderr, "error: kvm run failed %s\n", |
2271 | strerror(-run_ret)); | |
dae02ba5 LV |
2272 | #ifdef TARGET_PPC |
2273 | if (run_ret == -EBUSY) { | |
2274 | fprintf(stderr, | |
2275 | "This is probably because your SMT is enabled.\n" | |
2276 | "VCPU can only run on primary threads with all " | |
2277 | "secondary threads offline.\n"); | |
2278 | } | |
2279 | #endif | |
a85e130e PB |
2280 | ret = -1; |
2281 | break; | |
05330448 AL |
2282 | } |
2283 | ||
b76ac80a | 2284 | trace_kvm_run_exit(cpu->cpu_index, run->exit_reason); |
05330448 AL |
2285 | switch (run->exit_reason) { |
2286 | case KVM_EXIT_IO: | |
8c0d577e | 2287 | DPRINTF("handle_io\n"); |
80b7d2ef | 2288 | /* Called outside BQL */ |
4c663752 | 2289 | kvm_handle_io(run->io.port, attrs, |
b30e93e9 JK |
2290 | (uint8_t *)run + run->io.data_offset, |
2291 | run->io.direction, | |
2292 | run->io.size, | |
2293 | run->io.count); | |
d73cd8f4 | 2294 | ret = 0; |
05330448 AL |
2295 | break; |
2296 | case KVM_EXIT_MMIO: | |
8c0d577e | 2297 | DPRINTF("handle_mmio\n"); |
de7ea885 | 2298 | /* Called outside BQL */ |
4c663752 PB |
2299 | address_space_rw(&address_space_memory, |
2300 | run->mmio.phys_addr, attrs, | |
2301 | run->mmio.data, | |
2302 | run->mmio.len, | |
2303 | run->mmio.is_write); | |
d73cd8f4 | 2304 | ret = 0; |
05330448 AL |
2305 | break; |
2306 | case KVM_EXIT_IRQ_WINDOW_OPEN: | |
8c0d577e | 2307 | DPRINTF("irq_window_open\n"); |
d73cd8f4 | 2308 | ret = EXCP_INTERRUPT; |
05330448 AL |
2309 | break; |
2310 | case KVM_EXIT_SHUTDOWN: | |
8c0d577e | 2311 | DPRINTF("shutdown\n"); |
cf83f140 | 2312 | qemu_system_reset_request(SHUTDOWN_CAUSE_GUEST_RESET); |
d73cd8f4 | 2313 | ret = EXCP_INTERRUPT; |
05330448 AL |
2314 | break; |
2315 | case KVM_EXIT_UNKNOWN: | |
bb44e0d1 JK |
2316 | fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "\n", |
2317 | (uint64_t)run->hw.hardware_exit_reason); | |
73aaec4a | 2318 | ret = -1; |
05330448 | 2319 | break; |
7c80eef8 | 2320 | case KVM_EXIT_INTERNAL_ERROR: |
5326ab55 | 2321 | ret = kvm_handle_internal_error(cpu, run); |
7c80eef8 | 2322 | break; |
99040447 PS |
2323 | case KVM_EXIT_SYSTEM_EVENT: |
2324 | switch (run->system_event.type) { | |
2325 | case KVM_SYSTEM_EVENT_SHUTDOWN: | |
cf83f140 | 2326 | qemu_system_shutdown_request(SHUTDOWN_CAUSE_GUEST_SHUTDOWN); |
99040447 PS |
2327 | ret = EXCP_INTERRUPT; |
2328 | break; | |
2329 | case KVM_SYSTEM_EVENT_RESET: | |
cf83f140 | 2330 | qemu_system_reset_request(SHUTDOWN_CAUSE_GUEST_RESET); |
99040447 PS |
2331 | ret = EXCP_INTERRUPT; |
2332 | break; | |
7c207b90 | 2333 | case KVM_SYSTEM_EVENT_CRASH: |
d187e08d | 2334 | kvm_cpu_synchronize_state(cpu); |
7c207b90 | 2335 | qemu_mutex_lock_iothread(); |
