]>
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" |
05330448 | 25 | #include "sysemu.h" |
d33a1810 | 26 | #include "hw/hw.h" |
04fa27f5 | 27 | #include "hw/msi.h" |
e22a25c9 | 28 | #include "gdbstub.h" |
05330448 | 29 | #include "kvm.h" |
8369e01c | 30 | #include "bswap.h" |
a01672d3 | 31 | #include "memory.h" |
80a1ea37 | 32 | #include "exec-memory.h" |
05330448 | 33 | |
d2f2b8a7 SH |
34 | /* This check must be after config-host.h is included */ |
35 | #ifdef CONFIG_EVENTFD | |
36 | #include <sys/eventfd.h> | |
37 | #endif | |
38 | ||
93148aa5 | 39 | /* KVM uses PAGE_SIZE in its definition of COALESCED_MMIO_MAX */ |
f65ed4c1 AL |
40 | #define PAGE_SIZE TARGET_PAGE_SIZE |
41 | ||
05330448 AL |
42 | //#define DEBUG_KVM |
43 | ||
44 | #ifdef DEBUG_KVM | |
8c0d577e | 45 | #define DPRINTF(fmt, ...) \ |
05330448 AL |
46 | do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0) |
47 | #else | |
8c0d577e | 48 | #define DPRINTF(fmt, ...) \ |
05330448 AL |
49 | do { } while (0) |
50 | #endif | |
51 | ||
04fa27f5 JK |
52 | #define KVM_MSI_HASHTAB_SIZE 256 |
53 | ||
34fc643f AL |
54 | typedef struct KVMSlot |
55 | { | |
c227f099 AL |
56 | target_phys_addr_t start_addr; |
57 | ram_addr_t memory_size; | |
9f213ed9 | 58 | void *ram; |
34fc643f AL |
59 | int slot; |
60 | int flags; | |
61 | } KVMSlot; | |
05330448 | 62 | |
5832d1f2 AL |
63 | typedef struct kvm_dirty_log KVMDirtyLog; |
64 | ||
04fa27f5 JK |
65 | typedef struct KVMMSIRoute { |
66 | struct kvm_irq_routing_entry kroute; | |
67 | QTAILQ_ENTRY(KVMMSIRoute) entry; | |
68 | } KVMMSIRoute; | |
69 | ||
05330448 AL |
70 | struct KVMState |
71 | { | |
72 | KVMSlot slots[32]; | |
73 | int fd; | |
74 | int vmfd; | |
f65ed4c1 | 75 | int coalesced_mmio; |
62a2744c | 76 | struct kvm_coalesced_mmio_ring *coalesced_mmio_ring; |
1cae88b9 | 77 | bool coalesced_flush_in_progress; |
e69917e2 | 78 | int broken_set_mem_region; |
4495d6a7 | 79 | int migration_log; |
a0fb002c | 80 | int vcpu_events; |
b0b1d690 | 81 | int robust_singlestep; |
ff44f1a3 | 82 | int debugregs; |
e22a25c9 AL |
83 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
84 | struct kvm_sw_breakpoint_head kvm_sw_breakpoints; | |
85 | #endif | |
8a7c7393 | 86 | int pit_state2; |
f1665b21 | 87 | int xsave, xcrs; |
d2f2b8a7 | 88 | int many_ioeventfds; |
92e4b519 DG |
89 | /* The man page (and posix) say ioctl numbers are signed int, but |
90 | * they're not. Linux, glibc and *BSD all treat ioctl numbers as | |
91 | * unsigned, and treating them as signed here can break things */ | |
92 | unsigned irqchip_inject_ioctl; | |
84b058d7 JK |
93 | #ifdef KVM_CAP_IRQ_ROUTING |
94 | struct kvm_irq_routing *irq_routes; | |
95 | int nr_allocated_irq_routes; | |
96 | uint32_t *used_gsi_bitmap; | |
4e2e4e63 | 97 | unsigned int gsi_count; |
04fa27f5 | 98 | QTAILQ_HEAD(msi_hashtab, KVMMSIRoute) msi_hashtab[KVM_MSI_HASHTAB_SIZE]; |
4a3adebb | 99 | bool direct_msi; |
84b058d7 | 100 | #endif |
05330448 AL |
101 | }; |
102 | ||
6a7af8cb | 103 | KVMState *kvm_state; |
3d4b2649 | 104 | bool kvm_kernel_irqchip; |
05330448 | 105 | |
94a8d39a JK |
106 | static const KVMCapabilityInfo kvm_required_capabilites[] = { |
107 | KVM_CAP_INFO(USER_MEMORY), | |
108 | KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS), | |
109 | KVM_CAP_LAST_INFO | |
110 | }; | |
111 | ||
05330448 AL |
112 | static KVMSlot *kvm_alloc_slot(KVMState *s) |
113 | { | |
114 | int i; | |
115 | ||
116 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
a426e122 | 117 | if (s->slots[i].memory_size == 0) { |
05330448 | 118 | return &s->slots[i]; |
a426e122 | 119 | } |
05330448 AL |
120 | } |
121 | ||
d3f8d37f AL |
122 | fprintf(stderr, "%s: no free slot available\n", __func__); |
123 | abort(); | |
124 | } | |
125 | ||
126 | static KVMSlot *kvm_lookup_matching_slot(KVMState *s, | |
c227f099 AL |
127 | target_phys_addr_t start_addr, |
128 | target_phys_addr_t end_addr) | |
d3f8d37f AL |
129 | { |
130 | int i; | |
131 | ||
132 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
133 | KVMSlot *mem = &s->slots[i]; | |
134 | ||
135 | if (start_addr == mem->start_addr && | |
136 | end_addr == mem->start_addr + mem->memory_size) { | |
137 | return mem; | |
138 | } | |
139 | } | |
140 | ||
05330448 AL |
141 | return NULL; |
142 | } | |
143 | ||
6152e2ae AL |
144 | /* |
145 | * Find overlapping slot with lowest start address | |
146 | */ | |
147 | static KVMSlot *kvm_lookup_overlapping_slot(KVMState *s, | |
c227f099 AL |
148 | target_phys_addr_t start_addr, |
149 | target_phys_addr_t end_addr) | |
05330448 | 150 | { |
6152e2ae | 151 | KVMSlot *found = NULL; |
05330448 AL |
152 | int i; |
153 | ||
154 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
155 | KVMSlot *mem = &s->slots[i]; | |
156 | ||
6152e2ae AL |
157 | if (mem->memory_size == 0 || |
158 | (found && found->start_addr < mem->start_addr)) { | |
159 | continue; | |
160 | } | |
161 | ||
162 | if (end_addr > mem->start_addr && | |
163 | start_addr < mem->start_addr + mem->memory_size) { | |
164 | found = mem; | |
165 | } | |
05330448 AL |
166 | } |
167 | ||
6152e2ae | 168 | return found; |
05330448 AL |
169 | } |
170 | ||
9f213ed9 AK |
171 | int kvm_physical_memory_addr_from_host(KVMState *s, void *ram, |
172 | target_phys_addr_t *phys_addr) | |
983dfc3b HY |
173 | { |
174 | int i; | |
175 | ||
176 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
177 | KVMSlot *mem = &s->slots[i]; | |
178 | ||
9f213ed9 AK |
179 | if (ram >= mem->ram && ram < mem->ram + mem->memory_size) { |
180 | *phys_addr = mem->start_addr + (ram - mem->ram); | |
983dfc3b HY |
181 | return 1; |
182 | } | |
183 | } | |
184 | ||
185 | return 0; | |
186 | } | |
187 | ||
5832d1f2 AL |
188 | static int kvm_set_user_memory_region(KVMState *s, KVMSlot *slot) |
189 | { | |
190 | struct kvm_userspace_memory_region mem; | |
191 | ||
192 | mem.slot = slot->slot; | |
193 | mem.guest_phys_addr = slot->start_addr; | |
194 | mem.memory_size = slot->memory_size; | |
9f213ed9 | 195 | mem.userspace_addr = (unsigned long)slot->ram; |
5832d1f2 | 196 | mem.flags = slot->flags; |
4495d6a7 JK |
197 | if (s->migration_log) { |
198 | mem.flags |= KVM_MEM_LOG_DIRTY_PAGES; | |
199 | } | |
5832d1f2 AL |
200 | return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem); |
201 | } | |
202 | ||
8d2ba1fb JK |
203 | static void kvm_reset_vcpu(void *opaque) |
204 | { | |
9349b4f9 | 205 | CPUArchState *env = opaque; |
8d2ba1fb | 206 | |
caa5af0f | 207 | kvm_arch_reset_vcpu(env); |
8d2ba1fb | 208 | } |
5832d1f2 | 209 | |
9349b4f9 | 210 | int kvm_init_vcpu(CPUArchState *env) |
05330448 AL |
211 | { |
212 | KVMState *s = kvm_state; | |
213 | long mmap_size; | |
214 | int ret; | |
215 | ||
8c0d577e | 216 | DPRINTF("kvm_init_vcpu\n"); |
05330448 | 217 | |
984b5181 | 218 | ret = kvm_vm_ioctl(s, KVM_CREATE_VCPU, env->cpu_index); |
05330448 | 219 | if (ret < 0) { |
8c0d577e | 220 | DPRINTF("kvm_create_vcpu failed\n"); |
05330448 AL |
221 | goto err; |
222 | } | |
223 | ||
224 | env->kvm_fd = ret; | |
225 | env->kvm_state = s; | |
d841b6c4 | 226 | env->kvm_vcpu_dirty = 1; |
05330448 AL |
227 | |
228 | mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0); | |
229 | if (mmap_size < 0) { | |
748a680b | 230 | ret = mmap_size; |
8c0d577e | 231 | DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n"); |
05330448 AL |
232 | goto err; |
233 | } | |
234 | ||
235 | env->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED, | |
236 | env->kvm_fd, 0); | |
237 | if (env->kvm_run == MAP_FAILED) { | |
238 | ret = -errno; | |
8c0d577e | 239 | DPRINTF("mmap'ing vcpu state failed\n"); |
05330448 AL |
240 | goto err; |
241 | } | |
242 | ||
a426e122 JK |
243 | if (s->coalesced_mmio && !s->coalesced_mmio_ring) { |
244 | s->coalesced_mmio_ring = | |
245 | (void *)env->kvm_run + s->coalesced_mmio * PAGE_SIZE; | |
246 | } | |
62a2744c | 247 | |
05330448 | 248 | ret = kvm_arch_init_vcpu(env); |
8d2ba1fb | 249 | if (ret == 0) { |
a08d4367 | 250 | qemu_register_reset(kvm_reset_vcpu, env); |
caa5af0f | 251 | kvm_arch_reset_vcpu(env); |
8d2ba1fb | 252 | } |
05330448 AL |
253 | err: |
254 | return ret; | |
255 | } | |
256 | ||
5832d1f2 AL |
