xref: /openbmc/qemu/target/ppc/kvm.c (revision b097ba37)
1 /*
2  * PowerPC implementation of KVM hooks
3  *
4  * Copyright IBM Corp. 2007
5  * Copyright (C) 2011 Freescale Semiconductor, Inc.
6  *
7  * Authors:
8  *  Jerone Young <jyoung5@us.ibm.com>
9  *  Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
10  *  Hollis Blanchard <hollisb@us.ibm.com>
11  *
12  * This work is licensed under the terms of the GNU GPL, version 2 or later.
13  * See the COPYING file in the top-level directory.
14  *
15  */
16 
17 #include "qemu/osdep.h"
18 #include <dirent.h>
19 #include <sys/ioctl.h>
20 #include <sys/vfs.h>
21 
22 #include <linux/kvm.h>
23 
24 #include "qemu-common.h"
25 #include "qapi/error.h"
26 #include "qemu/error-report.h"
27 #include "cpu.h"
28 #include "cpu-models.h"
29 #include "qemu/timer.h"
30 #include "sysemu/sysemu.h"
31 #include "sysemu/hw_accel.h"
32 #include "kvm_ppc.h"
33 #include "sysemu/cpus.h"
34 #include "sysemu/device_tree.h"
35 #include "mmu-hash64.h"
36 
37 #include "hw/sysbus.h"
38 #include "hw/ppc/spapr.h"
39 #include "hw/ppc/spapr_cpu_core.h"
40 #include "hw/ppc/ppc.h"
41 #include "sysemu/watchdog.h"
42 #include "trace.h"
43 #include "exec/gdbstub.h"
44 #include "exec/memattrs.h"
45 #include "exec/ram_addr.h"
46 #include "sysemu/hostmem.h"
47 #include "qemu/cutils.h"
48 #include "qemu/mmap-alloc.h"
49 #include "elf.h"
50 #include "sysemu/kvm_int.h"
51 
52 #define PROC_DEVTREE_CPU      "/proc/device-tree/cpus/"
53 
54 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
55     KVM_CAP_LAST_INFO
56 };
57 
58 static int cap_interrupt_unset;
59 static int cap_interrupt_level;
60 static int cap_segstate;
61 static int cap_booke_sregs;
62 static int cap_ppc_smt;
63 static int cap_ppc_smt_possible;
64 static int cap_spapr_tce;
65 static int cap_spapr_tce_64;
66 static int cap_spapr_multitce;
67 static int cap_spapr_vfio;
68 static int cap_hior;
69 static int cap_one_reg;
70 static int cap_epr;
71 static int cap_ppc_watchdog;
72 static int cap_papr;
73 static int cap_htab_fd;
74 static int cap_fixup_hcalls;
75 static int cap_htm;             /* Hardware transactional memory support */
76 static int cap_mmu_radix;
77 static int cap_mmu_hash_v3;
78 static int cap_resize_hpt;
79 static int cap_ppc_pvr_compat;
80 static int cap_ppc_safe_cache;
81 static int cap_ppc_safe_bounds_check;
82 static int cap_ppc_safe_indirect_branch;
83 static int cap_ppc_count_cache_flush_assist;
84 static int cap_ppc_nested_kvm_hv;
85 static int cap_large_decr;
86 
87 static uint32_t debug_inst_opcode;
88 
89 /*
90  * XXX We have a race condition where we actually have a level triggered
91  *     interrupt, but the infrastructure can't expose that yet, so the guest
92  *     takes but ignores it, goes to sleep and never gets notified that there's
93  *     still an interrupt pending.
94  *
95  *     As a quick workaround, let's just wake up again 20 ms after we injected
96  *     an interrupt. That way we can assure that we're always reinjecting
97  *     interrupts in case the guest swallowed them.
98  */
99 static QEMUTimer *idle_timer;
100 
101 static void kvm_kick_cpu(void *opaque)
102 {
103     PowerPCCPU *cpu = opaque;
104 
105     qemu_cpu_kick(CPU(cpu));
106 }
107 
108 /*
109  * Check whether we are running with KVM-PR (instead of KVM-HV).  This
110  * should only be used for fallback tests - generally we should use
111  * explicit capabilities for the features we want, rather than
112  * assuming what is/isn't available depending on the KVM variant.
113  */
114 static bool kvmppc_is_pr(KVMState *ks)
115 {
116     /* Assume KVM-PR if the GET_PVINFO capability is available */
117     return kvm_vm_check_extension(ks, KVM_CAP_PPC_GET_PVINFO) != 0;
118 }
119 
120 static int kvm_ppc_register_host_cpu_type(MachineState *ms);
121 static void kvmppc_get_cpu_characteristics(KVMState *s);
122 static int kvmppc_get_dec_bits(void);
123 
124 int kvm_arch_init(MachineState *ms, KVMState *s)
125 {
126     cap_interrupt_unset = kvm_check_extension(s, KVM_CAP_PPC_UNSET_IRQ);
127     cap_interrupt_level = kvm_check_extension(s, KVM_CAP_PPC_IRQ_LEVEL);
128     cap_segstate = kvm_check_extension(s, KVM_CAP_PPC_SEGSTATE);
129     cap_booke_sregs = kvm_check_extension(s, KVM_CAP_PPC_BOOKE_SREGS);
130     cap_ppc_smt_possible = kvm_vm_check_extension(s, KVM_CAP_PPC_SMT_POSSIBLE);
131     cap_spapr_tce = kvm_check_extension(s, KVM_CAP_SPAPR_TCE);
132     cap_spapr_tce_64 = kvm_check_extension(s, KVM_CAP_SPAPR_TCE_64);
133     cap_spapr_multitce = kvm_check_extension(s, KVM_CAP_SPAPR_MULTITCE);
134     cap_spapr_vfio = kvm_vm_check_extension(s, KVM_CAP_SPAPR_TCE_VFIO);
135     cap_one_reg = kvm_check_extension(s, KVM_CAP_ONE_REG);
136     cap_hior = kvm_check_extension(s, KVM_CAP_PPC_HIOR);
137     cap_epr = kvm_check_extension(s, KVM_CAP_PPC_EPR);
138     cap_ppc_watchdog = kvm_check_extension(s, KVM_CAP_PPC_BOOKE_WATCHDOG);
139     /*
140      * Note: we don't set cap_papr here, because this capability is
141      * only activated after this by kvmppc_set_papr()
142      */
143     cap_htab_fd = kvm_vm_check_extension(s, KVM_CAP_PPC_HTAB_FD);
144     cap_fixup_hcalls = kvm_check_extension(s, KVM_CAP_PPC_FIXUP_HCALL);
145     cap_ppc_smt = kvm_vm_check_extension(s, KVM_CAP_PPC_SMT);
146     cap_htm = kvm_vm_check_extension(s, KVM_CAP_PPC_HTM);
147     cap_mmu_radix = kvm_vm_check_extension(s, KVM_CAP_PPC_MMU_RADIX);
148     cap_mmu_hash_v3 = kvm_vm_check_extension(s, KVM_CAP_PPC_MMU_HASH_V3);
149     cap_resize_hpt = kvm_vm_check_extension(s, KVM_CAP_SPAPR_RESIZE_HPT);
150     kvmppc_get_cpu_characteristics(s);
151     cap_ppc_nested_kvm_hv = kvm_vm_check_extension(s, KVM_CAP_PPC_NESTED_HV);
152     cap_large_decr = kvmppc_get_dec_bits();
153     /*
154      * Note: setting it to false because there is not such capability
155      * in KVM at this moment.
156      *
157      * TODO: call kvm_vm_check_extension() with the right capability
158      * after the kernel starts implementing it.
159      */
160     cap_ppc_pvr_compat = false;
161 
162     if (!cap_interrupt_level) {
163         fprintf(stderr, "KVM: Couldn't find level irq capability. Expect the "
164                         "VM to stall at times!\n");
165     }
166 
167     kvm_ppc_register_host_cpu_type(ms);
168 
169     return 0;
170 }
171 
172 int kvm_arch_irqchip_create(MachineState *ms, KVMState *s)
173 {
174     return 0;
175 }
176 
177 static int kvm_arch_sync_sregs(PowerPCCPU *cpu)
178 {
179     CPUPPCState *cenv = &cpu->env;
180     CPUState *cs = CPU(cpu);
181     struct kvm_sregs sregs;
182     int ret;
183 
184     if (cenv->excp_model == POWERPC_EXCP_BOOKE) {
185         /*
186          * What we're really trying to say is "if we're on BookE, we
187          * use the native PVR for now". This is the only sane way to
188          * check it though, so we potentially confuse users that they
189          * can run BookE guests on BookS. Let's hope nobody dares
190          * enough :)
191          */
192         return 0;
193     } else {
194         if (!cap_segstate) {
195             fprintf(stderr, "kvm error: missing PVR setting capability\n");
196             return -ENOSYS;
197         }
198     }
199 
200     ret = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs);
201     if (ret) {
202         return ret;
203     }
204 
205     sregs.pvr = cenv->spr[SPR_PVR];
206     return kvm_vcpu_ioctl(cs, KVM_SET_SREGS, &sregs);
207 }
208 
209 /* Set up a shared TLB array with KVM */
210 static int kvm_booke206_tlb_init(PowerPCCPU *cpu)
211 {
212     CPUPPCState *env = &cpu->env;
213     CPUState *cs = CPU(cpu);
214     struct kvm_book3e_206_tlb_params params = {};
215     struct kvm_config_tlb cfg = {};
216     unsigned int entries = 0;
217     int ret, i;
218 
219     if (!kvm_enabled() ||
220         !kvm_check_extension(cs->kvm_state, KVM_CAP_SW_TLB)) {
221         return 0;
222     }
223 
224     assert(ARRAY_SIZE(params.tlb_sizes) == BOOKE206_MAX_TLBN);
225 
226     for (i = 0; i < BOOKE206_MAX_TLBN; i++) {
227         params.tlb_sizes[i] = booke206_tlb_size(env, i);
228         params.tlb_ways[i] = booke206_tlb_ways(env, i);
229         entries += params.tlb_sizes[i];
230     }
231 
232     assert(entries == env->nb_tlb);
233     assert(sizeof(struct kvm_book3e_206_tlb_entry) == sizeof(ppcmas_tlb_t));
234 
235     env->tlb_dirty = true;
236 
237     cfg.array = (uintptr_t)env->tlb.tlbm;
238     cfg.array_len = sizeof(ppcmas_tlb_t) * entries;
239     cfg.params = (uintptr_t)&params;
240     cfg.mmu_type = KVM_MMU_FSL_BOOKE_NOHV;
241 
242     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_SW_TLB, 0, (uintptr_t)&cfg);
243     if (ret < 0) {
244         fprintf(stderr, "%s: couldn't enable KVM_CAP_SW_TLB: %s\n",
245                 __func__, strerror(-ret));
246         return ret;
247     }
248 
249     env->kvm_sw_tlb = true;
250     return 0;
251 }
252 
253 
254 #if defined(TARGET_PPC64)
255 static void kvm_get_smmu_info(struct kvm_ppc_smmu_info *info, Error **errp)
256 {
257     int ret;
258 
259     assert(kvm_state != NULL);
260 
261     if (!kvm_check_extension(kvm_state, KVM_CAP_PPC_GET_SMMU_INFO)) {
262         error_setg(errp, "KVM doesn't expose the MMU features it supports");
263         error_append_hint(errp, "Consider switching to a newer KVM\n");
264         return;
265     }
266 
267     ret = kvm_vm_ioctl(kvm_state, KVM_PPC_GET_SMMU_INFO, info);
268     if (ret == 0) {
269         return;
270     }
271 
272     error_setg_errno(errp, -ret,
273                      "KVM failed to provide the MMU features it supports");
274 }
275 
276 struct ppc_radix_page_info *kvm_get_radix_page_info(void)
277 {
278     KVMState *s = KVM_STATE(current_machine->accelerator);
279     struct ppc_radix_page_info *radix_page_info;
280     struct kvm_ppc_rmmu_info rmmu_info;
281     int i;
282 
283     if (!kvm_check_extension(s, KVM_CAP_PPC_MMU_RADIX)) {
284         return NULL;
285     }
286     if (kvm_vm_ioctl(s, KVM_PPC_GET_RMMU_INFO, &rmmu_info)) {
287         return NULL;
288     }
289     radix_page_info = g_malloc0(sizeof(*radix_page_info));
290     radix_page_info->count = 0;
291     for (i = 0; i < PPC_PAGE_SIZES_MAX_SZ; i++) {
292         if (rmmu_info.ap_encodings[i]) {
293             radix_page_info->entries[i] = rmmu_info.ap_encodings[i];
294             radix_page_info->count++;
295         }
296     }
297     return radix_page_info;
298 }
299 
300 target_ulong kvmppc_configure_v3_mmu(PowerPCCPU *cpu,
301                                      bool radix, bool gtse,
302                                      uint64_t proc_tbl)
303 {
304     CPUState *cs = CPU(cpu);
305     int ret;
306     uint64_t flags = 0;
307     struct kvm_ppc_mmuv3_cfg cfg = {
308         .process_table = proc_tbl,
309     };
310 
311     if (radix) {
312         flags |= KVM_PPC_MMUV3_RADIX;
313     }
314     if (gtse) {
315         flags |= KVM_PPC_MMUV3_GTSE;
316     }
317     cfg.flags = flags;
318     ret = kvm_vm_ioctl(cs->kvm_state, KVM_PPC_CONFIGURE_V3_MMU, &cfg);
319     switch (ret) {
320     case 0:
321         return H_SUCCESS;
322     case -EINVAL:
323         return H_PARAMETER;
324     case -ENODEV:
325         return H_NOT_AVAILABLE;
326     default:
327         return H_HARDWARE;
328     }
329 }
330 
331 bool kvmppc_hpt_needs_host_contiguous_pages(void)
332 {
333     static struct kvm_ppc_smmu_info smmu_info;
334 
335     if (!kvm_enabled()) {
336         return false;
337     }
338 
339     kvm_get_smmu_info(&smmu_info, &error_fatal);
340     return !!(smmu_info.flags & KVM_PPC_PAGE_SIZES_REAL);
341 }
342 
343 void kvm_check_mmu(PowerPCCPU *cpu, Error **errp)
344 {
345     struct kvm_ppc_smmu_info smmu_info;
346     int iq, ik, jq, jk;
347     Error *local_err = NULL;
348 
349     /* For now, we only have anything to check on hash64 MMUs */
350     if (!cpu->hash64_opts || !kvm_enabled()) {
351         return;
352     }
353 
354     kvm_get_smmu_info(&smmu_info, &local_err);
355     if (local_err) {
356         error_propagate(errp, local_err);
357         return;
358     }
359 
360     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)
361         && !(smmu_info.flags & KVM_PPC_1T_SEGMENTS)) {
362         error_setg(errp,
363                    "KVM does not support 1TiB segments which guest expects");
364         return;
365     }
366 
367     if (smmu_info.slb_size < cpu->hash64_opts->slb_size) {
368         error_setg(errp, "KVM only supports %u SLB entries, but guest needs %u",
369                    smmu_info.slb_size, cpu->hash64_opts->slb_size);
370         return;
371     }
372 
373     /*
374      * Verify that every pagesize supported by the cpu model is
375      * supported by KVM with the same encodings
376      */
377     for (iq = 0; iq < ARRAY_SIZE(cpu->hash64_opts->sps); iq++) {
378         PPCHash64SegmentPageSizes *qsps = &cpu->hash64_opts->sps[iq];
379         struct kvm_ppc_one_seg_page_size *ksps;
380 
381         for (ik = 0; ik < ARRAY_SIZE(smmu_info.sps); ik++) {
382             if (qsps->page_shift == smmu_info.sps[ik].page_shift) {
383                 break;
384             }
385         }
386         if (ik >= ARRAY_SIZE(smmu_info.sps)) {
387             error_setg(errp, "KVM doesn't support for base page shift %u",
388                        qsps->page_shift);
389             return;
390         }
391 
392         ksps = &smmu_info.sps[ik];
393         if (ksps->slb_enc != qsps->slb_enc) {
394             error_setg(errp,
395 "KVM uses SLB encoding 0x%x for page shift %u, but guest expects 0x%x",
396                        ksps->slb_enc, ksps->page_shift, qsps->slb_enc);
397             return;
398         }
399 
400         for (jq = 0; jq < ARRAY_SIZE(qsps->enc); jq++) {
401             for (jk = 0; jk < ARRAY_SIZE(ksps->enc); jk++) {
402                 if (qsps->enc[jq].page_shift == ksps->enc[jk].page_shift) {
403                     break;
404                 }
405             }
406 
407             if (jk >= ARRAY_SIZE(ksps->enc)) {
408                 error_setg(errp, "KVM doesn't support page shift %u/%u",
409                            qsps->enc[jq].page_shift, qsps->page_shift);
410                 return;
411             }
412             if (qsps->enc[jq].pte_enc != ksps->enc[jk].pte_enc) {
413                 error_setg(errp,
414 "KVM uses PTE encoding 0x%x for page shift %u/%u, but guest expects 0x%x",
415                            ksps->enc[jk].pte_enc, qsps->enc[jq].page_shift,
416                            qsps->page_shift, qsps->enc[jq].pte_enc);
417                 return;
418             }
419         }
420     }
421 
422     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
423         /*
424          * Mostly what guest pagesizes we can use are related to the
425          * host pages used to map guest RAM, which is handled in the
426          * platform code. Cache-Inhibited largepages (64k) however are
427          * used for I/O, so if they're mapped to the host at all it
428          * will be a normal mapping, not a special hugepage one used
429          * for RAM.
