]>
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" |
1de7afc9 | 34 | #include "qemu/bswap.h" |
022c62cb | 35 | #include "exec/memory.h" |
747afd5b | 36 | #include "exec/ram_addr.h" |
1de7afc9 | 37 | #include "qemu/event_notifier.h" |
db725815 | 38 | #include "qemu/main-loop.h" |
92229a57 | 39 | #include "trace.h" |
197e3524 | 40 | #include "hw/irq.h" |
23b0898e | 41 | #include "qapi/visitor.h" |
11bc4a13 PB |
42 | #include "qapi/qapi-types-common.h" |
43 | #include "qapi/qapi-visit-common.h" | |
6b552b9b | 44 | #include "sysemu/reset.h" |
57038a92 CF |
45 | #include "qemu/guest-random.h" |
46 | #include "sysemu/hw_accel.h" | |
47 | #include "kvm-cpus.h" | |
baa60983 | 48 | #include "sysemu/dirtylimit.h" |
05330448 | 49 | |
135a129a | 50 | #include "hw/boards.h" |
cc01a3f4 | 51 | #include "monitor/stats.h" |
135a129a | 52 | |
d2f2b8a7 SH |
53 | /* This check must be after config-host.h is included */ |
54 | #ifdef CONFIG_EVENTFD | |
55 | #include <sys/eventfd.h> | |
56 | #endif | |
57 | ||
bc92e4e9 AJ |
58 | /* KVM uses PAGE_SIZE in its definition of KVM_COALESCED_MMIO_MAX. We |
59 | * need to use the real host PAGE_SIZE, as that's what KVM will use. | |
60 | */ | |
eb8b1a79 JY |
61 | #ifdef PAGE_SIZE |
62 | #undef PAGE_SIZE | |
63 | #endif | |
8e3b0cbb | 64 | #define PAGE_SIZE qemu_real_host_page_size() |
f65ed4c1 | 65 | |
fd2ddd16 ML |
66 | #ifndef KVM_GUESTDBG_BLOCKIRQ |
67 | #define KVM_GUESTDBG_BLOCKIRQ 0 | |
68 | #endif | |
69 | ||
05330448 AL |
70 | //#define DEBUG_KVM |
71 | ||
72 | #ifdef DEBUG_KVM | |
8c0d577e | 73 | #define DPRINTF(fmt, ...) \ |
05330448 AL |
74 | do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0) |
75 | #else | |
8c0d577e | 76 | #define DPRINTF(fmt, ...) \ |
05330448 AL |
77 | do { } while (0) |
78 | #endif | |
79 | ||
04fa27f5 JK |
80 | #define KVM_MSI_HASHTAB_SIZE 256 |
81 | ||
4c055ab5 GZ |
82 | struct KVMParkedVcpu { |
83 | unsigned long vcpu_id; | |
84 | int kvm_fd; | |
85 | QLIST_ENTRY(KVMParkedVcpu) node; | |
86 | }; | |
87 | ||
b4420f19 PX |
88 | enum KVMDirtyRingReaperState { |
89 | KVM_DIRTY_RING_REAPER_NONE = 0, | |
90 | /* The reaper is sleeping */ | |
91 | KVM_DIRTY_RING_REAPER_WAIT, | |
92 | /* The reaper is reaping for dirty pages */ | |
93 | KVM_DIRTY_RING_REAPER_REAPING, | |
94 | }; | |
95 | ||
96 | /* | |
97 | * KVM reaper instance, responsible for collecting the KVM dirty bits | |
98 | * via the dirty ring. | |
99 | */ | |
100 | struct KVMDirtyRingReaper { | |
101 | /* The reaper thread */ | |
102 | QemuThread reaper_thr; | |
103 | volatile uint64_t reaper_iteration; /* iteration number of reaper thr */ | |
104 | volatile enum KVMDirtyRingReaperState reaper_state; /* reap thr state */ | |
105 | }; | |
106 | ||
9d1c35df | 107 | struct KVMState |
05330448 | 108 | { |
fc02086b EH |
109 | AccelState parent_obj; |
110 | ||
fb541ca5 | 111 | int nr_slots; |
05330448 AL |
112 | int fd; |
113 | int vmfd; | |
f65ed4c1 | 114 | int coalesced_mmio; |
e6d34aee | 115 | int coalesced_pio; |
62a2744c | 116 | struct kvm_coalesced_mmio_ring *coalesced_mmio_ring; |
1cae88b9 | 117 | bool coalesced_flush_in_progress; |
a0fb002c | 118 | int vcpu_events; |
b0b1d690 | 119 | int robust_singlestep; |
ff44f1a3 | 120 | int debugregs; |
e22a25c9 | 121 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
b58deb34 | 122 | QTAILQ_HEAD(, kvm_sw_breakpoint) kvm_sw_breakpoints; |
e22a25c9 | 123 | #endif |
ebbfef2f | 124 | int max_nested_state_len; |
d2f2b8a7 | 125 | int many_ioeventfds; |
3ab73842 | 126 | int intx_set_mask; |
23b0898e | 127 | int kvm_shadow_mem; |
11bc4a13 PB |
128 | bool kernel_irqchip_allowed; |
129 | bool kernel_irqchip_required; | |
d1972be1 | 130 | OnOffAuto kernel_irqchip_split; |
62dd4eda | 131 | bool sync_mmu; |
494cd11d | 132 | uint64_t manual_dirty_log_protect; |
92e4b519 DG |
133 | /* The man page (and posix) say ioctl numbers are signed int, but |
134 | * they're not. Linux, glibc and *BSD all treat ioctl numbers as | |
135 | * unsigned, and treating them as signed here can break things */ | |
e333cd69 | 136 | unsigned irq_set_ioctl; |
aed6efb9 | 137 | unsigned int sigmask_len; |
197e3524 | 138 | GHashTable *gsimap; |
84b058d7 JK |
139 | #ifdef KVM_CAP_IRQ_ROUTING |
140 | struct kvm_irq_routing *irq_routes; | |
141 | int nr_allocated_irq_routes; | |
8269fb70 | 142 | unsigned long *used_gsi_bitmap; |
4e2e4e63 | 143 | unsigned int gsi_count; |
b58deb34 | 144 | QTAILQ_HEAD(, KVMMSIRoute) msi_hashtab[KVM_MSI_HASHTAB_SIZE]; |
84b058d7 | 145 | #endif |
7bbda04c | 146 | KVMMemoryListener memory_listener; |
4c055ab5 | 147 | QLIST_HEAD(, KVMParkedVcpu) kvm_parked_vcpus; |
b20e3780 | 148 | |
8072aae3 AK |
149 | /* For "info mtree -f" to tell if an MR is registered in KVM */ |
150 | int nr_as; | |
151 | struct KVMAs { | |
152 | KVMMemoryListener *ml; | |
153 | AddressSpace *as; | |
154 | } *as; | |
2ea5cb0a PX |
155 | uint64_t kvm_dirty_ring_bytes; /* Size of the per-vcpu dirty ring */ |
156 | uint32_t kvm_dirty_ring_size; /* Number of dirty GFNs per ring */ | |
b4420f19 | 157 | struct KVMDirtyRingReaper reaper; |
9d1c35df | 158 | }; |
05330448 | 159 | |
6a7af8cb | 160 | KVMState *kvm_state; |
3d4b2649 | 161 | bool kvm_kernel_irqchip; |
15eafc2e | 162 | bool kvm_split_irqchip; |
7ae26bd4 | 163 | bool kvm_async_interrupts_allowed; |
215e79c0 | 164 | bool kvm_halt_in_kernel_allowed; |
69e03ae6 | 165 | bool kvm_eventfds_allowed; |
cc7e0ddf | 166 | bool kvm_irqfds_allowed; |
f41389ae | 167 | bool kvm_resamplefds_allowed; |
614e41bc | 168 | bool kvm_msi_via_irqfd_allowed; |
f3e1bed8 | 169 | bool kvm_gsi_routing_allowed; |
76fe21de | 170 | bool kvm_gsi_direct_mapping; |
13eed94e | 171 | bool kvm_allowed; |
df9c8b75 | 172 | bool kvm_readonly_mem_allowed; |
d0a073a1 | 173 | bool kvm_vm_attributes_allowed; |
50bf31b9 | 174 | bool kvm_direct_msi_allowed; |
35108223 | 175 | bool kvm_ioeventfd_any_length_allowed; |
767a554a | 176 | bool kvm_msi_use_devid; |
12bc5b4c ML |
177 | bool kvm_has_guest_debug; |
178 | int kvm_sstep_flags; | |
cf0f7cf9 | 179 | static bool kvm_immediate_exit; |
023ae9a8 | 180 | static hwaddr kvm_max_slot_size = ~0; |
05330448 | 181 | |
94a8d39a JK |
182 | static const KVMCapabilityInfo kvm_required_capabilites[] = { |
183 | KVM_CAP_INFO(USER_MEMORY), | |
184 | KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS), | |
89de4b91 | 185 | KVM_CAP_INFO(JOIN_MEMORY_REGIONS_WORKS), |
94a8d39a JK |
186 | KVM_CAP_LAST_INFO |
187 | }; | |
188 | ||
3607715a DG |
189 | static NotifierList kvm_irqchip_change_notifiers = |
190 | NOTIFIER_LIST_INITIALIZER(kvm_irqchip_change_notifiers); | |
191 | ||
c82d9d43 PX |
192 | struct KVMResampleFd { |
193 | int gsi; | |
194 | EventNotifier *resample_event; | |
195 | QLIST_ENTRY(KVMResampleFd) node; | |
196 | }; | |
197 | typedef struct KVMResampleFd KVMResampleFd; | |
198 | ||
199 | /* | |
200 | * Only used with split irqchip where we need to do the resample fd | |
201 | * kick for the kernel from userspace. | |
202 | */ | |
203 | static QLIST_HEAD(, KVMResampleFd) kvm_resample_fd_list = | |
204 | QLIST_HEAD_INITIALIZER(kvm_resample_fd_list); | |
205 | ||
a2f77862 PX |
206 | static QemuMutex kml_slots_lock; |
207 | ||
208 | #define kvm_slots_lock() qemu_mutex_lock(&kml_slots_lock) | |
209 | #define kvm_slots_unlock() qemu_mutex_unlock(&kml_slots_lock) | |
36adac49 | 210 | |
ea776d15 PX |
211 | static void kvm_slot_init_dirty_bitmap(KVMSlot *mem); |
212 | ||
c82d9d43 PX |
213 | static inline void kvm_resample_fd_remove(int gsi) |
214 | { | |
215 | KVMResampleFd *rfd; | |
216 | ||
217 | QLIST_FOREACH(rfd, &kvm_resample_fd_list, node) { | |
218 | if (rfd->gsi == gsi) { | |
219 | QLIST_REMOVE(rfd, node); | |
220 | g_free(rfd); | |
221 | break; | |
222 | } | |
223 | } | |
224 | } | |
225 | ||
226 | static inline void kvm_resample_fd_insert(int gsi, EventNotifier *event) | |
227 | { | |
228 | KVMResampleFd *rfd = g_new0(KVMResampleFd, 1); | |
229 | ||
230 | rfd->gsi = gsi; | |
231 | rfd->resample_event = event; | |
232 | ||
233 | QLIST_INSERT_HEAD(&kvm_resample_fd_list, rfd, node); | |
234 | } | |
235 | ||
236 | void kvm_resample_fd_notify(int gsi) | |
237 | { | |
238 | KVMResampleFd *rfd; | |
239 | ||
240 | QLIST_FOREACH(rfd, &kvm_resample_fd_list, node) { | |
241 | if (rfd->gsi == gsi) { | |
242 | event_notifier_set(rfd->resample_event); | |
243 | trace_kvm_resample_fd_notify(gsi); | |
244 | return; | |
245 | } | |
246 | } | |
247 | } | |
248 | ||
44f2e6c1 BR |
249 | int kvm_get_max_memslots(void) |
250 | { | |
4f7f5893 | 251 | KVMState *s = KVM_STATE(current_accel()); |
44f2e6c1 BR |
252 | |
253 | return s->nr_slots; | |
254 | } | |
255 | ||
36adac49 | 256 | /* Called with KVMMemoryListener.slots_lock held */ |
7bbda04c | 257 | static KVMSlot *kvm_get_free_slot(KVMMemoryListener *kml) |
05330448 | 258 | { |
7bbda04c | 259 | KVMState *s = kvm_state; |
05330448 AL |
260 | int i; |
261 | ||
fb541ca5 | 262 | for (i = 0; i < s->nr_slots; i++) { |
7bbda04c PB |
263 | if (kml->slots[i].memory_size == 0) { |
264 | return &kml->slots[i]; | |
a426e122 | 265 | } |
05330448 AL |
266 | } |
267 | ||
b8865591 IM |
268 | return NULL; |
269 | } | |
270 | ||
271 | bool kvm_has_free_slot(MachineState *ms) | |
272 | { | |
7bbda04c | 273 | KVMState *s = KVM_STATE(ms->accelerator); |
36adac49 PX |
274 | bool result; |
275 | KVMMemoryListener *kml = &s->memory_listener; | |
276 | ||
a2f77862 | 277 | kvm_slots_lock(); |
36adac49 | 278 | result = !!kvm_get_free_slot(kml); |
a2f77862 | 279 | kvm_slots_unlock(); |
7bbda04c | 280 | |
36adac49 | 281 | return result; |
b8865591 IM |
282 | } |
283 | ||
36adac49 | 284 | /* Called with KVMMemoryListener.slots_lock held */ |
7bbda04c | 285 | static KVMSlot *kvm_alloc_slot(KVMMemoryListener *kml) |
b8865591 | 286 | { |
7bbda04c | 287 | KVMSlot *slot = kvm_get_free_slot(kml); |
b8865591 IM |
288 | |
289 | if (slot) { | |
290 | return slot; | |
291 | } | |
292 | ||
d3f8d37f AL |
293 | fprintf(stderr, "%s: no free slot available\n", __func__); |
294 | abort(); | |
295 | } | |
296 | ||
7bbda04c | 297 | static KVMSlot *kvm_lookup_matching_slot(KVMMemoryListener *kml, |
a8170e5e | 298 | hwaddr start_addr, |
2747e716 | 299 | hwaddr size) |
d3f8d37f | 300 | { |
7bbda04c | 301 | KVMState *s = kvm_state; |
d3f8d37f AL |
302 | int i; |
303 | ||
fb541ca5 | 304 | for (i = 0; i < s->nr_slots; i++) { |
7bbda04c | 305 | KVMSlot *mem = &kml->slots[i]; |
d3f8d37f | 306 | |
2747e716 | 307 | if (start_addr == mem->start_addr && size == mem->memory_size) { |
d3f8d37f AL |
308 | return mem; |
309 | } | |
310 | } | |
311 | ||
05330448 AL |
312 | return NULL; |
313 | } | |
314 | ||
5ea69c2e DH |
315 | /* |
316 | * Calculate and align the start address and the size of the section. | |
317 | * Return the size. If the size is 0, the aligned section is empty. | |
318 | */ | |
319 | static hwaddr kvm_align_section(MemoryRegionSection *section, | |
320 | hwaddr *start) | |
321 | { | |
322 | hwaddr size = int128_get64(section->size); | |
a6ffc423 | 323 | hwaddr delta, aligned; |
5ea69c2e DH |
324 | |
325 | /* kvm works in page size chunks, but the function may be called | |
326 | with sub-page size and unaligned start address. Pad the start | |
327 | address to next and truncate size to previous page boundary. */ | |
a6ffc423 | 328 | aligned = ROUND_UP(section->offset_within_address_space, |
8e3b0cbb | 329 | qemu_real_host_page_size()); |
a6ffc423 DH |
330 | delta = aligned - section->offset_within_address_space; |
331 | *start = aligned; | |
5ea69c2e DH |
332 | if (delta > size) { |
333 | return 0; | |
334 | } | |
5ea69c2e | 335 | |
8e3b0cbb | 336 | return (size - delta) & qemu_real_host_page_mask(); |
5ea69c2e DH |
337 | } |
338 | ||
9f213ed9 | 339 | int kvm_physical_memory_addr_from_host(KVMState *s, void *ram, |
a8170e5e | 340 | hwaddr *phys_addr) |
983dfc3b | 341 | { |
7bbda04c | 342 | KVMMemoryListener *kml = &s->memory_listener; |
36adac49 | 343 | int i, ret = 0; |
983dfc3b | 344 | |
a2f77862 | 345 | kvm_slots_lock(); |
fb541ca5 | 346 | for (i = 0; i < s->nr_slots; i++) { |
7bbda04c | 347 | KVMSlot *mem = &kml->slots[i]; |
983dfc3b | 348 | |
9f213ed9 AK |
349 | if (ram >= mem->ram && ram < mem->ram + mem->memory_size) { |
350 | *phys_addr = mem->start_addr + (ram - mem->ram); | |
36adac49 PX |
351 | ret = 1; |
352 | break; | |
983dfc3b HY |
353 | } |
354 | } | |
a2f77862 | 355 | kvm_slots_unlock(); |
983dfc3b | 356 | |
36adac49 | 357 | return ret; |
983dfc3b HY |
358 | } |
359 | ||
6c090d4a | 360 | static int kvm_set_user_memory_region(KVMMemoryListener *kml, KVMSlot *slot, bool new) |
5832d1f2 | 361 | { |
7bbda04c | 362 | KVMState *s = kvm_state; |
5832d1f2 | 363 | struct kvm_userspace_memory_region mem; |
fe29141b | 364 | int ret; |
5832d1f2 | 365 | |
38bfe691 | 366 | mem.slot = slot->slot | (kml->as_id << 16); |
5832d1f2 | 367 | mem.guest_phys_addr = slot->start_addr; |
9f213ed9 | 368 | mem.userspace_addr = (unsigned long)slot->ram; |
5832d1f2 | 369 | mem.flags = slot->flags; |
651eb0f4 | 370 | |
6c090d4a | 371 | if (slot->memory_size && !new && (mem.flags ^ slot->old_flags) & KVM_MEM_READONLY) { |
235e8982 JJ |
372 | /* Set the slot size to 0 before setting the slot to the desired |
373 | * value. This is needed based on KVM commit 75d61fbc. */ | |
374 | mem.memory_size = 0; | |
88cd34ee PMD |
375 | ret = kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem); |
376 | if (ret < 0) { | |
377 | goto err; | |
378 | } | |
235e8982 JJ |
379 | } |
380 | mem.memory_size = slot->memory_size; | |
fe29141b | 381 | ret = kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem); |
6c090d4a | 382 | slot->old_flags = mem.flags; |
88cd34ee | 383 | err: |
fe29141b AK |
384 | trace_kvm_set_user_memory(mem.slot, mem.flags, mem.guest_phys_addr, |
385 | mem.memory_size, mem.userspace_addr, ret); | |
88cd34ee PMD |
386 | if (ret < 0) { |
387 | error_report("%s: KVM_SET_USER_MEMORY_REGION failed, slot=%d," | |
388 | " start=0x%" PRIx64 ", size=0x%" PRIx64 ": %s", | |
389 | __func__, mem.slot, slot->start_addr, | |
390 | (uint64_t)mem.memory_size, strerror(errno)); | |
391 | } | |
fe29141b | 392 | return ret; |
5832d1f2 AL |
393 | } |
394 | ||
57038a92 | 395 | static int do_kvm_destroy_vcpu(CPUState *cpu) |
4c055ab5 GZ |
396 | { |
397 | KVMState *s = kvm_state; | |
398 | long mmap_size; | |
399 | struct KVMParkedVcpu *vcpu = NULL; | |
400 | int ret = 0; | |
401 | ||
402 | DPRINTF("kvm_destroy_vcpu\n"); | |
403 | ||
b1115c99 LA |
404 | ret = kvm_arch_destroy_vcpu(cpu); |
405 | if (ret < 0) { | |
406 | goto err; | |
407 | } | |
408 | ||
4c055ab5 GZ |
409 | mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0); |
410 | if (mmap_size < 0) { | |
411 | ret = mmap_size; | |
412 | DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n"); | |
413 | goto err; | |
414 | } | |
415 | ||
416 | ret = munmap(cpu->kvm_run, mmap_size); | |
417 | if (ret < 0) { | |
418 | goto err; | |
419 | } | |
420 | ||
b4420f19 | 421 | if (cpu->kvm_dirty_gfns) { |
dcafa248 | 422 | ret = munmap(cpu->kvm_dirty_gfns, s->kvm_dirty_ring_bytes); |
b4420f19 PX |
423 | if (ret < 0) { |
424 | goto err; | |
425 | } | |
426 | } | |
427 | ||
4c055ab5 GZ |
428 | vcpu = g_malloc0(sizeof(*vcpu)); |
429 | vcpu->vcpu_id = kvm_arch_vcpu_id(cpu); | |
430 | vcpu->kvm_fd = cpu->kvm_fd; | |
431 | QLIST_INSERT_HEAD(&kvm_state->kvm_parked_vcpus, vcpu, node); | |
432 | err: | |
433 | return ret; | |
434 | } | |
435 | ||
57038a92 CF |
436 | void kvm_destroy_vcpu(CPUState *cpu) |
437 | { | |
438 | if (do_kvm_destroy_vcpu(cpu) < 0) { | |
439 | error_report("kvm_destroy_vcpu failed"); | |
440 | exit(EXIT_FAILURE); | |
441 | } | |
442 | } | |
443 | ||
4c055ab5 GZ |
444 | static int kvm_get_vcpu(KVMState *s, unsigned long vcpu_id) |
445 | { | |
446 | struct KVMParkedVcpu *cpu; | |
447 | ||
448 | QLIST_FOREACH(cpu, &s->kvm_parked_vcpus, node) { | |
449 | if (cpu->vcpu_id == vcpu_id) { | |
450 | int kvm_fd; | |
451 | ||
452 | QLIST_REMOVE(cpu, node); | |
453 | kvm_fd = cpu->kvm_fd; | |
454 | g_free(cpu); | |
455 | return kvm_fd; | |
456 | } | |
457 | } | |
458 | ||
459 | return kvm_vm_ioctl(s, KVM_CREATE_VCPU, (void *)vcpu_id); | |
460 | } | |
461 | ||
d0a92b35 | 462 | int kvm_init_vcpu(CPUState *cpu, Error **errp) |
05330448 AL |
463 | { |
464 | KVMState *s = kvm_state; | |
465 | long mmap_size; | |
466 | int ret; | |
467 | ||
d0a92b35 | 468 | trace_kvm_init_vcpu(cpu->cpu_index, kvm_arch_vcpu_id(cpu)); |
05330448 | 469 | |
4c055ab5 | 470 | ret = kvm_get_vcpu(s, kvm_arch_vcpu_id(cpu)); |
05330448 | 471 | if (ret < 0) { |
d0a92b35 DDAG |
472 | error_setg_errno(errp, -ret, "kvm_init_vcpu: kvm_get_vcpu failed (%lu)", |
473 | kvm_arch_vcpu_id(cpu)); | |
05330448 AL |
474 | goto err; |
475 | } | |
476 | ||
8737c51c | 477 | cpu->kvm_fd = ret; |
a60f24b5 | 478 | cpu->kvm_state = s; |
99f31832 | 479 | cpu->vcpu_dirty = true; |
7786ae40 | 480 | cpu->dirty_pages = 0; |
baa60983 | 481 | cpu->throttle_us_per_full = 0; |
05330448 AL |
482 | |
483 | mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0); | |
484 | if (mmap_size < 0) { | |
748a680b | 485 | ret = mmap_size; |
d0a92b35 DDAG |
486 | error_setg_errno(errp, -mmap_size, |
487 | "kvm_init_vcpu: KVM_GET_VCPU_MMAP_SIZE failed"); | |
05330448 AL |
488 | goto err; |
489 | } | |
490 | ||
f7575c96 | 491 | cpu->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED, |
8737c51c | 492 | cpu->kvm_fd, 0); |
f7575c96 | 493 | if (cpu->kvm_run == MAP_FAILED) { |
05330448 | 494 | ret = -errno; |
d0a92b35 DDAG |
495 | error_setg_errno(errp, ret, |
496 | "kvm_init_vcpu: mmap'ing vcpu state failed (%lu)", | |
497 | kvm_arch_vcpu_id(cpu)); | |
05330448 AL |
498 | goto err; |
499 | } | |
500 | ||
a426e122 JK |
501 | if (s->coalesced_mmio && !s->coalesced_mmio_ring) { |
502 | s->coalesced_mmio_ring = | |
f7575c96 | 503 | (void *)cpu->kvm_run + s->coalesced_mmio * PAGE_SIZE; |
a426e122 | 504 | } |
62a2744c | 505 | |
b4420f19 PX |
506 | if (s->kvm_dirty_ring_size) { |
507 | /* Use MAP_SHARED to share pages with the kernel */ | |
dcafa248 | 508 | cpu->kvm_dirty_gfns = mmap(NULL, s->kvm_dirty_ring_bytes, |
b4420f19 PX |
509 | PROT_READ | PROT_WRITE, MAP_SHARED, |
510 | cpu->kvm_fd, | |
511 | PAGE_SIZE * KVM_DIRTY_LOG_PAGE_OFFSET); | |
512 | if (cpu->kvm_dirty_gfns == MAP_FAILED) { | |
513 | ret = -errno; | |
514 | DPRINTF("mmap'ing vcpu dirty gfns failed: %d\n", ret); | |
515 | goto err; | |
516 | } | |
517 | } | |
518 | ||
20d695a9 | 519 | ret = kvm_arch_init_vcpu(cpu); |
d0a92b35 DDAG |
520 | if (ret < 0) { |
521 | error_setg_errno(errp, -ret, | |
522 | "kvm_init_vcpu: kvm_arch_init_vcpu failed (%lu)", | |
523 | kvm_arch_vcpu_id(cpu)); | |
524 | } | |
05330448 AL |
525 | err: |
526 | return ret; | |
527 | } | |
528 | ||
5832d1f2 AL |
529 | /* |
530 | * dirty pages logging control | |
531 | */ | |
25254bbc | 532 | |
d6ff5cbc | 533 | static int kvm_mem_flags(MemoryRegion *mr) |
25254bbc | 534 | { |
d6ff5cbc | 535 | bool readonly = mr->readonly || memory_region_is_romd(mr); |
235e8982 | 536 | int flags = 0; |
d6ff5cbc AJ |
537 | |
538 | if (memory_region_get_dirty_log_mask(mr) != 0) { | |
