]>
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 | ||
16 | #include <sys/types.h> | |
17 | #include <sys/ioctl.h> | |
18 | #include <sys/mman.h> | |
984b5181 | 19 | #include <stdarg.h> |
05330448 AL |
20 | |
21 | #include <linux/kvm.h> | |
22 | ||
23 | #include "qemu-common.h" | |
85199474 | 24 | #include "qemu-barrier.h" |
ebd063d1 PB |
25 | #include "qemu-option.h" |
26 | #include "qemu-config.h" | |
05330448 | 27 | #include "sysemu.h" |
d33a1810 | 28 | #include "hw/hw.h" |
04fa27f5 | 29 | #include "hw/msi.h" |
e22a25c9 | 30 | #include "gdbstub.h" |
05330448 | 31 | #include "kvm.h" |
8369e01c | 32 | #include "bswap.h" |
a01672d3 | 33 | #include "memory.h" |
80a1ea37 | 34 | #include "exec-memory.h" |
753d5e14 | 35 | #include "event_notifier.h" |
05330448 | 36 | |
d2f2b8a7 SH |
37 | /* This check must be after config-host.h is included */ |
38 | #ifdef CONFIG_EVENTFD | |
39 | #include <sys/eventfd.h> | |
40 | #endif | |
41 | ||
7566759f CB |
42 | #ifdef CONFIG_VALGRIND_H |
43 | #include <valgrind/memcheck.h> | |
44 | #endif | |
45 | ||
93148aa5 | 46 | /* KVM uses PAGE_SIZE in its definition of COALESCED_MMIO_MAX */ |
f65ed4c1 AL |
47 | #define PAGE_SIZE TARGET_PAGE_SIZE |
48 | ||
05330448 AL |
49 | //#define DEBUG_KVM |
50 | ||
51 | #ifdef DEBUG_KVM | |
8c0d577e | 52 | #define DPRINTF(fmt, ...) \ |
05330448 AL |
53 | do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0) |
54 | #else | |
8c0d577e | 55 | #define DPRINTF(fmt, ...) \ |
05330448 AL |
56 | do { } while (0) |
57 | #endif | |
58 | ||
04fa27f5 JK |
59 | #define KVM_MSI_HASHTAB_SIZE 256 |
60 | ||
34fc643f AL |
61 | typedef struct KVMSlot |
62 | { | |
c227f099 AL |
63 | target_phys_addr_t start_addr; |
64 | ram_addr_t memory_size; | |
9f213ed9 | 65 | void *ram; |
34fc643f AL |
66 | int slot; |
67 | int flags; | |
68 | } KVMSlot; | |
05330448 | 69 | |
5832d1f2 AL |
70 | typedef struct kvm_dirty_log KVMDirtyLog; |
71 | ||
05330448 AL |
72 | struct KVMState |
73 | { | |
74 | KVMSlot slots[32]; | |
75 | int fd; | |
76 | int vmfd; | |
f65ed4c1 | 77 | int coalesced_mmio; |
62a2744c | 78 | struct kvm_coalesced_mmio_ring *coalesced_mmio_ring; |
1cae88b9 | 79 | bool coalesced_flush_in_progress; |
e69917e2 | 80 | int broken_set_mem_region; |
4495d6a7 | 81 | int migration_log; |
a0fb002c | 82 | int vcpu_events; |
b0b1d690 | 83 | int robust_singlestep; |
ff44f1a3 | 84 | int debugregs; |
e22a25c9 AL |
85 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
86 | struct kvm_sw_breakpoint_head kvm_sw_breakpoints; | |
87 | #endif | |
8a7c7393 | 88 | int pit_state2; |
f1665b21 | 89 | int xsave, xcrs; |
d2f2b8a7 | 90 | int many_ioeventfds; |
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 */ | |
94 | unsigned irqchip_inject_ioctl; | |
84b058d7 JK |
95 | #ifdef KVM_CAP_IRQ_ROUTING |
96 | struct kvm_irq_routing *irq_routes; | |
97 | int nr_allocated_irq_routes; | |
98 | uint32_t *used_gsi_bitmap; | |
4e2e4e63 | 99 | unsigned int gsi_count; |
04fa27f5 | 100 | QTAILQ_HEAD(msi_hashtab, KVMMSIRoute) msi_hashtab[KVM_MSI_HASHTAB_SIZE]; |
4a3adebb | 101 | bool direct_msi; |
84b058d7 | 102 | #endif |
05330448 AL |
103 | }; |
104 | ||
6a7af8cb | 105 | KVMState *kvm_state; |
3d4b2649 | 106 | bool kvm_kernel_irqchip; |
7ae26bd4 | 107 | bool kvm_async_interrupts_allowed; |
cc7e0ddf | 108 | bool kvm_irqfds_allowed; |
614e41bc | 109 | bool kvm_msi_via_irqfd_allowed; |
f3e1bed8 | 110 | bool kvm_gsi_routing_allowed; |
05330448 | 111 | |
94a8d39a JK |
112 | static const KVMCapabilityInfo kvm_required_capabilites[] = { |
113 | KVM_CAP_INFO(USER_MEMORY), | |
114 | KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS), | |
115 | KVM_CAP_LAST_INFO | |
116 | }; | |
117 | ||
05330448 AL |
118 | static KVMSlot *kvm_alloc_slot(KVMState *s) |
119 | { | |
120 | int i; | |
121 | ||
122 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
a426e122 | 123 | if (s->slots[i].memory_size == 0) { |
05330448 | 124 | return &s->slots[i]; |
a426e122 | 125 | } |
05330448 AL |
126 | } |
127 | ||
d3f8d37f AL |
128 | fprintf(stderr, "%s: no free slot available\n", __func__); |
129 | abort(); | |
130 | } | |
131 | ||
132 | static KVMSlot *kvm_lookup_matching_slot(KVMState *s, | |
c227f099 AL |
133 | target_phys_addr_t start_addr, |
134 | target_phys_addr_t end_addr) | |
d3f8d37f AL |
135 | { |
136 | int i; | |
137 | ||
138 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
139 | KVMSlot *mem = &s->slots[i]; | |
140 | ||
141 | if (start_addr == mem->start_addr && | |
142 | end_addr == mem->start_addr + mem->memory_size) { | |
143 | return mem; | |
144 | } | |
145 | } | |
146 | ||
05330448 AL |
147 | return NULL; |
148 | } | |
149 | ||
6152e2ae AL |
150 | /* |
151 | * Find overlapping slot with lowest start address | |
152 | */ | |
153 | static KVMSlot *kvm_lookup_overlapping_slot(KVMState *s, | |
c227f099 AL |
154 | target_phys_addr_t start_addr, |
155 | target_phys_addr_t end_addr) | |
05330448 | 156 | { |
6152e2ae | 157 | KVMSlot *found = NULL; |
05330448 AL |
158 | int i; |
159 | ||
160 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
161 | KVMSlot *mem = &s->slots[i]; | |
162 | ||
6152e2ae AL |
163 | if (mem->memory_size == 0 || |
164 | (found && found->start_addr < mem->start_addr)) { | |
165 | continue; | |
166 | } | |
167 | ||
168 | if (end_addr > mem->start_addr && | |
169 | start_addr < mem->start_addr + mem->memory_size) { | |
170 | found = mem; | |
171 | } | |
05330448 AL |
172 | } |
173 | ||
6152e2ae | 174 | return found; |
05330448 AL |
175 | } |
176 | ||
9f213ed9 AK |
177 | int kvm_physical_memory_addr_from_host(KVMState *s, void *ram, |
178 | target_phys_addr_t *phys_addr) | |
983dfc3b HY |
179 | { |
180 | int i; | |
181 | ||
182 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
183 | KVMSlot *mem = &s->slots[i]; | |
184 | ||
9f213ed9 AK |
185 | if (ram >= mem->ram && ram < mem->ram + mem->memory_size) { |
186 | *phys_addr = mem->start_addr + (ram - mem->ram); | |
983dfc3b HY |
187 | return 1; |
188 | } | |
189 | } | |
190 | ||
191 | return 0; | |
192 | } | |
193 | ||
5832d1f2 AL |
194 | static int kvm_set_user_memory_region(KVMState *s, KVMSlot *slot) |
195 | { | |
196 | struct kvm_userspace_memory_region mem; | |
197 | ||
198 | mem.slot = slot->slot; | |
199 | mem.guest_phys_addr = slot->start_addr; | |
200 | mem.memory_size = slot->memory_size; | |
9f213ed9 | 201 | mem.userspace_addr = (unsigned long)slot->ram; |
5832d1f2 | 202 | mem.flags = slot->flags; |
4495d6a7 JK |
203 | if (s->migration_log) { |
204 | mem.flags |= KVM_MEM_LOG_DIRTY_PAGES; | |
205 | } | |
5832d1f2 AL |
206 | return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem); |
207 | } | |
208 | ||
8d2ba1fb JK |
209 | static void kvm_reset_vcpu(void *opaque) |
210 | { | |
9349b4f9 | 211 | CPUArchState *env = opaque; |
8d2ba1fb | 212 | |
caa5af0f | 213 | kvm_arch_reset_vcpu(env); |
8d2ba1fb | 214 | } |
5832d1f2 | 215 | |
9349b4f9 | 216 | int kvm_init_vcpu(CPUArchState *env) |
05330448 AL |
217 | { |
218 | KVMState *s = kvm_state; | |
219 | long mmap_size; | |
220 | int ret; | |
221 | ||
8c0d577e | 222 | DPRINTF("kvm_init_vcpu\n"); |
05330448 | 223 | |
984b5181 | 224 | ret = kvm_vm_ioctl(s, KVM_CREATE_VCPU, env->cpu_index); |
05330448 | 225 | if (ret < 0) { |
8c0d577e | 226 | DPRINTF("kvm_create_vcpu failed\n"); |
05330448 AL |
227 | goto err; |
228 | } | |
229 | ||
230 | env->kvm_fd = ret; | |
231 | env->kvm_state = s; | |
d841b6c4 | 232 | env->kvm_vcpu_dirty = 1; |
05330448 AL |
233 | |
234 | mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0); | |
235 | if (mmap_size < 0) { | |
748a680b | 236 | ret = mmap_size; |
8c0d577e | 237 | DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n"); |
05330448 AL |
238 | goto err; |
239 | } | |
240 | ||
241 | env->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED, | |
242 | env->kvm_fd, 0); | |
243 | if (env->kvm_run == MAP_FAILED) { | |
244 | ret = -errno; | |
8c0d577e | 245 | DPRINTF("mmap'ing vcpu state failed\n"); |
05330448 AL |
246 | goto err; |
247 | } | |
248 | ||
a426e122 JK |
249 | if (s->coalesced_mmio && !s->coalesced_mmio_ring) { |
250 | s->coalesced_mmio_ring = | |
251 | (void *)env->kvm_run + s->coalesced_mmio * PAGE_SIZE; | |
252 | } | |
62a2744c | 253 | |
05330448 | 254 | ret = kvm_arch_init_vcpu(env); |
8d2ba1fb | 255 | if (ret == 0) { |
a08d4367 | 256 | qemu_register_reset(kvm_reset_vcpu, env); |
caa5af0f | 257 | kvm_arch_reset_vcpu(env); |
8d2ba1fb | 258 | } |
05330448 AL |
259 | err: |
260 | return ret; | |
261 | } | |
262 | ||
5832d1f2 AL |
263 | /* |
264 | * dirty pages logging control | |
265 | */ | |
25254bbc MT |
266 | |
267 | static int kvm_mem_flags(KVMState *s, bool log_dirty) | |
