xref: /openbmc/qemu/target/ppc/kvm.c (revision 64547a3b)
1 /*
2  * PowerPC implementation of KVM hooks
3  *
4  * Copyright IBM Corp. 2007
5  * Copyright (C) 2011 Freescale Semiconductor, Inc.
6  *
7  * Authors:
8  *  Jerone Young <jyoung5@us.ibm.com>
9  *  Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
10  *  Hollis Blanchard <hollisb@us.ibm.com>
11  *
12  * This work is licensed under the terms of the GNU GPL, version 2 or later.
13  * See the COPYING file in the top-level directory.
14  *
15  */
16 
17 #include "qemu/osdep.h"
18 #include <dirent.h>
19 #include <sys/ioctl.h>
20 #include <sys/vfs.h>
21 
22 #include <linux/kvm.h>
23 
24 #include "qemu-common.h"
25 #include "qapi/error.h"
26 #include "qemu/error-report.h"
27 #include "cpu.h"
28 #include "cpu-models.h"
29 #include "qemu/timer.h"
30 #include "sysemu/hw_accel.h"
31 #include "kvm_ppc.h"
32 #include "sysemu/cpus.h"
33 #include "sysemu/device_tree.h"
34 #include "mmu-hash64.h"
35 
36 #include "hw/sysbus.h"
37 #include "hw/ppc/spapr.h"
38 #include "hw/ppc/spapr_cpu_core.h"
39 #include "hw/hw.h"
40 #include "hw/ppc/ppc.h"
41 #include "migration/qemu-file-types.h"
42 #include "sysemu/watchdog.h"
43 #include "trace.h"
44 #include "exec/gdbstub.h"
45 #include "exec/memattrs.h"
46 #include "exec/ram_addr.h"
47 #include "sysemu/hostmem.h"
48 #include "qemu/cutils.h"
49 #include "qemu/main-loop.h"
50 #include "qemu/mmap-alloc.h"
51 #include "elf.h"
52 #include "sysemu/kvm_int.h"
53 
54 #define PROC_DEVTREE_CPU      "/proc/device-tree/cpus/"
55 
56 #define DEBUG_RETURN_GUEST 0
57 #define DEBUG_RETURN_GDB   1
58 
59 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
60     KVM_CAP_LAST_INFO
61 };
62 
63 static int cap_interrupt_unset;
64 static int cap_segstate;
65 static int cap_booke_sregs;
66 static int cap_ppc_smt;
67 static int cap_ppc_smt_possible;
68 static int cap_spapr_tce;
69 static int cap_spapr_tce_64;
70 static int cap_spapr_multitce;
71 static int cap_spapr_vfio;
72 static int cap_hior;
73 static int cap_one_reg;
74 static int cap_epr;
75 static int cap_ppc_watchdog;
76 static int cap_papr;
77 static int cap_htab_fd;
78 static int cap_fixup_hcalls;
79 static int cap_htm;             /* Hardware transactional memory support */
80 static int cap_mmu_radix;
81 static int cap_mmu_hash_v3;
82 static int cap_xive;
83 static int cap_resize_hpt;
84 static int cap_ppc_pvr_compat;
85 static int cap_ppc_safe_cache;
86 static int cap_ppc_safe_bounds_check;
87 static int cap_ppc_safe_indirect_branch;
88 static int cap_ppc_count_cache_flush_assist;
89 static int cap_ppc_nested_kvm_hv;
90 static int cap_large_decr;
91 
92 static uint32_t debug_inst_opcode;
93 
94 /*
95  * Check whether we are running with KVM-PR (instead of KVM-HV).  This
96  * should only be used for fallback tests - generally we should use
97  * explicit capabilities for the features we want, rather than
98  * assuming what is/isn't available depending on the KVM variant.
99  */
100 static bool kvmppc_is_pr(KVMState *ks)
101 {
102     /* Assume KVM-PR if the GET_PVINFO capability is available */
103     return kvm_vm_check_extension(ks, KVM_CAP_PPC_GET_PVINFO) != 0;
104 }
105 
106 static int kvm_ppc_register_host_cpu_type(void);
107 static void kvmppc_get_cpu_characteristics(KVMState *s);
108 static int kvmppc_get_dec_bits(void);
109 
110 int kvm_arch_init(MachineState *ms, KVMState *s)
111 {
112     cap_interrupt_unset = kvm_check_extension(s, KVM_CAP_PPC_UNSET_IRQ);
113     cap_segstate = kvm_check_extension(s, KVM_CAP_PPC_SEGSTATE);
114     cap_booke_sregs = kvm_check_extension(s, KVM_CAP_PPC_BOOKE_SREGS);
115     cap_ppc_smt_possible = kvm_vm_check_extension(s, KVM_CAP_PPC_SMT_POSSIBLE);
116     cap_spapr_tce = kvm_check_extension(s, KVM_CAP_SPAPR_TCE);
117     cap_spapr_tce_64 = kvm_check_extension(s, KVM_CAP_SPAPR_TCE_64);
118     cap_spapr_multitce = kvm_check_extension(s, KVM_CAP_SPAPR_MULTITCE);
119     cap_spapr_vfio = kvm_vm_check_extension(s, KVM_CAP_SPAPR_TCE_VFIO);
120     cap_one_reg = kvm_check_extension(s, KVM_CAP_ONE_REG);
121     cap_hior = kvm_check_extension(s, KVM_CAP_PPC_HIOR);
122     cap_epr = kvm_check_extension(s, KVM_CAP_PPC_EPR);
123     cap_ppc_watchdog = kvm_check_extension(s, KVM_CAP_PPC_BOOKE_WATCHDOG);
124     /*
125      * Note: we don't set cap_papr here, because this capability is
126      * only activated after this by kvmppc_set_papr()
127      */
128     cap_htab_fd = kvm_vm_check_extension(s, KVM_CAP_PPC_HTAB_FD);
129     cap_fixup_hcalls = kvm_check_extension(s, KVM_CAP_PPC_FIXUP_HCALL);
130     cap_ppc_smt = kvm_vm_check_extension(s, KVM_CAP_PPC_SMT);
131     cap_htm = kvm_vm_check_extension(s, KVM_CAP_PPC_HTM);
132     cap_mmu_radix = kvm_vm_check_extension(s, KVM_CAP_PPC_MMU_RADIX);
133     cap_mmu_hash_v3 = kvm_vm_check_extension(s, KVM_CAP_PPC_MMU_HASH_V3);
134     cap_xive = kvm_vm_check_extension(s, KVM_CAP_PPC_IRQ_XIVE);
135     cap_resize_hpt = kvm_vm_check_extension(s, KVM_CAP_SPAPR_RESIZE_HPT);
136     kvmppc_get_cpu_characteristics(s);
137     cap_ppc_nested_kvm_hv = kvm_vm_check_extension(s, KVM_CAP_PPC_NESTED_HV);
138     cap_large_decr = kvmppc_get_dec_bits();
139     /*
140      * Note: setting it to false because there is not such capability
141      * in KVM at this moment.
142      *
143      * TODO: call kvm_vm_check_extension() with the right capability
144      * after the kernel starts implementing it.
145      */
146     cap_ppc_pvr_compat = false;
147 
148     if (!kvm_check_extension(s, KVM_CAP_PPC_IRQ_LEVEL)) {
149         error_report("KVM: Host kernel doesn't have level irq capability");
150         exit(1);
151     }
152 
153     kvm_ppc_register_host_cpu_type();
154 
155     return 0;
156 }
157 
158 int kvm_arch_irqchip_create(KVMState *s)
159 {
160     return 0;
161 }
162 
163 static int kvm_arch_sync_sregs(PowerPCCPU *cpu)
164 {
165     CPUPPCState *cenv = &cpu->env;
166     CPUState *cs = CPU(cpu);
167     struct kvm_sregs sregs;
168     int ret;
169 
170     if (cenv->excp_model == POWERPC_EXCP_BOOKE) {
171         /*
172          * What we're really trying to say is "if we're on BookE, we
173          * use the native PVR for now". This is the only sane way to
174          * check it though, so we potentially confuse users that they
175          * can run BookE guests on BookS. Let's hope nobody dares
176          * enough :)
177          */
178         return 0;
179     } else {
180         if (!cap_segstate) {
181             fprintf(stderr, "kvm error: missing PVR setting capability\n");
182             return -ENOSYS;
183         }
184     }
185 
186     ret = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs);
187     if (ret) {
188         return ret;
189     }
190 
191     sregs.pvr = cenv->spr[SPR_PVR];
192     return kvm_vcpu_ioctl(cs, KVM_SET_SREGS, &sregs);
193 }
194 
195 /* Set up a shared TLB array with KVM */
196 static int kvm_booke206_tlb_init(PowerPCCPU *cpu)
197 {
198     CPUPPCState *env = &cpu->env;
199     CPUState *cs = CPU(cpu);
200     struct kvm_book3e_206_tlb_params params = {};
201     struct kvm_config_tlb cfg = {};
202     unsigned int entries = 0;
203     int ret, i;
204 
205     if (!kvm_enabled() ||
206         !kvm_check_extension(cs->kvm_state, KVM_CAP_SW_TLB)) {
207         return 0;
208     }
209 
210     assert(ARRAY_SIZE(params.tlb_sizes) == BOOKE206_MAX_TLBN);
211 
212     for (i = 0; i < BOOKE206_MAX_TLBN; i++) {
213         params.tlb_sizes[i] = booke206_tlb_size(env, i);
214         params.tlb_ways[i] = booke206_tlb_ways(env, i);
215         entries += params.tlb_sizes[i];
216     }
217 
218     assert(entries == env->nb_tlb);
219     assert(sizeof(struct kvm_book3e_206_tlb_entry) == sizeof(ppcmas_tlb_t));
220 
221     env->tlb_dirty = true;
222 
223     cfg.array = (uintptr_t)env->tlb.tlbm;
224     cfg.array_len = sizeof(ppcmas_tlb_t) * entries;
225     cfg.params = (uintptr_t)&params;
226     cfg.mmu_type = KVM_MMU_FSL_BOOKE_NOHV;
227 
228     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_SW_TLB, 0, (uintptr_t)&cfg);
229     if (ret < 0) {
230         fprintf(stderr, "%s: couldn't enable KVM_CAP_SW_TLB: %s\n",
231                 __func__, strerror(-ret));
232         return ret;
233     }
234 
235     env->kvm_sw_tlb = true;
236     return 0;
237 }
238 
239 
240 #if defined(TARGET_PPC64)
241 static void kvm_get_smmu_info(struct kvm_ppc_smmu_info *info, Error **errp)
242 {
243     int ret;
244 
245     assert(kvm_state != NULL);
246 
247     if (!kvm_check_extension(kvm_state, KVM_CAP_PPC_GET_SMMU_INFO)) {
248         error_setg(errp, "KVM doesn't expose the MMU features it supports");
249         error_append_hint(errp, "Consider switching to a newer KVM\n");
250         return;
251     }
252 
253     ret = kvm_vm_ioctl(kvm_state, KVM_PPC_GET_SMMU_INFO, info);
254     if (ret == 0) {
255         return;
256     }
257 
258     error_setg_errno(errp, -ret,
259                      "KVM failed to provide the MMU features it supports");
260 }
261 
262 struct ppc_radix_page_info *kvm_get_radix_page_info(void)
263 {
264     KVMState *s = KVM_STATE(current_accel());
265     struct ppc_radix_page_info *radix_page_info;
266     struct kvm_ppc_rmmu_info rmmu_info;
267     int i;
268 
269     if (!kvm_check_extension(s, KVM_CAP_PPC_MMU_RADIX)) {
270         return NULL;
271     }
272     if (kvm_vm_ioctl(s, KVM_PPC_GET_RMMU_INFO, &rmmu_info)) {
273         return NULL;
274     }
275     radix_page_info = g_malloc0(sizeof(*radix_page_info));
276     radix_page_info->count = 0;
277     for (i = 0; i < PPC_PAGE_SIZES_MAX_SZ; i++) {
278         if (rmmu_info.ap_encodings[i]) {
279             radix_page_info->entries[i] = rmmu_info.ap_encodings[i];
280             radix_page_info->count++;
281         }
282     }
283     return radix_page_info;
284 }
285 
286 target_ulong kvmppc_configure_v3_mmu(PowerPCCPU *cpu,
287                                      bool radix, bool gtse,
288                                      uint64_t proc_tbl)
289 {
290     CPUState *cs = CPU(cpu);
291     int ret;
292     uint64_t flags = 0;
293     struct kvm_ppc_mmuv3_cfg cfg = {
294         .process_table = proc_tbl,
295     };
296 
297     if (radix) {
298         flags |= KVM_PPC_MMUV3_RADIX;
299     }
300     if (gtse) {
301         flags |= KVM_PPC_MMUV3_GTSE;
302     }
303     cfg.flags = flags;
304     ret = kvm_vm_ioctl(cs->kvm_state, KVM_PPC_CONFIGURE_V3_MMU, &cfg);
305     switch (ret) {
306     case 0:
307         return H_SUCCESS;
308     case -EINVAL:
309         return H_PARAMETER;
310     case -ENODEV:
311         return H_NOT_AVAILABLE;
312     default:
313         return H_HARDWARE;
314     }
315 }
316 
317 bool kvmppc_hpt_needs_host_contiguous_pages(void)
318 {
319     static struct kvm_ppc_smmu_info smmu_info;
320 
321     if (!kvm_enabled()) {
322         return false;
323     }
324 
325     kvm_get_smmu_info(&smmu_info, &error_fatal);
326     return !!(smmu_info.flags & KVM_PPC_PAGE_SIZES_REAL);
327 }
328 
329 void kvm_check_mmu(PowerPCCPU *cpu, Error **errp)
330 {
331     struct kvm_ppc_smmu_info smmu_info;
332     int iq, ik, jq, jk;
333     Error *local_err = NULL;
334 
335     /* For now, we only have anything to check on hash64 MMUs */
336     if (!cpu->hash64_opts || !kvm_enabled()) {
337         return;
338     }
339 
340     kvm_get_smmu_info(&smmu_info, &local_err);
341     if (local_err) {
342         error_propagate(errp, local_err);
343         return;
344     }
345 
346     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)
347         && !(smmu_info.flags & KVM_PPC_1T_SEGMENTS)) {
348         error_setg(errp,
349                    "KVM does not support 1TiB segments which guest expects");
350         return;
351     }
352 
353     if (smmu_info.slb_size < cpu->hash64_opts->slb_size) {
354         error_setg(errp, "KVM only supports %u SLB entries, but guest needs %u",
355                    smmu_info.slb_size, cpu->hash64_opts->slb_size);
356         return;
357     }
358 
359     /*
360      * Verify that every pagesize supported by the cpu model is
361      * supported by KVM with the same encodings
362      */
363     for (iq = 0; iq < ARRAY_SIZE(cpu->hash64_opts->sps); iq++) {
364         PPCHash64SegmentPageSizes *qsps = &cpu->hash64_opts->sps[iq];
365         struct kvm_ppc_one_seg_page_size *ksps;
366 
367         for (ik = 0; ik < ARRAY_SIZE(smmu_info.sps); ik++) {
368             if (qsps->page_shift == smmu_info.sps[ik].page_shift) {
369                 break;
370             }
371         }
372         if (ik >= ARRAY_SIZE(smmu_info.sps)) {
373             error_setg(errp, "KVM doesn't support for base page shift %u",
374                        qsps->page_shift);
375             return;
376         }
377 
378         ksps = &smmu_info.sps[ik];
379         if (ksps->slb_enc != qsps->slb_enc) {
380             error_setg(errp,
381 "KVM uses SLB encoding 0x%x for page shift %u, but guest expects 0x%x",
382                        ksps->slb_enc, ksps->page_shift, qsps->slb_enc);
383             return;
384         }
385 
386         for (jq = 0; jq < ARRAY_SIZE(qsps->enc); jq++) {
387             for (jk = 0; jk < ARRAY_SIZE(ksps->enc); jk++) {
388                 if (qsps->enc[jq].page_shift == ksps->enc[jk].page_shift) {
389                     break;
390                 }
391             }
392 
393             if (jk >= ARRAY_SIZE(ksps->enc)) {
394                 error_setg(errp, "KVM doesn't support page shift %u/%u",
395                            qsps->enc[jq].page_shift, qsps->page_shift);
396                 return;
397             }
398             if (qsps->enc[jq].pte_enc != ksps->enc[jk].pte_enc) {
399                 error_setg(errp,
400 "KVM uses PTE encoding 0x%x for page shift %u/%u, but guest expects 0x%x",
401                            ksps->enc[jk].pte_enc, qsps->enc[jq].page_shift,
402                            qsps->page_shift, qsps->enc[jq].pte_enc);
403                 return;
404             }
405         }
406     }
407 
408     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
409         /*
410          * Mostly what guest pagesizes we can use are related to the
411          * host pages used to map guest RAM, which is handled in the
412          * platform code. Cache-Inhibited largepages (64k) however are
413          * used for I/O, so if they're mapped to the host at all it
414          * will be a normal mapping, not a special hugepage one used
415          * for RAM.
