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/module.h>
25 #include <linux/vmalloc.h>
26 #include <linux/hrtimer.h>
28 #include <linux/slab.h>
29 #include <asm/cputable.h>
30 #include <asm/uaccess.h>
31 #include <asm/kvm_ppc.h>
32 #include <asm/tlbflush.h>
33 #include <asm/cputhreads.h>
35 #include "../mm/mmu_decl.h"
37 #define CREATE_TRACE_POINTS
40 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
42 return !(v->arch.shared->msr & MSR_WE) ||
43 !!(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;
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)
177 return kvmppc_core_init_vm(kvm);
180 void kvm_arch_destroy_vm(struct kvm *kvm)
183 struct kvm_vcpu *vcpu;
185 kvm_for_each_vcpu(i, vcpu, kvm)
186 kvm_arch_vcpu_free(vcpu);
188 mutex_lock(&kvm->lock);
189 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
190 kvm->vcpus[i] = NULL;
192 atomic_set(&kvm->online_vcpus, 0);
194 kvmppc_core_destroy_vm(kvm);
196 mutex_unlock(&kvm->lock);
199 void kvm_arch_sync_events(struct kvm *kvm)
203 int kvm_dev_ioctl_check_extension(long ext)
209 case KVM_CAP_PPC_BOOKE_SREGS:
211 case KVM_CAP_PPC_SEGSTATE:
212 case KVM_CAP_PPC_HIOR:
213 case KVM_CAP_PPC_PAPR:
215 case KVM_CAP_PPC_UNSET_IRQ:
216 case KVM_CAP_PPC_IRQ_LEVEL:
217 case KVM_CAP_ENABLE_CAP:
220 #ifndef CONFIG_KVM_BOOK3S_64_HV
221 case KVM_CAP_PPC_PAIRED_SINGLES:
222 case KVM_CAP_PPC_OSI:
223 case KVM_CAP_PPC_GET_PVINFO:
226 case KVM_CAP_COALESCED_MMIO:
227 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
230 #ifdef CONFIG_KVM_BOOK3S_64_HV
231 case KVM_CAP_SPAPR_TCE:
234 case KVM_CAP_PPC_SMT:
235 r = threads_per_core;
237 case KVM_CAP_PPC_RMA:
239 /* PPC970 requires an RMA */
240 if (cpu_has_feature(CPU_FTR_ARCH_201))
252 long kvm_arch_dev_ioctl(struct file *filp,
253 unsigned int ioctl, unsigned long arg)
258 int kvm_arch_prepare_memory_region(struct kvm *kvm,
259 struct kvm_memory_slot *memslot,
260 struct kvm_memory_slot old,
261 struct kvm_userspace_memory_region *mem,
264 return kvmppc_core_prepare_memory_region(kvm, mem);
267 void kvm_arch_commit_memory_region(struct kvm *kvm,
268 struct kvm_userspace_memory_region *mem,
269 struct kvm_memory_slot old,
272 kvmppc_core_commit_memory_region(kvm, mem);
276 void kvm_arch_flush_shadow(struct kvm *kvm)
280 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
282 struct kvm_vcpu *vcpu;
283 vcpu = kvmppc_core_vcpu_create(kvm, id);
284 vcpu->arch.wqp = &vcpu->wq;
286 kvmppc_create_vcpu_debugfs(vcpu, id);
290 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
292 /* Make sure we're not using the vcpu anymore */
293 hrtimer_cancel(&vcpu->arch.dec_timer);
294 tasklet_kill(&vcpu->arch.tasklet);
296 kvmppc_remove_vcpu_debugfs(vcpu);
297 kvmppc_core_vcpu_free(vcpu);
300 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
302 kvm_arch_vcpu_free(vcpu);
305 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
307 return kvmppc_core_pending_dec(vcpu);
310 static void kvmppc_decrementer_func(unsigned long data)
312 struct kvm_vcpu *vcpu = (struct kvm_vcpu *)data;
314 kvmppc_core_queue_dec(vcpu);
316 if (waitqueue_active(vcpu->arch.wqp)) {
317 wake_up_interruptible(vcpu->arch.wqp);
318 vcpu->stat.halt_wakeup++;
323 * low level hrtimer wake routine. Because this runs in hardirq context
324 * we schedule a tasklet to do the real work.
326 enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
328 struct kvm_vcpu *vcpu;
330 vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
331 tasklet_schedule(&vcpu->arch.tasklet);
333 return HRTIMER_NORESTART;
336 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
338 hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
339 tasklet_init(&vcpu->arch.tasklet, kvmppc_decrementer_func, (ulong)vcpu);
340 vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
341 vcpu->arch.dec_expires = ~(u64)0;
343 #ifdef CONFIG_KVM_EXIT_TIMING
344 mutex_init(&vcpu->arch.exit_timing_lock);
350 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
352 kvmppc_mmu_destroy(vcpu);
355 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
359 * vrsave (formerly usprg0) isn't used by Linux, but may
360 * be used by the guest.
362 * On non-booke this is associated with Altivec and
363 * is handled by code in book3s.c.
365 mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
367 kvmppc_core_vcpu_load(vcpu, cpu);
368 vcpu->cpu = smp_processor_id();
371 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
373 kvmppc_core_vcpu_put(vcpu);
375 vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
380 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
381 struct kvm_guest_debug *dbg)
386 static void kvmppc_complete_dcr_load(struct kvm_vcpu *vcpu,
389 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, run->dcr.data);
392 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
395 u64 uninitialized_var(gpr);
397 if (run->mmio.len > sizeof(gpr)) {
398 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
402 if (vcpu->arch.mmio_is_bigendian) {
403 switch (run->mmio.len) {
404 case 8: gpr = *(u64 *)run->mmio.data; break;
405 case 4: gpr = *(u32 *)run->mmio.data; break;
406 case 2: gpr = *(u16 *)run->mmio.data; break;
407 case 1: gpr = *(u8 *)run->mmio.data; break;
410 /* Convert BE data from userland back to LE. */
411 switch (run->mmio.len) {
412 case 4: gpr = ld_le32((u32 *)run->mmio.data); break;
413 case 2: gpr = ld_le16((u16 *)run->mmio.data); break;
414 case 1: gpr = *(u8 *)run->mmio.data; break;
418 if (vcpu->arch.mmio_sign_extend) {
419 switch (run->mmio.len) {
434 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
436 switch (vcpu->arch.io_gpr & KVM_REG_EXT_MASK) {
438 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
441 vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_REG_MASK] = gpr;
443 #ifdef CONFIG_PPC_BOOK3S
445 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_REG_MASK] = gpr;
448 vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_REG_MASK] = gpr;
449 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_REG_MASK] = gpr;
457 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
458 unsigned int rt, unsigned int bytes, int is_bigendian)
460 if (bytes > sizeof(run->mmio.data)) {
461 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
465 run->mmio.phys_addr = vcpu->arch.paddr_accessed;
466 run->mmio.len = bytes;
467 run->mmio.is_write = 0;
469 vcpu->arch.io_gpr = rt;
470 vcpu->arch.mmio_is_bigendian = is_bigendian;
471 vcpu->mmio_needed = 1;
472 vcpu->mmio_is_write = 0;
473 vcpu->arch.mmio_sign_extend = 0;
475 return EMULATE_DO_MMIO;
478 /* Same as above, but sign extends */
479 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
480 unsigned int rt, unsigned int bytes, int is_bigendian)
484 r = kvmppc_handle_load(run, vcpu, rt, bytes, is_bigendian);
485 vcpu->arch.mmio_sign_extend = 1;
490 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
491 u64 val, unsigned int bytes, int is_bigendian)
493 void *data = run->mmio.data;
495 if (bytes > sizeof(run->mmio.data)) {
496 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
500 run->mmio.phys_addr = vcpu->arch.paddr_accessed;
501 run->mmio.len = bytes;
502 run->mmio.is_write = 1;
503 vcpu->mmio_needed = 1;
504 vcpu->mmio_is_write = 1;
506 /* Store the value at the lowest bytes in 'data'. */
509 case 8: *(u64 *)data = val; break;
510 case 4: *(u32 *)data = val; break;
511 case 2: *(u16 *)data = val; break;
512 case 1: *(u8 *)data = val; break;
515 /* Store LE value into 'data'. */
517 case 4: st_le32(data, val); break;
518 case 2: st_le16(data, val); break;
519 case 1: *(u8 *)data = val; break;
523 return EMULATE_DO_MMIO;
526 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
531 if (vcpu->sigset_active)
532 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
534 if (vcpu->mmio_needed) {
535 if (!vcpu->mmio_is_write)
536 kvmppc_complete_mmio_load(vcpu, run);
537 vcpu->mmio_needed = 0;
538 } else if (vcpu->arch.dcr_needed) {
539 if (!vcpu->arch.dcr_is_write)
540 kvmppc_complete_dcr_load(vcpu, run);
541 vcpu->arch.dcr_needed = 0;
542 } else if (vcpu->arch.osi_needed) {
543 u64 *gprs = run->osi.gprs;
546 for (i = 0; i < 32; i++)
547 kvmppc_set_gpr(vcpu, i, gprs[i]);
548 vcpu->arch.osi_needed = 0;
549 } else if (vcpu->arch.hcall_needed) {
552 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
553 for (i = 0; i < 9; ++i)
554 kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
555 vcpu->arch.hcall_needed = 0;
558 kvmppc_core_deliver_interrupts(vcpu);
560 r = kvmppc_vcpu_run(run, vcpu);
562 if (vcpu->sigset_active)
563 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
568 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
570 if (irq->irq == KVM_INTERRUPT_UNSET) {
571 kvmppc_core_dequeue_external(vcpu, irq);
575 kvmppc_core_queue_external(vcpu, irq);
577 if (waitqueue_active(vcpu->arch.wqp)) {
578 wake_up_interruptible(vcpu->arch.wqp);
579 vcpu->stat.halt_wakeup++;
580 } else if (vcpu->cpu != -1) {
581 smp_send_reschedule(vcpu->cpu);
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;
610 r = kvmppc_sanity_check(vcpu);
615 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
616 struct kvm_mp_state *mp_state)
621 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
622 struct kvm_mp_state *mp_state)
627 long kvm_arch_vcpu_ioctl(struct file *filp,
628 unsigned int ioctl, unsigned long arg)
630 struct kvm_vcpu *vcpu = filp->private_data;
631 void __user *argp = (void __user *)arg;
635 case KVM_INTERRUPT: {
636 struct kvm_interrupt irq;
638 if (copy_from_user(&irq, argp, sizeof(irq)))
640 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
646 struct kvm_enable_cap cap;
648 if (copy_from_user(&cap, argp, sizeof(cap)))
650 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
661 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
663 u32 inst_lis = 0x3c000000;
664 u32 inst_ori = 0x60000000;
665 u32 inst_nop = 0x60000000;
666 u32 inst_sc = 0x44000002;
667 u32 inst_imm_mask = 0xffff;
670 * The hypercall to get into KVM from within guest context is as
673 * lis r0, r0, KVM_SC_MAGIC_R0@h
674 * ori r0, KVM_SC_MAGIC_R0@l
678 pvinfo->hcall[0] = inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask);
679 pvinfo->hcall[1] = inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask);
680 pvinfo->hcall[2] = inst_sc;
681 pvinfo->hcall[3] = inst_nop;
686 long kvm_arch_vm_ioctl(struct file *filp,
687 unsigned int ioctl, unsigned long arg)
689 void __user *argp = (void __user *)arg;
693 case KVM_PPC_GET_PVINFO: {
694 struct kvm_ppc_pvinfo pvinfo;
695 memset(&pvinfo, 0, sizeof(pvinfo));
696 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
697 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
704 #ifdef CONFIG_KVM_BOOK3S_64_HV
705 case KVM_CREATE_SPAPR_TCE: {
706 struct kvm_create_spapr_tce create_tce;
707 struct kvm *kvm = filp->private_data;
710 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
712 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce);
716 case KVM_ALLOCATE_RMA: {
717 struct kvm *kvm = filp->private_data;
718 struct kvm_allocate_rma rma;
720 r = kvm_vm_ioctl_allocate_rma(kvm, &rma);
721 if (r >= 0 && copy_to_user(argp, &rma, sizeof(rma)))
725 #endif /* CONFIG_KVM_BOOK3S_64_HV */
735 int kvm_arch_init(void *opaque)
740 void kvm_arch_exit(void)