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 <asm/switch_to.h>
37 #include <asm/firmware.h>
38 #include <asm/hvcall.h>
39 #include <linux/gfp.h>
40 #include <linux/sched.h>
41 #include <linux/vmalloc.h>
42 #include <linux/highmem.h>
46 /* #define EXIT_DEBUG */
47 /* #define DEBUG_EXT */
49 static int kvmppc_handle_ext(struct kvm_vcpu
*vcpu
, unsigned int exit_nr
,
52 /* Some compatibility defines */
53 #ifdef CONFIG_PPC_BOOK3S_32
54 #define MSR_USER32 MSR_USER
55 #define MSR_USER64 MSR_USER
56 #define HW_PAGE_SIZE PAGE_SIZE
59 void kvmppc_core_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
61 #ifdef CONFIG_PPC_BOOK3S_64
62 struct kvmppc_book3s_shadow_vcpu
*svcpu
= svcpu_get(vcpu
);
63 memcpy(svcpu
->slb
, to_book3s(vcpu
)->slb_shadow
, sizeof(svcpu
->slb
));
64 memcpy(&get_paca()->shadow_vcpu
, to_book3s(vcpu
)->shadow_vcpu
,
65 sizeof(get_paca()->shadow_vcpu
));
66 svcpu
->slb_max
= to_book3s(vcpu
)->slb_shadow_max
;
69 vcpu
->cpu
= smp_processor_id();
70 #ifdef CONFIG_PPC_BOOK3S_32
71 current
->thread
.kvm_shadow_vcpu
= to_book3s(vcpu
)->shadow_vcpu
;
75 void kvmppc_core_vcpu_put(struct kvm_vcpu
*vcpu
)
77 #ifdef CONFIG_PPC_BOOK3S_64
78 struct kvmppc_book3s_shadow_vcpu
*svcpu
= svcpu_get(vcpu
);
79 memcpy(to_book3s(vcpu
)->slb_shadow
, svcpu
->slb
, sizeof(svcpu
->slb
));
80 memcpy(to_book3s(vcpu
)->shadow_vcpu
, &get_paca()->shadow_vcpu
,
81 sizeof(get_paca()->shadow_vcpu
));
82 to_book3s(vcpu
)->slb_shadow_max
= svcpu
->slb_max
;
86 kvmppc_giveup_ext(vcpu
, MSR_FP
| MSR_VEC
| MSR_VSX
);
90 int kvmppc_core_check_requests(struct kvm_vcpu
*vcpu
)
92 int r
= 1; /* Indicate we want to get back into the guest */
94 /* We misuse TLB_FLUSH to indicate that we want to clear
95 all shadow cache entries */
96 if (kvm_check_request(KVM_REQ_TLB_FLUSH
, vcpu
))
97 kvmppc_mmu_pte_flush(vcpu
, 0, 0);
102 /************* MMU Notifiers *************/
104 int kvm_unmap_hva(struct kvm
*kvm
, unsigned long hva
)
106 trace_kvm_unmap_hva(hva
);
109 * Flush all shadow tlb entries everywhere. This is slow, but
110 * we are 100% sure that we catch the to be unmapped page
112 kvm_flush_remote_tlbs(kvm
);
117 int kvm_unmap_hva_range(struct kvm
*kvm
, unsigned long start
, unsigned long end
)
119 /* kvm_unmap_hva flushes everything anyways */
120 kvm_unmap_hva(kvm
, start
);
125 int kvm_age_hva(struct kvm
*kvm
, unsigned long hva
)
127 /* XXX could be more clever ;) */
131 int kvm_test_age_hva(struct kvm
*kvm
, unsigned long hva
)
133 /* XXX could be more clever ;) */
137 void kvm_set_spte_hva(struct kvm
*kvm
, unsigned long hva
, pte_t pte
)
139 /* The page will get remapped properly on its next fault */
140 kvm_unmap_hva(kvm
, hva
);
143 /*****************************************/
145 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu
*vcpu
)
147 ulong smsr
= vcpu
->arch
.shared
->msr
;
149 /* Guest MSR values */
150 smsr
&= MSR_FE0
| MSR_FE1
| MSR_SF
| MSR_SE
| MSR_BE
;
151 /* Process MSR values */
152 smsr
|= MSR_ME
| MSR_RI
| MSR_IR
| MSR_DR
| MSR_PR
| MSR_EE
;
153 /* External providers the guest reserved */
154 smsr
|= (vcpu
->arch
.shared
->msr
& vcpu
->arch
.guest_owned_ext
);
155 /* 64-bit Process MSR values */
156 #ifdef CONFIG_PPC_BOOK3S_64
157 smsr
|= MSR_ISF
| MSR_HV
;
159 vcpu
->arch
.shadow_msr
= smsr
;
162 void kvmppc_set_msr(struct kvm_vcpu
*vcpu
, u64 msr
)
164 ulong old_msr
= vcpu
->arch
.shared
->msr
;
167 printk(KERN_INFO
"KVM: Set MSR to 0x%llx\n", msr
);
170 msr
&= to_book3s(vcpu
)->msr_mask
;
171 vcpu
->arch
.shared
->msr
= msr
;
172 kvmppc_recalc_shadow_msr(vcpu
);
175 if (!vcpu
->arch
.pending_exceptions
) {
176 kvm_vcpu_block(vcpu
);
177 clear_bit(KVM_REQ_UNHALT
, &vcpu
->requests
);
178 vcpu
->stat
.halt_wakeup
++;
180 /* Unset POW bit after we woke up */
182 vcpu
->arch
.shared
->msr
= msr
;
186 if ((vcpu
->arch
.shared
->msr
& (MSR_PR
|MSR_IR
|MSR_DR
)) !=
187 (old_msr
& (MSR_PR
|MSR_IR
|MSR_DR
))) {
188 kvmppc_mmu_flush_segments(vcpu
);
189 kvmppc_mmu_map_segment(vcpu
, kvmppc_get_pc(vcpu
));
191 /* Preload magic page segment when in kernel mode */
192 if (!(msr
& MSR_PR
) && vcpu
->arch
.magic_page_pa
) {
193 struct kvm_vcpu_arch
*a
= &vcpu
->arch
;
196 kvmppc_mmu_map_segment(vcpu
, a
->magic_page_ea
);
198 kvmppc_mmu_map_segment(vcpu
, a
->magic_page_pa
);
203 * When switching from 32 to 64-bit, we may have a stale 32-bit
204 * magic page around, we need to flush it. Typically 32-bit magic
205 * page will be instanciated when calling into RTAS. Note: We
206 * assume that such transition only happens while in kernel mode,
207 * ie, we never transition from user 32-bit to kernel 64-bit with
208 * a 32-bit magic page around.
210 if (vcpu
->arch
.magic_page_pa
&&
211 !(old_msr
& MSR_PR
) && !(old_msr
& MSR_SF
) && (msr
& MSR_SF
)) {
212 /* going from RTAS to normal kernel code */
213 kvmppc_mmu_pte_flush(vcpu
, (uint32_t)vcpu
->arch
.magic_page_pa
,
217 /* Preload FPU if it's enabled */
218 if (vcpu
->arch
.shared
->msr
& MSR_FP
)
219 kvmppc_handle_ext(vcpu
, BOOK3S_INTERRUPT_FP_UNAVAIL
, MSR_FP
);
222 void kvmppc_set_pvr(struct kvm_vcpu
*vcpu
, u32 pvr
)
226 vcpu
->arch
.hflags
&= ~BOOK3S_HFLAG_SLB
;
227 vcpu
->arch
.pvr
= pvr
;
228 #ifdef CONFIG_PPC_BOOK3S_64
229 if ((pvr
>= 0x330000) && (pvr
< 0x70330000)) {
230 kvmppc_mmu_book3s_64_init(vcpu
);
231 if (!to_book3s(vcpu
)->hior_explicit
)
232 to_book3s(vcpu
)->hior
= 0xfff00000;
233 to_book3s(vcpu
)->msr_mask
= 0xffffffffffffffffULL
;
234 vcpu
->arch
.cpu_type
= KVM_CPU_3S_64
;
238 kvmppc_mmu_book3s_32_init(vcpu
);
239 if (!to_book3s(vcpu
)->hior_explicit
)
240 to_book3s(vcpu
)->hior
= 0;
241 to_book3s(vcpu
)->msr_mask
= 0xffffffffULL
;
242 vcpu
->arch
.cpu_type
= KVM_CPU_3S_32
;
245 kvmppc_sanity_check(vcpu
);
247 /* If we are in hypervisor level on 970, we can tell the CPU to
248 * treat DCBZ as 32 bytes store */
249 vcpu
->arch
.hflags
&= ~BOOK3S_HFLAG_DCBZ32
;
250 if (vcpu
->arch
.mmu
.is_dcbz32(vcpu
) && (mfmsr() & MSR_HV
) &&
251 !strcmp(cur_cpu_spec
->platform
, "ppc970"))
252 vcpu
->arch
.hflags
|= BOOK3S_HFLAG_DCBZ32
;
254 /* Cell performs badly if MSR_FEx are set. So let's hope nobody
255 really needs them in a VM on Cell and force disable them. */
256 if (!strcmp(cur_cpu_spec
->platform
, "ppc-cell-be"))
257 to_book3s(vcpu
)->msr_mask
&= ~(MSR_FE0
| MSR_FE1
);
259 #ifdef CONFIG_PPC_BOOK3S_32
260 /* 32 bit Book3S always has 32 byte dcbz */
261 vcpu
->arch
.hflags
|= BOOK3S_HFLAG_DCBZ32
;
264 /* On some CPUs we can execute paired single operations natively */
265 asm ( "mfpvr %0" : "=r"(host_pvr
));
267 case 0x00080200: /* lonestar 2.0 */
268 case 0x00088202: /* lonestar 2.2 */
269 case 0x70000100: /* gekko 1.0 */
270 case 0x00080100: /* gekko 2.0 */
271 case 0x00083203: /* gekko 2.3a */
272 case 0x00083213: /* gekko 2.3b */
273 case 0x00083204: /* gekko 2.4 */
274 case 0x00083214: /* gekko 2.4e (8SE) - retail HW2 */
275 case 0x00087200: /* broadway */
276 vcpu
->arch
.hflags
|= BOOK3S_HFLAG_NATIVE_PS
;
277 /* Enable HID2.PSE - in case we need it later */
278 mtspr(SPRN_HID2_GEKKO
, mfspr(SPRN_HID2_GEKKO
) | (1 << 29));
282 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
283 * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
284 * emulate 32 bytes dcbz length.
286 * The Book3s_64 inventors also realized this case and implemented a special bit
287 * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
289 * My approach here is to patch the dcbz instruction on executing pages.
291 static void kvmppc_patch_dcbz(struct kvm_vcpu
*vcpu
, struct kvmppc_pte
*pte
)
298 hpage
= gfn_to_page(vcpu
->kvm
, pte
->raddr
>> PAGE_SHIFT
);
299 if (is_error_page(hpage
))
302 hpage_offset
= pte
->raddr
& ~PAGE_MASK
;
303 hpage_offset
&= ~0xFFFULL
;
307 page
= kmap_atomic(hpage
);
309 /* patch dcbz into reserved instruction, so we trap */
310 for (i
=hpage_offset
; i
< hpage_offset
+ (HW_PAGE_SIZE
/ 4); i
++)
311 if ((page
[i
] & 0xff0007ff) == INS_DCBZ
)
312 page
[i
] &= 0xfffffff7;
318 static int kvmppc_visible_gfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
320 ulong mp_pa
= vcpu
->arch
.magic_page_pa
;
322 if (!(vcpu
->arch
.shared
->msr
& MSR_SF
))
323 mp_pa
= (uint32_t)mp_pa
;
325 if (unlikely(mp_pa
) &&
326 unlikely((mp_pa
& KVM_PAM
) >> PAGE_SHIFT
== gfn
)) {
330 return kvm_is_visible_gfn(vcpu
->kvm
, gfn
);
333 int kvmppc_handle_pagefault(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
,
334 ulong eaddr
, int vec
)
336 bool data
= (vec
== BOOK3S_INTERRUPT_DATA_STORAGE
);
337 int r
= RESUME_GUEST
;
340 struct kvmppc_pte pte
;
341 bool is_mmio
= false;
342 bool dr
= (vcpu
->arch
.shared
->msr
& MSR_DR
) ? true : false;
343 bool ir
= (vcpu
->arch
.shared
->msr
& MSR_IR
) ? true : false;
346 relocated
= data
? dr
: ir
;
348 /* Resolve real address if translation turned on */
350 page_found
= vcpu
->arch
.mmu
.xlate(vcpu
, eaddr
, &pte
, data
);
352 pte
.may_execute
= true;
354 pte
.may_write
= true;
355 pte
.raddr
= eaddr
& KVM_PAM
;
357 pte
.vpage
= eaddr
>> 12;
360 switch (vcpu
->arch
.shared
->msr
& (MSR_DR
|MSR_IR
)) {
362 pte
.vpage
|= ((u64
)VSID_REAL
<< (SID_SHIFT
- 12));
366 vcpu
->arch
.mmu
.esid_to_vsid(vcpu
, eaddr
>> SID_SHIFT
, &vsid
);
368 if ((vcpu
->arch
.shared
->msr
& (MSR_DR
|MSR_IR
)) == MSR_DR
)
369 pte
.vpage
|= ((u64
)VSID_REAL_DR
<< (SID_SHIFT
- 12));
371 pte
.vpage
|= ((u64
)VSID_REAL_IR
<< (SID_SHIFT
- 12));
375 page_found
= -EINVAL
;
379 if (vcpu
->arch
.mmu
.is_dcbz32(vcpu
) &&
380 (!(vcpu
->arch
.hflags
& BOOK3S_HFLAG_DCBZ32
))) {
382 * If we do the dcbz hack, we have to NX on every execution,
383 * so we can patch the executing code. This renders our guest
386 pte
.may_execute
= !data
;
389 if (page_found
== -ENOENT
) {
390 /* Page not found in guest PTE entries */
391 struct kvmppc_book3s_shadow_vcpu
*svcpu
= svcpu_get(vcpu
);
392 vcpu
->arch
.shared
->dar
= kvmppc_get_fault_dar(vcpu
);
393 vcpu
->arch
.shared
->dsisr
= svcpu
->fault_dsisr
;
394 vcpu
->arch
.shared
->msr
|=
395 (svcpu
->shadow_srr1
& 0x00000000f8000000ULL
);
397 kvmppc_book3s_queue_irqprio(vcpu
, vec
);
398 } else if (page_found
== -EPERM
) {
399 /* Storage protection */
400 struct kvmppc_book3s_shadow_vcpu
*svcpu
= svcpu_get(vcpu
);
401 vcpu
->arch
.shared
->dar
= kvmppc_get_fault_dar(vcpu
);
402 vcpu
->arch
.shared
->dsisr
= svcpu
->fault_dsisr
& ~DSISR_NOHPTE
;
403 vcpu
->arch
.shared
->dsisr
|= DSISR_PROTFAULT
;
404 vcpu
->arch
.shared
->msr
|=
405 svcpu
->shadow_srr1
& 0x00000000f8000000ULL
;
407 kvmppc_book3s_queue_irqprio(vcpu
, vec
);
408 } else if (page_found
== -EINVAL
) {
409 /* Page not found in guest SLB */
410 vcpu
->arch
.shared
->dar
= kvmppc_get_fault_dar(vcpu
);
411 kvmppc_book3s_queue_irqprio(vcpu
, vec
+ 0x80);
412 } else if (!is_mmio
&&
413 kvmppc_visible_gfn(vcpu
, pte
.raddr
>> PAGE_SHIFT
)) {
414 /* The guest's PTE is not mapped yet. Map on the host */
415 kvmppc_mmu_map_page(vcpu
, &pte
);
417 vcpu
->stat
.sp_storage
++;
418 else if (vcpu
->arch
.mmu
.is_dcbz32(vcpu
) &&
419 (!(vcpu
->arch
.hflags
& BOOK3S_HFLAG_DCBZ32
)))
420 kvmppc_patch_dcbz(vcpu
, &pte
);
423 vcpu
->stat
.mmio_exits
++;
424 vcpu
->arch
.paddr_accessed
= pte
.raddr
;
425 vcpu
->arch
.vaddr_accessed
= pte
.eaddr
;
426 r
= kvmppc_emulate_mmio(run
, vcpu
);
427 if ( r
== RESUME_HOST_NV
)
434 static inline int get_fpr_index(int i
)
436 return i
* TS_FPRWIDTH
;
439 /* Give up external provider (FPU, Altivec, VSX) */
440 void kvmppc_giveup_ext(struct kvm_vcpu
*vcpu
, ulong msr
)
442 struct thread_struct
*t
= ¤t
->thread
;
443 u64
*vcpu_fpr
= vcpu
->arch
.fpr
;
445 u64
*vcpu_vsx
= vcpu
->arch
.vsr
;
447 u64
*thread_fpr
= (u64
*)t
->fpr
;
451 * VSX instructions can access FP and vector registers, so if
452 * we are giving up VSX, make sure we give up FP and VMX as well.
455 msr
|= MSR_FP
| MSR_VEC
;
457 msr
&= vcpu
->arch
.guest_owned_ext
;
462 printk(KERN_INFO
"Giving up ext 0x%lx\n", msr
);
467 * Note that on CPUs with VSX, giveup_fpu stores
468 * both the traditional FP registers and the added VSX
469 * registers into thread.fpr[].
472 for (i
= 0; i
< ARRAY_SIZE(vcpu
->arch
.fpr
); i
++)
473 vcpu_fpr
[i
] = thread_fpr
[get_fpr_index(i
)];
475 vcpu
->arch
.fpscr
= t
->fpscr
.val
;
478 if (cpu_has_feature(CPU_FTR_VSX
))
479 for (i
= 0; i
< ARRAY_SIZE(vcpu
->arch
.vsr
) / 2; i
++)
480 vcpu_vsx
[i
] = thread_fpr
[get_fpr_index(i
) + 1];
484 #ifdef CONFIG_ALTIVEC
486 giveup_altivec(current
);
487 memcpy(vcpu
->arch
.vr
, t
->vr
, sizeof(vcpu
->arch
.vr
));
488 vcpu
->arch
.vscr
= t
->vscr
;
492 vcpu
->arch
.guest_owned_ext
&= ~(msr
| MSR_VSX
);
493 kvmppc_recalc_shadow_msr(vcpu
);
496 static int kvmppc_read_inst(struct kvm_vcpu
*vcpu
)
498 ulong srr0
= kvmppc_get_pc(vcpu
);
499 u32 last_inst
= kvmppc_get_last_inst(vcpu
);
502 ret
= kvmppc_ld(vcpu
, &srr0
, sizeof(u32
), &last_inst
, false);
503 if (ret
== -ENOENT
) {
504 ulong msr
= vcpu
->arch
.shared
->msr
;
506 msr
= kvmppc_set_field(msr
, 33, 33, 1);
507 msr
= kvmppc_set_field(msr
, 34, 36, 0);
508 vcpu
->arch
.shared
->msr
= kvmppc_set_field(msr
, 42, 47, 0);
509 kvmppc_book3s_queue_irqprio(vcpu
, BOOK3S_INTERRUPT_INST_STORAGE
);
510 return EMULATE_AGAIN
;
516 static int kvmppc_check_ext(struct kvm_vcpu
*vcpu
, unsigned int exit_nr
)
519 /* Need to do paired single emulation? */
520 if (!(vcpu
->arch
.hflags
& BOOK3S_HFLAG_PAIRED_SINGLE
))
523 /* Read out the instruction */
524 if (kvmppc_read_inst(vcpu
) == EMULATE_DONE
)
525 /* Need to emulate */
528 return EMULATE_AGAIN
;
531 /* Handle external providers (FPU, Altivec, VSX) */
532 static int kvmppc_handle_ext(struct kvm_vcpu
*vcpu
, unsigned int exit_nr
,
535 struct thread_struct
*t
= ¤t
->thread
;
536 u64
*vcpu_fpr
= vcpu
->arch
.fpr
;
538 u64
*vcpu_vsx
= vcpu
->arch
.vsr
;
540 u64
*thread_fpr
= (u64
*)t
->fpr
;
543 /* When we have paired singles, we emulate in software */
544 if (vcpu
->arch
.hflags
& BOOK3S_HFLAG_PAIRED_SINGLE
)
547 if (!(vcpu
->arch
.shared
->msr
& msr
)) {
548 kvmppc_book3s_queue_irqprio(vcpu
, exit_nr
);
552 if (msr
== MSR_VSX
) {
553 /* No VSX? Give an illegal instruction interrupt */
555 if (!cpu_has_feature(CPU_FTR_VSX
))
558 kvmppc_core_queue_program(vcpu
, SRR1_PROGILL
);
563 * We have to load up all the FP and VMX registers before
564 * we can let the guest use VSX instructions.
566 msr
= MSR_FP
| MSR_VEC
| MSR_VSX
;
569 /* See if we already own all the ext(s) needed */
570 msr
&= ~vcpu
->arch
.guest_owned_ext
;
575 printk(KERN_INFO
"Loading up ext 0x%lx\n", msr
);
578 current
->thread
.regs
->msr
|= msr
;
581 for (i
= 0; i
< ARRAY_SIZE(vcpu
->arch
.fpr
); i
++)
582 thread_fpr
[get_fpr_index(i
)] = vcpu_fpr
[i
];
584 for (i
= 0; i
< ARRAY_SIZE(vcpu
->arch
.vsr
) / 2; i
++)
585 thread_fpr
[get_fpr_index(i
) + 1] = vcpu_vsx
[i
];
587 t
->fpscr
.val
= vcpu
->arch
.fpscr
;
589 kvmppc_load_up_fpu();
593 #ifdef CONFIG_ALTIVEC
594 memcpy(t
->vr
, vcpu
->arch
.vr
, sizeof(vcpu
->arch
.vr
));
595 t
->vscr
= vcpu
->arch
.vscr
;
597 kvmppc_load_up_altivec();
601 vcpu
->arch
.guest_owned_ext
|= msr
;
602 kvmppc_recalc_shadow_msr(vcpu
);
607 int kvmppc_handle_exit(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
,
608 unsigned int exit_nr
)
613 vcpu
->stat
.sum_exits
++;
615 run
->exit_reason
= KVM_EXIT_UNKNOWN
;
616 run
->ready_for_interrupt_injection
= 1;
618 /* We get here with MSR.EE=1 */
620 trace_kvm_exit(exit_nr
, vcpu
);
624 case BOOK3S_INTERRUPT_INST_STORAGE
:
626 struct kvmppc_book3s_shadow_vcpu
*svcpu
= svcpu_get(vcpu
);
627 ulong shadow_srr1
= svcpu
->shadow_srr1
;
628 vcpu
->stat
.pf_instruc
++;
630 #ifdef CONFIG_PPC_BOOK3S_32
631 /* We set segments as unused segments when invalidating them. So
632 * treat the respective fault as segment fault. */
633 if (svcpu
->sr
[kvmppc_get_pc(vcpu
) >> SID_SHIFT
] == SR_INVALID
) {
634 kvmppc_mmu_map_segment(vcpu
, kvmppc_get_pc(vcpu
));
642 /* only care about PTEG not found errors, but leave NX alone */
643 if (shadow_srr1
& 0x40000000) {
644 r
= kvmppc_handle_pagefault(run
, vcpu
, kvmppc_get_pc(vcpu
), exit_nr
);
645 vcpu
->stat
.sp_instruc
++;
646 } else if (vcpu
->arch
.mmu
.is_dcbz32(vcpu
) &&
647 (!(vcpu
->arch
.hflags
& BOOK3S_HFLAG_DCBZ32
))) {
649 * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
650 * so we can't use the NX bit inside the guest. Let's cross our fingers,
651 * that no guest that needs the dcbz hack does NX.
653 kvmppc_mmu_pte_flush(vcpu
, kvmppc_get_pc(vcpu
), ~0xFFFUL
);
656 vcpu
->arch
.shared
->msr
|= shadow_srr1
& 0x58000000;
657 kvmppc_book3s_queue_irqprio(vcpu
, exit_nr
);
662 case BOOK3S_INTERRUPT_DATA_STORAGE
:
664 ulong dar
= kvmppc_get_fault_dar(vcpu
);
665 struct kvmppc_book3s_shadow_vcpu
*svcpu
= svcpu_get(vcpu
);
666 u32 fault_dsisr
= svcpu
->fault_dsisr
;
667 vcpu
->stat
.pf_storage
++;
669 #ifdef CONFIG_PPC_BOOK3S_32
670 /* We set segments as unused segments when invalidating them. So
671 * treat the respective fault as segment fault. */
672 if ((svcpu
->sr
[dar
>> SID_SHIFT
]) == SR_INVALID
) {
673 kvmppc_mmu_map_segment(vcpu
, dar
);
681 /* The only case we need to handle is missing shadow PTEs */
682 if (fault_dsisr
& DSISR_NOHPTE
) {
683 r
= kvmppc_handle_pagefault(run
, vcpu
, dar
, exit_nr
);
685 vcpu
->arch
.shared
->dar
= dar
;
686 vcpu
->arch
.shared
->dsisr
= fault_dsisr
;
687 kvmppc_book3s_queue_irqprio(vcpu
, exit_nr
);
692 case BOOK3S_INTERRUPT_DATA_SEGMENT
:
693 if (kvmppc_mmu_map_segment(vcpu
, kvmppc_get_fault_dar(vcpu
)) < 0) {
694 vcpu
->arch
.shared
->dar
= kvmppc_get_fault_dar(vcpu
);
695 kvmppc_book3s_queue_irqprio(vcpu
,
696 BOOK3S_INTERRUPT_DATA_SEGMENT
);
700 case BOOK3S_INTERRUPT_INST_SEGMENT
:
701 if (kvmppc_mmu_map_segment(vcpu
, kvmppc_get_pc(vcpu
)) < 0) {
702 kvmppc_book3s_queue_irqprio(vcpu
,
703 BOOK3S_INTERRUPT_INST_SEGMENT
);
707 /* We're good on these - the host merely wanted to get our attention */
708 case BOOK3S_INTERRUPT_DECREMENTER
:
709 case BOOK3S_INTERRUPT_HV_DECREMENTER
:
710 vcpu
->stat
.dec_exits
++;
713 case BOOK3S_INTERRUPT_EXTERNAL
:
714 case BOOK3S_INTERRUPT_EXTERNAL_LEVEL
:
715 case BOOK3S_INTERRUPT_EXTERNAL_HV
:
716 vcpu
->stat
.ext_intr_exits
++;
719 case BOOK3S_INTERRUPT_PERFMON
:
722 case BOOK3S_INTERRUPT_PROGRAM
:
723 case BOOK3S_INTERRUPT_H_EMUL_ASSIST
:
725 enum emulation_result er
;
726 struct kvmppc_book3s_shadow_vcpu
*svcpu
;
730 svcpu
= svcpu_get(vcpu
);
731 flags
= svcpu
->shadow_srr1
& 0x1f0000ull
;
734 if (vcpu
->arch
.shared
->msr
& MSR_PR
) {
736 printk(KERN_INFO
"Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", kvmppc_get_pc(vcpu
), kvmppc_get_last_inst(vcpu
));
738 if ((kvmppc_get_last_inst(vcpu
) & 0xff0007ff) !=
739 (INS_DCBZ
& 0xfffffff7)) {
740 kvmppc_core_queue_program(vcpu
, flags
);
746 vcpu
->stat
.emulated_inst_exits
++;
747 er
= kvmppc_emulate_instruction(run
, vcpu
);
756 printk(KERN_CRIT
"%s: emulation at %lx failed (%08x)\n",
757 __func__
, kvmppc_get_pc(vcpu
), kvmppc_get_last_inst(vcpu
));
758 kvmppc_core_queue_program(vcpu
, flags
);
761 case EMULATE_DO_MMIO
:
762 run
->exit_reason
= KVM_EXIT_MMIO
;
765 case EMULATE_EXIT_USER
:
773 case BOOK3S_INTERRUPT_SYSCALL
:
774 if (vcpu
->arch
.papr_enabled
&&
775 (kvmppc_get_last_inst(vcpu
) == 0x44000022) &&
776 !(vcpu
->arch
.shared
->msr
& MSR_PR
)) {
777 /* SC 1 papr hypercalls */
778 ulong cmd
= kvmppc_get_gpr(vcpu
, 3);
781 #ifdef CONFIG_KVM_BOOK3S_64_PR
782 if (kvmppc_h_pr(vcpu
, cmd
) == EMULATE_DONE
) {
788 run
->papr_hcall
.nr
= cmd
;
789 for (i
= 0; i
< 9; ++i
) {
790 ulong gpr
= kvmppc_get_gpr(vcpu
, 4 + i
);
791 run
->papr_hcall
.args
[i
] = gpr
;
793 run
->exit_reason
= KVM_EXIT_PAPR_HCALL
;
794 vcpu
->arch
.hcall_needed
= 1;
796 } else if (vcpu
->arch
.osi_enabled
&&
797 (((u32
)kvmppc_get_gpr(vcpu
, 3)) == OSI_SC_MAGIC_R3
) &&
798 (((u32
)kvmppc_get_gpr(vcpu
, 4)) == OSI_SC_MAGIC_R4
)) {
800 u64
*gprs
= run
->osi
.gprs
;
803 run
->exit_reason
= KVM_EXIT_OSI
;
804 for (i
= 0; i
< 32; i
++)
805 gprs
[i
] = kvmppc_get_gpr(vcpu
, i
);
806 vcpu
->arch
.osi_needed
= 1;
808 } else if (!(vcpu
->arch
.shared
->msr
& MSR_PR
) &&
809 (((u32
)kvmppc_get_gpr(vcpu
, 0)) == KVM_SC_MAGIC_R0
)) {
810 /* KVM PV hypercalls */
811 kvmppc_set_gpr(vcpu
, 3, kvmppc_kvm_pv(vcpu
));
815 vcpu
->stat
.syscall_exits
++;
816 kvmppc_book3s_queue_irqprio(vcpu
, exit_nr
);
820 case BOOK3S_INTERRUPT_FP_UNAVAIL
:
821 case BOOK3S_INTERRUPT_ALTIVEC
:
822 case BOOK3S_INTERRUPT_VSX
:
827 case BOOK3S_INTERRUPT_FP_UNAVAIL
: ext_msr
= MSR_FP
; break;
828 case BOOK3S_INTERRUPT_ALTIVEC
: ext_msr
= MSR_VEC
; break;
829 case BOOK3S_INTERRUPT_VSX
: ext_msr
= MSR_VSX
; break;
832 switch (kvmppc_check_ext(vcpu
, exit_nr
)) {
834 /* everything ok - let's enable the ext */
835 r
= kvmppc_handle_ext(vcpu
, exit_nr
, ext_msr
);
838 /* we need to emulate this instruction */
839 goto program_interrupt
;
842 /* nothing to worry about - go again */
847 case BOOK3S_INTERRUPT_ALIGNMENT
:
848 if (kvmppc_read_inst(vcpu
) == EMULATE_DONE
) {
849 vcpu
->arch
.shared
->dsisr
= kvmppc_alignment_dsisr(vcpu
,
850 kvmppc_get_last_inst(vcpu
));
851 vcpu
->arch
.shared
->dar
= kvmppc_alignment_dar(vcpu
,
852 kvmppc_get_last_inst(vcpu
));
853 kvmppc_book3s_queue_irqprio(vcpu
, exit_nr
);
857 case BOOK3S_INTERRUPT_MACHINE_CHECK
:
858 case BOOK3S_INTERRUPT_TRACE
:
859 kvmppc_book3s_queue_irqprio(vcpu
, exit_nr
);
864 struct kvmppc_book3s_shadow_vcpu
*svcpu
= svcpu_get(vcpu
);
865 ulong shadow_srr1
= svcpu
->shadow_srr1
;
867 /* Ugh - bork here! What did we get? */
868 printk(KERN_EMERG
"exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
869 exit_nr
, kvmppc_get_pc(vcpu
), shadow_srr1
);
876 if (!(r
& RESUME_HOST
)) {
877 /* To avoid clobbering exit_reason, only check for signals if
878 * we aren't already exiting to userspace for some other
882 * Interrupts could be timers for the guest which we have to
883 * inject again, so let's postpone them until we're in the guest
884 * and if we really did time things so badly, then we just exit
885 * again due to a host external interrupt.
888 s
= kvmppc_prepare_to_enter(vcpu
);
893 kvmppc_lazy_ee_enable();
897 trace_kvm_book3s_reenter(r
, vcpu
);
902 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu
*vcpu
,
903 struct kvm_sregs
*sregs
)
905 struct kvmppc_vcpu_book3s
*vcpu3s
= to_book3s(vcpu
);
908 sregs
->pvr
= vcpu
->arch
.pvr
;
910 sregs
->u
.s
.sdr1
= to_book3s(vcpu
)->sdr1
;
911 if (vcpu
->arch
.hflags
& BOOK3S_HFLAG_SLB
) {
912 for (i
= 0; i
< 64; i
++) {
913 sregs
->u
.s
.ppc64
.slb
[i
].slbe
= vcpu
->arch
.slb
[i
].orige
| i
;
914 sregs
->u
.s
.ppc64
.slb
[i
].slbv
= vcpu
->arch
.slb
[i
].origv
;
917 for (i
= 0; i
< 16; i
++)
918 sregs
->u
.s
.ppc32
.sr
[i
] = vcpu
->arch
.shared
->sr
[i
];
920 for (i
= 0; i
< 8; i
++) {
921 sregs
->u
.s
.ppc32
.ibat
[i
] = vcpu3s
->ibat
[i
].raw
;
922 sregs
->u
.s
.ppc32
.dbat
[i
] = vcpu3s
->dbat
[i
].raw
;
929 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu
*vcpu
,
930 struct kvm_sregs
*sregs
)
932 struct kvmppc_vcpu_book3s
*vcpu3s
= to_book3s(vcpu
);
935 kvmppc_set_pvr(vcpu
, sregs
->pvr
);
937 vcpu3s
->sdr1
= sregs
->u
.s
.sdr1
;
938 if (vcpu
->arch
.hflags
& BOOK3S_HFLAG_SLB
) {
939 for (i
= 0; i
< 64; i
++) {
940 vcpu
->arch
.mmu
.slbmte(vcpu
, sregs
->u
.s
.ppc64
.slb
[i
].slbv
,
941 sregs
->u
.s
.ppc64
.slb
[i
].slbe
);
944 for (i
= 0; i
< 16; i
++) {
945 vcpu
->arch
.mmu
.mtsrin(vcpu
, i
, sregs
->u
.s
.ppc32
.sr
[i
]);
947 for (i
= 0; i
< 8; i
++) {
948 kvmppc_set_bat(vcpu
, &(vcpu3s
->ibat
[i
]), false,
949 (u32
)sregs
->u
.s
.ppc32
.ibat
[i
]);
950 kvmppc_set_bat(vcpu
, &(vcpu3s
->ibat
[i
]), true,
951 (u32
)(sregs
->u
.s
.ppc32
.ibat
[i
] >> 32));
952 kvmppc_set_bat(vcpu
, &(vcpu3s
->dbat
[i
]), false,
953 (u32
)sregs
->u
.s
.ppc32
.dbat
[i
]);
954 kvmppc_set_bat(vcpu
, &(vcpu3s
->dbat
[i
]), true,
955 (u32
)(sregs
->u
.s
.ppc32
.dbat
[i
] >> 32));
959 /* Flush the MMU after messing with the segments */
960 kvmppc_mmu_pte_flush(vcpu
, 0, 0);
965 int kvmppc_get_one_reg(struct kvm_vcpu
*vcpu
, u64 id
, union kvmppc_one_reg
*val
)
970 case KVM_REG_PPC_HIOR
:
971 *val
= get_reg_val(id
, to_book3s(vcpu
)->hior
);
974 case KVM_REG_PPC_VSR0
... KVM_REG_PPC_VSR31
: {
975 long int i
= id
- KVM_REG_PPC_VSR0
;
977 if (!cpu_has_feature(CPU_FTR_VSX
)) {
981 val
->vsxval
[0] = vcpu
->arch
.fpr
[i
];
982 val
->vsxval
[1] = vcpu
->arch
.vsr
[i
];
985 #endif /* CONFIG_VSX */
994 int kvmppc_set_one_reg(struct kvm_vcpu
*vcpu
, u64 id
, union kvmppc_one_reg
*val
)
999 case KVM_REG_PPC_HIOR
:
1000 to_book3s(vcpu
)->hior
= set_reg_val(id
, *val
);
1001 to_book3s(vcpu
)->hior_explicit
= true;
1004 case KVM_REG_PPC_VSR0
... KVM_REG_PPC_VSR31
: {
1005 long int i
= id
- KVM_REG_PPC_VSR0
;
1007 if (!cpu_has_feature(CPU_FTR_VSX
)) {
1011 vcpu
->arch
.fpr
[i
] = val
->vsxval
[0];
1012 vcpu
->arch
.vsr
[i
] = val
->vsxval
[1];
1015 #endif /* CONFIG_VSX */
1024 int kvmppc_core_check_processor_compat(void)
1029 struct kvm_vcpu
*kvmppc_core_vcpu_create(struct kvm
*kvm
, unsigned int id
)
1031 struct kvmppc_vcpu_book3s
*vcpu_book3s
;
1032 struct kvm_vcpu
*vcpu
;
1036 vcpu_book3s
= vzalloc(sizeof(struct kvmppc_vcpu_book3s
));
1040 vcpu_book3s
->shadow_vcpu
=
1041 kzalloc(sizeof(*vcpu_book3s
->shadow_vcpu
), GFP_KERNEL
);
1042 if (!vcpu_book3s
->shadow_vcpu
)
1045 vcpu
= &vcpu_book3s
->vcpu
;
1046 err
= kvm_vcpu_init(vcpu
, kvm
, id
);
1048 goto free_shadow_vcpu
;
1050 p
= __get_free_page(GFP_KERNEL
|__GFP_ZERO
);
1051 /* the real shared page fills the last 4k of our page */
1052 vcpu
->arch
.shared
= (void*)(p
+ PAGE_SIZE
- 4096);
1056 #ifdef CONFIG_PPC_BOOK3S_64
1057 /* default to book3s_64 (970fx) */
1058 vcpu
->arch
.pvr
= 0x3C0301;
1060 /* default to book3s_32 (750) */
1061 vcpu
->arch
.pvr
= 0x84202;
1063 kvmppc_set_pvr(vcpu
, vcpu
->arch
.pvr
);
1064 vcpu
->arch
.slb_nr
= 64;
1066 vcpu
->arch
.shadow_msr
= MSR_USER64
;
1068 err
= kvmppc_mmu_init(vcpu
);
1075 kvm_vcpu_uninit(vcpu
);
1077 kfree(vcpu_book3s
->shadow_vcpu
);
1081 return ERR_PTR(err
);
1084 void kvmppc_core_vcpu_free(struct kvm_vcpu
*vcpu
)
1086 struct kvmppc_vcpu_book3s
*vcpu_book3s
= to_book3s(vcpu
);
1088 free_page((unsigned long)vcpu
->arch
.shared
& PAGE_MASK
);
1089 kvm_vcpu_uninit(vcpu
);
1090 kfree(vcpu_book3s
->shadow_vcpu
);
1094 int kvmppc_vcpu_run(struct kvm_run
*kvm_run
, struct kvm_vcpu
*vcpu
)
1097 double fpr
[32][TS_FPRWIDTH
];
1100 #ifdef CONFIG_ALTIVEC
1103 unsigned long uninitialized_var(vrsave
);
1111 /* Check if we can run the vcpu at all */
1112 if (!vcpu
->arch
.sane
) {
1113 kvm_run
->exit_reason
= KVM_EXIT_INTERNAL_ERROR
;
1119 * Interrupts could be timers for the guest which we have to inject
1120 * again, so let's postpone them until we're in the guest and if we
1121 * really did time things so badly, then we just exit again due to
1122 * a host external interrupt.
1124 local_irq_disable();
1125 ret
= kvmppc_prepare_to_enter(vcpu
);
1131 /* Save FPU state in stack */
1132 if (current
->thread
.regs
->msr
& MSR_FP
)
1133 giveup_fpu(current
);
1134 memcpy(fpr
, current
->thread
.fpr
, sizeof(current
->thread
.fpr
));
1135 fpscr
= current
->thread
.fpscr
.val
;
1136 fpexc_mode
= current
->thread
.fpexc_mode
;
1138 #ifdef CONFIG_ALTIVEC
1139 /* Save Altivec state in stack */
1140 used_vr
= current
->thread
.used_vr
;
1142 if (current
->thread
.regs
->msr
& MSR_VEC
)
1143 giveup_altivec(current
);
1144 memcpy(vr
, current
->thread
.vr
, sizeof(current
->thread
.vr
));
1145 vscr
= current
->thread
.vscr
;
1146 vrsave
= current
->thread
.vrsave
;
1151 /* Save VSX state in stack */
1152 used_vsr
= current
->thread
.used_vsr
;
1153 if (used_vsr
&& (current
->thread
.regs
->msr
& MSR_VSX
))
1154 __giveup_vsx(current
);
1157 /* Remember the MSR with disabled extensions */
1158 ext_msr
= current
->thread
.regs
->msr
;
1160 /* Preload FPU if it's enabled */
1161 if (vcpu
->arch
.shared
->msr
& MSR_FP
)
1162 kvmppc_handle_ext(vcpu
, BOOK3S_INTERRUPT_FP_UNAVAIL
, MSR_FP
);
1164 kvmppc_lazy_ee_enable();
1166 ret
= __kvmppc_vcpu_run(kvm_run
, vcpu
);
1168 /* No need for kvm_guest_exit. It's done in handle_exit.
1169 We also get here with interrupts enabled. */
1171 /* Make sure we save the guest FPU/Altivec/VSX state */
1172 kvmppc_giveup_ext(vcpu
, MSR_FP
| MSR_VEC
| MSR_VSX
);
1174 current
->thread
.regs
->msr
= ext_msr
;
1176 /* Restore FPU/VSX state from stack */
1177 memcpy(current
->thread
.fpr
, fpr
, sizeof(current
->thread
.fpr
));
1178 current
->thread
.fpscr
.val
= fpscr
;
1179 current
->thread
.fpexc_mode
= fpexc_mode
;
1181 #ifdef CONFIG_ALTIVEC
1182 /* Restore Altivec state from stack */
1183 if (used_vr
&& current
->thread
.used_vr
) {
1184 memcpy(current
->thread
.vr
, vr
, sizeof(current
->thread
.vr
));
1185 current
->thread
.vscr
= vscr
;
1186 current
->thread
.vrsave
= vrsave
;
1188 current
->thread
.used_vr
= used_vr
;
1192 current
->thread
.used_vsr
= used_vsr
;
1196 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
1201 * Get (and clear) the dirty memory log for a memory slot.
1203 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
,
1204 struct kvm_dirty_log
*log
)
1206 struct kvm_memory_slot
*memslot
;
1207 struct kvm_vcpu
*vcpu
;
1213 mutex_lock(&kvm
->slots_lock
);
1215 r
= kvm_get_dirty_log(kvm
, log
, &is_dirty
);
1219 /* If nothing is dirty, don't bother messing with page tables. */
1221 memslot
= id_to_memslot(kvm
->memslots
, log
->slot
);
1223 ga
= memslot
->base_gfn
<< PAGE_SHIFT
;
1224 ga_end
= ga
+ (memslot
->npages
<< PAGE_SHIFT
);
1226 kvm_for_each_vcpu(n
, vcpu
, kvm
)
1227 kvmppc_mmu_pte_pflush(vcpu
, ga
, ga_end
);
1229 n
= kvm_dirty_bitmap_bytes(memslot
);
1230 memset(memslot
->dirty_bitmap
, 0, n
);
1235 mutex_unlock(&kvm
->slots_lock
);
1240 int kvm_vm_ioctl_get_smmu_info(struct kvm
*kvm
, struct kvm_ppc_smmu_info
*info
)
1242 info
->flags
= KVM_PPC_1T_SEGMENTS
;
1244 /* SLB is always 64 entries */
1245 info
->slb_size
= 64;
1247 /* Standard 4k base page size segment */
1248 info
->sps
[0].page_shift
= 12;
1249 info
->sps
[0].slb_enc
= 0;
1250 info
->sps
[0].enc
[0].page_shift
= 12;
1251 info
->sps
[0].enc
[0].pte_enc
= 0;
1253 /* Standard 16M large page size segment */
1254 info
->sps
[1].page_shift
= 24;
1255 info
->sps
[1].slb_enc
= SLB_VSID_L
;
1256 info
->sps
[1].enc
[0].page_shift
= 24;
1257 info
->sps
[1].enc
[0].pte_enc
= 0;
1261 #endif /* CONFIG_PPC64 */
1263 void kvmppc_core_free_memslot(struct kvm_memory_slot
*free
,
1264 struct kvm_memory_slot
*dont
)
1268 int kvmppc_core_create_memslot(struct kvm_memory_slot
*slot
,
1269 unsigned long npages
)
1274 int kvmppc_core_prepare_memory_region(struct kvm
*kvm
,
1275 struct kvm_memory_slot
*memslot
,
1276 struct kvm_userspace_memory_region
*mem
)
1281 void kvmppc_core_commit_memory_region(struct kvm
*kvm
,
1282 struct kvm_userspace_memory_region
*mem
,
1283 const struct kvm_memory_slot
*old
)
1287 void kvmppc_core_flush_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*memslot
)
1291 static unsigned int kvm_global_user_count
= 0;
1292 static DEFINE_SPINLOCK(kvm_global_user_count_lock
);
1294 int kvmppc_core_init_vm(struct kvm
*kvm
)
1297 INIT_LIST_HEAD(&kvm
->arch
.spapr_tce_tables
);
1298 INIT_LIST_HEAD(&kvm
->arch
.rtas_tokens
);
1301 if (firmware_has_feature(FW_FEATURE_SET_MODE
)) {
1302 spin_lock(&kvm_global_user_count_lock
);
1303 if (++kvm_global_user_count
== 1)
1304 pSeries_disable_reloc_on_exc();
1305 spin_unlock(&kvm_global_user_count_lock
);
1310 void kvmppc_core_destroy_vm(struct kvm
*kvm
)
1313 WARN_ON(!list_empty(&kvm
->arch
.spapr_tce_tables
));
1316 if (firmware_has_feature(FW_FEATURE_SET_MODE
)) {
1317 spin_lock(&kvm_global_user_count_lock
);
1318 BUG_ON(kvm_global_user_count
== 0);
1319 if (--kvm_global_user_count
== 0)
1320 pSeries_enable_reloc_on_exc();
1321 spin_unlock(&kvm_global_user_count_lock
);
1325 static int kvmppc_book3s_init(void)
1329 r
= kvm_init(NULL
, sizeof(struct kvmppc_vcpu_book3s
), 0,
1335 r
= kvmppc_mmu_hpte_sysinit();
1340 static void kvmppc_book3s_exit(void)
1342 kvmppc_mmu_hpte_sysexit();
1346 module_init(kvmppc_book3s_init
);
1347 module_exit(kvmppc_book3s_exit
);