xref: /openbmc/qemu/target/ppc/kvm.c (revision 243975c0)
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 "qapi/error.h"
25 #include "qemu/error-report.h"
26 #include "cpu.h"
27 #include "cpu-models.h"
28 #include "qemu/timer.h"
29 #include "sysemu/hw_accel.h"
30 #include "kvm_ppc.h"
31 #include "sysemu/cpus.h"
32 #include "sysemu/device_tree.h"
33 #include "mmu-hash64.h"
34 
35 #include "hw/ppc/spapr.h"
36 #include "hw/ppc/spapr_cpu_core.h"
37 #include "hw/hw.h"
38 #include "hw/ppc/ppc.h"
39 #include "migration/qemu-file-types.h"
40 #include "sysemu/watchdog.h"
41 #include "trace.h"
42 #include "exec/gdbstub.h"
43 #include "exec/memattrs.h"
44 #include "exec/ram_addr.h"
45 #include "sysemu/hostmem.h"
46 #include "qemu/cutils.h"
47 #include "qemu/main-loop.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 #define DEBUG_RETURN_GUEST 0
55 #define DEBUG_RETURN_GDB   1
56 
57 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
58     KVM_CAP_LAST_INFO
59 };
60 
61 static int cap_interrupt_unset;
62 static int cap_segstate;
63 static int cap_booke_sregs;
64 static int cap_ppc_smt;
65 static int cap_ppc_smt_possible;
66 static int cap_spapr_tce;
67 static int cap_spapr_tce_64;
68 static int cap_spapr_multitce;
69 static int cap_spapr_vfio;
70 static int cap_hior;
71 static int cap_one_reg;
72 static int cap_epr;
73 static int cap_ppc_watchdog;
74 static int cap_papr;
75 static int cap_htab_fd;
76 static int cap_fixup_hcalls;
77 static int cap_htm;             /* Hardware transactional memory support */
78 static int cap_mmu_radix;
79 static int cap_mmu_hash_v3;
80 static int cap_xive;
81 static int cap_resize_hpt;
82 static int cap_ppc_pvr_compat;
83 static int cap_ppc_safe_cache;
84 static int cap_ppc_safe_bounds_check;
85 static int cap_ppc_safe_indirect_branch;
86 static int cap_ppc_count_cache_flush_assist;
87 static int cap_ppc_nested_kvm_hv;
88 static int cap_large_decr;
89 static int cap_fwnmi;
90 static int cap_rpt_invalidate;
91 static int cap_ail_mode_3;
92 
93 static uint32_t debug_inst_opcode;
94 
95 /*
96  * Check whether we are running with KVM-PR (instead of KVM-HV).  This
97  * should only be used for fallback tests - generally we should use
98  * explicit capabilities for the features we want, rather than
99  * assuming what is/isn't available depending on the KVM variant.
100  */
101 static bool kvmppc_is_pr(KVMState *ks)
102 {
103     /* Assume KVM-PR if the GET_PVINFO capability is available */
104     return kvm_vm_check_extension(ks, KVM_CAP_PPC_GET_PVINFO) != 0;
105 }
106 
107 static int kvm_ppc_register_host_cpu_type(void);
108 static void kvmppc_get_cpu_characteristics(KVMState *s);
109 static int kvmppc_get_dec_bits(void);
110 
111 int kvm_arch_get_default_type(MachineState *ms)
112 {
113     return 0;
114 }
115 
116 int kvm_arch_init(MachineState *ms, KVMState *s)
117 {
118     cap_interrupt_unset = kvm_check_extension(s, KVM_CAP_PPC_UNSET_IRQ);
119     cap_segstate = kvm_check_extension(s, KVM_CAP_PPC_SEGSTATE);
120     cap_booke_sregs = kvm_check_extension(s, KVM_CAP_PPC_BOOKE_SREGS);
121     cap_ppc_smt_possible = kvm_vm_check_extension(s, KVM_CAP_PPC_SMT_POSSIBLE);
122     cap_spapr_tce = kvm_check_extension(s, KVM_CAP_SPAPR_TCE);
123     cap_spapr_tce_64 = kvm_check_extension(s, KVM_CAP_SPAPR_TCE_64);
124     cap_spapr_multitce = kvm_check_extension(s, KVM_CAP_SPAPR_MULTITCE);
125     cap_spapr_vfio = kvm_vm_check_extension(s, KVM_CAP_SPAPR_TCE_VFIO);
126     cap_one_reg = kvm_check_extension(s, KVM_CAP_ONE_REG);
127     cap_hior = kvm_check_extension(s, KVM_CAP_PPC_HIOR);
128     cap_epr = kvm_check_extension(s, KVM_CAP_PPC_EPR);
129     cap_ppc_watchdog = kvm_check_extension(s, KVM_CAP_PPC_BOOKE_WATCHDOG);
130     /*
131      * Note: we don't set cap_papr here, because this capability is
132      * only activated after this by kvmppc_set_papr()
133      */
134     cap_htab_fd = kvm_vm_check_extension(s, KVM_CAP_PPC_HTAB_FD);
135     cap_fixup_hcalls = kvm_check_extension(s, KVM_CAP_PPC_FIXUP_HCALL);
136     cap_ppc_smt = kvm_vm_check_extension(s, KVM_CAP_PPC_SMT);
137     cap_htm = kvm_vm_check_extension(s, KVM_CAP_PPC_HTM);
138     cap_mmu_radix = kvm_vm_check_extension(s, KVM_CAP_PPC_MMU_RADIX);
139     cap_mmu_hash_v3 = kvm_vm_check_extension(s, KVM_CAP_PPC_MMU_HASH_V3);
140     cap_xive = kvm_vm_check_extension(s, KVM_CAP_PPC_IRQ_XIVE);
141     cap_resize_hpt = kvm_vm_check_extension(s, KVM_CAP_SPAPR_RESIZE_HPT);
142     kvmppc_get_cpu_characteristics(s);
143     cap_ppc_nested_kvm_hv = kvm_vm_check_extension(s, KVM_CAP_PPC_NESTED_HV);
144     cap_large_decr = kvmppc_get_dec_bits();
145     cap_fwnmi = kvm_vm_check_extension(s, KVM_CAP_PPC_FWNMI);
146     /*
147      * Note: setting it to false because there is not such capability
148      * in KVM at this moment.
149      *
150      * TODO: call kvm_vm_check_extension() with the right capability
151      * after the kernel starts implementing it.
152      */
153     cap_ppc_pvr_compat = false;
154 
155     if (!kvm_check_extension(s, KVM_CAP_PPC_IRQ_LEVEL)) {
156         error_report("KVM: Host kernel doesn't have level irq capability");
157         exit(1);
158     }
159 
160     cap_rpt_invalidate = kvm_vm_check_extension(s, KVM_CAP_PPC_RPT_INVALIDATE);
161     cap_ail_mode_3 = kvm_vm_check_extension(s, KVM_CAP_PPC_AIL_MODE_3);
162     kvm_ppc_register_host_cpu_type();
163 
164     return 0;
165 }
166 
167 int kvm_arch_irqchip_create(KVMState *s)
168 {
169     return 0;
170 }
171 
172 static int kvm_arch_sync_sregs(PowerPCCPU *cpu)
173 {
174     CPUPPCState *cenv = &cpu->env;
175     CPUState *cs = CPU(cpu);
176     struct kvm_sregs sregs;
177     int ret;
178 
179     if (cenv->excp_model == POWERPC_EXCP_BOOKE) {
180         /*
181          * What we're really trying to say is "if we're on BookE, we
182          * use the native PVR for now". This is the only sane way to
183          * check it though, so we potentially confuse users that they
184          * can run BookE guests on BookS. Let's hope nobody dares
185          * enough :)
186          */
187         return 0;
188     } else {
189         if (!cap_segstate) {
190             fprintf(stderr, "kvm error: missing PVR setting capability\n");
191             return -ENOSYS;
192         }
193     }
194 
195     ret = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs);
196     if (ret) {
197         return ret;
198     }
199 
200     sregs.pvr = cenv->spr[SPR_PVR];
201     return kvm_vcpu_ioctl(cs, KVM_SET_SREGS, &sregs);
202 }
203 
204 /* Set up a shared TLB array with KVM */
205 static int kvm_booke206_tlb_init(PowerPCCPU *cpu)
206 {
207     CPUPPCState *env = &cpu->env;
208     CPUState *cs = CPU(cpu);
209     struct kvm_book3e_206_tlb_params params = {};
210     struct kvm_config_tlb cfg = {};
211     unsigned int entries = 0;
212     int ret, i;
213 
214     if (!kvm_enabled() ||
215         !kvm_check_extension(cs->kvm_state, KVM_CAP_SW_TLB)) {
216         return 0;
217     }
218 
219     assert(ARRAY_SIZE(params.tlb_sizes) == BOOKE206_MAX_TLBN);
220 
221     for (i = 0; i < BOOKE206_MAX_TLBN; i++) {
222         params.tlb_sizes[i] = booke206_tlb_size(env, i);
223         params.tlb_ways[i] = booke206_tlb_ways(env, i);
224         entries += params.tlb_sizes[i];
225     }
226 
227     assert(entries == env->nb_tlb);
228     assert(sizeof(struct kvm_book3e_206_tlb_entry) == sizeof(ppcmas_tlb_t));
229 
230     env->tlb_dirty = true;
231 
232     cfg.array = (uintptr_t)env->tlb.tlbm;
233     cfg.array_len = sizeof(ppcmas_tlb_t) * entries;
234     cfg.params = (uintptr_t)&params;
235     cfg.mmu_type = KVM_MMU_FSL_BOOKE_NOHV;
236 
237     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_SW_TLB, 0, (uintptr_t)&cfg);
238     if (ret < 0) {
239         fprintf(stderr, "%s: couldn't enable KVM_CAP_SW_TLB: %s\n",
240                 __func__, strerror(-ret));
241         return ret;
242     }
243 
244     env->kvm_sw_tlb = true;
245     return 0;
246 }
247 
248 
249 #if defined(TARGET_PPC64)
250 static void kvm_get_smmu_info(struct kvm_ppc_smmu_info *info, Error **errp)
251 {
252     int ret;
253 
254     assert(kvm_state != NULL);
255 
256     if (!kvm_check_extension(kvm_state, KVM_CAP_PPC_GET_SMMU_INFO)) {
257         error_setg(errp, "KVM doesn't expose the MMU features it supports");
258         error_append_hint(errp, "Consider switching to a newer KVM\n");
259         return;
260     }
261 
262     ret = kvm_vm_ioctl(kvm_state, KVM_PPC_GET_SMMU_INFO, info);
263     if (ret == 0) {
264         return;
265     }
266 
267     error_setg_errno(errp, -ret,
268                      "KVM failed to provide the MMU features it supports");
269 }
270 
271 struct ppc_radix_page_info *kvm_get_radix_page_info(void)
272 {
273     KVMState *s = KVM_STATE(current_accel());
274     struct ppc_radix_page_info *radix_page_info;
275     struct kvm_ppc_rmmu_info rmmu_info = { };
276     int i;
277 
278     if (!kvm_check_extension(s, KVM_CAP_PPC_MMU_RADIX)) {
279         return NULL;
280     }
281     if (kvm_vm_ioctl(s, KVM_PPC_GET_RMMU_INFO, &rmmu_info)) {
282         return NULL;
283     }
284     radix_page_info = g_malloc0(sizeof(*radix_page_info));
285     radix_page_info->count = 0;
286     for (i = 0; i < PPC_PAGE_SIZES_MAX_SZ; i++) {
287         if (rmmu_info.ap_encodings[i]) {
288             radix_page_info->entries[i] = rmmu_info.ap_encodings[i];
289             radix_page_info->count++;
290         }
291     }
292     return radix_page_info;
293 }
294 
295 target_ulong kvmppc_configure_v3_mmu(PowerPCCPU *cpu,
296                                      bool radix, bool gtse,
297                                      uint64_t proc_tbl)
298 {
299     CPUState *cs = CPU(cpu);
300     int ret;
301     uint64_t flags = 0;
302     struct kvm_ppc_mmuv3_cfg cfg = {
303         .process_table = proc_tbl,
304     };
305 
306     if (radix) {
307         flags |= KVM_PPC_MMUV3_RADIX;
308     }
309     if (gtse) {
310         flags |= KVM_PPC_MMUV3_GTSE;
311     }
312     cfg.flags = flags;
313     ret = kvm_vm_ioctl(cs->kvm_state, KVM_PPC_CONFIGURE_V3_MMU, &cfg);
314     switch (ret) {
315     case 0:
316         return H_SUCCESS;
317     case -EINVAL:
318         return H_PARAMETER;
319     case -ENODEV:
320         return H_NOT_AVAILABLE;
321     default:
322         return H_HARDWARE;
323     }
