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 kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
51 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
53 int nr = kvmppc_get_gpr(vcpu, 11);
55 unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
56 unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
57 unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
58 unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
61 if (!(vcpu->arch.shared->msr & MSR_SF)) {
70 case HC_VENDOR_KVM | KVM_HC_PPC_MAP_MAGIC_PAGE:
72 vcpu->arch.magic_page_pa = param1;
73 vcpu->arch.magic_page_ea = param2;
75 r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
80 case HC_VENDOR_KVM | KVM_HC_FEATURES:
82 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
83 /* XXX Missing magic page on 44x */
84 r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
87 /* Second return value is in r4 */
90 r = HC_EV_UNIMPLEMENTED;
94 kvmppc_set_gpr(vcpu, 4, r2);
99 int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
103 /* We have to know what CPU to virtualize */
107 /* PAPR only works with book3s_64 */
108 if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
111 #ifdef CONFIG_KVM_BOOK3S_64_HV
112 /* HV KVM can only do PAPR mode for now */
113 if (!vcpu->arch.papr_enabled)
117 #ifdef CONFIG_KVM_BOOKE_HV
118 if (!cpu_has_feature(CPU_FTR_EMB_HV))
126 return r ? 0 : -EINVAL;
129 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
131 enum emulation_result er;
134 er = kvmppc_emulate_instruction(run, vcpu);
137 /* Future optimization: only reload non-volatiles if they were
138 * actually modified. */
141 case EMULATE_DO_MMIO:
142 run->exit_reason = KVM_EXIT_MMIO;
143 /* We must reload nonvolatiles because "update" load/store
144 * instructions modify register state. */
145 /* Future optimization: only reload non-volatiles if they were
146 * actually modified. */
150 /* XXX Deliver Program interrupt to guest. */
151 printk(KERN_EMERG "%s: emulation failed (%08x)\n", __func__,
152 kvmppc_get_last_inst(vcpu));
162 int kvm_arch_hardware_enable(void *garbage)
167 void kvm_arch_hardware_disable(void *garbage)
171 int kvm_arch_hardware_setup(void)
176 void kvm_arch_hardware_unsetup(void)
180 void kvm_arch_check_processor_compat(void *rtn)
182 *(int *)rtn = kvmppc_core_check_processor_compat();
185 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
190 return kvmppc_core_init_vm(kvm);
193 void kvm_arch_destroy_vm(struct kvm *kvm)
196 struct kvm_vcpu *vcpu;
198 kvm_for_each_vcpu(i, vcpu, kvm)
199 kvm_arch_vcpu_free(vcpu);
201 mutex_lock(&kvm->lock);
202 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
203 kvm->vcpus[i] = NULL;
205 atomic_set(&kvm->online_vcpus, 0);
207 kvmppc_core_destroy_vm(kvm);
209 mutex_unlock(&kvm->lock);
212 void kvm_arch_sync_events(struct kvm *kvm)
216 int kvm_dev_ioctl_check_extension(long ext)
222 case KVM_CAP_PPC_BOOKE_SREGS:
224 case KVM_CAP_PPC_SEGSTATE:
225 case KVM_CAP_PPC_HIOR:
226 case KVM_CAP_PPC_PAPR:
228 case KVM_CAP_PPC_UNSET_IRQ:
229 case KVM_CAP_PPC_IRQ_LEVEL:
230 case KVM_CAP_ENABLE_CAP:
231 case KVM_CAP_ONE_REG:
234 #ifndef CONFIG_KVM_BOOK3S_64_HV
235 case KVM_CAP_PPC_PAIRED_SINGLES:
236 case KVM_CAP_PPC_OSI:
237 case KVM_CAP_PPC_GET_PVINFO:
238 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
243 case KVM_CAP_COALESCED_MMIO:
244 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
247 #ifdef CONFIG_PPC_BOOK3S_64
248 case KVM_CAP_SPAPR_TCE:
251 #endif /* CONFIG_PPC_BOOK3S_64 */
252 #ifdef CONFIG_KVM_BOOK3S_64_HV
253 case KVM_CAP_PPC_SMT:
254 r = threads_per_core;
256 case KVM_CAP_PPC_RMA:
258 /* PPC970 requires an RMA */
259 if (cpu_has_feature(CPU_FTR_ARCH_201))
262 case KVM_CAP_SYNC_MMU:
263 r = cpu_has_feature(CPU_FTR_ARCH_206) ? 1 : 0;
266 case KVM_CAP_NR_VCPUS:
268 * Recommending a number of CPUs is somewhat arbitrary; we
269 * return the number of present CPUs for -HV (since a host
270 * will have secondary threads "offline"), and for other KVM
271 * implementations just count online CPUs.
273 #ifdef CONFIG_KVM_BOOK3S_64_HV
274 r = num_present_cpus();
276 r = num_online_cpus();
279 case KVM_CAP_MAX_VCPUS:
282 #ifdef CONFIG_PPC_BOOK3S_64
283 case KVM_CAP_PPC_GET_SMMU_INFO:
295 long kvm_arch_dev_ioctl(struct file *filp,
296 unsigned int ioctl, unsigned long arg)
301 void kvm_arch_free_memslot(struct kvm_memory_slot *free,
302 struct kvm_memory_slot *dont)
306 int kvm_arch_create_memslot(struct kvm_memory_slot *slot, unsigned long npages)
311 int kvm_arch_prepare_memory_region(struct kvm *kvm,
312 struct kvm_memory_slot *memslot,
313 struct kvm_memory_slot old,
314 struct kvm_userspace_memory_region *mem,
317 return kvmppc_core_prepare_memory_region(kvm, mem);
320 void kvm_arch_commit_memory_region(struct kvm *kvm,
321 struct kvm_userspace_memory_region *mem,
322 struct kvm_memory_slot old,
325 kvmppc_core_commit_memory_region(kvm, mem);
329 void kvm_arch_flush_shadow(struct kvm *kvm)
333 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
335 struct kvm_vcpu *vcpu;
336 vcpu = kvmppc_core_vcpu_create(kvm, id);
338 vcpu->arch.wqp = &vcpu->wq;
339 kvmppc_create_vcpu_debugfs(vcpu, id);
344 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
346 /* Make sure we're not using the vcpu anymore */
347 hrtimer_cancel(&vcpu->arch.dec_timer);
348 tasklet_kill(&vcpu->arch.tasklet);
350 kvmppc_remove_vcpu_debugfs(vcpu);
351 kvmppc_core_vcpu_free(vcpu);
354 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
356 kvm_arch_vcpu_free(vcpu);
359 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
361 return kvmppc_core_pending_dec(vcpu);
365 * low level hrtimer wake routine. Because this runs in hardirq context
366 * we schedule a tasklet to do the real work.
368 enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
370 struct kvm_vcpu *vcpu;
372 vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
373 tasklet_schedule(&vcpu->arch.tasklet);
375 return HRTIMER_NORESTART;
378 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
380 hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
381 tasklet_init(&vcpu->arch.tasklet, kvmppc_decrementer_func, (ulong)vcpu);
382 vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
383 vcpu->arch.dec_expires = ~(u64)0;
385 #ifdef CONFIG_KVM_EXIT_TIMING
386 mutex_init(&vcpu->arch.exit_timing_lock);
392 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
394 kvmppc_mmu_destroy(vcpu);
397 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
401 * vrsave (formerly usprg0) isn't used by Linux, but may
402 * be used by the guest.
404 * On non-booke this is associated with Altivec and
405 * is handled by code in book3s.c.
407 mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
409 kvmppc_core_vcpu_load(vcpu, cpu);
410 vcpu->cpu = smp_processor_id();
413 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
415 kvmppc_core_vcpu_put(vcpu);
417 vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
422 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
423 struct kvm_guest_debug *dbg)
428 static void kvmppc_complete_dcr_load(struct kvm_vcpu *vcpu,
431 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, run->dcr.data);
434 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
437 u64 uninitialized_var(gpr);
439 if (run->mmio.len > sizeof(gpr)) {
440 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
444 if (vcpu->arch.mmio_is_bigendian) {
445 switch (run->mmio.len) {
446 case 8: gpr = *(u64 *)run->mmio.data; break;
447 case 4: gpr = *(u32 *)run->mmio.data; break;
448 case 2: gpr = *(u16 *)run->mmio.data; break;
449 case 1: gpr = *(u8 *)run->mmio.data; break;
452 /* Convert BE data from userland back to LE. */
453 switch (run->mmio.len) {
454 case 4: gpr = ld_le32((u32 *)run->mmio.data); break;
455 case 2: gpr = ld_le16((u16 *)run->mmio.data); break;
456 case 1: gpr = *(u8 *)run->mmio.data; break;
460 if (vcpu->arch.mmio_sign_extend) {
461 switch (run->mmio.len) {
476 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
478 switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
479 case KVM_MMIO_REG_GPR:
480 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
482 case KVM_MMIO_REG_FPR:
483 vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
485 #ifdef CONFIG_PPC_BOOK3S
486 case KVM_MMIO_REG_QPR:
487 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
489 case KVM_MMIO_REG_FQPR:
490 vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
491 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
499 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
500 unsigned int rt, unsigned int bytes, int is_bigendian)
502 if (bytes > sizeof(run->mmio.data)) {
503 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
507 run->mmio.phys_addr = vcpu->arch.paddr_accessed;
508 run->mmio.len = bytes;
509 run->mmio.is_write = 0;
511 vcpu->arch.io_gpr = rt;
512 vcpu->arch.mmio_is_bigendian = is_bigendian;
513 vcpu->mmio_needed = 1;
514 vcpu->mmio_is_write = 0;
515 vcpu->arch.mmio_sign_extend = 0;
517 return EMULATE_DO_MMIO;
520 /* Same as above, but sign extends */
521 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
522 unsigned int rt, unsigned int bytes, int is_bigendian)
526 r = kvmppc_handle_load(run, vcpu, rt, bytes, is_bigendian);
527 vcpu->arch.mmio_sign_extend = 1;
532 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
533 u64 val, unsigned int bytes, int is_bigendian)
535 void *data = run->mmio.data;
537 if (bytes > sizeof(run->mmio.data)) {
538 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
542 run->mmio.phys_addr = vcpu->arch.paddr_accessed;
543 run->mmio.len = bytes;
544 run->mmio.is_write = 1;
545 vcpu->mmio_needed = 1;
546 vcpu->mmio_is_write = 1;
548 /* Store the value at the lowest bytes in 'data'. */
551 case 8: *(u64 *)data = val; break;
552 case 4: *(u32 *)data = val; break;
553 case 2: *(u16 *)data = val; break;
554 case 1: *(u8 *)data = val; break;
557 /* Store LE value into 'data'. */
559 case 4: st_le32(data, val); break;
560 case 2: st_le16(data, val); break;
561 case 1: *(u8 *)data = val; break;
565 return EMULATE_DO_MMIO;
568 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
573 if (vcpu->sigset_active)
574 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
576 if (vcpu->mmio_needed) {
577 if (!vcpu->mmio_is_write)
578 kvmppc_complete_mmio_load(vcpu, run);
579 vcpu->mmio_needed = 0;
580 } else if (vcpu->arch.dcr_needed) {
581 if (!vcpu->arch.dcr_is_write)
582 kvmppc_complete_dcr_load(vcpu, run);
583 vcpu->arch.dcr_needed = 0;
584 } else if (vcpu->arch.osi_needed) {
585 u64 *gprs = run->osi.gprs;
588 for (i = 0; i < 32; i++)
589 kvmppc_set_gpr(vcpu, i, gprs[i]);
590 vcpu->arch.osi_needed = 0;
591 } else if (vcpu->arch.hcall_needed) {
594 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
595 for (i = 0; i < 9; ++i)
596 kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
597 vcpu->arch.hcall_needed = 0;
600 r = kvmppc_vcpu_run(run, vcpu);
602 if (vcpu->sigset_active)
603 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
608 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
610 if (irq->irq == KVM_INTERRUPT_UNSET) {
611 kvmppc_core_dequeue_external(vcpu, irq);
615 kvmppc_core_queue_external(vcpu, irq);
622 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
623 struct kvm_enable_cap *cap)
631 case KVM_CAP_PPC_OSI:
633 vcpu->arch.osi_enabled = true;
635 case KVM_CAP_PPC_PAPR:
637 vcpu->arch.papr_enabled = true;
639 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
640 case KVM_CAP_SW_TLB: {
641 struct kvm_config_tlb cfg;
642 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
645 if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
648 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
658 r = kvmppc_sanity_check(vcpu);
663 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
664 struct kvm_mp_state *mp_state)
669 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
670 struct kvm_mp_state *mp_state)
675 long kvm_arch_vcpu_ioctl(struct file *filp,
676 unsigned int ioctl, unsigned long arg)
678 struct kvm_vcpu *vcpu = filp->private_data;
679 void __user *argp = (void __user *)arg;
683 case KVM_INTERRUPT: {
684 struct kvm_interrupt irq;
686 if (copy_from_user(&irq, argp, sizeof(irq)))
688 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
694 struct kvm_enable_cap cap;
696 if (copy_from_user(&cap, argp, sizeof(cap)))
698 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
702 case KVM_SET_ONE_REG:
703 case KVM_GET_ONE_REG:
705 struct kvm_one_reg reg;
707 if (copy_from_user(®, argp, sizeof(reg)))
709 if (ioctl == KVM_SET_ONE_REG)
710 r = kvm_vcpu_ioctl_set_one_reg(vcpu, ®);
712 r = kvm_vcpu_ioctl_get_one_reg(vcpu, ®);
716 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
717 case KVM_DIRTY_TLB: {
718 struct kvm_dirty_tlb dirty;
720 if (copy_from_user(&dirty, argp, sizeof(dirty)))
722 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
734 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
736 return VM_FAULT_SIGBUS;
739 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
741 u32 inst_lis = 0x3c000000;
742 u32 inst_ori = 0x60000000;
743 u32 inst_nop = 0x60000000;
744 u32 inst_sc = 0x44000002;
745 u32 inst_imm_mask = 0xffff;
748 * The hypercall to get into KVM from within guest context is as
751 * lis r0, r0, KVM_SC_MAGIC_R0@h
752 * ori r0, KVM_SC_MAGIC_R0@l
756 pvinfo->hcall[0] = inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask);
757 pvinfo->hcall[1] = inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask);
758 pvinfo->hcall[2] = inst_sc;
759 pvinfo->hcall[3] = inst_nop;
764 long kvm_arch_vm_ioctl(struct file *filp,
765 unsigned int ioctl, unsigned long arg)
767 void __user *argp = (void __user *)arg;
771 case KVM_PPC_GET_PVINFO: {
772 struct kvm_ppc_pvinfo pvinfo;
773 memset(&pvinfo, 0, sizeof(pvinfo));
774 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
775 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
782 #ifdef CONFIG_PPC_BOOK3S_64
783 case KVM_CREATE_SPAPR_TCE: {
784 struct kvm_create_spapr_tce create_tce;
785 struct kvm *kvm = filp->private_data;
788 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
790 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce);
793 #endif /* CONFIG_PPC_BOOK3S_64 */
795 #ifdef CONFIG_KVM_BOOK3S_64_HV
796 case KVM_ALLOCATE_RMA: {
797 struct kvm *kvm = filp->private_data;
798 struct kvm_allocate_rma rma;
800 r = kvm_vm_ioctl_allocate_rma(kvm, &rma);
801 if (r >= 0 && copy_to_user(argp, &rma, sizeof(rma)))
805 #endif /* CONFIG_KVM_BOOK3S_64_HV */
807 #ifdef CONFIG_PPC_BOOK3S_64
808 case KVM_PPC_GET_SMMU_INFO: {
809 struct kvm *kvm = filp->private_data;
810 struct kvm_ppc_smmu_info info;
812 memset(&info, 0, sizeof(info));
813 r = kvm_vm_ioctl_get_smmu_info(kvm, &info);
814 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
818 #endif /* CONFIG_PPC_BOOK3S_64 */
827 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
828 static unsigned long nr_lpids;
830 long kvmppc_alloc_lpid(void)
835 lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
836 if (lpid >= nr_lpids) {
837 pr_err("%s: No LPIDs free\n", __func__);
840 } while (test_and_set_bit(lpid, lpid_inuse));
845 void kvmppc_claim_lpid(long lpid)
847 set_bit(lpid, lpid_inuse);
850 void kvmppc_free_lpid(long lpid)
852 clear_bit(lpid, lpid_inuse);
855 void kvmppc_init_lpid(unsigned long nr_lpids_param)
857 nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
858 memset(lpid_inuse, 0, sizeof(lpid_inuse));
861 int kvm_arch_init(void *opaque)
866 void kvm_arch_exit(void)