root / target-ppc / kvm.c @ 997aba8e
History | View | Annotate | Download (48.1 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 |
|
69 |
/* XXX We have a race condition where we actually have a level triggered
|
70 |
* interrupt, but the infrastructure can't expose that yet, so the guest
|
71 |
* takes but ignores it, goes to sleep and never gets notified that there's
|
72 |
* still an interrupt pending.
|
73 |
*
|
74 |
* As a quick workaround, let's just wake up again 20 ms after we injected
|
75 |
* an interrupt. That way we can assure that we're always reinjecting
|
76 |
* interrupts in case the guest swallowed them.
|
77 |
*/
|
78 |
static QEMUTimer *idle_timer;
|
79 |
|
80 |
static void kvm_kick_cpu(void *opaque) |
81 |
{ |
82 |
PowerPCCPU *cpu = opaque; |
83 |
|
84 |
qemu_cpu_kick(CPU(cpu)); |
85 |
} |
86 |
|
87 |
static int kvm_ppc_register_host_cpu_type(void); |
88 |
|
89 |
int kvm_arch_init(KVMState *s)
|
90 |
{ |
91 |
cap_interrupt_unset = kvm_check_extension(s, KVM_CAP_PPC_UNSET_IRQ); |
92 |
cap_interrupt_level = kvm_check_extension(s, KVM_CAP_PPC_IRQ_LEVEL); |
93 |
cap_segstate = kvm_check_extension(s, KVM_CAP_PPC_SEGSTATE); |
94 |
cap_booke_sregs = kvm_check_extension(s, KVM_CAP_PPC_BOOKE_SREGS); |
95 |
cap_ppc_smt = kvm_check_extension(s, KVM_CAP_PPC_SMT); |
96 |
cap_ppc_rma = kvm_check_extension(s, KVM_CAP_PPC_RMA); |
97 |
cap_spapr_tce = kvm_check_extension(s, KVM_CAP_SPAPR_TCE); |
98 |
cap_one_reg = kvm_check_extension(s, KVM_CAP_ONE_REG); |
99 |
cap_hior = kvm_check_extension(s, KVM_CAP_PPC_HIOR); |
100 |
cap_epr = kvm_check_extension(s, KVM_CAP_PPC_EPR); |
101 |
cap_ppc_watchdog = kvm_check_extension(s, KVM_CAP_PPC_BOOKE_WATCHDOG); |
102 |
/* Note: we don't set cap_papr here, because this capability is
|
103 |
* only activated after this by kvmppc_set_papr() */
|
104 |
|
105 |
if (!cap_interrupt_level) {
|
106 |
fprintf(stderr, "KVM: Couldn't find level irq capability. Expect the "
|
107 |
"VM to stall at times!\n");
|
108 |
} |
109 |
|
110 |
kvm_ppc_register_host_cpu_type(); |
111 |
|
112 |
return 0; |
113 |
} |
114 |
|
115 |
static int kvm_arch_sync_sregs(PowerPCCPU *cpu) |
116 |
{ |
117 |
CPUPPCState *cenv = &cpu->env; |
118 |
CPUState *cs = CPU(cpu); |
119 |
struct kvm_sregs sregs;
|
120 |
int ret;
|
121 |
|
122 |
if (cenv->excp_model == POWERPC_EXCP_BOOKE) {
|
123 |
/* What we're really trying to say is "if we're on BookE, we use
|
124 |
the native PVR for now". This is the only sane way to check
|
125 |
it though, so we potentially confuse users that they can run
|
126 |
BookE guests on BookS. Let's hope nobody dares enough :) */
|
127 |
return 0; |
128 |
} else {
|
129 |
if (!cap_segstate) {
|
130 |
fprintf(stderr, "kvm error: missing PVR setting capability\n");
|
131 |
return -ENOSYS;
|
132 |
} |
133 |
} |
134 |
|
135 |
ret = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs); |
136 |
if (ret) {
|
137 |
return ret;
|
138 |
} |
139 |
|
140 |
sregs.pvr = cenv->spr[SPR_PVR]; |
141 |
return kvm_vcpu_ioctl(cs, KVM_SET_SREGS, &sregs);
|
142 |
} |
143 |
|
144 |
/* Set up a shared TLB array with KVM */
|
145 |
static int kvm_booke206_tlb_init(PowerPCCPU *cpu) |
146 |
{ |
147 |
CPUPPCState *env = &cpu->env; |
148 |
CPUState *cs = CPU(cpu); |
149 |
struct kvm_book3e_206_tlb_params params = {};
|
150 |
struct kvm_config_tlb cfg = {};
|
151 |
struct kvm_enable_cap encap = {};
|
152 |
unsigned int entries = 0; |
153 |
int ret, i;
|
154 |
|
155 |
if (!kvm_enabled() ||
|
156 |
!kvm_check_extension(cs->kvm_state, KVM_CAP_SW_TLB)) { |
157 |
return 0; |
158 |
} |
159 |
|
160 |
assert(ARRAY_SIZE(params.tlb_sizes) == BOOKE206_MAX_TLBN); |
161 |
|
162 |
for (i = 0; i < BOOKE206_MAX_TLBN; i++) { |
163 |
params.tlb_sizes[i] = booke206_tlb_size(env, i); |
164 |
params.tlb_ways[i] = booke206_tlb_ways(env, i); |
165 |
entries += params.tlb_sizes[i]; |
166 |
} |
167 |
|
168 |
assert(entries == env->nb_tlb); |
169 |
assert(sizeof(struct kvm_book3e_206_tlb_entry) == sizeof(ppcmas_tlb_t)); |
170 |
|
171 |
env->tlb_dirty = true;
|
172 |
|
173 |
cfg.array = (uintptr_t)env->tlb.tlbm; |
174 |
cfg.array_len = sizeof(ppcmas_tlb_t) * entries;
|
175 |
cfg.params = (uintptr_t)¶ms; |
176 |
cfg.mmu_type = KVM_MMU_FSL_BOOKE_NOHV; |
177 |
|
178 |
encap.cap = KVM_CAP_SW_TLB; |
179 |
encap.args[0] = (uintptr_t)&cfg;
|
180 |
|
181 |
ret = kvm_vcpu_ioctl(cs, KVM_ENABLE_CAP, &encap); |
182 |
if (ret < 0) { |
183 |
fprintf(stderr, "%s: couldn't enable KVM_CAP_SW_TLB: %s\n",
|
184 |
__func__, strerror(-ret)); |
185 |
return ret;
|
186 |
} |
187 |
|
188 |
env->kvm_sw_tlb = true;
|
189 |
return 0; |
190 |
} |
191 |
|
192 |
|
193 |
#if defined(TARGET_PPC64)
|
194 |
static void kvm_get_fallback_smmu_info(PowerPCCPU *cpu, |
195 |
struct kvm_ppc_smmu_info *info)
|
196 |
{ |
197 |
CPUPPCState *env = &cpu->env; |
198 |
CPUState *cs = CPU(cpu); |
199 |
|
200 |
memset(info, 0, sizeof(*info)); |
201 |
|
202 |
/* We don't have the new KVM_PPC_GET_SMMU_INFO ioctl, so
|
203 |
* need to "guess" what the supported page sizes are.
|
204 |
*
|
205 |
* For that to work we make a few assumptions:
|
206 |
*
|
207 |
* - If KVM_CAP_PPC_GET_PVINFO is supported we are running "PR"
|
208 |
* KVM which only supports 4K and 16M pages, but supports them
|
209 |
* regardless of the backing store characteritics. We also don't
|
210 |
* support 1T segments.
|
211 |
*
|
212 |
* This is safe as if HV KVM ever supports that capability or PR
|
213 |
* KVM grows supports for more page/segment sizes, those versions
|
214 |
* will have implemented KVM_CAP_PPC_GET_SMMU_INFO and thus we
|
215 |
* will not hit this fallback
|
216 |
*
|
217 |
* - Else we are running HV KVM. This means we only support page
|
218 |
* sizes that fit in the backing store. Additionally we only
|
219 |
* advertize 64K pages if the processor is ARCH 2.06 and we assume
|
220 |
* P7 encodings for the SLB and hash table. Here too, we assume
|
221 |
* support for any newer processor will mean a kernel that
|
222 |
* implements KVM_CAP_PPC_GET_SMMU_INFO and thus doesn't hit
|
223 |
* this fallback.