257 | /* |
258 | * dirty pages logging control | |
259 | */ | |
25254bbc MT |
260 | |
261 | static int kvm_mem_flags(KVMState *s, bool log_dirty) | |
262 | { | |
263 | return log_dirty ? KVM_MEM_LOG_DIRTY_PAGES : 0; | |
264 | } | |
265 | ||
266 | static int kvm_slot_dirty_pages_log_change(KVMSlot *mem, bool log_dirty) | |
5832d1f2 AL |
267 | { |
268 | KVMState *s = kvm_state; | |
25254bbc | 269 | int flags, mask = KVM_MEM_LOG_DIRTY_PAGES; |
4495d6a7 JK |
270 | int old_flags; |
271 | ||
4495d6a7 | 272 | old_flags = mem->flags; |
5832d1f2 | 273 | |
25254bbc | 274 | flags = (mem->flags & ~mask) | kvm_mem_flags(s, log_dirty); |
5832d1f2 AL |
275 | mem->flags = flags; |
276 | ||
4495d6a7 JK |
277 | /* If nothing changed effectively, no need to issue ioctl */ |
278 | if (s->migration_log) { | |
279 | flags |= KVM_MEM_LOG_DIRTY_PAGES; | |
280 | } | |
25254bbc | 281 | |
4495d6a7 | 282 | if (flags == old_flags) { |
25254bbc | 283 | return 0; |
4495d6a7 JK |
284 | } |
285 | ||
5832d1f2 AL |
286 | return kvm_set_user_memory_region(s, mem); |
287 | } | |
288 | ||
25254bbc MT |
289 | static int kvm_dirty_pages_log_change(target_phys_addr_t phys_addr, |
290 | ram_addr_t size, bool log_dirty) | |
291 | { | |
292 | KVMState *s = kvm_state; | |
293 | KVMSlot *mem = kvm_lookup_matching_slot(s, phys_addr, phys_addr + size); | |
294 | ||
295 | if (mem == NULL) { | |
296 | fprintf(stderr, "BUG: %s: invalid parameters " TARGET_FMT_plx "-" | |
297 | TARGET_FMT_plx "\n", __func__, phys_addr, | |
298 | (target_phys_addr_t)(phys_addr + size - 1)); | |
299 | return -EINVAL; | |
300 | } | |
301 | return kvm_slot_dirty_pages_log_change(mem, log_dirty); | |
302 | } | |
303 | ||
a01672d3 AK |
304 | static void kvm_log_start(MemoryListener *listener, |
305 | MemoryRegionSection *section) | |
5832d1f2 | 306 | { |
a01672d3 AK |
307 | int r; |
308 | ||
309 | r = kvm_dirty_pages_log_change(section->offset_within_address_space, | |
310 | section->size, true); | |
311 | if (r < 0) { | |
312 | abort(); | |
313 | } | |
5832d1f2 AL |
314 | } |
315 | ||
a01672d3 AK |
316 | static void kvm_log_stop(MemoryListener *listener, |
317 | MemoryRegionSection *section) | |
5832d1f2 | 318 | { |
a01672d3 AK |
319 | int r; |
320 | ||
321 | r = kvm_dirty_pages_log_change(section->offset_within_address_space, | |
322 | section->size, false); | |
323 | if (r < 0) { | |
324 | abort(); | |
325 | } | |
5832d1f2 AL |
326 | } |
327 | ||
7b8f3b78 | 328 | static int kvm_set_migration_log(int enable) |
4495d6a7 JK |
329 | { |
330 | KVMState *s = kvm_state; | |
331 | KVMSlot *mem; | |
332 | int i, err; | |
333 | ||
334 | s->migration_log = enable; | |
335 | ||
336 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { | |
337 | mem = &s->slots[i]; | |
338 | ||
70fedd76 AW |
339 | if (!mem->memory_size) { |
340 | continue; | |
341 | } | |
4495d6a7 JK |
342 | if (!!(mem->flags & KVM_MEM_LOG_DIRTY_PAGES) == enable) { |
343 | continue; | |
344 | } | |
345 | err = kvm_set_user_memory_region(s, mem); | |
346 | if (err) { | |
347 | return err; | |
348 | } | |
349 | } | |
350 | return 0; | |
351 | } | |
352 | ||
8369e01c | 353 | /* get kvm's dirty pages bitmap and update qemu's */ |
ffcde12f AK |
354 | static int kvm_get_dirty_pages_log_range(MemoryRegionSection *section, |
355 | unsigned long *bitmap) | |
96c1606b | 356 | { |
8369e01c | 357 | unsigned int i, j; |
aa90fec7 BH |
358 | unsigned long page_number, c; |
359 | target_phys_addr_t addr, addr1; | |
ffcde12f | 360 | unsigned int len = ((section->size / TARGET_PAGE_SIZE) + HOST_LONG_BITS - 1) / HOST_LONG_BITS; |
3145fcb6 | 361 | unsigned long hpratio = getpagesize() / TARGET_PAGE_SIZE; |
8369e01c MT |
362 | |
363 | /* | |
364 | * bitmap-traveling is faster than memory-traveling (for addr...) | |
365 | * especially when most of the memory is not dirty. | |
366 | */ | |
367 | for (i = 0; i < len; i++) { | |
368 | if (bitmap[i] != 0) { | |
369 | c = leul_to_cpu(bitmap[i]); | |
370 | do { | |
371 | j = ffsl(c) - 1; | |
372 | c &= ~(1ul << j); | |
3145fcb6 | 373 | page_number = (i * HOST_LONG_BITS + j) * hpratio; |
8369e01c | 374 | addr1 = page_number * TARGET_PAGE_SIZE; |
ffcde12f | 375 | addr = section->offset_within_region + addr1; |
3145fcb6 DG |
376 | memory_region_set_dirty(section->mr, addr, |
377 | TARGET_PAGE_SIZE * hpratio); | |
8369e01c MT |
378 | } while (c != 0); |
379 | } | |
380 | } | |
381 | return 0; | |
96c1606b AG |
382 | } |
383 | ||
8369e01c MT |
384 | #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1)) |
385 | ||
5832d1f2 AL |
386 | /** |
387 | * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space | |
fd4aa979 BS |
388 | * This function updates qemu's dirty bitmap using |
389 | * memory_region_set_dirty(). This means all bits are set | |
390 | * to dirty. | |
5832d1f2 | 391 | * |
d3f8d37f | 392 | * @start_add: start of logged region. |
5832d1f2 AL |
393 | * @end_addr: end of logged region. |
394 | */ | |
ffcde12f | 395 | static int kvm_physical_sync_dirty_bitmap(MemoryRegionSection *section) |
5832d1f2 AL |
396 | { |
397 | KVMState *s = kvm_state; | |
151f7749 | 398 | unsigned long size, allocated_size = 0; |
151f7749 JK |
399 | KVMDirtyLog d; |
400 | KVMSlot *mem; | |
401 | int ret = 0; | |
ffcde12f AK |
402 | target_phys_addr_t start_addr = section->offset_within_address_space; |
403 | target_phys_addr_t end_addr = start_addr + section->size; | |
5832d1f2 | 404 | |
151f7749 JK |
405 | d.dirty_bitmap = NULL; |
406 | while (start_addr < end_addr) { | |
407 | mem = kvm_lookup_overlapping_slot(s, start_addr, end_addr); | |
408 | if (mem == NULL) { | |
409 | break; | |
410 | } | |
5832d1f2 | 411 | |
51b0c606 MT |
412 | /* XXX bad kernel interface alert |
413 | * For dirty bitmap, kernel allocates array of size aligned to | |
414 | * bits-per-long. But for case when the kernel is 64bits and | |
415 | * the userspace is 32bits, userspace can't align to the same | |
416 | * bits-per-long, since sizeof(long) is different between kernel | |
417 | * and user space. This way, userspace will provide buffer which | |
418 | * may be 4 bytes less than the kernel will use, resulting in | |
419 | * userspace memory corruption (which is not detectable by valgrind | |
420 | * too, in most cases). | |
421 | * So for now, let's align to 64 instead of HOST_LONG_BITS here, in | |
422 | * a hope that sizeof(long) wont become >8 any time soon. | |
423 | */ | |
424 | size = ALIGN(((mem->memory_size) >> TARGET_PAGE_BITS), | |
425 | /*HOST_LONG_BITS*/ 64) / 8; | |
151f7749 | 426 | if (!d.dirty_bitmap) { |
7267c094 | 427 | d.dirty_bitmap = g_malloc(size); |
151f7749 | 428 | } else if (size > allocated_size) { |
7267c094 | 429 | d.dirty_bitmap = g_realloc(d.dirty_bitmap, size); |
151f7749 JK |
430 | } |
431 | allocated_size = size; | |
432 | memset(d.dirty_bitmap, 0, allocated_size); | |
5832d1f2 | 433 | |
151f7749 | 434 | d.slot = mem->slot; |
5832d1f2 | 435 | |
6e489f3f | 436 | if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) { |
8c0d577e | 437 | DPRINTF("ioctl failed %d\n", errno); |
151f7749 JK |
438 | ret = -1; |
439 | break; | |
440 | } | |
5832d1f2 | 441 | |
ffcde12f | 442 | kvm_get_dirty_pages_log_range(section, d.dirty_bitmap); |
8369e01c | 443 | start_addr = mem->start_addr + mem->memory_size; |
5832d1f2 | 444 | } |
7267c094 | 445 | g_free(d.dirty_bitmap); |
151f7749 JK |
446 | |
447 | return ret; | |
5832d1f2 AL |
448 | } |
449 | ||
c227f099 | 450 | int kvm_coalesce_mmio_region(target_phys_addr_t start, ram_addr_t size) |
f65ed4c1 AL |
451 | { |
452 | int ret = -ENOSYS; | |
f65ed4c1 AL |
453 | KVMState *s = kvm_state; |
454 | ||
455 | if (s->coalesced_mmio) { | |
456 | struct kvm_coalesced_mmio_zone zone; | |
457 | ||
458 | zone.addr = start; | |
459 | zone.size = size; | |
7e680753 | 460 | zone.pad = 0; |
f65ed4c1 AL |
461 | |
462 | ret = kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone); | |
463 | } | |
f65ed4c1 AL |
464 | |
465 | return ret; | |
466 | } | |
467 | ||
c227f099 | 468 | int kvm_uncoalesce_mmio_region(target_phys_addr_t start, ram_addr_t size) |
f65ed4c1 AL |
469 | { |
470 | int ret = -ENOSYS; | |
f65ed4c1 AL |
471 | KVMState *s = kvm_state; |
