xref: /openbmc/qemu/target/ppc/kvm.c (revision 59a3a1c0)
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_interrupt_level;
62 static int cap_segstate;
63 static int cap_booke_sregs;
64 static int cap_ppc_smt;
65 static int cap_ppc_smt_possible;
66 static int cap_spapr_tce;
67 static int cap_spapr_tce_64;
68 static int cap_spapr_multitce;
69 static int cap_spapr_vfio;
70 static int cap_hior;
71 static int cap_one_reg;
72 static int cap_epr;
73 static int cap_ppc_watchdog;
74 static int cap_papr;
75 static int cap_htab_fd;
76 static int cap_fixup_hcalls;
77 static int cap_htm;             /* Hardware transactional memory support */
78 static int cap_mmu_radix;
79 static int cap_mmu_hash_v3;
80 static int cap_xive;
81 static int cap_resize_hpt;
82 static int cap_ppc_pvr_compat;
83 static int cap_ppc_safe_cache;
84 static int cap_ppc_safe_bounds_check;
85 static int cap_ppc_safe_indirect_branch;
86 static int cap_ppc_count_cache_flush_assist;
87 static int cap_ppc_nested_kvm_hv;
88 static int cap_large_decr;
89 
90 static uint32_t debug_inst_opcode;
91 
92 /*
93  * XXX We have a race condition where we actually have a level triggered
94  *     interrupt, but the infrastructure can't expose that yet, so the guest
95  *     takes but ignores it, goes to sleep and never gets notified that there's
96  *     still an interrupt pending.
97  *
98  *     As a quick workaround, let's just wake up again 20 ms after we injected
99  *     an interrupt. That way we can assure that we're always reinjecting
100  *     interrupts in case the guest swallowed them.
101  */
102 static QEMUTimer *idle_timer;
103 
104 static void kvm_kick_cpu(void *opaque)
105 {
106     PowerPCCPU *cpu = opaque;
107 
108     qemu_cpu_kick(CPU(cpu));
109 }
110 
111 /*
112  * Check whether we are running with KVM-PR (instead of KVM-HV).  This
113  * should only be used for fallback tests - generally we should use
114  * explicit capabilities for the features we want, rather than
115  * assuming what is/isn't available depending on the KVM variant.
116  */
117 static bool kvmppc_is_pr(KVMState *ks)
118 {
119     /* Assume KVM-PR if the GET_PVINFO capability is available */
120     return kvm_vm_check_extension(ks, KVM_CAP_PPC_GET_PVINFO) != 0;
121 }
122 
123 static int kvm_ppc_register_host_cpu_type(MachineState *ms);
124 static void kvmppc_get_cpu_characteristics(KVMState *s);
125 static int kvmppc_get_dec_bits(void);
126 
127 int kvm_arch_init(MachineState *ms, KVMState *s)
128 {
129     cap_interrupt_unset = kvm_check_extension(s, KVM_CAP_PPC_UNSET_IRQ);
130     cap_interrupt_level = kvm_check_extension(s, KVM_CAP_PPC_IRQ_LEVEL);
131     cap_segstate = kvm_check_extension(s, KVM_CAP_PPC_SEGSTATE);
132     cap_booke_sregs = kvm_check_extension(s, KVM_CAP_PPC_BOOKE_SREGS);
133     cap_ppc_smt_possible = kvm_vm_check_extension(s, KVM_CAP_PPC_SMT_POSSIBLE);
134     cap_spapr_tce = kvm_check_extension(s, KVM_CAP_SPAPR_TCE);
135     cap_spapr_tce_64 = kvm_check_extension(s, KVM_CAP_SPAPR_TCE_64);
136     cap_spapr_multitce = kvm_check_extension(s, KVM_CAP_SPAPR_MULTITCE);
137     cap_spapr_vfio = kvm_vm_check_extension(s, KVM_CAP_SPAPR_TCE_VFIO);
138     cap_one_reg = kvm_check_extension(s, KVM_CAP_ONE_REG);
139     cap_hior = kvm_check_extension(s, KVM_CAP_PPC_HIOR);
140     cap_epr = kvm_check_extension(s, KVM_CAP_PPC_EPR);
141     cap_ppc_watchdog = kvm_check_extension(s, KVM_CAP_PPC_BOOKE_WATCHDOG);
142     /*
143      * Note: we don't set cap_papr here, because this capability is
144      * only activated after this by kvmppc_set_papr()
145      */
146     cap_htab_fd = kvm_vm_check_extension(s, KVM_CAP_PPC_HTAB_FD);
147     cap_fixup_hcalls = kvm_check_extension(s, KVM_CAP_PPC_FIXUP_HCALL);
148     cap_ppc_smt = kvm_vm_check_extension(s, KVM_CAP_PPC_SMT);
149     cap_htm = kvm_vm_check_extension(s, KVM_CAP_PPC_HTM);
150     cap_mmu_radix = kvm_vm_check_extension(s, KVM_CAP_PPC_MMU_RADIX);
151     cap_mmu_hash_v3 = kvm_vm_check_extension(s, KVM_CAP_PPC_MMU_HASH_V3);
152     cap_xive = kvm_vm_check_extension(s, KVM_CAP_PPC_IRQ_XIVE);
153     cap_resize_hpt = kvm_vm_check_extension(s, KVM_CAP_SPAPR_RESIZE_HPT);
154     kvmppc_get_cpu_characteristics(s);
155     cap_ppc_nested_kvm_hv = kvm_vm_check_extension(s, KVM_CAP_PPC_NESTED_HV);
156     cap_large_decr = kvmppc_get_dec_bits();
157     /*
158      * Note: setting it to false because there is not such capability
159      * in KVM at this moment.
160      *
161      * TODO: call kvm_vm_check_extension() with the right capability
162      * after the kernel starts implementing it.
163      */
164     cap_ppc_pvr_compat = false;
165 
166     if (!cap_interrupt_level) {
167         fprintf(stderr, "KVM: Couldn't find level irq capability. Expect the "
168                         "VM to stall at times!\n");
169     }
170 
171     kvm_ppc_register_host_cpu_type(ms);
172 
173     return 0;
174 }
175 
176 int kvm_arch_irqchip_create(MachineState *ms, KVMState *s)
177 {
178     return 0;
179 }
180 
181 static int kvm_arch_sync_sregs(PowerPCCPU *cpu)
182 {
183     CPUPPCState *cenv = &cpu->env;
184     CPUState *cs = CPU(cpu);
185     struct kvm_sregs sregs;
186     int ret;
187 
188     if (cenv->excp_model == POWERPC_EXCP_BOOKE) {
189         /*
190          * What we're really trying to say is "if we're on BookE, we
191          * use the native PVR for now". This is the only sane way to
192          * check it though, so we potentially confuse users that they
193          * can run BookE guests on BookS. Let's hope nobody dares
194          * enough :)
195          */
196         return 0;
197     } else {
198         if (!cap_segstate) {
199             fprintf(stderr, "kvm error: missing PVR setting capability\n");
200             return -ENOSYS;
201         }
202     }
203 
204     ret = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs);
205     if (ret) {
206         return ret;
207     }
208 
209     sregs.pvr = cenv->spr[SPR_PVR];
210     return kvm_vcpu_ioctl(cs, KVM_SET_SREGS, &sregs);
211 }
212 
213 /* Set up a shared TLB array with KVM */
214 static int kvm_booke206_tlb_init(PowerPCCPU *cpu)
215 {
216     CPUPPCState *env = &cpu->env;
217     CPUState *cs = CPU(cpu);
218     struct kvm_book3e_206_tlb_params params = {};
219     struct kvm_config_tlb cfg = {};
220     unsigned int entries = 0;
221     int ret, i;
222 
223     if (!kvm_enabled() ||
224         !kvm_check_extension(cs->kvm_state, KVM_CAP_SW_TLB)) {
225         return 0;
226     }
227 
228     assert(ARRAY_SIZE(params.tlb_sizes) == BOOKE206_MAX_TLBN);
229 
230     for (i = 0; i < BOOKE206_MAX_TLBN; i++) {
231         params.tlb_sizes[i] = booke206_tlb_size(env, i);
232         params.tlb_ways[i] = booke206_tlb_ways(env, i);
233         entries += params.tlb_sizes[i];
234     }
235 
236     assert(entries == env->nb_tlb);
237     assert(sizeof(struct kvm_book3e_206_tlb_entry) == sizeof(ppcmas_tlb_t));
238 
239     env->tlb_dirty = true;
240 
241     cfg.array = (uintptr_t)env->tlb.tlbm;
242     cfg.array_len = sizeof(ppcmas_tlb_t) * entries;
243     cfg.params = (uintptr_t)&params;
244     cfg.mmu_type = KVM_MMU_FSL_BOOKE_NOHV;
245 
246     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_SW_TLB, 0, (uintptr_t)&cfg);
247     if (ret < 0) {
248         fprintf(stderr, "%s: couldn't enable KVM_CAP_SW_TLB: %s\n",
249                 __func__, strerror(-ret));
250         return ret;
251     }
252 
253     env->kvm_sw_tlb = true;
254     return 0;
255 }
256 
257 
258 #if defined(TARGET_PPC64)
259 static void kvm_get_smmu_info(struct kvm_ppc_smmu_info *info, Error **errp)
260 {
261     int ret;
262 
263     assert(kvm_state != NULL);
264 
265     if (!kvm_check_extension(kvm_state, KVM_CAP_PPC_GET_SMMU_INFO)) {
266         error_setg(errp, "KVM doesn't expose the MMU features it supports");
267         error_append_hint(errp, "Consider switching to a newer KVM\n");
268         return;
269     }
270 
271     ret = kvm_vm_ioctl(kvm_state, KVM_PPC_GET_SMMU_INFO, info);
272     if (ret == 0) {
273         return;
274     }
275 
276     error_setg_errno(errp, -ret,
277                      "KVM failed to provide the MMU features it supports");
278 }
279 
280 struct ppc_radix_page_info *kvm_get_radix_page_info(void)
281 {
282     KVMState *s = KVM_STATE(current_machine->accelerator);
283     struct ppc_radix_page_info *radix_page_info;
284     struct kvm_ppc_rmmu_info rmmu_info;
285     int i;
286 
287     if (!kvm_check_extension(s, KVM_CAP_PPC_MMU_RADIX)) {
288         return NULL;
289     }
290     if (kvm_vm_ioctl(s, KVM_PPC_GET_RMMU_INFO, &rmmu_info)) {
291         return NULL;
292     }
293     radix_page_info = g_malloc0(sizeof(*radix_page_info));
294     radix_page_info->count = 0;
295     for (i = 0; i < PPC_PAGE_SIZES_MAX_SZ; i++) {
296         if (rmmu_info.ap_encodings[i]) {
297             radix_page_info->entries[i] = rmmu_info.ap_encodings[i];
298             radix_page_info->count++;
299         }
300     }
301     return radix_page_info;
302 }
303 
304 target_ulong kvmppc_configure_v3_mmu(PowerPCCPU *cpu,
305                                      bool radix, bool gtse,
306                                      uint64_t proc_tbl)
307 {
308     CPUState *cs = CPU(cpu);
309     int ret;
310     uint64_t flags = 0;
311     struct kvm_ppc_mmuv3_cfg cfg = {
312         .process_table = proc_tbl,
313     };
314 
315     if (radix) {
316         flags |= KVM_PPC_MMUV3_RADIX;
317     }
318     if (gtse) {
319         flags |= KVM_PPC_MMUV3_GTSE;
320     }
321     cfg.flags = flags;
322     ret = kvm_vm_ioctl(cs->kvm_state, KVM_PPC_CONFIGURE_V3_MMU, &cfg);
323     switch (ret) {
324     case 0:
325         return H_SUCCESS;
326     case -EINVAL:
327         return H_PARAMETER;
328     case -ENODEV:
329         return H_NOT_AVAILABLE;
330     default:
331         return H_HARDWARE;
332     }
333 }
334 
335 bool kvmppc_hpt_needs_host_contiguous_pages(void)
336 {
337     static struct kvm_ppc_smmu_info smmu_info;
338 
339     if (!kvm_enabled()) {
340         return false;
341     }
342 
343     kvm_get_smmu_info(&smmu_info, &error_fatal);
344     return !!(smmu_info.flags & KVM_PPC_PAGE_SIZES_REAL);
345 }
346 
347 void kvm_check_mmu(PowerPCCPU *cpu, Error **errp)
348 {
349     struct kvm_ppc_smmu_info smmu_info;
350     int iq, ik, jq, jk;
351     Error *local_err = NULL;
352 
353     /* For now, we only have anything to check on hash64 MMUs */
354     if (!cpu->hash64_opts || !kvm_enabled()) {
355         return;
356     }
357 
358     kvm_get_smmu_info(&smmu_info, &local_err);
359     if (local_err) {
360         error_propagate(errp, local_err);
361         return;
362     }
363 
364     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)
365         && !(smmu_info.flags & KVM_PPC_1T_SEGMENTS)) {
366         error_setg(errp,
367                    "KVM does not support 1TiB segments which guest expects");
368         return;
369     }
370 
371     if (smmu_info.slb_size < cpu->hash64_opts->slb_size) {
372         error_setg(errp, "KVM only supports %u SLB entries, but guest needs %u",
373                    smmu_info.slb_size, cpu->hash64_opts->slb_size);
374         return;
375     }
376 
377     /*
378      * Verify that every pagesize supported by the cpu model is
379      * supported by KVM with the same encodings
380      */
381     for (iq = 0; iq < ARRAY_SIZE(cpu->hash64_opts->sps); iq++) {
382         PPCHash64SegmentPageSizes *qsps = &cpu->hash64_opts->sps[iq];
383         struct kvm_ppc_one_seg_page_size *ksps;
384 
385         for (ik = 0; ik < ARRAY_SIZE(smmu_info.sps); ik++) {
386             if (qsps->page_shift == smmu_info.sps[ik].page_shift) {
387                 break;
388             }
389         }
390         if (ik >= ARRAY_SIZE(smmu_info.sps)) {
391             error_setg(errp, "KVM doesn't support for base page shift %u",
392                        qsps->page_shift);
393             return;
394         }
395 
396         ksps = &smmu_info.sps[ik];
397         if (ksps->slb_enc != qsps->slb_enc) {
398             error_setg(errp,
399 "KVM uses SLB encoding 0x%x for page shift %u, but guest expects 0x%x",
400                        ksps->slb_enc, ksps->page_shift, qsps->slb_enc);
401             return;
402         }
403 
404         for (jq = 0; jq < ARRAY_SIZE(qsps->enc); jq++) {
405             for (jk = 0; jk < ARRAY_SIZE(ksps->enc); jk++) {
406                 if (qsps->enc[jq].page_shift == ksps->enc[jk].page_shift) {
407                     break;
408                 }
409             }
410 
411             if (jk >= ARRAY_SIZE(ksps->enc)) {
412                 error_setg(errp, "KVM doesn't support page shift %u/%u",
413                            qsps->enc[jq].page_shift, qsps->page_shift);
414                 return;
415             }
416             if (qsps->enc[jq].pte_enc != ksps->enc[jk].pte_enc) {
417                 error_setg(errp,
418 "KVM uses PTE encoding 0x%x for page shift %u/%u, but guest expects 0x%x",
419                            ksps->enc[jk].pte_enc, qsps->enc[jq].page_shift,
420                            qsps->page_shift, qsps->enc[jq].pte_enc);
421                 return;
422             }
423         }
424     }
425 
426     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
427         /*
428          * Mostly what guest pagesizes we can use are related to the
429          * host pages used to map guest RAM, which is handled in the
430          * platform code. Cache-Inhibited largepages (64k) however are
431          * used for I/O, so if they're mapped to the host at all it
432          * will be a normal mapping, not a special hugepage one used
433          * for RAM.
