2 * This program is free software; you can redistribute it and/or modify
3 * it under the terms of the GNU General Public License, version 2, as
4 * published by the Free Software Foundation.
6 * This program is distributed in the hope that it will be useful,
7 * but WITHOUT ANY WARRANTY; without even the implied warranty of
8 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
9 * GNU General Public License for more details.
11 * You should have received a copy of the GNU General Public License
12 * along with this program; if not, write to the Free Software
13 * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
15 * Copyright IBM Corp. 2007
17 * Authors: Hollis Blanchard <hollisb@us.ibm.com>
18 * Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
21 #include <linux/errno.h>
22 #include <linux/err.h>
23 #include <linux/kvm_host.h>
24 #include <linux/vmalloc.h>
25 #include <linux/hrtimer.h>
27 #include <linux/slab.h>
28 #include <linux/file.h>
29 #include <linux/module.h>
30 #include <asm/cputable.h>
31 #include <asm/uaccess.h>
32 #include <asm/kvm_ppc.h>
33 #include <asm/tlbflush.h>
34 #include <asm/cputhreads.h>
35 #include <asm/irqflags.h>
38 #include "../mm/mmu_decl.h"
40 #define CREATE_TRACE_POINTS
43 struct kvmppc_ops *kvmppc_hv_ops;
44 EXPORT_SYMBOL_GPL(kvmppc_hv_ops);
45 struct kvmppc_ops *kvmppc_pr_ops;
46 EXPORT_SYMBOL_GPL(kvmppc_pr_ops);
49 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
51 return !!(v->arch.pending_exceptions) ||
55 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
61 * Common checks before entering the guest world. Call with interrupts
66 * == 1 if we're ready to go into guest state
67 * <= 0 if we need to go back to the host with return value
69 int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu)
73 WARN_ON(irqs_disabled());
84 if (signal_pending(current)) {
85 kvmppc_account_exit(vcpu, SIGNAL_EXITS);
86 vcpu->run->exit_reason = KVM_EXIT_INTR;
91 vcpu->mode = IN_GUEST_MODE;
94 * Reading vcpu->requests must happen after setting vcpu->mode,
95 * so we don't miss a request because the requester sees
96 * OUTSIDE_GUEST_MODE and assumes we'll be checking requests
97 * before next entering the guest (and thus doesn't IPI).
101 if (vcpu->requests) {
102 /* Make sure we process requests preemptable */
104 trace_kvm_check_requests(vcpu);
105 r = kvmppc_core_check_requests(vcpu);
112 if (kvmppc_core_prepare_to_enter(vcpu)) {
113 /* interrupts got enabled in between, so we
114 are back at square 1 */
126 EXPORT_SYMBOL_GPL(kvmppc_prepare_to_enter);
128 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
129 static void kvmppc_swab_shared(struct kvm_vcpu *vcpu)
131 struct kvm_vcpu_arch_shared *shared = vcpu->arch.shared;
134 shared->sprg0 = swab64(shared->sprg0);
135 shared->sprg1 = swab64(shared->sprg1);
136 shared->sprg2 = swab64(shared->sprg2);
137 shared->sprg3 = swab64(shared->sprg3);
138 shared->srr0 = swab64(shared->srr0);
139 shared->srr1 = swab64(shared->srr1);
140 shared->dar = swab64(shared->dar);
141 shared->msr = swab64(shared->msr);
142 shared->dsisr = swab32(shared->dsisr);
143 shared->int_pending = swab32(shared->int_pending);
144 for (i = 0; i < ARRAY_SIZE(shared->sr); i++)
145 shared->sr[i] = swab32(shared->sr[i]);
149 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
151 int nr = kvmppc_get_gpr(vcpu, 11);
153 unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
154 unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
155 unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
156 unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
157 unsigned long r2 = 0;
159 if (!(kvmppc_get_msr(vcpu) & MSR_SF)) {
161 param1 &= 0xffffffff;
162 param2 &= 0xffffffff;
163 param3 &= 0xffffffff;
164 param4 &= 0xffffffff;
168 case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE):
170 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
171 /* Book3S can be little endian, find it out here */
172 int shared_big_endian = true;
173 if (vcpu->arch.intr_msr & MSR_LE)
174 shared_big_endian = false;
175 if (shared_big_endian != vcpu->arch.shared_big_endian)
176 kvmppc_swab_shared(vcpu);
177 vcpu->arch.shared_big_endian = shared_big_endian;
180 if (!(param2 & MAGIC_PAGE_FLAG_NOT_MAPPED_NX)) {
182 * Older versions of the Linux magic page code had
183 * a bug where they would map their trampoline code
184 * NX. If that's the case, remove !PR NX capability.
