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 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu *vcpu)
93 ulong smsr = vcpu->arch.shared->msr;
95 /* Guest MSR values */
96 smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE | MSR_DE;
97 /* Process MSR values */
98 smsr |= MSR_ME | MSR_RI | MSR_IR | MSR_DR | MSR_PR | MSR_EE;
99 /* External providers the guest reserved */
100 smsr |= (vcpu->arch.shared->msr & vcpu->arch.guest_owned_ext);
101 /* 64-bit Process MSR values */
102 #ifdef CONFIG_PPC_BOOK3S_64
103 smsr |= MSR_ISF | MSR_HV;
105 vcpu->arch.shadow_msr = smsr;
108 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
110 ulong old_msr = vcpu->arch.shared->msr;
113 printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
116 msr &= to_book3s(vcpu)->msr_mask;
117 vcpu->arch.shared->msr = msr;
118 kvmppc_recalc_shadow_msr(vcpu);
121 if (!vcpu->arch.pending_exceptions) {
122 kvm_vcpu_block(vcpu);
123 vcpu->stat.halt_wakeup++;
125 /* Unset POW bit after we woke up */
127 vcpu->arch.shared->msr = msr;
131 if ((vcpu->arch.shared->msr & (MSR_PR|MSR_IR|MSR_DR)) !=
132 (old_msr & (MSR_PR|MSR_IR|MSR_DR))) {
133 kvmppc_mmu_flush_segments(vcpu);
134 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
136 /* Preload magic page segment when in kernel mode */
137 if (!(msr & MSR_PR) && vcpu->arch.magic_page_pa) {
138 struct kvm_vcpu_arch *a = &vcpu->arch;
141 kvmppc_mmu_map_segment(vcpu, a->magic_page_ea);
143 kvmppc_mmu_map_segment(vcpu, a->magic_page_pa);
147 /* Preload FPU if it's enabled */
148 if (vcpu->arch.shared->msr & MSR_FP)
149 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
152 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
156 vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
157 vcpu->arch.pvr = pvr;
158 #ifdef CONFIG_PPC_BOOK3S_64
159 if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
160 kvmppc_mmu_book3s_64_init(vcpu);
161 if (!to_book3s(vcpu)->hior_explicit)
162 to_book3s(vcpu)->hior = 0xfff00000;
163 to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
164 vcpu->arch.cpu_type = KVM_CPU_3S_64;
168 kvmppc_mmu_book3s_32_init(vcpu);
169 if (!to_book3s(vcpu)->hior_explicit)
170 to_book3s(vcpu)->hior = 0;
171 to_book3s(vcpu)->msr_mask = 0xffffffffULL;
172 vcpu->arch.cpu_type = KVM_CPU_3S_32;
175 kvmppc_sanity_check(vcpu);
177 /* If we are in hypervisor level on 970, we can tell the CPU to
178 * treat DCBZ as 32 bytes store */
179 vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
180 if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
181 !strcmp(cur_cpu_spec->platform, "ppc970"))
182 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
184 /* Cell performs badly if MSR_FEx are set. So let's hope nobody
185 really needs them in a VM on Cell and force disable them. */
186 if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be"))
187 to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1);
189 #ifdef CONFIG_PPC_BOOK3S_32
190 /* 32 bit Book3S always has 32 byte dcbz */
191 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
194 /* On some CPUs we can execute paired single operations natively */
195 asm ( "mfpvr %0" : "=r"(host_pvr));
197 case 0x00080200: /* lonestar 2.0 */
198 case 0x00088202: /* lonestar 2.2 */
199 case 0x70000100: /* gekko 1.0 */
200 case 0x00080100: /* gekko 2.0 */
201 case 0x00083203: /* gekko 2.3a */
202 case 0x00083213: /* gekko 2.3b */
203 case 0x00083204: /* gekko 2.4 */
204 case 0x00083214: /* gekko 2.4e (8SE) - retail HW2 */
205 case 0x00087200: /* broadway */
206 vcpu->arch.hflags |= BOOK3S_HFLAG_NATIVE_PS;
207 /* Enable HID2.PSE - in case we need it later */
208 mtspr(SPRN_HID2_GEKKO, mfspr(SPRN_HID2_GEKKO) | (1 << 29));
212 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
213 * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
214 * emulate 32 bytes dcbz length.
