migration/rdma: Plug memory leaks in qemu_rdma_registration_stop()
[qemu/armbru.git] / target / s390x / kvm.c
blobf2f75d2a57e84101b90fecd0abb2e7df8bd33764
1 /*
2 * QEMU S390x KVM implementation
4 * Copyright (c) 2009 Alexander Graf <agraf@suse.de>
5 * Copyright IBM Corp. 2012
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, see <http://www.gnu.org/licenses/>.
21 #include "qemu/osdep.h"
22 #include <sys/ioctl.h>
24 #include <linux/kvm.h>
25 #include <asm/ptrace.h>
27 #include "qemu-common.h"
28 #include "cpu.h"
29 #include "internal.h"
30 #include "kvm_s390x.h"
31 #include "sysemu/kvm_int.h"
32 #include "qapi/error.h"
33 #include "qemu/error-report.h"
34 #include "qemu/timer.h"
35 #include "qemu/units.h"
36 #include "qemu/main-loop.h"
37 #include "qemu/mmap-alloc.h"
38 #include "qemu/log.h"
39 #include "sysemu/sysemu.h"
40 #include "sysemu/hw_accel.h"
41 #include "sysemu/runstate.h"
42 #include "sysemu/device_tree.h"
43 #include "exec/gdbstub.h"
44 #include "exec/ram_addr.h"
45 #include "trace.h"
46 #include "hw/s390x/s390-pci-inst.h"
47 #include "hw/s390x/s390-pci-bus.h"
48 #include "hw/s390x/ipl.h"
49 #include "hw/s390x/ebcdic.h"
50 #include "exec/memattrs.h"
51 #include "hw/s390x/s390-virtio-ccw.h"
52 #include "hw/s390x/s390-virtio-hcall.h"
53 #include "hw/s390x/pv.h"
55 #ifndef DEBUG_KVM
56 #define DEBUG_KVM 0
57 #endif
59 #define DPRINTF(fmt, ...) do { \
60 if (DEBUG_KVM) { \
61 fprintf(stderr, fmt, ## __VA_ARGS__); \
62 } \
63 } while (0)
65 #define kvm_vm_check_mem_attr(s, attr) \
66 kvm_vm_check_attr(s, KVM_S390_VM_MEM_CTRL, attr)
68 #define IPA0_DIAG 0x8300
69 #define IPA0_SIGP 0xae00
70 #define IPA0_B2 0xb200
71 #define IPA0_B9 0xb900
72 #define IPA0_EB 0xeb00
73 #define IPA0_E3 0xe300
75 #define PRIV_B2_SCLP_CALL 0x20
76 #define PRIV_B2_CSCH 0x30
77 #define PRIV_B2_HSCH 0x31
78 #define PRIV_B2_MSCH 0x32
79 #define PRIV_B2_SSCH 0x33
80 #define PRIV_B2_STSCH 0x34
81 #define PRIV_B2_TSCH 0x35
82 #define PRIV_B2_TPI 0x36
83 #define PRIV_B2_SAL 0x37
84 #define PRIV_B2_RSCH 0x38
85 #define PRIV_B2_STCRW 0x39
86 #define PRIV_B2_STCPS 0x3a
87 #define PRIV_B2_RCHP 0x3b
88 #define PRIV_B2_SCHM 0x3c
89 #define PRIV_B2_CHSC 0x5f
90 #define PRIV_B2_SIGA 0x74
91 #define PRIV_B2_XSCH 0x76
93 #define PRIV_EB_SQBS 0x8a
94 #define PRIV_EB_PCISTB 0xd0
95 #define PRIV_EB_SIC 0xd1
97 #define PRIV_B9_EQBS 0x9c
98 #define PRIV_B9_CLP 0xa0
99 #define PRIV_B9_PCISTG 0xd0
100 #define PRIV_B9_PCILG 0xd2
101 #define PRIV_B9_RPCIT 0xd3
103 #define PRIV_E3_MPCIFC 0xd0
104 #define PRIV_E3_STPCIFC 0xd4
106 #define DIAG_TIMEREVENT 0x288
107 #define DIAG_IPL 0x308
108 #define DIAG_KVM_HYPERCALL 0x500
109 #define DIAG_KVM_BREAKPOINT 0x501
111 #define ICPT_INSTRUCTION 0x04
112 #define ICPT_PROGRAM 0x08
113 #define ICPT_EXT_INT 0x14
114 #define ICPT_WAITPSW 0x1c
115 #define ICPT_SOFT_INTERCEPT 0x24
116 #define ICPT_CPU_STOP 0x28
117 #define ICPT_OPEREXC 0x2c
118 #define ICPT_IO 0x40
119 #define ICPT_PV_INSTR 0x68
120 #define ICPT_PV_INSTR_NOTIFICATION 0x6c
122 #define NR_LOCAL_IRQS 32
124 * Needs to be big enough to contain max_cpus emergency signals
125 * and in addition NR_LOCAL_IRQS interrupts
127 #define VCPU_IRQ_BUF_SIZE(max_cpus) (sizeof(struct kvm_s390_irq) * \
128 (max_cpus + NR_LOCAL_IRQS))
130 * KVM does only support memory slots up to KVM_MEM_MAX_NR_PAGES pages
131 * as the dirty bitmap must be managed by bitops that take an int as
132 * position indicator. This would end at an unaligned address
133 * (0x7fffff00000). As future variants might provide larger pages
134 * and to make all addresses properly aligned, let us split at 4TB.
136 #define KVM_SLOT_MAX_BYTES (4UL * TiB)
138 static CPUWatchpoint hw_watchpoint;
140 * We don't use a list because this structure is also used to transmit the
141 * hardware breakpoints to the kernel.
143 static struct kvm_hw_breakpoint *hw_breakpoints;
144 static int nb_hw_breakpoints;
146 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
147 KVM_CAP_LAST_INFO
150 static int cap_sync_regs;
151 static int cap_async_pf;
152 static int cap_mem_op;
153 static int cap_s390_irq;
154 static int cap_ri;
155 static int cap_gs;
156 static int cap_hpage_1m;
157 static int cap_vcpu_resets;
158 static int cap_protected;
160 static int active_cmma;
162 static void *legacy_s390_alloc(size_t size, uint64_t *align, bool shared);
164 static int kvm_s390_query_mem_limit(uint64_t *memory_limit)
166 struct kvm_device_attr attr = {
167 .group = KVM_S390_VM_MEM_CTRL,
168 .attr = KVM_S390_VM_MEM_LIMIT_SIZE,
169 .addr = (uint64_t) memory_limit,
172 return kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
175 int kvm_s390_set_mem_limit(uint64_t new_limit, uint64_t *hw_limit)
177 int rc;
179 struct kvm_device_attr attr = {
180 .group = KVM_S390_VM_MEM_CTRL,
181 .attr = KVM_S390_VM_MEM_LIMIT_SIZE,
182 .addr = (uint64_t) &new_limit,
185 if (!kvm_vm_check_mem_attr(kvm_state, KVM_S390_VM_MEM_LIMIT_SIZE)) {
186 return 0;
189 rc = kvm_s390_query_mem_limit(hw_limit);
190 if (rc) {
191 return rc;
192 } else if (*hw_limit < new_limit) {
193 return -E2BIG;
196 return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
199 int kvm_s390_cmma_active(void)
201 return active_cmma;
204 static bool kvm_s390_cmma_available(void)
206 static bool initialized, value;
208 if (!initialized) {
209 initialized = true;
210 value = kvm_vm_check_mem_attr(kvm_state, KVM_S390_VM_MEM_ENABLE_CMMA) &&
211 kvm_vm_check_mem_attr(kvm_state, KVM_S390_VM_MEM_CLR_CMMA);
213 return value;
216 void kvm_s390_cmma_reset(void)
218 int rc;
219 struct kvm_device_attr attr = {
220 .group = KVM_S390_VM_MEM_CTRL,
221 .attr = KVM_S390_VM_MEM_CLR_CMMA,
224 if (!kvm_s390_cmma_active()) {
225 return;
228 rc = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
229 trace_kvm_clear_cmma(rc);
232 static void kvm_s390_enable_cmma(void)
234 int rc;
235 struct kvm_device_attr attr = {
236 .group = KVM_S390_VM_MEM_CTRL,
237 .attr = KVM_S390_VM_MEM_ENABLE_CMMA,
240 if (cap_hpage_1m) {
241 warn_report("CMM will not be enabled because it is not "
242 "compatible with huge memory backings.");
243 return;
245 rc = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
246 active_cmma = !rc;
247 trace_kvm_enable_cmma(rc);
250 static void kvm_s390_set_attr(uint64_t attr)
252 struct kvm_device_attr attribute = {
253 .group = KVM_S390_VM_CRYPTO,
254 .attr = attr,
257 int ret = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attribute);
259 if (ret) {
260 error_report("Failed to set crypto device attribute %lu: %s",
261 attr, strerror(-ret));
265 static void kvm_s390_init_aes_kw(void)
267 uint64_t attr = KVM_S390_VM_CRYPTO_DISABLE_AES_KW;
269 if (object_property_get_bool(OBJECT(qdev_get_machine()), "aes-key-wrap",
270 NULL)) {
271 attr = KVM_S390_VM_CRYPTO_ENABLE_AES_KW;
274 if (kvm_vm_check_attr(kvm_state, KVM_S390_VM_CRYPTO, attr)) {
275 kvm_s390_set_attr(attr);
279 static void kvm_s390_init_dea_kw(void)
281 uint64_t attr = KVM_S390_VM_CRYPTO_DISABLE_DEA_KW;
283 if (object_property_get_bool(OBJECT(qdev_get_machine()), "dea-key-wrap",
284 NULL)) {
285 attr = KVM_S390_VM_CRYPTO_ENABLE_DEA_KW;
288 if (kvm_vm_check_attr(kvm_state, KVM_S390_VM_CRYPTO, attr)) {
289 kvm_s390_set_attr(attr);
293 void kvm_s390_crypto_reset(void)
295 if (s390_has_feat(S390_FEAT_MSA_EXT_3)) {
296 kvm_s390_init_aes_kw();
297 kvm_s390_init_dea_kw();
301 void kvm_s390_set_max_pagesize(uint64_t pagesize, Error **errp)
303 if (pagesize == 4 * KiB) {
304 return;
307 if (!hpage_1m_allowed()) {
308 error_setg(errp, "This QEMU machine does not support huge page "
309 "mappings");
310 return;
313 if (pagesize != 1 * MiB) {
314 error_setg(errp, "Memory backing with 2G pages was specified, "
315 "but KVM does not support this memory backing");
316 return;
319 if (kvm_vm_enable_cap(kvm_state, KVM_CAP_S390_HPAGE_1M, 0)) {
320 error_setg(errp, "Memory backing with 1M pages was specified, "
321 "but KVM does not support this memory backing");
322 return;
325 cap_hpage_1m = 1;
328 int kvm_s390_get_hpage_1m(void)
330 return cap_hpage_1m;
333 static void ccw_machine_class_foreach(ObjectClass *oc, void *opaque)
335 MachineClass *mc = MACHINE_CLASS(oc);
337 mc->default_cpu_type = S390_CPU_TYPE_NAME("host");
340 int kvm_arch_init(MachineState *ms, KVMState *s)
342 object_class_foreach(ccw_machine_class_foreach, TYPE_S390_CCW_MACHINE,
343 false, NULL);
345 if (!kvm_check_extension(kvm_state, KVM_CAP_DEVICE_CTRL)) {
346 error_report("KVM is missing capability KVM_CAP_DEVICE_CTRL - "
347 "please use kernel 3.15 or newer");
348 return -1;
351 cap_sync_regs = kvm_check_extension(s, KVM_CAP_SYNC_REGS);
352 cap_async_pf = kvm_check_extension(s, KVM_CAP_ASYNC_PF);
353 cap_mem_op = kvm_check_extension(s, KVM_CAP_S390_MEM_OP);
354 cap_s390_irq = kvm_check_extension(s, KVM_CAP_S390_INJECT_IRQ);
355 cap_vcpu_resets = kvm_check_extension(s, KVM_CAP_S390_VCPU_RESETS);
356 cap_protected = kvm_check_extension(s, KVM_CAP_S390_PROTECTED);
358 if (!kvm_check_extension(s, KVM_CAP_S390_GMAP)
359 || !kvm_check_extension(s, KVM_CAP_S390_COW)) {
360 phys_mem_set_alloc(legacy_s390_alloc);
363 kvm_vm_enable_cap(s, KVM_CAP_S390_USER_SIGP, 0);
364 kvm_vm_enable_cap(s, KVM_CAP_S390_VECTOR_REGISTERS, 0);
365 kvm_vm_enable_cap(s, KVM_CAP_S390_USER_STSI, 0);
366 if (ri_allowed()) {
367 if (kvm_vm_enable_cap(s, KVM_CAP_S390_RI, 0) == 0) {
368 cap_ri = 1;
371 if (cpu_model_allowed()) {
372 if (kvm_vm_enable_cap(s, KVM_CAP_S390_GS, 0) == 0) {
373 cap_gs = 1;
378 * The migration interface for ais was introduced with kernel 4.13
379 * but the capability itself had been active since 4.12. As migration
380 * support is considered necessary, we only try to enable this for
381 * newer machine types if KVM_CAP_S390_AIS_MIGRATION is available.
383 if (cpu_model_allowed() && kvm_kernel_irqchip_allowed() &&
384 kvm_check_extension(s, KVM_CAP_S390_AIS_MIGRATION)) {
385 kvm_vm_enable_cap(s, KVM_CAP_S390_AIS, 0);
388 kvm_set_max_memslot_size(KVM_SLOT_MAX_BYTES);
389 return 0;
392 int kvm_arch_irqchip_create(KVMState *s)
394 return 0;
397 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
399 return cpu->cpu_index;
402 int kvm_arch_init_vcpu(CPUState *cs)
404 unsigned int max_cpus = MACHINE(qdev_get_machine())->smp.max_cpus;
405 S390CPU *cpu = S390_CPU(cs);
406 kvm_s390_set_cpu_state(cpu, cpu->env.cpu_state);
407 cpu->irqstate = g_malloc0(VCPU_IRQ_BUF_SIZE(max_cpus));
408 return 0;
411 int kvm_arch_destroy_vcpu(CPUState *cs)
413 S390CPU *cpu = S390_CPU(cs);
415 g_free(cpu->irqstate);
416 cpu->irqstate = NULL;
418 return 0;
421 static void kvm_s390_reset_vcpu(S390CPU *cpu, unsigned long type)
423 CPUState *cs = CPU(cpu);
426 * The reset call is needed here to reset in-kernel vcpu data that
427 * we can't access directly from QEMU (i.e. with older kernels
428 * which don't support sync_regs/ONE_REG). Before this ioctl
429 * cpu_synchronize_state() is called in common kvm code
430 * (kvm-all).
432 if (kvm_vcpu_ioctl(cs, type)) {
433 error_report("CPU reset failed on CPU %i type %lx",
434 cs->cpu_index, type);
438 void kvm_s390_reset_vcpu_initial(S390CPU *cpu)
440 kvm_s390_reset_vcpu(cpu, KVM_S390_INITIAL_RESET);
443 void kvm_s390_reset_vcpu_clear(S390CPU *cpu)
445 if (cap_vcpu_resets) {
446 kvm_s390_reset_vcpu(cpu, KVM_S390_CLEAR_RESET);
447 } else {
448 kvm_s390_reset_vcpu(cpu, KVM_S390_INITIAL_RESET);
452 void kvm_s390_reset_vcpu_normal(S390CPU *cpu)
454 if (cap_vcpu_resets) {
455 kvm_s390_reset_vcpu(cpu, KVM_S390_NORMAL_RESET);
459 static int can_sync_regs(CPUState *cs, int regs)
461 return cap_sync_regs && (cs->kvm_run->kvm_valid_regs & regs) == regs;
464 int kvm_arch_put_registers(CPUState *cs, int level)
466 S390CPU *cpu = S390_CPU(cs);
467 CPUS390XState *env = &cpu->env;
468 struct kvm_sregs sregs;
469 struct kvm_regs regs;
470 struct kvm_fpu fpu = {};
471 int r;
472 int i;
474 /* always save the PSW and the GPRS*/
475 cs->kvm_run->psw_addr = env->psw.addr;
476 cs->kvm_run->psw_mask = env->psw.mask;
478 if (can_sync_regs(cs, KVM_SYNC_GPRS)) {
479 for (i = 0; i < 16; i++) {
480 cs->kvm_run->s.regs.gprs[i] = env->regs[i];
481 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_GPRS;
483 } else {
484 for (i = 0; i < 16; i++) {
485 regs.gprs[i] = env->regs[i];
487 r = kvm_vcpu_ioctl(cs, KVM_SET_REGS, &regs);
488 if (r < 0) {
489 return r;
493 if (can_sync_regs(cs, KVM_SYNC_VRS)) {
494 for (i = 0; i < 32; i++) {
495 cs->kvm_run->s.regs.vrs[i][0] = env->vregs[i][0];
496 cs->kvm_run->s.regs.vrs[i][1] = env->vregs[i][1];
498 cs->kvm_run->s.regs.fpc = env->fpc;
499 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_VRS;
500 } else if (can_sync_regs(cs, KVM_SYNC_FPRS)) {
501 for (i = 0; i < 16; i++) {
502 cs->kvm_run->s.regs.fprs[i] = *get_freg(env, i);
504 cs->kvm_run->s.regs.fpc = env->fpc;
505 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_FPRS;
506 } else {
507 /* Floating point */
508 for (i = 0; i < 16; i++) {
509 fpu.fprs[i] = *get_freg(env, i);
511 fpu.fpc = env->fpc;
513 r = kvm_vcpu_ioctl(cs, KVM_SET_FPU, &fpu);
514 if (r < 0) {
515 return r;
519 /* Do we need to save more than that? */
520 if (level == KVM_PUT_RUNTIME_STATE) {
521 return 0;
524 if (can_sync_regs(cs, KVM_SYNC_ARCH0)) {
525 cs->kvm_run->s.regs.cputm = env->cputm;
526 cs->kvm_run->s.regs.ckc = env->ckc;
527 cs->kvm_run->s.regs.todpr = env->todpr;
528 cs->kvm_run->s.regs.gbea = env->gbea;
529 cs->kvm_run->s.regs.pp = env->pp;
530 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ARCH0;
531 } else {
533 * These ONE_REGS are not protected by a capability. As they are only
534 * necessary for migration we just trace a possible error, but don't
535 * return with an error return code.
537 kvm_set_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
538 kvm_set_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
539 kvm_set_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
540 kvm_set_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
541 kvm_set_one_reg(cs, KVM_REG_S390_PP, &env->pp);
544 if (can_sync_regs(cs, KVM_SYNC_RICCB)) {
545 memcpy(cs->kvm_run->s.regs.riccb, env->riccb, 64);
546 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_RICCB;
549 /* pfault parameters */
550 if (can_sync_regs(cs, KVM_SYNC_PFAULT)) {
551 cs->kvm_run->s.regs.pft = env->pfault_token;
552 cs->kvm_run->s.regs.pfs = env->pfault_select;
553 cs->kvm_run->s.regs.pfc = env->pfault_compare;
554 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_PFAULT;
555 } else if (cap_async_pf) {
556 r = kvm_set_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
557 if (r < 0) {
558 return r;
560 r = kvm_set_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
561 if (r < 0) {
562 return r;
564 r = kvm_set_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
565 if (r < 0) {
566 return r;
570 /* access registers and control registers*/
571 if (can_sync_regs(cs, KVM_SYNC_ACRS | KVM_SYNC_CRS)) {
572 for (i = 0; i < 16; i++) {
573 cs->kvm_run->s.regs.acrs[i] = env->aregs[i];
574 cs->kvm_run->s.regs.crs[i] = env->cregs[i];
576 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ACRS;
577 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_CRS;
578 } else {
579 for (i = 0; i < 16; i++) {
580 sregs.acrs[i] = env->aregs[i];
581 sregs.crs[i] = env->cregs[i];
583 r = kvm_vcpu_ioctl(cs, KVM_SET_SREGS, &sregs);
584 if (r < 0) {
585 return r;
589 if (can_sync_regs(cs, KVM_SYNC_GSCB)) {
590 memcpy(cs->kvm_run->s.regs.gscb, env->gscb, 32);
591 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_GSCB;
594 if (can_sync_regs(cs, KVM_SYNC_BPBC)) {
595 cs->kvm_run->s.regs.bpbc = env->bpbc;
596 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_BPBC;
599 if (can_sync_regs(cs, KVM_SYNC_ETOKEN)) {
600 cs->kvm_run->s.regs.etoken = env->etoken;
601 cs->kvm_run->s.regs.etoken_extension = env->etoken_extension;
602 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ETOKEN;
605 /* Finally the prefix */
606 if (can_sync_regs(cs, KVM_SYNC_PREFIX)) {
607 cs->kvm_run->s.regs.prefix = env->psa;
608 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_PREFIX;
609 } else {
610 /* prefix is only supported via sync regs */
612 return 0;
615 int kvm_arch_get_registers(CPUState *cs)
617 S390CPU *cpu = S390_CPU(cs);
618 CPUS390XState *env = &cpu->env;
619 struct kvm_sregs sregs;
620 struct kvm_regs regs;
621 struct kvm_fpu fpu;
622 int i, r;
624 /* get the PSW */
625 env->psw.addr = cs->kvm_run->psw_addr;
626 env->psw.mask = cs->kvm_run->psw_mask;
628 /* the GPRS */
629 if (can_sync_regs(cs, KVM_SYNC_GPRS)) {
630 for (i = 0; i < 16; i++) {
631 env->regs[i] = cs->kvm_run->s.regs.gprs[i];
633 } else {
634 r = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
635 if (r < 0) {
636 return r;
638 for (i = 0; i < 16; i++) {
639 env->regs[i] = regs.gprs[i];
643 /* The ACRS and CRS */
644 if (can_sync_regs(cs, KVM_SYNC_ACRS | KVM_SYNC_CRS)) {
645 for (i = 0; i < 16; i++) {
646 env->aregs[i] = cs->kvm_run->s.regs.acrs[i];
647 env->cregs[i] = cs->kvm_run->s.regs.crs[i];
649 } else {
650 r = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs);
651 if (r < 0) {
652 return r;
654 for (i = 0; i < 16; i++) {
655 env->aregs[i] = sregs.acrs[i];
656 env->cregs[i] = sregs.crs[i];
660 /* Floating point and vector registers */
661 if (can_sync_regs(cs, KVM_SYNC_VRS)) {
662 for (i = 0; i < 32; i++) {
663 env->vregs[i][0] = cs->kvm_run->s.regs.vrs[i][0];
664 env->vregs[i][1] = cs->kvm_run->s.regs.vrs[i][1];
666 env->fpc = cs->kvm_run->s.regs.fpc;
667 } else if (can_sync_regs(cs, KVM_SYNC_FPRS)) {
668 for (i = 0; i < 16; i++) {
669 *get_freg(env, i) = cs->kvm_run->s.regs.fprs[i];
671 env->fpc = cs->kvm_run->s.regs.fpc;
672 } else {
673 r = kvm_vcpu_ioctl(cs, KVM_GET_FPU, &fpu);
674 if (r < 0) {
675 return r;
677 for (i = 0; i < 16; i++) {
678 *get_freg(env, i) = fpu.fprs[i];
680 env->fpc = fpu.fpc;
683 /* The prefix */
684 if (can_sync_regs(cs, KVM_SYNC_PREFIX)) {
685 env->psa = cs->kvm_run->s.regs.prefix;
688 if (can_sync_regs(cs, KVM_SYNC_ARCH0)) {
689 env->cputm = cs->kvm_run->s.regs.cputm;
690 env->ckc = cs->kvm_run->s.regs.ckc;
691 env->todpr = cs->kvm_run->s.regs.todpr;
692 env->gbea = cs->kvm_run->s.regs.gbea;
693 env->pp = cs->kvm_run->s.regs.pp;
694 } else {
696 * These ONE_REGS are not protected by a capability. As they are only
697 * necessary for migration we just trace a possible error, but don't
698 * return with an error return code.
700 kvm_get_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
701 kvm_get_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
702 kvm_get_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
703 kvm_get_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
704 kvm_get_one_reg(cs, KVM_REG_S390_PP, &env->pp);
707 if (can_sync_regs(cs, KVM_SYNC_RICCB)) {
708 memcpy(env->riccb, cs->kvm_run->s.regs.riccb, 64);
711 if (can_sync_regs(cs, KVM_SYNC_GSCB)) {
712 memcpy(env->gscb, cs->kvm_run->s.regs.gscb, 32);
715 if (can_sync_regs(cs, KVM_SYNC_BPBC)) {
716 env->bpbc = cs->kvm_run->s.regs.bpbc;
719 if (can_sync_regs(cs, KVM_SYNC_ETOKEN)) {
720 env->etoken = cs->kvm_run->s.regs.etoken;
721 env->etoken_extension = cs->kvm_run->s.regs.etoken_extension;
724 /* pfault parameters */
725 if (can_sync_regs(cs, KVM_SYNC_PFAULT)) {
726 env->pfault_token = cs->kvm_run->s.regs.pft;
727 env->pfault_select = cs->kvm_run->s.regs.pfs;
728 env->pfault_compare = cs->kvm_run->s.regs.pfc;
729 } else if (cap_async_pf) {
730 r = kvm_get_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
731 if (r < 0) {
732 return r;
734 r = kvm_get_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
735 if (r < 0) {
736 return r;
738 r = kvm_get_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
739 if (r < 0) {
740 return r;
744 return 0;
747 int kvm_s390_get_clock(uint8_t *tod_high, uint64_t *tod_low)
749 int r;
750 struct kvm_device_attr attr = {
751 .group = KVM_S390_VM_TOD,
752 .attr = KVM_S390_VM_TOD_LOW,
753 .addr = (uint64_t)tod_low,
756 r = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
757 if (r) {
758 return r;
761 attr.attr = KVM_S390_VM_TOD_HIGH;
762 attr.addr = (uint64_t)tod_high;
763 return kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
766 int kvm_s390_get_clock_ext(uint8_t *tod_high, uint64_t *tod_low)
768 int r;
769 struct kvm_s390_vm_tod_clock gtod;
770 struct kvm_device_attr attr = {
771 .group = KVM_S390_VM_TOD,
772 .attr = KVM_S390_VM_TOD_EXT,
773 .addr = (uint64_t)&gtod,
776 r = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
777 *tod_high = gtod.epoch_idx;
778 *tod_low = gtod.tod;
780 return r;
783 int kvm_s390_set_clock(uint8_t tod_high, uint64_t tod_low)
785 int r;
786 struct kvm_device_attr attr = {
787 .group = KVM_S390_VM_TOD,
788 .attr = KVM_S390_VM_TOD_LOW,
789 .addr = (uint64_t)&tod_low,
792 r = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
793 if (r) {
794 return r;
797 attr.attr = KVM_S390_VM_TOD_HIGH;
798 attr.addr = (uint64_t)&tod_high;
799 return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
802 int kvm_s390_set_clock_ext(uint8_t tod_high, uint64_t tod_low)
804 struct kvm_s390_vm_tod_clock gtod = {
805 .epoch_idx = tod_high,
806 .tod = tod_low,
808 struct kvm_device_attr attr = {
809 .group = KVM_S390_VM_TOD,
810 .attr = KVM_S390_VM_TOD_EXT,
811 .addr = (uint64_t)&gtod,
814 return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
818 * kvm_s390_mem_op:
819 * @addr: the logical start address in guest memory
820 * @ar: the access register number
821 * @hostbuf: buffer in host memory. NULL = do only checks w/o copying
822 * @len: length that should be transferred
823 * @is_write: true = write, false = read
824 * Returns: 0 on success, non-zero if an exception or error occurred
826 * Use KVM ioctl to read/write from/to guest memory. An access exception
827 * is injected into the vCPU in case of translation errors.
829 int kvm_s390_mem_op(S390CPU *cpu, vaddr addr, uint8_t ar, void *hostbuf,
830 int len, bool is_write)
832 struct kvm_s390_mem_op mem_op = {
833 .gaddr = addr,
834 .flags = KVM_S390_MEMOP_F_INJECT_EXCEPTION,
835 .size = len,
836 .op = is_write ? KVM_S390_MEMOP_LOGICAL_WRITE
837 : KVM_S390_MEMOP_LOGICAL_READ,
838 .buf = (uint64_t)hostbuf,
839 .ar = ar,
841 int ret;
843 if (!cap_mem_op) {
844 return -ENOSYS;
846 if (!hostbuf) {
847 mem_op.flags |= KVM_S390_MEMOP_F_CHECK_ONLY;
850 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_S390_MEM_OP, &mem_op);
851 if (ret < 0) {
852 warn_report("KVM_S390_MEM_OP failed: %s", strerror(-ret));
854 return ret;
857 int kvm_s390_mem_op_pv(S390CPU *cpu, uint64_t offset, void *hostbuf,
858 int len, bool is_write)
860 struct kvm_s390_mem_op mem_op = {
861 .sida_offset = offset,
862 .size = len,
863 .op = is_write ? KVM_S390_MEMOP_SIDA_WRITE
864 : KVM_S390_MEMOP_SIDA_READ,
865 .buf = (uint64_t)hostbuf,
867 int ret;
869 if (!cap_mem_op || !cap_protected) {
870 return -ENOSYS;
873 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_S390_MEM_OP, &mem_op);
874 if (ret < 0) {
875 error_report("KVM_S390_MEM_OP failed: %s", strerror(-ret));
876 abort();
878 return ret;
882 * Legacy layout for s390:
883 * Older S390 KVM requires the topmost vma of the RAM to be
884 * smaller than an system defined value, which is at least 256GB.
885 * Larger systems have larger values. We put the guest between
886 * the end of data segment (system break) and this value. We
887 * use 32GB as a base to have enough room for the system break
888 * to grow. We also have to use MAP parameters that avoid
889 * read-only mapping of guest pages.
891 static void *legacy_s390_alloc(size_t size, uint64_t *align, bool shared)
893 static void *mem;
895 if (mem) {
896 /* we only support one allocation, which is enough for initial ram */
897 return NULL;
900 mem = mmap((void *) 0x800000000ULL, size,
901 PROT_EXEC|PROT_READ|PROT_WRITE,
902 MAP_SHARED | MAP_ANONYMOUS | MAP_FIXED, -1, 0);
903 if (mem == MAP_FAILED) {
904 mem = NULL;
906 if (mem && align) {
907 *align = QEMU_VMALLOC_ALIGN;
909 return mem;
912 static uint8_t const *sw_bp_inst;
913 static uint8_t sw_bp_ilen;
915 static void determine_sw_breakpoint_instr(void)
917 /* DIAG 501 is used for sw breakpoints with old kernels */
918 static const uint8_t diag_501[] = {0x83, 0x24, 0x05, 0x01};
919 /* Instruction 0x0000 is used for sw breakpoints with recent kernels */
920 static const uint8_t instr_0x0000[] = {0x00, 0x00};
922 if (sw_bp_inst) {
923 return;
925 if (kvm_vm_enable_cap(kvm_state, KVM_CAP_S390_USER_INSTR0, 0)) {
926 sw_bp_inst = diag_501;
927 sw_bp_ilen = sizeof(diag_501);
928 DPRINTF("KVM: will use 4-byte sw breakpoints.\n");
929 } else {
930 sw_bp_inst = instr_0x0000;
931 sw_bp_ilen = sizeof(instr_0x0000);
932 DPRINTF("KVM: will use 2-byte sw breakpoints.\n");
936 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
938 determine_sw_breakpoint_instr();
940 if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
941 sw_bp_ilen, 0) ||
942 cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)sw_bp_inst, sw_bp_ilen, 1)) {
943 return -EINVAL;
945 return 0;
948 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
950 uint8_t t[MAX_ILEN];
952 if (cpu_memory_rw_debug(cs, bp->pc, t, sw_bp_ilen, 0)) {
953 return -EINVAL;
954 } else if (memcmp(t, sw_bp_inst, sw_bp_ilen)) {
955 return -EINVAL;
956 } else if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
957 sw_bp_ilen, 1)) {
958 return -EINVAL;
961 return 0;
964 static struct kvm_hw_breakpoint *find_hw_breakpoint(target_ulong addr,
965 int len, int type)
967 int n;
969 for (n = 0; n < nb_hw_breakpoints; n++) {
970 if (hw_breakpoints[n].addr == addr && hw_breakpoints[n].type == type &&
971 (hw_breakpoints[n].len == len || len == -1)) {
972 return &hw_breakpoints[n];
976 return NULL;
979 static int insert_hw_breakpoint(target_ulong addr, int len, int type)
981 int size;
983 if (find_hw_breakpoint(addr, len, type)) {
984 return -EEXIST;
987 size = (nb_hw_breakpoints + 1) * sizeof(struct kvm_hw_breakpoint);
989 if (!hw_breakpoints) {
990 nb_hw_breakpoints = 0;
991 hw_breakpoints = (struct kvm_hw_breakpoint *)g_try_malloc(size);
992 } else {
993 hw_breakpoints =
994 (struct kvm_hw_breakpoint *)g_try_realloc(hw_breakpoints, size);
997 if (!hw_breakpoints) {
998 nb_hw_breakpoints = 0;
999 return -ENOMEM;
1002 hw_breakpoints[nb_hw_breakpoints].addr = addr;
1003 hw_breakpoints[nb_hw_breakpoints].len = len;
1004 hw_breakpoints[nb_hw_breakpoints].type = type;
1006 nb_hw_breakpoints++;
1008 return 0;
1011 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
1012 target_ulong len, int type)
1014 switch (type) {
1015 case GDB_BREAKPOINT_HW:
1016 type = KVM_HW_BP;
1017 break;
1018 case GDB_WATCHPOINT_WRITE:
1019 if (len < 1) {
1020 return -EINVAL;
1022 type = KVM_HW_WP_WRITE;
1023 break;
1024 default:
1025 return -ENOSYS;
1027 return insert_hw_breakpoint(addr, len, type);
1030 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
1031 target_ulong len, int type)
1033 int size;
1034 struct kvm_hw_breakpoint *bp = find_hw_breakpoint(addr, len, type);
1036 if (bp == NULL) {
1037 return -ENOENT;
1040 nb_hw_breakpoints--;
1041 if (nb_hw_breakpoints > 0) {
1043 * In order to trim the array, move the last element to the position to
1044 * be removed - if necessary.
1046 if (bp != &hw_breakpoints[nb_hw_breakpoints]) {
1047 *bp = hw_breakpoints[nb_hw_breakpoints];
1049 size = nb_hw_breakpoints * sizeof(struct kvm_hw_breakpoint);
1050 hw_breakpoints =
1051 (struct kvm_hw_breakpoint *)g_realloc(hw_breakpoints, size);
1052 } else {
1053 g_free(hw_breakpoints);
1054 hw_breakpoints = NULL;
1057 return 0;
1060 void kvm_arch_remove_all_hw_breakpoints(void)
1062 nb_hw_breakpoints = 0;
1063 g_free(hw_breakpoints);
1064 hw_breakpoints = NULL;
1067 void kvm_arch_update_guest_debug(CPUState *cpu, struct kvm_guest_debug *dbg)
1069 int i;
1071 if (nb_hw_breakpoints > 0) {
1072 dbg->arch.nr_hw_bp = nb_hw_breakpoints;
1073 dbg->arch.hw_bp = hw_breakpoints;
1075 for (i = 0; i < nb_hw_breakpoints; ++i) {
1076 hw_breakpoints[i].phys_addr = s390_cpu_get_phys_addr_debug(cpu,
1077 hw_breakpoints[i].addr);
1079 dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
1080 } else {
1081 dbg->arch.nr_hw_bp = 0;
1082 dbg->arch.hw_bp = NULL;
1086 void kvm_arch_pre_run(CPUState *cpu, struct kvm_run *run)
1090 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
1092 return MEMTXATTRS_UNSPECIFIED;
1095 int kvm_arch_process_async_events(CPUState *cs)
1097 return cs->halted;
1100 static int s390_kvm_irq_to_interrupt(struct kvm_s390_irq *irq,
1101 struct kvm_s390_interrupt *interrupt)
1103 int r = 0;
1105 interrupt->type = irq->type;
1106 switch (irq->type) {
1107 case KVM_S390_INT_VIRTIO:
1108 interrupt->parm = irq->u.ext.ext_params;
1109 /* fall through */
1110 case KVM_S390_INT_PFAULT_INIT:
1111 case KVM_S390_INT_PFAULT_DONE:
1112 interrupt->parm64 = irq->u.ext.ext_params2;
1113 break;
1114 case KVM_S390_PROGRAM_INT:
1115 interrupt->parm = irq->u.pgm.code;
1116 break;
1117 case KVM_S390_SIGP_SET_PREFIX:
1118 interrupt->parm = irq->u.prefix.address;
1119 break;
1120 case KVM_S390_INT_SERVICE:
1121 interrupt->parm = irq->u.ext.ext_params;
1122 break;
1123 case KVM_S390_MCHK:
1124 interrupt->parm = irq->u.mchk.cr14;
1125 interrupt->parm64 = irq->u.mchk.mcic;
1126 break;
1127 case KVM_S390_INT_EXTERNAL_CALL:
1128 interrupt->parm = irq->u.extcall.code;
1129 break;
1130 case KVM_S390_INT_EMERGENCY:
1131 interrupt->parm = irq->u.emerg.code;
1132 break;
1133 case KVM_S390_SIGP_STOP:
1134 case KVM_S390_RESTART:
1135 break; /* These types have no parameters */
1136 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1137 interrupt->parm = irq->u.io.subchannel_id << 16;
1138 interrupt->parm |= irq->u.io.subchannel_nr;
1139 interrupt->parm64 = (uint64_t)irq->u.io.io_int_parm << 32;
1140 interrupt->parm64 |= irq->u.io.io_int_word;
1141 break;
1142 default:
1143 r = -EINVAL;
1144 break;
1146 return r;
1149 static void inject_vcpu_irq_legacy(CPUState *cs, struct kvm_s390_irq *irq)
1151 struct kvm_s390_interrupt kvmint = {};
1152 int r;
1154 r = s390_kvm_irq_to_interrupt(irq, &kvmint);
1155 if (r < 0) {
1156 fprintf(stderr, "%s called with bogus interrupt\n", __func__);
1157 exit(1);
1160 r = kvm_vcpu_ioctl(cs, KVM_S390_INTERRUPT, &kvmint);
1161 if (r < 0) {
1162 fprintf(stderr, "KVM failed to inject interrupt\n");
1163 exit(1);
1167 void kvm_s390_vcpu_interrupt(S390CPU *cpu, struct kvm_s390_irq *irq)
1169 CPUState *cs = CPU(cpu);
1170 int r;
1172 if (cap_s390_irq) {
1173 r = kvm_vcpu_ioctl(cs, KVM_S390_IRQ, irq);
1174 if (!r) {
1175 return;
1177 error_report("KVM failed to inject interrupt %llx", irq->type);
1178 exit(1);
1181 inject_vcpu_irq_legacy(cs, irq);
1184 void kvm_s390_floating_interrupt_legacy(struct kvm_s390_irq *irq)
1186 struct kvm_s390_interrupt kvmint = {};
1187 int r;
1189 r = s390_kvm_irq_to_interrupt(irq, &kvmint);
1190 if (r < 0) {
1191 fprintf(stderr, "%s called with bogus interrupt\n", __func__);
1192 exit(1);
1195 r = kvm_vm_ioctl(kvm_state, KVM_S390_INTERRUPT, &kvmint);
1196 if (r < 0) {
1197 fprintf(stderr, "KVM failed to inject interrupt\n");
1198 exit(1);
1202 void kvm_s390_program_interrupt(S390CPU *cpu, uint16_t code)
1204 struct kvm_s390_irq irq = {
1205 .type = KVM_S390_PROGRAM_INT,
1206 .u.pgm.code = code,
1208 qemu_log_mask(CPU_LOG_INT, "program interrupt at %#" PRIx64 "\n",
1209 cpu->env.psw.addr);
1210 kvm_s390_vcpu_interrupt(cpu, &irq);
1213 void kvm_s390_access_exception(S390CPU *cpu, uint16_t code, uint64_t te_code)
1215 struct kvm_s390_irq irq = {
1216 .type = KVM_S390_PROGRAM_INT,
1217 .u.pgm.code = code,
1218 .u.pgm.trans_exc_code = te_code,
1219 .u.pgm.exc_access_id = te_code & 3,
1222 kvm_s390_vcpu_interrupt(cpu, &irq);
1225 static void kvm_sclp_service_call(S390CPU *cpu, struct kvm_run *run,
1226 uint16_t ipbh0)
1228 CPUS390XState *env = &cpu->env;
1229 uint64_t sccb;
1230 uint32_t code;
1231 int r;
1233 sccb = env->regs[ipbh0 & 0xf];
1234 code = env->regs[(ipbh0 & 0xf0) >> 4];
1236 switch (run->s390_sieic.icptcode) {
1237 case ICPT_PV_INSTR_NOTIFICATION:
1238 g_assert(s390_is_pv());
1239 /* The notification intercepts are currently handled by KVM */
1240 error_report("unexpected SCLP PV notification");
1241 exit(1);
1242 break;
1243 case ICPT_PV_INSTR:
1244 g_assert(s390_is_pv());
1245 sclp_service_call_protected(env, sccb, code);
1246 /* Setting the CC is done by the Ultravisor. */
1247 break;
1248 case ICPT_INSTRUCTION:
1249 g_assert(!s390_is_pv());
1250 r = sclp_service_call(env, sccb, code);
1251 if (r < 0) {
1252 kvm_s390_program_interrupt(cpu, -r);
1253 return;
1255 setcc(cpu, r);
1259 static int handle_b2(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1261 CPUS390XState *env = &cpu->env;
1262 int rc = 0;
1263 uint16_t ipbh0 = (run->s390_sieic.ipb & 0xffff0000) >> 16;
1265 switch (ipa1) {
1266 case PRIV_B2_XSCH:
1267 ioinst_handle_xsch(cpu, env->regs[1], RA_IGNORED);
1268 break;
1269 case PRIV_B2_CSCH:
1270 ioinst_handle_csch(cpu, env->regs[1], RA_IGNORED);
1271 break;
1272 case PRIV_B2_HSCH:
1273 ioinst_handle_hsch(cpu, env->regs[1], RA_IGNORED);
1274 break;
1275 case PRIV_B2_MSCH:
1276 ioinst_handle_msch(cpu, env->regs[1], run->s390_sieic.ipb, RA_IGNORED);
1277 break;
1278 case PRIV_B2_SSCH:
1279 ioinst_handle_ssch(cpu, env->regs[1], run->s390_sieic.ipb, RA_IGNORED);
1280 break;
1281 case PRIV_B2_STCRW:
1282 ioinst_handle_stcrw(cpu, run->s390_sieic.ipb, RA_IGNORED);
1283 break;
1284 case PRIV_B2_STSCH:
1285 ioinst_handle_stsch(cpu, env->regs[1], run->s390_sieic.ipb, RA_IGNORED);
1286 break;
1287 case PRIV_B2_TSCH:
1288 /* We should only get tsch via KVM_EXIT_S390_TSCH. */
1289 fprintf(stderr, "Spurious tsch intercept\n");
1290 break;
1291 case PRIV_B2_CHSC:
1292 ioinst_handle_chsc(cpu, run->s390_sieic.ipb, RA_IGNORED);
1293 break;
1294 case PRIV_B2_TPI:
1295 /* This should have been handled by kvm already. */
1296 fprintf(stderr, "Spurious tpi intercept\n");
1297 break;
1298 case PRIV_B2_SCHM:
1299 ioinst_handle_schm(cpu, env->regs[1], env->regs[2],
1300 run->s390_sieic.ipb, RA_IGNORED);
1301 break;
1302 case PRIV_B2_RSCH:
1303 ioinst_handle_rsch(cpu, env->regs[1], RA_IGNORED);
1304 break;
1305 case PRIV_B2_RCHP:
1306 ioinst_handle_rchp(cpu, env->regs[1], RA_IGNORED);
1307 break;
1308 case PRIV_B2_STCPS:
1309 /* We do not provide this instruction, it is suppressed. */
1310 break;
1311 case PRIV_B2_SAL:
1312 ioinst_handle_sal(cpu, env->regs[1], RA_IGNORED);
1313 break;
1314 case PRIV_B2_SIGA:
1315 /* Not provided, set CC = 3 for subchannel not operational */
1316 setcc(cpu, 3);
1317 break;
1318 case PRIV_B2_SCLP_CALL:
1319 kvm_sclp_service_call(cpu, run, ipbh0);
1320 break;
1321 default:
1322 rc = -1;
1323 DPRINTF("KVM: unhandled PRIV: 0xb2%x\n", ipa1);
1324 break;
1327 return rc;
1330 static uint64_t get_base_disp_rxy(S390CPU *cpu, struct kvm_run *run,
1331 uint8_t *ar)
1333 CPUS390XState *env = &cpu->env;
1334 uint32_t x2 = (run->s390_sieic.ipa & 0x000f);
1335 uint32_t base2 = run->s390_sieic.ipb >> 28;
1336 uint32_t disp2 = ((run->s390_sieic.ipb & 0x0fff0000) >> 16) +
1337 ((run->s390_sieic.ipb & 0xff00) << 4);
1339 if (disp2 & 0x80000) {
1340 disp2 += 0xfff00000;
1342 if (ar) {
1343 *ar = base2;
1346 return (base2 ? env->regs[base2] : 0) +
1347 (x2 ? env->regs[x2] : 0) + (long)(int)disp2;
1350 static uint64_t get_base_disp_rsy(S390CPU *cpu, struct kvm_run *run,
1351 uint8_t *ar)
1353 CPUS390XState *env = &cpu->env;
1354 uint32_t base2 = run->s390_sieic.ipb >> 28;
1355 uint32_t disp2 = ((run->s390_sieic.ipb & 0x0fff0000) >> 16) +
1356 ((run->s390_sieic.ipb & 0xff00) << 4);
1358 if (disp2 & 0x80000) {
1359 disp2 += 0xfff00000;
1361 if (ar) {
1362 *ar = base2;
1365 return (base2 ? env->regs[base2] : 0) + (long)(int)disp2;
1368 static int kvm_clp_service_call(S390CPU *cpu, struct kvm_run *run)
1370 uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1372 if (s390_has_feat(S390_FEAT_ZPCI)) {
1373 return clp_service_call(cpu, r2, RA_IGNORED);
1374 } else {
1375 return -1;
1379 static int kvm_pcilg_service_call(S390CPU *cpu, struct kvm_run *run)
1381 uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1382 uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1384 if (s390_has_feat(S390_FEAT_ZPCI)) {
1385 return pcilg_service_call(cpu, r1, r2, RA_IGNORED);
1386 } else {
1387 return -1;
1391 static int kvm_pcistg_service_call(S390CPU *cpu, struct kvm_run *run)
1393 uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1394 uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1396 if (s390_has_feat(S390_FEAT_ZPCI)) {
1397 return pcistg_service_call(cpu, r1, r2, RA_IGNORED);
1398 } else {
1399 return -1;
1403 static int kvm_stpcifc_service_call(S390CPU *cpu, struct kvm_run *run)
1405 uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1406 uint64_t fiba;
1407 uint8_t ar;
1409 if (s390_has_feat(S390_FEAT_ZPCI)) {
1410 fiba = get_base_disp_rxy(cpu, run, &ar);
1412 return stpcifc_service_call(cpu, r1, fiba, ar, RA_IGNORED);
1413 } else {
1414 return -1;
1418 static int kvm_sic_service_call(S390CPU *cpu, struct kvm_run *run)
1420 CPUS390XState *env = &cpu->env;
1421 uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1422 uint8_t r3 = run->s390_sieic.ipa & 0x000f;
1423 uint8_t isc;
1424 uint16_t mode;
1425 int r;
1427 mode = env->regs[r1] & 0xffff;
1428 isc = (env->regs[r3] >> 27) & 0x7;
1429 r = css_do_sic(env, isc, mode);
1430 if (r) {
1431 kvm_s390_program_interrupt(cpu, -r);
1434 return 0;
1437 static int kvm_rpcit_service_call(S390CPU *cpu, struct kvm_run *run)
1439 uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1440 uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1442 if (s390_has_feat(S390_FEAT_ZPCI)) {
1443 return rpcit_service_call(cpu, r1, r2, RA_IGNORED);
1444 } else {
1445 return -1;
1449 static int kvm_pcistb_service_call(S390CPU *cpu, struct kvm_run *run)
1451 uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1452 uint8_t r3 = run->s390_sieic.ipa & 0x000f;
1453 uint64_t gaddr;
1454 uint8_t ar;
1456 if (s390_has_feat(S390_FEAT_ZPCI)) {
1457 gaddr = get_base_disp_rsy(cpu, run, &ar);
1459 return pcistb_service_call(cpu, r1, r3, gaddr, ar, RA_IGNORED);
1460 } else {
1461 return -1;
1465 static int kvm_mpcifc_service_call(S390CPU *cpu, struct kvm_run *run)
1467 uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1468 uint64_t fiba;
1469 uint8_t ar;
1471 if (s390_has_feat(S390_FEAT_ZPCI)) {
1472 fiba = get_base_disp_rxy(cpu, run, &ar);
1474 return mpcifc_service_call(cpu, r1, fiba, ar, RA_IGNORED);
1475 } else {
1476 return -1;
1480 static int handle_b9(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1482 int r = 0;
1484 switch (ipa1) {
1485 case PRIV_B9_CLP:
1486 r = kvm_clp_service_call(cpu, run);
1487 break;
1488 case PRIV_B9_PCISTG:
1489 r = kvm_pcistg_service_call(cpu, run);
1490 break;
1491 case PRIV_B9_PCILG:
1492 r = kvm_pcilg_service_call(cpu, run);
1493 break;
1494 case PRIV_B9_RPCIT:
1495 r = kvm_rpcit_service_call(cpu, run);
1496 break;
1497 case PRIV_B9_EQBS:
1498 /* just inject exception */
1499 r = -1;
1500 break;
1501 default:
1502 r = -1;
1503 DPRINTF("KVM: unhandled PRIV: 0xb9%x\n", ipa1);
1504 break;
1507 return r;
1510 static int handle_eb(S390CPU *cpu, struct kvm_run *run, uint8_t ipbl)
1512 int r = 0;
1514 switch (ipbl) {
1515 case PRIV_EB_PCISTB:
1516 r = kvm_pcistb_service_call(cpu, run);
1517 break;
1518 case PRIV_EB_SIC:
1519 r = kvm_sic_service_call(cpu, run);
1520 break;
1521 case PRIV_EB_SQBS:
1522 /* just inject exception */
1523 r = -1;
1524 break;
1525 default:
1526 r = -1;
1527 DPRINTF("KVM: unhandled PRIV: 0xeb%x\n", ipbl);
1528 break;
1531 return r;
1534 static int handle_e3(S390CPU *cpu, struct kvm_run *run, uint8_t ipbl)
1536 int r = 0;
1538 switch (ipbl) {
1539 case PRIV_E3_MPCIFC:
1540 r = kvm_mpcifc_service_call(cpu, run);
1541 break;
1542 case PRIV_E3_STPCIFC:
1543 r = kvm_stpcifc_service_call(cpu, run);
1544 break;
1545 default:
1546 r = -1;
1547 DPRINTF("KVM: unhandled PRIV: 0xe3%x\n", ipbl);
1548 break;
1551 return r;
1554 static int handle_hypercall(S390CPU *cpu, struct kvm_run *run)
1556 CPUS390XState *env = &cpu->env;
1557 int ret;
1559 ret = s390_virtio_hypercall(env);
1560 if (ret == -EINVAL) {
1561 kvm_s390_program_interrupt(cpu, PGM_SPECIFICATION);
1562 return 0;
1565 return ret;
1568 static void kvm_handle_diag_288(S390CPU *cpu, struct kvm_run *run)
1570 uint64_t r1, r3;
1571 int rc;
1573 r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1574 r3 = run->s390_sieic.ipa & 0x000f;
1575 rc = handle_diag_288(&cpu->env, r1, r3);
1576 if (rc) {
1577 kvm_s390_program_interrupt(cpu, PGM_SPECIFICATION);
1581 static void kvm_handle_diag_308(S390CPU *cpu, struct kvm_run *run)
1583 uint64_t r1, r3;
1585 r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1586 r3 = run->s390_sieic.ipa & 0x000f;
1587 handle_diag_308(&cpu->env, r1, r3, RA_IGNORED);
1590 static int handle_sw_breakpoint(S390CPU *cpu, struct kvm_run *run)
1592 CPUS390XState *env = &cpu->env;
1593 unsigned long pc;
1595 pc = env->psw.addr - sw_bp_ilen;
1596 if (kvm_find_sw_breakpoint(CPU(cpu), pc)) {
1597 env->psw.addr = pc;
1598 return EXCP_DEBUG;
1601 return -ENOENT;
1604 #define DIAG_KVM_CODE_MASK 0x000000000000ffff
1606 static int handle_diag(S390CPU *cpu, struct kvm_run *run, uint32_t ipb)
1608 int r = 0;
1609 uint16_t func_code;
1612 * For any diagnose call we support, bits 48-63 of the resulting
1613 * address specify the function code; the remainder is ignored.
1615 func_code = decode_basedisp_rs(&cpu->env, ipb, NULL) & DIAG_KVM_CODE_MASK;
1616 switch (func_code) {
1617 case DIAG_TIMEREVENT:
1618 kvm_handle_diag_288(cpu, run);
1619 break;
1620 case DIAG_IPL:
1621 kvm_handle_diag_308(cpu, run);
1622 break;
1623 case DIAG_KVM_HYPERCALL:
1624 r = handle_hypercall(cpu, run);
1625 break;
1626 case DIAG_KVM_BREAKPOINT:
1627 r = handle_sw_breakpoint(cpu, run);
1628 break;
1629 default:
1630 DPRINTF("KVM: unknown DIAG: 0x%x\n", func_code);
1631 kvm_s390_program_interrupt(cpu, PGM_SPECIFICATION);
1632 break;
1635 return r;
1638 static int kvm_s390_handle_sigp(S390CPU *cpu, uint8_t ipa1, uint32_t ipb)
1640 CPUS390XState *env = &cpu->env;
1641 const uint8_t r1 = ipa1 >> 4;
1642 const uint8_t r3 = ipa1 & 0x0f;
1643 int ret;
1644 uint8_t order;
1646 /* get order code */
1647 order = decode_basedisp_rs(env, ipb, NULL) & SIGP_ORDER_MASK;
1649 ret = handle_sigp(env, order, r1, r3);
1650 setcc(cpu, ret);
1651 return 0;
1654 static int handle_instruction(S390CPU *cpu, struct kvm_run *run)
1656 unsigned int ipa0 = (run->s390_sieic.ipa & 0xff00);
1657 uint8_t ipa1 = run->s390_sieic.ipa & 0x00ff;
1658 int r = -1;
1660 DPRINTF("handle_instruction 0x%x 0x%x\n",
1661 run->s390_sieic.ipa, run->s390_sieic.ipb);
1662 switch (ipa0) {
1663 case IPA0_B2:
1664 r = handle_b2(cpu, run, ipa1);
1665 break;
1666 case IPA0_B9:
1667 r = handle_b9(cpu, run, ipa1);
1668 break;
1669 case IPA0_EB:
1670 r = handle_eb(cpu, run, run->s390_sieic.ipb & 0xff);
1671 break;
1672 case IPA0_E3:
1673 r = handle_e3(cpu, run, run->s390_sieic.ipb & 0xff);
1674 break;
1675 case IPA0_DIAG:
1676 r = handle_diag(cpu, run, run->s390_sieic.ipb);
1677 break;
1678 case IPA0_SIGP:
1679 r = kvm_s390_handle_sigp(cpu, ipa1, run->s390_sieic.ipb);
1680 break;
1683 if (r < 0) {
1684 r = 0;
1685 kvm_s390_program_interrupt(cpu, PGM_OPERATION);
1688 return r;
1691 static void unmanageable_intercept(S390CPU *cpu, S390CrashReason reason,
1692 int pswoffset)
1694 CPUState *cs = CPU(cpu);
1696 s390_cpu_halt(cpu);
1697 cpu->env.crash_reason = reason;
1698 qemu_system_guest_panicked(cpu_get_crash_info(cs));
1701 /* try to detect pgm check loops */
1702 static int handle_oper_loop(S390CPU *cpu, struct kvm_run *run)
1704 CPUState *cs = CPU(cpu);
1705 PSW oldpsw, newpsw;
1707 newpsw.mask = ldq_phys(cs->as, cpu->env.psa +
1708 offsetof(LowCore, program_new_psw));
1709 newpsw.addr = ldq_phys(cs->as, cpu->env.psa +
1710 offsetof(LowCore, program_new_psw) + 8);
1711 oldpsw.mask = run->psw_mask;
1712 oldpsw.addr = run->psw_addr;
1714 * Avoid endless loops of operation exceptions, if the pgm new
1715 * PSW will cause a new operation exception.
1716 * The heuristic checks if the pgm new psw is within 6 bytes before
1717 * the faulting psw address (with same DAT, AS settings) and the
1718 * new psw is not a wait psw and the fault was not triggered by
1719 * problem state. In that case go into crashed state.
1722 if (oldpsw.addr - newpsw.addr <= 6 &&
1723 !(newpsw.mask & PSW_MASK_WAIT) &&
1724 !(oldpsw.mask & PSW_MASK_PSTATE) &&
1725 (newpsw.mask & PSW_MASK_ASC) == (oldpsw.mask & PSW_MASK_ASC) &&
1726 (newpsw.mask & PSW_MASK_DAT) == (oldpsw.mask & PSW_MASK_DAT)) {
1727 unmanageable_intercept(cpu, S390_CRASH_REASON_OPINT_LOOP,
1728 offsetof(LowCore, program_new_psw));
1729 return EXCP_HALTED;
1731 return 0;
1734 static int handle_intercept(S390CPU *cpu)
1736 CPUState *cs = CPU(cpu);
1737 struct kvm_run *run = cs->kvm_run;
1738 int icpt_code = run->s390_sieic.icptcode;
1739 int r = 0;
1741 DPRINTF("intercept: 0x%x (at 0x%lx)\n", icpt_code,
1742 (long)cs->kvm_run->psw_addr);
1743 switch (icpt_code) {
1744 case ICPT_INSTRUCTION:
1745 case ICPT_PV_INSTR:
1746 case ICPT_PV_INSTR_NOTIFICATION:
1747 r = handle_instruction(cpu, run);
1748 break;
1749 case ICPT_PROGRAM:
1750 unmanageable_intercept(cpu, S390_CRASH_REASON_PGMINT_LOOP,
1751 offsetof(LowCore, program_new_psw));
1752 r = EXCP_HALTED;
1753 break;
1754 case ICPT_EXT_INT:
1755 unmanageable_intercept(cpu, S390_CRASH_REASON_EXTINT_LOOP,
1756 offsetof(LowCore, external_new_psw));
1757 r = EXCP_HALTED;
1758 break;
1759 case ICPT_WAITPSW:
1760 /* disabled wait, since enabled wait is handled in kernel */
1761 s390_handle_wait(cpu);
1762 r = EXCP_HALTED;
1763 break;
1764 case ICPT_CPU_STOP:
1765 do_stop_interrupt(&cpu->env);
1766 r = EXCP_HALTED;
1767 break;
1768 case ICPT_OPEREXC:
1769 /* check for break points */
1770 r = handle_sw_breakpoint(cpu, run);
1771 if (r == -ENOENT) {
1772 /* Then check for potential pgm check loops */
1773 r = handle_oper_loop(cpu, run);
1774 if (r == 0) {
1775 kvm_s390_program_interrupt(cpu, PGM_OPERATION);
1778 break;
1779 case ICPT_SOFT_INTERCEPT:
1780 fprintf(stderr, "KVM unimplemented icpt SOFT\n");
1781 exit(1);
1782 break;
1783 case ICPT_IO:
1784 fprintf(stderr, "KVM unimplemented icpt IO\n");
1785 exit(1);
1786 break;
1787 default:
1788 fprintf(stderr, "Unknown intercept code: %d\n", icpt_code);
1789 exit(1);
1790 break;
1793 return r;
1796 static int handle_tsch(S390CPU *cpu)
1798 CPUState *cs = CPU(cpu);
1799 struct kvm_run *run = cs->kvm_run;
1800 int ret;
1802 ret = ioinst_handle_tsch(cpu, cpu->env.regs[1], run->s390_tsch.ipb,
1803 RA_IGNORED);
1804 if (ret < 0) {
1806 * Failure.
1807 * If an I/O interrupt had been dequeued, we have to reinject it.
1809 if (run->s390_tsch.dequeued) {
1810 s390_io_interrupt(run->s390_tsch.subchannel_id,
1811 run->s390_tsch.subchannel_nr,
1812 run->s390_tsch.io_int_parm,
1813 run->s390_tsch.io_int_word);
1815 ret = 0;
1817 return ret;
1820 static void insert_stsi_3_2_2(S390CPU *cpu, __u64 addr, uint8_t ar)
1822 const MachineState *ms = MACHINE(qdev_get_machine());
1823 uint16_t conf_cpus = 0, reserved_cpus = 0;
1824 SysIB_322 sysib;
1825 int del, i;
1827 if (s390_is_pv()) {
1828 s390_cpu_pv_mem_read(cpu, 0, &sysib, sizeof(sysib));
1829 } else if (s390_cpu_virt_mem_read(cpu, addr, ar, &sysib, sizeof(sysib))) {
1830 return;
1832 /* Shift the stack of Extended Names to prepare for our own data */
1833 memmove(&sysib.ext_names[1], &sysib.ext_names[0],
1834 sizeof(sysib.ext_names[0]) * (sysib.count - 1));
1835 /* First virt level, that doesn't provide Ext Names delimits stack. It is
1836 * assumed it's not capable of managing Extended Names for lower levels.
1838 for (del = 1; del < sysib.count; del++) {
1839 if (!sysib.vm[del].ext_name_encoding || !sysib.ext_names[del][0]) {
1840 break;
1843 if (del < sysib.count) {
1844 memset(sysib.ext_names[del], 0,
1845 sizeof(sysib.ext_names[0]) * (sysib.count - del));
1848 /* count the cpus and split them into configured and reserved ones */
1849 for (i = 0; i < ms->possible_cpus->len; i++) {
1850 if (ms->possible_cpus->cpus[i].cpu) {
1851 conf_cpus++;
1852 } else {
1853 reserved_cpus++;
1856 sysib.vm[0].total_cpus = conf_cpus + reserved_cpus;
1857 sysib.vm[0].conf_cpus = conf_cpus;
1858 sysib.vm[0].reserved_cpus = reserved_cpus;
1860 /* Insert short machine name in EBCDIC, padded with blanks */
1861 if (qemu_name) {
1862 memset(sysib.vm[0].name, 0x40, sizeof(sysib.vm[0].name));
1863 ebcdic_put(sysib.vm[0].name, qemu_name, MIN(sizeof(sysib.vm[0].name),
1864 strlen(qemu_name)));
1866 sysib.vm[0].ext_name_encoding = 2; /* 2 = UTF-8 */
1867 memset(sysib.ext_names[0], 0, sizeof(sysib.ext_names[0]));
1868 /* If hypervisor specifies zero Extended Name in STSI322 SYSIB, it's
1869 * considered by s390 as not capable of providing any Extended Name.
1870 * Therefore if no name was specified on qemu invocation, we go with the
1871 * same "KVMguest" default, which KVM has filled into short name field.
1873 if (qemu_name) {
1874 strncpy((char *)sysib.ext_names[0], qemu_name,
1875 sizeof(sysib.ext_names[0]));
1876 } else {
1877 strcpy((char *)sysib.ext_names[0], "KVMguest");
1879 /* Insert UUID */
1880 memcpy(sysib.vm[0].uuid, &qemu_uuid, sizeof(sysib.vm[0].uuid));
1882 if (s390_is_pv()) {
1883 s390_cpu_pv_mem_write(cpu, 0, &sysib, sizeof(sysib));
1884 } else {
1885 s390_cpu_virt_mem_write(cpu, addr, ar, &sysib, sizeof(sysib));
1889 static int handle_stsi(S390CPU *cpu)
1891 CPUState *cs = CPU(cpu);
1892 struct kvm_run *run = cs->kvm_run;
1894 switch (run->s390_stsi.fc) {
1895 case 3:
1896 if (run->s390_stsi.sel1 != 2 || run->s390_stsi.sel2 != 2) {
1897 return 0;
1899 /* Only sysib 3.2.2 needs post-handling for now. */
1900 insert_stsi_3_2_2(cpu, run->s390_stsi.addr, run->s390_stsi.ar);
1901 return 0;
1902 default:
1903 return 0;
1907 static int kvm_arch_handle_debug_exit(S390CPU *cpu)
1909 CPUState *cs = CPU(cpu);
1910 struct kvm_run *run = cs->kvm_run;
1912 int ret = 0;
1913 struct kvm_debug_exit_arch *arch_info = &run->debug.arch;
1915 switch (arch_info->type) {
1916 case KVM_HW_WP_WRITE:
1917 if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
1918 cs->watchpoint_hit = &hw_watchpoint;
1919 hw_watchpoint.vaddr = arch_info->addr;
1920 hw_watchpoint.flags = BP_MEM_WRITE;
1921 ret = EXCP_DEBUG;
1923 break;
1924 case KVM_HW_BP:
1925 if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
1926 ret = EXCP_DEBUG;
1928 break;
1929 case KVM_SINGLESTEP:
1930 if (cs->singlestep_enabled) {
1931 ret = EXCP_DEBUG;
1933 break;
1934 default:
1935 ret = -ENOSYS;
1938 return ret;
1941 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1943 S390CPU *cpu = S390_CPU(cs);
1944 int ret = 0;
1946 qemu_mutex_lock_iothread();
1948 kvm_cpu_synchronize_state(cs);
1950 switch (run->exit_reason) {
1951 case KVM_EXIT_S390_SIEIC:
1952 ret = handle_intercept(cpu);
1953 break;
1954 case KVM_EXIT_S390_RESET:
1955 s390_ipl_reset_request(cs, S390_RESET_REIPL);
1956 break;
1957 case KVM_EXIT_S390_TSCH:
1958 ret = handle_tsch(cpu);
1959 break;
1960 case KVM_EXIT_S390_STSI:
1961 ret = handle_stsi(cpu);
1962 break;
1963 case KVM_EXIT_DEBUG:
1964 ret = kvm_arch_handle_debug_exit(cpu);
1965 break;
1966 default:
1967 fprintf(stderr, "Unknown KVM exit: %d\n", run->exit_reason);
1968 break;
1970 qemu_mutex_unlock_iothread();
1972 if (ret == 0) {
1973 ret = EXCP_INTERRUPT;
1975 return ret;
1978 bool kvm_arch_stop_on_emulation_error(CPUState *cpu)
1980 return true;
1983 void kvm_s390_enable_css_support(S390CPU *cpu)
1985 int r;
1987 /* Activate host kernel channel subsystem support. */
1988 r = kvm_vcpu_enable_cap(CPU(cpu), KVM_CAP_S390_CSS_SUPPORT, 0);
1989 assert(r == 0);
1992 void kvm_arch_init_irq_routing(KVMState *s)
1995 * Note that while irqchip capabilities generally imply that cpustates
1996 * are handled in-kernel, it is not true for s390 (yet); therefore, we
1997 * have to override the common code kvm_halt_in_kernel_allowed setting.
1999 if (kvm_check_extension(s, KVM_CAP_IRQ_ROUTING)) {
2000 kvm_gsi_routing_allowed = true;
2001 kvm_halt_in_kernel_allowed = false;
2005 int kvm_s390_assign_subch_ioeventfd(EventNotifier *notifier, uint32_t sch,
2006 int vq, bool assign)
2008 struct kvm_ioeventfd kick = {
2009 .flags = KVM_IOEVENTFD_FLAG_VIRTIO_CCW_NOTIFY |
2010 KVM_IOEVENTFD_FLAG_DATAMATCH,
2011 .fd = event_notifier_get_fd(notifier),
2012 .datamatch = vq,
2013 .addr = sch,
2014 .len = 8,
2016 trace_kvm_assign_subch_ioeventfd(kick.fd, kick.addr, assign,
2017 kick.datamatch);
2018 if (!kvm_check_extension(kvm_state, KVM_CAP_IOEVENTFD)) {
2019 return -ENOSYS;
2021 if (!assign) {
2022 kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
2024 return kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
2027 int kvm_s390_get_ri(void)
2029 return cap_ri;
2032 int kvm_s390_get_gs(void)
2034 return cap_gs;
2037 int kvm_s390_set_cpu_state(S390CPU *cpu, uint8_t cpu_state)
2039 struct kvm_mp_state mp_state = {};
2040 int ret;
2042 /* the kvm part might not have been initialized yet */
2043 if (CPU(cpu)->kvm_state == NULL) {
2044 return 0;
2047 switch (cpu_state) {
2048 case S390_CPU_STATE_STOPPED:
2049 mp_state.mp_state = KVM_MP_STATE_STOPPED;
2050 break;
2051 case S390_CPU_STATE_CHECK_STOP:
2052 mp_state.mp_state = KVM_MP_STATE_CHECK_STOP;
2053 break;
2054 case S390_CPU_STATE_OPERATING:
2055 mp_state.mp_state = KVM_MP_STATE_OPERATING;
2056 break;
2057 case S390_CPU_STATE_LOAD:
2058 mp_state.mp_state = KVM_MP_STATE_LOAD;
2059 break;
2060 default:
2061 error_report("Requested CPU state is not a valid S390 CPU state: %u",
2062 cpu_state);
2063 exit(1);
2066 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
2067 if (ret) {
2068 trace_kvm_failed_cpu_state_set(CPU(cpu)->cpu_index, cpu_state,
2069 strerror(-ret));
2072 return ret;
2075 void kvm_s390_vcpu_interrupt_pre_save(S390CPU *cpu)
2077 unsigned int max_cpus = MACHINE(qdev_get_machine())->smp.max_cpus;
2078 struct kvm_s390_irq_state irq_state = {
2079 .buf = (uint64_t) cpu->irqstate,
2080 .len = VCPU_IRQ_BUF_SIZE(max_cpus),
2082 CPUState *cs = CPU(cpu);
2083 int32_t bytes;
2085 if (!kvm_check_extension(kvm_state, KVM_CAP_S390_IRQ_STATE)) {
2086 return;
2089 bytes = kvm_vcpu_ioctl(cs, KVM_S390_GET_IRQ_STATE, &irq_state);
2090 if (bytes < 0) {
2091 cpu->irqstate_saved_size = 0;
2092 error_report("Migration of interrupt state failed");
2093 return;
2096 cpu->irqstate_saved_size = bytes;
2099 int kvm_s390_vcpu_interrupt_post_load(S390CPU *cpu)
2101 CPUState *cs = CPU(cpu);
2102 struct kvm_s390_irq_state irq_state = {
2103 .buf = (uint64_t) cpu->irqstate,
2104 .len = cpu->irqstate_saved_size,
2106 int r;
2108 if (cpu->irqstate_saved_size == 0) {
2109 return 0;
2112 if (!kvm_check_extension(kvm_state, KVM_CAP_S390_IRQ_STATE)) {
2113 return -ENOSYS;
2116 r = kvm_vcpu_ioctl(cs, KVM_S390_SET_IRQ_STATE, &irq_state);
2117 if (r) {
2118 error_report("Setting interrupt state failed %d", r);
2120 return r;
2123 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
2124 uint64_t address, uint32_t data, PCIDevice *dev)
2126 S390PCIBusDevice *pbdev;
2127 uint32_t vec = data & ZPCI_MSI_VEC_MASK;
2129 if (!dev) {
2130 DPRINTF("add_msi_route no pci device\n");
2131 return -ENODEV;
2134 pbdev = s390_pci_find_dev_by_target(s390_get_phb(), DEVICE(dev)->id);
2135 if (!pbdev) {
2136 DPRINTF("add_msi_route no zpci device\n");
2137 return -ENODEV;
2140 route->type = KVM_IRQ_ROUTING_S390_ADAPTER;
2141 route->flags = 0;
2142 route->u.adapter.summary_addr = pbdev->routes.adapter.summary_addr;
2143 route->u.adapter.ind_addr = pbdev->routes.adapter.ind_addr;
2144 route->u.adapter.summary_offset = pbdev->routes.adapter.summary_offset;
2145 route->u.adapter.ind_offset = pbdev->routes.adapter.ind_offset + vec;
2146 route->u.adapter.adapter_id = pbdev->routes.adapter.adapter_id;
2147 return 0;
2150 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
2151 int vector, PCIDevice *dev)
2153 return 0;
2156 int kvm_arch_release_virq_post(int virq)
2158 return 0;
2161 int kvm_arch_msi_data_to_gsi(uint32_t data)
2163 abort();
2166 static int query_cpu_subfunc(S390FeatBitmap features)
2168 struct kvm_s390_vm_cpu_subfunc prop = {};
2169 struct kvm_device_attr attr = {
2170 .group = KVM_S390_VM_CPU_MODEL,
2171 .attr = KVM_S390_VM_CPU_MACHINE_SUBFUNC,
2172 .addr = (uint64_t) &prop,
2174 int rc;
2176 rc = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
2177 if (rc) {
2178 return rc;
2182 * We're going to add all subfunctions now, if the corresponding feature
2183 * is available that unlocks the query functions.
2185 s390_add_from_feat_block(features, S390_FEAT_TYPE_PLO, prop.plo);
2186 if (test_bit(S390_FEAT_TOD_CLOCK_STEERING, features)) {
2187 s390_add_from_feat_block(features, S390_FEAT_TYPE_PTFF, prop.ptff);
2189 if (test_bit(S390_FEAT_MSA, features)) {
2190 s390_add_from_feat_block(features, S390_FEAT_TYPE_KMAC, prop.kmac);
2191 s390_add_from_feat_block(features, S390_FEAT_TYPE_KMC, prop.kmc);
2192 s390_add_from_feat_block(features, S390_FEAT_TYPE_KM, prop.km);
2193 s390_add_from_feat_block(features, S390_FEAT_TYPE_KIMD, prop.kimd);
2194 s390_add_from_feat_block(features, S390_FEAT_TYPE_KLMD, prop.klmd);
2196 if (test_bit(S390_FEAT_MSA_EXT_3, features)) {
2197 s390_add_from_feat_block(features, S390_FEAT_TYPE_PCKMO, prop.pckmo);
2199 if (test_bit(S390_FEAT_MSA_EXT_4, features)) {
2200 s390_add_from_feat_block(features, S390_FEAT_TYPE_KMCTR, prop.kmctr);
2201 s390_add_from_feat_block(features, S390_FEAT_TYPE_KMF, prop.kmf);
2202 s390_add_from_feat_block(features, S390_FEAT_TYPE_KMO, prop.kmo);
2203 s390_add_from_feat_block(features, S390_FEAT_TYPE_PCC, prop.pcc);
2205 if (test_bit(S390_FEAT_MSA_EXT_5, features)) {
2206 s390_add_from_feat_block(features, S390_FEAT_TYPE_PPNO, prop.ppno);
2208 if (test_bit(S390_FEAT_MSA_EXT_8, features)) {
2209 s390_add_from_feat_block(features, S390_FEAT_TYPE_KMA, prop.kma);
2211 if (test_bit(S390_FEAT_MSA_EXT_9, features)) {
2212 s390_add_from_feat_block(features, S390_FEAT_TYPE_KDSA, prop.kdsa);
2214 if (test_bit(S390_FEAT_ESORT_BASE, features)) {
2215 s390_add_from_feat_block(features, S390_FEAT_TYPE_SORTL, prop.sortl);
2217 if (test_bit(S390_FEAT_DEFLATE_BASE, features)) {
2218 s390_add_from_feat_block(features, S390_FEAT_TYPE_DFLTCC, prop.dfltcc);
2220 return 0;
2223 static int configure_cpu_subfunc(const S390FeatBitmap features)
2225 struct kvm_s390_vm_cpu_subfunc prop = {};
2226 struct kvm_device_attr attr = {
2227 .group = KVM_S390_VM_CPU_MODEL,
2228 .attr = KVM_S390_VM_CPU_PROCESSOR_SUBFUNC,
2229 .addr = (uint64_t) &prop,
2232 if (!kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2233 KVM_S390_VM_CPU_PROCESSOR_SUBFUNC)) {
2234 /* hardware support might be missing, IBC will handle most of this */
2235 return 0;
2238 s390_fill_feat_block(features, S390_FEAT_TYPE_PLO, prop.plo);
2239 if (test_bit(S390_FEAT_TOD_CLOCK_STEERING, features)) {
2240 s390_fill_feat_block(features, S390_FEAT_TYPE_PTFF, prop.ptff);
2242 if (test_bit(S390_FEAT_MSA, features)) {
2243 s390_fill_feat_block(features, S390_FEAT_TYPE_KMAC, prop.kmac);
2244 s390_fill_feat_block(features, S390_FEAT_TYPE_KMC, prop.kmc);
2245 s390_fill_feat_block(features, S390_FEAT_TYPE_KM, prop.km);
2246 s390_fill_feat_block(features, S390_FEAT_TYPE_KIMD, prop.kimd);
2247 s390_fill_feat_block(features, S390_FEAT_TYPE_KLMD, prop.klmd);
2249 if (test_bit(S390_FEAT_MSA_EXT_3, features)) {
2250 s390_fill_feat_block(features, S390_FEAT_TYPE_PCKMO, prop.pckmo);
2252 if (test_bit(S390_FEAT_MSA_EXT_4, features)) {
2253 s390_fill_feat_block(features, S390_FEAT_TYPE_KMCTR, prop.kmctr);
2254 s390_fill_feat_block(features, S390_FEAT_TYPE_KMF, prop.kmf);
2255 s390_fill_feat_block(features, S390_FEAT_TYPE_KMO, prop.kmo);
2256 s390_fill_feat_block(features, S390_FEAT_TYPE_PCC, prop.pcc);
2258 if (test_bit(S390_FEAT_MSA_EXT_5, features)) {
2259 s390_fill_feat_block(features, S390_FEAT_TYPE_PPNO, prop.ppno);
2261 if (test_bit(S390_FEAT_MSA_EXT_8, features)) {
2262 s390_fill_feat_block(features, S390_FEAT_TYPE_KMA, prop.kma);
2264 if (test_bit(S390_FEAT_MSA_EXT_9, features)) {
2265 s390_fill_feat_block(features, S390_FEAT_TYPE_KDSA, prop.kdsa);
2267 if (test_bit(S390_FEAT_ESORT_BASE, features)) {
2268 s390_fill_feat_block(features, S390_FEAT_TYPE_SORTL, prop.sortl);
2270 if (test_bit(S390_FEAT_DEFLATE_BASE, features)) {
2271 s390_fill_feat_block(features, S390_FEAT_TYPE_DFLTCC, prop.dfltcc);
2273 return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
2276 static int kvm_to_feat[][2] = {
2277 { KVM_S390_VM_CPU_FEAT_ESOP, S390_FEAT_ESOP },
2278 { KVM_S390_VM_CPU_FEAT_SIEF2, S390_FEAT_SIE_F2 },
2279 { KVM_S390_VM_CPU_FEAT_64BSCAO , S390_FEAT_SIE_64BSCAO },
2280 { KVM_S390_VM_CPU_FEAT_SIIF, S390_FEAT_SIE_SIIF },
2281 { KVM_S390_VM_CPU_FEAT_GPERE, S390_FEAT_SIE_GPERE },
2282 { KVM_S390_VM_CPU_FEAT_GSLS, S390_FEAT_SIE_GSLS },
2283 { KVM_S390_VM_CPU_FEAT_IB, S390_FEAT_SIE_IB },
2284 { KVM_S390_VM_CPU_FEAT_CEI, S390_FEAT_SIE_CEI },
2285 { KVM_S390_VM_CPU_FEAT_IBS, S390_FEAT_SIE_IBS },
2286 { KVM_S390_VM_CPU_FEAT_SKEY, S390_FEAT_SIE_SKEY },
2287 { KVM_S390_VM_CPU_FEAT_CMMA, S390_FEAT_SIE_CMMA },
2288 { KVM_S390_VM_CPU_FEAT_PFMFI, S390_FEAT_SIE_PFMFI},
2289 { KVM_S390_VM_CPU_FEAT_SIGPIF, S390_FEAT_SIE_SIGPIF},
2290 { KVM_S390_VM_CPU_FEAT_KSS, S390_FEAT_SIE_KSS},
2293 static int query_cpu_feat(S390FeatBitmap features)
2295 struct kvm_s390_vm_cpu_feat prop = {};
2296 struct kvm_device_attr attr = {
2297 .group = KVM_S390_VM_CPU_MODEL,
2298 .attr = KVM_S390_VM_CPU_MACHINE_FEAT,
2299 .addr = (uint64_t) &prop,
2301 int rc;
2302 int i;
2304 rc = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
2305 if (rc) {
2306 return rc;
2309 for (i = 0; i < ARRAY_SIZE(kvm_to_feat); i++) {
2310 if (test_be_bit(kvm_to_feat[i][0], (uint8_t *) prop.feat)) {
2311 set_bit(kvm_to_feat[i][1], features);
2314 return 0;
2317 static int configure_cpu_feat(const S390FeatBitmap features)
2319 struct kvm_s390_vm_cpu_feat prop = {};
2320 struct kvm_device_attr attr = {
2321 .group = KVM_S390_VM_CPU_MODEL,
2322 .attr = KVM_S390_VM_CPU_PROCESSOR_FEAT,
2323 .addr = (uint64_t) &prop,
2325 int i;
2327 for (i = 0; i < ARRAY_SIZE(kvm_to_feat); i++) {
2328 if (test_bit(kvm_to_feat[i][1], features)) {
2329 set_be_bit(kvm_to_feat[i][0], (uint8_t *) prop.feat);
2332 return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
2335 bool kvm_s390_cpu_models_supported(void)
2337 if (!cpu_model_allowed()) {
2338 /* compatibility machines interfere with the cpu model */
2339 return false;
2341 return kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2342 KVM_S390_VM_CPU_MACHINE) &&
2343 kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2344 KVM_S390_VM_CPU_PROCESSOR) &&
2345 kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2346 KVM_S390_VM_CPU_MACHINE_FEAT) &&
2347 kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2348 KVM_S390_VM_CPU_PROCESSOR_FEAT) &&
2349 kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2350 KVM_S390_VM_CPU_MACHINE_SUBFUNC);
2353 void kvm_s390_get_host_cpu_model(S390CPUModel *model, Error **errp)
2355 struct kvm_s390_vm_cpu_machine prop = {};
2356 struct kvm_device_attr attr = {
2357 .group = KVM_S390_VM_CPU_MODEL,
2358 .attr = KVM_S390_VM_CPU_MACHINE,
2359 .addr = (uint64_t) &prop,
2361 uint16_t unblocked_ibc = 0, cpu_type = 0;
2362 int rc;
2364 memset(model, 0, sizeof(*model));
2366 if (!kvm_s390_cpu_models_supported()) {
2367 error_setg(errp, "KVM doesn't support CPU models");
2368 return;
2371 /* query the basic cpu model properties */
2372 rc = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
2373 if (rc) {
2374 error_setg(errp, "KVM: Error querying host CPU model: %d", rc);
2375 return;
2378 cpu_type = cpuid_type(prop.cpuid);
2379 if (has_ibc(prop.ibc)) {
2380 model->lowest_ibc = lowest_ibc(prop.ibc);
2381 unblocked_ibc = unblocked_ibc(prop.ibc);
2383 model->cpu_id = cpuid_id(prop.cpuid);
2384 model->cpu_id_format = cpuid_format(prop.cpuid);
2385 model->cpu_ver = 0xff;
2387 /* get supported cpu features indicated via STFL(E) */
2388 s390_add_from_feat_block(model->features, S390_FEAT_TYPE_STFL,
2389 (uint8_t *) prop.fac_mask);
2390 /* dat-enhancement facility 2 has no bit but was introduced with stfle */
2391 if (test_bit(S390_FEAT_STFLE, model->features)) {
2392 set_bit(S390_FEAT_DAT_ENH_2, model->features);
2394 /* get supported cpu features indicated e.g. via SCLP */
2395 rc = query_cpu_feat(model->features);
2396 if (rc) {
2397 error_setg(errp, "KVM: Error querying CPU features: %d", rc);
2398 return;
2400 /* get supported cpu subfunctions indicated via query / test bit */
2401 rc = query_cpu_subfunc(model->features);
2402 if (rc) {
2403 error_setg(errp, "KVM: Error querying CPU subfunctions: %d", rc);
2404 return;
2407 /* PTFF subfunctions might be indicated although kernel support missing */
2408 if (!test_bit(S390_FEAT_MULTIPLE_EPOCH, model->features)) {
2409 clear_bit(S390_FEAT_PTFF_QSIE, model->features);
2410 clear_bit(S390_FEAT_PTFF_QTOUE, model->features);
2411 clear_bit(S390_FEAT_PTFF_STOE, model->features);
2412 clear_bit(S390_FEAT_PTFF_STOUE, model->features);
2415 /* with cpu model support, CMM is only indicated if really available */
2416 if (kvm_s390_cmma_available()) {
2417 set_bit(S390_FEAT_CMM, model->features);
2418 } else {
2419 /* no cmm -> no cmm nt */
2420 clear_bit(S390_FEAT_CMM_NT, model->features);
2423 /* bpb needs kernel support for migration, VSIE and reset */
2424 if (!kvm_check_extension(kvm_state, KVM_CAP_S390_BPB)) {
2425 clear_bit(S390_FEAT_BPB, model->features);
2429 * If we have support for protected virtualization, indicate
2430 * the protected virtualization IPL unpack facility.
2432 if (cap_protected) {
2433 set_bit(S390_FEAT_UNPACK, model->features);
2436 /* We emulate a zPCI bus and AEN, therefore we don't need HW support */
2437 set_bit(S390_FEAT_ZPCI, model->features);
2438 set_bit(S390_FEAT_ADAPTER_EVENT_NOTIFICATION, model->features);
2440 if (s390_known_cpu_type(cpu_type)) {
2441 /* we want the exact model, even if some features are missing */
2442 model->def = s390_find_cpu_def(cpu_type, ibc_gen(unblocked_ibc),
2443 ibc_ec_ga(unblocked_ibc), NULL);
2444 } else {
2445 /* model unknown, e.g. too new - search using features */
2446 model->def = s390_find_cpu_def(0, ibc_gen(unblocked_ibc),
2447 ibc_ec_ga(unblocked_ibc),
2448 model->features);
2450 if (!model->def) {
2451 error_setg(errp, "KVM: host CPU model could not be identified");
2452 return;
2454 /* for now, we can only provide the AP feature with HW support */
2455 if (kvm_vm_check_attr(kvm_state, KVM_S390_VM_CRYPTO,
2456 KVM_S390_VM_CRYPTO_ENABLE_APIE)) {
2457 set_bit(S390_FEAT_AP, model->features);
2459 /* strip of features that are not part of the maximum model */
2460 bitmap_and(model->features, model->features, model->def->full_feat,
2461 S390_FEAT_MAX);
2464 static void kvm_s390_configure_apie(bool interpret)
2466 uint64_t attr = interpret ? KVM_S390_VM_CRYPTO_ENABLE_APIE :
2467 KVM_S390_VM_CRYPTO_DISABLE_APIE;
2469 if (kvm_vm_check_attr(kvm_state, KVM_S390_VM_CRYPTO, attr)) {
2470 kvm_s390_set_attr(attr);
2474 void kvm_s390_apply_cpu_model(const S390CPUModel *model, Error **errp)
2476 struct kvm_s390_vm_cpu_processor prop = {
2477 .fac_list = { 0 },
2479 struct kvm_device_attr attr = {
2480 .group = KVM_S390_VM_CPU_MODEL,
2481 .attr = KVM_S390_VM_CPU_PROCESSOR,
2482 .addr = (uint64_t) &prop,
2484 int rc;
2486 if (!model) {
2487 /* compatibility handling if cpu models are disabled */
2488 if (kvm_s390_cmma_available()) {
2489 kvm_s390_enable_cmma();
2491 return;
2493 if (!kvm_s390_cpu_models_supported()) {
2494 error_setg(errp, "KVM doesn't support CPU models");
2495 return;
2497 prop.cpuid = s390_cpuid_from_cpu_model(model);
2498 prop.ibc = s390_ibc_from_cpu_model(model);
2499 /* configure cpu features indicated via STFL(e) */
2500 s390_fill_feat_block(model->features, S390_FEAT_TYPE_STFL,
2501 (uint8_t *) prop.fac_list);
2502 rc = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
2503 if (rc) {
2504 error_setg(errp, "KVM: Error configuring the CPU model: %d", rc);
2505 return;
2507 /* configure cpu features indicated e.g. via SCLP */
2508 rc = configure_cpu_feat(model->features);
2509 if (rc) {
2510 error_setg(errp, "KVM: Error configuring CPU features: %d", rc);
2511 return;
2513 /* configure cpu subfunctions indicated via query / test bit */
2514 rc = configure_cpu_subfunc(model->features);
2515 if (rc) {
2516 error_setg(errp, "KVM: Error configuring CPU subfunctions: %d", rc);
2517 return;
2519 /* enable CMM via CMMA */
2520 if (test_bit(S390_FEAT_CMM, model->features)) {
2521 kvm_s390_enable_cmma();
2524 if (test_bit(S390_FEAT_AP, model->features)) {
2525 kvm_s390_configure_apie(true);
2529 void kvm_s390_restart_interrupt(S390CPU *cpu)
2531 struct kvm_s390_irq irq = {
2532 .type = KVM_S390_RESTART,
2535 kvm_s390_vcpu_interrupt(cpu, &irq);
2538 void kvm_s390_stop_interrupt(S390CPU *cpu)
2540 struct kvm_s390_irq irq = {
2541 .type = KVM_S390_SIGP_STOP,
2544 kvm_s390_vcpu_interrupt(cpu, &irq);