416          */
417         if (qemu_real_host_page_size < 0x10000) {
418             error_setg(errp,
419                        "KVM can't supply 64kiB CI pages, which guest expects");
420         }
421     }
422 }
423 #endif /* !defined (TARGET_PPC64) */
424 
425 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
426 {
427     return POWERPC_CPU(cpu)->vcpu_id;
428 }
429 
430 /*
431  * e500 supports 2 h/w breakpoint and 2 watchpoint.  book3s supports
432  * only 1 watchpoint, so array size of 4 is sufficient for now.
433  */
434 #define MAX_HW_BKPTS 4
435 
436 static struct HWBreakpoint {
437     target_ulong addr;
438     int type;
439 } hw_debug_points[MAX_HW_BKPTS];
440 
441 static CPUWatchpoint hw_watchpoint;
442 
443 /* Default there is no breakpoint and watchpoint supported */
444 static int max_hw_breakpoint;
445 static int max_hw_watchpoint;
446 static int nb_hw_breakpoint;
447 static int nb_hw_watchpoint;
448 
449 static void kvmppc_hw_debug_points_init(CPUPPCState *cenv)
450 {
451     if (cenv->excp_model == POWERPC_EXCP_BOOKE) {
452         max_hw_breakpoint = 2;
453         max_hw_watchpoint = 2;
454     }
455 
456     if ((max_hw_breakpoint + max_hw_watchpoint) > MAX_HW_BKPTS) {
457         fprintf(stderr, "Error initializing h/w breakpoints\n");
458         return;
459     }
460 }
461 
462 int kvm_arch_init_vcpu(CPUState *cs)
463 {
464     PowerPCCPU *cpu = POWERPC_CPU(cs);
465     CPUPPCState *cenv = &cpu->env;
466     int ret;
467 
468     /* Synchronize sregs with kvm */
469     ret = kvm_arch_sync_sregs(cpu);
470     if (ret) {
471         if (ret == -EINVAL) {
472             error_report("Register sync failed... If you're using kvm-hv.ko,"
473                          " only \"-cpu host\" is possible");
474         }
475         return ret;
476     }
477 
478     switch (cenv->mmu_model) {
479     case POWERPC_MMU_BOOKE206:
480         /* This target supports access to KVM's guest TLB */
481         ret = kvm_booke206_tlb_init(cpu);
482         break;
483     case POWERPC_MMU_2_07:
484         if (!cap_htm && !kvmppc_is_pr(cs->kvm_state)) {
485             /*
486              * KVM-HV has transactional memory on POWER8 also without
487              * the KVM_CAP_PPC_HTM extension, so enable it here
488              * instead as long as it's availble to userspace on the
489              * host.
490              */
491             if (qemu_getauxval(AT_HWCAP2) & PPC_FEATURE2_HAS_HTM) {
492                 cap_htm = true;
493             }
494         }
495         break;
496     default:
497         break;
498     }
499 
500     kvm_get_one_reg(cs, KVM_REG_PPC_DEBUG_INST, &debug_inst_opcode);
501     kvmppc_hw_debug_points_init(cenv);
502 
503     return ret;
504 }
505 
506 int kvm_arch_destroy_vcpu(CPUState *cs)
507 {
508     return 0;
509 }
510 
511 static void kvm_sw_tlb_put(PowerPCCPU *cpu)
512 {
513     CPUPPCState *env = &cpu->env;
514     CPUState *cs = CPU(cpu);
515     struct kvm_dirty_tlb dirty_tlb;
516     unsigned char *bitmap;
517     int ret;
518 
519     if (!env->kvm_sw_tlb) {
520         return;
521     }
522 
523     bitmap = g_malloc((env->nb_tlb + 7) / 8);
524     memset(bitmap, 0xFF, (env->nb_tlb + 7) / 8);
525 
526     dirty_tlb.bitmap = (uintptr_t)bitmap;
527     dirty_tlb.num_dirty = env->nb_tlb;
528 
529     ret = kvm_vcpu_ioctl(cs, KVM_DIRTY_TLB, &dirty_tlb);
530     if (ret) {
531         fprintf(stderr, "%s: KVM_DIRTY_TLB: %s\n",
532                 __func__, strerror(-ret));
533     }
534 
535     g_free(bitmap);
536 }
537 
538 static void kvm_get_one_spr(CPUState *cs, uint64_t id, int spr)
539 {
540     PowerPCCPU *cpu = POWERPC_CPU(cs);
541     CPUPPCState *env = &cpu->env;
542     union {
543         uint32_t u32;
544         uint64_t u64;
545     } val;
546     struct kvm_one_reg reg = {
547         .id = id,
548         .addr = (uintptr_t) &val,
549     };
550     int ret;
551 
552     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
553     if (ret != 0) {
554         trace_kvm_failed_spr_get(spr, strerror(errno));
555     } else {
556         switch (id & KVM_REG_SIZE_MASK) {
557         case KVM_REG_SIZE_U32:
558             env->spr[spr] = val.u32;
559             break;
560 
561         case KVM_REG_SIZE_U64:
562             env->spr[spr] = val.u64;
563             break;
564 
565         default:
566             /* Don't handle this size yet */
567             abort();
568         }
569     }
570 }
571 
572 static void kvm_put_one_spr(CPUState *cs, uint64_t id, int spr)
573 {
574     PowerPCCPU *cpu = POWERPC_CPU(cs);
575     CPUPPCState *env = &cpu->env;
576     union {
577         uint32_t u32;
578         uint64_t u64;
579     } val;
580     struct kvm_one_reg reg = {
581         .id = id,
582         .addr = (uintptr_t) &val,
583     };
584     int ret;
585 
586     switch (id & KVM_REG_SIZE_MASK) {
587     case KVM_REG_SIZE_U32:
588         val.u32 = env->spr[spr];
589         break;
590 
591     case KVM_REG_SIZE_U64:
592         val.u64 = env->spr[spr];
593         break;
594 
595     default:
596         /* Don't handle this size yet */
597         abort();
598     }
599 
600     ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
601     if (ret != 0) {
602         trace_kvm_failed_spr_set(spr, strerror(errno));
603     }
604 }
605 
606 static int kvm_put_fp(CPUState *cs)
607 {
608     PowerPCCPU *cpu = POWERPC_CPU(cs);
609     CPUPPCState *env = &cpu->env;
610     struct kvm_one_reg reg;
611     int i;
612     int ret;
613 
614     if (env->insns_flags & PPC_FLOAT) {
615         uint64_t fpscr = env->fpscr;
616         bool vsx = !!(env->insns_flags2 & PPC2_VSX);
617 
618         reg.id = KVM_REG_PPC_FPSCR;
619         reg.addr = (uintptr_t)&fpscr;
620         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
621         if (ret < 0) {
622             trace_kvm_failed_fpscr_set(strerror(errno));
623             return ret;
624         }
625 
626         for (i = 0; i < 32; i++) {
627             uint64_t vsr[2];
628             uint64_t *fpr = cpu_fpr_ptr(&cpu->env, i);
629             uint64_t *vsrl = cpu_vsrl_ptr(&cpu->env, i);
630 
631 #ifdef HOST_WORDS_BIGENDIAN
632             vsr[0] = float64_val(*fpr);
633             vsr[1] = *vsrl;
634 #else
635             vsr[0] = *vsrl;
636             vsr[1] = float64_val(*fpr);
637 #endif
638             reg.addr = (uintptr_t) &vsr;
639             reg.id = vsx ? KVM_REG_PPC_VSR(i) : KVM_REG_PPC_FPR(i);
640 
641             ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
642             if (ret < 0) {
643                 trace_kvm_failed_fp_set(vsx ? "VSR" : "FPR", i,
644                                         strerror(errno));
645                 return ret;
646             }
647         }
648     }
649 
650     if (env->insns_flags & PPC_ALTIVEC) {
651         reg.id = KVM_REG_PPC_VSCR;
652         reg.addr = (uintptr_t)&env->vscr;
653         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
654         if (ret < 0) {
655             trace_kvm_failed_vscr_set(strerror(errno));
656             return ret;
657         }
658 
659         for (i = 0; i < 32; i++) {
660             reg.id = KVM_REG_PPC_VR(i);
661             reg.addr = (uintptr_t)cpu_avr_ptr(env, i);
662             ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
663             if (ret < 0) {
664                 trace_kvm_failed_vr_set(i, strerror(errno));
665                 return ret;
666             }
667         }
668     }
669 
670     return 0;
671 }
672 
673 static int kvm_get_fp(CPUState *cs)
674 {
675     PowerPCCPU *cpu = POWERPC_CPU(cs);
676     CPUPPCState *env = &cpu->env;
677     struct kvm_one_reg reg;
678     int i;
679     int ret;
680 
681     if (env->insns_flags & PPC_FLOAT) {
682         uint64_t fpscr;
683         bool vsx = !!(env->insns_flags2 & PPC2_VSX);
684 
685         reg.id = KVM_REG_PPC_FPSCR;
686         reg.addr = (uintptr_t)&fpscr;
687         ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
688         if (ret < 0) {
689             trace_kvm_failed_fpscr_get(strerror(errno));
690             return ret;
691         } else {
692             env->fpscr = fpscr;
693         }
694 
695         for (i = 0; i < 32; i++) {
696             uint64_t vsr[2];
697             uint64_t *fpr = cpu_fpr_ptr(&cpu->env, i);
698             uint64_t *vsrl = cpu_vsrl_ptr(&cpu->env, i);
699 
700             reg.addr = (uintptr_t) &vsr;
701             reg.id = vsx ? KVM_REG_PPC_VSR(i) : KVM_REG_PPC_FPR(i);
702 
703             ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
704             if (ret < 0) {
705                 trace_kvm_failed_fp_get(vsx ? "VSR" : "FPR", i,
706                                         strerror(errno));
707                 return ret;
708             } else {
709 #ifdef HOST_WORDS_BIGENDIAN
710                 *fpr = vsr[0];
711                 if (vsx) {
712                     *vsrl = vsr[1];
713                 }
714 #else
715                 *fpr = vsr[1];
716                 if (vsx) {
717                     *vsrl = vsr[0];
718                 }
719 #endif
720             }
721         }
722     }
723 
724     if (env->insns_flags & PPC_ALTIVEC) {
725         reg.id = KVM_REG_PPC_VSCR;
726         reg.addr = (uintptr_t)&env->vscr;
727         ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
728         if (ret < 0) {
729             trace_kvm_failed_vscr_get(strerror(errno));
730             return ret;
731         }
732 
733         for (i = 0; i < 32; i++) {
734             reg.id = KVM_REG_PPC_VR(i);
735             reg.addr = (uintptr_t)cpu_avr_ptr(env, i);
736             ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
