root / target-ppc / kvm.c @ 1de7afc9
History | View | Annotate | Download (33.3 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.h" |
28 |
#include "kvm.h" |
29 |
#include "kvm_ppc.h" |
30 |
#include "cpu.h" |
31 |
#include "cpus.h" |
32 |
#include "device_tree.h" |
33 |
#include "hw/sysbus.h" |
34 |
#include "hw/spapr.h" |
35 |
|
36 |
#include "hw/sysbus.h" |
37 |
#include "hw/spapr.h" |
38 |
#include "hw/spapr_vio.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 |
|
65 |
/* XXX We have a race condition where we actually have a level triggered
|
66 |
* interrupt, but the infrastructure can't expose that yet, so the guest
|
67 |
* takes but ignores it, goes to sleep and never gets notified that there's
|
68 |
* still an interrupt pending.
|
69 |
*
|
70 |
* As a quick workaround, let's just wake up again 20 ms after we injected
|
71 |
* an interrupt. That way we can assure that we're always reinjecting
|
72 |
* interrupts in case the guest swallowed them.
|
73 |
*/
|
74 |
static QEMUTimer *idle_timer;
|
75 |
|
76 |
static void kvm_kick_cpu(void *opaque) |
77 |
{ |
78 |
PowerPCCPU *cpu = opaque; |
79 |
|
80 |
qemu_cpu_kick(CPU(cpu)); |
81 |
} |
82 |
|
83 |
int kvm_arch_init(KVMState *s)
|
84 |
{ |
85 |
cap_interrupt_unset = kvm_check_extension(s, KVM_CAP_PPC_UNSET_IRQ); |
86 |
cap_interrupt_level = kvm_check_extension(s, KVM_CAP_PPC_IRQ_LEVEL); |
87 |
cap_segstate = kvm_check_extension(s, KVM_CAP_PPC_SEGSTATE); |
88 |
cap_booke_sregs = kvm_check_extension(s, KVM_CAP_PPC_BOOKE_SREGS); |
89 |
cap_ppc_smt = kvm_check_extension(s, KVM_CAP_PPC_SMT); |
90 |
cap_ppc_rma = kvm_check_extension(s, KVM_CAP_PPC_RMA); |
91 |
cap_spapr_tce = kvm_check_extension(s, KVM_CAP_SPAPR_TCE); |
92 |
cap_hior = kvm_check_extension(s, KVM_CAP_PPC_HIOR); |
93 |
|
94 |
if (!cap_interrupt_level) {
|
95 |
fprintf(stderr, "KVM: Couldn't find level irq capability. Expect the "
|
96 |
"VM to stall at times!\n");
|
97 |
} |
98 |
|
99 |
return 0; |
100 |
} |
101 |
|
102 |
static int kvm_arch_sync_sregs(CPUPPCState *cenv) |
103 |
{ |
104 |
struct kvm_sregs sregs;
|
105 |
int ret;
|
106 |
|
107 |
if (cenv->excp_model == POWERPC_EXCP_BOOKE) {
|
108 |
/* What we're really trying to say is "if we're on BookE, we use
|
109 |
the native PVR for now". This is the only sane way to check
|
110 |
it though, so we potentially confuse users that they can run
|
111 |
BookE guests on BookS. Let's hope nobody dares enough :) */
|
112 |
return 0; |
113 |
} else {
|
114 |
if (!cap_segstate) {
|
115 |
fprintf(stderr, "kvm error: missing PVR setting capability\n");
|
116 |
return -ENOSYS;
|
117 |
} |
118 |
} |
119 |
|
120 |
ret = kvm_vcpu_ioctl(cenv, KVM_GET_SREGS, &sregs); |
121 |
if (ret) {
|
122 |
return ret;
|
123 |
} |
124 |
|
125 |
sregs.pvr = cenv->spr[SPR_PVR]; |
126 |
return kvm_vcpu_ioctl(cenv, KVM_SET_SREGS, &sregs);
|
127 |
} |
128 |
|
129 |
/* Set up a shared TLB array with KVM */
|
130 |
static int kvm_booke206_tlb_init(CPUPPCState *env) |
131 |
{ |
132 |
struct kvm_book3e_206_tlb_params params = {};
|
133 |
struct kvm_config_tlb cfg = {};
|
134 |
struct kvm_enable_cap encap = {};
|
135 |
unsigned int entries = 0; |
136 |
int ret, i;
|
137 |
|
138 |
if (!kvm_enabled() ||
|
139 |
!kvm_check_extension(env->kvm_state, KVM_CAP_SW_TLB)) { |
140 |
return 0; |
141 |
} |
142 |
|
143 |
assert(ARRAY_SIZE(params.tlb_sizes) == BOOKE206_MAX_TLBN); |
144 |
|
145 |
for (i = 0; i < BOOKE206_MAX_TLBN; i++) { |
146 |
params.tlb_sizes[i] = booke206_tlb_size(env, i); |
147 |
params.tlb_ways[i] = booke206_tlb_ways(env, i); |
148 |
entries += params.tlb_sizes[i]; |
149 |
} |
150 |
|
151 |
assert(entries == env->nb_tlb); |
152 |
assert(sizeof(struct kvm_book3e_206_tlb_entry) == sizeof(ppcmas_tlb_t)); |
153 |
|
154 |
env->tlb_dirty = true;
|
155 |
|
156 |
cfg.array = (uintptr_t)env->tlb.tlbm; |
157 |
cfg.array_len = sizeof(ppcmas_tlb_t) * entries;
|
158 |
cfg.params = (uintptr_t)¶ms; |
159 |
cfg.mmu_type = KVM_MMU_FSL_BOOKE_NOHV; |
160 |
|
161 |
encap.cap = KVM_CAP_SW_TLB; |
162 |
encap.args[0] = (uintptr_t)&cfg;
|
163 |
|
164 |
ret = kvm_vcpu_ioctl(env, KVM_ENABLE_CAP, &encap); |
165 |
if (ret < 0) { |
166 |
fprintf(stderr, "%s: couldn't enable KVM_CAP_SW_TLB: %s\n",
|
167 |
__func__, strerror(-ret)); |
168 |
return ret;
|
169 |
} |
170 |
|
171 |
env->kvm_sw_tlb = true;
|
172 |
return 0; |
173 |
} |
174 |
|
175 |
|
176 |
#if defined(TARGET_PPC64)
|
177 |
static void kvm_get_fallback_smmu_info(CPUPPCState *env, |
178 |
struct kvm_ppc_smmu_info *info)
|
179 |
{ |
180 |
memset(info, 0, sizeof(*info)); |
181 |
|
182 |
/* We don't have the new KVM_PPC_GET_SMMU_INFO ioctl, so
|
183 |
* need to "guess" what the supported page sizes are.
|
184 |
*
|
185 |
* For that to work we make a few assumptions:
|
186 |
*
|
187 |
* - If KVM_CAP_PPC_GET_PVINFO is supported we are running "PR"
|
188 |
* KVM which only supports 4K and 16M pages, but supports them
|
189 |
* regardless of the backing store characteritics. We also don't
|
190 |
* support 1T segments.
|
191 |
*
|
192 |
* This is safe as if HV KVM ever supports that capability or PR
|
193 |
* KVM grows supports for more page/segment sizes, those versions
|
194 |
* will have implemented KVM_CAP_PPC_GET_SMMU_INFO and thus we
|
195 |
* will not hit this fallback
|
196 |
*
|
197 |
* - Else we are running HV KVM. This means we only support page
|
198 |
* sizes that fit in the backing store. Additionally we only
|
199 |
* advertize 64K pages if the processor is ARCH 2.06 and we assume
|
200 |
* P7 encodings for the SLB and hash table. Here too, we assume
|
201 |
* support for any newer processor will mean a kernel that
|
202 |
* implements KVM_CAP_PPC_GET_SMMU_INFO and thus doesn't hit
|
203 |
* this fallback.
|
204 |
*/
|
205 |
if (kvm_check_extension(env->kvm_state, KVM_CAP_PPC_GET_PVINFO)) {
|
206 |
/* No flags */
|
207 |
info->flags = 0;
|
208 |
info->slb_size = 64;
|
209 |
|
210 |
/* Standard 4k base page size segment */
|
211 |
info->sps[0].page_shift = 12; |
212 |
info->sps[0].slb_enc = 0; |
213 |
info->sps[0].enc[0].page_shift = 12; |
214 |
info->sps[0].enc[0].pte_enc = 0; |
215 |
|
216 |
/* Standard 16M large page size segment */
|
217 |
info->sps[1].page_shift = 24; |
218 |
info->sps[1].slb_enc = SLB_VSID_L;
|
219 |
info->sps[1].enc[0].page_shift = 24; |
220 |
info->sps[1].enc[0].pte_enc = 0; |
221 |
} else {
|
222 |
int i = 0; |
223 |
|
224 |
/* HV KVM has backing store size restrictions */
|
225 |
info->flags = KVM_PPC_PAGE_SIZES_REAL; |
226 |
|
227 |
if (env->mmu_model & POWERPC_MMU_1TSEG) {
|
228 |
info->flags |= KVM_PPC_1T_SEGMENTS; |
229 |
} |
230 |
|
231 |
if (env->mmu_model == POWERPC_MMU_2_06) {
|
232 |
info->slb_size = 32;
|
233 |
} else {
|
234 |
info->slb_size = 64;
|
235 |
} |
236 |
|
237 |
/* Standard 4k base page size segment */
|
238 |
info->sps[i].page_shift = 12;
|
239 |
info->sps[i].slb_enc = 0;
|
240 |
info->sps[i].enc[0].page_shift = 12; |
241 |
info->sps[i].enc[0].pte_enc = 0; |
242 |
i++; |
243 |
|
244 |
/* 64K on MMU 2.06 */
|
245 |
if (env->mmu_model == POWERPC_MMU_2_06) {
|
246 |
info->sps[i].page_shift = 16;
|
247 |
info->sps[i].slb_enc = 0x110;
|
248 |
info->sps[i].enc[0].page_shift = 16; |
249 |
info->sps[i].enc[0].pte_enc = 1; |
250 |
i++; |
251 |
} |
252 |
|
253 |
/* Standard 16M large page size segment */
|
254 |
info->sps[i].page_shift = 24;
|
255 |
info->sps[i].slb_enc = SLB_VSID_L; |
256 |
info->sps[i].enc[0].page_shift = 24; |
257 |
info->sps[i].enc[0].pte_enc = 0; |
258 |
} |
259 |
} |
260 |
|
261 |
static void kvm_get_smmu_info(CPUPPCState *env, struct kvm_ppc_smmu_info *info) |
262 |
{ |
263 |
int ret;
|
264 |
|
265 |
if (kvm_check_extension(env->kvm_state, KVM_CAP_PPC_GET_SMMU_INFO)) {
|
266 |
ret = kvm_vm_ioctl(env->kvm_state, KVM_PPC_GET_SMMU_INFO, info); |
267 |
if (ret == 0) { |
268 |
return;
|
269 |
} |
270 |
} |
271 |
|
272 |
kvm_get_fallback_smmu_info(env, info); |
273 |
} |
274 |
|
275 |
static long getrampagesize(void) |
276 |
{ |
277 |
struct statfs fs;
|
278 |
int ret;
|
279 |
|
280 |
if (!mem_path) {
|
281 |
/* guest RAM is backed by normal anonymous pages */
|
282 |
return getpagesize();
|
283 |
} |
284 |
|
285 |
do {
|
286 |
ret = statfs(mem_path, &fs); |
287 |
} while (ret != 0 && errno == EINTR); |
288 |
|
289 |
if (ret != 0) { |
290 |
fprintf(stderr, "Couldn't statfs() memory path: %s\n",
|
291 |
strerror(errno)); |
292 |
exit(1);
|
293 |
} |
294 |
|
295 |
#define HUGETLBFS_MAGIC 0x958458f6 |
296 |
|
297 |
if (fs.f_type != HUGETLBFS_MAGIC) {
|
298 |
/* Explicit mempath, but it's ordinary pages */
|
299 |
return getpagesize();
|
300 |
} |
301 |
|
302 |
/* It's hugepage, return the huge page size */
|
303 |
return fs.f_bsize;
|
304 |
} |
305 |
|
306 |
static bool kvm_valid_page_size(uint32_t flags, long rampgsize, uint32_t shift) |
307 |
{ |
308 |
if (!(flags & KVM_PPC_PAGE_SIZES_REAL)) {
|
309 |
return true; |
310 |
} |
311 |
|
312 |
return (1ul << shift) <= rampgsize; |
313 |
} |
314 |
|
315 |
static void kvm_fixup_page_sizes(CPUPPCState *env) |
316 |
{ |
317 |
static struct kvm_ppc_smmu_info smmu_info; |
318 |
static bool has_smmu_info; |
319 |
long rampagesize;
|
320 |
int iq, ik, jq, jk;
|
321 |
|
322 |
/* We only handle page sizes for 64-bit server guests for now */
|
323 |
if (!(env->mmu_model & POWERPC_MMU_64)) {
|
324 |
return;
|
325 |
} |
326 |
|
327 |
/* Collect MMU info from kernel if not already */
|
328 |
if (!has_smmu_info) {
|
329 |
kvm_get_smmu_info(env, &smmu_info); |
330 |
has_smmu_info = true;
|
331 |
} |
332 |
|
333 |
rampagesize = getrampagesize(); |
334 |
|
335 |
/* Convert to QEMU form */
|
336 |
memset(&env->sps, 0, sizeof(env->sps)); |
337 |
|
338 |
for (ik = iq = 0; ik < KVM_PPC_PAGE_SIZES_MAX_SZ; ik++) { |
339 |
struct ppc_one_seg_page_size *qsps = &env->sps.sps[iq];
|
340 |
struct kvm_ppc_one_seg_page_size *ksps = &smmu_info.sps[ik];
|
341 |
|
342 |
if (!kvm_valid_page_size(smmu_info.flags, rampagesize,
|
343 |
ksps->page_shift)) { |
344 |
continue;
|
345 |
} |
346 |
qsps->page_shift = ksps->page_shift; |
347 |
qsps->slb_enc = ksps->slb_enc; |
348 |
for (jk = jq = 0; jk < KVM_PPC_PAGE_SIZES_MAX_SZ; jk++) { |
349 |
if (!kvm_valid_page_size(smmu_info.flags, rampagesize,
|
350 |
ksps->enc[jk].page_shift)) { |
351 |
continue;
|
352 |
} |
353 |
qsps->enc[jq].page_shift = ksps->enc[jk].page_shift; |
354 |
qsps->enc[jq].pte_enc = ksps->enc[jk].pte_enc; |
355 |
if (++jq >= PPC_PAGE_SIZES_MAX_SZ) {
|
356 |
break;
|
357 |
} |
358 |
} |
359 |
if (++iq >= PPC_PAGE_SIZES_MAX_SZ) {
|
360 |
break;
|
361 |
} |
362 |
} |
363 |
env->slb_nr = smmu_info.slb_size; |
364 |
if (smmu_info.flags & KVM_PPC_1T_SEGMENTS) {
|
365 |
env->mmu_model |= POWERPC_MMU_1TSEG; |
366 |
} else {
|
367 |
env->mmu_model &= ~POWERPC_MMU_1TSEG; |
368 |
} |
369 |
} |
370 |
#else /* defined (TARGET_PPC64) */ |
371 |
|
372 |
static inline void kvm_fixup_page_sizes(CPUPPCState *env) |
373 |
{ |
374 |
} |
375 |
|
376 |
#endif /* !defined (TARGET_PPC64) */ |
377 |
|
378 |
int kvm_arch_init_vcpu(CPUPPCState *cenv)
|
379 |
{ |
380 |
PowerPCCPU *cpu = ppc_env_get_cpu(cenv); |
381 |
int ret;
|
382 |
|
383 |
/* Gather server mmu info from KVM and update the CPU state */
|
384 |
kvm_fixup_page_sizes(cenv); |
385 |
|
386 |
/* Synchronize sregs with kvm */
|
387 |
ret = kvm_arch_sync_sregs(cenv); |
388 |
if (ret) {
|
389 |
return ret;
|
390 |
} |
391 |
|
392 |
idle_timer = qemu_new_timer_ns(vm_clock, kvm_kick_cpu, cpu); |
393 |
|
394 |
/* Some targets support access to KVM's guest TLB. */
|
395 |
switch (cenv->mmu_model) {
|
396 |
case POWERPC_MMU_BOOKE206:
|
397 |
ret = kvm_booke206_tlb_init(cenv); |
398 |
break;
|
399 |
default:
|
400 |
break;
|
401 |
} |
402 |
|
403 |
return ret;
|
404 |
} |
405 |
|
406 |
void kvm_arch_reset_vcpu(CPUPPCState *env)
|
407 |
{ |
408 |
} |
409 |
|
410 |
static void kvm_sw_tlb_put(CPUPPCState *env) |
411 |
{ |
412 |
struct kvm_dirty_tlb dirty_tlb;
|
413 |
unsigned char *bitmap; |
414 |
int ret;
|
415 |
|
416 |
if (!env->kvm_sw_tlb) {
|
417 |
return;
|
418 |
} |
419 |
|
420 |
bitmap = g_malloc((env->nb_tlb + 7) / 8); |
421 |
memset(bitmap, 0xFF, (env->nb_tlb + 7) / 8); |
422 |
|
423 |
dirty_tlb.bitmap = (uintptr_t)bitmap; |
424 |
dirty_tlb.num_dirty = env->nb_tlb; |
425 |
|
426 |
ret = kvm_vcpu_ioctl(env, KVM_DIRTY_TLB, &dirty_tlb); |
427 |
if (ret) {
|
428 |
fprintf(stderr, "%s: KVM_DIRTY_TLB: %s\n",
|
429 |
__func__, strerror(-ret)); |
430 |
} |
431 |
|
432 |
g_free(bitmap); |
433 |
} |
434 |
|
435 |
int kvm_arch_put_registers(CPUPPCState *env, int level) |
436 |
{ |
437 |
struct kvm_regs regs;
|
438 |
int ret;
|
439 |
int i;
|
440 |
|
441 |
ret = kvm_vcpu_ioctl(env, KVM_GET_REGS, ®s); |
442 |
if (ret < 0) |
443 |
return ret;
|
444 |
|
445 |
regs.ctr = env->ctr; |
446 |
regs.lr = env->lr; |
447 |
regs.xer = env->xer; |
448 |
regs.msr = env->msr; |
449 |
regs.pc = env->nip; |
450 |
|
451 |
regs.srr0 = env->spr[SPR_SRR0]; |
452 |
regs.srr1 = env->spr[SPR_SRR1]; |
453 |
|
454 |
regs.sprg0 = env->spr[SPR_SPRG0]; |
455 |
regs.sprg1 = env->spr[SPR_SPRG1]; |
456 |
regs.sprg2 = env->spr[SPR_SPRG2]; |
457 |
regs.sprg3 = env->spr[SPR_SPRG3]; |
458 |
regs.sprg4 = env->spr[SPR_SPRG4]; |
459 |
regs.sprg5 = env->spr[SPR_SPRG5]; |
460 |
regs.sprg6 = env->spr[SPR_SPRG6]; |
461 |
regs.sprg7 = env->spr[SPR_SPRG7]; |
462 |
|
463 |
regs.pid = env->spr[SPR_BOOKE_PID]; |
464 |
|
465 |
for (i = 0;i < 32; i++) |
466 |
regs.gpr[i] = env->gpr[i]; |
467 |
|
468 |
ret = kvm_vcpu_ioctl(env, KVM_SET_REGS, ®s); |
469 |
if (ret < 0) |
470 |
return ret;
|
471 |
|
472 |
if (env->tlb_dirty) {
|
473 |
kvm_sw_tlb_put(env); |
474 |
env->tlb_dirty = false;
|
475 |
} |
476 |
|
477 |
if (cap_segstate && (level >= KVM_PUT_RESET_STATE)) {
|
478 |
struct kvm_sregs sregs;
|
479 |
|
480 |
sregs.pvr = env->spr[SPR_PVR]; |
481 |
|
482 |
sregs.u.s.sdr1 = env->spr[SPR_SDR1]; |
483 |
|
484 |
/* Sync SLB */
|
485 |
#ifdef TARGET_PPC64
|
486 |
for (i = 0; i < 64; i++) { |
487 |
sregs.u.s.ppc64.slb[i].slbe = env->slb[i].esid; |
488 |
sregs.u.s.ppc64.slb[i].slbv = env->slb[i].vsid; |
489 |
} |
490 |
#endif
|
491 |
|
492 |
/* Sync SRs */
|
493 |
for (i = 0; i < 16; i++) { |
494 |
sregs.u.s.ppc32.sr[i] = env->sr[i]; |
495 |
} |
496 |
|
497 |
/* Sync BATs */
|
498 |
for (i = 0; i < 8; i++) { |
499 |
/* Beware. We have to swap upper and lower bits here */
|
500 |
sregs.u.s.ppc32.dbat[i] = ((uint64_t)env->DBAT[0][i] << 32) |
501 |
| env->DBAT[1][i];
|
502 |
sregs.u.s.ppc32.ibat[i] = ((uint64_t)env->IBAT[0][i] << 32) |
503 |
| env->IBAT[1][i];
|
504 |
} |
505 |
|
506 |
ret = kvm_vcpu_ioctl(env, KVM_SET_SREGS, &sregs); |
507 |
if (ret) {
|
508 |
return ret;
|
509 |
} |
510 |
} |
511 |
|
512 |
if (cap_hior && (level >= KVM_PUT_RESET_STATE)) {
|
513 |
uint64_t hior = env->spr[SPR_HIOR]; |
514 |
struct kvm_one_reg reg = {
|
515 |
.id = KVM_REG_PPC_HIOR, |
516 |
.addr = (uintptr_t) &hior, |
517 |
}; |
518 |
|
519 |
ret = kvm_vcpu_ioctl(env, KVM_SET_ONE_REG, ®); |
520 |
if (ret) {
|
521 |
return ret;
|
522 |
} |
523 |
} |
524 |
|
525 |
return ret;
|
526 |
} |
527 |
|
528 |
int kvm_arch_get_registers(CPUPPCState *env)
|
529 |
{ |
530 |
struct kvm_regs regs;
|
531 |
struct kvm_sregs sregs;
|
532 |
uint32_t cr; |
533 |
int i, ret;
|
534 |
|
535 |
ret = kvm_vcpu_ioctl(env, KVM_GET_REGS, ®s); |
536 |
if (ret < 0) |
537 |
return ret;
|
538 |
|
539 |
cr = regs.cr; |
540 |
for (i = 7; i >= 0; i--) { |
541 |
env->crf[i] = cr & 15;
|
542 |
cr >>= 4;
|
543 |
} |
544 |
|
545 |
env->ctr = regs.ctr; |
546 |
env->lr = regs.lr; |
547 |
env->xer = regs.xer; |
548 |
env->msr = regs.msr; |
549 |
env->nip = regs.pc; |
550 |
|
551 |
env->spr[SPR_SRR0] = regs.srr0; |
552 |
env->spr[SPR_SRR1] = regs.srr1; |
553 |
|
554 |
env->spr[SPR_SPRG0] = regs.sprg0; |
555 |
env->spr[SPR_SPRG1] = regs.sprg1; |
556 |
env->spr[SPR_SPRG2] = regs.sprg2; |
557 |
env->spr[SPR_SPRG3] = regs.sprg3; |
558 |
env->spr[SPR_SPRG4] = regs.sprg4; |
559 |
env->spr[SPR_SPRG5] = regs.sprg5; |
560 |
env->spr[SPR_SPRG6] = regs.sprg6; |
561 |
env->spr[SPR_SPRG7] = regs.sprg7; |
562 |
|
563 |
env->spr[SPR_BOOKE_PID] = regs.pid; |
564 |
|
565 |
for (i = 0;i < 32; i++) |
566 |
env->gpr[i] = regs.gpr[i]; |
567 |
|
568 |
if (cap_booke_sregs) {
|
569 |
ret = kvm_vcpu_ioctl(env, KVM_GET_SREGS, &sregs); |
570 |
if (ret < 0) { |
571 |
return ret;
|
572 |
} |
573 |
|
574 |
if (sregs.u.e.features & KVM_SREGS_E_BASE) {
|
575 |
env->spr[SPR_BOOKE_CSRR0] = sregs.u.e.csrr0; |
576 |
env->spr[SPR_BOOKE_CSRR1] = sregs.u.e.csrr1; |
577 |
env->spr[SPR_BOOKE_ESR] = sregs.u.e.esr; |
578 |
env->spr[SPR_BOOKE_DEAR] = sregs.u.e.dear; |
579 |
env->spr[SPR_BOOKE_MCSR] = sregs.u.e.mcsr; |
580 |
env->spr[SPR_BOOKE_TSR] = sregs.u.e.tsr; |
581 |
env->spr[SPR_BOOKE_TCR] = sregs.u.e.tcr; |
582 |
env->spr[SPR_DECR] = sregs.u.e.dec; |
583 |
env->spr[SPR_TBL] = sregs.u.e.tb & 0xffffffff;
|
584 |
env->spr[SPR_TBU] = sregs.u.e.tb >> 32;
|
585 |
env->spr[SPR_VRSAVE] = sregs.u.e.vrsave; |
586 |
} |
587 |
|
588 |
if (sregs.u.e.features & KVM_SREGS_E_ARCH206) {
|
589 |
env->spr[SPR_BOOKE_PIR] = sregs.u.e.pir; |
590 |
env->spr[SPR_BOOKE_MCSRR0] = sregs.u.e.mcsrr0; |
591 |
env->spr[SPR_BOOKE_MCSRR1] = sregs.u.e.mcsrr1; |
592 |
env->spr[SPR_BOOKE_DECAR] = sregs.u.e.decar; |
593 |
env->spr[SPR_BOOKE_IVPR] = sregs.u.e.ivpr; |
594 |
} |
595 |
|
596 |
if (sregs.u.e.features & KVM_SREGS_E_64) {
|
597 |
env->spr[SPR_BOOKE_EPCR] = sregs.u.e.epcr; |
598 |
} |
599 |
|
600 |
if (sregs.u.e.features & KVM_SREGS_E_SPRG8) {
|
601 |
env->spr[SPR_BOOKE_SPRG8] = sregs.u.e.sprg8; |
602 |
} |
603 |
|
604 |
if (sregs.u.e.features & KVM_SREGS_E_IVOR) {
|
605 |
env->spr[SPR_BOOKE_IVOR0] = sregs.u.e.ivor_low[0];
|
606 |
env->spr[SPR_BOOKE_IVOR1] = sregs.u.e.ivor_low[1];
|
607 |
env->spr[SPR_BOOKE_IVOR2] = sregs.u.e.ivor_low[2];
|
608 |
env->spr[SPR_BOOKE_IVOR3] = sregs.u.e.ivor_low[3];
|
609 |
env->spr[SPR_BOOKE_IVOR4] = sregs.u.e.ivor_low[4];
|
610 |
env->spr[SPR_BOOKE_IVOR5] = sregs.u.e.ivor_low[5];
|
611 |
env->spr[SPR_BOOKE_IVOR6] = sregs.u.e.ivor_low[6];
|
612 |
env->spr[SPR_BOOKE_IVOR7] = sregs.u.e.ivor_low[7];
|
613 |
env->spr[SPR_BOOKE_IVOR8] = sregs.u.e.ivor_low[8];
|
614 |
env->spr[SPR_BOOKE_IVOR9] = sregs.u.e.ivor_low[9];
|
615 |
env->spr[SPR_BOOKE_IVOR10] = sregs.u.e.ivor_low[10];
|
616 |
env->spr[SPR_BOOKE_IVOR11] = sregs.u.e.ivor_low[11];
|
617 |
env->spr[SPR_BOOKE_IVOR12] = sregs.u.e.ivor_low[12];
|
618 |
env->spr[SPR_BOOKE_IVOR13] = sregs.u.e.ivor_low[13];
|
619 |
env->spr[SPR_BOOKE_IVOR14] = sregs.u.e.ivor_low[14];
|
620 |
env->spr[SPR_BOOKE_IVOR15] = sregs.u.e.ivor_low[15];
|
621 |
|
622 |
if (sregs.u.e.features & KVM_SREGS_E_SPE) {
|
623 |
env->spr[SPR_BOOKE_IVOR32] = sregs.u.e.ivor_high[0];
|
624 |
env->spr[SPR_BOOKE_IVOR33] = sregs.u.e.ivor_high[1];
|
625 |
env->spr[SPR_BOOKE_IVOR34] = sregs.u.e.ivor_high[2];
|
626 |
} |
627 |
|
628 |
if (sregs.u.e.features & KVM_SREGS_E_PM) {
|
629 |
env->spr[SPR_BOOKE_IVOR35] = sregs.u.e.ivor_high[3];
|
630 |
} |
631 |
|
632 |
if (sregs.u.e.features & KVM_SREGS_E_PC) {
|
633 |
env->spr[SPR_BOOKE_IVOR36] = sregs.u.e.ivor_high[4];
|
634 |
env->spr[SPR_BOOKE_IVOR37] = sregs.u.e.ivor_high[5];
|
635 |
} |
636 |
} |
637 |
|
638 |
if (sregs.u.e.features & KVM_SREGS_E_ARCH206_MMU) {
|
639 |
env->spr[SPR_BOOKE_MAS0] = sregs.u.e.mas0; |
640 |
env->spr[SPR_BOOKE_MAS1] = sregs.u.e.mas1; |
641 |
env->spr[SPR_BOOKE_MAS2] = sregs.u.e.mas2; |
642 |
env->spr[SPR_BOOKE_MAS3] = sregs.u.e.mas7_3 & 0xffffffff;
|
643 |
env->spr[SPR_BOOKE_MAS4] = sregs.u.e.mas4; |
644 |
env->spr[SPR_BOOKE_MAS6] = sregs.u.e.mas6; |
645 |
env->spr[SPR_BOOKE_MAS7] = sregs.u.e.mas7_3 >> 32;
|
646 |
env->spr[SPR_MMUCFG] = sregs.u.e.mmucfg; |
647 |
env->spr[SPR_BOOKE_TLB0CFG] = sregs.u.e.tlbcfg[0];
|
648 |
env->spr[SPR_BOOKE_TLB1CFG] = sregs.u.e.tlbcfg[1];
|
649 |
} |
650 |
|
651 |
if (sregs.u.e.features & KVM_SREGS_EXP) {
|
652 |
env->spr[SPR_BOOKE_EPR] = sregs.u.e.epr; |
653 |
} |
654 |
|
655 |
if (sregs.u.e.features & KVM_SREGS_E_PD) {
|
656 |
env->spr[SPR_BOOKE_EPLC] = sregs.u.e.eplc; |
657 |
env->spr[SPR_BOOKE_EPSC] = sregs.u.e.epsc; |
658 |
} |
659 |
|
660 |
if (sregs.u.e.impl_id == KVM_SREGS_E_IMPL_FSL) {
|
661 |
env->spr[SPR_E500_SVR] = sregs.u.e.impl.fsl.svr; |
662 |
env->spr[SPR_Exxx_MCAR] = sregs.u.e.impl.fsl.mcar; |
663 |
env->spr[SPR_HID0] = sregs.u.e.impl.fsl.hid0; |
664 |
|
665 |
if (sregs.u.e.impl.fsl.features & KVM_SREGS_E_FSL_PIDn) {
|
666 |
env->spr[SPR_BOOKE_PID1] = sregs.u.e.impl.fsl.pid1; |
667 |
env->spr[SPR_BOOKE_PID2] = sregs.u.e.impl.fsl.pid2; |
668 |
} |
669 |
} |
670 |
} |
671 |
|
672 |
if (cap_segstate) {
|
673 |
ret = kvm_vcpu_ioctl(env, KVM_GET_SREGS, &sregs); |
674 |
if (ret < 0) { |
675 |
return ret;
|
676 |
} |
677 |
|
678 |
ppc_store_sdr1(env, sregs.u.s.sdr1); |
679 |
|
680 |
/* Sync SLB */
|
681 |
#ifdef TARGET_PPC64
|
682 |
for (i = 0; i < 64; i++) { |
683 |
ppc_store_slb(env, sregs.u.s.ppc64.slb[i].slbe, |
684 |
sregs.u.s.ppc64.slb[i].slbv); |
685 |
} |
686 |
#endif
|
687 |
|
688 |
/* Sync SRs */
|
689 |
for (i = 0; i < 16; i++) { |
690 |
env->sr[i] = sregs.u.s.ppc32.sr[i]; |
691 |
} |
692 |
|
693 |
/* Sync BATs */
|
694 |
for (i = 0; i < 8; i++) { |
695 |
env->DBAT[0][i] = sregs.u.s.ppc32.dbat[i] & 0xffffffff; |
696 |
env->DBAT[1][i] = sregs.u.s.ppc32.dbat[i] >> 32; |
697 |
env->IBAT[0][i] = sregs.u.s.ppc32.ibat[i] & 0xffffffff; |
698 |
env->IBAT[1][i] = sregs.u.s.ppc32.ibat[i] >> 32; |
699 |
} |
700 |
} |
701 |
|
702 |
return 0; |
703 |
} |
704 |
|
705 |
int kvmppc_set_interrupt(CPUPPCState *env, int irq, int level) |
706 |
{ |
707 |
unsigned virq = level ? KVM_INTERRUPT_SET_LEVEL : KVM_INTERRUPT_UNSET;
|
708 |
|
709 |
if (irq != PPC_INTERRUPT_EXT) {
|
710 |
return 0; |
711 |
} |
712 |
|
713 |
if (!kvm_enabled() || !cap_interrupt_unset || !cap_interrupt_level) {
|
714 |
return 0; |
715 |
} |
716 |
|
717 |
kvm_vcpu_ioctl(env, KVM_INTERRUPT, &virq); |
718 |
|
719 |
return 0; |
720 |
} |
721 |
|
722 |
#if defined(TARGET_PPCEMB)
|
723 |
#define PPC_INPUT_INT PPC40x_INPUT_INT
|
724 |
#elif defined(TARGET_PPC64)
|
725 |
#define PPC_INPUT_INT PPC970_INPUT_INT
|
726 |
#else
|
727 |
#define PPC_INPUT_INT PPC6xx_INPUT_INT
|
728 |
#endif
|
729 |
|
730 |
void kvm_arch_pre_run(CPUPPCState *env, struct kvm_run *run) |
731 |
{ |
732 |
int r;
|
733 |
unsigned irq;
|
734 |
|
735 |
/* PowerPC QEMU tracks the various core input pins (interrupt, critical
|
736 |
* interrupt, reset, etc) in PPC-specific env->irq_input_state. */
|
737 |
if (!cap_interrupt_level &&
|
738 |
run->ready_for_interrupt_injection && |
739 |
(env->interrupt_request & CPU_INTERRUPT_HARD) && |
740 |
(env->irq_input_state & (1<<PPC_INPUT_INT)))
|
741 |
{ |
742 |
/* For now KVM disregards the 'irq' argument. However, in the
|
743 |
* future KVM could cache it in-kernel to avoid a heavyweight exit
|
744 |
* when reading the UIC.
|
745 |
*/
|
746 |
irq = KVM_INTERRUPT_SET; |
747 |
|
748 |
dprintf("injected interrupt %d\n", irq);
|
749 |
r = kvm_vcpu_ioctl(env, KVM_INTERRUPT, &irq); |
750 |
if (r < 0) |
751 |
printf("cpu %d fail inject %x\n", env->cpu_index, irq);
|
752 |
|
753 |
/* Always wake up soon in case the interrupt was level based */
|
754 |
qemu_mod_timer(idle_timer, qemu_get_clock_ns(vm_clock) + |
755 |
(get_ticks_per_sec() / 50));
|
756 |
} |
757 |
|
758 |
/* We don't know if there are more interrupts pending after this. However,
|
759 |
* the guest will return to userspace in the course of handling this one
|
760 |
* anyways, so we will get a chance to deliver the rest. */
|
761 |
} |
762 |
|
763 |
void kvm_arch_post_run(CPUPPCState *env, struct kvm_run *run) |
764 |
{ |
765 |
} |
766 |
|
767 |
int kvm_arch_process_async_events(CPUPPCState *env)
|
768 |
{ |
769 |
return env->halted;
|
770 |
} |
771 |
|
772 |
static int kvmppc_handle_halt(CPUPPCState *env) |
773 |
{ |
774 |
if (!(env->interrupt_request & CPU_INTERRUPT_HARD) && (msr_ee)) {
|
775 |
env->halted = 1;
|
776 |
env->exception_index = EXCP_HLT; |
777 |
} |
778 |
|
779 |
return 0; |
780 |
} |
781 |
|
782 |
/* map dcr access to existing qemu dcr emulation */
|
783 |
static int kvmppc_handle_dcr_read(CPUPPCState *env, uint32_t dcrn, uint32_t *data) |
784 |
{ |
785 |
if (ppc_dcr_read(env->dcr_env, dcrn, data) < 0) |
786 |
fprintf(stderr, "Read to unhandled DCR (0x%x)\n", dcrn);
|
787 |
|
788 |
return 0; |
789 |
} |
790 |
|
791 |
static int kvmppc_handle_dcr_write(CPUPPCState *env, uint32_t dcrn, uint32_t data) |
792 |
{ |
793 |
if (ppc_dcr_write(env->dcr_env, dcrn, data) < 0) |
794 |
fprintf(stderr, "Write to unhandled DCR (0x%x)\n", dcrn);
|
795 |
|
796 |
return 0; |
797 |
} |
798 |
|
799 |
int kvm_arch_handle_exit(CPUPPCState *env, struct kvm_run *run) |
800 |
{ |
801 |
int ret;
|
802 |
|
803 |
switch (run->exit_reason) {
|
804 |
case KVM_EXIT_DCR:
|
805 |
if (run->dcr.is_write) {
|
806 |
dprintf("handle dcr write\n");
|
807 |
ret = kvmppc_handle_dcr_write(env, run->dcr.dcrn, run->dcr.data); |
808 |
} else {
|
809 |
dprintf("handle dcr read\n");
|
810 |
ret = kvmppc_handle_dcr_read(env, run->dcr.dcrn, &run->dcr.data); |
811 |
} |
812 |
break;
|
813 |
case KVM_EXIT_HLT:
|
814 |
dprintf("handle halt\n");
|
815 |
ret = kvmppc_handle_halt(env); |
816 |
break;
|
817 |
#ifdef CONFIG_PSERIES
|
818 |
case KVM_EXIT_PAPR_HCALL:
|
819 |
dprintf("handle PAPR hypercall\n");
|
820 |
run->papr_hcall.ret = spapr_hypercall(ppc_env_get_cpu(env), |
821 |
run->papr_hcall.nr, |
822 |
run->papr_hcall.args); |
823 |
ret = 0;
|
824 |
break;
|
825 |
#endif
|
826 |
default:
|
827 |
fprintf(stderr, "KVM: unknown exit reason %d\n", run->exit_reason);
|
828 |
ret = -1;
|
829 |
break;
|
830 |
} |
831 |
|
832 |
return ret;
|
833 |
} |
834 |
|
835 |
static int read_cpuinfo(const char *field, char *value, int len) |
836 |
{ |
837 |
FILE *f; |
838 |
int ret = -1; |
839 |
int field_len = strlen(field);
|
840 |
char line[512]; |
841 |
|
842 |
f = fopen("/proc/cpuinfo", "r"); |
843 |
if (!f) {
|
844 |
return -1; |
845 |
} |
846 |
|
847 |
do {
|
848 |
if(!fgets(line, sizeof(line), f)) { |
849 |
break;
|
850 |
} |
851 |
if (!strncmp(line, field, field_len)) {
|
852 |
pstrcpy(value, len, line); |
853 |
ret = 0;
|
854 |
break;
|
855 |
} |
856 |
} while(*line);
|
857 |
|
858 |
fclose(f); |
859 |
|
860 |
return ret;
|
861 |
} |
862 |
|
863 |
uint32_t kvmppc_get_tbfreq(void)
|
864 |
{ |
865 |
char line[512]; |
866 |
char *ns;
|
867 |
uint32_t retval = get_ticks_per_sec(); |
868 |
|
869 |
if (read_cpuinfo("timebase", line, sizeof(line))) { |
870 |
return retval;
|
871 |
} |
872 |
|
873 |
if (!(ns = strchr(line, ':'))) { |
874 |
return retval;
|
875 |
} |
876 |
|
877 |
ns++; |
878 |
|
879 |
retval = atoi(ns); |
880 |
return retval;
|
881 |
} |
882 |
|
883 |
/* Try to find a device tree node for a CPU with clock-frequency property */
|
884 |
static int kvmppc_find_cpu_dt(char *buf, int buf_len) |
885 |
{ |
886 |
struct dirent *dirp;
|
887 |
DIR *dp; |
888 |
|
889 |
if ((dp = opendir(PROC_DEVTREE_CPU)) == NULL) { |
890 |
printf("Can't open directory " PROC_DEVTREE_CPU "\n"); |
891 |
return -1; |
892 |
} |
893 |
|
894 |
buf[0] = '\0'; |
895 |
while ((dirp = readdir(dp)) != NULL) { |
896 |
FILE *f; |
897 |
snprintf(buf, buf_len, "%s%s/clock-frequency", PROC_DEVTREE_CPU,
|
898 |
dirp->d_name); |
899 |
f = fopen(buf, "r");
|
900 |
if (f) {
|
901 |
snprintf(buf, buf_len, "%s%s", PROC_DEVTREE_CPU, dirp->d_name);
|
902 |
fclose(f); |
903 |
break;
|
904 |
} |
905 |
buf[0] = '\0'; |
906 |
} |
907 |
closedir(dp); |
908 |
if (buf[0] == '\0') { |
909 |
printf("Unknown host!\n");
|
910 |
return -1; |
911 |
} |
912 |
|
913 |
return 0; |
914 |
} |
915 |
|
916 |
/* Read a CPU node property from the host device tree that's a single
|
917 |
* integer (32-bit or 64-bit). Returns 0 if anything goes wrong
|
918 |
* (can't find or open the property, or doesn't understand the
|
919 |
* format) */
|
920 |
static uint64_t kvmppc_read_int_cpu_dt(const char *propname) |
921 |
{ |
922 |
char buf[PATH_MAX];
|
923 |
union {
|
924 |
uint32_t v32; |
925 |
uint64_t v64; |
926 |
} u; |
927 |
FILE *f; |
928 |
int len;
|
929 |
|
930 |
if (kvmppc_find_cpu_dt(buf, sizeof(buf))) { |
931 |
return -1; |
932 |
} |
933 |
|
934 |
strncat(buf, "/", sizeof(buf) - strlen(buf)); |
935 |
strncat(buf, propname, sizeof(buf) - strlen(buf));
|
936 |
|
937 |
f = fopen(buf, "rb");
|
938 |
if (!f) {
|
939 |
return -1; |
940 |
} |
941 |
|
942 |
len = fread(&u, 1, sizeof(u), f); |
943 |
fclose(f); |
944 |
switch (len) {
|
945 |
case 4: |
946 |
/* property is a 32-bit quantity */
|
947 |
return be32_to_cpu(u.v32);
|
948 |
case 8: |
949 |
return be64_to_cpu(u.v64);
|
950 |
} |
951 |
|
952 |
return 0; |
953 |
} |
954 |
|
955 |
uint64_t kvmppc_get_clockfreq(void)
|
956 |
{ |
957 |
return kvmppc_read_int_cpu_dt("clock-frequency"); |
958 |
} |
959 |
|
960 |
uint32_t kvmppc_get_vmx(void)
|
961 |
{ |
962 |
return kvmppc_read_int_cpu_dt("ibm,vmx"); |
963 |
} |
964 |
|
965 |
uint32_t kvmppc_get_dfp(void)
|
966 |
{ |
967 |
return kvmppc_read_int_cpu_dt("ibm,dfp"); |
968 |
} |
969 |
|
970 |
int kvmppc_get_hypercall(CPUPPCState *env, uint8_t *buf, int buf_len) |
971 |
{ |
972 |
uint32_t *hc = (uint32_t*)buf; |
973 |
|
974 |
struct kvm_ppc_pvinfo pvinfo;
|
975 |
|
976 |
if (kvm_check_extension(env->kvm_state, KVM_CAP_PPC_GET_PVINFO) &&
|
977 |
!kvm_vm_ioctl(env->kvm_state, KVM_PPC_GET_PVINFO, &pvinfo)) { |
978 |
memcpy(buf, pvinfo.hcall, buf_len); |
979 |
|
980 |
return 0; |
981 |
} |
982 |
|
983 |
/*
|
984 |
* Fallback to always fail hypercalls:
|
985 |
*
|
986 |
* li r3, -1
|
987 |
* nop
|
988 |
* nop
|
989 |
* nop
|
990 |
*/
|
991 |
|
992 |
hc[0] = 0x3860ffff; |
993 |
hc[1] = 0x60000000; |
994 |
hc[2] = 0x60000000; |
995 |
hc[3] = 0x60000000; |
996 |
|
997 |
return 0; |
998 |
} |
999 |
|
1000 |
void kvmppc_set_papr(CPUPPCState *env)
|
1001 |
{ |
1002 |
struct kvm_enable_cap cap = {};
|
1003 |
int ret;
|
1004 |
|
1005 |
cap.cap = KVM_CAP_PPC_PAPR; |
1006 |
ret = kvm_vcpu_ioctl(env, KVM_ENABLE_CAP, &cap); |
1007 |
|
1008 |
if (ret) {
|
1009 |
cpu_abort(env, "This KVM version does not support PAPR\n");
|
1010 |
} |
1011 |
} |
1012 |
|
1013 |
int kvmppc_smt_threads(void) |
1014 |
{ |
1015 |
return cap_ppc_smt ? cap_ppc_smt : 1; |
1016 |
} |
1017 |
|
1018 |
#ifdef TARGET_PPC64
|
1019 |
off_t kvmppc_alloc_rma(const char *name, MemoryRegion *sysmem) |
1020 |
{ |
1021 |
void *rma;
|
1022 |
off_t size; |
1023 |
int fd;
|
1024 |
struct kvm_allocate_rma ret;
|
1025 |
MemoryRegion *rma_region; |
1026 |
|
1027 |
/* If cap_ppc_rma == 0, contiguous RMA allocation is not supported
|
1028 |
* if cap_ppc_rma == 1, contiguous RMA allocation is supported, but
|
1029 |
* not necessary on this hardware
|
1030 |
* if cap_ppc_rma == 2, contiguous RMA allocation is needed on this hardware
|
1031 |
*
|
1032 |
* FIXME: We should allow the user to force contiguous RMA
|
1033 |
* allocation in the cap_ppc_rma==1 case.
|
1034 |
*/
|
1035 |
if (cap_ppc_rma < 2) { |
1036 |
return 0; |
1037 |
} |
1038 |
|
1039 |
fd = kvm_vm_ioctl(kvm_state, KVM_ALLOCATE_RMA, &ret); |
1040 |
if (fd < 0) { |
1041 |
fprintf(stderr, "KVM: Error on KVM_ALLOCATE_RMA: %s\n",
|
1042 |
strerror(errno)); |
1043 |
return -1; |
1044 |
} |
1045 |
|
1046 |
size = MIN(ret.rma_size, 256ul << 20); |
1047 |
|
1048 |
rma = mmap(NULL, size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0); |
1049 |
if (rma == MAP_FAILED) {
|
1050 |
fprintf(stderr, "KVM: Error mapping RMA: %s\n", strerror(errno));
|
1051 |
return -1; |
1052 |
}; |
1053 |
|
1054 |
rma_region = g_new(MemoryRegion, 1);
|
1055 |
memory_region_init_ram_ptr(rma_region, name, size, rma); |
1056 |
vmstate_register_ram_global(rma_region); |
1057 |
memory_region_add_subregion(sysmem, 0, rma_region);
|
1058 |
|
1059 |
return size;
|
1060 |
} |
1061 |
|
1062 |
uint64_t kvmppc_rma_size(uint64_t current_size, unsigned int hash_shift) |
1063 |
{ |
1064 |
if (cap_ppc_rma >= 2) { |
1065 |
return current_size;
|
1066 |
} |
1067 |
return MIN(current_size,
|
1068 |
getrampagesize() << (hash_shift - 7));
|
1069 |
} |
1070 |
#endif
|
1071 |
|
1072 |
void *kvmppc_create_spapr_tce(uint32_t liobn, uint32_t window_size, int *pfd) |
1073 |
{ |
1074 |
struct kvm_create_spapr_tce args = {
|
1075 |
.liobn = liobn, |
1076 |
.window_size = window_size, |
1077 |
}; |
1078 |
long len;
|
1079 |
int fd;
|
1080 |
void *table;
|
1081 |
|
1082 |
/* Must set fd to -1 so we don't try to munmap when called for
|
1083 |
* destroying the table, which the upper layers -will- do
|
1084 |
*/
|
1085 |
*pfd = -1;
|
1086 |
if (!cap_spapr_tce) {
|
1087 |
return NULL; |
1088 |
} |
1089 |
|
1090 |
fd = kvm_vm_ioctl(kvm_state, KVM_CREATE_SPAPR_TCE, &args); |
1091 |
if (fd < 0) { |
1092 |
fprintf(stderr, "KVM: Failed to create TCE table for liobn 0x%x\n",
|
1093 |
liobn); |
1094 |
return NULL; |
1095 |
} |
1096 |
|
1097 |
len = (window_size / SPAPR_TCE_PAGE_SIZE) * sizeof(sPAPRTCE);
|
1098 |
/* FIXME: round this up to page size */
|
1099 |
|
1100 |
table = mmap(NULL, len, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0); |
1101 |
if (table == MAP_FAILED) {
|
1102 |
fprintf(stderr, "KVM: Failed to map TCE table for liobn 0x%x\n",
|
1103 |
liobn); |
1104 |
close(fd); |
1105 |
return NULL; |
1106 |
} |
1107 |
|
1108 |
*pfd = fd; |
1109 |
return table;
|
1110 |
} |
1111 |
|
1112 |
int kvmppc_remove_spapr_tce(void *table, int fd, uint32_t window_size) |
1113 |
{ |
1114 |
long len;
|
1115 |
|
1116 |
if (fd < 0) { |
1117 |
return -1; |
1118 |
} |
1119 |
|
1120 |
len = (window_size / SPAPR_TCE_PAGE_SIZE)*sizeof(sPAPRTCE);
|
1121 |
if ((munmap(table, len) < 0) || |
1122 |
(close(fd) < 0)) {
|
1123 |
fprintf(stderr, "KVM: Unexpected error removing TCE table: %s",
|
1124 |
strerror(errno)); |
1125 |
/* Leak the table */
|
1126 |
} |
1127 |
|
1128 |
return 0; |
1129 |
} |
1130 |
|
1131 |
int kvmppc_reset_htab(int shift_hint) |
1132 |
{ |
1133 |
uint32_t shift = shift_hint; |
1134 |
|
1135 |
if (!kvm_enabled()) {
|
1136 |
/* Full emulation, tell caller to allocate htab itself */
|
1137 |
return 0; |
1138 |
} |
1139 |
if (kvm_check_extension(kvm_state, KVM_CAP_PPC_ALLOC_HTAB)) {
|
1140 |
int ret;
|
1141 |
ret = kvm_vm_ioctl(kvm_state, KVM_PPC_ALLOCATE_HTAB, &shift); |
1142 |
if (ret == -ENOTTY) {
|
1143 |
/* At least some versions of PR KVM advertise the
|
1144 |
* capability, but don't implement the ioctl(). Oops.
|
1145 |
* Return 0 so that we allocate the htab in qemu, as is
|
1146 |
* correct for PR. */
|
1147 |
return 0; |
1148 |
} else if (ret < 0) { |
1149 |
return ret;
|
1150 |
} |
1151 |
return shift;
|
1152 |
} |
1153 |
|
1154 |
/* We have a kernel that predates the htab reset calls. For PR
|
1155 |
* KVM, we need to allocate the htab ourselves, for an HV KVM of
|
1156 |
* this era, it has allocated a 16MB fixed size hash table
|
1157 |
* already. Kernels of this era have the GET_PVINFO capability
|
1158 |
* only on PR, so we use this hack to determine the right
|
1159 |
* answer */
|
1160 |
if (kvm_check_extension(kvm_state, KVM_CAP_PPC_GET_PVINFO)) {
|
1161 |
/* PR - tell caller to allocate htab */
|
1162 |
return 0; |
1163 |
} else {
|
1164 |
/* HV - assume 16MB kernel allocated htab */
|
1165 |
return 24; |
1166 |
} |
1167 |
} |
1168 |
|
1169 |
static inline uint32_t mfpvr(void) |
1170 |
{ |
1171 |
uint32_t pvr; |
1172 |
|
1173 |
asm ("mfpvr %0" |
1174 |
: "=r"(pvr));
|
1175 |
return pvr;
|
1176 |
} |
1177 |
|
1178 |
static void alter_insns(uint64_t *word, uint64_t flags, bool on) |
1179 |
{ |
1180 |
if (on) {
|
1181 |
*word |= flags; |
1182 |
} else {
|
1183 |
*word &= ~flags; |
1184 |
} |
1185 |
} |
1186 |
|
1187 |
const ppc_def_t *kvmppc_host_cpu_def(void) |
1188 |
{ |
1189 |
uint32_t host_pvr = mfpvr(); |
1190 |
const ppc_def_t *base_spec;
|
1191 |
ppc_def_t *spec; |
1192 |
uint32_t vmx = kvmppc_get_vmx(); |
1193 |
uint32_t dfp = kvmppc_get_dfp(); |
1194 |
|
1195 |
base_spec = ppc_find_by_pvr(host_pvr); |
1196 |
|
1197 |
spec = g_malloc0(sizeof(*spec));
|
1198 |
memcpy(spec, base_spec, sizeof(*spec));
|
1199 |
|
1200 |
/* Now fix up the spec with information we can query from the host */
|
1201 |
|
1202 |
if (vmx != -1) { |
1203 |
/* Only override when we know what the host supports */
|
1204 |
alter_insns(&spec->insns_flags, PPC_ALTIVEC, vmx > 0);
|
1205 |
alter_insns(&spec->insns_flags2, PPC2_VSX, vmx > 1);
|
1206 |
} |
1207 |
if (dfp != -1) { |
1208 |
/* Only override when we know what the host supports */
|
1209 |
alter_insns(&spec->insns_flags2, PPC2_DFP, dfp); |
1210 |
} |
1211 |
|
1212 |
return spec;
|
1213 |
} |
1214 |
|
1215 |
int kvmppc_fixup_cpu(CPUPPCState *env)
|
1216 |
{ |
1217 |
int smt;
|
1218 |
|
1219 |
/* Adjust cpu index for SMT */
|
1220 |
smt = kvmppc_smt_threads(); |
1221 |
env->cpu_index = (env->cpu_index / smp_threads) * smt |
1222 |
+ (env->cpu_index % smp_threads); |
1223 |
|
1224 |
return 0; |
1225 |
} |
1226 |
|
1227 |
|
1228 |
bool kvm_arch_stop_on_emulation_error(CPUPPCState *env)
|
1229 |
{ |
1230 |
return true; |
1231 |
} |
1232 |
|
1233 |
int kvm_arch_on_sigbus_vcpu(CPUPPCState *env, int code, void *addr) |
1234 |
{ |
1235 |
return 1; |
1236 |
} |
1237 |
|
1238 |
int kvm_arch_on_sigbus(int code, void *addr) |
1239 |
{ |
1240 |
return 1; |
1241 |
} |