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>
34 #include "../mm/mmu_decl.h"
36 #define CREATE_TRACE_POINTS
39 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
41 #ifndef CONFIG_KVM_BOOK3S_64_HV
42 return !(v->arch.shared->msr & MSR_WE) ||
43 !!(v->arch.pending_exceptions);
45 return !(v->arch.ceded) || !!(v->arch.pending_exceptions);
49 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
51 int nr = kvmppc_get_gpr(vcpu, 11);
53 unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
54 unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
55 unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
56 unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
59 if (!(vcpu->arch.shared->msr & MSR_SF)) {
68 case HC_VENDOR_KVM | KVM_HC_PPC_MAP_MAGIC_PAGE:
70 vcpu->arch.magic_page_pa = param1;
71 vcpu->arch.magic_page_ea = param2;
73 r2 = KVM_MAGIC_FEAT_SR;
78 case HC_VENDOR_KVM | KVM_HC_FEATURES:
80 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500)
81 /* XXX Missing magic page on 44x */
82 r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
85 /* Second return value is in r4 */
88 r = HC_EV_UNIMPLEMENTED;
92 kvmppc_set_gpr(vcpu, 4, r2);
97 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
99 enum emulation_result er;
102 er = kvmppc_emulate_instruction(run, vcpu);
105 /* Future optimization: only reload non-volatiles if they were
106 * actually modified. */
109 case EMULATE_DO_MMIO:
110 run->exit_reason = KVM_EXIT_MMIO;
111 /* We must reload nonvolatiles because "update" load/store
112 * instructions modify register state. */
113 /* Future optimization: only reload non-volatiles if they were
114 * actually modified. */
118 /* XXX Deliver Program interrupt to guest. */
119 printk(KERN_EMERG "%s: emulation failed (%08x)\n", __func__,
120 kvmppc_get_last_inst(vcpu));
130 int kvm_arch_hardware_enable(void *garbage)
135 void kvm_arch_hardware_disable(void *garbage)
139 int kvm_arch_hardware_setup(void)
144 void kvm_arch_hardware_unsetup(void)
148 void kvm_arch_check_processor_compat(void *rtn)
150 *(int *)rtn = kvmppc_core_check_processor_compat();
153 int kvm_arch_init_vm(struct kvm *kvm)
155 return kvmppc_core_init_vm(kvm);
158 void kvm_arch_destroy_vm(struct kvm *kvm)
161 struct kvm_vcpu *vcpu;
163 kvm_for_each_vcpu(i, vcpu, kvm)
164 kvm_arch_vcpu_free(vcpu);
166 mutex_lock(&kvm->lock);
167 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
168 kvm->vcpus[i] = NULL;
170 atomic_set(&kvm->online_vcpus, 0);
172 kvmppc_core_destroy_vm(kvm);
174 mutex_unlock(&kvm->lock);
177 void kvm_arch_sync_events(struct kvm *kvm)
181 int kvm_dev_ioctl_check_extension(long ext)
187 case KVM_CAP_PPC_BOOKE_SREGS:
189 case KVM_CAP_PPC_SEGSTATE:
191 case KVM_CAP_PPC_UNSET_IRQ:
192 case KVM_CAP_PPC_IRQ_LEVEL:
193 case KVM_CAP_ENABLE_CAP:
196 #ifndef CONFIG_KVM_BOOK3S_64_HV
197 case KVM_CAP_PPC_PAIRED_SINGLES:
198 case KVM_CAP_PPC_OSI:
199 case KVM_CAP_PPC_GET_PVINFO:
202 case KVM_CAP_COALESCED_MMIO:
203 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
214 long kvm_arch_dev_ioctl(struct file *filp,
215 unsigned int ioctl, unsigned long arg)
220 int kvm_arch_prepare_memory_region(struct kvm *kvm,
221 struct kvm_memory_slot *memslot,
222 struct kvm_memory_slot old,
223 struct kvm_userspace_memory_region *mem,
226 return kvmppc_core_prepare_memory_region(kvm, mem);
229 void kvm_arch_commit_memory_region(struct kvm *kvm,
230 struct kvm_userspace_memory_region *mem,
231 struct kvm_memory_slot old,
234 kvmppc_core_commit_memory_region(kvm, mem);
238 void kvm_arch_flush_shadow(struct kvm *kvm)
242 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
244 struct kvm_vcpu *vcpu;
245 vcpu = kvmppc_core_vcpu_create(kvm, id);
247 kvmppc_create_vcpu_debugfs(vcpu, id);
251 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
253 /* Make sure we're not using the vcpu anymore */
254 hrtimer_cancel(&vcpu->arch.dec_timer);
255 tasklet_kill(&vcpu->arch.tasklet);
257 kvmppc_remove_vcpu_debugfs(vcpu);
258 kvmppc_core_vcpu_free(vcpu);
261 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
263 kvm_arch_vcpu_free(vcpu);
266 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
268 return kvmppc_core_pending_dec(vcpu);
271 static void kvmppc_decrementer_func(unsigned long data)
273 struct kvm_vcpu *vcpu = (struct kvm_vcpu *)data;
275 kvmppc_core_queue_dec(vcpu);
277 if (waitqueue_active(&vcpu->wq)) {
278 wake_up_interruptible(&vcpu->wq);
279 vcpu->stat.halt_wakeup++;
284 * low level hrtimer wake routine. Because this runs in hardirq context
285 * we schedule a tasklet to do the real work.
287 enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
289 struct kvm_vcpu *vcpu;
291 vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
292 tasklet_schedule(&vcpu->arch.tasklet);
294 return HRTIMER_NORESTART;
297 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
299 hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
300 tasklet_init(&vcpu->arch.tasklet, kvmppc_decrementer_func, (ulong)vcpu);
301 vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
302 vcpu->arch.dec_expires = ~(u64)0;
304 #ifdef CONFIG_KVM_EXIT_TIMING
305 mutex_init(&vcpu->arch.exit_timing_lock);
311 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
313 kvmppc_mmu_destroy(vcpu);
316 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
320 * vrsave (formerly usprg0) isn't used by Linux, but may
321 * be used by the guest.
323 * On non-booke this is associated with Altivec and
324 * is handled by code in book3s.c.
326 mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
328 kvmppc_core_vcpu_load(vcpu, cpu);
329 vcpu->cpu = smp_processor_id();
332 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
334 kvmppc_core_vcpu_put(vcpu);
336 vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
341 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
342 struct kvm_guest_debug *dbg)
347 static void kvmppc_complete_dcr_load(struct kvm_vcpu *vcpu,
350 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, run->dcr.data);
353 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
356 u64 uninitialized_var(gpr);
358 if (run->mmio.len > sizeof(gpr)) {
359 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
363 if (vcpu->arch.mmio_is_bigendian) {
364 switch (run->mmio.len) {
365 case 8: gpr = *(u64 *)run->mmio.data; break;
366 case 4: gpr = *(u32 *)run->mmio.data; break;
367 case 2: gpr = *(u16 *)run->mmio.data; break;
368 case 1: gpr = *(u8 *)run->mmio.data; break;
371 /* Convert BE data from userland back to LE. */
372 switch (run->mmio.len) {
373 case 4: gpr = ld_le32((u32 *)run->mmio.data); break;
374 case 2: gpr = ld_le16((u16 *)run->mmio.data); break;
375 case 1: gpr = *(u8 *)run->mmio.data; break;
379 if (vcpu->arch.mmio_sign_extend) {
380 switch (run->mmio.len) {
395 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
397 switch (vcpu->arch.io_gpr & KVM_REG_EXT_MASK) {
399 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
402 vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_REG_MASK] = gpr;
404 #ifdef CONFIG_PPC_BOOK3S
406 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_REG_MASK] = gpr;
409 vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_REG_MASK] = gpr;
410 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_REG_MASK] = gpr;
418 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
419 unsigned int rt, unsigned int bytes, int is_bigendian)
421 if (bytes > sizeof(run->mmio.data)) {
422 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
426 run->mmio.phys_addr = vcpu->arch.paddr_accessed;
427 run->mmio.len = bytes;
428 run->mmio.is_write = 0;
430 vcpu->arch.io_gpr = rt;
431 vcpu->arch.mmio_is_bigendian = is_bigendian;
432 vcpu->mmio_needed = 1;
433 vcpu->mmio_is_write = 0;
434 vcpu->arch.mmio_sign_extend = 0;
436 return EMULATE_DO_MMIO;
439 /* Same as above, but sign extends */
440 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
441 unsigned int rt, unsigned int bytes, int is_bigendian)
445 r = kvmppc_handle_load(run, vcpu, rt, bytes, is_bigendian);
446 vcpu->arch.mmio_sign_extend = 1;
451 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
452 u64 val, unsigned int bytes, int is_bigendian)
454 void *data = run->mmio.data;
456 if (bytes > sizeof(run->mmio.data)) {
457 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
461 run->mmio.phys_addr = vcpu->arch.paddr_accessed;
462 run->mmio.len = bytes;
463 run->mmio.is_write = 1;
464 vcpu->mmio_needed = 1;
465 vcpu->mmio_is_write = 1;
467 /* Store the value at the lowest bytes in 'data'. */
470 case 8: *(u64 *)data = val; break;
471 case 4: *(u32 *)data = val; break;
472 case 2: *(u16 *)data = val; break;
473 case 1: *(u8 *)data = val; break;
476 /* Store LE value into 'data'. */
478 case 4: st_le32(data, val); break;
479 case 2: st_le16(data, val); break;
480 case 1: *(u8 *)data = val; break;
484 return EMULATE_DO_MMIO;
487 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
492 if (vcpu->sigset_active)
493 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
495 if (vcpu->mmio_needed) {
496 if (!vcpu->mmio_is_write)
497 kvmppc_complete_mmio_load(vcpu, run);
498 vcpu->mmio_needed = 0;
499 } else if (vcpu->arch.dcr_needed) {
500 if (!vcpu->arch.dcr_is_write)
501 kvmppc_complete_dcr_load(vcpu, run);
502 vcpu->arch.dcr_needed = 0;
503 } else if (vcpu->arch.osi_needed) {
504 u64 *gprs = run->osi.gprs;
507 for (i = 0; i < 32; i++)
508 kvmppc_set_gpr(vcpu, i, gprs[i]);
509 vcpu->arch.osi_needed = 0;
510 } else if (vcpu->arch.hcall_needed) {
513 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
514 for (i = 0; i < 9; ++i)
515 kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
516 vcpu->arch.hcall_needed = 0;
519 kvmppc_core_deliver_interrupts(vcpu);
521 r = kvmppc_vcpu_run(run, vcpu);
523 if (vcpu->sigset_active)
524 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
529 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
531 if (irq->irq == KVM_INTERRUPT_UNSET)
532 kvmppc_core_dequeue_external(vcpu, irq);
534 kvmppc_core_queue_external(vcpu, irq);
536 if (waitqueue_active(&vcpu->wq)) {
537 wake_up_interruptible(&vcpu->wq);
538 vcpu->stat.halt_wakeup++;
539 } else if (vcpu->cpu != -1) {
540 smp_send_reschedule(vcpu->cpu);
546 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
547 struct kvm_enable_cap *cap)
555 case KVM_CAP_PPC_OSI:
557 vcpu->arch.osi_enabled = true;
567 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
568 struct kvm_mp_state *mp_state)
573 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
574 struct kvm_mp_state *mp_state)
579 long kvm_arch_vcpu_ioctl(struct file *filp,
580 unsigned int ioctl, unsigned long arg)
582 struct kvm_vcpu *vcpu = filp->private_data;
583 void __user *argp = (void __user *)arg;
587 case KVM_INTERRUPT: {
588 struct kvm_interrupt irq;
590 if (copy_from_user(&irq, argp, sizeof(irq)))
592 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
598 struct kvm_enable_cap cap;
600 if (copy_from_user(&cap, argp, sizeof(cap)))
602 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
613 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
615 u32 inst_lis = 0x3c000000;
616 u32 inst_ori = 0x60000000;
617 u32 inst_nop = 0x60000000;
618 u32 inst_sc = 0x44000002;
619 u32 inst_imm_mask = 0xffff;
622 * The hypercall to get into KVM from within guest context is as
625 * lis r0, r0, KVM_SC_MAGIC_R0@h
626 * ori r0, KVM_SC_MAGIC_R0@l
630 pvinfo->hcall[0] = inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask);
631 pvinfo->hcall[1] = inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask);
632 pvinfo->hcall[2] = inst_sc;
633 pvinfo->hcall[3] = inst_nop;
638 long kvm_arch_vm_ioctl(struct file *filp,
639 unsigned int ioctl, unsigned long arg)
641 void __user *argp = (void __user *)arg;
645 case KVM_PPC_GET_PVINFO: {
646 struct kvm_ppc_pvinfo pvinfo;
647 memset(&pvinfo, 0, sizeof(pvinfo));
648 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
649 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
664 int kvm_arch_init(void *opaque)
669 void kvm_arch_exit(void)