|
224 |
*/
|
225 |
if (kvm_check_extension(cs->kvm_state, KVM_CAP_PPC_GET_PVINFO)) {
|
226 |
/* No flags */
|
227 |
info->flags = 0;
|
228 |
info->slb_size = 64;
|
229 |
|
230 |
/* Standard 4k base page size segment */
|
231 |
info->sps[0].page_shift = 12; |
232 |
info->sps[0].slb_enc = 0; |
233 |
info->sps[0].enc[0].page_shift = 12; |
234 |
info->sps[0].enc[0].pte_enc = 0; |
235 |
|
236 |
/* Standard 16M large page size segment */
|
237 |
info->sps[1].page_shift = 24; |
238 |
info->sps[1].slb_enc = SLB_VSID_L;
|
239 |
info->sps[1].enc[0].page_shift = 24; |
240 |
info->sps[1].enc[0].pte_enc = 0; |
241 |
} else {
|
242 |
int i = 0; |
243 |
|
244 |
/* HV KVM has backing store size restrictions */
|
245 |
info->flags = KVM_PPC_PAGE_SIZES_REAL; |
246 |
|
247 |
if (env->mmu_model & POWERPC_MMU_1TSEG) {
|
248 |
info->flags |= KVM_PPC_1T_SEGMENTS; |
249 |
} |
250 |
|
251 |
if (env->mmu_model == POWERPC_MMU_2_06) {
|
252 |
info->slb_size = 32;
|
253 |
} else {
|
254 |
info->slb_size = 64;
|
255 |
} |
256 |
|
257 |
/* Standard 4k base page size segment */
|
258 |
info->sps[i].page_shift = 12;
|
259 |
info->sps[i].slb_enc = 0;
|
260 |
info->sps[i].enc[0].page_shift = 12; |
261 |
info->sps[i].enc[0].pte_enc = 0; |
262 |
i++; |
263 |
|
264 |
/* 64K on MMU 2.06 */
|
265 |
if (env->mmu_model == POWERPC_MMU_2_06) {
|
266 |
info->sps[i].page_shift = 16;
|
267 |
info->sps[i].slb_enc = 0x110;
|
268 |
info->sps[i].enc[0].page_shift = 16; |
269 |
info->sps[i].enc[0].pte_enc = 1; |
270 |
i++; |
271 |
} |
272 |
|
273 |
/* Standard 16M large page size segment */
|
274 |
info->sps[i].page_shift = 24;
|
275 |
info->sps[i].slb_enc = SLB_VSID_L; |
276 |
info->sps[i].enc[0].page_shift = 24; |
277 |
info->sps[i].enc[0].pte_enc = 0; |
278 |
} |
279 |
} |
280 |
|
281 |
static void kvm_get_smmu_info(PowerPCCPU *cpu, struct kvm_ppc_smmu_info *info) |
282 |
{ |
283 |
CPUState *cs = CPU(cpu); |
284 |
int ret;
|
285 |
|
286 |
if (kvm_check_extension(cs->kvm_state, KVM_CAP_PPC_GET_SMMU_INFO)) {
|
287 |
ret = kvm_vm_ioctl(cs->kvm_state, KVM_PPC_GET_SMMU_INFO, info); |
288 |
if (ret == 0) { |
289 |
return;
|
290 |
} |
291 |
} |
292 |
|
293 |
kvm_get_fallback_smmu_info(cpu, info); |
294 |
} |
295 |
|
296 |
static long getrampagesize(void) |
297 |
{ |
298 |
struct statfs fs;
|
299 |
int ret;
|
300 |
|
301 |
if (!mem_path) {
|
302 |
/* guest RAM is backed by normal anonymous pages */
|
303 |
return getpagesize();
|
304 |
} |
305 |
|
306 |
do {
|
307 |
ret = statfs(mem_path, &fs); |
308 |
} while (ret != 0 && errno == EINTR); |
309 |
|
310 |
if (ret != 0) { |
311 |
fprintf(stderr, "Couldn't statfs() memory path: %s\n",
|
312 |
strerror(errno)); |
313 |
exit(1);
|
314 |
} |
315 |
|
316 |
#define HUGETLBFS_MAGIC 0x958458f6 |
317 |
|
318 |
if (fs.f_type != HUGETLBFS_MAGIC) {
|
319 |
/* Explicit mempath, but it's ordinary pages */
|
320 |
return getpagesize();
|
321 |
} |
322 |
|
323 |
/* It's hugepage, return the huge page size */
|
324 |
return fs.f_bsize;
|
325 |
} |
326 |
|
327 |
static bool kvm_valid_page_size(uint32_t flags, long rampgsize, uint32_t shift) |
328 |
{ |
329 |
if (!(flags & KVM_PPC_PAGE_SIZES_REAL)) {
|
330 |
return true; |
331 |
} |
332 |
|
333 |
return (1ul << shift) <= rampgsize; |
334 |
} |
335 |
|
336 |
static void kvm_fixup_page_sizes(PowerPCCPU *cpu) |
337 |
{ |
338 |
static struct kvm_ppc_smmu_info smmu_info; |
339 |
static bool has_smmu_info; |
340 |
CPUPPCState *env = &cpu->env; |
341 |
long rampagesize;
|
342 |
int iq, ik, jq, jk;
|
343 |
|
344 |
/* We only handle page sizes for 64-bit server guests for now */
|
345 |
if (!(env->mmu_model & POWERPC_MMU_64)) {
|
346 |
return;
|
347 |
} |
348 |
|
349 |
/* Collect MMU info from kernel if not already */
|
350 |
if (!has_smmu_info) {
|
351 |
kvm_get_smmu_info(cpu, &smmu_info); |
352 |
has_smmu_info = true;
|
353 |
} |
354 |
|
355 |
rampagesize = getrampagesize(); |
356 |
|
357 |
/* Convert to QEMU form */
|
358 |
memset(&env->sps, 0, sizeof(env->sps)); |
359 |
|
360 |
for (ik = iq = 0; ik < KVM_PPC_PAGE_SIZES_MAX_SZ; ik++) { |
361 |
struct ppc_one_seg_page_size *qsps = &env->sps.sps[iq];
|
362 |
struct kvm_ppc_one_seg_page_size *ksps = &smmu_info.sps[ik];
|
363 |
|
364 |
if (!kvm_valid_page_size(smmu_info.flags, rampagesize,
|
365 |
ksps->page_shift)) { |
366 |
continue;
|
367 |
} |
368 |
qsps->page_shift = ksps->page_shift; |
369 |
qsps->slb_enc = ksps->slb_enc; |
370 |
for (jk = jq = 0; jk < KVM_PPC_PAGE_SIZES_MAX_SZ; jk++) { |
371 |
if (!kvm_valid_page_size(smmu_info.flags, rampagesize,
|
372 |
ksps->enc[jk].page_shift)) { |
373 |
continue;
|
374 |
} |
375 |
qsps->enc[jq].page_shift = ksps->enc[jk].page_shift; |
376 |
qsps->enc[jq].pte_enc = ksps->enc[jk].pte_enc; |
377 |
if (++jq >= PPC_PAGE_SIZES_MAX_SZ) {
|
378 |
break;
|
379 |
} |
380 |
} |
381 |
if (++iq >= PPC_PAGE_SIZES_MAX_SZ) {
|
382 |
break;
|
383 |
} |
384 |
} |
385 |
env->slb_nr = smmu_info.slb_size; |
386 |
if (smmu_info.flags & KVM_PPC_1T_SEGMENTS) {
|
387 |
env->mmu_model |= POWERPC_MMU_1TSEG; |
388 |
} else {
|
389 |
env->mmu_model &= ~POWERPC_MMU_1TSEG; |
390 |
} |
391 |
} |
392 |
#else /* defined (TARGET_PPC64) */ |
393 |
|
394 |
static inline void kvm_fixup_page_sizes(PowerPCCPU *cpu) |
395 |
{ |
396 |
} |
397 |
|
398 |
#endif /* !defined (TARGET_PPC64) */ |
399 |
|
400 |
unsigned long kvm_arch_vcpu_id(CPUState *cpu) |
401 |
{ |
402 |
return cpu->cpu_index;
|
403 |
} |
404 |
|
405 |
int kvm_arch_init_vcpu(CPUState *cs)
|
406 |
{ |
407 |
PowerPCCPU *cpu = POWERPC_CPU(cs); |
408 |
CPUPPCState *cenv = &cpu->env; |
409 |
int ret;
|
410 |
|
411 |
/* Gather server mmu info from KVM and update the CPU state */
|
412 |
kvm_fixup_page_sizes(cpu); |
413 |
|
414 |
/* Synchronize sregs with kvm */
|
415 |
ret = kvm_arch_sync_sregs(cpu); |
416 |
if (ret) {
|
417 |
return ret;
|
418 |
} |
419 |
|
420 |
idle_timer = qemu_new_timer_ns(vm_clock, kvm_kick_cpu, cpu); |
421 |
|
422 |
/* Some targets support access to KVM's guest TLB. */
|
423 |
switch (cenv->mmu_model) {
|
424 |
case POWERPC_MMU_BOOKE206:
|
425 |
ret = kvm_booke206_tlb_init(cpu); |
426 |
break;
|
427 |
default:
|
428 |
break;
|
429 |
} |
430 |
|
431 |
return ret;
|
432 |
} |
433 |
|
434 |
void kvm_arch_reset_vcpu(CPUState *cpu)
|
435 |
{ |
436 |
} |
437 |
|
438 |
static void kvm_sw_tlb_put(PowerPCCPU *cpu) |
439 |
{ |
440 |
CPUPPCState *env = &cpu->env; |
441 |
CPUState *cs = CPU(cpu); |
442 |
struct kvm_dirty_tlb dirty_tlb;
|
443 |
unsigned char *bitmap; |
444 |
int ret;
|
445 |
|
446 |
if (!env->kvm_sw_tlb) {
|
447 |
return;
|
448 |
} |
449 |
|
450 |
bitmap = g_malloc((env->nb_tlb + 7) / 8); |
451 |
memset(bitmap, 0xFF, (env->nb_tlb + 7) / 8); |
452 |
|
453 |
dirty_tlb.bitmap = (uintptr_t)bitmap; |
454 |
dirty_tlb.num_dirty = env->nb_tlb; |
455 |
|
456 |
ret = kvm_vcpu_ioctl(cs, KVM_DIRTY_TLB, &dirty_tlb); |
457 |
if (ret) {
|
458 |
fprintf(stderr, "%s: KVM_DIRTY_TLB: %s\n",
|
459 |
__func__, strerror(-ret)); |
460 |
} |
461 |
|
462 |
g_free(bitmap); |
463 |
} |
464 |
|
465 |
