xref: /openbmc/qemu/target/ppc/kvm.c (revision 69242e7e)
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 static int cap_fwnmi;
92 static int cap_rpt_invalidate;
93 
94 static uint32_t debug_inst_opcode;
95 
96 /*
97  * Check whether we are running with KVM-PR (instead of KVM-HV).  This
98  * should only be used for fallback tests - generally we should use
99  * explicit capabilities for the features we want, rather than
100  * assuming what is/isn't available depending on the KVM variant.
101  */
102 static bool kvmppc_is_pr(KVMState *ks)
103 {
104     /* Assume KVM-PR if the GET_PVINFO capability is available */
105     return kvm_vm_check_extension(ks, KVM_CAP_PPC_GET_PVINFO) != 0;
106 }
107 
108 static int kvm_ppc_register_host_cpu_type(void);
109 static void kvmppc_get_cpu_characteristics(KVMState *s);
110 static int kvmppc_get_dec_bits(void);
111 
112 int kvm_arch_init(MachineState *ms, KVMState *s)
113 {
114     cap_interrupt_unset = kvm_check_extension(s, KVM_CAP_PPC_UNSET_IRQ);
115     cap_segstate = kvm_check_extension(s, KVM_CAP_PPC_SEGSTATE);
116     cap_booke_sregs = kvm_check_extension(s, KVM_CAP_PPC_BOOKE_SREGS);
117     cap_ppc_smt_possible = kvm_vm_check_extension(s, KVM_CAP_PPC_SMT_POSSIBLE);
118     cap_spapr_tce = kvm_check_extension(s, KVM_CAP_SPAPR_TCE);
119     cap_spapr_tce_64 = kvm_check_extension(s, KVM_CAP_SPAPR_TCE_64);
120     cap_spapr_multitce = kvm_check_extension(s, KVM_CAP_SPAPR_MULTITCE);
121     cap_spapr_vfio = kvm_vm_check_extension(s, KVM_CAP_SPAPR_TCE_VFIO);
122     cap_one_reg = kvm_check_extension(s, KVM_CAP_ONE_REG);
123     cap_hior = kvm_check_extension(s, KVM_CAP_PPC_HIOR);
124     cap_epr = kvm_check_extension(s, KVM_CAP_PPC_EPR);
125     cap_ppc_watchdog = kvm_check_extension(s, KVM_CAP_PPC_BOOKE_WATCHDOG);
126     /*
127      * Note: we don't set cap_papr here, because this capability is
128      * only activated after this by kvmppc_set_papr()
129      */
130     cap_htab_fd = kvm_vm_check_extension(s, KVM_CAP_PPC_HTAB_FD);
131     cap_fixup_hcalls = kvm_check_extension(s, KVM_CAP_PPC_FIXUP_HCALL);
132     cap_ppc_smt = kvm_vm_check_extension(s, KVM_CAP_PPC_SMT);
133     cap_htm = kvm_vm_check_extension(s, KVM_CAP_PPC_HTM);
134     cap_mmu_radix = kvm_vm_check_extension(s, KVM_CAP_PPC_MMU_RADIX);
135     cap_mmu_hash_v3 = kvm_vm_check_extension(s, KVM_CAP_PPC_MMU_HASH_V3);
136     cap_xive = kvm_vm_check_extension(s, KVM_CAP_PPC_IRQ_XIVE);
137     cap_resize_hpt = kvm_vm_check_extension(s, KVM_CAP_SPAPR_RESIZE_HPT);
138     kvmppc_get_cpu_characteristics(s);
139     cap_ppc_nested_kvm_hv = kvm_vm_check_extension(s, KVM_CAP_PPC_NESTED_HV);
140     cap_large_decr = kvmppc_get_dec_bits();
141     cap_fwnmi = kvm_vm_check_extension(s, KVM_CAP_PPC_FWNMI);
142     /*
143      * Note: setting it to false because there is not such capability
144      * in KVM at this moment.
145      *
146      * TODO: call kvm_vm_check_extension() with the right capability
147      * after the kernel starts implementing it.
148      */
149     cap_ppc_pvr_compat = false;
150 
151     if (!kvm_check_extension(s, KVM_CAP_PPC_IRQ_LEVEL)) {
152         error_report("KVM: Host kernel doesn't have level irq capability");
153         exit(1);
154     }
155 
156     cap_rpt_invalidate = kvm_vm_check_extension(s, KVM_CAP_PPC_RPT_INVALIDATE);
157     kvm_ppc_register_host_cpu_type();
158 
159     return 0;
160 }
161 
162 int kvm_arch_irqchip_create(KVMState *s)
163 {
164     return 0;
165 }
166 
167 static int kvm_arch_sync_sregs(PowerPCCPU *cpu)
168 {
169     CPUPPCState *cenv = &cpu->env;
170     CPUState *cs = CPU(cpu);
171     struct kvm_sregs sregs;
172     int ret;
173 
174     if (cenv->excp_model == POWERPC_EXCP_BOOKE) {
175         /*
176          * What we're really trying to say is "if we're on BookE, we
177          * use the native PVR for now". This is the only sane way to
178          * check it though, so we potentially confuse users that they
179          * can run BookE guests on BookS. Let's hope nobody dares
180          * enough :)
181          */
182         return 0;
183     } else {
184         if (!cap_segstate) {
185             fprintf(stderr, "kvm error: missing PVR setting capability\n");
186             return -ENOSYS;
187         }
188     }
189 
190     ret = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs);
191     if (ret) {
192         return ret;
193     }
194 
195     sregs.pvr = cenv->spr[SPR_PVR];
196     return kvm_vcpu_ioctl(cs, KVM_SET_SREGS, &sregs);
197 }
198 
199 /* Set up a shared TLB array with KVM */
200 static int kvm_booke206_tlb_init(PowerPCCPU *cpu)
201 {
202     CPUPPCState *env = &cpu->env;
203     CPUState *cs = CPU(cpu);
204     struct kvm_book3e_206_tlb_params params = {};
205     struct kvm_config_tlb cfg = {};
206     unsigned int entries = 0;
207     int ret, i;
208 
209     if (!kvm_enabled() ||
210         !kvm_check_extension(cs->kvm_state, KVM_CAP_SW_TLB)) {
211         return 0;
212     }
213 
214     assert(ARRAY_SIZE(params.tlb_sizes) == BOOKE206_MAX_TLBN);
215 
216     for (i = 0; i < BOOKE206_MAX_TLBN; i++) {
217         params.tlb_sizes[i] = booke206_tlb_size(env, i);
218         params.tlb_ways[i] = booke206_tlb_ways(env, i);
219         entries += params.tlb_sizes[i];
220     }
221 
222     assert(entries == env->nb_tlb);
223     assert(sizeof(struct kvm_book3e_206_tlb_entry) == sizeof(ppcmas_tlb_t));
224 
225     env->tlb_dirty = true;
226 
227     cfg.array = (uintptr_t)env->tlb.tlbm;
228     cfg.array_len = sizeof(ppcmas_tlb_t) * entries;
229     cfg.params = (uintptr_t)&params;
230     cfg.mmu_type = KVM_MMU_FSL_BOOKE_NOHV;
231 
232     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_SW_TLB, 0, (uintptr_t)&cfg);
233     if (ret < 0) {
234         fprintf(stderr, "%s: couldn't enable KVM_CAP_SW_TLB: %s\n",
235                 __func__, strerror(-ret));
236         return ret;
237     }
238 
239     env->kvm_sw_tlb = true;
240     return 0;
241 }
242 
243 
244 #if defined(TARGET_PPC64)
245 static void kvm_get_smmu_info(struct kvm_ppc_smmu_info *info, Error **errp)
246 {
247     int ret;
248 
249     assert(kvm_state != NULL);
250 
251     if (!kvm_check_extension(kvm_state, KVM_CAP_PPC_GET_SMMU_INFO)) {
252         error_setg(errp, "KVM doesn't expose the MMU features it supports");
253         error_append_hint(errp, "Consider switching to a newer KVM\n");
254         return;
255     }
256 
257     ret = kvm_vm_ioctl(kvm_state, KVM_PPC_GET_SMMU_INFO, info);
258     if (ret == 0) {
259         return;
260     }
261 
262     error_setg_errno(errp, -ret,
263                      "KVM failed to provide the MMU features it supports");
264 }
265 
266 struct ppc_radix_page_info *kvm_get_radix_page_info(void)
267 {
268     KVMState *s = KVM_STATE(current_accel());
269     struct ppc_radix_page_info *radix_page_info;
270     struct kvm_ppc_rmmu_info rmmu_info;
271     int i;
272 
273     if (!kvm_check_extension(s, KVM_CAP_PPC_MMU_RADIX)) {
274         return NULL;
275     }
276     if (kvm_vm_ioctl(s, KVM_PPC_GET_RMMU_INFO, &rmmu_info)) {
277         return NULL;
278     }
279     radix_page_info = g_malloc0(sizeof(*radix_page_info));
280     radix_page_info->count = 0;
281     for (i = 0; i < PPC_PAGE_SIZES_MAX_SZ; i++) {
282         if (rmmu_info.ap_encodings[i]) {
283             radix_page_info->entries[i] = rmmu_info.ap_encodings[i];
284             radix_page_info->count++;
285         }
286     }
287     return radix_page_info;
288 }
289 
290 target_ulong kvmppc_configure_v3_mmu(PowerPCCPU *cpu,
291                                      bool radix, bool gtse,
292                                      uint64_t proc_tbl)
293 {
294     CPUState *cs = CPU(cpu);
295     int ret;
296     uint64_t flags = 0;
297     struct kvm_ppc_mmuv3_cfg cfg = {
298         .process_table = proc_tbl,
299     };
300 
301     if (radix) {
302         flags |= KVM_PPC_MMUV3_RADIX;
303     }
304     if (gtse) {
305         flags |= KVM_PPC_MMUV3_GTSE;
306     }
307     cfg.flags = flags;
308     ret = kvm_vm_ioctl(cs->kvm_state, KVM_PPC_CONFIGURE_V3_MMU, &cfg);
309     switch (ret) {
310     case 0:
311         return H_SUCCESS;
312     case -EINVAL:
313         return H_PARAMETER;
314     case -ENODEV:
315         return H_NOT_AVAILABLE;
316     default:
317         return H_HARDWARE;
318     }
319 }
320 
321 bool kvmppc_hpt_needs_host_contiguous_pages(void)
322 {
323     static struct kvm_ppc_smmu_info smmu_info;
324 
325     if (!kvm_enabled()) {
326         return false;
327     }
328 
329     kvm_get_smmu_info(&smmu_info, &error_fatal);
330     return !!(smmu_info.flags & KVM_PPC_PAGE_SIZES_REAL);
331 }
332 
333 void kvm_check_mmu(PowerPCCPU *cpu, Error **errp)
334 {
335     struct kvm_ppc_smmu_info smmu_info;
336     int iq, ik, jq, jk;
337     Error *local_err = NULL;
338 
339     /* For now, we only have anything to check on hash64 MMUs */
340     if (!cpu->hash64_opts || !kvm_enabled()) {
341         return;
342     }
343 
344     kvm_get_smmu_info(&smmu_info, &local_err);
345     if (local_err) {
346         error_propagate(errp, local_err);
347         return;
348     }
349 
350     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)
351         && !(smmu_info.flags & KVM_PPC_1T_SEGMENTS)) {
352         error_setg(errp,
353                    "KVM does not support 1TiB segments which guest expects");
354         return;
355     }
356 
357     if (smmu_info.slb_size < cpu->hash64_opts->slb_size) {
358         error_setg(errp, "KVM only supports %u SLB entries, but guest needs %u",
359                    smmu_info.slb_size, cpu->hash64_opts->slb_size);
360         return;
361     }
362 
363     /*
364      * Verify that every pagesize supported by the cpu model is
365      * supported by KVM with the same encodings
366      */
367     for (iq = 0; iq < ARRAY_SIZE(cpu->hash64_opts->sps); iq++) {
368         PPCHash64SegmentPageSizes *qsps = &cpu->hash64_opts->sps[iq];
369         struct kvm_ppc_one_seg_page_size *ksps;
370 
371         for (ik = 0; ik < ARRAY_SIZE(smmu_info.sps); ik++) {
372             if (qsps->page_shift == smmu_info.sps[ik].page_shift) {
373                 break;
374             }
375         }
376         if (ik >= ARRAY_SIZE(smmu_info.sps)) {
377             error_setg(errp, "KVM doesn't support for base page shift %u",
378                        qsps->page_shift);
379             return;
380         }
381 
382         ksps = &smmu_info.sps[ik];
383         if (ksps->slb_enc != qsps->slb_enc) {
384             error_setg(errp,
385 "KVM uses SLB encoding 0x%x for page shift %u, but guest expects 0x%x",
386                        ksps->slb_enc, ksps->page_shift, qsps->slb_enc);
387             return;
388         }
389 
390         for (jq = 0; jq < ARRAY_SIZE(qsps->enc); jq++) {
391             for (jk = 0; jk < ARRAY_SIZE(ksps->enc); jk++) {
392                 if (qsps->enc[jq].page_shift == ksps->enc[jk].page_shift) {
393                     break;
394                 }
395             }
396 
397             if (jk >= ARRAY_SIZE(ksps->enc)) {
398                 error_setg(errp, "KVM doesn't support page shift %u/%u",
399                            qsps->enc[jq].page_shift, qsps->page_shift);
400                 return;
401             }
402             if (qsps->enc[jq].pte_enc != ksps->enc[jk].pte_enc) {
403                 error_setg(errp,
404 "KVM uses PTE encoding 0x%x for page shift %u/%u, but guest expects 0x%x",
405                            ksps->enc[jk].pte_enc, qsps->enc[jq].page_shift,
406                            qsps->page_shift, qsps->enc[jq].pte_enc);
407                 return;
408             }
409         }
410     }
411 
412     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
413         /*
414          * Mostly what guest pagesizes we can use are related to the
415          * host pages used to map guest RAM, which is handled in the
416          * platform code. Cache-Inhibited largepages (64k) however are
417          * used for I/O, so if they're mapped to the host at all it
418          * will be a normal mapping, not a special hugepage one used
419          * for RAM.
