]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - arch/powerpc/kvm/powerpc.c
projects / mirror_ubuntu-artful-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
KVM: PPC: Book3S HV: Virtualize doorbell facility on POWER9
[mirror_ubuntu-artful-kernel.git] / arch / powerpc / kvm / powerpc.c
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
50 struct kvmppc_ops *kvmppc_hv_ops;
51 EXPORT_SYMBOL_GPL(kvmppc_hv_ops);
52 struct kvmppc_ops *kvmppc_pr_ops;
53 EXPORT_SYMBOL_GPL(kvmppc_pr_ops);
54
55
56 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
57 {
58 return !!(v->arch.pending_exceptions) ||
59 v->requests;
60 }
61
62 int 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 */
76 int 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 }
136 EXPORT_SYMBOL_GPL(kvmppc_prepare_to_enter);
137
138 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
139 static 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
159 int 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 }
249 EXPORT_SYMBOL_GPL(kvmppc_kvm_pv);
250
251 int 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
274 out:
275 vcpu->arch.sane = r;
276 return r ? 0 : -EINVAL;
277 }
278 EXPORT_SYMBOL_GPL(kvmppc_sanity_check);
279
280 int 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 }
320 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio);
321
322 int 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 }
356 EXPORT_SYMBOL_GPL(kvmppc_st);
357
358 int 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 }
395 EXPORT_SYMBOL_GPL(kvmppc_ld);
396
397 int kvm_arch_hardware_enable(void)
398 {
399 return 0;
400 }
401
402 int kvm_arch_hardware_setup(void)
403 {
404 return 0;
405 }
406
407 void kvm_arch_check_processor_compat(void *rtn)
408 {
409 *(int *)rtn = kvmppc_core_check_processor_compat();
410 }
411
412 int 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);
441 err_out:
442 return -EINVAL;
443 }
444
445 bool kvm_arch_has_vcpu_debugfs(void)
446 {
447 return false;
448 }
449
450 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
451 {
452 return 0;
453 }
454
455 void 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
487 int 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_RMA:
570 r = 0;
571 break;
572 case KVM_CAP_PPC_HWRNG:
573 r = kvmppc_hwrng_present();
574 break;
575 case KVM_CAP_PPC_MMU_RADIX:
576 r = !!(hv_enabled && radix_enabled());
577 break;
578 case KVM_CAP_PPC_MMU_HASH_V3:
579 r = !!(hv_enabled && !radix_enabled() &&
580 cpu_has_feature(CPU_FTR_ARCH_300));
581 break;
582 #endif
583 case KVM_CAP_SYNC_MMU:
584 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
585 r = hv_enabled;
586 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
587 r = 1;
588 #else
589 r = 0;
590 #endif
591 break;
592 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
593 case KVM_CAP_PPC_HTAB_FD:
594 r = hv_enabled;
595 break;
596 #endif
597 case KVM_CAP_NR_VCPUS:
598 /*
599 * Recommending a number of CPUs is somewhat arbitrary; we
600 * return the number of present CPUs for -HV (since a host
601 * will have secondary threads "offline"), and for other KVM
602 * implementations just count online CPUs.
603 */
604 if (hv_enabled)
605 r = num_present_cpus();
606 else
607 r = num_online_cpus();
608 break;
609 case KVM_CAP_NR_MEMSLOTS:
610 r = KVM_USER_MEM_SLOTS;
611 break;
612 case KVM_CAP_MAX_VCPUS:
613 r = KVM_MAX_VCPUS;
614 break;
615 #ifdef CONFIG_PPC_BOOK3S_64
616 case KVM_CAP_PPC_GET_SMMU_INFO:
617 r = 1;
618 break;
619 case KVM_CAP_SPAPR_MULTITCE:
620 r = 1;
621 break;
622 case KVM_CAP_SPAPR_RESIZE_HPT:
623 /* Disable this on POWER9 until code handles new HPTE format */
624 r = !!hv_enabled && !cpu_has_feature(CPU_FTR_ARCH_300);
625 break;
626 #endif
627 case KVM_CAP_PPC_HTM:
628 r = cpu_has_feature(CPU_FTR_TM_COMP) &&
629 is_kvmppc_hv_enabled(kvm);
630 break;
631 default:
632 r = 0;
633 break;
634 }
635 return r;
636
637 }
638
639 long kvm_arch_dev_ioctl(struct file *filp,
640 unsigned int ioctl, unsigned long arg)
641 {
642 return -EINVAL;
643 }
644
645 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
646 struct kvm_memory_slot *dont)
647 {
648 kvmppc_core_free_memslot(kvm, free, dont);
649 }
650
651 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
652 unsigned long npages)
653 {
654 return kvmppc_core_create_memslot(kvm, slot, npages);
655 }
656
657 int kvm_arch_prepare_memory_region(struct kvm *kvm,
658 struct kvm_memory_slot *memslot,
659 const struct kvm_userspace_memory_region *mem,
660 enum kvm_mr_change change)
661 {
662 return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
