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 <linux/gfp.h>
38 #include <linux/sched.h>
39 #include <linux/vmalloc.h>
40 #include <linux/highmem.h>
44 /* #define EXIT_DEBUG */
45 /* #define DEBUG_EXT */
47 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
50 /* Some compatibility defines */
51 #ifdef CONFIG_PPC_BOOK3S_32
52 #define MSR_USER32 MSR_USER
53 #define MSR_USER64 MSR_USER
54 #define HW_PAGE_SIZE PAGE_SIZE
55 #define __hard_irq_disable local_irq_disable
56 #define __hard_irq_enable local_irq_enable
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;
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);
87 kvmppc_giveup_ext(vcpu, MSR_VEC);
88 kvmppc_giveup_ext(vcpu, MSR_VSX);
91 void kvmppc_core_check_requests(struct kvm_vcpu *vcpu)
95 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu *vcpu)
97 ulong smsr = vcpu->arch.shared->msr;
99 /* Guest MSR values */
100 smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE | MSR_DE;
101 /* Process MSR values */
102 smsr |= MSR_ME | MSR_RI | MSR_IR | MSR_DR | MSR_PR | MSR_EE;
103 /* External providers the guest reserved */
104 smsr |= (vcpu->arch.shared->msr & vcpu->arch.guest_owned_ext);
105 /* 64-bit Process MSR values */
106 #ifdef CONFIG_PPC_BOOK3S_64
107 smsr |= MSR_ISF | MSR_HV;
109 vcpu->arch.shadow_msr = smsr;
112 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
114 ulong old_msr = vcpu->arch.shared->msr;
117 printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
120 msr &= to_book3s(vcpu)->msr_mask;
121 vcpu->arch.shared->msr = msr;
122 kvmppc_recalc_shadow_msr(vcpu);
125 if (!vcpu->arch.pending_exceptions) {
126 kvm_vcpu_block(vcpu);
127 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
128 vcpu->stat.halt_wakeup++;
130 /* Unset POW bit after we woke up */
132 vcpu->arch.shared->msr = msr;
136 if ((vcpu->arch.shared->msr & (MSR_PR|MSR_IR|MSR_DR)) !=
137 (old_msr & (MSR_PR|MSR_IR|MSR_DR))) {
138 kvmppc_mmu_flush_segments(vcpu);
139 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
141 /* Preload magic page segment when in kernel mode */
142 if (!(msr & MSR_PR) && vcpu->arch.magic_page_pa) {
143 struct kvm_vcpu_arch *a = &vcpu->arch;
146 kvmppc_mmu_map_segment(vcpu, a->magic_page_ea);
148 kvmppc_mmu_map_segment(vcpu, a->magic_page_pa);
153 * When switching from 32 to 64-bit, we may have a stale 32-bit
154 * magic page around, we need to flush it. Typically 32-bit magic
155 * page will be instanciated when calling into RTAS. Note: We
156 * assume that such transition only happens while in kernel mode,
157 * ie, we never transition from user 32-bit to kernel 64-bit with
158 * a 32-bit magic page around.
