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>
47 #define CREATE_TRACE_POINTS
50 /* #define EXIT_DEBUG */
51 /* #define DEBUG_EXT */
53 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
56 /* Some compatibility defines */
57 #ifdef CONFIG_PPC_BOOK3S_32
58 #define MSR_USER32 MSR_USER
59 #define MSR_USER64 MSR_USER
60 #define HW_PAGE_SIZE PAGE_SIZE
63 static void kvmppc_core_vcpu_load_pr(struct kvm_vcpu *vcpu, int cpu)
65 #ifdef CONFIG_PPC_BOOK3S_64
66 struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
67 memcpy(svcpu->slb, to_book3s(vcpu)->slb_shadow, sizeof(svcpu->slb));
68 svcpu->slb_max = to_book3s(vcpu)->slb_shadow_max;
71 vcpu->cpu = smp_processor_id();
72 #ifdef CONFIG_PPC_BOOK3S_32
73 current->thread.kvm_shadow_vcpu = vcpu->arch.shadow_vcpu;
77 static void kvmppc_core_vcpu_put_pr(struct kvm_vcpu *vcpu)
79 #ifdef CONFIG_PPC_BOOK3S_64
80 struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
81 memcpy(to_book3s(vcpu)->slb_shadow, svcpu->slb, sizeof(svcpu->slb));
82 to_book3s(vcpu)->slb_shadow_max = svcpu->slb_max;
86 kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
90 /* Copy data needed by real-mode code from vcpu to shadow vcpu */
91 void kvmppc_copy_to_svcpu(struct kvmppc_book3s_shadow_vcpu *svcpu,
92 struct kvm_vcpu *vcpu)
94 svcpu->gpr[0] = vcpu->arch.gpr[0];
95 svcpu->gpr[1] = vcpu->arch.gpr[1];
96 svcpu->gpr[2] = vcpu->arch.gpr[2];
97 svcpu->gpr[3] = vcpu->arch.gpr[3];
98 svcpu->gpr[4] = vcpu->arch.gpr[4];
99 svcpu->gpr[5] = vcpu->arch.gpr[5];
100 svcpu->gpr[6] = vcpu->arch.gpr[6];
101 svcpu->gpr[7] = vcpu->arch.gpr[7];
102 svcpu->gpr[8] = vcpu->arch.gpr[8];
103 svcpu->gpr[9] = vcpu->arch.gpr[9];
104 svcpu->gpr[10] = vcpu->arch.gpr[10];
105 svcpu->gpr[11] = vcpu->arch.gpr[11];
106 svcpu->gpr[12] = vcpu->arch.gpr[12];
107 svcpu->gpr[13] = vcpu->arch.gpr[13];
108 svcpu->cr = vcpu->arch.cr;
109 svcpu->xer = vcpu->arch.xer;
110 svcpu->ctr = vcpu->arch.ctr;
111 svcpu->lr = vcpu->arch.lr;
112 svcpu->pc = vcpu->arch.pc;
115 /* Copy data touched by real-mode code from shadow vcpu back to vcpu */
116 void kvmppc_copy_from_svcpu(struct kvm_vcpu *vcpu,
117 struct kvmppc_book3s_shadow_vcpu *svcpu)
119 vcpu->arch.gpr[0] = svcpu->gpr[0];
120 vcpu->arch.gpr[1] = svcpu->gpr[1];
121 vcpu->arch.gpr[2] = svcpu->gpr[2];
122 vcpu->arch.gpr[3] = svcpu->gpr[3];
123 vcpu->arch.gpr[4] = svcpu->gpr[4];
124 vcpu->arch.gpr[5] = svcpu->gpr[5];
125 vcpu->arch.gpr[6] = svcpu->gpr[6];
126 vcpu->arch.gpr[7] = svcpu->gpr[7];
127 vcpu->arch.gpr[8] = svcpu->gpr[8];
128 vcpu->arch.gpr[9] = svcpu->gpr[9];
129 vcpu->arch.gpr[10] = svcpu->gpr[10];
130 vcpu->arch.gpr[11] = svcpu->gpr[11];
131 vcpu->arch.gpr[12] = svcpu->gpr[12];
132 vcpu->arch.gpr[13] = svcpu->gpr[13];
133 vcpu->arch.cr = svcpu->cr;
134 vcpu->arch.xer = svcpu->xer;
135 vcpu->arch.ctr = svcpu->ctr;
136 vcpu->arch.lr = svcpu->lr;
137 vcpu->arch.pc = svcpu->pc;
138 vcpu->arch.shadow_srr1 = svcpu->shadow_srr1;
139 vcpu->arch.fault_dar = svcpu->fault_dar;
140 vcpu->arch.fault_dsisr = svcpu->fault_dsisr;
141 vcpu->arch.last_inst = svcpu->last_inst;
144 static int kvmppc_core_check_requests_pr(struct kvm_vcpu *vcpu)
146 int r = 1; /* Indicate we want to get back into the guest */
148 /* We misuse TLB_FLUSH to indicate that we want to clear
149 all shadow cache entries */
150 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
151 kvmppc_mmu_pte_flush(vcpu, 0, 0);
156 /************* MMU Notifiers *************/
157 static void do_kvm_unmap_hva(struct kvm *kvm, unsigned long start,
161 struct kvm_vcpu *vcpu;
162 struct kvm_memslots *slots;
163 struct kvm_memory_slot *memslot;
165 slots = kvm_memslots(kvm);
166 kvm_for_each_memslot(memslot, slots) {
167 unsigned long hva_start, hva_end;
170 hva_start = max(start, memslot->userspace_addr);
171 hva_end = min(end, memslot->userspace_addr +
172 (memslot->npages << PAGE_SHIFT));
173 if (hva_start >= hva_end)
176 * {gfn(page) | page intersects with [hva_start, hva_end)} =
177 * {gfn, gfn+1, ..., gfn_end-1}.