c86f106b | 2336 | qemu_system_guest_panicked(cpu_get_crash_info(cpu)); |
7c207b90 AS |
2337 | qemu_mutex_unlock_iothread(); |
2338 | ret = 0; | |
2339 | break; | |
99040447 PS |
2340 | default: |
2341 | DPRINTF("kvm_arch_handle_exit\n"); | |
2342 | ret = kvm_arch_handle_exit(cpu, run); | |
2343 | break; | |
2344 | } | |
2345 | break; | |
05330448 | 2346 | default: |
8c0d577e | 2347 | DPRINTF("kvm_arch_handle_exit\n"); |
20d695a9 | 2348 | ret = kvm_arch_handle_exit(cpu, run); |
05330448 AL |
2349 | break; |
2350 | } | |
d73cd8f4 | 2351 | } while (ret == 0); |
05330448 | 2352 | |
1d78a3c3 | 2353 | cpu_exec_end(cpu); |
4b8523ee JK |
2354 | qemu_mutex_lock_iothread(); |
2355 | ||
73aaec4a | 2356 | if (ret < 0) { |
90c84c56 | 2357 | cpu_dump_state(cpu, stderr, CPU_DUMP_CODE); |
0461d5a6 | 2358 | vm_stop(RUN_STATE_INTERNAL_ERROR); |
becfc390 AL |
2359 | } |
2360 | ||
c5c6679d | 2361 | atomic_set(&cpu->exit_request, 0); |
05330448 AL |
2362 | return ret; |
2363 | } | |
2364 | ||
984b5181 | 2365 | int kvm_ioctl(KVMState *s, int type, ...) |
05330448 AL |
2366 | { |
2367 | int ret; | |
984b5181 AL |
2368 | void *arg; |
2369 | va_list ap; | |
05330448 | 2370 | |
984b5181 AL |
2371 | va_start(ap, type); |
2372 | arg = va_arg(ap, void *); | |
2373 | va_end(ap); | |
2374 | ||
9c775729 | 2375 | trace_kvm_ioctl(type, arg); |
984b5181 | 2376 | ret = ioctl(s->fd, type, arg); |
a426e122 | 2377 | if (ret == -1) { |
05330448 | 2378 | ret = -errno; |
a426e122 | 2379 | } |
05330448 AL |
2380 | return ret; |
2381 | } | |
2382 | ||
984b5181 | 2383 | int kvm_vm_ioctl(KVMState *s, int type, ...) |
05330448 AL |
2384 | { |
2385 | int ret; | |
984b5181 AL |
2386 | void *arg; |
2387 | va_list ap; | |
2388 | ||
2389 | va_start(ap, type); | |
2390 | arg = va_arg(ap, void *); | |
2391 | va_end(ap); | |
05330448 | 2392 | |
9c775729 | 2393 | trace_kvm_vm_ioctl(type, arg); |
984b5181 | 2394 | ret = ioctl(s->vmfd, type, arg); |
a426e122 | 2395 | if (ret == -1) { |
05330448 | 2396 | ret = -errno; |
a426e122 | 2397 | } |
05330448 AL |
2398 | return ret; |
2399 | } | |
2400 | ||
1bc22652 | 2401 | int kvm_vcpu_ioctl(CPUState *cpu, int type, ...) |
05330448 AL |
2402 | { |
2403 | int ret; | |
984b5181 AL |
2404 | void *arg; |
2405 | va_list ap; | |
2406 | ||
2407 | va_start(ap, type); | |
2408 | arg = va_arg(ap, void *); | |
2409 | va_end(ap); | |
05330448 | 2410 | |
9c775729 | 2411 | trace_kvm_vcpu_ioctl(cpu->cpu_index, type, arg); |
8737c51c | 2412 | ret = ioctl(cpu->kvm_fd, type, arg); |
a426e122 | 2413 | if (ret == -1) { |
05330448 | 2414 | ret = -errno; |
a426e122 | 2415 | } |
05330448 AL |
2416 | return ret; |
2417 | } | |
bd322087 | 2418 | |
0a6a7cca CD |
2419 | int kvm_device_ioctl(int fd, int type, ...) |
2420 | { | |
