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 <linux/irqbypass.h>
31 #include <linux/kvm_irqfd.h>
32 #include <asm/cputable.h>
33 #include <linux/uaccess.h>
34 #include <asm/kvm_ppc.h>
35 #include <asm/tlbflush.h>
36 #include <asm/cputhreads.h>
37 #include <asm/irqflags.h>
38 #include <asm/iommu.h>
41 #include "../mm/mmu_decl.h"
43 #define CREATE_TRACE_POINTS
46 struct kvmppc_ops
*kvmppc_hv_ops
;
47 EXPORT_SYMBOL_GPL(kvmppc_hv_ops
);
48 struct kvmppc_ops
*kvmppc_pr_ops
;
49 EXPORT_SYMBOL_GPL(kvmppc_pr_ops
);
52 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*v
)
54 return !!(v
->arch
.pending_exceptions
) ||
58 int kvm_arch_vcpu_should_kick(struct kvm_vcpu
*vcpu
)
64 * Common checks before entering the guest world. Call with interrupts
69 * == 1 if we're ready to go into guest state
70 * <= 0 if we need to go back to the host with return value
72 int kvmppc_prepare_to_enter(struct kvm_vcpu
*vcpu
)
76 WARN_ON(irqs_disabled());
87 if (signal_pending(current
)) {
88 kvmppc_account_exit(vcpu
, SIGNAL_EXITS
);
89 vcpu
->run
->exit_reason
= KVM_EXIT_INTR
;
94 vcpu
->mode
= IN_GUEST_MODE
;
97 * Reading vcpu->requests must happen after setting vcpu->mode,
98 * so we don't miss a request because the requester sees
99 * OUTSIDE_GUEST_MODE and assumes we'll be checking requests
100 * before next entering the guest (and thus doesn't IPI).
101 * This also orders the write to mode from any reads
102 * to the page tables done while the VCPU is running.
103 * Please see the comment in kvm_flush_remote_tlbs.
107 if (vcpu
->requests
) {
108 /* Make sure we process requests preemptable */
110 trace_kvm_check_requests(vcpu
);
111 r
= kvmppc_core_check_requests(vcpu
);
118 if (kvmppc_core_prepare_to_enter(vcpu
)) {
119 /* interrupts got enabled in between, so we
120 are back at square 1 */
124 guest_enter_irqoff();
132 EXPORT_SYMBOL_GPL(kvmppc_prepare_to_enter
);
134 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
135 static void kvmppc_swab_shared(struct kvm_vcpu
*vcpu
)
137 struct kvm_vcpu_arch_shared
*shared
= vcpu
->arch
.shared
;
140 shared
->sprg0
= swab64(shared
->sprg0
);
141 shared
->sprg1
= swab64(shared
->sprg1
);
142 shared
->sprg2
= swab64(shared
->sprg2
);
143 shared
->sprg3
= swab64(shared
->sprg3
);
144 shared
->srr0
= swab64(shared
->srr0
);
145 shared
->srr1
= swab64(shared
->srr1
);
146 shared
->dar
= swab64(shared
->dar
);
147 shared
->msr
= swab64(shared
->msr
);
148 shared
->dsisr
= swab32(shared
->dsisr
);
149 shared
->int_pending
= swab32(shared
->int_pending
);
150 for (i
= 0; i
< ARRAY_SIZE(shared
->sr
); i
++)
151 shared
->sr
[i
] = swab32(shared
->sr
[i
]);
155 int kvmppc_kvm_pv(struct kvm_vcpu
*vcpu
)
157 int nr
= kvmppc_get_gpr(vcpu
, 11);
159 unsigned long __maybe_unused param1
= kvmppc_get_gpr(vcpu
, 3);
160 unsigned long __maybe_unused param2
= kvmppc_get_gpr(vcpu
, 4);
161 unsigned long __maybe_unused param3
= kvmppc_get_gpr(vcpu
, 5);
162 unsigned long __maybe_unused param4
= kvmppc_get_gpr(vcpu
, 6);
163 unsigned long r2
= 0;
165 if (!(kvmppc_get_msr(vcpu
) & MSR_SF
)) {
167 param1
&= 0xffffffff;
168 param2
&= 0xffffffff;
169 param3
&= 0xffffffff;
170 param4
&= 0xffffffff;
174 case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE
):
176 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
177 /* Book3S can be little endian, find it out here */
178 int shared_big_endian
= true;
179 if (vcpu
->arch
.intr_msr
& MSR_LE
)
180 shared_big_endian
= false;
181 if (shared_big_endian
!= vcpu
->arch
.shared_big_endian
)
182 kvmppc_swab_shared(vcpu
);
183 vcpu
->arch
.shared_big_endian
= shared_big_endian
;
186 if (!(param2
& MAGIC_PAGE_FLAG_NOT_MAPPED_NX
)) {
188 * Older versions of the Linux magic page code had
189 * a bug where they would map their trampoline code
190 * NX. If that's the case, remove !PR NX capability.
192 vcpu
->arch
.disable_kernel_nx
= true;
193 kvm_make_request(KVM_REQ_TLB_FLUSH
, vcpu
);
196 vcpu
->arch
.magic_page_pa
= param1
& ~0xfffULL
;
197 vcpu
->arch
.magic_page_ea
= param2
& ~0xfffULL
;
199 #ifdef CONFIG_PPC_64K_PAGES
201 * Make sure our 4k magic page is in the same window of a 64k
202 * page within the guest and within the host's page.
