]>
Commit | Line | Data |
---|---|---|
494b00c7 CD |
1 | /* |
2 | * ARM implementation of KVM hooks | |
3 | * | |
4 | * Copyright Christoffer Dall 2009-2010 | |
5 | * | |
6 | * This work is licensed under the terms of the GNU GPL, version 2 or later. | |
7 | * See the COPYING file in the top-level directory. | |
8 | * | |
9 | */ | |
10 | ||
74c21bd0 | 11 | #include "qemu/osdep.h" |
494b00c7 | 12 | #include <sys/ioctl.h> |
494b00c7 CD |
13 | |
14 | #include <linux/kvm.h> | |
15 | ||
16 | #include "qemu-common.h" | |
17 | #include "qemu/timer.h" | |
2ecb2027 | 18 | #include "qemu/error-report.h" |
db725815 | 19 | #include "qemu/main-loop.h" |
494b00c7 CD |
20 | #include "sysemu/sysemu.h" |
21 | #include "sysemu/kvm.h" | |
a27382e2 | 22 | #include "sysemu/kvm_int.h" |
eb035b48 | 23 | #include "kvm_arm.h" |
494b00c7 | 24 | #include "cpu.h" |
b05c81d2 | 25 | #include "trace.h" |
38df27c8 | 26 | #include "internals.h" |
b05c81d2 | 27 | #include "hw/pci/pci.h" |
4c663752 | 28 | #include "exec/memattrs.h" |
4344af65 | 29 | #include "exec/address-spaces.h" |
15eafc2e | 30 | #include "hw/boards.h" |
64552b6b | 31 | #include "hw/irq.h" |
03dd024f | 32 | #include "qemu/log.h" |
494b00c7 CD |
33 | |
34 | const KVMCapabilityInfo kvm_arch_required_capabilities[] = { | |
35 | KVM_CAP_LAST_INFO | |
36 | }; | |
37 | ||
1a1753f7 | 38 | static bool cap_has_mp_state; |
202ccb6b | 39 | static bool cap_has_inject_serror_esr; |
1a1753f7 | 40 | |
c4487d76 PM |
41 | static ARMHostCPUFeatures arm_host_cpu_features; |
42 | ||
228d5e04 PS |
43 | int kvm_arm_vcpu_init(CPUState *cs) |
44 | { | |
45 | ARMCPU *cpu = ARM_CPU(cs); | |
46 | struct kvm_vcpu_init init; | |
47 | ||
48 | init.target = cpu->kvm_target; | |
49 | memcpy(init.features, cpu->kvm_init_features, sizeof(init.features)); | |
50 | ||
51 | return kvm_vcpu_ioctl(cs, KVM_ARM_VCPU_INIT, &init); | |
52 | } | |
53 | ||
202ccb6b DG |
54 | void kvm_arm_init_serror_injection(CPUState *cs) |
55 | { | |
56 | cap_has_inject_serror_esr = kvm_check_extension(cs->kvm_state, | |
57 | KVM_CAP_ARM_INJECT_SERROR_ESR); | |
58 | } | |
59 | ||
a96c0514 PM |
60 | bool kvm_arm_create_scratch_host_vcpu(const uint32_t *cpus_to_try, |
61 | int *fdarray, | |
62 | struct kvm_vcpu_init *init) | |
63 | { | |
64 | int ret, kvmfd = -1, vmfd = -1, cpufd = -1; | |
65 | ||
66 | kvmfd = qemu_open("/dev/kvm", O_RDWR); | |
67 | if (kvmfd < 0) { | |
68 | goto err; | |
69 | } | |
70 | vmfd = ioctl(kvmfd, KVM_CREATE_VM, 0); | |
71 | if (vmfd < 0) { | |
72 | goto err; | |
73 | } | |
74 | cpufd = ioctl(vmfd, KVM_CREATE_VCPU, 0); | |
75 | if (cpufd < 0) { | |
76 | goto err; | |
77 | } | |
78 | ||
2f340e9c PX |
79 | if (!init) { |
80 | /* Caller doesn't want the VCPU to be initialized, so skip it */ | |
81 | goto finish; | |
82 | } | |
83 | ||
a96c0514 PM |
84 | ret = ioctl(vmfd, KVM_ARM_PREFERRED_TARGET, init); |
85 | if (ret >= 0) { | |
86 | ret = ioctl(cpufd, KVM_ARM_VCPU_INIT, init); | |
87 | if (ret < 0) { | |
88 | goto err; | |
89 | } | |
2f340e9c | 90 | } else if (cpus_to_try) { |
a96c0514 PM |
91 | /* Old kernel which doesn't know about the |
92 | * PREFERRED_TARGET ioctl: we know it will only support | |
93 | * creating one kind of guest CPU which is its preferred | |
94 | * CPU type. | |
95 | */ | |
96 | while (*cpus_to_try != QEMU_KVM_ARM_TARGET_NONE) { | |
97 | init->target = *cpus_to_try++; | |
