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 hva)
300 /* XXX could be more clever ;) */
304 static int kvm_test_age_hva_pr(struct kvm *kvm, unsigned long hva)
306 /* XXX could be more clever ;) */
310 static void kvm_set_spte_hva_pr(struct kvm *kvm, unsigned long hva, pte_t pte)
312 /* The page will get remapped properly on its next fault */
313 do_kvm_unmap_hva(kvm, hva, hva + PAGE_SIZE);
316 /*****************************************/
318 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu *vcpu)
320 ulong guest_msr = kvmppc_get_msr(vcpu);
321 ulong smsr = guest_msr;
323 /* Guest MSR values */
324 smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE | MSR_LE;
325 /* Process MSR values */
326 smsr |= MSR_ME | MSR_RI | MSR_IR | MSR_DR | MSR_PR | MSR_EE;
327 /* External providers the guest reserved */
328 smsr |= (guest_msr & vcpu->arch.guest_owned_ext);
329 /* 64-bit Process MSR values */
330 #ifdef CONFIG_PPC_BOOK3S_64
331 smsr |= MSR_ISF | MSR_HV;
333 vcpu->arch.shadow_msr = smsr;
336 static void kvmppc_set_msr_pr(struct kvm_vcpu *vcpu, u64 msr)
338 ulong old_msr = kvmppc_get_msr(vcpu);
341 printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
344 msr &= to_book3s(vcpu)->msr_mask;
345 kvmppc_set_msr_fast(vcpu, msr);
346 kvmppc_recalc_shadow_msr(vcpu);
349 if (!vcpu->arch.pending_exceptions) {
350 kvm_vcpu_block(vcpu);
351 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
352 vcpu->stat.halt_wakeup++;
354 /* Unset POW bit after we woke up */
356 kvmppc_set_msr_fast(vcpu, msr);
360 if (kvmppc_is_split_real(vcpu))
361 kvmppc_fixup_split_real(vcpu);
363 kvmppc_unfixup_split_real(vcpu);
365 if ((kvmppc_get_msr(vcpu) & (MSR_PR|MSR_IR|MSR_DR)) !=
366 (old_msr & (MSR_PR|MSR_IR|MSR_DR))) {
367 kvmppc_mmu_flush_segments(vcpu);
368 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
370 /* Preload magic page segment when in kernel mode */
371 if (!(msr & MSR_PR) && vcpu->arch.magic_page_pa) {
372 struct kvm_vcpu_arch *a = &vcpu->arch;
375 kvmppc_mmu_map_segment(vcpu, a->magic_page_ea);
377 kvmppc_mmu_map_segment(vcpu, a->magic_page_pa);
382 * When switching from 32 to 64-bit, we may have a stale 32-bit
383 * magic page around, we need to flush it. Typically 32-bit magic
384 * page will be instanciated when calling into RTAS. Note: We
385 * assume that such transition only happens while in kernel mode,
386 * ie, we never transition from user 32-bit to kernel 64-bit with
387 * a 32-bit magic page around.
389 if (vcpu->arch.magic_page_pa &&
390 !(old_msr & MSR_PR) && !(old_msr & MSR_SF) && (msr & MSR_SF)) {
391 /* going from RTAS to normal kernel code */
392 kvmppc_mmu_pte_flush(vcpu, (uint32_t)vcpu->arch.magic_page_pa,
396 /* Preload FPU if it's enabled */
397 if (kvmppc_get_msr(vcpu) & MSR_FP)
398 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
401 void kvmppc_set_pvr_pr(struct kvm_vcpu *vcpu, u32 pvr)
405 vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
406 vcpu->arch.pvr = pvr;
407 #ifdef CONFIG_PPC_BOOK3S_64
408 if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
409 kvmppc_mmu_book3s_64_init(vcpu);
410 if (!to_book3s(vcpu)->hior_explicit)
411 to_book3s(vcpu)->hior = 0xfff00000;
412 to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
413 vcpu->arch.cpu_type = KVM_CPU_3S_64;
417 kvmppc_mmu_book3s_32_init(vcpu);
418 if (!to_book3s(vcpu)->hior_explicit)
419 to_book3s(vcpu)->hior = 0;
420 to_book3s(vcpu)->msr_mask = 0xffffffffULL;
421 vcpu->arch.cpu_type = KVM_CPU_3S_32;
424 kvmppc_sanity_check(vcpu);
426 /* If we are in hypervisor level on 970, we can tell the CPU to
427 * treat DCBZ as 32 bytes store */
428 vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
429 if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
430 !strcmp(cur_cpu_spec->platform, "ppc970"))
431 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
433 /* Cell performs badly if MSR_FEx are set. So let's hope nobody
434 really needs them in a VM on Cell and force disable them. */
435 if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be"))
436 to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1);
439 * If they're asking for POWER6 or later, set the flag
440 * indicating that we can do multiple large page sizes
442 * Also set the flag that indicates that tlbie has the large
443 * page bit in the RB operand instead of the instruction.