2421 | int ret; | |
2422 | void *arg; | |
2423 | va_list ap; | |
2424 | ||
2425 | va_start(ap, type); | |
2426 | arg = va_arg(ap, void *); | |
2427 | va_end(ap); | |
2428 | ||
2429 | trace_kvm_device_ioctl(fd, type, arg); | |
2430 | ret = ioctl(fd, type, arg); | |
2431 | if (ret == -1) { | |
2432 | ret = -errno; | |
2433 | } | |
2434 | return ret; | |
2435 | } | |
2436 | ||
d0a073a1 DD |
2437 | int kvm_vm_check_attr(KVMState *s, uint32_t group, uint64_t attr) |
2438 | { | |
2439 | int ret; | |
2440 | struct kvm_device_attr attribute = { | |
2441 | .group = group, | |
2442 | .attr = attr, | |
2443 | }; | |
2444 | ||
2445 | if (!kvm_vm_attributes_allowed) { | |
2446 | return 0; | |
2447 | } | |
2448 | ||
2449 | ret = kvm_vm_ioctl(s, KVM_HAS_DEVICE_ATTR, &attribute); | |
2450 | /* kvm returns 0 on success for HAS_DEVICE_ATTR */ | |
2451 | return ret ? 0 : 1; | |
2452 | } | |
2453 | ||
4b3cfe72 PF |
2454 | int kvm_device_check_attr(int dev_fd, uint32_t group, uint64_t attr) |
2455 | { | |
2456 | struct kvm_device_attr attribute = { | |
2457 | .group = group, | |
2458 | .attr = attr, | |
2459 | .flags = 0, | |
2460 | }; | |
2461 | ||
2462 | return kvm_device_ioctl(dev_fd, KVM_HAS_DEVICE_ATTR, &attribute) ? 0 : 1; | |
2463 | } | |
2464 | ||
556969e9 EA |
2465 | int kvm_device_access(int fd, int group, uint64_t attr, |
2466 | void *val, bool write, Error **errp) | |
4b3cfe72 PF |
2467 | { |
2468 | struct kvm_device_attr kvmattr; | |
2469 | int err; | |
2470 | ||
2471 | kvmattr.flags = 0; | |
2472 | kvmattr.group = group; | |
2473 | kvmattr.attr = attr; | |
2474 | kvmattr.addr = (uintptr_t)val; | |
2475 | ||
2476 | err = kvm_device_ioctl(fd, | |
2477 | write ? KVM_SET_DEVICE_ATTR : KVM_GET_DEVICE_ATTR, | |
2478 | &kvmattr); | |
2479 | if (err < 0) { | |
556969e9 EA |
2480 | error_setg_errno(errp, -err, |
2481 | "KVM_%s_DEVICE_ATTR failed: Group %d " | |
2482 | "attr 0x%016" PRIx64, | |
2483 | write ? "SET" : "GET", group, attr); | |
4b3cfe72 | 2484 | } |
556969e9 | 2485 | return err; |
4b3cfe72 PF |
2486 | } |
2487 | ||
62dd4eda | 2488 | bool kvm_has_sync_mmu(void) |
bd322087 | 2489 | { |
62dd4eda | 2490 | return kvm_state->sync_mmu; |
bd322087 | 2491 | } |
e22a25c9 | 2492 | |
a0fb002c JK |
2493 | int kvm_has_vcpu_events(void) |
2494 | { | |
2495 | return kvm_state->vcpu_events; | |
2496 | } | |
2497 | ||
b0b1d690 JK |
2498 | int kvm_has_robust_singlestep(void) |
2499 | { | |
2500 | return kvm_state->robust_singlestep; | |
2501 | } | |
2502 | ||
ff44f1a3 JK |
2503 | int kvm_has_debugregs(void) |
2504 | { | |
2505 | return kvm_state->debugregs; | |
2506 | } | |
2507 | ||
ebbfef2f LA |
2508 | int kvm_max_nested_state_length(void) |
2509 | { | |
2510 | return kvm_state->max_nested_state_len; | |
