]>
Commit | Line | Data |
---|---|---|
05330448 AL |
1 | /* |
2 | * QEMU KVM support | |
3 | * | |
4 | * Copyright IBM, Corp. 2008 | |
5832d1f2 | 5 | * Red Hat, Inc. 2008 |
05330448 AL |
6 | * |
7 | * Authors: | |
8 | * Anthony Liguori <aliguori@us.ibm.com> | |
5832d1f2 | 9 | * Glauber Costa <gcosta@redhat.com> |
05330448 AL |
10 | * |
11 | * This work is licensed under the terms of the GNU GPL, version 2 or later. | |
12 | * See the COPYING file in the top-level directory. | |
13 | * | |
14 | */ | |
15 | ||
d38ea87a | 16 | #include "qemu/osdep.h" |
05330448 | 17 | #include <sys/ioctl.h> |
05330448 AL |
18 | |
19 | #include <linux/kvm.h> | |
20 | ||
21 | #include "qemu-common.h" | |
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" |
d33a1810 | 27 | #include "hw/hw.h" |
a2cb15b0 | 28 | #include "hw/pci/msi.h" |
d1f6af6a | 29 | #include "hw/pci/msix.h" |
d426d9fb | 30 | #include "hw/s390x/adapter.h" |
022c62cb | 31 | #include "exec/gdbstub.h" |
8571ed35 | 32 | #include "sysemu/kvm_int.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" |
022c62cb | 37 | #include "exec/address-spaces.h" |
1de7afc9 | 38 | #include "qemu/event_notifier.h" |
0ab8ed18 | 39 | #include "trace-root.h" |
197e3524 | 40 | #include "hw/irq.h" |
05330448 | 41 | |
135a129a AK |
42 | #include "hw/boards.h" |
43 | ||
d2f2b8a7 SH |
44 | /* This check must be after config-host.h is included */ |
45 | #ifdef CONFIG_EVENTFD | |
46 | #include <sys/eventfd.h> | |
47 | #endif | |
48 | ||
bc92e4e9 AJ |
49 | /* KVM uses PAGE_SIZE in its definition of KVM_COALESCED_MMIO_MAX. We |
50 | * need to use the real host PAGE_SIZE, as that's what KVM will use. | |
51 | */ | |
52 | #define PAGE_SIZE getpagesize() | |
f65ed4c1 | 53 | |
05330448 AL |
54 | //#define DEBUG_KVM |
55 | ||
56 | #ifdef DEBUG_KVM | |
8c0d577e | 57 | #define DPRINTF(fmt, ...) \ |
05330448 AL |
58 | do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0) |
59 | #else | |
8c0d577e | 60 | #define DPRINTF(fmt, ...) \ |
05330448 AL |
61 | do { } while (0) |
62 | #endif | |
63 | ||
04fa27f5 JK |
64 | #define KVM_MSI_HASHTAB_SIZE 256 |
65 | ||
4c055ab5 GZ |
66 | struct KVMParkedVcpu { |
67 | unsigned long vcpu_id; | |
68 | int kvm_fd; | |
69 | QLIST_ENTRY(KVMParkedVcpu) node; | |
70 | }; | |
71 | ||
9d1c35df | 72 | struct KVMState |
05330448 | 73 | { |
fc02086b EH |
74 | AccelState parent_obj; |
75 | ||
fb541ca5 | 76 | int nr_slots; |
05330448 AL |
77 | int fd; |
78 | int vmfd; | |
f65ed4c1 | 79 | int coalesced_mmio; |
62a2744c | 80 | struct kvm_coalesced_mmio_ring *coalesced_mmio_ring; |
1cae88b9 | 81 | bool coalesced_flush_in_progress; |
e69917e2 | 82 | int broken_set_mem_region; |
a0fb002c | 83 | int vcpu_events; |
b0b1d690 | 84 | int robust_singlestep; |
ff44f1a3 | 85 | int debugregs; |
e22a25c9 AL |
86 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
87 | struct kvm_sw_breakpoint_head kvm_sw_breakpoints; | |
88 | #endif | |
d2f2b8a7 | 89 | int many_ioeventfds; |
3ab73842 | 90 | int intx_set_mask; |
92e4b519 DG |
91 | /* The man page (and posix) say ioctl numbers are signed int, but |
92 | * they're not. Linux, glibc and *BSD all treat ioctl numbers as | |
93 | * unsigned, and treating them as signed here can break things */ | |
e333cd69 | 94 | unsigned irq_set_ioctl; |
aed6efb9 | 95 | unsigned int sigmask_len; |
197e3524 | 96 | GHashTable *gsimap; |
84b058d7 JK |
97 | #ifdef KVM_CAP_IRQ_ROUTING |
98 | struct kvm_irq_routing *irq_routes; | |
99 | int nr_allocated_irq_routes; | |
8269fb70 | 100 | unsigned long *used_gsi_bitmap; |
4e2e4e63 | 101 | unsigned int gsi_count; |
04fa27f5 | 102 | QTAILQ_HEAD(msi_hashtab, KVMMSIRoute) msi_hashtab[KVM_MSI_HASHTAB_SIZE]; |
84b058d7 | 103 | #endif |
7bbda04c | 104 | KVMMemoryListener memory_listener; |
4c055ab5 | 105 | QLIST_HEAD(, KVMParkedVcpu) kvm_parked_vcpus; |
9d1c35df | 106 | }; |
05330448 | 107 | |
6a7af8cb | 108 | KVMState *kvm_state; |
3d4b2649 | 109 | bool kvm_kernel_irqchip; |
15eafc2e | 110 | bool kvm_split_irqchip; |
7ae26bd4 | 111 | bool kvm_async_interrupts_allowed; |
215e79c0 | 112 | bool kvm_halt_in_kernel_allowed; |
69e03ae6 | 113 | bool kvm_eventfds_allowed; |
cc7e0ddf | 114 | bool kvm_irqfds_allowed; |
f41389ae | 115 | bool kvm_resamplefds_allowed; |
614e41bc | 116 | bool kvm_msi_via_irqfd_allowed; |
f3e1bed8 | 117 | bool kvm_gsi_routing_allowed; |
76fe21de | 118 | bool kvm_gsi_direct_mapping; |
13eed94e | 119 | bool kvm_allowed; |
df9c8b75 | 120 | bool kvm_readonly_mem_allowed; |
d0a073a1 | 121 | bool kvm_vm_attributes_allowed; |
50bf31b9 | 122 | bool kvm_direct_msi_allowed; |
35108223 | 123 | bool kvm_ioeventfd_any_length_allowed; |
767a554a | 124 | bool kvm_msi_use_devid; |
cf0f7cf9 | 125 | static bool kvm_immediate_exit; |
05330448 | 126 | |
94a8d39a JK |
127 | static const KVMCapabilityInfo kvm_required_capabilites[] = { |
128 | KVM_CAP_INFO(USER_MEMORY), | |
129 | KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS), | |
130 | KVM_CAP_LAST_INFO | |
131 | }; | |
132 | ||
44f2e6c1 BR |
133 | int kvm_get_max_memslots(void) |
134 | { | |
135 | KVMState *s = KVM_STATE(current_machine->accelerator); | |
136 | ||
137 | return s->nr_slots; | |
138 | } | |
139 | ||
7bbda04c | 140 | static KVMSlot *kvm_get_free_slot(KVMMemoryListener *kml) |
05330448 | 141 | { |
7bbda04c | 142 | KVMState *s = kvm_state; |
05330448 AL |
143 | int i; |
144 | ||
fb541ca5 | 145 | for (i = 0; i < s->nr_slots; i++) { |
7bbda04c PB |
146 | if (kml->slots[i].memory_size == 0) { |
147 | return &kml->slots[i]; | |
a426e122 | 148 | } |
05330448 AL |
149 | } |
150 | ||
b8865591 IM |
151 | return NULL; |
152 | } | |
153 | ||
154 | bool kvm_has_free_slot(MachineState *ms) | |
155 | { | |
7bbda04c PB |
156 | KVMState *s = KVM_STATE(ms->accelerator); |
157 | ||
158 | return kvm_get_free_slot(&s->memory_listener); | |
b8865591 IM |
159 | } |
160 | ||
7bbda04c | 161 | static KVMSlot *kvm_alloc_slot(KVMMemoryListener *kml) |
b8865591 | 162 | { |
7bbda04c | 163 | KVMSlot *slot = kvm_get_free_slot(kml); |
b8865591 IM |
164 | |
165 | if (slot) { | |
166 | return slot; | |
167 | } | |
168 | ||
d3f8d37f AL |
169 | fprintf(stderr, "%s: no free slot available\n", __func__); |
170 | abort(); | |
171 | } | |
172 | ||
7bbda04c | 173 | static KVMSlot *kvm_lookup_matching_slot(KVMMemoryListener *kml, |
a8170e5e AK |
174 | hwaddr start_addr, |
175 | hwaddr end_addr) | |
d3f8d37f | 176 | { |
7bbda04c | 177 | KVMState *s = kvm_state; |
d3f8d37f AL |
178 | int i; |
179 | ||
fb541ca5 | 180 | for (i = 0; i < s->nr_slots; i++) { |
7bbda04c | 181 | KVMSlot *mem = &kml->slots[i]; |
d3f8d37f AL |
182 | |
183 | if (start_addr == mem->start_addr && | |
184 | end_addr == mem->start_addr + mem->memory_size) { | |
185 | return mem; | |
186 | } | |
187 | } | |
188 | ||
05330448 AL |
189 | return NULL; |
190 | } | |
191 | ||
6152e2ae AL |
192 | /* |
193 | * Find overlapping slot with lowest start address | |
194 | */ | |
7bbda04c | 195 | static KVMSlot *kvm_lookup_overlapping_slot(KVMMemoryListener *kml, |
a8170e5e AK |
196 | hwaddr start_addr, |
197 | hwaddr end_addr) | |
05330448 | 198 | { |
7bbda04c | 199 | KVMState *s = kvm_state; |
6152e2ae | 200 | KVMSlot *found = NULL; |
05330448 AL |
201 | int i; |
202 | ||
fb541ca5 | 203 | for (i = 0; i < s->nr_slots; i++) { |
7bbda04c | 204 | KVMSlot *mem = &kml->slots[i]; |
05330448 | 205 | |
6152e2ae AL |
206 | if (mem->memory_size == 0 || |
207 | (found && found->start_addr < mem->start_addr)) { | |
208 | continue; | |
209 | } | |
210 | ||
211 | if (end_addr > mem->start_addr && | |
212 | start_addr < mem->start_addr + mem->memory_size) { | |
213 | found = mem; | |
214 | } | |
05330448 AL |
215 | } |
216 | ||
6152e2ae | 217 | return found; |
05330448 AL |
218 | } |
219 | ||
9f213ed9 | 220 | int kvm_physical_memory_addr_from_host(KVMState *s, void *ram, |
a8170e5e | 221 | hwaddr *phys_addr) |
983dfc3b | 222 | { |
7bbda04c | 223 | KVMMemoryListener *kml = &s->memory_listener; |
983dfc3b HY |
224 | int i; |
225 | ||
fb541ca5 | 226 | for (i = 0; i < s->nr_slots; i++) { |
7bbda04c | 227 | KVMSlot *mem = &kml->slots[i]; |
983dfc3b | 228 | |
9f213ed9 AK |
229 | if (ram >= mem->ram && ram < mem->ram + mem->memory_size) { |
230 | *phys_addr = mem->start_addr + (ram - mem->ram); | |
983dfc3b HY |
231 | return 1; |
232 | } | |
233 | } | |
234 | ||
235 | return 0; | |
236 | } | |
237 | ||
7bbda04c | 238 | static int kvm_set_user_memory_region(KVMMemoryListener *kml, KVMSlot *slot) |
5832d1f2 | 239 | { |
7bbda04c | 240 | KVMState *s = kvm_state; |
5832d1f2 AL |
241 | struct kvm_userspace_memory_region mem; |
242 | ||
38bfe691 | 243 | mem.slot = slot->slot | (kml->as_id << 16); |
5832d1f2 | 244 | mem.guest_phys_addr = slot->start_addr; |
9f213ed9 | 245 | mem.userspace_addr = (unsigned long)slot->ram; |
5832d1f2 | 246 | mem.flags = slot->flags; |
651eb0f4 XG |
247 | |
248 | if (slot->memory_size && mem.flags & KVM_MEM_READONLY) { | |
235e8982 JJ |
249 | /* Set the slot size to 0 before setting the slot to the desired |
250 | * value. This is needed based on KVM commit 75d61fbc. */ | |
251 | mem.memory_size = 0; | |
252 | kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem); | |
253 | } | |
254 | mem.memory_size = slot->memory_size; | |
5832d1f2 AL |
255 | return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem); |
256 | } | |
257 | ||
4c055ab5 GZ |
258 | int kvm_destroy_vcpu(CPUState *cpu) |
259 | { | |
260 | KVMState *s = kvm_state; | |
261 | long mmap_size; | |
262 | struct KVMParkedVcpu *vcpu = NULL; | |
263 | int ret = 0; | |
264 | ||
265 | DPRINTF("kvm_destroy_vcpu\n"); | |
266 | ||
267 | mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0); | |
268 | if (mmap_size < 0) { | |
269 | ret = mmap_size; | |
270 | DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n"); | |
271 | goto err; | |
272 | } | |
273 | ||
274 | ret = munmap(cpu->kvm_run, mmap_size); | |
275 | if (ret < 0) { | |
276 | goto err; | |
277 | } | |
278 | ||
279 | vcpu = g_malloc0(sizeof(*vcpu)); | |
280 | vcpu->vcpu_id = kvm_arch_vcpu_id(cpu); | |
281 | vcpu->kvm_fd = cpu->kvm_fd; | |
282 | QLIST_INSERT_HEAD(&kvm_state->kvm_parked_vcpus, vcpu, node); | |
283 | err: | |
284 | return ret; | |
285 | } | |
286 | ||
287 | static int kvm_get_vcpu(KVMState *s, unsigned long vcpu_id) | |
288 | { | |
289 | struct KVMParkedVcpu *cpu; | |
290 | ||
291 | QLIST_FOREACH(cpu, &s->kvm_parked_vcpus, node) { | |
292 | if (cpu->vcpu_id == vcpu_id) { | |
293 | int kvm_fd; | |
294 | ||
295 | QLIST_REMOVE(cpu, node); | |
296 | kvm_fd = cpu->kvm_fd; | |
297 | g_free(cpu); | |
298 | return kvm_fd; | |
299 | } | |
300 | } | |
301 | ||
302 | return kvm_vm_ioctl(s, KVM_CREATE_VCPU, (void *)vcpu_id); | |
303 | } | |
304 | ||
504134d2 | 305 | int kvm_init_vcpu(CPUState *cpu) |
05330448 AL |
306 | { |
307 | KVMState *s = kvm_state; | |
308 | long mmap_size; | |
309 | int ret; | |
310 | ||
8c0d577e | 311 | DPRINTF("kvm_init_vcpu\n"); |
05330448 | 312 | |
4c055ab5 | 313 | ret = kvm_get_vcpu(s, kvm_arch_vcpu_id(cpu)); |
05330448 | 314 | if (ret < 0) { |
8c0d577e | 315 | DPRINTF("kvm_create_vcpu failed\n"); |
05330448 AL |
316 | goto err; |
317 | } | |
318 | ||
8737c51c | 319 | cpu->kvm_fd = ret; |
a60f24b5 | 320 | cpu->kvm_state = s; |
20d695a9 | 321 | cpu->kvm_vcpu_dirty = true; |
05330448 AL |
322 | |
323 | mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0); | |
324 | if (mmap_size < 0) { | |
748a680b | 325 | ret = mmap_size; |
8c0d577e | 326 | DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n"); |
05330448 AL |
327 | goto err; |
328 | } | |
329 | ||
f7575c96 | 330 | cpu->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED, |
8737c51c | 331 | cpu->kvm_fd, 0); |
f7575c96 | 332 | if (cpu->kvm_run == MAP_FAILED) { |
05330448 | 333 | ret = -errno; |
8c0d577e | 334 | DPRINTF("mmap'ing vcpu state failed\n"); |
05330448 AL |
335 | goto err; |
336 | } | |
337 | ||
a426e122 JK |
338 | if (s->coalesced_mmio && !s->coalesced_mmio_ring) { |
339 | s->coalesced_mmio_ring = | |
