root / target-ppc / kvm.c @ d83af167
History | View | Annotate | Download (51.1 kB)
| 1 |
/*
|
|---|---|
| 2 |
* PowerPC implementation of KVM hooks
|
| 3 |
*
|
| 4 |
* Copyright IBM Corp. 2007
|
| 5 |
* Copyright (C) 2011 Freescale Semiconductor, Inc.
|
| 6 |
*
|
| 7 |
* Authors:
|
| 8 |
* Jerone Young <jyoung5@us.ibm.com>
|
| 9 |
* Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
|
| 10 |
* Hollis Blanchard <hollisb@us.ibm.com>
|
| 11 |
*
|
| 12 |
* This work is licensed under the terms of the GNU GPL, version 2 or later.
|
| 13 |
* See the COPYING file in the top-level directory.
|
| 14 |
*
|
| 15 |
*/
|
| 16 |
|
| 17 |
#include <dirent.h> |
| 18 |
#include <sys/types.h> |
| 19 |
#include <sys/ioctl.h> |
| 20 |
#include <sys/mman.h> |
| 21 |
#include <sys/vfs.h> |
| 22 |
|
| 23 |
#include <linux/kvm.h> |
| 24 |
|
| 25 |
#include "qemu-common.h" |
| 26 |
#include "qemu/timer.h" |
| 27 |
#include "sysemu/sysemu.h" |
| 28 |
#include "sysemu/kvm.h" |
| 29 |
#include "kvm_ppc.h" |
| 30 |
#include "cpu.h" |
| 31 |
#include "sysemu/cpus.h" |
| 32 |
#include "sysemu/device_tree.h" |
| 33 |
#include "mmu-hash64.h" |
| 34 |
|
| 35 |
#include "hw/sysbus.h" |
| 36 |
#include "hw/ppc/spapr.h" |
| 37 |
#include "hw/ppc/spapr_vio.h" |
| 38 |
#include "sysemu/watchdog.h" |
| 39 |
|
| 40 |
//#define DEBUG_KVM
|
| 41 |
|
| 42 |
#ifdef DEBUG_KVM
|
| 43 |
#define DPRINTF(fmt, ...) \
|
| 44 |
do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0) |
| 45 |
#else
|
| 46 |
#define DPRINTF(fmt, ...) \
|
| 47 |
do { } while (0) |
| 48 |
#endif
|
| 49 |
|
| 50 |
#define PROC_DEVTREE_CPU "/proc/device-tree/cpus/" |
| 51 |
|
| 52 |
const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
|
| 53 |
KVM_CAP_LAST_INFO |
| 54 |
}; |
| 55 |
|
| 56 |
static int cap_interrupt_unset = false; |
| 57 |
static int cap_interrupt_level = false; |
| 58 |
static int cap_segstate; |
| 59 |
static int cap_booke_sregs; |
| 60 |
static int cap_ppc_smt; |
| 61 |
static int cap_ppc_rma; |
| 62 |
static int cap_spapr_tce; |
| 63 |
static int cap_hior; |
| 64 |
static int cap_one_reg; |
| 65 |
static int cap_epr; |
| 66 |
static int cap_ppc_watchdog; |
| 67 |
static int cap_papr; |
| 68 |
static int cap_htab_fd; |
| 69 |
|
| 70 |
/* XXX We have a race condition where we actually have a level triggered
|
| 71 |
* interrupt, but the infrastructure can't expose that yet, so the guest
|
| 72 |
* takes but ignores it, goes to sleep and never gets notified that there's
|
| 73 |
* still an interrupt pending.
|
| 74 |
*
|
| 75 |
* As a quick workaround, let's just wake up again 20 ms after we injected
|
| 76 |
* an interrupt. That way we can assure that we're always reinjecting
|
| 77 |
* interrupts in case the guest swallowed them.
|
| 78 |
*/
|
| 79 |
static QEMUTimer *idle_timer;
|
| 80 |
|
| 81 |
static void kvm_kick_cpu(void *opaque) |
| 82 |
{
|
| 83 |
PowerPCCPU *cpu = opaque; |
| 84 |
|
| 85 |
qemu_cpu_kick(CPU(cpu)); |
| 86 |
} |
| 87 |
|
| 88 |
static int kvm_ppc_register_host_cpu_type(void); |
| 89 |
|
| 90 |
int kvm_arch_init(KVMState *s)
|
| 91 |
{
|
| 92 |
cap_interrupt_unset = kvm_check_extension(s, KVM_CAP_PPC_UNSET_IRQ); |
| 93 |
cap_interrupt_level = kvm_check_extension(s, KVM_CAP_PPC_IRQ_LEVEL); |
| 94 |
cap_segstate = kvm_check_extension(s, KVM_CAP_PPC_SEGSTATE); |
| 95 |
cap_booke_sregs = kvm_check_extension(s, KVM_CAP_PPC_BOOKE_SREGS); |
| 96 |
cap_ppc_smt = kvm_check_extension(s, KVM_CAP_PPC_SMT); |
| 97 |
cap_ppc_rma = kvm_check_extension(s, KVM_CAP_PPC_RMA); |
| 98 |
cap_spapr_tce = kvm_check_extension(s, KVM_CAP_SPAPR_TCE); |
| 99 |
cap_one_reg = kvm_check_extension(s, KVM_CAP_ONE_REG); |
| 100 |
cap_hior = kvm_check_extension(s, KVM_CAP_PPC_HIOR); |
| 101 |
cap_epr = kvm_check_extension(s, KVM_CAP_PPC_EPR); |
| 102 |
cap_ppc_watchdog = kvm_check_extension(s, KVM_CAP_PPC_BOOKE_WATCHDOG); |
| 103 |
/* Note: we don't set cap_papr here, because this capability is
|
| 104 |
* only activated after this by kvmppc_set_papr() */
|
| 105 |
cap_htab_fd = kvm_check_extension(s, KVM_CAP_PPC_HTAB_FD); |
| 106 |
|
| 107 |
if (!cap_interrupt_level) {
|
| 108 |
fprintf(stderr, "KVM: Couldn't find level irq capability. Expect the "
|
| 109 |
"VM to stall at times!\n");
|
| 110 |
} |
| 111 |
|
| 112 |
kvm_ppc_register_host_cpu_type(); |
| 113 |
|
| 114 |
return 0; |
| 115 |
} |
| 116 |
|
| 117 |
static int kvm_arch_sync_sregs(PowerPCCPU *cpu) |
| 118 |
{
|
| 119 |
CPUPPCState *cenv = &cpu->env; |
| 120 |
CPUState *cs = CPU(cpu); |
| 121 |
struct kvm_sregs sregs;
|
| 122 |
int ret;
|
| 123 |
|
| 124 |
if (cenv->excp_model == POWERPC_EXCP_BOOKE) {
|
| 125 |
/* What we're really trying to say is "if we're on BookE, we use
|
| 126 |
the native PVR for now". This is the only sane way to check
|
| 127 |
it though, so we potentially confuse users that they can run
|
| 128 |
BookE guests on BookS. Let's hope nobody dares enough :) */
|
| 129 |
return 0; |
| 130 |
} else {
|
| 131 |
if (!cap_segstate) {
|
| 132 |
fprintf(stderr, "kvm error: missing PVR setting capability\n");
|
| 133 |
return -ENOSYS;
|
| 134 |
} |
| 135 |
} |
| 136 |
|
| 137 |
ret = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs); |
| 138 |
if (ret) {
|
| 139 |
return ret;
|
| 140 |
} |
| 141 |
|
| 142 |
sregs.pvr = cenv->spr[SPR_PVR]; |
| 143 |
return kvm_vcpu_ioctl(cs, KVM_SET_SREGS, &sregs);
|
| 144 |
} |
| 145 |
|
| 146 |
/* Set up a shared TLB array with KVM */
|
| 147 |
static int kvm_booke206_tlb_init(PowerPCCPU *cpu) |
| 148 |
{
|
| 149 |
CPUPPCState *env = &cpu->env; |
| 150 |
CPUState *cs = CPU(cpu); |
| 151 |
struct kvm_book3e_206_tlb_params params = {};
|
| 152 |
struct kvm_config_tlb cfg = {};
|
| 153 |
struct kvm_enable_cap encap = {};
|
| 154 |
unsigned int entries = 0; |
| 155 |
int ret, i;
|
| 156 |
|
| 157 |
if (!kvm_enabled() ||
|
| 158 |
!kvm_check_extension(cs->kvm_state, KVM_CAP_SW_TLB)) {
|
| 159 |
return 0; |
| 160 |
} |
| 161 |
|
| 162 |
assert(ARRAY_SIZE(params.tlb_sizes) == BOOKE206_MAX_TLBN); |
| 163 |
|
| 164 |
for (i = 0; i < BOOKE206_MAX_TLBN; i++) { |
| 165 |
params.tlb_sizes[i] = booke206_tlb_size(env, i); |
| 166 |
params.tlb_ways[i] = booke206_tlb_ways(env, i); |
| 167 |
entries += params.tlb_sizes[i]; |
| 168 |
} |
| 169 |
|
| 170 |
assert(entries == env->nb_tlb); |
| 171 |
assert(sizeof(struct kvm_book3e_206_tlb_entry) == sizeof(ppcmas_tlb_t)); |
| 172 |
|
| 173 |
env->tlb_dirty = true;
|
| 174 |
|
| 175 |
cfg.array = (uintptr_t)env->tlb.tlbm; |
| 176 |
cfg.array_len = sizeof(ppcmas_tlb_t) * entries;
|
| 177 |
cfg.params = (uintptr_t)¶ms; |
| 178 |
cfg.mmu_type = KVM_MMU_FSL_BOOKE_NOHV; |
| 179 |
|
| 180 |
encap.cap = KVM_CAP_SW_TLB; |
| 181 |
encap.args[0] = (uintptr_t)&cfg;
|
| 182 |
|
| 183 |
ret = kvm_vcpu_ioctl(cs, KVM_ENABLE_CAP, &encap); |
| 184 |
if (ret < 0) { |
| 185 |
fprintf(stderr, "%s: couldn't enable KVM_CAP_SW_TLB: %s\n",
|
| 186 |
__func__, strerror(-ret)); |
| 187 |
return ret;
|
| 188 |
} |
| 189 |
|
| 190 |
env->kvm_sw_tlb = true;
|
| 191 |
return 0; |
| 192 |
} |
| 193 |
|
| 194 |
|
| 195 |
#if defined(TARGET_PPC64)
|
| 196 |
static void kvm_get_fallback_smmu_info(PowerPCCPU *cpu, |
| 197 |
struct kvm_ppc_smmu_info *info)
|
| 198 |
{
|
| 199 |
CPUPPCState *env = &cpu->env; |
| 200 |
CPUState *cs = CPU(cpu); |
| 201 |
|
| 202 |
memset(info, 0, sizeof(*info)); |
| 203 |
|
| 204 |
/* We don't have the new KVM_PPC_GET_SMMU_INFO ioctl, so
|
| 205 |
* need to "guess" what the supported page sizes are.
|
| 206 |
*
|
| 207 |
* For that to work we make a few assumptions:
|
| 208 |
*
|
| 209 |
* - If KVM_CAP_PPC_GET_PVINFO is supported we are running "PR"
|
| 210 |
* KVM which only supports 4K and 16M pages, but supports them
|
| 211 |
* regardless of the backing store characteritics. We also don't
|
| 212 |
* support 1T segments.
|
| 213 |
*
|
| 214 |
* This is safe as if HV KVM ever supports that capability or PR
|
| 215 |
* KVM grows supports for more page/segment sizes, those versions
|
| 216 |
* will have implemented KVM_CAP_PPC_GET_SMMU_INFO and thus we
|
| 217 |
* will not hit this fallback
|
| 218 |
*
|
| 219 |
* - Else we are running HV KVM. This means we only support page
|
| 220 |
* sizes that fit in the backing store. Additionally we only
|
| 221 |
* advertize 64K pages if the processor is ARCH 2.06 and we assume
|
| 222 |
* P7 encodings for the SLB and hash table. Here too, we assume
|
| 223 |
* support for any newer processor will mean a kernel that
|
| 224 |
* implements KVM_CAP_PPC_GET_SMMU_INFO and thus doesn't hit
|
| 225 |
* this fallback.
|
| 226 |
*/
|
| 227 |
if (kvm_check_extension(cs->kvm_state, KVM_CAP_PPC_GET_PVINFO)) {
|
| 228 |
/* No flags */
|
| 229 |
info->flags = 0;
|
| 230 |
info->slb_size = 64;
|
| 231 |
|
| 232 |
/* Standard 4k base page size segment */
|
| 233 |
info->sps[0].page_shift = 12; |
| 234 |
info->sps[0].slb_enc = 0; |
| 235 |
info->sps[0].enc[0].page_shift = 12; |
| 236 |
info->sps[0].enc[0].pte_enc = 0; |
| 237 |
|
| 238 |
/* Standard 16M large page size segment */
|
| 239 |
info->sps[1].page_shift = 24; |
| 240 |
info->sps[1].slb_enc = SLB_VSID_L;
|
| 241 |
info->sps[1].enc[0].page_shift = 24; |
| 242 |
info->sps[1].enc[0].pte_enc = 0; |
| 243 |
} else {
|
| 244 |
int i = 0; |
| 245 |
|
| 246 |
/* HV KVM has backing store size restrictions */
|
| 247 |
info->flags = KVM_PPC_PAGE_SIZES_REAL; |
| 248 |
|
| 249 |
if (env->mmu_model & POWERPC_MMU_1TSEG) {
|
| 250 |
info->flags |= KVM_PPC_1T_SEGMENTS; |
| 251 |
} |
| 252 |
|
| 253 |
if (env->mmu_model == POWERPC_MMU_2_06) {
|
| 254 |
info->slb_size = 32;
|
| 255 |
} else {
|
| 256 |
info->slb_size = 64;
|
| 257 |
} |
| 258 |
|
| 259 |
/* Standard 4k base page size segment */
|
| 260 |
info->sps[i].page_shift = 12;
|
| 261 |
info->sps[i].slb_enc = 0;
|
| 262 |
info->sps[i].enc[0].page_shift = 12; |
| 263 |
info->sps[i].enc[0].pte_enc = 0; |
| 264 |
i++; |
| 265 |
|
| 266 |
/* 64K on MMU 2.06 */
|
| 267 |
if (env->mmu_model == POWERPC_MMU_2_06) {
|
| 268 |
info->sps[i].page_shift = 16;
|
| 269 |
info->sps[i].slb_enc = 0x110;
|
| 270 |
info->sps[i].enc[0].page_shift = 16; |
| 271 |
info->sps[i].enc[0].pte_enc = 1; |
| 272 |
i++; |
| 273 |
} |
| 274 |
|
| 275 |
/* Standard 16M large page size segment */
|
| 276 |
info->sps[i].page_shift = 24;
|
| 277 |
info->sps[i].slb_enc = SLB_VSID_L; |
| 278 |
info->sps[i].enc[0].page_shift = 24; |
| 279 |
info->sps[i].enc[0].pte_enc = 0; |
| 280 |
} |
| 281 |
} |
| 282 |
|
| 283 |
static void kvm_get_smmu_info(PowerPCCPU *cpu, struct kvm_ppc_smmu_info *info) |
| 284 |
{
|
| 285 |
CPUState *cs = CPU(cpu); |
| 286 |
int ret;
|
| 287 |
|
| 288 |
if (kvm_check_extension(cs->kvm_state, KVM_CAP_PPC_GET_SMMU_INFO)) {
|
| 289 |
ret = kvm_vm_ioctl(cs->kvm_state, KVM_PPC_GET_SMMU_INFO, info); |
| 290 |
if (ret == 0) { |
| 291 |
return;
|
| 292 |
} |
| 293 |
} |
| 294 |
|
| 295 |
kvm_get_fallback_smmu_info(cpu, info); |
| 296 |
} |
| 297 |
|
| 298 |
static long getrampagesize(void) |
| 299 |
{
|
| 300 |
struct statfs fs;
|
| 301 |
int ret;
|
| 302 |
|
| 303 |
if (!mem_path) {
|
| 304 |
/* guest RAM is backed by normal anonymous pages */
|
| 305 |
return getpagesize();
|
| 306 |
} |
| 307 |
|
| 308 |
do {
|
| 309 |
ret = statfs(mem_path, &fs); |
| 310 |
} while (ret != 0 && errno == EINTR); |
| 311 |
|
| 312 |
if (ret != 0) { |
| 313 |
fprintf(stderr, "Couldn't statfs() memory path: %s\n",
|
| 314 |
strerror(errno)); |
| 315 |
exit(1);
|
| 316 |
} |
| 317 |
|
| 318 |
#define HUGETLBFS_MAGIC 0x958458f6 |
| 319 |
|
| 320 |
if (fs.f_type != HUGETLBFS_MAGIC) {
|
| 321 |
/* Explicit mempath, but it's ordinary pages */
|
| 322 |
return getpagesize();
|
| 323 |
} |
| 324 |
|
| 325 |
/* It's hugepage, return the huge page size */
|
| 326 |
return fs.f_bsize;
|
| 327 |
} |
| 328 |
|
| 329 |
static bool kvm_valid_page_size(uint32_t flags, long rampgsize, uint32_t shift) |
| 330 |
{
|
| 331 |
if (!(flags & KVM_PPC_PAGE_SIZES_REAL)) {
|
| 332 |
return true; |
| 333 |
} |
| 334 |
|
| 335 |
return (1ul << shift) <= rampgsize; |
| 336 |
} |
| 337 |
|
| 338 |
static void kvm_fixup_page_sizes(PowerPCCPU *cpu) |
| 339 |
{
|
| 340 |
static struct kvm_ppc_smmu_info smmu_info; |
| 341 |
static bool has_smmu_info; |
| 342 |
CPUPPCState *env = &cpu->env; |
| 343 |
long rampagesize;
|
| 344 |
int iq, ik, jq, jk;
|
| 345 |
|
| 346 |
/* We only handle page sizes for 64-bit server guests for now */
|
| 347 |
if (!(env->mmu_model & POWERPC_MMU_64)) {
|
| 348 |
return;
|
| 349 |
} |
| 350 |
|
| 351 |
/* Collect MMU info from kernel if not already */
|
| 352 |
if (!has_smmu_info) {
|
| 353 |
kvm_get_smmu_info(cpu, &smmu_info); |
| 354 |
has_smmu_info = true;
|
| 355 |
} |
| 356 |
|
| 357 |
rampagesize = getrampagesize(); |
| 358 |
|
| 359 |
/* Convert to QEMU form */
|
| 360 |
memset(&env->sps, 0, sizeof(env->sps)); |
| 361 |
|
| 362 |
for (ik = iq = 0; ik < KVM_PPC_PAGE_SIZES_MAX_SZ; ik++) { |
| 363 |
struct ppc_one_seg_page_size *qsps = &env->sps.sps[iq];
|
| 364 |
struct kvm_ppc_one_seg_page_size *ksps = &smmu_info.sps[ik];
|
| 365 |
|
| 366 |
if (!kvm_valid_page_size(smmu_info.flags, rampagesize,
|
| 367 |
ksps->page_shift)) {
|
| 368 |
continue;
|
| 369 |
} |
| 370 |
qsps->page_shift = ksps->page_shift; |
| 371 |
qsps->slb_enc = ksps->slb_enc; |
| 372 |
for (jk = jq = 0; jk < KVM_PPC_PAGE_SIZES_MAX_SZ; jk++) { |
| 373 |
if (!kvm_valid_page_size(smmu_info.flags, rampagesize,
|
| 374 |
ksps->enc[jk].page_shift)) {
|
| 375 |
continue;
|
| 376 |
} |
| 377 |
qsps->enc[jq].page_shift = ksps->enc[jk].page_shift; |
| 378 |
qsps->enc[jq].pte_enc = ksps->enc[jk].pte_enc; |
| 379 |
if (++jq >= PPC_PAGE_SIZES_MAX_SZ) {
|
| 380 |
break;
|
| 381 |
} |
| 382 |
} |
| 383 |
if (++iq >= PPC_PAGE_SIZES_MAX_SZ) {
|
| 384 |
break;
|
| 385 |
} |
| 386 |
} |
| 387 |
env->slb_nr = smmu_info.slb_size; |
| 388 |
if (smmu_info.flags & KVM_PPC_1T_SEGMENTS) {
|
| 389 |
env->mmu_model |= POWERPC_MMU_1TSEG; |
| 390 |
} else {
|
| 391 |
env->mmu_model &= ~POWERPC_MMU_1TSEG; |
| 392 |
} |
| 393 |
} |
| 394 |
#else /* defined (TARGET_PPC64) */ |
| 395 |
|
| 396 |
static inline void kvm_fixup_page_sizes(PowerPCCPU *cpu) |
| 397 |
{
|
| 398 |
} |
| 399 |
|
| 400 |
#endif /* !defined (TARGET_PPC64) */ |
| 401 |
|
| 402 |
unsigned long kvm_arch_vcpu_id(CPUState *cpu) |
| 403 |
{
|
| 404 |
return cpu->cpu_index;
|
| 405 |
} |
| 406 |
|
| 407 |
int kvm_arch_init_vcpu(CPUState *cs)
|
| 408 |
{
|
| 409 |
PowerPCCPU *cpu = POWERPC_CPU(cs); |
| 410 |
CPUPPCState *cenv = &cpu->env; |
| 411 |
int ret;
|
| 412 |
|
| 413 |
/* Gather server mmu info from KVM and update the CPU state */
|
| 414 |
kvm_fixup_page_sizes(cpu); |
| 415 |
|
| 416 |
/* Synchronize sregs with kvm */
|
| 417 |
ret = kvm_arch_sync_sregs(cpu); |
| 418 |
if (ret) {
|
| 419 |
return ret;
|
| 420 |
} |
| 421 |
|
| 422 |
idle_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, kvm_kick_cpu, cpu); |
| 423 |
|
| 424 |
/* Some targets support access to KVM's guest TLB. */
|
| 425 |
switch (cenv->mmu_model) {
|
| 426 |
case POWERPC_MMU_BOOKE206:
|
| 427 |
ret = kvm_booke206_tlb_init(cpu); |
| 428 |
break;
|
| 429 |
default:
|
| 430 |
break;
|
| 431 |
} |
| 432 |
|
| 433 |
return ret;
|
| 434 |
} |
| 435 |
|
| 436 |
void kvm_arch_reset_vcpu(CPUState *cpu)
|
| 437 |
{
|
| 438 |
} |
| 439 |
|
| 440 |
static void kvm_sw_tlb_put(PowerPCCPU *cpu) |
| 441 |
{
|
| 442 |
CPUPPCState *env = &cpu->env; |
| 443 |
CPUState *cs = CPU(cpu); |
| 444 |
struct kvm_dirty_tlb dirty_tlb;
|
| 445 |
unsigned char *bitmap; |
| 446 |
int ret;
|
| 447 |
|
| 448 |
if (!env->kvm_sw_tlb) {
|
| 449 |
return;
|
| 450 |
} |
| 451 |
|
| 452 |
bitmap = g_malloc((env->nb_tlb + 7) / 8); |
| 453 |
memset(bitmap, 0xFF, (env->nb_tlb + 7) / 8); |
| 454 |
|
| 455 |
dirty_tlb.bitmap = (uintptr_t)bitmap; |
| 456 |
dirty_tlb.num_dirty = env->nb_tlb; |
| 457 |
|
| 458 |
ret = kvm_vcpu_ioctl(cs, KVM_DIRTY_TLB, &dirty_tlb); |
| 459 |
if (ret) {
|
| 460 |
fprintf(stderr, "%s: KVM_DIRTY_TLB: %s\n",
|
| 461 |
__func__, strerror(-ret)); |
| 462 |
} |
| 463 |
|
| 464 |
g_free(bitmap); |
| 465 |
} |
| 466 |
|
| 467 |
static void kvm_get_one_spr(CPUState *cs, uint64_t id, int spr) |
| 468 |
{
|
| 469 |
PowerPCCPU *cpu = POWERPC_CPU(cs); |
