]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame_incremental - arch/powerpc/kvm/powerpc.c
KVM: PPC: Book3S HV: Add capability to report possible virtual SMT modes
[mirror_ubuntu-bionic-kernel.git] / arch / powerpc / kvm / powerpc.c
... / ...
CommitLineData
1/*
2 * This program is free software; you can redistribute it and/or modify
3 * it under the terms of the GNU General Public License, version 2, as
4 * published by the Free Software Foundation.
5 *
6 * This program is distributed in the hope that it will be useful,
7 * but WITHOUT ANY WARRANTY; without even the implied warranty of
8 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
9 * GNU General Public License for more details.
10 *
11 * You should have received a copy of the GNU General Public License
12 * along with this program; if not, write to the Free Software
13 * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
14 *
15 * Copyright IBM Corp. 2007
16 *
17 * Authors: Hollis Blanchard <hollisb@us.ibm.com>
18 * Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
19 */
20
21#include <linux/errno.h>
22#include <linux/err.h>
23#include <linux/kvm_host.h>
24#include <linux/vmalloc.h>
25#include <linux/hrtimer.h>
26#include <linux/sched/signal.h>
27#include <linux/fs.h>
28#include <linux/slab.h>
29#include <linux/file.h>
30#include <linux/module.h>
31#include <linux/irqbypass.h>
32#include <linux/kvm_irqfd.h>
33#include <asm/cputable.h>
34#include <linux/uaccess.h>
35#include <asm/kvm_ppc.h>
36#include <asm/tlbflush.h>
37#include <asm/cputhreads.h>
38#include <asm/irqflags.h>
39#include <asm/iommu.h>
40#include <asm/switch_to.h>
41#include <asm/xive.h>
42
43#include "timing.h"
44#include "irq.h"
45#include "../mm/mmu_decl.h"
46
47#define CREATE_TRACE_POINTS
48#include "trace.h"
49
50struct kvmppc_ops *kvmppc_hv_ops;
51EXPORT_SYMBOL_GPL(kvmppc_hv_ops);
52struct kvmppc_ops *kvmppc_pr_ops;
53EXPORT_SYMBOL_GPL(kvmppc_pr_ops);
54
55
56int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
57{
58 return !!(v->arch.pending_exceptions) ||
59 v->requests;
60}
61
62int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
63{
64 return 1;
65}
66
67/*
68 * Common checks before entering the guest world. Call with interrupts
69 * disabled.
70 *
71 * returns:
72 *
73 * == 1 if we're ready to go into guest state
74 * <= 0 if we need to go back to the host with return value
75 */
76int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu)
77{
78 int r;
79
80 WARN_ON(irqs_disabled());
81 hard_irq_disable();
82
83 while (true) {
84 if (need_resched()) {
85 local_irq_enable();
86 cond_resched();
87 hard_irq_disable();
88 continue;
89 }
90
91 if (signal_pending(current)) {
92 kvmppc_account_exit(vcpu, SIGNAL_EXITS);
93 vcpu->run->exit_reason = KVM_EXIT_INTR;
94 r = -EINTR;
95 break;
96 }
97
98 vcpu->mode = IN_GUEST_MODE;
99
100 /*
101 * Reading vcpu->requests must happen after setting vcpu->mode,
102 * so we don't miss a request because the requester sees
103 * OUTSIDE_GUEST_MODE and assumes we'll be checking requests
104 * before next entering the guest (and thus doesn't IPI).
105 * This also orders the write to mode from any reads
106 * to the page tables done while the VCPU is running.
107 * Please see the comment in kvm_flush_remote_tlbs.
108 */
109 smp_mb();
110
111 if (vcpu->requests) {
112 /* Make sure we process requests preemptable */
113 local_irq_enable();
114 trace_kvm_check_requests(vcpu);
115 r = kvmppc_core_check_requests(vcpu);
116 hard_irq_disable();
117 if (r > 0)
118 continue;
119 break;
120 }
121
122 if (kvmppc_core_prepare_to_enter(vcpu)) {
123 /* interrupts got enabled in between, so we
124 are back at square 1 */
125 continue;
126 }
127
128 guest_enter_irqoff();
129 return 1;
130 }
131
132 /* return to host */
133 local_irq_enable();
134 return r;
135}
136EXPORT_SYMBOL_GPL(kvmppc_prepare_to_enter);
137
138#if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
139static void kvmppc_swab_shared(struct kvm_vcpu *vcpu)
140{
141 struct kvm_vcpu_arch_shared *shared = vcpu->arch.shared;
142 int i;
143
144 shared->sprg0 = swab64(shared->sprg0);
145 shared->sprg1 = swab64(shared->sprg1);
146 shared->sprg2 = swab64(shared->sprg2);
147 shared->sprg3 = swab64(shared->sprg3);
148 shared->srr0 = swab64(shared->srr0);
149 shared->srr1 = swab64(shared->srr1);
150 shared->dar = swab64(shared->dar);
151 shared->msr = swab64(shared->msr);
152 shared->dsisr = swab32(shared->dsisr);
153 shared->int_pending = swab32(shared->int_pending);
154 for (i = 0; i < ARRAY_SIZE(shared->sr); i++)
155 shared->sr[i] = swab32(shared->sr[i]);
156}
157#endif
158
159int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
160{
161 int nr = kvmppc_get_gpr(vcpu, 11);
162 int r;
163 unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
164 unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
165 unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
166 unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
167 unsigned long r2 = 0;
168
169 if (!(kvmppc_get_msr(vcpu) & MSR_SF)) {
170 /* 32 bit mode */
171 param1 &= 0xffffffff;
172 param2 &= 0xffffffff;
173 param3 &= 0xffffffff;
174 param4 &= 0xffffffff;
175 }
176
177 switch (nr) {
178 case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE):
179 {
180#if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
181 /* Book3S can be little endian, find it out here */
182 int shared_big_endian = true;
183 if (vcpu->arch.intr_msr & MSR_LE)
184 shared_big_endian = false;
185 if (shared_big_endian != vcpu->arch.shared_big_endian)
186 kvmppc_swab_shared(vcpu);
187 vcpu->arch.shared_big_endian = shared_big_endian;
188#endif
189
190 if (!(param2 & MAGIC_PAGE_FLAG_NOT_MAPPED_NX)) {
191 /*
192 * Older versions of the Linux magic page code had
193 * a bug where they would map their trampoline code
194 * NX. If that's the case, remove !PR NX capability.
195 */
196 vcpu->arch.disable_kernel_nx = true;
197 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
198 }
199
200 vcpu->arch.magic_page_pa = param1 & ~0xfffULL;
201 vcpu->arch.magic_page_ea = param2 & ~0xfffULL;
202
203#ifdef CONFIG_PPC_64K_PAGES
204 /*
205 * Make sure our 4k magic page is in the same window of a 64k
206 * page within the guest and within the host's page.