539 | flags |= KVM_MEM_LOG_DIRTY_PAGES; | |
540 | } | |
235e8982 JJ |
541 | if (readonly && kvm_readonly_mem_allowed) { |
542 | flags |= KVM_MEM_READONLY; | |
543 | } | |
544 | return flags; | |
25254bbc MT |
545 | } |
546 | ||
36adac49 | 547 | /* Called with KVMMemoryListener.slots_lock held */ |
7bbda04c PB |
548 | static int kvm_slot_update_flags(KVMMemoryListener *kml, KVMSlot *mem, |
549 | MemoryRegion *mr) | |
5832d1f2 | 550 | { |
d6ff5cbc | 551 | mem->flags = kvm_mem_flags(mr); |
5832d1f2 | 552 | |
4495d6a7 | 553 | /* If nothing changed effectively, no need to issue ioctl */ |
6c090d4a | 554 | if (mem->flags == mem->old_flags) { |
25254bbc | 555 | return 0; |
4495d6a7 JK |
556 | } |
557 | ||
ea776d15 | 558 | kvm_slot_init_dirty_bitmap(mem); |
6c090d4a | 559 | return kvm_set_user_memory_region(kml, mem, false); |
5832d1f2 AL |
560 | } |
561 | ||
7bbda04c PB |
562 | static int kvm_section_update_flags(KVMMemoryListener *kml, |
563 | MemoryRegionSection *section) | |
25254bbc | 564 | { |
023ae9a8 | 565 | hwaddr start_addr, size, slot_size; |
343562e8 | 566 | KVMSlot *mem; |
36adac49 | 567 | int ret = 0; |
25254bbc | 568 | |
343562e8 DH |
569 | size = kvm_align_section(section, &start_addr); |
570 | if (!size) { | |
ea8cb1a8 | 571 | return 0; |
25254bbc | 572 | } |
343562e8 | 573 | |
a2f77862 | 574 | kvm_slots_lock(); |
36adac49 | 575 | |
023ae9a8 IM |
576 | while (size && !ret) { |
577 | slot_size = MIN(kvm_max_slot_size, size); | |
578 | mem = kvm_lookup_matching_slot(kml, start_addr, slot_size); | |
579 | if (!mem) { | |
580 | /* We don't have a slot if we want to trap every access. */ | |
581 | goto out; | |
582 | } | |
343562e8 | 583 | |
023ae9a8 IM |
584 | ret = kvm_slot_update_flags(kml, mem, section->mr); |
585 | start_addr += slot_size; | |
586 | size -= slot_size; | |
587 | } | |
36adac49 PX |
588 | |
589 | out: | |
a2f77862 | 590 | kvm_slots_unlock(); |
36adac49 | 591 | return ret; |
25254bbc MT |
592 | } |
593 | ||
a01672d3 | 594 | static void kvm_log_start(MemoryListener *listener, |
b2dfd71c PB |
595 | MemoryRegionSection *section, |
596 | int old, int new) | |
5832d1f2 | 597 | { |
7bbda04c | 598 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); |
a01672d3 AK |
599 | int r; |
600 | ||
b2dfd71c PB |
601 | if (old != 0) { |
602 | return; | |
603 | } | |
604 | ||
7bbda04c | 605 | r = kvm_section_update_flags(kml, section); |
a01672d3 AK |
606 | if (r < 0) { |
607 | abort(); | |
608 | } | |
5832d1f2 AL |
609 | } |
610 | ||
a01672d3 | 611 | static void kvm_log_stop(MemoryListener *listener, |
b2dfd71c PB |
612 | MemoryRegionSection *section, |
613 | int old, int new) | |
5832d1f2 | 614 | { |
7bbda04c | 615 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); |
a01672d3 AK |
616 | int r; |
617 | ||
b2dfd71c PB |
618 | if (new != 0) { |
619 | return; | |
620 | } | |
621 | ||
7bbda04c | 622 | r = kvm_section_update_flags(kml, section); |
a01672d3 AK |
623 | if (r < 0) { |
624 | abort(); | |
625 | } | |
5832d1f2 AL |
626 | } |
627 | ||
8369e01c | 628 | /* get kvm's dirty pages bitmap and update qemu's */ |
2c20b27e | 629 | static void kvm_slot_sync_dirty_pages(KVMSlot *slot) |
96c1606b | 630 | { |
2c20b27e | 631 | ram_addr_t start = slot->ram_start_offset; |
8e3b0cbb | 632 | ram_addr_t pages = slot->memory_size / qemu_real_host_page_size(); |
5ff7fb77 | 633 | |
2c20b27e | 634 | cpu_physical_memory_set_dirty_lebitmap(slot->dirty_bmap, start, pages); |
96c1606b AG |
635 | } |
636 | ||
b4420f19 PX |
637 | static void kvm_slot_reset_dirty_pages(KVMSlot *slot) |
638 | { | |
639 | memset(slot->dirty_bmap, 0, slot->dirty_bmap_size); | |
640 | } | |
641 | ||
8369e01c MT |
642 | #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1)) |
643 | ||
9b3a31c7 | 644 | /* Allocate the dirty bitmap for a slot */ |
ea776d15 | 645 | static void kvm_slot_init_dirty_bitmap(KVMSlot *mem) |
9b3a31c7 | 646 | { |
ea776d15 PX |
647 | if (!(mem->flags & KVM_MEM_LOG_DIRTY_PAGES) || mem->dirty_bmap) { |
648 | return; | |
649 | } | |
650 | ||
9b3a31c7 DDAG |
651 | /* |
652 | * XXX bad kernel interface alert | |
653 | * For dirty bitmap, kernel allocates array of size aligned to | |
654 | * bits-per-long. But for case when the kernel is 64bits and | |
655 | * the userspace is 32bits, userspace can't align to the same | |
656 | * bits-per-long, since sizeof(long) is different between kernel | |
657 | * and user space. This way, userspace will provide buffer which | |
658 | * may be 4 bytes less than the kernel will use, resulting in | |
659 | * userspace memory corruption (which is not detectable by valgrind | |
660 | * too, in most cases). | |
661 | * So for now, let's align to 64 instead of HOST_LONG_BITS here, in | |
662 | * a hope that sizeof(long) won't become >8 any time soon. | |
e0a8f993 KZ |
663 | * |
664 | * Note: the granule of kvm dirty log is qemu_real_host_page_size. | |
665 | * And mem->memory_size is aligned to it (otherwise this mem can't | |
666 | * be registered to KVM). | |
9b3a31c7 | 667 | */ |
8e3b0cbb | 668 | hwaddr bitmap_size = ALIGN(mem->memory_size / qemu_real_host_page_size(), |
9b3a31c7 DDAG |
669 | /*HOST_LONG_BITS*/ 64) / 8; |
670 | mem->dirty_bmap = g_malloc0(bitmap_size); | |
563d32ba | 671 | mem->dirty_bmap_size = bitmap_size; |
9b3a31c7 DDAG |
672 | } |
673 | ||
e65e5f50 PX |
674 | /* |
675 | * Sync dirty bitmap from kernel to KVMSlot.dirty_bmap, return true if | |
676 | * succeeded, false otherwise | |
677 | */ | |
678 | static bool kvm_slot_get_dirty_log(KVMState *s, KVMSlot *slot) | |
679 | { | |
680 | struct kvm_dirty_log d = {}; | |
681 | int ret; | |
682 | ||
683 | d.dirty_bitmap = slot->dirty_bmap; | |
684 | d.slot = slot->slot | (slot->as_id << 16); | |
685 | ret = kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d); | |
686 | ||
687 | if (ret == -ENOENT) { | |
688 | /* kernel does not have dirty bitmap in this slot */ | |
689 | ret = 0; | |
690 | } | |
691 | if (ret) { | |
692 | error_report_once("%s: KVM_GET_DIRTY_LOG failed with %d", | |
693 | __func__, ret); | |
694 | } | |
695 | return ret == 0; | |
696 | } | |
697 | ||
b4420f19 PX |
698 | /* Should be with all slots_lock held for the address spaces. */ |
699 | static void kvm_dirty_ring_mark_page(KVMState *s, uint32_t as_id, | |
700 | uint32_t slot_id, uint64_t offset) | |
701 | { | |
702 | KVMMemoryListener *kml; | |
703 | KVMSlot *mem; | |
704 | ||
705 | if (as_id >= s->nr_as) { | |
706 | return; | |
707 | } | |
708 | ||
709 | kml = s->as[as_id].ml; | |
710 | mem = &kml->slots[slot_id]; | |
711 | ||
712 | if (!mem->memory_size || offset >= | |
8e3b0cbb | 713 | (mem->memory_size / qemu_real_host_page_size())) { |
b4420f19 PX |
714 | return; |
715 | } | |
716 | ||
717 | set_bit(offset, mem->dirty_bmap); | |
718 | } | |
719 | ||
720 | static bool dirty_gfn_is_dirtied(struct kvm_dirty_gfn *gfn) | |
721 | { | |
722 | return gfn->flags == KVM_DIRTY_GFN_F_DIRTY; | |
723 | } | |
724 | ||
725 | static void dirty_gfn_set_collected(struct kvm_dirty_gfn *gfn) | |
726 | { | |
727 | gfn->flags = KVM_DIRTY_GFN_F_RESET; | |
728 | } | |
729 | ||
730 | /* | |
731 | * Should be with all slots_lock held for the address spaces. It returns the | |
732 | * dirty page we've collected on this dirty ring. | |
733 | */ | |
734 | static uint32_t kvm_dirty_ring_reap_one(KVMState *s, CPUState *cpu) | |
735 | { | |
736 | struct kvm_dirty_gfn *dirty_gfns = cpu->kvm_dirty_gfns, *cur; | |
737 | uint32_t ring_size = s->kvm_dirty_ring_size; | |
738 | uint32_t count = 0, fetch = cpu->kvm_fetch_index; | |
739 | ||
740 | assert(dirty_gfns && ring_size); | |
741 | trace_kvm_dirty_ring_reap_vcpu(cpu->cpu_index); | |
742 | ||
743 | while (true) { | |
744 | cur = &dirty_gfns[fetch % ring_size]; | |
745 | if (!dirty_gfn_is_dirtied(cur)) { | |
746 | break; | |
747 | } | |
748 | kvm_dirty_ring_mark_page(s, cur->slot >> 16, cur->slot & 0xffff, | |
749 | cur->offset); | |
750 | dirty_gfn_set_collected(cur); | |
751 | trace_kvm_dirty_ring_page(cpu->cpu_index, fetch, cur->offset); | |
752 | fetch++; | |
753 | count++; | |
754 | } | |
755 | cpu->kvm_fetch_index = fetch; | |
7786ae40 | 756 | cpu->dirty_pages += count; |
b4420f19 PX |
757 | |
758 | return count; | |
759 | } | |
760 | ||
761 | /* Must be with slots_lock held */ | |
1667e2b9 | 762 | static uint64_t kvm_dirty_ring_reap_locked(KVMState *s, CPUState* cpu) |
b4420f19 PX |
763 | { |
764 | int ret; | |
b4420f19 PX |
765 | uint64_t total = 0; |
766 | int64_t stamp; | |
767 | ||
768 | stamp = get_clock(); | |
769 | ||
1667e2b9 HH |
770 | if (cpu) { |
771 | total = kvm_dirty_ring_reap_one(s, cpu); | |
772 | } else { | |
773 | CPU_FOREACH(cpu) { | |
774 | total += kvm_dirty_ring_reap_one(s, cpu); | |
775 | } | |
b4420f19 PX |
776 | } |
777 | ||
778 | if (total) { | |
779 | ret = kvm_vm_ioctl(s, KVM_RESET_DIRTY_RINGS); | |
780 | assert(ret == total); | |
781 | } | |
782 | ||
783 | stamp = get_clock() - stamp; | |
784 | ||
785 | if (total) { | |
786 | trace_kvm_dirty_ring_reap(total, stamp / 1000); | |
787 | } | |
788 | ||
789 | return total; | |
790 | } | |
791 | ||
792 | /* | |
793 | * Currently for simplicity, we must hold BQL before calling this. We can | |
794 | * consider to drop the BQL if we're clear with all the race conditions. | |
795 | */ | |
1667e2b9 | 796 | static uint64_t kvm_dirty_ring_reap(KVMState *s, CPUState *cpu) |
b4420f19 PX |
797 | { |
798 | uint64_t total; | |
799 | ||
800 | /* | |
801 | * We need to lock all kvm slots for all address spaces here, | |
802 | * because: | |
803 | * | |
804 | * (1) We need to mark dirty for dirty bitmaps in multiple slots | |
805 | * and for tons of pages, so it's better to take the lock here | |
806 | * once rather than once per page. And more importantly, | |
807 | * | |
808 | * (2) We must _NOT_ publish dirty bits to the other threads | |
809 | * (e.g., the migration thread) via the kvm memory slot dirty | |
810 | * bitmaps before correctly re-protect those dirtied pages. | |
811 | * Otherwise we can have potential risk of data corruption if | |
812 | * the page data is read in the other thread before we do | |
813 | * reset below. | |
814 | */ | |
815 | kvm_slots_lock(); | |
1667e2b9 | 816 | total = kvm_dirty_ring_reap_locked(s, cpu); |
b4420f19 PX |
817 | kvm_slots_unlock(); |
818 | ||
819 | return total; | |
820 | } | |
821 | ||
822 | static void do_kvm_cpu_synchronize_kick(CPUState *cpu, run_on_cpu_data arg) | |
823 | { | |
824 | /* No need to do anything */ | |
825 | } | |
826 | ||
827 | /* | |
828 | * Kick all vcpus out in a synchronized way. When returned, we | |
829 | * guarantee that every vcpu has been kicked and at least returned to | |
830 | * userspace once. | |
831 | */ | |
832 | static void kvm_cpu_synchronize_kick_all(void) | |
833 | { | |
834 | CPUState *cpu; | |
835 | ||
836 | CPU_FOREACH(cpu) { | |
837 | run_on_cpu(cpu, do_kvm_cpu_synchronize_kick, RUN_ON_CPU_NULL); | |
838 | } | |
839 | } | |
840 | ||
841 | /* | |
842 | * Flush all the existing dirty pages to the KVM slot buffers. When | |
843 | * this call returns, we guarantee that all the touched dirty pages | |
844 | * before calling this function have been put into the per-kvmslot | |
845 | * dirty bitmap. | |
846 | * | |
847 | * This function must be called with BQL held. | |
848 | */ | |
849 | static void kvm_dirty_ring_flush(void) | |
850 | { | |
851 | trace_kvm_dirty_ring_flush(0); | |
852 | /* | |
853 | * The function needs to be serialized. Since this function | |
854 | * should always be with BQL held, serialization is guaranteed. | |
855 | * However, let's be sure of it. | |
856 | */ | |
857 | assert(qemu_mutex_iothread_locked()); | |
858 | /* | |
859 | * First make sure to flush the hardware buffers by kicking all | |
860 | * vcpus out in a synchronous way. | |
861 | */ | |
862 | kvm_cpu_synchronize_kick_all(); | |
1667e2b9 | 863 | kvm_dirty_ring_reap(kvm_state, NULL); |
b4420f19 PX |
864 | trace_kvm_dirty_ring_flush(1); |
865 | } | |
866 | ||
5832d1f2 | 867 | /** |
4a12a11a | 868 | * kvm_physical_sync_dirty_bitmap - Sync dirty bitmap from kernel space |
5832d1f2 | 869 | * |
4a12a11a PX |
870 | * This function will first try to fetch dirty bitmap from the kernel, |
871 | * and then updates qemu's dirty bitmap. | |
872 | * | |
36adac49 PX |
873 | * NOTE: caller must be with kml->slots_lock held. |
874 | * | |
4a12a11a PX |
875 | * @kml: the KVM memory listener object |
876 | * @section: the memory section to sync the dirty bitmap with | |
5832d1f2 | 877 | */ |
e65e5f50 PX |
878 | static void kvm_physical_sync_dirty_bitmap(KVMMemoryListener *kml, |
879 | MemoryRegionSection *section) | |
5832d1f2 AL |
880 | { |
881 | KVMState *s = kvm_state; | |
151f7749 | 882 | KVMSlot *mem; |
67548f09 | 883 | hwaddr start_addr, size; |
2c20b27e | 884 | hwaddr slot_size; |
67548f09 DH |
885 | |
886 | size = kvm_align_section(section, &start_addr); | |
023ae9a8 | 887 | while (size) { |
023ae9a8 IM |
888 | slot_size = MIN(kvm_max_slot_size, size); |
889 | mem = kvm_lookup_matching_slot(kml, start_addr, slot_size); | |
67548f09 | 890 | if (!mem) { |
e377e87c | 891 | /* We don't have a slot if we want to trap every access. */ |
e65e5f50 | 892 | return; |
151f7749 | 893 | } |
e65e5f50 | 894 | if (kvm_slot_get_dirty_log(s, mem)) { |
2c20b27e | 895 | kvm_slot_sync_dirty_pages(mem); |
151f7749 | 896 | } |
023ae9a8 IM |
897 | start_addr += slot_size; |
898 | size -= slot_size; | |
5832d1f2 | 899 | } |
5832d1f2 AL |
900 | } |
901 | ||
ff4aa114 PX |
902 | /* Alignment requirement for KVM_CLEAR_DIRTY_LOG - 64 pages */ |
903 | #define KVM_CLEAR_LOG_SHIFT 6 | |
8e3b0cbb | 904 | #define KVM_CLEAR_LOG_ALIGN (qemu_real_host_page_size() << KVM_CLEAR_LOG_SHIFT) |
ff4aa114 PX |
905 | #define KVM_CLEAR_LOG_MASK (-KVM_CLEAR_LOG_ALIGN) |
906 | ||
4222147d PB |
907 | static int kvm_log_clear_one_slot(KVMSlot *mem, int as_id, uint64_t start, |
908 | uint64_t size) | |
ff4aa114 PX |
909 | { |
910 | KVMState *s = kvm_state; | |
4222147d | 911 | uint64_t end, bmap_start, start_delta, bmap_npages; |
ff4aa114 | 912 | struct kvm_clear_dirty_log d; |
8e3b0cbb | 913 | unsigned long *bmap_clear = NULL, psize = qemu_real_host_page_size(); |
4222147d | 914 | int ret; |
ff4aa114 PX |
915 | |
916 | /* | |
917 | * We need to extend either the start or the size or both to | |
918 | * satisfy the KVM interface requirement. Firstly, do the start | |
919 | * page alignment on 64 host pages | |
920 | */ | |
84516e5b PB |
921 | bmap_start = start & KVM_CLEAR_LOG_MASK; |
922 | start_delta = start - bmap_start; | |
ff4aa114 PX |
923 | bmap_start /= psize; |
924 | ||
925 | /* | |
926 | * The kernel interface has restriction on the size too, that either: | |
927 | * | |
928 | * (1) the size is 64 host pages aligned (just like the start), or | |
929 | * (2) the size fills up until the end of the KVM memslot. | |
930 | */ | |
931 | bmap_npages = DIV_ROUND_UP(size + start_delta, KVM_CLEAR_LOG_ALIGN) | |
932 | << KVM_CLEAR_LOG_SHIFT; | |
933 | end = mem->memory_size / psize; | |
934 | if (bmap_npages > end - bmap_start) { | |
935 | bmap_npages = end - bmap_start; | |
936 | } | |
937 | start_delta /= psize; | |
938 | ||
939 | /* | |
940 | * Prepare the bitmap to clear dirty bits. Here we must guarantee | |
941 | * that we won't clear any unknown dirty bits otherwise we might | |
942 | * accidentally clear some set bits which are not yet synced from | |
943 | * the kernel into QEMU's bitmap, then we'll lose track of the | |
944 | * guest modifications upon those pages (which can directly lead | |
945 | * to guest data loss or panic after migration). | |
946 | * | |
947 | * Layout of the KVMSlot.dirty_bmap: | |
948 | * | |
949 | * |<-------- bmap_npages -----------..>| | |
950 | * [1] | |
951 | * start_delta size | |
952 | * |----------------|-------------|------------------|------------| | |
953 | * ^ ^ ^ ^ | |
954 | * | | | | | |
955 | * start bmap_start (start) end | |
956 | * of memslot of memslot | |
957 | * | |
958 | * [1] bmap_npages can be aligned to either 64 pages or the end of slot | |
959 | */ | |
960 | ||
961 | assert(bmap_start % BITS_PER_LONG == 0); | |
962 | /* We should never do log_clear before log_sync */ | |
963 | assert(mem->dirty_bmap); | |
4054adbd | 964 | if (start_delta || bmap_npages - size / psize) { |
ff4aa114 PX |
965 | /* Slow path - we need to manipulate a temp bitmap */ |
966 | bmap_clear = bitmap_new(bmap_npages); | |
967 | bitmap_copy_with_src_offset(bmap_clear, mem->dirty_bmap, | |
968 | bmap_start, start_delta + size / psize); | |
969 | /* | |
970 | * We need to fill the holes at start because that was not | |
971 | * specified by the caller and we extended the bitmap only for | |
972 | * 64 pages alignment | |
973 | */ | |
974 | bitmap_clear(bmap_clear, 0, start_delta); | |
975 | d.dirty_bitmap = bmap_clear; | |
976 | } else { | |
4054adbd ZY |
977 | /* |
978 | * Fast path - both start and size align well with BITS_PER_LONG | |
979 | * (or the end of memory slot) | |
980 | */ | |
ff4aa114 PX |
981 | d.dirty_bitmap = mem->dirty_bmap + BIT_WORD(bmap_start); |
982 | } | |
983 | ||
984 | d.first_page = bmap_start; | |
985 | /* It should never overflow. If it happens, say something */ | |
986 | assert(bmap_npages <= UINT32_MAX); | |
987 | d.num_pages = bmap_npages; | |
4222147d | 988 | d.slot = mem->slot | (as_id << 16); |
ff4aa114 | 989 | |
38e0b790 TH |
990 | ret = kvm_vm_ioctl(s, KVM_CLEAR_DIRTY_LOG, &d); |
991 | if (ret < 0 && ret != -ENOENT) { | |
ff4aa114 PX |
992 | error_report("%s: KVM_CLEAR_DIRTY_LOG failed, slot=%d, " |
993 | "start=0x%"PRIx64", size=0x%"PRIx32", errno=%d", | |
994 | __func__, d.slot, (uint64_t)d.first_page, | |
995 | (uint32_t)d.num_pages, ret); | |
996 | } else { | |
997 | ret = 0; | |
998 | trace_kvm_clear_dirty_log(d.slot, d.first_page, d.num_pages); | |
999 | } | |
1000 | ||
1001 | /* | |
1002 | * After we have updated the remote dirty bitmap, we update the | |
1003 | * cached bitmap as well for the memslot, then if another user | |
1004 | * clears the same region we know we shouldn't clear it again on | |
1005 | * the remote otherwise it's data loss as well. | |
1006 | */ | |
1007 | bitmap_clear(mem->dirty_bmap, bmap_start + start_delta, | |
1008 | size / psize); | |
1009 | /* This handles the NULL case well */ | |
1010 | g_free(bmap_clear); | |
4222147d PB |
1011 | return ret; |
1012 | } | |
1013 | ||
1014 | ||
1015 | /** | |
1016 | * kvm_physical_log_clear - Clear the kernel's dirty bitmap for range | |
1017 | * | |
1018 | * NOTE: this will be a no-op if we haven't enabled manual dirty log | |
1019 | * protection in the host kernel because in that case this operation | |
1020 | * will be done within log_sync(). | |
1021 | * | |
1022 | * @kml: the kvm memory listener | |
1023 | * @section: the memory range to clear dirty bitmap | |
1024 | */ | |
1025 | static int kvm_physical_log_clear(KVMMemoryListener *kml, | |
1026 | MemoryRegionSection *section) | |
1027 | { | |
1028 | KVMState *s = kvm_state; | |
84516e5b PB |
1029 | uint64_t start, size, offset, count; |