f7575c96 | 340 | (void *)cpu->kvm_run + s->coalesced_mmio * PAGE_SIZE; |
a426e122 | 341 | } |
62a2744c | 342 | |
20d695a9 | 343 | ret = kvm_arch_init_vcpu(cpu); |
05330448 AL |
344 | err: |
345 | return ret; | |
346 | } | |
347 | ||
5832d1f2 AL |
348 | /* |
349 | * dirty pages logging control | |
350 | */ | |
25254bbc | 351 | |
d6ff5cbc | 352 | static int kvm_mem_flags(MemoryRegion *mr) |
25254bbc | 353 | { |
d6ff5cbc | 354 | bool readonly = mr->readonly || memory_region_is_romd(mr); |
235e8982 | 355 | int flags = 0; |
d6ff5cbc AJ |
356 | |
357 | if (memory_region_get_dirty_log_mask(mr) != 0) { | |
358 | flags |= KVM_MEM_LOG_DIRTY_PAGES; | |
359 | } | |
235e8982 JJ |
360 | if (readonly && kvm_readonly_mem_allowed) { |
361 | flags |= KVM_MEM_READONLY; | |
362 | } | |
363 | return flags; | |
25254bbc MT |
364 | } |
365 | ||
7bbda04c PB |
366 | static int kvm_slot_update_flags(KVMMemoryListener *kml, KVMSlot *mem, |
367 | MemoryRegion *mr) | |
5832d1f2 | 368 | { |
4495d6a7 JK |
369 | int old_flags; |
370 | ||
4495d6a7 | 371 | old_flags = mem->flags; |
d6ff5cbc | 372 | mem->flags = kvm_mem_flags(mr); |
5832d1f2 | 373 | |
4495d6a7 | 374 | /* If nothing changed effectively, no need to issue ioctl */ |
d6ff5cbc | 375 | if (mem->flags == old_flags) { |
25254bbc | 376 | return 0; |
4495d6a7 JK |
377 | } |
378 | ||
7bbda04c | 379 | return kvm_set_user_memory_region(kml, mem); |
5832d1f2 AL |
380 | } |
381 | ||
7bbda04c PB |
382 | static int kvm_section_update_flags(KVMMemoryListener *kml, |
383 | MemoryRegionSection *section) | |
25254bbc | 384 | { |
d6ff5cbc AJ |
385 | hwaddr phys_addr = section->offset_within_address_space; |
386 | ram_addr_t size = int128_get64(section->size); | |
7bbda04c | 387 | KVMSlot *mem = kvm_lookup_matching_slot(kml, phys_addr, phys_addr + size); |
25254bbc MT |
388 | |
389 | if (mem == NULL) { | |
ea8cb1a8 PB |
390 | return 0; |
391 | } else { | |
7bbda04c | 392 | return kvm_slot_update_flags(kml, mem, section->mr); |
25254bbc | 393 | } |
25254bbc MT |
394 | } |
395 | ||
a01672d3 | 396 | static void kvm_log_start(MemoryListener *listener, |
b2dfd71c PB |
397 | MemoryRegionSection *section, |
398 | int old, int new) | |
5832d1f2 | 399 | { |
7bbda04c | 400 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); |
a01672d3 AK |
401 | int r; |
402 | ||
b2dfd71c PB |
403 | if (old != 0) { |
404 | return; | |
405 | } | |
406 | ||
7bbda04c | 407 | r = kvm_section_update_flags(kml, section); |
a01672d3 AK |
408 | if (r < 0) { |
409 | abort(); | |
410 | } | |
5832d1f2 AL |
411 | } |
412 | ||
a01672d3 | 413 | static void kvm_log_stop(MemoryListener *listener, |
b2dfd71c PB |
414 | MemoryRegionSection *section, |
415 | int old, int new) | |
5832d1f2 | 416 | { |
7bbda04c | 417 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); |
a01672d3 AK |
418 | int r; |
419 | ||
b2dfd71c PB |
420 | if (new != 0) { |
421 | return; | |
422 | } | |
423 | ||
7bbda04c | 424 | r = kvm_section_update_flags(kml, section); |
a01672d3 AK |
425 | if (r < 0) { |
426 | abort(); | |
427 | } | |
5832d1f2 AL |
428 | } |
429 | ||
8369e01c | 430 | /* get kvm's dirty pages bitmap and update qemu's */ |
ffcde12f AK |
431 | static int kvm_get_dirty_pages_log_range(MemoryRegionSection *section, |
432 | unsigned long *bitmap) | |
96c1606b | 433 | { |
8e41fb63 FZ |
434 | ram_addr_t start = section->offset_within_region + |
435 | memory_region_get_ram_addr(section->mr); | |
5ff7fb77 JQ |
436 | ram_addr_t pages = int128_get64(section->size) / getpagesize(); |
437 | ||
438 | cpu_physical_memory_set_dirty_lebitmap(bitmap, start, pages); | |
8369e01c | 439 | return 0; |
96c1606b AG |
440 | } |
441 | ||
8369e01c MT |
442 | #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1)) |
443 | ||
5832d1f2 AL |
444 | /** |
445 | * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space | |
fd4aa979 BS |
446 | * This function updates qemu's dirty bitmap using |
447 | * memory_region_set_dirty(). This means all bits are set | |
448 | * to dirty. | |
5832d1f2 | 449 | * |
d3f8d37f | 450 | * @start_add: start of logged region. |
5832d1f2 AL |
451 | * @end_addr: end of logged region. |
452 | */ | |
7bbda04c PB |
453 | static int kvm_physical_sync_dirty_bitmap(KVMMemoryListener *kml, |
454 | MemoryRegionSection *section) | |
5832d1f2 AL |
455 | { |
456 | KVMState *s = kvm_state; | |
151f7749 | 457 | unsigned long size, allocated_size = 0; |
714f78c5 | 458 | struct kvm_dirty_log d = {}; |
151f7749 JK |
459 | KVMSlot *mem; |
460 | int ret = 0; | |
a8170e5e | 461 | hwaddr start_addr = section->offset_within_address_space; |
052e87b0 | 462 | hwaddr end_addr = start_addr + int128_get64(section->size); |
5832d1f2 | 463 | |
151f7749 JK |
464 | d.dirty_bitmap = NULL; |
465 | while (start_addr < end_addr) { | |
7bbda04c | 466 | mem = kvm_lookup_overlapping_slot(kml, start_addr, end_addr); |
151f7749 JK |
467 | if (mem == NULL) { |
468 | break; | |
469 | } | |
5832d1f2 | 470 | |
51b0c606 MT |
471 | /* XXX bad kernel interface alert |
472 | * For dirty bitmap, kernel allocates array of size aligned to | |
473 | * bits-per-long. But for case when the kernel is 64bits and | |
474 | * the userspace is 32bits, userspace can't align to the same | |
475 | * bits-per-long, since sizeof(long) is different between kernel | |
476 | * and user space. This way, userspace will provide buffer which | |
477 | * may be 4 bytes less than the kernel will use, resulting in | |
478 | * userspace memory corruption (which is not detectable by valgrind | |
479 | * too, in most cases). | |
480 | * So for now, let's align to 64 instead of HOST_LONG_BITS here, in | |
cb8d4c8f | 481 | * a hope that sizeof(long) won't become >8 any time soon. |
51b0c606 MT |
482 | */ |
483 | size = ALIGN(((mem->memory_size) >> TARGET_PAGE_BITS), | |
484 | /*HOST_LONG_BITS*/ 64) / 8; | |
151f7749 | 485 | if (!d.dirty_bitmap) { |
7267c094 | 486 | d.dirty_bitmap = g_malloc(size); |
151f7749 | 487 | } else if (size > allocated_size) { |
7267c094 | 488 | d.dirty_bitmap = g_realloc(d.dirty_bitmap, size); |
151f7749 JK |
489 | } |
490 | allocated_size = size; | |
491 | memset(d.dirty_bitmap, 0, allocated_size); | |
5832d1f2 | 492 | |
38bfe691 | 493 | d.slot = mem->slot | (kml->as_id << 16); |
50212d63 | 494 | if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) { |
8c0d577e | 495 | DPRINTF("ioctl failed %d\n", errno); |
151f7749 JK |
496 | ret = -1; |
497 | break; | |
498 | } | |
5832d1f2 | 499 | |
ffcde12f | 500 | kvm_get_dirty_pages_log_range(section, d.dirty_bitmap); |
8369e01c | 501 | start_addr = mem->start_addr + mem->memory_size; |
5832d1f2 | 502 | } |
7267c094 | 503 | g_free(d.dirty_bitmap); |
151f7749 JK |
504 | |
505 | return ret; | |
5832d1f2 AL |
506 | } |
507 | ||
95d2994a AK |
508 | static void kvm_coalesce_mmio_region(MemoryListener *listener, |
509 | MemoryRegionSection *secion, | |
a8170e5e | 510 | hwaddr start, hwaddr size) |
f65ed4c1 | 511 | { |
f65ed4c1 AL |
512 | KVMState *s = kvm_state; |
513 | ||
514 | if (s->coalesced_mmio) { | |
515 | struct kvm_coalesced_mmio_zone zone; | |
516 | ||
517 | zone.addr = start; | |
518 | zone.size = size; | |
7e680753 | 519 | zone.pad = 0; |
f65ed4c1 | 520 | |
95d2994a | 521 | (void)kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone); |
f65ed4c1 | 522 | } |
f65ed4c1 AL |
523 | } |
524 | ||
95d2994a AK |
525 | static void kvm_uncoalesce_mmio_region(MemoryListener *listener, |
526 | MemoryRegionSection *secion, | |
a8170e5e | 527 | hwaddr start, hwaddr size) |
f65ed4c1 | 528 | { |
f65ed4c1 AL |
529 | KVMState *s = kvm_state; |
530 | ||
531 | if (s->coalesced_mmio) { | |
532 | struct kvm_coalesced_mmio_zone zone; | |
533 | ||
534 | zone.addr = start; | |
535 | zone.size = size; | |
7e680753 | 536 | zone.pad = 0; |
f65ed4c1 | 537 | |
95d2994a | 538 | (void)kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone); |
f65ed4c1 | 539 | } |
f65ed4c1 AL |
540 | } |
541 | ||
ad7b8b33 AL |
542 | int kvm_check_extension(KVMState *s, unsigned int extension) |
543 | { | |
544 | int ret; | |
545 | ||
546 | ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension); | |
547 | if (ret < 0) { | |
548 | ret = 0; | |
549 | } | |
550 | ||
551 | return ret; | |
552 | } | |
553 | ||
7d0a07fa AG |
554 | int kvm_vm_check_extension(KVMState *s, unsigned int extension) |
555 | { | |
556 | int ret; | |
557 | ||
558 | ret = kvm_vm_ioctl(s, KVM_CHECK_EXTENSION, extension); | |
559 | if (ret < 0) { | |
560 | /* VM wide version not implemented, use global one instead */ | |
561 | ret = kvm_check_extension(s, extension); | |
562 | } | |
563 | ||
564 | return ret; | |
565 | } | |
566 | ||
b680c5ba GK |
567 | static uint32_t adjust_ioeventfd_endianness(uint32_t val, uint32_t size) |
568 | { | |
569 | #if defined(HOST_WORDS_BIGENDIAN) != defined(TARGET_WORDS_BIGENDIAN) | |
570 | /* The kernel expects ioeventfd values in HOST_WORDS_BIGENDIAN | |
571 | * endianness, but the memory core hands them in target endianness. | |
572 | * For example, PPC is always treated as big-endian even if running | |
573 | * on KVM and on PPC64LE. Correct here. | |
574 | */ | |
575 | switch (size) { | |
576 | case 2: | |
577 | val = bswap16(val); | |
578 | break; | |
579 | case 4: | |
580 | val = bswap32(val); | |
581 | break; | |
582 | } | |
583 | #endif | |
584 | return val; | |
585 | } | |
586 | ||
584f2be7 | 587 | static int kvm_set_ioeventfd_mmio(int fd, hwaddr addr, uint32_t val, |
41cb62c2 | 588 | bool assign, uint32_t size, bool datamatch) |
500ffd4a MT |
589 | { |
590 | int ret; | |
03a96b83 TH |
591 | struct kvm_ioeventfd iofd = { |
592 | .datamatch = datamatch ? adjust_ioeventfd_endianness(val, size) : 0, | |
593 | .addr = addr, | |
594 | .len = size, | |
595 | .flags = 0, | |
596 | .fd = fd, | |
597 | }; | |
500ffd4a MT |
598 | |
599 | if (!kvm_enabled()) { | |
600 | return -ENOSYS; | |
601 | } | |
602 | ||
41cb62c2 MT |
603 | if (datamatch) { |
604 | iofd.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH; | |
605 | } | |
500ffd4a MT |
606 | if (!assign) { |
607 | iofd.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; | |
608 | } | |
609 | ||
610 | ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd); | |
611 | ||
612 | if (ret < 0) { | |
613 | return -errno; | |
614 | } | |
615 | ||
616 | return 0; | |
617 | } | |
618 | ||
44c3f8f7 | 619 | static int kvm_set_ioeventfd_pio(int fd, uint16_t addr, uint16_t val, |
41cb62c2 | 620 | bool assign, uint32_t size, bool datamatch) |
500ffd4a MT |
621 | { |
622 | struct kvm_ioeventfd kick = { | |
b680c5ba | 623 | .datamatch = datamatch ? adjust_ioeventfd_endianness(val, size) : 0, |
500ffd4a | 624 | .addr = addr, |
41cb62c2 | 625 | .flags = KVM_IOEVENTFD_FLAG_PIO, |
44c3f8f7 | 626 | .len = size, |
500ffd4a MT |
627 | .fd = fd, |
628 | }; | |
629 | int r; | |
630 | if (!kvm_enabled()) { | |
631 | return -ENOSYS; | |
632 | } | |
41cb62c2 MT |
633 | if (datamatch) { |
634 | kick.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH; | |
635 | } | |
500ffd4a MT |
636 | if (!assign) { |
637 | kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; | |
638 | } | |
639 | r = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick); | |
640 | if (r < 0) { | |
641 | return r; | |
642 | } | |
643 | return 0; | |
644 | } | |
645 | ||
646 | ||
d2f2b8a7 SH |
647 | static int kvm_check_many_ioeventfds(void) |
648 | { | |
d0dcac83 SH |
649 | /* Userspace can use ioeventfd for io notification. This requires a host |
650 | * that supports eventfd(2) and an I/O thread; since eventfd does not | |
651 | * support SIGIO it cannot interrupt the vcpu. | |
652 | * | |
653 | * Older kernels have a 6 device limit on the KVM io bus. Find out so we | |
d2f2b8a7 SH |
654 | * can avoid creating too many ioeventfds. |
655 | */ | |
12d4536f | 656 | #if defined(CONFIG_EVENTFD) |
d2f2b8a7 SH |
657 | int ioeventfds[7]; |
658 | int i, ret = 0; | |
659 | for (i = 0; i < ARRAY_SIZE(ioeventfds); i++) { | |
660 | ioeventfds[i] = eventfd(0, EFD_CLOEXEC); | |
