of: MSI: Simplify irqdomain lookup
[linux/fpc-iii.git] / arch / powerpc / kvm / powerpc.c
blob6fd2405c7f4aa2e5bdc5ae417a409b33b6c5f7bd
1 /*
2 * This program is free software; you can redistribute it and/or modify
3 * it under the terms of the GNU General Public License, version 2, as
4 * published by the Free Software Foundation.
6 * This program is distributed in the hope that it will be useful,
7 * but WITHOUT ANY WARRANTY; without even the implied warranty of
8 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
9 * GNU General Public License for more details.
11 * You should have received a copy of the GNU General Public License
12 * along with this program; if not, write to the Free Software
13 * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
15 * Copyright IBM Corp. 2007
17 * Authors: Hollis Blanchard <hollisb@us.ibm.com>
18 * Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
21 #include <linux/errno.h>
22 #include <linux/err.h>
23 #include <linux/kvm_host.h>
24 #include <linux/vmalloc.h>
25 #include <linux/hrtimer.h>
26 #include <linux/fs.h>
27 #include <linux/slab.h>
28 #include <linux/file.h>
29 #include <linux/module.h>
30 #include <asm/cputable.h>
31 #include <asm/uaccess.h>
32 #include <asm/kvm_ppc.h>
33 #include <asm/tlbflush.h>
34 #include <asm/cputhreads.h>
35 #include <asm/irqflags.h>
36 #include "timing.h"
37 #include "irq.h"
38 #include "../mm/mmu_decl.h"
40 #define CREATE_TRACE_POINTS
41 #include "trace.h"
43 struct kvmppc_ops *kvmppc_hv_ops;
44 EXPORT_SYMBOL_GPL(kvmppc_hv_ops);
45 struct kvmppc_ops *kvmppc_pr_ops;
46 EXPORT_SYMBOL_GPL(kvmppc_pr_ops);
49 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
51 return !!(v->arch.pending_exceptions) ||
52 v->requests;
55 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
57 return 1;
61 * Common checks before entering the guest world. Call with interrupts
62 * disabled.
64 * returns:
66 * == 1 if we're ready to go into guest state
67 * <= 0 if we need to go back to the host with return value
69 int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu)
71 int r;
73 WARN_ON(irqs_disabled());
74 hard_irq_disable();
76 while (true) {
77 if (need_resched()) {
78 local_irq_enable();
79 cond_resched();
80 hard_irq_disable();
81 continue;
84 if (signal_pending(current)) {
85 kvmppc_account_exit(vcpu, SIGNAL_EXITS);
86 vcpu->run->exit_reason = KVM_EXIT_INTR;
87 r = -EINTR;
88 break;
91 vcpu->mode = IN_GUEST_MODE;
94 * Reading vcpu->requests must happen after setting vcpu->mode,
95 * so we don't miss a request because the requester sees
96 * OUTSIDE_GUEST_MODE and assumes we'll be checking requests
97 * before next entering the guest (and thus doesn't IPI).
99 smp_mb();
101 if (vcpu->requests) {
102 /* Make sure we process requests preemptable */
103 local_irq_enable();
104 trace_kvm_check_requests(vcpu);
105 r = kvmppc_core_check_requests(vcpu);
106 hard_irq_disable();
107 if (r > 0)
108 continue;
109 break;
112 if (kvmppc_core_prepare_to_enter(vcpu)) {
113 /* interrupts got enabled in between, so we
114 are back at square 1 */
115 continue;
118 __kvm_guest_enter();
119 return 1;
122 /* return to host */
123 local_irq_enable();
124 return r;
126 EXPORT_SYMBOL_GPL(kvmppc_prepare_to_enter);
128 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
129 static void kvmppc_swab_shared(struct kvm_vcpu *vcpu)
131 struct kvm_vcpu_arch_shared *shared = vcpu->arch.shared;
132 int i;
134 shared->sprg0 = swab64(shared->sprg0);
135 shared->sprg1 = swab64(shared->sprg1);
136 shared->sprg2 = swab64(shared->sprg2);
137 shared->sprg3 = swab64(shared->sprg3);
138 shared->srr0 = swab64(shared->srr0);
139 shared->srr1 = swab64(shared->srr1);
140 shared->dar = swab64(shared->dar);
141 shared->msr = swab64(shared->msr);
142 shared->dsisr = swab32(shared->dsisr);
143 shared->int_pending = swab32(shared->int_pending);
144 for (i = 0; i < ARRAY_SIZE(shared->sr); i++)
145 shared->sr[i] = swab32(shared->sr[i]);
147 #endif
149 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
151 int nr = kvmppc_get_gpr(vcpu, 11);
152 int r;
153 unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
154 unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
155 unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
156 unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
157 unsigned long r2 = 0;
159 if (!(kvmppc_get_msr(vcpu) & MSR_SF)) {
160 /* 32 bit mode */
161 param1 &= 0xffffffff;
162 param2 &= 0xffffffff;
163 param3 &= 0xffffffff;
164 param4 &= 0xffffffff;
167 switch (nr) {
168 case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE):
170 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
171 /* Book3S can be little endian, find it out here */
172 int shared_big_endian = true;
173 if (vcpu->arch.intr_msr & MSR_LE)
174 shared_big_endian = false;
175 if (shared_big_endian != vcpu->arch.shared_big_endian)
176 kvmppc_swab_shared(vcpu);
177 vcpu->arch.shared_big_endian = shared_big_endian;
178 #endif
180 if (!(param2 & MAGIC_PAGE_FLAG_NOT_MAPPED_NX)) {
182 * Older versions of the Linux magic page code had
183 * a bug where they would map their trampoline code
184 * NX. If that's the case, remove !PR NX capability.
