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>
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>
38 #include "../mm/mmu_decl.h"
40 #define CREATE_TRACE_POINTS
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) ||
55 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
61 * Common checks before entering the guest world. Call with interrupts
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)
73 WARN_ON(irqs_disabled());
84 if (signal_pending(current)) {
85 kvmppc_account_exit(vcpu, SIGNAL_EXITS);
86 vcpu->run->exit_reason = KVM_EXIT_INTR;
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).
101 if (vcpu->requests) {
102 /* Make sure we process requests preemptable */
104 trace_kvm_check_requests(vcpu);
105 r = kvmppc_core_check_requests(vcpu);
112 if (kvmppc_core_prepare_to_enter(vcpu)) {
113 /* interrupts got enabled in between, so we
114 are back at square 1 */
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;
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]);
149 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
151 int nr = kvmppc_get_gpr(vcpu, 11);
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)) {
161 param1 &= 0xffffffff;
162 param2 &= 0xffffffff;
163 param3 &= 0xffffffff;
164 param4 &= 0xffffffff;
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;
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 r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
198 case KVM_HCALL_TOKEN(KVM_HC_FEATURES):
200 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
201 /* XXX Missing magic page on 44x */
202 r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
205 /* Second return value is in r4 */
207 case EV_HCALL_TOKEN(EV_IDLE):
209 kvm_vcpu_block(vcpu);
210 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
213 r = EV_UNIMPLEMENTED;
217 kvmppc_set_gpr(vcpu, 4, r2);
221 EXPORT_SYMBOL_GPL(kvmppc_kvm_pv);
223 int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
227 /* We have to know what CPU to virtualize */
231 /* PAPR only works with book3s_64 */
232 if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
235 /* HV KVM can only do PAPR mode for now */
236 if (!vcpu->arch.papr_enabled && is_kvmppc_hv_enabled(vcpu->kvm))
239 #ifdef CONFIG_KVM_BOOKE_HV
240 if (!cpu_has_feature(CPU_FTR_EMB_HV))
248 return r ? 0 : -EINVAL;
250 EXPORT_SYMBOL_GPL(kvmppc_sanity_check);
252 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
254 enum emulation_result er;
257 er = kvmppc_emulate_instruction(run, vcpu);
260 /* Future optimization: only reload non-volatiles if they were
261 * actually modified. */
264 case EMULATE_DO_MMIO:
265 run->exit_reason = KVM_EXIT_MMIO;
266 /* We must reload nonvolatiles because "update" load/store
267 * instructions modify register state. */
268 /* Future optimization: only reload non-volatiles if they were
269 * actually modified. */
273 /* XXX Deliver Program interrupt to guest. */
274 printk(KERN_EMERG "%s: emulation failed (%08x)\n", __func__,
275 kvmppc_get_last_inst(vcpu));
285 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio);
287 int kvm_arch_hardware_enable(void *garbage)
292 void kvm_arch_hardware_disable(void *garbage)
296 int kvm_arch_hardware_setup(void)
301 void kvm_arch_hardware_unsetup(void)
305 void kvm_arch_check_processor_compat(void *rtn)
307 *(int *)rtn = kvmppc_core_check_processor_compat();
310 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
312 struct kvmppc_ops *kvm_ops = NULL;
314 * if we have both HV and PR enabled, default is HV
318 kvm_ops = kvmppc_hv_ops;
320 kvm_ops = kvmppc_pr_ops;
323 } else if (type == KVM_VM_PPC_HV) {
326 kvm_ops = kvmppc_hv_ops;
327 } else if (type == KVM_VM_PPC_PR) {
330 kvm_ops = kvmppc_pr_ops;
334 if (kvm_ops->owner && !try_module_get(kvm_ops->owner))
337 kvm->arch.kvm_ops = kvm_ops;
338 return kvmppc_core_init_vm(kvm);
343 void kvm_arch_destroy_vm(struct kvm *kvm)
346 struct kvm_vcpu *vcpu;
348 kvm_for_each_vcpu(i, vcpu, kvm)
349 kvm_arch_vcpu_free(vcpu);
351 mutex_lock(&kvm->lock);
352 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
353 kvm->vcpus[i] = NULL;
355 atomic_set(&kvm->online_vcpus, 0);
357 kvmppc_core_destroy_vm(kvm);
359 mutex_unlock(&kvm->lock);
361 /* drop the module reference */
362 module_put(kvm->arch.kvm_ops->owner);
365 void kvm_arch_sync_events(struct kvm *kvm)
369 int kvm_dev_ioctl_check_extension(long ext)
373 * Should some of this be vm ioctl ? is it possible now ?