445 switch (PVR_VER(pvr)) {
450 vcpu->arch.hflags |= BOOK3S_HFLAG_MULTI_PGSIZE |
451 BOOK3S_HFLAG_NEW_TLBIE;
455 #ifdef CONFIG_PPC_BOOK3S_32
456 /* 32 bit Book3S always has 32 byte dcbz */
457 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
460 /* On some CPUs we can execute paired single operations natively */
461 asm ( "mfpvr %0" : "=r"(host_pvr));
463 case 0x00080200: /* lonestar 2.0 */
464 case 0x00088202: /* lonestar 2.2 */
465 case 0x70000100: /* gekko 1.0 */
466 case 0x00080100: /* gekko 2.0 */
467 case 0x00083203: /* gekko 2.3a */
468 case 0x00083213: /* gekko 2.3b */
469 case 0x00083204: /* gekko 2.4 */
470 case 0x00083214: /* gekko 2.4e (8SE) - retail HW2 */
471 case 0x00087200: /* broadway */
472 vcpu->arch.hflags |= BOOK3S_HFLAG_NATIVE_PS;
473 /* Enable HID2.PSE - in case we need it later */
474 mtspr(SPRN_HID2_GEKKO, mfspr(SPRN_HID2_GEKKO) | (1 << 29));
478 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
479 * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
480 * emulate 32 bytes dcbz length.
482 * The Book3s_64 inventors also realized this case and implemented a special bit
483 * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
485 * My approach here is to patch the dcbz instruction on executing pages.
487 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
494 hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
495 if (is_error_page(hpage))
498 hpage_offset = pte->raddr & ~PAGE_MASK;
499 hpage_offset &= ~0xFFFULL;
503 page = kmap_atomic(hpage);
505 /* patch dcbz into reserved instruction, so we trap */
506 for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
507 if ((be32_to_cpu(page[i]) & 0xff0007ff) == INS_DCBZ)
508 page[i] &= cpu_to_be32(0xfffffff7);
514 static int kvmppc_visible_gpa(struct kvm_vcpu *vcpu, gpa_t gpa)
516 ulong mp_pa = vcpu->arch.magic_page_pa;
518 if (!(kvmppc_get_msr(vcpu) & MSR_SF))
519 mp_pa = (uint32_t)mp_pa;
522 if (unlikely(mp_pa) && unlikely((mp_pa & KVM_PAM) == (gpa & KVM_PAM))) {
526 return kvm_is_visible_gfn(vcpu->kvm, gpa >> PAGE_SHIFT);
529 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
530 ulong eaddr, int vec)
532 bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
533 bool iswrite = false;
534 int r = RESUME_GUEST;
537 struct kvmppc_pte pte;
538 bool is_mmio = false;
539 bool dr = (kvmppc_get_msr(vcpu) & MSR_DR) ? true : false;
540 bool ir = (kvmppc_get_msr(vcpu) & MSR_IR) ? true : false;
543 relocated = data ? dr : ir;
544 if (data && (vcpu->arch.fault_dsisr & DSISR_ISSTORE))
547 /* Resolve real address if translation turned on */
549 page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data, iswrite);
551 pte.may_execute = true;
553 pte.may_write = true;
554 pte.raddr = eaddr & KVM_PAM;
556 pte.vpage = eaddr >> 12;
557 pte.page_size = MMU_PAGE_64K;
560 switch (kvmppc_get_msr(vcpu) & (MSR_DR|MSR_IR)) {
562 pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
566 (vcpu->arch.hflags & BOOK3S_HFLAG_SPLIT_HACK) &&
567 ((pte.raddr & SPLIT_HACK_MASK) == SPLIT_HACK_OFFS))
568 pte.raddr &= ~SPLIT_HACK_MASK;
571 vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
573 if ((kvmppc_get_msr(vcpu) & (MSR_DR|MSR_IR)) == MSR_DR)
574 pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
576 pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
580 page_found = -EINVAL;
584 if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
585 (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
587 * If we do the dcbz hack, we have to NX on every execution,
588 * so we can patch the executing code. This renders our guest
591 pte.may_execute = !data;
594 if (page_found == -ENOENT) {
595 /* Page not found in guest PTE entries */
596 u64 ssrr1 = vcpu->arch.shadow_srr1;
597 u64 msr = kvmppc_get_msr(vcpu);
598 kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
599 kvmppc_set_dsisr(vcpu, vcpu->arch.fault_dsisr);
600 kvmppc_set_msr_fast(vcpu, msr | (ssrr1 & 0xf8000000ULL));
601 kvmppc_book3s_queue_irqprio(vcpu, vec);
602 } else if (page_found == -EPERM) {
603 /* Storage protection */
604 u32 dsisr = vcpu->arch.fault_dsisr;
605 u64 ssrr1 = vcpu->arch.shadow_srr1;
606 u64 msr = kvmppc_get_msr(vcpu);
607 kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
608 dsisr = (dsisr & ~DSISR_NOHPTE) | DSISR_PROTFAULT;
609 kvmppc_set_dsisr(vcpu, dsisr);
610 kvmppc_set_msr_fast(vcpu, msr | (ssrr1 & 0xf8000000ULL));
611 kvmppc_book3s_queue_irqprio(vcpu, vec);
612 } else if (page_found == -EINVAL) {
613 /* Page not found in guest SLB */
614 kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
615 kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
616 } else if (!is_mmio &&
617 kvmppc_visible_gpa(vcpu, pte.raddr)) {
618 if (data && !(vcpu->arch.fault_dsisr & DSISR_NOHPTE)) {
620 * There is already a host HPTE there, presumably
621 * a read-only one for a page the guest thinks
622 * is writable, so get rid of it first.
