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>
43 #include <linux/module.h>
44 #include <linux/miscdevice.h>
48 #define CREATE_TRACE_POINTS
51 /* #define EXIT_DEBUG */
52 /* #define DEBUG_EXT */
54 static int kvmppc_handle_ext(struct kvm_vcpu
*vcpu
, unsigned int exit_nr
,
56 static void kvmppc_giveup_fac(struct kvm_vcpu
*vcpu
, ulong fac
);
58 /* Some compatibility defines */
59 #ifdef CONFIG_PPC_BOOK3S_32
60 #define MSR_USER32 MSR_USER
61 #define MSR_USER64 MSR_USER
62 #define HW_PAGE_SIZE PAGE_SIZE
65 static bool kvmppc_is_split_real(struct kvm_vcpu
*vcpu
)
67 ulong msr
= kvmppc_get_msr(vcpu
);
68 return (msr
& (MSR_IR
|MSR_DR
)) == MSR_DR
;
71 static void kvmppc_fixup_split_real(struct kvm_vcpu
*vcpu
)
73 ulong msr
= kvmppc_get_msr(vcpu
);
74 ulong pc
= kvmppc_get_pc(vcpu
);
76 /* We are in DR only split real mode */
77 if ((msr
& (MSR_IR
|MSR_DR
)) != MSR_DR
)
80 /* We have not fixed up the guest already */
81 if (vcpu
->arch
.hflags
& BOOK3S_HFLAG_SPLIT_HACK
)
84 /* The code is in fixupable address space */
85 if (pc
& SPLIT_HACK_MASK
)
88 vcpu
->arch
.hflags
|= BOOK3S_HFLAG_SPLIT_HACK
;
89 kvmppc_set_pc(vcpu
, pc
| SPLIT_HACK_OFFS
);
92 void kvmppc_unfixup_split_real(struct kvm_vcpu
*vcpu
);
94 static void kvmppc_core_vcpu_load_pr(struct kvm_vcpu
*vcpu
, int cpu
)
96 #ifdef CONFIG_PPC_BOOK3S_64
97 struct kvmppc_book3s_shadow_vcpu
*svcpu
= svcpu_get(vcpu
);
98 memcpy(svcpu
->slb
, to_book3s(vcpu
)->slb_shadow
, sizeof(svcpu
->slb
));
99 svcpu
->slb_max
= to_book3s(vcpu
)->slb_shadow_max
;
104 /* Disable AIL if supported */
105 if (cpu_has_feature(CPU_FTR_HVMODE
) &&
106 cpu_has_feature(CPU_FTR_ARCH_207S
))
107 mtspr(SPRN_LPCR
, mfspr(SPRN_LPCR
) & ~LPCR_AIL
);
109 vcpu
->cpu
= smp_processor_id();
110 #ifdef CONFIG_PPC_BOOK3S_32
111 current
->thread
.kvm_shadow_vcpu
= vcpu
->arch
.shadow_vcpu
;
114 if (kvmppc_is_split_real(vcpu
))
115 kvmppc_fixup_split_real(vcpu
);
118 static void kvmppc_core_vcpu_put_pr(struct kvm_vcpu
*vcpu
)
120 #ifdef CONFIG_PPC_BOOK3S_64
121 struct kvmppc_book3s_shadow_vcpu
*svcpu
= svcpu_get(vcpu
);
123 kvmppc_copy_from_svcpu(vcpu
, svcpu
);
125 memcpy(to_book3s(vcpu
)->slb_shadow
, svcpu
->slb
, sizeof(svcpu
->slb
));
126 to_book3s(vcpu
)->slb_shadow_max
= svcpu
->slb_max
;
130 if (kvmppc_is_split_real(vcpu
))
131 kvmppc_unfixup_split_real(vcpu
);
133 kvmppc_giveup_ext(vcpu
, MSR_FP
| MSR_VEC
| MSR_VSX
);
134 kvmppc_giveup_fac(vcpu
, FSCR_TAR_LG
);
136 /* Enable AIL if supported */
137 if (cpu_has_feature(CPU_FTR_HVMODE
) &&
138 cpu_has_feature(CPU_FTR_ARCH_207S
))
139 mtspr(SPRN_LPCR
, mfspr(SPRN_LPCR
) | LPCR_AIL_3
);
144 /* Copy data needed by real-mode code from vcpu to shadow vcpu */
145 void kvmppc_copy_to_svcpu(struct kvmppc_book3s_shadow_vcpu
*svcpu
,
146 struct kvm_vcpu
*vcpu
)
148 svcpu
->gpr
[0] = vcpu
->arch
.gpr
[0];
149 svcpu
->gpr
[1] = vcpu
->arch
.gpr
[1];
150 svcpu
->gpr
[2] = vcpu
->arch
.gpr
[2];
151 svcpu
->gpr
[3] = vcpu
->arch
.gpr
[3];
152 svcpu
->gpr
[4] = vcpu
->arch
.gpr
[4];
153 svcpu
->gpr
[5] = vcpu
->arch
.gpr
[5];
154 svcpu
->gpr
[6] = vcpu
->arch
.gpr
[6];
155 svcpu
->gpr
[7] = vcpu
->arch
.gpr
[7];
156 svcpu
->gpr
[8] = vcpu
->arch
.gpr
[8];
157 svcpu
->gpr
[9] = vcpu
->arch
.gpr
[9];
158 svcpu
->gpr
[10] = vcpu
->arch
.gpr
[10];
159 svcpu
->gpr
[11] = vcpu
->arch
.gpr
[11];
160 svcpu
->gpr
[12] = vcpu
->arch
.gpr
[12];
161 svcpu
->gpr
[13] = vcpu
->arch
.gpr
[13];
162 svcpu
->cr
= vcpu
->arch
.cr
;
163 svcpu
->xer
= vcpu
->arch
.xer
;
164 svcpu
->ctr
= vcpu
->arch
.ctr
;
165 svcpu
->lr
= vcpu
->arch
.lr
;
166 svcpu
->pc
= vcpu
->arch
.pc
;
167 #ifdef CONFIG_PPC_BOOK3S_64
168 svcpu
->shadow_fscr
= vcpu
->arch
.shadow_fscr
;
171 * Now also save the current time base value. We use this
172 * to find the guest purr and spurr value.
174 vcpu
->arch
.entry_tb
= get_tb();
175 vcpu
->arch
.entry_vtb
= get_vtb();
176 if (cpu_has_feature(CPU_FTR_ARCH_207S
))
177 vcpu
->arch
.entry_ic
= mfspr(SPRN_IC
);
178 svcpu
->in_use
= true;
181 /* Copy data touched by real-mode code from shadow vcpu back to vcpu */
182 void kvmppc_copy_from_svcpu(struct kvm_vcpu
*vcpu
,
183 struct kvmppc_book3s_shadow_vcpu
*svcpu
)
186 * vcpu_put would just call us again because in_use hasn't
192 * Maybe we were already preempted and synced the svcpu from
193 * our preempt notifiers. Don't bother touching this svcpu then.
198 vcpu
->arch
.gpr
[0] = svcpu
->gpr
[0];
199 vcpu
->arch
.gpr
[1] = svcpu
->gpr
[1];
200 vcpu
->arch
.gpr
[2] = svcpu
->gpr
[2];
201 vcpu
->arch
.gpr
[3] = svcpu
->gpr
[3];
202 vcpu
->arch
.gpr
[4] = svcpu
->gpr
[4];
203 vcpu
->arch
.gpr
[5] = svcpu
->gpr
[5];
204 vcpu
->arch
.gpr
[6] = svcpu
->gpr
[6];
205 vcpu
->arch
.gpr
[7] = svcpu
->gpr
[7];
206 vcpu
->arch
.gpr
[8] = svcpu
->gpr
[8];
207 vcpu
->arch
.gpr
[9] = svcpu
->gpr
[9];
208 vcpu
->arch
.gpr
[10] = svcpu
->gpr
[10];
209 vcpu
->arch
.gpr
[11] = svcpu
->gpr
[11];
210 vcpu
->arch
.gpr
[12] = svcpu
->gpr
[12];
211 vcpu
->arch
.gpr
[13] = svcpu
->gpr
[13];
212 vcpu
->arch
.cr
= svcpu
->cr
;
213 vcpu
->arch
.xer
= svcpu
->xer
;
214 vcpu
->arch
.ctr
= svcpu
->ctr
;
215 vcpu
->arch
.lr
= svcpu
->lr
;
216 vcpu
->arch
.pc
= svcpu
->pc
;
217 vcpu
->arch
.shadow_srr1
= svcpu
->shadow_srr1
;
218 vcpu
->arch
.fault_dar
= svcpu
->fault_dar
;
219 vcpu
->arch
.fault_dsisr
= svcpu
->fault_dsisr
;
220 vcpu
->arch
.last_inst
= svcpu
->last_inst
;
221 #ifdef CONFIG_PPC_BOOK3S_64
222 vcpu
->arch
.shadow_fscr
= svcpu
->shadow_fscr
;
225 * Update purr and spurr using time base on exit.
227 vcpu
->arch
.purr
+= get_tb() - vcpu
->arch
.entry_tb
;
228 vcpu
->arch
.spurr
+= get_tb() - vcpu
->arch
.entry_tb
;
229 vcpu
->arch
.vtb
+= get_vtb() - vcpu
->arch
.entry_vtb
;
230 if (cpu_has_feature(CPU_FTR_ARCH_207S
))
231 vcpu
->arch
.ic
+= mfspr(SPRN_IC
) - vcpu
->arch
.entry_ic
;
232 svcpu
->in_use
= false;
238 static int kvmppc_core_check_requests_pr(struct kvm_vcpu
*vcpu
)
240 int r
= 1; /* Indicate we want to get back into the guest */
242 /* We misuse TLB_FLUSH to indicate that we want to clear
243 all shadow cache entries */
244 if (kvm_check_request(KVM_REQ_TLB_FLUSH
, vcpu
))
245 kvmppc_mmu_pte_flush(vcpu
, 0, 0);
250 /************* MMU Notifiers *************/
251 static void do_kvm_unmap_hva(struct kvm
*kvm
, unsigned long start
,
255 struct kvm_vcpu
*vcpu
;
256 struct kvm_memslots
*slots
;
257 struct kvm_memory_slot
*memslot
;
259 slots
= kvm_memslots(kvm
);
260 kvm_for_each_memslot(memslot
, slots
) {
261 unsigned long hva_start
, hva_end
;
264 hva_start
= max(start
, memslot
->userspace_addr
);
265 hva_end
= min(end
, memslot
->userspace_addr
+
266 (memslot
->npages
<< PAGE_SHIFT
));
267 if (hva_start
>= hva_end
)
270 * {gfn(page) | page intersects with [hva_start, hva_end)} =
271 * {gfn, gfn+1, ..., gfn_end-1}.
273 gfn
= hva_to_gfn_memslot(hva_start
, memslot
);
274 gfn_end
= hva_to_gfn_memslot(hva_end
+ PAGE_SIZE
- 1, memslot
);
275 kvm_for_each_vcpu(i
, vcpu
, kvm
)
276 kvmppc_mmu_pte_pflush(vcpu
, gfn
<< PAGE_SHIFT
,
277 gfn_end
<< PAGE_SHIFT
);
281 static int kvm_unmap_hva_pr(struct kvm
*kvm
, unsigned long hva
)
283 trace_kvm_unmap_hva(hva
);
285 do_kvm_unmap_hva(kvm
, hva
, hva
+ PAGE_SIZE
);
290 static int kvm_unmap_hva_range_pr(struct kvm
*kvm
, unsigned long start
,
293 do_kvm_unmap_hva(kvm
, start
, end
);
298 static int kvm_age_hva_pr(struct kvm
*kvm
, unsigned long start
,
301 /* XXX could be more clever ;) */
305 static int kvm_test_age_hva_pr(struct kvm
*kvm
, unsigned long hva
)
307 /* XXX could be more clever ;) */
311 static void kvm_set_spte_hva_pr(struct kvm
*kvm
, unsigned long hva
, pte_t pte
)
313 /* The page will get remapped properly on its next fault */
314 do_kvm_unmap_hva(kvm
, hva
, hva
+ PAGE_SIZE
);
317 /*****************************************/
319 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu
*vcpu
)
321 ulong guest_msr
= kvmppc_get_msr(vcpu
);
322 ulong smsr
= guest_msr
;
324 /* Guest MSR values */
325 smsr
&= MSR_FE0
| MSR_FE1
| MSR_SF
| MSR_SE
| MSR_BE
| MSR_LE
;
326 /* Process MSR values */
327 smsr
|= MSR_ME
| MSR_RI
| MSR_IR
| MSR_DR
| MSR_PR
| MSR_EE
;
328 /* External providers the guest reserved */
329 smsr
|= (guest_msr
& vcpu
->arch
.guest_owned_ext
);
330 /* 64-bit Process MSR values */
331 #ifdef CONFIG_PPC_BOOK3S_64
332 smsr
|= MSR_ISF
| MSR_HV
;
334 vcpu
->arch
.shadow_msr
= smsr
;
337 static void kvmppc_set_msr_pr(struct kvm_vcpu
*vcpu
, u64 msr
)
339 ulong old_msr
= kvmppc_get_msr(vcpu
);
342 printk(KERN_INFO
"KVM: Set MSR to 0x%llx\n", msr
);
345 msr
&= to_book3s(vcpu
)->msr_mask
;
346 kvmppc_set_msr_fast(vcpu
, msr
);
347 kvmppc_recalc_shadow_msr(vcpu
);
350 if (!vcpu
->arch
.pending_exceptions
) {
351 kvm_vcpu_block(vcpu
);
352 clear_bit(KVM_REQ_UNHALT
, &vcpu
->requests
);
353 vcpu
->stat
.halt_wakeup
++;
355 /* Unset POW bit after we woke up */
357 kvmppc_set_msr_fast(vcpu
, msr
);
361 if (kvmppc_is_split_real(vcpu
))
362 kvmppc_fixup_split_real(vcpu
);
364 kvmppc_unfixup_split_real(vcpu
);
366 if ((kvmppc_get_msr(vcpu
) & (MSR_PR
|MSR_IR
|MSR_DR
)) !=
367 (old_msr
& (MSR_PR
|MSR_IR
|MSR_DR
))) {
368 kvmppc_mmu_flush_segments(vcpu
);
369 kvmppc_mmu_map_segment(vcpu
, kvmppc_get_pc(vcpu
));
371 /* Preload magic page segment when in kernel mode */
372 if (!(msr
& MSR_PR
) && vcpu
->arch
.magic_page_pa
) {
373 struct kvm_vcpu_arch
*a
= &vcpu
->arch
;
376 kvmppc_mmu_map_segment(vcpu
, a
->magic_page_ea
);
378 kvmppc_mmu_map_segment(vcpu
, a
->magic_page_pa
);
383 * When switching from 32 to 64-bit, we may have a stale 32-bit
384 * magic page around, we need to flush it. Typically 32-bit magic
385 * page will be instanciated when calling into RTAS. Note: We
386 * assume that such transition only happens while in kernel mode,
387 * ie, we never transition from user 32-bit to kernel 64-bit with
388 * a 32-bit magic page around.
390 if (vcpu
->arch
.magic_page_pa
&&
391 !(old_msr
& MSR_PR
) && !(old_msr
& MSR_SF
) && (msr
& MSR_SF
)) {
392 /* going from RTAS to normal kernel code */
393 kvmppc_mmu_pte_flush(vcpu
, (uint32_t)vcpu
->arch
.magic_page_pa
,
397 /* Preload FPU if it's enabled */
398 if (kvmppc_get_msr(vcpu
) & MSR_FP
)
399 kvmppc_handle_ext(vcpu
, BOOK3S_INTERRUPT_FP_UNAVAIL
, MSR_FP
);
402 void kvmppc_set_pvr_pr(struct kvm_vcpu
*vcpu
, u32 pvr
)
406 vcpu
->arch
.hflags
&= ~BOOK3S_HFLAG_SLB
;
407 vcpu
->arch
.pvr
= pvr
;
408 #ifdef CONFIG_PPC_BOOK3S_64
409 if ((pvr
>= 0x330000) && (pvr
< 0x70330000)) {
410 kvmppc_mmu_book3s_64_init(vcpu
);
411 if (!to_book3s(vcpu
)->hior_explicit
)
412 to_book3s(vcpu
)->hior
= 0xfff00000;
413 to_book3s(vcpu
)->msr_mask
= 0xffffffffffffffffULL
;
414 vcpu
->arch
.cpu_type
= KVM_CPU_3S_64
;
418 kvmppc_mmu_book3s_32_init(vcpu
);
419 if (!to_book3s(vcpu
)->hior_explicit
)
420 to_book3s(vcpu
)->hior
= 0;
421 to_book3s(vcpu
)->msr_mask
= 0xffffffffULL
;
422 vcpu
->arch
.cpu_type
= KVM_CPU_3S_32
;
425 kvmppc_sanity_check(vcpu
);
427 /* If we are in hypervisor level on 970, we can tell the CPU to
428 * treat DCBZ as 32 bytes store */
429 vcpu
->arch
.hflags
&= ~BOOK3S_HFLAG_DCBZ32
;
430 if (vcpu
->arch
.mmu
.is_dcbz32(vcpu
) && (mfmsr() & MSR_HV
) &&
431 !strcmp(cur_cpu_spec
->platform
, "ppc970"))
432 vcpu
->arch
.hflags
|= BOOK3S_HFLAG_DCBZ32
;
434 /* Cell performs badly if MSR_FEx are set. So let's hope nobody
435 really needs them in a VM on Cell and force disable them. */
436 if (!strcmp(cur_cpu_spec
->platform
, "ppc-cell-be"))
437 to_book3s(vcpu
)->msr_mask
&= ~(MSR_FE0
| MSR_FE1
);
440 * If they're asking for POWER6 or later, set the flag
441 * indicating that we can do multiple large page sizes
443 * Also set the flag that indicates that tlbie has the large
444 * page bit in the RB operand instead of the instruction.
446 switch (PVR_VER(pvr
)) {
451 vcpu
->arch
.hflags
|= BOOK3S_HFLAG_MULTI_PGSIZE
|
452 BOOK3S_HFLAG_NEW_TLBIE
;
456 #ifdef CONFIG_PPC_BOOK3S_32
457 /* 32 bit Book3S always has 32 byte dcbz */
458 vcpu
->arch
.hflags
|= BOOK3S_HFLAG_DCBZ32
;
461 /* On some CPUs we can execute paired single operations natively */
462 asm ( "mfpvr %0" : "=r"(host_pvr
));
464 case 0x00080200: /* lonestar 2.0 */
465 case 0x00088202: /* lonestar 2.2 */
466 case 0x70000100: /* gekko 1.0 */
467 case 0x00080100: /* gekko 2.0 */
468 case 0x00083203: /* gekko 2.3a */
469 case 0x00083213: /* gekko 2.3b */
470 case 0x00083204: /* gekko 2.4 */
471 case 0x00083214: /* gekko 2.4e (8SE) - retail HW2 */
472 case 0x00087200: /* broadway */
473 vcpu
->arch
.hflags
|= BOOK3S_HFLAG_NATIVE_PS
;
474 /* Enable HID2.PSE - in case we need it later */
475 mtspr(SPRN_HID2_GEKKO
, mfspr(SPRN_HID2_GEKKO
) | (1 << 29));
479 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
480 * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
481 * emulate 32 bytes dcbz length.
483 * The Book3s_64 inventors also realized this case and implemented a special bit
484 * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
486 * My approach here is to patch the dcbz instruction on executing pages.
488 static void kvmppc_patch_dcbz(struct kvm_vcpu
*vcpu
, struct kvmppc_pte
*pte
)
495 hpage
= gfn_to_page(vcpu
->kvm
, pte
->raddr
>> PAGE_SHIFT
);
496 if (is_error_page(hpage
))
499 hpage_offset
= pte
->raddr
& ~PAGE_MASK
;
500 hpage_offset
&= ~0xFFFULL
;
504 page
= kmap_atomic(hpage
);
506 /* patch dcbz into reserved instruction, so we trap */
507 for (i
=hpage_offset
; i
< hpage_offset
+ (HW_PAGE_SIZE
/ 4); i
++)
508 if ((be32_to_cpu(page
[i
]) & 0xff0007ff) == INS_DCBZ
)
509 page
[i
] &= cpu_to_be32(0xfffffff7);
515 static int kvmppc_visible_gpa(struct kvm_vcpu
*vcpu
, gpa_t gpa
)
517 ulong mp_pa
= vcpu
->arch
.magic_page_pa
;
519 if (!(kvmppc_get_msr(vcpu
) & MSR_SF
))
520 mp_pa
= (uint32_t)mp_pa
;
523 if (unlikely(mp_pa
) && unlikely((mp_pa
& KVM_PAM
) == (gpa
& KVM_PAM
))) {
527 return kvm_is_visible_gfn(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
530 int kvmppc_handle_pagefault(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
,
531 ulong eaddr
, int vec
)
533 bool data
= (vec
== BOOK3S_INTERRUPT_DATA_STORAGE
);
534 bool iswrite
= false;
535 int r
= RESUME_GUEST
;
538 struct kvmppc_pte pte
;
539 bool is_mmio
= false;
540 bool dr
= (kvmppc_get_msr(vcpu
) & MSR_DR
) ? true : false;
541 bool ir
= (kvmppc_get_msr(vcpu
) & MSR_IR
) ? true : false;
544 relocated
= data
? dr
: ir
;
545 if (data
&& (vcpu
->arch
.fault_dsisr
& DSISR_ISSTORE
))
548 /* Resolve real address if translation turned on */
550 page_found
= vcpu
->arch
.mmu
.xlate(vcpu
, eaddr
, &pte
, data
, iswrite
);
552 pte
.may_execute
= true;
554 pte
.may_write
= true;
555 pte
.raddr
= eaddr
& KVM_PAM
;
557 pte
.vpage
= eaddr
>> 12;
558 pte
.page_size
= MMU_PAGE_64K
;
561 switch (kvmppc_get_msr(vcpu
) & (MSR_DR
|MSR_IR
)) {
563 pte
.vpage
|= ((u64
)VSID_REAL
<< (SID_SHIFT
- 12));
567 (vcpu
->arch
.hflags
& BOOK3S_HFLAG_SPLIT_HACK
) &&
568 ((pte
.raddr
& SPLIT_HACK_MASK
) == SPLIT_HACK_OFFS
))
569 pte
.raddr
&= ~SPLIT_HACK_MASK
;
572 vcpu
->arch
.mmu
.esid_to_vsid(vcpu
, eaddr
>> SID_SHIFT
, &vsid
);
574 if ((kvmppc_get_msr(vcpu
) & (MSR_DR
|MSR_IR
)) == MSR_DR
)
575 pte
.vpage
|= ((u64
)VSID_REAL_DR
<< (SID_SHIFT
- 12));
577 pte
.vpage
|= ((u64
)VSID_REAL_IR
<< (SID_SHIFT
- 12));
581 page_found
= -EINVAL
;
585 if (vcpu
->arch
.mmu
.is_dcbz32(vcpu
) &&
586 (!(vcpu
->arch
.hflags
& BOOK3S_HFLAG_DCBZ32
))) {
588 * If we do the dcbz hack, we have to NX on every execution,
589 * so we can patch the executing code. This renders our guest
592 pte
.may_execute
= !data
;
595 if (page_found
== -ENOENT
) {
596 /* Page not found in guest PTE entries */
597 u64 ssrr1
= vcpu
->arch
.shadow_srr1
;
598 u64 msr
= kvmppc_get_msr(vcpu
);
599 kvmppc_set_dar(vcpu
, kvmppc_get_fault_dar(vcpu
));
600 kvmppc_set_dsisr(vcpu
, vcpu
->arch
.fault_dsisr
);
601 kvmppc_set_msr_fast(vcpu
, msr
| (ssrr1
& 0xf8000000ULL
));
602 kvmppc_book3s_queue_irqprio(vcpu
, vec
);
603 } else if (page_found
== -EPERM
) {
604 /* Storage protection */
605 u32 dsisr
= vcpu
->arch
.fault_dsisr
;
606 u64 ssrr1
= vcpu
->arch
.shadow_srr1
;
607 u64 msr
= kvmppc_get_msr(vcpu
);
608 kvmppc_set_dar(vcpu
, kvmppc_get_fault_dar(vcpu
));
609 dsisr
= (dsisr
& ~DSISR_NOHPTE
) | DSISR_PROTFAULT
;
610 kvmppc_set_dsisr(vcpu
, dsisr
);
611 kvmppc_set_msr_fast(vcpu
, msr
| (ssrr1
& 0xf8000000ULL
));
612 kvmppc_book3s_queue_irqprio(vcpu
, vec
);
613 } else if (page_found
== -EINVAL
) {
614 /* Page not found in guest SLB */
615 kvmppc_set_dar(vcpu
, kvmppc_get_fault_dar(vcpu
));
616 kvmppc_book3s_queue_irqprio(vcpu
, vec
+ 0x80);
617 } else if (!is_mmio
&&
618 kvmppc_visible_gpa(vcpu
, pte
.raddr
)) {
619 if (data
&& !(vcpu
->arch
.fault_dsisr
& DSISR_NOHPTE
)) {
621 * There is already a host HPTE there, presumably
622 * a read-only one for a page the guest thinks
623 * is writable, so get rid of it first.
625 kvmppc_mmu_unmap_page(vcpu
, &pte
);
627 /* The guest's PTE is not mapped yet. Map on the host */
628 kvmppc_mmu_map_page(vcpu
, &pte
, iswrite
);
630 vcpu
->stat
.sp_storage
++;
631 else if (vcpu
->arch
.mmu
.is_dcbz32(vcpu
) &&
632 (!(vcpu
->arch
.hflags
& BOOK3S_HFLAG_DCBZ32
)))
633 kvmppc_patch_dcbz(vcpu
, &pte
);
636 vcpu
->stat
.mmio_exits
++;
637 vcpu
->arch
.paddr_accessed
= pte
.raddr
;
638 vcpu
->arch
.vaddr_accessed
= pte
.eaddr
;
639 r
= kvmppc_emulate_mmio(run
, vcpu
);
640 if ( r
== RESUME_HOST_NV
)
647 /* Give up external provider (FPU, Altivec, VSX) */
648 void kvmppc_giveup_ext(struct kvm_vcpu
*vcpu
, ulong msr
)
650 struct thread_struct
*t
= ¤t
->thread
;
653 * VSX instructions can access FP and vector registers, so if
654 * we are giving up VSX, make sure we give up FP and VMX as well.
657 msr
|= MSR_FP
| MSR_VEC
;
659 msr
&= vcpu
->arch
.guest_owned_ext
;
664 printk(KERN_INFO
"Giving up ext 0x%lx\n", msr
);
669 * Note that on CPUs with VSX, giveup_fpu stores
670 * both the traditional FP registers and the added VSX
671 * registers into thread.fp_state.fpr[].
673 if (t
->regs
->msr
& MSR_FP
)
675 t
->fp_save_area
= NULL
;
678 #ifdef CONFIG_ALTIVEC
680 if (current
->thread
.regs
->msr
& MSR_VEC
)
681 giveup_altivec(current
);
682 t
->vr_save_area
= NULL
;
686 vcpu
->arch
.guest_owned_ext
&= ~(msr
| MSR_VSX
);
687 kvmppc_recalc_shadow_msr(vcpu
);
690 /* Give up facility (TAR / EBB / DSCR) */
691 static void kvmppc_giveup_fac(struct kvm_vcpu
*vcpu
, ulong fac
)
693 #ifdef CONFIG_PPC_BOOK3S_64
694 if (!(vcpu
->arch
.shadow_fscr
& (1ULL << fac
))) {
695 /* Facility not available to the guest, ignore giveup request*/
701 vcpu
->arch
.tar
= mfspr(SPRN_TAR
);
702 mtspr(SPRN_TAR
, current
->thread
.tar
);
703 vcpu
->arch
.shadow_fscr
&= ~FSCR_TAR
;
709 /* Handle external providers (FPU, Altivec, VSX) */
710 static int kvmppc_handle_ext(struct kvm_vcpu
*vcpu
, unsigned int exit_nr
,
713 struct thread_struct
*t
= ¤t
->thread
;
715 /* When we have paired singles, we emulate in software */
716 if (vcpu
->arch
.hflags
& BOOK3S_HFLAG_PAIRED_SINGLE
)
719 if (!(kvmppc_get_msr(vcpu
) & msr
)) {
720 kvmppc_book3s_queue_irqprio(vcpu
, exit_nr
);
724 if (msr
== MSR_VSX
) {
725 /* No VSX? Give an illegal instruction interrupt */
727 if (!cpu_has_feature(CPU_FTR_VSX
))
730 kvmppc_core_queue_program(vcpu
, SRR1_PROGILL
);
735 * We have to load up all the FP and VMX registers before
736 * we can let the guest use VSX instructions.
738 msr
= MSR_FP
| MSR_VEC
| MSR_VSX
;
741 /* See if we already own all the ext(s) needed */
742 msr
&= ~vcpu
->arch
.guest_owned_ext
;
747 printk(KERN_INFO
"Loading up ext 0x%lx\n", msr
);
753 load_fp_state(&vcpu
->arch
.fp
);
754 t
->fp_save_area
= &vcpu
->arch
.fp
;
759 #ifdef CONFIG_ALTIVEC
761 enable_kernel_altivec();
762 load_vr_state(&vcpu
->arch
.vr
);
763 t
->vr_save_area
= &vcpu
->arch
.vr
;
769 vcpu
->arch
.guest_owned_ext
|= msr
;
770 kvmppc_recalc_shadow_msr(vcpu
);
776 * Kernel code using FP or VMX could have flushed guest state to
777 * the thread_struct; if so, get it back now.
779 static void kvmppc_handle_lost_ext(struct kvm_vcpu
*vcpu
)
781 unsigned long lost_ext
;
783 lost_ext
= vcpu
->arch
.guest_owned_ext
& ~current
->thread
.regs
->msr
;
787 if (lost_ext
& MSR_FP
) {
790 load_fp_state(&vcpu
->arch
.fp
);
793 #ifdef CONFIG_ALTIVEC
794 if (lost_ext
& MSR_VEC
) {
796 enable_kernel_altivec();
797 load_vr_state(&vcpu
->arch
.vr
);
801 current
->thread
.regs
->msr
|= lost_ext
;
804 #ifdef CONFIG_PPC_BOOK3S_64
806 static void kvmppc_trigger_fac_interrupt(struct kvm_vcpu
*vcpu
, ulong fac
)
808 /* Inject the Interrupt Cause field and trigger a guest interrupt */
809 vcpu
->arch
.fscr
&= ~(0xffULL
<< 56);
810 vcpu
->arch
.fscr
|= (fac
<< 56);
811 kvmppc_book3s_queue_irqprio(vcpu
, BOOK3S_INTERRUPT_FAC_UNAVAIL
);
814 static void kvmppc_emulate_fac(struct kvm_vcpu
*vcpu
, ulong fac
)
816 enum emulation_result er
= EMULATE_FAIL
;
818 if (!(kvmppc_get_msr(vcpu
) & MSR_PR
))
819 er
= kvmppc_emulate_instruction(vcpu
->run
, vcpu
);
821 if ((er
!= EMULATE_DONE
) && (er
!= EMULATE_AGAIN
)) {
822 /* Couldn't emulate, trigger interrupt in guest */
823 kvmppc_trigger_fac_interrupt(vcpu
, fac
);
827 /* Enable facilities (TAR, EBB, DSCR) for the guest */
828 static int kvmppc_handle_fac(struct kvm_vcpu
*vcpu
, ulong fac
)
830 bool guest_fac_enabled
;
831 BUG_ON(!cpu_has_feature(CPU_FTR_ARCH_207S
));
834 * Not every facility is enabled by FSCR bits, check whether the
835 * guest has this facility enabled at all.
840 guest_fac_enabled
= (vcpu
->arch
.fscr
& (1ULL << fac
));
843 guest_fac_enabled
= kvmppc_get_msr(vcpu
) & MSR_TM
;
846 guest_fac_enabled
= false;
850 if (!guest_fac_enabled
) {
851 /* Facility not enabled by the guest */
852 kvmppc_trigger_fac_interrupt(vcpu
, fac
);
858 /* TAR switching isn't lazy in Linux yet */
859 current
->thread
.tar
= mfspr(SPRN_TAR
);
860 mtspr(SPRN_TAR
, vcpu
->arch
.tar
);
861 vcpu
->arch
.shadow_fscr
|= FSCR_TAR
;
864 kvmppc_emulate_fac(vcpu
, fac
);
871 void kvmppc_set_fscr(struct kvm_vcpu
*vcpu
, u64 fscr
)
873 if ((vcpu
->arch
.fscr
& FSCR_TAR
) && !(fscr
& FSCR_TAR
)) {
874 /* TAR got dropped, drop it in shadow too */
875 kvmppc_giveup_fac(vcpu
, FSCR_TAR_LG
);
877 vcpu
->arch
.fscr
= fscr
;
881 int kvmppc_handle_exit_pr(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
,
882 unsigned int exit_nr
)
887 vcpu
->stat
.sum_exits
++;
889 run
->exit_reason
= KVM_EXIT_UNKNOWN
;
890 run
->ready_for_interrupt_injection
= 1;
892 /* We get here with MSR.EE=1 */
894 trace_kvm_exit(exit_nr
, vcpu
);
898 case BOOK3S_INTERRUPT_INST_STORAGE
:
900 ulong shadow_srr1
= vcpu
->arch
.shadow_srr1
;
901 vcpu
->stat
.pf_instruc
++;
903 if (kvmppc_is_split_real(vcpu
))
904 kvmppc_fixup_split_real(vcpu
);
906 #ifdef CONFIG_PPC_BOOK3S_32
907 /* We set segments as unused segments when invalidating them. So
908 * treat the respective fault as segment fault. */
910 struct kvmppc_book3s_shadow_vcpu
*svcpu
;
913 svcpu
= svcpu_get(vcpu
);
914 sr
= svcpu
->sr
[kvmppc_get_pc(vcpu
) >> SID_SHIFT
];
916 if (sr
== SR_INVALID
) {
917 kvmppc_mmu_map_segment(vcpu
, kvmppc_get_pc(vcpu
));
924 /* only care about PTEG not found errors, but leave NX alone */
925 if (shadow_srr1
& 0x40000000) {
926 int idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
927 r
= kvmppc_handle_pagefault(run
, vcpu
, kvmppc_get_pc(vcpu
), exit_nr
);
928 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
929 vcpu
->stat
.sp_instruc
++;
930 } else if (vcpu
->arch
.mmu
.is_dcbz32(vcpu
) &&
931 (!(vcpu
->arch
.hflags
& BOOK3S_HFLAG_DCBZ32
))) {
933 * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
934 * so we can't use the NX bit inside the guest. Let's cross our fingers,
935 * that no guest that needs the dcbz hack does NX.
937 kvmppc_mmu_pte_flush(vcpu
, kvmppc_get_pc(vcpu
), ~0xFFFUL
);
940 u64 msr
= kvmppc_get_msr(vcpu
);
941 msr
|= shadow_srr1
& 0x58000000;
942 kvmppc_set_msr_fast(vcpu
, msr
);
943 kvmppc_book3s_queue_irqprio(vcpu
, exit_nr
);
948 case BOOK3S_INTERRUPT_DATA_STORAGE
:
950 ulong dar
= kvmppc_get_fault_dar(vcpu
);
951 u32 fault_dsisr
= vcpu
->arch
.fault_dsisr
;
952 vcpu
->stat
.pf_storage
++;
954 #ifdef CONFIG_PPC_BOOK3S_32
955 /* We set segments as unused segments when invalidating them. So
956 * treat the respective fault as segment fault. */
958 struct kvmppc_book3s_shadow_vcpu
*svcpu
;
961 svcpu
= svcpu_get(vcpu
);
962 sr
= svcpu
->sr
[dar
>> SID_SHIFT
];
964 if (sr
== SR_INVALID
) {
965 kvmppc_mmu_map_segment(vcpu
, dar
);
973 * We need to handle missing shadow PTEs, and
974 * protection faults due to us mapping a page read-only
975 * when the guest thinks it is writable.
977 if (fault_dsisr
& (DSISR_NOHPTE
| DSISR_PROTFAULT
)) {
978 int idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
979 r
= kvmppc_handle_pagefault(run
, vcpu
, dar
, exit_nr
);
980 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
982 kvmppc_set_dar(vcpu
, dar
);
983 kvmppc_set_dsisr(vcpu
, fault_dsisr
);
984 kvmppc_book3s_queue_irqprio(vcpu
, exit_nr
);
989 case BOOK3S_INTERRUPT_DATA_SEGMENT
:
990 if (kvmppc_mmu_map_segment(vcpu
, kvmppc_get_fault_dar(vcpu
)) < 0) {
991 kvmppc_set_dar(vcpu
, kvmppc_get_fault_dar(vcpu
));
992 kvmppc_book3s_queue_irqprio(vcpu
,
993 BOOK3S_INTERRUPT_DATA_SEGMENT
);
997 case BOOK3S_INTERRUPT_INST_SEGMENT
:
998 if (kvmppc_mmu_map_segment(vcpu
, kvmppc_get_pc(vcpu
)) < 0) {
999 kvmppc_book3s_queue_irqprio(vcpu
,
1000 BOOK3S_INTERRUPT_INST_SEGMENT
);
1004 /* We're good on these - the host merely wanted to get our attention */
1005 case BOOK3S_INTERRUPT_DECREMENTER
:
1006 case BOOK3S_INTERRUPT_HV_DECREMENTER
:
1007 case BOOK3S_INTERRUPT_DOORBELL
:
1008 case BOOK3S_INTERRUPT_H_DOORBELL
:
1009 vcpu
->stat
.dec_exits
++;
1012 case BOOK3S_INTERRUPT_EXTERNAL
:
1013 case BOOK3S_INTERRUPT_EXTERNAL_LEVEL
:
1014 case BOOK3S_INTERRUPT_EXTERNAL_HV
:
1015 vcpu
->stat
.ext_intr_exits
++;
1018 case BOOK3S_INTERRUPT_PERFMON
:
1021 case BOOK3S_INTERRUPT_PROGRAM
:
1022 case BOOK3S_INTERRUPT_H_EMUL_ASSIST
:
1024 enum emulation_result er
;
1030 flags
= vcpu
->arch
.shadow_srr1
& 0x1f0000ull
;
1032 emul
= kvmppc_get_last_inst(vcpu
, INST_GENERIC
, &last_inst
);
1033 if (emul
!= EMULATE_DONE
) {
1038 if (kvmppc_get_msr(vcpu
) & MSR_PR
) {
1040 pr_info("Userspace triggered 0x700 exception at\n 0x%lx (0x%x)\n",
1041 kvmppc_get_pc(vcpu
), last_inst
);
1043 if ((last_inst
& 0xff0007ff) !=
1044 (INS_DCBZ
& 0xfffffff7)) {
1045 kvmppc_core_queue_program(vcpu
, flags
);
1051 vcpu
->stat
.emulated_inst_exits
++;
1052 er
= kvmppc_emulate_instruction(run
, vcpu
);
1055 r
= RESUME_GUEST_NV
;
1061 printk(KERN_CRIT
"%s: emulation at %lx failed (%08x)\n",
1062 __func__
, kvmppc_get_pc(vcpu
), last_inst
);
1063 kvmppc_core_queue_program(vcpu
, flags
);
1066 case EMULATE_DO_MMIO
:
1067 run
->exit_reason
= KVM_EXIT_MMIO
;
1070 case EMULATE_EXIT_USER
:
1078 case BOOK3S_INTERRUPT_SYSCALL
:
1083 /* Get last sc for papr */
1084 if (vcpu
->arch
.papr_enabled
) {
1085 /* The sc instuction points SRR0 to the next inst */
1086 emul
= kvmppc_get_last_inst(vcpu
, INST_SC
, &last_sc
);
1087 if (emul
!= EMULATE_DONE
) {
1088 kvmppc_set_pc(vcpu
, kvmppc_get_pc(vcpu
) - 4);
1094 if (vcpu
->arch
.papr_enabled
&&
1095 (last_sc
== 0x44000022) &&
1096 !(kvmppc_get_msr(vcpu
) & MSR_PR
)) {
1097 /* SC 1 papr hypercalls */
1098 ulong cmd
= kvmppc_get_gpr(vcpu
, 3);
1101 #ifdef CONFIG_PPC_BOOK3S_64
1102 if (kvmppc_h_pr(vcpu
, cmd
) == EMULATE_DONE
) {
1108 run
->papr_hcall
.nr
= cmd
;
1109 for (i
= 0; i
< 9; ++i
) {
1110 ulong gpr
= kvmppc_get_gpr(vcpu
, 4 + i
);
1111 run
->papr_hcall
.args
[i
] = gpr
;
1113 run
->exit_reason
= KVM_EXIT_PAPR_HCALL
;
1114 vcpu
->arch
.hcall_needed
= 1;
1116 } else if (vcpu
->arch
.osi_enabled
&&
1117 (((u32
)kvmppc_get_gpr(vcpu
, 3)) == OSI_SC_MAGIC_R3
) &&
1118 (((u32
)kvmppc_get_gpr(vcpu
, 4)) == OSI_SC_MAGIC_R4
)) {
1119 /* MOL hypercalls */
1120 u64
*gprs
= run
->osi
.gprs
;
1123 run
->exit_reason
= KVM_EXIT_OSI
;
1124 for (i
= 0; i
< 32; i
++)
1125 gprs
[i
] = kvmppc_get_gpr(vcpu
, i
);
1126 vcpu
->arch
.osi_needed
= 1;
1128 } else if (!(kvmppc_get_msr(vcpu
) & MSR_PR
) &&
1129 (((u32
)kvmppc_get_gpr(vcpu
, 0)) == KVM_SC_MAGIC_R0
)) {
1130 /* KVM PV hypercalls */
1131 kvmppc_set_gpr(vcpu
, 3, kvmppc_kvm_pv(vcpu
));
1134 /* Guest syscalls */
1135 vcpu
->stat
.syscall_exits
++;
1136 kvmppc_book3s_queue_irqprio(vcpu
, exit_nr
);
1141 case BOOK3S_INTERRUPT_FP_UNAVAIL
:
1142 case BOOK3S_INTERRUPT_ALTIVEC
:
1143 case BOOK3S_INTERRUPT_VSX
:
1149 if (vcpu
->arch
.hflags
& BOOK3S_HFLAG_PAIRED_SINGLE
) {
1150 /* Do paired single instruction emulation */
1151 emul
= kvmppc_get_last_inst(vcpu
, INST_GENERIC
,
1153 if (emul
== EMULATE_DONE
)
1154 goto program_interrupt
;
1161 /* Enable external provider */
1163 case BOOK3S_INTERRUPT_FP_UNAVAIL
:
1167 case BOOK3S_INTERRUPT_ALTIVEC
:
1171 case BOOK3S_INTERRUPT_VSX
:
1176 r
= kvmppc_handle_ext(vcpu
, exit_nr
, ext_msr
);
1179 case BOOK3S_INTERRUPT_ALIGNMENT
:
1182 int emul
= kvmppc_get_last_inst(vcpu
, INST_GENERIC
, &last_inst
);
1184 if (emul
== EMULATE_DONE
) {
1188 dsisr
= kvmppc_alignment_dsisr(vcpu
, last_inst
);
1189 dar
= kvmppc_alignment_dar(vcpu
, last_inst
);
1191 kvmppc_set_dsisr(vcpu
, dsisr
);
1192 kvmppc_set_dar(vcpu
, dar
);
1194 kvmppc_book3s_queue_irqprio(vcpu
, exit_nr
);
1199 #ifdef CONFIG_PPC_BOOK3S_64
1200 case BOOK3S_INTERRUPT_FAC_UNAVAIL
:
1201 kvmppc_handle_fac(vcpu
, vcpu
->arch
.shadow_fscr
>> 56);
1205 case BOOK3S_INTERRUPT_MACHINE_CHECK
:
1206 case BOOK3S_INTERRUPT_TRACE
:
1207 kvmppc_book3s_queue_irqprio(vcpu
, exit_nr
);
1212 ulong shadow_srr1
= vcpu
->arch
.shadow_srr1
;
1213 /* Ugh - bork here! What did we get? */
1214 printk(KERN_EMERG
"exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
1215 exit_nr
, kvmppc_get_pc(vcpu
), shadow_srr1
);
1222 if (!(r
& RESUME_HOST
)) {
1223 /* To avoid clobbering exit_reason, only check for signals if
1224 * we aren't already exiting to userspace for some other
1228 * Interrupts could be timers for the guest which we have to
1229 * inject again, so let's postpone them until we're in the guest
1230 * and if we really did time things so badly, then we just exit
1231 * again due to a host external interrupt.
1233 s
= kvmppc_prepare_to_enter(vcpu
);
1237 /* interrupts now hard-disabled */
1238 kvmppc_fix_ee_before_entry();
1241 kvmppc_handle_lost_ext(vcpu
);
1244 trace_kvm_book3s_reenter(r
, vcpu
);
1249 static int kvm_arch_vcpu_ioctl_get_sregs_pr(struct kvm_vcpu
*vcpu
,
1250 struct kvm_sregs
*sregs
)
1252 struct kvmppc_vcpu_book3s
*vcpu3s
= to_book3s(vcpu
);
1255 sregs
->pvr
= vcpu
->arch
.pvr
;
1257 sregs
->u
.s
.sdr1
= to_book3s(vcpu
)->sdr1
;
1258 if (vcpu
->arch
.hflags
& BOOK3S_HFLAG_SLB
) {
1259 for (i
= 0; i
< 64; i
++) {
1260 sregs
->u
.s
.ppc64
.slb
[i
].slbe
= vcpu
->arch
.slb
[i
].orige
| i
;
1261 sregs
->u
.s
.ppc64
.slb
[i
].slbv
= vcpu
->arch
.slb
[i
].origv
;
1264 for (i
= 0; i
< 16; i
++)
1265 sregs
->u
.s
.ppc32
.sr
[i
] = kvmppc_get_sr(vcpu
, i
);
1267 for (i
= 0; i
< 8; i
++) {
1268 sregs
->u
.s
.ppc32
.ibat
[i
] = vcpu3s
->ibat
[i
].raw
;
1269 sregs
->u
.s
.ppc32
.dbat
[i
] = vcpu3s
->dbat
[i
].raw
;
1276 static int kvm_arch_vcpu_ioctl_set_sregs_pr(struct kvm_vcpu
*vcpu
,
1277 struct kvm_sregs
*sregs
)
1279 struct kvmppc_vcpu_book3s
*vcpu3s
= to_book3s(vcpu
);
1282 kvmppc_set_pvr_pr(vcpu
, sregs
->pvr
);
1284 vcpu3s
->sdr1
= sregs
->u
.s
.sdr1
;
1285 if (vcpu
->arch
.hflags
& BOOK3S_HFLAG_SLB
) {
1286 for (i
= 0; i
< 64; i
++) {
1287 vcpu
->arch
.mmu
.slbmte(vcpu
, sregs
->u
.s
.ppc64
.slb
[i
].slbv
,
1288 sregs
->u
.s
.ppc64
.slb
[i
].slbe
);
1291 for (i
= 0; i
< 16; i
++) {
1292 vcpu
->arch
.mmu
.mtsrin(vcpu
, i
, sregs
->u
.s
.ppc32
.sr
[i
]);
1294 for (i
= 0; i
< 8; i
++) {
1295 kvmppc_set_bat(vcpu
, &(vcpu3s
->ibat
[i
]), false,
1296 (u32
)sregs
->u
.s
.ppc32
.ibat
[i
]);
1297 kvmppc_set_bat(vcpu
, &(vcpu3s
->ibat
[i
]), true,
1298 (u32
)(sregs
->u
.s
.ppc32
.ibat
[i
] >> 32));
1299 kvmppc_set_bat(vcpu
, &(vcpu3s
->dbat
[i
]), false,
1300 (u32
)sregs
->u
.s
.ppc32
.dbat
[i
]);
1301 kvmppc_set_bat(vcpu
, &(vcpu3s
->dbat
[i
]), true,
1302 (u32
)(sregs
->u
.s
.ppc32
.dbat
[i
] >> 32));
1306 /* Flush the MMU after messing with the segments */
1307 kvmppc_mmu_pte_flush(vcpu
, 0, 0);
1312 static int kvmppc_get_one_reg_pr(struct kvm_vcpu
*vcpu
, u64 id
,
1313 union kvmppc_one_reg
*val
)
1318 case KVM_REG_PPC_DEBUG_INST
:
1319 *val
= get_reg_val(id
, KVMPPC_INST_SW_BREAKPOINT
);
1321 case KVM_REG_PPC_HIOR
:
1322 *val
= get_reg_val(id
, to_book3s(vcpu
)->hior
);
1324 case KVM_REG_PPC_LPCR
:
1325 case KVM_REG_PPC_LPCR_64
:
1327 * We are only interested in the LPCR_ILE bit
1329 if (vcpu
->arch
.intr_msr
& MSR_LE
)
1330 *val
= get_reg_val(id
, LPCR_ILE
);
1332 *val
= get_reg_val(id
, 0);
1342 static void kvmppc_set_lpcr_pr(struct kvm_vcpu
*vcpu
, u64 new_lpcr
)
1344 if (new_lpcr
& LPCR_ILE
)
1345 vcpu
->arch
.intr_msr
|= MSR_LE
;
1347 vcpu
->arch
.intr_msr
&= ~MSR_LE
;
1350 static int kvmppc_set_one_reg_pr(struct kvm_vcpu
*vcpu
, u64 id
,
1351 union kvmppc_one_reg
*val
)
1356 case KVM_REG_PPC_HIOR
:
1357 to_book3s(vcpu
)->hior
= set_reg_val(id
, *val
);
1358 to_book3s(vcpu
)->hior_explicit
= true;
1360 case KVM_REG_PPC_LPCR
:
1361 case KVM_REG_PPC_LPCR_64
:
1362 kvmppc_set_lpcr_pr(vcpu
, set_reg_val(id
, *val
));
1372 static struct kvm_vcpu
*kvmppc_core_vcpu_create_pr(struct kvm
*kvm
,
1375 struct kvmppc_vcpu_book3s
*vcpu_book3s
;
1376 struct kvm_vcpu
*vcpu
;
1380 vcpu
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL
);
1384 vcpu_book3s
= vzalloc(sizeof(struct kvmppc_vcpu_book3s
));
1387 vcpu
->arch
.book3s
= vcpu_book3s
;
1389 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1390 vcpu
->arch
.shadow_vcpu
=
1391 kzalloc(sizeof(*vcpu
->arch
.shadow_vcpu
), GFP_KERNEL
);
1392 if (!vcpu
->arch
.shadow_vcpu
)
1396 err
= kvm_vcpu_init(vcpu
, kvm
, id
);
1398 goto free_shadow_vcpu
;
1401 p
= __get_free_page(GFP_KERNEL
|__GFP_ZERO
);
1404 vcpu
->arch
.shared
= (void *)p
;
1405 #ifdef CONFIG_PPC_BOOK3S_64
1406 /* Always start the shared struct in native endian mode */
1407 #ifdef __BIG_ENDIAN__
1408 vcpu
->arch
.shared_big_endian
= true;
1410 vcpu
->arch
.shared_big_endian
= false;
1414 * Default to the same as the host if we're on sufficiently
1415 * recent machine that we have 1TB segments;
1416 * otherwise default to PPC970FX.
1418 vcpu
->arch
.pvr
= 0x3C0301;
1419 if (mmu_has_feature(MMU_FTR_1T_SEGMENT
))
1420 vcpu
->arch
.pvr
= mfspr(SPRN_PVR
);
1421 vcpu
->arch
.intr_msr
= MSR_SF
;
1423 /* default to book3s_32 (750) */
1424 vcpu
->arch
.pvr
= 0x84202;
1426 kvmppc_set_pvr_pr(vcpu
, vcpu
->arch
.pvr
);
1427 vcpu
->arch
.slb_nr
= 64;
1429 vcpu
->arch
.shadow_msr
= MSR_USER64
& ~MSR_LE
;
1431 err
= kvmppc_mmu_init(vcpu
);
1438 kvm_vcpu_uninit(vcpu
);
1440 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1441 kfree(vcpu
->arch
.shadow_vcpu
);
1446 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
1448 return ERR_PTR(err
);
1451 static void kvmppc_core_vcpu_free_pr(struct kvm_vcpu
*vcpu
)
1453 struct kvmppc_vcpu_book3s
*vcpu_book3s
= to_book3s(vcpu
);
1455 free_page((unsigned long)vcpu
->arch
.shared
& PAGE_MASK
);
1456 kvm_vcpu_uninit(vcpu
);
1457 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1458 kfree(vcpu
->arch
.shadow_vcpu
);
1461 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
1464 static int kvmppc_vcpu_run_pr(struct kvm_run
*kvm_run
, struct kvm_vcpu
*vcpu
)
1467 #ifdef CONFIG_ALTIVEC
1468 unsigned long uninitialized_var(vrsave
);
1471 /* Check if we can run the vcpu at all */
1472 if (!vcpu
->arch
.sane
) {
1473 kvm_run
->exit_reason
= KVM_EXIT_INTERNAL_ERROR
;
1479 * Interrupts could be timers for the guest which we have to inject
1480 * again, so let's postpone them until we're in the guest and if we
1481 * really did time things so badly, then we just exit again due to
1482 * a host external interrupt.
1484 ret
= kvmppc_prepare_to_enter(vcpu
);
1487 /* interrupts now hard-disabled */
1489 /* Save FPU state in thread_struct */
1490 if (current
->thread
.regs
->msr
& MSR_FP
)
1491 giveup_fpu(current
);
1493 #ifdef CONFIG_ALTIVEC
1494 /* Save Altivec state in thread_struct */
1495 if (current
->thread
.regs
->msr
& MSR_VEC
)
1496 giveup_altivec(current
);
1500 /* Save VSX state in thread_struct */
1501 if (current
->thread
.regs
->msr
& MSR_VSX
)
1502 __giveup_vsx(current
);
1505 /* Preload FPU if it's enabled */
1506 if (kvmppc_get_msr(vcpu
) & MSR_FP
)
1507 kvmppc_handle_ext(vcpu
, BOOK3S_INTERRUPT_FP_UNAVAIL
, MSR_FP
);
1509 kvmppc_fix_ee_before_entry();
1511 ret
= __kvmppc_vcpu_run(kvm_run
, vcpu
);
1513 /* No need for kvm_guest_exit. It's done in handle_exit.
1514 We also get here with interrupts enabled. */
1516 /* Make sure we save the guest FPU/Altivec/VSX state */
1517 kvmppc_giveup_ext(vcpu
, MSR_FP
| MSR_VEC
| MSR_VSX
);
1519 /* Make sure we save the guest TAR/EBB/DSCR state */
1520 kvmppc_giveup_fac(vcpu
, FSCR_TAR_LG
);
1523 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
1528 * Get (and clear) the dirty memory log for a memory slot.
1530 static int kvm_vm_ioctl_get_dirty_log_pr(struct kvm
*kvm
,
1531 struct kvm_dirty_log
*log
)
1533 struct kvm_memory_slot
*memslot
;
1534 struct kvm_vcpu
*vcpu
;
1540 mutex_lock(&kvm
->slots_lock
);
1542 r
= kvm_get_dirty_log(kvm
, log
, &is_dirty
);
1546 /* If nothing is dirty, don't bother messing with page tables. */
1548 memslot
= id_to_memslot(kvm
->memslots
, log
->slot
);
1550 ga
= memslot
->base_gfn
<< PAGE_SHIFT
;
1551 ga_end
= ga
+ (memslot
->npages
<< PAGE_SHIFT
);
1553 kvm_for_each_vcpu(n
, vcpu
, kvm
)
1554 kvmppc_mmu_pte_pflush(vcpu
, ga
, ga_end
);
1556 n
= kvm_dirty_bitmap_bytes(memslot
);
1557 memset(memslot
->dirty_bitmap
, 0, n
);
1562 mutex_unlock(&kvm
->slots_lock
);
1566 static void kvmppc_core_flush_memslot_pr(struct kvm
*kvm
,
1567 struct kvm_memory_slot
*memslot
)
1572 static int kvmppc_core_prepare_memory_region_pr(struct kvm
*kvm
,
1573 struct kvm_memory_slot
*memslot
,
1574 struct kvm_userspace_memory_region
*mem
)
1579 static void kvmppc_core_commit_memory_region_pr(struct kvm
*kvm
,
1580 struct kvm_userspace_memory_region
*mem
,
1581 const struct kvm_memory_slot
*old
)
1586 static void kvmppc_core_free_memslot_pr(struct kvm_memory_slot
*free
,
1587 struct kvm_memory_slot
*dont
)
1592 static int kvmppc_core_create_memslot_pr(struct kvm_memory_slot
*slot
,
1593 unsigned long npages
)
1600 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm
*kvm
,
1601 struct kvm_ppc_smmu_info
*info
)
1604 struct kvm_vcpu
*vcpu
;
1608 /* SLB is always 64 entries */
1609 info
->slb_size
= 64;
1611 /* Standard 4k base page size segment */
1612 info
->sps
[0].page_shift
= 12;
1613 info
->sps
[0].slb_enc
= 0;
1614 info
->sps
[0].enc
[0].page_shift
= 12;
1615 info
->sps
[0].enc
[0].pte_enc
= 0;
1618 * 64k large page size.
1619 * We only want to put this in if the CPUs we're emulating
1620 * support it, but unfortunately we don't have a vcpu easily
1621 * to hand here to test. Just pick the first vcpu, and if
1622 * that doesn't exist yet, report the minimum capability,
1623 * i.e., no 64k pages.
1624 * 1T segment support goes along with 64k pages.
1627 vcpu
= kvm_get_vcpu(kvm
, 0);
1628 if (vcpu
&& (vcpu
->arch
.hflags
& BOOK3S_HFLAG_MULTI_PGSIZE
)) {
1629 info
->flags
= KVM_PPC_1T_SEGMENTS
;
1630 info
->sps
[i
].page_shift
= 16;
1631 info
->sps
[i
].slb_enc
= SLB_VSID_L
| SLB_VSID_LP_01
;
1632 info
->sps
[i
].enc
[0].page_shift
= 16;
1633 info
->sps
[i
].enc
[0].pte_enc
= 1;
1637 /* Standard 16M large page size segment */
1638 info
->sps
[i
].page_shift
= 24;
1639 info
->sps
[i
].slb_enc
= SLB_VSID_L
;
1640 info
->sps
[i
].enc
[0].page_shift
= 24;
1641 info
->sps
[i
].enc
[0].pte_enc
= 0;
1646 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm
*kvm
,
1647 struct kvm_ppc_smmu_info
*info
)
1649 /* We should not get called */
1652 #endif /* CONFIG_PPC64 */
1654 static unsigned int kvm_global_user_count
= 0;
1655 static DEFINE_SPINLOCK(kvm_global_user_count_lock
);
1657 static int kvmppc_core_init_vm_pr(struct kvm
*kvm
)
1659 mutex_init(&kvm
->arch
.hpt_mutex
);
1661 #ifdef CONFIG_PPC_BOOK3S_64
1662 /* Start out with the default set of hcalls enabled */
1663 kvmppc_pr_init_default_hcalls(kvm
);
1666 if (firmware_has_feature(FW_FEATURE_SET_MODE
)) {
1667 spin_lock(&kvm_global_user_count_lock
);
1668 if (++kvm_global_user_count
== 1)
1669 pSeries_disable_reloc_on_exc();
1670 spin_unlock(&kvm_global_user_count_lock
);
1675 static void kvmppc_core_destroy_vm_pr(struct kvm
*kvm
)
1678 WARN_ON(!list_empty(&kvm
->arch
.spapr_tce_tables
));
1681 if (firmware_has_feature(FW_FEATURE_SET_MODE
)) {
1682 spin_lock(&kvm_global_user_count_lock
);
1683 BUG_ON(kvm_global_user_count
== 0);
1684 if (--kvm_global_user_count
== 0)
1685 pSeries_enable_reloc_on_exc();
1686 spin_unlock(&kvm_global_user_count_lock
);
1690 static int kvmppc_core_check_processor_compat_pr(void)
1692 /* we are always compatible */
1696 static long kvm_arch_vm_ioctl_pr(struct file
*filp
,
1697 unsigned int ioctl
, unsigned long arg
)
1702 static struct kvmppc_ops kvm_ops_pr
= {
1703 .get_sregs
= kvm_arch_vcpu_ioctl_get_sregs_pr
,
1704 .set_sregs
= kvm_arch_vcpu_ioctl_set_sregs_pr
,
1705 .get_one_reg
= kvmppc_get_one_reg_pr
,
1706 .set_one_reg
= kvmppc_set_one_reg_pr
,
1707 .vcpu_load
= kvmppc_core_vcpu_load_pr
,
1708 .vcpu_put
= kvmppc_core_vcpu_put_pr
,
1709 .set_msr
= kvmppc_set_msr_pr
,
1710 .vcpu_run
= kvmppc_vcpu_run_pr
,
1711 .vcpu_create
= kvmppc_core_vcpu_create_pr
,
1712 .vcpu_free
= kvmppc_core_vcpu_free_pr
,
1713 .check_requests
= kvmppc_core_check_requests_pr
,
1714 .get_dirty_log
= kvm_vm_ioctl_get_dirty_log_pr
,
1715 .flush_memslot
= kvmppc_core_flush_memslot_pr
,
1716 .prepare_memory_region
= kvmppc_core_prepare_memory_region_pr
,
1717 .commit_memory_region
= kvmppc_core_commit_memory_region_pr
,
1718 .unmap_hva
= kvm_unmap_hva_pr
,
1719 .unmap_hva_range
= kvm_unmap_hva_range_pr
,
1720 .age_hva
= kvm_age_hva_pr
,
1721 .test_age_hva
= kvm_test_age_hva_pr
,
1722 .set_spte_hva
= kvm_set_spte_hva_pr
,
1723 .mmu_destroy
= kvmppc_mmu_destroy_pr
,
1724 .free_memslot
= kvmppc_core_free_memslot_pr
,
1725 .create_memslot
= kvmppc_core_create_memslot_pr
,
1726 .init_vm
= kvmppc_core_init_vm_pr
,
1727 .destroy_vm
= kvmppc_core_destroy_vm_pr
,
1728 .get_smmu_info
= kvm_vm_ioctl_get_smmu_info_pr
,
1729 .emulate_op
= kvmppc_core_emulate_op_pr
,
1730 .emulate_mtspr
= kvmppc_core_emulate_mtspr_pr
,
1731 .emulate_mfspr
= kvmppc_core_emulate_mfspr_pr
,
1732 .fast_vcpu_kick
= kvm_vcpu_kick
,
1733 .arch_vm_ioctl
= kvm_arch_vm_ioctl_pr
,
1734 #ifdef CONFIG_PPC_BOOK3S_64
1735 .hcall_implemented
= kvmppc_hcall_impl_pr
,
1740 int kvmppc_book3s_init_pr(void)
1744 r
= kvmppc_core_check_processor_compat_pr();
1748 kvm_ops_pr
.owner
= THIS_MODULE
;
1749 kvmppc_pr_ops
= &kvm_ops_pr
;
1751 r
= kvmppc_mmu_hpte_sysinit();
1755 void kvmppc_book3s_exit_pr(void)
1757 kvmppc_pr_ops
= NULL
;
1758 kvmppc_mmu_hpte_sysexit();
1762 * We only support separate modules for book3s 64
1764 #ifdef CONFIG_PPC_BOOK3S_64
1766 module_init(kvmppc_book3s_init_pr
);
1767 module_exit(kvmppc_book3s_exit_pr
);
1769 MODULE_LICENSE("GPL");
1770 MODULE_ALIAS_MISCDEV(KVM_MINOR
);
1771 MODULE_ALIAS("devname:kvm");