2511 | } | |
2512 | ||
d2f2b8a7 SH |
2513 | int kvm_has_many_ioeventfds(void) |
2514 | { | |
2515 | if (!kvm_enabled()) { | |
2516 | return 0; | |
2517 | } | |
2518 | return kvm_state->many_ioeventfds; | |
2519 | } | |
2520 | ||
84b058d7 JK |
2521 | int kvm_has_gsi_routing(void) |
2522 | { | |
a9c5eb0d | 2523 | #ifdef KVM_CAP_IRQ_ROUTING |
84b058d7 | 2524 | return kvm_check_extension(kvm_state, KVM_CAP_IRQ_ROUTING); |
a9c5eb0d AG |
2525 | #else |
2526 | return false; | |
2527 | #endif | |
84b058d7 JK |
2528 | } |
2529 | ||
3ab73842 JK |
2530 | int kvm_has_intx_set_mask(void) |
2531 | { | |
2532 | return kvm_state->intx_set_mask; | |
2533 | } | |
2534 | ||
5d721b78 AG |
2535 | bool kvm_arm_supports_user_irq(void) |
2536 | { | |
2537 | return kvm_check_extension(kvm_state, KVM_CAP_ARM_USER_IRQ); | |
2538 | } | |
2539 | ||
e22a25c9 | 2540 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
a60f24b5 | 2541 | struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *cpu, |
e22a25c9 AL |
2542 | target_ulong pc) |
2543 | { | |
2544 | struct kvm_sw_breakpoint *bp; | |
2545 | ||
a60f24b5 | 2546 | QTAILQ_FOREACH(bp, &cpu->kvm_state->kvm_sw_breakpoints, entry) { |
a426e122 | 2547 | if (bp->pc == pc) { |
e22a25c9 | 2548 | return bp; |
a426e122 | 2549 | } |
e22a25c9 AL |
2550 | } |
2551 | return NULL; | |
2552 | } | |
2553 | ||
a60f24b5 | 2554 | int kvm_sw_breakpoints_active(CPUState *cpu) |
e22a25c9 | 2555 | { |
a60f24b5 | 2556 | return !QTAILQ_EMPTY(&cpu->kvm_state->kvm_sw_breakpoints); |
e22a25c9 AL |
2557 | } |
2558 | ||
452e4751 GC |
2559 | struct kvm_set_guest_debug_data { |
2560 | struct kvm_guest_debug dbg; | |
452e4751 GC |
2561 | int err; |
2562 | }; | |
2563 | ||
14e6fe12 | 2564 | static void kvm_invoke_set_guest_debug(CPUState *cpu, run_on_cpu_data data) |
452e4751 | 2565 | { |
14e6fe12 PB |
2566 | struct kvm_set_guest_debug_data *dbg_data = |
2567 | (struct kvm_set_guest_debug_data *) data.host_ptr; | |
b3807725 | 2568 | |
3c0ed2a3 | 2569 | dbg_data->err = kvm_vcpu_ioctl(cpu, KVM_SET_GUEST_DEBUG, |
a60f24b5 | 2570 | &dbg_data->dbg); |
452e4751 GC |
2571 | } |
2572 | ||
38e478ec | 2573 | int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap) |
e22a25c9 | 2574 | { |
452e4751 | 2575 | struct kvm_set_guest_debug_data data; |
e22a25c9 | 2576 | |
b0b1d690 | 2577 | data.dbg.control = reinject_trap; |
e22a25c9 | 2578 | |
ed2803da | 2579 | if (cpu->singlestep_enabled) { |
b0b1d690 JK |
2580 | data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP; |
2581 | } | |
20d695a9 | 2582 | kvm_arch_update_guest_debug(cpu, &data.dbg); |
e22a25c9 | 2583 | |
14e6fe12 PB |
2584 | run_on_cpu(cpu, kvm_invoke_set_guest_debug, |
2585 | RUN_ON_CPU_HOST_PTR(&data)); | |