663 }
664
665 void kvm_arch_commit_memory_region(struct kvm *kvm,
666 const struct kvm_userspace_memory_region *mem,
667 const struct kvm_memory_slot *old,
668 const struct kvm_memory_slot *new,
669 enum kvm_mr_change change)
670 {
671 kvmppc_core_commit_memory_region(kvm, mem, old, new);
672 }
673
674 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
675 struct kvm_memory_slot *slot)
676 {
677 kvmppc_core_flush_memslot(kvm, slot);
678 }
679
680 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
681 {
682 struct kvm_vcpu *vcpu;
683 vcpu = kvmppc_core_vcpu_create(kvm, id);
684 if (!IS_ERR(vcpu)) {
685 vcpu->arch.wqp = &vcpu->wq;
686 kvmppc_create_vcpu_debugfs(vcpu, id);
687 }
688 return vcpu;
689 }
690
691 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
692 {
693 }
694
695 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
696 {
697 /* Make sure we're not using the vcpu anymore */
698 hrtimer_cancel(&vcpu->arch.dec_timer);
699
700 kvmppc_remove_vcpu_debugfs(vcpu);
701
702 switch (vcpu->arch.irq_type) {
703 case KVMPPC_IRQ_MPIC:
704 kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
705 break;
706 case KVMPPC_IRQ_XICS:
707 if (xive_enabled())
708 kvmppc_xive_cleanup_vcpu(vcpu);
709 else
710 kvmppc_xics_free_icp(vcpu);
711 break;
712 }
713
714 kvmppc_core_vcpu_free(vcpu);
715 }
716
717 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
718 {
719 kvm_arch_vcpu_free(vcpu);
720 }
721
722 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
723 {
724 return kvmppc_core_pending_dec(vcpu);
725 }
726
727 static enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
728 {
729 struct kvm_vcpu *vcpu;
730
731 vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
732 kvmppc_decrementer_func(vcpu);
733
734 return HRTIMER_NORESTART;
735 }
736
737 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
738 {
739 int ret;
740
741 hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
742 vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
743 vcpu->arch.dec_expires = ~(u64)0;
744
745 #ifdef CONFIG_KVM_EXIT_TIMING
746 mutex_init(&vcpu->arch.exit_timing_lock);
747 #endif
748 ret = kvmppc_subarch_vcpu_init(vcpu);
749 return ret;
750 }
751
752 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
753 {
754 kvmppc_mmu_destroy(vcpu);
755 kvmppc_subarch_vcpu_uninit(vcpu);
756 }
757
758 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
759 {
760 #ifdef CONFIG_BOOKE
761 /*
762 * vrsave (formerly usprg0) isn't used by Linux, but may
763 * be used by the guest.
764 *
765 * On non-booke this is associated with Altivec and
766 * is handled by code in book3s.c.
767 */
768 mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
769 #endif
770 kvmppc_core_vcpu_load(vcpu, cpu);
771 }
772
773 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
774 {
775 kvmppc_core_vcpu_put(vcpu);
776 #ifdef CONFIG_BOOKE
777 vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
778 #endif
779 }
780
781 /*
782 * irq_bypass_add_producer and irq_bypass_del_producer are only
783 * useful if the architecture supports PCI passthrough.
784 * irq_bypass_stop and irq_bypass_start are not needed and so
785 * kvm_ops are not defined for them.
786 */
787 bool kvm_arch_has_irq_bypass(void)
788 {
789 return ((kvmppc_hv_ops && kvmppc_hv_ops->irq_bypass_add_producer) ||
790 (kvmppc_pr_ops && kvmppc_pr_ops->irq_bypass_add_producer));
791 }
792
793 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
794 struct irq_bypass_producer *prod)
795 {
796 struct kvm_kernel_irqfd *irqfd =
797 container_of(cons, struct kvm_kernel_irqfd, consumer);
798 struct kvm *kvm = irqfd->kvm;
799
800 if (kvm->arch.kvm_ops->irq_bypass_add_producer)
801 return kvm->arch.kvm_ops->irq_bypass_add_producer(cons, prod);
802
803 return 0;
804 }
805
806 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
807 struct irq_bypass_producer *prod)
808 {
809 struct kvm_kernel_irqfd *irqfd =
810 container_of(cons, struct kvm_kernel_irqfd, consumer);
811 struct kvm *kvm = irqfd->kvm;
812
813 if (kvm->arch.kvm_ops->irq_bypass_del_producer)
814 kvm->arch.kvm_ops->irq_bypass_del_producer(cons, prod);
815 }
816
817 #ifdef CONFIG_VSX
818 static inline int kvmppc_get_vsr_dword_offset(int index)
819 {
820 int offset;
821
822 if ((index != 0) && (index != 1))
823 return -1;
824
825 #ifdef __BIG_ENDIAN
826 offset = index;
827 #else
828 offset = 1 - index;
829 #endif
830
831 return offset;
832 }
833
834 static inline int kvmppc_get_vsr_word_offset(int index)
835 {
836 int offset;
837
838 if ((index > 3) || (index < 0))
839 return -1;
840
841 #ifdef __BIG_ENDIAN
842 offset = index;
843 #else
844 offset = 3 - index;
845 #endif
846 return offset;
847 }
848
849 static inline void kvmppc_set_vsr_dword(struct kvm_vcpu *vcpu,
850 u64 gpr)
851 {
852 union kvmppc_one_reg val;
853 int offset = kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
854 int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
855
856 if (offset == -1)
857 return;
858
859 if (vcpu->arch.mmio_vsx_tx_sx_enabled) {
860 val.vval = VCPU_VSX_VR(vcpu, index);
861 val.vsxval[offset] = gpr;
862 VCPU_VSX_VR(vcpu, index) = val.vval;
863 } else {
864 VCPU_VSX_FPR(vcpu, index, offset) = gpr;
865 }
866 }
867
868 static inline void kvmppc_set_vsr_dword_dump(struct kvm_vcpu *vcpu,
869 u64 gpr)
870 {
871 union kvmppc_one_reg val;
872 int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
873
874 if (vcpu->arch.mmio_vsx_tx_sx_enabled) {
875 val.vval = VCPU_VSX_VR(vcpu, index);
876 val.vsxval[0] = gpr;
877 val.vsxval[1] = gpr;
878 VCPU_VSX_VR(vcpu, index) = val.vval;
879 } else {
880 VCPU_VSX_FPR(vcpu, index, 0) = gpr;
881 VCPU_VSX_FPR(vcpu, index, 1) = gpr;
882 }
883 }
884
885 static inline void kvmppc_set_vsr_word(struct kvm_vcpu *vcpu,
886 u32 gpr32)
887 {
888 union kvmppc_one_reg val;
889 int offset = kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
890 int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
891 int dword_offset, word_offset;
892
893 if (offset == -1)
894 return;
895
896 if (vcpu->arch.mmio_vsx_tx_sx_enabled) {
897 val.vval = VCPU_VSX_VR(vcpu, index);
898 val.vsx32val[offset] = gpr32;
899 VCPU_VSX_VR(vcpu, index) = val.vval;
900 } else {
901 dword_offset = offset / 2;
902 word_offset = offset % 2;
903 val.vsxval[0] = VCPU_VSX_FPR(vcpu, index, dword_offset);
904 val.vsx32val[word_offset] = gpr32;
905 VCPU_VSX_FPR(vcpu, index, dword_offset) = val.vsxval[0];
906 }
907 }
908 #endif /* CONFIG_VSX */
909
910 #ifdef CONFIG_PPC_FPU
911 static inline u64 sp_to_dp(u32 fprs)
912 {
913 u64 fprd;
914
915 preempt_disable();
916 enable_kernel_fp();
917 asm ("lfs%U1%X1 0,%1; stfd%U0%X0 0,%0" : "=m" (fprd) : "m" (fprs)
918 : "fr0");
919 preempt_enable();
920 return fprd;
921 }
922
923 static inline u32 dp_to_sp(u64 fprd)
924 {
925 u32 fprs;
926
927 preempt_disable();
928 enable_kernel_fp();
929 asm ("lfd%U1%X1 0,%1; stfs%U0%X0 0,%0" : "=m" (fprs) : "m" (fprd)
930 : "fr0");
931 preempt_enable();
932 return fprs;
933 }
934
935 #else
936 #define sp_to_dp(x) (x)
937 #define dp_to_sp(x) (x)
938 #endif /* CONFIG_PPC_FPU */
939
940 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
941 struct kvm_run *run)
942 {
943 u64 uninitialized_var(gpr);
944
945 if (run->mmio.len > sizeof(gpr)) {
946 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
947 return;
948 }
949
950 if (!vcpu->arch.mmio_host_swabbed) {
951 switch (run->mmio.len) {
952 case 8: gpr = *(u64 *)run->mmio.data; break;
953 case 4: gpr = *(u32 *)run->mmio.data; break;
954 case 2: gpr = *(u16 *)run->mmio.data; break;
955 case 1: gpr = *(u8 *)run->mmio.data; break;
956 }
957 } else {
958 switch (run->mmio.len) {
959 case 8: gpr = swab64(*(u64 *)run->mmio.data); break;
960 case 4: gpr = swab32(*(u32 *)run->mmio.data); break;
961 case 2: gpr = swab16(*(u16 *)run->mmio.data); break;
962 case 1: gpr = *(u8 *)run->mmio.data; break;
963 }
964 }
965
966 /* conversion between single and double precision */
967 if ((vcpu->arch.mmio_sp64_extend) && (run->mmio.len == 4))
968 gpr = sp_to_dp(gpr);
969
970 if (vcpu->arch.mmio_sign_extend) {
971 switch (run->mmio.len) {
972 #ifdef CONFIG_PPC64
973 case 4:
974 gpr = (s64)(s32)gpr;
975 break;
976 #endif
977 case 2:
978 gpr = (s64)(s16)gpr;
979 break;
980 case 1:
981 gpr = (s64)(s8)gpr;
982 break;
983 }
984 }
985
986 switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
987 case KVM_MMIO_REG_GPR:
988 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
989 break;
990 case KVM_MMIO_REG_FPR:
991 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
992 break;
993 #ifdef CONFIG_PPC_BOOK3S
994 case KVM_MMIO_REG_QPR:
995 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
996 break;
997 case KVM_MMIO_REG_FQPR:
998 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
999 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1000 break;
1001 #endif
1002 #ifdef CONFIG_VSX
1003 case KVM_MMIO_REG_VSX:
1004 if (vcpu->arch.mmio_vsx_copy_type == KVMPPC_VSX_COPY_DWORD)
1005 kvmppc_set_vsr_dword(vcpu, gpr);
1006 else if (vcpu->arch.mmio_vsx_copy_type == KVMPPC_VSX_COPY_WORD)
1007 kvmppc_set_vsr_word(vcpu, gpr);
1008 else if (vcpu->arch.mmio_vsx_copy_type ==
