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/sched/signal.h>
28 #include <linux/slab.h>
29 #include <linux/file.h>
30 #include <linux/module.h>
31 #include <linux/irqbypass.h>
32 #include <linux/kvm_irqfd.h>
33 #include <asm/cputable.h>
34 #include <linux/uaccess.h>
35 #include <asm/kvm_ppc.h>
36 #include <asm/tlbflush.h>
37 #include <asm/cputhreads.h>
38 #include <asm/irqflags.h>
39 #include <asm/iommu.h>
40 #include <asm/switch_to.h>
42 #ifdef CONFIG_PPC_PSERIES
43 #include <asm/hvcall.h>
44 #include <asm/plpar_wrappers.h>
49 #include "../mm/mmu_decl.h"
51 #define CREATE_TRACE_POINTS
54 struct kvmppc_ops *kvmppc_hv_ops;
55 EXPORT_SYMBOL_GPL(kvmppc_hv_ops);
56 struct kvmppc_ops *kvmppc_pr_ops;
57 EXPORT_SYMBOL_GPL(kvmppc_pr_ops);
60 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
62 return !!(v->arch.pending_exceptions) || kvm_request_pending(v);
65 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
70 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
76 * Common checks before entering the guest world. Call with interrupts
81 * == 1 if we're ready to go into guest state
82 * <= 0 if we need to go back to the host with return value
84 int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu)
88 WARN_ON(irqs_disabled());
99 if (signal_pending(current)) {
100 kvmppc_account_exit(vcpu, SIGNAL_EXITS);
101 vcpu->run->exit_reason = KVM_EXIT_INTR;
106 vcpu->mode = IN_GUEST_MODE;
109 * Reading vcpu->requests must happen after setting vcpu->mode,
110 * so we don't miss a request because the requester sees
111 * OUTSIDE_GUEST_MODE and assumes we'll be checking requests
112 * before next entering the guest (and thus doesn't IPI).
113 * This also orders the write to mode from any reads
114 * to the page tables done while the VCPU is running.
115 * Please see the comment in kvm_flush_remote_tlbs.
119 if (kvm_request_pending(vcpu)) {
120 /* Make sure we process requests preemptable */
122 trace_kvm_check_requests(vcpu);
123 r = kvmppc_core_check_requests(vcpu);
130 if (kvmppc_core_prepare_to_enter(vcpu)) {
131 /* interrupts got enabled in between, so we
132 are back at square 1 */
136 guest_enter_irqoff();
144 EXPORT_SYMBOL_GPL(kvmppc_prepare_to_enter);
146 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
147 static void kvmppc_swab_shared(struct kvm_vcpu *vcpu)
149 struct kvm_vcpu_arch_shared *shared = vcpu->arch.shared;
152 shared->sprg0 = swab64(shared->sprg0);
153 shared->sprg1 = swab64(shared->sprg1);
154 shared->sprg2 = swab64(shared->sprg2);
155 shared->sprg3 = swab64(shared->sprg3);
156 shared->srr0 = swab64(shared->srr0);
157 shared->srr1 = swab64(shared->srr1);
158 shared->dar = swab64(shared->dar);
159 shared->msr = swab64(shared->msr);
160 shared->dsisr = swab32(shared->dsisr);
161 shared->int_pending = swab32(shared->int_pending);
162 for (i = 0; i < ARRAY_SIZE(shared->sr); i++)
163 shared->sr[i] = swab32(shared->sr[i]);
167 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
169 int nr = kvmppc_get_gpr(vcpu, 11);
171 unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
172 unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
173 unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
174 unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
175 unsigned long r2 = 0;
177 if (!(kvmppc_get_msr(vcpu) & MSR_SF)) {
179 param1 &= 0xffffffff;
180 param2 &= 0xffffffff;
181 param3 &= 0xffffffff;
182 param4 &= 0xffffffff;
186 case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE):
188 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
189 /* Book3S can be little endian, find it out here */
190 int shared_big_endian = true;
191 if (vcpu->arch.intr_msr & MSR_LE)
192 shared_big_endian = false;
193 if (shared_big_endian != vcpu->arch.shared_big_endian)
194 kvmppc_swab_shared(vcpu);
195 vcpu->arch.shared_big_endian = shared_big_endian;
198 if (!(param2 & MAGIC_PAGE_FLAG_NOT_MAPPED_NX)) {
200 * Older versions of the Linux magic page code had
201 * a bug where they would map their trampoline code
202 * NX. If that's the case, remove !PR NX capability.
204 vcpu->arch.disable_kernel_nx = true;
205 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
208 vcpu->arch.magic_page_pa = param1 & ~0xfffULL;
209 vcpu->arch.magic_page_ea = param2 & ~0xfffULL;
211 #ifdef CONFIG_PPC_64K_PAGES
213 * Make sure our 4k magic page is in the same window of a 64k
214 * page within the guest and within the host's page.