179 gfn = hva_to_gfn_memslot(hva_start, memslot);
180 gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot);
181 kvm_for_each_vcpu(i, vcpu, kvm)
182 kvmppc_mmu_pte_pflush(vcpu, gfn << PAGE_SHIFT,
183 gfn_end << PAGE_SHIFT);
187 static int kvm_unmap_hva_pr(struct kvm *kvm, unsigned long hva)
189 trace_kvm_unmap_hva(hva);
191 do_kvm_unmap_hva(kvm, hva, hva + PAGE_SIZE);
196 static int kvm_unmap_hva_range_pr(struct kvm *kvm, unsigned long start,
199 do_kvm_unmap_hva(kvm, start, end);
204 static int kvm_age_hva_pr(struct kvm *kvm, unsigned long hva)
206 /* XXX could be more clever ;) */
210 static int kvm_test_age_hva_pr(struct kvm *kvm, unsigned long hva)
212 /* XXX could be more clever ;) */
216 static void kvm_set_spte_hva_pr(struct kvm *kvm, unsigned long hva, pte_t pte)
218 /* The page will get remapped properly on its next fault */
219 do_kvm_unmap_hva(kvm, hva, hva + PAGE_SIZE);
222 /*****************************************/
224 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu *vcpu)
226 ulong smsr = vcpu->arch.shared->msr;
228 /* Guest MSR values */
229 smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE;
230 /* Process MSR values */
231 smsr |= MSR_ME | MSR_RI | MSR_IR | MSR_DR | MSR_PR | MSR_EE;
232 /* External providers the guest reserved */
233 smsr |= (vcpu->arch.shared->msr & vcpu->arch.guest_owned_ext);
234 /* 64-bit Process MSR values */
235 #ifdef CONFIG_PPC_BOOK3S_64
236 smsr |= MSR_ISF | MSR_HV;
238 vcpu->arch.shadow_msr = smsr;
241 static void kvmppc_set_msr_pr(struct kvm_vcpu *vcpu, u64 msr)
243 ulong old_msr = vcpu->arch.shared->msr;
246 printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
249 msr &= to_book3s(vcpu)->msr_mask;
250 vcpu->arch.shared->msr = msr;
251 kvmppc_recalc_shadow_msr(vcpu);
254 if (!vcpu->arch.pending_exceptions) {
255 kvm_vcpu_block(vcpu);
256 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
257 vcpu->stat.halt_wakeup++;
259 /* Unset POW bit after we woke up */
261 vcpu->arch.shared->msr = msr;
265 if ((vcpu->arch.shared->msr & (MSR_PR|MSR_IR|MSR_DR)) !=
266 (old_msr & (MSR_PR|MSR_IR|MSR_DR))) {
267 kvmppc_mmu_flush_segments(vcpu);
268 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
270 /* Preload magic page segment when in kernel mode */
271 if (!(msr & MSR_PR) && vcpu->arch.magic_page_pa) {
272 struct kvm_vcpu_arch *a = &vcpu->arch;
275 kvmppc_mmu_map_segment(vcpu, a->magic_page_ea);
277 kvmppc_mmu_map_segment(vcpu, a->magic_page_pa);
282 * When switching from 32 to 64-bit, we may have a stale 32-bit
283 * magic page around, we need to flush it. Typically 32-bit magic
284 * page will be instanciated when calling into RTAS. Note: We
285 * assume that such transition only happens while in kernel mode,
286 * ie, we never transition from user 32-bit to kernel 64-bit with
287 * a 32-bit magic page around.
289 if (vcpu->arch.magic_page_pa &&
290 !(old_msr & MSR_PR) && !(old_msr & MSR_SF) && (msr & MSR_SF)) {
291 /* going from RTAS to normal kernel code */
292 kvmppc_mmu_pte_flush(vcpu, (uint32_t)vcpu->arch.magic_page_pa,
296 /* Preload FPU if it's enabled */
297 if (vcpu->arch.shared->msr & MSR_FP)
298 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
301 void kvmppc_set_pvr_pr(struct kvm_vcpu *vcpu, u32 pvr)
305 vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
306 vcpu->arch.pvr = pvr;
307 #ifdef CONFIG_PPC_BOOK3S_64
308 if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
309 kvmppc_mmu_book3s_64_init(vcpu);
310 if (!to_book3s(vcpu)->hior_explicit)
311 to_book3s(vcpu)->hior = 0xfff00000;
312 to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
313 vcpu->arch.cpu_type = KVM_CPU_3S_64;
317 kvmppc_mmu_book3s_32_init(vcpu);
318 if (!to_book3s(vcpu)->hior_explicit)
319 to_book3s(vcpu)->hior = 0;
320 to_book3s(vcpu)->msr_mask = 0xffffffffULL;
321 vcpu->arch.cpu_type = KVM_CPU_3S_32;
324 kvmppc_sanity_check(vcpu);
326 /* If we are in hypervisor level on 970, we can tell the CPU to
327 * treat DCBZ as 32 bytes store */
328 vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
329 if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
330 !strcmp(cur_cpu_spec->platform, "ppc970"))
331 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
333 /* Cell performs badly if MSR_FEx are set. So let's hope nobody
334 really needs them in a VM on Cell and force disable them. */
335 if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be"))
336 to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1);
339 * If they're asking for POWER6 or later, set the flag
340 * indicating that we can do multiple large page sizes
342 * Also set the flag that indicates that tlbie has the large
343 * page bit in the RB operand instead of the instruction.
