Add linux-next specific files for 20110831
[linux-2.6/next.git] / arch / powerpc / kvm / powerpc.c
blob3cef31e7c6767800fe1b281a5d4d9dec4a59f412
1 /*
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>
26 #include <linux/fs.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>
33 #include "timing.h"
34 #include "../mm/mmu_decl.h"
36 #define CREATE_TRACE_POINTS
37 #include "trace.h"
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);
44 #else
45 return !(v->arch.ceded) || !!(v->arch.pending_exceptions);
46 #endif
49 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
51 int nr = kvmppc_get_gpr(vcpu, 11);
52 int r;
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);
57 unsigned long r2 = 0;
59 if (!(vcpu->arch.shared->msr & MSR_SF)) {
60 /* 32 bit mode */
61 param1 &= 0xffffffff;
62 param2 &= 0xffffffff;
63 param3 &= 0xffffffff;
64 param4 &= 0xffffffff;
67 switch (nr) {
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;
75 r = HC_EV_SUCCESS;
76 break;
78 case HC_VENDOR_KVM | KVM_HC_FEATURES:
79 r = HC_EV_SUCCESS;
80 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500)
81 /* XXX Missing magic page on 44x */
82 r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
83 #endif
85 /* Second return value is in r4 */
86 break;
87 default:
88 r = HC_EV_UNIMPLEMENTED;
89 break;
92 kvmppc_set_gpr(vcpu, 4, r2);
94 return r;
97 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
99 enum emulation_result er;
100 int r;
102 er = kvmppc_emulate_instruction(run, vcpu);
103 switch (er) {
104 case EMULATE_DONE:
105 /* Future optimization: only reload non-volatiles if they were
106 * actually modified. */
107 r = RESUME_GUEST_NV;
108 break;
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. */
115 r = RESUME_HOST_NV;
116 break;
117 case EMULATE_FAIL:
118 /* XXX Deliver Program interrupt to guest. */
119 printk(KERN_EMERG "%s: emulation failed (%08x)\n", __func__,
120 kvmppc_get_last_inst(vcpu));
121 r = RESUME_HOST;
122 break;
123 default:
124 BUG();
127 return r;
130 int kvm_arch_hardware_enable(void *garbage)
132 return 0;
135 void kvm_arch_hardware_disable(void *garbage)
139 int kvm_arch_hardware_setup(void)
141 return 0;
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)
160 unsigned int i;
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)
183 int r;
185 switch (ext) {
186 #ifdef CONFIG_BOOKE
187 case KVM_CAP_PPC_BOOKE_SREGS:
188 #else
189 case KVM_CAP_PPC_SEGSTATE:
190 #endif
191 case KVM_CAP_PPC_UNSET_IRQ:
192 case KVM_CAP_PPC_IRQ_LEVEL:
193 case KVM_CAP_ENABLE_CAP:
194 r = 1;
195 break;
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:
200 r = 1;
201 break;
202 case KVM_CAP_COALESCED_MMIO:
203 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
204 break;
205 #endif
206 #ifdef CONFIG_KVM_BOOK3S_64_HV
207 case KVM_CAP_SPAPR_TCE:
208 r = 1;
209 break;
210 case KVM_CAP_PPC_SMT:
211 r = threads_per_core;
212 break;
213 case KVM_CAP_PPC_RMA:
214 r = 1;
215 /* PPC970 requires an RMA */
216 if (cpu_has_feature(CPU_FTR_ARCH_201))
217 r = 2;
218 break;
219 #endif
220 default:
221 r = 0;
222 break;
224 return r;
228 long kvm_arch_dev_ioctl(struct file *filp,
229 unsigned int ioctl, unsigned long arg)
231 return -EINVAL;
234 int kvm_arch_prepare_memory_region(struct kvm *kvm,
235 struct kvm_memory_slot *memslot,
236 struct kvm_memory_slot old,
237 struct kvm_userspace_memory_region *mem,
238 int user_alloc)
240 return kvmppc_core_prepare_memory_region(kvm, mem);
243 void kvm_arch_commit_memory_region(struct kvm *kvm,
244 struct kvm_userspace_memory_region *mem,
245 struct kvm_memory_slot old,
246 int user_alloc)
248 kvmppc_core_commit_memory_region(kvm, mem);
252 void kvm_arch_flush_shadow(struct kvm *kvm)
256 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
258 struct kvm_vcpu *vcpu;
259 vcpu = kvmppc_core_vcpu_create(kvm, id);
260 if (!IS_ERR(vcpu))
261 kvmppc_create_vcpu_debugfs(vcpu, id);
262 return vcpu;
265 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
267 /* Make sure we're not using the vcpu anymore */
268 hrtimer_cancel(&vcpu->arch.dec_timer);
269 tasklet_kill(&vcpu->arch.tasklet);
271 kvmppc_remove_vcpu_debugfs(vcpu);
272 kvmppc_core_vcpu_free(vcpu);
275 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
277 kvm_arch_vcpu_free(vcpu);
280 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
282 return kvmppc_core_pending_dec(vcpu);
285 static void kvmppc_decrementer_func(unsigned long data)
287 struct kvm_vcpu *vcpu = (struct kvm_vcpu *)data;
289 kvmppc_core_queue_dec(vcpu);
291 if (waitqueue_active(&vcpu->wq)) {
292 wake_up_interruptible(&vcpu->wq);
293 vcpu->stat.halt_wakeup++;
298 * low level hrtimer wake routine. Because this runs in hardirq context
299 * we schedule a tasklet to do the real work.
301 enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
303 struct kvm_vcpu *vcpu;
305 vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
306 tasklet_schedule(&vcpu->arch.tasklet);
308 return HRTIMER_NORESTART;
311 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
313 hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
314 tasklet_init(&vcpu->arch.tasklet, kvmppc_decrementer_func, (ulong)vcpu);
315 vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
316 vcpu->arch.dec_expires = ~(u64)0;
318 #ifdef CONFIG_KVM_EXIT_TIMING
319 mutex_init(&vcpu->arch.exit_timing_lock);
320 #endif
322 return 0;
325 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
327 kvmppc_mmu_destroy(vcpu);
330 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
332 #ifdef CONFIG_BOOKE
334 * vrsave (formerly usprg0) isn't used by Linux, but may
335 * be used by the guest.
337 * On non-booke this is associated with Altivec and
338 * is handled by code in book3s.c.
340 mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
341 #endif
342 kvmppc_core_vcpu_load(vcpu, cpu);
343 vcpu->cpu = smp_processor_id();
346 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
348 kvmppc_core_vcpu_put(vcpu);
349 #ifdef CONFIG_BOOKE
350 vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
351 #endif
352 vcpu->cpu = -1;
355 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
356 struct kvm_guest_debug *dbg)
358 return -EINVAL;
361 static void kvmppc_complete_dcr_load(struct kvm_vcpu *vcpu,
362 struct kvm_run *run)
364 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, run->dcr.data);
367 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
368 struct kvm_run *run)
370 u64 uninitialized_var(gpr);
372 if (run->mmio.len > sizeof(gpr)) {
373 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
374 return;
377 if (vcpu->arch.mmio_is_bigendian) {
378 switch (run->mmio.len) {
379 case 8: gpr = *(u64 *)run->mmio.data; break;
380 case 4: gpr = *(u32 *)run->mmio.data; break;
381 case 2: gpr = *(u16 *)run->mmio.data; break;
382 case 1: gpr = *(u8 *)run->mmio.data; break;
384 } else {
385 /* Convert BE data from userland back to LE. */
386 switch (run->mmio.len) {
387 case 4: gpr = ld_le32((u32 *)run->mmio.data); break;
388 case 2: gpr = ld_le16((u16 *)run->mmio.data); break;
389 case 1: gpr = *(u8 *)run->mmio.data; break;
393 if (vcpu->arch.mmio_sign_extend) {
394 switch (run->mmio.len) {
395 #ifdef CONFIG_PPC64
396 case 4:
397 gpr = (s64)(s32)gpr;
398 break;
399 #endif
400 case 2:
401 gpr = (s64)(s16)gpr;
402 break;
403 case 1:
404 gpr = (s64)(s8)gpr;
405 break;
409 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
411 switch (vcpu->arch.io_gpr & KVM_REG_EXT_MASK) {
412 case KVM_REG_GPR:
413 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
414 break;
415 case KVM_REG_FPR:
416 vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_REG_MASK] = gpr;
417 break;
418 #ifdef CONFIG_PPC_BOOK3S
419 case KVM_REG_QPR:
420 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_REG_MASK] = gpr;
421 break;
422 case KVM_REG_FQPR:
423 vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_REG_MASK] = gpr;
424 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_REG_MASK] = gpr;
425 break;
426 #endif
427 default:
428 BUG();
432 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
433 unsigned int rt, unsigned int bytes, int is_bigendian)
435 if (bytes > sizeof(run->mmio.data)) {
436 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
437 run->mmio.len);
440 run->mmio.phys_addr = vcpu->arch.paddr_accessed;
441 run->mmio.len = bytes;
442 run->mmio.is_write = 0;
444 vcpu->arch.io_gpr = rt;
445 vcpu->arch.mmio_is_bigendian = is_bigendian;
446 vcpu->mmio_needed = 1;
447 vcpu->mmio_is_write = 0;
448 vcpu->arch.mmio_sign_extend = 0;
450 return EMULATE_DO_MMIO;
453 /* Same as above, but sign extends */
454 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
455 unsigned int rt, unsigned int bytes, int is_bigendian)
457 int r;
459 r = kvmppc_handle_load(run, vcpu, rt, bytes, is_bigendian);
460 vcpu->arch.mmio_sign_extend = 1;
462 return r;
465 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
466 u64 val, unsigned int bytes, int is_bigendian)
468 void *data = run->mmio.data;
470 if (bytes > sizeof(run->mmio.data)) {
471 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
472 run->mmio.len);
475 run->mmio.phys_addr = vcpu->arch.paddr_accessed;
476 run->mmio.len = bytes;
477 run->mmio.is_write = 1;
478 vcpu->mmio_needed = 1;
479 vcpu->mmio_is_write = 1;
481 /* Store the value at the lowest bytes in 'data'. */
482 if (is_bigendian) {
483 switch (bytes) {
484 case 8: *(u64 *)data = val; break;
485 case 4: *(u32 *)data = val; break;
486 case 2: *(u16 *)data = val; break;
487 case 1: *(u8 *)data = val; break;
489 } else {
490 /* Store LE value into 'data'. */
491 switch (bytes) {
492 case 4: st_le32(data, val); break;
493 case 2: st_le16(data, val); break;
494 case 1: *(u8 *)data = val; break;
498 return EMULATE_DO_MMIO;
501 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
503 int r;
504 sigset_t sigsaved;
506 if (vcpu->sigset_active)
507 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
509 if (vcpu->mmio_needed) {
510 if (!vcpu->mmio_is_write)
511 kvmppc_complete_mmio_load(vcpu, run);
512 vcpu->mmio_needed = 0;
513 } else if (vcpu->arch.dcr_needed) {
514 if (!vcpu->arch.dcr_is_write)
515 kvmppc_complete_dcr_load(vcpu, run);
516 vcpu->arch.dcr_needed = 0;
517 } else if (vcpu->arch.osi_needed) {
518 u64 *gprs = run->osi.gprs;
519 int i;
521 for (i = 0; i < 32; i++)
522 kvmppc_set_gpr(vcpu, i, gprs[i]);
523 vcpu->arch.osi_needed = 0;
524 } else if (vcpu->arch.hcall_needed) {
525 int i;
527 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
528 for (i = 0; i < 9; ++i)
529 kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
530 vcpu->arch.hcall_needed = 0;
533 kvmppc_core_deliver_interrupts(vcpu);
535 r = kvmppc_vcpu_run(run, vcpu);
537 if (vcpu->sigset_active)
538 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
540 return r;
543 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
545 if (irq->irq == KVM_INTERRUPT_UNSET)
546 kvmppc_core_dequeue_external(vcpu, irq);
547 else
548 kvmppc_core_queue_external(vcpu, irq);
550 if (waitqueue_active(&vcpu->wq)) {
551 wake_up_interruptible(&vcpu->wq);
552 vcpu->stat.halt_wakeup++;
553 } else if (vcpu->cpu != -1) {
554 smp_send_reschedule(vcpu->cpu);
557 return 0;
560 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
561 struct kvm_enable_cap *cap)
563 int r;
565 if (cap->flags)
566 return -EINVAL;
568 switch (cap->cap) {
569 case KVM_CAP_PPC_OSI:
570 r = 0;
571 vcpu->arch.osi_enabled = true;
572 break;
573 default:
574 r = -EINVAL;
575 break;
578 return r;
581 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
582 struct kvm_mp_state *mp_state)
584 return -EINVAL;
587 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
588 struct kvm_mp_state *mp_state)
590 return -EINVAL;
593 long kvm_arch_vcpu_ioctl(struct file *filp,
594 unsigned int ioctl, unsigned long arg)
596 struct kvm_vcpu *vcpu = filp->private_data;
597 void __user *argp = (void __user *)arg;
598 long r;
600 switch (ioctl) {
601 case KVM_INTERRUPT: {
602 struct kvm_interrupt irq;
603 r = -EFAULT;
604 if (copy_from_user(&irq, argp, sizeof(irq)))
605 goto out;
606 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
607 goto out;
610 case KVM_ENABLE_CAP:
612 struct kvm_enable_cap cap;
613 r = -EFAULT;
614 if (copy_from_user(&cap, argp, sizeof(cap)))
615 goto out;
616 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
617 break;
619 default:
620 r = -EINVAL;
623 out:
624 return r;
627 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
629 u32 inst_lis = 0x3c000000;
630 u32 inst_ori = 0x60000000;
631 u32 inst_nop = 0x60000000;
632 u32 inst_sc = 0x44000002;
633 u32 inst_imm_mask = 0xffff;
636 * The hypercall to get into KVM from within guest context is as
637 * follows:
639 * lis r0, r0, KVM_SC_MAGIC_R0@h
640 * ori r0, KVM_SC_MAGIC_R0@l
641 * sc
642 * nop
644 pvinfo->hcall[0] = inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask);
645 pvinfo->hcall[1] = inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask);
646 pvinfo->hcall[2] = inst_sc;
647 pvinfo->hcall[3] = inst_nop;
649 return 0;
652 long kvm_arch_vm_ioctl(struct file *filp,
653 unsigned int ioctl, unsigned long arg)
655 void __user *argp = (void __user *)arg;
656 long r;
658 switch (ioctl) {
659 case KVM_PPC_GET_PVINFO: {
660 struct kvm_ppc_pvinfo pvinfo;
661 memset(&pvinfo, 0, sizeof(pvinfo));
662 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
663 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
664 r = -EFAULT;
665 goto out;
668 break;
670 #ifdef CONFIG_KVM_BOOK3S_64_HV
671 case KVM_CREATE_SPAPR_TCE: {
672 struct kvm_create_spapr_tce create_tce;
673 struct kvm *kvm = filp->private_data;
675 r = -EFAULT;
676 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
677 goto out;
678 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce);
679 goto out;
682 case KVM_ALLOCATE_RMA: {
683 struct kvm *kvm = filp->private_data;
684 struct kvm_allocate_rma rma;
686 r = kvm_vm_ioctl_allocate_rma(kvm, &rma);
687 if (r >= 0 && copy_to_user(argp, &rma, sizeof(rma)))
688 r = -EFAULT;
689 break;
691 #endif /* CONFIG_KVM_BOOK3S_64_HV */
693 default:
694 r = -ENOTTY;
697 out:
698 return r;
701 int kvm_arch_init(void *opaque)
703 return 0;
706 void kvm_arch_exit(void)