2 * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
5 * Alexander Graf <agraf@suse.de>
6 * Kevin Wolf <mail@kevin-wolf.de>
7 * Paul Mackerras <paulus@samba.org>
10 * Functions relating to running KVM on Book 3S processors where
11 * we don't have access to hypervisor mode, and we run the guest
12 * in problem state (user mode).
14 * This file is derived from arch/powerpc/kvm/44x.c,
15 * by Hollis Blanchard <hollisb@us.ibm.com>.
17 * This program is free software; you can redistribute it and/or modify
18 * it under the terms of the GNU General Public License, version 2, as
19 * published by the Free Software Foundation.
22 #include <linux/kvm_host.h>
23 #include <linux/export.h>
24 #include <linux/err.h>
25 #include <linux/slab.h>
28 #include <asm/cputable.h>
29 #include <asm/cacheflush.h>
30 #include <asm/tlbflush.h>
31 #include <asm/uaccess.h>
33 #include <asm/kvm_ppc.h>
34 #include <asm/kvm_book3s.h>
35 #include <asm/mmu_context.h>
36 #include <linux/gfp.h>
37 #include <linux/sched.h>
38 #include <linux/vmalloc.h>
39 #include <linux/highmem.h>
43 /* #define EXIT_DEBUG */
44 /* #define DEBUG_EXT */
46 static int kvmppc_handle_ext(struct kvm_vcpu
*vcpu
, unsigned int exit_nr
,
49 /* Some compatibility defines */
50 #ifdef CONFIG_PPC_BOOK3S_32
51 #define MSR_USER32 MSR_USER
52 #define MSR_USER64 MSR_USER
53 #define HW_PAGE_SIZE PAGE_SIZE
56 void kvmppc_core_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
58 #ifdef CONFIG_PPC_BOOK3S_64
59 memcpy(to_svcpu(vcpu
)->slb
, to_book3s(vcpu
)->slb_shadow
, sizeof(to_svcpu(vcpu
)->slb
));
60 memcpy(&get_paca()->shadow_vcpu
, to_book3s(vcpu
)->shadow_vcpu
,
61 sizeof(get_paca()->shadow_vcpu
));
62 to_svcpu(vcpu
)->slb_max
= to_book3s(vcpu
)->slb_shadow_max
;
65 #ifdef CONFIG_PPC_BOOK3S_32
66 current
->thread
.kvm_shadow_vcpu
= to_book3s(vcpu
)->shadow_vcpu
;
70 void kvmppc_core_vcpu_put(struct kvm_vcpu
*vcpu
)
72 #ifdef CONFIG_PPC_BOOK3S_64
73 memcpy(to_book3s(vcpu
)->slb_shadow
, to_svcpu(vcpu
)->slb
, sizeof(to_svcpu(vcpu
)->slb
));
74 memcpy(to_book3s(vcpu
)->shadow_vcpu
, &get_paca()->shadow_vcpu
,
75 sizeof(get_paca()->shadow_vcpu
));
76 to_book3s(vcpu
)->slb_shadow_max
= to_svcpu(vcpu
)->slb_max
;
79 kvmppc_giveup_ext(vcpu
, MSR_FP
);
80 kvmppc_giveup_ext(vcpu
, MSR_VEC
);
81 kvmppc_giveup_ext(vcpu
, MSR_VSX
);
84 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu
*vcpu
)
86 ulong smsr
= vcpu
->arch
.shared
->msr
;
88 /* Guest MSR values */
89 smsr
&= MSR_FE0
| MSR_FE1
| MSR_SF
| MSR_SE
| MSR_BE
| MSR_DE
;
90 /* Process MSR values */
91 smsr
|= MSR_ME
| MSR_RI
| MSR_IR
| MSR_DR
| MSR_PR
| MSR_EE
;
92 /* External providers the guest reserved */
93 smsr
|= (vcpu
->arch
.shared
->msr
& vcpu
->arch
.guest_owned_ext
);
94 /* 64-bit Process MSR values */
95 #ifdef CONFIG_PPC_BOOK3S_64
96 smsr
|= MSR_ISF
| MSR_HV
;
98 vcpu
->arch
.shadow_msr
= smsr
;
101 void kvmppc_set_msr(struct kvm_vcpu
*vcpu
, u64 msr
)
103 ulong old_msr
= vcpu
->arch
.shared
->msr
;
106 printk(KERN_INFO
"KVM: Set MSR to 0x%llx\n", msr
);
109 msr
&= to_book3s(vcpu
)->msr_mask
;
110 vcpu
->arch
.shared
->msr
= msr
;
111 kvmppc_recalc_shadow_msr(vcpu