204 if ((vcpu
->arch
.magic_page_pa
& 0xf000) !=
205 ((ulong
)vcpu
->arch
.shared
& 0xf000)) {
206 void *old_shared
= vcpu
->arch
.shared
;
207 ulong shared
= (ulong
)vcpu
->arch
.shared
;
211 shared
|= vcpu
->arch
.magic_page_pa
& 0xf000;
212 new_shared
= (void*)shared
;
213 memcpy(new_shared
, old_shared
, 0x1000);
214 vcpu
->arch
.shared
= new_shared
;
218 r2
= KVM_MAGIC_FEAT_SR
| KVM_MAGIC_FEAT_MAS0_TO_SPRG7
;
223 case KVM_HCALL_TOKEN(KVM_HC_FEATURES
):
225 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
226 r2
|= (1 << KVM_FEATURE_MAGIC_PAGE
);
229 /* Second return value is in r4 */
231 case EV_HCALL_TOKEN(EV_IDLE
):
233 kvm_vcpu_block(vcpu
);
234 clear_bit(KVM_REQ_UNHALT
, &vcpu
->requests
);
237 r
= EV_UNIMPLEMENTED
;
241 kvmppc_set_gpr(vcpu
, 4, r2
);
245 EXPORT_SYMBOL_GPL(kvmppc_kvm_pv
);
247 int kvmppc_sanity_check(struct kvm_vcpu
*vcpu
)
251 /* We have to know what CPU to virtualize */
255 /* PAPR only works with book3s_64 */
256 if ((vcpu
->arch
.cpu_type
!= KVM_CPU_3S_64
) && vcpu
->arch
.papr_enabled
)
259 /* HV KVM can only do PAPR mode for now */
260 if (!vcpu
->arch
.papr_enabled
&& is_kvmppc_hv_enabled(vcpu
->kvm
))
263 #ifdef CONFIG_KVM_BOOKE_HV
264 if (!cpu_has_feature(CPU_FTR_EMB_HV
))
272 return r
? 0 : -EINVAL
;
274 EXPORT_SYMBOL_GPL(kvmppc_sanity_check
);
276 int kvmppc_emulate_mmio(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
)
278 enum emulation_result er
;
281 er
= kvmppc_emulate_loadstore(vcpu
);
284 /* Future optimization: only reload non-volatiles if they were
285 * actually modified. */
291 case EMULATE_DO_MMIO
:
292 run
->exit_reason
= KVM_EXIT_MMIO
;
293 /* We must reload nonvolatiles because "update" load/store
294 * instructions modify register state. */
295 /* Future optimization: only reload non-volatiles if they were
296 * actually modified. */
303 kvmppc_get_last_inst(vcpu
, INST_GENERIC
, &last_inst
);
304 /* XXX Deliver Program interrupt to guest. */
305 pr_emerg("%s: emulation failed (%08x)\n", __func__
, last_inst
);
316 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio
);
318 int kvmppc_st(struct kvm_vcpu
*vcpu
, ulong
*eaddr
, int size
, void *ptr
,
321 ulong mp_pa
= vcpu
->arch
.magic_page_pa
& KVM_PAM
& PAGE_MASK
;
322 struct kvmppc_pte pte
;
327 r
= kvmppc_xlate(vcpu
, *eaddr
, data
? XLATE_DATA
: XLATE_INST
,
337 /* Magic page override */
338 if (kvmppc_supports_magic_page(vcpu
) && mp_pa
&&
339 ((pte
.raddr
& KVM_PAM
& PAGE_MASK
) == mp_pa
) &&
340 !(kvmppc_get_msr(vcpu
) & MSR_PR
)) {
341 void *magic
= vcpu
->arch
.shared
;
342 magic
+= pte
.eaddr
& 0xfff;
343 memcpy(magic
, ptr
, size
);
347 if (kvm_write_guest(vcpu
->kvm
, pte
.raddr
, ptr
, size
))
348 return EMULATE_DO_MMIO
;
352 EXPORT_SYMBOL_GPL(kvmppc_st
);
354 int kvmppc_ld(struct kvm_vcpu
*vcpu
, ulong
*eaddr
, int size
, void *ptr
,
357 ulong mp_pa
= vcpu
->arch
.magic_page_pa
& KVM_PAM
& PAGE_MASK
;
358 struct kvmppc_pte pte
;
363 rc
= kvmppc_xlate(vcpu
, *eaddr
, data
? XLATE_DATA
: XLATE_INST
,
373 if (!data
&& !pte
.may_execute
)
376 /* Magic page override */
377 if (kvmppc_supports_magic_page(vcpu
) && mp_pa
&&
378 ((pte
.raddr
& KVM_PAM
& PAGE_MASK
) == mp_pa
) &&
379 !(kvmppc_get_msr(vcpu
) & MSR_PR
)) {
380 void *magic
= vcpu
->arch
.shared
;
381 magic
+= pte
.eaddr
& 0xfff;
382 memcpy(ptr
, magic
, size
);
386 if (kvm_read_guest(vcpu
->kvm
, pte
.raddr
, ptr
, size
))
387 return EMULATE_DO_MMIO
;
391 EXPORT_SYMBOL_GPL(kvmppc_ld
);
393 int kvm_arch_hardware_enable(void)
398 int kvm_arch_hardware_setup(void)
403 void kvm_arch_check_processor_compat(void *rtn
)
405 *(int *)rtn
= kvmppc_core_check_processor_compat();
408 int kvm_arch_init_vm(struct kvm
*kvm
, unsigned long type
)
410 struct kvmppc_ops
*kvm_ops
= NULL
;
412 * if we have both HV and PR enabled, default is HV
416 kvm_ops
= kvmppc_hv_ops
;
418 kvm_ops
= kvmppc_pr_ops
;
421 } else if (type
== KVM_VM_PPC_HV
) {
424 kvm_ops
= kvmppc_hv_ops
;
425 } else if (type
== KVM_VM_PPC_PR
) {
428 kvm_ops
= kvmppc_pr_ops
;
432 if (kvm_ops
->owner
&& !try_module_get(kvm_ops
->owner
))
435 kvm
->arch
.kvm_ops
= kvm_ops
;
436 return kvmppc_core_init_vm(kvm
);
441 bool kvm_arch_has_vcpu_debugfs(void)
446 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu
*vcpu
)
451 void kvm_arch_destroy_vm(struct kvm
*kvm
)
454 struct kvm_vcpu
*vcpu
;
456 #ifdef CONFIG_KVM_XICS
458 * We call kick_all_cpus_sync() to ensure that all
459 * CPUs have executed any pending IPIs before we
460 * continue and free VCPUs structures below.