345 switch (PVR_VER(pvr)) {
350 vcpu->arch.hflags |= BOOK3S_HFLAG_MULTI_PGSIZE |
351 BOOK3S_HFLAG_NEW_TLBIE;
355 #ifdef CONFIG_PPC_BOOK3S_32
356 /* 32 bit Book3S always has 32 byte dcbz */
357 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
360 /* On some CPUs we can execute paired single operations natively */
361 asm ( "mfpvr %0" : "=r"(host_pvr));
363 case 0x00080200: /* lonestar 2.0 */
364 case 0x00088202: /* lonestar 2.2 */
365 case 0x70000100: /* gekko 1.0 */
366 case 0x00080100: /* gekko 2.0 */
367 case 0x00083203: /* gekko 2.3a */
368 case 0x00083213: /* gekko 2.3b */
369 case 0x00083204: /* gekko 2.4 */
370 case 0x00083214: /* gekko 2.4e (8SE) - retail HW2 */
371 case 0x00087200: /* broadway */
372 vcpu->arch.hflags |= BOOK3S_HFLAG_NATIVE_PS;
373 /* Enable HID2.PSE - in case we need it later */
374 mtspr(SPRN_HID2_GEKKO, mfspr(SPRN_HID2_GEKKO) | (1 << 29));
378 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
379 * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
380 * emulate 32 bytes dcbz length.
382 * The Book3s_64 inventors also realized this case and implemented a special bit
383 * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
385 * My approach here is to patch the dcbz instruction on executing pages.
387 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
394 hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
395 if (is_error_page(hpage))
398 hpage_offset = pte->raddr & ~PAGE_MASK;
399 hpage_offset &= ~0xFFFULL;
403 page = kmap_atomic(hpage);
405 /* patch dcbz into reserved instruction, so we trap */
406 for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
407 if ((page[i] & 0xff0007ff) == INS_DCBZ)
408 page[i] &= 0xfffffff7;
414 static int kvmppc_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
416 ulong mp_pa = vcpu->arch.magic_page_pa;
418 if (!(vcpu->arch.shared->msr & MSR_SF))
419 mp_pa = (uint32_t)mp_pa;
421 if (unlikely(mp_pa) &&
422 unlikely((mp_pa & KVM_PAM) >> PAGE_SHIFT == gfn)) {
426 return kvm_is_visible_gfn(vcpu->kvm, gfn);
429 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
430 ulong eaddr, int vec)
432 bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
433 bool iswrite = false;
434 int r = RESUME_GUEST;
437 struct kvmppc_pte pte;
438 bool is_mmio = false;
439 bool dr = (vcpu->arch.shared->msr & MSR_DR) ? true : false;
440 bool ir = (vcpu->arch.shared->msr & MSR_IR) ? true : false;
443 relocated = data ? dr : ir;
444 if (data && (vcpu->arch.fault_dsisr & DSISR_ISSTORE))
447 /* Resolve real address if translation turned on */
449 page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data, iswrite);
451 pte.may_execute = true;
453 pte.may_write = true;
454 pte.raddr = eaddr & KVM_PAM;
456 pte.vpage = eaddr >> 12;
457 pte.page_size = MMU_PAGE_64K;
460 switch (vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) {
462 pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
466 vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
468 if ((vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) == MSR_DR)
469 pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
471 pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
475 page_found = -EINVAL;
479 if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
480 (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
482 * If we do the dcbz hack, we have to NX on every execution,
483 * so we can patch the executing code. This renders our guest
486 pte.may_execute = !data;
489 if (page_found == -ENOENT) {
490 /* Page not found in guest PTE entries */
491 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
492 vcpu->arch.shared->dsisr = vcpu->arch.fault_dsisr;
493 vcpu->arch.shared->msr |=
494 vcpu->arch.shadow_srr1 & 0x00000000f8000000ULL;
495 kvmppc_book3s_queue_irqprio(vcpu, vec);
496 } else if (page_found == -EPERM) {
497 /* Storage protection */
498 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
499 vcpu->arch.shared->dsisr = vcpu->arch.fault_dsisr & ~DSISR_NOHPTE;
500 vcpu->arch.shared->dsisr |= DSISR_PROTFAULT;
501 vcpu->arch.shared->msr |=
502 vcpu->arch.shadow_srr1 & 0x00000000f8000000ULL;
503 kvmppc_book3s_queue_irqprio(vcpu, vec);
504 } else if (page_found == -EINVAL) {
505 /* Page not found in guest SLB */
506 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
507 kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
508 } else if (!is_mmio &&
509 kvmppc_visible_gfn(vcpu, pte.raddr >> PAGE_SHIFT)) {
510 if (data && !(vcpu->arch.fault_dsisr & DSISR_NOHPTE)) {
512 * There is already a host HPTE there, presumably
513 * a read-only one for a page the guest thinks
514 * is writable, so get rid of it first.
516 kvmppc_mmu_unmap_page(vcpu, &pte);
518 /* The guest's PTE is not mapped yet. Map on the host */
519 kvmppc_mmu_map_page(vcpu, &pte, iswrite);
521 vcpu->stat.sp_storage++;
522 else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
523 (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
524 kvmppc_patch_dcbz(vcpu, &pte);
527 vcpu->stat.mmio_exits++;
528 vcpu->arch.paddr_accessed = pte.raddr;
529 vcpu->arch.vaddr_accessed = pte.eaddr;
530 r = kvmppc_emulate_mmio(run, vcpu);
531 if ( r == RESUME_HOST_NV )
538 static inline int get_fpr_index(int i)
540 return i * TS_FPRWIDTH;
543 /* Give up external provider (FPU, Altivec, VSX) */
544 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
546 struct thread_struct *t = ¤t->thread;
547 u64 *vcpu_fpr = vcpu->arch.fpr;
549 u64 *vcpu_vsx = vcpu->arch.vsr;
551 u64 *thread_fpr = &t->fp_state.fpr[0][0];
555 * VSX instructions can access FP and vector registers, so if
556 * we are giving up VSX, make sure we give up FP and VMX as well.