624 kvmppc_mmu_unmap_page(vcpu, &pte);
626 /* The guest's PTE is not mapped yet. Map on the host */
627 kvmppc_mmu_map_page(vcpu, &pte, iswrite);
629 vcpu->stat.sp_storage++;
630 else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
631 (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
632 kvmppc_patch_dcbz(vcpu, &pte);
635 vcpu->stat.mmio_exits++;
636 vcpu->arch.paddr_accessed = pte.raddr;
637 vcpu->arch.vaddr_accessed = pte.eaddr;
638 r = kvmppc_emulate_mmio(run, vcpu);
639 if ( r == RESUME_HOST_NV )
646 static inline int get_fpr_index(int i)
648 return i * TS_FPRWIDTH;
651 /* Give up external provider (FPU, Altivec, VSX) */
652 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
654 struct thread_struct *t = ¤t->thread;
657 * VSX instructions can access FP and vector registers, so if
658 * we are giving up VSX, make sure we give up FP and VMX as well.
661 msr |= MSR_FP | MSR_VEC;
663 msr &= vcpu->arch.guest_owned_ext;
668 printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
673 * Note that on CPUs with VSX, giveup_fpu stores
674 * both the traditional FP registers and the added VSX
675 * registers into thread.fp_state.fpr[].
677 if (t->regs->msr & MSR_FP)
679 t->fp_save_area = NULL;
682 #ifdef CONFIG_ALTIVEC
684 if (current->thread.regs->msr & MSR_VEC)
685 giveup_altivec(current);
686 t->vr_save_area = NULL;
690 vcpu->arch.guest_owned_ext &= ~(msr | MSR_VSX);
691 kvmppc_recalc_shadow_msr(vcpu);
694 /* Give up facility (TAR / EBB / DSCR) */
695 static void kvmppc_giveup_fac(struct kvm_vcpu *vcpu, ulong fac)
697 #ifdef CONFIG_PPC_BOOK3S_64
698 if (!(vcpu->arch.shadow_fscr & (1ULL << fac))) {
699 /* Facility not available to the guest, ignore giveup request*/
705 vcpu->arch.tar = mfspr(SPRN_TAR);
706 mtspr(SPRN_TAR, current->thread.tar);
707 vcpu->arch.shadow_fscr &= ~FSCR_TAR;
713 /* Handle external providers (FPU, Altivec, VSX) */
714 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
717 struct thread_struct *t = ¤t->thread;
719 /* When we have paired singles, we emulate in software */
720 if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
723 if (!(kvmppc_get_msr(vcpu) & msr)) {
724 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
728 if (msr == MSR_VSX) {
729 /* No VSX? Give an illegal instruction interrupt */
731 if (!cpu_has_feature(CPU_FTR_VSX))
734 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
739 * We have to load up all the FP and VMX registers before
740 * we can let the guest use VSX instructions.
742 msr = MSR_FP | MSR_VEC | MSR_VSX;
745 /* See if we already own all the ext(s) needed */
746 msr &= ~vcpu->arch.guest_owned_ext;
751 printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
757 load_fp_state(&vcpu->arch.fp);
758 t->fp_save_area = &vcpu->arch.fp;
763 #ifdef CONFIG_ALTIVEC
765 enable_kernel_altivec();
766 load_vr_state(&vcpu->arch.vr);
767 t->vr_save_area = &vcpu->arch.vr;
773 vcpu->arch.guest_owned_ext |= msr;
774 kvmppc_recalc_shadow_msr(vcpu);
780 * Kernel code using FP or VMX could have flushed guest state to
781 * the thread_struct; if so, get it back now.
783 static void kvmppc_handle_lost_ext(struct kvm_vcpu *vcpu)
785 unsigned long lost_ext;
787 lost_ext = vcpu->arch.guest_owned_ext & ~current->thread.regs->msr;
791 if (lost_ext & MSR_FP) {
794 load_fp_state(&vcpu->arch.fp);
797 #ifdef CONFIG_ALTIVEC
798 if (lost_ext & MSR_VEC) {
800 enable_kernel_altivec();
801 load_vr_state(&vcpu->arch.vr);
805 current->thread.regs->msr |= lost_ext;
808 #ifdef CONFIG_PPC_BOOK3S_64
810 static void kvmppc_trigger_fac_interrupt(struct kvm_vcpu *vcpu, ulong fac)
812 /* Inject the Interrupt Cause field and trigger a guest interrupt */
813 vcpu->arch.fscr &= ~(0xffULL << 56);
814 vcpu->arch.fscr |= (fac << 56);
815 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_FAC_UNAVAIL);
818 static void kvmppc_emulate_fac(struct kvm_vcpu *vcpu, ulong fac)
820 enum emulation_result er = EMULATE_FAIL;
822 if (!(kvmppc_get_msr(vcpu) & MSR_PR))
823 er = kvmppc_emulate_instruction(vcpu->run, vcpu);
825 if ((er != EMULATE_DONE) && (er != EMULATE_AGAIN)) {
826 /* Couldn't emulate, trigger interrupt in guest */
827 kvmppc_trigger_fac_interrupt(vcpu, fac);
831 /* Enable facilities (TAR, EBB, DSCR) for the guest */
832 static int kvmppc_handle_fac(struct kvm_vcpu *vcpu, ulong fac)
834 bool guest_fac_enabled;
835 BUG_ON(!cpu_has_feature(CPU_FTR_ARCH_207S));
838 * Not every facility is enabled by FSCR bits, check whether the
839 * guest has this facility enabled at all.