216 * The Book3s_64 inventors also realized this case and implemented a special bit
217 * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
219 * My approach here is to patch the dcbz instruction on executing pages.
221 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
228 hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
229 if (is_error_page(hpage)) {
230 kvm_release_page_clean(hpage);
234 hpage_offset = pte->raddr & ~PAGE_MASK;
235 hpage_offset &= ~0xFFFULL;
239 page = kmap_atomic(hpage);
241 /* patch dcbz into reserved instruction, so we trap */
242 for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
243 if ((page[i] & 0xff0007ff) == INS_DCBZ)
244 page[i] &= 0xfffffff7;
250 static int kvmppc_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
252 ulong mp_pa = vcpu->arch.magic_page_pa;
254 if (unlikely(mp_pa) &&
255 unlikely((mp_pa & KVM_PAM) >> PAGE_SHIFT == gfn)) {
259 return kvm_is_visible_gfn(vcpu->kvm, gfn);
262 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
263 ulong eaddr, int vec)
265 bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
266 int r = RESUME_GUEST;
269 struct kvmppc_pte pte;
270 bool is_mmio = false;
271 bool dr = (vcpu->arch.shared->msr & MSR_DR) ? true : false;
272 bool ir = (vcpu->arch.shared->msr & MSR_IR) ? true : false;
275 relocated = data ? dr : ir;
277 /* Resolve real address if translation turned on */
279 page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data);
281 pte.may_execute = true;
283 pte.may_write = true;
284 pte.raddr = eaddr & KVM_PAM;
286 pte.vpage = eaddr >> 12;
289 switch (vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) {
291 pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
295 vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
297 if ((vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) == MSR_DR)
298 pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
300 pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
304 page_found = -EINVAL;
308 if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
309 (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
311 * If we do the dcbz hack, we have to NX on every execution,
312 * so we can patch the executing code. This renders our guest
315 pte.may_execute = !data;
318 if (page_found == -ENOENT) {
319 /* Page not found in guest PTE entries */
320 struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
321 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
322 vcpu->arch.shared->dsisr = svcpu->fault_dsisr;
323 vcpu->arch.shared->msr |=
324 (svcpu->shadow_srr1 & 0x00000000f8000000ULL);
326 kvmppc_book3s_queue_irqprio(vcpu, vec);
327 } else if (page_found == -EPERM) {
328 /* Storage protection */
329 struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
330 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
331 vcpu->arch.shared->dsisr = svcpu->fault_dsisr & ~DSISR_NOHPTE;
332 vcpu->arch.shared->dsisr |= DSISR_PROTFAULT;
333 vcpu->arch.shared->msr |=
334 svcpu->shadow_srr1 & 0x00000000f8000000ULL;
336 kvmppc_book3s_queue_irqprio(vcpu, vec);
337 } else if (page_found == -EINVAL) {
338 /* Page not found in guest SLB */
339 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
340 kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
341 } else if (!is_mmio &&
342 kvmppc_visible_gfn(vcpu, pte.raddr >> PAGE_SHIFT)) {
343 /* The guest's PTE is not mapped yet. Map on the host */
344 kvmppc_mmu_map_page(vcpu, &pte);
346 vcpu->stat.sp_storage++;
347 else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
348 (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
349 kvmppc_patch_dcbz(vcpu, &pte);
352 vcpu->stat.mmio_exits++;
353 vcpu->arch.paddr_accessed = pte.raddr;
354 r = kvmppc_emulate_mmio(run, vcpu);
355 if ( r == RESUME_HOST_NV )
362 static inline int get_fpr_index(int i)
370 /* Give up external provider (FPU, Altivec, VSX) */
371 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
373 struct thread_struct *t = ¤t->thread;
374 u64 *vcpu_fpr = vcpu->arch.fpr;
