xref: /openbmc/qemu/target/ppc/kvm.c (revision 06831001)
1 /*
2  * PowerPC implementation of KVM hooks
3  *
4  * Copyright IBM Corp. 2007
5  * Copyright (C) 2011 Freescale Semiconductor, Inc.
6  *
7  * Authors:
8  *  Jerone Young <jyoung5@us.ibm.com>
9  *  Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
10  *  Hollis Blanchard <hollisb@us.ibm.com>
11  *
12  * This work is licensed under the terms of the GNU GPL, version 2 or later.
13  * See the COPYING file in the top-level directory.
14  *
15  */
16 
17 #include "qemu/osdep.h"
18 #include <dirent.h>
19 #include <sys/ioctl.h>
20 #include <sys/vfs.h>
21 
22 #include <linux/kvm.h>
23 
24 #include "qapi/error.h"
25 #include "qemu/error-report.h"
26 #include "cpu.h"
27 #include "cpu-models.h"
28 #include "qemu/timer.h"
29 #include "sysemu/hw_accel.h"
30 #include "kvm_ppc.h"
31 #include "sysemu/cpus.h"
32 #include "sysemu/device_tree.h"
33 #include "mmu-hash64.h"
34 
35 #include "hw/ppc/spapr.h"
36 #include "hw/ppc/spapr_cpu_core.h"
37 #include "hw/hw.h"
38 #include "hw/ppc/ppc.h"
39 #include "migration/qemu-file-types.h"
40 #include "sysemu/watchdog.h"
41 #include "trace.h"
42 #include "exec/gdbstub.h"
43 #include "exec/memattrs.h"
44 #include "exec/ram_addr.h"
45 #include "sysemu/hostmem.h"
46 #include "qemu/cutils.h"
47 #include "qemu/main-loop.h"
48 #include "qemu/mmap-alloc.h"
49 #include "elf.h"
50 #include "sysemu/kvm_int.h"
51 
52 #define PROC_DEVTREE_CPU      "/proc/device-tree/cpus/"
53 
54 #define DEBUG_RETURN_GUEST 0
55 #define DEBUG_RETURN_GDB   1
56 
57 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
58     KVM_CAP_LAST_INFO
59 };
60 
61 static int cap_interrupt_unset;
62 static int cap_segstate;
63 static int cap_booke_sregs;
64 static int cap_ppc_smt;
65 static int cap_ppc_smt_possible;
66 static int cap_spapr_tce;
67 static int cap_spapr_tce_64;
68 static int cap_spapr_multitce;
69 static int cap_spapr_vfio;
70 static int cap_hior;
71 static int cap_one_reg;
72 static int cap_epr;
73 static int cap_ppc_watchdog;
74 static int cap_papr;
75 static int cap_htab_fd;
76 static int cap_fixup_hcalls;
77 static int cap_htm;             /* Hardware transactional memory support */
78 static int cap_mmu_radix;
79 static int cap_mmu_hash_v3;
80 static int cap_xive;
81 static int cap_resize_hpt;
82 static int cap_ppc_pvr_compat;
83 static int cap_ppc_safe_cache;
84 static int cap_ppc_safe_bounds_check;
85 static int cap_ppc_safe_indirect_branch;
86 static int cap_ppc_count_cache_flush_assist;
87 static int cap_ppc_nested_kvm_hv;
88 static int cap_large_decr;
89 static int cap_fwnmi;
90 static int cap_rpt_invalidate;
91 static int cap_ail_mode_3;
92 
93 static uint32_t debug_inst_opcode;
94 
95 /*
96  * Check whether we are running with KVM-PR (instead of KVM-HV).  This
97  * should only be used for fallback tests - generally we should use
98  * explicit capabilities for the features we want, rather than
99  * assuming what is/isn't available depending on the KVM variant.
100  */
101 static bool kvmppc_is_pr(KVMState *ks)
102 {
103     /* Assume KVM-PR if the GET_PVINFO capability is available */
104     return kvm_vm_check_extension(ks, KVM_CAP_PPC_GET_PVINFO) != 0;
105 }
106 
107 static int kvm_ppc_register_host_cpu_type(void);
108 static void kvmppc_get_cpu_characteristics(KVMState *s);
109 static int kvmppc_get_dec_bits(void);
110 
111 int kvm_arch_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     cap_ail_mode_3 = kvm_vm_check_extension(s, KVM_CAP_PPC_AIL_MODE_3);
157     kvm_ppc_register_host_cpu_type();
158 
159     return 0;
160 }
161 
162 int kvm_arch_irqchip_create(KVMState *s)
163 {
164     return 0;
165 }
166 
167 static int kvm_arch_sync_sregs(PowerPCCPU *cpu)
168 {
169     CPUPPCState *cenv = &cpu->env;
170     CPUState *cs = CPU(cpu);
171     struct kvm_sregs sregs;
172     int ret;
173 
174     if (cenv->excp_model == POWERPC_EXCP_BOOKE) {
175         /*
176          * What we're really trying to say is "if we're on BookE, we
177          * use the native PVR for now". This is the only sane way to
178          * check it though, so we potentially confuse users that they
179          * can run BookE guests on BookS. Let's hope nobody dares
180          * enough :)
181          */
182         return 0;
183     } else {
184         if (!cap_segstate) {
185             fprintf(stderr, "kvm error: missing PVR setting capability\n");
186             return -ENOSYS;
187         }
188     }
189 
190     ret = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs);
191     if (ret) {
192         return ret;
193     }
194 
195     sregs.pvr = cenv->spr[SPR_PVR];
196     return kvm_vcpu_ioctl(cs, KVM_SET_SREGS, &sregs);
197 }
198 
199 /* Set up a shared TLB array with KVM */
200 static int kvm_booke206_tlb_init(PowerPCCPU *cpu)
201 {
202     CPUPPCState *env = &cpu->env;
203     CPUState *cs = CPU(cpu);
204     struct kvm_book3e_206_tlb_params params = {};
205     struct kvm_config_tlb cfg = {};
206     unsigned int entries = 0;
207     int ret, i;
208 
209     if (!kvm_enabled() ||
210         !kvm_check_extension(cs->kvm_state, KVM_CAP_SW_TLB)) {
211         return 0;
212     }
213 
214     assert(ARRAY_SIZE(params.tlb_sizes) == BOOKE206_MAX_TLBN);
215 
216     for (i = 0; i < BOOKE206_MAX_TLBN; i++) {
217         params.tlb_sizes[i] = booke206_tlb_size(env, i);
218         params.tlb_ways[i] = booke206_tlb_ways(env, i);
219         entries += params.tlb_sizes[i];
220     }
221 
222     assert(entries == env->nb_tlb);
223     assert(sizeof(struct kvm_book3e_206_tlb_entry) == sizeof(ppcmas_tlb_t));
224 
225     env->tlb_dirty = true;
226 
227     cfg.array = (uintptr_t)env->tlb.tlbm;
228     cfg.array_len = sizeof(ppcmas_tlb_t) * entries;
229     cfg.params = (uintptr_t)&params;
230     cfg.mmu_type = KVM_MMU_FSL_BOOKE_NOHV;
231 
232     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_SW_TLB, 0, (uintptr_t)&cfg);
233     if (ret < 0) {
234         fprintf(stderr, "%s: couldn't enable KVM_CAP_SW_TLB: %s\n",
235                 __func__, strerror(-ret));
236         return ret;
237     }
238 
239     env->kvm_sw_tlb = true;
240     return 0;
241 }
242 
243 
244 #if defined(TARGET_PPC64)
245 static void kvm_get_smmu_info(struct kvm_ppc_smmu_info *info, Error **errp)
246 {
247     int ret;
248 
249     assert(kvm_state != NULL);
250 
251     if (!kvm_check_extension(kvm_state, KVM_CAP_PPC_GET_SMMU_INFO)) {
252         error_setg(errp, "KVM doesn't expose the MMU features it supports");
253         error_append_hint(errp, "Consider switching to a newer KVM\n");
254         return;
255     }
256 
257     ret = kvm_vm_ioctl(kvm_state, KVM_PPC_GET_SMMU_INFO, info);
258     if (ret == 0) {
259         return;
260     }
261 
262     error_setg_errno(errp, -ret,
263                      "KVM failed to provide the MMU features it supports");
264 }
265 
266 struct ppc_radix_page_info *kvm_get_radix_page_info(void)
267 {
268     KVMState *s = KVM_STATE(current_accel());
269     struct ppc_radix_page_info *radix_page_info;
270     struct kvm_ppc_rmmu_info rmmu_info = { };
271     int i;
272 
273     if (!kvm_check_extension(s, KVM_CAP_PPC_MMU_RADIX)) {
274         return NULL;
275     }
276     if (kvm_vm_ioctl(s, KVM_PPC_GET_RMMU_INFO, &rmmu_info)) {
277         return NULL;
278     }
279     radix_page_info = g_malloc0(sizeof(*radix_page_info));
280     radix_page_info->count = 0;
281     for (i = 0; i < PPC_PAGE_SIZES_MAX_SZ; i++) {
282         if (rmmu_info.ap_encodings[i]) {
283             radix_page_info->entries[i] = rmmu_info.ap_encodings[i];
284             radix_page_info->count++;
285         }
286     }
287     return radix_page_info;
288 }
289 
290 target_ulong kvmppc_configure_v3_mmu(PowerPCCPU *cpu,
291                                      bool radix, bool gtse,
292                                      uint64_t proc_tbl)
293 {
294     CPUState *cs = CPU(cpu);
295     int ret;
296     uint64_t flags = 0;
297     struct kvm_ppc_mmuv3_cfg cfg = {
298         .process_table = proc_tbl,
299     };
300 
301     if (radix) {
302         flags |= KVM_PPC_MMUV3_RADIX;
303     }
304     if (gtse) {
305         flags |= KVM_PPC_MMUV3_GTSE;
306     }
307     cfg.flags = flags;
308     ret = kvm_vm_ioctl(cs->kvm_state, KVM_PPC_CONFIGURE_V3_MMU, &cfg);
309     switch (ret) {
310     case 0:
311         return H_SUCCESS;
312     case -EINVAL:
313         return H_PARAMETER;
314     case -ENODEV:
315         return H_NOT_AVAILABLE;
316     default:
317         return H_HARDWARE;
318     }
319 }
320 
321 bool kvmppc_hpt_needs_host_contiguous_pages(void)
322 {
323     static struct kvm_ppc_smmu_info smmu_info;
324 
325     if (!kvm_enabled()) {
326         return false;
327     }
328 
329     kvm_get_smmu_info(&smmu_info, &error_fatal);
330     return !!(smmu_info.flags & KVM_PPC_PAGE_SIZES_REAL);
331 }
332 
333 void kvm_check_mmu(PowerPCCPU *cpu, Error **errp)
334 {
335     struct kvm_ppc_smmu_info smmu_info;
336     int iq, ik, jq, jk;
337     Error *local_err = NULL;
338 
339     /* For now, we only have anything to check on hash64 MMUs */
340     if (!cpu->hash64_opts || !kvm_enabled()) {
341         return;
342     }
343 
344     kvm_get_smmu_info(&smmu_info, &local_err);
345     if (local_err) {
346         error_propagate(errp, local_err);
347         return;
348     }
349 
350     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)
351         && !(smmu_info.flags & KVM_PPC_1T_SEGMENTS)) {
352         error_setg(errp,
353                    "KVM does not support 1TiB segments which guest expects");
354         return;
355     }
356 
357     if (smmu_info.slb_size < cpu->hash64_opts->slb_size) {
358         error_setg(errp, "KVM only supports %u SLB entries, but guest needs %u",
359                    smmu_info.slb_size, cpu->hash64_opts->slb_size);
360         return;
361     }
362 
363     /*
364      * Verify that every pagesize supported by the cpu model is
365      * supported by KVM with the same encodings
366      */
367     for (iq = 0; iq < ARRAY_SIZE(cpu->hash64_opts->sps); iq++) {
368         PPCHash64SegmentPageSizes *qsps = &cpu->hash64_opts->sps[iq];
369         struct kvm_ppc_one_seg_page_size *ksps;
370 
371         for (ik = 0; ik < ARRAY_SIZE(smmu_info.sps); ik++) {
372             if (qsps->page_shift == smmu_info.sps[ik].page_shift) {
373                 break;
374             }
375         }
376         if (ik >= ARRAY_SIZE(smmu_info.sps)) {
377             error_setg(errp, "KVM doesn't support for base page shift %u",
378                        qsps->page_shift);
379             return;
380         }
381 
382         ksps = &smmu_info.sps[ik];
383         if (ksps->slb_enc != qsps->slb_enc) {
384             error_setg(errp,
385 "KVM uses SLB encoding 0x%x for page shift %u, but guest expects 0x%x",
386                        ksps->slb_enc, ksps->page_shift, qsps->slb_enc);
387             return;
388         }
389 
390         for (jq = 0; jq < ARRAY_SIZE(qsps->enc); jq++) {
391             for (jk = 0; jk < ARRAY_SIZE(ksps->enc); jk++) {
392                 if (qsps->enc[jq].page_shift == ksps->enc[jk].page_shift) {
393                     break;
394                 }
395             }
396 
397             if (jk >= ARRAY_SIZE(ksps->enc)) {
398                 error_setg(errp, "KVM doesn't support page shift %u/%u",
399                            qsps->enc[jq].page_shift, qsps->page_shift);
400                 return;
401             }
402             if (qsps->enc[jq].pte_enc != ksps->enc[jk].pte_enc) {
403                 error_setg(errp,
404 "KVM uses PTE encoding 0x%x for page shift %u/%u, but guest expects 0x%x",
405                            ksps->enc[jk].pte_enc, qsps->enc[jq].page_shift,
406                            qsps->page_shift, qsps->enc[jq].pte_enc);
407                 return;
408             }
409         }
410     }
411 
412     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
413         /*
414          * Mostly what guest pagesizes we can use are related to the
415          * host pages used to map guest RAM, which is handled in the
416          * platform code. Cache-Inhibited largepages (64k) however are
417          * used for I/O, so if they're mapped to the host at all it
418          * will be a normal mapping, not a special hugepage one used
419          * for RAM.
420          */
421         if (qemu_real_host_page_size() < 0x10000) {
422             error_setg(errp,
423                        "KVM can't supply 64kiB CI pages, which guest expects");
424         }
425     }
426 }
427 #endif /* !defined (TARGET_PPC64) */
428 
429 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
430 {
431     return POWERPC_CPU(cpu)->vcpu_id;
432 }
433 
434 /*
435  * e500 supports 2 h/w breakpoint and 2 watchpoint.  book3s supports
436  * only 1 watchpoint, so array size of 4 is sufficient for now.
437  */
438 #define MAX_HW_BKPTS 4
439 
440 static struct HWBreakpoint {
441     target_ulong addr;
442     int type;
443 } hw_debug_points[MAX_HW_BKPTS];
444 
445 static CPUWatchpoint hw_watchpoint;
446 
447 /* Default there is no breakpoint and watchpoint supported */
448 static int max_hw_breakpoint;
449 static int max_hw_watchpoint;
450 static int nb_hw_breakpoint;
451 static int nb_hw_watchpoint;
452 
453 static void kvmppc_hw_debug_points_init(CPUPPCState *cenv)
454 {
455     if (cenv->excp_model == POWERPC_EXCP_BOOKE) {
456         max_hw_breakpoint = 2;
457         max_hw_watchpoint = 2;
458     }
459 
460     if ((max_hw_breakpoint + max_hw_watchpoint) > MAX_HW_BKPTS) {
461         fprintf(stderr, "Error initializing h/w breakpoints\n");
462         return;
463     }
464 }
465 
466 int kvm_arch_init_vcpu(CPUState *cs)
467 {
468     PowerPCCPU *cpu = POWERPC_CPU(cs);
469     CPUPPCState *cenv = &cpu->env;
470     int ret;
471 
472     /* Synchronize sregs with kvm */
473     ret = kvm_arch_sync_sregs(cpu);
474     if (ret) {
475         if (ret == -EINVAL) {
476             error_report("Register sync failed... If you're using kvm-hv.ko,"
477                          " only \"-cpu host\" is possible");
478         }
479         return ret;
480     }
481 
482     switch (cenv->mmu_model) {
483     case POWERPC_MMU_BOOKE206:
484         /* This target supports access to KVM's guest TLB */
485         ret = kvm_booke206_tlb_init(cpu);
486         break;
487     case POWERPC_MMU_2_07:
488         if (!cap_htm && !kvmppc_is_pr(cs->kvm_state)) {
489             /*
490              * KVM-HV has transactional memory on POWER8 also without
491              * the KVM_CAP_PPC_HTM extension, so enable it here
492              * instead as long as it's available to userspace on the
493              * host.
494              */
495             if (qemu_getauxval(AT_HWCAP2) & PPC_FEATURE2_HAS_HTM) {
496                 cap_htm = true;
497             }
498         }
499         break;
500     default:
501         break;
502     }
503 
504     kvm_get_one_reg(cs, KVM_REG_PPC_DEBUG_INST, &debug_inst_opcode);
505     kvmppc_hw_debug_points_init(cenv);
506 
507     return ret;
508 }
509 
510 int kvm_arch_destroy_vcpu(CPUState *cs)
511 {
512     return 0;
513 }
514 
515 static void kvm_sw_tlb_put(PowerPCCPU *cpu)
516 {
517     CPUPPCState *env = &cpu->env;
518     CPUState *cs = CPU(cpu);
519     struct kvm_dirty_tlb dirty_tlb;
520     unsigned char *bitmap;
521     int ret;
522 
523     if (!env->kvm_sw_tlb) {
524         return;
525     }
526 
527     bitmap = g_malloc((env->nb_tlb + 7) / 8);
528     memset(bitmap, 0xFF, (env->nb_tlb + 7) / 8);
529 
530     dirty_tlb.bitmap = (uintptr_t)bitmap;
531     dirty_tlb.num_dirty = env->nb_tlb;
532 
533     ret = kvm_vcpu_ioctl(cs, KVM_DIRTY_TLB, &dirty_tlb);
534     if (ret) {
535         fprintf(stderr, "%s: KVM_DIRTY_TLB: %s\n",
536                 __func__, strerror(-ret));
537     }
538 
539     g_free(bitmap);
540 }
541 
542 static void kvm_get_one_spr(CPUState *cs, uint64_t id, int spr)
543 {
544     PowerPCCPU *cpu = POWERPC_CPU(cs);
545     CPUPPCState *env = &cpu->env;
546     /* Init 'val' to avoid "uninitialised value" Valgrind warnings */
547     union {
548         uint32_t u32;
549         uint64_t u64;
550     } val = { };
551     struct kvm_one_reg reg = {
552         .id = id,
553         .addr = (uintptr_t) &val,
554     };
555     int ret;
556 
557     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
558     if (ret != 0) {
559         trace_kvm_failed_spr_get(spr, strerror(errno));
560     } else {
561         switch (id & KVM_REG_SIZE_MASK) {
562         case KVM_REG_SIZE_U32:
563             env->spr[spr] = val.u32;
564             break;
565 
566         case KVM_REG_SIZE_U64:
567             env->spr[spr] = val.u64;
568             break;
569 
570         default:
571             /* Don't handle this size yet */
572             abort();
573         }
574     }
575 }
576 
577 static void kvm_put_one_spr(CPUState *cs, uint64_t id, int spr)
578 {
579     PowerPCCPU *cpu = POWERPC_CPU(cs);
580     CPUPPCState *env = &cpu->env;
581     union {
582         uint32_t u32;
583         uint64_t u64;
584     } val;
585     struct kvm_one_reg reg = {
586         .id = id,
587         .addr = (uintptr_t) &val,
588     };
589     int ret;
590 
591     switch (id & KVM_REG_SIZE_MASK) {
592     case KVM_REG_SIZE_U32:
593         val.u32 = env->spr[spr];
594         break;
595 
596     case KVM_REG_SIZE_U64:
597         val.u64 = env->spr[spr];
598         break;
599 
600     default:
601         /* Don't handle this size yet */
602         abort();
603     }
604 
605     ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
606     if (ret != 0) {
607         trace_kvm_failed_spr_set(spr, strerror(errno));
608     }
609 }
610 
611 static int kvm_put_fp(CPUState *cs)
612 {
613     PowerPCCPU *cpu = POWERPC_CPU(cs);
614     CPUPPCState *env = &cpu->env;
615     struct kvm_one_reg reg;
616     int i;
617     int ret;
618 
619     if (env->insns_flags & PPC_FLOAT) {
620         uint64_t fpscr = env->fpscr;
621         bool vsx = !!(env->insns_flags2 & PPC2_VSX);
622 
623         reg.id = KVM_REG_PPC_FPSCR;
624         reg.addr = (uintptr_t)&fpscr;
625         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
626         if (ret < 0) {
627             trace_kvm_failed_fpscr_set(strerror(errno));
628             return ret;
629         }
630 
631         for (i = 0; i < 32; i++) {
632             uint64_t vsr[2];
633             uint64_t *fpr = cpu_fpr_ptr(&cpu->env, i);
634             uint64_t *vsrl = cpu_vsrl_ptr(&cpu->env, i);
635 
636 #if HOST_BIG_ENDIAN
637             vsr[0] = float64_val(*fpr);
638             vsr[1] = *vsrl;
639 #else
640             vsr[0] = *vsrl;
641             vsr[1] = float64_val(*fpr);
642 #endif
643             reg.addr = (uintptr_t) &vsr;
644             reg.id = vsx ? KVM_REG_PPC_VSR(i) : KVM_REG_PPC_FPR(i);
645 
646             ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
647             if (ret < 0) {
648                 trace_kvm_failed_fp_set(vsx ? "VSR" : "FPR", i,
649                                         strerror(errno));
650                 return ret;
651             }
652         }
653     }
654 
655     if (env->insns_flags & PPC_ALTIVEC) {
656         reg.id = KVM_REG_PPC_VSCR;
657         reg.addr = (uintptr_t)&env->vscr;
658         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
659         if (ret < 0) {
660             trace_kvm_failed_vscr_set(strerror(errno));
661             return ret;
662         }
663 
664         for (i = 0; i < 32; i++) {
665             reg.id = KVM_REG_PPC_VR(i);
666             reg.addr = (uintptr_t)cpu_avr_ptr(env, i);
667             ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
668             if (ret < 0) {
669                 trace_kvm_failed_vr_set(i, strerror(errno));
670                 return ret;
671             }
672         }
673     }
674 
675     return 0;
676 }
677 
678 static int kvm_get_fp(CPUState *cs)
679 {
680     PowerPCCPU *cpu = POWERPC_CPU(cs);
681     CPUPPCState *env = &cpu->env;
682     struct kvm_one_reg reg;
683     int i;
684     int ret;
685 
686     if (env->insns_flags & PPC_FLOAT) {
687         uint64_t fpscr;
688         bool vsx = !!(env->insns_flags2 & PPC2_VSX);
689 
690         reg.id = KVM_REG_PPC_FPSCR;
691         reg.addr = (uintptr_t)&fpscr;
692         ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
693         if (ret < 0) {
694             trace_kvm_failed_fpscr_get(strerror(errno));
695             return ret;
696         } else {
697             env->fpscr = fpscr;
698         }
699 
700         for (i = 0; i < 32; i++) {
701             uint64_t vsr[2];
702             uint64_t *fpr = cpu_fpr_ptr(&cpu->env, i);
703             uint64_t *vsrl = cpu_vsrl_ptr(&cpu->env, i);
704 
705             reg.addr = (uintptr_t) &vsr;
706             reg.id = vsx ? KVM_REG_PPC_VSR(i) : KVM_REG_PPC_FPR(i);
707 
708             ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
709             if (ret < 0) {
710                 trace_kvm_failed_fp_get(vsx ? "VSR" : "FPR", i,
711                                         strerror(errno));
712                 return ret;
713             } else {
714 #if HOST_BIG_ENDIAN
715                 *fpr = vsr[0];
716                 if (vsx) {
717                     *vsrl = vsr[1];
718                 }
719 #else
720                 *fpr = vsr[1];
721                 if (vsx) {
722                     *vsrl = vsr[0];
723                 }
724 #endif
725             }
726         }
727     }
728 
729     if (env->insns_flags & PPC_ALTIVEC) {
730         reg.id = KVM_REG_PPC_VSCR;
731         reg.addr = (uintptr_t)&env->vscr;
732         ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
733         if (ret < 0) {
734             trace_kvm_failed_vscr_get(strerror(errno));
735             return ret;
736         }
737 
738         for (i = 0; i < 32; i++) {
739             reg.id = KVM_REG_PPC_VR(i);
740             reg.addr = (uintptr_t)cpu_avr_ptr(env, i);
741             ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
742             if (ret < 0) {
743                 trace_kvm_failed_vr_get(i, strerror(errno));
744                 return ret;
745             }
746         }
747     }
748 
749     return 0;
750 }
751 
752 #if defined(TARGET_PPC64)
753 static int kvm_get_vpa(CPUState *cs)
754 {
755     PowerPCCPU *cpu = POWERPC_CPU(cs);
756     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
757     struct kvm_one_reg reg;
758     int ret;
759 
760     reg.id = KVM_REG_PPC_VPA_ADDR;
761     reg.addr = (uintptr_t)&spapr_cpu->vpa_addr;
762     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
763     if (ret < 0) {
764         trace_kvm_failed_vpa_addr_get(strerror(errno));
765         return ret;
766     }
767 
768     assert((uintptr_t)&spapr_cpu->slb_shadow_size
769            == ((uintptr_t)&spapr_cpu->slb_shadow_addr + 8));
770     reg.id = KVM_REG_PPC_VPA_SLB;
771     reg.addr = (uintptr_t)&spapr_cpu->slb_shadow_addr;
772     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
773     if (ret < 0) {
774         trace_kvm_failed_slb_get(strerror(errno));
775         return ret;
776     }
777 
778     assert((uintptr_t)&spapr_cpu->dtl_size
779            == ((uintptr_t)&spapr_cpu->dtl_addr + 8));
780     reg.id = KVM_REG_PPC_VPA_DTL;
781     reg.addr = (uintptr_t)&spapr_cpu->dtl_addr;
782     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
783     if (ret < 0) {
784         trace_kvm_failed_dtl_get(strerror(errno));
785         return ret;
786     }
787 
788     return 0;
789 }
790 
791 static int kvm_put_vpa(CPUState *cs)
792 {
793     PowerPCCPU *cpu = POWERPC_CPU(cs);
794     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
795     struct kvm_one_reg reg;
796     int ret;
797 
798     /*
799      * SLB shadow or DTL can't be registered unless a master VPA is
800      * registered.  That means when restoring state, if a VPA *is*
801      * registered, we need to set that up first.  If not, we need to
802      * deregister the others before deregistering the master VPA
803      */
804     assert(spapr_cpu->vpa_addr
805            || !(spapr_cpu->slb_shadow_addr || spapr_cpu->dtl_addr));
806 
807     if (spapr_cpu->vpa_addr) {
808         reg.id = KVM_REG_PPC_VPA_ADDR;
809         reg.addr = (uintptr_t)&spapr_cpu->vpa_addr;
810         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
811         if (ret < 0) {
812             trace_kvm_failed_vpa_addr_set(strerror(errno));
813             return ret;
814         }
815     }
816 
817     assert((uintptr_t)&spapr_cpu->slb_shadow_size
818            == ((uintptr_t)&spapr_cpu->slb_shadow_addr + 8));
819     reg.id = KVM_REG_PPC_VPA_SLB;
820     reg.addr = (uintptr_t)&spapr_cpu->slb_shadow_addr;
821     ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
822     if (ret < 0) {
823         trace_kvm_failed_slb_set(strerror(errno));
824         return ret;
825     }
826 
827     assert((uintptr_t)&spapr_cpu->dtl_size
828            == ((uintptr_t)&spapr_cpu->dtl_addr + 8));
829     reg.id = KVM_REG_PPC_VPA_DTL;
830     reg.addr = (uintptr_t)&spapr_cpu->dtl_addr;
831     ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
832     if (ret < 0) {
833         trace_kvm_failed_dtl_set(strerror(errno));
834         return ret;
835     }
836 
837     if (!spapr_cpu->vpa_addr) {
838         reg.id = KVM_REG_PPC_VPA_ADDR;
839         reg.addr = (uintptr_t)&spapr_cpu->vpa_addr;
840         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
841         if (ret < 0) {
842             trace_kvm_failed_null_vpa_addr_set(strerror(errno));
843             return ret;
844         }
845     }
846 
847     return 0;
848 }
849 #endif /* TARGET_PPC64 */
850 
851 int kvmppc_put_books_sregs(PowerPCCPU *cpu)
852 {
853     CPUPPCState *env = &cpu->env;
854     struct kvm_sregs sregs = { };
855     int i;
856 
857     sregs.pvr = env->spr[SPR_PVR];
858 
859     if (cpu->vhyp) {
860         PPCVirtualHypervisorClass *vhc =
861             PPC_VIRTUAL_HYPERVISOR_GET_CLASS(cpu->vhyp);
862         sregs.u.s.sdr1 = vhc->encode_hpt_for_kvm_pr(cpu->vhyp);
863     } else {
864         sregs.u.s.sdr1 = env->spr[SPR_SDR1];
865     }
866 
867     /* Sync SLB */
868 #ifdef TARGET_PPC64
869     for (i = 0; i < ARRAY_SIZE(env->slb); i++) {
870         sregs.u.s.ppc64.slb[i].slbe = env->slb[i].esid;
871         if (env->slb[i].esid & SLB_ESID_V) {
872             sregs.u.s.ppc64.slb[i].slbe |= i;
873         }
874         sregs.u.s.ppc64.slb[i].slbv = env->slb[i].vsid;
875     }
876 #endif
877 
878     /* Sync SRs */
879     for (i = 0; i < 16; i++) {
880         sregs.u.s.ppc32.sr[i] = env->sr[i];
881     }
882 
883     /* Sync BATs */
884     for (i = 0; i < 8; i++) {
885         /* Beware. We have to swap upper and lower bits here */
886         sregs.u.s.ppc32.dbat[i] = ((uint64_t)env->DBAT[0][i] << 32)
887             | env->DBAT[1][i];
888         sregs.u.s.ppc32.ibat[i] = ((uint64_t)env->IBAT[0][i] << 32)
889             | env->IBAT[1][i];
890     }
891 
892     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_SREGS, &sregs);
893 }
894 
895 int kvm_arch_put_registers(CPUState *cs, int level)
896 {
897     PowerPCCPU *cpu = POWERPC_CPU(cs);
898     CPUPPCState *env = &cpu->env;
899     struct kvm_regs regs;
900     int ret;
901     int i;
902 
903     ret = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
904     if (ret < 0) {
905         return ret;
906     }
907 
908     regs.ctr = env->ctr;
909     regs.lr  = env->lr;
910     regs.xer = cpu_read_xer(env);
911     regs.msr = env->msr;
912     regs.pc = env->nip;
913 
914     regs.srr0 = env->spr[SPR_SRR0];
915     regs.srr1 = env->spr[SPR_SRR1];
916 
917     regs.sprg0 = env->spr[SPR_SPRG0];
918     regs.sprg1 = env->spr[SPR_SPRG1];
919     regs.sprg2 = env->spr[SPR_SPRG2];
920     regs.sprg3 = env->spr[SPR_SPRG3];
921     regs.sprg4 = env->spr[SPR_SPRG4];
922     regs.sprg5 = env->spr[SPR_SPRG5];
923     regs.sprg6 = env->spr[SPR_SPRG6];
924     regs.sprg7 = env->spr[SPR_SPRG7];
925 
926     regs.pid = env->spr[SPR_BOOKE_PID];
927 
928     for (i = 0; i < 32; i++) {
929         regs.gpr[i] = env->gpr[i];
930     }
931 
932     regs.cr = ppc_get_cr(env);
933 
934     ret = kvm_vcpu_ioctl(cs, KVM_SET_REGS, &regs);
935     if (ret < 0) {
936         return ret;
937     }
938 
939     kvm_put_fp(cs);
940 
941     if (env->tlb_dirty) {
942         kvm_sw_tlb_put(cpu);
943         env->tlb_dirty = false;
944     }
945 
946     if (cap_segstate && (level >= KVM_PUT_RESET_STATE)) {
947         ret = kvmppc_put_books_sregs(cpu);
948         if (ret < 0) {
949             return ret;
950         }
951     }
952 
953     if (cap_hior && (level >= KVM_PUT_RESET_STATE)) {
954         kvm_put_one_spr(cs, KVM_REG_PPC_HIOR, SPR_HIOR);
955     }
956 
957     if (cap_one_reg) {
958         int i;
959 
960         /*
961          * We deliberately ignore errors here, for kernels which have
962          * the ONE_REG calls, but don't support the specific
963          * registers, there's a reasonable chance things will still
964          * work, at least until we try to migrate.
965          */
966         for (i = 0; i < 1024; i++) {
967             uint64_t id = env->spr_cb[i].one_reg_id;
968 
969             if (id != 0) {
970                 kvm_put_one_spr(cs, id, i);
971             }
972         }
973 
974 #ifdef TARGET_PPC64
975         if (FIELD_EX64(env->msr, MSR, TS)) {
976             for (i = 0; i < ARRAY_SIZE(env->tm_gpr); i++) {
977                 kvm_set_one_reg(cs, KVM_REG_PPC_TM_GPR(i), &env->tm_gpr[i]);
978             }
979             for (i = 0; i < ARRAY_SIZE(env->tm_vsr); i++) {
980                 kvm_set_one_reg(cs, KVM_REG_PPC_TM_VSR(i), &env->tm_vsr[i]);
981             }
982             kvm_set_one_reg(cs, KVM_REG_PPC_TM_CR, &env->tm_cr);
983             kvm_set_one_reg(cs, KVM_REG_PPC_TM_LR, &env->tm_lr);
984             kvm_set_one_reg(cs, KVM_REG_PPC_TM_CTR, &env->tm_ctr);
985             kvm_set_one_reg(cs, KVM_REG_PPC_TM_FPSCR, &env->tm_fpscr);
986             kvm_set_one_reg(cs, KVM_REG_PPC_TM_AMR, &env->tm_amr);
987             kvm_set_one_reg(cs, KVM_REG_PPC_TM_PPR, &env->tm_ppr);
988             kvm_set_one_reg(cs, KVM_REG_PPC_TM_VRSAVE, &env->tm_vrsave);
989             kvm_set_one_reg(cs, KVM_REG_PPC_TM_VSCR, &env->tm_vscr);
990             kvm_set_one_reg(cs, KVM_REG_PPC_TM_DSCR, &env->tm_dscr);
991             kvm_set_one_reg(cs, KVM_REG_PPC_TM_TAR, &env->tm_tar);
992         }
993 
994         if (cap_papr) {
995             if (kvm_put_vpa(cs) < 0) {
996                 trace_kvm_failed_put_vpa();
997             }
998         }
999 
1000         kvm_set_one_reg(cs, KVM_REG_PPC_TB_OFFSET, &env->tb_env->tb_offset);
1001 
1002         if (level > KVM_PUT_RUNTIME_STATE) {
1003             kvm_put_one_spr(cs, KVM_REG_PPC_DPDES, SPR_DPDES);
1004         }
1005 #endif /* TARGET_PPC64 */
1006     }
1007 
1008     return ret;
1009 }
1010 
1011 static void kvm_sync_excp(CPUPPCState *env, int vector, int ivor)
1012 {
1013      env->excp_vectors[vector] = env->spr[ivor] + env->spr[SPR_BOOKE_IVPR];
1014 }
1015 
1016 static int kvmppc_get_booke_sregs(PowerPCCPU *cpu)
1017 {
1018     CPUPPCState *env = &cpu->env;
1019     struct kvm_sregs sregs;
1020     int ret;
1021 
1022     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_SREGS, &sregs);
1023     if (ret < 0) {
1024         return ret;
1025     }
1026 
1027     if (sregs.u.e.features & KVM_SREGS_E_BASE) {
1028         env->spr[SPR_BOOKE_CSRR0] = sregs.u.e.csrr0;
1029         env->spr[SPR_BOOKE_CSRR1] = sregs.u.e.csrr1;
1030         env->spr[SPR_BOOKE_ESR] = sregs.u.e.esr;
1031         env->spr[SPR_BOOKE_DEAR] = sregs.u.e.dear;
1032         env->spr[SPR_BOOKE_MCSR] = sregs.u.e.mcsr;
1033         env->spr[SPR_BOOKE_TSR] = sregs.u.e.tsr;
1034         env->spr[SPR_BOOKE_TCR] = sregs.u.e.tcr;
1035         env->spr[SPR_DECR] = sregs.u.e.dec;
1036         env->spr[SPR_TBL] = sregs.u.e.tb & 0xffffffff;
1037         env->spr[SPR_TBU] = sregs.u.e.tb >> 32;
1038         env->spr[SPR_VRSAVE] = sregs.u.e.vrsave;
1039     }
1040 
1041     if (sregs.u.e.features & KVM_SREGS_E_ARCH206) {
1042         env->spr[SPR_BOOKE_PIR] = sregs.u.e.pir;
1043         env->spr[SPR_BOOKE_MCSRR0] = sregs.u.e.mcsrr0;
1044         env->spr[SPR_BOOKE_MCSRR1] = sregs.u.e.mcsrr1;
1045         env->spr[SPR_BOOKE_DECAR] = sregs.u.e.decar;
1046         env->spr[SPR_BOOKE_IVPR] = sregs.u.e.ivpr;
1047     }
1048 
1049     if (sregs.u.e.features & KVM_SREGS_E_64) {
1050         env->spr[SPR_BOOKE_EPCR] = sregs.u.e.epcr;
1051     }
1052 
1053     if (sregs.u.e.features & KVM_SREGS_E_SPRG8) {
1054         env->spr[SPR_BOOKE_SPRG8] = sregs.u.e.sprg8;
1055     }
1056 
1057     if (sregs.u.e.features & KVM_SREGS_E_IVOR) {
1058         env->spr[SPR_BOOKE_IVOR0] = sregs.u.e.ivor_low[0];
1059         kvm_sync_excp(env, POWERPC_EXCP_CRITICAL,  SPR_BOOKE_IVOR0);
1060         env->spr[SPR_BOOKE_IVOR1] = sregs.u.e.ivor_low[1];
1061         kvm_sync_excp(env, POWERPC_EXCP_MCHECK,  SPR_BOOKE_IVOR1);
1062         env->spr[SPR_BOOKE_IVOR2] = sregs.u.e.ivor_low[2];
1063         kvm_sync_excp(env, POWERPC_EXCP_DSI,  SPR_BOOKE_IVOR2);
1064         env->spr[SPR_BOOKE_IVOR3] = sregs.u.e.ivor_low[3];
1065         kvm_sync_excp(env, POWERPC_EXCP_ISI,  SPR_BOOKE_IVOR3);
1066         env->spr[SPR_BOOKE_IVOR4] = sregs.u.e.ivor_low[4];
1067         kvm_sync_excp(env, POWERPC_EXCP_EXTERNAL,  SPR_BOOKE_IVOR4);
1068         env->spr[SPR_BOOKE_IVOR5] = sregs.u.e.ivor_low[5];
1069         kvm_sync_excp(env, POWERPC_EXCP_ALIGN,  SPR_BOOKE_IVOR5);
1070         env->spr[SPR_BOOKE_IVOR6] = sregs.u.e.ivor_low[6];
1071         kvm_sync_excp(env, POWERPC_EXCP_PROGRAM,  SPR_BOOKE_IVOR6);
1072         env->spr[SPR_BOOKE_IVOR7] = sregs.u.e.ivor_low[7];
1073         kvm_sync_excp(env, POWERPC_EXCP_FPU,  SPR_BOOKE_IVOR7);
1074         env->spr[SPR_BOOKE_IVOR8] = sregs.u.e.ivor_low[8];
1075         kvm_sync_excp(env, POWERPC_EXCP_SYSCALL,  SPR_BOOKE_IVOR8);
1076         env->spr[SPR_BOOKE_IVOR9] = sregs.u.e.ivor_low[9];
1077         kvm_sync_excp(env, POWERPC_EXCP_APU,  SPR_BOOKE_IVOR9);
1078         env->spr[SPR_BOOKE_IVOR10] = sregs.u.e.ivor_low[10];
1079         kvm_sync_excp(env, POWERPC_EXCP_DECR,  SPR_BOOKE_IVOR10);
1080         env->spr[SPR_BOOKE_IVOR11] = sregs.u.e.ivor_low[11];
1081         kvm_sync_excp(env, POWERPC_EXCP_FIT,  SPR_BOOKE_IVOR11);
1082         env->spr[SPR_BOOKE_IVOR12] = sregs.u.e.ivor_low[12];
1083         kvm_sync_excp(env, POWERPC_EXCP_WDT,  SPR_BOOKE_IVOR12);
1084         env->spr[SPR_BOOKE_IVOR13] = sregs.u.e.ivor_low[13];
1085         kvm_sync_excp(env, POWERPC_EXCP_DTLB,  SPR_BOOKE_IVOR13);
1086         env->spr[SPR_BOOKE_IVOR14] = sregs.u.e.ivor_low[14];
1087         kvm_sync_excp(env, POWERPC_EXCP_ITLB,  SPR_BOOKE_IVOR14);
1088         env->spr[SPR_BOOKE_IVOR15] = sregs.u.e.ivor_low[15];
1089         kvm_sync_excp(env, POWERPC_EXCP_DEBUG,  SPR_BOOKE_IVOR15);
1090 
1091         if (sregs.u.e.features & KVM_SREGS_E_SPE) {
1092             env->spr[SPR_BOOKE_IVOR32] = sregs.u.e.ivor_high[0];
1093             kvm_sync_excp(env, POWERPC_EXCP_SPEU,  SPR_BOOKE_IVOR32);
1094             env->spr[SPR_BOOKE_IVOR33] = sregs.u.e.ivor_high[1];
1095             kvm_sync_excp(env, POWERPC_EXCP_EFPDI,  SPR_BOOKE_IVOR33);
1096             env->spr[SPR_BOOKE_IVOR34] = sregs.u.e.ivor_high[2];
1097             kvm_sync_excp(env, POWERPC_EXCP_EFPRI,  SPR_BOOKE_IVOR34);
1098         }
1099 
1100         if (sregs.u.e.features & KVM_SREGS_E_PM) {
1101             env->spr[SPR_BOOKE_IVOR35] = sregs.u.e.ivor_high[3];
1102             kvm_sync_excp(env, POWERPC_EXCP_EPERFM,  SPR_BOOKE_IVOR35);
1103         }
1104 
1105         if (sregs.u.e.features & KVM_SREGS_E_PC) {
1106             env->spr[SPR_BOOKE_IVOR36] = sregs.u.e.ivor_high[4];
1107             kvm_sync_excp(env, POWERPC_EXCP_DOORI,  SPR_BOOKE_IVOR36);
1108             env->spr[SPR_BOOKE_IVOR37] = sregs.u.e.ivor_high[5];
1109             kvm_sync_excp(env, POWERPC_EXCP_DOORCI, SPR_BOOKE_IVOR37);
1110         }
1111     }
1112 
1113     if (sregs.u.e.features & KVM_SREGS_E_ARCH206_MMU) {
1114         env->spr[SPR_BOOKE_MAS0] = sregs.u.e.mas0;
1115         env->spr[SPR_BOOKE_MAS1] = sregs.u.e.mas1;
1116         env->spr[SPR_BOOKE_MAS2] = sregs.u.e.mas2;
1117         env->spr[SPR_BOOKE_MAS3] = sregs.u.e.mas7_3 & 0xffffffff;
1118         env->spr[SPR_BOOKE_MAS4] = sregs.u.e.mas4;
1119         env->spr[SPR_BOOKE_MAS6] = sregs.u.e.mas6;
1120         env->spr[SPR_BOOKE_MAS7] = sregs.u.e.mas7_3 >> 32;
1121         env->spr[SPR_MMUCFG] = sregs.u.e.mmucfg;
1122         env->spr[SPR_BOOKE_TLB0CFG] = sregs.u.e.tlbcfg[0];
1123         env->spr[SPR_BOOKE_TLB1CFG] = sregs.u.e.tlbcfg[1];
1124     }
1125 
1126     if (sregs.u.e.features & KVM_SREGS_EXP) {
1127         env->spr[SPR_BOOKE_EPR] = sregs.u.e.epr;
1128     }
1129 
1130     if (sregs.u.e.features & KVM_SREGS_E_PD) {
1131         env->spr[SPR_BOOKE_EPLC] = sregs.u.e.eplc;
1132         env->spr[SPR_BOOKE_EPSC] = sregs.u.e.epsc;
1133     }
1134 
1135     if (sregs.u.e.impl_id == KVM_SREGS_E_IMPL_FSL) {
1136         env->spr[SPR_E500_SVR] = sregs.u.e.impl.fsl.svr;
1137         env->spr[SPR_Exxx_MCAR] = sregs.u.e.impl.fsl.mcar;
1138         env->spr[SPR_HID0] = sregs.u.e.impl.fsl.hid0;
1139 
1140         if (sregs.u.e.impl.fsl.features & KVM_SREGS_E_FSL_PIDn) {
1141             env->spr[SPR_BOOKE_PID1] = sregs.u.e.impl.fsl.pid1;
1142             env->spr[SPR_BOOKE_PID2] = sregs.u.e.impl.fsl.pid2;
1143         }
1144     }
1145 
1146     return 0;
1147 }
1148 
1149 static int kvmppc_get_books_sregs(PowerPCCPU *cpu)
1150 {
1151     CPUPPCState *env = &cpu->env;
1152     struct kvm_sregs sregs;
1153     int ret;
1154     int i;
1155 
1156     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_SREGS, &sregs);
1157     if (ret < 0) {
1158         return ret;
1159     }
1160 
1161     if (!cpu->vhyp) {
1162         ppc_store_sdr1(env, sregs.u.s.sdr1);
1163     }
1164 
1165     /* Sync SLB */
1166 #ifdef TARGET_PPC64
1167     /*
1168      * The packed SLB array we get from KVM_GET_SREGS only contains
1169      * information about valid entries. So we flush our internal copy
1170      * to get rid of stale ones, then put all valid SLB entries back
1171      * in.
1172      */
1173     memset(env->slb, 0, sizeof(env->slb));
1174     for (i = 0; i < ARRAY_SIZE(env->slb); i++) {
1175         target_ulong rb = sregs.u.s.ppc64.slb[i].slbe;
1176         target_ulong rs = sregs.u.s.ppc64.slb[i].slbv;
1177         /*
1178          * Only restore valid entries
1179          */
1180         if (rb & SLB_ESID_V) {
1181             ppc_store_slb(cpu, rb & 0xfff, rb & ~0xfffULL, rs);
1182         }
1183     }
1184 #endif
1185 
1186     /* Sync SRs */
1187     for (i = 0; i < 16; i++) {
1188         env->sr[i] = sregs.u.s.ppc32.sr[i];
1189     }
1190 
1191     /* Sync BATs */
1192     for (i = 0; i < 8; i++) {
1193         env->DBAT[0][i] = sregs.u.s.ppc32.dbat[i] & 0xffffffff;
1194         env->DBAT[1][i] = sregs.u.s.ppc32.dbat[i] >> 32;
1195         env->IBAT[0][i] = sregs.u.s.ppc32.ibat[i] & 0xffffffff;
1196         env->IBAT[1][i] = sregs.u.s.ppc32.ibat[i] >> 32;
1197     }
1198 
1199     return 0;
1200 }
1201 
1202 int kvm_arch_get_registers(CPUState *cs)
1203 {
1204     PowerPCCPU *cpu = POWERPC_CPU(cs);
1205     CPUPPCState *env = &cpu->env;
1206     struct kvm_regs regs;
1207     int i, ret;
1208 
1209     ret = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
1210     if (ret < 0) {
1211         return ret;
1212     }
1213 
1214     ppc_set_cr(env, regs.cr);
1215     env->ctr = regs.ctr;
1216     env->lr = regs.lr;
1217     cpu_write_xer(env, regs.xer);
1218     env->msr = regs.msr;
1219     env->nip = regs.pc;
1220 
1221     env->spr[SPR_SRR0] = regs.srr0;
1222     env->spr[SPR_SRR1] = regs.srr1;
1223 
1224     env->spr[SPR_SPRG0] = regs.sprg0;
1225     env->spr[SPR_SPRG1] = regs.sprg1;
1226     env->spr[SPR_SPRG2] = regs.sprg2;
1227     env->spr[SPR_SPRG3] = regs.sprg3;
1228     env->spr[SPR_SPRG4] = regs.sprg4;
1229     env->spr[SPR_SPRG5] = regs.sprg5;
1230     env->spr[SPR_SPRG6] = regs.sprg6;
1231     env->spr[SPR_SPRG7] = regs.sprg7;
1232 
1233     env->spr[SPR_BOOKE_PID] = regs.pid;
1234 
1235     for (i = 0; i < 32; i++) {
1236         env->gpr[i] = regs.gpr[i];
1237     }
1238 
1239     kvm_get_fp(cs);
1240 
1241     if (cap_booke_sregs) {
1242         ret = kvmppc_get_booke_sregs(cpu);
1243         if (ret < 0) {
1244             return ret;
1245         }
1246     }
1247 
1248     if (cap_segstate) {
1249         ret = kvmppc_get_books_sregs(cpu);
1250         if (ret < 0) {
1251             return ret;
1252         }
1253     }
1254 
1255     if (cap_hior) {
1256         kvm_get_one_spr(cs, KVM_REG_PPC_HIOR, SPR_HIOR);
1257     }
1258 
1259     if (cap_one_reg) {
1260         int i;
1261 
1262         /*
1263          * We deliberately ignore errors here, for kernels which have
1264          * the ONE_REG calls, but don't support the specific
1265          * registers, there's a reasonable chance things will still
1266          * work, at least until we try to migrate.
1267          */
1268         for (i = 0; i < 1024; i++) {
1269             uint64_t id = env->spr_cb[i].one_reg_id;
1270 
1271             if (id != 0) {
1272                 kvm_get_one_spr(cs, id, i);
1273             }
1274         }
1275 
1276 #ifdef TARGET_PPC64
1277         if (FIELD_EX64(env->msr, MSR, TS)) {
1278             for (i = 0; i < ARRAY_SIZE(env->tm_gpr); i++) {
1279                 kvm_get_one_reg(cs, KVM_REG_PPC_TM_GPR(i), &env->tm_gpr[i]);
1280             }
1281             for (i = 0; i < ARRAY_SIZE(env->tm_vsr); i++) {
1282                 kvm_get_one_reg(cs, KVM_REG_PPC_TM_VSR(i), &env->tm_vsr[i]);
1283             }
1284             kvm_get_one_reg(cs, KVM_REG_PPC_TM_CR, &env->tm_cr);
1285             kvm_get_one_reg(cs, KVM_REG_PPC_TM_LR, &env->tm_lr);
1286             kvm_get_one_reg(cs, KVM_REG_PPC_TM_CTR, &env->tm_ctr);
1287             kvm_get_one_reg(cs, KVM_REG_PPC_TM_FPSCR, &env->tm_fpscr);
1288             kvm_get_one_reg(cs, KVM_REG_PPC_TM_AMR, &env->tm_amr);
1289             kvm_get_one_reg(cs, KVM_REG_PPC_TM_PPR, &env->tm_ppr);
1290             kvm_get_one_reg(cs, KVM_REG_PPC_TM_VRSAVE, &env->tm_vrsave);
1291             kvm_get_one_reg(cs, KVM_REG_PPC_TM_VSCR, &env->tm_vscr);
1292             kvm_get_one_reg(cs, KVM_REG_PPC_TM_DSCR, &env->tm_dscr);
1293             kvm_get_one_reg(cs, KVM_REG_PPC_TM_TAR, &env->tm_tar);
1294         }
1295 
1296         if (cap_papr) {
1297             if (kvm_get_vpa(cs) < 0) {
1298                 trace_kvm_failed_get_vpa();
1299             }
1300         }
1301 
1302         kvm_get_one_reg(cs, KVM_REG_PPC_TB_OFFSET, &env->tb_env->tb_offset);
1303         kvm_get_one_spr(cs, KVM_REG_PPC_DPDES, SPR_DPDES);
1304 #endif
1305     }
1306 
1307     return 0;
1308 }
1309 
1310 int kvmppc_set_interrupt(PowerPCCPU *cpu, int irq, int level)
1311 {
1312     unsigned virq = level ? KVM_INTERRUPT_SET_LEVEL : KVM_INTERRUPT_UNSET;
1313 
1314     if (irq != PPC_INTERRUPT_EXT) {
1315         return 0;
1316     }
1317 
1318     if (!kvm_enabled() || !cap_interrupt_unset) {
1319         return 0;
1320     }
1321 
1322     kvm_vcpu_ioctl(CPU(cpu), KVM_INTERRUPT, &virq);
1323 
1324     return 0;
1325 }
1326 
1327 void kvm_arch_pre_run(CPUState *cs, struct kvm_run *run)
1328 {
1329     return;
1330 }
1331 
1332 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
1333 {
1334     return MEMTXATTRS_UNSPECIFIED;
1335 }
1336 
1337 int kvm_arch_process_async_events(CPUState *cs)
1338 {
1339     return cs->halted;
1340 }
1341 
1342 static int kvmppc_handle_halt(PowerPCCPU *cpu)
1343 {
1344     CPUState *cs = CPU(cpu);
1345     CPUPPCState *env = &cpu->env;
1346 
1347     if (!(cs->interrupt_request & CPU_INTERRUPT_HARD) &&
1348         FIELD_EX64(env->msr, MSR, EE)) {
1349         cs->halted = 1;
1350         cs->exception_index = EXCP_HLT;
1351     }
1352 
1353     return 0;
1354 }
1355 
1356 /* map dcr access to existing qemu dcr emulation */
1357 static int kvmppc_handle_dcr_read(CPUPPCState *env,
1358                                   uint32_t dcrn, uint32_t *data)
1359 {
1360     if (ppc_dcr_read(env->dcr_env, dcrn, data) < 0) {
1361         fprintf(stderr, "Read to unhandled DCR (0x%x)\n", dcrn);
1362     }
1363 
1364     return 0;
1365 }
1366 
1367 static int kvmppc_handle_dcr_write(CPUPPCState *env,
1368                                    uint32_t dcrn, uint32_t data)
1369 {
1370     if (ppc_dcr_write(env->dcr_env, dcrn, data) < 0) {
1371         fprintf(stderr, "Write to unhandled DCR (0x%x)\n", dcrn);
1372     }
1373 
1374     return 0;
1375 }
1376 
1377 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
1378 {
1379     /* Mixed endian case is not handled */
1380     uint32_t sc = debug_inst_opcode;
1381 
1382     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
1383                             sizeof(sc), 0) ||
1384         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&sc, sizeof(sc), 1)) {
1385         return -EINVAL;
1386     }
1387 
1388     return 0;
1389 }
1390 
1391 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
1392 {
1393     uint32_t sc;
1394 
1395     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&sc, sizeof(sc), 0) ||
1396         sc != debug_inst_opcode ||
1397         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
1398                             sizeof(sc), 1)) {
1399         return -EINVAL;
1400     }
1401 
1402     return 0;
1403 }
1404 
1405 static int find_hw_breakpoint(target_ulong addr, int type)
1406 {
1407     int n;
1408 
1409     assert((nb_hw_breakpoint + nb_hw_watchpoint)
1410            <= ARRAY_SIZE(hw_debug_points));
1411 
1412     for (n = 0; n < nb_hw_breakpoint + nb_hw_watchpoint; n++) {
1413         if (hw_debug_points[n].addr == addr &&
1414              hw_debug_points[n].type == type) {
1415             return n;
1416         }
1417     }
1418 
1419     return -1;
1420 }
1421 
1422 static int find_hw_watchpoint(target_ulong addr, int *flag)
1423 {
1424     int n;
1425 
1426     n = find_hw_breakpoint(addr, GDB_WATCHPOINT_ACCESS);
1427     if (n >= 0) {
1428         *flag = BP_MEM_ACCESS;
1429         return n;
1430     }
1431 
1432     n = find_hw_breakpoint(addr, GDB_WATCHPOINT_WRITE);
1433     if (n >= 0) {
1434         *flag = BP_MEM_WRITE;
1435         return n;
1436     }
1437 
1438     n = find_hw_breakpoint(addr, GDB_WATCHPOINT_READ);
1439     if (n >= 0) {
1440         *flag = BP_MEM_READ;
1441         return n;
1442     }
1443 
1444     return -1;
1445 }
1446 
1447 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
1448                                   target_ulong len, int type)
1449 {
1450     if ((nb_hw_breakpoint + nb_hw_watchpoint) >= ARRAY_SIZE(hw_debug_points)) {
1451         return -ENOBUFS;
1452     }
1453 
1454     hw_debug_points[nb_hw_breakpoint + nb_hw_watchpoint].addr = addr;
1455     hw_debug_points[nb_hw_breakpoint + nb_hw_watchpoint].type = type;
1456 
1457     switch (type) {
1458     case GDB_BREAKPOINT_HW:
1459         if (nb_hw_breakpoint >= max_hw_breakpoint) {
1460             return -ENOBUFS;
1461         }
1462 
1463         if (find_hw_breakpoint(addr, type) >= 0) {
1464             return -EEXIST;
1465         }
1466 
1467         nb_hw_breakpoint++;
1468         break;
1469 
1470     case GDB_WATCHPOINT_WRITE:
1471     case GDB_WATCHPOINT_READ:
1472     case GDB_WATCHPOINT_ACCESS:
1473         if (nb_hw_watchpoint >= max_hw_watchpoint) {
1474             return -ENOBUFS;
1475         }
1476 
1477         if (find_hw_breakpoint(addr, type) >= 0) {
1478             return -EEXIST;
1479         }
1480 
1481         nb_hw_watchpoint++;
1482         break;
1483 
1484     default:
1485         return -ENOSYS;
1486     }
1487 
1488     return 0;
1489 }
1490 
1491 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
1492                                   target_ulong len, int type)
1493 {
1494     int n;
1495 
1496     n = find_hw_breakpoint(addr, type);
1497     if (n < 0) {
1498         return -ENOENT;
1499     }
1500 
1501     switch (type) {
1502     case GDB_BREAKPOINT_HW:
1503         nb_hw_breakpoint--;
1504         break;
1505 
1506     case GDB_WATCHPOINT_WRITE:
1507     case GDB_WATCHPOINT_READ:
1508     case GDB_WATCHPOINT_ACCESS:
1509         nb_hw_watchpoint--;
1510         break;
1511 
1512     default:
1513         return -ENOSYS;
1514     }
1515     hw_debug_points[n] = hw_debug_points[nb_hw_breakpoint + nb_hw_watchpoint];
1516 
1517     return 0;
1518 }
1519 
1520 void kvm_arch_remove_all_hw_breakpoints(void)
1521 {
1522     nb_hw_breakpoint = nb_hw_watchpoint = 0;
1523 }
1524 
1525 void kvm_arch_update_guest_debug(CPUState *cs, struct kvm_guest_debug *dbg)
1526 {
1527     int n;
1528 
1529     /* Software Breakpoint updates */
1530     if (kvm_sw_breakpoints_active(cs)) {
1531         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP;
1532     }
1533 
1534     assert((nb_hw_breakpoint + nb_hw_watchpoint)
1535            <= ARRAY_SIZE(hw_debug_points));
1536     assert((nb_hw_breakpoint + nb_hw_watchpoint) <= ARRAY_SIZE(dbg->arch.bp));
1537 
1538     if (nb_hw_breakpoint + nb_hw_watchpoint > 0) {
1539         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
1540         memset(dbg->arch.bp, 0, sizeof(dbg->arch.bp));
1541         for (n = 0; n < nb_hw_breakpoint + nb_hw_watchpoint; n++) {
1542             switch (hw_debug_points[n].type) {
1543             case GDB_BREAKPOINT_HW:
1544                 dbg->arch.bp[n].type = KVMPPC_DEBUG_BREAKPOINT;
1545                 break;
1546             case GDB_WATCHPOINT_WRITE:
1547                 dbg->arch.bp[n].type = KVMPPC_DEBUG_WATCH_WRITE;
1548                 break;
1549             case GDB_WATCHPOINT_READ:
1550                 dbg->arch.bp[n].type = KVMPPC_DEBUG_WATCH_READ;
1551                 break;
1552             case GDB_WATCHPOINT_ACCESS:
1553                 dbg->arch.bp[n].type = KVMPPC_DEBUG_WATCH_WRITE |
1554                                         KVMPPC_DEBUG_WATCH_READ;
1555                 break;
1556             default:
1557                 cpu_abort(cs, "Unsupported breakpoint type\n");
1558             }
1559             dbg->arch.bp[n].addr = hw_debug_points[n].addr;
1560         }
1561     }
1562 }
1563 
1564 static int kvm_handle_hw_breakpoint(CPUState *cs,
1565                                     struct kvm_debug_exit_arch *arch_info)
1566 {
1567     int handle = DEBUG_RETURN_GUEST;
1568     int n;
1569     int flag = 0;
1570 
1571     if (nb_hw_breakpoint + nb_hw_watchpoint > 0) {
1572         if (arch_info->status & KVMPPC_DEBUG_BREAKPOINT) {
1573             n = find_hw_breakpoint(arch_info->address, GDB_BREAKPOINT_HW);
1574             if (n >= 0) {
1575                 handle = DEBUG_RETURN_GDB;
1576             }
1577         } else if (arch_info->status & (KVMPPC_DEBUG_WATCH_READ |
1578                                         KVMPPC_DEBUG_WATCH_WRITE)) {
1579             n = find_hw_watchpoint(arch_info->address,  &flag);
1580             if (n >= 0) {
1581                 handle = DEBUG_RETURN_GDB;
1582                 cs->watchpoint_hit = &hw_watchpoint;
1583                 hw_watchpoint.vaddr = hw_debug_points[n].addr;
1584                 hw_watchpoint.flags = flag;
1585             }
1586         }
1587     }
1588     return handle;
1589 }
1590 
1591 static int kvm_handle_singlestep(void)
1592 {
1593     return DEBUG_RETURN_GDB;
1594 }
1595 
1596 static int kvm_handle_sw_breakpoint(void)
1597 {
1598     return DEBUG_RETURN_GDB;
1599 }
1600 
1601 static int kvm_handle_debug(PowerPCCPU *cpu, struct kvm_run *run)
1602 {
1603     CPUState *cs = CPU(cpu);
1604     CPUPPCState *env = &cpu->env;
1605     struct kvm_debug_exit_arch *arch_info = &run->debug.arch;
1606 
1607     if (cs->singlestep_enabled) {
1608         return kvm_handle_singlestep();
1609     }
1610 
1611     if (arch_info->status) {
1612         return kvm_handle_hw_breakpoint(cs, arch_info);
1613     }
1614 
1615     if (kvm_find_sw_breakpoint(cs, arch_info->address)) {
1616         return kvm_handle_sw_breakpoint();
1617     }
1618 
1619     /*
1620      * QEMU is not able to handle debug exception, so inject
1621      * program exception to guest;
1622      * Yes program exception NOT debug exception !!
1623      * When QEMU is using debug resources then debug exception must
1624      * be always set. To achieve this we set MSR_DE and also set
1625      * MSRP_DEP so guest cannot change MSR_DE.
1626      * When emulating debug resource for guest we want guest
1627      * to control MSR_DE (enable/disable debug interrupt on need).
1628      * Supporting both configurations are NOT possible.
1629      * So the result is that we cannot share debug resources
1630      * between QEMU and Guest on BOOKE architecture.
1631      * In the current design QEMU gets the priority over guest,
1632      * this means that if QEMU is using debug resources then guest
1633      * cannot use them;
1634      * For software breakpoint QEMU uses a privileged instruction;
1635      * So there cannot be any reason that we are here for guest
1636      * set debug exception, only possibility is guest executed a
1637      * privileged / illegal instruction and that's why we are
1638      * injecting a program interrupt.
1639      */
1640     cpu_synchronize_state(cs);
1641     /*
1642      * env->nip is PC, so increment this by 4 to use
1643      * ppc_cpu_do_interrupt(), which set srr0 = env->nip - 4.
1644      */
1645     env->nip += 4;
1646     cs->exception_index = POWERPC_EXCP_PROGRAM;
1647     env->error_code = POWERPC_EXCP_INVAL;
1648     ppc_cpu_do_interrupt(cs);
1649 
1650     return DEBUG_RETURN_GUEST;
1651 }
1652 
1653 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1654 {
1655     PowerPCCPU *cpu = POWERPC_CPU(cs);
1656     CPUPPCState *env = &cpu->env;
1657     int ret;
1658 
1659     qemu_mutex_lock_iothread();
1660 
1661     switch (run->exit_reason) {
1662     case KVM_EXIT_DCR:
1663         if (run->dcr.is_write) {
1664             trace_kvm_handle_dcr_write();
1665             ret = kvmppc_handle_dcr_write(env, run->dcr.dcrn, run->dcr.data);
1666         } else {
1667             trace_kvm_handle_dcr_read();
1668             ret = kvmppc_handle_dcr_read(env, run->dcr.dcrn, &run->dcr.data);
1669         }
1670         break;
1671     case KVM_EXIT_HLT:
1672         trace_kvm_handle_halt();
1673         ret = kvmppc_handle_halt(cpu);
1674         break;
1675 #if defined(TARGET_PPC64)
1676     case KVM_EXIT_PAPR_HCALL:
1677         trace_kvm_handle_papr_hcall(run->papr_hcall.nr);
1678         run->papr_hcall.ret = spapr_hypercall(cpu,
1679                                               run->papr_hcall.nr,
1680                                               run->papr_hcall.args);
1681         ret = 0;
1682         break;
1683 #endif
1684     case KVM_EXIT_EPR:
1685         trace_kvm_handle_epr();
1686         run->epr.epr = ldl_phys(cs->as, env->mpic_iack);
1687         ret = 0;
1688         break;
1689     case KVM_EXIT_WATCHDOG:
1690         trace_kvm_handle_watchdog_expiry();
1691         watchdog_perform_action();
1692         ret = 0;
1693         break;
1694 
1695     case KVM_EXIT_DEBUG:
1696         trace_kvm_handle_debug_exception();
1697         if (kvm_handle_debug(cpu, run)) {
1698             ret = EXCP_DEBUG;
1699             break;
1700         }
1701         /* re-enter, this exception was guest-internal */
1702         ret = 0;
1703         break;
1704 
1705 #if defined(TARGET_PPC64)
1706     case KVM_EXIT_NMI:
1707         trace_kvm_handle_nmi_exception();
1708         ret = kvm_handle_nmi(cpu, run);
1709         break;
1710 #endif
1711 
1712     default:
1713         fprintf(stderr, "KVM: unknown exit reason %d\n", run->exit_reason);
1714         ret = -1;
1715         break;
1716     }
1717 
1718     qemu_mutex_unlock_iothread();
1719     return ret;
1720 }
1721 
1722 int kvmppc_or_tsr_bits(PowerPCCPU *cpu, uint32_t tsr_bits)
1723 {
1724     CPUState *cs = CPU(cpu);
1725     uint32_t bits = tsr_bits;
1726     struct kvm_one_reg reg = {
1727         .id = KVM_REG_PPC_OR_TSR,
1728         .addr = (uintptr_t) &bits,
1729     };
1730 
1731     return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
1732 }
1733 
1734 int kvmppc_clear_tsr_bits(PowerPCCPU *cpu, uint32_t tsr_bits)
1735 {
1736 
1737     CPUState *cs = CPU(cpu);
1738     uint32_t bits = tsr_bits;
1739     struct kvm_one_reg reg = {
1740         .id = KVM_REG_PPC_CLEAR_TSR,
1741         .addr = (uintptr_t) &bits,
1742     };
1743 
1744     return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
1745 }
1746 
1747 int kvmppc_set_tcr(PowerPCCPU *cpu)
1748 {
1749     CPUState *cs = CPU(cpu);
1750     CPUPPCState *env = &cpu->env;
1751     uint32_t tcr = env->spr[SPR_BOOKE_TCR];
1752 
1753     struct kvm_one_reg reg = {
1754         .id = KVM_REG_PPC_TCR,
1755         .addr = (uintptr_t) &tcr,
1756     };
1757 
1758     return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
1759 }
1760 
1761 int kvmppc_booke_watchdog_enable(PowerPCCPU *cpu)
1762 {
1763     CPUState *cs = CPU(cpu);
1764     int ret;
1765 
1766     if (!kvm_enabled()) {
1767         return -1;
1768     }
1769 
1770     if (!cap_ppc_watchdog) {
1771         printf("warning: KVM does not support watchdog");
1772         return -1;
1773     }
1774 
1775     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_PPC_BOOKE_WATCHDOG, 0);
1776     if (ret < 0) {
1777         fprintf(stderr, "%s: couldn't enable KVM_CAP_PPC_BOOKE_WATCHDOG: %s\n",
1778                 __func__, strerror(-ret));
1779         return ret;
1780     }
1781 
1782     return ret;
1783 }
1784 
1785 static int read_cpuinfo(const char *field, char *value, int len)
1786 {
1787     FILE *f;
1788     int ret = -1;
1789     int field_len = strlen(field);
1790     char line[512];
1791 
1792     f = fopen("/proc/cpuinfo", "r");
1793     if (!f) {
1794         return -1;
1795     }
1796 
1797     do {
1798         if (!fgets(line, sizeof(line), f)) {
1799             break;
1800         }
1801         if (!strncmp(line, field, field_len)) {
1802             pstrcpy(value, len, line);
1803             ret = 0;
1804             break;
1805         }
1806     } while (*line);
1807 
1808     fclose(f);
1809 
1810     return ret;
1811 }
1812 
1813 static uint32_t kvmppc_get_tbfreq_procfs(void)
1814 {
1815     char line[512];
1816     char *ns;
1817     uint32_t tbfreq_fallback = NANOSECONDS_PER_SECOND;
1818     uint32_t tbfreq_procfs;
1819 
1820     if (read_cpuinfo("timebase", line, sizeof(line))) {
1821         return tbfreq_fallback;
1822     }
1823 
1824     ns = strchr(line, ':');
1825     if (!ns) {
1826         return tbfreq_fallback;
1827     }
1828 
1829     tbfreq_procfs = atoi(++ns);
1830 
1831     /* 0 is certainly not acceptable by the guest, return fallback value */
1832     return tbfreq_procfs ? tbfreq_procfs : tbfreq_fallback;
1833 }
1834 
1835 uint32_t kvmppc_get_tbfreq(void)
1836 {
1837     static uint32_t cached_tbfreq;
1838 
1839     if (!cached_tbfreq) {
1840         cached_tbfreq = kvmppc_get_tbfreq_procfs();
1841     }
1842 
1843     return cached_tbfreq;
1844 }
1845 
1846 bool kvmppc_get_host_serial(char **value)
1847 {
1848     return g_file_get_contents("/proc/device-tree/system-id", value, NULL,
1849                                NULL);
1850 }
1851 
1852 bool kvmppc_get_host_model(char **value)
1853 {
1854     return g_file_get_contents("/proc/device-tree/model", value, NULL, NULL);
1855 }
1856 
1857 /* Try to find a device tree node for a CPU with clock-frequency property */
1858 static int kvmppc_find_cpu_dt(char *buf, int buf_len)
1859 {
1860     struct dirent *dirp;
1861     DIR *dp;
1862 
1863     dp = opendir(PROC_DEVTREE_CPU);
1864     if (!dp) {
1865         printf("Can't open directory " PROC_DEVTREE_CPU "\n");
1866         return -1;
1867     }
1868 
1869     buf[0] = '\0';
1870     while ((dirp = readdir(dp)) != NULL) {
1871         FILE *f;
1872 
1873         /* Don't accidentally read from the current and parent directories */
1874         if (strcmp(dirp->d_name, ".") == 0 || strcmp(dirp->d_name, "..") == 0) {
1875             continue;
1876         }
1877 
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 = 0;
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 bool kvmppc_supports_ail_3(void)
2566 {
2567     return cap_ail_mode_3;
2568 }
2569 
2570 PowerPCCPUClass *kvm_ppc_get_host_cpu_class(void)
2571 {
2572     uint32_t host_pvr = mfpvr();
2573     PowerPCCPUClass *pvr_pcc;
2574 
2575     pvr_pcc = ppc_cpu_class_by_pvr(host_pvr);
2576     if (pvr_pcc == NULL) {
2577         pvr_pcc = ppc_cpu_class_by_pvr_mask(host_pvr);
2578     }
2579 
2580     return pvr_pcc;
2581 }
2582 
2583 static void pseries_machine_class_fixup(ObjectClass *oc, void *opaque)
2584 {
2585     MachineClass *mc = MACHINE_CLASS(oc);
2586 
2587     mc->default_cpu_type = TYPE_HOST_POWERPC_CPU;
2588 }
2589 
2590 static int kvm_ppc_register_host_cpu_type(void)
2591 {
2592     TypeInfo type_info = {
2593         .name = TYPE_HOST_POWERPC_CPU,
2594         .class_init = kvmppc_host_cpu_class_init,
2595     };
2596     PowerPCCPUClass *pvr_pcc;
2597     ObjectClass *oc;
2598     DeviceClass *dc;
2599     int i;
2600 
2601     pvr_pcc = kvm_ppc_get_host_cpu_class();
2602     if (pvr_pcc == NULL) {
2603         return -1;
2604     }
2605     type_info.parent = object_class_get_name(OBJECT_CLASS(pvr_pcc));
2606     type_register(&type_info);
2607     /* override TCG default cpu type with 'host' cpu model */
2608     object_class_foreach(pseries_machine_class_fixup, TYPE_SPAPR_MACHINE,
2609                          false, NULL);
2610 
2611     oc = object_class_by_name(type_info.name);
2612     g_assert(oc);
2613 
2614     /*
2615      * Update generic CPU family class alias (e.g. on a POWER8NVL host,
2616      * we want "POWER8" to be a "family" alias that points to the current
2617      * host CPU type, too)
2618      */
2619     dc = DEVICE_CLASS(ppc_cpu_get_family_class(pvr_pcc));
2620     for (i = 0; ppc_cpu_aliases[i].alias != NULL; i++) {
2621         if (strcasecmp(ppc_cpu_aliases[i].alias, dc->desc) == 0) {
2622             char *suffix;
2623 
2624             ppc_cpu_aliases[i].model = g_strdup(object_class_get_name(oc));
2625             suffix = strstr(ppc_cpu_aliases[i].model, POWERPC_CPU_TYPE_SUFFIX);
2626             if (suffix) {
2627                 *suffix = 0;
2628             }
2629             break;
2630         }
2631     }
2632 
2633     return 0;
2634 }
2635 
2636 int kvmppc_define_rtas_kernel_token(uint32_t token, const char *function)
2637 {
2638     struct kvm_rtas_token_args args = {
2639         .token = token,
2640     };
2641 
2642     if (!kvm_check_extension(kvm_state, KVM_CAP_PPC_RTAS)) {
2643         return -ENOENT;
2644     }
2645 
2646     strncpy(args.name, function, sizeof(args.name) - 1);
2647 
2648     return kvm_vm_ioctl(kvm_state, KVM_PPC_RTAS_DEFINE_TOKEN, &args);
2649 }
2650 
2651 int kvmppc_get_htab_fd(bool write, uint64_t index, Error **errp)
2652 {
2653     struct kvm_get_htab_fd s = {
2654         .flags = write ? KVM_GET_HTAB_WRITE : 0,
2655         .start_index = index,
2656     };
2657     int ret;
2658 
2659     if (!cap_htab_fd) {
2660         error_setg(errp, "KVM version doesn't support %s the HPT",
2661                    write ? "writing" : "reading");
2662         return -ENOTSUP;
2663     }
2664 
2665     ret = kvm_vm_ioctl(kvm_state, KVM_PPC_GET_HTAB_FD, &s);
2666     if (ret < 0) {
2667         error_setg(errp, "Unable to open fd for %s HPT %s KVM: %s",
2668                    write ? "writing" : "reading", write ? "to" : "from",
2669                    strerror(errno));
2670         return -errno;
2671     }
2672 
2673     return ret;
2674 }
2675 
2676 int kvmppc_save_htab(QEMUFile *f, int fd, size_t bufsize, int64_t max_ns)
2677 {
2678     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2679     uint8_t buf[bufsize];
2680     ssize_t rc;
2681 
2682     do {
2683         rc = read(fd, buf, bufsize);
2684         if (rc < 0) {
2685             fprintf(stderr, "Error reading data from KVM HTAB fd: %s\n",
2686                     strerror(errno));
2687             return rc;
2688         } else if (rc) {
2689             uint8_t *buffer = buf;
2690             ssize_t n = rc;
2691             while (n) {
2692                 struct kvm_get_htab_header *head =
2693                     (struct kvm_get_htab_header *) buffer;
2694                 size_t chunksize = sizeof(*head) +
2695                      HASH_PTE_SIZE_64 * head->n_valid;
2696 
2697                 qemu_put_be32(f, head->index);
2698                 qemu_put_be16(f, head->n_valid);
2699                 qemu_put_be16(f, head->n_invalid);
2700                 qemu_put_buffer(f, (void *)(head + 1),
2701                                 HASH_PTE_SIZE_64 * head->n_valid);
2702 
2703                 buffer += chunksize;
2704                 n -= chunksize;
2705             }
2706         }
2707     } while ((rc != 0)
2708              && ((max_ns < 0) ||
2709                  ((qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) < max_ns)));
2710 
2711     return (rc == 0) ? 1 : 0;
2712 }
2713 
2714 int kvmppc_load_htab_chunk(QEMUFile *f, int fd, uint32_t index,
2715                            uint16_t n_valid, uint16_t n_invalid, Error **errp)
2716 {
2717     struct kvm_get_htab_header *buf;
2718     size_t chunksize = sizeof(*buf) + n_valid * HASH_PTE_SIZE_64;
2719     ssize_t rc;
2720 
2721     buf = alloca(chunksize);
2722     buf->index = index;
2723     buf->n_valid = n_valid;
2724     buf->n_invalid = n_invalid;
2725 
2726     qemu_get_buffer(f, (void *)(buf + 1), HASH_PTE_SIZE_64 * n_valid);
2727 
2728     rc = write(fd, buf, chunksize);
2729     if (rc < 0) {
2730         error_setg_errno(errp, errno, "Error writing the KVM hash table");
2731         return -errno;
2732     }
2733     if (rc != chunksize) {
2734         /* We should never get a short write on a single chunk */
2735         error_setg(errp, "Short write while restoring the KVM hash table");
2736         return -ENOSPC;
2737     }
2738     return 0;
2739 }
2740 
2741 bool kvm_arch_stop_on_emulation_error(CPUState *cpu)
2742 {
2743     return true;
2744 }
2745 
2746 void kvm_arch_init_irq_routing(KVMState *s)
2747 {
2748 }
2749 
2750 void kvmppc_read_hptes(ppc_hash_pte64_t *hptes, hwaddr ptex, int n)
2751 {
2752     int fd, rc;
2753     int i;
2754 
2755     fd = kvmppc_get_htab_fd(false, ptex, &error_abort);
2756 
2757     i = 0;
2758     while (i < n) {
2759         struct kvm_get_htab_header *hdr;
2760         int m = n < HPTES_PER_GROUP ? n : HPTES_PER_GROUP;
2761         char buf[sizeof(*hdr) + m * HASH_PTE_SIZE_64];
2762 
2763         rc = read(fd, buf, sizeof(buf));
2764         if (rc < 0) {
2765             hw_error("kvmppc_read_hptes: Unable to read HPTEs");
2766         }
2767 
2768         hdr = (struct kvm_get_htab_header *)buf;
2769         while ((i < n) && ((char *)hdr < (buf + rc))) {
2770             int invalid = hdr->n_invalid, valid = hdr->n_valid;
2771 
2772             if (hdr->index != (ptex + i)) {
2773                 hw_error("kvmppc_read_hptes: Unexpected HPTE index %"PRIu32
2774                          " != (%"HWADDR_PRIu" + %d", hdr->index, ptex, i);
2775             }
2776 
2777             if (n - i < valid) {
2778                 valid = n - i;
2779             }
2780             memcpy(hptes + i, hdr + 1, HASH_PTE_SIZE_64 * valid);
2781             i += valid;
2782 
2783             if ((n - i) < invalid) {
2784                 invalid = n - i;
2785             }
2786             memset(hptes + i, 0, invalid * HASH_PTE_SIZE_64);
2787             i += invalid;
2788 
2789             hdr = (struct kvm_get_htab_header *)
2790                 ((char *)(hdr + 1) + HASH_PTE_SIZE_64 * hdr->n_valid);
2791         }
2792     }
2793 
2794     close(fd);
2795 }
2796 
2797 void kvmppc_write_hpte(hwaddr ptex, uint64_t pte0, uint64_t pte1)
2798 {
2799     int fd, rc;
2800     struct {
2801         struct kvm_get_htab_header hdr;
2802         uint64_t pte0;
2803         uint64_t pte1;
2804     } buf;
2805 
2806     fd = kvmppc_get_htab_fd(true, 0 /* Ignored */, &error_abort);
2807 
2808     buf.hdr.n_valid = 1;
2809     buf.hdr.n_invalid = 0;
2810     buf.hdr.index = ptex;
2811     buf.pte0 = cpu_to_be64(pte0);
2812     buf.pte1 = cpu_to_be64(pte1);
2813 
2814     rc = write(fd, &buf, sizeof(buf));
2815     if (rc != sizeof(buf)) {
2816         hw_error("kvmppc_write_hpte: Unable to update KVM HPT");
2817     }
2818     close(fd);
2819 }
2820 
2821 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
2822                              uint64_t address, uint32_t data, PCIDevice *dev)
2823 {
2824     return 0;
2825 }
2826 
2827 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
2828                                 int vector, PCIDevice *dev)
2829 {
2830     return 0;
2831 }
2832 
2833 int kvm_arch_release_virq_post(int virq)
2834 {
2835     return 0;
2836 }
2837 
2838 int kvm_arch_msi_data_to_gsi(uint32_t data)
2839 {
2840     return data & 0xffff;
2841 }
2842 
2843 #if defined(TARGET_PPC64)
2844 int kvm_handle_nmi(PowerPCCPU *cpu, struct kvm_run *run)
2845 {
2846     uint16_t flags = run->flags & KVM_RUN_PPC_NMI_DISP_MASK;
2847 
2848     cpu_synchronize_state(CPU(cpu));
2849 
2850     spapr_mce_req_event(cpu, flags == KVM_RUN_PPC_NMI_DISP_FULLY_RECOV);
2851 
2852     return 0;
2853 }
2854 #endif
2855 
2856 int kvmppc_enable_hwrng(void)
2857 {
2858     if (!kvm_enabled() || !kvm_check_extension(kvm_state, KVM_CAP_PPC_HWRNG)) {
2859         return -1;
2860     }
2861 
2862     return kvmppc_enable_hcall(kvm_state, H_RANDOM);
2863 }
2864 
2865 void kvmppc_check_papr_resize_hpt(Error **errp)
2866 {
2867     if (!kvm_enabled()) {
2868         return; /* No KVM, we're good */
2869     }
2870 
2871     if (cap_resize_hpt) {
2872         return; /* Kernel has explicit support, we're good */
2873     }
2874 
2875     /* Otherwise fallback on looking for PR KVM */
2876     if (kvmppc_is_pr(kvm_state)) {
2877         return;
2878     }
2879 
2880     error_setg(errp,
2881                "Hash page table resizing not available with this KVM version");
2882 }
2883 
2884 int kvmppc_resize_hpt_prepare(PowerPCCPU *cpu, target_ulong flags, int shift)
2885 {
2886     CPUState *cs = CPU(cpu);
2887     struct kvm_ppc_resize_hpt rhpt = {
2888         .flags = flags,
2889         .shift = shift,
2890     };
2891 
2892     if (!cap_resize_hpt) {
2893         return -ENOSYS;
2894     }
2895 
2896     return kvm_vm_ioctl(cs->kvm_state, KVM_PPC_RESIZE_HPT_PREPARE, &rhpt);
2897 }
2898 
2899 int kvmppc_resize_hpt_commit(PowerPCCPU *cpu, target_ulong flags, int shift)
2900 {
2901     CPUState *cs = CPU(cpu);
2902     struct kvm_ppc_resize_hpt rhpt = {
2903         .flags = flags,
2904         .shift = shift,
2905     };
2906 
2907     if (!cap_resize_hpt) {
2908         return -ENOSYS;
2909     }
2910 
2911     return kvm_vm_ioctl(cs->kvm_state, KVM_PPC_RESIZE_HPT_COMMIT, &rhpt);
2912 }
2913 
2914 /*
2915  * This is a helper function to detect a post migration scenario
2916  * in which a guest, running as KVM-HV, freezes in cpu_post_load because
2917  * the guest kernel can't handle a PVR value other than the actual host
2918  * PVR in KVM_SET_SREGS, even if pvr_match() returns true.
2919  *
2920  * If we don't have cap_ppc_pvr_compat and we're not running in PR
2921  * (so, we're HV), return true. The workaround itself is done in
2922  * cpu_post_load.
2923  *
2924  * The order here is important: we'll only check for KVM PR as a
2925  * fallback if the guest kernel can't handle the situation itself.
2926  * We need to avoid as much as possible querying the running KVM type
2927  * in QEMU level.
2928  */
2929 bool kvmppc_pvr_workaround_required(PowerPCCPU *cpu)
2930 {
2931     CPUState *cs = CPU(cpu);
2932 
2933     if (!kvm_enabled()) {
2934         return false;
2935     }
2936 
2937     if (cap_ppc_pvr_compat) {
2938         return false;
2939     }
2940 
2941     return !kvmppc_is_pr(cs->kvm_state);
2942 }
2943 
2944 void kvmppc_set_reg_ppc_online(PowerPCCPU *cpu, unsigned int online)
2945 {
2946     CPUState *cs = CPU(cpu);
2947 
2948     if (kvm_enabled()) {
2949         kvm_set_one_reg(cs, KVM_REG_PPC_ONLINE, &online);
2950     }
2951 }
2952 
2953 void kvmppc_set_reg_tb_offset(PowerPCCPU *cpu, int64_t tb_offset)
2954 {
2955     CPUState *cs = CPU(cpu);
2956 
2957     if (kvm_enabled()) {
2958         kvm_set_one_reg(cs, KVM_REG_PPC_TB_OFFSET, &tb_offset);
2959     }
2960 }
2961 
2962 bool kvm_arch_cpu_check_are_resettable(void)
2963 {
2964     return true;
2965 }
2966 
2967 void kvm_arch_accel_class_init(ObjectClass *oc)
2968 {
2969 }
2970