186 vcpu->arch.disable_kernel_nx = true;
187 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
190 vcpu->arch.magic_page_pa = param1 & ~0xfffULL;
191 vcpu->arch.magic_page_ea = param2 & ~0xfffULL;
193 #ifdef CONFIG_PPC_64K_PAGES
195 * Make sure our 4k magic page is in the same window of a 64k
196 * page within the guest and within the host's page.
198 if ((vcpu->arch.magic_page_pa & 0xf000) !=
199 ((ulong)vcpu->arch.shared & 0xf000)) {
200 void *old_shared = vcpu->arch.shared;
201 ulong shared = (ulong)vcpu->arch.shared;
202 void *new_shared;
204 shared &= PAGE_MASK;
205 shared |= vcpu->arch.magic_page_pa & 0xf000;
206 new_shared = (void*)shared;
207 memcpy(new_shared, old_shared, 0x1000);
208 vcpu->arch.shared = new_shared;
210 #endif
212 r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
214 r = EV_SUCCESS;
215 break;
217 case KVM_HCALL_TOKEN(KVM_HC_FEATURES):
218 r = EV_SUCCESS;
219 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
220 r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
221 #endif
223 /* Second return value is in r4 */
224 break;
225 case EV_HCALL_TOKEN(EV_IDLE):
226 r = EV_SUCCESS;
227 kvm_vcpu_block(vcpu);
228 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
229 break;
230 default:
231 r = EV_UNIMPLEMENTED;
232 break;
235 kvmppc_set_gpr(vcpu, 4, r2);
237 return r;
239 EXPORT_SYMBOL_GPL(kvmppc_kvm_pv);
241 int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
243 int r = false;
245 /* We have to know what CPU to virtualize */
246 if (!vcpu->arch.pvr)
247 goto out;
249 /* PAPR only works with book3s_64 */
250 if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
251 goto out;
253 /* HV KVM can only do PAPR mode for now */
254 if (!vcpu->arch.papr_enabled && is_kvmppc_hv_enabled(vcpu->kvm))
255 goto out;
257 #ifdef CONFIG_KVM_BOOKE_HV
258 if (!cpu_has_feature(CPU_FTR_EMB_HV))
259 goto out;
260 #endif
262 r = true;
264 out:
265 vcpu->arch.sane = r;
266 return r ? 0 : -EINVAL;
268 EXPORT_SYMBOL_GPL(kvmppc_sanity_check);
270 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
272 enum emulation_result er;
273 int r;
275 er = kvmppc_emulate_loadstore(vcpu);
276 switch (er) {
277 case EMULATE_DONE:
278 /* Future optimization: only reload non-volatiles if they were
279 * actually modified. */
280 r = RESUME_GUEST_NV;
281 break;
282 case EMULATE_AGAIN:
283 r = RESUME_GUEST;
284 break;
285 case EMULATE_DO_MMIO:
286 run->exit_reason = KVM_EXIT_MMIO;
287 /* We must reload nonvolatiles because "update" load/store
288 * instructions modify register state. */
289 /* Future optimization: only reload non-volatiles if they were
290 * actually modified. */
291 r = RESUME_HOST_NV;
292 break;
293 case EMULATE_FAIL:
295 u32 last_inst;
297 kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
298 /* XXX Deliver Program interrupt to guest. */
299 pr_emerg("%s: emulation failed (%08x)\n", __func__, last_inst);
300 r = RESUME_HOST;
301 break;
303 default:
304 WARN_ON(1);
305 r = RESUME_GUEST;
308 return r;
310 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio);
312 int kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
313 bool data)
315 ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
316 struct kvmppc_pte pte;
317 int r;