462 if (is_kvmppc_hv_enabled(kvm
))
463 kick_all_cpus_sync();
466 kvm_for_each_vcpu(i
, vcpu
, kvm
)
467 kvm_arch_vcpu_free(vcpu
);
469 mutex_lock(&kvm
->lock
);
470 for (i
= 0; i
< atomic_read(&kvm
->online_vcpus
); i
++)
471 kvm
->vcpus
[i
] = NULL
;
473 atomic_set(&kvm
->online_vcpus
, 0);
475 kvmppc_core_destroy_vm(kvm
);
477 mutex_unlock(&kvm
->lock
);
479 /* drop the module reference */
480 module_put(kvm
->arch
.kvm_ops
->owner
);
483 int kvm_vm_ioctl_check_extension(struct kvm
*kvm
, long ext
)
486 /* Assume we're using HV mode when the HV module is loaded */
487 int hv_enabled
= kvmppc_hv_ops
? 1 : 0;
491 * Hooray - we know which VM type we're running on. Depend on
492 * that rather than the guess above.
494 hv_enabled
= is_kvmppc_hv_enabled(kvm
);
499 case KVM_CAP_PPC_BOOKE_SREGS
:
500 case KVM_CAP_PPC_BOOKE_WATCHDOG
:
501 case KVM_CAP_PPC_EPR
:
503 case KVM_CAP_PPC_SEGSTATE
:
504 case KVM_CAP_PPC_HIOR
:
505 case KVM_CAP_PPC_PAPR
:
507 case KVM_CAP_PPC_UNSET_IRQ
:
508 case KVM_CAP_PPC_IRQ_LEVEL
:
509 case KVM_CAP_ENABLE_CAP
:
510 case KVM_CAP_ENABLE_CAP_VM
:
511 case KVM_CAP_ONE_REG
:
512 case KVM_CAP_IOEVENTFD
:
513 case KVM_CAP_DEVICE_CTRL
:
514 case KVM_CAP_IMMEDIATE_EXIT
:
517 case KVM_CAP_PPC_PAIRED_SINGLES
:
518 case KVM_CAP_PPC_OSI
:
519 case KVM_CAP_PPC_GET_PVINFO
:
520 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
523 /* We support this only for PR */
526 #ifdef CONFIG_KVM_MMIO
527 case KVM_CAP_COALESCED_MMIO
:
528 r
= KVM_COALESCED_MMIO_PAGE_OFFSET
;
531 #ifdef CONFIG_KVM_MPIC
532 case KVM_CAP_IRQ_MPIC
:
537 #ifdef CONFIG_PPC_BOOK3S_64
538 case KVM_CAP_SPAPR_TCE
:
539 case KVM_CAP_SPAPR_TCE_64
:
540 case KVM_CAP_PPC_RTAS
:
541 case KVM_CAP_PPC_FIXUP_HCALL
:
542 case KVM_CAP_PPC_ENABLE_HCALL
:
543 #ifdef CONFIG_KVM_XICS
544 case KVM_CAP_IRQ_XICS
:
549 case KVM_CAP_PPC_ALLOC_HTAB
:
552 #endif /* CONFIG_PPC_BOOK3S_64 */
553 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
554 case KVM_CAP_PPC_SMT
:
557 if (cpu_has_feature(CPU_FTR_ARCH_300
))
560 r
= threads_per_subcore
;
563 case KVM_CAP_PPC_RMA
:
566 case KVM_CAP_PPC_HWRNG
:
567 r
= kvmppc_hwrng_present();
569 case KVM_CAP_PPC_MMU_RADIX
:
570 r
= !!(hv_enabled
&& radix_enabled());
572 case KVM_CAP_PPC_MMU_HASH_V3
:
573 r
= !!(hv_enabled
&& !radix_enabled() &&
574 cpu_has_feature(CPU_FTR_ARCH_300
));
577 case KVM_CAP_SYNC_MMU
:
578 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
580 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
586 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
587 case KVM_CAP_PPC_HTAB_FD
:
591 case KVM_CAP_NR_VCPUS
:
593 * Recommending a number of CPUs is somewhat arbitrary; we
594 * return the number of present CPUs for -HV (since a host
595 * will have secondary threads "offline"), and for other KVM
596 * implementations just count online CPUs.