160 if (vcpu->arch.magic_page_pa &&
161 !(old_msr & MSR_PR) && !(old_msr & MSR_SF) && (msr & MSR_SF)) {
162 /* going from RTAS to normal kernel code */
163 kvmppc_mmu_pte_flush(vcpu, (uint32_t)vcpu->arch.magic_page_pa,
167 /* Preload FPU if it's enabled */
168 if (vcpu->arch.shared->msr & MSR_FP)
169 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
172 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
176 vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
177 vcpu->arch.pvr = pvr;
178 #ifdef CONFIG_PPC_BOOK3S_64
179 if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
180 kvmppc_mmu_book3s_64_init(vcpu);
181 if (!to_book3s(vcpu)->hior_explicit)
182 to_book3s(vcpu)->hior = 0xfff00000;
183 to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
184 vcpu->arch.cpu_type = KVM_CPU_3S_64;
188 kvmppc_mmu_book3s_32_init(vcpu);
189 if (!to_book3s(vcpu)->hior_explicit)
190 to_book3s(vcpu)->hior = 0;
191 to_book3s(vcpu)->msr_mask = 0xffffffffULL;
192 vcpu->arch.cpu_type = KVM_CPU_3S_32;
195 kvmppc_sanity_check(vcpu);
197 /* If we are in hypervisor level on 970, we can tell the CPU to
198 * treat DCBZ as 32 bytes store */
199 vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
200 if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
201 !strcmp(cur_cpu_spec->platform, "ppc970"))
202 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
204 /* Cell performs badly if MSR_FEx are set. So let's hope nobody
205 really needs them in a VM on Cell and force disable them. */
206 if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be"))
207 to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1);
209 #ifdef CONFIG_PPC_BOOK3S_32
210 /* 32 bit Book3S always has 32 byte dcbz */
211 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
214 /* On some CPUs we can execute paired single operations natively */
215 asm ( "mfpvr %0" : "=r"(host_pvr));
217 case 0x00080200: /* lonestar 2.0 */
218 case 0x00088202: /* lonestar 2.2 */
219 case 0x70000100: /* gekko 1.0 */
220 case 0x00080100: /* gekko 2.0 */
221 case 0x00083203: /* gekko 2.3a */
222 case 0x00083213: /* gekko 2.3b */
223 case 0x00083204: /* gekko 2.4 */
224 case 0x00083214: /* gekko 2.4e (8SE) - retail HW2 */
225 case 0x00087200: /* broadway */
226 vcpu->arch.hflags |= BOOK3S_HFLAG_NATIVE_PS;
227 /* Enable HID2.PSE - in case we need it later */
228 mtspr(SPRN_HID2_GEKKO, mfspr(SPRN_HID2_GEKKO) | (1 << 29));
232 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
233 * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
234 * emulate 32 bytes dcbz length.
236 * The Book3s_64 inventors also realized this case and implemented a special bit
237 * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
239 * My approach here is to patch the dcbz instruction on executing pages.
241 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
248 hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
249 if (is_error_page(hpage))
252 hpage_offset = pte->raddr & ~PAGE_MASK;
253 hpage_offset &= ~0xFFFULL;
257 page = kmap_atomic(hpage);
259 /* patch dcbz into reserved instruction, so we trap */
260 for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
261 if ((page[i] & 0xff0007ff) == INS_DCBZ)
262 page[i] &= 0xfffffff7;
268 static int kvmppc_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
270 ulong mp_pa = vcpu->arch.magic_page_pa;
272 if (!(vcpu->arch.shared->msr & MSR_SF))
273 mp_pa = (uint32_t)mp_pa;
275 if (unlikely(mp_pa) &&
276 unlikely((mp_pa & KVM_PAM) >> PAGE_SHIFT == gfn)) {
280 return kvm_is_visible_gfn(vcpu->kvm, gfn);
283 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
284 ulong eaddr, int vec)
286 bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
287 int r = RESUME_GUEST;
290 struct kvmppc_pte pte;
291 bool is_mmio = false;
292 bool dr = (vcpu->arch.shared->msr & MSR_DR) ? true : false;
293 bool ir = (vcpu->arch.shared->msr & MSR_IR) ? true : false;
296 relocated = data ? dr : ir;
298 /* Resolve real address if translation turned on */
