]> git.proxmox.com Git - mirror_qemu.git/blob - target/arm/kvm32.c
Merge tag 'patchew/20200219160953.13771-1-imammedo@redhat.com' of https://github...
[mirror_qemu.git] / target / arm / kvm32.c
1 /*
2 * ARM implementation of KVM hooks, 32 bit specific code.
3 *
4 * Copyright Christoffer Dall 2009-2010
5 *
6 * This work is licensed under the terms of the GNU GPL, version 2 or later.
7 * See the COPYING file in the top-level directory.
8 *
9 */
10
11 #include "qemu/osdep.h"
12 #include <sys/ioctl.h>
13
14 #include <linux/kvm.h>
15
16 #include "qemu-common.h"
17 #include "cpu.h"
18 #include "qemu/timer.h"
19 #include "sysemu/runstate.h"
20 #include "sysemu/kvm.h"
21 #include "kvm_arm.h"
22 #include "internals.h"
23 #include "qemu/log.h"
24
25 static inline void set_feature(uint64_t *features, int feature)
26 {
27 *features |= 1ULL << feature;
28 }
29
30 static int read_sys_reg32(int fd, uint32_t *pret, uint64_t id)
31 {
32 struct kvm_one_reg idreg = { .id = id, .addr = (uintptr_t)pret };
33
34 assert((id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U32);
35 return ioctl(fd, KVM_GET_ONE_REG, &idreg);
36 }
37
38 bool kvm_arm_get_host_cpu_features(ARMHostCPUFeatures *ahcf)
39 {
40 /* Identify the feature bits corresponding to the host CPU, and
41 * fill out the ARMHostCPUClass fields accordingly. To do this
42 * we have to create a scratch VM, create a single CPU inside it,
43 * and then query that CPU for the relevant ID registers.
44 */
45 int err = 0, fdarray[3];
46 uint32_t midr, id_pfr0;
47 uint64_t features = 0;
48
49 /* Old kernels may not know about the PREFERRED_TARGET ioctl: however
50 * we know these will only support creating one kind of guest CPU,
51 * which is its preferred CPU type.
52 */
53 static const uint32_t cpus_to_try[] = {
54 QEMU_KVM_ARM_TARGET_CORTEX_A15,
55 QEMU_KVM_ARM_TARGET_NONE
56 };
57 /*
58 * target = -1 informs kvm_arm_create_scratch_host_vcpu()
59 * to use the preferred target
60 */
61 struct kvm_vcpu_init init = { .target = -1, };
62
63 if (!kvm_arm_create_scratch_host_vcpu(cpus_to_try, fdarray, &init)) {
64 return false;
65 }
66
67 ahcf->target = init.target;
68
69 /* This is not strictly blessed by the device tree binding docs yet,
70 * but in practice the kernel does not care about this string so
71 * there is no point maintaining an KVM_ARM_TARGET_* -> string table.
72 */
73 ahcf->dtb_compatible = "arm,arm-v7";
74
75 err |= read_sys_reg32(fdarray[2], &midr, ARM_CP15_REG32(0, 0, 0, 0));
76 err |= read_sys_reg32(fdarray[2], &id_pfr0, ARM_CP15_REG32(0, 0, 1, 0));
77
78 err |= read_sys_reg32(fdarray[2], &ahcf->isar.id_isar0,
79 ARM_CP15_REG32(0, 0, 2, 0));
80 err |= read_sys_reg32(fdarray[2], &ahcf->isar.id_isar1,
81 ARM_CP15_REG32(0, 0, 2, 1));
82 err |= read_sys_reg32(fdarray[2], &ahcf->isar.id_isar2,
83 ARM_CP15_REG32(0, 0, 2, 2));
84 err |= read_sys_reg32(fdarray[2], &ahcf->isar.id_isar3,
85 ARM_CP15_REG32(0, 0, 2, 3));
86 err |= read_sys_reg32(fdarray[2], &ahcf->isar.id_isar4,
87 ARM_CP15_REG32(0, 0, 2, 4));
88 err |= read_sys_reg32(fdarray[2], &ahcf->isar.id_isar5,
89 ARM_CP15_REG32(0, 0, 2, 5));
90 if (read_sys_reg32(fdarray[2], &ahcf->isar.id_isar6,
91 ARM_CP15_REG32(0, 0, 2, 7))) {
92 /*
93 * Older kernels don't support reading ID_ISAR6. This register was
94 * only introduced in ARMv8, so we can assume that it is zero on a
95 * CPU that a kernel this old is running on.
96 */
97 ahcf->isar.id_isar6 = 0;
98 }
99
100 err |= read_sys_reg32(fdarray[2], &ahcf->isar.id_dfr0,
101 ARM_CP15_REG32(0, 0, 1, 2));
102
103 err |= read_sys_reg32(fdarray[2], &ahcf->isar.mvfr0,
104 KVM_REG_ARM | KVM_REG_SIZE_U32 |
105 KVM_REG_ARM_VFP | KVM_REG_ARM_VFP_MVFR0);
106 err |= read_sys_reg32(fdarray[2], &ahcf->isar.mvfr1,
107 KVM_REG_ARM | KVM_REG_SIZE_U32 |
108 KVM_REG_ARM_VFP | KVM_REG_ARM_VFP_MVFR1);
109 /*
110 * FIXME: There is not yet a way to read MVFR2.
111 * Fortunately there is not yet anything in there that affects migration.
112 */
113
114 err |= read_sys_reg32(fdarray[2], &ahcf->isar.id_mmfr0,
115 ARM_CP15_REG32(0, 0, 1, 4));
116 err |= read_sys_reg32(fdarray[2], &ahcf->isar.id_mmfr1,
117 ARM_CP15_REG32(0, 0, 1, 5));
118 err |= read_sys_reg32(fdarray[2], &ahcf->isar.id_mmfr2,
119 ARM_CP15_REG32(0, 0, 1, 6));
120 err |= read_sys_reg32(fdarray[2], &ahcf->isar.id_mmfr3,
121 ARM_CP15_REG32(0, 0, 1, 7));
122 if (read_sys_reg32(fdarray[2], &ahcf->isar.id_mmfr4,
123 ARM_CP15_REG32(0, 0, 2, 6))) {
124 /*
125 * Older kernels don't support reading ID_MMFR4 (a new in v8
126 * register); assume it's zero.
127 */
128 ahcf->isar.id_mmfr4 = 0;
129 }
130
131 /*
132 * There is no way to read DBGDIDR, because currently 32-bit KVM
133 * doesn't implement debug at all. Leave it at zero.
134 */
135
136 kvm_arm_destroy_scratch_host_vcpu(fdarray);
137
138 if (err < 0) {
139 return false;
140 }
141
142 /* Now we've retrieved all the register information we can
143 * set the feature bits based on the ID register fields.
144 * We can assume any KVM supporting CPU is at least a v7
145 * with VFPv3, virtualization extensions, and the generic
146 * timers; this in turn implies most of the other feature
147 * bits, but a few must be tested.
148 */
149 set_feature(&features, ARM_FEATURE_V7VE);
150 set_feature(&features, ARM_FEATURE_VFP3);
151 set_feature(&features, ARM_FEATURE_GENERIC_TIMER);
152
153 if (extract32(id_pfr0, 12, 4) == 1) {
154 set_feature(&features, ARM_FEATURE_THUMB2EE);
155 }
156 if (extract32(ahcf->isar.mvfr1, 12, 4) == 1) {
157 set_feature(&features, ARM_FEATURE_NEON);
158 }
159 if (extract32(ahcf->isar.mvfr1, 28, 4) == 1) {
160 /* FMAC support implies VFPv4 */
161 set_feature(&features, ARM_FEATURE_VFP4);
162 }
163
164 ahcf->features = features;
165
166 return true;
167 }
168
169 bool kvm_arm_reg_syncs_via_cpreg_list(uint64_t regidx)
170 {
171 /* Return true if the regidx is a register we should synchronize
172 * via the cpreg_tuples array (ie is not a core reg we sync by
173 * hand in kvm_arch_get/put_registers())
174 */
175 switch (regidx & KVM_REG_ARM_COPROC_MASK) {
176 case KVM_REG_ARM_CORE:
177 case KVM_REG_ARM_VFP:
178 return false;
179 default:
180 return true;
181 }
182 }
183
184 typedef struct CPRegStateLevel {
185 uint64_t regidx;
186 int level;
187 } CPRegStateLevel;
188
189 /* All coprocessor registers not listed in the following table are assumed to
190 * be of the level KVM_PUT_RUNTIME_STATE. If a register should be written less
191 * often, you must add it to this table with a state of either
192 * KVM_PUT_RESET_STATE or KVM_PUT_FULL_STATE.
193 */
194 static const CPRegStateLevel non_runtime_cpregs[] = {
195 { KVM_REG_ARM_TIMER_CNT, KVM_PUT_FULL_STATE },
196 };
197
198 int kvm_arm_cpreg_level(uint64_t regidx)
199 {
200 int i;
201
202 for (i = 0; i < ARRAY_SIZE(non_runtime_cpregs); i++) {
203 const CPRegStateLevel *l = &non_runtime_cpregs[i];
204 if (l->regidx == regidx) {
205 return l->level;
206 }
207 }
208
209 return KVM_PUT_RUNTIME_STATE;
210 }
211
212 #define ARM_CPU_ID_MPIDR 0, 0, 0, 5
213
214 int kvm_arch_init_vcpu(CPUState *cs)
215 {
216 int ret;
217 uint64_t v;
218 uint32_t mpidr;
219 struct kvm_one_reg r;
220 ARMCPU *cpu = ARM_CPU(cs);
221
222 if (cpu->kvm_target == QEMU_KVM_ARM_TARGET_NONE) {
223 fprintf(stderr, "KVM is not supported for this guest CPU type\n");
224 return -EINVAL;
225 }
226
227 qemu_add_vm_change_state_handler(kvm_arm_vm_state_change, cs);
228
229 /* Determine init features for this CPU */
230 memset(cpu->kvm_init_features, 0, sizeof(cpu->kvm_init_features));
231 if (cpu->start_powered_off) {
232 cpu->kvm_init_features[0] |= 1 << KVM_ARM_VCPU_POWER_OFF;
233 }
234 if (kvm_check_extension(cs->kvm_state, KVM_CAP_ARM_PSCI_0_2)) {
235 cpu->psci_version = 2;
236 cpu->kvm_init_features[0] |= 1 << KVM_ARM_VCPU_PSCI_0_2;
237 }
238
239 /* Do KVM_ARM_VCPU_INIT ioctl */
240 ret = kvm_arm_vcpu_init(cs);
241 if (ret) {
242 return ret;
243 }
244
245 /* Query the kernel to make sure it supports 32 VFP
246 * registers: QEMU's "cortex-a15" CPU is always a
247 * VFP-D32 core. The simplest way to do this is just
248 * to attempt to read register d31.
249 */
250 r.id = KVM_REG_ARM | KVM_REG_SIZE_U64 | KVM_REG_ARM_VFP | 31;
251 r.addr = (uintptr_t)(&v);
252 ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &r);
253 if (ret == -ENOENT) {
254 return -EINVAL;
255 }
256
257 /*
258 * When KVM is in use, PSCI is emulated in-kernel and not by qemu.
259 * Currently KVM has its own idea about MPIDR assignment, so we
260 * override our defaults with what we get from KVM.
261 */
262 ret = kvm_get_one_reg(cs, ARM_CP15_REG32(ARM_CPU_ID_MPIDR), &mpidr);
263 if (ret) {
264 return ret;
265 }
266 cpu->mp_affinity = mpidr & ARM32_AFFINITY_MASK;
267
268 /* Check whether userspace can specify guest syndrome value */
269 kvm_arm_init_serror_injection(cs);
270
271 return kvm_arm_init_cpreg_list(cpu);
272 }
273
274 int kvm_arch_destroy_vcpu(CPUState *cs)
275 {
276 return 0;
277 }
278
279 typedef struct Reg {
280 uint64_t id;
281 int offset;
282 } Reg;
283
284 #define COREREG(KERNELNAME, QEMUFIELD) \
285 { \
286 KVM_REG_ARM | KVM_REG_SIZE_U32 | \
287 KVM_REG_ARM_CORE | KVM_REG_ARM_CORE_REG(KERNELNAME), \
288 offsetof(CPUARMState, QEMUFIELD) \
289 }
290
291 #define VFPSYSREG(R) \
292 { \
293 KVM_REG_ARM | KVM_REG_SIZE_U32 | KVM_REG_ARM_VFP | \
294 KVM_REG_ARM_VFP_##R, \
295 offsetof(CPUARMState, vfp.xregs[ARM_VFP_##R]) \
296 }
297
298 /* Like COREREG, but handle fields which are in a uint64_t in CPUARMState. */
299 #define COREREG64(KERNELNAME, QEMUFIELD) \
300 { \
301 KVM_REG_ARM | KVM_REG_SIZE_U32 | \
302 KVM_REG_ARM_CORE | KVM_REG_ARM_CORE_REG(KERNELNAME), \
303 offsetoflow32(CPUARMState, QEMUFIELD) \
304 }
305
306 static const Reg regs[] = {
307 /* R0_usr .. R14_usr */
308 COREREG(usr_regs.uregs[0], regs[0]),
309 COREREG(usr_regs.uregs[1], regs[1]),
310 COREREG(usr_regs.uregs[2], regs[2]),
311 COREREG(usr_regs.uregs[3], regs[3]),
312 COREREG(usr_regs.uregs[4], regs[4]),
313 COREREG(usr_regs.uregs[5], regs[5]),
314 COREREG(usr_regs.uregs[6], regs[6]),
315 COREREG(usr_regs.uregs[7], regs[7]),
316 COREREG(usr_regs.uregs[8], usr_regs[0]),
317 COREREG(usr_regs.uregs[9], usr_regs[1]),
318 COREREG(usr_regs.uregs[10], usr_regs[2]),
319 COREREG(usr_regs.uregs[11], usr_regs[3]),
320 COREREG(usr_regs.uregs[12], usr_regs[4]),
321 COREREG(usr_regs.uregs[13], banked_r13[BANK_USRSYS]),
322 COREREG(usr_regs.uregs[14], banked_r14[BANK_USRSYS]),
323 /* R13, R14, SPSR for SVC, ABT, UND, IRQ banks */
324 COREREG(svc_regs[0], banked_r13[BANK_SVC]),
325 COREREG(svc_regs[1], banked_r14[BANK_SVC]),
326 COREREG64(svc_regs[2], banked_spsr[BANK_SVC]),
327 COREREG(abt_regs[0], banked_r13[BANK_ABT]),
328 COREREG(abt_regs[1], banked_r14[BANK_ABT]),
329 COREREG64(abt_regs[2], banked_spsr[BANK_ABT]),
330 COREREG(und_regs[0], banked_r13[BANK_UND]),
331 COREREG(und_regs[1], banked_r14[BANK_UND]),
332 COREREG64(und_regs[2], banked_spsr[BANK_UND]),
333 COREREG(irq_regs[0], banked_r13[BANK_IRQ]),
334 COREREG(irq_regs[1], banked_r14[BANK_IRQ]),
335 COREREG64(irq_regs[2], banked_spsr[BANK_IRQ]),
336 /* R8_fiq .. R14_fiq and SPSR_fiq */
337 COREREG(fiq_regs[0], fiq_regs[0]),
338 COREREG(fiq_regs[1], fiq_regs[1]),
339 COREREG(fiq_regs[2], fiq_regs[2]),
340 COREREG(fiq_regs[3], fiq_regs[3]),
341 COREREG(fiq_regs[4], fiq_regs[4]),
342 COREREG(fiq_regs[5], banked_r13[BANK_FIQ]),
343 COREREG(fiq_regs[6], banked_r14[BANK_FIQ]),
344 COREREG64(fiq_regs[7], banked_spsr[BANK_FIQ]),
345 /* R15 */
346 COREREG(usr_regs.uregs[15], regs[15]),
347 /* VFP system registers */
348 VFPSYSREG(FPSID),
349 VFPSYSREG(MVFR1),
350 VFPSYSREG(MVFR0),
351 VFPSYSREG(FPEXC),
352 VFPSYSREG(FPINST),
353 VFPSYSREG(FPINST2),
354 };
355
356 int kvm_arch_put_registers(CPUState *cs, int level)
357 {
358 ARMCPU *cpu = ARM_CPU(cs);
359 CPUARMState *env = &cpu->env;
360 struct kvm_one_reg r;
361 int mode, bn;
362 int ret, i;
363 uint32_t cpsr, fpscr;
364
365 /* Make sure the banked regs are properly set */
366 mode = env->uncached_cpsr & CPSR_M;
367 bn = bank_number(mode);
368 if (mode == ARM_CPU_MODE_FIQ) {
369 memcpy(env->fiq_regs, env->regs + 8, 5 * sizeof(uint32_t));
370 } else {
371 memcpy(env->usr_regs, env->regs + 8, 5 * sizeof(uint32_t));
372 }
373 env->banked_r13[bn] = env->regs[13];
374 env->banked_spsr[bn] = env->spsr;
375 env->banked_r14[r14_bank_number(mode)] = env->regs[14];
376
377 /* Now we can safely copy stuff down to the kernel */
378 for (i = 0; i < ARRAY_SIZE(regs); i++) {
379 r.id = regs[i].id;
380 r.addr = (uintptr_t)(env) + regs[i].offset;
381 ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &r);
382 if (ret) {
383 return ret;
384 }
385 }
386
387 /* Special cases which aren't a single CPUARMState field */
388 cpsr = cpsr_read(env);
389 r.id = KVM_REG_ARM | KVM_REG_SIZE_U32 |
390 KVM_REG_ARM_CORE | KVM_REG_ARM_CORE_REG(usr_regs.ARM_cpsr);
391 r.addr = (uintptr_t)(&cpsr);
392 ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &r);
393 if (ret) {
394 return ret;
395 }
396
397 /* VFP registers */
398 r.id = KVM_REG_ARM | KVM_REG_SIZE_U64 | KVM_REG_ARM_VFP;
399 for (i = 0; i < 32; i++) {
400 r.addr = (uintptr_t)aa32_vfp_dreg(env, i);
401 ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &r);
402 if (ret) {
403 return ret;
404 }
405 r.id++;
406 }
407
408 r.id = KVM_REG_ARM | KVM_REG_SIZE_U32 | KVM_REG_ARM_VFP |
409 KVM_REG_ARM_VFP_FPSCR;
410 fpscr = vfp_get_fpscr(env);
411 r.addr = (uintptr_t)&fpscr;
412 ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &r);
413 if (ret) {
414 return ret;
415 }
416
417 ret = kvm_put_vcpu_events(cpu);
418 if (ret) {
419 return ret;
420 }
421
422 write_cpustate_to_list(cpu, true);
423
424 if (!write_list_to_kvmstate(cpu, level)) {
425 return EINVAL;
426 }
427
428 kvm_arm_sync_mpstate_to_kvm(cpu);
429
430 return ret;
431 }
432
433 int kvm_arch_get_registers(CPUState *cs)
434 {
435 ARMCPU *cpu = ARM_CPU(cs);
436 CPUARMState *env = &cpu->env;
437 struct kvm_one_reg r;
438 int mode, bn;
439 int ret, i;
440 uint32_t cpsr, fpscr;
441
442 for (i = 0; i < ARRAY_SIZE(regs); i++) {
443 r.id = regs[i].id;
444 r.addr = (uintptr_t)(env) + regs[i].offset;
445 ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &r);
446 if (ret) {
447 return ret;
448 }
449 }
450
451 /* Special cases which aren't a single CPUARMState field */
452 r.id = KVM_REG_ARM | KVM_REG_SIZE_U32 |
453 KVM_REG_ARM_CORE | KVM_REG_ARM_CORE_REG(usr_regs.ARM_cpsr);
454 r.addr = (uintptr_t)(&cpsr);
455 ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &r);
456 if (ret) {
457 return ret;
458 }
459 cpsr_write(env, cpsr, 0xffffffff, CPSRWriteRaw);
460
461 /* Make sure the current mode regs are properly set */
462 mode = env->uncached_cpsr & CPSR_M;
463 bn = bank_number(mode);
464 if (mode == ARM_CPU_MODE_FIQ) {
465 memcpy(env->regs + 8, env->fiq_regs, 5 * sizeof(uint32_t));
466 } else {
467 memcpy(env->regs + 8, env->usr_regs, 5 * sizeof(uint32_t));
468 }
469 env->regs[13] = env->banked_r13[bn];
470 env->spsr = env->banked_spsr[bn];
471 env->regs[14] = env->banked_r14[r14_bank_number(mode)];
472
473 /* VFP registers */
474 r.id = KVM_REG_ARM | KVM_REG_SIZE_U64 | KVM_REG_ARM_VFP;
475 for (i = 0; i < 32; i++) {
476 r.addr = (uintptr_t)aa32_vfp_dreg(env, i);
477 ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &r);
478 if (ret) {
479 return ret;
480 }
481 r.id++;
482 }
483
484 r.id = KVM_REG_ARM | KVM_REG_SIZE_U32 | KVM_REG_ARM_VFP |
485 KVM_REG_ARM_VFP_FPSCR;
486 r.addr = (uintptr_t)&fpscr;
487 ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &r);
488 if (ret) {
489 return ret;
490 }
491 vfp_set_fpscr(env, fpscr);
492
493 ret = kvm_get_vcpu_events(cpu);
494 if (ret) {
495 return ret;
496 }
497
498 if (!write_kvmstate_to_list(cpu)) {
499 return EINVAL;
500 }
501 /* Note that it's OK to have registers which aren't in CPUState,
502 * so we can ignore a failure return here.
503 */
504 write_list_to_cpustate(cpu);
505
506 kvm_arm_sync_mpstate_to_qemu(cpu);
507
508 return 0;
509 }
510
511 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
512 {
513 qemu_log_mask(LOG_UNIMP, "%s: guest debug not yet implemented\n", __func__);
514 return -EINVAL;
515 }
516
517 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
518 {
519 qemu_log_mask(LOG_UNIMP, "%s: guest debug not yet implemented\n", __func__);
520 return -EINVAL;
521 }
522
523 bool kvm_arm_handle_debug(CPUState *cs, struct kvm_debug_exit_arch *debug_exit)
524 {
525 qemu_log_mask(LOG_UNIMP, "%s: guest debug not yet implemented\n", __func__);
526 return false;
527 }
528
529 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
530 target_ulong len, int type)
531 {
532 qemu_log_mask(LOG_UNIMP, "%s: not implemented\n", __func__);
533 return -EINVAL;
534 }
535
536 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
537 target_ulong len, int type)
538 {
539 qemu_log_mask(LOG_UNIMP, "%s: not implemented\n", __func__);
540 return -EINVAL;
541 }
542
543 void kvm_arch_remove_all_hw_breakpoints(void)
544 {
545 qemu_log_mask(LOG_UNIMP, "%s: not implemented\n", __func__);
546 }
547
548 void kvm_arm_copy_hw_debug_data(struct kvm_guest_debug_arch *ptr)
549 {
550 qemu_log_mask(LOG_UNIMP, "%s: not implemented\n", __func__);
551 }
552
553 bool kvm_arm_hw_debug_active(CPUState *cs)
554 {
555 return false;
556 }
557
558 void kvm_arm_pmu_set_irq(CPUState *cs, int irq)
559 {
560 qemu_log_mask(LOG_UNIMP, "%s: not implemented\n", __func__);
561 }
562
563 void kvm_arm_pmu_init(CPUState *cs)
564 {
565 qemu_log_mask(LOG_UNIMP, "%s: not implemented\n", __func__);
566 }