599 r
= num_present_cpus();
601 r
= num_online_cpus();
603 case KVM_CAP_NR_MEMSLOTS
:
604 r
= KVM_USER_MEM_SLOTS
;
606 case KVM_CAP_MAX_VCPUS
:
609 #ifdef CONFIG_PPC_BOOK3S_64
610 case KVM_CAP_PPC_GET_SMMU_INFO
:
613 case KVM_CAP_SPAPR_MULTITCE
:
616 case KVM_CAP_SPAPR_RESIZE_HPT
:
617 /* Disable this on POWER9 until code handles new HPTE format */
618 r
= !!hv_enabled
&& !cpu_has_feature(CPU_FTR_ARCH_300
);
621 case KVM_CAP_PPC_HTM
:
622 r
= cpu_has_feature(CPU_FTR_TM_COMP
) &&
623 is_kvmppc_hv_enabled(kvm
);
633 long kvm_arch_dev_ioctl(struct file
*filp
,
634 unsigned int ioctl
, unsigned long arg
)
639 void kvm_arch_free_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*free
,
640 struct kvm_memory_slot
*dont
)
642 kvmppc_core_free_memslot(kvm
, free
, dont
);
645 int kvm_arch_create_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*slot
,
646 unsigned long npages
)
648 return kvmppc_core_create_memslot(kvm
, slot
, npages
);
651 int kvm_arch_prepare_memory_region(struct kvm
*kvm
,
652 struct kvm_memory_slot
*memslot
,
653 const struct kvm_userspace_memory_region
*mem
,
654 enum kvm_mr_change change
)
656 return kvmppc_core_prepare_memory_region(kvm
, memslot
, mem
);
659 void kvm_arch_commit_memory_region(struct kvm
*kvm
,
660 const struct kvm_userspace_memory_region
*mem
,
661 const struct kvm_memory_slot
*old
,
662 const struct kvm_memory_slot
*new,
663 enum kvm_mr_change change
)
665 kvmppc_core_commit_memory_region(kvm
, mem
, old
, new);
668 void kvm_arch_flush_shadow_memslot(struct kvm
*kvm
,
669 struct kvm_memory_slot
*slot
)
671 kvmppc_core_flush_memslot(kvm
, slot
);
674 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
, unsigned int id
)
676 struct kvm_vcpu
*vcpu
;
677 vcpu
= kvmppc_core_vcpu_create(kvm
, id
);
679 vcpu
->arch
.wqp
= &vcpu
->wq
;
680 kvmppc_create_vcpu_debugfs(vcpu
, id
);
685 void kvm_arch_vcpu_postcreate(struct kvm_vcpu
*vcpu
)
689 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
691 /* Make sure we're not using the vcpu anymore */
692 hrtimer_cancel(&vcpu
->arch
.dec_timer
);
694 kvmppc_remove_vcpu_debugfs(vcpu
);
696 switch (vcpu
->arch
.irq_type
) {
697 case KVMPPC_IRQ_MPIC
:
698 kvmppc_mpic_disconnect_vcpu(vcpu
->arch
.mpic
, vcpu
);
700 case KVMPPC_IRQ_XICS
:
701 kvmppc_xics_free_icp(vcpu
);
705 kvmppc_core_vcpu_free(vcpu
);
708 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
710 kvm_arch_vcpu_free(vcpu
);
713 int kvm_cpu_has_pending_timer(struct kvm_vcpu
*vcpu
)
715 return kvmppc_core_pending_dec(vcpu
);
718 static enum hrtimer_restart
kvmppc_decrementer_wakeup(struct hrtimer
*timer
)
720 struct kvm_vcpu
*vcpu
;
722 vcpu
= container_of(timer
, struct kvm_vcpu
, arch
.dec_timer
);
723 kvmppc_decrementer_func(vcpu
);
725 return HRTIMER_NORESTART
;
728 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
732 hrtimer_init(&vcpu
->arch
.dec_timer
, CLOCK_REALTIME
, HRTIMER_MODE_ABS
);
733 vcpu
->arch
.dec_timer
.function
= kvmppc_decrementer_wakeup
;
734 vcpu
->arch
.dec_expires
= ~(u64
)0;
736 #ifdef CONFIG_KVM_EXIT_TIMING
737 mutex_init(&vcpu
->arch
.exit_timing_lock
);
739 ret
= kvmppc_subarch_vcpu_init(vcpu
);
743 void kvm_arch_vcpu_uninit(struct kvm_vcpu
*vcpu
)
745 kvmppc_mmu_destroy(vcpu
);
746 kvmppc_subarch_vcpu_uninit(vcpu
);
749 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
753 * vrsave (formerly usprg0) isn't used by Linux, but may
754 * be used by the guest.
756 * On non-booke this is associated with Altivec and
757 * is handled by code in book3s.c.
759 mtspr(SPRN_VRSAVE
, vcpu
->arch
.vrsave
);
761 kvmppc_core_vcpu_load(vcpu
, cpu
);
764 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
766 kvmppc_core_vcpu_put(vcpu
);
768 vcpu
->arch
.vrsave
= mfspr(SPRN_VRSAVE
);
773 * irq_bypass_add_producer and irq_bypass_del_producer are only
774 * useful if the architecture supports PCI passthrough.
775 * irq_bypass_stop and irq_bypass_start are not needed and so
776 * kvm_ops are not defined for them.
