xref: /qemu/target/ppc/kvm.c (revision 123eec655287e43e0e86154e8093a394aefa3958)
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/sysemu.h"
31 #include "sysemu/hw_accel.h"
32 #include "kvm_ppc.h"
33 #include "sysemu/cpus.h"
34 #include "sysemu/device_tree.h"
35 #include "mmu-hash64.h"
36 
37 #include "hw/sysbus.h"
38 #include "hw/ppc/spapr.h"
39 #include "hw/ppc/spapr_vio.h"
40 #include "hw/ppc/spapr_cpu_core.h"
41 #include "hw/ppc/ppc.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/mmap-alloc.h"
50 #include "elf.h"
51 #include "sysemu/kvm_int.h"
52 
53 //#define DEBUG_KVM
54 
55 #ifdef DEBUG_KVM
56 #define DPRINTF(fmt, ...) \
57     do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
58 #else
59 #define DPRINTF(fmt, ...) \
60     do { } while (0)
61 #endif
62 
63 #define PROC_DEVTREE_CPU      "/proc/device-tree/cpus/"
64 
65 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
66     KVM_CAP_LAST_INFO
67 };
68 
69 static int cap_interrupt_unset = false;
70 static int cap_interrupt_level = false;
71 static int cap_segstate;
72 static int cap_booke_sregs;
73 static int cap_ppc_smt;
74 static int cap_ppc_smt_possible;
75 static int cap_spapr_tce;
76 static int cap_spapr_tce_64;
77 static int cap_spapr_multitce;
78 static int cap_spapr_vfio;
79 static int cap_hior;
80 static int cap_one_reg;
81 static int cap_epr;
82 static int cap_ppc_watchdog;
83 static int cap_papr;
84 static int cap_htab_fd;
85 static int cap_fixup_hcalls;
86 static int cap_htm;             /* Hardware transactional memory support */
87 static int cap_mmu_radix;
88 static int cap_mmu_hash_v3;
89 static int cap_resize_hpt;
90 static int cap_ppc_pvr_compat;
91 static int cap_ppc_safe_cache;
92 static int cap_ppc_safe_bounds_check;
93 static int cap_ppc_safe_indirect_branch;
94 
95 static uint32_t debug_inst_opcode;
96 
97 /* XXX We have a race condition where we actually have a level triggered
98  *     interrupt, but the infrastructure can't expose that yet, so the guest
99  *     takes but ignores it, goes to sleep and never gets notified that there's
100  *     still an interrupt pending.
101  *
102  *     As a quick workaround, let's just wake up again 20 ms after we injected
103  *     an interrupt. That way we can assure that we're always reinjecting
104  *     interrupts in case the guest swallowed them.
105  */
106 static QEMUTimer *idle_timer;
107 
108 static void kvm_kick_cpu(void *opaque)
109 {
110     PowerPCCPU *cpu = opaque;
111 
112     qemu_cpu_kick(CPU(cpu));
113 }
114 
115 /* Check whether we are running with KVM-PR (instead of KVM-HV).  This
116  * should only be used for fallback tests - generally we should use
117  * explicit capabilities for the features we want, rather than
118  * assuming what is/isn't available depending on the KVM variant. */
119 static bool kvmppc_is_pr(KVMState *ks)
120 {
121     /* Assume KVM-PR if the GET_PVINFO capability is available */
122     return kvm_vm_check_extension(ks, KVM_CAP_PPC_GET_PVINFO) != 0;
123 }
124 
125 static int kvm_ppc_register_host_cpu_type(MachineState *ms);
126 static void kvmppc_get_cpu_characteristics(KVMState *s);
127 
128 int kvm_arch_init(MachineState *ms, KVMState *s)
129 {
130     cap_interrupt_unset = kvm_check_extension(s, KVM_CAP_PPC_UNSET_IRQ);
131     cap_interrupt_level = kvm_check_extension(s, KVM_CAP_PPC_IRQ_LEVEL);
132     cap_segstate = kvm_check_extension(s, KVM_CAP_PPC_SEGSTATE);
133     cap_booke_sregs = kvm_check_extension(s, KVM_CAP_PPC_BOOKE_SREGS);
134     cap_ppc_smt_possible = kvm_vm_check_extension(s, KVM_CAP_PPC_SMT_POSSIBLE);
135     cap_spapr_tce = kvm_check_extension(s, KVM_CAP_SPAPR_TCE);
136     cap_spapr_tce_64 = kvm_check_extension(s, KVM_CAP_SPAPR_TCE_64);
137     cap_spapr_multitce = kvm_check_extension(s, KVM_CAP_SPAPR_MULTITCE);
138     cap_spapr_vfio = kvm_vm_check_extension(s, KVM_CAP_SPAPR_TCE_VFIO);
139     cap_one_reg = kvm_check_extension(s, KVM_CAP_ONE_REG);
140     cap_hior = kvm_check_extension(s, KVM_CAP_PPC_HIOR);
141     cap_epr = kvm_check_extension(s, KVM_CAP_PPC_EPR);
142     cap_ppc_watchdog = kvm_check_extension(s, KVM_CAP_PPC_BOOKE_WATCHDOG);
143     /* Note: we don't set cap_papr here, because this capability is
144      * only activated after this by kvmppc_set_papr() */
145     cap_htab_fd = kvm_vm_check_extension(s, KVM_CAP_PPC_HTAB_FD);
146     cap_fixup_hcalls = kvm_check_extension(s, KVM_CAP_PPC_FIXUP_HCALL);
147     cap_ppc_smt = kvm_vm_check_extension(s, KVM_CAP_PPC_SMT);
148     cap_htm = kvm_vm_check_extension(s, KVM_CAP_PPC_HTM);
149     cap_mmu_radix = kvm_vm_check_extension(s, KVM_CAP_PPC_MMU_RADIX);
150     cap_mmu_hash_v3 = kvm_vm_check_extension(s, KVM_CAP_PPC_MMU_HASH_V3);
151     cap_resize_hpt = kvm_vm_check_extension(s, KVM_CAP_SPAPR_RESIZE_HPT);
152     kvmppc_get_cpu_characteristics(s);
153     /*
154      * Note: setting it to false because there is not such capability
155      * in KVM at this moment.
156      *
157      * TODO: call kvm_vm_check_extension() with the right capability
158      * after the kernel starts implementing it.*/
159     cap_ppc_pvr_compat = false;
160 
161     if (!cap_interrupt_level) {
162         fprintf(stderr, "KVM: Couldn't find level irq capability. Expect the "
163                         "VM to stall at times!\n");
164     }
165 
166     kvm_ppc_register_host_cpu_type(ms);
167 
168     return 0;
169 }
170 
171 int kvm_arch_irqchip_create(MachineState *ms, KVMState *s)
172 {
173     return 0;
174 }
175 
176 static int kvm_arch_sync_sregs(PowerPCCPU *cpu)
177 {
178     CPUPPCState *cenv = &cpu->env;
179     CPUState *cs = CPU(cpu);
180     struct kvm_sregs sregs;
181     int ret;
182 
183     if (cenv->excp_model == POWERPC_EXCP_BOOKE) {
184         /* What we're really trying to say is "if we're on BookE, we use
185            the native PVR for now". This is the only sane way to check
186            it though, so we potentially confuse users that they can run
187            BookE guests on BookS. Let's hope nobody dares enough :) */
188         return 0;
189     } else {
190         if (!cap_segstate) {
191             fprintf(stderr, "kvm error: missing PVR setting capability\n");
192             return -ENOSYS;
193         }
194     }
195 
196     ret = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs);
197     if (ret) {
198         return ret;
199     }
200 
201     sregs.pvr = cenv->spr[SPR_PVR];
202     return kvm_vcpu_ioctl(cs, KVM_SET_SREGS, &sregs);
203 }
204 
205 /* Set up a shared TLB array with KVM */
206 static int kvm_booke206_tlb_init(PowerPCCPU *cpu)
207 {
208     CPUPPCState *env = &cpu->env;
209     CPUState *cs = CPU(cpu);
210     struct kvm_book3e_206_tlb_params params = {};
211     struct kvm_config_tlb cfg = {};
212     unsigned int entries = 0;
213     int ret, i;
214 
215     if (!kvm_enabled() ||
216         !kvm_check_extension(cs->kvm_state, KVM_CAP_SW_TLB)) {
217         return 0;
218     }
219 
220     assert(ARRAY_SIZE(params.tlb_sizes) == BOOKE206_MAX_TLBN);
221 
222     for (i = 0; i < BOOKE206_MAX_TLBN; i++) {
223         params.tlb_sizes[i] = booke206_tlb_size(env, i);
224         params.tlb_ways[i] = booke206_tlb_ways(env, i);
225         entries += params.tlb_sizes[i];
226     }
227 
228     assert(entries == env->nb_tlb);
229     assert(sizeof(struct kvm_book3e_206_tlb_entry) == sizeof(ppcmas_tlb_t));
230 
231     env->tlb_dirty = true;
232 
233     cfg.array = (uintptr_t)env->tlb.tlbm;
234     cfg.array_len = sizeof(ppcmas_tlb_t) * entries;
235     cfg.params = (uintptr_t)&params;
236     cfg.mmu_type = KVM_MMU_FSL_BOOKE_NOHV;
237 
238     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_SW_TLB, 0, (uintptr_t)&cfg);
239     if (ret < 0) {
240         fprintf(stderr, "%s: couldn't enable KVM_CAP_SW_TLB: %s\n",
241                 __func__, strerror(-ret));
242         return ret;
243     }
244 
245     env->kvm_sw_tlb = true;
246     return 0;
247 }
248 
249 
250 #if defined(TARGET_PPC64)
251 static void kvm_get_fallback_smmu_info(PowerPCCPU *cpu,
252                                        struct kvm_ppc_smmu_info *info)
253 {
254     CPUPPCState *env = &cpu->env;
255     CPUState *cs = CPU(cpu);
256 
257     memset(info, 0, sizeof(*info));
258 
259     /* We don't have the new KVM_PPC_GET_SMMU_INFO ioctl, so
260      * need to "guess" what the supported page sizes are.
261      *
262      * For that to work we make a few assumptions:
263      *
264      * - Check whether we are running "PR" KVM which only supports 4K
265      *   and 16M pages, but supports them regardless of the backing
266      *   store characteritics. We also don't support 1T segments.
267      *
268      *   This is safe as if HV KVM ever supports that capability or PR
269      *   KVM grows supports for more page/segment sizes, those versions
270      *   will have implemented KVM_CAP_PPC_GET_SMMU_INFO and thus we
271      *   will not hit this fallback
272      *
273      * - Else we are running HV KVM. This means we only support page
274      *   sizes that fit in the backing store. Additionally we only
275      *   advertize 64K pages if the processor is ARCH 2.06 and we assume
276      *   P7 encodings for the SLB and hash table. Here too, we assume
277      *   support for any newer processor will mean a kernel that
278      *   implements KVM_CAP_PPC_GET_SMMU_INFO and thus doesn't hit
279      *   this fallback.
