2 * PowerPC implementation of KVM hooks
4 * Copyright IBM Corp. 2007
5 * Copyright (C) 2011 Freescale Semiconductor, Inc.
8 * Jerone Young <jyoung5@us.ibm.com>
9 * Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
10 * Hollis Blanchard <hollisb@us.ibm.com>
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 "qemu/error-report.h"
27 #include "cpu-models.h"
28 #include "qemu/timer.h"
29 #include "sysemu/sysemu.h"
30 #include "sysemu/hw_accel.h"
32 #include "sysemu/cpus.h"
33 #include "sysemu/device_tree.h"
34 #include "mmu-hash64.h"
36 #include "hw/sysbus.h"
37 #include "hw/ppc/spapr.h"
38 #include "hw/ppc/spapr_vio.h"
39 #include "hw/ppc/spapr_cpu_core.h"
40 #include "hw/ppc/ppc.h"
41 #include "sysemu/watchdog.h"
43 #include "exec/gdbstub.h"
44 #include "exec/memattrs.h"
45 #include "exec/ram_addr.h"
46 #include "sysemu/hostmem.h"
47 #include "qemu/cutils.h"
48 #include "qemu/mmap-alloc.h"
49 #if defined(TARGET_PPC64)
50 #include "hw/ppc/spapr_cpu_core.h"
53 #include "sysemu/kvm_int.h"
58 #define DPRINTF(fmt, ...) \
59 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
61 #define DPRINTF(fmt, ...) \
65 #define PROC_DEVTREE_CPU "/proc/device-tree/cpus/"
67 const KVMCapabilityInfo kvm_arch_required_capabilities
[] = {
71 static int cap_interrupt_unset
= false;
72 static int cap_interrupt_level
= false;
73 static int cap_segstate
;
74 static int cap_booke_sregs
;
75 static int cap_ppc_smt
;
76 static int cap_ppc_rma
;
77 static int cap_spapr_tce
;
78 static int cap_spapr_tce_64
;
79 static int cap_spapr_multitce
;
80 static int cap_spapr_vfio
;
82 static int cap_one_reg
;
84 static int cap_ppc_watchdog
;
86 static int cap_htab_fd
;
87 static int cap_fixup_hcalls
;
88 static int cap_htm
; /* Hardware transactional memory support */
89 static int cap_mmu_radix
;
90 static int cap_mmu_hash_v3
;
92 static uint32_t debug_inst_opcode
;
94 /* XXX We have a race condition where we actually have a level triggered
95 * interrupt, but the infrastructure can't expose that yet, so the guest
96 * takes but ignores it, goes to sleep and never gets notified that there's
97 * still an interrupt pending.
99 * As a quick workaround, let's just wake up again 20 ms after we injected
100 * an interrupt. That way we can assure that we're always reinjecting
101 * interrupts in case the guest swallowed them.
103 static QEMUTimer
*idle_timer
;
105 static void kvm_kick_cpu(void *opaque
)
107 PowerPCCPU
*cpu
= opaque
;
109 qemu_cpu_kick(CPU(cpu
));
112 /* Check whether we are running with KVM-PR (instead of KVM-HV). This
113 * should only be used for fallback tests - generally we should use
114 * explicit capabilities for the features we want, rather than
115 * assuming what is/isn't available depending on the KVM variant. */
116 static bool kvmppc_is_pr(KVMState
*ks
)
118 /* Assume KVM-PR if the GET_PVINFO capability is available */
119 return kvm_check_extension(ks
, KVM_CAP_PPC_GET_PVINFO
) != 0;
122 static int kvm_ppc_register_host_cpu_type(void);
124 int kvm_arch_init(MachineState
*ms
, KVMState
*s
)
126 cap_interrupt_unset
= kvm_check_extension(s
, KVM_CAP_PPC_UNSET_IRQ
);
127 cap_interrupt_level
= kvm_check_extension(s
, KVM_CAP_PPC_IRQ_LEVEL
);
128 cap_segstate
= kvm_check_extension(s
, KVM_CAP_PPC_SEGSTATE
);
129 cap_booke_sregs
= kvm_check_extension(s
, KVM_CAP_PPC_BOOKE_SREGS
);
130 cap_ppc_smt
= kvm_check_extension(s
, KVM_CAP_PPC_SMT
);
131 cap_ppc_rma
= kvm_check_extension(s
, KVM_CAP_PPC_RMA
);
132 cap_spapr_tce
= kvm_check_extension(s
, KVM_CAP_SPAPR_TCE
);
133 cap_spapr_tce_64
= kvm_check_extension(s
, KVM_CAP_SPAPR_TCE_64
);
134 cap_spapr_multitce
= kvm_check_extension(s
, KVM_CAP_SPAPR_MULTITCE
);
135 cap_spapr_vfio
= false;
136 cap_one_reg
= kvm_check_extension(s
, KVM_CAP_ONE_REG
);
137 cap_hior
= kvm_check_extension(s
, KVM_CAP_PPC_HIOR
);
138 cap_epr
= kvm_check_extension(s
, KVM_CAP_PPC_EPR
);
139 cap_ppc_watchdog
= kvm_check_extension(s
, KVM_CAP_PPC_BOOKE_WATCHDOG
);
140 /* Note: we don't set cap_papr here, because this capability is
141 * only activated after this by kvmppc_set_papr() */
142 cap_htab_fd
= kvm_check_extension(s
, KVM_CAP_PPC_HTAB_FD
);
143 cap_fixup_hcalls
= kvm_check_extension(s
, KVM_CAP_PPC_FIXUP_HCALL
);
144 cap_htm
= kvm_vm_check_extension(s
, KVM_CAP_PPC_HTM
);
145 cap_mmu_radix
= kvm_vm_check_extension(s
, KVM_CAP_PPC_MMU_RADIX
);
146 cap_mmu_hash_v3
= kvm_vm_check_extension(s
, KVM_CAP_PPC_MMU_HASH_V3
);
148 if (!cap_interrupt_level
) {
149 fprintf(stderr
, "KVM: Couldn't find level irq capability. Expect the "
150 "VM to stall at times!\n");
153 kvm_ppc_register_host_cpu_type();
158 int kvm_arch_irqchip_create(MachineState
*ms
, KVMState
*s
)
163 static int kvm_arch_sync_sregs(PowerPCCPU
*cpu
)
165 CPUPPCState
*cenv
= &cpu
->env
;
166 CPUState
*cs
= CPU(cpu
);
167 struct kvm_sregs sregs
;
170 if (cenv
->excp_model
== POWERPC_EXCP_BOOKE
) {
171 /* What we're really trying to say is "if we're on BookE, we use
172 the native PVR for now". This is the only sane way to check
173 it though, so we potentially confuse users that they can run
174 BookE guests on BookS. Let's hope nobody dares enough :) */
178 fprintf(stderr
, "kvm error: missing PVR setting capability\n");
183 ret
= kvm_vcpu_ioctl(cs
, KVM_GET_SREGS
, &sregs
);
188 sregs
.pvr
= cenv
->spr
[SPR_PVR
];
189 return kvm_vcpu_ioctl(cs
, KVM_SET_SREGS
, &sregs
);
192 /* Set up a shared TLB array with KVM */
193 static int kvm_booke206_tlb_init(PowerPCCPU
*cpu
)
195 CPUPPCState
*env
= &cpu
->env
;
196 CPUState
*cs
= CPU(cpu
);
197 struct kvm_book3e_206_tlb_params params
= {};
198 struct kvm_config_tlb cfg
= {};
199 unsigned int entries
= 0;
202 if (!kvm_enabled() ||
203 !kvm_check_extension(cs
->kvm_state
, KVM_CAP_SW_TLB
)) {
207 assert(ARRAY_SIZE(params
.tlb_sizes
) == BOOKE206_MAX_TLBN
);
209 for (i
= 0; i
< BOOKE206_MAX_TLBN
; i
++) {
210 params
.tlb_sizes
[i
] = booke206_tlb_size(env
, i
);
211 params
.tlb_ways
[i
] = booke206_tlb_ways(env
, i
);
212 entries
+= params
.tlb_sizes
[i
];
215 assert(entries
== env
->nb_tlb
);
216 assert(sizeof(struct kvm_book3e_206_tlb_entry
) == sizeof(ppcmas_tlb_t
));
218 env
->tlb_dirty
= true;
220 cfg
.array
= (uintptr_t)env
->tlb
.tlbm
;
221 cfg
.array_len
= sizeof(ppcmas_tlb_t
) * entries
;
222 cfg
.params
= (uintptr_t)¶ms
;
223 cfg
.mmu_type
= KVM_MMU_FSL_BOOKE_NOHV
;
225 ret
= kvm_vcpu_enable_cap(cs
, KVM_CAP_SW_TLB
, 0, (uintptr_t)&cfg
);
227 fprintf(stderr
, "%s: couldn't enable KVM_CAP_SW_TLB: %s\n",
228 __func__
, strerror(-ret
));
232 env
->kvm_sw_tlb
= true;
237 #if defined(TARGET_PPC64)
238 static void kvm_get_fallback_smmu_info(PowerPCCPU
*cpu
,
239 struct kvm_ppc_smmu_info
*info
)
241 CPUPPCState
*env
= &cpu
->env
;
242 CPUState
*cs
= CPU(cpu
);
244 memset(info
, 0, sizeof(*info
));
246 /* We don't have the new KVM_PPC_GET_SMMU_INFO ioctl, so
247 * need to "guess" what the supported page sizes are.
249 * For that to work we make a few assumptions:
251 * - Check whether we are running "PR" KVM which only supports 4K
252 * and 16M pages, but supports them regardless of the backing
253 * store characteritics. We also don't support 1T segments.
255 * This is safe as if HV KVM ever supports that capability or PR
256 * KVM grows supports for more page/segment sizes, those versions
257 * will have implemented KVM_CAP_PPC_GET_SMMU_INFO and thus we
258 * will not hit this fallback
260 * - Else we are running HV KVM. This means we only support page
261 * sizes that fit in the backing store. Additionally we only
262 * advertize 64K pages if the processor is ARCH 2.06 and we assume
263 * P7 encodings for the SLB and hash table. Here too, we assume
264 * support for any newer processor will mean a kernel that
265 * implements KVM_CAP_PPC_GET_SMMU_INFO and thus doesn't hit
268 if (kvmppc_is_pr(cs
->kvm_state
)) {
273 /* Standard 4k base page size segment */
274 info
->sps
[0].page_shift
= 12;
275 info
->sps
[0].slb_enc
= 0;
276 info
->sps
[0].enc
[0].page_shift
= 12;
277 info
->sps
[0].enc
[0].pte_enc
= 0;
279 /* Standard 16M large page size segment */
280 info
->sps
[1].page_shift
= 24;
281 info
->sps
[1].slb_enc
= SLB_VSID_L
;
282 info
->sps
[1].enc
[0].page_shift
= 24;
283 info
->sps
[1].enc
[0].pte_enc
= 0;
287 /* HV KVM has backing store size restrictions */
288 info
->flags
= KVM_PPC_PAGE_SIZES_REAL
;
290 if (env
->mmu_model
& POWERPC_MMU_1TSEG
) {
291 info
->flags
|= KVM_PPC_1T_SEGMENTS
;
294 if (POWERPC_MMU_VER(env
->mmu_model
) == POWERPC_MMU_VER_2_06
||
295 POWERPC_MMU_VER(env
->mmu_model
) == POWERPC_MMU_VER_2_07
) {
301 /* Standard 4k base page size segment */
302 info
->sps
[i
].page_shift
= 12;
303 info
->sps
[i
].slb_enc
= 0;
304 info
->sps
[i
].enc
[0].page_shift
= 12;
305 info
->sps
[i
].enc
[0].pte_enc
= 0;
308 /* 64K on MMU 2.06 and later */
309 if (POWERPC_MMU_VER(env
->mmu_model
) == POWERPC_MMU_VER_2_06
||
310 POWERPC_MMU_VER(env
->mmu_model
) == POWERPC_MMU_VER_2_07
) {
311 info
->sps
[i
].page_shift
= 16;
312 info
->sps
[i
].slb_enc
= 0x110;
313 info
->sps
[i
].enc
[0].page_shift
= 16;
314 info
->sps
[i
].enc
[0].pte_enc
= 1;
318 /* Standard 16M large page size segment */
319 info
->sps
[i
].page_shift
= 24;
320 info
->sps
[i
].slb_enc
= SLB_VSID_L
;
321 info
->sps
[i
].enc
[0].page_shift
= 24;
322 info
->sps
[i
].enc
[0].pte_enc
= 0;
326 static void kvm_get_smmu_info(PowerPCCPU
*cpu
, struct kvm_ppc_smmu_info
*info
)
328 CPUState
*cs
= CPU(cpu
);
331 if (kvm_check_extension(cs
->kvm_state
, KVM_CAP_PPC_GET_SMMU_INFO
)) {
332 ret
= kvm_vm_ioctl(cs
->kvm_state
, KVM_PPC_GET_SMMU_INFO
, info
);
338 kvm_get_fallback_smmu_info(cpu
, info
);
341 struct ppc_radix_page_info
*kvm_get_radix_page_info(void)
343 KVMState
*s
= KVM_STATE(current_machine
->accelerator
);
344 struct ppc_radix_page_info
*radix_page_info
;
345 struct kvm_ppc_rmmu_info rmmu_info
;
348 if (!kvm_check_extension(s
, KVM_CAP_PPC_MMU_RADIX
)) {
351 if (kvm_vm_ioctl(s
, KVM_PPC_GET_RMMU_INFO
, &rmmu_info
)) {
354 radix_page_info
= g_malloc0(sizeof(*radix_page_info
));
355 radix_page_info
->count
= 0;
356 for (i
= 0; i
< PPC_PAGE_SIZES_MAX_SZ
; i
++) {
357 if (rmmu_info
.ap_encodings
[i
]) {
358 radix_page_info
->entries
[i
] = rmmu_info
.ap_encodings
[i
];
359 radix_page_info
->count
++;
362 return radix_page_info
;
365 target_ulong
kvmppc_configure_v3_mmu(PowerPCCPU
*cpu
,
366 bool radix
, bool gtse
,
369 CPUState
*cs
= CPU(cpu
);
372 struct kvm_ppc_mmuv3_cfg cfg
= {
373 .process_table
= proc_tbl
,
377 flags
|= KVM_PPC_MMUV3_RADIX
;
380 flags
|= KVM_PPC_MMUV3_GTSE
;
383 ret
= kvm_vm_ioctl(cs
->kvm_state
, KVM_PPC_CONFIGURE_V3_MMU
, &cfg
);
390 return H_NOT_AVAILABLE
;
396 static bool kvm_valid_page_size(uint32_t flags
, long rampgsize
, uint32_t shift
)
398 if (!(flags
& KVM_PPC_PAGE_SIZES_REAL
)) {
402 return (1ul << shift
) <= rampgsize
;
405 static long max_cpu_page_size
;
407 static void kvm_fixup_page_sizes(PowerPCCPU
*cpu
)
409 static struct kvm_ppc_smmu_info smmu_info
;
410 static bool has_smmu_info
;
411 CPUPPCState
*env
= &cpu
->env
;
413 bool has_64k_pages
= false;
415 /* We only handle page sizes for 64-bit server guests for now */
416 if (!(env
->mmu_model
& POWERPC_MMU_64
)) {
420 /* Collect MMU info from kernel if not already */
421 if (!has_smmu_info
) {
422 kvm_get_smmu_info(cpu
, &smmu_info
);
423 has_smmu_info
= true;
426 if (!max_cpu_page_size
) {
427 max_cpu_page_size
= qemu_getrampagesize();
430 /* Convert to QEMU form */
431 memset(&env
->sps
, 0, sizeof(env
->sps
));
433 /* If we have HV KVM, we need to forbid CI large pages if our
434 * host page size is smaller than 64K.
436 if (smmu_info
.flags
& KVM_PPC_PAGE_SIZES_REAL
) {
437 env
->ci_large_pages
= getpagesize() >= 0x10000;
441 * XXX This loop should be an entry wide AND of the capabilities that
442 * the selected CPU has with the capabilities that KVM supports.
444 for (ik
= iq
= 0; ik
< KVM_PPC_PAGE_SIZES_MAX_SZ
; ik
++) {
445 struct ppc_one_seg_page_size
*qsps
= &env
->sps
.sps
[iq
];
446 struct kvm_ppc_one_seg_page_size
*ksps
= &smmu_info
.sps
[ik
];
448 if (!kvm_valid_page_size(smmu_info
.flags
, max_cpu_page_size
,
452 qsps
->page_shift
= ksps
->page_shift
;
453 qsps
->slb_enc
= ksps
->slb_enc
;
454 for (jk
= jq
= 0; jk
< KVM_PPC_PAGE_SIZES_MAX_SZ
; jk
++) {
455 if (!kvm_valid_page_size(smmu_info
.flags
, max_cpu_page_size
,
456 ksps
->enc
[jk
].page_shift
)) {
459 if (ksps
->enc
[jk
].page_shift
== 16) {
460 has_64k_pages
= true;
462 qsps
->enc
[jq
].page_shift
= ksps
->enc
[jk
].page_shift
;
463 qsps
->enc
[jq
].pte_enc
= ksps
->enc
[jk
].pte_enc
;
464 if (++jq
>= PPC_PAGE_SIZES_MAX_SZ
) {
468 if (++iq
>= PPC_PAGE_SIZES_MAX_SZ
) {
472 env
->slb_nr
= smmu_info
.slb_size
;
473 if (!(smmu_info
.flags
& KVM_PPC_1T_SEGMENTS
)) {
474 env
->mmu_model
&= ~POWERPC_MMU_1TSEG
;
476 if (!has_64k_pages
) {
477 env
->mmu_model
&= ~POWERPC_MMU_64K
;
481 bool kvmppc_is_mem_backend_page_size_ok(char *obj_path
)
483 Object
*mem_obj
= object_resolve_path(obj_path
, NULL
);
484 char *mempath
= object_property_get_str(mem_obj
, "mem-path", NULL
);
488 pagesize
= qemu_mempath_getpagesize(mempath
);
490 pagesize
= getpagesize();
493 return pagesize
>= max_cpu_page_size
;
496 #else /* defined (TARGET_PPC64) */
498 static inline void kvm_fixup_page_sizes(PowerPCCPU
*cpu
)
502 bool kvmppc_is_mem_backend_page_size_ok(char *obj_path
)
507 #endif /* !defined (TARGET_PPC64) */
509 unsigned long kvm_arch_vcpu_id(CPUState
*cpu
)
511 return ppc_get_vcpu_dt_id(POWERPC_CPU(cpu
));
514 /* e500 supports 2 h/w breakpoint and 2 watchpoint.
515 * book3s supports only 1 watchpoint, so array size
516 * of 4 is sufficient for now.
518 #define MAX_HW_BKPTS 4
520 static struct HWBreakpoint
{
523 } hw_debug_points
[MAX_HW_BKPTS
];
525 static CPUWatchpoint hw_watchpoint
;
527 /* Default there is no breakpoint and watchpoint supported */
528 static int max_hw_breakpoint
;
529 static int max_hw_watchpoint
;
530 static int nb_hw_breakpoint
;
531 static int nb_hw_watchpoint
;
533 static void kvmppc_hw_debug_points_init(CPUPPCState
*cenv
)
535 if (cenv
->excp_model
== POWERPC_EXCP_BOOKE
) {
536 max_hw_breakpoint
= 2;
537 max_hw_watchpoint
= 2;
540 if ((max_hw_breakpoint
+ max_hw_watchpoint
) > MAX_HW_BKPTS
) {
541 fprintf(stderr
, "Error initializing h/w breakpoints\n");
546 int kvm_arch_init_vcpu(CPUState
*cs
)
548 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
549 CPUPPCState
*cenv
= &cpu
->env
;
552 /* Gather server mmu info from KVM and update the CPU state */
553 kvm_fixup_page_sizes(cpu
);
555 /* Synchronize sregs with kvm */
556 ret
= kvm_arch_sync_sregs(cpu
);
558 if (ret
== -EINVAL
) {
559 error_report("Register sync failed... If you're using kvm-hv.ko,"
560 " only \"-cpu host\" is possible");
565 idle_timer
= timer_new_ns(QEMU_CLOCK_VIRTUAL
, kvm_kick_cpu
, cpu
);
567 switch (cenv
->mmu_model
) {
568 case POWERPC_MMU_BOOKE206
:
569 /* This target supports access to KVM's guest TLB */
570 ret
= kvm_booke206_tlb_init(cpu
);
572 case POWERPC_MMU_2_07
:
573 if (!cap_htm
&& !kvmppc_is_pr(cs
->kvm_state
)) {
574 /* KVM-HV has transactional memory on POWER8 also without the
575 * KVM_CAP_PPC_HTM extension, so enable it here instead as
576 * long as it's availble to userspace on the host. */
577 if (qemu_getauxval(AT_HWCAP2
) & PPC_FEATURE2_HAS_HTM
) {
586 kvm_get_one_reg(cs
, KVM_REG_PPC_DEBUG_INST
, &debug_inst_opcode
);
587 kvmppc_hw_debug_points_init(cenv
);
592 static void kvm_sw_tlb_put(PowerPCCPU
*cpu
)
594 CPUPPCState
*env
= &cpu
->env
;
595 CPUState
*cs
= CPU(cpu
);
596 struct kvm_dirty_tlb dirty_tlb
;
597 unsigned char *bitmap
;
600 if (!env
->kvm_sw_tlb
) {
604 bitmap
= g_malloc((env
->nb_tlb
+ 7) / 8);
605 memset(bitmap
, 0xFF, (env
->nb_tlb
+ 7) / 8);
607 dirty_tlb
.bitmap
= (uintptr_t)bitmap
;
608 dirty_tlb
.num_dirty
= env
->nb_tlb
;
610 ret
= kvm_vcpu_ioctl(cs
, KVM_DIRTY_TLB
, &dirty_tlb
);
612 fprintf(stderr
, "%s: KVM_DIRTY_TLB: %s\n",
613 __func__
, strerror(-ret
));
619 static void kvm_get_one_spr(CPUState
*cs
, uint64_t id
, int spr
)
621 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
622 CPUPPCState
*env
= &cpu
->env
;
627 struct kvm_one_reg reg
= {
629 .addr
= (uintptr_t) &val
,
633 ret
= kvm_vcpu_ioctl(cs
, KVM_GET_ONE_REG
, ®
);
635 trace_kvm_failed_spr_get(spr
, strerror(errno
));
637 switch (id
& KVM_REG_SIZE_MASK
) {
638 case KVM_REG_SIZE_U32
:
639 env
->spr
[spr
] = val
.u32
;
642 case KVM_REG_SIZE_U64
:
643 env
->spr
[spr
] = val
.u64
;
647 /* Don't handle this size yet */
653 static void kvm_put_one_spr(CPUState
*cs
, uint64_t id
, int spr
)
655 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
656 CPUPPCState
*env
= &cpu
->env
;
661 struct kvm_one_reg reg
= {
663 .addr
= (uintptr_t) &val
,
667 switch (id
& KVM_REG_SIZE_MASK
) {
668 case KVM_REG_SIZE_U32
:
669 val
.u32
= env
->spr
[spr
];
672 case KVM_REG_SIZE_U64
:
673 val
.u64
= env
->spr
[spr
];
677 /* Don't handle this size yet */
681 ret
= kvm_vcpu_ioctl(cs
, KVM_SET_ONE_REG
, ®
);
683 trace_kvm_failed_spr_set(spr
, strerror(errno
));
687 static int kvm_put_fp(CPUState
*cs
)
689 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
690 CPUPPCState
*env
= &cpu
->env
;
691 struct kvm_one_reg reg
;
695 if (env
->insns_flags
& PPC_FLOAT
) {
696 uint64_t fpscr
= env
->fpscr
;
697 bool vsx
= !!(env
->insns_flags2
& PPC2_VSX
);
699 reg
.id
= KVM_REG_PPC_FPSCR
;
700 reg
.addr
= (uintptr_t)&fpscr
;
701 ret
= kvm_vcpu_ioctl(cs
, KVM_SET_ONE_REG
, ®
);
703 DPRINTF("Unable to set FPSCR to KVM: %s\n", strerror(errno
));
707 for (i
= 0; i
< 32; i
++) {
710 #ifdef HOST_WORDS_BIGENDIAN
711 vsr
[0] = float64_val(env
->fpr
[i
]);
712 vsr
[1] = env
->vsr
[i
];
714 vsr
[0] = env
->vsr
[i
];
715 vsr
[1] = float64_val(env
->fpr
[i
]);
717 reg
.addr
= (uintptr_t) &vsr
;
718 reg
.id
= vsx
? KVM_REG_PPC_VSR(i
) : KVM_REG_PPC_FPR(i
);
720 ret
= kvm_vcpu_ioctl(cs
, KVM_SET_ONE_REG
, ®
);
722 DPRINTF("Unable to set %s%d to KVM: %s\n", vsx
? "VSR" : "FPR",
729 if (env
->insns_flags
& PPC_ALTIVEC
) {
730 reg
.id
= KVM_REG_PPC_VSCR
;
731 reg
.addr
= (uintptr_t)&env
->vscr
;
732 ret
= kvm_vcpu_ioctl(cs
, KVM_SET_ONE_REG
, ®
);
734 DPRINTF("Unable to set VSCR to KVM: %s\n", strerror(errno
));
738 for (i
= 0; i
< 32; i
++) {
739 reg
.id
= KVM_REG_PPC_VR(i
);
740 reg
.addr
= (uintptr_t)&env
->avr
[i
];
741 ret
= kvm_vcpu_ioctl(cs
, KVM_SET_ONE_REG
, ®
);
743 DPRINTF("Unable to set VR%d to KVM: %s\n", i
, strerror(errno
));
752 static int kvm_get_fp(CPUState
*cs
)
754 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
755 CPUPPCState
*env
= &cpu
->env
;
756 struct kvm_one_reg reg
;
760 if (env
->insns_flags
& PPC_FLOAT
) {
762 bool vsx
= !!(env
->insns_flags2
& PPC2_VSX
);
764 reg
.id
= KVM_REG_PPC_FPSCR
;
765 reg
.addr
= (uintptr_t)&fpscr
;
766 ret
= kvm_vcpu_ioctl(cs
, KVM_GET_ONE_REG
, ®
);
768 DPRINTF("Unable to get FPSCR from KVM: %s\n", strerror(errno
));
774 for (i
= 0; i
< 32; i
++) {
777 reg
.addr
= (uintptr_t) &vsr
;
778 reg
.id
= vsx
? KVM_REG_PPC_VSR(i
) : KVM_REG_PPC_FPR(i
);
780 ret
= kvm_vcpu_ioctl(cs
, KVM_GET_ONE_REG
, ®
);
782 DPRINTF("Unable to get %s%d from KVM: %s\n",
783 vsx
? "VSR" : "FPR", i
, strerror(errno
));
786 #ifdef HOST_WORDS_BIGENDIAN
787 env
->fpr
[i
] = vsr
[0];
789 env
->vsr
[i
] = vsr
[1];
792 env
->fpr
[i
] = vsr
[1];
794 env
->vsr
[i
] = vsr
[0];
801 if (env
->insns_flags
& PPC_ALTIVEC
) {
802 reg
.id
= KVM_REG_PPC_VSCR
;
803 reg
.addr
= (uintptr_t)&env
->vscr
;
804 ret
= kvm_vcpu_ioctl(cs
, KVM_GET_ONE_REG
, ®
);
806 DPRINTF("Unable to get VSCR from KVM: %s\n", strerror(errno
));
810 for (i
= 0; i
< 32; i
++) {
811 reg
.id
= KVM_REG_PPC_VR(i
);
812 reg
.addr
= (uintptr_t)&env
->avr
[i
];
813 ret
= kvm_vcpu_ioctl(cs
, KVM_GET_ONE_REG
, ®
);
815 DPRINTF("Unable to get VR%d from KVM: %s\n",
825 #if defined(TARGET_PPC64)
826 static int kvm_get_vpa(CPUState
*cs
)
828 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
829 CPUPPCState
*env
= &cpu
->env
;
830 struct kvm_one_reg reg
;
833 reg
.id
= KVM_REG_PPC_VPA_ADDR
;
834 reg
.addr
= (uintptr_t)&env
->vpa_addr
;
835 ret
= kvm_vcpu_ioctl(cs
, KVM_GET_ONE_REG
, ®
);
837 DPRINTF("Unable to get VPA address from KVM: %s\n", strerror(errno
));
841 assert((uintptr_t)&env
->slb_shadow_size
842 == ((uintptr_t)&env
->slb_shadow_addr
+ 8));
843 reg
.id
= KVM_REG_PPC_VPA_SLB
;
844 reg
.addr
= (uintptr_t)&env
->slb_shadow_addr
;
845 ret
= kvm_vcpu_ioctl(cs
, KVM_GET_ONE_REG
, ®
);
847 DPRINTF("Unable to get SLB shadow state from KVM: %s\n",
852 assert((uintptr_t)&env
->dtl_size
== ((uintptr_t)&env
->dtl_addr
+ 8));
853 reg
.id
= KVM_REG_PPC_VPA_DTL
;
854 reg
.addr
= (uintptr_t)&env
->dtl_addr
;
855 ret
= kvm_vcpu_ioctl(cs
, KVM_GET_ONE_REG
, ®
);
857 DPRINTF("Unable to get dispatch trace log state from KVM: %s\n",
865 static int kvm_put_vpa(CPUState
*cs
)
867 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
868 CPUPPCState
*env
= &cpu
->env
;
869 struct kvm_one_reg reg
;
872 /* SLB shadow or DTL can't be registered unless a master VPA is
873 * registered. That means when restoring state, if a VPA *is*
874 * registered, we need to set that up first. If not, we need to
875 * deregister the others before deregistering the master VPA */
876 assert(env
->vpa_addr
|| !(env
->slb_shadow_addr
|| env
->dtl_addr
));
879 reg
.id
= KVM_REG_PPC_VPA_ADDR
;
880 reg
.addr
= (uintptr_t)&env
->vpa_addr
;
881 ret
= kvm_vcpu_ioctl(cs
, KVM_SET_ONE_REG
, ®
);
883 DPRINTF("Unable to set VPA address to KVM: %s\n", strerror(errno
));
888 assert((uintptr_t)&env
->slb_shadow_size
889 == ((uintptr_t)&env
->slb_shadow_addr
+ 8));
890 reg
.id
= KVM_REG_PPC_VPA_SLB
;
891 reg
.addr
= (uintptr_t)&env
->slb_shadow_addr
;
892 ret
= kvm_vcpu_ioctl(cs
, KVM_SET_ONE_REG
, ®
);
894 DPRINTF("Unable to set SLB shadow state to KVM: %s\n", strerror(errno
));
898 assert((uintptr_t)&env
->dtl_size
== ((uintptr_t)&env
->dtl_addr
+ 8));
899 reg
.id
= KVM_REG_PPC_VPA_DTL
;
900 reg
.addr
= (uintptr_t)&env
->dtl_addr
;
901 ret
= kvm_vcpu_ioctl(cs
, KVM_SET_ONE_REG
, ®
);
903 DPRINTF("Unable to set dispatch trace log state to KVM: %s\n",
908 if (!env
->vpa_addr
) {
909 reg
.id
= KVM_REG_PPC_VPA_ADDR
;
910 reg
.addr
= (uintptr_t)&env
->vpa_addr
;
911 ret
= kvm_vcpu_ioctl(cs
, KVM_SET_ONE_REG
, ®
);
913 DPRINTF("Unable to set VPA address to KVM: %s\n", strerror(errno
));
920 #endif /* TARGET_PPC64 */
922 int kvmppc_put_books_sregs(PowerPCCPU
*cpu
)
924 CPUPPCState
*env
= &cpu
->env
;
925 struct kvm_sregs sregs
;
928 sregs
.pvr
= env
->spr
[SPR_PVR
];
930 sregs
.u
.s
.sdr1
= env
->spr
[SPR_SDR1
];
934 for (i
= 0; i
< ARRAY_SIZE(env
->slb
); i
++) {
935 sregs
.u
.s
.ppc64
.slb
[i
].slbe
= env
->slb
[i
].esid
;
936 if (env
->slb
[i
].esid
& SLB_ESID_V
) {
937 sregs
.u
.s
.ppc64
.slb
[i
].slbe
|= i
;
939 sregs
.u
.s
.ppc64
.slb
[i
].slbv
= env
->slb
[i
].vsid
;
944 for (i
= 0; i
< 16; i
++) {
945 sregs
.u
.s
.ppc32
.sr
[i
] = env
->sr
[i
];
949 for (i
= 0; i
< 8; i
++) {
950 /* Beware. We have to swap upper and lower bits here */
951 sregs
.u
.s
.ppc32
.dbat
[i
] = ((uint64_t)env
->DBAT
[0][i
] << 32)
953 sregs
.u
.s
.ppc32
.ibat
[i
] = ((uint64_t)env
->IBAT
[0][i
] << 32)
957 return kvm_vcpu_ioctl(CPU(cpu
), KVM_SET_SREGS
, &sregs
);
960 int kvm_arch_put_registers(CPUState
*cs
, int level
)
962 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
963 CPUPPCState
*env
= &cpu
->env
;
964 struct kvm_regs regs
;
968 ret
= kvm_vcpu_ioctl(cs
, KVM_GET_REGS
, ®s
);
975 regs
.xer
= cpu_read_xer(env
);
979 regs
.srr0
= env
->spr
[SPR_SRR0
];
980 regs
.srr1
= env
->spr
[SPR_SRR1
];
982 regs
.sprg0
= env
->spr
[SPR_SPRG0
];
983 regs
.sprg1
= env
->spr
[SPR_SPRG1
];
984 regs
.sprg2
= env
->spr
[SPR_SPRG2
];
985 regs
.sprg3
= env
->spr
[SPR_SPRG3
];
986 regs
.sprg4
= env
->spr
[SPR_SPRG4
];
987 regs
.sprg5
= env
->spr
[SPR_SPRG5
];
988 regs
.sprg6
= env
->spr
[SPR_SPRG6
];
989 regs
.sprg7
= env
->spr
[SPR_SPRG7
];
991 regs
.pid
= env
->spr
[SPR_BOOKE_PID
];
993 for (i
= 0;i
< 32; i
++)
994 regs
.gpr
[i
] = env
->gpr
[i
];
997 for (i
= 0; i
< 8; i
++) {
998 regs
.cr
|= (env
->crf
[i
] & 15) << (4 * (7 - i
));
1001 ret
= kvm_vcpu_ioctl(cs
, KVM_SET_REGS
, ®s
);
1007 if (env
->tlb_dirty
) {
1008 kvm_sw_tlb_put(cpu
);
1009 env
->tlb_dirty
= false;
1012 if (cap_segstate
&& (level
>= KVM_PUT_RESET_STATE
)) {
1013 ret
= kvmppc_put_books_sregs(cpu
);
1019 if (cap_hior
&& (level
>= KVM_PUT_RESET_STATE
)) {
1020 kvm_put_one_spr(cs
, KVM_REG_PPC_HIOR
, SPR_HIOR
);
1026 /* We deliberately ignore errors here, for kernels which have
1027 * the ONE_REG calls, but don't support the specific
1028 * registers, there's a reasonable chance things will still
1029 * work, at least until we try to migrate. */
1030 for (i
= 0; i
< 1024; i
++) {
1031 uint64_t id
= env
->spr_cb
[i
].one_reg_id
;
1034 kvm_put_one_spr(cs
, id
, i
);
1040 for (i
= 0; i
< ARRAY_SIZE(env
->tm_gpr
); i
++) {
1041 kvm_set_one_reg(cs
, KVM_REG_PPC_TM_GPR(i
), &env
->tm_gpr
[i
]);
1043 for (i
= 0; i
< ARRAY_SIZE(env
->tm_vsr
); i
++) {
1044 kvm_set_one_reg(cs
, KVM_REG_PPC_TM_VSR(i
), &env
->tm_vsr
[i
]);
1046 kvm_set_one_reg(cs
, KVM_REG_PPC_TM_CR
, &env
->tm_cr
);
1047 kvm_set_one_reg(cs
, KVM_REG_PPC_TM_LR
, &env
->tm_lr
);
1048 kvm_set_one_reg(cs
, KVM_REG_PPC_TM_CTR
, &env
->tm_ctr
);
1049 kvm_set_one_reg(cs
, KVM_REG_PPC_TM_FPSCR
, &env
->tm_fpscr
);
1050 kvm_set_one_reg(cs
, KVM_REG_PPC_TM_AMR
, &env
->tm_amr
);
1051 kvm_set_one_reg(cs
, KVM_REG_PPC_TM_PPR
, &env
->tm_ppr
);
1052 kvm_set_one_reg(cs
, KVM_REG_PPC_TM_VRSAVE
, &env
->tm_vrsave
);
1053 kvm_set_one_reg(cs
, KVM_REG_PPC_TM_VSCR
, &env
->tm_vscr
);
1054 kvm_set_one_reg(cs
, KVM_REG_PPC_TM_DSCR
, &env
->tm_dscr
);
1055 kvm_set_one_reg(cs
, KVM_REG_PPC_TM_TAR
, &env
->tm_tar
);
1059 if (kvm_put_vpa(cs
) < 0) {
1060 DPRINTF("Warning: Unable to set VPA information to KVM\n");
1064 kvm_set_one_reg(cs
, KVM_REG_PPC_TB_OFFSET
, &env
->tb_env
->tb_offset
);
1065 #endif /* TARGET_PPC64 */
1071 static void kvm_sync_excp(CPUPPCState
*env
, int vector
, int ivor
)
1073 env
->excp_vectors
[vector
] = env
->spr
[ivor
] + env
->spr
[SPR_BOOKE_IVPR
];
1076 static int kvmppc_get_booke_sregs(PowerPCCPU
*cpu
)
1078 CPUPPCState
*env
= &cpu
->env
;
1079 struct kvm_sregs sregs
;
1082 ret
= kvm_vcpu_ioctl(CPU(cpu
), KVM_GET_SREGS
, &sregs
);
1087 if (sregs
.u
.e
.features
& KVM_SREGS_E_BASE
) {
1088 env
->spr
[SPR_BOOKE_CSRR0
] = sregs
.u
.e
.csrr0
;
1089 env
->spr
[SPR_BOOKE_CSRR1
] = sregs
.u
.e
.csrr1
;
1090 env
->spr
[SPR_BOOKE_ESR
] = sregs
.u
.e
.esr
;
1091 env
->spr
[SPR_BOOKE_DEAR
] = sregs
.u
.e
.dear
;
1092 env
->spr
[SPR_BOOKE_MCSR
] = sregs
.u
.e
.mcsr
;
1093 env
->spr
[SPR_BOOKE_TSR
] = sregs
.u
.e
.tsr
;
1094 env
->spr
[SPR_BOOKE_TCR
] = sregs
.u
.e
.tcr
;
1095 env
->spr
[SPR_DECR
] = sregs
.u
.e
.dec
;
1096 env
->spr
[SPR_TBL
] = sregs
.u
.e
.tb
& 0xffffffff;
1097 env
->spr
[SPR_TBU
] = sregs
.u
.e
.tb
>> 32;
1098 env
->spr
[SPR_VRSAVE
] = sregs
.u
.e
.vrsave
;
1101 if (sregs
.u
.e
.features
& KVM_SREGS_E_ARCH206
) {
1102 env
->spr
[SPR_BOOKE_PIR
] = sregs
.u
.e
.pir
;
1103 env
->spr
[SPR_BOOKE_MCSRR0
] = sregs
.u
.e
.mcsrr0
;
1104 env
->spr
[SPR_BOOKE_MCSRR1
] = sregs
.u
.e
.mcsrr1
;
1105 env
->spr
[SPR_BOOKE_DECAR
] = sregs
.u
.e
.decar
;
1106 env
->spr
[SPR_BOOKE_IVPR
] = sregs
.u
.e
.ivpr
;
1109 if (sregs
.u
.e
.features
& KVM_SREGS_E_64
) {
1110 env
->spr
[SPR_BOOKE_EPCR
] = sregs
.u
.e
.epcr
;
1113 if (sregs
.u
.e
.features
& KVM_SREGS_E_SPRG8
) {
1114 env
->spr
[SPR_BOOKE_SPRG8
] = sregs
.u
.e
.sprg8
;
1117 if (sregs
.u
.e
.features
& KVM_SREGS_E_IVOR
) {
1118 env
->spr
[SPR_BOOKE_IVOR0
] = sregs
.u
.e
.ivor_low
[0];
1119 kvm_sync_excp(env
, POWERPC_EXCP_CRITICAL
, SPR_BOOKE_IVOR0
);
1120 env
->spr
[SPR_BOOKE_IVOR1
] = sregs
.u
.e
.ivor_low
[1];
1121 kvm_sync_excp(env
, POWERPC_EXCP_MCHECK
, SPR_BOOKE_IVOR1
);
1122 env
->spr
[SPR_BOOKE_IVOR2
] = sregs
.u
.e
.ivor_low
[2];
1123 kvm_sync_excp(env
, POWERPC_EXCP_DSI
, SPR_BOOKE_IVOR2
);
1124 env
->spr
[SPR_BOOKE_IVOR3
] = sregs
.u
.e
.ivor_low
[3];
1125 kvm_sync_excp(env
, POWERPC_EXCP_ISI
, SPR_BOOKE_IVOR3
);
1126 env
->spr
[SPR_BOOKE_IVOR4
] = sregs
.u
.e
.ivor_low
[4];
1127 kvm_sync_excp(env
, POWERPC_EXCP_EXTERNAL
, SPR_BOOKE_IVOR4
);
1128 env
->spr
[SPR_BOOKE_IVOR5
] = sregs
.u
.e
.ivor_low
[5];
1129 kvm_sync_excp(env
, POWERPC_EXCP_ALIGN
, SPR_BOOKE_IVOR5
);
1130 env
->spr
[SPR_BOOKE_IVOR6
] = sregs
.u
.e
.ivor_low
[6];
1131 kvm_sync_excp(env
, POWERPC_EXCP_PROGRAM
, SPR_BOOKE_IVOR6
);
1132 env
->spr
[SPR_BOOKE_IVOR7
] = sregs
.u
.e
.ivor_low
[7];
1133 kvm_sync_excp(env
, POWERPC_EXCP_FPU
, SPR_BOOKE_IVOR7
);
1134 env
->spr
[SPR_BOOKE_IVOR8
] = sregs
.u
.e
.ivor_low
[8];
1135 kvm_sync_excp(env
, POWERPC_EXCP_SYSCALL
, SPR_BOOKE_IVOR8
);
1136 env
->spr
[SPR_BOOKE_IVOR9
] = sregs
.u
.e
.ivor_low
[9];
1137 kvm_sync_excp(env
, POWERPC_EXCP_APU
, SPR_BOOKE_IVOR9
);
1138 env
->spr
[SPR_BOOKE_IVOR10
] = sregs
.u
.e
.ivor_low
[10];
1139 kvm_sync_excp(env
, POWERPC_EXCP_DECR
, SPR_BOOKE_IVOR10
);
1140 env
->spr
[SPR_BOOKE_IVOR11
] = sregs
.u
.e
.ivor_low
[11];
1141 kvm_sync_excp(env
, POWERPC_EXCP_FIT
, SPR_BOOKE_IVOR11
);
1142 env
->spr
[SPR_BOOKE_IVOR12
] = sregs
.u
.e
.ivor_low
[12];
1143 kvm_sync_excp(env
, POWERPC_EXCP_WDT
, SPR_BOOKE_IVOR12
);
1144 env
->spr
[SPR_BOOKE_IVOR13
] = sregs
.u
.e
.ivor_low
[13];
1145 kvm_sync_excp(env
, POWERPC_EXCP_DTLB
, SPR_BOOKE_IVOR13
);
1146 env
->spr
[SPR_BOOKE_IVOR14
] = sregs
.u
.e
.ivor_low
[14];
1147 kvm_sync_excp(env
, POWERPC_EXCP_ITLB
, SPR_BOOKE_IVOR14
);
1148 env
->spr
[SPR_BOOKE_IVOR15
] = sregs
.u
.e
.ivor_low
[15];
1149 kvm_sync_excp(env
, POWERPC_EXCP_DEBUG
, SPR_BOOKE_IVOR15
);
1151 if (sregs
.u
.e
.features
& KVM_SREGS_E_SPE
) {
1152 env
->spr
[SPR_BOOKE_IVOR32
] = sregs
.u
.e
.ivor_high
[0];
1153 kvm_sync_excp(env
, POWERPC_EXCP_SPEU
, SPR_BOOKE_IVOR32
);
1154 env
->spr
[SPR_BOOKE_IVOR33
] = sregs
.u
.e
.ivor_high
[1];
1155 kvm_sync_excp(env
, POWERPC_EXCP_EFPDI
, SPR_BOOKE_IVOR33
);
1156 env
->spr
[SPR_BOOKE_IVOR34
] = sregs
.u
.e
.ivor_high
[2];
1157 kvm_sync_excp(env
, POWERPC_EXCP_EFPRI
, SPR_BOOKE_IVOR34
);
1160 if (sregs
.u
.e
.features
& KVM_SREGS_E_PM
) {
1161 env
->spr
[SPR_BOOKE_IVOR35
] = sregs
.u
.e
.ivor_high
[3];
1162 kvm_sync_excp(env
, POWERPC_EXCP_EPERFM
, SPR_BOOKE_IVOR35
);
1165 if (sregs
.u
.e
.features
& KVM_SREGS_E_PC
) {
1166 env
->spr
[SPR_BOOKE_IVOR36
] = sregs
.u
.e
.ivor_high
[4];
1167 kvm_sync_excp(env
, POWERPC_EXCP_DOORI
, SPR_BOOKE_IVOR36
);
1168 env
->spr
[SPR_BOOKE_IVOR37
] = sregs
.u
.e
.ivor_high
[5];
1169 kvm_sync_excp(env
, POWERPC_EXCP_DOORCI
, SPR_BOOKE_IVOR37
);
1173 if (sregs
.u
.e
.features
& KVM_SREGS_E_ARCH206_MMU
) {
1174 env
->spr
[SPR_BOOKE_MAS0
] = sregs
.u
.e
.mas0
;
1175 env
->spr
[SPR_BOOKE_MAS1
] = sregs
.u
.e
.mas1
;
1176 env
->spr
[SPR_BOOKE_MAS2
] = sregs
.u
.e
.mas2
;
1177 env
->spr
[SPR_BOOKE_MAS3
] = sregs
.u
.e
.mas7_3
& 0xffffffff;
1178 env
->spr
[SPR_BOOKE_MAS4
] = sregs
.u
.e
.mas4
;
1179 env
->spr
[SPR_BOOKE_MAS6
] = sregs
.u
.e
.mas6
;
1180 env
->spr
[SPR_BOOKE_MAS7
] = sregs
.u
.e
.mas7_3
>> 32;
1181 env
->spr
[SPR_MMUCFG
] = sregs
.u
.e
.mmucfg
;
1182 env
->spr
[SPR_BOOKE_TLB0CFG
] = sregs
.u
.e
.tlbcfg
[0];
1183 env
->spr
[SPR_BOOKE_TLB1CFG
] = sregs
.u
.e
.tlbcfg
[1];
1186 if (sregs
.u
.e
.features
& KVM_SREGS_EXP
) {
1187 env
->spr
[SPR_BOOKE_EPR
] = sregs
.u
.e
.epr
;
1190 if (sregs
.u
.e
.features
& KVM_SREGS_E_PD
) {
1191 env
->spr
[SPR_BOOKE_EPLC
] = sregs
.u
.e
.eplc
;
1192 env
->spr
[SPR_BOOKE_EPSC
] = sregs
.u
.e
.epsc
;
1195 if (sregs
.u
.e
.impl_id
== KVM_SREGS_E_IMPL_FSL
) {
1196 env
->spr
[SPR_E500_SVR
] = sregs
.u
.e
.impl
.fsl
.svr
;
1197 env
->spr
[SPR_Exxx_MCAR
] = sregs
.u
.e
.impl
.fsl
.mcar
;
1198 env
->spr
[SPR_HID0
] = sregs
.u
.e
.impl
.fsl
.hid0
;
1200 if (sregs
.u
.e
.impl
.fsl
.features
& KVM_SREGS_E_FSL_PIDn
) {
1201 env
->spr
[SPR_BOOKE_PID1
] = sregs
.u
.e
.impl
.fsl
.pid1
;
1202 env
->spr
[SPR_BOOKE_PID2
] = sregs
.u
.e
.impl
.fsl
.pid2
;
1209 static int kvmppc_get_books_sregs(PowerPCCPU
*cpu
)
1211 CPUPPCState
*env
= &cpu
->env
;
1212 struct kvm_sregs sregs
;
1216 ret
= kvm_vcpu_ioctl(CPU(cpu
), KVM_GET_SREGS
, &sregs
);
1222 ppc_store_sdr1(env
, sregs
.u
.s
.sdr1
);
1228 * The packed SLB array we get from KVM_GET_SREGS only contains
1229 * information about valid entries. So we flush our internal copy
1230 * to get rid of stale ones, then put all valid SLB entries back
1233 memset(env
->slb
, 0, sizeof(env
->slb
));
1234 for (i
= 0; i
< ARRAY_SIZE(env
->slb
); i
++) {
1235 target_ulong rb
= sregs
.u
.s
.ppc64
.slb
[i
].slbe
;
1236 target_ulong rs
= sregs
.u
.s
.ppc64
.slb
[i
].slbv
;
1238 * Only restore valid entries
1240 if (rb
& SLB_ESID_V
) {
1241 ppc_store_slb(cpu
, rb
& 0xfff, rb
& ~0xfffULL
, rs
);
1247 for (i
= 0; i
< 16; i
++) {
1248 env
->sr
[i
] = sregs
.u
.s
.ppc32
.sr
[i
];
1252 for (i
= 0; i
< 8; i
++) {
1253 env
->DBAT
[0][i
] = sregs
.u
.s
.ppc32
.dbat
[i
] & 0xffffffff;
1254 env
->DBAT
[1][i
] = sregs
.u
.s
.ppc32
.dbat
[i
] >> 32;
1255 env
->IBAT
[0][i
] = sregs
.u
.s
.ppc32
.ibat
[i
] & 0xffffffff;
1256 env
->IBAT
[1][i
] = sregs
.u
.s
.ppc32
.ibat
[i
] >> 32;
1262 int kvm_arch_get_registers(CPUState
*cs
)
1264 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
1265 CPUPPCState
*env
= &cpu
->env
;
1266 struct kvm_regs regs
;
1270 ret
= kvm_vcpu_ioctl(cs
, KVM_GET_REGS
, ®s
);
1275 for (i
= 7; i
>= 0; i
--) {
1276 env
->crf
[i
] = cr
& 15;
1280 env
->ctr
= regs
.ctr
;
1282 cpu_write_xer(env
, regs
.xer
);
1283 env
->msr
= regs
.msr
;
1286 env
->spr
[SPR_SRR0
] = regs
.srr0
;
1287 env
->spr
[SPR_SRR1
] = regs
.srr1
;
1289 env
->spr
[SPR_SPRG0
] = regs
.sprg0
;
1290 env
->spr
[SPR_SPRG1
] = regs
.sprg1
;
1291 env
->spr
[SPR_SPRG2
] = regs
.sprg2
;
1292 env
->spr
[SPR_SPRG3
] = regs
.sprg3
;
1293 env
->spr
[SPR_SPRG4
] = regs
.sprg4
;
1294 env
->spr
[SPR_SPRG5
] = regs
.sprg5
;
1295 env
->spr
[SPR_SPRG6
] = regs
.sprg6
;
1296 env
->spr
[SPR_SPRG7
] = regs
.sprg7
;
1298 env
->spr
[SPR_BOOKE_PID
] = regs
.pid
;
1300 for (i
= 0;i
< 32; i
++)
1301 env
->gpr
[i
] = regs
.gpr
[i
];
1305 if (cap_booke_sregs
) {
1306 ret
= kvmppc_get_booke_sregs(cpu
);
1313 ret
= kvmppc_get_books_sregs(cpu
);
1320 kvm_get_one_spr(cs
, KVM_REG_PPC_HIOR
, SPR_HIOR
);
1326 /* We deliberately ignore errors here, for kernels which have
1327 * the ONE_REG calls, but don't support the specific
1328 * registers, there's a reasonable chance things will still
1329 * work, at least until we try to migrate. */
1330 for (i
= 0; i
< 1024; i
++) {
1331 uint64_t id
= env
->spr_cb
[i
].one_reg_id
;
1334 kvm_get_one_spr(cs
, id
, i
);
1340 for (i
= 0; i
< ARRAY_SIZE(env
->tm_gpr
); i
++) {
1341 kvm_get_one_reg(cs
, KVM_REG_PPC_TM_GPR(i
), &env
->tm_gpr
[i
]);
1343 for (i
= 0; i
< ARRAY_SIZE(env
->tm_vsr
); i
++) {
1344 kvm_get_one_reg(cs
, KVM_REG_PPC_TM_VSR(i
), &env
->tm_vsr
[i
]);
1346 kvm_get_one_reg(cs
, KVM_REG_PPC_TM_CR
, &env
->tm_cr
);
1347 kvm_get_one_reg(cs
, KVM_REG_PPC_TM_LR
, &env
->tm_lr
);
1348 kvm_get_one_reg(cs
, KVM_REG_PPC_TM_CTR
, &env
->tm_ctr
);
1349 kvm_get_one_reg(cs
, KVM_REG_PPC_TM_FPSCR
, &env
->tm_fpscr
);
1350 kvm_get_one_reg(cs
, KVM_REG_PPC_TM_AMR
, &env
->tm_amr
);
1351 kvm_get_one_reg(cs
, KVM_REG_PPC_TM_PPR
, &env
->tm_ppr
);
1352 kvm_get_one_reg(cs
, KVM_REG_PPC_TM_VRSAVE
, &env
->tm_vrsave
);
1353 kvm_get_one_reg(cs
, KVM_REG_PPC_TM_VSCR
, &env
->tm_vscr
);
1354 kvm_get_one_reg(cs
, KVM_REG_PPC_TM_DSCR
, &env
->tm_dscr
);
1355 kvm_get_one_reg(cs
, KVM_REG_PPC_TM_TAR
, &env
->tm_tar
);
1359 if (kvm_get_vpa(cs
) < 0) {
1360 DPRINTF("Warning: Unable to get VPA information from KVM\n");
1364 kvm_get_one_reg(cs
, KVM_REG_PPC_TB_OFFSET
, &env
->tb_env
->tb_offset
);
1371 int kvmppc_set_interrupt(PowerPCCPU
*cpu
, int irq
, int level
)
1373 unsigned virq
= level
? KVM_INTERRUPT_SET_LEVEL
: KVM_INTERRUPT_UNSET
;
1375 if (irq
!= PPC_INTERRUPT_EXT
) {
1379 if (!kvm_enabled() || !cap_interrupt_unset
|| !cap_interrupt_level
) {
1383 kvm_vcpu_ioctl(CPU(cpu
), KVM_INTERRUPT
, &virq
);
1388 #if defined(TARGET_PPCEMB)
1389 #define PPC_INPUT_INT PPC40x_INPUT_INT
1390 #elif defined(TARGET_PPC64)
1391 #define PPC_INPUT_INT PPC970_INPUT_INT
1393 #define PPC_INPUT_INT PPC6xx_INPUT_INT
1396 void kvm_arch_pre_run(CPUState
*cs
, struct kvm_run
*run
)
1398 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
1399 CPUPPCState
*env
= &cpu
->env
;
1403 qemu_mutex_lock_iothread();
1405 /* PowerPC QEMU tracks the various core input pins (interrupt, critical
1406 * interrupt, reset, etc) in PPC-specific env->irq_input_state. */
1407 if (!cap_interrupt_level
&&
1408 run
->ready_for_interrupt_injection
&&
1409 (cs
->interrupt_request
& CPU_INTERRUPT_HARD
) &&
1410 (env
->irq_input_state
& (1<<PPC_INPUT_INT
)))
1412 /* For now KVM disregards the 'irq' argument. However, in the
1413 * future KVM could cache it in-kernel to avoid a heavyweight exit
1414 * when reading the UIC.
1416 irq
= KVM_INTERRUPT_SET
;
1418 DPRINTF("injected interrupt %d\n", irq
);
1419 r
= kvm_vcpu_ioctl(cs
, KVM_INTERRUPT
, &irq
);
1421 printf("cpu %d fail inject %x\n", cs
->cpu_index
, irq
);
1424 /* Always wake up soon in case the interrupt was level based */
1425 timer_mod(idle_timer
, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL
) +
1426 (NANOSECONDS_PER_SECOND
/ 50));
1429 /* We don't know if there are more interrupts pending after this. However,
1430 * the guest will return to userspace in the course of handling this one
1431 * anyways, so we will get a chance to deliver the rest. */
1433 qemu_mutex_unlock_iothread();
1436 MemTxAttrs
kvm_arch_post_run(CPUState
*cs
, struct kvm_run
*run
)
1438 return MEMTXATTRS_UNSPECIFIED
;
1441 int kvm_arch_process_async_events(CPUState
*cs
)
1446 static int kvmppc_handle_halt(PowerPCCPU
*cpu
)
1448 CPUState
*cs
= CPU(cpu
);
1449 CPUPPCState
*env
= &cpu
->env
;
1451 if (!(cs
->interrupt_request
& CPU_INTERRUPT_HARD
) && (msr_ee
)) {
1453 cs
->exception_index
= EXCP_HLT
;
1459 /* map dcr access to existing qemu dcr emulation */
1460 static int kvmppc_handle_dcr_read(CPUPPCState
*env
, uint32_t dcrn
, uint32_t *data
)
1462 if (ppc_dcr_read(env
->dcr_env
, dcrn
, data
) < 0)
1463 fprintf(stderr
, "Read to unhandled DCR (0x%x)\n", dcrn
);
1468 static int kvmppc_handle_dcr_write(CPUPPCState
*env
, uint32_t dcrn
, uint32_t data
)
1470 if (ppc_dcr_write(env
->dcr_env
, dcrn
, data
) < 0)
1471 fprintf(stderr
, "Write to unhandled DCR (0x%x)\n", dcrn
);
1476 int kvm_arch_insert_sw_breakpoint(CPUState
*cs
, struct kvm_sw_breakpoint
*bp
)
1478 /* Mixed endian case is not handled */
1479 uint32_t sc
= debug_inst_opcode
;
1481 if (cpu_memory_rw_debug(cs
, bp
->pc
, (uint8_t *)&bp
->saved_insn
,
1483 cpu_memory_rw_debug(cs
, bp
->pc
, (uint8_t *)&sc
, sizeof(sc
), 1)) {
1490 int kvm_arch_remove_sw_breakpoint(CPUState
*cs
, struct kvm_sw_breakpoint
*bp
)
1494 if (cpu_memory_rw_debug(cs
, bp
->pc
, (uint8_t *)&sc
, sizeof(sc
), 0) ||
1495 sc
!= debug_inst_opcode
||
1496 cpu_memory_rw_debug(cs
, bp
->pc
, (uint8_t *)&bp
->saved_insn
,
1504 static int find_hw_breakpoint(target_ulong addr
, int type
)
1508 assert((nb_hw_breakpoint
+ nb_hw_watchpoint
)
1509 <= ARRAY_SIZE(hw_debug_points
));
1511 for (n
= 0; n
< nb_hw_breakpoint
+ nb_hw_watchpoint
; n
++) {
1512 if (hw_debug_points
[n
].addr
== addr
&&
1513 hw_debug_points
[n
].type
== type
) {
1521 static int find_hw_watchpoint(target_ulong addr
, int *flag
)
1525 n
= find_hw_breakpoint(addr
, GDB_WATCHPOINT_ACCESS
);
1527 *flag
= BP_MEM_ACCESS
;
1531 n
= find_hw_breakpoint(addr
, GDB_WATCHPOINT_WRITE
);
1533 *flag
= BP_MEM_WRITE
;
1537 n
= find_hw_breakpoint(addr
, GDB_WATCHPOINT_READ
);
1539 *flag
= BP_MEM_READ
;
1546 int kvm_arch_insert_hw_breakpoint(target_ulong addr
,
1547 target_ulong len
, int type
)
1549 if ((nb_hw_breakpoint
+ nb_hw_watchpoint
) >= ARRAY_SIZE(hw_debug_points
)) {
1553 hw_debug_points
[nb_hw_breakpoint
+ nb_hw_watchpoint
].addr
= addr
;
1554 hw_debug_points
[nb_hw_breakpoint
+ nb_hw_watchpoint
].type
= type
;
1557 case GDB_BREAKPOINT_HW
:
1558 if (nb_hw_breakpoint
>= max_hw_breakpoint
) {
1562 if (find_hw_breakpoint(addr
, type
) >= 0) {
1569 case GDB_WATCHPOINT_WRITE
:
1570 case GDB_WATCHPOINT_READ
:
1571 case GDB_WATCHPOINT_ACCESS
:
1572 if (nb_hw_watchpoint
>= max_hw_watchpoint
) {
1576 if (find_hw_breakpoint(addr
, type
) >= 0) {
1590 int kvm_arch_remove_hw_breakpoint(target_ulong addr
,
1591 target_ulong len
, int type
)
1595 n
= find_hw_breakpoint(addr
, type
);
1601 case GDB_BREAKPOINT_HW
:
1605 case GDB_WATCHPOINT_WRITE
:
1606 case GDB_WATCHPOINT_READ
:
1607 case GDB_WATCHPOINT_ACCESS
:
1614 hw_debug_points
[n
] = hw_debug_points
[nb_hw_breakpoint
+ nb_hw_watchpoint
];
1619 void kvm_arch_remove_all_hw_breakpoints(void)
1621 nb_hw_breakpoint
= nb_hw_watchpoint
= 0;
1624 void kvm_arch_update_guest_debug(CPUState
*cs
, struct kvm_guest_debug
*dbg
)
1628 /* Software Breakpoint updates */
1629 if (kvm_sw_breakpoints_active(cs
)) {
1630 dbg
->control
|= KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_USE_SW_BP
;
1633 assert((nb_hw_breakpoint
+ nb_hw_watchpoint
)
1634 <= ARRAY_SIZE(hw_debug_points
));
1635 assert((nb_hw_breakpoint
+ nb_hw_watchpoint
) <= ARRAY_SIZE(dbg
->arch
.bp
));
1637 if (nb_hw_breakpoint
+ nb_hw_watchpoint
> 0) {
1638 dbg
->control
|= KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_USE_HW_BP
;
1639 memset(dbg
->arch
.bp
, 0, sizeof(dbg
->arch
.bp
));
1640 for (n
= 0; n
< nb_hw_breakpoint
+ nb_hw_watchpoint
; n
++) {
1641 switch (hw_debug_points
[n
].type
) {
1642 case GDB_BREAKPOINT_HW
:
1643 dbg
->arch
.bp
[n
].type
= KVMPPC_DEBUG_BREAKPOINT
;
1645 case GDB_WATCHPOINT_WRITE
:
1646 dbg
->arch
.bp
[n
].type
= KVMPPC_DEBUG_WATCH_WRITE
;
1648 case GDB_WATCHPOINT_READ
:
1649 dbg
->arch
.bp
[n
].type
= KVMPPC_DEBUG_WATCH_READ
;
1651 case GDB_WATCHPOINT_ACCESS
:
1652 dbg
->arch
.bp
[n
].type
= KVMPPC_DEBUG_WATCH_WRITE
|
1653 KVMPPC_DEBUG_WATCH_READ
;
1656 cpu_abort(cs
, "Unsupported breakpoint type\n");
1658 dbg
->arch
.bp
[n
].addr
= hw_debug_points
[n
].addr
;
1663 static int kvm_handle_debug(PowerPCCPU
*cpu
, struct kvm_run
*run
)
1665 CPUState
*cs
= CPU(cpu
);
1666 CPUPPCState
*env
= &cpu
->env
;
1667 struct kvm_debug_exit_arch
*arch_info
= &run
->debug
.arch
;
1672 if (cs
->singlestep_enabled
) {
1674 } else if (arch_info
->status
) {
1675 if (nb_hw_breakpoint
+ nb_hw_watchpoint
> 0) {
1676 if (arch_info
->status
& KVMPPC_DEBUG_BREAKPOINT
) {
1677 n
= find_hw_breakpoint(arch_info
->address
, GDB_BREAKPOINT_HW
);
1681 } else if (arch_info
->status
& (KVMPPC_DEBUG_WATCH_READ
|
1682 KVMPPC_DEBUG_WATCH_WRITE
)) {
1683 n
= find_hw_watchpoint(arch_info
->address
, &flag
);
1686 cs
->watchpoint_hit
= &hw_watchpoint
;
1687 hw_watchpoint
.vaddr
= hw_debug_points
[n
].addr
;
1688 hw_watchpoint
.flags
= flag
;
1692 } else if (kvm_find_sw_breakpoint(cs
, arch_info
->address
)) {
1695 /* QEMU is not able to handle debug exception, so inject
1696 * program exception to guest;
1697 * Yes program exception NOT debug exception !!
1698 * When QEMU is using debug resources then debug exception must
1699 * be always set. To achieve this we set MSR_DE and also set
1700 * MSRP_DEP so guest cannot change MSR_DE.
1701 * When emulating debug resource for guest we want guest
1702 * to control MSR_DE (enable/disable debug interrupt on need).
1703 * Supporting both configurations are NOT possible.
1704 * So the result is that we cannot share debug resources
1705 * between QEMU and Guest on BOOKE architecture.
1706 * In the current design QEMU gets the priority over guest,
1707 * this means that if QEMU is using debug resources then guest
1709 * For software breakpoint QEMU uses a privileged instruction;
1710 * So there cannot be any reason that we are here for guest
1711 * set debug exception, only possibility is guest executed a
1712 * privileged / illegal instruction and that's why we are
1713 * injecting a program interrupt.
1716 cpu_synchronize_state(cs
);
1717 /* env->nip is PC, so increment this by 4 to use
1718 * ppc_cpu_do_interrupt(), which set srr0 = env->nip - 4.
1721 cs
->exception_index
= POWERPC_EXCP_PROGRAM
;
1722 env
->error_code
= POWERPC_EXCP_INVAL
;
1723 ppc_cpu_do_interrupt(cs
);
1729 int kvm_arch_handle_exit(CPUState
*cs
, struct kvm_run
*run
)
1731 PowerPCCPU
*cpu
= POWERPC_CPU(cs
);
1732 CPUPPCState
*env
= &cpu
->env
;
1735 qemu_mutex_lock_iothread();
1737 switch (run
->exit_reason
) {
1739 if (run
->dcr
.is_write
) {
1740 DPRINTF("handle dcr write\n");
1741 ret
= kvmppc_handle_dcr_write(env
, run
->dcr
.dcrn
, run
->dcr
.data
);
1743 DPRINTF("handle dcr read\n");
1744 ret
= kvmppc_handle_dcr_read(env
, run
->dcr
.dcrn
, &run
->dcr
.data
);
1748 DPRINTF("handle halt\n");
1749 ret
= kvmppc_handle_halt(cpu
);
1751 #if defined(TARGET_PPC64)
1752 case KVM_EXIT_PAPR_HCALL
:
1753 DPRINTF("handle PAPR hypercall\n");
1754 run
->papr_hcall
.ret
= spapr_hypercall(cpu
,
1756 run
->papr_hcall
.args
);
1761 DPRINTF("handle epr\n");
1762 run
->epr
.epr
= ldl_phys(cs
->as
, env
->mpic_iack
);
1765 case KVM_EXIT_WATCHDOG
:
1766 DPRINTF("handle watchdog expiry\n");
1767 watchdog_perform_action();
1771 case KVM_EXIT_DEBUG
:
1772 DPRINTF("handle debug exception\n");
1773 if (kvm_handle_debug(cpu
, run
)) {
1777 /* re-enter, this exception was guest-internal */
1782 fprintf(stderr
, "KVM: unknown exit reason %d\n", run
->exit_reason
);
1787 qemu_mutex_unlock_iothread();
1791 int kvmppc_or_tsr_bits(PowerPCCPU
*cpu
, uint32_t tsr_bits
)
1793 CPUState
*cs
= CPU(cpu
);
1794 uint32_t bits
= tsr_bits
;
1795 struct kvm_one_reg reg
= {
1796 .id
= KVM_REG_PPC_OR_TSR
,
1797 .addr
= (uintptr_t) &bits
,
1800 return kvm_vcpu_ioctl(cs
, KVM_SET_ONE_REG
, ®
);
1803 int kvmppc_clear_tsr_bits(PowerPCCPU
*cpu
, uint32_t tsr_bits
)
1806 CPUState
*cs
= CPU(cpu
);
1807 uint32_t bits
= tsr_bits
;
1808 struct kvm_one_reg reg
= {
1809 .id
= KVM_REG_PPC_CLEAR_TSR
,
1810 .addr
= (uintptr_t) &bits
,
1813 return kvm_vcpu_ioctl(cs
, KVM_SET_ONE_REG
, ®
);
1816 int kvmppc_set_tcr(PowerPCCPU
*cpu
)
1818 CPUState
*cs
= CPU(cpu
);
1819 CPUPPCState
*env
= &cpu
->env
;
1820 uint32_t tcr
= env
->spr
[SPR_BOOKE_TCR
];
1822 struct kvm_one_reg reg
= {
1823 .id
= KVM_REG_PPC_TCR
,
1824 .addr
= (uintptr_t) &tcr
,
1827 return kvm_vcpu_ioctl(cs
, KVM_SET_ONE_REG
, ®
);
1830 int kvmppc_booke_watchdog_enable(PowerPCCPU
*cpu
)
1832 CPUState
*cs
= CPU(cpu
);
1835 if (!kvm_enabled()) {
1839 if (!cap_ppc_watchdog
) {
1840 printf("warning: KVM does not support watchdog");
1844 ret
= kvm_vcpu_enable_cap(cs
, KVM_CAP_PPC_BOOKE_WATCHDOG
, 0);
1846 fprintf(stderr
, "%s: couldn't enable KVM_CAP_PPC_BOOKE_WATCHDOG: %s\n",
1847 __func__
, strerror(-ret
));
1854 static int read_cpuinfo(const char *field
, char *value
, int len
)
1858 int field_len
= strlen(field
);
1861 f
= fopen("/proc/cpuinfo", "r");
1867 if (!fgets(line
, sizeof(line
), f
)) {
1870 if (!strncmp(line
, field
, field_len
)) {
1871 pstrcpy(value
, len
, line
);
1882 uint32_t kvmppc_get_tbfreq(void)
1886 uint32_t retval
= NANOSECONDS_PER_SECOND
;
1888 if (read_cpuinfo("timebase", line
, sizeof(line
))) {
1892 if (!(ns
= strchr(line
, ':'))) {
1901 bool kvmppc_get_host_serial(char **value
)
1903 return g_file_get_contents("/proc/device-tree/system-id", value
, NULL
,
1907 bool kvmppc_get_host_model(char **value
)
1909 return g_file_get_contents("/proc/device-tree/model", value
, NULL
, NULL
);
1912 /* Try to find a device tree node for a CPU with clock-frequency property */
1913 static int kvmppc_find_cpu_dt(char *buf
, int buf_len
)
1915 struct dirent
*dirp
;
1918 if ((dp
= opendir(PROC_DEVTREE_CPU
)) == NULL
) {
1919 printf("Can't open directory " PROC_DEVTREE_CPU
"\n");
1924 while ((dirp
= readdir(dp
)) != NULL
) {
1926 snprintf(buf
, buf_len
, "%s%s/clock-frequency", PROC_DEVTREE_CPU
,
1928 f
= fopen(buf
, "r");
1930 snprintf(buf
, buf_len
, "%s%s", PROC_DEVTREE_CPU
, dirp
->d_name
);
1937 if (buf
[0] == '\0') {
1938 printf("Unknown host!\n");
1945 static uint64_t kvmppc_read_int_dt(const char *filename
)
1954 f
= fopen(filename
, "rb");
1959 len
= fread(&u
, 1, sizeof(u
), f
);
1963 /* property is a 32-bit quantity */
1964 return be32_to_cpu(u
.v32
);
1966 return be64_to_cpu(u
.v64
);
1972 /* Read a CPU node property from the host device tree that's a single
1973 * integer (32-bit or 64-bit). Returns 0 if anything goes wrong
1974 * (can't find or open the property, or doesn't understand the
1976 static uint64_t kvmppc_read_int_cpu_dt(const char *propname
)
1978 char buf
[PATH_MAX
], *tmp
;
1981 if (kvmppc_find_cpu_dt(buf
, sizeof(buf
))) {
1985 tmp
= g_strdup_printf("%s/%s", buf
, propname
);
1986 val
= kvmppc_read_int_dt(tmp
);
1992 uint64_t kvmppc_get_clockfreq(void)
1994 return kvmppc_read_int_cpu_dt("clock-frequency");
1997 uint32_t kvmppc_get_vmx(void)
1999 return kvmppc_read_int_cpu_dt("ibm,vmx");
2002 uint32_t kvmppc_get_dfp(void)
2004 return kvmppc_read_int_cpu_dt("ibm,dfp");
2007 static int kvmppc_get_pvinfo(CPUPPCState
*env
, struct kvm_ppc_pvinfo
*pvinfo
)
2009 PowerPCCPU
*cpu
= ppc_env_get_cpu(env
);
2010 CPUState
*cs
= CPU(cpu
);
2012 if (kvm_vm_check_extension(cs
->kvm_state
, KVM_CAP_PPC_GET_PVINFO
) &&
2013 !kvm_vm_ioctl(cs
->kvm_state
, KVM_PPC_GET_PVINFO
, pvinfo
)) {
2020 int kvmppc_get_hasidle(CPUPPCState
*env
)
2022 struct kvm_ppc_pvinfo pvinfo
;
2024 if (!kvmppc_get_pvinfo(env
, &pvinfo
) &&
2025 (pvinfo
.flags
& KVM_PPC_PVINFO_FLAGS_EV_IDLE
)) {
2032 int kvmppc_get_hypercall(CPUPPCState
*env
, uint8_t *buf
, int buf_len
)
2034 uint32_t *hc
= (uint32_t*)buf
;
2035 struct kvm_ppc_pvinfo pvinfo
;
2037 if (!kvmppc_get_pvinfo(env
, &pvinfo
)) {
2038 memcpy(buf
, pvinfo
.hcall
, buf_len
);
2043 * Fallback to always fail hypercalls regardless of endianness:
2045 * tdi 0,r0,72 (becomes b .+8 in wrong endian, nop in good endian)
2047 * b .+8 (becomes nop in wrong endian)
2048 * bswap32(li r3, -1)
2051 hc
[0] = cpu_to_be32(0x08000048);
2052 hc
[1] = cpu_to_be32(0x3860ffff);
2053 hc
[2] = cpu_to_be32(0x48000008);
2054 hc
[3] = cpu_to_be32(bswap32(0x3860ffff));
2059 static inline int kvmppc_enable_hcall(KVMState
*s
, target_ulong hcall
)
2061 return kvm_vm_enable_cap(s
, KVM_CAP_PPC_ENABLE_HCALL
, 0, hcall
, 1);
2064 void kvmppc_enable_logical_ci_hcalls(void)
2067 * FIXME: it would be nice if we could detect the cases where
2068 * we're using a device which requires the in kernel
2069 * implementation of these hcalls, but the kernel lacks them and
2070 * produce a warning.
2072 kvmppc_enable_hcall(kvm_state
, H_LOGICAL_CI_LOAD
);
2073 kvmppc_enable_hcall(kvm_state
, H_LOGICAL_CI_STORE
);
2076 void kvmppc_enable_set_mode_hcall(void)
2078 kvmppc_enable_hcall(kvm_state
, H_SET_MODE
);
2081 void kvmppc_enable_clear_ref_mod_hcalls(void)
2083 kvmppc_enable_hcall(kvm_state
, H_CLEAR_REF
);
2084 kvmppc_enable_hcall(kvm_state
, H_CLEAR_MOD
);
2087 void kvmppc_set_papr(PowerPCCPU
*cpu
)
2089 CPUState
*cs
= CPU(cpu
);
2092 ret
= kvm_vcpu_enable_cap(cs
, KVM_CAP_PPC_PAPR
, 0);
2094 error_report("This vCPU type or KVM version does not support PAPR");
2098 /* Update the capability flag so we sync the right information
2103 int kvmppc_set_compat(PowerPCCPU
*cpu
, uint32_t compat_pvr
)
2105 return kvm_set_one_reg(CPU(cpu
), KVM_REG_PPC_ARCH_COMPAT
, &compat_pvr
);
2108 void kvmppc_set_mpic_proxy(PowerPCCPU
*cpu
, int mpic_proxy
)
2110 CPUState
*cs
= CPU(cpu
);
2113 ret
= kvm_vcpu_enable_cap(cs
, KVM_CAP_PPC_EPR
, 0, mpic_proxy
);
2114 if (ret
&& mpic_proxy
) {
2115 error_report("This KVM version does not support EPR");
2120 int kvmppc_smt_threads(void)
2122 return cap_ppc_smt
? cap_ppc_smt
: 1;
2126 off_t
kvmppc_alloc_rma(void **rma
)
2130 struct kvm_allocate_rma ret
;
2132 /* If cap_ppc_rma == 0, contiguous RMA allocation is not supported
2133 * if cap_ppc_rma == 1, contiguous RMA allocation is supported, but
2134 * not necessary on this hardware
2135 * if cap_ppc_rma == 2, contiguous RMA allocation is needed on this hardware
2137 * FIXME: We should allow the user to force contiguous RMA
2138 * allocation in the cap_ppc_rma==1 case.
2140 if (cap_ppc_rma
< 2) {
2144 fd
= kvm_vm_ioctl(kvm_state
, KVM_ALLOCATE_RMA
, &ret
);
2146 fprintf(stderr
, "KVM: Error on KVM_ALLOCATE_RMA: %s\n",
2151 size
= MIN(ret
.rma_size
, 256ul << 20);
2153 *rma
= mmap(NULL
, size
, PROT_READ
|PROT_WRITE
, MAP_SHARED
, fd
, 0);
2154 if (*rma
== MAP_FAILED
) {
2155 fprintf(stderr
, "KVM: Error mapping RMA: %s\n", strerror(errno
));
2162 uint64_t kvmppc_rma_size(uint64_t current_size
, unsigned int hash_shift
)
2164 struct kvm_ppc_smmu_info info
;
2165 long rampagesize
, best_page_shift
;
2168 if (cap_ppc_rma
>= 2) {
2169 return current_size
;
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;
2178 for (i
= 0; i
< KVM_PPC_PAGE_SIZES_MAX_SZ
; i
++) {
2179 struct kvm_ppc_one_seg_page_size
*sps
= &info
.sps
[i
];
2181 if (!sps
->page_shift
) {
2185 if ((sps
->page_shift
> best_page_shift
)
2186 && ((1UL << sps
->page_shift
) <= rampagesize
)) {
2187 best_page_shift
= sps
->page_shift
;
2191 return MIN(current_size
,
2192 1ULL << (best_page_shift
+ hash_shift
- 7));
2196 bool kvmppc_spapr_use_multitce(void)
2198 return cap_spapr_multitce
;
2201 int kvmppc_spapr_enable_inkernel_multitce(void)
2205 ret
= kvm_vm_enable_cap(kvm_state
, KVM_CAP_PPC_ENABLE_HCALL
, 0,
2206 H_PUT_TCE_INDIRECT
, 1);
2208 ret
= kvm_vm_enable_cap(kvm_state
, KVM_CAP_PPC_ENABLE_HCALL
, 0,
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
)
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
2227 if (!cap_spapr_tce
|| (need_vfio
&& !cap_spapr_vfio
)) {
2231 if (cap_spapr_tce_64
) {
2232 struct kvm_create_spapr_tce_64 args
= {
2234 .page_shift
= page_shift
,
2235 .offset
= bus_offset
>> page_shift
,
2239 fd
= kvm_vm_ioctl(kvm_state
, KVM_CREATE_SPAPR_TCE_64
, &args
);
2242 "KVM: Failed to create TCE64 table for liobn 0x%x\n",
2246 } else if (cap_spapr_tce
) {
2247 uint64_t window_size
= (uint64_t) nb_table
<< page_shift
;
2248 struct kvm_create_spapr_tce args
= {
2250 .window_size
= window_size
,
2252 if ((window_size
!= args
.window_size
) || bus_offset
) {
2255 fd
= kvm_vm_ioctl(kvm_state
, KVM_CREATE_SPAPR_TCE
, &args
);
2257 fprintf(stderr
, "KVM: Failed to create TCE table for liobn 0x%x\n",
2265 len
= nb_table
* sizeof(uint64_t);
2266 /* FIXME: round this up to page size */
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",
2280 int kvmppc_remove_spapr_tce(void *table
, int fd
, uint32_t nb_table
)
2288 len
= nb_table
* sizeof(uint64_t);
2289 if ((munmap(table
, len
) < 0) ||
2291 fprintf(stderr
, "KVM: Unexpected error removing TCE table: %s",
2293 /* Leak the table */
2299 int kvmppc_reset_htab(int shift_hint
)
2301 uint32_t shift
= shift_hint
;
2303 if (!kvm_enabled()) {
2304 /* Full emulation, tell caller to allocate htab itself */
2307 if (kvm_check_extension(kvm_state
, KVM_CAP_PPC_ALLOC_HTAB
)) {
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. */
2316 } else if (ret
< 0) {
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 */
2329 /* HV - assume 16MB kernel allocated htab */
2334 static inline uint32_t mfpvr(void)
2343 static void alter_insns(uint64_t *word
, uint64_t flags
, bool on
)
2352 static void kvmppc_host_cpu_class_init(ObjectClass
*oc
, void *data
)
2354 PowerPCCPUClass
*pcc
= POWERPC_CPU_CLASS(oc
);
2355 uint32_t vmx
= kvmppc_get_vmx();
2356 uint32_t dfp
= kvmppc_get_dfp();
2357 uint32_t dcache_size
= kvmppc_read_int_cpu_dt("d-cache-size");
2358 uint32_t icache_size
= kvmppc_read_int_cpu_dt("i-cache-size");
2360 /* Now fix up the class with information we can query from the host */
2364 /* Only override when we know what the host supports */
2365 alter_insns(&pcc
->insns_flags
, PPC_ALTIVEC
, vmx
> 0);
2366 alter_insns(&pcc
->insns_flags2
, PPC2_VSX
, vmx
> 1);
2369 /* Only override when we know what the host supports */
2370 alter_insns(&pcc
->insns_flags2
, PPC2_DFP
, dfp
);
2373 if (dcache_size
!= -1) {
2374 pcc
->l1_dcache_size
= dcache_size
;
2377 if (icache_size
!= -1) {
2378 pcc
->l1_icache_size
= icache_size
;
2381 #if defined(TARGET_PPC64)
2382 pcc
->radix_page_info
= kvm_get_radix_page_info();
2384 if ((pcc
->pvr
& 0xffffff00) == CPU_POWERPC_POWER9_DD1
) {
2386 * POWER9 DD1 has some bugs which make it not really ISA 3.00
2387 * compliant. More importantly, advertising ISA 3.00
2388 * architected mode may prevent guests from activating
2389 * necessary DD1 workarounds.
2391 pcc
->pcr_supported
&= ~(PCR_COMPAT_3_00
| PCR_COMPAT_2_07
2392 | PCR_COMPAT_2_06
| PCR_COMPAT_2_05
);
2394 #endif /* defined(TARGET_PPC64) */
2397 bool kvmppc_has_cap_epr(void)
2402 bool kvmppc_has_cap_htab_fd(void)
2407 bool kvmppc_has_cap_fixup_hcalls(void)
2409 return cap_fixup_hcalls
;
2412 bool kvmppc_has_cap_htm(void)
2417 bool kvmppc_has_cap_mmu_radix(void)
2419 return cap_mmu_radix
;
2422 bool kvmppc_has_cap_mmu_hash_v3(void)
2424 return cap_mmu_hash_v3
;
2427 PowerPCCPUClass
*kvm_ppc_get_host_cpu_class(void)
2429 uint32_t host_pvr
= mfpvr();
2430 PowerPCCPUClass
*pvr_pcc
;
2432 pvr_pcc
= ppc_cpu_class_by_pvr(host_pvr
);
2433 if (pvr_pcc
== NULL
) {
2434 pvr_pcc
= ppc_cpu_class_by_pvr_mask(host_pvr
);
2440 static int kvm_ppc_register_host_cpu_type(void)
2442 TypeInfo type_info
= {
2443 .name
= TYPE_HOST_POWERPC_CPU
,
2444 .class_init
= kvmppc_host_cpu_class_init
,
2446 PowerPCCPUClass
*pvr_pcc
;
2450 pvr_pcc
= kvm_ppc_get_host_cpu_class();
2451 if (pvr_pcc
== NULL
) {
2454 type_info
.parent
= object_class_get_name(OBJECT_CLASS(pvr_pcc
));
2455 type_register(&type_info
);
2457 #if defined(TARGET_PPC64)
2458 type_info
.name
= g_strdup_printf("%s-"TYPE_SPAPR_CPU_CORE
, "host");
2459 type_info
.parent
= TYPE_SPAPR_CPU_CORE
,
2460 type_info
.instance_size
= sizeof(sPAPRCPUCore
);
2461 type_info
.instance_init
= NULL
;
2462 type_info
.class_init
= spapr_cpu_core_class_init
;
2463 type_info
.class_data
= (void *) "host";
2464 type_register(&type_info
);
2465 g_free((void *)type_info
.name
);
2469 * Update generic CPU family class alias (e.g. on a POWER8NVL host,
2470 * we want "POWER8" to be a "family" alias that points to the current
2471 * host CPU type, too)
2473 dc
= DEVICE_CLASS(ppc_cpu_get_family_class(pvr_pcc
));
2474 for (i
= 0; ppc_cpu_aliases
[i
].alias
!= NULL
; i
++) {
2475 if (strcmp(ppc_cpu_aliases
[i
].alias
, dc
->desc
) == 0) {
2476 ObjectClass
*oc
= OBJECT_CLASS(pvr_pcc
);
2479 ppc_cpu_aliases
[i
].model
= g_strdup(object_class_get_name(oc
));
2480 suffix
= strstr(ppc_cpu_aliases
[i
].model
, "-"TYPE_POWERPC_CPU
);
2484 ppc_cpu_aliases
[i
].oc
= oc
;
2492 int kvmppc_define_rtas_kernel_token(uint32_t token
, const char *function
)
2494 struct kvm_rtas_token_args args
= {
2498 if (!kvm_check_extension(kvm_state
, KVM_CAP_PPC_RTAS
)) {
2502 strncpy(args
.name
, function
, sizeof(args
.name
));
2504 return kvm_vm_ioctl(kvm_state
, KVM_PPC_RTAS_DEFINE_TOKEN
, &args
);
2507 int kvmppc_get_htab_fd(bool write
)
2509 struct kvm_get_htab_fd s
= {
2510 .flags
= write
? KVM_GET_HTAB_WRITE
: 0,
2515 fprintf(stderr
, "KVM version doesn't support saving the hash table\n");
2519 return kvm_vm_ioctl(kvm_state
, KVM_PPC_GET_HTAB_FD
, &s
);
2522 int kvmppc_save_htab(QEMUFile
*f
, int fd
, size_t bufsize
, int64_t max_ns
)
2524 int64_t starttime
= qemu_clock_get_ns(QEMU_CLOCK_REALTIME
);
2525 uint8_t buf
[bufsize
];
2529 rc
= read(fd
, buf
, bufsize
);
2531 fprintf(stderr
, "Error reading data from KVM HTAB fd: %s\n",
2535 uint8_t *buffer
= buf
;
2538 struct kvm_get_htab_header
*head
=
2539 (struct kvm_get_htab_header
*) buffer
;
2540 size_t chunksize
= sizeof(*head
) +
2541 HASH_PTE_SIZE_64
* head
->n_valid
;
2543 qemu_put_be32(f
, head
->index
);
2544 qemu_put_be16(f
, head
->n_valid
);
2545 qemu_put_be16(f
, head
->n_invalid
);
2546 qemu_put_buffer(f
, (void *)(head
+ 1),
2547 HASH_PTE_SIZE_64
* head
->n_valid
);
2549 buffer
+= chunksize
;
2555 || ((qemu_clock_get_ns(QEMU_CLOCK_REALTIME
) - starttime
) < max_ns
)));
2557 return (rc
== 0) ? 1 : 0;
2560 int kvmppc_load_htab_chunk(QEMUFile
*f
, int fd
, uint32_t index
,
2561 uint16_t n_valid
, uint16_t n_invalid
)
2563 struct kvm_get_htab_header
*buf
;
2564 size_t chunksize
= sizeof(*buf
) + n_valid
*HASH_PTE_SIZE_64
;
2567 buf
= alloca(chunksize
);
2569 buf
->n_valid
= n_valid
;
2570 buf
->n_invalid
= n_invalid
;
2572 qemu_get_buffer(f
, (void *)(buf
+ 1), HASH_PTE_SIZE_64
*n_valid
);
2574 rc
= write(fd
, buf
, chunksize
);
2576 fprintf(stderr
, "Error writing KVM hash table: %s\n",
2580 if (rc
!= chunksize
) {
2581 /* We should never get a short write on a single chunk */
2582 fprintf(stderr
, "Short write, restoring KVM hash table\n");
2588 bool kvm_arch_stop_on_emulation_error(CPUState
*cpu
)
2593 void kvm_arch_init_irq_routing(KVMState
*s
)
2597 void kvmppc_read_hptes(ppc_hash_pte64_t
*hptes
, hwaddr ptex
, int n
)
2599 struct kvm_get_htab_fd ghf
= {
2601 .start_index
= ptex
,
2606 fd
= kvm_vm_ioctl(kvm_state
, KVM_PPC_GET_HTAB_FD
, &ghf
);
2608 hw_error("kvmppc_read_hptes: Unable to open HPT fd");
2613 struct kvm_get_htab_header
*hdr
;
2614 int m
= n
< HPTES_PER_GROUP
? n
: HPTES_PER_GROUP
;
2615 char buf
[sizeof(*hdr
) + m
* HASH_PTE_SIZE_64
];
2617 rc
= read(fd
, buf
, sizeof(buf
));
2619 hw_error("kvmppc_read_hptes: Unable to read HPTEs");
2622 hdr
= (struct kvm_get_htab_header
*)buf
;
2623 while ((i
< n
) && ((char *)hdr
< (buf
+ rc
))) {
2624 int invalid
= hdr
->n_invalid
;
2626 if (hdr
->index
!= (ptex
+ i
)) {
2627 hw_error("kvmppc_read_hptes: Unexpected HPTE index %"PRIu32
2628 " != (%"HWADDR_PRIu
" + %d", hdr
->index
, ptex
, i
);
2631 memcpy(hptes
+ i
, hdr
+ 1, HASH_PTE_SIZE_64
* hdr
->n_valid
);
2634 if ((n
- i
) < invalid
) {
2637 memset(hptes
+ i
, 0, invalid
* HASH_PTE_SIZE_64
);
2638 i
+= hdr
->n_invalid
;
2640 hdr
= (struct kvm_get_htab_header
*)
2641 ((char *)(hdr
+ 1) + HASH_PTE_SIZE_64
* hdr
->n_valid
);
2648 void kvmppc_write_hpte(hwaddr ptex
, uint64_t pte0
, uint64_t pte1
)
2651 struct kvm_get_htab_fd ghf
;
2653 struct kvm_get_htab_header hdr
;
2659 ghf
.start_index
= 0; /* Ignored */
2660 fd
= kvm_vm_ioctl(kvm_state
, KVM_PPC_GET_HTAB_FD
, &ghf
);
2662 hw_error("kvmppc_write_hpte: Unable to open HPT fd");
2665 buf
.hdr
.n_valid
= 1;
2666 buf
.hdr
.n_invalid
= 0;
2667 buf
.hdr
.index
= ptex
;
2668 buf
.pte0
= cpu_to_be64(pte0
);
2669 buf
.pte1
= cpu_to_be64(pte1
);
2671 rc
= write(fd
, &buf
, sizeof(buf
));
2672 if (rc
!= sizeof(buf
)) {
2673 hw_error("kvmppc_write_hpte: Unable to update KVM HPT");
2678 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry
*route
,
2679 uint64_t address
, uint32_t data
, PCIDevice
*dev
)
2684 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry
*route
,
2685 int vector
, PCIDevice
*dev
)
2690 int kvm_arch_release_virq_post(int virq
)
2695 int kvm_arch_msi_data_to_gsi(uint32_t data
)
2697 return data
& 0xffff;
2700 int kvmppc_enable_hwrng(void)
2702 if (!kvm_enabled() || !kvm_check_extension(kvm_state
, KVM_CAP_PPC_HWRNG
)) {
2706 return kvmppc_enable_hcall(kvm_state
, H_RANDOM
);