737             if (ret < 0) {
738                 trace_kvm_failed_vr_get(i, strerror(errno));
739                 return ret;
740             }
741         }
742     }
743 
744     return 0;
745 }
746 
747 #if defined(TARGET_PPC64)
748 static int kvm_get_vpa(CPUState *cs)
749 {
750     PowerPCCPU *cpu = POWERPC_CPU(cs);
751     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
752     struct kvm_one_reg reg;
753     int ret;
754 
755     reg.id = KVM_REG_PPC_VPA_ADDR;
756     reg.addr = (uintptr_t)&spapr_cpu->vpa_addr;
757     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
758     if (ret < 0) {
759         trace_kvm_failed_vpa_addr_get(strerror(errno));
760         return ret;
761     }
762 
763     assert((uintptr_t)&spapr_cpu->slb_shadow_size
764            == ((uintptr_t)&spapr_cpu->slb_shadow_addr + 8));
765     reg.id = KVM_REG_PPC_VPA_SLB;
766     reg.addr = (uintptr_t)&spapr_cpu->slb_shadow_addr;
767     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
768     if (ret < 0) {
769         trace_kvm_failed_slb_get(strerror(errno));
770         return ret;
771     }
772 
773     assert((uintptr_t)&spapr_cpu->dtl_size
774            == ((uintptr_t)&spapr_cpu->dtl_addr + 8));
775     reg.id = KVM_REG_PPC_VPA_DTL;
776     reg.addr = (uintptr_t)&spapr_cpu->dtl_addr;
777     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
778     if (ret < 0) {
779         trace_kvm_failed_dtl_get(strerror(errno));
780         return ret;
781     }
782 
783     return 0;
784 }
785 
786 static int kvm_put_vpa(CPUState *cs)
787 {
788     PowerPCCPU *cpu = POWERPC_CPU(cs);
789     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
790     struct kvm_one_reg reg;
791     int ret;
792 
793     /*
794      * SLB shadow or DTL can't be registered unless a master VPA is
795      * registered.  That means when restoring state, if a VPA *is*
796      * registered, we need to set that up first.  If not, we need to
797      * deregister the others before deregistering the master VPA
798      */
799     assert(spapr_cpu->vpa_addr
800            || !(spapr_cpu->slb_shadow_addr || spapr_cpu->dtl_addr));
801 
802     if (spapr_cpu->vpa_addr) {
803         reg.id = KVM_REG_PPC_VPA_ADDR;
804         reg.addr = (uintptr_t)&spapr_cpu->vpa_addr;
805         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
806         if (ret < 0) {
807             trace_kvm_failed_vpa_addr_set(strerror(errno));
808             return ret;
809         }
810     }
811 
812     assert((uintptr_t)&spapr_cpu->slb_shadow_size
813            == ((uintptr_t)&spapr_cpu->slb_shadow_addr + 8));
814     reg.id = KVM_REG_PPC_VPA_SLB;
815     reg.addr = (uintptr_t)&spapr_cpu->slb_shadow_addr;
816     ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
817     if (ret < 0) {
818         trace_kvm_failed_slb_set(strerror(errno));
819         return ret;
820     }
821 
822     assert((uintptr_t)&spapr_cpu->dtl_size
823            == ((uintptr_t)&spapr_cpu->dtl_addr + 8));
824     reg.id = KVM_REG_PPC_VPA_DTL;
825     reg.addr = (uintptr_t)&spapr_cpu->dtl_addr;
826     ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
827     if (ret < 0) {
828         trace_kvm_failed_dtl_set(strerror(errno));
829         return ret;
830     }
831 
832     if (!spapr_cpu->vpa_addr) {
833         reg.id = KVM_REG_PPC_VPA_ADDR;
834         reg.addr = (uintptr_t)&spapr_cpu->vpa_addr;
835         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
836         if (ret < 0) {
837             trace_kvm_failed_null_vpa_addr_set(strerror(errno));
838             return ret;
839         }
840     }
841 
842     return 0;
843 }
844 #endif /* TARGET_PPC64 */
845 
846 int kvmppc_put_books_sregs(PowerPCCPU *cpu)
847 {
848     CPUPPCState *env = &cpu->env;
849     struct kvm_sregs sregs;
850     int i;
851 
852     sregs.pvr = env->spr[SPR_PVR];
853 
854     if (cpu->vhyp) {
855         PPCVirtualHypervisorClass *vhc =
856             PPC_VIRTUAL_HYPERVISOR_GET_CLASS(cpu->vhyp);
857         sregs.u.s.sdr1 = vhc->encode_hpt_for_kvm_pr(cpu->vhyp);
858     } else {
859         sregs.u.s.sdr1 = env->spr[SPR_SDR1];
860     }
861 
862     /* Sync SLB */
863 #ifdef TARGET_PPC64
864     for (i = 0; i < ARRAY_SIZE(env->slb); i++) {
865         sregs.u.s.ppc64.slb[i].slbe = env->slb[i].esid;
866         if (env->slb[i].esid & SLB_ESID_V) {
867             sregs.u.s.ppc64.slb[i].slbe |= i;
868         }
869         sregs.u.s.ppc64.slb[i].slbv = env->slb[i].vsid;
870     }
871 #endif
872 
873     /* Sync SRs */
874     for (i = 0; i < 16; i++) {
875         sregs.u.s.ppc32.sr[i] = env->sr[i];
876     }
877 
878     /* Sync BATs */
879     for (i = 0; i < 8; i++) {
880         /* Beware. We have to swap upper and lower bits here */
881         sregs.u.s.ppc32.dbat[i] = ((uint64_t)env->DBAT[0][i] << 32)
882             | env->DBAT[1][i];
883         sregs.u.s.ppc32.ibat[i] = ((uint64_t)env->IBAT[0][i] << 32)
884             | env->IBAT[1][i];
885     }
886 
887     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_SREGS, &sregs);
888 }
889 
890 int kvm_arch_put_registers(CPUState *cs, int level)
891 {
892     PowerPCCPU *cpu = POWERPC_CPU(cs);
893     CPUPPCState *env = &cpu->env;
894     struct kvm_regs regs;
895     int ret;
896     int i;
897 
898     ret = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
899     if (ret < 0) {
900         return ret;
901     }
902 
903     regs.ctr = env->ctr;
904     regs.lr  = env->lr;
905     regs.xer = cpu_read_xer(env);
906     regs.msr = env->msr;
907     regs.pc = env->nip;
908 
909     regs.srr0 = env->spr[SPR_SRR0];
910     regs.srr1 = env->spr[SPR_SRR1];
911 
912     regs.sprg0 = env->spr[SPR_SPRG0];
913     regs.sprg1 = env->spr[SPR_SPRG1];
914     regs.sprg2 = env->spr[SPR_SPRG2];
915     regs.sprg3 = env->spr[SPR_SPRG3];
916     regs.sprg4 = env->spr[SPR_SPRG4];
917     regs.sprg5 = env->spr[SPR_SPRG5];
918     regs.sprg6 = env->spr[SPR_SPRG6];
919     regs.sprg7 = env->spr[SPR_SPRG7];
920 
921     regs.pid = env->spr[SPR_BOOKE_PID];
922 
923     for (i = 0; i < 32; i++) {
924         regs.gpr[i] = env->gpr[i];
925     }
926 
927     regs.cr = 0;
928     for (i = 0; i < 8; i++) {
929         regs.cr |= (env->crf[i] & 15) << (4 * (7 - i));
930     }
931 
932     ret = kvm_vcpu_ioctl(cs, KVM_SET_REGS, &regs);
933     if (ret < 0) {
934         return ret;
935     }
936 
937     kvm_put_fp(cs);
938 
939     if (env->tlb_dirty) {
940         kvm_sw_tlb_put(cpu);
941         env->tlb_dirty = false;
942     }
943 
944     if (cap_segstate && (level >= KVM_PUT_RESET_STATE)) {
945         ret = kvmppc_put_books_sregs(cpu);
946         if (ret < 0) {
947             return ret;
948         }
949     }
950 
951     if (cap_hior && (level >= KVM_PUT_RESET_STATE)) {
952         kvm_put_one_spr(cs, KVM_REG_PPC_HIOR, SPR_HIOR);
953     }
954 
955     if (cap_one_reg) {
956         int i;
957 
958         /*
959          * We deliberately ignore errors here, for kernels which have
960          * the ONE_REG calls, but don't support the specific
961          * registers, there's a reasonable chance things will still
962          * work, at least until we try to migrate.
963          */
964         for (i = 0; i < 1024; i++) {
965             uint64_t id = env->spr_cb[i].one_reg_id;
966 
967             if (id != 0) {
968                 kvm_put_one_spr(cs, id, i);
969             }
970         }
971 
972 #ifdef TARGET_PPC64
973         if (msr_ts) {
974             for (i = 0; i < ARRAY_SIZE(env->tm_gpr); i++) {
975                 kvm_set_one_reg(cs, KVM_REG_PPC_TM_GPR(i), &env->tm_gpr[i]);
976             }
977             for (i = 0; i < ARRAY_SIZE(env->tm_vsr); i++) {
978                 kvm_set_one_reg(cs, KVM_REG_PPC_TM_VSR(i), &env->tm_vsr[i]);
979             }
980             kvm_set_one_reg(cs, KVM_REG_PPC_TM_CR, &env->tm_cr);
981             kvm_set_one_reg(cs, KVM_REG_PPC_TM_LR, &env->tm_lr);
982             kvm_set_one_reg(cs, KVM_REG_PPC_TM_CTR, &env->tm_ctr);
983             kvm_set_one_reg(cs, KVM_REG_PPC_TM_FPSCR, &env->tm_fpscr);
984             kvm_set_one_reg(cs, KVM_REG_PPC_TM_AMR, &env->tm_amr);
985             kvm_set_one_reg(cs, KVM_REG_PPC_TM_PPR, &env->tm_ppr);
986             kvm_set_one_reg(cs, KVM_REG_PPC_TM_VRSAVE, &env->tm_vrsave);
987             kvm_set_one_reg(cs, KVM_REG_PPC_TM_VSCR, &env->tm_vscr);
988             kvm_set_one_reg(cs, KVM_REG_PPC_TM_DSCR, &env->tm_dscr);
989             kvm_set_one_reg(cs, KVM_REG_PPC_TM_TAR, &env->tm_tar);
990         }
991 
992         if (cap_papr) {
993             if (kvm_put_vpa(cs) < 0) {
994                 trace_kvm_failed_put_vpa();
995             }
996         }
997 
998         kvm_set_one_reg(cs, KVM_REG_PPC_TB_OFFSET, &env->tb_env->tb_offset);
999 
1000         if (level > KVM_PUT_RUNTIME_STATE) {
1001             kvm_put_one_spr(cs, KVM_REG_PPC_DPDES, SPR_DPDES);
1002         }
1003 #endif /* TARGET_PPC64 */
1004     }
1005 
1006     return ret;
1007 }
1008 
1009 static void kvm_sync_excp(CPUPPCState *env, int vector, int ivor)
1010 {
1011      env->excp_vectors[vector] = env->spr[ivor] + env->spr[SPR_BOOKE_IVPR];
1012 }
1013 
1014 static int kvmppc_get_booke_sregs(PowerPCCPU *cpu)
1015 {
1016     CPUPPCState *env = &cpu->env;
1017     struct kvm_sregs sregs;
1018     int ret;
1019 
1020     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_SREGS, &sregs);
1021     if (ret < 0) {
1022         return ret;
1023     }
1024 
1025     if (sregs.u.e.features & KVM_SREGS_E_BASE) {
1026         env->spr[SPR_BOOKE_CSRR0] = sregs.u.e.csrr0;
1027         env->spr[SPR_BOOKE_CSRR1] = sregs.u.e.csrr1;