324 }
325 
326 bool kvmppc_hpt_needs_host_contiguous_pages(void)
327 {
328     static struct kvm_ppc_smmu_info smmu_info;
329 
330     if (!kvm_enabled()) {
331         return false;
332     }
333 
334     kvm_get_smmu_info(&smmu_info, &error_fatal);
335     return !!(smmu_info.flags & KVM_PPC_PAGE_SIZES_REAL);
336 }
337 
338 void kvm_check_mmu(PowerPCCPU *cpu, Error **errp)
339 {
340     struct kvm_ppc_smmu_info smmu_info;
341     int iq, ik, jq, jk;
342     Error *local_err = NULL;
343 
344     /* For now, we only have anything to check on hash64 MMUs */
345     if (!cpu->hash64_opts || !kvm_enabled()) {
346         return;
347     }
348 
349     kvm_get_smmu_info(&smmu_info, &local_err);
350     if (local_err) {
351         error_propagate(errp, local_err);
352         return;
353     }
354 
355     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)
356         && !(smmu_info.flags & KVM_PPC_1T_SEGMENTS)) {
357         error_setg(errp,
358                    "KVM does not support 1TiB segments which guest expects");
359         return;
360     }
361 
362     if (smmu_info.slb_size < cpu->hash64_opts->slb_size) {
363         error_setg(errp, "KVM only supports %u SLB entries, but guest needs %u",
364                    smmu_info.slb_size, cpu->hash64_opts->slb_size);
365         return;
366     }
367 
368     /*
369      * Verify that every pagesize supported by the cpu model is
370      * supported by KVM with the same encodings
371      */
372     for (iq = 0; iq < ARRAY_SIZE(cpu->hash64_opts->sps); iq++) {
373         PPCHash64SegmentPageSizes *qsps = &cpu->hash64_opts->sps[iq];
374         struct kvm_ppc_one_seg_page_size *ksps;
375 
376         for (ik = 0; ik < ARRAY_SIZE(smmu_info.sps); ik++) {
377             if (qsps->page_shift == smmu_info.sps[ik].page_shift) {
378                 break;
379             }
380         }
381         if (ik >= ARRAY_SIZE(smmu_info.sps)) {
382             error_setg(errp, "KVM doesn't support for base page shift %u",
383                        qsps->page_shift);
384             return;
385         }
386 
387         ksps = &smmu_info.sps[ik];
388         if (ksps->slb_enc != qsps->slb_enc) {
389             error_setg(errp,
390 "KVM uses SLB encoding 0x%x for page shift %u, but guest expects 0x%x",
391                        ksps->slb_enc, ksps->page_shift, qsps->slb_enc);
392             return;
393         }
394 
395         for (jq = 0; jq < ARRAY_SIZE(qsps->enc); jq++) {
396             for (jk = 0; jk < ARRAY_SIZE(ksps->enc); jk++) {
397                 if (qsps->enc[jq].page_shift == ksps->enc[jk].page_shift) {
398                     break;
399                 }
400             }
401 
402             if (jk >= ARRAY_SIZE(ksps->enc)) {
403                 error_setg(errp, "KVM doesn't support page shift %u/%u",
404                            qsps->enc[jq].page_shift, qsps->page_shift);
405                 return;
406             }
407             if (qsps->enc[jq].pte_enc != ksps->enc[jk].pte_enc) {
408                 error_setg(errp,
409 "KVM uses PTE encoding 0x%x for page shift %u/%u, but guest expects 0x%x",
410                            ksps->enc[jk].pte_enc, qsps->enc[jq].page_shift,
411                            qsps->page_shift, qsps->enc[jq].pte_enc);
412                 return;
413             }
414         }
415     }
416 
417     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
418         /*
419          * Mostly what guest pagesizes we can use are related to the
420          * host pages used to map guest RAM, which is handled in the
421          * platform code. Cache-Inhibited largepages (64k) however are
422          * used for I/O, so if they're mapped to the host at all it
423          * will be a normal mapping, not a special hugepage one used
424          * for RAM.
425          */
426         if (qemu_real_host_page_size() < 0x10000) {
427             error_setg(errp,
428                        "KVM can't supply 64kiB CI pages, which guest expects");
429         }
430     }
431 }
432 #endif /* !defined (TARGET_PPC64) */
433 
434 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
435 {
436     return POWERPC_CPU(cpu)->vcpu_id;
437 }
438 
439 /*
440  * e500 supports 2 h/w breakpoint and 2 watchpoint.  book3s supports
441  * only 1 watchpoint, so array size of 4 is sufficient for now.
442  */
443 #define MAX_HW_BKPTS 4
444 
445 static struct HWBreakpoint {
446     target_ulong addr;
447     int type;
448 } hw_debug_points[MAX_HW_BKPTS];
449 
450 static CPUWatchpoint hw_watchpoint;
451 
452 /* Default there is no breakpoint and watchpoint supported */
453 static int max_hw_breakpoint;
454 static int max_hw_watchpoint;
455 static int nb_hw_breakpoint;
456 static int nb_hw_watchpoint;
457 
458 static void kvmppc_hw_debug_points_init(CPUPPCState *cenv)
459 {
460     if (cenv->excp_model == POWERPC_EXCP_BOOKE) {
461         max_hw_breakpoint = 2;
462         max_hw_watchpoint = 2;
463     }
464 
465     if ((max_hw_breakpoint + max_hw_watchpoint) > MAX_HW_BKPTS) {
466         fprintf(stderr, "Error initializing h/w breakpoints\n");
467         return;
468     }
469 }
470 
471 int kvm_arch_init_vcpu(CPUState *cs)
472 {
473     PowerPCCPU *cpu = POWERPC_CPU(cs);
474     CPUPPCState *cenv = &cpu->env;
475     int ret;
476 
477     /* Synchronize sregs with kvm */
478     ret = kvm_arch_sync_sregs(cpu);
479     if (ret) {
480         if (ret == -EINVAL) {
481             error_report("Register sync failed... If you're using kvm-hv.ko,"
482                          " only \"-cpu host\" is possible");
483         }
484         return ret;
485     }
486 
487     switch (cenv->mmu_model) {
488     case POWERPC_MMU_BOOKE206:
489         /* This target supports access to KVM's guest TLB */
490         ret = kvm_booke206_tlb_init(cpu);
491         break;
492     case POWERPC_MMU_2_07:
493         if (!cap_htm && !kvmppc_is_pr(cs->kvm_state)) {
494             /*
495              * KVM-HV has transactional memory on POWER8 also without
496              * the KVM_CAP_PPC_HTM extension, so enable it here
497              * instead as long as it's available to userspace on the
498              * host.
499              */
500             if (qemu_getauxval(AT_HWCAP2) & PPC_FEATURE2_HAS_HTM) {
501                 cap_htm = true;
502             }
503         }
504         break;
505     default:
506         break;
507     }
508 
509     kvm_get_one_reg(cs, KVM_REG_PPC_DEBUG_INST, &debug_inst_opcode);
510     kvmppc_hw_debug_points_init(cenv);
511 
512     return ret;
513 }
514 
515 int kvm_arch_destroy_vcpu(CPUState *cs)
516 {
517     return 0;
518 }
519 
520 static void kvm_sw_tlb_put(PowerPCCPU *cpu)
521 {
522     CPUPPCState *env = &cpu->env;
523     CPUState *cs = CPU(cpu);
524     struct kvm_dirty_tlb dirty_tlb;
525     unsigned char *bitmap;
526     int ret;
527 
528     if (!env->kvm_sw_tlb) {
529         return;
530     }
531 
532     bitmap = g_malloc((env->nb_tlb + 7) / 8);
533     memset(bitmap, 0xFF, (env->nb_tlb + 7) / 8);
534 
535     dirty_tlb.bitmap = (uintptr_t)bitmap;
536     dirty_tlb.num_dirty = env->nb_tlb;
537 
538     ret = kvm_vcpu_ioctl(cs, KVM_DIRTY_TLB, &dirty_tlb);
539     if (ret) {
540         fprintf(stderr, "%s: KVM_DIRTY_TLB: %s\n",
541                 __func__, strerror(-ret));
542     }
543 
544     g_free(bitmap);
545 }
546 
547 static void kvm_get_one_spr(CPUState *cs, uint64_t id, int spr)
548 {
549     PowerPCCPU *cpu = POWERPC_CPU(cs);
550     CPUPPCState *env = &cpu->env;
551     /* Init 'val' to avoid "uninitialised value" Valgrind warnings */
552     union {
553         uint32_t u32;
554         uint64_t u64;
555     } val = { };
556     struct kvm_one_reg reg = {
557         .id = id,
558         .addr = (uintptr_t) &val,
559     };
560     int ret;
561 
562     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
563     if (ret != 0) {
564         trace_kvm_failed_spr_get(spr, strerror(errno));
565     } else {
566         switch (id & KVM_REG_SIZE_MASK) {
567         case KVM_REG_SIZE_U32:
568             env->spr[spr] = val.u32;
569             break;
570 
571         case KVM_REG_SIZE_U64:
572             env->spr[spr] = val.u64;
573             break;
574 
575         default:
576             /* Don't handle this size yet */
577             abort();
578         }
579     }
580 }
581 
582 static void kvm_put_one_spr(CPUState *cs, uint64_t id, int spr)
583 {
584     PowerPCCPU *cpu = POWERPC_CPU(cs);
585     CPUPPCState *env = &cpu->env;
586     union {
587         uint32_t u32;
588         uint64_t u64;
589     } val;
590     struct kvm_one_reg reg = {
591         .id = id,
592         .addr = (uintptr_t) &val,
593     };
594     int ret;
595 
596     switch (id & KVM_REG_SIZE_MASK) {
597     case KVM_REG_SIZE_U32:
598         val.u32 = env->spr[spr];
599         break;
600 
601     case KVM_REG_SIZE_U64:
602         val.u64 = env->spr[spr];
603         break;
604 
605     default:
606         /* Don't handle this size yet */
607         abort();
608     }
609 
610     ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
611     if (ret != 0) {
612         trace_kvm_failed_spr_set(spr, strerror(errno));
613     }
614 }
615 
616 static int kvm_put_fp(CPUState *cs)
617 {
618     PowerPCCPU *cpu = POWERPC_CPU(cs);
619     CPUPPCState *env = &cpu->env;
620     struct kvm_one_reg reg;
621     int i;
622     int ret;
623 
624     if (env->insns_flags & PPC_FLOAT) {
625         uint64_t fpscr = env->fpscr;
626         bool vsx = !!(env->insns_flags2 & PPC2_VSX);
627 
628         reg.id = KVM_REG_PPC_FPSCR;
629         reg.addr = (uintptr_t)&fpscr;
630         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
631         if (ret < 0) {
632             trace_kvm_failed_fpscr_set(strerror(errno));
633             return ret;
634         }
635 
636         for (i = 0; i < 32; i++) {
637             uint64_t vsr[2];
638             uint64_t *fpr = cpu_fpr_ptr(&cpu->env, i);
639             uint64_t *vsrl = cpu_vsrl_ptr(&cpu->env, i);
640 
641 #if HOST_BIG_ENDIAN
642             vsr[0] = float64_val(*fpr);
643             vsr[1] = *vsrl;
644 #else
645             vsr[0] = *vsrl;
646             vsr[1] = float64_val(*fpr);
647 #endif
648             reg.addr = (uintptr_t) &vsr;
649             reg.id = vsx ? KVM_REG_PPC_VSR(i) : KVM_REG_PPC_FPR(i);