434          */
435         if (getpagesize() < 0x10000) {
436             error_setg(errp,
437                        "KVM can't supply 64kiB CI pages, which guest expects");
438         }
439     }
440 }
441 #endif /* !defined (TARGET_PPC64) */
442 
443 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
444 {
445     return POWERPC_CPU(cpu)->vcpu_id;
446 }
447 
448 /*
449  * e500 supports 2 h/w breakpoint and 2 watchpoint.  book3s supports
450  * only 1 watchpoint, so array size of 4 is sufficient for now.
451  */
452 #define MAX_HW_BKPTS 4
453 
454 static struct HWBreakpoint {
455     target_ulong addr;
456     int type;
457 } hw_debug_points[MAX_HW_BKPTS];
458 
459 static CPUWatchpoint hw_watchpoint;
460 
461 /* Default there is no breakpoint and watchpoint supported */
462 static int max_hw_breakpoint;
463 static int max_hw_watchpoint;
464 static int nb_hw_breakpoint;
465 static int nb_hw_watchpoint;
466 
467 static void kvmppc_hw_debug_points_init(CPUPPCState *cenv)
468 {
469     if (cenv->excp_model == POWERPC_EXCP_BOOKE) {
470         max_hw_breakpoint = 2;
471         max_hw_watchpoint = 2;
472     }
473 
474     if ((max_hw_breakpoint + max_hw_watchpoint) > MAX_HW_BKPTS) {
475         fprintf(stderr, "Error initializing h/w breakpoints\n");
476         return;
477     }
478 }
479 
480 int kvm_arch_init_vcpu(CPUState *cs)
481 {
482     PowerPCCPU *cpu = POWERPC_CPU(cs);
483     CPUPPCState *cenv = &cpu->env;
484     int ret;
485 
486     /* Synchronize sregs with kvm */
487     ret = kvm_arch_sync_sregs(cpu);
488     if (ret) {
489         if (ret == -EINVAL) {
490             error_report("Register sync failed... If you're using kvm-hv.ko,"
491                          " only \"-cpu host\" is possible");
492         }
493         return ret;
494     }
495 
496     idle_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, kvm_kick_cpu, cpu);
497 
498     switch (cenv->mmu_model) {
499     case POWERPC_MMU_BOOKE206:
500         /* This target supports access to KVM's guest TLB */
501         ret = kvm_booke206_tlb_init(cpu);
502         break;
503     case POWERPC_MMU_2_07:
504         if (!cap_htm && !kvmppc_is_pr(cs->kvm_state)) {
505             /*
506              * KVM-HV has transactional memory on POWER8 also without
507              * the KVM_CAP_PPC_HTM extension, so enable it here
508              * instead as long as it's availble to userspace on the
509              * host.
510              */
511             if (qemu_getauxval(AT_HWCAP2) & PPC_FEATURE2_HAS_HTM) {
512                 cap_htm = true;
513             }
514         }
515         break;
516     default:
517         break;
518     }
519 
520     kvm_get_one_reg(cs, KVM_REG_PPC_DEBUG_INST, &debug_inst_opcode);
521     kvmppc_hw_debug_points_init(cenv);
522 
523     return ret;
524 }
525 
526 int kvm_arch_destroy_vcpu(CPUState *cs)
527 {
528     return 0;
529 }
530 
531 static void kvm_sw_tlb_put(PowerPCCPU *cpu)
532 {
533     CPUPPCState *env = &cpu->env;
534     CPUState *cs = CPU(cpu);
535     struct kvm_dirty_tlb dirty_tlb;
536     unsigned char *bitmap;
537     int ret;
538 
539     if (!env->kvm_sw_tlb) {
540         return;
541     }
542 
543     bitmap = g_malloc((env->nb_tlb + 7) / 8);
544     memset(bitmap, 0xFF, (env->nb_tlb + 7) / 8);
545 
546     dirty_tlb.bitmap = (uintptr_t)bitmap;
547     dirty_tlb.num_dirty = env->nb_tlb;
548 
549     ret = kvm_vcpu_ioctl(cs, KVM_DIRTY_TLB, &dirty_tlb);
550     if (ret) {
551         fprintf(stderr, "%s: KVM_DIRTY_TLB: %s\n",
552                 __func__, strerror(-ret));
553     }
554 
555     g_free(bitmap);
556 }
557 
558 static void kvm_get_one_spr(CPUState *cs, uint64_t id, int spr)
559 {
560     PowerPCCPU *cpu = POWERPC_CPU(cs);
561     CPUPPCState *env = &cpu->env;
562     union {
563         uint32_t u32;
564         uint64_t u64;
565     } val;
566     struct kvm_one_reg reg = {
567         .id = id,
568         .addr = (uintptr_t) &val,
569     };
570     int ret;
571 
572     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
573     if (ret != 0) {
574         trace_kvm_failed_spr_get(spr, strerror(errno));
575     } else {
576         switch (id & KVM_REG_SIZE_MASK) {
577         case KVM_REG_SIZE_U32:
578             env->spr[spr] = val.u32;
579             break;
580 
581         case KVM_REG_SIZE_U64:
582             env->spr[spr] = val.u64;
583             break;
584 
585         default:
586             /* Don't handle this size yet */
587             abort();
588         }
589     }
590 }
591 
592 static void kvm_put_one_spr(CPUState *cs, uint64_t id, int spr)
593 {
594     PowerPCCPU *cpu = POWERPC_CPU(cs);
595     CPUPPCState *env = &cpu->env;
596     union {
597         uint32_t u32;
598         uint64_t u64;
599     } val;
600     struct kvm_one_reg reg = {
601         .id = id,
602         .addr = (uintptr_t) &val,
603     };
604     int ret;
605 
606     switch (id & KVM_REG_SIZE_MASK) {
607     case KVM_REG_SIZE_U32:
608         val.u32 = env->spr[spr];
609         break;
610 
611     case KVM_REG_SIZE_U64:
612         val.u64 = env->spr[spr];
613         break;
614 
615     default:
616         /* Don't handle this size yet */
617         abort();
618     }
619 
620     ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
621     if (ret != 0) {
622         trace_kvm_failed_spr_set(spr, strerror(errno));
623     }
624 }
625 
626 static int kvm_put_fp(CPUState *cs)
627 {
628     PowerPCCPU *cpu = POWERPC_CPU(cs);
629     CPUPPCState *env = &cpu->env;
630     struct kvm_one_reg reg;
631     int i;
632     int ret;
633 
634     if (env->insns_flags & PPC_FLOAT) {
635         uint64_t fpscr = env->fpscr;
636         bool vsx = !!(env->insns_flags2 & PPC2_VSX);
637 
638         reg.id = KVM_REG_PPC_FPSCR;
639         reg.addr = (uintptr_t)&fpscr;
640         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
641         if (ret < 0) {
642             trace_kvm_failed_fpscr_set(strerror(errno));
643             return ret;
644         }
645 
646         for (i = 0; i < 32; i++) {
647             uint64_t vsr[2];
648             uint64_t *fpr = cpu_fpr_ptr(&cpu->env, i);
649             uint64_t *vsrl = cpu_vsrl_ptr(&cpu->env, i);
650 
651 #ifdef HOST_WORDS_BIGENDIAN
652             vsr[0] = float64_val(*fpr);
653             vsr[1] = *vsrl;
654 #else
655             vsr[0] = *vsrl;
656             vsr[1] = float64_val(*fpr);
657 #endif
658             reg.addr = (uintptr_t) &vsr;
659             reg.id = vsx ? KVM_REG_PPC_VSR(i) : KVM_REG_PPC_FPR(i);
660 
661             ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
662             if (ret < 0) {
663                 trace_kvm_failed_fp_set(vsx ? "VSR" : "FPR", i,
664                                         strerror(errno));
665                 return ret;
666             }
667         }
668     }
669 
670     if (env->insns_flags & PPC_ALTIVEC) {
671         reg.id = KVM_REG_PPC_VSCR;
672         reg.addr = (uintptr_t)&env->vscr;
673         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
674         if (ret < 0) {
675             trace_kvm_failed_vscr_set(strerror(errno));
676             return ret;
677         }
678 
679         for (i = 0; i < 32; i++) {
680             reg.id = KVM_REG_PPC_VR(i);
681             reg.addr = (uintptr_t)cpu_avr_ptr(env, i);
682             ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
683             if (ret < 0) {
684                 trace_kvm_failed_vr_set(i, strerror(errno));
685                 return ret;
686             }
687         }
688     }
689 
690     return 0;
691 }
692 
693 static int kvm_get_fp(CPUState *cs)
694 {
695     PowerPCCPU *cpu = POWERPC_CPU(cs);
696     CPUPPCState *env = &cpu->env;
697     struct kvm_one_reg reg;
698     int i;
699     int ret;
700 
701     if (env->insns_flags & PPC_FLOAT) {
702         uint64_t fpscr;
703         bool vsx = !!(env->insns_flags2 & PPC2_VSX);
704 
705         reg.id = KVM_REG_PPC_FPSCR;
706         reg.addr = (uintptr_t)&fpscr;
707         ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
708         if (ret < 0) {
709             trace_kvm_failed_fpscr_get(strerror(errno));
710             return ret;
711         } else {
712             env->fpscr = fpscr;
713         }
714 
715         for (i = 0; i < 32; i++) {
716             uint64_t vsr[2];
717             uint64_t *fpr = cpu_fpr_ptr(&cpu->env, i);
718             uint64_t *vsrl = cpu_vsrl_ptr(&cpu->env, i);
719 
720             reg.addr = (uintptr_t) &vsr;
721             reg.id = vsx ? KVM_REG_PPC_VSR(i) : KVM_REG_PPC_FPR(i);
722 
723             ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
724             if (ret < 0) {
725                 trace_kvm_failed_fp_get(vsx ? "VSR" : "FPR", i,
726                                         strerror(errno));
727                 return ret;
728             } else {
729 #ifdef HOST_WORDS_BIGENDIAN
730                 *fpr = vsr[0];
731                 if (vsx) {
732                     *vsrl = vsr[1];
733                 }
734 #else
735                 *fpr = vsr[1];
736                 if (vsx) {
737                     *vsrl = vsr[0];
738                 }
739 #endif
740             }
741         }
742     }
743 
744     if (env->insns_flags & PPC_ALTIVEC) {
745         reg.id = KVM_REG_PPC_VSCR;
746         reg.addr = (uintptr_t)&env->vscr;
747         ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
748         if (ret < 0) {
749             trace_kvm_failed_vscr_get(strerror(errno));
750             return ret;
751         }
752 
753         for (i = 0; i < 32; i++) {
754             reg.id = KVM_REG_PPC_VR(i);
755             reg.addr = (uintptr_t)cpu_avr_ptr(env, i);
756             ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
757             if (ret < 0) {
758                 trace_kvm_failed_vr_get(i, strerror(errno));
759                 return ret;
760             }
761         }
762     }
763 
764     return 0;
765 }
766 
767 #if defined(TARGET_PPC64)
768 static int kvm_get_vpa(CPUState *cs)
769 {
770     PowerPCCPU *cpu = POWERPC_CPU(cs);
771     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
772     struct kvm_one_reg reg;
773     int ret;
774 
775     reg.id = KVM_REG_PPC_VPA_ADDR;
776     reg.addr = (uintptr_t)&spapr_cpu->vpa_addr;
777     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
778     if (ret < 0) {
779         trace_kvm_failed_vpa_addr_get(strerror(errno));
780         return ret;
781     }
782 
783     assert((uintptr_t)&spapr_cpu->slb_shadow_size
784            == ((uintptr_t)&spapr_cpu->slb_shadow_addr + 8));
785     reg.id = KVM_REG_PPC_VPA_SLB;
786     reg.addr = (uintptr_t)&spapr_cpu->slb_shadow_addr;
787     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
788     if (ret < 0) {
789         trace_kvm_failed_slb_get(strerror(errno));
790         return ret;
791     }
792 
793     assert((uintptr_t)&spapr_cpu->dtl_size
794            == ((uintptr_t)&spapr_cpu->dtl_addr + 8));
