xref: /openbmc/qemu/target/ppc/kvm.c (revision 4e6b1384)
1 /*
2  * PowerPC implementation of KVM hooks
3  *
4  * Copyright IBM Corp. 2007
5  * Copyright (C) 2011 Freescale Semiconductor, Inc.
6  *
7  * Authors:
8  *  Jerone Young <jyoung5@us.ibm.com>
9  *  Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
10  *  Hollis Blanchard <hollisb@us.ibm.com>
11  *
12  * This work is licensed under the terms of the GNU GPL, version 2 or later.
13  * See the COPYING file in the top-level directory.
14  *
15  */
16 
17 #include "qemu/osdep.h"
18 #include <dirent.h>
19 #include <sys/ioctl.h>
20 #include <sys/vfs.h>
21 
22 #include <linux/kvm.h>
23 
24 #include "qemu-common.h"
25 #include "qapi/error.h"
26 #include "qemu/error-report.h"
27 #include "cpu.h"
28 #include "cpu-models.h"
29 #include "qemu/timer.h"
30 #include "sysemu/hw_accel.h"
31 #include "kvm_ppc.h"
32 #include "sysemu/cpus.h"
33 #include "sysemu/device_tree.h"
34 #include "mmu-hash64.h"
35 
36 #include "hw/sysbus.h"
37 #include "hw/ppc/spapr.h"
38 #include "hw/ppc/spapr_cpu_core.h"
39 #include "hw/hw.h"
40 #include "hw/ppc/ppc.h"
41 #include "migration/qemu-file-types.h"
42 #include "sysemu/watchdog.h"
43 #include "trace.h"
44 #include "exec/gdbstub.h"
45 #include "exec/memattrs.h"
46 #include "exec/ram_addr.h"
47 #include "sysemu/hostmem.h"
48 #include "qemu/cutils.h"
49 #include "qemu/main-loop.h"
50 #include "qemu/mmap-alloc.h"
51 #include "elf.h"
52 #include "sysemu/kvm_int.h"
53 
54 #define PROC_DEVTREE_CPU      "/proc/device-tree/cpus/"
55 
56 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
57     KVM_CAP_LAST_INFO
58 };
59 
60 static int cap_interrupt_unset;
61 static int cap_segstate;
62 static int cap_booke_sregs;
63 static int cap_ppc_smt;
64 static int cap_ppc_smt_possible;
65 static int cap_spapr_tce;
66 static int cap_spapr_tce_64;
67 static int cap_spapr_multitce;
68 static int cap_spapr_vfio;
69 static int cap_hior;
70 static int cap_one_reg;
71 static int cap_epr;
72 static int cap_ppc_watchdog;
73 static int cap_papr;
74 static int cap_htab_fd;
75 static int cap_fixup_hcalls;
76 static int cap_htm;             /* Hardware transactional memory support */
77 static int cap_mmu_radix;
78 static int cap_mmu_hash_v3;
79 static int cap_xive;
80 static int cap_resize_hpt;
81 static int cap_ppc_pvr_compat;
82 static int cap_ppc_safe_cache;
83 static int cap_ppc_safe_bounds_check;
84 static int cap_ppc_safe_indirect_branch;
85 static int cap_ppc_count_cache_flush_assist;
86 static int cap_ppc_nested_kvm_hv;
87 static int cap_large_decr;
88 
89 static uint32_t debug_inst_opcode;
90 
91 /*
92  * Check whether we are running with KVM-PR (instead of KVM-HV).  This
93  * should only be used for fallback tests - generally we should use
94  * explicit capabilities for the features we want, rather than
95  * assuming what is/isn't available depending on the KVM variant.
96  */
97 static bool kvmppc_is_pr(KVMState *ks)
98 {
99     /* Assume KVM-PR if the GET_PVINFO capability is available */
100     return kvm_vm_check_extension(ks, KVM_CAP_PPC_GET_PVINFO) != 0;
101 }
102 
103 static int kvm_ppc_register_host_cpu_type(void);
104 static void kvmppc_get_cpu_characteristics(KVMState *s);
105 static int kvmppc_get_dec_bits(void);
106 
107 int kvm_arch_init(MachineState *ms, KVMState *s)
108 {
109     cap_interrupt_unset = kvm_check_extension(s, KVM_CAP_PPC_UNSET_IRQ);
110     cap_segstate = kvm_check_extension(s, KVM_CAP_PPC_SEGSTATE);
111     cap_booke_sregs = kvm_check_extension(s, KVM_CAP_PPC_BOOKE_SREGS);
112     cap_ppc_smt_possible = kvm_vm_check_extension(s, KVM_CAP_PPC_SMT_POSSIBLE);
113     cap_spapr_tce = kvm_check_extension(s, KVM_CAP_SPAPR_TCE);
114     cap_spapr_tce_64 = kvm_check_extension(s, KVM_CAP_SPAPR_TCE_64);
115     cap_spapr_multitce = kvm_check_extension(s, KVM_CAP_SPAPR_MULTITCE);
116     cap_spapr_vfio = kvm_vm_check_extension(s, KVM_CAP_SPAPR_TCE_VFIO);
117     cap_one_reg = kvm_check_extension(s, KVM_CAP_ONE_REG);
118     cap_hior = kvm_check_extension(s, KVM_CAP_PPC_HIOR);
119     cap_epr = kvm_check_extension(s, KVM_CAP_PPC_EPR);
120     cap_ppc_watchdog = kvm_check_extension(s, KVM_CAP_PPC_BOOKE_WATCHDOG);
121     /*
122      * Note: we don't set cap_papr here, because this capability is
123      * only activated after this by kvmppc_set_papr()
124      */
125     cap_htab_fd = kvm_vm_check_extension(s, KVM_CAP_PPC_HTAB_FD);
126     cap_fixup_hcalls = kvm_check_extension(s, KVM_CAP_PPC_FIXUP_HCALL);
127     cap_ppc_smt = kvm_vm_check_extension(s, KVM_CAP_PPC_SMT);
128     cap_htm = kvm_vm_check_extension(s, KVM_CAP_PPC_HTM);
129     cap_mmu_radix = kvm_vm_check_extension(s, KVM_CAP_PPC_MMU_RADIX);
130     cap_mmu_hash_v3 = kvm_vm_check_extension(s, KVM_CAP_PPC_MMU_HASH_V3);
131     cap_xive = kvm_vm_check_extension(s, KVM_CAP_PPC_IRQ_XIVE);
132     cap_resize_hpt = kvm_vm_check_extension(s, KVM_CAP_SPAPR_RESIZE_HPT);
133     kvmppc_get_cpu_characteristics(s);
134     cap_ppc_nested_kvm_hv = kvm_vm_check_extension(s, KVM_CAP_PPC_NESTED_HV);
135     cap_large_decr = kvmppc_get_dec_bits();
136     /*
137      * Note: setting it to false because there is not such capability
138      * in KVM at this moment.
139      *
140      * TODO: call kvm_vm_check_extension() with the right capability
141      * after the kernel starts implementing it.
142      */
143     cap_ppc_pvr_compat = false;
144 
145     if (!kvm_check_extension(s, KVM_CAP_PPC_IRQ_LEVEL)) {
146         error_report("KVM: Host kernel doesn't have level irq capability");
147         exit(1);
148     }
149 
150     kvm_ppc_register_host_cpu_type();
151 
152     return 0;
153 }
154 
155 int kvm_arch_irqchip_create(KVMState *s)
156 {
157     return 0;
158 }
159 
160 static int kvm_arch_sync_sregs(PowerPCCPU *cpu)
161 {
162     CPUPPCState *cenv = &cpu->env;
163     CPUState *cs = CPU(cpu);
164     struct kvm_sregs sregs;
165     int ret;
166 
167     if (cenv->excp_model == POWERPC_EXCP_BOOKE) {
168         /*
169          * What we're really trying to say is "if we're on BookE, we
170          * use the native PVR for now". This is the only sane way to
171          * check it though, so we potentially confuse users that they
172          * can run BookE guests on BookS. Let's hope nobody dares
173          * enough :)
174          */
175         return 0;
176     } else {
177         if (!cap_segstate) {
178             fprintf(stderr, "kvm error: missing PVR setting capability\n");
179             return -ENOSYS;
180         }
181     }
182 
183     ret = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs);
184     if (ret) {
185         return ret;
186     }
187 
188     sregs.pvr = cenv->spr[SPR_PVR];
189     return kvm_vcpu_ioctl(cs, KVM_SET_SREGS, &sregs);
190 }
191 
192 /* Set up a shared TLB array with KVM */
193 static int kvm_booke206_tlb_init(PowerPCCPU *cpu)
194 {
195     CPUPPCState *env = &cpu->env;
196     CPUState *cs = CPU(cpu);
197     struct kvm_book3e_206_tlb_params params = {};
198     struct kvm_config_tlb cfg = {};
199     unsigned int entries = 0;
200     int ret, i;
201 
202     if (!kvm_enabled() ||
203         !kvm_check_extension(cs->kvm_state, KVM_CAP_SW_TLB)) {
204         return 0;
205     }
206 
207     assert(ARRAY_SIZE(params.tlb_sizes) == BOOKE206_MAX_TLBN);
208 
209     for (i = 0; i < BOOKE206_MAX_TLBN; i++) {
210         params.tlb_sizes[i] = booke206_tlb_size(env, i);
211         params.tlb_ways[i] = booke206_tlb_ways(env, i);
212         entries += params.tlb_sizes[i];
213     }
214 
215     assert(entries == env->nb_tlb);
216     assert(sizeof(struct kvm_book3e_206_tlb_entry) == sizeof(ppcmas_tlb_t));
217 
218     env->tlb_dirty = true;
219 
220     cfg.array = (uintptr_t)env->tlb.tlbm;
221     cfg.array_len = sizeof(ppcmas_tlb_t) * entries;
222     cfg.params = (uintptr_t)&params;
223     cfg.mmu_type = KVM_MMU_FSL_BOOKE_NOHV;
224 
225     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_SW_TLB, 0, (uintptr_t)&cfg);
226     if (ret < 0) {
227         fprintf(stderr, "%s: couldn't enable KVM_CAP_SW_TLB: %s\n",
228                 __func__, strerror(-ret));
229         return ret;
230     }
231 
232     env->kvm_sw_tlb = true;
233     return 0;
234 }
235 
236 
237 #if defined(TARGET_PPC64)
238 static void kvm_get_smmu_info(struct kvm_ppc_smmu_info *info, Error **errp)
239 {
240     int ret;
241 
242     assert(kvm_state != NULL);
243 
244     if (!kvm_check_extension(kvm_state, KVM_CAP_PPC_GET_SMMU_INFO)) {
245         error_setg(errp, "KVM doesn't expose the MMU features it supports");
246         error_append_hint(errp, "Consider switching to a newer KVM\n");
247         return;
248     }
249 
250     ret = kvm_vm_ioctl(kvm_state, KVM_PPC_GET_SMMU_INFO, info);
251     if (ret == 0) {
252         return;
253     }
254 
255     error_setg_errno(errp, -ret,
256                      "KVM failed to provide the MMU features it supports");
257 }
258 
259 struct ppc_radix_page_info *kvm_get_radix_page_info(void)
260 {
261     KVMState *s = KVM_STATE(current_machine->accelerator);
262     struct ppc_radix_page_info *radix_page_info;
263     struct kvm_ppc_rmmu_info rmmu_info;
264     int i;
265 
266     if (!kvm_check_extension(s, KVM_CAP_PPC_MMU_RADIX)) {
267         return NULL;
268     }
269     if (kvm_vm_ioctl(s, KVM_PPC_GET_RMMU_INFO, &rmmu_info)) {
270         return NULL;
271     }
272     radix_page_info = g_malloc0(sizeof(*radix_page_info));
273     radix_page_info->count = 0;
274     for (i = 0; i < PPC_PAGE_SIZES_MAX_SZ; i++) {
275         if (rmmu_info.ap_encodings[i]) {
276             radix_page_info->entries[i] = rmmu_info.ap_encodings[i];
277             radix_page_info->count++;
278         }
279     }
280     return radix_page_info;
281 }
282 
283 target_ulong kvmppc_configure_v3_mmu(PowerPCCPU *cpu,
284                                      bool radix, bool gtse,
285                                      uint64_t proc_tbl)
286 {
287     CPUState *cs = CPU(cpu);
288     int ret;
289     uint64_t flags = 0;
290     struct kvm_ppc_mmuv3_cfg cfg = {
291         .process_table = proc_tbl,
292     };
293 
294     if (radix) {
295         flags |= KVM_PPC_MMUV3_RADIX;
296     }
297     if (gtse) {
298         flags |= KVM_PPC_MMUV3_GTSE;
299     }
300     cfg.flags = flags;
301     ret = kvm_vm_ioctl(cs->kvm_state, KVM_PPC_CONFIGURE_V3_MMU, &cfg);
302     switch (ret) {
303     case 0:
304         return H_SUCCESS;
305     case -EINVAL:
306         return H_PARAMETER;
307     case -ENODEV:
308         return H_NOT_AVAILABLE;
309     default:
310         return H_HARDWARE;
311     }
312 }
313 
314 bool kvmppc_hpt_needs_host_contiguous_pages(void)
315 {
316     static struct kvm_ppc_smmu_info smmu_info;
317 
318     if (!kvm_enabled()) {
319         return false;
320     }
321 
322     kvm_get_smmu_info(&smmu_info, &error_fatal);
323     return !!(smmu_info.flags & KVM_PPC_PAGE_SIZES_REAL);
324 }
325 
326 void kvm_check_mmu(PowerPCCPU *cpu, Error **errp)
327 {
328     struct kvm_ppc_smmu_info smmu_info;
329     int iq, ik, jq, jk;
330     Error *local_err = NULL;
331 
332     /* For now, we only have anything to check on hash64 MMUs */
333     if (!cpu->hash64_opts || !kvm_enabled()) {
334         return;
335     }
336 
337     kvm_get_smmu_info(&smmu_info, &local_err);
338     if (local_err) {
339         error_propagate(errp, local_err);
340         return;
341     }
342 
343     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)
344         && !(smmu_info.flags & KVM_PPC_1T_SEGMENTS)) {
345         error_setg(errp,
346                    "KVM does not support 1TiB segments which guest expects");
347         return;
348     }
349 
350     if (smmu_info.slb_size < cpu->hash64_opts->slb_size) {
351         error_setg(errp, "KVM only supports %u SLB entries, but guest needs %u",
352                    smmu_info.slb_size, cpu->hash64_opts->slb_size);
353         return;
354     }
355 
356     /*
357      * Verify that every pagesize supported by the cpu model is
358      * supported by KVM with the same encodings
359      */
360     for (iq = 0; iq < ARRAY_SIZE(cpu->hash64_opts->sps); iq++) {
361         PPCHash64SegmentPageSizes *qsps = &cpu->hash64_opts->sps[iq];
362         struct kvm_ppc_one_seg_page_size *ksps;
363 
364         for (ik = 0; ik < ARRAY_SIZE(smmu_info.sps); ik++) {
365             if (qsps->page_shift == smmu_info.sps[ik].page_shift) {
366                 break;
367             }
368         }
369         if (ik >= ARRAY_SIZE(smmu_info.sps)) {
370             error_setg(errp, "KVM doesn't support for base page shift %u",
371                        qsps->page_shift);
372             return;
373         }
374 
375         ksps = &smmu_info.sps[ik];
376         if (ksps->slb_enc != qsps->slb_enc) {
377             error_setg(errp,
378 "KVM uses SLB encoding 0x%x for page shift %u, but guest expects 0x%x",
379                        ksps->slb_enc, ksps->page_shift, qsps->slb_enc);
380             return;
381         }
382 
383         for (jq = 0; jq < ARRAY_SIZE(qsps->enc); jq++) {
384             for (jk = 0; jk < ARRAY_SIZE(ksps->enc); jk++) {
385                 if (qsps->enc[jq].page_shift == ksps->enc[jk].page_shift) {
386                     break;
387                 }
388             }
389 
390             if (jk >= ARRAY_SIZE(ksps->enc)) {
391                 error_setg(errp, "KVM doesn't support page shift %u/%u",
392                            qsps->enc[jq].page_shift, qsps->page_shift);
393                 return;
394             }
395             if (qsps->enc[jq].pte_enc != ksps->enc[jk].pte_enc) {
396                 error_setg(errp,
397 "KVM uses PTE encoding 0x%x for page shift %u/%u, but guest expects 0x%x",
398                            ksps->enc[jk].pte_enc, qsps->enc[jq].page_shift,
399                            qsps->page_shift, qsps->enc[jq].pte_enc);
400                 return;
401             }
402         }
403     }
404 
405     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
406         /*
407          * Mostly what guest pagesizes we can use are related to the
408          * host pages used to map guest RAM, which is handled in the
409          * platform code. Cache-Inhibited largepages (64k) however are
410          * used for I/O, so if they're mapped to the host at all it
411          * will be a normal mapping, not a special hugepage one used
412          * for RAM.
413          */
414         if (qemu_real_host_page_size < 0x10000) {
415             error_setg(errp,
416                        "KVM can't supply 64kiB CI pages, which guest expects");
417         }
418     }
419 }
420 #endif /* !defined (TARGET_PPC64) */
421 
422 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
423 {
424     return POWERPC_CPU(cpu)->vcpu_id;
425 }
426 
427 /*
428  * e500 supports 2 h/w breakpoint and 2 watchpoint.  book3s supports
429  * only 1 watchpoint, so array size of 4 is sufficient for now.
430  */
431 #define MAX_HW_BKPTS 4
432 
433 static struct HWBreakpoint {
434     target_ulong addr;
435     int type;
436 } hw_debug_points[MAX_HW_BKPTS];
437 
438 static CPUWatchpoint hw_watchpoint;
439 
440 /* Default there is no breakpoint and watchpoint supported */
441 static int max_hw_breakpoint;
442 static int max_hw_watchpoint;
443 static int nb_hw_breakpoint;
444 static int nb_hw_watchpoint;
445 
446 static void kvmppc_hw_debug_points_init(CPUPPCState *cenv)
447 {
448     if (cenv->excp_model == POWERPC_EXCP_BOOKE) {
449         max_hw_breakpoint = 2;
450         max_hw_watchpoint = 2;
451     }
452 
453     if ((max_hw_breakpoint + max_hw_watchpoint) > MAX_HW_BKPTS) {
454         fprintf(stderr, "Error initializing h/w breakpoints\n");
455         return;
456     }
457 }
458 
459 int kvm_arch_init_vcpu(CPUState *cs)
460 {
461     PowerPCCPU *cpu = POWERPC_CPU(cs);
462     CPUPPCState *cenv = &cpu->env;
463     int ret;
464 
465     /* Synchronize sregs with kvm */
466     ret = kvm_arch_sync_sregs(cpu);
467     if (ret) {
468         if (ret == -EINVAL) {
469             error_report("Register sync failed... If you're using kvm-hv.ko,"
470                          " only \"-cpu host\" is possible");
471         }
472         return ret;
473     }
474 
475     switch (cenv->mmu_model) {
476     case POWERPC_MMU_BOOKE206:
477         /* This target supports access to KVM's guest TLB */
478         ret = kvm_booke206_tlb_init(cpu);
479         break;
480     case POWERPC_MMU_2_07:
481         if (!cap_htm && !kvmppc_is_pr(cs->kvm_state)) {
482             /*
483              * KVM-HV has transactional memory on POWER8 also without
484              * the KVM_CAP_PPC_HTM extension, so enable it here
485              * instead as long as it's availble to userspace on the
486              * host.
487              */
488             if (qemu_getauxval(AT_HWCAP2) & PPC_FEATURE2_HAS_HTM) {
489                 cap_htm = true;
490             }
491         }
492         break;
493     default:
494         break;
495     }
496 
497     kvm_get_one_reg(cs, KVM_REG_PPC_DEBUG_INST, &debug_inst_opcode);
498     kvmppc_hw_debug_points_init(cenv);
499 
500     return ret;
501 }
502 
503 int kvm_arch_destroy_vcpu(CPUState *cs)
504 {
505     return 0;
506 }
507 
508 static void kvm_sw_tlb_put(PowerPCCPU *cpu)
509 {
510     CPUPPCState *env = &cpu->env;
511     CPUState *cs = CPU(cpu);
512     struct kvm_dirty_tlb dirty_tlb;
513     unsigned char *bitmap;
514     int ret;
515 
516     if (!env->kvm_sw_tlb) {
517         return;
518     }
519 
520     bitmap = g_malloc((env->nb_tlb + 7) / 8);
521     memset(bitmap, 0xFF, (env->nb_tlb + 7) / 8);
522 
523     dirty_tlb.bitmap = (uintptr_t)bitmap;
524     dirty_tlb.num_dirty = env->nb_tlb;
525 
526     ret = kvm_vcpu_ioctl(cs, KVM_DIRTY_TLB, &dirty_tlb);
527     if (ret) {
528         fprintf(stderr, "%s: KVM_DIRTY_TLB: %s\n",
529                 __func__, strerror(-ret));
530     }
531 
532     g_free(bitmap);
533 }
534 
535 static void kvm_get_one_spr(CPUState *cs, uint64_t id, int spr)
536 {
537     PowerPCCPU *cpu = POWERPC_CPU(cs);
538     CPUPPCState *env = &cpu->env;
539     union {
540         uint32_t u32;
541         uint64_t u64;
542     } val;
543     struct kvm_one_reg reg = {
544         .id = id,
545         .addr = (uintptr_t) &val,
546     };
547     int ret;
548 
549     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
550     if (ret != 0) {
551         trace_kvm_failed_spr_get(spr, strerror(errno));
552     } else {
553         switch (id & KVM_REG_SIZE_MASK) {
554         case KVM_REG_SIZE_U32:
555             env->spr[spr] = val.u32;
556             break;
557 
558         case KVM_REG_SIZE_U64:
559             env->spr[spr] = val.u64;
560             break;
561 
562         default:
563             /* Don't handle this size yet */
564             abort();
565         }
566     }
567 }
568 
569 static void kvm_put_one_spr(CPUState *cs, uint64_t id, int spr)
570 {
571     PowerPCCPU *cpu = POWERPC_CPU(cs);
572     CPUPPCState *env = &cpu->env;
573     union {
574         uint32_t u32;
575         uint64_t u64;
576     } val;
577     struct kvm_one_reg reg = {
578         .id = id,
579         .addr = (uintptr_t) &val,
580     };
581     int ret;
582 
583     switch (id & KVM_REG_SIZE_MASK) {
584     case KVM_REG_SIZE_U32:
585         val.u32 = env->spr[spr];
586         break;
587 
588     case KVM_REG_SIZE_U64:
589         val.u64 = env->spr[spr];
590         break;
591 
592     default:
593         /* Don't handle this size yet */
594         abort();
595     }
596 
597     ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
598     if (ret != 0) {
599         trace_kvm_failed_spr_set(spr, strerror(errno));
600     }
601 }
602 
603 static int kvm_put_fp(CPUState *cs)
604 {
605     PowerPCCPU *cpu = POWERPC_CPU(cs);
606     CPUPPCState *env = &cpu->env;
607     struct kvm_one_reg reg;
608     int i;
609     int ret;
610 
611     if (env->insns_flags & PPC_FLOAT) {
612         uint64_t fpscr = env->fpscr;
613         bool vsx = !!(env->insns_flags2 & PPC2_VSX);
614 
615         reg.id = KVM_REG_PPC_FPSCR;
616         reg.addr = (uintptr_t)&fpscr;
617         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
618         if (ret < 0) {
619             trace_kvm_failed_fpscr_set(strerror(errno));
620             return ret;
621         }
622 
623         for (i = 0; i < 32; i++) {
624             uint64_t vsr[2];
625             uint64_t *fpr = cpu_fpr_ptr(&cpu->env, i);
626             uint64_t *vsrl = cpu_vsrl_ptr(&cpu->env, i);
627 
628 #ifdef HOST_WORDS_BIGENDIAN
629             vsr[0] = float64_val(*fpr);
630             vsr[1] = *vsrl;
631 #else
632             vsr[0] = *vsrl;
633             vsr[1] = float64_val(*fpr);
634 #endif
635             reg.addr = (uintptr_t) &vsr;
636             reg.id = vsx ? KVM_REG_PPC_VSR(i) : KVM_REG_PPC_FPR(i);