559 msr |= MSR_FP | MSR_VEC;
561 msr &= vcpu->arch.guest_owned_ext;
566 printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
571 * Note that on CPUs with VSX, giveup_fpu stores
572 * both the traditional FP registers and the added VSX
573 * registers into thread.fp_state.fpr[].
575 if (current->thread.regs->msr & MSR_FP)
577 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
578 vcpu_fpr[i] = thread_fpr[get_fpr_index(i)];
580 vcpu->arch.fpscr = t->fp_state.fpscr;
583 if (cpu_has_feature(CPU_FTR_VSX))
584 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr) / 2; i++)
585 vcpu_vsx[i] = thread_fpr[get_fpr_index(i) + 1];
589 #ifdef CONFIG_ALTIVEC
591 if (current->thread.regs->msr & MSR_VEC)
592 giveup_altivec(current);
593 memcpy(vcpu->arch.vr, t->vr_state.vr, sizeof(vcpu->arch.vr));
594 vcpu->arch.vscr = t->vr_state.vscr;
598 vcpu->arch.guest_owned_ext &= ~(msr | MSR_VSX);
599 kvmppc_recalc_shadow_msr(vcpu);
602 static int kvmppc_read_inst(struct kvm_vcpu *vcpu)
604 ulong srr0 = kvmppc_get_pc(vcpu);
605 u32 last_inst = kvmppc_get_last_inst(vcpu);
608 ret = kvmppc_ld(vcpu, &srr0, sizeof(u32), &last_inst, false);
609 if (ret == -ENOENT) {
610 ulong msr = vcpu->arch.shared->msr;
612 msr = kvmppc_set_field(msr, 33, 33, 1);
613 msr = kvmppc_set_field(msr, 34, 36, 0);
614 vcpu->arch.shared->msr = kvmppc_set_field(msr, 42, 47, 0);
615 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_INST_STORAGE);
616 return EMULATE_AGAIN;
622 static int kvmppc_check_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr)
625 /* Need to do paired single emulation? */
626 if (!(vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE))
629 /* Read out the instruction */
630 if (kvmppc_read_inst(vcpu) == EMULATE_DONE)
631 /* Need to emulate */
634 return EMULATE_AGAIN;
637 /* Handle external providers (FPU, Altivec, VSX) */
638 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
641 struct thread_struct *t = ¤t->thread;
642 u64 *vcpu_fpr = vcpu->arch.fpr;
644 u64 *vcpu_vsx = vcpu->arch.vsr;
646 u64 *thread_fpr = &t->fp_state.fpr[0][0];
649 /* When we have paired singles, we emulate in software */
650 if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
653 if (!(vcpu->arch.shared->msr & msr)) {
654 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
658 if (msr == MSR_VSX) {
659 /* No VSX? Give an illegal instruction interrupt */
661 if (!cpu_has_feature(CPU_FTR_VSX))
664 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
669 * We have to load up all the FP and VMX registers before
670 * we can let the guest use VSX instructions.
672 msr = MSR_FP | MSR_VEC | MSR_VSX;
675 /* See if we already own all the ext(s) needed */
676 msr &= ~vcpu->arch.guest_owned_ext;
681 printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
685 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
686 thread_fpr[get_fpr_index(i)] = vcpu_fpr[i];
688 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr) / 2; i++)
689 thread_fpr[get_fpr_index(i) + 1] = vcpu_vsx[i];
691 t->fp_state.fpscr = vcpu->arch.fpscr;
693 kvmppc_load_up_fpu();
697 #ifdef CONFIG_ALTIVEC
698 memcpy(t->vr_state.vr, vcpu->arch.vr, sizeof(vcpu->arch.vr));
699 t->vr_state.vscr = vcpu->arch.vscr;
701 kvmppc_load_up_altivec();
705 current->thread.regs->msr |= msr;
706 vcpu->arch.guest_owned_ext |= msr;
707 kvmppc_recalc_shadow_msr(vcpu);
713 * Kernel code using FP or VMX could have flushed guest state to
714 * the thread_struct; if so, get it back now.
716 static void kvmppc_handle_lost_ext(struct kvm_vcpu *vcpu)
718 unsigned long lost_ext;
720 lost_ext = vcpu->arch.guest_owned_ext & ~current->thread.regs->msr;
724 if (lost_ext & MSR_FP)
725 kvmppc_load_up_fpu();
726 #ifdef CONFIG_ALTIVEC
727 if (lost_ext & MSR_VEC)
728 kvmppc_load_up_altivec();
730 current->thread.regs->msr |= lost_ext;
733 int kvmppc_handle_exit_pr(struct kvm_run *run, struct kvm_vcpu *vcpu,
734 unsigned int exit_nr)
739 vcpu->stat.sum_exits++;
741 run->exit_reason = KVM_EXIT_UNKNOWN;
742 run->ready_for_interrupt_injection = 1;
744 /* We get here with MSR.EE=1 */
746 trace_kvm_exit(exit_nr, vcpu);
750 case BOOK3S_INTERRUPT_INST_STORAGE:
752 ulong shadow_srr1 = vcpu->arch.shadow_srr1;
753 vcpu->stat.pf_instruc++;
755 #ifdef CONFIG_PPC_BOOK3S_32
756 /* We set segments as unused segments when invalidating them. So
757 * treat the respective fault as segment fault. */
759 struct kvmppc_book3s_shadow_vcpu *svcpu;
762 svcpu = svcpu_get(vcpu);
763 sr = svcpu->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT];
765 if (sr == SR_INVALID) {
766 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
773 /* only care about PTEG not found errors, but leave NX alone */
774 if (shadow_srr1 & 0x40000000) {
775 int idx = srcu_read_lock(&vcpu->kvm->srcu);
776 r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr);
777 srcu_read_unlock(&vcpu->kvm->srcu, idx);
778 vcpu->stat.sp_instruc++;
779 } else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
780 (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
782 * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
783 * so we can't use the NX bit inside the guest. Let's cross our fingers,
784 * that no guest that needs the dcbz hack does NX.