);
114 if (!vcpu
->arch
.pending_exceptions
) {
115 kvm_vcpu_block(vcpu
);
116 vcpu
->stat
.halt_wakeup
++;
118 /* Unset POW bit after we woke up */
120 vcpu
->arch
.shared
->msr
= msr
;
124 if ((vcpu
->arch
.shared
->msr
& (MSR_PR
|MSR_IR
|MSR_DR
)) !=
125 (old_msr
& (MSR_PR
|MSR_IR
|MSR_DR
))) {
126 kvmppc_mmu_flush_segments(vcpu
);
127 kvmppc_mmu_map_segment(vcpu
, kvmppc_get_pc(vcpu
));
129 /* Preload magic page segment when in kernel mode */
130 if (!(msr
& MSR_PR
) && vcpu
->arch
.magic_page_pa
) {
131 struct kvm_vcpu_arch
*a
= &vcpu
->arch
;
134 kvmppc_mmu_map_segment(vcpu
, a
->magic_page_ea
);
136 kvmppc_mmu_map_segment(vcpu
, a
->magic_page_pa
);
140 /* Preload FPU if it's enabled */
141 if (vcpu
->arch
.shared
->msr
& MSR_FP
)
142 kvmppc_handle_ext(vcpu
, BOOK3S_INTERRUPT_FP_UNAVAIL
, MSR_FP
);
145 void kvmppc_set_pvr(struct kvm_vcpu
*vcpu
, u32 pvr
)
149 vcpu
->arch
.hflags
&= ~BOOK3S_HFLAG_SLB
;
150 vcpu
->arch
.pvr
= pvr
;
151 #ifdef CONFIG_PPC_BOOK3S_64
152 if ((pvr
>= 0x330000) && (pvr
< 0x70330000)) {
153 kvmppc_mmu_book3s_64_init(vcpu
);
154 to_book3s(vcpu
)->hior
= 0xfff00000;
155 to_book3s(vcpu
)->msr_mask
= 0xffffffffffffffffULL
;
156 vcpu
->arch
.cpu_type
= KVM_CPU_3S_64
;
160 kvmppc_mmu_book3s_32_init(vcpu
);
161 to_book3s(vcpu
)->hior
= 0;
162 to_book3s(vcpu
)->msr_mask
= 0xffffffffULL
;
163 vcpu
->arch
.cpu_type
= KVM_CPU_3S_32
;
166 kvmppc_sanity_check(vcpu
);
168 /* If we are in hypervisor level on 970, we can tell the CPU to
169 * treat DCBZ as 32 bytes store */
170 vcpu
->arch
.hflags
&= ~BOOK3S_HFLAG_DCBZ32
;
171 if (vcpu
->arch
.mmu
.is_dcbz32(vcpu
) && (mfmsr() & MSR_HV
) &&
172 !strcmp(cur_cpu_spec
->platform
, "ppc970"))
173 vcpu
->arch
.hflags
|= BOOK3S_HFLAG_DCBZ32
;
175 /* Cell performs badly if MSR_FEx are set. So let's hope nobody
176 really needs them in a VM on Cell and force disable them. */
177 if (!strcmp(cur_cpu_spec
->platform
, "ppc-cell-be"))
178 to_book3s(vcpu
)->msr_mask
&= ~(MSR_FE0
| MSR_FE1
);
180 #ifdef CONFIG_PPC_BOOK3S_32
181 /* 32 bit Book3S always has 32 byte dcbz */
182 vcpu
->arch
.hflags
|= BOOK3S_HFLAG_DCBZ32
;
185 /* On some CPUs we can execute paired single operations natively */
186 asm ( "mfpvr %0" : "=r"(host_pvr
));
188 case 0x00080200: /* lonestar 2.0 */
189 case 0x00088202: /* lonestar 2.2 */
190 case 0x70000100: /* gekko 1.0 */
191 case 0x00080100: /* gekko 2.0 */
192 case 0x00083203: /* gekko 2.3a */
193 case 0x00083213: /* gekko 2.3b */
194 case 0x00083204: /* gekko 2.4 */
195 case 0x00083214: /* gekko 2.4e (8SE) - retail HW2 */
196 case 0x00087200: /* broadway */
197 vcpu
->arch
.hflags
|= BOOK3S_HFLAG_NATIVE_PS
;
198 /* Enable HID2.PSE - in case we need it later */
199 mtspr(SPRN_HID2_GEKKO
, mfspr(SPRN_HID2_GEKKO
) | (1 << 29));
203 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
204 * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
205 * emulate 32 bytes dcbz length.
207 * The Book3s_64 inventors also realized this case and implemented a special bit
208 * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
210 * My approach here is to patch the dcbz instruction on executing pages.