280      */
281     if (kvmppc_is_pr(cs->kvm_state)) {
282         /* No flags */
283         info->flags = 0;
284         info->slb_size = 64;
285 
286         /* Standard 4k base page size segment */
287         info->sps[0].page_shift = 12;
288         info->sps[0].slb_enc = 0;
289         info->sps[0].enc[0].page_shift = 12;
290         info->sps[0].enc[0].pte_enc = 0;
291 
292         /* Standard 16M large page size segment */
293         info->sps[1].page_shift = 24;
294         info->sps[1].slb_enc = SLB_VSID_L;
295         info->sps[1].enc[0].page_shift = 24;
296         info->sps[1].enc[0].pte_enc = 0;
297     } else {
298         int i = 0;
299 
300         /* HV KVM has backing store size restrictions */
301         info->flags = KVM_PPC_PAGE_SIZES_REAL;
302 
303         if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
304             info->flags |= KVM_PPC_1T_SEGMENTS;
305         }
306 
307         if (env->mmu_model == POWERPC_MMU_2_06 ||
308             env->mmu_model == POWERPC_MMU_2_07) {
309             info->slb_size = 32;
310         } else {
311             info->slb_size = 64;
312         }
313 
314         /* Standard 4k base page size segment */
315         info->sps[i].page_shift = 12;
316         info->sps[i].slb_enc = 0;
317         info->sps[i].enc[0].page_shift = 12;
318         info->sps[i].enc[0].pte_enc = 0;
319         i++;
320 
321         /* 64K on MMU 2.06 and later */
322         if (env->mmu_model == POWERPC_MMU_2_06 ||
323             env->mmu_model == POWERPC_MMU_2_07) {
324             info->sps[i].page_shift = 16;
325             info->sps[i].slb_enc = 0x110;
326             info->sps[i].enc[0].page_shift = 16;
327             info->sps[i].enc[0].pte_enc = 1;
328             i++;
329         }
330 
331         /* Standard 16M large page size segment */
332         info->sps[i].page_shift = 24;
333         info->sps[i].slb_enc = SLB_VSID_L;
334         info->sps[i].enc[0].page_shift = 24;
335         info->sps[i].enc[0].pte_enc = 0;
336     }
337 }
338 
339 static void kvm_get_smmu_info(PowerPCCPU *cpu, struct kvm_ppc_smmu_info *info)
340 {
341     CPUState *cs = CPU(cpu);
342     int ret;
343 
344     if (kvm_check_extension(cs->kvm_state, KVM_CAP_PPC_GET_SMMU_INFO)) {
345         ret = kvm_vm_ioctl(cs->kvm_state, KVM_PPC_GET_SMMU_INFO, info);
346         if (ret == 0) {
347             return;
348         }
349     }
350 
351     kvm_get_fallback_smmu_info(cpu, info);
352 }
353 
354 struct ppc_radix_page_info *kvm_get_radix_page_info(void)
355 {
356     KVMState *s = KVM_STATE(current_machine->accelerator);
357     struct ppc_radix_page_info *radix_page_info;
358     struct kvm_ppc_rmmu_info rmmu_info;
359     int i;
360 
361     if (!kvm_check_extension(s, KVM_CAP_PPC_MMU_RADIX)) {
362         return NULL;
363     }
364     if (kvm_vm_ioctl(s, KVM_PPC_GET_RMMU_INFO, &rmmu_info)) {
365         return NULL;
366     }
367     radix_page_info = g_malloc0(sizeof(*radix_page_info));
368     radix_page_info->count = 0;
369     for (i = 0; i < PPC_PAGE_SIZES_MAX_SZ; i++) {
370         if (rmmu_info.ap_encodings[i]) {
371             radix_page_info->entries[i] = rmmu_info.ap_encodings[i];
372             radix_page_info->count++;
373         }
374     }
375     return radix_page_info;
376 }
377 
378 target_ulong kvmppc_configure_v3_mmu(PowerPCCPU *cpu,
379                                      bool radix, bool gtse,
380                                      uint64_t proc_tbl)
381 {
382     CPUState *cs = CPU(cpu);
383     int ret;
384     uint64_t flags = 0;
385     struct kvm_ppc_mmuv3_cfg cfg = {
386         .process_table = proc_tbl,
387     };
388 
389     if (radix) {
390         flags |= KVM_PPC_MMUV3_RADIX;
391     }
392     if (gtse) {
393         flags |= KVM_PPC_MMUV3_GTSE;
394     }
395     cfg.flags = flags;
396     ret = kvm_vm_ioctl(cs->kvm_state, KVM_PPC_CONFIGURE_V3_MMU, &cfg);
397     switch (ret) {
398     case 0:
399         return H_SUCCESS;
400     case -EINVAL:
401         return H_PARAMETER;
402     case -ENODEV:
403         return H_NOT_AVAILABLE;
404     default:
405         return H_HARDWARE;
406     }
407 }
408 
409 bool kvmppc_hpt_needs_host_contiguous_pages(void)
410 {
411     PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
412     static struct kvm_ppc_smmu_info smmu_info;
413 
414     if (!kvm_enabled()) {
415         return false;
416     }
417 
418     kvm_get_smmu_info(cpu, &smmu_info);
419     return !!(smmu_info.flags & KVM_PPC_PAGE_SIZES_REAL);
420 }
421 
422 static bool kvm_valid_page_size(uint32_t flags, long rampgsize, uint32_t shift)
423 {
424     if (!kvmppc_hpt_needs_host_contiguous_pages()) {
425         return true;
426     }
427 
428     return (1ul << shift) <= rampgsize;
429 }
430 
431 static long max_cpu_page_size;
432 
433 static void kvm_fixup_page_sizes(PowerPCCPU *cpu)
434 {
435     static struct kvm_ppc_smmu_info smmu_info;
436     static bool has_smmu_info;
437     CPUPPCState *env = &cpu->env;
438     int iq, ik, jq, jk;
439 
440     /* We only handle page sizes for 64-bit server guests for now */
441     if (!(env->mmu_model & POWERPC_MMU_64)) {
442         return;
443     }
444 
445     /* Collect MMU info from kernel if not already */
446     if (!has_smmu_info) {
447         kvm_get_smmu_info(cpu, &smmu_info);
448         has_smmu_info = true;
449     }
450 
451     if (!max_cpu_page_size) {
452         max_cpu_page_size = qemu_getrampagesize();
453     }
454 
455     /* Convert to QEMU form */
456     memset(cpu->hash64_opts->sps, 0, sizeof(*cpu->hash64_opts->sps));
457 
458     /* If we have HV KVM, we need to forbid CI large pages if our
459      * host page size is smaller than 64K.
460      */
461     if (kvmppc_hpt_needs_host_contiguous_pages()) {
462         if (getpagesize() >= 0x10000) {
463             cpu->hash64_opts->flags |= PPC_HASH64_CI_LARGEPAGE;
464         } else {
465             cpu->hash64_opts->flags &= ~PPC_HASH64_CI_LARGEPAGE;
466         }
467     }
468 
469     /*
470      * XXX This loop should be an entry wide AND of the capabilities that
471      *     the selected CPU has with the capabilities that KVM supports.
472      */
473     for (ik = iq = 0; ik < KVM_PPC_PAGE_SIZES_MAX_SZ; ik++) {
474         PPCHash64SegmentPageSizes *qsps = &cpu->hash64_opts->sps[iq];
475         struct kvm_ppc_one_seg_page_size *ksps = &smmu_info.sps[ik];
476 
477         if (!kvm_valid_page_size(smmu_info.flags, max_cpu_page_size,
478                                  ksps->page_shift)) {
479             continue;
480         }
481         qsps->page_shift = ksps->page_shift;
482         qsps->slb_enc = ksps->slb_enc;
483         for (jk = jq = 0; jk < KVM_PPC_PAGE_SIZES_MAX_SZ; jk++) {
484             if (!kvm_valid_page_size(smmu_info.flags, max_cpu_page_size,
485                                      ksps->enc[jk].page_shift)) {
486                 continue;
487             }
488             qsps->enc[jq].page_shift = ksps->enc[jk].page_shift;
489             qsps->enc[jq].pte_enc = ksps->enc[jk].pte_enc;
490             if (++jq >= PPC_PAGE_SIZES_MAX_SZ) {
491                 break;
492             }
493         }
494         if (++iq >= PPC_PAGE_SIZES_MAX_SZ) {
495             break;
496         }
497     }
498     cpu->hash64_opts->slb_size = smmu_info.slb_size;
499     if (!(smmu_info.flags & KVM_PPC_1T_SEGMENTS)) {
500         cpu->hash64_opts->flags &= ~PPC_HASH64_1TSEG;
501     }
502 }
503 #else /* defined (TARGET_PPC64) */
504 
505 static inline void kvm_fixup_page_sizes(PowerPCCPU *cpu)
506 {
507 }
508 
509 #endif /* !defined (TARGET_PPC64) */
510 
511 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
512 {
513     return POWERPC_CPU(cpu)->vcpu_id;
514 }
515 
516 /* e500 supports 2 h/w breakpoint and 2 watchpoint.
517  * book3s supports only 1 watchpoint, so array size
518  * of 4 is sufficient for now.
519  */
520 #define MAX_HW_BKPTS 4
521 
522 static struct HWBreakpoint {
523     target_ulong addr;
524     int type;
525 } hw_debug_points[MAX_HW_BKPTS];
526 
527 static CPUWatchpoint hw_watchpoint;
528 
529 /* Default there is no breakpoint and watchpoint supported */
530 static int max_hw_breakpoint;
531 static int max_hw_watchpoint;
532 static int nb_hw_breakpoint;
533 static int nb_hw_watchpoint;
534 
535 static void kvmppc_hw_debug_points_init(CPUPPCState *cenv)
536 {
537     if (cenv->excp_model == POWERPC_EXCP_BOOKE) {
538         max_hw_breakpoint = 2;
539         max_hw_watchpoint = 2;
540     }
541 
542     if ((max_hw_breakpoint + max_hw_watchpoint) > MAX_HW_BKPTS) {
543         fprintf(stderr, "Error initializing h/w breakpoints\n");
544         return;
545     }
546 }
547 
548 int kvm_arch_init_vcpu(CPUState *cs)
549 {
550     PowerPCCPU *cpu = POWERPC_CPU(cs);
551     CPUPPCState *cenv = &cpu->env;
552     int ret;
553 
554     /* Gather server mmu info from KVM and update the CPU state */
555     kvm_fixup_page_sizes(cpu);
556 
557     /* Synchronize sregs with kvm */
558     ret = kvm_arch_sync_sregs(cpu);
559     if (ret) {
560         if (ret == -EINVAL) {
561             error_report("Register sync failed... If you're using kvm-hv.ko,"
562                          " only \"-cpu host\" is possible");
563         }
564         return ret;
565     }
566 
567     idle_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, kvm_kick_cpu, cpu);
568 
569     switch (cenv->mmu_model) {
570     case POWERPC_MMU_BOOKE206:
571         /* This target supports access to KVM's guest TLB */
572         ret = kvm_booke206_tlb_init(cpu);
573         break;
574     case POWERPC_MMU_2_07:
575         if (!cap_htm && !kvmppc_is_pr(cs->kvm_state)) {
576             /* KVM-HV has transactional memory on POWER8 also without the
577              * KVM_CAP_PPC_HTM extension, so enable it here instead as
578              * long as it's availble to userspace on the host. */
579             if (qemu_getauxval(AT_HWCAP2) & PPC_FEATURE2_HAS_HTM) {
580                 cap_htm = true;
581             }
582         }
583         break;
584     default:
585         break;
586     }
587 
588     kvm_get_one_reg(cs, KVM_REG_PPC_DEBUG_INST, &debug_inst_opcode);
589     kvmppc_hw_debug_points_init(cenv);
590 
591     return ret;
592 }
593 
594 static void kvm_sw_tlb_put(PowerPCCPU *cpu)
595 {
596     CPUPPCState *env = &cpu->env;
597     CPUState *cs = CPU(cpu);
598     struct kvm_dirty_tlb dirty_tlb;
599     unsigned char *bitmap;
600     int ret;
601 
602     if (!env->kvm_sw_tlb) {
603         return;
604     }
605 
606     bitmap = g_malloc((env->nb_tlb + 7) / 8);
607     memset(bitmap, 0xFF, (env->nb_tlb + 7) / 8);
608 
609     dirty_tlb.bitmap = (uintptr_t)bitmap;
610     dirty_tlb.num_dirty = env->nb_tlb;
611 
612     ret = kvm_vcpu_ioctl(cs, KVM_DIRTY_TLB, &dirty_tlb);
613     if (ret) {
614         fprintf(stderr, "%s: KVM_DIRTY_TLB: %s\n",
615                 __func__, strerror(-ret));
616     }
617 
618     g_free(bitmap);
619 }
620 
621 static void kvm_get_one_spr(CPUState *cs, uint64_t id, int spr)
622 {
623     PowerPCCPU *cpu = POWERPC_CPU(cs);
624     CPUPPCState *env = &cpu->env;
625     union {
626         uint32_t u32;
627         uint64_t u64;