650 
651             ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
652             if (ret < 0) {
653                 trace_kvm_failed_fp_set(vsx ? "VSR" : "FPR", i,
654                                         strerror(errno));
655                 return ret;
656             }
657         }
658     }
659 
660     if (env->insns_flags & PPC_ALTIVEC) {
661         reg.id = KVM_REG_PPC_VSCR;
662         reg.addr = (uintptr_t)&env->vscr;
663         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
664         if (ret < 0) {
665             trace_kvm_failed_vscr_set(strerror(errno));
666             return ret;
667         }
668 
669         for (i = 0; i < 32; i++) {
670             reg.id = KVM_REG_PPC_VR(i);
671             reg.addr = (uintptr_t)cpu_avr_ptr(env, i);
672             ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
673             if (ret < 0) {
674                 trace_kvm_failed_vr_set(i, strerror(errno));
675                 return ret;
676             }
677         }
678     }
679 
680     return 0;
681 }
682 
683 static int kvm_get_fp(CPUState *cs)
684 {
685     PowerPCCPU *cpu = POWERPC_CPU(cs);
686     CPUPPCState *env = &cpu->env;
687     struct kvm_one_reg reg;
688     int i;
689     int ret;
690 
691     if (env->insns_flags & PPC_FLOAT) {
692         uint64_t fpscr;
693         bool vsx = !!(env->insns_flags2 & PPC2_VSX);
694 
695         reg.id = KVM_REG_PPC_FPSCR;
696         reg.addr = (uintptr_t)&fpscr;
697         ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
698         if (ret < 0) {
699             trace_kvm_failed_fpscr_get(strerror(errno));
700             return ret;
701         } else {
702             env->fpscr = fpscr;
703         }
704 
705         for (i = 0; i < 32; i++) {
706             uint64_t vsr[2];
707             uint64_t *fpr = cpu_fpr_ptr(&cpu->env, i);
708             uint64_t *vsrl = cpu_vsrl_ptr(&cpu->env, i);
709 
710             reg.addr = (uintptr_t) &vsr;
711             reg.id = vsx ? KVM_REG_PPC_VSR(i) : KVM_REG_PPC_FPR(i);
712 
713             ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
714             if (ret < 0) {
715                 trace_kvm_failed_fp_get(vsx ? "VSR" : "FPR", i,
716                                         strerror(errno));
717                 return ret;
718             } else {
719 #if HOST_BIG_ENDIAN
720                 *fpr = vsr[0];
721                 if (vsx) {
722                     *vsrl = vsr[1];
723                 }
724 #else
725                 *fpr = vsr[1];
726                 if (vsx) {
727                     *vsrl = vsr[0];
728                 }
729 #endif
730             }
731         }
732     }
733 
734     if (env->insns_flags & PPC_ALTIVEC) {
735         reg.id = KVM_REG_PPC_VSCR;
736         reg.addr = (uintptr_t)&env->vscr;
737         ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
738         if (ret < 0) {
739             trace_kvm_failed_vscr_get(strerror(errno));
740             return ret;
741         }
742 
743         for (i = 0; i < 32; i++) {
744             reg.id = KVM_REG_PPC_VR(i);
745             reg.addr = (uintptr_t)cpu_avr_ptr(env, i);
746             ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
747             if (ret < 0) {
748                 trace_kvm_failed_vr_get(i, strerror(errno));
749                 return ret;
750             }
751         }
752     }
753 
754     return 0;
755 }
756 
757 #if defined(TARGET_PPC64)
758 static int kvm_get_vpa(CPUState *cs)
759 {
760     PowerPCCPU *cpu = POWERPC_CPU(cs);
761     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
762     struct kvm_one_reg reg;
763     int ret;
764 
765     reg.id = KVM_REG_PPC_VPA_ADDR;
766     reg.addr = (uintptr_t)&spapr_cpu->vpa_addr;
767     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
768     if (ret < 0) {
769         trace_kvm_failed_vpa_addr_get(strerror(errno));
770         return ret;
771     }
772 
773     assert((uintptr_t)&spapr_cpu->slb_shadow_size
774            == ((uintptr_t)&spapr_cpu->slb_shadow_addr + 8));
775     reg.id = KVM_REG_PPC_VPA_SLB;
776     reg.addr = (uintptr_t)&spapr_cpu->slb_shadow_addr;
777     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
778     if (ret < 0) {
779         trace_kvm_failed_slb_get(strerror(errno));
780         return ret;
781     }
782 
783     assert((uintptr_t)&spapr_cpu->dtl_size
784            == ((uintptr_t)&spapr_cpu->dtl_addr + 8));
785     reg.id = KVM_REG_PPC_VPA_DTL;
786     reg.addr = (uintptr_t)&spapr_cpu->dtl_addr;
787     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
788     if (ret < 0) {
789         trace_kvm_failed_dtl_get(strerror(errno));
790         return ret;
791     }
792 
793     return 0;
794 }
795 
796 static int kvm_put_vpa(CPUState *cs)
797 {
798     PowerPCCPU *cpu = POWERPC_CPU(cs);
799     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
800     struct kvm_one_reg reg;
801     int ret;
802 
803     /*
804      * SLB shadow or DTL can't be registered unless a master VPA is
805      * registered.  That means when restoring state, if a VPA *is*
806      * registered, we need to set that up first.  If not, we need to
807      * deregister the others before deregistering the master VPA
808      */
809     assert(spapr_cpu->vpa_addr
810            || !(spapr_cpu->slb_shadow_addr || spapr_cpu->dtl_addr));
811 
812     if (spapr_cpu->vpa_addr) {
813         reg.id = KVM_REG_PPC_VPA_ADDR;
814         reg.addr = (uintptr_t)&spapr_cpu->vpa_addr;
815         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
816         if (ret < 0) {
817             trace_kvm_failed_vpa_addr_set(strerror(errno));
818             return ret;
819         }
820     }
821 
822     assert((uintptr_t)&spapr_cpu->slb_shadow_size
823            == ((uintptr_t)&spapr_cpu->slb_shadow_addr + 8));
824     reg.id = KVM_REG_PPC_VPA_SLB;
825     reg.addr = (uintptr_t)&spapr_cpu->slb_shadow_addr;
826     ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
827     if (ret < 0) {
828         trace_kvm_failed_slb_set(strerror(errno));
829         return ret;
830     }
831 
832     assert((uintptr_t)&spapr_cpu->dtl_size
833            == ((uintptr_t)&spapr_cpu->dtl_addr + 8));
834     reg.id = KVM_REG_PPC_VPA_DTL;
835     reg.addr = (uintptr_t)&spapr_cpu->dtl_addr;
836     ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
837     if (ret < 0) {
838         trace_kvm_failed_dtl_set(strerror(errno));
839         return ret;
840     }
841 
842     if (!spapr_cpu->vpa_addr) {
843         reg.id = KVM_REG_PPC_VPA_ADDR;
844         reg.addr = (uintptr_t)&spapr_cpu->vpa_addr;
845         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
846         if (ret < 0) {
847             trace_kvm_failed_null_vpa_addr_set(strerror(errno));
848             return ret;
849         }
850     }
851 
852     return 0;
853 }
854 #endif /* TARGET_PPC64 */
855 
856 int kvmppc_put_books_sregs(PowerPCCPU *cpu)
857 {
858     CPUPPCState *env = &cpu->env;
859     struct kvm_sregs sregs = { };
860     int i;
861 
862     sregs.pvr = env->spr[SPR_PVR];
863 
864     if (cpu->vhyp) {
865         PPCVirtualHypervisorClass *vhc =
866             PPC_VIRTUAL_HYPERVISOR_GET_CLASS(cpu->vhyp);
867         sregs.u.s.sdr1 = vhc->encode_hpt_for_kvm_pr(cpu->vhyp);
868     } else {
869         sregs.u.s.sdr1 = env->spr[SPR_SDR1];
870     }
871 
872     /* Sync SLB */
873 #ifdef TARGET_PPC64
874     for (i = 0; i < ARRAY_SIZE(env->slb); i++) {
875         sregs.u.s.ppc64.slb[i].slbe = env->slb[i].esid;
876         if (env->slb[i].esid & SLB_ESID_V) {
877             sregs.u.s.ppc64.slb[i].slbe |= i;
878         }
879         sregs.u.s.ppc64.slb[i].slbv = env->slb[i].vsid;
880     }
881 #endif
882 
883     /* Sync SRs */
884     for (i = 0; i < 16; i++) {
885         sregs.u.s.ppc32.sr[i] = env->sr[i];
886     }
887 
888     /* Sync BATs */
889     for (i = 0; i < 8; i++) {
890         /* Beware. We have to swap upper and lower bits here */
891         sregs.u.s.ppc32.dbat[i] = ((uint64_t)env->DBAT[0][i] << 32)
892             | env->DBAT[1][i];
893         sregs.u.s.ppc32.ibat[i] = ((uint64_t)env->IBAT[0][i] << 32)
894             | env->IBAT[1][i];
895     }
896 
897     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_SREGS, &sregs);
898 }
899 
900 int kvm_arch_put_registers(CPUState *cs, int level)
901 {
902     PowerPCCPU *cpu = POWERPC_CPU(cs);
903     CPUPPCState *env = &cpu->env;
904     struct kvm_regs regs;
905     int ret;
906     int i;
907 
908     ret = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
909     if (ret < 0) {
910         return ret;
911     }
912 
913     regs.ctr = env->ctr;
914     regs.lr  = env->lr;
915     regs.xer = cpu_read_xer(env);
916     regs.msr = env->msr;
917     regs.pc = env->nip;
918 
919     regs.srr0 = env->spr[SPR_SRR0];
920     regs.srr1 = env->spr[SPR_SRR1];
921 
922     regs.sprg0 = env->spr[SPR_SPRG0];
923     regs.sprg1 = env->spr[SPR_SPRG1];
924     regs.sprg2 = env->spr[SPR_SPRG2];
925     regs.sprg3 = env->spr[SPR_SPRG3];
926     regs.sprg4 = env->spr[SPR_SPRG4];
927     regs.sprg5 = env->spr[SPR_SPRG5];
928     regs.sprg6 = env->spr[SPR_SPRG6];
929     regs.sprg7 = env->spr[SPR_SPRG7];
930 
931     regs.pid = env->spr[SPR_BOOKE_PID];
932 
933     for (i = 0; i < 32; i++) {
934         regs.gpr[i] = env->gpr[i];
935     }
936 
937     regs.cr = ppc_get_cr(env);
938 
939     ret = kvm_vcpu_ioctl(cs, KVM_SET_REGS, &regs);
940     if (ret < 0) {
941         return ret;
942     }
943 
944     kvm_put_fp(cs);
945 
946     if (env->tlb_dirty) {
947         kvm_sw_tlb_put(cpu);
948         env->tlb_dirty = false;
949     }
950 
951     if (cap_segstate && (level >= KVM_PUT_RESET_STATE)) {
952         ret = kvmppc_put_books_sregs(cpu);
953         if (ret < 0) {
954             return ret;
955         }
956     }
957 
958     if (cap_hior && (level >= KVM_PUT_RESET_STATE)) {
959         kvm_put_one_spr(cs, KVM_REG_PPC_HIOR, SPR_HIOR);
960     }
961 
962     if (cap_one_reg) {
963         int i;
964 
965         /*
966          * We deliberately ignore errors here, for kernels which have
967          * the ONE_REG calls, but don't support the specific
968          * registers, there's a reasonable chance things will still
969          * work, at least until we try to migrate.
970          */
971         for (i = 0; i < 1024; i++) {
972             uint64_t id = env->spr_cb[i].one_reg_id;
973 
974             if (id != 0) {
975                 kvm_put_one_spr(cs, id, i);
976             }
977         }
978 
979 #ifdef TARGET_PPC64
980         if (FIELD_EX64(env->msr, MSR, TS)) {
981             for (i = 0; i < ARRAY_SIZE(env->tm_gpr); i++) {
982                 kvm_set_one_reg(cs, KVM_REG_PPC_TM_GPR(i), &env->tm_gpr[i]);
983             }
984             for (i = 0; i < ARRAY_SIZE(env->tm_vsr); i++) {
985                 kvm_set_one_reg(cs, KVM_REG_PPC_TM_VSR(i), &env->tm_vsr[i]);
986             }
987             kvm_set_one_reg(cs, KVM_REG_PPC_TM_CR, &env->tm_cr);
988             kvm_set_one_reg(cs, KVM_REG_PPC_TM_LR, &env->tm_lr);
989             kvm_set_one_reg(cs, KVM_REG_PPC_TM_CTR, &env->tm_ctr);
990             kvm_set_one_reg(cs, KVM_REG_PPC_TM_FPSCR, &env->tm_fpscr);
991             kvm_set_one_reg(cs, KVM_REG_PPC_TM_AMR, &env->tm_amr);
992             kvm_set_one_reg(cs, KVM_REG_PPC_TM_PPR, &env->tm_ppr);
993             kvm_set_one_reg(cs, KVM_REG_PPC_TM_VRSAVE, &env->tm_vrsave);
994             kvm_set_one_reg(cs, KVM_REG_PPC_TM_VSCR, &env->tm_vscr);
995             kvm_set_one_reg(cs, KVM_REG_PPC_TM_DSCR, &env->tm_dscr);
996             kvm_set_one_reg(cs, KVM_REG_PPC_TM_TAR, &env->tm_tar);
997         }
998 
999         if (cap_papr) {
1000             if (kvm_put_vpa(cs) < 0) {
1001                 trace_kvm_failed_put_vpa();
1002             }
1003         }
1004 
1005         kvm_set_one_reg(cs, KVM_REG_PPC_TB_OFFSET, &env->tb_env->tb_offset);
1006 
1007         if (level > KVM_PUT_RUNTIME_STATE) {
1008             kvm_put_one_spr(cs, KVM_REG_PPC_DPDES, SPR_DPDES);
1009         }
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     int i, ret;
1213 
1214     ret = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
1215     if (ret < 0) {
1216         return ret;
1217     }
1218 
1219     ppc_set_cr(env, regs.cr);
1220     env->ctr = regs.ctr;
1221     env->lr = regs.lr;
1222     cpu_write_xer(env, regs.xer);
1223     env->msr = regs.msr;
1224     env->nip = regs.pc;
1225 
1226     env->spr[SPR_SRR0] = regs.srr0;
1227     env->spr[SPR_SRR1] = regs.srr1;
1228 
1229     env->spr[SPR_SPRG0] = regs.sprg0;
1230     env->spr[SPR_SPRG1] = regs.sprg1;
1231     env->spr[SPR_SPRG2] = regs.sprg2;
1232     env->spr[SPR_SPRG3] = regs.sprg3;
1233     env->spr[SPR_SPRG4] = regs.sprg4;
1234     env->spr[SPR_SPRG5] = regs.sprg5;
1235     env->spr[SPR_SPRG6] = regs.sprg6;
1236     env->spr[SPR_SPRG7] = regs.sprg7;
1237 
1238     env->spr[SPR_BOOKE_PID] = regs.pid;
1239 
1240     for (i = 0; i < 32; i++) {
1241         env->gpr[i] = regs.gpr[i];
1242     }
1243 
1244     kvm_get_fp(cs);
1245 
1246     if (cap_booke_sregs) {
1247         ret = kvmppc_get_booke_sregs(cpu);
1248         if (ret < 0) {
1249             return ret;
1250         }
1251     }
1252 
1253     if (cap_segstate) {
1254         ret = kvmppc_get_books_sregs(cpu);
1255         if (ret < 0) {
1256             return ret;
1257         }
1258     }
1259 
1260     if (cap_hior) {
1261         kvm_get_one_spr(cs, KVM_REG_PPC_HIOR, SPR_HIOR);
1262     }
1263 
1264     if (cap_one_reg) {
1265         int i;
1266 
1267         /*
1268          * We deliberately ignore errors here, for kernels which have
1269          * the ONE_REG calls, but don't support the specific
1270          * registers, there's a reasonable chance things will still
1271          * work, at least until we try to migrate.
1272          */
1273         for (i = 0; i < 1024; i++) {
1274             uint64_t id = env->spr_cb[i].one_reg_id;
1275 
1276             if (id != 0) {
1277                 kvm_get_one_spr(cs, id, i);
1278             }
1279         }
1280 
1281 #ifdef TARGET_PPC64
1282         if (FIELD_EX64(env->msr, MSR, TS)) {
1283             for (i = 0; i < ARRAY_SIZE(env->tm_gpr); i++) {
1284                 kvm_get_one_reg(cs, KVM_REG_PPC_TM_GPR(i), &env->tm_gpr[i]);
1285             }
1286             for (i = 0; i < ARRAY_SIZE(env->tm_vsr); i++) {
1287                 kvm_get_one_reg(cs, KVM_REG_PPC_TM_VSR(i), &env->tm_vsr[i]);
1288             }
1289             kvm_get_one_reg(cs, KVM_REG_PPC_TM_CR, &env->tm_cr);
1290             kvm_get_one_reg(cs, KVM_REG_PPC_TM_LR, &env->tm_lr);
1291             kvm_get_one_reg(cs, KVM_REG_PPC_TM_CTR, &env->tm_ctr);
1292             kvm_get_one_reg(cs, KVM_REG_PPC_TM_FPSCR, &env->tm_fpscr);
1293             kvm_get_one_reg(cs, KVM_REG_PPC_TM_AMR, &env->tm_amr);
1294             kvm_get_one_reg(cs, KVM_REG_PPC_TM_PPR, &env->tm_ppr);
1295             kvm_get_one_reg(cs, KVM_REG_PPC_TM_VRSAVE, &env->tm_vrsave);
1296             kvm_get_one_reg(cs, KVM_REG_PPC_TM_VSCR, &env->tm_vscr);
1297             kvm_get_one_reg(cs, KVM_REG_PPC_TM_DSCR, &env->tm_dscr);
1298             kvm_get_one_reg(cs, KVM_REG_PPC_TM_TAR, &env->tm_tar);
1299         }
1300 
1301         if (cap_papr) {
1302             if (kvm_get_vpa(cs) < 0) {
1303                 trace_kvm_failed_get_vpa();
1304             }
1305         }
1306 
1307         kvm_get_one_reg(cs, KVM_REG_PPC_TB_OFFSET, &env->tb_env->tb_offset);
1308         kvm_get_one_spr(cs, KVM_REG_PPC_DPDES, SPR_DPDES);
1309 #endif
1310     }
1311 
1312     return 0;
1313 }
1314 
1315 int kvmppc_set_interrupt(PowerPCCPU *cpu, int irq, int level)
1316 {
1317     unsigned virq = level ? KVM_INTERRUPT_SET_LEVEL : KVM_INTERRUPT_UNSET;
1318 
1319     if (irq != PPC_INTERRUPT_EXT) {
1320         return 0;
1321     }
1322 
1323     if (!kvm_enabled() || !cap_interrupt_unset) {
1324         return 0;
1325     }
1326 
1327     kvm_vcpu_ioctl(CPU(cpu), KVM_INTERRUPT, &virq);
1328 
1329     return 0;
1330 }
1331 
1332 void kvm_arch_pre_run(CPUState *cs, struct kvm_run *run)
1333 {
1334     return;
1335 }
1336 
1337 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
1338 {
1339     return MEMTXATTRS_UNSPECIFIED;
1340 }
1341 
1342 int kvm_arch_process_async_events(CPUState *cs)
1343 {
1344     return cs->halted;
1345 }
1346 
1347 static int kvmppc_handle_halt(PowerPCCPU *cpu)
1348 {
1349     CPUState *cs = CPU(cpu);
1350     CPUPPCState *env = &cpu->env;
1351 
1352     if (!(cs->interrupt_request & CPU_INTERRUPT_HARD) &&
1353         FIELD_EX64(env->msr, MSR, EE)) {
1354         cs->halted = 1;
1355         cs->exception_index = EXCP_HLT;
1356     }
1357 
1358     return 0;
1359 }
1360 
1361 /* map dcr access to existing qemu dcr emulation */
1362 static int kvmppc_handle_dcr_read(CPUPPCState *env,
1363                                   uint32_t dcrn, uint32_t *data)
1364 {
1365     if (ppc_dcr_read(env->dcr_env, dcrn, data) < 0) {
1366         fprintf(stderr, "Read to unhandled DCR (0x%x)\n", dcrn);
1367     }
1368 
1369     return 0;
1370 }
1371 
1372 static int kvmppc_handle_dcr_write(CPUPPCState *env,
1373                                    uint32_t dcrn, uint32_t data)
1374 {
1375     if (ppc_dcr_write(env->dcr_env, dcrn, data) < 0) {
1376         fprintf(stderr, "Write to unhandled DCR (0x%x)\n", dcrn);
1377     }
1378 
1379     return 0;
1380 }
1381 
1382 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
1383 {
1384     /* Mixed endian case is not handled */
1385     uint32_t sc = debug_inst_opcode;
1386 
1387     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
1388                             sizeof(sc), 0) ||
1389         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&sc, sizeof(sc), 1)) {
1390         return -EINVAL;
1391     }
1392 
1393     return 0;
1394 }
1395 
1396 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
1397 {
1398     uint32_t sc;
1399 
1400     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&sc, sizeof(sc), 0) ||
1401         sc != debug_inst_opcode ||
1402         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
1403                             sizeof(sc), 1)) {
1404         return -EINVAL;
1405     }
1406 
1407     return 0;
1408 }
1409 
1410 static int find_hw_breakpoint(target_ulong addr, int type)
1411 {
1412     int n;
1413 
1414     assert((nb_hw_breakpoint + nb_hw_watchpoint)
1415            <= ARRAY_SIZE(hw_debug_points));
1416 
1417     for (n = 0; n < nb_hw_breakpoint + nb_hw_watchpoint; n++) {
1418         if (hw_debug_points[n].addr == addr &&
1419              hw_debug_points[n].type == type) {
1420             return n;
1421         }
1422     }
1423 
1424     return -1;
1425 }
1426 
1427 static int find_hw_watchpoint(target_ulong addr, int *flag)
1428 {
1429     int n;
1430 
1431     n = find_hw_breakpoint(addr, GDB_WATCHPOINT_ACCESS);
1432     if (n >= 0) {
1433         *flag = BP_MEM_ACCESS;
1434         return n;
1435     }
1436 
1437     n = find_hw_breakpoint(addr, GDB_WATCHPOINT_WRITE);
1438     if (n >= 0) {
1439         *flag = BP_MEM_WRITE;
1440         return n;
1441     }
1442 
1443     n = find_hw_breakpoint(addr, GDB_WATCHPOINT_READ);
1444     if (n >= 0) {
1445         *flag = BP_MEM_READ;
1446         return n;
1447     }
1448 
1449     return -1;
1450 }
1451 
1452 int kvm_arch_insert_hw_breakpoint(vaddr addr, vaddr len, int type)
1453 {
1454     const unsigned breakpoint_index = nb_hw_breakpoint + nb_hw_watchpoint;
1455     if (breakpoint_index >= ARRAY_SIZE(hw_debug_points)) {
1456         return -ENOBUFS;
1457     }
1458 
1459     hw_debug_points[breakpoint_index].addr = addr;
1460     hw_debug_points[breakpoint_index].type = type;
1461 
1462     switch (type) {
1463     case GDB_BREAKPOINT_HW:
1464         if (nb_hw_breakpoint >= max_hw_breakpoint) {
1465             return -ENOBUFS;
1466         }
1467 
1468         if (find_hw_breakpoint(addr, type) >= 0) {
1469             return -EEXIST;
1470         }
1471 
1472         nb_hw_breakpoint++;
1473         break;
1474 
1475     case GDB_WATCHPOINT_WRITE:
1476     case GDB_WATCHPOINT_READ:
1477     case GDB_WATCHPOINT_ACCESS:
1478         if (nb_hw_watchpoint >= max_hw_watchpoint) {
1479             return -ENOBUFS;
1480         }
1481 
1482         if (find_hw_breakpoint(addr, type) >= 0) {
1483             return -EEXIST;
1484         }
1485 
1486         nb_hw_watchpoint++;
1487         break;
1488 
1489     default:
1490         return -ENOSYS;
1491     }
1492 
1493     return 0;
1494 }
1495 
1496 int kvm_arch_remove_hw_breakpoint(vaddr addr, vaddr len, int type)
1497 {
1498     int n;
1499 
1500     n = find_hw_breakpoint(addr, type);
1501     if (n < 0) {
1502         return -ENOENT;
1503     }
1504 
1505     switch (type) {
1506     case GDB_BREAKPOINT_HW:
1507         nb_hw_breakpoint--;
1508         break;
1509 
1510     case GDB_WATCHPOINT_WRITE:
1511     case GDB_WATCHPOINT_READ:
1512     case GDB_WATCHPOINT_ACCESS:
1513         nb_hw_watchpoint--;
1514         break;
1515 
1516     default:
1517         return -ENOSYS;
1518     }
1519     hw_debug_points[n] = hw_debug_points[nb_hw_breakpoint + nb_hw_watchpoint];
1520 
1521     return 0;
1522 }
1523 
1524 void kvm_arch_remove_all_hw_breakpoints(void)
1525 {
1526     nb_hw_breakpoint = nb_hw_watchpoint = 0;
1527 }
1528 
1529 void kvm_arch_update_guest_debug(CPUState *cs, struct kvm_guest_debug *dbg)
1530 {
1531     int n;
1532 
1533     /* Software Breakpoint updates */
1534     if (kvm_sw_breakpoints_active(cs)) {
1535         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP;
1536     }
1537 
1538     assert((nb_hw_breakpoint + nb_hw_watchpoint)
1539            <= ARRAY_SIZE(hw_debug_points));
1540     assert((nb_hw_breakpoint + nb_hw_watchpoint) <= ARRAY_SIZE(dbg->arch.bp));
1541 
1542     if (nb_hw_breakpoint + nb_hw_watchpoint > 0) {
1543         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
1544         memset(dbg->arch.bp, 0, sizeof(dbg->arch.bp));
1545         for (n = 0; n < nb_hw_breakpoint + nb_hw_watchpoint; n++) {
1546             switch (hw_debug_points[n].type) {
1547             case GDB_BREAKPOINT_HW:
1548                 dbg->arch.bp[n].type = KVMPPC_DEBUG_BREAKPOINT;
1549                 break;
1550             case GDB_WATCHPOINT_WRITE:
1551                 dbg->arch.bp[n].type = KVMPPC_DEBUG_WATCH_WRITE;
1552                 break;
1553             case GDB_WATCHPOINT_READ:
1554                 dbg->arch.bp[n].type = KVMPPC_DEBUG_WATCH_READ;
1555                 break;
1556             case GDB_WATCHPOINT_ACCESS:
1557                 dbg->arch.bp[n].type = KVMPPC_DEBUG_WATCH_WRITE |
1558                                         KVMPPC_DEBUG_WATCH_READ;
1559                 break;
1560             default:
1561                 cpu_abort(cs, "Unsupported breakpoint type\n");
1562             }
1563             dbg->arch.bp[n].addr = hw_debug_points[n].addr;
1564         }
1565     }
1566 }
1567 
1568 static int kvm_handle_hw_breakpoint(CPUState *cs,
1569                                     struct kvm_debug_exit_arch *arch_info)
1570 {
1571     int handle = DEBUG_RETURN_GUEST;
1572     int n;
1573     int flag = 0;
1574 
1575     if (nb_hw_breakpoint + nb_hw_watchpoint > 0) {
1576         if (arch_info->status & KVMPPC_DEBUG_BREAKPOINT) {
1577             n = find_hw_breakpoint(arch_info->address, GDB_BREAKPOINT_HW);
1578             if (n >= 0) {
1579                 handle = DEBUG_RETURN_GDB;
1580             }
1581         } else if (arch_info->status & (KVMPPC_DEBUG_WATCH_READ |
1582                                         KVMPPC_DEBUG_WATCH_WRITE)) {
1583             n = find_hw_watchpoint(arch_info->address,  &flag);
1584             if (n >= 0) {
1585                 handle = DEBUG_RETURN_GDB;
1586                 cs->watchpoint_hit = &hw_watchpoint;
1587                 hw_watchpoint.vaddr = hw_debug_points[n].addr;
1588                 hw_watchpoint.flags = flag;
1589             }
1590         }
1591     }
1592     return handle;
1593 }
1594 
1595 static int kvm_handle_singlestep(void)
1596 {
1597     return DEBUG_RETURN_GDB;
1598 }
1599 
1600 static int kvm_handle_sw_breakpoint(void)
1601 {
1602     return DEBUG_RETURN_GDB;
1603 }
1604 
1605 static int kvm_handle_debug(PowerPCCPU *cpu, struct kvm_run *run)
1606 {
1607     CPUState *cs = CPU(cpu);
1608     CPUPPCState *env = &cpu->env;
1609     struct kvm_debug_exit_arch *arch_info = &run->debug.arch;
1610 
1611     if (cs->singlestep_enabled) {
1612         return kvm_handle_singlestep();
1613     }
1614 
1615     if (arch_info->status) {
1616         return kvm_handle_hw_breakpoint(cs, arch_info);
1617     }
1618 
1619     if (kvm_find_sw_breakpoint(cs, arch_info->address)) {
1620         return kvm_handle_sw_breakpoint();
1621     }
1622 
1623     /*
1624      * QEMU is not able to handle debug exception, so inject
1625      * program exception to guest;
1626      * Yes program exception NOT debug exception !!
1627      * When QEMU is using debug resources then debug exception must
1628      * be always set. To achieve this we set MSR_DE and also set
1629      * MSRP_DEP so guest cannot change MSR_DE.
1630      * When emulating debug resource for guest we want guest
1631      * to control MSR_DE (enable/disable debug interrupt on need).
1632      * Supporting both configurations are NOT possible.
1633      * So the result is that we cannot share debug resources
1634      * between QEMU and Guest on BOOKE architecture.
1635      * In the current design QEMU gets the priority over guest,
1636      * this means that if QEMU is using debug resources then guest
1637      * cannot use them;
1638      * For software breakpoint QEMU uses a privileged instruction;
1639      * So there cannot be any reason that we are here for guest
1640      * set debug exception, only possibility is guest executed a
1641      * privileged / illegal instruction and that's why we are
1642      * injecting a program interrupt.
1643      */
1644     cpu_synchronize_state(cs);
1645     /*
1646      * env->nip is PC, so increment this by 4 to use
1647      * ppc_cpu_do_interrupt(), which set srr0 = env->nip - 4.
1648      */
1649     env->nip += 4;
1650     cs->exception_index = POWERPC_EXCP_PROGRAM;
1651     env->error_code = POWERPC_EXCP_INVAL;
1652     ppc_cpu_do_interrupt(cs);
1653 
1654     return DEBUG_RETURN_GUEST;
1655 }
1656 
1657 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1658 {
1659     PowerPCCPU *cpu = POWERPC_CPU(cs);
1660     CPUPPCState *env = &cpu->env;
1661     int ret;
1662 
1663     qemu_mutex_lock_iothread();
1664 
1665     switch (run->exit_reason) {
1666     case KVM_EXIT_DCR:
1667         if (run->dcr.is_write) {
1668             trace_kvm_handle_dcr_write();
1669             ret = kvmppc_handle_dcr_write(env, run->dcr.dcrn, run->dcr.data);
1670         } else {
1671             trace_kvm_handle_dcr_read();
1672             ret = kvmppc_handle_dcr_read(env, run->dcr.dcrn, &run->dcr.data);
1673         }
1674         break;
1675     case KVM_EXIT_HLT:
1676         trace_kvm_handle_halt();
1677         ret = kvmppc_handle_halt(cpu);
1678         break;
1679 #if defined(TARGET_PPC64)
1680     case KVM_EXIT_PAPR_HCALL:
1681         trace_kvm_handle_papr_hcall(run->papr_hcall.nr);
1682         run->papr_hcall.ret = spapr_hypercall(cpu,
1683                                               run->papr_hcall.nr,
1684                                               run->papr_hcall.args);
1685         ret = 0;
1686         break;
1687 #endif
1688     case KVM_EXIT_EPR:
1689         trace_kvm_handle_epr();
1690         run->epr.epr = ldl_phys(cs->as, env->mpic_iack);
1691         ret = 0;
1692         break;
1693     case KVM_EXIT_WATCHDOG:
1694         trace_kvm_handle_watchdog_expiry();
1695         watchdog_perform_action();
1696         ret = 0;
1697         break;
1698 
1699     case KVM_EXIT_DEBUG:
1700         trace_kvm_handle_debug_exception();
1701         if (kvm_handle_debug(cpu, run)) {
1702             ret = EXCP_DEBUG;
1703             break;
1704         }
1705         /* re-enter, this exception was guest-internal */
1706         ret = 0;
1707         break;
1708 
1709 #if defined(TARGET_PPC64)
1710     case KVM_EXIT_NMI:
1711         trace_kvm_handle_nmi_exception();
1712         ret = kvm_handle_nmi(cpu, run);
1713         break;
1714 #endif
1715 
1716     default:
1717         fprintf(stderr, "KVM: unknown exit reason %d\n", run->exit_reason);
1718         ret = -1;
1719         break;
1720     }
1721 
1722     qemu_mutex_unlock_iothread();
1723     return ret;
1724 }
1725 
1726 int kvmppc_or_tsr_bits(PowerPCCPU *cpu, uint32_t tsr_bits)
1727 {
1728     CPUState *cs = CPU(cpu);
1729     uint32_t bits = tsr_bits;
1730     struct kvm_one_reg reg = {
1731         .id = KVM_REG_PPC_OR_TSR,
1732         .addr = (uintptr_t) &bits,
1733     };
1734 
1735     if (!kvm_enabled()) {
1736         return 0;
1737     }
1738 
1739     return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
1740 }
1741 
1742 int kvmppc_clear_tsr_bits(PowerPCCPU *cpu, uint32_t tsr_bits)
1743 {
1744 
1745     CPUState *cs = CPU(cpu);
1746     uint32_t bits = tsr_bits;
1747     struct kvm_one_reg reg = {
1748         .id = KVM_REG_PPC_CLEAR_TSR,
1749         .addr = (uintptr_t) &bits,
1750     };
1751 
1752     if (!kvm_enabled()) {
1753         return 0;
1754     }
1755 
1756     return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
1757 }
1758 
1759 int kvmppc_set_tcr(PowerPCCPU *cpu)
1760 {
1761     CPUState *cs = CPU(cpu);
1762     CPUPPCState *env = &cpu->env;
1763     uint32_t tcr = env->spr[SPR_BOOKE_TCR];
1764 
1765     struct kvm_one_reg reg = {
1766         .id = KVM_REG_PPC_TCR,
1767         .addr = (uintptr_t) &tcr,
1768     };
1769 
1770     if (!kvm_enabled()) {
1771         return 0;
1772     }
1773 
1774     return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
1775 }
1776 
1777 int kvmppc_booke_watchdog_enable(PowerPCCPU *cpu)
1778 {
1779     CPUState *cs = CPU(cpu);
1780     int ret;
1781 
1782     if (!kvm_enabled()) {
1783         return -1;
1784     }
1785 
1786     if (!cap_ppc_watchdog) {
1787         printf("warning: KVM does not support watchdog");
1788         return -1;
1789     }
1790 
1791     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_PPC_BOOKE_WATCHDOG, 0);
1792     if (ret < 0) {
1793         fprintf(stderr, "%s: couldn't enable KVM_CAP_PPC_BOOKE_WATCHDOG: %s\n",
1794                 __func__, strerror(-ret));
1795         return ret;
1796     }
1797 
1798     return ret;
1799 }
1800 
1801 static int read_cpuinfo(const char *field, char *value, int len)
1802 {
1803     FILE *f;
1804     int ret = -1;
1805     int field_len = strlen(field);
1806     char line[512];
1807 
1808     f = fopen("/proc/cpuinfo", "r");
1809     if (!f) {
1810         return -1;
1811     }
1812 
1813     do {
1814         if (!fgets(line, sizeof(line), f)) {
1815             break;
1816         }
1817         if (!strncmp(line, field, field_len)) {
1818             pstrcpy(value, len, line);
1819             ret = 0;
1820             break;
1821         }
1822     } while (*line);
1823 
1824     fclose(f);
1825 
1826     return ret;
1827 }
1828 
1829 static uint32_t kvmppc_get_tbfreq_procfs(void)
1830 {
1831     char line[512];
1832     char *ns;
1833     uint32_t tbfreq_fallback = NANOSECONDS_PER_SECOND;
1834     uint32_t tbfreq_procfs;
1835 
1836     if (read_cpuinfo("timebase", line, sizeof(line))) {
1837         return tbfreq_fallback;
1838     }
1839 
1840     ns = strchr(line, ':');
1841     if (!ns) {
1842         return tbfreq_fallback;
1843     }
1844 
1845     tbfreq_procfs = atoi(++ns);
1846 
1847     /* 0 is certainly not acceptable by the guest, return fallback value */
1848     return tbfreq_procfs ? tbfreq_procfs : tbfreq_fallback;
1849 }
1850 
1851 uint32_t kvmppc_get_tbfreq(void)
1852 {
1853     static uint32_t cached_tbfreq;
1854 
1855     if (!cached_tbfreq) {
1856         cached_tbfreq = kvmppc_get_tbfreq_procfs();
1857     }
1858 
1859     return cached_tbfreq;
1860 }
1861 
1862 bool kvmppc_get_host_serial(char **value)
1863 {
1864     return g_file_get_contents("/proc/device-tree/system-id", value, NULL,
1865                                NULL);
1866 }
1867 
1868 bool kvmppc_get_host_model(char **value)
1869 {
1870     return g_file_get_contents("/proc/device-tree/model", value, NULL, NULL);
1871 }
1872 
1873 /* Try to find a device tree node for a CPU with clock-frequency property */
1874 static int kvmppc_find_cpu_dt(char *buf, int buf_len)
1875 {
1876     struct dirent *dirp;
1877     DIR *dp;
1878 
1879     dp = opendir(PROC_DEVTREE_CPU);
1880     if (!dp) {
1881         printf("Can't open directory " PROC_DEVTREE_CPU "\n");
1882         return -1;
1883     }
1884 
1885     buf[0] = '\0';
1886     while ((dirp = readdir(dp)) != NULL) {
1887         FILE *f;
1888 
1889         /* Don't accidentally read from the current and parent directories */
1890         if (strcmp(dirp->d_name, ".") == 0 || strcmp(dirp->d_name, "..") == 0) {
1891             continue;
1892         }
1893 
1894         snprintf(buf, buf_len, "%s%s/clock-frequency", PROC_DEVTREE_CPU,
1895                  dirp->d_name);
1896         f = fopen(buf, "r");
1897         if (f) {
1898             snprintf(buf, buf_len, "%s%s", PROC_DEVTREE_CPU, dirp->d_name);
1899             fclose(f);
1900             break;
1901         }
1902         buf[0] = '\0';
1903     }
1904     closedir(dp);
1905     if (buf[0] == '\0') {
1906         printf("Unknown host!\n");
1907         return -1;
1908     }
1909 
1910     return 0;
1911 }
1912 
1913 static uint64_t kvmppc_read_int_dt(const char *filename)
1914 {
1915     union {
1916         uint32_t v32;
1917         uint64_t v64;
1918     } u;
1919     FILE *f;
1920     int len;
1921 
1922     f = fopen(filename, "rb");
1923     if (!f) {
1924         return -1;
1925     }
1926 
1927     len = fread(&u, 1, sizeof(u), f);
1928     fclose(f);
1929     switch (len) {
1930     case 4:
1931         /* property is a 32-bit quantity */
1932         return be32_to_cpu(u.v32);
1933     case 8:
1934         return be64_to_cpu(u.v64);
1935     }
1936 
1937     return 0;
1938 }
1939 
1940 /*
1941  * Read a CPU node property from the host device tree that's a single
1942  * integer (32-bit or 64-bit).  Returns 0 if anything goes wrong
1943  * (can't find or open the property, or doesn't understand the format)
1944  */
1945 static uint64_t kvmppc_read_int_cpu_dt(const char *propname)
1946 {
1947     char buf[PATH_MAX], *tmp;
1948     uint64_t val;
1949 
1950     if (kvmppc_find_cpu_dt(buf, sizeof(buf))) {
1951         return -1;
1952     }
1953 
1954     tmp = g_strdup_printf("%s/%s", buf, propname);
1955     val = kvmppc_read_int_dt(tmp);
1956     g_free(tmp);
1957 
1958     return val;
1959 }
1960 
1961 uint64_t kvmppc_get_clockfreq(void)
1962 {
1963     return kvmppc_read_int_cpu_dt("clock-frequency");
1964 }
1965 
1966 static int kvmppc_get_dec_bits(void)
1967 {
1968     int nr_bits = kvmppc_read_int_cpu_dt("ibm,dec-bits");
1969 
1970     if (nr_bits > 0) {
1971         return nr_bits;
1972     }
1973     return 0;
1974 }
1975 
1976 static int kvmppc_get_pvinfo(CPUPPCState *env, struct kvm_ppc_pvinfo *pvinfo)
1977 {
1978     CPUState *cs = env_cpu(env);
1979 
1980     if (kvm_vm_check_extension(cs->kvm_state, KVM_CAP_PPC_GET_PVINFO) &&
1981         !kvm_vm_ioctl(cs->kvm_state, KVM_PPC_GET_PVINFO, pvinfo)) {
1982         return 0;
1983     }
1984 
1985     return 1;
1986 }
1987 
1988 int kvmppc_get_hasidle(CPUPPCState *env)
1989 {
1990     struct kvm_ppc_pvinfo pvinfo;
1991 
1992     if (!kvmppc_get_pvinfo(env, &pvinfo) &&
1993         (pvinfo.flags & KVM_PPC_PVINFO_FLAGS_EV_IDLE)) {
1994         return 1;
1995     }
1996 
1997     return 0;
1998 }
1999 
2000 int kvmppc_get_hypercall(CPUPPCState *env, uint8_t *buf, int buf_len)
2001 {
2002     uint32_t *hc = (uint32_t *)buf;
2003     struct kvm_ppc_pvinfo pvinfo;
2004 
2005     if (!kvmppc_get_pvinfo(env, &pvinfo)) {
2006         memcpy(buf, pvinfo.hcall, buf_len);
2007         return 0;
2008     }
2009 
2010     /*
2011      * Fallback to always fail hypercalls regardless of endianness:
2012      *
2013      *     tdi 0,r0,72 (becomes b .+8 in wrong endian, nop in good endian)
2014      *     li r3, -1
2015      *     b .+8       (becomes nop in wrong endian)
2016      *     bswap32(li r3, -1)
2017      */
2018 
2019     hc[0] = cpu_to_be32(0x08000048);
2020     hc[1] = cpu_to_be32(0x3860ffff);
2021     hc[2] = cpu_to_be32(0x48000008);
2022     hc[3] = cpu_to_be32(bswap32(0x3860ffff));
2023 
2024     return 1;
2025 }
2026 
2027 static inline int kvmppc_enable_hcall(KVMState *s, target_ulong hcall)
2028 {
2029     return kvm_vm_enable_cap(s, KVM_CAP_PPC_ENABLE_HCALL, 0, hcall, 1);
2030 }
2031 
2032 void kvmppc_enable_logical_ci_hcalls(void)
2033 {
2034     /*
2035      * FIXME: it would be nice if we could detect the cases where
2036      * we're using a device which requires the in kernel
2037      * implementation of these hcalls, but the kernel lacks them and
2038      * produce a warning.
2039      */
2040     kvmppc_enable_hcall(kvm_state, H_LOGICAL_CI_LOAD);
2041     kvmppc_enable_hcall(kvm_state, H_LOGICAL_CI_STORE);
2042 }
2043 
2044 void kvmppc_enable_set_mode_hcall(void)
2045 {
2046     kvmppc_enable_hcall(kvm_state, H_SET_MODE);
2047 }
2048 
2049 void kvmppc_enable_clear_ref_mod_hcalls(void)
2050 {
2051     kvmppc_enable_hcall(kvm_state, H_CLEAR_REF);
2052     kvmppc_enable_hcall(kvm_state, H_CLEAR_MOD);
2053 }
2054 
2055 void kvmppc_enable_h_page_init(void)
2056 {
2057     kvmppc_enable_hcall(kvm_state, H_PAGE_INIT);
2058 }
2059 
2060 void kvmppc_enable_h_rpt_invalidate(void)
2061 {
2062     kvmppc_enable_hcall(kvm_state, H_RPT_INVALIDATE);
2063 }
2064 
2065 void kvmppc_set_papr(PowerPCCPU *cpu)
2066 {
2067     CPUState *cs = CPU(cpu);
2068     int ret;
2069 
2070     if (!kvm_enabled()) {
2071         return;
2072     }
2073 
2074     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_PPC_PAPR, 0);
2075     if (ret) {
2076         error_report("This vCPU type or KVM version does not support PAPR");
2077         exit(1);
2078     }
2079 
2080     /*
2081      * Update the capability flag so we sync the right information
2082      * with kvm
2083      */
2084     cap_papr = 1;
2085 }
2086 
2087 int kvmppc_set_compat(PowerPCCPU *cpu, uint32_t compat_pvr)
2088 {
2089     return kvm_set_one_reg(CPU(cpu), KVM_REG_PPC_ARCH_COMPAT, &compat_pvr);
2090 }
2091 
2092 void kvmppc_set_mpic_proxy(PowerPCCPU *cpu, int mpic_proxy)
2093 {
2094     CPUState *cs = CPU(cpu);
2095     int ret;
2096 
2097     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_PPC_EPR, 0, mpic_proxy);
2098     if (ret && mpic_proxy) {
2099         error_report("This KVM version does not support EPR");
2100         exit(1);
2101     }
2102 }
2103 
2104 bool kvmppc_get_fwnmi(void)
2105 {
2106     return cap_fwnmi;
2107 }
2108 
2109 int kvmppc_set_fwnmi(PowerPCCPU *cpu)
2110 {
2111     CPUState *cs = CPU(cpu);
2112 
2113     return kvm_vcpu_enable_cap(cs, KVM_CAP_PPC_FWNMI, 0);
2114 }
2115 
2116 int kvmppc_smt_threads(void)
2117 {
2118     return cap_ppc_smt ? cap_ppc_smt : 1;
2119 }
2120 
2121 int kvmppc_set_smt_threads(int smt)
2122 {
2123     int ret;
2124 
2125     ret = kvm_vm_enable_cap(kvm_state, KVM_CAP_PPC_SMT, 0, smt, 0);
2126     if (!ret) {
2127         cap_ppc_smt = smt;
2128     }
2129     return ret;
2130 }
2131 
2132 void kvmppc_error_append_smt_possible_hint(Error *const *errp)
2133 {
2134     int i;
2135     GString *g;
2136     char *s;
2137 
2138     assert(kvm_enabled());
2139     if (cap_ppc_smt_possible) {
2140         g = g_string_new("Available VSMT modes:");
2141         for (i = 63; i >= 0; i--) {
2142             if ((1UL << i) & cap_ppc_smt_possible) {
2143                 g_string_append_printf(g, " %lu", (1UL << i));
2144             }
2145         }
2146         s = g_string_free(g, false);
2147         error_append_hint(errp, "%s.\n", s);
2148         g_free(s);
2149     } else {
2150         error_append_hint(errp,
2151                           "This KVM seems to be too old to support VSMT.\n");
2152     }
2153 }
2154 
2155 
2156 #ifdef TARGET_PPC64
2157 uint64_t kvmppc_vrma_limit(unsigned int hash_shift)
2158 {
2159     struct kvm_ppc_smmu_info info;
2160     long rampagesize, best_page_shift;
2161     int i;
2162 
2163     /*
2164      * Find the largest hardware supported page size that's less than
2165      * or equal to the (logical) backing page size of guest RAM
2166      */
2167     kvm_get_smmu_info(&info, &error_fatal);
2168     rampagesize = qemu_minrampagesize();
2169     best_page_shift = 0;
2170 
2171     for (i = 0; i < KVM_PPC_PAGE_SIZES_MAX_SZ; i++) {
2172         struct kvm_ppc_one_seg_page_size *sps = &info.sps[i];
2173 
2174         if (!sps->page_shift) {
2175             continue;
2176         }
2177 
2178         if ((sps->page_shift > best_page_shift)
2179             && ((1UL << sps->page_shift) <= rampagesize)) {
2180             best_page_shift = sps->page_shift;
2181         }
2182     }
2183 
2184     return 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 bool kvmppc_has_cap_xive(void)
2482 {
2483     return cap_xive;
2484 }
2485 
2486 static void kvmppc_get_cpu_characteristics(KVMState *s)
2487 {
2488     struct kvm_ppc_cpu_char c;
2489     int ret;
2490 
2491     /* Assume broken */
2492     cap_ppc_safe_cache = 0;
2493     cap_ppc_safe_bounds_check = 0;
2494     cap_ppc_safe_indirect_branch = 0;
2495 
2496     ret = kvm_vm_check_extension(s, KVM_CAP_PPC_GET_CPU_CHAR);
2497     if (!ret) {
2498         return;
2499     }
2500     ret = kvm_vm_ioctl(s, KVM_PPC_GET_CPU_CHAR, &c);
2501     if (ret < 0) {
2502         return;
2503     }
2504 
2505     cap_ppc_safe_cache = parse_cap_ppc_safe_cache(c);
2506     cap_ppc_safe_bounds_check = parse_cap_ppc_safe_bounds_check(c);
2507     cap_ppc_safe_indirect_branch = parse_cap_ppc_safe_indirect_branch(c);
2508     cap_ppc_count_cache_flush_assist =
2509         parse_cap_ppc_count_cache_flush_assist(c);
2510 }
2511 
2512 int kvmppc_get_cap_safe_cache(void)
2513 {
2514     return cap_ppc_safe_cache;
2515 }
2516 
2517 int kvmppc_get_cap_safe_bounds_check(void)
2518 {
2519     return cap_ppc_safe_bounds_check;
2520 }
2521 
2522 int kvmppc_get_cap_safe_indirect_branch(void)
2523 {
2524     return cap_ppc_safe_indirect_branch;
2525 }
2526 
2527 int kvmppc_get_cap_count_cache_flush_assist(void)
2528 {
2529     return cap_ppc_count_cache_flush_assist;
2530 }
2531 
2532 bool kvmppc_has_cap_nested_kvm_hv(void)
2533 {
2534     return !!cap_ppc_nested_kvm_hv;
2535 }
2536 
2537 int kvmppc_set_cap_nested_kvm_hv(int enable)
2538 {
2539     return kvm_vm_enable_cap(kvm_state, KVM_CAP_PPC_NESTED_HV, 0, enable);
2540 }
2541 
2542 bool kvmppc_has_cap_spapr_vfio(void)
2543 {
2544     return cap_spapr_vfio;
2545 }
2546 
2547 int kvmppc_get_cap_large_decr(void)
2548 {
2549     return cap_large_decr;
2550 }
2551 
2552 int kvmppc_enable_cap_large_decr(PowerPCCPU *cpu, int enable)
2553 {
2554     CPUState *cs = CPU(cpu);
2555     uint64_t lpcr = 0;
2556 
2557     kvm_get_one_reg(cs, KVM_REG_PPC_LPCR_64, &lpcr);
2558     /* Do we need to modify the LPCR? */
2559     if (!!(lpcr & LPCR_LD) != !!enable) {
2560         if (enable) {
2561             lpcr |= LPCR_LD;
2562         } else {
2563             lpcr &= ~LPCR_LD;
2564         }
2565         kvm_set_one_reg(cs, KVM_REG_PPC_LPCR_64, &lpcr);
2566         kvm_get_one_reg(cs, KVM_REG_PPC_LPCR_64, &lpcr);
2567 
2568         if (!!(lpcr & LPCR_LD) != !!enable) {
2569             return -1;
2570         }
2571     }
2572 
2573     return 0;
2574 }
2575 
2576 int kvmppc_has_cap_rpt_invalidate(void)
2577 {
2578     return cap_rpt_invalidate;
2579 }
2580 
2581 bool kvmppc_supports_ail_3(void)
2582 {
2583     return cap_ail_mode_3;
2584 }
2585 
2586 PowerPCCPUClass *kvm_ppc_get_host_cpu_class(void)
2587 {
2588     uint32_t host_pvr = mfpvr();
2589     PowerPCCPUClass *pvr_pcc;
2590 
2591     pvr_pcc = ppc_cpu_class_by_pvr(host_pvr);
2592     if (pvr_pcc == NULL) {
2593         pvr_pcc = ppc_cpu_class_by_pvr_mask(host_pvr);
2594     }
2595 
2596     return pvr_pcc;
2597 }
2598 
2599 static void pseries_machine_class_fixup(ObjectClass *oc, void *opaque)
2600 {
2601     MachineClass *mc = MACHINE_CLASS(oc);
2602 
2603     mc->default_cpu_type = TYPE_HOST_POWERPC_CPU;
2604 }
2605 
2606 static int kvm_ppc_register_host_cpu_type(void)
2607 {
2608     TypeInfo type_info = {
2609         .name = TYPE_HOST_POWERPC_CPU,
2610         .class_init = kvmppc_host_cpu_class_init,
2611     };
2612     PowerPCCPUClass *pvr_pcc;
2613     ObjectClass *oc;
2614     DeviceClass *dc;
2615     int i;
2616 
2617     pvr_pcc = kvm_ppc_get_host_cpu_class();
2618     if (pvr_pcc == NULL) {
2619         return -1;
2620     }
2621     type_info.parent = object_class_get_name(OBJECT_CLASS(pvr_pcc));
2622     type_register(&type_info);
2623     /* override TCG default cpu type with 'host' cpu model */
2624     object_class_foreach(pseries_machine_class_fixup, TYPE_SPAPR_MACHINE,
2625                          false, NULL);
2626 
2627     oc = object_class_by_name(type_info.name);
2628     g_assert(oc);
2629 
2630     /*
2631      * Update generic CPU family class alias (e.g. on a POWER8NVL host,
2632      * we want "POWER8" to be a "family" alias that points to the current
2633      * host CPU type, too)
2634      */
2635     dc = DEVICE_CLASS(ppc_cpu_get_family_class(pvr_pcc));
2636     for (i = 0; ppc_cpu_aliases[i].alias != NULL; i++) {
2637         if (strcasecmp(ppc_cpu_aliases[i].alias, dc->desc) == 0) {
2638             char *suffix;
2639 
2640             ppc_cpu_aliases[i].model = g_strdup(object_class_get_name(oc));
2641             suffix = strstr(ppc_cpu_aliases[i].model, POWERPC_CPU_TYPE_SUFFIX);
2642             if (suffix) {
2643                 *suffix = 0;
2644             }
2645             break;
2646         }
2647     }
2648 
2649     return 0;
2650 }
2651 
2652 int kvmppc_define_rtas_kernel_token(uint32_t token, const char *function)
2653 {
2654     struct kvm_rtas_token_args args = {
2655         .token = token,
2656     };
2657 
2658     if (!kvm_check_extension(kvm_state, KVM_CAP_PPC_RTAS)) {
2659         return -ENOENT;
2660     }
2661 
2662     strncpy(args.name, function, sizeof(args.name) - 1);
2663 
2664     return kvm_vm_ioctl(kvm_state, KVM_PPC_RTAS_DEFINE_TOKEN, &args);
2665 }
2666 
2667 int kvmppc_get_htab_fd(bool write, uint64_t index, Error **errp)
2668 {
2669     struct kvm_get_htab_fd s = {
2670         .flags = write ? KVM_GET_HTAB_WRITE : 0,
2671         .start_index = index,
2672     };
2673     int ret;
2674 
2675     if (!cap_htab_fd) {
2676         error_setg(errp, "KVM version doesn't support %s the HPT",
2677                    write ? "writing" : "reading");
2678         return -ENOTSUP;
2679     }
2680 
2681     ret = kvm_vm_ioctl(kvm_state, KVM_PPC_GET_HTAB_FD, &s);
2682     if (ret < 0) {
2683         error_setg(errp, "Unable to open fd for %s HPT %s KVM: %s",
2684                    write ? "writing" : "reading", write ? "to" : "from",
2685                    strerror(errno));
2686         return -errno;
2687     }
2688 
2689     return ret;
2690 }
2691 
2692 int kvmppc_save_htab(QEMUFile *f, int fd, size_t bufsize, int64_t max_ns)
2693 {
2694     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2695     uint8_t buf[bufsize];
2696     ssize_t rc;
2697 
2698     do {
2699         rc = read(fd, buf, bufsize);
2700         if (rc < 0) {
2701             fprintf(stderr, "Error reading data from KVM HTAB fd: %s\n",
2702                     strerror(errno));
2703             return rc;
2704         } else if (rc) {
2705             uint8_t *buffer = buf;
2706             ssize_t n = rc;
2707             while (n) {
2708                 struct kvm_get_htab_header *head =
2709                     (struct kvm_get_htab_header *) buffer;
2710                 size_t chunksize = sizeof(*head) +
2711                      HASH_PTE_SIZE_64 * head->n_valid;
2712 
2713                 qemu_put_be32(f, head->index);
2714                 qemu_put_be16(f, head->n_valid);
2715                 qemu_put_be16(f, head->n_invalid);
2716                 qemu_put_buffer(f, (void *)(head + 1),
2717                                 HASH_PTE_SIZE_64 * head->n_valid);
2718 
2719                 buffer += chunksize;
2720                 n -= chunksize;
2721             }
2722         }
2723     } while ((rc != 0)
2724              && ((max_ns < 0) ||
2725                  ((qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) < max_ns)));
2726 
2727     return (rc == 0) ? 1 : 0;
2728 }
2729 
2730 int kvmppc_load_htab_chunk(QEMUFile *f, int fd, uint32_t index,
2731                            uint16_t n_valid, uint16_t n_invalid, Error **errp)
2732 {
2733     struct kvm_get_htab_header *buf;
2734     size_t chunksize = sizeof(*buf) + n_valid * HASH_PTE_SIZE_64;
2735     ssize_t rc;
2736 
2737     buf = alloca(chunksize);
2738     buf->index = index;
2739     buf->n_valid = n_valid;
2740     buf->n_invalid = n_invalid;
2741 
2742     qemu_get_buffer(f, (void *)(buf + 1), HASH_PTE_SIZE_64 * n_valid);
2743 
2744     rc = write(fd, buf, chunksize);
2745     if (rc < 0) {
2746         error_setg_errno(errp, errno, "Error writing the KVM hash table");
2747         return -errno;
2748     }
2749     if (rc != chunksize) {
2750         /* We should never get a short write on a single chunk */
2751         error_setg(errp, "Short write while restoring the KVM hash table");
2752         return -ENOSPC;
2753     }
2754     return 0;
2755 }
2756 
2757 bool kvm_arch_stop_on_emulation_error(CPUState *cpu)
2758 {
2759     return true;
2760 }
2761 
2762 void kvm_arch_init_irq_routing(KVMState *s)
2763 {
2764 }
2765 
2766 void kvmppc_read_hptes(ppc_hash_pte64_t *hptes, hwaddr ptex, int n)
2767 {
2768     int fd, rc;
2769     int i;
2770 
2771     fd = kvmppc_get_htab_fd(false, ptex, &error_abort);
2772 
2773     i = 0;
2774     while (i < n) {
2775         struct kvm_get_htab_header *hdr;
2776         int m = n < HPTES_PER_GROUP ? n : HPTES_PER_GROUP;
2777         char buf[sizeof(*hdr) + m * HASH_PTE_SIZE_64];
2778 
2779         rc = read(fd, buf, sizeof(buf));
2780         if (rc < 0) {
2781             hw_error("kvmppc_read_hptes: Unable to read HPTEs");
2782         }
2783 
2784         hdr = (struct kvm_get_htab_header *)buf;
2785         while ((i < n) && ((char *)hdr < (buf + rc))) {
2786             int invalid = hdr->n_invalid, valid = hdr->n_valid;
2787 
2788             if (hdr->index != (ptex + i)) {
2789                 hw_error("kvmppc_read_hptes: Unexpected HPTE index %"PRIu32
2790                          " != (%"HWADDR_PRIu" + %d", hdr->index, ptex, i);
2791             }
2792 
2793             if (n - i < valid) {
2794                 valid = n - i;
2795             }
2796             memcpy(hptes + i, hdr + 1, HASH_PTE_SIZE_64 * valid);
2797             i += valid;
2798 
2799             if ((n - i) < invalid) {
2800                 invalid = n - i;
2801             }
2802             memset(hptes + i, 0, invalid * HASH_PTE_SIZE_64);
2803             i += invalid;
2804 
2805             hdr = (struct kvm_get_htab_header *)
2806                 ((char *)(hdr + 1) + HASH_PTE_SIZE_64 * hdr->n_valid);
2807         }
2808     }
2809 
2810     close(fd);
2811 }
2812 
2813 void kvmppc_write_hpte(hwaddr ptex, uint64_t pte0, uint64_t pte1)
2814 {
2815     int fd, rc;
2816     struct {
2817         struct kvm_get_htab_header hdr;
2818         uint64_t pte0;
2819         uint64_t pte1;
2820     } buf;
2821 
2822     fd = kvmppc_get_htab_fd(true, 0 /* Ignored */, &error_abort);
2823 
2824     buf.hdr.n_valid = 1;
2825     buf.hdr.n_invalid = 0;
2826     buf.hdr.index = ptex;
2827     buf.pte0 = cpu_to_be64(pte0);
2828     buf.pte1 = cpu_to_be64(pte1);
2829 
2830     rc = write(fd, &buf, sizeof(buf));
2831     if (rc != sizeof(buf)) {
2832         hw_error("kvmppc_write_hpte: Unable to update KVM HPT");
2833     }
2834     close(fd);
2835 }
2836 
2837 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
2838                              uint64_t address, uint32_t data, PCIDevice *dev)
2839 {
2840     return 0;
2841 }
2842 
2843 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
2844                                 int vector, PCIDevice *dev)
2845 {
2846     return 0;
2847 }
2848 
2849 int kvm_arch_release_virq_post(int virq)
2850 {
2851     return 0;
2852 }
2853 
2854 int kvm_arch_msi_data_to_gsi(uint32_t data)
2855 {
2856     return data & 0xffff;
2857 }
2858 
2859 #if defined(TARGET_PPC64)
2860 int kvm_handle_nmi(PowerPCCPU *cpu, struct kvm_run *run)
2861 {
2862     uint16_t flags = run->flags & KVM_RUN_PPC_NMI_DISP_MASK;
2863 
2864     cpu_synchronize_state(CPU(cpu));
2865 
2866     spapr_mce_req_event(cpu, flags == KVM_RUN_PPC_NMI_DISP_FULLY_RECOV);
2867 
2868     return 0;
2869 }
2870 #endif
2871 
2872 int kvmppc_enable_hwrng(void)
2873 {
2874     if (!kvm_enabled() || !kvm_check_extension(kvm_state, KVM_CAP_PPC_HWRNG)) {
2875         return -1;
2876     }
2877 
2878     return kvmppc_enable_hcall(kvm_state, H_RANDOM);
2879 }
2880 
2881 void kvmppc_check_papr_resize_hpt(Error **errp)
2882 {
2883     if (!kvm_enabled()) {
2884         return; /* No KVM, we're good */
2885     }
2886 
2887     if (cap_resize_hpt) {
2888         return; /* Kernel has explicit support, we're good */
2889     }
2890 
2891     /* Otherwise fallback on looking for PR KVM */
2892     if (kvmppc_is_pr(kvm_state)) {
2893         return;
2894     }
2895 
2896     error_setg(errp,
2897                "Hash page table resizing not available with this KVM version");
2898 }
2899 
2900 int kvmppc_resize_hpt_prepare(PowerPCCPU *cpu, target_ulong flags, int shift)
2901 {
2902     CPUState *cs = CPU(cpu);
2903     struct kvm_ppc_resize_hpt rhpt = {
2904         .flags = flags,
2905         .shift = shift,
2906     };
2907 
2908     if (!cap_resize_hpt) {
2909         return -ENOSYS;
2910     }
2911 
2912     return kvm_vm_ioctl(cs->kvm_state, KVM_PPC_RESIZE_HPT_PREPARE, &rhpt);
2913 }
2914 
2915 int kvmppc_resize_hpt_commit(PowerPCCPU *cpu, target_ulong flags, int shift)
2916 {
2917     CPUState *cs = CPU(cpu);
2918     struct kvm_ppc_resize_hpt rhpt = {
2919         .flags = flags,
2920         .shift = shift,
2921     };
2922 
2923     if (!cap_resize_hpt) {
2924         return -ENOSYS;
2925     }
2926 
2927     return kvm_vm_ioctl(cs->kvm_state, KVM_PPC_RESIZE_HPT_COMMIT, &rhpt);
2928 }
2929 
2930 /*
2931  * This is a helper function to detect a post migration scenario
2932  * in which a guest, running as KVM-HV, freezes in cpu_post_load because
2933  * the guest kernel can't handle a PVR value other than the actual host
2934  * PVR in KVM_SET_SREGS, even if pvr_match() returns true.
2935  *
2936  * If we don't have cap_ppc_pvr_compat and we're not running in PR
2937  * (so, we're HV), return true. The workaround itself is done in
2938  * cpu_post_load.
2939  *
2940  * The order here is important: we'll only check for KVM PR as a
2941  * fallback if the guest kernel can't handle the situation itself.
2942  * We need to avoid as much as possible querying the running KVM type
2943  * in QEMU level.
2944  */
2945 bool kvmppc_pvr_workaround_required(PowerPCCPU *cpu)
2946 {
2947     CPUState *cs = CPU(cpu);
2948 
2949     if (!kvm_enabled()) {
2950         return false;
2951     }
2952 
2953     if (cap_ppc_pvr_compat) {
2954         return false;
2955     }
2956 
2957     return !kvmppc_is_pr(cs->kvm_state);
2958 }
2959 
2960 void kvmppc_set_reg_ppc_online(PowerPCCPU *cpu, unsigned int online)
2961 {
2962     CPUState *cs = CPU(cpu);
2963 
2964     if (kvm_enabled()) {
2965         kvm_set_one_reg(cs, KVM_REG_PPC_ONLINE, &online);
2966     }
2967 }
2968 
2969 void kvmppc_set_reg_tb_offset(PowerPCCPU *cpu, int64_t tb_offset)
2970 {
2971     CPUState *cs = CPU(cpu);
2972 
2973     if (kvm_enabled()) {
2974         kvm_set_one_reg(cs, KVM_REG_PPC_TB_OFFSET, &tb_offset);
2975     }
2976 }
2977 
2978 bool kvm_arch_cpu_check_are_resettable(void)
2979 {
2980     return true;
2981 }
2982 
2983 void kvm_arch_accel_class_init(ObjectClass *oc)
2984 {
2985 }
2986