661 | if (ioeventfds[i] < 0) { | |
662 | break; | |
663 | } | |
41cb62c2 | 664 | ret = kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, true, 2, true); |
d2f2b8a7 SH |
665 | if (ret < 0) { |
666 | close(ioeventfds[i]); | |
667 | break; | |
668 | } | |
669 | } | |
670 | ||
671 | /* Decide whether many devices are supported or not */ | |
672 | ret = i == ARRAY_SIZE(ioeventfds); | |
673 | ||
674 | while (i-- > 0) { | |
41cb62c2 | 675 | kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, false, 2, true); |
d2f2b8a7 SH |
676 | close(ioeventfds[i]); |
677 | } | |
678 | return ret; | |
679 | #else | |
680 | return 0; | |
681 | #endif | |
682 | } | |
683 | ||
94a8d39a JK |
684 | static const KVMCapabilityInfo * |
685 | kvm_check_extension_list(KVMState *s, const KVMCapabilityInfo *list) | |
686 | { | |
687 | while (list->name) { | |
688 | if (!kvm_check_extension(s, list->value)) { | |
689 | return list; | |
690 | } | |
691 | list++; | |
692 | } | |
693 | return NULL; | |
694 | } | |
695 | ||
7bbda04c PB |
696 | static void kvm_set_phys_mem(KVMMemoryListener *kml, |
697 | MemoryRegionSection *section, bool add) | |
46dbef6a MT |
698 | { |
699 | KVMState *s = kvm_state; | |
46dbef6a MT |
700 | KVMSlot *mem, old; |
701 | int err; | |
a01672d3 | 702 | MemoryRegion *mr = section->mr; |
235e8982 | 703 | bool writeable = !mr->readonly && !mr->rom_device; |
a8170e5e | 704 | hwaddr start_addr = section->offset_within_address_space; |
052e87b0 | 705 | ram_addr_t size = int128_get64(section->size); |
9f213ed9 | 706 | void *ram = NULL; |
8f6f962b | 707 | unsigned delta; |
46dbef6a | 708 | |
14542fea | 709 | /* kvm works in page size chunks, but the function may be called |
f2a64032 AG |
710 | with sub-page size and unaligned start address. Pad the start |
711 | address to next and truncate size to previous page boundary. */ | |
b232c785 AK |
712 | delta = qemu_real_host_page_size - (start_addr & ~qemu_real_host_page_mask); |
713 | delta &= ~qemu_real_host_page_mask; | |
8f6f962b AK |
714 | if (delta > size) { |
715 | return; | |
716 | } | |
717 | start_addr += delta; | |
718 | size -= delta; | |
b232c785 AK |
719 | size &= qemu_real_host_page_mask; |
720 | if (!size || (start_addr & ~qemu_real_host_page_mask)) { | |
8f6f962b AK |
721 | return; |
722 | } | |
46dbef6a | 723 | |
a01672d3 | 724 | if (!memory_region_is_ram(mr)) { |
235e8982 JJ |
725 | if (writeable || !kvm_readonly_mem_allowed) { |
726 | return; | |
727 | } else if (!mr->romd_mode) { | |
728 | /* If the memory device is not in romd_mode, then we actually want | |
729 | * to remove the kvm memory slot so all accesses will trap. */ | |
730 | add = false; | |
731 | } | |
9f213ed9 AK |
732 | } |
733 | ||
8f6f962b | 734 | ram = memory_region_get_ram_ptr(mr) + section->offset_within_region + delta; |
a01672d3 | 735 | |
46dbef6a | 736 | while (1) { |
7bbda04c | 737 | mem = kvm_lookup_overlapping_slot(kml, start_addr, start_addr + size); |
46dbef6a MT |
738 | if (!mem) { |
739 | break; | |
740 | } | |
741 | ||
a01672d3 | 742 | if (add && start_addr >= mem->start_addr && |
46dbef6a | 743 | (start_addr + size <= mem->start_addr + mem->memory_size) && |
9f213ed9 | 744 | (ram - start_addr == mem->ram - mem->start_addr)) { |
46dbef6a | 745 | /* The new slot fits into the existing one and comes with |
25254bbc | 746 | * identical parameters - update flags and done. */ |
7bbda04c | 747 | kvm_slot_update_flags(kml, mem, mr); |
46dbef6a MT |
748 | return; |
749 | } | |
750 | ||
751 | old = *mem; | |
752 | ||
1bfbac4e | 753 | if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) { |
7bbda04c | 754 | kvm_physical_sync_dirty_bitmap(kml, section); |
3fbffb62 AK |
755 | } |
756 | ||
46dbef6a MT |
757 | /* unregister the overlapping slot */ |
758 | mem->memory_size = 0; | |
7bbda04c | 759 | err = kvm_set_user_memory_region(kml, mem); |
46dbef6a MT |
760 | if (err) { |
761 | fprintf(stderr, "%s: error unregistering overlapping slot: %s\n", | |
762 | __func__, strerror(-err)); | |
763 | abort(); | |
764 | } | |
765 | ||
766 | /* Workaround for older KVM versions: we can't join slots, even not by | |
767 | * unregistering the previous ones and then registering the larger | |
768 | * slot. We have to maintain the existing fragmentation. Sigh. | |
769 | * | |
770 | * This workaround assumes that the new slot starts at the same | |
771 | * address as the first existing one. If not or if some overlapping | |
772 | * slot comes around later, we will fail (not seen in practice so far) | |
773 | * - and actually require a recent KVM version. */ | |
774 | if (s->broken_set_mem_region && | |
a01672d3 | 775 | old.start_addr == start_addr && old.memory_size < size && add) { |
7bbda04c | 776 | mem = kvm_alloc_slot(kml); |
46dbef6a MT |
777 | mem->memory_size = old.memory_size; |
778 | mem->start_addr = old.start_addr; | |
9f213ed9 | 779 | mem->ram = old.ram; |
d6ff5cbc | 780 | mem->flags = kvm_mem_flags(mr); |
46dbef6a | 781 | |
7bbda04c | 782 | err = kvm_set_user_memory_region(kml, mem); |
46dbef6a MT |
783 | if (err) { |
784 | fprintf(stderr, "%s: error updating slot: %s\n", __func__, | |
785 | strerror(-err)); | |
786 | abort(); | |
787 | } | |
788 | ||
789 | start_addr += old.memory_size; | |
9f213ed9 | 790 | ram += old.memory_size; |
46dbef6a MT |
791 | size -= old.memory_size; |
792 | continue; | |
793 | } | |
794 | ||
795 | /* register prefix slot */ | |
796 | if (old.start_addr < start_addr) { | |
7bbda04c | 797 | mem = kvm_alloc_slot(kml); |
46dbef6a MT |
798 | mem->memory_size = start_addr - old.start_addr; |
799 | mem->start_addr = old.start_addr; | |
9f213ed9 | 800 | mem->ram = old.ram; |
d6ff5cbc | 801 | mem->flags = kvm_mem_flags(mr); |
46dbef6a | 802 | |
7bbda04c | 803 | err = kvm_set_user_memory_region(kml, mem); |
46dbef6a MT |
804 | if (err) { |
805 | fprintf(stderr, "%s: error registering prefix slot: %s\n", | |
806 | __func__, strerror(-err)); | |
d4d6868f AG |
807 | #ifdef TARGET_PPC |
808 | fprintf(stderr, "%s: This is probably because your kernel's " \ | |
809 | "PAGE_SIZE is too big. Please try to use 4k " \ | |
810 | "PAGE_SIZE!\n", __func__); | |
811 | #endif | |
46dbef6a MT |
812 | abort(); |
813 | } | |
814 | } | |
815 | ||
816 | /* register suffix slot */ | |
817 | if (old.start_addr + old.memory_size > start_addr + size) { | |
818 | ram_addr_t size_delta; | |
819 | ||
7bbda04c | 820 | mem = kvm_alloc_slot(kml); |
46dbef6a MT |
821 | mem->start_addr = start_addr + size; |
822 | size_delta = mem->start_addr - old.start_addr; | |
823 | mem->memory_size = old.memory_size - size_delta; | |
9f213ed9 | 824 | mem->ram = old.ram + size_delta; |
d6ff5cbc | 825 | mem->flags = kvm_mem_flags(mr); |
46dbef6a | 826 | |
7bbda04c | 827 | err = kvm_set_user_memory_region(kml, mem); |
46dbef6a MT |
828 | if (err) { |
829 | fprintf(stderr, "%s: error registering suffix slot: %s\n", | |
830 | __func__, strerror(-err)); | |
831 | abort(); | |
832 | } | |
833 | } | |
834 | } | |
835 | ||
836 | /* in case the KVM bug workaround already "consumed" the new slot */ | |
a426e122 | 837 | if (!size) { |
46dbef6a | 838 | return; |
a426e122 | 839 | } |
a01672d3 | 840 | if (!add) { |
46dbef6a | 841 | return; |
a426e122 | 842 | } |
7bbda04c | 843 | mem = kvm_alloc_slot(kml); |
46dbef6a MT |
844 | mem->memory_size = size; |
845 | mem->start_addr = start_addr; | |
9f213ed9 | 846 | mem->ram = ram; |
d6ff5cbc | 847 | mem->flags = kvm_mem_flags(mr); |
46dbef6a | 848 | |
7bbda04c | 849 | err = kvm_set_user_memory_region(kml, mem); |
46dbef6a MT |
850 | if (err) { |
851 | fprintf(stderr, "%s: error registering slot: %s\n", __func__, | |
852 | strerror(-err)); | |
853 | abort(); | |
854 | } | |
855 | } | |
856 | ||
a01672d3 AK |
857 | static void kvm_region_add(MemoryListener *listener, |
858 | MemoryRegionSection *section) | |
859 | { | |
7bbda04c PB |
860 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); |
861 | ||
dfde4e6e | 862 | memory_region_ref(section->mr); |
7bbda04c | 863 | kvm_set_phys_mem(kml, section, true); |
a01672d3 AK |
864 | } |
865 | ||
866 | static void kvm_region_del(MemoryListener *listener, | |
867 | MemoryRegionSection *section) | |
868 | { | |
7bbda04c PB |
869 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); |
870 | ||
871 | kvm_set_phys_mem(kml, section, false); | |
dfde4e6e | 872 | memory_region_unref(section->mr); |
a01672d3 AK |
873 | } |
874 | ||
875 | static void kvm_log_sync(MemoryListener *listener, | |
876 | MemoryRegionSection *section) | |
7b8f3b78 | 877 | { |
7bbda04c | 878 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); |
a01672d3 AK |
879 | int r; |
880 | ||
7bbda04c | 881 | r = kvm_physical_sync_dirty_bitmap(kml, section); |
a01672d3 AK |
882 | if (r < 0) { |
883 | abort(); | |
884 | } | |
7b8f3b78 MT |
885 | } |
886 | ||
d22b096e AK |
887 | static void kvm_mem_ioeventfd_add(MemoryListener *listener, |
888 | MemoryRegionSection *section, | |
889 | bool match_data, uint64_t data, | |
890 | EventNotifier *e) | |
891 | { | |
892 | int fd = event_notifier_get_fd(e); | |
80a1ea37 AK |
893 | int r; |
894 | ||
4b8f1c88 | 895 | r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space, |
052e87b0 PB |
896 | data, true, int128_get64(section->size), |
897 | match_data); | |
80a1ea37 | 898 | if (r < 0) { |
fa4ba923 AK |
899 | fprintf(stderr, "%s: error adding ioeventfd: %s\n", |
900 | __func__, strerror(-r)); | |
80a1ea37 AK |
901 | abort(); |
902 | } | |
903 | } | |
904 | ||
d22b096e AK |
905 | static void kvm_mem_ioeventfd_del(MemoryListener *listener, |
906 | MemoryRegionSection *section, | |
907 | bool match_data, uint64_t data, | |
908 | EventNotifier *e) | |
80a1ea37 | 909 | { |
d22b096e | 910 | int fd = event_notifier_get_fd(e); |
80a1ea37 AK |
911 | int r; |
912 | ||
4b8f1c88 | 913 | r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space, |
052e87b0 PB |
914 | data, false, int128_get64(section->size), |
915 | match_data); | |
80a1ea37 AK |
916 | if (r < 0) { |
917 | abort(); | |
918 | } | |
919 | } | |
920 | ||
d22b096e AK |
921 | static void kvm_io_ioeventfd_add(MemoryListener *listener, |
922 | MemoryRegionSection *section, | |
923 | bool match_data, uint64_t data, | |
924 | EventNotifier *e) | |
80a1ea37 | 925 | { |
d22b096e | 926 | int fd = event_notifier_get_fd(e); |
80a1ea37 AK |
927 | int r; |
928 | ||
44c3f8f7 | 929 | r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space, |
052e87b0 PB |
930 | data, true, int128_get64(section->size), |
931 | match_data); | |
80a1ea37 | 932 | if (r < 0) { |
fa4ba923 AK |
933 | fprintf(stderr, "%s: error adding ioeventfd: %s\n", |
934 | __func__, strerror(-r)); | |
80a1ea37 AK |
935 | abort(); |
936 | } | |
937 | } | |
938 | ||
d22b096e AK |
939 | static void kvm_io_ioeventfd_del(MemoryListener *listener, |
940 | MemoryRegionSection *section, | |
941 | bool match_data, uint64_t data, | |
942 | EventNotifier *e) | |
80a1ea37 AK |
943 | |
944 | { | |
d22b096e | 945 | int fd = event_notifier_get_fd(e); |
80a1ea37 AK |
946 | int r; |
947 | ||
44c3f8f7 | 948 | r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space, |
052e87b0 PB |
949 | data, false, int128_get64(section->size), |
950 | match_data); | |
80a1ea37 AK |
951 | if (r < 0) { |
952 | abort(); | |
953 | } | |
954 | } | |
955 | ||
38bfe691 PB |
956 | void kvm_memory_listener_register(KVMState *s, KVMMemoryListener *kml, |
957 | AddressSpace *as, int as_id) | |
7bbda04c PB |
958 | { |
959 | int i; | |
960 | ||
961 | kml->slots = g_malloc0(s->nr_slots * sizeof(KVMSlot)); | |
38bfe691 | 962 | kml->as_id = as_id; |
7bbda04c PB |
963 | |
964 | for (i = 0; i < s->nr_slots; i++) { | |
965 | kml->slots[i].slot = i; | |
966 | } | |
967 | ||
968 | kml->listener.region_add = kvm_region_add; | |
969 | kml->listener.region_del = kvm_region_del; | |
970 | kml->listener.log_start = kvm_log_start; | |
971 | kml->listener.log_stop = kvm_log_stop; | |
972 | kml->listener.log_sync = kvm_log_sync; | |