430          */
431         if (getpagesize() < 0x10000) {
432             error_setg(errp,
433                        "KVM can't supply 64kiB CI pages, which guest expects");
434         }
435     }
436 }
437 #endif /* !defined (TARGET_PPC64) */
438 
439 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
440 {
441     return POWERPC_CPU(cpu)->vcpu_id;
442 }
443 
444 /*
445  * e500 supports 2 h/w breakpoint and 2 watchpoint.  book3s supports
446  * only 1 watchpoint, so array size of 4 is sufficient for now.
447  */
448 #define MAX_HW_BKPTS 4
449 
450 static struct HWBreakpoint {
451     target_ulong addr;
452     int type;
453 } hw_debug_points[MAX_HW_BKPTS];
454 
455 static CPUWatchpoint hw_watchpoint;
456 
457 /* Default there is no breakpoint and watchpoint supported */
458 static int max_hw_breakpoint;
459 static int max_hw_watchpoint;
460 static int nb_hw_breakpoint;
461 static int nb_hw_watchpoint;
462 
463 static void kvmppc_hw_debug_points_init(CPUPPCState *cenv)
464 {
465     if (cenv->excp_model == POWERPC_EXCP_BOOKE) {
466         max_hw_breakpoint = 2;
467         max_hw_watchpoint = 2;
468     }
469 
470     if ((max_hw_breakpoint + max_hw_watchpoint) > MAX_HW_BKPTS) {
471         fprintf(stderr, "Error initializing h/w breakpoints\n");
472         return;
473     }
474 }
475 
476 int kvm_arch_init_vcpu(CPUState *cs)
477 {
478     PowerPCCPU *cpu = POWERPC_CPU(cs);
479     CPUPPCState *cenv = &cpu->env;
480     int ret;
481 
482     /* Synchronize sregs with kvm */
483     ret = kvm_arch_sync_sregs(cpu);
484     if (ret) {
485         if (ret == -EINVAL) {
486             error_report("Register sync failed... If you're using kvm-hv.ko,"
487                          " only \"-cpu host\" is possible");
488         }
489         return ret;
490     }
491 
492     idle_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, kvm_kick_cpu, cpu);
493 
494     switch (cenv->mmu_model) {
495     case POWERPC_MMU_BOOKE206:
496         /* This target supports access to KVM's guest TLB */
497         ret = kvm_booke206_tlb_init(cpu);
498         break;
499     case POWERPC_MMU_2_07:
500         if (!cap_htm && !kvmppc_is_pr(cs->kvm_state)) {
501             /*
502              * KVM-HV has transactional memory on POWER8 also without
503              * the KVM_CAP_PPC_HTM extension, so enable it here
504              * instead as long as it's availble to userspace on the
505              * host.
506              */
507             if (qemu_getauxval(AT_HWCAP2) & PPC_FEATURE2_HAS_HTM) {
508                 cap_htm = true;
509             }
510         }
511         break;
512     default:
513         break;
514     }
515 
516     kvm_get_one_reg(cs, KVM_REG_PPC_DEBUG_INST, &debug_inst_opcode);
517     kvmppc_hw_debug_points_init(cenv);
518 
519     return ret;
520 }
521 
522 static void kvm_sw_tlb_put(PowerPCCPU *cpu)
523 {
524     CPUPPCState *env = &cpu->env;
525     CPUState *cs = CPU(cpu);
526     struct kvm_dirty_tlb dirty_tlb;
527     unsigned char *bitmap;
528     int ret;
529 
530     if (!env->kvm_sw_tlb) {
531         return;
532     }
533 
534     bitmap = g_malloc((env->nb_tlb + 7) / 8);
535     memset(bitmap, 0xFF, (env->nb_tlb + 7) / 8);
536 
537     dirty_tlb.bitmap = (uintptr_t)bitmap;
538     dirty_tlb.num_dirty = env->nb_tlb;
539 
540     ret = kvm_vcpu_ioctl(cs, KVM_DIRTY_TLB, &dirty_tlb);
541     if (ret) {
542         fprintf(stderr, "%s: KVM_DIRTY_TLB: %s\n",
543                 __func__, strerror(-ret));
544     }
545 
546     g_free(bitmap);
547 }
548 
549 static void kvm_get_one_spr(CPUState *cs, uint64_t id, int spr)
550 {
551     PowerPCCPU *cpu = POWERPC_CPU(cs);
552     CPUPPCState *env = &cpu->env;
553     union {
554         uint32_t u32;
555         uint64_t u64;
556     } val;
557     struct kvm_one_reg reg = {
558         .id = id,
559         .addr = (uintptr_t) &val,
560     };
561     int ret;
562 
563     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
564     if (ret != 0) {
565         trace_kvm_failed_spr_get(spr, strerror(errno));
566     } else {
567         switch (id & KVM_REG_SIZE_MASK) {
568         case KVM_REG_SIZE_U32:
569             env->spr[spr] = val.u32;
570             break;
571 
572         case KVM_REG_SIZE_U64:
573             env->spr[spr] = val.u64;
574             break;
575 
576         default:
577             /* Don't handle this size yet */
578             abort();
579         }
580     }
581 }
582 
583 static void kvm_put_one_spr(CPUState *cs, uint64_t id, int spr)
584 {
585     PowerPCCPU *cpu = POWERPC_CPU(cs);
586     CPUPPCState *env = &cpu->env;
587     union {
588         uint32_t u32;
589         uint64_t u64;
590     } val;
591     struct kvm_one_reg reg = {
592         .id = id,
593         .addr = (uintptr_t) &val,
594     };
595     int ret;
596 
597     switch (id & KVM_REG_SIZE_MASK) {
598     case KVM_REG_SIZE_U32:
599         val.u32 = env->spr[spr];
600         break;
601 
602     case KVM_REG_SIZE_U64:
603         val.u64 = env->spr[spr];
604         break;
605 
606     default:
607         /* Don't handle this size yet */
608         abort();
609     }
610 
611     ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
612     if (ret != 0) {
613         trace_kvm_failed_spr_set(spr, strerror(errno));
614     }
615 }
616 
617 static int kvm_put_fp(CPUState *cs)
618 {
619     PowerPCCPU *cpu = POWERPC_CPU(cs);
620     CPUPPCState *env = &cpu->env;
621     struct kvm_one_reg reg;
622     int i;
623     int ret;
624 
625     if (env->insns_flags & PPC_FLOAT) {
626         uint64_t fpscr = env->fpscr;
627         bool vsx = !!(env->insns_flags2 & PPC2_VSX);
628 
629         reg.id = KVM_REG_PPC_FPSCR;
630         reg.addr = (uintptr_t)&fpscr;
631         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
632         if (ret < 0) {
633             trace_kvm_failed_fpscr_set(strerror(errno));
634             return ret;
635         }
636 
637         for (i = 0; i < 32; i++) {
638             uint64_t vsr[2];
639             uint64_t *fpr = cpu_fpr_ptr(&cpu->env, i);
640             uint64_t *vsrl = cpu_vsrl_ptr(&cpu->env, i);
641 
642 #ifdef HOST_WORDS_BIGENDIAN
643             vsr[0] = float64_val(*fpr);
644             vsr[1] = *vsrl;
645 #else
646             vsr[0] = *vsrl;
647             vsr[1] = float64_val(*fpr);
648 #endif
649             reg.addr = (uintptr_t) &vsr;
650             reg.id = vsx ? KVM_REG_PPC_VSR(i) : KVM_REG_PPC_FPR(i);
651 
652             ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
653             if (ret < 0) {
654                 trace_kvm_failed_fp_set(vsx ? "VSR" : "FPR", i,
655                                         strerror(errno));
656                 return ret;
657             }
658         }
659     }
660 
661     if (env->insns_flags & PPC_ALTIVEC) {
662         reg.id = KVM_REG_PPC_VSCR;
663         reg.addr = (uintptr_t)&env->vscr;
664         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
665         if (ret < 0) {
666             trace_kvm_failed_vscr_set(strerror(errno));
667             return ret;
668         }
669 
670         for (i = 0; i < 32; i++) {
671             reg.id = KVM_REG_PPC_VR(i);
672             reg.addr = (uintptr_t)cpu_avr_ptr(env, i);
673             ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
674             if (ret < 0) {
675                 trace_kvm_failed_vr_set(i, strerror(errno));
676                 return ret;
677             }
678         }
679     }
680 
681     return 0;
682 }
683 
684 static int kvm_get_fp(CPUState *cs)
685 {
686     PowerPCCPU *cpu = POWERPC_CPU(cs);
687     CPUPPCState *env = &cpu->env;
688     struct kvm_one_reg reg;
689     int i;
690     int ret;
691 
692     if (env->insns_flags & PPC_FLOAT) {
693         uint64_t fpscr;
694         bool vsx = !!(env->insns_flags2 & PPC2_VSX);
695 
696         reg.id = KVM_REG_PPC_FPSCR;
697         reg.addr = (uintptr_t)&fpscr;
698         ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
699         if (ret < 0) {
700             trace_kvm_failed_fpscr_get(strerror(errno));
701             return ret;
702         } else {
703             env->fpscr = fpscr;
704         }
705 
706         for (i = 0; i < 32; i++) {
707             uint64_t vsr[2];
708             uint64_t *fpr = cpu_fpr_ptr(&cpu->env, i);
709             uint64_t *vsrl = cpu_vsrl_ptr(&cpu->env, i);
710 
711             reg.addr = (uintptr_t) &vsr;
712             reg.id = vsx ? KVM_REG_PPC_VSR(i) : KVM_REG_PPC_FPR(i);
713 
714             ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
715             if (ret < 0) {
716                 trace_kvm_failed_fp_get(vsx ? "VSR" : "FPR", i,
717                                         strerror(errno));
718                 return ret;
719             } else {
720 #ifdef HOST_WORDS_BIGENDIAN
721                 *fpr = vsr[0];
722                 if (vsx) {
723                     *vsrl = vsr[1];
724                 }
725 #else
726                 *fpr = vsr[1];
727                 if (vsx) {
728                     *vsrl = vsr[0];
729                 }
730 #endif
731             }
732         }
733     }
734 
735     if (env->insns_flags & PPC_ALTIVEC) {
736         reg.id = KVM_REG_PPC_VSCR;
737         reg.addr = (uintptr_t)&env->vscr;
738         ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
739         if (ret < 0) {
740             trace_kvm_failed_vscr_get(strerror(errno));
741             return ret;
742         }
743 
744         for (i = 0; i < 32; i++) {
745             reg.id = KVM_REG_PPC_VR(i);
746             reg.addr = (uintptr_t)cpu_avr_ptr(env, i);
747             ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
748             if (ret < 0) {
749                 trace_kvm_failed_vr_get(i, strerror(errno));
750                 return ret;
751             }
752         }
753     }
754 
755     return 0;
756 }
757 
758 #if defined(TARGET_PPC64)
759 static int kvm_get_vpa(CPUState *cs)
760 {
761     PowerPCCPU *cpu = POWERPC_CPU(cs);
762     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
763     struct kvm_one_reg reg;
764     int ret;
765 
766     reg.id = KVM_REG_PPC_VPA_ADDR;
767     reg.addr = (uintptr_t)&spapr_cpu->vpa_addr;
768     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
769     if (ret < 0) {
770         trace_kvm_failed_vpa_addr_get(strerror(errno));
771         return ret;
772     }
773 
774     assert((uintptr_t)&spapr_cpu->slb_shadow_size
775            == ((uintptr_t)&spapr_cpu->slb_shadow_addr + 8));