786 kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
789 vcpu->arch.shared->msr |= shadow_srr1 & 0x58000000;
790 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
795 case BOOK3S_INTERRUPT_DATA_STORAGE:
797 ulong dar = kvmppc_get_fault_dar(vcpu);
798 u32 fault_dsisr = vcpu->arch.fault_dsisr;
799 vcpu->stat.pf_storage++;
801 #ifdef CONFIG_PPC_BOOK3S_32
802 /* We set segments as unused segments when invalidating them. So
803 * treat the respective fault as segment fault. */
805 struct kvmppc_book3s_shadow_vcpu *svcpu;
808 svcpu = svcpu_get(vcpu);
809 sr = svcpu->sr[dar >> SID_SHIFT];
811 if (sr == SR_INVALID) {
812 kvmppc_mmu_map_segment(vcpu, dar);
820 * We need to handle missing shadow PTEs, and
821 * protection faults due to us mapping a page read-only
822 * when the guest thinks it is writable.
824 if (fault_dsisr & (DSISR_NOHPTE | DSISR_PROTFAULT)) {
825 int idx = srcu_read_lock(&vcpu->kvm->srcu);
826 r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr);
827 srcu_read_unlock(&vcpu->kvm->srcu, idx);
829 vcpu->arch.shared->dar = dar;
830 vcpu->arch.shared->dsisr = fault_dsisr;
831 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
836 case BOOK3S_INTERRUPT_DATA_SEGMENT:
837 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
838 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
839 kvmppc_book3s_queue_irqprio(vcpu,
840 BOOK3S_INTERRUPT_DATA_SEGMENT);
844 case BOOK3S_INTERRUPT_INST_SEGMENT:
845 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
846 kvmppc_book3s_queue_irqprio(vcpu,
847 BOOK3S_INTERRUPT_INST_SEGMENT);
851 /* We're good on these - the host merely wanted to get our attention */
852 case BOOK3S_INTERRUPT_DECREMENTER:
853 case BOOK3S_INTERRUPT_HV_DECREMENTER:
854 vcpu->stat.dec_exits++;
857 case BOOK3S_INTERRUPT_EXTERNAL:
858 case BOOK3S_INTERRUPT_EXTERNAL_LEVEL:
859 case BOOK3S_INTERRUPT_EXTERNAL_HV:
860 vcpu->stat.ext_intr_exits++;
863 case BOOK3S_INTERRUPT_PERFMON:
866 case BOOK3S_INTERRUPT_PROGRAM:
867 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
869 enum emulation_result er;
873 flags = vcpu->arch.shadow_srr1 & 0x1f0000ull;
875 if (vcpu->arch.shared->msr & MSR_PR) {
877 printk(KERN_INFO "Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
879 if ((kvmppc_get_last_inst(vcpu) & 0xff0007ff) !=
880 (INS_DCBZ & 0xfffffff7)) {
881 kvmppc_core_queue_program(vcpu, flags);
887 vcpu->stat.emulated_inst_exits++;
888 er = kvmppc_emulate_instruction(run, vcpu);
897 printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
898 __func__, kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
899 kvmppc_core_queue_program(vcpu, flags);
902 case EMULATE_DO_MMIO:
903 run->exit_reason = KVM_EXIT_MMIO;
906 case EMULATE_EXIT_USER:
914 case BOOK3S_INTERRUPT_SYSCALL:
915 if (vcpu->arch.papr_enabled &&
916 (kvmppc_get_last_sc(vcpu) == 0x44000022) &&
917 !(vcpu->arch.shared->msr & MSR_PR)) {
918 /* SC 1 papr hypercalls */
919 ulong cmd = kvmppc_get_gpr(vcpu, 3);
922 #ifdef CONFIG_PPC_BOOK3S_64
923 if (kvmppc_h_pr(vcpu, cmd) == EMULATE_DONE) {
929 run->papr_hcall.nr = cmd;
930 for (i = 0; i < 9; ++i) {
931 ulong gpr = kvmppc_get_gpr(vcpu, 4 + i);
932 run->papr_hcall.args[i] = gpr;
934 run->exit_reason = KVM_EXIT_PAPR_HCALL;
935 vcpu->arch.hcall_needed = 1;
937 } else if (vcpu->arch.osi_enabled &&
938 (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
939 (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
941 u64 *gprs = run->osi.gprs;
944 run->exit_reason = KVM_EXIT_OSI;
945 for (i = 0; i < 32; i++)
946 gprs[i] = kvmppc_get_gpr(vcpu, i);
947 vcpu->arch.osi_needed = 1;
949 } else if (!(vcpu->arch.shared->msr & MSR_PR) &&
950 (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
951 /* KVM PV hypercalls */
952 kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
956 vcpu->stat.syscall_exits++;
957 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
961 case BOOK3S_INTERRUPT_FP_UNAVAIL:
962 case BOOK3S_INTERRUPT_ALTIVEC:
963 case BOOK3S_INTERRUPT_VSX:
968 case BOOK3S_INTERRUPT_FP_UNAVAIL: ext_msr = MSR_FP; break;
969 case BOOK3S_INTERRUPT_ALTIVEC: ext_msr = MSR_VEC; break;
970 case BOOK3S_INTERRUPT_VSX: ext_msr = MSR_VSX; break;
973 switch (kvmppc_check_ext(vcpu, exit_nr)) {
975 /* everything ok - let's enable the ext */
976 r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
979 /* we need to emulate this instruction */
980 goto program_interrupt;
983 /* nothing to worry about - go again */
988 case BOOK3S_INTERRUPT_ALIGNMENT:
989 if (kvmppc_read_inst(vcpu) == EMULATE_DONE) {
990 vcpu->arch.shared->dsisr = kvmppc_alignment_dsisr(vcpu,
991 kvmppc_get_last_inst(vcpu));
992 vcpu->arch.shared->dar = kvmppc_alignment_dar(vcpu,
993 kvmppc_get_last_inst(vcpu));
994 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
998 case BOOK3S_INTERRUPT_MACHINE_CHECK:
999 case BOOK3S_INTERRUPT_TRACE:
1000 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1005 ulong shadow_srr1 = vcpu->arch.shadow_srr1;
1006 /* Ugh - bork here! What did we get? */
1007 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
1008 exit_nr, kvmppc_get_pc(vcpu), shadow_srr1);
1015 if (!(r & RESUME_HOST)) {
1016 /* To avoid clobbering exit_reason, only check for signals if
1017 * we aren't already exiting to userspace for some other
1021 * Interrupts could be timers for the guest which we have to
1022 * inject again, so let's postpone them until we're in the guest
1023 * and if we really did time things so badly, then we just exit
1024 * again due to a host external interrupt.