268 | { | |
269 | return log_dirty ? KVM_MEM_LOG_DIRTY_PAGES : 0; | |
270 | } | |
271 | ||
272 | static int kvm_slot_dirty_pages_log_change(KVMSlot *mem, bool log_dirty) | |
5832d1f2 AL |
273 | { |
274 | KVMState *s = kvm_state; | |
25254bbc | 275 | int flags, mask = KVM_MEM_LOG_DIRTY_PAGES; |
4495d6a7 JK |
276 | int old_flags; |
277 | ||
4495d6a7 | 278 | old_flags = mem->flags; |
5832d1f2 | 279 | |
25254bbc | 280 | flags = (mem->flags & ~mask) | kvm_mem_flags(s, log_dirty); |
5832d1f2 AL |
281 | mem->flags = flags; |
282 | ||
4495d6a7 JK |
283 | /* If nothing changed effectively, no need to issue ioctl */ |
284 | if (s->migration_log) { | |
285 | flags |= KVM_MEM_LOG_DIRTY_PAGES; | |
286 | } | |
25254bbc | 287 | |
4495d6a7 | 288 | if (flags == old_flags) { |
25254bbc | 289 | return 0; |
4495d6a7 JK |
290 | } |
291 | ||
5832d1f2 AL |
292 | return kvm_set_user_memory_region(s, mem); |
293 | } | |
294 | ||
25254bbc MT |
295 | static int kvm_dirty_pages_log_change(target_phys_addr_t phys_addr, |
296 | ram_addr_t size, bool log_dirty) | |
297 | { | |
298 | KVMState *s = kvm_state; | |
299 | KVMSlot *mem = kvm_lookup_matching_slot(s, phys_addr, phys_addr + size); | |
300 | ||
301 | if (mem == NULL) { | |
302 | fprintf(stderr, "BUG: %s: invalid parameters " TARGET_FMT_plx "-" | |
303 | TARGET_FMT_plx "\n", __func__, phys_addr, | |
304 | (target_phys_addr_t)(phys_addr + size - 1)); | |
305 | return -EINVAL; | |
306 | } | |
307 | return kvm_slot_dirty_pages_log_change(mem, log_dirty); | |
308 | } | |
309 | ||
a01672d3 AK |
310 | static void kvm_log_start(MemoryListener *listener, |
311 | MemoryRegionSection *section) | |
5832d1f2 | 312 | { |
a01672d3 AK |
313 | int r; |
314 | ||
315 | r = kvm_dirty_pages_log_change(section->offset_within_address_space, | |
316 | section->size, true); | |
317 | if (r < 0) { | |
318 | abort(); | |
319 | } | |
5832d1f2 AL |
320 | } |
321 | ||
a01672d3 AK |
322 | static void kvm_log_stop(MemoryListener *listener, |
323 | MemoryRegionSection *section) | |
5832d1f2 | 324 | { |
a01672d3 AK |
325 | int r; |
326 | ||
327 | r = kvm_dirty_pages_log_change(section->offset_within_address_space, | |
328 | section->size, false); | |
329 | if (r < 0) { | |
330 | abort(); | |
331 | } | |
5832d1f2 AL |
332 | } |
333 | ||
7b8f3b78 | 334 | static int kvm_set_migration_log(int enable) |
4495d6a7 JK |
335 | { |
336 | KVMState *s = kvm_state; | |
337 | KVMSlot *mem; | |
338 | int i, err; | |
339 | ||
340 | s->migration_log = enable; | |
341 | ||
342 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
343 | mem = &s->slots[i]; | |
344 | ||
70fedd76 AW |
345 | if (!mem->memory_size) { |
346 | continue; | |
347 | } | |
4495d6a7 JK |
348 | if (!!(mem->flags & KVM_MEM_LOG_DIRTY_PAGES) == enable) { |
349 | continue; | |
350 | } | |
351 | err = kvm_set_user_memory_region(s, mem); | |
352 | if (err) { | |
353 | return err; | |
354 | } | |
355 | } | |
356 | return 0; | |
357 | } | |
358 | ||
8369e01c | 359 | /* get kvm's dirty pages bitmap and update qemu's */ |
ffcde12f AK |
360 | static int kvm_get_dirty_pages_log_range(MemoryRegionSection *section, |
361 | unsigned long *bitmap) | |
96c1606b | 362 | { |
8369e01c | 363 | unsigned int i, j; |
aa90fec7 BH |
364 | unsigned long page_number, c; |
365 | target_phys_addr_t addr, addr1; | |
ffcde12f | 366 | unsigned int len = ((section->size / TARGET_PAGE_SIZE) + HOST_LONG_BITS - 1) / HOST_LONG_BITS; |
3145fcb6 | 367 | unsigned long hpratio = getpagesize() / TARGET_PAGE_SIZE; |
8369e01c MT |
368 | |
369 | /* | |
370 | * bitmap-traveling is faster than memory-traveling (for addr...) | |
371 | * especially when most of the memory is not dirty. | |
372 | */ | |
373 | for (i = 0; i < len; i++) { | |
374 | if (bitmap[i] != 0) { | |
375 | c = leul_to_cpu(bitmap[i]); | |
376 | do { | |
377 | j = ffsl(c) - 1; | |
378 | c &= ~(1ul << j); | |
3145fcb6 | 379 | page_number = (i * HOST_LONG_BITS + j) * hpratio; |
8369e01c | 380 | addr1 = page_number * TARGET_PAGE_SIZE; |
ffcde12f | 381 | addr = section->offset_within_region + addr1; |
3145fcb6 DG |
382 | memory_region_set_dirty(section->mr, addr, |
383 | TARGET_PAGE_SIZE * hpratio); | |
8369e01c MT |
384 | } while (c != 0); |
385 | } | |
386 | } | |
387 | return 0; | |
96c1606b AG |
388 | } |
389 | ||
8369e01c MT |
390 | #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1)) |
391 | ||
5832d1f2 AL |
392 | /** |
393 | * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space | |
fd4aa979 BS |
394 | * This function updates qemu's dirty bitmap using |
395 | * memory_region_set_dirty(). This means all bits are set | |
396 | * to dirty. | |
5832d1f2 | 397 | * |
d3f8d37f | 398 | * @start_add: start of logged region. |
5832d1f2 AL |
399 | * @end_addr: end of logged region. |
400 | */ | |
ffcde12f | 401 | static int kvm_physical_sync_dirty_bitmap(MemoryRegionSection *section) |
5832d1f2 AL |
402 | { |
403 | KVMState *s = kvm_state; | |
151f7749 | 404 | unsigned long size, allocated_size = 0; |
151f7749 JK |
405 | KVMDirtyLog d; |
406 | KVMSlot *mem; | |
407 | int ret = 0; | |
ffcde12f AK |
408 | target_phys_addr_t start_addr = section->offset_within_address_space; |
409 | target_phys_addr_t end_addr = start_addr + section->size; | |
5832d1f2 | 410 | |
151f7749 JK |
411 | d.dirty_bitmap = NULL; |
412 | while (start_addr < end_addr) { | |
413 | mem = kvm_lookup_overlapping_slot(s, start_addr, end_addr); | |
414 | if (mem == NULL) { | |
415 | break; | |
416 | } | |
5832d1f2 | 417 | |
51b0c606 MT |
418 | /* XXX bad kernel interface alert |
419 | * For dirty bitmap, kernel allocates array of size aligned to | |
420 | * bits-per-long. But for case when the kernel is 64bits and | |
421 | * the userspace is 32bits, userspace can't align to the same | |
422 | * bits-per-long, since sizeof(long) is different between kernel | |
423 | * and user space. This way, userspace will provide buffer which | |
424 | * may be 4 bytes less than the kernel will use, resulting in | |
425 | * userspace memory corruption (which is not detectable by valgrind | |
426 | * too, in most cases). | |
427 | * So for now, let's align to 64 instead of HOST_LONG_BITS here, in | |
428 | * a hope that sizeof(long) wont become >8 any time soon. | |
429 | */ | |
430 | size = ALIGN(((mem->memory_size) >> TARGET_PAGE_BITS), | |
431 | /*HOST_LONG_BITS*/ 64) / 8; | |
151f7749 | 432 | if (!d.dirty_bitmap) { |
7267c094 | 433 | d.dirty_bitmap = g_malloc(size); |
151f7749 | 434 | } else if (size > allocated_size) { |
7267c094 | 435 | d.dirty_bitmap = g_realloc(d.dirty_bitmap, size); |
151f7749 JK |
436 | } |
437 | allocated_size = size; | |
438 | memset(d.dirty_bitmap, 0, allocated_size); | |
5832d1f2 | 439 | |
151f7749 | 440 | d.slot = mem->slot; |
5832d1f2 | 441 | |
6e489f3f | 442 | if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) { |
8c0d577e | 443 | DPRINTF("ioctl failed %d\n", errno); |
151f7749 JK |
444 | ret = -1; |
445 | break; | |
446 | } | |
5832d1f2 | 447 | |
ffcde12f | 448 | kvm_get_dirty_pages_log_range(section, d.dirty_bitmap); |
8369e01c | 449 | start_addr = mem->start_addr + mem->memory_size; |
5832d1f2 | 450 | } |
7267c094 | 451 | g_free(d.dirty_bitmap); |
151f7749 JK |
452 | |
453 | return ret; | |
5832d1f2 AL |
454 | } |
455 | ||
c227f099 | 456 | int kvm_coalesce_mmio_region(target_phys_addr_t start, ram_addr_t size) |
f65ed4c1 AL |
457 | { |
458 | int ret = -ENOSYS; | |
f65ed4c1 AL |
459 | KVMState *s = kvm_state; |
460 | ||
461 | if (s->coalesced_mmio) { | |
462 | struct kvm_coalesced_mmio_zone zone; | |
463 | ||
464 | zone.addr = start; | |
465 | zone.size = size; | |
7e680753 | 466 | zone.pad = 0; |
f65ed4c1 AL |
467 | |
468 | ret = kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone); | |
469 | } | |
f65ed4c1 AL |
470 | |
471 | return ret; | |
472 | } | |
473 | ||
c227f099 | 474 | int kvm_uncoalesce_mmio_region(target_phys_addr_t start, ram_addr_t size) |
f65ed4c1 AL |
475 | { |
476 | int ret = -ENOSYS; | |
f65ed4c1 AL |
477 | KVMState *s = kvm_state; |
478 | ||
479 | if (s->coalesced_mmio) { | |
480 | struct kvm_coalesced_mmio_zone zone; | |
481 | ||
482 | zone.addr = start; | |
483 | zone.size = size; | |
7e680753 | 484 | zone.pad = 0; |
f65ed4c1 AL |
485 | |
486 | ret = kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone); | |
487 | } | |
f65ed4c1 AL |
488 | |
489 | return ret; | |
490 | } | |
491 | ||
ad7b8b33 AL |
492 | int kvm_check_extension(KVMState *s, unsigned int extension) |
493 | { | |
494 | int ret; | |
495 | ||
496 | ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension); | |
497 | if (ret < 0) { | |
498 | ret = 0; | |
499 | } | |
500 | ||
501 | return ret; | |
502 | } | |
503 | ||