472 | ||
473 | if (s->coalesced_mmio) { | |
474 | struct kvm_coalesced_mmio_zone zone; | |
475 | ||
476 | zone.addr = start; | |
477 | zone.size = size; | |
7e680753 | 478 | zone.pad = 0; |
f65ed4c1 AL |
479 | |
480 | ret = kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone); | |
481 | } | |
f65ed4c1 AL |
482 | |
483 | return ret; | |
484 | } | |
485 | ||
ad7b8b33 AL |
486 | int kvm_check_extension(KVMState *s, unsigned int extension) |
487 | { | |
488 | int ret; | |
489 | ||
490 | ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension); | |
491 | if (ret < 0) { | |
492 | ret = 0; | |
493 | } | |
494 | ||
495 | return ret; | |
496 | } | |
497 | ||
d2f2b8a7 SH |
498 | static int kvm_check_many_ioeventfds(void) |
499 | { | |
d0dcac83 SH |
500 | /* Userspace can use ioeventfd for io notification. This requires a host |
501 | * that supports eventfd(2) and an I/O thread; since eventfd does not | |
502 | * support SIGIO it cannot interrupt the vcpu. | |
503 | * | |
504 | * Older kernels have a 6 device limit on the KVM io bus. Find out so we | |
d2f2b8a7 SH |
505 | * can avoid creating too many ioeventfds. |
506 | */ | |
12d4536f | 507 | #if defined(CONFIG_EVENTFD) |
d2f2b8a7 SH |
508 | int ioeventfds[7]; |
509 | int i, ret = 0; | |
510 | for (i = 0; i < ARRAY_SIZE(ioeventfds); i++) { | |
511 | ioeventfds[i] = eventfd(0, EFD_CLOEXEC); | |
512 | if (ioeventfds[i] < 0) { | |
513 | break; | |
514 | } | |
515 | ret = kvm_set_ioeventfd_pio_word(ioeventfds[i], 0, i, true); | |
516 | if (ret < 0) { | |
517 | close(ioeventfds[i]); | |
518 | break; | |
519 | } | |
520 | } | |
521 | ||
522 | /* Decide whether many devices are supported or not */ | |
523 | ret = i == ARRAY_SIZE(ioeventfds); | |
524 | ||
525 | while (i-- > 0) { | |
526 | kvm_set_ioeventfd_pio_word(ioeventfds[i], 0, i, false); | |
527 | close(ioeventfds[i]); | |
528 | } | |
529 | return ret; | |
530 | #else | |
531 | return 0; | |
532 | #endif | |
533 | } | |
534 | ||
94a8d39a JK |
535 | static const KVMCapabilityInfo * |
536 | kvm_check_extension_list(KVMState *s, const KVMCapabilityInfo *list) | |
537 | { | |
538 | while (list->name) { | |
539 | if (!kvm_check_extension(s, list->value)) { | |
540 | return list; | |
541 | } | |
542 | list++; | |
543 | } | |
544 | return NULL; | |
545 | } | |
546 | ||
a01672d3 | 547 | static void kvm_set_phys_mem(MemoryRegionSection *section, bool add) |
46dbef6a MT |
548 | { |
549 | KVMState *s = kvm_state; | |
46dbef6a MT |
550 | KVMSlot *mem, old; |
551 | int err; | |
a01672d3 AK |
552 | MemoryRegion *mr = section->mr; |
553 | bool log_dirty = memory_region_is_logging(mr); | |
554 | target_phys_addr_t start_addr = section->offset_within_address_space; | |
555 | ram_addr_t size = section->size; | |
9f213ed9 | 556 | void *ram = NULL; |
8f6f962b | 557 | unsigned delta; |
46dbef6a | 558 | |
14542fea GN |
559 | /* kvm works in page size chunks, but the function may be called |
560 | with sub-page size and unaligned start address. */ | |
8f6f962b AK |
561 | delta = TARGET_PAGE_ALIGN(size) - size; |
562 | if (delta > size) { | |
563 | return; | |
564 | } | |
565 | start_addr += delta; | |
566 | size -= delta; | |
567 | size &= TARGET_PAGE_MASK; | |
568 | if (!size || (start_addr & ~TARGET_PAGE_MASK)) { | |
569 | return; | |
570 | } | |
46dbef6a | 571 | |
a01672d3 AK |
572 | if (!memory_region_is_ram(mr)) { |
573 | return; | |
9f213ed9 AK |
574 | } |
575 | ||
8f6f962b | 576 | ram = memory_region_get_ram_ptr(mr) + section->offset_within_region + delta; |
a01672d3 | 577 | |
46dbef6a MT |
578 | while (1) { |
579 | mem = kvm_lookup_overlapping_slot(s, start_addr, start_addr + size); | |
580 | if (!mem) { | |
581 | break; | |
582 | } | |
583 | ||
a01672d3 | 584 | if (add && start_addr >= mem->start_addr && |
46dbef6a | 585 | (start_addr + size <= mem->start_addr + mem->memory_size) && |
9f213ed9 | 586 | (ram - start_addr == mem->ram - mem->start_addr)) { |
46dbef6a | 587 | /* The new slot fits into the existing one and comes with |
25254bbc MT |
588 | * identical parameters - update flags and done. */ |
589 | kvm_slot_dirty_pages_log_change(mem, log_dirty); | |
46dbef6a MT |
590 | return; |
591 | } | |
592 | ||
593 | old = *mem; | |
594 | ||
3fbffb62 AK |
595 | if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) { |
596 | kvm_physical_sync_dirty_bitmap(section); | |
597 | } | |
598 | ||
46dbef6a MT |
599 | /* unregister the overlapping slot */ |
600 | mem->memory_size = 0; | |
601 | err = kvm_set_user_memory_region(s, mem); | |
602 | if (err) { | |
603 | fprintf(stderr, "%s: error unregistering overlapping slot: %s\n", | |
604 | __func__, strerror(-err)); | |
605 | abort(); | |
606 | } | |
607 | ||
608 | /* Workaround for older KVM versions: we can't join slots, even not by | |
609 | * unregistering the previous ones and then registering the larger | |
610 | * slot. We have to maintain the existing fragmentation. Sigh. | |
611 | * | |
612 | * This workaround assumes that the new slot starts at the same | |
613 | * address as the first existing one. If not or if some overlapping | |
614 | * slot comes around later, we will fail (not seen in practice so far) | |
615 | * - and actually require a recent KVM version. */ | |
616 | if (s->broken_set_mem_region && | |
a01672d3 | 617 | old.start_addr == start_addr && old.memory_size < size && add) { |
46dbef6a MT |
618 | mem = kvm_alloc_slot(s); |
619 | mem->memory_size = old.memory_size; | |
620 | mem->start_addr = old.start_addr; | |
9f213ed9 | 621 | mem->ram = old.ram; |
25254bbc | 622 | mem->flags = kvm_mem_flags(s, log_dirty); |
46dbef6a MT |
623 | |
624 | err = kvm_set_user_memory_region(s, mem); | |
625 | if (err) { | |
626 | fprintf(stderr, "%s: error updating slot: %s\n", __func__, | |
627 | strerror(-err)); | |
628 | abort(); | |
629 | } | |
630 | ||
631 | start_addr += old.memory_size; | |
9f213ed9 | 632 | ram += old.memory_size; |
46dbef6a MT |
633 | size -= old.memory_size; |
634 | continue; | |
635 | } | |
636 | ||
637 | /* register prefix slot */ | |
638 | if (old.start_addr < start_addr) { | |
639 | mem = kvm_alloc_slot(s); | |
640 | mem->memory_size = start_addr - old.start_addr; | |
641 | mem->start_addr = old.start_addr; | |
9f213ed9 | 642 | mem->ram = old.ram; |
25254bbc | 643 | mem->flags = kvm_mem_flags(s, log_dirty); |
46dbef6a MT |
644 | |
645 | err = kvm_set_user_memory_region(s, mem); | |
646 | if (err) { | |
647 | fprintf(stderr, "%s: error registering prefix slot: %s\n", | |
648 | __func__, strerror(-err)); | |
d4d6868f AG |
649 | #ifdef TARGET_PPC |
650 | fprintf(stderr, "%s: This is probably because your kernel's " \ | |
651 | "PAGE_SIZE is too big. Please try to use 4k " \ | |
652 | "PAGE_SIZE!\n", __func__); | |
653 | #endif | |
46dbef6a MT |
654 | abort(); |
655 | } | |
656 | } | |
657 | ||
658 | /* register suffix slot */ | |
659 | if (old.start_addr + old.memory_size > start_addr + size) { | |
660 | ram_addr_t size_delta; | |
661 | ||
662 | mem = kvm_alloc_slot(s); | |
663 | mem->start_addr = start_addr + size; | |
664 | size_delta = mem->start_addr - old.start_addr; | |
665 | mem->memory_size = old.memory_size - size_delta; | |
9f213ed9 | 666 | mem->ram = old.ram + size_delta; |
25254bbc | 667 | mem->flags = kvm_mem_flags(s, log_dirty); |
46dbef6a MT |
668 | |
669 | err = kvm_set_user_memory_region(s, mem); | |
670 | if (err) { | |
671 | fprintf(stderr, "%s: error registering suffix slot: %s\n", | |
672 | __func__, strerror(-err)); | |
673 | abort(); | |
674 | } | |
675 | } | |
676 | } | |
677 | ||
678 | /* in case the KVM bug workaround already "consumed" the new slot */ | |
a426e122 | 679 | if (!size) { |
46dbef6a | 680 | return; |
a426e122 | 681 | } |
a01672d3 | 682 | if (!add) { |
46dbef6a | 683 | return; |
a426e122 | 684 | } |
46dbef6a MT |
685 | mem = kvm_alloc_slot(s); |
686 | mem->memory_size = size; | |
687 | mem->start_addr = start_addr; | |
9f213ed9 | 688 | mem->ram = ram; |
25254bbc | 689 | mem->flags = kvm_mem_flags(s, log_dirty); |
46dbef6a MT |
690 | |
691 | err = kvm_set_user_memory_region(s, mem); | |
692 | if (err) { | |
693 | fprintf(stderr, "%s: error registering slot: %s\n", __func__, | |
694 | strerror(-err)); | |
695 | abort(); | |
696 | } | |
697 | } | |
698 | ||
50c1e149 AK |
699 | static void kvm_begin(MemoryListener *listener) |
700 | { | |
701 | } | |
702 | ||
703 | static void kvm_commit(MemoryListener *listener) | |
704 | { | |
705 | } | |
706 | ||
a01672d3 AK |
707 | static void kvm_region_add(MemoryListener *listener, |
708 | MemoryRegionSection *section) | |
709 | { | |
710 | kvm_set_phys_mem(section, true); | |
711 | } | |
712 | ||
713 | static void kvm_region_del(MemoryListener *listener, | |
714 | MemoryRegionSection *section) | |
715 | { | |
716 | kvm_set_phys_mem(section, false); | |
717 | } | |
718 | ||
50c1e149 AK |
719 | static void kvm_region_nop(MemoryListener *listener, |
720 | MemoryRegionSection *section) | |
721 | { | |
722 | } | |
723 | ||
a01672d3 AK |
724 | static void kvm_log_sync(MemoryListener *listener, |
725 | MemoryRegionSection *section) | |
7b8f3b78 | 726 | { |
a01672d3 AK |
727 | int r; |
728 | ||
ffcde12f | 729 | r = kvm_physical_sync_dirty_bitmap(section); |
a01672d3 AK |
730 | if (r < 0) { |
731 | abort(); | |
732 | } | |
7b8f3b78 MT |
733 | } |
734 | ||
a01672d3 | 735 | static void kvm_log_global_start(struct MemoryListener *listener) |
7b8f3b78 | 736 | { |
a01672d3 AK |
737 | int r; |
738 | ||
739 | r = kvm_set_migration_log(1); | |
740 | assert(r >= 0); | |
7b8f3b78 MT |
741 | } |
742 | ||
a01672d3 | 743 | static void kvm_log_global_stop(struct MemoryListener *listener) |
7b8f3b78 | 744 | { |
a01672d3 AK |
745 | int r; |
746 | ||
747 | r = kvm_set_migration_log(0); | |
748 | assert(r >= 0); | |
7b8f3b78 MT |
749 | } |
750 | ||
80a1ea37 AK |
751 | static void kvm_mem_ioeventfd_add(MemoryRegionSection *section, |
752 | bool match_data, uint64_t data, int fd) | |
753 | { | |
754 | int r; | |
755 | ||
4b8f1c88 | 756 | assert(match_data && section->size <= 8); |
80a1ea37 | 757 | |
4b8f1c88 MT |
758 | r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space, |
759 | data, true, section->size); | |
80a1ea37 AK |
760 | if (r < 0) { |
761 | abort(); | |
762 | } | |
763 | } | |
764 | ||
765 | static void kvm_mem_ioeventfd_del(MemoryRegionSection *section, | |
766 | bool match_data, uint64_t data, int fd) | |
767 | { | |
768 | int r; | |
769 | ||
4b8f1c88 MT |
770 | r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space, |
771 | data, false, section->size); | |
80a1ea37 AK |
772 | if (r < 0) { |
773 | abort(); | |
774 | } | |
775 | } | |
776 | ||
777 | static void kvm_io_ioeventfd_add(MemoryRegionSection *section, | |
778 | bool match_data, uint64_t data, int fd) | |
779 | { | |
780 | int r; | |
781 | ||
782 | assert(match_data && section->size == 2); | |
783 | ||
784 | r = kvm_set_ioeventfd_pio_word(fd, section->offset_within_address_space, | |
785 | data, true); | |
786 | if (r < 0) { | |
787 | abort(); | |
788 | } | |
789 | } | |
790 | ||
791 | static void kvm_io_ioeventfd_del(MemoryRegionSection *section, | |
792 | bool match_data, uint64_t data, int fd) | |
793 | ||
794 | { | |
795 | int r; | |
796 | ||
797 | r = kvm_set_ioeventfd_pio_word(fd, section->offset_within_address_space, | |
798 | data, false); | |
799 | if (r < 0) { | |
800 | abort(); | |
801 | } | |
802 | } | |
803 | ||
804 | static void kvm_eventfd_add(MemoryListener *listener, | |
805 | MemoryRegionSection *section, | |
806 | bool match_data, uint64_t data, int fd) | |
807 | { | |
808 | if (section->address_space == get_system_memory()) { | |
809 | kvm_mem_ioeventfd_add(section, match_data, data, fd); | |
810 | } else { | |
811 | kvm_io_ioeventfd_add(section, match_data, data, fd); | |
812 | } | |
813 | } | |
814 | ||
815 | static void kvm_eventfd_del(MemoryListener *listener, | |
816 | MemoryRegionSection *section, | |
817 | bool match_data, uint64_t data, int fd) | |
818 | { | |
819 | if (section->address_space == get_system_memory()) { | |
820 | kvm_mem_ioeventfd_del(section, match_data, data, fd); | |
821 | } else { | |
822 | kvm_io_ioeventfd_del(section, match_data, data, fd); | |
823 | } | |
824 | } | |
825 | ||
a01672d3 | 826 | static MemoryListener kvm_memory_listener = { |
50c1e149 AK |
827 | .begin = kvm_begin, |
828 | .commit = kvm_commit, | |
a01672d3 AK |
829 | .region_add = kvm_region_add, |
830 | .region_del = kvm_region_del, | |
50c1e149 | 831 | .region_nop = kvm_region_nop, |
e5896b12 AP |
832 | .log_start = kvm_log_start, |
833 | .log_stop = kvm_log_stop, | |
a01672d3 AK |
834 | .log_sync = kvm_log_sync, |
835 | .log_global_start = kvm_log_global_start, | |
836 | .log_global_stop = kvm_log_global_stop, | |
80a1ea37 AK |
837 | .eventfd_add = kvm_eventfd_add, |
838 | .eventfd_del = kvm_eventfd_del, | |
72e22d2f | 839 | .priority = 10, |
7b8f3b78 MT |
840 | }; |
841 | ||
9349b4f9 | 842 | static void kvm_handle_interrupt(CPUArchState *env, int mask) |
aa7f74d1 JK |
843 | { |
844 | env->interrupt_request |= mask; | |
845 | ||
846 | if (!qemu_cpu_is_self(env)) { | |
847 | qemu_cpu_kick(env); | |
848 | } | |
849 | } | |
850 | ||
84b058d7 JK |
851 | int kvm_irqchip_set_irq(KVMState *s, int irq, int level) |
852 | { | |
853 | struct kvm_irq_level event; | |
854 | int ret; | |
855 | ||
3d4b2649 | 856 | assert(kvm_irqchip_in_kernel()); |
84b058d7 JK |
857 | |
858 | event.level = level; | |
859 | event.irq = irq; | |
860 | ret = kvm_vm_ioctl(s, s->irqchip_inject_ioctl, &event); | |
861 | if (ret < 0) { | |
862 | perror("kvm_set_irqchip_line"); | |
863 | abort(); | |
864 | } | |
865 | ||
866 | return (s->irqchip_inject_ioctl == KVM_IRQ_LINE) ? 1 : event.status; | |
867 | } | |
868 | ||
869 | #ifdef KVM_CAP_IRQ_ROUTING | |
870 | static void set_gsi(KVMState *s, unsigned int gsi) | |
871 | { | |
84b058d7 JK |
872 | s->used_gsi_bitmap[gsi / 32] |= 1U << (gsi % 32); |
873 | } | |
874 | ||
04fa27f5 JK |
875 | static void clear_gsi(KVMState *s, unsigned int gsi) |
876 | { | |
877 | s->used_gsi_bitmap[gsi / 32] &= ~(1U << (gsi % 32)); | |
878 | } | |
879 | ||
84b058d7 JK |
880 | static void kvm_init_irq_routing(KVMState *s) |
881 | { | |
04fa27f5 | 882 | int gsi_count, i; |
84b058d7 JK |
883 | |
884 | gsi_count = kvm_check_extension(s, KVM_CAP_IRQ_ROUTING); | |
885 | if (gsi_count > 0) { | |
886 | unsigned int gsi_bits, i; | |
887 | ||
888 | /* Round up so we can search ints using ffs */ | |
bc8c6788 | 889 | gsi_bits = ALIGN(gsi_count, 32); |
84b058d7 | 890 | s->used_gsi_bitmap = g_malloc0(gsi_bits / 8); |
4e2e4e63 | 891 | s->gsi_count = gsi_count; |
84b058d7 JK |
892 | |
893 | /* Mark any over-allocated bits as already in use */ | |
894 | for (i = gsi_count; i < gsi_bits; i++) { | |
895 | set_gsi(s, i); | |
896 | } | |
897 | } | |
898 | ||
899 | s->irq_routes = g_malloc0(sizeof(*s->irq_routes)); | |
900 | s->nr_allocated_irq_routes = 0; | |
901 | ||
4a3adebb JK |
902 | if (!s->direct_msi) { |
903 | for (i = 0; i < KVM_MSI_HASHTAB_SIZE; i++) { | |
904 | QTAILQ_INIT(&s->msi_hashtab[i]); | |
905 | } | |
04fa27f5 JK |
906 | } |
907 | ||
84b058d7 JK |
908 | kvm_arch_init_irq_routing(s); |
909 | } | |
910 | ||
911 | static void kvm_add_routing_entry(KVMState *s, | |
912 | struct kvm_irq_routing_entry *entry) | |
913 | { | |
914 | struct kvm_irq_routing_entry *new; | |
915 | int n, size; | |
916 | ||
917 | if (s->irq_routes->nr == s->nr_allocated_irq_routes) { | |
918 | n = s->nr_allocated_irq_routes * 2; | |
919 | if (n < 64) { | |
920 | n = 64; | |
921 | } | |
922 | size = sizeof(struct kvm_irq_routing); | |
923 | size += n * sizeof(*new); | |
924 | s->irq_routes = g_realloc(s->irq_routes, size); | |
925 | s->nr_allocated_irq_routes = n; | |
926 | } | |
927 | n = s->irq_routes->nr++; | |
928 | new = &s->irq_routes->entries[n]; | |
929 | memset(new, 0, sizeof(*new)); | |
930 | new->gsi = entry->gsi; | |
931 | new->type = entry->type; | |
932 | new->flags = entry->flags; | |
933 | new->u = entry->u; | |
934 | ||
935 | set_gsi(s, entry->gsi); | |
936 | } | |
937 | ||
1df186df | 938 | void kvm_irqchip_add_irq_route(KVMState *s, int irq, int irqchip, int pin) |
84b058d7 JK |
939 | { |
940 | struct kvm_irq_routing_entry e; | |
941 | ||
4e2e4e63 JK |
942 | assert(pin < s->gsi_count); |
943 | ||
84b058d7 JK |
944 | e.gsi = irq; |