1030 | KVMSlot *mem; | |
87287ac0 | 1031 | int ret = 0, i; |
4222147d PB |
1032 | |
1033 | if (!s->manual_dirty_log_protect) { | |
1034 | /* No need to do explicit clear */ | |
87287ac0 | 1035 | return ret; |
4222147d PB |
1036 | } |
1037 | ||
1038 | start = section->offset_within_address_space; | |
1039 | size = int128_get64(section->size); | |
1040 | ||
1041 | if (!size) { | |
1042 | /* Nothing more we can do... */ | |
87287ac0 | 1043 | return ret; |
4222147d PB |
1044 | } |
1045 | ||
a2f77862 | 1046 | kvm_slots_lock(); |
4222147d | 1047 | |
4222147d PB |
1048 | for (i = 0; i < s->nr_slots; i++) { |
1049 | mem = &kml->slots[i]; | |
84516e5b PB |
1050 | /* Discard slots that are empty or do not overlap the section */ |
1051 | if (!mem->memory_size || | |
1052 | mem->start_addr > start + size - 1 || | |
1053 | start > mem->start_addr + mem->memory_size - 1) { | |
1054 | continue; | |
1055 | } | |
1056 | ||
1057 | if (start >= mem->start_addr) { | |
1058 | /* The slot starts before section or is aligned to it. */ | |
1059 | offset = start - mem->start_addr; | |
1060 | count = MIN(mem->memory_size - offset, size); | |
1061 | } else { | |
1062 | /* The slot starts after section. */ | |
1063 | offset = 0; | |
1064 | count = MIN(mem->memory_size, size - (mem->start_addr - start)); | |
1065 | } | |
1066 | ret = kvm_log_clear_one_slot(mem, kml->as_id, offset, count); | |
1067 | if (ret < 0) { | |
4222147d PB |
1068 | break; |
1069 | } | |
1070 | } | |
1071 | ||
a2f77862 | 1072 | kvm_slots_unlock(); |
ff4aa114 PX |
1073 | |
1074 | return ret; | |
1075 | } | |
1076 | ||
95d2994a AK |
1077 | static void kvm_coalesce_mmio_region(MemoryListener *listener, |
1078 | MemoryRegionSection *secion, | |
a8170e5e | 1079 | hwaddr start, hwaddr size) |
f65ed4c1 | 1080 | { |
f65ed4c1 AL |
1081 | KVMState *s = kvm_state; |
1082 | ||
1083 | if (s->coalesced_mmio) { | |
1084 | struct kvm_coalesced_mmio_zone zone; | |
1085 | ||
1086 | zone.addr = start; | |
1087 | zone.size = size; | |
7e680753 | 1088 | zone.pad = 0; |
f65ed4c1 | 1089 | |
95d2994a | 1090 | (void)kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone); |
f65ed4c1 | 1091 | } |
f65ed4c1 AL |
1092 | } |
1093 | ||
95d2994a AK |
1094 | static void kvm_uncoalesce_mmio_region(MemoryListener *listener, |
1095 | MemoryRegionSection *secion, | |
a8170e5e | 1096 | hwaddr start, hwaddr size) |
f65ed4c1 | 1097 | { |
f65ed4c1 AL |
1098 | KVMState *s = kvm_state; |
1099 | ||
1100 | if (s->coalesced_mmio) { | |
1101 | struct kvm_coalesced_mmio_zone zone; | |
1102 | ||
1103 | zone.addr = start; | |
1104 | zone.size = size; | |
7e680753 | 1105 | zone.pad = 0; |
f65ed4c1 | 1106 | |
95d2994a | 1107 | (void)kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone); |
f65ed4c1 | 1108 | } |
f65ed4c1 AL |
1109 | } |
1110 | ||
e6d34aee PH |
1111 | static void kvm_coalesce_pio_add(MemoryListener *listener, |
1112 | MemoryRegionSection *section, | |
1113 | hwaddr start, hwaddr size) | |
1114 | { | |
1115 | KVMState *s = kvm_state; | |
1116 | ||
1117 | if (s->coalesced_pio) { | |
1118 | struct kvm_coalesced_mmio_zone zone; | |
1119 | ||
1120 | zone.addr = start; | |
1121 | zone.size = size; | |
1122 | zone.pio = 1; | |
1123 | ||
1124 | (void)kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone); | |
1125 | } | |
1126 | } | |
1127 | ||
1128 | static void kvm_coalesce_pio_del(MemoryListener *listener, | |
1129 | MemoryRegionSection *section, | |
1130 | hwaddr start, hwaddr size) | |
1131 | { | |
1132 | KVMState *s = kvm_state; | |
1133 | ||
1134 | if (s->coalesced_pio) { | |
1135 | struct kvm_coalesced_mmio_zone zone; | |
1136 | ||
1137 | zone.addr = start; | |
1138 | zone.size = size; | |
1139 | zone.pio = 1; | |
1140 | ||
1141 | (void)kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone); | |
1142 | } | |
1143 | } | |
1144 | ||
1145 | static MemoryListener kvm_coalesced_pio_listener = { | |
142518bd | 1146 | .name = "kvm-coalesced-pio", |
e6d34aee PH |
1147 | .coalesced_io_add = kvm_coalesce_pio_add, |
1148 | .coalesced_io_del = kvm_coalesce_pio_del, | |
1149 | }; | |
1150 | ||
ad7b8b33 AL |
1151 | int kvm_check_extension(KVMState *s, unsigned int extension) |
1152 | { | |
1153 | int ret; | |
1154 | ||
1155 | ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension); | |
1156 | if (ret < 0) { | |
1157 | ret = 0; | |
1158 | } | |
1159 | ||
1160 | return ret; | |
1161 | } | |
1162 | ||
7d0a07fa AG |
1163 | int kvm_vm_check_extension(KVMState *s, unsigned int extension) |
1164 | { | |
1165 | int ret; | |
1166 | ||
1167 | ret = kvm_vm_ioctl(s, KVM_CHECK_EXTENSION, extension); | |
1168 | if (ret < 0) { | |
1169 | /* VM wide version not implemented, use global one instead */ | |
1170 | ret = kvm_check_extension(s, extension); | |
1171 | } | |
1172 | ||
1173 | return ret; | |
1174 | } | |
1175 | ||
6b552b9b DG |
1176 | typedef struct HWPoisonPage { |
1177 | ram_addr_t ram_addr; | |
1178 | QLIST_ENTRY(HWPoisonPage) list; | |
1179 | } HWPoisonPage; | |
1180 | ||
1181 | static QLIST_HEAD(, HWPoisonPage) hwpoison_page_list = | |
1182 | QLIST_HEAD_INITIALIZER(hwpoison_page_list); | |
1183 | ||
1184 | static void kvm_unpoison_all(void *param) | |
1185 | { | |
1186 | HWPoisonPage *page, *next_page; | |
1187 | ||
1188 | QLIST_FOREACH_SAFE(page, &hwpoison_page_list, list, next_page) { | |
1189 | QLIST_REMOVE(page, list); | |
1190 | qemu_ram_remap(page->ram_addr, TARGET_PAGE_SIZE); | |
1191 | g_free(page); | |
1192 | } | |
1193 | } | |
1194 | ||
1195 | void kvm_hwpoison_page_add(ram_addr_t ram_addr) | |
1196 | { | |
1197 | HWPoisonPage *page; | |
1198 | ||
1199 | QLIST_FOREACH(page, &hwpoison_page_list, list) { | |
1200 | if (page->ram_addr == ram_addr) { | |
1201 | return; | |
1202 | } | |
1203 | } | |
1204 | page = g_new(HWPoisonPage, 1); | |
1205 | page->ram_addr = ram_addr; | |
1206 | QLIST_INSERT_HEAD(&hwpoison_page_list, page, list); | |
1207 | } | |
1208 | ||
b680c5ba GK |
1209 | static uint32_t adjust_ioeventfd_endianness(uint32_t val, uint32_t size) |
1210 | { | |
ee3eb3a7 | 1211 | #if HOST_BIG_ENDIAN != TARGET_BIG_ENDIAN |
e03b5686 | 1212 | /* The kernel expects ioeventfd values in HOST_BIG_ENDIAN |
b680c5ba GK |
1213 | * endianness, but the memory core hands them in target endianness. |
1214 | * For example, PPC is always treated as big-endian even if running | |
1215 | * on KVM and on PPC64LE. Correct here. | |
1216 | */ | |
1217 | switch (size) { | |
1218 | case 2: | |
1219 | val = bswap16(val); | |
1220 | break; | |
1221 | case 4: | |
1222 | val = bswap32(val); | |
1223 | break; | |
1224 | } | |
1225 | #endif | |
1226 | return val; | |
1227 | } | |
1228 | ||
584f2be7 | 1229 | static int kvm_set_ioeventfd_mmio(int fd, hwaddr addr, uint32_t val, |
41cb62c2 | 1230 | bool assign, uint32_t size, bool datamatch) |
500ffd4a MT |
1231 | { |
1232 | int ret; | |
03a96b83 TH |
1233 | struct kvm_ioeventfd iofd = { |
1234 | .datamatch = datamatch ? adjust_ioeventfd_endianness(val, size) : 0, | |
1235 | .addr = addr, | |
1236 | .len = size, | |
1237 | .flags = 0, | |
1238 | .fd = fd, | |
1239 | }; | |
500ffd4a | 1240 | |
876d16cd DDAG |
1241 | trace_kvm_set_ioeventfd_mmio(fd, (uint64_t)addr, val, assign, size, |
1242 | datamatch); | |
500ffd4a MT |
1243 | if (!kvm_enabled()) { |
1244 | return -ENOSYS; | |
1245 | } | |
1246 | ||
41cb62c2 MT |
1247 | if (datamatch) { |
1248 | iofd.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH; | |
1249 | } | |
500ffd4a MT |
1250 | if (!assign) { |
1251 | iofd.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; | |
1252 | } | |
1253 | ||
1254 | ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd); | |
1255 | ||
1256 | if (ret < 0) { | |
1257 | return -errno; | |
1258 | } | |
1259 | ||
1260 | return 0; | |
1261 | } | |
1262 | ||
44c3f8f7 | 1263 | static int kvm_set_ioeventfd_pio(int fd, uint16_t addr, uint16_t val, |
41cb62c2 | 1264 | bool assign, uint32_t size, bool datamatch) |
500ffd4a MT |
1265 | { |
1266 | struct kvm_ioeventfd kick = { | |
b680c5ba | 1267 | .datamatch = datamatch ? adjust_ioeventfd_endianness(val, size) : 0, |
500ffd4a | 1268 | .addr = addr, |
41cb62c2 | 1269 | .flags = KVM_IOEVENTFD_FLAG_PIO, |
44c3f8f7 | 1270 | .len = size, |
500ffd4a MT |
1271 | .fd = fd, |
1272 | }; | |
1273 | int r; | |
876d16cd | 1274 | trace_kvm_set_ioeventfd_pio(fd, addr, val, assign, size, datamatch); |
500ffd4a MT |
1275 | if (!kvm_enabled()) { |
1276 | return -ENOSYS; | |
1277 | } | |
41cb62c2 MT |
1278 | if (datamatch) { |
1279 | kick.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH; | |
1280 | } | |
500ffd4a MT |
1281 | if (!assign) { |
1282 | kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; | |
1283 | } | |
1284 | r = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick); | |
1285 | if (r < 0) { | |
1286 | return r; | |
1287 | } | |
1288 | return 0; | |
1289 | } | |
1290 | ||
1291 | ||
d2f2b8a7 SH |
1292 | static int kvm_check_many_ioeventfds(void) |
1293 | { | |
d0dcac83 SH |
1294 | /* Userspace can use ioeventfd for io notification. This requires a host |
1295 | * that supports eventfd(2) and an I/O thread; since eventfd does not | |
1296 | * support SIGIO it cannot interrupt the vcpu. | |
1297 | * | |
1298 | * Older kernels have a 6 device limit on the KVM io bus. Find out so we | |
d2f2b8a7 SH |
1299 | * can avoid creating too many ioeventfds. |
1300 | */ | |
12d4536f | 1301 | #if defined(CONFIG_EVENTFD) |
d2f2b8a7 SH |
1302 | int ioeventfds[7]; |
1303 | int i, ret = 0; | |
1304 | for (i = 0; i < ARRAY_SIZE(ioeventfds); i++) { | |
1305 | ioeventfds[i] = eventfd(0, EFD_CLOEXEC); | |
1306 | if (ioeventfds[i] < 0) { | |
1307 | break; | |
1308 | } | |
41cb62c2 | 1309 | ret = kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, true, 2, true); |
d2f2b8a7 SH |
1310 | if (ret < 0) { |
1311 | close(ioeventfds[i]); | |
1312 | break; | |
1313 | } | |
1314 | } | |
1315 | ||
1316 | /* Decide whether many devices are supported or not */ | |
1317 | ret = i == ARRAY_SIZE(ioeventfds); | |
1318 | ||
1319 | while (i-- > 0) { | |
41cb62c2 | 1320 | kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, false, 2, true); |
d2f2b8a7 SH |
1321 | close(ioeventfds[i]); |
1322 | } | |
1323 | return ret; | |
1324 | #else | |
1325 | return 0; | |
1326 | #endif | |
1327 | } | |
1328 | ||
94a8d39a JK |
1329 | static const KVMCapabilityInfo * |
1330 | kvm_check_extension_list(KVMState *s, const KVMCapabilityInfo *list) | |
1331 | { | |
1332 | while (list->name) { | |
1333 | if (!kvm_check_extension(s, list->value)) { | |
1334 | return list; | |
1335 | } | |
1336 | list++; | |
1337 | } | |
1338 | return NULL; | |
1339 | } | |
1340 | ||
023ae9a8 IM |
1341 | void kvm_set_max_memslot_size(hwaddr max_slot_size) |
1342 | { | |
1343 | g_assert( | |
8e3b0cbb | 1344 | ROUND_UP(max_slot_size, qemu_real_host_page_size()) == max_slot_size |
023ae9a8 IM |
1345 | ); |
1346 | kvm_max_slot_size = max_slot_size; | |
1347 | } | |
1348 | ||
7bbda04c PB |
1349 | static void kvm_set_phys_mem(KVMMemoryListener *kml, |
1350 | MemoryRegionSection *section, bool add) | |
46dbef6a | 1351 | { |
f357f564 | 1352 | KVMSlot *mem; |
46dbef6a | 1353 | int err; |
a01672d3 | 1354 | MemoryRegion *mr = section->mr; |
9323e79f | 1355 | bool writable = !mr->readonly && !mr->rom_device; |
2c20b27e PX |
1356 | hwaddr start_addr, size, slot_size, mr_offset; |
1357 | ram_addr_t ram_start_offset; | |
5ea69c2e | 1358 | void *ram; |
46dbef6a | 1359 | |
a01672d3 | 1360 | if (!memory_region_is_ram(mr)) { |
9323e79f | 1361 | if (writable || !kvm_readonly_mem_allowed) { |
235e8982 JJ |
1362 | return; |
1363 | } else if (!mr->romd_mode) { | |
1364 | /* If the memory device is not in romd_mode, then we actually want | |
1365 | * to remove the kvm memory slot so all accesses will trap. */ | |
1366 | add = false; | |
1367 | } | |
9f213ed9 AK |
1368 | } |
1369 | ||
5ea69c2e DH |
1370 | size = kvm_align_section(section, &start_addr); |
1371 | if (!size) { | |
1372 | return; | |
1373 | } | |
1374 | ||
2c20b27e PX |
1375 | /* The offset of the kvmslot within the memory region */ |
1376 | mr_offset = section->offset_within_region + start_addr - | |
1377 | section->offset_within_address_space; | |
1378 | ||
1379 | /* use aligned delta to align the ram address and offset */ | |
1380 | ram = memory_region_get_ram_ptr(mr) + mr_offset; | |
1381 | ram_start_offset = memory_region_get_ram_addr(mr) + mr_offset; | |
a01672d3 | 1382 | |
a2f77862 | 1383 | kvm_slots_lock(); |
36adac49 | 1384 | |
f357f564 | 1385 | if (!add) { |
023ae9a8 IM |
1386 | do { |
1387 | slot_size = MIN(kvm_max_slot_size, size); | |
1388 | mem = kvm_lookup_matching_slot(kml, start_addr, slot_size); | |
1389 | if (!mem) { | |
1390 | goto out; | |
1391 | } | |
1392 | if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) { | |
b4420f19 PX |
1393 | /* |
1394 | * NOTE: We should be aware of the fact that here we're only | |
1395 | * doing a best effort to sync dirty bits. No matter whether | |
1396 | * we're using dirty log or dirty ring, we ignored two facts: | |
1397 | * | |
1398 | * (1) dirty bits can reside in hardware buffers (PML) | |
1399 | * | |
1400 | * (2) after we collected dirty bits here, pages can be dirtied | |
1401 | * again before we do the final KVM_SET_USER_MEMORY_REGION to | |
1402 | * remove the slot. | |
1403 | * | |
1404 | * Not easy. Let's cross the fingers until it's fixed. | |
1405 | */ | |
1406 | if (kvm_state->kvm_dirty_ring_size) { | |
1667e2b9 | 1407 | kvm_dirty_ring_reap_locked(kvm_state, NULL); |
b4420f19 PX |
1408 | } else { |
1409 | kvm_slot_get_dirty_log(kvm_state, mem); | |
1410 | } | |
29b7e8be | 1411 | kvm_slot_sync_dirty_pages(mem); |
023ae9a8 | 1412 | } |
3fbffb62 | 1413 | |
023ae9a8 IM |
1414 | /* unregister the slot */ |
1415 | g_free(mem->dirty_bmap); | |
1416 | mem->dirty_bmap = NULL; | |
1417 | mem->memory_size = 0; | |
1418 | mem->flags = 0; | |
1419 | err = kvm_set_user_memory_region(kml, mem, false); | |
1420 | if (err) { | |
1421 | fprintf(stderr, "%s: error unregistering slot: %s\n", | |
1422 | __func__, strerror(-err)); | |
1423 | abort(); | |
1424 | } | |
1425 | start_addr += slot_size; | |
1426 | size -= slot_size; | |
1427 | } while (size); | |
36adac49 | 1428 | goto out; |
46dbef6a MT |
1429 | } |
1430 | ||
f357f564 | 1431 | /* register the new slot */ |
023ae9a8 IM |
1432 | do { |
1433 | slot_size = MIN(kvm_max_slot_size, size); | |
1434 | mem = kvm_alloc_slot(kml); | |
e65e5f50 | 1435 | mem->as_id = kml->as_id; |
023ae9a8 IM |
1436 | mem->memory_size = slot_size; |
1437 | mem->start_addr = start_addr; | |
2c20b27e | 1438 | mem->ram_start_offset = ram_start_offset; |
023ae9a8 IM |
1439 | mem->ram = ram; |
1440 | mem->flags = kvm_mem_flags(mr); | |
ea776d15 | 1441 | kvm_slot_init_dirty_bitmap(mem); |
023ae9a8 IM |
1442 | err = kvm_set_user_memory_region(kml, mem, true); |
1443 | if (err) { | |
1444 | fprintf(stderr, "%s: error registering slot: %s\n", __func__, | |
1445 | strerror(-err)); | |
1446 | abort(); | |
1447 | } | |
1448 | start_addr += slot_size; | |
2c20b27e | 1449 | ram_start_offset += slot_size; |
023ae9a8 IM |
1450 | ram += slot_size; |
1451 | size -= slot_size; | |
1452 | } while (size); | |
36adac49 PX |
1453 | |
1454 | out: | |
a2f77862 | 1455 | kvm_slots_unlock(); |
46dbef6a MT |
1456 | } |
1457 | ||
b4420f19 PX |
1458 | static void *kvm_dirty_ring_reaper_thread(void *data) |
1459 | { | |
1460 | KVMState *s = data; | |
1461 | struct KVMDirtyRingReaper *r = &s->reaper; | |
1462 | ||
1463 | rcu_register_thread(); | |
1464 | ||
1465 | trace_kvm_dirty_ring_reaper("init"); | |
1466 | ||
1467 | while (true) { | |
1468 | r->reaper_state = KVM_DIRTY_RING_REAPER_WAIT; | |
1469 | trace_kvm_dirty_ring_reaper("wait"); | |
1470 | /* | |
1471 | * TODO: provide a smarter timeout rather than a constant? | |
1472 | */ | |
1473 | sleep(1); | |
1474 | ||
baa60983 HH |
1475 | /* keep sleeping so that dirtylimit not be interfered by reaper */ |
1476 | if (dirtylimit_in_service()) { | |
1477 | continue; | |
1478 | } | |
1479 | ||
b4420f19 PX |
1480 | trace_kvm_dirty_ring_reaper("wakeup"); |
1481 | r->reaper_state = KVM_DIRTY_RING_REAPER_REAPING; | |
1482 | ||
1483 | qemu_mutex_lock_iothread(); | |
1667e2b9 | 1484 | kvm_dirty_ring_reap(s, NULL); |
b4420f19 PX |
1485 | qemu_mutex_unlock_iothread(); |
1486 | ||
1487 | r->reaper_iteration++; | |
1488 | } | |
1489 | ||
1490 | trace_kvm_dirty_ring_reaper("exit"); | |
1491 | ||
1492 | rcu_unregister_thread(); | |
1493 | ||
1494 | return NULL; | |
1495 | } | |
1496 | ||
1497 | static int kvm_dirty_ring_reaper_init(KVMState *s) | |
1498 | { | |
1499 | struct KVMDirtyRingReaper *r = &s->reaper; | |
1500 | ||
1501 | qemu_thread_create(&r->reaper_thr, "kvm-reaper", | |
1502 | kvm_dirty_ring_reaper_thread, | |
1503 | s, QEMU_THREAD_JOINABLE); | |
1504 | ||
1505 | return 0; | |
1506 | } | |
1507 | ||
a01672d3 AK |
1508 | static void kvm_region_add(MemoryListener *listener, |
1509 | MemoryRegionSection *section) | |
1510 | { | |
7bbda04c PB |
1511 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); |
1512 | ||
dfde4e6e | 1513 | memory_region_ref(section->mr); |
7bbda04c | 1514 | kvm_set_phys_mem(kml, section, true); |
a01672d3 AK |
1515 | } |
1516 | ||
1517 | static void kvm_region_del(MemoryListener *listener, | |
1518 | MemoryRegionSection *section) | |
1519 | { | |
7bbda04c PB |
1520 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); |
1521 | ||
1522 | kvm_set_phys_mem(kml, section, false); | |
dfde4e6e | 1523 | memory_region_unref(section->mr); |
a01672d3 AK |
1524 | } |
1525 | ||
1526 | static void kvm_log_sync(MemoryListener *listener, | |
1527 | MemoryRegionSection *section) | |
7b8f3b78 | 1528 | { |
7bbda04c | 1529 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); |
a01672d3 | 1530 | |
a2f77862 | 1531 | kvm_slots_lock(); |
e65e5f50 | 1532 | kvm_physical_sync_dirty_bitmap(kml, section); |
a2f77862 | 1533 | kvm_slots_unlock(); |
7b8f3b78 MT |
1534 | } |
1535 | ||
b4420f19 PX |
1536 | static void kvm_log_sync_global(MemoryListener *l) |
1537 | { | |
1538 | KVMMemoryListener *kml = container_of(l, KVMMemoryListener, listener); | |
1539 | KVMState *s = kvm_state; | |
1540 | KVMSlot *mem; | |
1541 | int i; | |
1542 | ||
1543 | /* Flush all kernel dirty addresses into KVMSlot dirty bitmap */ | |
1544 | kvm_dirty_ring_flush(); | |
1545 | ||
1546 | /* | |
1547 | * TODO: make this faster when nr_slots is big while there are | |
1548 | * only a few used slots (small VMs). | |
1549 | */ | |
1550 | kvm_slots_lock(); | |
1551 | for (i = 0; i < s->nr_slots; i++) { | |
1552 | mem = &kml->slots[i]; | |
1553 | if (mem->memory_size && mem->flags & KVM_MEM_LOG_DIRTY_PAGES) { | |
1554 | kvm_slot_sync_dirty_pages(mem); | |
1555 | /* | |
1556 | * This is not needed by KVM_GET_DIRTY_LOG because the | |
1557 | * ioctl will unconditionally overwrite the whole region. | |
1558 | * However kvm dirty ring has no such side effect. | |
1559 | */ | |
1560 | kvm_slot_reset_dirty_pages(mem); | |
1561 | } | |
1562 | } | |
1563 | kvm_slots_unlock(); | |
1564 | } | |
1565 | ||
ff4aa114 PX |
1566 | static void kvm_log_clear(MemoryListener *listener, |
1567 | MemoryRegionSection *section) | |
1568 | { | |
1569 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); | |
1570 | int r; | |
1571 | ||
1572 | r = kvm_physical_log_clear(kml, section); | |
1573 | if (r < 0) { | |
1574 | error_report_once("%s: kvm log clear failed: mr=%s " | |
1575 | "offset=%"HWADDR_PRIx" size=%"PRIx64, __func__, | |
1576 | section->mr->name, section->offset_within_region, | |
1577 | int128_get64(section->size)); | |
1578 | abort(); | |
1579 | } | |
1580 | } | |
1581 | ||
d22b096e AK |
1582 | static void kvm_mem_ioeventfd_add(MemoryListener *listener, |
1583 | MemoryRegionSection *section, | |
1584 | bool match_data, uint64_t data, | |
1585 | EventNotifier *e) | |
1586 | { | |
1587 | int fd = event_notifier_get_fd(e); | |
80a1ea37 AK |
1588 | int r; |
1589 | ||
4b8f1c88 | 1590 | r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space, |
052e87b0 PB |
1591 | data, true, int128_get64(section->size), |
1592 | match_data); | |
80a1ea37 | 1593 | if (r < 0) { |
e346bcbf YK |
1594 | fprintf(stderr, "%s: error adding ioeventfd: %s (%d)\n", |
1595 | __func__, strerror(-r), -r); | |
80a1ea37 AK |
1596 | abort(); |
1597 | } | |
1598 | } | |
1599 | ||
d22b096e AK |
1600 | static void kvm_mem_ioeventfd_del(MemoryListener *listener, |
1601 | MemoryRegionSection *section, | |
1602 | bool match_data, uint64_t data, | |
1603 | EventNotifier *e) | |
80a1ea37 | 1604 | { |
d22b096e | 1605 | int fd = event_notifier_get_fd(e); |
80a1ea37 AK |
1606 | int r; |
1607 | ||
4b8f1c88 | 1608 | r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space, |
052e87b0 PB |
1609 | data, false, int128_get64(section->size), |
1610 | match_data); | |
80a1ea37 | 1611 | if (r < 0) { |
e346bcbf YK |
1612 | fprintf(stderr, "%s: error deleting ioeventfd: %s (%d)\n", |
1613 | __func__, strerror(-r), -r); | |
80a1ea37 AK |
1614 | abort(); |
1615 | } | |
1616 | } | |
1617 | ||
d22b096e AK |
1618 | static void kvm_io_ioeventfd_add(MemoryListener *listener, |
1619 | MemoryRegionSection *section, | |
1620 | bool match_data, uint64_t data, | |
1621 | EventNotifier *e) | |
80a1ea37 | 1622 | { |
d22b096e | 1623 | int fd = event_notifier_get_fd(e); |
80a1ea37 AK |
1624 | int r; |
1625 | ||
44c3f8f7 | 1626 | r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space, |
052e87b0 PB |
1627 | data, true, int128_get64(section->size), |
1628 | match_data); | |
80a1ea37 | 1629 | if (r < 0) { |
e346bcbf YK |
1630 | fprintf(stderr, "%s: error adding ioeventfd: %s (%d)\n", |
1631 | __func__, strerror(-r), -r); | |
80a1ea37 AK |
1632 | abort(); |
1633 | } | |
1634 | } | |
1635 | ||
d22b096e AK |
1636 | static void kvm_io_ioeventfd_del(MemoryListener *listener, |
1637 | MemoryRegionSection *section, | |
1638 | bool match_data, uint64_t data, | |
1639 | EventNotifier *e) | |
80a1ea37 AK |
1640 | |
1641 | { | |
d22b096e | 1642 | int fd = event_notifier_get_fd(e); |
80a1ea37 AK |
1643 | int r; |
1644 | ||
44c3f8f7 | 1645 | r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space, |
052e87b0 PB |
1646 | data, false, int128_get64(section->size), |
1647 | match_data); | |
80a1ea37 | 1648 | if (r < 0) { |
e346bcbf YK |
1649 | fprintf(stderr, "%s: error deleting ioeventfd: %s (%d)\n", |
1650 | __func__, strerror(-r), -r); | |
80a1ea37 AK |
1651 | abort(); |
1652 | } | |
1653 | } | |
1654 | ||
38bfe691 | 1655 | void kvm_memory_listener_register(KVMState *s, KVMMemoryListener *kml, |
142518bd | 1656 | AddressSpace *as, int as_id, const char *name) |
7bbda04c PB |
1657 | { |
1658 | int i; | |
1659 | ||
b21e2380 | 1660 | kml->slots = g_new0(KVMSlot, s->nr_slots); |
38bfe691 | 1661 | kml->as_id = as_id; |
7bbda04c PB |
1662 | |
1663 | for (i = 0; i < s->nr_slots; i++) { | |
1664 | kml->slots[i].slot = i; | |
1665 | } | |
1666 | ||
1667 | kml->listener.region_add = kvm_region_add; | |
1668 | kml->listener.region_del = kvm_region_del; | |
1669 | kml->listener.log_start = kvm_log_start; | |
1670 | kml->listener.log_stop = kvm_log_stop; | |
7bbda04c | 1671 | kml->listener.priority = 10; |
142518bd | 1672 | kml->listener.name = name; |
7bbda04c | 1673 | |
b4420f19 PX |
1674 | if (s->kvm_dirty_ring_size) { |
1675 | kml->listener.log_sync_global = kvm_log_sync_global; | |
1676 | } else { | |
1677 | kml->listener.log_sync = kvm_log_sync; | |
1678 | kml->listener.log_clear = kvm_log_clear; | |
1679 | } | |
1680 | ||
7bbda04c | 1681 | memory_listener_register(&kml->listener, as); |
8072aae3 AK |
1682 | |
1683 | for (i = 0; i < s->nr_as; ++i) { | |
1684 | if (!s->as[i].as) { | |
1685 | s->as[i].as = as; | |
1686 | s->as[i].ml = kml; | |
1687 | break; | |
1688 | } | |
1689 | } | |
7bbda04c | 1690 | } |
d22b096e AK |
1691 | |
1692 | static MemoryListener kvm_io_listener = { | |
142518bd | 1693 | .name = "kvm-io", |
d22b096e AK |
1694 | .eventfd_add = kvm_io_ioeventfd_add, |
1695 | .eventfd_del = kvm_io_ioeventfd_del, | |
72e22d2f | 1696 | .priority = 10, |
7b8f3b78 MT |
1697 | }; |
1698 | ||
3889c3fa | 1699 | int kvm_set_irq(KVMState *s, int irq, int level) |
84b058d7 JK |
1700 | { |
1701 | struct kvm_irq_level event; | |
1702 | int ret; | |
1703 | ||
7ae26bd4 | 1704 | assert(kvm_async_interrupts_enabled()); |
84b058d7 JK |
1705 | |
1706 | event.level = level; | |
1707 | event.irq = irq; | |
e333cd69 | 1708 | ret = kvm_vm_ioctl(s, s->irq_set_ioctl, &event); |
84b058d7 | 1709 | if (ret < 0) { |
3889c3fa | 1710 | perror("kvm_set_irq"); |
84b058d7 JK |
1711 | abort(); |
1712 | } | |
1713 | ||
e333cd69 | 1714 | return (s->irq_set_ioctl == KVM_IRQ_LINE) ? 1 : event.status; |
84b058d7 JK |
1715 | } |
1716 | ||
1717 | #ifdef KVM_CAP_IRQ_ROUTING | |
d3d3bef0 JK |
1718 | typedef struct KVMMSIRoute { |
1719 | struct kvm_irq_routing_entry kroute; | |
1720 | QTAILQ_ENTRY(KVMMSIRoute) entry; | |
1721 | } KVMMSIRoute; | |
1722 | ||
84b058d7 JK |
1723 | static void set_gsi(KVMState *s, unsigned int gsi) |
1724 | { | |
8269fb70 | 1725 | set_bit(gsi, s->used_gsi_bitmap); |
84b058d7 JK |
1726 | } |
1727 | ||
04fa27f5 JK |
1728 | static void clear_gsi(KVMState *s, unsigned int gsi) |
1729 | { | |
8269fb70 | 1730 | clear_bit(gsi, s->used_gsi_bitmap); |
04fa27f5 JK |
1731 | } |
1732 | ||
7b774593 | 1733 | void kvm_init_irq_routing(KVMState *s) |
84b058d7 | 1734 | { |
04fa27f5 | 1735 | int gsi_count, i; |
84b058d7 | 1736 | |
00008418 | 1737 | gsi_count = kvm_check_extension(s, KVM_CAP_IRQ_ROUTING) - 1; |
84b058d7 | 1738 | if (gsi_count > 0) { |
84b058d7 | 1739 | /* Round up so we can search ints using ffs */ |
8269fb70 | 1740 | s->used_gsi_bitmap = bitmap_new(gsi_count); |
4e2e4e63 | 1741 | s->gsi_count = gsi_count; |
84b058d7 JK |
1742 | } |
1743 | ||
1744 | s->irq_routes = g_malloc0(sizeof(*s->irq_routes)); | |
1745 | s->nr_allocated_irq_routes = 0; | |
1746 | ||
50bf31b9 | 1747 | if (!kvm_direct_msi_allowed) { |
4a3adebb JK |
1748 | for (i = 0; i < KVM_MSI_HASHTAB_SIZE; i++) { |
1749 | QTAILQ_INIT(&s->msi_hashtab[i]); | |
1750 | } | |
04fa27f5 JK |
1751 | } |
1752 | ||
84b058d7 JK |
1753 | kvm_arch_init_irq_routing(s); |
1754 | } | |
1755 | ||
cb925cf9 | 1756 | void kvm_irqchip_commit_routes(KVMState *s) |
e7b20308 JK |
1757 | { |
1758 | int ret; | |
1759 | ||
7005f7f8 PX |
1760 | if (kvm_gsi_direct_mapping()) { |
1761 | return; | |
1762 | } | |
1763 | ||
1764 | if (!kvm_gsi_routing_enabled()) { | |
1765 | return; | |
1766 | } | |
1767 | ||
e7b20308 | 1768 | s->irq_routes->flags = 0; |
54a6c11b | 1769 | trace_kvm_irqchip_commit_routes(); |
e7b20308 JK |
1770 | ret = kvm_vm_ioctl(s, KVM_SET_GSI_ROUTING, s->irq_routes); |
1771 | assert(ret == 0); | |
1772 | } | |
1773 | ||
84b058d7 JK |
1774 | static void kvm_add_routing_entry(KVMState *s, |
1775 | struct kvm_irq_routing_entry *entry) | |
1776 | { | |
1777 | struct kvm_irq_routing_entry *new; | |
1778 | int n, size; | |
1779 | ||
1780 | if (s->irq_routes->nr == s->nr_allocated_irq_routes) { | |
1781 | n = s->nr_allocated_irq_routes * 2; | |
1782 | if (n < 64) { | |
1783 | n = 64; | |
1784 | } | |
1785 | size = sizeof(struct kvm_irq_routing); | |
1786 | size += n * sizeof(*new); | |
1787 | s->irq_routes = g_realloc(s->irq_routes, size); | |
1788 | s->nr_allocated_irq_routes = n; | |
1789 | } | |
1790 | n = s->irq_routes->nr++; | |
1791 | new = &s->irq_routes->entries[n]; | |
0fbc2074 MT |
1792 | |
1793 | *new = *entry; | |
84b058d7 JK |
1794 | |
1795 | set_gsi(s, entry->gsi); | |
1796 | } | |
1797 | ||
cc57407e JK |
1798 | static int kvm_update_routing_entry(KVMState *s, |
1799 | struct kvm_irq_routing_entry *new_entry) | |
1800 | { | |
1801 | struct kvm_irq_routing_entry *entry; | |
1802 | int n; | |
1803 | ||
1804 | for (n = 0; n < s->irq_routes->nr; n++) { | |
1805 | entry = &s->irq_routes->entries[n]; | |
1806 | if (entry->gsi != new_entry->gsi) { | |
1807 | continue; | |
1808 | } | |
1809 | ||
40509f7f MT |
1810 | if(!memcmp(entry, new_entry, sizeof *entry)) { |
1811 | return 0; | |
1812 | } | |
1813 | ||
0fbc2074 | 1814 | *entry = *new_entry; |
cc57407e | 1815 | |
cc57407e JK |
1816 | return 0; |
1817 | } | |
1818 | ||
1819 | return -ESRCH; | |
1820 | } | |
1821 | ||
1df186df | 1822 | void kvm_irqchip_add_irq_route(KVMState *s, int irq, int irqchip, int pin) |
84b058d7 | 1823 | { |
0fbc2074 | 1824 | struct kvm_irq_routing_entry e = {}; |
84b058d7 | 1825 | |
4e2e4e63 JK |
1826 | assert(pin < s->gsi_count); |
1827 | ||
84b058d7 JK |
1828 | e.gsi = irq; |
1829 | e.type = KVM_IRQ_ROUTING_IRQCHIP; | |
1830 | e.flags = 0; | |
1831 | e.u.irqchip.irqchip = irqchip; | |
1832 | e.u.irqchip.pin = pin; | |
1833 | kvm_add_routing_entry(s, &e); | |
1834 | } | |
1835 | ||
1e2aa8be | 1836 | void kvm_irqchip_release_virq(KVMState *s, int virq) |
04fa27f5 JK |
1837 | { |
1838 | struct kvm_irq_routing_entry *e; | |
1839 | int i; | |
1840 | ||
76fe21de AK |
1841 | if (kvm_gsi_direct_mapping()) { |
1842 | return; | |
1843 | } | |
1844 | ||
04fa27f5 JK |
1845 | for (i = 0; i < s->irq_routes->nr; i++) { |
1846 | e = &s->irq_routes->entries[i]; | |
1847 | if (e->gsi == virq) { | |
1848 | s->irq_routes->nr--; | |
1849 | *e = s->irq_routes->entries[s->irq_routes->nr]; | |
1850 | } | |
1851 | } | |
1852 | clear_gsi(s, virq); | |
38d87493 | 1853 | kvm_arch_release_virq_post(virq); |
9ba35d0b | 1854 | trace_kvm_irqchip_release_virq(virq); |
04fa27f5 JK |
1855 | } |
1856 | ||
3607715a DG |
1857 | void kvm_irqchip_add_change_notifier(Notifier *n) |
1858 | { | |
1859 | notifier_list_add(&kvm_irqchip_change_notifiers, n); | |
1860 | } | |
1861 | ||
1862 | void kvm_irqchip_remove_change_notifier(Notifier *n) | |
1863 | { | |
1864 | notifier_remove(n); | |
1865 | } | |
1866 | ||
1867 | void kvm_irqchip_change_notify(void) | |
1868 | { | |
1869 | notifier_list_notify(&kvm_irqchip_change_notifiers, NULL); | |
1870 | } | |
1871 | ||
04fa27f5 JK |
1872 | static unsigned int kvm_hash_msi(uint32_t data) |
1873 | { | |
1874 | /* This is optimized for IA32 MSI layout. However, no other arch shall | |
1875 | * repeat the mistake of not providing a direct MSI injection API. */ | |
1876 | return data & 0xff; | |
1877 | } | |
1878 | ||
1879 | static void kvm_flush_dynamic_msi_routes(KVMState *s) | |
1880 | { | |
1881 | KVMMSIRoute *route, *next; | |
1882 | unsigned int hash; | |
1883 | ||
1884 | for (hash = 0; hash < KVM_MSI_HASHTAB_SIZE; hash++) { | |
1885 | QTAILQ_FOREACH_SAFE(route, &s->msi_hashtab[hash], entry, next) { | |
1886 | kvm_irqchip_release_virq(s, route->kroute.gsi); | |
1887 | QTAILQ_REMOVE(&s->msi_hashtab[hash], route, entry); | |
1888 | g_free(route); | |
1889 | } | |
1890 | } | |
1891 | } | |
1892 | ||
1893 | static int kvm_irqchip_get_virq(KVMState *s) | |
1894 | { | |
8269fb70 | 1895 | int next_virq; |
04fa27f5 | 1896 | |
bdf02631 WM |
1897 | /* |
1898 | * PIC and IOAPIC share the first 16 GSI numbers, thus the available | |
1899 | * GSI numbers are more than the number of IRQ route. Allocating a GSI | |
1900 | * number can succeed even though a new route entry cannot be added. | |
1901 | * When this happens, flush dynamic MSI entries to free IRQ route entries. | |
1902 | */ | |
50bf31b9 | 1903 | if (!kvm_direct_msi_allowed && s->irq_routes->nr == s->gsi_count) { |
bdf02631 WM |
1904 | kvm_flush_dynamic_msi_routes(s); |
1905 | } | |
1906 | ||
04fa27f5 | 1907 | /* Return the lowest unused GSI in the bitmap */ |
8269fb70 WY |
1908 | next_virq = find_first_zero_bit(s->used_gsi_bitmap, s->gsi_count); |
1909 | if (next_virq >= s->gsi_count) { | |
1910 | return -ENOSPC; | |
1911 | } else { | |
1912 | return next_virq; | |
04fa27f5 | 1913 | } |
04fa27f5 JK |
1914 | } |
1915 | ||
1916 | static KVMMSIRoute *kvm_lookup_msi_route(KVMState *s, MSIMessage msg) | |
1917 | { | |
1918 | unsigned int hash = kvm_hash_msi(msg.data); | |
1919 | KVMMSIRoute *route; | |
1920 | ||
1921 | QTAILQ_FOREACH(route, &s->msi_hashtab[hash], entry) { | |
1922 | if (route->kroute.u.msi.address_lo == (uint32_t)msg.address && | |
1923 | route->kroute.u.msi.address_hi == (msg.address >> 32) && | |
d07cc1f1 | 1924 | route->kroute.u.msi.data == le32_to_cpu(msg.data)) { |
04fa27f5 JK |
1925 | return route; |
1926 | } | |
1927 | } | |
1928 | return NULL; | |
1929 | } | |
1930 | ||
1931 | int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg) | |
1932 | { | |
4a3adebb | 1933 | struct kvm_msi msi; |
04fa27f5 JK |
1934 | KVMMSIRoute *route; |
1935 | ||
50bf31b9 | 1936 | if (kvm_direct_msi_allowed) { |
4a3adebb JK |
1937 | msi.address_lo = (uint32_t)msg.address; |
1938 | msi.address_hi = msg.address >> 32; | |
d07cc1f1 | 1939 | msi.data = le32_to_cpu(msg.data); |
4a3adebb JK |
1940 | msi.flags = 0; |
1941 | memset(msi.pad, 0, sizeof(msi.pad)); | |
1942 | ||
1943 | return kvm_vm_ioctl(s, KVM_SIGNAL_MSI, &msi); | |
1944 | } | |
1945 | ||
04fa27f5 JK |
1946 | route = kvm_lookup_msi_route(s, msg); |
1947 | if (!route) { | |
e7b20308 | 1948 | int virq; |
04fa27f5 JK |
1949 | |
1950 | virq = kvm_irqchip_get_virq(s); | |
1951 | if (virq < 0) { | |
1952 | return virq; | |
1953 | } | |
1954 | ||
b21e2380 | 1955 | route = g_new0(KVMMSIRoute, 1); |
04fa27f5 JK |
1956 | route->kroute.gsi = virq; |
1957 | route->kroute.type = KVM_IRQ_ROUTING_MSI; | |
1958 | route->kroute.flags = 0; | |
1959 | route->kroute.u.msi.address_lo = (uint32_t)msg.address; | |
1960 | route->kroute.u.msi.address_hi = msg.address >> 32; | |
d07cc1f1 | 1961 | route->kroute.u.msi.data = le32_to_cpu(msg.data); |
04fa27f5 JK |
1962 | |
1963 | kvm_add_routing_entry(s, &route->kroute); | |
cb925cf9 | 1964 | kvm_irqchip_commit_routes(s); |
04fa27f5 JK |
1965 | |
1966 | QTAILQ_INSERT_TAIL(&s->msi_hashtab[kvm_hash_msi(msg.data)], route, | |
1967 | entry); | |
04fa27f5 JK |
1968 | } |
1969 | ||
1970 | assert(route->kroute.type == KVM_IRQ_ROUTING_MSI); | |
1971 | ||
3889c3fa | 1972 | return kvm_set_irq(s, route->kroute.gsi, 1); |
04fa27f5 JK |
1973 | } |
1974 | ||
def4c557 | 1975 | int kvm_irqchip_add_msi_route(KVMRouteChange *c, int vector, PCIDevice *dev) |
92b4e489 | 1976 | { |
0fbc2074 | 1977 | struct kvm_irq_routing_entry kroute = {}; |
92b4e489 | 1978 | int virq; |
def4c557 | 1979 | KVMState *s = c->s; |
d1f6af6a PX |
1980 | MSIMessage msg = {0, 0}; |
1981 | ||
88c725c7 | 1982 | if (pci_available && dev) { |
e1d4fb2d | 1983 | msg = pci_get_msi_message(dev, vector); |
d1f6af6a | 1984 | } |
92b4e489 | 1985 | |
76fe21de | 1986 | if (kvm_gsi_direct_mapping()) { |
1850b6b7 | 1987 | return kvm_arch_msi_data_to_gsi(msg.data); |
76fe21de AK |
1988 | } |
1989 | ||
f3e1bed8 | 1990 | if (!kvm_gsi_routing_enabled()) { |
92b4e489 JK |
1991 | return -ENOSYS; |
1992 | } | |
1993 | ||
1994 | virq = kvm_irqchip_get_virq(s); | |
1995 | if (virq < 0) { | |
1996 | return virq; | |
1997 | } | |
1998 | ||
1999 | kroute.gsi = virq; | |
2000 | kroute.type = KVM_IRQ_ROUTING_MSI; | |
2001 | kroute.flags = 0; | |
2002 | kroute.u.msi.address_lo = (uint32_t)msg.address; | |
2003 | kroute.u.msi.address_hi = msg.address >> 32; | |
d07cc1f1 | 2004 | kroute.u.msi.data = le32_to_cpu(msg.data); |
88c725c7 | 2005 | if (pci_available && kvm_msi_devid_required()) { |
767a554a PF |
2006 | kroute.flags = KVM_MSI_VALID_DEVID; |
2007 | kroute.u.msi.devid = pci_requester_id(dev); | |
2008 | } | |
dc9f06ca | 2009 | if (kvm_arch_fixup_msi_route(&kroute, msg.address, msg.data, dev)) { |
9e03a040 FB |
2010 | kvm_irqchip_release_virq(s, virq); |
2011 | return -EINVAL; | |
2012 | } | |
92b4e489 | 2013 | |
9ba35d0b PX |
2014 | trace_kvm_irqchip_add_msi_route(dev ? dev->name : (char *)"N/A", |
2015 | vector, virq); | |
54a6c11b | 2016 | |
92b4e489 | 2017 | kvm_add_routing_entry(s, &kroute); |
38d87493 | 2018 | kvm_arch_add_msi_route_post(&kroute, vector, dev); |
def4c557 | 2019 | c->changes++; |
92b4e489 JK |
2020 | |
2021 | return virq; | |
2022 | } | |
2023 | ||
dc9f06ca PF |
2024 | int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg, |
2025 | PCIDevice *dev) | |
cc57407e | 2026 | { |
0fbc2074 | 2027 | struct kvm_irq_routing_entry kroute = {}; |
cc57407e | 2028 | |
76fe21de AK |
2029 | if (kvm_gsi_direct_mapping()) { |
2030 | return 0; | |
2031 | } | |
2032 | ||
cc57407e JK |
2033 | if (!kvm_irqchip_in_kernel()) { |
2034 | return -ENOSYS; | |
2035 | } | |
2036 | ||
2037 | kroute.gsi = virq; | |
2038 | kroute.type = KVM_IRQ_ROUTING_MSI; | |
2039 | kroute.flags = 0; | |
2040 | kroute.u.msi.address_lo = (uint32_t)msg.address; | |
2041 | kroute.u.msi.address_hi = msg.address >> 32; | |
d07cc1f1 | 2042 | kroute.u.msi.data = le32_to_cpu(msg.data); |
88c725c7 | 2043 | if (pci_available && kvm_msi_devid_required()) { |
767a554a PF |
2044 | kroute.flags = KVM_MSI_VALID_DEVID; |
2045 | kroute.u.msi.devid = pci_requester_id(dev); | |
2046 | } | |
dc9f06ca | 2047 | if (kvm_arch_fixup_msi_route(&kroute, msg.address, msg.data, dev)) { |
9e03a040 FB |
2048 | return -EINVAL; |
2049 | } | |
cc57407e | 2050 | |
54a6c11b PX |
2051 | trace_kvm_irqchip_update_msi_route(virq); |
2052 | ||
cc57407e JK |
2053 | return kvm_update_routing_entry(s, &kroute); |
2054 | } | |
2055 | ||
ff66ba87 PX |
2056 | static int kvm_irqchip_assign_irqfd(KVMState *s, EventNotifier *event, |
2057 | EventNotifier *resample, int virq, | |
ca916d37 | 2058 | bool assign) |
39853bbc | 2059 | { |
ff66ba87 PX |
2060 | int fd = event_notifier_get_fd(event); |
2061 | int rfd = resample ? event_notifier_get_fd(resample) : -1; | |
2062 | ||
39853bbc JK |
2063 | struct kvm_irqfd irqfd = { |
2064 | .fd = fd, | |
2065 | .gsi = virq, | |
2066 | .flags = assign ? 0 : KVM_IRQFD_FLAG_DEASSIGN, | |
2067 | }; | |
2068 | ||
ca916d37 | 2069 | if (rfd != -1) { |
c82d9d43 PX |
2070 | assert(assign); |
2071 | if (kvm_irqchip_is_split()) { | |
2072 | /* | |
2073 | * When the slow irqchip (e.g. IOAPIC) is in the | |
2074 | * userspace, KVM kernel resamplefd will not work because | |
2075 | * the EOI of the interrupt will be delivered to userspace | |
2076 | * instead, so the KVM kernel resamplefd kick will be | |
2077 | * skipped. The userspace here mimics what the kernel | |
2078 | * provides with resamplefd, remember the resamplefd and | |