319 vcpu->stat.st++;
321 r = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
322 XLATE_WRITE, &pte);
323 if (r < 0)
324 return r;
326 *eaddr = pte.raddr;
328 if (!pte.may_write)
329 return -EPERM;
331 /* Magic page override */
332 if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
333 ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
334 !(kvmppc_get_msr(vcpu) & MSR_PR)) {
335 void *magic = vcpu->arch.shared;
336 magic += pte.eaddr & 0xfff;
337 memcpy(magic, ptr, size);
338 return EMULATE_DONE;
341 if (kvm_write_guest(vcpu->kvm, pte.raddr, ptr, size))
342 return EMULATE_DO_MMIO;
344 return EMULATE_DONE;
346 EXPORT_SYMBOL_GPL(kvmppc_st);
348 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
349 bool data)
351 ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
352 struct kvmppc_pte pte;
353 int rc;
355 vcpu->stat.ld++;
357 rc = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
358 XLATE_READ, &pte);
359 if (rc)
360 return rc;
362 *eaddr = pte.raddr;
364 if (!pte.may_read)
365 return -EPERM;
367 if (!data && !pte.may_execute)
368 return -ENOEXEC;
370 /* Magic page override */
371 if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
372 ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
373 !(kvmppc_get_msr(vcpu) & MSR_PR)) {
374 void *magic = vcpu->arch.shared;
375 magic += pte.eaddr & 0xfff;
376 memcpy(ptr, magic, size);
377 return EMULATE_DONE;
380 if (kvm_read_guest(vcpu->kvm, pte.raddr, ptr, size))
381 return EMULATE_DO_MMIO;
383 return EMULATE_DONE;
385 EXPORT_SYMBOL_GPL(kvmppc_ld);
387 int kvm_arch_hardware_enable(void)
389 return 0;
392 int kvm_arch_hardware_setup(void)
394 return 0;
397 void kvm_arch_check_processor_compat(void *rtn)
399 *(int *)rtn = kvmppc_core_check_processor_compat();
402 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
404 struct kvmppc_ops *kvm_ops = NULL;
406 * if we have both HV and PR enabled, default is HV
408 if (type == 0) {
409 if (kvmppc_hv_ops)
410 kvm_ops = kvmppc_hv_ops;
411 else
412 kvm_ops = kvmppc_pr_ops;
413 if (!kvm_ops)
414 goto err_out;
415 } else if (type == KVM_VM_PPC_HV) {
416 if (!kvmppc_hv_ops)
417 goto err_out;
418 kvm_ops = kvmppc_hv_ops;
419 } else if (type == KVM_VM_PPC_PR) {
420 if (!kvmppc_pr_ops)
421 goto err_out;
422 kvm_ops = kvmppc_pr_ops;
423 } else
424 goto err_out;
426 if (kvm_ops->owner && !try_module_get(kvm_ops->owner))
427 return -ENOENT;
429 kvm->arch.kvm_ops = kvm_ops;
430 return kvmppc_core_init_vm(kvm);
431 err_out:
432 return -EINVAL;
435 void kvm_arch_destroy_vm(struct kvm *kvm)
437 unsigned int i;
438 struct kvm_vcpu *vcpu;
440 kvm_for_each_vcpu(i, vcpu, kvm)
441 kvm_arch_vcpu_free(vcpu);
443 mutex_lock(&kvm->lock);
444 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
445 kvm->vcpus[i] = NULL;
447 atomic_set(&kvm->online_vcpus, 0);
449 kvmppc_core_destroy_vm(kvm);
451 mutex_unlock(&kvm->lock);
453 /* drop the module reference */
454 module_put(kvm->arch.kvm_ops->owner);
457 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
459 int r;
460 /* Assume we're using HV mode when the HV module is loaded */
461 int hv_enabled = kvmppc_hv_ops ? 1 : 0;
463 if (kvm) {
465 * Hooray - we know which VM type we're running on. Depend on
466 * that rather than the guess above.