376 u64 *vcpu_vsx = vcpu->arch.vsr;
378 u64 *thread_fpr = (u64*)t->fpr;
381 if (!(vcpu->arch.guest_owned_ext & msr))
385 printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
391 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
392 vcpu_fpr[i] = thread_fpr[get_fpr_index(i)];
394 vcpu->arch.fpscr = t->fpscr.val;
397 #ifdef CONFIG_ALTIVEC
398 giveup_altivec(current);
399 memcpy(vcpu->arch.vr, t->vr, sizeof(vcpu->arch.vr));
400 vcpu->arch.vscr = t->vscr;
405 __giveup_vsx(current);
406 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++)
407 vcpu_vsx[i] = thread_fpr[get_fpr_index(i) + 1];
414 vcpu->arch.guest_owned_ext &= ~msr;
415 current->thread.regs->msr &= ~msr;
416 kvmppc_recalc_shadow_msr(vcpu);
419 static int kvmppc_read_inst(struct kvm_vcpu *vcpu)
421 ulong srr0 = kvmppc_get_pc(vcpu);
422 u32 last_inst = kvmppc_get_last_inst(vcpu);
425 ret = kvmppc_ld(vcpu, &srr0, sizeof(u32), &last_inst, false);
426 if (ret == -ENOENT) {
427 ulong msr = vcpu->arch.shared->msr;
429 msr = kvmppc_set_field(msr, 33, 33, 1);
430 msr = kvmppc_set_field(msr, 34, 36, 0);
431 vcpu->arch.shared->msr = kvmppc_set_field(msr, 42, 47, 0);
432 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_INST_STORAGE);
433 return EMULATE_AGAIN;
439 static int kvmppc_check_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr)
442 /* Need to do paired single emulation? */
443 if (!(vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE))
446 /* Read out the instruction */
447 if (kvmppc_read_inst(vcpu) == EMULATE_DONE)
448 /* Need to emulate */
451 return EMULATE_AGAIN;
454 /* Handle external providers (FPU, Altivec, VSX) */
455 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
458 struct thread_struct *t = ¤t->thread;
459 u64 *vcpu_fpr = vcpu->arch.fpr;
461 u64 *vcpu_vsx = vcpu->arch.vsr;
463 u64 *thread_fpr = (u64*)t->fpr;
466 /* When we have paired singles, we emulate in software */
467 if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
470 if (!(vcpu->arch.shared->msr & msr)) {
471 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
475 /* We already own the ext */
476 if (vcpu->arch.guest_owned_ext & msr) {
481 printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
484 current->thread.regs->msr |= msr;
488 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
489 thread_fpr[get_fpr_index(i)] = vcpu_fpr[i];
491 t->fpscr.val = vcpu->arch.fpscr;
493 kvmppc_load_up_fpu();
496 #ifdef CONFIG_ALTIVEC
497 memcpy(t->vr, vcpu->arch.vr, sizeof(vcpu->arch.vr));
498 t->vscr = vcpu->arch.vscr;
500 kvmppc_load_up_altivec();
505 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++)
506 thread_fpr[get_fpr_index(i) + 1] = vcpu_vsx[i];
507 kvmppc_load_up_vsx();
514 vcpu->arch.guest_owned_ext |= msr;
516 kvmppc_recalc_shadow_msr(vcpu);
521 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
522 unsigned int exit_nr)
526 vcpu->stat.sum_exits++;
528 run->exit_reason = KVM_EXIT_UNKNOWN;
529 run->ready_for_interrupt_injection = 1;
531 trace_kvm_book3s_exit(exit_nr, vcpu);
535 case BOOK3S_INTERRUPT_INST_STORAGE:
537 struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
538 ulong shadow_srr1 = svcpu->shadow_srr1;
539 vcpu->stat.pf_instruc++;
541 #ifdef CONFIG_PPC_BOOK3S_32
542 /* We set segments as unused segments when invalidating them. So
543 * treat the respective fault as segment fault. */
544 if (svcpu->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT] == SR_INVALID) {
545 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
553 /* only care about PTEG not found errors, but leave NX alone */
554 if (shadow_srr1 & 0x40000000) {
555 r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr);
556 vcpu->stat.sp_instruc++;
557 } else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
558 (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
560 * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
561 * so we can't use the NX bit inside the guest. Let's cross our fingers,
562 * that no guest that needs the dcbz hack does NX.