207 */
208 if ((vcpu->arch.magic_page_pa & 0xf000) !=
209 ((ulong)vcpu->arch.shared & 0xf000)) {
210 void *old_shared = vcpu->arch.shared;
211 ulong shared = (ulong)vcpu->arch.shared;
212 void *new_shared;
213
214 shared &= PAGE_MASK;
215 shared |= vcpu->arch.magic_page_pa & 0xf000;
216 new_shared = (void*)shared;
217 memcpy(new_shared, old_shared, 0x1000);
218 vcpu->arch.shared = new_shared;
219 }
220#endif
221
222 r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
223
224 r = EV_SUCCESS;
225 break;
226 }
227 case KVM_HCALL_TOKEN(KVM_HC_FEATURES):
228 r = EV_SUCCESS;
229#if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
230 r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
231#endif
232
233 /* Second return value is in r4 */
234 break;
235 case EV_HCALL_TOKEN(EV_IDLE):
236 r = EV_SUCCESS;
237 kvm_vcpu_block(vcpu);
238 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
239 break;
240 default:
241 r = EV_UNIMPLEMENTED;
242 break;
243 }
244
245 kvmppc_set_gpr(vcpu, 4, r2);
246
247 return r;
248}
249EXPORT_SYMBOL_GPL(kvmppc_kvm_pv);
250
251int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
252{
253 int r = false;
254
255 /* We have to know what CPU to virtualize */
256 if (!vcpu->arch.pvr)
257 goto out;
258
259 /* PAPR only works with book3s_64 */
260 if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
261 goto out;
262
263 /* HV KVM can only do PAPR mode for now */
264 if (!vcpu->arch.papr_enabled && is_kvmppc_hv_enabled(vcpu->kvm))
265 goto out;
266
267#ifdef CONFIG_KVM_BOOKE_HV
268 if (!cpu_has_feature(CPU_FTR_EMB_HV))
269 goto out;
270#endif
271
272 r = true;
273
274out:
275 vcpu->arch.sane = r;
276 return r ? 0 : -EINVAL;
277}
278EXPORT_SYMBOL_GPL(kvmppc_sanity_check);
279
280int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
281{
282 enum emulation_result er;
283 int r;
284
285 er = kvmppc_emulate_loadstore(vcpu);
286 switch (er) {
287 case EMULATE_DONE:
288 /* Future optimization: only reload non-volatiles if they were
289 * actually modified. */
290 r = RESUME_GUEST_NV;
291 break;
292 case EMULATE_AGAIN:
293 r = RESUME_GUEST;
294 break;
295 case EMULATE_DO_MMIO:
296 run->exit_reason = KVM_EXIT_MMIO;
297 /* We must reload nonvolatiles because "update" load/store
298 * instructions modify register state. */
299 /* Future optimization: only reload non-volatiles if they were
300 * actually modified. */
301 r = RESUME_HOST_NV;
302 break;
303 case EMULATE_FAIL:
304 {
305 u32 last_inst;
306
307 kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
308 /* XXX Deliver Program interrupt to guest. */
309 pr_emerg("%s: emulation failed (%08x)\n", __func__, last_inst);
310 r = RESUME_HOST;
311 break;
312 }
313 default:
314 WARN_ON(1);
315 r = RESUME_GUEST;
316 }
317
318 return r;
319}
320EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio);
321
322int kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
323 bool data)
324{
325 ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
326 struct kvmppc_pte pte;
327 int r;
328
329 vcpu->stat.st++;
330
331 r = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
332 XLATE_WRITE, &pte);
333 if (r < 0)
334 return r;
335
336 *eaddr = pte.raddr;
337
338 if (!pte.may_write)
339 return -EPERM;
340
341 /* Magic page override */
342 if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
343 ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
344 !(kvmppc_get_msr(vcpu) & MSR_PR)) {
345 void *magic = vcpu->arch.shared;
346 magic += pte.eaddr & 0xfff;
347 memcpy(magic, ptr, size);
348 return EMULATE_DONE;
349 }
350
351 if (kvm_write_guest(vcpu->kvm, pte.raddr, ptr, size))
352 return EMULATE_DO_MMIO;
353
354 return EMULATE_DONE;
355}
356EXPORT_SYMBOL_GPL(kvmppc_st);
357
358int kvmppc_ld(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
359 bool data)
360{
361 ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
362 struct kvmppc_pte pte;
363 int rc;
364
365 vcpu->stat.ld++;
366
367 rc = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
368 XLATE_READ, &pte);
369 if (rc)
370 return rc;
371
372 *eaddr = pte.raddr;
373
374 if (!pte.may_read)
375 return -EPERM;
376
377 if (!data && !pte.may_execute)
378 return -ENOEXEC;
379
380 /* Magic page override */
381 if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
382 ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
383 !(kvmppc_get_msr(vcpu) & MSR_PR)) {
384 void *magic = vcpu->arch.shared;
385 magic += pte.eaddr & 0xfff;
386 memcpy(ptr, magic, size);
387 return EMULATE_DONE;
388 }
389
390 if (kvm_read_guest(vcpu->kvm, pte.raddr, ptr, size))
391 return EMULATE_DO_MMIO;
392
393 return EMULATE_DONE;
394}
395EXPORT_SYMBOL_GPL(kvmppc_ld);
396
397int kvm_arch_hardware_enable(void)
398{
399 return 0;
400}
401
402int kvm_arch_hardware_setup(void)
403{
404 return 0;
405}
406
407void kvm_arch_check_processor_compat(void *rtn)
408{
409 *(int *)rtn = kvmppc_core_check_processor_compat();
410}
411
412int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
413{
414 struct kvmppc_ops *kvm_ops = NULL;
415 /*
416 * if we have both HV and PR enabled, default is HV
417 */
418 if (type == 0) {
419 if (kvmppc_hv_ops)
420 kvm_ops = kvmppc_hv_ops;
421 else
422 kvm_ops = kvmppc_pr_ops;
423 if (!kvm_ops)
424 goto err_out;
425 } else if (type == KVM_VM_PPC_HV) {
426 if (!kvmppc_hv_ops)
427 goto err_out;
428 kvm_ops = kvmppc_hv_ops;
429 } else if (type == KVM_VM_PPC_PR) {
430 if (!kvmppc_pr_ops)
431 goto err_out;
432 kvm_ops = kvmppc_pr_ops;
433 } else
434 goto err_out;
435
436 if (kvm_ops->owner && !try_module_get(kvm_ops->owner))
437 return -ENOENT;
438
439 kvm->arch.kvm_ops = kvm_ops;
440 return kvmppc_core_init_vm(kvm);
441err_out:
442 return -EINVAL;
443}
444
445bool kvm_arch_has_vcpu_debugfs(void)
446{
447 return false;
448}
449
450int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
451{
452 return 0;
453}
454
455void kvm_arch_destroy_vm(struct kvm *kvm)
456{
457 unsigned int i;
458 struct kvm_vcpu *vcpu;
459
460#ifdef CONFIG_KVM_XICS
461 /*
462 * We call kick_all_cpus_sync() to ensure that all
463 * CPUs have executed any pending IPIs before we
464 * continue and free VCPUs structures below.
465 */
466 if (is_kvmppc_hv_enabled(kvm))
467 kick_all_cpus_sync();
468#endif
469
470 kvm_for_each_vcpu(i, vcpu, kvm)
471 kvm_arch_vcpu_free(vcpu);
472
473 mutex_lock(&kvm->lock);
474 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
475 kvm->vcpus[i] = NULL;
476
477 atomic_set(&kvm->online_vcpus, 0);
478
479 kvmppc_core_destroy_vm(kvm);
480
481 mutex_unlock(&kvm->lock);
482
483 /* drop the module reference */
484 module_put(kvm->arch.kvm_ops->owner);
485}
486
487int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
488{
489 int r;
490 /* Assume we're using HV mode when the HV module is loaded */
491 int hv_enabled = kvmppc_hv_ops ? 1 : 0;
492
493 if (kvm) {
494 /*
495 * Hooray - we know which VM type we're running on. Depend on
496 * that rather than the guess above.