468 hv_enabled = is_kvmppc_hv_enabled(kvm);
471 switch (ext) {
472 #ifdef CONFIG_BOOKE
473 case KVM_CAP_PPC_BOOKE_SREGS:
474 case KVM_CAP_PPC_BOOKE_WATCHDOG:
475 case KVM_CAP_PPC_EPR:
476 #else
477 case KVM_CAP_PPC_SEGSTATE:
478 case KVM_CAP_PPC_HIOR:
479 case KVM_CAP_PPC_PAPR:
480 #endif
481 case KVM_CAP_PPC_UNSET_IRQ:
482 case KVM_CAP_PPC_IRQ_LEVEL:
483 case KVM_CAP_ENABLE_CAP:
484 case KVM_CAP_ENABLE_CAP_VM:
485 case KVM_CAP_ONE_REG:
486 case KVM_CAP_IOEVENTFD:
487 case KVM_CAP_DEVICE_CTRL:
488 r = 1;
489 break;
490 case KVM_CAP_PPC_PAIRED_SINGLES:
491 case KVM_CAP_PPC_OSI:
492 case KVM_CAP_PPC_GET_PVINFO:
493 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
494 case KVM_CAP_SW_TLB:
495 #endif
496 /* We support this only for PR */
497 r = !hv_enabled;
498 break;
499 #ifdef CONFIG_KVM_MMIO
500 case KVM_CAP_COALESCED_MMIO:
501 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
502 break;
503 #endif
504 #ifdef CONFIG_KVM_MPIC
505 case KVM_CAP_IRQ_MPIC:
506 r = 1;
507 break;
508 #endif
510 #ifdef CONFIG_PPC_BOOK3S_64
511 case KVM_CAP_SPAPR_TCE:
512 case KVM_CAP_PPC_ALLOC_HTAB:
513 case KVM_CAP_PPC_RTAS:
514 case KVM_CAP_PPC_FIXUP_HCALL:
515 case KVM_CAP_PPC_ENABLE_HCALL:
516 #ifdef CONFIG_KVM_XICS
517 case KVM_CAP_IRQ_XICS:
518 #endif
519 r = 1;
520 break;
521 #endif /* CONFIG_PPC_BOOK3S_64 */
522 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
523 case KVM_CAP_PPC_SMT:
524 if (hv_enabled)
525 r = threads_per_subcore;
526 else
527 r = 0;
528 break;
529 case KVM_CAP_PPC_RMA:
530 r = 0;
531 break;
532 case KVM_CAP_PPC_HWRNG:
533 r = kvmppc_hwrng_present();
534 break;
535 #endif
536 case KVM_CAP_SYNC_MMU:
537 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
538 r = hv_enabled;
539 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
540 r = 1;
541 #else
542 r = 0;
543 #endif
544 break;
545 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
546 case KVM_CAP_PPC_HTAB_FD:
547 r = hv_enabled;
548 break;
549 #endif
550 case KVM_CAP_NR_VCPUS:
552 * Recommending a number of CPUs is somewhat arbitrary; we
553 * return the number of present CPUs for -HV (since a host
554 * will have secondary threads "offline"), and for other KVM
555 * implementations just count online CPUs.
557 if (hv_enabled)
558 r = num_present_cpus();
559 else
560 r = num_online_cpus();
561 break;
562 case KVM_CAP_NR_MEMSLOTS:
563 r = KVM_USER_MEM_SLOTS;
564 break;
565 case KVM_CAP_MAX_VCPUS:
566 r = KVM_MAX_VCPUS;
567 break;
568 #ifdef CONFIG_PPC_BOOK3S_64
569 case KVM_CAP_PPC_GET_SMMU_INFO:
570 r = 1;
571 break;
572 #endif
573 default:
574 r = 0;
575 break;
577 return r;
581 long kvm_arch_dev_ioctl(struct file *filp,
582 unsigned int ioctl, unsigned long arg)
584 return -EINVAL;
587 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
588 struct kvm_memory_slot *dont)
590 kvmppc_core_free_memslot(kvm, free, dont);
593 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
594 unsigned long npages)
596 return kvmppc_core_create_memslot(kvm, slot, npages);
599 int kvm_arch_prepare_memory_region(struct kvm *kvm,
600 struct kvm_memory_slot *memslot,
601 const struct kvm_userspace_memory_region *mem,
602 enum kvm_mr_change change)
604 return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
607 void kvm_arch_commit_memory_region(struct kvm *kvm,
608 const struct kvm_userspace_memory_region *mem,
609 const struct kvm_memory_slot *old,
610 const struct kvm_memory_slot *new,
611 enum kvm_mr_change change)
613 kvmppc_core_commit_memory_region(kvm, mem, old, new);
616 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
617 struct kvm_memory_slot *slot)
619 kvmppc_core_flush_memslot(kvm, slot);
622 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
624 struct kvm_vcpu *vcpu;
625 vcpu = kvmppc_core_vcpu_create(kvm, id);
626 if (!IS_ERR(vcpu)) {
627 vcpu->arch.wqp = &vcpu->wq;
628 kvmppc_create_vcpu_debugfs(vcpu, id);
630 return vcpu;
633 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
637 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
639 /* Make sure we're not using the vcpu anymore */
640 hrtimer_cancel(&vcpu->arch.dec_timer);
642 kvmppc_remove_vcpu_debugfs(vcpu);
644 switch (vcpu->arch.irq_type) {
645 case KVMPPC_IRQ_MPIC:
646 kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
647 break;
648 case KVMPPC_IRQ_XICS:
649 kvmppc_xics_free_icp(vcpu);
650 break;
653 kvmppc_core_vcpu_free(vcpu);
656 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
658 kvm_arch_vcpu_free(vcpu);
661 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
663 return kvmppc_core_pending_dec(vcpu);
666 static enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
668 struct kvm_vcpu *vcpu;
670 vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
671 kvmppc_decrementer_func(vcpu);
673 return HRTIMER_NORESTART;
676 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
678 int ret;
680 hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
681 vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
682 vcpu->arch.dec_expires = ~(u64)0;
684 #ifdef CONFIG_KVM_EXIT_TIMING
685 mutex_init(&vcpu->arch.exit_timing_lock);
686 #endif
687 ret = kvmppc_subarch_vcpu_init(vcpu);
688 return ret;
691 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
693 kvmppc_mmu_destroy(vcpu);
694 kvmppc_subarch_vcpu_uninit(vcpu);
697 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
699 #ifdef CONFIG_BOOKE
701 * vrsave (formerly usprg0) isn't used by Linux, but may
702 * be used by the guest.