1009 KVMPPC_VSX_COPY_DWORD_LOAD_DUMP)
1010 kvmppc_set_vsr_dword_dump(vcpu, gpr);
1011 break;
1012 #endif
1013 default:
1014 BUG();
1015 }
1016 }
1017
1018 static int __kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1019 unsigned int rt, unsigned int bytes,
1020 int is_default_endian, int sign_extend)
1021 {
1022 int idx, ret;
1023 bool host_swabbed;
1024
1025 /* Pity C doesn't have a logical XOR operator */
1026 if (kvmppc_need_byteswap(vcpu)) {
1027 host_swabbed = is_default_endian;
1028 } else {
1029 host_swabbed = !is_default_endian;
1030 }
1031
1032 if (bytes > sizeof(run->mmio.data)) {
1033 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1034 run->mmio.len);
1035 }
1036
1037 run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1038 run->mmio.len = bytes;
1039 run->mmio.is_write = 0;
1040
1041 vcpu->arch.io_gpr = rt;
1042 vcpu->arch.mmio_host_swabbed = host_swabbed;
1043 vcpu->mmio_needed = 1;
1044 vcpu->mmio_is_write = 0;
1045 vcpu->arch.mmio_sign_extend = sign_extend;
1046
1047 idx = srcu_read_lock(&vcpu->kvm->srcu);
1048
1049 ret = kvm_io_bus_read(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1050 bytes, &run->mmio.data);
1051
1052 srcu_read_unlock(&vcpu->kvm->srcu, idx);
1053
1054 if (!ret) {
1055 kvmppc_complete_mmio_load(vcpu, run);
1056 vcpu->mmio_needed = 0;
1057 return EMULATE_DONE;
1058 }
1059
1060 return EMULATE_DO_MMIO;
1061 }
1062
1063 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1064 unsigned int rt, unsigned int bytes,
1065 int is_default_endian)
1066 {
1067 return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 0);
1068 }
1069 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
1070
1071 /* Same as above, but sign extends */
1072 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
1073 unsigned int rt, unsigned int bytes,
1074 int is_default_endian)
1075 {
1076 return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 1);
1077 }
1078
1079 #ifdef CONFIG_VSX
1080 int kvmppc_handle_vsx_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1081 unsigned int rt, unsigned int bytes,
1082 int is_default_endian, int mmio_sign_extend)
1083 {
1084 enum emulation_result emulated = EMULATE_DONE;
1085
1086 /* Currently, mmio_vsx_copy_nums only allowed to be less than 4 */
1087 if ( (vcpu->arch.mmio_vsx_copy_nums > 4) ||
1088 (vcpu->arch.mmio_vsx_copy_nums < 0) ) {
1089 return EMULATE_FAIL;
1090 }
1091
1092 while (vcpu->arch.mmio_vsx_copy_nums) {
1093 emulated = __kvmppc_handle_load(run, vcpu, rt, bytes,
1094 is_default_endian, mmio_sign_extend);
1095
1096 if (emulated != EMULATE_DONE)
1097 break;
1098
1099 vcpu->arch.paddr_accessed += run->mmio.len;
1100
1101 vcpu->arch.mmio_vsx_copy_nums--;
1102 vcpu->arch.mmio_vsx_offset++;
1103 }
1104 return emulated;
1105 }
1106 #endif /* CONFIG_VSX */
1107
1108 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1109 u64 val, unsigned int bytes, int is_default_endian)
1110 {
1111 void *data = run->mmio.data;
1112 int idx, ret;
1113 bool host_swabbed;
1114
1115 /* Pity C doesn't have a logical XOR operator */
1116 if (kvmppc_need_byteswap(vcpu)) {
1117 host_swabbed = is_default_endian;
1118 } else {
1119 host_swabbed = !is_default_endian;
1120 }
1121
1122 if (bytes > sizeof(run->mmio.data)) {
1123 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1124 run->mmio.len);
1125 }
1126
1127 run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1128 run->mmio.len = bytes;
1129 run->mmio.is_write = 1;
1130 vcpu->mmio_needed = 1;
1131 vcpu->mmio_is_write = 1;
1132
1133 if ((vcpu->arch.mmio_sp64_extend) && (bytes == 4))
1134 val = dp_to_sp(val);
1135
1136 /* Store the value at the lowest bytes in 'data'. */
1137 if (!host_swabbed) {
1138 switch (bytes) {
1139 case 8: *(u64 *)data = val; break;
1140 case 4: *(u32 *)data = val; break;
1141 case 2: *(u16 *)data = val; break;
1142 case 1: *(u8 *)data = val; break;
1143 }
1144 } else {
1145 switch (bytes) {
1146 case 8: *(u64 *)data = swab64(val); break;
1147 case 4: *(u32 *)data = swab32(val); break;
1148 case 2: *(u16 *)data = swab16(val); break;
1149 case 1: *(u8 *)data = val; break;
1150 }
1151 }
1152
1153 idx = srcu_read_lock(&vcpu->kvm->srcu);
1154
1155 ret = kvm_io_bus_write(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1156 bytes, &run->mmio.data);
1157
1158 srcu_read_unlock(&vcpu->kvm->srcu, idx);
1159
1160 if (!ret) {
1161 vcpu->mmio_needed = 0;
1162 return EMULATE_DONE;
1163 }
1164
1165 return EMULATE_DO_MMIO;
1166 }
1167 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
1168
1169 #ifdef CONFIG_VSX
1170 static inline int kvmppc_get_vsr_data(struct kvm_vcpu *vcpu, int rs, u64 *val)
1171 {
1172 u32 dword_offset, word_offset;
1173 union kvmppc_one_reg reg;
1174 int vsx_offset = 0;
1175 int copy_type = vcpu->arch.mmio_vsx_copy_type;
1176 int result = 0;
1177
1178 switch (copy_type) {
1179 case KVMPPC_VSX_COPY_DWORD:
1180 vsx_offset =
1181 kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
1182
1183 if (vsx_offset == -1) {
1184 result = -1;
1185 break;
1186 }
1187
1188 if (!vcpu->arch.mmio_vsx_tx_sx_enabled) {
1189 *val = VCPU_VSX_FPR(vcpu, rs, vsx_offset);
1190 } else {
1191 reg.vval = VCPU_VSX_VR(vcpu, rs);
1192 *val = reg.vsxval[vsx_offset];
1193 }
1194 break;
1195
1196 case KVMPPC_VSX_COPY_WORD:
1197 vsx_offset =
1198 kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
1199
1200 if (vsx_offset == -1) {
1201 result = -1;
1202 break;
1203 }
1204
1205 if (!vcpu->arch.mmio_vsx_tx_sx_enabled) {
1206 dword_offset = vsx_offset / 2;
1207 word_offset = vsx_offset % 2;
1208 reg.vsxval[0] = VCPU_VSX_FPR(vcpu, rs, dword_offset);
1209 *val = reg.vsx32val[word_offset];
1210 } else {
1211 reg.vval = VCPU_VSX_VR(vcpu, rs);
1212 *val = reg.vsx32val[vsx_offset];
1213 }
1214 break;
1215
1216 default:
1217 result = -1;
1218 break;
1219 }
1220
1221 return result;
1222 }
1223
1224 int kvmppc_handle_vsx_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1225 int rs, unsigned int bytes, int is_default_endian)
1226 {
1227 u64 val;
1228 enum emulation_result emulated = EMULATE_DONE;
1229
1230 vcpu->arch.io_gpr = rs;
1231
1232 /* Currently, mmio_vsx_copy_nums only allowed to be less than 4 */
1233 if ( (vcpu->arch.mmio_vsx_copy_nums > 4) ||
1234 (vcpu->arch.mmio_vsx_copy_nums < 0) ) {
1235 return EMULATE_FAIL;
1236 }
1237
1238 while (vcpu->arch.mmio_vsx_copy_nums) {
1239 if (kvmppc_get_vsr_data(vcpu, rs, &val) == -1)
1240 return EMULATE_FAIL;
1241
1242 emulated = kvmppc_handle_store(run, vcpu,
1243 val, bytes, is_default_endian);
1244
1245 if (emulated != EMULATE_DONE)
1246 break;
1247
1248 vcpu->arch.paddr_accessed += run->mmio.len;
1249
1250 vcpu->arch.mmio_vsx_copy_nums--;
1251 vcpu->arch.mmio_vsx_offset++;
1252 }
1253
1254 return emulated;
1255 }
1256
1257 static int kvmppc_emulate_mmio_vsx_loadstore(struct kvm_vcpu *vcpu,
1258 struct kvm_run *run)
1259 {
1260 enum emulation_result emulated = EMULATE_FAIL;
1261 int r;
1262
1263 vcpu->arch.paddr_accessed += run->mmio.len;
1264
1265 if (!vcpu->mmio_is_write) {
1266 emulated = kvmppc_handle_vsx_load(run, vcpu, vcpu->arch.io_gpr,
1267 run->mmio.len, 1, vcpu->arch.mmio_sign_extend);
1268 } else {
1269 emulated = kvmppc_handle_vsx_store(run, vcpu,
1270 vcpu->arch.io_gpr, run->mmio.len, 1);
1271 }
1272
1273 switch (emulated) {
1274 case EMULATE_DO_MMIO:
1275 run->exit_reason = KVM_EXIT_MMIO;
1276 r = RESUME_HOST;
1277 break;
1278 case EMULATE_FAIL:
1279 pr_info("KVM: MMIO emulation failed (VSX repeat)\n");
1280 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1281 run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1282 r = RESUME_HOST;
1283 break;
1284 default:
1285 r = RESUME_GUEST;
1286 break;
1287 }
1288 return r;
1289 }
1290 #endif /* CONFIG_VSX */
1291
1292 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1293 {
1294 int r = 0;
1295 union kvmppc_one_reg val;
1296 int size;
1297
1298 size = one_reg_size(reg->id);
1299 if (size > sizeof(val))
1300 return -EINVAL;
1301
1302 r = kvmppc_get_one_reg(vcpu, reg->id, &val);
1303 if (r == -EINVAL) {
1304 r = 0;
1305 switch (reg->id) {
1306 #ifdef CONFIG_ALTIVEC
1307 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1308 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1309 r = -ENXIO;
1310 break;
1311 }
1312 val.vval = vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0];
1313 break;
1314 case KVM_REG_PPC_VSCR:
1315 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1316 r = -ENXIO;
1317 break;
1318 }
1319 val = get_reg_val(reg->id, vcpu->arch.vr.vscr.u[3]);
1320 break;
1321 case KVM_REG_PPC_VRSAVE:
1322 val = get_reg_val(reg->id, vcpu->arch.vrsave);
1323 break;
1324 #endif /* CONFIG_ALTIVEC */
1325 default:
1326 r = -EINVAL;
1327 break;
1328 }
1329 }
1330
1331 if (r)
1332 return r;
1333
1334 if (copy_to_user((char __user *)(unsigned long)reg->addr, &val, size))
1335 r = -EFAULT;
1336
1337 return r;
1338 }
1339
1340 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1341 {
1342 int r;
1343 union kvmppc_one_reg val;
1344 int size;
1345
1346 size = one_reg_size(reg->id);
1347 if (size > sizeof(val))
1348 return -EINVAL;
1349
1350 if (copy_from_user(&val, (char __user *)(unsigned long)reg->addr, size))
1351 return -EFAULT;
1352
1353 r = kvmppc_set_one_reg(vcpu, reg->id, &val);
1354 if (r == -EINVAL) {
1355 r = 0;
1356 switch (reg->id) {
1357 #ifdef CONFIG_ALTIVEC
1358 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1359 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1360 r = -ENXIO;
1361 break;
1362 }
1363 vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0] = val.vval;
1364 break;
1365 case KVM_REG_PPC_VSCR:
1366 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1367 r = -ENXIO;
1368 break;
1369 }
1370 vcpu->arch.vr.vscr.u[3] = set_reg_val(reg->id, val);
1371 break;
1372 case KVM_REG_PPC_VRSAVE:
1373 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1374 r = -ENXIO;
1375 break;
1376 }
1377 vcpu->arch.vrsave = set_reg_val(reg->id, val);
1378 break;
1379 #endif /* CONFIG_ALTIVEC */
1380 default:
1381 r = -EINVAL;
1382 break;
1383 }
1384 }
1385
1386 return r;
1387 }
1388
1389 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
1390 {
1391 int r;
1392 sigset_t sigsaved;
1393
1394 if (vcpu->mmio_needed) {
1395 vcpu->mmio_needed = 0;
1396 if (!vcpu->mmio_is_write)
1397 kvmppc_complete_mmio_load(vcpu, run);
1398 #ifdef CONFIG_VSX
1399 if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1400 vcpu->arch.mmio_vsx_copy_nums--;
1401 vcpu->arch.mmio_vsx_offset++;
1402 }
1403
1404 if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1405 r = kvmppc_emulate_mmio_vsx_loadstore(vcpu, run);
1406 if (r == RESUME_HOST) {
1407 vcpu->mmio_needed = 1;
1408 return r;
1409 }
1410 }
1411 #endif
1412 } else if (vcpu->arch.osi_needed) {
1413 u64 *gprs = run->osi.gprs;
1414 int i;
1415
1416 for (i = 0; i < 32; i++)
1417 kvmppc_set_gpr(vcpu, i, gprs[i]);
1418 vcpu->arch.osi_needed = 0;
1419 } else if (vcpu->arch.hcall_needed) {
1420 int i;
1421
1422 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
1423 for (i = 0; i < 9; ++i)
1424 kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
1425 vcpu->arch.hcall_needed = 0;
1426 #ifdef CONFIG_BOOKE
1427 } else if (vcpu->arch.epr_needed) {
1428 kvmppc_set_epr(vcpu, run->epr.epr);
1429 vcpu->arch.epr_needed = 0;
1430 #endif
1431 }
1432
1433 if (vcpu->sigset_active)
1434 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1435
1436 if (run->immediate_exit)
1437 r = -EINTR;
1438 else
1439 r = kvmppc_vcpu_run(run, vcpu);
1440
1441 if (vcpu->sigset_active)
1442 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1443
1444 return r;
1445 }
1446
1447 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
1448 {
1449 if (irq->irq == KVM_INTERRUPT_UNSET) {
1450 kvmppc_core_dequeue_external(vcpu);
1451 return 0;
1452 }
1453
1454 kvmppc_core_queue_external(vcpu, irq);
1455
1456 kvm_vcpu_kick(vcpu);
1457
1458 return 0;
1459 }
1460
1461 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
1462 struct kvm_enable_cap *cap)
1463 {
1464 int r;
1465
1466 if (cap->flags)
1467 return -EINVAL;
1468
1469 switch (cap->cap) {
1470 case KVM_CAP_PPC_OSI:
1471 r = 0;
1472 vcpu->arch.osi_enabled = true;
1473 break;
1474 case KVM_CAP_PPC_PAPR:
1475 r = 0;
1476 vcpu->arch.papr_enabled = true;
1477 break;
1478 case KVM_CAP_PPC_EPR:
1479 r = 0;
1480 if (cap->args[0])
1481 vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
1482 else
1483 vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
1484 break;
1485 #ifdef CONFIG_BOOKE
1486 case KVM_CAP_PPC_BOOKE_WATCHDOG:
1487 r = 0;
1488 vcpu->arch.watchdog_enabled = true;
1489 break;
1490 #endif
1491 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1492 case KVM_CAP_SW_TLB: {
1493 struct kvm_config_tlb cfg;
1494 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
1495
1496 r = -EFAULT;
1497 if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
1498 break;
1499
1500 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
1501 break;
1502 }
1503 #endif
1504 #ifdef CONFIG_KVM_MPIC
1505 case KVM_CAP_IRQ_MPIC: {
1506 struct fd f;
1507 struct kvm_device *dev;
1508
1509 r = -EBADF;
1510 f = fdget(cap->args[0]);
1511 if (!f.file)
1512 break;
1513
1514 r = -EPERM;
1515 dev = kvm_device_from_filp(f.file);
1516 if (dev)
1517 r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
1518
1519 fdput(f);
1520 break;
1521 }
1522 #endif
1523 #ifdef CONFIG_KVM_XICS
1524 case KVM_CAP_IRQ_XICS: {
1525 struct fd f;
1526 struct kvm_device *dev;
1527
1528 r = -EBADF;
1529 f = fdget(cap->args[0]);
1530 if (!f.file)
1531 break;
1532
1533 r = -EPERM;
1534 dev = kvm_device_from_filp(f.file);
1535 if (dev) {
1536 if (xive_enabled())
1537 r = kvmppc_xive_connect_vcpu(dev, vcpu, cap->args[1]);
1538 else
1539 r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
1540 }
1541
1542 fdput(f);
1543 break;
1544 }
1545 #endif /* CONFIG_KVM_XICS */
1546 default:
1547 r = -EINVAL;
1548 break;
1549 }
1550
1551 if (!r)
1552 r = kvmppc_sanity_check(vcpu);
1553
1554 return r;
1555 }
1556
1557 bool kvm_arch_intc_initialized(struct kvm *kvm)
1558 {
1559 #ifdef CONFIG_KVM_MPIC
1560 if (kvm->arch.mpic)
1561 return true;
1562 #endif
1563 #ifdef CONFIG_KVM_XICS
1564 if (kvm->arch.xics || kvm->arch.xive)
1565 return true;
1566 #endif
1567 return false;
1568 }
1569
1570 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1571 struct kvm_mp_state *mp_state)
1572 {
1573 return -EINVAL;
1574 }
1575
1576 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1577 struct kvm_mp_state *mp_state)
1578 {
1579 return -EINVAL;
1580 }
1581
1582 long kvm_arch_vcpu_ioctl(struct file *filp,
1583 unsigned int ioctl, unsigned long arg)
1584 {
1585 struct kvm_vcpu *vcpu = filp->private_data;
1586 void __user *argp = (void __user *)arg;
1587 long r;
1588
1589 switch (ioctl) {
1590 case KVM_INTERRUPT: {
1591 struct kvm_interrupt irq;
1592 r = -EFAULT;
1593 if (copy_from_user(&irq, argp, sizeof(irq)))
1594 goto out;
1595 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1596 goto out;
1597 }
1598
1599 case KVM_ENABLE_CAP:
1600 {
1601 struct kvm_enable_cap cap;
1602 r = -EFAULT;
1603 if (copy_from_user(&cap, argp, sizeof(cap)))
1604 goto out;
1605 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
1606 break;
1607 }
1608
1609 case KVM_SET_ONE_REG:
1610 case KVM_GET_ONE_REG:
1611 {
1612 struct kvm_one_reg reg;
1613 r = -EFAULT;
1614 if (copy_from_user(&reg, argp, sizeof(reg)))
1615 goto out;
1616 if (ioctl == KVM_SET_ONE_REG)
1617 r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
1618 else
1619 r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
1620 break;
1621 }
1622
1623 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1624 case KVM_DIRTY_TLB: {
1625 struct kvm_dirty_tlb dirty;
1626 r = -EFAULT;
1627 if (copy_from_user(&dirty, argp, sizeof(dirty)))
1628 goto out;
1629 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
1630 break;
1631 }
1632 #endif
1633 default:
1634 r = -EINVAL;
1635 }
1636
1637 out:
1638 return r;
1639 }
1640
1641 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
1642 {
1643 return VM_FAULT_SIGBUS;
1644 }
1645
1646 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
1647 {
1648 u32 inst_nop = 0x60000000;
1649 #ifdef CONFIG_KVM_BOOKE_HV
1650 u32 inst_sc1 = 0x44000022;
1651 pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
1652 pvinfo->hcall[1] = cpu_to_be32(inst_nop);
1653 pvinfo->hcall[2] = cpu_to_be32(inst_nop);
1654 pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1655 #else
1656 u32 inst_lis = 0x3c000000;
1657 u32 inst_ori = 0x60000000;
1658 u32 inst_sc = 0x44000002;
1659 u32 inst_imm_mask = 0xffff;
1660
1661 /*
1662 * The hypercall to get into KVM from within guest context is as
1663 * follows:
1664 *
1665 * lis r0, r0, KVM_SC_MAGIC_R0@h
1666 * ori r0, KVM_SC_MAGIC_R0@l
1667 * sc
1668 * nop
1669 */
1670 pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
1671 pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
1672 pvinfo->hcall[2] = cpu_to_be32(inst_sc);
1673 pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1674 #endif
1675
1676 pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
1677
1678 return 0;
1679 }
1680
1681 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
1682 bool line_status)
1683 {
1684 if (!irqchip_in_kernel(kvm))
1685 return -ENXIO;
1686
1687 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1688 irq_event->irq, irq_event->level,
1689 line_status);
1690 return 0;
1691 }
1692
1693
1694 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
1695 struct kvm_enable_cap *cap)
1696 {
1697 int r;
1698
1699 if (cap->flags)
1700 return -EINVAL;
1701
1702 switch (cap->cap) {
1703 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
1704 case KVM_CAP_PPC_ENABLE_HCALL: {
1705 unsigned long hcall = cap->args[0];
1706
1707 r = -EINVAL;
1708 if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
1709 cap->args[1] > 1)
1710 break;
1711 if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
1712 break;
1713 if (cap->args[1])
1714 set_bit(hcall / 4, kvm->arch.enabled_hcalls);
1715 else
1716 clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
1717 r = 0;
1718 break;
1719 }
1720 case KVM_CAP_PPC_SMT: {
1721 unsigned long mode = cap->args[0];
1722 unsigned long flags = cap->args[1];
1723
1724 r = -EINVAL;
1725 if (kvm->arch.kvm_ops->set_smt_mode)
1726 r = kvm->arch.kvm_ops->set_smt_mode(kvm, mode, flags);
1727 break;
1728 }
1729 #endif
1730 default:
1731 r = -EINVAL;
1732 break;
1733 }
1734
1735 return r;
1736 }
1737
1738 long kvm_arch_vm_ioctl(struct file *filp,
1739 unsigned int ioctl, unsigned long arg)
1740 {
1741 struct kvm *kvm __maybe_unused = filp->private_data;
1742 void __user *argp = (void __user *)arg;
1743 long r;
1744
1745 switch (ioctl) {
1746 case KVM_PPC_GET_PVINFO: {
1747 struct kvm_ppc_pvinfo pvinfo;
1748 memset(&pvinfo, 0, sizeof(pvinfo));
1749 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
1750 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
1751 r = -EFAULT;
1752 goto out;
1753 }
1754
1755 break;
1756 }
1757 case KVM_ENABLE_CAP:
1758 {
1759 struct kvm_enable_cap cap;
1760 r = -EFAULT;
1761 if (copy_from_user(&cap, argp, sizeof(cap)))
1762 goto out;
1763 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
1764 break;
1765 }
1766 #ifdef CONFIG_SPAPR_TCE_IOMMU
1767 case KVM_CREATE_SPAPR_TCE_64: {
1768 struct kvm_create_spapr_tce_64 create_tce_64;
1769
1770 r = -EFAULT;
1771 if (copy_from_user(&create_tce_64, argp, sizeof(create_tce_64)))
1772 goto out;
1773 if (create_tce_64.flags) {
1774 r = -EINVAL;
1775 goto out;
1776 }
1777 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
1778 goto out;
1779 }
1780 case KVM_CREATE_SPAPR_TCE: {
1781 struct kvm_create_spapr_tce create_tce;
1782 struct kvm_create_spapr_tce_64 create_tce_64;
1783
1784 r = -EFAULT;
1785 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
1786 goto out;
1787
1788 create_tce_64.liobn = create_tce.liobn;
1789 create_tce_64.page_shift = IOMMU_PAGE_SHIFT_4K;
1790 create_tce_64.offset = 0;
1791 create_tce_64.size = create_tce.window_size >>
1792 IOMMU_PAGE_SHIFT_4K;
1793 create_tce_64.flags = 0;
1794 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
1795 goto out;
1796 }
1797 #endif
1798 #ifdef CONFIG_PPC_BOOK3S_64
1799 case KVM_PPC_GET_SMMU_INFO: {
1800 struct kvm_ppc_smmu_info info;
1801 struct kvm *kvm = filp->private_data;
1802
1803 memset(&info, 0, sizeof(info));
1804 r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
1805 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
1806 r = -EFAULT;
1807 break;
1808 }
1809 case KVM_PPC_RTAS_DEFINE_TOKEN: {
1810 struct kvm *kvm = filp->private_data;
1811
1812 r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
1813 break;
1814 }
1815 case KVM_PPC_CONFIGURE_V3_MMU: {
1816 struct kvm *kvm = filp->private_data;
1817 struct kvm_ppc_mmuv3_cfg cfg;
1818
1819 r = -EINVAL;
1820 if (!kvm->arch.kvm_ops->configure_mmu)
1821 goto out;
1822 r = -EFAULT;
1823 if (copy_from_user(&cfg, argp, sizeof(cfg)))
1824 goto out;
1825 r = kvm->arch.kvm_ops->configure_mmu(kvm, &cfg);
1826 break;
1827 }
1828 case KVM_PPC_GET_RMMU_INFO: {
1829 struct kvm *kvm = filp->private_data;
1830 struct kvm_ppc_rmmu_info info;
1831
1832 r = -EINVAL;
1833 if (!kvm->arch.kvm_ops->get_rmmu_info)
1834 goto out;
1835 r = kvm->arch.kvm_ops->get_rmmu_info(kvm, &info);
1836 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
1837 r = -EFAULT;
1838 break;
1839 }
1840 default: {
1841 struct kvm *kvm = filp->private_data;
1842 r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
1843 }
1844 #else /* CONFIG_PPC_BOOK3S_64 */
1845 default:
1846 r = -ENOTTY;
1847 #endif
1848 }
1849 out:
1850 return r;
1851 }
1852
1853 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
1854 static unsigned long nr_lpids;
1855
1856 long kvmppc_alloc_lpid(void)
1857 {
1858 long lpid;
1859
1860 do {
1861 lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
1862 if (lpid >= nr_lpids) {
1863 pr_err("%s: No LPIDs free\n", __func__);
1864 return -ENOMEM;
1865 }
1866 } while (test_and_set_bit(lpid, lpid_inuse));
1867
1868 return lpid;
1869 }
1870 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
1871
1872 void kvmppc_claim_lpid(long lpid)
1873 {
1874 set_bit(lpid, lpid_inuse);
1875 }
1876 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
1877
1878 void kvmppc_free_lpid(long lpid)
1879 {
1880 clear_bit(lpid, lpid_inuse);
1881 }
1882 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
1883
1884 void kvmppc_init_lpid(unsigned long nr_lpids_param)
1885 {
1886 nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
1887 memset(lpid_inuse, 0, sizeof(lpid_inuse));
1888 }
1889 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
1890
1891 int kvm_arch_init(void *opaque)
1892 {
1893 return 0;
1894 }
1895
1896 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr);
Ubuntu Artful kernel mirror