pc: reserve more memory for ACPI for new machine types
[qemu.git] / target-s390x / kvm.c
bloba32d91aa013626730a72c2c083a0daf43809811c
1 /*
2 * QEMU S390x KVM implementation
4 * Copyright (c) 2009 Alexander Graf <agraf@suse.de>
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 <sys/types.h>
25 #include <sys/ioctl.h>
26 #include <sys/mman.h>
28 #include <linux/kvm.h>
29 #include <asm/ptrace.h>
31 #include "qemu-common.h"
32 #include "qemu/timer.h"
33 #include "sysemu/sysemu.h"
34 #include "sysemu/kvm.h"
35 #include "hw/hw.h"
36 #include "cpu.h"
37 #include "sysemu/device_tree.h"
38 #include "qapi/qmp/qjson.h"
39 #include "monitor/monitor.h"
40 #include "exec/gdbstub.h"
41 #include "trace.h"
42 #include "qapi-event.h"
44 /* #define DEBUG_KVM */
46 #ifdef DEBUG_KVM
47 #define DPRINTF(fmt, ...) \
48 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
49 #else
50 #define DPRINTF(fmt, ...) \
51 do { } while (0)
52 #endif
54 #define IPA0_DIAG 0x8300
55 #define IPA0_SIGP 0xae00
56 #define IPA0_B2 0xb200
57 #define IPA0_B9 0xb900
58 #define IPA0_EB 0xeb00
60 #define PRIV_B2_SCLP_CALL 0x20
61 #define PRIV_B2_CSCH 0x30
62 #define PRIV_B2_HSCH 0x31
63 #define PRIV_B2_MSCH 0x32
64 #define PRIV_B2_SSCH 0x33
65 #define PRIV_B2_STSCH 0x34
66 #define PRIV_B2_TSCH 0x35
67 #define PRIV_B2_TPI 0x36
68 #define PRIV_B2_SAL 0x37
69 #define PRIV_B2_RSCH 0x38
70 #define PRIV_B2_STCRW 0x39
71 #define PRIV_B2_STCPS 0x3a
72 #define PRIV_B2_RCHP 0x3b
73 #define PRIV_B2_SCHM 0x3c
74 #define PRIV_B2_CHSC 0x5f
75 #define PRIV_B2_SIGA 0x74
76 #define PRIV_B2_XSCH 0x76
78 #define PRIV_EB_SQBS 0x8a
80 #define PRIV_B9_EQBS 0x9c
82 #define DIAG_IPL 0x308
83 #define DIAG_KVM_HYPERCALL 0x500
84 #define DIAG_KVM_BREAKPOINT 0x501
86 #define ICPT_INSTRUCTION 0x04
87 #define ICPT_PROGRAM 0x08
88 #define ICPT_EXT_INT 0x14
89 #define ICPT_WAITPSW 0x1c
90 #define ICPT_SOFT_INTERCEPT 0x24
91 #define ICPT_CPU_STOP 0x28
92 #define ICPT_IO 0x40
94 static CPUWatchpoint hw_watchpoint;
96 * We don't use a list because this structure is also used to transmit the
97 * hardware breakpoints to the kernel.
99 static struct kvm_hw_breakpoint *hw_breakpoints;
100 static int nb_hw_breakpoints;
102 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
103 KVM_CAP_LAST_INFO
106 static int cap_sync_regs;
107 static int cap_async_pf;
109 static void *legacy_s390_alloc(size_t size);
111 static int kvm_s390_check_clear_cmma(KVMState *s)
113 struct kvm_device_attr attr = {
114 .group = KVM_S390_VM_MEM_CTRL,
115 .attr = KVM_S390_VM_MEM_CLR_CMMA,
118 return kvm_vm_ioctl(s, KVM_HAS_DEVICE_ATTR, &attr);
121 static int kvm_s390_check_enable_cmma(KVMState *s)
123 struct kvm_device_attr attr = {
124 .group = KVM_S390_VM_MEM_CTRL,
125 .attr = KVM_S390_VM_MEM_ENABLE_CMMA,
128 return kvm_vm_ioctl(s, KVM_HAS_DEVICE_ATTR, &attr);
131 void kvm_s390_clear_cmma_callback(void *opaque)
133 int rc;
134 KVMState *s = opaque;
135 struct kvm_device_attr attr = {
136 .group = KVM_S390_VM_MEM_CTRL,
137 .attr = KVM_S390_VM_MEM_CLR_CMMA,
140 rc = kvm_vm_ioctl(s, KVM_SET_DEVICE_ATTR, &attr);
141 trace_kvm_clear_cmma(rc);
144 static void kvm_s390_enable_cmma(KVMState *s)
146 int rc;
147 struct kvm_device_attr attr = {
148 .group = KVM_S390_VM_MEM_CTRL,
149 .attr = KVM_S390_VM_MEM_ENABLE_CMMA,
152 if (kvm_s390_check_enable_cmma(s) || kvm_s390_check_clear_cmma(s)) {
153 return;
156 rc = kvm_vm_ioctl(s, KVM_SET_DEVICE_ATTR, &attr);
157 if (!rc) {
158 qemu_register_reset(kvm_s390_clear_cmma_callback, s);
160 trace_kvm_enable_cmma(rc);
163 int kvm_arch_init(KVMState *s)
165 cap_sync_regs = kvm_check_extension(s, KVM_CAP_SYNC_REGS);
166 cap_async_pf = kvm_check_extension(s, KVM_CAP_ASYNC_PF);
168 if (kvm_check_extension(s, KVM_CAP_VM_ATTRIBUTES)) {
169 kvm_s390_enable_cmma(s);
172 if (!kvm_check_extension(s, KVM_CAP_S390_GMAP)
173 || !kvm_check_extension(s, KVM_CAP_S390_COW)) {
174 phys_mem_set_alloc(legacy_s390_alloc);
176 return 0;
179 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
181 return cpu->cpu_index;
184 int kvm_arch_init_vcpu(CPUState *cpu)
186 /* nothing todo yet */
187 return 0;
190 void kvm_s390_reset_vcpu(S390CPU *cpu)
192 CPUState *cs = CPU(cpu);
194 /* The initial reset call is needed here to reset in-kernel
195 * vcpu data that we can't access directly from QEMU
196 * (i.e. with older kernels which don't support sync_regs/ONE_REG).
197 * Before this ioctl cpu_synchronize_state() is called in common kvm
198 * code (kvm-all) */
199 if (kvm_vcpu_ioctl(cs, KVM_S390_INITIAL_RESET, NULL)) {
200 perror("Can't reset vcpu\n");
204 int kvm_arch_put_registers(CPUState *cs, int level)
206 S390CPU *cpu = S390_CPU(cs);
207 CPUS390XState *env = &cpu->env;
208 struct kvm_sregs sregs;
209 struct kvm_regs regs;
210 struct kvm_fpu fpu;
211 int r;
212 int i;
214 /* always save the PSW and the GPRS*/
215 cs->kvm_run->psw_addr = env->psw.addr;
216 cs->kvm_run->psw_mask = env->psw.mask;
218 if (cap_sync_regs && cs->kvm_run->kvm_valid_regs & KVM_SYNC_GPRS) {
219 for (i = 0; i < 16; i++) {
220 cs->kvm_run->s.regs.gprs[i] = env->regs[i];
221 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_GPRS;
223 } else {
224 for (i = 0; i < 16; i++) {
225 regs.gprs[i] = env->regs[i];
227 r = kvm_vcpu_ioctl(cs, KVM_SET_REGS, &regs);
228 if (r < 0) {
229 return r;
233 /* Floating point */
234 for (i = 0; i < 16; i++) {
235 fpu.fprs[i] = env->fregs[i].ll;
237 fpu.fpc = env->fpc;
239 r = kvm_vcpu_ioctl(cs, KVM_SET_FPU, &fpu);
240 if (r < 0) {
241 return r;
244 /* Do we need to save more than that? */
245 if (level == KVM_PUT_RUNTIME_STATE) {
246 return 0;
250 * These ONE_REGS are not protected by a capability. As they are only
251 * necessary for migration we just trace a possible error, but don't
252 * return with an error return code.
254 kvm_set_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
255 kvm_set_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
256 kvm_set_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
257 kvm_set_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
258 kvm_set_one_reg(cs, KVM_REG_S390_PP, &env->pp);
260 if (cap_async_pf) {
261 r = kvm_set_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
262 if (r < 0) {
263 return r;
265 r = kvm_set_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
266 if (r < 0) {
267 return r;
269 r = kvm_set_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
270 if (r < 0) {
271 return r;
275 if (cap_sync_regs &&
276 cs->kvm_run->kvm_valid_regs & KVM_SYNC_ACRS &&
277 cs->kvm_run->kvm_valid_regs & KVM_SYNC_CRS) {
278 for (i = 0; i < 16; i++) {
279 cs->kvm_run->s.regs.acrs[i] = env->aregs[i];
280 cs->kvm_run->s.regs.crs[i] = env->cregs[i];
282 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ACRS;
283 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_CRS;
284 } else {
285 for (i = 0; i < 16; i++) {
286 sregs.acrs[i] = env->aregs[i];
287 sregs.crs[i] = env->cregs[i];
289 r = kvm_vcpu_ioctl(cs, KVM_SET_SREGS, &sregs);
290 if (r < 0) {
291 return r;
295 /* Finally the prefix */
296 if (cap_sync_regs && cs->kvm_run->kvm_valid_regs & KVM_SYNC_PREFIX) {
297 cs->kvm_run->s.regs.prefix = env->psa;
298 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_PREFIX;
299 } else {
300 /* prefix is only supported via sync regs */
302 return 0;
305 int kvm_arch_get_registers(CPUState *cs)
307 S390CPU *cpu = S390_CPU(cs);
308 CPUS390XState *env = &cpu->env;
309 struct kvm_sregs sregs;
310 struct kvm_regs regs;
311 struct kvm_fpu fpu;
312 int i, r;
314 /* get the PSW */
315 env->psw.addr = cs->kvm_run->psw_addr;
316 env->psw.mask = cs->kvm_run->psw_mask;
318 /* the GPRS */
319 if (cap_sync_regs && cs->kvm_run->kvm_valid_regs & KVM_SYNC_GPRS) {
320 for (i = 0; i < 16; i++) {
321 env->regs[i] = cs->kvm_run->s.regs.gprs[i];
323 } else {
324 r = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
325 if (r < 0) {
326 return r;
328 for (i = 0; i < 16; i++) {
329 env->regs[i] = regs.gprs[i];
333 /* The ACRS and CRS */
334 if (cap_sync_regs &&
335 cs->kvm_run->kvm_valid_regs & KVM_SYNC_ACRS &&
336 cs->kvm_run->kvm_valid_regs & KVM_SYNC_CRS) {
337 for (i = 0; i < 16; i++) {
338 env->aregs[i] = cs->kvm_run->s.regs.acrs[i];
339 env->cregs[i] = cs->kvm_run->s.regs.crs[i];
341 } else {
342 r = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs);
343 if (r < 0) {
344 return r;
346 for (i = 0; i < 16; i++) {
347 env->aregs[i] = sregs.acrs[i];
348 env->cregs[i] = sregs.crs[i];
352 /* Floating point */
353 r = kvm_vcpu_ioctl(cs, KVM_GET_FPU, &fpu);
354 if (r < 0) {
355 return r;
357 for (i = 0; i < 16; i++) {
358 env->fregs[i].ll = fpu.fprs[i];
360 env->fpc = fpu.fpc;
362 /* The prefix */
363 if (cap_sync_regs && cs->kvm_run->kvm_valid_regs & KVM_SYNC_PREFIX) {
364 env->psa = cs->kvm_run->s.regs.prefix;
368 * These ONE_REGS are not protected by a capability. As they are only
369 * necessary for migration we just trace a possible error, but don't
370 * return with an error return code.
372 kvm_get_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
373 kvm_get_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
374 kvm_get_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
375 kvm_get_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
376 kvm_get_one_reg(cs, KVM_REG_S390_PP, &env->pp);
378 if (cap_async_pf) {
379 r = kvm_get_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
380 if (r < 0) {
381 return r;
383 r = kvm_get_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
384 if (r < 0) {
385 return r;
387 r = kvm_get_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
388 if (r < 0) {
389 return r;
393 return 0;
397 * Legacy layout for s390:
398 * Older S390 KVM requires the topmost vma of the RAM to be
399 * smaller than an system defined value, which is at least 256GB.
400 * Larger systems have larger values. We put the guest between
401 * the end of data segment (system break) and this value. We
402 * use 32GB as a base to have enough room for the system break
403 * to grow. We also have to use MAP parameters that avoid
404 * read-only mapping of guest pages.
406 static void *legacy_s390_alloc(size_t size)
408 void *mem;
410 mem = mmap((void *) 0x800000000ULL, size,
411 PROT_EXEC|PROT_READ|PROT_WRITE,
412 MAP_SHARED | MAP_ANONYMOUS | MAP_FIXED, -1, 0);
413 return mem == MAP_FAILED ? NULL : mem;
416 /* DIAG 501 is used for sw breakpoints */
417 static const uint8_t diag_501[] = {0x83, 0x24, 0x05, 0x01};
419 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
422 if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
423 sizeof(diag_501), 0) ||
424 cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)diag_501,
425 sizeof(diag_501), 1)) {
426 return -EINVAL;
428 return 0;
431 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
433 uint8_t t[sizeof(diag_501)];
435 if (cpu_memory_rw_debug(cs, bp->pc, t, sizeof(diag_501), 0)) {
436 return -EINVAL;
437 } else if (memcmp(t, diag_501, sizeof(diag_501))) {
438 return -EINVAL;
439 } else if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
440 sizeof(diag_501), 1)) {
441 return -EINVAL;
444 return 0;
447 static struct kvm_hw_breakpoint *find_hw_breakpoint(target_ulong addr,
448 int len, int type)
450 int n;
452 for (n = 0; n < nb_hw_breakpoints; n++) {
453 if (hw_breakpoints[n].addr == addr && hw_breakpoints[n].type == type &&
454 (hw_breakpoints[n].len == len || len == -1)) {
455 return &hw_breakpoints[n];
459 return NULL;
462 static int insert_hw_breakpoint(target_ulong addr, int len, int type)
464 int size;
466 if (find_hw_breakpoint(addr, len, type)) {
467 return -EEXIST;
470 size = (nb_hw_breakpoints + 1) * sizeof(struct kvm_hw_breakpoint);
472 if (!hw_breakpoints) {
473 nb_hw_breakpoints = 0;
474 hw_breakpoints = (struct kvm_hw_breakpoint *)g_try_malloc(size);
475 } else {
476 hw_breakpoints =
477 (struct kvm_hw_breakpoint *)g_try_realloc(hw_breakpoints, size);
480 if (!hw_breakpoints) {
481 nb_hw_breakpoints = 0;
482 return -ENOMEM;
485 hw_breakpoints[nb_hw_breakpoints].addr = addr;
486 hw_breakpoints[nb_hw_breakpoints].len = len;
487 hw_breakpoints[nb_hw_breakpoints].type = type;
489 nb_hw_breakpoints++;
491 return 0;
494 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
495 target_ulong len, int type)
497 switch (type) {
498 case GDB_BREAKPOINT_HW:
499 type = KVM_HW_BP;
500 break;
501 case GDB_WATCHPOINT_WRITE:
502 if (len < 1) {
503 return -EINVAL;
505 type = KVM_HW_WP_WRITE;
506 break;
507 default:
508 return -ENOSYS;
510 return insert_hw_breakpoint(addr, len, type);
513 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
514 target_ulong len, int type)
516 int size;
517 struct kvm_hw_breakpoint *bp = find_hw_breakpoint(addr, len, type);
519 if (bp == NULL) {
520 return -ENOENT;
523 nb_hw_breakpoints--;
524 if (nb_hw_breakpoints > 0) {
526 * In order to trim the array, move the last element to the position to
527 * be removed - if necessary.
529 if (bp != &hw_breakpoints[nb_hw_breakpoints]) {
530 *bp = hw_breakpoints[nb_hw_breakpoints];
532 size = nb_hw_breakpoints * sizeof(struct kvm_hw_breakpoint);
533 hw_breakpoints =
534 (struct kvm_hw_breakpoint *)g_realloc(hw_breakpoints, size);
535 } else {
536 g_free(hw_breakpoints);
537 hw_breakpoints = NULL;
540 return 0;
543 void kvm_arch_remove_all_hw_breakpoints(void)
545 nb_hw_breakpoints = 0;
546 g_free(hw_breakpoints);
547 hw_breakpoints = NULL;
550 void kvm_arch_update_guest_debug(CPUState *cpu, struct kvm_guest_debug *dbg)
552 int i;
554 if (nb_hw_breakpoints > 0) {
555 dbg->arch.nr_hw_bp = nb_hw_breakpoints;
556 dbg->arch.hw_bp = hw_breakpoints;
558 for (i = 0; i < nb_hw_breakpoints; ++i) {
559 hw_breakpoints[i].phys_addr = s390_cpu_get_phys_addr_debug(cpu,
560 hw_breakpoints[i].addr);
562 dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
563 } else {
564 dbg->arch.nr_hw_bp = 0;
565 dbg->arch.hw_bp = NULL;
569 void kvm_arch_pre_run(CPUState *cpu, struct kvm_run *run)
573 void kvm_arch_post_run(CPUState *cpu, struct kvm_run *run)
577 int kvm_arch_process_async_events(CPUState *cs)
579 return cs->halted;
582 static int s390_kvm_irq_to_interrupt(struct kvm_s390_irq *irq,
583 struct kvm_s390_interrupt *interrupt)
585 int r = 0;
587 interrupt->type = irq->type;
588 switch (irq->type) {
589 case KVM_S390_INT_VIRTIO:
590 interrupt->parm = irq->u.ext.ext_params;
591 /* fall through */
592 case KVM_S390_INT_PFAULT_INIT:
593 case KVM_S390_INT_PFAULT_DONE:
594 interrupt->parm64 = irq->u.ext.ext_params2;
595 break;
596 case KVM_S390_PROGRAM_INT:
597 interrupt->parm = irq->u.pgm.code;
598 break;
599 case KVM_S390_SIGP_SET_PREFIX:
600 interrupt->parm = irq->u.prefix.address;
601 break;
602 case KVM_S390_INT_SERVICE:
603 interrupt->parm = irq->u.ext.ext_params;
604 break;
605 case KVM_S390_MCHK:
606 interrupt->parm = irq->u.mchk.cr14;
607 interrupt->parm64 = irq->u.mchk.mcic;
608 break;
609 case KVM_S390_INT_EXTERNAL_CALL:
610 interrupt->parm = irq->u.extcall.code;
611 break;
612 case KVM_S390_INT_EMERGENCY:
613 interrupt->parm = irq->u.emerg.code;
614 break;
615 case KVM_S390_SIGP_STOP:
616 case KVM_S390_RESTART:
617 break; /* These types have no parameters */
618 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
619 interrupt->parm = irq->u.io.subchannel_id << 16;
620 interrupt->parm |= irq->u.io.subchannel_nr;
621 interrupt->parm64 = (uint64_t)irq->u.io.io_int_parm << 32;
622 interrupt->parm64 |= irq->u.io.io_int_word;
623 break;
624 default:
625 r = -EINVAL;
626 break;
628 return r;
631 void kvm_s390_vcpu_interrupt(S390CPU *cpu, struct kvm_s390_irq *irq)
633 struct kvm_s390_interrupt kvmint = {};
634 CPUState *cs = CPU(cpu);
635 int r;
637 r = s390_kvm_irq_to_interrupt(irq, &kvmint);
638 if (r < 0) {
639 fprintf(stderr, "%s called with bogus interrupt\n", __func__);
640 exit(1);
643 r = kvm_vcpu_ioctl(cs, KVM_S390_INTERRUPT, &kvmint);
644 if (r < 0) {
645 fprintf(stderr, "KVM failed to inject interrupt\n");
646 exit(1);
650 static void __kvm_s390_floating_interrupt(struct kvm_s390_irq *irq)
652 struct kvm_s390_interrupt kvmint = {};
653 int r;
655 r = s390_kvm_irq_to_interrupt(irq, &kvmint);
656 if (r < 0) {
657 fprintf(stderr, "%s called with bogus interrupt\n", __func__);
658 exit(1);
661 r = kvm_vm_ioctl(kvm_state, KVM_S390_INTERRUPT, &kvmint);
662 if (r < 0) {
663 fprintf(stderr, "KVM failed to inject interrupt\n");
664 exit(1);
668 void kvm_s390_floating_interrupt(struct kvm_s390_irq *irq)
670 static bool use_flic = true;
671 int r;
673 if (use_flic) {
674 r = kvm_s390_inject_flic(irq);
675 if (r == -ENOSYS) {
676 use_flic = false;
678 if (!r) {
679 return;
682 __kvm_s390_floating_interrupt(irq);
685 void kvm_s390_virtio_irq(int config_change, uint64_t token)
687 struct kvm_s390_irq irq = {
688 .type = KVM_S390_INT_VIRTIO,
689 .u.ext.ext_params = config_change,
690 .u.ext.ext_params2 = token,
693 kvm_s390_floating_interrupt(&irq);
696 void kvm_s390_service_interrupt(uint32_t parm)
698 struct kvm_s390_irq irq = {
699 .type = KVM_S390_INT_SERVICE,
700 .u.ext.ext_params = parm,
703 kvm_s390_floating_interrupt(&irq);
706 static void enter_pgmcheck(S390CPU *cpu, uint16_t code)
708 struct kvm_s390_irq irq = {
709 .type = KVM_S390_PROGRAM_INT,
710 .u.pgm.code = code,
713 kvm_s390_vcpu_interrupt(cpu, &irq);
716 static int kvm_sclp_service_call(S390CPU *cpu, struct kvm_run *run,
717 uint16_t ipbh0)
719 CPUS390XState *env = &cpu->env;
720 uint64_t sccb;
721 uint32_t code;
722 int r = 0;
724 cpu_synchronize_state(CPU(cpu));
725 sccb = env->regs[ipbh0 & 0xf];
726 code = env->regs[(ipbh0 & 0xf0) >> 4];
728 r = sclp_service_call(env, sccb, code);
729 if (r < 0) {
730 enter_pgmcheck(cpu, -r);
731 } else {
732 setcc(cpu, r);
735 return 0;
738 static int handle_b2(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
740 CPUS390XState *env = &cpu->env;
741 int rc = 0;
742 uint16_t ipbh0 = (run->s390_sieic.ipb & 0xffff0000) >> 16;
744 cpu_synchronize_state(CPU(cpu));
746 switch (ipa1) {
747 case PRIV_B2_XSCH:
748 ioinst_handle_xsch(cpu, env->regs[1]);
749 break;
750 case PRIV_B2_CSCH:
751 ioinst_handle_csch(cpu, env->regs[1]);
752 break;
753 case PRIV_B2_HSCH:
754 ioinst_handle_hsch(cpu, env->regs[1]);
755 break;
756 case PRIV_B2_MSCH:
757 ioinst_handle_msch(cpu, env->regs[1], run->s390_sieic.ipb);
758 break;
759 case PRIV_B2_SSCH:
760 ioinst_handle_ssch(cpu, env->regs[1], run->s390_sieic.ipb);
761 break;
762 case PRIV_B2_STCRW:
763 ioinst_handle_stcrw(cpu, run->s390_sieic.ipb);
764 break;
765 case PRIV_B2_STSCH:
766 ioinst_handle_stsch(cpu, env->regs[1], run->s390_sieic.ipb);
767 break;
768 case PRIV_B2_TSCH:
769 /* We should only get tsch via KVM_EXIT_S390_TSCH. */
770 fprintf(stderr, "Spurious tsch intercept\n");
771 break;
772 case PRIV_B2_CHSC:
773 ioinst_handle_chsc(cpu, run->s390_sieic.ipb);
774 break;
775 case PRIV_B2_TPI:
776 /* This should have been handled by kvm already. */
777 fprintf(stderr, "Spurious tpi intercept\n");
778 break;
779 case PRIV_B2_SCHM:
780 ioinst_handle_schm(cpu, env->regs[1], env->regs[2],
781 run->s390_sieic.ipb);
782 break;
783 case PRIV_B2_RSCH:
784 ioinst_handle_rsch(cpu, env->regs[1]);
785 break;
786 case PRIV_B2_RCHP:
787 ioinst_handle_rchp(cpu, env->regs[1]);
788 break;
789 case PRIV_B2_STCPS:
790 /* We do not provide this instruction, it is suppressed. */
791 break;
792 case PRIV_B2_SAL:
793 ioinst_handle_sal(cpu, env->regs[1]);
794 break;
795 case PRIV_B2_SIGA:
796 /* Not provided, set CC = 3 for subchannel not operational */
797 setcc(cpu, 3);
798 break;
799 case PRIV_B2_SCLP_CALL:
800 rc = kvm_sclp_service_call(cpu, run, ipbh0);
801 break;
802 default:
803 rc = -1;
804 DPRINTF("KVM: unhandled PRIV: 0xb2%x\n", ipa1);
805 break;
808 return rc;
811 static int handle_b9(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
813 int r = 0;
815 switch (ipa1) {
816 case PRIV_B9_EQBS:
817 /* just inject exception */
818 r = -1;
819 break;
820 default:
821 r = -1;
822 DPRINTF("KVM: unhandled PRIV: 0xb9%x\n", ipa1);
823 break;
826 return r;
829 static int handle_eb(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
831 int r = 0;
833 switch (ipa1) {
834 case PRIV_EB_SQBS:
835 /* just inject exception */
836 r = -1;
837 break;
838 default:
839 r = -1;
840 DPRINTF("KVM: unhandled PRIV: 0xeb%x\n", ipa1);
841 break;
844 return r;
847 static int handle_hypercall(S390CPU *cpu, struct kvm_run *run)
849 CPUS390XState *env = &cpu->env;
850 int ret;
852 cpu_synchronize_state(CPU(cpu));
853 ret = s390_virtio_hypercall(env);
854 if (ret == -EINVAL) {
855 enter_pgmcheck(cpu, PGM_SPECIFICATION);
856 return 0;
859 return ret;
862 static void kvm_handle_diag_308(S390CPU *cpu, struct kvm_run *run)
864 uint64_t r1, r3;
866 cpu_synchronize_state(CPU(cpu));
867 r1 = (run->s390_sieic.ipa & 0x00f0) >> 8;
868 r3 = run->s390_sieic.ipa & 0x000f;
869 handle_diag_308(&cpu->env, r1, r3);
872 static int handle_sw_breakpoint(S390CPU *cpu, struct kvm_run *run)
874 CPUS390XState *env = &cpu->env;
875 unsigned long pc;
877 cpu_synchronize_state(CPU(cpu));
879 pc = env->psw.addr - 4;
880 if (kvm_find_sw_breakpoint(CPU(cpu), pc)) {
881 env->psw.addr = pc;
882 return EXCP_DEBUG;
885 return -ENOENT;
888 #define DIAG_KVM_CODE_MASK 0x000000000000ffff
890 static int handle_diag(S390CPU *cpu, struct kvm_run *run, uint32_t ipb)
892 int r = 0;
893 uint16_t func_code;
896 * For any diagnose call we support, bits 48-63 of the resulting
897 * address specify the function code; the remainder is ignored.
899 func_code = decode_basedisp_rs(&cpu->env, ipb) & DIAG_KVM_CODE_MASK;
900 switch (func_code) {
901 case DIAG_IPL:
902 kvm_handle_diag_308(cpu, run);
903 break;
904 case DIAG_KVM_HYPERCALL:
905 r = handle_hypercall(cpu, run);
906 break;
907 case DIAG_KVM_BREAKPOINT:
908 r = handle_sw_breakpoint(cpu, run);
909 break;
910 default:
911 DPRINTF("KVM: unknown DIAG: 0x%x\n", func_code);
912 r = -1;
913 break;
916 return r;
919 static int kvm_s390_cpu_start(S390CPU *cpu)
921 s390_add_running_cpu(cpu);
922 qemu_cpu_kick(CPU(cpu));
923 DPRINTF("DONE: KVM cpu start: %p\n", &cpu->env);
924 return 0;
927 int kvm_s390_cpu_restart(S390CPU *cpu)
929 struct kvm_s390_irq irq = {
930 .type = KVM_S390_RESTART,
933 kvm_s390_vcpu_interrupt(cpu, &irq);
934 s390_add_running_cpu(cpu);
935 qemu_cpu_kick(CPU(cpu));
936 DPRINTF("DONE: KVM cpu restart: %p\n", &cpu->env);
937 return 0;
940 static void sigp_initial_cpu_reset(void *arg)
942 CPUState *cpu = arg;
943 S390CPUClass *scc = S390_CPU_GET_CLASS(cpu);
945 cpu_synchronize_state(cpu);
946 scc->initial_cpu_reset(cpu);
949 static void sigp_cpu_reset(void *arg)
951 CPUState *cpu = arg;
952 S390CPUClass *scc = S390_CPU_GET_CLASS(cpu);
954 cpu_synchronize_state(cpu);
955 scc->cpu_reset(cpu);
958 #define SIGP_ORDER_MASK 0x000000ff
960 static int handle_sigp(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
962 CPUS390XState *env = &cpu->env;
963 uint8_t order_code;
964 uint16_t cpu_addr;
965 S390CPU *target_cpu;
966 uint64_t *statusreg = &env->regs[ipa1 >> 4];
967 int cc;
969 cpu_synchronize_state(CPU(cpu));
971 /* get order code */
972 order_code = decode_basedisp_rs(env, run->s390_sieic.ipb) & SIGP_ORDER_MASK;
974 cpu_addr = env->regs[ipa1 & 0x0f];
975 target_cpu = s390_cpu_addr2state(cpu_addr);
976 if (target_cpu == NULL) {
977 cc = 3; /* not operational */
978 goto out;
981 switch (order_code) {
982 case SIGP_START:
983 cc = kvm_s390_cpu_start(target_cpu);
984 break;
985 case SIGP_RESTART:
986 cc = kvm_s390_cpu_restart(target_cpu);
987 break;
988 case SIGP_SET_ARCH:
989 *statusreg &= 0xffffffff00000000UL;
990 *statusreg |= SIGP_STAT_INVALID_PARAMETER;
991 cc = 1; /* status stored */
992 break;
993 case SIGP_INITIAL_CPU_RESET:
994 run_on_cpu(CPU(target_cpu), sigp_initial_cpu_reset, CPU(target_cpu));
995 cc = 0;
996 break;
997 case SIGP_CPU_RESET:
998 run_on_cpu(CPU(target_cpu), sigp_cpu_reset, CPU(target_cpu));
999 cc = 0;
1000 break;
1001 default:
1002 DPRINTF("KVM: unknown SIGP: 0x%x\n", order_code);
1003 *statusreg &= 0xffffffff00000000UL;
1004 *statusreg |= SIGP_STAT_INVALID_ORDER;
1005 cc = 1; /* status stored */
1006 break;
1009 out:
1010 setcc(cpu, cc);
1011 return 0;
1014 static int handle_instruction(S390CPU *cpu, struct kvm_run *run)
1016 unsigned int ipa0 = (run->s390_sieic.ipa & 0xff00);
1017 uint8_t ipa1 = run->s390_sieic.ipa & 0x00ff;
1018 int r = -1;
1020 DPRINTF("handle_instruction 0x%x 0x%x\n",
1021 run->s390_sieic.ipa, run->s390_sieic.ipb);
1022 switch (ipa0) {
1023 case IPA0_B2:
1024 r = handle_b2(cpu, run, ipa1);
1025 break;
1026 case IPA0_B9:
1027 r = handle_b9(cpu, run, ipa1);
1028 break;
1029 case IPA0_EB:
1030 r = handle_eb(cpu, run, ipa1);
1031 break;
1032 case IPA0_DIAG:
1033 r = handle_diag(cpu, run, run->s390_sieic.ipb);
1034 break;
1035 case IPA0_SIGP:
1036 r = handle_sigp(cpu, run, ipa1);
1037 break;
1040 if (r < 0) {
1041 r = 0;
1042 enter_pgmcheck(cpu, 0x0001);
1045 return r;
1048 static bool is_special_wait_psw(CPUState *cs)
1050 /* signal quiesce */
1051 return cs->kvm_run->psw_addr == 0xfffUL;
1054 static void guest_panicked(void)
1056 qapi_event_send_guest_panicked(GUEST_PANIC_ACTION_PAUSE,
1057 &error_abort);
1058 vm_stop(RUN_STATE_GUEST_PANICKED);
1061 static void unmanageable_intercept(S390CPU *cpu, const char *str, int pswoffset)
1063 CPUState *cs = CPU(cpu);
1065 error_report("Unmanageable %s! CPU%i new PSW: 0x%016lx:%016lx",
1066 str, cs->cpu_index, ldq_phys(cs->as, cpu->env.psa + pswoffset),
1067 ldq_phys(cs->as, cpu->env.psa + pswoffset + 8));
1068 s390_del_running_cpu(cpu);
1069 guest_panicked();
1072 static int handle_intercept(S390CPU *cpu)
1074 CPUState *cs = CPU(cpu);
1075 struct kvm_run *run = cs->kvm_run;
1076 int icpt_code = run->s390_sieic.icptcode;
1077 int r = 0;
1079 DPRINTF("intercept: 0x%x (at 0x%lx)\n", icpt_code,
1080 (long)cs->kvm_run->psw_addr);
1081 switch (icpt_code) {
1082 case ICPT_INSTRUCTION:
1083 r = handle_instruction(cpu, run);
1084 break;
1085 case ICPT_PROGRAM:
1086 unmanageable_intercept(cpu, "program interrupt",
1087 offsetof(LowCore, program_new_psw));
1088 r = EXCP_HALTED;
1089 break;
1090 case ICPT_EXT_INT:
1091 unmanageable_intercept(cpu, "external interrupt",
1092 offsetof(LowCore, external_new_psw));
1093 r = EXCP_HALTED;
1094 break;
1095 case ICPT_WAITPSW:
1096 /* disabled wait, since enabled wait is handled in kernel */
1097 if (s390_del_running_cpu(cpu) == 0) {
1098 if (is_special_wait_psw(cs)) {
1099 qemu_system_shutdown_request();
1100 } else {
1101 guest_panicked();
1104 r = EXCP_HALTED;
1105 break;
1106 case ICPT_CPU_STOP:
1107 if (s390_del_running_cpu(cpu) == 0) {
1108 qemu_system_shutdown_request();
1110 r = EXCP_HALTED;
1111 break;
1112 case ICPT_SOFT_INTERCEPT:
1113 fprintf(stderr, "KVM unimplemented icpt SOFT\n");
1114 exit(1);
1115 break;
1116 case ICPT_IO:
1117 fprintf(stderr, "KVM unimplemented icpt IO\n");
1118 exit(1);
1119 break;
1120 default:
1121 fprintf(stderr, "Unknown intercept code: %d\n", icpt_code);
1122 exit(1);
1123 break;
1126 return r;
1129 static int handle_tsch(S390CPU *cpu)
1131 CPUS390XState *env = &cpu->env;
1132 CPUState *cs = CPU(cpu);
1133 struct kvm_run *run = cs->kvm_run;
1134 int ret;
1136 cpu_synchronize_state(cs);
1138 ret = ioinst_handle_tsch(env, env->regs[1], run->s390_tsch.ipb);
1139 if (ret >= 0) {
1140 /* Success; set condition code. */
1141 setcc(cpu, ret);
1142 ret = 0;
1143 } else if (ret < -1) {
1145 * Failure.
1146 * If an I/O interrupt had been dequeued, we have to reinject it.
1148 if (run->s390_tsch.dequeued) {
1149 kvm_s390_io_interrupt(run->s390_tsch.subchannel_id,
1150 run->s390_tsch.subchannel_nr,
1151 run->s390_tsch.io_int_parm,
1152 run->s390_tsch.io_int_word);
1154 ret = 0;
1156 return ret;
1159 static int kvm_arch_handle_debug_exit(S390CPU *cpu)
1161 CPUState *cs = CPU(cpu);
1162 struct kvm_run *run = cs->kvm_run;
1164 int ret = 0;
1165 struct kvm_debug_exit_arch *arch_info = &run->debug.arch;
1167 switch (arch_info->type) {
1168 case KVM_HW_WP_WRITE:
1169 if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
1170 cs->watchpoint_hit = &hw_watchpoint;
1171 hw_watchpoint.vaddr = arch_info->addr;
1172 hw_watchpoint.flags = BP_MEM_WRITE;
1173 ret = EXCP_DEBUG;
1175 break;
1176 case KVM_HW_BP:
1177 if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
1178 ret = EXCP_DEBUG;
1180 break;
1181 case KVM_SINGLESTEP:
1182 if (cs->singlestep_enabled) {
1183 ret = EXCP_DEBUG;
1185 break;
1186 default:
1187 ret = -ENOSYS;
1190 return ret;
1193 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1195 S390CPU *cpu = S390_CPU(cs);
1196 int ret = 0;
1198 switch (run->exit_reason) {
1199 case KVM_EXIT_S390_SIEIC:
1200 ret = handle_intercept(cpu);
1201 break;
1202 case KVM_EXIT_S390_RESET:
1203 qemu_system_reset_request();
1204 break;
1205 case KVM_EXIT_S390_TSCH:
1206 ret = handle_tsch(cpu);
1207 break;
1208 case KVM_EXIT_DEBUG:
1209 ret = kvm_arch_handle_debug_exit(cpu);
1210 break;
1211 default:
1212 fprintf(stderr, "Unknown KVM exit: %d\n", run->exit_reason);
1213 break;
1216 if (ret == 0) {
1217 ret = EXCP_INTERRUPT;
1219 return ret;
1222 bool kvm_arch_stop_on_emulation_error(CPUState *cpu)
1224 return true;
1227 int kvm_arch_on_sigbus_vcpu(CPUState *cpu, int code, void *addr)
1229 return 1;
1232 int kvm_arch_on_sigbus(int code, void *addr)
1234 return 1;
1237 void kvm_s390_io_interrupt(uint16_t subchannel_id,
1238 uint16_t subchannel_nr, uint32_t io_int_parm,
1239 uint32_t io_int_word)
1241 struct kvm_s390_irq irq = {
1242 .u.io.subchannel_id = subchannel_id,
1243 .u.io.subchannel_nr = subchannel_nr,
1244 .u.io.io_int_parm = io_int_parm,
1245 .u.io.io_int_word = io_int_word,
1248 if (io_int_word & IO_INT_WORD_AI) {
1249 irq.type = KVM_S390_INT_IO(1, 0, 0, 0);
1250 } else {
1251 irq.type = ((subchannel_id & 0xff00) << 24) |
1252 ((subchannel_id & 0x00060) << 22) | (subchannel_nr << 16);
1254 kvm_s390_floating_interrupt(&irq);
1257 void kvm_s390_crw_mchk(void)
1259 struct kvm_s390_irq irq = {
1260 .type = KVM_S390_MCHK,
1261 .u.mchk.cr14 = 1 << 28,
1262 .u.mchk.mcic = 0x00400f1d40330000,
1264 kvm_s390_floating_interrupt(&irq);
1267 void kvm_s390_enable_css_support(S390CPU *cpu)
1269 int r;
1271 /* Activate host kernel channel subsystem support. */
1272 r = kvm_vcpu_enable_cap(CPU(cpu), KVM_CAP_S390_CSS_SUPPORT, 0);
1273 assert(r == 0);
1276 void kvm_arch_init_irq_routing(KVMState *s)
1279 * Note that while irqchip capabilities generally imply that cpustates
1280 * are handled in-kernel, it is not true for s390 (yet); therefore, we
1281 * have to override the common code kvm_halt_in_kernel_allowed setting.
1283 if (kvm_check_extension(s, KVM_CAP_IRQ_ROUTING)) {
1284 kvm_irqfds_allowed = true;
1285 kvm_gsi_routing_allowed = true;
1286 kvm_halt_in_kernel_allowed = false;
1290 int kvm_s390_assign_subch_ioeventfd(EventNotifier *notifier, uint32_t sch,
1291 int vq, bool assign)
1293 struct kvm_ioeventfd kick = {
1294 .flags = KVM_IOEVENTFD_FLAG_VIRTIO_CCW_NOTIFY |
1295 KVM_IOEVENTFD_FLAG_DATAMATCH,
1296 .fd = event_notifier_get_fd(notifier),
1297 .datamatch = vq,
1298 .addr = sch,
1299 .len = 8,
1301 if (!kvm_check_extension(kvm_state, KVM_CAP_IOEVENTFD)) {
1302 return -ENOSYS;
1304 if (!assign) {
1305 kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
1307 return kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);