]>
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> |
b4420f19 | 18 | #include <poll.h> |
05330448 AL |
19 | |
20 | #include <linux/kvm.h> | |
21 | ||
1de7afc9 PB |
22 | #include "qemu/atomic.h" |
23 | #include "qemu/option.h" | |
24 | #include "qemu/config-file.h" | |
4b3cfe72 | 25 | #include "qemu/error-report.h" |
556969e9 | 26 | #include "qapi/error.h" |
a2cb15b0 | 27 | #include "hw/pci/msi.h" |
d1f6af6a | 28 | #include "hw/pci/msix.h" |
d426d9fb | 29 | #include "hw/s390x/adapter.h" |
022c62cb | 30 | #include "exec/gdbstub.h" |
8571ed35 | 31 | #include "sysemu/kvm_int.h" |
54d31236 | 32 | #include "sysemu/runstate.h" |
d2528bdc | 33 | #include "sysemu/cpus.h" |
f39b7d2b | 34 | #include "sysemu/accel-blocker.h" |
1de7afc9 | 35 | #include "qemu/bswap.h" |
022c62cb | 36 | #include "exec/memory.h" |
747afd5b | 37 | #include "exec/ram_addr.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" |
23b0898e | 42 | #include "qapi/visitor.h" |
11bc4a13 PB |
43 | #include "qapi/qapi-types-common.h" |
44 | #include "qapi/qapi-visit-common.h" | |
6b552b9b | 45 | #include "sysemu/reset.h" |
57038a92 CF |
46 | #include "qemu/guest-random.h" |
47 | #include "sysemu/hw_accel.h" | |
48 | #include "kvm-cpus.h" | |
baa60983 | 49 | #include "sysemu/dirtylimit.h" |
f39b7d2b | 50 | #include "qemu/range.h" |
05330448 | 51 | |
135a129a | 52 | #include "hw/boards.h" |
aa09b3d5 | 53 | #include "sysemu/stats.h" |
135a129a | 54 | |
d2f2b8a7 SH |
55 | /* This check must be after config-host.h is included */ |
56 | #ifdef CONFIG_EVENTFD | |
57 | #include <sys/eventfd.h> | |
58 | #endif | |
59 | ||
bc92e4e9 AJ |
60 | /* KVM uses PAGE_SIZE in its definition of KVM_COALESCED_MMIO_MAX. We |
61 | * need to use the real host PAGE_SIZE, as that's what KVM will use. | |
62 | */ | |
eb8b1a79 JY |
63 | #ifdef PAGE_SIZE |
64 | #undef PAGE_SIZE | |
65 | #endif | |
8e3b0cbb | 66 | #define PAGE_SIZE qemu_real_host_page_size() |
f65ed4c1 | 67 | |
fd2ddd16 ML |
68 | #ifndef KVM_GUESTDBG_BLOCKIRQ |
69 | #define KVM_GUESTDBG_BLOCKIRQ 0 | |
70 | #endif | |
71 | ||
05330448 AL |
72 | //#define DEBUG_KVM |
73 | ||
74 | #ifdef DEBUG_KVM | |
8c0d577e | 75 | #define DPRINTF(fmt, ...) \ |
05330448 AL |
76 | do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0) |
77 | #else | |
8c0d577e | 78 | #define DPRINTF(fmt, ...) \ |
05330448 AL |
79 | do { } while (0) |
80 | #endif | |
81 | ||
4c055ab5 GZ |
82 | struct KVMParkedVcpu { |
83 | unsigned long vcpu_id; | |
84 | int kvm_fd; | |
85 | QLIST_ENTRY(KVMParkedVcpu) node; | |
86 | }; | |
87 | ||
6a7af8cb | 88 | KVMState *kvm_state; |
3d4b2649 | 89 | bool kvm_kernel_irqchip; |
15eafc2e | 90 | bool kvm_split_irqchip; |
7ae26bd4 | 91 | bool kvm_async_interrupts_allowed; |
215e79c0 | 92 | bool kvm_halt_in_kernel_allowed; |
69e03ae6 | 93 | bool kvm_eventfds_allowed; |
cc7e0ddf | 94 | bool kvm_irqfds_allowed; |
f41389ae | 95 | bool kvm_resamplefds_allowed; |
614e41bc | 96 | bool kvm_msi_via_irqfd_allowed; |
f3e1bed8 | 97 | bool kvm_gsi_routing_allowed; |
76fe21de | 98 | bool kvm_gsi_direct_mapping; |
13eed94e | 99 | bool kvm_allowed; |
df9c8b75 | 100 | bool kvm_readonly_mem_allowed; |
d0a073a1 | 101 | bool kvm_vm_attributes_allowed; |
50bf31b9 | 102 | bool kvm_direct_msi_allowed; |
35108223 | 103 | bool kvm_ioeventfd_any_length_allowed; |
767a554a | 104 | bool kvm_msi_use_devid; |
12bc5b4c | 105 | bool kvm_has_guest_debug; |
3b7a9388 | 106 | static int kvm_sstep_flags; |
cf0f7cf9 | 107 | static bool kvm_immediate_exit; |
023ae9a8 | 108 | static hwaddr kvm_max_slot_size = ~0; |
05330448 | 109 | |
94a8d39a JK |
110 | static const KVMCapabilityInfo kvm_required_capabilites[] = { |
111 | KVM_CAP_INFO(USER_MEMORY), | |
112 | KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS), | |
89de4b91 | 113 | KVM_CAP_INFO(JOIN_MEMORY_REGIONS_WORKS), |
94a8d39a JK |
114 | KVM_CAP_LAST_INFO |
115 | }; | |
116 | ||
3607715a DG |
117 | static NotifierList kvm_irqchip_change_notifiers = |
118 | NOTIFIER_LIST_INITIALIZER(kvm_irqchip_change_notifiers); | |
119 | ||
c82d9d43 PX |
120 | struct KVMResampleFd { |
121 | int gsi; | |
122 | EventNotifier *resample_event; | |
123 | QLIST_ENTRY(KVMResampleFd) node; | |
124 | }; | |
125 | typedef struct KVMResampleFd KVMResampleFd; | |
126 | ||
127 | /* | |
128 | * Only used with split irqchip where we need to do the resample fd | |
129 | * kick for the kernel from userspace. | |
130 | */ | |
131 | static QLIST_HEAD(, KVMResampleFd) kvm_resample_fd_list = | |
132 | QLIST_HEAD_INITIALIZER(kvm_resample_fd_list); | |
133 | ||
a2f77862 PX |
134 | static QemuMutex kml_slots_lock; |
135 | ||
136 | #define kvm_slots_lock() qemu_mutex_lock(&kml_slots_lock) | |
137 | #define kvm_slots_unlock() qemu_mutex_unlock(&kml_slots_lock) | |
36adac49 | 138 | |
ea776d15 PX |
139 | static void kvm_slot_init_dirty_bitmap(KVMSlot *mem); |
140 | ||
c82d9d43 PX |
141 | static inline void kvm_resample_fd_remove(int gsi) |
142 | { | |
143 | KVMResampleFd *rfd; | |
144 | ||
145 | QLIST_FOREACH(rfd, &kvm_resample_fd_list, node) { | |
146 | if (rfd->gsi == gsi) { | |
147 | QLIST_REMOVE(rfd, node); | |
148 | g_free(rfd); | |
149 | break; | |
150 | } | |
151 | } | |
152 | } | |
153 | ||
154 | static inline void kvm_resample_fd_insert(int gsi, EventNotifier *event) | |
155 | { | |
156 | KVMResampleFd *rfd = g_new0(KVMResampleFd, 1); | |
157 | ||
158 | rfd->gsi = gsi; | |
159 | rfd->resample_event = event; | |
160 | ||
161 | QLIST_INSERT_HEAD(&kvm_resample_fd_list, rfd, node); | |
162 | } | |
163 | ||
164 | void kvm_resample_fd_notify(int gsi) | |
165 | { | |
166 | KVMResampleFd *rfd; | |
167 | ||
168 | QLIST_FOREACH(rfd, &kvm_resample_fd_list, node) { | |
169 | if (rfd->gsi == gsi) { | |
170 | event_notifier_set(rfd->resample_event); | |
171 | trace_kvm_resample_fd_notify(gsi); | |
172 | return; | |
173 | } | |
174 | } | |
175 | } | |
176 | ||
44f2e6c1 BR |
177 | int kvm_get_max_memslots(void) |
178 | { | |
4f7f5893 | 179 | KVMState *s = KVM_STATE(current_accel()); |
44f2e6c1 BR |
180 | |
181 | return s->nr_slots; | |
182 | } | |
183 | ||
36adac49 | 184 | /* Called with KVMMemoryListener.slots_lock held */ |
7bbda04c | 185 | static KVMSlot *kvm_get_free_slot(KVMMemoryListener *kml) |
05330448 | 186 | { |
7bbda04c | 187 | KVMState *s = kvm_state; |
05330448 AL |
188 | int i; |
189 | ||
fb541ca5 | 190 | for (i = 0; i < s->nr_slots; i++) { |
7bbda04c PB |
191 | if (kml->slots[i].memory_size == 0) { |
192 | return &kml->slots[i]; | |
a426e122 | 193 | } |
05330448 AL |
194 | } |
195 | ||
b8865591 IM |
196 | return NULL; |
197 | } | |
198 | ||
199 | bool kvm_has_free_slot(MachineState *ms) | |
200 | { | |
7bbda04c | 201 | KVMState *s = KVM_STATE(ms->accelerator); |
36adac49 PX |
202 | bool result; |
203 | KVMMemoryListener *kml = &s->memory_listener; | |
204 | ||
a2f77862 | 205 | kvm_slots_lock(); |
36adac49 | 206 | result = !!kvm_get_free_slot(kml); |
a2f77862 | 207 | kvm_slots_unlock(); |
7bbda04c | 208 | |
36adac49 | 209 | return result; |
b8865591 IM |
210 | } |
211 | ||
36adac49 | 212 | /* Called with KVMMemoryListener.slots_lock held */ |
7bbda04c | 213 | static KVMSlot *kvm_alloc_slot(KVMMemoryListener *kml) |
b8865591 | 214 | { |
7bbda04c | 215 | KVMSlot *slot = kvm_get_free_slot(kml); |
b8865591 IM |
216 | |
217 | if (slot) { | |
218 | return slot; | |
219 | } | |
220 | ||
d3f8d37f AL |
221 | fprintf(stderr, "%s: no free slot available\n", __func__); |
222 | abort(); | |
223 | } | |
224 | ||
7bbda04c | 225 | static KVMSlot *kvm_lookup_matching_slot(KVMMemoryListener *kml, |
a8170e5e | 226 | hwaddr start_addr, |
2747e716 | 227 | hwaddr size) |
d3f8d37f | 228 | { |
7bbda04c | 229 | KVMState *s = kvm_state; |
d3f8d37f AL |
230 | int i; |
231 | ||
fb541ca5 | 232 | for (i = 0; i < s->nr_slots; i++) { |
7bbda04c | 233 | KVMSlot *mem = &kml->slots[i]; |
d3f8d37f | 234 | |
2747e716 | 235 | if (start_addr == mem->start_addr && size == mem->memory_size) { |
d3f8d37f AL |
236 | return mem; |
237 | } | |
238 | } | |
239 | ||
05330448 AL |
240 | return NULL; |
241 | } | |
242 | ||
5ea69c2e DH |
243 | /* |
244 | * Calculate and align the start address and the size of the section. | |
245 | * Return the size. If the size is 0, the aligned section is empty. | |
246 | */ | |
247 | static hwaddr kvm_align_section(MemoryRegionSection *section, | |
248 | hwaddr *start) | |
249 | { | |
250 | hwaddr size = int128_get64(section->size); | |
a6ffc423 | 251 | hwaddr delta, aligned; |
5ea69c2e DH |
252 | |
253 | /* kvm works in page size chunks, but the function may be called | |
254 | with sub-page size and unaligned start address. Pad the start | |
255 | address to next and truncate size to previous page boundary. */ | |
a6ffc423 | 256 | aligned = ROUND_UP(section->offset_within_address_space, |
8e3b0cbb | 257 | qemu_real_host_page_size()); |
a6ffc423 DH |
258 | delta = aligned - section->offset_within_address_space; |
259 | *start = aligned; | |
5ea69c2e DH |
260 | if (delta > size) { |
261 | return 0; | |
262 | } | |
5ea69c2e | 263 | |
8e3b0cbb | 264 | return (size - delta) & qemu_real_host_page_mask(); |
5ea69c2e DH |
265 | } |
266 | ||
9f213ed9 | 267 | int kvm_physical_memory_addr_from_host(KVMState *s, void *ram, |
a8170e5e | 268 | hwaddr *phys_addr) |
983dfc3b | 269 | { |
7bbda04c | 270 | KVMMemoryListener *kml = &s->memory_listener; |
36adac49 | 271 | int i, ret = 0; |
983dfc3b | 272 | |
a2f77862 | 273 | kvm_slots_lock(); |
fb541ca5 | 274 | for (i = 0; i < s->nr_slots; i++) { |
7bbda04c | 275 | KVMSlot *mem = &kml->slots[i]; |
983dfc3b | 276 | |
9f213ed9 AK |
277 | if (ram >= mem->ram && ram < mem->ram + mem->memory_size) { |
278 | *phys_addr = mem->start_addr + (ram - mem->ram); | |
36adac49 PX |
279 | ret = 1; |
280 | break; | |
983dfc3b HY |
281 | } |
282 | } | |
a2f77862 | 283 | kvm_slots_unlock(); |
983dfc3b | 284 | |
36adac49 | 285 | return ret; |
983dfc3b HY |
286 | } |
287 | ||
6c090d4a | 288 | static int kvm_set_user_memory_region(KVMMemoryListener *kml, KVMSlot *slot, bool new) |
5832d1f2 | 289 | { |
7bbda04c | 290 | KVMState *s = kvm_state; |
5832d1f2 | 291 | struct kvm_userspace_memory_region mem; |
fe29141b | 292 | int ret; |
5832d1f2 | 293 | |
38bfe691 | 294 | mem.slot = slot->slot | (kml->as_id << 16); |
5832d1f2 | 295 | mem.guest_phys_addr = slot->start_addr; |
9f213ed9 | 296 | mem.userspace_addr = (unsigned long)slot->ram; |
5832d1f2 | 297 | mem.flags = slot->flags; |
651eb0f4 | 298 | |
6c090d4a | 299 | if (slot->memory_size && !new && (mem.flags ^ slot->old_flags) & KVM_MEM_READONLY) { |
235e8982 JJ |
300 | /* Set the slot size to 0 before setting the slot to the desired |
301 | * value. This is needed based on KVM commit 75d61fbc. */ | |
302 | mem.memory_size = 0; | |
88cd34ee PMD |
303 | ret = kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem); |
304 | if (ret < 0) { | |
305 | goto err; | |
306 | } | |
235e8982 JJ |
307 | } |
308 | mem.memory_size = slot->memory_size; | |
fe29141b | 309 | ret = kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem); |
6c090d4a | 310 | slot->old_flags = mem.flags; |
88cd34ee | 311 | err: |
fe29141b AK |
312 | trace_kvm_set_user_memory(mem.slot, mem.flags, mem.guest_phys_addr, |
313 | mem.memory_size, mem.userspace_addr, ret); | |
88cd34ee PMD |
314 | if (ret < 0) { |
315 | error_report("%s: KVM_SET_USER_MEMORY_REGION failed, slot=%d," | |
316 | " start=0x%" PRIx64 ", size=0x%" PRIx64 ": %s", | |
317 | __func__, mem.slot, slot->start_addr, | |
318 | (uint64_t)mem.memory_size, strerror(errno)); | |
319 | } | |
fe29141b | 320 | return ret; |
5832d1f2 AL |
321 | } |
322 | ||
57038a92 | 323 | static int do_kvm_destroy_vcpu(CPUState *cpu) |
4c055ab5 GZ |
324 | { |
325 | KVMState *s = kvm_state; | |
326 | long mmap_size; | |
327 | struct KVMParkedVcpu *vcpu = NULL; | |
328 | int ret = 0; | |
329 | ||
330 | DPRINTF("kvm_destroy_vcpu\n"); | |
331 | ||
b1115c99 LA |
332 | ret = kvm_arch_destroy_vcpu(cpu); |
333 | if (ret < 0) { | |
334 | goto err; | |
335 | } | |
336 | ||
4c055ab5 GZ |
337 | mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0); |
338 | if (mmap_size < 0) { | |
339 | ret = mmap_size; | |
340 | DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n"); | |
341 | goto err; | |
342 | } | |
343 | ||
344 | ret = munmap(cpu->kvm_run, mmap_size); | |
345 | if (ret < 0) { | |
346 | goto err; | |
347 | } | |
348 | ||
b4420f19 | 349 | if (cpu->kvm_dirty_gfns) { |
dcafa248 | 350 | ret = munmap(cpu->kvm_dirty_gfns, s->kvm_dirty_ring_bytes); |
b4420f19 PX |
351 | if (ret < 0) { |
352 | goto err; | |
353 | } | |
354 | } | |
355 | ||
4c055ab5 GZ |
356 | vcpu = g_malloc0(sizeof(*vcpu)); |
357 | vcpu->vcpu_id = kvm_arch_vcpu_id(cpu); | |
358 | vcpu->kvm_fd = cpu->kvm_fd; | |
359 | QLIST_INSERT_HEAD(&kvm_state->kvm_parked_vcpus, vcpu, node); | |
360 | err: | |
361 | return ret; | |
362 | } | |
363 | ||
57038a92 CF |
364 | void kvm_destroy_vcpu(CPUState *cpu) |
365 | { | |
366 | if (do_kvm_destroy_vcpu(cpu) < 0) { | |
367 | error_report("kvm_destroy_vcpu failed"); | |
368 | exit(EXIT_FAILURE); | |
369 | } | |
370 | } | |
371 | ||
4c055ab5 GZ |
372 | static int kvm_get_vcpu(KVMState *s, unsigned long vcpu_id) |
373 | { | |
374 | struct KVMParkedVcpu *cpu; | |
375 | ||
376 | QLIST_FOREACH(cpu, &s->kvm_parked_vcpus, node) { | |
377 | if (cpu->vcpu_id == vcpu_id) { | |
378 | int kvm_fd; | |
379 | ||
380 | QLIST_REMOVE(cpu, node); | |
381 | kvm_fd = cpu->kvm_fd; | |
382 | g_free(cpu); | |
383 | return kvm_fd; | |
384 | } | |
385 | } | |
386 | ||
387 | return kvm_vm_ioctl(s, KVM_CREATE_VCPU, (void *)vcpu_id); | |
388 | } | |
389 | ||
d0a92b35 | 390 | int kvm_init_vcpu(CPUState *cpu, Error **errp) |
05330448 AL |
391 | { |
392 | KVMState *s = kvm_state; | |
393 | long mmap_size; | |
394 | int ret; | |
395 | ||
d0a92b35 | 396 | trace_kvm_init_vcpu(cpu->cpu_index, kvm_arch_vcpu_id(cpu)); |
05330448 | 397 | |
4c055ab5 | 398 | ret = kvm_get_vcpu(s, kvm_arch_vcpu_id(cpu)); |
05330448 | 399 | if (ret < 0) { |
d0a92b35 DDAG |
400 | error_setg_errno(errp, -ret, "kvm_init_vcpu: kvm_get_vcpu failed (%lu)", |
401 | kvm_arch_vcpu_id(cpu)); | |
05330448 AL |
402 | goto err; |
403 | } | |
404 | ||
8737c51c | 405 | cpu->kvm_fd = ret; |
a60f24b5 | 406 | cpu->kvm_state = s; |
99f31832 | 407 | cpu->vcpu_dirty = true; |
7786ae40 | 408 | cpu->dirty_pages = 0; |
baa60983 | 409 | cpu->throttle_us_per_full = 0; |
05330448 AL |
410 | |
411 | mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0); | |
412 | if (mmap_size < 0) { | |
748a680b | 413 | ret = mmap_size; |
d0a92b35 DDAG |
414 | error_setg_errno(errp, -mmap_size, |
415 | "kvm_init_vcpu: KVM_GET_VCPU_MMAP_SIZE failed"); | |
05330448 AL |
416 | goto err; |
417 | } | |
418 | ||
f7575c96 | 419 | cpu->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED, |
8737c51c | 420 | cpu->kvm_fd, 0); |
f7575c96 | 421 | if (cpu->kvm_run == MAP_FAILED) { |
05330448 | 422 | ret = -errno; |
d0a92b35 DDAG |
423 | error_setg_errno(errp, ret, |
424 | "kvm_init_vcpu: mmap'ing vcpu state failed (%lu)", | |
425 | kvm_arch_vcpu_id(cpu)); | |
05330448 AL |
426 | goto err; |
427 | } | |
428 | ||
a426e122 JK |
429 | if (s->coalesced_mmio && !s->coalesced_mmio_ring) { |
430 | s->coalesced_mmio_ring = | |
f7575c96 | 431 | (void *)cpu->kvm_run + s->coalesced_mmio * PAGE_SIZE; |
a426e122 | 432 | } |
62a2744c | 433 | |
b4420f19 PX |
434 | if (s->kvm_dirty_ring_size) { |
435 | /* Use MAP_SHARED to share pages with the kernel */ | |
dcafa248 | 436 | cpu->kvm_dirty_gfns = mmap(NULL, s->kvm_dirty_ring_bytes, |
b4420f19 PX |
437 | PROT_READ | PROT_WRITE, MAP_SHARED, |
438 | cpu->kvm_fd, | |
439 | PAGE_SIZE * KVM_DIRTY_LOG_PAGE_OFFSET); | |
440 | if (cpu->kvm_dirty_gfns == MAP_FAILED) { | |
441 | ret = -errno; | |
442 | DPRINTF("mmap'ing vcpu dirty gfns failed: %d\n", ret); | |
443 | goto err; | |
444 | } | |
445 | } | |
446 | ||
20d695a9 | 447 | ret = kvm_arch_init_vcpu(cpu); |
d0a92b35 DDAG |
448 | if (ret < 0) { |
449 | error_setg_errno(errp, -ret, | |
450 | "kvm_init_vcpu: kvm_arch_init_vcpu failed (%lu)", | |
451 | kvm_arch_vcpu_id(cpu)); | |
452 | } | |
3b6f4852 MT |
453 | cpu->kvm_vcpu_stats_fd = kvm_vcpu_ioctl(cpu, KVM_GET_STATS_FD, NULL); |
454 | ||
05330448 AL |
455 | err: |
456 | return ret; | |
457 | } | |
458 | ||
5832d1f2 AL |
459 | /* |
460 | * dirty pages logging control | |
461 | */ | |
25254bbc | 462 | |
d6ff5cbc | 463 | static int kvm_mem_flags(MemoryRegion *mr) |
25254bbc | 464 | { |
d6ff5cbc | 465 | bool readonly = mr->readonly || memory_region_is_romd(mr); |
235e8982 | 466 | int flags = 0; |
d6ff5cbc AJ |
467 | |
468 | if (memory_region_get_dirty_log_mask(mr) != 0) { | |
469 | flags |= KVM_MEM_LOG_DIRTY_PAGES; | |
470 | } | |
235e8982 JJ |
471 | if (readonly && kvm_readonly_mem_allowed) { |
472 | flags |= KVM_MEM_READONLY; | |
473 | } | |
474 | return flags; | |
25254bbc MT |
475 | } |
476 | ||
36adac49 | 477 | /* Called with KVMMemoryListener.slots_lock held */ |
7bbda04c PB |
478 | static int kvm_slot_update_flags(KVMMemoryListener *kml, KVMSlot *mem, |
479 | MemoryRegion *mr) | |
5832d1f2 | 480 | { |
d6ff5cbc | 481 | mem->flags = kvm_mem_flags(mr); |
5832d1f2 | 482 | |
4495d6a7 | 483 | /* If nothing changed effectively, no need to issue ioctl */ |
6c090d4a | 484 | if (mem->flags == mem->old_flags) { |
25254bbc | 485 | return 0; |
4495d6a7 JK |
486 | } |
487 | ||
ea776d15 | 488 | kvm_slot_init_dirty_bitmap(mem); |
6c090d4a | 489 | return kvm_set_user_memory_region(kml, mem, false); |
5832d1f2 AL |
490 | } |
491 | ||
7bbda04c PB |
492 | static int kvm_section_update_flags(KVMMemoryListener *kml, |
493 | MemoryRegionSection *section) | |
25254bbc | 494 | { |
023ae9a8 | 495 | hwaddr start_addr, size, slot_size; |
343562e8 | 496 | KVMSlot *mem; |
36adac49 | 497 | int ret = 0; |
25254bbc | 498 | |
343562e8 DH |
499 | size = kvm_align_section(section, &start_addr); |
500 | if (!size) { | |
ea8cb1a8 | 501 | return 0; |
25254bbc | 502 | } |
343562e8 | 503 | |
a2f77862 | 504 | kvm_slots_lock(); |
36adac49 | 505 | |
023ae9a8 IM |
506 | while (size && !ret) { |
507 | slot_size = MIN(kvm_max_slot_size, size); | |
508 | mem = kvm_lookup_matching_slot(kml, start_addr, slot_size); | |
509 | if (!mem) { | |
510 | /* We don't have a slot if we want to trap every access. */ | |
511 | goto out; | |
512 | } | |
343562e8 | 513 | |
023ae9a8 IM |
514 | ret = kvm_slot_update_flags(kml, mem, section->mr); |
515 | start_addr += slot_size; | |
516 | size -= slot_size; | |
517 | } | |
36adac49 PX |
518 | |
519 | out: | |
a2f77862 | 520 | kvm_slots_unlock(); |
36adac49 | 521 | return ret; |
25254bbc MT |
522 | } |
523 | ||
a01672d3 | 524 | static void kvm_log_start(MemoryListener *listener, |
b2dfd71c PB |
525 | MemoryRegionSection *section, |
526 | int old, int new) | |
5832d1f2 | 527 | { |
7bbda04c | 528 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); |
a01672d3 AK |
529 | int r; |
530 | ||
b2dfd71c PB |
531 | if (old != 0) { |
532 | return; | |
533 | } | |
534 | ||
7bbda04c | 535 | r = kvm_section_update_flags(kml, section); |
a01672d3 AK |
536 | if (r < 0) { |
537 | abort(); | |
538 | } | |
5832d1f2 AL |
539 | } |
540 | ||
a01672d3 | 541 | static void kvm_log_stop(MemoryListener *listener, |
b2dfd71c PB |
542 | MemoryRegionSection *section, |
543 | int old, int new) | |
5832d1f2 | 544 | { |
7bbda04c | 545 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); |
a01672d3 AK |
546 | int r; |
547 | ||
b2dfd71c PB |
548 | if (new != 0) { |
549 | return; | |
550 | } | |
551 | ||
7bbda04c | 552 | r = kvm_section_update_flags(kml, section); |
a01672d3 AK |
553 | if (r < 0) { |
554 | abort(); | |
555 | } | |
5832d1f2 AL |
556 | } |
557 | ||
8369e01c | 558 | /* get kvm's dirty pages bitmap and update qemu's */ |
2c20b27e | 559 | static void kvm_slot_sync_dirty_pages(KVMSlot *slot) |
96c1606b | 560 | { |
2c20b27e | 561 | ram_addr_t start = slot->ram_start_offset; |
8e3b0cbb | 562 | ram_addr_t pages = slot->memory_size / qemu_real_host_page_size(); |
5ff7fb77 | 563 | |
2c20b27e | 564 | cpu_physical_memory_set_dirty_lebitmap(slot->dirty_bmap, start, pages); |
96c1606b AG |
565 | } |
566 | ||
b4420f19 PX |
567 | static void kvm_slot_reset_dirty_pages(KVMSlot *slot) |
568 | { | |
569 | memset(slot->dirty_bmap, 0, slot->dirty_bmap_size); | |
570 | } | |
571 | ||
8369e01c MT |
572 | #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1)) |
573 | ||
9b3a31c7 | 574 | /* Allocate the dirty bitmap for a slot */ |
ea776d15 | 575 | static void kvm_slot_init_dirty_bitmap(KVMSlot *mem) |
9b3a31c7 | 576 | { |
ea776d15 PX |
577 | if (!(mem->flags & KVM_MEM_LOG_DIRTY_PAGES) || mem->dirty_bmap) { |
578 | return; | |
579 | } | |
580 | ||
9b3a31c7 DDAG |
581 | /* |
582 | * XXX bad kernel interface alert | |
583 | * For dirty bitmap, kernel allocates array of size aligned to | |
584 | * bits-per-long. But for case when the kernel is 64bits and | |
585 | * the userspace is 32bits, userspace can't align to the same | |
586 | * bits-per-long, since sizeof(long) is different between kernel | |
587 | * and user space. This way, userspace will provide buffer which | |
588 | * may be 4 bytes less than the kernel will use, resulting in | |
589 | * userspace memory corruption (which is not detectable by valgrind | |
590 | * too, in most cases). | |
591 | * So for now, let's align to 64 instead of HOST_LONG_BITS here, in | |
592 | * a hope that sizeof(long) won't become >8 any time soon. | |
e0a8f993 KZ |
593 | * |
594 | * Note: the granule of kvm dirty log is qemu_real_host_page_size. | |
595 | * And mem->memory_size is aligned to it (otherwise this mem can't | |
596 | * be registered to KVM). | |
9b3a31c7 | 597 | */ |
8e3b0cbb | 598 | hwaddr bitmap_size = ALIGN(mem->memory_size / qemu_real_host_page_size(), |
9b3a31c7 DDAG |
599 | /*HOST_LONG_BITS*/ 64) / 8; |
600 | mem->dirty_bmap = g_malloc0(bitmap_size); | |
563d32ba | 601 | mem->dirty_bmap_size = bitmap_size; |
9b3a31c7 DDAG |
602 | } |
603 | ||
e65e5f50 PX |
604 | /* |
605 | * Sync dirty bitmap from kernel to KVMSlot.dirty_bmap, return true if | |
606 | * succeeded, false otherwise | |
607 | */ | |
608 | static bool kvm_slot_get_dirty_log(KVMState *s, KVMSlot *slot) | |
609 | { | |
610 | struct kvm_dirty_log d = {}; | |
611 | int ret; | |
612 | ||
613 | d.dirty_bitmap = slot->dirty_bmap; | |
614 | d.slot = slot->slot | (slot->as_id << 16); | |
615 | ret = kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d); | |
616 | ||
617 | if (ret == -ENOENT) { | |
618 | /* kernel does not have dirty bitmap in this slot */ | |
619 | ret = 0; | |
620 | } | |
621 | if (ret) { | |
622 | error_report_once("%s: KVM_GET_DIRTY_LOG failed with %d", | |
623 | __func__, ret); | |
624 | } | |
625 | return ret == 0; | |
626 | } | |
627 | ||
b4420f19 PX |
628 | /* Should be with all slots_lock held for the address spaces. */ |
629 | static void kvm_dirty_ring_mark_page(KVMState *s, uint32_t as_id, | |
630 | uint32_t slot_id, uint64_t offset) | |
631 | { | |
632 | KVMMemoryListener *kml; | |
633 | KVMSlot *mem; | |
634 | ||
635 | if (as_id >= s->nr_as) { | |
636 | return; | |
637 | } | |
638 | ||
639 | kml = s->as[as_id].ml; | |
640 | mem = &kml->slots[slot_id]; | |
641 | ||
642 | if (!mem->memory_size || offset >= | |
8e3b0cbb | 643 | (mem->memory_size / qemu_real_host_page_size())) { |
b4420f19 PX |
644 | return; |
645 | } | |
646 | ||
647 | set_bit(offset, mem->dirty_bmap); | |
648 | } | |
649 | ||
650 | static bool dirty_gfn_is_dirtied(struct kvm_dirty_gfn *gfn) | |
651 | { | |
4802bf91 PB |
652 | /* |
653 | * Read the flags before the value. Pairs with barrier in | |
654 | * KVM's kvm_dirty_ring_push() function. | |
655 | */ | |
656 | return qatomic_load_acquire(&gfn->flags) == KVM_DIRTY_GFN_F_DIRTY; | |
b4420f19 PX |
657 | } |
658 | ||
659 | static void dirty_gfn_set_collected(struct kvm_dirty_gfn *gfn) | |
660 | { | |
52281c6d PB |
661 | /* |
662 | * Use a store-release so that the CPU that executes KVM_RESET_DIRTY_RINGS | |
663 | * sees the full content of the ring: | |
664 | * | |
665 | * CPU0 CPU1 CPU2 | |
666 | * ------------------------------------------------------------------------------ | |
667 | * fill gfn0 | |
668 | * store-rel flags for gfn0 | |
669 | * load-acq flags for gfn0 | |
670 | * store-rel RESET for gfn0 | |
671 | * ioctl(RESET_RINGS) | |
672 | * load-acq flags for gfn0 | |
673 | * check if flags have RESET | |
674 | * | |
675 | * The synchronization goes from CPU2 to CPU0 to CPU1. | |
676 | */ | |
677 | qatomic_store_release(&gfn->flags, KVM_DIRTY_GFN_F_RESET); | |
b4420f19 PX |
678 | } |
679 | ||
680 | /* | |
681 | * Should be with all slots_lock held for the address spaces. It returns the | |
682 | * dirty page we've collected on this dirty ring. | |
683 | */ | |
684 | static uint32_t kvm_dirty_ring_reap_one(KVMState *s, CPUState *cpu) | |
685 | { | |
686 | struct kvm_dirty_gfn *dirty_gfns = cpu->kvm_dirty_gfns, *cur; | |
687 | uint32_t ring_size = s->kvm_dirty_ring_size; | |
688 | uint32_t count = 0, fetch = cpu->kvm_fetch_index; | |
689 | ||
56adee40 PX |
690 | /* |
691 | * It's possible that we race with vcpu creation code where the vcpu is | |
692 | * put onto the vcpus list but not yet initialized the dirty ring | |
693 | * structures. If so, skip it. | |
694 | */ | |
695 | if (!cpu->created) { | |
696 | return 0; | |
697 | } | |
698 | ||
b4420f19 PX |
699 | assert(dirty_gfns && ring_size); |
700 | trace_kvm_dirty_ring_reap_vcpu(cpu->cpu_index); | |
701 | ||
702 | while (true) { | |
703 | cur = &dirty_gfns[fetch % ring_size]; | |
704 | if (!dirty_gfn_is_dirtied(cur)) { | |
705 | break; | |
706 | } | |
707 | kvm_dirty_ring_mark_page(s, cur->slot >> 16, cur->slot & 0xffff, | |
708 | cur->offset); | |
709 | dirty_gfn_set_collected(cur); | |
710 | trace_kvm_dirty_ring_page(cpu->cpu_index, fetch, cur->offset); | |
711 | fetch++; | |
712 | count++; | |
713 | } | |
714 | cpu->kvm_fetch_index = fetch; | |
7786ae40 | 715 | cpu->dirty_pages += count; |
b4420f19 PX |
716 | |
717 | return count; | |
718 | } | |
719 | ||
720 | /* Must be with slots_lock held */ | |
1667e2b9 | 721 | static uint64_t kvm_dirty_ring_reap_locked(KVMState *s, CPUState* cpu) |
b4420f19 PX |
722 | { |
723 | int ret; | |
b4420f19 PX |
724 | uint64_t total = 0; |
725 | int64_t stamp; | |
726 | ||
727 | stamp = get_clock(); | |
728 | ||
1667e2b9 HH |
729 | if (cpu) { |
730 | total = kvm_dirty_ring_reap_one(s, cpu); | |
731 | } else { | |
732 | CPU_FOREACH(cpu) { | |
733 | total += kvm_dirty_ring_reap_one(s, cpu); | |
734 | } | |
b4420f19 PX |
735 | } |
736 | ||
737 | if (total) { | |
738 | ret = kvm_vm_ioctl(s, KVM_RESET_DIRTY_RINGS); | |
739 | assert(ret == total); | |
740 | } | |
741 | ||
742 | stamp = get_clock() - stamp; | |
743 | ||
744 | if (total) { | |
745 | trace_kvm_dirty_ring_reap(total, stamp / 1000); | |
746 | } | |
747 | ||
748 | return total; | |
749 | } | |
750 | ||
751 | /* | |
752 | * Currently for simplicity, we must hold BQL before calling this. We can | |
753 | * consider to drop the BQL if we're clear with all the race conditions. | |
754 | */ | |
1667e2b9 | 755 | static uint64_t kvm_dirty_ring_reap(KVMState *s, CPUState *cpu) |
b4420f19 PX |
756 | { |
757 | uint64_t total; | |
758 | ||
759 | /* | |
760 | * We need to lock all kvm slots for all address spaces here, | |
761 | * because: | |
762 | * | |
763 | * (1) We need to mark dirty for dirty bitmaps in multiple slots | |
764 | * and for tons of pages, so it's better to take the lock here | |
765 | * once rather than once per page. And more importantly, | |
766 | * | |
767 | * (2) We must _NOT_ publish dirty bits to the other threads | |
768 | * (e.g., the migration thread) via the kvm memory slot dirty | |
769 | * bitmaps before correctly re-protect those dirtied pages. | |
770 | * Otherwise we can have potential risk of data corruption if | |
771 | * the page data is read in the other thread before we do | |
772 | * reset below. | |
773 | */ | |
774 | kvm_slots_lock(); | |
1667e2b9 | 775 | total = kvm_dirty_ring_reap_locked(s, cpu); |
b4420f19 PX |
776 | kvm_slots_unlock(); |
777 | ||
778 | return total; | |
779 | } | |
780 | ||
781 | static void do_kvm_cpu_synchronize_kick(CPUState *cpu, run_on_cpu_data arg) | |
782 | { | |
783 | /* No need to do anything */ | |
784 | } | |
785 | ||
786 | /* | |
787 | * Kick all vcpus out in a synchronized way. When returned, we | |
788 | * guarantee that every vcpu has been kicked and at least returned to | |
789 | * userspace once. | |
790 | */ | |
791 | static void kvm_cpu_synchronize_kick_all(void) | |
792 | { | |
793 | CPUState *cpu; | |
794 | ||
795 | CPU_FOREACH(cpu) { | |
796 | run_on_cpu(cpu, do_kvm_cpu_synchronize_kick, RUN_ON_CPU_NULL); | |
797 | } | |
798 | } | |
799 | ||
800 | /* | |
801 | * Flush all the existing dirty pages to the KVM slot buffers. When | |
802 | * this call returns, we guarantee that all the touched dirty pages | |
803 | * before calling this function have been put into the per-kvmslot | |
804 | * dirty bitmap. | |
805 | * | |
806 | * This function must be called with BQL held. | |
807 | */ | |
808 | static void kvm_dirty_ring_flush(void) | |
809 | { | |
810 | trace_kvm_dirty_ring_flush(0); | |
811 | /* | |
812 | * The function needs to be serialized. Since this function | |
813 | * should always be with BQL held, serialization is guaranteed. | |
814 | * However, let's be sure of it. | |
815 | */ | |
816 | assert(qemu_mutex_iothread_locked()); | |
817 | /* | |
818 | * First make sure to flush the hardware buffers by kicking all | |
819 | * vcpus out in a synchronous way. | |
820 | */ | |
821 | kvm_cpu_synchronize_kick_all(); | |
1667e2b9 | 822 | kvm_dirty_ring_reap(kvm_state, NULL); |
b4420f19 PX |
823 | trace_kvm_dirty_ring_flush(1); |
824 | } | |
825 | ||
5832d1f2 | 826 | /** |
4a12a11a | 827 | * kvm_physical_sync_dirty_bitmap - Sync dirty bitmap from kernel space |
5832d1f2 | 828 | * |
4a12a11a PX |
829 | * This function will first try to fetch dirty bitmap from the kernel, |
830 | * and then updates qemu's dirty bitmap. | |
831 | * | |
36adac49 PX |
832 | * NOTE: caller must be with kml->slots_lock held. |
833 | * | |
4a12a11a PX |
834 | * @kml: the KVM memory listener object |
835 | * @section: the memory section to sync the dirty bitmap with | |
5832d1f2 | 836 | */ |
e65e5f50 PX |
837 | static void kvm_physical_sync_dirty_bitmap(KVMMemoryListener *kml, |
838 | MemoryRegionSection *section) | |
5832d1f2 AL |
839 | { |
840 | KVMState *s = kvm_state; | |
151f7749 | 841 | KVMSlot *mem; |
67548f09 | 842 | hwaddr start_addr, size; |
2c20b27e | 843 | hwaddr slot_size; |
67548f09 DH |
844 | |
845 | size = kvm_align_section(section, &start_addr); | |
023ae9a8 | 846 | while (size) { |
023ae9a8 IM |
847 | slot_size = MIN(kvm_max_slot_size, size); |
848 | mem = kvm_lookup_matching_slot(kml, start_addr, slot_size); | |
67548f09 | 849 | if (!mem) { |
e377e87c | 850 | /* We don't have a slot if we want to trap every access. */ |
e65e5f50 | 851 | return; |
151f7749 | 852 | } |
e65e5f50 | 853 | if (kvm_slot_get_dirty_log(s, mem)) { |
2c20b27e | 854 | kvm_slot_sync_dirty_pages(mem); |
151f7749 | 855 | } |
023ae9a8 IM |
856 | start_addr += slot_size; |
857 | size -= slot_size; | |
5832d1f2 | 858 | } |
5832d1f2 AL |
859 | } |
860 | ||
ff4aa114 PX |
861 | /* Alignment requirement for KVM_CLEAR_DIRTY_LOG - 64 pages */ |
862 | #define KVM_CLEAR_LOG_SHIFT 6 | |
8e3b0cbb | 863 | #define KVM_CLEAR_LOG_ALIGN (qemu_real_host_page_size() << KVM_CLEAR_LOG_SHIFT) |
ff4aa114 PX |
864 | #define KVM_CLEAR_LOG_MASK (-KVM_CLEAR_LOG_ALIGN) |
865 | ||
4222147d PB |
866 | static int kvm_log_clear_one_slot(KVMSlot *mem, int as_id, uint64_t start, |
867 | uint64_t size) | |
ff4aa114 PX |
868 | { |
869 | KVMState *s = kvm_state; | |
4222147d | 870 | uint64_t end, bmap_start, start_delta, bmap_npages; |
ff4aa114 | 871 | struct kvm_clear_dirty_log d; |
8e3b0cbb | 872 | unsigned long *bmap_clear = NULL, psize = qemu_real_host_page_size(); |
4222147d | 873 | int ret; |
ff4aa114 PX |
874 | |
875 | /* | |
876 | * We need to extend either the start or the size or both to | |
877 | * satisfy the KVM interface requirement. Firstly, do the start | |
878 | * page alignment on 64 host pages | |
879 | */ | |
84516e5b PB |
880 | bmap_start = start & KVM_CLEAR_LOG_MASK; |
881 | start_delta = start - bmap_start; | |
ff4aa114 PX |
882 | bmap_start /= psize; |
883 | ||
884 | /* | |
885 | * The kernel interface has restriction on the size too, that either: | |
886 | * | |
887 | * (1) the size is 64 host pages aligned (just like the start), or | |
888 | * (2) the size fills up until the end of the KVM memslot. | |
889 | */ | |
890 | bmap_npages = DIV_ROUND_UP(size + start_delta, KVM_CLEAR_LOG_ALIGN) | |
891 | << KVM_CLEAR_LOG_SHIFT; | |
892 | end = mem->memory_size / psize; | |
893 | if (bmap_npages > end - bmap_start) { | |
894 | bmap_npages = end - bmap_start; | |
895 | } | |
896 | start_delta /= psize; | |
897 | ||
898 | /* | |
899 | * Prepare the bitmap to clear dirty bits. Here we must guarantee | |
900 | * that we won't clear any unknown dirty bits otherwise we might | |
901 | * accidentally clear some set bits which are not yet synced from | |
902 | * the kernel into QEMU's bitmap, then we'll lose track of the | |
903 | * guest modifications upon those pages (which can directly lead | |
904 | * to guest data loss or panic after migration). | |
905 | * | |
906 | * Layout of the KVMSlot.dirty_bmap: | |
907 | * | |
908 | * |<-------- bmap_npages -----------..>| | |
909 | * [1] | |
910 | * start_delta size | |
911 | * |----------------|-------------|------------------|------------| | |
912 | * ^ ^ ^ ^ | |
913 | * | | | | | |
914 | * start bmap_start (start) end | |
915 | * of memslot of memslot | |
916 | * | |
917 | * [1] bmap_npages can be aligned to either 64 pages or the end of slot | |
918 | */ | |
919 | ||
920 | assert(bmap_start % BITS_PER_LONG == 0); | |
921 | /* We should never do log_clear before log_sync */ | |
922 | assert(mem->dirty_bmap); | |
4054adbd | 923 | if (start_delta || bmap_npages - size / psize) { |
ff4aa114 PX |
924 | /* Slow path - we need to manipulate a temp bitmap */ |
925 | bmap_clear = bitmap_new(bmap_npages); | |
926 | bitmap_copy_with_src_offset(bmap_clear, mem->dirty_bmap, | |
927 | bmap_start, start_delta + size / psize); | |
928 | /* | |
929 | * We need to fill the holes at start because that was not | |
930 | * specified by the caller and we extended the bitmap only for | |
931 | * 64 pages alignment | |
932 | */ | |
933 | bitmap_clear(bmap_clear, 0, start_delta); | |
934 | d.dirty_bitmap = bmap_clear; | |
935 | } else { | |
4054adbd ZY |
936 | /* |
937 | * Fast path - both start and size align well with BITS_PER_LONG | |
938 | * (or the end of memory slot) | |
939 | */ | |
ff4aa114 PX |
940 | d.dirty_bitmap = mem->dirty_bmap + BIT_WORD(bmap_start); |
941 | } | |
942 | ||
943 | d.first_page = bmap_start; | |
944 | /* It should never overflow. If it happens, say something */ | |
945 | assert(bmap_npages <= UINT32_MAX); | |
946 | d.num_pages = bmap_npages; | |
4222147d | 947 | d.slot = mem->slot | (as_id << 16); |
ff4aa114 | 948 | |
38e0b790 TH |
949 | ret = kvm_vm_ioctl(s, KVM_CLEAR_DIRTY_LOG, &d); |
950 | if (ret < 0 && ret != -ENOENT) { | |
ff4aa114 PX |
951 | error_report("%s: KVM_CLEAR_DIRTY_LOG failed, slot=%d, " |
952 | "start=0x%"PRIx64", size=0x%"PRIx32", errno=%d", | |
953 | __func__, d.slot, (uint64_t)d.first_page, | |
954 | (uint32_t)d.num_pages, ret); | |
955 | } else { | |
956 | ret = 0; | |
957 | trace_kvm_clear_dirty_log(d.slot, d.first_page, d.num_pages); | |
958 | } | |
959 | ||
960 | /* | |
961 | * After we have updated the remote dirty bitmap, we update the | |
962 | * cached bitmap as well for the memslot, then if another user | |
963 | * clears the same region we know we shouldn't clear it again on | |
964 | * the remote otherwise it's data loss as well. | |
965 | */ | |
966 | bitmap_clear(mem->dirty_bmap, bmap_start + start_delta, | |
967 | size / psize); | |
968 | /* This handles the NULL case well */ | |
969 | g_free(bmap_clear); | |
4222147d PB |
970 | return ret; |
971 | } | |
972 | ||
973 | ||
974 | /** | |
975 | * kvm_physical_log_clear - Clear the kernel's dirty bitmap for range | |
976 | * | |
977 | * NOTE: this will be a no-op if we haven't enabled manual dirty log | |
978 | * protection in the host kernel because in that case this operation | |
979 | * will be done within log_sync(). | |
980 | * | |
981 | * @kml: the kvm memory listener | |
982 | * @section: the memory range to clear dirty bitmap | |
983 | */ | |
984 | static int kvm_physical_log_clear(KVMMemoryListener *kml, | |
985 | MemoryRegionSection *section) | |
986 | { | |
987 | KVMState *s = kvm_state; | |
84516e5b PB |
988 | uint64_t start, size, offset, count; |
989 | KVMSlot *mem; | |
87287ac0 | 990 | int ret = 0, i; |
4222147d PB |
991 | |
992 | if (!s->manual_dirty_log_protect) { | |
993 | /* No need to do explicit clear */ | |
87287ac0 | 994 | return ret; |
4222147d PB |
995 | } |
996 | ||
997 | start = section->offset_within_address_space; | |
998 | size = int128_get64(section->size); | |
999 | ||
1000 | if (!size) { | |
1001 | /* Nothing more we can do... */ | |
87287ac0 | 1002 | return ret; |
4222147d PB |
1003 | } |
1004 | ||
a2f77862 | 1005 | kvm_slots_lock(); |
4222147d | 1006 | |
4222147d PB |
1007 | for (i = 0; i < s->nr_slots; i++) { |
1008 | mem = &kml->slots[i]; | |
84516e5b PB |
1009 | /* Discard slots that are empty or do not overlap the section */ |
1010 | if (!mem->memory_size || | |
1011 | mem->start_addr > start + size - 1 || | |
1012 | start > mem->start_addr + mem->memory_size - 1) { | |
1013 | continue; | |
1014 | } | |
1015 | ||
1016 | if (start >= mem->start_addr) { | |
1017 | /* The slot starts before section or is aligned to it. */ | |
1018 | offset = start - mem->start_addr; | |
1019 | count = MIN(mem->memory_size - offset, size); | |
1020 | } else { | |
1021 | /* The slot starts after section. */ | |
1022 | offset = 0; | |
1023 | count = MIN(mem->memory_size, size - (mem->start_addr - start)); | |
1024 | } | |
1025 | ret = kvm_log_clear_one_slot(mem, kml->as_id, offset, count); | |
1026 | if (ret < 0) { | |
4222147d PB |
1027 | break; |
1028 | } | |
1029 | } | |
1030 | ||
a2f77862 | 1031 | kvm_slots_unlock(); |
ff4aa114 PX |
1032 | |
1033 | return ret; | |
1034 | } | |
1035 | ||
95d2994a AK |
1036 | static void kvm_coalesce_mmio_region(MemoryListener *listener, |
1037 | MemoryRegionSection *secion, | |
a8170e5e | 1038 | hwaddr start, hwaddr size) |
f65ed4c1 | 1039 | { |
f65ed4c1 AL |
1040 | KVMState *s = kvm_state; |
1041 | ||
1042 | if (s->coalesced_mmio) { | |
1043 | struct kvm_coalesced_mmio_zone zone; | |
1044 | ||
1045 | zone.addr = start; | |
1046 | zone.size = size; | |
7e680753 | 1047 | zone.pad = 0; |
f65ed4c1 | 1048 | |
95d2994a | 1049 | (void)kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone); |
f65ed4c1 | 1050 | } |
f65ed4c1 AL |
1051 | } |
1052 | ||
95d2994a AK |
1053 | static void kvm_uncoalesce_mmio_region(MemoryListener *listener, |
1054 | MemoryRegionSection *secion, | |
a8170e5e | 1055 | hwaddr start, hwaddr size) |
f65ed4c1 | 1056 | { |
f65ed4c1 AL |
1057 | KVMState *s = kvm_state; |
1058 | ||
1059 | if (s->coalesced_mmio) { | |
1060 | struct kvm_coalesced_mmio_zone zone; | |
1061 | ||
1062 | zone.addr = start; | |
1063 | zone.size = size; | |
7e680753 | 1064 | zone.pad = 0; |
f65ed4c1 | 1065 | |
95d2994a | 1066 | (void)kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone); |
f65ed4c1 | 1067 | } |
f65ed4c1 AL |
1068 | } |
1069 | ||
e6d34aee PH |
1070 | static void kvm_coalesce_pio_add(MemoryListener *listener, |
1071 | MemoryRegionSection *section, | |
1072 | hwaddr start, hwaddr size) | |
1073 | { | |
1074 | KVMState *s = kvm_state; | |
1075 | ||
1076 | if (s->coalesced_pio) { | |
1077 | struct kvm_coalesced_mmio_zone zone; | |
1078 | ||
1079 | zone.addr = start; | |
1080 | zone.size = size; | |
1081 | zone.pio = 1; | |
1082 | ||
1083 | (void)kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone); | |
1084 | } | |
1085 | } | |
1086 | ||
1087 | static void kvm_coalesce_pio_del(MemoryListener *listener, | |
1088 | MemoryRegionSection *section, | |
1089 | hwaddr start, hwaddr size) | |
1090 | { | |
1091 | KVMState *s = kvm_state; | |
1092 | ||
1093 | if (s->coalesced_pio) { | |
1094 | struct kvm_coalesced_mmio_zone zone; | |
1095 | ||
1096 | zone.addr = start; | |
1097 | zone.size = size; | |
1098 | zone.pio = 1; | |
1099 | ||
1100 | (void)kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone); | |
1101 | } | |
1102 | } | |
1103 | ||
1104 | static MemoryListener kvm_coalesced_pio_listener = { | |
142518bd | 1105 | .name = "kvm-coalesced-pio", |
e6d34aee PH |
1106 | .coalesced_io_add = kvm_coalesce_pio_add, |
1107 | .coalesced_io_del = kvm_coalesce_pio_del, | |
14a868c6 | 1108 | .priority = MEMORY_LISTENER_PRIORITY_MIN, |
e6d34aee PH |
1109 | }; |
1110 | ||
ad7b8b33 AL |
1111 | int kvm_check_extension(KVMState *s, unsigned int extension) |
1112 | { | |
1113 | int ret; | |
1114 | ||
1115 | ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension); | |
1116 | if (ret < 0) { | |
1117 | ret = 0; | |
1118 | } | |
1119 | ||
1120 | return ret; | |
1121 | } | |
1122 | ||
7d0a07fa AG |
1123 | int kvm_vm_check_extension(KVMState *s, unsigned int extension) |
1124 | { | |
1125 | int ret; | |
1126 | ||
1127 | ret = kvm_vm_ioctl(s, KVM_CHECK_EXTENSION, extension); | |
1128 | if (ret < 0) { | |
1129 | /* VM wide version not implemented, use global one instead */ | |
1130 | ret = kvm_check_extension(s, extension); | |
1131 | } | |
1132 | ||
1133 | return ret; | |
1134 | } | |
1135 | ||
6b552b9b DG |
1136 | typedef struct HWPoisonPage { |
1137 | ram_addr_t ram_addr; | |
1138 | QLIST_ENTRY(HWPoisonPage) list; | |
1139 | } HWPoisonPage; | |
1140 | ||
1141 | static QLIST_HEAD(, HWPoisonPage) hwpoison_page_list = | |
1142 | QLIST_HEAD_INITIALIZER(hwpoison_page_list); | |
1143 | ||
1144 | static void kvm_unpoison_all(void *param) | |
1145 | { | |
1146 | HWPoisonPage *page, *next_page; | |
1147 | ||
1148 | QLIST_FOREACH_SAFE(page, &hwpoison_page_list, list, next_page) { | |
1149 | QLIST_REMOVE(page, list); | |
1150 | qemu_ram_remap(page->ram_addr, TARGET_PAGE_SIZE); | |
1151 | g_free(page); | |
1152 | } | |
1153 | } | |
1154 | ||
1155 | void kvm_hwpoison_page_add(ram_addr_t ram_addr) | |
1156 | { | |
1157 | HWPoisonPage *page; | |
1158 | ||
1159 | QLIST_FOREACH(page, &hwpoison_page_list, list) { | |
1160 | if (page->ram_addr == ram_addr) { | |
1161 | return; | |
1162 | } | |
1163 | } | |
1164 | page = g_new(HWPoisonPage, 1); | |
1165 | page->ram_addr = ram_addr; | |
1166 | QLIST_INSERT_HEAD(&hwpoison_page_list, page, list); | |
1167 | } | |
1168 | ||
b680c5ba GK |
1169 | static uint32_t adjust_ioeventfd_endianness(uint32_t val, uint32_t size) |
1170 | { | |
ee3eb3a7 | 1171 | #if HOST_BIG_ENDIAN != TARGET_BIG_ENDIAN |
e03b5686 | 1172 | /* The kernel expects ioeventfd values in HOST_BIG_ENDIAN |
b680c5ba GK |
1173 | * endianness, but the memory core hands them in target endianness. |
1174 | * For example, PPC is always treated as big-endian even if running | |
1175 | * on KVM and on PPC64LE. Correct here. | |
1176 | */ | |
1177 | switch (size) { | |
1178 | case 2: | |
1179 | val = bswap16(val); | |
1180 | break; | |
1181 | case 4: | |
1182 | val = bswap32(val); | |
1183 | break; | |
1184 | } | |
1185 | #endif | |
1186 | return val; | |
1187 | } | |
1188 | ||
584f2be7 | 1189 | static int kvm_set_ioeventfd_mmio(int fd, hwaddr addr, uint32_t val, |
41cb62c2 | 1190 | bool assign, uint32_t size, bool datamatch) |
500ffd4a MT |
1191 | { |
1192 | int ret; | |
03a96b83 TH |
1193 | struct kvm_ioeventfd iofd = { |
1194 | .datamatch = datamatch ? adjust_ioeventfd_endianness(val, size) : 0, | |
1195 | .addr = addr, | |
1196 | .len = size, | |
1197 | .flags = 0, | |
1198 | .fd = fd, | |
1199 | }; | |
500ffd4a | 1200 | |
876d16cd DDAG |
1201 | trace_kvm_set_ioeventfd_mmio(fd, (uint64_t)addr, val, assign, size, |
1202 | datamatch); | |
500ffd4a MT |
1203 | if (!kvm_enabled()) { |
1204 | return -ENOSYS; | |
1205 | } | |
1206 | ||
41cb62c2 MT |
1207 | if (datamatch) { |
1208 | iofd.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH; | |
1209 | } | |
500ffd4a MT |
1210 | if (!assign) { |
1211 | iofd.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; | |
1212 | } | |
1213 | ||
1214 | ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd); | |
1215 | ||
1216 | if (ret < 0) { | |
1217 | return -errno; | |
1218 | } | |
1219 | ||
1220 | return 0; | |
1221 | } | |
1222 | ||
44c3f8f7 | 1223 | static int kvm_set_ioeventfd_pio(int fd, uint16_t addr, uint16_t val, |
41cb62c2 | 1224 | bool assign, uint32_t size, bool datamatch) |
500ffd4a MT |
1225 | { |
1226 | struct kvm_ioeventfd kick = { | |
b680c5ba | 1227 | .datamatch = datamatch ? adjust_ioeventfd_endianness(val, size) : 0, |
500ffd4a | 1228 | .addr = addr, |
41cb62c2 | 1229 | .flags = KVM_IOEVENTFD_FLAG_PIO, |
44c3f8f7 | 1230 | .len = size, |
500ffd4a MT |
1231 | .fd = fd, |
1232 | }; | |
1233 | int r; | |
876d16cd | 1234 | trace_kvm_set_ioeventfd_pio(fd, addr, val, assign, size, datamatch); |
500ffd4a MT |
1235 | if (!kvm_enabled()) { |
1236 | return -ENOSYS; | |
1237 | } | |
41cb62c2 MT |
1238 | if (datamatch) { |
1239 | kick.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH; | |
1240 | } | |
500ffd4a MT |
1241 | if (!assign) { |
1242 | kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; | |
1243 | } | |
1244 | r = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick); | |
1245 | if (r < 0) { | |
1246 | return r; | |
1247 | } | |
1248 | return 0; | |
1249 | } | |
1250 | ||
1251 | ||
d2f2b8a7 SH |
1252 | static int kvm_check_many_ioeventfds(void) |
1253 | { | |
d0dcac83 SH |
1254 | /* Userspace can use ioeventfd for io notification. This requires a host |
1255 | * that supports eventfd(2) and an I/O thread; since eventfd does not | |
1256 | * support SIGIO it cannot interrupt the vcpu. | |
1257 | * | |
1258 | * Older kernels have a 6 device limit on the KVM io bus. Find out so we | |
d2f2b8a7 SH |
1259 | * can avoid creating too many ioeventfds. |
1260 | */ | |
12d4536f | 1261 | #if defined(CONFIG_EVENTFD) |
d2f2b8a7 SH |
1262 | int ioeventfds[7]; |
1263 | int i, ret = 0; | |
1264 | for (i = 0; i < ARRAY_SIZE(ioeventfds); i++) { | |
1265 | ioeventfds[i] = eventfd(0, EFD_CLOEXEC); | |
1266 | if (ioeventfds[i] < 0) { | |
1267 | break; | |
1268 | } | |
41cb62c2 | 1269 | ret = kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, true, 2, true); |
d2f2b8a7 SH |
1270 | if (ret < 0) { |
1271 | close(ioeventfds[i]); | |
1272 | break; | |
1273 | } | |
1274 | } | |
1275 | ||
1276 | /* Decide whether many devices are supported or not */ | |
1277 | ret = i == ARRAY_SIZE(ioeventfds); | |
1278 | ||
1279 | while (i-- > 0) { | |
41cb62c2 | 1280 | kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, false, 2, true); |
d2f2b8a7 SH |
1281 | close(ioeventfds[i]); |
1282 | } | |
1283 | return ret; | |
1284 | #else | |
1285 | return 0; | |
1286 | #endif | |
1287 | } | |
1288 | ||
94a8d39a JK |
1289 | static const KVMCapabilityInfo * |
1290 | kvm_check_extension_list(KVMState *s, const KVMCapabilityInfo *list) | |
1291 | { | |
1292 | while (list->name) { | |
1293 | if (!kvm_check_extension(s, list->value)) { | |
1294 | return list; | |
1295 | } | |
1296 | list++; | |
1297 | } | |
1298 | return NULL; | |
1299 | } | |
1300 | ||
023ae9a8 IM |
1301 | void kvm_set_max_memslot_size(hwaddr max_slot_size) |
1302 | { | |
1303 | g_assert( | |
8e3b0cbb | 1304 | ROUND_UP(max_slot_size, qemu_real_host_page_size()) == max_slot_size |
023ae9a8 IM |
1305 | ); |
1306 | kvm_max_slot_size = max_slot_size; | |
1307 | } | |
1308 | ||
f39b7d2b | 1309 | /* Called with KVMMemoryListener.slots_lock held */ |
7bbda04c PB |
1310 | static void kvm_set_phys_mem(KVMMemoryListener *kml, |
1311 | MemoryRegionSection *section, bool add) | |
46dbef6a | 1312 | { |
f357f564 | 1313 | KVMSlot *mem; |
46dbef6a | 1314 | int err; |
a01672d3 | 1315 | MemoryRegion *mr = section->mr; |
9323e79f | 1316 | bool writable = !mr->readonly && !mr->rom_device; |
2c20b27e PX |
1317 | hwaddr start_addr, size, slot_size, mr_offset; |
1318 | ram_addr_t ram_start_offset; | |
5ea69c2e | 1319 | void *ram; |
46dbef6a | 1320 | |
a01672d3 | 1321 | if (!memory_region_is_ram(mr)) { |
9323e79f | 1322 | if (writable || !kvm_readonly_mem_allowed) { |
235e8982 JJ |
1323 | return; |
1324 | } else if (!mr->romd_mode) { | |
1325 | /* If the memory device is not in romd_mode, then we actually want | |
1326 | * to remove the kvm memory slot so all accesses will trap. */ | |
1327 | add = false; | |
1328 | } | |
9f213ed9 AK |
1329 | } |
1330 | ||
5ea69c2e DH |
1331 | size = kvm_align_section(section, &start_addr); |
1332 | if (!size) { | |
1333 | return; | |
1334 | } | |
1335 | ||
2c20b27e PX |
1336 | /* The offset of the kvmslot within the memory region */ |
1337 | mr_offset = section->offset_within_region + start_addr - | |
1338 | section->offset_within_address_space; | |
1339 | ||
1340 | /* use aligned delta to align the ram address and offset */ | |
1341 | ram = memory_region_get_ram_ptr(mr) + mr_offset; | |
1342 | ram_start_offset = memory_region_get_ram_addr(mr) + mr_offset; | |
a01672d3 | 1343 | |
f357f564 | 1344 | if (!add) { |
023ae9a8 IM |
1345 | do { |
1346 | slot_size = MIN(kvm_max_slot_size, size); | |
1347 | mem = kvm_lookup_matching_slot(kml, start_addr, slot_size); | |
1348 | if (!mem) { | |
f39b7d2b | 1349 | return; |
023ae9a8 IM |
1350 | } |
1351 | if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) { | |
b4420f19 PX |
1352 | /* |
1353 | * NOTE: We should be aware of the fact that here we're only | |
1354 | * doing a best effort to sync dirty bits. No matter whether | |
1355 | * we're using dirty log or dirty ring, we ignored two facts: | |
1356 | * | |
1357 | * (1) dirty bits can reside in hardware buffers (PML) | |
1358 | * | |
1359 | * (2) after we collected dirty bits here, pages can be dirtied | |
1360 | * again before we do the final KVM_SET_USER_MEMORY_REGION to | |
1361 | * remove the slot. | |
1362 | * | |
1363 | * Not easy. Let's cross the fingers until it's fixed. | |
1364 | */ | |
1365 | if (kvm_state->kvm_dirty_ring_size) { | |
1667e2b9 | 1366 | kvm_dirty_ring_reap_locked(kvm_state, NULL); |
b20cc776 GS |
1367 | if (kvm_state->kvm_dirty_ring_with_bitmap) { |
1368 | kvm_slot_sync_dirty_pages(mem); | |
1369 | kvm_slot_get_dirty_log(kvm_state, mem); | |
1370 | } | |
b4420f19 PX |
1371 | } else { |
1372 | kvm_slot_get_dirty_log(kvm_state, mem); | |
1373 | } | |
29b7e8be | 1374 | kvm_slot_sync_dirty_pages(mem); |
023ae9a8 | 1375 | } |
3fbffb62 | 1376 | |
023ae9a8 IM |
1377 | /* unregister the slot */ |
1378 | g_free(mem->dirty_bmap); | |
1379 | mem->dirty_bmap = NULL; | |
1380 | mem->memory_size = 0; | |
1381 | mem->flags = 0; | |
1382 | err = kvm_set_user_memory_region(kml, mem, false); | |
1383 | if (err) { | |
1384 | fprintf(stderr, "%s: error unregistering slot: %s\n", | |
1385 | __func__, strerror(-err)); | |
1386 | abort(); | |
1387 | } | |
1388 | start_addr += slot_size; | |
1389 | size -= slot_size; | |
1390 | } while (size); | |
f39b7d2b | 1391 | return; |
46dbef6a MT |
1392 | } |
1393 | ||
f357f564 | 1394 | /* register the new slot */ |
023ae9a8 IM |
1395 | do { |
1396 | slot_size = MIN(kvm_max_slot_size, size); | |
1397 | mem = kvm_alloc_slot(kml); | |
e65e5f50 | 1398 | mem->as_id = kml->as_id; |
023ae9a8 IM |
1399 | mem->memory_size = slot_size; |
1400 | mem->start_addr = start_addr; | |
2c20b27e | 1401 | mem->ram_start_offset = ram_start_offset; |
023ae9a8 IM |
1402 | mem->ram = ram; |
1403 | mem->flags = kvm_mem_flags(mr); | |
ea776d15 | 1404 | kvm_slot_init_dirty_bitmap(mem); |
023ae9a8 IM |
1405 | err = kvm_set_user_memory_region(kml, mem, true); |
1406 | if (err) { | |
1407 | fprintf(stderr, "%s: error registering slot: %s\n", __func__, | |
1408 | strerror(-err)); | |
1409 | abort(); | |
1410 | } | |
1411 | start_addr += slot_size; | |
2c20b27e | 1412 | ram_start_offset += slot_size; |
023ae9a8 IM |
1413 | ram += slot_size; |
1414 | size -= slot_size; | |
1415 | } while (size); | |
46dbef6a MT |
1416 | } |
1417 | ||
b4420f19 PX |
1418 | static void *kvm_dirty_ring_reaper_thread(void *data) |
1419 | { | |
1420 | KVMState *s = data; | |
1421 | struct KVMDirtyRingReaper *r = &s->reaper; | |
1422 | ||
1423 | rcu_register_thread(); | |
1424 | ||
1425 | trace_kvm_dirty_ring_reaper("init"); | |
1426 | ||
1427 | while (true) { | |
1428 | r->reaper_state = KVM_DIRTY_RING_REAPER_WAIT; | |
1429 | trace_kvm_dirty_ring_reaper("wait"); | |
1430 | /* | |
1431 | * TODO: provide a smarter timeout rather than a constant? | |
1432 | */ | |
1433 | sleep(1); | |
1434 | ||
baa60983 HH |
1435 | /* keep sleeping so that dirtylimit not be interfered by reaper */ |
1436 | if (dirtylimit_in_service()) { | |
1437 | continue; | |
1438 | } | |
1439 | ||
b4420f19 PX |
1440 | trace_kvm_dirty_ring_reaper("wakeup"); |
1441 | r->reaper_state = KVM_DIRTY_RING_REAPER_REAPING; | |
1442 | ||
1443 | qemu_mutex_lock_iothread(); | |
1667e2b9 | 1444 | kvm_dirty_ring_reap(s, NULL); |
b4420f19 PX |
1445 | qemu_mutex_unlock_iothread(); |
1446 | ||
1447 | r->reaper_iteration++; | |
1448 | } | |
1449 | ||
1450 | trace_kvm_dirty_ring_reaper("exit"); | |
1451 | ||
1452 | rcu_unregister_thread(); | |
1453 | ||
1454 | return NULL; | |
1455 | } | |
1456 | ||
1457 | static int kvm_dirty_ring_reaper_init(KVMState *s) | |
1458 | { | |
1459 | struct KVMDirtyRingReaper *r = &s->reaper; | |
1460 | ||
1461 | qemu_thread_create(&r->reaper_thr, "kvm-reaper", | |
1462 | kvm_dirty_ring_reaper_thread, | |
1463 | s, QEMU_THREAD_JOINABLE); | |
1464 | ||
1465 | return 0; | |
1466 | } | |
1467 | ||
3794cb94 GS |
1468 | static int kvm_dirty_ring_init(KVMState *s) |
1469 | { | |
1470 | uint32_t ring_size = s->kvm_dirty_ring_size; | |
1471 | uint64_t ring_bytes = ring_size * sizeof(struct kvm_dirty_gfn); | |
856e23a0 | 1472 | unsigned int capability = KVM_CAP_DIRTY_LOG_RING; |
3794cb94 GS |
1473 | int ret; |
1474 | ||
1475 | s->kvm_dirty_ring_size = 0; | |
1476 | s->kvm_dirty_ring_bytes = 0; | |
1477 | ||
1478 | /* Bail if the dirty ring size isn't specified */ | |
1479 | if (!ring_size) { | |
1480 | return 0; | |
1481 | } | |
1482 | ||
1483 | /* | |
1484 | * Read the max supported pages. Fall back to dirty logging mode | |
1485 | * if the dirty ring isn't supported. | |
1486 | */ | |
856e23a0 GS |
1487 | ret = kvm_vm_check_extension(s, capability); |
1488 | if (ret <= 0) { | |
1489 | capability = KVM_CAP_DIRTY_LOG_RING_ACQ_REL; | |
1490 | ret = kvm_vm_check_extension(s, capability); | |
1491 | } | |
1492 | ||
3794cb94 GS |
1493 | if (ret <= 0) { |
1494 | warn_report("KVM dirty ring not available, using bitmap method"); | |
1495 | return 0; | |
1496 | } | |
1497 | ||
1498 | if (ring_bytes > ret) { | |
1499 | error_report("KVM dirty ring size %" PRIu32 " too big " | |
1500 | "(maximum is %ld). Please use a smaller value.", | |
1501 | ring_size, (long)ret / sizeof(struct kvm_dirty_gfn)); | |
1502 | return -EINVAL; | |
1503 | } | |
1504 | ||
856e23a0 | 1505 | ret = kvm_vm_enable_cap(s, capability, 0, ring_bytes); |
3794cb94 GS |
1506 | if (ret) { |
1507 | error_report("Enabling of KVM dirty ring failed: %s. " | |
1508 | "Suggested minimum value is 1024.", strerror(-ret)); | |
1509 | return -EIO; | |
1510 | } | |
1511 | ||
856e23a0 GS |
1512 | /* Enable the backup bitmap if it is supported */ |
1513 | ret = kvm_vm_check_extension(s, KVM_CAP_DIRTY_LOG_RING_WITH_BITMAP); | |
1514 | if (ret > 0) { | |
1515 | ret = kvm_vm_enable_cap(s, KVM_CAP_DIRTY_LOG_RING_WITH_BITMAP, 0); | |
1516 | if (ret) { | |
1517 | error_report("Enabling of KVM dirty ring's backup bitmap failed: " | |
1518 | "%s. ", strerror(-ret)); | |
1519 | return -EIO; | |
1520 | } | |
1521 | ||
1522 | s->kvm_dirty_ring_with_bitmap = true; | |
1523 | } | |
1524 | ||
3794cb94 GS |
1525 | s->kvm_dirty_ring_size = ring_size; |
1526 | s->kvm_dirty_ring_bytes = ring_bytes; | |
1527 | ||
1528 | return 0; | |
1529 | } | |
1530 | ||
a01672d3 AK |
1531 | static void kvm_region_add(MemoryListener *listener, |
1532 | MemoryRegionSection *section) | |
1533 | { | |
7bbda04c | 1534 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); |
f39b7d2b DH |
1535 | KVMMemoryUpdate *update; |
1536 | ||
1537 | update = g_new0(KVMMemoryUpdate, 1); | |
1538 | update->section = *section; | |
7bbda04c | 1539 | |
f39b7d2b | 1540 | QSIMPLEQ_INSERT_TAIL(&kml->transaction_add, update, next); |
a01672d3 AK |
1541 | } |
1542 | ||
1543 | static void kvm_region_del(MemoryListener *listener, | |
1544 | MemoryRegionSection *section) | |
1545 | { | |
7bbda04c | 1546 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); |
f39b7d2b DH |
1547 | KVMMemoryUpdate *update; |
1548 | ||
1549 | update = g_new0(KVMMemoryUpdate, 1); | |
1550 | update->section = *section; | |
1551 | ||
1552 | QSIMPLEQ_INSERT_TAIL(&kml->transaction_del, update, next); | |
1553 | } | |
1554 | ||
1555 | static void kvm_region_commit(MemoryListener *listener) | |
1556 | { | |
1557 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, | |
1558 | listener); | |
1559 | KVMMemoryUpdate *u1, *u2; | |
1560 | bool need_inhibit = false; | |
1561 | ||
1562 | if (QSIMPLEQ_EMPTY(&kml->transaction_add) && | |
1563 | QSIMPLEQ_EMPTY(&kml->transaction_del)) { | |
1564 | return; | |
1565 | } | |
1566 | ||
1567 | /* | |
1568 | * We have to be careful when regions to add overlap with ranges to remove. | |
1569 | * We have to simulate atomic KVM memslot updates by making sure no ioctl() | |
1570 | * is currently active. | |
1571 | * | |
1572 | * The lists are order by addresses, so it's easy to find overlaps. | |
1573 | */ | |
1574 | u1 = QSIMPLEQ_FIRST(&kml->transaction_del); | |
1575 | u2 = QSIMPLEQ_FIRST(&kml->transaction_add); | |
1576 | while (u1 && u2) { | |
1577 | Range r1, r2; | |
1578 | ||
1579 | range_init_nofail(&r1, u1->section.offset_within_address_space, | |
1580 | int128_get64(u1->section.size)); | |
1581 | range_init_nofail(&r2, u2->section.offset_within_address_space, | |
1582 | int128_get64(u2->section.size)); | |
1583 | ||
1584 | if (range_overlaps_range(&r1, &r2)) { | |
1585 | need_inhibit = true; | |
1586 | break; | |
1587 | } | |
1588 | if (range_lob(&r1) < range_lob(&r2)) { | |
1589 | u1 = QSIMPLEQ_NEXT(u1, next); | |
1590 | } else { | |
1591 | u2 = QSIMPLEQ_NEXT(u2, next); | |
1592 | } | |
1593 | } | |
1594 | ||
1595 | kvm_slots_lock(); | |
1596 | if (need_inhibit) { | |
1597 | accel_ioctl_inhibit_begin(); | |