795     reg.id = KVM_REG_PPC_VPA_DTL;
796     reg.addr = (uintptr_t)&spapr_cpu->dtl_addr;
797     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
798     if (ret < 0) {
799         trace_kvm_failed_dtl_get(strerror(errno));
800         return ret;
801     }
802 
803     return 0;
804 }
805 
806 static int kvm_put_vpa(CPUState *cs)
807 {
808     PowerPCCPU *cpu = POWERPC_CPU(cs);
809     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
810     struct kvm_one_reg reg;
811     int ret;
812 
813     /*
814      * SLB shadow or DTL can't be registered unless a master VPA is
815      * registered.  That means when restoring state, if a VPA *is*
816      * registered, we need to set that up first.  If not, we need to
817      * deregister the others before deregistering the master VPA
818      */
819     assert(spapr_cpu->vpa_addr
820            || !(spapr_cpu->slb_shadow_addr || spapr_cpu->dtl_addr));
821 
822     if (spapr_cpu->vpa_addr) {
823         reg.id = KVM_REG_PPC_VPA_ADDR;
824         reg.addr = (uintptr_t)&spapr_cpu->vpa_addr;
825         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
826         if (ret < 0) {
827             trace_kvm_failed_vpa_addr_set(strerror(errno));
828             return ret;
829         }
830     }
831 
832     assert((uintptr_t)&spapr_cpu->slb_shadow_size
833            == ((uintptr_t)&spapr_cpu->slb_shadow_addr + 8));
834     reg.id = KVM_REG_PPC_VPA_SLB;
835     reg.addr = (uintptr_t)&spapr_cpu->slb_shadow_addr;
836     ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
837     if (ret < 0) {
838         trace_kvm_failed_slb_set(strerror(errno));
839         return ret;
840     }
841 
842     assert((uintptr_t)&spapr_cpu->dtl_size
843            == ((uintptr_t)&spapr_cpu->dtl_addr + 8));
844     reg.id = KVM_REG_PPC_VPA_DTL;
845     reg.addr = (uintptr_t)&spapr_cpu->dtl_addr;
846     ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
847     if (ret < 0) {
848         trace_kvm_failed_dtl_set(strerror(errno));
849         return ret;
850     }
851 
852     if (!spapr_cpu->vpa_addr) {
853         reg.id = KVM_REG_PPC_VPA_ADDR;
854         reg.addr = (uintptr_t)&spapr_cpu->vpa_addr;
855         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
856         if (ret < 0) {
857             trace_kvm_failed_null_vpa_addr_set(strerror(errno));
858             return ret;
859         }
860     }
861 
862     return 0;
863 }
864 #endif /* TARGET_PPC64 */
865 
866 int kvmppc_put_books_sregs(PowerPCCPU *cpu)
867 {
868     CPUPPCState *env = &cpu->env;
869     struct kvm_sregs sregs;
870     int i;
871 
872     sregs.pvr = env->spr[SPR_PVR];
873 
874     if (cpu->vhyp) {
875         PPCVirtualHypervisorClass *vhc =
876             PPC_VIRTUAL_HYPERVISOR_GET_CLASS(cpu->vhyp);
877         sregs.u.s.sdr1 = vhc->encode_hpt_for_kvm_pr(cpu->vhyp);
878     } else {
879         sregs.u.s.sdr1 = env->spr[SPR_SDR1];
880     }
881 
882     /* Sync SLB */
883 #ifdef TARGET_PPC64
884     for (i = 0; i < ARRAY_SIZE(env->slb); i++) {
885         sregs.u.s.ppc64.slb[i].slbe = env->slb[i].esid;
886         if (env->slb[i].esid & SLB_ESID_V) {
887             sregs.u.s.ppc64.slb[i].slbe |= i;
888         }
889         sregs.u.s.ppc64.slb[i].slbv = env->slb[i].vsid;
890     }
891 #endif
892 
893     /* Sync SRs */
894     for (i = 0; i < 16; i++) {
895         sregs.u.s.ppc32.sr[i] = env->sr[i];
896     }
897 
898     /* Sync BATs */
899     for (i = 0; i < 8; i++) {
900         /* Beware. We have to swap upper and lower bits here */
901         sregs.u.s.ppc32.dbat[i] = ((uint64_t)env->DBAT[0][i] << 32)
902             | env->DBAT[1][i];
903         sregs.u.s.ppc32.ibat[i] = ((uint64_t)env->IBAT[0][i] << 32)
904             | env->IBAT[1][i];
905     }
906 
907     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_SREGS, &sregs);
908 }
909 
910 int kvm_arch_put_registers(CPUState *cs, int level)
911 {
912     PowerPCCPU *cpu = POWERPC_CPU(cs);
913     CPUPPCState *env = &cpu->env;
914     struct kvm_regs regs;
915     int ret;
916     int i;
917 
918     ret = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
919     if (ret < 0) {
920         return ret;
921     }
922 
923     regs.ctr = env->ctr;
924     regs.lr  = env->lr;
925     regs.xer = cpu_read_xer(env);
926     regs.msr = env->msr;
927     regs.pc = env->nip;
928 
929     regs.srr0 = env->spr[SPR_SRR0];
930     regs.srr1 = env->spr[SPR_SRR1];
931 
932     regs.sprg0 = env->spr[SPR_SPRG0];
933     regs.sprg1 = env->spr[SPR_SPRG1];
934     regs.sprg2 = env->spr[SPR_SPRG2];
935     regs.sprg3 = env->spr[SPR_SPRG3];
936     regs.sprg4 = env->spr[SPR_SPRG4];
937     regs.sprg5 = env->spr[SPR_SPRG5];
938     regs.sprg6 = env->spr[SPR_SPRG6];
939     regs.sprg7 = env->spr[SPR_SPRG7];
940 
941     regs.pid = env->spr[SPR_BOOKE_PID];
942 
943     for (i = 0; i < 32; i++) {
944         regs.gpr[i] = env->gpr[i];
945     }
946 
947     regs.cr = 0;
948     for (i = 0; i < 8; i++) {
949         regs.cr |= (env->crf[i] & 15) << (4 * (7 - i));
950     }
951 
952     ret = kvm_vcpu_ioctl(cs, KVM_SET_REGS, &regs);
953     if (ret < 0) {
954         return ret;
955     }
956 
957     kvm_put_fp(cs);
958 
959     if (env->tlb_dirty) {
960         kvm_sw_tlb_put(cpu);
961         env->tlb_dirty = false;
962     }
963 
964     if (cap_segstate && (level >= KVM_PUT_RESET_STATE)) {
965         ret = kvmppc_put_books_sregs(cpu);
966         if (ret < 0) {
967             return ret;
968         }
969     }
970 
971     if (cap_hior && (level >= KVM_PUT_RESET_STATE)) {
972         kvm_put_one_spr(cs, KVM_REG_PPC_HIOR, SPR_HIOR);
973     }
974 
975     if (cap_one_reg) {
976         int i;
977 
978         /*
979          * We deliberately ignore errors here, for kernels which have
980          * the ONE_REG calls, but don't support the specific
981          * registers, there's a reasonable chance things will still
982          * work, at least until we try to migrate.
983          */
984         for (i = 0; i < 1024; i++) {
985             uint64_t id = env->spr_cb[i].one_reg_id;
986 
987             if (id != 0) {
988                 kvm_put_one_spr(cs, id, i);
989             }
990         }
991 
992 #ifdef TARGET_PPC64
993         if (msr_ts) {
994             for (i = 0; i < ARRAY_SIZE(env->tm_gpr); i++) {
995                 kvm_set_one_reg(cs, KVM_REG_PPC_TM_GPR(i), &env->tm_gpr[i]);
996             }
997             for (i = 0; i < ARRAY_SIZE(env->tm_vsr); i++) {
998                 kvm_set_one_reg(cs, KVM_REG_PPC_TM_VSR(i), &env->tm_vsr[i]);
999             }
1000             kvm_set_one_reg(cs, KVM_REG_PPC_TM_CR, &env->tm_cr);
1001             kvm_set_one_reg(cs, KVM_REG_PPC_TM_LR, &env->tm_lr);
1002             kvm_set_one_reg(cs, KVM_REG_PPC_TM_CTR, &env->tm_ctr);
1003             kvm_set_one_reg(cs, KVM_REG_PPC_TM_FPSCR, &env->tm_fpscr);
1004             kvm_set_one_reg(cs, KVM_REG_PPC_TM_AMR, &env->tm_amr);
1005             kvm_set_one_reg(cs, KVM_REG_PPC_TM_PPR, &env->tm_ppr);
1006             kvm_set_one_reg(cs, KVM_REG_PPC_TM_VRSAVE, &env->tm_vrsave);
1007             kvm_set_one_reg(cs, KVM_REG_PPC_TM_VSCR, &env->tm_vscr);
1008             kvm_set_one_reg(cs, KVM_REG_PPC_TM_DSCR, &env->tm_dscr);
1009             kvm_set_one_reg(cs, KVM_REG_PPC_TM_TAR, &env->tm_tar);
1010         }
1011 
1012         if (cap_papr) {
1013             if (kvm_put_vpa(cs) < 0) {
1014                 trace_kvm_failed_put_vpa();
1015             }
1016         }
1017 
1018         kvm_set_one_reg(cs, KVM_REG_PPC_TB_OFFSET, &env->tb_env->tb_offset);
1019 #endif /* TARGET_PPC64 */
1020     }
1021 
1022     return ret;
1023 }
1024 
1025 static void kvm_sync_excp(CPUPPCState *env, int vector, int ivor)
1026 {
1027      env->excp_vectors[vector] = env->spr[ivor] + env->spr[SPR_BOOKE_IVPR];
1028 }
1029 
1030 static int kvmppc_get_booke_sregs(PowerPCCPU *cpu)
1031 {
1032     CPUPPCState *env = &cpu->env;
1033     struct kvm_sregs sregs;
1034     int ret;
1035 
1036     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_SREGS, &sregs);
1037     if (ret < 0) {
1038         return ret;
1039     }
1040 
1041     if (sregs.u.e.features & KVM_SREGS_E_BASE) {
1042         env->spr[SPR_BOOKE_CSRR0] = sregs.u.e.csrr0;
1043         env->spr[SPR_BOOKE_CSRR1] = sregs.u.e.csrr1;
1044         env->spr[SPR_BOOKE_ESR] = sregs.u.e.esr;
1045         env->spr[SPR_BOOKE_DEAR] = sregs.u.e.dear;
1046         env->spr[SPR_BOOKE_MCSR] = sregs.u.e.mcsr;
1047         env->spr[SPR_BOOKE_TSR] = sregs.u.e.tsr;
1048         env->spr[SPR_BOOKE_TCR] = sregs.u.e.tcr;
1049         env->spr[SPR_DECR] = sregs.u.e.dec;
1050         env->spr[SPR_TBL] = sregs.u.e.tb & 0xffffffff;
1051         env->spr[SPR_TBU] = sregs.u.e.tb >> 32;
1052         env->spr[SPR_VRSAVE] = sregs.u.e.vrsave;
1053     }
1054 
1055     if (sregs.u.e.features & KVM_SREGS_E_ARCH206) {
1056         env->spr[SPR_BOOKE_PIR] = sregs.u.e.pir;
1057         env->spr[SPR_BOOKE_MCSRR0] = sregs.u.e.mcsrr0;
1058         env->spr[SPR_BOOKE_MCSRR1] = sregs.u.e.mcsrr1;
1059         env->spr[SPR_BOOKE_DECAR] = sregs.u.e.decar;
1060         env->spr[SPR_BOOKE_IVPR] = sregs.u.e.ivpr;
1061     }
1062 
1063     if (sregs.u.e.features & KVM_SREGS_E_64) {
1064         env->spr[SPR_BOOKE_EPCR] = sregs.u.e.epcr;
1065     }
1066 
1067     if (sregs.u.e.features & KVM_SREGS_E_SPRG8) {
1068         env->spr[SPR_BOOKE_SPRG8] = sregs.u.e.sprg8;
1069     }
1070 
1071     if (sregs.u.e.features & KVM_SREGS_E_IVOR) {
1072         env->spr[SPR_BOOKE_IVOR0] = sregs.u.e.ivor_low[0];
1073         kvm_sync_excp(env, POWERPC_EXCP_CRITICAL,  SPR_BOOKE_IVOR0);
1074         env->spr[SPR_BOOKE_IVOR1] = sregs.u.e.ivor_low[1];
1075         kvm_sync_excp(env, POWERPC_EXCP_MCHECK,  SPR_BOOKE_IVOR1);
1076         env->spr[SPR_BOOKE_IVOR2] = sregs.u.e.ivor_low[2];
1077         kvm_sync_excp(env, POWERPC_EXCP_DSI,  SPR_BOOKE_IVOR2);
1078         env->spr[SPR_BOOKE_IVOR3] = sregs.u.e.ivor_low[3];
1079         kvm_sync_excp(env, POWERPC_EXCP_ISI,  SPR_BOOKE_IVOR3);
1080         env->spr[SPR_BOOKE_IVOR4] = sregs.u.e.ivor_low[4];
1081         kvm_sync_excp(env, POWERPC_EXCP_EXTERNAL,  SPR_BOOKE_IVOR4);
1082         env->spr[SPR_BOOKE_IVOR5] = sregs.u.e.ivor_low[5];
1083         kvm_sync_excp(env, POWERPC_EXCP_ALIGN,  SPR_BOOKE_IVOR5);
1084         env->spr[SPR_BOOKE_IVOR6] = sregs.u.e.ivor_low[6];
1085         kvm_sync_excp(env, POWERPC_EXCP_PROGRAM,  SPR_BOOKE_IVOR6);
1086         env->spr[SPR_BOOKE_IVOR7] = sregs.u.e.ivor_low[7];
1087         kvm_sync_excp(env, POWERPC_EXCP_FPU,  SPR_BOOKE_IVOR7);
1088         env->spr[SPR_BOOKE_IVOR8] = sregs.u.e.ivor_low[8];
1089         kvm_sync_excp(env, POWERPC_EXCP_SYSCALL,  SPR_BOOKE_IVOR8);
1090         env->spr[SPR_BOOKE_IVOR9] = sregs.u.e.ivor_low[9];
1091         kvm_sync_excp(env, POWERPC_EXCP_APU,  SPR_BOOKE_IVOR9);
1092         env->spr[SPR_BOOKE_IVOR10] = sregs.u.e.ivor_low[10];
1093         kvm_sync_excp(env, POWERPC_EXCP_DECR,  SPR_BOOKE_IVOR10);