1028         env->spr[SPR_BOOKE_ESR] = sregs.u.e.esr;
1029         env->spr[SPR_BOOKE_DEAR] = sregs.u.e.dear;
1030         env->spr[SPR_BOOKE_MCSR] = sregs.u.e.mcsr;
1031         env->spr[SPR_BOOKE_TSR] = sregs.u.e.tsr;
1032         env->spr[SPR_BOOKE_TCR] = sregs.u.e.tcr;
1033         env->spr[SPR_DECR] = sregs.u.e.dec;
1034         env->spr[SPR_TBL] = sregs.u.e.tb & 0xffffffff;
1035         env->spr[SPR_TBU] = sregs.u.e.tb >> 32;
1036         env->spr[SPR_VRSAVE] = sregs.u.e.vrsave;
1037     }
1038 
1039     if (sregs.u.e.features & KVM_SREGS_E_ARCH206) {
1040         env->spr[SPR_BOOKE_PIR] = sregs.u.e.pir;
1041         env->spr[SPR_BOOKE_MCSRR0] = sregs.u.e.mcsrr0;
1042         env->spr[SPR_BOOKE_MCSRR1] = sregs.u.e.mcsrr1;
1043         env->spr[SPR_BOOKE_DECAR] = sregs.u.e.decar;
1044         env->spr[SPR_BOOKE_IVPR] = sregs.u.e.ivpr;
1045     }
1046 
1047     if (sregs.u.e.features & KVM_SREGS_E_64) {
1048         env->spr[SPR_BOOKE_EPCR] = sregs.u.e.epcr;
1049     }
1050 
1051     if (sregs.u.e.features & KVM_SREGS_E_SPRG8) {
1052         env->spr[SPR_BOOKE_SPRG8] = sregs.u.e.sprg8;
1053     }
1054 
1055     if (sregs.u.e.features & KVM_SREGS_E_IVOR) {
1056         env->spr[SPR_BOOKE_IVOR0] = sregs.u.e.ivor_low[0];
1057         kvm_sync_excp(env, POWERPC_EXCP_CRITICAL,  SPR_BOOKE_IVOR0);
1058         env->spr[SPR_BOOKE_IVOR1] = sregs.u.e.ivor_low[1];
1059         kvm_sync_excp(env, POWERPC_EXCP_MCHECK,  SPR_BOOKE_IVOR1);
1060         env->spr[SPR_BOOKE_IVOR2] = sregs.u.e.ivor_low[2];
1061         kvm_sync_excp(env, POWERPC_EXCP_DSI,  SPR_BOOKE_IVOR2);
1062         env->spr[SPR_BOOKE_IVOR3] = sregs.u.e.ivor_low[3];
1063         kvm_sync_excp(env, POWERPC_EXCP_ISI,  SPR_BOOKE_IVOR3);
1064         env->spr[SPR_BOOKE_IVOR4] = sregs.u.e.ivor_low[4];
1065         kvm_sync_excp(env, POWERPC_EXCP_EXTERNAL,  SPR_BOOKE_IVOR4);
1066         env->spr[SPR_BOOKE_IVOR5] = sregs.u.e.ivor_low[5];
1067         kvm_sync_excp(env, POWERPC_EXCP_ALIGN,  SPR_BOOKE_IVOR5);
1068         env->spr[SPR_BOOKE_IVOR6] = sregs.u.e.ivor_low[6];
1069         kvm_sync_excp(env, POWERPC_EXCP_PROGRAM,  SPR_BOOKE_IVOR6);
1070         env->spr[SPR_BOOKE_IVOR7] = sregs.u.e.ivor_low[7];
1071         kvm_sync_excp(env, POWERPC_EXCP_FPU,  SPR_BOOKE_IVOR7);
1072         env->spr[SPR_BOOKE_IVOR8] = sregs.u.e.ivor_low[8];
1073         kvm_sync_excp(env, POWERPC_EXCP_SYSCALL,  SPR_BOOKE_IVOR8);
1074         env->spr[SPR_BOOKE_IVOR9] = sregs.u.e.ivor_low[9];
1075         kvm_sync_excp(env, POWERPC_EXCP_APU,  SPR_BOOKE_IVOR9);
1076         env->spr[SPR_BOOKE_IVOR10] = sregs.u.e.ivor_low[10];
1077         kvm_sync_excp(env, POWERPC_EXCP_DECR,  SPR_BOOKE_IVOR10);
1078         env->spr[SPR_BOOKE_IVOR11] = sregs.u.e.ivor_low[11];
1079         kvm_sync_excp(env, POWERPC_EXCP_FIT,  SPR_BOOKE_IVOR11);
1080         env->spr[SPR_BOOKE_IVOR12] = sregs.u.e.ivor_low[12];
1081         kvm_sync_excp(env, POWERPC_EXCP_WDT,  SPR_BOOKE_IVOR12);
1082         env->spr[SPR_BOOKE_IVOR13] = sregs.u.e.ivor_low[13];
1083         kvm_sync_excp(env, POWERPC_EXCP_DTLB,  SPR_BOOKE_IVOR13);
1084         env->spr[SPR_BOOKE_IVOR14] = sregs.u.e.ivor_low[14];
1085         kvm_sync_excp(env, POWERPC_EXCP_ITLB,  SPR_BOOKE_IVOR14);
1086         env->spr[SPR_BOOKE_IVOR15] = sregs.u.e.ivor_low[15];
1087         kvm_sync_excp(env, POWERPC_EXCP_DEBUG,  SPR_BOOKE_IVOR15);
1088 
1089         if (sregs.u.e.features & KVM_SREGS_E_SPE) {
1090             env->spr[SPR_BOOKE_IVOR32] = sregs.u.e.ivor_high[0];
1091             kvm_sync_excp(env, POWERPC_EXCP_SPEU,  SPR_BOOKE_IVOR32);
1092             env->spr[SPR_BOOKE_IVOR33] = sregs.u.e.ivor_high[1];
1093             kvm_sync_excp(env, POWERPC_EXCP_EFPDI,  SPR_BOOKE_IVOR33);
1094             env->spr[SPR_BOOKE_IVOR34] = sregs.u.e.ivor_high[2];
1095             kvm_sync_excp(env, POWERPC_EXCP_EFPRI,  SPR_BOOKE_IVOR34);
1096         }
1097 
1098         if (sregs.u.e.features & KVM_SREGS_E_PM) {
1099             env->spr[SPR_BOOKE_IVOR35] = sregs.u.e.ivor_high[3];
1100             kvm_sync_excp(env, POWERPC_EXCP_EPERFM,  SPR_BOOKE_IVOR35);
1101         }
1102 
1103         if (sregs.u.e.features & KVM_SREGS_E_PC) {
1104             env->spr[SPR_BOOKE_IVOR36] = sregs.u.e.ivor_high[4];
1105             kvm_sync_excp(env, POWERPC_EXCP_DOORI,  SPR_BOOKE_IVOR36);
1106             env->spr[SPR_BOOKE_IVOR37] = sregs.u.e.ivor_high[5];
1107             kvm_sync_excp(env, POWERPC_EXCP_DOORCI, SPR_BOOKE_IVOR37);
1108         }
1109     }
1110 
1111     if (sregs.u.e.features & KVM_SREGS_E_ARCH206_MMU) {
1112         env->spr[SPR_BOOKE_MAS0] = sregs.u.e.mas0;
1113         env->spr[SPR_BOOKE_MAS1] = sregs.u.e.mas1;
1114         env->spr[SPR_BOOKE_MAS2] = sregs.u.e.mas2;
1115         env->spr[SPR_BOOKE_MAS3] = sregs.u.e.mas7_3 & 0xffffffff;
1116         env->spr[SPR_BOOKE_MAS4] = sregs.u.e.mas4;
1117         env->spr[SPR_BOOKE_MAS6] = sregs.u.e.mas6;
1118         env->spr[SPR_BOOKE_MAS7] = sregs.u.e.mas7_3 >> 32;
1119         env->spr[SPR_MMUCFG] = sregs.u.e.mmucfg;
1120         env->spr[SPR_BOOKE_TLB0CFG] = sregs.u.e.tlbcfg[0];
1121         env->spr[SPR_BOOKE_TLB1CFG] = sregs.u.e.tlbcfg[1];
1122     }
1123 
1124     if (sregs.u.e.features & KVM_SREGS_EXP) {
1125         env->spr[SPR_BOOKE_EPR] = sregs.u.e.epr;
1126     }
1127 
1128     if (sregs.u.e.features & KVM_SREGS_E_PD) {
1129         env->spr[SPR_BOOKE_EPLC] = sregs.u.e.eplc;
1130         env->spr[SPR_BOOKE_EPSC] = sregs.u.e.epsc;
1131     }
1132 
1133     if (sregs.u.e.impl_id == KVM_SREGS_E_IMPL_FSL) {
1134         env->spr[SPR_E500_SVR] = sregs.u.e.impl.fsl.svr;
1135         env->spr[SPR_Exxx_MCAR] = sregs.u.e.impl.fsl.mcar;
1136         env->spr[SPR_HID0] = sregs.u.e.impl.fsl.hid0;
1137 
1138         if (sregs.u.e.impl.fsl.features & KVM_SREGS_E_FSL_PIDn) {
1139             env->spr[SPR_BOOKE_PID1] = sregs.u.e.impl.fsl.pid1;
1140             env->spr[SPR_BOOKE_PID2] = sregs.u.e.impl.fsl.pid2;
1141         }
1142     }
1143 
1144     return 0;
1145 }
1146 
1147 static int kvmppc_get_books_sregs(PowerPCCPU *cpu)
1148 {
1149     CPUPPCState *env = &cpu->env;
1150     struct kvm_sregs sregs;
1151     int ret;
1152     int i;
1153 
1154     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_SREGS, &sregs);
1155     if (ret < 0) {
1156         return ret;
1157     }
1158 
1159     if (!cpu->vhyp) {
1160         ppc_store_sdr1(env, sregs.u.s.sdr1);
1161     }
1162 
1163     /* Sync SLB */
1164 #ifdef TARGET_PPC64
1165     /*
1166      * The packed SLB array we get from KVM_GET_SREGS only contains
1167      * information about valid entries. So we flush our internal copy
1168      * to get rid of stale ones, then put all valid SLB entries back
1169      * in.
1170      */
1171     memset(env->slb, 0, sizeof(env->slb));
1172     for (i = 0; i < ARRAY_SIZE(env->slb); i++) {
1173         target_ulong rb = sregs.u.s.ppc64.slb[i].slbe;
1174         target_ulong rs = sregs.u.s.ppc64.slb[i].slbv;
1175         /*
1176          * Only restore valid entries
1177          */
1178         if (rb & SLB_ESID_V) {
1179             ppc_store_slb(cpu, rb & 0xfff, rb & ~0xfffULL, rs);
1180         }
1181     }
1182 #endif
1183 
1184     /* Sync SRs */
1185     for (i = 0; i < 16; i++) {
1186         env->sr[i] = sregs.u.s.ppc32.sr[i];
1187     }
1188 
1189     /* Sync BATs */
1190     for (i = 0; i < 8; i++) {
1191         env->DBAT[0][i] = sregs.u.s.ppc32.dbat[i] & 0xffffffff;
1192         env->DBAT[1][i] = sregs.u.s.ppc32.dbat[i] >> 32;
1193         env->IBAT[0][i] = sregs.u.s.ppc32.ibat[i] & 0xffffffff;
1194         env->IBAT[1][i] = sregs.u.s.ppc32.ibat[i] >> 32;
1195     }
1196 
1197     return 0;
1198 }
1199 
1200 int kvm_arch_get_registers(CPUState *cs)
1201 {
1202     PowerPCCPU *cpu = POWERPC_CPU(cs);
1203     CPUPPCState *env = &cpu->env;
1204     struct kvm_regs regs;
1205     uint32_t cr;
1206     int i, ret;
1207 
1208     ret = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
1209     if (ret < 0) {
1210         return ret;
1211     }
1212 
1213     cr = regs.cr;
1214     for (i = 7; i >= 0; i--) {
1215         env->crf[i] = cr & 15;
1216         cr >>= 4;
1217     }
1218 
1219     env->ctr = regs.ctr;
1220     env->lr = regs.lr;
1221     cpu_write_xer(env, regs.xer);
1222     env->msr = regs.msr;
1223     env->nip = regs.pc;
1224 
1225     env->spr[SPR_SRR0] = regs.srr0;
1226     env->spr[SPR_SRR1] = regs.srr1;
1227 
1228     env->spr[SPR_SPRG0] = regs.sprg0;
1229     env->spr[SPR_SPRG1] = regs.sprg1;
1230     env->spr[SPR_SPRG2] = regs.sprg2;
1231     env->spr[SPR_SPRG3] = regs.sprg3;
1232     env->spr[SPR_SPRG4] = regs.sprg4;
1233     env->spr[SPR_SPRG5] = regs.sprg5;
1234     env->spr[SPR_SPRG6] = regs.sprg6;
1235     env->spr[SPR_SPRG7] = regs.sprg7;
1236 
1237     env->spr[SPR_BOOKE_PID] = regs.pid;
1238 
1239     for (i = 0; i < 32; i++) {
1240         env->gpr[i] = regs.gpr[i];
1241     }
1242 
1243     kvm_get_fp(cs);
1244 
1245     if (cap_booke_sregs) {
1246         ret = kvmppc_get_booke_sregs(cpu);
1247         if (ret < 0) {
1248             return ret;
1249         }
1250     }
1251 
1252     if (cap_segstate) {
1253         ret = kvmppc_get_books_sregs(cpu);
1254         if (ret < 0) {
1255             return ret;
1256         }
1257     }
1258 
1259     if (cap_hior) {
1260         kvm_get_one_spr(cs, KVM_REG_PPC_HIOR, SPR_HIOR);
1261     }
1262 
1263     if (cap_one_reg) {
1264         int i;
1265 
1266         /*
1267          * We deliberately ignore errors here, for kernels which have
1268          * the ONE_REG calls, but don't support the specific
1269          * registers, there's a reasonable chance things will still
1270          * work, at least until we try to migrate.