186 vcpu->arch.disable_kernel_nx = true;
187 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
190 vcpu->arch.magic_page_pa = param1 & ~0xfffULL;
191 vcpu->arch.magic_page_ea = param2 & ~0xfffULL;
193 #ifdef CONFIG_PPC_64K_PAGES
195 * Make sure our 4k magic page is in the same window of a 64k
196 * page within the guest and within the host's page.
198 if ((vcpu->arch.magic_page_pa & 0xf000) !=
199 ((ulong)vcpu->arch.shared & 0xf000)) {
200 void *old_shared = vcpu->arch.shared;
201 ulong shared = (ulong)vcpu->arch.shared;
205 shared |= vcpu->arch.magic_page_pa & 0xf000;
206 new_shared = (void*)shared;
207 memcpy(new_shared, old_shared, 0x1000);
208 vcpu->arch.shared = new_shared;
212 r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
217 case KVM_HCALL_TOKEN(KVM_HC_FEATURES):
219 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
220 r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
223 /* Second return value is in r4 */
225 case EV_HCALL_TOKEN(EV_IDLE):
227 kvm_vcpu_block(vcpu);
228 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
231 r = EV_UNIMPLEMENTED;
235 kvmppc_set_gpr(vcpu, 4, r2);
239 EXPORT_SYMBOL_GPL(kvmppc_kvm_pv);
241 int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
245 /* We have to know what CPU to virtualize */
249 /* PAPR only works with book3s_64 */
250 if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
253 /* HV KVM can only do PAPR mode for now */
254 if (!vcpu->arch.papr_enabled && is_kvmppc_hv_enabled(vcpu->kvm))
257 #ifdef CONFIG_KVM_BOOKE_HV
258 if (!cpu_has_feature(CPU_FTR_EMB_HV))
266 return r ? 0 : -EINVAL;
268 EXPORT_SYMBOL_GPL(kvmppc_sanity_check);
270 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
272 enum emulation_result er;
275 er = kvmppc_emulate_instruction(run, vcpu);
278 /* Future optimization: only reload non-volatiles if they were
279 * actually modified. */
285 case EMULATE_DO_MMIO:
286 run->exit_reason = KVM_EXIT_MMIO;
287 /* We must reload nonvolatiles because "update" load/store
288 * instructions modify register state. */
289 /* Future optimization: only reload non-volatiles if they were
290 * actually modified. */
297 kvmppc_get_last_inst(vcpu, false, &last_inst);
298 /* XXX Deliver Program interrupt to guest. */
299 pr_emerg("%s: emulation failed (%08x)\n", __func__, last_inst);
310 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio);
312 int kvm_arch_hardware_enable(void *garbage)
317 void kvm_arch_hardware_disable(void *garbage)
321 int kvm_arch_hardware_setup(void)
326 void kvm_arch_hardware_unsetup(void)
330 void kvm_arch_check_processor_compat(void *rtn)
332 *(int *)rtn = kvmppc_core_check_processor_compat();
335 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
337 struct kvmppc_ops *kvm_ops = NULL;
339 * if we have both HV and PR enabled, default is HV