98 | memset(init->features, 0, sizeof(init->features)); | |
99 | ret = ioctl(cpufd, KVM_ARM_VCPU_INIT, init); | |
100 | if (ret >= 0) { | |
101 | break; | |
102 | } | |
103 | } | |
104 | if (ret < 0) { | |
105 | goto err; | |
106 | } | |
2f340e9c PX |
107 | } else { |
108 | /* Treat a NULL cpus_to_try argument the same as an empty | |
109 | * list, which means we will fail the call since this must | |
110 | * be an old kernel which doesn't support PREFERRED_TARGET. | |
111 | */ | |
112 | goto err; | |
a96c0514 PM |
113 | } |
114 | ||
2f340e9c | 115 | finish: |
a96c0514 PM |
116 | fdarray[0] = kvmfd; |
117 | fdarray[1] = vmfd; | |
118 | fdarray[2] = cpufd; | |
119 | ||
120 | return true; | |
121 | ||
122 | err: | |
123 | if (cpufd >= 0) { | |
124 | close(cpufd); | |
125 | } | |
126 | if (vmfd >= 0) { | |
127 | close(vmfd); | |
128 | } | |
129 | if (kvmfd >= 0) { | |
130 | close(kvmfd); | |
131 | } | |
132 | ||
133 | return false; | |
134 | } | |
135 | ||
136 | void kvm_arm_destroy_scratch_host_vcpu(int *fdarray) | |
137 | { | |
138 | int i; | |
139 | ||
140 | for (i = 2; i >= 0; i--) { | |
141 | close(fdarray[i]); | |
142 | } | |
143 | } | |
144 | ||
c4487d76 | 145 | void kvm_arm_set_cpu_features_from_host(ARMCPU *cpu) |
a96c0514 | 146 | { |
c4487d76 | 147 | CPUARMState *env = &cpu->env; |
a96c0514 | 148 | |
c4487d76 PM |
149 | if (!arm_host_cpu_features.dtb_compatible) { |
150 | if (!kvm_enabled() || | |
151 | !kvm_arm_get_host_cpu_features(&arm_host_cpu_features)) { | |
152 | /* We can't report this error yet, so flag that we need to | |
153 | * in arm_cpu_realizefn(). | |
154 | */ | |
155 | cpu->kvm_target = QEMU_KVM_ARM_TARGET_NONE; | |
156 | cpu->host_cpu_probe_failed = true; | |
157 | return; | |
158 | } | |
a96c0514 | 159 | } |
c4487d76 PM |
160 | |
161 | cpu->kvm_target = arm_host_cpu_features.target; | |
162 | cpu->dtb_compatible = arm_host_cpu_features.dtb_compatible; | |
4674097c | 163 | cpu->isar = arm_host_cpu_features.isar; |
c4487d76 | 164 | env->features = arm_host_cpu_features.features; |
a96c0514 PM |
165 | } |
166 | ||
a27382e2 EA |
167 | int kvm_arm_get_max_vm_ipa_size(MachineState *ms) |
168 | { | |
169 | KVMState *s = KVM_STATE(ms->accelerator); | |
170 | int ret; | |
171 | ||
172 | ret = kvm_check_extension(s, KVM_CAP_ARM_VM_IPA_SIZE); | |
173 | return ret > 0 ? ret : 40; | |
174 | } | |
175 | ||
b16565b3 | 176 | int kvm_arch_init(MachineState *ms, KVMState *s) |
494b00c7 CD |
177 | { |
178 | /* For ARM interrupt delivery is always asynchronous, | |
179 | * whether we are using an in-kernel VGIC or not. | |
180 | */ | |
181 | kvm_async_interrupts_allowed = true; | |
a96c0514 | 182 | |
5d721b78 AG |
183 | /* |
184 | * PSCI wakes up secondary cores, so we always need to | |
185 | * have vCPUs waiting in kernel space | |
186 | */ | |
187 | kvm_halt_in_kernel_allowed = true; | |
188 | ||
1a1753f7 AB |
189 | cap_has_mp_state = kvm_check_extension(s, KVM_CAP_MP_STATE); |
190 | ||
494b00c7 CD |
191 | return 0; |
192 | } | |
193 | ||
194 | unsigned long kvm_arch_vcpu_id(CPUState *cpu) | |
195 | { | |
196 | return cpu->cpu_index; | |
197 | } | |
198 | ||
eb035b48 PM |
199 | /* We track all the KVM devices which need their memory addresses |
200 | * passing to the kernel in a list of these structures. | |
201 | * When board init is complete we run through the list and | |