d2f2b8a7 SH |
504 | static int kvm_check_many_ioeventfds(void) |
505 | { | |
d0dcac83 SH |
506 | /* Userspace can use ioeventfd for io notification. This requires a host |
507 | * that supports eventfd(2) and an I/O thread; since eventfd does not | |
508 | * support SIGIO it cannot interrupt the vcpu. | |
509 | * | |
510 | * Older kernels have a 6 device limit on the KVM io bus. Find out so we | |
d2f2b8a7 SH |
511 | * can avoid creating too many ioeventfds. |
512 | */ | |
12d4536f | 513 | #if defined(CONFIG_EVENTFD) |
d2f2b8a7 SH |
514 | int ioeventfds[7]; |
515 | int i, ret = 0; | |
516 | for (i = 0; i < ARRAY_SIZE(ioeventfds); i++) { | |
517 | ioeventfds[i] = eventfd(0, EFD_CLOEXEC); | |
518 | if (ioeventfds[i] < 0) { | |
519 | break; | |
520 | } | |
521 | ret = kvm_set_ioeventfd_pio_word(ioeventfds[i], 0, i, true); | |
522 | if (ret < 0) { | |
523 | close(ioeventfds[i]); | |
524 | break; | |
525 | } | |
526 | } | |
527 | ||
528 | /* Decide whether many devices are supported or not */ | |
529 | ret = i == ARRAY_SIZE(ioeventfds); | |
530 | ||
531 | while (i-- > 0) { | |
532 | kvm_set_ioeventfd_pio_word(ioeventfds[i], 0, i, false); | |
533 | close(ioeventfds[i]); | |
534 | } | |
535 | return ret; | |
536 | #else | |
537 | return 0; | |
538 | #endif | |
539 | } | |
540 | ||
94a8d39a JK |
541 | static const KVMCapabilityInfo * |
542 | kvm_check_extension_list(KVMState *s, const KVMCapabilityInfo *list) | |
543 | { | |
544 | while (list->name) { | |
545 | if (!kvm_check_extension(s, list->value)) { | |
546 | return list; | |
547 | } | |
548 | list++; | |
549 | } | |
550 | return NULL; | |
551 | } | |
552 | ||
a01672d3 | 553 | static void kvm_set_phys_mem(MemoryRegionSection *section, bool add) |
46dbef6a MT |
554 | { |
555 | KVMState *s = kvm_state; | |
46dbef6a MT |
556 | KVMSlot *mem, old; |
557 | int err; | |
a01672d3 AK |
558 | MemoryRegion *mr = section->mr; |
559 | bool log_dirty = memory_region_is_logging(mr); | |
560 | target_phys_addr_t start_addr = section->offset_within_address_space; | |
561 | ram_addr_t size = section->size; | |
9f213ed9 | 562 | void *ram = NULL; |
8f6f962b | 563 | unsigned delta; |
46dbef6a | 564 | |
14542fea GN |
565 | /* kvm works in page size chunks, but the function may be called |
566 | with sub-page size and unaligned start address. */ | |
8f6f962b AK |
567 | delta = TARGET_PAGE_ALIGN(size) - size; |
568 | if (delta > size) { | |
569 | return; | |
570 | } | |
571 | start_addr += delta; | |
572 | size -= delta; | |
573 | size &= TARGET_PAGE_MASK; | |
574 | if (!size || (start_addr & ~TARGET_PAGE_MASK)) { | |
575 | return; | |
576 | } | |
46dbef6a | 577 | |
a01672d3 AK |
578 | if (!memory_region_is_ram(mr)) { |
579 | return; | |
9f213ed9 AK |
580 | } |
581 | ||
8f6f962b | 582 | ram = memory_region_get_ram_ptr(mr) + section->offset_within_region + delta; |
a01672d3 | 583 | |
46dbef6a MT |
584 | while (1) { |
585 | mem = kvm_lookup_overlapping_slot(s, start_addr, start_addr + size); | |
586 | if (!mem) { | |
587 | break; | |
588 | } | |
589 | ||
a01672d3 | 590 | if (add && start_addr >= mem->start_addr && |
46dbef6a | 591 | (start_addr + size <= mem->start_addr + mem->memory_size) && |
9f213ed9 | 592 | (ram - start_addr == mem->ram - mem->start_addr)) { |
46dbef6a | 593 | /* The new slot fits into the existing one and comes with |
25254bbc MT |
594 | * identical parameters - update flags and done. */ |
595 | kvm_slot_dirty_pages_log_change(mem, log_dirty); | |
46dbef6a MT |
596 | return; |
597 | } | |
598 | ||
599 | old = *mem; | |
600 | ||
3fbffb62 AK |
601 | if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) { |
602 | kvm_physical_sync_dirty_bitmap(section); | |
603 | } | |
604 | ||
46dbef6a MT |
605 | /* unregister the overlapping slot */ |
606 | mem->memory_size = 0; | |
607 | err = kvm_set_user_memory_region(s, mem); | |
608 | if (err) { | |
609 | fprintf(stderr, "%s: error unregistering overlapping slot: %s\n", | |
610 | __func__, strerror(-err)); | |
611 | abort(); | |
612 | } | |
613 | ||
614 | /* Workaround for older KVM versions: we can't join slots, even not by | |
615 | * unregistering the previous ones and then registering the larger | |
616 | * slot. We have to maintain the existing fragmentation. Sigh. | |
617 | * | |
618 | * This workaround assumes that the new slot starts at the same | |
619 | * address as the first existing one. If not or if some overlapping | |
620 | * slot comes around later, we will fail (not seen in practice so far) | |
621 | * - and actually require a recent KVM version. */ | |
622 | if (s->broken_set_mem_region && | |
a01672d3 | 623 | old.start_addr == start_addr && old.memory_size < size && add) { |
46dbef6a MT |
624 | mem = kvm_alloc_slot(s); |
625 | mem->memory_size = old.memory_size; | |
626 | mem->start_addr = old.start_addr; | |
9f213ed9 | 627 | mem->ram = old.ram; |
25254bbc | 628 | mem->flags = kvm_mem_flags(s, log_dirty); |
46dbef6a MT |
629 | |
630 | err = kvm_set_user_memory_region(s, mem); | |
631 | if (err) { | |
632 | fprintf(stderr, "%s: error updating slot: %s\n", __func__, | |
633 | strerror(-err)); | |
634 | abort(); | |
635 | } | |
636 | ||
637 | start_addr += old.memory_size; | |
9f213ed9 | 638 | ram += old.memory_size; |
46dbef6a MT |
639 | size -= old.memory_size; |
640 | continue; | |
641 | } | |
642 | ||
643 | /* register prefix slot */ | |
644 | if (old.start_addr < start_addr) { | |
645 | mem = kvm_alloc_slot(s); | |
646 | mem->memory_size = start_addr - old.start_addr; | |
647 | mem->start_addr = old.start_addr; | |
9f213ed9 | 648 | mem->ram = old.ram; |
25254bbc | 649 | mem->flags = kvm_mem_flags(s, log_dirty); |
46dbef6a MT |
650 | |
651 | err = kvm_set_user_memory_region(s, mem); | |
652 | if (err) { | |
653 | fprintf(stderr, "%s: error registering prefix slot: %s\n", | |
654 | __func__, strerror(-err)); | |
d4d6868f AG |
655 | #ifdef TARGET_PPC |
656 | fprintf(stderr, "%s: This is probably because your kernel's " \ | |
657 | "PAGE_SIZE is too big. Please try to use 4k " \ | |
658 | "PAGE_SIZE!\n", __func__); | |
659 | #endif | |
46dbef6a MT |
660 | abort(); |
661 | } | |
662 | } | |
663 | ||
664 | /* register suffix slot */ | |
665 | if (old.start_addr + old.memory_size > start_addr + size) { | |
666 | ram_addr_t size_delta; | |
667 | ||
668 | mem = kvm_alloc_slot(s); | |
669 | mem->start_addr = start_addr + size; | |
670 | size_delta = mem->start_addr - old.start_addr; | |
671 | mem->memory_size = old.memory_size - size_delta; | |
9f213ed9 | 672 | mem->ram = old.ram + size_delta; |
25254bbc | 673 | mem->flags = kvm_mem_flags(s, log_dirty); |
46dbef6a MT |
674 | |
675 | err = kvm_set_user_memory_region(s, mem); | |
676 | if (err) { | |
677 | fprintf(stderr, "%s: error registering suffix slot: %s\n", | |
678 | __func__, strerror(-err)); | |
679 | abort(); | |
680 | } | |
681 | } | |
682 | } | |
683 | ||
684 | /* in case the KVM bug workaround already "consumed" the new slot */ | |
a426e122 | 685 | if (!size) { |
46dbef6a | 686 | return; |
a426e122 | 687 | } |
a01672d3 | 688 | if (!add) { |
46dbef6a | 689 | return; |
a426e122 | 690 | } |
46dbef6a MT |
691 | mem = kvm_alloc_slot(s); |
692 | mem->memory_size = size; | |
693 | mem->start_addr = start_addr; | |
9f213ed9 | 694 | mem->ram = ram; |
25254bbc | 695 | mem->flags = kvm_mem_flags(s, log_dirty); |
46dbef6a MT |
696 | |
697 | err = kvm_set_user_memory_region(s, mem); | |
698 | if (err) { | |
699 | fprintf(stderr, "%s: error registering slot: %s\n", __func__, | |
700 | strerror(-err)); | |
701 | abort(); | |
702 | } | |
703 | } | |
704 | ||
50c1e149 AK |
705 | static void kvm_begin(MemoryListener *listener) |
706 | { | |
707 | } | |
708 | ||
709 | static void kvm_commit(MemoryListener *listener) | |
710 | { | |
711 | } | |
712 | ||
a01672d3 AK |
713 | static void kvm_region_add(MemoryListener *listener, |
714 | MemoryRegionSection *section) | |
715 | { | |
716 | kvm_set_phys_mem(section, true); | |
717 | } | |
718 | ||
719 | static void kvm_region_del(MemoryListener *listener, | |
720 | MemoryRegionSection *section) | |
721 | { | |
722 | kvm_set_phys_mem(section, false); | |
723 | } | |
724 | ||
50c1e149 AK |
725 | static void kvm_region_nop(MemoryListener *listener, |
726 | MemoryRegionSection *section) | |
727 | { | |
728 | } | |
729 | ||
a01672d3 AK |
730 | static void kvm_log_sync(MemoryListener *listener, |
731 | MemoryRegionSection *section) | |
7b8f3b78 | 732 | { |
a01672d3 AK |
733 | int r; |
734 | ||
ffcde12f | 735 | r = kvm_physical_sync_dirty_bitmap(section); |
a01672d3 AK |
736 | if (r < 0) { |
737 | abort(); | |
738 | } | |
7b8f3b78 MT |
739 | } |
740 | ||