375 int hv_enabled = kvmppc_hv_ops ? 1 : 0;
379 case KVM_CAP_PPC_BOOKE_SREGS:
380 case KVM_CAP_PPC_BOOKE_WATCHDOG:
381 case KVM_CAP_PPC_EPR:
383 case KVM_CAP_PPC_SEGSTATE:
384 case KVM_CAP_PPC_HIOR:
385 case KVM_CAP_PPC_PAPR:
387 case KVM_CAP_PPC_UNSET_IRQ:
388 case KVM_CAP_PPC_IRQ_LEVEL:
389 case KVM_CAP_ENABLE_CAP:
390 case KVM_CAP_ENABLE_CAP_VM:
391 case KVM_CAP_ONE_REG:
392 case KVM_CAP_IOEVENTFD:
393 case KVM_CAP_DEVICE_CTRL:
396 case KVM_CAP_PPC_PAIRED_SINGLES:
397 case KVM_CAP_PPC_OSI:
398 case KVM_CAP_PPC_GET_PVINFO:
399 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
402 /* We support this only for PR */
405 #ifdef CONFIG_KVM_MMIO
406 case KVM_CAP_COALESCED_MMIO:
407 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
410 #ifdef CONFIG_KVM_MPIC
411 case KVM_CAP_IRQ_MPIC:
416 #ifdef CONFIG_PPC_BOOK3S_64
417 case KVM_CAP_SPAPR_TCE:
418 case KVM_CAP_PPC_ALLOC_HTAB:
419 case KVM_CAP_PPC_RTAS:
420 case KVM_CAP_PPC_FIXUP_HCALL:
421 case KVM_CAP_PPC_ENABLE_HCALL:
422 #ifdef CONFIG_KVM_XICS
423 case KVM_CAP_IRQ_XICS:
427 #endif /* CONFIG_PPC_BOOK3S_64 */
428 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
429 case KVM_CAP_PPC_SMT:
431 r = threads_per_subcore;
435 case KVM_CAP_PPC_RMA:
437 /* PPC970 requires an RMA */
438 if (r && cpu_has_feature(CPU_FTR_ARCH_201))
442 case KVM_CAP_SYNC_MMU:
443 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
445 r = cpu_has_feature(CPU_FTR_ARCH_206) ? 1 : 0;
448 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
454 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
455 case KVM_CAP_PPC_HTAB_FD:
459 case KVM_CAP_NR_VCPUS:
461 * Recommending a number of CPUs is somewhat arbitrary; we
462 * return the number of present CPUs for -HV (since a host
463 * will have secondary threads "offline"), and for other KVM
464 * implementations just count online CPUs.
467 r = num_present_cpus();
469 r = num_online_cpus();
471 case KVM_CAP_MAX_VCPUS:
474 #ifdef CONFIG_PPC_BOOK3S_64
475 case KVM_CAP_PPC_GET_SMMU_INFO:
487 long kvm_arch_dev_ioctl(struct file *filp,
488 unsigned int ioctl, unsigned long arg)
493 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
494 struct kvm_memory_slot *dont)
496 kvmppc_core_free_memslot(kvm, free, dont);
499 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
500 unsigned long npages)
502 return kvmppc_core_create_memslot(kvm, slot, npages);
505 void kvm_arch_memslots_updated(struct kvm *kvm)
509 int kvm_arch_prepare_memory_region(struct kvm *kvm,
510 struct kvm_memory_slot *memslot,
511 struct kvm_userspace_memory_region *mem,
512 enum kvm_mr_change change)
514 return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
517 void kvm_arch_commit_memory_region(struct kvm *kvm,
518 struct kvm_userspace_memory_region *mem,
519 const struct kvm_memory_slot *old,
520 enum kvm_mr_change change)
522 kvmppc_core_commit_memory_region(kvm, mem, old);
525 void kvm_arch_flush_shadow_all(struct kvm *kvm)
529 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
530 struct kvm_memory_slot *slot)
532 kvmppc_core_flush_memslot(kvm, slot);
535 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
537 struct kvm_vcpu *vcpu;
538 vcpu = kvmppc_core_vcpu_create(kvm, id);
540 vcpu->arch.wqp = &vcpu->wq;
541 kvmppc_create_vcpu_debugfs(vcpu, id);
546 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
551 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
553 /* Make sure we're not using the vcpu anymore */
554 hrtimer_cancel(&vcpu->arch.dec_timer);
555 tasklet_kill(&vcpu->arch.tasklet);
557 kvmppc_remove_vcpu_debugfs(vcpu);
559 switch (vcpu->arch.irq_type) {
560 case KVMPPC_IRQ_MPIC:
561 kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
563 case KVMPPC_IRQ_XICS:
564 kvmppc_xics_free_icp(vcpu);
568 kvmppc_core_vcpu_free(vcpu);
571 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
573 kvm_arch_vcpu_free(vcpu);
576 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
578 return kvmppc_core_pending_dec(vcpu);
582 * low level hrtimer wake routine. Because this runs in hardirq context
583 * we schedule a tasklet to do the real work.