202 | * tell the kernel the base addresses of the memory regions. | |
203 | * We use a MemoryListener to track mapping and unmapping of | |
204 | * the regions during board creation, so the board models don't | |
205 | * need to do anything special for the KVM case. | |
19d1bd0b EA |
206 | * |
207 | * Sometimes the address must be OR'ed with some other fields | |
208 | * (for example for KVM_VGIC_V3_ADDR_TYPE_REDIST_REGION). | |
209 | * @kda_addr_ormask aims at storing the value of those fields. | |
eb035b48 PM |
210 | */ |
211 | typedef struct KVMDevice { | |
212 | struct kvm_arm_device_addr kda; | |
1da41cc1 | 213 | struct kvm_device_attr kdattr; |
19d1bd0b | 214 | uint64_t kda_addr_ormask; |
eb035b48 PM |
215 | MemoryRegion *mr; |
216 | QSLIST_ENTRY(KVMDevice) entries; | |
1da41cc1 | 217 | int dev_fd; |
eb035b48 PM |
218 | } KVMDevice; |
219 | ||
b58deb34 | 220 | static QSLIST_HEAD(, KVMDevice) kvm_devices_head; |
eb035b48 PM |
221 | |
222 | static void kvm_arm_devlistener_add(MemoryListener *listener, | |
223 | MemoryRegionSection *section) | |
224 | { | |
225 | KVMDevice *kd; | |
226 | ||
227 | QSLIST_FOREACH(kd, &kvm_devices_head, entries) { | |
228 | if (section->mr == kd->mr) { | |
229 | kd->kda.addr = section->offset_within_address_space; | |
230 | } | |
231 | } | |
232 | } | |
233 | ||
234 | static void kvm_arm_devlistener_del(MemoryListener *listener, | |
235 | MemoryRegionSection *section) | |
236 | { | |
237 | KVMDevice *kd; | |
238 | ||
239 | QSLIST_FOREACH(kd, &kvm_devices_head, entries) { | |
240 | if (section->mr == kd->mr) { | |
241 | kd->kda.addr = -1; | |
242 | } | |
243 | } | |
244 | } | |
245 | ||
246 | static MemoryListener devlistener = { | |
247 | .region_add = kvm_arm_devlistener_add, | |
248 | .region_del = kvm_arm_devlistener_del, | |
249 | }; | |
250 | ||
1da41cc1 CD |
251 | static void kvm_arm_set_device_addr(KVMDevice *kd) |
252 | { | |
253 | struct kvm_device_attr *attr = &kd->kdattr; | |
254 | int ret; | |
255 | ||
256 | /* If the device control API is available and we have a device fd on the | |
257 | * KVMDevice struct, let's use the newer API | |
258 | */ | |
259 | if (kd->dev_fd >= 0) { | |
260 | uint64_t addr = kd->kda.addr; | |
19d1bd0b EA |
261 | |
262 | addr |= kd->kda_addr_ormask; | |
1da41cc1 CD |
263 | attr->addr = (uintptr_t)&addr; |
264 | ret = kvm_device_ioctl(kd->dev_fd, KVM_SET_DEVICE_ATTR, attr); | |
265 | } else { | |
266 | ret = kvm_vm_ioctl(kvm_state, KVM_ARM_SET_DEVICE_ADDR, &kd->kda); | |
267 | } | |
268 | ||
269 | if (ret < 0) { | |
270 | fprintf(stderr, "Failed to set device address: %s\n", | |
271 | strerror(-ret)); | |
272 | abort(); | |
273 | } | |
274 | } | |
275 | ||
eb035b48 PM |
276 | static void kvm_arm_machine_init_done(Notifier *notifier, void *data) |
277 | { | |
278 | KVMDevice *kd, *tkd; | |
279 | ||
eb035b48 PM |
280 | QSLIST_FOREACH_SAFE(kd, &kvm_devices_head, entries, tkd) { |
281 | if (kd->kda.addr != -1) { | |
1da41cc1 | 282 | kvm_arm_set_device_addr(kd); |
eb035b48 | 283 | } |
dfde4e6e | 284 | memory_region_unref(kd->mr); |
5ff9aaab | 285 | QSLIST_REMOVE_HEAD(&kvm_devices_head, entries); |
eb035b48 PM |
286 | g_free(kd); |
287 | } | |
0bbe4354 | 288 | memory_listener_unregister(&devlistener); |
eb035b48 PM |
289 | } |
290 | ||
291 | static Notifier notify = { | |
292 | .notify = kvm_arm_machine_init_done, | |
293 | }; | |
294 | ||