1598 | } | |
1599 | ||
1600 | /* Remove all memslots before adding the new ones. */ | |
1601 | while (!QSIMPLEQ_EMPTY(&kml->transaction_del)) { | |
1602 | u1 = QSIMPLEQ_FIRST(&kml->transaction_del); | |
1603 | QSIMPLEQ_REMOVE_HEAD(&kml->transaction_del, next); | |
7bbda04c | 1604 | |
f39b7d2b DH |
1605 | kvm_set_phys_mem(kml, &u1->section, false); |
1606 | memory_region_unref(u1->section.mr); | |
1607 | ||
1608 | g_free(u1); | |
1609 | } | |
1610 | while (!QSIMPLEQ_EMPTY(&kml->transaction_add)) { | |
1611 | u1 = QSIMPLEQ_FIRST(&kml->transaction_add); | |
1612 | QSIMPLEQ_REMOVE_HEAD(&kml->transaction_add, next); | |
1613 | ||
1614 | memory_region_ref(u1->section.mr); | |
1615 | kvm_set_phys_mem(kml, &u1->section, true); | |
1616 | ||
1617 | g_free(u1); | |
1618 | } | |
1619 | ||
1620 | if (need_inhibit) { | |
1621 | accel_ioctl_inhibit_end(); | |
1622 | } | |
1623 | kvm_slots_unlock(); | |
a01672d3 AK |
1624 | } |
1625 | ||
1626 | static void kvm_log_sync(MemoryListener *listener, | |
1627 | MemoryRegionSection *section) | |
7b8f3b78 | 1628 | { |
7bbda04c | 1629 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); |
a01672d3 | 1630 | |
a2f77862 | 1631 | kvm_slots_lock(); |
e65e5f50 | 1632 | kvm_physical_sync_dirty_bitmap(kml, section); |
a2f77862 | 1633 | kvm_slots_unlock(); |
7b8f3b78 MT |
1634 | } |
1635 | ||
1e493be5 | 1636 | static void kvm_log_sync_global(MemoryListener *l, bool last_stage) |
b4420f19 PX |
1637 | { |
1638 | KVMMemoryListener *kml = container_of(l, KVMMemoryListener, listener); | |
1639 | KVMState *s = kvm_state; | |
1640 | KVMSlot *mem; | |
1641 | int i; | |
1642 | ||
1643 | /* Flush all kernel dirty addresses into KVMSlot dirty bitmap */ | |
1644 | kvm_dirty_ring_flush(); | |
1645 | ||
1646 | /* | |
1647 | * TODO: make this faster when nr_slots is big while there are | |
1648 | * only a few used slots (small VMs). | |
1649 | */ | |
1650 | kvm_slots_lock(); | |
1651 | for (i = 0; i < s->nr_slots; i++) { | |
1652 | mem = &kml->slots[i]; | |
1653 | if (mem->memory_size && mem->flags & KVM_MEM_LOG_DIRTY_PAGES) { | |
1654 | kvm_slot_sync_dirty_pages(mem); | |
b20cc776 GS |
1655 | |
1656 | if (s->kvm_dirty_ring_with_bitmap && last_stage && | |
1657 | kvm_slot_get_dirty_log(s, mem)) { | |
1658 | kvm_slot_sync_dirty_pages(mem); | |
1659 | } | |
1660 | ||
b4420f19 PX |
1661 | /* |
1662 | * This is not needed by KVM_GET_DIRTY_LOG because the | |
1663 | * ioctl will unconditionally overwrite the whole region. | |
1664 | * However kvm dirty ring has no such side effect. | |
1665 | */ | |
1666 | kvm_slot_reset_dirty_pages(mem); | |
1667 | } | |
1668 | } | |
1669 | kvm_slots_unlock(); | |
1670 | } | |
1671 | ||
ff4aa114 PX |
1672 | static void kvm_log_clear(MemoryListener *listener, |
1673 | MemoryRegionSection *section) | |
1674 | { | |
1675 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); | |
1676 | int r; | |
1677 | ||
1678 | r = kvm_physical_log_clear(kml, section); | |
1679 | if (r < 0) { | |
1680 | error_report_once("%s: kvm log clear failed: mr=%s " | |
1681 | "offset=%"HWADDR_PRIx" size=%"PRIx64, __func__, | |
1682 | section->mr->name, section->offset_within_region, | |
1683 | int128_get64(section->size)); | |
1684 | abort(); | |
1685 | } | |
1686 | } | |
1687 | ||
d22b096e AK |
1688 | static void kvm_mem_ioeventfd_add(MemoryListener *listener, |
1689 | MemoryRegionSection *section, | |
1690 | bool match_data, uint64_t data, | |
1691 | EventNotifier *e) | |
1692 | { | |
1693 | int fd = event_notifier_get_fd(e); | |
80a1ea37 AK |
1694 | int r; |
1695 | ||
4b8f1c88 | 1696 | r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space, |
052e87b0 PB |
1697 | data, true, int128_get64(section->size), |
1698 | match_data); | |
80a1ea37 | 1699 | if (r < 0) { |
e346bcbf YK |
1700 | fprintf(stderr, "%s: error adding ioeventfd: %s (%d)\n", |
1701 | __func__, strerror(-r), -r); | |
80a1ea37 AK |
1702 | abort(); |
1703 | } | |
1704 | } | |
1705 | ||
d22b096e AK |
1706 | static void kvm_mem_ioeventfd_del(MemoryListener *listener, |
1707 | MemoryRegionSection *section, | |
1708 | bool match_data, uint64_t data, | |
1709 | EventNotifier *e) | |
80a1ea37 | 1710 | { |
d22b096e | 1711 | int fd = event_notifier_get_fd(e); |
80a1ea37 AK |
1712 | int r; |
1713 | ||
4b8f1c88 | 1714 | r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space, |
052e87b0 PB |
1715 | data, false, int128_get64(section->size), |
1716 | match_data); | |
80a1ea37 | 1717 | if (r < 0) { |
e346bcbf YK |
1718 | fprintf(stderr, "%s: error deleting ioeventfd: %s (%d)\n", |
1719 | __func__, strerror(-r), -r); | |
80a1ea37 AK |
1720 | abort(); |
1721 | } | |
1722 | } | |
1723 | ||
d22b096e AK |
1724 | static void kvm_io_ioeventfd_add(MemoryListener *listener, |
1725 | MemoryRegionSection *section, | |
1726 | bool match_data, uint64_t data, | |
1727 | EventNotifier *e) | |
80a1ea37 | 1728 | { |
d22b096e | 1729 | int fd = event_notifier_get_fd(e); |
80a1ea37 AK |
1730 | int r; |
1731 | ||
44c3f8f7 | 1732 | r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space, |
052e87b0 PB |
1733 | data, true, int128_get64(section->size), |
1734 | match_data); | |
80a1ea37 | 1735 | if (r < 0) { |
e346bcbf YK |
1736 | fprintf(stderr, "%s: error adding ioeventfd: %s (%d)\n", |
1737 | __func__, strerror(-r), -r); | |
80a1ea37 AK |
1738 | abort(); |
1739 | } | |
1740 | } | |
1741 | ||
d22b096e AK |
1742 | static void kvm_io_ioeventfd_del(MemoryListener *listener, |
1743 | MemoryRegionSection *section, | |
1744 | bool match_data, uint64_t data, | |
1745 | EventNotifier *e) | |
80a1ea37 AK |
1746 | |
1747 | { | |
d22b096e | 1748 | int fd = event_notifier_get_fd(e); |
80a1ea37 AK |
1749 | int r; |
1750 | ||
44c3f8f7 | 1751 | r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space, |
052e87b0 PB |
1752 | data, false, int128_get64(section->size), |
1753 | match_data); | |
80a1ea37 | 1754 | if (r < 0) { |
e346bcbf YK |
1755 | fprintf(stderr, "%s: error deleting ioeventfd: %s (%d)\n", |
1756 | __func__, strerror(-r), -r); | |
80a1ea37 AK |
1757 | abort(); |
1758 | } | |
1759 | } | |
1760 | ||
38bfe691 | 1761 | void kvm_memory_listener_register(KVMState *s, KVMMemoryListener *kml, |
142518bd | 1762 | AddressSpace *as, int as_id, const char *name) |
7bbda04c PB |
1763 | { |
1764 | int i; | |
1765 | ||
b21e2380 | 1766 | kml->slots = g_new0(KVMSlot, s->nr_slots); |
38bfe691 | 1767 | kml->as_id = as_id; |
7bbda04c PB |
1768 | |
1769 | for (i = 0; i < s->nr_slots; i++) { | |
1770 | kml->slots[i].slot = i; | |
1771 | } | |
1772 | ||
f39b7d2b DH |
1773 | QSIMPLEQ_INIT(&kml->transaction_add); |
1774 | QSIMPLEQ_INIT(&kml->transaction_del); | |
1775 | ||
7bbda04c PB |
1776 | kml->listener.region_add = kvm_region_add; |
1777 | kml->listener.region_del = kvm_region_del; | |
f39b7d2b | 1778 | kml->listener.commit = kvm_region_commit; |
7bbda04c PB |
1779 | kml->listener.log_start = kvm_log_start; |
1780 | kml->listener.log_stop = kvm_log_stop; | |
5369a36c | 1781 | kml->listener.priority = MEMORY_LISTENER_PRIORITY_ACCEL; |
142518bd | 1782 | kml->listener.name = name; |
7bbda04c | 1783 | |
b4420f19 PX |
1784 | if (s->kvm_dirty_ring_size) { |
1785 | kml->listener.log_sync_global = kvm_log_sync_global; | |
1786 | } else { | |
1787 | kml->listener.log_sync = kvm_log_sync; | |
1788 | kml->listener.log_clear = kvm_log_clear; | |
1789 | } | |
1790 | ||
7bbda04c | 1791 | memory_listener_register(&kml->listener, as); |
8072aae3 AK |
1792 | |
1793 | for (i = 0; i < s->nr_as; ++i) { | |
1794 | if (!s->as[i].as) { | |
1795 | s->as[i].as = as; | |
1796 | s->as[i].ml = kml; | |
1797 | break; | |
1798 | } | |
1799 | } | |
7bbda04c | 1800 | } |
d22b096e AK |
1801 | |
1802 | static MemoryListener kvm_io_listener = { | |
142518bd | 1803 | .name = "kvm-io", |
d22b096e AK |
1804 | .eventfd_add = kvm_io_ioeventfd_add, |
1805 | .eventfd_del = kvm_io_ioeventfd_del, | |
8be0461d | 1806 | .priority = MEMORY_LISTENER_PRIORITY_DEV_BACKEND, |
7b8f3b78 MT |
1807 | }; |
1808 | ||
3889c3fa | 1809 | int kvm_set_irq(KVMState *s, int irq, int level) |
84b058d7 JK |
1810 | { |
1811 | struct kvm_irq_level event; | |
1812 | int ret; | |
1813 | ||
7ae26bd4 | 1814 | assert(kvm_async_interrupts_enabled()); |
84b058d7 JK |
1815 | |
1816 | event.level = level; | |
1817 | event.irq = irq; | |
e333cd69 | 1818 | ret = kvm_vm_ioctl(s, s->irq_set_ioctl, &event); |
84b058d7 | 1819 | if (ret < 0) { |
3889c3fa | 1820 | perror("kvm_set_irq"); |
84b058d7 JK |
1821 | abort(); |
1822 | } | |
1823 | ||
e333cd69 | 1824 | return (s->irq_set_ioctl == KVM_IRQ_LINE) ? 1 : event.status; |
84b058d7 JK |
1825 | } |
1826 | ||
1827 | #ifdef KVM_CAP_IRQ_ROUTING | |
d3d3bef0 JK |
1828 | typedef struct KVMMSIRoute { |
1829 | struct kvm_irq_routing_entry kroute; | |
1830 | QTAILQ_ENTRY(KVMMSIRoute) entry; | |
1831 | } KVMMSIRoute; | |
1832 | ||
84b058d7 JK |
1833 | static void set_gsi(KVMState *s, unsigned int gsi) |
1834 | { | |
8269fb70 | 1835 | set_bit(gsi, s->used_gsi_bitmap); |
84b058d7 JK |
1836 | } |
1837 | ||
04fa27f5 JK |
1838 | static void clear_gsi(KVMState *s, unsigned int gsi) |
1839 | { | |
8269fb70 | 1840 | clear_bit(gsi, s->used_gsi_bitmap); |
04fa27f5 JK |
1841 | } |
1842 | ||
7b774593 | 1843 | void kvm_init_irq_routing(KVMState *s) |
84b058d7 | 1844 | { |
04fa27f5 | 1845 | int gsi_count, i; |
84b058d7 | 1846 | |
00008418 | 1847 | gsi_count = kvm_check_extension(s, KVM_CAP_IRQ_ROUTING) - 1; |
84b058d7 | 1848 | if (gsi_count > 0) { |
84b058d7 | 1849 | /* Round up so we can search ints using ffs */ |
8269fb70 | 1850 | s->used_gsi_bitmap = bitmap_new(gsi_count); |
4e2e4e63 | 1851 | s->gsi_count = gsi_count; |
84b058d7 JK |
1852 | } |
1853 | ||
1854 | s->irq_routes = g_malloc0(sizeof(*s->irq_routes)); | |
1855 | s->nr_allocated_irq_routes = 0; | |
1856 | ||
50bf31b9 | 1857 | if (!kvm_direct_msi_allowed) { |
4a3adebb JK |
1858 | for (i = 0; i < KVM_MSI_HASHTAB_SIZE; i++) { |
1859 | QTAILQ_INIT(&s->msi_hashtab[i]); | |
1860 | } | |
04fa27f5 JK |
1861 | } |
1862 | ||
84b058d7 JK |
1863 | kvm_arch_init_irq_routing(s); |
1864 | } | |
1865 | ||
cb925cf9 | 1866 | void kvm_irqchip_commit_routes(KVMState *s) |
e7b20308 JK |
1867 | { |
1868 | int ret; | |
1869 | ||
7005f7f8 PX |
1870 | if (kvm_gsi_direct_mapping()) { |
1871 | return; | |
1872 | } | |
1873 | ||
1874 | if (!kvm_gsi_routing_enabled()) { | |
1875 | return; | |
1876 | } | |
1877 | ||
e7b20308 | 1878 | s->irq_routes->flags = 0; |
54a6c11b | 1879 | trace_kvm_irqchip_commit_routes(); |
e7b20308 JK |
1880 | ret = kvm_vm_ioctl(s, KVM_SET_GSI_ROUTING, s->irq_routes); |
1881 | assert(ret == 0); | |
1882 | } | |
1883 | ||
84b058d7 JK |
1884 | static void kvm_add_routing_entry(KVMState *s, |
1885 | struct kvm_irq_routing_entry *entry) | |
1886 | { | |
1887 | struct kvm_irq_routing_entry *new; | |
1888 | int n, size; | |
1889 | ||
1890 | if (s->irq_routes->nr == s->nr_allocated_irq_routes) { | |
1891 | n = s->nr_allocated_irq_routes * 2; | |
1892 | if (n < 64) { | |
1893 | n = 64; | |
1894 | } | |
1895 | size = sizeof(struct kvm_irq_routing); | |
1896 | size += n * sizeof(*new); | |
1897 | s->irq_routes = g_realloc(s->irq_routes, size); | |
1898 | s->nr_allocated_irq_routes = n; | |
1899 | } | |
1900 | n = s->irq_routes->nr++; | |
1901 | new = &s->irq_routes->entries[n]; | |
0fbc2074 MT |
1902 | |
1903 | *new = *entry; | |
84b058d7 JK |
1904 | |
1905 | set_gsi(s, entry->gsi); | |
1906 | } | |
1907 | ||
cc57407e JK |
1908 | static int kvm_update_routing_entry(KVMState *s, |
1909 | struct kvm_irq_routing_entry *new_entry) | |
1910 | { | |
1911 | struct kvm_irq_routing_entry *entry; | |
1912 | int n; | |
1913 | ||
1914 | for (n = 0; n < s->irq_routes->nr; n++) { | |
1915 | entry = &s->irq_routes->entries[n]; | |
1916 | if (entry->gsi != new_entry->gsi) { | |
1917 | continue; | |
1918 | } | |
1919 | ||
40509f7f MT |
1920 | if(!memcmp(entry, new_entry, sizeof *entry)) { |
1921 | return 0; | |
1922 | } | |
1923 | ||
0fbc2074 | 1924 | *entry = *new_entry; |
cc57407e | 1925 | |
cc57407e JK |
1926 | return 0; |
1927 | } | |
1928 | ||
1929 | return -ESRCH; | |
1930 | } | |
1931 | ||
1df186df | 1932 | void kvm_irqchip_add_irq_route(KVMState *s, int irq, int irqchip, int pin) |
84b058d7 | 1933 | { |
0fbc2074 | 1934 | struct kvm_irq_routing_entry e = {}; |
84b058d7 | 1935 | |
4e2e4e63 JK |
1936 | assert(pin < s->gsi_count); |
1937 | ||
84b058d7 JK |
1938 | e.gsi = irq; |
1939 | e.type = KVM_IRQ_ROUTING_IRQCHIP; | |
1940 | e.flags = 0; | |
1941 | e.u.irqchip.irqchip = irqchip; | |
1942 | e.u.irqchip.pin = pin; | |
1943 | kvm_add_routing_entry(s, &e); | |
1944 | } | |
1945 | ||
1e2aa8be | 1946 | void kvm_irqchip_release_virq(KVMState *s, int virq) |
04fa27f5 JK |
1947 | { |
1948 | struct kvm_irq_routing_entry *e; | |
1949 | int i; | |
1950 | ||
76fe21de AK |
1951 | if (kvm_gsi_direct_mapping()) { |
1952 | return; | |
1953 | } | |
1954 | ||
04fa27f5 JK |
1955 | for (i = 0; i < s->irq_routes->nr; i++) { |
1956 | e = &s->irq_routes->entries[i]; | |
1957 | if (e->gsi == virq) { | |
1958 | s->irq_routes->nr--; | |
1959 | *e = s->irq_routes->entries[s->irq_routes->nr]; | |
1960 | } | |
1961 | } | |
1962 | clear_gsi(s, virq); | |
38d87493 | 1963 | kvm_arch_release_virq_post(virq); |
9ba35d0b | 1964 | trace_kvm_irqchip_release_virq(virq); |
04fa27f5 JK |
1965 | } |
1966 | ||
3607715a DG |
1967 | void kvm_irqchip_add_change_notifier(Notifier *n) |
1968 | { | |
1969 | notifier_list_add(&kvm_irqchip_change_notifiers, n); | |
1970 | } | |
1971 | ||
1972 | void kvm_irqchip_remove_change_notifier(Notifier *n) | |
1973 | { | |
1974 | notifier_remove(n); | |
1975 | } | |
1976 | ||
1977 | void kvm_irqchip_change_notify(void) | |
1978 | { | |
1979 | notifier_list_notify(&kvm_irqchip_change_notifiers, NULL); | |
1980 | } | |
1981 | ||
04fa27f5 JK |
1982 | static unsigned int kvm_hash_msi(uint32_t data) |
1983 | { | |
1984 | /* This is optimized for IA32 MSI layout. However, no other arch shall | |
1985 | * repeat the mistake of not providing a direct MSI injection API. */ | |
1986 | return data & 0xff; | |
1987 | } | |
1988 | ||
1989 | static void kvm_flush_dynamic_msi_routes(KVMState *s) | |
1990 | { | |
1991 | KVMMSIRoute *route, *next; | |
1992 | unsigned int hash; | |
1993 | ||
1994 | for (hash = 0; hash < KVM_MSI_HASHTAB_SIZE; hash++) { | |
1995 | QTAILQ_FOREACH_SAFE(route, &s->msi_hashtab[hash], entry, next) { | |
1996 | kvm_irqchip_release_virq(s, route->kroute.gsi); | |
1997 | QTAILQ_REMOVE(&s->msi_hashtab[hash], route, entry); | |
1998 | g_free(route); | |
1999 | } | |
2000 | } | |
2001 | } | |
2002 | ||
2003 | static int kvm_irqchip_get_virq(KVMState *s) | |
2004 | { | |
8269fb70 | 2005 | int next_virq; |
04fa27f5 | 2006 | |
bdf02631 WM |
2007 | /* |
2008 | * PIC and IOAPIC share the first 16 GSI numbers, thus the available | |
2009 | * GSI numbers are more than the number of IRQ route. Allocating a GSI | |
2010 | * number can succeed even though a new route entry cannot be added. | |
2011 | * When this happens, flush dynamic MSI entries to free IRQ route entries. | |
2012 | */ | |
50bf31b9 | 2013 | if (!kvm_direct_msi_allowed && s->irq_routes->nr == s->gsi_count) { |
bdf02631 WM |
2014 | kvm_flush_dynamic_msi_routes(s); |
2015 | } | |
2016 | ||
04fa27f5 | 2017 | /* Return the lowest unused GSI in the bitmap */ |
8269fb70 WY |
2018 | next_virq = find_first_zero_bit(s->used_gsi_bitmap, s->gsi_count); |
2019 | if (next_virq >= s->gsi_count) { | |
2020 | return -ENOSPC; | |
2021 | } else { | |
2022 | return next_virq; | |
04fa27f5 | 2023 | } |
04fa27f5 JK |
2024 | } |
2025 | ||
2026 | static KVMMSIRoute *kvm_lookup_msi_route(KVMState *s, MSIMessage msg) | |
2027 | { | |
2028 | unsigned int hash = kvm_hash_msi(msg.data); | |
2029 | KVMMSIRoute *route; | |
2030 | ||
2031 | QTAILQ_FOREACH(route, &s->msi_hashtab[hash], entry) { | |
2032 | if (route->kroute.u.msi.address_lo == (uint32_t)msg.address && | |
2033 | route->kroute.u.msi.address_hi == (msg.address >> 32) && | |
d07cc1f1 | 2034 | route->kroute.u.msi.data == le32_to_cpu(msg.data)) { |
04fa27f5 JK |
2035 | return route; |
2036 | } | |
2037 | } | |
2038 | return NULL; | |
2039 | } | |
2040 | ||
2041 | int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg) | |
2042 | { | |
4a3adebb | 2043 | struct kvm_msi msi; |
04fa27f5 JK |
2044 | KVMMSIRoute *route; |
2045 | ||
50bf31b9 | 2046 | if (kvm_direct_msi_allowed) { |
4a3adebb JK |
2047 | msi.address_lo = (uint32_t)msg.address; |
2048 | msi.address_hi = msg.address >> 32; | |
d07cc1f1 | 2049 | msi.data = le32_to_cpu(msg.data); |
4a3adebb JK |
2050 | msi.flags = 0; |
2051 | memset(msi.pad, 0, sizeof(msi.pad)); | |
2052 | ||
2053 | return kvm_vm_ioctl(s, KVM_SIGNAL_MSI, &msi); | |
2054 | } | |
2055 | ||
04fa27f5 JK |
2056 | route = kvm_lookup_msi_route(s, msg); |
2057 | if (!route) { | |
e7b20308 | 2058 | int virq; |
04fa27f5 JK |
2059 | |
2060 | virq = kvm_irqchip_get_virq(s); | |
2061 | if (virq < 0) { | |
2062 | return virq; | |
2063 | } | |
2064 | ||
b21e2380 | 2065 | route = g_new0(KVMMSIRoute, 1); |
04fa27f5 JK |
2066 | route->kroute.gsi = virq; |
2067 | route->kroute.type = KVM_IRQ_ROUTING_MSI; | |
2068 | route->kroute.flags = 0; | |
2069 | route->kroute.u.msi.address_lo = (uint32_t)msg.address; | |
2070 | route->kroute.u.msi.address_hi = msg.address >> 32; | |
d07cc1f1 | 2071 | route->kroute.u.msi.data = le32_to_cpu(msg.data); |
04fa27f5 JK |
2072 | |
2073 | kvm_add_routing_entry(s, &route->kroute); | |
cb925cf9 | 2074 | kvm_irqchip_commit_routes(s); |
04fa27f5 JK |
2075 | |
2076 | QTAILQ_INSERT_TAIL(&s->msi_hashtab[kvm_hash_msi(msg.data)], route, | |
2077 | entry); | |
04fa27f5 JK |
2078 | } |
2079 | ||
2080 | assert(route->kroute.type == KVM_IRQ_ROUTING_MSI); | |
2081 | ||
3889c3fa | 2082 | return kvm_set_irq(s, route->kroute.gsi, 1); |
04fa27f5 JK |
2083 | } |
2084 | ||
def4c557 | 2085 | int kvm_irqchip_add_msi_route(KVMRouteChange *c, int vector, PCIDevice *dev) |
92b4e489 | 2086 | { |
0fbc2074 | 2087 | struct kvm_irq_routing_entry kroute = {}; |
92b4e489 | 2088 | int virq; |
def4c557 | 2089 | KVMState *s = c->s; |
d1f6af6a PX |
2090 | MSIMessage msg = {0, 0}; |
2091 | ||
88c725c7 | 2092 | if (pci_available && dev) { |
e1d4fb2d | 2093 | msg = pci_get_msi_message(dev, vector); |
d1f6af6a | 2094 | } |
92b4e489 | 2095 | |
76fe21de | 2096 | if (kvm_gsi_direct_mapping()) { |
1850b6b7 | 2097 | return kvm_arch_msi_data_to_gsi(msg.data); |
76fe21de AK |
2098 | } |
2099 | ||
f3e1bed8 | 2100 | if (!kvm_gsi_routing_enabled()) { |
92b4e489 JK |
2101 | return -ENOSYS; |
2102 | } | |
2103 | ||
2104 | virq = kvm_irqchip_get_virq(s); | |
2105 | if (virq < 0) { | |
2106 | return virq; | |
2107 | } | |
2108 | ||
2109 | kroute.gsi = virq; | |
2110 | kroute.type = KVM_IRQ_ROUTING_MSI; | |
2111 | kroute.flags = 0; | |
2112 | kroute.u.msi.address_lo = (uint32_t)msg.address; | |
2113 | kroute.u.msi.address_hi = msg.address >> 32; | |
d07cc1f1 | 2114 | kroute.u.msi.data = le32_to_cpu(msg.data); |
88c725c7 | 2115 | if (pci_available && kvm_msi_devid_required()) { |
767a554a PF |
2116 | kroute.flags = KVM_MSI_VALID_DEVID; |
2117 | kroute.u.msi.devid = pci_requester_id(dev); | |
2118 | } | |
dc9f06ca | 2119 | if (kvm_arch_fixup_msi_route(&kroute, msg.address, msg.data, dev)) { |
9e03a040 FB |
2120 | kvm_irqchip_release_virq(s, virq); |
2121 | return -EINVAL; | |
2122 | } | |
92b4e489 | 2123 | |
9ba35d0b PX |
2124 | trace_kvm_irqchip_add_msi_route(dev ? dev->name : (char *)"N/A", |
2125 | vector, virq); | |
54a6c11b | 2126 | |
92b4e489 | 2127 | kvm_add_routing_entry(s, &kroute); |