585 enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
587 struct kvm_vcpu *vcpu;
589 vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
590 tasklet_schedule(&vcpu->arch.tasklet);
592 return HRTIMER_NORESTART;
595 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
599 hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
600 tasklet_init(&vcpu->arch.tasklet, kvmppc_decrementer_func, (ulong)vcpu);
601 vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
602 vcpu->arch.dec_expires = ~(u64)0;
604 #ifdef CONFIG_KVM_EXIT_TIMING
605 mutex_init(&vcpu->arch.exit_timing_lock);
607 ret = kvmppc_subarch_vcpu_init(vcpu);
611 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
613 kvmppc_mmu_destroy(vcpu);
614 kvmppc_subarch_vcpu_uninit(vcpu);
617 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
621 * vrsave (formerly usprg0) isn't used by Linux, but may
622 * be used by the guest.
624 * On non-booke this is associated with Altivec and
625 * is handled by code in book3s.c.
627 mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
629 kvmppc_core_vcpu_load(vcpu, cpu);
632 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
634 kvmppc_core_vcpu_put(vcpu);
636 vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
640 static void kvmppc_complete_dcr_load(struct kvm_vcpu *vcpu,
643 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, run->dcr.data);
646 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
649 u64 uninitialized_var(gpr);
651 if (run->mmio.len > sizeof(gpr)) {
652 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
656 if (vcpu->arch.mmio_is_bigendian) {
657 switch (run->mmio.len) {
658 case 8: gpr = *(u64 *)run->mmio.data; break;
659 case 4: gpr = *(u32 *)run->mmio.data; break;
660 case 2: gpr = *(u16 *)run->mmio.data; break;
661 case 1: gpr = *(u8 *)run->mmio.data; break;
664 /* Convert BE data from userland back to LE. */
665 switch (run->mmio.len) {
666 case 4: gpr = ld_le32((u32 *)run->mmio.data); break;
667 case 2: gpr = ld_le16((u16 *)run->mmio.data); break;
668 case 1: gpr = *(u8 *)run->mmio.data; break;
672 if (vcpu->arch.mmio_sign_extend) {
673 switch (run->mmio.len) {
688 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
690 switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
691 case KVM_MMIO_REG_GPR:
692 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
694 case KVM_MMIO_REG_FPR:
695 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
697 #ifdef CONFIG_PPC_BOOK3S
698 case KVM_MMIO_REG_QPR:
699 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
701 case KVM_MMIO_REG_FQPR:
702 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
703 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
711 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
712 unsigned int rt, unsigned int bytes,
713 int is_default_endian)
718 if (kvmppc_need_byteswap(vcpu)) {
719 /* Default endianness is "little endian". */
720 is_bigendian = !is_default_endian;
722 /* Default endianness is "big endian". */
723 is_bigendian = is_default_endian;
726 if (bytes > sizeof(run->mmio.data)) {
727 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
731 run->mmio.phys_addr = vcpu->arch.paddr_accessed;
732 run->mmio.len = bytes;
733 run->mmio.is_write = 0;
735 vcpu->arch.io_gpr = rt;
736 vcpu->arch.mmio_is_bigendian = is_bigendian;
737 vcpu->mmio_needed = 1;
738 vcpu->mmio_is_write = 0;
739 vcpu->arch.mmio_sign_extend = 0;
741 idx = srcu_read_lock(&vcpu->kvm->srcu);
743 ret = kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, run->mmio.phys_addr,
744 bytes, &run->mmio.data);
746 srcu_read_unlock(&vcpu->kvm->srcu, idx);
749 kvmppc_complete_mmio_load(vcpu, run);
750 vcpu->mmio_needed = 0;
754 return EMULATE_DO_MMIO;
756 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
758 /* Same as above, but sign extends */
759 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
760 unsigned int rt, unsigned int bytes,
761 int is_default_endian)
765 vcpu->arch.mmio_sign_extend = 1;
766 r = kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian);
771 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
772 u64 val, unsigned int bytes, int is_default_endian)
774 void *data = run->mmio.data;
778 if (kvmppc_need_byteswap(vcpu)) {
779 /* Default endianness is "little endian". */
780 is_bigendian = !is_default_endian;