564 kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
567 vcpu->arch.shared->msr |= shadow_srr1 & 0x58000000;
568 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
573 case BOOK3S_INTERRUPT_DATA_STORAGE:
575 ulong dar = kvmppc_get_fault_dar(vcpu);
576 struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
577 u32 fault_dsisr = svcpu->fault_dsisr;
578 vcpu->stat.pf_storage++;
580 #ifdef CONFIG_PPC_BOOK3S_32
581 /* We set segments as unused segments when invalidating them. So
582 * treat the respective fault as segment fault. */
583 if ((svcpu->sr[dar >> SID_SHIFT]) == SR_INVALID) {
584 kvmppc_mmu_map_segment(vcpu, dar);
592 /* The only case we need to handle is missing shadow PTEs */
593 if (fault_dsisr & DSISR_NOHPTE) {
594 r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr);
596 vcpu->arch.shared->dar = dar;
597 vcpu->arch.shared->dsisr = fault_dsisr;
598 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
603 case BOOK3S_INTERRUPT_DATA_SEGMENT:
604 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
605 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
606 kvmppc_book3s_queue_irqprio(vcpu,
607 BOOK3S_INTERRUPT_DATA_SEGMENT);
611 case BOOK3S_INTERRUPT_INST_SEGMENT:
612 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
613 kvmppc_book3s_queue_irqprio(vcpu,
614 BOOK3S_INTERRUPT_INST_SEGMENT);
618 /* We're good on these - the host merely wanted to get our attention */
619 case BOOK3S_INTERRUPT_DECREMENTER:
620 vcpu->stat.dec_exits++;
623 case BOOK3S_INTERRUPT_EXTERNAL:
624 vcpu->stat.ext_intr_exits++;
627 case BOOK3S_INTERRUPT_PERFMON:
630 case BOOK3S_INTERRUPT_PROGRAM:
632 enum emulation_result er;
633 struct kvmppc_book3s_shadow_vcpu *svcpu;
637 svcpu = svcpu_get(vcpu);
638 flags = svcpu->shadow_srr1 & 0x1f0000ull;
641 if (vcpu->arch.shared->msr & MSR_PR) {
643 printk(KERN_INFO "Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
645 if ((kvmppc_get_last_inst(vcpu) & 0xff0007ff) !=
646 (INS_DCBZ & 0xfffffff7)) {
647 kvmppc_core_queue_program(vcpu, flags);
653 vcpu->stat.emulated_inst_exits++;
654 er = kvmppc_emulate_instruction(run, vcpu);
663 printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
664 __func__, kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
665 kvmppc_core_queue_program(vcpu, flags);
668 case EMULATE_DO_MMIO:
669 run->exit_reason = KVM_EXIT_MMIO;
677 case BOOK3S_INTERRUPT_SYSCALL:
678 if (vcpu->arch.papr_enabled &&
679 (kvmppc_get_last_inst(vcpu) == 0x44000022) &&
680 !(vcpu->arch.shared->msr & MSR_PR)) {
681 /* SC 1 papr hypercalls */
682 ulong cmd = kvmppc_get_gpr(vcpu, 3);
685 #ifdef CONFIG_KVM_BOOK3S_64_PR
686 if (kvmppc_h_pr(vcpu, cmd) == EMULATE_DONE) {
692 run->papr_hcall.nr = cmd;
693 for (i = 0; i < 9; ++i) {
694 ulong gpr = kvmppc_get_gpr(vcpu, 4 + i);
695 run->papr_hcall.args[i] = gpr;
697 run->exit_reason = KVM_EXIT_PAPR_HCALL;
698 vcpu->arch.hcall_needed = 1;
700 } else if (vcpu->arch.osi_enabled &&
701 (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
702 (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
704 u64 *gprs = run->osi.gprs;
707 run->exit_reason = KVM_EXIT_OSI;
708 for (i = 0; i < 32; i++)
709 gprs[i] = kvmppc_get_gpr(vcpu, i);
710 vcpu->arch.osi_needed = 1;
712 } else if (!(vcpu->arch.shared->msr & MSR_PR) &&
713 (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
714 /* KVM PV hypercalls */
715 kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
719 vcpu->stat.syscall_exits++;
720 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
724 case BOOK3S_INTERRUPT_FP_UNAVAIL:
725 case BOOK3S_INTERRUPT_ALTIVEC:
726 case BOOK3S_INTERRUPT_VSX:
731 case BOOK3S_INTERRUPT_FP_UNAVAIL: ext_msr = MSR_FP; break;
732 case BOOK3S_INTERRUPT_ALTIVEC: ext_msr = MSR_VEC; break;
733 case BOOK3S_INTERRUPT_VSX: ext_msr = MSR_VSX; break;
736 switch (kvmppc_check_ext(vcpu, exit_nr)) {
738 /* everything ok - let's enable the ext */
739 r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
742 /* we need to emulate this instruction */
743 goto program_interrupt;
746 /* nothing to worry about - go again */
751 case BOOK3S_INTERRUPT_ALIGNMENT:
752 if (kvmppc_read_inst(vcpu) == EMULATE_DONE) {
753 vcpu->arch.shared->dsisr = kvmppc_alignment_dsisr(vcpu,
754 kvmppc_get_last_inst(vcpu));
755 vcpu->arch.shared->dar = kvmppc_alignment_dar(vcpu,
756 kvmppc_get_last_inst(vcpu));
757 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
761 case BOOK3S_INTERRUPT_MACHINE_CHECK:
762 case BOOK3S_INTERRUPT_TRACE:
763 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
768 struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
769 ulong shadow_srr1 = svcpu->shadow_srr1;
771 /* Ugh - bork here! What did we get? */
772 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
773 exit_nr, kvmppc_get_pc(vcpu), shadow_srr1);
781 if (!(r & RESUME_HOST)) {
782 /* To avoid clobbering exit_reason, only check for signals if
783 * we aren't already exiting to userspace for some other
787 * Interrupts could be timers for the guest which we have to
788 * inject again, so let's postpone them until we're in the guest
789 * and if we really did time things so badly, then we just exit
790 * again due to a host external interrupt.
792 __hard_irq_disable();
793 if (signal_pending(current)) {
796 printk(KERN_EMERG "KVM: Going back to host\n");
798 vcpu->stat.signal_exits++;
799 run->exit_reason = KVM_EXIT_INTR;
802 /* In case an interrupt came in that was triggered
803 * from userspace (like DEC), we need to check what
805 kvmppc_core_prepare_to_enter(vcpu);
809 trace_kvm_book3s_reenter(r, vcpu);
814 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
815 struct kvm_sregs *sregs)
817 struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
820 sregs->pvr = vcpu->arch.pvr;
822 sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
823 if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
824 for (i = 0; i < 64; i++) {
825 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige | i;
826 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
829 for (i = 0; i < 16; i++)
830 sregs->u.s.ppc32.sr[i] = vcpu->arch.shared->sr[i];
832 for (i = 0; i < 8; i++) {
833 sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
834 sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
841 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
842 struct kvm_sregs *sregs)
844 struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
847 kvmppc_set_pvr(vcpu, sregs->pvr);
849 vcpu3s->sdr1 = sregs->u.s.sdr1;
850 if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
851 for (i = 0; i < 64; i++) {
852 vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
853 sregs->u.s.ppc64.slb[i].slbe);
856 for (i = 0; i < 16; i++) {
857 vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
859 for (i = 0; i < 8; i++) {
860 kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
861 (u32)sregs->u.s.ppc32.ibat[i]);
862 kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
863 (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
864 kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
865 (u32)sregs->u.s.ppc32.dbat[i]);
866 kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
867 (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
871 /* Flush the MMU after messing with the segments */
872 kvmppc_mmu_pte_flush(vcpu, 0, 0);
877 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
882 case KVM_REG_PPC_HIOR:
883 r = copy_to_user((u64 __user *)(long)reg->addr,
884 &to_book3s(vcpu)->hior, sizeof(u64));
893 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
898 case KVM_REG_PPC_HIOR:
899 r = copy_from_user(&to_book3s(vcpu)->hior,
900 (u64 __user *)(long)reg->addr, sizeof(u64));
902 to_book3s(vcpu)->hior_explicit = true;
911 int kvmppc_core_check_processor_compat(void)
916 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
918 struct kvmppc_vcpu_book3s *vcpu_book3s;
919 struct kvm_vcpu *vcpu;
923 vcpu_book3s = vzalloc(sizeof(struct kvmppc_vcpu_book3s));
927 vcpu_book3s->shadow_vcpu = (struct kvmppc_book3s_shadow_vcpu *)
928 kzalloc(sizeof(*vcpu_book3s->shadow_vcpu), GFP_KERNEL);
929 if (!vcpu_book3s->shadow_vcpu)
932 vcpu = &vcpu_book3s->vcpu;
933 err = kvm_vcpu_init(vcpu, kvm, id);
935 goto free_shadow_vcpu;
937 p = __get_free_page(GFP_KERNEL|__GFP_ZERO);
938 /* the real shared page fills the last 4k of our page */
939 vcpu->arch.shared = (void*)(p + PAGE_SIZE - 4096);
943 #ifdef CONFIG_PPC_BOOK3S_64
944 /* default to book3s_64 (970fx) */
945 vcpu->arch.pvr = 0x3C0301;
947 /* default to book3s_32 (750) */
948 vcpu->arch.pvr = 0x84202;
950 kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
951 vcpu->arch.slb_nr = 64;
953 vcpu->arch.shadow_msr = MSR_USER64;
955 err = kvmppc_mmu_init(vcpu);
962 kvm_vcpu_uninit(vcpu);
964 kfree(vcpu_book3s->shadow_vcpu);
971 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
973 struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
975 free_page((unsigned long)vcpu->arch.shared & PAGE_MASK);
976 kvm_vcpu_uninit(vcpu);
977 kfree(vcpu_book3s->shadow_vcpu);
981 int kvmppc_vcpu_run(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
984 double fpr[32][TS_FPRWIDTH];
987 #ifdef CONFIG_ALTIVEC
990 unsigned long uninitialized_var(vrsave);
1000 /* Check if we can run the vcpu at all */
1001 if (!vcpu->arch.sane) {
1002 kvm_run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1007 kvmppc_core_prepare_to_enter(vcpu);
1010 * Interrupts could be timers for the guest which we have to inject
1011 * again, so let's postpone them until we're in the guest and if we
1012 * really did time things so badly, then we just exit again due to
1013 * a host external interrupt.
1015 __hard_irq_disable();
1017 /* No need to go into the guest when all we do is going out */
1018 if (signal_pending(current)) {
1019 __hard_irq_enable();
1020 kvm_run->exit_reason = KVM_EXIT_INTR;
1025 /* Save FPU state in stack */
1026 if (current->thread.regs->msr & MSR_FP)
1027 giveup_fpu(current);
1028 memcpy(fpr, current->thread.fpr, sizeof(current->thread.fpr));
1029 fpscr = current->thread.fpscr.val;
1030 fpexc_mode = current->thread.fpexc_mode;
1032 #ifdef CONFIG_ALTIVEC
1033 /* Save Altivec state in stack */
1034 used_vr = current->thread.used_vr;
1036 if (current->thread.regs->msr & MSR_VEC)
1037 giveup_altivec(current);
1038 memcpy(vr, current->thread.vr, sizeof(current->thread.vr));
1039 vscr = current->thread.vscr;
1040 vrsave = current->thread.vrsave;
1045 /* Save VSX state in stack */
1046 used_vsr = current->thread.used_vsr;
1047 if (used_vsr && (current->thread.regs->msr & MSR_VSX))
1048 __giveup_vsx(current);
1051 /* Remember the MSR with disabled extensions */
1052 ext_msr = current->thread.regs->msr;
1054 /* Preload FPU if it's enabled */
1055 if (vcpu->arch.shared->msr & MSR_FP)
1056 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
1060 ret = __kvmppc_vcpu_run(kvm_run, vcpu);
1064 current->thread.regs->msr = ext_msr;
1066 /* Make sure we save the guest FPU/Altivec/VSX state */
1067 kvmppc_giveup_ext(vcpu, MSR_FP);
1068 kvmppc_giveup_ext(vcpu, MSR_VEC);
1069 kvmppc_giveup_ext(vcpu, MSR_VSX);
1071 /* Restore FPU state from stack */
1072 memcpy(current->thread.fpr, fpr, sizeof(current->thread.fpr));
1073 current->thread.fpscr.val = fpscr;
1074 current->thread.fpexc_mode = fpexc_mode;
1076 #ifdef CONFIG_ALTIVEC
1077 /* Restore Altivec state from stack */
1078 if (used_vr && current->thread.used_vr) {
1079 memcpy(current->thread.vr, vr, sizeof(current->thread.vr));
1080 current->thread.vscr = vscr;
1081 current->thread.vrsave = vrsave;
1083 current->thread.used_vr = used_vr;
1087 current->thread.used_vsr = used_vsr;
1096 * Get (and clear) the dirty memory log for a memory slot.
1098 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1099 struct kvm_dirty_log *log)
1101 struct kvm_memory_slot *memslot;
1102 struct kvm_vcpu *vcpu;
1108 mutex_lock(&kvm->slots_lock);
1110 r = kvm_get_dirty_log(kvm, log, &is_dirty);
1114 /* If nothing is dirty, don't bother messing with page tables. */
1116 memslot = id_to_memslot(kvm->memslots, log->slot);
1118 ga = memslot->base_gfn << PAGE_SHIFT;
1119 ga_end = ga + (memslot->npages << PAGE_SHIFT);
1121 kvm_for_each_vcpu(n, vcpu, kvm)
1122 kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
1124 n = kvm_dirty_bitmap_bytes(memslot);
1125 memset(memslot->dirty_bitmap, 0, n);
1130 mutex_unlock(&kvm->slots_lock);
1134 int kvmppc_core_prepare_memory_region(struct kvm *kvm,
1135 struct kvm_userspace_memory_region *mem)
1140 void kvmppc_core_commit_memory_region(struct kvm *kvm,
1141 struct kvm_userspace_memory_region *mem)
1145 int kvmppc_core_init_vm(struct kvm *kvm)
1150 void kvmppc_core_destroy_vm(struct kvm *kvm)
1154 static int kvmppc_book3s_init(void)
1158 r = kvm_init(NULL, sizeof(struct kvmppc_vcpu_book3s), 0,
1164 r = kvmppc_mmu_hpte_sysinit();
1169 static void kvmppc_book3s_exit(void)
1171 kvmppc_mmu_hpte_sysexit();
1175 module_init(kvmppc_book3s_init);
1176 module_exit(kvmppc_book3s_exit);