343 kvm_ops = kvmppc_hv_ops;
345 kvm_ops = kvmppc_pr_ops;
348 } else if (type == KVM_VM_PPC_HV) {
351 kvm_ops = kvmppc_hv_ops;
352 } else if (type == KVM_VM_PPC_PR) {
355 kvm_ops = kvmppc_pr_ops;
359 if (kvm_ops->owner && !try_module_get(kvm_ops->owner))
362 kvm->arch.kvm_ops = kvm_ops;
363 return kvmppc_core_init_vm(kvm);
368 void kvm_arch_destroy_vm(struct kvm *kvm)
371 struct kvm_vcpu *vcpu;
373 kvm_for_each_vcpu(i, vcpu, kvm)
374 kvm_arch_vcpu_free(vcpu);
376 mutex_lock(&kvm->lock);
377 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
378 kvm->vcpus[i] = NULL;
380 atomic_set(&kvm->online_vcpus, 0);
382 kvmppc_core_destroy_vm(kvm);
384 mutex_unlock(&kvm->lock);
386 /* drop the module reference */
387 module_put(kvm->arch.kvm_ops->owner);
390 void kvm_arch_sync_events(struct kvm *kvm)
394 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
398 * Should some of this be vm ioctl ? is it possible now ?
400 int hv_enabled = kvmppc_hv_ops ? 1 : 0;
404 case KVM_CAP_PPC_BOOKE_SREGS:
405 case KVM_CAP_PPC_BOOKE_WATCHDOG:
406 case KVM_CAP_PPC_EPR:
408 case KVM_CAP_PPC_SEGSTATE:
409 case KVM_CAP_PPC_HIOR:
410 case KVM_CAP_PPC_PAPR:
412 case KVM_CAP_PPC_UNSET_IRQ:
413 case KVM_CAP_PPC_IRQ_LEVEL:
414 case KVM_CAP_ENABLE_CAP:
415 case KVM_CAP_ENABLE_CAP_VM:
416 case KVM_CAP_ONE_REG:
417 case KVM_CAP_IOEVENTFD:
418 case KVM_CAP_DEVICE_CTRL:
421 case KVM_CAP_PPC_PAIRED_SINGLES:
422 case KVM_CAP_PPC_OSI:
423 case KVM_CAP_PPC_GET_PVINFO:
424 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
427 /* We support this only for PR */
430 #ifdef CONFIG_KVM_MMIO
431 case KVM_CAP_COALESCED_MMIO:
432 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
435 #ifdef CONFIG_KVM_MPIC
436 case KVM_CAP_IRQ_MPIC:
441 #ifdef CONFIG_PPC_BOOK3S_64
442 case KVM_CAP_SPAPR_TCE:
443 case KVM_CAP_PPC_ALLOC_HTAB:
444 case KVM_CAP_PPC_RTAS:
445 case KVM_CAP_PPC_FIXUP_HCALL:
446 case KVM_CAP_PPC_ENABLE_HCALL:
447 #ifdef CONFIG_KVM_XICS
448 case KVM_CAP_IRQ_XICS:
452 #endif /* CONFIG_PPC_BOOK3S_64 */
453 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
454 case KVM_CAP_PPC_SMT:
456 r = threads_per_subcore;
460 case KVM_CAP_PPC_RMA:
462 /* PPC970 requires an RMA */
463 if (r && cpu_has_feature(CPU_FTR_ARCH_201))
467 case KVM_CAP_SYNC_MMU:
468 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
470 r = cpu_has_feature(CPU_FTR_ARCH_206) ? 1 : 0;
473 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
479 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
480 case KVM_CAP_PPC_HTAB_FD:
484 case KVM_CAP_NR_VCPUS:
486 * Recommending a number of CPUs is somewhat arbitrary; we
487 * return the number of present CPUs for -HV (since a host
488 * will have secondary threads "offline"), and for other KVM
489 * implementations just count online CPUs.