1094         env->spr[SPR_BOOKE_IVOR11] = sregs.u.e.ivor_low[11];
1095         kvm_sync_excp(env, POWERPC_EXCP_FIT,  SPR_BOOKE_IVOR11);
1096         env->spr[SPR_BOOKE_IVOR12] = sregs.u.e.ivor_low[12];
1097         kvm_sync_excp(env, POWERPC_EXCP_WDT,  SPR_BOOKE_IVOR12);
1098         env->spr[SPR_BOOKE_IVOR13] = sregs.u.e.ivor_low[13];
1099         kvm_sync_excp(env, POWERPC_EXCP_DTLB,  SPR_BOOKE_IVOR13);
1100         env->spr[SPR_BOOKE_IVOR14] = sregs.u.e.ivor_low[14];
1101         kvm_sync_excp(env, POWERPC_EXCP_ITLB,  SPR_BOOKE_IVOR14);
1102         env->spr[SPR_BOOKE_IVOR15] = sregs.u.e.ivor_low[15];
1103         kvm_sync_excp(env, POWERPC_EXCP_DEBUG,  SPR_BOOKE_IVOR15);
1104 
1105         if (sregs.u.e.features & KVM_SREGS_E_SPE) {
1106             env->spr[SPR_BOOKE_IVOR32] = sregs.u.e.ivor_high[0];
1107             kvm_sync_excp(env, POWERPC_EXCP_SPEU,  SPR_BOOKE_IVOR32);
1108             env->spr[SPR_BOOKE_IVOR33] = sregs.u.e.ivor_high[1];
1109             kvm_sync_excp(env, POWERPC_EXCP_EFPDI,  SPR_BOOKE_IVOR33);
1110             env->spr[SPR_BOOKE_IVOR34] = sregs.u.e.ivor_high[2];
1111             kvm_sync_excp(env, POWERPC_EXCP_EFPRI,  SPR_BOOKE_IVOR34);
1112         }
1113 
1114         if (sregs.u.e.features & KVM_SREGS_E_PM) {
1115             env->spr[SPR_BOOKE_IVOR35] = sregs.u.e.ivor_high[3];
1116             kvm_sync_excp(env, POWERPC_EXCP_EPERFM,  SPR_BOOKE_IVOR35);
1117         }
1118 
1119         if (sregs.u.e.features & KVM_SREGS_E_PC) {
1120             env->spr[SPR_BOOKE_IVOR36] = sregs.u.e.ivor_high[4];
1121             kvm_sync_excp(env, POWERPC_EXCP_DOORI,  SPR_BOOKE_IVOR36);
1122             env->spr[SPR_BOOKE_IVOR37] = sregs.u.e.ivor_high[5];
1123             kvm_sync_excp(env, POWERPC_EXCP_DOORCI, SPR_BOOKE_IVOR37);
1124         }
1125     }
1126 
1127     if (sregs.u.e.features & KVM_SREGS_E_ARCH206_MMU) {
1128         env->spr[SPR_BOOKE_MAS0] = sregs.u.e.mas0;
1129         env->spr[SPR_BOOKE_MAS1] = sregs.u.e.mas1;
1130         env->spr[SPR_BOOKE_MAS2] = sregs.u.e.mas2;
1131         env->spr[SPR_BOOKE_MAS3] = sregs.u.e.mas7_3 & 0xffffffff;
1132         env->spr[SPR_BOOKE_MAS4] = sregs.u.e.mas4;
1133         env->spr[SPR_BOOKE_MAS6] = sregs.u.e.mas6;
1134         env->spr[SPR_BOOKE_MAS7] = sregs.u.e.mas7_3 >> 32;
1135         env->spr[SPR_MMUCFG] = sregs.u.e.mmucfg;
1136         env->spr[SPR_BOOKE_TLB0CFG] = sregs.u.e.tlbcfg[0];
1137         env->spr[SPR_BOOKE_TLB1CFG] = sregs.u.e.tlbcfg[1];
1138     }
1139 
1140     if (sregs.u.e.features & KVM_SREGS_EXP) {
1141         env->spr[SPR_BOOKE_EPR] = sregs.u.e.epr;
1142     }
1143 
1144     if (sregs.u.e.features & KVM_SREGS_E_PD) {
1145         env->spr[SPR_BOOKE_EPLC] = sregs.u.e.eplc;
1146         env->spr[SPR_BOOKE_EPSC] = sregs.u.e.epsc;
1147     }
1148 
1149     if (sregs.u.e.impl_id == KVM_SREGS_E_IMPL_FSL) {
1150         env->spr[SPR_E500_SVR] = sregs.u.e.impl.fsl.svr;
1151         env->spr[SPR_Exxx_MCAR] = sregs.u.e.impl.fsl.mcar;
1152         env->spr[SPR_HID0] = sregs.u.e.impl.fsl.hid0;
1153 
1154         if (sregs.u.e.impl.fsl.features & KVM_SREGS_E_FSL_PIDn) {
1155             env->spr[SPR_BOOKE_PID1] = sregs.u.e.impl.fsl.pid1;
1156             env->spr[SPR_BOOKE_PID2] = sregs.u.e.impl.fsl.pid2;
1157         }
1158     }
1159 
1160     return 0;
1161 }
1162 
1163 static int kvmppc_get_books_sregs(PowerPCCPU *cpu)
1164 {
1165     CPUPPCState *env = &cpu->env;
1166     struct kvm_sregs sregs;
1167     int ret;
1168     int i;
1169 
1170     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_SREGS, &sregs);
1171     if (ret < 0) {
1172         return ret;
1173     }
1174 
1175     if (!cpu->vhyp) {
1176         ppc_store_sdr1(env, sregs.u.s.sdr1);
1177     }
1178 
1179     /* Sync SLB */
1180 #ifdef TARGET_PPC64
1181     /*
1182      * The packed SLB array we get from KVM_GET_SREGS only contains
1183      * information about valid entries. So we flush our internal copy
1184      * to get rid of stale ones, then put all valid SLB entries back
1185      * in.
1186      */
1187     memset(env->slb, 0, sizeof(env->slb));
1188     for (i = 0; i < ARRAY_SIZE(env->slb); i++) {
1189         target_ulong rb = sregs.u.s.ppc64.slb[i].slbe;
1190         target_ulong rs = sregs.u.s.ppc64.slb[i].slbv;
1191         /*
1192          * Only restore valid entries
1193          */
1194         if (rb & SLB_ESID_V) {
1195             ppc_store_slb(cpu, rb & 0xfff, rb & ~0xfffULL, rs);
1196         }
1197     }
1198 #endif
1199 
1200     /* Sync SRs */
1201     for (i = 0; i < 16; i++) {
1202         env->sr[i] = sregs.u.s.ppc32.sr[i];
1203     }
1204 
1205     /* Sync BATs */
1206     for (i = 0; i < 8; i++) {
1207         env->DBAT[0][i] = sregs.u.s.ppc32.dbat[i] & 0xffffffff;
1208         env->DBAT[1][i] = sregs.u.s.ppc32.dbat[i] >> 32;
1209         env->IBAT[0][i] = sregs.u.s.ppc32.ibat[i] & 0xffffffff;
1210         env->IBAT[1][i] = sregs.u.s.ppc32.ibat[i] >> 32;
1211     }
1212 
1213     return 0;
1214 }
1215 
1216 int kvm_arch_get_registers(CPUState *cs)
1217 {
1218     PowerPCCPU *cpu = POWERPC_CPU(cs);
1219     CPUPPCState *env = &cpu->env;
1220     struct kvm_regs regs;
1221     uint32_t cr;
1222     int i, ret;
1223 
1224     ret = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
1225     if (ret < 0) {
1226         return ret;
1227     }
1228 
1229     cr = regs.cr;
1230     for (i = 7; i >= 0; i--) {
1231         env->crf[i] = cr & 15;
1232         cr >>= 4;
1233     }
1234 
1235     env->ctr = regs.ctr;
1236     env->lr = regs.lr;
1237     cpu_write_xer(env, regs.xer);
1238     env->msr = regs.msr;
1239     env->nip = regs.pc;
1240 
1241     env->spr[SPR_SRR0] = regs.srr0;
1242     env->spr[SPR_SRR1] = regs.srr1;
1243 
1244     env->spr[SPR_SPRG0] = regs.sprg0;
1245     env->spr[SPR_SPRG1] = regs.sprg1;
1246     env->spr[SPR_SPRG2] = regs.sprg2;
1247     env->spr[SPR_SPRG3] = regs.sprg3;
1248     env->spr[SPR_SPRG4] = regs.sprg4;
1249     env->spr[SPR_SPRG5] = regs.sprg5;
1250     env->spr[SPR_SPRG6] = regs.sprg6;
1251     env->spr[SPR_SPRG7] = regs.sprg7;
1252 
1253     env->spr[SPR_BOOKE_PID] = regs.pid;
1254 
1255     for (i = 0; i < 32; i++) {
1256         env->gpr[i] = regs.gpr[i];
1257     }
1258 
1259     kvm_get_fp(cs);
1260 
1261     if (cap_booke_sregs) {
1262         ret = kvmppc_get_booke_sregs(cpu);
1263         if (ret < 0) {
1264             return ret;
1265         }
1266     }
1267 
1268     if (cap_segstate) {
1269         ret = kvmppc_get_books_sregs(cpu);
1270         if (ret < 0) {
1271             return ret;
1272         }
1273     }
1274 
1275     if (cap_hior) {
1276         kvm_get_one_spr(cs, KVM_REG_PPC_HIOR, SPR_HIOR);
1277     }
1278 
1279     if (cap_one_reg) {
1280         int i;
1281 
1282         /*
1283          * We deliberately ignore errors here, for kernels which have
1284          * the ONE_REG calls, but don't support the specific
1285          * registers, there's a reasonable chance things will still
1286          * work, at least until we try to migrate.
1287          */
1288         for (i = 0; i < 1024; i++) {
1289             uint64_t id = env->spr_cb[i].one_reg_id;
1290 
1291             if (id != 0) {
1292                 kvm_get_one_spr(cs, id, i);
1293             }
1294         }
1295 
1296 #ifdef TARGET_PPC64
1297         if (msr_ts) {
1298             for (i = 0; i < ARRAY_SIZE(env->tm_gpr); i++) {
1299                 kvm_get_one_reg(cs, KVM_REG_PPC_TM_GPR(i), &env->tm_gpr[i]);
1300             }
1301             for (i = 0; i < ARRAY_SIZE(env->tm_vsr); i++) {
1302                 kvm_get_one_reg(cs, KVM_REG_PPC_TM_VSR(i), &env->tm_vsr[i]);
1303             }
1304             kvm_get_one_reg(cs, KVM_REG_PPC_TM_CR, &env->tm_cr);
1305             kvm_get_one_reg(cs, KVM_REG_PPC_TM_LR, &env->tm_lr);
1306             kvm_get_one_reg(cs, KVM_REG_PPC_TM_CTR, &env->tm_ctr);
1307             kvm_get_one_reg(cs, KVM_REG_PPC_TM_FPSCR, &env->tm_fpscr);
1308             kvm_get_one_reg(cs, KVM_REG_PPC_TM_AMR, &env->tm_amr);
1309             kvm_get_one_reg(cs, KVM_REG_PPC_TM_PPR, &env->tm_ppr);
1310             kvm_get_one_reg(cs, KVM_REG_PPC_TM_VRSAVE, &env->tm_vrsave);
1311             kvm_get_one_reg(cs, KVM_REG_PPC_TM_VSCR, &env->tm_vscr);
1312             kvm_get_one_reg(cs, KVM_REG_PPC_TM_DSCR, &env->tm_dscr);
1313             kvm_get_one_reg(cs, KVM_REG_PPC_TM_TAR, &env->tm_tar);
1314         }
1315 
1316         if (cap_papr) {
1317             if (kvm_get_vpa(cs) < 0) {
1318                 trace_kvm_failed_get_vpa();
1319             }
1320         }
1321 
1322         kvm_get_one_reg(cs, KVM_REG_PPC_TB_OFFSET, &env->tb_env->tb_offset);
1323 #endif
1324     }
1325 
1326     return 0;
1327 }
1328 
1329 int kvmppc_set_interrupt(PowerPCCPU *cpu, int irq, int level)
1330 {
1331     unsigned virq = level ? KVM_INTERRUPT_SET_LEVEL : KVM_INTERRUPT_UNSET;
1332 
1333     if (irq != PPC_INTERRUPT_EXT) {
1334         return 0;
1335     }
1336 
1337     if (!kvm_enabled() || !cap_interrupt_unset || !cap_interrupt_level) {
1338         return 0;
1339     }
1340 
1341     kvm_vcpu_ioctl(CPU(cpu), KVM_INTERRUPT, &virq);
1342 
1343     return 0;
1344 }
1345 
1346 #if defined(TARGET_PPC64)
1347 #define PPC_INPUT_INT PPC970_INPUT_INT
1348 #else
1349 #define PPC_INPUT_INT PPC6xx_INPUT_INT
1350 #endif
1351 
1352 void kvm_arch_pre_run(CPUState *cs, struct kvm_run *run)
1353 {
1354     PowerPCCPU *cpu = POWERPC_CPU(cs);
1355     CPUPPCState *env = &cpu->env;
1356     int r;
1357     unsigned irq;
1358 
1359     qemu_mutex_lock_iothread();
1360 
1361     /*
1362      * PowerPC QEMU tracks the various core input pins (interrupt,
1363      * critical interrupt, reset, etc) in PPC-specific
1364      * env->irq_input_state.
1365      */
1366     if (!cap_interrupt_level &&
1367         run->ready_for_interrupt_injection &&
1368         (cs->interrupt_request & CPU_INTERRUPT_HARD) &&
1369         (env->irq_input_state & (1 << PPC_INPUT_INT)))
1370     {
1371         /*
1372          * For now KVM disregards the 'irq' argument. However, in the
1373          * future KVM could cache it in-kernel to avoid a heavyweight
1374          * exit when reading the UIC.
1375          */
1376         irq = KVM_INTERRUPT_SET;
1377 
1378         trace_kvm_injected_interrupt(irq);
1379         r = kvm_vcpu_ioctl(cs, KVM_INTERRUPT, &irq);
1380         if (r < 0) {
1381             printf("cpu %d fail inject %x\n", cs->cpu_index, irq);
1382         }
1383 
1384         /* Always wake up soon in case the interrupt was level based */
1385         timer_mod(idle_timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) +
1386                        (NANOSECONDS_PER_SECOND / 50));
1387     }
1388 
1389     /*
1390      * We don't know if there are more interrupts pending after
1391      * this. However, the guest will return to userspace in the course
1392      * of handling this one anyways, so we will get a chance to
1393      * deliver the rest.
1394      */
1395 
1396     qemu_mutex_unlock_iothread();
1397 }
1398 
1399 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
1400 {
1401     return MEMTXATTRS_UNSPECIFIED;
1402 }
1403 
1404 int kvm_arch_process_async_events(CPUState *cs)
1405 {
1406     return cs->halted;
1407 }
1408 
1409 static int kvmppc_handle_halt(PowerPCCPU *cpu)
1410 {
1411     CPUState *cs = CPU(cpu);
1412     CPUPPCState *env = &cpu->env;
1413 
1414     if (!(cs->interrupt_request & CPU_INTERRUPT_HARD) && (msr_ee)) {
1415         cs->halted = 1;
1416         cs->exception_index = EXCP_HLT;
1417     }
1418 
1419     return 0;
1420 }
1421 
1422 /* map dcr access to existing qemu dcr emulation */
1423 static int kvmppc_handle_dcr_read(CPUPPCState *env,
1424                                   uint32_t dcrn, uint32_t *data)
1425 {
1426     if (ppc_dcr_read(env->dcr_env, dcrn, data) < 0) {
1427         fprintf(stderr, "Read to unhandled DCR (0x%x)\n", dcrn);
1428     }
1429 
1430     return 0;
1431 }
1432 
1433 static int kvmppc_handle_dcr_write(CPUPPCState *env,
1434                                    uint32_t dcrn, uint32_t data)
1435 {
1436     if (ppc_dcr_write(env->dcr_env, dcrn, data) < 0) {
1437         fprintf(stderr, "Write to unhandled DCR (0x%x)\n", dcrn);
1438     }
1439 
1440     return 0;
1441 }
1442 
1443 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
1444 {
1445     /* Mixed endian case is not handled */
1446     uint32_t sc = debug_inst_opcode;
1447 
1448     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
1449                             sizeof(sc), 0) ||
1450         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&sc, sizeof(sc), 1)) {
1451         return -EINVAL;
1452     }
1453 
1454     return 0;
1455 }
1456 
1457 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
1458 {
1459     uint32_t sc;
1460 
1461     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&sc, sizeof(sc), 0) ||
1462         sc != debug_inst_opcode ||
1463         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
1464                             sizeof(sc), 1)) {
1465         return -EINVAL;
1466     }
1467 
1468     return 0;
1469 }
1470 
1471 static int find_hw_breakpoint(target_ulong addr, int type)
1472 {
1473     int n;
1474 
1475     assert((nb_hw_breakpoint + nb_hw_watchpoint)
1476            <= ARRAY_SIZE(hw_debug_points));
1477 
1478     for (n = 0; n < nb_hw_breakpoint + nb_hw_watchpoint; n++) {
1479         if (hw_debug_points[n].addr == addr &&
1480              hw_debug_points[n].type == type) {
1481             return n;
1482         }
1483     }
1484 
1485     return -1;
1486 }
1487 
1488 static int find_hw_watchpoint(target_ulong addr, int *flag)
1489 {
1490     int n;
1491 
1492     n = find_hw_breakpoint(addr, GDB_WATCHPOINT_ACCESS);
1493     if (n >= 0) {
1494         *flag = BP_MEM_ACCESS;
1495         return n;
1496     }
1497 
1498     n = find_hw_breakpoint(addr, GDB_WATCHPOINT_WRITE);
1499     if (n >= 0) {
1500         *flag = BP_MEM_WRITE;
1501         return n;
1502     }
1503 
1504     n = find_hw_breakpoint(addr, GDB_WATCHPOINT_READ);
1505     if (n >= 0) {
1506         *flag = BP_MEM_READ;
1507         return n;
1508     }
1509 
1510     return -1;
1511 }
1512 
1513 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
1514                                   target_ulong len, int type)
1515 {
1516     if ((nb_hw_breakpoint + nb_hw_watchpoint) >= ARRAY_SIZE(hw_debug_points)) {
1517         return -ENOBUFS;
1518     }
1519 
1520     hw_debug_points[nb_hw_breakpoint + nb_hw_watchpoint].addr = addr;
1521     hw_debug_points[nb_hw_breakpoint + nb_hw_watchpoint].type = type;
1522 
1523     switch (type) {
1524     case GDB_BREAKPOINT_HW:
1525         if (nb_hw_breakpoint >= max_hw_breakpoint) {
1526             return -ENOBUFS;
1527         }
1528 
1529         if (find_hw_breakpoint(addr, type) >= 0) {
1530             return -EEXIST;
1531         }
1532 
1533         nb_hw_breakpoint++;
1534         break;
1535 
1536     case GDB_WATCHPOINT_WRITE:
1537     case GDB_WATCHPOINT_READ:
1538     case GDB_WATCHPOINT_ACCESS:
1539         if (nb_hw_watchpoint >= max_hw_watchpoint) {
1540             return -ENOBUFS;
1541         }
1542 
1543         if (find_hw_breakpoint(addr, type) >= 0) {
1544             return -EEXIST;
1545         }
1546 
1547         nb_hw_watchpoint++;
1548         break;
1549 
1550     default:
1551         return -ENOSYS;
1552     }
1553 
1554     return 0;
1555 }
1556 
1557 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
1558                                   target_ulong len, int type)
1559 {
1560     int n;
1561 
1562     n = find_hw_breakpoint(addr, type);
1563     if (n < 0) {
1564         return -ENOENT;
1565     }
1566 
1567     switch (type) {
1568     case GDB_BREAKPOINT_HW:
1569         nb_hw_breakpoint--;
1570         break;
1571 
1572     case GDB_WATCHPOINT_WRITE:
1573     case GDB_WATCHPOINT_READ:
1574     case GDB_WATCHPOINT_ACCESS:
1575         nb_hw_watchpoint--;
1576         break;
1577 
1578     default:
1579         return -ENOSYS;
1580     }
1581     hw_debug_points[n] = hw_debug_points[nb_hw_breakpoint + nb_hw_watchpoint];
1582 
1583     return 0;
1584 }
1585 
1586 void kvm_arch_remove_all_hw_breakpoints(void)
1587 {
1588     nb_hw_breakpoint = nb_hw_watchpoint = 0;
1589 }
1590 
1591 void kvm_arch_update_guest_debug(CPUState *cs, struct kvm_guest_debug *dbg)
1592 {
1593     int n;
1594 
1595     /* Software Breakpoint updates */
1596     if (kvm_sw_breakpoints_active(cs)) {
1597         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP;
1598     }
1599 
1600     assert((nb_hw_breakpoint + nb_hw_watchpoint)
1601            <= ARRAY_SIZE(hw_debug_points));
1602     assert((nb_hw_breakpoint + nb_hw_watchpoint) <= ARRAY_SIZE(dbg->arch.bp));
1603 
1604     if (nb_hw_breakpoint + nb_hw_watchpoint > 0) {
1605         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
1606         memset(dbg->arch.bp, 0, sizeof(dbg->arch.bp));
1607         for (n = 0; n < nb_hw_breakpoint + nb_hw_watchpoint; n++) {
1608             switch (hw_debug_points[n].type) {
1609             case GDB_BREAKPOINT_HW:
1610                 dbg->arch.bp[n].type = KVMPPC_DEBUG_BREAKPOINT;
1611                 break;
1612             case GDB_WATCHPOINT_WRITE:
1613                 dbg->arch.bp[n].type = KVMPPC_DEBUG_WATCH_WRITE;
1614                 break;
1615             case GDB_WATCHPOINT_READ:
1616                 dbg->arch.bp[n].type = KVMPPC_DEBUG_WATCH_READ;
1617                 break;
1618             case GDB_WATCHPOINT_ACCESS:
1619                 dbg->arch.bp[n].type = KVMPPC_DEBUG_WATCH_WRITE |
1620                                         KVMPPC_DEBUG_WATCH_READ;
1621                 break;
1622             default:
1623                 cpu_abort(cs, "Unsupported breakpoint type\n");
1624             }
1625             dbg->arch.bp[n].addr = hw_debug_points[n].addr;
1626         }
1627     }
1628 }
1629 
1630 static int kvm_handle_hw_breakpoint(CPUState *cs,
1631                                     struct kvm_debug_exit_arch *arch_info)
1632 {
1633     int handle = 0;
1634     int n;
1635     int flag = 0;
1636 
1637     if (nb_hw_breakpoint + nb_hw_watchpoint > 0) {
1638         if (arch_info->status & KVMPPC_DEBUG_BREAKPOINT) {
1639             n = find_hw_breakpoint(arch_info->address, GDB_BREAKPOINT_HW);
1640             if (n >= 0) {
1641                 handle = 1;
1642             }
1643         } else if (arch_info->status & (KVMPPC_DEBUG_WATCH_READ |
1644                                         KVMPPC_DEBUG_WATCH_WRITE)) {
1645             n = find_hw_watchpoint(arch_info->address,  &flag);
1646             if (n >= 0) {
1647                 handle = 1;
1648                 cs->watchpoint_hit = &hw_watchpoint;
1649                 hw_watchpoint.vaddr = hw_debug_points[n].addr;
1650                 hw_watchpoint.flags = flag;
1651             }
1652         }
1653     }
1654     return handle;
1655 }
1656 
1657 static int kvm_handle_singlestep(void)
1658 {
1659     return 1;
1660 }
1661 
1662 static int kvm_handle_sw_breakpoint(void)
1663 {
1664     return 1;
1665 }
1666 
1667 static int kvm_handle_debug(PowerPCCPU *cpu, struct kvm_run *run)
1668 {
1669     CPUState *cs = CPU(cpu);
1670     CPUPPCState *env = &cpu->env;
1671     struct kvm_debug_exit_arch *arch_info = &run->debug.arch;
1672 
1673     if (cs->singlestep_enabled) {
1674         return kvm_handle_singlestep();
1675     }
1676 
1677     if (arch_info->status) {
1678         return kvm_handle_hw_breakpoint(cs, arch_info);
1679     }
1680 
1681     if (kvm_find_sw_breakpoint(cs, arch_info->address)) {
1682         return kvm_handle_sw_breakpoint();
1683     }
1684 
1685     /*
1686      * QEMU is not able to handle debug exception, so inject
1687      * program exception to guest;
1688      * Yes program exception NOT debug exception !!
1689      * When QEMU is using debug resources then debug exception must
1690      * be always set. To achieve this we set MSR_DE and also set
1691      * MSRP_DEP so guest cannot change MSR_DE.
1692      * When emulating debug resource for guest we want guest
1693      * to control MSR_DE (enable/disable debug interrupt on need).
1694      * Supporting both configurations are NOT possible.
1695      * So the result is that we cannot share debug resources
1696      * between QEMU and Guest on BOOKE architecture.
1697      * In the current design QEMU gets the priority over guest,
1698      * this means that if QEMU is using debug resources then guest
1699      * cannot use them;
1700      * For software breakpoint QEMU uses a privileged instruction;
1701      * So there cannot be any reason that we are here for guest
1702      * set debug exception, only possibility is guest executed a
1703      * privileged / illegal instruction and that's why we are
1704      * injecting a program interrupt.
1705      */
1706     cpu_synchronize_state(cs);
1707     /*
1708      * env->nip is PC, so increment this by 4 to use
1709      * ppc_cpu_do_interrupt(), which set srr0 = env->nip - 4.
1710      */
1711     env->nip += 4;
1712     cs->exception_index = POWERPC_EXCP_PROGRAM;
1713     env->error_code = POWERPC_EXCP_INVAL;
1714     ppc_cpu_do_interrupt(cs);
1715 
1716     return 0;
1717 }
1718 
1719 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1720 {
1721     PowerPCCPU *cpu = POWERPC_CPU(cs);
1722     CPUPPCState *env = &cpu->env;
1723     int ret;
1724 
1725     qemu_mutex_lock_iothread();
1726 
1727     switch (run->exit_reason) {
1728     case KVM_EXIT_DCR:
1729         if (run->dcr.is_write) {
1730             trace_kvm_handle_dcr_write();
1731             ret = kvmppc_handle_dcr_write(env, run->dcr.dcrn, run->dcr.data);
1732         } else {
1733             trace_kvm_handle_dcr_read();
1734             ret = kvmppc_handle_dcr_read(env, run->dcr.dcrn, &run->dcr.data);
1735         }
1736         break;
1737     case KVM_EXIT_HLT:
1738         trace_kvm_handle_halt();
1739         ret = kvmppc_handle_halt(cpu);
1740         break;
1741 #if defined(TARGET_PPC64)
1742     case KVM_EXIT_PAPR_HCALL:
1743         trace_kvm_handle_papr_hcall();
1744         run->papr_hcall.ret = spapr_hypercall(cpu,
1745                                               run->papr_hcall.nr,
1746                                               run->papr_hcall.args);
1747         ret = 0;
1748         break;
1749 #endif
1750     case KVM_EXIT_EPR:
1751         trace_kvm_handle_epr();
1752         run->epr.epr = ldl_phys(cs->as, env->mpic_iack);
1753         ret = 0;
1754         break;
1755     case KVM_EXIT_WATCHDOG:
1756         trace_kvm_handle_watchdog_expiry();
1757         watchdog_perform_action();
1758         ret = 0;
1759         break;
1760 
1761     case KVM_EXIT_DEBUG:
1762         trace_kvm_handle_debug_exception();
1763         if (kvm_handle_debug(cpu, run)) {
1764             ret = EXCP_DEBUG;
1765             break;
1766         }
1767         /* re-enter, this exception was guest-internal */
1768         ret = 0;
1769         break;
1770 
1771     default:
1772         fprintf(stderr, "KVM: unknown exit reason %d\n", run->exit_reason);
1773         ret = -1;
1774         break;
1775     }
1776 
1777     qemu_mutex_unlock_iothread();
1778     return ret;
1779 }
1780 
1781 int kvmppc_or_tsr_bits(PowerPCCPU *cpu, uint32_t tsr_bits)
1782 {
1783     CPUState *cs = CPU(cpu);
1784     uint32_t bits = tsr_bits;
1785     struct kvm_one_reg reg = {
1786         .id = KVM_REG_PPC_OR_TSR,
1787         .addr = (uintptr_t) &bits,
1788     };
1789 
1790     return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
1791 }
1792 
1793 int kvmppc_clear_tsr_bits(PowerPCCPU *cpu, uint32_t tsr_bits)
1794 {
1795 
1796     CPUState *cs = CPU(cpu);
1797     uint32_t bits = tsr_bits;
1798     struct kvm_one_reg reg = {
1799         .id = KVM_REG_PPC_CLEAR_TSR,
1800         .addr = (uintptr_t) &bits,
1801     };
1802 
1803     return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
1804 }
1805 
1806 int kvmppc_set_tcr(PowerPCCPU *cpu)
1807 {
1808     CPUState *cs = CPU(cpu);
1809     CPUPPCState *env = &cpu->env;
1810     uint32_t tcr = env->spr[SPR_BOOKE_TCR];
1811 
1812     struct kvm_one_reg reg = {
1813         .id = KVM_REG_PPC_TCR,
1814         .addr = (uintptr_t) &tcr,
1815     };
1816 
1817     return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
1818 }
1819 
1820 int kvmppc_booke_watchdog_enable(PowerPCCPU *cpu)
1821 {
1822     CPUState *cs = CPU(cpu);
1823     int ret;
1824 
1825     if (!kvm_enabled()) {
1826         return -1;
1827     }
1828 
1829     if (!cap_ppc_watchdog) {
1830         printf("warning: KVM does not support watchdog");
1831         return -1;
1832     }
1833 
1834     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_PPC_BOOKE_WATCHDOG, 0);
1835     if (ret < 0) {
1836         fprintf(stderr, "%s: couldn't enable KVM_CAP_PPC_BOOKE_WATCHDOG: %s\n",
1837                 __func__, strerror(-ret));
1838         return ret;
1839     }
1840 
1841     return ret;
1842 }
1843 
1844 static int read_cpuinfo(const char *field, char *value, int len)
1845 {
1846     FILE *f;
1847     int ret = -1;
1848     int field_len = strlen(field);
1849     char line[512];
1850 
1851     f = fopen("/proc/cpuinfo", "r");
1852     if (!f) {
1853         return -1;
1854     }
1855 
1856     do {
1857         if (!fgets(line, sizeof(line), f)) {
1858             break;
1859         }
1860         if (!strncmp(line, field, field_len)) {
1861             pstrcpy(value, len, line);
1862             ret = 0;
1863             break;
1864         }
1865     } while (*line);
1866 
1867     fclose(f);
1868 
1869     return ret;
1870 }
1871 
1872 uint32_t kvmppc_get_tbfreq(void)
1873 {
1874     char line[512];
1875     char *ns;
1876     uint32_t retval = NANOSECONDS_PER_SECOND;
1877 
1878     if (read_cpuinfo("timebase", line, sizeof(line))) {
1879         return retval;
1880     }
1881 
1882     ns = strchr(line, ':');
1883     if (!ns) {
1884         return retval;
1885     }
1886 
1887     ns++;
1888 
1889     return atoi(ns);
1890 }
1891 
1892 bool kvmppc_get_host_serial(char **value)
1893 {
1894     return g_file_get_contents("/proc/device-tree/system-id", value, NULL,
1895                                NULL);
1896 }
1897 
1898 bool kvmppc_get_host_model(char **value)
1899 {
1900     return g_file_get_contents("/proc/device-tree/model", value, NULL, NULL);
1901 }
1902 
1903 /* Try to find a device tree node for a CPU with clock-frequency property */
1904 static int kvmppc_find_cpu_dt(char *buf, int buf_len)
1905 {
1906     struct dirent *dirp;
1907     DIR *dp;
1908 
1909     dp = opendir(PROC_DEVTREE_CPU);
1910     if (!dp) {
1911         printf("Can't open directory " PROC_DEVTREE_CPU "\n");
1912         return -1;
1913     }
1914 
1915     buf[0] = '\0';
1916     while ((dirp = readdir(dp)) != NULL) {
1917         FILE *f;
1918         snprintf(buf, buf_len, "%s%s/clock-frequency", PROC_DEVTREE_CPU,
1919                  dirp->d_name);
1920         f = fopen(buf, "r");
1921         if (f) {
1922             snprintf(buf, buf_len, "%s%s", PROC_DEVTREE_CPU, dirp->d_name);
1923             fclose(f);
1924             break;
1925         }
1926         buf[0] = '\0';
1927     }
1928     closedir(dp);
1929     if (buf[0] == '\0') {
1930         printf("Unknown host!\n");
1931         return -1;
1932     }
1933 
1934     return 0;
1935 }
1936 
1937 static uint64_t kvmppc_read_int_dt(const char *filename)
1938 {
1939     union {
1940         uint32_t v32;
1941         uint64_t v64;
1942     } u;
1943     FILE *f;
1944     int len;
1945 
1946     f = fopen(filename, "rb");
1947     if (!f) {
1948         return -1;
1949     }
1950 
1951     len = fread(&u, 1, sizeof(u), f);
1952     fclose(f);
1953     switch (len) {
1954     case 4:
1955         /* property is a 32-bit quantity */
1956         return be32_to_cpu(u.v32);
1957     case 8:
1958         return be64_to_cpu(u.v64);
1959     }
1960 
1961     return 0;
1962 }
1963 
1964 /*
1965  * Read a CPU node property from the host device tree that's a single
1966  * integer (32-bit or 64-bit).  Returns 0 if anything goes wrong
1967  * (can't find or open the property, or doesn't understand the format)
1968  */
1969 static uint64_t kvmppc_read_int_cpu_dt(const char *propname)
1970 {
1971     char buf[PATH_MAX], *tmp;
1972     uint64_t val;
1973 
1974     if (kvmppc_find_cpu_dt(buf, sizeof(buf))) {
1975         return -1;
1976     }
1977 
1978     tmp = g_strdup_printf("%s/%s", buf, propname);
1979     val = kvmppc_read_int_dt(tmp);
1980     g_free(tmp);
1981 
1982     return val;
1983 }
1984 
1985 uint64_t kvmppc_get_clockfreq(void)
1986 {
1987     return kvmppc_read_int_cpu_dt("clock-frequency");
1988 }
1989 
1990 static int kvmppc_get_dec_bits(void)
1991 {
1992     int nr_bits = kvmppc_read_int_cpu_dt("ibm,dec-bits");
1993 
1994     if (nr_bits > 0) {
1995         return nr_bits;
1996     }
1997     return 0;
1998 }
1999 
2000 static int kvmppc_get_pvinfo(CPUPPCState *env, struct kvm_ppc_pvinfo *pvinfo)
2001 {
2002     CPUState *cs = env_cpu(env);
2003 
2004     if (kvm_vm_check_extension(cs->kvm_state, KVM_CAP_PPC_GET_PVINFO) &&
2005         !kvm_vm_ioctl(cs->kvm_state, KVM_PPC_GET_PVINFO, pvinfo)) {
2006         return 0;
2007     }
2008 
2009     return 1;
2010 }
2011 
2012 int kvmppc_get_hasidle(CPUPPCState *env)
2013 {
2014     struct kvm_ppc_pvinfo pvinfo;
2015 
2016     if (!kvmppc_get_pvinfo(env, &pvinfo) &&
2017         (pvinfo.flags & KVM_PPC_PVINFO_FLAGS_EV_IDLE)) {
2018         return 1;
2019     }
2020 
2021     return 0;
2022 }
2023 
2024 int kvmppc_get_hypercall(CPUPPCState *env, uint8_t *buf, int buf_len)
2025 {
2026     uint32_t *hc = (uint32_t *)buf;
2027     struct kvm_ppc_pvinfo pvinfo;
2028 
2029     if (!kvmppc_get_pvinfo(env, &pvinfo)) {
2030         memcpy(buf, pvinfo.hcall, buf_len);
2031         return 0;
2032     }
2033 
2034     /*
2035      * Fallback to always fail hypercalls regardless of endianness:
2036      *
2037      *     tdi 0,r0,72 (becomes b .+8 in wrong endian, nop in good endian)
2038      *     li r3, -1
2039      *     b .+8       (becomes nop in wrong endian)
2040      *     bswap32(li r3, -1)
2041      */
2042 
2043     hc[0] = cpu_to_be32(0x08000048);
2044     hc[1] = cpu_to_be32(0x3860ffff);
2045     hc[2] = cpu_to_be32(0x48000008);
2046     hc[3] = cpu_to_be32(bswap32(0x3860ffff));
2047 
2048     return 1;
2049 }
2050 
2051 static inline int kvmppc_enable_hcall(KVMState *s, target_ulong hcall)
2052 {
2053     return kvm_vm_enable_cap(s, KVM_CAP_PPC_ENABLE_HCALL, 0, hcall, 1);
2054 }
2055 
2056 void kvmppc_enable_logical_ci_hcalls(void)
2057 {
2058     /*
2059      * FIXME: it would be nice if we could detect the cases where
2060      * we're using a device which requires the in kernel
2061      * implementation of these hcalls, but the kernel lacks them and
2062      * produce a warning.
2063      */
2064     kvmppc_enable_hcall(kvm_state, H_LOGICAL_CI_LOAD);
2065     kvmppc_enable_hcall(kvm_state, H_LOGICAL_CI_STORE);
2066 }
2067 
2068 void kvmppc_enable_set_mode_hcall(void)
2069 {
2070     kvmppc_enable_hcall(kvm_state, H_SET_MODE);
2071 }
2072 
2073 void kvmppc_enable_clear_ref_mod_hcalls(void)
2074 {
2075     kvmppc_enable_hcall(kvm_state, H_CLEAR_REF);
2076     kvmppc_enable_hcall(kvm_state, H_CLEAR_MOD);
2077 }
2078 
2079 void kvmppc_enable_h_page_init(void)
2080 {
2081     kvmppc_enable_hcall(kvm_state, H_PAGE_INIT);
2082 }
2083 
2084 void kvmppc_set_papr(PowerPCCPU *cpu)
2085 {
2086     CPUState *cs = CPU(cpu);
2087     int ret;
2088 
2089     if (!kvm_enabled()) {
2090         return;
2091     }
2092 
2093     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_PPC_PAPR, 0);
2094     if (ret) {
2095         error_report("This vCPU type or KVM version does not support PAPR");
2096         exit(1);
2097     }
2098 
2099     /*
2100      * Update the capability flag so we sync the right information
2101      * with kvm
2102      */
2103     cap_papr = 1;
2104 }
2105 
2106 int kvmppc_set_compat(PowerPCCPU *cpu, uint32_t compat_pvr)
2107 {
2108     return kvm_set_one_reg(CPU(cpu), KVM_REG_PPC_ARCH_COMPAT, &compat_pvr);
2109 }
2110 
2111 void kvmppc_set_mpic_proxy(PowerPCCPU *cpu, int mpic_proxy)
2112 {
2113     CPUState *cs = CPU(cpu);
2114     int ret;
2115 
2116     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_PPC_EPR, 0, mpic_proxy);
2117     if (ret && mpic_proxy) {
2118         error_report("This KVM version does not support EPR");
2119         exit(1);
2120     }
2121 }
2122 
2123 int kvmppc_smt_threads(void)
2124 {
2125     return cap_ppc_smt ? cap_ppc_smt : 1;
2126 }
2127 
2128 int kvmppc_set_smt_threads(int smt)
2129 {
2130     int ret;
2131 
2132     ret = kvm_vm_enable_cap(kvm_state, KVM_CAP_PPC_SMT, 0, smt, 0);
2133     if (!ret) {
2134         cap_ppc_smt = smt;
2135     }
2136     return ret;
2137 }
2138 
2139 void kvmppc_hint_smt_possible(Error **errp)
2140 {
2141     int i;
2142     GString *g;
2143     char *s;
2144 
2145     assert(kvm_enabled());
2146     if (cap_ppc_smt_possible) {
2147         g = g_string_new("Available VSMT modes:");
2148         for (i = 63; i >= 0; i--) {
2149             if ((1UL << i) & cap_ppc_smt_possible) {
2150                 g_string_append_printf(g, " %lu", (1UL << i));
2151             }
2152         }
2153         s = g_string_free(g, false);
2154         error_append_hint(errp, "%s.\n", s);
2155         g_free(s);
2156     } else {
2157         error_append_hint(errp,
2158                           "This KVM seems to be too old to support VSMT.\n");
2159     }
2160 }
2161 
2162 
2163 #ifdef TARGET_PPC64
2164 uint64_t kvmppc_rma_size(uint64_t current_size, unsigned int hash_shift)
2165 {
2166     struct kvm_ppc_smmu_info info;
2167     long rampagesize, best_page_shift;
2168     int i;
2169 
2170     /*
2171      * Find the largest hardware supported page size that's less than
2172      * or equal to the (logical) backing page size of guest RAM
2173      */
2174     kvm_get_smmu_info(&info, &error_fatal);
2175     rampagesize = qemu_minrampagesize();
2176     best_page_shift = 0;
2177 
2178     for (i = 0; i < KVM_PPC_PAGE_SIZES_MAX_SZ; i++) {
2179         struct kvm_ppc_one_seg_page_size *sps = &info.sps[i];
2180 
2181         if (!sps->page_shift) {
2182             continue;
2183         }
2184 
2185         if ((sps->page_shift > best_page_shift)
2186             && ((1UL << sps->page_shift) <= rampagesize)) {
2187             best_page_shift = sps->page_shift;
2188         }
2189     }
2190 
2191     return MIN(current_size,
2192                1ULL << (best_page_shift + hash_shift - 7));
2193 }
2194 #endif
2195 
2196 bool kvmppc_spapr_use_multitce(void)
2197 {
2198     return cap_spapr_multitce;
2199 }
2200 
2201 int kvmppc_spapr_enable_inkernel_multitce(void)
2202 {
2203     int ret;
2204 
2205     ret = kvm_vm_enable_cap(kvm_state, KVM_CAP_PPC_ENABLE_HCALL, 0,
2206                             H_PUT_TCE_INDIRECT, 1);
2207     if (!ret) {
2208         ret = kvm_vm_enable_cap(kvm_state, KVM_CAP_PPC_ENABLE_HCALL, 0,
2209                                 H_STUFF_TCE, 1);
2210     }
2211 
2212     return ret;
2213 }
2214 
2215 void *kvmppc_create_spapr_tce(uint32_t liobn, uint32_t page_shift,
2216                               uint64_t bus_offset, uint32_t nb_table,
2217                               int *pfd, bool need_vfio)
2218 {
2219     long len;
2220     int fd;
2221     void *table;
2222 
2223     /*
2224      * Must set fd to -1 so we don't try to munmap when called for
2225      * destroying the table, which the upper layers -will- do
2226      */
2227     *pfd = -1;
2228     if (!cap_spapr_tce || (need_vfio && !cap_spapr_vfio)) {
2229         return NULL;
2230     }
2231 
2232     if (cap_spapr_tce_64) {
2233         struct kvm_create_spapr_tce_64 args = {
2234             .liobn = liobn,
2235             .page_shift = page_shift,
2236             .offset = bus_offset >> page_shift,
2237             .size = nb_table,
2238             .flags = 0
2239         };
2240         fd = kvm_vm_ioctl(kvm_state, KVM_CREATE_SPAPR_TCE_64, &args);
2241         if (fd < 0) {
2242             fprintf(stderr,
2243                     "KVM: Failed to create TCE64 table for liobn 0x%x\n",
2244                     liobn);
2245             return NULL;
2246         }
2247     } else if (cap_spapr_tce) {
2248         uint64_t window_size = (uint64_t) nb_table << page_shift;
2249         struct kvm_create_spapr_tce args = {
2250             .liobn = liobn,
2251             .window_size = window_size,
2252         };
2253         if ((window_size != args.window_size) || bus_offset) {
2254             return NULL;
2255         }
2256         fd = kvm_vm_ioctl(kvm_state, KVM_CREATE_SPAPR_TCE, &args);
2257         if (fd < 0) {
2258             fprintf(stderr, "KVM: Failed to create TCE table for liobn 0x%x\n",
2259                     liobn);
2260             return NULL;
2261         }
2262     } else {
2263         return NULL;
2264     }
2265 
2266     len = nb_table * sizeof(uint64_t);
2267     /* FIXME: round this up to page size */
2268 
2269     table = mmap(NULL, len, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
2270     if (table == MAP_FAILED) {
2271         fprintf(stderr, "KVM: Failed to map TCE table for liobn 0x%x\n",
2272                 liobn);
2273         close(fd);
2274         return NULL;
2275     }
2276 
2277     *pfd = fd;
2278     return table;
2279 }
2280 
2281 int kvmppc_remove_spapr_tce(void *table, int fd, uint32_t nb_table)
2282 {
2283     long len;
2284 
2285     if (fd < 0) {
2286         return -1;
2287     }
2288 
2289     len = nb_table * sizeof(uint64_t);
2290     if ((munmap(table, len) < 0) ||
2291         (close(fd) < 0)) {
2292         fprintf(stderr, "KVM: Unexpected error removing TCE table: %s",
2293                 strerror(errno));
2294         /* Leak the table */
2295     }
2296 
2297     return 0;
2298 }
2299 
2300 int kvmppc_reset_htab(int shift_hint)
2301 {
2302     uint32_t shift = shift_hint;
2303 
2304     if (!kvm_enabled()) {
2305         /* Full emulation, tell caller to allocate htab itself */
2306         return 0;
2307     }
2308     if (kvm_vm_check_extension(kvm_state, KVM_CAP_PPC_ALLOC_HTAB)) {
2309         int ret;
2310         ret = kvm_vm_ioctl(kvm_state, KVM_PPC_ALLOCATE_HTAB, &shift);
2311         if (ret == -ENOTTY) {
2312             /*
2313              * At least some versions of PR KVM advertise the
2314              * capability, but don't implement the ioctl().  Oops.
2315              * Return 0 so that we allocate the htab in qemu, as is
2316              * correct for PR.
2317              */
2318             return 0;
2319         } else if (ret < 0) {
2320             return ret;
2321         }
2322         return shift;
2323     }
2324 
2325     /*
2326      * We have a kernel that predates the htab reset calls.  For PR
2327      * KVM, we need to allocate the htab ourselves, for an HV KVM of
2328      * this era, it has allocated a 16MB fixed size hash table
2329      * already.
2330      */
2331     if (kvmppc_is_pr(kvm_state)) {
2332         /* PR - tell caller to allocate htab */
2333         return 0;
2334     } else {
2335         /* HV - assume 16MB kernel allocated htab */
2336         return 24;
2337     }
2338 }
2339 
2340 static inline uint32_t mfpvr(void)
2341 {
2342     uint32_t pvr;
2343 
2344     asm ("mfpvr %0"
2345          : "=r"(pvr));
2346     return pvr;
2347 }
2348 
2349 static void alter_insns(uint64_t *word, uint64_t flags, bool on)
2350 {
2351     if (on) {
2352         *word |= flags;
2353     } else {
2354         *word &= ~flags;
2355     }
2356 }
2357 
2358 static void kvmppc_host_cpu_class_init(ObjectClass *oc, void *data)
2359 {
2360     PowerPCCPUClass *pcc = POWERPC_CPU_CLASS(oc);
2361     uint32_t dcache_size = kvmppc_read_int_cpu_dt("d-cache-size");
2362     uint32_t icache_size = kvmppc_read_int_cpu_dt("i-cache-size");
2363 
2364     /* Now fix up the class with information we can query from the host */
2365     pcc->pvr = mfpvr();
2366 
2367     alter_insns(&pcc->insns_flags, PPC_ALTIVEC,
2368                 qemu_getauxval(AT_HWCAP) & PPC_FEATURE_HAS_ALTIVEC);
2369     alter_insns(&pcc->insns_flags2, PPC2_VSX,
2370                 qemu_getauxval(AT_HWCAP) & PPC_FEATURE_HAS_VSX);
2371     alter_insns(&pcc->insns_flags2, PPC2_DFP,
2372                 qemu_getauxval(AT_HWCAP) & PPC_FEATURE_HAS_DFP);
2373 
2374     if (dcache_size != -1) {
2375         pcc->l1_dcache_size = dcache_size;
2376     }
2377 
2378     if (icache_size != -1) {
2379         pcc->l1_icache_size = icache_size;
2380     }
2381 
2382 #if defined(TARGET_PPC64)
2383     pcc->radix_page_info = kvm_get_radix_page_info();
2384 
2385     if ((pcc->pvr & 0xffffff00) == CPU_POWERPC_POWER9_DD1) {
2386         /*
2387          * POWER9 DD1 has some bugs which make it not really ISA 3.00
2388          * compliant.  More importantly, advertising ISA 3.00
2389          * architected mode may prevent guests from activating
2390          * necessary DD1 workarounds.
2391          */
2392         pcc->pcr_supported &= ~(PCR_COMPAT_3_00 | PCR_COMPAT_2_07
2393                                 | PCR_COMPAT_2_06 | PCR_COMPAT_2_05);
2394     }
2395 #endif /* defined(TARGET_PPC64) */
2396 }
2397 
2398 bool kvmppc_has_cap_epr(void)
2399 {
2400     return cap_epr;
2401 }
2402 
2403 bool kvmppc_has_cap_fixup_hcalls(void)
2404 {
2405     return cap_fixup_hcalls;
2406 }
2407 
2408 bool kvmppc_has_cap_htm(void)
2409 {
2410     return cap_htm;
2411 }
2412 
2413 bool kvmppc_has_cap_mmu_radix(void)
2414 {
2415     return cap_mmu_radix;
2416 }
2417 
2418 bool kvmppc_has_cap_mmu_hash_v3(void)
2419 {
2420     return cap_mmu_hash_v3;
2421 }
2422 
2423 static bool kvmppc_power8_host(void)
2424 {
2425     bool ret = false;
2426 #ifdef TARGET_PPC64
2427     {
2428         uint32_t base_pvr = CPU_POWERPC_POWER_SERVER_MASK & mfpvr();
2429         ret = (base_pvr == CPU_POWERPC_POWER8E_BASE) ||
2430               (base_pvr == CPU_POWERPC_POWER8NVL_BASE) ||
2431               (base_pvr == CPU_POWERPC_POWER8_BASE);
2432     }
2433 #endif /* TARGET_PPC64 */
2434     return ret;
2435 }
2436 
2437 static int parse_cap_ppc_safe_cache(struct kvm_ppc_cpu_char c)
2438 {
2439     bool l1d_thread_priv_req = !kvmppc_power8_host();
2440 
2441     if (~c.behaviour & c.behaviour_mask & H_CPU_BEHAV_L1D_FLUSH_PR) {
2442         return 2;
2443     } else if ((!l1d_thread_priv_req ||
2444                 c.character & c.character_mask & H_CPU_CHAR_L1D_THREAD_PRIV) &&
2445                (c.character & c.character_mask
2446                 & (H_CPU_CHAR_L1D_FLUSH_ORI30 | H_CPU_CHAR_L1D_FLUSH_TRIG2))) {
2447         return 1;
2448     }
2449 
2450     return 0;
2451 }
2452 
2453 static int parse_cap_ppc_safe_bounds_check(struct kvm_ppc_cpu_char c)
2454 {
2455     if (~c.behaviour & c.behaviour_mask & H_CPU_BEHAV_BNDS_CHK_SPEC_BAR) {
2456         return 2;
2457     } else if (c.character & c.character_mask & H_CPU_CHAR_SPEC_BAR_ORI31) {
2458         return 1;
2459     }
2460 
2461     return 0;
2462 }
2463 
2464 static int parse_cap_ppc_safe_indirect_branch(struct kvm_ppc_cpu_char c)
2465 {
2466     if ((~c.behaviour & c.behaviour_mask & H_CPU_BEHAV_FLUSH_COUNT_CACHE) &&
2467         (~c.character & c.character_mask & H_CPU_CHAR_CACHE_COUNT_DIS) &&
2468         (~c.character & c.character_mask & H_CPU_CHAR_BCCTRL_SERIALISED)) {
2469         return SPAPR_CAP_FIXED_NA;
2470     } else if (c.behaviour & c.behaviour_mask & H_CPU_BEHAV_FLUSH_COUNT_CACHE) {
2471         return SPAPR_CAP_WORKAROUND;
2472     } else if (c.character & c.character_mask & H_CPU_CHAR_CACHE_COUNT_DIS) {
2473         return  SPAPR_CAP_FIXED_CCD;
2474     } else if (c.character & c.character_mask & H_CPU_CHAR_BCCTRL_SERIALISED) {
2475         return SPAPR_CAP_FIXED_IBS;
2476     }
2477 
2478     return 0;
2479 }
2480 
2481 static int parse_cap_ppc_count_cache_flush_assist(struct kvm_ppc_cpu_char c)
2482 {
2483     if (c.character & c.character_mask & H_CPU_CHAR_BCCTR_FLUSH_ASSIST) {
2484         return 1;
2485     }
2486     return 0;
2487 }
2488 
2489 bool kvmppc_has_cap_xive(void)
2490 {
2491     return cap_xive;
2492 }
2493 
2494 static void kvmppc_get_cpu_characteristics(KVMState *s)
2495 {
2496     struct kvm_ppc_cpu_char c;
2497     int ret;
2498 
2499     /* Assume broken */
2500     cap_ppc_safe_cache = 0;
2501     cap_ppc_safe_bounds_check = 0;
2502     cap_ppc_safe_indirect_branch = 0;
2503 
2504     ret = kvm_vm_check_extension(s, KVM_CAP_PPC_GET_CPU_CHAR);
2505     if (!ret) {
2506         return;
2507     }
2508     ret = kvm_vm_ioctl(s, KVM_PPC_GET_CPU_CHAR, &c);
2509     if (ret < 0) {
2510         return;
2511     }
2512 
2513     cap_ppc_safe_cache = parse_cap_ppc_safe_cache(c);
2514     cap_ppc_safe_bounds_check = parse_cap_ppc_safe_bounds_check(c);
2515     cap_ppc_safe_indirect_branch = parse_cap_ppc_safe_indirect_branch(c);
2516     cap_ppc_count_cache_flush_assist =
2517         parse_cap_ppc_count_cache_flush_assist(c);
2518 }
2519 
2520 int kvmppc_get_cap_safe_cache(void)
2521 {
2522     return cap_ppc_safe_cache;
2523 }
2524 
2525 int kvmppc_get_cap_safe_bounds_check(void)
2526 {
2527     return cap_ppc_safe_bounds_check;
2528 }
2529 
2530 int kvmppc_get_cap_safe_indirect_branch(void)
2531 {
2532     return cap_ppc_safe_indirect_branch;
2533 }
2534 
2535 int kvmppc_get_cap_count_cache_flush_assist(void)
2536 {
2537     return cap_ppc_count_cache_flush_assist;
2538 }
2539 
2540 bool kvmppc_has_cap_nested_kvm_hv(void)
2541 {
2542     return !!cap_ppc_nested_kvm_hv;
2543 }
2544 
2545 int kvmppc_set_cap_nested_kvm_hv(int enable)
2546 {
2547     return kvm_vm_enable_cap(kvm_state, KVM_CAP_PPC_NESTED_HV, 0, enable);
2548 }
2549 
2550 bool kvmppc_has_cap_spapr_vfio(void)
2551 {
2552     return cap_spapr_vfio;
2553 }
2554 
2555 int kvmppc_get_cap_large_decr(void)
2556 {
2557     return cap_large_decr;
2558 }
2559 
2560 int kvmppc_enable_cap_large_decr(PowerPCCPU *cpu, int enable)
2561 {
2562     CPUState *cs = CPU(cpu);
2563     uint64_t lpcr;
2564 
2565     kvm_get_one_reg(cs, KVM_REG_PPC_LPCR_64, &lpcr);
2566     /* Do we need to modify the LPCR? */
2567     if (!!(lpcr & LPCR_LD) != !!enable) {
2568         if (enable) {
2569             lpcr |= LPCR_LD;
2570         } else {
2571             lpcr &= ~LPCR_LD;
2572         }
2573         kvm_set_one_reg(cs, KVM_REG_PPC_LPCR_64, &lpcr);
2574         kvm_get_one_reg(cs, KVM_REG_PPC_LPCR_64, &lpcr);
2575 
2576         if (!!(lpcr & LPCR_LD) != !!enable) {
2577             return -1;
2578         }
2579     }
2580 
2581     return 0;
2582 }
2583 
2584 PowerPCCPUClass *kvm_ppc_get_host_cpu_class(void)
2585 {
2586     uint32_t host_pvr = mfpvr();
2587     PowerPCCPUClass *pvr_pcc;
2588 
2589     pvr_pcc = ppc_cpu_class_by_pvr(host_pvr);
2590     if (pvr_pcc == NULL) {
2591         pvr_pcc = ppc_cpu_class_by_pvr_mask(host_pvr);
2592     }
2593 
2594     return pvr_pcc;
2595 }
2596 
2597 static int kvm_ppc_register_host_cpu_type(MachineState *ms)
2598 {
2599     TypeInfo type_info = {
2600         .name = TYPE_HOST_POWERPC_CPU,
2601         .class_init = kvmppc_host_cpu_class_init,
2602     };
2603     MachineClass *mc = MACHINE_GET_CLASS(ms);
2604     PowerPCCPUClass *pvr_pcc;
2605     ObjectClass *oc;
2606     DeviceClass *dc;
2607     int i;
2608 
2609     pvr_pcc = kvm_ppc_get_host_cpu_class();
2610     if (pvr_pcc == NULL) {
2611         return -1;
2612     }
2613     type_info.parent = object_class_get_name(OBJECT_CLASS(pvr_pcc));
2614     type_register(&type_info);
2615     if (object_dynamic_cast(OBJECT(ms), TYPE_SPAPR_MACHINE)) {
2616         /* override TCG default cpu type with 'host' cpu model */
2617         mc->default_cpu_type = TYPE_HOST_POWERPC_CPU;
2618     }
2619 
2620     oc = object_class_by_name(type_info.name);
2621     g_assert(oc);
2622 
2623     /*
2624      * Update generic CPU family class alias (e.g. on a POWER8NVL host,
2625      * we want "POWER8" to be a "family" alias that points to the current
2626      * host CPU type, too)
2627      */
2628     dc = DEVICE_CLASS(ppc_cpu_get_family_class(pvr_pcc));
2629     for (i = 0; ppc_cpu_aliases[i].alias != NULL; i++) {
2630         if (strcasecmp(ppc_cpu_aliases[i].alias, dc->desc) == 0) {
2631             char *suffix;
2632 
2633             ppc_cpu_aliases[i].model = g_strdup(object_class_get_name(oc));
2634             suffix = strstr(ppc_cpu_aliases[i].model, POWERPC_CPU_TYPE_SUFFIX);
2635             if (suffix) {
2636                 *suffix = 0;
2637             }
2638             break;
2639         }
2640     }
2641 
2642     return 0;
2643 }
2644 
2645 int kvmppc_define_rtas_kernel_token(uint32_t token, const char *function)
2646 {
2647     struct kvm_rtas_token_args args = {
2648         .token = token,
2649     };
2650 
2651     if (!kvm_check_extension(kvm_state, KVM_CAP_PPC_RTAS)) {
2652         return -ENOENT;
2653     }
2654 
2655     strncpy(args.name, function, sizeof(args.name) - 1);
2656 
2657     return kvm_vm_ioctl(kvm_state, KVM_PPC_RTAS_DEFINE_TOKEN, &args);
2658 }
2659 
2660 int kvmppc_get_htab_fd(bool write, uint64_t index, Error **errp)
2661 {
2662     struct kvm_get_htab_fd s = {
2663         .flags = write ? KVM_GET_HTAB_WRITE : 0,
2664         .start_index = index,
2665     };
2666     int ret;
2667 
2668     if (!cap_htab_fd) {
2669         error_setg(errp, "KVM version doesn't support %s the HPT",
2670                    write ? "writing" : "reading");
2671         return -ENOTSUP;
2672     }
2673 
2674     ret = kvm_vm_ioctl(kvm_state, KVM_PPC_GET_HTAB_FD, &s);
2675     if (ret < 0) {
2676         error_setg(errp, "Unable to open fd for %s HPT %s KVM: %s",
2677                    write ? "writing" : "reading", write ? "to" : "from",
2678                    strerror(errno));
2679         return -errno;
2680     }
2681 
2682     return ret;
2683 }
2684 
2685 int kvmppc_save_htab(QEMUFile *f, int fd, size_t bufsize, int64_t max_ns)
2686 {
2687     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2688     uint8_t buf[bufsize];
2689     ssize_t rc;
2690 
2691     do {
2692         rc = read(fd, buf, bufsize);
2693         if (rc < 0) {
2694             fprintf(stderr, "Error reading data from KVM HTAB fd: %s\n",
2695                     strerror(errno));
2696             return rc;
2697         } else if (rc) {
2698             uint8_t *buffer = buf;
2699             ssize_t n = rc;
2700             while (n) {
2701                 struct kvm_get_htab_header *head =
2702                     (struct kvm_get_htab_header *) buffer;
2703                 size_t chunksize = sizeof(*head) +
2704                      HASH_PTE_SIZE_64 * head->n_valid;
2705 
2706                 qemu_put_be32(f, head->index);
2707                 qemu_put_be16(f, head->n_valid);
2708                 qemu_put_be16(f, head->n_invalid);
2709                 qemu_put_buffer(f, (void *)(head + 1),
2710                                 HASH_PTE_SIZE_64 * head->n_valid);
2711 
2712                 buffer += chunksize;
2713                 n -= chunksize;
2714             }
2715         }
2716     } while ((rc != 0)
2717              && ((max_ns < 0) ||
2718                  ((qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) < max_ns)));
2719 
2720     return (rc == 0) ? 1 : 0;
2721 }
2722 
2723 int kvmppc_load_htab_chunk(QEMUFile *f, int fd, uint32_t index,
2724                            uint16_t n_valid, uint16_t n_invalid)
2725 {
2726     struct kvm_get_htab_header *buf;
2727     size_t chunksize = sizeof(*buf) + n_valid * HASH_PTE_SIZE_64;
2728     ssize_t rc;
2729 
2730     buf = alloca(chunksize);
2731     buf->index = index;
2732     buf->n_valid = n_valid;
2733     buf->n_invalid = n_invalid;
2734 
2735     qemu_get_buffer(f, (void *)(buf + 1), HASH_PTE_SIZE_64 * n_valid);
2736 
2737     rc = write(fd, buf, chunksize);
2738     if (rc < 0) {
2739         fprintf(stderr, "Error writing KVM hash table: %s\n",
2740                 strerror(errno));
2741         return rc;
2742     }
2743     if (rc != chunksize) {
2744         /* We should never get a short write on a single chunk */
2745         fprintf(stderr, "Short write, restoring KVM hash table\n");
2746         return -1;
2747     }
2748     return 0;
2749 }
2750 
2751 bool kvm_arch_stop_on_emulation_error(CPUState *cpu)
2752 {
2753     return true;
2754 }
2755 
2756 void kvm_arch_init_irq_routing(KVMState *s)
2757 {
2758 }
2759 
2760 void kvmppc_read_hptes(ppc_hash_pte64_t *hptes, hwaddr ptex, int n)
2761 {
2762     int fd, rc;
2763     int i;
2764 
2765     fd = kvmppc_get_htab_fd(false, ptex, &error_abort);
2766 
2767     i = 0;
2768     while (i < n) {
2769         struct kvm_get_htab_header *hdr;
2770         int m = n < HPTES_PER_GROUP ? n : HPTES_PER_GROUP;
2771         char buf[sizeof(*hdr) + m * HASH_PTE_SIZE_64];
2772 
2773         rc = read(fd, buf, sizeof(buf));
2774         if (rc < 0) {
2775             hw_error("kvmppc_read_hptes: Unable to read HPTEs");
2776         }
2777 
2778         hdr = (struct kvm_get_htab_header *)buf;
2779         while ((i < n) && ((char *)hdr < (buf + rc))) {
2780             int invalid = hdr->n_invalid, valid = hdr->n_valid;
2781 
2782             if (hdr->index != (ptex + i)) {
2783                 hw_error("kvmppc_read_hptes: Unexpected HPTE index %"PRIu32
2784                          " != (%"HWADDR_PRIu" + %d", hdr->index, ptex, i);
2785             }
2786 
2787             if (n - i < valid) {
2788                 valid = n - i;
2789             }
2790             memcpy(hptes + i, hdr + 1, HASH_PTE_SIZE_64 * valid);
2791             i += valid;
2792 
2793             if ((n - i) < invalid) {
2794                 invalid = n - i;
2795             }
2796             memset(hptes + i, 0, invalid * HASH_PTE_SIZE_64);
2797             i += invalid;
2798 
2799             hdr = (struct kvm_get_htab_header *)
2800                 ((char *)(hdr + 1) + HASH_PTE_SIZE_64 * hdr->n_valid);
2801         }
2802     }
2803 
2804     close(fd);
2805 }
2806 
2807 void kvmppc_write_hpte(hwaddr ptex, uint64_t pte0, uint64_t pte1)
2808 {
2809     int fd, rc;
2810     struct {
2811         struct kvm_get_htab_header hdr;
2812         uint64_t pte0;
2813         uint64_t pte1;
2814     } buf;
2815 
2816     fd = kvmppc_get_htab_fd(true, 0 /* Ignored */, &error_abort);
2817 
2818     buf.hdr.n_valid = 1;
2819     buf.hdr.n_invalid = 0;
2820     buf.hdr.index = ptex;
2821     buf.pte0 = cpu_to_be64(pte0);
2822     buf.pte1 = cpu_to_be64(pte1);
2823 
2824     rc = write(fd, &buf, sizeof(buf));
2825     if (rc != sizeof(buf)) {
2826         hw_error("kvmppc_write_hpte: Unable to update KVM HPT");
2827     }
2828     close(fd);
2829 }
2830 
2831 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
2832                              uint64_t address, uint32_t data, PCIDevice *dev)
2833 {
2834     return 0;
2835 }
2836 
2837 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
2838                                 int vector, PCIDevice *dev)
2839 {
2840     return 0;
2841 }
2842 
2843 int kvm_arch_release_virq_post(int virq)
2844 {
2845     return 0;
2846 }
2847 
2848 int kvm_arch_msi_data_to_gsi(uint32_t data)
2849 {
2850     return data & 0xffff;
2851 }
2852 
2853 int kvmppc_enable_hwrng(void)
2854 {
2855     if (!kvm_enabled() || !kvm_check_extension(kvm_state, KVM_CAP_PPC_HWRNG)) {
2856         return -1;
2857     }
2858 
2859     return kvmppc_enable_hcall(kvm_state, H_RANDOM);
2860 }
2861 
2862 void kvmppc_check_papr_resize_hpt(Error **errp)
2863 {
2864     if (!kvm_enabled()) {
2865         return; /* No KVM, we're good */
2866     }
2867 
2868     if (cap_resize_hpt) {
2869         return; /* Kernel has explicit support, we're good */
2870     }
2871 
2872     /* Otherwise fallback on looking for PR KVM */
2873     if (kvmppc_is_pr(kvm_state)) {
2874         return;
2875     }
2876 
2877     error_setg(errp,
2878                "Hash page table resizing not available with this KVM version");
2879 }
2880 
2881 int kvmppc_resize_hpt_prepare(PowerPCCPU *cpu, target_ulong flags, int shift)
2882 {
2883     CPUState *cs = CPU(cpu);
2884     struct kvm_ppc_resize_hpt rhpt = {
2885         .flags = flags,
2886         .shift = shift,
2887     };
2888 
2889     if (!cap_resize_hpt) {
2890         return -ENOSYS;
2891     }
2892 
2893     return kvm_vm_ioctl(cs->kvm_state, KVM_PPC_RESIZE_HPT_PREPARE, &rhpt);
2894 }
2895 
2896 int kvmppc_resize_hpt_commit(PowerPCCPU *cpu, target_ulong flags, int shift)
2897 {
2898     CPUState *cs = CPU(cpu);
2899     struct kvm_ppc_resize_hpt rhpt = {
2900         .flags = flags,
2901         .shift = shift,
2902     };
2903 
2904     if (!cap_resize_hpt) {
2905         return -ENOSYS;
2906     }
2907 
2908     return kvm_vm_ioctl(cs->kvm_state, KVM_PPC_RESIZE_HPT_COMMIT, &rhpt);
2909 }
2910 
2911 /*
2912  * This is a helper function to detect a post migration scenario
2913  * in which a guest, running as KVM-HV, freezes in cpu_post_load because
2914  * the guest kernel can't handle a PVR value other than the actual host
2915  * PVR in KVM_SET_SREGS, even if pvr_match() returns true.
2916  *
2917  * If we don't have cap_ppc_pvr_compat and we're not running in PR
2918  * (so, we're HV), return true. The workaround itself is done in
2919  * cpu_post_load.
2920  *
2921  * The order here is important: we'll only check for KVM PR as a
2922  * fallback if the guest kernel can't handle the situation itself.
2923  * We need to avoid as much as possible querying the running KVM type
2924  * in QEMU level.
2925  */
2926 bool kvmppc_pvr_workaround_required(PowerPCCPU *cpu)
2927 {
2928     CPUState *cs = CPU(cpu);
2929 
2930     if (!kvm_enabled()) {
2931         return false;
2932     }
2933 
2934     if (cap_ppc_pvr_compat) {
2935         return false;
2936     }
2937 
2938     return !kvmppc_is_pr(cs->kvm_state);
2939 }
2940 
2941 void kvmppc_set_reg_ppc_online(PowerPCCPU *cpu, unsigned int online)
2942 {
2943     CPUState *cs = CPU(cpu);
2944 
2945     if (kvm_enabled()) {
2946         kvm_set_one_reg(cs, KVM_REG_PPC_ONLINE, &online);
2947     }
2948 }
2949 
2950 void kvmppc_set_reg_tb_offset(PowerPCCPU *cpu, int64_t tb_offset)
2951 {
2952     CPUState *cs = CPU(cpu);
2953 
2954     if (kvm_enabled()) {
2955         kvm_set_one_reg(cs, KVM_REG_PPC_TB_OFFSET, &tb_offset);
2956     }
2957 }
2958