| 470 |
CPUPPCState *env = &cpu->env; |
| 471 |
union {
|
| 472 |
uint32_t u32; |
| 473 |
uint64_t u64; |
| 474 |
} val; |
| 475 |
struct kvm_one_reg reg = {
|
| 476 |
.id = id, |
| 477 |
.addr = (uintptr_t) &val, |
| 478 |
}; |
| 479 |
int ret;
|
| 480 |
|
| 481 |
ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®); |
| 482 |
if (ret != 0) { |
| 483 |
fprintf(stderr, "Warning: Unable to retrieve SPR %d from KVM: %s\n",
|
| 484 |
spr, strerror(errno)); |
| 485 |
} else {
|
| 486 |
switch (id & KVM_REG_SIZE_MASK) {
|
| 487 |
case KVM_REG_SIZE_U32:
|
| 488 |
env->spr[spr] = val.u32; |
| 489 |
break;
|
| 490 |
|
| 491 |
case KVM_REG_SIZE_U64:
|
| 492 |
env->spr[spr] = val.u64; |
| 493 |
break;
|
| 494 |
|
| 495 |
default:
|
| 496 |
/* Don't handle this size yet */
|
| 497 |
abort(); |
| 498 |
} |
| 499 |
} |
| 500 |
} |
| 501 |
|
| 502 |
static void kvm_put_one_spr(CPUState *cs, uint64_t id, int spr) |
| 503 |
{
|
| 504 |
PowerPCCPU *cpu = POWERPC_CPU(cs); |
| 505 |
CPUPPCState *env = &cpu->env; |
| 506 |
union {
|
| 507 |
uint32_t u32; |
| 508 |
uint64_t u64; |
| 509 |
} val; |
| 510 |
struct kvm_one_reg reg = {
|
| 511 |
.id = id, |
| 512 |
.addr = (uintptr_t) &val, |
| 513 |
}; |
| 514 |
int ret;
|
| 515 |
|
| 516 |
switch (id & KVM_REG_SIZE_MASK) {
|
| 517 |
case KVM_REG_SIZE_U32:
|
| 518 |
val.u32 = env->spr[spr]; |
| 519 |
break;
|
| 520 |
|
| 521 |
case KVM_REG_SIZE_U64:
|
| 522 |
val.u64 = env->spr[spr]; |
| 523 |
break;
|
| 524 |
|
| 525 |
default:
|
| 526 |
/* Don't handle this size yet */
|
| 527 |
abort(); |
| 528 |
} |
| 529 |
|
| 530 |
ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®); |
| 531 |
if (ret != 0) { |
| 532 |
fprintf(stderr, "Warning: Unable to set SPR %d to KVM: %s\n",
|
| 533 |
spr, strerror(errno)); |
| 534 |
} |
| 535 |
} |
| 536 |
|
| 537 |
static int kvm_put_fp(CPUState *cs) |
| 538 |
{
|
| 539 |
PowerPCCPU *cpu = POWERPC_CPU(cs); |
| 540 |
CPUPPCState *env = &cpu->env; |
| 541 |
struct kvm_one_reg reg;
|
| 542 |
int i;
|
| 543 |
int ret;
|
| 544 |
|
| 545 |
if (env->insns_flags & PPC_FLOAT) {
|
| 546 |
uint64_t fpscr = env->fpscr; |
| 547 |
bool vsx = !!(env->insns_flags2 & PPC2_VSX);
|
| 548 |
|
| 549 |
reg.id = KVM_REG_PPC_FPSCR; |
| 550 |
reg.addr = (uintptr_t)&fpscr; |
| 551 |
ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®); |
| 552 |
if (ret < 0) { |
| 553 |
DPRINTF("Unable to set FPSCR to KVM: %s\n", strerror(errno));
|
| 554 |
return ret;
|
| 555 |
} |
| 556 |
|
| 557 |
for (i = 0; i < 32; i++) { |
| 558 |
uint64_t vsr[2];
|
| 559 |
|
| 560 |
vsr[0] = float64_val(env->fpr[i]);
|
| 561 |
vsr[1] = env->vsr[i];
|
| 562 |
reg.addr = (uintptr_t) &vsr; |
| 563 |
reg.id = vsx ? KVM_REG_PPC_VSR(i) : KVM_REG_PPC_FPR(i); |
| 564 |
|
| 565 |
ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®); |
| 566 |
if (ret < 0) { |
| 567 |
DPRINTF("Unable to set %s%d to KVM: %s\n", vsx ? "VSR" : "FPR", |
| 568 |
i, strerror(errno)); |
| 569 |
return ret;
|
| 570 |
} |
| 571 |
} |
| 572 |
} |
| 573 |
|
| 574 |
if (env->insns_flags & PPC_ALTIVEC) {
|
| 575 |
reg.id = KVM_REG_PPC_VSCR; |
| 576 |
reg.addr = (uintptr_t)&env->vscr; |
| 577 |
ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®); |
| 578 |
if (ret < 0) { |
| 579 |
DPRINTF("Unable to set VSCR to KVM: %s\n", strerror(errno));
|
| 580 |
return ret;
|
| 581 |
} |
| 582 |
|
| 583 |
for (i = 0; i < 32; i++) { |
| 584 |
reg.id = KVM_REG_PPC_VR(i); |
| 585 |
reg.addr = (uintptr_t)&env->avr[i]; |
| 586 |
ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®); |
| 587 |
if (ret < 0) { |
| 588 |
DPRINTF("Unable to set VR%d to KVM: %s\n", i, strerror(errno));
|
| 589 |
return ret;
|
| 590 |
} |
| 591 |
} |
| 592 |
} |
| 593 |
|
| 594 |
return 0; |
| 595 |
} |
| 596 |
|
| 597 |
static int kvm_get_fp(CPUState *cs) |
| 598 |
{
|
| 599 |
PowerPCCPU *cpu = POWERPC_CPU(cs); |
| 600 |
CPUPPCState *env = &cpu->env; |
| 601 |
struct kvm_one_reg reg;
|
| 602 |
int i;
|
| 603 |
int ret;
|
| 604 |
|
| 605 |
if (env->insns_flags & PPC_FLOAT) {
|
| 606 |
uint64_t fpscr; |
| 607 |
bool vsx = !!(env->insns_flags2 & PPC2_VSX);
|
| 608 |
|
| 609 |
reg.id = KVM_REG_PPC_FPSCR; |
| 610 |
reg.addr = (uintptr_t)&fpscr; |
| 611 |
ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®); |
| 612 |
if (ret < 0) { |
| 613 |
DPRINTF("Unable to get FPSCR from KVM: %s\n", strerror(errno));
|
| 614 |
return ret;
|
| 615 |
} else {
|
| 616 |
env->fpscr = fpscr; |
| 617 |
} |
| 618 |
|
| 619 |
for (i = 0; i < 32; i++) { |
| 620 |
uint64_t vsr[2];
|
| 621 |
|
| 622 |
reg.addr = (uintptr_t) &vsr; |
| 623 |
reg.id = vsx ? KVM_REG_PPC_VSR(i) : KVM_REG_PPC_FPR(i); |
| 624 |
|
| 625 |
ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®); |
| 626 |
if (ret < 0) { |
| 627 |
DPRINTF("Unable to get %s%d from KVM: %s\n",
|
| 628 |
vsx ? "VSR" : "FPR", i, strerror(errno)); |
| 629 |
return ret;
|
| 630 |
} else {
|
| 631 |
env->fpr[i] = vsr[0];
|
| 632 |
if (vsx) {
|
| 633 |
env->vsr[i] = vsr[1];
|
| 634 |
} |
| 635 |
} |
| 636 |
} |
| 637 |
} |
| 638 |
|
| 639 |
if (env->insns_flags & PPC_ALTIVEC) {
|
| 640 |
reg.id = KVM_REG_PPC_VSCR; |
| 641 |
reg.addr = (uintptr_t)&env->vscr; |
| 642 |
ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®); |
| 643 |
if (ret < 0) { |
| 644 |
DPRINTF("Unable to get VSCR from KVM: %s\n", strerror(errno));
|
| 645 |
return ret;
|
| 646 |
} |
| 647 |
|
| 648 |
for (i = 0; i < 32; i++) { |
| 649 |
reg.id = KVM_REG_PPC_VR(i); |
| 650 |
reg.addr = (uintptr_t)&env->avr[i]; |
| 651 |
ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®); |
| 652 |
if (ret < 0) { |
| 653 |
DPRINTF("Unable to get VR%d from KVM: %s\n",
|
| 654 |
i, strerror(errno)); |
| 655 |
return ret;
|
| 656 |
} |
| 657 |
} |
| 658 |
} |
| 659 |
|
| 660 |
return 0; |
| 661 |
} |
| 662 |
|
| 663 |
#if defined(TARGET_PPC64)
|
| 664 |
static int kvm_get_vpa(CPUState *cs) |
| 665 |
{
|
| 666 |
PowerPCCPU *cpu = POWERPC_CPU(cs); |
| 667 |
CPUPPCState *env = &cpu->env; |
| 668 |
struct kvm_one_reg reg;
|
| 669 |
int ret;
|
| 670 |
|
| 671 |
reg.id = KVM_REG_PPC_VPA_ADDR; |
| 672 |
reg.addr = (uintptr_t)&env->vpa_addr; |
| 673 |
ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®); |
| 674 |
if (ret < 0) { |
| 675 |
DPRINTF("Unable to get VPA address from KVM: %s\n", strerror(errno));
|
| 676 |
return ret;
|
| 677 |
} |
| 678 |
|
| 679 |
assert((uintptr_t)&env->slb_shadow_size |
| 680 |
== ((uintptr_t)&env->slb_shadow_addr + 8));
|
| 681 |
reg.id = KVM_REG_PPC_VPA_SLB; |
| 682 |
reg.addr = (uintptr_t)&env->slb_shadow_addr; |
| 683 |
ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®); |
| 684 |
if (ret < 0) { |
| 685 |
DPRINTF("Unable to get SLB shadow state from KVM: %s\n",
|
| 686 |
strerror(errno)); |
| 687 |
return ret;
|
| 688 |
} |
| 689 |
|
| 690 |
assert((uintptr_t)&env->dtl_size == ((uintptr_t)&env->dtl_addr + 8));
|
| 691 |
reg.id = KVM_REG_PPC_VPA_DTL; |
| 692 |
reg.addr = (uintptr_t)&env->dtl_addr; |
| 693 |
ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®); |
| 694 |
if (ret < 0) { |
| 695 |
DPRINTF("Unable to get dispatch trace log state from KVM: %s\n",
|
| 696 |
strerror(errno)); |
| 697 |
return ret;
|
| 698 |
} |
| 699 |
|
| 700 |
return 0; |
| 701 |
} |
| 702 |
|
| 703 |
static int kvm_put_vpa(CPUState *cs) |
| 704 |
{
|
| 705 |
PowerPCCPU *cpu = POWERPC_CPU(cs); |
| 706 |
CPUPPCState *env = &cpu->env; |
| 707 |
struct kvm_one_reg reg;
|
| 708 |
int ret;
|
| 709 |
|
| 710 |
/* SLB shadow or DTL can't be registered unless a master VPA is
|
| 711 |
* registered. That means when restoring state, if a VPA *is*
|
| 712 |
* registered, we need to set that up first. If not, we need to
|
| 713 |
* deregister the others before deregistering the master VPA */
|
| 714 |
assert(env->vpa_addr || !(env->slb_shadow_addr || env->dtl_addr)); |
| 715 |
|
| 716 |
if (env->vpa_addr) {
|
| 717 |
reg.id = KVM_REG_PPC_VPA_ADDR; |
| 718 |
reg.addr = (uintptr_t)&env->vpa_addr; |
| 719 |
ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®); |
| 720 |
if (ret < 0) { |
| 721 |
DPRINTF("Unable to set VPA address to KVM: %s\n", strerror(errno));
|
| 722 |
return ret;
|
| 723 |
} |
| 724 |
} |
| 725 |
|
| 726 |
assert((uintptr_t)&env->slb_shadow_size |
| 727 |
== ((uintptr_t)&env->slb_shadow_addr + 8));
|
| 728 |
reg.id = KVM_REG_PPC_VPA_SLB; |
| 729 |
reg.addr = (uintptr_t)&env->slb_shadow_addr; |
| 730 |
ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®); |
| 731 |
if (ret < 0) { |
| 732 |
DPRINTF("Unable to set SLB shadow state to KVM: %s\n", strerror(errno));
|
| 733 |
return ret;
|
| 734 |
} |
| 735 |
|
| 736 |
assert((uintptr_t)&env->dtl_size == ((uintptr_t)&env->dtl_addr + 8));
|
| 737 |
reg.id = KVM_REG_PPC_VPA_DTL; |
| 738 |
reg.addr = (uintptr_t)&env->dtl_addr; |
| 739 |
ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®); |
| 740 |
if (ret < 0) { |
| 741 |
DPRINTF("Unable to set dispatch trace log state to KVM: %s\n",
|
| 742 |
strerror(errno)); |
| 743 |
return ret;
|
| 744 |
} |
| 745 |
|
| 746 |
if (!env->vpa_addr) {
|
| 747 |
reg.id = KVM_REG_PPC_VPA_ADDR; |
| 748 |
reg.addr = (uintptr_t)&env->vpa_addr; |
| 749 |
ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®); |
| 750 |
if (ret < 0) { |
| 751 |
DPRINTF("Unable to set VPA address to KVM: %s\n", strerror(errno));
|
| 752 |
return ret;
|
| 753 |
} |
| 754 |
} |
| 755 |
|
| 756 |
return 0; |
| 757 |
} |
| 758 |
#endif /* TARGET_PPC64 */ |
| 759 |
|
| 760 |
int kvm_arch_put_registers(CPUState *cs, int level) |
| 761 |
{
|
| 762 |
PowerPCCPU *cpu = POWERPC_CPU(cs); |
| 763 |
CPUPPCState *env = &cpu->env; |
| 764 |
struct kvm_regs regs;
|
| 765 |
int ret;
|
| 766 |
int i;
|
| 767 |
|
| 768 |
ret = kvm_vcpu_ioctl(cs, KVM_GET_REGS, ®s); |
| 769 |
if (ret < 0) { |
| 770 |
return ret;
|
| 771 |
} |
| 772 |
|
| 773 |
regs.ctr = env->ctr; |
| 774 |
regs.lr = env->lr; |
| 775 |
regs.xer = cpu_read_xer(env); |
| 776 |
regs.msr = env->msr; |
| 777 |
regs.pc = env->nip; |
| 778 |
|
| 779 |
regs.srr0 = env->spr[SPR_SRR0]; |
| 780 |
regs.srr1 = env->spr[SPR_SRR1]; |
| 781 |
|
| 782 |
regs.sprg0 = env->spr[SPR_SPRG0]; |
| 783 |
regs.sprg1 = env->spr[SPR_SPRG1]; |
| 784 |
regs.sprg2 = env->spr[SPR_SPRG2]; |
| 785 |
regs.sprg3 = env->spr[SPR_SPRG3]; |
| 786 |
regs.sprg4 = env->spr[SPR_SPRG4]; |
| 787 |
regs.sprg5 = env->spr[SPR_SPRG5]; |
| 788 |
regs.sprg6 = env->spr[SPR_SPRG6]; |
| 789 |
regs.sprg7 = env->spr[SPR_SPRG7]; |
| 790 |
|
| 791 |
regs.pid = env->spr[SPR_BOOKE_PID]; |
| 792 |
|
| 793 |
for (i = 0;i < 32; i++) |
| 794 |
regs.gpr[i] = env->gpr[i]; |
| 795 |
|
| 796 |
regs.cr = 0;
|
| 797 |
for (i = 0; i < 8; i++) { |
| 798 |
regs.cr |= (env->crf[i] & 15) << (4 * (7 - i)); |
| 799 |
} |
| 800 |
|
| 801 |
ret = kvm_vcpu_ioctl(cs, KVM_SET_REGS, ®s); |
| 802 |
if (ret < 0) |
| 803 |
return ret;
|
| 804 |
|
| 805 |
kvm_put_fp(cs); |
| 806 |
|
| 807 |
if (env->tlb_dirty) {
|
| 808 |
kvm_sw_tlb_put(cpu); |
| 809 |
env->tlb_dirty = false;
|
| 810 |
} |
| 811 |
|
| 812 |
if (cap_segstate && (level >= KVM_PUT_RESET_STATE)) {
|
| 813 |
struct kvm_sregs sregs;
|
| 814 |
|
| 815 |
sregs.pvr = env->spr[SPR_PVR]; |
| 816 |
|
| 817 |
sregs.u.s.sdr1 = env->spr[SPR_SDR1]; |
| 818 |
|
| 819 |
/* Sync SLB */
|
| 820 |
#ifdef TARGET_PPC64
|
| 821 |
for (i = 0; i < ARRAY_SIZE(env->slb); i++) { |
| 822 |
sregs.u.s.ppc64.slb[i].slbe = env->slb[i].esid; |
| 823 |
sregs.u.s.ppc64.slb[i].slbv = env->slb[i].vsid; |
| 824 |
} |
| 825 |
#endif
|
| 826 |
|
| 827 |
/* Sync SRs */
|
| 828 |
for (i = 0; i < 16; i++) { |
| 829 |
sregs.u.s.ppc32.sr[i] = env->sr[i]; |
| 830 |
} |
| 831 |
|
| 832 |
/* Sync BATs */
|
| 833 |
for (i = 0; i < 8; i++) { |
| 834 |
/* Beware. We have to swap upper and lower bits here */
|
| 835 |
sregs.u.s.ppc32.dbat[i] = ((uint64_t)env->DBAT[0][i] << 32) |
| 836 |
| env->DBAT[1][i];
|
| 837 |
sregs.u.s.ppc32.ibat[i] = ((uint64_t)env->IBAT[0][i] << 32) |
| 838 |
| env->IBAT[1][i];
|
| 839 |
} |
| 840 |
|
| 841 |
ret = kvm_vcpu_ioctl(cs, KVM_SET_SREGS, &sregs); |
| 842 |
if (ret) {
|
| 843 |
return ret;
|
| 844 |
} |
| 845 |
} |
| 846 |
|
| 847 |
if (cap_hior && (level >= KVM_PUT_RESET_STATE)) {
|
| 848 |
kvm_put_one_spr(cs, KVM_REG_PPC_HIOR, SPR_HIOR); |
| 849 |
} |
| 850 |
|
| 851 |
if (cap_one_reg) {
|
| 852 |
int i;
|
| 853 |
|
| 854 |
/* We deliberately ignore errors here, for kernels which have
|
| 855 |
* the ONE_REG calls, but don't support the specific
|
| 856 |
* registers, there's a reasonable chance things will still
|
| 857 |
* work, at least until we try to migrate. */
|
| 858 |
for (i = 0; i < 1024; i++) { |
| 859 |
uint64_t id = env->spr_cb[i].one_reg_id; |
| 860 |
|
| 861 |
if (id != 0) { |
| 862 |
kvm_put_one_spr(cs, id, i); |
| 863 |
} |
| 864 |
} |
| 865 |
|
| 866 |
#ifdef TARGET_PPC64
|
| 867 |
if (cap_papr) {
|
| 868 |
if (kvm_put_vpa(cs) < 0) { |
| 869 |
DPRINTF("Warning: Unable to set VPA information to KVM\n");
|
| 870 |
} |
| 871 |
} |
| 872 |
#endif /* TARGET_PPC64 */ |
| 873 |
} |
| 874 |
|
| 875 |
return ret;
|
| 876 |
} |
| 877 |
|
| 878 |
int kvm_arch_get_registers(CPUState *cs)
|
| 879 |
{
|
| 880 |
PowerPCCPU *cpu = POWERPC_CPU(cs); |
| 881 |
CPUPPCState *env = &cpu->env; |
| 882 |
struct kvm_regs regs;
|
| 883 |
struct kvm_sregs sregs;
|
| 884 |
uint32_t cr; |
| 885 |
int i, ret;
|
| 886 |
|
| 887 |
ret = kvm_vcpu_ioctl(cs, KVM_GET_REGS, ®s); |
| 888 |
if (ret < 0) |
| 889 |
return ret;
|
| 890 |
|
| 891 |
cr = regs.cr; |
| 892 |
for (i = 7; i >= 0; i--) { |
| 893 |
env->crf[i] = cr & 15;
|
| 894 |
cr >>= 4;
|
| 895 |
} |
| 896 |
|
| 897 |
env->ctr = regs.ctr; |
| 898 |
env->lr = regs.lr; |
| 899 |
cpu_write_xer(env, regs.xer); |
| 900 |
env->msr = regs.msr; |
| 901 |
env->nip = regs.pc; |
| 902 |
|
| 903 |
env->spr[SPR_SRR0] = regs.srr0; |
| 904 |
env->spr[SPR_SRR1] = regs.srr1; |
| 905 |
|
| 906 |
env->spr[SPR_SPRG0] = regs.sprg0; |
| 907 |
env->spr[SPR_SPRG1] = regs.sprg1; |
| 908 |
env->spr[SPR_SPRG2] = regs.sprg2; |
| 909 |
env->spr[SPR_SPRG3] = regs.sprg3; |
| 910 |
env->spr[SPR_SPRG4] = regs.sprg4; |
| 911 |
env->spr[SPR_SPRG5] = regs.sprg5; |
| 912 |
env->spr[SPR_SPRG6] = regs.sprg6; |
| 913 |
env->spr[SPR_SPRG7] = regs.sprg7; |
| 914 |
|
| 915 |
env->spr[SPR_BOOKE_PID] = regs.pid; |
| 916 |
|
| 917 |
for (i = 0;i < 32; i++) |
| 918 |
env->gpr[i] = regs.gpr[i]; |
| 919 |
|
| 920 |
kvm_get_fp(cs); |
| 921 |
|
| 922 |
if (cap_booke_sregs) {
|
| 923 |
ret = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs); |
| 924 |
if (ret < 0) { |
| 925 |
return ret;
|
| 926 |
} |
| 927 |
|
| 928 |
if (sregs.u.e.features & KVM_SREGS_E_BASE) {
|
| 929 |
env->spr[SPR_BOOKE_CSRR0] = sregs.u.e.csrr0; |
| 930 |
env->spr[SPR_BOOKE_CSRR1] = sregs.u.e.csrr1; |
| 931 |
env->spr[SPR_BOOKE_ESR] = sregs.u.e.esr; |
| 932 |
env->spr[SPR_BOOKE_DEAR] = sregs.u.e.dear; |
| 933 |
env->spr[SPR_BOOKE_MCSR] = sregs.u.e.mcsr; |
| 934 |
env->spr[SPR_BOOKE_TSR] = sregs.u.e.tsr; |
| 935 |
env->spr[SPR_BOOKE_TCR] = sregs.u.e.tcr; |
| 936 |
env->spr[SPR_DECR] = sregs.u.e.dec; |
| 937 |
env->spr[SPR_TBL] = sregs.u.e.tb & 0xffffffff;
|
| 938 |
env->spr[SPR_TBU] = sregs.u.e.tb >> 32;
|
| 939 |
env->spr[SPR_VRSAVE] = sregs.u.e.vrsave; |
| 940 |
} |
| 941 |
|
| 942 |
if (sregs.u.e.features & KVM_SREGS_E_ARCH206) {
|
| 943 |
env->spr[SPR_BOOKE_PIR] = sregs.u.e.pir; |
| 944 |
env->spr[SPR_BOOKE_MCSRR0] = sregs.u.e.mcsrr0; |
| 945 |
env->spr[SPR_BOOKE_MCSRR1] = sregs.u.e.mcsrr1; |
| 946 |
env->spr[SPR_BOOKE_DECAR] = sregs.u.e.decar; |
| 947 |
env->spr[SPR_BOOKE_IVPR] = sregs.u.e.ivpr; |
| 948 |
} |
| 949 |
|
| 950 |
if (sregs.u.e.features & KVM_SREGS_E_64) {
|
| 951 |
env->spr[SPR_BOOKE_EPCR] = sregs.u.e.epcr; |
| 952 |
} |
| 953 |
|
| 954 |
if (sregs.u.e.features & KVM_SREGS_E_SPRG8) {
|
| 955 |
env->spr[SPR_BOOKE_SPRG8] = sregs.u.e.sprg8; |
| 956 |
} |
| 957 |
|
| 958 |
if (sregs.u.e.features & KVM_SREGS_E_IVOR) {
|
| 959 |
env->spr[SPR_BOOKE_IVOR0] = sregs.u.e.ivor_low[0];
|
| 960 |
env->spr[SPR_BOOKE_IVOR1] = sregs.u.e.ivor_low[1];
|
| 961 |
env->spr[SPR_BOOKE_IVOR2] = sregs.u.e.ivor_low[2];
|
| 962 |
env->spr[SPR_BOOKE_IVOR3] = sregs.u.e.ivor_low[3];
|
| 963 |
env->spr[SPR_BOOKE_IVOR4] = sregs.u.e.ivor_low[4];
|
| 964 |
env->spr[SPR_BOOKE_IVOR5] = sregs.u.e.ivor_low[5];
|
| 965 |
env->spr[SPR_BOOKE_IVOR6] = sregs.u.e.ivor_low[6];
|
| 966 |
env->spr[SPR_BOOKE_IVOR7] = sregs.u.e.ivor_low[7];
|
| 967 |
env->spr[SPR_BOOKE_IVOR8] = sregs.u.e.ivor_low[8];
|
| 968 |
env->spr[SPR_BOOKE_IVOR9] = sregs.u.e.ivor_low[9];
|
| 969 |
env->spr[SPR_BOOKE_IVOR10] = sregs.u.e.ivor_low[10];
|
| 970 |
env->spr[SPR_BOOKE_IVOR11] = sregs.u.e.ivor_low[11];
|
| 971 |
env->spr[SPR_BOOKE_IVOR12] = sregs.u.e.ivor_low[12];
|
| 972 |
env->spr[SPR_BOOKE_IVOR13] = sregs.u.e.ivor_low[13];
|
| 973 |
env->spr[SPR_BOOKE_IVOR14] = sregs.u.e.ivor_low[14];
|
| 974 |
env->spr[SPR_BOOKE_IVOR15] = sregs.u.e.ivor_low[15];
|
| 975 |
|
| 976 |
if (sregs.u.e.features & KVM_SREGS_E_SPE) {
|
| 977 |
env->spr[SPR_BOOKE_IVOR32] = sregs.u.e.ivor_high[0];
|
| 978 |
env->spr[SPR_BOOKE_IVOR33] = sregs.u.e.ivor_high[1];
|
| 979 |
env->spr[SPR_BOOKE_IVOR34] = sregs.u.e.ivor_high[2];
|
| 980 |
} |
| 981 |
|
| 982 |
if (sregs.u.e.features & KVM_SREGS_E_PM) {
|
| 983 |
env->spr[SPR_BOOKE_IVOR35] = sregs.u.e.ivor_high[3];
|
| 984 |
} |
| 985 |
|
| 986 |
if (sregs.u.e.features & KVM_SREGS_E_PC) {
|
| 987 |
env->spr[SPR_BOOKE_IVOR36] = sregs.u.e.ivor_high[4];
|
| 988 |
env->spr[SPR_BOOKE_IVOR37] = sregs.u.e.ivor_high[5];
|
| 989 |
} |
| 990 |
} |
| 991 |
|
| 992 |
if (sregs.u.e.features & KVM_SREGS_E_ARCH206_MMU) {
|
| 993 |
env->spr[SPR_BOOKE_MAS0] = sregs.u.e.mas0; |
| 994 |
env->spr[SPR_BOOKE_MAS1] = sregs.u.e.mas1; |
| 995 |
env->spr[SPR_BOOKE_MAS2] = sregs.u.e.mas2; |
| 996 |
env->spr[SPR_BOOKE_MAS3] = sregs.u.e.mas7_3 & 0xffffffff;
|
| 997 |
env->spr[SPR_BOOKE_MAS4] = sregs.u.e.mas4; |
| 998 |
env->spr[SPR_BOOKE_MAS6] = sregs.u.e.mas6; |
| 999 |
env->spr[SPR_BOOKE_MAS7] = sregs.u.e.mas7_3 >> 32;
|
| 1000 |
env->spr[SPR_MMUCFG] = sregs.u.e.mmucfg; |
| 1001 |
env->spr[SPR_BOOKE_TLB0CFG] = sregs.u.e.tlbcfg[0];
|
| 1002 |
env->spr[SPR_BOOKE_TLB1CFG] = sregs.u.e.tlbcfg[1];
|
| 1003 |
} |
| 1004 |
|
| 1005 |
if (sregs.u.e.features & KVM_SREGS_EXP) {
|
| 1006 |
env->spr[SPR_BOOKE_EPR] = sregs.u.e.epr; |
| 1007 |
} |
| 1008 |
|
| 1009 |
if (sregs.u.e.features & KVM_SREGS_E_PD) {
|
| 1010 |
env->spr[SPR_BOOKE_EPLC] = sregs.u.e.eplc; |
| 1011 |
env->spr[SPR_BOOKE_EPSC] = sregs.u.e.epsc; |
| 1012 |
} |
| 1013 |
|
| 1014 |
if (sregs.u.e.impl_id == KVM_SREGS_E_IMPL_FSL) {
|
| 1015 |
env->spr[SPR_E500_SVR] = sregs.u.e.impl.fsl.svr; |
| 1016 |
env->spr[SPR_Exxx_MCAR] = sregs.u.e.impl.fsl.mcar; |
| 1017 |
env->spr[SPR_HID0] = sregs.u.e.impl.fsl.hid0; |
| 1018 |
|
| 1019 |
if (sregs.u.e.impl.fsl.features & KVM_SREGS_E_FSL_PIDn) {
|
| 1020 |
env->spr[SPR_BOOKE_PID1] = sregs.u.e.impl.fsl.pid1; |
| 1021 |
env->spr[SPR_BOOKE_PID2] = sregs.u.e.impl.fsl.pid2; |
| 1022 |
} |
| 1023 |
} |
| 1024 |
} |
| 1025 |
|
| 1026 |
if (cap_segstate) {
|
| 1027 |
ret = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs); |
| 1028 |
if (ret < 0) { |
| 1029 |
return ret;
|
| 1030 |
} |
| 1031 |
|
| 1032 |
ppc_store_sdr1(env, sregs.u.s.sdr1); |
| 1033 |
|
| 1034 |
/* Sync SLB */
|
| 1035 |
#ifdef TARGET_PPC64
|
| 1036 |
/*
|
| 1037 |
* The packed SLB array we get from KVM_GET_SREGS only contains
|
| 1038 |
* information about valid entries. So we flush our internal
|
| 1039 |
* copy to get rid of stale ones, then put all valid SLB entries
|
| 1040 |
* back in.
|
| 1041 |
*/
|
| 1042 |
memset(env->slb, 0, sizeof(env->slb)); |
| 1043 |
for (i = 0; i < ARRAY_SIZE(env->slb); i++) { |
| 1044 |
target_ulong rb = sregs.u.s.ppc64.slb[i].slbe; |
| 1045 |
target_ulong rs = sregs.u.s.ppc64.slb[i].slbv; |
| 1046 |
/*
|
| 1047 |
* Only restore valid entries
|
| 1048 |
*/
|
| 1049 |
if (rb & SLB_ESID_V) {
|
| 1050 |
ppc_store_slb(env, rb, rs); |
| 1051 |
} |
| 1052 |
} |
| 1053 |
#endif
|
| 1054 |
|
| 1055 |
/* Sync SRs */
|
| 1056 |
for (i = 0; i < 16; i++) { |
| 1057 |
env->sr[i] = sregs.u.s.ppc32.sr[i]; |
| 1058 |
} |
| 1059 |
|
| 1060 |
/* Sync BATs */
|
| 1061 |
for (i = 0; i < 8; i++) { |
| 1062 |
env->DBAT[0][i] = sregs.u.s.ppc32.dbat[i] & 0xffffffff; |
| 1063 |
env->DBAT[1][i] = sregs.u.s.ppc32.dbat[i] >> 32; |
| 1064 |
env->IBAT[0][i] = sregs.u.s.ppc32.ibat[i] & 0xffffffff; |
| 1065 |
env->IBAT[1][i] = sregs.u.s.ppc32.ibat[i] >> 32; |
| 1066 |
} |
| 1067 |
} |
| 1068 |
|
| 1069 |
if (cap_hior) {
|
| 1070 |
kvm_get_one_spr(cs, KVM_REG_PPC_HIOR, SPR_HIOR); |
| 1071 |
} |
| 1072 |
|
| 1073 |
if (cap_one_reg) {
|
| 1074 |
int i;
|
| 1075 |
|
| 1076 |
/* We deliberately ignore errors here, for kernels which have
|
| 1077 |
* the ONE_REG calls, but don't support the specific
|
| 1078 |
* registers, there's a reasonable chance things will still
|
| 1079 |
* work, at least until we try to migrate. */
|
| 1080 |
for (i = 0; i < 1024; i++) { |
| 1081 |
uint64_t id = env->spr_cb[i].one_reg_id; |
| 1082 |
|
| 1083 |
if (id != 0) { |
| 1084 |
kvm_get_one_spr(cs, id, i); |
| 1085 |
} |
| 1086 |
} |
| 1087 |
|
| 1088 |
#ifdef TARGET_PPC64
|
| 1089 |
if (cap_papr) {
|
| 1090 |
if (kvm_get_vpa(cs) < 0) { |
| 1091 |
DPRINTF("Warning: Unable to get VPA information from KVM\n");
|
| 1092 |
} |
| 1093 |
} |
| 1094 |
#endif
|
| 1095 |
} |
| 1096 |
|
| 1097 |
return 0; |
| 1098 |
} |
| 1099 |
|
| 1100 |
int kvmppc_set_interrupt(PowerPCCPU *cpu, int irq, int level) |
| 1101 |
{
|
| 1102 |
unsigned virq = level ? KVM_INTERRUPT_SET_LEVEL : KVM_INTERRUPT_UNSET;
|
| 1103 |
|
| 1104 |
if (irq != PPC_INTERRUPT_EXT) {
|
| 1105 |
return 0; |
| 1106 |
} |
| 1107 |
|
| 1108 |
if (!kvm_enabled() || !cap_interrupt_unset || !cap_interrupt_level) {
|
| 1109 |
return 0; |
| 1110 |
} |
| 1111 |
|
| 1112 |
kvm_vcpu_ioctl(CPU(cpu), KVM_INTERRUPT, &virq); |
| 1113 |
|
| 1114 |
return 0; |
| 1115 |
} |
| 1116 |
|
| 1117 |
#if defined(TARGET_PPCEMB)
|
| 1118 |
#define PPC_INPUT_INT PPC40x_INPUT_INT
|
| 1119 |
#elif defined(TARGET_PPC64)
|
| 1120 |
#define PPC_INPUT_INT PPC970_INPUT_INT
|
| 1121 |
#else
|
| 1122 |
#define PPC_INPUT_INT PPC6xx_INPUT_INT
|
| 1123 |
#endif
|
| 1124 |
|
| 1125 |
void kvm_arch_pre_run(CPUState *cs, struct kvm_run *run) |
| 1126 |
{
|
| 1127 |
PowerPCCPU *cpu = POWERPC_CPU(cs); |
| 1128 |
CPUPPCState *env = &cpu->env; |
| 1129 |
int r;
|
| 1130 |
unsigned irq;
|
| 1131 |
|
| 1132 |
/* PowerPC QEMU tracks the various core input pins (interrupt, critical
|
| 1133 |
* interrupt, reset, etc) in PPC-specific env->irq_input_state. */
|
| 1134 |
if (!cap_interrupt_level &&
|
| 1135 |
run->ready_for_interrupt_injection && |
| 1136 |
(cs->interrupt_request & CPU_INTERRUPT_HARD) && |
| 1137 |
(env->irq_input_state & (1<<PPC_INPUT_INT)))
|
| 1138 |
{
|
| 1139 |
/* For now KVM disregards the 'irq' argument. However, in the
|
| 1140 |
* future KVM could cache it in-kernel to avoid a heavyweight exit
|
| 1141 |
* when reading the UIC.
|
| 1142 |
*/
|
| 1143 |
irq = KVM_INTERRUPT_SET; |
| 1144 |
|
| 1145 |
DPRINTF("injected interrupt %d\n", irq);
|
| 1146 |
r = kvm_vcpu_ioctl(cs, KVM_INTERRUPT, &irq); |
| 1147 |
if (r < 0) { |
| 1148 |
printf("cpu %d fail inject %x\n", cs->cpu_index, irq);
|
| 1149 |
} |
| 1150 |
|
| 1151 |
/* Always wake up soon in case the interrupt was level based */
|
| 1152 |
timer_mod(idle_timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + |
| 1153 |
(get_ticks_per_sec() / 50));
|
| 1154 |
} |
| 1155 |
|
| 1156 |
/* We don't know if there are more interrupts pending after this. However,
|
| 1157 |
* the guest will return to userspace in the course of handling this one
|
| 1158 |
* anyways, so we will get a chance to deliver the rest. */
|
| 1159 |
} |
| 1160 |
|
| 1161 |
void kvm_arch_post_run(CPUState *cpu, struct kvm_run *run) |
| 1162 |
{
|
| 1163 |
} |
| 1164 |
|
| 1165 |
int kvm_arch_process_async_events(CPUState *cs)
|
| 1166 |
{
|
| 1167 |
return cs->halted;
|
| 1168 |
} |
| 1169 |
|
| 1170 |
static int kvmppc_handle_halt(PowerPCCPU *cpu) |
| 1171 |
{
|
| 1172 |
CPUState *cs = CPU(cpu); |
| 1173 |
CPUPPCState *env = &cpu->env; |
| 1174 |
|
| 1175 |
if (!(cs->interrupt_request & CPU_INTERRUPT_HARD) && (msr_ee)) {
|
| 1176 |
cs->halted = 1;
|
| 1177 |
env->exception_index = EXCP_HLT; |
| 1178 |
} |
| 1179 |
|
| 1180 |
return 0; |
| 1181 |
} |
| 1182 |
|
| 1183 |
/* map dcr access to existing qemu dcr emulation */
|
| 1184 |
static int kvmppc_handle_dcr_read(CPUPPCState *env, uint32_t dcrn, uint32_t *data) |
| 1185 |
{
|
| 1186 |
if (ppc_dcr_read(env->dcr_env, dcrn, data) < 0) |
| 1187 |
fprintf(stderr, "Read to unhandled DCR (0x%x)\n", dcrn);
|
| 1188 |
|
| 1189 |
return 0; |
| 1190 |
} |
| 1191 |
|
| 1192 |
static int kvmppc_handle_dcr_write(CPUPPCState *env, uint32_t dcrn, uint32_t data) |
| 1193 |
{
|
| 1194 |
if (ppc_dcr_write(env->dcr_env, dcrn, data) < 0) |
| 1195 |
fprintf(stderr, "Write to unhandled DCR (0x%x)\n", dcrn);
|
| 1196 |
|
| 1197 |
return 0; |
| 1198 |
} |
| 1199 |
|
| 1200 |
int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run) |
| 1201 |
{
|
| 1202 |
PowerPCCPU *cpu = POWERPC_CPU(cs); |
| 1203 |
CPUPPCState *env = &cpu->env; |
| 1204 |
int ret;
|
| 1205 |
|
| 1206 |
switch (run->exit_reason) {
|
| 1207 |
case KVM_EXIT_DCR:
|
| 1208 |
if (run->dcr.is_write) {
|
| 1209 |
DPRINTF("handle dcr write\n");
|
| 1210 |
ret = kvmppc_handle_dcr_write(env, run->dcr.dcrn, run->dcr.data); |
| 1211 |
} else {
|
| 1212 |
DPRINTF("handle dcr read\n");
|
| 1213 |
ret = kvmppc_handle_dcr_read(env, run->dcr.dcrn, &run->dcr.data); |
| 1214 |
} |
| 1215 |
break;
|
| 1216 |
case KVM_EXIT_HLT:
|
| 1217 |
DPRINTF("handle halt\n");
|
| 1218 |
ret = kvmppc_handle_halt(cpu); |
| 1219 |
break;
|
| 1220 |
#if defined(TARGET_PPC64)
|
| 1221 |
case KVM_EXIT_PAPR_HCALL:
|
| 1222 |
DPRINTF("handle PAPR hypercall\n");
|
| 1223 |
run->papr_hcall.ret = spapr_hypercall(cpu, |
| 1224 |
run->papr_hcall.nr, |
| 1225 |
run->papr_hcall.args); |
| 1226 |
ret = 0;
|
| 1227 |
break;
|
| 1228 |
#endif
|
| 1229 |
case KVM_EXIT_EPR:
|
| 1230 |
DPRINTF("handle epr\n");
|
| 1231 |
run->epr.epr = ldl_phys(env->mpic_iack); |
| 1232 |
ret = 0;
|
| 1233 |
break;
|
| 1234 |
case KVM_EXIT_WATCHDOG:
|
| 1235 |
DPRINTF("handle watchdog expiry\n");
|
| 1236 |
watchdog_perform_action(); |
| 1237 |
ret = 0;
|
| 1238 |
break;
|
| 1239 |
|
| 1240 |
default:
|
| 1241 |
fprintf(stderr, "KVM: unknown exit reason %d\n", run->exit_reason);
|
| 1242 |
ret = -1;
|
| 1243 |
break;
|
| 1244 |
} |
| 1245 |
|
| 1246 |
return ret;
|
| 1247 |
} |
| 1248 |
|
| 1249 |
int kvmppc_or_tsr_bits(PowerPCCPU *cpu, uint32_t tsr_bits)
|
| 1250 |
{
|
| 1251 |
CPUState *cs = CPU(cpu); |
| 1252 |
uint32_t bits = tsr_bits; |
| 1253 |
struct kvm_one_reg reg = {
|
| 1254 |
.id = KVM_REG_PPC_OR_TSR, |
| 1255 |
.addr = (uintptr_t) &bits, |
| 1256 |
}; |
| 1257 |
|
| 1258 |
return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
|
| 1259 |
} |
| 1260 |
|
| 1261 |
int kvmppc_clear_tsr_bits(PowerPCCPU *cpu, uint32_t tsr_bits)
|
| 1262 |
{
|
| 1263 |
|
| 1264 |
CPUState *cs = CPU(cpu); |
| 1265 |
uint32_t bits = tsr_bits; |
| 1266 |
struct kvm_one_reg reg = {
|
| 1267 |
.id = KVM_REG_PPC_CLEAR_TSR, |
| 1268 |
.addr = (uintptr_t) &bits, |
| 1269 |
}; |
| 1270 |
|
| 1271 |
return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
|
| 1272 |
} |
| 1273 |
|
| 1274 |
int kvmppc_set_tcr(PowerPCCPU *cpu)
|
| 1275 |
{
|
| 1276 |
CPUState *cs = CPU(cpu); |
| 1277 |
CPUPPCState *env = &cpu->env; |
| 1278 |
uint32_t tcr = env->spr[SPR_BOOKE_TCR]; |
| 1279 |
|
| 1280 |
struct kvm_one_reg reg = {
|
| 1281 |
.id = KVM_REG_PPC_TCR, |
| 1282 |
.addr = (uintptr_t) &tcr, |
| 1283 |
}; |
| 1284 |
|
| 1285 |
return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
|
| 1286 |
} |
| 1287 |
|
| 1288 |
int kvmppc_booke_watchdog_enable(PowerPCCPU *cpu)
|
| 1289 |
{
|
| 1290 |
CPUState *cs = CPU(cpu); |
| 1291 |
struct kvm_enable_cap encap = {};
|
| 1292 |
int ret;
|
| 1293 |
|
| 1294 |
if (!kvm_enabled()) {
|
| 1295 |
return -1; |
| 1296 |
} |
| 1297 |
|
| 1298 |
if (!cap_ppc_watchdog) {
|
| 1299 |
printf("warning: KVM does not support watchdog");
|
| 1300 |
return -1; |
| 1301 |
} |
| 1302 |
|
| 1303 |
encap.cap = KVM_CAP_PPC_BOOKE_WATCHDOG; |
| 1304 |
ret = kvm_vcpu_ioctl(cs, KVM_ENABLE_CAP, &encap); |
| 1305 |
if (ret < 0) { |
| 1306 |
fprintf(stderr, "%s: couldn't enable KVM_CAP_PPC_BOOKE_WATCHDOG: %s\n",
|
| 1307 |
__func__, strerror(-ret)); |
| 1308 |
return ret;
|
| 1309 |
} |
| 1310 |
|
| 1311 |
return ret;
|
| 1312 |
} |
| 1313 |
|
| 1314 |
static int read_cpuinfo(const char *field, char *value, int len) |
| 1315 |
{
|
| 1316 |
FILE *f; |
| 1317 |
int ret = -1; |
| 1318 |
int field_len = strlen(field);
|
| 1319 |
char line[512]; |
| 1320 |
|
| 1321 |
f = fopen("/proc/cpuinfo", "r"); |
| 1322 |
if (!f) {
|
| 1323 |
return -1; |
| 1324 |
} |
| 1325 |
|
| 1326 |
do {
|
| 1327 |
if(!fgets(line, sizeof(line), f)) { |
| 1328 |
break;
|
| 1329 |
} |
| 1330 |
if (!strncmp(line, field, field_len)) {
|
| 1331 |
pstrcpy(value, len, line); |
| 1332 |
ret = 0;
|
| 1333 |
break;
|
| 1334 |
} |
| 1335 |
} while(*line);
|
| 1336 |
|
| 1337 |
fclose(f); |
| 1338 |
|
| 1339 |
return ret;
|
| 1340 |
} |
| 1341 |
|
| 1342 |
uint32_t kvmppc_get_tbfreq(void)
|
| 1343 |
{
|
| 1344 |
char line[512]; |
| 1345 |
char *ns;
|
| 1346 |
uint32_t retval = get_ticks_per_sec(); |
| 1347 |
|
| 1348 |
if (read_cpuinfo("timebase", line, sizeof(line))) { |
| 1349 |
return retval;
|
| 1350 |
} |
| 1351 |
|
| 1352 |
if (!(ns = strchr(line, ':'))) { |
| 1353 |
return retval;
|
| 1354 |
} |
| 1355 |
|
| 1356 |
ns++; |
| 1357 |
|
| 1358 |
retval = atoi(ns); |
| 1359 |
return retval;
|
| 1360 |
} |
| 1361 |
|
| 1362 |
/* Try to find a device tree node for a CPU with clock-frequency property */
|
| 1363 |
static int kvmppc_find_cpu_dt(char *buf, int buf_len) |
| 1364 |
{
|
| 1365 |
struct dirent *dirp;
|
| 1366 |
DIR *dp; |
| 1367 |
|
| 1368 |
if ((dp = opendir(PROC_DEVTREE_CPU)) == NULL) { |
| 1369 |
printf("Can't open directory " PROC_DEVTREE_CPU "\n"); |
| 1370 |
return -1; |
| 1371 |
} |
| 1372 |
|
| 1373 |
buf[0] = '\0'; |
| 1374 |
while ((dirp = readdir(dp)) != NULL) { |
| 1375 |
FILE *f; |
| 1376 |
snprintf(buf, buf_len, "%s%s/clock-frequency", PROC_DEVTREE_CPU,
|
| 1377 |
dirp->d_name); |
| 1378 |
f = fopen(buf, "r");
|
| 1379 |
if (f) {
|
| 1380 |
snprintf(buf, buf_len, "%s%s", PROC_DEVTREE_CPU, dirp->d_name);
|
| 1381 |
fclose(f); |
| 1382 |
break;
|
| 1383 |
} |
| 1384 |
buf[0] = '\0'; |
| 1385 |
} |
| 1386 |
closedir(dp); |
| 1387 |
if (buf[0] == '\0') { |
| 1388 |
printf("Unknown host!\n");
|
| 1389 |
return -1; |
| 1390 |
} |
| 1391 |
|
| 1392 |
return 0; |
| 1393 |
} |
| 1394 |
|
| 1395 |
/* Read a CPU node property from the host device tree that's a single
|
| 1396 |
* integer (32-bit or 64-bit). Returns 0 if anything goes wrong
|
| 1397 |
* (can't find or open the property, or doesn't understand the
|
| 1398 |
* format) */
|
| 1399 |
static uint64_t kvmppc_read_int_cpu_dt(const char *propname) |
| 1400 |
{
|
| 1401 |
char buf[PATH_MAX];
|
| 1402 |
union {
|
| 1403 |
uint32_t v32; |
| 1404 |
uint64_t v64; |
| 1405 |
} u; |
| 1406 |
FILE *f; |
| 1407 |
int len;
|
| 1408 |
|
| 1409 |
if (kvmppc_find_cpu_dt(buf, sizeof(buf))) { |
| 1410 |
return -1; |
| 1411 |
} |
| 1412 |
|
| 1413 |
strncat(buf, "/", sizeof(buf) - strlen(buf)); |
| 1414 |
strncat(buf, propname, sizeof(buf) - strlen(buf));
|
| 1415 |
|
| 1416 |
f = fopen(buf, "rb");
|
| 1417 |
if (!f) {
|
| 1418 |
return -1; |
| 1419 |
} |
| 1420 |
|
| 1421 |
len = fread(&u, 1, sizeof(u), f); |
| 1422 |
fclose(f); |
| 1423 |
switch (len) {
|
| 1424 |
case 4: |
| 1425 |
/* property is a 32-bit quantity */
|
| 1426 |
return be32_to_cpu(u.v32);
|
| 1427 |
case 8: |
| 1428 |
return be64_to_cpu(u.v64);
|
| 1429 |
} |
| 1430 |
|
| 1431 |
return 0; |
| 1432 |
} |
| 1433 |
|
| 1434 |
uint64_t kvmppc_get_clockfreq(void)
|
| 1435 |
{
|
| 1436 |
return kvmppc_read_int_cpu_dt("clock-frequency"); |
| 1437 |
} |
| 1438 |
|
| 1439 |
uint32_t kvmppc_get_vmx(void)
|
| 1440 |
{
|
| 1441 |
return kvmppc_read_int_cpu_dt("ibm,vmx"); |
| 1442 |
} |
| 1443 |
|
| 1444 |
uint32_t kvmppc_get_dfp(void)
|
| 1445 |
{
|
| 1446 |
return kvmppc_read_int_cpu_dt("ibm,dfp"); |
| 1447 |
} |
| 1448 |
|
| 1449 |
static int kvmppc_get_pvinfo(CPUPPCState *env, struct kvm_ppc_pvinfo *pvinfo) |
| 1450 |
{
|
| 1451 |
PowerPCCPU *cpu = ppc_env_get_cpu(env); |
| 1452 |
CPUState *cs = CPU(cpu); |
| 1453 |
|
| 1454 |
if (kvm_check_extension(cs->kvm_state, KVM_CAP_PPC_GET_PVINFO) &&
|
| 1455 |
!kvm_vm_ioctl(cs->kvm_state, KVM_PPC_GET_PVINFO, pvinfo)) {
|
| 1456 |
return 0; |
| 1457 |
} |
| 1458 |
|
| 1459 |
return 1; |
| 1460 |
} |
| 1461 |
|
| 1462 |
int kvmppc_get_hasidle(CPUPPCState *env)
|
| 1463 |
{
|
| 1464 |
struct kvm_ppc_pvinfo pvinfo;
|
| 1465 |
|
| 1466 |
if (!kvmppc_get_pvinfo(env, &pvinfo) &&
|
| 1467 |
(pvinfo.flags & KVM_PPC_PVINFO_FLAGS_EV_IDLE)) {
|
| 1468 |
return 1; |
| 1469 |
} |
| 1470 |
|
| 1471 |
return 0; |
| 1472 |
} |
| 1473 |
|
| 1474 |
int kvmppc_get_hypercall(CPUPPCState *env, uint8_t *buf, int buf_len) |
| 1475 |
{
|
| 1476 |
uint32_t *hc = (uint32_t*)buf; |
| 1477 |
struct kvm_ppc_pvinfo pvinfo;
|
| 1478 |
|
| 1479 |
if (!kvmppc_get_pvinfo(env, &pvinfo)) {
|
| 1480 |
memcpy(buf, pvinfo.hcall, buf_len); |
| 1481 |
return 0; |
| 1482 |
} |
| 1483 |
|
| 1484 |
/*
|
| 1485 |
* Fallback to always fail hypercalls:
|
| 1486 |
*
|
| 1487 |
* li r3, -1
|
| 1488 |
* nop
|
| 1489 |
* nop
|
| 1490 |
* nop
|
| 1491 |
*/
|
| 1492 |
|
| 1493 |
hc[0] = 0x3860ffff; |
| 1494 |
hc[1] = 0x60000000; |
| 1495 |
hc[2] = 0x60000000; |
| 1496 |
hc[3] = 0x60000000; |
| 1497 |
|
| 1498 |
return 0; |
| 1499 |
} |
| 1500 |
|
| 1501 |
void kvmppc_set_papr(PowerPCCPU *cpu)
|
| 1502 |
{
|
| 1503 |
CPUPPCState *env = &cpu->env; |
| 1504 |
CPUState *cs = CPU(cpu); |
| 1505 |
struct kvm_enable_cap cap = {};
|
| 1506 |
int ret;
|
| 1507 |
|
| 1508 |
cap.cap = KVM_CAP_PPC_PAPR; |
| 1509 |
ret = kvm_vcpu_ioctl(cs, KVM_ENABLE_CAP, &cap); |
| 1510 |
|
| 1511 |
if (ret) {
|
| 1512 |
cpu_abort(env, "This KVM version does not support PAPR\n");
|
| 1513 |
} |
| 1514 |
|
| 1515 |
/* Update the capability flag so we sync the right information
|
| 1516 |
* with kvm */
|
| 1517 |
cap_papr = 1;
|
| 1518 |
} |
| 1519 |
|
| 1520 |
void kvmppc_set_mpic_proxy(PowerPCCPU *cpu, int mpic_proxy) |
| 1521 |
{
|
| 1522 |
CPUPPCState *env = &cpu->env; |
| 1523 |
CPUState *cs = CPU(cpu); |
| 1524 |
struct kvm_enable_cap cap = {};
|
| 1525 |
int ret;
|
| 1526 |
|
| 1527 |
cap.cap = KVM_CAP_PPC_EPR; |
| 1528 |
cap.args[0] = mpic_proxy;
|
| 1529 |
ret = kvm_vcpu_ioctl(cs, KVM_ENABLE_CAP, &cap); |
| 1530 |
|
| 1531 |
if (ret && mpic_proxy) {
|
| 1532 |
cpu_abort(env, "This KVM version does not support EPR\n");
|
| 1533 |
} |
| 1534 |
} |
| 1535 |
|
| 1536 |
int kvmppc_smt_threads(void) |
| 1537 |
{
|
| 1538 |
return cap_ppc_smt ? cap_ppc_smt : 1; |
| 1539 |
} |
| 1540 |
|
| 1541 |
#ifdef TARGET_PPC64
|
| 1542 |
off_t kvmppc_alloc_rma(const char *name, MemoryRegion *sysmem) |
| 1543 |
{
|
| 1544 |
void *rma;
|
| 1545 |
off_t size; |
| 1546 |
int fd;
|
| 1547 |
struct kvm_allocate_rma ret;
|
| 1548 |
MemoryRegion *rma_region; |
| 1549 |
|
| 1550 |
/* If cap_ppc_rma == 0, contiguous RMA allocation is not supported
|
| 1551 |
* if cap_ppc_rma == 1, contiguous RMA allocation is supported, but
|
| 1552 |
* not necessary on this hardware
|
| 1553 |
* if cap_ppc_rma == 2, contiguous RMA allocation is needed on this hardware
|
| 1554 |
*
|
| 1555 |
* FIXME: We should allow the user to force contiguous RMA
|
| 1556 |
* allocation in the cap_ppc_rma==1 case.
|
| 1557 |
*/
|
| 1558 |
if (cap_ppc_rma < 2) { |
| 1559 |
return 0; |
| 1560 |
} |
| 1561 |
|
| 1562 |
fd = kvm_vm_ioctl(kvm_state, KVM_ALLOCATE_RMA, &ret); |
| 1563 |
if (fd < 0) { |
| 1564 |
fprintf(stderr, "KVM: Error on KVM_ALLOCATE_RMA: %s\n",
|
| 1565 |
strerror(errno)); |
| 1566 |
return -1; |
| 1567 |
} |
| 1568 |
|
| 1569 |
size = MIN(ret.rma_size, 256ul << 20); |
| 1570 |
|
| 1571 |
rma = mmap(NULL, size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0); |
| 1572 |
if (rma == MAP_FAILED) {
|
| 1573 |
fprintf(stderr, "KVM: Error mapping RMA: %s\n", strerror(errno));
|
| 1574 |
return -1; |
| 1575 |
}; |
| 1576 |
|
| 1577 |
rma_region = g_new(MemoryRegion, 1);
|
| 1578 |
memory_region_init_ram_ptr(rma_region, NULL, name, size, rma);
|
| 1579 |
vmstate_register_ram_global(rma_region); |
| 1580 |
memory_region_add_subregion(sysmem, 0, rma_region);
|
| 1581 |
|
| 1582 |
return size;
|
| 1583 |
} |
| 1584 |
|
| 1585 |
uint64_t kvmppc_rma_size(uint64_t current_size, unsigned int hash_shift) |
| 1586 |
{
|
| 1587 |
struct kvm_ppc_smmu_info info;
|
| 1588 |
long rampagesize, best_page_shift;
|
| 1589 |
int i;
|
| 1590 |
|
| 1591 |
if (cap_ppc_rma >= 2) { |
| 1592 |
return current_size;
|
| 1593 |
} |
| 1594 |
|
| 1595 |
/* Find the largest hardware supported page size that's less than
|
| 1596 |
* or equal to the (logical) backing page size of guest RAM */
|
| 1597 |
kvm_get_smmu_info(POWERPC_CPU(first_cpu), &info); |
| 1598 |
rampagesize = getrampagesize(); |
| 1599 |
best_page_shift = 0;
|
| 1600 |
|
| 1601 |
for (i = 0; i < KVM_PPC_PAGE_SIZES_MAX_SZ; i++) { |
| 1602 |
struct kvm_ppc_one_seg_page_size *sps = &info.sps[i];
|
| 1603 |
|
| 1604 |
if (!sps->page_shift) {
|
| 1605 |
continue;
|
| 1606 |
} |
| 1607 |
|
| 1608 |
if ((sps->page_shift > best_page_shift)
|
| 1609 |
&& ((1UL << sps->page_shift) <= rampagesize)) {
|
| 1610 |
best_page_shift = sps->page_shift; |
| 1611 |
} |
| 1612 |
} |
| 1613 |
|
| 1614 |
return MIN(current_size,
|
| 1615 |
1ULL << (best_page_shift + hash_shift - 7)); |
| 1616 |
} |
| 1617 |
#endif
|
| 1618 |
|
| 1619 |
void *kvmppc_create_spapr_tce(uint32_t liobn, uint32_t window_size, int *pfd) |
| 1620 |
{
|
| 1621 |
struct kvm_create_spapr_tce args = {
|
| 1622 |
.liobn = liobn, |
| 1623 |
.window_size = window_size, |
| 1624 |
}; |
| 1625 |
long len;
|
| 1626 |
int fd;
|
| 1627 |
void *table;
|
| 1628 |
|
| 1629 |
/* Must set fd to -1 so we don't try to munmap when called for
|
| 1630 |
* destroying the table, which the upper layers -will- do
|
| 1631 |
*/
|
| 1632 |
*pfd = -1;
|
| 1633 |
if (!cap_spapr_tce) {
|
| 1634 |
return NULL; |
| 1635 |
} |
| 1636 |
|
| 1637 |
fd = kvm_vm_ioctl(kvm_state, KVM_CREATE_SPAPR_TCE, &args); |
| 1638 |
if (fd < 0) { |
| 1639 |
fprintf(stderr, "KVM: Failed to create TCE table for liobn 0x%x\n",
|
| 1640 |
liobn); |
| 1641 |
return NULL; |
| 1642 |
} |
| 1643 |
|
| 1644 |
len = (window_size / SPAPR_TCE_PAGE_SIZE) * sizeof(uint64_t);
|
| 1645 |
/* FIXME: round this up to page size */
|
| 1646 |
|
| 1647 |
table = mmap(NULL, len, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0); |
| 1648 |
if (table == MAP_FAILED) {
|
| 1649 |
fprintf(stderr, "KVM: Failed to map TCE table for liobn 0x%x\n",
|
| 1650 |
liobn); |
| 1651 |
close(fd); |
| 1652 |
return NULL; |
| 1653 |
} |
| 1654 |
|
| 1655 |
*pfd = fd; |
| 1656 |
return table;
|
| 1657 |
} |
| 1658 |
|
| 1659 |
int kvmppc_remove_spapr_tce(void *table, int fd, uint32_t window_size) |
| 1660 |
{
|
| 1661 |
long len;
|
| 1662 |
|
| 1663 |
if (fd < 0) { |
| 1664 |
return -1; |
| 1665 |
} |
| 1666 |
|
| 1667 |
len = (window_size / SPAPR_TCE_PAGE_SIZE)*sizeof(uint64_t);
|
| 1668 |
if ((munmap(table, len) < 0) || |
| 1669 |
(close(fd) < 0)) {
|
| 1670 |
fprintf(stderr, "KVM: Unexpected error removing TCE table: %s",
|
| 1671 |
strerror(errno)); |
| 1672 |
/* Leak the table */
|
| 1673 |
} |
| 1674 |
|
| 1675 |
return 0; |
| 1676 |
} |
| 1677 |
|
| 1678 |
int kvmppc_reset_htab(int shift_hint) |
| 1679 |
{
|
| 1680 |
uint32_t shift = shift_hint; |
| 1681 |
|
| 1682 |
if (!kvm_enabled()) {
|
| 1683 |
/* Full emulation, tell caller to allocate htab itself */
|
| 1684 |
return 0; |
| 1685 |
} |
| 1686 |
if (kvm_check_extension(kvm_state, KVM_CAP_PPC_ALLOC_HTAB)) {
|
| 1687 |
int ret;
|
| 1688 |
ret = kvm_vm_ioctl(kvm_state, KVM_PPC_ALLOCATE_HTAB, &shift); |
| 1689 |
if (ret == -ENOTTY) {
|
| 1690 |
/* At least some versions of PR KVM advertise the
|
| 1691 |
* capability, but don't implement the ioctl(). Oops.
|
| 1692 |
* Return 0 so that we allocate the htab in qemu, as is
|
| 1693 |
* correct for PR. */
|
| 1694 |
return 0; |
| 1695 |
} else if (ret < 0) { |
| 1696 |
return ret;
|
| 1697 |
} |
| 1698 |
return shift;
|
| 1699 |
} |
| 1700 |
|
| 1701 |
/* We have a kernel that predates the htab reset calls. For PR
|
| 1702 |
* KVM, we need to allocate the htab ourselves, for an HV KVM of
|
| 1703 |
* this era, it has allocated a 16MB fixed size hash table
|
| 1704 |
* already. Kernels of this era have the GET_PVINFO capability
|
| 1705 |
* only on PR, so we use this hack to determine the right
|
| 1706 |
* answer */
|
| 1707 |
if (kvm_check_extension(kvm_state, KVM_CAP_PPC_GET_PVINFO)) {
|
| 1708 |
/* PR - tell caller to allocate htab */
|
| 1709 |
return 0; |
| 1710 |
} else {
|
| 1711 |
/* HV - assume 16MB kernel allocated htab */
|
| 1712 |
return 24; |
| 1713 |
} |
| 1714 |
} |
| 1715 |
|
| 1716 |
static inline uint32_t mfpvr(void) |
| 1717 |
{
|
| 1718 |
uint32_t pvr; |
| 1719 |
|
| 1720 |
asm ("mfpvr %0" |
| 1721 |
: "=r"(pvr));
|
| 1722 |
return pvr;
|
| 1723 |
} |
| 1724 |
|
| 1725 |
static void alter_insns(uint64_t *word, uint64_t flags, bool on) |
| 1726 |
{
|
| 1727 |
if (on) {
|
| 1728 |
*word |= flags; |
| 1729 |
} else {
|
| 1730 |
*word &= ~flags; |
| 1731 |
} |
| 1732 |
} |
| 1733 |
|
| 1734 |
static void kvmppc_host_cpu_initfn(Object *obj) |
| 1735 |
{
|
| 1736 |
assert(kvm_enabled()); |
| 1737 |
} |
| 1738 |
|
| 1739 |
static void kvmppc_host_cpu_class_init(ObjectClass *oc, void *data) |
| 1740 |
{
|
| 1741 |
PowerPCCPUClass *pcc = POWERPC_CPU_CLASS(oc); |
| 1742 |
uint32_t vmx = kvmppc_get_vmx(); |
| 1743 |
uint32_t dfp = kvmppc_get_dfp(); |
| 1744 |
uint32_t dcache_size = kvmppc_read_int_cpu_dt("d-cache-size");
|
| 1745 |
uint32_t icache_size = kvmppc_read_int_cpu_dt("i-cache-size");
|
| 1746 |
|
| 1747 |
/* Now fix up the class with information we can query from the host */
|
| 1748 |
|
| 1749 |
if (vmx != -1) { |
| 1750 |
/* Only override when we know what the host supports */
|
| 1751 |
alter_insns(&pcc->insns_flags, PPC_ALTIVEC, vmx > 0);
|
| 1752 |
alter_insns(&pcc->insns_flags2, PPC2_VSX, vmx > 1);
|
| 1753 |
} |
| 1754 |
if (dfp != -1) { |
| 1755 |
/* Only override when we know what the host supports */
|
| 1756 |
alter_insns(&pcc->insns_flags2, PPC2_DFP, dfp); |
| 1757 |
} |
| 1758 |
|
| 1759 |
if (dcache_size != -1) { |
| 1760 |
pcc->l1_dcache_size = dcache_size; |
| 1761 |
} |
| 1762 |
|
| 1763 |
if (icache_size != -1) { |
| 1764 |
pcc->l1_icache_size = icache_size; |
| 1765 |
} |
| 1766 |
} |
| 1767 |
|
| 1768 |
int kvmppc_fixup_cpu(PowerPCCPU *cpu)
|
| 1769 |
{
|
| 1770 |
CPUState *cs = CPU(cpu); |
| 1771 |
int smt;
|
| 1772 |
|
| 1773 |
/* Adjust cpu index for SMT */
|
| 1774 |
smt = kvmppc_smt_threads(); |
| 1775 |
cs->cpu_index = (cs->cpu_index / smp_threads) * smt |
| 1776 |
+ (cs->cpu_index % smp_threads); |
| 1777 |
|
| 1778 |
return 0; |
| 1779 |
} |
| 1780 |
|
| 1781 |
bool kvmppc_has_cap_epr(void) |
| 1782 |
{
|
| 1783 |
return cap_epr;
|
| 1784 |
} |
| 1785 |
|
| 1786 |
static int kvm_ppc_register_host_cpu_type(void) |
| 1787 |
{
|
| 1788 |
TypeInfo type_info = {
|
| 1789 |
.name = TYPE_HOST_POWERPC_CPU, |
| 1790 |
.instance_init = kvmppc_host_cpu_initfn, |
| 1791 |
.class_init = kvmppc_host_cpu_class_init, |
| 1792 |
}; |
| 1793 |
uint32_t host_pvr = mfpvr(); |
| 1794 |
PowerPCCPUClass *pvr_pcc; |
| 1795 |
|
| 1796 |
pvr_pcc = ppc_cpu_class_by_pvr(host_pvr); |
| 1797 |
if (pvr_pcc == NULL) { |
| 1798 |
return -1; |
| 1799 |
} |
| 1800 |
type_info.parent = object_class_get_name(OBJECT_CLASS(pvr_pcc)); |
| 1801 |
type_register(&type_info); |
| 1802 |
return 0; |
| 1803 |
} |
| 1804 |
|
| 1805 |
int kvmppc_define_rtas_kernel_token(uint32_t token, const char *function) |
| 1806 |
{
|
| 1807 |
struct kvm_rtas_token_args args = {
|
| 1808 |
.token = token, |
| 1809 |
}; |
| 1810 |
|
| 1811 |
if (!kvm_check_extension(kvm_state, KVM_CAP_PPC_RTAS)) {
|
| 1812 |
return -ENOENT;
|
| 1813 |
} |
| 1814 |
|
| 1815 |
strncpy(args.name, function, sizeof(args.name));
|
| 1816 |
|
| 1817 |
return kvm_vm_ioctl(kvm_state, KVM_PPC_RTAS_DEFINE_TOKEN, &args);
|
| 1818 |
} |
| 1819 |
|
| 1820 |
int kvmppc_get_htab_fd(bool write) |
| 1821 |
{
|
| 1822 |
struct kvm_get_htab_fd s = {
|
| 1823 |
.flags = write ? KVM_GET_HTAB_WRITE : 0,
|
| 1824 |
.start_index = 0,
|
| 1825 |
}; |
| 1826 |
|
| 1827 |
if (!cap_htab_fd) {
|
| 1828 |
fprintf(stderr, "KVM version doesn't support saving the hash table\n");
|
| 1829 |
return -1; |
| 1830 |
} |
| 1831 |
|
| 1832 |
return kvm_vm_ioctl(kvm_state, KVM_PPC_GET_HTAB_FD, &s);
|
| 1833 |
} |
| 1834 |
|
| 1835 |
int kvmppc_save_htab(QEMUFile *f, int fd, size_t bufsize, int64_t max_ns) |
| 1836 |
{
|
| 1837 |
int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); |
| 1838 |
uint8_t buf[bufsize]; |
| 1839 |
ssize_t rc; |
| 1840 |
|
| 1841 |
do {
|
| 1842 |
rc = read(fd, buf, bufsize); |
| 1843 |
if (rc < 0) { |
| 1844 |
fprintf(stderr, "Error reading data from KVM HTAB fd: %s\n",
|
| 1845 |
strerror(errno)); |
| 1846 |
return rc;
|
| 1847 |
} else if (rc) { |
| 1848 |
/* Kernel already retuns data in BE format for the file */
|
| 1849 |
qemu_put_buffer(f, buf, rc); |
| 1850 |
} |
| 1851 |
} while ((rc != 0) |
| 1852 |
&& ((max_ns < 0)
|
| 1853 |
|| ((qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) < max_ns))); |
| 1854 |
|
| 1855 |
return (rc == 0) ? 1 : 0; |
| 1856 |
} |
| 1857 |
|
| 1858 |
int kvmppc_load_htab_chunk(QEMUFile *f, int fd, uint32_t index, |
| 1859 |
uint16_t n_valid, uint16_t n_invalid) |
| 1860 |
{
|
| 1861 |
struct kvm_get_htab_header *buf;
|
| 1862 |
size_t chunksize = sizeof(*buf) + n_valid*HASH_PTE_SIZE_64;
|
| 1863 |
ssize_t rc; |
| 1864 |
|
| 1865 |
buf = alloca(chunksize); |
| 1866 |
/* This is KVM on ppc, so this is all big-endian */
|
| 1867 |
buf->index = index; |
| 1868 |
buf->n_valid = n_valid; |
| 1869 |
buf->n_invalid = n_invalid; |
| 1870 |
|
| 1871 |
qemu_get_buffer(f, (void *)(buf + 1), HASH_PTE_SIZE_64*n_valid); |
| 1872 |
|
| 1873 |
rc = write(fd, buf, chunksize); |
| 1874 |
if (rc < 0) { |
| 1875 |
fprintf(stderr, "Error writing KVM hash table: %s\n",
|
| 1876 |
strerror(errno)); |
| 1877 |
return rc;
|
| 1878 |
} |
| 1879 |
if (rc != chunksize) {
|
| 1880 |
/* We should never get a short write on a single chunk */
|
| 1881 |
fprintf(stderr, "Short write, restoring KVM hash table\n");
|
| 1882 |
return -1; |
| 1883 |
} |
| 1884 |
return 0; |
| 1885 |
} |
| 1886 |
|
| 1887 |
bool kvm_arch_stop_on_emulation_error(CPUState *cpu)
|
| 1888 |
{
|
| 1889 |
return true; |
| 1890 |
} |
| 1891 |
|
| 1892 |
int kvm_arch_on_sigbus_vcpu(CPUState *cpu, int code, void *addr) |
| 1893 |
{
|
| 1894 |
return 1; |
| 1895 |
} |
| 1896 |
|
| 1897 |
int kvm_arch_on_sigbus(int code, void *addr) |
| 1898 |
{
|
| 1899 |
return 1; |
| 1900 |
} |
| 1901 |
|
| 1902 |
void kvm_arch_init_irq_routing(KVMState *s)
|
| 1903 |
{
|
| 1904 |
} |