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