973 | kml->listener.priority = 10; | |
974 | ||
975 | memory_listener_register(&kml->listener, as); | |
976 | } | |
d22b096e AK |
977 | |
978 | static MemoryListener kvm_io_listener = { | |
d22b096e AK |
979 | .eventfd_add = kvm_io_ioeventfd_add, |
980 | .eventfd_del = kvm_io_ioeventfd_del, | |
72e22d2f | 981 | .priority = 10, |
7b8f3b78 MT |
982 | }; |
983 | ||
c3affe56 | 984 | static void kvm_handle_interrupt(CPUState *cpu, int mask) |
aa7f74d1 | 985 | { |
259186a7 | 986 | cpu->interrupt_request |= mask; |
aa7f74d1 | 987 | |
60e82579 | 988 | if (!qemu_cpu_is_self(cpu)) { |
c08d7424 | 989 | qemu_cpu_kick(cpu); |
aa7f74d1 JK |
990 | } |
991 | } | |
992 | ||
3889c3fa | 993 | int kvm_set_irq(KVMState *s, int irq, int level) |
84b058d7 JK |
994 | { |
995 | struct kvm_irq_level event; | |
996 | int ret; | |
997 | ||
7ae26bd4 | 998 | assert(kvm_async_interrupts_enabled()); |
84b058d7 JK |
999 | |
1000 | event.level = level; | |
1001 | event.irq = irq; | |
e333cd69 | 1002 | ret = kvm_vm_ioctl(s, s->irq_set_ioctl, &event); |
84b058d7 | 1003 | if (ret < 0) { |
3889c3fa | 1004 | perror("kvm_set_irq"); |
84b058d7 JK |
1005 | abort(); |
1006 | } | |
1007 | ||
e333cd69 | 1008 | return (s->irq_set_ioctl == KVM_IRQ_LINE) ? 1 : event.status; |
84b058d7 JK |
1009 | } |
1010 | ||
1011 | #ifdef KVM_CAP_IRQ_ROUTING | |
d3d3bef0 JK |
1012 | typedef struct KVMMSIRoute { |
1013 | struct kvm_irq_routing_entry kroute; | |
1014 | QTAILQ_ENTRY(KVMMSIRoute) entry; | |
1015 | } KVMMSIRoute; | |
1016 | ||
84b058d7 JK |
1017 | static void set_gsi(KVMState *s, unsigned int gsi) |
1018 | { | |
8269fb70 | 1019 | set_bit(gsi, s->used_gsi_bitmap); |
84b058d7 JK |
1020 | } |
1021 | ||
04fa27f5 JK |
1022 | static void clear_gsi(KVMState *s, unsigned int gsi) |
1023 | { | |
8269fb70 | 1024 | clear_bit(gsi, s->used_gsi_bitmap); |
04fa27f5 JK |
1025 | } |
1026 | ||
7b774593 | 1027 | void kvm_init_irq_routing(KVMState *s) |
84b058d7 | 1028 | { |
04fa27f5 | 1029 | int gsi_count, i; |
84b058d7 | 1030 | |
00008418 | 1031 | gsi_count = kvm_check_extension(s, KVM_CAP_IRQ_ROUTING) - 1; |
84b058d7 | 1032 | if (gsi_count > 0) { |
84b058d7 | 1033 | /* Round up so we can search ints using ffs */ |
8269fb70 | 1034 | s->used_gsi_bitmap = bitmap_new(gsi_count); |
4e2e4e63 | 1035 | s->gsi_count = gsi_count; |
84b058d7 JK |
1036 | } |
1037 | ||
1038 | s->irq_routes = g_malloc0(sizeof(*s->irq_routes)); | |
1039 | s->nr_allocated_irq_routes = 0; | |
1040 | ||
50bf31b9 | 1041 | if (!kvm_direct_msi_allowed) { |
4a3adebb JK |
1042 | for (i = 0; i < KVM_MSI_HASHTAB_SIZE; i++) { |
1043 | QTAILQ_INIT(&s->msi_hashtab[i]); | |
1044 | } | |
04fa27f5 JK |
1045 | } |
1046 | ||
84b058d7 JK |
1047 | kvm_arch_init_irq_routing(s); |
1048 | } | |
1049 | ||
cb925cf9 | 1050 | void kvm_irqchip_commit_routes(KVMState *s) |
e7b20308 JK |
1051 | { |
1052 | int ret; | |
1053 | ||
7005f7f8 PX |
1054 | if (kvm_gsi_direct_mapping()) { |
1055 | return; | |
1056 | } | |
1057 | ||
1058 | if (!kvm_gsi_routing_enabled()) { | |
1059 | return; | |
1060 | } | |
1061 | ||
e7b20308 | 1062 | s->irq_routes->flags = 0; |
54a6c11b | 1063 | trace_kvm_irqchip_commit_routes(); |
e7b20308 JK |
1064 | ret = kvm_vm_ioctl(s, KVM_SET_GSI_ROUTING, s->irq_routes); |
1065 | assert(ret == 0); | |
1066 | } | |
1067 | ||
84b058d7 JK |
1068 | static void kvm_add_routing_entry(KVMState *s, |
1069 | struct kvm_irq_routing_entry *entry) | |
1070 | { | |
1071 | struct kvm_irq_routing_entry *new; | |
1072 | int n, size; | |
1073 | ||
1074 | if (s->irq_routes->nr == s->nr_allocated_irq_routes) { | |
1075 | n = s->nr_allocated_irq_routes * 2; | |
1076 | if (n < 64) { | |
1077 | n = 64; | |
1078 | } | |
1079 | size = sizeof(struct kvm_irq_routing); | |
1080 | size += n * sizeof(*new); | |
1081 | s->irq_routes = g_realloc(s->irq_routes, size); | |
1082 | s->nr_allocated_irq_routes = n; | |
1083 | } | |
1084 | n = s->irq_routes->nr++; | |
1085 | new = &s->irq_routes->entries[n]; | |
0fbc2074 MT |
1086 | |
1087 | *new = *entry; | |
84b058d7 JK |
1088 | |
1089 | set_gsi(s, entry->gsi); | |
1090 | } | |
1091 | ||
cc57407e JK |
1092 | static int kvm_update_routing_entry(KVMState *s, |
1093 | struct kvm_irq_routing_entry *new_entry) | |
1094 | { | |
1095 | struct kvm_irq_routing_entry *entry; | |
1096 | int n; | |
1097 | ||
1098 | for (n = 0; n < s->irq_routes->nr; n++) { | |
1099 | entry = &s->irq_routes->entries[n]; | |
1100 | if (entry->gsi != new_entry->gsi) { | |
1101 | continue; | |
1102 | } | |
1103 | ||
40509f7f MT |
1104 | if(!memcmp(entry, new_entry, sizeof *entry)) { |
1105 | return 0; | |
1106 | } | |
1107 | ||
0fbc2074 | 1108 | *entry = *new_entry; |
cc57407e | 1109 | |
cc57407e JK |
1110 | return 0; |
1111 | } | |
1112 | ||
1113 | return -ESRCH; | |
1114 | } | |
1115 | ||
1df186df | 1116 | void kvm_irqchip_add_irq_route(KVMState *s, int irq, int irqchip, int pin) |
84b058d7 | 1117 | { |
0fbc2074 | 1118 | struct kvm_irq_routing_entry e = {}; |
84b058d7 | 1119 | |
4e2e4e63 JK |
1120 | assert(pin < s->gsi_count); |
1121 | ||
84b058d7 JK |
1122 | e.gsi = irq; |
1123 | e.type = KVM_IRQ_ROUTING_IRQCHIP; | |
1124 | e.flags = 0; | |
1125 | e.u.irqchip.irqchip = irqchip; | |
1126 | e.u.irqchip.pin = pin; | |
1127 | kvm_add_routing_entry(s, &e); | |
1128 | } | |
1129 | ||
1e2aa8be | 1130 | void kvm_irqchip_release_virq(KVMState *s, int virq) |
04fa27f5 JK |
1131 | { |
1132 | struct kvm_irq_routing_entry *e; | |
1133 | int i; | |
1134 | ||
76fe21de AK |
1135 | if (kvm_gsi_direct_mapping()) { |
1136 | return; | |
1137 | } | |
1138 | ||
04fa27f5 JK |
1139 | for (i = 0; i < s->irq_routes->nr; i++) { |
1140 | e = &s->irq_routes->entries[i]; | |
1141 | if (e->gsi == virq) { | |
1142 | s->irq_routes->nr--; | |
1143 | *e = s->irq_routes->entries[s->irq_routes->nr]; | |
1144 | } | |
1145 | } | |
1146 | clear_gsi(s, virq); | |
38d87493 | 1147 | kvm_arch_release_virq_post(virq); |
9ba35d0b | 1148 | trace_kvm_irqchip_release_virq(virq); |
04fa27f5 JK |
1149 | } |
1150 | ||
1151 | static unsigned int kvm_hash_msi(uint32_t data) | |
1152 | { | |
1153 | /* This is optimized for IA32 MSI layout. However, no other arch shall | |
1154 | * repeat the mistake of not providing a direct MSI injection API. */ | |
1155 | return data & 0xff; | |
1156 | } | |
1157 | ||
1158 | static void kvm_flush_dynamic_msi_routes(KVMState *s) | |
1159 | { | |
1160 | KVMMSIRoute *route, *next; | |
1161 | unsigned int hash; | |
1162 | ||
1163 | for (hash = 0; hash < KVM_MSI_HASHTAB_SIZE; hash++) { | |
1164 | QTAILQ_FOREACH_SAFE(route, &s->msi_hashtab[hash], entry, next) { | |
1165 | kvm_irqchip_release_virq(s, route->kroute.gsi); | |
1166 | QTAILQ_REMOVE(&s->msi_hashtab[hash], route, entry); | |
1167 | g_free(route); | |
1168 | } | |
1169 | } | |
1170 | } | |
1171 | ||
1172 | static int kvm_irqchip_get_virq(KVMState *s) | |
1173 | { | |
8269fb70 | 1174 | int next_virq; |
04fa27f5 | 1175 | |
bdf02631 WM |
1176 | /* |
1177 | * PIC and IOAPIC share the first 16 GSI numbers, thus the available | |
1178 | * GSI numbers are more than the number of IRQ route. Allocating a GSI | |
1179 | * number can succeed even though a new route entry cannot be added. | |
1180 | * When this happens, flush dynamic MSI entries to free IRQ route entries. | |
1181 | */ | |
50bf31b9 | 1182 | if (!kvm_direct_msi_allowed && s->irq_routes->nr == s->gsi_count) { |
bdf02631 WM |
1183 | kvm_flush_dynamic_msi_routes(s); |
1184 | } | |
1185 | ||
04fa27f5 | 1186 | /* Return the lowest unused GSI in the bitmap */ |
8269fb70 WY |
1187 | next_virq = find_first_zero_bit(s->used_gsi_bitmap, s->gsi_count); |
1188 | if (next_virq >= s->gsi_count) { | |
1189 | return -ENOSPC; | |
1190 | } else { | |
1191 | return next_virq; | |
04fa27f5 | 1192 | } |
04fa27f5 JK |
1193 | } |
1194 | ||
1195 | static KVMMSIRoute *kvm_lookup_msi_route(KVMState *s, MSIMessage msg) | |
1196 | { | |
1197 | unsigned int hash = kvm_hash_msi(msg.data); | |
1198 | KVMMSIRoute *route; | |
1199 | ||
1200 | QTAILQ_FOREACH(route, &s->msi_hashtab[hash], entry) { | |
1201 | if (route->kroute.u.msi.address_lo == (uint32_t)msg.address && | |
1202 | route->kroute.u.msi.address_hi == (msg.address >> 32) && | |
d07cc1f1 | 1203 | route->kroute.u.msi.data == le32_to_cpu(msg.data)) { |
04fa27f5 JK |
1204 | return route; |
1205 | } | |
1206 | } | |
1207 | return NULL; | |
1208 | } | |
1209 | ||
1210 | int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg) | |
1211 | { | |
4a3adebb | 1212 | struct kvm_msi msi; |
04fa27f5 JK |
1213 | KVMMSIRoute *route; |
1214 | ||
50bf31b9 | 1215 | if (kvm_direct_msi_allowed) { |
4a3adebb JK |
1216 | msi.address_lo = (uint32_t)msg.address; |
1217 | msi.address_hi = msg.address >> 32; | |
d07cc1f1 | 1218 | msi.data = le32_to_cpu(msg.data); |
4a3adebb JK |
1219 | msi.flags = 0; |
1220 | memset(msi.pad, 0, sizeof(msi.pad)); | |
1221 | ||
1222 | return kvm_vm_ioctl(s, KVM_SIGNAL_MSI, &msi); | |
1223 | } | |
1224 | ||
04fa27f5 JK |
1225 | route = kvm_lookup_msi_route(s, msg); |
1226 | if (!route) { | |
e7b20308 | 1227 | int virq; |
04fa27f5 JK |
1228 | |
1229 | virq = kvm_irqchip_get_virq(s); | |
1230 | if (virq < 0) { | |
1231 | return virq; | |
1232 | } | |
1233 | ||
0fbc2074 | 1234 | route = g_malloc0(sizeof(KVMMSIRoute)); |
04fa27f5 JK |
1235 | route->kroute.gsi = virq; |
1236 | route->kroute.type = KVM_IRQ_ROUTING_MSI; | |
1237 | route->kroute.flags = 0; | |
1238 | route->kroute.u.msi.address_lo = (uint32_t)msg.address; | |
1239 | route->kroute.u.msi.address_hi = msg.address >> 32; | |
d07cc1f1 | 1240 | route->kroute.u.msi.data = le32_to_cpu(msg.data); |
04fa27f5 JK |
1241 | |
1242 | kvm_add_routing_entry(s, &route->kroute); | |
cb925cf9 | 1243 | kvm_irqchip_commit_routes(s); |
04fa27f5 JK |
1244 | |
1245 | QTAILQ_INSERT_TAIL(&s->msi_hashtab[kvm_hash_msi(msg.data)], route, | |
1246 | entry); | |
04fa27f5 JK |
1247 | } |
1248 | ||
1249 | assert(route->kroute.type == KVM_IRQ_ROUTING_MSI); | |
1250 | ||
3889c3fa | 1251 | return kvm_set_irq(s, route->kroute.gsi, 1); |
04fa27f5 JK |
1252 | } |
1253 | ||
d1f6af6a | 1254 | int kvm_irqchip_add_msi_route(KVMState *s, int vector, PCIDevice *dev) |
92b4e489 | 1255 | { |
0fbc2074 | 1256 | struct kvm_irq_routing_entry kroute = {}; |
92b4e489 | 1257 | int virq; |
d1f6af6a PX |
1258 | MSIMessage msg = {0, 0}; |
1259 | ||
1260 | if (dev) { | |
e1d4fb2d | 1261 | msg = pci_get_msi_message(dev, vector); |
d1f6af6a | 1262 | } |
92b4e489 | 1263 | |
76fe21de | 1264 | if (kvm_gsi_direct_mapping()) { |
1850b6b7 | 1265 | return kvm_arch_msi_data_to_gsi(msg.data); |
76fe21de AK |
1266 | } |
1267 | ||
f3e1bed8 | 1268 | if (!kvm_gsi_routing_enabled()) { |
92b4e489 JK |
1269 | return -ENOSYS; |
1270 | } | |
1271 | ||
1272 | virq = kvm_irqchip_get_virq(s); | |
1273 | if (virq < 0) { | |
1274 | return virq; | |
1275 | } | |
1276 | ||
1277 | kroute.gsi = virq; | |
1278 | kroute.type = KVM_IRQ_ROUTING_MSI; | |
1279 | kroute.flags = 0; | |
1280 | kroute.u.msi.address_lo = (uint32_t)msg.address; | |
1281 | kroute.u.msi.address_hi = msg.address >> 32; | |
d07cc1f1 | 1282 | kroute.u.msi.data = le32_to_cpu(msg.data); |
767a554a PF |
1283 | if (kvm_msi_devid_required()) { |
1284 | kroute.flags = KVM_MSI_VALID_DEVID; | |
1285 | kroute.u.msi.devid = pci_requester_id(dev); | |
1286 | } | |
dc9f06ca | 1287 | if (kvm_arch_fixup_msi_route(&kroute, msg.address, msg.data, dev)) { |
9e03a040 FB |
1288 | kvm_irqchip_release_virq(s, virq); |
1289 | return -EINVAL; | |
1290 | } | |
92b4e489 | 1291 | |
9ba35d0b PX |
1292 | trace_kvm_irqchip_add_msi_route(dev ? dev->name : (char *)"N/A", |
1293 | vector, virq); | |
54a6c11b | 1294 | |
92b4e489 | 1295 | kvm_add_routing_entry(s, &kroute); |
38d87493 | 1296 | kvm_arch_add_msi_route_post(&kroute, vector, dev); |
cb925cf9 | 1297 | kvm_irqchip_commit_routes(s); |
92b4e489 JK |
1298 | |
1299 | return virq; | |
1300 | } | |
1301 | ||
dc9f06ca PF |
1302 | int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg, |
1303 | PCIDevice *dev) | |
cc57407e | 1304 | { |