300 page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data);
302 pte.may_execute = true;
304 pte.may_write = true;
305 pte.raddr = eaddr & KVM_PAM;
307 pte.vpage = eaddr >> 12;
310 switch (vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) {
312 pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
316 vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
318 if ((vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) == MSR_DR)
319 pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
321 pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
325 page_found = -EINVAL;
329 if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
330 (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
332 * If we do the dcbz hack, we have to NX on every execution,
333 * so we can patch the executing code. This renders our guest
336 pte.may_execute = !data;
339 if (page_found == -ENOENT) {
340 /* Page not found in guest PTE entries */
341 struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
342 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
343 vcpu->arch.shared->dsisr = svcpu->fault_dsisr;
344 vcpu->arch.shared->msr |=
345 (svcpu->shadow_srr1 & 0x00000000f8000000ULL);
347 kvmppc_book3s_queue_irqprio(vcpu, vec);
348 } else if (page_found == -EPERM) {
349 /* Storage protection */
350 struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
351 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
352 vcpu->arch.shared->dsisr = svcpu->fault_dsisr & ~DSISR_NOHPTE;
353 vcpu->arch.shared->dsisr |= DSISR_PROTFAULT;
354 vcpu->arch.shared->msr |=
355 svcpu->shadow_srr1 & 0x00000000f8000000ULL;
357 kvmppc_book3s_queue_irqprio(vcpu, vec);
358 } else if (page_found == -EINVAL) {
359 /* Page not found in guest SLB */
360 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
361 kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
362 } else if (!is_mmio &&
363 kvmppc_visible_gfn(vcpu, pte.raddr >> PAGE_SHIFT)) {
364 /* The guest's PTE is not mapped yet. Map on the host */
365 kvmppc_mmu_map_page(vcpu, &pte);
367 vcpu->stat.sp_storage++;
368 else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
369 (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
370 kvmppc_patch_dcbz(vcpu, &pte);
373 vcpu->stat.mmio_exits++;
374 vcpu->arch.paddr_accessed = pte.raddr;
375 vcpu->arch.vaddr_accessed = pte.eaddr;
376 r = kvmppc_emulate_mmio(run, vcpu);
377 if ( r == RESUME_HOST_NV )
384 static inline int get_fpr_index(int i)
392 /* Give up external provider (FPU, Altivec, VSX) */
393 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
395 struct thread_struct *t = ¤t->thread;
396 u64 *vcpu_fpr = vcpu->arch.fpr;
398 u64 *vcpu_vsx = vcpu->arch.vsr;
400 u64 *thread_fpr = (u64*)t->fpr;
403 if (!(vcpu->arch.guest_owned_ext & msr))
407 printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
413 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
414 vcpu_fpr[i] = thread_fpr[get_fpr_index(i)];
416 vcpu->arch.fpscr = t->fpscr.val;
419 #ifdef CONFIG_ALTIVEC
420 giveup_altivec(current);
421 memcpy(vcpu->arch.vr, t->vr, sizeof(vcpu->arch.vr));
422 vcpu->arch.vscr = t->vscr;
427 __giveup_vsx(current);
428 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++)
429 vcpu_vsx[i] = thread_fpr[get_fpr_index(i) + 1];
436 vcpu->arch.guest_owned_ext &= ~msr;
437 current->thread.regs->msr &= ~msr;
438 kvmppc_recalc_shadow_msr(vcpu);
441 static int kvmppc_read_inst(struct kvm_vcpu *vcpu)
443 ulong srr0 = kvmppc_get_pc(vcpu);
444 u32 last_inst = kvmppc_get_last_inst(vcpu);
447 ret = kvmppc_ld(vcpu, &srr0, sizeof(u32), &last_inst, false);
448 if (ret == -ENOENT) {
449 ulong msr = vcpu->arch.shared->msr;
451 msr = kvmppc_set_field(msr, 33, 33, 1);
452 msr = kvmppc_set_field(msr, 34, 36, 0);
453 vcpu->arch.shared->msr = kvmppc_set_field(msr, 42, 47, 0);
454 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_INST_STORAGE);
455 return EMULATE_AGAIN;
461 static int kvmppc_check_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr)
464 /* Need to do paired single emulation? */