1271          */
1272         for (i = 0; i < 1024; i++) {
1273             uint64_t id = env->spr_cb[i].one_reg_id;
1274 
1275             if (id != 0) {
1276                 kvm_get_one_spr(cs, id, i);
1277             }
1278         }
1279 
1280 #ifdef TARGET_PPC64
1281         if (msr_ts) {
1282             for (i = 0; i < ARRAY_SIZE(env->tm_gpr); i++) {
1283                 kvm_get_one_reg(cs, KVM_REG_PPC_TM_GPR(i), &env->tm_gpr[i]);
1284             }
1285             for (i = 0; i < ARRAY_SIZE(env->tm_vsr); i++) {
1286                 kvm_get_one_reg(cs, KVM_REG_PPC_TM_VSR(i), &env->tm_vsr[i]);
1287             }
1288             kvm_get_one_reg(cs, KVM_REG_PPC_TM_CR, &env->tm_cr);
1289             kvm_get_one_reg(cs, KVM_REG_PPC_TM_LR, &env->tm_lr);
1290             kvm_get_one_reg(cs, KVM_REG_PPC_TM_CTR, &env->tm_ctr);
1291             kvm_get_one_reg(cs, KVM_REG_PPC_TM_FPSCR, &env->tm_fpscr);
1292             kvm_get_one_reg(cs, KVM_REG_PPC_TM_AMR, &env->tm_amr);
1293             kvm_get_one_reg(cs, KVM_REG_PPC_TM_PPR, &env->tm_ppr);
1294             kvm_get_one_reg(cs, KVM_REG_PPC_TM_VRSAVE, &env->tm_vrsave);
1295             kvm_get_one_reg(cs, KVM_REG_PPC_TM_VSCR, &env->tm_vscr);
1296             kvm_get_one_reg(cs, KVM_REG_PPC_TM_DSCR, &env->tm_dscr);
1297             kvm_get_one_reg(cs, KVM_REG_PPC_TM_TAR, &env->tm_tar);
1298         }
1299 
1300         if (cap_papr) {
1301             if (kvm_get_vpa(cs) < 0) {
1302                 trace_kvm_failed_get_vpa();
1303             }
1304         }
1305 
1306         kvm_get_one_reg(cs, KVM_REG_PPC_TB_OFFSET, &env->tb_env->tb_offset);
1307         kvm_get_one_spr(cs, KVM_REG_PPC_DPDES, SPR_DPDES);
1308 #endif
1309     }
1310 
1311     return 0;
1312 }
1313 
1314 int kvmppc_set_interrupt(PowerPCCPU *cpu, int irq, int level)
1315 {
1316     unsigned virq = level ? KVM_INTERRUPT_SET_LEVEL : KVM_INTERRUPT_UNSET;
1317 
1318     if (irq != PPC_INTERRUPT_EXT) {
1319         return 0;
1320     }
1321 
1322     if (!kvm_enabled() || !cap_interrupt_unset) {
1323         return 0;
1324     }
1325 
1326     kvm_vcpu_ioctl(CPU(cpu), KVM_INTERRUPT, &virq);
1327 
1328     return 0;
1329 }
1330 
1331 void kvm_arch_pre_run(CPUState *cs, struct kvm_run *run)
1332 {
1333     return;
1334 }
1335 
1336 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
1337 {
1338     return MEMTXATTRS_UNSPECIFIED;
1339 }
1340 
1341 int kvm_arch_process_async_events(CPUState *cs)
1342 {
1343     return cs->halted;
1344 }
1345 
1346 static int kvmppc_handle_halt(PowerPCCPU *cpu)
1347 {
1348     CPUState *cs = CPU(cpu);
1349     CPUPPCState *env = &cpu->env;
1350 
1351     if (!(cs->interrupt_request & CPU_INTERRUPT_HARD) && (msr_ee)) {
1352         cs->halted = 1;
1353         cs->exception_index = EXCP_HLT;
1354     }
1355 
1356     return 0;
1357 }
1358 
1359 /* map dcr access to existing qemu dcr emulation */
1360 static int kvmppc_handle_dcr_read(CPUPPCState *env,
1361                                   uint32_t dcrn, uint32_t *data)
1362 {
1363     if (ppc_dcr_read(env->dcr_env, dcrn, data) < 0) {
1364         fprintf(stderr, "Read to unhandled DCR (0x%x)\n", dcrn);
1365     }
1366 
1367     return 0;
1368 }
1369 
1370 static int kvmppc_handle_dcr_write(CPUPPCState *env,
1371                                    uint32_t dcrn, uint32_t data)
1372 {
1373     if (ppc_dcr_write(env->dcr_env, dcrn, data) < 0) {
1374         fprintf(stderr, "Write to unhandled DCR (0x%x)\n", dcrn);
1375     }
1376 
1377     return 0;
1378 }
1379 
1380 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
1381 {
1382     /* Mixed endian case is not handled */
1383     uint32_t sc = debug_inst_opcode;
1384 
1385     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
1386                             sizeof(sc), 0) ||
1387         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&sc, sizeof(sc), 1)) {
1388         return -EINVAL;
1389     }
1390 
1391     return 0;
1392 }
1393 
1394 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
1395 {
1396     uint32_t sc;
1397 
1398     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&sc, sizeof(sc), 0) ||
1399         sc != debug_inst_opcode ||
1400         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
1401                             sizeof(sc), 1)) {
1402         return -EINVAL;
1403     }
1404 
1405     return 0;
1406 }
1407 
1408 static int find_hw_breakpoint(target_ulong addr, int type)
1409 {
1410     int n;
1411 
1412     assert((nb_hw_breakpoint + nb_hw_watchpoint)
1413            <= ARRAY_SIZE(hw_debug_points));
1414 
1415     for (n = 0; n < nb_hw_breakpoint + nb_hw_watchpoint; n++) {
1416         if (hw_debug_points[n].addr == addr &&
1417              hw_debug_points[n].type == type) {
1418             return n;
1419         }
1420     }
1421 
1422     return -1;
1423 }
1424 
1425 static int find_hw_watchpoint(target_ulong addr, int *flag)
1426 {
1427     int n;
1428 
1429     n = find_hw_breakpoint(addr, GDB_WATCHPOINT_ACCESS);
1430     if (n >= 0) {
1431         *flag = BP_MEM_ACCESS;
1432         return n;
1433     }
1434 
1435     n = find_hw_breakpoint(addr, GDB_WATCHPOINT_WRITE);
1436     if (n >= 0) {
1437         *flag = BP_MEM_WRITE;
1438         return n;
1439     }
1440 
1441     n = find_hw_breakpoint(addr, GDB_WATCHPOINT_READ);
1442     if (n >= 0) {
1443         *flag = BP_MEM_READ;
1444         return n;
1445     }
1446 
1447     return -1;
1448 }
1449 
1450 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
1451                                   target_ulong len, int type)
1452 {
1453     if ((nb_hw_breakpoint + nb_hw_watchpoint) >= ARRAY_SIZE(hw_debug_points)) {
1454         return -ENOBUFS;
1455     }
1456 
1457     hw_debug_points[nb_hw_breakpoint + nb_hw_watchpoint].addr = addr;
1458     hw_debug_points[nb_hw_breakpoint + nb_hw_watchpoint].type = type;
1459 
1460     switch (type) {
1461     case GDB_BREAKPOINT_HW:
1462         if (nb_hw_breakpoint >= max_hw_breakpoint) {
1463             return -ENOBUFS;
1464         }
1465 
1466         if (find_hw_breakpoint(addr, type) >= 0) {
1467             return -EEXIST;
1468         }
1469 
1470         nb_hw_breakpoint++;
1471         break;
1472 
1473     case GDB_WATCHPOINT_WRITE:
1474     case GDB_WATCHPOINT_READ:
1475     case GDB_WATCHPOINT_ACCESS:
1476         if (nb_hw_watchpoint >= max_hw_watchpoint) {
1477             return -ENOBUFS;
1478         }
1479 
1480         if (find_hw_breakpoint(addr, type) >= 0) {
1481             return -EEXIST;
1482         }
1483 
1484         nb_hw_watchpoint++;
1485         break;
1486 
1487     default:
1488         return -ENOSYS;
1489     }
1490 
1491     return 0;
1492 }
1493 
1494 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
1495                                   target_ulong len, int type)
1496 {
1497     int n;
1498 
1499     n = find_hw_breakpoint(addr, type);
1500     if (n < 0) {
1501         return -ENOENT;
1502     }
1503 
1504     switch (type) {
1505     case GDB_BREAKPOINT_HW:
1506         nb_hw_breakpoint--;
1507         break;
1508 
1509     case GDB_WATCHPOINT_WRITE:
1510     case GDB_WATCHPOINT_READ:
1511     case GDB_WATCHPOINT_ACCESS:
1512         nb_hw_watchpoint--;
1513         break;
1514 
1515     default:
1516         return -ENOSYS;
1517     }
1518     hw_debug_points[n] = hw_debug_points[nb_hw_breakpoint + nb_hw_watchpoint];
1519 
1520     return 0;
1521 }
1522 
1523 void kvm_arch_remove_all_hw_breakpoints(void)
1524 {
1525     nb_hw_breakpoint = nb_hw_watchpoint = 0;
1526 }
1527 
1528 void kvm_arch_update_guest_debug(CPUState *cs, struct kvm_guest_debug *dbg)
1529 {
1530     int n;
1531 
1532     /* Software Breakpoint updates */
1533     if (kvm_sw_breakpoints_active(cs)) {
1534         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP;
1535     }
1536 
1537     assert((nb_hw_breakpoint + nb_hw_watchpoint)
1538            <= ARRAY_SIZE(hw_debug_points));
1539     assert((nb_hw_breakpoint + nb_hw_watchpoint) <= ARRAY_SIZE(dbg->arch.bp));
1540 
1541     if (nb_hw_breakpoint + nb_hw_watchpoint > 0) {
1542         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
1543         memset(dbg->arch.bp, 0, sizeof(dbg->arch.bp));
1544         for (n = 0; n < nb_hw_breakpoint + nb_hw_watchpoint; n++) {
1545             switch (hw_debug_points[n].type) {
1546             case GDB_BREAKPOINT_HW:
1547                 dbg->arch.bp[n].type = KVMPPC_DEBUG_BREAKPOINT;
1548                 break;
1549             case GDB_WATCHPOINT_WRITE:
1550                 dbg->arch.bp[n].type = KVMPPC_DEBUG_WATCH_WRITE;
1551                 break;
1552             case GDB_WATCHPOINT_READ:
1553                 dbg->arch.bp[n].type = KVMPPC_DEBUG_WATCH_READ;
1554                 break;
1555             case GDB_WATCHPOINT_ACCESS:
1556                 dbg->arch.bp[n].type = KVMPPC_DEBUG_WATCH_WRITE |
1557                                         KVMPPC_DEBUG_WATCH_READ;
1558                 break;
1559             default:
1560                 cpu_abort(cs, "Unsupported breakpoint type\n");
1561             }
1562             dbg->arch.bp[n].addr = hw_debug_points[n].addr;
1563         }
1564     }
1565 }
1566 
1567 static int kvm_handle_hw_breakpoint(CPUState *cs,
1568                                     struct kvm_debug_exit_arch *arch_info)
1569 {
1570     int handle = DEBUG_RETURN_GUEST;
1571     int n;
1572     int flag = 0;
1573 
1574     if (nb_hw_breakpoint + nb_hw_watchpoint > 0) {
1575         if (arch_info->status & KVMPPC_DEBUG_BREAKPOINT) {
1576             n = find_hw_breakpoint(arch_info->address, GDB_BREAKPOINT_HW);
1577             if (n >= 0) {
1578                 handle = DEBUG_RETURN_GDB;
1579             }
1580         } else if (arch_info->status & (KVMPPC_DEBUG_WATCH_READ |
1581                                         KVMPPC_DEBUG_WATCH_WRITE)) {
1582             n = find_hw_watchpoint(arch_info->address,  &flag);
1583             if (n >= 0) {
1584                 handle = DEBUG_RETURN_GDB;
1585                 cs->watchpoint_hit = &hw_watchpoint;
1586                 hw_watchpoint.vaddr = hw_debug_points[n].addr;
1587                 hw_watchpoint.flags = flag;
1588             }
1589         }
1590     }
1591     return handle;
1592 }
1593 
1594 static int kvm_handle_singlestep(void)
1595 {
1596     return DEBUG_RETURN_GDB;
1597 }
1598 
1599 static int kvm_handle_sw_breakpoint(void)
1600 {
1601     return DEBUG_RETURN_GDB;
1602 }
1603 
1604 static int kvm_handle_debug(PowerPCCPU *cpu, struct kvm_run *run)
1605 {
1606     CPUState *cs = CPU(cpu);
1607     CPUPPCState *env = &cpu->env;
1608     struct kvm_debug_exit_arch *arch_info = &run->debug.arch;
1609 
1610     if (cs->singlestep_enabled) {
1611         return kvm_handle_singlestep();
1612     }
1613 
1614     if (arch_info->status) {
1615         return kvm_handle_hw_breakpoint(cs, arch_info);
1616     }
1617 
1618     if (kvm_find_sw_breakpoint(cs, arch_info->address)) {
1619         return kvm_handle_sw_breakpoint();
1620     }
1621 
1622     /*
1623      * QEMU is not able to handle debug exception, so inject
1624      * program exception to guest;
1625      * Yes program exception NOT debug exception !!
1626      * When QEMU is using debug resources then debug exception must
1627      * be always set. To achieve this we set MSR_DE and also set
1628      * MSRP_DEP so guest cannot change MSR_DE.
1629      * When emulating debug resource for guest we want guest
1630      * to control MSR_DE (enable/disable debug interrupt on need).
1631      * Supporting both configurations are NOT possible.
1632      * So the result is that we cannot share debug resources
1633      * between QEMU and Guest on BOOKE architecture.
1634      * In the current design QEMU gets the priority over guest,
1635      * this means that if QEMU is using debug resources then guest
1636      * cannot use them;
1637      * For software breakpoint QEMU uses a privileged instruction;
1638      * So there cannot be any reason that we are here for guest
1639      * set debug exception, only possibility is guest executed a
1640      * privileged / illegal instruction and that's why we are
1641      * injecting a program interrupt.
1642      */
1643     cpu_synchronize_state(cs);
1644     /*
1645      * env->nip is PC, so increment this by 4 to use
1646      * ppc_cpu_do_interrupt(), which set srr0 = env->nip - 4.
1647      */
1648     env->nip += 4;
1649     cs->exception_index = POWERPC_EXCP_PROGRAM;
1650     env->error_code = POWERPC_EXCP_INVAL;
1651     ppc_cpu_do_interrupt(cs);
1652 
1653     return DEBUG_RETURN_GUEST;
1654 }
1655 
1656 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1657 {
1658     PowerPCCPU *cpu = POWERPC_CPU(cs);
1659     CPUPPCState *env = &cpu->env;
1660     int ret;
1661 
1662     qemu_mutex_lock_iothread();
1663 
1664     switch (run->exit_reason) {
1665     case KVM_EXIT_DCR:
1666         if (run->dcr.is_write) {
1667             trace_kvm_handle_dcr_write();
1668             ret = kvmppc_handle_dcr_write(env, run->dcr.dcrn, run->dcr.data);
1669         } else {
1670             trace_kvm_handle_dcr_read();
1671             ret = kvmppc_handle_dcr_read(env, run->dcr.dcrn, &run->dcr.data);
1672         }
1673         break;
1674     case KVM_EXIT_HLT:
1675         trace_kvm_handle_halt();
1676         ret = kvmppc_handle_halt(cpu);
1677         break;
1678 #if defined(TARGET_PPC64)
1679     case KVM_EXIT_PAPR_HCALL:
1680         trace_kvm_handle_papr_hcall();
1681         run->papr_hcall.ret = spapr_hypercall(cpu,
1682                                               run->papr_hcall.nr,
1683                                               run->papr_hcall.args);
1684         ret = 0;
1685         break;
1686 #endif
1687     case KVM_EXIT_EPR:
1688         trace_kvm_handle_epr();
1689         run->epr.epr = ldl_phys(cs->as, env->mpic_iack);
1690         ret = 0;
1691         break;
1692     case KVM_EXIT_WATCHDOG:
1693         trace_kvm_handle_watchdog_expiry();
1694         watchdog_perform_action();
1695         ret = 0;
1696         break;
1697 
1698     case KVM_EXIT_DEBUG:
1699         trace_kvm_handle_debug_exception();
1700         if (kvm_handle_debug(cpu, run)) {
1701             ret = EXCP_DEBUG;
1702             break;
1703         }
1704         /* re-enter, this exception was guest-internal */
1705         ret = 0;
1706         break;
1707 
1708 #if defined(TARGET_PPC64)
1709     case KVM_EXIT_NMI:
1710         trace_kvm_handle_nmi_exception();
1711         ret = kvm_handle_nmi(cpu, run);
1712         break;
1713 #endif
1714 
1715     default:
1716         fprintf(stderr, "KVM: unknown exit reason %d\n", run->exit_reason);
1717         ret = -1;
1718         break;
1719     }
1720 
1721     qemu_mutex_unlock_iothread();
1722     return ret;
1723 }
1724 
1725 int kvmppc_or_tsr_bits(PowerPCCPU *cpu, uint32_t tsr_bits)
1726 {
1727     CPUState *cs = CPU(cpu);
1728     uint32_t bits = tsr_bits;
1729     struct kvm_one_reg reg = {
1730         .id = KVM_REG_PPC_OR_TSR,
1731         .addr = (uintptr_t) &bits,
1732     };
1733 
1734     return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
1735 }
1736 
1737 int kvmppc_clear_tsr_bits(PowerPCCPU *cpu, uint32_t tsr_bits)
1738 {
1739 
1740     CPUState *cs = CPU(cpu);
1741     uint32_t bits = tsr_bits;
1742     struct kvm_one_reg reg = {
1743         .id = KVM_REG_PPC_CLEAR_TSR,
1744         .addr = (uintptr_t) &bits,
1745     };
1746 
1747     return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
1748 }
1749 
1750 int kvmppc_set_tcr(PowerPCCPU *cpu)
1751 {
1752     CPUState *cs = CPU(cpu);
1753     CPUPPCState *env = &cpu->env;
1754     uint32_t tcr = env->spr[SPR_BOOKE_TCR];
1755 
1756     struct kvm_one_reg reg = {
1757         .id = KVM_REG_PPC_TCR,
1758         .addr = (uintptr_t) &tcr,
1759     };
1760 
1761     return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
1762 }
1763 
1764 int kvmppc_booke_watchdog_enable(PowerPCCPU *cpu)
1765 {
1766     CPUState *cs = CPU(cpu);
1767     int ret;
1768 
1769     if (!kvm_enabled()) {
1770         return -1;
1771     }
1772 
1773     if (!cap_ppc_watchdog) {
1774         printf("warning: KVM does not support watchdog");
1775         return -1;
1776     }
1777 
1778     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_PPC_BOOKE_WATCHDOG, 0);
1779     if (ret < 0) {
1780         fprintf(stderr, "%s: couldn't enable KVM_CAP_PPC_BOOKE_WATCHDOG: %s\n",
1781                 __func__, strerror(-ret));
1782         return ret;
1783     }
1784 
1785     return ret;
1786 }
1787 
1788 static int read_cpuinfo(const char *field, char *value, int len)
1789 {
1790     FILE *f;
1791     int ret = -1;
1792     int field_len = strlen(field);
1793     char line[512];
1794 
1795     f = fopen("/proc/cpuinfo", "r");
1796     if (!f) {
1797         return -1;
1798     }
1799 
1800     do {
1801         if (!fgets(line, sizeof(line), f)) {
1802             break;
1803         }
1804         if (!strncmp(line, field, field_len)) {
1805             pstrcpy(value, len, line);
1806             ret = 0;
1807             break;
1808         }
1809     } while (*line);
1810 
1811     fclose(f);
1812 
1813     return ret;
1814 }
1815 
1816 uint32_t kvmppc_get_tbfreq(void)
1817 {
1818     char line[512];
1819     char *ns;
1820     uint32_t retval = NANOSECONDS_PER_SECOND;
1821 
1822     if (read_cpuinfo("timebase", line, sizeof(line))) {
1823         return retval;
1824     }
1825 
1826     ns = strchr(line, ':');
1827     if (!ns) {
1828         return retval;
1829     }
1830 
1831     ns++;
1832 
1833     return atoi(ns);
1834 }
1835 
1836 bool kvmppc_get_host_serial(char **value)
1837 {
1838     return g_file_get_contents("/proc/device-tree/system-id", value, NULL,
1839                                NULL);
1840 }
1841 
1842 bool kvmppc_get_host_model(char **value)
1843 {
1844     return g_file_get_contents("/proc/device-tree/model", value, NULL, NULL);
1845 }
1846 
1847 /* Try to find a device tree node for a CPU with clock-frequency property */
1848 static int kvmppc_find_cpu_dt(char *buf, int buf_len)
1849 {
1850     struct dirent *dirp;
1851     DIR *dp;
1852 
1853     dp = opendir(PROC_DEVTREE_CPU);
1854     if (!dp) {
1855         printf("Can't open directory " PROC_DEVTREE_CPU "\n");
1856         return -1;
1857     }
1858 
1859     buf[0] = '\0';
1860     while ((dirp = readdir(dp)) != NULL) {
1861         FILE *f;
1862         snprintf(buf, buf_len, "%s%s/clock-frequency", PROC_DEVTREE_CPU,
1863                  dirp->d_name);
1864         f = fopen(buf, "r");
1865         if (f) {
1866             snprintf(buf, buf_len, "%s%s", PROC_DEVTREE_CPU, dirp->d_name);
1867             fclose(f);
1868             break;
1869         }
1870         buf[0] = '\0';
1871     }
1872     closedir(dp);
1873     if (buf[0] == '\0') {
1874         printf("Unknown host!\n");
1875         return -1;
1876     }
1877 
1878     return 0;
1879 }
1880 
1881 static uint64_t kvmppc_read_int_dt(const char *filename)
1882 {
1883     union {
1884         uint32_t v32;
1885         uint64_t v64;
1886     } u;
1887     FILE *f;
1888     int len;
1889 
1890     f = fopen(filename, "rb");
1891     if (!f) {
1892         return -1;
1893     }
1894 
1895     len = fread(&u, 1, sizeof(u), f);
1896     fclose(f);
1897     switch (len) {
1898     case 4:
1899         /* property is a 32-bit quantity */
1900         return be32_to_cpu(u.v32);
1901     case 8:
1902         return be64_to_cpu(u.v64);
1903     }
1904 
1905     return 0;
1906 }
1907 
1908 /*
1909  * Read a CPU node property from the host device tree that's a single
1910  * integer (32-bit or 64-bit).  Returns 0 if anything goes wrong
1911  * (can't find or open the property, or doesn't understand the format)
1912  */
1913 static uint64_t kvmppc_read_int_cpu_dt(const char *propname)
1914 {
1915     char buf[PATH_MAX], *tmp;
1916     uint64_t val;
1917 
1918     if (kvmppc_find_cpu_dt(buf, sizeof(buf))) {
1919         return -1;
1920     }
1921 
1922     tmp = g_strdup_printf("%s/%s", buf, propname);
1923     val = kvmppc_read_int_dt(tmp);
1924     g_free(tmp);
1925 
1926     return val;
1927 }
1928 
1929 uint64_t kvmppc_get_clockfreq(void)
1930 {
1931     return kvmppc_read_int_cpu_dt("clock-frequency");
1932 }
1933 
1934 static int kvmppc_get_dec_bits(void)
1935 {
1936     int nr_bits = kvmppc_read_int_cpu_dt("ibm,dec-bits");
1937 
1938     if (nr_bits > 0) {
1939         return nr_bits;
1940     }
1941     return 0;
1942 }
1943 
1944 static int kvmppc_get_pvinfo(CPUPPCState *env, struct kvm_ppc_pvinfo *pvinfo)
1945 {
1946     CPUState *cs = env_cpu(env);
1947 
1948     if (kvm_vm_check_extension(cs->kvm_state, KVM_CAP_PPC_GET_PVINFO) &&
1949         !kvm_vm_ioctl(cs->kvm_state, KVM_PPC_GET_PVINFO, pvinfo)) {
1950         return 0;
1951     }
1952 
1953     return 1;
1954 }
1955 
1956 int kvmppc_get_hasidle(CPUPPCState *env)
1957 {
1958     struct kvm_ppc_pvinfo pvinfo;
1959 
1960     if (!kvmppc_get_pvinfo(env, &pvinfo) &&
1961         (pvinfo.flags & KVM_PPC_PVINFO_FLAGS_EV_IDLE)) {
1962         return 1;
1963     }
1964 
1965     return 0;
1966 }
1967 
1968 int kvmppc_get_hypercall(CPUPPCState *env, uint8_t *buf, int buf_len)
1969 {
1970     uint32_t *hc = (uint32_t *)buf;
1971     struct kvm_ppc_pvinfo pvinfo;
1972 
1973     if (!kvmppc_get_pvinfo(env, &pvinfo)) {
1974         memcpy(buf, pvinfo.hcall, buf_len);
1975         return 0;
1976     }
1977 
1978     /*
1979      * Fallback to always fail hypercalls regardless of endianness:
1980      *
1981      *     tdi 0,r0,72 (becomes b .+8 in wrong endian, nop in good endian)
1982      *     li r3, -1
1983      *     b .+8       (becomes nop in wrong endian)
1984      *     bswap32(li r3, -1)
1985      */
1986 
1987     hc[0] = cpu_to_be32(0x08000048);
1988     hc[1] = cpu_to_be32(0x3860ffff);
1989     hc[2] = cpu_to_be32(0x48000008);
1990     hc[3] = cpu_to_be32(bswap32(0x3860ffff));
1991 
1992     return 1;
1993 }
1994 
1995 static inline int kvmppc_enable_hcall(KVMState *s, target_ulong hcall)
1996 {
1997     return kvm_vm_enable_cap(s, KVM_CAP_PPC_ENABLE_HCALL, 0, hcall, 1);
1998 }
1999 
2000 void kvmppc_enable_logical_ci_hcalls(void)
2001 {
2002     /*
2003      * FIXME: it would be nice if we could detect the cases where
2004      * we're using a device which requires the in kernel
2005      * implementation of these hcalls, but the kernel lacks them and
2006      * produce a warning.
2007      */
2008     kvmppc_enable_hcall(kvm_state, H_LOGICAL_CI_LOAD);
2009     kvmppc_enable_hcall(kvm_state, H_LOGICAL_CI_STORE);
2010 }
2011 
2012 void kvmppc_enable_set_mode_hcall(void)
2013 {
2014     kvmppc_enable_hcall(kvm_state, H_SET_MODE);
2015 }
2016 
2017 void kvmppc_enable_clear_ref_mod_hcalls(void)
2018 {
2019     kvmppc_enable_hcall(kvm_state, H_CLEAR_REF);
2020     kvmppc_enable_hcall(kvm_state, H_CLEAR_MOD);
2021 }
2022 
2023 void kvmppc_enable_h_page_init(void)
2024 {
2025     kvmppc_enable_hcall(kvm_state, H_PAGE_INIT);
2026 }
2027 
2028 void kvmppc_set_papr(PowerPCCPU *cpu)
2029 {
2030     CPUState *cs = CPU(cpu);
2031     int ret;
2032 
2033     if (!kvm_enabled()) {
2034         return;
2035     }
2036 
2037     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_PPC_PAPR, 0);
2038     if (ret) {
2039         error_report("This vCPU type or KVM version does not support PAPR");
2040         exit(1);
2041     }
2042 
2043     /*
2044      * Update the capability flag so we sync the right information
2045      * with kvm
2046      */
2047     cap_papr = 1;
2048 }
2049 
2050 int kvmppc_set_compat(PowerPCCPU *cpu, uint32_t compat_pvr)
2051 {
2052     return kvm_set_one_reg(CPU(cpu), KVM_REG_PPC_ARCH_COMPAT, &compat_pvr);
2053 }
2054 
2055 void kvmppc_set_mpic_proxy(PowerPCCPU *cpu, int mpic_proxy)
2056 {
2057     CPUState *cs = CPU(cpu);
2058     int ret;
2059 
2060     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_PPC_EPR, 0, mpic_proxy);
2061     if (ret && mpic_proxy) {
2062         error_report("This KVM version does not support EPR");
2063         exit(1);
2064     }
2065 }
2066 
2067 int kvmppc_set_fwnmi(void)
2068 {
2069     PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
2070     CPUState *cs = CPU(cpu);
2071 
2072     return kvm_vcpu_enable_cap(cs, KVM_CAP_PPC_FWNMI, 0);
2073 }
2074 
2075 int kvmppc_smt_threads(void)
2076 {
2077     return cap_ppc_smt ? cap_ppc_smt : 1;
2078 }
2079 
2080 int kvmppc_set_smt_threads(int smt)
2081 {
2082     int ret;
2083 
2084     ret = kvm_vm_enable_cap(kvm_state, KVM_CAP_PPC_SMT, 0, smt, 0);
2085     if (!ret) {
2086         cap_ppc_smt = smt;
2087     }
2088     return ret;
2089 }
2090 
2091 void kvmppc_error_append_smt_possible_hint(Error *const *errp)
2092 {
2093     int i;
2094     GString *g;
2095     char *s;
2096 
2097     assert(kvm_enabled());
2098     if (cap_ppc_smt_possible) {
2099         g = g_string_new("Available VSMT modes:");
2100         for (i = 63; i >= 0; i--) {
2101             if ((1UL << i) & cap_ppc_smt_possible) {
2102                 g_string_append_printf(g, " %lu", (1UL << i));
2103             }
2104         }
2105         s = g_string_free(g, false);
2106         error_append_hint(errp, "%s.\n", s);
2107         g_free(s);
2108     } else {
2109         error_append_hint(errp,
2110                           "This KVM seems to be too old to support VSMT.\n");
2111     }
2112 }
2113 
2114 
2115 #ifdef TARGET_PPC64
2116 uint64_t kvmppc_vrma_limit(unsigned int hash_shift)
2117 {
2118     struct kvm_ppc_smmu_info info;
2119     long rampagesize, best_page_shift;
2120     int i;
2121 
2122     /*
2123      * Find the largest hardware supported page size that's less than
2124      * or equal to the (logical) backing page size of guest RAM
2125      */
2126     kvm_get_smmu_info(&info, &error_fatal);
2127     rampagesize = qemu_minrampagesize();
2128     best_page_shift = 0;
2129 
2130     for (i = 0; i < KVM_PPC_PAGE_SIZES_MAX_SZ; i++) {
2131         struct kvm_ppc_one_seg_page_size *sps = &info.sps[i];
2132 
2133         if (!sps->page_shift) {
2134             continue;
2135         }
2136 
2137         if ((sps->page_shift > best_page_shift)
2138             && ((1UL << sps->page_shift) <= rampagesize)) {
2139             best_page_shift = sps->page_shift;
2140         }
2141     }
2142 
2143     return 1ULL << (best_page_shift + hash_shift - 7);
2144 }
2145 #endif
2146 
2147 bool kvmppc_spapr_use_multitce(void)
2148 {
2149     return cap_spapr_multitce;
2150 }
2151 
2152 int kvmppc_spapr_enable_inkernel_multitce(void)
2153 {
2154     int ret;
2155 
2156     ret = kvm_vm_enable_cap(kvm_state, KVM_CAP_PPC_ENABLE_HCALL, 0,
2157                             H_PUT_TCE_INDIRECT, 1);
2158     if (!ret) {
2159         ret = kvm_vm_enable_cap(kvm_state, KVM_CAP_PPC_ENABLE_HCALL, 0,
2160                                 H_STUFF_TCE, 1);
2161     }
2162 
2163     return ret;
2164 }
2165 
2166 void *kvmppc_create_spapr_tce(uint32_t liobn, uint32_t page_shift,
2167                               uint64_t bus_offset, uint32_t nb_table,
2168                               int *pfd, bool need_vfio)
2169 {
2170     long len;
2171     int fd;
2172     void *table;
2173 
2174     /*
2175      * Must set fd to -1 so we don't try to munmap when called for
2176      * destroying the table, which the upper layers -will- do
2177      */
2178     *pfd = -1;
2179     if (!cap_spapr_tce || (need_vfio && !cap_spapr_vfio)) {
2180         return NULL;
2181     }
2182 
2183     if (cap_spapr_tce_64) {
2184         struct kvm_create_spapr_tce_64 args = {
2185             .liobn = liobn,
2186             .page_shift = page_shift,
2187             .offset = bus_offset >> page_shift,
2188             .size = nb_table,
2189             .flags = 0
2190         };
2191         fd = kvm_vm_ioctl(kvm_state, KVM_CREATE_SPAPR_TCE_64, &args);
2192         if (fd < 0) {
2193             fprintf(stderr,
2194                     "KVM: Failed to create TCE64 table for liobn 0x%x\n",
2195                     liobn);
2196             return NULL;
2197         }
2198     } else if (cap_spapr_tce) {
2199         uint64_t window_size = (uint64_t) nb_table << page_shift;
2200         struct kvm_create_spapr_tce args = {
2201             .liobn = liobn,
2202             .window_size = window_size,
2203         };
2204         if ((window_size != args.window_size) || bus_offset) {
2205             return NULL;
2206         }
2207         fd = kvm_vm_ioctl(kvm_state, KVM_CREATE_SPAPR_TCE, &args);
2208         if (fd < 0) {
2209             fprintf(stderr, "KVM: Failed to create TCE table for liobn 0x%x\n",
2210                     liobn);
2211             return NULL;
2212         }
2213     } else {
2214         return NULL;
2215     }
2216 
2217     len = nb_table * sizeof(uint64_t);
2218     /* FIXME: round this up to page size */
2219 
2220     table = mmap(NULL, len, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
2221     if (table == MAP_FAILED) {
2222         fprintf(stderr, "KVM: Failed to map TCE table for liobn 0x%x\n",
2223                 liobn);
2224         close(fd);
2225         return NULL;
2226     }
2227 
2228     *pfd = fd;
2229     return table;
2230 }
2231 
2232 int kvmppc_remove_spapr_tce(void *table, int fd, uint32_t nb_table)
2233 {
2234     long len;
2235 
2236     if (fd < 0) {
2237         return -1;
2238     }
2239 
2240     len = nb_table * sizeof(uint64_t);
2241     if ((munmap(table, len) < 0) ||
2242         (close(fd) < 0)) {
2243         fprintf(stderr, "KVM: Unexpected error removing TCE table: %s",
2244                 strerror(errno));
2245         /* Leak the table */
2246     }
2247 
2248     return 0;
2249 }
2250 
2251 int kvmppc_reset_htab(int shift_hint)
2252 {
2253     uint32_t shift = shift_hint;
2254 
2255     if (!kvm_enabled()) {
2256         /* Full emulation, tell caller to allocate htab itself */
2257         return 0;
2258     }
2259     if (kvm_vm_check_extension(kvm_state, KVM_CAP_PPC_ALLOC_HTAB)) {
2260         int ret;
2261         ret = kvm_vm_ioctl(kvm_state, KVM_PPC_ALLOCATE_HTAB, &shift);
2262         if (ret == -ENOTTY) {
2263             /*
2264              * At least some versions of PR KVM advertise the
2265              * capability, but don't implement the ioctl().  Oops.
2266              * Return 0 so that we allocate the htab in qemu, as is
2267              * correct for PR.
2268              */
2269             return 0;
2270         } else if (ret < 0) {
2271             return ret;
2272         }
2273         return shift;
2274     }
2275 
2276     /*
2277      * We have a kernel that predates the htab reset calls.  For PR
2278      * KVM, we need to allocate the htab ourselves, for an HV KVM of
2279      * this era, it has allocated a 16MB fixed size hash table
2280      * already.
2281      */
2282     if (kvmppc_is_pr(kvm_state)) {
2283         /* PR - tell caller to allocate htab */
2284         return 0;
2285     } else {
2286         /* HV - assume 16MB kernel allocated htab */
2287         return 24;
2288     }
2289 }
2290 
2291 static inline uint32_t mfpvr(void)
2292 {
2293     uint32_t pvr;
2294 
2295     asm ("mfpvr %0"
2296          : "=r"(pvr));
2297     return pvr;
2298 }
2299 
2300 static void alter_insns(uint64_t *word, uint64_t flags, bool on)
2301 {
2302     if (on) {
2303         *word |= flags;
2304     } else {
2305         *word &= ~flags;
2306     }
2307 }
2308 
2309 static void kvmppc_host_cpu_class_init(ObjectClass *oc, void *data)
2310 {
2311     PowerPCCPUClass *pcc = POWERPC_CPU_CLASS(oc);
2312     uint32_t dcache_size = kvmppc_read_int_cpu_dt("d-cache-size");
2313     uint32_t icache_size = kvmppc_read_int_cpu_dt("i-cache-size");
2314 
2315     /* Now fix up the class with information we can query from the host */
2316     pcc->pvr = mfpvr();
2317 
2318     alter_insns(&pcc->insns_flags, PPC_ALTIVEC,
2319                 qemu_getauxval(AT_HWCAP) & PPC_FEATURE_HAS_ALTIVEC);
2320     alter_insns(&pcc->insns_flags2, PPC2_VSX,
2321                 qemu_getauxval(AT_HWCAP) & PPC_FEATURE_HAS_VSX);
2322     alter_insns(&pcc->insns_flags2, PPC2_DFP,
2323                 qemu_getauxval(AT_HWCAP) & PPC_FEATURE_HAS_DFP);
2324 
2325     if (dcache_size != -1) {
2326         pcc->l1_dcache_size = dcache_size;
2327     }
2328 
2329     if (icache_size != -1) {
2330         pcc->l1_icache_size = icache_size;
2331     }
2332 
2333 #if defined(TARGET_PPC64)
2334     pcc->radix_page_info = kvm_get_radix_page_info();
2335 
2336     if ((pcc->pvr & 0xffffff00) == CPU_POWERPC_POWER9_DD1) {
2337         /*
2338          * POWER9 DD1 has some bugs which make it not really ISA 3.00
2339          * compliant.  More importantly, advertising ISA 3.00
2340          * architected mode may prevent guests from activating
2341          * necessary DD1 workarounds.
2342          */
2343         pcc->pcr_supported &= ~(PCR_COMPAT_3_00 | PCR_COMPAT_2_07
2344                                 | PCR_COMPAT_2_06 | PCR_COMPAT_2_05);
2345     }
2346 #endif /* defined(TARGET_PPC64) */
2347 }
2348 
2349 bool kvmppc_has_cap_epr(void)
2350 {
2351     return cap_epr;
2352 }
2353 
2354 bool kvmppc_has_cap_fixup_hcalls(void)
2355 {
2356     return cap_fixup_hcalls;
2357 }
2358 
2359 bool kvmppc_has_cap_htm(void)
2360 {
2361     return cap_htm;
2362 }
2363 
2364 bool kvmppc_has_cap_mmu_radix(void)
2365 {
2366     return cap_mmu_radix;
2367 }
2368 
2369 bool kvmppc_has_cap_mmu_hash_v3(void)
2370 {
2371     return cap_mmu_hash_v3;
2372 }
2373 
2374 static bool kvmppc_power8_host(void)
2375 {
2376     bool ret = false;
2377 #ifdef TARGET_PPC64
2378     {
2379         uint32_t base_pvr = CPU_POWERPC_POWER_SERVER_MASK & mfpvr();
2380         ret = (base_pvr == CPU_POWERPC_POWER8E_BASE) ||
2381               (base_pvr == CPU_POWERPC_POWER8NVL_BASE) ||
2382               (base_pvr == CPU_POWERPC_POWER8_BASE);
2383     }
2384 #endif /* TARGET_PPC64 */
2385     return ret;
2386 }
2387 
2388 static int parse_cap_ppc_safe_cache(struct kvm_ppc_cpu_char c)
2389 {
2390     bool l1d_thread_priv_req = !kvmppc_power8_host();
2391 
2392     if (~c.behaviour & c.behaviour_mask & H_CPU_BEHAV_L1D_FLUSH_PR) {
2393         return 2;
2394     } else if ((!l1d_thread_priv_req ||
2395                 c.character & c.character_mask & H_CPU_CHAR_L1D_THREAD_PRIV) &&
2396                (c.character & c.character_mask
2397                 & (H_CPU_CHAR_L1D_FLUSH_ORI30 | H_CPU_CHAR_L1D_FLUSH_TRIG2))) {
2398         return 1;
2399     }
2400 
2401     return 0;
2402 }
2403 
2404 static int parse_cap_ppc_safe_bounds_check(struct kvm_ppc_cpu_char c)
2405 {
2406     if (~c.behaviour & c.behaviour_mask & H_CPU_BEHAV_BNDS_CHK_SPEC_BAR) {
2407         return 2;
2408     } else if (c.character & c.character_mask & H_CPU_CHAR_SPEC_BAR_ORI31) {
2409         return 1;
2410     }
2411 
2412     return 0;
2413 }
2414 
2415 static int parse_cap_ppc_safe_indirect_branch(struct kvm_ppc_cpu_char c)
2416 {
2417     if ((~c.behaviour & c.behaviour_mask & H_CPU_BEHAV_FLUSH_COUNT_CACHE) &&
2418         (~c.character & c.character_mask & H_CPU_CHAR_CACHE_COUNT_DIS) &&
2419         (~c.character & c.character_mask & H_CPU_CHAR_BCCTRL_SERIALISED)) {
2420         return SPAPR_CAP_FIXED_NA;
2421     } else if (c.behaviour & c.behaviour_mask & H_CPU_BEHAV_FLUSH_COUNT_CACHE) {
2422         return SPAPR_CAP_WORKAROUND;
2423     } else if (c.character & c.character_mask & H_CPU_CHAR_CACHE_COUNT_DIS) {
2424         return  SPAPR_CAP_FIXED_CCD;
2425     } else if (c.character & c.character_mask & H_CPU_CHAR_BCCTRL_SERIALISED) {
2426         return SPAPR_CAP_FIXED_IBS;
2427     }
2428 
2429     return 0;
2430 }
2431 
2432 static int parse_cap_ppc_count_cache_flush_assist(struct kvm_ppc_cpu_char c)
2433 {
2434     if (c.character & c.character_mask & H_CPU_CHAR_BCCTR_FLUSH_ASSIST) {
2435         return 1;
2436     }
2437     return 0;
2438 }
2439 
2440 bool kvmppc_has_cap_xive(void)
2441 {
2442     return cap_xive;
2443 }
2444 
2445 static void kvmppc_get_cpu_characteristics(KVMState *s)
2446 {
2447     struct kvm_ppc_cpu_char c;
2448     int ret;
2449 
2450     /* Assume broken */
2451     cap_ppc_safe_cache = 0;
2452     cap_ppc_safe_bounds_check = 0;
2453     cap_ppc_safe_indirect_branch = 0;
2454 
2455     ret = kvm_vm_check_extension(s, KVM_CAP_PPC_GET_CPU_CHAR);
2456     if (!ret) {
2457         return;
2458     }
2459     ret = kvm_vm_ioctl(s, KVM_PPC_GET_CPU_CHAR, &c);
2460     if (ret < 0) {
2461         return;
2462     }
2463 
2464     cap_ppc_safe_cache = parse_cap_ppc_safe_cache(c);
2465     cap_ppc_safe_bounds_check = parse_cap_ppc_safe_bounds_check(c);
2466     cap_ppc_safe_indirect_branch = parse_cap_ppc_safe_indirect_branch(c);
2467     cap_ppc_count_cache_flush_assist =
2468         parse_cap_ppc_count_cache_flush_assist(c);
2469 }
2470 
2471 int kvmppc_get_cap_safe_cache(void)
2472 {
2473     return cap_ppc_safe_cache;
2474 }
2475 
2476 int kvmppc_get_cap_safe_bounds_check(void)
2477 {
2478     return cap_ppc_safe_bounds_check;
2479 }
2480 
2481 int kvmppc_get_cap_safe_indirect_branch(void)
2482 {
2483     return cap_ppc_safe_indirect_branch;
2484 }
2485 
2486 int kvmppc_get_cap_count_cache_flush_assist(void)
2487 {
2488     return cap_ppc_count_cache_flush_assist;
2489 }
2490 
2491 bool kvmppc_has_cap_nested_kvm_hv(void)
2492 {
2493     return !!cap_ppc_nested_kvm_hv;
2494 }
2495 
2496 int kvmppc_set_cap_nested_kvm_hv(int enable)
2497 {
2498     return kvm_vm_enable_cap(kvm_state, KVM_CAP_PPC_NESTED_HV, 0, enable);
2499 }
2500 
2501 bool kvmppc_has_cap_spapr_vfio(void)
2502 {
2503     return cap_spapr_vfio;
2504 }
2505 
2506 int kvmppc_get_cap_large_decr(void)
2507 {
2508     return cap_large_decr;
2509 }
2510 
2511 int kvmppc_enable_cap_large_decr(PowerPCCPU *cpu, int enable)
2512 {
2513     CPUState *cs = CPU(cpu);
2514     uint64_t lpcr;
2515 
2516     kvm_get_one_reg(cs, KVM_REG_PPC_LPCR_64, &lpcr);
2517     /* Do we need to modify the LPCR? */
2518     if (!!(lpcr & LPCR_LD) != !!enable) {
2519         if (enable) {
2520             lpcr |= LPCR_LD;
2521         } else {
2522             lpcr &= ~LPCR_LD;
2523         }
2524         kvm_set_one_reg(cs, KVM_REG_PPC_LPCR_64, &lpcr);
2525         kvm_get_one_reg(cs, KVM_REG_PPC_LPCR_64, &lpcr);
2526 
2527         if (!!(lpcr & LPCR_LD) != !!enable) {
2528             return -1;
2529         }
2530     }
2531 
2532     return 0;
2533 }
2534 
2535 PowerPCCPUClass *kvm_ppc_get_host_cpu_class(void)
2536 {
2537     uint32_t host_pvr = mfpvr();
2538     PowerPCCPUClass *pvr_pcc;
2539 
2540     pvr_pcc = ppc_cpu_class_by_pvr(host_pvr);
2541     if (pvr_pcc == NULL) {
2542         pvr_pcc = ppc_cpu_class_by_pvr_mask(host_pvr);
2543     }
2544 
2545     return pvr_pcc;
2546 }
2547 
2548 static void pseries_machine_class_fixup(ObjectClass *oc, void *opaque)
2549 {
2550     MachineClass *mc = MACHINE_CLASS(oc);
2551 
2552     mc->default_cpu_type = TYPE_HOST_POWERPC_CPU;
2553 }
2554 
2555 static int kvm_ppc_register_host_cpu_type(void)
2556 {
2557     TypeInfo type_info = {
2558         .name = TYPE_HOST_POWERPC_CPU,
2559         .class_init = kvmppc_host_cpu_class_init,
2560     };
2561     PowerPCCPUClass *pvr_pcc;
2562     ObjectClass *oc;
2563     DeviceClass *dc;
2564     int i;
2565 
2566     pvr_pcc = kvm_ppc_get_host_cpu_class();
2567     if (pvr_pcc == NULL) {
2568         return -1;
2569     }
2570     type_info.parent = object_class_get_name(OBJECT_CLASS(pvr_pcc));
2571     type_register(&type_info);
2572     /* override TCG default cpu type with 'host' cpu model */
2573     object_class_foreach(pseries_machine_class_fixup, TYPE_SPAPR_MACHINE,
2574                          false, NULL);
2575 
2576     oc = object_class_by_name(type_info.name);
2577     g_assert(oc);
2578 
2579     /*
2580      * Update generic CPU family class alias (e.g. on a POWER8NVL host,
2581      * we want "POWER8" to be a "family" alias that points to the current
2582      * host CPU type, too)
2583      */
2584     dc = DEVICE_CLASS(ppc_cpu_get_family_class(pvr_pcc));
2585     for (i = 0; ppc_cpu_aliases[i].alias != NULL; i++) {
2586         if (strcasecmp(ppc_cpu_aliases[i].alias, dc->desc) == 0) {
2587             char *suffix;
2588 
2589             ppc_cpu_aliases[i].model = g_strdup(object_class_get_name(oc));
2590             suffix = strstr(ppc_cpu_aliases[i].model, POWERPC_CPU_TYPE_SUFFIX);
2591             if (suffix) {
2592                 *suffix = 0;
2593             }
2594             break;
2595         }
2596     }
2597 
2598     return 0;
2599 }
2600 
2601 int kvmppc_define_rtas_kernel_token(uint32_t token, const char *function)
2602 {
2603     struct kvm_rtas_token_args args = {