945 | e.type = KVM_IRQ_ROUTING_IRQCHIP; | |
946 | e.flags = 0; | |
947 | e.u.irqchip.irqchip = irqchip; | |
948 | e.u.irqchip.pin = pin; | |
949 | kvm_add_routing_entry(s, &e); | |
950 | } | |
951 | ||
952 | int kvm_irqchip_commit_routes(KVMState *s) | |
953 | { | |
954 | s->irq_routes->flags = 0; | |
955 | return kvm_vm_ioctl(s, KVM_SET_GSI_ROUTING, s->irq_routes); | |
956 | } | |
957 | ||
04fa27f5 JK |
958 | static void kvm_irqchip_release_virq(KVMState *s, int virq) |
959 | { | |
960 | struct kvm_irq_routing_entry *e; | |
961 | int i; | |
962 | ||
963 | for (i = 0; i < s->irq_routes->nr; i++) { | |
964 | e = &s->irq_routes->entries[i]; | |
965 | if (e->gsi == virq) { | |
966 | s->irq_routes->nr--; | |
967 | *e = s->irq_routes->entries[s->irq_routes->nr]; | |
968 | } | |
969 | } | |
970 | clear_gsi(s, virq); | |
971 | } | |
972 | ||
973 | static unsigned int kvm_hash_msi(uint32_t data) | |
974 | { | |
975 | /* This is optimized for IA32 MSI layout. However, no other arch shall | |
976 | * repeat the mistake of not providing a direct MSI injection API. */ | |
977 | return data & 0xff; | |
978 | } | |
979 | ||
980 | static void kvm_flush_dynamic_msi_routes(KVMState *s) | |
981 | { | |
982 | KVMMSIRoute *route, *next; | |
983 | unsigned int hash; | |
984 | ||
985 | for (hash = 0; hash < KVM_MSI_HASHTAB_SIZE; hash++) { | |
986 | QTAILQ_FOREACH_SAFE(route, &s->msi_hashtab[hash], entry, next) { | |
987 | kvm_irqchip_release_virq(s, route->kroute.gsi); | |
988 | QTAILQ_REMOVE(&s->msi_hashtab[hash], route, entry); | |
989 | g_free(route); | |
990 | } | |
991 | } | |
992 | } | |
993 | ||
994 | static int kvm_irqchip_get_virq(KVMState *s) | |
995 | { | |
996 | uint32_t *word = s->used_gsi_bitmap; | |
997 | int max_words = ALIGN(s->gsi_count, 32) / 32; | |
998 | int i, bit; | |
999 | bool retry = true; | |
1000 | ||
1001 | again: | |
1002 | /* Return the lowest unused GSI in the bitmap */ | |
1003 | for (i = 0; i < max_words; i++) { | |
1004 | bit = ffs(~word[i]); | |
1005 | if (!bit) { | |
1006 | continue; | |
1007 | } | |
1008 | ||
1009 | return bit - 1 + i * 32; | |
1010 | } | |
4a3adebb | 1011 | if (!s->direct_msi && retry) { |
04fa27f5 JK |
1012 | retry = false; |
1013 | kvm_flush_dynamic_msi_routes(s); | |
1014 | goto again; | |
1015 | } | |
1016 | return -ENOSPC; | |
1017 | ||
1018 | } | |
1019 | ||
1020 | static KVMMSIRoute *kvm_lookup_msi_route(KVMState *s, MSIMessage msg) | |
1021 | { | |
1022 | unsigned int hash = kvm_hash_msi(msg.data); | |
1023 | KVMMSIRoute *route; | |
1024 | ||
1025 | QTAILQ_FOREACH(route, &s->msi_hashtab[hash], entry) { | |
1026 | if (route->kroute.u.msi.address_lo == (uint32_t)msg.address && | |
1027 | route->kroute.u.msi.address_hi == (msg.address >> 32) && | |
1028 | route->kroute.u.msi.data == msg.data) { | |
1029 | return route; | |
1030 | } | |
1031 | } | |
1032 | return NULL; | |
1033 | } | |
1034 | ||
1035 | int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg) | |
1036 | { | |
4a3adebb | 1037 | struct kvm_msi msi; |
04fa27f5 JK |
1038 | KVMMSIRoute *route; |
1039 | ||
4a3adebb JK |
1040 | if (s->direct_msi) { |
1041 | msi.address_lo = (uint32_t)msg.address; | |
1042 | msi.address_hi = msg.address >> 32; | |
1043 | msi.data = msg.data; | |
1044 | msi.flags = 0; | |
1045 | memset(msi.pad, 0, sizeof(msi.pad)); | |
1046 | ||
1047 | return kvm_vm_ioctl(s, KVM_SIGNAL_MSI, &msi); | |
1048 | } | |
1049 | ||
04fa27f5 JK |
1050 | route = kvm_lookup_msi_route(s, msg); |
1051 | if (!route) { | |
1052 | int virq, ret; | |
1053 | ||
1054 | virq = kvm_irqchip_get_virq(s); | |
1055 | if (virq < 0) { | |
1056 | return virq; | |
1057 | } | |
1058 | ||
1059 | route = g_malloc(sizeof(KVMMSIRoute)); | |
1060 | route->kroute.gsi = virq; | |
1061 | route->kroute.type = KVM_IRQ_ROUTING_MSI; | |
1062 | route->kroute.flags = 0; | |
1063 | route->kroute.u.msi.address_lo = (uint32_t)msg.address; | |
1064 | route->kroute.u.msi.address_hi = msg.address >> 32; | |
1065 | route->kroute.u.msi.data = msg.data; | |
1066 | ||
1067 | kvm_add_routing_entry(s, &route->kroute); | |
1068 | ||
1069 | QTAILQ_INSERT_TAIL(&s->msi_hashtab[kvm_hash_msi(msg.data)], route, | |
1070 | entry); | |
1071 | ||
1072 | ret = kvm_irqchip_commit_routes(s); | |
1073 | if (ret < 0) { | |
1074 | return ret; | |
1075 | } | |
1076 | } | |
1077 | ||
1078 | assert(route->kroute.type == KVM_IRQ_ROUTING_MSI); | |
1079 | ||
1080 | return kvm_irqchip_set_irq(s, route->kroute.gsi, 1); | |
1081 | } | |
1082 | ||
92b4e489 JK |
1083 | int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg) |
1084 | { | |
1085 | struct kvm_irq_routing_entry kroute; | |
1086 | int virq; | |
1087 | ||
1088 | if (!kvm_irqchip_in_kernel()) { | |
1089 | return -ENOSYS; | |
1090 | } | |
1091 | ||
1092 | virq = kvm_irqchip_get_virq(s); | |
1093 | if (virq < 0) { | |
1094 | return virq; | |
1095 | } | |
1096 | ||
1097 | kroute.gsi = virq; | |
1098 | kroute.type = KVM_IRQ_ROUTING_MSI; | |
1099 | kroute.flags = 0; | |
1100 | kroute.u.msi.address_lo = (uint32_t)msg.address; | |
1101 | kroute.u.msi.address_hi = msg.address >> 32; | |
1102 | kroute.u.msi.data = msg.data; | |
1103 | ||
1104 | kvm_add_routing_entry(s, &kroute); | |
1105 | ||
1106 | return virq; | |
1107 | } | |
1108 | ||
84b058d7 JK |
1109 | #else /* !KVM_CAP_IRQ_ROUTING */ |
1110 | ||
1111 | static void kvm_init_irq_routing(KVMState *s) | |
1112 | { | |
1113 | } | |
04fa27f5 JK |
1114 | |
1115 | int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg) | |
1116 | { | |
1117 | abort(); | |
1118 | } | |
92b4e489 JK |
1119 | |
1120 | int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg) | |
1121 | { | |
1122 | abort(); | |
1123 | } | |
84b058d7 JK |
1124 | #endif /* !KVM_CAP_IRQ_ROUTING */ |
1125 | ||
1126 | static int kvm_irqchip_create(KVMState *s) | |
1127 | { | |
1128 | QemuOptsList *list = qemu_find_opts("machine"); | |
1129 | int ret; | |
1130 | ||
1131 | if (QTAILQ_EMPTY(&list->head) || | |
1132 | !qemu_opt_get_bool(QTAILQ_FIRST(&list->head), | |
a24b9106 | 1133 | "kernel_irqchip", true) || |
84b058d7 JK |
1134 | !kvm_check_extension(s, KVM_CAP_IRQCHIP)) { |
1135 | return 0; | |
1136 | } | |
1137 | ||
1138 | ret = kvm_vm_ioctl(s, KVM_CREATE_IRQCHIP); | |
1139 | if (ret < 0) { | |
1140 | fprintf(stderr, "Create kernel irqchip failed\n"); | |
1141 | return ret; | |
1142 | } | |
1143 | ||
1144 | s->irqchip_inject_ioctl = KVM_IRQ_LINE; | |
1145 | if (kvm_check_extension(s, KVM_CAP_IRQ_INJECT_STATUS)) { | |
1146 | s->irqchip_inject_ioctl = KVM_IRQ_LINE_STATUS; | |
1147 | } | |
3d4b2649 | 1148 | kvm_kernel_irqchip = true; |
84b058d7 JK |
1149 | |
1150 | kvm_init_irq_routing(s); | |
1151 | ||
1152 | return 0; | |
1153 | } | |
1154 | ||
cad1e282 | 1155 | int kvm_init(void) |
05330448 | 1156 | { |
168ccc11 JK |
1157 | static const char upgrade_note[] = |
1158 | "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n" | |
1159 | "(see http://sourceforge.net/projects/kvm).\n"; | |
05330448 | 1160 | KVMState *s; |
94a8d39a | 1161 | const KVMCapabilityInfo *missing_cap; |
05330448 AL |
1162 | int ret; |
1163 | int i; | |
1164 | ||
7267c094 | 1165 | s = g_malloc0(sizeof(KVMState)); |
05330448 | 1166 | |
3145fcb6 DG |
1167 | /* |
1168 | * On systems where the kernel can support different base page | |
1169 | * sizes, host page size may be different from TARGET_PAGE_SIZE, | |
1170 | * even with KVM. TARGET_PAGE_SIZE is assumed to be the minimum | |
1171 | * page size for the system though. | |
1172 | */ | |
1173 | assert(TARGET_PAGE_SIZE <= getpagesize()); | |
1174 | ||
e22a25c9 | 1175 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
72cf2d4f | 1176 | QTAILQ_INIT(&s->kvm_sw_breakpoints); |
e22a25c9 | 1177 | #endif |
a426e122 | 1178 | for (i = 0; i < ARRAY_SIZE(s->slots); i++) { |
05330448 | 1179 | s->slots[i].slot = i; |
a426e122 | 1180 | } |
05330448 | 1181 | s->vmfd = -1; |
40ff6d7e | 1182 | s->fd = qemu_open("/dev/kvm", O_RDWR); |
05330448 AL |
1183 | if (s->fd == -1) { |
1184 | fprintf(stderr, "Could not access KVM kernel module: %m\n"); | |
1185 | ret = -errno; | |
1186 | goto err; | |
1187 | } | |
1188 | ||
1189 | ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0); | |
1190 | if (ret < KVM_API_VERSION) { | |
a426e122 | 1191 | if (ret > 0) { |
05330448 | 1192 | ret = -EINVAL; |
a426e122 | 1193 | } |
05330448 AL |
1194 | fprintf(stderr, "kvm version too old\n"); |
1195 | goto err; | |
1196 | } | |
1197 | ||
1198 | if (ret > KVM_API_VERSION) { | |
1199 | ret = -EINVAL; | |
1200 | fprintf(stderr, "kvm version not supported\n"); | |
1201 | goto err; | |
1202 | } | |
1203 | ||
1204 | s->vmfd = kvm_ioctl(s, KVM_CREATE_VM, 0); | |
0104dcac AG |
1205 | if (s->vmfd < 0) { |
1206 | #ifdef TARGET_S390X | |
1207 | fprintf(stderr, "Please add the 'switch_amode' kernel parameter to " | |
1208 | "your host kernel command line\n"); | |
1209 | #endif | |
db9eae1c | 1210 | ret = s->vmfd; |
05330448 | 1211 | goto err; |
0104dcac | 1212 | } |
05330448 | 1213 | |
94a8d39a JK |
1214 | missing_cap = kvm_check_extension_list(s, kvm_required_capabilites); |
1215 | if (!missing_cap) { | |
1216 | missing_cap = | |
1217 | kvm_check_extension_list(s, kvm_arch_required_capabilities); | |
05330448 | 1218 | } |
94a8d39a | 1219 | if (missing_cap) { |
ad7b8b33 | 1220 | ret = -EINVAL; |
94a8d39a JK |
1221 | fprintf(stderr, "kvm does not support %s\n%s", |
1222 | missing_cap->name, upgrade_note); | |
d85dc283 AL |
1223 | goto err; |
1224 | } | |
1225 | ||
ad7b8b33 | 1226 | s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO); |
f65ed4c1 | 1227 | |
e69917e2 | 1228 | s->broken_set_mem_region = 1; |
14a09518 | 1229 | ret = kvm_check_extension(s, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS); |
e69917e2 JK |
1230 | if (ret > 0) { |
1231 | s->broken_set_mem_region = 0; | |
1232 | } | |
e69917e2 | 1233 | |
a0fb002c JK |
1234 | #ifdef KVM_CAP_VCPU_EVENTS |
1235 | s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS); | |
1236 | #endif | |
1237 | ||
b0b1d690 JK |
1238 | s->robust_singlestep = |
1239 | kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP); | |
b0b1d690 | 1240 | |
ff44f1a3 JK |
1241 | #ifdef KVM_CAP_DEBUGREGS |
1242 | s->debugregs = kvm_check_extension(s, KVM_CAP_DEBUGREGS); | |
1243 | #endif | |
1244 | ||
f1665b21 SY |
1245 | #ifdef KVM_CAP_XSAVE |
1246 | s->xsave = kvm_check_extension(s, KVM_CAP_XSAVE); | |
1247 | #endif | |
1248 | ||
f1665b21 SY |
1249 | #ifdef KVM_CAP_XCRS |
1250 | s->xcrs = kvm_check_extension(s, KVM_CAP_XCRS); | |
1251 | #endif | |
1252 | ||
8a7c7393 JK |
1253 | #ifdef KVM_CAP_PIT_STATE2 |
1254 | s->pit_state2 = kvm_check_extension(s, KVM_CAP_PIT_STATE2); | |
1255 | #endif | |
1256 | ||
4a3adebb JK |
1257 | s->direct_msi = (kvm_check_extension(s, KVM_CAP_SIGNAL_MSI) > 0); |
1258 | ||
cad1e282 | 1259 | ret = kvm_arch_init(s); |
a426e122 | 1260 | if (ret < 0) { |
05330448 | 1261 | goto err; |
a426e122 | 1262 | } |
05330448 | 1263 | |
84b058d7 JK |
1264 | ret = kvm_irqchip_create(s); |
1265 | if (ret < 0) { | |
1266 | goto err; | |
1267 | } | |
1268 | ||
05330448 | 1269 | kvm_state = s; |
7376e582 | 1270 | memory_listener_register(&kvm_memory_listener, NULL); |
05330448 | 1271 | |
d2f2b8a7 SH |
1272 | s->many_ioeventfds = kvm_check_many_ioeventfds(); |
1273 | ||
aa7f74d1 JK |
1274 | cpu_interrupt_handler = kvm_handle_interrupt; |
1275 | ||
05330448 AL |
1276 | return 0; |
1277 | ||
1278 | err: | |
1279 | if (s) { | |
db9eae1c | 1280 | if (s->vmfd >= 0) { |
05330448 | 1281 | close(s->vmfd); |
a426e122 JK |
1282 | } |
1283 | if (s->fd != -1) { | |
05330448 | 1284 | close(s->fd); |
a426e122 | 1285 | } |
05330448 | 1286 | } |
7267c094 | 1287 | g_free(s); |
05330448 AL |
1288 | |
1289 | return ret; | |
1290 | } | |
1291 | ||
b30e93e9 JK |
1292 | static void kvm_handle_io(uint16_t port, void *data, int direction, int size, |
1293 | uint32_t count) | |
05330448 AL |
1294 | { |
1295 | int i; | |
1296 | uint8_t *ptr = data; | |
1297 | ||
1298 | for (i = 0; i < count; i++) { | |
1299 | if (direction == KVM_EXIT_IO_IN) { | |
1300 | switch (size) { | |
1301 | case 1: | |
afcea8cb | 1302 | stb_p(ptr, cpu_inb(port)); |
05330448 AL |
1303 | break; |
1304 | case 2: | |
afcea8cb | 1305 | stw_p(ptr, cpu_inw(port)); |
05330448 AL |
1306 | break; |
1307 | case 4: | |
afcea8cb | 1308 | stl_p(ptr, cpu_inl(port)); |
05330448 AL |
1309 | break; |
1310 | } | |
1311 | } else { | |
1312 | switch (size) { | |
1313 | case 1: | |
afcea8cb | 1314 | cpu_outb(port, ldub_p(ptr)); |
05330448 AL |
1315 | break; |
1316 | case 2: | |
afcea8cb | 1317 | cpu_outw(port, lduw_p(ptr)); |
05330448 AL |
1318 | break; |
1319 | case 4: | |
afcea8cb | 1320 | cpu_outl(port, ldl_p(ptr)); |
05330448 AL |
1321 | break; |
1322 | } | |
1323 | } | |
1324 | ||
1325 | ptr += size; | |
1326 | } | |
05330448 AL |
1327 | } |
1328 | ||
9349b4f9 | 1329 | static int kvm_handle_internal_error(CPUArchState *env, struct kvm_run *run) |
7c80eef8 | 1330 | { |
bb44e0d1 | 1331 | fprintf(stderr, "KVM internal error."); |
7c80eef8 MT |
1332 | if (kvm_check_extension(kvm_state, KVM_CAP_INTERNAL_ERROR_DATA)) { |
1333 | int i; | |
1334 | ||
bb44e0d1 | 1335 | fprintf(stderr, " Suberror: %d\n", run->internal.suberror); |
7c80eef8 MT |
1336 | for (i = 0; i < run->internal.ndata; ++i) { |
1337 | fprintf(stderr, "extra data[%d]: %"PRIx64"\n", | |
1338 | i, (uint64_t)run->internal.data[i]); | |
1339 | } | |
bb44e0d1 JK |
1340 | } else { |
1341 | fprintf(stderr, "\n"); | |
7c80eef8 | 1342 | } |
7c80eef8 MT |
1343 | if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) { |
1344 | fprintf(stderr, "emulation failure\n"); | |
a426e122 | 1345 | if (!kvm_arch_stop_on_emulation_error(env)) { |
f5c848ee | 1346 | cpu_dump_state(env, stderr, fprintf, CPU_DUMP_CODE); |
d73cd8f4 | 1347 | return EXCP_INTERRUPT; |
a426e122 | 1348 | } |
7c80eef8 MT |
1349 | } |
1350 | /* FIXME: Should trigger a qmp message to let management know | |
1351 | * something went wrong. | |
1352 | */ | |
73aaec4a | 1353 | return -1; |
7c80eef8 | 1354 | } |
7c80eef8 | 1355 | |
62a2744c | 1356 | void kvm_flush_coalesced_mmio_buffer(void) |
f65ed4c1 | 1357 | { |
f65ed4c1 | 1358 | KVMState *s = kvm_state; |
1cae88b9 AK |
1359 | |
1360 | if (s->coalesced_flush_in_progress) { | |
1361 | return; | |
1362 | } | |
1363 | ||
1364 | s->coalesced_flush_in_progress = true; | |
1365 | ||
62a2744c SY |
1366 | if (s->coalesced_mmio_ring) { |
1367 | struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring; | |
f65ed4c1 AL |
1368 | while (ring->first != ring->last) { |
1369 | struct kvm_coalesced_mmio *ent; | |
1370 | ||
1371 | ent = &ring->coalesced_mmio[ring->first]; | |
1372 | ||
1373 | cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len); | |
85199474 | 1374 | smp_wmb(); |
f65ed4c1 AL |
1375 | ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX; |
1376 | } | |
1377 | } | |
1cae88b9 AK |
1378 | |
1379 | s->coalesced_flush_in_progress = false; | |
f65ed4c1 AL |
1380 | } |
1381 | ||
2705d56a | 1382 | static void do_kvm_cpu_synchronize_state(void *_env) |
4c0960c0 | 1383 | { |
9349b4f9 | 1384 | CPUArchState *env = _env; |
2705d56a | 1385 | |
9ded2744 | 1386 | if (!env->kvm_vcpu_dirty) { |
4c0960c0 | 1387 | kvm_arch_get_registers(env); |
9ded2744 | 1388 | env->kvm_vcpu_dirty = 1; |
4c0960c0 AK |
1389 | } |
1390 | } | |
1391 | ||
9349b4f9 | 1392 | void kvm_cpu_synchronize_state(CPUArchState *env) |
2705d56a | 1393 | { |
a426e122 | 1394 | if (!env->kvm_vcpu_dirty) { |
2705d56a | 1395 | run_on_cpu(env, do_kvm_cpu_synchronize_state, env); |
a426e122 | 1396 | } |
2705d56a JK |
1397 | } |
1398 | ||
9349b4f9 | 1399 | void kvm_cpu_synchronize_post_reset(CPUArchState *env) |
ea375f9a JK |
1400 | { |
1401 | kvm_arch_put_registers(env, KVM_PUT_RESET_STATE); | |
1402 | env->kvm_vcpu_dirty = 0; | |
1403 | } | |
1404 | ||
9349b4f9 | 1405 | void kvm_cpu_synchronize_post_init(CPUArchState *env) |
ea375f9a JK |
1406 | { |
1407 | kvm_arch_put_registers(env, KVM_PUT_FULL_STATE); | |
1408 | env->kvm_vcpu_dirty = 0; | |
1409 | } | |
1410 | ||
9349b4f9 | 1411 | int kvm_cpu_exec(CPUArchState *env) |
05330448 AL |
1412 | { |
1413 | struct kvm_run *run = env->kvm_run; | |
7cbb533f | 1414 | int ret, run_ret; |
05330448 | 1415 | |
8c0d577e | 1416 | DPRINTF("kvm_cpu_exec()\n"); |
05330448 | 1417 | |
99036865 | 1418 | if (kvm_arch_process_async_events(env)) { |
9ccfac9e | 1419 | env->exit_request = 0; |
6792a57b | 1420 | return EXCP_HLT; |
9ccfac9e | 1421 | } |
0af691d7 | 1422 | |
9ccfac9e | 1423 | do { |
9ded2744 | 1424 | if (env->kvm_vcpu_dirty) { |
ea375f9a | 1425 | kvm_arch_put_registers(env, KVM_PUT_RUNTIME_STATE); |
9ded2744 | 1426 | env->kvm_vcpu_dirty = 0; |
4c0960c0 AK |
1427 | } |
1428 | ||
8c14c173 | 1429 | kvm_arch_pre_run(env, run); |
9ccfac9e JK |
1430 | if (env->exit_request) { |
1431 | DPRINTF("interrupt exit requested\n"); | |
1432 | /* | |
1433 | * KVM requires us to reenter the kernel after IO exits to complete | |
1434 | * instruction emulation. This self-signal will ensure that we | |
1435 | * leave ASAP again. | |
1436 | */ | |
1437 | qemu_cpu_kick_self(); | |
1438 | } | |
d549db5a | 1439 | qemu_mutex_unlock_iothread(); |
9ccfac9e | 1440 | |
7cbb533f | 1441 | run_ret = kvm_vcpu_ioctl(env, KVM_RUN, 0); |
9ccfac9e | 1442 | |
d549db5a | 1443 | qemu_mutex_lock_iothread(); |
05330448 AL |
1444 | kvm_arch_post_run(env, run); |
1445 | ||
b0c883b5 JK |
1446 | kvm_flush_coalesced_mmio_buffer(); |
1447 | ||
7cbb533f | 1448 | if (run_ret < 0) { |
dc77d341 JK |
1449 | if (run_ret == -EINTR || run_ret == -EAGAIN) { |
1450 | DPRINTF("io window exit\n"); | |
d73cd8f4 | 1451 | ret = EXCP_INTERRUPT; |
dc77d341 JK |
1452 | break; |
1453 | } | |
7b011fbc ME |
1454 | fprintf(stderr, "error: kvm run failed %s\n", |
1455 | strerror(-run_ret)); | |
05330448 AL |
1456 | abort(); |
1457 | } | |
1458 | ||
05330448 AL |
1459 | switch (run->exit_reason) { |
1460 | case KVM_EXIT_IO: | |
8c0d577e | 1461 | DPRINTF("handle_io\n"); |
b30e93e9 JK |
1462 | kvm_handle_io(run->io.port, |
1463 | (uint8_t *)run + run->io.data_offset, | |
1464 | run->io.direction, | |
1465 | run->io.size, | |
1466 | run->io.count); | |
d73cd8f4 | 1467 | ret = 0; |
05330448 AL |
1468 | break; |
1469 | case KVM_EXIT_MMIO: | |
8c0d577e | 1470 | DPRINTF("handle_mmio\n"); |
05330448 AL |
1471 | cpu_physical_memory_rw(run->mmio.phys_addr, |
1472 | run->mmio.data, | |
1473 | run->mmio.len, | |
1474 | run->mmio.is_write); | |
d73cd8f4 | 1475 | ret = 0; |
05330448 AL |
1476 | break; |
1477 | case KVM_EXIT_IRQ_WINDOW_OPEN: | |
8c0d577e | 1478 | DPRINTF("irq_window_open\n"); |
d73cd8f4 | 1479 | ret = EXCP_INTERRUPT; |
05330448 AL |
1480 | break; |
1481 | case KVM_EXIT_SHUTDOWN: | |
8c0d577e | 1482 | DPRINTF("shutdown\n"); |
05330448 | 1483 | qemu_system_reset_request(); |
d73cd8f4 | 1484 | ret = EXCP_INTERRUPT; |
05330448 AL |
1485 | break; |
1486 | case KVM_EXIT_UNKNOWN: | |
bb44e0d1 JK |
1487 | fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "\n", |
1488 | (uint64_t)run->hw.hardware_exit_reason); | |
73aaec4a | 1489 | ret = -1; |
05330448 | 1490 | break; |
7c80eef8 | 1491 | case KVM_EXIT_INTERNAL_ERROR: |
73aaec4a | 1492 | ret = kvm_handle_internal_error(env, run); |
7c80eef8 | 1493 | break; |
05330448 | 1494 | default: |
8c0d577e | 1495 | DPRINTF("kvm_arch_handle_exit\n"); |
05330448 AL |
1496 | ret = kvm_arch_handle_exit(env, run); |
1497 | break; | |
1498 | } | |
d73cd8f4 | 1499 | } while (ret == 0); |
05330448 | 1500 | |
73aaec4a | 1501 | if (ret < 0) { |
f5c848ee | 1502 | cpu_dump_state(env, stderr, fprintf, CPU_DUMP_CODE); |
0461d5a6 | 1503 | vm_stop(RUN_STATE_INTERNAL_ERROR); |
becfc390 AL |
1504 | } |
1505 | ||
6792a57b | 1506 | env->exit_request = 0; |
05330448 AL |
1507 | return ret; |
1508 | } | |
1509 | ||
984b5181 | 1510 | int kvm_ioctl(KVMState *s, int type, ...) |
05330448 AL |
1511 | { |
1512 | int ret; | |
984b5181 AL |
1513 | void *arg; |
1514 | va_list ap; | |
05330448 | 1515 | |
984b5181 AL |
1516 | va_start(ap, type); |
1517 | arg = va_arg(ap, void *); | |
1518 | va_end(ap); | |
1519 | ||
1520 | ret = ioctl(s->fd, type, arg); | |
a426e122 | 1521 | if (ret == -1) { |
05330448 | 1522 | ret = -errno; |
a426e122 | 1523 | } |
05330448 AL |
1524 | return ret; |
1525 | } | |
1526 | ||
984b5181 | 1527 | int kvm_vm_ioctl(KVMState *s, int type, ...) |
05330448 AL |
1528 | { |
1529 | int ret; | |
984b5181 AL |
1530 | void *arg; |
1531 | va_list ap; | |
1532 | ||
1533 | va_start(ap, type); | |
1534 | arg = va_arg(ap, void *); | |
1535 | va_end(ap); | |
05330448 | 1536 | |
984b5181 | 1537 | ret = ioctl(s->vmfd, type, arg); |
a426e122 | 1538 | if (ret == -1) { |
05330448 | 1539 | ret = -errno; |
a426e122 | 1540 | } |
05330448 AL |
1541 | return ret; |
1542 | } | |
1543 | ||
9349b4f9 | 1544 | int kvm_vcpu_ioctl(CPUArchState *env, int type, ...) |
05330448 AL |
1545 | { |
1546 | int ret; | |
984b5181 AL |
1547 | void *arg; |
1548 | va_list ap; | |
1549 | ||
1550 | va_start(ap, type); | |
1551 | arg = va_arg(ap, void *); | |
1552 | va_end(ap); | |
05330448 | 1553 | |
984b5181 | 1554 | ret = ioctl(env->kvm_fd, type, arg); |
a426e122 | 1555 | if (ret == -1) { |
05330448 | 1556 | ret = -errno; |
a426e122 | 1557 | } |
05330448 AL |
1558 | return ret; |
1559 | } | |
bd322087 AL |
1560 | |
1561 | int kvm_has_sync_mmu(void) | |
1562 | { | |
94a8d39a | 1563 | return kvm_check_extension(kvm_state, KVM_CAP_SYNC_MMU); |
bd322087 | 1564 | } |
e22a25c9 | 1565 | |
a0fb002c JK |
1566 | int kvm_has_vcpu_events(void) |
1567 | { | |
1568 | return kvm_state->vcpu_events; | |
1569 | } | |
1570 | ||
b0b1d690 JK |
1571 | int kvm_has_robust_singlestep(void) |
1572 | { | |
1573 | return kvm_state->robust_singlestep; | |
1574 | } | |
1575 | ||
ff44f1a3 JK |
1576 | int kvm_has_debugregs(void) |
1577 | { | |
1578 | return kvm_state->debugregs; | |
1579 | } | |
1580 | ||
f1665b21 SY |
1581 | int kvm_has_xsave(void) |
1582 | { | |
1583 | return kvm_state->xsave; | |
1584 | } | |
1585 | ||
1586 | int kvm_has_xcrs(void) | |
1587 | { | |
1588 | return kvm_state->xcrs; | |
1589 | } | |
1590 | ||
8a7c7393 JK |
1591 | int kvm_has_pit_state2(void) |
1592 | { | |
1593 | return kvm_state->pit_state2; | |
1594 | } | |
1595 | ||
d2f2b8a7 SH |
1596 | int kvm_has_many_ioeventfds(void) |
1597 | { | |
1598 | if (!kvm_enabled()) { | |
1599 | return 0; | |
1600 | } | |
1601 | return kvm_state->many_ioeventfds; | |
1602 | } | |
1603 | ||
84b058d7 JK |
1604 | int kvm_has_gsi_routing(void) |
1605 | { | |
a9c5eb0d | 1606 | #ifdef KVM_CAP_IRQ_ROUTING |
84b058d7 | 1607 | return kvm_check_extension(kvm_state, KVM_CAP_IRQ_ROUTING); |
a9c5eb0d AG |
1608 | #else |
1609 | return false; | |
1610 | #endif | |
84b058d7 JK |
1611 | } |
1612 | ||
9b5b76d4 JK |
1613 | int kvm_allows_irq0_override(void) |
1614 | { | |
3d4b2649 | 1615 | return !kvm_irqchip_in_kernel() || kvm_has_gsi_routing(); |
9b5b76d4 JK |
1616 | } |
1617 | ||
6f0437e8 JK |
1618 | void kvm_setup_guest_memory(void *start, size_t size) |
1619 | { | |
1620 | if (!kvm_has_sync_mmu()) { | |
e78815a5 | 1621 | int ret = qemu_madvise(start, size, QEMU_MADV_DONTFORK); |
6f0437e8 JK |
1622 | |
1623 | if (ret) { | |
e78815a5 AF |
1624 | perror("qemu_madvise"); |
1625 | fprintf(stderr, | |
1626 | "Need MADV_DONTFORK in absence of synchronous KVM MMU\n"); | |
6f0437e8 JK |
1627 | exit(1); |
1628 | } | |
6f0437e8 JK |
1629 | } |
1630 | } | |
1631 | ||
e22a25c9 | 1632 | #ifdef KVM_CAP_SET_GUEST_DEBUG |
9349b4f9 | 1633 | struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUArchState *env, |
e22a25c9 AL |
1634 | target_ulong pc) |
1635 | { | |
1636 | struct kvm_sw_breakpoint *bp; | |
1637 | ||
72cf2d4f | 1638 | QTAILQ_FOREACH(bp, &env->kvm_state->kvm_sw_breakpoints, entry) { |
a426e122 | 1639 | if (bp->pc == pc) { |
e22a25c9 | 1640 | return bp; |
a426e122 | 1641 | } |
e22a25c9 AL |
1642 | } |
1643 | return NULL; | |
1644 | } | |
1645 | ||
9349b4f9 | 1646 | int kvm_sw_breakpoints_active(CPUArchState *env) |
e22a25c9 | 1647 | { |
72cf2d4f | 1648 | return !QTAILQ_EMPTY(&env->kvm_state->kvm_sw_breakpoints); |
e22a25c9 AL |
1649 | } |
1650 | ||
452e4751 GC |
1651 | struct kvm_set_guest_debug_data { |
1652 | struct kvm_guest_debug dbg; | |
9349b4f9 | 1653 | CPUArchState *env; |
452e4751 GC |
1654 | int err; |
1655 | }; | |
1656 | ||
1657 | static void kvm_invoke_set_guest_debug(void *data) | |
1658 | { | |
1659 | struct kvm_set_guest_debug_data *dbg_data = data; | |
9349b4f9 | 1660 | CPUArchState *env = dbg_data->env; |
b3807725 | 1661 | |
b3807725 | 1662 | dbg_data->err = kvm_vcpu_ioctl(env, KVM_SET_GUEST_DEBUG, &dbg_data->dbg); |
452e4751 GC |
1663 | } |
1664 | ||
9349b4f9 | 1665 | int kvm_update_guest_debug(CPUArchState *env, unsigned long reinject_trap) |
e22a25c9 | 1666 | { |
452e4751 | 1667 | struct kvm_set_guest_debug_data data; |
e22a25c9 | 1668 | |
b0b1d690 | 1669 | data.dbg.control = reinject_trap; |
e22a25c9 | 1670 | |
b0b1d690 JK |
1671 | if (env->singlestep_enabled) { |
1672 | data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP; | |
1673 | } | |
452e4751 | 1674 | kvm_arch_update_guest_debug(env, &data.dbg); |
452e4751 | 1675 | data.env = env; |
e22a25c9 | 1676 | |
be41cbe0 | 1677 | run_on_cpu(env, kvm_invoke_set_guest_debug, &data); |
452e4751 | 1678 | return data.err; |
e22a25c9 AL |
1679 | } |
1680 | ||
9349b4f9 | 1681 | int kvm_insert_breakpoint(CPUArchState *current_env, target_ulong addr, |
e22a25c9 AL |
1682 | target_ulong len, int type) |
1683 | { | |
1684 | struct kvm_sw_breakpoint *bp; | |
9349b4f9 | 1685 | CPUArchState *env; |
e22a25c9 AL |
1686 | int err; |
1687 | ||
1688 | if (type == GDB_BREAKPOINT_SW) { | |
1689 | bp = kvm_find_sw_breakpoint(current_env, addr); | |
1690 | if (bp) { | |
1691 | bp->use_count++; | |
1692 | return 0; | |
1693 | } | |
1694 | ||
7267c094 | 1695 | bp = g_malloc(sizeof(struct kvm_sw_breakpoint)); |
a426e122 | 1696 | if (!bp) { |
e22a25c9 | 1697 | return -ENOMEM; |
a426e122 | 1698 | } |
e22a25c9 AL |
1699 | |
1700 | bp->pc = addr; | |
1701 | bp->use_count = 1; | |
1702 | err = kvm_arch_insert_sw_breakpoint(current_env, bp); | |
1703 | if (err) { | |
7267c094 | 1704 | g_free(bp); |
e22a25c9 AL |
1705 | return err; |
1706 | } | |
1707 | ||
72cf2d4f | 1708 | QTAILQ_INSERT_HEAD(¤t_env->kvm_state->kvm_sw_breakpoints, |
e22a25c9 AL |
1709 | bp, entry); |
1710 | } else { | |
1711 | err = kvm_arch_insert_hw_breakpoint(addr, len, type); | |
a426e122 | 1712 | if (err) { |
e22a25c9 | 1713 | return err; |
a426e122 | 1714 | } |
e22a25c9 AL |
1715 | } |
1716 | ||
1717 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
1718 | err = kvm_update_guest_debug(env, 0); | |
a426e122 | 1719 | if (err) { |
e22a25c9 | 1720 | return err; |
a426e122 | 1721 | } |
e22a25c9 AL |
1722 | } |
1723 | return 0; | |
1724 | } | |
1725 | ||
9349b4f9 | 1726 | int kvm_remove_breakpoint(CPUArchState *current_env, target_ulong addr, |
e22a25c9 AL |
1727 | target_ulong len, int type) |
1728 | { | |
1729 | struct kvm_sw_breakpoint *bp; | |
9349b4f9 | 1730 | CPUArchState *env; |
e22a25c9 AL |
1731 | int err; |
1732 | ||
1733 | if (type == GDB_BREAKPOINT_SW) { | |
1734 | bp = kvm_find_sw_breakpoint(current_env, addr); | |
a426e122 | 1735 | if (!bp) { |
e22a25c9 | 1736 | return -ENOENT; |
a426e122 | 1737 | } |
e22a25c9 AL |
1738 | |
1739 | if (bp->use_count > 1) { | |
1740 | bp->use_count--; | |
1741 | return 0; | |
1742 | } | |
1743 | ||
1744 | err = kvm_arch_remove_sw_breakpoint(current_env, bp); | |
a426e122 | 1745 | if (err) { |
e22a25c9 | 1746 | return err; |
a426e122 | 1747 | } |
e22a25c9 | 1748 | |
72cf2d4f | 1749 | QTAILQ_REMOVE(¤t_env->kvm_state->kvm_sw_breakpoints, bp, entry); |
7267c094 | 1750 | g_free(bp); |
e22a25c9 AL |
1751 | } else { |
1752 | err = kvm_arch_remove_hw_breakpoint(addr, len, type); | |
a426e122 | 1753 | if (err) { |
e22a25c9 | 1754 | return err; |
a426e122 | 1755 | } |
e22a25c9 AL |
1756 | } |
1757 | ||
1758 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
1759 | err = kvm_update_guest_debug(env, 0); | |
a426e122 | 1760 | if (err) { |
e22a25c9 | 1761 | return err; |
a426e122 | 1762 | } |
e22a25c9 AL |
1763 | } |
1764 | return 0; | |
1765 | } | |
1766 | ||
9349b4f9 | 1767 | void kvm_remove_all_breakpoints(CPUArchState *current_env) |
e22a25c9 AL |
1768 | { |
1769 | struct kvm_sw_breakpoint *bp, *next; | |
1770 | KVMState *s = current_env->kvm_state; | |
9349b4f9 | 1771 | CPUArchState *env; |
e22a25c9 | 1772 | |
72cf2d4f | 1773 | QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) { |
e22a25c9 AL |
1774 | if (kvm_arch_remove_sw_breakpoint(current_env, bp) != 0) { |
1775 | /* Try harder to find a CPU that currently sees the breakpoint. */ | |
1776 | for (env = first_cpu; env != NULL; env = env->next_cpu) { | |
a426e122 | 1777 | if (kvm_arch_remove_sw_breakpoint(env, bp) == 0) { |
e22a25c9 | 1778 | break; |
a426e122 | 1779 | } |
e22a25c9 AL |
1780 | } |
1781 | } | |
1782 | } | |
1783 | kvm_arch_remove_all_hw_breakpoints(); | |
1784 | ||
a426e122 | 1785 | for (env = first_cpu; env != NULL; env = env->next_cpu) { |
e22a25c9 | 1786 | kvm_update_guest_debug(env, 0); |
a426e122 | 1787 | } |
e22a25c9 AL |
1788 | } |
1789 | ||
1790 | #else /* !KVM_CAP_SET_GUEST_DEBUG */ | |
1791 | ||
9349b4f9 | 1792 | int kvm_update_guest_debug(CPUArchState *env, unsigned long reinject_trap) |
e22a25c9 AL |
1793 | { |
1794 | return -EINVAL; | |
1795 | } | |
1796 | ||
9349b4f9 | 1797 | int kvm_insert_breakpoint(CPUArchState *current_env, target_ulong addr, |
e22a25c9 AL |
1798 | target_ulong len, int type) |
1799 | { | |
1800 | return -EINVAL; | |
1801 | } | |
1802 | ||
9349b4f9 | 1803 | int kvm_remove_breakpoint(CPUArchState *current_env, target_ulong addr, |
e22a25c9 AL |
1804 | target_ulong len, int type) |
1805 | { | |
1806 | return -EINVAL; | |
1807 | } | |
1808 | ||
9349b4f9 | 1809 | void kvm_remove_all_breakpoints(CPUArchState *current_env) |
e22a25c9 AL |
1810 | { |
1811 | } | |
1812 | #endif /* !KVM_CAP_SET_GUEST_DEBUG */ | |
cc84de95 | 1813 | |
9349b4f9 | 1814 | int kvm_set_signal_mask(CPUArchState *env, const sigset_t *sigset) |
cc84de95 MT |
1815 | { |
1816 | struct kvm_signal_mask *sigmask; | |
1817 | int r; | |
1818 | ||
a426e122 | 1819 | if (!sigset) { |
cc84de95 | 1820 | return kvm_vcpu_ioctl(env, KVM_SET_SIGNAL_MASK, NULL); |
a426e122 | 1821 | } |
cc84de95 | 1822 | |
7267c094 | 1823 | sigmask = g_malloc(sizeof(*sigmask) + sizeof(*sigset)); |
cc84de95 MT |
1824 | |
1825 | sigmask->len = 8; | |
1826 | memcpy(sigmask->sigset, sigset, sizeof(*sigset)); | |
1827 | r = kvm_vcpu_ioctl(env, KVM_SET_SIGNAL_MASK, sigmask); | |
7267c094 | 1828 | g_free(sigmask); |
cc84de95 MT |
1829 | |
1830 | return r; | |
1831 | } | |
ca821806 | 1832 | |
4b8f1c88 MT |
1833 | int kvm_set_ioeventfd_mmio(int fd, uint32_t addr, uint32_t val, bool assign, |
1834 | uint32_t size) | |
44f1a3d8 | 1835 | { |
44f1a3d8 CM |
1836 | int ret; |
1837 | struct kvm_ioeventfd iofd; | |
1838 | ||
1839 | iofd.datamatch = val; | |
1840 | iofd.addr = addr; | |
4b8f1c88 | 1841 | iofd.len = size; |
44f1a3d8 CM |
1842 | iofd.flags = KVM_IOEVENTFD_FLAG_DATAMATCH; |
1843 | iofd.fd = fd; | |
1844 | ||
1845 | if (!kvm_enabled()) { | |
1846 | return -ENOSYS; | |
1847 | } | |
1848 | ||
1849 | if (!assign) { | |
1850 | iofd.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; | |
1851 | } | |
1852 | ||
1853 | ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd); | |
1854 | ||
1855 | if (ret < 0) { | |
1856 | return -errno; | |
1857 | } | |
1858 | ||
1859 | return 0; | |
44f1a3d8 CM |
1860 | } |
1861 | ||
ca821806 MT |
1862 | int kvm_set_ioeventfd_pio_word(int fd, uint16_t addr, uint16_t val, bool assign) |
1863 | { | |
1864 | struct kvm_ioeventfd kick = { | |
1865 | .datamatch = val, | |
1866 | .addr = addr, | |
1867 | .len = 2, | |
1868 | .flags = KVM_IOEVENTFD_FLAG_DATAMATCH | KVM_IOEVENTFD_FLAG_PIO, | |
1869 | .fd = fd, | |
1870 | }; | |
1871 | int r; | |
a426e122 | 1872 | if (!kvm_enabled()) { |
ca821806 | 1873 | return -ENOSYS; |
a426e122 JK |
1874 | } |
1875 | if (!assign) { | |
ca821806 | 1876 | kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; |
a426e122 | 1877 | } |
ca821806 | 1878 | r = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick); |
a426e122 | 1879 | if (r < 0) { |
ca821806 | 1880 | return r; |
a426e122 | 1881 | } |
ca821806 | 1882 | return 0; |
98c8573e | 1883 | } |
a1b87fe0 | 1884 | |
9349b4f9 | 1885 | int kvm_on_sigbus_vcpu(CPUArchState *env, int code, void *addr) |
a1b87fe0 JK |
1886 | { |
1887 | return kvm_arch_on_sigbus_vcpu(env, code, addr); | |
1888 | } | |
1889 | ||
1890 | int kvm_on_sigbus(int code, void *addr) | |
1891 | { | |
1892 | return kvm_arch_on_sigbus(code, addr); | |
1893 | } |