1026 local_irq_disable();
1027 s = kvmppc_prepare_to_enter(vcpu);
1032 kvmppc_fix_ee_before_entry();
1034 kvmppc_handle_lost_ext(vcpu);
1037 trace_kvm_book3s_reenter(r, vcpu);
1042 static int kvm_arch_vcpu_ioctl_get_sregs_pr(struct kvm_vcpu *vcpu,
1043 struct kvm_sregs *sregs)
1045 struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1048 sregs->pvr = vcpu->arch.pvr;
1050 sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
1051 if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1052 for (i = 0; i < 64; i++) {
1053 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige | i;
1054 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1057 for (i = 0; i < 16; i++)
1058 sregs->u.s.ppc32.sr[i] = vcpu->arch.shared->sr[i];
1060 for (i = 0; i < 8; i++) {
1061 sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
1062 sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
1069 static int kvm_arch_vcpu_ioctl_set_sregs_pr(struct kvm_vcpu *vcpu,
1070 struct kvm_sregs *sregs)
1072 struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1075 kvmppc_set_pvr_pr(vcpu, sregs->pvr);
1077 vcpu3s->sdr1 = sregs->u.s.sdr1;
1078 if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1079 for (i = 0; i < 64; i++) {
1080 vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
1081 sregs->u.s.ppc64.slb[i].slbe);
1084 for (i = 0; i < 16; i++) {
1085 vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
1087 for (i = 0; i < 8; i++) {
1088 kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
1089 (u32)sregs->u.s.ppc32.ibat[i]);
1090 kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
1091 (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
1092 kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
1093 (u32)sregs->u.s.ppc32.dbat[i]);
1094 kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
1095 (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
1099 /* Flush the MMU after messing with the segments */
1100 kvmppc_mmu_pte_flush(vcpu, 0, 0);
1105 static int kvmppc_get_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1106 union kvmppc_one_reg *val)
1111 case KVM_REG_PPC_HIOR:
1112 *val = get_reg_val(id, to_book3s(vcpu)->hior);
1115 case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31: {
1116 long int i = id - KVM_REG_PPC_VSR0;
1118 if (!cpu_has_feature(CPU_FTR_VSX)) {
1122 val->vsxval[0] = vcpu->arch.fpr[i];
1123 val->vsxval[1] = vcpu->arch.vsr[i];
1126 #endif /* CONFIG_VSX */
1135 static int kvmppc_set_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1136 union kvmppc_one_reg *val)
1141 case KVM_REG_PPC_HIOR:
1142 to_book3s(vcpu)->hior = set_reg_val(id, *val);
1143 to_book3s(vcpu)->hior_explicit = true;
1146 case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31: {
1147 long int i = id - KVM_REG_PPC_VSR0;
1149 if (!cpu_has_feature(CPU_FTR_VSX)) {
1153 vcpu->arch.fpr[i] = val->vsxval[0];
1154 vcpu->arch.vsr[i] = val->vsxval[1];
1157 #endif /* CONFIG_VSX */
1166 static struct kvm_vcpu *kvmppc_core_vcpu_create_pr(struct kvm *kvm,
1169 struct kvmppc_vcpu_book3s *vcpu_book3s;
1170 struct kvm_vcpu *vcpu;
1174 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1178 vcpu_book3s = vzalloc(sizeof(struct kvmppc_vcpu_book3s));
1181 vcpu->arch.book3s = vcpu_book3s;
1183 #ifdef CONFIG_KVM_BOOK3S_32
1184 vcpu->arch.shadow_vcpu =
1185 kzalloc(sizeof(*vcpu->arch.shadow_vcpu), GFP_KERNEL);
1186 if (!vcpu->arch.shadow_vcpu)
1190 err = kvm_vcpu_init(vcpu, kvm, id);
1192 goto free_shadow_vcpu;
1195 p = __get_free_page(GFP_KERNEL|__GFP_ZERO);
1198 /* the real shared page fills the last 4k of our page */
1199 vcpu->arch.shared = (void *)(p + PAGE_SIZE - 4096);
1201 #ifdef CONFIG_PPC_BOOK3S_64
1203 * Default to the same as the host if we're on sufficiently
1204 * recent machine that we have 1TB segments;
1205 * otherwise default to PPC970FX.
