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>
44 #include "../mm/mmu_decl.h"
46 #define CREATE_TRACE_POINTS
49 struct kvmppc_ops
*kvmppc_hv_ops
;
50 EXPORT_SYMBOL_GPL(kvmppc_hv_ops
);
51 struct kvmppc_ops
*kvmppc_pr_ops
;
52 EXPORT_SYMBOL_GPL(kvmppc_pr_ops
);
55 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*v
)
57 return !!(v
->arch
.pending_exceptions
) ||
61 int kvm_arch_vcpu_should_kick(struct kvm_vcpu
*vcpu
)
67 * Common checks before entering the guest world. Call with interrupts
72 * == 1 if we're ready to go into guest state
73 * <= 0 if we need to go back to the host with return value
75 int kvmppc_prepare_to_enter(struct kvm_vcpu
*vcpu
)
79 WARN_ON(irqs_disabled());
90 if (signal_pending(current
)) {
91 kvmppc_account_exit(vcpu
, SIGNAL_EXITS
);
92 vcpu
->run
->exit_reason
= KVM_EXIT_INTR
;
97 vcpu
->mode
= IN_GUEST_MODE
;
100 * Reading vcpu->requests must happen after setting vcpu->mode,
101 * so we don't miss a request because the requester sees
102 * OUTSIDE_GUEST_MODE and assumes we'll be checking requests
103 * before next entering the guest (and thus doesn't IPI).
104 * This also orders the write to mode from any reads
105 * to the page tables done while the VCPU is running.
106 * Please see the comment in kvm_flush_remote_tlbs.
110 if (vcpu
->requests
) {
111 /* Make sure we process requests preemptable */
113 trace_kvm_check_requests(vcpu
);
114 r
= kvmppc_core_check_requests(vcpu
);
121 if (kvmppc_core_prepare_to_enter(vcpu
)) {
122 /* interrupts got enabled in between, so we
123 are back at square 1 */
127 guest_enter_irqoff();
135 EXPORT_SYMBOL_GPL(kvmppc_prepare_to_enter
);
137 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
138 static void kvmppc_swab_shared(struct kvm_vcpu
*vcpu
)
140 struct kvm_vcpu_arch_shared
*shared
= vcpu
->arch
.shared
;
143 shared
->sprg0
= swab64(shared
->sprg0
);
144 shared
->sprg1
= swab64(shared
->sprg1
);
145 shared
->sprg2
= swab64(shared
->sprg2
);
146 shared
->sprg3
= swab64(shared
->sprg3
);
147 shared
->srr0
= swab64(shared
->srr0
);
148 shared
->srr1
= swab64(shared
->srr1
);
149 shared
->dar
= swab64(shared
->dar
);
150 shared
->msr
= swab64(shared
->msr
);
151 shared
->dsisr
= swab32(shared
->dsisr
);
152 shared
->int_pending
= swab32(shared
->int_pending
);
153 for (i
= 0; i
< ARRAY_SIZE(shared
->sr
); i
++)
154 shared
->sr
[i
] = swab32(shared
->sr
[i
]);
158 int kvmppc_kvm_pv(struct kvm_vcpu
*vcpu
)
160 int nr
= kvmppc_get_gpr(vcpu
, 11);
162 unsigned long __maybe_unused param1
= kvmppc_get_gpr(vcpu
, 3);
163 unsigned long __maybe_unused param2
= kvmppc_get_gpr(vcpu
, 4);
164 unsigned long __maybe_unused param3
= kvmppc_get_gpr(vcpu
, 5);
165 unsigned long __maybe_unused param4
= kvmppc_get_gpr(vcpu
, 6);
166 unsigned long r2
= 0;
168 if (!(kvmppc_get_msr(vcpu
) & MSR_SF
)) {
170 param1
&= 0xffffffff;
171 param2
&= 0xffffffff;
172 param3
&= 0xffffffff;
173 param4
&= 0xffffffff;
177 case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE
):
179 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
180 /* Book3S can be little endian, find it out here */
181 int shared_big_endian
= true;
182 if (vcpu
->arch
.intr_msr
& MSR_LE
)
183 shared_big_endian
= false;
184 if (shared_big_endian
!= vcpu
->arch
.shared_big_endian
)
185 kvmppc_swab_shared(vcpu
);
186 vcpu
->arch
.shared_big_endian
= shared_big_endian
;
189 if (!(param2
& MAGIC_PAGE_FLAG_NOT_MAPPED_NX
)) {
191 * Older versions of the Linux magic page code had
192 * a bug where they would map their trampoline code
193 * NX. If that's the case, remove !PR NX capability.
195 vcpu
->arch
.disable_kernel_nx
= true;
196 kvm_make_request(KVM_REQ_TLB_FLUSH
, vcpu
);
199 vcpu
->arch
.magic_page_pa
= param1
& ~0xfffULL
;
200 vcpu
->arch
.magic_page_ea
= param2
& ~0xfffULL
;
202 #ifdef CONFIG_PPC_64K_PAGES
204 * Make sure our 4k magic page is in the same window of a 64k
205 * page within the guest and within the host's page.