216 if ((vcpu->arch.magic_page_pa & 0xf000) !=
217 ((ulong)vcpu->arch.shared & 0xf000)) {
218 void *old_shared = vcpu->arch.shared;
219 ulong shared = (ulong)vcpu->arch.shared;
223 shared |= vcpu->arch.magic_page_pa & 0xf000;
224 new_shared = (void*)shared;
225 memcpy(new_shared, old_shared, 0x1000);
226 vcpu->arch.shared = new_shared;
230 r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
235 case KVM_HCALL_TOKEN(KVM_HC_FEATURES):
237 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
238 r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
241 /* Second return value is in r4 */
243 case EV_HCALL_TOKEN(EV_IDLE):
245 kvm_vcpu_block(vcpu);
246 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
249 r = EV_UNIMPLEMENTED;
253 kvmppc_set_gpr(vcpu, 4, r2);
257 EXPORT_SYMBOL_GPL(kvmppc_kvm_pv);
259 int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
263 /* We have to know what CPU to virtualize */
267 /* PAPR only works with book3s_64 */
268 if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
271 /* HV KVM can only do PAPR mode for now */
272 if (!vcpu->arch.papr_enabled && is_kvmppc_hv_enabled(vcpu->kvm))
275 #ifdef CONFIG_KVM_BOOKE_HV
276 if (!cpu_has_feature(CPU_FTR_EMB_HV))
284 return r ? 0 : -EINVAL;
286 EXPORT_SYMBOL_GPL(kvmppc_sanity_check);
288 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
290 enum emulation_result er;
293 er = kvmppc_emulate_loadstore(vcpu);
296 /* Future optimization: only reload non-volatiles if they were
297 * actually modified. */
303 case EMULATE_DO_MMIO:
304 run->exit_reason = KVM_EXIT_MMIO;
305 /* We must reload nonvolatiles because "update" load/store
306 * instructions modify register state. */
307 /* Future optimization: only reload non-volatiles if they were
308 * actually modified. */
315 kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
316 /* XXX Deliver Program interrupt to guest. */
317 pr_emerg("%s: emulation failed (%08x)\n", __func__, last_inst);
328 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio);
330 int kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
333 ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
334 struct kvmppc_pte pte;
339 r = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
349 /* Magic page override */
350 if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
351 ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
352 !(kvmppc_get_msr(vcpu) & MSR_PR)) {
353 void *magic = vcpu->arch.shared;
354 magic += pte.eaddr & 0xfff;
355 memcpy(magic, ptr, size);
359 if (kvm_write_guest(vcpu->kvm, pte.raddr, ptr, size))
360 return EMULATE_DO_MMIO;
364 EXPORT_SYMBOL_GPL(kvmppc_st);
366 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
369 ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
370 struct kvmppc_pte pte;
375 rc = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
385 if (!data && !pte.may_execute)
388 /* Magic page override */
389 if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
390 ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
391 !(kvmppc_get_msr(vcpu) & MSR_PR)) {
392 void *magic = vcpu->arch.shared;
393 magic += pte.eaddr & 0xfff;
394 memcpy(ptr, magic, size);
398 if (kvm_read_guest(vcpu->kvm, pte.raddr, ptr, size))
399 return EMULATE_DO_MMIO;
403 EXPORT_SYMBOL_GPL(kvmppc_ld);
405 int kvm_arch_hardware_enable(void)
410 int kvm_arch_hardware_setup(void)
415 void kvm_arch_check_processor_compat(void *rtn)
417 *(int *)rtn = kvmppc_core_check_processor_compat();
420 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
422 struct kvmppc_ops *kvm_ops = NULL;
424 * if we have both HV and PR enabled, default is HV
428 kvm_ops = kvmppc_hv_ops;
430 kvm_ops = kvmppc_pr_ops;
433 } else if (type == KVM_VM_PPC_HV) {
436 kvm_ops = kvmppc_hv_ops;
437 } else if (type == KVM_VM_PPC_PR) {
440 kvm_ops = kvmppc_pr_ops;
444 if (kvm_ops->owner && !try_module_get(kvm_ops->owner))
447 kvm->arch.kvm_ops = kvm_ops;
448 return kvmppc_core_init_vm(kvm);
453 bool kvm_arch_has_vcpu_debugfs(void)
458 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
463 void kvm_arch_destroy_vm(struct kvm *kvm)
466 struct kvm_vcpu *vcpu;
468 #ifdef CONFIG_KVM_XICS
470 * We call kick_all_cpus_sync() to ensure that all
471 * CPUs have executed any pending IPIs before we
472 * continue and free VCPUs structures below.
474 if (is_kvmppc_hv_enabled(kvm))
475 kick_all_cpus_sync();
478 kvm_for_each_vcpu(i, vcpu, kvm)
479 kvm_arch_vcpu_free(vcpu);
481 mutex_lock(&kvm->lock);
482 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
483 kvm->vcpus[i] = NULL;
485 atomic_set(&kvm->online_vcpus, 0);
487 kvmppc_core_destroy_vm(kvm);
489 mutex_unlock(&kvm->lock);
491 /* drop the module reference */
492 module_put(kvm->arch.kvm_ops->owner);
495 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
498 /* Assume we're using HV mode when the HV module is loaded */
499 int hv_enabled = kvmppc_hv_ops ? 1 : 0;
503 * Hooray - we know which VM type we're running on. Depend on
504 * that rather than the guess above.