776     reg.id = KVM_REG_PPC_VPA_SLB;
777     reg.addr = (uintptr_t)&spapr_cpu->slb_shadow_addr;
778     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
779     if (ret < 0) {
780         trace_kvm_failed_slb_get(strerror(errno));
781         return ret;
782     }
783 
784     assert((uintptr_t)&spapr_cpu->dtl_size
785            == ((uintptr_t)&spapr_cpu->dtl_addr + 8));
786     reg.id = KVM_REG_PPC_VPA_DTL;
787     reg.addr = (uintptr_t)&spapr_cpu->dtl_addr;
788     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
789     if (ret < 0) {
790         trace_kvm_failed_dtl_get(strerror(errno));
791         return ret;
792     }
793 
794     return 0;
795 }
796 
797 static int kvm_put_vpa(CPUState *cs)
798 {
799     PowerPCCPU *cpu = POWERPC_CPU(cs);
800     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
801     struct kvm_one_reg reg;
802     int ret;
803 
804     /*
805      * SLB shadow or DTL can't be registered unless a master VPA is
806      * registered.  That means when restoring state, if a VPA *is*
807      * registered, we need to set that up first.  If not, we need to
808      * deregister the others before deregistering the master VPA
809      */
810     assert(spapr_cpu->vpa_addr
811            || !(spapr_cpu->slb_shadow_addr || spapr_cpu->dtl_addr));
812 
813     if (spapr_cpu->vpa_addr) {
814         reg.id = KVM_REG_PPC_VPA_ADDR;
815         reg.addr = (uintptr_t)&spapr_cpu->vpa_addr;
816         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
817         if (ret < 0) {
818             trace_kvm_failed_vpa_addr_set(strerror(errno));
819             return ret;
820         }
821     }
822 
823     assert((uintptr_t)&spapr_cpu->slb_shadow_size
824            == ((uintptr_t)&spapr_cpu->slb_shadow_addr + 8));
825     reg.id = KVM_REG_PPC_VPA_SLB;
826     reg.addr = (uintptr_t)&spapr_cpu->slb_shadow_addr;
827     ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
828     if (ret < 0) {
829         trace_kvm_failed_slb_set(strerror(errno));
830         return ret;
831     }
832 
833     assert((uintptr_t)&spapr_cpu->dtl_size
834            == ((uintptr_t)&spapr_cpu->dtl_addr + 8));
835     reg.id = KVM_REG_PPC_VPA_DTL;
836     reg.addr = (uintptr_t)&spapr_cpu->dtl_addr;
837     ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
838     if (ret < 0) {
839         trace_kvm_failed_dtl_set(strerror(errno));
840         return ret;
841     }
842 
843     if (!spapr_cpu->vpa_addr) {
844         reg.id = KVM_REG_PPC_VPA_ADDR;
845         reg.addr = (uintptr_t)&spapr_cpu->vpa_addr;
846         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
847         if (ret < 0) {
848             trace_kvm_failed_null_vpa_addr_set(strerror(errno));
849             return ret;
850         }
851     }
852 
853     return 0;
854 }
855 #endif /* TARGET_PPC64 */
856 
857 int kvmppc_put_books_sregs(PowerPCCPU *cpu)
858 {
859     CPUPPCState *env = &cpu->env;
860     struct kvm_sregs sregs;
861     int i;
862 
863     sregs.pvr = env->spr[SPR_PVR];
864 
865     if (cpu->vhyp) {
866         PPCVirtualHypervisorClass *vhc =
867             PPC_VIRTUAL_HYPERVISOR_GET_CLASS(cpu->vhyp);
868         sregs.u.s.sdr1 = vhc->encode_hpt_for_kvm_pr(cpu->vhyp);
869     } else {
870         sregs.u.s.sdr1 = env->spr[SPR_SDR1];
871     }
872 
873     /* Sync SLB */
874 #ifdef TARGET_PPC64
875     for (i = 0; i < ARRAY_SIZE(env->slb); i++) {
876         sregs.u.s.ppc64.slb[i].slbe = env->slb[i].esid;
877         if (env->slb[i].esid & SLB_ESID_V) {
878             sregs.u.s.ppc64.slb[i].slbe |= i;
879         }
880         sregs.u.s.ppc64.slb[i].slbv = env->slb[i].vsid;
881     }
882 #endif
883 
884     /* Sync SRs */
885     for (i = 0; i < 16; i++) {
886         sregs.u.s.ppc32.sr[i] = env->sr[i];
887     }
888 
889     /* Sync BATs */
890     for (i = 0; i < 8; i++) {
891         /* Beware. We have to swap upper and lower bits here */
892         sregs.u.s.ppc32.dbat[i] = ((uint64_t)env->DBAT[0][i] << 32)
893             | env->DBAT[1][i];
894         sregs.u.s.ppc32.ibat[i] = ((uint64_t)env->IBAT[0][i] << 32)
895             | env->IBAT[1][i];
896     }
897 
898     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_SREGS, &sregs);
899 }
900 
901 int kvm_arch_put_registers(CPUState *cs, int level)
902 {
903     PowerPCCPU *cpu = POWERPC_CPU(cs);
904     CPUPPCState *env = &cpu->env;
905     struct kvm_regs regs;
906     int ret;
907     int i;
908 
909     ret = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
910     if (ret < 0) {
911         return ret;
912     }
913 
914     regs.ctr = env->ctr;
915     regs.lr  = env->lr;
916     regs.xer = cpu_read_xer(env);
917     regs.msr = env->msr;
918     regs.pc = env->nip;
919 
920     regs.srr0 = env->spr[SPR_SRR0];
921     regs.srr1 = env->spr[SPR_SRR1];
922 
923     regs.sprg0 = env->spr[SPR_SPRG0];
924     regs.sprg1 = env->spr[SPR_SPRG1];
925     regs.sprg2 = env->spr[SPR_SPRG2];
926     regs.sprg3 = env->spr[SPR_SPRG3];
927     regs.sprg4 = env->spr[SPR_SPRG4];
928     regs.sprg5 = env->spr[SPR_SPRG5];
929     regs.sprg6 = env->spr[SPR_SPRG6];
930     regs.sprg7 = env->spr[SPR_SPRG7];
931 
932     regs.pid = env->spr[SPR_BOOKE_PID];
933 
934     for (i = 0; i < 32; i++) {
935         regs.gpr[i] = env->gpr[i];
936     }
937 
938     regs.cr = 0;
939     for (i = 0; i < 8; i++) {
940         regs.cr |= (env->crf[i] & 15) << (4 * (7 - i));
941     }
942 
943     ret = kvm_vcpu_ioctl(cs, KVM_SET_REGS, &regs);
944     if (ret < 0) {
945         return ret;
946     }
947 
948     kvm_put_fp(cs);
949 
950     if (env->tlb_dirty) {
951         kvm_sw_tlb_put(cpu);
952         env->tlb_dirty = false;
953     }
954 
955     if (cap_segstate && (level >= KVM_PUT_RESET_STATE)) {
956         ret = kvmppc_put_books_sregs(cpu);
957         if (ret < 0) {
958             return ret;
959         }
960     }
961 
962     if (cap_hior && (level >= KVM_PUT_RESET_STATE)) {
963         kvm_put_one_spr(cs, KVM_REG_PPC_HIOR, SPR_HIOR);
964     }
965 
966     if (cap_one_reg) {
967         int i;
968 
969         /*
970          * We deliberately ignore errors here, for kernels which have
971          * the ONE_REG calls, but don't support the specific
972          * registers, there's a reasonable chance things will still
973          * work, at least until we try to migrate.
974          */
975         for (i = 0; i < 1024; i++) {
976             uint64_t id = env->spr_cb[i].one_reg_id;
977 
978             if (id != 0) {
979                 kvm_put_one_spr(cs, id, i);
980             }
981         }
982 
983 #ifdef TARGET_PPC64
984         if (msr_ts) {
985             for (i = 0; i < ARRAY_SIZE(env->tm_gpr); i++) {
986                 kvm_set_one_reg(cs, KVM_REG_PPC_TM_GPR(i), &env->tm_gpr[i]);
987             }
988             for (i = 0; i < ARRAY_SIZE(env->tm_vsr); i++) {
989                 kvm_set_one_reg(cs, KVM_REG_PPC_TM_VSR(i), &env->tm_vsr[i]);
990             }
991             kvm_set_one_reg(cs, KVM_REG_PPC_TM_CR, &env->tm_cr);
992             kvm_set_one_reg(cs, KVM_REG_PPC_TM_LR, &env->tm_lr);
993             kvm_set_one_reg(cs, KVM_REG_PPC_TM_CTR, &env->tm_ctr);
994             kvm_set_one_reg(cs, KVM_REG_PPC_TM_FPSCR, &env->tm_fpscr);
995             kvm_set_one_reg(cs, KVM_REG_PPC_TM_AMR, &env->tm_amr);
996             kvm_set_one_reg(cs, KVM_REG_PPC_TM_PPR, &env->tm_ppr);
997             kvm_set_one_reg(cs, KVM_REG_PPC_TM_VRSAVE, &env->tm_vrsave);
998             kvm_set_one_reg(cs, KVM_REG_PPC_TM_VSCR, &env->tm_vscr);
999             kvm_set_one_reg(cs, KVM_REG_PPC_TM_DSCR, &env->tm_dscr);
1000             kvm_set_one_reg(cs, KVM_REG_PPC_TM_TAR, &env->tm_tar);
1001         }
1002 
1003         if (cap_papr) {
1004             if (kvm_put_vpa(cs) < 0) {
1005                 trace_kvm_failed_put_vpa();
1006             }
1007         }
1008 
1009         kvm_set_one_reg(cs, KVM_REG_PPC_TB_OFFSET, &env->tb_env->tb_offset);
1010 #endif /* TARGET_PPC64 */
1011     }
1012 
1013     return ret;
1014 }
1015 
1016 static void kvm_sync_excp(CPUPPCState *env, int vector, int ivor)
1017 {
1018      env->excp_vectors[vector] = env->spr[ivor] + env->spr[SPR_BOOKE_IVPR];
1019 }
1020 
1021 static int kvmppc_get_booke_sregs(PowerPCCPU *cpu)
1022 {
1023     CPUPPCState *env = &cpu->env;
1024     struct kvm_sregs sregs;
1025     int ret;
1026 
1027     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_SREGS, &sregs);
1028     if (ret < 0) {
1029         return ret;
1030     }
1031 
1032     if (sregs.u.e.features & KVM_SREGS_E_BASE) {
1033         env->spr[SPR_BOOKE_CSRR0] = sregs.u.e.csrr0;
1034         env->spr[SPR_BOOKE_CSRR1] = sregs.u.e.csrr1;
1035         env->spr[SPR_BOOKE_ESR] = sregs.u.e.esr;
1036         env->spr[SPR_BOOKE_DEAR] = sregs.u.e.dear;
1037         env->spr[SPR_BOOKE_MCSR] = sregs.u.e.mcsr;
1038         env->spr[SPR_BOOKE_TSR] = sregs.u.e.tsr;
1039         env->spr[SPR_BOOKE_TCR] = sregs.u.e.tcr;
1040         env->spr[SPR_DECR] = sregs.u.e.dec;
1041         env->spr[SPR_TBL] = sregs.u.e.tb & 0xffffffff;
1042         env->spr[SPR_TBU] = sregs.u.e.tb >> 32;
1043         env->spr[SPR_VRSAVE] = sregs.u.e.vrsave;
1044     }
1045 
1046     if (sregs.u.e.features & KVM_SREGS_E_ARCH206) {
1047         env->spr[SPR_BOOKE_PIR] = sregs.u.e.pir;
1048         env->spr[SPR_BOOKE_MCSRR0] = sregs.u.e.mcsrr0;
1049         env->spr[SPR_BOOKE_MCSRR1] = sregs.u.e.mcsrr1;
1050         env->spr[SPR_BOOKE_DECAR] = sregs.u.e.decar;
1051         env->spr[SPR_BOOKE_IVPR] = sregs.u.e.ivpr;
1052     }
1053 
1054     if (sregs.u.e.features & KVM_SREGS_E_64) {
1055         env->spr[SPR_BOOKE_EPCR] = sregs.u.e.epcr;
1056     }
1057 
1058     if (sregs.u.e.features & KVM_SREGS_E_SPRG8) {
1059         env->spr[SPR_BOOKE_SPRG8] = sregs.u.e.sprg8;
1060     }
1061 
1062     if (sregs.u.e.features & KVM_SREGS_E_IVOR) {
1063         env->spr[SPR_BOOKE_IVOR0] = sregs.u.e.ivor_low[0];
1064         kvm_sync_excp(env, POWERPC_EXCP_CRITICAL,  SPR_BOOKE_IVOR0);
1065         env->spr[SPR_BOOKE_IVOR1] = sregs.u.e.ivor_low[1];
1066         kvm_sync_excp(env, POWERPC_EXCP_MCHECK,  SPR_BOOKE_IVOR1);
1067         env->spr[SPR_BOOKE_IVOR2] = sregs.u.e.ivor_low[2];
1068         kvm_sync_excp(env, POWERPC_EXCP_DSI,  SPR_BOOKE_IVOR2);
1069         env->spr[SPR_BOOKE_IVOR3] = sregs.u.e.ivor_low[3];
1070         kvm_sync_excp(env, POWERPC_EXCP_ISI,  SPR_BOOKE_IVOR3);
1071         env->spr[SPR_BOOKE_IVOR4] = sregs.u.e.ivor_low[4];
1072         kvm_sync_excp(env, POWERPC_EXCP_EXTERNAL,  SPR_BOOKE_IVOR4);
