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_HIOR:
216 case KVM_CAP_PPC_PAPR:
218 case KVM_CAP_PPC_UNSET_IRQ:
219 case KVM_CAP_PPC_IRQ_LEVEL:
220 case KVM_CAP_ENABLE_CAP:
221 case KVM_CAP_ONE_REG:
224 #ifndef CONFIG_KVM_BOOK3S_64_HV
225 case KVM_CAP_PPC_PAIRED_SINGLES:
226 case KVM_CAP_PPC_OSI:
227 case KVM_CAP_PPC_GET_PVINFO:
228 #ifdef CONFIG_KVM_E500
233 case KVM_CAP_COALESCED_MMIO:
234 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
237 #ifdef CONFIG_KVM_BOOK3S_64_HV
238 case KVM_CAP_SPAPR_TCE:
241 case KVM_CAP_PPC_SMT:
242 r = threads_per_core;
244 case KVM_CAP_PPC_RMA:
246 /* PPC970 requires an RMA */
247 if (cpu_has_feature(CPU_FTR_ARCH_201))
250 case KVM_CAP_SYNC_MMU:
251 r = cpu_has_feature(CPU_FTR_ARCH_206) ? 1 : 0;
254 case KVM_CAP_NR_VCPUS:
256 * Recommending a number of CPUs is somewhat arbitrary; we
257 * return the number of present CPUs for -HV (since a host
258 * will have secondary threads "offline"), and for other KVM
259 * implementations just count online CPUs.
261 #ifdef CONFIG_KVM_BOOK3S_64_HV
262 r = num_present_cpus();
264 r = num_online_cpus();
267 case KVM_CAP_MAX_VCPUS:
278 long kvm_arch_dev_ioctl(struct file *filp,
279 unsigned int ioctl, unsigned long arg)
284 int kvm_arch_prepare_memory_region(struct kvm *kvm,
285 struct kvm_memory_slot *memslot,
286 struct kvm_memory_slot old,
287 struct kvm_userspace_memory_region *mem,
290 return kvmppc_core_prepare_memory_region(kvm, mem);
293 void kvm_arch_commit_memory_region(struct kvm *kvm,
294 struct kvm_userspace_memory_region *mem,
295 struct kvm_memory_slot old,
298 kvmppc_core_commit_memory_region(kvm, mem);
302 void kvm_arch_flush_shadow(struct kvm *kvm)
306 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
308 struct kvm_vcpu *vcpu;
309 vcpu = kvmppc_core_vcpu_create(kvm, id);
311 vcpu->arch.wqp = &vcpu->wq;
312 kvmppc_create_vcpu_debugfs(vcpu, id);
317 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
319 /* Make sure we're not using the vcpu anymore */
320 hrtimer_cancel(&vcpu->arch.dec_timer);
321 tasklet_kill(&vcpu->arch.tasklet);
323 kvmppc_remove_vcpu_debugfs(vcpu);
324 kvmppc_core_vcpu_free(vcpu);
327 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
329 kvm_arch_vcpu_free(vcpu);
332 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
334 return kvmppc_core_pending_dec(vcpu);
338 * low level hrtimer wake routine. Because this runs in hardirq context
339 * we schedule a tasklet to do the real work.
341 enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
343 struct kvm_vcpu *vcpu;
345 vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
346 tasklet_schedule(&vcpu->arch.tasklet);
348 return HRTIMER_NORESTART;
351 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
353 hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
354 tasklet_init(&vcpu->arch.tasklet, kvmppc_decrementer_func, (ulong)vcpu);
355 vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
356 vcpu->arch.dec_expires = ~(u64)0;
358 #ifdef CONFIG_KVM_EXIT_TIMING
359 mutex_init(&vcpu->arch.exit_timing_lock);
365 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
367 kvmppc_mmu_destroy(vcpu);
370 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
374 * vrsave (formerly usprg0) isn't used by Linux, but may
375 * be used by the guest.
377 * On non-booke this is associated with Altivec and
378 * is handled by code in book3s.c.
380 mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
382 kvmppc_core_vcpu_load(vcpu, cpu);
383 vcpu->cpu = smp_processor_id();
386 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
388 kvmppc_core_vcpu_put(vcpu);
390 vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
395 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
396 struct kvm_guest_debug *dbg)
401 static void kvmppc_complete_dcr_load(struct kvm_vcpu *vcpu,
404 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, run->dcr.data);
407 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
410 u64 uninitialized_var(gpr);
412 if (run->mmio.len > sizeof(gpr)) {
413 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
417 if (vcpu->arch.mmio_is_bigendian) {
418 switch (run->mmio.len) {
419 case 8: gpr = *(u64 *)run->mmio.data; break;
420 case 4: gpr = *(u32 *)run->mmio.data; break;
421 case 2: gpr = *(u16 *)run->mmio.data; break;
422 case 1: gpr = *(u8 *)run->mmio.data; break;
425 /* Convert BE data from userland back to LE. */
426 switch (run->mmio.len) {
427 case 4: gpr = ld_le32((u32 *)run->mmio.data); break;
428 case 2: gpr = ld_le16((u16 *)run->mmio.data); break;
429 case 1: gpr = *(u8 *)run->mmio.data; break;
433 if (vcpu->arch.mmio_sign_extend) {
434 switch (run->mmio.len) {
449 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
451 switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
452 case KVM_MMIO_REG_GPR:
453 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
455 case KVM_MMIO_REG_FPR:
456 vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
458 #ifdef CONFIG_PPC_BOOK3S
459 case KVM_MMIO_REG_QPR:
460 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
462 case KVM_MMIO_REG_FQPR:
463 vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
464 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
472 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
473 unsigned int rt, unsigned int bytes, int is_bigendian)
475 if (bytes > sizeof(run->mmio.data)) {
476 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
480 run->mmio.phys_addr = vcpu->arch.paddr_accessed;
481 run->mmio.len = bytes;
482 run->mmio.is_write = 0;
484 vcpu->arch.io_gpr = rt;
485 vcpu->arch.mmio_is_bigendian = is_bigendian;
486 vcpu->mmio_needed = 1;
487 vcpu->mmio_is_write = 0;
488 vcpu->arch.mmio_sign_extend = 0;
490 return EMULATE_DO_MMIO;
493 /* Same as above, but sign extends */
494 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
495 unsigned int rt, unsigned int bytes, int is_bigendian)
499 r = kvmppc_handle_load(run, vcpu, rt, bytes, is_bigendian);
500 vcpu->arch.mmio_sign_extend = 1;
505 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
506 u64 val, unsigned int bytes, int is_bigendian)
508 void *data = run->mmio.data;
510 if (bytes > sizeof(run->mmio.data)) {
511 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
515 run->mmio.phys_addr = vcpu->arch.paddr_accessed;
516 run->mmio.len = bytes;
517 run->mmio.is_write = 1;
518 vcpu->mmio_needed = 1;
519 vcpu->mmio_is_write = 1;
521 /* Store the value at the lowest bytes in 'data'. */
524 case 8: *(u64 *)data = val; break;
525 case 4: *(u32 *)data = val; break;
526 case 2: *(u16 *)data = val; break;
527 case 1: *(u8 *)data = val; break;
530 /* Store LE value into 'data'. */
532 case 4: st_le32(data, val); break;
533 case 2: st_le16(data, val); break;
534 case 1: *(u8 *)data = val; break;
538 return EMULATE_DO_MMIO;
541 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
546 if (vcpu->sigset_active)
547 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
549 if (vcpu->mmio_needed) {
550 if (!vcpu->mmio_is_write)
551 kvmppc_complete_mmio_load(vcpu, run);
552 vcpu->mmio_needed = 0;
553 } else if (vcpu->arch.dcr_needed) {
554 if (!vcpu->arch.dcr_is_write)
555 kvmppc_complete_dcr_load(vcpu, run);
556 vcpu->arch.dcr_needed = 0;
557 } else if (vcpu->arch.osi_needed) {
558 u64 *gprs = run->osi.gprs;
561 for (i = 0; i < 32; i++)
562 kvmppc_set_gpr(vcpu, i, gprs[i]);
563 vcpu->arch.osi_needed = 0;
564 } else if (vcpu->arch.hcall_needed) {
567 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
568 for (i = 0; i < 9; ++i)
569 kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
570 vcpu->arch.hcall_needed = 0;
573 r = kvmppc_vcpu_run(run, vcpu);
575 if (vcpu->sigset_active)
576 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
581 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
587 if (waitqueue_active(vcpu->arch.wqp)) {
588 wake_up_interruptible(vcpu->arch.wqp);
589 vcpu->stat.halt_wakeup++;
590 } else if (cpu != me && cpu != -1) {
591 smp_send_reschedule(vcpu->cpu);
596 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
598 if (irq->irq == KVM_INTERRUPT_UNSET) {
599 kvmppc_core_dequeue_external(vcpu, irq);
603 kvmppc_core_queue_external(vcpu, irq);
609 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
610 struct kvm_enable_cap *cap)
618 case KVM_CAP_PPC_OSI:
620 vcpu->arch.osi_enabled = true;
622 case KVM_CAP_PPC_PAPR:
624 vcpu->arch.papr_enabled = true;
626 #ifdef CONFIG_KVM_E500
627 case KVM_CAP_SW_TLB: {
628 struct kvm_config_tlb cfg;
629 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
632 if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
635 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
645 r = kvmppc_sanity_check(vcpu);
650 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
651 struct kvm_mp_state *mp_state)
656 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
657 struct kvm_mp_state *mp_state)
662 long kvm_arch_vcpu_ioctl(struct file *filp,
663 unsigned int ioctl, unsigned long arg)
665 struct kvm_vcpu *vcpu = filp->private_data;
666 void __user *argp = (void __user *)arg;
670 case KVM_INTERRUPT: {
671 struct kvm_interrupt irq;
673 if (copy_from_user(&irq, argp, sizeof(irq)))
675 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
681 struct kvm_enable_cap cap;
683 if (copy_from_user(&cap, argp, sizeof(cap)))
685 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
689 case KVM_SET_ONE_REG:
690 case KVM_GET_ONE_REG:
692 struct kvm_one_reg reg;
694 if (copy_from_user(®, argp, sizeof(reg)))
696 if (ioctl == KVM_SET_ONE_REG)
697 r = kvm_vcpu_ioctl_set_one_reg(vcpu, ®);
699 r = kvm_vcpu_ioctl_get_one_reg(vcpu, ®);
703 #ifdef CONFIG_KVM_E500
704 case KVM_DIRTY_TLB: {
705 struct kvm_dirty_tlb dirty;
707 if (copy_from_user(&dirty, argp, sizeof(dirty)))
709 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
722 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
724 return VM_FAULT_SIGBUS;
727 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
729 u32 inst_lis = 0x3c000000;
730 u32 inst_ori = 0x60000000;
731 u32 inst_nop = 0x60000000;
732 u32 inst_sc = 0x44000002;
733 u32 inst_imm_mask = 0xffff;
736 * The hypercall to get into KVM from within guest context is as
739 * lis r0, r0, KVM_SC_MAGIC_R0@h
740 * ori r0, KVM_SC_MAGIC_R0@l
744 pvinfo->hcall[0] = inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask);
745 pvinfo->hcall[1] = inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask);
746 pvinfo->hcall[2] = inst_sc;
747 pvinfo->hcall[3] = inst_nop;
752 long kvm_arch_vm_ioctl(struct file *filp,
753 unsigned int ioctl, unsigned long arg)
755 void __user *argp = (void __user *)arg;
759 case KVM_PPC_GET_PVINFO: {
760 struct kvm_ppc_pvinfo pvinfo;
761 memset(&pvinfo, 0, sizeof(pvinfo));
762 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
763 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
770 #ifdef CONFIG_KVM_BOOK3S_64_HV
771 case KVM_CREATE_SPAPR_TCE: {
772 struct kvm_create_spapr_tce create_tce;
773 struct kvm *kvm = filp->private_data;
776 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
778 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce);
782 case KVM_ALLOCATE_RMA: {
783 struct kvm *kvm = filp->private_data;
784 struct kvm_allocate_rma rma;
786 r = kvm_vm_ioctl_allocate_rma(kvm, &rma);
787 if (r >= 0 && copy_to_user(argp, &rma, sizeof(rma)))
791 #endif /* CONFIG_KVM_BOOK3S_64_HV */
801 int kvm_arch_init(void *opaque)
806 void kvm_arch_exit(void)