1da41cc1 | 295 | void kvm_arm_register_device(MemoryRegion *mr, uint64_t devid, uint64_t group, |
19d1bd0b | 296 | uint64_t attr, int dev_fd, uint64_t addr_ormask) |
eb035b48 PM |
297 | { |
298 | KVMDevice *kd; | |
299 | ||
300 | if (!kvm_irqchip_in_kernel()) { | |
301 | return; | |
302 | } | |
303 | ||
304 | if (QSLIST_EMPTY(&kvm_devices_head)) { | |
4344af65 | 305 | memory_listener_register(&devlistener, &address_space_memory); |
eb035b48 PM |
306 | qemu_add_machine_init_done_notifier(¬ify); |
307 | } | |
308 | kd = g_new0(KVMDevice, 1); | |
309 | kd->mr = mr; | |
310 | kd->kda.id = devid; | |
311 | kd->kda.addr = -1; | |
1da41cc1 CD |
312 | kd->kdattr.flags = 0; |
313 | kd->kdattr.group = group; | |
314 | kd->kdattr.attr = attr; | |
315 | kd->dev_fd = dev_fd; | |
19d1bd0b | 316 | kd->kda_addr_ormask = addr_ormask; |
eb035b48 | 317 | QSLIST_INSERT_HEAD(&kvm_devices_head, kd, entries); |
dfde4e6e | 318 | memory_region_ref(kd->mr); |
eb035b48 PM |
319 | } |
320 | ||
38df27c8 AB |
321 | static int compare_u64(const void *a, const void *b) |
322 | { | |
323 | if (*(uint64_t *)a > *(uint64_t *)b) { | |
324 | return 1; | |
325 | } | |
326 | if (*(uint64_t *)a < *(uint64_t *)b) { | |
327 | return -1; | |
328 | } | |
329 | return 0; | |
330 | } | |
331 | ||
c8a44709 | 332 | /* Initialize the ARMCPU cpreg list according to the kernel's |
38df27c8 AB |
333 | * definition of what CPU registers it knows about (and throw away |
334 | * the previous TCG-created cpreg list). | |
335 | */ | |
336 | int kvm_arm_init_cpreg_list(ARMCPU *cpu) | |
337 | { | |
338 | struct kvm_reg_list rl; | |
339 | struct kvm_reg_list *rlp; | |
340 | int i, ret, arraylen; | |
341 | CPUState *cs = CPU(cpu); | |
342 | ||
343 | rl.n = 0; | |
344 | ret = kvm_vcpu_ioctl(cs, KVM_GET_REG_LIST, &rl); | |
345 | if (ret != -E2BIG) { | |
346 | return ret; | |
347 | } | |
348 | rlp = g_malloc(sizeof(struct kvm_reg_list) + rl.n * sizeof(uint64_t)); | |
349 | rlp->n = rl.n; | |
350 | ret = kvm_vcpu_ioctl(cs, KVM_GET_REG_LIST, rlp); | |
351 | if (ret) { | |
352 | goto out; | |
353 | } | |
354 | /* Sort the list we get back from the kernel, since cpreg_tuples | |
355 | * must be in strictly ascending order. | |
356 | */ | |
357 | qsort(&rlp->reg, rlp->n, sizeof(rlp->reg[0]), compare_u64); | |
358 | ||
359 | for (i = 0, arraylen = 0; i < rlp->n; i++) { | |
360 | if (!kvm_arm_reg_syncs_via_cpreg_list(rlp->reg[i])) { | |
361 | continue; | |
362 | } | |
363 | switch (rlp->reg[i] & KVM_REG_SIZE_MASK) { | |
364 | case KVM_REG_SIZE_U32: | |
365 | case KVM_REG_SIZE_U64: | |
366 | break; | |
367 | default: | |
368 | fprintf(stderr, "Can't handle size of register in kernel list\n"); | |
369 | ret = -EINVAL; | |
370 | goto out; | |
371 | } | |
372 | ||
373 | arraylen++; | |
374 | } | |
375 | ||
376 | cpu->cpreg_indexes = g_renew(uint64_t, cpu->cpreg_indexes, arraylen); | |
377 | cpu->cpreg_values = g_renew(uint64_t, cpu->cpreg_values, arraylen); | |
378 | cpu->cpreg_vmstate_indexes = g_renew(uint64_t, cpu->cpreg_vmstate_indexes, | |
379 | arraylen); | |
380 | cpu->cpreg_vmstate_values = g_renew(uint64_t, cpu->cpreg_vmstate_values, | |
381 | arraylen); | |
382 | cpu->cpreg_array_len = arraylen; | |
383 | cpu->cpreg_vmstate_array_len = arraylen; | |
384 | ||
385 | for (i = 0, arraylen = 0; i < rlp->n; i++) { | |
386 | uint64_t regidx = rlp->reg[i]; | |