1207 vcpu->arch.pvr = 0x3C0301;
1208 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1209 vcpu->arch.pvr = mfspr(SPRN_PVR);
1211 /* default to book3s_32 (750) */
1212 vcpu->arch.pvr = 0x84202;
1214 kvmppc_set_pvr_pr(vcpu, vcpu->arch.pvr);
1215 vcpu->arch.slb_nr = 64;
1217 vcpu->arch.shadow_msr = MSR_USER64;
1219 err = kvmppc_mmu_init(vcpu);
1226 kvm_vcpu_uninit(vcpu);
1228 #ifdef CONFIG_KVM_BOOK3S_32
1229 kfree(vcpu->arch.shadow_vcpu);
1234 kmem_cache_free(kvm_vcpu_cache, vcpu);
1236 return ERR_PTR(err);
1239 static void kvmppc_core_vcpu_free_pr(struct kvm_vcpu *vcpu)
1241 struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
1243 free_page((unsigned long)vcpu->arch.shared & PAGE_MASK);
1244 kvm_vcpu_uninit(vcpu);
1245 #ifdef CONFIG_KVM_BOOK3S_32
1246 kfree(vcpu->arch.shadow_vcpu);
1249 kmem_cache_free(kvm_vcpu_cache, vcpu);
1252 static int kvmppc_vcpu_run_pr(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1255 struct thread_fp_state fp;
1257 #ifdef CONFIG_ALTIVEC
1258 struct thread_vr_state vr;
1259 unsigned long uninitialized_var(vrsave);
1267 /* Check if we can run the vcpu at all */
1268 if (!vcpu->arch.sane) {
1269 kvm_run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1275 * Interrupts could be timers for the guest which we have to inject
1276 * again, so let's postpone them until we're in the guest and if we
1277 * really did time things so badly, then we just exit again due to
1278 * a host external interrupt.
1280 local_irq_disable();
1281 ret = kvmppc_prepare_to_enter(vcpu);
1287 /* Save FPU state in stack */
1288 if (current->thread.regs->msr & MSR_FP)
1289 giveup_fpu(current);
1290 fp = current->thread.fp_state;
1291 fpexc_mode = current->thread.fpexc_mode;
1293 #ifdef CONFIG_ALTIVEC
1294 /* Save Altivec state in stack */
1295 used_vr = current->thread.used_vr;
1297 if (current->thread.regs->msr & MSR_VEC)
1298 giveup_altivec(current);
1299 vr = current->thread.vr_state;
1300 vrsave = current->thread.vrsave;
1305 /* Save VSX state in stack */
1306 used_vsr = current->thread.used_vsr;
1307 if (used_vsr && (current->thread.regs->msr & MSR_VSX))
1308 __giveup_vsx(current);
1311 /* Remember the MSR with disabled extensions */
1312 ext_msr = current->thread.regs->msr;
1314 /* Preload FPU if it's enabled */
1315 if (vcpu->arch.shared->msr & MSR_FP)
1316 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
1318 kvmppc_fix_ee_before_entry();
1320 ret = __kvmppc_vcpu_run(kvm_run, vcpu);
1322 /* No need for kvm_guest_exit. It's done in handle_exit.
1323 We also get here with interrupts enabled. */
1325 /* Make sure we save the guest FPU/Altivec/VSX state */
1326 kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
1328 current->thread.regs->msr = ext_msr;
1330 /* Restore FPU/VSX state from stack */
1331 current->thread.fp_state = fp;
1332 current->thread.fpexc_mode = fpexc_mode;
1334 #ifdef CONFIG_ALTIVEC
1335 /* Restore Altivec state from stack */
1336 if (used_vr && current->thread.used_vr) {
1337 current->thread.vr_state = vr;
1338 current->thread.vrsave = vrsave;
1340 current->thread.used_vr = used_vr;
1344 current->thread.used_vsr = used_vsr;
1348 vcpu->mode = OUTSIDE_GUEST_MODE;
1353 * Get (and clear) the dirty memory log for a memory slot.
1355 static int kvm_vm_ioctl_get_dirty_log_pr(struct kvm *kvm,
1356 struct kvm_dirty_log *log)
1358 struct kvm_memory_slot *memslot;
1359 struct kvm_vcpu *vcpu;
1365 mutex_lock(&kvm->slots_lock);
1367 r = kvm_get_dirty_log(kvm, log, &is_dirty);
1371 /* If nothing is dirty, don't bother messing with page tables. */
1373 memslot = id_to_memslot(kvm->memslots, log->slot);
1375 ga = memslot->base_gfn << PAGE_SHIFT;
1376 ga_end = ga + (memslot->npages << PAGE_SHIFT);
1378 kvm_for_each_vcpu(n, vcpu, kvm)
1379 kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
1381 n = kvm_dirty_bitmap_bytes(memslot);
1382 memset(memslot->dirty_bitmap, 0, n);
1387 mutex_unlock(&kvm->slots_lock);
1391 static void kvmppc_core_flush_memslot_pr(struct kvm *kvm,
1392 struct kvm_memory_slot *memslot)
1397 static int kvmppc_core_prepare_memory_region_pr(struct kvm *kvm,
1398 struct kvm_memory_slot *memslot,
1399 struct kvm_userspace_memory_region *mem)
1404 static void kvmppc_core_commit_memory_region_pr(struct kvm *kvm,
1405 struct kvm_userspace_memory_region *mem,
1406 const struct kvm_memory_slot *old)
1411 static void kvmppc_core_free_memslot_pr(struct kvm_memory_slot *free,
1412 struct kvm_memory_slot *dont)
1417 static int kvmppc_core_create_memslot_pr(struct kvm_memory_slot *slot,
1418 unsigned long npages)
1425 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1426 struct kvm_ppc_smmu_info *info)
1429 struct kvm_vcpu *vcpu;
1433 /* SLB is always 64 entries */
1434 info->slb_size = 64;
1436 /* Standard 4k base page size segment */
1437 info->sps[0].page_shift = 12;
1438 info->sps[0].slb_enc = 0;
1439 info->sps[0].enc[0].page_shift = 12;
1440 info->sps[0].enc[0].pte_enc = 0;
1443 * 64k large page size.
1444 * We only want to put this in if the CPUs we're emulating
1445 * support it, but unfortunately we don't have a vcpu easily
1446 * to hand here to test. Just pick the first vcpu, and if
1447 * that doesn't exist yet, report the minimum capability,
1448 * i.e., no 64k pages.
1449 * 1T segment support goes along with 64k pages.
1452 vcpu = kvm_get_vcpu(kvm, 0);
1453 if (vcpu && (vcpu->arch.hflags & BOOK3S_HFLAG_MULTI_PGSIZE)) {
1454 info->flags = KVM_PPC_1T_SEGMENTS;
1455 info->sps[i].page_shift = 16;
1456 info->sps[i].slb_enc = SLB_VSID_L | SLB_VSID_LP_01;
1457 info->sps[i].enc[0].page_shift = 16;
1458 info->sps[i].enc[0].pte_enc = 1;
1462 /* Standard 16M large page size segment */
1463 info->sps[i].page_shift = 24;
1464 info->sps[i].slb_enc = SLB_VSID_L;
1465 info->sps[i].enc[0].page_shift = 24;
1466 info->sps[i].enc[0].pte_enc = 0;
1471 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1472 struct kvm_ppc_smmu_info *info)
1474 /* We should not get called */
1477 #endif /* CONFIG_PPC64 */
1479 static unsigned int kvm_global_user_count = 0;
1480 static DEFINE_SPINLOCK(kvm_global_user_count_lock);
1482 static int kvmppc_core_init_vm_pr(struct kvm *kvm)
1484 mutex_init(&kvm->arch.hpt_mutex);
1486 if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
1487 spin_lock(&kvm_global_user_count_lock);
1488 if (++kvm_global_user_count == 1)
1489 pSeries_disable_reloc_on_exc();
1490 spin_unlock(&kvm_global_user_count_lock);
1495 static void kvmppc_core_destroy_vm_pr(struct kvm *kvm)
1498 WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
1501 if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
1502 spin_lock(&kvm_global_user_count_lock);
1503 BUG_ON(kvm_global_user_count == 0);
1504 if (--kvm_global_user_count == 0)
1505 pSeries_enable_reloc_on_exc();
1506 spin_unlock(&kvm_global_user_count_lock);
1510 static int kvmppc_core_check_processor_compat_pr(void)
1512 /* we are always compatible */
1516 static long kvm_arch_vm_ioctl_pr(struct file *filp,
1517 unsigned int ioctl, unsigned long arg)
1522 static struct kvmppc_ops kvm_ops_pr = {
1523 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_pr,
1524 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_pr,
1525 .get_one_reg = kvmppc_get_one_reg_pr,
1526 .set_one_reg = kvmppc_set_one_reg_pr,
1527 .vcpu_load = kvmppc_core_vcpu_load_pr,
1528 .vcpu_put = kvmppc_core_vcpu_put_pr,
1529 .set_msr = kvmppc_set_msr_pr,
1530 .vcpu_run = kvmppc_vcpu_run_pr,
1531 .vcpu_create = kvmppc_core_vcpu_create_pr,
1532 .vcpu_free = kvmppc_core_vcpu_free_pr,
1533 .check_requests = kvmppc_core_check_requests_pr,
1534 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_pr,
1535 .flush_memslot = kvmppc_core_flush_memslot_pr,
1536 .prepare_memory_region = kvmppc_core_prepare_memory_region_pr,
1537 .commit_memory_region = kvmppc_core_commit_memory_region_pr,
1538 .unmap_hva = kvm_unmap_hva_pr,
1539 .unmap_hva_range = kvm_unmap_hva_range_pr,
1540 .age_hva = kvm_age_hva_pr,
1541 .test_age_hva = kvm_test_age_hva_pr,
1542 .set_spte_hva = kvm_set_spte_hva_pr,
1543 .mmu_destroy = kvmppc_mmu_destroy_pr,
1544 .free_memslot = kvmppc_core_free_memslot_pr,
1545 .create_memslot = kvmppc_core_create_memslot_pr,
1546 .init_vm = kvmppc_core_init_vm_pr,
1547 .destroy_vm = kvmppc_core_destroy_vm_pr,
1548 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_pr,
1549 .emulate_op = kvmppc_core_emulate_op_pr,
1550 .emulate_mtspr = kvmppc_core_emulate_mtspr_pr,
1551 .emulate_mfspr = kvmppc_core_emulate_mfspr_pr,
1552 .fast_vcpu_kick = kvm_vcpu_kick,
1553 .arch_vm_ioctl = kvm_arch_vm_ioctl_pr,
1557 int kvmppc_book3s_init_pr(void)
1561 r = kvmppc_core_check_processor_compat_pr();
1565 kvm_ops_pr.owner = THIS_MODULE;
1566 kvmppc_pr_ops = &kvm_ops_pr;
1568 r = kvmppc_mmu_hpte_sysinit();
1572 void kvmppc_book3s_exit_pr(void)
1574 kvmppc_pr_ops = NULL;
1575 kvmppc_mmu_hpte_sysexit();
1579 * We only support separate modules for book3s 64
1581 #ifdef CONFIG_PPC_BOOK3S_64
1583 module_init(kvmppc_book3s_init_pr);
1584 module_exit(kvmppc_book3s_exit_pr);
1586 MODULE_LICENSE("GPL");