704 * On non-booke this is associated with Altivec and
705 * is handled by code in book3s.c.
707 mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
708 #endif
709 kvmppc_core_vcpu_load(vcpu, cpu);
712 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
714 kvmppc_core_vcpu_put(vcpu);
715 #ifdef CONFIG_BOOKE
716 vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
717 #endif
720 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
721 struct kvm_run *run)
723 u64 uninitialized_var(gpr);
725 if (run->mmio.len > sizeof(gpr)) {
726 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
727 return;
730 if (!vcpu->arch.mmio_host_swabbed) {
731 switch (run->mmio.len) {
732 case 8: gpr = *(u64 *)run->mmio.data; break;
733 case 4: gpr = *(u32 *)run->mmio.data; break;
734 case 2: gpr = *(u16 *)run->mmio.data; break;
735 case 1: gpr = *(u8 *)run->mmio.data; break;
737 } else {
738 switch (run->mmio.len) {
739 case 8: gpr = swab64(*(u64 *)run->mmio.data); break;
740 case 4: gpr = swab32(*(u32 *)run->mmio.data); break;
741 case 2: gpr = swab16(*(u16 *)run->mmio.data); break;
742 case 1: gpr = *(u8 *)run->mmio.data; break;
746 if (vcpu->arch.mmio_sign_extend) {
747 switch (run->mmio.len) {
748 #ifdef CONFIG_PPC64
749 case 4:
750 gpr = (s64)(s32)gpr;
751 break;
752 #endif
753 case 2:
754 gpr = (s64)(s16)gpr;
755 break;
756 case 1:
757 gpr = (s64)(s8)gpr;
758 break;
762 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
764 switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
765 case KVM_MMIO_REG_GPR:
766 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
767 break;
768 case KVM_MMIO_REG_FPR:
769 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
770 break;
771 #ifdef CONFIG_PPC_BOOK3S
772 case KVM_MMIO_REG_QPR:
773 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
774 break;
775 case KVM_MMIO_REG_FQPR:
776 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
777 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
778 break;
779 #endif
780 default:
781 BUG();
785 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
786 unsigned int rt, unsigned int bytes,
787 int is_default_endian)
789 int idx, ret;
790 bool host_swabbed;
792 /* Pity C doesn't have a logical XOR operator */
793 if (kvmppc_need_byteswap(vcpu)) {
794 host_swabbed = is_default_endian;
795 } else {
796 host_swabbed = !is_default_endian;
799 if (bytes > sizeof(run->mmio.data)) {
800 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
801 run->mmio.len);
804 run->mmio.phys_addr = vcpu->arch.paddr_accessed;
805 run->mmio.len = bytes;
806 run->mmio.is_write = 0;
808 vcpu->arch.io_gpr = rt;
809 vcpu->arch.mmio_host_swabbed = host_swabbed;
810 vcpu->mmio_needed = 1;
811 vcpu->mmio_is_write = 0;
812 vcpu->arch.mmio_sign_extend = 0;
814 idx = srcu_read_lock(&vcpu->kvm->srcu);
816 ret = kvm_io_bus_read(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
817 bytes, &run->mmio.data);
819 srcu_read_unlock(&vcpu->kvm->srcu, idx);
821 if (!ret) {
822 kvmppc_complete_mmio_load(vcpu, run);
823 vcpu->mmio_needed = 0;
824 return EMULATE_DONE;
827 return EMULATE_DO_MMIO;
829 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
831 /* Same as above, but sign extends */
832 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
833 unsigned int rt, unsigned int bytes,
834 int is_default_endian)
836 int r;
838 vcpu->arch.mmio_sign_extend = 1;
839 r = kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian);
841 return r;
844 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
845 u64 val, unsigned int bytes, int is_default_endian)
847 void *data = run->mmio.data;
848 int idx, ret;
849 bool host_swabbed;
851 /* Pity C doesn't have a logical XOR operator */
852 if (kvmppc_need_byteswap(vcpu)) {
853 host_swabbed = is_default_endian;
854 } else {
855 host_swabbed = !is_default_endian;
858 if (bytes > sizeof(run->mmio.data)) {
859 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
860 run->mmio.len);
863 run->mmio.phys_addr = vcpu->arch.paddr_accessed;
864 run->mmio.len = bytes;
865 run->mmio.is_write = 1;
866 vcpu->mmio_needed = 1;
867 vcpu->mmio_is_write = 1;
869 /* Store the value at the lowest bytes in 'data'. */