465 if (!(vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE))
468 /* Read out the instruction */
469 if (kvmppc_read_inst(vcpu) == EMULATE_DONE)
470 /* Need to emulate */
473 return EMULATE_AGAIN;
476 /* Handle external providers (FPU, Altivec, VSX) */
477 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
480 struct thread_struct *t = ¤t->thread;
481 u64 *vcpu_fpr = vcpu->arch.fpr;
483 u64 *vcpu_vsx = vcpu->arch.vsr;
485 u64 *thread_fpr = (u64*)t->fpr;
488 /* When we have paired singles, we emulate in software */
489 if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
492 if (!(vcpu->arch.shared->msr & msr)) {
493 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
497 /* We already own the ext */
498 if (vcpu->arch.guest_owned_ext & msr) {
503 printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
506 current->thread.regs->msr |= msr;
510 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
511 thread_fpr[get_fpr_index(i)] = vcpu_fpr[i];
513 t->fpscr.val = vcpu->arch.fpscr;
515 kvmppc_load_up_fpu();
518 #ifdef CONFIG_ALTIVEC
519 memcpy(t->vr, vcpu->arch.vr, sizeof(vcpu->arch.vr));
520 t->vscr = vcpu->arch.vscr;
522 kvmppc_load_up_altivec();
527 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++)
528 thread_fpr[get_fpr_index(i) + 1] = vcpu_vsx[i];
529 kvmppc_load_up_vsx();
536 vcpu->arch.guest_owned_ext |= msr;
538 kvmppc_recalc_shadow_msr(vcpu);
543 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
544 unsigned int exit_nr)
548 vcpu->stat.sum_exits++;
550 run->exit_reason = KVM_EXIT_UNKNOWN;
551 run->ready_for_interrupt_injection = 1;
553 /* We get here with MSR.EE=0, so enable it to be a nice citizen */
556 trace_kvm_exit(exit_nr, vcpu);
560 case BOOK3S_INTERRUPT_INST_STORAGE:
562 struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
563 ulong shadow_srr1 = svcpu->shadow_srr1;
564 vcpu->stat.pf_instruc++;
566 #ifdef CONFIG_PPC_BOOK3S_32
567 /* We set segments as unused segments when invalidating them. So
568 * treat the respective fault as segment fault. */
569 if (svcpu->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT] == SR_INVALID) {
570 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
578 /* only care about PTEG not found errors, but leave NX alone */
579 if (shadow_srr1 & 0x40000000) {
580 r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr);
581 vcpu->stat.sp_instruc++;
582 } else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
583 (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
585 * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
586 * so we can't use the NX bit inside the guest. Let's cross our fingers,
587 * that no guest that needs the dcbz hack does NX.
589 kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
592 vcpu->arch.shared->msr |= shadow_srr1 & 0x58000000;
593 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
598 case BOOK3S_INTERRUPT_DATA_STORAGE:
600 ulong dar = kvmppc_get_fault_dar(vcpu);
601 struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
602 u32 fault_dsisr = svcpu->fault_dsisr;
603 vcpu->stat.pf_storage++;
605 #ifdef CONFIG_PPC_BOOK3S_32
606 /* We set segments as unused segments when invalidating them. So
607 * treat the respective fault as segment fault. */
608 if ((svcpu->sr[dar >> SID_SHIFT]) == SR_INVALID) {
609 kvmppc_mmu_map_segment(vcpu, dar);
617 /* The only case we need to handle is missing shadow PTEs */
618 if (fault_dsisr & DSISR_NOHPTE) {
619 r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr);
621 vcpu->arch.shared->dar = dar;
622 vcpu->arch.shared->dsisr = fault_dsisr;
623 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
628 case BOOK3S_INTERRUPT_DATA_SEGMENT:
629 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
630 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
631 kvmppc_book3s_queue_irqprio(vcpu,
632 BOOK3S_INTERRUPT_DATA_SEGMENT);
636 case BOOK3S_INTERRUPT_INST_SEGMENT:
637 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
638 kvmppc_book3s_queue_irqprio(vcpu,
639 BOOK3S_INTERRUPT_INST_SEGMENT);