1073         env->spr[SPR_BOOKE_IVOR5] = sregs.u.e.ivor_low[5];
1074         kvm_sync_excp(env, POWERPC_EXCP_ALIGN,  SPR_BOOKE_IVOR5);
1075         env->spr[SPR_BOOKE_IVOR6] = sregs.u.e.ivor_low[6];
1076         kvm_sync_excp(env, POWERPC_EXCP_PROGRAM,  SPR_BOOKE_IVOR6);
1077         env->spr[SPR_BOOKE_IVOR7] = sregs.u.e.ivor_low[7];
1078         kvm_sync_excp(env, POWERPC_EXCP_FPU,  SPR_BOOKE_IVOR7);
1079         env->spr[SPR_BOOKE_IVOR8] = sregs.u.e.ivor_low[8];
1080         kvm_sync_excp(env, POWERPC_EXCP_SYSCALL,  SPR_BOOKE_IVOR8);
1081         env->spr[SPR_BOOKE_IVOR9] = sregs.u.e.ivor_low[9];
1082         kvm_sync_excp(env, POWERPC_EXCP_APU,  SPR_BOOKE_IVOR9);
1083         env->spr[SPR_BOOKE_IVOR10] = sregs.u.e.ivor_low[10];
1084         kvm_sync_excp(env, POWERPC_EXCP_DECR,  SPR_BOOKE_IVOR10);
1085         env->spr[SPR_BOOKE_IVOR11] = sregs.u.e.ivor_low[11];
1086         kvm_sync_excp(env, POWERPC_EXCP_FIT,  SPR_BOOKE_IVOR11);
1087         env->spr[SPR_BOOKE_IVOR12] = sregs.u.e.ivor_low[12];
1088         kvm_sync_excp(env, POWERPC_EXCP_WDT,  SPR_BOOKE_IVOR12);
1089         env->spr[SPR_BOOKE_IVOR13] = sregs.u.e.ivor_low[13];
1090         kvm_sync_excp(env, POWERPC_EXCP_DTLB,  SPR_BOOKE_IVOR13);
1091         env->spr[SPR_BOOKE_IVOR14] = sregs.u.e.ivor_low[14];
1092         kvm_sync_excp(env, POWERPC_EXCP_ITLB,  SPR_BOOKE_IVOR14);
1093         env->spr[SPR_BOOKE_IVOR15] = sregs.u.e.ivor_low[15];
1094         kvm_sync_excp(env, POWERPC_EXCP_DEBUG,  SPR_BOOKE_IVOR15);
1095 
1096         if (sregs.u.e.features & KVM_SREGS_E_SPE) {
1097             env->spr[SPR_BOOKE_IVOR32] = sregs.u.e.ivor_high[0];
1098             kvm_sync_excp(env, POWERPC_EXCP_SPEU,  SPR_BOOKE_IVOR32);
1099             env->spr[SPR_BOOKE_IVOR33] = sregs.u.e.ivor_high[1];
1100             kvm_sync_excp(env, POWERPC_EXCP_EFPDI,  SPR_BOOKE_IVOR33);
1101             env->spr[SPR_BOOKE_IVOR34] = sregs.u.e.ivor_high[2];
1102             kvm_sync_excp(env, POWERPC_EXCP_EFPRI,  SPR_BOOKE_IVOR34);
1103         }
1104 
1105         if (sregs.u.e.features & KVM_SREGS_E_PM) {
1106             env->spr[SPR_BOOKE_IVOR35] = sregs.u.e.ivor_high[3];
1107             kvm_sync_excp(env, POWERPC_EXCP_EPERFM,  SPR_BOOKE_IVOR35);
1108         }
1109 
1110         if (sregs.u.e.features & KVM_SREGS_E_PC) {
1111             env->spr[SPR_BOOKE_IVOR36] = sregs.u.e.ivor_high[4];
1112             kvm_sync_excp(env, POWERPC_EXCP_DOORI,  SPR_BOOKE_IVOR36);
1113             env->spr[SPR_BOOKE_IVOR37] = sregs.u.e.ivor_high[5];
1114             kvm_sync_excp(env, POWERPC_EXCP_DOORCI, SPR_BOOKE_IVOR37);
1115         }
1116     }
1117 
1118     if (sregs.u.e.features & KVM_SREGS_E_ARCH206_MMU) {
1119         env->spr[SPR_BOOKE_MAS0] = sregs.u.e.mas0;
1120         env->spr[SPR_BOOKE_MAS1] = sregs.u.e.mas1;
1121         env->spr[SPR_BOOKE_MAS2] = sregs.u.e.mas2;
1122         env->spr[SPR_BOOKE_MAS3] = sregs.u.e.mas7_3 & 0xffffffff;
1123         env->spr[SPR_BOOKE_MAS4] = sregs.u.e.mas4;
1124         env->spr[SPR_BOOKE_MAS6] = sregs.u.e.mas6;
1125         env->spr[SPR_BOOKE_MAS7] = sregs.u.e.mas7_3 >> 32;
1126         env->spr[SPR_MMUCFG] = sregs.u.e.mmucfg;
1127         env->spr[SPR_BOOKE_TLB0CFG] = sregs.u.e.tlbcfg[0];
1128         env->spr[SPR_BOOKE_TLB1CFG] = sregs.u.e.tlbcfg[1];
1129     }
1130 
1131     if (sregs.u.e.features & KVM_SREGS_EXP) {
1132         env->spr[SPR_BOOKE_EPR] = sregs.u.e.epr;
1133     }
1134 
1135     if (sregs.u.e.features & KVM_SREGS_E_PD) {
1136         env->spr[SPR_BOOKE_EPLC] = sregs.u.e.eplc;
1137         env->spr[SPR_BOOKE_EPSC] = sregs.u.e.epsc;
1138     }
1139 
1140     if (sregs.u.e.impl_id == KVM_SREGS_E_IMPL_FSL) {
1141         env->spr[SPR_E500_SVR] = sregs.u.e.impl.fsl.svr;
1142         env->spr[SPR_Exxx_MCAR] = sregs.u.e.impl.fsl.mcar;
1143         env->spr[SPR_HID0] = sregs.u.e.impl.fsl.hid0;
1144 
1145         if (sregs.u.e.impl.fsl.features & KVM_SREGS_E_FSL_PIDn) {
1146             env->spr[SPR_BOOKE_PID1] = sregs.u.e.impl.fsl.pid1;
1147             env->spr[SPR_BOOKE_PID2] = sregs.u.e.impl.fsl.pid2;
1148         }
1149     }
1150 
1151     return 0;
1152 }
1153 
1154 static int kvmppc_get_books_sregs(PowerPCCPU *cpu)
1155 {
1156     CPUPPCState *env = &cpu->env;
1157     struct kvm_sregs sregs;
1158     int ret;
1159     int i;
1160 
1161     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_SREGS, &sregs);
1162     if (ret < 0) {
1163         return ret;
1164     }
1165 
1166     if (!cpu->vhyp) {
1167         ppc_store_sdr1(env, sregs.u.s.sdr1);
1168     }
1169 
1170     /* Sync SLB */
1171 #ifdef TARGET_PPC64
1172     /*
1173      * The packed SLB array we get from KVM_GET_SREGS only contains
1174      * information about valid entries. So we flush our internal copy
1175      * to get rid of stale ones, then put all valid SLB entries back
1176      * in.
1177      */
1178     memset(env->slb, 0, sizeof(env->slb));
1179     for (i = 0; i < ARRAY_SIZE(env->slb); i++) {
1180         target_ulong rb = sregs.u.s.ppc64.slb[i].slbe;
1181         target_ulong rs = sregs.u.s.ppc64.slb[i].slbv;
1182         /*
1183          * Only restore valid entries
1184          */
1185         if (rb & SLB_ESID_V) {
1186             ppc_store_slb(cpu, rb & 0xfff, rb & ~0xfffULL, rs);
1187         }
1188     }
1189 #endif
1190 
1191     /* Sync SRs */
1192     for (i = 0; i < 16; i++) {
1193         env->sr[i] = sregs.u.s.ppc32.sr[i];
1194     }
1195 
1196     /* Sync BATs */
1197     for (i = 0; i < 8; i++) {
1198         env->DBAT[0][i] = sregs.u.s.ppc32.dbat[i] & 0xffffffff;
1199         env->DBAT[1][i] = sregs.u.s.ppc32.dbat[i] >> 32;
1200         env->IBAT[0][i] = sregs.u.s.ppc32.ibat[i] & 0xffffffff;
1201         env->IBAT[1][i] = sregs.u.s.ppc32.ibat[i] >> 32;
1202     }
1203 
1204     return 0;
1205 }
1206 
1207 int kvm_arch_get_registers(CPUState *cs)
1208 {
1209     PowerPCCPU *cpu = POWERPC_CPU(cs);
1210     CPUPPCState *env = &cpu->env;
1211     struct kvm_regs regs;
1212     uint32_t cr;
1213     int i, ret;
1214 
1215     ret = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
1216     if (ret < 0) {
1217         return ret;
1218     }
1219 
1220     cr = regs.cr;
1221     for (i = 7; i >= 0; i--) {
1222         env->crf[i] = cr & 15;
1223         cr >>= 4;
1224     }
1225 
1226     env->ctr = regs.ctr;
1227     env->lr = regs.lr;
1228     cpu_write_xer(env, regs.xer);
1229     env->msr = regs.msr;
1230     env->nip = regs.pc;
1231 
1232     env->spr[SPR_SRR0] = regs.srr0;
1233     env->spr[SPR_SRR1] = regs.srr1;
1234 
1235     env->spr[SPR_SPRG0] = regs.sprg0;
1236     env->spr[SPR_SPRG1] = regs.sprg1;
1237     env->spr[SPR_SPRG2] = regs.sprg2;
1238     env->spr[SPR_SPRG3] = regs.sprg3;
1239     env->spr[SPR_SPRG4] = regs.sprg4;
1240     env->spr[SPR_SPRG5] = regs.sprg5;
1241     env->spr[SPR_SPRG6] = regs.sprg6;
1242     env->spr[SPR_SPRG7] = regs.sprg7;
1243 
1244     env->spr[SPR_BOOKE_PID] = regs.pid;
1245 
1246     for (i = 0; i < 32; i++) {
1247         env->gpr[i] = regs.gpr[i];
1248     }
1249 
1250     kvm_get_fp(cs);
1251 
1252     if (cap_booke_sregs) {
1253         ret = kvmppc_get_booke_sregs(cpu);
1254         if (ret < 0) {
1255             return ret;
1256         }
1257     }
1258 
1259     if (cap_segstate) {
1260         ret = kvmppc_get_books_sregs(cpu);
1261         if (ret < 0) {
1262             return ret;
1263         }
1264     }
1265 
1266     if (cap_hior) {
1267         kvm_get_one_spr(cs, KVM_REG_PPC_HIOR, SPR_HIOR);
1268     }
1269 
1270     if (cap_one_reg) {
1271         int i;
1272 
1273         /*
1274          * We deliberately ignore errors here, for kernels which have
1275          * the ONE_REG calls, but don't support the specific
1276          * registers, there's a reasonable chance things will still
1277          * work, at least until we try to migrate.
1278          */
1279         for (i = 0; i < 1024; i++) {
1280             uint64_t id = env->spr_cb[i].one_reg_id;
1281 
1282             if (id != 0) {
1283                 kvm_get_one_spr(cs, id, i);
1284             }
1285         }
1286 
1287 #ifdef TARGET_PPC64
1288         if (msr_ts) {
1289             for (i = 0; i < ARRAY_SIZE(env->tm_gpr); i++) {
1290                 kvm_get_one_reg(cs, KVM_REG_PPC_TM_GPR(i), &env->tm_gpr[i]);
1291             }
1292             for (i = 0; i < ARRAY_SIZE(env->tm_vsr); i++) {
1293                 kvm_get_one_reg(cs, KVM_REG_PPC_TM_VSR(i), &env->tm_vsr[i]);
1294             }
1295             kvm_get_one_reg(cs, KVM_REG_PPC_TM_CR, &env->tm_cr);
1296             kvm_get_one_reg(cs, KVM_REG_PPC_TM_LR, &env->tm_lr);
1297             kvm_get_one_reg(cs, KVM_REG_PPC_TM_CTR, &env->tm_ctr);
1298             kvm_get_one_reg(cs, KVM_REG_PPC_TM_FPSCR, &env->tm_fpscr);
1299             kvm_get_one_reg(cs, KVM_REG_PPC_TM_AMR, &env->tm_amr);
1300             kvm_get_one_reg(cs, KVM_REG_PPC_TM_PPR, &env->tm_ppr);
1301             kvm_get_one_reg(cs, KVM_REG_PPC_TM_VRSAVE, &env->tm_vrsave);
1302             kvm_get_one_reg(cs, KVM_REG_PPC_TM_VSCR, &env->tm_vscr);
1303             kvm_get_one_reg(cs, KVM_REG_PPC_TM_DSCR, &env->tm_dscr);
1304             kvm_get_one_reg(cs, KVM_REG_PPC_TM_TAR, &env->tm_tar);
1305         }
1306 
1307         if (cap_papr) {
1308             if (kvm_get_vpa(cs) < 0) {
1309                 trace_kvm_failed_get_vpa();
1310             }
1311         }
1312 
1313         kvm_get_one_reg(cs, KVM_REG_PPC_TB_OFFSET, &env->tb_env->tb_offset);
1314 #endif
1315     }
1316 
1317     return 0;
1318 }
1319 
1320 int kvmppc_set_interrupt(PowerPCCPU *cpu, int irq, int level)
1321 {
1322     unsigned virq = level ? KVM_INTERRUPT_SET_LEVEL : KVM_INTERRUPT_UNSET;
1323 
1324     if (irq != PPC_INTERRUPT_EXT) {
1325         return 0;
1326     }
1327 
1328     if (!kvm_enabled() || !cap_interrupt_unset || !cap_interrupt_level) {
1329         return 0;
1330     }
1331 
1332     kvm_vcpu_ioctl(CPU(cpu), KVM_INTERRUPT, &virq);
1333 
1334     return 0;
1335 }
1336 
1337 #if defined(TARGET_PPC64)
1338 #define PPC_INPUT_INT PPC970_INPUT_INT
1339 #else
1340 #define PPC_INPUT_INT PPC6xx_INPUT_INT
1341 #endif
1342 
1343 void kvm_arch_pre_run(CPUState *cs, struct kvm_run *run)
1344 {
1345     PowerPCCPU *cpu = POWERPC_CPU(cs);
1346     CPUPPCState *env = &cpu->env;
1347     int r;
1348     unsigned irq;
1349 
1350     qemu_mutex_lock_iothread();
1351 
1352     /*
1353      * PowerPC QEMU tracks the various core input pins (interrupt,
1354      * critical interrupt, reset, etc) in PPC-specific
1355      * env->irq_input_state.
1356      */
1357     if (!cap_interrupt_level &&
1358         run->ready_for_interrupt_injection &&
1359         (cs->interrupt_request & CPU_INTERRUPT_HARD) &&
1360         (env->irq_input_state & (1 << PPC_INPUT_INT)))
1361     {
1362         /*
1363          * For now KVM disregards the 'irq' argument. However, in the
1364          * future KVM could cache it in-kernel to avoid a heavyweight
1365          * exit when reading the UIC.
1366          */
1367         irq = KVM_INTERRUPT_SET;
1368 
1369         trace_kvm_injected_interrupt(irq);
1370         r = kvm_vcpu_ioctl(cs, KVM_INTERRUPT, &irq);
1371         if (r < 0) {
1372             printf("cpu %d fail inject %x\n", cs->cpu_index, irq);
1373         }
1374 
1375         /* Always wake up soon in case the interrupt was level based */
1376         timer_mod(idle_timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) +
1377                        (NANOSECONDS_PER_SECOND / 50));
1378     }
1379 
1380     /*
1381      * We don't know if there are more interrupts pending after
1382      * this. However, the guest will return to userspace in the course
1383      * of handling this one anyways, so we will get a chance to
1384      * deliver the rest.
1385      */
1386 
1387     qemu_mutex_unlock_iothread();
1388 }
1389 
1390 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
1391 {
1392     return MEMTXATTRS_UNSPECIFIED;
1393 }
1394 
1395 int kvm_arch_process_async_events(CPUState *cs)
1396 {
1397     return cs->halted;
1398 }
1399 
1400 static int kvmppc_handle_halt(PowerPCCPU *cpu)
1401 {
1402     CPUState *cs = CPU(cpu);
1403     CPUPPCState *env = &cpu->env;
1404 
1405     if (!(cs->interrupt_request & CPU_INTERRUPT_HARD) && (msr_ee)) {
1406         cs->halted = 1;
1407         cs->exception_index = EXCP_HLT;
1408     }
1409 
1410     return 0;
1411 }
1412 
1413 /* map dcr access to existing qemu dcr emulation */
1414 static int kvmppc_handle_dcr_read(CPUPPCState *env,
1415                                   uint32_t dcrn, uint32_t *data)
1416 {
1417     if (ppc_dcr_read(env->dcr_env, dcrn, data) < 0) {
1418         fprintf(stderr, "Read to unhandled DCR (0x%x)\n", dcrn);
1419     }
1420 
1421     return 0;
1422 }
1423 
1424 static int kvmppc_handle_dcr_write(CPUPPCState *env,
1425                                    uint32_t dcrn, uint32_t data)
1426 {
1427     if (ppc_dcr_write(env->dcr_env, dcrn, data) < 0) {
1428         fprintf(stderr, "Write to unhandled DCR (0x%x)\n", dcrn);
1429     }
1430 
1431     return 0;
1432 }
1433 
1434 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
1435 {
1436     /* Mixed endian case is not handled */
1437     uint32_t sc = debug_inst_opcode;
1438 
1439     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
1440                             sizeof(sc), 0) ||
1441         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&sc, sizeof(sc), 1)) {
1442         return -EINVAL;
1443     }
1444 
1445     return 0;
1446 }
1447 
1448 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
1449 {
1450     uint32_t sc;
1451 
1452     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&sc, sizeof(sc), 0) ||
1453         sc != debug_inst_opcode ||
1454         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
1455                             sizeof(sc), 1)) {
1456         return -EINVAL;
1457     }
1458 
1459     return 0;
1460 }
1461 
1462 static int find_hw_breakpoint(target_ulong addr, int type)
1463 {
1464     int n;
1465 
1466     assert((nb_hw_breakpoint + nb_hw_watchpoint)
1467            <= ARRAY_SIZE(hw_debug_points));
1468 
1469     for (n = 0; n < nb_hw_breakpoint + nb_hw_watchpoint; n++) {
1470         if (hw_debug_points[n].addr == addr &&
1471              hw_debug_points[n].type == type) {
1472             return n;
1473         }
1474     }
1475 
1476     return -1;
1477 }
1478 
1479 static int find_hw_watchpoint(target_ulong addr, int *flag)
1480 {
1481     int n;
1482 
1483     n = find_hw_breakpoint(addr, GDB_WATCHPOINT_ACCESS);
1484     if (n >= 0) {
1485         *flag = BP_MEM_ACCESS;
1486         return n;
1487     }
1488 
1489     n = find_hw_breakpoint(addr, GDB_WATCHPOINT_WRITE);
1490     if (n >= 0) {
1491         *flag = BP_MEM_WRITE;
1492         return n;
1493     }
1494 
1495     n = find_hw_breakpoint(addr, GDB_WATCHPOINT_READ);
1496     if (n >= 0) {
1497         *flag = BP_MEM_READ;
1498         return n;
1499     }
1500 
1501     return -1;
1502 }
1503 
1504 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
1505                                   target_ulong len, int type)
1506 {
1507     if ((nb_hw_breakpoint + nb_hw_watchpoint) >= ARRAY_SIZE(hw_debug_points)) {
1508         return -ENOBUFS;
1509     }
1510 
1511     hw_debug_points[nb_hw_breakpoint + nb_hw_watchpoint].addr = addr;
1512     hw_debug_points[nb_hw_breakpoint + nb_hw_watchpoint].type = type;
1513 
1514     switch (type) {
1515     case GDB_BREAKPOINT_HW:
1516         if (nb_hw_breakpoint >= max_hw_breakpoint) {
1517             return -ENOBUFS;
1518         }
1519 
1520         if (find_hw_breakpoint(addr, type) >= 0) {
1521             return -EEXIST;
1522         }
1523 
1524         nb_hw_breakpoint++;
1525         break;
1526 
1527     case GDB_WATCHPOINT_WRITE:
1528     case GDB_WATCHPOINT_READ:
1529     case GDB_WATCHPOINT_ACCESS:
1530         if (nb_hw_watchpoint >= max_hw_watchpoint) {
1531             return -ENOBUFS;
1532         }
1533 
1534         if (find_hw_breakpoint(addr, type) >= 0) {
1535             return -EEXIST;
1536         }
1537 
1538         nb_hw_watchpoint++;
1539         break;
1540 
1541     default:
1542         return -ENOSYS;
1543     }
1544 
1545     return 0;
1546 }
1547 
1548 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
1549                                   target_ulong len, int type)
1550 {
1551     int n;
1552 
1553     n = find_hw_breakpoint(addr, type);
1554     if (n < 0) {
1555         return -ENOENT;
1556     }
1557 
1558     switch (type) {
1559     case GDB_BREAKPOINT_HW:
1560         nb_hw_breakpoint--;
1561         break;
1562 
1563     case GDB_WATCHPOINT_WRITE:
1564     case GDB_WATCHPOINT_READ:
1565     case GDB_WATCHPOINT_ACCESS:
1566         nb_hw_watchpoint--;
1567         break;
1568 
1569     default:
1570         return -ENOSYS;
1571     }
1572     hw_debug_points[n] = hw_debug_points[nb_hw_breakpoint + nb_hw_watchpoint];
1573 
1574     return 0;
1575 }
1576 
1577 void kvm_arch_remove_all_hw_breakpoints(void)
1578 {
1579     nb_hw_breakpoint = nb_hw_watchpoint = 0;
1580 }
1581 
1582 void kvm_arch_update_guest_debug(CPUState *cs, struct kvm_guest_debug *dbg)
1583 {
1584     int n;
1585 
1586     /* Software Breakpoint updates */
1587     if (kvm_sw_breakpoints_active(cs)) {
1588         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP;
1589     }
1590 
1591     assert((nb_hw_breakpoint + nb_hw_watchpoint)
1592            <= ARRAY_SIZE(hw_debug_points));
1593     assert((nb_hw_breakpoint + nb_hw_watchpoint) <= ARRAY_SIZE(dbg->arch.bp));
1594 
1595     if (nb_hw_breakpoint + nb_hw_watchpoint > 0) {
1596         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
1597         memset(dbg->arch.bp, 0, sizeof(dbg->arch.bp));
1598         for (n = 0; n < nb_hw_breakpoint + nb_hw_watchpoint; n++) {
1599             switch (hw_debug_points[n].type) {
1600             case GDB_BREAKPOINT_HW:
1601                 dbg->arch.bp[n].type = KVMPPC_DEBUG_BREAKPOINT;
1602                 break;
1603             case GDB_WATCHPOINT_WRITE:
1604                 dbg->arch.bp[n].type = KVMPPC_DEBUG_WATCH_WRITE;
1605                 break;
1606             case GDB_WATCHPOINT_READ:
1607                 dbg->arch.bp[n].type = KVMPPC_DEBUG_WATCH_READ;
1608                 break;
1609             case GDB_WATCHPOINT_ACCESS:
1610                 dbg->arch.bp[n].type = KVMPPC_DEBUG_WATCH_WRITE |
1611                                         KVMPPC_DEBUG_WATCH_READ;
1612                 break;
1613             default:
1614                 cpu_abort(cs, "Unsupported breakpoint type\n");
1615             }
1616             dbg->arch.bp[n].addr = hw_debug_points[n].addr;
1617         }
1618     }
1619 }
1620 
1621 static int kvm_handle_hw_breakpoint(CPUState *cs,
1622                                     struct kvm_debug_exit_arch *arch_info)
1623 {
1624     int handle = 0;
1625     int n;
1626     int flag = 0;
1627 
1628     if (nb_hw_breakpoint + nb_hw_watchpoint > 0) {
1629         if (arch_info->status & KVMPPC_DEBUG_BREAKPOINT) {
1630             n = find_hw_breakpoint(arch_info->address, GDB_BREAKPOINT_HW);
1631             if (n >= 0) {
1632                 handle = 1;
1633             }
1634         } else if (arch_info->status & (KVMPPC_DEBUG_WATCH_READ |
1635                                         KVMPPC_DEBUG_WATCH_WRITE)) {
1636             n = find_hw_watchpoint(arch_info->address,  &flag);
1637             if (n >= 0) {
1638                 handle = 1;
1639                 cs->watchpoint_hit = &hw_watchpoint;
1640                 hw_watchpoint.vaddr = hw_debug_points[n].addr;
1641                 hw_watchpoint.flags = flag;
1642             }
1643         }
1644     }
1645     return handle;
1646 }
1647 
1648 static int kvm_handle_singlestep(void)
1649 {
1650     return 1;
1651 }
1652 
1653 static int kvm_handle_sw_breakpoint(void)
1654 {
1655     return 1;
1656 }
1657 
1658 static int kvm_handle_debug(PowerPCCPU *cpu, struct kvm_run *run)
1659 {
1660     CPUState *cs = CPU(cpu);
1661     CPUPPCState *env = &cpu->env;
1662     struct kvm_debug_exit_arch *arch_info = &run->debug.arch;
1663 
1664     if (cs->singlestep_enabled) {
1665         return kvm_handle_singlestep();
1666     }
1667 
1668     if (arch_info->status) {
1669         return kvm_handle_hw_breakpoint(cs, arch_info);
1670     }
1671 
1672     if (kvm_find_sw_breakpoint(cs, arch_info->address)) {
1673         return kvm_handle_sw_breakpoint();
1674     }
1675 
1676     /*
1677      * QEMU is not able to handle debug exception, so inject
1678      * program exception to guest;
1679      * Yes program exception NOT debug exception !!
1680      * When QEMU is using debug resources then debug exception must
1681      * be always set. To achieve this we set MSR_DE and also set
1682      * MSRP_DEP so guest cannot change MSR_DE.
1683      * When emulating debug resource for guest we want guest
1684      * to control MSR_DE (enable/disable debug interrupt on need).
1685      * Supporting both configurations are NOT possible.
1686      * So the result is that we cannot share debug resources
1687      * between QEMU and Guest on BOOKE architecture.
1688      * In the current design QEMU gets the priority over guest,
1689      * this means that if QEMU is using debug resources then guest
1690      * cannot use them;
1691      * For software breakpoint QEMU uses a privileged instruction;
1692      * So there cannot be any reason that we are here for guest
1693      * set debug exception, only possibility is guest executed a
1694      * privileged / illegal instruction and that's why we are
1695      * injecting a program interrupt.
1696      */
1697     cpu_synchronize_state(cs);
1698     /*
1699      * env->nip is PC, so increment this by 4 to use
1700      * ppc_cpu_do_interrupt(), which set srr0 = env->nip - 4.