387 | if (!kvm_arm_reg_syncs_via_cpreg_list(regidx)) { | |
388 | continue; | |
389 | } | |
390 | cpu->cpreg_indexes[arraylen] = regidx; | |
391 | arraylen++; | |
392 | } | |
393 | assert(cpu->cpreg_array_len == arraylen); | |
394 | ||
395 | if (!write_kvmstate_to_list(cpu)) { | |
396 | /* Shouldn't happen unless kernel is inconsistent about | |
397 | * what registers exist. | |
398 | */ | |
399 | fprintf(stderr, "Initial read of kernel register state failed\n"); | |
400 | ret = -EINVAL; | |
401 | goto out; | |
402 | } | |
403 | ||
404 | out: | |
405 | g_free(rlp); | |
406 | return ret; | |
407 | } | |
408 | ||
ff047453 PM |
409 | bool write_kvmstate_to_list(ARMCPU *cpu) |
410 | { | |
411 | CPUState *cs = CPU(cpu); | |
412 | int i; | |
413 | bool ok = true; | |
414 | ||
415 | for (i = 0; i < cpu->cpreg_array_len; i++) { | |
416 | struct kvm_one_reg r; | |
417 | uint64_t regidx = cpu->cpreg_indexes[i]; | |
418 | uint32_t v32; | |
419 | int ret; | |
420 | ||
421 | r.id = regidx; | |
422 | ||
423 | switch (regidx & KVM_REG_SIZE_MASK) { | |
424 | case KVM_REG_SIZE_U32: | |
425 | r.addr = (uintptr_t)&v32; | |
426 | ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &r); | |
427 | if (!ret) { | |
428 | cpu->cpreg_values[i] = v32; | |
429 | } | |
430 | break; | |
431 | case KVM_REG_SIZE_U64: | |
432 | r.addr = (uintptr_t)(cpu->cpreg_values + i); | |
433 | ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &r); | |
434 | break; | |
435 | default: | |
436 | abort(); | |
437 | } | |
438 | if (ret) { | |
439 | ok = false; | |
440 | } | |
441 | } | |
442 | return ok; | |
443 | } | |
444 | ||
4b7a6bf4 | 445 | bool write_list_to_kvmstate(ARMCPU *cpu, int level) |
ff047453 PM |
446 | { |
447 | CPUState *cs = CPU(cpu); | |
448 | int i; | |
449 | bool ok = true; | |
450 | ||
451 | for (i = 0; i < cpu->cpreg_array_len; i++) { | |
452 | struct kvm_one_reg r; | |
453 | uint64_t regidx = cpu->cpreg_indexes[i]; | |
454 | uint32_t v32; | |
455 | int ret; | |
456 | ||
4b7a6bf4 CD |
457 | if (kvm_arm_cpreg_level(regidx) > level) { |
458 | continue; | |
459 | } | |
460 | ||
ff047453 PM |
461 | r.id = regidx; |
462 | switch (regidx & KVM_REG_SIZE_MASK) { | |
463 | case KVM_REG_SIZE_U32: | |
464 | v32 = cpu->cpreg_values[i]; | |
465 | r.addr = (uintptr_t)&v32; | |
466 | break; | |
467 | case KVM_REG_SIZE_U64: | |
468 | r.addr = (uintptr_t)(cpu->cpreg_values + i); | |
469 | break; | |
470 | default: | |
471 | abort(); | |
472 | } | |
473 | ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &r); | |
474 | if (ret) { | |
475 | /* We might fail for "unknown register" and also for | |
476 | * "you tried to set a register which is constant with | |
477 | * a different value from what it actually contains". | |
478 | */ | |
479 | ok = false; | |
480 | } | |
481 | } | |
482 | return ok; | |
483 | } | |
484 | ||
38df27c8 AB |
485 | void kvm_arm_reset_vcpu(ARMCPU *cpu) |
486 | { | |
25f2895e CD |
487 | int ret; |
488 | ||
38df27c8 AB |
489 | /* Re-init VCPU so that all registers are set to |
490 | * their respective reset values. | |
491 | */ | |
25f2895e CD |
492 | ret = kvm_arm_vcpu_init(CPU(cpu)); |
493 | if (ret < 0) { | |
494 | fprintf(stderr, "kvm_arm_vcpu_init failed: %s\n", strerror(-ret)); | |
495 | abort(); | |
496 | } | |
497 | if (!write_kvmstate_to_list(cpu)) { | |
498 | fprintf(stderr, "write_kvmstate_to_list failed\n"); | |
499 | abort(); | |
500 | } | |
b698e4ee PM |
501 | /* |
502 | * Sync the reset values also into the CPUState. This is necessary | |
503 | * because the next thing we do will be a kvm_arch_put_registers() | |
504 | * which will update the list values from the CPUState before copying | |
505 | * the list values back to KVM. It's OK to ignore failure returns here | |
506 | * for the same reason we do so in kvm_arch_get_registers(). | |
507 | */ | |
508 | write_list_to_cpustate(cpu); | |
38df27c8 AB |
509 | } |
510 | ||
1a1753f7 AB |
511 | /* |
512 | * Update KVM's MP_STATE based on what QEMU thinks it is | |
513 | */ | |
514 | int kvm_arm_sync_mpstate_to_kvm(ARMCPU *cpu) | |
515 | { | |
516 | if (cap_has_mp_state) { | |
517 | struct kvm_mp_state mp_state = { | |
062ba099 AB |
518 | .mp_state = (cpu->power_state == PSCI_OFF) ? |
519 | KVM_MP_STATE_STOPPED : KVM_MP_STATE_RUNNABLE | |
1a1753f7 AB |
520 | }; |
521 | int ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state); | |
522 | if (ret) { | |
523 | fprintf(stderr, "%s: failed to set MP_STATE %d/%s\n", | |
524 | __func__, ret, strerror(-ret)); | |
525 | return -1; | |
526 | } | |
527 | } | |
528 | ||
529 | return 0; | |
530 | } | |
531 | ||
532 | /* | |
533 | * Sync the KVM MP_STATE into QEMU | |
534 | */ | |
535 | int kvm_arm_sync_mpstate_to_qemu(ARMCPU *cpu) | |
536 | { | |
537 | if (cap_has_mp_state) { | |
538 | struct kvm_mp_state mp_state; | |
539 | int ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_MP_STATE, &mp_state); | |
540 | if (ret) { | |
541 | fprintf(stderr, "%s: failed to get MP_STATE %d/%s\n", | |
542 | __func__, ret, strerror(-ret)); | |
543 | abort(); | |
544 | } | |
062ba099 AB |
545 | cpu->power_state = (mp_state.mp_state == KVM_MP_STATE_STOPPED) ? |
546 | PSCI_OFF : PSCI_ON; | |
1a1753f7 AB |
547 | } |
548 | ||
549 | return 0; | |
550 | } | |
551 | ||
202ccb6b DG |
552 | int kvm_put_vcpu_events(ARMCPU *cpu) |
553 | { | |
554 | CPUARMState *env = &cpu->env; | |
555 | struct kvm_vcpu_events events; | |
556 | int ret; | |
557 | ||
558 | if (!kvm_has_vcpu_events()) { | |
559 | return 0; | |
560 | } | |
561 | ||
562 | memset(&events, 0, sizeof(events)); | |
563 | events.exception.serror_pending = env->serror.pending; | |
564 | ||
565 | /* Inject SError to guest with specified syndrome if host kernel | |
566 | * supports it, otherwise inject SError without syndrome. | |
567 | */ | |
568 | if (cap_has_inject_serror_esr) { | |
569 | events.exception.serror_has_esr = env->serror.has_esr; | |
570 | events.exception.serror_esr = env->serror.esr; | |
571 | } | |
572 | ||
573 | ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_VCPU_EVENTS, &events); | |
574 | if (ret) { | |
575 | error_report("failed to put vcpu events"); | |
576 | } | |
577 | ||
578 | return ret; | |
579 | } | |
580 | ||
581 | int kvm_get_vcpu_events(ARMCPU *cpu) | |
582 | { | |
583 | CPUARMState *env = &cpu->env; | |
584 | struct kvm_vcpu_events events; | |
585 | int ret; | |
586 | ||
587 | if (!kvm_has_vcpu_events()) { | |
588 | return 0; | |
589 | } | |
590 | ||
591 | memset(&events, 0, sizeof(events)); | |
592 | ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_VCPU_EVENTS, &events); | |
593 | if (ret) { | |
594 | error_report("failed to get vcpu events"); | |
595 | return ret; | |
596 | } | |
597 | ||
598 | env->serror.pending = events.exception.serror_pending; | |
599 | env->serror.has_esr = events.exception.serror_has_esr; | |
600 | env->serror.esr = events.exception.serror_esr; | |
601 | ||
602 | return 0; | |
603 | } | |
604 | ||
494b00c7 CD |
605 | void kvm_arch_pre_run(CPUState *cs, struct kvm_run *run) |
606 | { | |
607 | } | |
608 | ||
4c663752 | 609 | MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run) |
494b00c7 | 610 | { |
5d721b78 AG |
611 | ARMCPU *cpu; |
612 | uint32_t switched_level; | |
613 | ||
614 | if (kvm_irqchip_in_kernel()) { | |
615 | /* | |
616 | * We only need to sync timer states with user-space interrupt | |
617 | * controllers, so return early and save cycles if we don't. | |
618 | */ | |
619 | return MEMTXATTRS_UNSPECIFIED; | |
620 | } | |
621 | ||
622 | cpu = ARM_CPU(cs); | |
623 | ||
624 | /* Synchronize our shadowed in-kernel device irq lines with the kvm ones */ | |
625 | if (run->s.regs.device_irq_level != cpu->device_irq_level) { | |
626 | switched_level = cpu->device_irq_level ^ run->s.regs.device_irq_level; | |
627 | ||
628 | qemu_mutex_lock_iothread(); | |
629 | ||
630 | if (switched_level & KVM_ARM_DEV_EL1_VTIMER) { | |
631 | qemu_set_irq(cpu->gt_timer_outputs[GTIMER_VIRT], | |
632 | !!(run->s.regs.device_irq_level & | |
633 | KVM_ARM_DEV_EL1_VTIMER)); | |
634 | switched_level &= ~KVM_ARM_DEV_EL1_VTIMER; | |
635 | } | |
636 | ||
637 | if (switched_level & KVM_ARM_DEV_EL1_PTIMER) { | |
638 | qemu_set_irq(cpu->gt_timer_outputs[GTIMER_PHYS], | |
639 | !!(run->s.regs.device_irq_level & | |
640 | KVM_ARM_DEV_EL1_PTIMER)); | |
641 | switched_level &= ~KVM_ARM_DEV_EL1_PTIMER; | |
642 | } | |
643 | ||
b1659527 AJ |
644 | if (switched_level & KVM_ARM_DEV_PMU) { |
645 | qemu_set_irq(cpu->pmu_interrupt, | |
646 | !!(run->s.regs.device_irq_level & KVM_ARM_DEV_PMU)); | |
647 | switched_level &= ~KVM_ARM_DEV_PMU; | |
648 | } | |
5d721b78 AG |
649 | |
650 | if (switched_level) { | |
651 | qemu_log_mask(LOG_UNIMP, "%s: unhandled in-kernel device IRQ %x\n", | |
652 | __func__, switched_level); | |
653 | } | |
654 | ||
655 | /* We also mark unknown levels as processed to not waste cycles */ | |
656 | cpu->device_irq_level = run->s.regs.device_irq_level; | |
657 | qemu_mutex_unlock_iothread(); | |
658 | } | |
659 | ||
4c663752 | 660 | return MEMTXATTRS_UNSPECIFIED; |
494b00c7 CD |
661 | } |
662 | ||
2ecb2027 | 663 | |
494b00c7 CD |
664 | int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run) |
665 | { | |
2ecb2027 AB |
666 | int ret = 0; |
667 | ||
668 | switch (run->exit_reason) { | |
669 | case KVM_EXIT_DEBUG: | |
670 | if (kvm_arm_handle_debug(cs, &run->debug.arch)) { | |
671 | ret = EXCP_DEBUG; | |
672 | } /* otherwise return to guest */ | |
673 | break; | |
674 | default: | |
675 | qemu_log_mask(LOG_UNIMP, "%s: un-handled exit reason %d\n", | |
676 | __func__, run->exit_reason); | |
677 | break; | |
678 | } | |
679 | return ret; | |
494b00c7 CD |
680 | } |
681 | ||
494b00c7 CD |
682 | bool kvm_arch_stop_on_emulation_error(CPUState *cs) |
683 | { | |
684 | return true; | |
685 | } | |
686 | ||
687 | int kvm_arch_process_async_events(CPUState *cs) | |
688 | { | |
689 | return 0; | |
690 | } | |
691 | ||
2ecb2027 AB |
692 | /* The #ifdef protections are until 32bit headers are imported and can |
693 | * be removed once both 32 and 64 bit reach feature parity. | |
694 | */ | |
494b00c7 CD |