782 /* Default endianness is "big endian". */
783 is_bigendian = is_default_endian;
786 if (bytes > sizeof(run->mmio.data)) {
787 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
791 run->mmio.phys_addr = vcpu->arch.paddr_accessed;
792 run->mmio.len = bytes;
793 run->mmio.is_write = 1;
794 vcpu->mmio_needed = 1;
795 vcpu->mmio_is_write = 1;
797 /* Store the value at the lowest bytes in 'data'. */
800 case 8: *(u64 *)data = val; break;
801 case 4: *(u32 *)data = val; break;
802 case 2: *(u16 *)data = val; break;
803 case 1: *(u8 *)data = val; break;
806 /* Store LE value into 'data'. */
808 case 4: st_le32(data, val); break;
809 case 2: st_le16(data, val); break;
810 case 1: *(u8 *)data = val; break;
814 idx = srcu_read_lock(&vcpu->kvm->srcu);
816 ret = kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, run->mmio.phys_addr,
817 bytes, &run->mmio.data);
819 srcu_read_unlock(&vcpu->kvm->srcu, idx);
822 vcpu->mmio_needed = 0;
826 return EMULATE_DO_MMIO;
828 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
830 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
835 if (vcpu->sigset_active)
836 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
838 if (vcpu->mmio_needed) {
839 if (!vcpu->mmio_is_write)
840 kvmppc_complete_mmio_load(vcpu, run);
841 vcpu->mmio_needed = 0;
842 } else if (vcpu->arch.dcr_needed) {
843 if (!vcpu->arch.dcr_is_write)
844 kvmppc_complete_dcr_load(vcpu, run);
845 vcpu->arch.dcr_needed = 0;
846 } else if (vcpu->arch.osi_needed) {
847 u64 *gprs = run->osi.gprs;
850 for (i = 0; i < 32; i++)
851 kvmppc_set_gpr(vcpu, i, gprs[i]);
852 vcpu->arch.osi_needed = 0;
853 } else if (vcpu->arch.hcall_needed) {
856 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
857 for (i = 0; i < 9; ++i)
858 kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
859 vcpu->arch.hcall_needed = 0;
861 } else if (vcpu->arch.epr_needed) {
862 kvmppc_set_epr(vcpu, run->epr.epr);
863 vcpu->arch.epr_needed = 0;
867 r = kvmppc_vcpu_run(run, vcpu);
869 if (vcpu->sigset_active)
870 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
875 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
877 if (irq->irq == KVM_INTERRUPT_UNSET) {
878 kvmppc_core_dequeue_external(vcpu);
882 kvmppc_core_queue_external(vcpu, irq);
889 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
890 struct kvm_enable_cap *cap)
898 case KVM_CAP_PPC_OSI:
900 vcpu->arch.osi_enabled = true;
902 case KVM_CAP_PPC_PAPR:
904 vcpu->arch.papr_enabled = true;
906 case KVM_CAP_PPC_EPR:
909 vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
911 vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
914 case KVM_CAP_PPC_BOOKE_WATCHDOG:
916 vcpu->arch.watchdog_enabled = true;
919 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
920 case KVM_CAP_SW_TLB: {
921 struct kvm_config_tlb cfg;
922 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
925 if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
928 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
932 #ifdef CONFIG_KVM_MPIC
933 case KVM_CAP_IRQ_MPIC: {
935 struct kvm_device *dev;
938 f = fdget(cap->args[0]);
943 dev = kvm_device_from_filp(f.file);
945 r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
951 #ifdef CONFIG_KVM_XICS
952 case KVM_CAP_IRQ_XICS: {
954 struct kvm_device *dev;
957 f = fdget(cap->args[0]);
962 dev = kvm_device_from_filp(f.file);
964 r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
969 #endif /* CONFIG_KVM_XICS */
976 r = kvmppc_sanity_check(vcpu);
981 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
982 struct kvm_mp_state *mp_state)
987 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
988 struct kvm_mp_state *mp_state)
993 long kvm_arch_vcpu_ioctl(struct file *filp,
994 unsigned int ioctl, unsigned long arg)
996 struct kvm_vcpu *vcpu = filp->private_data;
997 void __user *argp = (void __user *)arg;
1001 case KVM_INTERRUPT: {
1002 struct kvm_interrupt irq;
1004 if (copy_from_user(&irq, argp, sizeof(irq)))
1006 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1010 case KVM_ENABLE_CAP:
1012 struct kvm_enable_cap cap;
1014 if (copy_from_user(&cap, argp, sizeof(cap)))
1016 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
1020 case KVM_SET_ONE_REG:
1021 case KVM_GET_ONE_REG:
1023 struct kvm_one_reg reg;
1025 if (copy_from_user(®, argp, sizeof(reg)))
1027 if (ioctl == KVM_SET_ONE_REG)
1028 r = kvm_vcpu_ioctl_set_one_reg(vcpu, ®);
1030 r = kvm_vcpu_ioctl_get_one_reg(vcpu, ®);
1034 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1035 case KVM_DIRTY_TLB: {
1036 struct kvm_dirty_tlb dirty;
1038 if (copy_from_user(&dirty, argp, sizeof(dirty)))
1040 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
1052 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
1054 return VM_FAULT_SIGBUS;
1057 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
1059 u32 inst_nop = 0x60000000;
1060 #ifdef CONFIG_KVM_BOOKE_HV
1061 u32 inst_sc1 = 0x44000022;
1062 pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
1063 pvinfo->hcall[1] = cpu_to_be32(inst_nop);
1064 pvinfo->hcall[2] = cpu_to_be32(inst_nop);
1065 pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1067 u32 inst_lis = 0x3c000000;
1068 u32 inst_ori = 0x60000000;
1069 u32 inst_sc = 0x44000002;
1070 u32 inst_imm_mask = 0xffff;
1073 * The hypercall to get into KVM from within guest context is as
1076 * lis r0, r0, KVM_SC_MAGIC_R0@h
1077 * ori r0, KVM_SC_MAGIC_R0@l
1081 pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
1082 pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
1083 pvinfo->hcall[2] = cpu_to_be32(inst_sc);
1084 pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1087 pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
1092 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
1095 if (!irqchip_in_kernel(kvm))
1098 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1099 irq_event->irq, irq_event->level,
1105 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
1106 struct kvm_enable_cap *cap)
1114 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
1115 case KVM_CAP_PPC_ENABLE_HCALL: {
1116 unsigned long hcall = cap->args[0];
1119 if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
1122 if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
1125 set_bit(hcall / 4, kvm->arch.enabled_hcalls);
1127 clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
1140 long kvm_arch_vm_ioctl(struct file *filp,
1141 unsigned int ioctl, unsigned long arg)
1143 struct kvm *kvm __maybe_unused = filp->private_data;
1144 void __user *argp = (void __user *)arg;
1148 case KVM_PPC_GET_PVINFO: {
1149 struct kvm_ppc_pvinfo pvinfo;
1150 memset(&pvinfo, 0, sizeof(pvinfo));
1151 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
1152 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
1159 case KVM_ENABLE_CAP:
1161 struct kvm_enable_cap cap;
1163 if (copy_from_user(&cap, argp, sizeof(cap)))
1165 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
1168 #ifdef CONFIG_PPC_BOOK3S_64
1169 case KVM_CREATE_SPAPR_TCE: {
1170 struct kvm_create_spapr_tce create_tce;
1173 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
1175 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce);
1178 case KVM_PPC_GET_SMMU_INFO: {
1179 struct kvm_ppc_smmu_info info;
1180 struct kvm *kvm = filp->private_data;
1182 memset(&info, 0, sizeof(info));
1183 r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
1184 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
1188 case KVM_PPC_RTAS_DEFINE_TOKEN: {
1189 struct kvm *kvm = filp->private_data;
1191 r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
1195 struct kvm *kvm = filp->private_data;
1196 r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
1198 #else /* CONFIG_PPC_BOOK3S_64 */
1207 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
1208 static unsigned long nr_lpids;
1210 long kvmppc_alloc_lpid(void)
1215 lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
1216 if (lpid >= nr_lpids) {
1217 pr_err("%s: No LPIDs free\n", __func__);
1220 } while (test_and_set_bit(lpid, lpid_inuse));
1224 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
1226 void kvmppc_claim_lpid(long lpid)
1228 set_bit(lpid, lpid_inuse);
1230 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
1232 void kvmppc_free_lpid(long lpid)
1234 clear_bit(lpid, lpid_inuse);
1236 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
1238 void kvmppc_init_lpid(unsigned long nr_lpids_param)
1240 nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
1241 memset(lpid_inuse, 0, sizeof(lpid_inuse));
1243 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
1245 int kvm_arch_init(void *opaque)
1250 void kvm_arch_exit(void)