506 hv_enabled = is_kvmppc_hv_enabled(kvm);
511 case KVM_CAP_PPC_BOOKE_SREGS:
512 case KVM_CAP_PPC_BOOKE_WATCHDOG:
513 case KVM_CAP_PPC_EPR:
515 case KVM_CAP_PPC_SEGSTATE:
516 case KVM_CAP_PPC_HIOR:
517 case KVM_CAP_PPC_PAPR:
519 case KVM_CAP_PPC_UNSET_IRQ:
520 case KVM_CAP_PPC_IRQ_LEVEL:
521 case KVM_CAP_ENABLE_CAP:
522 case KVM_CAP_ENABLE_CAP_VM:
523 case KVM_CAP_ONE_REG:
524 case KVM_CAP_IOEVENTFD:
525 case KVM_CAP_DEVICE_CTRL:
526 case KVM_CAP_IMMEDIATE_EXIT:
529 case KVM_CAP_PPC_PAIRED_SINGLES:
530 case KVM_CAP_PPC_OSI:
531 case KVM_CAP_PPC_GET_PVINFO:
532 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
535 /* We support this only for PR */
538 #ifdef CONFIG_KVM_MPIC
539 case KVM_CAP_IRQ_MPIC:
544 #ifdef CONFIG_PPC_BOOK3S_64
545 case KVM_CAP_SPAPR_TCE:
546 case KVM_CAP_SPAPR_TCE_64:
548 case KVM_CAP_SPAPR_TCE_VFIO:
549 case KVM_CAP_PPC_RTAS:
550 case KVM_CAP_PPC_FIXUP_HCALL:
551 case KVM_CAP_PPC_ENABLE_HCALL:
552 #ifdef CONFIG_KVM_XICS
553 case KVM_CAP_IRQ_XICS:
555 case KVM_CAP_PPC_GET_CPU_CHAR:
559 case KVM_CAP_PPC_ALLOC_HTAB:
562 #endif /* CONFIG_PPC_BOOK3S_64 */
563 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
564 case KVM_CAP_PPC_SMT:
567 if (kvm->arch.emul_smt_mode > 1)
568 r = kvm->arch.emul_smt_mode;
570 r = kvm->arch.smt_mode;
571 } else if (hv_enabled) {
572 if (cpu_has_feature(CPU_FTR_ARCH_300))
575 r = threads_per_subcore;
578 case KVM_CAP_PPC_SMT_POSSIBLE:
581 if (!cpu_has_feature(CPU_FTR_ARCH_300))
582 r = ((threads_per_subcore << 1) - 1);
584 /* P9 can emulate dbells, so allow any mode */
588 case KVM_CAP_PPC_RMA:
591 case KVM_CAP_PPC_HWRNG:
592 r = kvmppc_hwrng_present();
594 case KVM_CAP_PPC_MMU_RADIX:
595 r = !!(hv_enabled && radix_enabled());
597 case KVM_CAP_PPC_MMU_HASH_V3:
598 r = !!(hv_enabled && cpu_has_feature(CPU_FTR_ARCH_300));
601 case KVM_CAP_SYNC_MMU:
602 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
604 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
610 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
611 case KVM_CAP_PPC_HTAB_FD:
615 case KVM_CAP_NR_VCPUS:
617 * Recommending a number of CPUs is somewhat arbitrary; we
618 * return the number of present CPUs for -HV (since a host
619 * will have secondary threads "offline"), and for other KVM
620 * implementations just count online CPUs.
623 r = num_present_cpus();
625 r = num_online_cpus();
627 case KVM_CAP_NR_MEMSLOTS:
628 r = KVM_USER_MEM_SLOTS;
630 case KVM_CAP_MAX_VCPUS:
633 #ifdef CONFIG_PPC_BOOK3S_64
634 case KVM_CAP_PPC_GET_SMMU_INFO:
637 case KVM_CAP_SPAPR_MULTITCE:
640 case KVM_CAP_SPAPR_RESIZE_HPT:
641 /* Disable this on POWER9 until code handles new HPTE format */
642 r = !!hv_enabled && !cpu_has_feature(CPU_FTR_ARCH_300);
645 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
646 case KVM_CAP_PPC_FWNMI:
650 case KVM_CAP_PPC_HTM:
652 (cur_cpu_spec->cpu_user_features2 & PPC_FEATURE2_HTM_COMP);
662 long kvm_arch_dev_ioctl(struct file *filp,
663 unsigned int ioctl, unsigned long arg)
668 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
669 struct kvm_memory_slot *dont)
671 kvmppc_core_free_memslot(kvm, free, dont);
674 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
675 unsigned long npages)
677 return kvmppc_core_create_memslot(kvm, slot, npages);
680 int kvm_arch_prepare_memory_region(struct kvm *kvm,
681 struct kvm_memory_slot *memslot,
682 const struct kvm_userspace_memory_region *mem,
683 enum kvm_mr_change change)
685 return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
688 void kvm_arch_commit_memory_region(struct kvm *kvm,
689 const struct kvm_userspace_memory_region *mem,
690 const struct kvm_memory_slot *old,
691 const struct kvm_memory_slot *new,
692 enum kvm_mr_change change)
694 kvmppc_core_commit_memory_region(kvm, mem, old, new);
697 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
698 struct kvm_memory_slot *slot)
700 kvmppc_core_flush_memslot(kvm, slot);
703 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
705 struct kvm_vcpu *vcpu;
706 vcpu = kvmppc_core_vcpu_create(kvm, id);
708 vcpu->arch.wqp = &vcpu->wq;
709 kvmppc_create_vcpu_debugfs(vcpu, id);
714 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
718 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
720 /* Make sure we're not using the vcpu anymore */
721 hrtimer_cancel(&vcpu->arch.dec_timer);
723 kvmppc_remove_vcpu_debugfs(vcpu);
725 switch (vcpu->arch.irq_type) {
726 case KVMPPC_IRQ_MPIC:
727 kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
729 case KVMPPC_IRQ_XICS:
731 kvmppc_xive_cleanup_vcpu(vcpu);
733 kvmppc_xics_free_icp(vcpu);
737 kvmppc_core_vcpu_free(vcpu);
740 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
742 kvm_arch_vcpu_free(vcpu);
745 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
747 return kvmppc_core_pending_dec(vcpu);
750 static enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
752 struct kvm_vcpu *vcpu;
754 vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
755 kvmppc_decrementer_func(vcpu);
757 return HRTIMER_NORESTART;
760 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
764 hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
765 vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
766 vcpu->arch.dec_expires = get_tb();
768 #ifdef CONFIG_KVM_EXIT_TIMING
769 mutex_init(&vcpu->arch.exit_timing_lock);
771 ret = kvmppc_subarch_vcpu_init(vcpu);
775 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
777 kvmppc_mmu_destroy(vcpu);
778 kvmppc_subarch_vcpu_uninit(vcpu);
781 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
785 * vrsave (formerly usprg0) isn't used by Linux, but may
786 * be used by the guest.
788 * On non-booke this is associated with Altivec and
789 * is handled by code in book3s.c.
791 mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
793 kvmppc_core_vcpu_load(vcpu, cpu);
796 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
798 kvmppc_core_vcpu_put(vcpu);
800 vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
805 * irq_bypass_add_producer and irq_bypass_del_producer are only
806 * useful if the architecture supports PCI passthrough.
807 * irq_bypass_stop and irq_bypass_start are not needed and so
808 * kvm_ops are not defined for them.
810 bool kvm_arch_has_irq_bypass(void)
812 return ((kvmppc_hv_ops && kvmppc_hv_ops->irq_bypass_add_producer) ||
813 (kvmppc_pr_ops && kvmppc_pr_ops->irq_bypass_add_producer));
816 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
817 struct irq_bypass_producer *prod)
819 struct kvm_kernel_irqfd *irqfd =
820 container_of(cons, struct kvm_kernel_irqfd, consumer);
821 struct kvm *kvm = irqfd->kvm;
823 if (kvm->arch.kvm_ops->irq_bypass_add_producer)
824 return kvm->arch.kvm_ops->irq_bypass_add_producer(cons, prod);
829 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
830 struct irq_bypass_producer *prod)
832 struct kvm_kernel_irqfd *irqfd =
833 container_of(cons, struct kvm_kernel_irqfd, consumer);
834 struct kvm *kvm = irqfd->kvm;
836 if (kvm->arch.kvm_ops->irq_bypass_del_producer)
837 kvm->arch.kvm_ops->irq_bypass_del_producer(cons, prod);
841 static inline int kvmppc_get_vsr_dword_offset(int index)
845 if ((index != 0) && (index != 1))
857 static inline int kvmppc_get_vsr_word_offset(int index)
861 if ((index > 3) || (index < 0))
872 static inline void kvmppc_set_vsr_dword(struct kvm_vcpu *vcpu,
875 union kvmppc_one_reg val;
876 int offset = kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
877 int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
882 if (vcpu->arch.mmio_vsx_tx_sx_enabled) {
883 val.vval = VCPU_VSX_VR(vcpu, index);
884 val.vsxval[offset] = gpr;
885 VCPU_VSX_VR(vcpu, index) = val.vval;
887 VCPU_VSX_FPR(vcpu, index, offset) = gpr;
891 static inline void kvmppc_set_vsr_dword_dump(struct kvm_vcpu *vcpu,
894 union kvmppc_one_reg val;
895 int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
897 if (vcpu->arch.mmio_vsx_tx_sx_enabled) {
898 val.vval = VCPU_VSX_VR(vcpu, index);
901 VCPU_VSX_VR(vcpu, index) = val.vval;
903 VCPU_VSX_FPR(vcpu, index, 0) = gpr;
904 VCPU_VSX_FPR(vcpu, index, 1) = gpr;
908 static inline void kvmppc_set_vsr_word(struct kvm_vcpu *vcpu,
911 union kvmppc_one_reg val;
912 int offset = kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
913 int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
914 int dword_offset, word_offset;
919 if (vcpu->arch.mmio_vsx_tx_sx_enabled) {
920 val.vval = VCPU_VSX_VR(vcpu, index);
921 val.vsx32val[offset] = gpr32;
922 VCPU_VSX_VR(vcpu, index) = val.vval;
924 dword_offset = offset / 2;
925 word_offset = offset % 2;
926 val.vsxval[0] = VCPU_VSX_FPR(vcpu, index, dword_offset);
927 val.vsx32val[word_offset] = gpr32;
928 VCPU_VSX_FPR(vcpu, index, dword_offset) = val.vsxval[0];
931 #endif /* CONFIG_VSX */
933 #ifdef CONFIG_PPC_FPU
934 static inline u64 sp_to_dp(u32 fprs)
940 asm ("lfs%U1%X1 0,%1; stfd%U0%X0 0,%0" : "=m" (fprd) : "m" (fprs)
946 static inline u32 dp_to_sp(u64 fprd)
952 asm ("lfd%U1%X1 0,%1; stfs%U0%X0 0,%0" : "=m" (fprs) : "m" (fprd)
959 #define sp_to_dp(x) (x)
960 #define dp_to_sp(x) (x)
961 #endif /* CONFIG_PPC_FPU */
963 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
966 u64 uninitialized_var(gpr);
968 if (run->mmio.len > sizeof(gpr)) {
969 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
973 if (!vcpu->arch.mmio_host_swabbed) {
974 switch (run->mmio.len) {
975 case 8: gpr = *(u64 *)run->mmio.data; break;
976 case 4: gpr = *(u32 *)run->mmio.data; break;
977 case 2: gpr = *(u16 *)run->mmio.data; break;
978 case 1: gpr = *(u8 *)run->mmio.data; break;
981 switch (run->mmio.len) {
982 case 8: gpr = swab64(*(u64 *)run->mmio.data); break;
983 case 4: gpr = swab32(*(u32 *)run->mmio.data); break;
984 case 2: gpr = swab16(*(u16 *)run->mmio.data); break;
985 case 1: gpr = *(u8 *)run->mmio.data; break;
989 /* conversion between single and double precision */
990 if ((vcpu->arch.mmio_sp64_extend) && (run->mmio.len == 4))
993 if (vcpu->arch.mmio_sign_extend) {
994 switch (run->mmio.len) {
1001 gpr = (s64)(s16)gpr;
1009 switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
1010 case KVM_MMIO_REG_GPR:
1011 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
1013 case KVM_MMIO_REG_FPR:
1014 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1016 #ifdef CONFIG_PPC_BOOK3S
1017 case KVM_MMIO_REG_QPR:
1018 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1020 case KVM_MMIO_REG_FQPR:
1021 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1022 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1026 case KVM_MMIO_REG_VSX:
1027 if (vcpu->arch.mmio_vsx_copy_type == KVMPPC_VSX_COPY_DWORD)
1028 kvmppc_set_vsr_dword(vcpu, gpr);
1029 else if (vcpu->arch.mmio_vsx_copy_type == KVMPPC_VSX_COPY_WORD)
1030 kvmppc_set_vsr_word(vcpu, gpr);
1031 else if (vcpu->arch.mmio_vsx_copy_type ==
1032 KVMPPC_VSX_COPY_DWORD_LOAD_DUMP)
1033 kvmppc_set_vsr_dword_dump(vcpu, gpr);
1041 static int __kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1042 unsigned int rt, unsigned int bytes,
1043 int is_default_endian, int sign_extend)
1048 /* Pity C doesn't have a logical XOR operator */
1049 if (kvmppc_need_byteswap(vcpu)) {
1050 host_swabbed = is_default_endian;
1052 host_swabbed = !is_default_endian;
1055 if (bytes > sizeof(run->mmio.data)) {
1056 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1060 run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1061 run->mmio.len = bytes;
1062 run->mmio.is_write = 0;
1064 vcpu->arch.io_gpr = rt;
1065 vcpu->arch.mmio_host_swabbed = host_swabbed;
1066 vcpu->mmio_needed = 1;
1067 vcpu->mmio_is_write = 0;
1068 vcpu->arch.mmio_sign_extend = sign_extend;
1070 idx = srcu_read_lock(&vcpu->kvm->srcu);
1072 ret = kvm_io_bus_read(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1073 bytes, &run->mmio.data);
1075 srcu_read_unlock(&vcpu->kvm->srcu, idx);
1078 kvmppc_complete_mmio_load(vcpu, run);
1079 vcpu->mmio_needed = 0;
1080 return EMULATE_DONE;
1083 return EMULATE_DO_MMIO;
1086 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1087 unsigned int rt, unsigned int bytes,
1088 int is_default_endian)
1090 return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 0);
1092 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
1094 /* Same as above, but sign extends */
1095 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
1096 unsigned int rt, unsigned int bytes,
1097 int is_default_endian)
1099 return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 1);
1103 int kvmppc_handle_vsx_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1104 unsigned int rt, unsigned int bytes,
1105 int is_default_endian, int mmio_sign_extend)
1107 enum emulation_result emulated = EMULATE_DONE;
1109 /* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1110 if (vcpu->arch.mmio_vsx_copy_nums > 4)
1111 return EMULATE_FAIL;
1113 while (vcpu->arch.mmio_vsx_copy_nums) {
1114 emulated = __kvmppc_handle_load(run, vcpu, rt, bytes,
1115 is_default_endian, mmio_sign_extend);
1117 if (emulated != EMULATE_DONE)
1120 vcpu->arch.paddr_accessed += run->mmio.len;
1122 vcpu->arch.mmio_vsx_copy_nums--;
1123 vcpu->arch.mmio_vsx_offset++;
1127 #endif /* CONFIG_VSX */
1129 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1130 u64 val, unsigned int bytes, int is_default_endian)
1132 void *data = run->mmio.data;
1136 /* Pity C doesn't have a logical XOR operator */
1137 if (kvmppc_need_byteswap(vcpu)) {
1138 host_swabbed = is_default_endian;
1140 host_swabbed = !is_default_endian;
1143 if (bytes > sizeof(run->mmio.data)) {
1144 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1148 run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1149 run->mmio.len = bytes;
1150 run->mmio.is_write = 1;
1151 vcpu->mmio_needed = 1;
1152 vcpu->mmio_is_write = 1;
1154 if ((vcpu->arch.mmio_sp64_extend) && (bytes == 4))
1155 val = dp_to_sp(val);
1157 /* Store the value at the lowest bytes in 'data'. */
1158 if (!host_swabbed) {
1160 case 8: *(u64 *)data = val; break;
1161 case 4: *(u32 *)data = val; break;
1162 case 2: *(u16 *)data = val; break;
1163 case 1: *(u8 *)data = val; break;
1167 case 8: *(u64 *)data = swab64(val); break;
1168 case 4: *(u32 *)data = swab32(val); break;
1169 case 2: *(u16 *)data = swab16(val); break;
1170 case 1: *(u8 *)data = val; break;
1174 idx = srcu_read_lock(&vcpu->kvm->srcu);
1176 ret = kvm_io_bus_write(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1177 bytes, &run->mmio.data);
1179 srcu_read_unlock(&vcpu->kvm->srcu, idx);
1182 vcpu->mmio_needed = 0;
1183 return EMULATE_DONE;
1186 return EMULATE_DO_MMIO;
1188 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
1191 static inline int kvmppc_get_vsr_data(struct kvm_vcpu *vcpu, int rs, u64 *val)
1193 u32 dword_offset, word_offset;
1194 union kvmppc_one_reg reg;
1196 int copy_type = vcpu->arch.mmio_vsx_copy_type;
1199 switch (copy_type) {
1200 case KVMPPC_VSX_COPY_DWORD:
1202 kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
1204 if (vsx_offset == -1) {
1209 if (!vcpu->arch.mmio_vsx_tx_sx_enabled) {
1210 *val = VCPU_VSX_FPR(vcpu, rs, vsx_offset);
1212 reg.vval = VCPU_VSX_VR(vcpu, rs);
1213 *val = reg.vsxval[vsx_offset];
1217 case KVMPPC_VSX_COPY_WORD:
1219 kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
1221 if (vsx_offset == -1) {
1226 if (!vcpu->arch.mmio_vsx_tx_sx_enabled) {
1227 dword_offset = vsx_offset / 2;
1228 word_offset = vsx_offset % 2;
1229 reg.vsxval[0] = VCPU_VSX_FPR(vcpu, rs, dword_offset);
1230 *val = reg.vsx32val[word_offset];
1232 reg.vval = VCPU_VSX_VR(vcpu, rs);
1233 *val = reg.vsx32val[vsx_offset];
1245 int kvmppc_handle_vsx_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1246 int rs, unsigned int bytes, int is_default_endian)
1249 enum emulation_result emulated = EMULATE_DONE;
1251 vcpu->arch.io_gpr = rs;
1253 /* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1254 if (vcpu->arch.mmio_vsx_copy_nums > 4)
1255 return EMULATE_FAIL;
1257 while (vcpu->arch.mmio_vsx_copy_nums) {
1258 if (kvmppc_get_vsr_data(vcpu, rs, &val) == -1)
1259 return EMULATE_FAIL;
1261 emulated = kvmppc_handle_store(run, vcpu,
1262 val, bytes, is_default_endian);
1264 if (emulated != EMULATE_DONE)
1267 vcpu->arch.paddr_accessed += run->mmio.len;
1269 vcpu->arch.mmio_vsx_copy_nums--;
1270 vcpu->arch.mmio_vsx_offset++;
1276 static int kvmppc_emulate_mmio_vsx_loadstore(struct kvm_vcpu *vcpu,
1277 struct kvm_run *run)
1279 enum emulation_result emulated = EMULATE_FAIL;
1282 vcpu->arch.paddr_accessed += run->mmio.len;
1284 if (!vcpu->mmio_is_write) {
1285 emulated = kvmppc_handle_vsx_load(run, vcpu, vcpu->arch.io_gpr,
1286 run->mmio.len, 1, vcpu->arch.mmio_sign_extend);
1288 emulated = kvmppc_handle_vsx_store(run, vcpu,
1289 vcpu->arch.io_gpr, run->mmio.len, 1);
1293 case EMULATE_DO_MMIO:
1294 run->exit_reason = KVM_EXIT_MMIO;
1298 pr_info("KVM: MMIO emulation failed (VSX repeat)\n");
1299 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1300 run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1309 #endif /* CONFIG_VSX */
1311 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1314 union kvmppc_one_reg val;
1317 size = one_reg_size(reg->id);
1318 if (size > sizeof(val))
1321 r = kvmppc_get_one_reg(vcpu, reg->id, &val);
1325 #ifdef CONFIG_ALTIVEC
1326 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1327 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1331 val.vval = vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0];
1333 case KVM_REG_PPC_VSCR:
1334 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1338 val = get_reg_val(reg->id, vcpu->arch.vr.vscr.u[3]);
1340 case KVM_REG_PPC_VRSAVE:
1341 val = get_reg_val(reg->id, vcpu->arch.vrsave);
1343 #endif /* CONFIG_ALTIVEC */
1353 if (copy_to_user((char __user *)(unsigned long)reg->addr, &val, size))
1359 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1362 union kvmppc_one_reg val;
1365 size = one_reg_size(reg->id);
1366 if (size > sizeof(val))
1369 if (copy_from_user(&val, (char __user *)(unsigned long)reg->addr, size))
1372 r = kvmppc_set_one_reg(vcpu, reg->id, &val);
1376 #ifdef CONFIG_ALTIVEC
1377 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1378 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1382 vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0] = val.vval;
1384 case KVM_REG_PPC_VSCR:
1385 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1389 vcpu->arch.vr.vscr.u[3] = set_reg_val(reg->id, val);
1391 case KVM_REG_PPC_VRSAVE:
1392 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1396 vcpu->arch.vrsave = set_reg_val(reg->id, val);
1398 #endif /* CONFIG_ALTIVEC */
1408 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
1414 if (vcpu->mmio_needed) {
1415 vcpu->mmio_needed = 0;
1416 if (!vcpu->mmio_is_write)
1417 kvmppc_complete_mmio_load(vcpu, run);
1419 if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1420 vcpu->arch.mmio_vsx_copy_nums--;
1421 vcpu->arch.mmio_vsx_offset++;
1424 if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1425 r = kvmppc_emulate_mmio_vsx_loadstore(vcpu, run);
1426 if (r == RESUME_HOST) {
1427 vcpu->mmio_needed = 1;
1432 } else if (vcpu->arch.osi_needed) {
1433 u64 *gprs = run->osi.gprs;
1436 for (i = 0; i < 32; i++)
1437 kvmppc_set_gpr(vcpu, i, gprs[i]);
1438 vcpu->arch.osi_needed = 0;
1439 } else if (vcpu->arch.hcall_needed) {
1442 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
1443 for (i = 0; i < 9; ++i)
1444 kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
1445 vcpu->arch.hcall_needed = 0;
1447 } else if (vcpu->arch.epr_needed) {
1448 kvmppc_set_epr(vcpu, run->epr.epr);
1449 vcpu->arch.epr_needed = 0;
1453 kvm_sigset_activate(vcpu);
1455 if (run->immediate_exit)
1458 r = kvmppc_vcpu_run(run, vcpu);
1460 kvm_sigset_deactivate(vcpu);
1467 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
1469 if (irq->irq == KVM_INTERRUPT_UNSET) {
1470 kvmppc_core_dequeue_external(vcpu);
1474 kvmppc_core_queue_external(vcpu, irq);
1476 kvm_vcpu_kick(vcpu);
1481 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
1482 struct kvm_enable_cap *cap)
1490 case KVM_CAP_PPC_OSI:
1492 vcpu->arch.osi_enabled = true;
1494 case KVM_CAP_PPC_PAPR:
1496 vcpu->arch.papr_enabled = true;
1498 case KVM_CAP_PPC_EPR:
1501 vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
1503 vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
1506 case KVM_CAP_PPC_BOOKE_WATCHDOG:
1508 vcpu->arch.watchdog_enabled = true;
1511 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1512 case KVM_CAP_SW_TLB: {
1513 struct kvm_config_tlb cfg;
1514 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
1517 if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
1520 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
1524 #ifdef CONFIG_KVM_MPIC
1525 case KVM_CAP_IRQ_MPIC: {
1527 struct kvm_device *dev;
1530 f = fdget(cap->args[0]);
1535 dev = kvm_device_from_filp(f.file);
1537 r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
1543 #ifdef CONFIG_KVM_XICS
1544 case KVM_CAP_IRQ_XICS: {
1546 struct kvm_device *dev;
1549 f = fdget(cap->args[0]);
1554 dev = kvm_device_from_filp(f.file);
1557 r = kvmppc_xive_connect_vcpu(dev, vcpu, cap->args[1]);
1559 r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
1565 #endif /* CONFIG_KVM_XICS */
1566 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1567 case KVM_CAP_PPC_FWNMI:
1569 if (!is_kvmppc_hv_enabled(vcpu->kvm))
1572 vcpu->kvm->arch.fwnmi_enabled = true;
1574 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
1581 r = kvmppc_sanity_check(vcpu);
1586 bool kvm_arch_intc_initialized(struct kvm *kvm)
1588 #ifdef CONFIG_KVM_MPIC
1592 #ifdef CONFIG_KVM_XICS
1593 if (kvm->arch.xics || kvm->arch.xive)
1599 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1600 struct kvm_mp_state *mp_state)
1605 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1606 struct kvm_mp_state *mp_state)
1611 long kvm_arch_vcpu_async_ioctl(struct file *filp,
1612 unsigned int ioctl, unsigned long arg)
1614 struct kvm_vcpu *vcpu = filp->private_data;
1615 void __user *argp = (void __user *)arg;
1617 if (ioctl == KVM_INTERRUPT) {
1618 struct kvm_interrupt irq;
1619 if (copy_from_user(&irq, argp, sizeof(irq)))
1621 return kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1623 return -ENOIOCTLCMD;
1626 long kvm_arch_vcpu_ioctl(struct file *filp,
1627 unsigned int ioctl, unsigned long arg)
1629 struct kvm_vcpu *vcpu = filp->private_data;
1630 void __user *argp = (void __user *)arg;
1636 case KVM_ENABLE_CAP:
1638 struct kvm_enable_cap cap;
1640 if (copy_from_user(&cap, argp, sizeof(cap)))
1642 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
1646 case KVM_SET_ONE_REG:
1647 case KVM_GET_ONE_REG:
1649 struct kvm_one_reg reg;
1651 if (copy_from_user(®, argp, sizeof(reg)))
1653 if (ioctl == KVM_SET_ONE_REG)
1654 r = kvm_vcpu_ioctl_set_one_reg(vcpu, ®);
1656 r = kvm_vcpu_ioctl_get_one_reg(vcpu, ®);
1660 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1661 case KVM_DIRTY_TLB: {
1662 struct kvm_dirty_tlb dirty;
1664 if (copy_from_user(&dirty, argp, sizeof(dirty)))
1666 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
1679 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
1681 return VM_FAULT_SIGBUS;
1684 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
1686 u32 inst_nop = 0x60000000;
1687 #ifdef CONFIG_KVM_BOOKE_HV
1688 u32 inst_sc1 = 0x44000022;
1689 pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
1690 pvinfo->hcall[1] = cpu_to_be32(inst_nop);
1691 pvinfo->hcall[2] = cpu_to_be32(inst_nop);
1692 pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1694 u32 inst_lis = 0x3c000000;
1695 u32 inst_ori = 0x60000000;
1696 u32 inst_sc = 0x44000002;
1697 u32 inst_imm_mask = 0xffff;
1700 * The hypercall to get into KVM from within guest context is as
1703 * lis r0, r0, KVM_SC_MAGIC_R0@h
1704 * ori r0, KVM_SC_MAGIC_R0@l
1708 pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
1709 pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
1710 pvinfo->hcall[2] = cpu_to_be32(inst_sc);
1711 pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1714 pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
1719 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
1722 if (!irqchip_in_kernel(kvm))
1725 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1726 irq_event->irq, irq_event->level,
1732 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
1733 struct kvm_enable_cap *cap)
1741 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
1742 case KVM_CAP_PPC_ENABLE_HCALL: {
1743 unsigned long hcall = cap->args[0];
1746 if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
1749 if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
1752 set_bit(hcall / 4, kvm->arch.enabled_hcalls);
1754 clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
1758 case KVM_CAP_PPC_SMT: {
1759 unsigned long mode = cap->args[0];
1760 unsigned long flags = cap->args[1];
1763 if (kvm->arch.kvm_ops->set_smt_mode)
1764 r = kvm->arch.kvm_ops->set_smt_mode(kvm, mode, flags);
1776 #ifdef CONFIG_PPC_BOOK3S_64
1778 * These functions check whether the underlying hardware is safe
1779 * against attacks based on observing the effects of speculatively
1780 * executed instructions, and whether it supplies instructions for
1781 * use in workarounds. The information comes from firmware, either
1782 * via the device tree on powernv platforms or from an hcall on
1783 * pseries platforms.
1785 #ifdef CONFIG_PPC_PSERIES
1786 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
1788 struct h_cpu_char_result c;
1791 if (!machine_is(pseries))
1794 rc = plpar_get_cpu_characteristics(&c);
1795 if (rc == H_SUCCESS) {
1796 cp->character = c.character;
1797 cp->behaviour = c.behaviour;
1798 cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
1799 KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
1800 KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
1801 KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
1802 KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
1803 KVM_PPC_CPU_CHAR_BR_HINT_HONOURED |
1804 KVM_PPC_CPU_CHAR_MTTRIG_THR_RECONF |
1805 KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
1806 cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
1807 KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
1808 KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
1813 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
1819 static inline bool have_fw_feat(struct device_node *fw_features,
1820 const char *state, const char *name)
1822 struct device_node *np;
1825 np = of_get_child_by_name(fw_features, name);
1827 r = of_property_read_bool(np, state);
1833 static int kvmppc_get_cpu_char(struct kvm_ppc_cpu_char *cp)
1835 struct device_node *np, *fw_features;
1838 memset(cp, 0, sizeof(*cp));
1839 r = pseries_get_cpu_char(cp);
1843 np = of_find_node_by_name(NULL, "ibm,opal");
1845 fw_features = of_get_child_by_name(np, "fw-features");
1849 if (have_fw_feat(fw_features, "enabled",
1850 "inst-spec-barrier-ori31,31,0"))
1851 cp->character |= KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31;
1852 if (have_fw_feat(fw_features, "enabled",
1853 "fw-bcctrl-serialized"))
1854 cp->character |= KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED;
1855 if (have_fw_feat(fw_features, "enabled",
1856 "inst-l1d-flush-ori30,30,0"))
1857 cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30;
1858 if (have_fw_feat(fw_features, "enabled",
1859 "inst-l1d-flush-trig2"))
1860 cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2;
1861 if (have_fw_feat(fw_features, "enabled",
1862 "fw-l1d-thread-split"))
1863 cp->character |= KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV;
1864 if (have_fw_feat(fw_features, "enabled",
1865 "fw-count-cache-disabled"))
1866 cp->character |= KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
1867 cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
1868 KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
1869 KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
1870 KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
1871 KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
1872 KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
1874 if (have_fw_feat(fw_features, "enabled",
1875 "speculation-policy-favor-security"))
1876 cp->behaviour |= KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY;
1877 if (!have_fw_feat(fw_features, "disabled",
1878 "needs-l1d-flush-msr-pr-0-to-1"))
1879 cp->behaviour |= KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR;
1880 if (!have_fw_feat(fw_features, "disabled",
1881 "needs-spec-barrier-for-bound-checks"))
1882 cp->behaviour |= KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
1883 cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
1884 KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
1885 KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
1887 of_node_put(fw_features);
1894 long kvm_arch_vm_ioctl(struct file *filp,
1895 unsigned int ioctl, unsigned long arg)
1897 struct kvm *kvm __maybe_unused = filp->private_data;
1898 void __user *argp = (void __user *)arg;
1902 case KVM_PPC_GET_PVINFO: {
1903 struct kvm_ppc_pvinfo pvinfo;
1904 memset(&pvinfo, 0, sizeof(pvinfo));
1905 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
1906 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
1913 case KVM_ENABLE_CAP:
1915 struct kvm_enable_cap cap;
1917 if (copy_from_user(&cap, argp, sizeof(cap)))
1919 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
1922 #ifdef CONFIG_SPAPR_TCE_IOMMU
1923 case KVM_CREATE_SPAPR_TCE_64: {
1924 struct kvm_create_spapr_tce_64 create_tce_64;
1927 if (copy_from_user(&create_tce_64, argp, sizeof(create_tce_64)))
1929 if (create_tce_64.flags) {
1933 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
1936 case KVM_CREATE_SPAPR_TCE: {
1937 struct kvm_create_spapr_tce create_tce;
1938 struct kvm_create_spapr_tce_64 create_tce_64;
1941 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
1944 create_tce_64.liobn = create_tce.liobn;
1945 create_tce_64.page_shift = IOMMU_PAGE_SHIFT_4K;
1946 create_tce_64.offset = 0;
1947 create_tce_64.size = create_tce.window_size >>
1948 IOMMU_PAGE_SHIFT_4K;
1949 create_tce_64.flags = 0;
1950 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
1954 #ifdef CONFIG_PPC_BOOK3S_64
1955 case KVM_PPC_GET_SMMU_INFO: {
1956 struct kvm_ppc_smmu_info info;
1957 struct kvm *kvm = filp->private_data;
1959 memset(&info, 0, sizeof(info));
1960 r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
1961 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
1965 case KVM_PPC_RTAS_DEFINE_TOKEN: {
1966 struct kvm *kvm = filp->private_data;
1968 r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
1971 case KVM_PPC_CONFIGURE_V3_MMU: {
1972 struct kvm *kvm = filp->private_data;
1973 struct kvm_ppc_mmuv3_cfg cfg;
1976 if (!kvm->arch.kvm_ops->configure_mmu)
1979 if (copy_from_user(&cfg, argp, sizeof(cfg)))
1981 r = kvm->arch.kvm_ops->configure_mmu(kvm, &cfg);
1984 case KVM_PPC_GET_RMMU_INFO: {
1985 struct kvm *kvm = filp->private_data;
1986 struct kvm_ppc_rmmu_info info;
1989 if (!kvm->arch.kvm_ops->get_rmmu_info)
1991 r = kvm->arch.kvm_ops->get_rmmu_info(kvm, &info);
1992 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
1996 case KVM_PPC_GET_CPU_CHAR: {
1997 struct kvm_ppc_cpu_char cpuchar;
1999 r = kvmppc_get_cpu_char(&cpuchar);
2000 if (r >= 0 && copy_to_user(argp, &cpuchar, sizeof(cpuchar)))
2005 struct kvm *kvm = filp->private_data;
2006 r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
2008 #else /* CONFIG_PPC_BOOK3S_64 */
2017 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
2018 static unsigned long nr_lpids;
2020 long kvmppc_alloc_lpid(void)
2025 lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
2026 if (lpid >= nr_lpids) {
2027 pr_err("%s: No LPIDs free\n", __func__);
2030 } while (test_and_set_bit(lpid, lpid_inuse));
2034 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
2036 void kvmppc_claim_lpid(long lpid)
2038 set_bit(lpid, lpid_inuse);
2040 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
2042 void kvmppc_free_lpid(long lpid)
2044 clear_bit(lpid, lpid_inuse);
2046 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
2048 void kvmppc_init_lpid(unsigned long nr_lpids_param)
2050 nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
2051 memset(lpid_inuse, 0, sizeof(lpid_inuse));
2053 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
2055 int kvm_arch_init(void *opaque)
2060 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr);