xref: /openbmc/qemu/target/ppc/kvm.c (revision a46d410c)
1 /*
2  * PowerPC implementation of KVM hooks
3  *
4  * Copyright IBM Corp. 2007
5  * Copyright (C) 2011 Freescale Semiconductor, Inc.
6  *
7  * Authors:
8  *  Jerone Young <jyoung5@us.ibm.com>
9  *  Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
10  *  Hollis Blanchard <hollisb@us.ibm.com>
11  *
12  * This work is licensed under the terms of the GNU GPL, version 2 or later.
13  * See the COPYING file in the top-level directory.
14  *
15  */
16 
17 #include "qemu/osdep.h"
18 #include <dirent.h>
19 #include <sys/ioctl.h>
20 #include <sys/vfs.h>
21 
22 #include <linux/kvm.h>
23 
24 #include "qapi/error.h"
25 #include "qemu/error-report.h"
26 #include "cpu.h"
27 #include "cpu-models.h"
28 #include "qemu/timer.h"
29 #include "sysemu/hw_accel.h"
30 #include "kvm_ppc.h"
31 #include "sysemu/cpus.h"
32 #include "sysemu/device_tree.h"
33 #include "mmu-hash64.h"
34 
35 #include "hw/sysbus.h"
36 #include "hw/ppc/spapr.h"
37 #include "hw/ppc/spapr_cpu_core.h"
38 #include "hw/hw.h"
39 #include "hw/ppc/ppc.h"
40 #include "migration/qemu-file-types.h"
41 #include "sysemu/watchdog.h"
42 #include "trace.h"
43 #include "exec/gdbstub.h"
44 #include "exec/memattrs.h"
45 #include "exec/ram_addr.h"
46 #include "sysemu/hostmem.h"
47 #include "qemu/cutils.h"
48 #include "qemu/main-loop.h"
49 #include "qemu/mmap-alloc.h"
50 #include "elf.h"
51 #include "sysemu/kvm_int.h"
52 
53 #define PROC_DEVTREE_CPU      "/proc/device-tree/cpus/"
54 
55 #define DEBUG_RETURN_GUEST 0
56 #define DEBUG_RETURN_GDB   1
57 
58 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
59     KVM_CAP_LAST_INFO
60 };
61 
62 static int cap_interrupt_unset;
63 static int cap_segstate;
64 static int cap_booke_sregs;
65 static int cap_ppc_smt;
66 static int cap_ppc_smt_possible;
67 static int cap_spapr_tce;
68 static int cap_spapr_tce_64;
69 static int cap_spapr_multitce;
70 static int cap_spapr_vfio;
71 static int cap_hior;
72 static int cap_one_reg;
73 static int cap_epr;
74 static int cap_ppc_watchdog;
75 static int cap_papr;
76 static int cap_htab_fd;
77 static int cap_fixup_hcalls;
78 static int cap_htm;             /* Hardware transactional memory support */
79 static int cap_mmu_radix;
80 static int cap_mmu_hash_v3;
81 static int cap_xive;
82 static int cap_resize_hpt;
83 static int cap_ppc_pvr_compat;
84 static int cap_ppc_safe_cache;
85 static int cap_ppc_safe_bounds_check;
86 static int cap_ppc_safe_indirect_branch;
87 static int cap_ppc_count_cache_flush_assist;
88 static int cap_ppc_nested_kvm_hv;
89 static int cap_large_decr;
90 static int cap_fwnmi;
91 static int cap_rpt_invalidate;
92 
93 static uint32_t debug_inst_opcode;
94 
95 /*
96  * Check whether we are running with KVM-PR (instead of KVM-HV).  This
97  * should only be used for fallback tests - generally we should use
98  * explicit capabilities for the features we want, rather than
99  * assuming what is/isn't available depending on the KVM variant.
100  */
101 static bool kvmppc_is_pr(KVMState *ks)
102 {
103     /* Assume KVM-PR if the GET_PVINFO capability is available */
104     return kvm_vm_check_extension(ks, KVM_CAP_PPC_GET_PVINFO) != 0;
105 }
106 
107 static int kvm_ppc_register_host_cpu_type(void);
108 static void kvmppc_get_cpu_characteristics(KVMState *s);
109 static int kvmppc_get_dec_bits(void);
110 
111 int kvm_arch_init(MachineState *ms, KVMState *s)
112 {
113     cap_interrupt_unset = kvm_check_extension(s, KVM_CAP_PPC_UNSET_IRQ);
114     cap_segstate = kvm_check_extension(s, KVM_CAP_PPC_SEGSTATE);
115     cap_booke_sregs = kvm_check_extension(s, KVM_CAP_PPC_BOOKE_SREGS);
116     cap_ppc_smt_possible = kvm_vm_check_extension(s, KVM_CAP_PPC_SMT_POSSIBLE);
117     cap_spapr_tce = kvm_check_extension(s, KVM_CAP_SPAPR_TCE);
118     cap_spapr_tce_64 = kvm_check_extension(s, KVM_CAP_SPAPR_TCE_64);
119     cap_spapr_multitce = kvm_check_extension(s, KVM_CAP_SPAPR_MULTITCE);
120     cap_spapr_vfio = kvm_vm_check_extension(s, KVM_CAP_SPAPR_TCE_VFIO);
121     cap_one_reg = kvm_check_extension(s, KVM_CAP_ONE_REG);
122     cap_hior = kvm_check_extension(s, KVM_CAP_PPC_HIOR);
123     cap_epr = kvm_check_extension(s, KVM_CAP_PPC_EPR);
124     cap_ppc_watchdog = kvm_check_extension(s, KVM_CAP_PPC_BOOKE_WATCHDOG);
125     /*
126      * Note: we don't set cap_papr here, because this capability is
127      * only activated after this by kvmppc_set_papr()
128      */
129     cap_htab_fd = kvm_vm_check_extension(s, KVM_CAP_PPC_HTAB_FD);
130     cap_fixup_hcalls = kvm_check_extension(s, KVM_CAP_PPC_FIXUP_HCALL);
131     cap_ppc_smt = kvm_vm_check_extension(s, KVM_CAP_PPC_SMT);
132     cap_htm = kvm_vm_check_extension(s, KVM_CAP_PPC_HTM);
133     cap_mmu_radix = kvm_vm_check_extension(s, KVM_CAP_PPC_MMU_RADIX);
134     cap_mmu_hash_v3 = kvm_vm_check_extension(s, KVM_CAP_PPC_MMU_HASH_V3);
135     cap_xive = kvm_vm_check_extension(s, KVM_CAP_PPC_IRQ_XIVE);
136     cap_resize_hpt = kvm_vm_check_extension(s, KVM_CAP_SPAPR_RESIZE_HPT);
137     kvmppc_get_cpu_characteristics(s);
138     cap_ppc_nested_kvm_hv = kvm_vm_check_extension(s, KVM_CAP_PPC_NESTED_HV);
139     cap_large_decr = kvmppc_get_dec_bits();
140     cap_fwnmi = kvm_vm_check_extension(s, KVM_CAP_PPC_FWNMI);
141     /*
142      * Note: setting it to false because there is not such capability
143      * in KVM at this moment.
144      *
145      * TODO: call kvm_vm_check_extension() with the right capability
146      * after the kernel starts implementing it.
147      */
148     cap_ppc_pvr_compat = false;
149 
150     if (!kvm_check_extension(s, KVM_CAP_PPC_IRQ_LEVEL)) {
151         error_report("KVM: Host kernel doesn't have level irq capability");
152         exit(1);
153     }
154 
155     cap_rpt_invalidate = kvm_vm_check_extension(s, KVM_CAP_PPC_RPT_INVALIDATE);
156     kvm_ppc_register_host_cpu_type();
157 
158     return 0;
159 }
160 
161 int kvm_arch_irqchip_create(KVMState *s)
162 {
163     return 0;
164 }
165 
166 static int kvm_arch_sync_sregs(PowerPCCPU *cpu)
167 {
168     CPUPPCState *cenv = &cpu->env;
169     CPUState *cs = CPU(cpu);
170     struct kvm_sregs sregs;
171     int ret;
172 
173     if (cenv->excp_model == POWERPC_EXCP_BOOKE) {
174         /*
175          * What we're really trying to say is "if we're on BookE, we
176          * use the native PVR for now". This is the only sane way to
177          * check it though, so we potentially confuse users that they
178          * can run BookE guests on BookS. Let's hope nobody dares
179          * enough :)
180          */
181         return 0;
182     } else {
183         if (!cap_segstate) {
184             fprintf(stderr, "kvm error: missing PVR setting capability\n");
185             return -ENOSYS;
186         }
187     }
188 
189     ret = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs);
190     if (ret) {
191         return ret;
192     }
193 
194     sregs.pvr = cenv->spr[SPR_PVR];
195     return kvm_vcpu_ioctl(cs, KVM_SET_SREGS, &sregs);
196 }
197 
198 /* Set up a shared TLB array with KVM */
199 static int kvm_booke206_tlb_init(PowerPCCPU *cpu)
200 {
201     CPUPPCState *env = &cpu->env;
202     CPUState *cs = CPU(cpu);
203     struct kvm_book3e_206_tlb_params params = {};
204     struct kvm_config_tlb cfg = {};
205     unsigned int entries = 0;
206     int ret, i;
207 
208     if (!kvm_enabled() ||
209         !kvm_check_extension(cs->kvm_state, KVM_CAP_SW_TLB)) {
210         return 0;
211     }
212 
213     assert(ARRAY_SIZE(params.tlb_sizes) == BOOKE206_MAX_TLBN);
214 
215     for (i = 0; i < BOOKE206_MAX_TLBN; i++) {
216         params.tlb_sizes[i] = booke206_tlb_size(env, i);
217         params.tlb_ways[i] = booke206_tlb_ways(env, i);
218         entries += params.tlb_sizes[i];
219     }
220 
221     assert(entries == env->nb_tlb);
222     assert(sizeof(struct kvm_book3e_206_tlb_entry) == sizeof(ppcmas_tlb_t));
223 
224     env->tlb_dirty = true;
225 
226     cfg.array = (uintptr_t)env->tlb.tlbm;
227     cfg.array_len = sizeof(ppcmas_tlb_t) * entries;
228     cfg.params = (uintptr_t)&params;
229     cfg.mmu_type = KVM_MMU_FSL_BOOKE_NOHV;
230 
231     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_SW_TLB, 0, (uintptr_t)&cfg);
232     if (ret < 0) {
233         fprintf(stderr, "%s: couldn't enable KVM_CAP_SW_TLB: %s\n",
234                 __func__, strerror(-ret));
235         return ret;
236     }
237 
238     env->kvm_sw_tlb = true;
239     return 0;
240 }
241 
242 
243 #if defined(TARGET_PPC64)
244 static void kvm_get_smmu_info(struct kvm_ppc_smmu_info *info, Error **errp)
245 {
246     int ret;
247 
248     assert(kvm_state != NULL);
249 
250     if (!kvm_check_extension(kvm_state, KVM_CAP_PPC_GET_SMMU_INFO)) {
251         error_setg(errp, "KVM doesn't expose the MMU features it supports");
252         error_append_hint(errp, "Consider switching to a newer KVM\n");
253         return;
254     }
255 
256     ret = kvm_vm_ioctl(kvm_state, KVM_PPC_GET_SMMU_INFO, info);
257     if (ret == 0) {
258         return;
259     }
260 
261     error_setg_errno(errp, -ret,
262                      "KVM failed to provide the MMU features it supports");
263 }
264 
265 struct ppc_radix_page_info *kvm_get_radix_page_info(void)
266 {
267     KVMState *s = KVM_STATE(current_accel());
268     struct ppc_radix_page_info *radix_page_info;
269     struct kvm_ppc_rmmu_info rmmu_info;
270     int i;
271 
272     if (!kvm_check_extension(s, KVM_CAP_PPC_MMU_RADIX)) {
273         return NULL;
274     }
275     if (kvm_vm_ioctl(s, KVM_PPC_GET_RMMU_INFO, &rmmu_info)) {
276         return NULL;
277     }
278     radix_page_info = g_malloc0(sizeof(*radix_page_info));
279     radix_page_info->count = 0;
280     for (i = 0; i < PPC_PAGE_SIZES_MAX_SZ; i++) {
281         if (rmmu_info.ap_encodings[i]) {
282             radix_page_info->entries[i] = rmmu_info.ap_encodings[i];
283             radix_page_info->count++;
284         }
285     }
286     return radix_page_info;
287 }
288 
289 target_ulong kvmppc_configure_v3_mmu(PowerPCCPU *cpu,
290                                      bool radix, bool gtse,
291                                      uint64_t proc_tbl)
292 {
293     CPUState *cs = CPU(cpu);
294     int ret;
295     uint64_t flags = 0;
296     struct kvm_ppc_mmuv3_cfg cfg = {
297         .process_table = proc_tbl,
298     };
299 
300     if (radix) {
301         flags |= KVM_PPC_MMUV3_RADIX;
302     }
303     if (gtse) {
304         flags |= KVM_PPC_MMUV3_GTSE;
305     }
306     cfg.flags = flags;
307     ret = kvm_vm_ioctl(cs->kvm_state, KVM_PPC_CONFIGURE_V3_MMU, &cfg);
308     switch (ret) {
309     case 0:
310         return H_SUCCESS;
311     case -EINVAL:
312         return H_PARAMETER;
313     case -ENODEV:
314         return H_NOT_AVAILABLE;
315     default:
316         return H_HARDWARE;
317     }
318 }
319 
320 bool kvmppc_hpt_needs_host_contiguous_pages(void)
321 {
322     static struct kvm_ppc_smmu_info smmu_info;
323 
324     if (!kvm_enabled()) {
325         return false;
326     }
327 
328     kvm_get_smmu_info(&smmu_info, &error_fatal);
329     return !!(smmu_info.flags & KVM_PPC_PAGE_SIZES_REAL);
330 }
331 
332 void kvm_check_mmu(PowerPCCPU *cpu, Error **errp)
333 {
334     struct kvm_ppc_smmu_info smmu_info;
335     int iq, ik, jq, jk;
336     Error *local_err = NULL;
337 
338     /* For now, we only have anything to check on hash64 MMUs */
339     if (!cpu->hash64_opts || !kvm_enabled()) {
340         return;
341     }
342 
343     kvm_get_smmu_info(&smmu_info, &local_err);
344     if (local_err) {
345         error_propagate(errp, local_err);
346         return;
347     }
348 
349     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)
350         && !(smmu_info.flags & KVM_PPC_1T_SEGMENTS)) {
351         error_setg(errp,
352                    "KVM does not support 1TiB segments which guest expects");
353         return;
354     }
355 
356     if (smmu_info.slb_size < cpu->hash64_opts->slb_size) {
357         error_setg(errp, "KVM only supports %u SLB entries, but guest needs %u",
358                    smmu_info.slb_size, cpu->hash64_opts->slb_size);
359         return;
360     }
361 
362     /*
363      * Verify that every pagesize supported by the cpu model is
364      * supported by KVM with the same encodings
365      */
366     for (iq = 0; iq < ARRAY_SIZE(cpu->hash64_opts->sps); iq++) {
367         PPCHash64SegmentPageSizes *qsps = &cpu->hash64_opts->sps[iq];
368         struct kvm_ppc_one_seg_page_size *ksps;
369 
370         for (ik = 0; ik < ARRAY_SIZE(smmu_info.sps); ik++) {
371             if (qsps->page_shift == smmu_info.sps[ik].page_shift) {
372                 break;
373             }
374         }
375         if (ik >= ARRAY_SIZE(smmu_info.sps)) {
376             error_setg(errp, "KVM doesn't support for base page shift %u",
377                        qsps->page_shift);
378             return;
379         }
380 
381         ksps = &smmu_info.sps[ik];
382         if (ksps->slb_enc != qsps->slb_enc) {
383             error_setg(errp,
384 "KVM uses SLB encoding 0x%x for page shift %u, but guest expects 0x%x",
385                        ksps->slb_enc, ksps->page_shift, qsps->slb_enc);
386             return;
387         }
388 
389         for (jq = 0; jq < ARRAY_SIZE(qsps->enc); jq++) {
390             for (jk = 0; jk < ARRAY_SIZE(ksps->enc); jk++) {
391                 if (qsps->enc[jq].page_shift == ksps->enc[jk].page_shift) {
392                     break;
393                 }
394             }
395 
396             if (jk >= ARRAY_SIZE(ksps->enc)) {
397                 error_setg(errp, "KVM doesn't support page shift %u/%u",
398                            qsps->enc[jq].page_shift, qsps->page_shift);
399                 return;
400             }
401             if (qsps->enc[jq].pte_enc != ksps->enc[jk].pte_enc) {
402                 error_setg(errp,
403 "KVM uses PTE encoding 0x%x for page shift %u/%u, but guest expects 0x%x",
404                            ksps->enc[jk].pte_enc, qsps->enc[jq].page_shift,
405                            qsps->page_shift, qsps->enc[jq].pte_enc);
406                 return;
407             }
408         }
409     }
410 
411     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
412         /*
413          * Mostly what guest pagesizes we can use are related to the
414          * host pages used to map guest RAM, which is handled in the
415          * platform code. Cache-Inhibited largepages (64k) however are
416          * used for I/O, so if they're mapped to the host at all it
417          * will be a normal mapping, not a special hugepage one used
418          * for RAM.