2079 | * kick it when we receive EOI of this IRQ. | |
2080 | * | |
2081 | * This is hackery because IOAPIC is mostly bypassed | |
2082 | * (except EOI broadcasts) when irqfd is used. However | |
2083 | * this can bring much performance back for split irqchip | |
2084 | * with INTx IRQs (for VFIO, this gives 93% perf of the | |
2085 | * full fast path, which is 46% perf boost comparing to | |
2086 | * the INTx slow path). | |
2087 | */ | |
2088 | kvm_resample_fd_insert(virq, resample); | |
2089 | } else { | |
2090 | irqfd.flags |= KVM_IRQFD_FLAG_RESAMPLE; | |
2091 | irqfd.resamplefd = rfd; | |
2092 | } | |
2093 | } else if (!assign) { | |
2094 | if (kvm_irqchip_is_split()) { | |
2095 | kvm_resample_fd_remove(virq); | |
2096 | } | |
ca916d37 VM |
2097 | } |
2098 | ||
cc7e0ddf | 2099 | if (!kvm_irqfds_enabled()) { |
39853bbc JK |
2100 | return -ENOSYS; |
2101 | } | |
2102 | ||
2103 | return kvm_vm_ioctl(s, KVM_IRQFD, &irqfd); | |
2104 | } | |
2105 | ||
d426d9fb CH |
2106 | int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter) |
2107 | { | |
e9af2fef | 2108 | struct kvm_irq_routing_entry kroute = {}; |
d426d9fb CH |
2109 | int virq; |
2110 | ||
2111 | if (!kvm_gsi_routing_enabled()) { | |
2112 | return -ENOSYS; | |
2113 | } | |
2114 | ||
2115 | virq = kvm_irqchip_get_virq(s); | |
2116 | if (virq < 0) { | |
2117 | return virq; | |
2118 | } | |
2119 | ||
2120 | kroute.gsi = virq; | |
2121 | kroute.type = KVM_IRQ_ROUTING_S390_ADAPTER; | |
2122 | kroute.flags = 0; | |
2123 | kroute.u.adapter.summary_addr = adapter->summary_addr; | |
2124 | kroute.u.adapter.ind_addr = adapter->ind_addr; | |
2125 | kroute.u.adapter.summary_offset = adapter->summary_offset; | |
2126 | kroute.u.adapter.ind_offset = adapter->ind_offset; | |
2127 | kroute.u.adapter.adapter_id = adapter->adapter_id; | |
2128 | ||
2129 | kvm_add_routing_entry(s, &kroute); | |
d426d9fb CH |
2130 | |
2131 | return virq; | |
2132 | } | |
2133 | ||
977a8d9c AS |
2134 | int kvm_irqchip_add_hv_sint_route(KVMState *s, uint32_t vcpu, uint32_t sint) |
2135 | { | |
2136 | struct kvm_irq_routing_entry kroute = {}; | |
2137 | int virq; | |
2138 | ||
2139 | if (!kvm_gsi_routing_enabled()) { | |
2140 | return -ENOSYS; | |
2141 | } | |
2142 | if (!kvm_check_extension(s, KVM_CAP_HYPERV_SYNIC)) { | |
2143 | return -ENOSYS; | |
2144 | } | |
2145 | virq = kvm_irqchip_get_virq(s); | |
2146 | if (virq < 0) { | |
2147 | return virq; | |
2148 | } | |
2149 | ||
2150 | kroute.gsi = virq; | |
2151 | kroute.type = KVM_IRQ_ROUTING_HV_SINT; | |
2152 | kroute.flags = 0; | |
2153 | kroute.u.hv_sint.vcpu = vcpu; | |
2154 | kroute.u.hv_sint.sint = sint; | |
2155 | ||
2156 | kvm_add_routing_entry(s, &kroute); | |
2157 | kvm_irqchip_commit_routes(s); | |
2158 | ||
2159 | return virq; | |
2160 | } | |
2161 | ||
84b058d7 JK |
2162 | #else /* !KVM_CAP_IRQ_ROUTING */ |
2163 | ||
7b774593 | 2164 | void kvm_init_irq_routing(KVMState *s) |
84b058d7 JK |
2165 | { |
2166 | } | |
04fa27f5 | 2167 | |
d3d3bef0 JK |
2168 | void kvm_irqchip_release_virq(KVMState *s, int virq) |
2169 | { | |
2170 | } | |
2171 | ||
04fa27f5 JK |
2172 | int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg) |
2173 | { | |
2174 | abort(); | |
2175 | } | |
92b4e489 | 2176 | |
def4c557 | 2177 | int kvm_irqchip_add_msi_route(KVMRouteChange *c, int vector, PCIDevice *dev) |
92b4e489 | 2178 | { |
df410675 | 2179 | return -ENOSYS; |
92b4e489 | 2180 | } |
39853bbc | 2181 | |
d426d9fb CH |
2182 | int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter) |
2183 | { | |
2184 | return -ENOSYS; | |
2185 | } | |
2186 | ||
977a8d9c AS |
2187 | int kvm_irqchip_add_hv_sint_route(KVMState *s, uint32_t vcpu, uint32_t sint) |
2188 | { | |
2189 | return -ENOSYS; | |
2190 | } | |
2191 | ||
ff66ba87 PX |
2192 | static int kvm_irqchip_assign_irqfd(KVMState *s, EventNotifier *event, |
2193 | EventNotifier *resample, int virq, | |
2194 | bool assign) | |
39853bbc JK |
2195 | { |
2196 | abort(); | |
2197 | } | |
dabe3143 MT |
2198 | |
2199 | int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg) | |
2200 | { | |
2201 | return -ENOSYS; | |
2202 | } | |
84b058d7 JK |
2203 | #endif /* !KVM_CAP_IRQ_ROUTING */ |
2204 | ||
1c9b71a7 EA |
2205 | int kvm_irqchip_add_irqfd_notifier_gsi(KVMState *s, EventNotifier *n, |
2206 | EventNotifier *rn, int virq) | |
39853bbc | 2207 | { |
ff66ba87 | 2208 | return kvm_irqchip_assign_irqfd(s, n, rn, virq, true); |
39853bbc JK |
2209 | } |
2210 | ||
1c9b71a7 EA |
2211 | int kvm_irqchip_remove_irqfd_notifier_gsi(KVMState *s, EventNotifier *n, |
2212 | int virq) | |
15b2bd18 | 2213 | { |
ff66ba87 | 2214 | return kvm_irqchip_assign_irqfd(s, n, NULL, virq, false); |
15b2bd18 PB |
2215 | } |
2216 | ||
197e3524 EA |
2217 | int kvm_irqchip_add_irqfd_notifier(KVMState *s, EventNotifier *n, |
2218 | EventNotifier *rn, qemu_irq irq) | |
2219 | { | |
2220 | gpointer key, gsi; | |
2221 | gboolean found = g_hash_table_lookup_extended(s->gsimap, irq, &key, &gsi); | |
2222 | ||
2223 | if (!found) { | |
2224 | return -ENXIO; | |
2225 | } | |
2226 | return kvm_irqchip_add_irqfd_notifier_gsi(s, n, rn, GPOINTER_TO_INT(gsi)); | |
2227 | } | |
2228 | ||
2229 | int kvm_irqchip_remove_irqfd_notifier(KVMState *s, EventNotifier *n, | |
2230 | qemu_irq irq) | |
2231 | { | |
2232 | gpointer key, gsi; | |
2233 | gboolean found = g_hash_table_lookup_extended(s->gsimap, irq, &key, &gsi); | |
2234 | ||
2235 | if (!found) { | |
2236 | return -ENXIO; | |
2237 | } | |
2238 | return kvm_irqchip_remove_irqfd_notifier_gsi(s, n, GPOINTER_TO_INT(gsi)); | |
2239 | } | |
2240 | ||
2241 | void kvm_irqchip_set_qemuirq_gsi(KVMState *s, qemu_irq irq, int gsi) | |
2242 | { | |
2243 | g_hash_table_insert(s->gsimap, irq, GINT_TO_POINTER(gsi)); | |
2244 | } | |
2245 | ||
4376c40d | 2246 | static void kvm_irqchip_create(KVMState *s) |
84b058d7 | 2247 | { |
84b058d7 JK |
2248 | int ret; |
2249 | ||
d1972be1 | 2250 | assert(s->kernel_irqchip_split != ON_OFF_AUTO_AUTO); |
8db4936b PB |
2251 | if (kvm_check_extension(s, KVM_CAP_IRQCHIP)) { |
2252 | ; | |
2253 | } else if (kvm_check_extension(s, KVM_CAP_S390_IRQCHIP)) { | |
2254 | ret = kvm_vm_enable_cap(s, KVM_CAP_S390_IRQCHIP, 0); | |
2255 | if (ret < 0) { | |
2256 | fprintf(stderr, "Enable kernel irqchip failed: %s\n", strerror(-ret)); | |
2257 | exit(1); | |
2258 | } | |
2259 | } else { | |
2260 | return; | |
84b058d7 JK |
2261 | } |
2262 | ||
d6032e06 CD |
2263 | /* First probe and see if there's a arch-specific hook to create the |
2264 | * in-kernel irqchip for us */ | |
4376c40d | 2265 | ret = kvm_arch_irqchip_create(s); |
8db4936b | 2266 | if (ret == 0) { |
d1972be1 | 2267 | if (s->kernel_irqchip_split == ON_OFF_AUTO_ON) { |
47c182fe | 2268 | error_report("Split IRQ chip mode not supported."); |
15eafc2e PB |
2269 | exit(1); |
2270 | } else { | |
2271 | ret = kvm_vm_ioctl(s, KVM_CREATE_IRQCHIP); | |
2272 | } | |
8db4936b PB |
2273 | } |
2274 | if (ret < 0) { | |
2275 | fprintf(stderr, "Create kernel irqchip failed: %s\n", strerror(-ret)); | |
2276 | exit(1); | |
84b058d7 JK |
2277 | } |
2278 | ||
3d4b2649 | 2279 | kvm_kernel_irqchip = true; |
7ae26bd4 PM |
2280 | /* If we have an in-kernel IRQ chip then we must have asynchronous |
2281 | * interrupt delivery (though the reverse is not necessarily true) | |
2282 | */ | |
2283 | kvm_async_interrupts_allowed = true; | |
215e79c0 | 2284 | kvm_halt_in_kernel_allowed = true; |
84b058d7 JK |
2285 | |
2286 | kvm_init_irq_routing(s); | |
2287 | ||
197e3524 | 2288 | s->gsimap = g_hash_table_new(g_direct_hash, g_direct_equal); |
84b058d7 JK |
2289 | } |
2290 | ||
670436ce AJ |
2291 | /* Find number of supported CPUs using the recommended |
2292 | * procedure from the kernel API documentation to cope with | |
2293 | * older kernels that may be missing capabilities. | |
2294 | */ | |
2295 | static int kvm_recommended_vcpus(KVMState *s) | |
3ed444e9 | 2296 | { |
11748ba7 | 2297 | int ret = kvm_vm_check_extension(s, KVM_CAP_NR_VCPUS); |
670436ce AJ |
2298 | return (ret) ? ret : 4; |
2299 | } | |
3ed444e9 | 2300 | |
670436ce AJ |
2301 | static int kvm_max_vcpus(KVMState *s) |
2302 | { | |
2303 | int ret = kvm_check_extension(s, KVM_CAP_MAX_VCPUS); | |
2304 | return (ret) ? ret : kvm_recommended_vcpus(s); | |
3ed444e9 DH |
2305 | } |
2306 | ||
f31e3266 GK |
2307 | static int kvm_max_vcpu_id(KVMState *s) |
2308 | { | |
2309 | int ret = kvm_check_extension(s, KVM_CAP_MAX_VCPU_ID); | |
2310 | return (ret) ? ret : kvm_max_vcpus(s); | |
2311 | } | |
2312 | ||
41264b38 GK |
2313 | bool kvm_vcpu_id_is_valid(int vcpu_id) |
2314 | { | |
4f7f5893 | 2315 | KVMState *s = KVM_STATE(current_accel()); |
f31e3266 | 2316 | return vcpu_id >= 0 && vcpu_id < kvm_max_vcpu_id(s); |
41264b38 GK |
2317 | } |
2318 | ||
7786ae40 HH |
2319 | bool kvm_dirty_ring_enabled(void) |
2320 | { | |
2321 | return kvm_state->kvm_dirty_ring_size ? true : false; | |
2322 | } | |
2323 | ||
467ef823 | 2324 | static void query_stats_cb(StatsResultList **result, StatsTarget target, |
cf7405bc | 2325 | strList *names, strList *targets, Error **errp); |
cc01a3f4 MK |
2326 | static void query_stats_schemas_cb(StatsSchemaList **result, Error **errp); |
2327 | ||
4a06a7cc HH |
2328 | uint32_t kvm_dirty_ring_size(void) |
2329 | { | |
2330 | return kvm_state->kvm_dirty_ring_size; | |
2331 | } | |
2332 | ||
f6a1ef64 | 2333 | static int kvm_init(MachineState *ms) |
05330448 | 2334 | { |
f6a1ef64 | 2335 | MachineClass *mc = MACHINE_GET_CLASS(ms); |
168ccc11 JK |
2336 | static const char upgrade_note[] = |
2337 | "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n" | |
2338 | "(see http://sourceforge.net/projects/kvm).\n"; | |
670436ce AJ |
2339 | struct { |
2340 | const char *name; | |
2341 | int num; | |
2342 | } num_cpus[] = { | |
5cc8767d LX |
2343 | { "SMP", ms->smp.cpus }, |
2344 | { "hotpluggable", ms->smp.max_cpus }, | |
670436ce AJ |
2345 | { NULL, } |
2346 | }, *nc = num_cpus; | |
2347 | int soft_vcpus_limit, hard_vcpus_limit; | |
05330448 | 2348 | KVMState *s; |
94a8d39a | 2349 | const KVMCapabilityInfo *missing_cap; |
05330448 | 2350 | int ret; |
7bbda04c | 2351 | int type = 0; |
494cd11d | 2352 | uint64_t dirty_log_manual_caps; |
05330448 | 2353 | |
a2f77862 PX |
2354 | qemu_mutex_init(&kml_slots_lock); |
2355 | ||
fc02086b | 2356 | s = KVM_STATE(ms->accelerator); |
05330448 | 2357 | |
3145fcb6 DG |
2358 | /* |
2359 | * On systems where the kernel can support different base page | |
2360 | * sizes, host page size may be different from TARGET_PAGE_SIZE, | |
2361 | * even with KVM. TARGET_PAGE_SIZE is assumed to be the minimum | |
2362 | * page size for the system though. | |
2363 | */ | |
8e3b0cbb | 2364 | assert(TARGET_PAGE_SIZE <= qemu_real_host_page_size()); |
3145fcb6 | 2365 | |
aed6efb9 JH |
2366 | s->sigmask_len = 8; |
2367 | ||
e22a25c9 | 2368 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
72cf2d4f | 2369 | QTAILQ_INIT(&s->kvm_sw_breakpoints); |
e22a25c9 | 2370 | #endif |
4c055ab5 | 2371 | QLIST_INIT(&s->kvm_parked_vcpus); |
448058aa | 2372 | s->fd = qemu_open_old("/dev/kvm", O_RDWR); |
05330448 AL |
2373 | if (s->fd == -1) { |
2374 | fprintf(stderr, "Could not access KVM kernel module: %m\n"); | |
2375 | ret = -errno; | |
2376 | goto err; | |
2377 | } | |
2378 | ||
2379 | ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0); | |
2380 | if (ret < KVM_API_VERSION) { | |
0e1dac6c | 2381 | if (ret >= 0) { |
05330448 | 2382 | ret = -EINVAL; |
a426e122 | 2383 | } |
05330448 AL |
2384 | fprintf(stderr, "kvm version too old\n"); |
2385 | goto err; | |
2386 | } | |
2387 | ||
2388 | if (ret > KVM_API_VERSION) { | |
2389 | ret = -EINVAL; | |
2390 | fprintf(stderr, "kvm version not supported\n"); | |
2391 | goto err; | |
2392 | } | |
2393 | ||
cf0f7cf9 | 2394 | kvm_immediate_exit = kvm_check_extension(s, KVM_CAP_IMMEDIATE_EXIT); |
fb541ca5 AW |
2395 | s->nr_slots = kvm_check_extension(s, KVM_CAP_NR_MEMSLOTS); |
2396 | ||
2397 | /* If unspecified, use the default value */ | |
2398 | if (!s->nr_slots) { | |
2399 | s->nr_slots = 32; | |
2400 | } | |
2401 | ||
8072aae3 AK |
2402 | s->nr_as = kvm_check_extension(s, KVM_CAP_MULTI_ADDRESS_SPACE); |
2403 | if (s->nr_as <= 1) { | |
2404 | s->nr_as = 1; | |
2405 | } | |
2406 | s->as = g_new0(struct KVMAs, s->nr_as); | |
2407 | ||
f2ce39b4 PB |
2408 | if (object_property_find(OBJECT(current_machine), "kvm-type")) { |
2409 | g_autofree char *kvm_type = object_property_get_str(OBJECT(current_machine), | |
2410 | "kvm-type", | |
2411 | &error_abort); | |
dc0ca80e | 2412 | type = mc->kvm_type(ms, kvm_type); |
516fc0a0 AJ |
2413 | } else if (mc->kvm_type) { |
2414 | type = mc->kvm_type(ms, NULL); | |
135a129a AK |
2415 | } |
2416 | ||
94ccff13 | 2417 | do { |
135a129a | 2418 | ret = kvm_ioctl(s, KVM_CREATE_VM, type); |
94ccff13 TK |
2419 | } while (ret == -EINTR); |
2420 | ||
2421 | if (ret < 0) { | |
521f438e | 2422 | fprintf(stderr, "ioctl(KVM_CREATE_VM) failed: %d %s\n", -ret, |
94ccff13 TK |
2423 | strerror(-ret)); |
2424 | ||
0104dcac | 2425 | #ifdef TARGET_S390X |
2c80e996 CH |
2426 | if (ret == -EINVAL) { |
2427 | fprintf(stderr, | |
2428 | "Host kernel setup problem detected. Please verify:\n"); | |
2429 | fprintf(stderr, "- for kernels supporting the switch_amode or" | |
2430 | " user_mode parameters, whether\n"); | |
2431 | fprintf(stderr, | |
2432 | " user space is running in primary address space\n"); | |
2433 | fprintf(stderr, | |
2434 | "- for kernels supporting the vm.allocate_pgste sysctl, " | |
2435 | "whether it is enabled\n"); | |
2436 | } | |
380e4929 FR |
2437 | #elif defined(TARGET_PPC) |
2438 | if (ret == -EINVAL) { | |
2439 | fprintf(stderr, | |
2440 | "PPC KVM module is not loaded. Try modprobe kvm_%s.\n", | |
2441 | (type == 2) ? "pr" : "hv"); | |
2442 | } | |
0104dcac | 2443 | #endif |
05330448 | 2444 | goto err; |
0104dcac | 2445 | } |
05330448 | 2446 | |
94ccff13 | 2447 | s->vmfd = ret; |
11748ba7 GK |
2448 | |
2449 | /* check the vcpu limits */ | |
2450 | soft_vcpus_limit = kvm_recommended_vcpus(s); | |
2451 | hard_vcpus_limit = kvm_max_vcpus(s); | |
2452 | ||
2453 | while (nc->name) { | |
2454 | if (nc->num > soft_vcpus_limit) { | |
2455 | warn_report("Number of %s cpus requested (%d) exceeds " | |
2456 | "the recommended cpus supported by KVM (%d)", | |
2457 | nc->name, nc->num, soft_vcpus_limit); | |
2458 | ||
2459 | if (nc->num > hard_vcpus_limit) { | |
2460 | fprintf(stderr, "Number of %s cpus requested (%d) exceeds " | |
2461 | "the maximum cpus supported by KVM (%d)\n", | |
2462 | nc->name, nc->num, hard_vcpus_limit); | |
2463 | exit(1); | |
2464 | } | |
2465 | } | |
2466 | nc++; | |
2467 | } | |
2468 | ||
94a8d39a JK |
2469 | missing_cap = kvm_check_extension_list(s, kvm_required_capabilites); |
2470 | if (!missing_cap) { | |
2471 | missing_cap = | |
2472 | kvm_check_extension_list(s, kvm_arch_required_capabilities); | |
05330448 | 2473 | } |
94a8d39a | 2474 | if (missing_cap) { |
ad7b8b33 | 2475 | ret = -EINVAL; |
94a8d39a JK |
2476 | fprintf(stderr, "kvm does not support %s\n%s", |
2477 | missing_cap->name, upgrade_note); | |
d85dc283 AL |
2478 | goto err; |
2479 | } | |
2480 | ||
ad7b8b33 | 2481 | s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO); |
e6d34aee PH |
2482 | s->coalesced_pio = s->coalesced_mmio && |
2483 | kvm_check_extension(s, KVM_CAP_COALESCED_PIO); | |
f65ed4c1 | 2484 | |
b4420f19 PX |
2485 | /* |
2486 | * Enable KVM dirty ring if supported, otherwise fall back to | |
2487 | * dirty logging mode | |
2488 | */ | |
2489 | if (s->kvm_dirty_ring_size > 0) { | |
2490 | uint64_t ring_bytes; | |
2491 | ||
2492 | ring_bytes = s->kvm_dirty_ring_size * sizeof(struct kvm_dirty_gfn); | |
2493 | ||
2494 | /* Read the max supported pages */ | |
2495 | ret = kvm_vm_check_extension(s, KVM_CAP_DIRTY_LOG_RING); | |
2496 | if (ret > 0) { | |
2497 | if (ring_bytes > ret) { | |
2498 | error_report("KVM dirty ring size %" PRIu32 " too big " | |
2499 | "(maximum is %ld). Please use a smaller value.", | |
2500 | s->kvm_dirty_ring_size, | |
2501 | (long)ret / sizeof(struct kvm_dirty_gfn)); | |
2502 | ret = -EINVAL; | |
2503 | goto err; | |
2504 | } | |
2505 | ||
2506 | ret = kvm_vm_enable_cap(s, KVM_CAP_DIRTY_LOG_RING, 0, ring_bytes); | |
2507 | if (ret) { | |
2508 | error_report("Enabling of KVM dirty ring failed: %s. " | |
7916b5fc | 2509 | "Suggested minimum value is 1024.", strerror(-ret)); |
b4420f19 PX |
2510 | goto err; |
2511 | } | |
2512 | ||
2513 | s->kvm_dirty_ring_bytes = ring_bytes; | |
2514 | } else { | |
2515 | warn_report("KVM dirty ring not available, using bitmap method"); | |
2516 | s->kvm_dirty_ring_size = 0; | |
2517 | } | |
2518 | } | |
2519 | ||
a81a5926 PX |
2520 | /* |
2521 | * KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2 is not needed when dirty ring is | |
2522 | * enabled. More importantly, KVM_DIRTY_LOG_INITIALLY_SET will assume no | |
2523 | * page is wr-protected initially, which is against how kvm dirty ring is | |
2524 | * usage - kvm dirty ring requires all pages are wr-protected at the very | |
2525 | * beginning. Enabling this feature for dirty ring causes data corruption. | |
b4420f19 PX |
2526 | * |
2527 | * TODO: Without KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2 and kvm clear dirty log, | |
2528 | * we may expect a higher stall time when starting the migration. In the | |
2529 | * future we can enable KVM_CLEAR_DIRTY_LOG to work with dirty ring too: | |
2530 | * instead of clearing dirty bit, it can be a way to explicitly wr-protect | |
2531 | * guest pages. | |
a81a5926 PX |
2532 | */ |
2533 | if (!s->kvm_dirty_ring_size) { | |
2534 | dirty_log_manual_caps = | |
2535 | kvm_check_extension(s, KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2); | |
2536 | dirty_log_manual_caps &= (KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE | | |
2537 | KVM_DIRTY_LOG_INITIALLY_SET); | |
2538 | s->manual_dirty_log_protect = dirty_log_manual_caps; | |
2539 | if (dirty_log_manual_caps) { | |
2540 | ret = kvm_vm_enable_cap(s, KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2, 0, | |
2541 | dirty_log_manual_caps); | |
2542 | if (ret) { | |
2543 | warn_report("Trying to enable capability %"PRIu64" of " | |
2544 | "KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2 but failed. " | |
2545 | "Falling back to the legacy mode. ", | |
2546 | dirty_log_manual_caps); | |
2547 | s->manual_dirty_log_protect = 0; | |
2548 | } | |
ff4aa114 PX |
2549 | } |
2550 | } | |
2551 | ||
a0fb002c JK |
2552 | #ifdef KVM_CAP_VCPU_EVENTS |
2553 | s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS); | |
2554 | #endif | |
2555 | ||
b0b1d690 JK |
2556 | s->robust_singlestep = |
2557 | kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP); | |
b0b1d690 | 2558 | |
ff44f1a3 JK |
2559 | #ifdef KVM_CAP_DEBUGREGS |
2560 | s->debugregs = kvm_check_extension(s, KVM_CAP_DEBUGREGS); | |
2561 | #endif | |
2562 | ||
ebbfef2f LA |
2563 | s->max_nested_state_len = kvm_check_extension(s, KVM_CAP_NESTED_STATE); |
2564 | ||
d3d3bef0 | 2565 | #ifdef KVM_CAP_IRQ_ROUTING |