static void kvm_get_one_spr(CPUState *cs, uint64_t id, int spr) |
466 |
{ |
467 |
PowerPCCPU *cpu = POWERPC_CPU(cs); |
468 |
CPUPPCState *env = &cpu->env; |
469 |
union {
|
470 |
uint32_t u32; |
471 |
uint64_t u64; |
472 |
} val; |
473 |
struct kvm_one_reg reg = {
|
474 |
.id = id, |
475 |
.addr = (uintptr_t) &val, |
476 |
}; |
477 |
int ret;
|
478 |
|
479 |
ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®); |
480 |
if (ret != 0) { |
481 |
fprintf(stderr, "Warning: Unable to retrieve SPR %d from KVM: %s\n",
|
482 |
spr, strerror(errno)); |
483 |
} else {
|
484 |
switch (id & KVM_REG_SIZE_MASK) {
|
485 |
case KVM_REG_SIZE_U32:
|
486 |
env->spr[spr] = val.u32; |
487 |
break;
|
488 |
|
489 |
case KVM_REG_SIZE_U64:
|
490 |
env->spr[spr] = val.u64; |
491 |
break;
|
492 |
|
493 |
default:
|
494 |
/* Don't handle this size yet */
|
495 |
abort(); |
496 |
} |
497 |
} |
498 |
} |
499 |
|
500 |
static void kvm_put_one_spr(CPUState *cs, uint64_t id, int spr) |
501 |
{ |
502 |
PowerPCCPU *cpu = POWERPC_CPU(cs); |
503 |
CPUPPCState *env = &cpu->env; |
504 |
union {
|
505 |
uint32_t u32; |
506 |
uint64_t u64; |
507 |
} val; |
508 |
struct kvm_one_reg reg = {
|
509 |
.id = id, |
510 |
.addr = (uintptr_t) &val, |
511 |
}; |
512 |
int ret;
|
513 |
|
514 |
switch (id & KVM_REG_SIZE_MASK) {
|
515 |
case KVM_REG_SIZE_U32:
|
516 |
val.u32 = env->spr[spr]; |
517 |
break;
|
518 |
|
519 |
case KVM_REG_SIZE_U64:
|
520 |
val.u64 = env->spr[spr]; |
521 |
break;
|
522 |
|
523 |
default:
|
524 |
/* Don't handle this size yet */
|
525 |
abort(); |
526 |
} |
527 |
|
528 |
ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®); |
529 |
if (ret != 0) { |
530 |
fprintf(stderr, "Warning: Unable to set SPR %d to KVM: %s\n",
|
531 |
spr, strerror(errno)); |
532 |
} |
533 |
} |
534 |
|
535 |
static int kvm_put_fp(CPUState *cs) |
536 |
{ |
537 |
PowerPCCPU *cpu = POWERPC_CPU(cs); |
538 |
CPUPPCState *env = &cpu->env; |
539 |
struct kvm_one_reg reg;
|
540 |
int i;
|
541 |
int ret;
|
542 |
|
543 |
if (env->insns_flags & PPC_FLOAT) {
|
544 |
uint64_t fpscr = env->fpscr; |
545 |
bool vsx = !!(env->insns_flags2 & PPC2_VSX);
|
546 |
|
547 |
reg.id = KVM_REG_PPC_FPSCR; |
548 |
reg.addr = (uintptr_t)&fpscr; |
549 |
ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®); |
550 |
if (ret < 0) { |
551 |
dprintf("Unable to set FPSCR to KVM: %s\n", strerror(errno));
|
552 |
return ret;
|
553 |
} |
554 |
|
555 |
for (i = 0; i < 32; i++) { |
556 |
uint64_t vsr[2];
|
557 |
|
558 |
vsr[0] = float64_val(env->fpr[i]);
|
559 |
vsr[1] = env->vsr[i];
|
560 |
reg.addr = (uintptr_t) &vsr; |
561 |
reg.id = vsx ? KVM_REG_PPC_VSR(i) : KVM_REG_PPC_FPR(i); |
562 |
|
563 |
ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®); |
564 |
if (ret < 0) { |
565 |
dprintf("Unable to set %s%d to KVM: %s\n", vsx ? "VSR" : "FPR", |
566 |
i, strerror(errno)); |
567 |
return ret;
|
568 |
} |
569 |
} |
570 |
} |
571 |
|
572 |
if (env->insns_flags & PPC_ALTIVEC) {
|
573 |
reg.id = KVM_REG_PPC_VSCR; |
574 |
reg.addr = (uintptr_t)&env->vscr; |
575 |
ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®); |
576 |
if (ret < 0) { |
577 |
dprintf("Unable to set VSCR to KVM: %s\n", strerror(errno));
|
578 |
return ret;
|
579 |
} |
580 |
|
581 |
for (i = 0; i < 32; i++) { |
582 |
reg.id = KVM_REG_PPC_VR(i); |
583 |
reg.addr = (uintptr_t)&env->avr[i]; |
584 |
ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®); |
585 |
if (ret < 0) { |
586 |
dprintf("Unable to set VR%d to KVM: %s\n", i, strerror(errno));
|
587 |
return ret;
|
588 |
} |
589 |
} |
590 |
} |
591 |
|
592 |
return 0; |
593 |
} |
594 |
|
595 |
static int kvm_get_fp(CPUState *cs) |
596 |
{ |
597 |
PowerPCCPU *cpu = POWERPC_CPU(cs); |
598 |
CPUPPCState *env = &cpu->env; |
599 |
struct kvm_one_reg reg;
|
600 |
int i;
|
601 |
int ret;
|
602 |
|
603 |
if (env->insns_flags & PPC_FLOAT) {
|
604 |
uint64_t fpscr; |
605 |
bool vsx = !!(env->insns_flags2 & PPC2_VSX);
|
606 |
|
607 |
reg.id = KVM_REG_PPC_FPSCR; |
608 |
reg.addr = (uintptr_t)&fpscr; |
609 |
ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®); |
610 |
if (ret < 0) { |
611 |
dprintf("Unable to get FPSCR from KVM: %s\n", strerror(errno));
|
612 |
return ret;
|
613 |
} else {
|
614 |
env->fpscr = fpscr; |
615 |
} |
616 |
|
617 |
for (i = 0; i < 32; i++) { |
618 |
uint64_t vsr[2];
|
619 |
|
620 |
reg.addr = (uintptr_t) &vsr; |
621 |
reg.id = vsx ? KVM_REG_PPC_VSR(i) : KVM_REG_PPC_FPR(i); |
622 |
|
623 |
ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®); |
624 |
if (ret < 0) { |
625 |
dprintf("Unable to get %s%d from KVM: %s\n",
|
626 |
vsx ? "VSR" : "FPR", i, strerror(errno)); |
627 |
return ret;
|
628 |
} else {
|
629 |
env->fpr[i] = vsr[0];
|
630 |
if (vsx) {
|
631 |
env->vsr[i] = vsr[1];
|
632 |
} |
633 |
} |
634 |
} |
635 |
} |
636 |
|
637 |
if (env->insns_flags & PPC_ALTIVEC) {
|
638 |
reg.id = KVM_REG_PPC_VSCR; |
639 |
reg.addr = (uintptr_t)&env->vscr; |
640 |
ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®); |
641 |
if (ret < 0) { |
642 |
dprintf("Unable to get VSCR from KVM: %s\n", strerror(errno));
|
643 |
return ret;
|
644 |
} |
645 |
|
646 |
for (i = 0; i < 32; i++) { |
647 |
reg.id = KVM_REG_PPC_VR(i); |
648 |
reg.addr = (uintptr_t)&env->avr[i]; |
649 |
ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®); |
650 |
if (ret < 0) { |
651 |
dprintf("Unable to get VR%d from KVM: %s\n",
|
652 |
i, strerror(errno)); |
653 |
return ret;
|
654 |
} |
655 |
} |
656 |
} |
657 |
|
658 |
return 0; |
659 |
} |
660 |
|
661 |
#if defined(TARGET_PPC64)
|
662 |
static int kvm_get_vpa(CPUState *cs) |
663 |
{ |
664 |
PowerPCCPU *cpu = POWERPC_CPU(cs); |
665 |
CPUPPCState *env = &cpu->env; |
666 |
struct kvm_one_reg reg;
|
667 |
int ret;
|
668 |
|
669 |
reg.id = KVM_REG_PPC_VPA_ADDR; |
670 |
reg.addr = (uintptr_t)&env->vpa_addr; |
671 |
ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®); |
672 |
if (ret < 0) { |
673 |
dprintf("Unable to get VPA address from KVM: %s\n", strerror(errno));
|
674 |
return ret;
|
675 |
} |
676 |
|
677 |
assert((uintptr_t)&env->slb_shadow_size |
678 |
== ((uintptr_t)&env->slb_shadow_addr + 8));
|
679 |
reg.id = KVM_REG_PPC_VPA_SLB; |
680 |
reg.addr = (uintptr_t)&env->slb_shadow_addr; |
681 |
ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®); |
682 |
if (ret < 0) { |
683 |
dprintf("Unable to get SLB shadow state from KVM: %s\n",
|
684 |
strerror(errno)); |
685 |
return ret;
|
686 |
} |
687 |
|
688 |
assert((uintptr_t)&env->dtl_size == ((uintptr_t)&env->dtl_addr + 8));
|
689 |
reg.id = KVM_REG_PPC_VPA_DTL; |
690 |
reg.addr = (uintptr_t)&env->dtl_addr; |
691 |
ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®); |
692 |
if (ret < 0) { |
693 |
dprintf("Unable to get dispatch trace log state from KVM: %s\n",
|
694 |
strerror(errno)); |
695 |
return ret;
|
696 |
} |
697 |
|
698 |
return 0; |
699 |
} |
700 |
|
701 |
static int kvm_put_vpa(CPUState *cs) |
702 |
{ |
703 |
PowerPCCPU *cpu = POWERPC_CPU(cs); |
704 |
CPUPPCState *env = &cpu->env; |
705 |
struct kvm_one_reg reg;
|
706 |
int ret;
|
707 |
|
708 |
/* SLB shadow or DTL can't be registered unless a master VPA is
|
709 |
* registered. That means when restoring state, if a VPA *is*
|
710 |
* registered, we need to set that up first. If not, we need to
|
711 |
* deregister the others before deregistering the master VPA */
|
712 |
assert(env->vpa_addr || !(env->slb_shadow_addr || env->dtl_addr)); |
713 |
|
714 |
if (env->vpa_addr) {
|
715 |
reg.id = KVM_REG_PPC_VPA_ADDR; |
716 |
reg.addr = (uintptr_t)&env->vpa_addr; |
717 |
ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®); |
718 |
if (ret < 0) { |
719 |
dprintf("Unable to set VPA address to KVM: %s\n", strerror(errno));
|
720 |
return ret;
|
721 |
} |
722 |
} |
723 |
|
724 |
assert((uintptr_t)&env->slb_shadow_size |
725 |
== ((uintptr_t)&env->slb_shadow_addr + 8));
|
726 |
reg.id = KVM_REG_PPC_VPA_SLB; |
727 |
reg.addr = (uintptr_t)&env->slb_shadow_addr; |
728 |
ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®); |
729 |
if (ret < 0) { |
730 |
dprintf("Unable to set SLB shadow state to KVM: %s\n", strerror(errno));
|
731 |
return ret;
|
732 |
} |
733 |
|
734 |
assert((uintptr_t)&env->dtl_size == ((uintptr_t)&env->dtl_addr + 8));
|
735 |
reg.id = KVM_REG_PPC_VPA_DTL; |
736 |
reg.addr = (uintptr_t)&env->dtl_addr; |
737 |
ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®); |
738 |
if (ret < 0) { |
739 |
dprintf("Unable to set dispatch trace log state to KVM: %s\n",
|
740 |
strerror(errno)); |
741 |
return ret;
|
742 |
} |
743 |
|
744 |
if (!env->vpa_addr) {
|
745 |
reg.id = KVM_REG_PPC_VPA_ADDR; |
746 |
reg.addr = (uintptr_t)&env->vpa_addr; |
747 |
ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®); |
748 |
if (ret < 0) { |
749 |
dprintf("Unable to set VPA address to KVM: %s\n", strerror(errno));
|
750 |
return ret;
|
751 |
} |
752 |
} |
753 |
|
754 |
return 0; |
755 |
} |
756 |
#endif /* TARGET_PPC64 */ |
757 |
|
758 |
int kvm_arch_put_registers(CPUState *cs, int level) |
759 |
{ |
760 |
PowerPCCPU *cpu = POWERPC_CPU(cs); |
761 |
CPUPPCState *env = &cpu->env; |
762 |
struct kvm_regs regs;
|
763 |
int ret;
|
764 |
int i;
|
765 |
|
766 |
ret = kvm_vcpu_ioctl(cs, KVM_GET_REGS, ®s); |
767 |
if (ret < 0) { |
768 |
return ret;
|
769 |
} |
770 |
|
771 |
regs.ctr = env->ctr; |
772 |
regs.lr = env->lr; |
773 |
regs.xer = cpu_read_xer(env); |
774 |
regs.msr = env->msr; |
775 |
regs.pc = env->nip; |
776 |
|
777 |
regs.srr0 = env->spr[SPR_SRR0]; |
778 |
regs.srr1 = env->spr[SPR_SRR1]; |
779 |
|
780 |
regs.sprg0 = env->spr[SPR_SPRG0]; |
781 |
regs.sprg1 = env->spr[SPR_SPRG1]; |
782 |
regs.sprg2 = env->spr[SPR_SPRG2]; |
783 |
regs.sprg3 = env->spr[SPR_SPRG3]; |
784 |
regs.sprg4 = env->spr[SPR_SPRG4]; |
785 |
regs.sprg5 = env->spr[SPR_SPRG5]; |
786 |
regs.sprg6 = env->spr[SPR_SPRG6]; |
787 |
regs.sprg7 = env->spr[SPR_SPRG7]; |
788 |
|
789 |
regs.pid = env->spr[SPR_BOOKE_PID]; |
790 |
|
791 |
for (i = 0;i < 32; i++) |
792 |
regs.gpr[i] = env->gpr[i]; |
793 |
|
794 |
ret = kvm_vcpu_ioctl(cs, KVM_SET_REGS, ®s); |
795 |
if (ret < 0) |
796 |
return ret;
|
797 |
|
798 |
kvm_put_fp(cs); |
799 |
|
800 |
if (env->tlb_dirty) {
|
801 |
kvm_sw_tlb_put(cpu); |
802 |
env->tlb_dirty = false;
|
803 |
} |
804 |
|
805 |
if (cap_segstate && (level >= KVM_PUT_RESET_STATE)) {
|
806 |
struct kvm_sregs sregs;
|
807 |
|
808 |
sregs.pvr = env->spr[SPR_PVR]; |
809 |
|
810 |
sregs.u.s.sdr1 = env->spr[SPR_SDR1]; |
811 |
|
812 |
/* Sync SLB */
|
813 |
#ifdef TARGET_PPC64
|
814 |
for (i = 0; i < 64; i++) { |
815 |
sregs.u.s.ppc64.slb[i].slbe = env->slb[i].esid; |
816 |
sregs.u.s.ppc64.slb[i].slbv = env->slb[i].vsid; |
817 |
} |
818 |
#endif
|
819 |
|
820 |
/* Sync SRs */
|
821 |
for (i = 0; i < 16; i++) { |
822 |
sregs.u.s.ppc32.sr[i] = env->sr[i]; |
823 |
} |
824 |
|
825 |
/* Sync BATs */
|
826 |
for (i = 0; i < 8; i++) { |
827 |
/* Beware. We have to swap upper and lower bits here */
|
828 |
sregs.u.s.ppc32.dbat[i] = ((uint64_t)env->DBAT[0][i] << 32) |
829 |
| env->DBAT[1][i];
|
830 |
sregs.u.s.ppc32.ibat[i] = ((uint64_t)env->IBAT[0][i] << 32) |
831 |
| env->IBAT[1][i];
|
832 |
} |
833 |
|
834 |
ret = kvm_vcpu_ioctl(cs, KVM_SET_SREGS, &sregs); |
835 |
if (ret) {
|
836 |
return ret;
|
837 |
} |
838 |
} |
839 |
|
840 |
if (cap_hior && (level >= KVM_PUT_RESET_STATE)) {
|
841 |
kvm_put_one_spr(cs, KVM_REG_PPC_HIOR, SPR_HIOR); |
842 |
} |
843 |
|
844 |
if (cap_one_reg) {
|
845 |
int i;
|
846 |
|
847 |
/* We deliberately ignore errors here, for kernels which have
|
848 |
* the ONE_REG calls, but don't support the specific
|
849 |
* registers, there's a reasonable chance things will still
|
850 |
* work, at least until we try to migrate. */
|
851 |
for (i = 0; i < 1024; i++) { |
852 |
uint64_t id = env->spr_cb[i].one_reg_id; |
853 |
|
854 |
if (id != 0) { |
855 |
kvm_put_one_spr(cs, id, i); |
856 |
} |
857 |
} |
858 |
|
859 |
#ifdef TARGET_PPC64
|
860 |
if (cap_papr) {
|
861 |
if (kvm_put_vpa(cs) < 0) { |
862 |
dprintf("Warning: Unable to set VPA information to KVM\n");
|
863 |
} |
864 |
} |
865 |
#endif /* TARGET_PPC64 */ |
866 |
} |
867 |
|
868 |
return ret;
|
869 |
} |
870 |
|
871 |
int kvm_arch_get_registers(CPUState *cs)
|
872 |
{ |
873 |
PowerPCCPU *cpu = POWERPC_CPU(cs); |
874 |
CPUPPCState *env = &cpu->env; |
875 |
struct kvm_regs regs;
|
876 |
struct kvm_sregs sregs;
|
877 |
uint32_t cr; |
878 |
int i, ret;
|
879 |
|
880 |
ret = kvm_vcpu_ioctl(cs, KVM_GET_REGS, ®s); |
881 |
if (ret < 0) |
882 |
return ret;
|
883 |
|
884 |
cr = regs.cr; |
885 |
for (i = 7; i >= 0; i--) { |
886 |
env->crf[i] = cr & 15;
|
887 |
cr >>= 4;
|
888 |
} |
889 |
|
890 |
env->ctr = regs.ctr; |
891 |
env->lr = regs.lr; |
892 |
cpu_write_xer(env, regs.xer); |
893 |
env->msr = regs.msr; |
894 |
env->nip = regs.pc; |
895 |
|
896 |
env->spr[SPR_SRR0] = regs.srr0; |
897 |
env->spr[SPR_SRR1] = regs.srr1; |
898 |
|
899 |
env->spr[SPR_SPRG0] = regs.sprg0; |
900 |
env->spr[SPR_SPRG1] = regs.sprg1; |
901 |
env->spr[SPR_SPRG2] = regs.sprg2; |
902 |
env->spr[SPR_SPRG3] = regs.sprg3; |
903 |
env->spr[SPR_SPRG4] = regs.sprg4; |
904 |
env->spr[SPR_SPRG5] = regs.sprg5; |
905 |
env->spr[SPR_SPRG6] = regs.sprg6; |
906 |
env->spr[SPR_SPRG7] = regs.sprg7; |
907 |
|
908 |
env->spr[SPR_BOOKE_PID] = regs.pid; |
909 |
|
910 |
for (i = 0;i < 32; i++) |
911 |
env->gpr[i] = regs.gpr[i]; |
912 |
|
913 |
kvm_get_fp(cs); |
914 |
|
915 |
if (cap_booke_sregs) {
|
916 |
ret = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs); |
917 |
if (ret < 0) { |
918 |
return ret;
|
919 |
} |
920 |
|
921 |
if (sregs.u.e.features & KVM_SREGS_E_BASE) {
|
922 |
env->spr[SPR_BOOKE_CSRR0] = sregs.u.e.csrr0; |
923 |
env->spr[SPR_BOOKE_CSRR1] = sregs.u.e.csrr1; |
924 |
env->spr[SPR_BOOKE_ESR] = sregs.u.e.esr; |
925 |
env->spr[SPR_BOOKE_DEAR] = sregs.u.e.dear; |
926 |
env->spr[SPR_BOOKE_MCSR] = sregs.u.e.mcsr; |
927 |
env->spr[SPR_BOOKE_TSR] = sregs.u.e.tsr; |
928 |
env->spr[SPR_BOOKE_TCR] = sregs.u.e.tcr; |
929 |
env->spr[SPR_DECR] = sregs.u.e.dec; |
930 |
env->spr[SPR_TBL] = sregs.u.e.tb & 0xffffffff;
|
931 |
env->spr[SPR_TBU] = sregs.u.e.tb >> 32;
|
932 |
env->spr[SPR_VRSAVE] = sregs.u.e.vrsave; |
933 |
} |
934 |
|
935 |
if (sregs.u.e.features & KVM_SREGS_E_ARCH206) {
|
936 |
env->spr[SPR_BOOKE_PIR] = sregs.u.e.pir; |
937 |
env->spr[SPR_BOOKE_MCSRR0] = sregs.u.e.mcsrr0; |
938 |
env->spr[SPR_BOOKE_MCSRR1] = sregs.u.e.mcsrr1; |
939 |
env->spr[SPR_BOOKE_DECAR] = sregs.u.e.decar; |
940 |
env->spr[SPR_BOOKE_IVPR] = sregs.u.e.ivpr; |
941 |
} |
942 |
|
943 |
if (sregs.u.e.features & KVM_SREGS_E_64) {
|
944 |
env->spr[SPR_BOOKE_EPCR] = sregs.u.e.epcr; |
945 |
} |
946 |
|
947 |
if (sregs.u.e.features & KVM_SREGS_E_SPRG8) {
|
948 |
env->spr[SPR_BOOKE_SPRG8] = sregs.u.e.sprg8; |
949 |
} |
950 |
|
951 |
if (sregs.u.e.features & KVM_SREGS_E_IVOR) {
|
952 |
env->spr[SPR_BOOKE_IVOR0] = sregs.u.e.ivor_low[0];
|
953 |
env->spr[SPR_BOOKE_IVOR1] = sregs.u.e.ivor_low[1];
|
954 |
env->spr[SPR_BOOKE_IVOR2] = sregs.u.e.ivor_low[2];
|
955 |
env->spr[SPR_BOOKE_IVOR3] = sregs.u.e.ivor_low[3];
|
956 |
env->spr[SPR_BOOKE_IVOR4] = sregs.u.e.ivor_low[4];
|
957 |
env->spr[SPR_BOOKE_IVOR5] = sregs.u.e.ivor_low[5];
|
958 |
env->spr[SPR_BOOKE_IVOR6] = sregs.u.e.ivor_low[6];
|
959 |
env->spr[SPR_BOOKE_IVOR7] = sregs.u.e.ivor_low[7];
|
960 |
env->spr[SPR_BOOKE_IVOR8] = sregs.u.e.ivor_low[8];
|
961 |
env->spr[SPR_BOOKE_IVOR9] = sregs.u.e.ivor_low[9];
|
962 |
env->spr[SPR_BOOKE_IVOR10] = sregs.u.e.ivor_low[10];
|
963 |
env->spr[SPR_BOOKE_IVOR11] = sregs.u.e.ivor_low[11];
|
964 |
env->spr[SPR_BOOKE_IVOR12] = sregs.u.e.ivor_low[12];
|
965 |
env->spr[SPR_BOOKE_IVOR13] = sregs.u.e.ivor_low[13];
|
966 |
env->spr[SPR_BOOKE_IVOR14] = sregs.u.e.ivor_low[14];
|
967 |
env->spr[SPR_BOOKE_IVOR15] = sregs.u.e.ivor_low[15];
|
968 |
|
969 |
if (sregs.u.e.features & KVM_SREGS_E_SPE) {
|
970 |
env->spr[SPR_BOOKE_IVOR32] = sregs.u.e.ivor_high[0];
|
971 |
env->spr[SPR_BOOKE_IVOR33] = sregs.u.e.ivor_high[1];
|
972 |
env->spr[SPR_BOOKE_IVOR34] = sregs.u.e.ivor_high[2];
|
973 |
} |
974 |
|
975 |
if (sregs.u.e.features & KVM_SREGS_E_PM) {
|
976 |
env->spr[SPR_BOOKE_IVOR35] = sregs.u.e.ivor_high[3];
|
977 |
} |
978 |
|
979 |
if (sregs.u.e.features & KVM_SREGS_E_PC) {
|
980 |
env->spr[SPR_BOOKE_IVOR36] = sregs.u.e.ivor_high[4];
|
981 |
env->spr[SPR_BOOKE_IVOR37] = sregs.u.e.ivor_high[5];
|
982 |
} |
983 |
} |
984 |
|
985 |
if (sregs.u.e.features & KVM_SREGS_E_ARCH206_MMU) {
|
986 |
env->spr[SPR_BOOKE_MAS0] = sregs.u.e.mas0; |
987 |
env->spr[SPR_BOOKE_MAS1] = sregs.u.e.mas1; |
988 |
env->spr[SPR_BOOKE_MAS2] = sregs.u.e.mas2; |
989 |
env->spr[SPR_BOOKE_MAS3] = sregs.u.e.mas7_3 & 0xffffffff;
|
990 |
env->spr[SPR_BOOKE_MAS4] = sregs.u.e.mas4; |
991 |
env->spr[SPR_BOOKE_MAS6] = sregs.u.e.mas6; |
992 |
env->spr[SPR_BOOKE_MAS7] = sregs.u.e.mas7_3 >> 32;
|
993 |
env->spr[SPR_MMUCFG] = sregs.u.e.mmucfg; |
994 |
env->spr[SPR_BOOKE_TLB0CFG] = sregs.u.e.tlbcfg[0];
|
995 |
env->spr[SPR_BOOKE_TLB1CFG] = sregs.u.e.tlbcfg[1];
|
996 |
} |
997 |
|
998 |
if (sregs.u.e.features & KVM_SREGS_EXP) {
|
999 |
env->spr[SPR_BOOKE_EPR] = sregs.u.e.epr; |
1000 |
} |
1001 |
|
1002 |
if (sregs.u.e.features & KVM_SREGS_E_PD) {
|
1003 |
env->spr[SPR_BOOKE_EPLC] = sregs.u.e.eplc; |
1004 |
env->spr[SPR_BOOKE_EPSC] = sregs.u.e.epsc; |
1005 |
} |
1006 |
|
1007 |
if (sregs.u.e.impl_id == KVM_SREGS_E_IMPL_FSL) {
|
1008 |
env->spr[SPR_E500_SVR] = sregs.u.e.impl.fsl.svr; |
1009 |
env->spr[SPR_Exxx_MCAR] = sregs.u.e.impl.fsl.mcar; |
1010 |
env->spr[SPR_HID0] = sregs.u.e.impl.fsl.hid0; |
1011 |
|
1012 |
if (sregs.u.e.impl.fsl.features & KVM_SREGS_E_FSL_PIDn) {
|
1013 |
env->spr[SPR_BOOKE_PID1] = sregs.u.e.impl.fsl.pid1; |
1014 |
env->spr[SPR_BOOKE_PID2] = sregs.u.e.impl.fsl.pid2; |
1015 |
} |
1016 |
} |
1017 |
} |
1018 |
|
1019 |
if (cap_segstate) {
|
1020 |
ret = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs); |
1021 |
if (ret < 0) { |
1022 |
return ret;
|
1023 |
} |
1024 |
|
1025 |
ppc_store_sdr1(env, sregs.u.s.sdr1); |
1026 |
|
1027 |
/* Sync SLB */
|
1028 |
#ifdef TARGET_PPC64
|
1029 |
for (i = 0; i < 64; i++) { |
1030 |
ppc_store_slb(env, sregs.u.s.ppc64.slb[i].slbe, |
1031 |
sregs.u.s.ppc64.slb[i].slbv); |
1032 |
} |
1033 |
#endif
|
1034 |
|
1035 |
/* Sync SRs */
|
1036 |
for (i = 0; i < 16; i++) { |
1037 |
env->sr[i] = sregs.u.s.ppc32.sr[i]; |
1038 |
} |
1039 |
|
1040 |
/* Sync BATs */
|
1041 |
for (i = 0; i < 8; i++) { |
1042 |
env->DBAT[0][i] = sregs.u.s.ppc32.dbat[i] & 0xffffffff; |
1043 |
env->DBAT[1][i] = sregs.u.s.ppc32.dbat[i] >> 32; |
1044 |
env->IBAT[0][i] = sregs.u.s.ppc32.ibat[i] & 0xffffffff; |
1045 |
env->IBAT[1][i] = sregs.u.s.ppc32.ibat[i] >> 32; |
1046 |
} |
1047 |
} |
1048 |
|
1049 |
if (cap_hior) {
|
1050 |
kvm_get_one_spr(cs, KVM_REG_PPC_HIOR, SPR_HIOR); |
1051 |
} |
1052 |
|
1053 |
if (cap_one_reg) {
|
1054 |
int i;
|
1055 |
|
1056 |
/* We deliberately ignore errors here, for kernels which have
|
1057 |
* the ONE_REG calls, but don't support the specific
|
1058 |
* registers, there's a reasonable chance things will still
|
1059 |
* work, at least until we try to migrate. */
|
1060 |
for (i = 0; i < 1024; i++) { |
1061 |
uint64_t id = env->spr_cb[i].one_reg_id; |
1062 |
|
1063 |
if (id != 0) { |
1064 |
kvm_get_one_spr(cs, id, i); |
1065 |
} |
1066 |
} |
1067 |
|
1068 |
#ifdef TARGET_PPC64
|
1069 |
if (cap_papr) {
|
1070 |
if (kvm_get_vpa(cs) < 0) { |
1071 |
dprintf("Warning: Unable to get VPA information from KVM\n");
|
1072 |
} |
1073 |
} |
1074 |
#endif
|
1075 |
} |
1076 |
|
1077 |
return 0; |
1078 |
} |
1079 |
|
1080 |
int kvmppc_set_interrupt(PowerPCCPU *cpu, int irq, int level) |
1081 |
{ |
1082 |
unsigned virq = level ? KVM_INTERRUPT_SET_LEVEL : KVM_INTERRUPT_UNSET;
|
1083 |
|
1084 |
if (irq != PPC_INTERRUPT_EXT) {
|
1085 |
return 0; |
1086 |
} |
1087 |
|
1088 |
if (!kvm_enabled() || !cap_interrupt_unset || !cap_interrupt_level) {
|
1089 |
return 0; |
1090 |
} |
1091 |
|
1092 |
kvm_vcpu_ioctl(CPU(cpu), KVM_INTERRUPT, &virq); |
1093 |
|
1094 |
return 0; |
1095 |
} |
1096 |
|
1097 |
#if defined(TARGET_PPCEMB)
|
1098 |
#define PPC_INPUT_INT PPC40x_INPUT_INT
|
1099 |
#elif defined(TARGET_PPC64)
|
1100 |
#define PPC_INPUT_INT PPC970_INPUT_INT
|
1101 |
#else
|
1102 |
#define PPC_INPUT_INT PPC6xx_INPUT_INT
|
1103 |
#endif
|
1104 |
|
1105 |
void kvm_arch_pre_run(CPUState *cs, struct kvm_run *run) |
1106 |
{ |
1107 |
PowerPCCPU *cpu = POWERPC_CPU(cs); |
1108 |
CPUPPCState *env = &cpu->env; |
1109 |
int r;
|
1110 |
unsigned irq;
|
1111 |
|
1112 |
/* PowerPC QEMU tracks the various core input pins (interrupt, critical
|
1113 |
* interrupt, reset, etc) in PPC-specific env->irq_input_state. */
|
1114 |
if (!cap_interrupt_level &&
|
1115 |
run->ready_for_interrupt_injection && |
1116 |
(cs->interrupt_request & CPU_INTERRUPT_HARD) && |
1117 |
(env->irq_input_state & (1<<PPC_INPUT_INT)))
|
1118 |
{ |
1119 |
/* For now KVM disregards the 'irq' argument. However, in the
|
1120 |
* future KVM could cache it in-kernel to avoid a heavyweight exit
|
1121 |
* when reading the UIC.
|
1122 |
*/
|
1123 |
irq = KVM_INTERRUPT_SET; |
1124 |
|
1125 |
dprintf("injected interrupt %d\n", irq);
|
1126 |
r = kvm_vcpu_ioctl(cs, KVM_INTERRUPT, &irq); |
1127 |
if (r < 0) { |
1128 |
printf("cpu %d fail inject %x\n", cs->cpu_index, irq);
|
1129 |
} |
1130 |
|
1131 |
/* Always wake up soon in case the interrupt was level based */
|
1132 |
qemu_mod_timer(idle_timer, qemu_get_clock_ns(vm_clock) + |
1133 |
(get_ticks_per_sec() / 50));
|
1134 |
} |
1135 |
|
1136 |
/* We don't know if there are more interrupts pending after this. However,
|
1137 |
* the guest will return to userspace in the course of handling this one
|
1138 |
* anyways, so we will get a chance to deliver the rest. */
|
1139 |
} |
1140 |
|
1141 |
void kvm_arch_post_run(CPUState *cpu, struct kvm_run *run) |
1142 |
{ |
1143 |
} |
1144 |
|
1145 |
int kvm_arch_process_async_events(CPUState *cs)
|
1146 |
{ |
1147 |
return cs->halted;
|
1148 |
} |
1149 |
|
1150 |
static int kvmppc_handle_halt(PowerPCCPU *cpu) |
1151 |
{ |
1152 |
CPUState *cs = CPU(cpu); |
1153 |
CPUPPCState *env = &cpu->env; |
1154 |
|
1155 |
if (!(cs->interrupt_request & CPU_INTERRUPT_HARD) && (msr_ee)) {
|
1156 |
cs->halted = 1;
|
1157 |
env->exception_index = EXCP_HLT; |
1158 |
} |
1159 |
|
1160 |
return 0; |
1161 |
} |
1162 |
|
1163 |
/* map dcr access to existing qemu dcr emulation */
|
1164 |
static int kvmppc_handle_dcr_read(CPUPPCState *env, uint32_t dcrn, uint32_t *data) |
1165 |
{ |
1166 |
if (ppc_dcr_read(env->dcr_env, dcrn, data) < 0) |
1167 |
fprintf(stderr, "Read to unhandled DCR (0x%x)\n", dcrn);
|
1168 |
|
1169 |
return 0; |
1170 |
} |
1171 |
|
1172 |
static int kvmppc_handle_dcr_write(CPUPPCState *env, uint32_t dcrn, uint32_t data) |
1173 |
{ |
1174 |
if (ppc_dcr_write(env->dcr_env, dcrn, data) < 0) |
1175 |
fprintf(stderr, "Write to unhandled DCR (0x%x)\n", dcrn);
|
1176 |
|
1177 |
return 0; |
1178 |
} |
1179 |
|
1180 |
int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run) |
1181 |
{ |
1182 |
PowerPCCPU *cpu = POWERPC_CPU(cs); |
1183 |
CPUPPCState *env = &cpu->env; |
1184 |
int ret;
|
1185 |
|
1186 |
switch (run->exit_reason) {
|
1187 |
case KVM_EXIT_DCR:
|
1188 |
if (run->dcr.is_write) {
|
1189 |
dprintf("handle dcr write\n");
|
1190 |
ret = kvmppc_handle_dcr_write(env, run->dcr.dcrn, run->dcr.data); |
1191 |
} else {
|
1192 |
dprintf("handle dcr read\n");
|
1193 |
ret = kvmppc_handle_dcr_read(env, run->dcr.dcrn, &run->dcr.data); |
1194 |
} |
1195 |
break;
|
1196 |
case KVM_EXIT_HLT:
|
1197 |
dprintf("handle halt\n");
|
1198 |
ret = kvmppc_handle_halt(cpu); |
1199 |
break;
|
1200 |
#if defined(TARGET_PPC64)
|
1201 |
case KVM_EXIT_PAPR_HCALL:
|
1202 |
dprintf("handle PAPR hypercall\n");
|
1203 |
run->papr_hcall.ret = spapr_hypercall(cpu, |
1204 |
run->papr_hcall.nr, |
1205 |
run->papr_hcall.args); |
1206 |
ret = 0;
|
1207 |
break;
|
1208 |
#endif
|
1209 |
case KVM_EXIT_EPR:
|
1210 |
dprintf("handle epr\n");
|
1211 |
run->epr.epr = ldl_phys(env->mpic_iack); |
1212 |
ret = 0;
|
1213 |
break;
|
1214 |
case KVM_EXIT_WATCHDOG:
|
1215 |
dprintf("handle watchdog expiry\n");
|
1216 |
watchdog_perform_action(); |
1217 |
ret = 0;
|
1218 |
break;
|
1219 |
|
1220 |
default:
|
1221 |
fprintf(stderr, "KVM: unknown exit reason %d\n", run->exit_reason);
|
1222 |
ret = -1;
|
1223 |
break;
|
1224 |
} |
1225 |
|
1226 |
return ret;
|
1227 |
} |
1228 |
|
1229 |
int kvmppc_or_tsr_bits(PowerPCCPU *cpu, uint32_t tsr_bits)
|
1230 |
{ |
1231 |
CPUState *cs = CPU(cpu); |
1232 |
uint32_t bits = tsr_bits; |
1233 |
struct kvm_one_reg reg = {
|
1234 |
.id = KVM_REG_PPC_OR_TSR, |
1235 |
.addr = (uintptr_t) &bits, |
1236 |
}; |
1237 |
|
1238 |
return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
|
1239 |
} |
1240 |
|
1241 |
int kvmppc_clear_tsr_bits(PowerPCCPU *cpu, uint32_t tsr_bits)
|
1242 |
{ |
1243 |
|
1244 |
CPUState *cs = CPU(cpu); |
1245 |
uint32_t bits = tsr_bits; |
1246 |
struct kvm_one_reg reg = {
|
1247 |
.id = KVM_REG_PPC_CLEAR_TSR, |
1248 |
.addr = (uintptr_t) &bits, |
1249 |
}; |
1250 |
|
1251 |
return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
|
1252 |
} |
1253 |
|
1254 |
int kvmppc_set_tcr(PowerPCCPU *cpu)
|
1255 |
{ |
1256 |
CPUState *cs = CPU(cpu); |
1257 |
CPUPPCState *env = &cpu->env; |
1258 |
uint32_t tcr = env->spr[SPR_BOOKE_TCR]; |
1259 |
|
1260 |
struct kvm_one_reg reg = {
|
1261 |
.id = KVM_REG_PPC_TCR, |
1262 |
.addr = (uintptr_t) &tcr, |
1263 |
}; |
1264 |
|
1265 |
return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
|
1266 |
} |
1267 |
|
1268 |
int kvmppc_booke_watchdog_enable(PowerPCCPU *cpu)
|
1269 |
{ |
1270 |
CPUState *cs = CPU(cpu); |
1271 |
struct kvm_enable_cap encap = {};
|
1272 |
int ret;
|
1273 |
|
1274 |
if (!kvm_enabled()) {
|
1275 |
return -1; |
1276 |
} |
1277 |
|
1278 |
if (!cap_ppc_watchdog) {
|
1279 |
printf("warning: KVM does not support watchdog");
|
1280 |
return -1; |
1281 |
} |
1282 |
|
1283 |
encap.cap = KVM_CAP_PPC_BOOKE_WATCHDOG; |
1284 |
ret = kvm_vcpu_ioctl(cs, KVM_ENABLE_CAP, &encap); |
1285 |
if (ret < 0) { |
1286 |
fprintf(stderr, "%s: couldn't enable KVM_CAP_PPC_BOOKE_WATCHDOG: %s\n",
|
1287 |
__func__, strerror(-ret)); |
1288 |
return ret;
|
1289 |
} |
1290 |
|
1291 |
return ret;
|
1292 |
} |
1293 |
|
1294 |
static int read_cpuinfo(const char *field, char *value, int len) |
1295 |
{ |
1296 |
FILE *f; |
1297 |
int ret = -1; |
1298 |
int field_len = strlen(field);
|
1299 |
char line[512]; |
1300 |
|
1301 |
f = fopen("/proc/cpuinfo", "r"); |
1302 |
if (!f) {
|
1303 |
return -1; |
1304 |
} |
1305 |
|
1306 |
do {
|
1307 |
if(!fgets(line, sizeof(line), f)) { |
1308 |
break;
|
1309 |
} |
1310 |
if (!strncmp(line, field, field_len)) {
|
1311 |
pstrcpy(value, len, line); |
1312 |
ret = 0;
|
1313 |
break;
|
1314 |
} |
1315 |
} while(*line);
|
1316 |
|
1317 |
fclose(f); |
1318 |
|
1319 |
return ret;
|
1320 |
} |
1321 |
|
1322 |
uint32_t kvmppc_get_tbfreq(void)
|
1323 |
{ |
1324 |
char line[512]; |
1325 |
char *ns;
|
1326 |
uint32_t retval = get_ticks_per_sec(); |
1327 |
|
1328 |
if (read_cpuinfo("timebase", line, sizeof(line))) { |
1329 |
return retval;
|
1330 |
} |
1331 |
|
1332 |
if (!(ns = strchr(line, ':'))) { |
1333 |
return retval;
|
1334 |
} |
1335 |
|
1336 |
ns++; |
1337 |
|
1338 |
retval = atoi(ns); |
1339 |
return retval;
|
1340 |
} |
1341 |
|
1342 |
/* Try to find a device tree node for a CPU with clock-frequency property */
|
1343 |
static int kvmppc_find_cpu_dt(char *buf, int buf_len) |
1344 |
{ |
1345 |
struct dirent *dirp;
|
1346 |
DIR *dp; |
1347 |
|
1348 |
if ((dp = opendir(PROC_DEVTREE_CPU)) == NULL) { |
1349 |
printf("Can't open directory " PROC_DEVTREE_CPU "\n"); |
1350 |
return -1; |
1351 |
} |
1352 |
|
1353 |
buf[0] = '\0'; |
1354 |
while ((dirp = readdir(dp)) != NULL) { |
1355 |
FILE *f; |
1356 |
snprintf(buf, buf_len, "%s%s/clock-frequency", PROC_DEVTREE_CPU,
|
1357 |
dirp->d_name); |
1358 |
f = fopen(buf, "r");
|
1359 |
if (f) {
|
1360 |
snprintf(buf, buf_len, "%s%s", PROC_DEVTREE_CPU, dirp->d_name);
|
1361 |
fclose(f); |
1362 |
break;
|
1363 |
} |
1364 |
buf[0] = '\0'; |
1365 |
} |
1366 |
closedir(dp); |
1367 |
if (buf[0] == '\0') { |
1368 |
printf("Unknown host!\n");
|
1369 |
return -1; |
1370 |
} |
1371 |
|
1372 |
return 0; |
1373 |
} |
1374 |
|
1375 |
/* Read a CPU node property from the host device tree that's a single
|
1376 |
* integer (32-bit or 64-bit). Returns 0 if anything goes wrong
|
1377 |
* (can't find or open the property, or doesn't understand the
|
1378 |
* format) */
|
1379 |
static uint64_t kvmppc_read_int_cpu_dt(const char *propname) |
1380 |
{ |
1381 |
char buf[PATH_MAX];
|
1382 |
union {
|
1383 |
uint32_t v32; |
1384 |
uint64_t v64; |
1385 |
} u; |
1386 |
FILE *f; |
1387 |
int len;
|
1388 |
|
1389 |
if (kvmppc_find_cpu_dt(buf, sizeof(buf))) { |
1390 |
return -1; |
1391 |
} |
1392 |
|
1393 |
strncat(buf, "/", sizeof(buf) - strlen(buf)); |
1394 |
strncat(buf, propname, sizeof(buf) - strlen(buf));
|
1395 |
|
1396 |
f = fopen(buf, "rb");
|
1397 |
if (!f) {
|
1398 |
return -1; |
1399 |
} |
1400 |
|
1401 |
len = fread(&u, 1, sizeof(u), f); |
1402 |
fclose(f); |
1403 |
switch (len) {
|
1404 |
case 4: |
1405 |
/* property is a 32-bit quantity */
|
1406 |
return be32_to_cpu(u.v32);
|
1407 |
case 8: |
1408 |
return be64_to_cpu(u.v64);
|
1409 |
} |
1410 |
|
1411 |
return 0; |
1412 |
} |
1413 |
|
1414 |
uint64_t kvmppc_get_clockfreq(void)
|
1415 |
{ |
1416 |
return kvmppc_read_int_cpu_dt("clock-frequency"); |
1417 |
} |
1418 |
|
1419 |
uint32_t kvmppc_get_vmx(void)
|
1420 |
{ |
1421 |
return kvmppc_read_int_cpu_dt("ibm,vmx"); |
1422 |
} |
1423 |
|
1424 |
uint32_t kvmppc_get_dfp(void)
|
1425 |
{ |
1426 |
return kvmppc_read_int_cpu_dt("ibm,dfp"); |
1427 |
} |
1428 |
|
1429 |
static int kvmppc_get_pvinfo(CPUPPCState *env, struct kvm_ppc_pvinfo *pvinfo) |
1430 |
{ |
1431 |
PowerPCCPU *cpu = ppc_env_get_cpu(env); |
1432 |
CPUState *cs = CPU(cpu); |
1433 |
|
1434 |
if (kvm_check_extension(cs->kvm_state, KVM_CAP_PPC_GET_PVINFO) &&
|
1435 |
!kvm_vm_ioctl(cs->kvm_state, KVM_PPC_GET_PVINFO, pvinfo)) { |
1436 |
return 0; |
1437 |
} |
1438 |
|
1439 |
return 1; |
1440 |
} |
1441 |
|
1442 |
int kvmppc_get_hasidle(CPUPPCState *env)
|
1443 |
{ |
1444 |
struct kvm_ppc_pvinfo pvinfo;
|
1445 |
|
1446 |
if (!kvmppc_get_pvinfo(env, &pvinfo) &&
|
1447 |
(pvinfo.flags & KVM_PPC_PVINFO_FLAGS_EV_IDLE)) { |
1448 |
return 1; |
1449 |
} |
1450 |
|
1451 |
return 0; |
1452 |
} |
1453 |
|
1454 |
int kvmppc_get_hypercall(CPUPPCState *env, uint8_t *buf, int buf_len) |
1455 |
{ |
1456 |
uint32_t *hc = (uint32_t*)buf; |
1457 |
struct kvm_ppc_pvinfo pvinfo;
|
1458 |
|
1459 |
if (!kvmppc_get_pvinfo(env, &pvinfo)) {
|
1460 |
memcpy(buf, pvinfo.hcall, buf_len); |
1461 |
return 0; |
1462 |
} |
1463 |
|
1464 |
/*
|
1465 |
* Fallback to always fail hypercalls:
|
1466 |
*
|
1467 |
* li r3, -1
|
1468 |
* nop
|
1469 |
* nop
|
1470 |
* nop
|
1471 |
*/
|
1472 |
|
1473 |
hc[0] = 0x3860ffff; |
1474 |
hc[1] = 0x60000000; |
1475 |
hc[2] = 0x60000000; |
1476 |
hc[3] = 0x60000000; |
1477 |
|
1478 |
return 0; |
1479 |
} |
1480 |
|
1481 |
void kvmppc_set_papr(PowerPCCPU *cpu)
|
1482 |
{ |
1483 |
CPUPPCState *env = &cpu->env; |
1484 |
CPUState *cs = CPU(cpu); |
1485 |
struct kvm_enable_cap cap = {};
|
1486 |
int ret;
|
1487 |
|
1488 |
cap.cap = KVM_CAP_PPC_PAPR; |
1489 |
ret = kvm_vcpu_ioctl(cs, KVM_ENABLE_CAP, &cap); |
1490 |
|
1491 |
if (ret) {
|
1492 |
cpu_abort(env, "This KVM version does not support PAPR\n");
|
1493 |
} |
1494 |
|
1495 |
/* Update the capability flag so we sync the right information
|
1496 |
* with kvm */
|
1497 |
cap_papr = 1;
|
1498 |
} |
1499 |
|
1500 |
void kvmppc_set_mpic_proxy(PowerPCCPU *cpu, int mpic_proxy) |
1501 |
{ |
1502 |
CPUPPCState *env = &cpu->env; |
1503 |
CPUState *cs = CPU(cpu); |
1504 |
struct kvm_enable_cap cap = {};
|
1505 |
int ret;
|
1506 |
|
1507 |
cap.cap = KVM_CAP_PPC_EPR; |
1508 |
cap.args[0] = mpic_proxy;
|
1509 |
ret = kvm_vcpu_ioctl(cs, KVM_ENABLE_CAP, &cap); |
1510 |
|
1511 |
if (ret && mpic_proxy) {
|
1512 |
cpu_abort(env, "This KVM version does not support EPR\n");
|
1513 |
} |
1514 |
} |
1515 |
|
1516 |
int kvmppc_smt_threads(void) |
1517 |
{ |
1518 |
return cap_ppc_smt ? cap_ppc_smt : 1; |
1519 |
} |
1520 |
|
1521 |
#ifdef TARGET_PPC64
|
1522 |
off_t kvmppc_alloc_rma(const char *name, MemoryRegion *sysmem) |
1523 |
{ |
1524 |
void *rma;
|
1525 |
off_t size; |
1526 |
int fd;
|
1527 |
struct kvm_allocate_rma ret;
|
1528 |
MemoryRegion *rma_region; |
1529 |
|
1530 |
/* If cap_ppc_rma == 0, contiguous RMA allocation is not supported
|
1531 |
* if cap_ppc_rma == 1, contiguous RMA allocation is supported, but
|
1532 |
* not necessary on this hardware
|
1533 |
* if cap_ppc_rma == 2, contiguous RMA allocation is needed on this hardware
|
1534 |
*
|
1535 |
* FIXME: We should allow the user to force contiguous RMA
|
1536 |
* allocation in the cap_ppc_rma==1 case.
|
1537 |
*/
|
1538 |
if (cap_ppc_rma < 2) { |
1539 |
return 0; |
1540 |
} |
1541 |
|
1542 |
fd = kvm_vm_ioctl(kvm_state, KVM_ALLOCATE_RMA, &ret); |
1543 |
if (fd < 0) { |
1544 |
fprintf(stderr, "KVM: Error on KVM_ALLOCATE_RMA: %s\n",
|
1545 |
strerror(errno)); |
1546 |
return -1; |
1547 |
} |
1548 |
|
1549 |
size = MIN(ret.rma_size, 256ul << 20); |
1550 |
|
1551 |
rma = mmap(NULL, size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0); |
1552 |
if (rma == MAP_FAILED) {
|
1553 |
fprintf(stderr, "KVM: Error mapping RMA: %s\n", strerror(errno));
|
1554 |
return -1; |
1555 |
}; |
1556 |
|
1557 |
rma_region = g_new(MemoryRegion, 1);
|
1558 |
memory_region_init_ram_ptr(rma_region, name, size, rma); |
1559 |
vmstate_register_ram_global(rma_region); |
1560 |
memory_region_add_subregion(sysmem, 0, rma_region);
|
1561 |
|
1562 |
return size;
|
1563 |
} |
1564 |
|
1565 |
uint64_t kvmppc_rma_size(uint64_t current_size, unsigned int hash_shift) |
1566 |
{ |
1567 |
struct kvm_ppc_smmu_info info;
|
1568 |
long rampagesize, best_page_shift;
|
1569 |
int i;
|
1570 |
|
1571 |
if (cap_ppc_rma >= 2) { |
1572 |
return current_size;
|
1573 |
} |
1574 |
|
1575 |
/* Find the largest hardware supported page size that's less than
|
1576 |
* or equal to the (logical) backing page size of guest RAM */
|
1577 |
kvm_get_smmu_info(ppc_env_get_cpu(first_cpu), &info); |
1578 |
rampagesize = getrampagesize(); |
1579 |
best_page_shift = 0;
|
1580 |
|
1581 |
for (i = 0; i < KVM_PPC_PAGE_SIZES_MAX_SZ; i++) { |
1582 |
struct kvm_ppc_one_seg_page_size *sps = &info.sps[i];
|
1583 |
|
1584 |
if (!sps->page_shift) {
|
1585 |
continue;
|
1586 |
} |
1587 |
|
1588 |
if ((sps->page_shift > best_page_shift)
|
1589 |
&& ((1UL << sps->page_shift) <= rampagesize)) {
|
1590 |
best_page_shift = sps->page_shift; |
1591 |
} |
1592 |
} |
1593 |
|
1594 |
return MIN(current_size,
|
1595 |
1ULL << (best_page_shift + hash_shift - 7)); |
1596 |
} |
1597 |
#endif
|
1598 |
|
1599 |
void *kvmppc_create_spapr_tce(uint32_t liobn, uint32_t window_size, int *pfd) |
1600 |
{ |
1601 |
struct kvm_create_spapr_tce args = {
|
1602 |
.liobn = liobn, |
1603 |
.window_size = window_size, |
1604 |
}; |
1605 |
long len;
|
1606 |
int fd;
|
1607 |
void *table;
|
1608 |
|
1609 |
/* Must set fd to -1 so we don't try to munmap when called for
|
1610 |
* destroying the table, which the upper layers -will- do
|
1611 |
*/
|
1612 |
*pfd = -1;
|
1613 |
if (!cap_spapr_tce) {
|
1614 |
return NULL; |
1615 |
} |
1616 |
|
1617 |
fd = kvm_vm_ioctl(kvm_state, KVM_CREATE_SPAPR_TCE, &args); |
1618 |
if (fd < 0) { |
1619 |
fprintf(stderr, "KVM: Failed to create TCE table for liobn 0x%x\n",
|
1620 |
liobn); |
1621 |
return NULL; |
1622 |
} |
1623 |
|
1624 |
len = (window_size / SPAPR_TCE_PAGE_SIZE) * sizeof(sPAPRTCE);
|
1625 |
/* FIXME: round this up to page size */
|
1626 |
|
1627 |
table = mmap(NULL, len, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0); |
1628 |
if (table == MAP_FAILED) {
|
1629 |
fprintf(stderr, "KVM: Failed to map TCE table for liobn 0x%x\n",
|
1630 |
liobn); |
1631 |
close(fd); |
1632 |
return NULL; |
1633 |
} |
1634 |
|
1635 |
*pfd = fd; |
1636 |
return table;
|
1637 |
} |
1638 |
|
1639 |
int kvmppc_remove_spapr_tce(void *table, int fd, uint32_t window_size) |
1640 |
{ |
1641 |
long len;
|
1642 |
|
1643 |
if (fd < 0) { |
1644 |
return -1; |
1645 |
} |
1646 |
|
1647 |
len = (window_size / SPAPR_TCE_PAGE_SIZE)*sizeof(sPAPRTCE);
|
1648 |
if ((munmap(table, len) < 0) || |
1649 |
(close(fd) < 0)) {
|
1650 |
fprintf(stderr, "KVM: Unexpected error removing TCE table: %s",
|
1651 |
strerror(errno)); |
1652 |
/* Leak the table */
|
1653 |
} |
1654 |
|
1655 |
return 0; |
1656 |
} |
1657 |
|
1658 |
int kvmppc_reset_htab(int shift_hint) |
1659 |
{ |
1660 |
uint32_t shift = shift_hint; |
1661 |
|
1662 |
if (!kvm_enabled()) {
|
1663 |
/* Full emulation, tell caller to allocate htab itself */
|
1664 |
return 0; |
1665 |
} |
1666 |
if (kvm_check_extension(kvm_state, KVM_CAP_PPC_ALLOC_HTAB)) {
|
1667 |
int ret;
|
1668 |
ret = kvm_vm_ioctl(kvm_state, KVM_PPC_ALLOCATE_HTAB, &shift); |
1669 |
if (ret == -ENOTTY) {
|
1670 |
/* At least some versions of PR KVM advertise the
|
1671 |
* capability, but don't implement the ioctl(). Oops.
|
1672 |
* Return 0 so that we allocate the htab in qemu, as is
|
1673 |
* correct for PR. */
|
1674 |
return 0; |
1675 |
} else if (ret < 0) { |
1676 |
return ret;
|
1677 |
} |
1678 |
return shift;
|
1679 |
} |
1680 |
|
1681 |
/* We have a kernel that predates the htab reset calls. For PR
|
1682 |
* KVM, we need to allocate the htab ourselves, for an HV KVM of
|
1683 |
* this era, it has allocated a 16MB fixed size hash table
|
1684 |
* already. Kernels of this era have the GET_PVINFO capability
|
1685 |
* only on PR, so we use this hack to determine the right
|
1686 |
* answer */
|
1687 |
if (kvm_check_extension(kvm_state, KVM_CAP_PPC_GET_PVINFO)) {
|
1688 |
/* PR - tell caller to allocate htab */
|
1689 |
return 0; |
1690 |
} else {
|
1691 |
/* HV - assume 16MB kernel allocated htab */
|
1692 |
return 24; |
1693 |
} |
1694 |
} |
1695 |
|
1696 |
static inline uint32_t mfpvr(void) |
1697 |
{ |
1698 |
uint32_t pvr; |
1699 |
|
1700 |
asm ("mfpvr %0" |
1701 |
: "=r"(pvr));
|
1702 |
return pvr;
|
1703 |
} |
1704 |
|
1705 |
static void alter_insns(uint64_t *word, uint64_t flags, bool on) |
1706 |
{ |
1707 |
if (on) {
|
1708 |
*word |= flags; |
1709 |
} else {
|
1710 |
*word &= ~flags; |
1711 |
} |
1712 |
} |
1713 |
|
1714 |
static void kvmppc_host_cpu_initfn(Object *obj) |
1715 |
{ |
1716 |
assert(kvm_enabled()); |
1717 |
} |
1718 |
|
1719 |
static void kvmppc_host_cpu_class_init(ObjectClass *oc, void *data) |
1720 |
{ |
1721 |
PowerPCCPUClass *pcc = POWERPC_CPU_CLASS(oc); |
1722 |
uint32_t vmx = kvmppc_get_vmx(); |
1723 |
uint32_t dfp = kvmppc_get_dfp(); |
1724 |
uint32_t dcache_size = kvmppc_read_int_cpu_dt("d-cache-size");
|
1725 |
uint32_t icache_size = kvmppc_read_int_cpu_dt("i-cache-size");
|
1726 |
|
1727 |
/* Now fix up the class with information we can query from the host */
|
1728 |
|
1729 |
if (vmx != -1) { |
1730 |
/* Only override when we know what the host supports */
|
1731 |
alter_insns(&pcc->insns_flags, PPC_ALTIVEC, vmx > 0);
|
1732 |
alter_insns(&pcc->insns_flags2, PPC2_VSX, vmx > 1);
|
1733 |
} |
1734 |
if (dfp != -1) { |
1735 |
/* Only override when we know what the host supports */
|
1736 |
alter_insns(&pcc->insns_flags2, PPC2_DFP, dfp); |
1737 |
} |
1738 |
|
1739 |
if (dcache_size != -1) { |
1740 |
pcc->l1_dcache_size = dcache_size; |
1741 |
} |
1742 |
|
1743 |
if (icache_size != -1) { |
1744 |
pcc->l1_icache_size = icache_size; |
1745 |
} |
1746 |
} |
1747 |
|
1748 |
int kvmppc_fixup_cpu(PowerPCCPU *cpu)
|
1749 |
{ |
1750 |
CPUState *cs = CPU(cpu); |
1751 |
int smt;
|
1752 |
|
1753 |
/* Adjust cpu index for SMT */
|
1754 |
smt = kvmppc_smt_threads(); |
1755 |
cs->cpu_index = (cs->cpu_index / smp_threads) * smt |
1756 |
+ (cs->cpu_index % smp_threads); |
1757 |
|
1758 |
return 0; |
1759 |
} |
1760 |
|
1761 |
bool kvmppc_has_cap_epr(void) |
1762 |
{ |
1763 |
return cap_epr;
|
1764 |
} |
1765 |
|
1766 |
static int kvm_ppc_register_host_cpu_type(void) |
1767 |
{ |
1768 |
TypeInfo type_info = { |
1769 |
.name = TYPE_HOST_POWERPC_CPU, |
1770 |
.instance_init = kvmppc_host_cpu_initfn, |
1771 |
.class_init = kvmppc_host_cpu_class_init, |
1772 |
}; |
1773 |
uint32_t host_pvr = mfpvr(); |
1774 |
PowerPCCPUClass *pvr_pcc; |
1775 |
|
1776 |
pvr_pcc = ppc_cpu_class_by_pvr(host_pvr); |
1777 |
if (pvr_pcc == NULL) { |
1778 |
return -1; |
1779 |
} |
1780 |
type_info.parent = object_class_get_name(OBJECT_CLASS(pvr_pcc)); |
1781 |
type_register(&type_info); |
1782 |
return 0; |
1783 |
} |
1784 |
|
1785 |
|
1786 |
bool kvm_arch_stop_on_emulation_error(CPUState *cpu)
|
1787 |
{ |
1788 |
return true; |
1789 |
} |
1790 |
|
1791 |
int kvm_arch_on_sigbus_vcpu(CPUState *cpu, int code, void *addr) |
1792 |
{ |
1793 |
return 1; |
1794 |
} |
1795 |
|
1796 |
int kvm_arch_on_sigbus(int code, void *addr) |
1797 |
{ |
1798 |
return 1; |
1799 |
} |