452e4751 | 2586 | return data.err; |
e22a25c9 AL |
2587 | } |
2588 | ||
62278814 | 2589 | int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr, |
e22a25c9 AL |
2590 | target_ulong len, int type) |
2591 | { | |
2592 | struct kvm_sw_breakpoint *bp; | |
e22a25c9 AL |
2593 | int err; |
2594 | ||
2595 | if (type == GDB_BREAKPOINT_SW) { | |
80b7cd73 | 2596 | bp = kvm_find_sw_breakpoint(cpu, addr); |
e22a25c9 AL |
2597 | if (bp) { |
2598 | bp->use_count++; | |
2599 | return 0; | |
2600 | } | |
2601 | ||
7267c094 | 2602 | bp = g_malloc(sizeof(struct kvm_sw_breakpoint)); |
e22a25c9 AL |
2603 | bp->pc = addr; |
2604 | bp->use_count = 1; | |
80b7cd73 | 2605 | err = kvm_arch_insert_sw_breakpoint(cpu, bp); |
e22a25c9 | 2606 | if (err) { |
7267c094 | 2607 | g_free(bp); |
e22a25c9 AL |
2608 | return err; |
2609 | } | |
2610 | ||
80b7cd73 | 2611 | QTAILQ_INSERT_HEAD(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry); |
e22a25c9 AL |
2612 | } else { |
2613 | err = kvm_arch_insert_hw_breakpoint(addr, len, type); | |
a426e122 | 2614 | if (err) { |
e22a25c9 | 2615 | return err; |
a426e122 | 2616 | } |
e22a25c9 AL |
2617 | } |
2618 | ||
bdc44640 | 2619 | CPU_FOREACH(cpu) { |
38e478ec | 2620 | err = kvm_update_guest_debug(cpu, 0); |
a426e122 | 2621 | if (err) { |
e22a25c9 | 2622 | return err; |
a426e122 | 2623 | } |
e22a25c9 AL |
2624 | } |
2625 | return 0; | |
2626 | } | |
2627 | ||
62278814 | 2628 | int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr, |
e22a25c9 AL |
2629 | target_ulong len, int type) |
2630 | { | |
2631 | struct kvm_sw_breakpoint *bp; | |
e22a25c9 AL |
2632 | int err; |
2633 | ||
2634 | if (type == GDB_BREAKPOINT_SW) { | |
80b7cd73 | 2635 | bp = kvm_find_sw_breakpoint(cpu, addr); |
a426e122 | 2636 | if (!bp) { |
e22a25c9 | 2637 | return -ENOENT; |
a426e122 | 2638 | } |
e22a25c9 AL |
2639 | |
2640 | if (bp->use_count > 1) { | |
2641 | bp->use_count--; | |
2642 | return 0; | |
2643 | } | |
2644 | ||
80b7cd73 | 2645 | err = kvm_arch_remove_sw_breakpoint(cpu, bp); |
a426e122 | 2646 | if (err) { |
e22a25c9 | 2647 | return err; |
a426e122 | 2648 | } |
e22a25c9 | 2649 | |
80b7cd73 | 2650 | QTAILQ_REMOVE(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry); |
7267c094 | 2651 | g_free(bp); |
e22a25c9 AL |
2652 | } else { |
2653 | err = kvm_arch_remove_hw_breakpoint(addr, len, type); | |
a426e122 | 2654 | if (err) { |
e22a25c9 | 2655 | return err; |
a426e122 | 2656 | } |
e22a25c9 AL |
2657 | } |
2658 | ||
bdc44640 | 2659 | CPU_FOREACH(cpu) { |
38e478ec | 2660 | err = kvm_update_guest_debug(cpu, 0); |
a426e122 | 2661 | if (err) { |
e22a25c9 | 2662 | return err; |
a426e122 | 2663 | } |
e22a25c9 AL |
2664 | } |
2665 | return 0; | |
2666 | } | |
2667 | ||
