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