a01672d3 | 741 | static void kvm_log_global_start(struct MemoryListener *listener) |
7b8f3b78 | 742 | { |
a01672d3 AK |
743 | int r; |
744 | ||
745 | r = kvm_set_migration_log(1); | |
746 | assert(r >= 0); | |
7b8f3b78 MT |
747 | } |
748 | ||
a01672d3 | 749 | static void kvm_log_global_stop(struct MemoryListener *listener) |
7b8f3b78 | 750 | { |
a01672d3 AK |
751 | int r; |
752 | ||
753 | r = kvm_set_migration_log(0); | |
754 | assert(r >= 0); | |
7b8f3b78 MT |
755 | } |
756 | ||
80a1ea37 AK |
757 | static void kvm_mem_ioeventfd_add(MemoryRegionSection *section, |
758 | bool match_data, uint64_t data, int fd) | |
759 | { | |
760 | int r; | |
761 | ||
4b8f1c88 | 762 | assert(match_data && section->size <= 8); |
80a1ea37 | 763 | |
4b8f1c88 MT |
764 | r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space, |
765 | data, true, section->size); | |
80a1ea37 AK |
766 | if (r < 0) { |
767 | abort(); | |
768 | } | |
769 | } | |
770 | ||
771 | static void kvm_mem_ioeventfd_del(MemoryRegionSection *section, | |
772 | bool match_data, uint64_t data, int fd) | |
773 | { | |
774 | int r; | |
775 | ||
4b8f1c88 MT |
776 | r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space, |
777 | data, false, section->size); | |
80a1ea37 AK |
778 | if (r < 0) { |
779 | abort(); | |
780 | } | |
781 | } | |
782 | ||
783 | static void kvm_io_ioeventfd_add(MemoryRegionSection *section, | |
784 | bool match_data, uint64_t data, int fd) | |
785 | { | |
786 | int r; | |
787 | ||
788 | assert(match_data && section->size == 2); | |
789 | ||
790 | r = kvm_set_ioeventfd_pio_word(fd, section->offset_within_address_space, | |
791 | data, true); | |
792 | if (r < 0) { | |
793 | abort(); | |
794 | } | |
795 | } | |
796 | ||
797 | static void kvm_io_ioeventfd_del(MemoryRegionSection *section, | |
798 | bool match_data, uint64_t data, int fd) | |
799 | ||
800 | { | |
801 | int r; | |
802 | ||
803 | r = kvm_set_ioeventfd_pio_word(fd, section->offset_within_address_space, | |
804 | data, false); | |
805 | if (r < 0) { | |
806 | abort(); | |
807 | } | |
808 | } | |
809 | ||
810 | static void kvm_eventfd_add(MemoryListener *listener, | |
811 | MemoryRegionSection *section, | |
753d5e14 PB |
812 | bool match_data, uint64_t data, |
813 | EventNotifier *e) | |
80a1ea37 AK |
814 | { |
815 | if (section->address_space == get_system_memory()) { | |
753d5e14 PB |
816 | kvm_mem_ioeventfd_add(section, match_data, data, |
817 | event_notifier_get_fd(e)); | |
80a1ea37 | 818 | } else { |
753d5e14 PB |
819 | kvm_io_ioeventfd_add(section, match_data, data, |
820 | event_notifier_get_fd(e)); | |
80a1ea37 AK |
821 | } |
822 | } | |
823 | ||
824 | static void kvm_eventfd_del(MemoryListener *listener, | |
825 | MemoryRegionSection *section, | |
753d5e14 PB |
826 | bool match_data, uint64_t data, |
827 | EventNotifier *e) | |
80a1ea37 AK |
828 | { |
829 | if (section->address_space == get_system_memory()) { | |
753d5e14 PB |
830 | kvm_mem_ioeventfd_del(section, match_data, data, |
831 | event_notifier_get_fd(e)); | |
80a1ea37 | 832 | } else { |
753d5e14 PB |
833 | kvm_io_ioeventfd_del(section, match_data, data, |
834 | event_notifier_get_fd(e)); | |
80a1ea37 AK |
835 | } |
836 | } | |
837 | ||
a01672d3 | 838 | static MemoryListener kvm_memory_listener = { |
50c1e149 AK |
839 | .begin = kvm_begin, |
840 | .commit = kvm_commit, | |
a01672d3 AK |
841 | .region_add = kvm_region_add, |
842 | .region_del = kvm_region_del, | |
50c1e149 | 843 | .region_nop = kvm_region_nop, |
e5896b12 AP |
844 | .log_start = kvm_log_start, |
845 | .log_stop = kvm_log_stop, | |
a01672d3 AK |
846 | .log_sync = kvm_log_sync, |
847 | .log_global_start = kvm_log_global_start, | |
848 | .log_global_stop = kvm_log_global_stop, | |
80a1ea37 AK |
849 | .eventfd_add = kvm_eventfd_add, |
850 | .eventfd_del = kvm_eventfd_del, | |
72e22d2f | 851 | .priority = 10, |
7b8f3b78 MT |
852 | }; |
853 | ||
9349b4f9 | 854 | static void kvm_handle_interrupt(CPUArchState *env, int mask) |
aa7f74d1 JK |
855 | { |
856 | env->interrupt_request |= mask; | |
857 | ||
858 | if (!qemu_cpu_is_self(env)) { | |
859 | qemu_cpu_kick(env); | |
860 | } | |
861 | } | |
862 | ||
3889c3fa | 863 | int kvm_set_irq(KVMState *s, int irq, int level) |
84b058d7 JK |
864 | { |
865 | struct kvm_irq_level event; | |
866 | int ret; | |
867 | ||
7ae26bd4 | 868 | assert(kvm_async_interrupts_enabled()); |
84b058d7 JK |
869 | |
870 | event.level = level; | |
871 | event.irq = irq; | |
872 | ret = kvm_vm_ioctl(s, s->irqchip_inject_ioctl, &event); | |
873 | if (ret < 0) { | |
3889c3fa | 874 | perror("kvm_set_irq"); |
84b058d7 JK |
875 | abort(); |
876 | } | |
877 | ||
878 | return (s->irqchip_inject_ioctl == KVM_IRQ_LINE) ? 1 : event.status; | |
879 | } | |
880 | ||
881 | #ifdef KVM_CAP_IRQ_ROUTING | |
d3d3bef0 JK |
882 | typedef struct KVMMSIRoute { |
883 | struct kvm_irq_routing_entry kroute; | |
884 | QTAILQ_ENTRY(KVMMSIRoute) entry; | |
885 | } KVMMSIRoute; | |
886 | ||
84b058d7 JK |
887 | static void set_gsi(KVMState *s, unsigned int gsi) |
888 | { | |
84b058d7 JK |
889 | s->used_gsi_bitmap[gsi / 32] |= 1U << (gsi % 32); |
890 | } | |
891 | ||
04fa27f5 JK |
892 | static void clear_gsi(KVMState *s, unsigned int gsi) |
893 | { | |
894 | s->used_gsi_bitmap[gsi / 32] &= ~(1U << (gsi % 32)); | |
895 | } | |
896 | ||
84b058d7 JK |
897 | static void kvm_init_irq_routing(KVMState *s) |
898 | { | |
04fa27f5 | 899 | int gsi_count, i; |
84b058d7 JK |
900 | |
901 | gsi_count = kvm_check_extension(s, KVM_CAP_IRQ_ROUTING); | |
902 | if (gsi_count > 0) { | |
903 | unsigned int gsi_bits, i; | |
904 | ||
905 | /* Round up so we can search ints using ffs */ | |
bc8c6788 | 906 | gsi_bits = ALIGN(gsi_count, 32); |
84b058d7 | 907 | s->used_gsi_bitmap = g_malloc0(gsi_bits / 8); |
4e2e4e63 | 908 | s->gsi_count = gsi_count; |
84b058d7 JK |
909 | |
910 | /* Mark any over-allocated bits as already in use */ | |
911 | for (i = gsi_count; i < gsi_bits; i++) { | |
912 | set_gsi(s, i); | |
913 | } | |
914 | } | |
915 | ||
916 | s->irq_routes = g_malloc0(sizeof(*s->irq_routes)); | |
917 | s->nr_allocated_irq_routes = 0; | |
918 | ||
4a3adebb JK |
919 | if (!s->direct_msi) { |
920 | for (i = 0; i < KVM_MSI_HASHTAB_SIZE; i++) { | |
921 | QTAILQ_INIT(&s->msi_hashtab[i]); | |
922 | } | |
04fa27f5 JK |
923 | } |
924 | ||
84b058d7 JK |
925 | kvm_arch_init_irq_routing(s); |
926 | } | |
927 | ||
e7b20308 JK |
928 | static void kvm_irqchip_commit_routes(KVMState *s) |
929 | { | |
930 | int ret; | |
931 | ||
932 | s->irq_routes->flags = 0; | |
933 | ret = kvm_vm_ioctl(s, KVM_SET_GSI_ROUTING, s->irq_routes); | |
934 | assert(ret == 0); | |
935 | } | |
936 | ||
84b058d7 JK |
937 | static void kvm_add_routing_entry(KVMState *s, |
938 | struct kvm_irq_routing_entry *entry) | |
939 | { | |
940 | struct kvm_irq_routing_entry *new; | |
941 | int n, size; | |
942 | ||
943 | if (s->irq_routes->nr == s->nr_allocated_irq_routes) { | |
944 | n = s->nr_allocated_irq_routes * 2; | |
945 | if (n < 64) { | |
946 | n = 64; | |
947 | } | |
948 | size = sizeof(struct kvm_irq_routing); | |
949 | size += n * sizeof(*new); | |
950 | s->irq_routes = g_realloc(s->irq_routes, size); | |
951 | s->nr_allocated_irq_routes = n; | |
952 | } | |
953 | n = s->irq_routes->nr++; | |
954 | new = &s->irq_routes->entries[n]; | |
955 | memset(new, 0, sizeof(*new)); | |
956 | new->gsi = entry->gsi; | |
957 | new->type = entry->type; | |
958 | new->flags = entry->flags; | |
959 | new->u = entry->u; | |
960 | ||
961 | set_gsi(s, entry->gsi); | |
e7b20308 JK |
962 | |
963 | kvm_irqchip_commit_routes(s); | |
84b058d7 JK |
964 | } |
965 | ||
1df186df | 966 | void kvm_irqchip_add_irq_route(KVMState *s, int irq, int irqchip, int pin) |
84b058d7 JK |
967 | { |
968 | struct kvm_irq_routing_entry e; | |
969 | ||
4e2e4e63 JK |
970 | assert(pin < s->gsi_count); |
971 | ||
84b058d7 JK |
972 | e.gsi = irq; |
973 | e.type = KVM_IRQ_ROUTING_IRQCHIP; | |
974 | e.flags = 0; | |
975 | e.u.irqchip.irqchip = irqchip; | |
976 | e.u.irqchip.pin = pin; | |
977 | kvm_add_routing_entry(s, &e); | |
978 | } | |
979 | ||
1e2aa8be | 980 | void kvm_irqchip_release_virq(KVMState *s, int virq) |
04fa27f5 JK |
981 | { |
982 | struct kvm_irq_routing_entry *e; | |
983 | int i; | |
984 | ||
985 | for (i = 0; i < s->irq_routes->nr; i++) { | |
986 | e = &s->irq_routes->entries[i]; | |
987 | if (e->gsi == virq) { | |
988 | s->irq_routes->nr--; | |
989 | *e = s->irq_routes->entries[s->irq_routes->nr]; | |
990 | } | |
991 | } | |
992 | clear_gsi(s, virq); | |
e7b20308 JK |
993 | |
994 | kvm_irqchip_commit_routes(s); | |
04fa27f5 JK |
995 | } |
996 | ||
997 | static unsigned int kvm_hash_msi(uint32_t data) | |
998 | { | |