637 
638             ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
639             if (ret < 0) {
640                 trace_kvm_failed_fp_set(vsx ? "VSR" : "FPR", i,
641                                         strerror(errno));
642                 return ret;
643             }
644         }
645     }
646 
647     if (env->insns_flags & PPC_ALTIVEC) {
648         reg.id = KVM_REG_PPC_VSCR;
649         reg.addr = (uintptr_t)&env->vscr;
650         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
651         if (ret < 0) {
652             trace_kvm_failed_vscr_set(strerror(errno));
653             return ret;
654         }
655 
656         for (i = 0; i < 32; i++) {
657             reg.id = KVM_REG_PPC_VR(i);
658             reg.addr = (uintptr_t)cpu_avr_ptr(env, i);
659             ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
660             if (ret < 0) {
661                 trace_kvm_failed_vr_set(i, strerror(errno));
662                 return ret;
663             }
664         }
665     }
666 
667     return 0;
668 }
669 
670 static int kvm_get_fp(CPUState *cs)
671 {
672     PowerPCCPU *cpu = POWERPC_CPU(cs);
673     CPUPPCState *env = &cpu->env;
674     struct kvm_one_reg reg;
675     int i;
676     int ret;
677 
678     if (env->insns_flags & PPC_FLOAT) {
679         uint64_t fpscr;
680         bool vsx = !!(env->insns_flags2 & PPC2_VSX);
681 
682         reg.id = KVM_REG_PPC_FPSCR;
683         reg.addr = (uintptr_t)&fpscr;
684         ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
685         if (ret < 0) {
686             trace_kvm_failed_fpscr_get(strerror(errno));
687             return ret;
688         } else {
689             env->fpscr = fpscr;
690         }
691 
692         for (i = 0; i < 32; i++) {
693             uint64_t vsr[2];
694             uint64_t *fpr = cpu_fpr_ptr(&cpu->env, i);
695             uint64_t *vsrl = cpu_vsrl_ptr(&cpu->env, i);
696 
697             reg.addr = (uintptr_t) &vsr;
698             reg.id = vsx ? KVM_REG_PPC_VSR(i) : KVM_REG_PPC_FPR(i);
699 
700             ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
701             if (ret < 0) {
702                 trace_kvm_failed_fp_get(vsx ? "VSR" : "FPR", i,
703                                         strerror(errno));
704                 return ret;
705             } else {
706 #ifdef HOST_WORDS_BIGENDIAN
707                 *fpr = vsr[0];
708                 if (vsx) {
709                     *vsrl = vsr[1];
710                 }
711 #else
712                 *fpr = vsr[1];
713                 if (vsx) {
714                     *vsrl = vsr[0];
715                 }
716 #endif
717             }
718         }
719     }
720 
721     if (env->insns_flags & PPC_ALTIVEC) {
722         reg.id = KVM_REG_PPC_VSCR;
723         reg.addr = (uintptr_t)&env->vscr;
724         ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
725         if (ret < 0) {
726             trace_kvm_failed_vscr_get(strerror(errno));
727             return ret;
728         }
729 
730         for (i = 0; i < 32; i++) {
731             reg.id = KVM_REG_PPC_VR(i);
732             reg.addr = (uintptr_t)cpu_avr_ptr(env, i);
733             ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
734             if (ret < 0) {
735                 trace_kvm_failed_vr_get(i, strerror(errno));
736                 return ret;
737             }
738         }
739     }
740 
741     return 0;
742 }
743 
744 #if defined(TARGET_PPC64)
745 static int kvm_get_vpa(CPUState *cs)
746 {
747     PowerPCCPU *cpu = POWERPC_CPU(cs);
748     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
749     struct kvm_one_reg reg;
750     int ret;
751 
752     reg.id = KVM_REG_PPC_VPA_ADDR;
753     reg.addr = (uintptr_t)&spapr_cpu->vpa_addr;
754     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
755     if (ret < 0) {
756         trace_kvm_failed_vpa_addr_get(strerror(errno));
757         return ret;
758     }
759 
760     assert((uintptr_t)&spapr_cpu->slb_shadow_size
761            == ((uintptr_t)&spapr_cpu->slb_shadow_addr + 8));
762     reg.id = KVM_REG_PPC_VPA_SLB;
763     reg.addr = (uintptr_t)&spapr_cpu->slb_shadow_addr;
764     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
765     if (ret < 0) {
766         trace_kvm_failed_slb_get(strerror(errno));
767         return ret;
768     }
769 
770     assert((uintptr_t)&spapr_cpu->dtl_size
771            == ((uintptr_t)&spapr_cpu->dtl_addr + 8));
772     reg.id = KVM_REG_PPC_VPA_DTL;
773     reg.addr = (uintptr_t)&spapr_cpu->dtl_addr;
774     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
775     if (ret < 0) {
776         trace_kvm_failed_dtl_get(strerror(errno));
777         return ret;
778     }
779 
780     return 0;
781 }
782 
783 static int kvm_put_vpa(CPUState *cs)
784 {
785     PowerPCCPU *cpu = POWERPC_CPU(cs);
786     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
787     struct kvm_one_reg reg;
788     int ret;
789 
790     /*
791      * SLB shadow or DTL can't be registered unless a master VPA is
792      * registered.  That means when restoring state, if a VPA *is*
793      * registered, we need to set that up first.  If not, we need to
794      * deregister the others before deregistering the master VPA
795      */
796     assert(spapr_cpu->vpa_addr
797            || !(spapr_cpu->slb_shadow_addr || spapr_cpu->dtl_addr));
798 
799     if (spapr_cpu->vpa_addr) {
800         reg.id = KVM_REG_PPC_VPA_ADDR;
801         reg.addr = (uintptr_t)&spapr_cpu->vpa_addr;
802         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
803         if (ret < 0) {
804             trace_kvm_failed_vpa_addr_set(strerror(errno));
805             return ret;
806         }
807     }
808 
809     assert((uintptr_t)&spapr_cpu->slb_shadow_size
810            == ((uintptr_t)&spapr_cpu->slb_shadow_addr + 8));
811     reg.id = KVM_REG_PPC_VPA_SLB;
812     reg.addr = (uintptr_t)&spapr_cpu->slb_shadow_addr;
813     ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
814     if (ret < 0) {
815         trace_kvm_failed_slb_set(strerror(errno));
816         return ret;
817     }
818 
819     assert((uintptr_t)&spapr_cpu->dtl_size
820            == ((uintptr_t)&spapr_cpu->dtl_addr + 8));
821     reg.id = KVM_REG_PPC_VPA_DTL;
822     reg.addr = (uintptr_t)&spapr_cpu->dtl_addr;
823     ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
824     if (ret < 0) {
825         trace_kvm_failed_dtl_set(strerror(errno));
826         return ret;
827     }
828 
829     if (!spapr_cpu->vpa_addr) {
830         reg.id = KVM_REG_PPC_VPA_ADDR;
831         reg.addr = (uintptr_t)&spapr_cpu->vpa_addr;
832         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
833         if (ret < 0) {
834             trace_kvm_failed_null_vpa_addr_set(strerror(errno));
835             return ret;
836         }
837     }
838 
839     return 0;
840 }
841 #endif /* TARGET_PPC64 */
842 
843 int kvmppc_put_books_sregs(PowerPCCPU *cpu)
844 {
845     CPUPPCState *env = &cpu->env;
846     struct kvm_sregs sregs;
847     int i;
848 
849     sregs.pvr = env->spr[SPR_PVR];
850 
851     if (cpu->vhyp) {
852         PPCVirtualHypervisorClass *vhc =
853             PPC_VIRTUAL_HYPERVISOR_GET_CLASS(cpu->vhyp);
854         sregs.u.s.sdr1 = vhc->encode_hpt_for_kvm_pr(cpu->vhyp);
855     } else {
856         sregs.u.s.sdr1 = env->spr[SPR_SDR1];
857     }
858 
859     /* Sync SLB */
860 #ifdef TARGET_PPC64
861     for (i = 0; i < ARRAY_SIZE(env->slb); i++) {
862         sregs.u.s.ppc64.slb[i].slbe = env->slb[i].esid;
863         if (env->slb[i].esid & SLB_ESID_V) {
864             sregs.u.s.ppc64.slb[i].slbe |= i;
865         }
866         sregs.u.s.ppc64.slb[i].slbv = env->slb[i].vsid;
867     }
868 #endif
869 
870     /* Sync SRs */
871     for (i = 0; i < 16; i++) {
872         sregs.u.s.ppc32.sr[i] = env->sr[i];
873     }
874 
875     /* Sync BATs */
876     for (i = 0; i < 8; i++) {
877         /* Beware. We have to swap upper and lower bits here */
878         sregs.u.s.ppc32.dbat[i] = ((uint64_t)env->DBAT[0][i] << 32)
879             | env->DBAT[1][i];
880         sregs.u.s.ppc32.ibat[i] = ((uint64_t)env->IBAT[0][i] << 32)
881             | env->IBAT[1][i];
882     }
883 
884     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_SREGS, &sregs);
885 }
886 
887 int kvm_arch_put_registers(CPUState *cs, int level)
888 {
889     PowerPCCPU *cpu = POWERPC_CPU(cs);
890     CPUPPCState *env = &cpu->env;
891     struct kvm_regs regs;
892     int ret;
893     int i;
894 
895     ret = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
896     if (ret < 0) {
897         return ret;
898     }
899 
900     regs.ctr = env->ctr;
901     regs.lr  = env->lr;
902     regs.xer = cpu_read_xer(env);
903     regs.msr = env->msr;
904     regs.pc = env->nip;
905 
906     regs.srr0 = env->spr[SPR_SRR0];
907     regs.srr1 = env->spr[SPR_SRR1];
908 
909     regs.sprg0 = env->spr[SPR_SPRG0];
910     regs.sprg1 = env->spr[SPR_SPRG1];
911     regs.sprg2 = env->spr[SPR_SPRG2];
912     regs.sprg3 = env->spr[SPR_SPRG3];
913     regs.sprg4 = env->spr[SPR_SPRG4];
914     regs.sprg5 = env->spr[SPR_SPRG5];
915     regs.sprg6 = env->spr[SPR_SPRG6];
916     regs.sprg7 = env->spr[SPR_SPRG7];
917 
918     regs.pid = env->spr[SPR_BOOKE_PID];
919 
920     for (i = 0; i < 32; i++) {
921         regs.gpr[i] = env->gpr[i];
922     }
923 
924     regs.cr = 0;
925     for (i = 0; i < 8; i++) {
926         regs.cr |= (env->crf[i] & 15) << (4 * (7 - i));
927     }
928 
929     ret = kvm_vcpu_ioctl(cs, KVM_SET_REGS, &regs);
930     if (ret < 0) {
931         return ret;
932     }
933 
934     kvm_put_fp(cs);
935 
936     if (env->tlb_dirty) {
937         kvm_sw_tlb_put(cpu);
938         env->tlb_dirty = false;
939     }
940 
941     if (cap_segstate && (level >= KVM_PUT_RESET_STATE)) {
942         ret = kvmppc_put_books_sregs(cpu);
943         if (ret < 0) {
944             return ret;
945         }
946     }
947 
948     if (cap_hior && (level >= KVM_PUT_RESET_STATE)) {
949         kvm_put_one_spr(cs, KVM_REG_PPC_HIOR, SPR_HIOR);
950     }
951 
952     if (cap_one_reg) {
953         int i;
954 
955         /*
956          * We deliberately ignore errors here, for kernels which have
957          * the ONE_REG calls, but don't support the specific
958          * registers, there's a reasonable chance things will still
959          * work, at least until we try to migrate.
960          */
961         for (i = 0; i < 1024; i++) {
962             uint64_t id = env->spr_cb[i].one_reg_id;
963 
964             if (id != 0) {
965                 kvm_put_one_spr(cs, id, i);
966             }
967         }
968 
969 #ifdef TARGET_PPC64
970         if (msr_ts) {
971             for (i = 0; i < ARRAY_SIZE(env->tm_gpr); i++) {
972                 kvm_set_one_reg(cs, KVM_REG_PPC_TM_GPR(i), &env->tm_gpr[i]);
973             }
974             for (i = 0; i < ARRAY_SIZE(env->tm_vsr); i++) {
975                 kvm_set_one_reg(cs, KVM_REG_PPC_TM_VSR(i), &env->tm_vsr[i]);
976             }
977             kvm_set_one_reg(cs, KVM_REG_PPC_TM_CR, &env->tm_cr);
978             kvm_set_one_reg(cs, KVM_REG_PPC_TM_LR, &env->tm_lr);
979             kvm_set_one_reg(cs, KVM_REG_PPC_TM_CTR, &env->tm_ctr);
980             kvm_set_one_reg(cs, KVM_REG_PPC_TM_FPSCR, &env->tm_fpscr);
981             kvm_set_one_reg(cs, KVM_REG_PPC_TM_AMR, &env->tm_amr);
982             kvm_set_one_reg(cs, KVM_REG_PPC_TM_PPR, &env->tm_ppr);
983             kvm_set_one_reg(cs, KVM_REG_PPC_TM_VRSAVE, &env->tm_vrsave);
984             kvm_set_one_reg(cs, KVM_REG_PPC_TM_VSCR, &env->tm_vscr);
985             kvm_set_one_reg(cs, KVM_REG_PPC_TM_DSCR, &env->tm_dscr);
986             kvm_set_one_reg(cs, KVM_REG_PPC_TM_TAR, &env->tm_tar);
987         }
988 
989         if (cap_papr) {
990             if (kvm_put_vpa(cs) < 0) {
991                 trace_kvm_failed_put_vpa();
992             }
993         }
994 
995         kvm_set_one_reg(cs, KVM_REG_PPC_TB_OFFSET, &env->tb_env->tb_offset);
996 
997         if (level > KVM_PUT_RUNTIME_STATE) {
998             kvm_put_one_spr(cs, KVM_REG_PPC_DPDES, SPR_DPDES);
999         }
1000 #endif /* TARGET_PPC64 */
1001     }
1002 
1003     return ret;
1004 }
1005 
1006 static void kvm_sync_excp(CPUPPCState *env, int vector, int ivor)
1007 {
1008      env->excp_vectors[vector] = env->spr[ivor] + env->spr[SPR_BOOKE_IVPR];
1009 }
1010 
1011 static int kvmppc_get_booke_sregs(PowerPCCPU *cpu)
1012 {
1013     CPUPPCState *env = &cpu->env;
1014     struct kvm_sregs sregs;
1015     int ret;
1016 
1017     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_SREGS, &sregs);
1018     if (ret < 0) {
1019         return ret;
1020     }
1021 
1022     if (sregs.u.e.features & KVM_SREGS_E_BASE) {
1023         env->spr[SPR_BOOKE_CSRR0] = sregs.u.e.csrr0;
1024         env->spr[SPR_BOOKE_CSRR1] = sregs.u.e.csrr1;
1025         env->spr[SPR_BOOKE_ESR] = sregs.u.e.esr;
1026         env->spr[SPR_BOOKE_DEAR] = sregs.u.e.dear;
1027         env->spr[SPR_BOOKE_MCSR] = sregs.u.e.mcsr;
1028         env->spr[SPR_BOOKE_TSR] = sregs.u.e.tsr;
1029         env->spr[SPR_BOOKE_TCR] = sregs.u.e.tcr;
1030         env->spr[SPR_DECR] = sregs.u.e.dec;
1031         env->spr[SPR_TBL] = sregs.u.e.tb & 0xffffffff;
1032         env->spr[SPR_TBU] = sregs.u.e.tb >> 32;
1033         env->spr[SPR_VRSAVE] = sregs.u.e.vrsave;
1034     }
1035 
1036     if (sregs.u.e.features & KVM_SREGS_E_ARCH206) {
1037         env->spr[SPR_BOOKE_PIR] = sregs.u.e.pir;
1038         env->spr[SPR_BOOKE_MCSRR0] = sregs.u.e.mcsrr0;
1039         env->spr[SPR_BOOKE_MCSRR1] = sregs.u.e.mcsrr1;
1040         env->spr[SPR_BOOKE_DECAR] = sregs.u.e.decar;
1041         env->spr[SPR_BOOKE_IVPR] = sregs.u.e.ivpr;
1042     }
1043 
1044     if (sregs.u.e.features & KVM_SREGS_E_64) {
1045         env->spr[SPR_BOOKE_EPCR] = sregs.u.e.epcr;
1046     }
1047 
1048     if (sregs.u.e.features & KVM_SREGS_E_SPRG8) {
1049         env->spr[SPR_BOOKE_SPRG8] = sregs.u.e.sprg8;
1050     }
1051 
1052     if (sregs.u.e.features & KVM_SREGS_E_IVOR) {
1053         env->spr[SPR_BOOKE_IVOR0] = sregs.u.e.ivor_low[0];
1054         kvm_sync_excp(env, POWERPC_EXCP_CRITICAL,  SPR_BOOKE_IVOR0);
1055         env->spr[SPR_BOOKE_IVOR1] = sregs.u.e.ivor_low[1];
1056         kvm_sync_excp(env, POWERPC_EXCP_MCHECK,  SPR_BOOKE_IVOR1);
1057         env->spr[SPR_BOOKE_IVOR2] = sregs.u.e.ivor_low[2];
1058         kvm_sync_excp(env, POWERPC_EXCP_DSI,  SPR_BOOKE_IVOR2);
1059         env->spr[SPR_BOOKE_IVOR3] = sregs.u.e.ivor_low[3];
1060         kvm_sync_excp(env, POWERPC_EXCP_ISI,  SPR_BOOKE_IVOR3);
1061         env->spr[SPR_BOOKE_IVOR4] = sregs.u.e.ivor_low[4];
1062         kvm_sync_excp(env, POWERPC_EXCP_EXTERNAL,  SPR_BOOKE_IVOR4);
1063         env->spr[SPR_BOOKE_IVOR5] = sregs.u.e.ivor_low[5];
1064         kvm_sync_excp(env, POWERPC_EXCP_ALIGN,  SPR_BOOKE_IVOR5);
1065         env->spr[SPR_BOOKE_IVOR6] = sregs.u.e.ivor_low[6];
1066         kvm_sync_excp(env, POWERPC_EXCP_PROGRAM,  SPR_BOOKE_IVOR6);
1067         env->spr[SPR_BOOKE_IVOR7] = sregs.u.e.ivor_low[7];
1068         kvm_sync_excp(env, POWERPC_EXCP_FPU,  SPR_BOOKE_IVOR7);
1069         env->spr[SPR_BOOKE_IVOR8] = sregs.u.e.ivor_low[8];
1070         kvm_sync_excp(env, POWERPC_EXCP_SYSCALL,  SPR_BOOKE_IVOR8);
1071         env->spr[SPR_BOOKE_IVOR9] = sregs.u.e.ivor_low[9];
1072         kvm_sync_excp(env, POWERPC_EXCP_APU,  SPR_BOOKE_IVOR9);
1073         env->spr[SPR_BOOKE_IVOR10] = sregs.u.e.ivor_low[10];
1074         kvm_sync_excp(env, POWERPC_EXCP_DECR,  SPR_BOOKE_IVOR10);
1075         env->spr[SPR_BOOKE_IVOR11] = sregs.u.e.ivor_low[11];
1076         kvm_sync_excp(env, POWERPC_EXCP_FIT,  SPR_BOOKE_IVOR11);
1077         env->spr[SPR_BOOKE_IVOR12] = sregs.u.e.ivor_low[12];
1078         kvm_sync_excp(env, POWERPC_EXCP_WDT,  SPR_BOOKE_IVOR12);
1079         env->spr[SPR_BOOKE_IVOR13] = sregs.u.e.ivor_low[13];
1080         kvm_sync_excp(env, POWERPC_EXCP_DTLB,  SPR_BOOKE_IVOR13);
1081         env->spr[SPR_BOOKE_IVOR14] = sregs.u.e.ivor_low[14];
1082         kvm_sync_excp(env, POWERPC_EXCP_ITLB,  SPR_BOOKE_IVOR14);
1083         env->spr[SPR_BOOKE_IVOR15] = sregs.u.e.ivor_low[15];
1084         kvm_sync_excp(env, POWERPC_EXCP_DEBUG,  SPR_BOOKE_IVOR15);
1085 
1086         if (sregs.u.e.features & KVM_SREGS_E_SPE) {
1087             env->spr[SPR_BOOKE_IVOR32] = sregs.u.e.ivor_high[0];
1088             kvm_sync_excp(env, POWERPC_EXCP_SPEU,  SPR_BOOKE_IVOR32);
1089             env->spr[SPR_BOOKE_IVOR33] = sregs.u.e.ivor_high[1];
1090             kvm_sync_excp(env, POWERPC_EXCP_EFPDI,  SPR_BOOKE_IVOR33);
1091             env->spr[SPR_BOOKE_IVOR34] = sregs.u.e.ivor_high[2];
1092             kvm_sync_excp(env, POWERPC_EXCP_EFPRI,  SPR_BOOKE_IVOR34);
1093         }
1094 
1095         if (sregs.u.e.features & KVM_SREGS_E_PM) {
1096             env->spr[SPR_BOOKE_IVOR35] = sregs.u.e.ivor_high[3];
1097             kvm_sync_excp(env, POWERPC_EXCP_EPERFM,  SPR_BOOKE_IVOR35);
1098         }
1099 
1100         if (sregs.u.e.features & KVM_SREGS_E_PC) {
1101             env->spr[SPR_BOOKE_IVOR36] = sregs.u.e.ivor_high[4];
1102             kvm_sync_excp(env, POWERPC_EXCP_DOORI,  SPR_BOOKE_IVOR36);
1103             env->spr[SPR_BOOKE_IVOR37] = sregs.u.e.ivor_high[5];
1104             kvm_sync_excp(env, POWERPC_EXCP_DOORCI, SPR_BOOKE_IVOR37);
1105         }
1106     }
1107 
1108     if (sregs.u.e.features & KVM_SREGS_E_ARCH206_MMU) {
1109         env->spr[SPR_BOOKE_MAS0] = sregs.u.e.mas0;
1110         env->spr[SPR_BOOKE_MAS1] = sregs.u.e.mas1;
1111         env->spr[SPR_BOOKE_MAS2] = sregs.u.e.mas2;
1112         env->spr[SPR_BOOKE_MAS3] = sregs.u.e.mas7_3 & 0xffffffff;
1113         env->spr[SPR_BOOKE_MAS4] = sregs.u.e.mas4;
1114         env->spr[SPR_BOOKE_MAS6] = sregs.u.e.mas6;
1115         env->spr[SPR_BOOKE_MAS7] = sregs.u.e.mas7_3 >> 32;
1116         env->spr[SPR_MMUCFG] = sregs.u.e.mmucfg;
1117         env->spr[SPR_BOOKE_TLB0CFG] = sregs.u.e.tlbcfg[0];
1118         env->spr[SPR_BOOKE_TLB1CFG] = sregs.u.e.tlbcfg[1];
1119     }
1120 
1121     if (sregs.u.e.features & KVM_SREGS_EXP) {
1122         env->spr[SPR_BOOKE_EPR] = sregs.u.e.epr;
1123     }
1124 
1125     if (sregs.u.e.features & KVM_SREGS_E_PD) {
1126         env->spr[SPR_BOOKE_EPLC] = sregs.u.e.eplc;
1127         env->spr[SPR_BOOKE_EPSC] = sregs.u.e.epsc;
1128     }
1129 
1130     if (sregs.u.e.impl_id == KVM_SREGS_E_IMPL_FSL) {
1131         env->spr[SPR_E500_SVR] = sregs.u.e.impl.fsl.svr;
1132         env->spr[SPR_Exxx_MCAR] = sregs.u.e.impl.fsl.mcar;
1133         env->spr[SPR_HID0] = sregs.u.e.impl.fsl.hid0;
1134 
1135         if (sregs.u.e.impl.fsl.features & KVM_SREGS_E_FSL_PIDn) {
1136             env->spr[SPR_BOOKE_PID1] = sregs.u.e.impl.fsl.pid1;
1137             env->spr[SPR_BOOKE_PID2] = sregs.u.e.impl.fsl.pid2;
1138         }
1139     }
1140 
1141     return 0;
1142 }
1143 
1144 static int kvmppc_get_books_sregs(PowerPCCPU *cpu)
1145 {
1146     CPUPPCState *env = &cpu->env;
1147     struct kvm_sregs sregs;
1148     int ret;
1149     int i;
1150 
1151     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_SREGS, &sregs);
1152     if (ret < 0) {
1153         return ret;
1154     }
1155 
1156     if (!cpu->vhyp) {
1157         ppc_store_sdr1(env, sregs.u.s.sdr1);
1158     }
1159 
1160     /* Sync SLB */
1161 #ifdef TARGET_PPC64
1162     /*
1163      * The packed SLB array we get from KVM_GET_SREGS only contains
1164      * information about valid entries. So we flush our internal copy
1165      * to get rid of stale ones, then put all valid SLB entries back
1166      * in.
1167      */
1168     memset(env->slb, 0, sizeof(env->slb));
1169     for (i = 0; i < ARRAY_SIZE(env->slb); i++) {
1170         target_ulong rb = sregs.u.s.ppc64.slb[i].slbe;
1171         target_ulong rs = sregs.u.s.ppc64.slb[i].slbv;
1172         /*
1173          * Only restore valid entries
1174          */
1175         if (rb & SLB_ESID_V) {
1176             ppc_store_slb(cpu, rb & 0xfff, rb & ~0xfffULL, rs);
1177         }
1178     }
1179 #endif
1180 
1181     /* Sync SRs */
1182     for (i = 0; i < 16; i++) {
1183         env->sr[i] = sregs.u.s.ppc32.sr[i];
1184     }
1185 
1186     /* Sync BATs */
1187     for (i = 0; i < 8; i++) {
1188         env->DBAT[0][i] = sregs.u.s.ppc32.dbat[i] & 0xffffffff;
1189         env->DBAT[1][i] = sregs.u.s.ppc32.dbat[i] >> 32;
1190         env->IBAT[0][i] = sregs.u.s.ppc32.ibat[i] & 0xffffffff;
1191         env->IBAT[1][i] = sregs.u.s.ppc32.ibat[i] >> 32;
1192     }
1193 
1194     return 0;
1195 }
1196 
1197 int kvm_arch_get_registers(CPUState *cs)
1198 {
1199     PowerPCCPU *cpu = POWERPC_CPU(cs);
1200     CPUPPCState *env = &cpu->env;
1201     struct kvm_regs regs;
1202     uint32_t cr;
1203     int i, ret;
1204 
1205     ret = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
1206     if (ret < 0) {
1207         return ret;
1208     }
1209 
1210     cr = regs.cr;
1211     for (i = 7; i >= 0; i--) {
1212         env->crf[i] = cr & 15;
1213         cr >>= 4;
1214     }
1215 
1216     env->ctr = regs.ctr;
1217     env->lr = regs.lr;
1218     cpu_write_xer(env, regs.xer);
1219     env->msr = regs.msr;
1220     env->nip = regs.pc;
1221 
1222     env->spr[SPR_SRR0] = regs.srr0;
1223     env->spr[SPR_SRR1] = regs.srr1;
1224 
1225     env->spr[SPR_SPRG0] = regs.sprg0;
1226     env->spr[SPR_SPRG1] = regs.sprg1;
1227     env->spr[SPR_SPRG2] = regs.sprg2;
1228     env->spr[SPR_SPRG3] = regs.sprg3;
1229     env->spr[SPR_SPRG4] = regs.sprg4;
1230     env->spr[SPR_SPRG5] = regs.sprg5;
1231     env->spr[SPR_SPRG6] = regs.sprg6;
1232     env->spr[SPR_SPRG7] = regs.sprg7;
1233 
1234     env->spr[SPR_BOOKE_PID] = regs.pid;
1235 
1236     for (i = 0; i < 32; i++) {
1237         env->gpr[i] = regs.gpr[i];
1238     }
1239 
1240     kvm_get_fp(cs);
1241 
1242     if (cap_booke_sregs) {
1243         ret = kvmppc_get_booke_sregs(cpu);
1244         if (ret < 0) {
1245             return ret;
1246         }
1247     }
1248 
1249     if (cap_segstate) {
1250         ret = kvmppc_get_books_sregs(cpu);
1251         if (ret < 0) {
1252             return ret;
1253         }
1254     }
1255 
1256     if (cap_hior) {
1257         kvm_get_one_spr(cs, KVM_REG_PPC_HIOR, SPR_HIOR);
1258     }
1259 
1260     if (cap_one_reg) {
1261         int i;
1262 
1263         /*
1264          * We deliberately ignore errors here, for kernels which have
1265          * the ONE_REG calls, but don't support the specific
1266          * registers, there's a reasonable chance things will still
1267          * work, at least until we try to migrate.
1268          */
1269         for (i = 0; i < 1024; i++) {
1270             uint64_t id = env->spr_cb[i].one_reg_id;
1271 
1272             if (id != 0) {
1273                 kvm_get_one_spr(cs, id, i);
1274             }
1275         }
1276 
1277 #ifdef TARGET_PPC64
1278         if (msr_ts) {
1279             for (i = 0; i < ARRAY_SIZE(env->tm_gpr); i++) {
1280                 kvm_get_one_reg(cs, KVM_REG_PPC_TM_GPR(i), &env->tm_gpr[i]);
1281             }
1282             for (i = 0; i < ARRAY_SIZE(env->tm_vsr); i++) {
1283                 kvm_get_one_reg(cs, KVM_REG_PPC_TM_VSR(i), &env->tm_vsr[i]);
1284             }
1285             kvm_get_one_reg(cs, KVM_REG_PPC_TM_CR, &env->tm_cr);
1286             kvm_get_one_reg(cs, KVM_REG_PPC_TM_LR, &env->tm_lr);
1287             kvm_get_one_reg(cs, KVM_REG_PPC_TM_CTR, &env->tm_ctr);
1288             kvm_get_one_reg(cs, KVM_REG_PPC_TM_FPSCR, &env->tm_fpscr);
1289             kvm_get_one_reg(cs, KVM_REG_PPC_TM_AMR, &env->tm_amr);
1290             kvm_get_one_reg(cs, KVM_REG_PPC_TM_PPR, &env->tm_ppr);
1291             kvm_get_one_reg(cs, KVM_REG_PPC_TM_VRSAVE, &env->tm_vrsave);
1292             kvm_get_one_reg(cs, KVM_REG_PPC_TM_VSCR, &env->tm_vscr);
1293             kvm_get_one_reg(cs, KVM_REG_PPC_TM_DSCR, &env->tm_dscr);
1294             kvm_get_one_reg(cs, KVM_REG_PPC_TM_TAR, &env->tm_tar);
1295         }
1296 
1297         if (cap_papr) {
1298             if (kvm_get_vpa(cs) < 0) {
1299                 trace_kvm_failed_get_vpa();
1300             }
1301         }
1302 
1303         kvm_get_one_reg(cs, KVM_REG_PPC_TB_OFFSET, &env->tb_env->tb_offset);
1304         kvm_get_one_spr(cs, KVM_REG_PPC_DPDES, SPR_DPDES);
1305 #endif
1306     }
1307 
1308     return 0;
1309 }
1310 
1311 int kvmppc_set_interrupt(PowerPCCPU *cpu, int irq, int level)
1312 {
1313     unsigned virq = level ? KVM_INTERRUPT_SET_LEVEL : KVM_INTERRUPT_UNSET;
1314 
1315     if (irq != PPC_INTERRUPT_EXT) {
1316         return 0;
1317     }
1318 
1319     if (!kvm_enabled() || !cap_interrupt_unset) {
1320         return 0;
1321     }
1322 
1323     kvm_vcpu_ioctl(CPU(cpu), KVM_INTERRUPT, &virq);
1324 
1325     return 0;
1326 }
1327 
1328 void kvm_arch_pre_run(CPUState *cs, struct kvm_run *run)
1329 {
1330     return;
1331 }
1332 
1333 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
1334 {
1335     return MEMTXATTRS_UNSPECIFIED;
1336 }
1337 
1338 int kvm_arch_process_async_events(CPUState *cs)
1339 {
1340     return cs->halted;
1341 }
1342 
1343 static int kvmppc_handle_halt(PowerPCCPU *cpu)
1344 {
1345     CPUState *cs = CPU(cpu);
1346     CPUPPCState *env = &cpu->env;
1347 
1348     if (!(cs->interrupt_request & CPU_INTERRUPT_HARD) && (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 = 0;
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 = 1;
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 = 1;
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 1;
1594 }
1595 
1596 static int kvm_handle_sw_breakpoint(void)
1597 {
1598     return 1;
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 0;
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();
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     default:
1706         fprintf(stderr, "KVM: unknown exit reason %d\n", run->exit_reason);
1707         ret = -1;
1708         break;
1709     }
1710 
1711     qemu_mutex_unlock_iothread();
1712     return ret;
1713 }
1714 
1715 int kvmppc_or_tsr_bits(PowerPCCPU *cpu, uint32_t tsr_bits)
1716 {
1717     CPUState *cs = CPU(cpu);
1718     uint32_t bits = tsr_bits;
1719     struct kvm_one_reg reg = {
1720         .id = KVM_REG_PPC_OR_TSR,
1721         .addr = (uintptr_t) &bits,
1722     };
1723 
1724     return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
1725 }
1726 
1727 int kvmppc_clear_tsr_bits(PowerPCCPU *cpu, uint32_t tsr_bits)
1728 {
1729 
1730     CPUState *cs = CPU(cpu);
1731     uint32_t bits = tsr_bits;
1732     struct kvm_one_reg reg = {
1733         .id = KVM_REG_PPC_CLEAR_TSR,
1734         .addr = (uintptr_t) &bits,
1735     };
1736 
1737     return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
1738 }
1739 
1740 int kvmppc_set_tcr(PowerPCCPU *cpu)
1741 {
1742     CPUState *cs = CPU(cpu);
1743     CPUPPCState *env = &cpu->env;
1744     uint32_t tcr = env->spr[SPR_BOOKE_TCR];
1745 
1746     struct kvm_one_reg reg = {
1747         .id = KVM_REG_PPC_TCR,
1748         .addr = (uintptr_t) &tcr,
1749     };
1750 
1751     return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
1752 }
1753 
1754 int kvmppc_booke_watchdog_enable(PowerPCCPU *cpu)
1755 {
1756     CPUState *cs = CPU(cpu);
1757     int ret;
1758 
1759     if (!kvm_enabled()) {
1760         return -1;
1761     }
1762 
1763     if (!cap_ppc_watchdog) {
1764         printf("warning: KVM does not support watchdog");
1765         return -1;
1766     }
1767 
1768     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_PPC_BOOKE_WATCHDOG, 0);
1769     if (ret < 0) {
1770         fprintf(stderr, "%s: couldn't enable KVM_CAP_PPC_BOOKE_WATCHDOG: %s\n",
1771                 __func__, strerror(-ret));
1772         return ret;
1773     }
1774 
1775     return ret;
1776 }
1777 
1778 static int read_cpuinfo(const char *field, char *value, int len)
1779 {
1780     FILE *f;
1781     int ret = -1;
1782     int field_len = strlen(field);
1783     char line[512];
1784 
1785     f = fopen("/proc/cpuinfo", "r");
1786     if (!f) {
1787         return -1;
1788     }
1789 
1790     do {
1791         if (!fgets(line, sizeof(line), f)) {
1792             break;
1793         }
1794         if (!strncmp(line, field, field_len)) {
1795             pstrcpy(value, len, line);
1796             ret = 0;
1797             break;
1798         }
1799     } while (*line);
1800 
1801     fclose(f);
1802 
1803     return ret;
1804 }
1805 
1806 uint32_t kvmppc_get_tbfreq(void)
1807 {
1808     char line[512];
1809     char *ns;
1810     uint32_t retval = NANOSECONDS_PER_SECOND;
1811 
1812     if (read_cpuinfo("timebase", line, sizeof(line))) {
1813         return retval;
1814     }
1815 
1816     ns = strchr(line, ':');
1817     if (!ns) {
1818         return retval;
1819     }
1820 
1821     ns++;
1822 
1823     return atoi(ns);
1824 }
1825 
1826 bool kvmppc_get_host_serial(char **value)
1827 {
1828     return g_file_get_contents("/proc/device-tree/system-id", value, NULL,
1829                                NULL);
1830 }
1831 
1832 bool kvmppc_get_host_model(char **value)
1833 {
1834     return g_file_get_contents("/proc/device-tree/model", value, NULL, NULL);
1835 }
1836 
1837 /* Try to find a device tree node for a CPU with clock-frequency property */
1838 static int kvmppc_find_cpu_dt(char *buf, int buf_len)
1839 {
1840     struct dirent *dirp;
1841     DIR *dp;
1842 
1843     dp = opendir(PROC_DEVTREE_CPU);
1844     if (!dp) {
1845         printf("Can't open directory " PROC_DEVTREE_CPU "\n");
1846         return -1;
1847     }
1848 
1849     buf[0] = '\0';
1850     while ((dirp = readdir(dp)) != NULL) {
1851         FILE *f;
1852         snprintf(buf, buf_len, "%s%s/clock-frequency", PROC_DEVTREE_CPU,
1853                  dirp->d_name);
1854         f = fopen(buf, "r");
1855         if (f) {
1856             snprintf(buf, buf_len, "%s%s", PROC_DEVTREE_CPU, dirp->d_name);
1857             fclose(f);
1858             break;
1859         }
1860         buf[0] = '\0';
1861     }
1862     closedir(dp);
1863     if (buf[0] == '\0') {
1864         printf("Unknown host!\n");
1865         return -1;
1866     }
1867 
1868     return 0;
1869 }
1870 
1871 static uint64_t kvmppc_read_int_dt(const char *filename)
1872 {
1873     union {
1874         uint32_t v32;
1875         uint64_t v64;
1876     } u;
1877     FILE *f;
1878     int len;
1879 
1880     f = fopen(filename, "rb");
1881     if (!f) {
1882         return -1;
1883     }
1884 
1885     len = fread(&u, 1, sizeof(u), f);
1886     fclose(f);
1887     switch (len) {
1888     case 4:
1889         /* property is a 32-bit quantity */
1890         return be32_to_cpu(u.v32);
1891     case 8:
1892         return be64_to_cpu(u.v64);
1893     }
1894 
1895     return 0;
1896 }
1897 
1898 /*
1899  * Read a CPU node property from the host device tree that's a single
1900  * integer (32-bit or 64-bit).  Returns 0 if anything goes wrong
1901  * (can't find or open the property, or doesn't understand the format)
1902  */
1903 static uint64_t kvmppc_read_int_cpu_dt(const char *propname)
1904 {
1905     char buf[PATH_MAX], *tmp;
1906     uint64_t val;
1907 
1908     if (kvmppc_find_cpu_dt(buf, sizeof(buf))) {
1909         return -1;
1910     }
1911 
1912     tmp = g_strdup_printf("%s/%s", buf, propname);
1913     val = kvmppc_read_int_dt(tmp);
1914     g_free(tmp);
1915 
1916     return val;
1917 }
1918 
1919 uint64_t kvmppc_get_clockfreq(void)
1920 {
1921     return kvmppc_read_int_cpu_dt("clock-frequency");
1922 }
1923 
1924 static int kvmppc_get_dec_bits(void)
1925 {
1926     int nr_bits = kvmppc_read_int_cpu_dt("ibm,dec-bits");
1927 
1928     if (nr_bits > 0) {
1929         return nr_bits;
1930     }
1931     return 0;
1932 }
1933 
1934 static int kvmppc_get_pvinfo(CPUPPCState *env, struct kvm_ppc_pvinfo *pvinfo)
1935 {
1936     CPUState *cs = env_cpu(env);
1937 
1938     if (kvm_vm_check_extension(cs->kvm_state, KVM_CAP_PPC_GET_PVINFO) &&
1939         !kvm_vm_ioctl(cs->kvm_state, KVM_PPC_GET_PVINFO, pvinfo)) {
1940         return 0;
1941     }
1942 
1943     return 1;
1944 }
1945 
1946 int kvmppc_get_hasidle(CPUPPCState *env)
1947 {
1948     struct kvm_ppc_pvinfo pvinfo;
1949 
1950     if (!kvmppc_get_pvinfo(env, &pvinfo) &&
1951         (pvinfo.flags & KVM_PPC_PVINFO_FLAGS_EV_IDLE)) {
1952         return 1;
1953     }
1954 
1955     return 0;
1956 }
1957 
1958 int kvmppc_get_hypercall(CPUPPCState *env, uint8_t *buf, int buf_len)
1959 {
1960     uint32_t *hc = (uint32_t *)buf;
1961     struct kvm_ppc_pvinfo pvinfo;
1962 
1963     if (!kvmppc_get_pvinfo(env, &pvinfo)) {
1964         memcpy(buf, pvinfo.hcall, buf_len);
1965         return 0;
1966     }
1967 
1968     /*
1969      * Fallback to always fail hypercalls regardless of endianness:
1970      *
1971      *     tdi 0,r0,72 (becomes b .+8 in wrong endian, nop in good endian)
1972      *     li r3, -1
1973      *     b .+8       (becomes nop in wrong endian)
1974      *     bswap32(li r3, -1)
1975      */
1976 
1977     hc[0] = cpu_to_be32(0x08000048);
1978     hc[1] = cpu_to_be32(0x3860ffff);
1979     hc[2] = cpu_to_be32(0x48000008);
1980     hc[3] = cpu_to_be32(bswap32(0x3860ffff));
1981 
1982     return 1;
1983 }
1984 
1985 static inline int kvmppc_enable_hcall(KVMState *s, target_ulong hcall)
1986 {
1987     return kvm_vm_enable_cap(s, KVM_CAP_PPC_ENABLE_HCALL, 0, hcall, 1);
1988 }
1989 
1990 void kvmppc_enable_logical_ci_hcalls(void)
1991 {
1992     /*
1993      * FIXME: it would be nice if we could detect the cases where
1994      * we're using a device which requires the in kernel
1995      * implementation of these hcalls, but the kernel lacks them and
1996      * produce a warning.