419          */
420         if (qemu_real_host_page_size() < 0x10000) {
421             error_setg(errp,
422                        "KVM can't supply 64kiB CI pages, which guest expects");
423         }
424     }
425 }
426 #endif /* !defined (TARGET_PPC64) */
427 
428 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
429 {
430     return POWERPC_CPU(cpu)->vcpu_id;
431 }
432 
433 /*
434  * e500 supports 2 h/w breakpoint and 2 watchpoint.  book3s supports
435  * only 1 watchpoint, so array size of 4 is sufficient for now.
436  */
437 #define MAX_HW_BKPTS 4
438 
439 static struct HWBreakpoint {
440     target_ulong addr;
441     int type;
442 } hw_debug_points[MAX_HW_BKPTS];
443 
444 static CPUWatchpoint hw_watchpoint;
445 
446 /* Default there is no breakpoint and watchpoint supported */
447 static int max_hw_breakpoint;
448 static int max_hw_watchpoint;
449 static int nb_hw_breakpoint;
450 static int nb_hw_watchpoint;
451 
452 static void kvmppc_hw_debug_points_init(CPUPPCState *cenv)
453 {
454     if (cenv->excp_model == POWERPC_EXCP_BOOKE) {
455         max_hw_breakpoint = 2;
456         max_hw_watchpoint = 2;
457     }
458 
459     if ((max_hw_breakpoint + max_hw_watchpoint) > MAX_HW_BKPTS) {
460         fprintf(stderr, "Error initializing h/w breakpoints\n");
461         return;
462     }
463 }
464 
465 int kvm_arch_init_vcpu(CPUState *cs)
466 {
467     PowerPCCPU *cpu = POWERPC_CPU(cs);
468     CPUPPCState *cenv = &cpu->env;
469     int ret;
470 
471     /* Synchronize sregs with kvm */
472     ret = kvm_arch_sync_sregs(cpu);
473     if (ret) {
474         if (ret == -EINVAL) {
475             error_report("Register sync failed... If you're using kvm-hv.ko,"
476                          " only \"-cpu host\" is possible");
477         }
478         return ret;
479     }
480 
481     switch (cenv->mmu_model) {
482     case POWERPC_MMU_BOOKE206:
483         /* This target supports access to KVM's guest TLB */
484         ret = kvm_booke206_tlb_init(cpu);
485         break;
486     case POWERPC_MMU_2_07:
487         if (!cap_htm && !kvmppc_is_pr(cs->kvm_state)) {
488             /*
489              * KVM-HV has transactional memory on POWER8 also without
490              * the KVM_CAP_PPC_HTM extension, so enable it here
491              * instead as long as it's available to userspace on the
492              * host.
493              */
494             if (qemu_getauxval(AT_HWCAP2) & PPC_FEATURE2_HAS_HTM) {
495                 cap_htm = true;
496             }
497         }
498         break;
499     default:
500         break;
501     }
502 
503     kvm_get_one_reg(cs, KVM_REG_PPC_DEBUG_INST, &debug_inst_opcode);
504     kvmppc_hw_debug_points_init(cenv);
505 
506     return ret;
507 }
508 
509 int kvm_arch_destroy_vcpu(CPUState *cs)
510 {
511     return 0;
512 }
513 
514 static void kvm_sw_tlb_put(PowerPCCPU *cpu)
515 {
516     CPUPPCState *env = &cpu->env;
517     CPUState *cs = CPU(cpu);
518     struct kvm_dirty_tlb dirty_tlb;
519     unsigned char *bitmap;
520     int ret;
521 
522     if (!env->kvm_sw_tlb) {
523         return;
524     }
525 
526     bitmap = g_malloc((env->nb_tlb + 7) / 8);
527     memset(bitmap, 0xFF, (env->nb_tlb + 7) / 8);
528 
529     dirty_tlb.bitmap = (uintptr_t)bitmap;
530     dirty_tlb.num_dirty = env->nb_tlb;
531 
532     ret = kvm_vcpu_ioctl(cs, KVM_DIRTY_TLB, &dirty_tlb);
533     if (ret) {
534         fprintf(stderr, "%s: KVM_DIRTY_TLB: %s\n",
535                 __func__, strerror(-ret));
536     }
537 
538     g_free(bitmap);
539 }
540 
541 static void kvm_get_one_spr(CPUState *cs, uint64_t id, int spr)
542 {
543     PowerPCCPU *cpu = POWERPC_CPU(cs);
544     CPUPPCState *env = &cpu->env;
545     union {
546         uint32_t u32;
547         uint64_t u64;
548     } val;
549     struct kvm_one_reg reg = {
550         .id = id,
551         .addr = (uintptr_t) &val,
552     };
553     int ret;
554 
555     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
556     if (ret != 0) {
557         trace_kvm_failed_spr_get(spr, strerror(errno));
558     } else {
559         switch (id & KVM_REG_SIZE_MASK) {
560         case KVM_REG_SIZE_U32:
561             env->spr[spr] = val.u32;
562             break;
563 
564         case KVM_REG_SIZE_U64:
565             env->spr[spr] = val.u64;
566             break;
567 
568         default:
569             /* Don't handle this size yet */
570             abort();
571         }
572     }
573 }
574 
575 static void kvm_put_one_spr(CPUState *cs, uint64_t id, int spr)
576 {
577     PowerPCCPU *cpu = POWERPC_CPU(cs);
578     CPUPPCState *env = &cpu->env;
579     union {
580         uint32_t u32;
581         uint64_t u64;
582     } val;
583     struct kvm_one_reg reg = {
584         .id = id,
585         .addr = (uintptr_t) &val,
586     };
587     int ret;
588 
589     switch (id & KVM_REG_SIZE_MASK) {
590     case KVM_REG_SIZE_U32:
591         val.u32 = env->spr[spr];
592         break;
593 
594     case KVM_REG_SIZE_U64:
595         val.u64 = env->spr[spr];
596         break;
597 
598     default:
599         /* Don't handle this size yet */
600         abort();
601     }
602 
603     ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
604     if (ret != 0) {
605         trace_kvm_failed_spr_set(spr, strerror(errno));
606     }
607 }
608 
609 static int kvm_put_fp(CPUState *cs)
610 {
611     PowerPCCPU *cpu = POWERPC_CPU(cs);
612     CPUPPCState *env = &cpu->env;
613     struct kvm_one_reg reg;
614     int i;
615     int ret;
616 
617     if (env->insns_flags & PPC_FLOAT) {
618         uint64_t fpscr = env->fpscr;
619         bool vsx = !!(env->insns_flags2 & PPC2_VSX);
620 
621         reg.id = KVM_REG_PPC_FPSCR;
622         reg.addr = (uintptr_t)&fpscr;
623         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
624         if (ret < 0) {
625             trace_kvm_failed_fpscr_set(strerror(errno));
626             return ret;
627         }
628 
629         for (i = 0; i < 32; i++) {
630             uint64_t vsr[2];
631             uint64_t *fpr = cpu_fpr_ptr(&cpu->env, i);
632             uint64_t *vsrl = cpu_vsrl_ptr(&cpu->env, i);
633 
634 #if HOST_BIG_ENDIAN
635             vsr[0] = float64_val(*fpr);
636             vsr[1] = *vsrl;
637 #else
638             vsr[0] = *vsrl;
639             vsr[1] = float64_val(*fpr);
640 #endif
641             reg.addr = (uintptr_t) &vsr;
642             reg.id = vsx ? KVM_REG_PPC_VSR(i) : KVM_REG_PPC_FPR(i);
643 
644             ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
645             if (ret < 0) {
646                 trace_kvm_failed_fp_set(vsx ? "VSR" : "FPR", i,
647                                         strerror(errno));
648                 return ret;
649             }
650         }
651     }
652 
653     if (env->insns_flags & PPC_ALTIVEC) {
654         reg.id = KVM_REG_PPC_VSCR;
655         reg.addr = (uintptr_t)&env->vscr;
656         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
657         if (ret < 0) {
658             trace_kvm_failed_vscr_set(strerror(errno));
659             return ret;
660         }
661 
662         for (i = 0; i < 32; i++) {
663             reg.id = KVM_REG_PPC_VR(i);
664             reg.addr = (uintptr_t)cpu_avr_ptr(env, i);
665             ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
666             if (ret < 0) {
667                 trace_kvm_failed_vr_set(i, strerror(errno));
668                 return ret;
669             }
670         }
671     }
672 
673     return 0;
674 }
675 
676 static int kvm_get_fp(CPUState *cs)
677 {
678     PowerPCCPU *cpu = POWERPC_CPU(cs);
679     CPUPPCState *env = &cpu->env;
680     struct kvm_one_reg reg;
681     int i;
682     int ret;
683 
684     if (env->insns_flags & PPC_FLOAT) {
685         uint64_t fpscr;
686         bool vsx = !!(env->insns_flags2 & PPC2_VSX);
687 
688         reg.id = KVM_REG_PPC_FPSCR;
689         reg.addr = (uintptr_t)&fpscr;
690         ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
691         if (ret < 0) {
692             trace_kvm_failed_fpscr_get(strerror(errno));
693             return ret;
694         } else {
695             env->fpscr = fpscr;
696         }
697 
698         for (i = 0; i < 32; i++) {
699             uint64_t vsr[2];
700             uint64_t *fpr = cpu_fpr_ptr(&cpu->env, i);
701             uint64_t *vsrl = cpu_vsrl_ptr(&cpu->env, i);
702 
703             reg.addr = (uintptr_t) &vsr;
704             reg.id = vsx ? KVM_REG_PPC_VSR(i) : KVM_REG_PPC_FPR(i);
705 
706             ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
707             if (ret < 0) {
708                 trace_kvm_failed_fp_get(vsx ? "VSR" : "FPR", i,
709                                         strerror(errno));
710                 return ret;
711             } else {
712 #if HOST_BIG_ENDIAN
713                 *fpr = vsr[0];
714                 if (vsx) {
715                     *vsrl = vsr[1];
716                 }
717 #else
718                 *fpr = vsr[1];
719                 if (vsx) {
720                     *vsrl = vsr[0];
721                 }
722 #endif
723             }
724         }
725     }
726 
727     if (env->insns_flags & PPC_ALTIVEC) {
728         reg.id = KVM_REG_PPC_VSCR;
729         reg.addr = (uintptr_t)&env->vscr;
730         ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
731         if (ret < 0) {
732             trace_kvm_failed_vscr_get(strerror(errno));
733             return ret;
734         }
735 
736         for (i = 0; i < 32; i++) {
737             reg.id = KVM_REG_PPC_VR(i);
738             reg.addr = (uintptr_t)cpu_avr_ptr(env, i);
739             ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
740             if (ret < 0) {
741                 trace_kvm_failed_vr_get(i, strerror(errno));
742                 return ret;
743             }
744         }
745     }
746 
747     return 0;
748 }
749 
750 #if defined(TARGET_PPC64)
751 static int kvm_get_vpa(CPUState *cs)
752 {
753     PowerPCCPU *cpu = POWERPC_CPU(cs);
754     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
755     struct kvm_one_reg reg;
756     int ret;
757 
758     reg.id = KVM_REG_PPC_VPA_ADDR;
759     reg.addr = (uintptr_t)&spapr_cpu->vpa_addr;
760     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
761     if (ret < 0) {
762         trace_kvm_failed_vpa_addr_get(strerror(errno));
763         return ret;
764     }
765 
766     assert((uintptr_t)&spapr_cpu->slb_shadow_size
767            == ((uintptr_t)&spapr_cpu->slb_shadow_addr + 8));
768     reg.id = KVM_REG_PPC_VPA_SLB;
769     reg.addr = (uintptr_t)&spapr_cpu->slb_shadow_addr;
770     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
771     if (ret < 0) {
772         trace_kvm_failed_slb_get(strerror(errno));
773         return ret;
774     }
775 
776     assert((uintptr_t)&spapr_cpu->dtl_size
777            == ((uintptr_t)&spapr_cpu->dtl_addr + 8));
778     reg.id = KVM_REG_PPC_VPA_DTL;
779     reg.addr = (uintptr_t)&spapr_cpu->dtl_addr;
780     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
781     if (ret < 0) {
782         trace_kvm_failed_dtl_get(strerror(errno));
783         return ret;
784     }
785 
786     return 0;
787 }
788 
789 static int kvm_put_vpa(CPUState *cs)
790 {
791     PowerPCCPU *cpu = POWERPC_CPU(cs);
792     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
793     struct kvm_one_reg reg;
794     int ret;
795 
796     /*
797      * SLB shadow or DTL can't be registered unless a master VPA is
798      * registered.  That means when restoring state, if a VPA *is*
799      * registered, we need to set that up first.  If not, we need to
800      * deregister the others before deregistering the master VPA
801      */
802     assert(spapr_cpu->vpa_addr
803            || !(spapr_cpu->slb_shadow_addr || spapr_cpu->dtl_addr));
804 
805     if (spapr_cpu->vpa_addr) {
806         reg.id = KVM_REG_PPC_VPA_ADDR;
807         reg.addr = (uintptr_t)&spapr_cpu->vpa_addr;
808         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
809         if (ret < 0) {
810             trace_kvm_failed_vpa_addr_set(strerror(errno));
811             return ret;
812         }
813     }
814 
815     assert((uintptr_t)&spapr_cpu->slb_shadow_size
816            == ((uintptr_t)&spapr_cpu->slb_shadow_addr + 8));
817     reg.id = KVM_REG_PPC_VPA_SLB;
818     reg.addr = (uintptr_t)&spapr_cpu->slb_shadow_addr;
819     ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
820     if (ret < 0) {
821         trace_kvm_failed_slb_set(strerror(errno));
822         return ret;
823     }
824 
825     assert((uintptr_t)&spapr_cpu->dtl_size
826            == ((uintptr_t)&spapr_cpu->dtl_addr + 8));
827     reg.id = KVM_REG_PPC_VPA_DTL;
828     reg.addr = (uintptr_t)&spapr_cpu->dtl_addr;
829     ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
830     if (ret < 0) {
831         trace_kvm_failed_dtl_set(strerror(errno));
832         return ret;
833     }
834 
835     if (!spapr_cpu->vpa_addr) {
836         reg.id = KVM_REG_PPC_VPA_ADDR;
837         reg.addr = (uintptr_t)&spapr_cpu->vpa_addr;
838         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
839         if (ret < 0) {
840             trace_kvm_failed_null_vpa_addr_set(strerror(errno));
841             return ret;
842         }
843     }
844 
845     return 0;
846 }
847 #endif /* TARGET_PPC64 */
848 
849 int kvmppc_put_books_sregs(PowerPCCPU *cpu)
850 {
851     CPUPPCState *env = &cpu->env;
852     struct kvm_sregs sregs;
853     int i;
854 
855     sregs.pvr = env->spr[SPR_PVR];
856 
857     if (cpu->vhyp) {
858         PPCVirtualHypervisorClass *vhc =
859             PPC_VIRTUAL_HYPERVISOR_GET_CLASS(cpu->vhyp);
860         sregs.u.s.sdr1 = vhc->encode_hpt_for_kvm_pr(cpu->vhyp);
861     } else {
862         sregs.u.s.sdr1 = env->spr[SPR_SDR1];
863     }
864 
865     /* Sync SLB */
866 #ifdef TARGET_PPC64
867     for (i = 0; i < ARRAY_SIZE(env->slb); i++) {
868         sregs.u.s.ppc64.slb[i].slbe = env->slb[i].esid;
869         if (env->slb[i].esid & SLB_ESID_V) {
870             sregs.u.s.ppc64.slb[i].slbe |= i;
871         }
872         sregs.u.s.ppc64.slb[i].slbv = env->slb[i].vsid;
873     }
874 #endif
875 
876     /* Sync SRs */
877     for (i = 0; i < 16; i++) {
878         sregs.u.s.ppc32.sr[i] = env->sr[i];
879     }
880 
881     /* Sync BATs */
882     for (i = 0; i < 8; i++) {
883         /* Beware. We have to swap upper and lower bits here */
884         sregs.u.s.ppc32.dbat[i] = ((uint64_t)env->DBAT[0][i] << 32)
885             | env->DBAT[1][i];
886         sregs.u.s.ppc32.ibat[i] = ((uint64_t)env->IBAT[0][i] << 32)
887             | env->IBAT[1][i];
888     }
889 
890     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_SREGS, &sregs);
891 }
892 
893 int kvm_arch_put_registers(CPUState *cs, int level)
894 {
895     PowerPCCPU *cpu = POWERPC_CPU(cs);
896     CPUPPCState *env = &cpu->env;
897     struct kvm_regs regs;
898     int ret;
899     int i;
900 
901     ret = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
902     if (ret < 0) {
903         return ret;
904     }
905 
906     regs.ctr = env->ctr;
907     regs.lr  = env->lr;
908     regs.xer = cpu_read_xer(env);
909     regs.msr = env->msr;
910     regs.pc = env->nip;
911 
912     regs.srr0 = env->spr[SPR_SRR0];
913     regs.srr1 = env->spr[SPR_SRR1];
914 
915     regs.sprg0 = env->spr[SPR_SPRG0];
916     regs.sprg1 = env->spr[SPR_SPRG1];
917     regs.sprg2 = env->spr[SPR_SPRG2];
918     regs.sprg3 = env->spr[SPR_SPRG3];
919     regs.sprg4 = env->spr[SPR_SPRG4];
920     regs.sprg5 = env->spr[SPR_SPRG5];
921     regs.sprg6 = env->spr[SPR_SPRG6];
922     regs.sprg7 = env->spr[SPR_SPRG7];
923 
924     regs.pid = env->spr[SPR_BOOKE_PID];
925 
926     for (i = 0; i < 32; i++) {
927         regs.gpr[i] = env->gpr[i];
928     }
929 
930     regs.cr = 0;
931     for (i = 0; i < 8; i++) {
932         regs.cr |= (env->crf[i] & 15) << (4 * (7 - i));
933     }
934 
935     ret = kvm_vcpu_ioctl(cs, KVM_SET_REGS, &regs);
936     if (ret < 0) {
937         return ret;
938     }
939 
940     kvm_put_fp(cs);
941 
942     if (env->tlb_dirty) {
943         kvm_sw_tlb_put(cpu);
944         env->tlb_dirty = false;
945     }
946 
947     if (cap_segstate && (level >= KVM_PUT_RESET_STATE)) {
948         ret = kvmppc_put_books_sregs(cpu);
949         if (ret < 0) {
950             return ret;
951         }
952     }
953 
954     if (cap_hior && (level >= KVM_PUT_RESET_STATE)) {
955         kvm_put_one_spr(cs, KVM_REG_PPC_HIOR, SPR_HIOR);
956     }
957 
958     if (cap_one_reg) {
959         int i;
960 
961         /*
962          * We deliberately ignore errors here, for kernels which have
963          * the ONE_REG calls, but don't support the specific
964          * registers, there's a reasonable chance things will still
965          * work, at least until we try to migrate.
966          */
967         for (i = 0; i < 1024; i++) {
968             uint64_t id = env->spr_cb[i].one_reg_id;
969 
970             if (id != 0) {
971                 kvm_put_one_spr(cs, id, i);
972             }
973         }
974 
975 #ifdef TARGET_PPC64
976         if (msr_ts) {
977             for (i = 0; i < ARRAY_SIZE(env->tm_gpr); i++) {
978                 kvm_set_one_reg(cs, KVM_REG_PPC_TM_GPR(i), &env->tm_gpr[i]);
979             }
980             for (i = 0; i < ARRAY_SIZE(env->tm_vsr); i++) {
981                 kvm_set_one_reg(cs, KVM_REG_PPC_TM_VSR(i), &env->tm_vsr[i]);
982             }
983             kvm_set_one_reg(cs, KVM_REG_PPC_TM_CR, &env->tm_cr);
984             kvm_set_one_reg(cs, KVM_REG_PPC_TM_LR, &env->tm_lr);
985             kvm_set_one_reg(cs, KVM_REG_PPC_TM_CTR, &env->tm_ctr);
986             kvm_set_one_reg(cs, KVM_REG_PPC_TM_FPSCR, &env->tm_fpscr);
987             kvm_set_one_reg(cs, KVM_REG_PPC_TM_AMR, &env->tm_amr);
988             kvm_set_one_reg(cs, KVM_REG_PPC_TM_PPR, &env->tm_ppr);
989             kvm_set_one_reg(cs, KVM_REG_PPC_TM_VRSAVE, &env->tm_vrsave);
990             kvm_set_one_reg(cs, KVM_REG_PPC_TM_VSCR, &env->tm_vscr);
991             kvm_set_one_reg(cs, KVM_REG_PPC_TM_DSCR, &env->tm_dscr);
992             kvm_set_one_reg(cs, KVM_REG_PPC_TM_TAR, &env->tm_tar);
993         }
994 
995         if (cap_papr) {
996             if (kvm_put_vpa(cs) < 0) {
997                 trace_kvm_failed_put_vpa();
998             }
999         }
1000 
1001         kvm_set_one_reg(cs, KVM_REG_PPC_TB_OFFSET, &env->tb_env->tb_offset);
1002 
1003         if (level > KVM_PUT_RUNTIME_STATE) {
1004             kvm_put_one_spr(cs, KVM_REG_PPC_DPDES, SPR_DPDES);
1005         }
1006 #endif /* TARGET_PPC64 */
1007     }
1008 
1009     return ret;
1010 }
1011 
1012 static void kvm_sync_excp(CPUPPCState *env, int vector, int ivor)
1013 {
1014      env->excp_vectors[vector] = env->spr[ivor] + env->spr[SPR_BOOKE_IVPR];
1015 }
1016 
1017 static int kvmppc_get_booke_sregs(PowerPCCPU *cpu)
1018 {
1019     CPUPPCState *env = &cpu->env;
1020     struct kvm_sregs sregs;
1021     int ret;
1022 
1023     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_SREGS, &sregs);
1024     if (ret < 0) {
1025         return ret;
1026     }
1027 
1028     if (sregs.u.e.features & KVM_SREGS_E_BASE) {
1029         env->spr[SPR_BOOKE_CSRR0] = sregs.u.e.csrr0;
1030         env->spr[SPR_BOOKE_CSRR1] = sregs.u.e.csrr1;
1031         env->spr[SPR_BOOKE_ESR] = sregs.u.e.esr;
1032         env->spr[SPR_BOOKE_DEAR] = sregs.u.e.dear;
1033         env->spr[SPR_BOOKE_MCSR] = sregs.u.e.mcsr;
1034         env->spr[SPR_BOOKE_TSR] = sregs.u.e.tsr;
1035         env->spr[SPR_BOOKE_TCR] = sregs.u.e.tcr;
1036         env->spr[SPR_DECR] = sregs.u.e.dec;
1037         env->spr[SPR_TBL] = sregs.u.e.tb & 0xffffffff;
1038         env->spr[SPR_TBU] = sregs.u.e.tb >> 32;
1039         env->spr[SPR_VRSAVE] = sregs.u.e.vrsave;
1040     }
1041 
1042     if (sregs.u.e.features & KVM_SREGS_E_ARCH206) {
1043         env->spr[SPR_BOOKE_PIR] = sregs.u.e.pir;
1044         env->spr[SPR_BOOKE_MCSRR0] = sregs.u.e.mcsrr0;
1045         env->spr[SPR_BOOKE_MCSRR1] = sregs.u.e.mcsrr1;
1046         env->spr[SPR_BOOKE_DECAR] = sregs.u.e.decar;
1047         env->spr[SPR_BOOKE_IVPR] = sregs.u.e.ivpr;
1048     }
1049 
1050     if (sregs.u.e.features & KVM_SREGS_E_64) {
1051         env->spr[SPR_BOOKE_EPCR] = sregs.u.e.epcr;
1052     }
1053 
1054     if (sregs.u.e.features & KVM_SREGS_E_SPRG8) {
1055         env->spr[SPR_BOOKE_SPRG8] = sregs.u.e.sprg8;
1056     }
1057 
1058     if (sregs.u.e.features & KVM_SREGS_E_IVOR) {
1059         env->spr[SPR_BOOKE_IVOR0] = sregs.u.e.ivor_low[0];
1060         kvm_sync_excp(env, POWERPC_EXCP_CRITICAL,  SPR_BOOKE_IVOR0);
1061         env->spr[SPR_BOOKE_IVOR1] = sregs.u.e.ivor_low[1];
1062         kvm_sync_excp(env, POWERPC_EXCP_MCHECK,  SPR_BOOKE_IVOR1);
1063         env->spr[SPR_BOOKE_IVOR2] = sregs.u.e.ivor_low[2];
1064         kvm_sync_excp(env, POWERPC_EXCP_DSI,  SPR_BOOKE_IVOR2);
1065         env->spr[SPR_BOOKE_IVOR3] = sregs.u.e.ivor_low[3];
1066         kvm_sync_excp(env, POWERPC_EXCP_ISI,  SPR_BOOKE_IVOR3);
1067         env->spr[SPR_BOOKE_IVOR4] = sregs.u.e.ivor_low[4];
1068         kvm_sync_excp(env, POWERPC_EXCP_EXTERNAL,  SPR_BOOKE_IVOR4);
1069         env->spr[SPR_BOOKE_IVOR5] = sregs.u.e.ivor_low[5];
1070         kvm_sync_excp(env, POWERPC_EXCP_ALIGN,  SPR_BOOKE_IVOR5);
1071         env->spr[SPR_BOOKE_IVOR6] = sregs.u.e.ivor_low[6];
1072         kvm_sync_excp(env, POWERPC_EXCP_PROGRAM,  SPR_BOOKE_IVOR6);
1073         env->spr[SPR_BOOKE_IVOR7] = sregs.u.e.ivor_low[7];
1074         kvm_sync_excp(env, POWERPC_EXCP_FPU,  SPR_BOOKE_IVOR7);
1075         env->spr[SPR_BOOKE_IVOR8] = sregs.u.e.ivor_low[8];
1076         kvm_sync_excp(env, POWERPC_EXCP_SYSCALL,  SPR_BOOKE_IVOR8);
1077         env->spr[SPR_BOOKE_IVOR9] = sregs.u.e.ivor_low[9];
1078         kvm_sync_excp(env, POWERPC_EXCP_APU,  SPR_BOOKE_IVOR9);
1079         env->spr[SPR_BOOKE_IVOR10] = sregs.u.e.ivor_low[10];
1080         kvm_sync_excp(env, POWERPC_EXCP_DECR,  SPR_BOOKE_IVOR10);
1081         env->spr[SPR_BOOKE_IVOR11] = sregs.u.e.ivor_low[11];
1082         kvm_sync_excp(env, POWERPC_EXCP_FIT,  SPR_BOOKE_IVOR11);
1083         env->spr[SPR_BOOKE_IVOR12] = sregs.u.e.ivor_low[12];
1084         kvm_sync_excp(env, POWERPC_EXCP_WDT,  SPR_BOOKE_IVOR12);
1085         env->spr[SPR_BOOKE_IVOR13] = sregs.u.e.ivor_low[13];
1086         kvm_sync_excp(env, POWERPC_EXCP_DTLB,  SPR_BOOKE_IVOR13);
1087         env->spr[SPR_BOOKE_IVOR14] = sregs.u.e.ivor_low[14];
1088         kvm_sync_excp(env, POWERPC_EXCP_ITLB,  SPR_BOOKE_IVOR14);
1089         env->spr[SPR_BOOKE_IVOR15] = sregs.u.e.ivor_low[15];
1090         kvm_sync_excp(env, POWERPC_EXCP_DEBUG,  SPR_BOOKE_IVOR15);
1091 
1092         if (sregs.u.e.features & KVM_SREGS_E_SPE) {
1093             env->spr[SPR_BOOKE_IVOR32] = sregs.u.e.ivor_high[0];
1094             kvm_sync_excp(env, POWERPC_EXCP_SPEU,  SPR_BOOKE_IVOR32);
1095             env->spr[SPR_BOOKE_IVOR33] = sregs.u.e.ivor_high[1];
1096             kvm_sync_excp(env, POWERPC_EXCP_EFPDI,  SPR_BOOKE_IVOR33);
1097             env->spr[SPR_BOOKE_IVOR34] = sregs.u.e.ivor_high[2];
1098             kvm_sync_excp(env, POWERPC_EXCP_EFPRI,  SPR_BOOKE_IVOR34);
1099         }
1100 
1101         if (sregs.u.e.features & KVM_SREGS_E_PM) {
1102             env->spr[SPR_BOOKE_IVOR35] = sregs.u.e.ivor_high[3];
1103             kvm_sync_excp(env, POWERPC_EXCP_EPERFM,  SPR_BOOKE_IVOR35);
1104         }
1105 
1106         if (sregs.u.e.features & KVM_SREGS_E_PC) {
1107             env->spr[SPR_BOOKE_IVOR36] = sregs.u.e.ivor_high[4];
1108             kvm_sync_excp(env, POWERPC_EXCP_DOORI,  SPR_BOOKE_IVOR36);
1109             env->spr[SPR_BOOKE_IVOR37] = sregs.u.e.ivor_high[5];
1110             kvm_sync_excp(env, POWERPC_EXCP_DOORCI, SPR_BOOKE_IVOR37);
1111         }
1112     }
1113 
1114     if (sregs.u.e.features & KVM_SREGS_E_ARCH206_MMU) {
1115         env->spr[SPR_BOOKE_MAS0] = sregs.u.e.mas0;
1116         env->spr[SPR_BOOKE_MAS1] = sregs.u.e.mas1;
1117         env->spr[SPR_BOOKE_MAS2] = sregs.u.e.mas2;
1118         env->spr[SPR_BOOKE_MAS3] = sregs.u.e.mas7_3 & 0xffffffff;
1119         env->spr[SPR_BOOKE_MAS4] = sregs.u.e.mas4;
1120         env->spr[SPR_BOOKE_MAS6] = sregs.u.e.mas6;
1121         env->spr[SPR_BOOKE_MAS7] = sregs.u.e.mas7_3 >> 32;
1122         env->spr[SPR_MMUCFG] = sregs.u.e.mmucfg;
1123         env->spr[SPR_BOOKE_TLB0CFG] = sregs.u.e.tlbcfg[0];
1124         env->spr[SPR_BOOKE_TLB1CFG] = sregs.u.e.tlbcfg[1];
1125     }
1126 
1127     if (sregs.u.e.features & KVM_SREGS_EXP) {
1128         env->spr[SPR_BOOKE_EPR] = sregs.u.e.epr;
1129     }
1130 
1131     if (sregs.u.e.features & KVM_SREGS_E_PD) {
1132         env->spr[SPR_BOOKE_EPLC] = sregs.u.e.eplc;
1133         env->spr[SPR_BOOKE_EPSC] = sregs.u.e.epsc;
1134     }
1135 
1136     if (sregs.u.e.impl_id == KVM_SREGS_E_IMPL_FSL) {
1137         env->spr[SPR_E500_SVR] = sregs.u.e.impl.fsl.svr;
1138         env->spr[SPR_Exxx_MCAR] = sregs.u.e.impl.fsl.mcar;
1139         env->spr[SPR_HID0] = sregs.u.e.impl.fsl.hid0;
1140 
1141         if (sregs.u.e.impl.fsl.features & KVM_SREGS_E_FSL_PIDn) {
1142             env->spr[SPR_BOOKE_PID1] = sregs.u.e.impl.fsl.pid1;
1143             env->spr[SPR_BOOKE_PID2] = sregs.u.e.impl.fsl.pid2;
1144         }
1145     }
1146 
1147     return 0;
1148 }
1149 
1150 static int kvmppc_get_books_sregs(PowerPCCPU *cpu)
1151 {
1152     CPUPPCState *env = &cpu->env;
1153     struct kvm_sregs sregs;
1154     int ret;
1155     int i;
1156 
1157     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_SREGS, &sregs);
1158     if (ret < 0) {
1159         return ret;
1160     }
1161 
1162     if (!cpu->vhyp) {
1163         ppc_store_sdr1(env, sregs.u.s.sdr1);
1164     }
1165 
1166     /* Sync SLB */
1167 #ifdef TARGET_PPC64
1168     /*
1169      * The packed SLB array we get from KVM_GET_SREGS only contains
1170      * information about valid entries. So we flush our internal copy
1171      * to get rid of stale ones, then put all valid SLB entries back
1172      * in.
1173      */
1174     memset(env->slb, 0, sizeof(env->slb));
1175     for (i = 0; i < ARRAY_SIZE(env->slb); i++) {
1176         target_ulong rb = sregs.u.s.ppc64.slb[i].slbe;
1177         target_ulong rs = sregs.u.s.ppc64.slb[i].slbv;
1178         /*
1179          * Only restore valid entries
1180          */
1181         if (rb & SLB_ESID_V) {
1182             ppc_store_slb(cpu, rb & 0xfff, rb & ~0xfffULL, rs);
1183         }
1184     }
1185 #endif
1186 
1187     /* Sync SRs */
1188     for (i = 0; i < 16; i++) {
1189         env->sr[i] = sregs.u.s.ppc32.sr[i];
1190     }
1191 
1192     /* Sync BATs */
1193     for (i = 0; i < 8; i++) {
1194         env->DBAT[0][i] = sregs.u.s.ppc32.dbat[i] & 0xffffffff;
1195         env->DBAT[1][i] = sregs.u.s.ppc32.dbat[i] >> 32;
1196         env->IBAT[0][i] = sregs.u.s.ppc32.ibat[i] & 0xffffffff;
1197         env->IBAT[1][i] = sregs.u.s.ppc32.ibat[i] >> 32;
1198     }
1199 
1200     return 0;
1201 }
1202 
1203 int kvm_arch_get_registers(CPUState *cs)
1204 {
1205     PowerPCCPU *cpu = POWERPC_CPU(cs);
1206     CPUPPCState *env = &cpu->env;
1207     struct kvm_regs regs;
1208     uint32_t cr;
1209     int i, ret;
1210 
1211     ret = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
1212     if (ret < 0) {
1213         return ret;
1214     }
1215 
1216     cr = regs.cr;
1217     for (i = 7; i >= 0; i--) {
1218         env->crf[i] = cr & 15;
1219         cr >>= 4;
1220     }
1221 
1222     env->ctr = regs.ctr;
1223     env->lr = regs.lr;
1224     cpu_write_xer(env, regs.xer);
1225     env->msr = regs.msr;
1226     env->nip = regs.pc;
1227 
1228     env->spr[SPR_SRR0] = regs.srr0;
1229     env->spr[SPR_SRR1] = regs.srr1;
1230 
1231     env->spr[SPR_SPRG0] = regs.sprg0;
1232     env->spr[SPR_SPRG1] = regs.sprg1;
1233     env->spr[SPR_SPRG2] = regs.sprg2;
1234     env->spr[SPR_SPRG3] = regs.sprg3;
1235     env->spr[SPR_SPRG4] = regs.sprg4;
1236     env->spr[SPR_SPRG5] = regs.sprg5;
1237     env->spr[SPR_SPRG6] = regs.sprg6;
1238     env->spr[SPR_SPRG7] = regs.sprg7;
1239 
1240     env->spr[SPR_BOOKE_PID] = regs.pid;
1241 
1242     for (i = 0; i < 32; i++) {
1243         env->gpr[i] = regs.gpr[i];
1244     }
1245 
1246     kvm_get_fp(cs);
1247 
1248     if (cap_booke_sregs) {
1249         ret = kvmppc_get_booke_sregs(cpu);
1250         if (ret < 0) {
1251             return ret;
1252         }
1253     }
1254 
1255     if (cap_segstate) {
1256         ret = kvmppc_get_books_sregs(cpu);
1257         if (ret < 0) {
1258             return ret;
1259         }
1260     }
1261 
1262     if (cap_hior) {
1263         kvm_get_one_spr(cs, KVM_REG_PPC_HIOR, SPR_HIOR);
1264     }
1265 
1266     if (cap_one_reg) {
1267         int i;
1268 
1269         /*
1270          * We deliberately ignore errors here, for kernels which have
1271          * the ONE_REG calls, but don't support the specific
1272          * registers, there's a reasonable chance things will still
1273          * work, at least until we try to migrate.
1274          */
1275         for (i = 0; i < 1024; i++) {
1276             uint64_t id = env->spr_cb[i].one_reg_id;
1277 
1278             if (id != 0) {
1279                 kvm_get_one_spr(cs, id, i);
1280             }
1281         }
1282 
1283 #ifdef TARGET_PPC64
1284         if (msr_ts) {
1285             for (i = 0; i < ARRAY_SIZE(env->tm_gpr); i++) {
1286                 kvm_get_one_reg(cs, KVM_REG_PPC_TM_GPR(i), &env->tm_gpr[i]);
1287             }
1288             for (i = 0; i < ARRAY_SIZE(env->tm_vsr); i++) {
1289                 kvm_get_one_reg(cs, KVM_REG_PPC_TM_VSR(i), &env->tm_vsr[i]);
1290             }
1291             kvm_get_one_reg(cs, KVM_REG_PPC_TM_CR, &env->tm_cr);
1292             kvm_get_one_reg(cs, KVM_REG_PPC_TM_LR, &env->tm_lr);
1293             kvm_get_one_reg(cs, KVM_REG_PPC_TM_CTR, &env->tm_ctr);
1294             kvm_get_one_reg(cs, KVM_REG_PPC_TM_FPSCR, &env->tm_fpscr);
1295             kvm_get_one_reg(cs, KVM_REG_PPC_TM_AMR, &env->tm_amr);
1296             kvm_get_one_reg(cs, KVM_REG_PPC_TM_PPR, &env->tm_ppr);
1297             kvm_get_one_reg(cs, KVM_REG_PPC_TM_VRSAVE, &env->tm_vrsave);
1298             kvm_get_one_reg(cs, KVM_REG_PPC_TM_VSCR, &env->tm_vscr);
1299             kvm_get_one_reg(cs, KVM_REG_PPC_TM_DSCR, &env->tm_dscr);
1300             kvm_get_one_reg(cs, KVM_REG_PPC_TM_TAR, &env->tm_tar);
1301         }
1302 
1303         if (cap_papr) {
1304             if (kvm_get_vpa(cs) < 0) {
1305                 trace_kvm_failed_get_vpa();
1306             }
1307         }
1308 
1309         kvm_get_one_reg(cs, KVM_REG_PPC_TB_OFFSET, &env->tb_env->tb_offset);
1310         kvm_get_one_spr(cs, KVM_REG_PPC_DPDES, SPR_DPDES);
1311 #endif
1312     }
1313 
1314     return 0;
1315 }
1316 
1317 int kvmppc_set_interrupt(PowerPCCPU *cpu, int irq, int level)
1318 {
1319     unsigned virq = level ? KVM_INTERRUPT_SET_LEVEL : KVM_INTERRUPT_UNSET;
1320 
1321     if (irq != PPC_INTERRUPT_EXT) {
1322         return 0;
1323     }
1324 
1325     if (!kvm_enabled() || !cap_interrupt_unset) {
1326         return 0;
1327     }
1328 
1329     kvm_vcpu_ioctl(CPU(cpu), KVM_INTERRUPT, &virq);
1330 
1331     return 0;
1332 }
1333 
1334 void kvm_arch_pre_run(CPUState *cs, struct kvm_run *run)
1335 {
1336     return;
1337 }
1338 
1339 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
1340 {
1341     return MEMTXATTRS_UNSPECIFIED;
1342 }
1343 
1344 int kvm_arch_process_async_events(CPUState *cs)
1345 {
1346     return cs->halted;
1347 }
1348 
1349 static int kvmppc_handle_halt(PowerPCCPU *cpu)
1350 {
1351     CPUState *cs = CPU(cpu);
1352     CPUPPCState *env = &cpu->env;
1353 
1354     if (!(cs->interrupt_request & CPU_INTERRUPT_HARD) && (msr_ee)) {
1355         cs->halted = 1;
1356         cs->exception_index = EXCP_HLT;
1357     }
1358 
1359     return 0;
1360 }
1361 
1362 /* map dcr access to existing qemu dcr emulation */
1363 static int kvmppc_handle_dcr_read(CPUPPCState *env,
1364                                   uint32_t dcrn, uint32_t *data)
1365 {
1366     if (ppc_dcr_read(env->dcr_env, dcrn, data) < 0) {
1367         fprintf(stderr, "Read to unhandled DCR (0x%x)\n", dcrn);
1368     }
1369 
1370     return 0;
1371 }
1372 
1373 static int kvmppc_handle_dcr_write(CPUPPCState *env,
1374                                    uint32_t dcrn, uint32_t data)
1375 {
1376     if (ppc_dcr_write(env->dcr_env, dcrn, data) < 0) {
1377         fprintf(stderr, "Write to unhandled DCR (0x%x)\n", dcrn);
1378     }
1379 
1380     return 0;
1381 }
1382 
1383 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
1384 {
1385     /* Mixed endian case is not handled */
1386     uint32_t sc = debug_inst_opcode;
1387 
1388     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
1389                             sizeof(sc), 0) ||
1390         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&sc, sizeof(sc), 1)) {
1391         return -EINVAL;
1392     }
1393 
1394     return 0;
1395 }
1396 
1397 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
1398 {
1399     uint32_t sc;
1400 
1401     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&sc, sizeof(sc), 0) ||
1402         sc != debug_inst_opcode ||
1403         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
1404                             sizeof(sc), 1)) {
1405         return -EINVAL;
1406     }
1407 
1408     return 0;
1409 }
1410 
1411 static int find_hw_breakpoint(target_ulong addr, int type)
1412 {
1413     int n;
1414 
1415     assert((nb_hw_breakpoint + nb_hw_watchpoint)
1416            <= ARRAY_SIZE(hw_debug_points));
1417 
1418     for (n = 0; n < nb_hw_breakpoint + nb_hw_watchpoint; n++) {
1419         if (hw_debug_points[n].addr == addr &&
1420              hw_debug_points[n].type == type) {
1421             return n;
1422         }
1423     }
1424 
1425     return -1;
1426 }
1427 
1428 static int find_hw_watchpoint(target_ulong addr, int *flag)
1429 {
1430     int n;
1431 
1432     n = find_hw_breakpoint(addr, GDB_WATCHPOINT_ACCESS);
1433     if (n >= 0) {
1434         *flag = BP_MEM_ACCESS;
1435         return n;
1436     }
1437 
1438     n = find_hw_breakpoint(addr, GDB_WATCHPOINT_WRITE);
1439     if (n >= 0) {
1440         *flag = BP_MEM_WRITE;
1441         return n;
1442     }
1443 
1444     n = find_hw_breakpoint(addr, GDB_WATCHPOINT_READ);
1445     if (n >= 0) {
1446         *flag = BP_MEM_READ;
1447         return n;
1448     }
1449 
1450     return -1;
1451 }
1452 
1453 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
1454                                   target_ulong len, int type)
1455 {
1456     if ((nb_hw_breakpoint + nb_hw_watchpoint) >= ARRAY_SIZE(hw_debug_points)) {
1457         return -ENOBUFS;
1458     }
1459 
1460     hw_debug_points[nb_hw_breakpoint + nb_hw_watchpoint].addr = addr;
1461     hw_debug_points[nb_hw_breakpoint + nb_hw_watchpoint].type = type;
1462 
1463     switch (type) {
1464     case GDB_BREAKPOINT_HW:
1465         if (nb_hw_breakpoint >= max_hw_breakpoint) {
1466             return -ENOBUFS;
1467         }
1468 
1469         if (find_hw_breakpoint(addr, type) >= 0) {
1470             return -EEXIST;
1471         }
1472 
1473         nb_hw_breakpoint++;
1474         break;
1475 
1476     case GDB_WATCHPOINT_WRITE:
1477     case GDB_WATCHPOINT_READ:
1478     case GDB_WATCHPOINT_ACCESS:
1479         if (nb_hw_watchpoint >= max_hw_watchpoint) {
1480             return -ENOBUFS;
1481         }
1482 
1483         if (find_hw_breakpoint(addr, type) >= 0) {
1484             return -EEXIST;
1485         }
1486 
1487         nb_hw_watchpoint++;
1488         break;
1489 
1490     default:
1491         return -ENOSYS;
1492     }
1493 
1494     return 0;
1495 }
1496 
1497 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
1498                                   target_ulong len, int type)
1499 {
1500     int n;
1501 
1502     n = find_hw_breakpoint(addr, type);
1503     if (n < 0) {
1504         return -ENOENT;
1505     }
1506 
1507     switch (type) {
1508     case GDB_BREAKPOINT_HW:
1509         nb_hw_breakpoint--;
1510         break;
1511 
1512     case GDB_WATCHPOINT_WRITE:
1513     case GDB_WATCHPOINT_READ:
1514     case GDB_WATCHPOINT_ACCESS:
1515         nb_hw_watchpoint--;
1516         break;
1517 
1518     default:
1519         return -ENOSYS;
1520     }
1521     hw_debug_points[n] = hw_debug_points[nb_hw_breakpoint + nb_hw_watchpoint];
1522 
1523     return 0;
1524 }
1525 
1526 void kvm_arch_remove_all_hw_breakpoints(void)
1527 {
1528     nb_hw_breakpoint = nb_hw_watchpoint = 0;
1529 }
1530 
1531 void kvm_arch_update_guest_debug(CPUState *cs, struct kvm_guest_debug *dbg)
1532 {
1533     int n;
1534 
1535     /* Software Breakpoint updates */
1536     if (kvm_sw_breakpoints_active(cs)) {
1537         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP;
1538     }
1539 
1540     assert((nb_hw_breakpoint + nb_hw_watchpoint)
1541            <= ARRAY_SIZE(hw_debug_points));
1542     assert((nb_hw_breakpoint + nb_hw_watchpoint) <= ARRAY_SIZE(dbg->arch.bp));
1543 
1544     if (nb_hw_breakpoint + nb_hw_watchpoint > 0) {
1545         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
1546         memset(dbg->arch.bp, 0, sizeof(dbg->arch.bp));
1547         for (n = 0; n < nb_hw_breakpoint + nb_hw_watchpoint; n++) {
1548             switch (hw_debug_points[n].type) {
1549             case GDB_BREAKPOINT_HW:
1550                 dbg->arch.bp[n].type = KVMPPC_DEBUG_BREAKPOINT;
1551                 break;
1552             case GDB_WATCHPOINT_WRITE:
1553                 dbg->arch.bp[n].type = KVMPPC_DEBUG_WATCH_WRITE;
1554                 break;
1555             case GDB_WATCHPOINT_READ:
1556                 dbg->arch.bp[n].type = KVMPPC_DEBUG_WATCH_READ;
1557                 break;
1558             case GDB_WATCHPOINT_ACCESS:
1559                 dbg->arch.bp[n].type = KVMPPC_DEBUG_WATCH_WRITE |
1560                                         KVMPPC_DEBUG_WATCH_READ;
1561                 break;
1562             default:
1563                 cpu_abort(cs, "Unsupported breakpoint type\n");
1564             }
1565             dbg->arch.bp[n].addr = hw_debug_points[n].addr;
1566         }
1567     }
1568 }
1569 
1570 static int kvm_handle_hw_breakpoint(CPUState *cs,
1571                                     struct kvm_debug_exit_arch *arch_info)
1572 {
1573     int handle = DEBUG_RETURN_GUEST;
1574     int n;
1575     int flag = 0;
1576 
1577     if (nb_hw_breakpoint + nb_hw_watchpoint > 0) {
1578         if (arch_info->status & KVMPPC_DEBUG_BREAKPOINT) {
1579             n = find_hw_breakpoint(arch_info->address, GDB_BREAKPOINT_HW);
1580             if (n >= 0) {
1581                 handle = DEBUG_RETURN_GDB;
1582             }
1583         } else if (arch_info->status & (KVMPPC_DEBUG_WATCH_READ |
1584                                         KVMPPC_DEBUG_WATCH_WRITE)) {
1585             n = find_hw_watchpoint(arch_info->address,  &flag);
1586             if (n >= 0) {
1587                 handle = DEBUG_RETURN_GDB;
1588                 cs->watchpoint_hit = &hw_watchpoint;
1589                 hw_watchpoint.vaddr = hw_debug_points[n].addr;
1590                 hw_watchpoint.flags = flag;
1591             }
1592         }
1593     }
1594     return handle;
1595 }
1596 
1597 static int kvm_handle_singlestep(void)
1598 {
1599     return DEBUG_RETURN_GDB;
1600 }
1601 
1602 static int kvm_handle_sw_breakpoint(void)
1603 {
1604     return DEBUG_RETURN_GDB;
1605 }
1606 
1607 static int kvm_handle_debug(PowerPCCPU *cpu, struct kvm_run *run)
1608 {
1609     CPUState *cs = CPU(cpu);
1610     CPUPPCState *env = &cpu->env;
1611     struct kvm_debug_exit_arch *arch_info = &run->debug.arch;
1612 
1613     if (cs->singlestep_enabled) {
1614         return kvm_handle_singlestep();
1615     }
1616 
1617     if (arch_info->status) {
1618         return kvm_handle_hw_breakpoint(cs, arch_info);
1619     }
1620 
1621     if (kvm_find_sw_breakpoint(cs, arch_info->address)) {
1622         return kvm_handle_sw_breakpoint();
1623     }
1624 
1625     /*
1626      * QEMU is not able to handle debug exception, so inject
1627      * program exception to guest;
1628      * Yes program exception NOT debug exception !!
1629      * When QEMU is using debug resources then debug exception must
1630      * be always set. To achieve this we set MSR_DE and also set
1631      * MSRP_DEP so guest cannot change MSR_DE.
1632      * When emulating debug resource for guest we want guest
1633      * to control MSR_DE (enable/disable debug interrupt on need).
1634      * Supporting both configurations are NOT possible.
1635      * So the result is that we cannot share debug resources
1636      * between QEMU and Guest on BOOKE architecture.
1637      * In the current design QEMU gets the priority over guest,
1638      * this means that if QEMU is using debug resources then guest
1639      * cannot use them;
1640      * For software breakpoint QEMU uses a privileged instruction;
1641      * So there cannot be any reason that we are here for guest
1642      * set debug exception, only possibility is guest executed a
1643      * privileged / illegal instruction and that's why we are
1644      * injecting a program interrupt.
1645      */
1646     cpu_synchronize_state(cs);
1647     /*
1648      * env->nip is PC, so increment this by 4 to use
1649      * ppc_cpu_do_interrupt(), which set srr0 = env->nip - 4.
1650      */
1651     env->nip += 4;
1652     cs->exception_index = POWERPC_EXCP_PROGRAM;
1653     env->error_code = POWERPC_EXCP_INVAL;
1654     ppc_cpu_do_interrupt(cs);
1655 
1656     return DEBUG_RETURN_GUEST;
1657 }
1658 
1659 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1660 {
1661     PowerPCCPU *cpu = POWERPC_CPU(cs);
1662     CPUPPCState *env = &cpu->env;
1663     int ret;
1664 
1665     qemu_mutex_lock_iothread();
1666 
1667     switch (run->exit_reason) {
1668     case KVM_EXIT_DCR:
1669         if (run->dcr.is_write) {
1670             trace_kvm_handle_dcr_write();
1671             ret = kvmppc_handle_dcr_write(env, run->dcr.dcrn, run->dcr.data);
1672         } else {
1673             trace_kvm_handle_dcr_read();
1674             ret = kvmppc_handle_dcr_read(env, run->dcr.dcrn, &run->dcr.data);
1675         }
1676         break;
1677     case KVM_EXIT_HLT:
1678         trace_kvm_handle_halt();
1679         ret = kvmppc_handle_halt(cpu);
1680         break;
1681 #if defined(TARGET_PPC64)
1682     case KVM_EXIT_PAPR_HCALL:
1683         trace_kvm_handle_papr_hcall(run->papr_hcall.nr);
1684         run->papr_hcall.ret = spapr_hypercall(cpu,
1685                                               run->papr_hcall.nr,
1686                                               run->papr_hcall.args);
1687         ret = 0;
1688         break;
1689 #endif
1690     case KVM_EXIT_EPR:
1691         trace_kvm_handle_epr();
1692         run->epr.epr = ldl_phys(cs->as, env->mpic_iack);
1693         ret = 0;
1694         break;
1695     case KVM_EXIT_WATCHDOG:
1696         trace_kvm_handle_watchdog_expiry();
1697         watchdog_perform_action();
1698         ret = 0;
1699         break;
1700 
1701     case KVM_EXIT_DEBUG:
1702         trace_kvm_handle_debug_exception();
1703         if (kvm_handle_debug(cpu, run)) {
1704             ret = EXCP_DEBUG;
1705             break;
1706         }
1707         /* re-enter, this exception was guest-internal */
1708         ret = 0;
1709         break;
1710 
1711 #if defined(TARGET_PPC64)
1712     case KVM_EXIT_NMI:
1713         trace_kvm_handle_nmi_exception();
1714         ret = kvm_handle_nmi(cpu, run);
1715         break;
1716 #endif
1717 
1718     default:
1719         fprintf(stderr, "KVM: unknown exit reason %d\n", run->exit_reason);
1720         ret = -1;
1721         break;
1722     }
1723 
1724     qemu_mutex_unlock_iothread();
1725     return ret;
1726 }
1727 
1728 int kvmppc_or_tsr_bits(PowerPCCPU *cpu, uint32_t tsr_bits)
1729 {
1730     CPUState *cs = CPU(cpu);
1731     uint32_t bits = tsr_bits;
1732     struct kvm_one_reg reg = {
1733         .id = KVM_REG_PPC_OR_TSR,
1734         .addr = (uintptr_t) &bits,
1735     };
1736 
1737     return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
1738 }
1739 
1740 int kvmppc_clear_tsr_bits(PowerPCCPU *cpu, uint32_t tsr_bits)
1741 {
1742 
1743     CPUState *cs = CPU(cpu);
1744     uint32_t bits = tsr_bits;
1745     struct kvm_one_reg reg = {
1746         .id = KVM_REG_PPC_CLEAR_TSR,
1747         .addr = (uintptr_t) &bits,
1748     };
1749 
1750     return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
1751 }
1752 
1753 int kvmppc_set_tcr(PowerPCCPU *cpu)
1754 {
1755     CPUState *cs = CPU(cpu);
1756     CPUPPCState *env = &cpu->env;
1757     uint32_t tcr = env->spr[SPR_BOOKE_TCR];
1758 
1759     struct kvm_one_reg reg = {
1760         .id = KVM_REG_PPC_TCR,
1761         .addr = (uintptr_t) &tcr,
1762     };
1763 
1764     return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
1765 }
1766 
1767 int kvmppc_booke_watchdog_enable(PowerPCCPU *cpu)
1768 {
1769     CPUState *cs = CPU(cpu);
1770     int ret;
1771 
1772     if (!kvm_enabled()) {
1773         return -1;
1774     }
1775 
1776     if (!cap_ppc_watchdog) {
1777         printf("warning: KVM does not support watchdog");
1778         return -1;
1779     }
1780 
1781     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_PPC_BOOKE_WATCHDOG, 0);
1782     if (ret < 0) {
1783         fprintf(stderr, "%s: couldn't enable KVM_CAP_PPC_BOOKE_WATCHDOG: %s\n",
1784                 __func__, strerror(-ret));
1785         return ret;
1786     }
1787 
1788     return ret;
1789 }
1790 
1791 static int read_cpuinfo(const char *field, char *value, int len)
1792 {
1793     FILE *f;
1794     int ret = -1;
1795     int field_len = strlen(field);
1796     char line[512];
1797 
1798     f = fopen("/proc/cpuinfo", "r");
1799     if (!f) {
1800         return -1;
1801     }
1802 
1803     do {
1804         if (!fgets(line, sizeof(line), f)) {
1805             break;
1806         }
1807         if (!strncmp(line, field, field_len)) {
1808             pstrcpy(value, len, line);
1809             ret = 0;
1810             break;
1811         }
1812     } while (*line);
1813 
1814     fclose(f);
1815 
1816     return ret;
1817 }
1818 
1819 static uint32_t kvmppc_get_tbfreq_procfs(void)
1820 {
1821     char line[512];
1822     char *ns;
1823     uint32_t tbfreq_fallback = NANOSECONDS_PER_SECOND;
1824     uint32_t tbfreq_procfs;
1825 
1826     if (read_cpuinfo("timebase", line, sizeof(line))) {
1827         return tbfreq_fallback;
1828     }
1829 
1830     ns = strchr(line, ':');
1831     if (!ns) {
1832         return tbfreq_fallback;
1833     }
1834 
1835     tbfreq_procfs = atoi(++ns);
1836 
1837     /* 0 is certainly not acceptable by the guest, return fallback value */
1838     return tbfreq_procfs ? tbfreq_procfs : tbfreq_fallback;
1839 }
1840 
1841 uint32_t kvmppc_get_tbfreq(void)
1842 {
1843     static uint32_t cached_tbfreq;
1844 
1845     if (!cached_tbfreq) {
1846         cached_tbfreq = kvmppc_get_tbfreq_procfs();
1847     }
1848 
1849     return cached_tbfreq;
1850 }
1851 
1852 bool kvmppc_get_host_serial(char **value)
1853 {
1854     return g_file_get_contents("/proc/device-tree/system-id", value, NULL,
1855                                NULL);
1856 }
1857 
1858 bool kvmppc_get_host_model(char **value)
1859 {
1860     return g_file_get_contents("/proc/device-tree/model", value, NULL, NULL);
1861 }
1862 
1863 /* Try to find a device tree node for a CPU with clock-frequency property */
1864 static int kvmppc_find_cpu_dt(char *buf, int buf_len)
1865 {
1866     struct dirent *dirp;
1867     DIR *dp;
1868 
1869     dp = opendir(PROC_DEVTREE_CPU);
1870     if (!dp) {
1871         printf("Can't open directory " PROC_DEVTREE_CPU "\n");
1872         return -1;
1873     }
1874 
1875     buf[0] = '\0';
1876     while ((dirp = readdir(dp)) != NULL) {
1877         FILE *f;
1878         snprintf(buf, buf_len, "%s%s/clock-frequency", PROC_DEVTREE_CPU,
1879                  dirp->d_name);
1880         f = fopen(buf, "r");
1881         if (f) {
1882             snprintf(buf, buf_len, "%s%s", PROC_DEVTREE_CPU, dirp->d_name);
1883             fclose(f);
1884             break;
1885         }
1886         buf[0] = '\0';
1887     }
1888     closedir(dp);
1889     if (buf[0] == '\0') {
1890         printf("Unknown host!\n");
1891         return -1;
1892     }
1893 
1894     return 0;
1895 }
1896 
1897 static uint64_t kvmppc_read_int_dt(const char *filename)
1898 {
1899     union {
1900         uint32_t v32;
1901         uint64_t v64;
1902     } u;
1903     FILE *f;
1904     int len;
1905 
1906     f = fopen(filename, "rb");
1907     if (!f) {
1908         return -1;
1909     }
1910 
1911     len = fread(&u, 1, sizeof(u), f);
1912     fclose(f);
1913     switch (len) {
1914     case 4:
1915         /* property is a 32-bit quantity */
1916         return be32_to_cpu(u.v32);
1917     case 8:
1918         return be64_to_cpu(u.v64);
1919     }
1920 
1921     return 0;
1922 }
1923 
1924 /*
1925  * Read a CPU node property from the host device tree that's a single
1926  * integer (32-bit or 64-bit).  Returns 0 if anything goes wrong
1927  * (can't find or open the property, or doesn't understand the format)
1928  */
1929 static uint64_t kvmppc_read_int_cpu_dt(const char *propname)
1930 {
1931     char buf[PATH_MAX], *tmp;
1932     uint64_t val;
1933 
1934     if (kvmppc_find_cpu_dt(buf, sizeof(buf))) {
1935         return -1;
1936     }
1937 
1938     tmp = g_strdup_printf("%s/%s", buf, propname);
1939     val = kvmppc_read_int_dt(tmp);
1940     g_free(tmp);
1941 
1942     return val;
1943 }
1944 
1945 uint64_t kvmppc_get_clockfreq(void)
1946 {
1947     return kvmppc_read_int_cpu_dt("clock-frequency");
1948 }
1949 
1950 static int kvmppc_get_dec_bits(void)
1951 {
1952     int nr_bits = kvmppc_read_int_cpu_dt("ibm,dec-bits");
1953 
1954     if (nr_bits > 0) {
1955         return nr_bits;
1956     }
1957     return 0;
1958 }
1959 
1960 static int kvmppc_get_pvinfo(CPUPPCState *env, struct kvm_ppc_pvinfo *pvinfo)
1961 {
1962     CPUState *cs = env_cpu(env);
1963 
1964     if (kvm_vm_check_extension(cs->kvm_state, KVM_CAP_PPC_GET_PVINFO) &&
1965         !kvm_vm_ioctl(cs->kvm_state, KVM_PPC_GET_PVINFO, pvinfo)) {
1966         return 0;
1967     }
1968 
1969     return 1;
1970 }
1971 
1972 int kvmppc_get_hasidle(CPUPPCState *env)
1973 {
1974     struct kvm_ppc_pvinfo pvinfo;
1975 
1976     if (!kvmppc_get_pvinfo(env, &pvinfo) &&
1977         (pvinfo.flags & KVM_PPC_PVINFO_FLAGS_EV_IDLE)) {
1978         return 1;
1979     }
1980 
1981     return 0;
1982 }
1983 
1984 int kvmppc_get_hypercall(CPUPPCState *env, uint8_t *buf, int buf_len)
1985 {
1986     uint32_t *hc = (uint32_t *)buf;
1987     struct kvm_ppc_pvinfo pvinfo;
1988 
1989     if (!kvmppc_get_pvinfo(env, &pvinfo)) {
1990         memcpy(buf, pvinfo.hcall, buf_len);
1991         return 0;
1992     }
1993 
1994     /*
1995      * Fallback to always fail hypercalls regardless of endianness:
1996      *
1997      *     tdi 0,r0,72 (becomes b .+8 in wrong endian, nop in good endian)
1998      *     li r3, -1
1999      *     b .+8       (becomes nop in wrong endian)
2000      *     bswap32(li r3, -1)
2001      */
2002 
2003     hc[0] = cpu_to_be32(0x08000048);
2004     hc[1] = cpu_to_be32(0x3860ffff);
2005     hc[2] = cpu_to_be32(0x48000008);
2006     hc[3] = cpu_to_be32(bswap32(0x3860ffff));
2007 
2008     return 1;
2009 }
2010 
2011 static inline int kvmppc_enable_hcall(KVMState *s, target_ulong hcall)
2012 {
2013     return kvm_vm_enable_cap(s, KVM_CAP_PPC_ENABLE_HCALL, 0, hcall, 1);
2014 }
2015 
2016 void kvmppc_enable_logical_ci_hcalls(void)
2017 {
2018     /*
2019      * FIXME: it would be nice if we could detect the cases where
2020      * we're using a device which requires the in kernel
2021      * implementation of these hcalls, but the kernel lacks them and
2022      * produce a warning.
2023      */
2024     kvmppc_enable_hcall(kvm_state, H_LOGICAL_CI_LOAD);
2025     kvmppc_enable_hcall(kvm_state, H_LOGICAL_CI_STORE);
2026 }
2027 
2028 void kvmppc_enable_set_mode_hcall(void)
2029 {
2030     kvmppc_enable_hcall(kvm_state, H_SET_MODE);
2031 }
2032 
2033 void kvmppc_enable_clear_ref_mod_hcalls(void)
2034 {
2035     kvmppc_enable_hcall(kvm_state, H_CLEAR_REF);
2036     kvmppc_enable_hcall(kvm_state, H_CLEAR_MOD);
2037 }
2038 
2039 void kvmppc_enable_h_page_init(void)
2040 {
2041     kvmppc_enable_hcall(kvm_state, H_PAGE_INIT);
2042 }
2043 
2044 void kvmppc_enable_h_rpt_invalidate(void)
2045 {
2046     kvmppc_enable_hcall(kvm_state, H_RPT_INVALIDATE);
2047 }
2048 
2049 void kvmppc_set_papr(PowerPCCPU *cpu)
2050 {
2051     CPUState *cs = CPU(cpu);
2052     int ret;
2053 
2054     if (!kvm_enabled()) {
2055         return;
2056     }
2057 
2058     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_PPC_PAPR, 0);
2059     if (ret) {
2060         error_report("This vCPU type or KVM version does not support PAPR");
2061         exit(1);
2062     }
2063 
2064     /*
2065      * Update the capability flag so we sync the right information
2066      * with kvm
2067      */
2068     cap_papr = 1;
2069 }
2070 
2071 int kvmppc_set_compat(PowerPCCPU *cpu, uint32_t compat_pvr)
2072 {
2073     return kvm_set_one_reg(CPU(cpu), KVM_REG_PPC_ARCH_COMPAT, &compat_pvr);
2074 }
2075 
2076 void kvmppc_set_mpic_proxy(PowerPCCPU *cpu, int mpic_proxy)
2077 {
2078     CPUState *cs = CPU(cpu);
2079     int ret;
2080 
2081     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_PPC_EPR, 0, mpic_proxy);
2082     if (ret && mpic_proxy) {
2083         error_report("This KVM version does not support EPR");
2084         exit(1);
2085     }
2086 }
2087 
2088 bool kvmppc_get_fwnmi(void)
2089 {
2090     return cap_fwnmi;
2091 }
2092 
2093 int kvmppc_set_fwnmi(PowerPCCPU *cpu)
2094 {
2095     CPUState *cs = CPU(cpu);
2096 
2097     return kvm_vcpu_enable_cap(cs, KVM_CAP_PPC_FWNMI, 0);
2098 }
2099 
2100 int kvmppc_smt_threads(void)
2101 {
2102     return cap_ppc_smt ? cap_ppc_smt : 1;
2103 }
2104 
2105 int kvmppc_set_smt_threads(int smt)
2106 {
2107     int ret;
2108 
2109     ret = kvm_vm_enable_cap(kvm_state, KVM_CAP_PPC_SMT, 0, smt, 0);
2110     if (!ret) {
2111         cap_ppc_smt = smt;
2112     }
2113     return ret;
2114 }
2115 
2116 void kvmppc_error_append_smt_possible_hint(Error *const *errp)
2117 {
2118     int i;
2119     GString *g;
2120     char *s;
2121 
2122     assert(kvm_enabled());
2123     if (cap_ppc_smt_possible) {
2124         g = g_string_new("Available VSMT modes:");
2125         for (i = 63; i >= 0; i--) {
2126             if ((1UL << i) & cap_ppc_smt_possible) {
2127                 g_string_append_printf(g, " %lu", (1UL << i));
2128             }
2129         }
2130         s = g_string_free(g, false);
2131         error_append_hint(errp, "%s.\n", s);
2132         g_free(s);
2133     } else {
2134         error_append_hint(errp,
2135                           "This KVM seems to be too old to support VSMT.\n");
2136     }
2137 }
2138 
2139 
2140 #ifdef TARGET_PPC64
2141 uint64_t kvmppc_vrma_limit(unsigned int hash_shift)
2142 {
2143     struct kvm_ppc_smmu_info info;
2144     long rampagesize, best_page_shift;
2145     int i;
2146 
2147     /*
2148      * Find the largest hardware supported page size that's less than
2149      * or equal to the (logical) backing page size of guest RAM
2150      */
2151     kvm_get_smmu_info(&info, &error_fatal);
2152     rampagesize = qemu_minrampagesize();
2153     best_page_shift = 0;
2154 
2155     for (i = 0; i < KVM_PPC_PAGE_SIZES_MAX_SZ; i++) {
2156         struct kvm_ppc_one_seg_page_size *sps = &info.sps[i];
2157 
2158         if (!sps->page_shift) {
2159             continue;
2160         }
2161 
2162         if ((sps->page_shift > best_page_shift)
2163             && ((1UL << sps->page_shift) <= rampagesize)) {
2164             best_page_shift = sps->page_shift;
2165         }
2166     }
2167 
2168     return 1ULL << (best_page_shift + hash_shift - 7);
2169 }
2170 #endif
2171 
2172 bool kvmppc_spapr_use_multitce(void)
2173 {
2174     return cap_spapr_multitce;
2175 }
2176 
2177 int kvmppc_spapr_enable_inkernel_multitce(void)
2178 {
2179     int ret;
2180 
2181     ret = kvm_vm_enable_cap(kvm_state, KVM_CAP_PPC_ENABLE_HCALL, 0,
2182                             H_PUT_TCE_INDIRECT, 1);
2183     if (!ret) {
2184         ret = kvm_vm_enable_cap(kvm_state, KVM_CAP_PPC_ENABLE_HCALL, 0,
2185                                 H_STUFF_TCE, 1);
2186     }
2187 
2188     return ret;
2189 }
2190 
2191 void *kvmppc_create_spapr_tce(uint32_t liobn, uint32_t page_shift,
2192                               uint64_t bus_offset, uint32_t nb_table,
2193                               int *pfd, bool need_vfio)
2194 {
2195     long len;
2196     int fd;
2197     void *table;
2198 
2199     /*
2200      * Must set fd to -1 so we don't try to munmap when called for
2201      * destroying the table, which the upper layers -will- do
2202      */
2203     *pfd = -1;
2204     if (!cap_spapr_tce || (need_vfio && !cap_spapr_vfio)) {
2205         return NULL;
2206     }
2207 
2208     if (cap_spapr_tce_64) {
2209         struct kvm_create_spapr_tce_64 args = {
2210             .liobn = liobn,
2211             .page_shift = page_shift,
2212             .offset = bus_offset >> page_shift,
2213             .size = nb_table,
2214             .flags = 0
2215         };
2216         fd = kvm_vm_ioctl(kvm_state, KVM_CREATE_SPAPR_TCE_64, &args);
2217         if (fd < 0) {
2218             fprintf(stderr,
2219                     "KVM: Failed to create TCE64 table for liobn 0x%x\n",
2220                     liobn);
2221             return NULL;
2222         }
2223     } else if (cap_spapr_tce) {
2224         uint64_t window_size = (uint64_t) nb_table << page_shift;
2225         struct kvm_create_spapr_tce args = {
2226             .liobn = liobn,
2227             .window_size = window_size,
2228         };
2229         if ((window_size != args.window_size) || bus_offset) {
2230             return NULL;
2231         }
2232         fd = kvm_vm_ioctl(kvm_state, KVM_CREATE_SPAPR_TCE, &args);
2233         if (fd < 0) {
2234             fprintf(stderr, "KVM: Failed to create TCE table for liobn 0x%x\n",
2235                     liobn);
2236             return NULL;
2237         }
2238     } else {
2239         return NULL;
2240     }
2241 
2242     len = nb_table * sizeof(uint64_t);
2243     /* FIXME: round this up to page size */
2244 
2245     table = mmap(NULL, len, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
2246     if (table == MAP_FAILED) {
2247         fprintf(stderr, "KVM: Failed to map TCE table for liobn 0x%x\n",
2248                 liobn);
2249         close(fd);
2250         return NULL;
2251     }
2252 
2253     *pfd = fd;
2254     return table;
2255 }
2256 
2257 int kvmppc_remove_spapr_tce(void *table, int fd, uint32_t nb_table)
2258 {
2259     long len;
2260 
2261     if (fd < 0) {
2262         return -1;
2263     }
2264 
2265     len = nb_table * sizeof(uint64_t);
2266     if ((munmap(table, len) < 0) ||
2267         (close(fd) < 0)) {
2268         fprintf(stderr, "KVM: Unexpected error removing TCE table: %s",
2269                 strerror(errno));
2270         /* Leak the table */
2271     }
2272 
2273     return 0;
2274 }
2275 
2276 int kvmppc_reset_htab(int shift_hint)
2277 {
2278     uint32_t shift = shift_hint;
2279 
2280     if (!kvm_enabled()) {
2281         /* Full emulation, tell caller to allocate htab itself */
2282         return 0;
2283     }
2284     if (kvm_vm_check_extension(kvm_state, KVM_CAP_PPC_ALLOC_HTAB)) {
2285         int ret;
2286         ret = kvm_vm_ioctl(kvm_state, KVM_PPC_ALLOCATE_HTAB, &shift);
2287         if (ret == -ENOTTY) {
2288             /*
2289              * At least some versions of PR KVM advertise the
2290              * capability, but don't implement the ioctl().  Oops.
2291              * Return 0 so that we allocate the htab in qemu, as is
2292              * correct for PR.
2293              */
2294             return 0;
2295         } else if (ret < 0) {
2296             return ret;
2297         }
2298         return shift;
2299     }
2300 
2301     /*
2302      * We have a kernel that predates the htab reset calls.  For PR
2303      * KVM, we need to allocate the htab ourselves, for an HV KVM of
2304      * this era, it has allocated a 16MB fixed size hash table
2305      * already.
2306      */
2307     if (kvmppc_is_pr(kvm_state)) {
2308         /* PR - tell caller to allocate htab */
2309         return 0;
2310     } else {
2311         /* HV - assume 16MB kernel allocated htab */
2312         return 24;
2313     }
2314 }
2315 
2316 static inline uint32_t mfpvr(void)
2317 {
2318     uint32_t pvr;
2319 
2320     asm ("mfpvr %0"
2321          : "=r"(pvr));
2322     return pvr;
2323 }
2324 
2325 static void alter_insns(uint64_t *word, uint64_t flags, bool on)
2326 {
2327     if (on) {
2328         *word |= flags;
2329     } else {
2330         *word &= ~flags;
2331     }
2332 }
2333 
2334 static void kvmppc_host_cpu_class_init(ObjectClass *oc, void *data)
2335 {
2336     PowerPCCPUClass *pcc = POWERPC_CPU_CLASS(oc);
2337     uint32_t dcache_size = kvmppc_read_int_cpu_dt("d-cache-size");
2338     uint32_t icache_size = kvmppc_read_int_cpu_dt("i-cache-size");
2339 
2340     /* Now fix up the class with information we can query from the host */
2341     pcc->pvr = mfpvr();
2342 
2343     alter_insns(&pcc->insns_flags, PPC_ALTIVEC,
2344                 qemu_getauxval(AT_HWCAP) & PPC_FEATURE_HAS_ALTIVEC);
2345     alter_insns(&pcc->insns_flags2, PPC2_VSX,
2346                 qemu_getauxval(AT_HWCAP) & PPC_FEATURE_HAS_VSX);
2347     alter_insns(&pcc->insns_flags2, PPC2_DFP,
2348                 qemu_getauxval(AT_HWCAP) & PPC_FEATURE_HAS_DFP);
2349 
2350     if (dcache_size != -1) {
2351         pcc->l1_dcache_size = dcache_size;
2352     }
2353 
2354     if (icache_size != -1) {
2355         pcc->l1_icache_size = icache_size;
2356     }
2357 
2358 #if defined(TARGET_PPC64)
2359     pcc->radix_page_info = kvm_get_radix_page_info();
2360 
2361     if ((pcc->pvr & 0xffffff00) == CPU_POWERPC_POWER9_DD1) {
2362         /*
2363          * POWER9 DD1 has some bugs which make it not really ISA 3.00
2364          * compliant.  More importantly, advertising ISA 3.00
2365          * architected mode may prevent guests from activating
2366          * necessary DD1 workarounds.
2367          */
2368         pcc->pcr_supported &= ~(PCR_COMPAT_3_00 | PCR_COMPAT_2_07
2369                                 | PCR_COMPAT_2_06 | PCR_COMPAT_2_05);
2370     }
2371 #endif /* defined(TARGET_PPC64) */
2372 }
2373 
2374 bool kvmppc_has_cap_epr(void)
2375 {
2376     return cap_epr;
2377 }
2378 
2379 bool kvmppc_has_cap_fixup_hcalls(void)
2380 {
2381     return cap_fixup_hcalls;
2382 }
2383 
2384 bool kvmppc_has_cap_htm(void)
2385 {
2386     return cap_htm;
2387 }
2388 
2389 bool kvmppc_has_cap_mmu_radix(void)
2390 {
2391     return cap_mmu_radix;
2392 }
2393 
2394 bool kvmppc_has_cap_mmu_hash_v3(void)
2395 {
2396     return cap_mmu_hash_v3;
2397 }
2398 
2399 static bool kvmppc_power8_host(void)
2400 {
2401     bool ret = false;
2402 #ifdef TARGET_PPC64
2403     {
2404         uint32_t base_pvr = CPU_POWERPC_POWER_SERVER_MASK & mfpvr();
2405         ret = (base_pvr == CPU_POWERPC_POWER8E_BASE) ||
2406               (base_pvr == CPU_POWERPC_POWER8NVL_BASE) ||
2407               (base_pvr == CPU_POWERPC_POWER8_BASE);
2408     }
2409 #endif /* TARGET_PPC64 */
2410     return ret;
2411 }
2412 
2413 static int parse_cap_ppc_safe_cache(struct kvm_ppc_cpu_char c)
2414 {
2415     bool l1d_thread_priv_req = !kvmppc_power8_host();
2416 
2417     if (~c.behaviour & c.behaviour_mask & H_CPU_BEHAV_L1D_FLUSH_PR) {
2418         return 2;
2419     } else if ((!l1d_thread_priv_req ||
2420                 c.character & c.character_mask & H_CPU_CHAR_L1D_THREAD_PRIV) &&
2421                (c.character & c.character_mask
2422                 & (H_CPU_CHAR_L1D_FLUSH_ORI30 | H_CPU_CHAR_L1D_FLUSH_TRIG2))) {
2423         return 1;
2424     }
2425 
2426     return 0;
2427 }
2428 
2429 static int parse_cap_ppc_safe_bounds_check(struct kvm_ppc_cpu_char c)
2430 {
2431     if (~c.behaviour & c.behaviour_mask & H_CPU_BEHAV_BNDS_CHK_SPEC_BAR) {
2432         return 2;
2433     } else if (c.character & c.character_mask & H_CPU_CHAR_SPEC_BAR_ORI31) {
2434         return 1;
2435     }
2436 
2437     return 0;
2438 }
2439 
2440 static int parse_cap_ppc_safe_indirect_branch(struct kvm_ppc_cpu_char c)
2441 {
2442     if ((~c.behaviour & c.behaviour_mask & H_CPU_BEHAV_FLUSH_COUNT_CACHE) &&
2443         (~c.character & c.character_mask & H_CPU_CHAR_CACHE_COUNT_DIS) &&
2444         (~c.character & c.character_mask & H_CPU_CHAR_BCCTRL_SERIALISED)) {
2445         return SPAPR_CAP_FIXED_NA;
2446     } else if (c.behaviour & c.behaviour_mask & H_CPU_BEHAV_FLUSH_COUNT_CACHE) {
2447         return SPAPR_CAP_WORKAROUND;
2448     } else if (c.character & c.character_mask & H_CPU_CHAR_CACHE_COUNT_DIS) {
2449         return  SPAPR_CAP_FIXED_CCD;
2450     } else if (c.character & c.character_mask & H_CPU_CHAR_BCCTRL_SERIALISED) {
2451         return SPAPR_CAP_FIXED_IBS;
2452     }
2453 
2454     return 0;
2455 }
2456 
2457 static int parse_cap_ppc_count_cache_flush_assist(struct kvm_ppc_cpu_char c)
2458 {
2459     if (c.character & c.character_mask & H_CPU_CHAR_BCCTR_FLUSH_ASSIST) {
2460         return 1;
2461     }
2462     return 0;
2463 }
2464 
2465 bool kvmppc_has_cap_xive(void)
2466 {
2467     return cap_xive;
2468 }
2469 
2470 static void kvmppc_get_cpu_characteristics(KVMState *s)
2471 {
2472     struct kvm_ppc_cpu_char c;
2473     int ret;
2474 
2475     /* Assume broken */
2476     cap_ppc_safe_cache = 0;
2477     cap_ppc_safe_bounds_check = 0;
2478     cap_ppc_safe_indirect_branch = 0;
2479 
2480     ret = kvm_vm_check_extension(s, KVM_CAP_PPC_GET_CPU_CHAR);
2481     if (!ret) {
2482         return;
2483     }
2484     ret = kvm_vm_ioctl(s, KVM_PPC_GET_CPU_CHAR, &c);
2485     if (ret < 0) {
2486         return;
2487     }
2488 
2489     cap_ppc_safe_cache = parse_cap_ppc_safe_cache(c);
2490     cap_ppc_safe_bounds_check = parse_cap_ppc_safe_bounds_check(c);
2491     cap_ppc_safe_indirect_branch = parse_cap_ppc_safe_indirect_branch(c);
2492     cap_ppc_count_cache_flush_assist =
2493         parse_cap_ppc_count_cache_flush_assist(c);
2494 }
2495 
2496 int kvmppc_get_cap_safe_cache(void)
2497 {
2498     return cap_ppc_safe_cache;
2499 }
2500 
2501 int kvmppc_get_cap_safe_bounds_check(void)
2502 {
2503     return cap_ppc_safe_bounds_check;
2504 }
2505 
2506 int kvmppc_get_cap_safe_indirect_branch(void)
2507 {
2508     return cap_ppc_safe_indirect_branch;
2509 }
2510 
2511 int kvmppc_get_cap_count_cache_flush_assist(void)
2512 {
2513     return cap_ppc_count_cache_flush_assist;
2514 }
2515 
2516 bool kvmppc_has_cap_nested_kvm_hv(void)
2517 {
2518     return !!cap_ppc_nested_kvm_hv;
2519 }
2520 
2521 int kvmppc_set_cap_nested_kvm_hv(int enable)
2522 {
2523     return kvm_vm_enable_cap(kvm_state, KVM_CAP_PPC_NESTED_HV, 0, enable);
2524 }
2525 
2526 bool kvmppc_has_cap_spapr_vfio(void)
2527 {
2528     return cap_spapr_vfio;
2529 }
2530 
2531 int kvmppc_get_cap_large_decr(void)
2532 {
2533     return cap_large_decr;
2534 }
2535 
2536 int kvmppc_enable_cap_large_decr(PowerPCCPU *cpu, int enable)
2537 {
2538     CPUState *cs = CPU(cpu);
2539     uint64_t lpcr;
2540 
2541     kvm_get_one_reg(cs, KVM_REG_PPC_LPCR_64, &lpcr);
2542     /* Do we need to modify the LPCR? */
2543     if (!!(lpcr & LPCR_LD) != !!enable) {
2544         if (enable) {
2545             lpcr |= LPCR_LD;
2546         } else {
2547             lpcr &= ~LPCR_LD;
2548         }
2549         kvm_set_one_reg(cs, KVM_REG_PPC_LPCR_64, &lpcr);
2550         kvm_get_one_reg(cs, KVM_REG_PPC_LPCR_64, &lpcr);
2551 
2552         if (!!(lpcr & LPCR_LD) != !!enable) {
2553             return -1;
2554         }
2555     }
2556 
2557     return 0;
2558 }
2559 
2560 int kvmppc_has_cap_rpt_invalidate(void)
2561 {
2562     return cap_rpt_invalidate;
2563 }
2564 
2565 PowerPCCPUClass *kvm_ppc_get_host_cpu_class(void)
2566 {
2567     uint32_t host_pvr = mfpvr();
2568     PowerPCCPUClass *pvr_pcc;
2569 
2570     pvr_pcc = ppc_cpu_class_by_pvr(host_pvr);
2571     if (pvr_pcc == NULL) {
2572         pvr_pcc = ppc_cpu_class_by_pvr_mask(host_pvr);
2573     }
2574 
2575     return pvr_pcc;
2576 }
2577 
2578 static void pseries_machine_class_fixup(ObjectClass *oc, void *opaque)
2579 {
2580     MachineClass *mc = MACHINE_CLASS(oc);
2581 
2582     mc->default_cpu_type = TYPE_HOST_POWERPC_CPU;
2583 }
2584 
2585 static int kvm_ppc_register_host_cpu_type(void)
2586 {
2587     TypeInfo type_info = {
2588         .name = TYPE_HOST_POWERPC_CPU,
2589         .class_init = kvmppc_host_cpu_class_init,
2590     };
2591     PowerPCCPUClass *pvr_pcc;
2592     ObjectClass *oc;
2593     DeviceClass *dc;
2594     int i;
2595 
2596     pvr_pcc = kvm_ppc_get_host_cpu_class();
2597     if (pvr_pcc == NULL) {
2598         return -1;
2599     }
2600     type_info.parent = object_class_get_name(OBJECT_CLASS(pvr_pcc));
2601     type_register(&type_info);
2602     /* override TCG default cpu type with 'host' cpu model */
2603     object_class_foreach(pseries_machine_class_fixup, TYPE_SPAPR_MACHINE,
2604                          false, NULL);
2605 
2606     oc = object_class_by_name(type_info.name);
2607     g_assert(oc);
2608 
2609     /*
2610      * Update generic CPU family class alias (e.g. on a POWER8NVL host,
2611      * we want "POWER8" to be a "family" alias that points to the current
2612      * host CPU type, too)
2613      */
2614     dc = DEVICE_CLASS(ppc_cpu_get_family_class(pvr_pcc));
2615     for (i = 0; ppc_cpu_aliases[i].alias != NULL; i++) {
2616         if (strcasecmp(ppc_cpu_aliases[i].alias, dc->desc) == 0) {
2617             char *suffix;
2618 
2619             ppc_cpu_aliases[i].model = g_strdup(object_class_get_name(oc));
2620             suffix = strstr(ppc_cpu_aliases[i].model, POWERPC_CPU_TYPE_SUFFIX);
2621             if (suffix) {
2622                 *suffix = 0;
2623             }