778 bool kvm_arch_has_irq_bypass(void)
780 return ((kvmppc_hv_ops
&& kvmppc_hv_ops
->irq_bypass_add_producer
) ||
781 (kvmppc_pr_ops
&& kvmppc_pr_ops
->irq_bypass_add_producer
));
784 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer
*cons
,
785 struct irq_bypass_producer
*prod
)
787 struct kvm_kernel_irqfd
*irqfd
=
788 container_of(cons
, struct kvm_kernel_irqfd
, consumer
);
789 struct kvm
*kvm
= irqfd
->kvm
;
791 if (kvm
->arch
.kvm_ops
->irq_bypass_add_producer
)
792 return kvm
->arch
.kvm_ops
->irq_bypass_add_producer(cons
, prod
);
797 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer
*cons
,
798 struct irq_bypass_producer
*prod
)
800 struct kvm_kernel_irqfd
*irqfd
=
801 container_of(cons
, struct kvm_kernel_irqfd
, consumer
);
802 struct kvm
*kvm
= irqfd
->kvm
;
804 if (kvm
->arch
.kvm_ops
->irq_bypass_del_producer
)
805 kvm
->arch
.kvm_ops
->irq_bypass_del_producer(cons
, prod
);
808 static void kvmppc_complete_mmio_load(struct kvm_vcpu
*vcpu
,
811 u64
uninitialized_var(gpr
);
813 if (run
->mmio
.len
> sizeof(gpr
)) {
814 printk(KERN_ERR
"bad MMIO length: %d\n", run
->mmio
.len
);
818 if (!vcpu
->arch
.mmio_host_swabbed
) {
819 switch (run
->mmio
.len
) {
820 case 8: gpr
= *(u64
*)run
->mmio
.data
; break;
821 case 4: gpr
= *(u32
*)run
->mmio
.data
; break;
822 case 2: gpr
= *(u16
*)run
->mmio
.data
; break;
823 case 1: gpr
= *(u8
*)run
->mmio
.data
; break;
826 switch (run
->mmio
.len
) {
827 case 8: gpr
= swab64(*(u64
*)run
->mmio
.data
); break;
828 case 4: gpr
= swab32(*(u32
*)run
->mmio
.data
); break;
829 case 2: gpr
= swab16(*(u16
*)run
->mmio
.data
); break;
830 case 1: gpr
= *(u8
*)run
->mmio
.data
; break;
834 if (vcpu
->arch
.mmio_sign_extend
) {
835 switch (run
->mmio
.len
) {
850 kvmppc_set_gpr(vcpu
, vcpu
->arch
.io_gpr
, gpr
);
852 switch (vcpu
->arch
.io_gpr
& KVM_MMIO_REG_EXT_MASK
) {
853 case KVM_MMIO_REG_GPR
:
854 kvmppc_set_gpr(vcpu
, vcpu
->arch
.io_gpr
, gpr
);
856 case KVM_MMIO_REG_FPR
:
857 VCPU_FPR(vcpu
, vcpu
->arch
.io_gpr
& KVM_MMIO_REG_MASK
) = gpr
;
859 #ifdef CONFIG_PPC_BOOK3S
860 case KVM_MMIO_REG_QPR
:
861 vcpu
->arch
.qpr
[vcpu
->arch
.io_gpr
& KVM_MMIO_REG_MASK
] = gpr
;
863 case KVM_MMIO_REG_FQPR
:
864 VCPU_FPR(vcpu
, vcpu
->arch
.io_gpr
& KVM_MMIO_REG_MASK
) = gpr
;
865 vcpu
->arch
.qpr
[vcpu
->arch
.io_gpr
& KVM_MMIO_REG_MASK
] = gpr
;
873 static int __kvmppc_handle_load(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
,
874 unsigned int rt
, unsigned int bytes
,
875 int is_default_endian
, int sign_extend
)
880 /* Pity C doesn't have a logical XOR operator */
881 if (kvmppc_need_byteswap(vcpu
)) {
882 host_swabbed
= is_default_endian
;
884 host_swabbed
= !is_default_endian
;
887 if (bytes
> sizeof(run
->mmio
.data
)) {
888 printk(KERN_ERR
"%s: bad MMIO length: %d\n", __func__
,
892 run
->mmio
.phys_addr
= vcpu
->arch
.paddr_accessed
;
893 run
->mmio
.len
= bytes
;
894 run
->mmio
.is_write
= 0;
896 vcpu
->arch
.io_gpr
= rt
;
897 vcpu
->arch
.mmio_host_swabbed
= host_swabbed
;
898 vcpu
->mmio_needed
= 1;
899 vcpu
->mmio_is_write
= 0;
900 vcpu
->arch
.mmio_sign_extend
= sign_extend
;
902 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
904 ret
= kvm_io_bus_read(vcpu
, KVM_MMIO_BUS
, run
->mmio
.phys_addr
,
905 bytes
, &run
->mmio
.data
);
907 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
910 kvmppc_complete_mmio_load(vcpu
, run
);
911 vcpu
->mmio_needed
= 0;
915 return EMULATE_DO_MMIO
;
918 int kvmppc_handle_load(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
,
919 unsigned int rt
, unsigned int bytes
,
920 int is_default_endian
)
922 return __kvmppc_handle_load(run
, vcpu
, rt
, bytes
, is_default_endian
, 0);
924 EXPORT_SYMBOL_GPL(kvmppc_handle_load
);
926 /* Same as above, but sign extends */
927 int kvmppc_handle_loads(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
,
928 unsigned int rt
, unsigned int bytes
,
929 int is_default_endian
)
931 return __kvmppc_handle_load(run
, vcpu
, rt
, bytes
, is_default_endian
, 1);
934 int kvmppc_handle_store(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
,
935 u64 val
, unsigned int bytes
, int is_default_endian
)
937 void *data
= run
->mmio
.data
;
941 /* Pity C doesn't have a logical XOR operator */
942 if (kvmppc_need_byteswap(vcpu
)) {
943 host_swabbed
= is_default_endian
;
945 host_swabbed
= !is_default_endian
;
948 if (bytes
> sizeof(run
->mmio
.data
)) {
949 printk(KERN_ERR
"%s: bad MMIO length: %d\n", __func__
,
953 run
->mmio
.phys_addr
= vcpu
->arch
.paddr_accessed
;
954 run
->mmio
.len
= bytes
;
955 run
->mmio
.is_write
= 1;
956 vcpu
->mmio_needed
= 1;
957 vcpu
->mmio_is_write
= 1;
959 /* Store the value at the lowest bytes in 'data'. */
962 case 8: *(u64
*)data
= val
; break;
963 case 4: *(u32
*)data
= val
; break;
964 case 2: *(u16
*)data
= val
; break;
965 case 1: *(u8
*)data
= val
; break;
969 case 8: *(u64
*)data
= swab64(val
); break;
970 case 4: *(u32
*)data
= swab32(val
); break;
971 case 2: *(u16
*)data
= swab16(val
); break;
972 case 1: *(u8
*)data
= val
; break;
976 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
978 ret
= kvm_io_bus_write(vcpu
, KVM_MMIO_BUS
, run
->mmio
.phys_addr
,
979 bytes
, &run
->mmio
.data
);
981 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
984 vcpu
->mmio_needed
= 0;
988 return EMULATE_DO_MMIO
;
990 EXPORT_SYMBOL_GPL(kvmppc_handle_store
);
992 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu
*vcpu
, struct kvm_one_reg
*reg
)
995 union kvmppc_one_reg val
;
998 size
= one_reg_size(reg
->id
);
999 if (size
> sizeof(val
))
1002 r
= kvmppc_get_one_reg(vcpu
, reg
->id
, &val
);
1006 #ifdef CONFIG_ALTIVEC
1007 case KVM_REG_PPC_VR0
... KVM_REG_PPC_VR31
:
1008 if (!cpu_has_feature(CPU_FTR_ALTIVEC
)) {
1012 val
.vval
= vcpu
->arch
.vr
.vr
[reg
->id
- KVM_REG_PPC_VR0
];
1014 case KVM_REG_PPC_VSCR
:
1015 if (!cpu_has_feature(CPU_FTR_ALTIVEC
)) {
1019 val
= get_reg_val(reg
->id
, vcpu
->arch
.vr
.vscr
.u
[3]);
1021 case KVM_REG_PPC_VRSAVE
:
1022 val
= get_reg_val(reg
->id
, vcpu
->arch
.vrsave
);
1024 #endif /* CONFIG_ALTIVEC */
1034 if (copy_to_user((char __user
*)(unsigned long)reg
->addr
, &val
, size
))
1040 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu
*vcpu
, struct kvm_one_reg
*reg
)
1043 union kvmppc_one_reg val
;
1046 size
= one_reg_size(reg
->id
);
1047 if (size
> sizeof(val
))
1050 if (copy_from_user(&val
, (char __user
*)(unsigned long)reg
->addr
, size
))
1053 r
= kvmppc_set_one_reg(vcpu
, reg
->id
, &val
);
1057 #ifdef CONFIG_ALTIVEC
1058 case KVM_REG_PPC_VR0
... KVM_REG_PPC_VR31
:
1059 if (!cpu_has_feature(CPU_FTR_ALTIVEC
)) {
1063 vcpu
->arch
.vr
.vr
[reg
->id
- KVM_REG_PPC_VR0
] = val
.vval
;
1065 case KVM_REG_PPC_VSCR
:
1066 if (!cpu_has_feature(CPU_FTR_ALTIVEC
)) {
1070 vcpu
->arch
.vr
.vscr
.u
[3] = set_reg_val(reg
->id
, val
);
1072 case KVM_REG_PPC_VRSAVE
:
1073 if (!cpu_has_feature(CPU_FTR_ALTIVEC
)) {
1077 vcpu
->arch
.vrsave
= set_reg_val(reg
->id
, val
);
1079 #endif /* CONFIG_ALTIVEC */
1089 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
)
1094 if (vcpu
->sigset_active
)
1095 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
1097 if (vcpu
->mmio_needed
) {
1098 if (!vcpu
->mmio_is_write
)
1099 kvmppc_complete_mmio_load(vcpu
, run
);
1100 vcpu
->mmio_needed
= 0;
1101 } else if (vcpu
->arch
.osi_needed
) {
1102 u64
*gprs
= run
->osi
.gprs
;
1105 for (i
= 0; i
< 32; i
++)
1106 kvmppc_set_gpr(vcpu
, i
, gprs
[i
]);
1107 vcpu
->arch
.osi_needed
= 0;
1108 } else if (vcpu
->arch
.hcall_needed
) {
1111 kvmppc_set_gpr(vcpu
, 3, run
->papr_hcall
.ret
);
1112 for (i
= 0; i
< 9; ++i
)
1113 kvmppc_set_gpr(vcpu
, 4 + i
, run
->papr_hcall
.args
[i
]);
1114 vcpu
->arch
.hcall_needed
= 0;
1116 } else if (vcpu
->arch
.epr_needed
) {
1117 kvmppc_set_epr(vcpu
, run
->epr
.epr
);
1118 vcpu
->arch
.epr_needed
= 0;
1122 if (run
->immediate_exit
)
1125 r
= kvmppc_vcpu_run(run
, vcpu
);
1127 if (vcpu
->sigset_active
)
1128 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
1133 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu
*vcpu
, struct kvm_interrupt
*irq
)
1135 if (irq
->irq
== KVM_INTERRUPT_UNSET
) {
1136 kvmppc_core_dequeue_external(vcpu
);
1140 kvmppc_core_queue_external(vcpu
, irq
);
1142 kvm_vcpu_kick(vcpu
);
1147 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu
*vcpu
,
1148 struct kvm_enable_cap
*cap
)
1156 case KVM_CAP_PPC_OSI
:
1158 vcpu
->arch
.osi_enabled
= true;
1160 case KVM_CAP_PPC_PAPR
:
1162 vcpu
->arch
.papr_enabled
= true;
1164 case KVM_CAP_PPC_EPR
:
1167 vcpu
->arch
.epr_flags
|= KVMPPC_EPR_USER
;
1169 vcpu
->arch
.epr_flags
&= ~KVMPPC_EPR_USER
;
1172 case KVM_CAP_PPC_BOOKE_WATCHDOG
:
1174 vcpu
->arch
.watchdog_enabled
= true;
1177 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1178 case KVM_CAP_SW_TLB
: {
1179 struct kvm_config_tlb cfg
;
1180 void __user
*user_ptr
= (void __user
*)(uintptr_t)cap
->args
[0];
1183 if (copy_from_user(&cfg
, user_ptr
, sizeof(cfg
)))
1186 r
= kvm_vcpu_ioctl_config_tlb(vcpu
, &cfg
);
1190 #ifdef CONFIG_KVM_MPIC
1191 case KVM_CAP_IRQ_MPIC
: {
1193 struct kvm_device
*dev
;
1196 f
= fdget(cap
->args
[0]);
1201 dev
= kvm_device_from_filp(f
.file
);
1203 r
= kvmppc_mpic_connect_vcpu(dev
, vcpu
, cap
->args
[1]);
1209 #ifdef CONFIG_KVM_XICS
1210 case KVM_CAP_IRQ_XICS
: {
1212 struct kvm_device
*dev
;
1215 f
= fdget(cap
->args
[0]);
1220 dev
= kvm_device_from_filp(f
.file
);
1222 r
= kvmppc_xics_connect_vcpu(dev
, vcpu
, cap
->args
[1]);
1227 #endif /* CONFIG_KVM_XICS */
1234 r
= kvmppc_sanity_check(vcpu
);
1239 bool kvm_arch_intc_initialized(struct kvm
*kvm
)
1241 #ifdef CONFIG_KVM_MPIC
1245 #ifdef CONFIG_KVM_XICS
1252 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
1253 struct kvm_mp_state
*mp_state
)
1258 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
1259 struct kvm_mp_state
*mp_state
)
1264 long kvm_arch_vcpu_ioctl(struct file
*filp
,
1265 unsigned int ioctl
, unsigned long arg
)
1267 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1268 void __user
*argp
= (void __user
*)arg
;
1272 case KVM_INTERRUPT
: {
1273 struct kvm_interrupt irq
;
1275 if (copy_from_user(&irq
, argp
, sizeof(irq
)))
1277 r
= kvm_vcpu_ioctl_interrupt(vcpu
, &irq
);
1281 case KVM_ENABLE_CAP
:
1283 struct kvm_enable_cap cap
;
1285 if (copy_from_user(&cap
, argp
, sizeof(cap
)))
1287 r
= kvm_vcpu_ioctl_enable_cap(vcpu
, &cap
);
1291 case KVM_SET_ONE_REG
:
1292 case KVM_GET_ONE_REG
:
1294 struct kvm_one_reg reg
;
1296 if (copy_from_user(®
, argp
, sizeof(reg
)))
1298 if (ioctl
== KVM_SET_ONE_REG
)
1299 r
= kvm_vcpu_ioctl_set_one_reg(vcpu
, ®
);
1301 r
= kvm_vcpu_ioctl_get_one_reg(vcpu
, ®
);
1305 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1306 case KVM_DIRTY_TLB
: {
1307 struct kvm_dirty_tlb dirty
;
1309 if (copy_from_user(&dirty
, argp
, sizeof(dirty
)))
1311 r
= kvm_vcpu_ioctl_dirty_tlb(vcpu
, &dirty
);
1323 int kvm_arch_vcpu_fault(struct kvm_vcpu
*vcpu
, struct vm_fault
*vmf
)
1325 return VM_FAULT_SIGBUS
;
1328 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo
*pvinfo
)
1330 u32 inst_nop
= 0x60000000;
1331 #ifdef CONFIG_KVM_BOOKE_HV
1332 u32 inst_sc1
= 0x44000022;
1333 pvinfo
->hcall
[0] = cpu_to_be32(inst_sc1
);
1334 pvinfo
->hcall
[1] = cpu_to_be32(inst_nop
);
1335 pvinfo
->hcall
[2] = cpu_to_be32(inst_nop
);
1336 pvinfo
->hcall
[3] = cpu_to_be32(inst_nop
);
1338 u32 inst_lis
= 0x3c000000;
1339 u32 inst_ori
= 0x60000000;
1340 u32 inst_sc
= 0x44000002;
1341 u32 inst_imm_mask
= 0xffff;
1344 * The hypercall to get into KVM from within guest context is as
1347 * lis r0, r0, KVM_SC_MAGIC_R0@h
1348 * ori r0, KVM_SC_MAGIC_R0@l
1352 pvinfo
->hcall
[0] = cpu_to_be32(inst_lis
| ((KVM_SC_MAGIC_R0
>> 16) & inst_imm_mask
));
1353 pvinfo
->hcall
[1] = cpu_to_be32(inst_ori
| (KVM_SC_MAGIC_R0
& inst_imm_mask
));
1354 pvinfo
->hcall
[2] = cpu_to_be32(inst_sc
);
1355 pvinfo
->hcall
[3] = cpu_to_be32(inst_nop
);
1358 pvinfo
->flags
= KVM_PPC_PVINFO_FLAGS_EV_IDLE
;
1363 int kvm_vm_ioctl_irq_line(struct kvm
*kvm
, struct kvm_irq_level
*irq_event
,
1366 if (!irqchip_in_kernel(kvm
))
1369 irq_event
->status
= kvm_set_irq(kvm
, KVM_USERSPACE_IRQ_SOURCE_ID
,
1370 irq_event
->irq
, irq_event
->level
,
1376 static int kvm_vm_ioctl_enable_cap(struct kvm
*kvm
,
1377 struct kvm_enable_cap
*cap
)
1385 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
1386 case KVM_CAP_PPC_ENABLE_HCALL
: {
1387 unsigned long hcall
= cap
->args
[0];
1390 if (hcall
> MAX_HCALL_OPCODE
|| (hcall
& 3) ||
1393 if (!kvmppc_book3s_hcall_implemented(kvm
, hcall
))
1396 set_bit(hcall
/ 4, kvm
->arch
.enabled_hcalls
);
1398 clear_bit(hcall
/ 4, kvm
->arch
.enabled_hcalls
);
1411 long kvm_arch_vm_ioctl(struct file
*filp
,
1412 unsigned int ioctl
, unsigned long arg
)
1414 struct kvm
*kvm __maybe_unused
= filp
->private_data
;
1415 void __user
*argp
= (void __user
*)arg
;
1419 case KVM_PPC_GET_PVINFO
: {
1420 struct kvm_ppc_pvinfo pvinfo
;
1421 memset(&pvinfo
, 0, sizeof(pvinfo
));
1422 r
= kvm_vm_ioctl_get_pvinfo(&pvinfo
);
1423 if (copy_to_user(argp
, &pvinfo
, sizeof(pvinfo
))) {
1430 case KVM_ENABLE_CAP
:
1432 struct kvm_enable_cap cap
;
1434 if (copy_from_user(&cap
, argp
, sizeof(cap
)))
1436 r
= kvm_vm_ioctl_enable_cap(kvm
, &cap
);
1439 #ifdef CONFIG_PPC_BOOK3S_64
1440 case KVM_CREATE_SPAPR_TCE_64
: {
1441 struct kvm_create_spapr_tce_64 create_tce_64
;
1444 if (copy_from_user(&create_tce_64
, argp
, sizeof(create_tce_64
)))
1446 if (create_tce_64
.flags
) {
1450 r
= kvm_vm_ioctl_create_spapr_tce(kvm
, &create_tce_64
);
1453 case KVM_CREATE_SPAPR_TCE
: {
1454 struct kvm_create_spapr_tce create_tce
;
1455 struct kvm_create_spapr_tce_64 create_tce_64
;
1458 if (copy_from_user(&create_tce
, argp
, sizeof(create_tce
)))
1461 create_tce_64
.liobn
= create_tce
.liobn
;
1462 create_tce_64
.page_shift
= IOMMU_PAGE_SHIFT_4K
;
1463 create_tce_64
.offset
= 0;
1464 create_tce_64
.size
= create_tce
.window_size
>>
1465 IOMMU_PAGE_SHIFT_4K
;
1466 create_tce_64
.flags
= 0;
1467 r
= kvm_vm_ioctl_create_spapr_tce(kvm
, &create_tce_64
);
1470 case KVM_PPC_GET_SMMU_INFO
: {
1471 struct kvm_ppc_smmu_info info
;
1472 struct kvm
*kvm
= filp
->private_data
;
1474 memset(&info
, 0, sizeof(info
));
1475 r
= kvm
->arch
.kvm_ops
->get_smmu_info(kvm
, &info
);
1476 if (r
>= 0 && copy_to_user(argp
, &info
, sizeof(info
)))
1480 case KVM_PPC_RTAS_DEFINE_TOKEN
: {
1481 struct kvm
*kvm
= filp
->private_data
;
1483 r
= kvm_vm_ioctl_rtas_define_token(kvm
, argp
);
1486 case KVM_PPC_CONFIGURE_V3_MMU
: {
1487 struct kvm
*kvm
= filp
->private_data
;
1488 struct kvm_ppc_mmuv3_cfg cfg
;
1491 if (!kvm
->arch
.kvm_ops
->configure_mmu
)
1494 if (copy_from_user(&cfg
, argp
, sizeof(cfg
)))
1496 r
= kvm
->arch
.kvm_ops
->configure_mmu(kvm
, &cfg
);
1499 case KVM_PPC_GET_RMMU_INFO
: {
1500 struct kvm
*kvm
= filp
->private_data
;
1501 struct kvm_ppc_rmmu_info info
;
1504 if (!kvm
->arch
.kvm_ops
->get_rmmu_info
)
1506 r
= kvm
->arch
.kvm_ops
->get_rmmu_info(kvm
, &info
);
1507 if (r
>= 0 && copy_to_user(argp
, &info
, sizeof(info
)))
1512 struct kvm
*kvm
= filp
->private_data
;
1513 r
= kvm
->arch
.kvm_ops
->arch_vm_ioctl(filp
, ioctl
, arg
);
1515 #else /* CONFIG_PPC_BOOK3S_64 */
1524 static unsigned long lpid_inuse
[BITS_TO_LONGS(KVMPPC_NR_LPIDS
)];
1525 static unsigned long nr_lpids
;
1527 long kvmppc_alloc_lpid(void)
1532 lpid
= find_first_zero_bit(lpid_inuse
, KVMPPC_NR_LPIDS
);
1533 if (lpid
>= nr_lpids
) {
1534 pr_err("%s: No LPIDs free\n", __func__
);
1537 } while (test_and_set_bit(lpid
, lpid_inuse
));
1541 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid
);
1543 void kvmppc_claim_lpid(long lpid
)
1545 set_bit(lpid
, lpid_inuse
);
1547 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid
);
1549 void kvmppc_free_lpid(long lpid
)
1551 clear_bit(lpid
, lpid_inuse
);
1553 EXPORT_SYMBOL_GPL(kvmppc_free_lpid
);
1555 void kvmppc_init_lpid(unsigned long nr_lpids_param
)
1557 nr_lpids
= min_t(unsigned long, KVMPPC_NR_LPIDS
, nr_lpids_param
);
1558 memset(lpid_inuse
, 0, sizeof(lpid_inuse
));
1560 EXPORT_SYMBOL_GPL(kvmppc_init_lpid
);
1562 int kvm_arch_init(void *opaque
)
1567 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr
);