?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'akpm'
[linux-2.6/next.git] / arch / powerpc / kvm / book3s_pr.c
blob86dd9ac4cd1b49f15785fbe3ed61a945651825cb
1 /*
2 * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
3 *
4 * Authors:
5 * Alexander Graf <agraf@suse.de>
6 * Kevin Wolf <mail@kevin-wolf.de>
7 * Paul Mackerras <paulus@samba.org>
8 *
9 * Description:
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).
13 *
14 * This file is derived from arch/powerpc/kvm/44x.c,
15 * by Hollis Blanchard <hollisb@us.ibm.com>.
16 *
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.
20 */
21
22 #include <linux/kvm_host.h>
23 #include <linux/export.h>
24 #include <linux/err.h>
25 #include <linux/slab.h>
26
27 #include <asm/reg.h>
28 #include <asm/cputable.h>
29 #include <asm/cacheflush.h>
30 #include <asm/tlbflush.h>
31 #include <asm/uaccess.h>
32 #include <asm/io.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>
40
41 #include "trace.h"
42
43 /* #define EXIT_DEBUG */
44 /* #define DEBUG_EXT */
45
46 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
47 ulong msr);
48
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
54 #endif
55
56 void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
57 {
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;
63 #endif
64
65 #ifdef CONFIG_PPC_BOOK3S_32
66 current->thread.kvm_shadow_vcpu = to_book3s(vcpu)->shadow_vcpu;
67 #endif
68 }
69
70 void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
71 {
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;
77 #endif
78
79 kvmppc_giveup_ext(vcpu, MSR_FP);
80 kvmppc_giveup_ext(vcpu, MSR_VEC);
81 kvmppc_giveup_ext(vcpu, MSR_VSX);
82 }
83
84 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu *vcpu)
85 {
86 ulong smsr = vcpu->arch.shared->msr;
87
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;
97 #endif
98 vcpu->arch.shadow_msr = smsr;
99 }
100
101 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
102 {
103 ulong old_msr = vcpu->arch.shared->msr;
104
105 #ifdef EXIT_DEBUG
106 printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
107 #endif
108
109 msr &= to_book3s(vcpu)->msr_mask;
110 vcpu->arch.shared->msr = msr;
111 kvmppc_recalc_shadow_msr(vcpu);
112
113 if (msr & MSR_POW) {
114 if (!vcpu->arch.pending_exceptions) {
115 kvm_vcpu_block(vcpu);
116 vcpu->stat.halt_wakeup++;
117
118 /* Unset POW bit after we woke up */
119 msr &= ~MSR_POW;
120 vcpu->arch.shared->msr = msr;
121 }
122 }
123
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));
128
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;
132
133 if (msr & MSR_DR)
134 kvmppc_mmu_map_segment(vcpu, a->magic_page_ea);
135 else
136 kvmppc_mmu_map_segment(vcpu, a->magic_page_pa);
137 }
138 }
139
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);
143 }
144
145 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
146 {
147 u32 host_pvr;
148
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 } else
157 #endif
158 {
159 kvmppc_mmu_book3s_32_init(vcpu);
160 to_book3s(vcpu)->hior = 0;
161 to_book3s(vcpu)->msr_mask = 0xffffffffULL;
162 }
163
164 /* If we are in hypervisor level on 970, we can tell the CPU to
165 * treat DCBZ as 32 bytes store */
166 vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
167 if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
168 !strcmp(cur_cpu_spec->platform, "ppc970"))
169 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
170
171 /* Cell performs badly if MSR_FEx are set. So let's hope nobody
172 really needs them in a VM on Cell and force disable them. */
173 if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be"))
174 to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1);
175
176 #ifdef CONFIG_PPC_BOOK3S_32
177 /* 32 bit Book3S always has 32 byte dcbz */
178 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
179 #endif
180
181 /* On some CPUs we can execute paired single operations natively */
182 asm ( "mfpvr %0" : "=r"(host_pvr));
183 switch (host_pvr) {
184 case 0x00080200: /* lonestar 2.0 */
185 case 0x00088202: /* lonestar 2.2 */
186 case 0x70000100: /* gekko 1.0 */
187 case 0x00080100: /* gekko 2.0 */
188 case 0x00083203: /* gekko 2.3a */
189 case 0x00083213: /* gekko 2.3b */
190 case 0x00083204: /* gekko 2.4 */
191 case 0x00083214: /* gekko 2.4e (8SE) - retail HW2 */
192 case 0x00087200: /* broadway */
193 vcpu->arch.hflags |= BOOK3S_HFLAG_NATIVE_PS;
194 /* Enable HID2.PSE - in case we need it later */
195 mtspr(SPRN_HID2_GEKKO, mfspr(SPRN_HID2_GEKKO) | (1 << 29));
196 }
197 }
198
199 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
200 * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
201 * emulate 32 bytes dcbz length.
202 *
203 * The Book3s_64 inventors also realized this case and implemented a special bit
204 * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
205 *
206 * My approach here is to patch the dcbz instruction on executing pages.
207 */
208 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
209 {
210 struct page *hpage;
211 u64 hpage_offset;
212 u32 *page;
213 int i;
214
215 hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
216 if (is_error_page(hpage)) {
217 kvm_release_page_clean(hpage);
218 return;
219 }
220
221 hpage_offset = pte->raddr & ~PAGE_MASK;
222 hpage_offset &= ~0xFFFULL;
223 hpage_offset /= 4;
224
225 get_page(hpage);
226 page = kmap_atomic(hpage, KM_USER0);
227
228 /* patch dcbz into reserved instruction, so we trap */
229 for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
230 if ((page[i] & 0xff0007ff) == INS_DCBZ)
231 page[i] &= 0xfffffff7;
232
233 kunmap_atomic(page, KM_USER0);
234 put_page(hpage);
235 }
236
237 static int kvmppc_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
238 {
239 ulong mp_pa = vcpu->arch.magic_page_pa;
240
241 if (unlikely(mp_pa) &&
242 unlikely((mp_pa & KVM_PAM) >> PAGE_SHIFT == gfn)) {
243 return 1;
244 }
245
246 return kvm_is_visible_gfn(vcpu->kvm, gfn);
247 }
248
249 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
250 ulong eaddr, int vec)
251 {
252 bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
253 int r = RESUME_GUEST;
254 int relocated;
255 int page_found = 0;
256 struct kvmppc_pte pte;
257 bool is_mmio = false;
258 bool dr = (vcpu->arch.shared->msr & MSR_DR) ? true : false;
259 bool ir = (vcpu->arch.shared->msr & MSR_IR) ? true : false;
260 u64 vsid;
261
262 relocated = data ? dr : ir;
263
264 /* Resolve real address if translation turned on */
265 if (relocated) {
266 page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data);
267 } else {
268 pte.may_execute = true;
269 pte.may_read = true;
270 pte.may_write = true;
271 pte.raddr = eaddr & KVM_PAM;
272 pte.eaddr = eaddr;
273 pte.vpage = eaddr >> 12;
274 }
275
276 switch (vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) {
277 case 0:
278 pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
279 break;
280 case MSR_DR:
281 case MSR_IR:
282 vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
283
284 if ((vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) == MSR_DR)
285 pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
286 else
287 pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
288 pte.vpage |= vsid;
289
290 if (vsid == -1)
291 page_found = -EINVAL;
292 break;
293 }
294
295 if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
296 (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
297 /*
298 * If we do the dcbz hack, we have to NX on every execution,
299 * so we can patch the executing code. This renders our guest
300 * NX-less.
301 */
302 pte.may_execute = !data;
303 }
304
305 if (page_found == -ENOENT) {
306 /* Page not found in guest PTE entries */
307 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
308 vcpu->arch.shared->dsisr = to_svcpu(vcpu)->fault_dsisr;
309 vcpu->arch.shared->msr |=
310 (to_svcpu(vcpu)->shadow_srr1 & 0x00000000f8000000ULL);
311 kvmppc_book3s_queue_irqprio(vcpu, vec);
312 } else if (page_found == -EPERM) {
313 /* Storage protection */
314 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
315 vcpu->arch.shared->dsisr =
316 to_svcpu(vcpu)->fault_dsisr & ~DSISR_NOHPTE;
317 vcpu->arch.shared->dsisr |= DSISR_PROTFAULT;
318 vcpu->arch.shared->msr |=
319 (to_svcpu(vcpu)->shadow_srr1 & 0x00000000f8000000ULL);
320 kvmppc_book3s_queue_irqprio(vcpu, vec);
321 } else if (page_found == -EINVAL) {
322 /* Page not found in guest SLB */
323 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
324 kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
325 } else if (!is_mmio &&
326 kvmppc_visible_gfn(vcpu, pte.raddr >> PAGE_SHIFT)) {
327 /* The guest's PTE is not mapped yet. Map on the host */
328 kvmppc_mmu_map_page(vcpu, &pte);
329 if (data)
330 vcpu->stat.sp_storage++;
331 else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
332 (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
333 kvmppc_patch_dcbz(vcpu, &pte);
334 } else {
335 /* MMIO */
336 vcpu->stat.mmio_exits++;
337 vcpu->arch.paddr_accessed = pte.raddr;
338 r = kvmppc_emulate_mmio(run, vcpu);
339 if ( r == RESUME_HOST_NV )
340 r = RESUME_HOST;
341 }
342
343 return r;
344 }
345
346 static inline int get_fpr_index(int i)
347 {
348 #ifdef CONFIG_VSX
349 i *= 2;
350 #endif
351 return i;
352 }
353
354 /* Give up external provider (FPU, Altivec, VSX) */
355 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
356 {
357 struct thread_struct *t = &current->thread;
358 u64 *vcpu_fpr = vcpu->arch.fpr;
359 #ifdef CONFIG_VSX
360 u64 *vcpu_vsx = vcpu->arch.vsr;
361 #endif
362 u64 *thread_fpr = (u64*)t->fpr;
363 int i;
364
365 if (!(vcpu->arch.guest_owned_ext & msr))
366 return;
367
368 #ifdef DEBUG_EXT
369 printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
370 #endif
371
372 switch (msr) {
373 case MSR_FP:
374 giveup_fpu(current);
375 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
376 vcpu_fpr[i] = thread_fpr[get_fpr_index(i)];
377
378 vcpu->arch.fpscr = t->fpscr.val;
379 break;
380 case MSR_VEC:
381 #ifdef CONFIG_ALTIVEC
382 giveup_altivec(current);
383 memcpy(vcpu->arch.vr, t->vr, sizeof(vcpu->arch.vr));
384 vcpu->arch.vscr = t->vscr;
385 #endif
386 break;
387 case MSR_VSX:
388 #ifdef CONFIG_VSX
389 __giveup_vsx(current);
390 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++)
391 vcpu_vsx[i] = thread_fpr[get_fpr_index(i) + 1];
392 #endif
393 break;
394 default:
395 BUG();
396 }
397
398 vcpu->arch.guest_owned_ext &= ~msr;
399 current->thread.regs->msr &= ~msr;
400 kvmppc_recalc_shadow_msr(vcpu);
401 }
402
403 static int kvmppc_read_inst(struct kvm_vcpu *vcpu)
404 {
405 ulong srr0 = kvmppc_get_pc(vcpu);
406 u32 last_inst = kvmppc_get_last_inst(vcpu);
407 int ret;
408
409 ret = kvmppc_ld(vcpu, &srr0, sizeof(u32), &last_inst, false);
410 if (ret == -ENOENT) {
411 ulong msr = vcpu->arch.shared->msr;
412
413 msr = kvmppc_set_field(msr, 33, 33, 1);
414 msr = kvmppc_set_field(msr, 34, 36, 0);
415 vcpu->arch.shared->msr = kvmppc_set_field(msr, 42, 47, 0);
416 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_INST_STORAGE);
417 return EMULATE_AGAIN;
418 }
419
420 return EMULATE_DONE;
421 }
422
423 static int kvmppc_check_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr)
424 {
425
426 /* Need to do paired single emulation? */
427 if (!(vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE))
428 return EMULATE_DONE;
429
430 /* Read out the instruction */
431 if (kvmppc_read_inst(vcpu) == EMULATE_DONE)
432 /* Need to emulate */
433 return EMULATE_FAIL;
434
435 return EMULATE_AGAIN;
436 }
437
438 /* Handle external providers (FPU, Altivec, VSX) */
439 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
440 ulong msr)
441 {
442 struct thread_struct *t = &current->thread;
443 u64 *vcpu_fpr = vcpu->arch.fpr;
444 #ifdef CONFIG_VSX
445 u64 *vcpu_vsx = vcpu->arch.vsr;
446 #endif
447 u64 *thread_fpr = (u64*)t->fpr;
448 int i;
449
450 /* When we have paired singles, we emulate in software */
451 if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
452 return RESUME_GUEST;
453
454 if (!(vcpu->arch.shared->msr & msr)) {
455 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
456 return RESUME_GUEST;
457 }
458
459 /* We already own the ext */
460 if (vcpu->arch.guest_owned_ext & msr) {
461 return RESUME_GUEST;
462 }
463
464 #ifdef DEBUG_EXT
465 printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
466 #endif
467
468 current->thread.regs->msr |= msr;
469
470 switch (msr) {
471 case MSR_FP:
472 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
473 thread_fpr[get_fpr_index(i)] = vcpu_fpr[i];
474
475 t->fpscr.val = vcpu->arch.fpscr;
476 t->fpexc_mode = 0;
477 kvmppc_load_up_fpu();
478 break;
479 case MSR_VEC:
480 #ifdef CONFIG_ALTIVEC
481 memcpy(t->vr, vcpu->arch.vr, sizeof(vcpu->arch.vr));
482 t->vscr = vcpu->arch.vscr;
483 t->vrsave = -1;
484 kvmppc_load_up_altivec();
485 #endif
486 break;
487 case MSR_VSX:
488 #ifdef CONFIG_VSX
489 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++)
490 thread_fpr[get_fpr_index(i) + 1] = vcpu_vsx[i];
491 kvmppc_load_up_vsx();
492 #endif
493 break;
494 default:
495 BUG();
496 }
497
498 vcpu->arch.guest_owned_ext |= msr;
499
500 kvmppc_recalc_shadow_msr(vcpu);
501
502 return RESUME_GUEST;
503 }
504
505 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
506 unsigned int exit_nr)
507 {
508 int r = RESUME_HOST;
509
510 vcpu->stat.sum_exits++;
511
512 run->exit_reason = KVM_EXIT_UNKNOWN;
513 run->ready_for_interrupt_injection = 1;
514
515 trace_kvm_book3s_exit(exit_nr, vcpu);
516 kvm_resched(vcpu);
517 switch (exit_nr) {
518 case BOOK3S_INTERRUPT_INST_STORAGE:
519 vcpu->stat.pf_instruc++;
520
521 #ifdef CONFIG_PPC_BOOK3S_32
522 /* We set segments as unused segments when invalidating them. So
523 * treat the respective fault as segment fault. */
524 if (to_svcpu(vcpu)->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT]
525 == SR_INVALID) {
526 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
527 r = RESUME_GUEST;
528 break;
529 }
530 #endif
531
532 /* only care about PTEG not found errors, but leave NX alone */
533 if (to_svcpu(vcpu)->shadow_srr1 & 0x40000000) {
534 r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr);
535 vcpu->stat.sp_instruc++;
536 } else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
537 (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
538 /*
539 * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
540 * so we can't use the NX bit inside the guest. Let's cross our fingers,
541 * that no guest that needs the dcbz hack does NX.
542 */
543 kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
544 r = RESUME_GUEST;
545 } else {
546 vcpu->arch.shared->msr |=
547 to_svcpu(vcpu)->shadow_srr1 & 0x58000000;
548 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
549 r = RESUME_GUEST;
550 }
551 break;
552 case BOOK3S_INTERRUPT_DATA_STORAGE:
553 {
554 ulong dar = kvmppc_get_fault_dar(vcpu);
555 vcpu->stat.pf_storage++;
556
557 #ifdef CONFIG_PPC_BOOK3S_32
558 /* We set segments as unused segments when invalidating them. So
559 * treat the respective fault as segment fault. */
560 if ((to_svcpu(vcpu)->sr[dar >> SID_SHIFT]) == SR_INVALID) {
561 kvmppc_mmu_map_segment(vcpu, dar);
562 r = RESUME_GUEST;
563 break;
564 }
565 #endif
566
567 /* The only case we need to handle is missing shadow PTEs */
568 if (to_svcpu(vcpu)->fault_dsisr & DSISR_NOHPTE) {
569 r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr);
570 } else {
571 vcpu->arch.shared->dar = dar;
572 vcpu->arch.shared->dsisr = to_svcpu(vcpu)->fault_dsisr;
573 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
574 r = RESUME_GUEST;
575 }
576 break;
577 }
578 case BOOK3S_INTERRUPT_DATA_SEGMENT:
579 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
580 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
581 kvmppc_book3s_queue_irqprio(vcpu,
582 BOOK3S_INTERRUPT_DATA_SEGMENT);
583 }
584 r = RESUME_GUEST;
585 break;
586 case BOOK3S_INTERRUPT_INST_SEGMENT:
587 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
588 kvmppc_book3s_queue_irqprio(vcpu,
589 BOOK3S_INTERRUPT_INST_SEGMENT);
590 }
591 r = RESUME_GUEST;
592 break;
593 /* We're good on these - the host merely wanted to get our attention */
594 case BOOK3S_INTERRUPT_DECREMENTER:
595 vcpu->stat.dec_exits++;
596 r = RESUME_GUEST;
597 break;
598 case BOOK3S_INTERRUPT_EXTERNAL:
599 vcpu->stat.ext_intr_exits++;
600 r = RESUME_GUEST;
601 break;
602 case BOOK3S_INTERRUPT_PERFMON:
603 r = RESUME_GUEST;
604 break;
605 case BOOK3S_INTERRUPT_PROGRAM:
606 {
607 enum emulation_result er;
608 ulong flags;
609
610 program_interrupt:
611 flags = to_svcpu(vcpu)->shadow_srr1 & 0x1f0000ull;
612
613 if (vcpu->arch.shared->msr & MSR_PR) {
614 #ifdef EXIT_DEBUG
615 printk(KERN_INFO "Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
616 #endif
617 if ((kvmppc_get_last_inst(vcpu) & 0xff0007ff) !=
618 (INS_DCBZ & 0xfffffff7)) {
619 kvmppc_core_queue_program(vcpu, flags);
620 r = RESUME_GUEST;
621 break;
622 }
623 }
624
625 vcpu->stat.emulated_inst_exits++;
626 er = kvmppc_emulate_instruction(run, vcpu);
627 switch (er) {
628 case EMULATE_DONE:
629 r = RESUME_GUEST_NV;
630 break;
631 case EMULATE_AGAIN:
632 r = RESUME_GUEST;
633 break;
634 case EMULATE_FAIL:
635 printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
636 __func__, kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
637 kvmppc_core_queue_program(vcpu, flags);
638 r = RESUME_GUEST;
639 break;
640 case EMULATE_DO_MMIO:
641 run->exit_reason = KVM_EXIT_MMIO;
642 r = RESUME_HOST_NV;
643 break;
644 default:
645 BUG();
646 }
647 break;
648 }
649 case BOOK3S_INTERRUPT_SYSCALL:
650 if (vcpu->arch.osi_enabled &&
651 (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
652 (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
653 /* MOL hypercalls */
654 u64 *gprs = run->osi.gprs;
655 int i;
656
657 run->exit_reason = KVM_EXIT_OSI;
658 for (i = 0; i < 32; i++)
659 gprs[i] = kvmppc_get_gpr(vcpu, i);
660 vcpu->arch.osi_needed = 1;
661 r = RESUME_HOST_NV;
662 } else if (!(vcpu->arch.shared->msr & MSR_PR) &&
663 (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
664 /* KVM PV hypercalls */
665 kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
666 r = RESUME_GUEST;
667 } else {
668 /* Guest syscalls */
669 vcpu->stat.syscall_exits++;
670 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
671 r = RESUME_GUEST;
672 }
673 break;
674 case BOOK3S_INTERRUPT_FP_UNAVAIL:
675 case BOOK3S_INTERRUPT_ALTIVEC:
676 case BOOK3S_INTERRUPT_VSX:
677 {
678 int ext_msr = 0;
679
680 switch (exit_nr) {
681 case BOOK3S_INTERRUPT_FP_UNAVAIL: ext_msr = MSR_FP; break;
682 case BOOK3S_INTERRUPT_ALTIVEC: ext_msr = MSR_VEC; break;
683 case BOOK3S_INTERRUPT_VSX: ext_msr = MSR_VSX; break;
684 }
685
686 switch (kvmppc_check_ext(vcpu, exit_nr)) {
687 case EMULATE_DONE:
688 /* everything ok - let's enable the ext */
689 r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
690 break;
691 case EMULATE_FAIL:
692 /* we need to emulate this instruction */
693 goto program_interrupt;
694 break;
695 default:
696 /* nothing to worry about - go again */
697 break;
698 }
699 break;
700 }
701 case BOOK3S_INTERRUPT_ALIGNMENT:
702 if (kvmppc_read_inst(vcpu) == EMULATE_DONE) {
703 vcpu->arch.shared->dsisr = kvmppc_alignment_dsisr(vcpu,
704 kvmppc_get_last_inst(vcpu));
705 vcpu->arch.shared->dar = kvmppc_alignment_dar(vcpu,
706 kvmppc_get_last_inst(vcpu));
707 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
708 }
709 r = RESUME_GUEST;
710 break;
711 case BOOK3S_INTERRUPT_MACHINE_CHECK:
712 case BOOK3S_INTERRUPT_TRACE:
713 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
714 r = RESUME_GUEST;
715 break;
716 default:
717 /* Ugh - bork here! What did we get? */
718 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
719 exit_nr, kvmppc_get_pc(vcpu), to_svcpu(vcpu)->shadow_srr1);
720 r = RESUME_HOST;
721 BUG();
722 break;
723 }
724
725
726 if (!(r & RESUME_HOST)) {
727 /* To avoid clobbering exit_reason, only check for signals if
728 * we aren't already exiting to userspace for some other
729 * reason. */
730 if (signal_pending(current)) {
731 #ifdef EXIT_DEBUG
732 printk(KERN_EMERG "KVM: Going back to host\n");
733 #endif
734 vcpu->stat.signal_exits++;
735 run->exit_reason = KVM_EXIT_INTR;
736 r = -EINTR;
737 } else {
738 /* In case an interrupt came in that was triggered
739 * from userspace (like DEC), we need to check what
740 * to inject now! */
741 kvmppc_core_deliver_interrupts(vcpu);
742 }
743 }
744
745 trace_kvm_book3s_reenter(r, vcpu);
746
747 return r;
748 }
749
750 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
751 struct kvm_sregs *sregs)
752 {
753 struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
754 int i;
755
756 sregs->pvr = vcpu->arch.pvr;
757
758 sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
759 if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
760 for (i = 0; i < 64; i++) {
761 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige | i;
762 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
763 }
764 } else {
765 for (i = 0; i < 16; i++)
766 sregs->u.s.ppc32.sr[i] = vcpu->arch.shared->sr[i];
767
768 for (i = 0; i < 8; i++) {
769 sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
770 sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
771 }
772 }
773
774 return 0;
775 }
776
777 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
778 struct kvm_sregs *sregs)
779 {
780 struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
781 int i;
782
783 kvmppc_set_pvr(vcpu, sregs->pvr);
784
785 vcpu3s->sdr1 = sregs->u.s.sdr1;
786 if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
787 for (i = 0; i < 64; i++) {
788 vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
789 sregs->u.s.ppc64.slb[i].slbe);
790 }
791 } else {
792 for (i = 0; i < 16; i++) {
793 vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
794 }
795 for (i = 0; i < 8; i++) {
796 kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
797 (u32)sregs->u.s.ppc32.ibat[i]);
798 kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
799 (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
800 kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
801 (u32)sregs->u.s.ppc32.dbat[i]);
802 kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
803 (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
804 }
805 }
806
807 /* Flush the MMU after messing with the segments */
808 kvmppc_mmu_pte_flush(vcpu, 0, 0);
809
810 return 0;
811 }
812
813 int kvmppc_core_check_processor_compat(void)
814 {
815 return 0;
816 }
817
818 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
819 {
820 struct kvmppc_vcpu_book3s *vcpu_book3s;
821 struct kvm_vcpu *vcpu;
822 int err = -ENOMEM;
823 unsigned long p;
824
825 vcpu_book3s = vzalloc(sizeof(struct kvmppc_vcpu_book3s));
826 if (!vcpu_book3s)
827 goto out;
828
829 vcpu_book3s->shadow_vcpu = (struct kvmppc_book3s_shadow_vcpu *)
830 kzalloc(sizeof(*vcpu_book3s->shadow_vcpu), GFP_KERNEL);
831 if (!vcpu_book3s->shadow_vcpu)
832 goto free_vcpu;
833
834 vcpu = &vcpu_book3s->vcpu;
835 err = kvm_vcpu_init(vcpu, kvm, id);
836 if (err)
837 goto free_shadow_vcpu;
838
839 p = __get_free_page(GFP_KERNEL|__GFP_ZERO);
840 /* the real shared page fills the last 4k of our page */
841 vcpu->arch.shared = (void*)(p + PAGE_SIZE - 4096);
842 if (!p)
843 goto uninit_vcpu;
844
845 vcpu->arch.host_retip = kvm_return_point;
846 vcpu->arch.host_msr = mfmsr();
847 #ifdef CONFIG_PPC_BOOK3S_64
848 /* default to book3s_64 (970fx) */
849 vcpu->arch.pvr = 0x3C0301;
850 #else
851 /* default to book3s_32 (750) */
852 vcpu->arch.pvr = 0x84202;
853 #endif
854 kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
855 vcpu->arch.slb_nr = 64;
856
857 /* remember where some real-mode handlers are */
858 vcpu->arch.trampoline_lowmem = __pa(kvmppc_handler_lowmem_trampoline);
859 vcpu->arch.trampoline_enter = __pa(kvmppc_handler_trampoline_enter);
860 vcpu->arch.highmem_handler = (ulong)kvmppc_handler_highmem;
861 #ifdef CONFIG_PPC_BOOK3S_64
862 vcpu->arch.rmcall = *(ulong*)kvmppc_rmcall;
863 #else
864 vcpu->arch.rmcall = (ulong)kvmppc_rmcall;
865 #endif
866
867 vcpu->arch.shadow_msr = MSR_USER64;
868
869 err = kvmppc_mmu_init(vcpu);
870 if (err < 0)
871 goto uninit_vcpu;
872
873 return vcpu;
874
875 uninit_vcpu:
876 kvm_vcpu_uninit(vcpu);
877 free_shadow_vcpu:
878 kfree(vcpu_book3s->shadow_vcpu);
879 free_vcpu:
880 vfree(vcpu_book3s);
881 out:
882 return ERR_PTR(err);
883 }
884
885 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
886 {
887 struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
888
889 free_page((unsigned long)vcpu->arch.shared & PAGE_MASK);
890 kvm_vcpu_uninit(vcpu);
891 kfree(vcpu_book3s->shadow_vcpu);
892 vfree(vcpu_book3s);
893 }
894
895 int kvmppc_vcpu_run(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
896 {
897 int ret;
898 double fpr[32][TS_FPRWIDTH];
899 unsigned int fpscr;
900 int fpexc_mode;
901 #ifdef CONFIG_ALTIVEC
902 vector128 vr[32];
903 vector128 vscr;
904 unsigned long uninitialized_var(vrsave);
905 int used_vr;
906 #endif
907 #ifdef CONFIG_VSX
908 int used_vsr;
909 #endif
910 ulong ext_msr;
911
912 /* No need to go into the guest when all we do is going out */
913 if (signal_pending(current)) {
914 kvm_run->exit_reason = KVM_EXIT_INTR;
915 return -EINTR;
916 }
917
918 /* Save FPU state in stack */
919 if (current->thread.regs->msr & MSR_FP)
920 giveup_fpu(current);
921 memcpy(fpr, current->thread.fpr, sizeof(current->thread.fpr));
922 fpscr = current->thread.fpscr.val;
923 fpexc_mode = current->thread.fpexc_mode;
924
925 #ifdef CONFIG_ALTIVEC
926 /* Save Altivec state in stack */
927 used_vr = current->thread.used_vr;
928 if (used_vr) {
929 if (current->thread.regs->msr & MSR_VEC)
930 giveup_altivec(current);
931 memcpy(vr, current->thread.vr, sizeof(current->thread.vr));
932 vscr = current->thread.vscr;
933 vrsave = current->thread.vrsave;
934 }
935 #endif
936
937 #ifdef CONFIG_VSX
938 /* Save VSX state in stack */
939 used_vsr = current->thread.used_vsr;
940 if (used_vsr && (current->thread.regs->msr & MSR_VSX))
941 __giveup_vsx(current);
942 #endif
943
944 /* Remember the MSR with disabled extensions */
945 ext_msr = current->thread.regs->msr;
946
947 /* Preload FPU if it's enabled */
948 if (vcpu->arch.shared->msr & MSR_FP)
949 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
950
951 kvm_guest_enter();
952
953 ret = __kvmppc_vcpu_run(kvm_run, vcpu);
954
955 kvm_guest_exit();
956
957 local_irq_disable();
958
959 current->thread.regs->msr = ext_msr;
960
961 /* Make sure we save the guest FPU/Altivec/VSX state */
962 kvmppc_giveup_ext(vcpu, MSR_FP);
963 kvmppc_giveup_ext(vcpu, MSR_VEC);
964 kvmppc_giveup_ext(vcpu, MSR_VSX);
965
966 /* Restore FPU state from stack */
967 memcpy(current->thread.fpr, fpr, sizeof(current->thread.fpr));
968 current->thread.fpscr.val = fpscr;
969 current->thread.fpexc_mode = fpexc_mode;
970
971 #ifdef CONFIG_ALTIVEC
972 /* Restore Altivec state from stack */
973 if (used_vr && current->thread.used_vr) {
974 memcpy(current->thread.vr, vr, sizeof(current->thread.vr));
975 current->thread.vscr = vscr;
976 current->thread.vrsave = vrsave;
977 }
978 current->thread.used_vr = used_vr;
979 #endif
980
981 #ifdef CONFIG_VSX
982 current->thread.used_vsr = used_vsr;
983 #endif
984
985 return ret;
986 }
987
988 int kvmppc_core_prepare_memory_region(struct kvm *kvm,
989 struct kvm_userspace_memory_region *mem)
990 {
991 return 0;
992 }
993
994 void kvmppc_core_commit_memory_region(struct kvm *kvm,
995 struct kvm_userspace_memory_region *mem)
996 {
997 }
998
999 int kvmppc_core_init_vm(struct kvm *kvm)
1000 {
1001 return 0;
1002 }
1003
1004 void kvmppc_core_destroy_vm(struct kvm *kvm)
1005 {
1006 }
1007
1008 static int kvmppc_book3s_init(void)
1009 {
1010 int r;
1011
1012 r = kvm_init(NULL, sizeof(struct kvmppc_vcpu_book3s), 0,
1013 THIS_MODULE);
1014
1015 if (r)
1016 return r;
1017
1018 r = kvmppc_mmu_hpte_sysinit();
1019
1020 return r;
1021 }
1022
1023 static void kvmppc_book3s_exit(void)
1024 {
1025 kvmppc_mmu_hpte_sysexit();
1026 kvm_exit();
1027 }
1028
1029 module_init(kvmppc_book3s_init);
1030 module_exit(kvmppc_book3s_exit);
linux-next