1d5791f4 | 2668 | void kvm_remove_all_breakpoints(CPUState *cpu) |
e22a25c9 AL |
2669 | { |
2670 | struct kvm_sw_breakpoint *bp, *next; | |
80b7cd73 | 2671 | KVMState *s = cpu->kvm_state; |
dc54e252 | 2672 | CPUState *tmpcpu; |
e22a25c9 | 2673 | |
72cf2d4f | 2674 | QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) { |
80b7cd73 | 2675 | if (kvm_arch_remove_sw_breakpoint(cpu, bp) != 0) { |
e22a25c9 | 2676 | /* Try harder to find a CPU that currently sees the breakpoint. */ |
dc54e252 CG |
2677 | CPU_FOREACH(tmpcpu) { |
2678 | if (kvm_arch_remove_sw_breakpoint(tmpcpu, bp) == 0) { | |
e22a25c9 | 2679 | break; |
a426e122 | 2680 | } |
e22a25c9 AL |
2681 | } |
2682 | } | |
78021d6d JK |
2683 | QTAILQ_REMOVE(&s->kvm_sw_breakpoints, bp, entry); |
2684 | g_free(bp); | |
e22a25c9 AL |
2685 | } |
2686 | kvm_arch_remove_all_hw_breakpoints(); | |
2687 | ||
bdc44640 | 2688 | CPU_FOREACH(cpu) { |
38e478ec | 2689 | kvm_update_guest_debug(cpu, 0); |
a426e122 | 2690 | } |
e22a25c9 AL |
2691 | } |
2692 | ||
2693 | #else /* !KVM_CAP_SET_GUEST_DEBUG */ | |
2694 | ||
38e478ec | 2695 | int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap) |
e22a25c9 AL |
2696 | { |
2697 | return -EINVAL; | |
2698 | } | |
2699 | ||
62278814 | 2700 | int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr, |
e22a25c9 AL |
2701 | target_ulong len, int type) |
2702 | { | |
2703 | return -EINVAL; | |
2704 | } | |
2705 | ||
62278814 | 2706 | int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr, |
e22a25c9 AL |
2707 | target_ulong len, int type) |
2708 | { | |
2709 | return -EINVAL; | |
2710 | } | |
2711 | ||
1d5791f4 | 2712 | void kvm_remove_all_breakpoints(CPUState *cpu) |
e22a25c9 AL |
2713 | { |
2714 | } | |
2715 | #endif /* !KVM_CAP_SET_GUEST_DEBUG */ | |
cc84de95 | 2716 | |
18268b60 | 2717 | static int kvm_set_signal_mask(CPUState *cpu, const sigset_t *sigset) |
cc84de95 | 2718 | { |
aed6efb9 | 2719 | KVMState *s = kvm_state; |
cc84de95 MT |
2720 | struct kvm_signal_mask *sigmask; |
2721 | int r; | |
2722 | ||
7267c094 | 2723 | sigmask = g_malloc(sizeof(*sigmask) + sizeof(*sigset)); |
cc84de95 | 2724 | |
aed6efb9 | 2725 | sigmask->len = s->sigmask_len; |
cc84de95 | 2726 | memcpy(sigmask->sigset, sigset, sizeof(*sigset)); |
1bc22652 | 2727 | r = kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, sigmask); |
7267c094 | 2728 | g_free(sigmask); |
cc84de95 MT |
2729 | |
2730 | return r; | |
2731 | } | |
4d39892c | 2732 | |
cf0f7cf9 | 2733 | static void kvm_ipi_signal(int sig) |
18268b60 | 2734 | { |
cf0f7cf9 PB |
2735 | if (current_cpu) { |
2736 | assert(kvm_immediate_exit); | |
2737 | kvm_cpu_kick(current_cpu); | |
2738 | } | |
18268b60 PB |
2739 | } |
2740 | ||
2741 | void kvm_init_cpu_signals(CPUState *cpu) | |
2742 | { | |
2743 | int r; | |
2744 | sigset_t set; | |
2745 | struct sigaction sigact; | |
2746 | ||
2747 | memset(&sigact, 0, sizeof(sigact)); | |
cf0f7cf9 | 2748 | sigact.sa_handler = kvm_ipi_signal; |
18268b60 PB |
2749 | sigaction(SIG_IPI, &sigact, NULL); |
2750 | ||
2751 | pthread_sigmask(SIG_BLOCK, NULL, &set); | |
2752 | #if defined KVM_HAVE_MCE_INJECTION | |
2753 | sigdelset(&set, SIGBUS); | |
2754 | pthread_sigmask(SIG_SETMASK, &set, NULL); | |
2755 | #endif | |
2756 | sigdelset(&set, SIG_IPI); | |
cf0f7cf9 PB |
2757 | if (kvm_immediate_exit) { |
2758 | r = pthread_sigmask(SIG_SETMASK, &set, NULL); | |
2759 | } else { | |
2760 | r = kvm_set_signal_mask(cpu, &set); | |
2761 | } | |
18268b60 PB |
2762 | if (r) { |
2763 | fprintf(stderr, "kvm_set_signal_mask: %s\n", strerror(-r)); | |
2764 | exit(1); | |
2765 | } | |
2766 | } | |
2767 | ||
2ae41db2 | 2768 | /* Called asynchronously in VCPU thread. */ |
290adf38 | 2769 | int kvm_on_sigbus_vcpu(CPUState *cpu, int code, void *addr) |
a1b87fe0 | 2770 | { |
2ae41db2 PB |
2771 | #ifdef KVM_HAVE_MCE_INJECTION |
2772 | if (have_sigbus_pending) { | |
2773 | return 1; | |
2774 | } | |
2775 | have_sigbus_pending = true; | |
2776 | pending_sigbus_addr = addr; | |
2777 | pending_sigbus_code = code; | |
2778 | atomic_set(&cpu->exit_request, 1); | |
2779 | return 0; | |
2780 | #else | |
2781 | return 1; | |
2782 | #endif | |
a1b87fe0 JK |
2783 | } |
2784 | ||
2ae41db2 | 2785 | /* Called synchronously (via signalfd) in main thread. */ |
a1b87fe0 JK |
2786 | int kvm_on_sigbus(int code, void *addr) |
2787 | { | |
2ae41db2 | 2788 | #ifdef KVM_HAVE_MCE_INJECTION |
4d39892c PB |
2789 | /* Action required MCE kills the process if SIGBUS is blocked. Because |
2790 | * that's what happens in the I/O thread, where we handle MCE via signalfd, | |
2791 | * we can only get action optional here. | |
2792 | */ | |
2793 | assert(code != BUS_MCEERR_AR); | |
2794 | kvm_arch_on_sigbus_vcpu(first_cpu, code, addr); | |
2795 | return 0; | |
2ae41db2 PB |
2796 | #else |
2797 | return 1; | |
2798 | #endif | |
a1b87fe0 | 2799 | } |
0a6a7cca CD |
2800 | |
2801 | int kvm_create_device(KVMState *s, uint64_t type, bool test) | |
2802 | { | |
2803 | int ret; | |
2804 | struct kvm_create_device create_dev; | |
2805 | ||
2806 | create_dev.type = type; | |
2807 | create_dev.fd = -1; | |
2808 | create_dev.flags = test ? KVM_CREATE_DEVICE_TEST : 0; | |
2809 | ||
2810 | if (!kvm_check_extension(s, KVM_CAP_DEVICE_CTRL)) { | |
2811 | return -ENOTSUP; | |
2812 | } | |
2813 | ||
2814 | ret = kvm_vm_ioctl(s, KVM_CREATE_DEVICE, &create_dev); | |
2815 | if (ret) { | |
2816 | return ret; | |
2817 | } | |
2818 | ||
