2 * QEMU S390x KVM implementation
5 * Copyright IBM Corp. 2012
7 * This library is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2 of the License, or (at your option) any later version.
12 * This library 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 * Lesser General Public License for more details.
17 * Contributions after 2012-10-29 are licensed under the terms of the
18 * GNU GPL, version 2 or (at your option) any later version.
20 * You should have received a copy of the GNU (Lesser) General Public
21 * License along with this library; if not, see <http://www.gnu.org/licenses/>.
24 #include "qemu/osdep.h"
25 #include <sys/ioctl.h>
27 #include <linux/kvm.h>
28 #include <asm/ptrace.h>
30 #include "qemu-common.h"
32 #include "qemu/error-report.h"
33 #include "qemu/timer.h"
34 #include "sysemu/sysemu.h"
35 #include "sysemu/kvm.h"
37 #include "sysemu/device_tree.h"
38 #include "qapi/qmp/qjson.h"
39 #include "exec/gdbstub.h"
40 #include "exec/address-spaces.h"
42 #include "qapi-event.h"
43 #include "hw/s390x/s390-pci-inst.h"
44 #include "hw/s390x/s390-pci-bus.h"
45 #include "hw/s390x/ipl.h"
46 #include "hw/s390x/ebcdic.h"
47 #include "exec/memattrs.h"
48 #include "hw/s390x/s390-virtio-ccw.h"
50 /* #define DEBUG_KVM */
53 #define DPRINTF(fmt, ...) \
54 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
56 #define DPRINTF(fmt, ...) \
60 #define kvm_vm_check_mem_attr(s, attr) \
61 kvm_vm_check_attr(s, KVM_S390_VM_MEM_CTRL, attr)
63 #define IPA0_DIAG 0x8300
64 #define IPA0_SIGP 0xae00
65 #define IPA0_B2 0xb200
66 #define IPA0_B9 0xb900
67 #define IPA0_EB 0xeb00
68 #define IPA0_E3 0xe300
70 #define PRIV_B2_SCLP_CALL 0x20
71 #define PRIV_B2_CSCH 0x30
72 #define PRIV_B2_HSCH 0x31
73 #define PRIV_B2_MSCH 0x32
74 #define PRIV_B2_SSCH 0x33
75 #define PRIV_B2_STSCH 0x34
76 #define PRIV_B2_TSCH 0x35
77 #define PRIV_B2_TPI 0x36
78 #define PRIV_B2_SAL 0x37
79 #define PRIV_B2_RSCH 0x38
80 #define PRIV_B2_STCRW 0x39
81 #define PRIV_B2_STCPS 0x3a
82 #define PRIV_B2_RCHP 0x3b
83 #define PRIV_B2_SCHM 0x3c
84 #define PRIV_B2_CHSC 0x5f
85 #define PRIV_B2_SIGA 0x74
86 #define PRIV_B2_XSCH 0x76
88 #define PRIV_EB_SQBS 0x8a
89 #define PRIV_EB_PCISTB 0xd0
90 #define PRIV_EB_SIC 0xd1
92 #define PRIV_B9_EQBS 0x9c
93 #define PRIV_B9_CLP 0xa0
94 #define PRIV_B9_PCISTG 0xd0
95 #define PRIV_B9_PCILG 0xd2
96 #define PRIV_B9_RPCIT 0xd3
98 #define PRIV_E3_MPCIFC 0xd0
99 #define PRIV_E3_STPCIFC 0xd4
101 #define DIAG_TIMEREVENT 0x288
102 #define DIAG_IPL 0x308
103 #define DIAG_KVM_HYPERCALL 0x500
104 #define DIAG_KVM_BREAKPOINT 0x501
106 #define ICPT_INSTRUCTION 0x04
107 #define ICPT_PROGRAM 0x08
108 #define ICPT_EXT_INT 0x14
109 #define ICPT_WAITPSW 0x1c
110 #define ICPT_SOFT_INTERCEPT 0x24
111 #define ICPT_CPU_STOP 0x28
112 #define ICPT_OPEREXC 0x2c
115 #define NR_LOCAL_IRQS 32
117 * Needs to be big enough to contain max_cpus emergency signals
118 * and in addition NR_LOCAL_IRQS interrupts
120 #define VCPU_IRQ_BUF_SIZE (sizeof(struct kvm_s390_irq) * \
121 (max_cpus + NR_LOCAL_IRQS))
123 static CPUWatchpoint hw_watchpoint;
125 * We don't use a list because this structure is also used to transmit the
126 * hardware breakpoints to the kernel.
128 static struct kvm_hw_breakpoint *hw_breakpoints;
129 static int nb_hw_breakpoints;
131 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
135 static int cap_sync_regs;
136 static int cap_async_pf;
137 static int cap_mem_op;
138 static int cap_s390_irq;
141 static void *legacy_s390_alloc(size_t size, uint64_t *align);
143 static int kvm_s390_query_mem_limit(KVMState *s, uint64_t *memory_limit)
145 struct kvm_device_attr attr = {
146 .group = KVM_S390_VM_MEM_CTRL,
147 .attr = KVM_S390_VM_MEM_LIMIT_SIZE,
148 .addr = (uint64_t) memory_limit,
151 return kvm_vm_ioctl(s, KVM_GET_DEVICE_ATTR, &attr);
154 int kvm_s390_set_mem_limit(KVMState *s, uint64_t new_limit, uint64_t *hw_limit)
158 struct kvm_device_attr attr = {
159 .group = KVM_S390_VM_MEM_CTRL,
160 .attr = KVM_S390_VM_MEM_LIMIT_SIZE,
161 .addr = (uint64_t) &new_limit,
164 if (!kvm_vm_check_mem_attr(s, KVM_S390_VM_MEM_LIMIT_SIZE)) {
168 rc = kvm_s390_query_mem_limit(s, hw_limit);
171 } else if (*hw_limit < new_limit) {
175 return kvm_vm_ioctl(s, KVM_SET_DEVICE_ATTR, &attr);
178 static bool kvm_s390_cmma_available(void)
180 static bool initialized, value;
184 value = kvm_vm_check_mem_attr(kvm_state, KVM_S390_VM_MEM_ENABLE_CMMA) &&
185 kvm_vm_check_mem_attr(kvm_state, KVM_S390_VM_MEM_CLR_CMMA);
190 void kvm_s390_cmma_reset(void)
193 struct kvm_device_attr attr = {
194 .group = KVM_S390_VM_MEM_CTRL,
195 .attr = KVM_S390_VM_MEM_CLR_CMMA,
198 if (!mem_path || !kvm_s390_cmma_available()) {
202 rc = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
203 trace_kvm_clear_cmma(rc);
206 static void kvm_s390_enable_cmma(void)
209 struct kvm_device_attr attr = {
210 .group = KVM_S390_VM_MEM_CTRL,
211 .attr = KVM_S390_VM_MEM_ENABLE_CMMA,
214 rc = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
215 trace_kvm_enable_cmma(rc);
218 static void kvm_s390_set_attr(uint64_t attr)
220 struct kvm_device_attr attribute = {
221 .group = KVM_S390_VM_CRYPTO,
225 int ret = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attribute);
228 error_report("Failed to set crypto device attribute %lu: %s",
229 attr, strerror(-ret));
233 static void kvm_s390_init_aes_kw(void)
235 uint64_t attr = KVM_S390_VM_CRYPTO_DISABLE_AES_KW;
237 if (object_property_get_bool(OBJECT(qdev_get_machine()), "aes-key-wrap",
239 attr = KVM_S390_VM_CRYPTO_ENABLE_AES_KW;
242 if (kvm_vm_check_attr(kvm_state, KVM_S390_VM_CRYPTO, attr)) {
243 kvm_s390_set_attr(attr);
247 static void kvm_s390_init_dea_kw(void)
249 uint64_t attr = KVM_S390_VM_CRYPTO_DISABLE_DEA_KW;
251 if (object_property_get_bool(OBJECT(qdev_get_machine()), "dea-key-wrap",
253 attr = KVM_S390_VM_CRYPTO_ENABLE_DEA_KW;
256 if (kvm_vm_check_attr(kvm_state, KVM_S390_VM_CRYPTO, attr)) {
257 kvm_s390_set_attr(attr);
261 void kvm_s390_crypto_reset(void)
263 if (s390_has_feat(S390_FEAT_MSA_EXT_3)) {
264 kvm_s390_init_aes_kw();
265 kvm_s390_init_dea_kw();
269 int kvm_arch_init(MachineState *ms, KVMState *s)
271 cap_sync_regs = kvm_check_extension(s, KVM_CAP_SYNC_REGS);
272 cap_async_pf = kvm_check_extension(s, KVM_CAP_ASYNC_PF);
273 cap_mem_op = kvm_check_extension(s, KVM_CAP_S390_MEM_OP);
274 cap_s390_irq = kvm_check_extension(s, KVM_CAP_S390_INJECT_IRQ);
276 if (!kvm_check_extension(s, KVM_CAP_S390_GMAP)
277 || !kvm_check_extension(s, KVM_CAP_S390_COW)) {
278 phys_mem_set_alloc(legacy_s390_alloc);
281 kvm_vm_enable_cap(s, KVM_CAP_S390_USER_SIGP, 0);
282 kvm_vm_enable_cap(s, KVM_CAP_S390_VECTOR_REGISTERS, 0);
283 kvm_vm_enable_cap(s, KVM_CAP_S390_USER_STSI, 0);
285 if (kvm_vm_enable_cap(s, KVM_CAP_S390_RI, 0) == 0) {
293 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
295 return cpu->cpu_index;
298 int kvm_arch_init_vcpu(CPUState *cs)
300 S390CPU *cpu = S390_CPU(cs);
301 kvm_s390_set_cpu_state(cpu, cpu->env.cpu_state);
302 cpu->irqstate = g_malloc0(VCPU_IRQ_BUF_SIZE);
306 void kvm_s390_reset_vcpu(S390CPU *cpu)
308 CPUState *cs = CPU(cpu);
310 /* The initial reset call is needed here to reset in-kernel
311 * vcpu data that we can't access directly from QEMU
312 * (i.e. with older kernels which don't support sync_regs/ONE_REG).
313 * Before this ioctl cpu_synchronize_state() is called in common kvm
315 if (kvm_vcpu_ioctl(cs, KVM_S390_INITIAL_RESET, NULL)) {
316 error_report("Initial CPU reset failed on CPU %i", cs->cpu_index);
320 static int can_sync_regs(CPUState *cs, int regs)
322 return cap_sync_regs && (cs->kvm_run->kvm_valid_regs & regs) == regs;
325 int kvm_arch_put_registers(CPUState *cs, int level)
327 S390CPU *cpu = S390_CPU(cs);
328 CPUS390XState *env = &cpu->env;
329 struct kvm_sregs sregs;
330 struct kvm_regs regs;
331 struct kvm_fpu fpu = {};
335 /* always save the PSW and the GPRS*/
336 cs->kvm_run->psw_addr = env->psw.addr;
337 cs->kvm_run->psw_mask = env->psw.mask;
339 if (can_sync_regs(cs, KVM_SYNC_GPRS)) {
340 for (i = 0; i < 16; i++) {
341 cs->kvm_run->s.regs.gprs[i] = env->regs[i];
342 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_GPRS;
345 for (i = 0; i < 16; i++) {
346 regs.gprs[i] = env->regs[i];
348 r = kvm_vcpu_ioctl(cs, KVM_SET_REGS, ®s);
354 if (can_sync_regs(cs, KVM_SYNC_VRS)) {
355 for (i = 0; i < 32; i++) {
356 cs->kvm_run->s.regs.vrs[i][0] = env->vregs[i][0].ll;
357 cs->kvm_run->s.regs.vrs[i][1] = env->vregs[i][1].ll;
359 cs->kvm_run->s.regs.fpc = env->fpc;
360 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_VRS;
361 } else if (can_sync_regs(cs, KVM_SYNC_FPRS)) {
362 for (i = 0; i < 16; i++) {
363 cs->kvm_run->s.regs.fprs[i] = get_freg(env, i)->ll;
365 cs->kvm_run->s.regs.fpc = env->fpc;
366 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_FPRS;
369 for (i = 0; i < 16; i++) {
370 fpu.fprs[i] = get_freg(env, i)->ll;
374 r = kvm_vcpu_ioctl(cs, KVM_SET_FPU, &fpu);
380 /* Do we need to save more than that? */
381 if (level == KVM_PUT_RUNTIME_STATE) {
385 if (can_sync_regs(cs, KVM_SYNC_ARCH0)) {
386 cs->kvm_run->s.regs.cputm = env->cputm;
387 cs->kvm_run->s.regs.ckc = env->ckc;
388 cs->kvm_run->s.regs.todpr = env->todpr;
389 cs->kvm_run->s.regs.gbea = env->gbea;
390 cs->kvm_run->s.regs.pp = env->pp;
391 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ARCH0;
394 * These ONE_REGS are not protected by a capability. As they are only
395 * necessary for migration we just trace a possible error, but don't
396 * return with an error return code.
398 kvm_set_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
399 kvm_set_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
400 kvm_set_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
401 kvm_set_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
402 kvm_set_one_reg(cs, KVM_REG_S390_PP, &env->pp);
405 if (can_sync_regs(cs, KVM_SYNC_RICCB)) {
406 memcpy(cs->kvm_run->s.regs.riccb, env->riccb, 64);
407 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_RICCB;
410 /* pfault parameters */
411 if (can_sync_regs(cs, KVM_SYNC_PFAULT)) {
412 cs->kvm_run->s.regs.pft = env->pfault_token;
413 cs->kvm_run->s.regs.pfs = env->pfault_select;
414 cs->kvm_run->s.regs.pfc = env->pfault_compare;
415 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_PFAULT;
416 } else if (cap_async_pf) {
417 r = kvm_set_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
421 r = kvm_set_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
425 r = kvm_set_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
431 /* access registers and control registers*/
432 if (can_sync_regs(cs, KVM_SYNC_ACRS | KVM_SYNC_CRS)) {
433 for (i = 0; i < 16; i++) {
434 cs->kvm_run->s.regs.acrs[i] = env->aregs[i];
435 cs->kvm_run->s.regs.crs[i] = env->cregs[i];
437 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ACRS;
438 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_CRS;
440 for (i = 0; i < 16; i++) {
441 sregs.acrs[i] = env->aregs[i];
442 sregs.crs[i] = env->cregs[i];
444 r = kvm_vcpu_ioctl(cs, KVM_SET_SREGS, &sregs);
450 /* Finally the prefix */
451 if (can_sync_regs(cs, KVM_SYNC_PREFIX)) {
452 cs->kvm_run->s.regs.prefix = env->psa;
453 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_PREFIX;
455 /* prefix is only supported via sync regs */
460 int kvm_arch_get_registers(CPUState *cs)
462 S390CPU *cpu = S390_CPU(cs);
463 CPUS390XState *env = &cpu->env;
464 struct kvm_sregs sregs;
465 struct kvm_regs regs;
470 env->psw.addr = cs->kvm_run->psw_addr;
471 env->psw.mask = cs->kvm_run->psw_mask;
474 if (can_sync_regs(cs, KVM_SYNC_GPRS)) {
475 for (i = 0; i < 16; i++) {
476 env->regs[i] = cs->kvm_run->s.regs.gprs[i];
479 r = kvm_vcpu_ioctl(cs, KVM_GET_REGS, ®s);
483 for (i = 0; i < 16; i++) {
484 env->regs[i] = regs.gprs[i];
488 /* The ACRS and CRS */
489 if (can_sync_regs(cs, KVM_SYNC_ACRS | KVM_SYNC_CRS)) {
490 for (i = 0; i < 16; i++) {
491 env->aregs[i] = cs->kvm_run->s.regs.acrs[i];
492 env->cregs[i] = cs->kvm_run->s.regs.crs[i];
495 r = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs);
499 for (i = 0; i < 16; i++) {
500 env->aregs[i] = sregs.acrs[i];
501 env->cregs[i] = sregs.crs[i];
505 /* Floating point and vector registers */
506 if (can_sync_regs(cs, KVM_SYNC_VRS)) {
507 for (i = 0; i < 32; i++) {
508 env->vregs[i][0].ll = cs->kvm_run->s.regs.vrs[i][0];
509 env->vregs[i][1].ll = cs->kvm_run->s.regs.vrs[i][1];
511 env->fpc = cs->kvm_run->s.regs.fpc;
512 } else if (can_sync_regs(cs, KVM_SYNC_FPRS)) {
513 for (i = 0; i < 16; i++) {
514 get_freg(env, i)->ll = cs->kvm_run->s.regs.fprs[i];
516 env->fpc = cs->kvm_run->s.regs.fpc;
518 r = kvm_vcpu_ioctl(cs, KVM_GET_FPU, &fpu);
522 for (i = 0; i < 16; i++) {
523 get_freg(env, i)->ll = fpu.fprs[i];
529 if (can_sync_regs(cs, KVM_SYNC_PREFIX)) {
530 env->psa = cs->kvm_run->s.regs.prefix;
533 if (can_sync_regs(cs, KVM_SYNC_ARCH0)) {
534 env->cputm = cs->kvm_run->s.regs.cputm;
535 env->ckc = cs->kvm_run->s.regs.ckc;
536 env->todpr = cs->kvm_run->s.regs.todpr;
537 env->gbea = cs->kvm_run->s.regs.gbea;
538 env->pp = cs->kvm_run->s.regs.pp;
541 * These ONE_REGS are not protected by a capability. As they are only
542 * necessary for migration we just trace a possible error, but don't
543 * return with an error return code.
545 kvm_get_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
546 kvm_get_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
547 kvm_get_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
548 kvm_get_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
549 kvm_get_one_reg(cs, KVM_REG_S390_PP, &env->pp);
552 if (can_sync_regs(cs, KVM_SYNC_RICCB)) {
553 memcpy(env->riccb, cs->kvm_run->s.regs.riccb, 64);
556 /* pfault parameters */
557 if (can_sync_regs(cs, KVM_SYNC_PFAULT)) {
558 env->pfault_token = cs->kvm_run->s.regs.pft;
559 env->pfault_select = cs->kvm_run->s.regs.pfs;
560 env->pfault_compare = cs->kvm_run->s.regs.pfc;
561 } else if (cap_async_pf) {
562 r = kvm_get_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
566 r = kvm_get_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
570 r = kvm_get_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
579 int kvm_s390_get_clock(uint8_t *tod_high, uint64_t *tod_low)
582 struct kvm_device_attr attr = {
583 .group = KVM_S390_VM_TOD,
584 .attr = KVM_S390_VM_TOD_LOW,
585 .addr = (uint64_t)tod_low,
588 r = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
593 attr.attr = KVM_S390_VM_TOD_HIGH;
594 attr.addr = (uint64_t)tod_high;
595 return kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
598 int kvm_s390_set_clock(uint8_t *tod_high, uint64_t *tod_low)
602 struct kvm_device_attr attr = {
603 .group = KVM_S390_VM_TOD,
604 .attr = KVM_S390_VM_TOD_LOW,
605 .addr = (uint64_t)tod_low,
608 r = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
613 attr.attr = KVM_S390_VM_TOD_HIGH;
614 attr.addr = (uint64_t)tod_high;
615 return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
620 * @addr: the logical start address in guest memory
621 * @ar: the access register number
622 * @hostbuf: buffer in host memory. NULL = do only checks w/o copying
623 * @len: length that should be transferred
624 * @is_write: true = write, false = read
625 * Returns: 0 on success, non-zero if an exception or error occurred
627 * Use KVM ioctl to read/write from/to guest memory. An access exception
628 * is injected into the vCPU in case of translation errors.
630 int kvm_s390_mem_op(S390CPU *cpu, vaddr addr, uint8_t ar, void *hostbuf,
631 int len, bool is_write)
633 struct kvm_s390_mem_op mem_op = {
635 .flags = KVM_S390_MEMOP_F_INJECT_EXCEPTION,
637 .op = is_write ? KVM_S390_MEMOP_LOGICAL_WRITE
638 : KVM_S390_MEMOP_LOGICAL_READ,
639 .buf = (uint64_t)hostbuf,
648 mem_op.flags |= KVM_S390_MEMOP_F_CHECK_ONLY;
651 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_S390_MEM_OP, &mem_op);
653 error_printf("KVM_S390_MEM_OP failed: %s\n", strerror(-ret));
659 * Legacy layout for s390:
660 * Older S390 KVM requires the topmost vma of the RAM to be
661 * smaller than an system defined value, which is at least 256GB.
662 * Larger systems have larger values. We put the guest between
663 * the end of data segment (system break) and this value. We
664 * use 32GB as a base to have enough room for the system break
665 * to grow. We also have to use MAP parameters that avoid
666 * read-only mapping of guest pages.
668 static void *legacy_s390_alloc(size_t size, uint64_t *align)
672 mem = mmap((void *) 0x800000000ULL, size,
673 PROT_EXEC|PROT_READ|PROT_WRITE,
674 MAP_SHARED | MAP_ANONYMOUS | MAP_FIXED, -1, 0);
675 return mem == MAP_FAILED ? NULL : mem;
678 static uint8_t const *sw_bp_inst;
679 static uint8_t sw_bp_ilen;
681 static void determine_sw_breakpoint_instr(void)
683 /* DIAG 501 is used for sw breakpoints with old kernels */
684 static const uint8_t diag_501[] = {0x83, 0x24, 0x05, 0x01};
685 /* Instruction 0x0000 is used for sw breakpoints with recent kernels */
686 static const uint8_t instr_0x0000[] = {0x00, 0x00};
691 if (kvm_vm_enable_cap(kvm_state, KVM_CAP_S390_USER_INSTR0, 0)) {
692 sw_bp_inst = diag_501;
693 sw_bp_ilen = sizeof(diag_501);
694 DPRINTF("KVM: will use 4-byte sw breakpoints.\n");
696 sw_bp_inst = instr_0x0000;
697 sw_bp_ilen = sizeof(instr_0x0000);
698 DPRINTF("KVM: will use 2-byte sw breakpoints.\n");
702 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
704 determine_sw_breakpoint_instr();
706 if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
708 cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)sw_bp_inst, sw_bp_ilen, 1)) {
714 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
718 if (cpu_memory_rw_debug(cs, bp->pc, t, sw_bp_ilen, 0)) {
720 } else if (memcmp(t, sw_bp_inst, sw_bp_ilen)) {
722 } else if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
730 static struct kvm_hw_breakpoint *find_hw_breakpoint(target_ulong addr,
735 for (n = 0; n < nb_hw_breakpoints; n++) {
736 if (hw_breakpoints[n].addr == addr && hw_breakpoints[n].type == type &&
737 (hw_breakpoints[n].len == len || len == -1)) {
738 return &hw_breakpoints[n];
745 static int insert_hw_breakpoint(target_ulong addr, int len, int type)
749 if (find_hw_breakpoint(addr, len, type)) {
753 size = (nb_hw_breakpoints + 1) * sizeof(struct kvm_hw_breakpoint);
755 if (!hw_breakpoints) {
756 nb_hw_breakpoints = 0;
757 hw_breakpoints = (struct kvm_hw_breakpoint *)g_try_malloc(size);
760 (struct kvm_hw_breakpoint *)g_try_realloc(hw_breakpoints, size);
763 if (!hw_breakpoints) {
764 nb_hw_breakpoints = 0;
768 hw_breakpoints[nb_hw_breakpoints].addr = addr;
769 hw_breakpoints[nb_hw_breakpoints].len = len;
770 hw_breakpoints[nb_hw_breakpoints].type = type;
777 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
778 target_ulong len, int type)
781 case GDB_BREAKPOINT_HW:
784 case GDB_WATCHPOINT_WRITE:
788 type = KVM_HW_WP_WRITE;
793 return insert_hw_breakpoint(addr, len, type);
796 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
797 target_ulong len, int type)
800 struct kvm_hw_breakpoint *bp = find_hw_breakpoint(addr, len, type);
807 if (nb_hw_breakpoints > 0) {
809 * In order to trim the array, move the last element to the position to
810 * be removed - if necessary.
812 if (bp != &hw_breakpoints[nb_hw_breakpoints]) {
813 *bp = hw_breakpoints[nb_hw_breakpoints];
815 size = nb_hw_breakpoints * sizeof(struct kvm_hw_breakpoint);
817 (struct kvm_hw_breakpoint *)g_realloc(hw_breakpoints, size);
819 g_free(hw_breakpoints);
820 hw_breakpoints = NULL;
826 void kvm_arch_remove_all_hw_breakpoints(void)
828 nb_hw_breakpoints = 0;
829 g_free(hw_breakpoints);
830 hw_breakpoints = NULL;
833 void kvm_arch_update_guest_debug(CPUState *cpu, struct kvm_guest_debug *dbg)
837 if (nb_hw_breakpoints > 0) {
838 dbg->arch.nr_hw_bp = nb_hw_breakpoints;
839 dbg->arch.hw_bp = hw_breakpoints;
841 for (i = 0; i < nb_hw_breakpoints; ++i) {
842 hw_breakpoints[i].phys_addr = s390_cpu_get_phys_addr_debug(cpu,
843 hw_breakpoints[i].addr);
845 dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
847 dbg->arch.nr_hw_bp = 0;
848 dbg->arch.hw_bp = NULL;
852 void kvm_arch_pre_run(CPUState *cpu, struct kvm_run *run)
856 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
858 return MEMTXATTRS_UNSPECIFIED;
861 int kvm_arch_process_async_events(CPUState *cs)
866 static int s390_kvm_irq_to_interrupt(struct kvm_s390_irq *irq,
867 struct kvm_s390_interrupt *interrupt)
871 interrupt->type = irq->type;
873 case KVM_S390_INT_VIRTIO:
874 interrupt->parm = irq->u.ext.ext_params;
876 case KVM_S390_INT_PFAULT_INIT:
877 case KVM_S390_INT_PFAULT_DONE:
878 interrupt->parm64 = irq->u.ext.ext_params2;
880 case KVM_S390_PROGRAM_INT:
881 interrupt->parm = irq->u.pgm.code;
883 case KVM_S390_SIGP_SET_PREFIX:
884 interrupt->parm = irq->u.prefix.address;
886 case KVM_S390_INT_SERVICE:
887 interrupt->parm = irq->u.ext.ext_params;
890 interrupt->parm = irq->u.mchk.cr14;
891 interrupt->parm64 = irq->u.mchk.mcic;
893 case KVM_S390_INT_EXTERNAL_CALL:
894 interrupt->parm = irq->u.extcall.code;
896 case KVM_S390_INT_EMERGENCY:
897 interrupt->parm = irq->u.emerg.code;
899 case KVM_S390_SIGP_STOP:
900 case KVM_S390_RESTART:
901 break; /* These types have no parameters */
902 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
903 interrupt->parm = irq->u.io.subchannel_id << 16;
904 interrupt->parm |= irq->u.io.subchannel_nr;
905 interrupt->parm64 = (uint64_t)irq->u.io.io_int_parm << 32;
906 interrupt->parm64 |= irq->u.io.io_int_word;
915 static void inject_vcpu_irq_legacy(CPUState *cs, struct kvm_s390_irq *irq)
917 struct kvm_s390_interrupt kvmint = {};
920 r = s390_kvm_irq_to_interrupt(irq, &kvmint);
922 fprintf(stderr, "%s called with bogus interrupt\n", __func__);
926 r = kvm_vcpu_ioctl(cs, KVM_S390_INTERRUPT, &kvmint);
928 fprintf(stderr, "KVM failed to inject interrupt\n");
933 void kvm_s390_vcpu_interrupt(S390CPU *cpu, struct kvm_s390_irq *irq)
935 CPUState *cs = CPU(cpu);
939 r = kvm_vcpu_ioctl(cs, KVM_S390_IRQ, irq);
943 error_report("KVM failed to inject interrupt %llx", irq->type);
947 inject_vcpu_irq_legacy(cs, irq);
950 static void __kvm_s390_floating_interrupt(struct kvm_s390_irq *irq)
952 struct kvm_s390_interrupt kvmint = {};
955 r = s390_kvm_irq_to_interrupt(irq, &kvmint);
957 fprintf(stderr, "%s called with bogus interrupt\n", __func__);
961 r = kvm_vm_ioctl(kvm_state, KVM_S390_INTERRUPT, &kvmint);
963 fprintf(stderr, "KVM failed to inject interrupt\n");
968 void kvm_s390_floating_interrupt(struct kvm_s390_irq *irq)
970 static bool use_flic = true;
974 r = kvm_s390_inject_flic(irq);
982 __kvm_s390_floating_interrupt(irq);
985 void kvm_s390_service_interrupt(uint32_t parm)
987 struct kvm_s390_irq irq = {
988 .type = KVM_S390_INT_SERVICE,
989 .u.ext.ext_params = parm,
992 kvm_s390_floating_interrupt(&irq);
995 static void enter_pgmcheck(S390CPU *cpu, uint16_t code)
997 struct kvm_s390_irq irq = {
998 .type = KVM_S390_PROGRAM_INT,
1002 kvm_s390_vcpu_interrupt(cpu, &irq);
1005 void kvm_s390_access_exception(S390CPU *cpu, uint16_t code, uint64_t te_code)
1007 struct kvm_s390_irq irq = {
1008 .type = KVM_S390_PROGRAM_INT,
1010 .u.pgm.trans_exc_code = te_code,
1011 .u.pgm.exc_access_id = te_code & 3,
1014 kvm_s390_vcpu_interrupt(cpu, &irq);
1017 static int kvm_sclp_service_call(S390CPU *cpu, struct kvm_run *run,
1020 CPUS390XState *env = &cpu->env;
1025 cpu_synchronize_state(CPU(cpu));
1026 sccb = env->regs[ipbh0 & 0xf];
1027 code = env->regs[(ipbh0 & 0xf0) >> 4];
1029 r = sclp_service_call(env, sccb, code);
1031 enter_pgmcheck(cpu, -r);
1039 static int handle_b2(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1041 CPUS390XState *env = &cpu->env;
1043 uint16_t ipbh0 = (run->s390_sieic.ipb & 0xffff0000) >> 16;
1045 cpu_synchronize_state(CPU(cpu));
1049 ioinst_handle_xsch(cpu, env->regs[1]);
1052 ioinst_handle_csch(cpu, env->regs[1]);
1055 ioinst_handle_hsch(cpu, env->regs[1]);
1058 ioinst_handle_msch(cpu, env->regs[1], run->s390_sieic.ipb);
1061 ioinst_handle_ssch(cpu, env->regs[1], run->s390_sieic.ipb);
1064 ioinst_handle_stcrw(cpu, run->s390_sieic.ipb);
1067 ioinst_handle_stsch(cpu, env->regs[1], run->s390_sieic.ipb);
1070 /* We should only get tsch via KVM_EXIT_S390_TSCH. */
1071 fprintf(stderr, "Spurious tsch intercept\n");
1074 ioinst_handle_chsc(cpu, run->s390_sieic.ipb);
1077 /* This should have been handled by kvm already. */
1078 fprintf(stderr, "Spurious tpi intercept\n");
1081 ioinst_handle_schm(cpu, env->regs[1], env->regs[2],
1082 run->s390_sieic.ipb);
1085 ioinst_handle_rsch(cpu, env->regs[1]);
1088 ioinst_handle_rchp(cpu, env->regs[1]);
1091 /* We do not provide this instruction, it is suppressed. */
1094 ioinst_handle_sal(cpu, env->regs[1]);
1097 /* Not provided, set CC = 3 for subchannel not operational */
1100 case PRIV_B2_SCLP_CALL:
1101 rc = kvm_sclp_service_call(cpu, run, ipbh0);
1105 DPRINTF("KVM: unhandled PRIV: 0xb2%x\n", ipa1);
1112 static uint64_t get_base_disp_rxy(S390CPU *cpu, struct kvm_run *run,
1115 CPUS390XState *env = &cpu->env;
1116 uint32_t x2 = (run->s390_sieic.ipa & 0x000f);
1117 uint32_t base2 = run->s390_sieic.ipb >> 28;
1118 uint32_t disp2 = ((run->s390_sieic.ipb & 0x0fff0000) >> 16) +
1119 ((run->s390_sieic.ipb & 0xff00) << 4);
1121 if (disp2 & 0x80000) {
1122 disp2 += 0xfff00000;
1128 return (base2 ? env->regs[base2] : 0) +
1129 (x2 ? env->regs[x2] : 0) + (long)(int)disp2;
1132 static uint64_t get_base_disp_rsy(S390CPU *cpu, struct kvm_run *run,
1135 CPUS390XState *env = &cpu->env;
1136 uint32_t base2 = run->s390_sieic.ipb >> 28;
1137 uint32_t disp2 = ((run->s390_sieic.ipb & 0x0fff0000) >> 16) +
1138 ((run->s390_sieic.ipb & 0xff00) << 4);
1140 if (disp2 & 0x80000) {
1141 disp2 += 0xfff00000;
1147 return (base2 ? env->regs[base2] : 0) + (long)(int)disp2;
1150 static int kvm_clp_service_call(S390CPU *cpu, struct kvm_run *run)
1152 uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1154 return clp_service_call(cpu, r2);
1157 static int kvm_pcilg_service_call(S390CPU *cpu, struct kvm_run *run)
1159 uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1160 uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1162 return pcilg_service_call(cpu, r1, r2);
1165 static int kvm_pcistg_service_call(S390CPU *cpu, struct kvm_run *run)
1167 uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1168 uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1170 return pcistg_service_call(cpu, r1, r2);
1173 static int kvm_stpcifc_service_call(S390CPU *cpu, struct kvm_run *run)
1175 uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1179 cpu_synchronize_state(CPU(cpu));
1180 fiba = get_base_disp_rxy(cpu, run, &ar);
1182 return stpcifc_service_call(cpu, r1, fiba, ar);
1185 static int kvm_sic_service_call(S390CPU *cpu, struct kvm_run *run)
1191 static int kvm_rpcit_service_call(S390CPU *cpu, struct kvm_run *run)
1193 uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1194 uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1196 return rpcit_service_call(cpu, r1, r2);
1199 static int kvm_pcistb_service_call(S390CPU *cpu, struct kvm_run *run)
1201 uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1202 uint8_t r3 = run->s390_sieic.ipa & 0x000f;
1206 cpu_synchronize_state(CPU(cpu));
1207 gaddr = get_base_disp_rsy(cpu, run, &ar);
1209 return pcistb_service_call(cpu, r1, r3, gaddr, ar);
1212 static int kvm_mpcifc_service_call(S390CPU *cpu, struct kvm_run *run)
1214 uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1218 cpu_synchronize_state(CPU(cpu));
1219 fiba = get_base_disp_rxy(cpu, run, &ar);
1221 return mpcifc_service_call(cpu, r1, fiba, ar);
1224 static int handle_b9(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1230 r = kvm_clp_service_call(cpu, run);
1232 case PRIV_B9_PCISTG:
1233 r = kvm_pcistg_service_call(cpu, run);
1236 r = kvm_pcilg_service_call(cpu, run);
1239 r = kvm_rpcit_service_call(cpu, run);
1242 /* just inject exception */
1247 DPRINTF("KVM: unhandled PRIV: 0xb9%x\n", ipa1);
1254 static int handle_eb(S390CPU *cpu, struct kvm_run *run, uint8_t ipbl)
1259 case PRIV_EB_PCISTB:
1260 r = kvm_pcistb_service_call(cpu, run);
1263 r = kvm_sic_service_call(cpu, run);
1266 /* just inject exception */
1271 DPRINTF("KVM: unhandled PRIV: 0xeb%x\n", ipbl);
1278 static int handle_e3(S390CPU *cpu, struct kvm_run *run, uint8_t ipbl)
1283 case PRIV_E3_MPCIFC:
1284 r = kvm_mpcifc_service_call(cpu, run);
1286 case PRIV_E3_STPCIFC:
1287 r = kvm_stpcifc_service_call(cpu, run);
1291 DPRINTF("KVM: unhandled PRIV: 0xe3%x\n", ipbl);
1298 static int handle_hypercall(S390CPU *cpu, struct kvm_run *run)
1300 CPUS390XState *env = &cpu->env;
1303 cpu_synchronize_state(CPU(cpu));
1304 ret = s390_virtio_hypercall(env);
1305 if (ret == -EINVAL) {
1306 enter_pgmcheck(cpu, PGM_SPECIFICATION);
1313 static void kvm_handle_diag_288(S390CPU *cpu, struct kvm_run *run)
1318 cpu_synchronize_state(CPU(cpu));
1319 r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1320 r3 = run->s390_sieic.ipa & 0x000f;
1321 rc = handle_diag_288(&cpu->env, r1, r3);
1323 enter_pgmcheck(cpu, PGM_SPECIFICATION);
1327 static void kvm_handle_diag_308(S390CPU *cpu, struct kvm_run *run)
1331 cpu_synchronize_state(CPU(cpu));
1332 r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1333 r3 = run->s390_sieic.ipa & 0x000f;
1334 handle_diag_308(&cpu->env, r1, r3);
1337 static int handle_sw_breakpoint(S390CPU *cpu, struct kvm_run *run)
1339 CPUS390XState *env = &cpu->env;
1342 cpu_synchronize_state(CPU(cpu));
1344 pc = env->psw.addr - sw_bp_ilen;
1345 if (kvm_find_sw_breakpoint(CPU(cpu), pc)) {
1353 #define DIAG_KVM_CODE_MASK 0x000000000000ffff
1355 static int handle_diag(S390CPU *cpu, struct kvm_run *run, uint32_t ipb)
1361 * For any diagnose call we support, bits 48-63 of the resulting
1362 * address specify the function code; the remainder is ignored.
1364 func_code = decode_basedisp_rs(&cpu->env, ipb, NULL) & DIAG_KVM_CODE_MASK;
1365 switch (func_code) {
1366 case DIAG_TIMEREVENT:
1367 kvm_handle_diag_288(cpu, run);
1370 kvm_handle_diag_308(cpu, run);
1372 case DIAG_KVM_HYPERCALL:
1373 r = handle_hypercall(cpu, run);
1375 case DIAG_KVM_BREAKPOINT:
1376 r = handle_sw_breakpoint(cpu, run);
1379 DPRINTF("KVM: unknown DIAG: 0x%x\n", func_code);
1380 enter_pgmcheck(cpu, PGM_SPECIFICATION);
1387 typedef struct SigpInfo {
1390 uint64_t *status_reg;
1393 static void set_sigp_status(SigpInfo *si, uint64_t status)
1395 *si->status_reg &= 0xffffffff00000000ULL;
1396 *si->status_reg |= status;
1397 si->cc = SIGP_CC_STATUS_STORED;
1400 static void sigp_start(CPUState *cs, void *arg)
1402 S390CPU *cpu = S390_CPU(cs);
1405 if (s390_cpu_get_state(cpu) != CPU_STATE_STOPPED) {
1406 si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1410 s390_cpu_set_state(CPU_STATE_OPERATING, cpu);
1411 si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1414 static void sigp_stop(CPUState *cs, void *arg)
1416 S390CPU *cpu = S390_CPU(cs);
1418 struct kvm_s390_irq irq = {
1419 .type = KVM_S390_SIGP_STOP,
1422 if (s390_cpu_get_state(cpu) != CPU_STATE_OPERATING) {
1423 si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1427 /* disabled wait - sleeping in user space */
1429 s390_cpu_set_state(CPU_STATE_STOPPED, cpu);
1431 /* execute the stop function */
1432 cpu->env.sigp_order = SIGP_STOP;
1433 kvm_s390_vcpu_interrupt(cpu, &irq);
1435 si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1438 #define ADTL_SAVE_AREA_SIZE 1024
1439 static int kvm_s390_store_adtl_status(S390CPU *cpu, hwaddr addr)
1442 hwaddr len = ADTL_SAVE_AREA_SIZE;
1444 mem = cpu_physical_memory_map(addr, &len, 1);
1448 if (len != ADTL_SAVE_AREA_SIZE) {
1449 cpu_physical_memory_unmap(mem, len, 1, 0);
1453 memcpy(mem, &cpu->env.vregs, 512);
1455 cpu_physical_memory_unmap(mem, len, 1, len);
1460 #define KVM_S390_STORE_STATUS_DEF_ADDR offsetof(LowCore, floating_pt_save_area)
1461 #define SAVE_AREA_SIZE 512
1462 static int kvm_s390_store_status(S390CPU *cpu, hwaddr addr, bool store_arch)
1464 static const uint8_t ar_id = 1;
1465 uint64_t ckc = cpu->env.ckc >> 8;
1468 hwaddr len = SAVE_AREA_SIZE;
1470 mem = cpu_physical_memory_map(addr, &len, 1);
1474 if (len != SAVE_AREA_SIZE) {
1475 cpu_physical_memory_unmap(mem, len, 1, 0);
1480 cpu_physical_memory_write(offsetof(LowCore, ar_access_id), &ar_id, 1);
1482 for (i = 0; i < 16; ++i) {
1483 *((uint64_t *)mem + i) = get_freg(&cpu->env, i)->ll;
1485 memcpy(mem + 128, &cpu->env.regs, 128);
1486 memcpy(mem + 256, &cpu->env.psw, 16);
1487 memcpy(mem + 280, &cpu->env.psa, 4);
1488 memcpy(mem + 284, &cpu->env.fpc, 4);
1489 memcpy(mem + 292, &cpu->env.todpr, 4);
1490 memcpy(mem + 296, &cpu->env.cputm, 8);
1491 memcpy(mem + 304, &ckc, 8);
1492 memcpy(mem + 320, &cpu->env.aregs, 64);
1493 memcpy(mem + 384, &cpu->env.cregs, 128);
1495 cpu_physical_memory_unmap(mem, len, 1, len);
1500 static void sigp_stop_and_store_status(CPUState *cs, void *arg)
1502 S390CPU *cpu = S390_CPU(cs);
1504 struct kvm_s390_irq irq = {
1505 .type = KVM_S390_SIGP_STOP,
1508 /* disabled wait - sleeping in user space */
1509 if (s390_cpu_get_state(cpu) == CPU_STATE_OPERATING && cs->halted) {
1510 s390_cpu_set_state(CPU_STATE_STOPPED, cpu);
1513 switch (s390_cpu_get_state(cpu)) {
1514 case CPU_STATE_OPERATING:
1515 cpu->env.sigp_order = SIGP_STOP_STORE_STATUS;
1516 kvm_s390_vcpu_interrupt(cpu, &irq);
1517 /* store will be performed when handling the stop intercept */
1519 case CPU_STATE_STOPPED:
1520 /* already stopped, just store the status */
1521 cpu_synchronize_state(cs);
1522 kvm_s390_store_status(cpu, KVM_S390_STORE_STATUS_DEF_ADDR, true);
1525 si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1528 static void sigp_store_status_at_address(CPUState *cs, void *arg)
1530 S390CPU *cpu = S390_CPU(cs);
1532 uint32_t address = si->param & 0x7ffffe00u;
1534 /* cpu has to be stopped */
1535 if (s390_cpu_get_state(cpu) != CPU_STATE_STOPPED) {
1536 set_sigp_status(si, SIGP_STAT_INCORRECT_STATE);
1540 cpu_synchronize_state(cs);
1542 if (kvm_s390_store_status(cpu, address, false)) {
1543 set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1546 si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1549 static void sigp_store_adtl_status(CPUState *cs, void *arg)
1551 S390CPU *cpu = S390_CPU(cs);
1554 if (!s390_has_feat(S390_FEAT_VECTOR)) {
1555 set_sigp_status(si, SIGP_STAT_INVALID_ORDER);
1559 /* cpu has to be stopped */
1560 if (s390_cpu_get_state(cpu) != CPU_STATE_STOPPED) {
1561 set_sigp_status(si, SIGP_STAT_INCORRECT_STATE);
1565 /* parameter must be aligned to 1024-byte boundary */
1566 if (si->param & 0x3ff) {
1567 set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1571 cpu_synchronize_state(cs);
1573 if (kvm_s390_store_adtl_status(cpu, si->param)) {
1574 set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1577 si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1580 static void sigp_restart(CPUState *cs, void *arg)
1582 S390CPU *cpu = S390_CPU(cs);
1584 struct kvm_s390_irq irq = {
1585 .type = KVM_S390_RESTART,
1588 switch (s390_cpu_get_state(cpu)) {
1589 case CPU_STATE_STOPPED:
1590 /* the restart irq has to be delivered prior to any other pending irq */
1591 cpu_synchronize_state(cs);
1592 do_restart_interrupt(&cpu->env);
1593 s390_cpu_set_state(CPU_STATE_OPERATING, cpu);
1595 case CPU_STATE_OPERATING:
1596 kvm_s390_vcpu_interrupt(cpu, &irq);
1599 si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1602 int kvm_s390_cpu_restart(S390CPU *cpu)
1606 run_on_cpu(CPU(cpu), sigp_restart, &si);
1607 DPRINTF("DONE: KVM cpu restart: %p\n", &cpu->env);
1611 static void sigp_initial_cpu_reset(CPUState *cs, void *arg)
1613 S390CPU *cpu = S390_CPU(cs);
1614 S390CPUClass *scc = S390_CPU_GET_CLASS(cpu);
1617 cpu_synchronize_state(cs);
1618 scc->initial_cpu_reset(cs);
1619 cpu_synchronize_post_reset(cs);
1620 si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1623 static void sigp_cpu_reset(CPUState *cs, void *arg)
1625 S390CPU *cpu = S390_CPU(cs);
1626 S390CPUClass *scc = S390_CPU_GET_CLASS(cpu);
1629 cpu_synchronize_state(cs);
1631 cpu_synchronize_post_reset(cs);
1632 si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1635 static void sigp_set_prefix(CPUState *cs, void *arg)
1637 S390CPU *cpu = S390_CPU(cs);
1639 uint32_t addr = si->param & 0x7fffe000u;
1641 cpu_synchronize_state(cs);
1643 if (!address_space_access_valid(&address_space_memory, addr,
1644 sizeof(struct LowCore), false)) {
1645 set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1649 /* cpu has to be stopped */
1650 if (s390_cpu_get_state(cpu) != CPU_STATE_STOPPED) {
1651 set_sigp_status(si, SIGP_STAT_INCORRECT_STATE);
1655 cpu->env.psa = addr;
1656 cpu_synchronize_post_init(cs);
1657 si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1660 static int handle_sigp_single_dst(S390CPU *dst_cpu, uint8_t order,
1661 uint64_t param, uint64_t *status_reg)
1665 .status_reg = status_reg,
1668 /* cpu available? */
1669 if (dst_cpu == NULL) {
1670 return SIGP_CC_NOT_OPERATIONAL;
1673 /* only resets can break pending orders */
1674 if (dst_cpu->env.sigp_order != 0 &&
1675 order != SIGP_CPU_RESET &&
1676 order != SIGP_INITIAL_CPU_RESET) {
1677 return SIGP_CC_BUSY;
1682 run_on_cpu(CPU(dst_cpu), sigp_start, &si);
1685 run_on_cpu(CPU(dst_cpu), sigp_stop, &si);
1688 run_on_cpu(CPU(dst_cpu), sigp_restart, &si);
1690 case SIGP_STOP_STORE_STATUS:
1691 run_on_cpu(CPU(dst_cpu), sigp_stop_and_store_status, &si);
1693 case SIGP_STORE_STATUS_ADDR:
1694 run_on_cpu(CPU(dst_cpu), sigp_store_status_at_address, &si);
1696 case SIGP_STORE_ADTL_STATUS:
1697 run_on_cpu(CPU(dst_cpu), sigp_store_adtl_status, &si);
1699 case SIGP_SET_PREFIX:
1700 run_on_cpu(CPU(dst_cpu), sigp_set_prefix, &si);
1702 case SIGP_INITIAL_CPU_RESET:
1703 run_on_cpu(CPU(dst_cpu), sigp_initial_cpu_reset, &si);
1705 case SIGP_CPU_RESET:
1706 run_on_cpu(CPU(dst_cpu), sigp_cpu_reset, &si);
1709 DPRINTF("KVM: unknown SIGP: 0x%x\n", order);
1710 set_sigp_status(&si, SIGP_STAT_INVALID_ORDER);
1716 static int sigp_set_architecture(S390CPU *cpu, uint32_t param,
1717 uint64_t *status_reg)
1722 /* due to the BQL, we are the only active cpu */
1723 CPU_FOREACH(cur_cs) {
1724 cur_cpu = S390_CPU(cur_cs);
1725 if (cur_cpu->env.sigp_order != 0) {
1726 return SIGP_CC_BUSY;
1728 cpu_synchronize_state(cur_cs);
1729 /* all but the current one have to be stopped */
1730 if (cur_cpu != cpu &&
1731 s390_cpu_get_state(cur_cpu) != CPU_STATE_STOPPED) {
1732 *status_reg &= 0xffffffff00000000ULL;
1733 *status_reg |= SIGP_STAT_INCORRECT_STATE;
1734 return SIGP_CC_STATUS_STORED;
1738 switch (param & 0xff) {
1739 case SIGP_MODE_ESA_S390:
1741 return SIGP_CC_NOT_OPERATIONAL;
1742 case SIGP_MODE_Z_ARCH_TRANS_ALL_PSW:
1743 case SIGP_MODE_Z_ARCH_TRANS_CUR_PSW:
1744 CPU_FOREACH(cur_cs) {
1745 cur_cpu = S390_CPU(cur_cs);
1746 cur_cpu->env.pfault_token = -1UL;
1750 *status_reg &= 0xffffffff00000000ULL;
1751 *status_reg |= SIGP_STAT_INVALID_PARAMETER;
1752 return SIGP_CC_STATUS_STORED;
1755 return SIGP_CC_ORDER_CODE_ACCEPTED;
1758 #define SIGP_ORDER_MASK 0x000000ff
1760 static int handle_sigp(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1762 CPUS390XState *env = &cpu->env;
1763 const uint8_t r1 = ipa1 >> 4;
1764 const uint8_t r3 = ipa1 & 0x0f;
1767 uint64_t *status_reg;
1769 S390CPU *dst_cpu = NULL;
1771 cpu_synchronize_state(CPU(cpu));
1773 /* get order code */
1774 order = decode_basedisp_rs(env, run->s390_sieic.ipb, NULL)
1776 status_reg = &env->regs[r1];
1777 param = (r1 % 2) ? env->regs[r1] : env->regs[r1 + 1];
1781 ret = sigp_set_architecture(cpu, param, status_reg);
1784 /* all other sigp orders target a single vcpu */
1785 dst_cpu = s390_cpu_addr2state(env->regs[r3]);
1786 ret = handle_sigp_single_dst(dst_cpu, order, param, status_reg);
1789 trace_kvm_sigp_finished(order, CPU(cpu)->cpu_index,
1790 dst_cpu ? CPU(dst_cpu)->cpu_index : -1, ret);
1800 static int handle_instruction(S390CPU *cpu, struct kvm_run *run)
1802 unsigned int ipa0 = (run->s390_sieic.ipa & 0xff00);
1803 uint8_t ipa1 = run->s390_sieic.ipa & 0x00ff;
1806 DPRINTF("handle_instruction 0x%x 0x%x\n",
1807 run->s390_sieic.ipa, run->s390_sieic.ipb);
1810 r = handle_b2(cpu, run, ipa1);
1813 r = handle_b9(cpu, run, ipa1);
1816 r = handle_eb(cpu, run, run->s390_sieic.ipb & 0xff);
1819 r = handle_e3(cpu, run, run->s390_sieic.ipb & 0xff);
1822 r = handle_diag(cpu, run, run->s390_sieic.ipb);
1825 r = handle_sigp(cpu, run, ipa1);
1831 enter_pgmcheck(cpu, 0x0001);
1837 static bool is_special_wait_psw(CPUState *cs)
1839 /* signal quiesce */
1840 return cs->kvm_run->psw_addr == 0xfffUL;
1843 static void unmanageable_intercept(S390CPU *cpu, const char *str, int pswoffset)
1845 CPUState *cs = CPU(cpu);
1847 error_report("Unmanageable %s! CPU%i new PSW: 0x%016lx:%016lx",
1848 str, cs->cpu_index, ldq_phys(cs->as, cpu->env.psa + pswoffset),
1849 ldq_phys(cs->as, cpu->env.psa + pswoffset + 8));
1851 qemu_system_guest_panicked();
1854 static int handle_intercept(S390CPU *cpu)
1856 CPUState *cs = CPU(cpu);
1857 struct kvm_run *run = cs->kvm_run;
1858 int icpt_code = run->s390_sieic.icptcode;
1861 DPRINTF("intercept: 0x%x (at 0x%lx)\n", icpt_code,
1862 (long)cs->kvm_run->psw_addr);
1863 switch (icpt_code) {
1864 case ICPT_INSTRUCTION:
1865 r = handle_instruction(cpu, run);
1868 unmanageable_intercept(cpu, "program interrupt",
1869 offsetof(LowCore, program_new_psw));
1873 unmanageable_intercept(cpu, "external interrupt",
1874 offsetof(LowCore, external_new_psw));
1878 /* disabled wait, since enabled wait is handled in kernel */
1879 cpu_synchronize_state(cs);
1880 if (s390_cpu_halt(cpu) == 0) {
1881 if (is_special_wait_psw(cs)) {
1882 qemu_system_shutdown_request();
1884 qemu_system_guest_panicked();
1890 if (s390_cpu_set_state(CPU_STATE_STOPPED, cpu) == 0) {
1891 qemu_system_shutdown_request();
1893 if (cpu->env.sigp_order == SIGP_STOP_STORE_STATUS) {
1894 kvm_s390_store_status(cpu, KVM_S390_STORE_STATUS_DEF_ADDR,
1897 cpu->env.sigp_order = 0;
1901 /* currently only instr 0x0000 after enabled via capability */
1902 r = handle_sw_breakpoint(cpu, run);
1904 enter_pgmcheck(cpu, PGM_OPERATION);
1908 case ICPT_SOFT_INTERCEPT:
1909 fprintf(stderr, "KVM unimplemented icpt SOFT\n");
1913 fprintf(stderr, "KVM unimplemented icpt IO\n");
1917 fprintf(stderr, "Unknown intercept code: %d\n", icpt_code);
1925 static int handle_tsch(S390CPU *cpu)
1927 CPUState *cs = CPU(cpu);
1928 struct kvm_run *run = cs->kvm_run;
1931 cpu_synchronize_state(cs);
1933 ret = ioinst_handle_tsch(cpu, cpu->env.regs[1], run->s390_tsch.ipb);
1937 * If an I/O interrupt had been dequeued, we have to reinject it.
1939 if (run->s390_tsch.dequeued) {
1940 kvm_s390_io_interrupt(run->s390_tsch.subchannel_id,
1941 run->s390_tsch.subchannel_nr,
1942 run->s390_tsch.io_int_parm,
1943 run->s390_tsch.io_int_word);
1950 static void insert_stsi_3_2_2(S390CPU *cpu, __u64 addr, uint8_t ar)
1952 struct sysib_322 sysib;
1955 if (s390_cpu_virt_mem_read(cpu, addr, ar, &sysib, sizeof(sysib))) {
1958 /* Shift the stack of Extended Names to prepare for our own data */
1959 memmove(&sysib.ext_names[1], &sysib.ext_names[0],
1960 sizeof(sysib.ext_names[0]) * (sysib.count - 1));
1961 /* First virt level, that doesn't provide Ext Names delimits stack. It is
1962 * assumed it's not capable of managing Extended Names for lower levels.
1964 for (del = 1; del < sysib.count; del++) {
1965 if (!sysib.vm[del].ext_name_encoding || !sysib.ext_names[del][0]) {
1969 if (del < sysib.count) {
1970 memset(sysib.ext_names[del], 0,
1971 sizeof(sysib.ext_names[0]) * (sysib.count - del));
1973 /* Insert short machine name in EBCDIC, padded with blanks */
1975 memset(sysib.vm[0].name, 0x40, sizeof(sysib.vm[0].name));
1976 ebcdic_put(sysib.vm[0].name, qemu_name, MIN(sizeof(sysib.vm[0].name),
1977 strlen(qemu_name)));
1979 sysib.vm[0].ext_name_encoding = 2; /* 2 = UTF-8 */
1980 memset(sysib.ext_names[0], 0, sizeof(sysib.ext_names[0]));
1981 /* If hypervisor specifies zero Extended Name in STSI322 SYSIB, it's
1982 * considered by s390 as not capable of providing any Extended Name.
1983 * Therefore if no name was specified on qemu invocation, we go with the
1984 * same "KVMguest" default, which KVM has filled into short name field.
1987 strncpy((char *)sysib.ext_names[0], qemu_name,
1988 sizeof(sysib.ext_names[0]));
1990 strcpy((char *)sysib.ext_names[0], "KVMguest");
1993 memcpy(sysib.vm[0].uuid, qemu_uuid, sizeof(sysib.vm[0].uuid));
1995 s390_cpu_virt_mem_write(cpu, addr, ar, &sysib, sizeof(sysib));
1998 static int handle_stsi(S390CPU *cpu)
2000 CPUState *cs = CPU(cpu);
2001 struct kvm_run *run = cs->kvm_run;
2003 switch (run->s390_stsi.fc) {
2005 if (run->s390_stsi.sel1 != 2 || run->s390_stsi.sel2 != 2) {
2008 /* Only sysib 3.2.2 needs post-handling for now. */
2009 insert_stsi_3_2_2(cpu, run->s390_stsi.addr, run->s390_stsi.ar);
2016 static int kvm_arch_handle_debug_exit(S390CPU *cpu)
2018 CPUState *cs = CPU(cpu);
2019 struct kvm_run *run = cs->kvm_run;
2022 struct kvm_debug_exit_arch *arch_info = &run->debug.arch;
2024 switch (arch_info->type) {
2025 case KVM_HW_WP_WRITE:
2026 if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
2027 cs->watchpoint_hit = &hw_watchpoint;
2028 hw_watchpoint.vaddr = arch_info->addr;
2029 hw_watchpoint.flags = BP_MEM_WRITE;
2034 if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
2038 case KVM_SINGLESTEP:
2039 if (cs->singlestep_enabled) {
2050 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
2052 S390CPU *cpu = S390_CPU(cs);
2055 qemu_mutex_lock_iothread();
2057 switch (run->exit_reason) {
2058 case KVM_EXIT_S390_SIEIC:
2059 ret = handle_intercept(cpu);
2061 case KVM_EXIT_S390_RESET:
2062 s390_reipl_request();
2064 case KVM_EXIT_S390_TSCH:
2065 ret = handle_tsch(cpu);
2067 case KVM_EXIT_S390_STSI:
2068 ret = handle_stsi(cpu);
2070 case KVM_EXIT_DEBUG:
2071 ret = kvm_arch_handle_debug_exit(cpu);
2074 fprintf(stderr, "Unknown KVM exit: %d\n", run->exit_reason);
2077 qemu_mutex_unlock_iothread();
2080 ret = EXCP_INTERRUPT;
2085 bool kvm_arch_stop_on_emulation_error(CPUState *cpu)
2090 int kvm_arch_on_sigbus_vcpu(CPUState *cpu, int code, void *addr)
2095 int kvm_arch_on_sigbus(int code, void *addr)
2100 void kvm_s390_io_interrupt(uint16_t subchannel_id,
2101 uint16_t subchannel_nr, uint32_t io_int_parm,
2102 uint32_t io_int_word)
2104 struct kvm_s390_irq irq = {
2105 .u.io.subchannel_id = subchannel_id,
2106 .u.io.subchannel_nr = subchannel_nr,
2107 .u.io.io_int_parm = io_int_parm,
2108 .u.io.io_int_word = io_int_word,
2111 if (io_int_word & IO_INT_WORD_AI) {
2112 irq.type = KVM_S390_INT_IO(1, 0, 0, 0);
2114 irq.type = KVM_S390_INT_IO(0, (subchannel_id & 0xff00) >> 8,
2115 (subchannel_id & 0x0006),
2118 kvm_s390_floating_interrupt(&irq);
2121 static uint64_t build_channel_report_mcic(void)
2125 /* subclass: indicate channel report pending */
2127 /* subclass modifiers: none */
2128 /* storage errors: none */
2129 /* validity bits: no damage */
2130 MCIC_VB_WP | MCIC_VB_MS | MCIC_VB_PM | MCIC_VB_IA | MCIC_VB_FP |
2131 MCIC_VB_GR | MCIC_VB_CR | MCIC_VB_ST | MCIC_VB_AR | MCIC_VB_PR |
2132 MCIC_VB_FC | MCIC_VB_CT | MCIC_VB_CC;
2133 if (s390_has_feat(S390_FEAT_VECTOR)) {
2139 void kvm_s390_crw_mchk(void)
2141 struct kvm_s390_irq irq = {
2142 .type = KVM_S390_MCHK,
2143 .u.mchk.cr14 = 1 << 28,
2144 .u.mchk.mcic = build_channel_report_mcic(),
2146 kvm_s390_floating_interrupt(&irq);
2149 void kvm_s390_enable_css_support(S390CPU *cpu)
2153 /* Activate host kernel channel subsystem support. */
2154 r = kvm_vcpu_enable_cap(CPU(cpu), KVM_CAP_S390_CSS_SUPPORT, 0);
2158 void kvm_arch_init_irq_routing(KVMState *s)
2161 * Note that while irqchip capabilities generally imply that cpustates
2162 * are handled in-kernel, it is not true for s390 (yet); therefore, we
2163 * have to override the common code kvm_halt_in_kernel_allowed setting.
2165 if (kvm_check_extension(s, KVM_CAP_IRQ_ROUTING)) {
2166 kvm_gsi_routing_allowed = true;
2167 kvm_halt_in_kernel_allowed = false;
2171 int kvm_s390_assign_subch_ioeventfd(EventNotifier *notifier, uint32_t sch,
2172 int vq, bool assign)
2174 struct kvm_ioeventfd kick = {
2175 .flags = KVM_IOEVENTFD_FLAG_VIRTIO_CCW_NOTIFY |
2176 KVM_IOEVENTFD_FLAG_DATAMATCH,
2177 .fd = event_notifier_get_fd(notifier),
2182 if (!kvm_check_extension(kvm_state, KVM_CAP_IOEVENTFD)) {
2186 kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
2188 return kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
2191 int kvm_s390_get_memslot_count(KVMState *s)
2193 return kvm_check_extension(s, KVM_CAP_NR_MEMSLOTS);
2196 int kvm_s390_get_ri(void)
2201 int kvm_s390_set_cpu_state(S390CPU *cpu, uint8_t cpu_state)
2203 struct kvm_mp_state mp_state = {};
2206 /* the kvm part might not have been initialized yet */
2207 if (CPU(cpu)->kvm_state == NULL) {
2211 switch (cpu_state) {
2212 case CPU_STATE_STOPPED:
2213 mp_state.mp_state = KVM_MP_STATE_STOPPED;
2215 case CPU_STATE_CHECK_STOP:
2216 mp_state.mp_state = KVM_MP_STATE_CHECK_STOP;
2218 case CPU_STATE_OPERATING:
2219 mp_state.mp_state = KVM_MP_STATE_OPERATING;
2221 case CPU_STATE_LOAD:
2222 mp_state.mp_state = KVM_MP_STATE_LOAD;
2225 error_report("Requested CPU state is not a valid S390 CPU state: %u",
2230 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
2232 trace_kvm_failed_cpu_state_set(CPU(cpu)->cpu_index, cpu_state,
2239 void kvm_s390_vcpu_interrupt_pre_save(S390CPU *cpu)
2241 struct kvm_s390_irq_state irq_state;
2242 CPUState *cs = CPU(cpu);
2245 if (!kvm_check_extension(kvm_state, KVM_CAP_S390_IRQ_STATE)) {
2249 irq_state.buf = (uint64_t) cpu->irqstate;
2250 irq_state.len = VCPU_IRQ_BUF_SIZE;
2252 bytes = kvm_vcpu_ioctl(cs, KVM_S390_GET_IRQ_STATE, &irq_state);
2254 cpu->irqstate_saved_size = 0;
2255 error_report("Migration of interrupt state failed");
2259 cpu->irqstate_saved_size = bytes;
2262 int kvm_s390_vcpu_interrupt_post_load(S390CPU *cpu)
2264 CPUState *cs = CPU(cpu);
2265 struct kvm_s390_irq_state irq_state;
2268 if (cpu->irqstate_saved_size == 0) {
2272 if (!kvm_check_extension(kvm_state, KVM_CAP_S390_IRQ_STATE)) {
2276 irq_state.buf = (uint64_t) cpu->irqstate;
2277 irq_state.len = cpu->irqstate_saved_size;
2279 r = kvm_vcpu_ioctl(cs, KVM_S390_SET_IRQ_STATE, &irq_state);
2281 error_report("Setting interrupt state failed %d", r);
2286 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
2287 uint64_t address, uint32_t data, PCIDevice *dev)
2289 S390PCIBusDevice *pbdev;
2290 uint32_t idx = data >> ZPCI_MSI_VEC_BITS;
2291 uint32_t vec = data & ZPCI_MSI_VEC_MASK;
2293 pbdev = s390_pci_find_dev_by_idx(idx);
2295 DPRINTF("add_msi_route no dev\n");
2299 pbdev->routes.adapter.ind_offset = vec;
2301 route->type = KVM_IRQ_ROUTING_S390_ADAPTER;
2303 route->u.adapter.summary_addr = pbdev->routes.adapter.summary_addr;
2304 route->u.adapter.ind_addr = pbdev->routes.adapter.ind_addr;
2305 route->u.adapter.summary_offset = pbdev->routes.adapter.summary_offset;
2306 route->u.adapter.ind_offset = pbdev->routes.adapter.ind_offset;
2307 route->u.adapter.adapter_id = pbdev->routes.adapter.adapter_id;
2311 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
2312 int vector, PCIDevice *dev)
2317 int kvm_arch_release_virq_post(int virq)
2322 int kvm_arch_msi_data_to_gsi(uint32_t data)
2327 static inline int test_bit_inv(long nr, const unsigned long *addr)
2329 return test_bit(BE_BIT_NR(nr), addr);
2332 static inline void set_bit_inv(long nr, unsigned long *addr)
2334 set_bit(BE_BIT_NR(nr), addr);
2337 static int query_cpu_subfunc(S390FeatBitmap features)
2339 struct kvm_s390_vm_cpu_subfunc prop;
2340 struct kvm_device_attr attr = {
2341 .group = KVM_S390_VM_CPU_MODEL,
2342 .attr = KVM_S390_VM_CPU_MACHINE_SUBFUNC,
2343 .addr = (uint64_t) &prop,
2347 rc = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
2353 * We're going to add all subfunctions now, if the corresponding feature
2354 * is available that unlocks the query functions.
2356 s390_add_from_feat_block(features, S390_FEAT_TYPE_PLO, prop.plo);
2357 if (test_bit(S390_FEAT_TOD_CLOCK_STEERING, features)) {
2358 s390_add_from_feat_block(features, S390_FEAT_TYPE_PTFF, prop.ptff);
2360 if (test_bit(S390_FEAT_MSA, features)) {
2361 s390_add_from_feat_block(features, S390_FEAT_TYPE_KMAC, prop.kmac);
2362 s390_add_from_feat_block(features, S390_FEAT_TYPE_KMC, prop.kmc);
2363 s390_add_from_feat_block(features, S390_FEAT_TYPE_KM, prop.km);
2364 s390_add_from_feat_block(features, S390_FEAT_TYPE_KIMD, prop.kimd);
2365 s390_add_from_feat_block(features, S390_FEAT_TYPE_KLMD, prop.klmd);
2367 if (test_bit(S390_FEAT_MSA_EXT_3, features)) {
2368 s390_add_from_feat_block(features, S390_FEAT_TYPE_PCKMO, prop.pckmo);
2370 if (test_bit(S390_FEAT_MSA_EXT_4, features)) {
2371 s390_add_from_feat_block(features, S390_FEAT_TYPE_KMCTR, prop.kmctr);
2372 s390_add_from_feat_block(features, S390_FEAT_TYPE_KMF, prop.kmf);
2373 s390_add_from_feat_block(features, S390_FEAT_TYPE_KMO, prop.kmo);
2374 s390_add_from_feat_block(features, S390_FEAT_TYPE_PCC, prop.pcc);
2376 if (test_bit(S390_FEAT_MSA_EXT_5, features)) {
2377 s390_add_from_feat_block(features, S390_FEAT_TYPE_PPNO, prop.ppno);
2382 static int configure_cpu_subfunc(const S390FeatBitmap features)
2384 struct kvm_s390_vm_cpu_subfunc prop = {};
2385 struct kvm_device_attr attr = {
2386 .group = KVM_S390_VM_CPU_MODEL,
2387 .attr = KVM_S390_VM_CPU_PROCESSOR_SUBFUNC,
2388 .addr = (uint64_t) &prop,
2391 if (!kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2392 KVM_S390_VM_CPU_PROCESSOR_SUBFUNC)) {
2393 /* hardware support might be missing, IBC will handle most of this */
2397 s390_fill_feat_block(features, S390_FEAT_TYPE_PLO, prop.plo);
2398 if (test_bit(S390_FEAT_TOD_CLOCK_STEERING, features)) {
2399 s390_fill_feat_block(features, S390_FEAT_TYPE_PTFF, prop.ptff);
2400 prop.ptff[0] |= 0x80; /* query is always available */
2402 if (test_bit(S390_FEAT_MSA, features)) {
2403 s390_fill_feat_block(features, S390_FEAT_TYPE_KMAC, prop.kmac);
2404 prop.kmac[0] |= 0x80; /* query is always available */
2405 s390_fill_feat_block(features, S390_FEAT_TYPE_KMC, prop.kmc);
2406 prop.kmc[0] |= 0x80; /* query is always available */
2407 s390_fill_feat_block(features, S390_FEAT_TYPE_KM, prop.km);
2408 prop.km[0] |= 0x80; /* query is always available */
2409 s390_fill_feat_block(features, S390_FEAT_TYPE_KIMD, prop.kimd);
2410 prop.kimd[0] |= 0x80; /* query is always available */
2411 s390_fill_feat_block(features, S390_FEAT_TYPE_KLMD, prop.klmd);
2412 prop.klmd[0] |= 0x80; /* query is always available */
2414 if (test_bit(S390_FEAT_MSA_EXT_3, features)) {
2415 s390_fill_feat_block(features, S390_FEAT_TYPE_PCKMO, prop.pckmo);
2416 prop.pckmo[0] |= 0x80; /* query is always available */
2418 if (test_bit(S390_FEAT_MSA_EXT_4, features)) {
2419 s390_fill_feat_block(features, S390_FEAT_TYPE_KMCTR, prop.kmctr);
2420 prop.kmctr[0] |= 0x80; /* query is always available */
2421 s390_fill_feat_block(features, S390_FEAT_TYPE_KMF, prop.kmf);
2422 prop.kmf[0] |= 0x80; /* query is always available */
2423 s390_fill_feat_block(features, S390_FEAT_TYPE_KMO, prop.kmo);
2424 prop.kmo[0] |= 0x80; /* query is always available */
2425 s390_fill_feat_block(features, S390_FEAT_TYPE_PCC, prop.pcc);
2426 prop.pcc[0] |= 0x80; /* query is always available */
2428 if (test_bit(S390_FEAT_MSA_EXT_5, features)) {
2429 s390_fill_feat_block(features, S390_FEAT_TYPE_PPNO, prop.ppno);
2430 prop.ppno[0] |= 0x80; /* query is always available */
2432 return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
2435 static int kvm_to_feat[][2] = {
2436 { KVM_S390_VM_CPU_FEAT_ESOP, S390_FEAT_ESOP },
2437 { KVM_S390_VM_CPU_FEAT_SIEF2, S390_FEAT_SIE_F2 },
2438 { KVM_S390_VM_CPU_FEAT_64BSCAO , S390_FEAT_SIE_64BSCAO },
2439 { KVM_S390_VM_CPU_FEAT_SIIF, S390_FEAT_SIE_SIIF },
2440 { KVM_S390_VM_CPU_FEAT_GPERE, S390_FEAT_SIE_GPERE },
2441 { KVM_S390_VM_CPU_FEAT_GSLS, S390_FEAT_SIE_GSLS },
2442 { KVM_S390_VM_CPU_FEAT_IB, S390_FEAT_SIE_IB },
2443 { KVM_S390_VM_CPU_FEAT_CEI, S390_FEAT_SIE_CEI },
2444 { KVM_S390_VM_CPU_FEAT_IBS, S390_FEAT_SIE_IBS },
2445 { KVM_S390_VM_CPU_FEAT_SKEY, S390_FEAT_SIE_SKEY },
2446 { KVM_S390_VM_CPU_FEAT_CMMA, S390_FEAT_SIE_CMMA },
2447 { KVM_S390_VM_CPU_FEAT_PFMFI, S390_FEAT_SIE_PFMFI},
2448 { KVM_S390_VM_CPU_FEAT_SIGPIF, S390_FEAT_SIE_SIGPIF},
2451 static int query_cpu_feat(S390FeatBitmap features)
2453 struct kvm_s390_vm_cpu_feat prop;
2454 struct kvm_device_attr attr = {
2455 .group = KVM_S390_VM_CPU_MODEL,
2456 .attr = KVM_S390_VM_CPU_MACHINE_FEAT,
2457 .addr = (uint64_t) &prop,
2462 rc = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
2467 for (i = 0; i < ARRAY_SIZE(kvm_to_feat); i++) {
2468 if (test_bit_inv(kvm_to_feat[i][0], (unsigned long *)prop.feat)) {
2469 set_bit(kvm_to_feat[i][1], features);
2475 static int configure_cpu_feat(const S390FeatBitmap features)
2477 struct kvm_s390_vm_cpu_feat prop = {};
2478 struct kvm_device_attr attr = {
2479 .group = KVM_S390_VM_CPU_MODEL,
2480 .attr = KVM_S390_VM_CPU_PROCESSOR_FEAT,
2481 .addr = (uint64_t) &prop,
2485 for (i = 0; i < ARRAY_SIZE(kvm_to_feat); i++) {
2486 if (test_bit(kvm_to_feat[i][1], features)) {
2487 set_bit_inv(kvm_to_feat[i][0], (unsigned long *)prop.feat);
2490 return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
2493 bool kvm_s390_cpu_models_supported(void)
2495 if (!cpu_model_allowed()) {
2496 /* compatibility machines interfere with the cpu model */
2499 return kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2500 KVM_S390_VM_CPU_MACHINE) &&
2501 kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2502 KVM_S390_VM_CPU_PROCESSOR) &&
2503 kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2504 KVM_S390_VM_CPU_MACHINE_FEAT) &&
2505 kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2506 KVM_S390_VM_CPU_PROCESSOR_FEAT) &&
2507 kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2508 KVM_S390_VM_CPU_MACHINE_SUBFUNC);
2511 void kvm_s390_get_host_cpu_model(S390CPUModel *model, Error **errp)
2513 struct kvm_s390_vm_cpu_machine prop = {};
2514 struct kvm_device_attr attr = {
2515 .group = KVM_S390_VM_CPU_MODEL,
2516 .attr = KVM_S390_VM_CPU_MACHINE,
2517 .addr = (uint64_t) &prop,
2519 uint16_t unblocked_ibc = 0, cpu_type = 0;
2522 memset(model, 0, sizeof(*model));
2524 if (!kvm_s390_cpu_models_supported()) {
2525 error_setg(errp, "KVM doesn't support CPU models");
2529 /* query the basic cpu model properties */
2530 rc = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
2532 error_setg(errp, "KVM: Error querying host CPU model: %d", rc);
2536 cpu_type = cpuid_type(prop.cpuid);
2537 if (has_ibc(prop.ibc)) {
2538 model->lowest_ibc = lowest_ibc(prop.ibc);
2539 unblocked_ibc = unblocked_ibc(prop.ibc);
2541 model->cpu_id = cpuid_id(prop.cpuid);
2542 model->cpu_ver = 0xff;
2544 /* get supported cpu features indicated via STFL(E) */
2545 s390_add_from_feat_block(model->features, S390_FEAT_TYPE_STFL,
2546 (uint8_t *) prop.fac_mask);
2547 /* dat-enhancement facility 2 has no bit but was introduced with stfle */
2548 if (test_bit(S390_FEAT_STFLE, model->features)) {
2549 set_bit(S390_FEAT_DAT_ENH_2, model->features);
2551 /* get supported cpu features indicated e.g. via SCLP */
2552 rc = query_cpu_feat(model->features);
2554 error_setg(errp, "KVM: Error querying CPU features: %d", rc);
2557 /* get supported cpu subfunctions indicated via query / test bit */
2558 rc = query_cpu_subfunc(model->features);
2560 error_setg(errp, "KVM: Error querying CPU subfunctions: %d", rc);
2564 /* with cpu model support, CMM is only indicated if really available */
2565 if (kvm_s390_cmma_available()) {
2566 set_bit(S390_FEAT_CMM, model->features);
2569 if (s390_known_cpu_type(cpu_type)) {
2570 /* we want the exact model, even if some features are missing */
2571 model->def = s390_find_cpu_def(cpu_type, ibc_gen(unblocked_ibc),
2572 ibc_ec_ga(unblocked_ibc), NULL);
2574 /* model unknown, e.g. too new - search using features */
2575 model->def = s390_find_cpu_def(0, ibc_gen(unblocked_ibc),
2576 ibc_ec_ga(unblocked_ibc),
2580 error_setg(errp, "KVM: host CPU model could not be identified");
2583 /* strip of features that are not part of the maximum model */
2584 bitmap_and(model->features, model->features, model->def->full_feat,
2588 void kvm_s390_apply_cpu_model(const S390CPUModel *model, Error **errp)
2590 struct kvm_s390_vm_cpu_processor prop = {
2593 struct kvm_device_attr attr = {
2594 .group = KVM_S390_VM_CPU_MODEL,
2595 .attr = KVM_S390_VM_CPU_PROCESSOR,
2596 .addr = (uint64_t) &prop,
2601 /* compatibility handling if cpu models are disabled */
2602 if (kvm_s390_cmma_available() && !mem_path) {
2603 kvm_s390_enable_cmma();
2607 if (!kvm_s390_cpu_models_supported()) {
2608 error_setg(errp, "KVM doesn't support CPU models");
2611 prop.cpuid = s390_cpuid_from_cpu_model(model);
2612 prop.ibc = s390_ibc_from_cpu_model(model);
2613 /* configure cpu features indicated via STFL(e) */
2614 s390_fill_feat_block(model->features, S390_FEAT_TYPE_STFL,
2615 (uint8_t *) prop.fac_list);
2616 rc = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
2618 error_setg(errp, "KVM: Error configuring the CPU model: %d", rc);
2621 /* configure cpu features indicated e.g. via SCLP */
2622 rc = configure_cpu_feat(model->features);
2624 error_setg(errp, "KVM: Error configuring CPU features: %d", rc);
2627 /* configure cpu subfunctions indicated via query / test bit */
2628 rc = configure_cpu_subfunc(model->features);
2630 error_setg(errp, "KVM: Error configuring CPU subfunctions: %d", rc);
2633 /* enable CMM via CMMA - disable on hugetlbfs */
2634 if (test_bit(S390_FEAT_CMM, model->features)) {
2636 error_report("Warning: CMM will not be enabled because it is not "
2637 "compatible to hugetlbfs.");
2639 kvm_s390_enable_cmma();