]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * QEMU KVM support | |
3 | * | |
4 | * Copyright IBM, Corp. 2008 | |
5 | * Red Hat, Inc. 2008 | |
6 | * | |
7 | * Authors: | |
8 | * Anthony Liguori <aliguori@us.ibm.com> | |
9 | * Glauber Costa <gcosta@redhat.com> | |
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> | |
19 | #include <stdarg.h> | |
20 | ||
21 | #include <linux/kvm.h> | |
22 | ||
23 | #include "qemu-common.h" | |
24 | #include "qemu/atomic.h" | |
25 | #include "qemu/option.h" | |
26 | #include "qemu/config-file.h" | |
27 | #include "qemu/error-report.h" | |
28 | #include "hw/hw.h" | |
29 | #include "hw/pci/msi.h" | |
30 | #include "hw/s390x/adapter.h" | |
31 | #include "exec/gdbstub.h" | |
32 | #include "sysemu/kvm_int.h" | |
33 | #include "qemu/bswap.h" | |
34 | #include "exec/memory.h" | |
35 | #include "exec/ram_addr.h" | |
36 | #include "exec/address-spaces.h" | |
37 | #include "qemu/event_notifier.h" | |
38 | #include "trace.h" | |
39 | #include "hw/irq.h" | |
40 | ||
41 | #include "hw/boards.h" | |
42 | ||
43 | /* This check must be after config-host.h is included */ | |
44 | #ifdef CONFIG_EVENTFD | |
45 | #include <sys/eventfd.h> | |
46 | #endif | |
47 | ||
48 | /* KVM uses PAGE_SIZE in its definition of COALESCED_MMIO_MAX */ | |
49 | #define PAGE_SIZE TARGET_PAGE_SIZE | |
50 | ||
51 | //#define DEBUG_KVM | |
52 | ||
53 | #ifdef DEBUG_KVM | |
54 | #define DPRINTF(fmt, ...) \ | |
55 | do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0) | |
56 | #else | |
57 | #define DPRINTF(fmt, ...) \ | |
58 | do { } while (0) | |
59 | #endif | |
60 | ||
61 | #define KVM_MSI_HASHTAB_SIZE 256 | |
62 | ||
63 | struct KVMState | |
64 | { | |
65 | AccelState parent_obj; | |
66 | ||
67 | int nr_slots; | |
68 | int fd; | |
69 | int vmfd; | |
70 | int coalesced_mmio; | |
71 | struct kvm_coalesced_mmio_ring *coalesced_mmio_ring; | |
72 | bool coalesced_flush_in_progress; | |
73 | int broken_set_mem_region; | |
74 | int vcpu_events; | |
75 | int robust_singlestep; | |
76 | int debugregs; | |
77 | #ifdef KVM_CAP_SET_GUEST_DEBUG | |
78 | struct kvm_sw_breakpoint_head kvm_sw_breakpoints; | |
79 | #endif | |
80 | int pit_state2; | |
81 | int xsave, xcrs; | |
82 | int many_ioeventfds; | |
83 | int intx_set_mask; | |
84 | /* The man page (and posix) say ioctl numbers are signed int, but | |
85 | * they're not. Linux, glibc and *BSD all treat ioctl numbers as | |
86 | * unsigned, and treating them as signed here can break things */ | |
87 | unsigned irq_set_ioctl; | |
88 | unsigned int sigmask_len; | |
89 | GHashTable *gsimap; | |
90 | #ifdef KVM_CAP_IRQ_ROUTING | |
91 | struct kvm_irq_routing *irq_routes; | |
92 | int nr_allocated_irq_routes; | |
93 | uint32_t *used_gsi_bitmap; | |
94 | unsigned int gsi_count; | |
95 | QTAILQ_HEAD(msi_hashtab, KVMMSIRoute) msi_hashtab[KVM_MSI_HASHTAB_SIZE]; | |
96 | bool direct_msi; | |
97 | #endif | |
98 | KVMMemoryListener memory_listener; | |
99 | }; | |
100 | ||
101 | KVMState *kvm_state; | |
102 | bool kvm_kernel_irqchip; | |
103 | bool kvm_async_interrupts_allowed; | |
104 | bool kvm_halt_in_kernel_allowed; | |
105 | bool kvm_eventfds_allowed; | |
106 | bool kvm_irqfds_allowed; | |
107 | bool kvm_resamplefds_allowed; | |
108 | bool kvm_msi_via_irqfd_allowed; | |
109 | bool kvm_gsi_routing_allowed; | |
110 | bool kvm_gsi_direct_mapping; | |
111 | bool kvm_allowed; | |
112 | bool kvm_readonly_mem_allowed; | |
113 | bool kvm_vm_attributes_allowed; | |
114 | ||
115 | static const KVMCapabilityInfo kvm_required_capabilites[] = { | |
116 | KVM_CAP_INFO(USER_MEMORY), | |
117 | KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS), | |
118 | KVM_CAP_LAST_INFO | |
119 | }; | |
120 | ||
121 | static KVMSlot *kvm_get_free_slot(KVMMemoryListener *kml) | |
122 | { | |
123 | KVMState *s = kvm_state; | |
124 | int i; | |
125 | ||
126 | for (i = 0; i < s->nr_slots; i++) { | |
127 | if (kml->slots[i].memory_size == 0) { | |
128 | return &kml->slots[i]; | |
129 | } | |
130 | } | |
131 | ||
132 | return NULL; | |
133 | } | |
134 | ||
135 | bool kvm_has_free_slot(MachineState *ms) | |
136 | { | |
137 | KVMState *s = KVM_STATE(ms->accelerator); | |
138 | ||
139 | return kvm_get_free_slot(&s->memory_listener); | |
140 | } | |
141 | ||
142 | static KVMSlot *kvm_alloc_slot(KVMMemoryListener *kml) | |
143 | { | |
144 | KVMSlot *slot = kvm_get_free_slot(kml); | |
145 | ||
146 | if (slot) { | |
147 | return slot; | |
148 | } | |
149 | ||
150 | fprintf(stderr, "%s: no free slot available\n", __func__); | |
151 | abort(); | |
152 | } | |
153 | ||
154 | static KVMSlot *kvm_lookup_matching_slot(KVMMemoryListener *kml, | |
155 | hwaddr start_addr, | |
156 | hwaddr end_addr) | |
157 | { | |
158 | KVMState *s = kvm_state; | |
159 | int i; | |
160 | ||
161 | for (i = 0; i < s->nr_slots; i++) { | |
162 | KVMSlot *mem = &kml->slots[i]; | |
163 | ||
164 | if (start_addr == mem->start_addr && | |
165 | end_addr == mem->start_addr + mem->memory_size) { | |
166 | return mem; | |
167 | } | |
168 | } | |
169 | ||
170 | return NULL; | |
171 | } | |
172 | ||
173 | /* | |
174 | * Find overlapping slot with lowest start address | |
175 | */ | |
176 | static KVMSlot *kvm_lookup_overlapping_slot(KVMMemoryListener *kml, | |
177 | hwaddr start_addr, | |
178 | hwaddr end_addr) | |
179 | { | |
180 | KVMState *s = kvm_state; | |
181 | KVMSlot *found = NULL; | |
182 | int i; | |
183 | ||
184 | for (i = 0; i < s->nr_slots; i++) { | |
185 | KVMSlot *mem = &kml->slots[i]; | |
186 | ||
187 | if (mem->memory_size == 0 || | |
188 | (found && found->start_addr < mem->start_addr)) { | |
189 | continue; | |
190 | } | |
191 | ||
192 | if (end_addr > mem->start_addr && | |
193 | start_addr < mem->start_addr + mem->memory_size) { | |
194 | found = mem; | |
195 | } | |
196 | } | |
197 | ||
198 | return found; | |
199 | } | |
200 | ||
201 | int kvm_physical_memory_addr_from_host(KVMState *s, void *ram, | |
202 | hwaddr *phys_addr) | |
203 | { | |
204 | KVMMemoryListener *kml = &s->memory_listener; | |
205 | int i; | |
206 | ||
207 | for (i = 0; i < s->nr_slots; i++) { | |
208 | KVMSlot *mem = &kml->slots[i]; | |
209 | ||
210 | if (ram >= mem->ram && ram < mem->ram + mem->memory_size) { | |
211 | *phys_addr = mem->start_addr + (ram - mem->ram); | |
212 | return 1; | |
213 | } | |
214 | } | |
215 | ||
216 | return 0; | |
217 | } | |
218 | ||
219 | static int kvm_set_user_memory_region(KVMMemoryListener *kml, KVMSlot *slot) | |
220 | { | |
221 | KVMState *s = kvm_state; | |
222 | struct kvm_userspace_memory_region mem; | |
223 | ||
224 | mem.slot = slot->slot | (kml->as_id << 16); | |
225 | mem.guest_phys_addr = slot->start_addr; | |
226 | mem.userspace_addr = (unsigned long)slot->ram; | |
227 | mem.flags = slot->flags; | |
228 | ||
229 | if (slot->memory_size && mem.flags & KVM_MEM_READONLY) { | |
230 | /* Set the slot size to 0 before setting the slot to the desired | |
231 | * value. This is needed based on KVM commit 75d61fbc. */ | |
232 | mem.memory_size = 0; | |
233 | kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem); | |
234 | } | |
235 | mem.memory_size = slot->memory_size; | |
236 | return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem); | |
237 | } | |
238 | ||
239 | int kvm_init_vcpu(CPUState *cpu) | |
240 | { | |
241 | KVMState *s = kvm_state; | |
242 | long mmap_size; | |
243 | int ret; | |
244 | ||
245 | DPRINTF("kvm_init_vcpu\n"); | |
246 | ||
247 | ret = kvm_vm_ioctl(s, KVM_CREATE_VCPU, (void *)kvm_arch_vcpu_id(cpu)); | |
248 | if (ret < 0) { | |
249 | DPRINTF("kvm_create_vcpu failed\n"); | |
250 | goto err; | |
251 | } | |
252 | ||
253 | cpu->kvm_fd = ret; | |
254 | cpu->kvm_state = s; | |
255 | cpu->kvm_vcpu_dirty = true; | |
256 | ||
257 | mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0); | |
258 | if (mmap_size < 0) { | |
259 | ret = mmap_size; | |
260 | DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n"); | |
261 | goto err; | |
262 | } | |
263 | ||
264 | cpu->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED, | |
265 | cpu->kvm_fd, 0); | |
266 | if (cpu->kvm_run == MAP_FAILED) { | |
267 | ret = -errno; | |
268 | DPRINTF("mmap'ing vcpu state failed\n"); | |
269 | goto err; | |
270 | } | |
271 | ||
272 | if (s->coalesced_mmio && !s->coalesced_mmio_ring) { | |
273 | s->coalesced_mmio_ring = | |
274 | (void *)cpu->kvm_run + s->coalesced_mmio * PAGE_SIZE; | |
275 | } | |
276 | ||
277 | ret = kvm_arch_init_vcpu(cpu); | |
278 | err: | |
279 | return ret; | |
280 | } | |
281 | ||
282 | /* | |
283 | * dirty pages logging control | |
284 | */ | |
285 | ||
286 | static int kvm_mem_flags(MemoryRegion *mr) | |
287 | { | |
288 | bool readonly = mr->readonly || memory_region_is_romd(mr); | |
289 | int flags = 0; | |
290 | ||
291 | if (memory_region_get_dirty_log_mask(mr) != 0) { | |
292 | flags |= KVM_MEM_LOG_DIRTY_PAGES; | |
293 | } | |
294 | if (readonly && kvm_readonly_mem_allowed) { | |
295 | flags |= KVM_MEM_READONLY; | |
296 | } | |
297 | return flags; | |
298 | } | |
299 | ||
300 | static int kvm_slot_update_flags(KVMMemoryListener *kml, KVMSlot *mem, | |
301 | MemoryRegion *mr) | |
302 | { | |
303 | int old_flags; | |
304 | ||
305 | old_flags = mem->flags; | |
306 | mem->flags = kvm_mem_flags(mr); | |
307 | ||
308 | /* If nothing changed effectively, no need to issue ioctl */ | |
309 | if (mem->flags == old_flags) { | |
310 | return 0; | |
311 | } | |
312 | ||
313 | return kvm_set_user_memory_region(kml, mem); | |
314 | } | |
315 | ||