628     } val;
629     struct kvm_one_reg reg = {
630         .id = id,
631         .addr = (uintptr_t) &val,
632     };
633     int ret;
634 
635     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
636     if (ret != 0) {
637         trace_kvm_failed_spr_get(spr, strerror(errno));
638     } else {
639         switch (id & KVM_REG_SIZE_MASK) {
640         case KVM_REG_SIZE_U32:
641             env->spr[spr] = val.u32;
642             break;
643 
644         case KVM_REG_SIZE_U64:
645             env->spr[spr] = val.u64;
646             break;
647 
648         default:
649             /* Don't handle this size yet */
650             abort();
651         }
652     }
653 }
654 
655 static void kvm_put_one_spr(CPUState *cs, uint64_t id, int spr)
656 {
657     PowerPCCPU *cpu = POWERPC_CPU(cs);
658     CPUPPCState *env = &cpu->env;
659     union {
660         uint32_t u32;
661         uint64_t u64;
662     } val;
663     struct kvm_one_reg reg = {
664         .id = id,
665         .addr = (uintptr_t) &val,
666     };
667     int ret;
668 
669     switch (id & KVM_REG_SIZE_MASK) {
670     case KVM_REG_SIZE_U32:
671         val.u32 = env->spr[spr];
672         break;
673 
674     case KVM_REG_SIZE_U64:
675         val.u64 = env->spr[spr];
676         break;
677 
678     default:
679         /* Don't handle this size yet */
680         abort();
681     }
682 
683     ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
684     if (ret != 0) {
685         trace_kvm_failed_spr_set(spr, strerror(errno));
686     }
687 }
688 
689 static int kvm_put_fp(CPUState *cs)
690 {
691     PowerPCCPU *cpu = POWERPC_CPU(cs);
692     CPUPPCState *env = &cpu->env;
693     struct kvm_one_reg reg;
694     int i;
695     int ret;
696 
697     if (env->insns_flags & PPC_FLOAT) {
698         uint64_t fpscr = env->fpscr;
699         bool vsx = !!(env->insns_flags2 & PPC2_VSX);
700 
701         reg.id = KVM_REG_PPC_FPSCR;
702         reg.addr = (uintptr_t)&fpscr;
703         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
704         if (ret < 0) {
705             DPRINTF("Unable to set FPSCR to KVM: %s\n", strerror(errno));
706             return ret;
707         }
708 
709         for (i = 0; i < 32; i++) {
710             uint64_t vsr[2];
711 
712 #ifdef HOST_WORDS_BIGENDIAN
713             vsr[0] = float64_val(env->fpr[i]);
714             vsr[1] = env->vsr[i];
715 #else
716             vsr[0] = env->vsr[i];
717             vsr[1] = float64_val(env->fpr[i]);
718 #endif
719             reg.addr = (uintptr_t) &vsr;
720             reg.id = vsx ? KVM_REG_PPC_VSR(i) : KVM_REG_PPC_FPR(i);
721 
722             ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
723             if (ret < 0) {
724                 DPRINTF("Unable to set %s%d to KVM: %s\n", vsx ? "VSR" : "FPR",
725                         i, strerror(errno));
726                 return ret;
727             }
728         }
729     }
730 
731     if (env->insns_flags & PPC_ALTIVEC) {
732         reg.id = KVM_REG_PPC_VSCR;
733         reg.addr = (uintptr_t)&env->vscr;
734         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
735         if (ret < 0) {
736             DPRINTF("Unable to set VSCR to KVM: %s\n", strerror(errno));
737             return ret;
738         }
739 
740         for (i = 0; i < 32; i++) {
741             reg.id = KVM_REG_PPC_VR(i);
742             reg.addr = (uintptr_t)&env->avr[i];
743             ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
744             if (ret < 0) {
745                 DPRINTF("Unable to set VR%d to KVM: %s\n", i, strerror(errno));
746                 return ret;
747             }
748         }
749     }
750 
751     return 0;
752 }
753 
754 static int kvm_get_fp(CPUState *cs)
755 {
756     PowerPCCPU *cpu = POWERPC_CPU(cs);
757     CPUPPCState *env = &cpu->env;
758     struct kvm_one_reg reg;
759     int i;
760     int ret;
761 
762     if (env->insns_flags & PPC_FLOAT) {
763         uint64_t fpscr;
764         bool vsx = !!(env->insns_flags2 & PPC2_VSX);
765 
766         reg.id = KVM_REG_PPC_FPSCR;
767         reg.addr = (uintptr_t)&fpscr;
768         ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
769         if (ret < 0) {
770             DPRINTF("Unable to get FPSCR from KVM: %s\n", strerror(errno));
771             return ret;
772         } else {
773             env->fpscr = fpscr;
774         }
775 
776         for (i = 0; i < 32; i++) {
777             uint64_t vsr[2];
778 
779             reg.addr = (uintptr_t) &vsr;
780             reg.id = vsx ? KVM_REG_PPC_VSR(i) : KVM_REG_PPC_FPR(i);
781 
782             ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
783             if (ret < 0) {
784                 DPRINTF("Unable to get %s%d from KVM: %s\n",
785                         vsx ? "VSR" : "FPR", i, strerror(errno));
786                 return ret;
787             } else {
788 #ifdef HOST_WORDS_BIGENDIAN
789                 env->fpr[i] = vsr[0];
790                 if (vsx) {
791                     env->vsr[i] = vsr[1];
792                 }
793 #else
794                 env->fpr[i] = vsr[1];
795                 if (vsx) {
796                     env->vsr[i] = vsr[0];
797                 }
798 #endif
799             }
800         }
801     }
802 
803     if (env->insns_flags & PPC_ALTIVEC) {
804         reg.id = KVM_REG_PPC_VSCR;
805         reg.addr = (uintptr_t)&env->vscr;
806         ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
807         if (ret < 0) {
808             DPRINTF("Unable to get VSCR from KVM: %s\n", strerror(errno));
809             return ret;
810         }
811 
812         for (i = 0; i < 32; i++) {
813             reg.id = KVM_REG_PPC_VR(i);
814             reg.addr = (uintptr_t)&env->avr[i];
815             ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
816             if (ret < 0) {
817                 DPRINTF("Unable to get VR%d from KVM: %s\n",
818                         i, strerror(errno));
819                 return ret;
820             }
821         }
822     }
823 
824     return 0;
825 }
826 
827 #if defined(TARGET_PPC64)
828 static int kvm_get_vpa(CPUState *cs)
829 {
830     PowerPCCPU *cpu = POWERPC_CPU(cs);
831     sPAPRCPUState *spapr_cpu = spapr_cpu_state(cpu);
832     struct kvm_one_reg reg;
833     int ret;
834 
835     reg.id = KVM_REG_PPC_VPA_ADDR;
836     reg.addr = (uintptr_t)&spapr_cpu->vpa_addr;
837     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
838     if (ret < 0) {
839         DPRINTF("Unable to get VPA address from KVM: %s\n", strerror(errno));
840         return ret;
841     }
842 
843     assert((uintptr_t)&spapr_cpu->slb_shadow_size
844            == ((uintptr_t)&spapr_cpu->slb_shadow_addr + 8));
845     reg.id = KVM_REG_PPC_VPA_SLB;
846     reg.addr = (uintptr_t)&spapr_cpu->slb_shadow_addr;
847     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
848     if (ret < 0) {
849         DPRINTF("Unable to get SLB shadow state from KVM: %s\n",
850                 strerror(errno));
851         return ret;
852     }
853 
854     assert((uintptr_t)&spapr_cpu->dtl_size
855            == ((uintptr_t)&spapr_cpu->dtl_addr + 8));
856     reg.id = KVM_REG_PPC_VPA_DTL;
857     reg.addr = (uintptr_t)&spapr_cpu->dtl_addr;
858     ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
859     if (ret < 0) {
860         DPRINTF("Unable to get dispatch trace log state from KVM: %s\n",
861                 strerror(errno));
862         return ret;
863     }
864 
865     return 0;
866 }
867 
868 static int kvm_put_vpa(CPUState *cs)
869 {
870     PowerPCCPU *cpu = POWERPC_CPU(cs);
871     sPAPRCPUState *spapr_cpu = spapr_cpu_state(cpu);
872     struct kvm_one_reg reg;
873     int ret;
874 
875     /* SLB shadow or DTL can't be registered unless a master VPA is
876      * registered.  That means when restoring state, if a VPA *is*
877      * registered, we need to set that up first.  If not, we need to
878      * deregister the others before deregistering the master VPA */
879     assert(spapr_cpu->vpa_addr
880            || !(spapr_cpu->slb_shadow_addr || spapr_cpu->dtl_addr));
881 
882     if (spapr_cpu->vpa_addr) {
883         reg.id = KVM_REG_PPC_VPA_ADDR;
884         reg.addr = (uintptr_t)&spapr_cpu->vpa_addr;
885         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
886         if (ret < 0) {
887             DPRINTF("Unable to set VPA address to KVM: %s\n", strerror(errno));
888             return ret;
889         }
890     }
891 
892     assert((uintptr_t)&spapr_cpu->slb_shadow_size
893            == ((uintptr_t)&spapr_cpu->slb_shadow_addr + 8));
894     reg.id = KVM_REG_PPC_VPA_SLB;
895     reg.addr = (uintptr_t)&spapr_cpu->slb_shadow_addr;
896     ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
897     if (ret < 0) {
898         DPRINTF("Unable to set SLB shadow state to KVM: %s\n", strerror(errno));
899         return ret;
900     }
901 
902     assert((uintptr_t)&spapr_cpu->dtl_size
903            == ((uintptr_t)&spapr_cpu->dtl_addr + 8));
904     reg.id = KVM_REG_PPC_VPA_DTL;
905     reg.addr = (uintptr_t)&spapr_cpu->dtl_addr;
906     ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
907     if (ret < 0) {
908         DPRINTF("Unable to set dispatch trace log state to KVM: %s\n",
909                 strerror(errno));
910         return ret;
911     }
912 
913     if (!spapr_cpu->vpa_addr) {
914         reg.id = KVM_REG_PPC_VPA_ADDR;
915         reg.addr = (uintptr_t)&spapr_cpu->vpa_addr;
916         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
917         if (ret < 0) {
918             DPRINTF("Unable to set VPA address to KVM: %s\n", strerror(errno));
919             return ret;
920         }
921     }
922 
923     return 0;
924 }
925 #endif /* TARGET_PPC64 */
926 
927 int kvmppc_put_books_sregs(PowerPCCPU *cpu)
928 {
929     CPUPPCState *env = &cpu->env;
930     struct kvm_sregs sregs;
931     int i;
932 
933     sregs.pvr = env->spr[SPR_PVR];
934 
935     if (cpu->vhyp) {
936         PPCVirtualHypervisorClass *vhc =
937             PPC_VIRTUAL_HYPERVISOR_GET_CLASS(cpu->vhyp);
938         sregs.u.s.sdr1 = vhc->encode_hpt_for_kvm_pr(cpu->vhyp);
939     } else {
940         sregs.u.s.sdr1 = env->spr[SPR_SDR1];
941     }
942 
943     /* Sync SLB */
944 #ifdef TARGET_PPC64
945     for (i = 0; i < ARRAY_SIZE(env->slb); i++) {
946         sregs.u.s.ppc64.slb[i].slbe = env->slb[i].esid;
947         if (env->slb[i].esid & SLB_ESID_V) {
948             sregs.u.s.ppc64.slb[i].slbe |= i;
949         }
950         sregs.u.s.ppc64.slb[i].slbv = env->slb[i].vsid;
951     }
952 #endif
953 
954     /* Sync SRs */
955     for (i = 0; i < 16; i++) {
956         sregs.u.s.ppc32.sr[i] = env->sr[i];
957     }
958 
959     /* Sync BATs */
960     for (i = 0; i < 8; i++) {
961         /* Beware. We have to swap upper and lower bits here */
962         sregs.u.s.ppc32.dbat[i] = ((uint64_t)env->DBAT[0][i] << 32)
963             | env->DBAT[1][i];
964         sregs.u.s.ppc32.ibat[i] = ((uint64_t)env->IBAT[0][i] << 32)
965             | env->IBAT[1][i];
966     }
967 
968     return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_SREGS, &sregs);
969 }