0fbc2074 | 1305 | struct kvm_irq_routing_entry kroute = {}; |
cc57407e | 1306 | |
76fe21de AK |
1307 | if (kvm_gsi_direct_mapping()) { |
1308 | return 0; | |
1309 | } | |
1310 | ||
cc57407e JK |
1311 | if (!kvm_irqchip_in_kernel()) { |
1312 | return -ENOSYS; | |
1313 | } | |
1314 | ||
1315 | kroute.gsi = virq; | |
1316 | kroute.type = KVM_IRQ_ROUTING_MSI; | |
1317 | kroute.flags = 0; | |
1318 | kroute.u.msi.address_lo = (uint32_t)msg.address; | |
1319 | kroute.u.msi.address_hi = msg.address >> 32; | |
d07cc1f1 | 1320 | kroute.u.msi.data = le32_to_cpu(msg.data); |
767a554a PF |
1321 | if (kvm_msi_devid_required()) { |
1322 | kroute.flags = KVM_MSI_VALID_DEVID; | |
1323 | kroute.u.msi.devid = pci_requester_id(dev); | |
1324 | } | |
dc9f06ca | 1325 | if (kvm_arch_fixup_msi_route(&kroute, msg.address, msg.data, dev)) { |
9e03a040 FB |
1326 | return -EINVAL; |
1327 | } | |
cc57407e | 1328 | |
54a6c11b PX |
1329 | trace_kvm_irqchip_update_msi_route(virq); |
1330 | ||
cc57407e JK |
1331 | return kvm_update_routing_entry(s, &kroute); |
1332 | } | |
1333 | ||
ca916d37 VM |
1334 | static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int rfd, int virq, |
1335 | bool assign) | |
39853bbc JK |
1336 | { |
1337 | struct kvm_irqfd irqfd = { | |
1338 | .fd = fd, | |
1339 | .gsi = virq, | |
1340 | .flags = assign ? 0 : KVM_IRQFD_FLAG_DEASSIGN, | |
1341 | }; | |
1342 | ||
ca916d37 VM |
1343 | if (rfd != -1) { |
1344 | irqfd.flags |= KVM_IRQFD_FLAG_RESAMPLE; | |
1345 | irqfd.resamplefd = rfd; | |
1346 | } | |
1347 | ||
cc7e0ddf | 1348 | if (!kvm_irqfds_enabled()) { |
39853bbc JK |
1349 | return -ENOSYS; |
1350 | } | |
1351 | ||
1352 | return kvm_vm_ioctl(s, KVM_IRQFD, &irqfd); | |
1353 | } | |
1354 | ||
d426d9fb CH |
1355 | int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter) |
1356 | { | |
e9af2fef | 1357 | struct kvm_irq_routing_entry kroute = {}; |
d426d9fb CH |
1358 | int virq; |
1359 | ||
1360 | if (!kvm_gsi_routing_enabled()) { | |
1361 | return -ENOSYS; | |
1362 | } | |
1363 | ||
1364 | virq = kvm_irqchip_get_virq(s); | |
1365 | if (virq < 0) { | |
1366 | return virq; | |
1367 | } | |
1368 | ||
1369 | kroute.gsi = virq; | |
1370 | kroute.type = KVM_IRQ_ROUTING_S390_ADAPTER; | |
1371 | kroute.flags = 0; | |
1372 | kroute.u.adapter.summary_addr = adapter->summary_addr; | |
1373 | kroute.u.adapter.ind_addr = adapter->ind_addr; | |
1374 | kroute.u.adapter.summary_offset = adapter->summary_offset; | |
1375 | kroute.u.adapter.ind_offset = adapter->ind_offset; | |
1376 | kroute.u.adapter.adapter_id = adapter->adapter_id; | |
1377 | ||
1378 | kvm_add_routing_entry(s, &kroute); | |
d426d9fb CH |
1379 | |
1380 | return virq; | |
1381 | } | |
1382 | ||
977a8d9c AS |
1383 | int kvm_irqchip_add_hv_sint_route(KVMState *s, uint32_t vcpu, uint32_t sint) |
1384 | { | |
1385 | struct kvm_irq_routing_entry kroute = {}; | |
1386 | int virq; | |
1387 | ||
1388 | if (!kvm_gsi_routing_enabled()) { | |
1389 | return -ENOSYS; | |
1390 | } | |
1391 | if (!kvm_check_extension(s, KVM_CAP_HYPERV_SYNIC)) { | |
1392 | return -ENOSYS; | |
1393 | } | |
1394 | virq = kvm_irqchip_get_virq(s); | |
1395 | if (virq < 0) { | |
1396 | return virq; | |
1397 | } | |
1398 | ||
1399 | kroute.gsi = virq; | |
1400 | kroute.type = KVM_IRQ_ROUTING_HV_SINT; | |
1401 | kroute.flags = 0; | |
1402 | kroute.u.hv_sint.vcpu = vcpu; | |
1403 | kroute.u.hv_sint.sint = sint; | |
1404 | ||
1405 | kvm_add_routing_entry(s, &kroute); | |
1406 | kvm_irqchip_commit_routes(s); | |
1407 | ||
1408 | return virq; | |
1409 | } | |
1410 | ||
84b058d7 JK |
1411 | #else /* !KVM_CAP_IRQ_ROUTING */ |
1412 | ||
7b774593 | 1413 | void kvm_init_irq_routing(KVMState *s) |
84b058d7 JK |
1414 | { |
1415 | } | |
04fa27f5 | 1416 | |
d3d3bef0 JK |
1417 | void kvm_irqchip_release_virq(KVMState *s, int virq) |
1418 | { | |
1419 | } | |
1420 | ||
04fa27f5 JK |
1421 | int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg) |
1422 | { | |
1423 | abort(); | |
1424 | } | |
92b4e489 | 1425 | |
d1f6af6a | 1426 | int kvm_irqchip_add_msi_route(KVMState *s, int vector, PCIDevice *dev) |
92b4e489 | 1427 | { |
df410675 | 1428 | return -ENOSYS; |
92b4e489 | 1429 | } |
39853bbc | 1430 | |
d426d9fb CH |
1431 | int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter) |
1432 | { | |
1433 | return -ENOSYS; | |
1434 | } | |
1435 | ||
977a8d9c AS |
1436 | int kvm_irqchip_add_hv_sint_route(KVMState *s, uint32_t vcpu, uint32_t sint) |
1437 | { | |
1438 | return -ENOSYS; | |
1439 | } | |
1440 | ||
39853bbc JK |
1441 | static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int virq, bool assign) |
1442 | { | |
1443 | abort(); | |
1444 | } | |
dabe3143 MT |
1445 | |
1446 | int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg) | |
1447 | { | |
1448 | return -ENOSYS; | |
1449 | } | |
84b058d7 JK |
1450 | #endif /* !KVM_CAP_IRQ_ROUTING */ |
1451 | ||
1c9b71a7 EA |
1452 | int kvm_irqchip_add_irqfd_notifier_gsi(KVMState *s, EventNotifier *n, |
1453 | EventNotifier *rn, int virq) | |
39853bbc | 1454 | { |
ca916d37 VM |
1455 | return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n), |
1456 | rn ? event_notifier_get_fd(rn) : -1, virq, true); | |
39853bbc JK |
1457 | } |
1458 | ||
1c9b71a7 EA |
1459 | int kvm_irqchip_remove_irqfd_notifier_gsi(KVMState *s, EventNotifier *n, |
1460 | int virq) | |
15b2bd18 | 1461 | { |
ca916d37 VM |
1462 | return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n), -1, virq, |
1463 | false); | |
15b2bd18 PB |
1464 | } |
1465 | ||
197e3524 EA |
1466 | int kvm_irqchip_add_irqfd_notifier(KVMState *s, EventNotifier *n, |
1467 | EventNotifier *rn, qemu_irq irq) | |
1468 | { | |
1469 | gpointer key, gsi; | |
1470 | gboolean found = g_hash_table_lookup_extended(s->gsimap, irq, &key, &gsi); | |
1471 | ||
1472 | if (!found) { | |
1473 | return -ENXIO; | |
1474 | } | |
1475 | return kvm_irqchip_add_irqfd_notifier_gsi(s, n, rn, GPOINTER_TO_INT(gsi)); | |
1476 | } | |
1477 | ||
1478 | int kvm_irqchip_remove_irqfd_notifier(KVMState *s, EventNotifier *n, | |
1479 | qemu_irq irq) | |
1480 | { | |
1481 | gpointer key, gsi; | |
1482 | gboolean found = g_hash_table_lookup_extended(s->gsimap, irq, &key, &gsi); | |
1483 | ||
1484 | if (!found) { | |
1485 | return -ENXIO; | |
1486 | } | |
1487 | return kvm_irqchip_remove_irqfd_notifier_gsi(s, n, GPOINTER_TO_INT(gsi)); | |
1488 | } | |
1489 | ||
1490 | void kvm_irqchip_set_qemuirq_gsi(KVMState *s, qemu_irq irq, int gsi) | |
1491 | { | |
1492 | g_hash_table_insert(s->gsimap, irq, GINT_TO_POINTER(gsi)); | |
1493 | } | |
1494 | ||
8db4936b | 1495 | static void kvm_irqchip_create(MachineState *machine, KVMState *s) |
84b058d7 | 1496 | { |
84b058d7 JK |
1497 | int ret; |
1498 | ||
8db4936b PB |
1499 | if (kvm_check_extension(s, KVM_CAP_IRQCHIP)) { |
1500 | ; | |
1501 | } else if (kvm_check_extension(s, KVM_CAP_S390_IRQCHIP)) { | |
1502 | ret = kvm_vm_enable_cap(s, KVM_CAP_S390_IRQCHIP, 0); | |
1503 | if (ret < 0) { | |
1504 | fprintf(stderr, "Enable kernel irqchip failed: %s\n", strerror(-ret)); | |
1505 | exit(1); | |
1506 | } | |
1507 | } else { | |
1508 | return; | |
84b058d7 JK |
1509 | } |
1510 | ||
d6032e06 CD |
1511 | /* First probe and see if there's a arch-specific hook to create the |
1512 | * in-kernel irqchip for us */ | |
15eafc2e | 1513 | ret = kvm_arch_irqchip_create(machine, s); |
8db4936b | 1514 | if (ret == 0) { |
15eafc2e PB |
1515 | if (machine_kernel_irqchip_split(machine)) { |
1516 | perror("Split IRQ chip mode not supported."); | |
1517 | exit(1); | |
1518 | } else { | |
1519 | ret = kvm_vm_ioctl(s, KVM_CREATE_IRQCHIP); | |
1520 | } | |
8db4936b PB |
1521 | } |
1522 | if (ret < 0) { | |
1523 | fprintf(stderr, "Create kernel irqchip failed: %s\n", strerror(-ret)); | |
1524 | exit(1); | |
84b058d7 JK |
1525 | } |
1526 | ||
3d4b2649 | 1527 | kvm_kernel_irqchip = true; |
7ae26bd4 PM |
1528 | /* If we have an in-kernel IRQ chip then we must have asynchronous |
1529 | * interrupt delivery (though the reverse is not necessarily true) | |
1530 | */ | |
1531 | kvm_async_interrupts_allowed = true; | |
215e79c0 | 1532 | kvm_halt_in_kernel_allowed = true; |
84b058d7 JK |
1533 | |
1534 | kvm_init_irq_routing(s); | |
1535 | ||
197e3524 | 1536 | s->gsimap = g_hash_table_new(g_direct_hash, g_direct_equal); |
84b058d7 JK |
1537 | } |
1538 | ||
670436ce AJ |
1539 | /* Find number of supported CPUs using the recommended |
1540 | * procedure from the kernel API documentation to cope with | |
1541 | * older kernels that may be missing capabilities. | |
1542 | */ | |
1543 | static int kvm_recommended_vcpus(KVMState *s) | |
3ed444e9 | 1544 | { |
670436ce AJ |
1545 | int ret = kvm_check_extension(s, KVM_CAP_NR_VCPUS); |
1546 | return (ret) ? ret : 4; | |
1547 | } | |
3ed444e9 | 1548 | |
670436ce AJ |
1549 | static int kvm_max_vcpus(KVMState *s) |
1550 | { | |
1551 | int ret = kvm_check_extension(s, KVM_CAP_MAX_VCPUS); | |
1552 | return (ret) ? ret : kvm_recommended_vcpus(s); | |
3ed444e9 DH |
1553 | } |
1554 | ||
f31e3266 GK |
1555 | static int kvm_max_vcpu_id(KVMState *s) |
1556 | { | |
1557 | int ret = kvm_check_extension(s, KVM_CAP_MAX_VCPU_ID); | |
1558 | return (ret) ? ret : kvm_max_vcpus(s); | |
1559 | } | |
1560 | ||
41264b38 GK |
1561 | bool kvm_vcpu_id_is_valid(int vcpu_id) |
1562 | { | |
1563 | KVMState *s = KVM_STATE(current_machine->accelerator); | |
f31e3266 | 1564 | return vcpu_id >= 0 && vcpu_id < kvm_max_vcpu_id(s); |
41264b38 GK |
1565 | } |
1566 | ||
f6a1ef64 | 1567 | static int kvm_init(MachineState *ms) |
05330448 | 1568 | { |
f6a1ef64 | 1569 | MachineClass *mc = MACHINE_GET_CLASS(ms); |
168ccc11 JK |
1570 | static const char upgrade_note[] = |
1571 | "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n" | |
1572 | "(see http://sourceforge.net/projects/kvm).\n"; | |
670436ce AJ |
1573 | struct { |
1574 | const char *name; | |
1575 | int num; | |
1576 | } num_cpus[] = { | |
1577 | { "SMP", smp_cpus }, | |
1578 | { "hotpluggable", max_cpus }, | |
1579 | { NULL, } | |
1580 | }, *nc = num_cpus; | |
1581 | int soft_vcpus_limit, hard_vcpus_limit; | |
05330448 | 1582 | KVMState *s; |
94a8d39a | 1583 | const KVMCapabilityInfo *missing_cap; |
05330448 | 1584 | int ret; |
7bbda04c | 1585 | int type = 0; |
135a129a | 1586 | const char *kvm_type; |
05330448 | 1587 | |
fc02086b | 1588 | s = KVM_STATE(ms->accelerator); |
05330448 | 1589 | |
3145fcb6 DG |
1590 | /* |
1591 | * On systems where the kernel can support different base page | |
1592 | * sizes, host page size may be different from TARGET_PAGE_SIZE, | |
1593 | * even with KVM. TARGET_PAGE_SIZE is assumed to be the minimum | |
1594 | * page size for the system though. | |
1595 | */ | |
1596 | assert(TARGET_PAGE_SIZE <= getpagesize()); | |
1597 | ||
aed6efb9 JH |
1598 | s->sigmask_len = 8; |
1599 | ||
e22a25c9 | 1600 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
72cf2d4f | 1601 | QTAILQ_INIT(&s->kvm_sw_breakpoints); |
e22a25c9 | 1602 | #endif |
4c055ab5 | 1603 | QLIST_INIT(&s->kvm_parked_vcpus); |
05330448 | 1604 | s->vmfd = -1; |
40ff6d7e | 1605 | s->fd = qemu_open("/dev/kvm", O_RDWR); |
05330448 AL |
1606 | if (s->fd == -1) { |
1607 | fprintf(stderr, "Could not access KVM kernel module: %m\n"); | |
1608 | ret = -errno; | |
1609 | goto err; | |
1610 | } | |
1611 | ||
1612 | ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0); | |
1613 | if (ret < KVM_API_VERSION) { | |
0e1dac6c | 1614 | if (ret >= 0) { |
05330448 | 1615 | ret = -EINVAL; |
a426e122 | 1616 | } |
05330448 AL |
1617 | fprintf(stderr, "kvm version too old\n"); |
1618 | goto err; | |
1619 | } | |
1620 | ||
1621 | if (ret > KVM_API_VERSION) { | |
1622 | ret = -EINVAL; | |
1623 | fprintf(stderr, "kvm version not supported\n"); | |
1624 | goto err; | |
1625 | } | |
1626 | ||
cf0f7cf9 | 1627 | kvm_immediate_exit = kvm_check_extension(s, KVM_CAP_IMMEDIATE_EXIT); |
fb541ca5 AW |
1628 | s->nr_slots = kvm_check_extension(s, KVM_CAP_NR_MEMSLOTS); |
1629 | ||
1630 | /* If unspecified, use the default value */ | |