212 static void kvmppc_patch_dcbz(struct kvm_vcpu
*vcpu
, struct kvmppc_pte
*pte
)
219 hpage
= gfn_to_page(vcpu
->kvm
, pte
->raddr
>> PAGE_SHIFT
);
220 if (is_error_page(hpage
)) {
221 kvm_release_page_clean(hpage
);
225 hpage_offset
= pte
->raddr
& ~PAGE_MASK
;
226 hpage_offset
&= ~0xFFFULL
;
230 page
= kmap_atomic(hpage
);
232 /* patch dcbz into reserved instruction, so we trap */
233 for (i
=hpage_offset
; i
< hpage_offset
+ (HW_PAGE_SIZE
/ 4); i
++)
234 if ((page
[i
] & 0xff0007ff) == INS_DCBZ
)
235 page
[i
] &= 0xfffffff7;
241 static int kvmppc_visible_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
243 ulong mp_pa
= vcpu
->arch
.magic_page_pa
;
245 if (unlikely(mp_pa
) &&
246 unlikely((mp_pa
& KVM_PAM
) >> PAGE_SHIFT
== gfn
)) {
250 return kvm_is_visible_gfn(vcpu
->kvm
, gfn
);
253 int kvmppc_handle_pagefault(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
,
254 ulong eaddr
, int vec
)
256 bool data
= (vec
== BOOK3S_INTERRUPT_DATA_STORAGE
);
257 int r
= RESUME_GUEST
;
260 struct kvmppc_pte pte
;
261 bool is_mmio
= false;
262 bool dr
= (vcpu
->arch
.shared
->msr
& MSR_DR
) ? true : false;
263 bool ir
= (vcpu
->arch
.shared
->msr
& MSR_IR
) ? true : false;
266 relocated
= data
? dr
: ir
;
268 /* Resolve real address if translation turned on */
270 page_found
= vcpu
->arch
.mmu
.xlate(vcpu
, eaddr
, &pte
, data
);
272 pte
.may_execute
= true;
274 pte
.may_write
= true;
275 pte
.raddr
= eaddr
& KVM_PAM
;
277 pte
.vpage
= eaddr
>> 12;
280 switch (vcpu
->arch
.shared
->msr
& (MSR_DR
|MSR_IR
)) {
282 pte
.vpage
|= ((u64
)VSID_REAL
<< (SID_SHIFT
- 12));
286 vcpu
->arch
.mmu
.esid_to_vsid(vcpu
, eaddr
>> SID_SHIFT
, &vsid
);
288 if ((vcpu
->arch
.shared
->msr
& (MSR_DR
|MSR_IR
)) == MSR_DR
)
289 pte
.vpage
|= ((u64
)VSID_REAL_DR
<< (SID_SHIFT
- 12));
291 pte
.vpage
|= ((u64
)VSID_REAL_IR
<< (SID_SHIFT
- 12));
295 page_found
= -EINVAL
;
299 if (vcpu
->arch
.mmu
.is_dcbz32(vcpu
) &&
300 (!(vcpu
->arch
.hflags
& BOOK3S_HFLAG_DCBZ32
))) {
302 * If we do the dcbz hack, we have to NX on every execution,
303 * so we can patch the executing code. This renders our guest
306 pte
.may_execute
= !data
;
309 if (page_found
== -ENOENT
) {
310 /* Page not found in guest PTE entries */
311 vcpu
->arch
.shared
->dar
= kvmppc_get_fault_dar(vcpu
);
312 vcpu
->arch
.shared
->dsisr
= to_svcpu(vcpu
)->fault_dsisr
;
313 vcpu
->arch
.shared
->msr
|=
314 (to_svcpu(vcpu
)->shadow_srr1
& 0x00000000f8000000ULL
);
315 kvmppc_book3s_queue_irqprio(vcpu
, vec
);
316 } else if (page_found
== -EPERM
) {
317 /* Storage protection */
318 vcpu
->arch
.shared
->dar
= kvmppc_get_fault_dar(vcpu
);
319 vcpu
->arch
.shared
->dsisr
=
320 to_svcpu(vcpu
)->fault_dsisr
& ~DSISR_NOHPTE
;
321 vcpu
->arch
.shared
->dsisr
|= DSISR_PROTFAULT
;
322 vcpu
->arch
.shared
->msr
|=
323 (to_svcpu(vcpu
)->shadow_srr1
& 0x00000000f8000000ULL
);
324 kvmppc_book3s_queue_irqprio(vcpu
, vec
);
325 } else if (page_found
== -EINVAL
) {
326 /* Page not found in guest SLB */
327 vcpu
->arch
.shared
->dar
= kvmppc_get_fault_dar(vcpu
);
328 kvmppc_book3s_queue_irqprio(vcpu
, vec
+ 0x80);
329 } else if (!is_mmio
&&
330 kvmppc_visible_gfn(vcpu
, pte
.raddr
>> PAGE_SHIFT
)) {