1701      */
1702     env->nip += 4;
1703     cs->exception_index = POWERPC_EXCP_PROGRAM;
1704     env->error_code = POWERPC_EXCP_INVAL;
1705     ppc_cpu_do_interrupt(cs);
1706 
1707     return 0;
1708 }
1709 
1710 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1711 {
1712     PowerPCCPU *cpu = POWERPC_CPU(cs);
1713     CPUPPCState *env = &cpu->env;
1714     int ret;
1715 
1716     qemu_mutex_lock_iothread();
1717 
1718     switch (run->exit_reason) {
1719     case KVM_EXIT_DCR:
1720         if (run->dcr.is_write) {
1721             trace_kvm_handle_dcr_write();
1722             ret = kvmppc_handle_dcr_write(env, run->dcr.dcrn, run->dcr.data);
1723         } else {
1724             trace_kvm_handle_drc_read();
1725             ret = kvmppc_handle_dcr_read(env, run->dcr.dcrn, &run->dcr.data);
1726         }
1727         break;
1728     case KVM_EXIT_HLT:
1729         trace_kvm_handle_halt();
1730         ret = kvmppc_handle_halt(cpu);
1731         break;
1732 #if defined(TARGET_PPC64)
1733     case KVM_EXIT_PAPR_HCALL:
1734         trace_kvm_handle_papr_hcall();
1735         run->papr_hcall.ret = spapr_hypercall(cpu,
1736                                               run->papr_hcall.nr,
1737                                               run->papr_hcall.args);
1738         ret = 0;
1739         break;
1740 #endif
1741     case KVM_EXIT_EPR:
1742         trace_kvm_handle_epr();
1743         run->epr.epr = ldl_phys(cs->as, env->mpic_iack);
1744         ret = 0;
1745         break;
1746     case KVM_EXIT_WATCHDOG:
1747         trace_kvm_handle_watchdog_expiry();
1748         watchdog_perform_action();
1749         ret = 0;
1750         break;
1751 
1752     case KVM_EXIT_DEBUG:
1753         trace_kvm_handle_debug_exception();
1754         if (kvm_handle_debug(cpu, run)) {
1755             ret = EXCP_DEBUG;
1756             break;
1757         }
1758         /* re-enter, this exception was guest-internal */
1759         ret = 0;
1760         break;
1761 
1762     default:
1763         fprintf(stderr, "KVM: unknown exit reason %d\n", run->exit_reason);
1764         ret = -1;
1765         break;
1766     }
1767 
1768     qemu_mutex_unlock_iothread();
1769     return ret;
1770 }
1771 
1772 int kvmppc_or_tsr_bits(PowerPCCPU *cpu, uint32_t tsr_bits)
1773 {
1774     CPUState *cs = CPU(cpu);
1775     uint32_t bits = tsr_bits;
1776     struct kvm_one_reg reg = {
1777         .id = KVM_REG_PPC_OR_TSR,
1778         .addr = (uintptr_t) &bits,
1779     };
1780 
1781     return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
1782 }
1783 
1784 int kvmppc_clear_tsr_bits(PowerPCCPU *cpu, uint32_t tsr_bits)
1785 {
1786 
1787     CPUState *cs = CPU(cpu);
1788     uint32_t bits = tsr_bits;
1789     struct kvm_one_reg reg = {
1790         .id = KVM_REG_PPC_CLEAR_TSR,
1791         .addr = (uintptr_t) &bits,
1792     };
1793 
1794     return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
1795 }
1796 
1797 int kvmppc_set_tcr(PowerPCCPU *cpu)
1798 {
1799     CPUState *cs = CPU(cpu);
1800     CPUPPCState *env = &cpu->env;
1801     uint32_t tcr = env->spr[SPR_BOOKE_TCR];
1802 
1803     struct kvm_one_reg reg = {
1804         .id = KVM_REG_PPC_TCR,
1805         .addr = (uintptr_t) &tcr,
1806     };
1807 
1808     return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
1809 }
1810 
1811 int kvmppc_booke_watchdog_enable(PowerPCCPU *cpu)
1812 {
1813     CPUState *cs = CPU(cpu);
1814     int ret;
1815 
1816     if (!kvm_enabled()) {
1817         return -1;
1818     }
1819 
1820     if (!cap_ppc_watchdog) {
1821         printf("warning: KVM does not support watchdog");
1822         return -1;
1823     }
1824 
1825     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_PPC_BOOKE_WATCHDOG, 0);
1826     if (ret < 0) {
1827         fprintf(stderr, "%s: couldn't enable KVM_CAP_PPC_BOOKE_WATCHDOG: %s\n",
1828                 __func__, strerror(-ret));
1829         return ret;
1830     }
1831 
1832     return ret;
1833 }
1834 
1835 static int read_cpuinfo(const char *field, char *value, int len)
1836 {
1837     FILE *f;
1838     int ret = -1;
1839     int field_len = strlen(field);
1840     char line[512];
1841 
1842     f = fopen("/proc/cpuinfo", "r");
1843     if (!f) {
1844         return -1;
1845     }
1846 
1847     do {
1848         if (!fgets(line, sizeof(line), f)) {
1849             break;
1850         }
1851         if (!strncmp(line, field, field_len)) {
1852             pstrcpy(value, len, line);
1853             ret = 0;
1854             break;
1855         }
1856     } while (*line);
1857 
1858     fclose(f);
1859 
1860     return ret;
1861 }
1862 
1863 uint32_t kvmppc_get_tbfreq(void)
1864 {
1865     char line[512];
1866     char *ns;
1867     uint32_t retval = NANOSECONDS_PER_SECOND;
1868 
1869     if (read_cpuinfo("timebase", line, sizeof(line))) {
1870         return retval;
1871     }
1872 
1873     ns = strchr(line, ':');
1874     if (!ns) {
1875         return retval;
1876     }
1877 
1878     ns++;
1879 
1880     return atoi(ns);
1881 }
1882 
1883 bool kvmppc_get_host_serial(char **value)
1884 {
1885     return g_file_get_contents("/proc/device-tree/system-id", value, NULL,
1886                                NULL);
1887 }
1888 
1889 bool kvmppc_get_host_model(char **value)
1890 {
1891     return g_file_get_contents("/proc/device-tree/model", value, NULL, NULL);
1892 }
1893 
1894 /* Try to find a device tree node for a CPU with clock-frequency property */
1895 static int kvmppc_find_cpu_dt(char *buf, int buf_len)
1896 {
1897     struct dirent *dirp;
1898     DIR *dp;
1899 
1900     dp = opendir(PROC_DEVTREE_CPU);
1901     if (!dp) {
1902         printf("Can't open directory " PROC_DEVTREE_CPU "\n");
1903         return -1;
1904     }
1905 
1906     buf[0] = '\0';
1907     while ((dirp = readdir(dp)) != NULL) {
1908         FILE *f;
1909         snprintf(buf, buf_len, "%s%s/clock-frequency", PROC_DEVTREE_CPU,
1910                  dirp->d_name);
1911         f = fopen(buf, "r");
1912         if (f) {
1913             snprintf(buf, buf_len, "%s%s", PROC_DEVTREE_CPU, dirp->d_name);
1914             fclose(f);
1915             break;
1916         }
1917         buf[0] = '\0';
1918     }
1919     closedir(dp);
1920     if (buf[0] == '\0') {
1921         printf("Unknown host!\n");
1922         return -1;
1923     }
1924 
1925     return 0;
1926 }
1927 
1928 static uint64_t kvmppc_read_int_dt(const char *filename)
1929 {
1930     union {
1931         uint32_t v32;
1932         uint64_t v64;
1933     } u;
1934     FILE *f;
1935     int len;
1936 
1937     f = fopen(filename, "rb");
1938     if (!f) {
1939         return -1;
1940     }
1941 
1942     len = fread(&u, 1, sizeof(u), f);
1943     fclose(f);
1944     switch (len) {
1945     case 4:
1946         /* property is a 32-bit quantity */
1947         return be32_to_cpu(u.v32);
1948     case 8:
1949         return be64_to_cpu(u.v64);
1950     }
1951 
1952     return 0;
1953 }
1954 
1955 /*
1956  * Read a CPU node property from the host device tree that's a single
1957  * integer (32-bit or 64-bit).  Returns 0 if anything goes wrong
1958  * (can't find or open the property, or doesn't understand the format)
1959  */
1960 static uint64_t kvmppc_read_int_cpu_dt(const char *propname)
1961 {
1962     char buf[PATH_MAX], *tmp;
1963     uint64_t val;
1964 
1965     if (kvmppc_find_cpu_dt(buf, sizeof(buf))) {
1966         return -1;
1967     }
1968 
1969     tmp = g_strdup_printf("%s/%s", buf, propname);
1970     val = kvmppc_read_int_dt(tmp);
1971     g_free(tmp);
1972 
1973     return val;
1974 }
1975 
1976 uint64_t kvmppc_get_clockfreq(void)
1977 {
1978     return kvmppc_read_int_cpu_dt("clock-frequency");
1979 }
1980 
1981 static int kvmppc_get_dec_bits(void)
1982 {
1983     int nr_bits = kvmppc_read_int_cpu_dt("ibm,dec-bits");
1984 
1985     if (nr_bits > 0) {
1986         return nr_bits;
1987     }
1988     return 0;
1989 }
1990 
1991 static int kvmppc_get_pvinfo(CPUPPCState *env, struct kvm_ppc_pvinfo *pvinfo)
1992  {
1993      PowerPCCPU *cpu = ppc_env_get_cpu(env);
1994      CPUState *cs = CPU(cpu);
1995 
1996     if (kvm_vm_check_extension(cs->kvm_state, KVM_CAP_PPC_GET_PVINFO) &&
1997         !kvm_vm_ioctl(cs->kvm_state, KVM_PPC_GET_PVINFO, pvinfo)) {
1998         return 0;
1999     }
2000 
2001     return 1;
2002 }
2003 
2004 int kvmppc_get_hasidle(CPUPPCState *env)
2005 {
2006     struct kvm_ppc_pvinfo pvinfo;
2007 
2008     if (!kvmppc_get_pvinfo(env, &pvinfo) &&
2009         (pvinfo.flags & KVM_PPC_PVINFO_FLAGS_EV_IDLE)) {
2010         return 1;
2011     }
2012 
2013     return 0;
2014 }
2015 
2016 int kvmppc_get_hypercall(CPUPPCState *env, uint8_t *buf, int buf_len)
2017 {
2018     uint32_t *hc = (uint32_t *)buf;
2019     struct kvm_ppc_pvinfo pvinfo;
2020 
2021     if (!kvmppc_get_pvinfo(env, &pvinfo)) {
2022         memcpy(buf, pvinfo.hcall, buf_len);
2023         return 0;
2024     }
2025 
2026     /*
2027      * Fallback to always fail hypercalls regardless of endianness:
2028      *
2029      *     tdi 0,r0,72 (becomes b .+8 in wrong endian, nop in good endian)
2030      *     li r3, -1
2031      *     b .+8       (becomes nop in wrong endian)
2032      *     bswap32(li r3, -1)
2033      */
2034 
2035     hc[0] = cpu_to_be32(0x08000048);
2036     hc[1] = cpu_to_be32(0x3860ffff);
2037     hc[2] = cpu_to_be32(0x48000008);
2038     hc[3] = cpu_to_be32(bswap32(0x3860ffff));
2039 
2040     return 1;
2041 }
2042 
2043 static inline int kvmppc_enable_hcall(KVMState *s, target_ulong hcall)
2044 {
2045     return kvm_vm_enable_cap(s, KVM_CAP_PPC_ENABLE_HCALL, 0, hcall, 1);
2046 }
2047 
2048 void kvmppc_enable_logical_ci_hcalls(void)
2049 {
2050     /*
2051      * FIXME: it would be nice if we could detect the cases where
2052      * we're using a device which requires the in kernel
2053      * implementation of these hcalls, but the kernel lacks them and
2054      * produce a warning.
2055      */
2056     kvmppc_enable_hcall(kvm_state, H_LOGICAL_CI_LOAD);
2057     kvmppc_enable_hcall(kvm_state, H_LOGICAL_CI_STORE);
2058 }
2059 
2060 void kvmppc_enable_set_mode_hcall(void)
2061 {
2062     kvmppc_enable_hcall(kvm_state, H_SET_MODE);
2063 }
2064 
2065 void kvmppc_enable_clear_ref_mod_hcalls(void)
2066 {
2067     kvmppc_enable_hcall(kvm_state, H_CLEAR_REF);
2068     kvmppc_enable_hcall(kvm_state, H_CLEAR_MOD);
2069 }
2070 
2071 void kvmppc_enable_h_page_init(void)
2072 {
2073     kvmppc_enable_hcall(kvm_state, H_PAGE_INIT);
2074 }
2075 
2076 void kvmppc_set_papr(PowerPCCPU *cpu)
2077 {
2078     CPUState *cs = CPU(cpu);
2079     int ret;
2080 
2081     if (!kvm_enabled()) {
2082         return;
2083     }
2084 
2085     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_PPC_PAPR, 0);
2086     if (ret) {
2087         error_report("This vCPU type or KVM version does not support PAPR");
2088         exit(1);
2089     }
2090 
2091     /*
2092      * Update the capability flag so we sync the right information
2093      * with kvm
2094      */
2095     cap_papr = 1;
2096 }
2097 
2098 int kvmppc_set_compat(PowerPCCPU *cpu, uint32_t compat_pvr)
2099 {
2100     return kvm_set_one_reg(CPU(cpu), KVM_REG_PPC_ARCH_COMPAT, &compat_pvr);
2101 }
2102 
2103 void kvmppc_set_mpic_proxy(PowerPCCPU *cpu, int mpic_proxy)
2104 {
2105     CPUState *cs = CPU(cpu);
2106     int ret;
2107 
2108     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_PPC_EPR, 0, mpic_proxy);
2109     if (ret && mpic_proxy) {
2110         error_report("This KVM version does not support EPR");
2111         exit(1);
2112     }
2113 }
2114 
2115 int kvmppc_smt_threads(void)
2116 {
2117     return cap_ppc_smt ? cap_ppc_smt : 1;
2118 }
2119 
2120 int kvmppc_set_smt_threads(int smt)
2121 {
2122     int ret;
2123 
2124     ret = kvm_vm_enable_cap(kvm_state, KVM_CAP_PPC_SMT, 0, smt, 0);
2125     if (!ret) {
2126         cap_ppc_smt = smt;
2127     }
2128     return ret;
2129 }
2130 
2131 void kvmppc_hint_smt_possible(Error **errp)
2132 {
2133     int i;
2134     GString *g;
2135     char *s;
2136 
2137     assert(kvm_enabled());
2138     if (cap_ppc_smt_possible) {
2139         g = g_string_new("Available VSMT modes:");
2140         for (i = 63; i >= 0; i--) {
2141             if ((1UL << i) & cap_ppc_smt_possible) {
2142                 g_string_append_printf(g, " %lu", (1UL << i));
2143             }
2144         }
2145         s = g_string_free(g, false);
2146         error_append_hint(errp, "%s.\n", s);
2147         g_free(s);
2148     } else {
2149         error_append_hint(errp,
2150                           "This KVM seems to be too old to support VSMT.\n");
2151     }
2152 }
2153 
2154 
2155 #ifdef TARGET_PPC64
2156 uint64_t kvmppc_rma_size(uint64_t current_size, unsigned int hash_shift)
2157 {
2158     struct kvm_ppc_smmu_info info;
2159     long rampagesize, best_page_shift;
2160     int i;
2161 
2162     /*
2163      * Find the largest hardware supported page size that's less than
2164      * or equal to the (logical) backing page size of guest RAM
2165      */
2166     kvm_get_smmu_info(&info, &error_fatal);
2167     rampagesize = qemu_minrampagesize();
2168     best_page_shift = 0;
2169 
2170     for (i = 0; i < KVM_PPC_PAGE_SIZES_MAX_SZ; i++) {
2171         struct kvm_ppc_one_seg_page_size *sps = &info.sps[i];
2172 
2173         if (!sps->page_shift) {
2174             continue;
2175         }
2176 
2177         if ((sps->page_shift > best_page_shift)
2178             && ((1UL << sps->page_shift) <= rampagesize)) {
2179             best_page_shift = sps->page_shift;
2180         }
2181     }
2182 
2183     return MIN(current_size,
2184                1ULL << (best_page_shift + hash_shift - 7));
2185 }
2186 #endif
2187 
2188 bool kvmppc_spapr_use_multitce(void)
2189 {
2190     return cap_spapr_multitce;
2191 }
2192 
2193 int kvmppc_spapr_enable_inkernel_multitce(void)
2194 {
2195     int ret;
2196 
2197     ret = kvm_vm_enable_cap(kvm_state, KVM_CAP_PPC_ENABLE_HCALL, 0,
2198                             H_PUT_TCE_INDIRECT, 1);
2199     if (!ret) {
2200         ret = kvm_vm_enable_cap(kvm_state, KVM_CAP_PPC_ENABLE_HCALL, 0,
2201                                 H_STUFF_TCE, 1);
2202     }
2203 
2204     return ret;
2205 }
2206 
2207 void *kvmppc_create_spapr_tce(uint32_t liobn, uint32_t page_shift,
2208                               uint64_t bus_offset, uint32_t nb_table,
2209                               int *pfd, bool need_vfio)
2210 {
2211     long len;
2212     int fd;
2213     void *table;
2214 
2215     /*
2216      * Must set fd to -1 so we don't try to munmap when called for
2217      * destroying the table, which the upper layers -will- do
2218      */
2219     *pfd = -1;
2220     if (!cap_spapr_tce || (need_vfio && !cap_spapr_vfio)) {
2221         return NULL;
2222     }
2223 
2224     if (cap_spapr_tce_64) {
2225         struct kvm_create_spapr_tce_64 args = {
2226             .liobn = liobn,
2227             .page_shift = page_shift,
2228             .offset = bus_offset >> page_shift,
2229             .size = nb_table,
2230             .flags = 0
2231         };
2232         fd = kvm_vm_ioctl(kvm_state, KVM_CREATE_SPAPR_TCE_64, &args);
2233         if (fd < 0) {
2234             fprintf(stderr,
2235                     "KVM: Failed to create TCE64 table for liobn 0x%x\n",
2236                     liobn);
2237             return NULL;
2238         }
2239     } else if (cap_spapr_tce) {
2240         uint64_t window_size = (uint64_t) nb_table << page_shift;
2241         struct kvm_create_spapr_tce args = {
2242             .liobn = liobn,
2243             .window_size = window_size,
2244         };
2245         if ((window_size != args.window_size) || bus_offset) {
2246             return NULL;
2247         }
2248         fd = kvm_vm_ioctl(kvm_state, KVM_CREATE_SPAPR_TCE, &args);
2249         if (fd < 0) {
2250             fprintf(stderr, "KVM: Failed to create TCE table for liobn 0x%x\n",
2251                     liobn);
2252             return NULL;
2253         }
2254     } else {
2255         return NULL;
2256     }
2257 
2258     len = nb_table * sizeof(uint64_t);
2259     /* FIXME: round this up to page size */
2260 
2261     table = mmap(NULL, len, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
2262     if (table == MAP_FAILED) {
2263         fprintf(stderr, "KVM: Failed to map TCE table for liobn 0x%x\n",
2264                 liobn);
2265         close(fd);
2266         return NULL;
2267     }
2268 
2269     *pfd = fd;
2270     return table;
2271 }
2272 
2273 int kvmppc_remove_spapr_tce(void *table, int fd, uint32_t nb_table)
2274 {
2275     long len;
2276 
2277     if (fd < 0) {
2278         return -1;
2279     }
2280 
2281     len = nb_table * sizeof(uint64_t);
2282     if ((munmap(table, len) < 0) ||
2283         (close(fd) < 0)) {
2284         fprintf(stderr, "KVM: Unexpected error removing TCE table: %s",
2285                 strerror(errno));
2286         /* Leak the table */
2287     }
2288 
2289     return 0;
2290 }
2291 
2292 int kvmppc_reset_htab(int shift_hint)
2293 {
2294     uint32_t shift = shift_hint;
2295 
2296     if (!kvm_enabled()) {
2297         /* Full emulation, tell caller to allocate htab itself */
2298         return 0;
2299     }
2300     if (kvm_vm_check_extension(kvm_state, KVM_CAP_PPC_ALLOC_HTAB)) {
2301         int ret;
2302         ret = kvm_vm_ioctl(kvm_state, KVM_PPC_ALLOCATE_HTAB, &shift);
2303         if (ret == -ENOTTY) {
2304             /*
2305              * At least some versions of PR KVM advertise the
2306              * capability, but don't implement the ioctl().  Oops.
2307              * Return 0 so that we allocate the htab in qemu, as is
2308              * correct for PR.
2309              */
2310             return 0;
2311         } else if (ret < 0) {
2312             return ret;
2313         }
2314         return shift;
2315     }
2316 
2317     /*
2318      * We have a kernel that predates the htab reset calls.  For PR
2319      * KVM, we need to allocate the htab ourselves, for an HV KVM of
2320      * this era, it has allocated a 16MB fixed size hash table
2321      * already.
2322      */
2323     if (kvmppc_is_pr(kvm_state)) {
2324         /* PR - tell caller to allocate htab */
2325         return 0;
2326     } else {
2327         /* HV - assume 16MB kernel allocated htab */
2328         return 24;
2329     }
2330 }
2331 
2332 static inline uint32_t mfpvr(void)
2333 {
2334     uint32_t pvr;
2335 
2336     asm ("mfpvr %0"
2337          : "=r"(pvr));
2338     return pvr;
2339 }
2340 
2341 static void alter_insns(uint64_t *word, uint64_t flags, bool on)
2342 {
2343     if (on) {
2344         *word |= flags;
2345     } else {
2346         *word &= ~flags;
2347     }
2348 }
2349 
2350 static void kvmppc_host_cpu_class_init(ObjectClass *oc, void *data)
2351 {
2352     PowerPCCPUClass *pcc = POWERPC_CPU_CLASS(oc);
2353     uint32_t dcache_size = kvmppc_read_int_cpu_dt("d-cache-size");
2354     uint32_t icache_size = kvmppc_read_int_cpu_dt("i-cache-size");
2355 
2356     /* Now fix up the class with information we can query from the host */
2357     pcc->pvr = mfpvr();
2358 
2359     alter_insns(&pcc->insns_flags, PPC_ALTIVEC,
2360                 qemu_getauxval(AT_HWCAP) & PPC_FEATURE_HAS_ALTIVEC);
2361     alter_insns(&pcc->insns_flags2, PPC2_VSX,
2362                 qemu_getauxval(AT_HWCAP) & PPC_FEATURE_HAS_VSX);
2363     alter_insns(&pcc->insns_flags2, PPC2_DFP,
2364                 qemu_getauxval(AT_HWCAP) & PPC_FEATURE_HAS_DFP);
2365 
2366     if (dcache_size != -1) {
2367         pcc->l1_dcache_size = dcache_size;
2368     }
2369 
2370     if (icache_size != -1) {
2371         pcc->l1_icache_size = icache_size;
2372     }
2373 
2374 #if defined(TARGET_PPC64)
2375     pcc->radix_page_info = kvm_get_radix_page_info();
2376 
2377     if ((pcc->pvr & 0xffffff00) == CPU_POWERPC_POWER9_DD1) {
2378         /*
2379          * POWER9 DD1 has some bugs which make it not really ISA 3.00
2380          * compliant.  More importantly, advertising ISA 3.00
2381          * architected mode may prevent guests from activating
2382          * necessary DD1 workarounds.
2383          */
2384         pcc->pcr_supported &= ~(PCR_COMPAT_3_00 | PCR_COMPAT_2_07
2385                                 | PCR_COMPAT_2_06 | PCR_COMPAT_2_05);
2386     }
2387 #endif /* defined(TARGET_PPC64) */
2388 }
2389 
2390 bool kvmppc_has_cap_epr(void)
2391 {
2392     return cap_epr;
2393 }
2394 
2395 bool kvmppc_has_cap_fixup_hcalls(void)
2396 {
2397     return cap_fixup_hcalls;
2398 }
2399 
2400 bool kvmppc_has_cap_htm(void)
2401 {
2402     return cap_htm;
2403 }
2404 
2405 bool kvmppc_has_cap_mmu_radix(void)
2406 {
2407     return cap_mmu_radix;
2408 }
2409 
2410 bool kvmppc_has_cap_mmu_hash_v3(void)
2411 {
2412     return cap_mmu_hash_v3;
2413 }
2414 
2415 static bool kvmppc_power8_host(void)
2416 {
2417     bool ret = false;
2418 #ifdef TARGET_PPC64
2419     {
2420         uint32_t base_pvr = CPU_POWERPC_POWER_SERVER_MASK & mfpvr();
2421         ret = (base_pvr == CPU_POWERPC_POWER8E_BASE) ||
2422               (base_pvr == CPU_POWERPC_POWER8NVL_BASE) ||
2423               (base_pvr == CPU_POWERPC_POWER8_BASE);
2424     }
2425 #endif /* TARGET_PPC64 */
2426     return ret;
2427 }
2428 
2429 static int parse_cap_ppc_safe_cache(struct kvm_ppc_cpu_char c)
2430 {
2431     bool l1d_thread_priv_req = !kvmppc_power8_host();
2432 
2433     if (~c.behaviour & c.behaviour_mask & H_CPU_BEHAV_L1D_FLUSH_PR) {
2434         return 2;
2435     } else if ((!l1d_thread_priv_req ||
2436                 c.character & c.character_mask & H_CPU_CHAR_L1D_THREAD_PRIV) &&
2437                (c.character & c.character_mask
2438                 & (H_CPU_CHAR_L1D_FLUSH_ORI30 | H_CPU_CHAR_L1D_FLUSH_TRIG2))) {
2439         return 1;
2440     }
2441 
2442     return 0;
2443 }
2444 
2445 static int parse_cap_ppc_safe_bounds_check(struct kvm_ppc_cpu_char c)
2446 {
2447     if (~c.behaviour & c.behaviour_mask & H_CPU_BEHAV_BNDS_CHK_SPEC_BAR) {
2448         return 2;
2449     } else if (c.character & c.character_mask & H_CPU_CHAR_SPEC_BAR_ORI31) {
2450         return 1;
2451     }
2452 
2453     return 0;
2454 }
2455 
2456 static int parse_cap_ppc_safe_indirect_branch(struct kvm_ppc_cpu_char c)
2457 {
2458     if ((~c.behaviour & c.behaviour_mask & H_CPU_BEHAV_FLUSH_COUNT_CACHE) &&
2459         (~c.character & c.character_mask & H_CPU_CHAR_CACHE_COUNT_DIS) &&
2460         (~c.character & c.character_mask & H_CPU_CHAR_BCCTRL_SERIALISED)) {
2461         return SPAPR_CAP_FIXED_NA;
2462     } else if (c.behaviour & c.behaviour_mask & H_CPU_BEHAV_FLUSH_COUNT_CACHE) {
2463         return SPAPR_CAP_WORKAROUND;
2464     } else if (c.character & c.character_mask & H_CPU_CHAR_CACHE_COUNT_DIS) {
2465         return  SPAPR_CAP_FIXED_CCD;
2466     } else if (c.character & c.character_mask & H_CPU_CHAR_BCCTRL_SERIALISED) {
2467         return SPAPR_CAP_FIXED_IBS;
2468     }
2469 
2470     return 0;
2471 }
2472 
2473 static int parse_cap_ppc_count_cache_flush_assist(struct kvm_ppc_cpu_char c)
2474 {
2475     if (c.character & c.character_mask & H_CPU_CHAR_BCCTR_FLUSH_ASSIST) {
2476         return 1;
2477     }
2478     return 0;
2479 }
2480 
2481 static void kvmppc_get_cpu_characteristics(KVMState *s)
2482 {
2483     struct kvm_ppc_cpu_char c;
2484     int ret;
2485 
2486     /* Assume broken */
2487     cap_ppc_safe_cache = 0;
2488     cap_ppc_safe_bounds_check = 0;
2489     cap_ppc_safe_indirect_branch = 0;
2490 
2491     ret = kvm_vm_check_extension(s, KVM_CAP_PPC_GET_CPU_CHAR);