2604         .token = token,
2605     };
2606 
2607     if (!kvm_check_extension(kvm_state, KVM_CAP_PPC_RTAS)) {
2608         return -ENOENT;
2609     }
2610 
2611     strncpy(args.name, function, sizeof(args.name) - 1);
2612 
2613     return kvm_vm_ioctl(kvm_state, KVM_PPC_RTAS_DEFINE_TOKEN, &args);
2614 }
2615 
2616 int kvmppc_get_htab_fd(bool write, uint64_t index, Error **errp)
2617 {
2618     struct kvm_get_htab_fd s = {
2619         .flags = write ? KVM_GET_HTAB_WRITE : 0,
2620         .start_index = index,
2621     };
2622     int ret;
2623 
2624     if (!cap_htab_fd) {
2625         error_setg(errp, "KVM version doesn't support %s the HPT",
2626                    write ? "writing" : "reading");
2627         return -ENOTSUP;
2628     }
2629 
2630     ret = kvm_vm_ioctl(kvm_state, KVM_PPC_GET_HTAB_FD, &s);
2631     if (ret < 0) {
2632         error_setg(errp, "Unable to open fd for %s HPT %s KVM: %s",
2633                    write ? "writing" : "reading", write ? "to" : "from",
2634                    strerror(errno));
2635         return -errno;
2636     }
2637 
2638     return ret;
2639 }
2640 
2641 int kvmppc_save_htab(QEMUFile *f, int fd, size_t bufsize, int64_t max_ns)
2642 {
2643     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2644     uint8_t buf[bufsize];
2645     ssize_t rc;
2646 
2647     do {
2648         rc = read(fd, buf, bufsize);
2649         if (rc < 0) {
2650             fprintf(stderr, "Error reading data from KVM HTAB fd: %s\n",
2651                     strerror(errno));
2652             return rc;
2653         } else if (rc) {
2654             uint8_t *buffer = buf;
2655             ssize_t n = rc;
2656             while (n) {
2657                 struct kvm_get_htab_header *head =
2658                     (struct kvm_get_htab_header *) buffer;
2659                 size_t chunksize = sizeof(*head) +
2660                      HASH_PTE_SIZE_64 * head->n_valid;
2661 
2662                 qemu_put_be32(f, head->index);
2663                 qemu_put_be16(f, head->n_valid);
2664                 qemu_put_be16(f, head->n_invalid);
2665                 qemu_put_buffer(f, (void *)(head + 1),
2666                                 HASH_PTE_SIZE_64 * head->n_valid);
2667 
2668                 buffer += chunksize;
2669                 n -= chunksize;
2670             }
2671         }
2672     } while ((rc != 0)
2673              && ((max_ns < 0) ||
2674                  ((qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) < max_ns)));
2675 
2676     return (rc == 0) ? 1 : 0;
2677 }
2678 
2679 int kvmppc_load_htab_chunk(QEMUFile *f, int fd, uint32_t index,
2680                            uint16_t n_valid, uint16_t n_invalid)
2681 {
2682     struct kvm_get_htab_header *buf;
2683     size_t chunksize = sizeof(*buf) + n_valid * HASH_PTE_SIZE_64;
2684     ssize_t rc;
2685 
2686     buf = alloca(chunksize);
2687     buf->index = index;
2688     buf->n_valid = n_valid;
2689     buf->n_invalid = n_invalid;
2690 
2691     qemu_get_buffer(f, (void *)(buf + 1), HASH_PTE_SIZE_64 * n_valid);
2692 
2693     rc = write(fd, buf, chunksize);
2694     if (rc < 0) {
2695         fprintf(stderr, "Error writing KVM hash table: %s\n",
2696                 strerror(errno));
2697         return rc;
2698     }
2699     if (rc != chunksize) {
2700         /* We should never get a short write on a single chunk */
2701         fprintf(stderr, "Short write, restoring KVM hash table\n");
2702         return -1;
2703     }
2704     return 0;
2705 }
2706 
2707 bool kvm_arch_stop_on_emulation_error(CPUState *cpu)
2708 {
2709     return true;
2710 }
2711 
2712 void kvm_arch_init_irq_routing(KVMState *s)
2713 {
2714 }
2715 
2716 void kvmppc_read_hptes(ppc_hash_pte64_t *hptes, hwaddr ptex, int n)
2717 {
2718     int fd, rc;
2719     int i;
2720 
2721     fd = kvmppc_get_htab_fd(false, ptex, &error_abort);
2722 
2723     i = 0;
2724     while (i < n) {
2725         struct kvm_get_htab_header *hdr;
2726         int m = n < HPTES_PER_GROUP ? n : HPTES_PER_GROUP;
2727         char buf[sizeof(*hdr) + m * HASH_PTE_SIZE_64];
2728 
2729         rc = read(fd, buf, sizeof(buf));
2730         if (rc < 0) {
2731             hw_error("kvmppc_read_hptes: Unable to read HPTEs");
2732         }
2733 
2734         hdr = (struct kvm_get_htab_header *)buf;
2735         while ((i < n) && ((char *)hdr < (buf + rc))) {
2736             int invalid = hdr->n_invalid, valid = hdr->n_valid;
2737 
2738             if (hdr->index != (ptex + i)) {
2739                 hw_error("kvmppc_read_hptes: Unexpected HPTE index %"PRIu32
2740                          " != (%"HWADDR_PRIu" + %d", hdr->index, ptex, i);
2741             }
2742 
2743             if (n - i < valid) {
2744                 valid = n - i;
2745             }
2746             memcpy(hptes + i, hdr + 1, HASH_PTE_SIZE_64 * valid);
2747             i += valid;
2748 
2749             if ((n - i) < invalid) {
2750                 invalid = n - i;
2751             }
2752             memset(hptes + i, 0, invalid * HASH_PTE_SIZE_64);
2753             i += invalid;
2754 
2755             hdr = (struct kvm_get_htab_header *)
2756                 ((char *)(hdr + 1) + HASH_PTE_SIZE_64 * hdr->n_valid);
2757         }
2758     }
2759 
2760     close(fd);
2761 }
2762 
2763 void kvmppc_write_hpte(hwaddr ptex, uint64_t pte0, uint64_t pte1)
2764 {
2765     int fd, rc;
2766     struct {
2767         struct kvm_get_htab_header hdr;
2768         uint64_t pte0;
2769         uint64_t pte1;
2770     } buf;
2771 
2772     fd = kvmppc_get_htab_fd(true, 0 /* Ignored */, &error_abort);
2773 
2774     buf.hdr.n_valid = 1;
2775     buf.hdr.n_invalid = 0;
2776     buf.hdr.index = ptex;
2777     buf.pte0 = cpu_to_be64(pte0);
2778     buf.pte1 = cpu_to_be64(pte1);
2779 
2780     rc = write(fd, &buf, sizeof(buf));
2781     if (rc != sizeof(buf)) {
2782         hw_error("kvmppc_write_hpte: Unable to update KVM HPT");
2783     }
2784     close(fd);
2785 }
2786 
2787 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
2788                              uint64_t address, uint32_t data, PCIDevice *dev)
2789 {
2790     return 0;
2791 }
2792 
2793 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
2794                                 int vector, PCIDevice *dev)
2795 {
2796     return 0;
2797 }
2798 
2799 int kvm_arch_release_virq_post(int virq)
2800 {
2801     return 0;
2802 }
2803 
2804 int kvm_arch_msi_data_to_gsi(uint32_t data)
2805 {
2806     return data & 0xffff;
2807 }
2808 
2809 #if defined(TARGET_PPC64)
2810 int kvm_handle_nmi(PowerPCCPU *cpu, struct kvm_run *run)
2811 {
2812     bool recovered = run->flags & KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
2813 
2814     cpu_synchronize_state(CPU(cpu));
2815 
2816     spapr_mce_req_event(cpu, recovered);
2817 
2818     return 0;
2819 }
2820 #endif
2821 
2822 int kvmppc_enable_hwrng(void)
2823 {
2824     if (!kvm_enabled() || !kvm_check_extension(kvm_state, KVM_CAP_PPC_HWRNG)) {
2825         return -1;
2826     }
2827 
2828     return kvmppc_enable_hcall(kvm_state, H_RANDOM);
2829 }
2830 
2831 void kvmppc_check_papr_resize_hpt(Error **errp)
2832 {
2833     if (!kvm_enabled()) {
2834         return; /* No KVM, we're good */
2835     }
2836 
2837     if (cap_resize_hpt) {
2838         return; /* Kernel has explicit support, we're good */
2839     }
2840 
2841     /* Otherwise fallback on looking for PR KVM */
2842     if (kvmppc_is_pr(kvm_state)) {
2843         return;
2844     }
2845 
2846     error_setg(errp,
2847                "Hash page table resizing not available with this KVM version");
2848 }
2849 
2850 int kvmppc_resize_hpt_prepare(PowerPCCPU *cpu, target_ulong flags, int shift)
2851 {
2852     CPUState *cs = CPU(cpu);
2853     struct kvm_ppc_resize_hpt rhpt = {
2854         .flags = flags,
2855         .shift = shift,
2856     };
2857 
2858     if (!cap_resize_hpt) {
2859         return -ENOSYS;
2860     }
2861 
2862     return kvm_vm_ioctl(cs->kvm_state, KVM_PPC_RESIZE_HPT_PREPARE, &rhpt);
2863 }
2864 
2865 int kvmppc_resize_hpt_commit(PowerPCCPU *cpu, target_ulong flags, int shift)
2866 {
2867     CPUState *cs = CPU(cpu);
2868     struct kvm_ppc_resize_hpt rhpt = {
2869         .flags = flags,
2870         .shift = shift,
2871     };
2872 
2873     if (!cap_resize_hpt) {
2874         return -ENOSYS;
2875     }
2876 
2877     return kvm_vm_ioctl(cs->kvm_state, KVM_PPC_RESIZE_HPT_COMMIT, &rhpt);
2878 }
2879 
2880 /*
2881  * This is a helper function to detect a post migration scenario
2882  * in which a guest, running as KVM-HV, freezes in cpu_post_load because
2883  * the guest kernel can't handle a PVR value other than the actual host
2884  * PVR in KVM_SET_SREGS, even if pvr_match() returns true.
2885  *
2886  * If we don't have cap_ppc_pvr_compat and we're not running in PR
2887  * (so, we're HV), return true. The workaround itself is done in
2888  * cpu_post_load.
2889  *
2890  * The order here is important: we'll only check for KVM PR as a
2891  * fallback if the guest kernel can't handle the situation itself.
2892  * We need to avoid as much as possible querying the running KVM type
2893  * in QEMU level.
2894  */
2895 bool kvmppc_pvr_workaround_required(PowerPCCPU *cpu)
2896 {
2897     CPUState *cs = CPU(cpu);
2898 
2899     if (!kvm_enabled()) {
2900         return false;
2901     }
2902 
2903     if (cap_ppc_pvr_compat) {
2904         return false;
2905     }
2906 
2907     return !kvmppc_is_pr(cs->kvm_state);
2908 }
2909 
2910 void kvmppc_set_reg_ppc_online(PowerPCCPU *cpu, unsigned int online)
2911 {
2912     CPUState *cs = CPU(cpu);
2913 
2914     if (kvm_enabled()) {
2915         kvm_set_one_reg(cs, KVM_REG_PPC_ONLINE, &online);
2916     }
2917 }
2918 
2919 void kvmppc_set_reg_tb_offset(PowerPCCPU *cpu, int64_t tb_offset)
2920 {
2921     CPUState *cs = CPU(cpu);
2922 
2923     if (kvm_enabled()) {
2924         kvm_set_one_reg(cs, KVM_REG_PPC_TB_OFFSET, &tb_offset);
2925     }
2926 }
2927 
2928 /*
2929  * Don't set error if KVM_PPC_SVM_OFF ioctl is invoked on kernels
2930  * that don't support this ioctl.
2931  */
2932 void kvmppc_svm_off(Error **errp)
2933 {
2934     int rc;
2935 
2936     if (!kvm_enabled()) {
2937         return;
2938     }
2939 
2940     rc = kvm_vm_ioctl(KVM_STATE(current_accel()), KVM_PPC_SVM_OFF);
2941     if (rc && rc != -ENOTTY) {
2942         error_setg_errno(errp, -rc, "KVM_PPC_SVM_OFF ioctl failed");
2943     }
2944 }
2945