2 * PowerPC implementation of KVM hooks
4 * Copyright IBM Corp. 2007
5 * Copyright (C) 2011 Freescale Semiconductor, Inc.
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.
17 #include "qemu/osdep.h"
19 #include <sys/ioctl.h>
22 #include <linux/kvm.h>
24 #include "qemu-common.h"
25 #include "qapi/error.h"
26 #include "qemu/error-report.h"
28 #include "cpu-models.h"
29 #include "qemu/timer.h"
30 #include "sysemu/sysemu.h"
31 #include "sysemu/hw_accel.h"
33 #include "sysemu/cpus.h"
34 #include "sysemu/device_tree.h"
35 #include "mmu-hash64.h"
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"
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"
51 #include "sysemu/kvm_int.h"
56 #define DPRINTF(fmt, ...) \
57 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
59 #define DPRINTF(fmt, ...) \
63 #define PROC_DEVTREE_CPU "/proc/device-tree/cpus/"
65 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
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_ppc_rma;
76 static int cap_spapr_tce;
77 static int cap_spapr_tce_64;
78 static int cap_spapr_multitce;
79 static int cap_spapr_vfio;
81 static int cap_one_reg;
83 static int cap_ppc_watchdog;
85 static int cap_htab_fd;
86 static int cap_fixup_hcalls;
87 static int cap_htm; /* Hardware transactional memory support */
88 static int cap_mmu_radix;
89 static int cap_mmu_hash_v3;
90 static int cap_resize_hpt;
91 static int cap_ppc_pvr_compat;
92 static int cap_ppc_safe_cache;
93 static int cap_ppc_safe_bounds_check;
94 static int cap_ppc_safe_indirect_branch;
96 static uint32_t debug_inst_opcode;
98 /* XXX We have a race condition where we actually have a level triggered
99 * interrupt, but the infrastructure can't expose that yet, so the guest
100 * takes but ignores it, goes to sleep and never gets notified that there's
101 * still an interrupt pending.
103 * As a quick workaround, let's just wake up again 20 ms after we injected
104 * an interrupt. That way we can assure that we're always reinjecting
105 * interrupts in case the guest swallowed them.
107 static QEMUTimer *idle_timer;
109 static void kvm_kick_cpu(void *opaque)
111 PowerPCCPU *cpu = opaque;
113 qemu_cpu_kick(CPU(cpu));
116 /* Check whether we are running with KVM-PR (instead of KVM-HV). This
117 * should only be used for fallback tests - generally we should use
118 * explicit capabilities for the features we want, rather than
119 * assuming what is/isn't available depending on the KVM variant. */
120 static bool kvmppc_is_pr(KVMState *ks)
122 /* Assume KVM-PR if the GET_PVINFO capability is available */
123 return kvm_vm_check_extension(ks, KVM_CAP_PPC_GET_PVINFO) != 0;
126 static int kvm_ppc_register_host_cpu_type(MachineState *ms);
127 static void kvmppc_get_cpu_characteristics(KVMState *s);
129 int kvm_arch_init(MachineState *ms, KVMState *s)
131 cap_interrupt_unset = kvm_check_extension(s, KVM_CAP_PPC_UNSET_IRQ);
132 cap_interrupt_level = kvm_check_extension(s, KVM_CAP_PPC_IRQ_LEVEL);
133 cap_segstate = kvm_check_extension(s, KVM_CAP_PPC_SEGSTATE);
134 cap_booke_sregs = kvm_check_extension(s, KVM_CAP_PPC_BOOKE_SREGS);
135 cap_ppc_smt_possible = kvm_vm_check_extension(s, KVM_CAP_PPC_SMT_POSSIBLE);
136 cap_ppc_rma = kvm_check_extension(s, KVM_CAP_PPC_RMA);
137 cap_spapr_tce = kvm_check_extension(s, KVM_CAP_SPAPR_TCE);
138 cap_spapr_tce_64 = kvm_check_extension(s, KVM_CAP_SPAPR_TCE_64);
139 cap_spapr_multitce = kvm_check_extension(s, KVM_CAP_SPAPR_MULTITCE);
140 cap_spapr_vfio = kvm_vm_check_extension(s, KVM_CAP_SPAPR_TCE_VFIO);
141 cap_one_reg = kvm_check_extension(s, KVM_CAP_ONE_REG);
142 cap_hior = kvm_check_extension(s, KVM_CAP_PPC_HIOR);
143 cap_epr = kvm_check_extension(s, KVM_CAP_PPC_EPR);
144 cap_ppc_watchdog = kvm_check_extension(s, KVM_CAP_PPC_BOOKE_WATCHDOG);
145 /* Note: we don't set cap_papr here, because this capability is
146 * only activated after this by kvmppc_set_papr() */
147 cap_htab_fd = kvm_vm_check_extension(s, KVM_CAP_PPC_HTAB_FD);
148 cap_fixup_hcalls = kvm_check_extension(s, KVM_CAP_PPC_FIXUP_HCALL);
149 cap_ppc_smt = kvm_vm_check_extension(s, KVM_CAP_PPC_SMT);
150 cap_htm = kvm_vm_check_extension(s, KVM_CAP_PPC_HTM);
151 cap_mmu_radix = kvm_vm_check_extension(s, KVM_CAP_PPC_MMU_RADIX);
152 cap_mmu_hash_v3 = kvm_vm_check_extension(s, KVM_CAP_PPC_MMU_HASH_V3);
153 cap_resize_hpt = kvm_vm_check_extension(s, KVM_CAP_SPAPR_RESIZE_HPT);
154 kvmppc_get_cpu_characteristics(s);
156 * Note: setting it to false because there is not such capability
157 * in KVM at this moment.
159 * TODO: call kvm_vm_check_extension() with the right capability
160 * after the kernel starts implementing it.*/
161 cap_ppc_pvr_compat = false;
163 if (!cap_interrupt_level) {
164 fprintf(stderr, "KVM: Couldn't find level irq capability. Expect the "
165 "VM to stall at times!\n");
168 kvm_ppc_register_host_cpu_type(ms);
173 int kvm_arch_irqchip_create(MachineState *ms, KVMState *s)
178 static int kvm_arch_sync_sregs(PowerPCCPU *cpu)
180 CPUPPCState *cenv = &cpu->env;
181 CPUState *cs = CPU(cpu);
182 struct kvm_sregs sregs;
185 if (cenv->excp_model == POWERPC_EXCP_BOOKE) {
186 /* What we're really trying to say is "if we're on BookE, we use
187 the native PVR for now". This is the only sane way to check
188 it though, so we potentially confuse users that they can run
189 BookE guests on BookS. Let's hope nobody dares enough :) */
193 fprintf(stderr, "kvm error: missing PVR setting capability\n");
198 ret = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs);
203 sregs.pvr = cenv->spr[SPR_PVR];
204 return kvm_vcpu_ioctl(cs, KVM_SET_SREGS, &sregs);
207 /* Set up a shared TLB array with KVM */
208 static int kvm_booke206_tlb_init(PowerPCCPU *cpu)
210 CPUPPCState *env = &cpu->env;
211 CPUState *cs = CPU(cpu);
212 struct kvm_book3e_206_tlb_params params = {};
213 struct kvm_config_tlb cfg = {};
214 unsigned int entries = 0;
217 if (!kvm_enabled() ||
218 !kvm_check_extension(cs->kvm_state, KVM_CAP_SW_TLB)) {
222 assert(ARRAY_SIZE(params.tlb_sizes) == BOOKE206_MAX_TLBN);
224 for (i = 0; i < BOOKE206_MAX_TLBN; i++) {
225 params.tlb_sizes[i] = booke206_tlb_size(env, i);
226 params.tlb_ways[i] = booke206_tlb_ways(env, i);
227 entries += params.tlb_sizes[i];
230 assert(entries == env->nb_tlb);
231 assert(sizeof(struct kvm_book3e_206_tlb_entry) == sizeof(ppcmas_tlb_t));
233 env->tlb_dirty = true;
235 cfg.array = (uintptr_t)env->tlb.tlbm;
236 cfg.array_len = sizeof(ppcmas_tlb_t) * entries;
237 cfg.params = (uintptr_t)¶ms;
238 cfg.mmu_type = KVM_MMU_FSL_BOOKE_NOHV;
240 ret = kvm_vcpu_enable_cap(cs, KVM_CAP_SW_TLB, 0, (uintptr_t)&cfg);
242 fprintf(stderr, "%s: couldn't enable KVM_CAP_SW_TLB: %s\n",
243 __func__, strerror(-ret));
247 env->kvm_sw_tlb = true;
252 #if defined(TARGET_PPC64)
253 static void kvm_get_fallback_smmu_info(PowerPCCPU *cpu,
254 struct kvm_ppc_smmu_info *info)
256 CPUPPCState *env = &cpu->env;
257 CPUState *cs = CPU(cpu);
259 memset(info, 0, sizeof(*info));
261 /* We don't have the new KVM_PPC_GET_SMMU_INFO ioctl, so
262 * need to "guess" what the supported page sizes are.
264 * For that to work we make a few assumptions:
266 * - Check whether we are running "PR" KVM which only supports 4K
267 * and 16M pages, but supports them regardless of the backing
268 * store characteritics. We also don't support 1T segments.
270 * This is safe as if HV KVM ever supports that capability or PR
271 * KVM grows supports for more page/segment sizes, those versions
272 * will have implemented KVM_CAP_PPC_GET_SMMU_INFO and thus we
273 * will not hit this fallback
275 * - Else we are running HV KVM. This means we only support page
276 * sizes that fit in the backing store. Additionally we only
277 * advertize 64K pages if the processor is ARCH 2.06 and we assume
278 * P7 encodings for the SLB and hash table. Here too, we assume
279 * support for any newer processor will mean a kernel that
280 * implements KVM_CAP_PPC_GET_SMMU_INFO and thus doesn't hit
283 if (kvmppc_is_pr(cs->kvm_state)) {
288 /* Standard 4k base page size segment */
289 info->sps[0].page_shift = 12;
290 info->sps[0].slb_enc = 0;
291 info->sps[0].enc[0].page_shift = 12;
292 info->sps[0].enc[0].pte_enc = 0;
294 /* Standard 16M large page size segment */
295 info->sps[1].page_shift = 24;
296 info->sps[1].slb_enc = SLB_VSID_L;
297 info->sps[1].enc[0].page_shift = 24;
298 info->sps[1].enc[0].pte_enc = 0;
302 /* HV KVM has backing store size restrictions */
303 info->flags = KVM_PPC_PAGE_SIZES_REAL;
305 if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
306 info->flags |= KVM_PPC_1T_SEGMENTS;
309 if (env->mmu_model == POWERPC_MMU_2_06 ||
310 env->mmu_model == POWERPC_MMU_2_07) {
316 /* Standard 4k base page size segment */
317 info->sps[i].page_shift = 12;
318 info->sps[i].slb_enc = 0;
319 info->sps[i].enc[0].page_shift = 12;
320 info->sps[i].enc[0].pte_enc = 0;
323 /* 64K on MMU 2.06 and later */
324 if (env->mmu_model == POWERPC_MMU_2_06 ||
325 env->mmu_model == POWERPC_MMU_2_07) {
326 info->sps[i].page_shift = 16;
327 info->sps[i].slb_enc = 0x110;
328 info->sps[i].enc[0].page_shift = 16;
329 info->sps[i].enc[0].pte_enc = 1;
333 /* Standard 16M large page size segment */
334 info->sps[i].page_shift = 24;
335 info->sps[i].slb_enc = SLB_VSID_L;
336 info->sps[i].enc[0].page_shift = 24;
337 info->sps[i].enc[0].pte_enc = 0;
341 static void kvm_get_smmu_info(PowerPCCPU *cpu, struct kvm_ppc_smmu_info *info)
343 CPUState *cs = CPU(cpu);
346 if (kvm_check_extension(cs->kvm_state, KVM_CAP_PPC_GET_SMMU_INFO)) {
347 ret = kvm_vm_ioctl(cs->kvm_state, KVM_PPC_GET_SMMU_INFO, info);
353 kvm_get_fallback_smmu_info(cpu, info);
356 struct ppc_radix_page_info *kvm_get_radix_page_info(void)
358 KVMState *s = KVM_STATE(current_machine->accelerator);
359 struct ppc_radix_page_info *radix_page_info;
360 struct kvm_ppc_rmmu_info rmmu_info;
363 if (!kvm_check_extension(s, KVM_CAP_PPC_MMU_RADIX)) {
366 if (kvm_vm_ioctl(s, KVM_PPC_GET_RMMU_INFO, &rmmu_info)) {
369 radix_page_info = g_malloc0(sizeof(*radix_page_info));
370 radix_page_info->count = 0;
371 for (i = 0; i < PPC_PAGE_SIZES_MAX_SZ; i++) {
372 if (rmmu_info.ap_encodings[i]) {
373 radix_page_info->entries[i] = rmmu_info.ap_encodings[i];
374 radix_page_info->count++;
377 return radix_page_info;
380 target_ulong kvmppc_configure_v3_mmu(PowerPCCPU *cpu,
381 bool radix, bool gtse,
384 CPUState *cs = CPU(cpu);
387 struct kvm_ppc_mmuv3_cfg cfg = {
388 .process_table = proc_tbl,
392 flags |= KVM_PPC_MMUV3_RADIX;
395 flags |= KVM_PPC_MMUV3_GTSE;
398 ret = kvm_vm_ioctl(cs->kvm_state, KVM_PPC_CONFIGURE_V3_MMU, &cfg);
405 return H_NOT_AVAILABLE;
411 static bool kvm_valid_page_size(uint32_t flags, long rampgsize, uint32_t shift)
413 if (!(flags & KVM_PPC_PAGE_SIZES_REAL)) {
417 return (1ul << shift) <= rampgsize;
420 static long max_cpu_page_size;
422 static void kvm_fixup_page_sizes(PowerPCCPU *cpu)
424 static struct kvm_ppc_smmu_info smmu_info;
425 static bool has_smmu_info;
426 CPUPPCState *env = &cpu->env;
429 /* We only handle page sizes for 64-bit server guests for now */
430 if (!(env->mmu_model & POWERPC_MMU_64)) {
434 /* Collect MMU info from kernel if not already */
435 if (!has_smmu_info) {
436 kvm_get_smmu_info(cpu, &smmu_info);
437 has_smmu_info = true;
440 if (!max_cpu_page_size) {
441 max_cpu_page_size = qemu_getrampagesize();
444 /* Convert to QEMU form */
445 memset(cpu->hash64_opts->sps, 0, sizeof(*cpu->hash64_opts->sps));
447 /* If we have HV KVM, we need to forbid CI large pages if our
448 * host page size is smaller than 64K.
450 if (smmu_info.flags & KVM_PPC_PAGE_SIZES_REAL) {
451 if (getpagesize() >= 0x10000) {
452 cpu->hash64_opts->flags |= PPC_HASH64_CI_LARGEPAGE;
454 cpu->hash64_opts->flags &= ~PPC_HASH64_CI_LARGEPAGE;
459 * XXX This loop should be an entry wide AND of the capabilities that
460 * the selected CPU has with the capabilities that KVM supports.
462 for (ik = iq = 0; ik < KVM_PPC_PAGE_SIZES_MAX_SZ; ik++) {
463 PPCHash64SegmentPageSizes *qsps = &cpu->hash64_opts->sps[iq];
464 struct kvm_ppc_one_seg_page_size *ksps = &smmu_info.sps[ik];
466 if (!kvm_valid_page_size(smmu_info.flags, max_cpu_page_size,
470 qsps->page_shift = ksps->page_shift;
471 qsps->slb_enc = ksps->slb_enc;
472 for (jk = jq = 0; jk < KVM_PPC_PAGE_SIZES_MAX_SZ; jk++) {
473 if (!kvm_valid_page_size(smmu_info.flags, max_cpu_page_size,
474 ksps->enc[jk].page_shift)) {
477 qsps->enc[jq].page_shift = ksps->enc[jk].page_shift;
478 qsps->enc[jq].pte_enc = ksps->enc[jk].pte_enc;
479 if (++jq >= PPC_PAGE_SIZES_MAX_SZ) {
483 if (++iq >= PPC_PAGE_SIZES_MAX_SZ) {
487 cpu->hash64_opts->slb_size = smmu_info.slb_size;
488 if (!(smmu_info.flags & KVM_PPC_1T_SEGMENTS)) {
489 cpu->hash64_opts->flags &= ~PPC_HASH64_1TSEG;
493 bool kvmppc_is_mem_backend_page_size_ok(const char *obj_path)
495 Object *mem_obj = object_resolve_path(obj_path, NULL);
496 long pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(mem_obj));
498 return pagesize >= max_cpu_page_size;
501 #else /* defined (TARGET_PPC64) */
503 static inline void kvm_fixup_page_sizes(PowerPCCPU *cpu)
507 bool kvmppc_is_mem_backend_page_size_ok(const char *obj_path)
512 #endif /* !defined (TARGET_PPC64) */
514 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
516 return POWERPC_CPU(cpu)->vcpu_id;
519 /* e500 supports 2 h/w breakpoint and 2 watchpoint.
520 * book3s supports only 1 watchpoint, so array size
521 * of 4 is sufficient for now.
523 #define MAX_HW_BKPTS 4
525 static struct HWBreakpoint {
528 } hw_debug_points[MAX_HW_BKPTS];
530 static CPUWatchpoint hw_watchpoint;
532 /* Default there is no breakpoint and watchpoint supported */
533 static int max_hw_breakpoint;
534 static int max_hw_watchpoint;
535 static int nb_hw_breakpoint;
536 static int nb_hw_watchpoint;
538 static void kvmppc_hw_debug_points_init(CPUPPCState *cenv)
540 if (cenv->excp_model == POWERPC_EXCP_BOOKE) {
541 max_hw_breakpoint = 2;
542 max_hw_watchpoint = 2;
545 if ((max_hw_breakpoint + max_hw_watchpoint) > MAX_HW_BKPTS) {
546 fprintf(stderr, "Error initializing h/w breakpoints\n");
551 int kvm_arch_init_vcpu(CPUState *cs)
553 PowerPCCPU *cpu = POWERPC_CPU(cs);
554 CPUPPCState *cenv = &cpu->env;
557 /* Gather server mmu info from KVM and update the CPU state */
558 kvm_fixup_page_sizes(cpu);
560 /* Synchronize sregs with kvm */
561 ret = kvm_arch_sync_sregs(cpu);
563 if (ret == -EINVAL) {
564 error_report("Register sync failed... If you're using kvm-hv.ko,"
565 " only \"-cpu host\" is possible");
570 idle_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, kvm_kick_cpu, cpu);
572 switch (cenv->mmu_model) {
573 case POWERPC_MMU_BOOKE206:
574 /* This target supports access to KVM's guest TLB */
575 ret = kvm_booke206_tlb_init(cpu);
577 case POWERPC_MMU_2_07:
578 if (!cap_htm && !kvmppc_is_pr(cs->kvm_state)) {
579 /* KVM-HV has transactional memory on POWER8 also without the
580 * KVM_CAP_PPC_HTM extension, so enable it here instead as
581 * long as it's availble to userspace on the host. */
582 if (qemu_getauxval(AT_HWCAP2) & PPC_FEATURE2_HAS_HTM) {
591 kvm_get_one_reg(cs, KVM_REG_PPC_DEBUG_INST, &debug_inst_opcode);
592 kvmppc_hw_debug_points_init(cenv);
597 static void kvm_sw_tlb_put(PowerPCCPU *cpu)
599 CPUPPCState *env = &cpu->env;
600 CPUState *cs = CPU(cpu);
601 struct kvm_dirty_tlb dirty_tlb;
602 unsigned char *bitmap;
605 if (!env->kvm_sw_tlb) {
609 bitmap = g_malloc((env->nb_tlb + 7) / 8);
610 memset(bitmap, 0xFF, (env->nb_tlb + 7) / 8);
612 dirty_tlb.bitmap = (uintptr_t)bitmap;
613 dirty_tlb.num_dirty = env->nb_tlb;
615 ret = kvm_vcpu_ioctl(cs, KVM_DIRTY_TLB, &dirty_tlb);
617 fprintf(stderr, "%s: KVM_DIRTY_TLB: %s\n",
618 __func__, strerror(-ret));
624 static void kvm_get_one_spr(CPUState *cs, uint64_t id, int spr)
626 PowerPCCPU *cpu = POWERPC_CPU(cs);
627 CPUPPCState *env = &cpu->env;
632 struct kvm_one_reg reg = {
634 .addr = (uintptr_t) &val,
638 ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®);
640 trace_kvm_failed_spr_get(spr, strerror(errno));
642 switch (id & KVM_REG_SIZE_MASK) {
643 case KVM_REG_SIZE_U32:
644 env->spr[spr] = val.u32;
647 case KVM_REG_SIZE_U64:
648 env->spr[spr] = val.u64;
652 /* Don't handle this size yet */
658 static void kvm_put_one_spr(CPUState *cs, uint64_t id, int spr)
660 PowerPCCPU *cpu = POWERPC_CPU(cs);
661 CPUPPCState *env = &cpu->env;
666 struct kvm_one_reg reg = {
668 .addr = (uintptr_t) &val,
672 switch (id & KVM_REG_SIZE_MASK) {
673 case KVM_REG_SIZE_U32:
674 val.u32 = env->spr[spr];
677 case KVM_REG_SIZE_U64:
678 val.u64 = env->spr[spr];
682 /* Don't handle this size yet */
686 ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
688 trace_kvm_failed_spr_set(spr, strerror(errno));
692 static int kvm_put_fp(CPUState *cs)
694 PowerPCCPU *cpu = POWERPC_CPU(cs);
695 CPUPPCState *env = &cpu->env;
696 struct kvm_one_reg reg;
700 if (env->insns_flags & PPC_FLOAT) {
701 uint64_t fpscr = env->fpscr;
702 bool vsx = !!(env->insns_flags2 & PPC2_VSX);
704 reg.id = KVM_REG_PPC_FPSCR;
705 reg.addr = (uintptr_t)&fpscr;
706 ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
708 DPRINTF("Unable to set FPSCR to KVM: %s\n", strerror(errno));
712 for (i = 0; i < 32; i++) {
715 #ifdef HOST_WORDS_BIGENDIAN
716 vsr[0] = float64_val(env->fpr[i]);
717 vsr[1] = env->vsr[i];
719 vsr[0] = env->vsr[i];
720 vsr[1] = float64_val(env->fpr[i]);
722 reg.addr = (uintptr_t) &vsr;
723 reg.id = vsx ? KVM_REG_PPC_VSR(i) : KVM_REG_PPC_FPR(i);
725 ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
727 DPRINTF("Unable to set %s%d to KVM: %s\n", vsx ? "VSR" : "FPR",
734 if (env->insns_flags & PPC_ALTIVEC) {
735 reg.id = KVM_REG_PPC_VSCR;
736 reg.addr = (uintptr_t)&env->vscr;
737 ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
739 DPRINTF("Unable to set VSCR to KVM: %s\n", strerror(errno));
743 for (i = 0; i < 32; i++) {
744 reg.id = KVM_REG_PPC_VR(i);
745 reg.addr = (uintptr_t)&env->avr[i];
746 ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
748 DPRINTF("Unable to set VR%d to KVM: %s\n", i, strerror(errno));
757 static int kvm_get_fp(CPUState *cs)
759 PowerPCCPU *cpu = POWERPC_CPU(cs);
760 CPUPPCState *env = &cpu->env;
761 struct kvm_one_reg reg;
765 if (env->insns_flags & PPC_FLOAT) {
767 bool vsx = !!(env->insns_flags2 & PPC2_VSX);
769 reg.id = KVM_REG_PPC_FPSCR;
770 reg.addr = (uintptr_t)&fpscr;
771 ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®);
773 DPRINTF("Unable to get FPSCR from KVM: %s\n", strerror(errno));
779 for (i = 0; i < 32; i++) {
782 reg.addr = (uintptr_t) &vsr;
783 reg.id = vsx ? KVM_REG_PPC_VSR(i) : KVM_REG_PPC_FPR(i);
785 ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®);
787 DPRINTF("Unable to get %s%d from KVM: %s\n",
788 vsx ? "VSR" : "FPR", i, strerror(errno));
791 #ifdef HOST_WORDS_BIGENDIAN
792 env->fpr[i] = vsr[0];
794 env->vsr[i] = vsr[1];
797 env->fpr[i] = vsr[1];
799 env->vsr[i] = vsr[0];
806 if (env->insns_flags & PPC_ALTIVEC) {
807 reg.id = KVM_REG_PPC_VSCR;
808 reg.addr = (uintptr_t)&env->vscr;
809 ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®);
811 DPRINTF("Unable to get VSCR from KVM: %s\n", strerror(errno));
815 for (i = 0; i < 32; i++) {
816 reg.id = KVM_REG_PPC_VR(i);
817 reg.addr = (uintptr_t)&env->avr[i];
818 ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®);
820 DPRINTF("Unable to get VR%d from KVM: %s\n",
830 #if defined(TARGET_PPC64)
831 static int kvm_get_vpa(CPUState *cs)
833 PowerPCCPU *cpu = POWERPC_CPU(cs);
834 CPUPPCState *env = &cpu->env;
835 struct kvm_one_reg reg;
838 reg.id = KVM_REG_PPC_VPA_ADDR;
839 reg.addr = (uintptr_t)&env->vpa_addr;
840 ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®);
842 DPRINTF("Unable to get VPA address from KVM: %s\n", strerror(errno));
846 assert((uintptr_t)&env->slb_shadow_size
847 == ((uintptr_t)&env->slb_shadow_addr + 8));
848 reg.id = KVM_REG_PPC_VPA_SLB;
849 reg.addr = (uintptr_t)&env->slb_shadow_addr;
850 ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®);
852 DPRINTF("Unable to get SLB shadow state from KVM: %s\n",
857 assert((uintptr_t)&env->dtl_size == ((uintptr_t)&env->dtl_addr + 8));
858 reg.id = KVM_REG_PPC_VPA_DTL;
859 reg.addr = (uintptr_t)&env->dtl_addr;
860 ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®);
862 DPRINTF("Unable to get dispatch trace log state from KVM: %s\n",
870 static int kvm_put_vpa(CPUState *cs)
872 PowerPCCPU *cpu = POWERPC_CPU(cs);
873 CPUPPCState *env = &cpu->env;
874 struct kvm_one_reg reg;
877 /* SLB shadow or DTL can't be registered unless a master VPA is
878 * registered. That means when restoring state, if a VPA *is*
879 * registered, we need to set that up first. If not, we need to
880 * deregister the others before deregistering the master VPA */
881 assert(env->vpa_addr || !(env->slb_shadow_addr || env->dtl_addr));
884 reg.id = KVM_REG_PPC_VPA_ADDR;
885 reg.addr = (uintptr_t)&env->vpa_addr;
886 ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
888 DPRINTF("Unable to set VPA address to KVM: %s\n", strerror(errno));
893 assert((uintptr_t)&env->slb_shadow_size
894 == ((uintptr_t)&env->slb_shadow_addr + 8));
895 reg.id = KVM_REG_PPC_VPA_SLB;
896 reg.addr = (uintptr_t)&env->slb_shadow_addr;
897 ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
899 DPRINTF("Unable to set SLB shadow state to KVM: %s\n", strerror(errno));
903 assert((uintptr_t)&env->dtl_size == ((uintptr_t)&env->dtl_addr + 8));
904 reg.id = KVM_REG_PPC_VPA_DTL;
905 reg.addr = (uintptr_t)&env->dtl_addr;
906 ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
908 DPRINTF("Unable to set dispatch trace log state to KVM: %s\n",
913 if (!env->vpa_addr) {
914 reg.id = KVM_REG_PPC_VPA_ADDR;
915 reg.addr = (uintptr_t)&env->vpa_addr;
916 ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
918 DPRINTF("Unable to set VPA address to KVM: %s\n", strerror(errno));
925 #endif /* TARGET_PPC64 */
927 int kvmppc_put_books_sregs(PowerPCCPU *cpu)
929 CPUPPCState *env = &cpu->env;
930 struct kvm_sregs sregs;
933 sregs.pvr = env->spr[SPR_PVR];
936 PPCVirtualHypervisorClass *vhc =
937 PPC_VIRTUAL_HYPERVISOR_GET_CLASS(cpu->vhyp);
938 sregs.u.s.sdr1 = vhc->encode_hpt_for_kvm_pr(cpu->vhyp);
940 sregs.u.s.sdr1 = env->spr[SPR_SDR1];
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;
950 sregs.u.s.ppc64.slb[i].slbv = env->slb[i].vsid;
955 for (i = 0; i < 16; i++) {
956 sregs.u.s.ppc32.sr[i] = env->sr[i];
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)
964 sregs.u.s.ppc32.ibat[i] = ((uint64_t)env->IBAT[0][i] << 32)
968 return kvm_vcpu_ioctl(CPU(cpu), KVM_SET_SREGS, &sregs);
971 int kvm_arch_put_registers(CPUState *cs, int level)
973 PowerPCCPU *cpu = POWERPC_CPU(cs);
974 CPUPPCState *env = &cpu->env;
975 struct kvm_regs regs;
979 ret = kvm_vcpu_ioctl(cs, KVM_GET_REGS, ®s);
986 regs.xer = cpu_read_xer(env);
990 regs.srr0 = env->spr[SPR_SRR0];
991 regs.srr1 = env->spr[SPR_SRR1];
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];
1002 regs.pid = env->spr[SPR_BOOKE_PID];
1004 for (i = 0;i < 32; i++)
1005 regs.gpr[i] = env->gpr[i];
1008 for (i = 0; i < 8; i++) {
1009 regs.cr |= (env->crf[i] & 15) << (4 * (7 - i));
1012 ret = kvm_vcpu_ioctl(cs, KVM_SET_REGS, ®s);
1018 if (env->tlb_dirty) {
1019 kvm_sw_tlb_put(cpu);
1020 env->tlb_dirty = false;
1023 if (cap_segstate && (level >= KVM_PUT_RESET_STATE)) {
1024 ret = kvmppc_put_books_sregs(cpu);
1030 if (cap_hior && (level >= KVM_PUT_RESET_STATE)) {
1031 kvm_put_one_spr(cs, KVM_REG_PPC_HIOR, SPR_HIOR);
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;
1045 kvm_put_one_spr(cs, id, i);
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]);
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]);
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);
1070 if (kvm_put_vpa(cs) < 0) {
1071 DPRINTF("Warning: Unable to set VPA information to KVM\n");
1075 kvm_set_one_reg(cs, KVM_REG_PPC_TB_OFFSET, &env->tb_env->tb_offset);
1076 #endif /* TARGET_PPC64 */
1082 static void kvm_sync_excp(CPUPPCState *env, int vector, int ivor)
1084 env->excp_vectors[vector] = env->spr[ivor] + env->spr[SPR_BOOKE_IVPR];
1087 static int kvmppc_get_booke_sregs(PowerPCCPU *cpu)
1089 CPUPPCState *env = &cpu->env;
1090 struct kvm_sregs sregs;
1093 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_SREGS, &sregs);
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;
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;
1120 if (sregs.u.e.features & KVM_SREGS_E_64) {
1121 env->spr[SPR_BOOKE_EPCR] = sregs.u.e.epcr;
1124 if (sregs.u.e.features & KVM_SREGS_E_SPRG8) {
1125 env->spr[SPR_BOOKE_SPRG8] = sregs.u.e.sprg8;
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);
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);
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);
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);
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];
1197 if (sregs.u.e.features & KVM_SREGS_EXP) {
1198 env->spr[SPR_BOOKE_EPR] = sregs.u.e.epr;
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;
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;
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;
1220 static int kvmppc_get_books_sregs(PowerPCCPU *cpu)
1222 CPUPPCState *env = &cpu->env;
1223 struct kvm_sregs sregs;
1227 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_SREGS, &sregs);
1233 ppc_store_sdr1(env, sregs.u.s.sdr1);
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
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;
1249 * Only restore valid entries
1251 if (rb & SLB_ESID_V) {
1252 ppc_store_slb(cpu, rb & 0xfff, rb & ~0xfffULL, rs);
1258 for (i = 0; i < 16; i++) {
1259 env->sr[i] = sregs.u.s.ppc32.sr[i];
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;
1273 int kvm_arch_get_registers(CPUState *cs)
1275 PowerPCCPU *cpu = POWERPC_CPU(cs);
1276 CPUPPCState *env = &cpu->env;
1277 struct kvm_regs regs;
1281 ret = kvm_vcpu_ioctl(cs, KVM_GET_REGS, ®s);
1286 for (i = 7; i >= 0; i--) {
1287 env->crf[i] = cr & 15;
1291 env->ctr = regs.ctr;
1293 cpu_write_xer(env, regs.xer);
1294 env->msr = regs.msr;
1297 env->spr[SPR_SRR0] = regs.srr0;
1298 env->spr[SPR_SRR1] = regs.srr1;
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;
1309 env->spr[SPR_BOOKE_PID] = regs.pid;
1311 for (i = 0;i < 32; i++)
1312 env->gpr[i] = regs.gpr[i];
1316 if (cap_booke_sregs) {
1317 ret = kvmppc_get_booke_sregs(cpu);
1324 ret = kvmppc_get_books_sregs(cpu);
1331 kvm_get_one_spr(cs, KVM_REG_PPC_HIOR, SPR_HIOR);
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;
1345 kvm_get_one_spr(cs, id, i);
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]);
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]);
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);
1370 if (kvm_get_vpa(cs) < 0) {
1371 DPRINTF("Warning: Unable to get VPA information from KVM\n");
1375 kvm_get_one_reg(cs, KVM_REG_PPC_TB_OFFSET, &env->tb_env->tb_offset);
1382 int kvmppc_set_interrupt(PowerPCCPU *cpu, int irq, int level)
1384 unsigned virq = level ? KVM_INTERRUPT_SET_LEVEL : KVM_INTERRUPT_UNSET;
1386 if (irq != PPC_INTERRUPT_EXT) {
1390 if (!kvm_enabled() || !cap_interrupt_unset || !cap_interrupt_level) {
1394 kvm_vcpu_ioctl(CPU(cpu), KVM_INTERRUPT, &virq);
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
1404 #define PPC_INPUT_INT PPC6xx_INPUT_INT
1407 void kvm_arch_pre_run(CPUState *cs, struct kvm_run *run)
1409 PowerPCCPU *cpu = POWERPC_CPU(cs);
1410 CPUPPCState *env = &cpu->env;
1414 qemu_mutex_lock_iothread();
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)))
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.
1427 irq = KVM_INTERRUPT_SET;
1429 DPRINTF("injected interrupt %d\n", irq);
1430 r = kvm_vcpu_ioctl(cs, KVM_INTERRUPT, &irq);
1432 printf("cpu %d fail inject %x\n", cs->cpu_index, irq);
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));
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. */
1444 qemu_mutex_unlock_iothread();
1447 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
1449 return MEMTXATTRS_UNSPECIFIED;
1452 int kvm_arch_process_async_events(CPUState *cs)
1457 static int kvmppc_handle_halt(PowerPCCPU *cpu)
1459 CPUState *cs = CPU(cpu);
1460 CPUPPCState *env = &cpu->env;
1462 if (!(cs->interrupt_request & CPU_INTERRUPT_HARD) && (msr_ee)) {
1464 cs->exception_index = EXCP_HLT;
1470 /* map dcr access to existing qemu dcr emulation */
1471 static int kvmppc_handle_dcr_read(CPUPPCState *env, uint32_t dcrn, uint32_t *data)
1473 if (ppc_dcr_read(env->dcr_env, dcrn, data) < 0)
1474 fprintf(stderr, "Read to unhandled DCR (0x%x)\n", dcrn);
1479 static int kvmppc_handle_dcr_write(CPUPPCState *env, uint32_t dcrn, uint32_t data)
1481 if (ppc_dcr_write(env->dcr_env, dcrn, data) < 0)
1482 fprintf(stderr, "Write to unhandled DCR (0x%x)\n", dcrn);
1487 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
1489 /* Mixed endian case is not handled */
1490 uint32_t sc = debug_inst_opcode;
1492 if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
1494 cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&sc, sizeof(sc), 1)) {
1501 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
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,
1515 static int find_hw_breakpoint(target_ulong addr, int type)
1519 assert((nb_hw_breakpoint + nb_hw_watchpoint)
1520 <= ARRAY_SIZE(hw_debug_points));
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) {
1532 static int find_hw_watchpoint(target_ulong addr, int *flag)
1536 n = find_hw_breakpoint(addr, GDB_WATCHPOINT_ACCESS);
1538 *flag = BP_MEM_ACCESS;
1542 n = find_hw_breakpoint(addr, GDB_WATCHPOINT_WRITE);
1544 *flag = BP_MEM_WRITE;
1548 n = find_hw_breakpoint(addr, GDB_WATCHPOINT_READ);
1550 *flag = BP_MEM_READ;
1557 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
1558 target_ulong len, int type)
1560 if ((nb_hw_breakpoint + nb_hw_watchpoint) >= ARRAY_SIZE(hw_debug_points)) {
1564 hw_debug_points[nb_hw_breakpoint + nb_hw_watchpoint].addr = addr;
1565 hw_debug_points[nb_hw_breakpoint + nb_hw_watchpoint].type = type;
1568 case GDB_BREAKPOINT_HW:
1569 if (nb_hw_breakpoint >= max_hw_breakpoint) {
1573 if (find_hw_breakpoint(addr, type) >= 0) {
1580 case GDB_WATCHPOINT_WRITE:
1581 case GDB_WATCHPOINT_READ:
1582 case GDB_WATCHPOINT_ACCESS:
1583 if (nb_hw_watchpoint >= max_hw_watchpoint) {
1587 if (find_hw_breakpoint(addr, type) >= 0) {
1601 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
1602 target_ulong len, int type)
1606 n = find_hw_breakpoint(addr, type);
1612 case GDB_BREAKPOINT_HW:
1616 case GDB_WATCHPOINT_WRITE:
1617 case GDB_WATCHPOINT_READ:
1618 case GDB_WATCHPOINT_ACCESS:
1625 hw_debug_points[n] = hw_debug_points[nb_hw_breakpoint + nb_hw_watchpoint];
1630 void kvm_arch_remove_all_hw_breakpoints(void)
1632 nb_hw_breakpoint = nb_hw_watchpoint = 0;
1635 void kvm_arch_update_guest_debug(CPUState *cs, struct kvm_guest_debug *dbg)
1639 /* Software Breakpoint updates */
1640 if (kvm_sw_breakpoints_active(cs)) {
1641 dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP;
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));
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;
1656 case GDB_WATCHPOINT_WRITE:
1657 dbg->arch.bp[n].type = KVMPPC_DEBUG_WATCH_WRITE;
1659 case GDB_WATCHPOINT_READ:
1660 dbg->arch.bp[n].type = KVMPPC_DEBUG_WATCH_READ;
1662 case GDB_WATCHPOINT_ACCESS:
1663 dbg->arch.bp[n].type = KVMPPC_DEBUG_WATCH_WRITE |
1664 KVMPPC_DEBUG_WATCH_READ;
1667 cpu_abort(cs, "Unsupported breakpoint type\n");
1669 dbg->arch.bp[n].addr = hw_debug_points[n].addr;
1674 static int kvm_handle_debug(PowerPCCPU *cpu, struct kvm_run *run)
1676 CPUState *cs = CPU(cpu);
1677 CPUPPCState *env = &cpu->env;
1678 struct kvm_debug_exit_arch *arch_info = &run->debug.arch;
1683 if (cs->singlestep_enabled) {
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);
1692 } else if (arch_info->status & (KVMPPC_DEBUG_WATCH_READ |
1693 KVMPPC_DEBUG_WATCH_WRITE)) {
1694 n = find_hw_watchpoint(arch_info->address, &flag);
1697 cs->watchpoint_hit = &hw_watchpoint;
1698 hw_watchpoint.vaddr = hw_debug_points[n].addr;
1699 hw_watchpoint.flags = flag;
1703 } else if (kvm_find_sw_breakpoint(cs, arch_info->address)) {
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
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.
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.
1732 cs->exception_index = POWERPC_EXCP_PROGRAM;
1733 env->error_code = POWERPC_EXCP_INVAL;
1734 ppc_cpu_do_interrupt(cs);
1740 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1742 PowerPCCPU *cpu = POWERPC_CPU(cs);
1743 CPUPPCState *env = &cpu->env;
1746 qemu_mutex_lock_iothread();
1748 switch (run->exit_reason) {
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);
1754 DPRINTF("handle dcr read\n");
1755 ret = kvmppc_handle_dcr_read(env, run->dcr.dcrn, &run->dcr.data);
1759 DPRINTF("handle halt\n");
1760 ret = kvmppc_handle_halt(cpu);
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,
1767 run->papr_hcall.args);
1772 DPRINTF("handle epr\n");
1773 run->epr.epr = ldl_phys(cs->as, env->mpic_iack);
1776 case KVM_EXIT_WATCHDOG:
1777 DPRINTF("handle watchdog expiry\n");
1778 watchdog_perform_action();
1782 case KVM_EXIT_DEBUG:
1783 DPRINTF("handle debug exception\n");
1784 if (kvm_handle_debug(cpu, run)) {
1788 /* re-enter, this exception was guest-internal */
1793 fprintf(stderr, "KVM: unknown exit reason %d\n", run->exit_reason);
1798 qemu_mutex_unlock_iothread();
1802 int kvmppc_or_tsr_bits(PowerPCCPU *cpu, uint32_t tsr_bits)
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,
1811 return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
1814 int kvmppc_clear_tsr_bits(PowerPCCPU *cpu, uint32_t tsr_bits)
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,
1824 return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
1827 int kvmppc_set_tcr(PowerPCCPU *cpu)
1829 CPUState *cs = CPU(cpu);
1830 CPUPPCState *env = &cpu->env;
1831 uint32_t tcr = env->spr[SPR_BOOKE_TCR];
1833 struct kvm_one_reg reg = {
1834 .id = KVM_REG_PPC_TCR,
1835 .addr = (uintptr_t) &tcr,
1838 return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
1841 int kvmppc_booke_watchdog_enable(PowerPCCPU *cpu)
1843 CPUState *cs = CPU(cpu);
1846 if (!kvm_enabled()) {
1850 if (!cap_ppc_watchdog) {
1851 printf("warning: KVM does not support watchdog");
1855 ret = kvm_vcpu_enable_cap(cs, KVM_CAP_PPC_BOOKE_WATCHDOG, 0);
1857 fprintf(stderr, "%s: couldn't enable KVM_CAP_PPC_BOOKE_WATCHDOG: %s\n",
1858 __func__, strerror(-ret));
1865 static int read_cpuinfo(const char *field, char *value, int len)
1869 int field_len = strlen(field);
1872 f = fopen("/proc/cpuinfo", "r");
1878 if (!fgets(line, sizeof(line), f)) {
1881 if (!strncmp(line, field, field_len)) {
1882 pstrcpy(value, len, line);
1893 uint32_t kvmppc_get_tbfreq(void)
1897 uint32_t retval = NANOSECONDS_PER_SECOND;
1899 if (read_cpuinfo("timebase", line, sizeof(line))) {
1903 if (!(ns = strchr(line, ':'))) {
1912 bool kvmppc_get_host_serial(char **value)
1914 return g_file_get_contents("/proc/device-tree/system-id", value, NULL,
1918 bool kvmppc_get_host_model(char **value)
1920 return g_file_get_contents("/proc/device-tree/model", value, NULL, NULL);
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)
1926 struct dirent *dirp;
1929 if ((dp = opendir(PROC_DEVTREE_CPU)) == NULL) {
1930 printf("Can't open directory " PROC_DEVTREE_CPU "\n");
1935 while ((dirp = readdir(dp)) != NULL) {
1937 snprintf(buf, buf_len, "%s%s/clock-frequency", PROC_DEVTREE_CPU,
1939 f = fopen(buf, "r");
1941 snprintf(buf, buf_len, "%s%s", PROC_DEVTREE_CPU, dirp->d_name);
1948 if (buf[0] == '\0') {
1949 printf("Unknown host!\n");
1956 static uint64_t kvmppc_read_int_dt(const char *filename)
1965 f = fopen(filename, "rb");
1970 len = fread(&u, 1, sizeof(u), f);
1974 /* property is a 32-bit quantity */
1975 return be32_to_cpu(u.v32);
1977 return be64_to_cpu(u.v64);
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
1987 static uint64_t kvmppc_read_int_cpu_dt(const char *propname)
1989 char buf[PATH_MAX], *tmp;
1992 if (kvmppc_find_cpu_dt(buf, sizeof(buf))) {
1996 tmp = g_strdup_printf("%s/%s", buf, propname);
1997 val = kvmppc_read_int_dt(tmp);
2003 uint64_t kvmppc_get_clockfreq(void)
2005 return kvmppc_read_int_cpu_dt("clock-frequency");
2008 static int kvmppc_get_pvinfo(CPUPPCState *env, struct kvm_ppc_pvinfo *pvinfo)
2010 PowerPCCPU *cpu = ppc_env_get_cpu(env);
2011 CPUState *cs = CPU(cpu);
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)) {
2021 int kvmppc_get_hasidle(CPUPPCState *env)
2023 struct kvm_ppc_pvinfo pvinfo;
2025 if (!kvmppc_get_pvinfo(env, &pvinfo) &&
2026 (pvinfo.flags & KVM_PPC_PVINFO_FLAGS_EV_IDLE)) {
2033 int kvmppc_get_hypercall(CPUPPCState *env, uint8_t *buf, int buf_len)
2035 uint32_t *hc = (uint32_t*)buf;
2036 struct kvm_ppc_pvinfo pvinfo;
2038 if (!kvmppc_get_pvinfo(env, &pvinfo)) {
2039 memcpy(buf, pvinfo.hcall, buf_len);
2044 * Fallback to always fail hypercalls regardless of endianness:
2046 * tdi 0,r0,72 (becomes b .+8 in wrong endian, nop in good endian)
2048 * b .+8 (becomes nop in wrong endian)
2049 * bswap32(li r3, -1)
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));
2060 static inline int kvmppc_enable_hcall(KVMState *s, target_ulong hcall)
2062 return kvm_vm_enable_cap(s, KVM_CAP_PPC_ENABLE_HCALL, 0, hcall, 1);
2065 void kvmppc_enable_logical_ci_hcalls(void)
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.
2073 kvmppc_enable_hcall(kvm_state, H_LOGICAL_CI_LOAD);
2074 kvmppc_enable_hcall(kvm_state, H_LOGICAL_CI_STORE);
2077 void kvmppc_enable_set_mode_hcall(void)
2079 kvmppc_enable_hcall(kvm_state, H_SET_MODE);
2082 void kvmppc_enable_clear_ref_mod_hcalls(void)
2084 kvmppc_enable_hcall(kvm_state, H_CLEAR_REF);
2085 kvmppc_enable_hcall(kvm_state, H_CLEAR_MOD);
2088 void kvmppc_set_papr(PowerPCCPU *cpu)
2090 CPUState *cs = CPU(cpu);
2093 ret = kvm_vcpu_enable_cap(cs, KVM_CAP_PPC_PAPR, 0);
2095 error_report("This vCPU type or KVM version does not support PAPR");
2099 /* Update the capability flag so we sync the right information
2104 int kvmppc_set_compat(PowerPCCPU *cpu, uint32_t compat_pvr)
2106 return kvm_set_one_reg(CPU(cpu), KVM_REG_PPC_ARCH_COMPAT, &compat_pvr);
2109 void kvmppc_set_mpic_proxy(PowerPCCPU *cpu, int mpic_proxy)
2111 CPUState *cs = CPU(cpu);
2114 ret = kvm_vcpu_enable_cap(cs, KVM_CAP_PPC_EPR, 0, mpic_proxy);
2115 if (ret && mpic_proxy) {
2116 error_report("This KVM version does not support EPR");
2121 int kvmppc_smt_threads(void)
2123 return cap_ppc_smt ? cap_ppc_smt : 1;
2126 int kvmppc_set_smt_threads(int smt)
2130 ret = kvm_vm_enable_cap(kvm_state, KVM_CAP_PPC_SMT, 0, smt, 0);
2137 void kvmppc_hint_smt_possible(Error **errp)
2143 assert(kvm_enabled());
2144 if (cap_ppc_smt_possible) {
2145 g = g_string_new("Available VSMT modes:");
2146 for (i = 63; i >= 0; i--) {
2147 if ((1UL << i) & cap_ppc_smt_possible) {
2148 g_string_append_printf(g, " %lu", (1UL << i));
2151 s = g_string_free(g, false);
2152 error_append_hint(errp, "%s.\n", s);
2155 error_append_hint(errp,
2156 "This KVM seems to be too old to support VSMT.\n");
2162 off_t kvmppc_alloc_rma(void **rma)
2166 struct kvm_allocate_rma ret;
2168 /* If cap_ppc_rma == 0, contiguous RMA allocation is not supported
2169 * if cap_ppc_rma == 1, contiguous RMA allocation is supported, but
2170 * not necessary on this hardware
2171 * if cap_ppc_rma == 2, contiguous RMA allocation is needed on this hardware
2173 * FIXME: We should allow the user to force contiguous RMA
2174 * allocation in the cap_ppc_rma==1 case.
2176 if (cap_ppc_rma < 2) {
2180 fd = kvm_vm_ioctl(kvm_state, KVM_ALLOCATE_RMA, &ret);
2182 fprintf(stderr, "KVM: Error on KVM_ALLOCATE_RMA: %s\n",
2187 size = MIN(ret.rma_size, 256ul << 20);
2189 *rma = mmap(NULL, size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
2190 if (*rma == MAP_FAILED) {
2191 fprintf(stderr, "KVM: Error mapping RMA: %s\n", strerror(errno));
2198 uint64_t kvmppc_rma_size(uint64_t current_size, unsigned int hash_shift)
2200 struct kvm_ppc_smmu_info info;
2201 long rampagesize, best_page_shift;
2204 if (cap_ppc_rma >= 2) {
2205 return current_size;
2208 /* Find the largest hardware supported page size that's less than
2209 * or equal to the (logical) backing page size of guest RAM */
2210 kvm_get_smmu_info(POWERPC_CPU(first_cpu), &info);
2211 rampagesize = qemu_getrampagesize();
2212 best_page_shift = 0;
2214 for (i = 0; i < KVM_PPC_PAGE_SIZES_MAX_SZ; i++) {
2215 struct kvm_ppc_one_seg_page_size *sps = &info.sps[i];
2217 if (!sps->page_shift) {
2221 if ((sps->page_shift > best_page_shift)
2222 && ((1UL << sps->page_shift) <= rampagesize)) {
2223 best_page_shift = sps->page_shift;
2227 return MIN(current_size,
2228 1ULL << (best_page_shift + hash_shift - 7));
2232 bool kvmppc_spapr_use_multitce(void)
2234 return cap_spapr_multitce;
2237 int kvmppc_spapr_enable_inkernel_multitce(void)
2241 ret = kvm_vm_enable_cap(kvm_state, KVM_CAP_PPC_ENABLE_HCALL, 0,
2242 H_PUT_TCE_INDIRECT, 1);
2244 ret = kvm_vm_enable_cap(kvm_state, KVM_CAP_PPC_ENABLE_HCALL, 0,
2251 void *kvmppc_create_spapr_tce(uint32_t liobn, uint32_t page_shift,
2252 uint64_t bus_offset, uint32_t nb_table,
2253 int *pfd, bool need_vfio)
2259 /* Must set fd to -1 so we don't try to munmap when called for
2260 * destroying the table, which the upper layers -will- do
2263 if (!cap_spapr_tce || (need_vfio && !cap_spapr_vfio)) {
2267 if (cap_spapr_tce_64) {
2268 struct kvm_create_spapr_tce_64 args = {
2270 .page_shift = page_shift,
2271 .offset = bus_offset >> page_shift,
2275 fd = kvm_vm_ioctl(kvm_state, KVM_CREATE_SPAPR_TCE_64, &args);
2278 "KVM: Failed to create TCE64 table for liobn 0x%x\n",
2282 } else if (cap_spapr_tce) {
2283 uint64_t window_size = (uint64_t) nb_table << page_shift;
2284 struct kvm_create_spapr_tce args = {
2286 .window_size = window_size,
2288 if ((window_size != args.window_size) || bus_offset) {
2291 fd = kvm_vm_ioctl(kvm_state, KVM_CREATE_SPAPR_TCE, &args);
2293 fprintf(stderr, "KVM: Failed to create TCE table for liobn 0x%x\n",
2301 len = nb_table * sizeof(uint64_t);
2302 /* FIXME: round this up to page size */
2304 table = mmap(NULL, len, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
2305 if (table == MAP_FAILED) {
2306 fprintf(stderr, "KVM: Failed to map TCE table for liobn 0x%x\n",
2316 int kvmppc_remove_spapr_tce(void *table, int fd, uint32_t nb_table)
2324 len = nb_table * sizeof(uint64_t);
2325 if ((munmap(table, len) < 0) ||
2327 fprintf(stderr, "KVM: Unexpected error removing TCE table: %s",
2329 /* Leak the table */
2335 int kvmppc_reset_htab(int shift_hint)
2337 uint32_t shift = shift_hint;
2339 if (!kvm_enabled()) {
2340 /* Full emulation, tell caller to allocate htab itself */
2343 if (kvm_vm_check_extension(kvm_state, KVM_CAP_PPC_ALLOC_HTAB)) {
2345 ret = kvm_vm_ioctl(kvm_state, KVM_PPC_ALLOCATE_HTAB, &shift);
2346 if (ret == -ENOTTY) {
2347 /* At least some versions of PR KVM advertise the
2348 * capability, but don't implement the ioctl(). Oops.
2349 * Return 0 so that we allocate the htab in qemu, as is
2350 * correct for PR. */
2352 } else if (ret < 0) {
2358 /* We have a kernel that predates the htab reset calls. For PR
2359 * KVM, we need to allocate the htab ourselves, for an HV KVM of
2360 * this era, it has allocated a 16MB fixed size hash table already. */
2361 if (kvmppc_is_pr(kvm_state)) {
2362 /* PR - tell caller to allocate htab */
2365 /* HV - assume 16MB kernel allocated htab */
2370 static inline uint32_t mfpvr(void)
2379 static void alter_insns(uint64_t *word, uint64_t flags, bool on)
2388 static void kvmppc_host_cpu_class_init(ObjectClass *oc, void *data)
2390 PowerPCCPUClass *pcc = POWERPC_CPU_CLASS(oc);
2391 uint32_t dcache_size = kvmppc_read_int_cpu_dt("d-cache-size");
2392 uint32_t icache_size = kvmppc_read_int_cpu_dt("i-cache-size");
2394 /* Now fix up the class with information we can query from the host */
2397 alter_insns(&pcc->insns_flags, PPC_ALTIVEC,
2398 qemu_getauxval(AT_HWCAP) & PPC_FEATURE_HAS_ALTIVEC);
2399 alter_insns(&pcc->insns_flags2, PPC2_VSX,
2400 qemu_getauxval(AT_HWCAP) & PPC_FEATURE_HAS_VSX);
2401 alter_insns(&pcc->insns_flags2, PPC2_DFP,
2402 qemu_getauxval(AT_HWCAP) & PPC_FEATURE_HAS_DFP);
2404 if (dcache_size != -1) {
2405 pcc->l1_dcache_size = dcache_size;
2408 if (icache_size != -1) {
2409 pcc->l1_icache_size = icache_size;
2412 #if defined(TARGET_PPC64)
2413 pcc->radix_page_info = kvm_get_radix_page_info();
2415 if ((pcc->pvr & 0xffffff00) == CPU_POWERPC_POWER9_DD1) {
2417 * POWER9 DD1 has some bugs which make it not really ISA 3.00
2418 * compliant. More importantly, advertising ISA 3.00
2419 * architected mode may prevent guests from activating
2420 * necessary DD1 workarounds.
2422 pcc->pcr_supported &= ~(PCR_COMPAT_3_00 | PCR_COMPAT_2_07
2423 | PCR_COMPAT_2_06 | PCR_COMPAT_2_05);
2425 #endif /* defined(TARGET_PPC64) */
2428 bool kvmppc_has_cap_epr(void)
2433 bool kvmppc_has_cap_fixup_hcalls(void)
2435 return cap_fixup_hcalls;
2438 bool kvmppc_has_cap_htm(void)
2443 bool kvmppc_has_cap_mmu_radix(void)
2445 return cap_mmu_radix;
2448 bool kvmppc_has_cap_mmu_hash_v3(void)
2450 return cap_mmu_hash_v3;
2453 static void kvmppc_get_cpu_characteristics(KVMState *s)
2455 struct kvm_ppc_cpu_char c;
2459 cap_ppc_safe_cache = 0;
2460 cap_ppc_safe_bounds_check = 0;
2461 cap_ppc_safe_indirect_branch = 0;
2463 ret = kvm_vm_check_extension(s, KVM_CAP_PPC_GET_CPU_CHAR);
2467 ret = kvm_vm_ioctl(s, KVM_PPC_GET_CPU_CHAR, &c);
2471 /* Parse and set cap_ppc_safe_cache */
2472 if (~c.behaviour & c.behaviour_mask & H_CPU_BEHAV_L1D_FLUSH_PR) {
2473 cap_ppc_safe_cache = 2;
2474 } else if ((c.character & c.character_mask & H_CPU_CHAR_L1D_THREAD_PRIV) &&
2475 (c.character & c.character_mask
2476 & (H_CPU_CHAR_L1D_FLUSH_ORI30 | H_CPU_CHAR_L1D_FLUSH_TRIG2))) {
2477 cap_ppc_safe_cache = 1;
2479 /* Parse and set cap_ppc_safe_bounds_check */
2480 if (~c.behaviour & c.behaviour_mask & H_CPU_BEHAV_BNDS_CHK_SPEC_BAR) {
2481 cap_ppc_safe_bounds_check = 2;
2482 } else if (c.character & c.character_mask & H_CPU_CHAR_SPEC_BAR_ORI31) {
2483 cap_ppc_safe_bounds_check = 1;
2485 /* Parse and set cap_ppc_safe_indirect_branch */
2486 if (c.character & c.character_mask & H_CPU_CHAR_CACHE_COUNT_DIS) {
2487 cap_ppc_safe_indirect_branch = SPAPR_CAP_FIXED_CCD;
2488 } else if (c.character & c.character_mask & H_CPU_CHAR_BCCTRL_SERIALISED) {
2489 cap_ppc_safe_indirect_branch = SPAPR_CAP_FIXED_IBS;
2493 int kvmppc_get_cap_safe_cache(void)
2495 return cap_ppc_safe_cache;
2498 int kvmppc_get_cap_safe_bounds_check(void)
2500 return cap_ppc_safe_bounds_check;
2503 int kvmppc_get_cap_safe_indirect_branch(void)
2505 return cap_ppc_safe_indirect_branch;
2508 bool kvmppc_has_cap_spapr_vfio(void)
2510 return cap_spapr_vfio;
2513 PowerPCCPUClass *kvm_ppc_get_host_cpu_class(void)
2515 uint32_t host_pvr = mfpvr();
2516 PowerPCCPUClass *pvr_pcc;
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);
2526 static int kvm_ppc_register_host_cpu_type(MachineState *ms)
2528 TypeInfo type_info = {
2529 .name = TYPE_HOST_POWERPC_CPU,
2530 .class_init = kvmppc_host_cpu_class_init,
2532 MachineClass *mc = MACHINE_GET_CLASS(ms);
2533 PowerPCCPUClass *pvr_pcc;
2538 pvr_pcc = kvm_ppc_get_host_cpu_class();
2539 if (pvr_pcc == NULL) {
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;
2549 oc = object_class_by_name(type_info.name);
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)
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) {
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);
2574 int kvmppc_define_rtas_kernel_token(uint32_t token, const char *function)
2576 struct kvm_rtas_token_args args = {
2580 if (!kvm_check_extension(kvm_state, KVM_CAP_PPC_RTAS)) {
2584 strncpy(args.name, function, sizeof(args.name));
2586 return kvm_vm_ioctl(kvm_state, KVM_PPC_RTAS_DEFINE_TOKEN, &args);
2589 int kvmppc_get_htab_fd(bool write, uint64_t index, Error **errp)
2591 struct kvm_get_htab_fd s = {
2592 .flags = write ? KVM_GET_HTAB_WRITE : 0,
2593 .start_index = index,
2598 error_setg(errp, "KVM version doesn't support %s the HPT",
2599 write ? "writing" : "reading");
2603 ret = kvm_vm_ioctl(kvm_state, KVM_PPC_GET_HTAB_FD, &s);
2605 error_setg(errp, "Unable to open fd for %s HPT %s KVM: %s",
2606 write ? "writing" : "reading", write ? "to" : "from",
2614 int kvmppc_save_htab(QEMUFile *f, int fd, size_t bufsize, int64_t max_ns)
2616 int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2617 uint8_t buf[bufsize];
2621 rc = read(fd, buf, bufsize);
2623 fprintf(stderr, "Error reading data from KVM HTAB fd: %s\n",
2627 uint8_t *buffer = buf;
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;
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);
2641 buffer += chunksize;
2647 || ((qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) < max_ns)));
2649 return (rc == 0) ? 1 : 0;
2652 int kvmppc_load_htab_chunk(QEMUFile *f, int fd, uint32_t index,
2653 uint16_t n_valid, uint16_t n_invalid)
2655 struct kvm_get_htab_header *buf;
2656 size_t chunksize = sizeof(*buf) + n_valid*HASH_PTE_SIZE_64;
2659 buf = alloca(chunksize);
2661 buf->n_valid = n_valid;
2662 buf->n_invalid = n_invalid;
2664 qemu_get_buffer(f, (void *)(buf + 1), HASH_PTE_SIZE_64*n_valid);
2666 rc = write(fd, buf, chunksize);
2668 fprintf(stderr, "Error writing KVM hash table: %s\n",
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");
2680 bool kvm_arch_stop_on_emulation_error(CPUState *cpu)
2685 void kvm_arch_init_irq_routing(KVMState *s)
2689 void kvmppc_read_hptes(ppc_hash_pte64_t *hptes, hwaddr ptex, int n)
2694 fd = kvmppc_get_htab_fd(false, ptex, &error_abort);
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];
2702 rc = read(fd, buf, sizeof(buf));
2704 hw_error("kvmppc_read_hptes: Unable to read HPTEs");
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;
2711 if (hdr->index != (ptex + i)) {
2712 hw_error("kvmppc_read_hptes: Unexpected HPTE index %"PRIu32
2713 " != (%"HWADDR_PRIu" + %d", hdr->index, ptex, i);
2716 if (n - i < valid) {
2719 memcpy(hptes + i, hdr + 1, HASH_PTE_SIZE_64 * valid);
2722 if ((n - i) < invalid) {
2725 memset(hptes + i, 0, invalid * HASH_PTE_SIZE_64);
2728 hdr = (struct kvm_get_htab_header *)
2729 ((char *)(hdr + 1) + HASH_PTE_SIZE_64 * hdr->n_valid);
2736 void kvmppc_write_hpte(hwaddr ptex, uint64_t pte0, uint64_t pte1)
2740 struct kvm_get_htab_header hdr;
2745 fd = kvmppc_get_htab_fd(true, 0 /* Ignored */, &error_abort);
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);
2753 rc = write(fd, &buf, sizeof(buf));
2754 if (rc != sizeof(buf)) {
2755 hw_error("kvmppc_write_hpte: Unable to update KVM HPT");
2760 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
2761 uint64_t address, uint32_t data, PCIDevice *dev)
2766 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
2767 int vector, PCIDevice *dev)
2772 int kvm_arch_release_virq_post(int virq)
2777 int kvm_arch_msi_data_to_gsi(uint32_t data)
2779 return data & 0xffff;
2782 int kvmppc_enable_hwrng(void)
2784 if (!kvm_enabled() || !kvm_check_extension(kvm_state, KVM_CAP_PPC_HWRNG)) {
2788 return kvmppc_enable_hcall(kvm_state, H_RANDOM);
2791 void kvmppc_check_papr_resize_hpt(Error **errp)
2793 if (!kvm_enabled()) {
2794 return; /* No KVM, we're good */
2797 if (cap_resize_hpt) {
2798 return; /* Kernel has explicit support, we're good */
2801 /* Otherwise fallback on looking for PR KVM */
2802 if (kvmppc_is_pr(kvm_state)) {
2807 "Hash page table resizing not available with this KVM version");
2810 int kvmppc_resize_hpt_prepare(PowerPCCPU *cpu, target_ulong flags, int shift)
2812 CPUState *cs = CPU(cpu);
2813 struct kvm_ppc_resize_hpt rhpt = {
2818 if (!cap_resize_hpt) {
2822 return kvm_vm_ioctl(cs->kvm_state, KVM_PPC_RESIZE_HPT_PREPARE, &rhpt);
2825 int kvmppc_resize_hpt_commit(PowerPCCPU *cpu, target_ulong flags, int shift)
2827 CPUState *cs = CPU(cpu);
2828 struct kvm_ppc_resize_hpt rhpt = {
2833 if (!cap_resize_hpt) {
2837 return kvm_vm_ioctl(cs->kvm_state, KVM_PPC_RESIZE_HPT_COMMIT, &rhpt);
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.
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
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
2855 bool kvmppc_pvr_workaround_required(PowerPCCPU *cpu)
2857 CPUState *cs = CPU(cpu);
2859 if (!kvm_enabled()) {
2863 if (cap_ppc_pvr_compat) {
2867 return !kvmppc_is_pr(cs->kvm_state);