492 r = num_present_cpus();
494 r = num_online_cpus();
496 case KVM_CAP_MAX_VCPUS:
499 #ifdef CONFIG_PPC_BOOK3S_64
500 case KVM_CAP_PPC_GET_SMMU_INFO:
512 long kvm_arch_dev_ioctl(struct file *filp,
513 unsigned int ioctl, unsigned long arg)
518 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
519 struct kvm_memory_slot *dont)
521 kvmppc_core_free_memslot(kvm, free, dont);
524 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
525 unsigned long npages)
527 return kvmppc_core_create_memslot(kvm, slot, npages);
530 void kvm_arch_memslots_updated(struct kvm *kvm)
534 int kvm_arch_prepare_memory_region(struct kvm *kvm,
535 struct kvm_memory_slot *memslot,
536 struct kvm_userspace_memory_region *mem,
537 enum kvm_mr_change change)
539 return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
542 void kvm_arch_commit_memory_region(struct kvm *kvm,
543 struct kvm_userspace_memory_region *mem,
544 const struct kvm_memory_slot *old,
545 enum kvm_mr_change change)
547 kvmppc_core_commit_memory_region(kvm, mem, old);
550 void kvm_arch_flush_shadow_all(struct kvm *kvm)
554 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
555 struct kvm_memory_slot *slot)
557 kvmppc_core_flush_memslot(kvm, slot);
560 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
562 struct kvm_vcpu *vcpu;
563 vcpu = kvmppc_core_vcpu_create(kvm, id);
565 vcpu->arch.wqp = &vcpu->wq;
566 kvmppc_create_vcpu_debugfs(vcpu, id);
571 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
576 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
578 /* Make sure we're not using the vcpu anymore */
579 hrtimer_cancel(&vcpu->arch.dec_timer);
580 tasklet_kill(&vcpu->arch.tasklet);
582 kvmppc_remove_vcpu_debugfs(vcpu);
584 switch (vcpu->arch.irq_type) {
585 case KVMPPC_IRQ_MPIC:
586 kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
588 case KVMPPC_IRQ_XICS:
589 kvmppc_xics_free_icp(vcpu);
593 kvmppc_core_vcpu_free(vcpu);
596 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
598 kvm_arch_vcpu_free(vcpu);
601 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
603 return kvmppc_core_pending_dec(vcpu);
607 * low level hrtimer wake routine. Because this runs in hardirq context
608 * we schedule a tasklet to do the real work.
610 enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
612 struct kvm_vcpu *vcpu;
614 vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
615 tasklet_schedule(&vcpu->arch.tasklet);
617 return HRTIMER_NORESTART;
620 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
624 hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
625 tasklet_init(&vcpu->arch.tasklet, kvmppc_decrementer_func, (ulong)vcpu);
626 vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
627 vcpu->arch.dec_expires = ~(u64)0;
629 #ifdef CONFIG_KVM_EXIT_TIMING
630 mutex_init(&vcpu->arch.exit_timing_lock);
632 ret = kvmppc_subarch_vcpu_init(vcpu);
636 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
638 kvmppc_mmu_destroy(vcpu);
639 kvmppc_subarch_vcpu_uninit(vcpu);
642 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
646 * vrsave (formerly usprg0) isn't used by Linux, but may
647 * be used by the guest.
649 * On non-booke this is associated with Altivec and
650 * is handled by code in book3s.c.