316 | static int kvm_section_update_flags(KVMMemoryListener *kml, | |
317 | MemoryRegionSection *section) | |
318 | { | |
319 | hwaddr phys_addr = section->offset_within_address_space; | |
320 | ram_addr_t size = int128_get64(section->size); | |
321 | KVMSlot *mem = kvm_lookup_matching_slot(kml, phys_addr, phys_addr + size); | |
322 | ||
323 | if (mem == NULL) { | |
324 | return 0; | |
325 | } else { | |
326 | return kvm_slot_update_flags(kml, mem, section->mr); | |
327 | } | |
328 | } | |
329 | ||
330 | static void kvm_log_start(MemoryListener *listener, | |
331 | MemoryRegionSection *section, | |
332 | int old, int new) | |
333 | { | |
334 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); | |
335 | int r; | |
336 | ||
337 | if (old != 0) { | |
338 | return; | |
339 | } | |
340 | ||
341 | r = kvm_section_update_flags(kml, section); | |
342 | if (r < 0) { | |
343 | abort(); | |
344 | } | |
345 | } | |
346 | ||
347 | static void kvm_log_stop(MemoryListener *listener, | |
348 | MemoryRegionSection *section, | |
349 | int old, int new) | |
350 | { | |
351 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); | |
352 | int r; | |
353 | ||
354 | if (new != 0) { | |
355 | return; | |
356 | } | |
357 | ||
358 | r = kvm_section_update_flags(kml, section); | |
359 | if (r < 0) { | |
360 | abort(); | |
361 | } | |
362 | } | |
363 | ||
364 | /* get kvm's dirty pages bitmap and update qemu's */ | |
365 | static int kvm_get_dirty_pages_log_range(MemoryRegionSection *section, | |
366 | unsigned long *bitmap) | |
367 | { | |
368 | ram_addr_t start = section->offset_within_region + section->mr->ram_addr; | |
369 | ram_addr_t pages = int128_get64(section->size) / getpagesize(); | |
370 | ||
371 | cpu_physical_memory_set_dirty_lebitmap(bitmap, start, pages); | |
372 | return 0; | |
373 | } | |
374 | ||
375 | #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1)) | |
376 | ||
377 | /** | |
378 | * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space | |
379 | * This function updates qemu's dirty bitmap using | |
380 | * memory_region_set_dirty(). This means all bits are set | |
381 | * to dirty. | |
382 | * | |
383 | * @start_add: start of logged region. | |
384 | * @end_addr: end of logged region. | |
385 | */ | |
386 | static int kvm_physical_sync_dirty_bitmap(KVMMemoryListener *kml, | |
387 | MemoryRegionSection *section) | |
388 | { | |
389 | KVMState *s = kvm_state; | |
390 | unsigned long size, allocated_size = 0; | |
391 | struct kvm_dirty_log d = {}; | |
392 | KVMSlot *mem; | |
393 | int ret = 0; | |
394 | hwaddr start_addr = section->offset_within_address_space; | |
395 | hwaddr end_addr = start_addr + int128_get64(section->size); | |
396 | ||
397 | d.dirty_bitmap = NULL; | |
398 | while (start_addr < end_addr) { | |
399 | mem = kvm_lookup_overlapping_slot(kml, start_addr, end_addr); | |
400 | if (mem == NULL) { | |
401 | break; | |
402 | } | |
403 | ||
404 | /* XXX bad kernel interface alert | |
405 | * For dirty bitmap, kernel allocates array of size aligned to | |
406 | * bits-per-long. But for case when the kernel is 64bits and | |
407 | * the userspace is 32bits, userspace can't align to the same | |
408 | * bits-per-long, since sizeof(long) is different between kernel | |
409 | * and user space. This way, userspace will provide buffer which | |
410 | * may be 4 bytes less than the kernel will use, resulting in | |
411 | * userspace memory corruption (which is not detectable by valgrind | |
412 | * too, in most cases). | |
413 | * So for now, let's align to 64 instead of HOST_LONG_BITS here, in | |
414 | * a hope that sizeof(long) wont become >8 any time soon. | |
415 | */ | |
416 | size = ALIGN(((mem->memory_size) >> TARGET_PAGE_BITS), | |
417 | /*HOST_LONG_BITS*/ 64) / 8; | |
418 | if (!d.dirty_bitmap) { | |
419 | d.dirty_bitmap = g_malloc(size); | |
420 | } else if (size > allocated_size) { | |
421 | d.dirty_bitmap = g_realloc(d.dirty_bitmap, size); | |
422 | } | |
423 | allocated_size = size; | |
424 | memset(d.dirty_bitmap, 0, allocated_size); | |
425 | ||
426 | d.slot = mem->slot | (kml->as_id << 16); | |
427 | if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) { | |
428 | DPRINTF("ioctl failed %d\n", errno); | |
429 | ret = -1; | |
430 | break; | |
431 | } | |
432 | ||
433 | kvm_get_dirty_pages_log_range(section, d.dirty_bitmap); | |
434 | start_addr = mem->start_addr + mem->memory_size; | |
435 | } | |
436 | g_free(d.dirty_bitmap); | |
437 | ||
438 | return ret; | |
439 | } | |
440 | ||
441 | static void kvm_coalesce_mmio_region(MemoryListener *listener, | |
442 | MemoryRegionSection *secion, | |
443 | hwaddr start, hwaddr size) | |
444 | { | |
445 | KVMState *s = kvm_state; | |
446 | ||
447 | if (s->coalesced_mmio) { | |
448 | struct kvm_coalesced_mmio_zone zone; | |
449 | ||
450 | zone.addr = start; | |
451 | zone.size = size; | |
452 | zone.pad = 0; | |
453 | ||
454 | (void)kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone); | |
455 | } | |
456 | } | |
457 | ||
458 | static void kvm_uncoalesce_mmio_region(MemoryListener *listener, | |
459 | MemoryRegionSection *secion, | |
460 | hwaddr start, hwaddr size) | |
461 | { | |
462 | KVMState *s = kvm_state; | |
463 | ||
464 | if (s->coalesced_mmio) { | |
465 | struct kvm_coalesced_mmio_zone zone; | |
466 | ||
467 | zone.addr = start; | |
468 | zone.size = size; | |
469 | zone.pad = 0; | |
470 | ||
471 | (void)kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone); | |
472 | } | |
473 | } | |
474 | ||
475 | int kvm_check_extension(KVMState *s, unsigned int extension) | |
476 | { | |
477 | int ret; | |
478 | ||
479 | ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension); | |
480 | if (ret < 0) { | |
481 | ret = 0; | |
482 | } | |
483 | ||
484 | return ret; | |
485 | } | |
486 | ||
487 | int kvm_vm_check_extension(KVMState *s, unsigned int extension) | |
488 | { | |
489 | int ret; | |
490 | ||
491 | ret = kvm_vm_ioctl(s, KVM_CHECK_EXTENSION, extension); | |
492 | if (ret < 0) { | |
493 | /* VM wide version not implemented, use global one instead */ | |
494 | ret = kvm_check_extension(s, extension); | |
495 | } | |
496 | ||
497 | return ret; | |
498 | } | |
499 | ||
500 | static uint32_t adjust_ioeventfd_endianness(uint32_t val, uint32_t size) | |
501 | { | |
502 | #if defined(HOST_WORDS_BIGENDIAN) != defined(TARGET_WORDS_BIGENDIAN) | |
503 | /* The kernel expects ioeventfd values in HOST_WORDS_BIGENDIAN | |
504 | * endianness, but the memory core hands them in target endianness. | |
505 | * For example, PPC is always treated as big-endian even if running | |
506 | * on KVM and on PPC64LE. Correct here. | |
507 | */ | |
508 | switch (size) { | |
509 | case 2: | |
510 | val = bswap16(val); | |
511 | break; | |
512 | case 4: | |
513 | val = bswap32(val); | |
514 | break; | |
515 | } | |
516 | #endif | |
517 | return val; | |
518 | } | |
519 | ||
520 | static int kvm_set_ioeventfd_mmio(int fd, hwaddr addr, uint32_t val, | |
521 | bool assign, uint32_t size, bool datamatch) | |
522 | { | |
523 | int ret; | |
524 | struct kvm_ioeventfd iofd = { | |
525 | .datamatch = datamatch ? adjust_ioeventfd_endianness(val, size) : 0, | |
526 | .addr = addr, | |
527 | .len = size, | |
528 | .flags = 0, | |
529 | .fd = fd, | |
530 | }; | |
531 | ||
532 | if (!kvm_enabled()) { | |
533 | return -ENOSYS; | |
534 | } | |
535 | ||
536 | if (datamatch) { | |
537 | iofd.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH; | |
538 | } | |
539 | if (!assign) { | |
540 | iofd.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; | |
541 | } | |
542 | ||
543 | ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd); | |
544 | ||
545 | if (ret < 0) { | |
546 | return -errno; | |
547 | } | |
548 | ||
549 | return 0; | |
550 | } | |
551 | ||
552 | static int kvm_set_ioeventfd_pio(int fd, uint16_t addr, uint16_t val, | |
553 | bool assign, uint32_t size, bool datamatch) | |
554 | { | |
555 | struct kvm_ioeventfd kick = { | |
556 | .datamatch = datamatch ? adjust_ioeventfd_endianness(val, size) : 0, | |
557 | .addr = addr, | |
558 | .flags = KVM_IOEVENTFD_FLAG_PIO, | |
559 | .len = size, | |
560 | .fd = fd, | |
561 | }; | |
562 | int r; | |
563 | if (!kvm_enabled()) { | |
564 | return -ENOSYS; | |
565 | } | |
566 | if (datamatch) { | |
567 | kick.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH; | |
568 | } | |
569 | if (!assign) { | |
570 | kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; | |
571 | } | |
572 | r = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick); | |
573 | if (r < 0) { | |
574 | return r; | |
575 | } | |
576 | return 0; | |
577 | } | |
578 | ||
579 | ||
580 | static int kvm_check_many_ioeventfds(void) | |
581 | { | |
582 | /* Userspace can use ioeventfd for io notification. This requires a host | |
583 | * that supports eventfd(2) and an I/O thread; since eventfd does not | |
584 | * support SIGIO it cannot interrupt the vcpu. | |
585 | * | |
586 | * Older kernels have a 6 device limit on the KVM io bus. Find out so we | |
587 | * can avoid creating too many ioeventfds. | |
588 | */ | |
589 | #if defined(CONFIG_EVENTFD) | |
590 | int ioeventfds[7]; | |
591 | int i, ret = 0; | |
592 | for (i = 0; i < ARRAY_SIZE(ioeventfds); i++) { | |
593 | ioeventfds[i] = eventfd(0, EFD_CLOEXEC); | |
594 | if (ioeventfds[i] < 0) { | |
595 | break; | |
596 | } | |
597 | ret = kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, true, 2, true); | |
598 | if (ret < 0) { | |