497 */
498 hv_enabled = is_kvmppc_hv_enabled(kvm);
499 }
500
501 switch (ext) {
502#ifdef CONFIG_BOOKE
503 case KVM_CAP_PPC_BOOKE_SREGS:
504 case KVM_CAP_PPC_BOOKE_WATCHDOG:
505 case KVM_CAP_PPC_EPR:
506#else
507 case KVM_CAP_PPC_SEGSTATE:
508 case KVM_CAP_PPC_HIOR:
509 case KVM_CAP_PPC_PAPR:
510#endif
511 case KVM_CAP_PPC_UNSET_IRQ:
512 case KVM_CAP_PPC_IRQ_LEVEL:
513 case KVM_CAP_ENABLE_CAP:
514 case KVM_CAP_ENABLE_CAP_VM:
515 case KVM_CAP_ONE_REG:
516 case KVM_CAP_IOEVENTFD:
517 case KVM_CAP_DEVICE_CTRL:
518 case KVM_CAP_IMMEDIATE_EXIT:
519 r = 1;
520 break;
521 case KVM_CAP_PPC_PAIRED_SINGLES:
522 case KVM_CAP_PPC_OSI:
523 case KVM_CAP_PPC_GET_PVINFO:
524#if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
525 case KVM_CAP_SW_TLB:
526#endif
527 /* We support this only for PR */
528 r = !hv_enabled;
529 break;
530#ifdef CONFIG_KVM_MPIC
531 case KVM_CAP_IRQ_MPIC:
532 r = 1;
533 break;
534#endif
535
536#ifdef CONFIG_PPC_BOOK3S_64
537 case KVM_CAP_SPAPR_TCE:
538 case KVM_CAP_SPAPR_TCE_64:
539 /* fallthrough */
540 case KVM_CAP_SPAPR_TCE_VFIO:
541 case KVM_CAP_PPC_RTAS:
542 case KVM_CAP_PPC_FIXUP_HCALL:
543 case KVM_CAP_PPC_ENABLE_HCALL:
544#ifdef CONFIG_KVM_XICS
545 case KVM_CAP_IRQ_XICS:
546#endif
547 r = 1;
548 break;
549
550 case KVM_CAP_PPC_ALLOC_HTAB:
551 r = hv_enabled;
552 break;
553#endif /* CONFIG_PPC_BOOK3S_64 */
554#ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
555 case KVM_CAP_PPC_SMT:
556 r = 0;
557 if (kvm) {
558 if (kvm->arch.emul_smt_mode > 1)
559 r = kvm->arch.emul_smt_mode;
560 else
561 r = kvm->arch.smt_mode;
562 } else if (hv_enabled) {
563 if (cpu_has_feature(CPU_FTR_ARCH_300))
564 r = 1;
565 else
566 r = threads_per_subcore;
567 }
568 break;
569 case KVM_CAP_PPC_SMT_POSSIBLE:
570 r = 1;
571 if (hv_enabled) {
572 if (!cpu_has_feature(CPU_FTR_ARCH_300))
573 r = ((threads_per_subcore << 1) - 1);
574 else
575 /* P9 can emulate dbells, so allow any mode */
576 r = 8 | 4 | 2 | 1;
577 }
578 break;
579 case KVM_CAP_PPC_RMA:
580 r = 0;
581 break;
582 case KVM_CAP_PPC_HWRNG:
583 r = kvmppc_hwrng_present();
584 break;
585 case KVM_CAP_PPC_MMU_RADIX:
586 r = !!(hv_enabled && radix_enabled());
587 break;
588 case KVM_CAP_PPC_MMU_HASH_V3:
589 r = !!(hv_enabled && !radix_enabled() &&
590 cpu_has_feature(CPU_FTR_ARCH_300));
591 break;
592#endif
593 case KVM_CAP_SYNC_MMU:
594#ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
595 r = hv_enabled;
596#elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
597 r = 1;
598#else
599 r = 0;
600#endif
601 break;
602#ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
603 case KVM_CAP_PPC_HTAB_FD:
604 r = hv_enabled;
605 break;
606#endif
607 case KVM_CAP_NR_VCPUS:
608 /*
609 * Recommending a number of CPUs is somewhat arbitrary; we
610 * return the number of present CPUs for -HV (since a host
611 * will have secondary threads "offline"), and for other KVM
612 * implementations just count online CPUs.
613 */
614 if (hv_enabled)
615 r = num_present_cpus();
616 else
617 r = num_online_cpus();
618 break;
619 case KVM_CAP_NR_MEMSLOTS:
620 r = KVM_USER_MEM_SLOTS;
621 break;
622 case KVM_CAP_MAX_VCPUS:
623 r = KVM_MAX_VCPUS;
624 break;
625#ifdef CONFIG_PPC_BOOK3S_64
626 case KVM_CAP_PPC_GET_SMMU_INFO:
627 r = 1;
628 break;
629 case KVM_CAP_SPAPR_MULTITCE:
630 r = 1;
631 break;
632 case KVM_CAP_SPAPR_RESIZE_HPT:
633 /* Disable this on POWER9 until code handles new HPTE format */
634 r = !!hv_enabled && !cpu_has_feature(CPU_FTR_ARCH_300);
635 break;
636#endif
637#ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
638 case KVM_CAP_PPC_FWNMI:
639 r = hv_enabled;
640 break;
641#endif
642 case KVM_CAP_PPC_HTM:
643 r = cpu_has_feature(CPU_FTR_TM_COMP) &&
644 is_kvmppc_hv_enabled(kvm);
645 break;
646 default:
647 r = 0;
648 break;
649 }
650 return r;
651
652}
653
654long kvm_arch_dev_ioctl(struct file *filp,
655 unsigned int ioctl, unsigned long arg)
656{
657 return -EINVAL;
658}
659
660void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
661 struct kvm_memory_slot *dont)
662{
663 kvmppc_core_free_memslot(kvm, free, dont);
664}
665
666int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
667 unsigned long npages)
668{
669 return kvmppc_core_create_memslot(kvm, slot, npages);
670}
671
672int kvm_arch_prepare_memory_region(struct kvm *kvm,
673 struct kvm_memory_slot *memslot,
674 const struct kvm_userspace_memory_region *mem,
675 enum kvm_mr_change change)
676{
677 return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
678}
679
680void kvm_arch_commit_memory_region(struct kvm *kvm,
681 const struct kvm_userspace_memory_region *mem,
682 const struct kvm_memory_slot *old,
683 const struct kvm_memory_slot *new,
684 enum kvm_mr_change change)
685{
686 kvmppc_core_commit_memory_region(kvm, mem, old, new);
687}
688
689void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
690 struct kvm_memory_slot *slot)
691{
692 kvmppc_core_flush_memslot(kvm, slot);
693}
694
695struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
696{
697 struct kvm_vcpu *vcpu;
698 vcpu = kvmppc_core_vcpu_create(kvm, id);
699 if (!IS_ERR(vcpu)) {
700 vcpu->arch.wqp = &vcpu->wq;
701 kvmppc_create_vcpu_debugfs(vcpu, id);
702 }
703 return vcpu;
704}
705
706void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
707{
708}
709
710void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
711{
712 /* Make sure we're not using the vcpu anymore */
713 hrtimer_cancel(&vcpu->arch.dec_timer);
714
715 kvmppc_remove_vcpu_debugfs(vcpu);
716
717 switch (vcpu->arch.irq_type) {
718 case KVMPPC_IRQ_MPIC:
719 kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
720 break;
721 case KVMPPC_IRQ_XICS:
722 if (xive_enabled())
723 kvmppc_xive_cleanup_vcpu(vcpu);
724 else
725 kvmppc_xics_free_icp(vcpu);
726 break;
727 }
728
729 kvmppc_core_vcpu_free(vcpu);
730}
731
732void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
733{
734 kvm_arch_vcpu_free(vcpu);
735}
736
737int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
738{
739 return kvmppc_core_pending_dec(vcpu);
740}
741
742static enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
743{
744 struct kvm_vcpu *vcpu;
745
746 vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
747 kvmppc_decrementer_func(vcpu);
748
749 return HRTIMER_NORESTART;
750}
751
752int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
753{
754 int ret;
755
756 hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
757 vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
758 vcpu->arch.dec_expires = ~(u64)0;
759
760#ifdef CONFIG_KVM_EXIT_TIMING
761 mutex_init(&vcpu->arch.exit_timing_lock);
762#endif
763 ret = kvmppc_subarch_vcpu_init(vcpu);
764 return ret;
765}
766
767void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
768{
769 kvmppc_mmu_destroy(vcpu);
770 kvmppc_subarch_vcpu_uninit(vcpu);
771}
772
773void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
774{
775#ifdef CONFIG_BOOKE
776 /*
777 * vrsave (formerly usprg0) isn't used by Linux, but may
778 * be used by the guest.
779 *
780 * On non-booke this is associated with Altivec and
781 * is handled by code in book3s.c.
782 */
783 mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
784#endif
785 kvmppc_core_vcpu_load(vcpu, cpu);
786}
787
788void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
789{
790 kvmppc_core_vcpu_put(vcpu);
791#ifdef CONFIG_BOOKE
792 vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
793#endif
794}
795
796/*
797 * irq_bypass_add_producer and irq_bypass_del_producer are only
798 * useful if the architecture supports PCI passthrough.
799 * irq_bypass_stop and irq_bypass_start are not needed and so
800 * kvm_ops are not defined for them.
801 */
802bool kvm_arch_has_irq_bypass(void)
803{
804 return ((kvmppc_hv_ops && kvmppc_hv_ops->irq_bypass_add_producer) ||
805 (kvmppc_pr_ops && kvmppc_pr_ops->irq_bypass_add_producer));
806}
807
808int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
809 struct irq_bypass_producer *prod)
810{
811 struct kvm_kernel_irqfd *irqfd =
812 container_of(cons, struct kvm_kernel_irqfd, consumer);
813 struct kvm *kvm = irqfd->kvm;
814
815 if (kvm->arch.kvm_ops->irq_bypass_add_producer)
816 return kvm->arch.kvm_ops->irq_bypass_add_producer(cons, prod);
817
818 return 0;
819}
820
821void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
822 struct irq_bypass_producer *prod)
823{
824 struct kvm_kernel_irqfd *irqfd =
825 container_of(cons, struct kvm_kernel_irqfd, consumer);
826 struct kvm *kvm = irqfd->kvm;
827
828 if (kvm->arch.kvm_ops->irq_bypass_del_producer)
829 kvm->arch.kvm_ops->irq_bypass_del_producer(cons, prod);
830}
831
832#ifdef CONFIG_VSX
833static inline int kvmppc_get_vsr_dword_offset(int index)
834{
835 int offset;
836
837 if ((index != 0) && (index != 1))
838 return -1;
839
840#ifdef __BIG_ENDIAN
841 offset = index;
842#else
843 offset = 1 - index;
844#endif
845
846 return offset;
847}
848
849static inline int kvmppc_get_vsr_word_offset(int index)
850{
851 int offset;
852
853 if ((index > 3) || (index < 0))
854 return -1;
855
856#ifdef __BIG_ENDIAN
857 offset = index;
858#else
859 offset = 3 - index;
860#endif
861 return offset;
862}
863
864static inline void kvmppc_set_vsr_dword(struct kvm_vcpu *vcpu,
865 u64 gpr)
866{
867 union kvmppc_one_reg val;
868 int offset = kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
869 int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
870
871 if (offset == -1)
872 return;
873
874 if (vcpu->arch.mmio_vsx_tx_sx_enabled) {
875 val.vval = VCPU_VSX_VR(vcpu, index);
876 val.vsxval[offset] = gpr;
877 VCPU_VSX_VR(vcpu, index) = val.vval;
878 } else {
879 VCPU_VSX_FPR(vcpu, index, offset) = gpr;
880 }
881}
882
883static inline void kvmppc_set_vsr_dword_dump(struct kvm_vcpu *vcpu,
884 u64 gpr)
885{
886 union kvmppc_one_reg val;
887 int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
888
889 if (vcpu->arch.mmio_vsx_tx_sx_enabled) {
890 val.vval = VCPU_VSX_VR(vcpu, index);
891 val.vsxval[0] = gpr;
892 val.vsxval[1] = gpr;
893 VCPU_VSX_VR(vcpu, index) = val.vval;
894 } else {
895 VCPU_VSX_FPR(vcpu, index, 0) = gpr;
896 VCPU_VSX_FPR(vcpu, index, 1) = gpr;
897 }
898}
899
900static inline void kvmppc_set_vsr_word(struct kvm_vcpu *vcpu,
901 u32 gpr32)
902{
903 union kvmppc_one_reg val;
904 int offset = kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
905 int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
906 int dword_offset, word_offset;
907
908 if (offset == -1)
909 return;
910
911 if (vcpu->arch.mmio_vsx_tx_sx_enabled) {
912 val.vval = VCPU_VSX_VR(vcpu, index);
913 val.vsx32val[offset] = gpr32;
914 VCPU_VSX_VR(vcpu, index) = val.vval;
915 } else {
916 dword_offset = offset / 2;
917 word_offset = offset % 2;
918 val.vsxval[0] = VCPU_VSX_FPR(vcpu, index, dword_offset);
919 val.vsx32val[word_offset] = gpr32;
920 VCPU_VSX_FPR(vcpu, index, dword_offset) = val.vsxval[0];
921 }
922}
923#endif /* CONFIG_VSX */
924
925#ifdef CONFIG_PPC_FPU
926static inline u64 sp_to_dp(u32 fprs)
927{
928 u64 fprd;
929
930 preempt_disable();
931 enable_kernel_fp();
932 asm ("lfs%U1%X1 0,%1; stfd%U0%X0 0,%0" : "=m" (fprd) : "m" (fprs)
933 : "fr0");
934 preempt_enable();
935 return fprd;
936}
937
938static inline u32 dp_to_sp(u64 fprd)
939{
940 u32 fprs;
941
942 preempt_disable();
943 enable_kernel_fp();
944 asm ("lfd%U1%X1 0,%1; stfs%U0%X0 0,%0" : "=m" (fprs) : "m" (fprd)
945 : "fr0");
946 preempt_enable();
947 return fprs;
948}
949
950#else
951#define sp_to_dp(x) (x)
952#define dp_to_sp(x) (x)
953#endif /* CONFIG_PPC_FPU */
954
955static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
956 struct kvm_run *run)
957{
958 u64 uninitialized_var(gpr);
959
960 if (run->mmio.len > sizeof(gpr)) {
961 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
962 return;
963 }
964
965 if (!vcpu->arch.mmio_host_swabbed) {
966 switch (run->mmio.len) {
967 case 8: gpr = *(u64 *)run->mmio.data; break;
968 case 4: gpr = *(u32 *)run->mmio.data; break;
969 case 2: gpr = *(u16 *)run->mmio.data; break;
970 case 1: gpr = *(u8 *)run->mmio.data; break;
971 }
972 } else {
973 switch (run->mmio.len) {
974 case 8: gpr = swab64(*(u64 *)run->mmio.data); break;
975 case 4: gpr = swab32(*(u32 *)run->mmio.data); break;
976 case 2: gpr = swab16(*(u16 *)run->mmio.data); break;
977 case 1: gpr = *(u8 *)run->mmio.data; break;
978 }
979 }
980
981 /* conversion between single and double precision */
982 if ((vcpu->arch.mmio_sp64_extend) && (run->mmio.len == 4))
983 gpr = sp_to_dp(gpr);
984
985 if (vcpu->arch.mmio_sign_extend) {
986 switch (run->mmio.len) {
987#ifdef CONFIG_PPC64
988 case 4:
989 gpr = (s64)(s32)gpr;
990 break;
991#endif
992 case 2:
993 gpr = (s64)(s16)gpr;
994 break;
995 case 1:
996 gpr = (s64)(s8)gpr;
997 break;
998 }
999 }
1000
1001 switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
1002 case KVM_MMIO_REG_GPR:
1003 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
1004 break;
1005 case KVM_MMIO_REG_FPR:
1006 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1007 break;
1008#ifdef CONFIG_PPC_BOOK3S
1009 case KVM_MMIO_REG_QPR:
1010 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1011 break;
1012 case KVM_MMIO_REG_FQPR:
1013 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1014 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1015 break;
1016#endif
1017#ifdef CONFIG_VSX
1018 case KVM_MMIO_REG_VSX:
1019 if (vcpu->arch.mmio_vsx_copy_type == KVMPPC_VSX_COPY_DWORD)
1020 kvmppc_set_vsr_dword(vcpu, gpr);
1021 else if (vcpu->arch.mmio_vsx_copy_type == KVMPPC_VSX_COPY_WORD)
1022 kvmppc_set_vsr_word(vcpu, gpr);
1023 else if (vcpu->arch.mmio_vsx_copy_type ==
1024 KVMPPC_VSX_COPY_DWORD_LOAD_DUMP)
1025 kvmppc_set_vsr_dword_dump(vcpu, gpr);
1026 break;
1027#endif
1028 default:
1029 BUG();
1030 }
1031}
1032
1033static int __kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1034 unsigned int rt, unsigned int bytes,
1035 int is_default_endian, int sign_extend)
1036{
1037 int idx, ret;
1038 bool host_swabbed;
1039
1040 /* Pity C doesn't have a logical XOR operator */
1041 if (kvmppc_need_byteswap(vcpu)) {
1042 host_swabbed = is_default_endian;
1043 } else {
1044 host_swabbed = !is_default_endian;
1045 }
1046
1047 if (bytes > sizeof(run->mmio.data)) {
1048 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1049 run->mmio.len);
1050 }
1051
1052 run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1053 run->mmio.len = bytes;
1054 run->mmio.is_write = 0;
1055
1056 vcpu->arch.io_gpr = rt;
1057 vcpu->arch.mmio_host_swabbed = host_swabbed;
1058 vcpu->mmio_needed = 1;
1059 vcpu->mmio_is_write = 0;
1060 vcpu->arch.mmio_sign_extend = sign_extend;
1061
1062 idx = srcu_read_lock(&vcpu->kvm->srcu);
1063
1064 ret = kvm_io_bus_read(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1065 bytes, &run->mmio.data);
1066
1067 srcu_read_unlock(&vcpu->kvm->srcu, idx);
1068
1069 if (!ret) {
1070 kvmppc_complete_mmio_load(vcpu, run);
1071 vcpu->mmio_needed = 0;
1072 return EMULATE_DONE;
1073 }
1074
1075 return EMULATE_DO_MMIO;
1076}
1077
1078int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1079 unsigned int rt, unsigned int bytes,
1080 int is_default_endian)
1081{
1082 return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 0);
1083}
1084EXPORT_SYMBOL_GPL(kvmppc_handle_load);
1085
1086/* Same as above, but sign extends */
1087int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
1088 unsigned int rt, unsigned int bytes,
1089 int is_default_endian)
1090{
1091 return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 1);
1092}
1093
1094#ifdef CONFIG_VSX
1095int kvmppc_handle_vsx_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1096 unsigned int rt, unsigned int bytes,
1097 int is_default_endian, int mmio_sign_extend)
1098{
1099 enum emulation_result emulated = EMULATE_DONE;
1100
1101 /* Currently, mmio_vsx_copy_nums only allowed to be less than 4 */
1102 if ( (vcpu->arch.mmio_vsx_copy_nums > 4) ||
1103 (vcpu->arch.mmio_vsx_copy_nums < 0) ) {
1104 return EMULATE_FAIL;
1105 }
1106
1107 while (vcpu->arch.mmio_vsx_copy_nums) {
1108 emulated = __kvmppc_handle_load(run, vcpu, rt, bytes,
1109 is_default_endian, mmio_sign_extend);
1110
1111 if (emulated != EMULATE_DONE)
1112 break;
1113
1114 vcpu->arch.paddr_accessed += run->mmio.len;
1115
1116 vcpu->arch.mmio_vsx_copy_nums--;
1117 vcpu->arch.mmio_vsx_offset++;
1118 }
1119 return emulated;
1120}
1121#endif /* CONFIG_VSX */
1122
1123int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1124 u64 val, unsigned int bytes, int is_default_endian)
1125{
1126 void *data = run->mmio.data;
1127 int idx, ret;
1128 bool host_swabbed;
1129
1130 /* Pity C doesn't have a logical XOR operator */
1131 if (kvmppc_need_byteswap(vcpu)) {
1132 host_swabbed = is_default_endian;
1133 } else {
1134 host_swabbed = !is_default_endian;
1135 }
1136
1137 if (bytes > sizeof(run->mmio.data)) {
1138 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1139 run->mmio.len);
1140 }
1141
1142 run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1143 run->mmio.len = bytes;
1144 run->mmio.is_write = 1;
1145 vcpu->mmio_needed = 1;
1146 vcpu->mmio_is_write = 1;
1147
1148 if ((vcpu->arch.mmio_sp64_extend) && (bytes == 4))
1149 val = dp_to_sp(val);
1150
1151 /* Store the value at the lowest bytes in 'data'. */
1152 if (!host_swabbed) {
1153 switch (bytes) {
1154 case 8: *(u64 *)data = val; break;
1155 case 4: *(u32 *)data = val; break;
1156 case 2: *(u16 *)data = val; break;
1157 case 1: *(u8 *)data = val; break;
1158 }
1159 } else {
1160 switch (bytes) {
1161 case 8: *(u64 *)data = swab64(val); break;
1162 case 4: *(u32 *)data = swab32(val); break;
1163 case 2: *(u16 *)data = swab16(val); break;
1164 case 1: *(u8 *)data = val; break;
1165 }
1166 }
1167
1168 idx = srcu_read_lock(&vcpu->kvm->srcu);
1169
1170 ret = kvm_io_bus_write(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1171 bytes, &run->mmio.data);
1172
1173 srcu_read_unlock(&vcpu->kvm->srcu, idx);
1174
1175 if (!ret) {
1176 vcpu->mmio_needed = 0;
1177 return EMULATE_DONE;
1178 }
1179
1180 return EMULATE_DO_MMIO;
1181}
1182EXPORT_SYMBOL_GPL(kvmppc_handle_store);
1183
1184#ifdef CONFIG_VSX
1185static inline int kvmppc_get_vsr_data(struct kvm_vcpu *vcpu, int rs, u64 *val)
1186{
1187 u32 dword_offset, word_offset;
1188 union kvmppc_one_reg reg;
1189 int vsx_offset = 0;
1190 int copy_type = vcpu->arch.mmio_vsx_copy_type;
1191 int result = 0;
1192
1193 switch (copy_type) {
1194 case KVMPPC_VSX_COPY_DWORD:
1195 vsx_offset =
1196 kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
1197
1198 if (vsx_offset == -1) {
1199 result = -1;
1200 break;
1201 }
1202
1203 if (!vcpu->arch.mmio_vsx_tx_sx_enabled) {
1204 *val = VCPU_VSX_FPR(vcpu, rs, vsx_offset);
1205 } else {
1206 reg.vval = VCPU_VSX_VR(vcpu, rs);
1207 *val = reg.vsxval[vsx_offset];
1208 }
1209 break;
1210
1211 case KVMPPC_VSX_COPY_WORD:
1212 vsx_offset =