331 /* The guest's PTE is not mapped yet. Map on the host */
332 kvmppc_mmu_map_page(vcpu
, &pte
);
334 vcpu
->stat
.sp_storage
++;
335 else if (vcpu
->arch
.mmu
.is_dcbz32(vcpu
) &&
336 (!(vcpu
->arch
.hflags
& BOOK3S_HFLAG_DCBZ32
)))
337 kvmppc_patch_dcbz(vcpu
, &pte
);
340 vcpu
->stat
.mmio_exits
++;
341 vcpu
->arch
.paddr_accessed
= pte
.raddr
;
342 r
= kvmppc_emulate_mmio(run
, vcpu
);
343 if ( r
== RESUME_HOST_NV
)
350 static inline int get_fpr_index(int i
)
358 /* Give up external provider (FPU, Altivec, VSX) */
359 void kvmppc_giveup_ext(struct kvm_vcpu
*vcpu
, ulong msr
)
361 struct thread_struct
*t
= ¤t
->thread
;
362 u64
*vcpu_fpr
= vcpu
->arch
.fpr
;
364 u64
*vcpu_vsx
= vcpu
->arch
.vsr
;
366 u64
*thread_fpr
= (u64
*)t
->fpr
;
369 if (!(vcpu
->arch
.guest_owned_ext
& msr
))
373 printk(KERN_INFO
"Giving up ext 0x%lx\n", msr
);
379 for (i
= 0; i
< ARRAY_SIZE(vcpu
->arch
.fpr
); i
++)
380 vcpu_fpr
[i
] = thread_fpr
[get_fpr_index(i
)];
382 vcpu
->arch
.fpscr
= t
->fpscr
.val
;
385 #ifdef CONFIG_ALTIVEC
386 giveup_altivec(current
);
387 memcpy(vcpu
->arch
.vr
, t
->vr
, sizeof(vcpu
->arch
.vr
));
388 vcpu
->arch
.vscr
= t
->vscr
;
393 __giveup_vsx(current
);
394 for (i
= 0; i
< ARRAY_SIZE(vcpu
->arch
.vsr
); i
++)
395 vcpu_vsx
[i
] = thread_fpr
[get_fpr_index(i
) + 1];
402 vcpu
->arch
.guest_owned_ext
&= ~msr
;
403 current
->thread
.regs
->msr
&= ~msr
;
404 kvmppc_recalc_shadow_msr(vcpu
);
407 static int kvmppc_read_inst(struct kvm_vcpu
*vcpu
)
409 ulong srr0
= kvmppc_get_pc(vcpu
);
410 u32 last_inst
= kvmppc_get_last_inst(vcpu
);
413 ret
= kvmppc_ld(vcpu
, &srr0
, sizeof(u32
), &last_inst
, false);
414 if (ret
== -ENOENT
) {
415 ulong msr
= vcpu
->arch
.shared
->msr
;
417 msr
= kvmppc_set_field(msr
, 33, 33, 1);
418 msr
= kvmppc_set_field(msr
, 34, 36, 0);
419 vcpu
->arch
.shared
->msr
= kvmppc_set_field(msr
, 42, 47, 0);
420 kvmppc_book3s_queue_irqprio(vcpu
, BOOK3S_INTERRUPT_INST_STORAGE
);
421 return EMULATE_AGAIN
;
427 static int kvmppc_check_ext(struct kvm_vcpu
*vcpu
, unsigned int exit_nr
)
430 /* Need to do paired single emulation? */
431 if (!(vcpu
->arch
.hflags
& BOOK3S_HFLAG_PAIRED_SINGLE
))
434 /* Read out the instruction */
435 if (kvmppc_read_inst(vcpu
) == EMULATE_DONE
)
436 /* Need to emulate */
439 return EMULATE_AGAIN
;
442 /* Handle external providers (FPU, Altivec, VSX) */
443 static int kvmppc_handle_ext(struct kvm_vcpu
*vcpu
, unsigned int exit_nr
,
446 struct thread_struct
*t
= ¤t
->thread
;
447 u64
*vcpu_fpr
= vcpu
->arch
.fpr
;
449 u64
*vcpu_vsx
= vcpu
->arch
.vsr
;
451 u64
*thread_fpr
= (u64
*)t
->fpr
;
454 /* When we have paired singles, we emulate in software */
455 if (vcpu
->arch
.hflags
& BOOK3S_HFLAG_PAIRED_SINGLE
)
458 if (!(vcpu
->arch
.shared
->msr
& msr
)) {
459 kvmppc_book3s_queue_irqprio(vcpu
, exit_nr
);
463 /* We already own the ext */
464 if (vcpu
->arch
.guest_owned_ext
& msr
) {
469 printk(KERN_INFO
"Loading up ext 0x%lx\n", msr
);
472 current
->thread
.regs
->msr
|= msr
;
476 for (i
= 0; i
< ARRAY_SIZE(vcpu
->arch
.fpr
); i
++)
477 thread_fpr
[get_fpr_index(i
)] = vcpu_fpr
[i
];
479 t
->fpscr
.val
= vcpu
->arch
.fpscr
;
481 kvmppc_load_up_fpu();
484 #ifdef CONFIG_ALTIVEC
485 memcpy(t
->vr
, vcpu
->arch
.vr