999 | /* This is optimized for IA32 MSI layout. However, no other arch shall | |
1000 | * repeat the mistake of not providing a direct MSI injection API. */ | |
1001 | return data & 0xff; | |
1002 | } | |
1003 | ||
1004 | static void kvm_flush_dynamic_msi_routes(KVMState *s) | |
1005 | { | |
1006 | KVMMSIRoute *route, *next; | |
1007 | unsigned int hash; | |
1008 | ||
1009 | for (hash = 0; hash < KVM_MSI_HASHTAB_SIZE; hash++) { | |
1010 | QTAILQ_FOREACH_SAFE(route, &s->msi_hashtab[hash], entry, next) { | |
1011 | kvm_irqchip_release_virq(s, route->kroute.gsi); | |
1012 | QTAILQ_REMOVE(&s->msi_hashtab[hash], route, entry); | |
1013 | g_free(route); | |
1014 | } | |
1015 | } | |
1016 | } | |
1017 | ||
1018 | static int kvm_irqchip_get_virq(KVMState *s) | |
1019 | { | |
1020 | uint32_t *word = s->used_gsi_bitmap; | |
1021 | int max_words = ALIGN(s->gsi_count, 32) / 32; | |
1022 | int i, bit; | |
1023 | bool retry = true; | |
1024 | ||
1025 | again: | |
1026 | /* Return the lowest unused GSI in the bitmap */ | |
1027 | for (i = 0; i < max_words; i++) { | |
1028 | bit = ffs(~word[i]); | |
1029 | if (!bit) { | |
1030 | continue; | |
1031 | } | |
1032 | ||
1033 | return bit - 1 + i * 32; | |
1034 | } | |
4a3adebb | 1035 | if (!s->direct_msi && retry) { |
04fa27f5 JK |
1036 | retry = false; |
1037 | kvm_flush_dynamic_msi_routes(s); | |
1038 | goto again; | |
1039 | } | |
1040 | return -ENOSPC; | |
1041 | ||
1042 | } | |
1043 | ||
1044 | static KVMMSIRoute *kvm_lookup_msi_route(KVMState *s, MSIMessage msg) | |
1045 | { | |
1046 | unsigned int hash = kvm_hash_msi(msg.data); | |
1047 | KVMMSIRoute *route; | |
1048 | ||
1049 | QTAILQ_FOREACH(route, &s->msi_hashtab[hash], entry) { | |
1050 | if (route->kroute.u.msi.address_lo == (uint32_t)msg.address && | |
1051 | route->kroute.u.msi.address_hi == (msg.address >> 32) && | |
1052 | route->kroute.u.msi.data == msg.data) { | |
1053 | return route; | |
1054 | } | |
1055 | } | |
1056 | return NULL; | |
1057 | } | |
1058 | ||
1059 | int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg) | |
1060 | { | |
4a3adebb | 1061 | struct kvm_msi msi; |
04fa27f5 JK |
1062 | KVMMSIRoute *route; |
1063 | ||
4a3adebb JK |
1064 | if (s->direct_msi) { |
1065 | msi.address_lo = (uint32_t)msg.address; | |
1066 | msi.address_hi = msg.address >> 32; | |
1067 | msi.data = msg.data; | |
1068 | msi.flags = 0; | |
1069 | memset(msi.pad, 0, sizeof(msi.pad)); | |
1070 | ||
1071 | return kvm_vm_ioctl(s, KVM_SIGNAL_MSI, &msi); | |
1072 | } | |
1073 | ||
04fa27f5 JK |
1074 | route = kvm_lookup_msi_route(s, msg); |
1075 | if (!route) { | |
e7b20308 | 1076 | int virq; |
04fa27f5 JK |
1077 | |
1078 | virq = kvm_irqchip_get_virq(s); | |
1079 | if (virq < 0) { | |
1080 | return virq; | |
1081 | } | |
1082 | ||
1083 | route = g_malloc(sizeof(KVMMSIRoute)); | |
1084 | route->kroute.gsi = virq; | |
1085 | route->kroute.type = KVM_IRQ_ROUTING_MSI; | |
1086 | route->kroute.flags = 0; | |
1087 | route->kroute.u.msi.address_lo = (uint32_t)msg.address; | |
1088 | route->kroute.u.msi.address_hi = msg.address >> 32; | |
1089 | route->kroute.u.msi.data = msg.data; | |
1090 | ||
1091 | kvm_add_routing_entry(s, &route->kroute); | |
1092 | ||
1093 | QTAILQ_INSERT_TAIL(&s->msi_hashtab[kvm_hash_msi(msg.data)], route, | |
1094 | entry); | |
04fa27f5 JK |
1095 | } |
1096 | ||
1097 | assert(route->kroute.type == KVM_IRQ_ROUTING_MSI); | |
1098 | ||
3889c3fa | 1099 | return kvm_set_irq(s, route->kroute.gsi, 1); |
04fa27f5 JK |
1100 | } |
1101 | ||
92b4e489 JK |
1102 | int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg) |
1103 | { | |
1104 | struct kvm_irq_routing_entry kroute; | |
1105 | int virq; | |
1106 | ||
f3e1bed8 | 1107 | if (!kvm_gsi_routing_enabled()) { |
92b4e489 JK |
1108 | return -ENOSYS; |
1109 | } | |
1110 | ||
1111 | virq = kvm_irqchip_get_virq(s); | |
1112 | if (virq < 0) { | |
1113 | return virq; | |
1114 | } | |
1115 | ||
1116 | kroute.gsi = virq; | |
1117 | kroute.type = KVM_IRQ_ROUTING_MSI; | |
1118 | kroute.flags = 0; | |
1119 | kroute.u.msi.address_lo = (uint32_t)msg.address; | |
1120 | kroute.u.msi.address_hi = msg.address >> 32; | |
1121 | kroute.u.msi.data = msg.data; | |
1122 | ||
1123 | kvm_add_routing_entry(s, &kroute); | |
1124 | ||
1125 | return virq; | |
1126 | } | |
1127 | ||
39853bbc JK |
1128 | static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int virq, bool assign) |
1129 | { | |
1130 | struct kvm_irqfd irqfd = { | |
1131 | .fd = fd, | |
1132 | .gsi = virq, | |
1133 | .flags = assign ? 0 : KVM_IRQFD_FLAG_DEASSIGN, | |
1134 | }; | |
1135 | ||
cc7e0ddf | 1136 | if (!kvm_irqfds_enabled()) { |
39853bbc JK |
1137 | return -ENOSYS; |
1138 | } | |
1139 | ||
1140 | return kvm_vm_ioctl(s, KVM_IRQFD, &irqfd); | |
1141 | } | |
1142 | ||
84b058d7 JK |
1143 | #else /* !KVM_CAP_IRQ_ROUTING */ |
1144 | ||
1145 | static void kvm_init_irq_routing(KVMState *s) | |
1146 | { | |
1147 | } | |
04fa27f5 | 1148 | |
d3d3bef0 JK |
1149 | void kvm_irqchip_release_virq(KVMState *s, int virq) |
1150 | { | |
1151 | } | |
1152 | ||
04fa27f5 JK |
1153 | int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg) |
1154 | { | |
1155 | abort(); | |
1156 | } | |
92b4e489 JK |
1157 | |
1158 | int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg) | |
1159 | { | |
df410675 | 1160 | return -ENOSYS; |
92b4e489 | 1161 | } |
39853bbc JK |
1162 | |
1163 | static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int virq, bool assign) | |
1164 | { | |
1165 | abort(); | |
1166 | } | |
84b058d7 JK |
1167 | #endif /* !KVM_CAP_IRQ_ROUTING */ |
1168 | ||
39853bbc JK |
1169 | int kvm_irqchip_add_irqfd(KVMState *s, int fd, int virq) |
1170 | { | |
1171 | return kvm_irqchip_assign_irqfd(s, fd, virq, true); | |
1172 | } | |
1173 | ||
15b2bd18 PB |
1174 | int kvm_irqchip_add_irq_notifier(KVMState *s, EventNotifier *n, int virq) |
1175 | { | |
1176 | return kvm_irqchip_add_irqfd(s, event_notifier_get_fd(n), virq); | |
1177 | } | |
1178 | ||
39853bbc JK |
1179 | int kvm_irqchip_remove_irqfd(KVMState *s, int fd, int virq) |
1180 | { | |
1181 | return kvm_irqchip_assign_irqfd(s, fd, virq, false); | |
1182 | } | |
1183 | ||
15b2bd18 PB |
1184 | int kvm_irqchip_remove_irq_notifier(KVMState *s, EventNotifier *n, int virq) |
1185 | { | |
1186 | return kvm_irqchip_remove_irqfd(s, event_notifier_get_fd(n), virq); | |
1187 | } | |
1188 | ||
84b058d7 JK |
1189 | static int kvm_irqchip_create(KVMState *s) |
1190 | { | |
1191 | QemuOptsList *list = qemu_find_opts("machine"); | |
1192 | int ret; | |
1193 | ||
1194 | if (QTAILQ_EMPTY(&list->head) || | |
1195 | !qemu_opt_get_bool(QTAILQ_FIRST(&list->head), | |
a24b9106 | 1196 | "kernel_irqchip", true) || |
84b058d7 JK |
1197 | !kvm_check_extension(s, KVM_CAP_IRQCHIP)) { |
1198 | return 0; | |
1199 | } | |
1200 | ||
1201 | ret = kvm_vm_ioctl(s, KVM_CREATE_IRQCHIP); | |
1202 | if (ret < 0) { | |
1203 | fprintf(stderr, "Create kernel irqchip failed\n"); | |
1204 | return ret; | |
1205 | } | |
1206 | ||
1207 | s->irqchip_inject_ioctl = KVM_IRQ_LINE; | |
1208 | if (kvm_check_extension(s, KVM_CAP_IRQ_INJECT_STATUS)) { | |
1209 | s->irqchip_inject_ioctl = KVM_IRQ_LINE_STATUS; | |
1210 | } | |
3d4b2649 | 1211 | kvm_kernel_irqchip = true; |
7ae26bd4 PM |
1212 | /* If we have an in-kernel IRQ chip then we must have asynchronous |
1213 | * interrupt delivery (though the reverse is not necessarily true) | |
1214 | */ | |
1215 | kvm_async_interrupts_allowed = true; | |
84b058d7 JK |
1216 | |
1217 | kvm_init_irq_routing(s); | |
1218 | ||
1219 | return 0; | |
1220 | } | |
1221 | ||
3ed444e9 DH |
1222 | static int kvm_max_vcpus(KVMState *s) |
1223 | { | |
1224 | int ret; | |
1225 | ||
1226 | /* Find number of supported CPUs using the recommended | |
1227 | * procedure from the kernel API documentation to cope with | |
1228 | * older kernels that may be missing capabilities. | |
1229 | */ | |
1230 | ret = kvm_check_extension(s, KVM_CAP_MAX_VCPUS); | |
1231 | if (ret) { | |
1232 | return ret; | |
1233 | } | |
1234 | ret = kvm_check_extension(s, KVM_CAP_NR_VCPUS); | |
1235 | if (ret) { | |
1236 | return ret; | |
1237 | } | |
1238 | ||
1239 | return 4; | |
1240 | } | |
1241 | ||
cad1e282 | 1242 | int kvm_init(void) |
05330448 | 1243 | { |
168ccc11 JK |
1244 | static const char upgrade_note[] = |
1245 | "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n" | |
1246 | "(see http://sourceforge.net/projects/kvm).\n"; | |
05330448 | 1247 | KVMState *s; |
94a8d39a | 1248 | const KVMCapabilityInfo *missing_cap; |
05330448 AL |
1249 | int ret; |
1250 | int i; | |
3ed444e9 | 1251 | int max_vcpus; |
05330448 | 1252 | |
7267c094 | 1253 | s = g_malloc0(sizeof(KVMState)); |
05330448 | 1254 | |
3145fcb6 DG |
1255 | /* |
1256 | * On systems where the kernel can support different base page | |
1257 | * sizes, host page size may be different from TARGET_PAGE_SIZE, | |
1258 | * even with KVM. TARGET_PAGE_SIZE is assumed to be the minimum | |
1259 | * page size for the system though. | |
1260 | */ | |
1261 | assert(TARGET_PAGE_SIZE <= getpagesize()); | |
1262 | ||
e22a25c9 | 1263 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
72cf2d4f | 1264 | QTAILQ_INIT(&s->kvm_sw_breakpoints); |
e22a25c9 | 1265 | #endif |
a426e122 | 1266 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { |
05330448 | 1267 | s->slots[i].slot = i; |
a426e122 | 1268 | } |
05330448 | 1269 | s->vmfd = -1; |
40ff6d7e | 1270 | s->fd = qemu_open("/dev/kvm", O_RDWR); |
05330448 AL |
1271 | if (s->fd == -1) { |
1272 | fprintf(stderr, "Could not access KVM kernel module: %m\n"); | |
1273 | ret = -errno; | |
1274 | goto err; | |
1275 | } | |
1276 | ||
1277 | ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0); | |
1278 | if (ret < KVM_API_VERSION) { | |
a426e122 | 1279 | if (ret > 0) { |
05330448 | 1280 | ret = -EINVAL; |
a426e122 | 1281 | } |
05330448 AL |
1282 | fprintf(stderr, "kvm version too old\n"); |
1283 | goto err; | |
1284 | } | |
1285 | ||
1286 | if (ret > KVM_API_VERSION) { | |
1287 | ret = -EINVAL; | |
1288 | fprintf(stderr, "kvm version not supported\n"); | |
1289 | goto err; | |
1290 | } | |
1291 | ||
3ed444e9 DH |
1292 | max_vcpus = kvm_max_vcpus(s); |
1293 | if (smp_cpus > max_vcpus) { | |
1294 | ret = -EINVAL; | |
1295 | fprintf(stderr, "Number of SMP cpus requested (%d) exceeds max cpus " | |
1296 | "supported by KVM (%d)\n", smp_cpus, max_vcpus); | |
1297 | goto err; | |
1298 | } | |
1299 | ||
05330448 | 1300 | s->vmfd = kvm_ioctl(s, KVM_CREATE_VM, 0); |
0104dcac AG |
1301 | if (s->vmfd < 0) { |
1302 | #ifdef TARGET_S390X | |
1303 | fprintf(stderr, "Please add the 'switch_amode' kernel parameter to " | |
1304 | "your host kernel command line\n"); | |
1305 | #endif | |
db9eae1c | 1306 | ret = s->vmfd; |
05330448 | 1307 | goto err; |
0104dcac | 1308 | } |
05330448 | 1309 | |
94a8d39a JK |
1310 | missing_cap = kvm_check_extension_list(s, kvm_required_capabilites); |
1311 | if (!missing_cap) { | |
1312 | missing_cap = | |
1313 | kvm_check_extension_list(s, kvm_arch_required_capabilities); | |
05330448 | 1314 | } |
94a8d39a | 1315 | if (missing_cap) { |
ad7b8b33 | 1316 | ret = -EINVAL; |
94a8d39a JK |
1317 | fprintf(stderr, "kvm does not support %s\n%s", |
1318 | missing_cap->name, upgrade_note); | |
d85dc283 AL |
1319 | goto err; |
1320 | } | |
1321 | ||
ad7b8b33 | 1322 | s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO); |
f65ed4c1 | 1323 | |
e69917e2 | 1324 | s->broken_set_mem_region = 1; |
14a09518 | 1325 | ret = kvm_check_extension(s, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS); |
e69917e2 JK |
1326 | if (ret > 0) { |
1327 | s->broken_set_mem_region = 0; | |
1328 | } | |
e69917e2 | 1329 | |
a0fb002c JK |
1330 | #ifdef KVM_CAP_VCPU_EVENTS |
1331 | s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS); | |
1332 | #endif | |
1333 | ||
b0b1d690 JK |
1334 | s->robust_singlestep = |
1335 | kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP); | |
b0b1d690 | 1336 | |
ff44f1a3 JK |
1337 | #ifdef KVM_CAP_DEBUGREGS |
1338 | s->debugregs = kvm_check_extension(s, KVM_CAP_DEBUGREGS); | |
1339 | #endif | |
1340 | ||
f1665b21 SY |
1341 | #ifdef KVM_CAP_XSAVE |
1342 | s->xsave = kvm_check_extension(s, KVM_CAP_XSAVE); | |
1343 | #endif | |
1344 | ||
f1665b21 SY |
1345 | #ifdef KVM_CAP_XCRS |
1346 | s->xcrs = kvm_check_extension(s, KVM_CAP_XCRS); | |
1347 | #endif | |
1348 | ||
8a7c7393 JK |
1349 | #ifdef KVM_CAP_PIT_STATE2 |
1350 | s->pit_state2 = kvm_check_extension(s, KVM_CAP_PIT_STATE2); | |
1351 | #endif | |
1352 | ||
d3d3bef0 | 1353 | #ifdef KVM_CAP_IRQ_ROUTING |
4a3adebb | 1354 | s->direct_msi = (kvm_check_extension(s, KVM_CAP_SIGNAL_MSI) > 0); |
d3d3bef0 | 1355 | #endif |
4a3adebb | 1356 | |
cad1e282 | 1357 | ret = kvm_arch_init(s); |
a426e122 | 1358 | if (ret < 0) { |
05330448 | 1359 | goto err; |
a426e122 | 1360 | } |
05330448 | 1361 | |
84b058d7 JK |
1362 | ret = kvm_irqchip_create(s); |
1363 | if (ret < 0) { | |
1364 | goto err; | |
1365 | } | |
1366 | ||
05330448 | 1367 | kvm_state = s; |
7376e582 | 1368 | memory_listener_register(&kvm_memory_listener, NULL); |
05330448 | 1369 | |
d2f2b8a7 SH |
1370 | s->many_ioeventfds = kvm_check_many_ioeventfds(); |
1371 | ||
aa7f74d1 JK |
1372 | cpu_interrupt_handler = kvm_handle_interrupt; |
1373 | ||
05330448 AL |
1374 | return 0; |
1375 | ||
1376 | err: | |
a78cedc0 SW |
1377 | if (s->vmfd >= 0) { |
1378 | close(s->vmfd); | |
1379 | } | |
1380 | if (s->fd != -1) { | |
1381 | close(s->fd); | |
05330448 | 1382 | } |
7267c094 | 1383 | g_free(s); |
05330448 AL |
1384 | |
1385 | return ret; | |
1386 | } | |
1387 | ||
b30e93e9 JK |
1388 | static void kvm_handle_io(uint16_t port, void *data, int direction, int size, |
1389 | uint32_t count) | |
05330448 AL |
1390 | { |
1391 | int i; | |
1392 | uint8_t *ptr = data; | |
1393 | ||
1394 | for (i = 0; i < count; i++) { | |
1395 | if (direction == KVM_EXIT_IO_IN) { | |
1396 | switch (size) { | |
1397 | case 1: | |
afcea8cb | 1398 | stb_p(ptr, cpu_inb(port)); |
05330448 AL |
1399 | break; |
1400 | case 2: | |
afcea8cb | 1401 | stw_p(ptr, cpu_inw(port)); |
05330448 AL |
1402 | break; |
1403 | case 4: | |
afcea8cb | 1404 | stl_p(ptr, cpu_inl(port)); |
05330448 AL |
1405 | break; |
1406 | } | |
1407 | } else { | |
1408 | switch (size) { | |
1409 | case 1: | |
afcea8cb | 1410 | cpu_outb(port, ldub_p(ptr)); |
05330448 AL |
1411 | break; |
1412 | case 2: | |
afcea8cb | 1413 | cpu_outw(port, lduw_p(ptr)); |
05330448 AL |
1414 | break; |
1415 | case 4: | |
afcea8cb | 1416 | cpu_outl(port, ldl_p(ptr)); |
05330448 AL |
1417 | break; |
1418 | } | |
1419 | } | |
1420 | ||
1421 | ptr += size; | |
1422 | } | |
05330448 AL |
1423 | } |
1424 | ||
9349b4f9 | 1425 | static int kvm_handle_internal_error(CPUArchState *env, struct kvm_run *run) |
7c80eef8 | 1426 | { |
bb44e0d1 | 1427 | fprintf(stderr, "KVM internal error."); |
7c80eef8 MT |
1428 | if (kvm_check_extension(kvm_state, KVM_CAP_INTERNAL_ERROR_DATA)) { |
1429 | int i; | |
1430 | ||
bb44e0d1 | 1431 | fprintf(stderr, " Suberror: %d\n", run->internal.suberror); |
7c80eef8 MT |
1432 | for (i = 0; i < run->internal.ndata; ++i) { |
1433 | fprintf(stderr, "extra data[%d]: %"PRIx64"\n", | |
1434 | i, (uint64_t)run->internal.data[i]); | |
1435 | } | |
bb44e0d1 JK |
1436 | } else { |
1437 | fprintf(stderr, "\n"); | |
7c80eef8 | 1438 | } |
7c80eef8 MT |
1439 | if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) { |
1440 | fprintf(stderr, "emulation failure\n"); | |
a426e122 | 1441 | if (!kvm_arch_stop_on_emulation_error(env)) { |
f5c848ee | 1442 | cpu_dump_state(env, stderr, fprintf, CPU_DUMP_CODE); |
d73cd8f4 | 1443 | return EXCP_INTERRUPT; |
a426e122 | 1444 | } |
7c80eef8 MT |
1445 | } |
1446 | /* FIXME: Should trigger a qmp message to let management know | |
1447 | * something went wrong. | |
1448 | */ | |
73aaec4a | 1449 | return -1; |
7c80eef8 | 1450 | } |
7c80eef8 | 1451 | |
62a2744c | 1452 | void kvm_flush_coalesced_mmio_buffer(void) |
f65ed4c1 | 1453 | { |
f65ed4c1 | 1454 | KVMState *s = kvm_state; |
1cae88b9 AK |
1455 | |
1456 | if (s->coalesced_flush_in_progress) { | |
1457 | return; | |
1458 | } | |
1459 | ||
1460 | s->coalesced_flush_in_progress = true; | |
1461 | ||
62a2744c SY |
1462 | if (s->coalesced_mmio_ring) { |
1463 | struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring; | |
f65ed4c1 AL |
1464 | while (ring->first != ring->last) { |
1465 | struct kvm_coalesced_mmio *ent; | |
1466 | ||
1467 | ent = &ring->coalesced_mmio[ring->first]; | |
1468 | ||
1469 | cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len); | |
85199474 | 1470 | smp_wmb(); |
f65ed4c1 AL |
1471 | ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX; |
1472 | } | |
1473 | } | |
1cae88b9 AK |
1474 | |
1475 | s->coalesced_flush_in_progress = false; | |
f65ed4c1 AL |
1476 | } |
1477 | ||
2705d56a | 1478 | static void do_kvm_cpu_synchronize_state(void *_env) |
4c0960c0 | 1479 | { |
9349b4f9 | 1480 | CPUArchState *env = _env; |
2705d56a | 1481 | |
9ded2744 | 1482 | if (!env->kvm_vcpu_dirty) { |
4c0960c0 | 1483 | kvm_arch_get_registers(env); |
9ded2744 | 1484 | env->kvm_vcpu_dirty = 1; |
4c0960c0 AK |
1485 | } |
1486 | } | |
1487 | ||
9349b4f9 | 1488 | void kvm_cpu_synchronize_state(CPUArchState *env) |
2705d56a | 1489 | { |
a426e122 | 1490 | if (!env->kvm_vcpu_dirty) { |
2705d56a | 1491 | run_on_cpu(env, do_kvm_cpu_synchronize_state, env); |
a426e122 | 1492 | } |
2705d56a JK |
1493 | } |
1494 | ||
9349b4f9 | 1495 | void kvm_cpu_synchronize_post_reset(CPUArchState *env) |
ea375f9a JK |
1496 | { |
1497 | kvm_arch_put_registers(env, KVM_PUT_RESET_STATE); | |
1498 | env->kvm_vcpu_dirty = 0; | |
1499 | } | |
1500 | ||
9349b4f9 | 1501 | void kvm_cpu_synchronize_post_init(CPUArchState *env) |
ea375f9a JK |
1502 | { |
1503 | kvm_arch_put_registers(env, KVM_PUT_FULL_STATE); | |
1504 | env->kvm_vcpu_dirty = 0; | |
1505 | } | |
1506 | ||
9349b4f9 | 1507 | int kvm_cpu_exec(CPUArchState *env) |
05330448 AL |
1508 | { |
1509 | struct kvm_run *run = env->kvm_run; | |
7cbb533f | 1510 | int ret, run_ret; |
05330448 | 1511 | |
8c0d577e | 1512 | DPRINTF("kvm_cpu_exec()\n"); |
05330448 | 1513 | |
99036865 | 1514 | if (kvm_arch_process_async_events(env)) { |
9ccfac9e | 1515 | env->exit_request = 0; |
6792a57b | 1516 | return EXCP_HLT; |
9ccfac9e | 1517 | } |
0af691d7 | 1518 | |
9ccfac9e | 1519 | do { |
9ded2744 | 1520 | if (env->kvm_vcpu_dirty) { |
ea375f9a | 1521 | kvm_arch_put_registers(env, KVM_PUT_RUNTIME_STATE); |
9ded2744 | 1522 | env->kvm_vcpu_dirty = 0; |
4c0960c0 AK |
1523 | } |
1524 | ||
8c14c173 | 1525 | kvm_arch_pre_run(env, run); |
9ccfac9e JK |
1526 | if (env->exit_request) { |
1527 | DPRINTF("interrupt exit requested\n"); | |
1528 | /* | |
1529 | * KVM requires us to reenter the kernel after IO exits to complete | |
1530 | * instruction emulation. This self-signal will ensure that we | |
1531 | * leave ASAP again. | |
1532 | */ | |
1533 | qemu_cpu_kick_self(); | |
1534 | } | |
d549db5a | 1535 | qemu_mutex_unlock_iothread(); |
9ccfac9e | 1536 | |
7cbb533f | 1537 | run_ret = kvm_vcpu_ioctl(env, KVM_RUN, 0); |
9ccfac9e | 1538 | |
d549db5a | 1539 | qemu_mutex_lock_iothread(); |
05330448 AL |
1540 | kvm_arch_post_run(env, run); |
1541 | ||
b0c883b5 JK |
1542 | kvm_flush_coalesced_mmio_buffer(); |
1543 | ||
7cbb533f | 1544 | if (run_ret < 0) { |
dc77d341 JK |
1545 | if (run_ret == -EINTR || run_ret == -EAGAIN) { |
1546 | DPRINTF("io window exit\n"); | |
d73cd8f4 | 1547 | ret = EXCP_INTERRUPT; |
dc77d341 JK |
1548 | break; |
1549 | } | |
7b011fbc ME |
1550 | fprintf(stderr, "error: kvm run failed %s\n", |
1551 | strerror(-run_ret)); | |
05330448 AL |
1552 | abort(); |
1553 | } | |
1554 | ||
05330448 AL |
1555 | switch (run->exit_reason) { |
1556 | case KVM_EXIT_IO: | |
8c0d577e | 1557 | DPRINTF("handle_io\n"); |
b30e93e9 JK |
1558 | kvm_handle_io(run->io.port, |
1559 | (uint8_t *)run + run->io.data_offset, | |
1560 | run->io.direction, | |
1561 | run->io.size, | |
1562 | run->io.count); | |
d73cd8f4 | 1563 | ret = 0; |
05330448 AL |
1564 | break; |
1565 | case KVM_EXIT_MMIO: | |
8c0d577e | 1566 | DPRINTF("handle_mmio\n"); |
05330448 AL |
1567 | cpu_physical_memory_rw(run->mmio.phys_addr, |
1568 | run->mmio.data, | |
1569 | run->mmio.len, | |
1570 | run->mmio.is_write); | |
d73cd8f4 | 1571 | ret = 0; |
05330448 AL |
1572 | break; |
1573 | case KVM_EXIT_IRQ_WINDOW_OPEN: | |
8c0d577e | 1574 | DPRINTF("irq_window_open\n"); |
d73cd8f4 | 1575 | ret = EXCP_INTERRUPT; |
05330448 AL |
1576 | break; |
1577 | case KVM_EXIT_SHUTDOWN: | |
8c0d577e | 1578 | DPRINTF("shutdown\n"); |
05330448 | 1579 | qemu_system_reset_request(); |
d73cd8f4 | 1580 | ret = EXCP_INTERRUPT; |
05330448 AL |
1581 | break; |
1582 | case KVM_EXIT_UNKNOWN: | |
bb44e0d1 JK |
1583 | fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "\n", |
1584 | (uint64_t)run->hw.hardware_exit_reason); | |
73aaec4a | 1585 | ret = -1; |
05330448 | 1586 | break; |
7c80eef8 | 1587 | case KVM_EXIT_INTERNAL_ERROR: |
73aaec4a | 1588 | ret = kvm_handle_internal_error(env, run); |
7c80eef8 | 1589 | break; |
05330448 | 1590 | default: |
8c0d577e | 1591 | DPRINTF("kvm_arch_handle_exit\n"); |
05330448 AL |
1592 | ret = kvm_arch_handle_exit(env, run); |
1593 | break; | |
1594 | } | |
d73cd8f4 | 1595 | } while (ret == 0); |
05330448 | 1596 | |
73aaec4a | 1597 | if (ret < 0) { |
f5c848ee | 1598 | cpu_dump_state(env, stderr, fprintf, CPU_DUMP_CODE); |
0461d5a6 | 1599 | vm_stop(RUN_STATE_INTERNAL_ERROR); |
becfc390 AL |
1600 | } |
1601 | ||
6792a57b | 1602 | env->exit_request = 0; |
05330448 AL |
1603 | return ret; |
1604 | } | |
1605 | ||
984b5181 | 1606 | int kvm_ioctl(KVMState *s, int type, ...) |
05330448 AL |
1607 | { |
1608 | int ret; | |
984b5181 AL |
1609 | void *arg; |
1610 | va_list ap; | |
05330448 | 1611 | |
984b5181 AL |
1612 | va_start(ap, type); |
1613 | arg = va_arg(ap, void *); | |
1614 | va_end(ap); | |
1615 | ||
1616 | ret = ioctl(s->fd, type, arg); | |
a426e122 | 1617 | if (ret == -1) { |
05330448 | 1618 | ret = -errno; |
a426e122 | 1619 | } |
05330448 AL |
1620 | return ret; |
1621 | } | |
1622 | ||
984b5181 | 1623 | int kvm_vm_ioctl(KVMState *s, int type, ...) |
05330448 AL |
1624 | { |
1625 | int ret; | |
984b5181 AL |
1626 | void *arg; |
1627 | va_list ap; | |
1628 | ||
1629 | va_start(ap, type); | |
1630 | arg = va_arg(ap, void *); | |
1631 | va_end(ap); | |
05330448 | 1632 | |
984b5181 | 1633 | ret = ioctl(s->vmfd, type, arg); |
a426e122 | 1634 | if (ret == -1) { |
05330448 | 1635 | ret = -errno; |
a426e122 | 1636 | } |
05330448 AL |
1637 | return ret; |
1638 | } | |
1639 | ||
9349b4f9 | 1640 | int kvm_vcpu_ioctl(CPUArchState *env, int type, ...) |
05330448 AL |
1641 | { |
1642 | int ret; | |
984b5181 AL |
1643 | void *arg; |
1644 | va_list ap; | |
1645 | ||
1646 | va_start(ap, type); | |
1647 | arg = va_arg(ap, void *); | |
1648 | va_end(ap); | |
05330448 | 1649 | |
984b5181 | 1650 | ret = ioctl(env->kvm_fd, type, arg); |
a426e122 | 1651 | if (ret == -1) { |
05330448 | 1652 | ret = -errno; |
a426e122 | 1653 | } |
05330448 AL |
1654 | return ret; |
1655 | } | |
bd322087 AL |
1656 | |
1657 | int kvm_has_sync_mmu(void) | |
1658 | { | |
94a8d39a | 1659 | return kvm_check_extension(kvm_state, KVM_CAP_SYNC_MMU); |
bd322087 | 1660 | } |
e22a25c9 | 1661 | |
a0fb002c JK |
1662 | int kvm_has_vcpu_events(void) |
1663 | { | |
1664 | return kvm_state->vcpu_events; | |
1665 | } | |
1666 | ||
b0b1d690 JK |
1667 | int kvm_has_robust_singlestep(void) |
1668 | { | |
1669 | return kvm_state->robust_singlestep; | |
1670 | } | |
1671 | ||
ff44f1a3 JK |
1672 | int kvm_has_debugregs(void) |
1673 | { | |
1674 | return kvm_state->debugregs; | |
1675 | } | |
1676 | ||
f1665b21 SY |
1677 | int kvm_has_xsave(void) |
1678 | { | |
1679 | return kvm_state->xsave; | |
1680 | } | |
1681 | ||
1682 | int kvm_has_xcrs(void) | |
1683 | { | |
1684 | return kvm_state->xcrs; | |
1685 | } | |
1686 | ||
8a7c7393 JK |
1687 | int kvm_has_pit_state2(void) |
1688 | { | |
1689 | return kvm_state->pit_state2; | |
1690 | } | |
1691 | ||
d2f2b8a7 SH |
1692 | int kvm_has_many_ioeventfds(void) |
1693 | { | |
1694 | if (!kvm_enabled()) { | |
1695 | return 0; | |
1696 | } | |
1697 | return kvm_state->many_ioeventfds; | |
1698 | } | |
1699 | ||
84b058d7 JK |
1700 | int kvm_has_gsi_routing(void) |
1701 | { | |
a9c5eb0d | 1702 | #ifdef KVM_CAP_IRQ_ROUTING |
84b058d7 | 1703 | return kvm_check_extension(kvm_state, KVM_CAP_IRQ_ROUTING); |
a9c5eb0d AG |
1704 | #else |
1705 | return false; | |
1706 | #endif | |
84b058d7 JK |
1707 | } |
1708 | ||
fdec9918 CB |
1709 | void *kvm_vmalloc(ram_addr_t size) |
1710 | { | |
1711 | #ifdef TARGET_S390X | |
1712 | void *mem; | |
1713 | ||
1714 | mem = kvm_arch_vmalloc(size); | |
1715 | if (mem) { | |
1716 | return mem; | |
1717 | } | |
1718 | #endif | |
1719 | return qemu_vmalloc(size); | |
1720 | } | |
1721 | ||
6f0437e8 JK |
1722 | void kvm_setup_guest_memory(void *start, size_t size) |
1723 | { | |
7566759f CB |
1724 | #ifdef CONFIG_VALGRIND_H |
1725 | VALGRIND_MAKE_MEM_DEFINED(start, size); | |
1726 | #endif | |
6f0437e8 | 1727 | if (!kvm_has_sync_mmu()) { |
e78815a5 | 1728 | int ret = qemu_madvise(start, size, QEMU_MADV_DONTFORK); |
6f0437e8 JK |
1729 | |
1730 | if (ret) { | |
e78815a5 AF |
1731 | perror("qemu_madvise"); |
1732 | fprintf(stderr, | |
1733 | "Need MADV_DONTFORK in absence of synchronous KVM MMU\n"); | |
6f0437e8 JK |
1734 | exit(1); |
1735 | } | |
6f0437e8 JK |
1736 | } |
1737 | } | |
1738 | ||
e22a25c9 | 1739 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
9349b4f9 | 1740 | struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUArchState *env, |
e22a25c9 AL |
1741 | target_ulong pc) |
1742 | { | |
1743 | struct kvm_sw_breakpoint *bp; | |
1744 | ||
72cf2d4f | 1745 | QTAILQ_FOREACH(bp, &env->kvm_state->kvm_sw_breakpoints, entry) { |
a426e122 | 1746 | if (bp->pc == pc) { |
e22a25c9 | 1747 | return bp; |
a426e122 | 1748 | } |
e22a25c9 AL |
1749 | } |
1750 | return NULL; | |
1751 | } | |
1752 | ||
9349b4f9 | 1753 | int kvm_sw_breakpoints_active(CPUArchState *env) |
e22a25c9 | 1754 | { |