207 if ((vcpu
->arch
.magic_page_pa
& 0xf000) !=
208 ((ulong
)vcpu
->arch
.shared
& 0xf000)) {
209 void *old_shared
= vcpu
->arch
.shared
;
210 ulong shared
= (ulong
)vcpu
->arch
.shared
;
214 shared
|= vcpu
->arch
.magic_page_pa
& 0xf000;
215 new_shared
= (void*)shared
;
216 memcpy(new_shared
, old_shared
, 0x1000);
217 vcpu
->arch
.shared
= new_shared
;
221 r2
= KVM_MAGIC_FEAT_SR
| KVM_MAGIC_FEAT_MAS0_TO_SPRG7
;
226 case KVM_HCALL_TOKEN(KVM_HC_FEATURES
):
228 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
229 r2
|= (1 << KVM_FEATURE_MAGIC_PAGE
);
232 /* Second return value is in r4 */
234 case EV_HCALL_TOKEN(EV_IDLE
):
236 kvm_vcpu_block(vcpu
);
237 clear_bit(KVM_REQ_UNHALT
, &vcpu
->requests
);
240 r
= EV_UNIMPLEMENTED
;
244 kvmppc_set_gpr(vcpu
, 4, r2
);
248 EXPORT_SYMBOL_GPL(kvmppc_kvm_pv
);
250 int kvmppc_sanity_check(struct kvm_vcpu
*vcpu
)
254 /* We have to know what CPU to virtualize */
258 /* PAPR only works with book3s_64 */
259 if ((vcpu
->arch
.cpu_type
!= KVM_CPU_3S_64
) && vcpu
->arch
.papr_enabled
)
262 /* HV KVM can only do PAPR mode for now */
263 if (!vcpu
->arch
.papr_enabled
&& is_kvmppc_hv_enabled(vcpu
->kvm
))
266 #ifdef CONFIG_KVM_BOOKE_HV
267 if (!cpu_has_feature(CPU_FTR_EMB_HV
))
275 return r
? 0 : -EINVAL
;
277 EXPORT_SYMBOL_GPL(kvmppc_sanity_check
);
279 int kvmppc_emulate_mmio(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
)
281 enum emulation_result er
;
284 er
= kvmppc_emulate_loadstore(vcpu
);
287 /* Future optimization: only reload non-volatiles if they were
288 * actually modified. */
294 case EMULATE_DO_MMIO
:
295 run
->exit_reason
= KVM_EXIT_MMIO
;
296 /* We must reload nonvolatiles because "update" load/store
297 * instructions modify register state. */
298 /* Future optimization: only reload non-volatiles if they were
299 * actually modified. */
306 kvmppc_get_last_inst(vcpu
, INST_GENERIC
, &last_inst
);
307 /* XXX Deliver Program interrupt to guest. */
308 pr_emerg("%s: emulation failed (%08x)\n", __func__
, last_inst
);
319 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio
);
321 int kvmppc_st(struct kvm_vcpu
*vcpu
, ulong
*eaddr
, int size
, void *ptr
,
324 ulong mp_pa
= vcpu
->arch
.magic_page_pa
& KVM_PAM
& PAGE_MASK
;
325 struct kvmppc_pte pte
;
330 r
= kvmppc_xlate(vcpu
, *eaddr
, data
? XLATE_DATA
: XLATE_INST
,
340 /* Magic page override */
341 if (kvmppc_supports_magic_page(vcpu
) && mp_pa
&&
342 ((pte
.raddr
& KVM_PAM
& PAGE_MASK
) == mp_pa
) &&
343 !(kvmppc_get_msr(vcpu
) & MSR_PR
)) {
344 void *magic
= vcpu
->arch
.shared
;
345 magic
+= pte
.eaddr
& 0xfff;
346 memcpy(magic
, ptr
, size
);
350 if (kvm_write_guest(vcpu
->kvm
, pte
.raddr
, ptr
, size
))
351 return EMULATE_DO_MMIO
;
355 EXPORT_SYMBOL_GPL(kvmppc_st
);
357 int kvmppc_ld(struct kvm_vcpu
*vcpu
, ulong
*eaddr
, int size
, void *ptr
,
360 ulong mp_pa
= vcpu
->arch
.magic_page_pa
& KVM_PAM
& PAGE_MASK
;
361 struct kvmppc_pte pte
;
366 rc
= kvmppc_xlate(vcpu
, *eaddr
, data
? XLATE_DATA
: XLATE_INST
,
376 if (!data
&& !pte
.may_execute
)
379 /* Magic page override */
380 if (kvmppc_supports_magic_page(vcpu
) && mp_pa
&&
381 ((pte
.raddr
& KVM_PAM
& PAGE_MASK
) == mp_pa
) &&
382 !(kvmppc_get_msr(vcpu
) & MSR_PR
)) {
383 void *magic
= vcpu
->arch
.shared
;
384 magic
+= pte
.eaddr
& 0xfff;
385 memcpy(ptr
, magic
, size
);
389 if (kvm_read_guest(vcpu
->kvm
, pte
.raddr
, ptr
, size
))
390 return EMULATE_DO_MMIO
;
394 EXPORT_SYMBOL_GPL(kvmppc_ld
);
396 int kvm_arch_hardware_enable(void)
401 int kvm_arch_hardware_setup(void)
406 void kvm_arch_check_processor_compat(void *rtn
)
408 *(int *)rtn
= kvmppc_core_check_processor_compat();
411 int kvm_arch_init_vm(struct kvm
*kvm
, unsigned long type
)
413 struct kvmppc_ops
*kvm_ops
= NULL
;
415 * if we have both HV and PR enabled, default is HV
419 kvm_ops
= kvmppc_hv_ops
;
421 kvm_ops
= kvmppc_pr_ops
;
424 } else if (type
== KVM_VM_PPC_HV
) {
427 kvm_ops
= kvmppc_hv_ops
;
428 } else if (type
== KVM_VM_PPC_PR
) {
431 kvm_ops
= kvmppc_pr_ops
;
435 if (kvm_ops
->owner
&& !try_module_get(kvm_ops
->owner
))
438 kvm
->arch
.kvm_ops
= kvm_ops
;
439 return kvmppc_core_init_vm(kvm
);
444 bool kvm_arch_has_vcpu_debugfs(void)
449 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu
*vcpu
)
454 void kvm_arch_destroy_vm(struct kvm
*kvm
)
457 struct kvm_vcpu
*vcpu
;
459 #ifdef CONFIG_KVM_XICS
461 * We call kick_all_cpus_sync() to ensure that all
462 * CPUs have executed any pending IPIs before we
463 * continue and free VCPUs structures below.