870 if (!host_swabbed) {
871 switch (bytes) {
872 case 8: *(u64 *)data = val; break;
873 case 4: *(u32 *)data = val; break;
874 case 2: *(u16 *)data = val; break;
875 case 1: *(u8 *)data = val; break;
877 } else {
878 switch (bytes) {
879 case 8: *(u64 *)data = swab64(val); break;
880 case 4: *(u32 *)data = swab32(val); break;
881 case 2: *(u16 *)data = swab16(val); break;
882 case 1: *(u8 *)data = val; break;
886 idx = srcu_read_lock(&vcpu->kvm->srcu);
888 ret = kvm_io_bus_write(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
889 bytes, &run->mmio.data);
891 srcu_read_unlock(&vcpu->kvm->srcu, idx);
893 if (!ret) {
894 vcpu->mmio_needed = 0;
895 return EMULATE_DONE;
898 return EMULATE_DO_MMIO;
900 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
902 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
904 int r = 0;
905 union kvmppc_one_reg val;
906 int size;
908 size = one_reg_size(reg->id);
909 if (size > sizeof(val))
910 return -EINVAL;
912 r = kvmppc_get_one_reg(vcpu, reg->id, &val);
913 if (r == -EINVAL) {
914 r = 0;
915 switch (reg->id) {
916 #ifdef CONFIG_ALTIVEC
917 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
918 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
919 r = -ENXIO;
920 break;
922 vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0] = val.vval;
923 break;
924 case KVM_REG_PPC_VSCR:
925 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
926 r = -ENXIO;
927 break;
929 vcpu->arch.vr.vscr.u[3] = set_reg_val(reg->id, val);
930 break;
931 case KVM_REG_PPC_VRSAVE:
932 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
933 r = -ENXIO;
934 break;
936 vcpu->arch.vrsave = set_reg_val(reg->id, val);
937 break;
938 #endif /* CONFIG_ALTIVEC */
939 default:
940 r = -EINVAL;
941 break;
945 if (r)
946 return r;
948 if (copy_to_user((char __user *)(unsigned long)reg->addr, &val, size))
949 r = -EFAULT;
951 return r;
954 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
956 int r;
957 union kvmppc_one_reg val;
958 int size;
960 size = one_reg_size(reg->id);
961 if (size > sizeof(val))
962 return -EINVAL;
964 if (copy_from_user(&val, (char __user *)(unsigned long)reg->addr, size))
965 return -EFAULT;
967 r = kvmppc_set_one_reg(vcpu, reg->id, &val);
968 if (r == -EINVAL) {
969 r = 0;
970 switch (reg->id) {
971 #ifdef CONFIG_ALTIVEC
972 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
973 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
974 r = -ENXIO;
975 break;
977 val.vval = vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0];
978 break;
979 case KVM_REG_PPC_VSCR:
980 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
981 r = -ENXIO;
982 break;
984 val = get_reg_val(reg->id, vcpu->arch.vr.vscr.u[3]);
985 break;
986 case KVM_REG_PPC_VRSAVE:
987 val = get_reg_val(reg->id, vcpu->arch.vrsave);
988 break;
989 #endif /* CONFIG_ALTIVEC */
990 default:
991 r = -EINVAL;
992 break;
996 return r;
999 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
1001 int r;
1002 sigset_t sigsaved;
1004 if (vcpu->sigset_active)
1005 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1007 if (vcpu->mmio_needed) {
1008 if (!vcpu->mmio_is_write)
1009 kvmppc_complete_mmio_load(vcpu, run);
1010 vcpu->mmio_needed = 0;
1011 } else if (vcpu->arch.osi_needed) {
1012 u64 *gprs = run->osi.gprs;
1013 int i;
1015 for (i = 0; i < 32; i++)
1016 kvmppc_set_gpr(vcpu, i, gprs[i]);
1017 vcpu->arch.osi_needed = 0;
1018 } else if (vcpu->arch.hcall_needed) {
1019 int i;
1021 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
1022 for (i = 0; i < 9; ++i)
1023 kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
1024 vcpu->arch.hcall_needed = 0;
1025 #ifdef CONFIG_BOOKE
1026 } else if (vcpu->arch.epr_needed) {
1027 kvmppc_set_epr(vcpu, run->epr.epr);
1028 vcpu->arch.epr_needed = 0;
1029 #endif
1032 r = kvmppc_vcpu_run(run, vcpu);
1034 if (vcpu->sigset_active)
1035 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1037 return r;
1040 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
1042 if (irq->irq == KVM_INTERRUPT_UNSET) {
1043 kvmppc_core_dequeue_external(vcpu);
1044 return 0;
1047 kvmppc_core_queue_external(vcpu, irq);
