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) /* XXX Missing magic page on BookE */
77 r2
|= (1 << KVM_FEATURE_MAGIC_PAGE
);
80 /* Second return value is in r4 */
83 r
= HC_EV_UNIMPLEMENTED
;
87 kvmppc_set_gpr(vcpu
, 4, r2
);
92 int kvmppc_emulate_mmio(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
)
94 enum emulation_result er
;
97 er
= kvmppc_emulate_instruction(run
, vcpu
);
100 /* Future optimization: only reload non-volatiles if they were
101 * actually modified. */
104 case EMULATE_DO_MMIO
:
105 run
->exit_reason
= KVM_EXIT_MMIO
;
106 /* We must reload nonvolatiles because "update" load/store
107 * instructions modify register state. */
108 /* Future optimization: only reload non-volatiles if they were
109 * actually modified. */
113 /* XXX Deliver Program interrupt to guest. */
114 printk(KERN_EMERG
"%s: emulation failed (%08x)\n", __func__
,
115 kvmppc_get_last_inst(vcpu
));
125 int kvm_arch_hardware_enable(void *garbage
)
130 void kvm_arch_hardware_disable(void *garbage
)
134 int kvm_arch_hardware_setup(void)
139 void kvm_arch_hardware_unsetup(void)
143 void kvm_arch_check_processor_compat(void *rtn
)
145 *(int *)rtn
= kvmppc_core_check_processor_compat();
148 int kvm_arch_init_vm(struct kvm
*kvm
)
153 void kvm_arch_destroy_vm(struct kvm
*kvm
)
156 struct kvm_vcpu
*vcpu
;
158 kvm_for_each_vcpu(i
, vcpu
, kvm
)
159 kvm_arch_vcpu_free(vcpu
);
161 mutex_lock(&kvm
->lock
);
162 for (i
= 0; i
< atomic_read(&kvm
->online_vcpus
); i
++)
163 kvm
->vcpus
[i
] = NULL
;
165 atomic_set(&kvm
->online_vcpus
, 0);
166 mutex_unlock(&kvm
->lock
);
169 void kvm_arch_sync_events(struct kvm
*kvm
)
173 int kvm_dev_ioctl_check_extension(long ext
)
179 case KVM_CAP_PPC_BOOKE_SREGS
:
181 case KVM_CAP_PPC_SEGSTATE
:
183 case KVM_CAP_PPC_PAIRED_SINGLES
:
184 case KVM_CAP_PPC_UNSET_IRQ
:
185 case KVM_CAP_PPC_IRQ_LEVEL
:
186 case KVM_CAP_ENABLE_CAP
:
187 case KVM_CAP_PPC_OSI
:
188 case KVM_CAP_PPC_GET_PVINFO
:
191 case KVM_CAP_COALESCED_MMIO
:
192 r
= KVM_COALESCED_MMIO_PAGE_OFFSET
;
202 long kvm_arch_dev_ioctl(struct file
*filp
,
203 unsigned int ioctl
, unsigned long arg
)
208 int kvm_arch_prepare_memory_region(struct kvm
*kvm
,
209 struct kvm_memory_slot
*memslot
,
210 struct kvm_memory_slot old
,
211 struct kvm_userspace_memory_region
*mem
,
217 void kvm_arch_commit_memory_region(struct kvm
*kvm
,
218 struct kvm_userspace_memory_region
*mem
,
219 struct kvm_memory_slot old
,
226 void kvm_arch_flush_shadow(struct kvm
*kvm
)
230 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
, unsigned int id
)
232 struct kvm_vcpu
*vcpu
;
233 vcpu
= kvmppc_core_vcpu_create(kvm
, id
);
235 kvmppc_create_vcpu_debugfs(vcpu
, id
);
239 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
241 /* Make sure we're not using the vcpu anymore */
242 hrtimer_cancel(&vcpu
->arch
.dec_timer
);
243 tasklet_kill(&vcpu
->arch
.tasklet
);
245 kvmppc_remove_vcpu_debugfs(vcpu
);
246 kvmppc_core_vcpu_free(vcpu
);
249 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
251 kvm_arch_vcpu_free(vcpu
);
254 int kvm_cpu_has_pending_timer(struct kvm_vcpu
*vcpu
)
256 return kvmppc_core_pending_dec(vcpu
);
259 static void kvmppc_decrementer_func(unsigned long data
)
261 struct kvm_vcpu
*vcpu
= (struct kvm_vcpu
*)data
;
263 kvmppc_core_queue_dec(vcpu
);
265 if (waitqueue_active(&vcpu
->wq
)) {
266 wake_up_interruptible(&vcpu
->wq
);
267 vcpu
->stat
.halt_wakeup
++;
272 * low level hrtimer wake routine. Because this runs in hardirq context
273 * we schedule a tasklet to do the real work.
275 enum hrtimer_restart
kvmppc_decrementer_wakeup(struct hrtimer
*timer
)
277 struct kvm_vcpu
*vcpu
;
279 vcpu
= container_of(timer
, struct kvm_vcpu
, arch
.dec_timer
);
280 tasklet_schedule(&vcpu
->arch
.tasklet
);
282 return HRTIMER_NORESTART
;
285 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
287 hrtimer_init(&vcpu
->arch
.dec_timer
, CLOCK_REALTIME
, HRTIMER_MODE_ABS
);
288 tasklet_init(&vcpu
->arch
.tasklet
, kvmppc_decrementer_func
, (ulong
)vcpu
);
289 vcpu
->arch
.dec_timer
.function
= kvmppc_decrementer_wakeup
;
291 #ifdef CONFIG_KVM_EXIT_TIMING
292 mutex_init(&vcpu
->arch
.exit_timing_lock
);
298 void kvm_arch_vcpu_uninit(struct kvm_vcpu
*vcpu
)
300 kvmppc_mmu_destroy(vcpu
);
303 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
307 * vrsave (formerly usprg0) isn't used by Linux, but may
308 * be used by the guest.
310 * On non-booke this is associated with Altivec and
311 * is handled by code in book3s.c.
313 mtspr(SPRN_VRSAVE
, vcpu
->arch
.vrsave
);
315 kvmppc_core_vcpu_load(vcpu
, cpu
);
318 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
320 kvmppc_core_vcpu_put(vcpu
);
322 vcpu
->arch
.vrsave
= mfspr(SPRN_VRSAVE
);
326 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu
*vcpu
,
327 struct kvm_guest_debug
*dbg
)
332 static void kvmppc_complete_dcr_load(struct kvm_vcpu
*vcpu
,
335 kvmppc_set_gpr(vcpu
, vcpu
->arch
.io_gpr
, run
->dcr
.data
);
338 static void kvmppc_complete_mmio_load(struct kvm_vcpu
*vcpu
,
341 u64
uninitialized_var(gpr
);
343 if (run
->mmio
.len
> sizeof(gpr
)) {
344 printk(KERN_ERR
"bad MMIO length: %d\n", run
->mmio
.len
);
348 if (vcpu
->arch
.mmio_is_bigendian
) {
349 switch (run
->mmio
.len
) {
350 case 8: gpr
= *(u64
*)run
->mmio
.data
; break;
351 case 4: gpr
= *(u32
*)run
->mmio
.data
; break;
352 case 2: gpr
= *(u16
*)run
->mmio
.data
; break;
353 case 1: gpr
= *(u8
*)run
->mmio
.data
; break;
356 /* Convert BE data from userland back to LE. */
357 switch (run
->mmio
.len
) {
358 case 4: gpr
= ld_le32((u32
*)run
->mmio
.data
); break;
359 case 2: gpr
= ld_le16((u16
*)run
->mmio
.data
); break;
360 case 1: gpr
= *(u8
*)run
->mmio
.data
; break;
364 if (vcpu
->arch
.mmio_sign_extend
) {
365 switch (run
->mmio
.len
) {
380 kvmppc_set_gpr(vcpu
, vcpu
->arch
.io_gpr
, gpr
);
382 switch (vcpu
->arch
.io_gpr
& KVM_REG_EXT_MASK
) {
384 kvmppc_set_gpr(vcpu
, vcpu
->arch
.io_gpr
, gpr
);
387 vcpu
->arch
.fpr
[vcpu
->arch
.io_gpr
& KVM_REG_MASK
] = gpr
;
389 #ifdef CONFIG_PPC_BOOK3S
391 vcpu
->arch
.qpr
[vcpu
->arch
.io_gpr
& KVM_REG_MASK
] = gpr
;
394 vcpu
->arch
.fpr
[vcpu
->arch
.io_gpr
& KVM_REG_MASK
] = gpr
;
395 vcpu
->arch
.qpr
[vcpu
->arch
.io_gpr
& KVM_REG_MASK
] = gpr
;
403 int kvmppc_handle_load(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
,
404 unsigned int rt
, unsigned int bytes
, int is_bigendian
)
406 if (bytes
> sizeof(run
->mmio
.data
)) {
407 printk(KERN_ERR
"%s: bad MMIO length: %d\n", __func__
,
411 run
->mmio
.phys_addr
= vcpu
->arch
.paddr_accessed
;
412 run
->mmio
.len
= bytes
;
413 run
->mmio
.is_write
= 0;
415 vcpu
->arch
.io_gpr
= rt
;
416 vcpu
->arch
.mmio_is_bigendian
= is_bigendian
;
417 vcpu
->mmio_needed
= 1;
418 vcpu
->mmio_is_write
= 0;
419 vcpu
->arch
.mmio_sign_extend
= 0;
421 return EMULATE_DO_MMIO
;
424 /* Same as above, but sign extends */
425 int kvmppc_handle_loads(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
,
426 unsigned int rt
, unsigned int bytes
, int is_bigendian
)
430 r
= kvmppc_handle_load(run
, vcpu
, rt
, bytes
, is_bigendian
);
431 vcpu
->arch
.mmio_sign_extend
= 1;
436 int kvmppc_handle_store(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
,
437 u64 val
, unsigned int bytes
, int is_bigendian
)
439 void *data
= run
->mmio
.data
;
441 if (bytes
> sizeof(run
->mmio
.data
)) {
442 printk(KERN_ERR
"%s: bad MMIO length: %d\n", __func__
,
446 run
->mmio
.phys_addr
= vcpu
->arch
.paddr_accessed
;
447 run
->mmio
.len
= bytes
;
448 run
->mmio
.is_write
= 1;
449 vcpu
->mmio_needed
= 1;
450 vcpu
->mmio_is_write
= 1;
452 /* Store the value at the lowest bytes in 'data'. */
455 case 8: *(u64
*)data
= val
; break;
456 case 4: *(u32
*)data
= val
; break;
457 case 2: *(u16
*)data
= val
; break;
458 case 1: *(u8
*)data
= val
; break;
461 /* Store LE value into 'data'. */
463 case 4: st_le32(data
, val
); break;
464 case 2: st_le16(data
, val
); break;
465 case 1: *(u8
*)data
= val
; break;
469 return EMULATE_DO_MMIO
;
472 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
)
477 if (vcpu
->sigset_active
)
478 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
480 if (vcpu
->mmio_needed
) {
481 if (!vcpu
->mmio_is_write
)
482 kvmppc_complete_mmio_load(vcpu
, run
);
483 vcpu
->mmio_needed
= 0;
484 } else if (vcpu
->arch
.dcr_needed
) {
485 if (!vcpu
->arch
.dcr_is_write
)
486 kvmppc_complete_dcr_load(vcpu
, run
);
487 vcpu
->arch
.dcr_needed
= 0;
488 } else if (vcpu
->arch
.osi_needed
) {
489 u64
*gprs
= run
->osi
.gprs
;
492 for (i
= 0; i
< 32; i
++)
493 kvmppc_set_gpr(vcpu
, i
, gprs
[i
]);
494 vcpu
->arch
.osi_needed
= 0;
497 kvmppc_core_deliver_interrupts(vcpu
);
501 r
= __kvmppc_vcpu_run(run
, vcpu
);
505 if (vcpu
->sigset_active
)
506 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
511 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu
*vcpu
, struct kvm_interrupt
*irq
)
513 if (irq
->irq
== KVM_INTERRUPT_UNSET
)
514 kvmppc_core_dequeue_external(vcpu
, irq
);
516 kvmppc_core_queue_external(vcpu
, irq
);
518 if (waitqueue_active(&vcpu
->wq
)) {
519 wake_up_interruptible(&vcpu
->wq
);
520 vcpu
->stat
.halt_wakeup
++;
526 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu
*vcpu
,
527 struct kvm_enable_cap
*cap
)
535 case KVM_CAP_PPC_OSI
:
537 vcpu
->arch
.osi_enabled
= true;
547 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
548 struct kvm_mp_state
*mp_state
)
553 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
554 struct kvm_mp_state
*mp_state
)
559 long kvm_arch_vcpu_ioctl(struct file
*filp
,
560 unsigned int ioctl
, unsigned long arg
)
562 struct kvm_vcpu
*vcpu
= filp
->private_data
;
563 void __user
*argp
= (void __user
*)arg
;
567 case KVM_INTERRUPT
: {
568 struct kvm_interrupt irq
;
570 if (copy_from_user(&irq
, argp
, sizeof(irq
)))
572 r
= kvm_vcpu_ioctl_interrupt(vcpu
, &irq
);
578 struct kvm_enable_cap cap
;
580 if (copy_from_user(&cap
, argp
, sizeof(cap
)))
582 r
= kvm_vcpu_ioctl_enable_cap(vcpu
, &cap
);
593 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo
*pvinfo
)
595 u32 inst_lis
= 0x3c000000;
596 u32 inst_ori
= 0x60000000;
597 u32 inst_nop
= 0x60000000;
598 u32 inst_sc
= 0x44000002;
599 u32 inst_imm_mask
= 0xffff;
602 * The hypercall to get into KVM from within guest context is as
605 * lis r0, r0, KVM_SC_MAGIC_R0@h
606 * ori r0, KVM_SC_MAGIC_R0@l
610 pvinfo
->hcall
[0] = inst_lis
| ((KVM_SC_MAGIC_R0
>> 16) & inst_imm_mask
);
611 pvinfo
->hcall
[1] = inst_ori
| (KVM_SC_MAGIC_R0
& inst_imm_mask
);
612 pvinfo
->hcall
[2] = inst_sc
;
613 pvinfo
->hcall
[3] = inst_nop
;
618 long kvm_arch_vm_ioctl(struct file
*filp
,
619 unsigned int ioctl
, unsigned long arg
)
621 void __user
*argp
= (void __user
*)arg
;
625 case KVM_PPC_GET_PVINFO
: {
626 struct kvm_ppc_pvinfo pvinfo
;
627 memset(&pvinfo
, 0, sizeof(pvinfo
));
628 r
= kvm_vm_ioctl_get_pvinfo(&pvinfo
);
629 if (copy_to_user(argp
, &pvinfo
, sizeof(pvinfo
))) {
644 int kvm_arch_init(void *opaque
)
649 void kvm_arch_exit(void)