50bf31b9 | 2566 | kvm_direct_msi_allowed = (kvm_check_extension(s, KVM_CAP_SIGNAL_MSI) > 0); |
d3d3bef0 | 2567 | #endif |
4a3adebb | 2568 | |
3ab73842 JK |
2569 | s->intx_set_mask = kvm_check_extension(s, KVM_CAP_PCI_2_3); |
2570 | ||
e333cd69 | 2571 | s->irq_set_ioctl = KVM_IRQ_LINE; |
8732fbd2 | 2572 | if (kvm_check_extension(s, KVM_CAP_IRQ_INJECT_STATUS)) { |
e333cd69 | 2573 | s->irq_set_ioctl = KVM_IRQ_LINE_STATUS; |
8732fbd2 PM |
2574 | } |
2575 | ||
df9c8b75 JJ |
2576 | kvm_readonly_mem_allowed = |
2577 | (kvm_check_extension(s, KVM_CAP_READONLY_MEM) > 0); | |
df9c8b75 | 2578 | |
69e03ae6 NN |
2579 | kvm_eventfds_allowed = |
2580 | (kvm_check_extension(s, KVM_CAP_IOEVENTFD) > 0); | |
2581 | ||
f41389ae EA |
2582 | kvm_irqfds_allowed = |
2583 | (kvm_check_extension(s, KVM_CAP_IRQFD) > 0); | |
2584 | ||
2585 | kvm_resamplefds_allowed = | |
2586 | (kvm_check_extension(s, KVM_CAP_IRQFD_RESAMPLE) > 0); | |
2587 | ||
d0a073a1 DD |
2588 | kvm_vm_attributes_allowed = |
2589 | (kvm_check_extension(s, KVM_CAP_VM_ATTRIBUTES) > 0); | |
2590 | ||
35108223 JW |
2591 | kvm_ioeventfd_any_length_allowed = |
2592 | (kvm_check_extension(s, KVM_CAP_IOEVENTFD_ANY_LENGTH) > 0); | |
2593 | ||
12bc5b4c ML |
2594 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
2595 | kvm_has_guest_debug = | |
2596 | (kvm_check_extension(s, KVM_CAP_SET_GUEST_DEBUG) > 0); | |
2597 | #endif | |
2598 | ||
2599 | kvm_sstep_flags = 0; | |
2600 | if (kvm_has_guest_debug) { | |
2601 | kvm_sstep_flags = SSTEP_ENABLE; | |
fd2ddd16 ML |
2602 | |
2603 | #if defined KVM_CAP_SET_GUEST_DEBUG2 | |
2604 | int guest_debug_flags = | |
2605 | kvm_check_extension(s, KVM_CAP_SET_GUEST_DEBUG2); | |
2606 | ||
2607 | if (guest_debug_flags & KVM_GUESTDBG_BLOCKIRQ) { | |
2608 | kvm_sstep_flags |= SSTEP_NOIRQ; | |
2609 | } | |
2610 | #endif | |
12bc5b4c ML |
2611 | } |
2612 | ||
d870cfde GA |
2613 | kvm_state = s; |
2614 | ||
b16565b3 | 2615 | ret = kvm_arch_init(ms, s); |
a426e122 | 2616 | if (ret < 0) { |
05330448 | 2617 | goto err; |
a426e122 | 2618 | } |
05330448 | 2619 | |
d1972be1 XL |
2620 | if (s->kernel_irqchip_split == ON_OFF_AUTO_AUTO) { |
2621 | s->kernel_irqchip_split = mc->default_kernel_irqchip_split ? ON_OFF_AUTO_ON : ON_OFF_AUTO_OFF; | |
2622 | } | |
2623 | ||
6b552b9b DG |
2624 | qemu_register_reset(kvm_unpoison_all, NULL); |
2625 | ||
11bc4a13 | 2626 | if (s->kernel_irqchip_allowed) { |
4376c40d | 2627 | kvm_irqchip_create(s); |
84b058d7 JK |
2628 | } |
2629 | ||
8c56c1a5 PF |
2630 | if (kvm_eventfds_allowed) { |
2631 | s->memory_listener.listener.eventfd_add = kvm_mem_ioeventfd_add; | |
2632 | s->memory_listener.listener.eventfd_del = kvm_mem_ioeventfd_del; | |
2633 | } | |
e6d34aee PH |
2634 | s->memory_listener.listener.coalesced_io_add = kvm_coalesce_mmio_region; |
2635 | s->memory_listener.listener.coalesced_io_del = kvm_uncoalesce_mmio_region; | |
7bbda04c PB |
2636 | |
2637 | kvm_memory_listener_register(s, &s->memory_listener, | |
142518bd | 2638 | &address_space_memory, 0, "kvm-memory"); |
f9b49088 EA |
2639 | if (kvm_eventfds_allowed) { |
2640 | memory_listener_register(&kvm_io_listener, | |
2641 | &address_space_io); | |
2642 | } | |
e6d34aee PH |
2643 | memory_listener_register(&kvm_coalesced_pio_listener, |
2644 | &address_space_io); | |
05330448 | 2645 | |
d2f2b8a7 SH |
2646 | s->many_ioeventfds = kvm_check_many_ioeventfds(); |
2647 | ||
62dd4eda | 2648 | s->sync_mmu = !!kvm_vm_check_extension(kvm_state, KVM_CAP_SYNC_MMU); |
f5948942 | 2649 | if (!s->sync_mmu) { |
956b109f DH |
2650 | ret = ram_block_discard_disable(true); |
2651 | assert(!ret); | |
f5948942 | 2652 | } |
b4420f19 PX |
2653 | |
2654 | if (s->kvm_dirty_ring_size) { | |
2655 | ret = kvm_dirty_ring_reaper_init(s); | |
2656 | if (ret) { | |
2657 | goto err; | |
2658 | } | |
2659 | } | |
2660 | ||
cc01a3f4 | 2661 | if (kvm_check_extension(kvm_state, KVM_CAP_BINARY_STATS_FD)) { |
068cc51d PB |
2662 | add_stats_callbacks(STATS_PROVIDER_KVM, query_stats_cb, |
2663 | query_stats_schemas_cb); | |
cc01a3f4 MK |
2664 | } |
2665 | ||
05330448 AL |
2666 | return 0; |
2667 | ||
2668 | err: | |
0e1dac6c | 2669 | assert(ret < 0); |
6d1cc321 SW |
2670 | if (s->vmfd >= 0) { |
2671 | close(s->vmfd); | |
2672 | } | |
2673 | if (s->fd != -1) { | |
2674 | close(s->fd); | |
05330448 | 2675 | } |
7bbda04c | 2676 | g_free(s->memory_listener.slots); |
05330448 AL |
2677 | |
2678 | return ret; | |
2679 | } | |
2680 | ||
aed6efb9 JH |
2681 | void kvm_set_sigmask_len(KVMState *s, unsigned int sigmask_len) |
2682 | { | |
2683 | s->sigmask_len = sigmask_len; | |
2684 | } | |
2685 | ||
4c663752 PB |
2686 | static void kvm_handle_io(uint16_t port, MemTxAttrs attrs, void *data, int direction, |
2687 | int size, uint32_t count) | |
05330448 AL |
2688 | { |
2689 | int i; | |
2690 | uint8_t *ptr = data; | |
2691 | ||
2692 | for (i = 0; i < count; i++) { | |
4c663752 | 2693 | address_space_rw(&address_space_io, port, attrs, |
5c9eb028 | 2694 | ptr, size, |
354678c5 | 2695 | direction == KVM_EXIT_IO_OUT); |
05330448 AL |
2696 | ptr += size; |
2697 | } | |
05330448 AL |
2698 | } |
2699 | ||
5326ab55 | 2700 | static int kvm_handle_internal_error(CPUState *cpu, struct kvm_run *run) |
7c80eef8 | 2701 | { |
977c7b6d RK |
2702 | fprintf(stderr, "KVM internal error. Suberror: %d\n", |
2703 | run->internal.suberror); | |
2704 | ||
7c80eef8 MT |
2705 | if (kvm_check_extension(kvm_state, KVM_CAP_INTERNAL_ERROR_DATA)) { |
2706 | int i; | |
2707 | ||
7c80eef8 | 2708 | for (i = 0; i < run->internal.ndata; ++i) { |
56567da3 | 2709 | fprintf(stderr, "extra data[%d]: 0x%016"PRIx64"\n", |
7c80eef8 MT |
2710 | i, (uint64_t)run->internal.data[i]); |
2711 | } | |
2712 | } | |
7c80eef8 MT |
2713 | if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) { |
2714 | fprintf(stderr, "emulation failure\n"); | |
20d695a9 | 2715 | if (!kvm_arch_stop_on_emulation_error(cpu)) { |
90c84c56 | 2716 | cpu_dump_state(cpu, stderr, CPU_DUMP_CODE); |
d73cd8f4 | 2717 | return EXCP_INTERRUPT; |
a426e122 | 2718 | } |
7c80eef8 MT |
2719 | } |
2720 | /* FIXME: Should trigger a qmp message to let management know | |
2721 | * something went wrong. | |
2722 | */ | |
73aaec4a | 2723 | return -1; |
7c80eef8 | 2724 | } |
7c80eef8 | 2725 | |
62a2744c | 2726 | void kvm_flush_coalesced_mmio_buffer(void) |
f65ed4c1 | 2727 | { |
f65ed4c1 | 2728 | KVMState *s = kvm_state; |
1cae88b9 AK |
2729 | |
2730 | if (s->coalesced_flush_in_progress) { | |
2731 | return; | |
2732 | } | |
2733 | ||
2734 | s->coalesced_flush_in_progress = true; | |
2735 | ||
62a2744c SY |
2736 | if (s->coalesced_mmio_ring) { |
2737 | struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring; | |
f65ed4c1 AL |
2738 | while (ring->first != ring->last) { |
2739 | struct kvm_coalesced_mmio *ent; | |
2740 | ||
2741 | ent = &ring->coalesced_mmio[ring->first]; | |
2742 | ||
e6d34aee | 2743 | if (ent->pio == 1) { |
19f70347 PM |
2744 | address_space_write(&address_space_io, ent->phys_addr, |
2745 | MEMTXATTRS_UNSPECIFIED, ent->data, | |
2746 | ent->len); | |
e6d34aee PH |
2747 | } else { |
2748 | cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len); | |
2749 | } | |
85199474 | 2750 | smp_wmb(); |
f65ed4c1 AL |
2751 | ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX; |
2752 | } | |
2753 | } | |
1cae88b9 AK |
2754 | |
2755 | s->coalesced_flush_in_progress = false; | |
f65ed4c1 AL |
2756 | } |
2757 | ||
92a5199b TL |
2758 | bool kvm_cpu_check_are_resettable(void) |
2759 | { | |
2760 | return kvm_arch_cpu_check_are_resettable(); | |
2761 | } | |
2762 | ||
14e6fe12 | 2763 | static void do_kvm_cpu_synchronize_state(CPUState *cpu, run_on_cpu_data arg) |
4c0960c0 | 2764 | { |
99f31832 | 2765 | if (!cpu->vcpu_dirty) { |
20d695a9 | 2766 | kvm_arch_get_registers(cpu); |
99f31832 | 2767 | cpu->vcpu_dirty = true; |
4c0960c0 AK |
2768 | } |
2769 | } | |
2770 | ||
dd1750d7 | 2771 | void kvm_cpu_synchronize_state(CPUState *cpu) |
2705d56a | 2772 | { |
99f31832 | 2773 | if (!cpu->vcpu_dirty) { |
14e6fe12 | 2774 | run_on_cpu(cpu, do_kvm_cpu_synchronize_state, RUN_ON_CPU_NULL); |
a426e122 | 2775 | } |
2705d56a JK |
2776 | } |
2777 | ||
14e6fe12 | 2778 | static void do_kvm_cpu_synchronize_post_reset(CPUState *cpu, run_on_cpu_data arg) |
ea375f9a | 2779 | { |
20d695a9 | 2780 | kvm_arch_put_registers(cpu, KVM_PUT_RESET_STATE); |
99f31832 | 2781 | cpu->vcpu_dirty = false; |
ea375f9a JK |
2782 | } |
2783 | ||
c8e2085d DH |
2784 | void kvm_cpu_synchronize_post_reset(CPUState *cpu) |
2785 | { | |
14e6fe12 | 2786 | run_on_cpu(cpu, do_kvm_cpu_synchronize_post_reset, RUN_ON_CPU_NULL); |
c8e2085d DH |
2787 | } |
2788 | ||
14e6fe12 | 2789 | static void do_kvm_cpu_synchronize_post_init(CPUState *cpu, run_on_cpu_data arg) |
ea375f9a | 2790 | { |
20d695a9 | 2791 | kvm_arch_put_registers(cpu, KVM_PUT_FULL_STATE); |
99f31832 | 2792 | cpu->vcpu_dirty = false; |
ea375f9a JK |
2793 | } |
2794 | ||
c8e2085d DH |
2795 | void kvm_cpu_synchronize_post_init(CPUState *cpu) |
2796 | { | |
14e6fe12 | 2797 | run_on_cpu(cpu, do_kvm_cpu_synchronize_post_init, RUN_ON_CPU_NULL); |
c8e2085d DH |
2798 | } |
2799 | ||
75e972da DG |
2800 | static void do_kvm_cpu_synchronize_pre_loadvm(CPUState *cpu, run_on_cpu_data arg) |
2801 | { | |
99f31832 | 2802 | cpu->vcpu_dirty = true; |
75e972da DG |
2803 | } |
2804 | ||
2805 | void kvm_cpu_synchronize_pre_loadvm(CPUState *cpu) | |
2806 | { | |
2807 | run_on_cpu(cpu, do_kvm_cpu_synchronize_pre_loadvm, RUN_ON_CPU_NULL); | |
2808 | } | |
2809 | ||
2ae41db2 PB |
2810 | #ifdef KVM_HAVE_MCE_INJECTION |
2811 | static __thread void *pending_sigbus_addr; | |
2812 | static __thread int pending_sigbus_code; | |
2813 | static __thread bool have_sigbus_pending; | |
2814 | #endif | |
2815 | ||
cf0f7cf9 PB |
2816 | static void kvm_cpu_kick(CPUState *cpu) |
2817 | { | |
d73415a3 | 2818 | qatomic_set(&cpu->kvm_run->immediate_exit, 1); |
cf0f7cf9 PB |
2819 | } |
2820 | ||
2821 | static void kvm_cpu_kick_self(void) | |
2822 | { | |
2823 | if (kvm_immediate_exit) { | |
2824 | kvm_cpu_kick(current_cpu); | |
2825 | } else { | |
2826 | qemu_cpu_kick_self(); | |
2827 | } | |
2828 | } | |
2829 | ||
18268b60 PB |
2830 | static void kvm_eat_signals(CPUState *cpu) |
2831 | { | |
2832 | struct timespec ts = { 0, 0 }; | |
2833 | siginfo_t siginfo; | |
2834 | sigset_t waitset; | |
2835 | sigset_t chkset; | |
2836 | int r; | |
2837 | ||
cf0f7cf9 | 2838 | if (kvm_immediate_exit) { |
d73415a3 | 2839 | qatomic_set(&cpu->kvm_run->immediate_exit, 0); |
cf0f7cf9 PB |
2840 | /* Write kvm_run->immediate_exit before the cpu->exit_request |
2841 | * write in kvm_cpu_exec. | |
2842 | */ | |
2843 | smp_wmb(); | |
2844 | return; | |
2845 | } | |
2846 | ||
18268b60 PB |
2847 | sigemptyset(&waitset); |
2848 | sigaddset(&waitset, SIG_IPI); | |
2849 | ||
2850 | do { | |
2851 | r = sigtimedwait(&waitset, &siginfo, &ts); | |
2852 | if (r == -1 && !(errno == EAGAIN || errno == EINTR)) { | |
2853 | perror("sigtimedwait"); | |
2854 | exit(1); | |
2855 | } | |
2856 | ||
2857 | r = sigpending(&chkset); | |
2858 | if (r == -1) { | |
2859 | perror("sigpending"); | |
2860 | exit(1); | |
2861 | } | |
2862 | } while (sigismember(&chkset, SIG_IPI)); | |
2863 | } | |
2864 | ||
1458c363 | 2865 | int kvm_cpu_exec(CPUState *cpu) |
05330448 | 2866 | { |
f7575c96 | 2867 | struct kvm_run *run = cpu->kvm_run; |
7cbb533f | 2868 | int ret, run_ret; |
05330448 | 2869 | |
8c0d577e | 2870 | DPRINTF("kvm_cpu_exec()\n"); |
05330448 | 2871 | |
20d695a9 | 2872 | if (kvm_arch_process_async_events(cpu)) { |
d73415a3 | 2873 | qatomic_set(&cpu->exit_request, 0); |
6792a57b | 2874 | return EXCP_HLT; |
9ccfac9e | 2875 | } |
0af691d7 | 2876 | |
4b8523ee | 2877 | qemu_mutex_unlock_iothread(); |
1d78a3c3 | 2878 | cpu_exec_start(cpu); |
4b8523ee | 2879 | |
9ccfac9e | 2880 | do { |
4c663752 PB |
2881 | MemTxAttrs attrs; |
2882 | ||
99f31832 | 2883 | if (cpu->vcpu_dirty) { |
20d695a9 | 2884 | kvm_arch_put_registers(cpu, KVM_PUT_RUNTIME_STATE); |
99f31832 | 2885 | cpu->vcpu_dirty = false; |
4c0960c0 AK |
2886 | } |
2887 | ||
20d695a9 | 2888 | kvm_arch_pre_run(cpu, run); |
d73415a3 | 2889 | if (qatomic_read(&cpu->exit_request)) { |
9ccfac9e JK |
2890 | DPRINTF("interrupt exit requested\n"); |
2891 | /* | |
2892 | * KVM requires us to reenter the kernel after IO exits to complete | |
2893 | * instruction emulation. This self-signal will ensure that we | |
2894 | * leave ASAP again. | |
2895 | */ | |
cf0f7cf9 | 2896 | kvm_cpu_kick_self(); |
9ccfac9e | 2897 | } |
9ccfac9e | 2898 | |
cf0f7cf9 PB |
2899 | /* Read cpu->exit_request before KVM_RUN reads run->immediate_exit. |
2900 | * Matching barrier in kvm_eat_signals. | |
2901 | */ | |
2902 | smp_rmb(); | |
2903 | ||
1bc22652 | 2904 | run_ret = kvm_vcpu_ioctl(cpu, KVM_RUN, 0); |
9ccfac9e | 2905 | |
4c663752 | 2906 | attrs = kvm_arch_post_run(cpu, run); |
05330448 | 2907 | |
2ae41db2 PB |
2908 | #ifdef KVM_HAVE_MCE_INJECTION |
2909 | if (unlikely(have_sigbus_pending)) { | |
2910 | qemu_mutex_lock_iothread(); | |
2911 | kvm_arch_on_sigbus_vcpu(cpu, pending_sigbus_code, | |
2912 | pending_sigbus_addr); | |
2913 | have_sigbus_pending = false; | |
2914 | qemu_mutex_unlock_iothread(); | |
2915 | } | |
2916 | #endif | |
2917 | ||
7cbb533f | 2918 | if (run_ret < 0) { |
dc77d341 JK |
2919 | if (run_ret == -EINTR || run_ret == -EAGAIN) { |
2920 | DPRINTF("io window exit\n"); | |
18268b60 | 2921 | kvm_eat_signals(cpu); |
d73cd8f4 | 2922 | ret = EXCP_INTERRUPT; |
dc77d341 JK |
2923 | break; |
2924 | } | |
7b011fbc ME |
2925 | fprintf(stderr, "error: kvm run failed %s\n", |
2926 | strerror(-run_ret)); | |
dae02ba5 LV |
2927 | #ifdef TARGET_PPC |
2928 | if (run_ret == -EBUSY) { | |
2929 | fprintf(stderr, | |
2930 | "This is probably because your SMT is enabled.\n" | |
2931 | "VCPU can only run on primary threads with all " | |
2932 | "secondary threads offline.\n"); | |
2933 | } | |
2934 | #endif | |
a85e130e PB |
2935 | ret = -1; |
2936 | break; | |
05330448 AL |
2937 | } |
2938 | ||
b76ac80a | 2939 | trace_kvm_run_exit(cpu->cpu_index, run->exit_reason); |
05330448 AL |
2940 | switch (run->exit_reason) { |
2941 | case KVM_EXIT_IO: | |
8c0d577e | 2942 | DPRINTF("handle_io\n"); |
80b7d2ef | 2943 | /* Called outside BQL */ |
4c663752 | 2944 | kvm_handle_io(run->io.port, attrs, |
b30e93e9 JK |
2945 | (uint8_t *)run + run->io.data_offset, |
2946 | run->io.direction, | |
2947 | run->io.size, | |
2948 | run->io.count); | |
d73cd8f4 | 2949 | ret = 0; |
05330448 AL |
2950 | break; |
2951 | case KVM_EXIT_MMIO: | |
8c0d577e | 2952 | DPRINTF("handle_mmio\n"); |
de7ea885 | 2953 | /* Called outside BQL */ |
4c663752 PB |
2954 | address_space_rw(&address_space_memory, |
2955 | run->mmio.phys_addr, attrs, | |
2956 | run->mmio.data, | |
2957 | run->mmio.len, | |
2958 | run->mmio.is_write); | |
d73cd8f4 | 2959 | ret = 0; |
05330448 AL |
2960 | break; |
2961 | case KVM_EXIT_IRQ_WINDOW_OPEN: | |
8c0d577e | 2962 | DPRINTF("irq_window_open\n"); |
d73cd8f4 | 2963 | ret = EXCP_INTERRUPT; |
05330448 AL |
2964 | break; |
2965 | case KVM_EXIT_SHUTDOWN: | |
8c0d577e | 2966 | DPRINTF("shutdown\n"); |
cf83f140 | 2967 | qemu_system_reset_request(SHUTDOWN_CAUSE_GUEST_RESET); |
d73cd8f4 | 2968 | ret = EXCP_INTERRUPT; |
05330448 AL |
2969 | break; |
2970 | case KVM_EXIT_UNKNOWN: | |
bb44e0d1 JK |
2971 | fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "\n", |
2972 | (uint64_t)run->hw.hardware_exit_reason); | |
73aaec4a | 2973 | ret = -1; |
05330448 | 2974 | break; |
7c80eef8 | 2975 | case KVM_EXIT_INTERNAL_ERROR: |
5326ab55 | 2976 | ret = kvm_handle_internal_error(cpu, run); |
7c80eef8 | 2977 | break; |
b4420f19 PX |
2978 | case KVM_EXIT_DIRTY_RING_FULL: |
2979 | /* | |
2980 | * We shouldn't continue if the dirty ring of this vcpu is | |
2981 | * still full. Got kicked by KVM_RESET_DIRTY_RINGS. | |
2982 | */ | |
2983 | trace_kvm_dirty_ring_full(cpu->cpu_index); | |
2984 | qemu_mutex_lock_iothread(); | |
baa60983 HH |
2985 | /* |
2986 | * We throttle vCPU by making it sleep once it exit from kernel | |
2987 | * due to dirty ring full. In the dirtylimit scenario, reaping | |
2988 | * all vCPUs after a single vCPU dirty ring get full result in | |
2989 | * the miss of sleep, so just reap the ring-fulled vCPU. | |
2990 | */ | |
2991 | if (dirtylimit_in_service()) { | |
2992 | kvm_dirty_ring_reap(kvm_state, cpu); | |
2993 | } else { | |
2994 | kvm_dirty_ring_reap(kvm_state, NULL); | |
2995 | } | |
b4420f19 | 2996 | qemu_mutex_unlock_iothread(); |
baa60983 | 2997 | dirtylimit_vcpu_execute(cpu); |
b4420f19 PX |
2998 | ret = 0; |
2999 | break; | |
99040447 PS |
3000 | case KVM_EXIT_SYSTEM_EVENT: |
3001 | switch (run->system_event.type) { | |
3002 | case KVM_SYSTEM_EVENT_SHUTDOWN: | |
cf83f140 | 3003 | qemu_system_shutdown_request(SHUTDOWN_CAUSE_GUEST_SHUTDOWN); |
99040447 PS |
3004 | ret = EXCP_INTERRUPT; |
3005 | break; | |
3006 | case KVM_SYSTEM_EVENT_RESET: | |
cf83f140 | 3007 | qemu_system_reset_request(SHUTDOWN_CAUSE_GUEST_RESET); |
99040447 PS |
3008 | ret = EXCP_INTERRUPT; |
3009 | break; | |
7c207b90 | 3010 | case KVM_SYSTEM_EVENT_CRASH: |
d187e08d | 3011 | kvm_cpu_synchronize_state(cpu); |
7c207b90 | 3012 | qemu_mutex_lock_iothread(); |
c86f106b | 3013 | qemu_system_guest_panicked(cpu_get_crash_info(cpu)); |
7c207b90 AS |
3014 | qemu_mutex_unlock_iothread(); |
3015 | ret = 0; | |
3016 | break; | |
99040447 PS |
3017 | default: |
3018 | DPRINTF("kvm_arch_handle_exit\n"); | |
3019 | ret = kvm_arch_handle_exit(cpu, run); | |
3020 | break; | |
3021 | } | |
3022 | break; | |
05330448 | 3023 | default: |
8c0d577e | 3024 | DPRINTF("kvm_arch_handle_exit\n"); |
20d695a9 | 3025 | ret = kvm_arch_handle_exit(cpu, run); |
05330448 AL |
3026 | break; |
3027 | } | |
d73cd8f4 | 3028 | } while (ret == 0); |
05330448 | 3029 | |
1d78a3c3 | 3030 | cpu_exec_end(cpu); |
4b8523ee JK |
3031 | qemu_mutex_lock_iothread(); |
3032 | ||
73aaec4a | 3033 | if (ret < 0) { |
90c84c56 | 3034 | cpu_dump_state(cpu, stderr, CPU_DUMP_CODE); |
0461d5a6 | 3035 | vm_stop(RUN_STATE_INTERNAL_ERROR); |
becfc390 AL |
3036 | } |
3037 | ||
d73415a3 | 3038 | qatomic_set(&cpu->exit_request, 0); |
05330448 AL |
3039 | return ret; |
3040 | } | |
3041 | ||
984b5181 | 3042 | int kvm_ioctl(KVMState *s, int type, ...) |
05330448 AL |
3043 | { |
3044 | int ret; | |
984b5181 AL |
3045 | void *arg; |
3046 | va_list ap; | |
05330448 | 3047 | |
984b5181 AL |
3048 | va_start(ap, type); |
3049 | arg = va_arg(ap, void *); | |
3050 | va_end(ap); | |
3051 | ||
9c775729 | 3052 | trace_kvm_ioctl(type, arg); |
984b5181 | 3053 | ret = ioctl(s->fd, type, arg); |
a426e122 | 3054 | if (ret == -1) { |
05330448 | 3055 | ret = -errno; |
a426e122 | 3056 | } |
05330448 AL |
3057 | return ret; |
3058 | } | |
3059 | ||
984b5181 | 3060 | int kvm_vm_ioctl(KVMState *s, int type, ...) |
05330448 AL |
3061 | { |
3062 | int ret; | |
984b5181 AL |
3063 | void *arg; |
3064 | va_list ap; | |
3065 | ||
3066 | va_start(ap, type); | |
3067 | arg = va_arg(ap, void *); | |
3068 | va_end(ap); | |
05330448 | 3069 | |
9c775729 | 3070 | trace_kvm_vm_ioctl(type, arg); |
984b5181 | 3071 | ret = ioctl(s->vmfd, type, arg); |
a426e122 | 3072 | if (ret == -1) { |
05330448 | 3073 | ret = -errno; |
a426e122 | 3074 | } |
05330448 AL |
3075 | return ret; |
3076 | } | |
3077 | ||
1bc22652 | 3078 | int kvm_vcpu_ioctl(CPUState *cpu, int type, ...) |
05330448 AL |
3079 | { |
3080 | int ret; | |
984b5181 AL |
3081 | void *arg; |
3082 | va_list ap; | |
3083 | ||
3084 | va_start(ap, type); | |
3085 | arg = va_arg(ap, void *); | |
3086 | va_end(ap); | |
05330448 | 3087 | |
9c775729 | 3088 | trace_kvm_vcpu_ioctl(cpu->cpu_index, type, arg); |
8737c51c | 3089 | ret = ioctl(cpu->kvm_fd, type, arg); |
a426e122 | 3090 | if (ret == -1) { |
05330448 | 3091 | ret = -errno; |
a426e122 | 3092 | } |
05330448 AL |
3093 | return ret; |
3094 | } | |
bd322087 | 3095 | |
0a6a7cca CD |
3096 | int kvm_device_ioctl(int fd, int type, ...) |
3097 | { | |
3098 | int ret; | |
3099 | void *arg; | |
3100 | va_list ap; | |
3101 | ||
3102 | va_start(ap, type); | |
3103 | arg = va_arg(ap, void *); | |
3104 | va_end(ap); | |
3105 | ||
3106 | trace_kvm_device_ioctl(fd, type, arg); | |
3107 | ret = ioctl(fd, type, arg); | |
3108 | if (ret == -1) { | |
3109 | ret = -errno; | |
3110 | } | |
3111 | return ret; | |
3112 | } | |
3113 | ||
d0a073a1 DD |
3114 | int kvm_vm_check_attr(KVMState *s, uint32_t group, uint64_t attr) |
3115 | { | |
3116 | int ret; | |
3117 | struct kvm_device_attr attribute = { | |
3118 | .group = group, | |
3119 | .attr = attr, | |
3120 | }; | |
3121 | ||
3122 | if (!kvm_vm_attributes_allowed) { | |
3123 | return 0; | |
3124 | } | |
3125 | ||
3126 | ret = kvm_vm_ioctl(s, KVM_HAS_DEVICE_ATTR, &attribute); | |
3127 | /* kvm returns 0 on success for HAS_DEVICE_ATTR */ | |
3128 | return ret ? 0 : 1; | |
3129 | } | |
3130 | ||
4b3cfe72 PF |
3131 | int kvm_device_check_attr(int dev_fd, uint32_t group, uint64_t attr) |
3132 | { | |
3133 | struct kvm_device_attr attribute = { | |
3134 | .group = group, | |
3135 | .attr = attr, | |
3136 | .flags = 0, | |
3137 | }; | |
3138 | ||
3139 | return kvm_device_ioctl(dev_fd, KVM_HAS_DEVICE_ATTR, &attribute) ? 0 : 1; | |
3140 | } | |
3141 | ||
556969e9 EA |
3142 | int kvm_device_access(int fd, int group, uint64_t attr, |
3143 | void *val, bool write, Error **errp) | |
4b3cfe72 PF |
3144 | { |
3145 | struct kvm_device_attr kvmattr; | |
3146 | int err; | |
3147 | ||
3148 | kvmattr.flags = 0; | |
3149 | kvmattr.group = group; | |
3150 | kvmattr.attr = attr; | |
3151 | kvmattr.addr = (uintptr_t)val; | |
3152 | ||
3153 | err = kvm_device_ioctl(fd, | |
3154 | write ? KVM_SET_DEVICE_ATTR : KVM_GET_DEVICE_ATTR, | |
3155 | &kvmattr); | |
3156 | if (err < 0) { | |
556969e9 EA |
3157 | error_setg_errno(errp, -err, |
3158 | "KVM_%s_DEVICE_ATTR failed: Group %d " | |
3159 | "attr 0x%016" PRIx64, | |
3160 | write ? "SET" : "GET", group, attr); | |
4b3cfe72 | 3161 | } |
556969e9 | 3162 | return err; |
4b3cfe72 PF |
3163 | } |
3164 | ||
62dd4eda | 3165 | bool kvm_has_sync_mmu(void) |
bd322087 | 3166 | { |
62dd4eda | 3167 | return kvm_state->sync_mmu; |
bd322087 | 3168 | } |
e22a25c9 | 3169 | |
a0fb002c JK |
3170 | int kvm_has_vcpu_events(void) |
3171 | { | |
3172 | return kvm_state->vcpu_events; | |
3173 | } | |
3174 | ||
b0b1d690 JK |
3175 | int kvm_has_robust_singlestep(void) |
3176 | { | |
3177 | return kvm_state->robust_singlestep; | |
3178 | } | |
3179 | ||
ff44f1a3 JK |
3180 | int kvm_has_debugregs(void) |
3181 | { | |
3182 | return kvm_state->debugregs; | |
3183 | } | |
3184 | ||
ebbfef2f LA |
3185 | int kvm_max_nested_state_length(void) |
3186 | { | |
3187 | return kvm_state->max_nested_state_len; | |
3188 | } | |
3189 | ||
d2f2b8a7 SH |
3190 | int kvm_has_many_ioeventfds(void) |
3191 | { | |
3192 | if (!kvm_enabled()) { | |
3193 | return 0; | |
3194 | } | |
3195 | return kvm_state->many_ioeventfds; | |
3196 | } | |
3197 | ||
84b058d7 JK |
3198 | int kvm_has_gsi_routing(void) |
3199 | { | |
a9c5eb0d | 3200 | #ifdef KVM_CAP_IRQ_ROUTING |
84b058d7 | 3201 | return kvm_check_extension(kvm_state, KVM_CAP_IRQ_ROUTING); |
a9c5eb0d AG |
3202 | #else |
3203 | return false; | |
3204 | #endif | |
84b058d7 JK |
3205 | } |
3206 | ||
3ab73842 JK |
3207 | int kvm_has_intx_set_mask(void) |
3208 | { | |
3209 | return kvm_state->intx_set_mask; | |
3210 | } | |
3211 | ||
5d721b78 AG |
3212 | bool kvm_arm_supports_user_irq(void) |
3213 | { | |
3214 | return kvm_check_extension(kvm_state, KVM_CAP_ARM_USER_IRQ); | |
3215 | } | |
3216 | ||
e22a25c9 | 3217 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
a60f24b5 | 3218 | struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *cpu, |
e22a25c9 AL |
3219 | target_ulong pc) |
3220 | { | |
3221 | struct kvm_sw_breakpoint *bp; | |
3222 | ||
a60f24b5 | 3223 | QTAILQ_FOREACH(bp, &cpu->kvm_state->kvm_sw_breakpoints, entry) { |
a426e122 | 3224 | if (bp->pc == pc) { |
e22a25c9 | 3225 | return bp; |
a426e122 | 3226 | } |
e22a25c9 AL |
3227 | } |
3228 | return NULL; | |
3229 | } | |
3230 | ||
a60f24b5 | 3231 | int kvm_sw_breakpoints_active(CPUState *cpu) |
e22a25c9 | 3232 | { |
a60f24b5 | 3233 | return !QTAILQ_EMPTY(&cpu->kvm_state->kvm_sw_breakpoints); |
e22a25c9 AL |
3234 | } |
3235 | ||
452e4751 GC |
3236 | struct kvm_set_guest_debug_data { |
3237 | struct kvm_guest_debug dbg; | |
452e4751 GC |
3238 | int err; |
3239 | }; | |
3240 | ||
14e6fe12 | 3241 | static void kvm_invoke_set_guest_debug(CPUState *cpu, run_on_cpu_data data) |
452e4751 | 3242 | { |
14e6fe12 PB |
3243 | struct kvm_set_guest_debug_data *dbg_data = |
3244 | (struct kvm_set_guest_debug_data *) data.host_ptr; | |
b3807725 | 3245 | |
3c0ed2a3 | 3246 | dbg_data->err = kvm_vcpu_ioctl(cpu, KVM_SET_GUEST_DEBUG, |
a60f24b5 | 3247 | &dbg_data->dbg); |
452e4751 GC |
3248 | } |
3249 | ||
38e478ec | 3250 | int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap) |
e22a25c9 | 3251 | { |
452e4751 | 3252 | struct kvm_set_guest_debug_data data; |
e22a25c9 | 3253 | |
b0b1d690 | 3254 | data.dbg.control = reinject_trap; |
e22a25c9 | 3255 | |
ed2803da | 3256 | if (cpu->singlestep_enabled) { |
b0b1d690 | 3257 | data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP; |
fd2ddd16 ML |
3258 | |
3259 | if (cpu->singlestep_enabled & SSTEP_NOIRQ) { | |
3260 | data.dbg.control |= KVM_GUESTDBG_BLOCKIRQ; | |
3261 | } | |
b0b1d690 | 3262 | } |
20d695a9 | 3263 | kvm_arch_update_guest_debug(cpu, &data.dbg); |
e22a25c9 | 3264 | |
14e6fe12 PB |
3265 | run_on_cpu(cpu, kvm_invoke_set_guest_debug, |
3266 | RUN_ON_CPU_HOST_PTR(&data)); | |
452e4751 | 3267 | return data.err; |
e22a25c9 AL |
3268 | } |
3269 | ||
62278814 | 3270 | int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr, |
e22a25c9 AL |
3271 | target_ulong len, int type) |
3272 | { | |
3273 | struct kvm_sw_breakpoint *bp; | |
e22a25c9 AL |
3274 | int err; |
3275 | ||
3276 | if (type == GDB_BREAKPOINT_SW) { | |
80b7cd73 | 3277 | bp = kvm_find_sw_breakpoint(cpu, addr); |
e22a25c9 AL |
3278 | if (bp) { |
3279 | bp->use_count++; | |
3280 | return 0; | |
3281 | } | |
3282 | ||
b21e2380 | 3283 | bp = g_new(struct kvm_sw_breakpoint, 1); |
e22a25c9 AL |
3284 | bp->pc = addr; |
3285 | bp->use_count = 1; | |
80b7cd73 | 3286 | err = kvm_arch_insert_sw_breakpoint(cpu, bp); |
e22a25c9 | 3287 | if (err) { |
7267c094 | 3288 | g_free(bp); |
e22a25c9 AL |
3289 | return err; |
3290 | } | |
3291 | ||
80b7cd73 | 3292 | QTAILQ_INSERT_HEAD(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry); |
e22a25c9 AL |
3293 | } else { |
3294 | err = kvm_arch_insert_hw_breakpoint(addr, len, type); | |
a426e122 | 3295 | if (err) { |
e22a25c9 | 3296 | return err; |
a426e122 | 3297 | } |
e22a25c9 AL |
3298 | } |
3299 | ||
bdc44640 | 3300 | CPU_FOREACH(cpu) { |
38e478ec | 3301 | err = kvm_update_guest_debug(cpu, 0); |
a426e122 | 3302 | if (err) { |
e22a25c9 | 3303 | return err; |
a426e122 | 3304 | } |
e22a25c9 AL |
3305 | } |
3306 | return 0; | |
3307 | } | |
3308 | ||
62278814 | 3309 | int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr, |
e22a25c9 AL |
3310 | target_ulong len, int type) |
3311 | { | |
3312 | struct kvm_sw_breakpoint *bp; | |
e22a25c9 AL |
3313 | int err; |
3314 | ||
3315 | if (type == GDB_BREAKPOINT_SW) { | |
80b7cd73 | 3316 | bp = kvm_find_sw_breakpoint(cpu, addr); |
a426e122 | 3317 | if (!bp) { |
e22a25c9 | 3318 | return -ENOENT; |
a426e122 | 3319 | } |
e22a25c9 AL |
3320 | |
3321 | if (bp->use_count > 1) { | |
3322 | bp->use_count--; | |
3323 | return 0; | |
3324 | } | |
3325 | ||
80b7cd73 | 3326 | err = kvm_arch_remove_sw_breakpoint(cpu, bp); |
a426e122 | 3327 | if (err) { |
e22a25c9 | 3328 | return err; |
a426e122 | 3329 | } |
e22a25c9 | 3330 | |
80b7cd73 | 3331 | QTAILQ_REMOVE(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry); |
7267c094 | 3332 | g_free(bp); |
e22a25c9 AL |
3333 | } else { |
3334 | err = kvm_arch_remove_hw_breakpoint(addr, len, type); | |
a426e122 | 3335 | if (err) { |
e22a25c9 | 3336 | return err; |
a426e122 | 3337 | } |
e22a25c9 AL |
3338 | } |
3339 | ||
bdc44640 | 3340 | CPU_FOREACH(cpu) { |
38e478ec | 3341 | err = kvm_update_guest_debug(cpu, 0); |
a426e122 | 3342 | if (err) { |
e22a25c9 | 3343 | return err; |
a426e122 | 3344 | } |
e22a25c9 AL |
3345 | } |
3346 | return 0; | |
3347 | } | |
3348 | ||
1d5791f4 | 3349 | void kvm_remove_all_breakpoints(CPUState *cpu) |
e22a25c9 AL |
3350 | { |
3351 | struct kvm_sw_breakpoint *bp, *next; | |
80b7cd73 | 3352 | KVMState *s = cpu->kvm_state; |
dc54e252 | 3353 | CPUState *tmpcpu; |
e22a25c9 | 3354 | |
72cf2d4f | 3355 | QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) { |
80b7cd73 | 3356 | if (kvm_arch_remove_sw_breakpoint(cpu, bp) != 0) { |
e22a25c9 | 3357 | /* Try harder to find a CPU that currently sees the breakpoint. */ |
dc54e252 CG |
3358 | CPU_FOREACH(tmpcpu) { |
3359 | if (kvm_arch_remove_sw_breakpoint(tmpcpu, bp) == 0) { | |
e22a25c9 | 3360 | break; |
a426e122 | 3361 | } |
e22a25c9 AL |
3362 | } |
3363 | } | |
78021d6d JK |
3364 | QTAILQ_REMOVE(&s->kvm_sw_breakpoints, bp, entry); |
3365 | g_free(bp); | |
e22a25c9 AL |
3366 | } |
3367 | kvm_arch_remove_all_hw_breakpoints(); | |
3368 | ||
bdc44640 | 3369 | CPU_FOREACH(cpu) { |
38e478ec | 3370 | kvm_update_guest_debug(cpu, 0); |
a426e122 | 3371 | } |
e22a25c9 AL |
3372 | } |
3373 | ||
3374 | #else /* !KVM_CAP_SET_GUEST_DEBUG */ | |
3375 | ||
38e478ec | 3376 | int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap) |
e22a25c9 AL |
3377 | { |
3378 | return -EINVAL; | |
3379 | } | |
3380 | ||
62278814 | 3381 | int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr, |
e22a25c9 AL |
3382 | target_ulong len, int type) |
3383 | { | |
3384 | return -EINVAL; | |
3385 | } | |
3386 | ||
62278814 | 3387 | int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr, |
e22a25c9 AL |
3388 | target_ulong len, int type) |
3389 | { | |
3390 | return -EINVAL; | |
3391 | } | |
3392 | ||
1d5791f4 | 3393 | void kvm_remove_all_breakpoints(CPUState *cpu) |
e22a25c9 AL |
3394 | { |
3395 | } | |
3396 | #endif /* !KVM_CAP_SET_GUEST_DEBUG */ | |
cc84de95 | 3397 | |
18268b60 | 3398 | static int kvm_set_signal_mask(CPUState *cpu, const sigset_t *sigset) |
cc84de95 | 3399 | { |
aed6efb9 | 3400 | KVMState *s = kvm_state; |
cc84de95 MT |
3401 | struct kvm_signal_mask *sigmask; |
3402 | int r; | |
3403 | ||
7267c094 | 3404 | sigmask = g_malloc(sizeof(*sigmask) + sizeof(*sigset)); |
cc84de95 | 3405 | |
aed6efb9 | 3406 | sigmask->len = s->sigmask_len; |
cc84de95 | 3407 | memcpy(sigmask->sigset, sigset, sizeof(*sigset)); |
1bc22652 | 3408 | r = kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, sigmask); |
7267c094 | 3409 | g_free(sigmask); |
cc84de95 MT |
3410 | |
3411 | return r; | |
3412 | } | |
4d39892c | 3413 | |
cf0f7cf9 | 3414 | static void kvm_ipi_signal(int sig) |
18268b60 | 3415 | { |
cf0f7cf9 PB |
3416 | if (current_cpu) { |
3417 | assert(kvm_immediate_exit); | |
3418 | kvm_cpu_kick(current_cpu); | |
3419 | } | |
18268b60 PB |
3420 | } |
3421 | ||
3422 | void kvm_init_cpu_signals(CPUState *cpu) | |
3423 | { | |
3424 | int r; | |
3425 | sigset_t set; | |
3426 | struct sigaction sigact; | |
3427 | ||
3428 | memset(&sigact, 0, sizeof(sigact)); | |
cf0f7cf9 | 3429 | sigact.sa_handler = kvm_ipi_signal; |
18268b60 PB |
3430 | sigaction(SIG_IPI, &sigact, NULL); |
3431 | ||
3432 | pthread_sigmask(SIG_BLOCK, NULL, &set); | |
3433 | #if defined KVM_HAVE_MCE_INJECTION | |
3434 | sigdelset(&set, SIGBUS); | |
3435 | pthread_sigmask(SIG_SETMASK, &set, NULL); | |
3436 | #endif | |
3437 | sigdelset(&set, SIG_IPI); | |
cf0f7cf9 PB |
3438 | if (kvm_immediate_exit) { |
3439 | r = pthread_sigmask(SIG_SETMASK, &set, NULL); | |
3440 | } else { | |
3441 | r = kvm_set_signal_mask(cpu, &set); | |
3442 | } | |
18268b60 PB |
3443 | if (r) { |
3444 | fprintf(stderr, "kvm_set_signal_mask: %s\n", strerror(-r)); | |
3445 | exit(1); | |
3446 | } | |
3447 | } | |
3448 | ||
2ae41db2 | 3449 | /* Called asynchronously in VCPU thread. */ |
290adf38 | 3450 | int kvm_on_sigbus_vcpu(CPUState *cpu, int code, void *addr) |
a1b87fe0 | 3451 | { |
2ae41db2 PB |
3452 | #ifdef KVM_HAVE_MCE_INJECTION |
3453 | if (have_sigbus_pending) { | |
3454 | return 1; | |
3455 | } | |
3456 | have_sigbus_pending = true; | |
3457 | pending_sigbus_addr = addr; | |
3458 | pending_sigbus_code = code; | |
d73415a3 | 3459 | qatomic_set(&cpu->exit_request, 1); |
2ae41db2 PB |
3460 | return 0; |
3461 | #else | |
3462 | return 1; | |
3463 | #endif | |
a1b87fe0 JK |
3464 | } |
3465 | ||
2ae41db2 | 3466 | /* Called synchronously (via signalfd) in main thread. */ |
a1b87fe0 JK |
3467 | int kvm_on_sigbus(int code, void *addr) |
3468 | { | |
2ae41db2 | 3469 | #ifdef KVM_HAVE_MCE_INJECTION |
4d39892c PB |
3470 | /* Action required MCE kills the process if SIGBUS is blocked. Because |
3471 | * that's what happens in the I/O thread, where we handle MCE via signalfd, | |
3472 | * we can only get action optional here. | |
3473 | */ | |
3474 | assert(code != BUS_MCEERR_AR); | |
3475 | kvm_arch_on_sigbus_vcpu(first_cpu, code, addr); | |
3476 | return 0; | |
2ae41db2 PB |
3477 | #else |
3478 | return 1; | |
3479 | #endif | |
a1b87fe0 | 3480 | } |
0a6a7cca CD |
3481 | |
3482 | int kvm_create_device(KVMState *s, uint64_t type, bool test) | |
3483 | { | |
3484 | int ret; | |
3485 | struct kvm_create_device create_dev; | |
3486 | ||
3487 | create_dev.type = type; | |
3488 | create_dev.fd = -1; | |
3489 | create_dev.flags = test ? KVM_CREATE_DEVICE_TEST : 0; | |
3490 | ||
3491 | if (!kvm_check_extension(s, KVM_CAP_DEVICE_CTRL)) { | |
3492 | return -ENOTSUP; | |
3493 | } | |
3494 | ||
3495 | ret = kvm_vm_ioctl(s, KVM_CREATE_DEVICE, &create_dev); | |
3496 | if (ret) { | |
3497 | return ret; | |
3498 | } | |
3499 | ||
3500 | return test ? 0 : create_dev.fd; | |
3501 | } | |
ada4135f | 3502 | |
29039acf PX |
3503 | bool kvm_device_supported(int vmfd, uint64_t type) |
3504 | { | |
3505 | struct kvm_create_device create_dev = { | |
3506 | .type = type, | |
3507 | .fd = -1, | |
3508 | .flags = KVM_CREATE_DEVICE_TEST, | |
3509 | }; | |
3510 | ||
3511 | if (ioctl(vmfd, KVM_CHECK_EXTENSION, KVM_CAP_DEVICE_CTRL) <= 0) { | |
3512 | return false; | |
3513 | } | |
3514 | ||
3515 | return (ioctl(vmfd, KVM_CREATE_DEVICE, &create_dev) >= 0); | |
3516 | } | |
3517 | ||
ada4135f CH |
3518 | int kvm_set_one_reg(CPUState *cs, uint64_t id, void *source) |
3519 | { | |
3520 | struct kvm_one_reg reg; | |
3521 | int r; | |
3522 | ||
3523 | reg.id = id; | |
3524 | reg.addr = (uintptr_t) source; | |
3525 | r = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®); | |
3526 | if (r) { | |
844a3d34 | 3527 | trace_kvm_failed_reg_set(id, strerror(-r)); |
ada4135f CH |
3528 | } |
3529 | return r; | |
3530 | } | |
3531 | ||
3532 | int kvm_get_one_reg(CPUState *cs, uint64_t id, void *target) | |
3533 | { | |
3534 | struct kvm_one_reg reg; | |
3535 | int r; | |
3536 | ||
3537 | reg.id = id; | |
3538 | reg.addr = (uintptr_t) target; | |
3539 | r = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®); | |
3540 | if (r) { | |
844a3d34 | 3541 | trace_kvm_failed_reg_get(id, strerror(-r)); |
ada4135f CH |
3542 | } |
3543 | return r; | |
3544 | } | |
782c3f29 | 3545 | |
8072aae3 AK |
3546 | static bool kvm_accel_has_memory(MachineState *ms, AddressSpace *as, |
3547 | hwaddr start_addr, hwaddr size) | |
3548 | { | |
3549 | KVMState *kvm = KVM_STATE(ms->accelerator); | |
3550 | int i; | |
3551 | ||
3552 | for (i = 0; i < kvm->nr_as; ++i) { | |
3553 | if (kvm->as[i].as == as && kvm->as[i].ml) { | |
023ae9a8 | 3554 | size = MIN(kvm_max_slot_size, size); |
8072aae3 AK |
3555 | return NULL != kvm_lookup_matching_slot(kvm->as[i].ml, |
3556 | start_addr, size); | |
3557 | } | |
3558 | } | |
3559 | ||
3560 | return false; | |
3561 | } | |
3562 | ||
23b0898e PB |
3563 | static void kvm_get_kvm_shadow_mem(Object *obj, Visitor *v, |
3564 | const char *name, void *opaque, | |
3565 | Error **errp) | |
3566 | { | |
3567 | KVMState *s = KVM_STATE(obj); | |
3568 | int64_t value = s->kvm_shadow_mem; | |
3569 | ||
3570 | visit_type_int(v, name, &value, errp); | |
3571 | } | |
3572 | ||
3573 | static void kvm_set_kvm_shadow_mem(Object *obj, Visitor *v, | |
3574 | const char *name, void *opaque, | |
3575 | Error **errp) | |
3576 | { | |
3577 | KVMState *s = KVM_STATE(obj); | |
23b0898e PB |
3578 | int64_t value; |
3579 | ||
70cbae42 PB |
3580 | if (s->fd != -1) { |
3581 | error_setg(errp, "Cannot set properties after the accelerator has been initialized"); | |
3582 | return; | |
3583 | } | |
3584 | ||
668f62ec | 3585 | if (!visit_type_int(v, name, &value, errp)) { |
23b0898e PB |
3586 | return; |
3587 | } | |
3588 | ||
3589 | s->kvm_shadow_mem = value; | |
3590 | } | |
3591 | ||
11bc4a13 PB |
3592 | static void kvm_set_kernel_irqchip(Object *obj, Visitor *v, |
3593 | const char *name, void *opaque, | |
3594 | Error **errp) | |
3595 | { | |
11bc4a13 PB |
3596 | KVMState *s = KVM_STATE(obj); |
3597 | OnOffSplit mode; | |
3598 | ||
70cbae42 PB |
3599 | if (s->fd != -1) { |
3600 | error_setg(errp, "Cannot set properties after the accelerator has been initialized"); | |
3601 | return; | |
3602 | } | |
3603 | ||
14217038 | 3604 | if (!visit_type_OnOffSplit(v, name, &mode, errp)) { |
11bc4a13 | 3605 | return; |
14217038 MA |
3606 | } |
3607 | switch (mode) { | |
3608 | case ON_OFF_SPLIT_ON: | |
3609 | s->kernel_irqchip_allowed = true; | |
3610 | s->kernel_irqchip_required = true; | |
3611 | s->kernel_irqchip_split = ON_OFF_AUTO_OFF; | |
3612 | break; | |
3613 | case ON_OFF_SPLIT_OFF: | |
3614 | s->kernel_irqchip_allowed = false; | |
3615 | s->kernel_irqchip_required = false; | |
3616 | s->kernel_irqchip_split = ON_OFF_AUTO_OFF; | |
3617 | break; | |
3618 | case ON_OFF_SPLIT_SPLIT: | |
3619 | s->kernel_irqchip_allowed = true; | |
3620 | s->kernel_irqchip_required = true; | |
3621 | s->kernel_irqchip_split = ON_OFF_AUTO_ON; | |
3622 | break; | |
3623 | default: | |
3624 | /* The value was checked in visit_type_OnOffSplit() above. If | |
3625 | * we get here, then something is wrong in QEMU. | |
3626 | */ | |
3627 | abort(); | |
11bc4a13 PB |
3628 | } |
3629 | } | |
3630 | ||
4376c40d PB |
3631 | bool kvm_kernel_irqchip_allowed(void) |
3632 | { | |
11bc4a13 | 3633 | return kvm_state->kernel_irqchip_allowed; |
4376c40d PB |
3634 | } |
3635 | ||
3636 | bool kvm_kernel_irqchip_required(void) | |
3637 | { | |
11bc4a13 | 3638 | return kvm_state->kernel_irqchip_required; |
4376c40d PB |
3639 | } |
3640 | ||
3641 | bool kvm_kernel_irqchip_split(void) | |
3642 | { | |
d1972be1 | 3643 | return kvm_state->kernel_irqchip_split == ON_OFF_AUTO_ON; |
4376c40d PB |
3644 | } |
3645 | ||
2ea5cb0a PX |
3646 | static void kvm_get_dirty_ring_size(Object *obj, Visitor *v, |
3647 | const char *name, void *opaque, | |
3648 | Error **errp) | |
3649 | { | |
3650 | KVMState *s = KVM_STATE(obj); | |
3651 | uint32_t value = s->kvm_dirty_ring_size; | |
3652 | ||
3653 | visit_type_uint32(v, name, &value, errp); | |
3654 | } | |
3655 | ||
3656 | static void kvm_set_dirty_ring_size(Object *obj, Visitor *v, | |
3657 | const char *name, void *opaque, | |
3658 | Error **errp) | |
3659 | { | |
3660 | KVMState *s = KVM_STATE(obj); | |
3661 | Error *error = NULL; | |
3662 | uint32_t value; | |
3663 | ||
3664 | if (s->fd != -1) { | |
3665 | error_setg(errp, "Cannot set properties after the accelerator has been initialized"); | |
3666 | return; | |
3667 | } | |
3668 | ||
3669 | visit_type_uint32(v, name, &value, &error); | |
3670 | if (error) { | |
3671 | error_propagate(errp, error); | |
3672 | return; | |
3673 | } | |
3674 | if (value & (value - 1)) { | |
3675 | error_setg(errp, "dirty-ring-size must be a power of two."); | |
3676 | return; | |
3677 | } | |
3678 | ||
3679 | s->kvm_dirty_ring_size = value; | |
3680 | } | |
3681 | ||
23b0898e PB |
3682 | static void kvm_accel_instance_init(Object *obj) |
3683 | { | |
3684 | KVMState *s = KVM_STATE(obj); | |
3685 | ||
70cbae42 PB |
3686 | s->fd = -1; |
3687 | s->vmfd = -1; | |
23b0898e | 3688 | s->kvm_shadow_mem = -1; |
d1972be1 XL |
3689 | s->kernel_irqchip_allowed = true; |
3690 | s->kernel_irqchip_split = ON_OFF_AUTO_AUTO; | |
2ea5cb0a PX |
3691 | /* KVM dirty ring is by default off */ |
3692 | s->kvm_dirty_ring_size = 0; | |
23b0898e PB |
3693 | } |
3694 | ||
782c3f29 EH |
3695 | static void kvm_accel_class_init(ObjectClass *oc, void *data) |
3696 | { | |
3697 | AccelClass *ac = ACCEL_CLASS(oc); | |
3698 | ac->name = "KVM"; | |
0d15da8e | 3699 | ac->init_machine = kvm_init; |
8072aae3 | 3700 | ac->has_memory = kvm_accel_has_memory; |
782c3f29 | 3701 | ac->allowed = &kvm_allowed; |
23b0898e | 3702 | |
11bc4a13 PB |
3703 | object_class_property_add(oc, "kernel-irqchip", "on|off|split", |
3704 | NULL, kvm_set_kernel_irqchip, | |
d2623129 | 3705 | NULL, NULL); |
11bc4a13 | 3706 | object_class_property_set_description(oc, "kernel-irqchip", |
7eecec7d | 3707 | "Configure KVM in-kernel irqchip"); |
11bc4a13 | 3708 | |
23b0898e PB |
3709 | object_class_property_add(oc, "kvm-shadow-mem", "int", |
3710 | kvm_get_kvm_shadow_mem, kvm_set_kvm_shadow_mem, | |
d2623129 | 3711 | NULL, NULL); |
23b0898e | 3712 | object_class_property_set_description(oc, "kvm-shadow-mem", |
7eecec7d | 3713 | "KVM shadow MMU size"); |
2ea5cb0a PX |
3714 | |
3715 | object_class_property_add(oc, "dirty-ring-size", "uint32", | |
3716 | kvm_get_dirty_ring_size, kvm_set_dirty_ring_size, | |
3717 | NULL, NULL); | |
3718 | object_class_property_set_description(oc, "dirty-ring-size", | |
3719 | "Size of KVM dirty page ring buffer (default: 0, i.e. use bitmap)"); | |
782c3f29 EH |
3720 | } |
3721 | ||
3722 | static const TypeInfo kvm_accel_type = { | |
3723 | .name = TYPE_KVM_ACCEL, | |
3724 | .parent = TYPE_ACCEL, | |
23b0898e | 3725 | .instance_init = kvm_accel_instance_init, |
782c3f29 | 3726 | .class_init = kvm_accel_class_init, |
fc02086b | 3727 | .instance_size = sizeof(KVMState), |
782c3f29 EH |
3728 | }; |
3729 | ||
3730 | static void kvm_type_init(void) | |
3731 | { | |
3732 | type_register_static(&kvm_accel_type); | |
3733 | } | |
3734 | ||
3735 | type_init(kvm_type_init); | |
cc01a3f4 MK |
3736 | |
3737 | typedef struct StatsArgs { | |
3738 | union StatsResultsType { | |
3739 | StatsResultList **stats; | |
3740 | StatsSchemaList **schema; | |
3741 | } result; | |
cf7405bc | 3742 | strList *names; |
cc01a3f4 MK |
3743 | Error **errp; |
3744 | } StatsArgs; | |
3745 | ||
3746 | static StatsList *add_kvmstat_entry(struct kvm_stats_desc *pdesc, | |
3747 | uint64_t *stats_data, | |
3748 | StatsList *stats_list, | |
3749 | Error **errp) | |
3750 | { | |
3751 | ||
3752 | Stats *stats; | |
3753 | uint64List *val_list = NULL; | |
3754 | ||
3755 | /* Only add stats that we understand. */ | |
3756 | switch (pdesc->flags & KVM_STATS_TYPE_MASK) { | |
3757 | case KVM_STATS_TYPE_CUMULATIVE: | |
3758 | case KVM_STATS_TYPE_INSTANT: | |
3759 | case KVM_STATS_TYPE_PEAK: | |
3760 | case KVM_STATS_TYPE_LINEAR_HIST: | |
3761 | case KVM_STATS_TYPE_LOG_HIST: | |
3762 | break; | |
3763 | default: | |
3764 | return stats_list; | |
3765 | } | |
3766 | ||
3767 | switch (pdesc->flags & KVM_STATS_UNIT_MASK) { | |
3768 | case KVM_STATS_UNIT_NONE: | |
3769 | case KVM_STATS_UNIT_BYTES: | |
3770 | case KVM_STATS_UNIT_CYCLES: | |
3771 | case KVM_STATS_UNIT_SECONDS: | |
105bb7cd | 3772 | case KVM_STATS_UNIT_BOOLEAN: |
cc01a3f4 MK |
3773 | break; |
3774 | default: | |
3775 | return stats_list; | |
3776 | } | |
3777 | ||
3778 | switch (pdesc->flags & KVM_STATS_BASE_MASK) { | |
3779 | case KVM_STATS_BASE_POW10: | |
3780 | case KVM_STATS_BASE_POW2: | |
3781 | break; | |
3782 | default: | |
3783 | return stats_list; | |
3784 | } | |
3785 | ||
3786 | /* Alloc and populate data list */ | |
3787 | stats = g_new0(Stats, 1); | |
3788 | stats->name = g_strdup(pdesc->name); | |
3789 | stats->value = g_new0(StatsValue, 1);; | |
3790 | ||
105bb7cd PB |
3791 | if ((pdesc->flags & KVM_STATS_UNIT_MASK) == KVM_STATS_UNIT_BOOLEAN) { |
3792 | stats->value->u.boolean = *stats_data; | |
3793 | stats->value->type = QTYPE_QBOOL; | |
3794 | } else if (pdesc->size == 1) { | |
cc01a3f4 MK |
3795 | stats->value->u.scalar = *stats_data; |
3796 | stats->value->type = QTYPE_QNUM; | |
3797 | } else { | |
3798 | int i; | |
3799 | for (i = 0; i < pdesc->size; i++) { | |
3800 | QAPI_LIST_PREPEND(val_list, stats_data[i]); | |
3801 | } | |
3802 | stats->value->u.list = val_list; | |
3803 | stats->value->type = QTYPE_QLIST; | |
3804 | } | |
3805 | ||
3806 | QAPI_LIST_PREPEND(stats_list, stats); | |
3807 | return stats_list; | |
3808 | } | |
3809 | ||
3810 | static StatsSchemaValueList *add_kvmschema_entry(struct kvm_stats_desc *pdesc, | |
3811 | StatsSchemaValueList *list, | |
3812 | Error **errp) | |
3813 | { | |
3814 | StatsSchemaValueList *schema_entry = g_new0(StatsSchemaValueList, 1); | |
3815 | schema_entry->value = g_new0(StatsSchemaValue, 1); | |
3816 | ||
3817 | switch (pdesc->flags & KVM_STATS_TYPE_MASK) { | |
3818 | case KVM_STATS_TYPE_CUMULATIVE: | |
3819 | schema_entry->value->type = STATS_TYPE_CUMULATIVE; | |
3820 | break; | |
3821 | case KVM_STATS_TYPE_INSTANT: | |
3822 | schema_entry->value->type = STATS_TYPE_INSTANT; | |
3823 | break; | |
3824 | case KVM_STATS_TYPE_PEAK: | |
3825 | schema_entry->value->type = STATS_TYPE_PEAK; | |
3826 | break; | |
3827 | case KVM_STATS_TYPE_LINEAR_HIST: | |
3828 | schema_entry->value->type = STATS_TYPE_LINEAR_HISTOGRAM; | |
3829 | schema_entry->value->bucket_size = pdesc->bucket_size; | |
3830 | schema_entry->value->has_bucket_size = true; | |
3831 | break; | |
3832 | case KVM_STATS_TYPE_LOG_HIST: | |
3833 | schema_entry->value->type = STATS_TYPE_LOG2_HISTOGRAM; | |
3834 | break; | |
3835 | default: | |
3836 | goto exit; | |
3837 | } | |
3838 | ||
3839 | switch (pdesc->flags & KVM_STATS_UNIT_MASK) { | |
3840 | case KVM_STATS_UNIT_NONE: | |
3841 | break; | |
105bb7cd PB |
3842 | case KVM_STATS_UNIT_BOOLEAN: |
3843 | schema_entry->value->has_unit = true; | |
3844 | schema_entry->value->unit = STATS_UNIT_BOOLEAN; | |
3845 | break; | |
cc01a3f4 MK |
3846 | case KVM_STATS_UNIT_BYTES: |
3847 | schema_entry->value->has_unit = true; | |
3848 | schema_entry->value->unit = STATS_UNIT_BYTES; | |
3849 | break; | |
3850 | case KVM_STATS_UNIT_CYCLES: | |
3851 | schema_entry->value->has_unit = true; | |
3852 | schema_entry->value->unit = STATS_UNIT_CYCLES; | |
3853 | break; | |
3854 | case KVM_STATS_UNIT_SECONDS: | |
3855 | schema_entry->value->has_unit = true; | |
3856 | schema_entry->value->unit = STATS_UNIT_SECONDS; | |
3857 | break; | |
3858 | default: | |
3859 | goto exit; | |
3860 | } | |
3861 | ||
3862 | schema_entry->value->exponent = pdesc->exponent; | |
3863 | if (pdesc->exponent) { | |
3864 | switch (pdesc->flags & KVM_STATS_BASE_MASK) { | |
3865 | case KVM_STATS_BASE_POW10: | |
3866 | schema_entry->value->has_base = true; | |
3867 | schema_entry->value->base = 10; | |
3868 | break; | |
3869 | case KVM_STATS_BASE_POW2: | |
3870 | schema_entry->value->has_base = true; | |
3871 | schema_entry->value->base = 2; | |
3872 | break; | |
3873 | default: | |
3874 | goto exit; | |
3875 | } | |
3876 | } | |
3877 | ||
3878 | schema_entry->value->name = g_strdup(pdesc->name); | |
3879 | schema_entry->next = list; | |
3880 | return schema_entry; | |
3881 | exit: | |
3882 | g_free(schema_entry->value); | |
3883 | g_free(schema_entry); | |
3884 | return list; | |
3885 | } | |
3886 | ||
3887 | /* Cached stats descriptors */ | |
3888 | typedef struct StatsDescriptors { | |
3889 | const char *ident; /* cache key, currently the StatsTarget */ | |
3890 | struct kvm_stats_desc *kvm_stats_desc; | |
3891 | struct kvm_stats_header *kvm_stats_header; | |
3892 | QTAILQ_ENTRY(StatsDescriptors) next; | |
3893 | } StatsDescriptors; | |
3894 | ||
3895 | static QTAILQ_HEAD(, StatsDescriptors) stats_descriptors = | |
3896 | QTAILQ_HEAD_INITIALIZER(stats_descriptors); | |
3897 | ||
3898 | /* | |
3899 | * Return the descriptors for 'target', that either have already been read | |
3900 | * or are retrieved from 'stats_fd'. | |
3901 | */ | |
3902 | static StatsDescriptors *find_stats_descriptors(StatsTarget target, int stats_fd, | |
3903 | Error **errp) | |
3904 | { | |
3905 | StatsDescriptors *descriptors; | |
3906 | const char *ident; | |
3907 | struct kvm_stats_desc *kvm_stats_desc; | |
3908 | struct kvm_stats_header *kvm_stats_header; | |
3909 | size_t size_desc; | |
3910 | ssize_t ret; | |
3911 | ||
3912 | ident = StatsTarget_str(target); | |
3913 | QTAILQ_FOREACH(descriptors, &stats_descriptors, next) { | |
3914 | if (g_str_equal(descriptors->ident, ident)) { | |
3915 | return descriptors; | |
3916 | } | |
3917 | } | |
3918 | ||
3919 | descriptors = g_new0(StatsDescriptors, 1); | |
3920 | ||
3921 | /* Read stats header */ | |
3922 | kvm_stats_header = g_malloc(sizeof(*kvm_stats_header)); | |
3923 | ret = read(stats_fd, kvm_stats_header, sizeof(*kvm_stats_header)); | |
3924 | if (ret != sizeof(*kvm_stats_header)) { | |
3925 | error_setg(errp, "KVM stats: failed to read stats header: " | |
3926 | "expected %zu actual %zu", | |
3927 | sizeof(*kvm_stats_header), ret); | |
f696b74b | 3928 | g_free(descriptors); |
cc01a3f4 MK |
3929 | return NULL; |
3930 | } | |
3931 | size_desc = sizeof(*kvm_stats_desc) + kvm_stats_header->name_size; | |
3932 | ||
3933 | /* Read stats descriptors */ | |
3934 | kvm_stats_desc = g_malloc0_n(kvm_stats_header->num_desc, size_desc); | |
3935 | ret = pread(stats_fd, kvm_stats_desc, | |
3936 | size_desc * kvm_stats_header->num_desc, | |
3937 | kvm_stats_header->desc_offset); | |
3938 | ||
3939 | if (ret != size_desc * kvm_stats_header->num_desc) { | |
3940 | error_setg(errp, "KVM stats: failed to read stats descriptors: " | |
3941 | "expected %zu actual %zu", | |
3942 | size_desc * kvm_stats_header->num_desc, ret); | |
3943 | g_free(descriptors); | |
3944 | g_free(kvm_stats_desc); | |
3945 | return NULL; | |
3946 | } | |
3947 | descriptors->kvm_stats_header = kvm_stats_header; | |
3948 | descriptors->kvm_stats_desc = kvm_stats_desc; | |
3949 | descriptors->ident = ident; | |
3950 | QTAILQ_INSERT_TAIL(&stats_descriptors, descriptors, next); | |
3951 | return descriptors; | |
3952 | } | |
3953 | ||
3954 | static void query_stats(StatsResultList **result, StatsTarget target, | |
cf7405bc | 3955 | strList *names, int stats_fd, Error **errp) |
cc01a3f4 MK |
3956 | { |
3957 | struct kvm_stats_desc *kvm_stats_desc; | |
3958 | struct kvm_stats_header *kvm_stats_header; | |
3959 | StatsDescriptors *descriptors; | |
3960 | g_autofree uint64_t *stats_data = NULL; | |
3961 | struct kvm_stats_desc *pdesc; | |
3962 | StatsList *stats_list = NULL; | |
3963 | size_t size_desc, size_data = 0; | |
3964 | ssize_t ret; | |
3965 | int i; | |
3966 | ||
3967 | descriptors = find_stats_descriptors(target, stats_fd, errp); | |
3968 | if (!descriptors) { | |
3969 | return; | |
3970 | } | |
3971 | ||
3972 | kvm_stats_header = descriptors->kvm_stats_header; | |
3973 | kvm_stats_desc = descriptors->kvm_stats_desc; | |
3974 | size_desc = sizeof(*kvm_stats_desc) + kvm_stats_header->name_size; | |
3975 | ||
3976 | /* Tally the total data size; read schema data */ | |
3977 | for (i = 0; i < kvm_stats_header->num_desc; ++i) { | |
3978 | pdesc = (void *)kvm_stats_desc + i * size_desc; | |
3979 | size_data += pdesc->size * sizeof(*stats_data); | |
3980 | } | |
3981 | ||
3982 | stats_data = g_malloc0(size_data); | |
3983 | ret = pread(stats_fd, stats_data, size_data, kvm_stats_header->data_offset); | |
3984 | ||
3985 | if (ret != size_data) { | |
3986 | error_setg(errp, "KVM stats: failed to read data: " | |
3987 | "expected %zu actual %zu", size_data, ret); | |
3988 | return; | |
3989 | } | |
3990 | ||
3991 | for (i = 0; i < kvm_stats_header->num_desc; ++i) { | |
3992 | uint64_t *stats; | |
3993 | pdesc = (void *)kvm_stats_desc + i * size_desc; | |
3994 | ||
3995 | /* Add entry to the list */ | |
3996 | stats = (void *)stats_data + pdesc->offset; | |
cf7405bc PB |
3997 | if (!apply_str_list_filter(pdesc->name, names)) { |
3998 | continue; | |
3999 | } | |
cc01a3f4 MK |
4000 | stats_list = add_kvmstat_entry(pdesc, stats, stats_list, errp); |
4001 | } | |
4002 | ||
4003 | if (!stats_list) { | |
4004 | return; | |
4005 | } | |
4006 | ||
4007 | switch (target) { | |
4008 | case STATS_TARGET_VM: | |
4009 | add_stats_entry(result, STATS_PROVIDER_KVM, NULL, stats_list); | |
4010 | break; | |
4011 | case STATS_TARGET_VCPU: | |
4012 | add_stats_entry(result, STATS_PROVIDER_KVM, | |
4013 | current_cpu->parent_obj.canonical_path, | |
4014 | stats_list); | |
4015 | break; | |
4016 | default: | |
d12dd9c7 | 4017 | g_assert_not_reached(); |
cc01a3f4 MK |
4018 | } |
4019 | } | |
4020 | ||
4021 | static void query_stats_schema(StatsSchemaList **result, StatsTarget target, | |
4022 | int stats_fd, Error **errp) | |
4023 | { | |
4024 | struct kvm_stats_desc *kvm_stats_desc; | |
4025 | struct kvm_stats_header *kvm_stats_header; | |
4026 | StatsDescriptors *descriptors; | |
4027 | struct kvm_stats_desc *pdesc; | |
4028 | StatsSchemaValueList *stats_list = NULL; | |
4029 | size_t size_desc; | |
4030 | int i; | |
4031 | ||
4032 | descriptors = find_stats_descriptors(target, stats_fd, errp); | |
4033 | if (!descriptors) { | |
4034 | return; | |
4035 | } | |
4036 | ||
4037 | kvm_stats_header = descriptors->kvm_stats_header; | |
4038 | kvm_stats_desc = descriptors->kvm_stats_desc; | |
4039 | size_desc = sizeof(*kvm_stats_desc) + kvm_stats_header->name_size; | |
4040 | ||
4041 | /* Tally the total data size; read schema data */ | |
4042 | for (i = 0; i < kvm_stats_header->num_desc; ++i) { | |
4043 | pdesc = (void *)kvm_stats_desc + i * size_desc; | |
4044 | stats_list = add_kvmschema_entry(pdesc, stats_list, errp); | |
4045 | } | |
4046 | ||
4047 | add_stats_schema(result, STATS_PROVIDER_KVM, target, stats_list); | |
4048 | } | |
4049 | ||
4050 | static void query_stats_vcpu(CPUState *cpu, run_on_cpu_data data) | |
4051 | { | |
4052 | StatsArgs *kvm_stats_args = (StatsArgs *) data.host_ptr; | |
4053 | int stats_fd = kvm_vcpu_ioctl(cpu, KVM_GET_STATS_FD, NULL); | |
4054 | Error *local_err = NULL; | |
4055 | ||
4056 | if (stats_fd == -1) { | |
4057 | error_setg_errno(&local_err, errno, "KVM stats: ioctl failed"); | |
4058 | error_propagate(kvm_stats_args->errp, local_err); | |
4059 | return; | |
4060 | } | |
cf7405bc PB |
4061 | query_stats(kvm_stats_args->result.stats, STATS_TARGET_VCPU, |
4062 | kvm_stats_args->names, stats_fd, kvm_stats_args->errp); | |
cc01a3f4 MK |
4063 | close(stats_fd); |
4064 | } | |
4065 | ||
4066 | static void query_stats_schema_vcpu(CPUState *cpu, run_on_cpu_data data) | |
4067 | { | |
4068 | StatsArgs *kvm_stats_args = (StatsArgs *) data.host_ptr; | |
4069 | int stats_fd = kvm_vcpu_ioctl(cpu, KVM_GET_STATS_FD, NULL); | |
4070 | Error *local_err = NULL; | |
4071 | ||
4072 | if (stats_fd == -1) { | |
4073 | error_setg_errno(&local_err, errno, "KVM stats: ioctl failed"); | |
4074 | error_propagate(kvm_stats_args->errp, local_err); | |
4075 | return; | |
4076 | } | |
4077 | query_stats_schema(kvm_stats_args->result.schema, STATS_TARGET_VCPU, stats_fd, | |
4078 | kvm_stats_args->errp); | |
4079 | close(stats_fd); | |
4080 | } | |
4081 | ||
467ef823 | 4082 | static void query_stats_cb(StatsResultList **result, StatsTarget target, |
cf7405bc | 4083 | strList *names, strList *targets, Error **errp) |
cc01a3f4 MK |
4084 | { |
4085 | KVMState *s = kvm_state; | |
4086 | CPUState *cpu; | |
4087 | int stats_fd; | |
4088 | ||
4089 | switch (target) { | |
4090 | case STATS_TARGET_VM: | |
4091 | { | |
4092 | stats_fd = kvm_vm_ioctl(s, KVM_GET_STATS_FD, NULL); | |
4093 | if (stats_fd == -1) { | |
4094 | error_setg_errno(errp, errno, "KVM stats: ioctl failed"); | |
4095 | return; | |
4096 | } | |
cf7405bc | 4097 | query_stats(result, target, names, stats_fd, errp); |
cc01a3f4 MK |
4098 | close(stats_fd); |
4099 | break; | |
4100 | } | |
4101 | case STATS_TARGET_VCPU: | |
4102 | { | |
4103 | StatsArgs stats_args; | |
4104 | stats_args.result.stats = result; | |
cf7405bc | 4105 | stats_args.names = names; |
cc01a3f4 MK |
4106 | stats_args.errp = errp; |
4107 | CPU_FOREACH(cpu) { | |
467ef823 PB |
4108 | if (!apply_str_list_filter(cpu->parent_obj.canonical_path, targets)) { |
4109 | continue; | |
4110 | } | |
cc01a3f4 MK |
4111 | run_on_cpu(cpu, query_stats_vcpu, RUN_ON_CPU_HOST_PTR(&stats_args)); |
4112 | } | |
4113 | break; | |
4114 | } | |
4115 | default: | |
4116 | break; | |
4117 | } | |
4118 | } | |
4119 | ||
4120 | void query_stats_schemas_cb(StatsSchemaList **result, Error **errp) | |
4121 | { | |
4122 | StatsArgs stats_args; | |
4123 | KVMState *s = kvm_state; | |
4124 | int stats_fd; | |
4125 | ||
4126 | stats_fd = kvm_vm_ioctl(s, KVM_GET_STATS_FD, NULL); | |
4127 | if (stats_fd == -1) { | |
4128 | error_setg_errno(errp, errno, "KVM stats: ioctl failed"); | |
4129 | return; | |
4130 | } | |
4131 | query_stats_schema(result, STATS_TARGET_VM, stats_fd, errp); | |
4132 | close(stats_fd); | |
4133 | ||
4134 | stats_args.result.schema = result; | |
4135 | stats_args.errp = errp; | |
4136 | run_on_cpu(first_cpu, query_stats_schema_vcpu, RUN_ON_CPU_HOST_PTR(&stats_args)); | |
4137 | } |