1049 kvm_vcpu_kick(vcpu);
1051 return 0;
1054 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
1055 struct kvm_enable_cap *cap)
1057 int r;
1059 if (cap->flags)
1060 return -EINVAL;
1062 switch (cap->cap) {
1063 case KVM_CAP_PPC_OSI:
1064 r = 0;
1065 vcpu->arch.osi_enabled = true;
1066 break;
1067 case KVM_CAP_PPC_PAPR:
1068 r = 0;
1069 vcpu->arch.papr_enabled = true;
1070 break;
1071 case KVM_CAP_PPC_EPR:
1072 r = 0;
1073 if (cap->args[0])
1074 vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
1075 else
1076 vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
1077 break;
1078 #ifdef CONFIG_BOOKE
1079 case KVM_CAP_PPC_BOOKE_WATCHDOG:
1080 r = 0;
1081 vcpu->arch.watchdog_enabled = true;
1082 break;
1083 #endif
1084 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1085 case KVM_CAP_SW_TLB: {
1086 struct kvm_config_tlb cfg;
1087 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
1089 r = -EFAULT;
1090 if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
1091 break;
1093 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
1094 break;
1096 #endif
1097 #ifdef CONFIG_KVM_MPIC
1098 case KVM_CAP_IRQ_MPIC: {
1099 struct fd f;
1100 struct kvm_device *dev;
1102 r = -EBADF;
1103 f = fdget(cap->args[0]);
1104 if (!f.file)
1105 break;
1107 r = -EPERM;
1108 dev = kvm_device_from_filp(f.file);
1109 if (dev)
1110 r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
1112 fdput(f);
1113 break;
1115 #endif
1116 #ifdef CONFIG_KVM_XICS
1117 case KVM_CAP_IRQ_XICS: {
1118 struct fd f;
1119 struct kvm_device *dev;
1121 r = -EBADF;
1122 f = fdget(cap->args[0]);
1123 if (!f.file)
1124 break;
1126 r = -EPERM;
1127 dev = kvm_device_from_filp(f.file);
1128 if (dev)
1129 r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
1131 fdput(f);
1132 break;
1134 #endif /* CONFIG_KVM_XICS */
1135 default:
1136 r = -EINVAL;
1137 break;
1140 if (!r)
1141 r = kvmppc_sanity_check(vcpu);
1143 return r;
1146 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1147 struct kvm_mp_state *mp_state)
1149 return -EINVAL;
1152 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1153 struct kvm_mp_state *mp_state)
1155 return -EINVAL;
1158 long kvm_arch_vcpu_ioctl(struct file *filp,
1159 unsigned int ioctl, unsigned long arg)
1161 struct kvm_vcpu *vcpu = filp->private_data;
1162 void __user *argp = (void __user *)arg;
1163 long r;
1165 switch (ioctl) {
1166 case KVM_INTERRUPT: {
1167 struct kvm_interrupt irq;
1168 r = -EFAULT;
1169 if (copy_from_user(&irq, argp, sizeof(irq)))
1170 goto out;
1171 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1172 goto out;
1175 case KVM_ENABLE_CAP:
1177 struct kvm_enable_cap cap;
1178 r = -EFAULT;
1179 if (copy_from_user(&cap, argp, sizeof(cap)))
1180 goto out;
1181 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
1182 break;
1185 case KVM_SET_ONE_REG:
1186 case KVM_GET_ONE_REG:
1188 struct kvm_one_reg reg;
1189 r = -EFAULT;
1190 if (copy_from_user(&reg, argp, sizeof(reg)))
1191 goto out;
1192 if (ioctl == KVM_SET_ONE_REG)
1193 r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
1194 else
1195 r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
1196 break;
1199 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1200 case KVM_DIRTY_TLB: {
1201 struct kvm_dirty_tlb dirty;
1202 r = -EFAULT;
1203 if (copy_from_user(&dirty, argp, sizeof(dirty)))
1204 goto out;
1205 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
1206 break;
1208 #endif
1209 default:
1210 r = -EINVAL;
1213 out:
1214 return r;
1217 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
1219 return VM_FAULT_SIGBUS;
1222 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
1224 u32 inst_nop = 0x60000000;
1225 #ifdef CONFIG_KVM_BOOKE_HV
1226 u32 inst_sc1 = 0x44000022;
1227 pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
1228 pvinfo->hcall[1] = cpu_to_be32(inst_nop);
1229 pvinfo->hcall[2] = cpu_to_be32(inst_nop);
1230 pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1231 #else
1232 u32 inst_lis = 0x3c000000;
1233 u32 inst_ori = 0x60000000;
1234 u32 inst_sc = 0x44000002;
1235 u32 inst_imm_mask = 0xffff;
1238 * The hypercall to get into KVM from within guest context is as