420          */
421         if (qemu_real_host_page_size() < 0x10000) {
422             error_setg(errp,
423                        "KVM can't supply 64kiB CI pages, which guest expects");
424         }
425     }
426 }
427 #endif /* !defined (TARGET_PPC64) */
428 
429 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
430 {
431     return POWERPC_CPU(cpu)->vcpu_id;
432 }
433 
434 /*
435  * e500 supports 2 h/w breakpoint and 2 watchpoint.  book3s supports
436  * only 1 watchpoint, so array size of 4 is sufficient for now.
437  */
438 #define MAX_HW_BKPTS 4
439 
440 static struct HWBreakpoint {
441     target_ulong addr;
442     int type;
443 } hw_debug_points[MAX_HW_BKPTS];
444 
445 static CPUWatchpoint hw_watchpoint;
446 
447 /* Default there is no breakpoint and watchpoint supported */
448 static int max_hw_breakpoint;
449 static int max_hw_watchpoint;
450 static int nb_hw_breakpoint;
451 static int nb_hw_watchpoint;
452 
453 static void kvmppc_hw_debug_points_init(CPUPPCState *cenv)
454 {
455     if (cenv->excp_model == POWERPC_EXCP_BOOKE) {
456         max_hw_breakpoint = 2;
457         max_hw_watchpoint = 2;
458     }
459 
460     if ((max_hw_breakpoint + max_hw_watchpoint) > MAX_HW_BKPTS) {
461         fprintf(stderr, "Error initializing h/w breakpoints\n");
462         return;
463     }
464 }
465 
466 int kvm_arch_init_vcpu(CPUState *cs)
467 {
468     PowerPCCPU *cpu = POWERPC_CPU(cs);
469     CPUPPCState *cenv = &cpu->env;
470     int ret;
471 
472     /* Synchronize sregs with kvm */
473     ret = kvm_arch_sync_sregs(cpu);
474     if (ret) {
475         if (ret == -EINVAL) {
476             error_report("Register sync failed... If you're using kvm-hv.ko,"
477                          " only \"-cpu host\" is possible");
478         }
479         return ret;
480     }
481 
482     switch (cenv->mmu_model) {
483     case POWERPC_MMU_BOOKE206:
484         /* This target supports access to KVM's guest TLB */
485         ret = kvm_booke206_tlb_init(cpu);
486         break;
487     case POWERPC_MMU_2_07:
488         if (!cap_htm && !kvmppc_is_pr(cs->kvm_state)) {
489             /*
490              * KVM-HV has transactional memory on POWER8 also without
491              * the KVM_CAP_PPC_HTM extension, so enable it here
492              * instead as long as it's available to userspace on the
493              * host.
494              */
495             if (qemu_getauxval(AT_HWCAP2) & PPC_FEATURE2_HAS_HTM) {
496                 cap_htm = true;
497             }
498         }
499         break;
500     default:
501         break;
502     }
503 
504     kvm_get_one_reg(cs, KVM_REG_PPC_DEBUG_INST, &debug_inst_opcode);
505     kvmppc_hw_debug_points_init(cenv);
506 
507     return ret;
508 }
509 
510 int kvm_arch_destroy_vcpu(CPUState *cs)
511 {
512     return 0;
513 }
514 
515 static void kvm_sw_tlb_put(PowerPCCPU *cpu)
516 {
517     CPUPPCState *env = &cpu->env;
518     CPUState *cs = CPU(cpu);
519     struct kvm_dirty_tlb dirty_tlb;
520     unsigned char *bitmap;
521     int ret;
522 
523     if (!env->kvm_sw_tlb) {
524         return;
525     }
526 
527     bitmap = g_malloc((env->nb_tlb + 7) / 8);
528     memset(bitmap, 0xFF, (env->nb_tlb + 7) / 8);
529 
530     dirty_tlb.bitmap = (uintptr_t)bitmap;
531     dirty_tlb.num_dirty = env->nb_tlb;
532 
533     ret = kvm_vcpu_ioctl(cs, KVM_DIRTY_TLB, &dirty_tlb);
534     if (ret) {
535         fprintf(stderr, "%s: KVM_DIRTY_TLB: %s\n",
536                 __func__, strerror(-ret));
537     }
538 
539     g_free(bitmap);
540 }
541 
542 static void kvm_get_one_spr(CPUState *cs, uint64_t id, int spr)
543 {
544     PowerPCCPU *cpu = POWERPC_CPU(cs);
545     CPUPPCState *env = &cpu->env;
546     union {
547         uint32_t u32;
548         uint64_t u64;
549     } val;
550     struct kvm_one_reg reg = {
551         .id = id,
552         .addr = (uintptr_t) &val,
553     };
554     int ret;
555 
556     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
557     if (ret != 0) {
558         trace_kvm_failed_spr_get(spr, strerror(errno));
559     } else {
560         switch (id & KVM_REG_SIZE_MASK) {
561         case KVM_REG_SIZE_U32:
562             env->spr[spr] = val.u32;
563             break;
564 
565         case KVM_REG_SIZE_U64:
566             env->spr[spr] = val.u64;
567             break;
568 
569         default:
570             /* Don't handle this size yet */
571             abort();
572         }
573     }
574 }
575 
576 static void kvm_put_one_spr(CPUState *cs, uint64_t id, int spr)
577 {
578     PowerPCCPU *cpu = POWERPC_CPU(cs);
579     CPUPPCState *env = &cpu->env;
580     union {
581         uint32_t u32;
582         uint64_t u64;
583     } val;
584     struct kvm_one_reg reg = {
585         .id = id,
586         .addr = (uintptr_t) &val,
587     };
588     int ret;
589 
590     switch (id & KVM_REG_SIZE_MASK) {
591     case KVM_REG_SIZE_U32:
592         val.u32 = env->spr[spr];
593         break;
594 
595     case KVM_REG_SIZE_U64:
596         val.u64 = env->spr[spr];
597         break;
598 
599     default:
600         /* Don't handle this size yet */
601         abort();
602     }
603 
604     ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
605     if (ret != 0) {
606         trace_kvm_failed_spr_set(spr, strerror(errno));
607     }
608 }
609 
610 static int kvm_put_fp(CPUState *cs)
611 {
612     PowerPCCPU *cpu = POWERPC_CPU(cs);
613     CPUPPCState *env = &cpu->env;
614     struct kvm_one_reg reg;
615     int i;
616     int ret;
617 
618     if (env->insns_flags & PPC_FLOAT) {
619         uint64_t fpscr = env->fpscr;
620         bool vsx = !!(env->insns_flags2 & PPC2_VSX);
621 
622         reg.id = KVM_REG_PPC_FPSCR;
623         reg.addr = (uintptr_t)&fpscr;
624         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
625         if (ret < 0) {
626             trace_kvm_failed_fpscr_set(strerror(errno));
627             return ret;
628         }
629 
630         for (i = 0; i < 32; i++) {
631             uint64_t vsr[2];
632             uint64_t *fpr = cpu_fpr_ptr(&cpu->env, i);
633             uint64_t *vsrl = cpu_vsrl_ptr(&cpu->env, i);
634 
635 #if HOST_BIG_ENDIAN
636             vsr[0] = float64_val(*fpr);
637             vsr[1] = *vsrl;
638 #else
639             vsr[0] = *vsrl;
640             vsr[1] = float64_val(*fpr);
641 #endif
642             reg.addr = (uintptr_t) &vsr;
643             reg.id = vsx ? KVM_REG_PPC_VSR(i) : KVM_REG_PPC_FPR(i);
644 
645             ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
646             if (ret < 0) {
647                 trace_kvm_failed_fp_set(vsx ? "VSR" : "FPR", i,
648                                         strerror(errno));
649                 return ret;
650             }
651         }
652     }
653 
654     if (env->insns_flags & PPC_ALTIVEC) {
655         reg.id = KVM_REG_PPC_VSCR;
656         reg.addr = (uintptr_t)&env->vscr;
657         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
658         if (ret < 0) {
659             trace_kvm_failed_vscr_set(strerror(errno));
660             return ret;
661         }
662 
663         for (i = 0; i < 32; i++) {
664             reg.id = KVM_REG_PPC_VR(i);
665             reg.addr = (uintptr_t)cpu_avr_ptr(env, i);
666             ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
667             if (ret < 0) {
668                 trace_kvm_failed_vr_set(i, strerror(errno));
669                 return ret;
670             }
671         }
672     }
673 
674     return 0;
675 }
676 
677 static int kvm_get_fp(CPUState *cs)
678 {
679     PowerPCCPU *cpu = POWERPC_CPU(cs);
680     CPUPPCState *env = &cpu->env;
681     struct kvm_one_reg reg;
682     int i;
683     int ret;
684 
685     if (env->insns_flags & PPC_FLOAT) {
686         uint64_t fpscr;
687         bool vsx = !!(env->insns_flags2 & PPC2_VSX);
688 
689         reg.id = KVM_REG_PPC_FPSCR;
690         reg.addr = (uintptr_t)&fpscr;
691         ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
692         if (ret < 0) {
693             trace_kvm_failed_fpscr_get(strerror(errno));
694             return ret;
695         } else {
696             env->fpscr = fpscr;
697         }
698 
699         for (i = 0; i < 32; i++) {
700             uint64_t vsr[2];
701             uint64_t *fpr = cpu_fpr_ptr(&cpu->env, i);
702             uint64_t *vsrl = cpu_vsrl_ptr(&cpu->env, i);
703 
704             reg.addr = (uintptr_t) &vsr;
705             reg.id = vsx ? KVM_REG_PPC_VSR(i) : KVM_REG_PPC_FPR(i);
706 
707             ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
708             if (ret < 0) {
709                 trace_kvm_failed_fp_get(vsx ? "VSR" : "FPR", i,
710                                         strerror(errno));
711                 return ret;
712             } else {
713 #if HOST_BIG_ENDIAN
714                 *fpr = vsr[0];
715                 if (vsx) {
716                     *vsrl = vsr[1];
717                 }
718 #else
719                 *fpr = vsr[1];
720                 if (vsx) {
721                     *vsrl = vsr[0];
722                 }
723 #endif
724             }
725         }
726     }
727 
728     if (env->insns_flags & PPC_ALTIVEC) {
729         reg.id = KVM_REG_PPC_VSCR;
730         reg.addr = (uintptr_t)&env->vscr;
731         ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
732         if (ret < 0) {
733             trace_kvm_failed_vscr_get(strerror(errno));
734             return ret;
735         }
736 
737         for (i = 0; i < 32; i++) {
738             reg.id = KVM_REG_PPC_VR(i);
739             reg.addr = (uintptr_t)cpu_avr_ptr(env, i);
740             ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
741             if (ret < 0) {
742                 trace_kvm_failed_vr_get(i, strerror(errno));
743                 return ret;
744             }
745         }
746     }
747 
748     return 0;
749 }
750 
751 #if defined(TARGET_PPC64)
752 static int kvm_get_vpa(CPUState *cs)
753 {
754     PowerPCCPU *cpu = POWERPC_CPU(cs);
755     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
756     struct kvm_one_reg reg;
757     int ret;
758 
759     reg.id = KVM_REG_PPC_VPA_ADDR;
760     reg.addr = (uintptr_t)&spapr_cpu->vpa_addr;
761     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
762     if (ret < 0) {
763         trace_kvm_failed_vpa_addr_get(strerror(errno));
764         return ret;
765     }
766 
767     assert((uintptr_t)&spapr_cpu->slb_shadow_size
768            == ((uintptr_t)&spapr_cpu->slb_shadow_addr + 8));
769     reg.id = KVM_REG_PPC_VPA_SLB;
770     reg.addr = (uintptr_t)&spapr_cpu->slb_shadow_addr;
771     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
772     if (ret < 0) {
773         trace_kvm_failed_slb_get(strerror(errno));
774         return ret;
775     }
776 
777     assert((uintptr_t)&spapr_cpu->dtl_size
778            == ((uintptr_t)&spapr_cpu->dtl_addr + 8));
779     reg.id = KVM_REG_PPC_VPA_DTL;
780     reg.addr = (uintptr_t)&spapr_cpu->dtl_addr;
781     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
782     if (ret < 0) {
783         trace_kvm_failed_dtl_get(strerror(errno));
784         return ret;
785     }
786 
787     return 0;
788 }
789 
790 static int kvm_put_vpa(CPUState *cs)
791 {
792     PowerPCCPU *cpu = POWERPC_CPU(cs);
793     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
794     struct kvm_one_reg reg;
795     int ret;
796 
797     /*
798      * SLB shadow or DTL can't be registered unless a master VPA is
799      * registered.  That means when restoring state, if a VPA *is*
800      * registered, we need to set that up first.  If not, we need to
801      * deregister the others before deregistering the master VPA
802      */
803     assert(spapr_cpu->vpa_addr
804            || !(spapr_cpu->slb_shadow_addr || spapr_cpu->dtl_addr));
805 
806     if (spapr_cpu->vpa_addr) {
807         reg.id = KVM_REG_PPC_VPA_ADDR;
808         reg.addr = (uintptr_t)&spapr_cpu->vpa_addr;