599 | close(ioeventfds[i]); | |
600 | break; | |
601 | } | |
602 | } | |
603 | ||
604 | /* Decide whether many devices are supported or not */ | |
605 | ret = i == ARRAY_SIZE(ioeventfds); | |
606 | ||
607 | while (i-- > 0) { | |
608 | kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, false, 2, true); | |
609 | close(ioeventfds[i]); | |
610 | } | |
611 | return ret; | |
612 | #else | |
613 | return 0; | |
614 | #endif | |
615 | } | |
616 | ||
617 | static const KVMCapabilityInfo * | |
618 | kvm_check_extension_list(KVMState *s, const KVMCapabilityInfo *list) | |
619 | { | |
620 | while (list->name) { | |
621 | if (!kvm_check_extension(s, list->value)) { | |
622 | return list; | |
623 | } | |
624 | list++; | |
625 | } | |
626 | return NULL; | |
627 | } | |
628 | ||
629 | static void kvm_set_phys_mem(KVMMemoryListener *kml, | |
630 | MemoryRegionSection *section, bool add) | |
631 | { | |
632 | KVMState *s = kvm_state; | |
633 | KVMSlot *mem, old; | |
634 | int err; | |
635 | MemoryRegion *mr = section->mr; | |
636 | bool writeable = !mr->readonly && !mr->rom_device; | |
637 | hwaddr start_addr = section->offset_within_address_space; | |
638 | ram_addr_t size = int128_get64(section->size); | |
639 | void *ram = NULL; | |
640 | unsigned delta; | |
641 | ||
642 | /* kvm works in page size chunks, but the function may be called | |
643 | with sub-page size and unaligned start address. Pad the start | |
644 | address to next and truncate size to previous page boundary. */ | |
645 | delta = (TARGET_PAGE_SIZE - (start_addr & ~TARGET_PAGE_MASK)); | |
646 | delta &= ~TARGET_PAGE_MASK; | |
647 | if (delta > size) { | |
648 | return; | |
649 | } | |
650 | start_addr += delta; | |
651 | size -= delta; | |
652 | size &= TARGET_PAGE_MASK; | |
653 | if (!size || (start_addr & ~TARGET_PAGE_MASK)) { | |
654 | return; | |
655 | } | |
656 | ||
657 | if (!memory_region_is_ram(mr)) { | |
658 | if (writeable || !kvm_readonly_mem_allowed) { | |
659 | return; | |
660 | } else if (!mr->romd_mode) { | |
661 | /* If the memory device is not in romd_mode, then we actually want | |
662 | * to remove the kvm memory slot so all accesses will trap. */ | |
663 | add = false; | |
664 | } | |
665 | } | |
666 | ||
667 | ram = memory_region_get_ram_ptr(mr) + section->offset_within_region + delta; | |
668 | ||
669 | while (1) { | |
670 | mem = kvm_lookup_overlapping_slot(kml, start_addr, start_addr + size); | |
671 | if (!mem) { | |
672 | break; | |
673 | } | |
674 | ||
675 | if (add && start_addr >= mem->start_addr && | |
676 | (start_addr + size <= mem->start_addr + mem->memory_size) && | |
677 | (ram - start_addr == mem->ram - mem->start_addr)) { | |
678 | /* The new slot fits into the existing one and comes with | |
679 | * identical parameters - update flags and done. */ | |
680 | kvm_slot_update_flags(kml, mem, mr); | |
681 | return; | |
682 | } | |
683 | ||
684 | old = *mem; | |
685 | ||
686 | if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) { | |
687 | kvm_physical_sync_dirty_bitmap(kml, section); | |
688 | } | |
689 | ||
690 | /* unregister the overlapping slot */ | |
691 | mem->memory_size = 0; | |
692 | err = kvm_set_user_memory_region(kml, mem); | |
693 | if (err) { | |
694 | fprintf(stderr, "%s: error unregistering overlapping slot: %s\n", | |
695 | __func__, strerror(-err)); | |
696 | abort(); | |
697 | } | |
698 | ||
699 | /* Workaround for older KVM versions: we can't join slots, even not by | |
700 | * unregistering the previous ones and then registering the larger | |
701 | * slot. We have to maintain the existing fragmentation. Sigh. | |
702 | * | |
703 | * This workaround assumes that the new slot starts at the same | |
704 | * address as the first existing one. If not or if some overlapping | |
705 | * slot comes around later, we will fail (not seen in practice so far) | |
706 | * - and actually require a recent KVM version. */ | |
707 | if (s->broken_set_mem_region && | |
708 | old.start_addr == start_addr && old.memory_size < size && add) { | |
709 | mem = kvm_alloc_slot(kml); | |
710 | mem->memory_size = old.memory_size; | |
711 | mem->start_addr = old.start_addr; | |
712 | mem->ram = old.ram; | |
713 | mem->flags = kvm_mem_flags(mr); | |
714 | ||
715 | err = kvm_set_user_memory_region(kml, mem); | |
716 | if (err) { | |
717 | fprintf(stderr, "%s: error updating slot: %s\n", __func__, | |
718 | strerror(-err)); | |
719 | abort(); | |
720 | } | |
721 | ||
722 | start_addr += old.memory_size; | |
723 | ram += old.memory_size; | |
724 | size -= old.memory_size; | |
725 | continue; | |
726 | } | |
727 | ||
728 | /* register prefix slot */ | |
729 | if (old.start_addr < start_addr) { | |
730 | mem = kvm_alloc_slot(kml); | |
731 | mem->memory_size = start_addr - old.start_addr; | |
732 | mem->start_addr = old.start_addr; | |
733 | mem->ram = old.ram; | |
734 | mem->flags = kvm_mem_flags(mr); | |
735 | ||
736 | err = kvm_set_user_memory_region(kml, mem); | |
737 | if (err) { | |
738 | fprintf(stderr, "%s: error registering prefix slot: %s\n", | |
739 | __func__, strerror(-err)); | |
740 | #ifdef TARGET_PPC | |
741 | fprintf(stderr, "%s: This is probably because your kernel's " \ | |
742 | "PAGE_SIZE is too big. Please try to use 4k " \ | |
743 | "PAGE_SIZE!\n", __func__); | |
744 | #endif | |
745 | abort(); | |
746 | } | |
747 | } | |
748 | ||
749 | /* register suffix slot */ | |
750 | if (old.start_addr + old.memory_size > start_addr + size) { | |
751 | ram_addr_t size_delta; | |
752 | ||
753 | mem = kvm_alloc_slot(kml); | |
754 | mem->start_addr = start_addr + size; | |
755 | size_delta = mem->start_addr - old.start_addr; | |
756 | mem->memory_size = old.memory_size - size_delta; | |
757 | mem->ram = old.ram + size_delta; | |
758 | mem->flags = kvm_mem_flags(mr); | |
759 | ||
760 | err = kvm_set_user_memory_region(kml, mem); | |
761 | if (err) { | |
762 | fprintf(stderr, "%s: error registering suffix slot: %s\n", | |
763 | __func__, strerror(-err)); | |
764 | abort(); | |
765 | } | |
766 | } | |
767 | } | |
768 | ||
769 | /* in case the KVM bug workaround already "consumed" the new slot */ | |
770 | if (!size) { | |
771 | return; | |
772 | } | |
773 | if (!add) { | |
774 | return; | |
775 | } | |
776 | mem = kvm_alloc_slot(kml); | |
777 | mem->memory_size = size; | |
778 | mem->start_addr = start_addr; | |
779 | mem->ram = ram; | |
780 | mem->flags = kvm_mem_flags(mr); | |
781 | ||
782 | err = kvm_set_user_memory_region(kml, mem); | |
783 | if (err) { | |
784 | fprintf(stderr, "%s: error registering slot: %s\n", __func__, | |
785 | strerror(-err)); | |
786 | abort(); | |
787 | } | |
788 | } | |
789 | ||
790 | static void kvm_region_add(MemoryListener *listener, | |
791 | MemoryRegionSection *section) | |
792 | { | |
793 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); | |
794 | ||
795 | memory_region_ref(section->mr); | |
796 | kvm_set_phys_mem(kml, section, true); | |
797 | } | |
798 | ||
799 | static void kvm_region_del(MemoryListener *listener, | |
800 | MemoryRegionSection *section) | |
801 | { | |
802 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); | |
803 | ||
804 | kvm_set_phys_mem(kml, section, false); | |
805 | memory_region_unref(section->mr); | |
806 | } | |
807 | ||
808 | static void kvm_log_sync(MemoryListener *listener, | |
809 | MemoryRegionSection *section) | |
810 | { | |
811 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); | |
812 | int r; | |
813 | ||
814 | r = kvm_physical_sync_dirty_bitmap(kml, section); | |
815 | if (r < 0) { | |
816 | abort(); | |
817 | } | |
818 | } | |
819 | ||
820 | static void kvm_mem_ioeventfd_add(MemoryListener *listener, | |
821 | MemoryRegionSection *section, | |
822 | bool match_data, uint64_t data, | |
823 | EventNotifier *e) | |
824 | { | |
825 | int fd = event_notifier_get_fd(e); | |
826 | int r; | |
827 | ||
828 | r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space, | |
829 | data, true, int128_get64(section->size), | |
830 | match_data); | |
831 | if (r < 0) { | |
832 | fprintf(stderr, "%s: error adding ioeventfd: %s\n", | |
833 | __func__, strerror(-r)); | |
834 | abort(); | |
835 | } | |
836 | } | |
837 | ||
838 | static void kvm_mem_ioeventfd_del(MemoryListener *listener, | |
839 | MemoryRegionSection *section, | |
840 | bool match_data, uint64_t data, | |
841 | EventNotifier *e) | |
842 | { | |
843 | int fd = event_notifier_get_fd(e); | |
844 | int r; | |
845 | ||
846 | r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space, | |
847 | data, false, int128_get64(section->size), | |
848 | match_data); | |
849 | if (r < 0) { | |
850 | abort(); | |
851 | } | |
852 | } | |
853 | ||
854 | static void kvm_io_ioeventfd_add(MemoryListener *listener, | |
855 | MemoryRegionSection *section, | |
856 | bool match_data, uint64_t data, | |
857 | EventNotifier *e) | |
858 | { | |
859 | int fd = event_notifier_get_fd(e); | |
860 | int r; | |
861 | ||
862 | r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space, | |
863 | data, true, int128_get64(section->size), | |
864 | match_data); | |
865 | if (r < 0) { | |
866 | fprintf(stderr, "%s: error adding ioeventfd: %s\n", | |
867 | __func__, strerror(-r)); | |
868 | abort(); | |
869 | } | |
870 | } | |
871 | ||
872 | static void kvm_io_ioeventfd_del(MemoryListener *listener, | |
873 | MemoryRegionSection *section, | |
874 | bool match_data, uint64_t data, | |
875 | EventNotifier *e) | |
876 | ||
877 | { | |
878 | int fd = event_notifier_get_fd(e); | |
879 | int r; | |
880 | ||
881 | r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space, | |
882 | data, false, int128_get64(section->size), | |
883 | match_data); | |
884 | if (r < 0) { | |
885 | abort(); | |
886 | } | |
887 | } | |
888 | ||
889 | void kvm_memory_listener_register(KVMState *s, KVMMemoryListener *kml, | |
890 | AddressSpace *as, int as_id) | |
891 | { | |
892 | int i; | |
893 | ||
894 | kml->slots = g_malloc0(s->nr_slots * sizeof(KVMSlot)); | |
895 | kml->as_id = as_id; | |
896 | ||
897 | for (i = 0; i < s->nr_slots; i++) { | |
898 | kml->slots[i].slot = i; | |
899 | } | |
900 | ||
901 | kml->listener.region_add = kvm_region_add; | |
902 | kml->listener.region_del = kvm_region_del; | |
903 | kml->listener.log_start = kvm_log_start; | |
904 | kml->listener.log_stop = kvm_log_stop; | |
905 | kml->listener.log_sync = kvm_log_sync; | |
906 | kml->listener.priority = 10; | |
907 | ||
908 | memory_listener_register(&kml->listener, as); | |
909 | } | |
910 | ||
911 | static MemoryListener kvm_io_listener = { | |
912 | .eventfd_add = kvm_io_ioeventfd_add, | |
913 | .eventfd_del = kvm_io_ioeventfd_del, | |
914 | .priority = 10, | |
915 | }; | |
916 | ||
917 | static void kvm_handle_interrupt(CPUState *cpu, int mask) | |
918 | { | |
919 | cpu->interrupt_request |= mask; | |
920 | ||
921 | if (!qemu_cpu_is_self(cpu)) { | |
922 | qemu_cpu_kick(cpu); | |
923 | } | |
924 | } | |
925 | ||
926 | int kvm_set_irq(KVMState *s, int irq, int level) | |
927 | { | |
928 | struct kvm_irq_level event; | |
929 | int ret; | |
930 | ||
931 | assert(kvm_async_interrupts_enabled()); | |
932 | ||
933 | event.level = level; | |
934 | event.irq = irq; | |
935 | ret = kvm_vm_ioctl(s, s->irq_set_ioctl, &event); | |
936 | if (ret < 0) { | |
937 | perror("kvm_set_irq"); | |
938 | abort(); | |
939 | } | |
940 | ||
941 | return (s->irq_set_ioctl == KVM_IRQ_LINE) ? 1 : event.status; | |
942 | } | |
943 | ||
944 | #ifdef KVM_CAP_IRQ_ROUTING | |
945 | typedef struct KVMMSIRoute { | |
946 | struct kvm_irq_routing_entry kroute; | |
947 | QTAILQ_ENTRY(KVMMSIRoute) entry; | |
948 | } KVMMSIRoute; | |
949 | ||
950 | static void set_gsi(KVMState *s, unsigned int gsi) | |
951 | { | |
952 | s->used_gsi_bitmap[gsi / 32] |= 1U << (gsi % 32); | |
953 | } | |
954 | ||
955 | static void clear_gsi(KVMState *s, unsigned int gsi) | |
956 | { | |
957 | s->used_gsi_bitmap[gsi / 32] &= ~(1U << (gsi % 32)); | |
958 | } | |
959 | ||
960 | void kvm_init_irq_routing(KVMState *s) | |
961 | { | |
962 | int gsi_count, i; | |
963 | ||
964 | gsi_count = kvm_check_extension(s, KVM_CAP_IRQ_ROUTING) - 1; | |
965 | if (gsi_count > 0) { | |
966 | unsigned int gsi_bits, i; | |
967 | ||
968 | /* Round up so we can search ints using ffs */ | |
969 | gsi_bits = ALIGN(gsi_count, 32); | |
970 | s->used_gsi_bitmap = g_malloc0(gsi_bits / 8); | |
971 | s->gsi_count = gsi_count; | |
972 | ||
973 | /* Mark any over-allocated bits as already in use */ | |
974 | for (i = gsi_count; i < gsi_bits; i++) { | |
975 | set_gsi(s, i); | |
976 | } | |
977 | } | |
978 | ||
979 | s->irq_routes = g_malloc0(sizeof(*s->irq_routes)); | |
980 | s->nr_allocated_irq_routes = 0; | |
981 | ||
982 | if (!s->direct_msi) { | |
983 | for (i = 0; i < KVM_MSI_HASHTAB_SIZE; i++) { | |
984 | QTAILQ_INIT(&s->msi_hashtab[i]); | |
985 | } | |
986 | } | |
987 | ||
988 | kvm_arch_init_irq_routing(s); | |
989 | } | |
990 | ||
991 | void kvm_irqchip_commit_routes(KVMState *s) | |
992 | { | |
993 | int ret; | |
994 | ||
995 | s->irq_routes->flags = 0; | |
996 | ret = kvm_vm_ioctl(s, KVM_SET_GSI_ROUTING, s->irq_routes); | |
997 | assert(ret == 0); | |
998 | } | |
999 | ||
1000 | static void kvm_add_routing_entry(KVMState *s, | |
1001 | struct kvm_irq_routing_entry *entry) | |
1002 | { | |
1003 | struct kvm_irq_routing_entry *new; | |
1004 | int n, size; | |
1005 | ||
1006 | if (s->irq_routes->nr == s->nr_allocated_irq_routes) { | |
1007 | n = s->nr_allocated_irq_routes * 2; | |
1008 | if (n < 64) { | |
1009 | n = 64; | |
1010 | } | |
1011 | size = sizeof(struct kvm_irq_routing); | |
1012 | size += n * sizeof(*new); | |
1013 | s->irq_routes = g_realloc(s->irq_routes, size); | |
1014 | s->nr_allocated_irq_routes = n; | |
1015 | } | |
1016 | n = s->irq_routes->nr++; | |
1017 | new = &s->irq_routes->entries[n]; | |
1018 | ||
1019 | *new = *entry; | |
1020 | ||
1021 | set_gsi(s, entry->gsi); | |
1022 | } | |
1023 | ||
1024 | static int kvm_update_routing_entry(KVMState *s, | |
1025 | struct kvm_irq_routing_entry *new_entry) | |
1026 | { | |
1027 | struct kvm_irq_routing_entry *entry; | |
1028 | int n; | |
1029 | ||
1030 | for (n = 0; n < s->irq_routes->nr; n++) { | |
1031 | entry = &s->irq_routes->entries[n]; | |
1032 | if (entry->gsi != new_entry->gsi) { | |
1033 | continue; | |
1034 | } | |
1035 | ||
1036 | if(!memcmp(entry, new_entry, sizeof *entry)) { | |
1037 | return 0; | |
1038 | } | |
1039 | ||
1040 | *entry = *new_entry; | |
1041 | ||
1042 | kvm_irqchip_commit_routes(s); | |
1043 | ||
1044 | return 0; | |
1045 | } | |
1046 | ||
1047 | return -ESRCH; | |
1048 | } | |
1049 | ||
1050 | void kvm_irqchip_add_irq_route(KVMState *s, int irq, int irqchip, int pin) | |
1051 | { | |
1052 | struct kvm_irq_routing_entry e = {}; | |
1053 | ||
1054 | assert(pin < s->gsi_count); | |
1055 | ||
1056 | e.gsi = irq; | |
1057 | e.type = KVM_IRQ_ROUTING_IRQCHIP; | |
1058 | e.flags = 0; | |
1059 | e.u.irqchip.irqchip = irqchip; | |
1060 | e.u.irqchip.pin = pin; | |
1061 | kvm_add_routing_entry(s, &e); | |
1062 | } | |
1063 | ||
1064 | void kvm_irqchip_release_virq(KVMState *s, int virq) | |
1065 | { | |
1066 | struct kvm_irq_routing_entry *e; | |
1067 | int i; | |
1068 | ||
1069 | if (kvm_gsi_direct_mapping()) { | |
1070 | return; | |
1071 | } | |
1072 | ||
1073 | for (i = 0; i < s->irq_routes->nr; i++) { | |
1074 | e = &s->irq_routes->entries[i]; | |
1075 | if (e->gsi == virq) { | |
1076 | s->irq_routes->nr--; | |
1077 | *e = s->irq_routes->entries[s->irq_routes->nr]; | |
1078 | } | |
1079 | } | |
1080 | clear_gsi(s, virq); | |
1081 | } | |
1082 | ||
1083 | static unsigned int kvm_hash_msi(uint32_t data) | |
1084 | { | |
1085 | /* This is optimized for IA32 MSI layout. However, no other arch shall | |
1086 | * repeat the mistake of not providing a direct MSI injection API. */ | |
1087 | return data & 0xff; | |
1088 | } | |
1089 | ||
1090 | static void kvm_flush_dynamic_msi_routes(KVMState *s) | |
1091 | { | |
1092 | KVMMSIRoute *route, *next; | |
1093 | unsigned int hash; | |
1094 | ||
1095 | for (hash = 0; hash < KVM_MSI_HASHTAB_SIZE; hash++) { | |
1096 | QTAILQ_FOREACH_SAFE(route, &s->msi_hashtab[hash], entry, next) { | |
1097 | kvm_irqchip_release_virq(s, route->kroute.gsi); | |
1098 | QTAILQ_REMOVE(&s->msi_hashtab[hash], route, entry); | |
1099 | g_free(route); | |
1100 | } | |
1101 | } | |
1102 | } | |
1103 | ||
1104 | static int kvm_irqchip_get_virq(KVMState *s) | |
1105 | { | |
1106 | uint32_t *word = s->used_gsi_bitmap; | |
1107 | int max_words = ALIGN(s->gsi_count, 32) / 32; | |
1108 | int i, zeroes; | |
1109 | ||
1110 | /* | |
1111 | * PIC and IOAPIC share the first 16 GSI numbers, thus the available | |
1112 | * GSI numbers are more than the number of IRQ route. Allocating a GSI | |
1113 | * number can succeed even though a new route entry cannot be added. | |
1114 | * When this happens, flush dynamic MSI entries to free IRQ route entries. | |
1115 | */ | |
1116 | if (!s->direct_msi && s->irq_routes->nr == s->gsi_count) { | |
1117 | kvm_flush_dynamic_msi_routes(s); | |
1118 | } | |
1119 | ||
1120 | /* Return the lowest unused GSI in the bitmap */ | |
1121 | for (i = 0; i < max_words; i++) { | |
1122 | zeroes = ctz32(~word[i]); | |
1123 | if (zeroes == 32) { | |
1124 | continue; | |
1125 | } | |
1126 | ||
1127 | return zeroes + i * 32; | |
1128 | } | |
1129 | return -ENOSPC; | |
1130 | ||
1131 | } | |
1132 | ||
1133 | static KVMMSIRoute *kvm_lookup_msi_route(KVMState *s, MSIMessage msg) | |
1134 | { | |
1135 | unsigned int hash = kvm_hash_msi(msg.data); | |
1136 | KVMMSIRoute *route; | |
1137 | ||
1138 | QTAILQ_FOREACH(route, &s->msi_hashtab[hash], entry) { | |
1139 | if (route->kroute.u.msi.address_lo == (uint32_t)msg.address && | |
1140 | route->kroute.u.msi.address_hi == (msg.address >> 32) && | |
1141 | route->kroute.u.msi.data == le32_to_cpu(msg.data)) { | |
1142 | return route; | |
1143 | } | |
1144 | } | |
1145 | return NULL; | |
1146 | } | |
1147 | ||
1148 | int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg) | |
1149 | { | |
1150 | struct kvm_msi msi; | |
1151 | KVMMSIRoute *route; | |
1152 | ||
1153 | if (s->direct_msi) { | |
1154 | msi.address_lo = (uint32_t)msg.address; | |
1155 | msi.address_hi = msg.address >> 32; | |
1156 | msi.data = le32_to_cpu(msg.data); | |
1157 | msi.flags = 0; | |
1158 | memset(msi.pad, 0, sizeof(msi.pad)); | |
1159 | ||
1160 | return kvm_vm_ioctl(s, KVM_SIGNAL_MSI, &msi); | |
1161 | } | |
1162 | ||
1163 | route = kvm_lookup_msi_route(s, msg); | |
1164 | if (!route) { | |
1165 | int virq; | |
1166 | ||
1167 | virq = kvm_irqchip_get_virq(s); | |
1168 | if (virq < 0) { | |
1169 | return virq; | |
1170 | } | |
1171 | ||
1172 | route = g_malloc0(sizeof(KVMMSIRoute)); | |
1173 | route->kroute.gsi = virq; | |
1174 | route->kroute.type = KVM_IRQ_ROUTING_MSI; | |
1175 | route->kroute.flags = 0; | |
1176 | route->kroute.u.msi.address_lo = (uint32_t)msg.address; | |
1177 | route->kroute.u.msi.address_hi = msg.address >> 32; | |
1178 | route->kroute.u.msi.data = le32_to_cpu(msg.data); | |
1179 | ||
1180 | kvm_add_routing_entry(s, &route->kroute); | |
1181 | kvm_irqchip_commit_routes(s); | |
1182 | ||
1183 | QTAILQ_INSERT_TAIL(&s->msi_hashtab[kvm_hash_msi(msg.data)], route, | |
1184 | entry); | |
1185 | } | |
1186 | ||
1187 | assert(route->kroute.type == KVM_IRQ_ROUTING_MSI); | |
1188 | ||
1189 | return kvm_set_irq(s, route->kroute.gsi, 1); | |
1190 | } | |
1191 | ||
1192 | int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg) | |
1193 | { | |
1194 | struct kvm_irq_routing_entry kroute = {}; | |
1195 | int virq; | |
1196 | ||
1197 | if (kvm_gsi_direct_mapping()) { | |
1198 | return kvm_arch_msi_data_to_gsi(msg.data); | |
1199 | } | |
1200 | ||
1201 | if (!kvm_gsi_routing_enabled()) { | |
1202 | return -ENOSYS; | |
1203 | } | |
1204 | ||
1205 | virq = kvm_irqchip_get_virq(s); | |
1206 | if (virq < 0) { | |
1207 | return virq; | |
1208 | } | |
1209 | ||
1210 | kroute.gsi = virq; | |
1211 | kroute.type = KVM_IRQ_ROUTING_MSI; | |
1212 | kroute.flags = 0; | |
1213 | kroute.u.msi.address_lo = (uint32_t)msg.address; | |
1214 | kroute.u.msi.address_hi = msg.address >> 32; | |
1215 | kroute.u.msi.data = le32_to_cpu(msg.data); | |
1216 | if (kvm_arch_fixup_msi_route(&kroute, msg.address, msg.data)) { | |
1217 | kvm_irqchip_release_virq(s, virq); | |
1218 | return -EINVAL; | |
1219 | } | |
1220 | ||
1221 | kvm_add_routing_entry(s, &kroute); | |
1222 | kvm_irqchip_commit_routes(s); | |
1223 | ||
1224 | return virq; | |
1225 | } | |
1226 | ||
1227 | int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg) | |
1228 | { | |
1229 | struct kvm_irq_routing_entry kroute = {}; | |
1230 | ||
1231 | if (kvm_gsi_direct_mapping()) { | |
1232 | return 0; | |
1233 | } | |
1234 | ||
1235 | if (!kvm_irqchip_in_kernel()) { | |
1236 | return -ENOSYS; | |
1237 | } | |
1238 | ||
1239 | kroute.gsi = virq; | |
1240 | kroute.type = KVM_IRQ_ROUTING_MSI; | |
1241 | kroute.flags = 0; | |
1242 | kroute.u.msi.address_lo = (uint32_t)msg.address; | |
1243 | kroute.u.msi.address_hi = msg.address >> 32; | |
1244 | kroute.u.msi.data = le32_to_cpu(msg.data); | |
1245 | if (kvm_arch_fixup_msi_route(&kroute, msg.address, msg.data)) { | |
1246 | return -EINVAL; | |
1247 | } | |
1248 | ||
1249 | return kvm_update_routing_entry(s, &kroute); | |
1250 | } | |
1251 | ||
1252 | static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int rfd, int virq, | |
1253 | bool assign) | |
1254 | { | |
1255 | struct kvm_irqfd irqfd = { | |
1256 | .fd = fd, | |
1257 | .gsi = virq, | |
1258 | .flags = assign ? 0 : KVM_IRQFD_FLAG_DEASSIGN, | |
1259 | }; | |
1260 | ||
1261 | if (rfd != -1) { | |
1262 | irqfd.flags |= KVM_IRQFD_FLAG_RESAMPLE; | |
1263 | irqfd.resamplefd = rfd; | |
1264 | } | |
1265 | ||
1266 | if (!kvm_irqfds_enabled()) { | |
1267 | return -ENOSYS; | |
1268 | } | |
1269 | ||
1270 | return kvm_vm_ioctl(s, KVM_IRQFD, &irqfd); | |
1271 | } | |
1272 | ||
1273 | int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter) | |
1274 | { | |
1275 | struct kvm_irq_routing_entry kroute = {}; | |
1276 | int virq; | |
1277 | ||
1278 | if (!kvm_gsi_routing_enabled()) { | |
1279 | return -ENOSYS; | |
1280 | } | |
1281 | ||
1282 | virq = kvm_irqchip_get_virq(s); | |
1283 | if (virq < 0) { | |
1284 | return virq; | |
1285 | } | |
1286 | ||
1287 | kroute.gsi = virq; | |
1288 | kroute.type = KVM_IRQ_ROUTING_S390_ADAPTER; | |
1289 | kroute.flags = 0; | |
1290 | kroute.u.adapter.summary_addr = adapter->summary_addr; | |
1291 | kroute.u.adapter.ind_addr = adapter->ind_addr; | |
1292 | kroute.u.adapter.summary_offset = adapter->summary_offset; | |
1293 | kroute.u.adapter.ind_offset = adapter->ind_offset; | |
1294 | kroute.u.adapter.adapter_id = adapter->adapter_id; | |
1295 | ||
1296 | kvm_add_routing_entry(s, &kroute); | |
1297 | ||
1298 | return virq; | |
1299 | } | |
1300 | ||
1301 | #else /* !KVM_CAP_IRQ_ROUTING */ | |
1302 | ||
1303 | void kvm_init_irq_routing(KVMState *s) | |
1304 | { | |
1305 | } | |
1306 | ||
1307 | void kvm_irqchip_release_virq(KVMState *s, int virq) | |
1308 | { | |
1309 | } | |
1310 | ||
1311 | int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg) | |
1312 | { | |
1313 | abort(); | |
1314 | } | |
1315 | ||
1316 | int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg) | |
1317 | { | |
1318 | return -ENOSYS; | |
1319 | } | |
1320 | ||
1321 | int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter) | |
1322 | { | |
1323 | return -ENOSYS; | |
1324 | } | |
1325 | ||
1326 | static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int virq, bool assign) | |
1327 | { | |
1328 | abort(); | |
1329 | } | |
1330 | ||
1331 | int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg) | |
1332 | { | |
1333 | return -ENOSYS; | |
1334 | } | |
1335 | #endif /* !KVM_CAP_IRQ_ROUTING */ | |
1336 | ||
1337 | int kvm_irqchip_add_irqfd_notifier_gsi(KVMState *s, EventNotifier *n, | |
1338 | EventNotifier *rn, int virq) | |
1339 | { | |
1340 | return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n), | |
1341 | rn ? event_notifier_get_fd(rn) : -1, virq, true); | |
1342 | } | |
1343 | ||
1344 | int kvm_irqchip_remove_irqfd_notifier_gsi(KVMState *s, EventNotifier *n, | |
1345 | int virq) | |
1346 | { | |
1347 | return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n), -1, virq, | |
1348 | false); | |
1349 | } | |
1350 | ||
1351 | int kvm_irqchip_add_irqfd_notifier(KVMState *s, EventNotifier *n, | |
1352 | EventNotifier *rn, qemu_irq irq) | |
1353 | { | |
1354 | gpointer key, gsi; | |
1355 | gboolean found = g_hash_table_lookup_extended(s->gsimap, irq, &key, &gsi); | |
1356 | ||
1357 | if (!found) { | |
1358 | return -ENXIO; | |
1359 | } | |
1360 | return kvm_irqchip_add_irqfd_notifier_gsi(s, n, rn, GPOINTER_TO_INT(gsi)); | |
1361 | } | |
1362 | ||
1363 | int kvm_irqchip_remove_irqfd_notifier(KVMState *s, EventNotifier *n, | |
1364 | qemu_irq irq) | |
1365 | { | |
1366 | gpointer key, gsi; | |
1367 | gboolean found = g_hash_table_lookup_extended(s->gsimap, irq, &key, &gsi); | |
1368 | ||
1369 | if (!found) { | |
1370 | return -ENXIO; | |
1371 | } | |
1372 | return kvm_irqchip_remove_irqfd_notifier_gsi(s, n, GPOINTER_TO_INT(gsi)); | |
1373 | } | |
1374 | ||
1375 | void kvm_irqchip_set_qemuirq_gsi(KVMState *s, qemu_irq irq, int gsi) | |
1376 | { | |
1377 | g_hash_table_insert(s->gsimap, irq, GINT_TO_POINTER(gsi)); | |
1378 | } | |
1379 | ||
1380 | static void kvm_irqchip_create(MachineState *machine, KVMState *s) | |
1381 | { | |
1382 | int ret; | |
1383 | ||
1384 | if (kvm_check_extension(s, KVM_CAP_IRQCHIP)) { | |
1385 | ; | |
1386 | } else if (kvm_check_extension(s, KVM_CAP_S390_IRQCHIP)) { | |
1387 | ret = kvm_vm_enable_cap(s, KVM_CAP_S390_IRQCHIP, 0); | |
1388 | if (ret < 0) { | |
1389 | fprintf(stderr, "Enable kernel irqchip failed: %s\n", strerror(-ret)); | |
1390 | exit(1); | |
1391 | } | |
1392 | } else { | |
1393 | return; | |
1394 | } | |
1395 | ||
1396 | /* First probe and see if there's a arch-specific hook to create the | |
1397 | * in-kernel irqchip for us */ | |
1398 | ret = kvm_arch_irqchip_create(s); | |
1399 | if (ret == 0) { | |
1400 | ret = kvm_vm_ioctl(s, KVM_CREATE_IRQCHIP); | |
1401 | } | |
1402 | if (ret < 0) { | |
1403 | fprintf(stderr, "Create kernel irqchip failed: %s\n", strerror(-ret)); | |
1404 | exit(1); | |
1405 | } | |
1406 | ||
1407 | kvm_kernel_irqchip = true; | |
1408 | /* If we have an in-kernel IRQ chip then we must have asynchronous | |
1409 | * interrupt delivery (though the reverse is not necessarily true) | |
1410 | */ | |
1411 | kvm_async_interrupts_allowed = true; | |
1412 | kvm_halt_in_kernel_allowed = true; | |
1413 | ||
1414 | kvm_init_irq_routing(s); | |
1415 | ||
1416 | s->gsimap = g_hash_table_new(g_direct_hash, g_direct_equal); | |
1417 | } | |
1418 | ||
1419 | /* Find number of supported CPUs using the recommended | |
1420 | * procedure from the kernel API documentation to cope with | |
1421 | * older kernels that may be missing capabilities. | |
1422 | */ | |
1423 | static int kvm_recommended_vcpus(KVMState *s) | |
1424 | { | |
1425 | int ret = kvm_check_extension(s, KVM_CAP_NR_VCPUS); | |
1426 | return (ret) ? ret : 4; | |
1427 | } | |
1428 | ||
1429 | static int kvm_max_vcpus(KVMState *s) | |
1430 | { | |
1431 | int ret = kvm_check_extension(s, KVM_CAP_MAX_VCPUS); | |
1432 | return (ret) ? ret : kvm_recommended_vcpus(s); | |
1433 | } | |
1434 | ||
1435 | static int kvm_init(MachineState *ms) | |
1436 | { | |
1437 | MachineClass *mc = MACHINE_GET_CLASS(ms); | |
1438 | static const char upgrade_note[] = | |
1439 | "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n" | |
1440 | "(see http://sourceforge.net/projects/kvm).\n"; | |
1441 | struct { | |
1442 | const char *name; | |
1443 | int num; | |
1444 | } num_cpus[] = { | |
1445 | { "SMP", smp_cpus }, | |
1446 | { "hotpluggable", max_cpus }, | |
1447 | { NULL, } | |
1448 | }, *nc = num_cpus; | |
1449 | int soft_vcpus_limit, hard_vcpus_limit; | |
1450 | KVMState *s; | |
1451 | const KVMCapabilityInfo *missing_cap; | |
1452 | int ret; | |
1453 | int type = 0; | |
1454 | const char *kvm_type; | |
1455 | ||
1456 | s = KVM_STATE(ms->accelerator); | |
1457 | ||
1458 | /* | |
1459 | * On systems where the kernel can support different base page | |
1460 | * sizes, host page size may be different from TARGET_PAGE_SIZE, | |
1461 | * even with KVM. TARGET_PAGE_SIZE is assumed to be the minimum | |
1462 | * page size for the system though. | |
1463 | */ | |
1464 | assert(TARGET_PAGE_SIZE <= getpagesize()); | |
1465 | page_size_init(); | |
1466 | ||
1467 | s->sigmask_len = 8; | |
1468 | ||
1469 | #ifdef KVM_CAP_SET_GUEST_DEBUG | |
1470 | QTAILQ_INIT(&s->kvm_sw_breakpoints); | |
1471 | #endif | |
1472 | s->vmfd = -1; | |
1473 | s->fd = qemu_open("/dev/kvm", O_RDWR); | |
1474 | if (s->fd == -1) { | |
1475 | fprintf(stderr, "Could not access KVM kernel module: %m\n"); | |
1476 | ret = -errno; | |
1477 | goto err; | |
1478 | } | |
1479 | ||
1480 | ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0); | |
1481 | if (ret < KVM_API_VERSION) { | |
1482 | if (ret >= 0) { | |
1483 | ret = -EINVAL; | |
1484 | } | |
1485 | fprintf(stderr, "kvm version too old\n"); | |
1486 | goto err; | |
1487 | } | |
1488 | ||
1489 | if (ret > KVM_API_VERSION) { | |
1490 | ret = -EINVAL; | |
1491 | fprintf(stderr, "kvm version not supported\n"); | |
1492 | goto err; | |
1493 | } | |
1494 | ||
1495 | s->nr_slots = kvm_check_extension(s, KVM_CAP_NR_MEMSLOTS); | |
1496 | ||
1497 | /* If unspecified, use the default value */ | |
1498 | if (!s->nr_slots) { | |
1499 | s->nr_slots = 32; | |
1500 | } | |
1501 | ||
1502 | /* check the vcpu limits */ | |
1503 | soft_vcpus_limit = kvm_recommended_vcpus(s); | |
1504 | hard_vcpus_limit = kvm_max_vcpus(s); | |
1505 | ||
1506 | while (nc->name) { | |
1507 | if (nc->num > soft_vcpus_limit) { | |
1508 | fprintf(stderr, | |
1509 | "Warning: Number of %s cpus requested (%d) exceeds " | |
1510 | "the recommended cpus supported by KVM (%d)\n", | |
1511 | nc->name, nc->num, soft_vcpus_limit); | |
1512 | ||
1513 | if (nc->num > hard_vcpus_limit) { | |
1514 | fprintf(stderr, "Number of %s cpus requested (%d) exceeds " | |
1515 | "the maximum cpus supported by KVM (%d)\n", | |
1516 | nc->name, nc->num, hard_vcpus_limit); | |
1517 | exit(1); | |
1518 | } | |
1519 | } | |
1520 | nc++; | |
1521 | } | |
1522 | ||
1523 | kvm_type = qemu_opt_get(qemu_get_machine_opts(), "kvm-type"); | |
1524 | if (mc->kvm_type) { | |
1525 | type = mc->kvm_type(kvm_type); | |
1526 | } else if (kvm_type) { | |
1527 | ret = -EINVAL; | |
1528 | fprintf(stderr, "Invalid argument kvm-type=%s\n", kvm_type); | |
1529 | goto err; | |
1530 | } | |
1531 | ||
1532 | do { | |
1533 | ret = kvm_ioctl(s, KVM_CREATE_VM, type); | |
1534 | } while (ret == -EINTR); | |
1535 | ||
1536 | if (ret < 0) { | |
1537 | fprintf(stderr, "ioctl(KVM_CREATE_VM) failed: %d %s\n", -ret, | |
1538 | strerror(-ret)); | |
1539 | ||
1540 | #ifdef TARGET_S390X | |
1541 | if (ret == -EINVAL) { | |
1542 | fprintf(stderr, | |
1543 | "Host kernel setup problem detected. Please verify:\n"); | |
1544 | fprintf(stderr, "- for kernels supporting the switch_amode or" | |
1545 | " user_mode parameters, whether\n"); | |
1546 | fprintf(stderr, | |
1547 | " user space is running in primary address space\n"); | |
1548 | fprintf(stderr, | |
1549 | "- for kernels supporting the vm.allocate_pgste sysctl, " | |
1550 | "whether it is enabled\n"); | |
1551 | } | |
1552 | #endif | |
1553 | goto err; | |
1554 | } | |
1555 | ||
1556 | s->vmfd = ret; | |
1557 | missing_cap = kvm_check_extension_list(s, kvm_required_capabilites); | |
1558 | if (!missing_cap) { | |
1559 | missing_cap = | |
1560 | kvm_check_extension_list(s, kvm_arch_required_capabilities); | |
1561 | } | |
1562 | if (missing_cap) { | |
1563 | ret = -EINVAL; | |
1564 | fprintf(stderr, "kvm does not support %s\n%s", | |
1565 | missing_cap->name, upgrade_note); | |
1566 | goto err; | |
1567 | } | |
1568 | ||
1569 | s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO); | |
1570 | ||
1571 | s->broken_set_mem_region = 1; | |
1572 | ret = kvm_check_extension(s, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS); | |
1573 | if (ret > 0) { | |
1574 | s->broken_set_mem_region = 0; | |
1575 | } | |
1576 | ||
1577 | #ifdef KVM_CAP_VCPU_EVENTS | |
1578 | s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS); | |
1579 | #endif | |
1580 | ||
1581 | s->robust_singlestep = | |
1582 | kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP); | |
1583 | ||
1584 | #ifdef KVM_CAP_DEBUGREGS | |
1585 | s->debugregs = kvm_check_extension(s, KVM_CAP_DEBUGREGS); | |
1586 | #endif | |
1587 | ||
1588 | #ifdef KVM_CAP_XSAVE | |
1589 | s->xsave = kvm_check_extension(s, KVM_CAP_XSAVE); | |
1590 | #endif | |
1591 | ||
1592 | #ifdef KVM_CAP_XCRS | |
1593 | s->xcrs = kvm_check_extension(s, KVM_CAP_XCRS); | |
1594 | #endif | |
1595 | ||
1596 | #ifdef KVM_CAP_PIT_STATE2 | |
1597 | s->pit_state2 = kvm_check_extension(s, KVM_CAP_PIT_STATE2); | |
1598 | #endif | |
1599 | ||
1600 | #ifdef KVM_CAP_IRQ_ROUTING | |
1601 | s->direct_msi = (kvm_check_extension(s, KVM_CAP_SIGNAL_MSI) > 0); | |
1602 | #endif | |
1603 | ||
1604 | s->intx_set_mask = kvm_check_extension(s, KVM_CAP_PCI_2_3); | |
1605 | ||
1606 | s->irq_set_ioctl = KVM_IRQ_LINE; | |
1607 | if (kvm_check_extension(s, KVM_CAP_IRQ_INJECT_STATUS)) { | |
1608 | s->irq_set_ioctl = KVM_IRQ_LINE_STATUS; | |
1609 | } | |
1610 | ||
1611 | #ifdef KVM_CAP_READONLY_MEM | |
1612 | kvm_readonly_mem_allowed = | |
1613 | (kvm_check_extension(s, KVM_CAP_READONLY_MEM) > 0); | |
1614 | #endif | |
1615 | ||
1616 | kvm_eventfds_allowed = | |
1617 | (kvm_check_extension(s, KVM_CAP_IOEVENTFD) > 0); | |
1618 | ||
1619 | kvm_irqfds_allowed = | |
1620 | (kvm_check_extension(s, KVM_CAP_IRQFD) > 0); | |
1621 | ||
1622 | kvm_resamplefds_allowed = | |
1623 | (kvm_check_extension(s, KVM_CAP_IRQFD_RESAMPLE) > 0); | |
1624 | ||
1625 | kvm_vm_attributes_allowed = | |
1626 | (kvm_check_extension(s, KVM_CAP_VM_ATTRIBUTES) > 0); | |
1627 | ||
1628 | ret = kvm_arch_init(ms, s); | |
1629 | if (ret < 0) { | |
1630 | goto err; | |
1631 | } | |
1632 | ||
1633 | if (machine_kernel_irqchip_allowed(ms)) { | |
1634 | kvm_irqchip_create(ms, s); | |
1635 | } | |
1636 | ||
1637 | kvm_state = s; | |
1638 | ||
1639 | s->memory_listener.listener.eventfd_add = kvm_mem_ioeventfd_add; | |
1640 | s->memory_listener.listener.eventfd_del = kvm_mem_ioeventfd_del; | |
1641 | s->memory_listener.listener.coalesced_mmio_add = kvm_coalesce_mmio_region; | |
1642 | s->memory_listener.listener.coalesced_mmio_del = kvm_uncoalesce_mmio_region; | |
1643 | ||
1644 | kvm_memory_listener_register(s, &s->memory_listener, | |
1645 | &address_space_memory, 0); | |
1646 | memory_listener_register(&kvm_io_listener, | |
1647 | &address_space_io); | |
1648 | ||
1649 | s->many_ioeventfds = kvm_check_many_ioeventfds(); | |
1650 | ||
1651 | cpu_interrupt_handler = kvm_handle_interrupt; | |
1652 | ||
1653 | return 0; | |
1654 | ||
1655 | err: | |
1656 | assert(ret < 0); | |
1657 | if (s->vmfd >= 0) { | |
1658 | close(s->vmfd); | |
1659 | } | |
1660 | if (s->fd != -1) { | |
1661 | close(s->fd); | |
1662 | } | |
1663 | g_free(s->memory_listener.slots); | |
1664 | ||
1665 | return ret; | |
1666 | } | |
1667 | ||
1668 | void kvm_set_sigmask_len(KVMState *s, unsigned int sigmask_len) | |
1669 | { | |
1670 | s->sigmask_len = sigmask_len; | |
1671 | } | |
1672 | ||
1673 | static void kvm_handle_io(uint16_t port, MemTxAttrs attrs, void *data, int direction, | |
1674 | int size, uint32_t count) | |
1675 | { | |
1676 | int i; | |
1677 | uint8_t *ptr = data; | |
1678 | ||
1679 | for (i = 0; i < count; i++) { | |
1680 | address_space_rw(&address_space_io, port, attrs, | |
1681 | ptr, size, | |
1682 | direction == KVM_EXIT_IO_OUT); | |
1683 | ptr += size; | |
1684 | } | |
1685 | } | |
1686 | ||
1687 | static int kvm_handle_internal_error(CPUState *cpu, struct kvm_run *run) | |
1688 | { | |
1689 | fprintf(stderr, "KVM internal error. Suberror: %d\n", | |
1690 | run->internal.suberror); | |
1691 | ||
1692 | if (kvm_check_extension(kvm_state, KVM_CAP_INTERNAL_ERROR_DATA)) { | |
1693 | int i; | |
1694 | ||
1695 | for (i = 0; i < run->internal.ndata; ++i) { | |
1696 | fprintf(stderr, "extra data[%d]: %"PRIx64"\n", | |
1697 | i, (uint64_t)run->internal.data[i]); | |
1698 | } | |
1699 | } | |
1700 | if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) { | |
1701 | fprintf(stderr, "emulation failure\n"); | |
1702 | if (!kvm_arch_stop_on_emulation_error(cpu)) { | |
1703 | cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE); | |
1704 | return EXCP_INTERRUPT; | |
1705 | } | |
1706 | } | |
1707 | /* FIXME: Should trigger a qmp message to let management know | |
1708 | * something went wrong. | |
1709 | */ | |
1710 | return -1; | |
1711 | } | |
1712 | ||
1713 | void kvm_flush_coalesced_mmio_buffer(void) | |
1714 | { | |
1715 | KVMState *s = kvm_state; | |
1716 | ||
1717 | if (s->coalesced_flush_in_progress) { | |
1718 | return; | |
1719 | } | |
1720 | ||
1721 | s->coalesced_flush_in_progress = true; | |
1722 | ||
1723 | if (s->coalesced_mmio_ring) { | |
1724 | struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring; | |
1725 | while (ring->first != ring->last) { | |
1726 | struct kvm_coalesced_mmio *ent; | |
1727 | ||
1728 | ent = &ring->coalesced_mmio[ring->first]; | |
1729 | ||
1730 | cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len); | |
1731 | smp_wmb(); | |
1732 | ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX; | |
1733 | } | |
1734 | } | |
1735 | ||
1736 | s->coalesced_flush_in_progress = false; | |
1737 | } | |
1738 | ||
1739 | static void do_kvm_cpu_synchronize_state(void *arg) | |
1740 | { | |
1741 | CPUState *cpu = arg; | |
1742 | ||
1743 | if (!cpu->kvm_vcpu_dirty) { | |
1744 | kvm_arch_get_registers(cpu); | |
1745 | cpu->kvm_vcpu_dirty = true; | |
1746 | } | |
1747 | } | |
1748 | ||
1749 | void kvm_cpu_synchronize_state(CPUState *cpu) | |
1750 | { | |
1751 | if (!cpu->kvm_vcpu_dirty) { | |
1752 | run_on_cpu(cpu, do_kvm_cpu_synchronize_state, cpu); | |
1753 | } | |
1754 | } | |
1755 | ||
1756 | static void do_kvm_cpu_synchronize_post_reset(void *arg) | |
1757 | { | |
1758 | CPUState *cpu = arg; | |
1759 | ||
1760 | kvm_arch_put_registers(cpu, KVM_PUT_RESET_STATE); | |
1761 | cpu->kvm_vcpu_dirty = false; | |
1762 | } | |
1763 | ||
1764 | void kvm_cpu_synchronize_post_reset(CPUState *cpu) | |
1765 | { | |
1766 | run_on_cpu(cpu, do_kvm_cpu_synchronize_post_reset, cpu); | |
1767 | } | |
1768 | ||
1769 | static void do_kvm_cpu_synchronize_post_init(void *arg) | |
1770 | { | |
1771 | CPUState *cpu = arg; | |
1772 | ||
1773 | kvm_arch_put_registers(cpu, KVM_PUT_FULL_STATE); | |
1774 | cpu->kvm_vcpu_dirty = false; | |
1775 | } | |
1776 | ||
1777 | void kvm_cpu_synchronize_post_init(CPUState *cpu) | |
1778 | { | |
1779 | run_on_cpu(cpu, do_kvm_cpu_synchronize_post_init, cpu); | |
1780 | } | |
1781 | ||
1782 | void kvm_cpu_clean_state(CPUState *cpu) | |
1783 | { | |
1784 | cpu->kvm_vcpu_dirty = false; | |
1785 | } | |
1786 | ||
1787 | int kvm_cpu_exec(CPUState *cpu) | |
1788 | { | |
1789 | struct kvm_run *run = cpu->kvm_run; | |
1790 | int ret, run_ret; | |
1791 | ||
1792 | DPRINTF("kvm_cpu_exec()\n"); | |
1793 | ||
1794 | if (kvm_arch_process_async_events(cpu)) { | |
1795 | cpu->exit_request = 0; | |
1796 | return EXCP_HLT; | |
1797 | } | |
1798 | ||
1799 | qemu_mutex_unlock_iothread(); | |
1800 | ||
1801 | do { | |
1802 | MemTxAttrs attrs; | |
1803 | ||
1804 | if (cpu->kvm_vcpu_dirty) { | |
1805 | kvm_arch_put_registers(cpu, KVM_PUT_RUNTIME_STATE); | |
1806 | cpu->kvm_vcpu_dirty = false; | |
1807 | } | |
1808 | ||
1809 | kvm_arch_pre_run(cpu, run); | |
1810 | if (cpu->exit_request) { | |
1811 | DPRINTF("interrupt exit requested\n"); | |
1812 | /* | |
1813 | * KVM requires us to reenter the kernel after IO exits to complete | |
1814 | * instruction emulation. This self-signal will ensure that we | |
1815 | * leave ASAP again. | |
1816 | */ | |
1817 | qemu_cpu_kick_self(); | |
1818 | } | |
1819 | ||
1820 | run_ret = kvm_vcpu_ioctl(cpu, KVM_RUN, 0); | |
1821 | ||
1822 | attrs = kvm_arch_post_run(cpu, run); | |
1823 | ||
1824 | if (run_ret < 0) { | |
1825 | if (run_ret == -EINTR || run_ret == -EAGAIN) { | |
1826 | DPRINTF("io window exit\n"); | |
1827 | ret = EXCP_INTERRUPT; | |
1828 | break; | |
1829 | } | |
1830 | fprintf(stderr, "error: kvm run failed %s\n", | |
1831 | strerror(-run_ret)); | |
1832 | #ifdef TARGET_PPC | |
1833 | if (run_ret == -EBUSY) { | |
1834 | fprintf(stderr, | |
1835 | "This is probably because your SMT is enabled.\n" | |
1836 | "VCPU can only run on primary threads with all " | |
1837 | "secondary threads offline.\n"); | |
1838 | } | |
1839 | #endif | |
1840 | ret = -1; | |
1841 | break; | |
1842 | } | |
1843 | ||
1844 | trace_kvm_run_exit(cpu->cpu_index, run->exit_reason); | |
1845 | switch (run->exit_reason) { | |
1846 | case KVM_EXIT_IO: | |
1847 | DPRINTF("handle_io\n"); | |
1848 | /* Called outside BQL */ | |
1849 | kvm_handle_io(run->io.port, attrs, | |
1850 | (uint8_t *)run + run->io.data_offset, | |
1851 | run->io.direction, | |
1852 | run->io.size, | |
1853 | run->io.count); | |
1854 | ret = 0; | |
1855 | break; | |
1856 | case KVM_EXIT_MMIO: | |
1857 | DPRINTF("handle_mmio\n"); | |
1858 | /* Called outside BQL */ | |
1859 | address_space_rw(&address_space_memory, | |
1860 | run->mmio.phys_addr, attrs, | |
1861 | run->mmio.data, | |
1862 | run->mmio.len, | |
1863 | run->mmio.is_write); | |
1864 | ret = 0; | |
1865 | break; | |
1866 | case KVM_EXIT_IRQ_WINDOW_OPEN: | |
1867 | DPRINTF("irq_window_open\n"); | |
1868 | ret = EXCP_INTERRUPT; | |
1869 | break; | |
1870 | case KVM_EXIT_SHUTDOWN: | |
1871 | DPRINTF("shutdown\n"); | |
1872 | qemu_system_reset_request(); | |
1873 | ret = EXCP_INTERRUPT; | |
1874 | break; | |
1875 | case KVM_EXIT_UNKNOWN: | |
1876 | fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "\n", | |
1877 | (uint64_t)run->hw.hardware_exit_reason); | |
1878 | ret = -1; | |
1879 | break; | |
1880 | case KVM_EXIT_INTERNAL_ERROR: | |
1881 | ret = kvm_handle_internal_error(cpu, run); | |
1882 | break; | |
1883 | case KVM_EXIT_SYSTEM_EVENT: | |
1884 | switch (run->system_event.type) { | |
1885 | case KVM_SYSTEM_EVENT_SHUTDOWN: | |
1886 | qemu_system_shutdown_request(); | |
1887 | ret = EXCP_INTERRUPT; | |
1888 | break; | |
1889 | case KVM_SYSTEM_EVENT_RESET: | |
1890 | qemu_system_reset_request(); | |
1891 | ret = EXCP_INTERRUPT; | |
1892 | break; | |
1893 | case KVM_SYSTEM_EVENT_CRASH: | |
1894 | qemu_mutex_lock_iothread(); | |
1895 | qemu_system_guest_panicked(); | |
1896 | qemu_mutex_unlock_iothread(); | |
1897 | ret = 0; | |
1898 | break; | |
1899 | default: | |
1900 | DPRINTF("kvm_arch_handle_exit\n"); | |
1901 | ret = kvm_arch_handle_exit(cpu, run); | |
1902 | break; | |
1903 | } | |
1904 | break; | |
1905 | default: | |
1906 | DPRINTF("kvm_arch_handle_exit\n"); | |
1907 | ret = kvm_arch_handle_exit(cpu, run); | |
1908 | break; | |
1909 | } | |
1910 | } while (ret == 0); | |
1911 | ||
1912 | qemu_mutex_lock_iothread(); | |
1913 | ||
1914 | if (ret < 0) { | |
1915 | cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE); | |
1916 | vm_stop(RUN_STATE_INTERNAL_ERROR); | |
1917 | } | |
1918 | ||
1919 | cpu->exit_request = 0; | |
1920 | return ret; | |
1921 | } | |
1922 | ||
1923 | int kvm_ioctl(KVMState *s, int type, ...) | |
1924 | { | |
1925 | int ret; | |
1926 | void *arg; | |
1927 | va_list ap; | |
1928 | ||
1929 | va_start(ap, type); | |
1930 | arg = va_arg(ap, void *); | |
1931 | va_end(ap); | |
1932 | ||
1933 | trace_kvm_ioctl(type, arg); | |
1934 | ret = ioctl(s->fd, type, arg); | |
1935 | if (ret == -1) { | |
1936 | ret = -errno; | |
1937 | } | |
1938 | return ret; | |
1939 | } | |
1940 | ||
1941 | int kvm_vm_ioctl(KVMState *s, int type, ...) | |
1942 | { | |
1943 | int ret; | |
1944 | void *arg; | |
1945 | va_list ap; | |
1946 | ||
1947 | va_start(ap, type); | |
1948 | arg = va_arg(ap, void *); | |
1949 | va_end(ap); | |
1950 | ||
1951 | trace_kvm_vm_ioctl(type, arg); | |
1952 | ret = ioctl(s->vmfd, type, arg); | |
1953 | if (ret == -1) { | |
1954 | ret = -errno; | |
1955 | } | |
1956 | return ret; | |
1957 | } | |
1958 | ||
1959 | int kvm_vcpu_ioctl(CPUState *cpu, int type, ...) | |
1960 | { | |
1961 | int ret; | |
1962 | void *arg; | |
1963 | va_list ap; | |
1964 | ||
1965 | va_start(ap, type); | |
1966 | arg = va_arg(ap, void *); | |
1967 | va_end(ap); | |
1968 | ||
1969 | trace_kvm_vcpu_ioctl(cpu->cpu_index, type, arg); | |
1970 | ret = ioctl(cpu->kvm_fd, type, arg); | |
1971 | if (ret == -1) { | |
1972 | ret = -errno; | |
1973 | } | |
1974 | return ret; | |
1975 | } | |
1976 | ||
1977 | int kvm_device_ioctl(int fd, int type, ...) | |
1978 | { | |
1979 | int ret; | |
1980 | void *arg; | |
1981 | va_list ap; | |
1982 | ||
1983 | va_start(ap, type); | |
1984 | arg = va_arg(ap, void *); | |
1985 | va_end(ap); | |
1986 | ||
1987 | trace_kvm_device_ioctl(fd, type, arg); | |
1988 | ret = ioctl(fd, type, arg); | |
1989 | if (ret == -1) { | |
1990 | ret = -errno; | |
1991 | } | |
1992 | return ret; | |
1993 | } | |
1994 | ||
1995 | int kvm_vm_check_attr(KVMState *s, uint32_t group, uint64_t attr) | |
1996 | { | |
1997 | int ret; | |
1998 | struct kvm_device_attr attribute = { | |
1999 | .group = group, | |
2000 | .attr = attr, | |
2001 | }; | |
2002 | ||
2003 | if (!kvm_vm_attributes_allowed) { | |
2004 | return 0; | |
2005 | } | |
2006 | ||
2007 | ret = kvm_vm_ioctl(s, KVM_HAS_DEVICE_ATTR, &attribute); | |
2008 | /* kvm returns 0 on success for HAS_DEVICE_ATTR */ | |
2009 | return ret ? 0 : 1; | |
2010 | } | |
2011 | ||
2012 | int kvm_device_check_attr(int dev_fd, uint32_t group, uint64_t attr) | |
2013 | { | |
2014 | struct kvm_device_attr attribute = { | |
2015 | .group = group, | |
2016 | .attr = attr, | |
2017 | .flags = 0, | |
2018 | }; | |
2019 | ||
2020 | return kvm_device_ioctl(dev_fd, KVM_HAS_DEVICE_ATTR, &attribute) ? 0 : 1; | |
2021 | } | |
2022 | ||
2023 | void kvm_device_access(int fd, int group, uint64_t attr, | |
2024 | void *val, bool write) | |
2025 | { | |
2026 | struct kvm_device_attr kvmattr; | |
2027 | int err; | |
2028 | ||
2029 | kvmattr.flags = 0; | |
2030 | kvmattr.group = group; | |
2031 | kvmattr.attr = attr; | |
2032 | kvmattr.addr = (uintptr_t)val; | |
2033 | ||
2034 | err = kvm_device_ioctl(fd, | |
2035 | write ? KVM_SET_DEVICE_ATTR : KVM_GET_DEVICE_ATTR, | |
2036 | &kvmattr); | |
2037 | if (err < 0) { | |
2038 | error_report("KVM_%s_DEVICE_ATTR failed: %s\n" | |
2039 | "Group %d attr 0x%016" PRIx64, write ? "SET" : "GET", | |
2040 | strerror(-err), group, attr); | |
2041 | abort(); | |
2042 | } | |
2043 | } | |
2044 | ||
2045 | int kvm_has_sync_mmu(void) | |
2046 | { | |
2047 | return kvm_check_extension(kvm_state, KVM_CAP_SYNC_MMU); | |
2048 | } | |
2049 | ||
2050 | int kvm_has_vcpu_events(void) | |
2051 | { | |
2052 | return kvm_state->vcpu_events; | |
2053 | } | |
2054 | ||
2055 | int kvm_has_robust_singlestep(void) | |
2056 | { | |
2057 | return kvm_state->robust_singlestep; | |
2058 | } | |
2059 | ||
2060 | int kvm_has_debugregs(void) | |
2061 | { | |
2062 | return kvm_state->debugregs; | |
2063 | } | |
2064 | ||
2065 | int kvm_has_xsave(void) | |
2066 | { | |
2067 | return kvm_state->xsave; | |
2068 | } | |
2069 | ||
2070 | int kvm_has_xcrs(void) | |
2071 | { | |
2072 | return kvm_state->xcrs; | |
2073 | } | |
2074 | ||
2075 | int kvm_has_pit_state2(void) | |
2076 | { | |
2077 | return kvm_state->pit_state2; | |
2078 | } | |
2079 | ||
2080 | int kvm_has_many_ioeventfds(void) | |
2081 | { | |
2082 | if (!kvm_enabled()) { | |
2083 | return 0; | |
2084 | } | |
2085 | return kvm_state->many_ioeventfds; | |
2086 | } | |
2087 | ||
2088 | int kvm_has_gsi_routing(void) | |
2089 | { | |
2090 | #ifdef KVM_CAP_IRQ_ROUTING | |
2091 | return kvm_check_extension(kvm_state, KVM_CAP_IRQ_ROUTING); | |
2092 | #else | |
2093 | return false; | |
2094 | #endif | |
2095 | } | |
2096 | ||
2097 | int kvm_has_intx_set_mask(void) | |
2098 | { | |
2099 | return kvm_state->intx_set_mask; | |
2100 | } | |
2101 | ||
2102 | void kvm_setup_guest_memory(void *start, size_t size) | |
2103 | { | |
2104 | if (!kvm_has_sync_mmu()) { | |
2105 | int ret = qemu_madvise(start, size, QEMU_MADV_DONTFORK); | |
2106 | ||
2107 | if (ret) { | |
2108 | perror("qemu_madvise"); | |
2109 | fprintf(stderr, | |
2110 | "Need MADV_DONTFORK in absence of synchronous KVM MMU\n"); | |
2111 | exit(1); | |
2112 | } | |
2113 | } | |
2114 | } | |
2115 | ||
2116 | #ifdef KVM_CAP_SET_GUEST_DEBUG | |
2117 | struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *cpu, | |
2118 | target_ulong pc) | |
2119 | { | |
2120 | struct kvm_sw_breakpoint *bp; | |
2121 | ||
2122 | QTAILQ_FOREACH(bp, &cpu->kvm_state->kvm_sw_breakpoints, entry) { | |
2123 | if (bp->pc == pc) { | |
2124 | return bp; | |
2125 | } | |
2126 | } | |
2127 | return NULL; | |
2128 | } | |
2129 | ||
2130 | int kvm_sw_breakpoints_active(CPUState *cpu) | |
2131 | { | |
2132 | return !QTAILQ_EMPTY(&cpu->kvm_state->kvm_sw_breakpoints); | |
2133 | } | |
2134 | ||
2135 | struct kvm_set_guest_debug_data { | |
2136 | struct kvm_guest_debug dbg; | |
2137 | CPUState *cpu; | |
2138 | int err; | |
2139 | }; | |
2140 | ||
2141 | static void kvm_invoke_set_guest_debug(void *data) | |
2142 | { | |
2143 | struct kvm_set_guest_debug_data *dbg_data = data; | |
2144 | ||
2145 | dbg_data->err = kvm_vcpu_ioctl(dbg_data->cpu, KVM_SET_GUEST_DEBUG, | |
2146 | &dbg_data->dbg); | |
2147 | } | |
2148 | ||
2149 | int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap) | |
2150 | { | |
2151 | struct kvm_set_guest_debug_data data; | |
2152 | ||
2153 | data.dbg.control = reinject_trap; | |
2154 | ||
2155 | if (cpu->singlestep_enabled) { | |
2156 | data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP; | |
2157 | } | |
2158 | kvm_arch_update_guest_debug(cpu, &data.dbg); | |
2159 | data.cpu = cpu; | |
2160 | ||
2161 | run_on_cpu(cpu, kvm_invoke_set_guest_debug, &data); | |
2162 | return data.err; | |
2163 | } | |
2164 | ||
2165 | int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr, | |
2166 | target_ulong len, int type) | |
2167 | { | |
2168 | struct kvm_sw_breakpoint *bp; | |
2169 | int err; | |
2170 | ||
2171 | if (type == GDB_BREAKPOINT_SW) { | |
2172 | bp = kvm_find_sw_breakpoint(cpu, addr); | |
2173 | if (bp) { | |
2174 | bp->use_count++; | |
2175 | return 0; | |
2176 | } | |
2177 | ||
2178 | bp = g_malloc(sizeof(struct kvm_sw_breakpoint)); | |
2179 | bp->pc = addr; | |
2180 | bp->use_count = 1; | |
2181 | err = kvm_arch_insert_sw_breakpoint(cpu, bp); | |
2182 | if (err) { | |
2183 | g_free(bp); | |
2184 | return err; | |
2185 | } | |
2186 | ||
2187 | QTAILQ_INSERT_HEAD(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry); | |
2188 | } else { | |
2189 | err = kvm_arch_insert_hw_breakpoint(addr, len, type); | |
2190 | if (err) { | |
2191 | return err; | |
2192 | } | |
2193 | } | |
2194 | ||
2195 | CPU_FOREACH(cpu) { | |
2196 | err = kvm_update_guest_debug(cpu, 0); | |
2197 | if (err) { | |
2198 | return err; | |
2199 | } | |
2200 | } | |
2201 | return 0; | |
2202 | } | |
2203 | ||
2204 | int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr, | |
2205 | target_ulong len, int type) | |
2206 | { | |
2207 | struct kvm_sw_breakpoint *bp; | |
2208 | int err; | |
2209 | ||
2210 | if (type == GDB_BREAKPOINT_SW) { | |
2211 | bp = kvm_find_sw_breakpoint(cpu, addr); | |
2212 | if (!bp) { | |
2213 | return -ENOENT; | |
2214 | } | |
2215 | ||
2216 | if (bp->use_count > 1) { | |
2217 | bp->use_count--; | |
2218 | return 0; | |
2219 | } | |
2220 | ||
2221 | err = kvm_arch_remove_sw_breakpoint(cpu, bp); | |
2222 | if (err) { | |
2223 | return err; | |
2224 | } | |
2225 | ||
2226 | QTAILQ_REMOVE(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry); | |
2227 | g_free(bp); | |
2228 | } else { | |
2229 | err = kvm_arch_remove_hw_breakpoint(addr, len, type); | |
2230 | if (err) { | |
2231 | return err; | |
2232 | } | |
2233 | } | |
2234 | ||
2235 | CPU_FOREACH(cpu) { | |
2236 | err = kvm_update_guest_debug(cpu, 0); | |
2237 | if (err) { | |
2238 | return err; | |
2239 | } | |
2240 | } | |
2241 | return 0; | |
2242 | } | |
2243 | ||
2244 | void kvm_remove_all_breakpoints(CPUState *cpu) | |
2245 | { | |
2246 | struct kvm_sw_breakpoint *bp, *next; | |
2247 | KVMState *s = cpu->kvm_state; | |
2248 | CPUState *tmpcpu; | |
2249 | ||
2250 | QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) { | |
2251 | if (kvm_arch_remove_sw_breakpoint(cpu, bp) != 0) { | |
2252 | /* Try harder to find a CPU that currently sees the breakpoint. */ | |
2253 | CPU_FOREACH(tmpcpu) { | |
2254 | if (kvm_arch_remove_sw_breakpoint(tmpcpu, bp) == 0) { | |
2255 | break; | |
2256 | } | |
2257 | } | |
2258 | } | |
2259 | QTAILQ_REMOVE(&s->kvm_sw_breakpoints, bp, entry); | |
2260 | g_free(bp); | |
2261 | } | |
2262 | kvm_arch_remove_all_hw_breakpoints(); | |
2263 | ||
2264 | CPU_FOREACH(cpu) { | |
2265 | kvm_update_guest_debug(cpu, 0); | |
2266 | } | |
2267 | } | |
2268 | ||
2269 | #else /* !KVM_CAP_SET_GUEST_DEBUG */ | |
2270 | ||
2271 | int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap) | |
2272 | { | |
2273 | return -EINVAL; | |
2274 | } | |
2275 | ||
2276 | int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr, | |
2277 | target_ulong len, int type) | |
2278 | { | |
2279 | return -EINVAL; | |
2280 | } | |
2281 | ||
2282 | int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr, | |
2283 | target_ulong len, int type) | |
2284 | { | |
2285 | return -EINVAL; | |
2286 | } | |
2287 | ||
2288 | void kvm_remove_all_breakpoints(CPUState *cpu) | |
2289 | { | |
2290 | } | |
2291 | #endif /* !KVM_CAP_SET_GUEST_DEBUG */ | |
2292 | ||
2293 | int kvm_set_signal_mask(CPUState *cpu, const sigset_t *sigset) | |
2294 | { | |
2295 | KVMState *s = kvm_state; | |
2296 | struct kvm_signal_mask *sigmask; | |
2297 | int r; | |
2298 | ||
2299 | if (!sigset) { | |
2300 | return kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, NULL); | |
2301 | } | |
2302 | ||
2303 | sigmask = g_malloc(sizeof(*sigmask) + sizeof(*sigset)); | |
2304 | ||
2305 | sigmask->len = s->sigmask_len; | |
2306 | memcpy(sigmask->sigset, sigset, sizeof(*sigset)); | |
2307 | r = kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, sigmask); | |
2308 | g_free(sigmask); | |
2309 | ||
2310 | return r; | |
2311 | } | |
2312 | int kvm_on_sigbus_vcpu(CPUState *cpu, int code, void *addr) | |
2313 | { | |
2314 | return kvm_arch_on_sigbus_vcpu(cpu, code, addr); | |
2315 | } | |
2316 | ||
2317 | int kvm_on_sigbus(int code, void *addr) | |
2318 | { | |
2319 | return kvm_arch_on_sigbus(code, addr); | |
2320 | } | |
2321 | ||
2322 | int kvm_create_device(KVMState *s, uint64_t type, bool test) | |
2323 | { | |
2324 | int ret; | |
2325 | struct kvm_create_device create_dev; | |
2326 | ||
2327 | create_dev.type = type; | |
2328 | create_dev.fd = -1; | |
2329 | create_dev.flags = test ? KVM_CREATE_DEVICE_TEST : 0; | |
2330 | ||
2331 | if (!kvm_check_extension(s, KVM_CAP_DEVICE_CTRL)) { | |
2332 | return -ENOTSUP; | |
2333 | } | |
2334 | ||
2335 | ret = kvm_vm_ioctl(s, KVM_CREATE_DEVICE, &create_dev); | |
2336 | if (ret) { | |
2337 | return ret; | |
2338 | } | |
2339 | ||
2340 | return test ? 0 : create_dev.fd; | |
2341 | } | |
2342 | ||
2343 | int kvm_set_one_reg(CPUState *cs, uint64_t id, void *source) | |
2344 | { | |
2345 | struct kvm_one_reg reg; | |
2346 | int r; | |
2347 | ||
2348 | reg.id = id; | |
2349 | reg.addr = (uintptr_t) source; | |
2350 | r = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®); | |
2351 | if (r) { | |
2352 | trace_kvm_failed_reg_set(id, strerror(r)); | |
2353 | } | |
2354 | return r; | |
2355 | } | |
2356 | ||
2357 | int kvm_get_one_reg(CPUState *cs, uint64_t id, void *target) | |
2358 | { | |
2359 | struct kvm_one_reg reg; | |
2360 | int r; | |
2361 | ||
2362 | reg.id = id; | |
2363 | reg.addr = (uintptr_t) target; | |
2364 | r = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®); | |
2365 | if (r) { | |
2366 | trace_kvm_failed_reg_get(id, strerror(r)); | |
2367 | } | |
2368 | return r; | |
2369 | } | |
2370 | ||
2371 | static void kvm_accel_class_init(ObjectClass *oc, void *data) | |
2372 | { | |
2373 | AccelClass *ac = ACCEL_CLASS(oc); | |
2374 | ac->name = "KVM"; | |
2375 | ac->init_machine = kvm_init; | |
2376 | ac->allowed = &kvm_allowed; | |
2377 | } | |
2378 | ||
2379 | static const TypeInfo kvm_accel_type = { | |
2380 | .name = TYPE_KVM_ACCEL, | |
2381 | .parent = TYPE_ACCEL, | |
2382 | .class_init = kvm_accel_class_init, | |
2383 | .instance_size = sizeof(KVMState), | |
2384 | }; | |
2385 | ||
2386 | static void kvm_type_init(void) | |
2387 | { | |
2388 | type_register_static(&kvm_accel_type); | |
2389 | } | |
2390 | ||
2391 | type_init(kvm_type_init); |