2624             break;
2625         }
2626     }
2627 
2628     return 0;
2629 }
2630 
2631 int kvmppc_define_rtas_kernel_token(uint32_t token, const char *function)
2632 {
2633     struct kvm_rtas_token_args args = {
2634         .token = token,
2635     };
2636 
2637     if (!kvm_check_extension(kvm_state, KVM_CAP_PPC_RTAS)) {
2638         return -ENOENT;
2639     }
2640 
2641     strncpy(args.name, function, sizeof(args.name) - 1);
2642 
2643     return kvm_vm_ioctl(kvm_state, KVM_PPC_RTAS_DEFINE_TOKEN, &args);
2644 }
2645 
2646 int kvmppc_get_htab_fd(bool write, uint64_t index, Error **errp)
2647 {
2648     struct kvm_get_htab_fd s = {
2649         .flags = write ? KVM_GET_HTAB_WRITE : 0,
2650         .start_index = index,
2651     };
2652     int ret;
2653 
2654     if (!cap_htab_fd) {
2655         error_setg(errp, "KVM version doesn't support %s the HPT",
2656                    write ? "writing" : "reading");
2657         return -ENOTSUP;
2658     }
2659 
2660     ret = kvm_vm_ioctl(kvm_state, KVM_PPC_GET_HTAB_FD, &s);
2661     if (ret < 0) {
2662         error_setg(errp, "Unable to open fd for %s HPT %s KVM: %s",
2663                    write ? "writing" : "reading", write ? "to" : "from",
2664                    strerror(errno));
2665         return -errno;
2666     }
2667 
2668     return ret;
2669 }
2670 
2671 int kvmppc_save_htab(QEMUFile *f, int fd, size_t bufsize, int64_t max_ns)
2672 {
2673     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2674     uint8_t buf[bufsize];
2675     ssize_t rc;
2676 
2677     do {
2678         rc = read(fd, buf, bufsize);
2679         if (rc < 0) {
2680             fprintf(stderr, "Error reading data from KVM HTAB fd: %s\n",
2681                     strerror(errno));
2682             return rc;
2683         } else if (rc) {
2684             uint8_t *buffer = buf;
2685             ssize_t n = rc;
2686             while (n) {
2687                 struct kvm_get_htab_header *head =
2688                     (struct kvm_get_htab_header *) buffer;
2689                 size_t chunksize = sizeof(*head) +
2690                      HASH_PTE_SIZE_64 * head->n_valid;
2691 
2692                 qemu_put_be32(f, head->index);
2693                 qemu_put_be16(f, head->n_valid);
2694                 qemu_put_be16(f, head->n_invalid);
2695                 qemu_put_buffer(f, (void *)(head + 1),
2696                                 HASH_PTE_SIZE_64 * head->n_valid);
2697 
2698                 buffer += chunksize;
2699                 n -= chunksize;
2700             }
2701         }
2702     } while ((rc != 0)
2703              && ((max_ns < 0) ||
2704                  ((qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) < max_ns)));
2705 
2706     return (rc == 0) ? 1 : 0;
2707 }
2708 
2709 int kvmppc_load_htab_chunk(QEMUFile *f, int fd, uint32_t index,
2710                            uint16_t n_valid, uint16_t n_invalid, Error **errp)
2711 {
2712     struct kvm_get_htab_header *buf;
2713     size_t chunksize = sizeof(*buf) + n_valid * HASH_PTE_SIZE_64;
2714     ssize_t rc;
2715 
2716     buf = alloca(chunksize);
2717     buf->index = index;
2718     buf->n_valid = n_valid;
2719     buf->n_invalid = n_invalid;
2720 
2721     qemu_get_buffer(f, (void *)(buf + 1), HASH_PTE_SIZE_64 * n_valid);
2722 
2723     rc = write(fd, buf, chunksize);
2724     if (rc < 0) {
2725         error_setg_errno(errp, errno, "Error writing the KVM hash table");
2726         return -errno;
2727     }
2728     if (rc != chunksize) {
2729         /* We should never get a short write on a single chunk */
2730         error_setg(errp, "Short write while restoring the KVM hash table");
2731         return -ENOSPC;
2732     }
2733     return 0;
2734 }
2735 
2736 bool kvm_arch_stop_on_emulation_error(CPUState *cpu)
2737 {
2738     return true;
2739 }
2740 
2741 void kvm_arch_init_irq_routing(KVMState *s)
2742 {
2743 }
2744 
2745 void kvmppc_read_hptes(ppc_hash_pte64_t *hptes, hwaddr ptex, int n)
2746 {
2747     int fd, rc;
2748     int i;
2749 
2750     fd = kvmppc_get_htab_fd(false, ptex, &error_abort);
2751 
2752     i = 0;
2753     while (i < n) {
2754         struct kvm_get_htab_header *hdr;
2755         int m = n < HPTES_PER_GROUP ? n : HPTES_PER_GROUP;
2756         char buf[sizeof(*hdr) + m * HASH_PTE_SIZE_64];
2757 
2758         rc = read(fd, buf, sizeof(buf));
2759         if (rc < 0) {
2760             hw_error("kvmppc_read_hptes: Unable to read HPTEs");
2761         }
2762 
2763         hdr = (struct kvm_get_htab_header *)buf;
2764         while ((i < n) && ((char *)hdr < (buf + rc))) {
2765             int invalid = hdr->n_invalid, valid = hdr->n_valid;
2766 
2767             if (hdr->index != (ptex + i)) {
2768                 hw_error("kvmppc_read_hptes: Unexpected HPTE index %"PRIu32
2769                          " != (%"HWADDR_PRIu" + %d", hdr->index, ptex, i);
2770             }
2771 
2772             if (n - i < valid) {
2773                 valid = n - i;
2774             }
2775             memcpy(hptes + i, hdr + 1, HASH_PTE_SIZE_64 * valid);
2776             i += valid;
2777 
2778             if ((n - i) < invalid) {
2779                 invalid = n - i;
2780             }
2781             memset(hptes + i, 0, invalid * HASH_PTE_SIZE_64);
2782             i += invalid;
2783 
2784             hdr = (struct kvm_get_htab_header *)
2785                 ((char *)(hdr + 1) + HASH_PTE_SIZE_64 * hdr->n_valid);
2786         }
2787     }
2788 
2789     close(fd);
2790 }
2791 
2792 void kvmppc_write_hpte(hwaddr ptex, uint64_t pte0, uint64_t pte1)
2793 {
2794     int fd, rc;
2795     struct {
2796         struct kvm_get_htab_header hdr;
2797         uint64_t pte0;
2798         uint64_t pte1;
2799     } buf;
2800 
2801     fd = kvmppc_get_htab_fd(true, 0 /* Ignored */, &error_abort);
2802 
2803     buf.hdr.n_valid = 1;
2804     buf.hdr.n_invalid = 0;
2805     buf.hdr.index = ptex;
2806     buf.pte0 = cpu_to_be64(pte0);
2807     buf.pte1 = cpu_to_be64(pte1);
2808 
2809     rc = write(fd, &buf, sizeof(buf));
2810     if (rc != sizeof(buf)) {
2811         hw_error("kvmppc_write_hpte: Unable to update KVM HPT");
2812     }
2813     close(fd);
2814 }
2815 
2816 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
2817                              uint64_t address, uint32_t data, PCIDevice *dev)
2818 {
2819     return 0;
2820 }
2821 
2822 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
2823                                 int vector, PCIDevice *dev)
2824 {
2825     return 0;
2826 }
2827 
2828 int kvm_arch_release_virq_post(int virq)
2829 {
2830     return 0;
2831 }
2832 
2833 int kvm_arch_msi_data_to_gsi(uint32_t data)
2834 {
2835     return data & 0xffff;
2836 }
2837 
2838 #if defined(TARGET_PPC64)
2839 int kvm_handle_nmi(PowerPCCPU *cpu, struct kvm_run *run)
2840 {
2841     uint16_t flags = run->flags & KVM_RUN_PPC_NMI_DISP_MASK;
2842 
2843     cpu_synchronize_state(CPU(cpu));
2844 
2845     spapr_mce_req_event(cpu, flags == KVM_RUN_PPC_NMI_DISP_FULLY_RECOV);
2846 
2847     return 0;
2848 }
2849 #endif
2850 
2851 int kvmppc_enable_hwrng(void)
2852 {
2853     if (!kvm_enabled() || !kvm_check_extension(kvm_state, KVM_CAP_PPC_HWRNG)) {
2854         return -1;
2855     }
2856 
2857     return kvmppc_enable_hcall(kvm_state, H_RANDOM);
2858 }
2859 
2860 void kvmppc_check_papr_resize_hpt(Error **errp)
2861 {
2862     if (!kvm_enabled()) {
2863         return; /* No KVM, we're good */
2864     }
2865 
2866     if (cap_resize_hpt) {
2867         return; /* Kernel has explicit support, we're good */
2868     }
2869 
2870     /* Otherwise fallback on looking for PR KVM */
2871     if (kvmppc_is_pr(kvm_state)) {
2872         return;
2873     }
2874 
2875     error_setg(errp,
2876                "Hash page table resizing not available with this KVM version");
2877 }
2878 
2879 int kvmppc_resize_hpt_prepare(PowerPCCPU *cpu, target_ulong flags, int shift)
2880 {
2881     CPUState *cs = CPU(cpu);
2882     struct kvm_ppc_resize_hpt rhpt = {
2883         .flags = flags,
2884         .shift = shift,
2885     };
2886 
2887     if (!cap_resize_hpt) {
2888         return -ENOSYS;
2889     }
2890 
2891     return kvm_vm_ioctl(cs->kvm_state, KVM_PPC_RESIZE_HPT_PREPARE, &rhpt);
2892 }
2893 
2894 int kvmppc_resize_hpt_commit(PowerPCCPU *cpu, target_ulong flags, int shift)
2895 {
2896     CPUState *cs = CPU(cpu);
2897     struct kvm_ppc_resize_hpt rhpt = {
2898         .flags = flags,
2899         .shift = shift,
2900     };
2901 
2902     if (!cap_resize_hpt) {
2903         return -ENOSYS;
2904     }
2905 
2906     return kvm_vm_ioctl(cs->kvm_state, KVM_PPC_RESIZE_HPT_COMMIT, &rhpt);
2907 }
2908 
2909 /*
2910  * This is a helper function to detect a post migration scenario
2911  * in which a guest, running as KVM-HV, freezes in cpu_post_load because
2912  * the guest kernel can't handle a PVR value other than the actual host
2913  * PVR in KVM_SET_SREGS, even if pvr_match() returns true.
2914  *
2915  * If we don't have cap_ppc_pvr_compat and we're not running in PR
2916  * (so, we're HV), return true. The workaround itself is done in
2917  * cpu_post_load.
2918  *
2919  * The order here is important: we'll only check for KVM PR as a
2920  * fallback if the guest kernel can't handle the situation itself.
2921  * We need to avoid as much as possible querying the running KVM type
2922  * in QEMU level.
2923  */
2924 bool kvmppc_pvr_workaround_required(PowerPCCPU *cpu)
2925 {
2926     CPUState *cs = CPU(cpu);
2927 
2928     if (!kvm_enabled()) {
2929         return false;
2930     }
2931 
2932     if (cap_ppc_pvr_compat) {
2933         return false;
2934     }
2935 
2936     return !kvmppc_is_pr(cs->kvm_state);
2937 }
2938 
2939 void kvmppc_set_reg_ppc_online(PowerPCCPU *cpu, unsigned int online)
2940 {
2941     CPUState *cs = CPU(cpu);
2942 
2943     if (kvm_enabled()) {
2944         kvm_set_one_reg(cs, KVM_REG_PPC_ONLINE, &online);
2945     }
2946 }
2947 
2948 void kvmppc_set_reg_tb_offset(PowerPCCPU *cpu, int64_t tb_offset)
2949 {
2950     CPUState *cs = CPU(cpu);
2951 
2952     if (kvm_enabled()) {
2953         kvm_set_one_reg(cs, KVM_REG_PPC_TB_OFFSET, &tb_offset);
2954     }
2955 }
2956 
2957 bool kvm_arch_cpu_check_are_resettable(void)
2958 {
2959     return true;
2960 }
2961