2 * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
5 * Alexander Graf <agraf@suse.de>
6 * Kevin Wolf <mail@kevin-wolf.de>
9 * This file is derived from arch/powerpc/kvm/44x.c,
10 * by Hollis Blanchard <hollisb@us.ibm.com>.
12 * This program is free software; you can redistribute it and/or modify
13 * it under the terms of the GNU General Public License, version 2, as
14 * published by the Free Software Foundation.
17 #include <linux/kvm_host.h>
18 #include <linux/err.h>
21 #include <asm/cputable.h>
22 #include <asm/cacheflush.h>
23 #include <asm/tlbflush.h>
24 #include <asm/uaccess.h>
26 #include <asm/kvm_ppc.h>
27 #include <asm/kvm_book3s.h>
28 #include <asm/mmu_context.h>
29 #include <linux/gfp.h>
30 #include <linux/sched.h>
31 #include <linux/vmalloc.h>
32 #include <linux/highmem.h>
34 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
36 /* #define EXIT_DEBUG */
37 /* #define EXIT_DEBUG_SIMPLE */
38 /* #define DEBUG_EXT */
40 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
43 struct kvm_stats_debugfs_item debugfs_entries[] = {
44 { "exits", VCPU_STAT(sum_exits) },
45 { "mmio", VCPU_STAT(mmio_exits) },
46 { "sig", VCPU_STAT(signal_exits) },
47 { "sysc", VCPU_STAT(syscall_exits) },
48 { "inst_emu", VCPU_STAT(emulated_inst_exits) },
49 { "dec", VCPU_STAT(dec_exits) },
50 { "ext_intr", VCPU_STAT(ext_intr_exits) },
51 { "queue_intr", VCPU_STAT(queue_intr) },
52 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
53 { "pf_storage", VCPU_STAT(pf_storage) },
54 { "sp_storage", VCPU_STAT(sp_storage) },
55 { "pf_instruc", VCPU_STAT(pf_instruc) },
56 { "sp_instruc", VCPU_STAT(sp_instruc) },
57 { "ld", VCPU_STAT(ld) },
58 { "ld_slow", VCPU_STAT(ld_slow) },
59 { "st", VCPU_STAT(st) },
60 { "st_slow", VCPU_STAT(st_slow) },
64 void kvmppc_core_load_host_debugstate(struct kvm_vcpu *vcpu)
68 void kvmppc_core_load_guest_debugstate(struct kvm_vcpu *vcpu)
72 void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
74 #ifdef CONFIG_PPC_BOOK3S_64
75 memcpy(to_svcpu(vcpu)->slb, to_book3s(vcpu)->slb_shadow, sizeof(to_svcpu(vcpu)->slb));
76 memcpy(&get_paca()->shadow_vcpu, to_book3s(vcpu)->shadow_vcpu,
77 sizeof(get_paca()->shadow_vcpu));
78 to_svcpu(vcpu)->slb_max = to_book3s(vcpu)->slb_shadow_max;
81 #ifdef CONFIG_PPC_BOOK3S_32
82 current->thread.kvm_shadow_vcpu = to_book3s(vcpu)->shadow_vcpu;
86 void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
88 #ifdef CONFIG_PPC_BOOK3S_64
89 memcpy(to_book3s(vcpu)->slb_shadow, to_svcpu(vcpu)->slb, sizeof(to_svcpu(vcpu)->slb));
90 memcpy(to_book3s(vcpu)->shadow_vcpu, &get_paca()->shadow_vcpu,
91 sizeof(get_paca()->shadow_vcpu));
92 to_book3s(vcpu)->slb_shadow_max = to_svcpu(vcpu)->slb_max;
95 kvmppc_giveup_ext(vcpu, MSR_FP);
96 kvmppc_giveup_ext(vcpu, MSR_VEC);
97 kvmppc_giveup_ext(vcpu, MSR_VSX);
100 #if defined(EXIT_DEBUG)
101 static u32 kvmppc_get_dec(struct kvm_vcpu *vcpu)
103 u64 jd = mftb() - vcpu->arch.dec_jiffies;
104 return vcpu->arch.dec - jd;
108 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu *vcpu)
110 vcpu->arch.shadow_msr = vcpu->arch.msr;
111 /* Guest MSR values */
112 vcpu->arch.shadow_msr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE |
114 /* Process MSR values */
115 vcpu->arch.shadow_msr |= MSR_ME | MSR_RI | MSR_IR | MSR_DR | MSR_PR |
117 /* External providers the guest reserved */
118 vcpu->arch.shadow_msr |= (vcpu->arch.msr & vcpu->arch.guest_owned_ext);
119 /* 64-bit Process MSR values */
120 #ifdef CONFIG_PPC_BOOK3S_64
121 vcpu->arch.shadow_msr |= MSR_ISF | MSR_HV;
125 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
127 ulong old_msr = vcpu->arch.msr;
130 printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
133 msr &= to_book3s(vcpu)->msr_mask;
134 vcpu->arch.msr = msr;
135 kvmppc_recalc_shadow_msr(vcpu);
137 if (msr & (MSR_WE|MSR_POW)) {
138 if (!vcpu->arch.pending_exceptions) {
139 kvm_vcpu_block(vcpu);
140 vcpu->stat.halt_wakeup++;
144 if (((vcpu->arch.msr & (MSR_IR|MSR_DR)) != (old_msr & (MSR_IR|MSR_DR))) ||
145 (vcpu->arch.msr & MSR_PR) != (old_msr & MSR_PR)) {
146 bool dr = (vcpu->arch.msr & MSR_DR) ? true : false;
147 bool ir = (vcpu->arch.msr & MSR_IR) ? true : false;
149 /* Flush split mode PTEs */
151 kvmppc_mmu_pte_vflush(vcpu, VSID_SPLIT_MASK,
154 kvmppc_mmu_flush_segments(vcpu);
155 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
158 /* Preload FPU if it's enabled */
159 if (vcpu->arch.msr & MSR_FP)
160 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
163 void kvmppc_inject_interrupt(struct kvm_vcpu *vcpu, int vec, u64 flags)
165 vcpu->arch.srr0 = kvmppc_get_pc(vcpu);
166 vcpu->arch.srr1 = vcpu->arch.msr | flags;
167 kvmppc_set_pc(vcpu, to_book3s(vcpu)->hior + vec);
168 vcpu->arch.mmu.reset_msr(vcpu);
171 static int kvmppc_book3s_vec2irqprio(unsigned int vec)
176 case 0x100: prio = BOOK3S_IRQPRIO_SYSTEM_RESET; break;
177 case 0x200: prio = BOOK3S_IRQPRIO_MACHINE_CHECK; break;
178 case 0x300: prio = BOOK3S_IRQPRIO_DATA_STORAGE; break;
179 case 0x380: prio = BOOK3S_IRQPRIO_DATA_SEGMENT; break;
180 case 0x400: prio = BOOK3S_IRQPRIO_INST_STORAGE; break;
181 case 0x480: prio = BOOK3S_IRQPRIO_INST_SEGMENT; break;
182 case 0x500: prio = BOOK3S_IRQPRIO_EXTERNAL; break;
183 case 0x600: prio = BOOK3S_IRQPRIO_ALIGNMENT; break;
184 case 0x700: prio = BOOK3S_IRQPRIO_PROGRAM; break;
185 case 0x800: prio = BOOK3S_IRQPRIO_FP_UNAVAIL; break;
186 case 0x900: prio = BOOK3S_IRQPRIO_DECREMENTER; break;
187 case 0xc00: prio = BOOK3S_IRQPRIO_SYSCALL; break;
188 case 0xd00: prio = BOOK3S_IRQPRIO_DEBUG; break;
189 case 0xf20: prio = BOOK3S_IRQPRIO_ALTIVEC; break;
190 case 0xf40: prio = BOOK3S_IRQPRIO_VSX; break;
191 default: prio = BOOK3S_IRQPRIO_MAX; break;
197 static void kvmppc_book3s_dequeue_irqprio(struct kvm_vcpu *vcpu,
200 clear_bit(kvmppc_book3s_vec2irqprio(vec),
201 &vcpu->arch.pending_exceptions);
204 void kvmppc_book3s_queue_irqprio(struct kvm_vcpu *vcpu, unsigned int vec)
206 vcpu->stat.queue_intr++;
208 set_bit(kvmppc_book3s_vec2irqprio(vec),
209 &vcpu->arch.pending_exceptions);
211 printk(KERN_INFO "Queueing interrupt %x\n", vec);
216 void kvmppc_core_queue_program(struct kvm_vcpu *vcpu, ulong flags)
218 to_book3s(vcpu)->prog_flags = flags;
219 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_PROGRAM);
222 void kvmppc_core_queue_dec(struct kvm_vcpu *vcpu)
224 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_DECREMENTER);
227 int kvmppc_core_pending_dec(struct kvm_vcpu *vcpu)
229 return test_bit(BOOK3S_INTERRUPT_DECREMENTER >> 7, &vcpu->arch.pending_exceptions);
232 void kvmppc_core_dequeue_dec(struct kvm_vcpu *vcpu)
234 kvmppc_book3s_dequeue_irqprio(vcpu, BOOK3S_INTERRUPT_DECREMENTER);
237 void kvmppc_core_queue_external(struct kvm_vcpu *vcpu,
238 struct kvm_interrupt *irq)
240 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_EXTERNAL);
243 void kvmppc_core_dequeue_external(struct kvm_vcpu *vcpu,
244 struct kvm_interrupt *irq)
246 kvmppc_book3s_dequeue_irqprio(vcpu, BOOK3S_INTERRUPT_EXTERNAL);
249 int kvmppc_book3s_irqprio_deliver(struct kvm_vcpu *vcpu, unsigned int priority)
256 case BOOK3S_IRQPRIO_DECREMENTER:
257 deliver = vcpu->arch.msr & MSR_EE;
258 vec = BOOK3S_INTERRUPT_DECREMENTER;
260 case BOOK3S_IRQPRIO_EXTERNAL:
261 deliver = vcpu->arch.msr & MSR_EE;
262 vec = BOOK3S_INTERRUPT_EXTERNAL;
264 case BOOK3S_IRQPRIO_SYSTEM_RESET:
265 vec = BOOK3S_INTERRUPT_SYSTEM_RESET;
267 case BOOK3S_IRQPRIO_MACHINE_CHECK:
268 vec = BOOK3S_INTERRUPT_MACHINE_CHECK;
270 case BOOK3S_IRQPRIO_DATA_STORAGE:
271 vec = BOOK3S_INTERRUPT_DATA_STORAGE;
273 case BOOK3S_IRQPRIO_INST_STORAGE:
274 vec = BOOK3S_INTERRUPT_INST_STORAGE;
276 case BOOK3S_IRQPRIO_DATA_SEGMENT:
277 vec = BOOK3S_INTERRUPT_DATA_SEGMENT;
279 case BOOK3S_IRQPRIO_INST_SEGMENT:
280 vec = BOOK3S_INTERRUPT_INST_SEGMENT;
282 case BOOK3S_IRQPRIO_ALIGNMENT:
283 vec = BOOK3S_INTERRUPT_ALIGNMENT;
285 case BOOK3S_IRQPRIO_PROGRAM:
286 vec = BOOK3S_INTERRUPT_PROGRAM;
287 flags = to_book3s(vcpu)->prog_flags;
289 case BOOK3S_IRQPRIO_VSX:
290 vec = BOOK3S_INTERRUPT_VSX;
292 case BOOK3S_IRQPRIO_ALTIVEC:
293 vec = BOOK3S_INTERRUPT_ALTIVEC;
295 case BOOK3S_IRQPRIO_FP_UNAVAIL:
296 vec = BOOK3S_INTERRUPT_FP_UNAVAIL;
298 case BOOK3S_IRQPRIO_SYSCALL:
299 vec = BOOK3S_INTERRUPT_SYSCALL;
301 case BOOK3S_IRQPRIO_DEBUG:
302 vec = BOOK3S_INTERRUPT_TRACE;
304 case BOOK3S_IRQPRIO_PERFORMANCE_MONITOR:
305 vec = BOOK3S_INTERRUPT_PERFMON;
309 printk(KERN_ERR "KVM: Unknown interrupt: 0x%x\n", priority);
314 printk(KERN_INFO "Deliver interrupt 0x%x? %x\n", vec, deliver);
318 kvmppc_inject_interrupt(vcpu, vec, flags);
323 void kvmppc_core_deliver_interrupts(struct kvm_vcpu *vcpu)
325 unsigned long *pending = &vcpu->arch.pending_exceptions;
326 unsigned int priority;
329 if (vcpu->arch.pending_exceptions)
330 printk(KERN_EMERG "KVM: Check pending: %lx\n", vcpu->arch.pending_exceptions);
332 priority = __ffs(*pending);
333 while (priority <= (sizeof(unsigned int) * 8)) {
334 if (kvmppc_book3s_irqprio_deliver(vcpu, priority) &&
335 (priority != BOOK3S_IRQPRIO_DECREMENTER)) {
336 /* DEC interrupts get cleared by mtdec */
337 clear_bit(priority, &vcpu->arch.pending_exceptions);
341 priority = find_next_bit(pending,
342 BITS_PER_BYTE * sizeof(*pending),
347 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
349 vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
350 vcpu->arch.pvr = pvr;
351 if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
352 kvmppc_mmu_book3s_64_init(vcpu);
353 to_book3s(vcpu)->hior = 0xfff00000;
354 to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
356 kvmppc_mmu_book3s_32_init(vcpu);
357 to_book3s(vcpu)->hior = 0;
358 to_book3s(vcpu)->msr_mask = 0xffffffffULL;
361 /* If we are in hypervisor level on 970, we can tell the CPU to
362 * treat DCBZ as 32 bytes store */
363 vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
364 if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
365 !strcmp(cur_cpu_spec->platform, "ppc970"))
366 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
368 /* Cell performs badly if MSR_FEx are set. So let's hope nobody
369 really needs them in a VM on Cell and force disable them. */
370 if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be"))
371 to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1);
374 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
375 * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
376 * emulate 32 bytes dcbz length.
378 * The Book3s_64 inventors also realized this case and implemented a special bit
379 * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
381 * My approach here is to patch the dcbz instruction on executing pages.
383 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
390 hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
391 if (is_error_page(hpage))
394 hpage_offset = pte->raddr & ~PAGE_MASK;
395 hpage_offset &= ~0xFFFULL;
399 page = kmap_atomic(hpage, KM_USER0);
401 /* patch dcbz into reserved instruction, so we trap */
402 for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
403 if ((page[i] & 0xff0007ff) == INS_DCBZ)
404 page[i] &= 0xfffffff7;
406 kunmap_atomic(page, KM_USER0);
410 static int kvmppc_xlate(struct kvm_vcpu *vcpu, ulong eaddr, bool data,
411 struct kvmppc_pte *pte)
413 int relocated = (vcpu->arch.msr & (data ? MSR_DR : MSR_IR));
417 r = vcpu->arch.mmu.xlate(vcpu, eaddr, pte, data);
420 pte->raddr = eaddr & 0xffffffff;
421 pte->vpage = VSID_REAL | eaddr >> 12;
422 pte->may_read = true;
423 pte->may_write = true;
424 pte->may_execute = true;
431 static hva_t kvmppc_bad_hva(void)
436 static hva_t kvmppc_pte_to_hva(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte,
441 if (read && !pte->may_read)
444 if (!read && !pte->may_write)
447 hpage = gfn_to_hva(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
448 if (kvm_is_error_hva(hpage))
451 return hpage | (pte->raddr & ~PAGE_MASK);
453 return kvmppc_bad_hva();
456 int kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
459 struct kvmppc_pte pte;
463 if (kvmppc_xlate(vcpu, *eaddr, data, &pte))
471 if (kvm_write_guest(vcpu->kvm, pte.raddr, ptr, size))
472 return EMULATE_DO_MMIO;
477 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
480 struct kvmppc_pte pte;
485 if (kvmppc_xlate(vcpu, *eaddr, data, &pte))
490 hva = kvmppc_pte_to_hva(vcpu, &pte, true);
491 if (kvm_is_error_hva(hva))
494 if (copy_from_user(ptr, (void __user *)hva, size)) {
495 printk(KERN_INFO "kvmppc_ld at 0x%lx failed\n", hva);
504 return EMULATE_DO_MMIO;
507 static int kvmppc_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
509 return kvm_is_visible_gfn(vcpu->kvm, gfn);
512 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
513 ulong eaddr, int vec)
515 bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
516 int r = RESUME_GUEST;
519 struct kvmppc_pte pte;
520 bool is_mmio = false;
521 bool dr = (vcpu->arch.msr & MSR_DR) ? true : false;
522 bool ir = (vcpu->arch.msr & MSR_IR) ? true : false;
524 relocated = data ? dr : ir;
526 /* Resolve real address if translation turned on */
528 page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data);
530 pte.may_execute = true;
532 pte.may_write = true;
533 pte.raddr = eaddr & 0xffffffff;
535 pte.vpage = eaddr >> 12;
538 switch (vcpu->arch.msr & (MSR_DR|MSR_IR)) {
540 pte.vpage |= VSID_REAL;
543 pte.vpage |= VSID_REAL_DR;
546 pte.vpage |= VSID_REAL_IR;
550 if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
551 (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
553 * If we do the dcbz hack, we have to NX on every execution,
554 * so we can patch the executing code. This renders our guest
557 pte.may_execute = !data;
560 if (page_found == -ENOENT) {
561 /* Page not found in guest PTE entries */
562 vcpu->arch.dear = kvmppc_get_fault_dar(vcpu);
563 to_book3s(vcpu)->dsisr = to_svcpu(vcpu)->fault_dsisr;
564 vcpu->arch.msr |= (to_svcpu(vcpu)->shadow_srr1 & 0x00000000f8000000ULL);
565 kvmppc_book3s_queue_irqprio(vcpu, vec);
566 } else if (page_found == -EPERM) {
567 /* Storage protection */
568 vcpu->arch.dear = kvmppc_get_fault_dar(vcpu);
569 to_book3s(vcpu)->dsisr = to_svcpu(vcpu)->fault_dsisr & ~DSISR_NOHPTE;
570 to_book3s(vcpu)->dsisr |= DSISR_PROTFAULT;
571 vcpu->arch.msr |= (to_svcpu(vcpu)->shadow_srr1 & 0x00000000f8000000ULL);
572 kvmppc_book3s_queue_irqprio(vcpu, vec);
573 } else if (page_found == -EINVAL) {
574 /* Page not found in guest SLB */
575 vcpu->arch.dear = kvmppc_get_fault_dar(vcpu);
576 kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
577 } else if (!is_mmio &&
578 kvmppc_visible_gfn(vcpu, pte.raddr >> PAGE_SHIFT)) {
579 /* The guest's PTE is not mapped yet. Map on the host */
580 kvmppc_mmu_map_page(vcpu, &pte);
582 vcpu->stat.sp_storage++;
583 else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
584 (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
585 kvmppc_patch_dcbz(vcpu, &pte);
588 vcpu->stat.mmio_exits++;
589 vcpu->arch.paddr_accessed = pte.raddr;
590 r = kvmppc_emulate_mmio(run, vcpu);
591 if ( r == RESUME_HOST_NV )
598 static inline int get_fpr_index(int i)
606 /* Give up external provider (FPU, Altivec, VSX) */
607 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
609 struct thread_struct *t = ¤t->thread;
610 u64 *vcpu_fpr = vcpu->arch.fpr;
612 u64 *vcpu_vsx = vcpu->arch.vsr;
614 u64 *thread_fpr = (u64*)t->fpr;
617 if (!(vcpu->arch.guest_owned_ext & msr))
621 printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
627 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
628 vcpu_fpr[i] = thread_fpr[get_fpr_index(i)];
630 vcpu->arch.fpscr = t->fpscr.val;
633 #ifdef CONFIG_ALTIVEC
634 giveup_altivec(current);
635 memcpy(vcpu->arch.vr, t->vr, sizeof(vcpu->arch.vr));
636 vcpu->arch.vscr = t->vscr;
641 __giveup_vsx(current);
642 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++)
643 vcpu_vsx[i] = thread_fpr[get_fpr_index(i) + 1];
650 vcpu->arch.guest_owned_ext &= ~msr;
651 current->thread.regs->msr &= ~msr;
652 kvmppc_recalc_shadow_msr(vcpu);
655 static int kvmppc_read_inst(struct kvm_vcpu *vcpu)
657 ulong srr0 = kvmppc_get_pc(vcpu);
658 u32 last_inst = kvmppc_get_last_inst(vcpu);
661 ret = kvmppc_ld(vcpu, &srr0, sizeof(u32), &last_inst, false);
662 if (ret == -ENOENT) {
663 vcpu->arch.msr = kvmppc_set_field(vcpu->arch.msr, 33, 33, 1);
664 vcpu->arch.msr = kvmppc_set_field(vcpu->arch.msr, 34, 36, 0);
665 vcpu->arch.msr = kvmppc_set_field(vcpu->arch.msr, 42, 47, 0);
666 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_INST_STORAGE);
667 return EMULATE_AGAIN;
673 static int kvmppc_check_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr)
676 /* Need to do paired single emulation? */
677 if (!(vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE))
680 /* Read out the instruction */
681 if (kvmppc_read_inst(vcpu) == EMULATE_DONE)
682 /* Need to emulate */
685 return EMULATE_AGAIN;
688 /* Handle external providers (FPU, Altivec, VSX) */
689 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
692 struct thread_struct *t = ¤t->thread;
693 u64 *vcpu_fpr = vcpu->arch.fpr;
695 u64 *vcpu_vsx = vcpu->arch.vsr;
697 u64 *thread_fpr = (u64*)t->fpr;
700 /* When we have paired singles, we emulate in software */
701 if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
704 if (!(vcpu->arch.msr & msr)) {
705 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
709 /* We already own the ext */
710 if (vcpu->arch.guest_owned_ext & msr) {
715 printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
718 current->thread.regs->msr |= msr;
722 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
723 thread_fpr[get_fpr_index(i)] = vcpu_fpr[i];
725 t->fpscr.val = vcpu->arch.fpscr;
727 kvmppc_load_up_fpu();
730 #ifdef CONFIG_ALTIVEC
731 memcpy(t->vr, vcpu->arch.vr, sizeof(vcpu->arch.vr));
732 t->vscr = vcpu->arch.vscr;
734 kvmppc_load_up_altivec();
739 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++)
740 thread_fpr[get_fpr_index(i) + 1] = vcpu_vsx[i];
741 kvmppc_load_up_vsx();
748 vcpu->arch.guest_owned_ext |= msr;
750 kvmppc_recalc_shadow_msr(vcpu);
755 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
756 unsigned int exit_nr)
760 vcpu->stat.sum_exits++;
762 run->exit_reason = KVM_EXIT_UNKNOWN;
763 run->ready_for_interrupt_injection = 1;
765 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | dar=0x%lx | dec=0x%x | msr=0x%lx\n",
766 exit_nr, kvmppc_get_pc(vcpu), kvmppc_get_fault_dar(vcpu),
767 kvmppc_get_dec(vcpu), to_svcpu(vcpu)->shadow_srr1);
768 #elif defined (EXIT_DEBUG_SIMPLE)
769 if ((exit_nr != 0x900) && (exit_nr != 0x500))
770 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | dar=0x%lx | msr=0x%lx\n",
771 exit_nr, kvmppc_get_pc(vcpu), kvmppc_get_fault_dar(vcpu),
776 case BOOK3S_INTERRUPT_INST_STORAGE:
777 vcpu->stat.pf_instruc++;
778 /* only care about PTEG not found errors, but leave NX alone */
779 if (to_svcpu(vcpu)->shadow_srr1 & 0x40000000) {
780 r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr);
781 vcpu->stat.sp_instruc++;
782 } else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
783 (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
785 * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
786 * so we can't use the NX bit inside the guest. Let's cross our fingers,
787 * that no guest that needs the dcbz hack does NX.
789 kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFULL);
792 vcpu->arch.msr |= to_svcpu(vcpu)->shadow_srr1 & 0x58000000;
793 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
794 kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFULL);
798 case BOOK3S_INTERRUPT_DATA_STORAGE:
800 ulong dar = kvmppc_get_fault_dar(vcpu);
801 vcpu->stat.pf_storage++;
802 /* The only case we need to handle is missing shadow PTEs */
803 if (to_svcpu(vcpu)->fault_dsisr & DSISR_NOHPTE) {
804 r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr);
806 vcpu->arch.dear = dar;
807 to_book3s(vcpu)->dsisr = to_svcpu(vcpu)->fault_dsisr;
808 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
809 kvmppc_mmu_pte_flush(vcpu, vcpu->arch.dear, ~0xFFFULL);
814 case BOOK3S_INTERRUPT_DATA_SEGMENT:
815 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
816 vcpu->arch.dear = kvmppc_get_fault_dar(vcpu);
817 kvmppc_book3s_queue_irqprio(vcpu,
818 BOOK3S_INTERRUPT_DATA_SEGMENT);
822 case BOOK3S_INTERRUPT_INST_SEGMENT:
823 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
824 kvmppc_book3s_queue_irqprio(vcpu,
825 BOOK3S_INTERRUPT_INST_SEGMENT);
829 /* We're good on these - the host merely wanted to get our attention */
830 case BOOK3S_INTERRUPT_DECREMENTER:
831 vcpu->stat.dec_exits++;
834 case BOOK3S_INTERRUPT_EXTERNAL:
835 vcpu->stat.ext_intr_exits++;
838 case BOOK3S_INTERRUPT_PROGRAM:
840 enum emulation_result er;
844 flags = to_svcpu(vcpu)->shadow_srr1 & 0x1f0000ull;
846 if (vcpu->arch.msr & MSR_PR) {
848 printk(KERN_INFO "Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
850 if ((kvmppc_get_last_inst(vcpu) & 0xff0007ff) !=
851 (INS_DCBZ & 0xfffffff7)) {
852 kvmppc_core_queue_program(vcpu, flags);
858 vcpu->stat.emulated_inst_exits++;
859 er = kvmppc_emulate_instruction(run, vcpu);
868 printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
869 __func__, kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
870 kvmppc_core_queue_program(vcpu, flags);
873 case EMULATE_DO_MMIO:
874 run->exit_reason = KVM_EXIT_MMIO;
882 case BOOK3S_INTERRUPT_SYSCALL:
883 // XXX make user settable
884 if (vcpu->arch.osi_enabled &&
885 (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
886 (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
887 u64 *gprs = run->osi.gprs;
890 run->exit_reason = KVM_EXIT_OSI;
891 for (i = 0; i < 32; i++)
892 gprs[i] = kvmppc_get_gpr(vcpu, i);
893 vcpu->arch.osi_needed = 1;
897 vcpu->stat.syscall_exits++;
898 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
902 case BOOK3S_INTERRUPT_FP_UNAVAIL:
903 case BOOK3S_INTERRUPT_ALTIVEC:
904 case BOOK3S_INTERRUPT_VSX:
909 case BOOK3S_INTERRUPT_FP_UNAVAIL: ext_msr = MSR_FP; break;
910 case BOOK3S_INTERRUPT_ALTIVEC: ext_msr = MSR_VEC; break;
911 case BOOK3S_INTERRUPT_VSX: ext_msr = MSR_VSX; break;
914 switch (kvmppc_check_ext(vcpu, exit_nr)) {
916 /* everything ok - let's enable the ext */
917 r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
920 /* we need to emulate this instruction */
921 goto program_interrupt;
924 /* nothing to worry about - go again */
929 case BOOK3S_INTERRUPT_ALIGNMENT:
930 if (kvmppc_read_inst(vcpu) == EMULATE_DONE) {
931 to_book3s(vcpu)->dsisr = kvmppc_alignment_dsisr(vcpu,
932 kvmppc_get_last_inst(vcpu));
933 vcpu->arch.dear = kvmppc_alignment_dar(vcpu,
934 kvmppc_get_last_inst(vcpu));
935 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
939 case BOOK3S_INTERRUPT_MACHINE_CHECK:
940 case BOOK3S_INTERRUPT_TRACE:
941 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
945 /* Ugh - bork here! What did we get? */
946 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
947 exit_nr, kvmppc_get_pc(vcpu), to_svcpu(vcpu)->shadow_srr1);
954 if (!(r & RESUME_HOST)) {
955 /* To avoid clobbering exit_reason, only check for signals if
956 * we aren't already exiting to userspace for some other
958 if (signal_pending(current)) {
960 printk(KERN_EMERG "KVM: Going back to host\n");
962 vcpu->stat.signal_exits++;
963 run->exit_reason = KVM_EXIT_INTR;
966 /* In case an interrupt came in that was triggered
967 * from userspace (like DEC), we need to check what
969 kvmppc_core_deliver_interrupts(vcpu);
974 printk(KERN_EMERG "KVM exit: vcpu=0x%p pc=0x%lx r=0x%x\n", vcpu, kvmppc_get_pc(vcpu), r);
980 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
985 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
991 regs->pc = kvmppc_get_pc(vcpu);
992 regs->cr = kvmppc_get_cr(vcpu);
993 regs->ctr = kvmppc_get_ctr(vcpu);
994 regs->lr = kvmppc_get_lr(vcpu);
995 regs->xer = kvmppc_get_xer(vcpu);
996 regs->msr = vcpu->arch.msr;
997 regs->srr0 = vcpu->arch.srr0;
998 regs->srr1 = vcpu->arch.srr1;
999 regs->pid = vcpu->arch.pid;
1000 regs->sprg0 = vcpu->arch.sprg0;
1001 regs->sprg1 = vcpu->arch.sprg1;
1002 regs->sprg2 = vcpu->arch.sprg2;
1003 regs->sprg3 = vcpu->arch.sprg3;
1004 regs->sprg5 = vcpu->arch.sprg4;
1005 regs->sprg6 = vcpu->arch.sprg5;
1006 regs->sprg7 = vcpu->arch.sprg6;
1008 for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
1009 regs->gpr[i] = kvmppc_get_gpr(vcpu, i);
1016 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1022 kvmppc_set_pc(vcpu, regs->pc);
1023 kvmppc_set_cr(vcpu, regs->cr);
1024 kvmppc_set_ctr(vcpu, regs->ctr);
1025 kvmppc_set_lr(vcpu, regs->lr);
1026 kvmppc_set_xer(vcpu, regs->xer);
1027 kvmppc_set_msr(vcpu, regs->msr);
1028 vcpu->arch.srr0 = regs->srr0;
1029 vcpu->arch.srr1 = regs->srr1;
1030 vcpu->arch.sprg0 = regs->sprg0;
1031 vcpu->arch.sprg1 = regs->sprg1;
1032 vcpu->arch.sprg2 = regs->sprg2;
1033 vcpu->arch.sprg3 = regs->sprg3;
1034 vcpu->arch.sprg5 = regs->sprg4;
1035 vcpu->arch.sprg6 = regs->sprg5;
1036 vcpu->arch.sprg7 = regs->sprg6;
1038 for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
1039 kvmppc_set_gpr(vcpu, i, regs->gpr[i]);
1046 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1047 struct kvm_sregs *sregs)
1049 struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1052 sregs->pvr = vcpu->arch.pvr;
1054 sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
1055 if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1056 for (i = 0; i < 64; i++) {
1057 sregs->u.s.ppc64.slb[i].slbe = vcpu3s->slb[i].orige | i;
1058 sregs->u.s.ppc64.slb[i].slbv = vcpu3s->slb[i].origv;
1061 for (i = 0; i < 16; i++) {
1062 sregs->u.s.ppc32.sr[i] = vcpu3s->sr[i].raw;
1063 sregs->u.s.ppc32.sr[i] = vcpu3s->sr[i].raw;
1065 for (i = 0; i < 8; i++) {
1066 sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
1067 sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
1073 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1074 struct kvm_sregs *sregs)
1076 struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1079 kvmppc_set_pvr(vcpu, sregs->pvr);
1081 vcpu3s->sdr1 = sregs->u.s.sdr1;
1082 if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1083 for (i = 0; i < 64; i++) {
1084 vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
1085 sregs->u.s.ppc64.slb[i].slbe);
1088 for (i = 0; i < 16; i++) {
1089 vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
1091 for (i = 0; i < 8; i++) {
1092 kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
1093 (u32)sregs->u.s.ppc32.ibat[i]);
1094 kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
1095 (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
1096 kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
1097 (u32)sregs->u.s.ppc32.dbat[i]);
1098 kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
1099 (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
1103 /* Flush the MMU after messing with the segments */
1104 kvmppc_mmu_pte_flush(vcpu, 0, 0);
1108 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1113 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1118 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1119 struct kvm_translation *tr)
1125 * Get (and clear) the dirty memory log for a memory slot.
1127 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1128 struct kvm_dirty_log *log)
1130 struct kvm_memory_slot *memslot;
1131 struct kvm_vcpu *vcpu;
1137 mutex_lock(&kvm->slots_lock);
1139 r = kvm_get_dirty_log(kvm, log, &is_dirty);
1143 /* If nothing is dirty, don't bother messing with page tables. */
1145 memslot = &kvm->memslots->memslots[log->slot];
1147 ga = memslot->base_gfn << PAGE_SHIFT;
1148 ga_end = ga + (memslot->npages << PAGE_SHIFT);
1150 kvm_for_each_vcpu(n, vcpu, kvm)
1151 kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
1153 n = kvm_dirty_bitmap_bytes(memslot);
1154 memset(memslot->dirty_bitmap, 0, n);
1159 mutex_unlock(&kvm->slots_lock);
1163 int kvmppc_core_check_processor_compat(void)
1168 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
1170 struct kvmppc_vcpu_book3s *vcpu_book3s;
1171 struct kvm_vcpu *vcpu;
1174 vcpu_book3s = vmalloc(sizeof(struct kvmppc_vcpu_book3s));
1178 memset(vcpu_book3s, 0, sizeof(struct kvmppc_vcpu_book3s));
1180 vcpu_book3s->shadow_vcpu = (struct kvmppc_book3s_shadow_vcpu *)
1181 kzalloc(sizeof(*vcpu_book3s->shadow_vcpu), GFP_KERNEL);
1182 if (!vcpu_book3s->shadow_vcpu)
1185 vcpu = &vcpu_book3s->vcpu;
1186 err = kvm_vcpu_init(vcpu, kvm, id);
1188 goto free_shadow_vcpu;
1190 vcpu->arch.host_retip = kvm_return_point;
1191 vcpu->arch.host_msr = mfmsr();
1192 /* default to book3s_64 (970fx) */
1193 vcpu->arch.pvr = 0x3C0301;
1194 kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
1195 vcpu_book3s->slb_nr = 64;
1197 /* remember where some real-mode handlers are */
1198 vcpu->arch.trampoline_lowmem = kvmppc_trampoline_lowmem;
1199 vcpu->arch.trampoline_enter = kvmppc_trampoline_enter;
1200 vcpu->arch.highmem_handler = (ulong)kvmppc_handler_highmem;
1201 vcpu->arch.rmcall = *(ulong*)kvmppc_rmcall;
1203 vcpu->arch.shadow_msr = MSR_USER64;
1205 err = kvmppc_mmu_init(vcpu);
1207 goto free_shadow_vcpu;
1212 kfree(vcpu_book3s->shadow_vcpu);
1216 return ERR_PTR(err);
1219 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
1221 struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
1223 kvm_vcpu_uninit(vcpu);
1224 kfree(vcpu_book3s->shadow_vcpu);
1228 extern int __kvmppc_vcpu_entry(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu);
1229 int __kvmppc_vcpu_run(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1232 struct thread_struct ext_bkp;
1233 #ifdef CONFIG_ALTIVEC
1234 bool save_vec = current->thread.used_vr;
1237 bool save_vsx = current->thread.used_vsr;
1241 /* No need to go into the guest when all we do is going out */
1242 if (signal_pending(current)) {
1243 kvm_run->exit_reason = KVM_EXIT_INTR;
1247 /* Save FPU state in stack */
1248 if (current->thread.regs->msr & MSR_FP)
1249 giveup_fpu(current);
1250 memcpy(ext_bkp.fpr, current->thread.fpr, sizeof(current->thread.fpr));
1251 ext_bkp.fpscr = current->thread.fpscr;
1252 ext_bkp.fpexc_mode = current->thread.fpexc_mode;
1254 #ifdef CONFIG_ALTIVEC
1255 /* Save Altivec state in stack */
1257 if (current->thread.regs->msr & MSR_VEC)
1258 giveup_altivec(current);
1259 memcpy(ext_bkp.vr, current->thread.vr, sizeof(ext_bkp.vr));
1260 ext_bkp.vscr = current->thread.vscr;
1261 ext_bkp.vrsave = current->thread.vrsave;
1263 ext_bkp.used_vr = current->thread.used_vr;
1267 /* Save VSX state in stack */
1268 if (save_vsx && (current->thread.regs->msr & MSR_VSX))
1269 __giveup_vsx(current);
1270 ext_bkp.used_vsr = current->thread.used_vsr;
1273 /* Remember the MSR with disabled extensions */
1274 ext_msr = current->thread.regs->msr;
1276 /* XXX we get called with irq disabled - change that! */
1279 /* Preload FPU if it's enabled */
1280 if (vcpu->arch.msr & MSR_FP)
1281 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
1283 ret = __kvmppc_vcpu_entry(kvm_run, vcpu);
1285 local_irq_disable();
1287 current->thread.regs->msr = ext_msr;
1289 /* Make sure we save the guest FPU/Altivec/VSX state */
1290 kvmppc_giveup_ext(vcpu, MSR_FP);
1291 kvmppc_giveup_ext(vcpu, MSR_VEC);
1292 kvmppc_giveup_ext(vcpu, MSR_VSX);
1294 /* Restore FPU state from stack */
1295 memcpy(current->thread.fpr, ext_bkp.fpr, sizeof(ext_bkp.fpr));
1296 current->thread.fpscr = ext_bkp.fpscr;
1297 current->thread.fpexc_mode = ext_bkp.fpexc_mode;
1299 #ifdef CONFIG_ALTIVEC
1300 /* Restore Altivec state from stack */
1301 if (save_vec && current->thread.used_vr) {
1302 memcpy(current->thread.vr, ext_bkp.vr, sizeof(ext_bkp.vr));
1303 current->thread.vscr = ext_bkp.vscr;
1304 current->thread.vrsave= ext_bkp.vrsave;
1306 current->thread.used_vr = ext_bkp.used_vr;
1310 current->thread.used_vsr = ext_bkp.used_vsr;
1316 static int kvmppc_book3s_init(void)
1318 return kvm_init(NULL, sizeof(struct kvmppc_vcpu_book3s), THIS_MODULE);
1321 static void kvmppc_book3s_exit(void)
1326 module_init(kvmppc_book3s_init);
1327 module_exit(kvmppc_book3s_exit);