844 guest_fac_enabled = (vcpu->arch.fscr & (1ULL << fac));
847 guest_fac_enabled = kvmppc_get_msr(vcpu) & MSR_TM;
850 guest_fac_enabled = false;
854 if (!guest_fac_enabled) {
855 /* Facility not enabled by the guest */
856 kvmppc_trigger_fac_interrupt(vcpu, fac);
862 /* TAR switching isn't lazy in Linux yet */
863 current->thread.tar = mfspr(SPRN_TAR);
864 mtspr(SPRN_TAR, vcpu->arch.tar);
865 vcpu->arch.shadow_fscr |= FSCR_TAR;
868 kvmppc_emulate_fac(vcpu, fac);
875 void kvmppc_set_fscr(struct kvm_vcpu *vcpu, u64 fscr)
877 if ((vcpu->arch.fscr & FSCR_TAR) && !(fscr & FSCR_TAR)) {
878 /* TAR got dropped, drop it in shadow too */
879 kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
881 vcpu->arch.fscr = fscr;
885 int kvmppc_handle_exit_pr(struct kvm_run *run, struct kvm_vcpu *vcpu,
886 unsigned int exit_nr)
891 vcpu->stat.sum_exits++;
893 run->exit_reason = KVM_EXIT_UNKNOWN;
894 run->ready_for_interrupt_injection = 1;
896 /* We get here with MSR.EE=1 */
898 trace_kvm_exit(exit_nr, vcpu);
902 case BOOK3S_INTERRUPT_INST_STORAGE:
904 ulong shadow_srr1 = vcpu->arch.shadow_srr1;
905 vcpu->stat.pf_instruc++;
907 if (kvmppc_is_split_real(vcpu))
908 kvmppc_fixup_split_real(vcpu);
910 #ifdef CONFIG_PPC_BOOK3S_32
911 /* We set segments as unused segments when invalidating them. So
912 * treat the respective fault as segment fault. */
914 struct kvmppc_book3s_shadow_vcpu *svcpu;
917 svcpu = svcpu_get(vcpu);
918 sr = svcpu->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT];
920 if (sr == SR_INVALID) {
921 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
928 /* only care about PTEG not found errors, but leave NX alone */
929 if (shadow_srr1 & 0x40000000) {
930 int idx = srcu_read_lock(&vcpu->kvm->srcu);
931 r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr);
932 srcu_read_unlock(&vcpu->kvm->srcu, idx);
933 vcpu->stat.sp_instruc++;
934 } else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
935 (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
937 * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
938 * so we can't use the NX bit inside the guest. Let's cross our fingers,
939 * that no guest that needs the dcbz hack does NX.
941 kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
944 u64 msr = kvmppc_get_msr(vcpu);
945 msr |= shadow_srr1 & 0x58000000;
946 kvmppc_set_msr_fast(vcpu, msr);
947 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
952 case BOOK3S_INTERRUPT_DATA_STORAGE:
954 ulong dar = kvmppc_get_fault_dar(vcpu);
955 u32 fault_dsisr = vcpu->arch.fault_dsisr;
956 vcpu->stat.pf_storage++;
958 #ifdef CONFIG_PPC_BOOK3S_32
959 /* We set segments as unused segments when invalidating them. So
960 * treat the respective fault as segment fault. */
962 struct kvmppc_book3s_shadow_vcpu *svcpu;
965 svcpu = svcpu_get(vcpu);
966 sr = svcpu->sr[dar >> SID_SHIFT];
968 if (sr == SR_INVALID) {
969 kvmppc_mmu_map_segment(vcpu, dar);
977 * We need to handle missing shadow PTEs, and
978 * protection faults due to us mapping a page read-only
979 * when the guest thinks it is writable.
981 if (fault_dsisr & (DSISR_NOHPTE | DSISR_PROTFAULT)) {
982 int idx = srcu_read_lock(&vcpu->kvm->srcu);
983 r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr);
984 srcu_read_unlock(&vcpu->kvm->srcu, idx);
986 kvmppc_set_dar(vcpu, dar);
987 kvmppc_set_dsisr(vcpu, fault_dsisr);
988 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
993 case BOOK3S_INTERRUPT_DATA_SEGMENT:
994 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
995 kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
996 kvmppc_book3s_queue_irqprio(vcpu,
997 BOOK3S_INTERRUPT_DATA_SEGMENT);
1001 case BOOK3S_INTERRUPT_INST_SEGMENT:
1002 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
1003 kvmppc_book3s_queue_irqprio(vcpu,
1004 BOOK3S_INTERRUPT_INST_SEGMENT);
1008 /* We're good on these - the host merely wanted to get our attention */
1009 case BOOK3S_INTERRUPT_DECREMENTER:
1010 case BOOK3S_INTERRUPT_HV_DECREMENTER:
1011 case BOOK3S_INTERRUPT_DOORBELL:
1012 case BOOK3S_INTERRUPT_H_DOORBELL:
1013 vcpu->stat.dec_exits++;
1016 case BOOK3S_INTERRUPT_EXTERNAL:
1017 case BOOK3S_INTERRUPT_EXTERNAL_LEVEL:
1018 case BOOK3S_INTERRUPT_EXTERNAL_HV:
1019 vcpu->stat.ext_intr_exits++;
1022 case BOOK3S_INTERRUPT_PERFMON:
1025 case BOOK3S_INTERRUPT_PROGRAM:
1026 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1028 enum emulation_result er;
1034 flags = vcpu->arch.shadow_srr1 & 0x1f0000ull;
1036 emul = kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
1037 if (emul != EMULATE_DONE) {
1042 if (kvmppc_get_msr(vcpu) & MSR_PR) {
1044 pr_info("Userspace triggered 0x700 exception at\n 0x%lx (0x%x)\n",
1045 kvmppc_get_pc(vcpu), last_inst);
1047 if ((last_inst & 0xff0007ff) !=
1048 (INS_DCBZ & 0xfffffff7)) {
1049 kvmppc_core_queue_program(vcpu, flags);
1055 vcpu->stat.emulated_inst_exits++;
1056 er = kvmppc_emulate_instruction(run, vcpu);
1059 r = RESUME_GUEST_NV;
1065 printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
1066 __func__, kvmppc_get_pc(vcpu), last_inst);
1067 kvmppc_core_queue_program(vcpu, flags);
1070 case EMULATE_DO_MMIO:
1071 run->exit_reason = KVM_EXIT_MMIO;
1074 case EMULATE_EXIT_USER:
1082 case BOOK3S_INTERRUPT_SYSCALL:
1087 /* Get last sc for papr */
1088 if (vcpu->arch.papr_enabled) {
1089 /* The sc instuction points SRR0 to the next inst */
1090 emul = kvmppc_get_last_inst(vcpu, INST_SC, &last_sc);
1091 if (emul != EMULATE_DONE) {
1092 kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) - 4);
1098 if (vcpu->arch.papr_enabled &&
1099 (last_sc == 0x44000022) &&
1100 !(kvmppc_get_msr(vcpu) & MSR_PR)) {
1101 /* SC 1 papr hypercalls */
1102 ulong cmd = kvmppc_get_gpr(vcpu, 3);
1105 #ifdef CONFIG_PPC_BOOK3S_64
1106 if (kvmppc_h_pr(vcpu, cmd) == EMULATE_DONE) {
1112 run->papr_hcall.nr = cmd;
1113 for (i = 0; i < 9; ++i) {
1114 ulong gpr = kvmppc_get_gpr(vcpu, 4 + i);
1115 run->papr_hcall.args[i] = gpr;
1117 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1118 vcpu->arch.hcall_needed = 1;
1120 } else if (vcpu->arch.osi_enabled &&
1121 (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
1122 (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
1123 /* MOL hypercalls */
1124 u64 *gprs = run->osi.gprs;
1127 run->exit_reason = KVM_EXIT_OSI;
1128 for (i = 0; i < 32; i++)
1129 gprs[i] = kvmppc_get_gpr(vcpu, i);
1130 vcpu->arch.osi_needed = 1;
1132 } else if (!(kvmppc_get_msr(vcpu) & MSR_PR) &&
1133 (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
1134 /* KVM PV hypercalls */
1135 kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
1138 /* Guest syscalls */
1139 vcpu->stat.syscall_exits++;
1140 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1145 case BOOK3S_INTERRUPT_FP_UNAVAIL:
1146 case BOOK3S_INTERRUPT_ALTIVEC:
1147 case BOOK3S_INTERRUPT_VSX:
1153 if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE) {
1154 /* Do paired single instruction emulation */
1155 emul = kvmppc_get_last_inst(vcpu, INST_GENERIC,
1157 if (emul == EMULATE_DONE)
1158 goto program_interrupt;
1165 /* Enable external provider */
1167 case BOOK3S_INTERRUPT_FP_UNAVAIL:
1171 case BOOK3S_INTERRUPT_ALTIVEC:
1175 case BOOK3S_INTERRUPT_VSX:
1180 r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
1183 case BOOK3S_INTERRUPT_ALIGNMENT:
1186 int emul = kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
1188 if (emul == EMULATE_DONE) {
1192 dsisr = kvmppc_alignment_dsisr(vcpu, last_inst);
1193 dar = kvmppc_alignment_dar(vcpu, last_inst);
1195 kvmppc_set_dsisr(vcpu, dsisr);
1196 kvmppc_set_dar(vcpu, dar);
1198 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1203 #ifdef CONFIG_PPC_BOOK3S_64
1204 case BOOK3S_INTERRUPT_FAC_UNAVAIL:
1205 kvmppc_handle_fac(vcpu, vcpu->arch.shadow_fscr >> 56);
1209 case BOOK3S_INTERRUPT_MACHINE_CHECK:
1210 case BOOK3S_INTERRUPT_TRACE:
1211 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1216 ulong shadow_srr1 = vcpu->arch.shadow_srr1;
1217 /* Ugh - bork here! What did we get? */
1218 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
1219 exit_nr, kvmppc_get_pc(vcpu), shadow_srr1);
1226 if (!(r & RESUME_HOST)) {
1227 /* To avoid clobbering exit_reason, only check for signals if
1228 * we aren't already exiting to userspace for some other
1232 * Interrupts could be timers for the guest which we have to
1233 * inject again, so let's postpone them until we're in the guest
1234 * and if we really did time things so badly, then we just exit
1235 * again due to a host external interrupt.
1237 s = kvmppc_prepare_to_enter(vcpu);
1241 /* interrupts now hard-disabled */
1242 kvmppc_fix_ee_before_entry();
1245 kvmppc_handle_lost_ext(vcpu);
1248 trace_kvm_book3s_reenter(r, vcpu);
1253 static int kvm_arch_vcpu_ioctl_get_sregs_pr(struct kvm_vcpu *vcpu,
1254 struct kvm_sregs *sregs)
1256 struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1259 sregs->pvr = vcpu->arch.pvr;
1261 sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
1262 if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1263 for (i = 0; i < 64; i++) {
1264 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige | i;
1265 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1268 for (i = 0; i < 16; i++)
1269 sregs->u.s.ppc32.sr[i] = kvmppc_get_sr(vcpu, i);
1271 for (i = 0; i < 8; i++) {
1272 sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
1273 sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
1280 static int kvm_arch_vcpu_ioctl_set_sregs_pr(struct kvm_vcpu *vcpu,
1281 struct kvm_sregs *sregs)
1283 struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1286 kvmppc_set_pvr_pr(vcpu, sregs->pvr);
1288 vcpu3s->sdr1 = sregs->u.s.sdr1;
1289 if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1290 for (i = 0; i < 64; i++) {
1291 vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
1292 sregs->u.s.ppc64.slb[i].slbe);
1295 for (i = 0; i < 16; i++) {
1296 vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
1298 for (i = 0; i < 8; i++) {
1299 kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
1300 (u32)sregs->u.s.ppc32.ibat[i]);
1301 kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
1302 (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
1303 kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
1304 (u32)sregs->u.s.ppc32.dbat[i]);
1305 kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
1306 (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
1310 /* Flush the MMU after messing with the segments */
1311 kvmppc_mmu_pte_flush(vcpu, 0, 0);
1316 static int kvmppc_get_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1317 union kvmppc_one_reg *val)
1322 case KVM_REG_PPC_HIOR:
1323 *val = get_reg_val(id, to_book3s(vcpu)->hior);
1325 case KVM_REG_PPC_LPCR:
1326 case KVM_REG_PPC_LPCR_64:
1328 * We are only interested in the LPCR_ILE bit
1330 if (vcpu->arch.intr_msr & MSR_LE)
1331 *val = get_reg_val(id, LPCR_ILE);
1333 *val = get_reg_val(id, 0);
1343 static void kvmppc_set_lpcr_pr(struct kvm_vcpu *vcpu, u64 new_lpcr)
1345 if (new_lpcr & LPCR_ILE)
1346 vcpu->arch.intr_msr |= MSR_LE;
1348 vcpu->arch.intr_msr &= ~MSR_LE;
1351 static int kvmppc_set_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1352 union kvmppc_one_reg *val)
1357 case KVM_REG_PPC_HIOR:
1358 to_book3s(vcpu)->hior = set_reg_val(id, *val);
1359 to_book3s(vcpu)->hior_explicit = true;
1361 case KVM_REG_PPC_LPCR:
1362 case KVM_REG_PPC_LPCR_64:
1363 kvmppc_set_lpcr_pr(vcpu, set_reg_val(id, *val));
1373 static struct kvm_vcpu *kvmppc_core_vcpu_create_pr(struct kvm *kvm,
1376 struct kvmppc_vcpu_book3s *vcpu_book3s;
1377 struct kvm_vcpu *vcpu;
1381 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1385 vcpu_book3s = vzalloc(sizeof(struct kvmppc_vcpu_book3s));
1388 vcpu->arch.book3s = vcpu_book3s;
1390 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1391 vcpu->arch.shadow_vcpu =
1392 kzalloc(sizeof(*vcpu->arch.shadow_vcpu), GFP_KERNEL);
1393 if (!vcpu->arch.shadow_vcpu)
1397 err = kvm_vcpu_init(vcpu, kvm, id);
1399 goto free_shadow_vcpu;
1402 p = __get_free_page(GFP_KERNEL|__GFP_ZERO);
1405 vcpu->arch.shared = (void *)p;
1406 #ifdef CONFIG_PPC_BOOK3S_64
1407 /* Always start the shared struct in native endian mode */
1408 #ifdef __BIG_ENDIAN__
1409 vcpu->arch.shared_big_endian = true;
1411 vcpu->arch.shared_big_endian = false;
1415 * Default to the same as the host if we're on sufficiently
1416 * recent machine that we have 1TB segments;
1417 * otherwise default to PPC970FX.
1419 vcpu->arch.pvr = 0x3C0301;
1420 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1421 vcpu->arch.pvr = mfspr(SPRN_PVR);
1422 vcpu->arch.intr_msr = MSR_SF;
1424 /* default to book3s_32 (750) */
1425 vcpu->arch.pvr = 0x84202;
1427 kvmppc_set_pvr_pr(vcpu, vcpu->arch.pvr);
1428 vcpu->arch.slb_nr = 64;
1430 vcpu->arch.shadow_msr = MSR_USER64 & ~MSR_LE;
1432 err = kvmppc_mmu_init(vcpu);
1439 kvm_vcpu_uninit(vcpu);
1441 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1442 kfree(vcpu->arch.shadow_vcpu);
1447 kmem_cache_free(kvm_vcpu_cache, vcpu);
1449 return ERR_PTR(err);
1452 static void kvmppc_core_vcpu_free_pr(struct kvm_vcpu *vcpu)
1454 struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
1456 free_page((unsigned long)vcpu->arch.shared & PAGE_MASK);
1457 kvm_vcpu_uninit(vcpu);
1458 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1459 kfree(vcpu->arch.shadow_vcpu);
1462 kmem_cache_free(kvm_vcpu_cache, vcpu);
1465 static int kvmppc_vcpu_run_pr(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1468 #ifdef CONFIG_ALTIVEC
1469 unsigned long uninitialized_var(vrsave);
1472 /* Check if we can run the vcpu at all */
1473 if (!vcpu->arch.sane) {
1474 kvm_run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1480 * Interrupts could be timers for the guest which we have to inject
1481 * again, so let's postpone them until we're in the guest and if we
1482 * really did time things so badly, then we just exit again due to
1483 * a host external interrupt.
1485 ret = kvmppc_prepare_to_enter(vcpu);
1488 /* interrupts now hard-disabled */
1490 /* Save FPU state in thread_struct */
1491 if (current->thread.regs->msr & MSR_FP)
1492 giveup_fpu(current);
1494 #ifdef CONFIG_ALTIVEC
1495 /* Save Altivec state in thread_struct */
1496 if (current->thread.regs->msr & MSR_VEC)
1497 giveup_altivec(current);
1501 /* Save VSX state in thread_struct */
1502 if (current->thread.regs->msr & MSR_VSX)
1503 __giveup_vsx(current);
1506 /* Preload FPU if it's enabled */
1507 if (kvmppc_get_msr(vcpu) & MSR_FP)
1508 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
1510 kvmppc_fix_ee_before_entry();
1512 ret = __kvmppc_vcpu_run(kvm_run, vcpu);
1514 /* No need for kvm_guest_exit. It's done in handle_exit.
1515 We also get here with interrupts enabled. */
1517 /* Make sure we save the guest FPU/Altivec/VSX state */
1518 kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
1520 /* Make sure we save the guest TAR/EBB/DSCR state */
1521 kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
1524 vcpu->mode = OUTSIDE_GUEST_MODE;
1529 * Get (and clear) the dirty memory log for a memory slot.
1531 static int kvm_vm_ioctl_get_dirty_log_pr(struct kvm *kvm,
1532 struct kvm_dirty_log *log)
1534 struct kvm_memory_slot *memslot;
1535 struct kvm_vcpu *vcpu;
1541 mutex_lock(&kvm->slots_lock);
1543 r = kvm_get_dirty_log(kvm, log, &is_dirty);
1547 /* If nothing is dirty, don't bother messing with page tables. */
1549 memslot = id_to_memslot(kvm->memslots, log->slot);
1551 ga = memslot->base_gfn << PAGE_SHIFT;
1552 ga_end = ga + (memslot->npages << PAGE_SHIFT);
1554 kvm_for_each_vcpu(n, vcpu, kvm)
1555 kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
1557 n = kvm_dirty_bitmap_bytes(memslot);
1558 memset(memslot->dirty_bitmap, 0, n);
1563 mutex_unlock(&kvm->slots_lock);
1567 static void kvmppc_core_flush_memslot_pr(struct kvm *kvm,
1568 struct kvm_memory_slot *memslot)
1573 static int kvmppc_core_prepare_memory_region_pr(struct kvm *kvm,
1574 struct kvm_memory_slot *memslot,
1575 struct kvm_userspace_memory_region *mem)
1580 static void kvmppc_core_commit_memory_region_pr(struct kvm *kvm,
1581 struct kvm_userspace_memory_region *mem,
1582 const struct kvm_memory_slot *old)
1587 static void kvmppc_core_free_memslot_pr(struct kvm_memory_slot *free,
1588 struct kvm_memory_slot *dont)
1593 static int kvmppc_core_create_memslot_pr(struct kvm_memory_slot *slot,
1594 unsigned long npages)
1601 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1602 struct kvm_ppc_smmu_info *info)
1605 struct kvm_vcpu *vcpu;
1609 /* SLB is always 64 entries */
1610 info->slb_size = 64;
1612 /* Standard 4k base page size segment */
1613 info->sps[0].page_shift = 12;
1614 info->sps[0].slb_enc = 0;
1615 info->sps[0].enc[0].page_shift = 12;
1616 info->sps[0].enc[0].pte_enc = 0;
1619 * 64k large page size.
1620 * We only want to put this in if the CPUs we're emulating
1621 * support it, but unfortunately we don't have a vcpu easily
1622 * to hand here to test. Just pick the first vcpu, and if
1623 * that doesn't exist yet, report the minimum capability,
1624 * i.e., no 64k pages.
1625 * 1T segment support goes along with 64k pages.
1628 vcpu = kvm_get_vcpu(kvm, 0);
1629 if (vcpu && (vcpu->arch.hflags & BOOK3S_HFLAG_MULTI_PGSIZE)) {
1630 info->flags = KVM_PPC_1T_SEGMENTS;
1631 info->sps[i].page_shift = 16;
1632 info->sps[i].slb_enc = SLB_VSID_L | SLB_VSID_LP_01;
1633 info->sps[i].enc[0].page_shift = 16;
1634 info->sps[i].enc[0].pte_enc = 1;
1638 /* Standard 16M large page size segment */
1639 info->sps[i].page_shift = 24;
1640 info->sps[i].slb_enc = SLB_VSID_L;
1641 info->sps[i].enc[0].page_shift = 24;
1642 info->sps[i].enc[0].pte_enc = 0;
1647 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1648 struct kvm_ppc_smmu_info *info)
1650 /* We should not get called */
1653 #endif /* CONFIG_PPC64 */
1655 static unsigned int kvm_global_user_count = 0;
1656 static DEFINE_SPINLOCK(kvm_global_user_count_lock);
1658 static int kvmppc_core_init_vm_pr(struct kvm *kvm)
1660 mutex_init(&kvm->arch.hpt_mutex);
1662 #ifdef CONFIG_PPC_BOOK3S_64
1663 /* Start out with the default set of hcalls enabled */
1664 kvmppc_pr_init_default_hcalls(kvm);
1667 if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
1668 spin_lock(&kvm_global_user_count_lock);
1669 if (++kvm_global_user_count == 1)
1670 pSeries_disable_reloc_on_exc();
1671 spin_unlock(&kvm_global_user_count_lock);
1676 static void kvmppc_core_destroy_vm_pr(struct kvm *kvm)
1679 WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
1682 if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
1683 spin_lock(&kvm_global_user_count_lock);
1684 BUG_ON(kvm_global_user_count == 0);
1685 if (--kvm_global_user_count == 0)
1686 pSeries_enable_reloc_on_exc();
1687 spin_unlock(&kvm_global_user_count_lock);
1691 static int kvmppc_core_check_processor_compat_pr(void)
1693 /* we are always compatible */
1697 static long kvm_arch_vm_ioctl_pr(struct file *filp,
1698 unsigned int ioctl, unsigned long arg)
1703 static struct kvmppc_ops kvm_ops_pr = {
1704 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_pr,
1705 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_pr,
1706 .get_one_reg = kvmppc_get_one_reg_pr,
1707 .set_one_reg = kvmppc_set_one_reg_pr,
1708 .vcpu_load = kvmppc_core_vcpu_load_pr,
1709 .vcpu_put = kvmppc_core_vcpu_put_pr,
1710 .set_msr = kvmppc_set_msr_pr,
1711 .vcpu_run = kvmppc_vcpu_run_pr,
1712 .vcpu_create = kvmppc_core_vcpu_create_pr,
1713 .vcpu_free = kvmppc_core_vcpu_free_pr,
1714 .check_requests = kvmppc_core_check_requests_pr,
1715 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_pr,
1716 .flush_memslot = kvmppc_core_flush_memslot_pr,
1717 .prepare_memory_region = kvmppc_core_prepare_memory_region_pr,
1718 .commit_memory_region = kvmppc_core_commit_memory_region_pr,
1719 .unmap_hva = kvm_unmap_hva_pr,
1720 .unmap_hva_range = kvm_unmap_hva_range_pr,
1721 .age_hva = kvm_age_hva_pr,
1722 .test_age_hva = kvm_test_age_hva_pr,
1723 .set_spte_hva = kvm_set_spte_hva_pr,
1724 .mmu_destroy = kvmppc_mmu_destroy_pr,
1725 .free_memslot = kvmppc_core_free_memslot_pr,
1726 .create_memslot = kvmppc_core_create_memslot_pr,
1727 .init_vm = kvmppc_core_init_vm_pr,
1728 .destroy_vm = kvmppc_core_destroy_vm_pr,
1729 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_pr,
1730 .emulate_op = kvmppc_core_emulate_op_pr,
1731 .emulate_mtspr = kvmppc_core_emulate_mtspr_pr,
1732 .emulate_mfspr = kvmppc_core_emulate_mfspr_pr,
1733 .fast_vcpu_kick = kvm_vcpu_kick,
1734 .arch_vm_ioctl = kvm_arch_vm_ioctl_pr,
1735 #ifdef CONFIG_PPC_BOOK3S_64
1736 .hcall_implemented = kvmppc_hcall_impl_pr,
1741 int kvmppc_book3s_init_pr(void)
1745 r = kvmppc_core_check_processor_compat_pr();
1749 kvm_ops_pr.owner = THIS_MODULE;
1750 kvmppc_pr_ops = &kvm_ops_pr;
1752 r = kvmppc_mmu_hpte_sysinit();
1756 void kvmppc_book3s_exit_pr(void)
1758 kvmppc_pr_ops = NULL;
1759 kvmppc_mmu_hpte_sysexit();
1763 * We only support separate modules for book3s 64
1765 #ifdef CONFIG_PPC_BOOK3S_64
1767 module_init(kvmppc_book3s_init_pr);
1768 module_exit(kvmppc_book3s_exit_pr);
1770 MODULE_LICENSE("GPL");
1771 MODULE_ALIAS_MISCDEV(KVM_MINOR);
1772 MODULE_ALIAS("devname:kvm");