695 | void kvm_arch_update_guest_debug(CPUState *cs, struct kvm_guest_debug *dbg) |
696 | { | |
2ecb2027 AB |
697 | #ifdef KVM_GUESTDBG_USE_SW_BP |
698 | if (kvm_sw_breakpoints_active(cs)) { | |
699 | dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP; | |
700 | } | |
701 | #endif | |
e4482ab7 AB |
702 | #ifdef KVM_GUESTDBG_USE_HW |
703 | if (kvm_arm_hw_debug_active(cs)) { | |
704 | dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW; | |
705 | kvm_arm_copy_hw_debug_data(&dbg->arch); | |
706 | } | |
707 | #endif | |
494b00c7 | 708 | } |
b3a1c626 AK |
709 | |
710 | void kvm_arch_init_irq_routing(KVMState *s) | |
711 | { | |
712 | } | |
1da41cc1 | 713 | |
15eafc2e | 714 | int kvm_arch_irqchip_create(MachineState *ms, KVMState *s) |
1da41cc1 | 715 | { |
15eafc2e PB |
716 | if (machine_kernel_irqchip_split(ms)) { |
717 | perror("-machine kernel_irqchip=split is not supported on ARM."); | |
718 | exit(1); | |
719 | } | |
720 | ||
1da41cc1 CD |
721 | /* If we can create the VGIC using the newer device control API, we |
722 | * let the device do this when it initializes itself, otherwise we | |
723 | * fall back to the old API */ | |
34e85cd9 PF |
724 | return kvm_check_extension(s, KVM_CAP_DEVICE_CTRL); |
725 | } | |
1da41cc1 | 726 | |
34e85cd9 PF |
727 | int kvm_arm_vgic_probe(void) |
728 | { | |
729 | if (kvm_create_device(kvm_state, | |
730 | KVM_DEV_TYPE_ARM_VGIC_V3, true) == 0) { | |
731 | return 3; | |
732 | } else if (kvm_create_device(kvm_state, | |
733 | KVM_DEV_TYPE_ARM_VGIC_V2, true) == 0) { | |
734 | return 2; | |
735 | } else { | |
736 | return 0; | |
1da41cc1 | 737 | } |
1da41cc1 | 738 | } |
9e03a040 FB |
739 | |
740 | int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route, | |
dc9f06ca | 741 | uint64_t address, uint32_t data, PCIDevice *dev) |
9e03a040 | 742 | { |
b05c81d2 EA |
743 | AddressSpace *as = pci_device_iommu_address_space(dev); |
744 | hwaddr xlat, len, doorbell_gpa; | |
745 | MemoryRegionSection mrs; | |
746 | MemoryRegion *mr; | |
747 | int ret = 1; | |
748 | ||
749 | if (as == &address_space_memory) { | |
750 | return 0; | |
751 | } | |
752 | ||
753 | /* MSI doorbell address is translated by an IOMMU */ | |
754 | ||
755 | rcu_read_lock(); | |
bc6b1cec PM |
756 | mr = address_space_translate(as, address, &xlat, &len, true, |
757 | MEMTXATTRS_UNSPECIFIED); | |
b05c81d2 EA |
758 | if (!mr) { |
759 | goto unlock; | |
760 | } | |
761 | mrs = memory_region_find(mr, xlat, 1); | |
762 | if (!mrs.mr) { | |
763 | goto unlock; | |
764 | } | |
765 | ||
766 | doorbell_gpa = mrs.offset_within_address_space; | |
767 | memory_region_unref(mrs.mr); | |
768 | ||
769 | route->u.msi.address_lo = doorbell_gpa; | |
770 | route->u.msi.address_hi = doorbell_gpa >> 32; | |
771 | ||
772 | trace_kvm_arm_fixup_msi_route(address, doorbell_gpa); | |
773 | ||
774 | ret = 0; | |
775 | ||
776 | unlock: | |
777 | rcu_read_unlock(); | |
778 | return ret; | |
9e03a040 | 779 | } |
1850b6b7 | 780 | |
38d87493 PX |
781 | int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route, |
782 | int vector, PCIDevice *dev) | |
783 | { | |
784 | return 0; | |
785 | } | |
786 | ||
787 | int kvm_arch_release_virq_post(int virq) | |
788 | { | |
789 | return 0; | |
790 | } | |
791 | ||
1850b6b7 EA |
792 | int kvm_arch_msi_data_to_gsi(uint32_t data) |
793 | { | |
794 | return (data - 32) & 0xffff; | |
795 | } |