72cf2d4f | 1755 | return !QTAILQ_EMPTY(&env->kvm_state->kvm_sw_breakpoints); |
e22a25c9 AL |
1756 | } |
1757 | ||
452e4751 GC |
1758 | struct kvm_set_guest_debug_data { |
1759 | struct kvm_guest_debug dbg; | |
9349b4f9 | 1760 | CPUArchState *env; |
452e4751 GC |
1761 | int err; |
1762 | }; | |
1763 | ||
1764 | static void kvm_invoke_set_guest_debug(void *data) | |
1765 | { | |
1766 | struct kvm_set_guest_debug_data *dbg_data = data; | |
9349b4f9 | 1767 | CPUArchState *env = dbg_data->env; |
b3807725 | 1768 | |
b3807725 | 1769 | dbg_data->err = kvm_vcpu_ioctl(env, KVM_SET_GUEST_DEBUG, &dbg_data->dbg); |
452e4751 GC |
1770 | } |
1771 | ||
9349b4f9 | 1772 | int kvm_update_guest_debug(CPUArchState *env, unsigned long reinject_trap) |
e22a25c9 | 1773 | { |
452e4751 | 1774 | struct kvm_set_guest_debug_data data; |
e22a25c9 | 1775 | |
b0b1d690 | 1776 | data.dbg.control = reinject_trap; |
e22a25c9 | 1777 | |
b0b1d690 JK |
1778 | if (env->singlestep_enabled) { |
1779 | data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP; | |
1780 | } | |
452e4751 | 1781 | kvm_arch_update_guest_debug(env, &data.dbg); |
452e4751 | 1782 | data.env = env; |
e22a25c9 | 1783 | |
be41cbe0 | 1784 | run_on_cpu(env, kvm_invoke_set_guest_debug, &data); |
452e4751 | 1785 | return data.err; |
e22a25c9 AL |
1786 | } |
1787 | ||
9349b4f9 | 1788 | int kvm_insert_breakpoint(CPUArchState *current_env, target_ulong addr, |
e22a25c9 AL |
1789 | target_ulong len, int type) |
1790 | { | |
1791 | struct kvm_sw_breakpoint *bp; | |
9349b4f9 | 1792 | CPUArchState *env; |
e22a25c9 AL |
1793 | int err; |
1794 | ||
1795 | if (type == GDB_BREAKPOINT_SW) { | |
1796 | bp = kvm_find_sw_breakpoint(current_env, addr); | |
1797 | if (bp) { | |
1798 | bp->use_count++; | |
1799 | return 0; | |
1800 | } | |
1801 | ||
7267c094 | 1802 | bp = g_malloc(sizeof(struct kvm_sw_breakpoint)); |
a426e122 | 1803 | if (!bp) { |
e22a25c9 | 1804 | return -ENOMEM; |
a426e122 | 1805 | } |
e22a25c9 AL |
1806 | |
1807 | bp->pc = addr; | |
1808 | bp->use_count = 1; | |
1809 | err = kvm_arch_insert_sw_breakpoint(current_env, bp); | |
1810 | if (err) { | |
7267c094 | 1811 | g_free(bp); |
e22a25c9 AL |
1812 | return err; |
1813 | } | |
1814 | ||
72cf2d4f | 1815 | QTAILQ_INSERT_HEAD(¤t_env->kvm_state->kvm_sw_breakpoints, |
e22a25c9 AL |
1816 | bp, entry); |
1817 | } else { | |
1818 | err = kvm_arch_insert_hw_breakpoint(addr, len, type); | |
a426e122 | 1819 | if (err) { |
e22a25c9 | 1820 | return err; |
a426e122 | 1821 | } |
e22a25c9 AL |
1822 | } |
1823 | ||
1824 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
1825 | err = kvm_update_guest_debug(env, 0); | |
a426e122 | 1826 | if (err) { |
e22a25c9 | 1827 | return err; |
a426e122 | 1828 | } |
e22a25c9 AL |
1829 | } |
1830 | return 0; | |
1831 | } | |
1832 | ||
9349b4f9 | 1833 | int kvm_remove_breakpoint(CPUArchState *current_env, target_ulong addr, |
e22a25c9 AL |
1834 | target_ulong len, int type) |
1835 | { | |
1836 | struct kvm_sw_breakpoint *bp; | |
9349b4f9 | 1837 | CPUArchState *env; |
e22a25c9 AL |
1838 | int err; |
1839 | ||
1840 | if (type == GDB_BREAKPOINT_SW) { | |
1841 | bp = kvm_find_sw_breakpoint(current_env, addr); | |
a426e122 | 1842 | if (!bp) { |
e22a25c9 | 1843 | return -ENOENT; |
a426e122 | 1844 | } |
e22a25c9 AL |
1845 | |
1846 | if (bp->use_count > 1) { | |
1847 | bp->use_count--; | |
1848 | return 0; | |
1849 | } | |
1850 | ||
1851 | err = kvm_arch_remove_sw_breakpoint(current_env, bp); | |
a426e122 | 1852 | if (err) { |
e22a25c9 | 1853 | return err; |
a426e122 | 1854 | } |
e22a25c9 | 1855 | |
72cf2d4f | 1856 | QTAILQ_REMOVE(¤t_env->kvm_state->kvm_sw_breakpoints, bp, entry); |
7267c094 | 1857 | g_free(bp); |
e22a25c9 AL |
1858 | } else { |
1859 | err = kvm_arch_remove_hw_breakpoint(addr, len, type); | |
a426e122 | 1860 | if (err) { |
e22a25c9 | 1861 | return err; |
a426e122 | 1862 | } |
e22a25c9 AL |
1863 | } |
1864 | ||
1865 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
1866 | err = kvm_update_guest_debug(env, 0); | |
a426e122 | 1867 | if (err) { |
e22a25c9 | 1868 | return err; |
a426e122 | 1869 | } |
e22a25c9 AL |
1870 | } |
1871 | return 0; | |
1872 | } | |
1873 | ||
9349b4f9 | 1874 | void kvm_remove_all_breakpoints(CPUArchState *current_env) |
e22a25c9 AL |
1875 | { |
1876 | struct kvm_sw_breakpoint *bp, *next; | |
1877 | KVMState *s = current_env->kvm_state; | |
9349b4f9 | 1878 | CPUArchState *env; |
e22a25c9 | 1879 | |
72cf2d4f | 1880 | QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) { |
e22a25c9 AL |
1881 | if (kvm_arch_remove_sw_breakpoint(current_env, bp) != 0) { |
1882 | /* Try harder to find a CPU that currently sees the breakpoint. */ | |
1883 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
a426e122 | 1884 | if (kvm_arch_remove_sw_breakpoint(env, bp) == 0) { |
e22a25c9 | 1885 | break; |
a426e122 | 1886 | } |
e22a25c9 AL |
1887 | } |
1888 | } | |
1889 | } | |
1890 | kvm_arch_remove_all_hw_breakpoints(); | |
1891 | ||
a426e122 | 1892 | for (env = first_cpu; env != NULL; env = env->next_cpu) { |
e22a25c9 | 1893 | kvm_update_guest_debug(env, 0); |
a426e122 | 1894 | } |
e22a25c9 AL |
1895 | } |
1896 | ||
1897 | #else /* !KVM_CAP_SET_GUEST_DEBUG */ | |
1898 | ||
9349b4f9 | 1899 | int kvm_update_guest_debug(CPUArchState *env, unsigned long reinject_trap) |
e22a25c9 AL |
1900 | { |
1901 | return -EINVAL; | |
1902 | } | |
1903 | ||
9349b4f9 | 1904 | int kvm_insert_breakpoint(CPUArchState *current_env, target_ulong addr, |
e22a25c9 AL |
1905 | target_ulong len, int type) |
1906 | { | |
1907 | return -EINVAL; | |
1908 | } | |
1909 | ||
9349b4f9 | 1910 | int kvm_remove_breakpoint(CPUArchState *current_env, target_ulong addr, |
e22a25c9 AL |
1911 | target_ulong len, int type) |
1912 | { | |
1913 | return -EINVAL; | |
1914 | } | |
1915 | ||
9349b4f9 | 1916 | void kvm_remove_all_breakpoints(CPUArchState *current_env) |
e22a25c9 AL |
1917 | { |
1918 | } | |
1919 | #endif /* !KVM_CAP_SET_GUEST_DEBUG */ | |
cc84de95 | 1920 | |
9349b4f9 | 1921 | int kvm_set_signal_mask(CPUArchState *env, const sigset_t *sigset) |
cc84de95 MT |
1922 | { |
1923 | struct kvm_signal_mask *sigmask; | |
1924 | int r; | |
1925 | ||
a426e122 | 1926 | if (!sigset) { |
cc84de95 | 1927 | return kvm_vcpu_ioctl(env, KVM_SET_SIGNAL_MASK, NULL); |
a426e122 | 1928 | } |
cc84de95 | 1929 | |
7267c094 | 1930 | sigmask = g_malloc(sizeof(*sigmask) + sizeof(*sigset)); |
cc84de95 MT |
1931 | |
1932 | sigmask->len = 8; | |
1933 | memcpy(sigmask->sigset, sigset, sizeof(*sigset)); | |
1934 | r = kvm_vcpu_ioctl(env, KVM_SET_SIGNAL_MASK, sigmask); | |
7267c094 | 1935 | g_free(sigmask); |
cc84de95 MT |
1936 | |
1937 | return r; | |
1938 | } | |
ca821806 | 1939 | |
4b8f1c88 MT |
1940 | int kvm_set_ioeventfd_mmio(int fd, uint32_t addr, uint32_t val, bool assign, |
1941 | uint32_t size) | |
44f1a3d8 | 1942 | { |
44f1a3d8 CM |
1943 | int ret; |
1944 | struct kvm_ioeventfd iofd; | |
1945 | ||
1946 | iofd.datamatch = val; | |
1947 | iofd.addr = addr; | |
4b8f1c88 | 1948 | iofd.len = size; |
44f1a3d8 CM |
1949 | iofd.flags = KVM_IOEVENTFD_FLAG_DATAMATCH; |
1950 | iofd.fd = fd; | |
1951 | ||
1952 | if (!kvm_enabled()) { | |
1953 | return -ENOSYS; | |
1954 | } | |
1955 | ||
1956 | if (!assign) { | |
1957 | iofd.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; | |
1958 | } | |
1959 | ||
1960 | ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd); | |
1961 | ||
1962 | if (ret < 0) { | |
1963 | return -errno; | |
1964 | } | |
1965 | ||
1966 | return 0; | |
44f1a3d8 CM |
1967 | } |
1968 | ||
ca821806 MT |
1969 | int kvm_set_ioeventfd_pio_word(int fd, uint16_t addr, uint16_t val, bool assign) |
1970 | { | |
1971 | struct kvm_ioeventfd kick = { | |
1972 | .datamatch = val, | |
1973 | .addr = addr, | |
1974 | .len = 2, | |
1975 | .flags = KVM_IOEVENTFD_FLAG_DATAMATCH | KVM_IOEVENTFD_FLAG_PIO, | |
1976 | .fd = fd, | |
1977 | }; | |
1978 | int r; | |
a426e122 | 1979 | if (!kvm_enabled()) { |
ca821806 | 1980 | return -ENOSYS; |
a426e122 JK |
1981 | } |
1982 | if (!assign) { | |
ca821806 | 1983 | kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; |
a426e122 | 1984 | } |
ca821806 | 1985 | r = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick); |
a426e122 | 1986 | if (r < 0) { |
ca821806 | 1987 | return r; |
a426e122 | 1988 | } |
ca821806 | 1989 | return 0; |
98c8573e | 1990 | } |
a1b87fe0 | 1991 | |
9349b4f9 | 1992 | int kvm_on_sigbus_vcpu(CPUArchState *env, int code, void *addr) |
a1b87fe0 JK |
1993 | { |
1994 | return kvm_arch_on_sigbus_vcpu(env, code, addr); | |
1995 | } | |
1996 | ||
1997 | int kvm_on_sigbus(int code, void *addr) | |
1998 | { | |
1999 | return kvm_arch_on_sigbus(code, addr); | |
2000 | } |