1239 * follows:
1241 * lis r0, r0, KVM_SC_MAGIC_R0@h
1242 * ori r0, KVM_SC_MAGIC_R0@l
1243 * sc
1244 * nop
1246 pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
1247 pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
1248 pvinfo->hcall[2] = cpu_to_be32(inst_sc);
1249 pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1250 #endif
1252 pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
1254 return 0;
1257 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
1258 bool line_status)
1260 if (!irqchip_in_kernel(kvm))
1261 return -ENXIO;
1263 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1264 irq_event->irq, irq_event->level,
1265 line_status);
1266 return 0;
1270 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
1271 struct kvm_enable_cap *cap)
1273 int r;
1275 if (cap->flags)
1276 return -EINVAL;
1278 switch (cap->cap) {
1279 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
1280 case KVM_CAP_PPC_ENABLE_HCALL: {
1281 unsigned long hcall = cap->args[0];
1283 r = -EINVAL;
1284 if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
1285 cap->args[1] > 1)
1286 break;
1287 if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
1288 break;
1289 if (cap->args[1])
1290 set_bit(hcall / 4, kvm->arch.enabled_hcalls);
1291 else
1292 clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
1293 r = 0;
1294 break;
1296 #endif
1297 default:
1298 r = -EINVAL;
1299 break;
1302 return r;
1305 long kvm_arch_vm_ioctl(struct file *filp,
1306 unsigned int ioctl, unsigned long arg)
1308 struct kvm *kvm __maybe_unused = filp->private_data;
1309 void __user *argp = (void __user *)arg;
1310 long r;
1312 switch (ioctl) {
1313 case KVM_PPC_GET_PVINFO: {
1314 struct kvm_ppc_pvinfo pvinfo;
1315 memset(&pvinfo, 0, sizeof(pvinfo));
1316 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
1317 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
1318 r = -EFAULT;
1319 goto out;
1322 break;
1324 case KVM_ENABLE_CAP:
1326 struct kvm_enable_cap cap;
1327 r = -EFAULT;
1328 if (copy_from_user(&cap, argp, sizeof(cap)))
1329 goto out;
1330 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
1331 break;
1333 #ifdef CONFIG_PPC_BOOK3S_64
1334 case KVM_CREATE_SPAPR_TCE: {
1335 struct kvm_create_spapr_tce create_tce;
1337 r = -EFAULT;
1338 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
1339 goto out;
1340 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce);
1341 goto out;
1343 case KVM_PPC_GET_SMMU_INFO: {
1344 struct kvm_ppc_smmu_info info;
1345 struct kvm *kvm = filp->private_data;
1347 memset(&info, 0, sizeof(info));
1348 r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
1349 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
1350 r = -EFAULT;
1351 break;
1353 case KVM_PPC_RTAS_DEFINE_TOKEN: {
1354 struct kvm *kvm = filp->private_data;
1356 r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
1357 break;
1359 default: {
1360 struct kvm *kvm = filp->private_data;
1361 r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
1363 #else /* CONFIG_PPC_BOOK3S_64 */
1364 default:
1365 r = -ENOTTY;
1366 #endif
1368 out:
1369 return r;
1372 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
1373 static unsigned long nr_lpids;
1375 long kvmppc_alloc_lpid(void)
1377 long lpid;
1379 do {
1380 lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
1381 if (lpid >= nr_lpids) {
1382 pr_err("%s: No LPIDs free\n", __func__);
1383 return -ENOMEM;
1385 } while (test_and_set_bit(lpid, lpid_inuse));
1387 return lpid;
1389 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
1391 void kvmppc_claim_lpid(long lpid)
1393 set_bit(lpid, lpid_inuse);
1395 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
1397 void kvmppc_free_lpid(long lpid)
1399 clear_bit(lpid, lpid_inuse);
1401 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
1403 void kvmppc_init_lpid(unsigned long nr_lpids_param)
1405 nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
1406 memset(lpid_inuse, 0, sizeof(lpid_inuse));
1408 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
1410 int kvm_arch_init(void *opaque)
1412 return 0;
1415 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr);