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 return !(v
->arch
.shared
->msr
& MSR_WE
) ||
42 !!(v
->arch
.pending_exceptions
);
45 int kvmppc_kvm_pv(struct kvm_vcpu
*vcpu
)
47 int nr
= kvmppc_get_gpr(vcpu
, 11);
49 unsigned long __maybe_unused param1
= kvmppc_get_gpr(vcpu
, 3);
50 unsigned long __maybe_unused param2
= kvmppc_get_gpr(vcpu
, 4);
51 unsigned long __maybe_unused param3
= kvmppc_get_gpr(vcpu
, 5);
52 unsigned long __maybe_unused param4
= kvmppc_get_gpr(vcpu
, 6);
55 if (!(vcpu
->arch
.shared
->msr
& MSR_SF
)) {
64 case HC_VENDOR_KVM
| KVM_HC_PPC_MAP_MAGIC_PAGE
:
66 vcpu
->arch
.magic_page_pa
= param1
;
67 vcpu
->arch
.magic_page_ea
= param2
;
69 r2
= KVM_MAGIC_FEAT_SR
;
74 case HC_VENDOR_KVM
| KVM_HC_FEATURES
:
76 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500)
77 /* XXX Missing magic page on 44x */
78 r2
|= (1 << KVM_FEATURE_MAGIC_PAGE
);
81 /* Second return value is in r4 */
84 r
= HC_EV_UNIMPLEMENTED
;
88 kvmppc_set_gpr(vcpu
, 4, r2
);
93 int kvmppc_emulate_mmio(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
)
95 enum emulation_result er
;
98 er
= kvmppc_emulate_instruction(run
, vcpu
);
101 /* Future optimization: only reload non-volatiles if they were
102 * actually modified. */
105 case EMULATE_DO_MMIO
:
106 run
->exit_reason
= KVM_EXIT_MMIO
;
107 /* We must reload nonvolatiles because "update" load/store
108 * instructions modify register state. */
109 /* Future optimization: only reload non-volatiles if they were
110 * actually modified. */
114 /* XXX Deliver Program interrupt to guest. */
115 printk(KERN_EMERG
"%s: emulation failed (%08x)\n", __func__
,
116 kvmppc_get_last_inst(vcpu
));
126 int kvm_arch_hardware_enable(void *garbage
)
131 void kvm_arch_hardware_disable(void *garbage
)
135 int kvm_arch_hardware_setup(void)
140 void kvm_arch_hardware_unsetup(void)
144 void kvm_arch_check_processor_compat(void *rtn
)
146 *(int *)rtn
= kvmppc_core_check_processor_compat();
149 int kvm_arch_init_vm(struct kvm
*kvm
)
154 void kvm_arch_destroy_vm(struct kvm
*kvm
)
157 struct kvm_vcpu
*vcpu
;
159 kvm_for_each_vcpu(i
, vcpu
, kvm
)
160 kvm_arch_vcpu_free(vcpu
);
162 mutex_lock(&kvm
->lock
);
163 for (i
= 0; i
< atomic_read(&kvm
->online_vcpus
); i
++)
164 kvm
->vcpus
[i
] = NULL
;
166 atomic_set(&kvm
->online_vcpus
, 0);
167 mutex_unlock(&kvm
->lock
);
170 void kvm_arch_sync_events(struct kvm
*kvm
)
174 int kvm_dev_ioctl_check_extension(long ext
)
180 case KVM_CAP_PPC_BOOKE_SREGS
:
182 case KVM_CAP_PPC_SEGSTATE
:
184 case KVM_CAP_PPC_PAIRED_SINGLES
:
185 case KVM_CAP_PPC_UNSET_IRQ
:
186 case KVM_CAP_PPC_IRQ_LEVEL
:
187 case KVM_CAP_ENABLE_CAP
:
188 case KVM_CAP_PPC_OSI
:
189 case KVM_CAP_PPC_GET_PVINFO
:
192 case KVM_CAP_COALESCED_MMIO
:
193 r
= KVM_COALESCED_MMIO_PAGE_OFFSET
;
203 long kvm_arch_dev_ioctl(struct file
*filp
,
204 unsigned int ioctl
, unsigned long arg
)
209 int kvm_arch_prepare_memory_region(struct kvm
*kvm
,
210 struct kvm_memory_slot
*memslot
,
211 struct kvm_memory_slot old
,
212 struct kvm_userspace_memory_region
*mem
,
218 void kvm_arch_commit_memory_region(struct kvm
*kvm
,
219 struct kvm_userspace_memory_region
*mem
,
220 struct kvm_memory_slot old
,
227 void kvm_arch_flush_shadow(struct kvm
*kvm
)
231 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
, unsigned int id
)
233 struct kvm_vcpu
*vcpu
;
234 vcpu
= kvmppc_core_vcpu_create(kvm
, id
);
236 kvmppc_create_vcpu_debugfs(vcpu
, id
);
240 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
242 /* Make sure we're not using the vcpu anymore */
243 hrtimer_cancel(&vcpu
->arch
.dec_timer
);
244 tasklet_kill(&vcpu
->arch
.tasklet
);
246 kvmppc_remove_vcpu_debugfs(vcpu
);
247 kvmppc_core_vcpu_free(vcpu
);
250 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
252 kvm_arch_vcpu_free(vcpu
);
255 int kvm_cpu_has_pending_timer(struct kvm_vcpu
*vcpu
)
257 return kvmppc_core_pending_dec(vcpu
);
260 static void kvmppc_decrementer_func(unsigned long data
)
262 struct kvm_vcpu
*vcpu
= (struct kvm_vcpu
*)data
;
264 kvmppc_core_queue_dec(vcpu
);
266 if (waitqueue_active(&vcpu
->wq
)) {
267 wake_up_interruptible(&vcpu
->wq
);
268 vcpu
->stat
.halt_wakeup
++;
273 * low level hrtimer wake routine. Because this runs in hardirq context
274 * we schedule a tasklet to do the real work.
276 enum hrtimer_restart
kvmppc_decrementer_wakeup(struct hrtimer
*timer
)
278 struct kvm_vcpu
*vcpu
;
280 vcpu
= container_of(timer
, struct kvm_vcpu
, arch
.dec_timer
);
281 tasklet_schedule(&vcpu
->arch
.tasklet
);
283 return HRTIMER_NORESTART
;
286 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
288 hrtimer_init(&vcpu
->arch
.dec_timer
, CLOCK_REALTIME
, HRTIMER_MODE_ABS
);
289 tasklet_init(&vcpu
->arch
.tasklet
, kvmppc_decrementer_func
, (ulong
)vcpu
);
290 vcpu
->arch
.dec_timer
.function
= kvmppc_decrementer_wakeup
;
292 #ifdef CONFIG_KVM_EXIT_TIMING
293 mutex_init(&vcpu
->arch
.exit_timing_lock
);
299 void kvm_arch_vcpu_uninit(struct kvm_vcpu
*vcpu
)
301 kvmppc_mmu_destroy(vcpu
);
304 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
308 * vrsave (formerly usprg0) isn't used by Linux, but may
309 * be used by the guest.
311 * On non-booke this is associated with Altivec and
312 * is handled by code in book3s.c.
314 mtspr(SPRN_VRSAVE
, vcpu
->arch
.vrsave
);
316 kvmppc_core_vcpu_load(vcpu
, cpu
);
319 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
321 kvmppc_core_vcpu_put(vcpu
);
323 vcpu
->arch
.vrsave
= mfspr(SPRN_VRSAVE
);
327 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu
*vcpu
,
328 struct kvm_guest_debug
*dbg
)
333 static void kvmppc_complete_dcr_load(struct kvm_vcpu
*vcpu
,
336 kvmppc_set_gpr(vcpu
, vcpu
->arch
.io_gpr
, run
->dcr
.data
);
339 static void kvmppc_complete_mmio_load(struct kvm_vcpu
*vcpu
,
342 u64
uninitialized_var(gpr
);
344 if (run
->mmio
.len
> sizeof(gpr
)) {
345 printk(KERN_ERR
"bad MMIO length: %d\n", run
->mmio
.len
);
349 if (vcpu
->arch
.mmio_is_bigendian
) {
350 switch (run
->mmio
.len
) {
351 case 8: gpr
= *(u64
*)run
->mmio
.data
; break;
352 case 4: gpr
= *(u32
*)run
->mmio
.data
; break;
353 case 2: gpr
= *(u16
*)run
->mmio
.data
; break;
354 case 1: gpr
= *(u8
*)run
->mmio
.data
; break;
357 /* Convert BE data from userland back to LE. */
358 switch (run
->mmio
.len
) {
359 case 4: gpr
= ld_le32((u32
*)run
->mmio
.data
); break;
360 case 2: gpr
= ld_le16((u16
*)run
->mmio
.data
); break;
361 case 1: gpr
= *(u8
*)run
->mmio
.data
; break;
365 if (vcpu
->arch
.mmio_sign_extend
) {
366 switch (run
->mmio
.len
) {
381 kvmppc_set_gpr(vcpu
, vcpu
->arch
.io_gpr
, gpr
);
383 switch (vcpu
->arch
.io_gpr
& KVM_REG_EXT_MASK
) {
385 kvmppc_set_gpr(vcpu
, vcpu
->arch
.io_gpr
, gpr
);
388 vcpu
->arch
.fpr
[vcpu
->arch
.io_gpr
& KVM_REG_MASK
] = gpr
;
390 #ifdef CONFIG_PPC_BOOK3S
392 vcpu
->arch
.qpr
[vcpu
->arch
.io_gpr
& KVM_REG_MASK
] = gpr
;
395 vcpu
->arch
.fpr
[vcpu
->arch
.io_gpr
& KVM_REG_MASK
] = gpr
;
396 vcpu
->arch
.qpr
[vcpu
->arch
.io_gpr
& KVM_REG_MASK
] = gpr
;
404 int kvmppc_handle_load(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
,
405 unsigned int rt
, unsigned int bytes
, int is_bigendian
)
407 if (bytes
> sizeof(run
->mmio
.data
)) {
408 printk(KERN_ERR
"%s: bad MMIO length: %d\n", __func__
,
412 run
->mmio
.phys_addr
= vcpu
->arch
.paddr_accessed
;
413 run
->mmio
.len
= bytes
;
414 run
->mmio
.is_write
= 0;
416 vcpu
->arch
.io_gpr
= rt
;
417 vcpu
->arch
.mmio_is_bigendian
= is_bigendian
;
418 vcpu
->mmio_needed
= 1;
419 vcpu
->mmio_is_write
= 0;
420 vcpu
->arch
.mmio_sign_extend
= 0;
422 return EMULATE_DO_MMIO
;
425 /* Same as above, but sign extends */
426 int kvmppc_handle_loads(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
,
427 unsigned int rt
, unsigned int bytes
, int is_bigendian
)
431 r
= kvmppc_handle_load(run
, vcpu
, rt
, bytes
, is_bigendian
);
432 vcpu
->arch
.mmio_sign_extend
= 1;
437 int kvmppc_handle_store(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
,
438 u64 val
, unsigned int bytes
, int is_bigendian
)
440 void *data
= run
->mmio
.data
;
442 if (bytes
> sizeof(run
->mmio
.data
)) {
443 printk(KERN_ERR
"%s: bad MMIO length: %d\n", __func__
,
447 run
->mmio
.phys_addr
= vcpu
->arch
.paddr_accessed
;
448 run
->mmio
.len
= bytes
;
449 run
->mmio
.is_write
= 1;
450 vcpu
->mmio_needed
= 1;
451 vcpu
->mmio_is_write
= 1;
453 /* Store the value at the lowest bytes in 'data'. */
456 case 8: *(u64
*)data
= val
; break;
457 case 4: *(u32
*)data
= val
; break;
458 case 2: *(u16
*)data
= val
; break;
459 case 1: *(u8
*)data
= val
; break;
462 /* Store LE value into 'data'. */
464 case 4: st_le32(data
, val
); break;
465 case 2: st_le16(data
, val
); break;
466 case 1: *(u8
*)data
= val
; break;
470 return EMULATE_DO_MMIO
;
473 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
)
478 if (vcpu
->sigset_active
)
479 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
481 if (vcpu
->mmio_needed
) {
482 if (!vcpu
->mmio_is_write
)
483 kvmppc_complete_mmio_load(vcpu
, run
);
484 vcpu
->mmio_needed
= 0;
485 } else if (vcpu
->arch
.dcr_needed
) {
486 if (!vcpu
->arch
.dcr_is_write
)
487 kvmppc_complete_dcr_load(vcpu
, run
);
488 vcpu
->arch
.dcr_needed
= 0;
489 } else if (vcpu
->arch
.osi_needed
) {
490 u64
*gprs
= run
->osi
.gprs
;
493 for (i
= 0; i
< 32; i
++)
494 kvmppc_set_gpr(vcpu
, i
, gprs
[i
]);
495 vcpu
->arch
.osi_needed
= 0;
498 kvmppc_core_deliver_interrupts(vcpu
);
502 r
= __kvmppc_vcpu_run(run
, vcpu
);
506 if (vcpu
->sigset_active
)
507 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
512 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu
*vcpu
, struct kvm_interrupt
*irq
)
514 if (irq
->irq
== KVM_INTERRUPT_UNSET
)
515 kvmppc_core_dequeue_external(vcpu
, irq
);
517 kvmppc_core_queue_external(vcpu
, irq
);
519 if (waitqueue_active(&vcpu
->wq
)) {
520 wake_up_interruptible(&vcpu
->wq
);
521 vcpu
->stat
.halt_wakeup
++;
527 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu
*vcpu
,
528 struct kvm_enable_cap
*cap
)
536 case KVM_CAP_PPC_OSI
:
538 vcpu
->arch
.osi_enabled
= true;
548 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
549 struct kvm_mp_state
*mp_state
)
554 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
555 struct kvm_mp_state
*mp_state
)
560 long kvm_arch_vcpu_ioctl(struct file
*filp
,
561 unsigned int ioctl
, unsigned long arg
)
563 struct kvm_vcpu
*vcpu
= filp
->private_data
;
564 void __user
*argp
= (void __user
*)arg
;
568 case KVM_INTERRUPT
: {
569 struct kvm_interrupt irq
;
571 if (copy_from_user(&irq
, argp
, sizeof(irq
)))
573 r
= kvm_vcpu_ioctl_interrupt(vcpu
, &irq
);
579 struct kvm_enable_cap cap
;
581 if (copy_from_user(&cap
, argp
, sizeof(cap
)))
583 r
= kvm_vcpu_ioctl_enable_cap(vcpu
, &cap
);
594 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo
*pvinfo
)
596 u32 inst_lis
= 0x3c000000;
597 u32 inst_ori
= 0x60000000;
598 u32 inst_nop
= 0x60000000;
599 u32 inst_sc
= 0x44000002;
600 u32 inst_imm_mask
= 0xffff;
603 * The hypercall to get into KVM from within guest context is as
606 * lis r0, r0, KVM_SC_MAGIC_R0@h
607 * ori r0, KVM_SC_MAGIC_R0@l
611 pvinfo
->hcall
[0] = inst_lis
| ((KVM_SC_MAGIC_R0
>> 16) & inst_imm_mask
);
612 pvinfo
->hcall
[1] = inst_ori
| (KVM_SC_MAGIC_R0
& inst_imm_mask
);
613 pvinfo
->hcall
[2] = inst_sc
;
614 pvinfo
->hcall
[3] = inst_nop
;
619 long kvm_arch_vm_ioctl(struct file
*filp
,
620 unsigned int ioctl
, unsigned long arg
)
622 void __user
*argp
= (void __user
*)arg
;
626 case KVM_PPC_GET_PVINFO
: {
627 struct kvm_ppc_pvinfo pvinfo
;
628 memset(&pvinfo
, 0, sizeof(pvinfo
));
629 r
= kvm_vm_ioctl_get_pvinfo(&pvinfo
);
630 if (copy_to_user(argp
, &pvinfo
, sizeof(pvinfo
))) {
645 int kvm_arch_init(void *opaque
)
650 void kvm_arch_exit(void)