38d87493 | 2128 | kvm_arch_add_msi_route_post(&kroute, vector, dev); |
def4c557 | 2129 | c->changes++; |
92b4e489 JK |
2130 | |
2131 | return virq; | |
2132 | } | |
2133 | ||
dc9f06ca PF |
2134 | int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg, |
2135 | PCIDevice *dev) | |
cc57407e | 2136 | { |
0fbc2074 | 2137 | struct kvm_irq_routing_entry kroute = {}; |
cc57407e | 2138 | |
76fe21de AK |
2139 | if (kvm_gsi_direct_mapping()) { |
2140 | return 0; | |
2141 | } | |
2142 | ||
cc57407e JK |
2143 | if (!kvm_irqchip_in_kernel()) { |
2144 | return -ENOSYS; | |
2145 | } | |
2146 | ||
2147 | kroute.gsi = virq; | |
2148 | kroute.type = KVM_IRQ_ROUTING_MSI; | |
2149 | kroute.flags = 0; | |
2150 | kroute.u.msi.address_lo = (uint32_t)msg.address; | |
2151 | kroute.u.msi.address_hi = msg.address >> 32; | |
d07cc1f1 | 2152 | kroute.u.msi.data = le32_to_cpu(msg.data); |
88c725c7 | 2153 | if (pci_available && kvm_msi_devid_required()) { |
767a554a PF |
2154 | kroute.flags = KVM_MSI_VALID_DEVID; |
2155 | kroute.u.msi.devid = pci_requester_id(dev); | |
2156 | } | |
dc9f06ca | 2157 | if (kvm_arch_fixup_msi_route(&kroute, msg.address, msg.data, dev)) { |
9e03a040 FB |
2158 | return -EINVAL; |
2159 | } | |
cc57407e | 2160 | |
54a6c11b PX |
2161 | trace_kvm_irqchip_update_msi_route(virq); |
2162 | ||
cc57407e JK |
2163 | return kvm_update_routing_entry(s, &kroute); |
2164 | } | |
2165 | ||
ff66ba87 PX |
2166 | static int kvm_irqchip_assign_irqfd(KVMState *s, EventNotifier *event, |
2167 | EventNotifier *resample, int virq, | |
ca916d37 | 2168 | bool assign) |
39853bbc | 2169 | { |
ff66ba87 PX |
2170 | int fd = event_notifier_get_fd(event); |
2171 | int rfd = resample ? event_notifier_get_fd(resample) : -1; | |
2172 | ||
39853bbc JK |
2173 | struct kvm_irqfd irqfd = { |
2174 | .fd = fd, | |
2175 | .gsi = virq, | |
2176 | .flags = assign ? 0 : KVM_IRQFD_FLAG_DEASSIGN, | |
2177 | }; | |
2178 | ||
ca916d37 | 2179 | if (rfd != -1) { |
c82d9d43 PX |
2180 | assert(assign); |
2181 | if (kvm_irqchip_is_split()) { | |
2182 | /* | |
2183 | * When the slow irqchip (e.g. IOAPIC) is in the | |
2184 | * userspace, KVM kernel resamplefd will not work because | |
2185 | * the EOI of the interrupt will be delivered to userspace | |
2186 | * instead, so the KVM kernel resamplefd kick will be | |
2187 | * skipped. The userspace here mimics what the kernel | |
2188 | * provides with resamplefd, remember the resamplefd and | |
2189 | * kick it when we receive EOI of this IRQ. | |
2190 | * | |
2191 | * This is hackery because IOAPIC is mostly bypassed | |
2192 | * (except EOI broadcasts) when irqfd is used. However | |
2193 | * this can bring much performance back for split irqchip | |
2194 | * with INTx IRQs (for VFIO, this gives 93% perf of the | |
2195 | * full fast path, which is 46% perf boost comparing to | |
2196 | * the INTx slow path). | |
2197 | */ | |
2198 | kvm_resample_fd_insert(virq, resample); | |
2199 | } else { | |
2200 | irqfd.flags |= KVM_IRQFD_FLAG_RESAMPLE; | |
2201 | irqfd.resamplefd = rfd; | |
2202 | } | |
2203 | } else if (!assign) { | |
2204 | if (kvm_irqchip_is_split()) { | |
2205 | kvm_resample_fd_remove(virq); | |
2206 | } | |
ca916d37 VM |
2207 | } |
2208 | ||
cc7e0ddf | 2209 | if (!kvm_irqfds_enabled()) { |
39853bbc JK |
2210 | return -ENOSYS; |
2211 | } | |
2212 | ||
2213 | return kvm_vm_ioctl(s, KVM_IRQFD, &irqfd); | |
2214 | } | |
2215 | ||
d426d9fb CH |
2216 | int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter) |
2217 | { | |
e9af2fef | 2218 | struct kvm_irq_routing_entry kroute = {}; |
d426d9fb CH |
2219 | int virq; |
2220 | ||
2221 | if (!kvm_gsi_routing_enabled()) { | |
2222 | return -ENOSYS; | |
2223 | } | |
2224 | ||
2225 | virq = kvm_irqchip_get_virq(s); | |
2226 | if (virq < 0) { | |
2227 | return virq; | |
2228 | } | |
2229 | ||
2230 | kroute.gsi = virq; | |
2231 | kroute.type = KVM_IRQ_ROUTING_S390_ADAPTER; | |
2232 | kroute.flags = 0; | |
2233 | kroute.u.adapter.summary_addr = adapter->summary_addr; | |
2234 | kroute.u.adapter.ind_addr = adapter->ind_addr; | |
2235 | kroute.u.adapter.summary_offset = adapter->summary_offset; | |
2236 | kroute.u.adapter.ind_offset = adapter->ind_offset; | |
2237 | kroute.u.adapter.adapter_id = adapter->adapter_id; | |
2238 | ||
2239 | kvm_add_routing_entry(s, &kroute); | |
d426d9fb CH |
2240 | |
2241 | return virq; | |
2242 | } | |
2243 | ||
977a8d9c AS |
2244 | int kvm_irqchip_add_hv_sint_route(KVMState *s, uint32_t vcpu, uint32_t sint) |
2245 | { | |
2246 | struct kvm_irq_routing_entry kroute = {}; | |
2247 | int virq; | |
2248 | ||
2249 | if (!kvm_gsi_routing_enabled()) { | |
2250 | return -ENOSYS; | |
2251 | } | |
2252 | if (!kvm_check_extension(s, KVM_CAP_HYPERV_SYNIC)) { | |
2253 | return -ENOSYS; | |
2254 | } | |
2255 | virq = kvm_irqchip_get_virq(s); | |
2256 | if (virq < 0) { | |
2257 | return virq; | |
2258 | } | |
2259 | ||
2260 | kroute.gsi = virq; | |
2261 | kroute.type = KVM_IRQ_ROUTING_HV_SINT; | |
2262 | kroute.flags = 0; | |
2263 | kroute.u.hv_sint.vcpu = vcpu; | |
2264 | kroute.u.hv_sint.sint = sint; | |
2265 | ||
2266 | kvm_add_routing_entry(s, &kroute); | |
2267 | kvm_irqchip_commit_routes(s); | |
2268 | ||
2269 | return virq; | |
2270 | } | |
2271 | ||
84b058d7 JK |
2272 | #else /* !KVM_CAP_IRQ_ROUTING */ |
2273 | ||
7b774593 | 2274 | void kvm_init_irq_routing(KVMState *s) |
84b058d7 JK |
2275 | { |
2276 | } | |
04fa27f5 | 2277 | |
d3d3bef0 JK |
2278 | void kvm_irqchip_release_virq(KVMState *s, int virq) |
2279 | { | |
2280 | } | |
2281 | ||
04fa27f5 JK |
2282 | int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg) |
2283 | { | |
2284 | abort(); | |
2285 | } | |
92b4e489 | 2286 | |
def4c557 | 2287 | int kvm_irqchip_add_msi_route(KVMRouteChange *c, int vector, PCIDevice *dev) |
92b4e489 | 2288 | { |
df410675 | 2289 | return -ENOSYS; |
92b4e489 | 2290 | } |
39853bbc | 2291 | |
d426d9fb CH |
2292 | int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter) |
2293 | { | |
2294 | return -ENOSYS; | |
2295 | } | |
2296 | ||
977a8d9c AS |
2297 | int kvm_irqchip_add_hv_sint_route(KVMState *s, uint32_t vcpu, uint32_t sint) |
2298 | { | |
2299 | return -ENOSYS; | |
2300 | } | |
2301 | ||
ff66ba87 PX |
2302 | static int kvm_irqchip_assign_irqfd(KVMState *s, EventNotifier *event, |
2303 | EventNotifier *resample, int virq, | |
2304 | bool assign) | |
39853bbc JK |
2305 | { |
2306 | abort(); | |
2307 | } | |
dabe3143 MT |
2308 | |
2309 | int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg) | |
2310 | { | |
2311 | return -ENOSYS; | |
2312 | } | |
84b058d7 JK |
2313 | #endif /* !KVM_CAP_IRQ_ROUTING */ |
2314 | ||
1c9b71a7 EA |
2315 | int kvm_irqchip_add_irqfd_notifier_gsi(KVMState *s, EventNotifier *n, |
2316 | EventNotifier *rn, int virq) | |
39853bbc | 2317 | { |
ff66ba87 | 2318 | return kvm_irqchip_assign_irqfd(s, n, rn, virq, true); |
39853bbc JK |
2319 | } |
2320 | ||
1c9b71a7 EA |
2321 | int kvm_irqchip_remove_irqfd_notifier_gsi(KVMState *s, EventNotifier *n, |
2322 | int virq) | |
15b2bd18 | 2323 | { |
ff66ba87 | 2324 | return kvm_irqchip_assign_irqfd(s, n, NULL, virq, false); |
15b2bd18 PB |
2325 | } |
2326 | ||
197e3524 EA |
2327 | int kvm_irqchip_add_irqfd_notifier(KVMState *s, EventNotifier *n, |
2328 | EventNotifier *rn, qemu_irq irq) | |
2329 | { | |
2330 | gpointer key, gsi; | |
2331 | gboolean found = g_hash_table_lookup_extended(s->gsimap, irq, &key, &gsi); | |
2332 | ||
2333 | if (!found) { | |
2334 | return -ENXIO; | |
2335 | } | |
2336 | return kvm_irqchip_add_irqfd_notifier_gsi(s, n, rn, GPOINTER_TO_INT(gsi)); | |
2337 | } | |
2338 | ||
2339 | int kvm_irqchip_remove_irqfd_notifier(KVMState *s, EventNotifier *n, | |
2340 | qemu_irq irq) | |
2341 | { | |
2342 | gpointer key, gsi; | |
2343 | gboolean found = g_hash_table_lookup_extended(s->gsimap, irq, &key, &gsi); | |
2344 | ||
2345 | if (!found) { | |
2346 | return -ENXIO; | |
2347 | } | |
2348 | return kvm_irqchip_remove_irqfd_notifier_gsi(s, n, GPOINTER_TO_INT(gsi)); | |
2349 | } | |
2350 | ||
2351 | void kvm_irqchip_set_qemuirq_gsi(KVMState *s, qemu_irq irq, int gsi) | |
2352 | { | |
2353 | g_hash_table_insert(s->gsimap, irq, GINT_TO_POINTER(gsi)); | |
2354 | } | |
2355 | ||
4376c40d | 2356 | static void kvm_irqchip_create(KVMState *s) |
84b058d7 | 2357 | { |
84b058d7 JK |
2358 | int ret; |
2359 | ||
d1972be1 | 2360 | assert(s->kernel_irqchip_split != ON_OFF_AUTO_AUTO); |
8db4936b PB |
2361 | if (kvm_check_extension(s, KVM_CAP_IRQCHIP)) { |
2362 | ; | |
2363 | } else if (kvm_check_extension(s, KVM_CAP_S390_IRQCHIP)) { | |
2364 | ret = kvm_vm_enable_cap(s, KVM_CAP_S390_IRQCHIP, 0); | |
2365 | if (ret < 0) { | |
2366 | fprintf(stderr, "Enable kernel irqchip failed: %s\n", strerror(-ret)); | |
2367 | exit(1); | |
2368 | } | |
2369 | } else { | |
2370 | return; | |
84b058d7 JK |
2371 | } |
2372 | ||
d6032e06 CD |
2373 | /* First probe and see if there's a arch-specific hook to create the |
2374 | * in-kernel irqchip for us */ | |
4376c40d | 2375 | ret = kvm_arch_irqchip_create(s); |
8db4936b | 2376 | if (ret == 0) { |
d1972be1 | 2377 | if (s->kernel_irqchip_split == ON_OFF_AUTO_ON) { |
47c182fe | 2378 | error_report("Split IRQ chip mode not supported."); |
15eafc2e PB |
2379 | exit(1); |
2380 | } else { | |
2381 | ret = kvm_vm_ioctl(s, KVM_CREATE_IRQCHIP); | |
2382 | } | |
8db4936b PB |
2383 | } |
2384 | if (ret < 0) { | |
2385 | fprintf(stderr, "Create kernel irqchip failed: %s\n", strerror(-ret)); | |
2386 | exit(1); | |
84b058d7 JK |
2387 | } |
2388 | ||
3d4b2649 | 2389 | kvm_kernel_irqchip = true; |
7ae26bd4 PM |
2390 | /* If we have an in-kernel IRQ chip then we must have asynchronous |
2391 | * interrupt delivery (though the reverse is not necessarily true) | |
2392 | */ | |
2393 | kvm_async_interrupts_allowed = true; | |
215e79c0 | 2394 | kvm_halt_in_kernel_allowed = true; |
84b058d7 JK |
2395 | |
2396 | kvm_init_irq_routing(s); | |
2397 | ||
197e3524 | 2398 | s->gsimap = g_hash_table_new(g_direct_hash, g_direct_equal); |
84b058d7 JK |
2399 | } |
2400 | ||
670436ce AJ |
2401 | /* Find number of supported CPUs using the recommended |
2402 | * procedure from the kernel API documentation to cope with | |
2403 | * older kernels that may be missing capabilities. | |
2404 | */ | |
2405 | static int kvm_recommended_vcpus(KVMState *s) | |
3ed444e9 | 2406 | { |
11748ba7 | 2407 | int ret = kvm_vm_check_extension(s, KVM_CAP_NR_VCPUS); |
670436ce AJ |
2408 | return (ret) ? ret : 4; |
2409 | } | |
3ed444e9 | 2410 | |
670436ce AJ |
2411 | static int kvm_max_vcpus(KVMState *s) |
2412 | { | |
2413 | int ret = kvm_check_extension(s, KVM_CAP_MAX_VCPUS); | |
2414 | return (ret) ? ret : kvm_recommended_vcpus(s); | |
3ed444e9 DH |
2415 | } |
2416 | ||
f31e3266 GK |
2417 | static int kvm_max_vcpu_id(KVMState *s) |
2418 | { | |
2419 | int ret = kvm_check_extension(s, KVM_CAP_MAX_VCPU_ID); | |
2420 | return (ret) ? ret : kvm_max_vcpus(s); | |
2421 | } | |
2422 | ||
41264b38 GK |
2423 | bool kvm_vcpu_id_is_valid(int vcpu_id) |
2424 | { | |
4f7f5893 | 2425 | KVMState *s = KVM_STATE(current_accel()); |
f31e3266 | 2426 | return vcpu_id >= 0 && vcpu_id < kvm_max_vcpu_id(s); |
41264b38 GK |
2427 | } |
2428 | ||
7786ae40 HH |
2429 | bool kvm_dirty_ring_enabled(void) |
2430 | { | |
2431 | return kvm_state->kvm_dirty_ring_size ? true : false; | |
2432 | } | |
2433 | ||
467ef823 | 2434 | static void query_stats_cb(StatsResultList **result, StatsTarget target, |
cf7405bc | 2435 | strList *names, strList *targets, Error **errp); |
cc01a3f4 MK |
2436 | static void query_stats_schemas_cb(StatsSchemaList **result, Error **errp); |
2437 | ||
4a06a7cc HH |
2438 | uint32_t kvm_dirty_ring_size(void) |
2439 | { | |
2440 | return kvm_state->kvm_dirty_ring_size; | |
2441 | } | |
2442 | ||
f6a1ef64 | 2443 | static int kvm_init(MachineState *ms) |
05330448 | 2444 | { |
f6a1ef64 | 2445 | MachineClass *mc = MACHINE_GET_CLASS(ms); |
168ccc11 JK |
2446 | static const char upgrade_note[] = |
2447 | "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n" | |
2448 | "(see http://sourceforge.net/projects/kvm).\n"; | |
2459d420 | 2449 | const struct { |
670436ce AJ |
2450 | const char *name; |
2451 | int num; | |
2452 | } num_cpus[] = { | |
5cc8767d LX |
2453 | { "SMP", ms->smp.cpus }, |
2454 | { "hotpluggable", ms->smp.max_cpus }, | |
2459d420 | 2455 | { /* end of list */ } |
670436ce AJ |
2456 | }, *nc = num_cpus; |
2457 | int soft_vcpus_limit, hard_vcpus_limit; | |
05330448 | 2458 | KVMState *s; |
94a8d39a | 2459 | const KVMCapabilityInfo *missing_cap; |
05330448 | 2460 | int ret; |
7bbda04c | 2461 | int type = 0; |
494cd11d | 2462 | uint64_t dirty_log_manual_caps; |
05330448 | 2463 | |
a2f77862 PX |
2464 | qemu_mutex_init(&kml_slots_lock); |
2465 | ||
fc02086b | 2466 | s = KVM_STATE(ms->accelerator); |
05330448 | 2467 | |
3145fcb6 DG |
2468 | /* |
2469 | * On systems where the kernel can support different base page | |
2470 | * sizes, host page size may be different from TARGET_PAGE_SIZE, | |
2471 | * even with KVM. TARGET_PAGE_SIZE is assumed to be the minimum | |
2472 | * page size for the system though. | |
2473 | */ | |
8e3b0cbb | 2474 | assert(TARGET_PAGE_SIZE <= qemu_real_host_page_size()); |
3145fcb6 | 2475 | |
aed6efb9 | 2476 | s->sigmask_len = 8; |
a27dd2de | 2477 | accel_blocker_init(); |
aed6efb9 | 2478 | |
e22a25c9 | 2479 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
72cf2d4f | 2480 | QTAILQ_INIT(&s->kvm_sw_breakpoints); |
e22a25c9 | 2481 | #endif |
4c055ab5 | 2482 | QLIST_INIT(&s->kvm_parked_vcpus); |
448058aa | 2483 | s->fd = qemu_open_old("/dev/kvm", O_RDWR); |
05330448 AL |
2484 | if (s->fd == -1) { |
2485 | fprintf(stderr, "Could not access KVM kernel module: %m\n"); | |
2486 | ret = -errno; | |
2487 | goto err; | |
2488 | } | |
2489 | ||
2490 | ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0); | |
2491 | if (ret < KVM_API_VERSION) { | |
0e1dac6c | 2492 | if (ret >= 0) { |
05330448 | 2493 | ret = -EINVAL; |
a426e122 | 2494 | } |
05330448 AL |
2495 | fprintf(stderr, "kvm version too old\n"); |
2496 | goto err; | |
2497 | } | |
2498 | ||
2499 | if (ret > KVM_API_VERSION) { | |
2500 | ret = -EINVAL; | |
2501 | fprintf(stderr, "kvm version not supported\n"); | |
2502 | goto err; | |
2503 | } | |
2504 | ||
cf0f7cf9 | 2505 | kvm_immediate_exit = kvm_check_extension(s, KVM_CAP_IMMEDIATE_EXIT); |
fb541ca5 AW |
2506 | s->nr_slots = kvm_check_extension(s, KVM_CAP_NR_MEMSLOTS); |
2507 | ||
2508 | /* If unspecified, use the default value */ | |
2509 | if (!s->nr_slots) { | |
2510 | s->nr_slots = 32; | |
2511 | } | |
2512 | ||
8072aae3 AK |
2513 | s->nr_as = kvm_check_extension(s, KVM_CAP_MULTI_ADDRESS_SPACE); |
2514 | if (s->nr_as <= 1) { | |
2515 | s->nr_as = 1; | |
2516 | } | |
2517 | s->as = g_new0(struct KVMAs, s->nr_as); | |
2518 | ||
f2ce39b4 PB |
2519 | if (object_property_find(OBJECT(current_machine), "kvm-type")) { |
2520 | g_autofree char *kvm_type = object_property_get_str(OBJECT(current_machine), | |
2521 | "kvm-type", | |
2522 | &error_abort); | |
dc0ca80e | 2523 | type = mc->kvm_type(ms, kvm_type); |
516fc0a0 AJ |
2524 | } else if (mc->kvm_type) { |
2525 | type = mc->kvm_type(ms, NULL); | |
135a129a AK |
2526 | } |
2527 | ||
94ccff13 | 2528 | do { |
135a129a | 2529 | ret = kvm_ioctl(s, KVM_CREATE_VM, type); |
94ccff13 TK |
2530 | } while (ret == -EINTR); |
2531 | ||
2532 | if (ret < 0) { | |
521f438e | 2533 | fprintf(stderr, "ioctl(KVM_CREATE_VM) failed: %d %s\n", -ret, |
94ccff13 TK |
2534 | strerror(-ret)); |
2535 | ||
0104dcac | 2536 | #ifdef TARGET_S390X |
2c80e996 CH |
2537 | if (ret == -EINVAL) { |
2538 | fprintf(stderr, | |
2539 | "Host kernel setup problem detected. Please verify:\n"); | |
2540 | fprintf(stderr, "- for kernels supporting the switch_amode or" | |
2541 | " user_mode parameters, whether\n"); | |
2542 | fprintf(stderr, | |
2543 | " user space is running in primary address space\n"); | |
2544 | fprintf(stderr, | |
2545 | "- for kernels supporting the vm.allocate_pgste sysctl, " | |
2546 | "whether it is enabled\n"); | |
2547 | } | |
380e4929 FR |
2548 | #elif defined(TARGET_PPC) |
2549 | if (ret == -EINVAL) { | |
2550 | fprintf(stderr, | |
2551 | "PPC KVM module is not loaded. Try modprobe kvm_%s.\n", | |
2552 | (type == 2) ? "pr" : "hv"); | |
2553 | } | |
0104dcac | 2554 | #endif |
05330448 | 2555 | goto err; |
0104dcac | 2556 | } |
05330448 | 2557 | |
94ccff13 | 2558 | s->vmfd = ret; |
11748ba7 GK |
2559 | |
2560 | /* check the vcpu limits */ | |
2561 | soft_vcpus_limit = kvm_recommended_vcpus(s); | |
2562 | hard_vcpus_limit = kvm_max_vcpus(s); | |
2563 | ||
2564 | while (nc->name) { | |
2565 | if (nc->num > soft_vcpus_limit) { | |
2566 | warn_report("Number of %s cpus requested (%d) exceeds " | |
2567 | "the recommended cpus supported by KVM (%d)", | |
2568 | nc->name, nc->num, soft_vcpus_limit); | |
2569 | ||
2570 | if (nc->num > hard_vcpus_limit) { | |
2571 | fprintf(stderr, "Number of %s cpus requested (%d) exceeds " | |
2572 | "the maximum cpus supported by KVM (%d)\n", | |
2573 | nc->name, nc->num, hard_vcpus_limit); | |
2574 | exit(1); | |
2575 | } | |
2576 | } | |
2577 | nc++; | |
2578 | } | |
2579 | ||
94a8d39a JK |
2580 | missing_cap = kvm_check_extension_list(s, kvm_required_capabilites); |
2581 | if (!missing_cap) { | |
2582 | missing_cap = | |
2583 | kvm_check_extension_list(s, kvm_arch_required_capabilities); | |
05330448 | 2584 | } |
94a8d39a | 2585 | if (missing_cap) { |
ad7b8b33 | 2586 | ret = -EINVAL; |
94a8d39a JK |
2587 | fprintf(stderr, "kvm does not support %s\n%s", |
2588 | missing_cap->name, upgrade_note); | |
d85dc283 AL |
2589 | goto err; |
2590 | } | |
2591 | ||
ad7b8b33 | 2592 | s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO); |
e6d34aee PH |
2593 | s->coalesced_pio = s->coalesced_mmio && |
2594 | kvm_check_extension(s, KVM_CAP_COALESCED_PIO); | |
f65ed4c1 | 2595 | |
b4420f19 PX |
2596 | /* |
2597 | * Enable KVM dirty ring if supported, otherwise fall back to | |
2598 | * dirty logging mode | |
2599 | */ | |
3794cb94 GS |
2600 | ret = kvm_dirty_ring_init(s); |
2601 | if (ret < 0) { | |
2602 | goto err; | |
b4420f19 PX |
2603 | } |
2604 | ||
a81a5926 PX |
2605 | /* |
2606 | * KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2 is not needed when dirty ring is | |
2607 | * enabled. More importantly, KVM_DIRTY_LOG_INITIALLY_SET will assume no | |
2608 | * page is wr-protected initially, which is against how kvm dirty ring is | |
2609 | * usage - kvm dirty ring requires all pages are wr-protected at the very | |
2610 | * beginning. Enabling this feature for dirty ring causes data corruption. | |
b4420f19 PX |
2611 | * |
2612 | * TODO: Without KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2 and kvm clear dirty log, | |
2613 | * we may expect a higher stall time when starting the migration. In the | |
2614 | * future we can enable KVM_CLEAR_DIRTY_LOG to work with dirty ring too: | |
2615 | * instead of clearing dirty bit, it can be a way to explicitly wr-protect | |
2616 | * guest pages. | |
a81a5926 PX |
2617 | */ |
2618 | if (!s->kvm_dirty_ring_size) { | |
2619 | dirty_log_manual_caps = | |
2620 | kvm_check_extension(s, KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2); | |
2621 | dirty_log_manual_caps &= (KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE | | |
2622 | KVM_DIRTY_LOG_INITIALLY_SET); | |
2623 | s->manual_dirty_log_protect = dirty_log_manual_caps; | |
2624 | if (dirty_log_manual_caps) { | |
2625 | ret = kvm_vm_enable_cap(s, KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2, 0, | |
2626 | dirty_log_manual_caps); | |
2627 | if (ret) { | |
2628 | warn_report("Trying to enable capability %"PRIu64" of " | |
2629 | "KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2 but failed. " | |
2630 | "Falling back to the legacy mode. ", | |
2631 | dirty_log_manual_caps); | |
2632 | s->manual_dirty_log_protect = 0; | |
2633 | } | |
ff4aa114 PX |
2634 | } |
2635 | } | |
2636 | ||
a0fb002c JK |
2637 | #ifdef KVM_CAP_VCPU_EVENTS |
2638 | s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS); | |
2639 | #endif | |
2640 | ||
b0b1d690 JK |
2641 | s->robust_singlestep = |
2642 | kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP); | |
b0b1d690 | 2643 | |
ff44f1a3 JK |
2644 | #ifdef KVM_CAP_DEBUGREGS |
2645 | s->debugregs = kvm_check_extension(s, KVM_CAP_DEBUGREGS); | |
2646 | #endif | |
2647 | ||
ebbfef2f LA |
2648 | s->max_nested_state_len = kvm_check_extension(s, KVM_CAP_NESTED_STATE); |
2649 | ||
d3d3bef0 | 2650 | #ifdef KVM_CAP_IRQ_ROUTING |
50bf31b9 | 2651 | kvm_direct_msi_allowed = (kvm_check_extension(s, KVM_CAP_SIGNAL_MSI) > 0); |
d3d3bef0 | 2652 | #endif |
4a3adebb | 2653 | |
3ab73842 JK |
2654 | s->intx_set_mask = kvm_check_extension(s, KVM_CAP_PCI_2_3); |
2655 | ||
e333cd69 | 2656 | s->irq_set_ioctl = KVM_IRQ_LINE; |
8732fbd2 | 2657 | if (kvm_check_extension(s, KVM_CAP_IRQ_INJECT_STATUS)) { |
e333cd69 | 2658 | s->irq_set_ioctl = KVM_IRQ_LINE_STATUS; |
8732fbd2 PM |
2659 | } |
2660 | ||
df9c8b75 JJ |
2661 | kvm_readonly_mem_allowed = |
2662 | (kvm_check_extension(s, KVM_CAP_READONLY_MEM) > 0); | |
df9c8b75 | 2663 | |
69e03ae6 NN |
2664 | kvm_eventfds_allowed = |
2665 | (kvm_check_extension(s, KVM_CAP_IOEVENTFD) > 0); | |
2666 | ||
f41389ae EA |
2667 | kvm_irqfds_allowed = |
2668 | (kvm_check_extension(s, KVM_CAP_IRQFD) > 0); | |
2669 | ||
2670 | kvm_resamplefds_allowed = | |
2671 | (kvm_check_extension(s, KVM_CAP_IRQFD_RESAMPLE) > 0); | |
2672 | ||
d0a073a1 DD |
2673 | kvm_vm_attributes_allowed = |
2674 | (kvm_check_extension(s, KVM_CAP_VM_ATTRIBUTES) > 0); | |
2675 | ||
35108223 JW |
2676 | kvm_ioeventfd_any_length_allowed = |
2677 | (kvm_check_extension(s, KVM_CAP_IOEVENTFD_ANY_LENGTH) > 0); | |
2678 | ||
12bc5b4c ML |
2679 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
2680 | kvm_has_guest_debug = | |
2681 | (kvm_check_extension(s, KVM_CAP_SET_GUEST_DEBUG) > 0); | |
2682 | #endif | |
2683 | ||
2684 | kvm_sstep_flags = 0; | |
2685 | if (kvm_has_guest_debug) { | |
2686 | kvm_sstep_flags = SSTEP_ENABLE; | |
fd2ddd16 ML |
2687 | |
2688 | #if defined KVM_CAP_SET_GUEST_DEBUG2 | |
2689 | int guest_debug_flags = | |
2690 | kvm_check_extension(s, KVM_CAP_SET_GUEST_DEBUG2); | |
2691 | ||
2692 | if (guest_debug_flags & KVM_GUESTDBG_BLOCKIRQ) { | |
2693 | kvm_sstep_flags |= SSTEP_NOIRQ; | |
2694 | } | |
2695 | #endif | |
12bc5b4c ML |
2696 | } |
2697 | ||
d870cfde GA |
2698 | kvm_state = s; |
2699 | ||
b16565b3 | 2700 | ret = kvm_arch_init(ms, s); |
a426e122 | 2701 | if (ret < 0) { |
05330448 | 2702 | goto err; |
a426e122 | 2703 | } |
05330448 | 2704 | |
d1972be1 XL |
2705 | if (s->kernel_irqchip_split == ON_OFF_AUTO_AUTO) { |
2706 | s->kernel_irqchip_split = mc->default_kernel_irqchip_split ? ON_OFF_AUTO_ON : ON_OFF_AUTO_OFF; | |
2707 | } | |
2708 | ||
6b552b9b DG |
2709 | qemu_register_reset(kvm_unpoison_all, NULL); |
2710 | ||
11bc4a13 | 2711 | if (s->kernel_irqchip_allowed) { |
4376c40d | 2712 | kvm_irqchip_create(s); |
84b058d7 JK |
2713 | } |
2714 | ||
8c56c1a5 PF |
2715 | if (kvm_eventfds_allowed) { |
2716 | s->memory_listener.listener.eventfd_add = kvm_mem_ioeventfd_add; | |
2717 | s->memory_listener.listener.eventfd_del = kvm_mem_ioeventfd_del; | |
2718 | } | |
e6d34aee PH |
2719 | s->memory_listener.listener.coalesced_io_add = kvm_coalesce_mmio_region; |
2720 | s->memory_listener.listener.coalesced_io_del = kvm_uncoalesce_mmio_region; | |
7bbda04c PB |
2721 | |
2722 | kvm_memory_listener_register(s, &s->memory_listener, | |
142518bd | 2723 | &address_space_memory, 0, "kvm-memory"); |
f9b49088 EA |
2724 | if (kvm_eventfds_allowed) { |
2725 | memory_listener_register(&kvm_io_listener, | |
2726 | &address_space_io); | |
2727 | } | |
e6d34aee PH |
2728 | memory_listener_register(&kvm_coalesced_pio_listener, |
2729 | &address_space_io); | |
05330448 | 2730 | |
d2f2b8a7 SH |
2731 | s->many_ioeventfds = kvm_check_many_ioeventfds(); |
2732 | ||
62dd4eda | 2733 | s->sync_mmu = !!kvm_vm_check_extension(kvm_state, KVM_CAP_SYNC_MMU); |
f5948942 | 2734 | if (!s->sync_mmu) { |
956b109f DH |
2735 | ret = ram_block_discard_disable(true); |
2736 | assert(!ret); | |
f5948942 | 2737 | } |
b4420f19 PX |
2738 | |
2739 | if (s->kvm_dirty_ring_size) { | |
2740 | ret = kvm_dirty_ring_reaper_init(s); | |
2741 | if (ret) { | |
2742 | goto err; | |
2743 | } | |
2744 | } | |
2745 | ||
cc01a3f4 | 2746 | if (kvm_check_extension(kvm_state, KVM_CAP_BINARY_STATS_FD)) { |
068cc51d PB |
2747 | add_stats_callbacks(STATS_PROVIDER_KVM, query_stats_cb, |
2748 | query_stats_schemas_cb); | |
cc01a3f4 MK |
2749 | } |
2750 | ||
05330448 AL |
2751 | return 0; |
2752 | ||
2753 | err: | |
0e1dac6c | 2754 | assert(ret < 0); |
6d1cc321 SW |
2755 | if (s->vmfd >= 0) { |
2756 | close(s->vmfd); | |
2757 | } | |
2758 | if (s->fd != -1) { | |
2759 | close(s->fd); | |
05330448 | 2760 | } |
7bbda04c | 2761 | g_free(s->memory_listener.slots); |
05330448 AL |
2762 | |
2763 | return ret; | |
2764 | } | |
2765 | ||
aed6efb9 JH |
2766 | void kvm_set_sigmask_len(KVMState *s, unsigned int sigmask_len) |
2767 | { | |
2768 | s->sigmask_len = sigmask_len; | |
2769 | } | |
2770 | ||
4c663752 PB |
2771 | static void kvm_handle_io(uint16_t port, MemTxAttrs attrs, void *data, int direction, |
2772 | int size, uint32_t count) | |
05330448 AL |
2773 | { |
2774 | int i; | |
2775 | uint8_t *ptr = data; | |
2776 | ||
2777 | for (i = 0; i < count; i++) { | |
4c663752 | 2778 | address_space_rw(&address_space_io, port, attrs, |
5c9eb028 | 2779 | ptr, size, |
354678c5 | 2780 | direction == KVM_EXIT_IO_OUT); |
05330448 AL |
2781 | ptr += size; |
2782 | } | |
05330448 AL |
2783 | } |
2784 | ||
5326ab55 | 2785 | static int kvm_handle_internal_error(CPUState *cpu, struct kvm_run *run) |
7c80eef8 | 2786 | { |
977c7b6d RK |
2787 | fprintf(stderr, "KVM internal error. Suberror: %d\n", |
2788 | run->internal.suberror); | |
2789 | ||
7c80eef8 MT |
2790 | if (kvm_check_extension(kvm_state, KVM_CAP_INTERNAL_ERROR_DATA)) { |
2791 | int i; | |
2792 | ||
7c80eef8 | 2793 | for (i = 0; i < run->internal.ndata; ++i) { |
56567da3 | 2794 | fprintf(stderr, "extra data[%d]: 0x%016"PRIx64"\n", |
7c80eef8 MT |
2795 | i, (uint64_t)run->internal.data[i]); |
2796 | } | |
2797 | } | |
7c80eef8 MT |
2798 | if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) { |
2799 | fprintf(stderr, "emulation failure\n"); | |
20d695a9 | 2800 | if (!kvm_arch_stop_on_emulation_error(cpu)) { |
90c84c56 | 2801 | cpu_dump_state(cpu, stderr, CPU_DUMP_CODE); |
d73cd8f4 | 2802 | return EXCP_INTERRUPT; |
a426e122 | 2803 | } |
7c80eef8 MT |
2804 | } |
2805 | /* FIXME: Should trigger a qmp message to let management know | |
2806 | * something went wrong. | |
2807 | */ | |
73aaec4a | 2808 | return -1; |
7c80eef8 | 2809 | } |
7c80eef8 | 2810 | |
62a2744c | 2811 | void kvm_flush_coalesced_mmio_buffer(void) |
f65ed4c1 | 2812 | { |
f65ed4c1 | 2813 | KVMState *s = kvm_state; |
1cae88b9 | 2814 | |
fe6bda58 | 2815 | if (!s || s->coalesced_flush_in_progress) { |
1cae88b9 AK |
2816 | return; |
2817 | } | |
2818 | ||
2819 | s->coalesced_flush_in_progress = true; | |
2820 | ||
62a2744c SY |
2821 | if (s->coalesced_mmio_ring) { |
2822 | struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring; | |
f65ed4c1 AL |
2823 | while (ring->first != ring->last) { |
2824 | struct kvm_coalesced_mmio *ent; | |
2825 | ||
2826 | ent = &ring->coalesced_mmio[ring->first]; | |
2827 | ||
e6d34aee | 2828 | if (ent->pio == 1) { |
19f70347 PM |
2829 | address_space_write(&address_space_io, ent->phys_addr, |
2830 | MEMTXATTRS_UNSPECIFIED, ent->data, | |
2831 | ent->len); | |
e6d34aee PH |
2832 | } else { |
2833 | cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len); | |
2834 | } | |
85199474 | 2835 | smp_wmb(); |
f65ed4c1 AL |
2836 | ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX; |
2837 | } | |
2838 | } | |
1cae88b9 AK |
2839 | |
2840 | s->coalesced_flush_in_progress = false; | |
f65ed4c1 AL |
2841 | } |
2842 | ||
92a5199b TL |
2843 | bool kvm_cpu_check_are_resettable(void) |
2844 | { | |
2845 | return kvm_arch_cpu_check_are_resettable(); | |
2846 | } | |
2847 | ||
14e6fe12 | 2848 | static void do_kvm_cpu_synchronize_state(CPUState *cpu, run_on_cpu_data arg) |
4c0960c0 | 2849 | { |
99f31832 | 2850 | if (!cpu->vcpu_dirty) { |
20d695a9 | 2851 | kvm_arch_get_registers(cpu); |
99f31832 | 2852 | cpu->vcpu_dirty = true; |
4c0960c0 AK |
2853 | } |
2854 | } | |
2855 | ||
dd1750d7 | 2856 | void kvm_cpu_synchronize_state(CPUState *cpu) |
2705d56a | 2857 | { |
99f31832 | 2858 | if (!cpu->vcpu_dirty) { |
14e6fe12 | 2859 | run_on_cpu(cpu, do_kvm_cpu_synchronize_state, RUN_ON_CPU_NULL); |
a426e122 | 2860 | } |
2705d56a JK |
2861 | } |
2862 | ||
14e6fe12 | 2863 | static void do_kvm_cpu_synchronize_post_reset(CPUState *cpu, run_on_cpu_data arg) |
ea375f9a | 2864 | { |
20d695a9 | 2865 | kvm_arch_put_registers(cpu, KVM_PUT_RESET_STATE); |
99f31832 | 2866 | cpu->vcpu_dirty = false; |
ea375f9a JK |
2867 | } |
2868 | ||
c8e2085d DH |
2869 | void kvm_cpu_synchronize_post_reset(CPUState *cpu) |
2870 | { | |
14e6fe12 | 2871 | run_on_cpu(cpu, do_kvm_cpu_synchronize_post_reset, RUN_ON_CPU_NULL); |
c8e2085d DH |
2872 | } |
2873 | ||
14e6fe12 | 2874 | static void do_kvm_cpu_synchronize_post_init(CPUState *cpu, run_on_cpu_data arg) |
ea375f9a | 2875 | { |
20d695a9 | 2876 | kvm_arch_put_registers(cpu, KVM_PUT_FULL_STATE); |
99f31832 | 2877 | cpu->vcpu_dirty = false; |
ea375f9a JK |
2878 | } |
2879 | ||
c8e2085d DH |
2880 | void kvm_cpu_synchronize_post_init(CPUState *cpu) |
2881 | { | |
14e6fe12 | 2882 | run_on_cpu(cpu, do_kvm_cpu_synchronize_post_init, RUN_ON_CPU_NULL); |
c8e2085d DH |
2883 | } |
2884 | ||
75e972da DG |
2885 | static void do_kvm_cpu_synchronize_pre_loadvm(CPUState *cpu, run_on_cpu_data arg) |
2886 | { | |
99f31832 | 2887 | cpu->vcpu_dirty = true; |
75e972da DG |
2888 | } |
2889 | ||
2890 | void kvm_cpu_synchronize_pre_loadvm(CPUState *cpu) | |
2891 | { | |
2892 | run_on_cpu(cpu, do_kvm_cpu_synchronize_pre_loadvm, RUN_ON_CPU_NULL); | |
2893 | } | |
2894 | ||
2ae41db2 PB |
2895 | #ifdef KVM_HAVE_MCE_INJECTION |
2896 | static __thread void *pending_sigbus_addr; | |
2897 | static __thread int pending_sigbus_code; | |
2898 | static __thread bool have_sigbus_pending; | |
2899 | #endif | |
2900 | ||
cf0f7cf9 PB |
2901 | static void kvm_cpu_kick(CPUState *cpu) |
2902 | { | |
d73415a3 | 2903 | qatomic_set(&cpu->kvm_run->immediate_exit, 1); |
cf0f7cf9 PB |
2904 | } |
2905 | ||
2906 | static void kvm_cpu_kick_self(void) | |
2907 | { | |
2908 | if (kvm_immediate_exit) { | |
2909 | kvm_cpu_kick(current_cpu); | |
2910 | } else { | |
2911 | qemu_cpu_kick_self(); | |
2912 | } | |
2913 | } | |
2914 | ||
18268b60 PB |
2915 | static void kvm_eat_signals(CPUState *cpu) |
2916 | { | |
2917 | struct timespec ts = { 0, 0 }; | |
2918 | siginfo_t siginfo; | |
2919 | sigset_t waitset; | |
2920 | sigset_t chkset; | |
2921 | int r; | |
2922 | ||
cf0f7cf9 | 2923 | if (kvm_immediate_exit) { |
d73415a3 | 2924 | qatomic_set(&cpu->kvm_run->immediate_exit, 0); |
cf0f7cf9 PB |
2925 | /* Write kvm_run->immediate_exit before the cpu->exit_request |
2926 | * write in kvm_cpu_exec. | |
2927 | */ | |
2928 | smp_wmb(); | |
2929 | return; | |
2930 | } | |
2931 | ||
18268b60 PB |
2932 | sigemptyset(&waitset); |
2933 | sigaddset(&waitset, SIG_IPI); | |
2934 | ||
2935 | do { | |
2936 | r = sigtimedwait(&waitset, &siginfo, &ts); | |
2937 | if (r == -1 && !(errno == EAGAIN || errno == EINTR)) { | |
2938 | perror("sigtimedwait"); | |
2939 | exit(1); | |
2940 | } | |
2941 | ||
2942 | r = sigpending(&chkset); | |
2943 | if (r == -1) { | |
2944 | perror("sigpending"); | |
2945 | exit(1); | |
2946 | } | |
2947 | } while (sigismember(&chkset, SIG_IPI)); | |
2948 | } | |
2949 | ||
1458c363 | 2950 | int kvm_cpu_exec(CPUState *cpu) |
05330448 | 2951 | { |
f7575c96 | 2952 | struct kvm_run *run = cpu->kvm_run; |
7cbb533f | 2953 | int ret, run_ret; |
05330448 | 2954 | |
8c0d577e | 2955 | DPRINTF("kvm_cpu_exec()\n"); |
05330448 | 2956 | |
20d695a9 | 2957 | if (kvm_arch_process_async_events(cpu)) { |
d73415a3 | 2958 | qatomic_set(&cpu->exit_request, 0); |
6792a57b | 2959 | return EXCP_HLT; |
9ccfac9e | 2960 | } |
0af691d7 | 2961 | |
4b8523ee | 2962 | qemu_mutex_unlock_iothread(); |
1d78a3c3 | 2963 | cpu_exec_start(cpu); |
4b8523ee | 2964 | |
9ccfac9e | 2965 | do { |
4c663752 PB |
2966 | MemTxAttrs attrs; |
2967 | ||
99f31832 | 2968 | if (cpu->vcpu_dirty) { |
20d695a9 | 2969 | kvm_arch_put_registers(cpu, KVM_PUT_RUNTIME_STATE); |
99f31832 | 2970 | cpu->vcpu_dirty = false; |
4c0960c0 AK |
2971 | } |
2972 | ||
20d695a9 | 2973 | kvm_arch_pre_run(cpu, run); |
d73415a3 | 2974 | if (qatomic_read(&cpu->exit_request)) { |
9ccfac9e JK |
2975 | DPRINTF("interrupt exit requested\n"); |
2976 | /* | |
2977 | * KVM requires us to reenter the kernel after IO exits to complete | |
2978 | * instruction emulation. This self-signal will ensure that we | |
2979 | * leave ASAP again. | |
2980 | */ | |
cf0f7cf9 | 2981 | kvm_cpu_kick_self(); |
9ccfac9e | 2982 | } |
9ccfac9e | 2983 | |
cf0f7cf9 PB |
2984 | /* Read cpu->exit_request before KVM_RUN reads run->immediate_exit. |
2985 | * Matching barrier in kvm_eat_signals. | |
2986 | */ | |
2987 | smp_rmb(); | |
2988 | ||
1bc22652 | 2989 | run_ret = kvm_vcpu_ioctl(cpu, KVM_RUN, 0); |
9ccfac9e | 2990 | |
4c663752 | 2991 | attrs = kvm_arch_post_run(cpu, run); |
05330448 | 2992 | |
2ae41db2 PB |
2993 | #ifdef KVM_HAVE_MCE_INJECTION |
2994 | if (unlikely(have_sigbus_pending)) { | |
2995 | qemu_mutex_lock_iothread(); | |
2996 | kvm_arch_on_sigbus_vcpu(cpu, pending_sigbus_code, | |
2997 | pending_sigbus_addr); | |
2998 | have_sigbus_pending = false; | |
2999 | qemu_mutex_unlock_iothread(); | |
3000 | } | |
3001 | #endif | |
3002 | ||
7cbb533f | 3003 | if (run_ret < 0) { |
dc77d341 JK |
3004 | if (run_ret == -EINTR || run_ret == -EAGAIN) { |
3005 | DPRINTF("io window exit\n"); | |
18268b60 | 3006 | kvm_eat_signals(cpu); |
d73cd8f4 | 3007 | ret = EXCP_INTERRUPT; |
dc77d341 JK |
3008 | break; |
3009 | } | |
7b011fbc ME |
3010 | fprintf(stderr, "error: kvm run failed %s\n", |
3011 | strerror(-run_ret)); | |
dae02ba5 LV |
3012 | #ifdef TARGET_PPC |
3013 | if (run_ret == -EBUSY) { | |
3014 | fprintf(stderr, | |
3015 | "This is probably because your SMT is enabled.\n" | |
3016 | "VCPU can only run on primary threads with all " | |
3017 | "secondary threads offline.\n"); | |
3018 | } | |
3019 | #endif | |
a85e130e PB |
3020 | ret = -1; |
3021 | break; | |
05330448 AL |
3022 | } |
3023 | ||
b76ac80a | 3024 | trace_kvm_run_exit(cpu->cpu_index, run->exit_reason); |
05330448 AL |
3025 | switch (run->exit_reason) { |
3026 | case KVM_EXIT_IO: | |
8c0d577e | 3027 | DPRINTF("handle_io\n"); |
80b7d2ef | 3028 | /* Called outside BQL */ |
4c663752 | 3029 | kvm_handle_io(run->io.port, attrs, |
b30e93e9 JK |
3030 | (uint8_t *)run + run->io.data_offset, |
3031 | run->io.direction, | |
3032 | run->io.size, | |
3033 | run->io.count); | |
d73cd8f4 | 3034 | ret = 0; |
05330448 AL |
3035 | break; |
3036 | case KVM_EXIT_MMIO: | |
8c0d577e | 3037 | DPRINTF("handle_mmio\n"); |
de7ea885 | 3038 | /* Called outside BQL */ |
4c663752 PB |
3039 | address_space_rw(&address_space_memory, |
3040 | run->mmio.phys_addr, attrs, | |
3041 | run->mmio.data, | |
3042 | run->mmio.len, | |
3043 | run->mmio.is_write); | |
d73cd8f4 | 3044 | ret = 0; |
05330448 AL |
3045 | break; |
3046 | case KVM_EXIT_IRQ_WINDOW_OPEN: | |
8c0d577e | 3047 | DPRINTF("irq_window_open\n"); |
d73cd8f4 | 3048 | ret = EXCP_INTERRUPT; |
05330448 AL |
3049 | break; |
3050 | case KVM_EXIT_SHUTDOWN: | |
8c0d577e | 3051 | DPRINTF("shutdown\n"); |
cf83f140 | 3052 | qemu_system_reset_request(SHUTDOWN_CAUSE_GUEST_RESET); |
d73cd8f4 | 3053 | ret = EXCP_INTERRUPT; |
05330448 AL |
3054 | break; |
3055 | case KVM_EXIT_UNKNOWN: | |
bb44e0d1 JK |
3056 | fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "\n", |
3057 | (uint64_t)run->hw.hardware_exit_reason); | |
73aaec4a | 3058 | ret = -1; |
05330448 | 3059 | break; |
7c80eef8 | 3060 | case KVM_EXIT_INTERNAL_ERROR: |
5326ab55 | 3061 | ret = kvm_handle_internal_error(cpu, run); |
7c80eef8 | 3062 | break; |
b4420f19 PX |
3063 | case KVM_EXIT_DIRTY_RING_FULL: |
3064 | /* | |
3065 | * We shouldn't continue if the dirty ring of this vcpu is | |
3066 | * still full. Got kicked by KVM_RESET_DIRTY_RINGS. | |
3067 | */ | |
3068 | trace_kvm_dirty_ring_full(cpu->cpu_index); | |
3069 | qemu_mutex_lock_iothread(); | |
baa60983 HH |
3070 | /* |
3071 | * We throttle vCPU by making it sleep once it exit from kernel | |
3072 | * due to dirty ring full. In the dirtylimit scenario, reaping | |
3073 | * all vCPUs after a single vCPU dirty ring get full result in | |
3074 | * the miss of sleep, so just reap the ring-fulled vCPU. | |
3075 | */ | |
3076 | if (dirtylimit_in_service()) { | |
3077 | kvm_dirty_ring_reap(kvm_state, cpu); | |
3078 | } else { | |
3079 | kvm_dirty_ring_reap(kvm_state, NULL); | |
3080 | } | |
b4420f19 | 3081 | qemu_mutex_unlock_iothread(); |
baa60983 | 3082 | dirtylimit_vcpu_execute(cpu); |
b4420f19 PX |
3083 | ret = 0; |
3084 | break; | |
99040447 PS |
3085 | case KVM_EXIT_SYSTEM_EVENT: |
3086 | switch (run->system_event.type) { | |
3087 | case KVM_SYSTEM_EVENT_SHUTDOWN: | |
cf83f140 | 3088 | qemu_system_shutdown_request(SHUTDOWN_CAUSE_GUEST_SHUTDOWN); |
99040447 PS |
3089 | ret = EXCP_INTERRUPT; |
3090 | break; | |
3091 | case KVM_SYSTEM_EVENT_RESET: | |
cf83f140 | 3092 | qemu_system_reset_request(SHUTDOWN_CAUSE_GUEST_RESET); |
99040447 PS |
3093 | ret = EXCP_INTERRUPT; |
3094 | break; | |
7c207b90 | 3095 | case KVM_SYSTEM_EVENT_CRASH: |
d187e08d | 3096 | kvm_cpu_synchronize_state(cpu); |
7c207b90 | 3097 | qemu_mutex_lock_iothread(); |
c86f106b | 3098 | qemu_system_guest_panicked(cpu_get_crash_info(cpu)); |
7c207b90 AS |
3099 | qemu_mutex_unlock_iothread(); |
3100 | ret = 0; | |
3101 | break; | |
99040447 PS |
3102 | default: |
3103 | DPRINTF("kvm_arch_handle_exit\n"); | |
3104 | ret = kvm_arch_handle_exit(cpu, run); | |
3105 | break; | |
3106 | } | |
3107 | break; | |
05330448 | 3108 | default: |
8c0d577e | 3109 | DPRINTF("kvm_arch_handle_exit\n"); |
20d695a9 | 3110 | ret = kvm_arch_handle_exit(cpu, run); |
05330448 AL |
3111 | break; |
3112 | } | |
d73cd8f4 | 3113 | } while (ret == 0); |
05330448 | 3114 | |
1d78a3c3 | 3115 | cpu_exec_end(cpu); |
4b8523ee JK |
3116 | qemu_mutex_lock_iothread(); |
3117 | ||
73aaec4a | 3118 | if (ret < 0) { |
90c84c56 | 3119 | cpu_dump_state(cpu, stderr, CPU_DUMP_CODE); |
0461d5a6 | 3120 | vm_stop(RUN_STATE_INTERNAL_ERROR); |
becfc390 AL |
3121 | } |
3122 | ||
d73415a3 | 3123 | qatomic_set(&cpu->exit_request, 0); |
05330448 AL |
3124 | return ret; |
3125 | } | |
3126 | ||
984b5181 | 3127 | int kvm_ioctl(KVMState *s, int type, ...) |
05330448 AL |
3128 | { |
3129 | int ret; | |
984b5181 AL |
3130 | void *arg; |
3131 | va_list ap; | |
05330448 | 3132 | |
984b5181 AL |
3133 | va_start(ap, type); |
3134 | arg = va_arg(ap, void *); | |
3135 | va_end(ap); | |
3136 | ||
9c775729 | 3137 | trace_kvm_ioctl(type, arg); |
984b5181 | 3138 | ret = ioctl(s->fd, type, arg); |
a426e122 | 3139 | if (ret == -1) { |
05330448 | 3140 | ret = -errno; |
a426e122 | 3141 | } |
05330448 AL |
3142 | return ret; |
3143 | } | |
3144 | ||
984b5181 | 3145 | int kvm_vm_ioctl(KVMState *s, int type, ...) |
05330448 AL |
3146 | { |
3147 | int ret; | |
984b5181 AL |
3148 | void *arg; |
3149 | va_list ap; | |
3150 | ||
3151 | va_start(ap, type); | |
3152 | arg = va_arg(ap, void *); | |
3153 | va_end(ap); | |
05330448 | 3154 | |
9c775729 | 3155 | trace_kvm_vm_ioctl(type, arg); |
a27dd2de | 3156 | accel_ioctl_begin(); |
984b5181 | 3157 | ret = ioctl(s->vmfd, type, arg); |
a27dd2de | 3158 | accel_ioctl_end(); |
a426e122 | 3159 | if (ret == -1) { |
05330448 | 3160 | ret = -errno; |
a426e122 | 3161 | } |
05330448 AL |
3162 | return ret; |
3163 | } | |
3164 | ||
1bc22652 | 3165 | int kvm_vcpu_ioctl(CPUState *cpu, int type, ...) |
05330448 AL |
3166 | { |
3167 | int ret; | |
984b5181 AL |
3168 | void *arg; |
3169 | va_list ap; | |
3170 | ||
3171 | va_start(ap, type); | |
3172 | arg = va_arg(ap, void *); | |
3173 | va_end(ap); | |
05330448 | 3174 | |
9c775729 | 3175 | trace_kvm_vcpu_ioctl(cpu->cpu_index, type, arg); |
a27dd2de | 3176 | accel_cpu_ioctl_begin(cpu); |
8737c51c | 3177 | ret = ioctl(cpu->kvm_fd, type, arg); |
a27dd2de | 3178 | accel_cpu_ioctl_end(cpu); |
a426e122 | 3179 | if (ret == -1) { |
05330448 | 3180 | ret = -errno; |
a426e122 | 3181 | } |
05330448 AL |
3182 | return ret; |
3183 | } | |
bd322087 | 3184 | |
0a6a7cca CD |
3185 | int kvm_device_ioctl(int fd, int type, ...) |
3186 | { | |
3187 | int ret; | |
3188 | void *arg; | |
3189 | va_list ap; | |
3190 | ||
3191 | va_start(ap, type); | |
3192 | arg = va_arg(ap, void *); | |
3193 | va_end(ap); | |
3194 | ||
3195 | trace_kvm_device_ioctl(fd, type, arg); | |
a27dd2de | 3196 | accel_ioctl_begin(); |
0a6a7cca | 3197 | ret = ioctl(fd, type, arg); |
a27dd2de | 3198 | accel_ioctl_end(); |
0a6a7cca CD |
3199 | if (ret == -1) { |
3200 | ret = -errno; | |
3201 | } | |
3202 | return ret; | |
3203 | } | |
3204 | ||
d0a073a1 DD |
3205 | int kvm_vm_check_attr(KVMState *s, uint32_t group, uint64_t attr) |
3206 | { | |
3207 | int ret; | |
3208 | struct kvm_device_attr attribute = { | |
3209 | .group = group, | |
3210 | .attr = attr, | |
3211 | }; | |
3212 | ||
3213 | if (!kvm_vm_attributes_allowed) { | |
3214 | return 0; | |
3215 | } | |
3216 | ||
3217 | ret = kvm_vm_ioctl(s, KVM_HAS_DEVICE_ATTR, &attribute); | |
3218 | /* kvm returns 0 on success for HAS_DEVICE_ATTR */ | |
3219 | return ret ? 0 : 1; | |
3220 | } | |
3221 | ||
4b3cfe72 PF |
3222 | int kvm_device_check_attr(int dev_fd, uint32_t group, uint64_t attr) |
3223 | { | |
3224 | struct kvm_device_attr attribute = { | |
3225 | .group = group, | |
3226 | .attr = attr, | |
3227 | .flags = 0, | |
3228 | }; | |
3229 | ||
3230 | return kvm_device_ioctl(dev_fd, KVM_HAS_DEVICE_ATTR, &attribute) ? 0 : 1; | |
3231 | } | |
3232 | ||
556969e9 EA |
3233 | int kvm_device_access(int fd, int group, uint64_t attr, |
3234 | void *val, bool write, Error **errp) | |
4b3cfe72 PF |
3235 | { |
3236 | struct kvm_device_attr kvmattr; | |
3237 | int err; | |
3238 | ||
3239 | kvmattr.flags = 0; | |
3240 | kvmattr.group = group; | |
3241 | kvmattr.attr = attr; | |
3242 | kvmattr.addr = (uintptr_t)val; | |
3243 | ||
3244 | err = kvm_device_ioctl(fd, | |
3245 | write ? KVM_SET_DEVICE_ATTR : KVM_GET_DEVICE_ATTR, | |
3246 | &kvmattr); | |
3247 | if (err < 0) { | |
556969e9 EA |
3248 | error_setg_errno(errp, -err, |
3249 | "KVM_%s_DEVICE_ATTR failed: Group %d " | |
3250 | "attr 0x%016" PRIx64, | |
3251 | write ? "SET" : "GET", group, attr); | |
4b3cfe72 | 3252 | } |
556969e9 | 3253 | return err; |
4b3cfe72 PF |
3254 | } |
3255 | ||
62dd4eda | 3256 | bool kvm_has_sync_mmu(void) |
bd322087 | 3257 | { |
62dd4eda | 3258 | return kvm_state->sync_mmu; |
bd322087 | 3259 | } |
e22a25c9 | 3260 | |
a0fb002c JK |
3261 | int kvm_has_vcpu_events(void) |
3262 | { | |
3263 | return kvm_state->vcpu_events; | |
3264 | } | |
3265 | ||
b0b1d690 JK |
3266 | int kvm_has_robust_singlestep(void) |
3267 | { | |
3268 | return kvm_state->robust_singlestep; | |
3269 | } | |
3270 | ||
ff44f1a3 JK |
3271 | int kvm_has_debugregs(void) |
3272 | { | |
3273 | return kvm_state->debugregs; | |
3274 | } | |
3275 | ||
ebbfef2f LA |
3276 | int kvm_max_nested_state_length(void) |
3277 | { | |
3278 | return kvm_state->max_nested_state_len; | |
3279 | } | |
3280 | ||
d2f2b8a7 SH |
3281 | int kvm_has_many_ioeventfds(void) |
3282 | { | |
3283 | if (!kvm_enabled()) { | |
3284 | return 0; | |
3285 | } | |
3286 | return kvm_state->many_ioeventfds; | |
3287 | } | |
3288 | ||
84b058d7 JK |
3289 | int kvm_has_gsi_routing(void) |
3290 | { | |
a9c5eb0d | 3291 | #ifdef KVM_CAP_IRQ_ROUTING |
84b058d7 | 3292 | return kvm_check_extension(kvm_state, KVM_CAP_IRQ_ROUTING); |
a9c5eb0d AG |
3293 | #else |
3294 | return false; | |
3295 | #endif | |
84b058d7 JK |
3296 | } |
3297 | ||
3ab73842 JK |
3298 | int kvm_has_intx_set_mask(void) |
3299 | { | |
3300 | return kvm_state->intx_set_mask; | |
3301 | } | |
3302 | ||
5d721b78 AG |
3303 | bool kvm_arm_supports_user_irq(void) |
3304 | { | |
3305 | return kvm_check_extension(kvm_state, KVM_CAP_ARM_USER_IRQ); | |
3306 | } | |
3307 | ||
e22a25c9 | 3308 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
a60f24b5 | 3309 | struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *cpu, |
e22a25c9 AL |
3310 | target_ulong pc) |
3311 | { | |
3312 | struct kvm_sw_breakpoint *bp; | |
3313 | ||
a60f24b5 | 3314 | QTAILQ_FOREACH(bp, &cpu->kvm_state->kvm_sw_breakpoints, entry) { |
a426e122 | 3315 | if (bp->pc == pc) { |
e22a25c9 | 3316 | return bp; |
a426e122 | 3317 | } |
e22a25c9 AL |
3318 | } |
3319 | return NULL; | |
3320 | } | |
3321 | ||
a60f24b5 | 3322 | int kvm_sw_breakpoints_active(CPUState *cpu) |
e22a25c9 | 3323 | { |
a60f24b5 | 3324 | return !QTAILQ_EMPTY(&cpu->kvm_state->kvm_sw_breakpoints); |
e22a25c9 AL |
3325 | } |
3326 | ||
452e4751 GC |
3327 | struct kvm_set_guest_debug_data { |
3328 | struct kvm_guest_debug dbg; | |
452e4751 GC |
3329 | int err; |
3330 | }; | |
3331 | ||
14e6fe12 | 3332 | static void kvm_invoke_set_guest_debug(CPUState *cpu, run_on_cpu_data data) |
452e4751 | 3333 | { |
14e6fe12 PB |
3334 | struct kvm_set_guest_debug_data *dbg_data = |
3335 | (struct kvm_set_guest_debug_data *) data.host_ptr; | |
b3807725 | 3336 | |
3c0ed2a3 | 3337 | dbg_data->err = kvm_vcpu_ioctl(cpu, KVM_SET_GUEST_DEBUG, |
a60f24b5 | 3338 | &dbg_data->dbg); |
452e4751 GC |
3339 | } |
3340 | ||
38e478ec | 3341 | int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap) |
e22a25c9 | 3342 | { |
452e4751 | 3343 | struct kvm_set_guest_debug_data data; |
e22a25c9 | 3344 | |
b0b1d690 | 3345 | data.dbg.control = reinject_trap; |
e22a25c9 | 3346 | |
ed2803da | 3347 | if (cpu->singlestep_enabled) { |
b0b1d690 | 3348 | data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP; |
fd2ddd16 ML |
3349 | |
3350 | if (cpu->singlestep_enabled & SSTEP_NOIRQ) { | |
3351 | data.dbg.control |= KVM_GUESTDBG_BLOCKIRQ; | |
3352 | } | |
b0b1d690 | 3353 | } |
20d695a9 | 3354 | kvm_arch_update_guest_debug(cpu, &data.dbg); |
e22a25c9 | 3355 | |
14e6fe12 PB |
3356 | run_on_cpu(cpu, kvm_invoke_set_guest_debug, |
3357 | RUN_ON_CPU_HOST_PTR(&data)); | |
452e4751 | 3358 | return data.err; |
e22a25c9 AL |
3359 | } |
3360 | ||
a48e7d9e AB |
3361 | bool kvm_supports_guest_debug(void) |
3362 | { | |
3363 | /* probed during kvm_init() */ | |
3364 | return kvm_has_guest_debug; | |
3365 | } | |
3366 | ||
55b5b8e9 | 3367 | int kvm_insert_breakpoint(CPUState *cpu, int type, vaddr addr, vaddr len) |
e22a25c9 AL |
3368 | { |
3369 | struct kvm_sw_breakpoint *bp; | |
e22a25c9 AL |
3370 | int err; |
3371 | ||
3372 | if (type == GDB_BREAKPOINT_SW) { | |
80b7cd73 | 3373 | bp = kvm_find_sw_breakpoint(cpu, addr); |
e22a25c9 AL |
3374 | if (bp) { |
3375 | bp->use_count++; | |
3376 | return 0; | |
3377 | } | |
3378 | ||
b21e2380 | 3379 | bp = g_new(struct kvm_sw_breakpoint, 1); |
e22a25c9 AL |
3380 | bp->pc = addr; |
3381 | bp->use_count = 1; | |
80b7cd73 | 3382 | err = kvm_arch_insert_sw_breakpoint(cpu, bp); |
e22a25c9 | 3383 | if (err) { |
7267c094 | 3384 | g_free(bp); |
e22a25c9 AL |
3385 | return err; |
3386 | } | |
3387 | ||
80b7cd73 | 3388 | QTAILQ_INSERT_HEAD(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry); |
e22a25c9 AL |
3389 | } else { |
3390 | err = kvm_arch_insert_hw_breakpoint(addr, len, type); | |
a426e122 | 3391 | if (err) { |
e22a25c9 | 3392 | return err; |
a426e122 | 3393 | } |
e22a25c9 AL |
3394 | } |
3395 | ||
bdc44640 | 3396 | CPU_FOREACH(cpu) { |
38e478ec | 3397 | err = kvm_update_guest_debug(cpu, 0); |
a426e122 | 3398 | if (err) { |
e22a25c9 | 3399 | return err; |
a426e122 | 3400 | } |
e22a25c9 AL |
3401 | } |
3402 | return 0; | |
3403 | } | |
3404 | ||
55b5b8e9 | 3405 | int kvm_remove_breakpoint(CPUState *cpu, int type, vaddr addr, vaddr len) |
e22a25c9 AL |
3406 | { |
3407 | struct kvm_sw_breakpoint *bp; | |
e22a25c9 AL |
3408 | int err; |
3409 | ||
3410 | if (type == GDB_BREAKPOINT_SW) { | |
80b7cd73 | 3411 | bp = kvm_find_sw_breakpoint(cpu, addr); |
a426e122 | 3412 | if (!bp) { |
e22a25c9 | 3413 | return -ENOENT; |
a426e122 | 3414 | } |
e22a25c9 AL |
3415 | |
3416 | if (bp->use_count > 1) { | |
3417 | bp->use_count--; | |
3418 | return 0; | |
3419 | } | |
3420 | ||
80b7cd73 | 3421 | err = kvm_arch_remove_sw_breakpoint(cpu, bp); |
a426e122 | 3422 | if (err) { |
e22a25c9 | 3423 | return err; |
a426e122 | 3424 | } |
e22a25c9 | 3425 | |
80b7cd73 | 3426 | QTAILQ_REMOVE(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry); |
7267c094 | 3427 | g_free(bp); |
e22a25c9 AL |
3428 | } else { |
3429 | err = kvm_arch_remove_hw_breakpoint(addr, len, type); | |
a426e122 | 3430 | if (err) { |
e22a25c9 | 3431 | return err; |
a426e122 | 3432 | } |
e22a25c9 AL |
3433 | } |
3434 | ||
bdc44640 | 3435 | CPU_FOREACH(cpu) { |
38e478ec | 3436 | err = kvm_update_guest_debug(cpu, 0); |
a426e122 | 3437 | if (err) { |
e22a25c9 | 3438 | return err; |
a426e122 | 3439 | } |
e22a25c9 AL |
3440 | } |
3441 | return 0; | |
3442 | } | |
3443 | ||
1d5791f4 | 3444 | void kvm_remove_all_breakpoints(CPUState *cpu) |
e22a25c9 AL |
3445 | { |
3446 | struct kvm_sw_breakpoint *bp, *next; | |
80b7cd73 | 3447 | KVMState *s = cpu->kvm_state; |
dc54e252 | 3448 | CPUState *tmpcpu; |
e22a25c9 | 3449 | |
72cf2d4f | 3450 | QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) { |
80b7cd73 | 3451 | if (kvm_arch_remove_sw_breakpoint(cpu, bp) != 0) { |
e22a25c9 | 3452 | /* Try harder to find a CPU that currently sees the breakpoint. */ |
dc54e252 CG |
3453 | CPU_FOREACH(tmpcpu) { |
3454 | if (kvm_arch_remove_sw_breakpoint(tmpcpu, bp) == 0) { | |
e22a25c9 | 3455 | break; |
a426e122 | 3456 | } |
e22a25c9 AL |
3457 | } |
3458 | } | |
78021d6d JK |
3459 | QTAILQ_REMOVE(&s->kvm_sw_breakpoints, bp, entry); |
3460 | g_free(bp); | |
e22a25c9 AL |
3461 | } |
3462 | kvm_arch_remove_all_hw_breakpoints(); | |
3463 | ||
bdc44640 | 3464 | CPU_FOREACH(cpu) { |
38e478ec | 3465 | kvm_update_guest_debug(cpu, 0); |
a426e122 | 3466 | } |
e22a25c9 AL |
3467 | } |
3468 | ||
e22a25c9 | 3469 | #endif /* !KVM_CAP_SET_GUEST_DEBUG */ |
cc84de95 | 3470 | |
18268b60 | 3471 | static int kvm_set_signal_mask(CPUState *cpu, const sigset_t *sigset) |
cc84de95 | 3472 | { |
aed6efb9 | 3473 | KVMState *s = kvm_state; |
cc84de95 MT |
3474 | struct kvm_signal_mask *sigmask; |
3475 | int r; | |
3476 | ||
7267c094 | 3477 | sigmask = g_malloc(sizeof(*sigmask) + sizeof(*sigset)); |
cc84de95 | 3478 | |
aed6efb9 | 3479 | sigmask->len = s->sigmask_len; |
cc84de95 | 3480 | memcpy(sigmask->sigset, sigset, sizeof(*sigset)); |
1bc22652 | 3481 | r = kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, sigmask); |
7267c094 | 3482 | g_free(sigmask); |
cc84de95 MT |
3483 | |
3484 | return r; | |
3485 | } | |
4d39892c | 3486 | |
cf0f7cf9 | 3487 | static void kvm_ipi_signal(int sig) |
18268b60 | 3488 | { |
cf0f7cf9 PB |
3489 | if (current_cpu) { |
3490 | assert(kvm_immediate_exit); | |
3491 | kvm_cpu_kick(current_cpu); | |
3492 | } | |
18268b60 PB |
3493 | } |
3494 | ||
3495 | void kvm_init_cpu_signals(CPUState *cpu) | |
3496 | { | |
3497 | int r; | |
3498 | sigset_t set; | |
3499 | struct sigaction sigact; | |
3500 | ||
3501 | memset(&sigact, 0, sizeof(sigact)); | |
cf0f7cf9 | 3502 | sigact.sa_handler = kvm_ipi_signal; |
18268b60 PB |
3503 | sigaction(SIG_IPI, &sigact, NULL); |
3504 | ||
3505 | pthread_sigmask(SIG_BLOCK, NULL, &set); | |
3506 | #if defined KVM_HAVE_MCE_INJECTION | |
3507 | sigdelset(&set, SIGBUS); | |
3508 | pthread_sigmask(SIG_SETMASK, &set, NULL); | |
3509 | #endif | |
3510 | sigdelset(&set, SIG_IPI); | |
cf0f7cf9 PB |
3511 | if (kvm_immediate_exit) { |
3512 | r = pthread_sigmask(SIG_SETMASK, &set, NULL); | |
3513 | } else { | |
3514 | r = kvm_set_signal_mask(cpu, &set); | |
3515 | } | |
18268b60 PB |
3516 | if (r) { |
3517 | fprintf(stderr, "kvm_set_signal_mask: %s\n", strerror(-r)); | |
3518 | exit(1); | |
3519 | } | |
3520 | } | |
3521 | ||
2ae41db2 | 3522 | /* Called asynchronously in VCPU thread. */ |
290adf38 | 3523 | int kvm_on_sigbus_vcpu(CPUState *cpu, int code, void *addr) |
a1b87fe0 | 3524 | { |
2ae41db2 PB |
3525 | #ifdef KVM_HAVE_MCE_INJECTION |
3526 | if (have_sigbus_pending) { | |
3527 | return 1; | |
3528 | } | |
3529 | have_sigbus_pending = true; | |
3530 | pending_sigbus_addr = addr; | |
3531 | pending_sigbus_code = code; | |
d73415a3 | 3532 | qatomic_set(&cpu->exit_request, 1); |
2ae41db2 PB |
3533 | return 0; |
3534 | #else | |
3535 | return 1; | |
3536 | #endif | |
a1b87fe0 JK |
3537 | } |
3538 | ||
2ae41db2 | 3539 | /* Called synchronously (via signalfd) in main thread. */ |
a1b87fe0 JK |
3540 | int kvm_on_sigbus(int code, void *addr) |
3541 | { | |
2ae41db2 | 3542 | #ifdef KVM_HAVE_MCE_INJECTION |
4d39892c PB |
3543 | /* Action required MCE kills the process if SIGBUS is blocked. Because |
3544 | * that's what happens in the I/O thread, where we handle MCE via signalfd, | |
3545 | * we can only get action optional here. | |
3546 | */ | |
3547 | assert(code != BUS_MCEERR_AR); | |
3548 | kvm_arch_on_sigbus_vcpu(first_cpu, code, addr); | |
3549 | return 0; | |
2ae41db2 PB |
3550 | #else |
3551 | return 1; | |
3552 | #endif | |
a1b87fe0 | 3553 | } |
0a6a7cca CD |
3554 | |
3555 | int kvm_create_device(KVMState *s, uint64_t type, bool test) | |
3556 | { | |
3557 | int ret; | |
3558 | struct kvm_create_device create_dev; | |
3559 | ||
3560 | create_dev.type = type; | |
3561 | create_dev.fd = -1; | |
3562 | create_dev.flags = test ? KVM_CREATE_DEVICE_TEST : 0; | |
3563 | ||
3564 | if (!kvm_check_extension(s, KVM_CAP_DEVICE_CTRL)) { | |
3565 | return -ENOTSUP; | |
3566 | } | |
3567 | ||
3568 | ret = kvm_vm_ioctl(s, KVM_CREATE_DEVICE, &create_dev); | |
3569 | if (ret) { | |
3570 | return ret; | |
3571 | } | |
3572 | ||
3573 | return test ? 0 : create_dev.fd; | |
3574 | } | |
ada4135f | 3575 | |
29039acf PX |
3576 | bool kvm_device_supported(int vmfd, uint64_t type) |
3577 | { | |
3578 | struct kvm_create_device create_dev = { | |
3579 | .type = type, | |
3580 | .fd = -1, | |
3581 | .flags = KVM_CREATE_DEVICE_TEST, | |
3582 | }; | |
3583 | ||
3584 | if (ioctl(vmfd, KVM_CHECK_EXTENSION, KVM_CAP_DEVICE_CTRL) <= 0) { | |
3585 | return false; | |
3586 | } | |
3587 | ||
3588 | return (ioctl(vmfd, KVM_CREATE_DEVICE, &create_dev) >= 0); | |
3589 | } | |
3590 | ||
ada4135f CH |
3591 | int kvm_set_one_reg(CPUState *cs, uint64_t id, void *source) |
3592 | { | |
3593 | struct kvm_one_reg reg; | |
3594 | int r; | |
3595 | ||
3596 | reg.id = id; | |
3597 | reg.addr = (uintptr_t) source; | |
3598 | r = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®); | |
3599 | if (r) { | |
844a3d34 | 3600 | trace_kvm_failed_reg_set(id, strerror(-r)); |
ada4135f CH |
3601 | } |
3602 | return r; | |
3603 | } | |
3604 | ||
3605 | int kvm_get_one_reg(CPUState *cs, uint64_t id, void *target) | |
3606 | { | |
3607 | struct kvm_one_reg reg; | |
3608 | int r; | |
3609 | ||
3610 | reg.id = id; | |
3611 | reg.addr = (uintptr_t) target; | |
3612 | r = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®); | |
3613 | if (r) { | |
844a3d34 | 3614 | trace_kvm_failed_reg_get(id, strerror(-r)); |
ada4135f CH |
3615 | } |
3616 | return r; | |
3617 | } | |
782c3f29 | 3618 | |
8072aae3 AK |
3619 | static bool kvm_accel_has_memory(MachineState *ms, AddressSpace *as, |
3620 | hwaddr start_addr, hwaddr size) | |
3621 | { | |
3622 | KVMState *kvm = KVM_STATE(ms->accelerator); | |
3623 | int i; | |
3624 | ||
3625 | for (i = 0; i < kvm->nr_as; ++i) { | |
3626 | if (kvm->as[i].as == as && kvm->as[i].ml) { | |
023ae9a8 | 3627 | size = MIN(kvm_max_slot_size, size); |
8072aae3 AK |
3628 | return NULL != kvm_lookup_matching_slot(kvm->as[i].ml, |
3629 | start_addr, size); | |
3630 | } | |
3631 | } | |
3632 | ||
3633 | return false; | |
3634 | } | |
3635 | ||
23b0898e PB |
3636 | static void kvm_get_kvm_shadow_mem(Object *obj, Visitor *v, |
3637 | const char *name, void *opaque, | |
3638 | Error **errp) | |
3639 | { | |
3640 | KVMState *s = KVM_STATE(obj); | |
3641 | int64_t value = s->kvm_shadow_mem; | |
3642 | ||
3643 | visit_type_int(v, name, &value, errp); | |
3644 | } | |
3645 | ||
3646 | static void kvm_set_kvm_shadow_mem(Object *obj, Visitor *v, | |
3647 | const char *name, void *opaque, | |
3648 | Error **errp) | |
3649 | { | |
3650 | KVMState *s = KVM_STATE(obj); | |
23b0898e PB |
3651 | int64_t value; |
3652 | ||
70cbae42 PB |
3653 | if (s->fd != -1) { |
3654 | error_setg(errp, "Cannot set properties after the accelerator has been initialized"); | |
3655 | return; | |
3656 | } | |
3657 | ||
668f62ec | 3658 | if (!visit_type_int(v, name, &value, errp)) { |
23b0898e PB |
3659 | return; |
3660 | } | |
3661 | ||
3662 | s->kvm_shadow_mem = value; | |
3663 | } | |
3664 | ||
11bc4a13 PB |
3665 | static void kvm_set_kernel_irqchip(Object *obj, Visitor *v, |
3666 | const char *name, void *opaque, | |
3667 | Error **errp) | |
3668 | { | |
11bc4a13 PB |
3669 | KVMState *s = KVM_STATE(obj); |
3670 | OnOffSplit mode; | |
3671 | ||
70cbae42 PB |
3672 | if (s->fd != -1) { |
3673 | error_setg(errp, "Cannot set properties after the accelerator has been initialized"); | |
3674 | return; | |
3675 | } | |
3676 | ||
14217038 | 3677 | if (!visit_type_OnOffSplit(v, name, &mode, errp)) { |
11bc4a13 | 3678 | return; |
14217038 MA |
3679 | } |
3680 | switch (mode) { | |
3681 | case ON_OFF_SPLIT_ON: | |
3682 | s->kernel_irqchip_allowed = true; | |
3683 | s->kernel_irqchip_required = true; | |
3684 | s->kernel_irqchip_split = ON_OFF_AUTO_OFF; | |
3685 | break; | |
3686 | case ON_OFF_SPLIT_OFF: | |
3687 | s->kernel_irqchip_allowed = false; | |
3688 | s->kernel_irqchip_required = false; | |
3689 | s->kernel_irqchip_split = ON_OFF_AUTO_OFF; | |
3690 | break; | |
3691 | case ON_OFF_SPLIT_SPLIT: | |
3692 | s->kernel_irqchip_allowed = true; | |
3693 | s->kernel_irqchip_required = true; | |
3694 | s->kernel_irqchip_split = ON_OFF_AUTO_ON; | |
3695 | break; | |
3696 | default: | |
3697 | /* The value was checked in visit_type_OnOffSplit() above. If | |
3698 | * we get here, then something is wrong in QEMU. | |
3699 | */ | |
3700 | abort(); | |
11bc4a13 PB |
3701 | } |
3702 | } | |
3703 | ||
4376c40d PB |
3704 | bool kvm_kernel_irqchip_allowed(void) |
3705 | { | |
11bc4a13 | 3706 | return kvm_state->kernel_irqchip_allowed; |
4376c40d PB |
3707 | } |
3708 | ||
3709 | bool kvm_kernel_irqchip_required(void) | |
3710 | { | |