2819 | return test ? 0 : create_dev.fd; | |
2820 | } | |
ada4135f | 2821 | |
29039acf PX |
2822 | bool kvm_device_supported(int vmfd, uint64_t type) |
2823 | { | |
2824 | struct kvm_create_device create_dev = { | |
2825 | .type = type, | |
2826 | .fd = -1, | |
2827 | .flags = KVM_CREATE_DEVICE_TEST, | |
2828 | }; | |
2829 | ||
2830 | if (ioctl(vmfd, KVM_CHECK_EXTENSION, KVM_CAP_DEVICE_CTRL) <= 0) { | |
2831 | return false; | |
2832 | } | |
2833 | ||
2834 | return (ioctl(vmfd, KVM_CREATE_DEVICE, &create_dev) >= 0); | |
2835 | } | |
2836 | ||
ada4135f CH |
2837 | int kvm_set_one_reg(CPUState *cs, uint64_t id, void *source) |
2838 | { | |
2839 | struct kvm_one_reg reg; | |
2840 | int r; | |
2841 | ||
2842 | reg.id = id; | |
2843 | reg.addr = (uintptr_t) source; | |
2844 | r = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®); | |
2845 | if (r) { | |
844a3d34 | 2846 | trace_kvm_failed_reg_set(id, strerror(-r)); |
ada4135f CH |
2847 | } |
2848 | return r; | |
2849 | } | |
2850 | ||
2851 | int kvm_get_one_reg(CPUState *cs, uint64_t id, void *target) | |
2852 | { | |
2853 | struct kvm_one_reg reg; | |
2854 | int r; | |
2855 | ||
2856 | reg.id = id; | |
2857 | reg.addr = (uintptr_t) target; | |
2858 | r = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®); | |
2859 | if (r) { | |
844a3d34 | 2860 | trace_kvm_failed_reg_get(id, strerror(-r)); |
ada4135f CH |
2861 | } |
2862 | return r; | |
2863 | } | |
782c3f29 | 2864 | |
8072aae3 AK |
2865 | static bool kvm_accel_has_memory(MachineState *ms, AddressSpace *as, |
2866 | hwaddr start_addr, hwaddr size) | |
2867 | { | |
2868 | KVMState *kvm = KVM_STATE(ms->accelerator); | |
2869 | int i; | |
2870 | ||
2871 | for (i = 0; i < kvm->nr_as; ++i) { | |
2872 | if (kvm->as[i].as == as && kvm->as[i].ml) { | |
2873 | return NULL != kvm_lookup_matching_slot(kvm->as[i].ml, | |
2874 | start_addr, size); | |
2875 | } | |
2876 | } | |
2877 | ||
2878 | return false; | |
2879 | } | |
2880 | ||
782c3f29 EH |
2881 | static void kvm_accel_class_init(ObjectClass *oc, void *data) |
2882 | { | |
2883 | AccelClass *ac = ACCEL_CLASS(oc); | |
2884 | ac->name = "KVM"; | |
0d15da8e | 2885 | ac->init_machine = kvm_init; |
8072aae3 | 2886 | ac->has_memory = kvm_accel_has_memory; |
782c3f29 EH |
2887 | ac->allowed = &kvm_allowed; |
2888 | } | |
2889 | ||
2890 | static const TypeInfo kvm_accel_type = { | |
2891 | .name = TYPE_KVM_ACCEL, | |
2892 | .parent = TYPE_ACCEL, | |
2893 | .class_init = kvm_accel_class_init, | |
fc02086b | 2894 | .instance_size = sizeof(KVMState), |
782c3f29 EH |
2895 | }; |
2896 | ||
2897 | static void kvm_type_init(void) | |
2898 | { | |
2899 | type_register_static(&kvm_accel_type); | |
2900 | } | |
2901 | ||
2902 | type_init(kvm_type_init); |