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 <asm/cputable.h>
29 #include <asm/uaccess.h>
30 #include <asm/kvm_ppc.h>
31 #include <asm/tlbflush.h>
32 #include <asm/cputhreads.h>
34 #include "../mm/mmu_decl.h"
36 #define CREATE_TRACE_POINTS
39 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
41 return !(v->arch.shared->msr & MSR_WE) ||
42 !!(v->arch.pending_exceptions) ||
46 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
48 int nr = kvmppc_get_gpr(vcpu, 11);
50 unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
51 unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
52 unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
53 unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
56 if (!(vcpu->arch.shared->msr & MSR_SF)) {
65 case HC_VENDOR_KVM | KVM_HC_PPC_MAP_MAGIC_PAGE:
67 vcpu->arch.magic_page_pa = param1;
68 vcpu->arch.magic_page_ea = param2;
70 r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
75 case HC_VENDOR_KVM | KVM_HC_FEATURES:
77 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500)
78 /* XXX Missing magic page on 44x */
79 r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
82 /* Second return value is in r4 */
85 r = HC_EV_UNIMPLEMENTED;
89 kvmppc_set_gpr(vcpu, 4, r2);
94 int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
98 /* We have to know what CPU to virtualize */
102 /* PAPR only works with book3s_64 */
103 if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
106 #ifdef CONFIG_KVM_BOOK3S_64_HV
107 /* HV KVM can only do PAPR mode for now */
108 if (!vcpu->arch.papr_enabled)
116 return r ? 0 : -EINVAL;
119 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
121 enum emulation_result er;
124 er = kvmppc_emulate_instruction(run, vcpu);
127 /* Future optimization: only reload non-volatiles if they were
128 * actually modified. */
131 case EMULATE_DO_MMIO:
132 run->exit_reason = KVM_EXIT_MMIO;
133 /* We must reload nonvolatiles because "update" load/store
134 * instructions modify register state. */
135 /* Future optimization: only reload non-volatiles if they were
136 * actually modified. */
140 /* XXX Deliver Program interrupt to guest. */
141 printk(KERN_EMERG "%s: emulation failed (%08x)\n", __func__,
142 kvmppc_get_last_inst(vcpu));
152 int kvm_arch_hardware_enable(void *garbage)
157 void kvm_arch_hardware_disable(void *garbage)
161 int kvm_arch_hardware_setup(void)
166 void kvm_arch_hardware_unsetup(void)
170 void kvm_arch_check_processor_compat(void *rtn)
172 *(int *)rtn = kvmppc_core_check_processor_compat();
175 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
180 return kvmppc_core_init_vm(kvm);
183 void kvm_arch_destroy_vm(struct kvm *kvm)
186 struct kvm_vcpu *vcpu;
188 kvm_for_each_vcpu(i, vcpu, kvm)
189 kvm_arch_vcpu_free(vcpu);
191 mutex_lock(&kvm->lock);
192 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
193 kvm->vcpus[i] = NULL;
195 atomic_set(&kvm->online_vcpus, 0);
197 kvmppc_core_destroy_vm(kvm);
199 mutex_unlock(&kvm->lock);
202 void kvm_arch_sync_events(struct kvm *kvm)
206 int kvm_dev_ioctl_check_extension(long ext)
212 case KVM_CAP_PPC_BOOKE_SREGS:
214 case KVM_CAP_PPC_SEGSTATE:
215 case KVM_CAP_PPC_PAPR:
217 case KVM_CAP_PPC_UNSET_IRQ:
218 case KVM_CAP_PPC_IRQ_LEVEL:
219 case KVM_CAP_ENABLE_CAP:
222 #ifndef CONFIG_KVM_BOOK3S_64_HV
223 case KVM_CAP_PPC_PAIRED_SINGLES:
224 case KVM_CAP_PPC_OSI:
225 case KVM_CAP_PPC_GET_PVINFO:
226 #ifdef CONFIG_KVM_E500
231 case KVM_CAP_COALESCED_MMIO:
232 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
235 #ifdef CONFIG_KVM_BOOK3S_64_HV
236 case KVM_CAP_SPAPR_TCE:
239 case KVM_CAP_PPC_SMT:
240 r = threads_per_core;
242 case KVM_CAP_PPC_RMA:
244 /* PPC970 requires an RMA */
245 if (cpu_has_feature(CPU_FTR_ARCH_201))
257 long kvm_arch_dev_ioctl(struct file *filp,
258 unsigned int ioctl, unsigned long arg)
263 int kvm_arch_prepare_memory_region(struct kvm *kvm,
264 struct kvm_memory_slot *memslot,
265 struct kvm_memory_slot old,
266 struct kvm_userspace_memory_region *mem,
269 return kvmppc_core_prepare_memory_region(kvm, mem);
272 void kvm_arch_commit_memory_region(struct kvm *kvm,
273 struct kvm_userspace_memory_region *mem,
274 struct kvm_memory_slot old,
277 kvmppc_core_commit_memory_region(kvm, mem);
281 void kvm_arch_flush_shadow(struct kvm *kvm)
285 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
287 struct kvm_vcpu *vcpu;
288 vcpu = kvmppc_core_vcpu_create(kvm, id);
289 vcpu->arch.wqp = &vcpu->wq;
291 kvmppc_create_vcpu_debugfs(vcpu, id);
295 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
297 /* Make sure we're not using the vcpu anymore */
298 hrtimer_cancel(&vcpu->arch.dec_timer);
299 tasklet_kill(&vcpu->arch.tasklet);
301 kvmppc_remove_vcpu_debugfs(vcpu);
302 kvmppc_core_vcpu_free(vcpu);
305 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
307 kvm_arch_vcpu_free(vcpu);
310 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
312 return kvmppc_core_pending_dec(vcpu);
316 * low level hrtimer wake routine. Because this runs in hardirq context
317 * we schedule a tasklet to do the real work.
319 enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
321 struct kvm_vcpu *vcpu;
323 vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
324 tasklet_schedule(&vcpu->arch.tasklet);
326 return HRTIMER_NORESTART;
329 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
331 hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
332 tasklet_init(&vcpu->arch.tasklet, kvmppc_decrementer_func, (ulong)vcpu);
333 vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
334 vcpu->arch.dec_expires = ~(u64)0;
336 #ifdef CONFIG_KVM_EXIT_TIMING
337 mutex_init(&vcpu->arch.exit_timing_lock);
343 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
345 kvmppc_mmu_destroy(vcpu);
348 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
352 * vrsave (formerly usprg0) isn't used by Linux, but may
353 * be used by the guest.
355 * On non-booke this is associated with Altivec and
356 * is handled by code in book3s.c.
358 mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
360 kvmppc_core_vcpu_load(vcpu, cpu);
361 vcpu->cpu = smp_processor_id();
364 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
366 kvmppc_core_vcpu_put(vcpu);
368 vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
373 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
374 struct kvm_guest_debug *dbg)
379 static void kvmppc_complete_dcr_load(struct kvm_vcpu *vcpu,
382 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, run->dcr.data);
385 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
388 u64 uninitialized_var(gpr);
390 if (run->mmio.len > sizeof(gpr)) {
391 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
395 if (vcpu->arch.mmio_is_bigendian) {
396 switch (run->mmio.len) {
397 case 8: gpr = *(u64 *)run->mmio.data; break;
398 case 4: gpr = *(u32 *)run->mmio.data; break;
399 case 2: gpr = *(u16 *)run->mmio.data; break;
400 case 1: gpr = *(u8 *)run->mmio.data; break;
403 /* Convert BE data from userland back to LE. */
404 switch (run->mmio.len) {
405 case 4: gpr = ld_le32((u32 *)run->mmio.data); break;
406 case 2: gpr = ld_le16((u16 *)run->mmio.data); break;
407 case 1: gpr = *(u8 *)run->mmio.data; break;
411 if (vcpu->arch.mmio_sign_extend) {
412 switch (run->mmio.len) {
427 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
429 switch (vcpu->arch.io_gpr & KVM_REG_EXT_MASK) {
431 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
434 vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_REG_MASK] = gpr;
436 #ifdef CONFIG_PPC_BOOK3S
438 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_REG_MASK] = gpr;
441 vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_REG_MASK] = gpr;
442 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_REG_MASK] = gpr;
450 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
451 unsigned int rt, unsigned int bytes, int is_bigendian)
453 if (bytes > sizeof(run->mmio.data)) {
454 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
458 run->mmio.phys_addr = vcpu->arch.paddr_accessed;
459 run->mmio.len = bytes;
460 run->mmio.is_write = 0;
462 vcpu->arch.io_gpr = rt;
463 vcpu->arch.mmio_is_bigendian = is_bigendian;
464 vcpu->mmio_needed = 1;
465 vcpu->mmio_is_write = 0;
466 vcpu->arch.mmio_sign_extend = 0;
468 return EMULATE_DO_MMIO;
471 /* Same as above, but sign extends */
472 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
473 unsigned int rt, unsigned int bytes, int is_bigendian)
477 r = kvmppc_handle_load(run, vcpu, rt, bytes, is_bigendian);
478 vcpu->arch.mmio_sign_extend = 1;
483 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
484 u64 val, unsigned int bytes, int is_bigendian)
486 void *data = run->mmio.data;
488 if (bytes > sizeof(run->mmio.data)) {
489 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
493 run->mmio.phys_addr = vcpu->arch.paddr_accessed;
494 run->mmio.len = bytes;
495 run->mmio.is_write = 1;
496 vcpu->mmio_needed = 1;
497 vcpu->mmio_is_write = 1;
499 /* Store the value at the lowest bytes in 'data'. */
502 case 8: *(u64 *)data = val; break;
503 case 4: *(u32 *)data = val; break;
504 case 2: *(u16 *)data = val; break;
505 case 1: *(u8 *)data = val; break;
508 /* Store LE value into 'data'. */
510 case 4: st_le32(data, val); break;
511 case 2: st_le16(data, val); break;
512 case 1: *(u8 *)data = val; break;
516 return EMULATE_DO_MMIO;
519 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
524 if (vcpu->sigset_active)
525 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
527 if (vcpu->mmio_needed) {
528 if (!vcpu->mmio_is_write)
529 kvmppc_complete_mmio_load(vcpu, run);
530 vcpu->mmio_needed = 0;
531 } else if (vcpu->arch.dcr_needed) {
532 if (!vcpu->arch.dcr_is_write)
533 kvmppc_complete_dcr_load(vcpu, run);
534 vcpu->arch.dcr_needed = 0;
535 } else if (vcpu->arch.osi_needed) {
536 u64 *gprs = run->osi.gprs;
539 for (i = 0; i < 32; i++)
540 kvmppc_set_gpr(vcpu, i, gprs[i]);
541 vcpu->arch.osi_needed = 0;
542 } else if (vcpu->arch.hcall_needed) {
545 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
546 for (i = 0; i < 9; ++i)
547 kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
548 vcpu->arch.hcall_needed = 0;
551 r = kvmppc_vcpu_run(run, vcpu);
553 if (vcpu->sigset_active)
554 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
559 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
565 if (waitqueue_active(vcpu->arch.wqp)) {
566 wake_up_interruptible(vcpu->arch.wqp);
567 vcpu->stat.halt_wakeup++;
568 } else if (cpu != me && cpu != -1) {
569 smp_send_reschedule(vcpu->cpu);
574 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
576 if (irq->irq == KVM_INTERRUPT_UNSET) {
577 kvmppc_core_dequeue_external(vcpu, irq);
581 kvmppc_core_queue_external(vcpu, irq);
587 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
588 struct kvm_enable_cap *cap)
596 case KVM_CAP_PPC_OSI:
598 vcpu->arch.osi_enabled = true;
600 case KVM_CAP_PPC_PAPR:
602 vcpu->arch.papr_enabled = true;
604 #ifdef CONFIG_KVM_E500
605 case KVM_CAP_SW_TLB: {
606 struct kvm_config_tlb cfg;
607 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
610 if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
613 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
623 r = kvmppc_sanity_check(vcpu);
628 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
629 struct kvm_mp_state *mp_state)
634 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
635 struct kvm_mp_state *mp_state)
640 long kvm_arch_vcpu_ioctl(struct file *filp,
641 unsigned int ioctl, unsigned long arg)
643 struct kvm_vcpu *vcpu = filp->private_data;
644 void __user *argp = (void __user *)arg;
648 case KVM_INTERRUPT: {
649 struct kvm_interrupt irq;
651 if (copy_from_user(&irq, argp, sizeof(irq)))
653 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
659 struct kvm_enable_cap cap;
661 if (copy_from_user(&cap, argp, sizeof(cap)))
663 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
667 #ifdef CONFIG_KVM_E500
668 case KVM_DIRTY_TLB: {
669 struct kvm_dirty_tlb dirty;
671 if (copy_from_user(&dirty, argp, sizeof(dirty)))
673 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
686 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
688 return VM_FAULT_SIGBUS;
691 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
693 u32 inst_lis = 0x3c000000;
694 u32 inst_ori = 0x60000000;
695 u32 inst_nop = 0x60000000;
696 u32 inst_sc = 0x44000002;
697 u32 inst_imm_mask = 0xffff;
700 * The hypercall to get into KVM from within guest context is as
703 * lis r0, r0, KVM_SC_MAGIC_R0@h
704 * ori r0, KVM_SC_MAGIC_R0@l
708 pvinfo->hcall[0] = inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask);
709 pvinfo->hcall[1] = inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask);
710 pvinfo->hcall[2] = inst_sc;
711 pvinfo->hcall[3] = inst_nop;
716 long kvm_arch_vm_ioctl(struct file *filp,
717 unsigned int ioctl, unsigned long arg)
719 void __user *argp = (void __user *)arg;
723 case KVM_PPC_GET_PVINFO: {
724 struct kvm_ppc_pvinfo pvinfo;
725 memset(&pvinfo, 0, sizeof(pvinfo));
726 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
727 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
734 #ifdef CONFIG_KVM_BOOK3S_64_HV
735 case KVM_CREATE_SPAPR_TCE: {
736 struct kvm_create_spapr_tce create_tce;
737 struct kvm *kvm = filp->private_data;
740 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
742 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce);
746 case KVM_ALLOCATE_RMA: {
747 struct kvm *kvm = filp->private_data;
748 struct kvm_allocate_rma rma;
750 r = kvm_vm_ioctl_allocate_rma(kvm, &rma);
751 if (r >= 0 && copy_to_user(argp, &rma, sizeof(rma)))
755 #endif /* CONFIG_KVM_BOOK3S_64_HV */
765 int kvm_arch_init(void *opaque)
770 void kvm_arch_exit(void)