, sizeof(vcpu
->arch
.vr
));
486 t
->vscr
= vcpu
->arch
.vscr
;
488 kvmppc_load_up_altivec();
493 for (i
= 0; i
< ARRAY_SIZE(vcpu
->arch
.vsr
); i
++)
494 thread_fpr
[get_fpr_index(i
) + 1] = vcpu_vsx
[i
];
495 kvmppc_load_up_vsx();
502 vcpu
->arch
.guest_owned_ext
|= msr
;
504 kvmppc_recalc_shadow_msr(vcpu
);
509 int kvmppc_handle_exit(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
,
510 unsigned int exit_nr
)
514 vcpu
->stat
.sum_exits
++;
516 run
->exit_reason
= KVM_EXIT_UNKNOWN
;
517 run
->ready_for_interrupt_injection
= 1;
519 trace_kvm_book3s_exit(exit_nr
, vcpu
);
522 case BOOK3S_INTERRUPT_INST_STORAGE
:
523 vcpu
->stat
.pf_instruc
++;
525 #ifdef CONFIG_PPC_BOOK3S_32
526 /* We set segments as unused segments when invalidating them. So
527 * treat the respective fault as segment fault. */
528 if (to_svcpu(vcpu
)->sr
[kvmppc_get_pc(vcpu
) >> SID_SHIFT
]
530 kvmppc_mmu_map_segment(vcpu
, kvmppc_get_pc(vcpu
));
536 /* only care about PTEG not found errors, but leave NX alone */
537 if (to_svcpu(vcpu
)->shadow_srr1
& 0x40000000) {
538 r
= kvmppc_handle_pagefault(run
, vcpu
, kvmppc_get_pc(vcpu
), exit_nr
);
539 vcpu
->stat
.sp_instruc
++;
540 } else if (vcpu
->arch
.mmu
.is_dcbz32(vcpu
) &&
541 (!(vcpu
->arch
.hflags
& BOOK3S_HFLAG_DCBZ32
))) {
543 * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
544 * so we can't use the NX bit inside the guest. Let's cross our fingers,
545 * that no guest that needs the dcbz hack does NX.
547 kvmppc_mmu_pte_flush(vcpu
, kvmppc_get_pc(vcpu
), ~0xFFFUL
);
550 vcpu
->arch
.shared
->msr
|=
551 to_svcpu(vcpu
)->shadow_srr1
& 0x58000000;
552 kvmppc_book3s_queue_irqprio(vcpu
, exit_nr
);
556 case BOOK3S_INTERRUPT_DATA_STORAGE
:
558 ulong dar
= kvmppc_get_fault_dar(vcpu
);
559 vcpu
->stat
.pf_storage
++;
561 #ifdef CONFIG_PPC_BOOK3S_32
562 /* We set segments as unused segments when invalidating them. So
563 * treat the respective fault as segment fault. */
564 if ((to_svcpu(vcpu
)->sr
[dar
>> SID_SHIFT
]) == SR_INVALID
) {
565 kvmppc_mmu_map_segment(vcpu
, dar
);
571 /* The only case we need to handle is missing shadow PTEs */
572 if (to_svcpu(vcpu
)->fault_dsisr
& DSISR_NOHPTE
) {
573 r
= kvmppc_handle_pagefault(run
, vcpu
, dar
, exit_nr
);
575 vcpu
->arch
.shared
->dar
= dar
;
576 vcpu
->arch
.shared
->dsisr
= to_svcpu(vcpu
)->fault_dsisr
;
577 kvmppc_book3s_queue_irqprio(vcpu
, exit_nr
);
582 case BOOK3S_INTERRUPT_DATA_SEGMENT
:
583 if (kvmppc_mmu_map_segment(vcpu
, kvmppc_get_fault_dar(vcpu
)) < 0) {
584 vcpu
->arch
.shared
->dar
= kvmppc_get_fault_dar(vcpu
);
585 kvmppc_book3s_queue_irqprio(vcpu
,
586 BOOK3S_INTERRUPT_DATA_SEGMENT
);
590 case BOOK3S_INTERRUPT_INST_SEGMENT
:
591 if (kvmppc_mmu_map_segment(vcpu
, kvmppc_get_pc(vcpu
)) < 0) {
592 kvmppc_book3s_queue_irqprio(vcpu
,
593 BOOK3S_INTERRUPT_INST_SEGMENT
);
597 /* We're good on these - the host merely wanted to get our attention */
598 case BOOK3S_INTERRUPT_DECREMENTER
:
599 vcpu
->stat
.dec_exits
++;
602 case BOOK3S_INTERRUPT_EXTERNAL
:
603 vcpu
->stat
.ext_intr_exits
++;
606 case BOOK3S_INTERRUPT_PERFMON
:
609 case BOOK3S_INTERRUPT_PROGRAM
:
611 enum emulation_result er
;
615 flags
= to_svcpu(vcpu
)->shadow_srr1
& 0x1f0000ull
;
617 if (vcpu
->arch
.shared
->msr
& MSR_PR
) {
619 printk(KERN_INFO
"Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", kvmppc_get_pc(vcpu
), kvmppc_get_last_inst(vcpu
));
621 if ((kvmppc_get_last_inst(vcpu
) & 0xff0007ff) !=
622 (INS_DCBZ
& 0xfffffff7)) {
623 kvmppc_core_queue_program(vcpu
, flags
);
629 vcpu
->stat
.emulated_inst_exits
++;
630 er
= kvmppc_emulate_instruction(run
, vcpu
);
639 printk(KERN_CRIT
"%s: emulation at %lx failed (%08x)\n",
640 __func__
, kvmppc_get_pc(vcpu
), kvmppc_get_last_inst(vcpu
));
641 kvmppc_core_queue_program(vcpu
, flags
);
644 case EMULATE_DO_MMIO
:
645 run
->exit_reason
= KVM_EXIT_MMIO
;
653 case BOOK3S_INTERRUPT_SYSCALL
:
654 if (vcpu
->arch
.papr_enabled
&&
655 (kvmppc_get_last_inst(vcpu
) == 0x44000022) &&
656 !(vcpu
->arch
.shared
->msr
& MSR_PR
)) {
657 /* SC 1 papr hypercalls */
658 ulong cmd
= kvmppc_get_gpr(vcpu
, 3);
661 #ifdef CONFIG_KVM_BOOK3S_64_PR
662 if (kvmppc_h_pr(vcpu
, cmd
) == EMULATE_DONE
) {
668 run
->papr_hcall
.nr
= cmd
;
669 for (i
= 0; i
< 9; ++i
) {
670 ulong gpr
= kvmppc_get_gpr(vcpu
, 4 + i
);
671 run
->papr_hcall
.args
[i
] = gpr
;
673 run
->exit_reason
= KVM_EXIT_PAPR_HCALL
;
674 vcpu
->arch
.hcall_needed
= 1;
676 } else if (vcpu
->arch
.osi_enabled
&&
677 (((u32
)kvmppc_get_gpr(vcpu
, 3)) == OSI_SC_MAGIC_R3
) &&
678 (((u32
)kvmppc_get_gpr(vcpu
, 4)) == OSI_SC_MAGIC_R4
)) {
680 u64
*gprs
= run
->osi
.gprs
;
683 run
->exit_reason
= KVM_EXIT_OSI
;
684 for (i
= 0; i
< 32; i
++)
685 gprs
[i
] = kvmppc_get_gpr(vcpu
, i
);
686 vcpu
->arch
.osi_needed
= 1;
688 } else if (!(vcpu
->arch
.shared
->msr
& MSR_PR
) &&
689 (((u32
)kvmppc_get_gpr(vcpu
, 0)) == KVM_SC_MAGIC_R0
)) {
690 /* KVM PV hypercalls */
691 kvmppc_set_gpr(vcpu
, 3, kvmppc_kvm_pv(vcpu
));
695 vcpu
->stat
.syscall_exits
++;
696 kvmppc_book3s_queue_irqprio(vcpu
, exit_nr
);
700 case BOOK3S_INTERRUPT_FP_UNAVAIL
:
701 case BOOK3S_INTERRUPT_ALTIVEC
:
702 case BOOK3S_INTERRUPT_VSX
:
707 case BOOK3S_INTERRUPT_FP_UNAVAIL
: ext_msr
= MSR_FP
; break;
708 case BOOK3S_INTERRUPT_ALTIVEC
: ext_msr
= MSR_VEC
; break;
709 case BOOK3S_INTERRUPT_VSX
: ext_msr
= MSR_VSX
; break;
712 switch (kvmppc_check_ext(vcpu
, exit_nr
)) {
714 /* everything ok - let's enable the ext */
715 r
= kvmppc_handle_ext(vcpu
, exit_nr
, ext_msr
);
718 /* we need to emulate this instruction */
719 goto program_interrupt
;
722 /* nothing to worry about - go again */
727 case BOOK3S_INTERRUPT_ALIGNMENT
:
728 if (kvmppc_read_inst(vcpu
) == EMULATE_DONE
) {
729 vcpu
->arch
.shared
->dsisr
= kvmppc_alignment_dsisr(vcpu
,
730 kvmppc_get_last_inst(vcpu
));
731 vcpu
->arch
.shared
->dar
= kvmppc_alignment_dar(vcpu
,
732 kvmppc_get_last_inst(vcpu
));
733 kvmppc_book3s_queue_irqprio(vcpu
, exit_nr
);
737 case BOOK3S_INTERRUPT_MACHINE_CHECK
:
738 case BOOK3S_INTERRUPT_TRACE
:
739 kvmppc_book3s_queue_irqprio(vcpu
, exit_nr
);
743 /* Ugh - bork here! What did we get? */
744 printk(KERN_EMERG
"exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
745 exit_nr
, kvmppc_get_pc(vcpu
), to_svcpu(vcpu
)->shadow_srr1
);
752 if (!(r
& RESUME_HOST
)) {
753 /* To avoid clobbering exit_reason, only check for signals if
754 * we aren't already exiting to userspace for some other
756 if (signal_pending(current
)) {
758 printk(KERN_EMERG
"KVM: Going back to host\n");
760 vcpu
->stat
.signal_exits
++;
761 run
->exit_reason
= KVM_EXIT_INTR
;
764 /* In case an interrupt came in that was triggered
765 * from userspace (like DEC), we need to check what
767 kvmppc_core_deliver_interrupts(vcpu
);
771 trace_kvm_book3s_reenter(r
, vcpu
);
776 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu
*vcpu
,
777 struct kvm_sregs
*sregs
)
779 struct kvmppc_vcpu_book3s
*vcpu3s
= to_book3s(vcpu
);
782 sregs
->pvr
= vcpu
->arch
.pvr
;
784 sregs
->u
.s
.sdr1
= to_book3s(vcpu
)->sdr1
;
785 if (vcpu
->arch
.hflags
& BOOK3S_HFLAG_SLB
) {
786 for (i
= 0; i
< 64; i
++) {
787 sregs
->u
.s
.ppc64
.slb
[i
].slbe
= vcpu
->arch
.slb
[i
].orige
| i
;
788 sregs
->u
.s
.ppc64
.slb
[i
].slbv
= vcpu
->arch
.slb
[i
].origv
;
791 for (i
= 0; i
< 16; i
++)
792 sregs
->u
.s
.ppc32
.sr
[i
] = vcpu
->arch
.shared
->sr
[i
];
794 for (i
= 0; i
< 8; i
++) {
795 sregs
->u
.s
.ppc32
.ibat
[i
] = vcpu3s
->ibat
[i
].raw
;
796 sregs
->u
.s
.ppc32
.dbat
[i
] = vcpu3s
->dbat
[i
].raw
;
803 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu
*vcpu
,
804 struct kvm_sregs
*sregs
)
806 struct kvmppc_vcpu_book3s
*vcpu3s
= to_book3s(vcpu
);
809 kvmppc_set_pvr(vcpu
, sregs
->pvr
);
811 vcpu3s
->sdr1
= sregs
->u
.s
.sdr1
;
812 if (vcpu
->arch
.hflags
& BOOK3S_HFLAG_SLB
) {
813 for (i
= 0; i
< 64; i
++) {
814 vcpu
->arch
.mmu
.slbmte(vcpu
, sregs
->u
.s
.ppc64
.slb
[i
].slbv
,
815 sregs
->u
.s
.ppc64
.slb
[i
].slbe
);
818 for (i
= 0; i
< 16; i
++) {
819 vcpu
->arch
.mmu
.mtsrin(vcpu
, i
, sregs
->u
.s
.ppc32
.sr
[i
]);
821 for (i
= 0; i
< 8; i
++) {
822 kvmppc_set_bat(vcpu
, &(vcpu3s
->ibat
[i
]), false,
823 (u32
)sregs
->u
.s
.ppc32
.ibat
[i
]);
824 kvmppc_set_bat(vcpu
, &(vcpu3s
->ibat
[i
]), true,
825 (u32
)(sregs
->u
.s
.ppc32
.ibat
[i
] >> 32));
826 kvmppc_set_bat(vcpu
, &(vcpu3s
->dbat
[i
]), false,
827 (u32
)sregs
->u
.s
.ppc32
.dbat
[i
]);
828 kvmppc_set_bat(vcpu
, &(vcpu3s
->dbat
[i
]), true,
829 (u32
)(sregs
->u
.s
.ppc32
.dbat
[i
] >> 32));
833 /* Flush the MMU after messing with the segments */
834 kvmppc_mmu_pte_flush(vcpu
, 0, 0);
839 int kvmppc_core_check_processor_compat(void)
844 struct kvm_vcpu
*kvmppc_core_vcpu_create(struct kvm
*kvm
, unsigned int id
)
846 struct kvmppc_vcpu_book3s
*vcpu_book3s
;
847 struct kvm_vcpu
*vcpu
;
851 vcpu_book3s
= vzalloc(sizeof(struct kvmppc_vcpu_book3s
));
855 vcpu_book3s
->shadow_vcpu
= (struct kvmppc_book3s_shadow_vcpu
*)
856 kzalloc(sizeof(*vcpu_book3s
->shadow_vcpu
), GFP_KERNEL
);
857 if (!vcpu_book3s
->shadow_vcpu
)
860 vcpu
= &vcpu_book3s
->vcpu
;
861 err
= kvm_vcpu_init(vcpu
, kvm
, id
);
863 goto free_shadow_vcpu
;
865 p
= __get_free_page(GFP_KERNEL
|__GFP_ZERO
);
866 /* the real shared page fills the last 4k of our page */
867 vcpu
->arch
.shared
= (void*)(p
+ PAGE_SIZE
- 4096);
871 #ifdef CONFIG_PPC_BOOK3S_64
872 /* default to book3s_64 (970fx) */
873 vcpu
->arch
.pvr
= 0x3C0301;
875 /* default to book3s_32 (750) */
876 vcpu
->arch
.pvr
= 0x84202;
878 kvmppc_set_pvr(vcpu
, vcpu
->arch
.pvr
);
879 vcpu
->arch
.slb_nr
= 64;
881 vcpu
->arch
.shadow_msr
= MSR_USER64
;
883 err
= kvmppc_mmu_init(vcpu
);
890 kvm_vcpu_uninit(vcpu
);
892 kfree(vcpu_book3s
->shadow_vcpu
);
899 void kvmppc_core_vcpu_free(struct kvm_vcpu
*vcpu
)
901 struct kvmppc_vcpu_book3s
*vcpu_book3s
= to_book3s(vcpu
);
903 free_page((unsigned long)vcpu
->arch
.shared
& PAGE_MASK
);
904 kvm_vcpu_uninit(vcpu
);
905 kfree(vcpu_book3s
->shadow_vcpu
);
909 int kvmppc_vcpu_run(struct kvm_run
*kvm_run
, struct kvm_vcpu
*vcpu
)
912 double fpr
[32][TS_FPRWIDTH
];
915 #ifdef CONFIG_ALTIVEC
918 unsigned long uninitialized_var(vrsave
);
926 /* Check if we can run the vcpu at all */
927 if (!vcpu
->arch
.sane
) {
928 kvm_run
->exit_reason
= KVM_EXIT_INTERNAL_ERROR
;
932 /* No need to go into the guest when all we do is going out */
933 if (signal_pending(current
)) {
934 kvm_run
->exit_reason
= KVM_EXIT_INTR
;
938 /* Save FPU state in stack */
939 if (current
->thread
.regs
->msr
& MSR_FP
)
941 memcpy(fpr
, current
->thread
.fpr
, sizeof(current
->thread
.fpr
));
942 fpscr
= current
->thread
.fpscr
.val
;
943 fpexc_mode
= current
->thread
.fpexc_mode
;
945 #ifdef CONFIG_ALTIVEC
946 /* Save Altivec state in stack */
947 used_vr
= current
->thread
.used_vr
;
949 if (current
->thread
.regs
->msr
& MSR_VEC
)
950 giveup_altivec(current
);
951 memcpy(vr
, current
->thread
.vr
, sizeof(current
->thread
.vr
));
952 vscr
= current
->thread
.vscr
;
953 vrsave
= current
->thread
.vrsave
;
958 /* Save VSX state in stack */
959 used_vsr
= current
->thread
.used_vsr
;
960 if (used_vsr
&& (current
->thread
.regs
->msr
& MSR_VSX
))
961 __giveup_vsx(current
);
964 /* Remember the MSR with disabled extensions */
965 ext_msr
= current
->thread
.regs
->msr
;
967 /* Preload FPU if it's enabled */
968 if (vcpu
->arch
.shared
->msr
& MSR_FP
)
969 kvmppc_handle_ext(vcpu
, BOOK3S_INTERRUPT_FP_UNAVAIL
, MSR_FP
);
973 ret
= __kvmppc_vcpu_run(kvm_run
, vcpu
);
979 current
->thread
.regs
->msr
= ext_msr
;
981 /* Make sure we save the guest FPU/Altivec/VSX state */
982 kvmppc_giveup_ext(vcpu
, MSR_FP
);
983 kvmppc_giveup_ext(vcpu
, MSR_VEC
);
984 kvmppc_giveup_ext(vcpu
, MSR_VSX
);
986 /* Restore FPU state from stack */
987 memcpy(current
->thread
.fpr
, fpr
, sizeof(current
->thread
.fpr
));
988 current
->thread
.fpscr
.val
= fpscr
;
989 current
->thread
.fpexc_mode
= fpexc_mode
;
991 #ifdef CONFIG_ALTIVEC
992 /* Restore Altivec state from stack */
993 if (used_vr
&& current
->thread
.used_vr
) {
994 memcpy(current
->thread
.vr
, vr
, sizeof(current
->thread
.vr
));
995 current
->thread
.vscr
= vscr
;
996 current
->thread
.vrsave
= vrsave
;
998 current
->thread
.used_vr
= used_vr
;
1002 current
->thread
.used_vsr
= used_vsr
;
1008 int kvmppc_core_prepare_memory_region(struct kvm
*kvm
,
1009 struct kvm_userspace_memory_region
*mem
)
1014 void kvmppc_core_commit_memory_region(struct kvm
*kvm
,
1015 struct kvm_userspace_memory_region
*mem
)
1019 int kvmppc_core_init_vm(struct kvm
*kvm
)
1024 void kvmppc_core_destroy_vm(struct kvm
*kvm
)
1028 static int kvmppc_book3s_init(void)
1032 r
= kvm_init(NULL
, sizeof(struct kvmppc_vcpu_book3s
), 0,
1038 r
= kvmppc_mmu_hpte_sysinit();
1043 static void kvmppc_book3s_exit(void)
1045 kvmppc_mmu_hpte_sysexit();
1049 module_init(kvmppc_book3s_init
);
1050 module_exit(kvmppc_book3s_exit
);