970 
971 int kvm_arch_put_registers(CPUState *cs, int level)
972 {
973     PowerPCCPU *cpu = POWERPC_CPU(cs);
974     CPUPPCState *env = &cpu->env;
975     struct kvm_regs regs;
976     int ret;
977     int i;
978 
979     ret = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
980     if (ret < 0) {
981         return ret;
982     }
983 
984     regs.ctr = env->ctr;
985     regs.lr  = env->lr;
986     regs.xer = cpu_read_xer(env);
987     regs.msr = env->msr;
988     regs.pc = env->nip;
989 
990     regs.srr0 = env->spr[SPR_SRR0];
991     regs.srr1 = env->spr[SPR_SRR1];
992 
993     regs.sprg0 = env->spr[SPR_SPRG0];
994     regs.sprg1 = env->spr[SPR_SPRG1];
995     regs.sprg2 = env->spr[SPR_SPRG2];
996     regs.sprg3 = env->spr[SPR_SPRG3];
997     regs.sprg4 = env->spr[SPR_SPRG4];
998     regs.sprg5 = env->spr[SPR_SPRG5];
999     regs.sprg6 = env->spr[SPR_SPRG6];
1000     regs.sprg7 = env->spr[SPR_SPRG7];
1001 
1002     regs.pid = env->spr[SPR_BOOKE_PID];
1003 
1004     for (i = 0;i < 32; i++)
1005         regs.gpr[i] = env->gpr[i];
1006 
1007     regs.cr = 0;
1008     for (i = 0; i < 8; i++) {
1009         regs.cr |= (env->crf[i] & 15) << (4 * (7 - i));
1010     }
1011 
1012     ret = kvm_vcpu_ioctl(cs, KVM_SET_REGS, &regs);
1013     if (ret < 0)
1014         return ret;
1015 
1016     kvm_put_fp(cs);
1017 
1018     if (env->tlb_dirty) {
1019         kvm_sw_tlb_put(cpu);
1020         env->tlb_dirty = false;
1021     }
1022 
1023     if (cap_segstate && (level >= KVM_PUT_RESET_STATE)) {
1024         ret = kvmppc_put_books_sregs(cpu);
1025         if (ret < 0) {
1026             return ret;
1027         }
1028     }
1029 
1030     if (cap_hior && (level >= KVM_PUT_RESET_STATE)) {
1031         kvm_put_one_spr(cs, KVM_REG_PPC_HIOR, SPR_HIOR);
1032     }
1033 
1034     if (cap_one_reg) {
1035         int i;
1036 
1037         /* We deliberately ignore errors here, for kernels which have
1038          * the ONE_REG calls, but don't support the specific
1039          * registers, there's a reasonable chance things will still
1040          * work, at least until we try to migrate. */
1041         for (i = 0; i < 1024; i++) {
1042             uint64_t id = env->spr_cb[i].one_reg_id;
1043 
1044             if (id != 0) {
1045                 kvm_put_one_spr(cs, id, i);
1046             }
1047         }
1048 
1049 #ifdef TARGET_PPC64
1050         if (msr_ts) {
1051             for (i = 0; i < ARRAY_SIZE(env->tm_gpr); i++) {
1052                 kvm_set_one_reg(cs, KVM_REG_PPC_TM_GPR(i), &env->tm_gpr[i]);
1053             }
1054             for (i = 0; i < ARRAY_SIZE(env->tm_vsr); i++) {
1055                 kvm_set_one_reg(cs, KVM_REG_PPC_TM_VSR(i), &env->tm_vsr[i]);
1056             }
1057             kvm_set_one_reg(cs, KVM_REG_PPC_TM_CR, &env->tm_cr);
1058             kvm_set_one_reg(cs, KVM_REG_PPC_TM_LR, &env->tm_lr);
1059             kvm_set_one_reg(cs, KVM_REG_PPC_TM_CTR, &env->tm_ctr);
1060             kvm_set_one_reg(cs, KVM_REG_PPC_TM_FPSCR, &env->tm_fpscr);
1061             kvm_set_one_reg(cs, KVM_REG_PPC_TM_AMR, &env->tm_amr);
1062             kvm_set_one_reg(cs, KVM_REG_PPC_TM_PPR, &env->tm_ppr);
1063             kvm_set_one_reg(cs, KVM_REG_PPC_TM_VRSAVE, &env->tm_vrsave);
1064             kvm_set_one_reg(cs, KVM_REG_PPC_TM_VSCR, &env->tm_vscr);
1065             kvm_set_one_reg(cs, KVM_REG_PPC_TM_DSCR, &env->tm_dscr);
1066             kvm_set_one_reg(cs, KVM_REG_PPC_TM_TAR, &env->tm_tar);
1067         }
1068 
1069         if (cap_papr) {
1070             if (kvm_put_vpa(cs) < 0) {
1071                 DPRINTF("Warning: Unable to set VPA information to KVM\n");
1072             }
1073         }
1074 
1075         kvm_set_one_reg(cs, KVM_REG_PPC_TB_OFFSET, &env->tb_env->tb_offset);
1076 #endif /* TARGET_PPC64 */
1077     }
1078 
1079     return ret;
1080 }
1081 
1082 static void kvm_sync_excp(CPUPPCState *env, int vector, int ivor)
1083 {
1084      env->excp_vectors[vector] = env->spr[ivor] + env->spr[SPR_BOOKE_IVPR];
1085 }
1086 
1087 static int kvmppc_get_booke_sregs(PowerPCCPU *cpu)
1088 {
1089     CPUPPCState *env = &cpu->env;
1090     struct kvm_sregs sregs;
1091     int ret;
1092 
1093     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_SREGS, &sregs);
1094     if (ret < 0) {
1095         return ret;
1096     }
1097 
1098     if (sregs.u.e.features & KVM_SREGS_E_BASE) {
1099         env->spr[SPR_BOOKE_CSRR0] = sregs.u.e.csrr0;
1100         env->spr[SPR_BOOKE_CSRR1] = sregs.u.e.csrr1;
1101         env->spr[SPR_BOOKE_ESR] = sregs.u.e.esr;
1102         env->spr[SPR_BOOKE_DEAR] = sregs.u.e.dear;
1103         env->spr[SPR_BOOKE_MCSR] = sregs.u.e.mcsr;
1104         env->spr[SPR_BOOKE_TSR] = sregs.u.e.tsr;
1105         env->spr[SPR_BOOKE_TCR] = sregs.u.e.tcr;
1106         env->spr[SPR_DECR] = sregs.u.e.dec;
1107         env->spr[SPR_TBL] = sregs.u.e.tb & 0xffffffff;
1108         env->spr[SPR_TBU] = sregs.u.e.tb >> 32;
1109         env->spr[SPR_VRSAVE] = sregs.u.e.vrsave;
1110     }
1111 
1112     if (sregs.u.e.features & KVM_SREGS_E_ARCH206) {
1113         env->spr[SPR_BOOKE_PIR] = sregs.u.e.pir;
1114         env->spr[SPR_BOOKE_MCSRR0] = sregs.u.e.mcsrr0;
1115         env->spr[SPR_BOOKE_MCSRR1] = sregs.u.e.mcsrr1;
1116         env->spr[SPR_BOOKE_DECAR] = sregs.u.e.decar;
1117         env->spr[SPR_BOOKE_IVPR] = sregs.u.e.ivpr;
1118     }
1119 
1120     if (sregs.u.e.features & KVM_SREGS_E_64) {
1121         env->spr[SPR_BOOKE_EPCR] = sregs.u.e.epcr;
1122     }
1123 
1124     if (sregs.u.e.features & KVM_SREGS_E_SPRG8) {
1125         env->spr[SPR_BOOKE_SPRG8] = sregs.u.e.sprg8;
1126     }
1127 
1128     if (sregs.u.e.features & KVM_SREGS_E_IVOR) {
1129         env->spr[SPR_BOOKE_IVOR0] = sregs.u.e.ivor_low[0];
1130         kvm_sync_excp(env, POWERPC_EXCP_CRITICAL,  SPR_BOOKE_IVOR0);
1131         env->spr[SPR_BOOKE_IVOR1] = sregs.u.e.ivor_low[1];
1132         kvm_sync_excp(env, POWERPC_EXCP_MCHECK,  SPR_BOOKE_IVOR1);
1133         env->spr[SPR_BOOKE_IVOR2] = sregs.u.e.ivor_low[2];
1134         kvm_sync_excp(env, POWERPC_EXCP_DSI,  SPR_BOOKE_IVOR2);
1135         env->spr[SPR_BOOKE_IVOR3] = sregs.u.e.ivor_low[3];
1136         kvm_sync_excp(env, POWERPC_EXCP_ISI,  SPR_BOOKE_IVOR3);
1137         env->spr[SPR_BOOKE_IVOR4] = sregs.u.e.ivor_low[4];
1138         kvm_sync_excp(env, POWERPC_EXCP_EXTERNAL,  SPR_BOOKE_IVOR4);
1139         env->spr[SPR_BOOKE_IVOR5] = sregs.u.e.ivor_low[5];
1140         kvm_sync_excp(env, POWERPC_EXCP_ALIGN,  SPR_BOOKE_IVOR5);
1141         env->spr[SPR_BOOKE_IVOR6] = sregs.u.e.ivor_low[6];
1142         kvm_sync_excp(env, POWERPC_EXCP_PROGRAM,  SPR_BOOKE_IVOR6);
1143         env->spr[SPR_BOOKE_IVOR7] = sregs.u.e.ivor_low[7];
1144         kvm_sync_excp(env, POWERPC_EXCP_FPU,  SPR_BOOKE_IVOR7);
1145         env->spr[SPR_BOOKE_IVOR8] = sregs.u.e.ivor_low[8];
1146         kvm_sync_excp(env, POWERPC_EXCP_SYSCALL,  SPR_BOOKE_IVOR8);
1147         env->spr[SPR_BOOKE_IVOR9] = sregs.u.e.ivor_low[9];
1148         kvm_sync_excp(env, POWERPC_EXCP_APU,  SPR_BOOKE_IVOR9);
1149         env->spr[SPR_BOOKE_IVOR10] = sregs.u.e.ivor_low[10];
1150         kvm_sync_excp(env, POWERPC_EXCP_DECR,  SPR_BOOKE_IVOR10);
1151         env->spr[SPR_BOOKE_IVOR11] = sregs.u.e.ivor_low[11];
1152         kvm_sync_excp(env, POWERPC_EXCP_FIT,  SPR_BOOKE_IVOR11);
1153         env->spr[SPR_BOOKE_IVOR12] = sregs.u.e.ivor_low[12];
1154         kvm_sync_excp(env, POWERPC_EXCP_WDT,  SPR_BOOKE_IVOR12);
1155         env->spr[SPR_BOOKE_IVOR13] = sregs.u.e.ivor_low[13];
1156         kvm_sync_excp(env, POWERPC_EXCP_DTLB,  SPR_BOOKE_IVOR13);
1157         env->spr[SPR_BOOKE_IVOR14] = sregs.u.e.ivor_low[14];
1158         kvm_sync_excp(env, POWERPC_EXCP_ITLB,  SPR_BOOKE_IVOR14);
1159         env->spr[SPR_BOOKE_IVOR15] = sregs.u.e.ivor_low[15];
1160         kvm_sync_excp(env, POWERPC_EXCP_DEBUG,  SPR_BOOKE_IVOR15);
1161 
1162         if (sregs.u.e.features & KVM_SREGS_E_SPE) {
1163             env->spr[SPR_BOOKE_IVOR32] = sregs.u.e.ivor_high[0];
1164             kvm_sync_excp(env, POWERPC_EXCP_SPEU,  SPR_BOOKE_IVOR32);
1165             env->spr[SPR_BOOKE_IVOR33] = sregs.u.e.ivor_high[1];
1166             kvm_sync_excp(env, POWERPC_EXCP_EFPDI,  SPR_BOOKE_IVOR33);
1167             env->spr[SPR_BOOKE_IVOR34] = sregs.u.e.ivor_high[2];
1168             kvm_sync_excp(env, POWERPC_EXCP_EFPRI,  SPR_BOOKE_IVOR34);
1169         }
1170 
1171         if (sregs.u.e.features & KVM_SREGS_E_PM) {
1172             env->spr[SPR_BOOKE_IVOR35] = sregs.u.e.ivor_high[3];
1173             kvm_sync_excp(env, POWERPC_EXCP_EPERFM,  SPR_BOOKE_IVOR35);
1174         }
1175 
1176         if (sregs.u.e.features & KVM_SREGS_E_PC) {
1177             env->spr[SPR_BOOKE_IVOR36] = sregs.u.e.ivor_high[4];
1178             kvm_sync_excp(env, POWERPC_EXCP_DOORI,  SPR_BOOKE_IVOR36);
1179             env->spr[SPR_BOOKE_IVOR37] = sregs.u.e.ivor_high[5];
1180             kvm_sync_excp(env, POWERPC_EXCP_DOORCI, SPR_BOOKE_IVOR37);
1181         }
1182     }
1183 
1184     if (sregs.u.e.features & KVM_SREGS_E_ARCH206_MMU) {
1185         env->spr[SPR_BOOKE_MAS0] = sregs.u.e.mas0;
1186         env->spr[SPR_BOOKE_MAS1] = sregs.u.e.mas1;
1187         env->spr[SPR_BOOKE_MAS2] = sregs.u.e.mas2;
1188         env->spr[SPR_BOOKE_MAS3] = sregs.u.e.mas7_3 & 0xffffffff;
1189         env->spr[SPR_BOOKE_MAS4] = sregs.u.e.mas4;
1190         env->spr[SPR_BOOKE_MAS6] = sregs.u.e.mas6;
1191         env->spr[SPR_BOOKE_MAS7] = sregs.u.e.mas7_3 >> 32;
1192         env->spr[SPR_MMUCFG] = sregs.u.e.mmucfg;
1193         env->spr[SPR_BOOKE_TLB0CFG] = sregs.u.e.tlbcfg[0];
1194         env->spr[SPR_BOOKE_TLB1CFG] = sregs.u.e.tlbcfg[1];
1195     }
1196 
1197     if (sregs.u.e.features & KVM_SREGS_EXP) {
1198         env->spr[SPR_BOOKE_EPR] = sregs.u.e.epr;
1199     }
1200 
1201     if (sregs.u.e.features & KVM_SREGS_E_PD) {
1202         env->spr[SPR_BOOKE_EPLC] = sregs.u.e.eplc;
1203         env->spr[SPR_BOOKE_EPSC] = sregs.u.e.epsc;
1204     }
1205 
1206     if (sregs.u.e.impl_id == KVM_SREGS_E_IMPL_FSL) {
1207         env->spr[SPR_E500_SVR] = sregs.u.e.impl.fsl.svr;
1208         env->spr[SPR_Exxx_MCAR] = sregs.u.e.impl.fsl.mcar;
1209         env->spr[SPR_HID0] = sregs.u.e.impl.fsl.hid0;
1210 
1211         if (sregs.u.e.impl.fsl.features & KVM_SREGS_E_FSL_PIDn) {
1212             env->spr[SPR_BOOKE_PID1] = sregs.u.e.impl.fsl.pid1;
1213             env->spr[SPR_BOOKE_PID2] = sregs.u.e.impl.fsl.pid2;
1214         }
1215     }
1216 
1217     return 0;
1218 }
1219 
1220 static int kvmppc_get_books_sregs(PowerPCCPU *cpu)
1221 {
1222     CPUPPCState *env = &cpu->env;
1223     struct kvm_sregs sregs;
1224     int ret;
1225     int i;
1226 
1227     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_SREGS, &sregs);
1228     if (ret < 0) {
1229         return ret;
1230     }
1231 
1232     if (!cpu->vhyp) {
1233         ppc_store_sdr1(env, sregs.u.s.sdr1);
1234     }
1235 
1236     /* Sync SLB */
1237 #ifdef TARGET_PPC64
1238     /*
1239      * The packed SLB array we get from KVM_GET_SREGS only contains
1240      * information about valid entries. So we flush our internal copy
1241      * to get rid of stale ones, then put all valid SLB entries back
1242      * in.
1243      */
1244     memset(env->slb, 0, sizeof(env->slb));
1245     for (i = 0; i < ARRAY_SIZE(env->slb); i++) {
1246         target_ulong rb = sregs.u.s.ppc64.slb[i].slbe;
1247         target_ulong rs = sregs.u.s.ppc64.slb[i].slbv;
1248         /*
1249          * Only restore valid entries
1250          */
1251         if (rb & SLB_ESID_V) {
1252             ppc_store_slb(cpu, rb & 0xfff, rb & ~0xfffULL, rs);
1253         }
1254     }
1255 #endif
1256 
1257     /* Sync SRs */
1258     for (i = 0; i < 16; i++) {
1259         env->sr[i] = sregs.u.s.ppc32.sr[i];
1260     }
1261 
1262     /* Sync BATs */
1263     for (i = 0; i < 8; i++) {
1264         env->DBAT[0][i] = sregs.u.s.ppc32.dbat[i] & 0xffffffff;
1265         env->DBAT[1][i] = sregs.u.s.ppc32.dbat[i] >> 32;
1266         env->IBAT[0][i] = sregs.u.s.ppc32.ibat[i] & 0xffffffff;
1267         env->IBAT[1][i] = sregs.u.s.ppc32.ibat[i] >> 32;
1268     }
1269 
1270     return 0;
1271 }
1272 
1273 int kvm_arch_get_registers(CPUState *cs)
1274 {
1275     PowerPCCPU *cpu = POWERPC_CPU(cs);
1276     CPUPPCState *env = &cpu->env;
1277     struct kvm_regs regs;
1278     uint32_t cr;
1279     int i, ret;
1280 
1281     ret = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
1282     if (ret < 0)
1283         return ret;
1284 
1285     cr = regs.cr;
1286     for (i = 7; i >= 0; i--) {
1287         env->crf[i] = cr & 15;
1288         cr >>= 4;
1289     }
1290 
1291     env->ctr = regs.ctr;
1292     env->lr = regs.lr;
1293     cpu_write_xer(env, regs.xer);
1294     env->msr = regs.msr;
1295     env->nip = regs.pc;
1296 
1297     env->spr[SPR_SRR0] = regs.srr0;
1298     env->spr[SPR_SRR1] = regs.srr1;
1299 
1300     env->spr[SPR_SPRG0] = regs.sprg0;
1301     env->spr[SPR_SPRG1] = regs.sprg1;
1302     env->spr[SPR_SPRG2] = regs.sprg2;
1303     env->spr[SPR_SPRG3] = regs.sprg3;
1304     env->spr[SPR_SPRG4] = regs.sprg4;
1305     env->spr[SPR_SPRG5] = regs.sprg5;
1306     env->spr[SPR_SPRG6] = regs.sprg6;
1307     env->spr[SPR_SPRG7] = regs.sprg7;
1308 
1309     env->spr[SPR_BOOKE_PID] = regs.pid;
1310 
1311     for (i = 0;i < 32; i++)
1312         env->gpr[i] = regs.gpr[i];
1313 
1314     kvm_get_fp(cs);
1315 
1316     if (cap_booke_sregs) {
1317         ret = kvmppc_get_booke_sregs(cpu);
1318         if (ret < 0) {
1319             return ret;
1320         }
1321     }
1322 
1323     if (cap_segstate) {
1324         ret = kvmppc_get_books_sregs(cpu);
1325         if (ret < 0) {
1326             return ret;
1327         }
1328     }
1329 
1330     if (cap_hior) {
1331         kvm_get_one_spr(cs, KVM_REG_PPC_HIOR, SPR_HIOR);
1332     }
1333 
1334     if (cap_one_reg) {
1335         int i;
1336 
1337         /* We deliberately ignore errors here, for kernels which have
1338          * the ONE_REG calls, but don't support the specific
1339          * registers, there's a reasonable chance things will still
1340          * work, at least until we try to migrate. */
1341         for (i = 0; i < 1024; i++) {
1342             uint64_t id = env->spr_cb[i].one_reg_id;
1343 
1344             if (id != 0) {
1345                 kvm_get_one_spr(cs, id, i);
1346             }
1347         }
1348 
1349 #ifdef TARGET_PPC64
1350         if (msr_ts) {
1351             for (i = 0; i < ARRAY_SIZE(env->tm_gpr); i++) {
1352                 kvm_get_one_reg(cs, KVM_REG_PPC_TM_GPR(i), &env->tm_gpr[i]);
1353             }
1354             for (i = 0; i < ARRAY_SIZE(env->tm_vsr); i++) {
1355                 kvm_get_one_reg(cs, KVM_REG_PPC_TM_VSR(i), &env->tm_vsr[i]);
1356             }
1357             kvm_get_one_reg(cs, KVM_REG_PPC_TM_CR, &env->tm_cr);
1358             kvm_get_one_reg(cs, KVM_REG_PPC_TM_LR, &env->tm_lr);
1359             kvm_get_one_reg(cs, KVM_REG_PPC_TM_CTR, &env->tm_ctr);
1360             kvm_get_one_reg(cs, KVM_REG_PPC_TM_FPSCR, &env->tm_fpscr);
1361             kvm_get_one_reg(cs, KVM_REG_PPC_TM_AMR, &env->tm_amr);
1362             kvm_get_one_reg(cs, KVM_REG_PPC_TM_PPR, &env->tm_ppr);
1363             kvm_get_one_reg(cs, KVM_REG_PPC_TM_VRSAVE, &env->tm_vrsave);
1364             kvm_get_one_reg(cs, KVM_REG_PPC_TM_VSCR, &env->tm_vscr);
1365             kvm_get_one_reg(cs, KVM_REG_PPC_TM_DSCR, &env->tm_dscr);
1366             kvm_get_one_reg(cs, KVM_REG_PPC_TM_TAR, &env->tm_tar);
1367         }
1368 
1369         if (cap_papr) {
1370             if (kvm_get_vpa(cs) < 0) {
1371                 DPRINTF("Warning: Unable to get VPA information from KVM\n");
1372             }
1373         }
1374 
1375         kvm_get_one_reg(cs, KVM_REG_PPC_TB_OFFSET, &env->tb_env->tb_offset);
1376 #endif
1377     }
1378 
1379     return 0;
1380 }
1381 
1382 int kvmppc_set_interrupt(PowerPCCPU *cpu, int irq, int level)
1383 {
1384     unsigned virq = level ? KVM_INTERRUPT_SET_LEVEL : KVM_INTERRUPT_UNSET;
1385 
1386     if (irq != PPC_INTERRUPT_EXT) {
1387         return 0;
1388     }
1389 
1390     if (!kvm_enabled() || !cap_interrupt_unset || !cap_interrupt_level) {
1391         return 0;
1392     }
1393 
1394     kvm_vcpu_ioctl(CPU(cpu), KVM_INTERRUPT, &virq);
1395 
1396     return 0;
1397 }
1398 
1399 #if defined(TARGET_PPCEMB)
1400 #define PPC_INPUT_INT PPC40x_INPUT_INT
1401 #elif defined(TARGET_PPC64)
1402 #define PPC_INPUT_INT PPC970_INPUT_INT
1403 #else
1404 #define PPC_INPUT_INT PPC6xx_INPUT_INT
1405 #endif
1406 
1407 void kvm_arch_pre_run(CPUState *cs, struct kvm_run *run)
1408 {
1409     PowerPCCPU *cpu = POWERPC_CPU(cs);
1410     CPUPPCState *env = &cpu->env;
1411     int r;
1412     unsigned irq;
1413 
1414     qemu_mutex_lock_iothread();
1415 
1416     /* PowerPC QEMU tracks the various core input pins (interrupt, critical
1417      * interrupt, reset, etc) in PPC-specific env->irq_input_state. */
1418     if (!cap_interrupt_level &&
1419         run->ready_for_interrupt_injection &&
1420         (cs->interrupt_request & CPU_INTERRUPT_HARD) &&
1421         (env->irq_input_state & (1<<PPC_INPUT_INT)))
1422     {
1423         /* For now KVM disregards the 'irq' argument. However, in the
1424          * future KVM could cache it in-kernel to avoid a heavyweight exit
1425          * when reading the UIC.
1426          */
1427         irq = KVM_INTERRUPT_SET;
1428 
1429         DPRINTF("injected interrupt %d\n", irq);
1430         r = kvm_vcpu_ioctl(cs, KVM_INTERRUPT, &irq);
1431         if (r < 0) {
1432             printf("cpu %d fail inject %x\n", cs->cpu_index, irq);
1433         }
1434 
1435         /* Always wake up soon in case the interrupt was level based */
1436         timer_mod(idle_timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) +
1437                        (NANOSECONDS_PER_SECOND / 50));
1438     }
1439 
1440     /* We don't know if there are more interrupts pending after this. However,
1441      * the guest will return to userspace in the course of handling this one
1442      * anyways, so we will get a chance to deliver the rest. */
1443 
1444     qemu_mutex_unlock_iothread();
1445 }
1446 
1447 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
1448 {
1449     return MEMTXATTRS_UNSPECIFIED;
1450 }
1451 
1452 int kvm_arch_process_async_events(CPUState *cs)
1453 {
1454     return cs->halted;
1455 }
1456 
1457 static int kvmppc_handle_halt(PowerPCCPU *cpu)
1458 {
1459     CPUState *cs = CPU(cpu);
1460     CPUPPCState *env = &cpu->env;
1461 
1462     if (!(cs->interrupt_request & CPU_INTERRUPT_HARD) && (msr_ee)) {
1463         cs->halted = 1;
1464         cs->exception_index = EXCP_HLT;
1465     }
1466 
1467     return 0;
1468 }
1469 
1470 /* map dcr access to existing qemu dcr emulation */
1471 static int kvmppc_handle_dcr_read(CPUPPCState *env, uint32_t dcrn, uint32_t *data)
1472 {
1473     if (ppc_dcr_read(env->dcr_env, dcrn, data) < 0)
1474         fprintf(stderr, "Read to unhandled DCR (0x%x)\n", dcrn);
1475 
1476     return 0;
1477 }
1478 
1479 static int kvmppc_handle_dcr_write(CPUPPCState *env, uint32_t dcrn, uint32_t data)
1480 {
1481     if (ppc_dcr_write(env->dcr_env, dcrn, data) < 0)
1482         fprintf(stderr, "Write to unhandled DCR (0x%x)\n", dcrn);
1483 
1484     return 0;
1485 }
1486 
1487 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
1488 {
1489     /* Mixed endian case is not handled */
1490     uint32_t sc = debug_inst_opcode;
1491 
1492     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
1493                             sizeof(sc), 0) ||
1494         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&sc, sizeof(sc), 1)) {
1495         return -EINVAL;
1496     }
1497 
1498     return 0;
1499 }
1500 
1501 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
1502 {
1503     uint32_t sc;
1504 
1505     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&sc, sizeof(sc), 0) ||
1506         sc != debug_inst_opcode ||
1507         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
1508                             sizeof(sc), 1)) {
1509         return -EINVAL;
1510     }
1511 
1512     return 0;
1513 }
1514 
1515 static int find_hw_breakpoint(target_ulong addr, int type)
1516 {
1517     int n;
1518 
1519     assert((nb_hw_breakpoint + nb_hw_watchpoint)
1520            <= ARRAY_SIZE(hw_debug_points));
1521 
1522     for (n = 0; n < nb_hw_breakpoint + nb_hw_watchpoint; n++) {
1523         if (hw_debug_points[n].addr == addr &&
1524              hw_debug_points[n].type == type) {
1525             return n;
1526         }
1527     }
1528 
1529     return -1;
1530 }
1531 
1532 static int find_hw_watchpoint(target_ulong addr, int *flag)
1533 {
1534     int n;
1535 
1536     n = find_hw_breakpoint(addr, GDB_WATCHPOINT_ACCESS);
1537     if (n >= 0) {
1538         *flag = BP_MEM_ACCESS;
1539         return n;
1540     }
1541 
1542     n = find_hw_breakpoint(addr, GDB_WATCHPOINT_WRITE);
1543     if (n >= 0) {
1544         *flag = BP_MEM_WRITE;
1545         return n;
1546     }
1547 
1548     n = find_hw_breakpoint(addr, GDB_WATCHPOINT_READ);
1549     if (n >= 0) {
1550         *flag = BP_MEM_READ;
1551         return n;
1552     }
1553 
1554     return -1;
1555 }
1556 
1557 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
1558                                   target_ulong len, int type)
1559 {
1560     if ((nb_hw_breakpoint + nb_hw_watchpoint) >= ARRAY_SIZE(hw_debug_points)) {
1561         return -ENOBUFS;
1562     }
1563 
1564     hw_debug_points[nb_hw_breakpoint + nb_hw_watchpoint].addr = addr;
1565     hw_debug_points[nb_hw_breakpoint + nb_hw_watchpoint].type = type;
1566 
1567     switch (type) {
1568     case GDB_BREAKPOINT_HW:
1569         if (nb_hw_breakpoint >= max_hw_breakpoint) {
1570             return -ENOBUFS;
1571         }
1572 
1573         if (find_hw_breakpoint(addr, type) >= 0) {
1574             return -EEXIST;
1575         }
1576 
1577         nb_hw_breakpoint++;
1578         break;
1579 
1580     case GDB_WATCHPOINT_WRITE:
1581     case GDB_WATCHPOINT_READ:
1582     case GDB_WATCHPOINT_ACCESS:
1583         if (nb_hw_watchpoint >= max_hw_watchpoint) {
1584             return -ENOBUFS;
1585         }
1586 
1587         if (find_hw_breakpoint(addr, type) >= 0) {
1588             return -EEXIST;
1589         }
1590 
1591         nb_hw_watchpoint++;
1592         break;
1593 
1594     default:
1595         return -ENOSYS;
1596     }
1597 
1598     return 0;
1599 }
1600 
1601 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
1602                                   target_ulong len, int type)
1603 {
1604     int n;
1605 
1606     n = find_hw_breakpoint(addr, type);
1607     if (n < 0) {
1608         return -ENOENT;
1609     }
1610 
1611     switch (type) {
1612     case GDB_BREAKPOINT_HW:
1613         nb_hw_breakpoint--;
1614         break;
1615 
1616     case GDB_WATCHPOINT_WRITE:
1617     case GDB_WATCHPOINT_READ:
1618     case GDB_WATCHPOINT_ACCESS:
1619         nb_hw_watchpoint--;
1620         break;
1621 
1622     default:
1623         return -ENOSYS;
1624     }
1625     hw_debug_points[n] = hw_debug_points[nb_hw_breakpoint + nb_hw_watchpoint];
1626 
1627     return 0;
1628 }
1629 
1630 void kvm_arch_remove_all_hw_breakpoints(void)
1631 {
1632     nb_hw_breakpoint = nb_hw_watchpoint = 0;
1633 }
1634 
1635 void kvm_arch_update_guest_debug(CPUState *cs, struct kvm_guest_debug *dbg)
1636 {
1637     int n;
1638 
1639     /* Software Breakpoint updates */
1640     if (kvm_sw_breakpoints_active(cs)) {
1641         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP;
1642     }
1643 
1644     assert((nb_hw_breakpoint + nb_hw_watchpoint)
1645            <= ARRAY_SIZE(hw_debug_points));
1646     assert((nb_hw_breakpoint + nb_hw_watchpoint) <= ARRAY_SIZE(dbg->arch.bp));
1647 
1648     if (nb_hw_breakpoint + nb_hw_watchpoint > 0) {
1649         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
1650         memset(dbg->arch.bp, 0, sizeof(dbg->arch.bp));
1651         for (n = 0; n < nb_hw_breakpoint + nb_hw_watchpoint; n++) {
1652             switch (hw_debug_points[n].type) {
1653             case GDB_BREAKPOINT_HW:
1654                 dbg->arch.bp[n].type = KVMPPC_DEBUG_BREAKPOINT;
1655                 break;
1656             case GDB_WATCHPOINT_WRITE:
1657                 dbg->arch.bp[n].type = KVMPPC_DEBUG_WATCH_WRITE;
1658                 break;
1659             case GDB_WATCHPOINT_READ:
1660                 dbg->arch.bp[n].type = KVMPPC_DEBUG_WATCH_READ;
1661                 break;
1662             case GDB_WATCHPOINT_ACCESS:
1663                 dbg->arch.bp[n].type = KVMPPC_DEBUG_WATCH_WRITE |
1664                                         KVMPPC_DEBUG_WATCH_READ;
1665                 break;
1666             default:
1667                 cpu_abort(cs, "Unsupported breakpoint type\n");
1668             }
1669             dbg->arch.bp[n].addr = hw_debug_points[n].addr;
1670         }
1671     }
1672 }
1673 
1674 static int kvm_handle_debug(PowerPCCPU *cpu, struct kvm_run *run)
1675 {
1676     CPUState *cs = CPU(cpu);
1677     CPUPPCState *env = &cpu->env;
1678     struct kvm_debug_exit_arch *arch_info = &run->debug.arch;
1679     int handle = 0;
1680     int n;
1681     int flag = 0;
1682 
1683     if (cs->singlestep_enabled) {
1684         handle = 1;
1685     } else if (arch_info->status) {
1686         if (nb_hw_breakpoint + nb_hw_watchpoint > 0) {
1687             if (arch_info->status & KVMPPC_DEBUG_BREAKPOINT) {
1688                 n = find_hw_breakpoint(arch_info->address, GDB_BREAKPOINT_HW);
1689                 if (n >= 0) {
1690                     handle = 1;
1691                 }
1692             } else if (arch_info->status & (KVMPPC_DEBUG_WATCH_READ |
1693                                             KVMPPC_DEBUG_WATCH_WRITE)) {
1694                 n = find_hw_watchpoint(arch_info->address,  &flag);
1695                 if (n >= 0) {
1696                     handle = 1;
1697                     cs->watchpoint_hit = &hw_watchpoint;
1698                     hw_watchpoint.vaddr = hw_debug_points[n].addr;
1699                     hw_watchpoint.flags = flag;
1700                 }
1701             }
1702         }
1703     } else if (kvm_find_sw_breakpoint(cs, arch_info->address)) {
1704         handle = 1;
1705     } else {
1706         /* QEMU is not able to handle debug exception, so inject
1707          * program exception to guest;
1708          * Yes program exception NOT debug exception !!
1709          * When QEMU is using debug resources then debug exception must
1710          * be always set. To achieve this we set MSR_DE and also set
1711          * MSRP_DEP so guest cannot change MSR_DE.
1712          * When emulating debug resource for guest we want guest
1713          * to control MSR_DE (enable/disable debug interrupt on need).
1714          * Supporting both configurations are NOT possible.
1715          * So the result is that we cannot share debug resources
1716          * between QEMU and Guest on BOOKE architecture.
1717          * In the current design QEMU gets the priority over guest,
1718          * this means that if QEMU is using debug resources then guest
1719          * cannot use them;
1720          * For software breakpoint QEMU uses a privileged instruction;
1721          * So there cannot be any reason that we are here for guest
1722          * set debug exception, only possibility is guest executed a
1723          * privileged / illegal instruction and that's why we are
1724          * injecting a program interrupt.
1725          */
1726 
1727         cpu_synchronize_state(cs);
1728         /* env->nip is PC, so increment this by 4 to use
1729          * ppc_cpu_do_interrupt(), which set srr0 = env->nip - 4.
1730          */
1731         env->nip += 4;
1732         cs->exception_index = POWERPC_EXCP_PROGRAM;
1733         env->error_code = POWERPC_EXCP_INVAL;
1734         ppc_cpu_do_interrupt(cs);
1735     }
1736 
1737     return handle;
1738 }
1739 
1740 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1741 {
1742     PowerPCCPU *cpu = POWERPC_CPU(cs);
1743     CPUPPCState *env = &cpu->env;
1744     int ret;
1745 
1746     qemu_mutex_lock_iothread();
1747 
1748     switch (run->exit_reason) {
1749     case KVM_EXIT_DCR:
1750         if (run->dcr.is_write) {
1751             DPRINTF("handle dcr write\n");
1752             ret = kvmppc_handle_dcr_write(env, run->dcr.dcrn, run->dcr.data);
1753         } else {
1754             DPRINTF("handle dcr read\n");
1755             ret = kvmppc_handle_dcr_read(env, run->dcr.dcrn, &run->dcr.data);
1756         }
1757         break;
1758     case KVM_EXIT_HLT:
1759         DPRINTF("handle halt\n");
1760         ret = kvmppc_handle_halt(cpu);
1761         break;
1762 #if defined(TARGET_PPC64)
1763     case KVM_EXIT_PAPR_HCALL:
1764         DPRINTF("handle PAPR hypercall\n");
1765         run->papr_hcall.ret = spapr_hypercall(cpu,
1766                                               run->papr_hcall.nr,
1767                                               run->papr_hcall.args);
1768         ret = 0;
1769         break;
1770 #endif
1771     case KVM_EXIT_EPR:
1772         DPRINTF("handle epr\n");
1773         run->epr.epr = ldl_phys(cs->as, env->mpic_iack);
1774         ret = 0;
1775         break;
1776     case KVM_EXIT_WATCHDOG:
1777         DPRINTF("handle watchdog expiry\n");
1778         watchdog_perform_action();
1779         ret = 0;
1780         break;
1781 
1782     case KVM_EXIT_DEBUG:
1783         DPRINTF("handle debug exception\n");
1784         if (kvm_handle_debug(cpu, run)) {
1785             ret = EXCP_DEBUG;
1786             break;
1787         }
1788         /* re-enter, this exception was guest-internal */
1789         ret = 0;
1790         break;
1791 
1792     default:
1793         fprintf(stderr, "KVM: unknown exit reason %d\n", run->exit_reason);
1794         ret = -1;
1795         break;
1796     }
1797 
1798     qemu_mutex_unlock_iothread();
1799     return ret;
1800 }
1801 
1802 int kvmppc_or_tsr_bits(PowerPCCPU *cpu, uint32_t tsr_bits)
1803 {
1804     CPUState *cs = CPU(cpu);
1805     uint32_t bits = tsr_bits;
1806     struct kvm_one_reg reg = {
1807         .id = KVM_REG_PPC_OR_TSR,
1808         .addr = (uintptr_t) &bits,
1809     };
1810 
1811     return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
1812 }
1813 
1814 int kvmppc_clear_tsr_bits(PowerPCCPU *cpu, uint32_t tsr_bits)
1815 {
1816 
1817     CPUState *cs = CPU(cpu);
1818     uint32_t bits = tsr_bits;
1819     struct kvm_one_reg reg = {
1820         .id = KVM_REG_PPC_CLEAR_TSR,
1821         .addr = (uintptr_t) &bits,
1822     };
1823 
1824     return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
1825 }
1826 
1827 int kvmppc_set_tcr(PowerPCCPU *cpu)
1828 {
1829     CPUState *cs = CPU(cpu);
1830     CPUPPCState *env = &cpu->env;
1831     uint32_t tcr = env->spr[SPR_BOOKE_TCR];
1832 
1833     struct kvm_one_reg reg = {
1834         .id = KVM_REG_PPC_TCR,
1835         .addr = (uintptr_t) &tcr,
1836     };
1837 
1838     return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
1839 }
1840 
1841 int kvmppc_booke_watchdog_enable(PowerPCCPU *cpu)
1842 {
1843     CPUState *cs = CPU(cpu);
1844     int ret;
1845 
1846     if (!kvm_enabled()) {
1847         return -1;
1848     }
1849 
1850     if (!cap_ppc_watchdog) {
1851         printf("warning: KVM does not support watchdog");
1852         return -1;
1853     }
1854 
1855     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_PPC_BOOKE_WATCHDOG, 0);
1856     if (ret < 0) {
1857         fprintf(stderr, "%s: couldn't enable KVM_CAP_PPC_BOOKE_WATCHDOG: %s\n",
1858                 __func__, strerror(-ret));
1859         return ret;
1860     }
1861 
1862     return ret;
1863 }
1864 
1865 static int read_cpuinfo(const char *field, char *value, int len)
1866 {
1867     FILE *f;
1868     int ret = -1;
1869     int field_len = strlen(field);
1870     char line[512];
1871 
1872     f = fopen("/proc/cpuinfo", "r");
1873     if (!f) {
1874         return -1;
1875     }
1876 
1877     do {
1878         if (!fgets(line, sizeof(line), f)) {
1879             break;
1880         }
1881         if (!strncmp(line, field, field_len)) {
1882             pstrcpy(value, len, line);
1883             ret = 0;
1884             break;
1885         }
1886     } while(*line);
1887 
1888     fclose(f);
1889 
1890     return ret;
1891 }
1892 
1893 uint32_t kvmppc_get_tbfreq(void)
1894 {
1895     char line[512];
1896     char *ns;
1897     uint32_t retval = NANOSECONDS_PER_SECOND;
1898 
1899     if (read_cpuinfo("timebase", line, sizeof(line))) {
1900         return retval;
1901     }
1902 
1903     if (!(ns = strchr(line, ':'))) {
1904         return retval;
1905     }
1906 
1907     ns++;
1908 
1909     return atoi(ns);
1910 }
1911 
1912 bool kvmppc_get_host_serial(char **value)
1913 {
1914     return g_file_get_contents("/proc/device-tree/system-id", value, NULL,
1915                                NULL);
1916 }
1917 
1918 bool kvmppc_get_host_model(char **value)
1919 {
1920     return g_file_get_contents("/proc/device-tree/model", value, NULL, NULL);
1921 }
1922 
1923 /* Try to find a device tree node for a CPU with clock-frequency property */
1924 static int kvmppc_find_cpu_dt(char *buf, int buf_len)
1925 {
1926     struct dirent *dirp;
1927     DIR *dp;
1928 
1929     if ((dp = opendir(PROC_DEVTREE_CPU)) == NULL) {
1930         printf("Can't open directory " PROC_DEVTREE_CPU "\n");
1931         return -1;
1932     }
1933 
1934     buf[0] = '\0';
1935     while ((dirp = readdir(dp)) != NULL) {
1936         FILE *f;
1937         snprintf(buf, buf_len, "%s%s/clock-frequency", PROC_DEVTREE_CPU,
1938                  dirp->d_name);
1939         f = fopen(buf, "r");
1940         if (f) {
1941             snprintf(buf, buf_len, "%s%s", PROC_DEVTREE_CPU, dirp->d_name);
1942             fclose(f);
1943             break;
1944         }
1945         buf[0] = '\0';
1946     }
1947     closedir(dp);
1948     if (buf[0] == '\0') {
1949         printf("Unknown host!\n");
1950         return -1;
1951     }
1952 
1953     return 0;
1954 }
1955 
1956 static uint64_t kvmppc_read_int_dt(const char *filename)
1957 {
1958     union {
1959         uint32_t v32;
1960         uint64_t v64;
1961     } u;
1962     FILE *f;
1963     int len;
1964 
1965     f = fopen(filename, "rb");
1966     if (!f) {
1967         return -1;
1968     }
1969 
1970     len = fread(&u, 1, sizeof(u), f);
1971     fclose(f);
1972     switch (len) {
1973     case 4:
1974         /* property is a 32-bit quantity */
1975         return be32_to_cpu(u.v32);
1976     case 8:
1977         return be64_to_cpu(u.v64);
1978     }
1979 
1980     return 0;
1981 }
1982 
1983 /* Read a CPU node property from the host device tree that's a single
1984  * integer (32-bit or 64-bit).  Returns 0 if anything goes wrong
1985  * (can't find or open the property, or doesn't understand the
1986  * format) */
1987 static uint64_t kvmppc_read_int_cpu_dt(const char *propname)
1988 {
1989     char buf[PATH_MAX], *tmp;
1990     uint64_t val;
1991 
1992     if (kvmppc_find_cpu_dt(buf, sizeof(buf))) {
1993         return -1;
1994     }
1995 
1996     tmp = g_strdup_printf("%s/%s", buf, propname);
1997     val = kvmppc_read_int_dt(tmp);
1998     g_free(tmp);
1999 
2000     return val;
2001 }
2002 
2003 uint64_t kvmppc_get_clockfreq(void)
2004 {
2005     return kvmppc_read_int_cpu_dt("clock-frequency");
2006 }
2007 
2008 static int kvmppc_get_pvinfo(CPUPPCState *env, struct kvm_ppc_pvinfo *pvinfo)
2009  {
2010      PowerPCCPU *cpu = ppc_env_get_cpu(env);
2011      CPUState *cs = CPU(cpu);
2012 
2013     if (kvm_vm_check_extension(cs->kvm_state, KVM_CAP_PPC_GET_PVINFO) &&
2014         !kvm_vm_ioctl(cs->kvm_state, KVM_PPC_GET_PVINFO, pvinfo)) {
2015         return 0;
2016     }
2017 
2018     return 1;
2019 }
2020 
2021 int kvmppc_get_hasidle(CPUPPCState *env)
2022 {
2023     struct kvm_ppc_pvinfo pvinfo;
2024 
2025     if (!kvmppc_get_pvinfo(env, &pvinfo) &&
2026         (pvinfo.flags & KVM_PPC_PVINFO_FLAGS_EV_IDLE)) {
2027         return 1;
2028     }
2029 
2030     return 0;
2031 }
2032 
2033 int kvmppc_get_hypercall(CPUPPCState *env, uint8_t *buf, int buf_len)
2034 {
2035     uint32_t *hc = (uint32_t*)buf;
2036     struct kvm_ppc_pvinfo pvinfo;
2037 
2038     if (!kvmppc_get_pvinfo(env, &pvinfo)) {
2039         memcpy(buf, pvinfo.hcall, buf_len);
2040         return 0;
2041     }
2042 
2043     /*
2044      * Fallback to always fail hypercalls regardless of endianness:
2045      *
2046      *     tdi 0,r0,72 (becomes b .+8 in wrong endian, nop in good endian)
2047      *     li r3, -1
2048      *     b .+8       (becomes nop in wrong endian)
2049      *     bswap32(li r3, -1)
2050      */
2051 
2052     hc[0] = cpu_to_be32(0x08000048);
2053     hc[1] = cpu_to_be32(0x3860ffff);
2054     hc[2] = cpu_to_be32(0x48000008);
2055     hc[3] = cpu_to_be32(bswap32(0x3860ffff));
2056 
2057     return 1;
2058 }
2059 
2060 static inline int kvmppc_enable_hcall(KVMState *s, target_ulong hcall)
2061 {
2062     return kvm_vm_enable_cap(s, KVM_CAP_PPC_ENABLE_HCALL, 0, hcall, 1);
2063 }
2064 
2065 void kvmppc_enable_logical_ci_hcalls(void)
2066 {
2067     /*
2068      * FIXME: it would be nice if we could detect the cases where
2069      * we're using a device which requires the in kernel
2070      * implementation of these hcalls, but the kernel lacks them and
2071      * produce a warning.
2072      */
2073     kvmppc_enable_hcall(kvm_state, H_LOGICAL_CI_LOAD);
2074     kvmppc_enable_hcall(kvm_state, H_LOGICAL_CI_STORE);
2075 }
2076 
2077 void kvmppc_enable_set_mode_hcall(void)
2078 {
2079     kvmppc_enable_hcall(kvm_state, H_SET_MODE);
2080 }
2081 
2082 void kvmppc_enable_clear_ref_mod_hcalls(void)
2083 {
2084     kvmppc_enable_hcall(kvm_state, H_CLEAR_REF);
2085     kvmppc_enable_hcall(kvm_state, H_CLEAR_MOD);
2086 }
2087 
2088 void kvmppc_set_papr(PowerPCCPU *cpu)
2089 {
2090     CPUState *cs = CPU(cpu);
2091     int ret;
2092 
2093     if (!kvm_enabled()) {
2094         return;
2095     }
2096 
2097     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_PPC_PAPR, 0);
2098     if (ret) {
2099         error_report("This vCPU type or KVM version does not support PAPR");
2100         exit(1);
2101     }
2102 
2103     /* Update the capability flag so we sync the right information
2104      * with kvm */
2105     cap_papr = 1;
2106 }
2107 
2108 int kvmppc_set_compat(PowerPCCPU *cpu, uint32_t compat_pvr)
2109 {
2110     return kvm_set_one_reg(CPU(cpu), KVM_REG_PPC_ARCH_COMPAT, &compat_pvr);
2111 }
2112 
2113 void kvmppc_set_mpic_proxy(PowerPCCPU *cpu, int mpic_proxy)
2114 {
2115     CPUState *cs = CPU(cpu);
2116     int ret;
2117 
2118     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_PPC_EPR, 0, mpic_proxy);
2119     if (ret && mpic_proxy) {
2120         error_report("This KVM version does not support EPR");
2121         exit(1);
2122     }
2123 }
2124 
2125 int kvmppc_smt_threads(void)
2126 {
2127     return cap_ppc_smt ? cap_ppc_smt : 1;
2128 }
2129 
2130 int kvmppc_set_smt_threads(int smt)
2131 {
2132     int ret;
2133 
2134     ret = kvm_vm_enable_cap(kvm_state, KVM_CAP_PPC_SMT, 0, smt, 0);
2135     if (!ret) {
2136         cap_ppc_smt = smt;
2137     }
2138     return ret;
2139 }
2140 
2141 void kvmppc_hint_smt_possible(Error **errp)
2142 {
2143     int i;
2144     GString *g;
2145     char *s;
2146 
2147     assert(kvm_enabled());
2148     if (cap_ppc_smt_possible) {
2149         g = g_string_new("Available VSMT modes:");
2150         for (i = 63; i >= 0; i--) {
2151             if ((1UL << i) & cap_ppc_smt_possible) {
2152                 g_string_append_printf(g, " %lu", (1UL << i));
2153             }
2154         }
2155         s = g_string_free(g, false);
2156         error_append_hint(errp, "%s.\n", s);
2157         g_free(s);
2158     } else {
2159         error_append_hint(errp,
2160                           "This KVM seems to be too old to support VSMT.\n");
2161     }
2162 }
2163 
2164 
2165 #ifdef TARGET_PPC64
2166 uint64_t kvmppc_rma_size(uint64_t current_size, unsigned int hash_shift)
2167 {
2168     struct kvm_ppc_smmu_info info;
2169     long rampagesize, best_page_shift;
2170     int i;
2171 
2172     /* Find the largest hardware supported page size that's less than
2173      * or equal to the (logical) backing page size of guest RAM */
2174     kvm_get_smmu_info(POWERPC_CPU(first_cpu), &info);
2175     rampagesize = qemu_getrampagesize();
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     /* Must set fd to -1 so we don't try to munmap when called for
2224      * destroying the table, which the upper layers -will- do
2225      */
2226     *pfd = -1;
2227     if (!cap_spapr_tce || (need_vfio && !cap_spapr_vfio)) {
2228         return NULL;
2229     }
2230 
2231     if (cap_spapr_tce_64) {
2232         struct kvm_create_spapr_tce_64 args = {
2233             .liobn = liobn,
2234             .page_shift = page_shift,
2235             .offset = bus_offset >> page_shift,
2236             .size = nb_table,
2237             .flags = 0
2238         };
2239         fd = kvm_vm_ioctl(kvm_state, KVM_CREATE_SPAPR_TCE_64, &args);
2240         if (fd < 0) {
2241             fprintf(stderr,
2242                     "KVM: Failed to create TCE64 table for liobn 0x%x\n",
2243                     liobn);
2244             return NULL;
2245         }
2246     } else if (cap_spapr_tce) {
2247         uint64_t window_size = (uint64_t) nb_table << page_shift;
2248         struct kvm_create_spapr_tce args = {
2249             .liobn = liobn,
2250             .window_size = window_size,
2251         };
2252         if ((window_size != args.window_size) || bus_offset) {
2253             return NULL;
2254         }
2255         fd = kvm_vm_ioctl(kvm_state, KVM_CREATE_SPAPR_TCE, &args);
2256         if (fd < 0) {
2257             fprintf(stderr, "KVM: Failed to create TCE table for liobn 0x%x\n",
2258                     liobn);
2259             return NULL;
2260         }
2261     } else {
2262         return NULL;
2263     }
2264 
2265     len = nb_table * sizeof(uint64_t);
2266     /* FIXME: round this up to page size */
2267 
2268     table = mmap(NULL, len, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
2269     if (table == MAP_FAILED) {
2270         fprintf(stderr, "KVM: Failed to map TCE table for liobn 0x%x\n",
2271                 liobn);
2272         close(fd);
2273         return NULL;
2274     }
2275 
2276     *pfd = fd;
2277     return table;
2278 }
2279 
2280 int kvmppc_remove_spapr_tce(void *table, int fd, uint32_t nb_table)
2281 {
2282     long len;
2283 
2284     if (fd < 0) {
2285         return -1;
2286     }
2287 
2288     len = nb_table * sizeof(uint64_t);
2289     if ((munmap(table, len) < 0) ||
2290         (close(fd) < 0)) {
2291         fprintf(stderr, "KVM: Unexpected error removing TCE table: %s",
2292                 strerror(errno));
2293         /* Leak the table */
2294     }
2295 
2296     return 0;
2297 }
2298 
2299 int kvmppc_reset_htab(int shift_hint)
2300 {
2301     uint32_t shift = shift_hint;
2302 
2303     if (!kvm_enabled()) {
2304         /* Full emulation, tell caller to allocate htab itself */
2305         return 0;
2306     }
2307     if (kvm_vm_check_extension(kvm_state, KVM_CAP_PPC_ALLOC_HTAB)) {
2308         int ret;
2309         ret = kvm_vm_ioctl(kvm_state, KVM_PPC_ALLOCATE_HTAB, &shift);
2310         if (ret == -ENOTTY) {
2311             /* At least some versions of PR KVM advertise the
2312              * capability, but don't implement the ioctl().  Oops.
2313              * Return 0 so that we allocate the htab in qemu, as is
2314              * correct for PR. */
2315             return 0;
2316         } else if (ret < 0) {
2317             return ret;
2318         }
2319         return shift;
2320     }
2321 
2322     /* We have a kernel that predates the htab reset calls.  For PR
2323      * KVM, we need to allocate the htab ourselves, for an HV KVM of
2324      * this era, it has allocated a 16MB fixed size hash table already. */
2325     if (kvmppc_is_pr(kvm_state)) {
2326         /* PR - tell caller to allocate htab */
2327         return 0;
2328     } else {
2329         /* HV - assume 16MB kernel allocated htab */
2330         return 24;
2331     }
2332 }
2333 
2334 static inline uint32_t mfpvr(void)
2335 {
2336     uint32_t pvr;
2337 
2338     asm ("mfpvr %0"
2339          : "=r"(pvr));
2340     return pvr;
2341 }
2342 
2343 static void alter_insns(uint64_t *word, uint64_t flags, bool on)
2344 {
2345     if (on) {
2346         *word |= flags;
2347     } else {
2348         *word &= ~flags;
2349     }
2350 }
2351 
2352 static void kvmppc_host_cpu_class_init(ObjectClass *oc, void *data)
2353 {
2354     PowerPCCPUClass *pcc = POWERPC_CPU_CLASS(oc);
2355     uint32_t dcache_size = kvmppc_read_int_cpu_dt("d-cache-size");
2356     uint32_t icache_size = kvmppc_read_int_cpu_dt("i-cache-size");
2357 
2358     /* Now fix up the class with information we can query from the host */
2359     pcc->pvr = mfpvr();
2360 
2361     alter_insns(&pcc->insns_flags, PPC_ALTIVEC,
2362                 qemu_getauxval(AT_HWCAP) & PPC_FEATURE_HAS_ALTIVEC);
2363     alter_insns(&pcc->insns_flags2, PPC2_VSX,
2364                 qemu_getauxval(AT_HWCAP) & PPC_FEATURE_HAS_VSX);
2365     alter_insns(&pcc->insns_flags2, PPC2_DFP,
2366                 qemu_getauxval(AT_HWCAP) & PPC_FEATURE_HAS_DFP);
2367 
2368     if (dcache_size != -1) {
2369         pcc->l1_dcache_size = dcache_size;
2370     }
2371 
2372     if (icache_size != -1) {
2373         pcc->l1_icache_size = icache_size;
2374     }
2375 
2376 #if defined(TARGET_PPC64)
2377     pcc->radix_page_info = kvm_get_radix_page_info();
2378 
2379     if ((pcc->pvr & 0xffffff00) == CPU_POWERPC_POWER9_DD1) {
2380         /*
2381          * POWER9 DD1 has some bugs which make it not really ISA 3.00
2382          * compliant.  More importantly, advertising ISA 3.00
2383          * architected mode may prevent guests from activating
2384          * necessary DD1 workarounds.
2385          */
2386         pcc->pcr_supported &= ~(PCR_COMPAT_3_00 | PCR_COMPAT_2_07
2387                                 | PCR_COMPAT_2_06 | PCR_COMPAT_2_05);
2388     }
2389 #endif /* defined(TARGET_PPC64) */
2390 }
2391 
2392 bool kvmppc_has_cap_epr(void)
2393 {
2394     return cap_epr;
2395 }
2396 
2397 bool kvmppc_has_cap_fixup_hcalls(void)
2398 {
2399     return cap_fixup_hcalls;
2400 }
2401 
2402 bool kvmppc_has_cap_htm(void)
2403 {
2404     return cap_htm;
2405 }
2406 
2407 bool kvmppc_has_cap_mmu_radix(void)
2408 {
2409     return cap_mmu_radix;
2410 }
2411 
2412 bool kvmppc_has_cap_mmu_hash_v3(void)
2413 {
2414     return cap_mmu_hash_v3;
2415 }
2416 
2417 static bool kvmppc_power8_host(void)
2418 {
2419     bool ret = false;
2420 #ifdef TARGET_PPC64
2421     {
2422         uint32_t base_pvr = CPU_POWERPC_POWER_SERVER_MASK & mfpvr();
2423         ret = (base_pvr == CPU_POWERPC_POWER8E_BASE) ||
2424               (base_pvr == CPU_POWERPC_POWER8NVL_BASE) ||
2425               (base_pvr == CPU_POWERPC_POWER8_BASE);
2426     }
2427 #endif /* TARGET_PPC64 */
2428     return ret;
2429 }
2430 
2431 static int parse_cap_ppc_safe_cache(struct kvm_ppc_cpu_char c)
2432 {
2433     bool l1d_thread_priv_req = !kvmppc_power8_host();
2434 
2435     if (~c.behaviour & c.behaviour_mask & H_CPU_BEHAV_L1D_FLUSH_PR) {
2436         return 2;
2437     } else if ((!l1d_thread_priv_req ||
2438                 c.character & c.character_mask & H_CPU_CHAR_L1D_THREAD_PRIV) &&
2439                (c.character & c.character_mask
2440                 & (H_CPU_CHAR_L1D_FLUSH_ORI30 | H_CPU_CHAR_L1D_FLUSH_TRIG2))) {
2441         return 1;
2442     }
2443 
2444     return 0;
2445 }
2446 
2447 static int parse_cap_ppc_safe_bounds_check(struct kvm_ppc_cpu_char c)
2448 {
2449     if (~c.behaviour & c.behaviour_mask & H_CPU_BEHAV_BNDS_CHK_SPEC_BAR) {
2450         return 2;
2451     } else if (c.character & c.character_mask & H_CPU_CHAR_SPEC_BAR_ORI31) {
2452         return 1;
2453     }
2454 
2455     return 0;
2456 }
2457 
2458 static int parse_cap_ppc_safe_indirect_branch(struct kvm_ppc_cpu_char c)
2459 {
2460     if (c.character & c.character_mask & H_CPU_CHAR_CACHE_COUNT_DIS) {
2461         return  SPAPR_CAP_FIXED_CCD;
2462     } else if (c.character & c.character_mask & H_CPU_CHAR_BCCTRL_SERIALISED) {
2463         return SPAPR_CAP_FIXED_IBS;
2464     }
2465 
2466     return 0;
2467 }
2468 
2469 static void kvmppc_get_cpu_characteristics(KVMState *s)
2470 {
2471     struct kvm_ppc_cpu_char c;
2472     int ret;
2473 
2474     /* Assume broken */
2475     cap_ppc_safe_cache = 0;
2476     cap_ppc_safe_bounds_check = 0;
2477     cap_ppc_safe_indirect_branch = 0;
2478 
2479     ret = kvm_vm_check_extension(s, KVM_CAP_PPC_GET_CPU_CHAR);
2480     if (!ret) {
2481         return;
2482     }
2483     ret = kvm_vm_ioctl(s, KVM_PPC_GET_CPU_CHAR, &c);
2484     if (ret < 0) {
2485         return;
2486     }
2487 
2488     cap_ppc_safe_cache = parse_cap_ppc_safe_cache(c);
2489     cap_ppc_safe_bounds_check = parse_cap_ppc_safe_bounds_check(c);
2490     cap_ppc_safe_indirect_branch = parse_cap_ppc_safe_indirect_branch(c);
2491 }
2492 
2493 int kvmppc_get_cap_safe_cache(void)
2494 {
2495     return cap_ppc_safe_cache;
2496 }
2497 
2498 int kvmppc_get_cap_safe_bounds_check(void)
2499 {
2500     return cap_ppc_safe_bounds_check;
2501 }
2502 
2503 int kvmppc_get_cap_safe_indirect_branch(void)
2504 {
2505     return cap_ppc_safe_indirect_branch;
2506 }
2507 
2508 bool kvmppc_has_cap_spapr_vfio(void)
2509 {
2510     return cap_spapr_vfio;
2511 }
2512 
2513 PowerPCCPUClass *kvm_ppc_get_host_cpu_class(void)
2514 {
2515     uint32_t host_pvr = mfpvr();
2516     PowerPCCPUClass *pvr_pcc;
2517 
2518     pvr_pcc = ppc_cpu_class_by_pvr(host_pvr);
2519     if (pvr_pcc == NULL) {
2520         pvr_pcc = ppc_cpu_class_by_pvr_mask(host_pvr);
2521     }
2522 
2523     return pvr_pcc;
2524 }
2525 
2526 static int kvm_ppc_register_host_cpu_type(MachineState *ms)
2527 {
2528     TypeInfo type_info = {
2529         .name = TYPE_HOST_POWERPC_CPU,
2530         .class_init = kvmppc_host_cpu_class_init,
2531     };
2532     MachineClass *mc = MACHINE_GET_CLASS(ms);
2533     PowerPCCPUClass *pvr_pcc;
2534     ObjectClass *oc;
2535     DeviceClass *dc;
2536     int i;
2537 
2538     pvr_pcc = kvm_ppc_get_host_cpu_class();
2539     if (pvr_pcc == NULL) {
2540         return -1;
2541     }
2542     type_info.parent = object_class_get_name(OBJECT_CLASS(pvr_pcc));
2543     type_register(&type_info);
2544     if (object_dynamic_cast(OBJECT(ms), TYPE_SPAPR_MACHINE)) {
2545         /* override TCG default cpu type with 'host' cpu model */
2546         mc->default_cpu_type = TYPE_HOST_POWERPC_CPU;
2547     }
2548 
2549     oc = object_class_by_name(type_info.name);
2550     g_assert(oc);
2551 
2552     /*
2553      * Update generic CPU family class alias (e.g. on a POWER8NVL host,
2554      * we want "POWER8" to be a "family" alias that points to the current
2555      * host CPU type, too)
2556      */
2557     dc = DEVICE_CLASS(ppc_cpu_get_family_class(pvr_pcc));
2558     for (i = 0; ppc_cpu_aliases[i].alias != NULL; i++) {
2559         if (strcasecmp(ppc_cpu_aliases[i].alias, dc->desc) == 0) {
2560             char *suffix;
2561 
2562             ppc_cpu_aliases[i].model = g_strdup(object_class_get_name(oc));
2563             suffix = strstr(ppc_cpu_aliases[i].model, POWERPC_CPU_TYPE_SUFFIX);
2564             if (suffix) {
2565                 *suffix = 0;
2566             }
2567             break;
2568         }
2569     }
2570 
2571     return 0;
2572 }
2573 
2574 int kvmppc_define_rtas_kernel_token(uint32_t token, const char *function)
2575 {
2576     struct kvm_rtas_token_args args = {
2577         .token = token,
2578     };
2579 
2580     if (!kvm_check_extension(kvm_state, KVM_CAP_PPC_RTAS)) {
2581         return -ENOENT;
2582     }
2583 
2584     strncpy(args.name, function, sizeof(args.name));
2585 
2586     return kvm_vm_ioctl(kvm_state, KVM_PPC_RTAS_DEFINE_TOKEN, &args);
2587 }
2588 
2589 int kvmppc_get_htab_fd(bool write, uint64_t index, Error **errp)
2590 {
2591     struct kvm_get_htab_fd s = {
2592         .flags = write ? KVM_GET_HTAB_WRITE : 0,
2593         .start_index = index,
2594     };
2595     int ret;
2596 
2597     if (!cap_htab_fd) {
2598         error_setg(errp, "KVM version doesn't support %s the HPT",
2599                    write ? "writing" : "reading");
2600         return -ENOTSUP;
2601     }
2602 
2603     ret = kvm_vm_ioctl(kvm_state, KVM_PPC_GET_HTAB_FD, &s);
2604     if (ret < 0) {
2605         error_setg(errp, "Unable to open fd for %s HPT %s KVM: %s",
2606                    write ? "writing" : "reading", write ? "to" : "from",
2607                    strerror(errno));
2608         return -errno;
2609     }
2610 
2611     return ret;
2612 }
2613 
2614 int kvmppc_save_htab(QEMUFile *f, int fd, size_t bufsize, int64_t max_ns)
2615 {
2616     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2617     uint8_t buf[bufsize];
2618     ssize_t rc;
2619 
2620     do {
2621         rc = read(fd, buf, bufsize);
2622         if (rc < 0) {
2623             fprintf(stderr, "Error reading data from KVM HTAB fd: %s\n",
2624                     strerror(errno));
2625             return rc;
2626         } else if (rc) {
2627             uint8_t *buffer = buf;
2628             ssize_t n = rc;
2629             while (n) {
2630                 struct kvm_get_htab_header *head =
2631                     (struct kvm_get_htab_header *) buffer;
2632                 size_t chunksize = sizeof(*head) +
2633                      HASH_PTE_SIZE_64 * head->n_valid;
2634 
2635                 qemu_put_be32(f, head->index);
2636                 qemu_put_be16(f, head->n_valid);
2637                 qemu_put_be16(f, head->n_invalid);
2638                 qemu_put_buffer(f, (void *)(head + 1),
2639                                 HASH_PTE_SIZE_64 * head->n_valid);
2640 
2641                 buffer += chunksize;
2642                 n -= chunksize;
2643             }
2644         }
2645     } while ((rc != 0)
2646              && ((max_ns < 0)
2647                  || ((qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) < max_ns)));
2648 
2649     return (rc == 0) ? 1 : 0;
2650 }
2651 
2652 int kvmppc_load_htab_chunk(QEMUFile *f, int fd, uint32_t index,
2653                            uint16_t n_valid, uint16_t n_invalid)
2654 {
2655     struct kvm_get_htab_header *buf;
2656     size_t chunksize = sizeof(*buf) + n_valid*HASH_PTE_SIZE_64;
2657     ssize_t rc;
2658 
2659     buf = alloca(chunksize);
2660     buf->index = index;
2661     buf->n_valid = n_valid;
2662     buf->n_invalid = n_invalid;
2663 
2664     qemu_get_buffer(f, (void *)(buf + 1), HASH_PTE_SIZE_64*n_valid);
2665 
2666     rc = write(fd, buf, chunksize);
2667     if (rc < 0) {
2668         fprintf(stderr, "Error writing KVM hash table: %s\n",
2669                 strerror(errno));
2670         return rc;
2671     }
2672     if (rc != chunksize) {
2673         /* We should never get a short write on a single chunk */
2674         fprintf(stderr, "Short write, restoring KVM hash table\n");
2675         return -1;
2676     }
2677     return 0;
2678 }
2679 
2680 bool kvm_arch_stop_on_emulation_error(CPUState *cpu)
2681 {
2682     return true;
2683 }
2684 
2685 void kvm_arch_init_irq_routing(KVMState *s)
2686 {
2687 }
2688 
2689 void kvmppc_read_hptes(ppc_hash_pte64_t *hptes, hwaddr ptex, int n)
2690 {
2691     int fd, rc;
2692     int i;
2693 
2694     fd = kvmppc_get_htab_fd(false, ptex, &error_abort);
2695 
2696     i = 0;
2697     while (i < n) {
2698         struct kvm_get_htab_header *hdr;
2699         int m = n < HPTES_PER_GROUP ? n : HPTES_PER_GROUP;
2700         char buf[sizeof(*hdr) + m * HASH_PTE_SIZE_64];
2701 
2702         rc = read(fd, buf, sizeof(buf));
2703         if (rc < 0) {
2704             hw_error("kvmppc_read_hptes: Unable to read HPTEs");
2705         }
2706 
2707         hdr = (struct kvm_get_htab_header *)buf;
2708         while ((i < n) && ((char *)hdr < (buf + rc))) {
2709             int invalid = hdr->n_invalid, valid = hdr->n_valid;
2710 
2711             if (hdr->index != (ptex + i)) {
2712                 hw_error("kvmppc_read_hptes: Unexpected HPTE index %"PRIu32
2713                          " != (%"HWADDR_PRIu" + %d", hdr->index, ptex, i);
2714             }
2715 
2716             if (n - i < valid) {
2717                 valid = n - i;
2718             }
2719             memcpy(hptes + i, hdr + 1, HASH_PTE_SIZE_64 * valid);
2720             i += valid;
2721 
2722             if ((n - i) < invalid) {
2723                 invalid = n - i;
2724             }
2725             memset(hptes + i, 0, invalid * HASH_PTE_SIZE_64);
2726             i += invalid;
2727 
2728             hdr = (struct kvm_get_htab_header *)
2729                 ((char *)(hdr + 1) + HASH_PTE_SIZE_64 * hdr->n_valid);
2730         }
2731     }
2732 
2733     close(fd);
2734 }
2735 
2736 void kvmppc_write_hpte(hwaddr ptex, uint64_t pte0, uint64_t pte1)
2737 {
2738     int fd, rc;
2739     struct {
2740         struct kvm_get_htab_header hdr;
2741         uint64_t pte0;
2742         uint64_t pte1;
2743     } buf;
2744 
2745     fd = kvmppc_get_htab_fd(true, 0 /* Ignored */, &error_abort);
2746 
2747     buf.hdr.n_valid = 1;
2748     buf.hdr.n_invalid = 0;
2749     buf.hdr.index = ptex;
2750     buf.pte0 = cpu_to_be64(pte0);
2751     buf.pte1 = cpu_to_be64(pte1);
2752 
2753     rc = write(fd, &buf, sizeof(buf));
2754     if (rc != sizeof(buf)) {
2755         hw_error("kvmppc_write_hpte: Unable to update KVM HPT");
2756     }
2757     close(fd);
2758 }
2759 
2760 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
2761                              uint64_t address, uint32_t data, PCIDevice *dev)
2762 {
2763     return 0;
2764 }
2765 
2766 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
2767                                 int vector, PCIDevice *dev)
2768 {
2769     return 0;
2770 }
2771 
2772 int kvm_arch_release_virq_post(int virq)
2773 {
2774     return 0;
2775 }
2776 
2777 int kvm_arch_msi_data_to_gsi(uint32_t data)
2778 {
2779     return data & 0xffff;
2780 }
2781 
2782 int kvmppc_enable_hwrng(void)
2783 {
2784     if (!kvm_enabled() || !kvm_check_extension(kvm_state, KVM_CAP_PPC_HWRNG)) {
2785         return -1;
2786     }
2787 
2788     return kvmppc_enable_hcall(kvm_state, H_RANDOM);
2789 }
2790 
2791 void kvmppc_check_papr_resize_hpt(Error **errp)
2792 {
2793     if (!kvm_enabled()) {
2794         return; /* No KVM, we're good */
2795     }
2796 
2797     if (cap_resize_hpt) {
2798         return; /* Kernel has explicit support, we're good */
2799     }
2800 
2801     /* Otherwise fallback on looking for PR KVM */
2802     if (kvmppc_is_pr(kvm_state)) {
2803         return;
2804     }
2805 
2806     error_setg(errp,
2807                "Hash page table resizing not available with this KVM version");
2808 }
2809 
2810 int kvmppc_resize_hpt_prepare(PowerPCCPU *cpu, target_ulong flags, int shift)
2811 {
2812     CPUState *cs = CPU(cpu);
2813     struct kvm_ppc_resize_hpt rhpt = {
2814         .flags = flags,
2815         .shift = shift,
2816     };
2817 
2818     if (!cap_resize_hpt) {
2819         return -ENOSYS;
2820     }
2821 
2822     return kvm_vm_ioctl(cs->kvm_state, KVM_PPC_RESIZE_HPT_PREPARE, &rhpt);
2823 }
2824 
2825 int kvmppc_resize_hpt_commit(PowerPCCPU *cpu, target_ulong flags, int shift)
2826 {
2827     CPUState *cs = CPU(cpu);
2828     struct kvm_ppc_resize_hpt rhpt = {
2829         .flags = flags,
2830         .shift = shift,
2831     };
2832 
2833     if (!cap_resize_hpt) {
2834         return -ENOSYS;
2835     }
2836 
2837     return kvm_vm_ioctl(cs->kvm_state, KVM_PPC_RESIZE_HPT_COMMIT, &rhpt);
2838 }
2839 
2840 /*
2841  * This is a helper function to detect a post migration scenario
2842  * in which a guest, running as KVM-HV, freezes in cpu_post_load because
2843  * the guest kernel can't handle a PVR value other than the actual host
2844  * PVR in KVM_SET_SREGS, even if pvr_match() returns true.
2845  *
2846  * If we don't have cap_ppc_pvr_compat and we're not running in PR
2847  * (so, we're HV), return true. The workaround itself is done in
2848  * cpu_post_load.
2849  *
2850  * The order here is important: we'll only check for KVM PR as a
2851  * fallback if the guest kernel can't handle the situation itself.
2852  * We need to avoid as much as possible querying the running KVM type
2853  * in QEMU level.
2854  */
2855 bool kvmppc_pvr_workaround_required(PowerPCCPU *cpu)
2856 {
2857     CPUState *cs = CPU(cpu);
2858 
2859     if (!kvm_enabled()) {
2860         return false;
2861     }
2862 
2863     if (cap_ppc_pvr_compat) {
2864         return false;
2865     }
2866 
2867     return !kvmppc_is_pr(cs->kvm_state);
2868 }
2869