1213 kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
1214
1215 if (vsx_offset == -1) {
1216 result = -1;
1217 break;
1218 }
1219
1220 if (!vcpu->arch.mmio_vsx_tx_sx_enabled) {
1221 dword_offset = vsx_offset / 2;
1222 word_offset = vsx_offset % 2;
1223 reg.vsxval[0] = VCPU_VSX_FPR(vcpu, rs, dword_offset);
1224 *val = reg.vsx32val[word_offset];
1225 } else {
1226 reg.vval = VCPU_VSX_VR(vcpu, rs);
1227 *val = reg.vsx32val[vsx_offset];
1228 }
1229 break;
1230
1231 default:
1232 result = -1;
1233 break;
1234 }
1235
1236 return result;
1237}
1238
1239int kvmppc_handle_vsx_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1240 int rs, unsigned int bytes, int is_default_endian)
1241{
1242 u64 val;
1243 enum emulation_result emulated = EMULATE_DONE;
1244
1245 vcpu->arch.io_gpr = rs;
1246
1247 /* Currently, mmio_vsx_copy_nums only allowed to be less than 4 */
1248 if ( (vcpu->arch.mmio_vsx_copy_nums > 4) ||
1249 (vcpu->arch.mmio_vsx_copy_nums < 0) ) {
1250 return EMULATE_FAIL;
1251 }
1252
1253 while (vcpu->arch.mmio_vsx_copy_nums) {
1254 if (kvmppc_get_vsr_data(vcpu, rs, &val) == -1)
1255 return EMULATE_FAIL;
1256
1257 emulated = kvmppc_handle_store(run, vcpu,
1258 val, bytes, is_default_endian);
1259
1260 if (emulated != EMULATE_DONE)
1261 break;
1262
1263 vcpu->arch.paddr_accessed += run->mmio.len;
1264
1265 vcpu->arch.mmio_vsx_copy_nums--;
1266 vcpu->arch.mmio_vsx_offset++;
1267 }
1268
1269 return emulated;
1270}
1271
1272static int kvmppc_emulate_mmio_vsx_loadstore(struct kvm_vcpu *vcpu,
1273 struct kvm_run *run)
1274{
1275 enum emulation_result emulated = EMULATE_FAIL;
1276 int r;
1277
1278 vcpu->arch.paddr_accessed += run->mmio.len;
1279
1280 if (!vcpu->mmio_is_write) {
1281 emulated = kvmppc_handle_vsx_load(run, vcpu, vcpu->arch.io_gpr,
1282 run->mmio.len, 1, vcpu->arch.mmio_sign_extend);
1283 } else {
1284 emulated = kvmppc_handle_vsx_store(run, vcpu,
1285 vcpu->arch.io_gpr, run->mmio.len, 1);
1286 }
1287
1288 switch (emulated) {
1289 case EMULATE_DO_MMIO:
1290 run->exit_reason = KVM_EXIT_MMIO;
1291 r = RESUME_HOST;
1292 break;
1293 case EMULATE_FAIL:
1294 pr_info("KVM: MMIO emulation failed (VSX repeat)\n");
1295 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1296 run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1297 r = RESUME_HOST;
1298 break;
1299 default:
1300 r = RESUME_GUEST;
1301 break;
1302 }
1303 return r;
1304}
1305#endif /* CONFIG_VSX */
1306
1307int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1308{
1309 int r = 0;
1310 union kvmppc_one_reg val;
1311 int size;
1312
1313 size = one_reg_size(reg->id);
1314 if (size > sizeof(val))
1315 return -EINVAL;
1316
1317 r = kvmppc_get_one_reg(vcpu, reg->id, &val);
1318 if (r == -EINVAL) {
1319 r = 0;
1320 switch (reg->id) {
1321#ifdef CONFIG_ALTIVEC
1322 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1323 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1324 r = -ENXIO;
1325 break;
1326 }
1327 val.vval = vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0];
1328 break;
1329 case KVM_REG_PPC_VSCR:
1330 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1331 r = -ENXIO;
1332 break;
1333 }
1334 val = get_reg_val(reg->id, vcpu->arch.vr.vscr.u[3]);
1335 break;
1336 case KVM_REG_PPC_VRSAVE:
1337 val = get_reg_val(reg->id, vcpu->arch.vrsave);
1338 break;
1339#endif /* CONFIG_ALTIVEC */
1340 default:
1341 r = -EINVAL;
1342 break;
1343 }
1344 }
1345
1346 if (r)
1347 return r;
1348
1349 if (copy_to_user((char __user *)(unsigned long)reg->addr, &val, size))
1350 r = -EFAULT;
1351
1352 return r;
1353}
1354
1355int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1356{
1357 int r;
1358 union kvmppc_one_reg val;
1359 int size;
1360
1361 size = one_reg_size(reg->id);
1362 if (size > sizeof(val))
1363 return -EINVAL;
1364
1365 if (copy_from_user(&val, (char __user *)(unsigned long)reg->addr, size))
1366 return -EFAULT;
1367
1368 r = kvmppc_set_one_reg(vcpu, reg->id, &val);
1369 if (r == -EINVAL) {
1370 r = 0;
1371 switch (reg->id) {
1372#ifdef CONFIG_ALTIVEC
1373 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1374 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1375 r = -ENXIO;
1376 break;
1377 }
1378 vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0] = val.vval;
1379 break;
1380 case KVM_REG_PPC_VSCR:
1381 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1382 r = -ENXIO;
1383 break;
1384 }
1385 vcpu->arch.vr.vscr.u[3] = set_reg_val(reg->id, val);
1386 break;
1387 case KVM_REG_PPC_VRSAVE:
1388 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1389 r = -ENXIO;
1390 break;
1391 }
1392 vcpu->arch.vrsave = set_reg_val(reg->id, val);
1393 break;
1394#endif /* CONFIG_ALTIVEC */
1395 default:
1396 r = -EINVAL;
1397 break;
1398 }
1399 }
1400
1401 return r;
1402}
1403
1404int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
1405{
1406 int r;
1407 sigset_t sigsaved;
1408
1409 if (vcpu->mmio_needed) {
1410 vcpu->mmio_needed = 0;
1411 if (!vcpu->mmio_is_write)
1412 kvmppc_complete_mmio_load(vcpu, run);
1413#ifdef CONFIG_VSX
1414 if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1415 vcpu->arch.mmio_vsx_copy_nums--;
1416 vcpu->arch.mmio_vsx_offset++;
1417 }
1418
1419 if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1420 r = kvmppc_emulate_mmio_vsx_loadstore(vcpu, run);
1421 if (r == RESUME_HOST) {
1422 vcpu->mmio_needed = 1;
1423 return r;
1424 }
1425 }
1426#endif
1427 } else if (vcpu->arch.osi_needed) {
1428 u64 *gprs = run->osi.gprs;
1429 int i;
1430
1431 for (i = 0; i < 32; i++)
1432 kvmppc_set_gpr(vcpu, i, gprs[i]);
1433 vcpu->arch.osi_needed = 0;
1434 } else if (vcpu->arch.hcall_needed) {
1435 int i;
1436
1437 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
1438 for (i = 0; i < 9; ++i)
1439 kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
1440 vcpu->arch.hcall_needed = 0;
1441#ifdef CONFIG_BOOKE
1442 } else if (vcpu->arch.epr_needed) {
1443 kvmppc_set_epr(vcpu, run->epr.epr);