809         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
810         if (ret < 0) {
811             trace_kvm_failed_vpa_addr_set(strerror(errno));
812             return ret;
813         }
814     }
815 
816     assert((uintptr_t)&spapr_cpu->slb_shadow_size
817            == ((uintptr_t)&spapr_cpu->slb_shadow_addr + 8));
818     reg.id = KVM_REG_PPC_VPA_SLB;
819     reg.addr = (uintptr_t)&spapr_cpu->slb_shadow_addr;
820     ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
821     if (ret < 0) {
822         trace_kvm_failed_slb_set(strerror(errno));
823         return ret;
824     }
825 
826     assert((uintptr_t)&spapr_cpu->dtl_size
827            == ((uintptr_t)&spapr_cpu->dtl_addr + 8));
828     reg.id = KVM_REG_PPC_VPA_DTL;
829     reg.addr = (uintptr_t)&spapr_cpu->dtl_addr;
830     ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
831     if (ret < 0) {
832         trace_kvm_failed_dtl_set(strerror(errno));
833         return ret;
834     }
835 
836     if (!spapr_cpu->vpa_addr) {
837         reg.id = KVM_REG_PPC_VPA_ADDR;
838         reg.addr = (uintptr_t)&spapr_cpu->vpa_addr;
839         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
840         if (ret < 0) {
841             trace_kvm_failed_null_vpa_addr_set(strerror(errno));
842             return ret;
843         }
844     }
845 
846     return 0;
847 }
848 #endif /* TARGET_PPC64 */
849 
850 int kvmppc_put_books_sregs(PowerPCCPU *cpu)
851 {
852     CPUPPCState *env = &cpu->env;
853     struct kvm_sregs sregs;
854     int i;
855 
856     sregs.pvr = env->spr[SPR_PVR];
857 
858     if (cpu->vhyp) {
859         PPCVirtualHypervisorClass *vhc =
860             PPC_VIRTUAL_HYPERVISOR_GET_CLASS(cpu->vhyp);
861         sregs.u.s.sdr1 = vhc->encode_hpt_for_kvm_pr(cpu->vhyp);
862     } else {
863         sregs.u.s.sdr1 = env->spr[SPR_SDR1];
864     }
865 
866     /* Sync SLB */
867 #ifdef TARGET_PPC64
868     for (i = 0; i < ARRAY_SIZE(env->slb); i++) {
869         sregs.u.s.ppc64.slb[i].slbe = env->slb[i].esid;
870         if (env->slb[i].esid & SLB_ESID_V) {
871             sregs.u.s.ppc64.slb[i].slbe |= i;
872         }
873         sregs.u.s.ppc64.slb[i].slbv = env->slb[i].vsid;
874     }
875 #endif
876 
877     /* Sync SRs */
878     for (i = 0; i < 16; i++) {
879         sregs.u.s.ppc32.sr[i] = env->sr[i];
880     }
881 
882     /* Sync BATs */
883     for (i = 0; i < 8; i++) {
884         /* Beware. We have to swap upper and lower bits here */
885         sregs.u.s.ppc32.dbat[i] = ((uint64_t)env->DBAT[0][i] << 32)
886             | env->DBAT[1][i];
887         sregs.u.s.ppc32.ibat[i] = ((uint64_t)env->IBAT[0][i] << 32)
888             | env->IBAT[1][i];
889     }
890 
891     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_SREGS, &sregs);
892 }
893 
894 int kvm_arch_put_registers(CPUState *cs, int level)
895 {
896     PowerPCCPU *cpu = POWERPC_CPU(cs);
897     CPUPPCState *env = &cpu->env;
898     struct kvm_regs regs;
899     int ret;
900     int i;
901 
902     ret = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
903     if (ret < 0) {
904         return ret;
905     }
906 
907     regs.ctr = env->ctr;
908     regs.lr  = env->lr;
909     regs.xer = cpu_read_xer(env);
910     regs.msr = env->msr;
911     regs.pc = env->nip;
912 
913     regs.srr0 = env->spr[SPR_SRR0];
914     regs.srr1 = env->spr[SPR_SRR1];
915 
916     regs.sprg0 = env->spr[SPR_SPRG0];
917     regs.sprg1 = env->spr[SPR_SPRG1];
918     regs.sprg2 = env->spr[SPR_SPRG2];
919     regs.sprg3 = env->spr[SPR_SPRG3];
920     regs.sprg4 = env->spr[SPR_SPRG4];
921     regs.sprg5 = env->spr[SPR_SPRG5];
922     regs.sprg6 = env->spr[SPR_SPRG6];
923     regs.sprg7 = env->spr[SPR_SPRG7];
924 
925     regs.pid = env->spr[SPR_BOOKE_PID];
926 
927     for (i = 0; i < 32; i++) {
928         regs.gpr[i] = env->gpr[i];
929     }
930 
931     regs.cr = 0;
932     for (i = 0; i < 8; i++) {
933         regs.cr |= (env->crf[i] & 15) << (4 * (7 - i));
934     }
935 
936     ret = kvm_vcpu_ioctl(cs, KVM_SET_REGS, &regs);
937     if (ret < 0) {
938         return ret;
939     }
940 
941     kvm_put_fp(cs);
942 
943     if (env->tlb_dirty) {
944         kvm_sw_tlb_put(cpu);
945         env->tlb_dirty = false;
946     }
947 
948     if (cap_segstate && (level >= KVM_PUT_RESET_STATE)) {
949         ret = kvmppc_put_books_sregs(cpu);
950         if (ret < 0) {
951             return ret;
952         }
953     }
954 
955     if (cap_hior && (level >= KVM_PUT_RESET_STATE)) {
956         kvm_put_one_spr(cs, KVM_REG_PPC_HIOR, SPR_HIOR);
957     }
958 
959     if (cap_one_reg) {
960         int i;
961 
962         /*
963          * We deliberately ignore errors here, for kernels which have
964          * the ONE_REG calls, but don't support the specific
965          * registers, there's a reasonable chance things will still
966          * work, at least until we try to migrate.
967          */
968         for (i = 0; i < 1024; i++) {
969             uint64_t id = env->spr_cb[i].one_reg_id;
970 
971             if (id != 0) {
972                 kvm_put_one_spr(cs, id, i);
973             }
974         }
975 
976 #ifdef TARGET_PPC64
977         if (msr_ts) {
978             for (i = 0; i < ARRAY_SIZE(env->tm_gpr); i++) {
979                 kvm_set_one_reg(cs, KVM_REG_PPC_TM_GPR(i), &env->tm_gpr[i]);
980             }
981             for (i = 0; i < ARRAY_SIZE(env->tm_vsr); i++) {
982                 kvm_set_one_reg(cs, KVM_REG_PPC_TM_VSR(i), &env->tm_vsr[i]);
983             }
984             kvm_set_one_reg(cs, KVM_REG_PPC_TM_CR, &env->tm_cr);
985             kvm_set_one_reg(cs, KVM_REG_PPC_TM_LR, &env->tm_lr);
986             kvm_set_one_reg(cs, KVM_REG_PPC_TM_CTR, &env->tm_ctr);
987             kvm_set_one_reg(cs, KVM_REG_PPC_TM_FPSCR, &env->tm_fpscr);
988             kvm_set_one_reg(cs, KVM_REG_PPC_TM_AMR, &env->tm_amr);
989             kvm_set_one_reg(cs, KVM_REG_PPC_TM_PPR, &env->tm_ppr);
990             kvm_set_one_reg(cs, KVM_REG_PPC_TM_VRSAVE, &env->tm_vrsave);
991             kvm_set_one_reg(cs, KVM_REG_PPC_TM_VSCR, &env->tm_vscr);
992             kvm_set_one_reg(cs, KVM_REG_PPC_TM_DSCR, &env->tm_dscr);
993             kvm_set_one_reg(cs, KVM_REG_PPC_TM_TAR, &env->tm_tar);
994         }
995 
996         if (cap_papr) {
997             if (kvm_put_vpa(cs) < 0) {
998                 trace_kvm_failed_put_vpa();
999             }
1000         }
1001 
1002         kvm_set_one_reg(cs, KVM_REG_PPC_TB_OFFSET, &env->tb_env->tb_offset);
1003 
1004         if (level > KVM_PUT_RUNTIME_STATE) {
1005             kvm_put_one_spr(cs, KVM_REG_PPC_DPDES, SPR_DPDES);
1006         }
1007 #endif /* TARGET_PPC64 */
1008     }
1009 
1010     return ret;
1011 }
1012 
1013 static void kvm_sync_excp(CPUPPCState *env, int vector, int ivor)
1014 {
1015      env->excp_vectors[vector] = env->spr[ivor] + env->spr[SPR_BOOKE_IVPR];
1016 }
1017 
1018 static int kvmppc_get_booke_sregs(PowerPCCPU *cpu)
1019 {
1020     CPUPPCState *env = &cpu->env;
1021     struct kvm_sregs sregs;
1022     int ret;
1023 
1024     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_SREGS, &sregs);
1025     if (ret < 0) {
1026         return ret;
1027     }
1028 
1029     if (sregs.u.e.features & KVM_SREGS_E_BASE) {
1030         env->spr[SPR_BOOKE_CSRR0] = sregs.u.e.csrr0;
1031         env->spr[SPR_BOOKE_CSRR1] = sregs.u.e.csrr1;
1032         env->spr[SPR_BOOKE_ESR] = sregs.u.e.esr;
1033         env->spr[SPR_BOOKE_DEAR] = sregs.u.e.dear;
1034         env->spr[SPR_BOOKE_MCSR] = sregs.u.e.mcsr;
1035         env->spr[SPR_BOOKE_TSR] = sregs.u.e.tsr;
1036         env->spr[SPR_BOOKE_TCR] = sregs.u.e.tcr;
1037         env->spr[SPR_DECR] = sregs.u.e.dec;
1038         env->spr[SPR_TBL] = sregs.u.e.tb & 0xffffffff;
1039         env->spr[SPR_TBU] = sregs.u.e.tb >> 32;
1040         env->spr[SPR_VRSAVE] = sregs.u.e.vrsave;
1041     }
1042 
1043     if (sregs.u.e.features & KVM_SREGS_E_ARCH206) {
1044         env->spr[SPR_BOOKE_PIR] = sregs.u.e.pir;
1045         env->spr[SPR_BOOKE_MCSRR0] = sregs.u.e.mcsrr0;
1046         env->spr[SPR_BOOKE_MCSRR1] = sregs.u.e.mcsrr1;
1047         env->spr[SPR_BOOKE_DECAR] = sregs.u.e.decar;
1048         env->spr[SPR_BOOKE_IVPR] = sregs.u.e.ivpr;
1049     }
1050 
1051     if (sregs.u.e.features & KVM_SREGS_E_64) {
1052         env->spr[SPR_BOOKE_EPCR] = sregs.u.e.epcr;
1053     }
1054 
1055     if (sregs.u.e.features & KVM_SREGS_E_SPRG8) {
1056         env->spr[SPR_BOOKE_SPRG8] = sregs.u.e.sprg8;
1057     }
1058 
1059     if (sregs.u.e.features & KVM_SREGS_E_IVOR) {
1060         env->spr[SPR_BOOKE_IVOR0] = sregs.u.e.ivor_low[0];
1061         kvm_sync_excp(env, POWERPC_EXCP_CRITICAL,  SPR_BOOKE_IVOR0);
1062         env->spr[SPR_BOOKE_IVOR1] = sregs.u.e.ivor_low[1];
1063         kvm_sync_excp(env, POWERPC_EXCP_MCHECK,  SPR_BOOKE_IVOR1);
1064         env->spr[SPR_BOOKE_IVOR2] = sregs.u.e.ivor_low[2];
1065         kvm_sync_excp(env, POWERPC_EXCP_DSI,  SPR_BOOKE_IVOR2);
1066         env->spr[SPR_BOOKE_IVOR3] = sregs.u.e.ivor_low[3];
1067         kvm_sync_excp(env, POWERPC_EXCP_ISI,  SPR_BOOKE_IVOR3);
1068         env->spr[SPR_BOOKE_IVOR4] = sregs.u.e.ivor_low[4];
1069         kvm_sync_excp(env, POWERPC_EXCP_EXTERNAL,  SPR_BOOKE_IVOR4);
1070         env->spr[SPR_BOOKE_IVOR5] = sregs.u.e.ivor_low[5];
1071         kvm_sync_excp(env, POWERPC_EXCP_ALIGN,  SPR_BOOKE_IVOR5);
1072         env->spr[SPR_BOOKE_IVOR6] = sregs.u.e.ivor_low[6];
1073         kvm_sync_excp(env, POWERPC_EXCP_PROGRAM,  SPR_BOOKE_IVOR6);
1074         env->spr[SPR_BOOKE_IVOR7] = sregs.u.e.ivor_low[7];
1075         kvm_sync_excp(env, POWERPC_EXCP_FPU,  SPR_BOOKE_IVOR7);
1076         env->spr[SPR_BOOKE_IVOR8] = sregs.u.e.ivor_low[8];
1077         kvm_sync_excp(env, POWERPC_EXCP_SYSCALL,  SPR_BOOKE_IVOR8);
1078         env->spr[SPR_BOOKE_IVOR9] = sregs.u.e.ivor_low[9];
1079         kvm_sync_excp(env, POWERPC_EXCP_APU,  SPR_BOOKE_IVOR9);
1080         env->spr[SPR_BOOKE_IVOR10] = sregs.u.e.ivor_low[10];
1081         kvm_sync_excp(env, POWERPC_EXCP_DECR,  SPR_BOOKE_IVOR10);
1082         env->spr[SPR_BOOKE_IVOR11] = sregs.u.e.ivor_low[11];
1083         kvm_sync_excp(env, POWERPC_EXCP_FIT,  SPR_BOOKE_IVOR11);
1084         env->spr[SPR_BOOKE_IVOR12] = sregs.u.e.ivor_low[12];
1085         kvm_sync_excp(env, POWERPC_EXCP_WDT,  SPR_BOOKE_IVOR12);
1086         env->spr[SPR_BOOKE_IVOR13] = sregs.u.e.ivor_low[13];
1087         kvm_sync_excp(env, POWERPC_EXCP_DTLB,  SPR_BOOKE_IVOR13);
1088         env->spr[SPR_BOOKE_IVOR14] = sregs.u.e.ivor_low[14];
1089         kvm_sync_excp(env, POWERPC_EXCP_ITLB,  SPR_BOOKE_IVOR14);
1090         env->spr[SPR_BOOKE_IVOR15] = sregs.u.e.ivor_low[15];
1091         kvm_sync_excp(env, POWERPC_EXCP_DEBUG,  SPR_BOOKE_IVOR15);
1092 
1093         if (sregs.u.e.features & KVM_SREGS_E_SPE) {
1094             env->spr[SPR_BOOKE_IVOR32] = sregs.u.e.ivor_high[0];
1095             kvm_sync_excp(env, POWERPC_EXCP_SPEU,  SPR_BOOKE_IVOR32);
1096             env->spr[SPR_BOOKE_IVOR33] = sregs.u.e.ivor_high[1];
1097             kvm_sync_excp(env, POWERPC_EXCP_EFPDI,  SPR_BOOKE_IVOR33);
1098             env->spr[SPR_BOOKE_IVOR34] = sregs.u.e.ivor_high[2];
1099             kvm_sync_excp(env, POWERPC_EXCP_EFPRI,  SPR_BOOKE_IVOR34);
1100         }
1101 
1102         if (sregs.u.e.features & KVM_SREGS_E_PM) {
1103             env->spr[SPR_BOOKE_IVOR35] = sregs.u.e.ivor_high[3];
1104             kvm_sync_excp(env, POWERPC_EXCP_EPERFM,  SPR_BOOKE_IVOR35);
1105         }
1106 
1107         if (sregs.u.e.features & KVM_SREGS_E_PC) {
1108             env->spr[SPR_BOOKE_IVOR36] = sregs.u.e.ivor_high[4];
1109             kvm_sync_excp(env, POWERPC_EXCP_DOORI,  SPR_BOOKE_IVOR36);
1110             env->spr[SPR_BOOKE_IVOR37] = sregs.u.e.ivor_high[5];
1111             kvm_sync_excp(env, POWERPC_EXCP_DOORCI, SPR_BOOKE_IVOR37);
1112         }
1113     }
1114 
1115     if (sregs.u.e.features & KVM_SREGS_E_ARCH206_MMU) {
1116         env->spr[SPR_BOOKE_MAS0] = sregs.u.e.mas0;
1117         env->spr[SPR_BOOKE_MAS1] = sregs.u.e.mas1;
1118         env->spr[SPR_BOOKE_MAS2] = sregs.u.e.mas2;
1119         env->spr[SPR_BOOKE_MAS3] = sregs.u.e.mas7_3 & 0xffffffff;
1120         env->spr[SPR_BOOKE_MAS4] = sregs.u.e.mas4;
1121         env->spr[SPR_BOOKE_MAS6] = sregs.u.e.mas6;
1122         env->spr[SPR_BOOKE_MAS7] = sregs.u.e.mas7_3 >> 32;
1123         env->spr[SPR_MMUCFG] = sregs.u.e.mmucfg;
1124         env->spr[SPR_BOOKE_TLB0CFG] = sregs.u.e.tlbcfg[0];
1125         env->spr[SPR_BOOKE_TLB1CFG] = sregs.u.e.tlbcfg[1];
1126     }
1127 
1128     if (sregs.u.e.features & KVM_SREGS_EXP) {
1129         env->spr[SPR_BOOKE_EPR] = sregs.u.e.epr;
1130     }
1131 
1132     if (sregs.u.e.features & KVM_SREGS_E_PD) {
1133         env->spr[SPR_BOOKE_EPLC] = sregs.u.e.eplc;
1134         env->spr[SPR_BOOKE_EPSC] = sregs.u.e.epsc;
1135     }
1136 
1137     if (sregs.u.e.impl_id == KVM_SREGS_E_IMPL_FSL) {
1138         env->spr[SPR_E500_SVR] = sregs.u.e.impl.fsl.svr;
1139         env->spr[SPR_Exxx_MCAR] = sregs.u.e.impl.fsl.mcar;
1140         env->spr[SPR_HID0] = sregs.u.e.impl.fsl.hid0;
1141 
1142         if (sregs.u.e.impl.fsl.features & KVM_SREGS_E_FSL_PIDn) {
1143             env->spr[SPR_BOOKE_PID1] = sregs.u.e.impl.fsl.pid1;
1144             env->spr[SPR_BOOKE_PID2] = sregs.u.e.impl.fsl.pid2;
1145         }
1146     }
1147 
1148     return 0;
1149 }
1150 
1151 static int kvmppc_get_books_sregs(PowerPCCPU *cpu)
1152 {
1153     CPUPPCState *env = &cpu->env;
1154     struct kvm_sregs sregs;
1155     int ret;
1156     int i;
1157 
1158     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_SREGS, &sregs);
1159     if (ret < 0) {
1160         return ret;
1161     }
1162 
1163     if (!cpu->vhyp) {
1164         ppc_store_sdr1(env, sregs.u.s.sdr1);
1165     }
1166 
1167     /* Sync SLB */
1168 #ifdef TARGET_PPC64
1169     /*
1170      * The packed SLB array we get from KVM_GET_SREGS only contains
1171      * information about valid entries. So we flush our internal copy
1172      * to get rid of stale ones, then put all valid SLB entries back
1173      * in.
1174      */
1175     memset(env->slb, 0, sizeof(env->slb));
1176     for (i = 0; i < ARRAY_SIZE(env->slb); i++) {
1177         target_ulong rb = sregs.u.s.ppc64.slb[i].slbe;
1178         target_ulong rs = sregs.u.s.ppc64.slb[i].slbv;
1179         /*
1180          * Only restore valid entries
1181          */
1182         if (rb & SLB_ESID_V) {
1183             ppc_store_slb(cpu, rb & 0xfff, rb & ~0xfffULL, rs);
1184         }
1185     }
1186 #endif
1187 
1188     /* Sync SRs */
1189     for (i = 0; i < 16; i++) {
1190         env->sr[i] = sregs.u.s.ppc32.sr[i];
1191     }
1192 
1193     /* Sync BATs */
1194     for (i = 0; i < 8; i++) {
1195         env->DBAT[0][i] = sregs.u.s.ppc32.dbat[i] & 0xffffffff;
1196         env->DBAT[1][i] = sregs.u.s.ppc32.dbat[i] >> 32;
1197         env->IBAT[0][i] = sregs.u.s.ppc32.ibat[i] & 0xffffffff;
1198         env->IBAT[1][i] = sregs.u.s.ppc32.ibat[i] >> 32;
1199     }
1200 
1201     return 0;
1202 }
1203 
1204 int kvm_arch_get_registers(CPUState *cs)
1205 {
1206     PowerPCCPU *cpu = POWERPC_CPU(cs);
1207     CPUPPCState *env = &cpu->env;
1208     struct kvm_regs regs;
1209     uint32_t cr;
1210     int i, ret;
1211 
1212     ret = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
1213     if (ret < 0) {
1214         return ret;
1215     }
1216 
1217     cr = regs.cr;
1218     for (i = 7; i >= 0; i--) {
1219         env->crf[i] = cr & 15;
1220         cr >>= 4;
1221     }
1222 
1223     env->ctr = regs.ctr;
1224     env->lr = regs.lr;
1225     cpu_write_xer(env, regs.xer);
1226     env->msr = regs.msr;
1227     env->nip = regs.pc;
1228 
1229     env->spr[SPR_SRR0] = regs.srr0;
1230     env->spr[SPR_SRR1] = regs.srr1;
1231 
1232     env->spr[SPR_SPRG0] = regs.sprg0;
1233     env->spr[SPR_SPRG1] = regs.sprg1;
1234     env->spr[SPR_SPRG2] = regs.sprg2;
1235     env->spr[SPR_SPRG3] = regs.sprg3;
1236     env->spr[SPR_SPRG4] = regs.sprg4;
1237     env->spr[SPR_SPRG5] = regs.sprg5;
1238     env->spr[SPR_SPRG6] = regs.sprg6;
1239     env->spr[SPR_SPRG7] = regs.sprg7;
1240 
1241     env->spr[SPR_BOOKE_PID] = regs.pid;
1242 
1243     for (i = 0; i < 32; i++) {
1244         env->gpr[i] = regs.gpr[i];
1245     }
1246 
1247     kvm_get_fp(cs);
1248 
1249     if (cap_booke_sregs) {
1250         ret = kvmppc_get_booke_sregs(cpu);
1251         if (ret < 0) {
1252             return ret;
1253         }
1254     }
1255 
1256     if (cap_segstate) {
1257         ret = kvmppc_get_books_sregs(cpu);
1258         if (ret < 0) {
1259             return ret;
1260         }
1261     }
1262 
1263     if (cap_hior) {
1264         kvm_get_one_spr(cs, KVM_REG_PPC_HIOR, SPR_HIOR);
1265     }
1266 
1267     if (cap_one_reg) {
1268         int i;
1269 
1270         /*
1271          * We deliberately ignore errors here, for kernels which have
1272          * the ONE_REG calls, but don't support the specific
1273          * registers, there's a reasonable chance things will still
1274          * work, at least until we try to migrate.
1275          */
1276         for (i = 0; i < 1024; i++) {
1277             uint64_t id = env->spr_cb[i].one_reg_id;
1278 
1279             if (id != 0) {
1280                 kvm_get_one_spr(cs, id, i);
1281             }
1282         }
1283 
1284 #ifdef TARGET_PPC64
1285         if (msr_ts) {
1286             for (i = 0; i < ARRAY_SIZE(env->tm_gpr); i++) {
1287                 kvm_get_one_reg(cs, KVM_REG_PPC_TM_GPR(i), &env->tm_gpr[i]);
1288             }
1289             for (i = 0; i < ARRAY_SIZE(env->tm_vsr); i++) {
1290                 kvm_get_one_reg(cs, KVM_REG_PPC_TM_VSR(i), &env->tm_vsr[i]);
1291             }
1292             kvm_get_one_reg(cs, KVM_REG_PPC_TM_CR, &env->tm_cr);
1293             kvm_get_one_reg(cs, KVM_REG_PPC_TM_LR, &env->tm_lr);
1294             kvm_get_one_reg(cs, KVM_REG_PPC_TM_CTR, &env->tm_ctr);
1295             kvm_get_one_reg(cs, KVM_REG_PPC_TM_FPSCR, &env->tm_fpscr);
1296             kvm_get_one_reg(cs, KVM_REG_PPC_TM_AMR, &env->tm_amr);
1297             kvm_get_one_reg(cs, KVM_REG_PPC_TM_PPR, &env->tm_ppr);
1298             kvm_get_one_reg(cs, KVM_REG_PPC_TM_VRSAVE, &env->tm_vrsave);
1299             kvm_get_one_reg(cs, KVM_REG_PPC_TM_VSCR, &env->tm_vscr);
1300             kvm_get_one_reg(cs, KVM_REG_PPC_TM_DSCR, &env->tm_dscr);
1301             kvm_get_one_reg(cs, KVM_REG_PPC_TM_TAR, &env->tm_tar);
1302         }
1303 
1304         if (cap_papr) {
1305             if (kvm_get_vpa(cs) < 0) {
1306                 trace_kvm_failed_get_vpa();
1307             }
1308         }
1309 
1310         kvm_get_one_reg(cs, KVM_REG_PPC_TB_OFFSET, &env->tb_env->tb_offset);
1311         kvm_get_one_spr(cs, KVM_REG_PPC_DPDES, SPR_DPDES);
1312 #endif
1313     }
1314 
1315     return 0;
1316 }
1317 
1318 int kvmppc_set_interrupt(PowerPCCPU *cpu, int irq, int level)
1319 {
1320     unsigned virq = level ? KVM_INTERRUPT_SET_LEVEL : KVM_INTERRUPT_UNSET;
1321 
1322     if (irq != PPC_INTERRUPT_EXT) {
1323         return 0;
1324     }
1325 
1326     if (!kvm_enabled() || !cap_interrupt_unset) {
1327         return 0;
1328     }
1329 
1330     kvm_vcpu_ioctl(CPU(cpu), KVM_INTERRUPT, &virq);
1331 
1332     return 0;
1333 }
1334 
1335 void kvm_arch_pre_run(CPUState *cs, struct kvm_run *run)
1336 {
1337     return;
1338 }
1339 
1340 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
1341 {
1342     return MEMTXATTRS_UNSPECIFIED;
1343 }
1344 
1345 int kvm_arch_process_async_events(CPUState *cs)
1346 {
1347     return cs->halted;
1348 }
1349 
1350 static int kvmppc_handle_halt(PowerPCCPU *cpu)
1351 {
1352     CPUState *cs = CPU(cpu);
1353     CPUPPCState *env = &cpu->env;
1354 
1355     if (!(cs->interrupt_request & CPU_INTERRUPT_HARD) && (msr_ee)) {
1356         cs->halted = 1;
1357         cs->exception_index = EXCP_HLT;
1358     }
1359 
1360     return 0;
1361 }
1362 
1363 /* map dcr access to existing qemu dcr emulation */
1364 static int kvmppc_handle_dcr_read(CPUPPCState *env,
1365                                   uint32_t dcrn, uint32_t *data)
1366 {
1367     if (ppc_dcr_read(env->dcr_env, dcrn, data) < 0) {
1368         fprintf(stderr, "Read to unhandled DCR (0x%x)\n", dcrn);
1369     }
1370 
1371     return 0;
1372 }
1373 
1374 static int kvmppc_handle_dcr_write(CPUPPCState *env,
1375                                    uint32_t dcrn, uint32_t data)
1376 {
1377     if (ppc_dcr_write(env->dcr_env, dcrn, data) < 0) {
1378         fprintf(stderr, "Write to unhandled DCR (0x%x)\n", dcrn);
1379     }
1380 
1381     return 0;
1382 }
1383 
1384 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
1385 {
1386     /* Mixed endian case is not handled */
1387     uint32_t sc = debug_inst_opcode;
1388 
1389     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
1390                             sizeof(sc), 0) ||
1391         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&sc, sizeof(sc), 1)) {
1392         return -EINVAL;
1393     }
1394 
1395     return 0;
1396 }
1397 
1398 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
1399 {
1400     uint32_t sc;
1401 
1402     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&sc, sizeof(sc), 0) ||
1403         sc != debug_inst_opcode ||
1404         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
1405                             sizeof(sc), 1)) {
1406         return -EINVAL;
1407     }
1408 
1409     return 0;
1410 }
1411 
1412 static int find_hw_breakpoint(target_ulong addr, int type)
1413 {
1414     int n;
1415 
1416     assert((nb_hw_breakpoint + nb_hw_watchpoint)
1417            <= ARRAY_SIZE(hw_debug_points));
1418 
1419     for (n = 0; n < nb_hw_breakpoint + nb_hw_watchpoint; n++) {
1420         if (hw_debug_points[n].addr == addr &&
1421              hw_debug_points[n].type == type) {
1422             return n;
1423         }
1424     }
1425 
1426     return -1;
1427 }
1428 
1429 static int find_hw_watchpoint(target_ulong addr, int *flag)
1430 {
1431     int n;
1432 
1433     n = find_hw_breakpoint(addr, GDB_WATCHPOINT_ACCESS);
1434     if (n >= 0) {
1435         *flag = BP_MEM_ACCESS;
1436         return n;
1437     }
1438 
1439     n = find_hw_breakpoint(addr, GDB_WATCHPOINT_WRITE);
1440     if (n >= 0) {
1441         *flag = BP_MEM_WRITE;
1442         return n;
1443     }
1444 
1445     n = find_hw_breakpoint(addr, GDB_WATCHPOINT_READ);
1446     if (n >= 0) {
1447         *flag = BP_MEM_READ;
1448         return n;
1449     }
1450 
1451     return -1;
1452 }
1453 
1454 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
1455                                   target_ulong len, int type)
1456 {
1457     if ((nb_hw_breakpoint + nb_hw_watchpoint) >= ARRAY_SIZE(hw_debug_points)) {
1458         return -ENOBUFS;
1459     }
1460 
1461     hw_debug_points[nb_hw_breakpoint + nb_hw_watchpoint].addr = addr;
1462     hw_debug_points[nb_hw_breakpoint + nb_hw_watchpoint].type = type;
1463 
1464     switch (type) {
1465     case GDB_BREAKPOINT_HW:
1466         if (nb_hw_breakpoint >= max_hw_breakpoint) {
1467             return -ENOBUFS;
1468         }
1469 
1470         if (find_hw_breakpoint(addr, type) >= 0) {
1471             return -EEXIST;
1472         }
1473 
1474         nb_hw_breakpoint++;
1475         break;
1476 
1477     case GDB_WATCHPOINT_WRITE:
1478     case GDB_WATCHPOINT_READ:
1479     case GDB_WATCHPOINT_ACCESS:
1480         if (nb_hw_watchpoint >= max_hw_watchpoint) {
1481             return -ENOBUFS;
1482         }
1483 
1484         if (find_hw_breakpoint(addr, type) >= 0) {
1485             return -EEXIST;
1486         }
1487 
1488         nb_hw_watchpoint++;
1489         break;
1490 
1491     default:
1492         return -ENOSYS;
1493     }
1494 
1495     return 0;
1496 }
1497 
1498 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
1499                                   target_ulong len, int type)
1500 {
1501     int n;
1502 
1503     n = find_hw_breakpoint(addr, type);
1504     if (n < 0) {
1505         return -ENOENT;
1506     }
1507 
1508     switch (type) {
1509     case GDB_BREAKPOINT_HW:
1510         nb_hw_breakpoint--;
1511         break;
1512 
1513     case GDB_WATCHPOINT_WRITE:
1514     case GDB_WATCHPOINT_READ:
1515     case GDB_WATCHPOINT_ACCESS:
1516         nb_hw_watchpoint--;
1517         break;
1518 
1519     default:
1520         return -ENOSYS;
1521     }
1522     hw_debug_points[n] = hw_debug_points[nb_hw_breakpoint + nb_hw_watchpoint];
1523 
1524     return 0;
1525 }
1526 
1527 void kvm_arch_remove_all_hw_breakpoints(void)
1528 {
1529     nb_hw_breakpoint = nb_hw_watchpoint = 0;
1530 }
1531 
1532 void kvm_arch_update_guest_debug(CPUState *cs, struct kvm_guest_debug *dbg)
1533 {
1534     int n;
1535 
1536     /* Software Breakpoint updates */
1537     if (kvm_sw_breakpoints_active(cs)) {
1538         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP;
1539     }
1540 
1541     assert((nb_hw_breakpoint + nb_hw_watchpoint)
1542            <= ARRAY_SIZE(hw_debug_points));
1543     assert((nb_hw_breakpoint + nb_hw_watchpoint) <= ARRAY_SIZE(dbg->arch.bp));
1544 
1545     if (nb_hw_breakpoint + nb_hw_watchpoint > 0) {
1546         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
1547         memset(dbg->arch.bp, 0, sizeof(dbg->arch.bp));
1548         for (n = 0; n < nb_hw_breakpoint + nb_hw_watchpoint; n++) {
1549             switch (hw_debug_points[n].type) {
1550             case GDB_BREAKPOINT_HW:
1551                 dbg->arch.bp[n].type = KVMPPC_DEBUG_BREAKPOINT;
1552                 break;
1553             case GDB_WATCHPOINT_WRITE:
1554                 dbg->arch.bp[n].type = KVMPPC_DEBUG_WATCH_WRITE;
1555                 break;
1556             case GDB_WATCHPOINT_READ:
1557                 dbg->arch.bp[n].type = KVMPPC_DEBUG_WATCH_READ;
1558                 break;
1559             case GDB_WATCHPOINT_ACCESS:
1560                 dbg->arch.bp[n].type = KVMPPC_DEBUG_WATCH_WRITE |
1561                                         KVMPPC_DEBUG_WATCH_READ;
1562                 break;
1563             default:
1564                 cpu_abort(cs, "Unsupported breakpoint type\n");
1565             }
1566             dbg->arch.bp[n].addr = hw_debug_points[n].addr;
1567         }
1568     }
1569 }
1570 
1571 static int kvm_handle_hw_breakpoint(CPUState *cs,
1572                                     struct kvm_debug_exit_arch *arch_info)
1573 {
1574     int handle = DEBUG_RETURN_GUEST;
1575     int n;
1576     int flag = 0;
1577 
1578     if (nb_hw_breakpoint + nb_hw_watchpoint > 0) {
1579         if (arch_info->status & KVMPPC_DEBUG_BREAKPOINT) {
1580             n = find_hw_breakpoint(arch_info->address, GDB_BREAKPOINT_HW);
1581             if (n >= 0) {
1582                 handle = DEBUG_RETURN_GDB;
1583             }
1584         } else if (arch_info->status & (KVMPPC_DEBUG_WATCH_READ |
1585                                         KVMPPC_DEBUG_WATCH_WRITE)) {
1586             n = find_hw_watchpoint(arch_info->address,  &flag);
1587             if (n >= 0) {
1588                 handle = DEBUG_RETURN_GDB;
1589                 cs->watchpoint_hit = &hw_watchpoint;
1590                 hw_watchpoint.vaddr = hw_debug_points[n].addr;
1591                 hw_watchpoint.flags = flag;
1592             }
1593         }
1594     }
1595     return handle;
1596 }
1597 
1598 static int kvm_handle_singlestep(void)
1599 {
1600     return DEBUG_RETURN_GDB;
1601 }
1602 
1603 static int kvm_handle_sw_breakpoint(void)
1604 {
1605     return DEBUG_RETURN_GDB;
1606 }
1607 
1608 static int kvm_handle_debug(PowerPCCPU *cpu, struct kvm_run *run)
1609 {
1610     CPUState *cs = CPU(cpu);
1611     CPUPPCState *env = &cpu->env;
1612     struct kvm_debug_exit_arch *arch_info = &run->debug.arch;
1613 
1614     if (cs->singlestep_enabled) {
1615         return kvm_handle_singlestep();
1616     }
1617 
1618     if (arch_info->status) {
1619         return kvm_handle_hw_breakpoint(cs, arch_info);
1620     }
1621 
1622     if (kvm_find_sw_breakpoint(cs, arch_info->address)) {
1623         return kvm_handle_sw_breakpoint();
1624     }
1625 
1626     /*
1627      * QEMU is not able to handle debug exception, so inject
1628      * program exception to guest;
1629      * Yes program exception NOT debug exception !!
1630      * When QEMU is using debug resources then debug exception must
1631      * be always set. To achieve this we set MSR_DE and also set
1632      * MSRP_DEP so guest cannot change MSR_DE.
1633      * When emulating debug resource for guest we want guest
1634      * to control MSR_DE (enable/disable debug interrupt on need).
1635      * Supporting both configurations are NOT possible.
1636      * So the result is that we cannot share debug resources
1637      * between QEMU and Guest on BOOKE architecture.
1638      * In the current design QEMU gets the priority over guest,
1639      * this means that if QEMU is using debug resources then guest
1640      * cannot use them;
1641      * For software breakpoint QEMU uses a privileged instruction;
1642      * So there cannot be any reason that we are here for guest
1643      * set debug exception, only possibility is guest executed a
1644      * privileged / illegal instruction and that's why we are
1645      * injecting a program interrupt.
1646      */
1647     cpu_synchronize_state(cs);
1648     /*
1649      * env->nip is PC, so increment this by 4 to use
1650      * ppc_cpu_do_interrupt(), which set srr0 = env->nip - 4.
1651      */
1652     env->nip += 4;
1653     cs->exception_index = POWERPC_EXCP_PROGRAM;
1654     env->error_code = POWERPC_EXCP_INVAL;
1655     ppc_cpu_do_interrupt(cs);
1656 
1657     return DEBUG_RETURN_GUEST;
1658 }
1659 
1660 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1661 {
1662     PowerPCCPU *cpu = POWERPC_CPU(cs);
1663     CPUPPCState *env = &cpu->env;
1664     int ret;
1665 
1666     qemu_mutex_lock_iothread();
1667 
1668     switch (run->exit_reason) {
1669     case KVM_EXIT_DCR:
1670         if (run->dcr.is_write) {
1671             trace_kvm_handle_dcr_write();
1672             ret = kvmppc_handle_dcr_write(env, run->dcr.dcrn, run->dcr.data);
1673         } else {
1674             trace_kvm_handle_dcr_read();
1675             ret = kvmppc_handle_dcr_read(env, run->dcr.dcrn, &run->dcr.data);
1676         }
1677         break;
1678     case KVM_EXIT_HLT:
1679         trace_kvm_handle_halt();
1680         ret = kvmppc_handle_halt(cpu);
1681         break;
1682 #if defined(TARGET_PPC64)
1683     case KVM_EXIT_PAPR_HCALL:
1684         trace_kvm_handle_papr_hcall();
1685         run->papr_hcall.ret = spapr_hypercall(cpu,
1686                                               run->papr_hcall.nr,
1687                                               run->papr_hcall.args);
1688         ret = 0;
1689         break;
1690 #endif
1691     case KVM_EXIT_EPR:
1692         trace_kvm_handle_epr();
1693         run->epr.epr = ldl_phys(cs->as, env->mpic_iack);
1694         ret = 0;
1695         break;
1696     case KVM_EXIT_WATCHDOG:
1697         trace_kvm_handle_watchdog_expiry();
1698         watchdog_perform_action();
1699         ret = 0;
1700         break;
1701 
1702     case KVM_EXIT_DEBUG:
1703         trace_kvm_handle_debug_exception();
1704         if (kvm_handle_debug(cpu, run)) {
1705             ret = EXCP_DEBUG;
1706             break;
1707         }
1708         /* re-enter, this exception was guest-internal */
1709         ret = 0;
1710         break;
1711 
1712 #if defined(TARGET_PPC64)
1713     case KVM_EXIT_NMI:
1714         trace_kvm_handle_nmi_exception();
1715         ret = kvm_handle_nmi(cpu, run);
1716         break;
1717 #endif
1718 
1719     default:
1720         fprintf(stderr, "KVM: unknown exit reason %d\n", run->exit_reason);
1721         ret = -1;
1722         break;
1723     }
1724 
1725     qemu_mutex_unlock_iothread();
1726     return ret;
1727 }
1728 
1729 int kvmppc_or_tsr_bits(PowerPCCPU *cpu, uint32_t tsr_bits)
1730 {
1731     CPUState *cs = CPU(cpu);
1732     uint32_t bits = tsr_bits;
1733     struct kvm_one_reg reg = {
1734         .id = KVM_REG_PPC_OR_TSR,
1735         .addr = (uintptr_t) &bits,
1736     };
1737 
1738     return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
1739 }
1740 
1741 int kvmppc_clear_tsr_bits(PowerPCCPU *cpu, uint32_t tsr_bits)
1742 {
1743 
1744     CPUState *cs = CPU(cpu);
1745     uint32_t bits = tsr_bits;
1746     struct kvm_one_reg reg = {
1747         .id = KVM_REG_PPC_CLEAR_TSR,
1748         .addr = (uintptr_t) &bits,
1749     };
1750 
1751     return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
1752 }
1753 
1754 int kvmppc_set_tcr(PowerPCCPU *cpu)
1755 {
1756     CPUState *cs = CPU(cpu);
1757     CPUPPCState *env = &cpu->env;
1758     uint32_t tcr = env->spr[SPR_BOOKE_TCR];
1759 
1760     struct kvm_one_reg reg = {
1761         .id = KVM_REG_PPC_TCR,
1762         .addr = (uintptr_t) &tcr,
1763     };
1764 
1765     return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
1766 }
1767 
1768 int kvmppc_booke_watchdog_enable(PowerPCCPU *cpu)
1769 {
1770     CPUState *cs = CPU(cpu);
1771     int ret;
1772 
1773     if (!kvm_enabled()) {
1774         return -1;
1775     }
1776 
1777     if (!cap_ppc_watchdog) {
1778         printf("warning: KVM does not support watchdog");
1779         return -1;
1780     }
1781 
1782     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_PPC_BOOKE_WATCHDOG, 0);
1783     if (ret < 0) {
1784         fprintf(stderr, "%s: couldn't enable KVM_CAP_PPC_BOOKE_WATCHDOG: %s\n",
1785                 __func__, strerror(-ret));
1786         return ret;
1787     }
1788 
1789     return ret;
1790 }
1791 
1792 static int read_cpuinfo(const char *field, char *value, int len)
1793 {
1794     FILE *f;
1795     int ret = -1;
1796     int field_len = strlen(field);
1797     char line[512];
1798 
1799     f = fopen("/proc/cpuinfo", "r");
1800     if (!f) {
1801         return -1;
1802     }
1803 
1804     do {
1805         if (!fgets(line, sizeof(line), f)) {
1806             break;
1807         }
1808         if (!strncmp(line, field, field_len)) {
1809             pstrcpy(value, len, line);
1810             ret = 0;
1811             break;
1812         }
1813     } while (*line);
1814 
1815     fclose(f);
1816 
1817     return ret;
1818 }
1819 
1820 static uint32_t kvmppc_get_tbfreq_procfs(void)
1821 {
1822     char line[512];
1823     char *ns;
1824     uint32_t tbfreq_fallback = NANOSECONDS_PER_SECOND;
1825     uint32_t tbfreq_procfs;
1826 
1827     if (read_cpuinfo("timebase", line, sizeof(line))) {
1828         return tbfreq_fallback;
1829     }
1830 
1831     ns = strchr(line, ':');
1832     if (!ns) {
1833         return tbfreq_fallback;
1834     }
1835 
1836     tbfreq_procfs = atoi(++ns);
1837 
1838     /* 0 is certainly not acceptable by the guest, return fallback value */
1839     return tbfreq_procfs ? tbfreq_procfs : tbfreq_fallback;
1840 }
1841 
1842 uint32_t kvmppc_get_tbfreq(void)
1843 {
1844     static uint32_t cached_tbfreq;
1845 
1846     if (!cached_tbfreq) {
1847         cached_tbfreq = kvmppc_get_tbfreq_procfs();
1848     }
1849 
1850     return cached_tbfreq;
1851 }
1852 
1853 bool kvmppc_get_host_serial(char **value)
1854 {
1855     return g_file_get_contents("/proc/device-tree/system-id", value, NULL,
1856                                NULL);
1857 }
1858 
1859 bool kvmppc_get_host_model(char **value)
1860 {
1861     return g_file_get_contents("/proc/device-tree/model", value, NULL, NULL);
1862 }
1863 
1864 /* Try to find a device tree node for a CPU with clock-frequency property */
1865 static int kvmppc_find_cpu_dt(char *buf, int buf_len)
1866 {
1867     struct dirent *dirp;
1868     DIR *dp;
1869 
1870     dp = opendir(PROC_DEVTREE_CPU);
1871     if (!dp) {
1872         printf("Can't open directory " PROC_DEVTREE_CPU "\n");
1873         return -1;
1874     }
1875 
1876     buf[0] = '\0';
1877     while ((dirp = readdir(dp)) != NULL) {
1878         FILE *f;
1879         snprintf(buf, buf_len, "%s%s/clock-frequency", PROC_DEVTREE_CPU,
1880                  dirp->d_name);
1881         f = fopen(buf, "r");
1882         if (f) {
1883             snprintf(buf, buf_len, "%s%s", PROC_DEVTREE_CPU, dirp->d_name);
1884             fclose(f);
1885             break;
1886         }
1887         buf[0] = '\0';
1888     }
1889     closedir(dp);
1890     if (buf[0] == '\0') {
1891         printf("Unknown host!\n");
1892         return -1;
1893     }
1894 
1895     return 0;
1896 }
1897 
1898 static uint64_t kvmppc_read_int_dt(const char *filename)
1899 {
1900     union {
1901         uint32_t v32;
1902         uint64_t v64;
1903     } u;
1904     FILE *f;
1905     int len;
1906 
1907     f = fopen(filename, "rb");
1908     if (!f) {
1909         return -1;
1910     }
1911 
1912     len = fread(&u, 1, sizeof(u), f);
1913     fclose(f);
1914     switch (len) {
1915     case 4:
1916         /* property is a 32-bit quantity */
1917         return be32_to_cpu(u.v32);
1918     case 8:
1919         return be64_to_cpu(u.v64);
1920     }
1921 
1922     return 0;
1923 }
1924 
1925 /*
1926  * Read a CPU node property from the host device tree that's a single
1927  * integer (32-bit or 64-bit).  Returns 0 if anything goes wrong
1928  * (can't find or open the property, or doesn't understand the format)
1929  */
1930 static uint64_t kvmppc_read_int_cpu_dt(const char *propname)
1931 {
1932     char buf[PATH_MAX], *tmp;
1933     uint64_t val;
1934 
1935     if (kvmppc_find_cpu_dt(buf, sizeof(buf))) {
1936         return -1;
1937     }
1938 
1939     tmp = g_strdup_printf("%s/%s", buf, propname);
1940     val = kvmppc_read_int_dt(tmp);
1941     g_free(tmp);
1942 
1943     return val;
1944 }
1945 
1946 uint64_t kvmppc_get_clockfreq(void)
1947 {
1948     return kvmppc_read_int_cpu_dt("clock-frequency");
1949 }
1950 
1951 static int kvmppc_get_dec_bits(void)
1952 {
1953     int nr_bits = kvmppc_read_int_cpu_dt("ibm,dec-bits");
1954 
1955     if (nr_bits > 0) {
1956         return nr_bits;
1957     }
1958     return 0;
1959 }
1960 
1961 static int kvmppc_get_pvinfo(CPUPPCState *env, struct kvm_ppc_pvinfo *pvinfo)
1962 {
1963     CPUState *cs = env_cpu(env);
1964 
1965     if (kvm_vm_check_extension(cs->kvm_state, KVM_CAP_PPC_GET_PVINFO) &&
1966         !kvm_vm_ioctl(cs->kvm_state, KVM_PPC_GET_PVINFO, pvinfo)) {
1967         return 0;
1968     }
1969 
1970     return 1;
1971 }
1972 
1973 int kvmppc_get_hasidle(CPUPPCState *env)
1974 {
1975     struct kvm_ppc_pvinfo pvinfo;
1976 
1977     if (!kvmppc_get_pvinfo(env, &pvinfo) &&
1978         (pvinfo.flags & KVM_PPC_PVINFO_FLAGS_EV_IDLE)) {
1979         return 1;
1980     }
1981 
1982     return 0;
1983 }
1984 
1985 int kvmppc_get_hypercall(CPUPPCState *env, uint8_t *buf, int buf_len)
1986 {
1987     uint32_t *hc = (uint32_t *)buf;
1988     struct kvm_ppc_pvinfo pvinfo;
1989 
1990     if (!kvmppc_get_pvinfo(env, &pvinfo)) {
1991         memcpy(buf, pvinfo.hcall, buf_len);
1992         return 0;
1993     }
1994 
1995     /*
1996      * Fallback to always fail hypercalls regardless of endianness:
1997      *
1998      *     tdi 0,r0,72 (becomes b .+8 in wrong endian, nop in good endian)
1999      *     li r3, -1
2000      *     b .+8       (becomes nop in wrong endian)
2001      *     bswap32(li r3, -1)
2002      */
2003 
2004     hc[0] = cpu_to_be32(0x08000048);
2005     hc[1] = cpu_to_be32(0x3860ffff);
2006     hc[2] = cpu_to_be32(0x48000008);
2007     hc[3] = cpu_to_be32(bswap32(0x3860ffff));
2008 
2009     return 1;
2010 }
2011 
2012 static inline int kvmppc_enable_hcall(KVMState *s, target_ulong hcall)
2013 {
2014     return kvm_vm_enable_cap(s, KVM_CAP_PPC_ENABLE_HCALL, 0, hcall, 1);
2015 }
2016 
2017 void kvmppc_enable_logical_ci_hcalls(void)
2018 {
2019     /*
2020      * FIXME: it would be nice if we could detect the cases where
2021      * we're using a device which requires the in kernel
2022      * implementation of these hcalls, but the kernel lacks them and
2023      * produce a warning.
2024      */
2025     kvmppc_enable_hcall(kvm_state, H_LOGICAL_CI_LOAD);
2026     kvmppc_enable_hcall(kvm_state, H_LOGICAL_CI_STORE);
2027 }
2028 
2029 void kvmppc_enable_set_mode_hcall(void)
2030 {
2031     kvmppc_enable_hcall(kvm_state, H_SET_MODE);
2032 }
2033 
2034 void kvmppc_enable_clear_ref_mod_hcalls(void)
2035 {
2036     kvmppc_enable_hcall(kvm_state, H_CLEAR_REF);
2037     kvmppc_enable_hcall(kvm_state, H_CLEAR_MOD);
2038 }
2039 
2040 void kvmppc_enable_h_page_init(void)
2041 {
2042     kvmppc_enable_hcall(kvm_state, H_PAGE_INIT);
2043 }
2044 
2045 void kvmppc_enable_h_rpt_invalidate(void)
2046 {
2047     kvmppc_enable_hcall(kvm_state, H_RPT_INVALIDATE);
2048 }
2049 
2050 void kvmppc_set_papr(PowerPCCPU *cpu)
2051 {
2052     CPUState *cs = CPU(cpu);
2053     int ret;
2054 
2055     if (!kvm_enabled()) {
2056         return;
2057     }
2058 
2059     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_PPC_PAPR, 0);
2060     if (ret) {
2061         error_report("This vCPU type or KVM version does not support PAPR");
2062         exit(1);
2063     }
2064 
2065     /*
2066      * Update the capability flag so we sync the right information
2067      * with kvm
2068      */
2069     cap_papr = 1;
2070 }
2071 
2072 int kvmppc_set_compat(PowerPCCPU *cpu, uint32_t compat_pvr)
2073 {
2074     return kvm_set_one_reg(CPU(cpu), KVM_REG_PPC_ARCH_COMPAT, &compat_pvr);
2075 }
2076 
2077 void kvmppc_set_mpic_proxy(PowerPCCPU *cpu, int mpic_proxy)
2078 {
2079     CPUState *cs = CPU(cpu);
2080     int ret;
2081 
2082     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_PPC_EPR, 0, mpic_proxy);
2083     if (ret && mpic_proxy) {
2084         error_report("This KVM version does not support EPR");
2085         exit(1);
2086     }
2087 }
2088 
2089 bool kvmppc_get_fwnmi(void)
2090 {
2091     return cap_fwnmi;
2092 }
2093 
2094 int kvmppc_set_fwnmi(PowerPCCPU *cpu)
2095 {
2096     CPUState *cs = CPU(cpu);
2097 
2098     return kvm_vcpu_enable_cap(cs, KVM_CAP_PPC_FWNMI, 0);
2099 }
2100 
2101 int kvmppc_smt_threads(void)
2102 {
2103     return cap_ppc_smt ? cap_ppc_smt : 1;
2104 }
2105 
2106 int kvmppc_set_smt_threads(int smt)
2107 {
2108     int ret;
2109 
2110     ret = kvm_vm_enable_cap(kvm_state, KVM_CAP_PPC_SMT, 0, smt, 0);
2111     if (!ret) {
2112         cap_ppc_smt = smt;
2113     }
2114     return ret;
2115 }
2116 
2117 void kvmppc_error_append_smt_possible_hint(Error *const *errp)
2118 {
2119     int i;
2120     GString *g;
2121     char *s;
2122 
2123     assert(kvm_enabled());
2124     if (cap_ppc_smt_possible) {
2125         g = g_string_new("Available VSMT modes:");
2126         for (i = 63; i >= 0; i--) {
2127             if ((1UL << i) & cap_ppc_smt_possible) {
2128                 g_string_append_printf(g, " %lu", (1UL << i));
2129             }
2130         }
2131         s = g_string_free(g, false);
2132         error_append_hint(errp, "%s.\n", s);
2133         g_free(s);
2134     } else {
2135         error_append_hint(errp,
2136                           "This KVM seems to be too old to support VSMT.\n");
2137     }
2138 }
2139 
2140 
2141 #ifdef TARGET_PPC64
2142 uint64_t kvmppc_vrma_limit(unsigned int hash_shift)
2143 {
2144     struct kvm_ppc_smmu_info info;
2145     long rampagesize, best_page_shift;
2146     int i;
2147 
2148     /*
2149      * Find the largest hardware supported page size that's less than
2150      * or equal to the (logical) backing page size of guest RAM
2151      */
2152     kvm_get_smmu_info(&info, &error_fatal);
2153     rampagesize = qemu_minrampagesize();
2154     best_page_shift = 0;
2155 
2156     for (i = 0; i < KVM_PPC_PAGE_SIZES_MAX_SZ; i++) {
2157         struct kvm_ppc_one_seg_page_size *sps = &info.sps[i];
2158 
2159         if (!sps->page_shift) {
2160             continue;
2161         }
2162 
2163         if ((sps->page_shift > best_page_shift)
2164             && ((1UL << sps->page_shift) <= rampagesize)) {
2165             best_page_shift = sps->page_shift;
2166         }
2167     }
2168 
2169     return 1ULL << (best_page_shift + hash_shift - 7);
2170 }
2171 #endif
2172 
2173 bool kvmppc_spapr_use_multitce(void)
2174 {
2175     return cap_spapr_multitce;
2176 }
2177 
2178 int kvmppc_spapr_enable_inkernel_multitce(void)
2179 {
2180     int ret;
2181 
2182     ret = kvm_vm_enable_cap(kvm_state, KVM_CAP_PPC_ENABLE_HCALL, 0,
2183                             H_PUT_TCE_INDIRECT, 1);
2184     if (!ret) {
2185         ret = kvm_vm_enable_cap(kvm_state, KVM_CAP_PPC_ENABLE_HCALL, 0,
2186                                 H_STUFF_TCE, 1);
2187     }
2188 
2189     return ret;
2190 }
2191 
2192 void *kvmppc_create_spapr_tce(uint32_t liobn, uint32_t page_shift,
2193                               uint64_t bus_offset, uint32_t nb_table,
2194                               int *pfd, bool need_vfio)
2195 {
2196     long len;
2197     int fd;
2198     void *table;
2199 
2200     /*
2201      * Must set fd to -1 so we don't try to munmap when called for
2202      * destroying the table, which the upper layers -will- do
2203      */
2204     *pfd = -1;
2205     if (!cap_spapr_tce || (need_vfio && !cap_spapr_vfio)) {
2206         return NULL;
2207     }
2208 
2209     if (cap_spapr_tce_64) {
2210         struct kvm_create_spapr_tce_64 args = {
2211             .liobn = liobn,
2212             .page_shift = page_shift,
2213             .offset = bus_offset >> page_shift,
2214             .size = nb_table,
2215             .flags = 0
2216         };
2217         fd = kvm_vm_ioctl(kvm_state, KVM_CREATE_SPAPR_TCE_64, &args);
2218         if (fd < 0) {
2219             fprintf(stderr,
2220                     "KVM: Failed to create TCE64 table for liobn 0x%x\n",
2221                     liobn);
2222             return NULL;
2223         }
2224     } else if (cap_spapr_tce) {
2225         uint64_t window_size = (uint64_t) nb_table << page_shift;
2226         struct kvm_create_spapr_tce args = {
2227             .liobn = liobn,
2228             .window_size = window_size,
2229         };
2230         if ((window_size != args.window_size) || bus_offset) {
2231             return NULL;
2232         }
2233         fd = kvm_vm_ioctl(kvm_state, KVM_CREATE_SPAPR_TCE, &args);
2234         if (fd < 0) {
2235             fprintf(stderr, "KVM: Failed to create TCE table for liobn 0x%x\n",
2236                     liobn);
2237             return NULL;
2238         }
2239     } else {
2240         return NULL;
2241     }
2242 
2243     len = nb_table * sizeof(uint64_t);
2244     /* FIXME: round this up to page size */
2245 
2246     table = mmap(NULL, len, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
2247     if (table == MAP_FAILED) {
2248         fprintf(stderr, "KVM: Failed to map TCE table for liobn 0x%x\n",
2249                 liobn);
2250         close(fd);
2251         return NULL;
2252     }
2253 
2254     *pfd = fd;
2255     return table;
2256 }
2257 
2258 int kvmppc_remove_spapr_tce(void *table, int fd, uint32_t nb_table)
2259 {
2260     long len;
2261 
2262     if (fd < 0) {
2263         return -1;
2264     }
2265 
2266     len = nb_table * sizeof(uint64_t);
2267     if ((munmap(table, len) < 0) ||
2268         (close(fd) < 0)) {
2269         fprintf(stderr, "KVM: Unexpected error removing TCE table: %s",
2270                 strerror(errno));
2271         /* Leak the table */
2272     }
2273 
2274     return 0;
2275 }
2276 
2277 int kvmppc_reset_htab(int shift_hint)
2278 {
2279     uint32_t shift = shift_hint;
2280 
2281     if (!kvm_enabled()) {
2282         /* Full emulation, tell caller to allocate htab itself */
2283         return 0;
2284     }
2285     if (kvm_vm_check_extension(kvm_state, KVM_CAP_PPC_ALLOC_HTAB)) {
2286         int ret;
2287         ret = kvm_vm_ioctl(kvm_state, KVM_PPC_ALLOCATE_HTAB, &shift);
2288         if (ret == -ENOTTY) {
2289             /*
2290              * At least some versions of PR KVM advertise the
2291              * capability, but don't implement the ioctl().  Oops.
2292              * Return 0 so that we allocate the htab in qemu, as is
2293              * correct for PR.
2294              */
2295             return 0;
2296         } else if (ret < 0) {
2297             return ret;
2298         }
2299         return shift;
2300     }
2301 
2302     /*
2303      * We have a kernel that predates the htab reset calls.  For PR
2304      * KVM, we need to allocate the htab ourselves, for an HV KVM of
2305      * this era, it has allocated a 16MB fixed size hash table
2306      * already.
2307      */
2308     if (kvmppc_is_pr(kvm_state)) {
2309         /* PR - tell caller to allocate htab */
2310         return 0;
2311     } else {
2312         /* HV - assume 16MB kernel allocated htab */
2313         return 24;
2314     }
2315 }
2316 
2317 static inline uint32_t mfpvr(void)
2318 {
2319     uint32_t pvr;
2320 
2321     asm ("mfpvr %0"
2322          : "=r"(pvr));
2323     return pvr;
2324 }
2325 
2326 static void alter_insns(uint64_t *word, uint64_t flags, bool on)
2327 {
2328     if (on) {
2329         *word |= flags;
2330     } else {
2331         *word &= ~flags;
2332     }
2333 }
2334 
2335 static void kvmppc_host_cpu_class_init(ObjectClass *oc, void *data)
2336 {
2337     PowerPCCPUClass *pcc = POWERPC_CPU_CLASS(oc);
2338     uint32_t dcache_size = kvmppc_read_int_cpu_dt("d-cache-size");
2339     uint32_t icache_size = kvmppc_read_int_cpu_dt("i-cache-size");
2340 
2341     /* Now fix up the class with information we can query from the host */
2342     pcc->pvr = mfpvr();
2343 
2344     alter_insns(&pcc->insns_flags, PPC_ALTIVEC,
2345                 qemu_getauxval(AT_HWCAP) & PPC_FEATURE_HAS_ALTIVEC);
2346     alter_insns(&pcc->insns_flags2, PPC2_VSX,
2347                 qemu_getauxval(AT_HWCAP) & PPC_FEATURE_HAS_VSX);
2348     alter_insns(&pcc->insns_flags2, PPC2_DFP,
2349                 qemu_getauxval(AT_HWCAP) & PPC_FEATURE_HAS_DFP);
2350 
2351     if (dcache_size != -1) {
2352         pcc->l1_dcache_size = dcache_size;
2353     }
2354 
2355     if (icache_size != -1) {
2356         pcc->l1_icache_size = icache_size;
2357     }
2358 
2359 #if defined(TARGET_PPC64)
2360     pcc->radix_page_info = kvm_get_radix_page_info();
2361 
2362     if ((pcc->pvr & 0xffffff00) == CPU_POWERPC_POWER9_DD1) {
2363         /*
2364          * POWER9 DD1 has some bugs which make it not really ISA 3.00
2365          * compliant.  More importantly, advertising ISA 3.00
2366          * architected mode may prevent guests from activating
2367          * necessary DD1 workarounds.
2368          */
2369         pcc->pcr_supported &= ~(PCR_COMPAT_3_00 | PCR_COMPAT_2_07
2370                                 | PCR_COMPAT_2_06 | PCR_COMPAT_2_05);
2371     }
2372 #endif /* defined(TARGET_PPC64) */
2373 }
2374 
2375 bool kvmppc_has_cap_epr(void)
2376 {
2377     return cap_epr;
2378 }
2379 
2380 bool kvmppc_has_cap_fixup_hcalls(void)
2381 {
2382     return cap_fixup_hcalls;
2383 }
2384 
2385 bool kvmppc_has_cap_htm(void)
2386 {
2387     return cap_htm;
2388 }
2389 
2390 bool kvmppc_has_cap_mmu_radix(void)
2391 {
2392     return cap_mmu_radix;
2393 }
2394 
2395 bool kvmppc_has_cap_mmu_hash_v3(void)
2396 {
2397     return cap_mmu_hash_v3;
2398 }
2399 
2400 static bool kvmppc_power8_host(void)
2401 {
2402     bool ret = false;
2403 #ifdef TARGET_PPC64
2404     {
2405         uint32_t base_pvr = CPU_POWERPC_POWER_SERVER_MASK & mfpvr();
2406         ret = (base_pvr == CPU_POWERPC_POWER8E_BASE) ||
2407               (base_pvr == CPU_POWERPC_POWER8NVL_BASE) ||
2408               (base_pvr == CPU_POWERPC_POWER8_BASE);
2409     }
2410 #endif /* TARGET_PPC64 */
2411     return ret;
2412 }
2413 
2414 static int parse_cap_ppc_safe_cache(struct kvm_ppc_cpu_char c)
2415 {
2416     bool l1d_thread_priv_req = !kvmppc_power8_host();
2417 
2418     if (~c.behaviour & c.behaviour_mask & H_CPU_BEHAV_L1D_FLUSH_PR) {
2419         return 2;
2420     } else if ((!l1d_thread_priv_req ||
2421                 c.character & c.character_mask & H_CPU_CHAR_L1D_THREAD_PRIV) &&
2422                (c.character & c.character_mask
2423                 & (H_CPU_CHAR_L1D_FLUSH_ORI30 | H_CPU_CHAR_L1D_FLUSH_TRIG2))) {
2424         return 1;
2425     }
2426 
2427     return 0;
2428 }
2429 
2430 static int parse_cap_ppc_safe_bounds_check(struct kvm_ppc_cpu_char c)
2431 {
2432     if (~c.behaviour & c.behaviour_mask & H_CPU_BEHAV_BNDS_CHK_SPEC_BAR) {
2433         return 2;
2434     } else if (c.character & c.character_mask & H_CPU_CHAR_SPEC_BAR_ORI31) {
2435         return 1;
2436     }
2437 
2438     return 0;
2439 }
2440 
2441 static int parse_cap_ppc_safe_indirect_branch(struct kvm_ppc_cpu_char c)
2442 {
2443     if ((~c.behaviour & c.behaviour_mask & H_CPU_BEHAV_FLUSH_COUNT_CACHE) &&
2444         (~c.character & c.character_mask & H_CPU_CHAR_CACHE_COUNT_DIS) &&
2445         (~c.character & c.character_mask & H_CPU_CHAR_BCCTRL_SERIALISED)) {
2446         return SPAPR_CAP_FIXED_NA;
2447     } else if (c.behaviour & c.behaviour_mask & H_CPU_BEHAV_FLUSH_COUNT_CACHE) {
2448         return SPAPR_CAP_WORKAROUND;
2449     } else if (c.character & c.character_mask & H_CPU_CHAR_CACHE_COUNT_DIS) {
2450         return  SPAPR_CAP_FIXED_CCD;
2451     } else if (c.character & c.character_mask & H_CPU_CHAR_BCCTRL_SERIALISED) {
2452         return SPAPR_CAP_FIXED_IBS;
2453     }
2454 
2455     return 0;
2456 }
2457 
2458 static int parse_cap_ppc_count_cache_flush_assist(struct kvm_ppc_cpu_char c)
2459 {
2460     if (c.character & c.character_mask & H_CPU_CHAR_BCCTR_FLUSH_ASSIST) {
2461         return 1;
2462     }
2463     return 0;
2464 }
2465 
2466 bool kvmppc_has_cap_xive(void)
2467 {
2468     return cap_xive;
2469 }
2470 
2471 static void kvmppc_get_cpu_characteristics(KVMState *s)
2472 {
2473     struct kvm_ppc_cpu_char c;
2474     int ret;
2475 
2476     /* Assume broken */
2477     cap_ppc_safe_cache = 0;
2478     cap_ppc_safe_bounds_check = 0;
2479     cap_ppc_safe_indirect_branch = 0;
2480 
2481     ret = kvm_vm_check_extension(s, KVM_CAP_PPC_GET_CPU_CHAR);
2482     if (!ret) {
2483         return;
2484     }
2485     ret = kvm_vm_ioctl(s, KVM_PPC_GET_CPU_CHAR, &c);
2486     if (ret < 0) {
2487         return;
2488     }
2489 
2490     cap_ppc_safe_cache = parse_cap_ppc_safe_cache(c);
2491     cap_ppc_safe_bounds_check = parse_cap_ppc_safe_bounds_check(c);
2492     cap_ppc_safe_indirect_branch = parse_cap_ppc_safe_indirect_branch(c);
2493     cap_ppc_count_cache_flush_assist =
2494         parse_cap_ppc_count_cache_flush_assist(c);
2495 }
2496 
2497 int kvmppc_get_cap_safe_cache(void)
2498 {
2499     return cap_ppc_safe_cache;
2500 }
2501 
2502 int kvmppc_get_cap_safe_bounds_check(void)
2503 {
2504     return cap_ppc_safe_bounds_check;
2505 }
2506 
2507 int kvmppc_get_cap_safe_indirect_branch(void)
2508 {
2509     return cap_ppc_safe_indirect_branch;
2510 }
2511 
2512 int kvmppc_get_cap_count_cache_flush_assist(void)
2513 {
2514     return cap_ppc_count_cache_flush_assist;
2515 }
2516 
2517 bool kvmppc_has_cap_nested_kvm_hv(void)
2518 {
2519     return !!cap_ppc_nested_kvm_hv;
2520 }
2521 
2522 int kvmppc_set_cap_nested_kvm_hv(int enable)
2523 {
2524     return kvm_vm_enable_cap(kvm_state, KVM_CAP_PPC_NESTED_HV, 0, enable);
2525 }
2526 
2527 bool kvmppc_has_cap_spapr_vfio(void)
2528 {
2529     return cap_spapr_vfio;
2530 }
2531 
2532 int kvmppc_get_cap_large_decr(void)
2533 {
2534     return cap_large_decr;
2535 }
2536 
2537 int kvmppc_enable_cap_large_decr(PowerPCCPU *cpu, int enable)
2538 {
2539     CPUState *cs = CPU(cpu);
2540     uint64_t lpcr;
2541 
2542     kvm_get_one_reg(cs, KVM_REG_PPC_LPCR_64, &lpcr);
2543     /* Do we need to modify the LPCR? */
2544     if (!!(lpcr & LPCR_LD) != !!enable) {
2545         if (enable) {
2546             lpcr |= LPCR_LD;
2547         } else {
2548             lpcr &= ~LPCR_LD;
2549         }
2550         kvm_set_one_reg(cs, KVM_REG_PPC_LPCR_64, &lpcr);
2551         kvm_get_one_reg(cs, KVM_REG_PPC_LPCR_64, &lpcr);
2552 
2553         if (!!(lpcr & LPCR_LD) != !!enable) {
2554             return -1;
2555         }
2556     }
2557 
2558     return 0;
2559 }
2560 
2561 int kvmppc_has_cap_rpt_invalidate(void)
2562 {
2563     return cap_rpt_invalidate;
2564 }
2565 
2566 PowerPCCPUClass *kvm_ppc_get_host_cpu_class(void)
2567 {
2568     uint32_t host_pvr = mfpvr();
2569     PowerPCCPUClass *pvr_pcc;
2570 
2571     pvr_pcc = ppc_cpu_class_by_pvr(host_pvr);
2572     if (pvr_pcc == NULL) {
2573         pvr_pcc = ppc_cpu_class_by_pvr_mask(host_pvr);
2574     }
2575 
2576     return pvr_pcc;
2577 }
2578 
2579 static void pseries_machine_class_fixup(ObjectClass *oc, void *opaque)
2580 {
2581     MachineClass *mc = MACHINE_CLASS(oc);
2582 
2583     mc->default_cpu_type = TYPE_HOST_POWERPC_CPU;
2584 }
2585 
2586 static int kvm_ppc_register_host_cpu_type(void)
2587 {
2588     TypeInfo type_info = {
2589         .name = TYPE_HOST_POWERPC_CPU,
2590         .class_init = kvmppc_host_cpu_class_init,
2591     };
2592     PowerPCCPUClass *pvr_pcc;
2593     ObjectClass *oc;
2594     DeviceClass *dc;
2595     int i;
2596 
2597     pvr_pcc = kvm_ppc_get_host_cpu_class();
2598     if (pvr_pcc == NULL) {
2599         return -1;
2600     }
2601     type_info.parent = object_class_get_name(OBJECT_CLASS(pvr_pcc));
2602     type_register(&type_info);
2603     /* override TCG default cpu type with 'host' cpu model */
2604     object_class_foreach(pseries_machine_class_fixup, TYPE_SPAPR_MACHINE,
2605                          false, NULL);
2606 
2607     oc = object_class_by_name(type_info.name);
2608     g_assert(oc);
2609 
2610     /*
2611      * Update generic CPU family class alias (e.g. on a POWER8NVL host,
2612      * we want "POWER8" to be a "family" alias that points to the current
2613      * host CPU type, too)
2614      */
2615     dc = DEVICE_CLASS(ppc_cpu_get_family_class(pvr_pcc));
2616     for (i = 0; ppc_cpu_aliases[i].alias != NULL; i++) {
2617         if (strcasecmp(ppc_cpu_aliases[i].alias, dc->desc) == 0) {
2618             char *suffix;
2619 
2620             ppc_cpu_aliases[i].model = g_strdup(object_class_get_name(oc));
2621             suffix = strstr(ppc_cpu_aliases[i].model, POWERPC_CPU_TYPE_SUFFIX);
2622             if (suffix) {
2623                 *suffix = 0;
2624             }
2625             break;
2626         }
2627     }
2628 
2629     return 0;
2630 }
2631 
2632 int kvmppc_define_rtas_kernel_token(uint32_t token, const char *function)
2633 {
2634     struct kvm_rtas_token_args args = {
2635         .token = token,
2636     };
2637 
2638     if (!kvm_check_extension(kvm_state, KVM_CAP_PPC_RTAS)) {
2639         return -ENOENT;
2640     }
2641 
2642     strncpy(args.name, function, sizeof(args.name) - 1);
2643 
2644     return kvm_vm_ioctl(kvm_state, KVM_PPC_RTAS_DEFINE_TOKEN, &args);
2645 }
2646 
2647 int kvmppc_get_htab_fd(bool write, uint64_t index, Error **errp)
2648 {
2649     struct kvm_get_htab_fd s = {
2650         .flags = write ? KVM_GET_HTAB_WRITE : 0,
2651         .start_index = index,
2652     };
2653     int ret;
2654 
2655     if (!cap_htab_fd) {
2656         error_setg(errp, "KVM version doesn't support %s the HPT",
2657                    write ? "writing" : "reading");
2658         return -ENOTSUP;
2659     }
2660 
2661     ret = kvm_vm_ioctl(kvm_state, KVM_PPC_GET_HTAB_FD, &s);
2662     if (ret < 0) {
2663         error_setg(errp, "Unable to open fd for %s HPT %s KVM: %s",
2664                    write ? "writing" : "reading", write ? "to" : "from",
2665                    strerror(errno));
2666         return -errno;
2667     }
2668 
2669     return ret;
2670 }
2671 
2672 int kvmppc_save_htab(QEMUFile *f, int fd, size_t bufsize, int64_t max_ns)
2673 {
2674     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2675     uint8_t buf[bufsize];
2676     ssize_t rc;
2677 
2678     do {
2679         rc = read(fd, buf, bufsize);
2680         if (rc < 0) {
2681             fprintf(stderr, "Error reading data from KVM HTAB fd: %s\n",
2682                     strerror(errno));
2683             return rc;
2684         } else if (rc) {
2685             uint8_t *buffer = buf;
2686             ssize_t n = rc;
2687             while (n) {
2688                 struct kvm_get_htab_header *head =
2689                     (struct kvm_get_htab_header *) buffer;
2690                 size_t chunksize = sizeof(*head) +
2691                      HASH_PTE_SIZE_64 * head->n_valid;
2692 
2693                 qemu_put_be32(f, head->index);
2694                 qemu_put_be16(f, head->n_valid);
2695                 qemu_put_be16(f, head->n_invalid);
2696                 qemu_put_buffer(f, (void *)(head + 1),
2697                                 HASH_PTE_SIZE_64 * head->n_valid);
2698 
2699                 buffer += chunksize;
2700                 n -= chunksize;
2701             }
2702         }
2703     } while ((rc != 0)
2704              && ((max_ns < 0) ||
2705                  ((qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) < max_ns)));
2706 
2707     return (rc == 0) ? 1 : 0;
2708 }
2709 
2710 int kvmppc_load_htab_chunk(QEMUFile *f, int fd, uint32_t index,
2711                            uint16_t n_valid, uint16_t n_invalid, Error **errp)
2712 {
2713     struct kvm_get_htab_header *buf;
2714     size_t chunksize = sizeof(*buf) + n_valid * HASH_PTE_SIZE_64;
2715     ssize_t rc;
2716 
2717     buf = alloca(chunksize);
2718     buf->index = index;
2719     buf->n_valid = n_valid;
2720     buf->n_invalid = n_invalid;
2721 
2722     qemu_get_buffer(f, (void *)(buf + 1), HASH_PTE_SIZE_64 * n_valid);
2723 
2724     rc = write(fd, buf, chunksize);
2725     if (rc < 0) {
2726         error_setg_errno(errp, errno, "Error writing the KVM hash table");
2727         return -errno;
2728     }
2729     if (rc != chunksize) {
2730         /* We should never get a short write on a single chunk */
2731         error_setg(errp, "Short write while restoring the KVM hash table");
2732         return -ENOSPC;
2733     }
2734     return 0;
2735 }
2736 
2737 bool kvm_arch_stop_on_emulation_error(CPUState *cpu)
2738 {
2739     return true;
2740 }
2741 
2742 void kvm_arch_init_irq_routing(KVMState *s)
2743 {
2744 }
2745 
2746 void kvmppc_read_hptes(ppc_hash_pte64_t *hptes, hwaddr ptex, int n)
2747 {
2748     int fd, rc;
2749     int i;
2750 
2751     fd = kvmppc_get_htab_fd(false, ptex, &error_abort);
2752 
2753     i = 0;
2754     while (i < n) {
2755         struct kvm_get_htab_header *hdr;
2756         int m = n < HPTES_PER_GROUP ? n : HPTES_PER_GROUP;
2757         char buf[sizeof(*hdr) + m * HASH_PTE_SIZE_64];
2758 
2759         rc = read(fd, buf, sizeof(buf));
2760         if (rc < 0) {
2761             hw_error("kvmppc_read_hptes: Unable to read HPTEs");
2762         }
2763 
2764         hdr = (struct kvm_get_htab_header *)buf;
2765         while ((i < n) && ((char *)hdr < (buf + rc))) {
2766             int invalid = hdr->n_invalid, valid = hdr->n_valid;
2767 
2768             if (hdr->index != (ptex + i)) {
2769                 hw_error("kvmppc_read_hptes: Unexpected HPTE index %"PRIu32
2770                          " != (%"HWADDR_PRIu" + %d", hdr->index, ptex, i);
2771             }
2772 
2773             if (n - i < valid) {
2774                 valid = n - i;
2775             }
2776             memcpy(hptes + i, hdr + 1, HASH_PTE_SIZE_64 * valid);
2777             i += valid;
2778 
2779             if ((n - i) < invalid) {
2780                 invalid = n - i;
2781             }
2782             memset(hptes + i, 0, invalid * HASH_PTE_SIZE_64);
2783             i += invalid;
2784 
2785             hdr = (struct kvm_get_htab_header *)
2786                 ((char *)(hdr + 1) + HASH_PTE_SIZE_64 * hdr->n_valid);
2787         }
2788     }
2789 
2790     close(fd);
2791 }
2792 
2793 void kvmppc_write_hpte(hwaddr ptex, uint64_t pte0, uint64_t pte1)
2794 {
2795     int fd, rc;
2796     struct {
2797         struct kvm_get_htab_header hdr;
2798         uint64_t pte0;
2799         uint64_t pte1;
2800     } buf;
2801 
2802     fd = kvmppc_get_htab_fd(true, 0 /* Ignored */, &error_abort);
2803 
2804     buf.hdr.n_valid = 1;
2805     buf.hdr.n_invalid = 0;
2806     buf.hdr.index = ptex;
2807     buf.pte0 = cpu_to_be64(pte0);
2808     buf.pte1 = cpu_to_be64(pte1);
2809 
2810     rc = write(fd, &buf, sizeof(buf));
2811     if (rc != sizeof(buf)) {
2812         hw_error("kvmppc_write_hpte: Unable to update KVM HPT");
2813     }
2814     close(fd);
2815 }
2816 
2817 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
2818                              uint64_t address, uint32_t data, PCIDevice *dev)
2819 {
2820     return 0;
2821 }
2822 
2823 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
2824                                 int vector, PCIDevice *dev)
2825 {
2826     return 0;
2827 }
2828 
2829 int kvm_arch_release_virq_post(int virq)
2830 {
2831     return 0;
2832 }
2833 
2834 int kvm_arch_msi_data_to_gsi(uint32_t data)
2835 {
2836     return data & 0xffff;
2837 }
2838 
2839 #if defined(TARGET_PPC64)
2840 int kvm_handle_nmi(PowerPCCPU *cpu, struct kvm_run *run)
2841 {
2842     uint16_t flags = run->flags & KVM_RUN_PPC_NMI_DISP_MASK;
2843 
2844     cpu_synchronize_state(CPU(cpu));
2845 
2846     spapr_mce_req_event(cpu, flags == KVM_RUN_PPC_NMI_DISP_FULLY_RECOV);
2847 
2848     return 0;
2849 }
2850 #endif
2851 
2852 int kvmppc_enable_hwrng(void)
2853 {
2854     if (!kvm_enabled() || !kvm_check_extension(kvm_state, KVM_CAP_PPC_HWRNG)) {
2855         return -1;
2856     }
2857 
2858     return kvmppc_enable_hcall(kvm_state, H_RANDOM);
2859 }
2860 
2861 void kvmppc_check_papr_resize_hpt(Error **errp)
2862 {
2863     if (!kvm_enabled()) {
2864         return; /* No KVM, we're good */
2865     }
2866 
2867     if (cap_resize_hpt) {
2868         return; /* Kernel has explicit support, we're good */
2869     }
2870 
2871     /* Otherwise fallback on looking for PR KVM */
2872     if (kvmppc_is_pr(kvm_state)) {
2873         return;
2874     }
2875 
2876     error_setg(errp,
2877                "Hash page table resizing not available with this KVM version");
2878 }
2879 
2880 int kvmppc_resize_hpt_prepare(PowerPCCPU *cpu, target_ulong flags, int shift)
2881 {
2882     CPUState *cs = CPU(cpu);
2883     struct kvm_ppc_resize_hpt rhpt = {
2884         .flags = flags,
2885         .shift = shift,
2886     };
2887 
2888     if (!cap_resize_hpt) {
2889         return -ENOSYS;
2890     }
2891 
2892     return kvm_vm_ioctl(cs->kvm_state, KVM_PPC_RESIZE_HPT_PREPARE, &rhpt);
2893 }
2894 
2895 int kvmppc_resize_hpt_commit(PowerPCCPU *cpu, target_ulong flags, int shift)
2896 {
2897     CPUState *cs = CPU(cpu);
2898     struct kvm_ppc_resize_hpt rhpt = {
2899         .flags = flags,
2900         .shift = shift,
2901     };
2902 
2903     if (!cap_resize_hpt) {
2904         return -ENOSYS;
2905     }
2906 
2907     return kvm_vm_ioctl(cs->kvm_state, KVM_PPC_RESIZE_HPT_COMMIT, &rhpt);
2908 }
2909 
2910 /*
2911  * This is a helper function to detect a post migration scenario
2912  * in which a guest, running as KVM-HV, freezes in cpu_post_load because
2913  * the guest kernel can't handle a PVR value other than the actual host
2914  * PVR in KVM_SET_SREGS, even if pvr_match() returns true.
2915  *
2916  * If we don't have cap_ppc_pvr_compat and we're not running in PR
2917  * (so, we're HV), return true. The workaround itself is done in
2918  * cpu_post_load.
2919  *
2920  * The order here is important: we'll only check for KVM PR as a
2921  * fallback if the guest kernel can't handle the situation itself.
2922  * We need to avoid as much as possible querying the running KVM type
2923  * in QEMU level.
2924  */
2925 bool kvmppc_pvr_workaround_required(PowerPCCPU *cpu)
2926 {
2927     CPUState *cs = CPU(cpu);
2928 
2929     if (!kvm_enabled()) {
2930         return false;
2931     }
2932 
2933     if (cap_ppc_pvr_compat) {
2934         return false;
2935     }
2936 
2937     return !kvmppc_is_pr(cs->kvm_state);
2938 }
2939 
2940 void kvmppc_set_reg_ppc_online(PowerPCCPU *cpu, unsigned int online)
2941 {
2942     CPUState *cs = CPU(cpu);
2943 
2944     if (kvm_enabled()) {
2945         kvm_set_one_reg(cs, KVM_REG_PPC_ONLINE, &online);
2946     }
2947 }
2948 
2949 void kvmppc_set_reg_tb_offset(PowerPCCPU *cpu, int64_t tb_offset)
2950 {
2951     CPUState *cs = CPU(cpu);
2952 
2953     if (kvm_enabled()) {
2954         kvm_set_one_reg(cs, KVM_REG_PPC_TB_OFFSET, &tb_offset);
2955     }
2956 }
2957 
2958 bool kvm_arch_cpu_check_are_resettable(void)
2959 {
2960     return true;
2961 }
2962