465 if (is_kvmppc_hv_enabled(kvm
))
466 kick_all_cpus_sync();
469 kvm_for_each_vcpu(i
, vcpu
, kvm
)
470 kvm_arch_vcpu_free(vcpu
);
472 mutex_lock(&kvm
->lock
);
473 for (i
= 0; i
< atomic_read(&kvm
->online_vcpus
); i
++)
474 kvm
->vcpus
[i
] = NULL
;
476 atomic_set(&kvm
->online_vcpus
, 0);
478 kvmppc_core_destroy_vm(kvm
);
480 mutex_unlock(&kvm
->lock
);
482 /* drop the module reference */
483 module_put(kvm
->arch
.kvm_ops
->owner
);
486 int kvm_vm_ioctl_check_extension(struct kvm
*kvm
, long ext
)
489 /* Assume we're using HV mode when the HV module is loaded */
490 int hv_enabled
= kvmppc_hv_ops
? 1 : 0;
494 * Hooray - we know which VM type we're running on. Depend on
495 * that rather than the guess above.
497 hv_enabled
= is_kvmppc_hv_enabled(kvm
);
502 case KVM_CAP_PPC_BOOKE_SREGS
:
503 case KVM_CAP_PPC_BOOKE_WATCHDOG
:
504 case KVM_CAP_PPC_EPR
:
506 case KVM_CAP_PPC_SEGSTATE
:
507 case KVM_CAP_PPC_HIOR
:
508 case KVM_CAP_PPC_PAPR
:
510 case KVM_CAP_PPC_UNSET_IRQ
:
511 case KVM_CAP_PPC_IRQ_LEVEL
:
512 case KVM_CAP_ENABLE_CAP
:
513 case KVM_CAP_ENABLE_CAP_VM
:
514 case KVM_CAP_ONE_REG
:
515 case KVM_CAP_IOEVENTFD
:
516 case KVM_CAP_DEVICE_CTRL
:
517 case KVM_CAP_IMMEDIATE_EXIT
:
520 case KVM_CAP_PPC_PAIRED_SINGLES
:
521 case KVM_CAP_PPC_OSI
:
522 case KVM_CAP_PPC_GET_PVINFO
:
523 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
526 /* We support this only for PR */
529 #ifdef CONFIG_KVM_MMIO
530 case KVM_CAP_COALESCED_MMIO
:
531 r
= KVM_COALESCED_MMIO_PAGE_OFFSET
;
534 #ifdef CONFIG_KVM_MPIC
535 case KVM_CAP_IRQ_MPIC
:
540 #ifdef CONFIG_PPC_BOOK3S_64
541 case KVM_CAP_SPAPR_TCE
:
542 case KVM_CAP_SPAPR_TCE_64
:
543 case KVM_CAP_PPC_RTAS
:
544 case KVM_CAP_PPC_FIXUP_HCALL
:
545 case KVM_CAP_PPC_ENABLE_HCALL
:
546 #ifdef CONFIG_KVM_XICS
547 case KVM_CAP_IRQ_XICS
:
552 case KVM_CAP_PPC_ALLOC_HTAB
:
555 #endif /* CONFIG_PPC_BOOK3S_64 */
556 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
557 case KVM_CAP_PPC_SMT
:
560 if (cpu_has_feature(CPU_FTR_ARCH_300
))
563 r
= threads_per_subcore
;
566 case KVM_CAP_PPC_RMA
:
569 case KVM_CAP_PPC_HWRNG
:
570 r
= kvmppc_hwrng_present();
572 case KVM_CAP_PPC_MMU_RADIX
:
573 r
= !!(hv_enabled
&& radix_enabled());
575 case KVM_CAP_PPC_MMU_HASH_V3
:
576 r
= !!(hv_enabled
&& !radix_enabled() &&
577 cpu_has_feature(CPU_FTR_ARCH_300
));
580 case KVM_CAP_SYNC_MMU
:
581 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
583 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
589 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
590 case KVM_CAP_PPC_HTAB_FD
:
594 case KVM_CAP_NR_VCPUS
:
596 * Recommending a number of CPUs is somewhat arbitrary; we
597 * return the number of present CPUs for -HV (since a host
598 * will have secondary threads "offline"), and for other KVM
599 * implementations just count online CPUs.
602 r
= num_present_cpus();
604 r
= num_online_cpus();
606 case KVM_CAP_NR_MEMSLOTS
:
607 r
= KVM_USER_MEM_SLOTS
;
609 case KVM_CAP_MAX_VCPUS
:
612 #ifdef CONFIG_PPC_BOOK3S_64
613 case KVM_CAP_PPC_GET_SMMU_INFO
:
616 case KVM_CAP_SPAPR_MULTITCE
:
619 case KVM_CAP_SPAPR_RESIZE_HPT
:
620 /* Disable this on POWER9 until code handles new HPTE format */
621 r
= !!hv_enabled
&& !cpu_has_feature(CPU_FTR_ARCH_300
);
624 case KVM_CAP_PPC_HTM
:
625 r
= cpu_has_feature(CPU_FTR_TM_COMP
) &&
626 is_kvmppc_hv_enabled(kvm
);
636 long kvm_arch_dev_ioctl(struct file
*filp
,
637 unsigned int ioctl
, unsigned long arg
)
642 void kvm_arch_free_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*free
,
643 struct kvm_memory_slot
*dont
)
645 kvmppc_core_free_memslot(kvm
, free
, dont
);
648 int kvm_arch_create_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*slot
,
649 unsigned long npages
)
651 return kvmppc_core_create_memslot(kvm
, slot
, npages
);
654 int kvm_arch_prepare_memory_region(struct kvm
*kvm
,
655 struct kvm_memory_slot
*memslot
,
656 const struct kvm_userspace_memory_region
*mem
,
657 enum kvm_mr_change change
)
659 return kvmppc_core_prepare_memory_region(kvm
, memslot
, mem
);
662 void kvm_arch_commit_memory_region(struct kvm
*kvm
,
663 const struct kvm_userspace_memory_region
*mem
,
664 const struct kvm_memory_slot
*old
,
665 const struct kvm_memory_slot
*new,
666 enum kvm_mr_change change
)
668 kvmppc_core_commit_memory_region(kvm
, mem
, old
, new);
671 void kvm_arch_flush_shadow_memslot(struct kvm
*kvm
,
672 struct kvm_memory_slot
*slot
)
674 kvmppc_core_flush_memslot(kvm
, slot
);
677 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
, unsigned int id
)
679 struct kvm_vcpu
*vcpu
;
680 vcpu
= kvmppc_core_vcpu_create(kvm
, id
);
682 vcpu
->arch
.wqp
= &vcpu
->wq
;
683 kvmppc_create_vcpu_debugfs(vcpu
, id
);
688 void kvm_arch_vcpu_postcreate(struct kvm_vcpu
*vcpu
)
692 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
694 /* Make sure we're not using the vcpu anymore */
695 hrtimer_cancel(&vcpu
->arch
.dec_timer
);
697 kvmppc_remove_vcpu_debugfs(vcpu
);
699 switch (vcpu
->arch
.irq_type
) {
700 case KVMPPC_IRQ_MPIC
:
701 kvmppc_mpic_disconnect_vcpu(vcpu
->arch
.mpic
, vcpu
);
703 case KVMPPC_IRQ_XICS
:
705 kvmppc_xive_cleanup_vcpu(vcpu
);
707 kvmppc_xics_free_icp(vcpu
);
711 kvmppc_core_vcpu_free(vcpu
);
714 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
716 kvm_arch_vcpu_free(vcpu
);
719 int kvm_cpu_has_pending_timer(struct kvm_vcpu
*vcpu
)
721 return kvmppc_core_pending_dec(vcpu
);
724 static enum hrtimer_restart
kvmppc_decrementer_wakeup(struct hrtimer
*timer
)
726 struct kvm_vcpu
*vcpu
;
728 vcpu
= container_of(timer
, struct kvm_vcpu
, arch
.dec_timer
);
729 kvmppc_decrementer_func(vcpu
);
731 return HRTIMER_NORESTART
;
734 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
738 hrtimer_init(&vcpu
->arch
.dec_timer
, CLOCK_REALTIME
, HRTIMER_MODE_ABS
);
739 vcpu
->arch
.dec_timer
.function
= kvmppc_decrementer_wakeup
;
740 vcpu
->arch
.dec_expires
= ~(u64
)0;
742 #ifdef CONFIG_KVM_EXIT_TIMING
743 mutex_init(&vcpu
->arch
.exit_timing_lock
);
745 ret
= kvmppc_subarch_vcpu_init(vcpu
);
749 void kvm_arch_vcpu_uninit(struct kvm_vcpu
*vcpu
)
751 kvmppc_mmu_destroy(vcpu
);
752 kvmppc_subarch_vcpu_uninit(vcpu
);
755 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
759 * vrsave (formerly usprg0) isn't used by Linux, but may
760 * be used by the guest.
762 * On non-booke this is associated with Altivec and
763 * is handled by code in book3s.c.
765 mtspr(SPRN_VRSAVE
, vcpu
->arch
.vrsave
);
767 kvmppc_core_vcpu_load(vcpu
, cpu
);
770 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
772 kvmppc_core_vcpu_put(vcpu
);
774 vcpu
->arch
.vrsave
= mfspr(SPRN_VRSAVE
);
779 * irq_bypass_add_producer and irq_bypass_del_producer are only
780 * useful if the architecture supports PCI passthrough.
781 * irq_bypass_stop and irq_bypass_start are not needed and so
782 * kvm_ops are not defined for them.
784 bool kvm_arch_has_irq_bypass(void)
786 return ((kvmppc_hv_ops
&& kvmppc_hv_ops
->irq_bypass_add_producer
) ||
787 (kvmppc_pr_ops
&& kvmppc_pr_ops
->irq_bypass_add_producer
));
790 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer
*cons
,
791 struct irq_bypass_producer
*prod
)
793 struct kvm_kernel_irqfd
*irqfd
=
794 container_of(cons
, struct kvm_kernel_irqfd
, consumer
);
795 struct kvm
*kvm
= irqfd
->kvm
;
797 if (kvm
->arch
.kvm_ops
->irq_bypass_add_producer
)
798 return kvm
->arch
.kvm_ops
->irq_bypass_add_producer(cons
, prod
);
803 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer
*cons
,
804 struct irq_bypass_producer
*prod
)
806 struct kvm_kernel_irqfd
*irqfd
=
807 container_of(cons
, struct kvm_kernel_irqfd
, consumer
);
808 struct kvm
*kvm
= irqfd
->kvm
;
810 if (kvm
->arch
.kvm_ops
->irq_bypass_del_producer
)
811 kvm
->arch
.kvm_ops
->irq_bypass_del_producer(cons
, prod
);
814 static void kvmppc_complete_mmio_load(struct kvm_vcpu
*vcpu
,
817 u64
uninitialized_var(gpr
);
819 if (run
->mmio
.len
> sizeof(gpr
)) {
820 printk(KERN_ERR
"bad MMIO length: %d\n", run
->mmio
.len
);
824 if (!vcpu
->arch
.mmio_host_swabbed
) {
825 switch (run
->mmio
.len
) {
826 case 8: gpr
= *(u64
*)run
->mmio
.data
; break;
827 case 4: gpr
= *(u32
*)run
->mmio
.data
; break;
828 case 2: gpr
= *(u16
*)run
->mmio
.data
; break;
829 case 1: gpr
= *(u8
*)run
->mmio
.data
; break;
832 switch (run
->mmio
.len
) {
833 case 8: gpr
= swab64(*(u64
*)run
->mmio
.data
); break;
834 case 4: gpr
= swab32(*(u32
*)run
->mmio
.data
); break;
835 case 2: gpr
= swab16(*(u16
*)run
->mmio
.data
); break;
836 case 1: gpr
= *(u8
*)run
->mmio
.data
; break;
840 if (vcpu
->arch
.mmio_sign_extend
) {
841 switch (run
->mmio
.len
) {
856 kvmppc_set_gpr(vcpu
, vcpu
->arch
.io_gpr
, gpr
);
858 switch (vcpu
->arch
.io_gpr
& KVM_MMIO_REG_EXT_MASK
) {
859 case KVM_MMIO_REG_GPR
:
860 kvmppc_set_gpr(vcpu
, vcpu
->arch
.io_gpr
, gpr
);
862 case KVM_MMIO_REG_FPR
:
863 VCPU_FPR(vcpu
, vcpu
->arch
.io_gpr
& KVM_MMIO_REG_MASK
) = gpr
;
865 #ifdef CONFIG_PPC_BOOK3S
866 case KVM_MMIO_REG_QPR
:
867 vcpu
->arch
.qpr
[vcpu
->arch
.io_gpr
& KVM_MMIO_REG_MASK
] = gpr
;
869 case KVM_MMIO_REG_FQPR
:
870 VCPU_FPR(vcpu
, vcpu
->arch
.io_gpr
& KVM_MMIO_REG_MASK
) = gpr
;
871 vcpu
->arch
.qpr
[vcpu
->arch
.io_gpr
& KVM_MMIO_REG_MASK
] = gpr
;
879 static int __kvmppc_handle_load(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
,
880 unsigned int rt
, unsigned int bytes
,
881 int is_default_endian
, int sign_extend
)
886 /* Pity C doesn't have a logical XOR operator */
887 if (kvmppc_need_byteswap(vcpu
)) {
888 host_swabbed
= is_default_endian
;
890 host_swabbed
= !is_default_endian
;
893 if (bytes
> sizeof(run
->mmio
.data
)) {
894 printk(KERN_ERR
"%s: bad MMIO length: %d\n", __func__
,
898 run
->mmio
.phys_addr
= vcpu
->arch
.paddr_accessed
;
899 run
->mmio
.len
= bytes
;
900 run
->mmio
.is_write
= 0;
902 vcpu
->arch
.io_gpr
= rt
;
903 vcpu
->arch
.mmio_host_swabbed
= host_swabbed
;
904 vcpu
->mmio_needed
= 1;
905 vcpu
->mmio_is_write
= 0;
906 vcpu
->arch
.mmio_sign_extend
= sign_extend
;
908 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
910 ret
= kvm_io_bus_read(vcpu
, KVM_MMIO_BUS
, run
->mmio
.phys_addr
,
911 bytes
, &run
->mmio
.data
);
913 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
916 kvmppc_complete_mmio_load(vcpu
, run
);
917 vcpu
->mmio_needed
= 0;
921 return EMULATE_DO_MMIO
;
924 int kvmppc_handle_load(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
,
925 unsigned int rt
, unsigned int bytes
,
926 int is_default_endian
)
928 return __kvmppc_handle_load(run
, vcpu
, rt
, bytes
, is_default_endian
, 0);
930 EXPORT_SYMBOL_GPL(kvmppc_handle_load
);
932 /* Same as above, but sign extends */
933 int kvmppc_handle_loads(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
,
934 unsigned int rt
, unsigned int bytes
,
935 int is_default_endian
)
937 return __kvmppc_handle_load(run
, vcpu
, rt
, bytes
, is_default_endian
, 1);
940 int kvmppc_handle_store(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
,
941 u64 val
, unsigned int bytes
, int is_default_endian
)
943 void *data
= run
->mmio
.data
;
947 /* Pity C doesn't have a logical XOR operator */
948 if (kvmppc_need_byteswap(vcpu
)) {
949 host_swabbed
= is_default_endian
;
951 host_swabbed
= !is_default_endian
;
954 if (bytes
> sizeof(run
->mmio
.data
)) {
955 printk(KERN_ERR
"%s: bad MMIO length: %d\n", __func__
,
959 run
->mmio
.phys_addr
= vcpu
->arch
.paddr_accessed
;
960 run
->mmio
.len
= bytes
;
961 run
->mmio
.is_write
= 1;
962 vcpu
->mmio_needed
= 1;
963 vcpu
->mmio_is_write
= 1;
965 /* Store the value at the lowest bytes in 'data'. */
968 case 8: *(u64
*)data
= val
; break;
969 case 4: *(u32
*)data
= val
; break;
970 case 2: *(u16
*)data
= val
; break;
971 case 1: *(u8
*)data
= val
; break;
975 case 8: *(u64
*)data
= swab64(val
); break;
976 case 4: *(u32
*)data
= swab32(val
); break;
977 case 2: *(u16
*)data
= swab16(val
); break;
978 case 1: *(u8
*)data
= val
; break;
982 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
984 ret
= kvm_io_bus_write(vcpu
, KVM_MMIO_BUS
, run
->mmio
.phys_addr
,
985 bytes
, &run
->mmio
.data
);
987 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
990 vcpu
->mmio_needed
= 0;
994 return EMULATE_DO_MMIO
;
996 EXPORT_SYMBOL_GPL(kvmppc_handle_store
);
998 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu
*vcpu
, struct kvm_one_reg
*reg
)
1001 union kvmppc_one_reg val
;
1004 size
= one_reg_size(reg
->id
);
1005 if (size
> sizeof(val
))
1008 r
= kvmppc_get_one_reg(vcpu
, reg
->id
, &val
);
1012 #ifdef CONFIG_ALTIVEC
1013 case KVM_REG_PPC_VR0
... KVM_REG_PPC_VR31
:
1014 if (!cpu_has_feature(CPU_FTR_ALTIVEC
)) {
1018 val
.vval
= vcpu
->arch
.vr
.vr
[reg
->id
- KVM_REG_PPC_VR0
];
1020 case KVM_REG_PPC_VSCR
:
1021 if (!cpu_has_feature(CPU_FTR_ALTIVEC
)) {
1025 val
= get_reg_val(reg
->id
, vcpu
->arch
.vr
.vscr
.u
[3]);
1027 case KVM_REG_PPC_VRSAVE
:
1028 val
= get_reg_val(reg
->id
, vcpu
->arch
.vrsave
);
1030 #endif /* CONFIG_ALTIVEC */
1040 if (copy_to_user((char __user
*)(unsigned long)reg
->addr
, &val
, size
))
1046 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu
*vcpu
, struct kvm_one_reg
*reg
)
1049 union kvmppc_one_reg val
;
1052 size
= one_reg_size(reg
->id
);
1053 if (size
> sizeof(val
))
1056 if (copy_from_user(&val
, (char __user
*)(unsigned long)reg
->addr
, size
))
1059 r
= kvmppc_set_one_reg(vcpu
, reg
->id
, &val
);
1063 #ifdef CONFIG_ALTIVEC
1064 case KVM_REG_PPC_VR0
... KVM_REG_PPC_VR31
:
1065 if (!cpu_has_feature(CPU_FTR_ALTIVEC
)) {
1069 vcpu
->arch
.vr
.vr
[reg
->id
- KVM_REG_PPC_VR0
] = val
.vval
;
1071 case KVM_REG_PPC_VSCR
:
1072 if (!cpu_has_feature(CPU_FTR_ALTIVEC
)) {
1076 vcpu
->arch
.vr
.vscr
.u
[3] = set_reg_val(reg
->id
, val
);
1078 case KVM_REG_PPC_VRSAVE
:
1079 if (!cpu_has_feature(CPU_FTR_ALTIVEC
)) {
1083 vcpu
->arch
.vrsave
= set_reg_val(reg
->id
, val
);
1085 #endif /* CONFIG_ALTIVEC */
1095 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
)
1100 if (vcpu
->sigset_active
)
1101 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
1103 if (vcpu
->mmio_needed
) {
1104 if (!vcpu
->mmio_is_write
)
1105 kvmppc_complete_mmio_load(vcpu
, run
);
1106 vcpu
->mmio_needed
= 0;
1107 } else if (vcpu
->arch
.osi_needed
) {
1108 u64
*gprs
= run
->osi
.gprs
;
1111 for (i
= 0; i
< 32; i
++)
1112 kvmppc_set_gpr(vcpu
, i
, gprs
[i
]);
1113 vcpu
->arch
.osi_needed
= 0;
1114 } else if (vcpu
->arch
.hcall_needed
) {
1117 kvmppc_set_gpr(vcpu
, 3, run
->papr_hcall
.ret
);
1118 for (i
= 0; i
< 9; ++i
)
1119 kvmppc_set_gpr(vcpu
, 4 + i
, run
->papr_hcall
.args
[i
]);
1120 vcpu
->arch
.hcall_needed
= 0;
1122 } else if (vcpu
->arch
.epr_needed
) {
1123 kvmppc_set_epr(vcpu
, run
->epr
.epr
);
1124 vcpu
->arch
.epr_needed
= 0;
1128 if (run
->immediate_exit
)
1131 r
= kvmppc_vcpu_run(run
, vcpu
);
1133 if (vcpu
->sigset_active
)
1134 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
1139 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu
*vcpu
, struct kvm_interrupt
*irq
)
1141 if (irq
->irq
== KVM_INTERRUPT_UNSET
) {
1142 kvmppc_core_dequeue_external(vcpu
);
1146 kvmppc_core_queue_external(vcpu
, irq
);
1148 kvm_vcpu_kick(vcpu
);
1153 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu
*vcpu
,
1154 struct kvm_enable_cap
*cap
)
1162 case KVM_CAP_PPC_OSI
:
1164 vcpu
->arch
.osi_enabled
= true;
1166 case KVM_CAP_PPC_PAPR
:
1168 vcpu
->arch
.papr_enabled
= true;
1170 case KVM_CAP_PPC_EPR
:
1173 vcpu
->arch
.epr_flags
|= KVMPPC_EPR_USER
;
1175 vcpu
->arch
.epr_flags
&= ~KVMPPC_EPR_USER
;
1178 case KVM_CAP_PPC_BOOKE_WATCHDOG
:
1180 vcpu
->arch
.watchdog_enabled
= true;
1183 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1184 case KVM_CAP_SW_TLB
: {
1185 struct kvm_config_tlb cfg
;
1186 void __user
*user_ptr
= (void __user
*)(uintptr_t)cap
->args
[0];
1189 if (copy_from_user(&cfg
, user_ptr
, sizeof(cfg
)))
1192 r
= kvm_vcpu_ioctl_config_tlb(vcpu
, &cfg
);
1196 #ifdef CONFIG_KVM_MPIC
1197 case KVM_CAP_IRQ_MPIC
: {
1199 struct kvm_device
*dev
;
1202 f
= fdget(cap
->args
[0]);
1207 dev
= kvm_device_from_filp(f
.file
);
1209 r
= kvmppc_mpic_connect_vcpu(dev
, vcpu
, cap
->args
[1]);
1215 #ifdef CONFIG_KVM_XICS
1216 case KVM_CAP_IRQ_XICS
: {
1218 struct kvm_device
*dev
;
1221 f
= fdget(cap
->args
[0]);
1226 dev
= kvm_device_from_filp(f
.file
);
1229 r
= kvmppc_xive_connect_vcpu(dev
, vcpu
, cap
->args
[1]);
1231 r
= kvmppc_xics_connect_vcpu(dev
, vcpu
, cap
->args
[1]);
1237 #endif /* CONFIG_KVM_XICS */
1244 r
= kvmppc_sanity_check(vcpu
);
1249 bool kvm_arch_intc_initialized(struct kvm
*kvm
)
1251 #ifdef CONFIG_KVM_MPIC
1255 #ifdef CONFIG_KVM_XICS
1256 if (kvm
->arch
.xics
|| kvm
->arch
.xive
)
1262 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
1263 struct kvm_mp_state
*mp_state
)
1268 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
1269 struct kvm_mp_state
*mp_state
)
1274 long kvm_arch_vcpu_ioctl(struct file
*filp
,
1275 unsigned int ioctl
, unsigned long arg
)
1277 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1278 void __user
*argp
= (void __user
*)arg
;
1282 case KVM_INTERRUPT
: {
1283 struct kvm_interrupt irq
;
1285 if (copy_from_user(&irq
, argp
, sizeof(irq
)))
1287 r
= kvm_vcpu_ioctl_interrupt(vcpu
, &irq
);
1291 case KVM_ENABLE_CAP
:
1293 struct kvm_enable_cap cap
;
1295 if (copy_from_user(&cap
, argp
, sizeof(cap
)))
1297 r
= kvm_vcpu_ioctl_enable_cap(vcpu
, &cap
);
1301 case KVM_SET_ONE_REG
:
1302 case KVM_GET_ONE_REG
:
1304 struct kvm_one_reg reg
;
1306 if (copy_from_user(®
, argp
, sizeof(reg
)))
1308 if (ioctl
== KVM_SET_ONE_REG
)
1309 r
= kvm_vcpu_ioctl_set_one_reg(vcpu
, ®
);
1311 r
= kvm_vcpu_ioctl_get_one_reg(vcpu
, ®
);
1315 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1316 case KVM_DIRTY_TLB
: {
1317 struct kvm_dirty_tlb dirty
;
1319 if (copy_from_user(&dirty
, argp
, sizeof(dirty
)))
1321 r
= kvm_vcpu_ioctl_dirty_tlb(vcpu
, &dirty
);
1333 int kvm_arch_vcpu_fault(struct kvm_vcpu
*vcpu
, struct vm_fault
*vmf
)
1335 return VM_FAULT_SIGBUS
;
1338 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo
*pvinfo
)
1340 u32 inst_nop
= 0x60000000;
1341 #ifdef CONFIG_KVM_BOOKE_HV
1342 u32 inst_sc1
= 0x44000022;
1343 pvinfo
->hcall
[0] = cpu_to_be32(inst_sc1
);
1344 pvinfo
->hcall
[1] = cpu_to_be32(inst_nop
);
1345 pvinfo
->hcall
[2] = cpu_to_be32(inst_nop
);
1346 pvinfo
->hcall
[3] = cpu_to_be32(inst_nop
);
1348 u32 inst_lis
= 0x3c000000;
1349 u32 inst_ori
= 0x60000000;
1350 u32 inst_sc
= 0x44000002;
1351 u32 inst_imm_mask
= 0xffff;
1354 * The hypercall to get into KVM from within guest context is as
1357 * lis r0, r0, KVM_SC_MAGIC_R0@h
1358 * ori r0, KVM_SC_MAGIC_R0@l
1362 pvinfo
->hcall
[0] = cpu_to_be32(inst_lis
| ((KVM_SC_MAGIC_R0
>> 16) & inst_imm_mask
));
1363 pvinfo
->hcall
[1] = cpu_to_be32(inst_ori
| (KVM_SC_MAGIC_R0
& inst_imm_mask
));
1364 pvinfo
->hcall
[2] = cpu_to_be32(inst_sc
);
1365 pvinfo
->hcall
[3] = cpu_to_be32(inst_nop
);
1368 pvinfo
->flags
= KVM_PPC_PVINFO_FLAGS_EV_IDLE
;
1373 int kvm_vm_ioctl_irq_line(struct kvm
*kvm
, struct kvm_irq_level
*irq_event
,
1376 if (!irqchip_in_kernel(kvm
))
1379 irq_event
->status
= kvm_set_irq(kvm
, KVM_USERSPACE_IRQ_SOURCE_ID
,
1380 irq_event
->irq
, irq_event
->level
,
1386 static int kvm_vm_ioctl_enable_cap(struct kvm
*kvm
,
1387 struct kvm_enable_cap
*cap
)
1395 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
1396 case KVM_CAP_PPC_ENABLE_HCALL
: {
1397 unsigned long hcall
= cap
->args
[0];
1400 if (hcall
> MAX_HCALL_OPCODE
|| (hcall
& 3) ||
1403 if (!kvmppc_book3s_hcall_implemented(kvm
, hcall
))
1406 set_bit(hcall
/ 4, kvm
->arch
.enabled_hcalls
);
1408 clear_bit(hcall
/ 4, kvm
->arch
.enabled_hcalls
);
1421 long kvm_arch_vm_ioctl(struct file
*filp
,
1422 unsigned int ioctl
, unsigned long arg
)
1424 struct kvm
*kvm __maybe_unused
= filp
->private_data
;
1425 void __user
*argp
= (void __user
*)arg
;
1429 case KVM_PPC_GET_PVINFO
: {
1430 struct kvm_ppc_pvinfo pvinfo
;
1431 memset(&pvinfo
, 0, sizeof(pvinfo
));
1432 r
= kvm_vm_ioctl_get_pvinfo(&pvinfo
);
1433 if (copy_to_user(argp
, &pvinfo
, sizeof(pvinfo
))) {
1440 case KVM_ENABLE_CAP
:
1442 struct kvm_enable_cap cap
;
1444 if (copy_from_user(&cap
, argp
, sizeof(cap
)))
1446 r
= kvm_vm_ioctl_enable_cap(kvm
, &cap
);
1449 #ifdef CONFIG_PPC_BOOK3S_64
1450 case KVM_CREATE_SPAPR_TCE_64
: {
1451 struct kvm_create_spapr_tce_64 create_tce_64
;
1454 if (copy_from_user(&create_tce_64
, argp
, sizeof(create_tce_64
)))
1456 if (create_tce_64
.flags
) {
1460 r
= kvm_vm_ioctl_create_spapr_tce(kvm
, &create_tce_64
);
1463 case KVM_CREATE_SPAPR_TCE
: {
1464 struct kvm_create_spapr_tce create_tce
;
1465 struct kvm_create_spapr_tce_64 create_tce_64
;
1468 if (copy_from_user(&create_tce
, argp
, sizeof(create_tce
)))
1471 create_tce_64
.liobn
= create_tce
.liobn
;
1472 create_tce_64
.page_shift
= IOMMU_PAGE_SHIFT_4K
;
1473 create_tce_64
.offset
= 0;
1474 create_tce_64
.size
= create_tce
.window_size
>>
1475 IOMMU_PAGE_SHIFT_4K
;
1476 create_tce_64
.flags
= 0;
1477 r
= kvm_vm_ioctl_create_spapr_tce(kvm
, &create_tce_64
);
1480 case KVM_PPC_GET_SMMU_INFO
: {
1481 struct kvm_ppc_smmu_info info
;
1482 struct kvm
*kvm
= filp
->private_data
;
1484 memset(&info
, 0, sizeof(info
));
1485 r
= kvm
->arch
.kvm_ops
->get_smmu_info(kvm
, &info
);
1486 if (r
>= 0 && copy_to_user(argp
, &info
, sizeof(info
)))
1490 case KVM_PPC_RTAS_DEFINE_TOKEN
: {
1491 struct kvm
*kvm
= filp
->private_data
;
1493 r
= kvm_vm_ioctl_rtas_define_token(kvm
, argp
);
1496 case KVM_PPC_CONFIGURE_V3_MMU
: {
1497 struct kvm
*kvm
= filp
->private_data
;
1498 struct kvm_ppc_mmuv3_cfg cfg
;
1501 if (!kvm
->arch
.kvm_ops
->configure_mmu
)
1504 if (copy_from_user(&cfg
, argp
, sizeof(cfg
)))
1506 r
= kvm
->arch
.kvm_ops
->configure_mmu(kvm
, &cfg
);
1509 case KVM_PPC_GET_RMMU_INFO
: {
1510 struct kvm
*kvm
= filp
->private_data
;
1511 struct kvm_ppc_rmmu_info info
;
1514 if (!kvm
->arch
.kvm_ops
->get_rmmu_info
)
1516 r
= kvm
->arch
.kvm_ops
->get_rmmu_info(kvm
, &info
);
1517 if (r
>= 0 && copy_to_user(argp
, &info
, sizeof(info
)))
1522 struct kvm
*kvm
= filp
->private_data
;
1523 r
= kvm
->arch
.kvm_ops
->arch_vm_ioctl(filp
, ioctl
, arg
);
1525 #else /* CONFIG_PPC_BOOK3S_64 */
1534 static unsigned long lpid_inuse
[BITS_TO_LONGS(KVMPPC_NR_LPIDS
)];
1535 static unsigned long nr_lpids
;
1537 long kvmppc_alloc_lpid(void)
1542 lpid
= find_first_zero_bit(lpid_inuse
, KVMPPC_NR_LPIDS
);
1543 if (lpid
>= nr_lpids
) {
1544 pr_err("%s: No LPIDs free\n", __func__
);
1547 } while (test_and_set_bit(lpid
, lpid_inuse
));
1551 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid
);
1553 void kvmppc_claim_lpid(long lpid
)
1555 set_bit(lpid
, lpid_inuse
);
1557 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid
);
1559 void kvmppc_free_lpid(long lpid
)
1561 clear_bit(lpid
, lpid_inuse
);
1563 EXPORT_SYMBOL_GPL(kvmppc_free_lpid
);
1565 void kvmppc_init_lpid(unsigned long nr_lpids_param
)
1567 nr_lpids
= min_t(unsigned long, KVMPPC_NR_LPIDS
, nr_lpids_param
);
1568 memset(lpid_inuse
, 0, sizeof(lpid_inuse
));
1570 EXPORT_SYMBOL_GPL(kvmppc_init_lpid
);
1572 int kvm_arch_init(void *opaque
)
1577 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr
);