643 /* We're good on these - the host merely wanted to get our attention */
644 case BOOK3S_INTERRUPT_DECREMENTER:
645 case BOOK3S_INTERRUPT_HV_DECREMENTER:
646 vcpu->stat.dec_exits++;
649 case BOOK3S_INTERRUPT_EXTERNAL:
650 case BOOK3S_INTERRUPT_EXTERNAL_LEVEL:
651 case BOOK3S_INTERRUPT_EXTERNAL_HV:
652 vcpu->stat.ext_intr_exits++;
655 case BOOK3S_INTERRUPT_PERFMON:
658 case BOOK3S_INTERRUPT_PROGRAM:
659 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
661 enum emulation_result er;
662 struct kvmppc_book3s_shadow_vcpu *svcpu;
666 svcpu = svcpu_get(vcpu);
667 flags = svcpu->shadow_srr1 & 0x1f0000ull;
670 if (vcpu->arch.shared->msr & MSR_PR) {
672 printk(KERN_INFO "Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
674 if ((kvmppc_get_last_inst(vcpu) & 0xff0007ff) !=
675 (INS_DCBZ & 0xfffffff7)) {
676 kvmppc_core_queue_program(vcpu, flags);
682 vcpu->stat.emulated_inst_exits++;
683 er = kvmppc_emulate_instruction(run, vcpu);
692 printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
693 __func__, kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
694 kvmppc_core_queue_program(vcpu, flags);
697 case EMULATE_DO_MMIO:
698 run->exit_reason = KVM_EXIT_MMIO;
706 case BOOK3S_INTERRUPT_SYSCALL:
707 if (vcpu->arch.papr_enabled &&
708 (kvmppc_get_last_inst(vcpu) == 0x44000022) &&
709 !(vcpu->arch.shared->msr & MSR_PR)) {
710 /* SC 1 papr hypercalls */
711 ulong cmd = kvmppc_get_gpr(vcpu, 3);
714 #ifdef CONFIG_KVM_BOOK3S_64_PR
715 if (kvmppc_h_pr(vcpu, cmd) == EMULATE_DONE) {
721 run->papr_hcall.nr = cmd;
722 for (i = 0; i < 9; ++i) {
723 ulong gpr = kvmppc_get_gpr(vcpu, 4 + i);
724 run->papr_hcall.args[i] = gpr;
726 run->exit_reason = KVM_EXIT_PAPR_HCALL;
727 vcpu->arch.hcall_needed = 1;
729 } else if (vcpu->arch.osi_enabled &&
730 (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
731 (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
733 u64 *gprs = run->osi.gprs;
736 run->exit_reason = KVM_EXIT_OSI;
737 for (i = 0; i < 32; i++)
738 gprs[i] = kvmppc_get_gpr(vcpu, i);
739 vcpu->arch.osi_needed = 1;
741 } else if (!(vcpu->arch.shared->msr & MSR_PR) &&
742 (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
743 /* KVM PV hypercalls */
744 kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
748 vcpu->stat.syscall_exits++;
749 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
753 case BOOK3S_INTERRUPT_FP_UNAVAIL:
754 case BOOK3S_INTERRUPT_ALTIVEC:
755 case BOOK3S_INTERRUPT_VSX:
760 case BOOK3S_INTERRUPT_FP_UNAVAIL: ext_msr = MSR_FP; break;
761 case BOOK3S_INTERRUPT_ALTIVEC: ext_msr = MSR_VEC; break;
762 case BOOK3S_INTERRUPT_VSX: ext_msr = MSR_VSX; break;
765 switch (kvmppc_check_ext(vcpu, exit_nr)) {
767 /* everything ok - let's enable the ext */
768 r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
771 /* we need to emulate this instruction */
772 goto program_interrupt;
775 /* nothing to worry about - go again */
780 case BOOK3S_INTERRUPT_ALIGNMENT:
781 if (kvmppc_read_inst(vcpu) == EMULATE_DONE) {
782 vcpu->arch.shared->dsisr = kvmppc_alignment_dsisr(vcpu,
783 kvmppc_get_last_inst(vcpu));
784 vcpu->arch.shared->dar = kvmppc_alignment_dar(vcpu,
785 kvmppc_get_last_inst(vcpu));
786 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
790 case BOOK3S_INTERRUPT_MACHINE_CHECK:
791 case BOOK3S_INTERRUPT_TRACE:
792 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
797 struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
798 ulong shadow_srr1 = svcpu->shadow_srr1;
800 /* Ugh - bork here! What did we get? */
801 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
802 exit_nr, kvmppc_get_pc(vcpu), shadow_srr1);
810 if (!(r & RESUME_HOST)) {
811 /* To avoid clobbering exit_reason, only check for signals if
812 * we aren't already exiting to userspace for some other
816 * Interrupts could be timers for the guest which we have to
817 * inject again, so let's postpone them until we're in the guest
818 * and if we really did time things so badly, then we just exit
819 * again due to a host external interrupt.
821 __hard_irq_disable();
822 if (kvmppc_prepare_to_enter(vcpu)) {
823 run->exit_reason = KVM_EXIT_INTR;
828 trace_kvm_book3s_reenter(r, vcpu);
833 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
834 struct kvm_sregs *sregs)
836 struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
839 sregs->pvr = vcpu->arch.pvr;
841 sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
842 if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
843 for (i = 0; i < 64; i++) {
844 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige | i;
845 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
848 for (i = 0; i < 16; i++)
849 sregs->u.s.ppc32.sr[i] = vcpu->arch.shared->sr[i];
851 for (i = 0; i < 8; i++) {
852 sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
853 sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
860 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
861 struct kvm_sregs *sregs)
863 struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
866 kvmppc_set_pvr(vcpu, sregs->pvr);
868 vcpu3s->sdr1 = sregs->u.s.sdr1;
869 if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
870 for (i = 0; i < 64; i++) {
871 vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
872 sregs->u.s.ppc64.slb[i].slbe);
875 for (i = 0; i < 16; i++) {
876 vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
878 for (i = 0; i < 8; i++) {
879 kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
880 (u32)sregs->u.s.ppc32.ibat[i]);
881 kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
882 (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
883 kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
884 (u32)sregs->u.s.ppc32.dbat[i]);
885 kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
886 (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
890 /* Flush the MMU after messing with the segments */
891 kvmppc_mmu_pte_flush(vcpu, 0, 0);
896 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
901 case KVM_REG_PPC_HIOR:
902 r = copy_to_user((u64 __user *)(long)reg->addr,
903 &to_book3s(vcpu)->hior, sizeof(u64));
912 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
917 case KVM_REG_PPC_HIOR:
918 r = copy_from_user(&to_book3s(vcpu)->hior,
919 (u64 __user *)(long)reg->addr, sizeof(u64));
921 to_book3s(vcpu)->hior_explicit = true;
930 int kvmppc_core_check_processor_compat(void)
935 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
937 struct kvmppc_vcpu_book3s *vcpu_book3s;
938 struct kvm_vcpu *vcpu;
942 vcpu_book3s = vzalloc(sizeof(struct kvmppc_vcpu_book3s));
946 vcpu_book3s->shadow_vcpu = (struct kvmppc_book3s_shadow_vcpu *)
947 kzalloc(sizeof(*vcpu_book3s->shadow_vcpu), GFP_KERNEL);
948 if (!vcpu_book3s->shadow_vcpu)
951 vcpu = &vcpu_book3s->vcpu;
952 err = kvm_vcpu_init(vcpu, kvm, id);
954 goto free_shadow_vcpu;
956 p = __get_free_page(GFP_KERNEL|__GFP_ZERO);
957 /* the real shared page fills the last 4k of our page */
958 vcpu->arch.shared = (void*)(p + PAGE_SIZE - 4096);
962 #ifdef CONFIG_PPC_BOOK3S_64
963 /* default to book3s_64 (970fx) */
964 vcpu->arch.pvr = 0x3C0301;
966 /* default to book3s_32 (750) */
967 vcpu->arch.pvr = 0x84202;
969 kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
970 vcpu->arch.slb_nr = 64;
972 vcpu->arch.shadow_msr = MSR_USER64;
974 err = kvmppc_mmu_init(vcpu);
981 kvm_vcpu_uninit(vcpu);
983 kfree(vcpu_book3s->shadow_vcpu);
990 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
992 struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
994 free_page((unsigned long)vcpu->arch.shared & PAGE_MASK);
995 kvm_vcpu_uninit(vcpu);
996 kfree(vcpu_book3s->shadow_vcpu);
1000 int kvmppc_vcpu_run(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1003 double fpr[32][TS_FPRWIDTH];
1006 #ifdef CONFIG_ALTIVEC
1009 unsigned long uninitialized_var(vrsave);
1019 /* Check if we can run the vcpu at all */
1020 if (!vcpu->arch.sane) {
1021 kvm_run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1027 * Interrupts could be timers for the guest which we have to inject
1028 * again, so let's postpone them until we're in the guest and if we
1029 * really did time things so badly, then we just exit again due to
1030 * a host external interrupt.
1032 __hard_irq_disable();
1033 if (kvmppc_prepare_to_enter(vcpu)) {
1034 __hard_irq_enable();
1035 kvm_run->exit_reason = KVM_EXIT_INTR;
1040 /* Save FPU state in stack */
1041 if (current->thread.regs->msr & MSR_FP)
1042 giveup_fpu(current);
1043 memcpy(fpr, current->thread.fpr, sizeof(current->thread.fpr));
1044 fpscr = current->thread.fpscr.val;
1045 fpexc_mode = current->thread.fpexc_mode;
1047 #ifdef CONFIG_ALTIVEC
1048 /* Save Altivec state in stack */
1049 used_vr = current->thread.used_vr;
1051 if (current->thread.regs->msr & MSR_VEC)
1052 giveup_altivec(current);
1053 memcpy(vr, current->thread.vr, sizeof(current->thread.vr));
1054 vscr = current->thread.vscr;
1055 vrsave = current->thread.vrsave;
1060 /* Save VSX state in stack */
1061 used_vsr = current->thread.used_vsr;
1062 if (used_vsr && (current->thread.regs->msr & MSR_VSX))
1063 __giveup_vsx(current);
1066 /* Remember the MSR with disabled extensions */
1067 ext_msr = current->thread.regs->msr;
1069 /* Preload FPU if it's enabled */
1070 if (vcpu->arch.shared->msr & MSR_FP)
1071 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
1075 ret = __kvmppc_vcpu_run(kvm_run, vcpu);
1079 current->thread.regs->msr = ext_msr;
1081 /* Make sure we save the guest FPU/Altivec/VSX state */
1082 kvmppc_giveup_ext(vcpu, MSR_FP);
1083 kvmppc_giveup_ext(vcpu, MSR_VEC);
1084 kvmppc_giveup_ext(vcpu, MSR_VSX);
1086 /* Restore FPU state from stack */
1087 memcpy(current->thread.fpr, fpr, sizeof(current->thread.fpr));
1088 current->thread.fpscr.val = fpscr;
1089 current->thread.fpexc_mode = fpexc_mode;
1091 #ifdef CONFIG_ALTIVEC
1092 /* Restore Altivec state from stack */
1093 if (used_vr && current->thread.used_vr) {
1094 memcpy(current->thread.vr, vr, sizeof(current->thread.vr));
1095 current->thread.vscr = vscr;
1096 current->thread.vrsave = vrsave;
1098 current->thread.used_vr = used_vr;
1102 current->thread.used_vsr = used_vsr;
1111 * Get (and clear) the dirty memory log for a memory slot.
1113 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1114 struct kvm_dirty_log *log)
1116 struct kvm_memory_slot *memslot;
1117 struct kvm_vcpu *vcpu;
1123 mutex_lock(&kvm->slots_lock);
1125 r = kvm_get_dirty_log(kvm, log, &is_dirty);
1129 /* If nothing is dirty, don't bother messing with page tables. */
1131 memslot = id_to_memslot(kvm->memslots, log->slot);
1133 ga = memslot->base_gfn << PAGE_SHIFT;
1134 ga_end = ga + (memslot->npages << PAGE_SHIFT);
1136 kvm_for_each_vcpu(n, vcpu, kvm)
1137 kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
1139 n = kvm_dirty_bitmap_bytes(memslot);
1140 memset(memslot->dirty_bitmap, 0, n);
1145 mutex_unlock(&kvm->slots_lock);
1150 int kvm_vm_ioctl_get_smmu_info(struct kvm *kvm, struct kvm_ppc_smmu_info *info)
1155 /* SLB is always 64 entries */
1156 info->slb_size = 64;
1158 /* Standard 4k base page size segment */
1159 info->sps[0].page_shift = 12;
1160 info->sps[0].slb_enc = 0;
1161 info->sps[0].enc[0].page_shift = 12;
1162 info->sps[0].enc[0].pte_enc = 0;
1164 /* Standard 16M large page size segment */
1165 info->sps[1].page_shift = 24;
1166 info->sps[1].slb_enc = SLB_VSID_L;
1167 info->sps[1].enc[0].page_shift = 24;
1168 info->sps[1].enc[0].pte_enc = 0;
1172 #endif /* CONFIG_PPC64 */
1174 int kvmppc_core_prepare_memory_region(struct kvm *kvm,
1175 struct kvm_userspace_memory_region *mem)
1180 void kvmppc_core_commit_memory_region(struct kvm *kvm,
1181 struct kvm_userspace_memory_region *mem)
1185 int kvmppc_core_init_vm(struct kvm *kvm)
1188 INIT_LIST_HEAD(&kvm->arch.spapr_tce_tables);
1194 void kvmppc_core_destroy_vm(struct kvm *kvm)
1197 WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
1201 static int kvmppc_book3s_init(void)
1205 r = kvm_init(NULL, sizeof(struct kvmppc_vcpu_book3s), 0,
1211 r = kvmppc_mmu_hpte_sysinit();
1216 static void kvmppc_book3s_exit(void)
1218 kvmppc_mmu_hpte_sysexit();
1222 module_init(kvmppc_book3s_init);
1223 module_exit(kvmppc_book3s_exit);