1997      */
1998     kvmppc_enable_hcall(kvm_state, H_LOGICAL_CI_LOAD);
1999     kvmppc_enable_hcall(kvm_state, H_LOGICAL_CI_STORE);
2000 }
2001 
2002 void kvmppc_enable_set_mode_hcall(void)
2003 {
2004     kvmppc_enable_hcall(kvm_state, H_SET_MODE);
2005 }
2006 
2007 void kvmppc_enable_clear_ref_mod_hcalls(void)
2008 {
2009     kvmppc_enable_hcall(kvm_state, H_CLEAR_REF);
2010     kvmppc_enable_hcall(kvm_state, H_CLEAR_MOD);
2011 }
2012 
2013 void kvmppc_enable_h_page_init(void)
2014 {
2015     kvmppc_enable_hcall(kvm_state, H_PAGE_INIT);
2016 }
2017 
2018 void kvmppc_set_papr(PowerPCCPU *cpu)
2019 {
2020     CPUState *cs = CPU(cpu);
2021     int ret;
2022 
2023     if (!kvm_enabled()) {
2024         return;
2025     }
2026 
2027     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_PPC_PAPR, 0);
2028     if (ret) {
2029         error_report("This vCPU type or KVM version does not support PAPR");
2030         exit(1);
2031     }
2032 
2033     /*
2034      * Update the capability flag so we sync the right information
2035      * with kvm
2036      */
2037     cap_papr = 1;
2038 }
2039 
2040 int kvmppc_set_compat(PowerPCCPU *cpu, uint32_t compat_pvr)
2041 {
2042     return kvm_set_one_reg(CPU(cpu), KVM_REG_PPC_ARCH_COMPAT, &compat_pvr);
2043 }
2044 
2045 void kvmppc_set_mpic_proxy(PowerPCCPU *cpu, int mpic_proxy)
2046 {
2047     CPUState *cs = CPU(cpu);
2048     int ret;
2049 
2050     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_PPC_EPR, 0, mpic_proxy);
2051     if (ret && mpic_proxy) {
2052         error_report("This KVM version does not support EPR");
2053         exit(1);
2054     }
2055 }
2056 
2057 int kvmppc_smt_threads(void)
2058 {
2059     return cap_ppc_smt ? cap_ppc_smt : 1;
2060 }
2061 
2062 int kvmppc_set_smt_threads(int smt)
2063 {
2064     int ret;
2065 
2066     ret = kvm_vm_enable_cap(kvm_state, KVM_CAP_PPC_SMT, 0, smt, 0);
2067     if (!ret) {
2068         cap_ppc_smt = smt;
2069     }
2070     return ret;
2071 }
2072 
2073 void kvmppc_error_append_smt_possible_hint(Error *const *errp)
2074 {
2075     int i;
2076     GString *g;
2077     char *s;
2078 
2079     assert(kvm_enabled());
2080     if (cap_ppc_smt_possible) {
2081         g = g_string_new("Available VSMT modes:");
2082         for (i = 63; i >= 0; i--) {
2083             if ((1UL << i) & cap_ppc_smt_possible) {
2084                 g_string_append_printf(g, " %lu", (1UL << i));
2085             }
2086         }
2087         s = g_string_free(g, false);
2088         error_append_hint(errp, "%s.\n", s);
2089         g_free(s);
2090     } else {
2091         error_append_hint(errp,
2092                           "This KVM seems to be too old to support VSMT.\n");
2093     }
2094 }
2095 
2096 
2097 #ifdef TARGET_PPC64
2098 uint64_t kvmppc_rma_size(uint64_t current_size, unsigned int hash_shift)
2099 {
2100     struct kvm_ppc_smmu_info info;
2101     long rampagesize, best_page_shift;
2102     int i;
2103 
2104     /*
2105      * Find the largest hardware supported page size that's less than
2106      * or equal to the (logical) backing page size of guest RAM
2107      */
2108     kvm_get_smmu_info(&info, &error_fatal);
2109     rampagesize = qemu_minrampagesize();
2110     best_page_shift = 0;
2111 
2112     for (i = 0; i < KVM_PPC_PAGE_SIZES_MAX_SZ; i++) {
2113         struct kvm_ppc_one_seg_page_size *sps = &info.sps[i];
2114 
2115         if (!sps->page_shift) {
2116             continue;
2117         }
2118 
2119         if ((sps->page_shift > best_page_shift)
2120             && ((1UL << sps->page_shift) <= rampagesize)) {
2121             best_page_shift = sps->page_shift;
2122         }
2123     }
2124 
2125     return MIN(current_size,
2126                1ULL << (best_page_shift + hash_shift - 7));
2127 }
2128 #endif
2129 
2130 bool kvmppc_spapr_use_multitce(void)
2131 {
2132     return cap_spapr_multitce;
2133 }
2134 
2135 int kvmppc_spapr_enable_inkernel_multitce(void)
2136 {
2137     int ret;
2138 
2139     ret = kvm_vm_enable_cap(kvm_state, KVM_CAP_PPC_ENABLE_HCALL, 0,
2140                             H_PUT_TCE_INDIRECT, 1);
2141     if (!ret) {
2142         ret = kvm_vm_enable_cap(kvm_state, KVM_CAP_PPC_ENABLE_HCALL, 0,
2143                                 H_STUFF_TCE, 1);
2144     }
2145 
2146     return ret;
2147 }
2148 
2149 void *kvmppc_create_spapr_tce(uint32_t liobn, uint32_t page_shift,
2150                               uint64_t bus_offset, uint32_t nb_table,
2151                               int *pfd, bool need_vfio)
2152 {
2153     long len;
2154     int fd;
2155     void *table;
2156 
2157     /*
2158      * Must set fd to -1 so we don't try to munmap when called for
2159      * destroying the table, which the upper layers -will- do
2160      */
2161     *pfd = -1;
2162     if (!cap_spapr_tce || (need_vfio && !cap_spapr_vfio)) {
2163         return NULL;
2164     }
2165 
2166     if (cap_spapr_tce_64) {
2167         struct kvm_create_spapr_tce_64 args = {
2168             .liobn = liobn,
2169             .page_shift = page_shift,
2170             .offset = bus_offset >> page_shift,
2171             .size = nb_table,
2172             .flags = 0
2173         };
2174         fd = kvm_vm_ioctl(kvm_state, KVM_CREATE_SPAPR_TCE_64, &args);
2175         if (fd < 0) {
2176             fprintf(stderr,
2177                     "KVM: Failed to create TCE64 table for liobn 0x%x\n",
2178                     liobn);
2179             return NULL;
2180         }
2181     } else if (cap_spapr_tce) {
2182         uint64_t window_size = (uint64_t) nb_table << page_shift;
2183         struct kvm_create_spapr_tce args = {
2184             .liobn = liobn,
2185             .window_size = window_size,
2186         };
2187         if ((window_size != args.window_size) || bus_offset) {
2188             return NULL;
2189         }
2190         fd = kvm_vm_ioctl(kvm_state, KVM_CREATE_SPAPR_TCE, &args);
2191         if (fd < 0) {
2192             fprintf(stderr, "KVM: Failed to create TCE table for liobn 0x%x\n",
2193                     liobn);
2194             return NULL;
2195         }
2196     } else {
2197         return NULL;
2198     }
2199 
2200     len = nb_table * sizeof(uint64_t);
2201     /* FIXME: round this up to page size */
2202 
2203     table = mmap(NULL, len, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
2204     if (table == MAP_FAILED) {
2205         fprintf(stderr, "KVM: Failed to map TCE table for liobn 0x%x\n",
2206                 liobn);
2207         close(fd);
2208         return NULL;
2209     }
2210 
2211     *pfd = fd;
2212     return table;
2213 }
2214 
2215 int kvmppc_remove_spapr_tce(void *table, int fd, uint32_t nb_table)
2216 {
2217     long len;
2218 
2219     if (fd < 0) {
2220         return -1;
2221     }
2222 
2223     len = nb_table * sizeof(uint64_t);
2224     if ((munmap(table, len) < 0) ||
2225         (close(fd) < 0)) {
2226         fprintf(stderr, "KVM: Unexpected error removing TCE table: %s",
2227                 strerror(errno));
2228         /* Leak the table */
2229     }
2230 
2231     return 0;
2232 }
2233 
2234 int kvmppc_reset_htab(int shift_hint)
2235 {
2236     uint32_t shift = shift_hint;
2237 
2238     if (!kvm_enabled()) {
2239         /* Full emulation, tell caller to allocate htab itself */
2240         return 0;
2241     }
2242     if (kvm_vm_check_extension(kvm_state, KVM_CAP_PPC_ALLOC_HTAB)) {
2243         int ret;
2244         ret = kvm_vm_ioctl(kvm_state, KVM_PPC_ALLOCATE_HTAB, &shift);
2245         if (ret == -ENOTTY) {
2246             /*
2247              * At least some versions of PR KVM advertise the
2248              * capability, but don't implement the ioctl().  Oops.
2249              * Return 0 so that we allocate the htab in qemu, as is
2250              * correct for PR.
2251              */
2252             return 0;
2253         } else if (ret < 0) {
2254             return ret;
2255         }
2256         return shift;
2257     }
2258 
2259     /*
2260      * We have a kernel that predates the htab reset calls.  For PR
2261      * KVM, we need to allocate the htab ourselves, for an HV KVM of
2262      * this era, it has allocated a 16MB fixed size hash table
2263      * already.
2264      */
2265     if (kvmppc_is_pr(kvm_state)) {
2266         /* PR - tell caller to allocate htab */
2267         return 0;
2268     } else {
2269         /* HV - assume 16MB kernel allocated htab */
2270         return 24;
2271     }
2272 }
2273 
2274 static inline uint32_t mfpvr(void)
2275 {
2276     uint32_t pvr;
2277 
2278     asm ("mfpvr %0"
2279          : "=r"(pvr));
2280     return pvr;
2281 }
2282 
2283 static void alter_insns(uint64_t *word, uint64_t flags, bool on)
2284 {
2285     if (on) {
2286         *word |= flags;
2287     } else {
2288         *word &= ~flags;
2289     }
2290 }
2291 
2292 static void kvmppc_host_cpu_class_init(ObjectClass *oc, void *data)
2293 {
2294     PowerPCCPUClass *pcc = POWERPC_CPU_CLASS(oc);
2295     uint32_t dcache_size = kvmppc_read_int_cpu_dt("d-cache-size");
2296     uint32_t icache_size = kvmppc_read_int_cpu_dt("i-cache-size");
2297 
2298     /* Now fix up the class with information we can query from the host */
2299     pcc->pvr = mfpvr();
2300 
2301     alter_insns(&pcc->insns_flags, PPC_ALTIVEC,
2302                 qemu_getauxval(AT_HWCAP) & PPC_FEATURE_HAS_ALTIVEC);
2303     alter_insns(&pcc->insns_flags2, PPC2_VSX,
2304                 qemu_getauxval(AT_HWCAP) & PPC_FEATURE_HAS_VSX);
2305     alter_insns(&pcc->insns_flags2, PPC2_DFP,
2306                 qemu_getauxval(AT_HWCAP) & PPC_FEATURE_HAS_DFP);
2307 
2308     if (dcache_size != -1) {
2309         pcc->l1_dcache_size = dcache_size;
2310     }
2311 
2312     if (icache_size != -1) {
2313         pcc->l1_icache_size = icache_size;
2314     }
2315 
2316 #if defined(TARGET_PPC64)
2317     pcc->radix_page_info = kvm_get_radix_page_info();
2318 
2319     if ((pcc->pvr & 0xffffff00) == CPU_POWERPC_POWER9_DD1) {
2320         /*
2321          * POWER9 DD1 has some bugs which make it not really ISA 3.00
2322          * compliant.  More importantly, advertising ISA 3.00
2323          * architected mode may prevent guests from activating
2324          * necessary DD1 workarounds.
2325          */
2326         pcc->pcr_supported &= ~(PCR_COMPAT_3_00 | PCR_COMPAT_2_07
2327                                 | PCR_COMPAT_2_06 | PCR_COMPAT_2_05);
2328     }
2329 #endif /* defined(TARGET_PPC64) */
2330 }
2331 
2332 bool kvmppc_has_cap_epr(void)
2333 {
2334     return cap_epr;
2335 }
2336 
2337 bool kvmppc_has_cap_fixup_hcalls(void)
2338 {
2339     return cap_fixup_hcalls;
2340 }
2341 
2342 bool kvmppc_has_cap_htm(void)
2343 {
2344     return cap_htm;
2345 }
2346 
2347 bool kvmppc_has_cap_mmu_radix(void)
2348 {
2349     return cap_mmu_radix;
2350 }
2351 
2352 bool kvmppc_has_cap_mmu_hash_v3(void)
2353 {
2354     return cap_mmu_hash_v3;
2355 }
2356 
2357 static bool kvmppc_power8_host(void)
2358 {
2359     bool ret = false;
2360 #ifdef TARGET_PPC64
2361     {
2362         uint32_t base_pvr = CPU_POWERPC_POWER_SERVER_MASK & mfpvr();
2363         ret = (base_pvr == CPU_POWERPC_POWER8E_BASE) ||
2364               (base_pvr == CPU_POWERPC_POWER8NVL_BASE) ||
2365               (base_pvr == CPU_POWERPC_POWER8_BASE);
2366     }
2367 #endif /* TARGET_PPC64 */
2368     return ret;
2369 }
2370 
2371 static int parse_cap_ppc_safe_cache(struct kvm_ppc_cpu_char c)
2372 {
2373     bool l1d_thread_priv_req = !kvmppc_power8_host();
2374 
2375     if (~c.behaviour & c.behaviour_mask & H_CPU_BEHAV_L1D_FLUSH_PR) {
2376         return 2;
2377     } else if ((!l1d_thread_priv_req ||
2378                 c.character & c.character_mask & H_CPU_CHAR_L1D_THREAD_PRIV) &&
2379                (c.character & c.character_mask
2380                 & (H_CPU_CHAR_L1D_FLUSH_ORI30 | H_CPU_CHAR_L1D_FLUSH_TRIG2))) {
2381         return 1;
2382     }
2383 
2384     return 0;
2385 }
2386 
2387 static int parse_cap_ppc_safe_bounds_check(struct kvm_ppc_cpu_char c)
2388 {
2389     if (~c.behaviour & c.behaviour_mask & H_CPU_BEHAV_BNDS_CHK_SPEC_BAR) {
2390         return 2;
2391     } else if (c.character & c.character_mask & H_CPU_CHAR_SPEC_BAR_ORI31) {
2392         return 1;
2393     }
2394 
2395     return 0;
2396 }
2397 
2398 static int parse_cap_ppc_safe_indirect_branch(struct kvm_ppc_cpu_char c)
2399 {
2400     if ((~c.behaviour & c.behaviour_mask & H_CPU_BEHAV_FLUSH_COUNT_CACHE) &&
2401         (~c.character & c.character_mask & H_CPU_CHAR_CACHE_COUNT_DIS) &&
2402         (~c.character & c.character_mask & H_CPU_CHAR_BCCTRL_SERIALISED)) {
2403         return SPAPR_CAP_FIXED_NA;
2404     } else if (c.behaviour & c.behaviour_mask & H_CPU_BEHAV_FLUSH_COUNT_CACHE) {
2405         return SPAPR_CAP_WORKAROUND;
2406     } else if (c.character & c.character_mask & H_CPU_CHAR_CACHE_COUNT_DIS) {
2407         return  SPAPR_CAP_FIXED_CCD;
2408     } else if (c.character & c.character_mask & H_CPU_CHAR_BCCTRL_SERIALISED) {
2409         return SPAPR_CAP_FIXED_IBS;
2410     }
2411 
2412     return 0;
2413 }
2414 
2415 static int parse_cap_ppc_count_cache_flush_assist(struct kvm_ppc_cpu_char c)
2416 {
2417     if (c.character & c.character_mask & H_CPU_CHAR_BCCTR_FLUSH_ASSIST) {
2418         return 1;
2419     }
2420     return 0;
2421 }
2422 
2423 bool kvmppc_has_cap_xive(void)
2424 {
2425     return cap_xive;
2426 }
2427 
2428 static void kvmppc_get_cpu_characteristics(KVMState *s)
2429 {
2430     struct kvm_ppc_cpu_char c;
2431     int ret;
2432 
2433     /* Assume broken */
2434     cap_ppc_safe_cache = 0;
2435     cap_ppc_safe_bounds_check = 0;
2436     cap_ppc_safe_indirect_branch = 0;
2437 
2438     ret = kvm_vm_check_extension(s, KVM_CAP_PPC_GET_CPU_CHAR);
2439     if (!ret) {
2440         return;
2441     }
2442     ret = kvm_vm_ioctl(s, KVM_PPC_GET_CPU_CHAR, &c);
2443     if (ret < 0) {
2444         return;
2445     }
2446 
2447     cap_ppc_safe_cache = parse_cap_ppc_safe_cache(c);
2448     cap_ppc_safe_bounds_check = parse_cap_ppc_safe_bounds_check(c);
2449     cap_ppc_safe_indirect_branch = parse_cap_ppc_safe_indirect_branch(c);
2450     cap_ppc_count_cache_flush_assist =
2451         parse_cap_ppc_count_cache_flush_assist(c);
2452 }
2453 
2454 int kvmppc_get_cap_safe_cache(void)
2455 {
2456     return cap_ppc_safe_cache;
2457 }
2458 
2459 int kvmppc_get_cap_safe_bounds_check(void)
2460 {
2461     return cap_ppc_safe_bounds_check;
2462 }
2463 
2464 int kvmppc_get_cap_safe_indirect_branch(void)
2465 {
2466     return cap_ppc_safe_indirect_branch;
2467 }
2468 
2469 int kvmppc_get_cap_count_cache_flush_assist(void)
2470 {
2471     return cap_ppc_count_cache_flush_assist;
2472 }
2473 
2474 bool kvmppc_has_cap_nested_kvm_hv(void)
2475 {
2476     return !!cap_ppc_nested_kvm_hv;
2477 }
2478 
2479 int kvmppc_set_cap_nested_kvm_hv(int enable)
2480 {
2481     return kvm_vm_enable_cap(kvm_state, KVM_CAP_PPC_NESTED_HV, 0, enable);
2482 }
2483 
2484 bool kvmppc_has_cap_spapr_vfio(void)
2485 {
2486     return cap_spapr_vfio;
2487 }
2488 
2489 int kvmppc_get_cap_large_decr(void)
2490 {
2491     return cap_large_decr;
2492 }
2493 
2494 int kvmppc_enable_cap_large_decr(PowerPCCPU *cpu, int enable)
2495 {
2496     CPUState *cs = CPU(cpu);
2497     uint64_t lpcr;
2498 
2499     kvm_get_one_reg(cs, KVM_REG_PPC_LPCR_64, &lpcr);
2500     /* Do we need to modify the LPCR? */
2501     if (!!(lpcr & LPCR_LD) != !!enable) {
2502         if (enable) {
2503             lpcr |= LPCR_LD;
2504         } else {
2505             lpcr &= ~LPCR_LD;
2506         }
2507         kvm_set_one_reg(cs, KVM_REG_PPC_LPCR_64, &lpcr);
2508         kvm_get_one_reg(cs, KVM_REG_PPC_LPCR_64, &lpcr);
2509 
2510         if (!!(lpcr & LPCR_LD) != !!enable) {
2511             return -1;
2512         }
2513     }
2514 
2515     return 0;
2516 }
2517 
2518 PowerPCCPUClass *kvm_ppc_get_host_cpu_class(void)
2519 {
2520     uint32_t host_pvr = mfpvr();
2521     PowerPCCPUClass *pvr_pcc;
2522 
2523     pvr_pcc = ppc_cpu_class_by_pvr(host_pvr);
2524     if (pvr_pcc == NULL) {
2525         pvr_pcc = ppc_cpu_class_by_pvr_mask(host_pvr);
2526     }
2527 
2528     return pvr_pcc;
2529 }
2530 
2531 static void pseries_machine_class_fixup(ObjectClass *oc, void *opaque)
2532 {
2533     MachineClass *mc = MACHINE_CLASS(oc);
2534 
2535     mc->default_cpu_type = TYPE_HOST_POWERPC_CPU;
2536 }
2537 
2538 static int kvm_ppc_register_host_cpu_type(void)
2539 {
2540     TypeInfo type_info = {
2541         .name = TYPE_HOST_POWERPC_CPU,
2542         .class_init = kvmppc_host_cpu_class_init,
2543     };
2544     PowerPCCPUClass *pvr_pcc;
2545     ObjectClass *oc;
2546     DeviceClass *dc;
2547     int i;
2548 
2549     pvr_pcc = kvm_ppc_get_host_cpu_class();
2550     if (pvr_pcc == NULL) {
2551         return -1;
2552     }
2553     type_info.parent = object_class_get_name(OBJECT_CLASS(pvr_pcc));
2554     type_register(&type_info);
2555     /* override TCG default cpu type with 'host' cpu model */
2556     object_class_foreach(pseries_machine_class_fixup, TYPE_SPAPR_MACHINE,
2557                          false, NULL);
2558 
2559     oc = object_class_by_name(type_info.name);
2560     g_assert(oc);
2561 
2562     /*
2563      * Update generic CPU family class alias (e.g. on a POWER8NVL host,
2564      * we want "POWER8" to be a "family" alias that points to the current
2565      * host CPU type, too)
2566      */
2567     dc = DEVICE_CLASS(ppc_cpu_get_family_class(pvr_pcc));
2568     for (i = 0; ppc_cpu_aliases[i].alias != NULL; i++) {
2569         if (strcasecmp(ppc_cpu_aliases[i].alias, dc->desc) == 0) {
2570             char *suffix;
2571 
2572             ppc_cpu_aliases[i].model = g_strdup(object_class_get_name(oc));
2573             suffix = strstr(ppc_cpu_aliases[i].model, POWERPC_CPU_TYPE_SUFFIX);
2574             if (suffix) {
2575                 *suffix = 0;
2576             }
2577             break;
2578         }
2579     }
2580 
2581     return 0;
2582 }
2583 
2584 int kvmppc_define_rtas_kernel_token(uint32_t token, const char *function)
2585 {
2586     struct kvm_rtas_token_args args = {
2587         .token = token,
2588     };
2589 
2590     if (!kvm_check_extension(kvm_state, KVM_CAP_PPC_RTAS)) {
2591         return -ENOENT;
2592     }
2593 
2594     strncpy(args.name, function, sizeof(args.name) - 1);
2595 
2596     return kvm_vm_ioctl(kvm_state, KVM_PPC_RTAS_DEFINE_TOKEN, &args);
2597 }
2598 
2599 int kvmppc_get_htab_fd(bool write, uint64_t index, Error **errp)
2600 {
2601     struct kvm_get_htab_fd s = {
2602         .flags = write ? KVM_GET_HTAB_WRITE : 0,
2603         .start_index = index,
2604     };
2605     int ret;
2606 
2607     if (!cap_htab_fd) {
2608         error_setg(errp, "KVM version doesn't support %s the HPT",
2609                    write ? "writing" : "reading");
2610         return -ENOTSUP;
2611     }
2612 
2613     ret = kvm_vm_ioctl(kvm_state, KVM_PPC_GET_HTAB_FD, &s);
2614     if (ret < 0) {
2615         error_setg(errp, "Unable to open fd for %s HPT %s KVM: %s",
2616                    write ? "writing" : "reading", write ? "to" : "from",
2617                    strerror(errno));
2618         return -errno;
2619     }
2620 
2621     return ret;
2622 }
2623 
2624 int kvmppc_save_htab(QEMUFile *f, int fd, size_t bufsize, int64_t max_ns)
2625 {
2626     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2627     uint8_t buf[bufsize];
2628     ssize_t rc;
2629 
2630     do {
2631         rc = read(fd, buf, bufsize);
2632         if (rc < 0) {
2633             fprintf(stderr, "Error reading data from KVM HTAB fd: %s\n",
2634                     strerror(errno));
2635             return rc;
2636         } else if (rc) {
2637             uint8_t *buffer = buf;
2638             ssize_t n = rc;
2639             while (n) {
2640                 struct kvm_get_htab_header *head =
2641                     (struct kvm_get_htab_header *) buffer;
2642                 size_t chunksize = sizeof(*head) +
2643                      HASH_PTE_SIZE_64 * head->n_valid;
2644 
2645                 qemu_put_be32(f, head->index);
2646                 qemu_put_be16(f, head->n_valid);
2647                 qemu_put_be16(f, head->n_invalid);
2648                 qemu_put_buffer(f, (void *)(head + 1),
2649                                 HASH_PTE_SIZE_64 * head->n_valid);
2650 
2651                 buffer += chunksize;
2652                 n -= chunksize;
2653             }
2654         }
2655     } while ((rc != 0)
2656              && ((max_ns < 0) ||
2657                  ((qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) < max_ns)));
2658 
2659     return (rc == 0) ? 1 : 0;
2660 }
2661 
2662 int kvmppc_load_htab_chunk(QEMUFile *f, int fd, uint32_t index,
2663                            uint16_t n_valid, uint16_t n_invalid)
2664 {
2665     struct kvm_get_htab_header *buf;
2666     size_t chunksize = sizeof(*buf) + n_valid * HASH_PTE_SIZE_64;
2667     ssize_t rc;
2668 
2669     buf = alloca(chunksize);
2670     buf->index = index;
2671     buf->n_valid = n_valid;
2672     buf->n_invalid = n_invalid;
2673 
2674     qemu_get_buffer(f, (void *)(buf + 1), HASH_PTE_SIZE_64 * n_valid);
2675 
2676     rc = write(fd, buf, chunksize);
2677     if (rc < 0) {
2678         fprintf(stderr, "Error writing KVM hash table: %s\n",
2679                 strerror(errno));
2680         return rc;
2681     }
2682     if (rc != chunksize) {
2683         /* We should never get a short write on a single chunk */
2684         fprintf(stderr, "Short write, restoring KVM hash table\n");
2685         return -1;
2686     }
2687     return 0;
2688 }
2689 
2690 bool kvm_arch_stop_on_emulation_error(CPUState *cpu)
2691 {
2692     return true;
2693 }
2694 
2695 void kvm_arch_init_irq_routing(KVMState *s)
2696 {
2697 }
2698 
2699 void kvmppc_read_hptes(ppc_hash_pte64_t *hptes, hwaddr ptex, int n)
2700 {
2701     int fd, rc;
2702     int i;
2703 
2704     fd = kvmppc_get_htab_fd(false, ptex, &error_abort);
2705 
2706     i = 0;
2707     while (i < n) {
2708         struct kvm_get_htab_header *hdr;
2709         int m = n < HPTES_PER_GROUP ? n : HPTES_PER_GROUP;
2710         char buf[sizeof(*hdr) + m * HASH_PTE_SIZE_64];
2711 
2712         rc = read(fd, buf, sizeof(buf));
2713         if (rc < 0) {
2714             hw_error("kvmppc_read_hptes: Unable to read HPTEs");
2715         }
2716 
2717         hdr = (struct kvm_get_htab_header *)buf;
2718         while ((i < n) && ((char *)hdr < (buf + rc))) {
2719             int invalid = hdr->n_invalid, valid = hdr->n_valid;
2720 
2721             if (hdr->index != (ptex + i)) {
2722                 hw_error("kvmppc_read_hptes: Unexpected HPTE index %"PRIu32
2723                          " != (%"HWADDR_PRIu" + %d", hdr->index, ptex, i);
2724             }
2725 
2726             if (n - i < valid) {
2727                 valid = n - i;
2728             }
2729             memcpy(hptes + i, hdr + 1, HASH_PTE_SIZE_64 * valid);
2730             i += valid;
2731 
2732             if ((n - i) < invalid) {
2733                 invalid = n - i;
2734             }
2735             memset(hptes + i, 0, invalid * HASH_PTE_SIZE_64);
2736             i += invalid;
2737 
2738             hdr = (struct kvm_get_htab_header *)
2739                 ((char *)(hdr + 1) + HASH_PTE_SIZE_64 * hdr->n_valid);
2740         }
2741     }
2742 
2743     close(fd);
2744 }
2745 
2746 void kvmppc_write_hpte(hwaddr ptex, uint64_t pte0, uint64_t pte1)
2747 {
2748     int fd, rc;
2749     struct {
2750         struct kvm_get_htab_header hdr;
2751         uint64_t pte0;
2752         uint64_t pte1;
2753     } buf;
2754 
2755     fd = kvmppc_get_htab_fd(true, 0 /* Ignored */, &error_abort);
2756 
2757     buf.hdr.n_valid = 1;
2758     buf.hdr.n_invalid = 0;
2759     buf.hdr.index = ptex;
2760     buf.pte0 = cpu_to_be64(pte0);
2761     buf.pte1 = cpu_to_be64(pte1);
2762 
2763     rc = write(fd, &buf, sizeof(buf));
2764     if (rc != sizeof(buf)) {
2765         hw_error("kvmppc_write_hpte: Unable to update KVM HPT");
2766     }
2767     close(fd);
2768 }
2769 
2770 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
2771                              uint64_t address, uint32_t data, PCIDevice *dev)
2772 {
2773     return 0;
2774 }
2775 
2776 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
2777                                 int vector, PCIDevice *dev)
2778 {
2779     return 0;
2780 }
2781 
2782 int kvm_arch_release_virq_post(int virq)
2783 {
2784     return 0;
2785 }
2786 
2787 int kvm_arch_msi_data_to_gsi(uint32_t data)
2788 {
2789     return data & 0xffff;
2790 }
2791 
2792 int kvmppc_enable_hwrng(void)
2793 {
2794     if (!kvm_enabled() || !kvm_check_extension(kvm_state, KVM_CAP_PPC_HWRNG)) {
2795         return -1;
2796     }
2797 
2798     return kvmppc_enable_hcall(kvm_state, H_RANDOM);
2799 }
2800 
2801 void kvmppc_check_papr_resize_hpt(Error **errp)
2802 {
2803     if (!kvm_enabled()) {
2804         return; /* No KVM, we're good */
2805     }
2806 
2807     if (cap_resize_hpt) {
2808         return; /* Kernel has explicit support, we're good */
2809     }
2810 
2811     /* Otherwise fallback on looking for PR KVM */
2812     if (kvmppc_is_pr(kvm_state)) {
2813         return;
2814     }
2815 
2816     error_setg(errp,
2817                "Hash page table resizing not available with this KVM version");
2818 }
2819 
2820 int kvmppc_resize_hpt_prepare(PowerPCCPU *cpu, target_ulong flags, int shift)
2821 {
2822     CPUState *cs = CPU(cpu);
2823     struct kvm_ppc_resize_hpt rhpt = {
2824         .flags = flags,
2825         .shift = shift,
2826     };
2827 
2828     if (!cap_resize_hpt) {
2829         return -ENOSYS;
2830     }
2831 
2832     return kvm_vm_ioctl(cs->kvm_state, KVM_PPC_RESIZE_HPT_PREPARE, &rhpt);
2833 }
2834 
2835 int kvmppc_resize_hpt_commit(PowerPCCPU *cpu, target_ulong flags, int shift)
2836 {
2837     CPUState *cs = CPU(cpu);
2838     struct kvm_ppc_resize_hpt rhpt = {
2839         .flags = flags,
2840         .shift = shift,
2841     };
2842 
2843     if (!cap_resize_hpt) {
2844         return -ENOSYS;
2845     }
2846 
2847     return kvm_vm_ioctl(cs->kvm_state, KVM_PPC_RESIZE_HPT_COMMIT, &rhpt);
2848 }
2849 
2850 /*
2851  * This is a helper function to detect a post migration scenario
2852  * in which a guest, running as KVM-HV, freezes in cpu_post_load because
2853  * the guest kernel can't handle a PVR value other than the actual host
2854  * PVR in KVM_SET_SREGS, even if pvr_match() returns true.
2855  *
2856  * If we don't have cap_ppc_pvr_compat and we're not running in PR
2857  * (so, we're HV), return true. The workaround itself is done in
2858  * cpu_post_load.
2859  *
2860  * The order here is important: we'll only check for KVM PR as a
2861  * fallback if the guest kernel can't handle the situation itself.
2862  * We need to avoid as much as possible querying the running KVM type
2863  * in QEMU level.
2864  */
2865 bool kvmppc_pvr_workaround_required(PowerPCCPU *cpu)
2866 {
2867     CPUState *cs = CPU(cpu);
2868 
2869     if (!kvm_enabled()) {
2870         return false;
2871     }
2872 
2873     if (cap_ppc_pvr_compat) {
2874         return false;
2875     }
2876 
2877     return !kvmppc_is_pr(cs->kvm_state);
2878 }
2879 
2880 void kvmppc_set_reg_ppc_online(PowerPCCPU *cpu, unsigned int online)
2881 {
2882     CPUState *cs = CPU(cpu);
2883 
2884     if (kvm_enabled()) {
2885         kvm_set_one_reg(cs, KVM_REG_PPC_ONLINE, &online);
2886     }
2887 }
2888 
2889 void kvmppc_set_reg_tb_offset(PowerPCCPU *cpu, int64_t tb_offset)
2890 {
2891     CPUState *cs = CPU(cpu);
2892 
2893     if (kvm_enabled()) {
2894         kvm_set_one_reg(cs, KVM_REG_PPC_TB_OFFSET, &tb_offset);
2895     }
2896 }
2897 
2898 /*
2899  * Don't set error if KVM_PPC_SVM_OFF ioctl is invoked on kernels
2900  * that don't support this ioctl.
2901  */
2902 void kvmppc_svm_off(Error **errp)
2903 {
2904     int rc;
2905 
2906     if (!kvm_enabled()) {
2907         return;
2908     }
2909 
2910     rc = kvm_vm_ioctl(KVM_STATE(current_machine->accelerator), KVM_PPC_SVM_OFF);
2911     if (rc && rc != -ENOTTY) {
2912         error_setg_errno(errp, -rc, "KVM_PPC_SVM_OFF ioctl failed");
2913     }
2914 }
2915