1631 | if (!s->nr_slots) { | |
1632 | s->nr_slots = 32; | |
1633 | } | |
1634 | ||
670436ce AJ |
1635 | /* check the vcpu limits */ |
1636 | soft_vcpus_limit = kvm_recommended_vcpus(s); | |
1637 | hard_vcpus_limit = kvm_max_vcpus(s); | |
3ed444e9 | 1638 | |
670436ce AJ |
1639 | while (nc->name) { |
1640 | if (nc->num > soft_vcpus_limit) { | |
1641 | fprintf(stderr, | |
1642 | "Warning: Number of %s cpus requested (%d) exceeds " | |
1643 | "the recommended cpus supported by KVM (%d)\n", | |
1644 | nc->name, nc->num, soft_vcpus_limit); | |
1645 | ||
1646 | if (nc->num > hard_vcpus_limit) { | |
670436ce AJ |
1647 | fprintf(stderr, "Number of %s cpus requested (%d) exceeds " |
1648 | "the maximum cpus supported by KVM (%d)\n", | |
1649 | nc->name, nc->num, hard_vcpus_limit); | |
9ba3cf54 | 1650 | exit(1); |
670436ce AJ |
1651 | } |
1652 | } | |
1653 | nc++; | |
7dc52526 MT |
1654 | } |
1655 | ||
135a129a | 1656 | kvm_type = qemu_opt_get(qemu_get_machine_opts(), "kvm-type"); |
f1e29879 MA |
1657 | if (mc->kvm_type) { |
1658 | type = mc->kvm_type(kvm_type); | |
135a129a | 1659 | } else if (kvm_type) { |
0e1dac6c | 1660 | ret = -EINVAL; |
135a129a AK |
1661 | fprintf(stderr, "Invalid argument kvm-type=%s\n", kvm_type); |
1662 | goto err; | |
1663 | } | |
1664 | ||
94ccff13 | 1665 | do { |
135a129a | 1666 | ret = kvm_ioctl(s, KVM_CREATE_VM, type); |
94ccff13 TK |
1667 | } while (ret == -EINTR); |
1668 | ||
1669 | if (ret < 0) { | |
521f438e | 1670 | fprintf(stderr, "ioctl(KVM_CREATE_VM) failed: %d %s\n", -ret, |
94ccff13 TK |
1671 | strerror(-ret)); |
1672 | ||
0104dcac | 1673 | #ifdef TARGET_S390X |
2c80e996 CH |
1674 | if (ret == -EINVAL) { |
1675 | fprintf(stderr, | |
1676 | "Host kernel setup problem detected. Please verify:\n"); | |
1677 | fprintf(stderr, "- for kernels supporting the switch_amode or" | |
1678 | " user_mode parameters, whether\n"); | |
1679 | fprintf(stderr, | |
1680 | " user space is running in primary address space\n"); | |
1681 | fprintf(stderr, | |
1682 | "- for kernels supporting the vm.allocate_pgste sysctl, " | |
1683 | "whether it is enabled\n"); | |
1684 | } | |
0104dcac | 1685 | #endif |
05330448 | 1686 | goto err; |
0104dcac | 1687 | } |
05330448 | 1688 | |
94ccff13 | 1689 | s->vmfd = ret; |
94a8d39a JK |
1690 | missing_cap = kvm_check_extension_list(s, kvm_required_capabilites); |
1691 | if (!missing_cap) { | |
1692 | missing_cap = | |
1693 | kvm_check_extension_list(s, kvm_arch_required_capabilities); | |
05330448 | 1694 | } |
94a8d39a | 1695 | if (missing_cap) { |
ad7b8b33 | 1696 | ret = -EINVAL; |
94a8d39a JK |
1697 | fprintf(stderr, "kvm does not support %s\n%s", |
1698 | missing_cap->name, upgrade_note); | |
d85dc283 AL |
1699 | goto err; |
1700 | } | |
1701 | ||
ad7b8b33 | 1702 | s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO); |
f65ed4c1 | 1703 | |
e69917e2 | 1704 | s->broken_set_mem_region = 1; |
14a09518 | 1705 | ret = kvm_check_extension(s, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS); |
e69917e2 JK |
1706 | if (ret > 0) { |
1707 | s->broken_set_mem_region = 0; | |
1708 | } | |
e69917e2 | 1709 | |
a0fb002c JK |
1710 | #ifdef KVM_CAP_VCPU_EVENTS |
1711 | s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS); | |
1712 | #endif | |
1713 | ||
b0b1d690 JK |
1714 | s->robust_singlestep = |
1715 | kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP); | |
b0b1d690 | 1716 | |
ff44f1a3 JK |
1717 | #ifdef KVM_CAP_DEBUGREGS |
1718 | s->debugregs = kvm_check_extension(s, KVM_CAP_DEBUGREGS); | |
1719 | #endif | |
1720 | ||
d3d3bef0 | 1721 | #ifdef KVM_CAP_IRQ_ROUTING |
50bf31b9 | 1722 | kvm_direct_msi_allowed = (kvm_check_extension(s, KVM_CAP_SIGNAL_MSI) > 0); |
d3d3bef0 | 1723 | #endif |
4a3adebb | 1724 | |
3ab73842 JK |
1725 | s->intx_set_mask = kvm_check_extension(s, KVM_CAP_PCI_2_3); |
1726 | ||
e333cd69 | 1727 | s->irq_set_ioctl = KVM_IRQ_LINE; |
8732fbd2 | 1728 | if (kvm_check_extension(s, KVM_CAP_IRQ_INJECT_STATUS)) { |
e333cd69 | 1729 | s->irq_set_ioctl = KVM_IRQ_LINE_STATUS; |
8732fbd2 PM |
1730 | } |
1731 | ||
df9c8b75 JJ |
1732 | #ifdef KVM_CAP_READONLY_MEM |
1733 | kvm_readonly_mem_allowed = | |
1734 | (kvm_check_extension(s, KVM_CAP_READONLY_MEM) > 0); | |
1735 | #endif | |
1736 | ||
69e03ae6 NN |
1737 | kvm_eventfds_allowed = |
1738 | (kvm_check_extension(s, KVM_CAP_IOEVENTFD) > 0); | |
1739 | ||
f41389ae EA |
1740 | kvm_irqfds_allowed = |
1741 | (kvm_check_extension(s, KVM_CAP_IRQFD) > 0); | |
1742 | ||
1743 | kvm_resamplefds_allowed = | |
1744 | (kvm_check_extension(s, KVM_CAP_IRQFD_RESAMPLE) > 0); | |
1745 | ||
d0a073a1 DD |
1746 | kvm_vm_attributes_allowed = |
1747 | (kvm_check_extension(s, KVM_CAP_VM_ATTRIBUTES) > 0); | |
1748 | ||
35108223 JW |
1749 | kvm_ioeventfd_any_length_allowed = |
1750 | (kvm_check_extension(s, KVM_CAP_IOEVENTFD_ANY_LENGTH) > 0); | |
1751 | ||
d870cfde GA |
1752 | kvm_state = s; |
1753 | ||
b16565b3 | 1754 | ret = kvm_arch_init(ms, s); |
a426e122 | 1755 | if (ret < 0) { |
05330448 | 1756 | goto err; |
a426e122 | 1757 | } |
05330448 | 1758 | |
8db4936b PB |
1759 | if (machine_kernel_irqchip_allowed(ms)) { |
1760 | kvm_irqchip_create(ms, s); | |
84b058d7 JK |
1761 | } |
1762 | ||
8c56c1a5 PF |
1763 | if (kvm_eventfds_allowed) { |
1764 | s->memory_listener.listener.eventfd_add = kvm_mem_ioeventfd_add; | |
1765 | s->memory_listener.listener.eventfd_del = kvm_mem_ioeventfd_del; | |
1766 | } | |
7bbda04c PB |
1767 | s->memory_listener.listener.coalesced_mmio_add = kvm_coalesce_mmio_region; |
1768 | s->memory_listener.listener.coalesced_mmio_del = kvm_uncoalesce_mmio_region; | |
1769 | ||
1770 | kvm_memory_listener_register(s, &s->memory_listener, | |
38bfe691 | 1771 | &address_space_memory, 0); |
7bbda04c PB |
1772 | memory_listener_register(&kvm_io_listener, |
1773 | &address_space_io); | |
05330448 | 1774 | |
d2f2b8a7 SH |
1775 | s->many_ioeventfds = kvm_check_many_ioeventfds(); |
1776 | ||
aa7f74d1 JK |
1777 | cpu_interrupt_handler = kvm_handle_interrupt; |
1778 | ||
05330448 AL |
1779 | return 0; |
1780 | ||
1781 | err: | |
0e1dac6c | 1782 | assert(ret < 0); |
6d1cc321 SW |
1783 | if (s->vmfd >= 0) { |
1784 | close(s->vmfd); | |
1785 | } | |
1786 | if (s->fd != -1) { | |
1787 | close(s->fd); | |
05330448 | 1788 | } |
7bbda04c | 1789 | g_free(s->memory_listener.slots); |
05330448 AL |
1790 | |
1791 | return ret; | |
1792 | } | |
1793 | ||
aed6efb9 JH |
1794 | void kvm_set_sigmask_len(KVMState *s, unsigned int sigmask_len) |
1795 | { | |
1796 | s->sigmask_len = sigmask_len; | |
1797 | } | |
1798 | ||
4c663752 PB |
1799 | static void kvm_handle_io(uint16_t port, MemTxAttrs attrs, void *data, int direction, |
1800 | int size, uint32_t count) | |
05330448 AL |
1801 | { |
1802 | int i; | |
1803 | uint8_t *ptr = data; | |
1804 | ||
1805 | for (i = 0; i < count; i++) { | |
4c663752 | 1806 | address_space_rw(&address_space_io, port, attrs, |
5c9eb028 | 1807 | ptr, size, |
354678c5 | 1808 | direction == KVM_EXIT_IO_OUT); |
05330448 AL |
1809 | ptr += size; |
1810 | } | |
05330448 AL |
1811 | } |
1812 | ||
5326ab55 | 1813 | static int kvm_handle_internal_error(CPUState *cpu, struct kvm_run *run) |
7c80eef8 | 1814 | { |
977c7b6d RK |
1815 | fprintf(stderr, "KVM internal error. Suberror: %d\n", |
1816 | run->internal.suberror); | |
1817 | ||
7c80eef8 MT |
1818 | if (kvm_check_extension(kvm_state, KVM_CAP_INTERNAL_ERROR_DATA)) { |
1819 | int i; | |
1820 | ||
7c80eef8 MT |
1821 | for (i = 0; i < run->internal.ndata; ++i) { |
1822 | fprintf(stderr, "extra data[%d]: %"PRIx64"\n", | |
1823 | i, (uint64_t)run->internal.data[i]); | |
1824 | } | |
1825 | } | |
7c80eef8 MT |
1826 | if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) { |
1827 | fprintf(stderr, "emulation failure\n"); | |
20d695a9 | 1828 | if (!kvm_arch_stop_on_emulation_error(cpu)) { |
878096ee | 1829 | cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE); |
d73cd8f4 | 1830 | return EXCP_INTERRUPT; |
a426e122 | 1831 | } |
7c80eef8 MT |
1832 | } |
1833 | /* FIXME: Should trigger a qmp message to let management know | |
1834 | * something went wrong. | |
1835 | */ | |
73aaec4a | 1836 | return -1; |
7c80eef8 | 1837 | } |
7c80eef8 | 1838 | |
62a2744c | 1839 | void kvm_flush_coalesced_mmio_buffer(void) |
f65ed4c1 | 1840 | { |
f65ed4c1 | 1841 | KVMState *s = kvm_state; |
1cae88b9 AK |
1842 | |
1843 | if (s->coalesced_flush_in_progress) { | |
1844 | return; | |
1845 | } | |
1846 | ||
1847 | s->coalesced_flush_in_progress = true; | |
1848 | ||
62a2744c SY |
1849 | if (s->coalesced_mmio_ring) { |
1850 | struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring; | |
f65ed4c1 AL |
1851 | while (ring->first != ring->last) { |
1852 | struct kvm_coalesced_mmio *ent; | |
1853 | ||
1854 | ent = &ring->coalesced_mmio[ring->first]; | |
1855 | ||
1856 | cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len); | |
85199474 | 1857 | smp_wmb(); |
f65ed4c1 AL |
1858 | ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX; |
1859 | } | |
1860 | } | |
1cae88b9 AK |
1861 | |
1862 | s->coalesced_flush_in_progress = false; | |
f65ed4c1 AL |
1863 | } |
1864 | ||
14e6fe12 | 1865 | static void do_kvm_cpu_synchronize_state(CPUState *cpu, run_on_cpu_data arg) |
4c0960c0 | 1866 | { |
20d695a9 AF |
1867 | if (!cpu->kvm_vcpu_dirty) { |
1868 | kvm_arch_get_registers(cpu); | |
1869 | cpu->kvm_vcpu_dirty = true; | |
4c0960c0 AK |
1870 | } |
1871 | } | |
1872 | ||
dd1750d7 | 1873 | void kvm_cpu_synchronize_state(CPUState *cpu) |
2705d56a | 1874 | { |
20d695a9 | 1875 | if (!cpu->kvm_vcpu_dirty) { |
14e6fe12 | 1876 | run_on_cpu(cpu, do_kvm_cpu_synchronize_state, RUN_ON_CPU_NULL); |
a426e122 | 1877 | } |
2705d56a JK |
1878 | } |
1879 | ||
14e6fe12 | 1880 | static void do_kvm_cpu_synchronize_post_reset(CPUState *cpu, run_on_cpu_data arg) |
ea375f9a | 1881 | { |
20d695a9 AF |
1882 | kvm_arch_put_registers(cpu, KVM_PUT_RESET_STATE); |
1883 | cpu->kvm_vcpu_dirty = false; | |
ea375f9a JK |
1884 | } |
1885 | ||
c8e2085d DH |
1886 | void kvm_cpu_synchronize_post_reset(CPUState *cpu) |
1887 | { | |
14e6fe12 | 1888 | run_on_cpu(cpu, do_kvm_cpu_synchronize_post_reset, RUN_ON_CPU_NULL); |
c8e2085d DH |
1889 | } |
1890 | ||
14e6fe12 | 1891 | static void do_kvm_cpu_synchronize_post_init(CPUState *cpu, run_on_cpu_data arg) |
ea375f9a | 1892 | { |
20d695a9 AF |
1893 | kvm_arch_put_registers(cpu, KVM_PUT_FULL_STATE); |
1894 | cpu->kvm_vcpu_dirty = false; | |
ea375f9a JK |
1895 | } |
1896 | ||
c8e2085d DH |
1897 | void kvm_cpu_synchronize_post_init(CPUState *cpu) |
1898 | { | |
14e6fe12 | 1899 | run_on_cpu(cpu, do_kvm_cpu_synchronize_post_init, RUN_ON_CPU_NULL); |
c8e2085d DH |
1900 | } |
1901 | ||
75e972da DG |
1902 | static void do_kvm_cpu_synchronize_pre_loadvm(CPUState *cpu, run_on_cpu_data arg) |
1903 | { | |
1904 | cpu->kvm_vcpu_dirty = true; | |
1905 | } | |
1906 | ||
1907 | void kvm_cpu_synchronize_pre_loadvm(CPUState *cpu) | |
1908 | { | |
1909 | run_on_cpu(cpu, do_kvm_cpu_synchronize_pre_loadvm, RUN_ON_CPU_NULL); | |
1910 | } | |
1911 | ||
2ae41db2 PB |
1912 | #ifdef KVM_HAVE_MCE_INJECTION |
1913 | static __thread void *pending_sigbus_addr; | |
1914 | static __thread int pending_sigbus_code; | |
1915 | static __thread bool have_sigbus_pending; | |
1916 | #endif | |
1917 | ||
cf0f7cf9 PB |
1918 | static void kvm_cpu_kick(CPUState *cpu) |
1919 | { | |
1920 | atomic_set(&cpu->kvm_run->immediate_exit, 1); | |
1921 | } | |
1922 | ||
1923 | static void kvm_cpu_kick_self(void) | |
1924 | { | |
1925 | if (kvm_immediate_exit) { | |
1926 | kvm_cpu_kick(current_cpu); | |
1927 | } else { | |
1928 | qemu_cpu_kick_self(); | |
1929 | } | |
1930 | } | |
1931 | ||
18268b60 PB |
1932 | static void kvm_eat_signals(CPUState *cpu) |
1933 | { | |
1934 | struct timespec ts = { 0, 0 }; | |
1935 | siginfo_t siginfo; | |
1936 | sigset_t waitset; | |
1937 | sigset_t chkset; | |
1938 | int r; | |
1939 | ||
cf0f7cf9 PB |
1940 | if (kvm_immediate_exit) { |
1941 | atomic_set(&cpu->kvm_run->immediate_exit, 0); | |
1942 | /* Write kvm_run->immediate_exit before the cpu->exit_request | |
1943 | * write in kvm_cpu_exec. | |
1944 | */ | |
1945 | smp_wmb(); | |
1946 | return; | |
1947 | } | |
1948 | ||
18268b60 PB |
1949 | sigemptyset(&waitset); |
1950 | sigaddset(&waitset, SIG_IPI); | |
1951 | ||
1952 | do { | |
1953 | r = sigtimedwait(&waitset, &siginfo, &ts); | |
1954 | if (r == -1 && !(errno == EAGAIN || errno == EINTR)) { | |
1955 | perror("sigtimedwait"); | |
1956 | exit(1); | |
1957 | } | |
1958 | ||
1959 | r = sigpending(&chkset); | |
1960 | if (r == -1) { | |
1961 | perror("sigpending"); | |
1962 | exit(1); | |
1963 | } | |
1964 | } while (sigismember(&chkset, SIG_IPI)); | |
1965 | } | |
1966 | ||
1458c363 | 1967 | int kvm_cpu_exec(CPUState *cpu) |
05330448 | 1968 | { |
f7575c96 | 1969 | struct kvm_run *run = cpu->kvm_run; |
7cbb533f | 1970 | int ret, run_ret; |
05330448 | 1971 | |
8c0d577e | 1972 | DPRINTF("kvm_cpu_exec()\n"); |
05330448 | 1973 | |
20d695a9 | 1974 | if (kvm_arch_process_async_events(cpu)) { |
c5c6679d | 1975 | atomic_set(&cpu->exit_request, 0); |
6792a57b | 1976 | return EXCP_HLT; |
9ccfac9e | 1977 | } |
0af691d7 | 1978 | |
4b8523ee JK |
1979 | qemu_mutex_unlock_iothread(); |
1980 | ||
9ccfac9e | 1981 | do { |
4c663752 PB |
1982 | MemTxAttrs attrs; |
1983 | ||
20d695a9 AF |
1984 | if (cpu->kvm_vcpu_dirty) { |
1985 | kvm_arch_put_registers(cpu, KVM_PUT_RUNTIME_STATE); | |
1986 | cpu->kvm_vcpu_dirty = false; | |
4c0960c0 AK |
1987 | } |
1988 | ||
20d695a9 | 1989 | kvm_arch_pre_run(cpu, run); |
c5c6679d | 1990 | if (atomic_read(&cpu->exit_request)) { |
9ccfac9e JK |
1991 | DPRINTF("interrupt exit requested\n"); |
1992 | /* | |
1993 | * KVM requires us to reenter the kernel after IO exits to complete | |
1994 | * instruction emulation. This self-signal will ensure that we | |
1995 | * leave ASAP again. | |
1996 | */ | |
cf0f7cf9 | 1997 | kvm_cpu_kick_self(); |
9ccfac9e | 1998 | } |
9ccfac9e | 1999 | |
cf0f7cf9 PB |
2000 | /* Read cpu->exit_request before KVM_RUN reads run->immediate_exit. |
2001 | * Matching barrier in kvm_eat_signals. | |
2002 | */ | |
2003 | smp_rmb(); | |
2004 | ||
1bc22652 | 2005 | run_ret = kvm_vcpu_ioctl(cpu, KVM_RUN, 0); |
9ccfac9e | 2006 | |
4c663752 | 2007 | attrs = kvm_arch_post_run(cpu, run); |
05330448 | 2008 | |
2ae41db2 PB |
2009 | #ifdef KVM_HAVE_MCE_INJECTION |
2010 | if (unlikely(have_sigbus_pending)) { | |
2011 | qemu_mutex_lock_iothread(); | |
2012 | kvm_arch_on_sigbus_vcpu(cpu, pending_sigbus_code, | |
2013 | pending_sigbus_addr); | |
2014 | have_sigbus_pending = false; | |
2015 | qemu_mutex_unlock_iothread(); | |
2016 | } | |
2017 | #endif | |
2018 | ||
7cbb533f | 2019 | if (run_ret < 0) { |
dc77d341 JK |
2020 | if (run_ret == -EINTR || run_ret == -EAGAIN) { |
2021 | DPRINTF("io window exit\n"); | |
18268b60 | 2022 | kvm_eat_signals(cpu); |
d73cd8f4 | 2023 | ret = EXCP_INTERRUPT; |
dc77d341 JK |
2024 | break; |
2025 | } | |
7b011fbc ME |
2026 | fprintf(stderr, "error: kvm run failed %s\n", |
2027 | strerror(-run_ret)); | |
dae02ba5 LV |
2028 | #ifdef TARGET_PPC |
2029 | if (run_ret == -EBUSY) { | |
2030 | fprintf(stderr, | |
2031 | "This is probably because your SMT is enabled.\n" | |
2032 | "VCPU can only run on primary threads with all " | |
2033 | "secondary threads offline.\n"); | |
2034 | } | |
2035 | #endif | |
a85e130e PB |
2036 | ret = -1; |
2037 | break; | |
05330448 AL |
2038 | } |
2039 | ||
b76ac80a | 2040 | trace_kvm_run_exit(cpu->cpu_index, run->exit_reason); |
05330448 AL |
2041 | switch (run->exit_reason) { |
2042 | case KVM_EXIT_IO: | |
8c0d577e | 2043 | DPRINTF("handle_io\n"); |
80b7d2ef | 2044 | /* Called outside BQL */ |
4c663752 | 2045 | kvm_handle_io(run->io.port, attrs, |
b30e93e9 JK |
2046 | (uint8_t *)run + run->io.data_offset, |
2047 | run->io.direction, | |
2048 | run->io.size, | |
2049 | run->io.count); | |
d73cd8f4 | 2050 | ret = 0; |
05330448 AL |
2051 | break; |
2052 | case KVM_EXIT_MMIO: | |
8c0d577e | 2053 | DPRINTF("handle_mmio\n"); |
de7ea885 | 2054 | /* Called outside BQL */ |
4c663752 PB |
2055 | address_space_rw(&address_space_memory, |
2056 | run->mmio.phys_addr, attrs, | |
2057 | run->mmio.data, | |
2058 | run->mmio.len, | |
2059 | run->mmio.is_write); | |
d73cd8f4 | 2060 | ret = 0; |
05330448 AL |
2061 | break; |
2062 | case KVM_EXIT_IRQ_WINDOW_OPEN: | |
8c0d577e | 2063 | DPRINTF("irq_window_open\n"); |
d73cd8f4 | 2064 | ret = EXCP_INTERRUPT; |
05330448 AL |
2065 | break; |
2066 | case KVM_EXIT_SHUTDOWN: | |
8c0d577e | 2067 | DPRINTF("shutdown\n"); |
cf83f140 | 2068 | qemu_system_reset_request(SHUTDOWN_CAUSE_GUEST_RESET); |
d73cd8f4 | 2069 | ret = EXCP_INTERRUPT; |
05330448 AL |
2070 | break; |
2071 | case KVM_EXIT_UNKNOWN: | |
bb44e0d1 JK |
2072 | fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "\n", |
2073 | (uint64_t)run->hw.hardware_exit_reason); | |
73aaec4a | 2074 | ret = -1; |
05330448 | 2075 | break; |
7c80eef8 | 2076 | case KVM_EXIT_INTERNAL_ERROR: |
5326ab55 | 2077 | ret = kvm_handle_internal_error(cpu, run); |
7c80eef8 | 2078 | break; |
99040447 PS |
2079 | case KVM_EXIT_SYSTEM_EVENT: |
2080 | switch (run->system_event.type) { | |
2081 | case KVM_SYSTEM_EVENT_SHUTDOWN: | |
cf83f140 | 2082 | qemu_system_shutdown_request(SHUTDOWN_CAUSE_GUEST_SHUTDOWN); |
99040447 PS |
2083 | ret = EXCP_INTERRUPT; |
2084 | break; | |
2085 | case KVM_SYSTEM_EVENT_RESET: | |
cf83f140 | 2086 | qemu_system_reset_request(SHUTDOWN_CAUSE_GUEST_RESET); |
99040447 PS |
2087 | ret = EXCP_INTERRUPT; |
2088 | break; | |
7c207b90 | 2089 | case KVM_SYSTEM_EVENT_CRASH: |
d187e08d | 2090 | kvm_cpu_synchronize_state(cpu); |
7c207b90 | 2091 | qemu_mutex_lock_iothread(); |
c86f106b | 2092 | qemu_system_guest_panicked(cpu_get_crash_info(cpu)); |
7c207b90 AS |
2093 | qemu_mutex_unlock_iothread(); |
2094 | ret = 0; | |
2095 | break; | |
99040447 PS |
2096 | default: |
2097 | DPRINTF("kvm_arch_handle_exit\n"); | |
2098 | ret = kvm_arch_handle_exit(cpu, run); | |
2099 | break; | |
2100 | } | |
2101 | break; | |
05330448 | 2102 | default: |
8c0d577e | 2103 | DPRINTF("kvm_arch_handle_exit\n"); |
20d695a9 | 2104 | ret = kvm_arch_handle_exit(cpu, run); |
05330448 AL |
2105 | break; |
2106 | } | |
d73cd8f4 | 2107 | } while (ret == 0); |
05330448 | 2108 | |
4b8523ee JK |
2109 | qemu_mutex_lock_iothread(); |
2110 | ||
73aaec4a | 2111 | if (ret < 0) { |
878096ee | 2112 | cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE); |
0461d5a6 | 2113 | vm_stop(RUN_STATE_INTERNAL_ERROR); |
becfc390 AL |
2114 | } |
2115 | ||
c5c6679d | 2116 | atomic_set(&cpu->exit_request, 0); |
05330448 AL |
2117 | return ret; |
2118 | } | |
2119 | ||
984b5181 | 2120 | int kvm_ioctl(KVMState *s, int type, ...) |
05330448 AL |
2121 | { |
2122 | int ret; | |
984b5181 AL |
2123 | void *arg; |
2124 | va_list ap; | |
05330448 | 2125 | |
984b5181 AL |
2126 | va_start(ap, type); |
2127 | arg = va_arg(ap, void *); | |
2128 | va_end(ap); | |
2129 | ||
9c775729 | 2130 | trace_kvm_ioctl(type, arg); |
984b5181 | 2131 | ret = ioctl(s->fd, type, arg); |
a426e122 | 2132 | if (ret == -1) { |
05330448 | 2133 | ret = -errno; |
a426e122 | 2134 | } |
05330448 AL |
2135 | return ret; |
2136 | } | |
2137 | ||
984b5181 | 2138 | int kvm_vm_ioctl(KVMState *s, int type, ...) |
05330448 AL |
2139 | { |
2140 | int ret; | |
984b5181 AL |
2141 | void *arg; |
2142 | va_list ap; | |
2143 | ||
2144 | va_start(ap, type); | |
2145 | arg = va_arg(ap, void *); | |
2146 | va_end(ap); | |
05330448 | 2147 | |
9c775729 | 2148 | trace_kvm_vm_ioctl(type, arg); |
984b5181 | 2149 | ret = ioctl(s->vmfd, type, arg); |
a426e122 | 2150 | if (ret == -1) { |
05330448 | 2151 | ret = -errno; |
a426e122 | 2152 | } |
05330448 AL |
2153 | return ret; |
2154 | } | |
2155 | ||
1bc22652 | 2156 | int kvm_vcpu_ioctl(CPUState *cpu, int type, ...) |
05330448 AL |
2157 | { |
2158 | int ret; | |
984b5181 AL |
2159 | void *arg; |
2160 | va_list ap; | |
2161 | ||
2162 | va_start(ap, type); | |
2163 | arg = va_arg(ap, void *); | |
2164 | va_end(ap); | |
05330448 | 2165 | |
9c775729 | 2166 | trace_kvm_vcpu_ioctl(cpu->cpu_index, type, arg); |
8737c51c | 2167 | ret = ioctl(cpu->kvm_fd, type, arg); |
a426e122 | 2168 | if (ret == -1) { |
05330448 | 2169 | ret = -errno; |
a426e122 | 2170 | } |
05330448 AL |
2171 | return ret; |
2172 | } | |
bd322087 | 2173 | |
0a6a7cca CD |
2174 | int kvm_device_ioctl(int fd, int type, ...) |
2175 | { | |
2176 | int ret; | |
2177 | void *arg; | |
2178 | va_list ap; | |
2179 | ||
2180 | va_start(ap, type); | |
2181 | arg = va_arg(ap, void *); | |
2182 | va_end(ap); | |
2183 | ||
2184 | trace_kvm_device_ioctl(fd, type, arg); | |
2185 | ret = ioctl(fd, type, arg); | |
2186 | if (ret == -1) { | |
2187 | ret = -errno; | |
2188 | } | |
2189 | return ret; | |
2190 | } | |
2191 | ||
d0a073a1 DD |
2192 | int kvm_vm_check_attr(KVMState *s, uint32_t group, uint64_t attr) |
2193 | { | |
2194 | int ret; | |
2195 | struct kvm_device_attr attribute = { | |
2196 | .group = group, | |
2197 | .attr = attr, | |
2198 | }; | |
2199 | ||
2200 | if (!kvm_vm_attributes_allowed) { | |
2201 | return 0; | |
2202 | } | |
2203 | ||
2204 | ret = kvm_vm_ioctl(s, KVM_HAS_DEVICE_ATTR, &attribute); | |
2205 | /* kvm returns 0 on success for HAS_DEVICE_ATTR */ | |
2206 | return ret ? 0 : 1; | |
2207 | } | |
2208 | ||
4b3cfe72 PF |
2209 | int kvm_device_check_attr(int dev_fd, uint32_t group, uint64_t attr) |
2210 | { | |
2211 | struct kvm_device_attr attribute = { | |
2212 | .group = group, | |
2213 | .attr = attr, | |
2214 | .flags = 0, | |
2215 | }; | |
2216 | ||
2217 | return kvm_device_ioctl(dev_fd, KVM_HAS_DEVICE_ATTR, &attribute) ? 0 : 1; | |
2218 | } | |
2219 | ||
556969e9 EA |
2220 | int kvm_device_access(int fd, int group, uint64_t attr, |
2221 | void *val, bool write, Error **errp) | |
4b3cfe72 PF |
2222 | { |
2223 | struct kvm_device_attr kvmattr; | |
2224 | int err; | |
2225 | ||
2226 | kvmattr.flags = 0; | |
2227 | kvmattr.group = group; | |
2228 | kvmattr.attr = attr; | |
2229 | kvmattr.addr = (uintptr_t)val; | |
2230 | ||
2231 | err = kvm_device_ioctl(fd, | |
2232 | write ? KVM_SET_DEVICE_ATTR : KVM_GET_DEVICE_ATTR, | |
2233 | &kvmattr); | |
2234 | if (err < 0) { | |
556969e9 EA |
2235 | error_setg_errno(errp, -err, |
2236 | "KVM_%s_DEVICE_ATTR failed: Group %d " | |
2237 | "attr 0x%016" PRIx64, | |
2238 | write ? "SET" : "GET", group, attr); | |
4b3cfe72 | 2239 | } |
556969e9 | 2240 | return err; |
4b3cfe72 PF |
2241 | } |
2242 | ||
c2cd627d | 2243 | /* Return 1 on success, 0 on failure */ |
bd322087 AL |
2244 | int kvm_has_sync_mmu(void) |
2245 | { | |
94a8d39a | 2246 | return kvm_check_extension(kvm_state, KVM_CAP_SYNC_MMU); |
bd322087 | 2247 | } |
e22a25c9 | 2248 | |
a0fb002c JK |
2249 | int kvm_has_vcpu_events(void) |
2250 | { | |
2251 | return kvm_state->vcpu_events; | |
2252 | } | |
2253 | ||
b0b1d690 JK |
2254 | int kvm_has_robust_singlestep(void) |
2255 | { | |
2256 | return kvm_state->robust_singlestep; | |
2257 | } | |
2258 | ||
ff44f1a3 JK |
2259 | int kvm_has_debugregs(void) |
2260 | { | |
2261 | return kvm_state->debugregs; | |
2262 | } | |
2263 | ||
d2f2b8a7 SH |
2264 | int kvm_has_many_ioeventfds(void) |
2265 | { | |
2266 | if (!kvm_enabled()) { | |
2267 | return 0; | |
2268 | } | |
2269 | return kvm_state->many_ioeventfds; | |
2270 | } | |
2271 | ||
84b058d7 JK |
2272 | int kvm_has_gsi_routing(void) |
2273 | { | |
a9c5eb0d | 2274 | #ifdef KVM_CAP_IRQ_ROUTING |
84b058d7 | 2275 | return kvm_check_extension(kvm_state, KVM_CAP_IRQ_ROUTING); |
a9c5eb0d AG |
2276 | #else |
2277 | return false; | |
2278 | #endif | |
84b058d7 JK |
2279 | } |
2280 | ||
3ab73842 JK |
2281 | int kvm_has_intx_set_mask(void) |
2282 | { | |
2283 | return kvm_state->intx_set_mask; | |
2284 | } | |
2285 | ||
e22a25c9 | 2286 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
a60f24b5 | 2287 | struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *cpu, |
e22a25c9 AL |
2288 | target_ulong pc) |
2289 | { | |
2290 | struct kvm_sw_breakpoint *bp; | |
2291 | ||
a60f24b5 | 2292 | QTAILQ_FOREACH(bp, &cpu->kvm_state->kvm_sw_breakpoints, entry) { |
a426e122 | 2293 | if (bp->pc == pc) { |
e22a25c9 | 2294 | return bp; |
a426e122 | 2295 | } |
e22a25c9 AL |
2296 | } |
2297 | return NULL; | |
2298 | } | |
2299 | ||
a60f24b5 | 2300 | int kvm_sw_breakpoints_active(CPUState *cpu) |
e22a25c9 | 2301 | { |
a60f24b5 | 2302 | return !QTAILQ_EMPTY(&cpu->kvm_state->kvm_sw_breakpoints); |
e22a25c9 AL |
2303 | } |
2304 | ||
452e4751 GC |
2305 | struct kvm_set_guest_debug_data { |
2306 | struct kvm_guest_debug dbg; | |
452e4751 GC |
2307 | int err; |
2308 | }; | |
2309 | ||
14e6fe12 | 2310 | static void kvm_invoke_set_guest_debug(CPUState *cpu, run_on_cpu_data data) |
452e4751 | 2311 | { |
14e6fe12 PB |
2312 | struct kvm_set_guest_debug_data *dbg_data = |
2313 | (struct kvm_set_guest_debug_data *) data.host_ptr; | |
b3807725 | 2314 | |
3c0ed2a3 | 2315 | dbg_data->err = kvm_vcpu_ioctl(cpu, KVM_SET_GUEST_DEBUG, |
a60f24b5 | 2316 | &dbg_data->dbg); |
452e4751 GC |
2317 | } |
2318 | ||
38e478ec | 2319 | int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap) |
e22a25c9 | 2320 | { |
452e4751 | 2321 | struct kvm_set_guest_debug_data data; |
e22a25c9 | 2322 | |
b0b1d690 | 2323 | data.dbg.control = reinject_trap; |
e22a25c9 | 2324 | |
ed2803da | 2325 | if (cpu->singlestep_enabled) { |
b0b1d690 JK |
2326 | data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP; |
2327 | } | |
20d695a9 | 2328 | kvm_arch_update_guest_debug(cpu, &data.dbg); |
e22a25c9 | 2329 | |
14e6fe12 PB |
2330 | run_on_cpu(cpu, kvm_invoke_set_guest_debug, |
2331 | RUN_ON_CPU_HOST_PTR(&data)); | |
452e4751 | 2332 | return data.err; |
e22a25c9 AL |
2333 | } |
2334 | ||
62278814 | 2335 | int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr, |
e22a25c9 AL |
2336 | target_ulong len, int type) |
2337 | { | |
2338 | struct kvm_sw_breakpoint *bp; | |
e22a25c9 AL |
2339 | int err; |
2340 | ||
2341 | if (type == GDB_BREAKPOINT_SW) { | |
80b7cd73 | 2342 | bp = kvm_find_sw_breakpoint(cpu, addr); |
e22a25c9 AL |
2343 | if (bp) { |
2344 | bp->use_count++; | |
2345 | return 0; | |
2346 | } | |
2347 | ||
7267c094 | 2348 | bp = g_malloc(sizeof(struct kvm_sw_breakpoint)); |
e22a25c9 AL |
2349 | bp->pc = addr; |
2350 | bp->use_count = 1; | |
80b7cd73 | 2351 | err = kvm_arch_insert_sw_breakpoint(cpu, bp); |
e22a25c9 | 2352 | if (err) { |
7267c094 | 2353 | g_free(bp); |
e22a25c9 AL |
2354 | return err; |
2355 | } | |
2356 | ||
80b7cd73 | 2357 | QTAILQ_INSERT_HEAD(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry); |
e22a25c9 AL |
2358 | } else { |
2359 | err = kvm_arch_insert_hw_breakpoint(addr, len, type); | |
a426e122 | 2360 | if (err) { |
e22a25c9 | 2361 | return err; |
a426e122 | 2362 | } |
e22a25c9 AL |
2363 | } |
2364 | ||
bdc44640 | 2365 | CPU_FOREACH(cpu) { |
38e478ec | 2366 | err = kvm_update_guest_debug(cpu, 0); |
a426e122 | 2367 | if (err) { |
e22a25c9 | 2368 | return err; |
a426e122 | 2369 | } |
e22a25c9 AL |
2370 | } |
2371 | return 0; | |
2372 | } | |
2373 | ||
62278814 | 2374 | int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr, |
e22a25c9 AL |
2375 | target_ulong len, int type) |
2376 | { | |
2377 | struct kvm_sw_breakpoint *bp; | |
e22a25c9 AL |
2378 | int err; |
2379 | ||
2380 | if (type == GDB_BREAKPOINT_SW) { | |
80b7cd73 | 2381 | bp = kvm_find_sw_breakpoint(cpu, addr); |
a426e122 | 2382 | if (!bp) { |
e22a25c9 | 2383 | return -ENOENT; |
a426e122 | 2384 | } |
e22a25c9 AL |
2385 | |
2386 | if (bp->use_count > 1) { | |
2387 | bp->use_count--; | |
2388 | return 0; | |
2389 | } | |
2390 | ||
80b7cd73 | 2391 | err = kvm_arch_remove_sw_breakpoint(cpu, bp); |
a426e122 | 2392 | if (err) { |
e22a25c9 | 2393 | return err; |
a426e122 | 2394 | } |
e22a25c9 | 2395 | |
80b7cd73 | 2396 | QTAILQ_REMOVE(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry); |
7267c094 | 2397 | g_free(bp); |
e22a25c9 AL |
2398 | } else { |
2399 | err = kvm_arch_remove_hw_breakpoint(addr, len, type); | |
a426e122 | 2400 | if (err) { |
e22a25c9 | 2401 | return err; |
a426e122 | 2402 | } |
e22a25c9 AL |
2403 | } |
2404 | ||
bdc44640 | 2405 | CPU_FOREACH(cpu) { |
38e478ec | 2406 | err = kvm_update_guest_debug(cpu, 0); |
a426e122 | 2407 | if (err) { |
e22a25c9 | 2408 | return err; |
a426e122 | 2409 | } |
e22a25c9 AL |
2410 | } |
2411 | return 0; | |
2412 | } | |
2413 | ||
1d5791f4 | 2414 | void kvm_remove_all_breakpoints(CPUState *cpu) |
e22a25c9 AL |
2415 | { |
2416 | struct kvm_sw_breakpoint *bp, *next; | |
80b7cd73 | 2417 | KVMState *s = cpu->kvm_state; |
dc54e252 | 2418 | CPUState *tmpcpu; |
e22a25c9 | 2419 | |
72cf2d4f | 2420 | QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) { |
80b7cd73 | 2421 | if (kvm_arch_remove_sw_breakpoint(cpu, bp) != 0) { |
e22a25c9 | 2422 | /* Try harder to find a CPU that currently sees the breakpoint. */ |
dc54e252 CG |
2423 | CPU_FOREACH(tmpcpu) { |
2424 | if (kvm_arch_remove_sw_breakpoint(tmpcpu, bp) == 0) { | |
e22a25c9 | 2425 | break; |
a426e122 | 2426 | } |
e22a25c9 AL |
2427 | } |
2428 | } | |
78021d6d JK |
2429 | QTAILQ_REMOVE(&s->kvm_sw_breakpoints, bp, entry); |
2430 | g_free(bp); | |
e22a25c9 AL |
2431 | } |
2432 | kvm_arch_remove_all_hw_breakpoints(); | |
2433 | ||
bdc44640 | 2434 | CPU_FOREACH(cpu) { |
38e478ec | 2435 | kvm_update_guest_debug(cpu, 0); |
a426e122 | 2436 | } |
e22a25c9 AL |
2437 | } |
2438 | ||
2439 | #else /* !KVM_CAP_SET_GUEST_DEBUG */ | |
2440 | ||
38e478ec | 2441 | int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap) |
e22a25c9 AL |
2442 | { |
2443 | return -EINVAL; | |
2444 | } | |
2445 | ||
62278814 | 2446 | int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr, |
e22a25c9 AL |
2447 | target_ulong len, int type) |
2448 | { | |
2449 | return -EINVAL; | |
2450 | } | |
2451 | ||
62278814 | 2452 | int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr, |
e22a25c9 AL |
2453 | target_ulong len, int type) |
2454 | { | |
2455 | return -EINVAL; | |
2456 | } | |
2457 | ||
1d5791f4 | 2458 | void kvm_remove_all_breakpoints(CPUState *cpu) |
e22a25c9 AL |
2459 | { |
2460 | } | |
2461 | #endif /* !KVM_CAP_SET_GUEST_DEBUG */ | |
cc84de95 | 2462 | |
18268b60 | 2463 | static int kvm_set_signal_mask(CPUState *cpu, const sigset_t *sigset) |
cc84de95 | 2464 | { |
aed6efb9 | 2465 | KVMState *s = kvm_state; |
cc84de95 MT |
2466 | struct kvm_signal_mask *sigmask; |
2467 | int r; | |
2468 | ||
7267c094 | 2469 | sigmask = g_malloc(sizeof(*sigmask) + sizeof(*sigset)); |
cc84de95 | 2470 | |
aed6efb9 | 2471 | sigmask->len = s->sigmask_len; |
cc84de95 | 2472 | memcpy(sigmask->sigset, sigset, sizeof(*sigset)); |
1bc22652 | 2473 | r = kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, sigmask); |
7267c094 | 2474 | g_free(sigmask); |
cc84de95 MT |
2475 | |
2476 | return r; | |
2477 | } | |
4d39892c | 2478 | |
cf0f7cf9 | 2479 | static void kvm_ipi_signal(int sig) |
18268b60 | 2480 | { |
cf0f7cf9 PB |
2481 | if (current_cpu) { |
2482 | assert(kvm_immediate_exit); | |
2483 | kvm_cpu_kick(current_cpu); | |
2484 | } | |
18268b60 PB |
2485 | } |
2486 | ||
2487 | void kvm_init_cpu_signals(CPUState *cpu) | |
2488 | { | |
2489 | int r; | |
2490 | sigset_t set; | |
2491 | struct sigaction sigact; | |
2492 | ||
2493 | memset(&sigact, 0, sizeof(sigact)); | |
cf0f7cf9 | 2494 | sigact.sa_handler = kvm_ipi_signal; |
18268b60 PB |
2495 | sigaction(SIG_IPI, &sigact, NULL); |
2496 | ||
2497 | pthread_sigmask(SIG_BLOCK, NULL, &set); | |
2498 | #if defined KVM_HAVE_MCE_INJECTION | |
2499 | sigdelset(&set, SIGBUS); | |
2500 | pthread_sigmask(SIG_SETMASK, &set, NULL); | |
2501 | #endif | |
2502 | sigdelset(&set, SIG_IPI); | |
cf0f7cf9 PB |
2503 | if (kvm_immediate_exit) { |
2504 | r = pthread_sigmask(SIG_SETMASK, &set, NULL); | |
2505 | } else { | |
2506 | r = kvm_set_signal_mask(cpu, &set); | |
2507 | } | |
18268b60 PB |
2508 | if (r) { |
2509 | fprintf(stderr, "kvm_set_signal_mask: %s\n", strerror(-r)); | |
2510 | exit(1); | |
2511 | } | |
2512 | } | |
2513 | ||
2ae41db2 | 2514 | /* Called asynchronously in VCPU thread. */ |
290adf38 | 2515 | int kvm_on_sigbus_vcpu(CPUState *cpu, int code, void *addr) |
a1b87fe0 | 2516 | { |
2ae41db2 PB |
2517 | #ifdef KVM_HAVE_MCE_INJECTION |
2518 | if (have_sigbus_pending) { | |
2519 | return 1; | |
2520 | } | |
2521 | have_sigbus_pending = true; | |
2522 | pending_sigbus_addr = addr; | |
2523 | pending_sigbus_code = code; | |
2524 | atomic_set(&cpu->exit_request, 1); | |
2525 | return 0; | |
2526 | #else | |
2527 | return 1; | |
2528 | #endif | |
a1b87fe0 JK |
2529 | } |
2530 | ||
2ae41db2 | 2531 | /* Called synchronously (via signalfd) in main thread. */ |
a1b87fe0 JK |
2532 | int kvm_on_sigbus(int code, void *addr) |
2533 | { | |
2ae41db2 | 2534 | #ifdef KVM_HAVE_MCE_INJECTION |
4d39892c PB |
2535 | /* Action required MCE kills the process if SIGBUS is blocked. Because |
2536 | * that's what happens in the I/O thread, where we handle MCE via signalfd, | |
2537 | * we can only get action optional here. | |
2538 | */ | |
2539 | assert(code != BUS_MCEERR_AR); | |
2540 | kvm_arch_on_sigbus_vcpu(first_cpu, code, addr); | |
2541 | return 0; | |
2ae41db2 PB |
2542 | #else |
2543 | return 1; | |
2544 | #endif | |
a1b87fe0 | 2545 | } |
0a6a7cca CD |
2546 | |
2547 | int kvm_create_device(KVMState *s, uint64_t type, bool test) | |
2548 | { | |
2549 | int ret; | |
2550 | struct kvm_create_device create_dev; | |
2551 | ||
2552 | create_dev.type = type; | |
2553 | create_dev.fd = -1; | |
2554 | create_dev.flags = test ? KVM_CREATE_DEVICE_TEST : 0; | |
2555 | ||
2556 | if (!kvm_check_extension(s, KVM_CAP_DEVICE_CTRL)) { | |
2557 | return -ENOTSUP; | |
2558 | } | |
2559 | ||
2560 | ret = kvm_vm_ioctl(s, KVM_CREATE_DEVICE, &create_dev); | |
2561 | if (ret) { | |
2562 | return ret; | |
2563 | } | |
2564 | ||
2565 | return test ? 0 : create_dev.fd; | |
2566 | } | |
ada4135f | 2567 | |
29039acf PX |
2568 | bool kvm_device_supported(int vmfd, uint64_t type) |
2569 | { | |
2570 | struct kvm_create_device create_dev = { | |
2571 | .type = type, | |
2572 | .fd = -1, | |
2573 | .flags = KVM_CREATE_DEVICE_TEST, | |
2574 | }; | |
2575 | ||
2576 | if (ioctl(vmfd, KVM_CHECK_EXTENSION, KVM_CAP_DEVICE_CTRL) <= 0) { | |
2577 | return false; | |
2578 | } | |
2579 | ||
2580 | return (ioctl(vmfd, KVM_CREATE_DEVICE, &create_dev) >= 0); | |
2581 | } | |
2582 | ||
ada4135f CH |
2583 | int kvm_set_one_reg(CPUState *cs, uint64_t id, void *source) |
2584 | { | |
2585 | struct kvm_one_reg reg; | |
2586 | int r; | |
2587 | ||
2588 | reg.id = id; | |
2589 | reg.addr = (uintptr_t) source; | |
2590 | r = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®); | |
2591 | if (r) { | |
844a3d34 | 2592 | trace_kvm_failed_reg_set(id, strerror(-r)); |
ada4135f CH |
2593 | } |
2594 | return r; | |
2595 | } | |
2596 | ||
2597 | int kvm_get_one_reg(CPUState *cs, uint64_t id, void *target) | |
2598 | { | |
2599 | struct kvm_one_reg reg; | |
2600 | int r; | |
2601 | ||
2602 | reg.id = id; | |
2603 | reg.addr = (uintptr_t) target; | |
2604 | r = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®); | |
2605 | if (r) { | |
844a3d34 | 2606 | trace_kvm_failed_reg_get(id, strerror(-r)); |
ada4135f CH |
2607 | } |
2608 | return r; | |
2609 | } | |
782c3f29 EH |
2610 | |
2611 | static void kvm_accel_class_init(ObjectClass *oc, void *data) | |
2612 | { | |
2613 | AccelClass *ac = ACCEL_CLASS(oc); | |
2614 | ac->name = "KVM"; | |
0d15da8e | 2615 | ac->init_machine = kvm_init; |
782c3f29 EH |
2616 | ac->allowed = &kvm_allowed; |
2617 | } | |
2618 | ||
2619 | static const TypeInfo kvm_accel_type = { | |
2620 | .name = TYPE_KVM_ACCEL, | |
2621 | .parent = TYPE_ACCEL, | |
2622 | .class_init = kvm_accel_class_init, | |
fc02086b | 2623 | .instance_size = sizeof(KVMState), |
782c3f29 EH |
2624 | }; |
2625 | ||
2626 | static void kvm_type_init(void) | |
2627 | { | |
2628 | type_register_static(&kvm_accel_type); | |
2629 | } | |
2630 | ||
2631 | type_init(kvm_type_init); |