11bc4a13 | 3711 | return kvm_state->kernel_irqchip_required; |
4376c40d PB |
3712 | } |
3713 | ||
3714 | bool kvm_kernel_irqchip_split(void) | |
3715 | { | |
d1972be1 | 3716 | return kvm_state->kernel_irqchip_split == ON_OFF_AUTO_ON; |
4376c40d PB |
3717 | } |
3718 | ||
2ea5cb0a PX |
3719 | static void kvm_get_dirty_ring_size(Object *obj, Visitor *v, |
3720 | const char *name, void *opaque, | |
3721 | Error **errp) | |
3722 | { | |
3723 | KVMState *s = KVM_STATE(obj); | |
3724 | uint32_t value = s->kvm_dirty_ring_size; | |
3725 | ||
3726 | visit_type_uint32(v, name, &value, errp); | |
3727 | } | |
3728 | ||
3729 | static void kvm_set_dirty_ring_size(Object *obj, Visitor *v, | |
3730 | const char *name, void *opaque, | |
3731 | Error **errp) | |
3732 | { | |
3733 | KVMState *s = KVM_STATE(obj); | |
2ea5cb0a PX |
3734 | uint32_t value; |
3735 | ||
3736 | if (s->fd != -1) { | |
3737 | error_setg(errp, "Cannot set properties after the accelerator has been initialized"); | |
3738 | return; | |
3739 | } | |
3740 | ||
d1c81c34 | 3741 | if (!visit_type_uint32(v, name, &value, errp)) { |
2ea5cb0a PX |
3742 | return; |
3743 | } | |
3744 | if (value & (value - 1)) { | |
3745 | error_setg(errp, "dirty-ring-size must be a power of two."); | |
3746 | return; | |
3747 | } | |
3748 | ||
3749 | s->kvm_dirty_ring_size = value; | |
3750 | } | |
3751 | ||
23b0898e PB |
3752 | static void kvm_accel_instance_init(Object *obj) |
3753 | { | |
3754 | KVMState *s = KVM_STATE(obj); | |
3755 | ||
70cbae42 PB |
3756 | s->fd = -1; |
3757 | s->vmfd = -1; | |
23b0898e | 3758 | s->kvm_shadow_mem = -1; |
d1972be1 XL |
3759 | s->kernel_irqchip_allowed = true; |
3760 | s->kernel_irqchip_split = ON_OFF_AUTO_AUTO; | |
2ea5cb0a PX |
3761 | /* KVM dirty ring is by default off */ |
3762 | s->kvm_dirty_ring_size = 0; | |
b20cc776 | 3763 | s->kvm_dirty_ring_with_bitmap = false; |
e2e69f6b CQ |
3764 | s->notify_vmexit = NOTIFY_VMEXIT_OPTION_RUN; |
3765 | s->notify_window = 0; | |
61491cf4 | 3766 | s->xen_version = 0; |
6f43f2ee | 3767 | s->xen_gnttab_max_frames = 64; |
e16aff4c | 3768 | s->xen_evtchn_max_pirq = 256; |
23b0898e PB |
3769 | } |
3770 | ||
3b7a9388 AB |
3771 | /** |
3772 | * kvm_gdbstub_sstep_flags(): | |
3773 | * | |
3774 | * Returns: SSTEP_* flags that KVM supports for guest debug. The | |
3775 | * support is probed during kvm_init() | |
3776 | */ | |
3777 | static int kvm_gdbstub_sstep_flags(void) | |
3778 | { | |
3779 | return kvm_sstep_flags; | |
3780 | } | |
3781 | ||
782c3f29 EH |
3782 | static void kvm_accel_class_init(ObjectClass *oc, void *data) |
3783 | { | |
3784 | AccelClass *ac = ACCEL_CLASS(oc); | |
3785 | ac->name = "KVM"; | |
0d15da8e | 3786 | ac->init_machine = kvm_init; |
8072aae3 | 3787 | ac->has_memory = kvm_accel_has_memory; |
782c3f29 | 3788 | ac->allowed = &kvm_allowed; |
3b7a9388 | 3789 | ac->gdbstub_supported_sstep_flags = kvm_gdbstub_sstep_flags; |
23b0898e | 3790 | |
11bc4a13 PB |
3791 | object_class_property_add(oc, "kernel-irqchip", "on|off|split", |
3792 | NULL, kvm_set_kernel_irqchip, | |
d2623129 | 3793 | NULL, NULL); |
11bc4a13 | 3794 | object_class_property_set_description(oc, "kernel-irqchip", |
7eecec7d | 3795 | "Configure KVM in-kernel irqchip"); |
11bc4a13 | 3796 | |
23b0898e PB |
3797 | object_class_property_add(oc, "kvm-shadow-mem", "int", |
3798 | kvm_get_kvm_shadow_mem, kvm_set_kvm_shadow_mem, | |
d2623129 | 3799 | NULL, NULL); |
23b0898e | 3800 | object_class_property_set_description(oc, "kvm-shadow-mem", |
7eecec7d | 3801 | "KVM shadow MMU size"); |
2ea5cb0a PX |
3802 | |
3803 | object_class_property_add(oc, "dirty-ring-size", "uint32", | |
3804 | kvm_get_dirty_ring_size, kvm_set_dirty_ring_size, | |
3805 | NULL, NULL); | |
3806 | object_class_property_set_description(oc, "dirty-ring-size", | |
3807 | "Size of KVM dirty page ring buffer (default: 0, i.e. use bitmap)"); | |
3dba0a33 PB |
3808 | |
3809 | kvm_arch_accel_class_init(oc); | |
782c3f29 EH |
3810 | } |
3811 | ||
3812 | static const TypeInfo kvm_accel_type = { | |
3813 | .name = TYPE_KVM_ACCEL, | |
3814 | .parent = TYPE_ACCEL, | |
23b0898e | 3815 | .instance_init = kvm_accel_instance_init, |
782c3f29 | 3816 | .class_init = kvm_accel_class_init, |
fc02086b | 3817 | .instance_size = sizeof(KVMState), |
782c3f29 EH |
3818 | }; |
3819 | ||
3820 | static void kvm_type_init(void) | |
3821 | { | |
3822 | type_register_static(&kvm_accel_type); | |
3823 | } | |
3824 | ||
3825 | type_init(kvm_type_init); | |
cc01a3f4 MK |
3826 | |
3827 | typedef struct StatsArgs { | |
3828 | union StatsResultsType { | |
3829 | StatsResultList **stats; | |
3830 | StatsSchemaList **schema; | |
3831 | } result; | |
cf7405bc | 3832 | strList *names; |
cc01a3f4 MK |
3833 | Error **errp; |
3834 | } StatsArgs; | |
3835 | ||
3836 | static StatsList *add_kvmstat_entry(struct kvm_stats_desc *pdesc, | |
3837 | uint64_t *stats_data, | |
3838 | StatsList *stats_list, | |
3839 | Error **errp) | |
3840 | { | |
3841 | ||
3842 | Stats *stats; | |
3843 | uint64List *val_list = NULL; | |
3844 | ||
3845 | /* Only add stats that we understand. */ | |
3846 | switch (pdesc->flags & KVM_STATS_TYPE_MASK) { | |
3847 | case KVM_STATS_TYPE_CUMULATIVE: | |
3848 | case KVM_STATS_TYPE_INSTANT: | |
3849 | case KVM_STATS_TYPE_PEAK: | |
3850 | case KVM_STATS_TYPE_LINEAR_HIST: | |
3851 | case KVM_STATS_TYPE_LOG_HIST: | |
3852 | break; | |
3853 | default: | |
3854 | return stats_list; | |
3855 | } | |
3856 | ||
3857 | switch (pdesc->flags & KVM_STATS_UNIT_MASK) { | |
3858 | case KVM_STATS_UNIT_NONE: | |
3859 | case KVM_STATS_UNIT_BYTES: | |
3860 | case KVM_STATS_UNIT_CYCLES: | |
3861 | case KVM_STATS_UNIT_SECONDS: | |
105bb7cd | 3862 | case KVM_STATS_UNIT_BOOLEAN: |
cc01a3f4 MK |
3863 | break; |
3864 | default: | |
3865 | return stats_list; | |
3866 | } | |
3867 | ||
3868 | switch (pdesc->flags & KVM_STATS_BASE_MASK) { | |
3869 | case KVM_STATS_BASE_POW10: | |
3870 | case KVM_STATS_BASE_POW2: | |
3871 | break; | |
3872 | default: | |
3873 | return stats_list; | |
3874 | } | |
3875 | ||
3876 | /* Alloc and populate data list */ | |
3877 | stats = g_new0(Stats, 1); | |
3878 | stats->name = g_strdup(pdesc->name); | |
3879 | stats->value = g_new0(StatsValue, 1);; | |
3880 | ||
105bb7cd PB |
3881 | if ((pdesc->flags & KVM_STATS_UNIT_MASK) == KVM_STATS_UNIT_BOOLEAN) { |
3882 | stats->value->u.boolean = *stats_data; | |
3883 | stats->value->type = QTYPE_QBOOL; | |
3884 | } else if (pdesc->size == 1) { | |
cc01a3f4 MK |
3885 | stats->value->u.scalar = *stats_data; |
3886 | stats->value->type = QTYPE_QNUM; | |
3887 | } else { | |
3888 | int i; | |
3889 | for (i = 0; i < pdesc->size; i++) { | |
3890 | QAPI_LIST_PREPEND(val_list, stats_data[i]); | |
3891 | } | |
3892 | stats->value->u.list = val_list; | |
3893 | stats->value->type = QTYPE_QLIST; | |
3894 | } | |
3895 | ||
3896 | QAPI_LIST_PREPEND(stats_list, stats); | |
3897 | return stats_list; | |
3898 | } | |
3899 | ||
3900 | static StatsSchemaValueList *add_kvmschema_entry(struct kvm_stats_desc *pdesc, | |
3901 | StatsSchemaValueList *list, | |
3902 | Error **errp) | |
3903 | { | |
3904 | StatsSchemaValueList *schema_entry = g_new0(StatsSchemaValueList, 1); | |
3905 | schema_entry->value = g_new0(StatsSchemaValue, 1); | |
3906 | ||
3907 | switch (pdesc->flags & KVM_STATS_TYPE_MASK) { | |
3908 | case KVM_STATS_TYPE_CUMULATIVE: | |
3909 | schema_entry->value->type = STATS_TYPE_CUMULATIVE; | |
3910 | break; | |
3911 | case KVM_STATS_TYPE_INSTANT: | |
3912 | schema_entry->value->type = STATS_TYPE_INSTANT; | |
3913 | break; | |
3914 | case KVM_STATS_TYPE_PEAK: | |
3915 | schema_entry->value->type = STATS_TYPE_PEAK; | |
3916 | break; | |
3917 | case KVM_STATS_TYPE_LINEAR_HIST: | |
3918 | schema_entry->value->type = STATS_TYPE_LINEAR_HISTOGRAM; | |
3919 | schema_entry->value->bucket_size = pdesc->bucket_size; | |
3920 | schema_entry->value->has_bucket_size = true; | |
3921 | break; | |
3922 | case KVM_STATS_TYPE_LOG_HIST: | |
3923 | schema_entry->value->type = STATS_TYPE_LOG2_HISTOGRAM; | |
3924 | break; | |
3925 | default: | |
3926 | goto exit; | |
3927 | } | |
3928 | ||
3929 | switch (pdesc->flags & KVM_STATS_UNIT_MASK) { | |
3930 | case KVM_STATS_UNIT_NONE: | |
3931 | break; | |
105bb7cd PB |
3932 | case KVM_STATS_UNIT_BOOLEAN: |
3933 | schema_entry->value->has_unit = true; | |
3934 | schema_entry->value->unit = STATS_UNIT_BOOLEAN; | |
3935 | break; | |
cc01a3f4 MK |
3936 | case KVM_STATS_UNIT_BYTES: |
3937 | schema_entry->value->has_unit = true; | |
3938 | schema_entry->value->unit = STATS_UNIT_BYTES; | |
3939 | break; | |
3940 | case KVM_STATS_UNIT_CYCLES: | |
3941 | schema_entry->value->has_unit = true; | |
3942 | schema_entry->value->unit = STATS_UNIT_CYCLES; | |
3943 | break; | |
3944 | case KVM_STATS_UNIT_SECONDS: | |
3945 | schema_entry->value->has_unit = true; | |
3946 | schema_entry->value->unit = STATS_UNIT_SECONDS; | |
3947 | break; | |
3948 | default: | |
3949 | goto exit; | |
3950 | } | |
3951 | ||
3952 | schema_entry->value->exponent = pdesc->exponent; | |
3953 | if (pdesc->exponent) { | |
3954 | switch (pdesc->flags & KVM_STATS_BASE_MASK) { | |
3955 | case KVM_STATS_BASE_POW10: | |
3956 | schema_entry->value->has_base = true; | |
3957 | schema_entry->value->base = 10; | |
3958 | break; | |
3959 | case KVM_STATS_BASE_POW2: | |
3960 | schema_entry->value->has_base = true; | |
3961 | schema_entry->value->base = 2; | |
3962 | break; | |
3963 | default: | |
3964 | goto exit; | |
3965 | } | |
3966 | } | |
3967 | ||
3968 | schema_entry->value->name = g_strdup(pdesc->name); | |
3969 | schema_entry->next = list; | |
3970 | return schema_entry; | |
3971 | exit: | |
3972 | g_free(schema_entry->value); | |
3973 | g_free(schema_entry); | |
3974 | return list; | |
3975 | } | |
3976 | ||
3977 | /* Cached stats descriptors */ | |
3978 | typedef struct StatsDescriptors { | |
3979 | const char *ident; /* cache key, currently the StatsTarget */ | |
3980 | struct kvm_stats_desc *kvm_stats_desc; | |
21adec30 | 3981 | struct kvm_stats_header kvm_stats_header; |
cc01a3f4 MK |
3982 | QTAILQ_ENTRY(StatsDescriptors) next; |
3983 | } StatsDescriptors; | |
3984 | ||
3985 | static QTAILQ_HEAD(, StatsDescriptors) stats_descriptors = | |
3986 | QTAILQ_HEAD_INITIALIZER(stats_descriptors); | |
3987 | ||
3988 | /* | |
3989 | * Return the descriptors for 'target', that either have already been read | |
3990 | * or are retrieved from 'stats_fd'. | |
3991 | */ | |
3992 | static StatsDescriptors *find_stats_descriptors(StatsTarget target, int stats_fd, | |
3993 | Error **errp) | |
3994 | { | |
3995 | StatsDescriptors *descriptors; | |
3996 | const char *ident; | |
3997 | struct kvm_stats_desc *kvm_stats_desc; | |
3998 | struct kvm_stats_header *kvm_stats_header; | |
3999 | size_t size_desc; | |
4000 | ssize_t ret; | |
4001 | ||
4002 | ident = StatsTarget_str(target); | |
4003 | QTAILQ_FOREACH(descriptors, &stats_descriptors, next) { | |
4004 | if (g_str_equal(descriptors->ident, ident)) { | |
4005 | return descriptors; | |
4006 | } | |
4007 | } | |
4008 | ||
4009 | descriptors = g_new0(StatsDescriptors, 1); | |
4010 | ||
4011 | /* Read stats header */ | |
21adec30 | 4012 | kvm_stats_header = &descriptors->kvm_stats_header; |
3b6f4852 | 4013 | ret = pread(stats_fd, kvm_stats_header, sizeof(*kvm_stats_header), 0); |
cc01a3f4 MK |
4014 | if (ret != sizeof(*kvm_stats_header)) { |
4015 | error_setg(errp, "KVM stats: failed to read stats header: " | |
4016 | "expected %zu actual %zu", | |
4017 | sizeof(*kvm_stats_header), ret); | |
f696b74b | 4018 | g_free(descriptors); |
cc01a3f4 MK |
4019 | return NULL; |
4020 | } | |
4021 | size_desc = sizeof(*kvm_stats_desc) + kvm_stats_header->name_size; | |
4022 | ||
4023 | /* Read stats descriptors */ | |
4024 | kvm_stats_desc = g_malloc0_n(kvm_stats_header->num_desc, size_desc); | |
4025 | ret = pread(stats_fd, kvm_stats_desc, | |
4026 | size_desc * kvm_stats_header->num_desc, | |
4027 | kvm_stats_header->desc_offset); | |
4028 | ||
4029 | if (ret != size_desc * kvm_stats_header->num_desc) { | |
4030 | error_setg(errp, "KVM stats: failed to read stats descriptors: " | |
4031 | "expected %zu actual %zu", | |
4032 | size_desc * kvm_stats_header->num_desc, ret); | |
4033 | g_free(descriptors); | |
4034 | g_free(kvm_stats_desc); | |
4035 | return NULL; | |
4036 | } | |
cc01a3f4 MK |
4037 | descriptors->kvm_stats_desc = kvm_stats_desc; |
4038 | descriptors->ident = ident; | |
4039 | QTAILQ_INSERT_TAIL(&stats_descriptors, descriptors, next); | |
4040 | return descriptors; | |
4041 | } | |
4042 | ||
4043 | static void query_stats(StatsResultList **result, StatsTarget target, | |
3b6f4852 MT |
4044 | strList *names, int stats_fd, CPUState *cpu, |
4045 | Error **errp) | |
cc01a3f4 MK |
4046 | { |
4047 | struct kvm_stats_desc *kvm_stats_desc; | |
4048 | struct kvm_stats_header *kvm_stats_header; | |
4049 | StatsDescriptors *descriptors; | |
4050 | g_autofree uint64_t *stats_data = NULL; | |
4051 | struct kvm_stats_desc *pdesc; | |
4052 | StatsList *stats_list = NULL; | |
4053 | size_t size_desc, size_data = 0; | |
4054 | ssize_t ret; | |
4055 | int i; | |
4056 | ||
4057 | descriptors = find_stats_descriptors(target, stats_fd, errp); | |
4058 | if (!descriptors) { | |
4059 | return; | |
4060 | } | |
4061 | ||
21adec30 | 4062 | kvm_stats_header = &descriptors->kvm_stats_header; |
cc01a3f4 MK |
4063 | kvm_stats_desc = descriptors->kvm_stats_desc; |
4064 | size_desc = sizeof(*kvm_stats_desc) + kvm_stats_header->name_size; | |
4065 | ||
4066 | /* Tally the total data size; read schema data */ | |
4067 | for (i = 0; i < kvm_stats_header->num_desc; ++i) { | |
4068 | pdesc = (void *)kvm_stats_desc + i * size_desc; | |
4069 | size_data += pdesc->size * sizeof(*stats_data); | |
4070 | } | |
4071 | ||
4072 | stats_data = g_malloc0(size_data); | |
4073 | ret = pread(stats_fd, stats_data, size_data, kvm_stats_header->data_offset); | |
4074 | ||
4075 | if (ret != size_data) { | |
4076 | error_setg(errp, "KVM stats: failed to read data: " | |
4077 | "expected %zu actual %zu", size_data, ret); | |
4078 | return; | |
4079 | } | |
4080 | ||
4081 | for (i = 0; i < kvm_stats_header->num_desc; ++i) { | |
4082 | uint64_t *stats; | |
4083 | pdesc = (void *)kvm_stats_desc + i * size_desc; | |
4084 | ||
4085 | /* Add entry to the list */ | |
4086 | stats = (void *)stats_data + pdesc->offset; | |
cf7405bc PB |
4087 | if (!apply_str_list_filter(pdesc->name, names)) { |
4088 | continue; | |
4089 | } | |
cc01a3f4 MK |
4090 | stats_list = add_kvmstat_entry(pdesc, stats, stats_list, errp); |
4091 | } | |
4092 | ||
4093 | if (!stats_list) { | |
4094 | return; | |
4095 | } | |
4096 | ||
4097 | switch (target) { | |
4098 | case STATS_TARGET_VM: | |
4099 | add_stats_entry(result, STATS_PROVIDER_KVM, NULL, stats_list); | |
4100 | break; | |
4101 | case STATS_TARGET_VCPU: | |
4102 | add_stats_entry(result, STATS_PROVIDER_KVM, | |
3b6f4852 | 4103 | cpu->parent_obj.canonical_path, |
cc01a3f4 MK |
4104 | stats_list); |
4105 | break; | |
4106 | default: | |
d12dd9c7 | 4107 | g_assert_not_reached(); |
cc01a3f4 MK |
4108 | } |
4109 | } | |
4110 | ||
4111 | static void query_stats_schema(StatsSchemaList **result, StatsTarget target, | |
4112 | int stats_fd, Error **errp) | |
4113 | { | |
4114 | struct kvm_stats_desc *kvm_stats_desc; | |
4115 | struct kvm_stats_header *kvm_stats_header; | |
4116 | StatsDescriptors *descriptors; | |
4117 | struct kvm_stats_desc *pdesc; | |
4118 | StatsSchemaValueList *stats_list = NULL; | |
4119 | size_t size_desc; | |
4120 | int i; | |
4121 | ||
4122 | descriptors = find_stats_descriptors(target, stats_fd, errp); | |
4123 | if (!descriptors) { | |
4124 | return; | |
4125 | } | |
4126 | ||
21adec30 | 4127 | kvm_stats_header = &descriptors->kvm_stats_header; |
cc01a3f4 MK |
4128 | kvm_stats_desc = descriptors->kvm_stats_desc; |
4129 | size_desc = sizeof(*kvm_stats_desc) + kvm_stats_header->name_size; | |
4130 | ||
4131 | /* Tally the total data size; read schema data */ | |
4132 | for (i = 0; i < kvm_stats_header->num_desc; ++i) { | |
4133 | pdesc = (void *)kvm_stats_desc + i * size_desc; | |
4134 | stats_list = add_kvmschema_entry(pdesc, stats_list, errp); | |
4135 | } | |
4136 | ||
4137 | add_stats_schema(result, STATS_PROVIDER_KVM, target, stats_list); | |
4138 | } | |
4139 | ||
3b6f4852 | 4140 | static void query_stats_vcpu(CPUState *cpu, StatsArgs *kvm_stats_args) |
cc01a3f4 | 4141 | { |
3b6f4852 | 4142 | int stats_fd = cpu->kvm_vcpu_stats_fd; |
cc01a3f4 MK |
4143 | Error *local_err = NULL; |
4144 | ||
4145 | if (stats_fd == -1) { | |
4146 | error_setg_errno(&local_err, errno, "KVM stats: ioctl failed"); | |
4147 | error_propagate(kvm_stats_args->errp, local_err); | |
4148 | return; | |
4149 | } | |
cf7405bc | 4150 | query_stats(kvm_stats_args->result.stats, STATS_TARGET_VCPU, |
3b6f4852 MT |
4151 | kvm_stats_args->names, stats_fd, cpu, |
4152 | kvm_stats_args->errp); | |
cc01a3f4 MK |
4153 | } |
4154 | ||
3b6f4852 | 4155 | static void query_stats_schema_vcpu(CPUState *cpu, StatsArgs *kvm_stats_args) |
cc01a3f4 | 4156 | { |
3b6f4852 | 4157 | int stats_fd = cpu->kvm_vcpu_stats_fd; |
cc01a3f4 MK |
4158 | Error *local_err = NULL; |
4159 | ||
4160 | if (stats_fd == -1) { | |
4161 | error_setg_errno(&local_err, errno, "KVM stats: ioctl failed"); | |
4162 | error_propagate(kvm_stats_args->errp, local_err); | |
4163 | return; | |
4164 | } | |
4165 | query_stats_schema(kvm_stats_args->result.schema, STATS_TARGET_VCPU, stats_fd, | |
4166 | kvm_stats_args->errp); | |
cc01a3f4 MK |
4167 | } |
4168 | ||
467ef823 | 4169 | static void query_stats_cb(StatsResultList **result, StatsTarget target, |
cf7405bc | 4170 | strList *names, strList *targets, Error **errp) |
cc01a3f4 MK |
4171 | { |
4172 | KVMState *s = kvm_state; | |
4173 | CPUState *cpu; | |
4174 | int stats_fd; | |
4175 | ||
4176 | switch (target) { | |
4177 | case STATS_TARGET_VM: | |
4178 | { | |
4179 | stats_fd = kvm_vm_ioctl(s, KVM_GET_STATS_FD, NULL); | |
4180 | if (stats_fd == -1) { | |
4181 | error_setg_errno(errp, errno, "KVM stats: ioctl failed"); | |
4182 | return; | |
4183 | } | |
3b6f4852 | 4184 | query_stats(result, target, names, stats_fd, NULL, errp); |
cc01a3f4 MK |
4185 | close(stats_fd); |
4186 | break; | |
4187 | } | |
4188 | case STATS_TARGET_VCPU: | |
4189 | { | |
4190 | StatsArgs stats_args; | |
4191 | stats_args.result.stats = result; | |
cf7405bc | 4192 | stats_args.names = names; |
cc01a3f4 MK |
4193 | stats_args.errp = errp; |
4194 | CPU_FOREACH(cpu) { | |
467ef823 PB |
4195 | if (!apply_str_list_filter(cpu->parent_obj.canonical_path, targets)) { |
4196 | continue; | |
4197 | } | |
3b6f4852 | 4198 | query_stats_vcpu(cpu, &stats_args); |
cc01a3f4 MK |
4199 | } |
4200 | break; | |
4201 | } | |
4202 | default: | |
4203 | break; | |
4204 | } | |
4205 | } | |
4206 | ||
4207 | void query_stats_schemas_cb(StatsSchemaList **result, Error **errp) | |
4208 | { | |
4209 | StatsArgs stats_args; | |
4210 | KVMState *s = kvm_state; | |
4211 | int stats_fd; | |
4212 | ||
4213 | stats_fd = kvm_vm_ioctl(s, KVM_GET_STATS_FD, NULL); | |
4214 | if (stats_fd == -1) { | |
4215 | error_setg_errno(errp, errno, "KVM stats: ioctl failed"); | |
4216 | return; | |
4217 | } | |
4218 | query_stats_schema(result, STATS_TARGET_VM, stats_fd, errp); | |
4219 | close(stats_fd); | |
4220 | ||
a9197ad2 PB |
4221 | if (first_cpu) { |
4222 | stats_args.result.schema = result; | |
4223 | stats_args.errp = errp; | |
3b6f4852 | 4224 | query_stats_schema_vcpu(first_cpu, &stats_args); |
a9197ad2 | 4225 | } |
cc01a3f4 | 4226 | } |