1444 vcpu->arch.epr_needed = 0;
1445#endif
1446 }
1447
1448 if (vcpu->sigset_active)
1449 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1450
1451 if (run->immediate_exit)
1452 r = -EINTR;
1453 else
1454 r = kvmppc_vcpu_run(run, vcpu);
1455
1456 if (vcpu->sigset_active)
1457 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1458
1459 return r;
1460}
1461
1462int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
1463{
1464 if (irq->irq == KVM_INTERRUPT_UNSET) {
1465 kvmppc_core_dequeue_external(vcpu);
1466 return 0;
1467 }
1468
1469 kvmppc_core_queue_external(vcpu, irq);
1470
1471 kvm_vcpu_kick(vcpu);
1472
1473 return 0;
1474}
1475
1476static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
1477 struct kvm_enable_cap *cap)
1478{
1479 int r;
1480
1481 if (cap->flags)
1482 return -EINVAL;
1483
1484 switch (cap->cap) {
1485 case KVM_CAP_PPC_OSI:
1486 r = 0;
1487 vcpu->arch.osi_enabled = true;
1488 break;
1489 case KVM_CAP_PPC_PAPR:
1490 r = 0;
1491 vcpu->arch.papr_enabled = true;
1492 break;
1493 case KVM_CAP_PPC_EPR:
1494 r = 0;
1495 if (cap->args[0])
1496 vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
1497 else
1498 vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
1499 break;
1500#ifdef CONFIG_BOOKE
1501 case KVM_CAP_PPC_BOOKE_WATCHDOG:
1502 r = 0;
1503 vcpu->arch.watchdog_enabled = true;
1504 break;
1505#endif
1506#if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1507 case KVM_CAP_SW_TLB: {
1508 struct kvm_config_tlb cfg;
1509 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
1510
1511 r = -EFAULT;
1512 if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
1513 break;
1514
1515 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
1516 break;
1517 }
1518#endif
1519#ifdef CONFIG_KVM_MPIC
1520 case KVM_CAP_IRQ_MPIC: {
1521 struct fd f;
1522 struct kvm_device *dev;
1523
1524 r = -EBADF;
1525 f = fdget(cap->args[0]);
1526 if (!f.file)
1527 break;
1528
1529 r = -EPERM;
1530 dev = kvm_device_from_filp(f.file);
1531 if (dev)
1532 r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
1533
1534 fdput(f);
1535 break;
1536 }
1537#endif
1538#ifdef CONFIG_KVM_XICS
1539 case KVM_CAP_IRQ_XICS: {
1540 struct fd f;
1541 struct kvm_device *dev;
1542
1543 r = -EBADF;
1544 f = fdget(cap->args[0]);
1545 if (!f.file)
1546 break;
1547
1548 r = -EPERM;
1549 dev = kvm_device_from_filp(f.file);
1550 if (dev) {
1551 if (xive_enabled())
1552 r = kvmppc_xive_connect_vcpu(dev, vcpu, cap->args[1]);
1553 else
1554 r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
1555 }
1556
1557 fdput(f);
1558 break;
1559 }
1560#endif /* CONFIG_KVM_XICS */
1561#ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1562 case KVM_CAP_PPC_FWNMI:
1563 r = -EINVAL;
1564 if (!is_kvmppc_hv_enabled(vcpu->kvm))
1565 break;
1566 r = 0;
1567 vcpu->kvm->arch.fwnmi_enabled = true;
1568 break;
1569#endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
1570 default:
1571 r = -EINVAL;
1572 break;
1573 }
1574
1575 if (!r)
1576 r = kvmppc_sanity_check(vcpu);
1577
1578 return r;
1579}
1580
1581bool kvm_arch_intc_initialized(struct kvm *kvm)
1582{
1583#ifdef CONFIG_KVM_MPIC
1584 if (kvm->arch.mpic)
1585 return true;
1586#endif
1587#ifdef CONFIG_KVM_XICS
1588 if (kvm->arch.xics || kvm->arch.xive)
1589 return true;
1590#endif
1591 return false;
1592}
1593
1594int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1595 struct kvm_mp_state *mp_state)
1596{
1597 return -EINVAL;
1598}
1599
1600int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1601 struct kvm_mp_state *mp_state)
1602{
1603 return -EINVAL;
1604}
1605
1606long kvm_arch_vcpu_ioctl(struct file *filp,
1607 unsigned int ioctl, unsigned long arg)
1608{
1609 struct kvm_vcpu *vcpu = filp->private_data;
1610 void __user *argp = (void __user *)arg;
1611 long r;
1612
1613 switch (ioctl) {
1614 case KVM_INTERRUPT: {
1615 struct kvm_interrupt irq;
1616 r = -EFAULT;
1617 if (copy_from_user(&irq, argp, sizeof(irq)))
1618 goto out;
1619 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1620 goto out;
1621 }
1622
1623 case KVM_ENABLE_CAP:
1624 {
1625 struct kvm_enable_cap cap;
1626 r = -EFAULT;
1627 if (copy_from_user(&cap, argp, sizeof(cap)))
1628 goto out;
1629 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
1630 break;
1631 }
1632
1633 case KVM_SET_ONE_REG:
1634 case KVM_GET_ONE_REG:
1635 {
1636 struct kvm_one_reg reg;
1637 r = -EFAULT;
1638 if (copy_from_user(&reg, argp, sizeof(reg)))
1639 goto out;
1640 if (ioctl == KVM_SET_ONE_REG)
1641 r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
1642 else
1643 r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
1644 break;
1645 }
1646
1647#if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1648 case KVM_DIRTY_TLB: {
1649 struct kvm_dirty_tlb dirty;
1650 r = -EFAULT;
1651 if (copy_from_user(&dirty, argp, sizeof(dirty)))
1652 goto out;
1653 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
1654 break;
1655 }
1656#endif
1657 default:
1658 r = -EINVAL;
1659 }
1660
1661out:
1662 return r;
1663}
1664
1665int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
1666{
1667 return VM_FAULT_SIGBUS;
1668}
1669
1670static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
1671{
1672 u32 inst_nop = 0x60000000;
1673#ifdef CONFIG_KVM_BOOKE_HV
1674 u32 inst_sc1 = 0x44000022;
1675 pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
1676 pvinfo->hcall[1] = cpu_to_be32(inst_nop);
1677 pvinfo->hcall[2] = cpu_to_be32(inst_nop);
1678 pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1679#else
1680 u32 inst_lis = 0x3c000000;
1681 u32 inst_ori = 0x60000000;
1682 u32 inst_sc = 0x44000002;
1683 u32 inst_imm_mask = 0xffff;
1684
1685 /*
1686 * The hypercall to get into KVM from within guest context is as
1687 * follows:
1688 *
1689 * lis r0, r0, KVM_SC_MAGIC_R0@h
1690 * ori r0, KVM_SC_MAGIC_R0@l
1691 * sc
1692 * nop
1693 */
1694 pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