652 mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
654 kvmppc_core_vcpu_load(vcpu, cpu);
657 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
659 kvmppc_core_vcpu_put(vcpu);
661 vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
665 static void kvmppc_complete_dcr_load(struct kvm_vcpu *vcpu,
668 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, run->dcr.data);
671 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
674 u64 uninitialized_var(gpr);
676 if (run->mmio.len > sizeof(gpr)) {
677 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
681 if (vcpu->arch.mmio_is_bigendian) {
682 switch (run->mmio.len) {
683 case 8: gpr = *(u64 *)run->mmio.data; break;
684 case 4: gpr = *(u32 *)run->mmio.data; break;
685 case 2: gpr = *(u16 *)run->mmio.data; break;
686 case 1: gpr = *(u8 *)run->mmio.data; break;
689 /* Convert BE data from userland back to LE. */
690 switch (run->mmio.len) {
691 case 4: gpr = ld_le32((u32 *)run->mmio.data); break;
692 case 2: gpr = ld_le16((u16 *)run->mmio.data); break;
693 case 1: gpr = *(u8 *)run->mmio.data; break;
697 if (vcpu->arch.mmio_sign_extend) {
698 switch (run->mmio.len) {
713 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
715 switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
716 case KVM_MMIO_REG_GPR:
717 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
719 case KVM_MMIO_REG_FPR:
720 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
722 #ifdef CONFIG_PPC_BOOK3S
723 case KVM_MMIO_REG_QPR:
724 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
726 case KVM_MMIO_REG_FQPR:
727 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
728 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
736 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
737 unsigned int rt, unsigned int bytes,
738 int is_default_endian)
743 if (kvmppc_need_byteswap(vcpu)) {
744 /* Default endianness is "little endian". */
745 is_bigendian = !is_default_endian;
747 /* Default endianness is "big endian". */
748 is_bigendian = is_default_endian;
751 if (bytes > sizeof(run->mmio.data)) {
752 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
756 run->mmio.phys_addr = vcpu->arch.paddr_accessed;
757 run->mmio.len = bytes;
758 run->mmio.is_write = 0;
760 vcpu->arch.io_gpr = rt;
761 vcpu->arch.mmio_is_bigendian = is_bigendian;
762 vcpu->mmio_needed = 1;
763 vcpu->mmio_is_write = 0;
764 vcpu->arch.mmio_sign_extend = 0;
766 idx = srcu_read_lock(&vcpu->kvm->srcu);
768 ret = kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, run->mmio.phys_addr,
769 bytes, &run->mmio.data);
771 srcu_read_unlock(&vcpu->kvm->srcu, idx);
774 kvmppc_complete_mmio_load(vcpu, run);
775 vcpu->mmio_needed = 0;
779 return EMULATE_DO_MMIO;
781 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
783 /* Same as above, but sign extends */
784 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
785 unsigned int rt, unsigned int bytes,
786 int is_default_endian)
790 vcpu->arch.mmio_sign_extend = 1;
791 r = kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian);
796 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
797 u64 val, unsigned int bytes, int is_default_endian)
799 void *data = run->mmio.data;
803 if (kvmppc_need_byteswap(vcpu)) {
804 /* Default endianness is "little endian". */
805 is_bigendian = !is_default_endian;
807 /* Default endianness is "big endian". */
808 is_bigendian = is_default_endian;
811 if (bytes > sizeof(run->mmio.data)) {
812 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
816 run->mmio.phys_addr = vcpu->arch.paddr_accessed;
817 run->mmio.len = bytes;
818 run->mmio.is_write = 1;
819 vcpu->mmio_needed = 1;
820 vcpu->mmio_is_write = 1;
822 /* Store the value at the lowest bytes in 'data'. */
825 case 8: *(u64 *)data = val; break;
826 case 4: *(u32 *)data = val; break;
827 case 2: *(u16 *)data = val; break;
828 case 1: *(u8 *)data = val; break;
831 /* Store LE value into 'data'. */
833 case 4: st_le32(data, val); break;
834 case 2: st_le16(data, val); break;
835 case 1: *(u8 *)data = val; break;
839 idx = srcu_read_lock(&vcpu->kvm->srcu);
841 ret = kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, run->mmio.phys_addr,
842 bytes, &run->mmio.data);
844 srcu_read_unlock(&vcpu->kvm->srcu, idx);
847 vcpu->mmio_needed = 0;
851 return EMULATE_DO_MMIO;
853 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
855 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
860 if (vcpu->sigset_active)
861 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
863 if (vcpu->mmio_needed) {
864 if (!vcpu->mmio_is_write)
865 kvmppc_complete_mmio_load(vcpu, run);
866 vcpu->mmio_needed = 0;
867 } else if (vcpu->arch.dcr_needed) {
868 if (!vcpu->arch.dcr_is_write)
869 kvmppc_complete_dcr_load(vcpu, run);
870 vcpu->arch.dcr_needed = 0;
871 } else if (vcpu->arch.osi_needed) {
872 u64 *gprs = run->osi.gprs;
875 for (i = 0; i < 32; i++)
876 kvmppc_set_gpr(vcpu, i, gprs[i]);
877 vcpu->arch.osi_needed = 0;
878 } else if (vcpu->arch.hcall_needed) {
881 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
882 for (i = 0; i < 9; ++i)
883 kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
884 vcpu->arch.hcall_needed = 0;
886 } else if (vcpu->arch.epr_needed) {
887 kvmppc_set_epr(vcpu, run->epr.epr);
888 vcpu->arch.epr_needed = 0;
892 r = kvmppc_vcpu_run(run, vcpu);
894 if (vcpu->sigset_active)
895 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
900 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
902 if (irq->irq == KVM_INTERRUPT_UNSET) {
903 kvmppc_core_dequeue_external(vcpu);
907 kvmppc_core_queue_external(vcpu, irq);
914 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
915 struct kvm_enable_cap *cap)
923 case KVM_CAP_PPC_OSI:
925 vcpu->arch.osi_enabled = true;
927 case KVM_CAP_PPC_PAPR:
929 vcpu->arch.papr_enabled = true;
931 case KVM_CAP_PPC_EPR:
934 vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
936 vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
939 case KVM_CAP_PPC_BOOKE_WATCHDOG:
941 vcpu->arch.watchdog_enabled = true;
944 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
945 case KVM_CAP_SW_TLB: {
946 struct kvm_config_tlb cfg;
947 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
950 if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
953 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
957 #ifdef CONFIG_KVM_MPIC
958 case KVM_CAP_IRQ_MPIC: {
960 struct kvm_device *dev;
963 f = fdget(cap->args[0]);
968 dev = kvm_device_from_filp(f.file);
970 r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
976 #ifdef CONFIG_KVM_XICS
977 case KVM_CAP_IRQ_XICS: {
979 struct kvm_device *dev;
982 f = fdget(cap->args[0]);
987 dev = kvm_device_from_filp(f.file);
989 r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
994 #endif /* CONFIG_KVM_XICS */
1001 r = kvmppc_sanity_check(vcpu);
1006 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1007 struct kvm_mp_state *mp_state)
1012 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1013 struct kvm_mp_state *mp_state)
1018 long kvm_arch_vcpu_ioctl(struct file *filp,
1019 unsigned int ioctl, unsigned long arg)
1021 struct kvm_vcpu *vcpu = filp->private_data;
1022 void __user *argp = (void __user *)arg;
1026 case KVM_INTERRUPT: {
1027 struct kvm_interrupt irq;
1029 if (copy_from_user(&irq, argp, sizeof(irq)))
1031 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1035 case KVM_ENABLE_CAP:
1037 struct kvm_enable_cap cap;
1039 if (copy_from_user(&cap, argp, sizeof(cap)))
1041 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
1045 case KVM_SET_ONE_REG:
1046 case KVM_GET_ONE_REG:
1048 struct kvm_one_reg reg;
1050 if (copy_from_user(®, argp, sizeof(reg)))
1052 if (ioctl == KVM_SET_ONE_REG)
1053 r = kvm_vcpu_ioctl_set_one_reg(vcpu, ®);
1055 r = kvm_vcpu_ioctl_get_one_reg(vcpu, ®);
1059 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1060 case KVM_DIRTY_TLB: {
1061 struct kvm_dirty_tlb dirty;
1063 if (copy_from_user(&dirty, argp, sizeof(dirty)))
1065 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
1077 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
1079 return VM_FAULT_SIGBUS;
1082 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
1084 u32 inst_nop = 0x60000000;
1085 #ifdef CONFIG_KVM_BOOKE_HV
1086 u32 inst_sc1 = 0x44000022;
1087 pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
1088 pvinfo->hcall[1] = cpu_to_be32(inst_nop);
1089 pvinfo->hcall[2] = cpu_to_be32(inst_nop);
1090 pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1092 u32 inst_lis = 0x3c000000;
1093 u32 inst_ori = 0x60000000;
1094 u32 inst_sc = 0x44000002;
1095 u32 inst_imm_mask = 0xffff;
1098 * The hypercall to get into KVM from within guest context is as
1101 * lis r0, r0, KVM_SC_MAGIC_R0@h
1102 * ori r0, KVM_SC_MAGIC_R0@l
1106 pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
1107 pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
1108 pvinfo->hcall[2] = cpu_to_be32(inst_sc);
1109 pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1112 pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
1117 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
1120 if (!irqchip_in_kernel(kvm))
1123 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1124 irq_event->irq, irq_event->level,
1130 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
1131 struct kvm_enable_cap *cap)
1139 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
1140 case KVM_CAP_PPC_ENABLE_HCALL: {
1141 unsigned long hcall = cap->args[0];
1144 if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
1147 if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
1150 set_bit(hcall / 4, kvm->arch.enabled_hcalls);
1152 clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
1165 long kvm_arch_vm_ioctl(struct file *filp,
1166 unsigned int ioctl, unsigned long arg)
1168 struct kvm *kvm __maybe_unused = filp->private_data;
1169 void __user *argp = (void __user *)arg;
1173 case KVM_PPC_GET_PVINFO: {
1174 struct kvm_ppc_pvinfo pvinfo;
1175 memset(&pvinfo, 0, sizeof(pvinfo));
1176 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
1177 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
1184 case KVM_ENABLE_CAP:
1186 struct kvm_enable_cap cap;
1188 if (copy_from_user(&cap, argp, sizeof(cap)))
1190 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
1193 #ifdef CONFIG_PPC_BOOK3S_64
1194 case KVM_CREATE_SPAPR_TCE: {
1195 struct kvm_create_spapr_tce create_tce;
1198 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
1200 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce);
1203 case KVM_PPC_GET_SMMU_INFO: {
1204 struct kvm_ppc_smmu_info info;
1205 struct kvm *kvm = filp->private_data;
1207 memset(&info, 0, sizeof(info));
1208 r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
1209 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
1213 case KVM_PPC_RTAS_DEFINE_TOKEN: {
1214 struct kvm *kvm = filp->private_data;
1216 r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
1220 struct kvm *kvm = filp->private_data;
1221 r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
1223 #else /* CONFIG_PPC_BOOK3S_64 */
1232 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
1233 static unsigned long nr_lpids;
1235 long kvmppc_alloc_lpid(void)
1240 lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
1241 if (lpid >= nr_lpids) {
1242 pr_err("%s: No LPIDs free\n", __func__);
1245 } while (test_and_set_bit(lpid, lpid_inuse));
1249 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
1251 void kvmppc_claim_lpid(long lpid)
1253 set_bit(lpid, lpid_inuse);
1255 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
1257 void kvmppc_free_lpid(long lpid)
1259 clear_bit(lpid, lpid_inuse);
1261 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
1263 void kvmppc_init_lpid(unsigned long nr_lpids_param)
1265 nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
1266 memset(lpid_inuse, 0, sizeof(lpid_inuse));
1268 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
1270 int kvm_arch_init(void *opaque)
1275 void kvm_arch_exit(void)