2492     if (!ret) {
2493         return;
2494     }
2495     ret = kvm_vm_ioctl(s, KVM_PPC_GET_CPU_CHAR, &c);
2496     if (ret < 0) {
2497         return;
2498     }
2499 
2500     cap_ppc_safe_cache = parse_cap_ppc_safe_cache(c);
2501     cap_ppc_safe_bounds_check = parse_cap_ppc_safe_bounds_check(c);
2502     cap_ppc_safe_indirect_branch = parse_cap_ppc_safe_indirect_branch(c);
2503     cap_ppc_count_cache_flush_assist =
2504         parse_cap_ppc_count_cache_flush_assist(c);
2505 }
2506 
2507 int kvmppc_get_cap_safe_cache(void)
2508 {
2509     return cap_ppc_safe_cache;
2510 }
2511 
2512 int kvmppc_get_cap_safe_bounds_check(void)
2513 {
2514     return cap_ppc_safe_bounds_check;
2515 }
2516 
2517 int kvmppc_get_cap_safe_indirect_branch(void)
2518 {
2519     return cap_ppc_safe_indirect_branch;
2520 }
2521 
2522 int kvmppc_get_cap_count_cache_flush_assist(void)
2523 {
2524     return cap_ppc_count_cache_flush_assist;
2525 }
2526 
2527 bool kvmppc_has_cap_nested_kvm_hv(void)
2528 {
2529     return !!cap_ppc_nested_kvm_hv;
2530 }
2531 
2532 int kvmppc_set_cap_nested_kvm_hv(int enable)
2533 {
2534     return kvm_vm_enable_cap(kvm_state, KVM_CAP_PPC_NESTED_HV, 0, enable);
2535 }
2536 
2537 bool kvmppc_has_cap_spapr_vfio(void)
2538 {
2539     return cap_spapr_vfio;
2540 }
2541 
2542 int kvmppc_get_cap_large_decr(void)
2543 {
2544     return cap_large_decr;
2545 }
2546 
2547 int kvmppc_enable_cap_large_decr(PowerPCCPU *cpu, int enable)
2548 {
2549     CPUState *cs = CPU(cpu);
2550     uint64_t lpcr;
2551 
2552     kvm_get_one_reg(cs, KVM_REG_PPC_LPCR_64, &lpcr);
2553     /* Do we need to modify the LPCR? */
2554     if (!!(lpcr & LPCR_LD) != !!enable) {
2555         if (enable) {
2556             lpcr |= LPCR_LD;
2557         } else {
2558             lpcr &= ~LPCR_LD;
2559         }
2560         kvm_set_one_reg(cs, KVM_REG_PPC_LPCR_64, &lpcr);
2561         kvm_get_one_reg(cs, KVM_REG_PPC_LPCR_64, &lpcr);
2562 
2563         if (!!(lpcr & LPCR_LD) != !!enable) {
2564             return -1;
2565         }
2566     }
2567 
2568     return 0;
2569 }
2570 
2571 PowerPCCPUClass *kvm_ppc_get_host_cpu_class(void)
2572 {
2573     uint32_t host_pvr = mfpvr();
2574     PowerPCCPUClass *pvr_pcc;
2575 
2576     pvr_pcc = ppc_cpu_class_by_pvr(host_pvr);
2577     if (pvr_pcc == NULL) {
2578         pvr_pcc = ppc_cpu_class_by_pvr_mask(host_pvr);
2579     }
2580 
2581     return pvr_pcc;
2582 }
2583 
2584 static int kvm_ppc_register_host_cpu_type(MachineState *ms)
2585 {
2586     TypeInfo type_info = {
2587         .name = TYPE_HOST_POWERPC_CPU,
2588         .class_init = kvmppc_host_cpu_class_init,
2589     };
2590     MachineClass *mc = MACHINE_GET_CLASS(ms);
2591     PowerPCCPUClass *pvr_pcc;
2592     ObjectClass *oc;
2593     DeviceClass *dc;
2594     int i;
2595 
2596     pvr_pcc = kvm_ppc_get_host_cpu_class();
2597     if (pvr_pcc == NULL) {
2598         return -1;
2599     }
2600     type_info.parent = object_class_get_name(OBJECT_CLASS(pvr_pcc));
2601     type_register(&type_info);
2602     if (object_dynamic_cast(OBJECT(ms), TYPE_SPAPR_MACHINE)) {
2603         /* override TCG default cpu type with 'host' cpu model */
2604         mc->default_cpu_type = TYPE_HOST_POWERPC_CPU;
2605     }
2606 
2607     oc = object_class_by_name(type_info.name);
2608     g_assert(oc);
2609 
2610     /*
2611      * Update generic CPU family class alias (e.g. on a POWER8NVL host,
2612      * we want "POWER8" to be a "family" alias that points to the current
2613      * host CPU type, too)
2614      */
2615     dc = DEVICE_CLASS(ppc_cpu_get_family_class(pvr_pcc));
2616     for (i = 0; ppc_cpu_aliases[i].alias != NULL; i++) {
2617         if (strcasecmp(ppc_cpu_aliases[i].alias, dc->desc) == 0) {
2618             char *suffix;
2619 
2620             ppc_cpu_aliases[i].model = g_strdup(object_class_get_name(oc));
2621             suffix = strstr(ppc_cpu_aliases[i].model, POWERPC_CPU_TYPE_SUFFIX);
2622             if (suffix) {
2623                 *suffix = 0;
2624             }
2625             break;
2626         }
2627     }
2628 
2629     return 0;
2630 }
2631 
2632 int kvmppc_define_rtas_kernel_token(uint32_t token, const char *function)
2633 {
2634     struct kvm_rtas_token_args args = {
2635         .token = token,
2636     };
2637 
2638     if (!kvm_check_extension(kvm_state, KVM_CAP_PPC_RTAS)) {
2639         return -ENOENT;
2640     }
2641 
2642     strncpy(args.name, function, sizeof(args.name));
2643 
2644     return kvm_vm_ioctl(kvm_state, KVM_PPC_RTAS_DEFINE_TOKEN, &args);
2645 }
2646 
2647 int kvmppc_get_htab_fd(bool write, uint64_t index, Error **errp)
2648 {
2649     struct kvm_get_htab_fd s = {
2650         .flags = write ? KVM_GET_HTAB_WRITE : 0,
2651         .start_index = index,
2652     };
2653     int ret;
2654 
2655     if (!cap_htab_fd) {
2656         error_setg(errp, "KVM version doesn't support %s the HPT",
2657                    write ? "writing" : "reading");
2658         return -ENOTSUP;
2659     }
2660 
2661     ret = kvm_vm_ioctl(kvm_state, KVM_PPC_GET_HTAB_FD, &s);
2662     if (ret < 0) {
2663         error_setg(errp, "Unable to open fd for %s HPT %s KVM: %s",
2664                    write ? "writing" : "reading", write ? "to" : "from",
2665                    strerror(errno));
2666         return -errno;
2667     }
2668 
2669     return ret;
2670 }
2671 
2672 int kvmppc_save_htab(QEMUFile *f, int fd, size_t bufsize, int64_t max_ns)
2673 {
2674     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2675     uint8_t buf[bufsize];
2676     ssize_t rc;
2677 
2678     do {
2679         rc = read(fd, buf, bufsize);
2680         if (rc < 0) {
2681             fprintf(stderr, "Error reading data from KVM HTAB fd: %s\n",
2682                     strerror(errno));
2683             return rc;
2684         } else if (rc) {
2685             uint8_t *buffer = buf;
2686             ssize_t n = rc;
2687             while (n) {
2688                 struct kvm_get_htab_header *head =
2689                     (struct kvm_get_htab_header *) buffer;
2690                 size_t chunksize = sizeof(*head) +
2691                      HASH_PTE_SIZE_64 * head->n_valid;
2692 
2693                 qemu_put_be32(f, head->index);
2694                 qemu_put_be16(f, head->n_valid);
2695                 qemu_put_be16(f, head->n_invalid);
2696                 qemu_put_buffer(f, (void *)(head + 1),
2697                                 HASH_PTE_SIZE_64 * head->n_valid);
2698 
2699                 buffer += chunksize;
2700                 n -= chunksize;
2701             }
2702         }
2703     } while ((rc != 0)
2704              && ((max_ns < 0) ||
2705                  ((qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) < max_ns)));
2706 
2707     return (rc == 0) ? 1 : 0;
2708 }
2709 
2710 int kvmppc_load_htab_chunk(QEMUFile *f, int fd, uint32_t index,
2711                            uint16_t n_valid, uint16_t n_invalid)
2712 {
2713     struct kvm_get_htab_header *buf;
2714     size_t chunksize = sizeof(*buf) + n_valid * HASH_PTE_SIZE_64;
2715     ssize_t rc;
2716 
2717     buf = alloca(chunksize);
2718     buf->index = index;
2719     buf->n_valid = n_valid;
2720     buf->n_invalid = n_invalid;
2721 
2722     qemu_get_buffer(f, (void *)(buf + 1), HASH_PTE_SIZE_64 * n_valid);
2723 
2724     rc = write(fd, buf, chunksize);
2725     if (rc < 0) {
2726         fprintf(stderr, "Error writing KVM hash table: %s\n",
2727                 strerror(errno));
2728         return rc;
2729     }
2730     if (rc != chunksize) {
2731         /* We should never get a short write on a single chunk */
2732         fprintf(stderr, "Short write, restoring KVM hash table\n");
2733         return -1;
2734     }
2735     return 0;
2736 }
2737 
2738 bool kvm_arch_stop_on_emulation_error(CPUState *cpu)
2739 {
2740     return true;
2741 }
2742 
2743 void kvm_arch_init_irq_routing(KVMState *s)
2744 {
2745 }
2746 
2747 void kvmppc_read_hptes(ppc_hash_pte64_t *hptes, hwaddr ptex, int n)
2748 {
2749     int fd, rc;
2750     int i;
2751 
2752     fd = kvmppc_get_htab_fd(false, ptex, &error_abort);
2753 
2754     i = 0;
2755     while (i < n) {
2756         struct kvm_get_htab_header *hdr;
2757         int m = n < HPTES_PER_GROUP ? n : HPTES_PER_GROUP;
2758         char buf[sizeof(*hdr) + m * HASH_PTE_SIZE_64];
2759 
2760         rc = read(fd, buf, sizeof(buf));
2761         if (rc < 0) {
2762             hw_error("kvmppc_read_hptes: Unable to read HPTEs");
2763         }
2764 
2765         hdr = (struct kvm_get_htab_header *)buf;
2766         while ((i < n) && ((char *)hdr < (buf + rc))) {
2767             int invalid = hdr->n_invalid, valid = hdr->n_valid;
2768 
2769             if (hdr->index != (ptex + i)) {
2770                 hw_error("kvmppc_read_hptes: Unexpected HPTE index %"PRIu32
2771                          " != (%"HWADDR_PRIu" + %d", hdr->index, ptex, i);
2772             }
2773 
2774             if (n - i < valid) {
2775                 valid = n - i;
2776             }
2777             memcpy(hptes + i, hdr + 1, HASH_PTE_SIZE_64 * valid);
2778             i += valid;
2779 
2780             if ((n - i) < invalid) {
2781                 invalid = n - i;
2782             }
2783             memset(hptes + i, 0, invalid * HASH_PTE_SIZE_64);
2784             i += invalid;
2785 
2786             hdr = (struct kvm_get_htab_header *)
2787                 ((char *)(hdr + 1) + HASH_PTE_SIZE_64 * hdr->n_valid);
2788         }
2789     }
2790 
2791     close(fd);
2792 }
2793 
2794 void kvmppc_write_hpte(hwaddr ptex, uint64_t pte0, uint64_t pte1)
2795 {
2796     int fd, rc;
2797     struct {
2798         struct kvm_get_htab_header hdr;
2799         uint64_t pte0;
2800         uint64_t pte1;
2801     } buf;
2802 
2803     fd = kvmppc_get_htab_fd(true, 0 /* Ignored */, &error_abort);
2804 
2805     buf.hdr.n_valid = 1;
2806     buf.hdr.n_invalid = 0;
2807     buf.hdr.index = ptex;
2808     buf.pte0 = cpu_to_be64(pte0);
2809     buf.pte1 = cpu_to_be64(pte1);
2810 
2811     rc = write(fd, &buf, sizeof(buf));
2812     if (rc != sizeof(buf)) {
2813         hw_error("kvmppc_write_hpte: Unable to update KVM HPT");
2814     }
2815     close(fd);
2816 }
2817 
2818 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
2819                              uint64_t address, uint32_t data, PCIDevice *dev)
2820 {
2821     return 0;
2822 }
2823 
2824 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
2825                                 int vector, PCIDevice *dev)
2826 {
2827     return 0;
2828 }
2829 
2830 int kvm_arch_release_virq_post(int virq)
2831 {
2832     return 0;
2833 }
2834 
2835 int kvm_arch_msi_data_to_gsi(uint32_t data)
2836 {
2837     return data & 0xffff;
2838 }
2839 
2840 int kvmppc_enable_hwrng(void)
2841 {
2842     if (!kvm_enabled() || !kvm_check_extension(kvm_state, KVM_CAP_PPC_HWRNG)) {
2843         return -1;
2844     }
2845 
2846     return kvmppc_enable_hcall(kvm_state, H_RANDOM);
2847 }
2848 
2849 void kvmppc_check_papr_resize_hpt(Error **errp)
2850 {
2851     if (!kvm_enabled()) {
2852         return; /* No KVM, we're good */
2853     }
2854 
2855     if (cap_resize_hpt) {
2856         return; /* Kernel has explicit support, we're good */
2857     }
2858 
2859     /* Otherwise fallback on looking for PR KVM */
2860     if (kvmppc_is_pr(kvm_state)) {
2861         return;
2862     }
2863 
2864     error_setg(errp,
2865                "Hash page table resizing not available with this KVM version");
2866 }
2867 
2868 int kvmppc_resize_hpt_prepare(PowerPCCPU *cpu, target_ulong flags, int shift)
2869 {
2870     CPUState *cs = CPU(cpu);
2871     struct kvm_ppc_resize_hpt rhpt = {
2872         .flags = flags,
2873         .shift = shift,
2874     };
2875 
2876     if (!cap_resize_hpt) {
2877         return -ENOSYS;
2878     }
2879 
2880     return kvm_vm_ioctl(cs->kvm_state, KVM_PPC_RESIZE_HPT_PREPARE, &rhpt);
2881 }
2882 
2883 int kvmppc_resize_hpt_commit(PowerPCCPU *cpu, target_ulong flags, int shift)
2884 {
2885     CPUState *cs = CPU(cpu);
2886     struct kvm_ppc_resize_hpt rhpt = {
2887         .flags = flags,
2888         .shift = shift,
2889     };
2890 
2891     if (!cap_resize_hpt) {
2892         return -ENOSYS;
2893     }
2894 
2895     return kvm_vm_ioctl(cs->kvm_state, KVM_PPC_RESIZE_HPT_COMMIT, &rhpt);
2896 }
2897 
2898 /*
2899  * This is a helper function to detect a post migration scenario
2900  * in which a guest, running as KVM-HV, freezes in cpu_post_load because
2901  * the guest kernel can't handle a PVR value other than the actual host
2902  * PVR in KVM_SET_SREGS, even if pvr_match() returns true.
2903  *
2904  * If we don't have cap_ppc_pvr_compat and we're not running in PR
2905  * (so, we're HV), return true. The workaround itself is done in
2906  * cpu_post_load.
2907  *
2908  * The order here is important: we'll only check for KVM PR as a
2909  * fallback if the guest kernel can't handle the situation itself.
2910  * We need to avoid as much as possible querying the running KVM type
2911  * in QEMU level.
2912  */
2913 bool kvmppc_pvr_workaround_required(PowerPCCPU *cpu)
2914 {
2915     CPUState *cs = CPU(cpu);
2916 
2917     if (!kvm_enabled()) {
2918         return false;
2919     }
2920 
2921     if (cap_ppc_pvr_compat) {
2922         return false;
2923     }
2924 
2925     return !kvmppc_is_pr(cs->kvm_state);
2926 }
2927 
2928 void kvmppc_set_reg_ppc_online(PowerPCCPU *cpu, unsigned int online)
2929 {
2930     CPUState *cs = CPU(cpu);
2931 
2932     if (kvm_enabled()) {
2933         kvm_set_one_reg(cs, KVM_REG_PPC_ONLINE, &online);
2934     }
2935 }
2936