1695 pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
1696 pvinfo->hcall[2] = cpu_to_be32(inst_sc);
1697 pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1698#endif
1699
1700 pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
1701
1702 return 0;
1703}
1704
1705int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
1706 bool line_status)
1707{
1708 if (!irqchip_in_kernel(kvm))
1709 return -ENXIO;
1710
1711 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1712 irq_event->irq, irq_event->level,
1713 line_status);
1714 return 0;
1715}
1716
1717
1718static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
1719 struct kvm_enable_cap *cap)
1720{
1721 int r;
1722
1723 if (cap->flags)
1724 return -EINVAL;
1725
1726 switch (cap->cap) {
1727#ifdef CONFIG_KVM_BOOK3S_64_HANDLER
1728 case KVM_CAP_PPC_ENABLE_HCALL: {
1729 unsigned long hcall = cap->args[0];
1730
1731 r = -EINVAL;
1732 if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
1733 cap->args[1] > 1)
1734 break;
1735 if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
1736 break;
1737 if (cap->args[1])
1738 set_bit(hcall / 4, kvm->arch.enabled_hcalls);
1739 else
1740 clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
1741 r = 0;
1742 break;
1743 }
1744 case KVM_CAP_PPC_SMT: {
1745 unsigned long mode = cap->args[0];
1746 unsigned long flags = cap->args[1];
1747
1748 r = -EINVAL;
1749 if (kvm->arch.kvm_ops->set_smt_mode)
1750 r = kvm->arch.kvm_ops->set_smt_mode(kvm, mode, flags);
1751 break;
1752 }
1753#endif
1754 default:
1755 r = -EINVAL;
1756 break;
1757 }
1758
1759 return r;
1760}
1761
1762long kvm_arch_vm_ioctl(struct file *filp,
1763 unsigned int ioctl, unsigned long arg)
1764{
1765 struct kvm *kvm __maybe_unused = filp->private_data;
1766 void __user *argp = (void __user *)arg;
1767 long r;
1768
1769 switch (ioctl) {
1770 case KVM_PPC_GET_PVINFO: {
1771 struct kvm_ppc_pvinfo pvinfo;
1772 memset(&pvinfo, 0, sizeof(pvinfo));
1773 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
1774 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
1775 r = -EFAULT;
1776 goto out;
1777 }
1778
1779 break;
1780 }
1781 case KVM_ENABLE_CAP:
1782 {
1783 struct kvm_enable_cap cap;
1784 r = -EFAULT;
1785 if (copy_from_user(&cap, argp, sizeof(cap)))
1786 goto out;
1787 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
1788 break;
1789 }
1790#ifdef CONFIG_SPAPR_TCE_IOMMU
1791 case KVM_CREATE_SPAPR_TCE_64: {
1792 struct kvm_create_spapr_tce_64 create_tce_64;
1793
1794 r = -EFAULT;
1795 if (copy_from_user(&create_tce_64, argp, sizeof(create_tce_64)))
1796 goto out;
1797 if (create_tce_64.flags) {
1798 r = -EINVAL;
1799 goto out;
1800 }
1801 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
1802 goto out;
1803 }
1804 case KVM_CREATE_SPAPR_TCE: {
1805 struct kvm_create_spapr_tce create_tce;
1806 struct kvm_create_spapr_tce_64 create_tce_64;
1807
1808 r = -EFAULT;
1809 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
1810 goto out;
1811
1812 create_tce_64.liobn = create_tce.liobn;
1813 create_tce_64.page_shift = IOMMU_PAGE_SHIFT_4K;
1814 create_tce_64.offset = 0;
1815 create_tce_64.size = create_tce.window_size >>
1816 IOMMU_PAGE_SHIFT_4K;
1817 create_tce_64.flags = 0;
1818 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
1819 goto out;
1820 }
1821#endif
1822#ifdef CONFIG_PPC_BOOK3S_64
1823 case KVM_PPC_GET_SMMU_INFO: {
1824 struct kvm_ppc_smmu_info info;
1825 struct kvm *kvm = filp->private_data;
1826
1827 memset(&info, 0, sizeof(info));
1828 r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
1829 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
1830 r = -EFAULT;
1831 break;
1832 }
1833 case KVM_PPC_RTAS_DEFINE_TOKEN: {
1834 struct kvm *kvm = filp->private_data;
1835
1836 r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
1837 break;
1838 }
1839 case KVM_PPC_CONFIGURE_V3_MMU: {
1840 struct kvm *kvm = filp->private_data;
1841 struct kvm_ppc_mmuv3_cfg cfg;
1842
1843 r = -EINVAL;
1844 if (!kvm->arch.kvm_ops->configure_mmu)
1845 goto out;
1846 r = -EFAULT;
1847 if (copy_from_user(&cfg, argp, sizeof(cfg)))
1848 goto out;
1849 r = kvm->arch.kvm_ops->configure_mmu(kvm, &cfg);
1850 break;
1851 }
1852 case KVM_PPC_GET_RMMU_INFO: {
1853 struct kvm *kvm = filp->private_data;
1854 struct kvm_ppc_rmmu_info info;
1855
1856 r = -EINVAL;
1857 if (!kvm->arch.kvm_ops->get_rmmu_info)
1858 goto out;
1859 r = kvm->arch.kvm_ops->get_rmmu_info(kvm, &info);
1860 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
1861 r = -EFAULT;
1862 break;
1863 }
1864 default: {
1865 struct kvm *kvm = filp->private_data;
1866 r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
1867 }
1868#else /* CONFIG_PPC_BOOK3S_64 */
1869 default:
1870 r = -ENOTTY;
1871#endif
1872 }
1873out:
1874 return r;
1875}
1876
1877static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
1878static unsigned long nr_lpids;
1879
1880long kvmppc_alloc_lpid(void)
1881{
1882 long lpid;
1883
1884 do {
1885 lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
1886 if (lpid >= nr_lpids) {
1887 pr_err("%s: No LPIDs free\n", __func__);
1888 return -ENOMEM;
1889 }
1890 } while (test_and_set_bit(lpid, lpid_inuse));
1891
1892 return lpid;
1893}
1894EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
1895
1896void kvmppc_claim_lpid(long lpid)
1897{
1898 set_bit(lpid, lpid_inuse);
1899}
1900EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
1901
1902void kvmppc_free_lpid(long lpid)
1903{
1904 clear_bit(lpid, lpid_inuse);
1905}
1906EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
1907
1908void kvmppc_init_lpid(unsigned long nr_lpids_param)
1909{
1910 nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
1911 memset(lpid_inuse, 0, sizeof(lpid_inuse));
1912}
1913EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
1914
1915int kvm_arch_init(void *opaque)
1916{
1917 return 0;
1918}
1919
1920EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr);