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 <sys/types.h>
25 #include <sys/ioctl.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"
37 #include "sysemu/device_tree.h"
38 #include "qapi/qmp/qjson.h"
39 #include "monitor/monitor.h"
40 #include "exec/gdbstub.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"
47 /* #define DEBUG_KVM */
50 #define DPRINTF(fmt, ...) \
51 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
53 #define DPRINTF(fmt, ...) \
57 #define IPA0_DIAG 0x8300
58 #define IPA0_SIGP 0xae00
59 #define IPA0_B2 0xb200
60 #define IPA0_B9 0xb900
61 #define IPA0_EB 0xeb00
62 #define IPA0_E3 0xe300
64 #define PRIV_B2_SCLP_CALL 0x20
65 #define PRIV_B2_CSCH 0x30
66 #define PRIV_B2_HSCH 0x31
67 #define PRIV_B2_MSCH 0x32
68 #define PRIV_B2_SSCH 0x33
69 #define PRIV_B2_STSCH 0x34
70 #define PRIV_B2_TSCH 0x35
71 #define PRIV_B2_TPI 0x36
72 #define PRIV_B2_SAL 0x37
73 #define PRIV_B2_RSCH 0x38
74 #define PRIV_B2_STCRW 0x39
75 #define PRIV_B2_STCPS 0x3a
76 #define PRIV_B2_RCHP 0x3b
77 #define PRIV_B2_SCHM 0x3c
78 #define PRIV_B2_CHSC 0x5f
79 #define PRIV_B2_SIGA 0x74
80 #define PRIV_B2_XSCH 0x76
82 #define PRIV_EB_SQBS 0x8a
83 #define PRIV_EB_PCISTB 0xd0
84 #define PRIV_EB_SIC 0xd1
86 #define PRIV_B9_EQBS 0x9c
87 #define PRIV_B9_CLP 0xa0
88 #define PRIV_B9_PCISTG 0xd0
89 #define PRIV_B9_PCILG 0xd2
90 #define PRIV_B9_RPCIT 0xd3
92 #define PRIV_E3_MPCIFC 0xd0
93 #define PRIV_E3_STPCIFC 0xd4
95 #define DIAG_IPL 0x308
96 #define DIAG_KVM_HYPERCALL 0x500
97 #define DIAG_KVM_BREAKPOINT 0x501
99 #define ICPT_INSTRUCTION 0x04
100 #define ICPT_PROGRAM 0x08
101 #define ICPT_EXT_INT 0x14
102 #define ICPT_WAITPSW 0x1c
103 #define ICPT_SOFT_INTERCEPT 0x24
104 #define ICPT_CPU_STOP 0x28
107 static CPUWatchpoint hw_watchpoint;
109 * We don't use a list because this structure is also used to transmit the
110 * hardware breakpoints to the kernel.
112 static struct kvm_hw_breakpoint *hw_breakpoints;
113 static int nb_hw_breakpoints;
115 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
119 static int cap_sync_regs;
120 static int cap_async_pf;
122 static void *legacy_s390_alloc(size_t size, uint64_t *align);
124 static int kvm_s390_check_clear_cmma(KVMState *s)
126 struct kvm_device_attr attr = {
127 .group = KVM_S390_VM_MEM_CTRL,
128 .attr = KVM_S390_VM_MEM_CLR_CMMA,
131 return kvm_vm_ioctl(s, KVM_HAS_DEVICE_ATTR, &attr);
134 static int kvm_s390_check_enable_cmma(KVMState *s)
136 struct kvm_device_attr attr = {
137 .group = KVM_S390_VM_MEM_CTRL,
138 .attr = KVM_S390_VM_MEM_ENABLE_CMMA,
141 return kvm_vm_ioctl(s, KVM_HAS_DEVICE_ATTR, &attr);
144 void kvm_s390_clear_cmma_callback(void *opaque)
147 KVMState *s = opaque;
148 struct kvm_device_attr attr = {
149 .group = KVM_S390_VM_MEM_CTRL,
150 .attr = KVM_S390_VM_MEM_CLR_CMMA,
153 rc = kvm_vm_ioctl(s, KVM_SET_DEVICE_ATTR, &attr);
154 trace_kvm_clear_cmma(rc);
157 static void kvm_s390_enable_cmma(KVMState *s)
160 struct kvm_device_attr attr = {
161 .group = KVM_S390_VM_MEM_CTRL,
162 .attr = KVM_S390_VM_MEM_ENABLE_CMMA,
165 if (kvm_s390_check_enable_cmma(s) || kvm_s390_check_clear_cmma(s)) {
169 rc = kvm_vm_ioctl(s, KVM_SET_DEVICE_ATTR, &attr);
171 qemu_register_reset(kvm_s390_clear_cmma_callback, s);
173 trace_kvm_enable_cmma(rc);
176 int kvm_arch_init(KVMState *s)
178 cap_sync_regs = kvm_check_extension(s, KVM_CAP_SYNC_REGS);
179 cap_async_pf = kvm_check_extension(s, KVM_CAP_ASYNC_PF);
181 if (kvm_check_extension(s, KVM_CAP_VM_ATTRIBUTES)) {
182 kvm_s390_enable_cmma(s);
185 if (!kvm_check_extension(s, KVM_CAP_S390_GMAP)
186 || !kvm_check_extension(s, KVM_CAP_S390_COW)) {
187 phys_mem_set_alloc(legacy_s390_alloc);
192 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
194 return cpu->cpu_index;
197 int kvm_arch_init_vcpu(CPUState *cs)
199 S390CPU *cpu = S390_CPU(cs);
200 kvm_s390_set_cpu_state(cpu, cpu->env.cpu_state);
204 void kvm_s390_reset_vcpu(S390CPU *cpu)
206 CPUState *cs = CPU(cpu);
208 /* The initial reset call is needed here to reset in-kernel
209 * vcpu data that we can't access directly from QEMU
210 * (i.e. with older kernels which don't support sync_regs/ONE_REG).
211 * Before this ioctl cpu_synchronize_state() is called in common kvm
213 if (kvm_vcpu_ioctl(cs, KVM_S390_INITIAL_RESET, NULL)) {
214 error_report("Initial CPU reset failed on CPU %i\n", cs->cpu_index);
218 static int can_sync_regs(CPUState *cs, int regs)
220 return cap_sync_regs && (cs->kvm_run->kvm_valid_regs & regs) == regs;
223 int kvm_arch_put_registers(CPUState *cs, int level)
225 S390CPU *cpu = S390_CPU(cs);
226 CPUS390XState *env = &cpu->env;
227 struct kvm_sregs sregs;
228 struct kvm_regs regs;
229 struct kvm_fpu fpu = {};
233 /* always save the PSW and the GPRS*/
234 cs->kvm_run->psw_addr = env->psw.addr;
235 cs->kvm_run->psw_mask = env->psw.mask;
237 if (can_sync_regs(cs, KVM_SYNC_GPRS)) {
238 for (i = 0; i < 16; i++) {
239 cs->kvm_run->s.regs.gprs[i] = env->regs[i];
240 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_GPRS;
243 for (i = 0; i < 16; i++) {
244 regs.gprs[i] = env->regs[i];
246 r = kvm_vcpu_ioctl(cs, KVM_SET_REGS, ®s);
253 for (i = 0; i < 16; i++) {
254 fpu.fprs[i] = env->fregs[i].ll;
258 r = kvm_vcpu_ioctl(cs, KVM_SET_FPU, &fpu);
263 /* Do we need to save more than that? */
264 if (level == KVM_PUT_RUNTIME_STATE) {
268 if (can_sync_regs(cs, KVM_SYNC_ARCH0)) {
269 cs->kvm_run->s.regs.cputm = env->cputm;
270 cs->kvm_run->s.regs.ckc = env->ckc;
271 cs->kvm_run->s.regs.todpr = env->todpr;
272 cs->kvm_run->s.regs.gbea = env->gbea;
273 cs->kvm_run->s.regs.pp = env->pp;
274 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ARCH0;
277 * These ONE_REGS are not protected by a capability. As they are only
278 * necessary for migration we just trace a possible error, but don't
279 * return with an error return code.
281 kvm_set_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
282 kvm_set_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
283 kvm_set_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
284 kvm_set_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
285 kvm_set_one_reg(cs, KVM_REG_S390_PP, &env->pp);
288 /* pfault parameters */
289 if (can_sync_regs(cs, KVM_SYNC_PFAULT)) {
290 cs->kvm_run->s.regs.pft = env->pfault_token;
291 cs->kvm_run->s.regs.pfs = env->pfault_select;
292 cs->kvm_run->s.regs.pfc = env->pfault_compare;
293 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_PFAULT;
294 } else if (cap_async_pf) {
295 r = kvm_set_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
299 r = kvm_set_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
303 r = kvm_set_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
309 /* access registers and control registers*/
310 if (can_sync_regs(cs, KVM_SYNC_ACRS | KVM_SYNC_CRS)) {
311 for (i = 0; i < 16; i++) {
312 cs->kvm_run->s.regs.acrs[i] = env->aregs[i];
313 cs->kvm_run->s.regs.crs[i] = env->cregs[i];
315 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ACRS;
316 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_CRS;
318 for (i = 0; i < 16; i++) {
319 sregs.acrs[i] = env->aregs[i];
320 sregs.crs[i] = env->cregs[i];
322 r = kvm_vcpu_ioctl(cs, KVM_SET_SREGS, &sregs);
328 /* Finally the prefix */
329 if (can_sync_regs(cs, KVM_SYNC_PREFIX)) {
330 cs->kvm_run->s.regs.prefix = env->psa;
331 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_PREFIX;
333 /* prefix is only supported via sync regs */
338 int kvm_arch_get_registers(CPUState *cs)
340 S390CPU *cpu = S390_CPU(cs);
341 CPUS390XState *env = &cpu->env;
342 struct kvm_sregs sregs;
343 struct kvm_regs regs;
348 env->psw.addr = cs->kvm_run->psw_addr;
349 env->psw.mask = cs->kvm_run->psw_mask;
352 if (can_sync_regs(cs, KVM_SYNC_GPRS)) {
353 for (i = 0; i < 16; i++) {
354 env->regs[i] = cs->kvm_run->s.regs.gprs[i];
357 r = kvm_vcpu_ioctl(cs, KVM_GET_REGS, ®s);
361 for (i = 0; i < 16; i++) {
362 env->regs[i] = regs.gprs[i];
366 /* The ACRS and CRS */
367 if (can_sync_regs(cs, KVM_SYNC_ACRS | KVM_SYNC_CRS)) {
368 for (i = 0; i < 16; i++) {
369 env->aregs[i] = cs->kvm_run->s.regs.acrs[i];
370 env->cregs[i] = cs->kvm_run->s.regs.crs[i];
373 r = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs);
377 for (i = 0; i < 16; i++) {
378 env->aregs[i] = sregs.acrs[i];
379 env->cregs[i] = sregs.crs[i];
384 r = kvm_vcpu_ioctl(cs, KVM_GET_FPU, &fpu);
388 for (i = 0; i < 16; i++) {
389 env->fregs[i].ll = fpu.fprs[i];
394 if (can_sync_regs(cs, KVM_SYNC_PREFIX)) {
395 env->psa = cs->kvm_run->s.regs.prefix;
398 if (can_sync_regs(cs, KVM_SYNC_ARCH0)) {
399 env->cputm = cs->kvm_run->s.regs.cputm;
400 env->ckc = cs->kvm_run->s.regs.ckc;
401 env->todpr = cs->kvm_run->s.regs.todpr;
402 env->gbea = cs->kvm_run->s.regs.gbea;
403 env->pp = cs->kvm_run->s.regs.pp;
406 * These ONE_REGS are not protected by a capability. As they are only
407 * necessary for migration we just trace a possible error, but don't
408 * return with an error return code.
410 kvm_get_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
411 kvm_get_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
412 kvm_get_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
413 kvm_get_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
414 kvm_get_one_reg(cs, KVM_REG_S390_PP, &env->pp);
417 /* pfault parameters */
418 if (can_sync_regs(cs, KVM_SYNC_PFAULT)) {
419 env->pfault_token = cs->kvm_run->s.regs.pft;
420 env->pfault_select = cs->kvm_run->s.regs.pfs;
421 env->pfault_compare = cs->kvm_run->s.regs.pfc;
422 } else if (cap_async_pf) {
423 r = kvm_get_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
427 r = kvm_get_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
431 r = kvm_get_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
441 * Legacy layout for s390:
442 * Older S390 KVM requires the topmost vma of the RAM to be
443 * smaller than an system defined value, which is at least 256GB.
444 * Larger systems have larger values. We put the guest between
445 * the end of data segment (system break) and this value. We
446 * use 32GB as a base to have enough room for the system break
447 * to grow. We also have to use MAP parameters that avoid
448 * read-only mapping of guest pages.
450 static void *legacy_s390_alloc(size_t size, uint64_t *align)
454 mem = mmap((void *) 0x800000000ULL, size,
455 PROT_EXEC|PROT_READ|PROT_WRITE,
456 MAP_SHARED | MAP_ANONYMOUS | MAP_FIXED, -1, 0);
457 return mem == MAP_FAILED ? NULL : mem;
460 /* DIAG 501 is used for sw breakpoints */
461 static const uint8_t diag_501[] = {0x83, 0x24, 0x05, 0x01};
463 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
466 if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
467 sizeof(diag_501), 0) ||
468 cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)diag_501,
469 sizeof(diag_501), 1)) {
475 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
477 uint8_t t[sizeof(diag_501)];
479 if (cpu_memory_rw_debug(cs, bp->pc, t, sizeof(diag_501), 0)) {
481 } else if (memcmp(t, diag_501, sizeof(diag_501))) {
483 } else if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
484 sizeof(diag_501), 1)) {
491 static struct kvm_hw_breakpoint *find_hw_breakpoint(target_ulong addr,
496 for (n = 0; n < nb_hw_breakpoints; n++) {
497 if (hw_breakpoints[n].addr == addr && hw_breakpoints[n].type == type &&
498 (hw_breakpoints[n].len == len || len == -1)) {
499 return &hw_breakpoints[n];
506 static int insert_hw_breakpoint(target_ulong addr, int len, int type)
510 if (find_hw_breakpoint(addr, len, type)) {
514 size = (nb_hw_breakpoints + 1) * sizeof(struct kvm_hw_breakpoint);
516 if (!hw_breakpoints) {
517 nb_hw_breakpoints = 0;
518 hw_breakpoints = (struct kvm_hw_breakpoint *)g_try_malloc(size);
521 (struct kvm_hw_breakpoint *)g_try_realloc(hw_breakpoints, size);
524 if (!hw_breakpoints) {
525 nb_hw_breakpoints = 0;
529 hw_breakpoints[nb_hw_breakpoints].addr = addr;
530 hw_breakpoints[nb_hw_breakpoints].len = len;
531 hw_breakpoints[nb_hw_breakpoints].type = type;
538 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
539 target_ulong len, int type)
542 case GDB_BREAKPOINT_HW:
545 case GDB_WATCHPOINT_WRITE:
549 type = KVM_HW_WP_WRITE;
554 return insert_hw_breakpoint(addr, len, type);
557 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
558 target_ulong len, int type)
561 struct kvm_hw_breakpoint *bp = find_hw_breakpoint(addr, len, type);
568 if (nb_hw_breakpoints > 0) {
570 * In order to trim the array, move the last element to the position to
571 * be removed - if necessary.
573 if (bp != &hw_breakpoints[nb_hw_breakpoints]) {
574 *bp = hw_breakpoints[nb_hw_breakpoints];
576 size = nb_hw_breakpoints * sizeof(struct kvm_hw_breakpoint);
578 (struct kvm_hw_breakpoint *)g_realloc(hw_breakpoints, size);
580 g_free(hw_breakpoints);
581 hw_breakpoints = NULL;
587 void kvm_arch_remove_all_hw_breakpoints(void)
589 nb_hw_breakpoints = 0;
590 g_free(hw_breakpoints);
591 hw_breakpoints = NULL;
594 void kvm_arch_update_guest_debug(CPUState *cpu, struct kvm_guest_debug *dbg)
598 if (nb_hw_breakpoints > 0) {
599 dbg->arch.nr_hw_bp = nb_hw_breakpoints;
600 dbg->arch.hw_bp = hw_breakpoints;
602 for (i = 0; i < nb_hw_breakpoints; ++i) {
603 hw_breakpoints[i].phys_addr = s390_cpu_get_phys_addr_debug(cpu,
604 hw_breakpoints[i].addr);
606 dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
608 dbg->arch.nr_hw_bp = 0;
609 dbg->arch.hw_bp = NULL;
613 void kvm_arch_pre_run(CPUState *cpu, struct kvm_run *run)
617 void kvm_arch_post_run(CPUState *cpu, struct kvm_run *run)
621 int kvm_arch_process_async_events(CPUState *cs)
626 static int s390_kvm_irq_to_interrupt(struct kvm_s390_irq *irq,
627 struct kvm_s390_interrupt *interrupt)
631 interrupt->type = irq->type;
633 case KVM_S390_INT_VIRTIO:
634 interrupt->parm = irq->u.ext.ext_params;
636 case KVM_S390_INT_PFAULT_INIT:
637 case KVM_S390_INT_PFAULT_DONE:
638 interrupt->parm64 = irq->u.ext.ext_params2;
640 case KVM_S390_PROGRAM_INT:
641 interrupt->parm = irq->u.pgm.code;
643 case KVM_S390_SIGP_SET_PREFIX:
644 interrupt->parm = irq->u.prefix.address;
646 case KVM_S390_INT_SERVICE:
647 interrupt->parm = irq->u.ext.ext_params;
650 interrupt->parm = irq->u.mchk.cr14;
651 interrupt->parm64 = irq->u.mchk.mcic;
653 case KVM_S390_INT_EXTERNAL_CALL:
654 interrupt->parm = irq->u.extcall.code;
656 case KVM_S390_INT_EMERGENCY:
657 interrupt->parm = irq->u.emerg.code;
659 case KVM_S390_SIGP_STOP:
660 case KVM_S390_RESTART:
661 break; /* These types have no parameters */
662 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
663 interrupt->parm = irq->u.io.subchannel_id << 16;
664 interrupt->parm |= irq->u.io.subchannel_nr;
665 interrupt->parm64 = (uint64_t)irq->u.io.io_int_parm << 32;
666 interrupt->parm64 |= irq->u.io.io_int_word;
675 void kvm_s390_vcpu_interrupt(S390CPU *cpu, struct kvm_s390_irq *irq)
677 struct kvm_s390_interrupt kvmint = {};
678 CPUState *cs = CPU(cpu);
681 r = s390_kvm_irq_to_interrupt(irq, &kvmint);
683 fprintf(stderr, "%s called with bogus interrupt\n", __func__);
687 r = kvm_vcpu_ioctl(cs, KVM_S390_INTERRUPT, &kvmint);
689 fprintf(stderr, "KVM failed to inject interrupt\n");
694 static void __kvm_s390_floating_interrupt(struct kvm_s390_irq *irq)
696 struct kvm_s390_interrupt kvmint = {};
699 r = s390_kvm_irq_to_interrupt(irq, &kvmint);
701 fprintf(stderr, "%s called with bogus interrupt\n", __func__);
705 r = kvm_vm_ioctl(kvm_state, KVM_S390_INTERRUPT, &kvmint);
707 fprintf(stderr, "KVM failed to inject interrupt\n");
712 void kvm_s390_floating_interrupt(struct kvm_s390_irq *irq)
714 static bool use_flic = true;
718 r = kvm_s390_inject_flic(irq);
726 __kvm_s390_floating_interrupt(irq);
729 void kvm_s390_virtio_irq(int config_change, uint64_t token)
731 struct kvm_s390_irq irq = {
732 .type = KVM_S390_INT_VIRTIO,
733 .u.ext.ext_params = config_change,
734 .u.ext.ext_params2 = token,
737 kvm_s390_floating_interrupt(&irq);
740 void kvm_s390_service_interrupt(uint32_t parm)
742 struct kvm_s390_irq irq = {
743 .type = KVM_S390_INT_SERVICE,
744 .u.ext.ext_params = parm,
747 kvm_s390_floating_interrupt(&irq);
750 static void enter_pgmcheck(S390CPU *cpu, uint16_t code)
752 struct kvm_s390_irq irq = {
753 .type = KVM_S390_PROGRAM_INT,
757 kvm_s390_vcpu_interrupt(cpu, &irq);
760 static int kvm_sclp_service_call(S390CPU *cpu, struct kvm_run *run,
763 CPUS390XState *env = &cpu->env;
768 cpu_synchronize_state(CPU(cpu));
769 sccb = env->regs[ipbh0 & 0xf];
770 code = env->regs[(ipbh0 & 0xf0) >> 4];
772 r = sclp_service_call(env, sccb, code);
774 enter_pgmcheck(cpu, -r);
782 static int handle_b2(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
784 CPUS390XState *env = &cpu->env;
786 uint16_t ipbh0 = (run->s390_sieic.ipb & 0xffff0000) >> 16;
788 cpu_synchronize_state(CPU(cpu));
792 ioinst_handle_xsch(cpu, env->regs[1]);
795 ioinst_handle_csch(cpu, env->regs[1]);
798 ioinst_handle_hsch(cpu, env->regs[1]);
801 ioinst_handle_msch(cpu, env->regs[1], run->s390_sieic.ipb);
804 ioinst_handle_ssch(cpu, env->regs[1], run->s390_sieic.ipb);
807 ioinst_handle_stcrw(cpu, run->s390_sieic.ipb);
810 ioinst_handle_stsch(cpu, env->regs[1], run->s390_sieic.ipb);
813 /* We should only get tsch via KVM_EXIT_S390_TSCH. */
814 fprintf(stderr, "Spurious tsch intercept\n");
817 ioinst_handle_chsc(cpu, run->s390_sieic.ipb);
820 /* This should have been handled by kvm already. */
821 fprintf(stderr, "Spurious tpi intercept\n");
824 ioinst_handle_schm(cpu, env->regs[1], env->regs[2],
825 run->s390_sieic.ipb);
828 ioinst_handle_rsch(cpu, env->regs[1]);
831 ioinst_handle_rchp(cpu, env->regs[1]);
834 /* We do not provide this instruction, it is suppressed. */
837 ioinst_handle_sal(cpu, env->regs[1]);
840 /* Not provided, set CC = 3 for subchannel not operational */
843 case PRIV_B2_SCLP_CALL:
844 rc = kvm_sclp_service_call(cpu, run, ipbh0);
848 DPRINTF("KVM: unhandled PRIV: 0xb2%x\n", ipa1);
855 static uint64_t get_base_disp_rxy(S390CPU *cpu, struct kvm_run *run)
857 CPUS390XState *env = &cpu->env;
858 uint32_t x2 = (run->s390_sieic.ipa & 0x000f);
859 uint32_t base2 = run->s390_sieic.ipb >> 28;
860 uint32_t disp2 = ((run->s390_sieic.ipb & 0x0fff0000) >> 16) +
861 ((run->s390_sieic.ipb & 0xff00) << 4);
863 if (disp2 & 0x80000) {
867 return (base2 ? env->regs[base2] : 0) +
868 (x2 ? env->regs[x2] : 0) + (long)(int)disp2;
871 static uint64_t get_base_disp_rsy(S390CPU *cpu, struct kvm_run *run)
873 CPUS390XState *env = &cpu->env;
874 uint32_t base2 = run->s390_sieic.ipb >> 28;
875 uint32_t disp2 = ((run->s390_sieic.ipb & 0x0fff0000) >> 16) +
876 ((run->s390_sieic.ipb & 0xff00) << 4);
878 if (disp2 & 0x80000) {
882 return (base2 ? env->regs[base2] : 0) + (long)(int)disp2;
885 static int kvm_clp_service_call(S390CPU *cpu, struct kvm_run *run)
887 uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
889 return clp_service_call(cpu, r2);
892 static int kvm_pcilg_service_call(S390CPU *cpu, struct kvm_run *run)
894 uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
895 uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
897 return pcilg_service_call(cpu, r1, r2);
900 static int kvm_pcistg_service_call(S390CPU *cpu, struct kvm_run *run)
902 uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
903 uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
905 return pcistg_service_call(cpu, r1, r2);
908 static int kvm_stpcifc_service_call(S390CPU *cpu, struct kvm_run *run)
910 uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
913 cpu_synchronize_state(CPU(cpu));
914 fiba = get_base_disp_rxy(cpu, run);
916 return stpcifc_service_call(cpu, r1, fiba);
919 static int kvm_sic_service_call(S390CPU *cpu, struct kvm_run *run)
925 static int kvm_rpcit_service_call(S390CPU *cpu, struct kvm_run *run)
927 uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
928 uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
930 return rpcit_service_call(cpu, r1, r2);
933 static int kvm_pcistb_service_call(S390CPU *cpu, struct kvm_run *run)
935 uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
936 uint8_t r3 = run->s390_sieic.ipa & 0x000f;
939 cpu_synchronize_state(CPU(cpu));
940 gaddr = get_base_disp_rsy(cpu, run);
942 return pcistb_service_call(cpu, r1, r3, gaddr);
945 static int kvm_mpcifc_service_call(S390CPU *cpu, struct kvm_run *run)
947 uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
950 cpu_synchronize_state(CPU(cpu));
951 fiba = get_base_disp_rxy(cpu, run);
953 return mpcifc_service_call(cpu, r1, fiba);
956 static int handle_b9(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
962 r = kvm_clp_service_call(cpu, run);
965 r = kvm_pcistg_service_call(cpu, run);
968 r = kvm_pcilg_service_call(cpu, run);
971 r = kvm_rpcit_service_call(cpu, run);
974 /* just inject exception */
979 DPRINTF("KVM: unhandled PRIV: 0xb9%x\n", ipa1);
986 static int handle_eb(S390CPU *cpu, struct kvm_run *run, uint8_t ipbl)
992 r = kvm_pcistb_service_call(cpu, run);
995 r = kvm_sic_service_call(cpu, run);
998 /* just inject exception */
1003 DPRINTF("KVM: unhandled PRIV: 0xeb%x\n", ipbl);
1010 static int handle_e3(S390CPU *cpu, struct kvm_run *run, uint8_t ipbl)
1015 case PRIV_E3_MPCIFC:
1016 r = kvm_mpcifc_service_call(cpu, run);
1018 case PRIV_E3_STPCIFC:
1019 r = kvm_stpcifc_service_call(cpu, run);
1023 DPRINTF("KVM: unhandled PRIV: 0xe3%x\n", ipbl);
1030 static int handle_hypercall(S390CPU *cpu, struct kvm_run *run)
1032 CPUS390XState *env = &cpu->env;
1035 cpu_synchronize_state(CPU(cpu));
1036 ret = s390_virtio_hypercall(env);
1037 if (ret == -EINVAL) {
1038 enter_pgmcheck(cpu, PGM_SPECIFICATION);
1045 static void kvm_handle_diag_308(S390CPU *cpu, struct kvm_run *run)
1049 cpu_synchronize_state(CPU(cpu));
1050 r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1051 r3 = run->s390_sieic.ipa & 0x000f;
1052 handle_diag_308(&cpu->env, r1, r3);
1055 static int handle_sw_breakpoint(S390CPU *cpu, struct kvm_run *run)
1057 CPUS390XState *env = &cpu->env;
1060 cpu_synchronize_state(CPU(cpu));
1062 pc = env->psw.addr - 4;
1063 if (kvm_find_sw_breakpoint(CPU(cpu), pc)) {
1071 #define DIAG_KVM_CODE_MASK 0x000000000000ffff
1073 static int handle_diag(S390CPU *cpu, struct kvm_run *run, uint32_t ipb)
1079 * For any diagnose call we support, bits 48-63 of the resulting
1080 * address specify the function code; the remainder is ignored.
1082 func_code = decode_basedisp_rs(&cpu->env, ipb) & DIAG_KVM_CODE_MASK;
1083 switch (func_code) {
1085 kvm_handle_diag_308(cpu, run);
1087 case DIAG_KVM_HYPERCALL:
1088 r = handle_hypercall(cpu, run);
1090 case DIAG_KVM_BREAKPOINT:
1091 r = handle_sw_breakpoint(cpu, run);
1094 DPRINTF("KVM: unknown DIAG: 0x%x\n", func_code);
1095 enter_pgmcheck(cpu, PGM_SPECIFICATION);
1102 static void sigp_cpu_start(void *arg)
1105 S390CPU *cpu = S390_CPU(cs);
1107 s390_cpu_set_state(CPU_STATE_OPERATING, cpu);
1108 DPRINTF("DONE: KVM cpu start: %p\n", &cpu->env);
1111 static void sigp_cpu_restart(void *arg)
1114 S390CPU *cpu = S390_CPU(cs);
1115 struct kvm_s390_irq irq = {
1116 .type = KVM_S390_RESTART,
1119 kvm_s390_vcpu_interrupt(cpu, &irq);
1120 s390_cpu_set_state(CPU_STATE_OPERATING, cpu);
1123 int kvm_s390_cpu_restart(S390CPU *cpu)
1125 run_on_cpu(CPU(cpu), sigp_cpu_restart, CPU(cpu));
1126 DPRINTF("DONE: KVM cpu restart: %p\n", &cpu->env);
1130 static void sigp_initial_cpu_reset(void *arg)
1132 CPUState *cpu = arg;
1133 S390CPUClass *scc = S390_CPU_GET_CLASS(cpu);
1135 cpu_synchronize_state(cpu);
1136 scc->initial_cpu_reset(cpu);
1137 cpu_synchronize_post_reset(cpu);
1140 static void sigp_cpu_reset(void *arg)
1142 CPUState *cpu = arg;
1143 S390CPUClass *scc = S390_CPU_GET_CLASS(cpu);
1145 cpu_synchronize_state(cpu);
1146 scc->cpu_reset(cpu);
1147 cpu_synchronize_post_reset(cpu);
1150 #define SIGP_ORDER_MASK 0x000000ff
1152 static int handle_sigp(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1154 CPUS390XState *env = &cpu->env;
1157 S390CPU *target_cpu;
1158 uint64_t *statusreg = &env->regs[ipa1 >> 4];
1161 cpu_synchronize_state(CPU(cpu));
1163 /* get order code */
1164 order_code = decode_basedisp_rs(env, run->s390_sieic.ipb) & SIGP_ORDER_MASK;
1166 cpu_addr = env->regs[ipa1 & 0x0f];
1167 target_cpu = s390_cpu_addr2state(cpu_addr);
1168 if (target_cpu == NULL) {
1169 cc = 3; /* not operational */
1173 switch (order_code) {
1175 run_on_cpu(CPU(target_cpu), sigp_cpu_start, CPU(target_cpu));
1179 run_on_cpu(CPU(target_cpu), sigp_cpu_restart, CPU(target_cpu));
1183 *statusreg &= 0xffffffff00000000UL;
1184 *statusreg |= SIGP_STAT_INVALID_PARAMETER;
1185 cc = 1; /* status stored */
1187 case SIGP_INITIAL_CPU_RESET:
1188 run_on_cpu(CPU(target_cpu), sigp_initial_cpu_reset, CPU(target_cpu));
1191 case SIGP_CPU_RESET:
1192 run_on_cpu(CPU(target_cpu), sigp_cpu_reset, CPU(target_cpu));
1196 DPRINTF("KVM: unknown SIGP: 0x%x\n", order_code);
1197 *statusreg &= 0xffffffff00000000UL;
1198 *statusreg |= SIGP_STAT_INVALID_ORDER;
1199 cc = 1; /* status stored */
1208 static int handle_instruction(S390CPU *cpu, struct kvm_run *run)
1210 unsigned int ipa0 = (run->s390_sieic.ipa & 0xff00);
1211 uint8_t ipa1 = run->s390_sieic.ipa & 0x00ff;
1214 DPRINTF("handle_instruction 0x%x 0x%x\n",
1215 run->s390_sieic.ipa, run->s390_sieic.ipb);
1218 r = handle_b2(cpu, run, ipa1);
1221 r = handle_b9(cpu, run, ipa1);
1224 r = handle_eb(cpu, run, run->s390_sieic.ipb & 0xff);
1227 r = handle_e3(cpu, run, run->s390_sieic.ipb & 0xff);
1230 r = handle_diag(cpu, run, run->s390_sieic.ipb);
1233 r = handle_sigp(cpu, run, ipa1);
1239 enter_pgmcheck(cpu, 0x0001);
1245 static bool is_special_wait_psw(CPUState *cs)
1247 /* signal quiesce */
1248 return cs->kvm_run->psw_addr == 0xfffUL;
1251 static void guest_panicked(void)
1253 qapi_event_send_guest_panicked(GUEST_PANIC_ACTION_PAUSE,
1255 vm_stop(RUN_STATE_GUEST_PANICKED);
1258 static void unmanageable_intercept(S390CPU *cpu, const char *str, int pswoffset)
1260 CPUState *cs = CPU(cpu);
1262 error_report("Unmanageable %s! CPU%i new PSW: 0x%016lx:%016lx",
1263 str, cs->cpu_index, ldq_phys(cs->as, cpu->env.psa + pswoffset),
1264 ldq_phys(cs->as, cpu->env.psa + pswoffset + 8));
1269 static int handle_intercept(S390CPU *cpu)
1271 CPUState *cs = CPU(cpu);
1272 struct kvm_run *run = cs->kvm_run;
1273 int icpt_code = run->s390_sieic.icptcode;
1276 DPRINTF("intercept: 0x%x (at 0x%lx)\n", icpt_code,
1277 (long)cs->kvm_run->psw_addr);
1278 switch (icpt_code) {
1279 case ICPT_INSTRUCTION:
1280 r = handle_instruction(cpu, run);
1283 unmanageable_intercept(cpu, "program interrupt",
1284 offsetof(LowCore, program_new_psw));
1288 unmanageable_intercept(cpu, "external interrupt",
1289 offsetof(LowCore, external_new_psw));
1293 /* disabled wait, since enabled wait is handled in kernel */
1294 cpu_synchronize_state(cs);
1295 if (s390_cpu_halt(cpu) == 0) {
1296 if (is_special_wait_psw(cs)) {
1297 qemu_system_shutdown_request();
1305 if (s390_cpu_set_state(CPU_STATE_STOPPED, cpu) == 0) {
1306 qemu_system_shutdown_request();
1310 case ICPT_SOFT_INTERCEPT:
1311 fprintf(stderr, "KVM unimplemented icpt SOFT\n");
1315 fprintf(stderr, "KVM unimplemented icpt IO\n");
1319 fprintf(stderr, "Unknown intercept code: %d\n", icpt_code);
1327 static int handle_tsch(S390CPU *cpu)
1329 CPUS390XState *env = &cpu->env;
1330 CPUState *cs = CPU(cpu);
1331 struct kvm_run *run = cs->kvm_run;
1334 cpu_synchronize_state(cs);
1336 ret = ioinst_handle_tsch(env, env->regs[1], run->s390_tsch.ipb);
1338 /* Success; set condition code. */
1341 } else if (ret < -1) {
1344 * If an I/O interrupt had been dequeued, we have to reinject it.
1346 if (run->s390_tsch.dequeued) {
1347 kvm_s390_io_interrupt(run->s390_tsch.subchannel_id,
1348 run->s390_tsch.subchannel_nr,
1349 run->s390_tsch.io_int_parm,
1350 run->s390_tsch.io_int_word);
1357 static int kvm_arch_handle_debug_exit(S390CPU *cpu)
1359 CPUState *cs = CPU(cpu);
1360 struct kvm_run *run = cs->kvm_run;
1363 struct kvm_debug_exit_arch *arch_info = &run->debug.arch;
1365 switch (arch_info->type) {
1366 case KVM_HW_WP_WRITE:
1367 if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
1368 cs->watchpoint_hit = &hw_watchpoint;
1369 hw_watchpoint.vaddr = arch_info->addr;
1370 hw_watchpoint.flags = BP_MEM_WRITE;
1375 if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
1379 case KVM_SINGLESTEP:
1380 if (cs->singlestep_enabled) {
1391 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1393 S390CPU *cpu = S390_CPU(cs);
1396 switch (run->exit_reason) {
1397 case KVM_EXIT_S390_SIEIC:
1398 ret = handle_intercept(cpu);
1400 case KVM_EXIT_S390_RESET:
1401 s390_reipl_request();
1403 case KVM_EXIT_S390_TSCH:
1404 ret = handle_tsch(cpu);
1406 case KVM_EXIT_DEBUG:
1407 ret = kvm_arch_handle_debug_exit(cpu);
1410 fprintf(stderr, "Unknown KVM exit: %d\n", run->exit_reason);
1415 ret = EXCP_INTERRUPT;
1420 bool kvm_arch_stop_on_emulation_error(CPUState *cpu)
1425 int kvm_arch_on_sigbus_vcpu(CPUState *cpu, int code, void *addr)
1430 int kvm_arch_on_sigbus(int code, void *addr)
1435 void kvm_s390_io_interrupt(uint16_t subchannel_id,
1436 uint16_t subchannel_nr, uint32_t io_int_parm,
1437 uint32_t io_int_word)
1439 struct kvm_s390_irq irq = {
1440 .u.io.subchannel_id = subchannel_id,
1441 .u.io.subchannel_nr = subchannel_nr,
1442 .u.io.io_int_parm = io_int_parm,
1443 .u.io.io_int_word = io_int_word,
1446 if (io_int_word & IO_INT_WORD_AI) {
1447 irq.type = KVM_S390_INT_IO(1, 0, 0, 0);
1449 irq.type = ((subchannel_id & 0xff00) << 24) |
1450 ((subchannel_id & 0x00060) << 22) | (subchannel_nr << 16);
1452 kvm_s390_floating_interrupt(&irq);
1455 void kvm_s390_crw_mchk(void)
1457 struct kvm_s390_irq irq = {
1458 .type = KVM_S390_MCHK,
1459 .u.mchk.cr14 = 1 << 28,
1460 .u.mchk.mcic = 0x00400f1d40330000ULL,
1462 kvm_s390_floating_interrupt(&irq);
1465 void kvm_s390_enable_css_support(S390CPU *cpu)
1469 /* Activate host kernel channel subsystem support. */
1470 r = kvm_vcpu_enable_cap(CPU(cpu), KVM_CAP_S390_CSS_SUPPORT, 0);
1474 void kvm_arch_init_irq_routing(KVMState *s)
1477 * Note that while irqchip capabilities generally imply that cpustates
1478 * are handled in-kernel, it is not true for s390 (yet); therefore, we
1479 * have to override the common code kvm_halt_in_kernel_allowed setting.
1481 if (kvm_check_extension(s, KVM_CAP_IRQ_ROUTING)) {
1482 kvm_gsi_routing_allowed = true;
1483 kvm_halt_in_kernel_allowed = false;
1487 int kvm_s390_assign_subch_ioeventfd(EventNotifier *notifier, uint32_t sch,
1488 int vq, bool assign)
1490 struct kvm_ioeventfd kick = {
1491 .flags = KVM_IOEVENTFD_FLAG_VIRTIO_CCW_NOTIFY |
1492 KVM_IOEVENTFD_FLAG_DATAMATCH,
1493 .fd = event_notifier_get_fd(notifier),
1498 if (!kvm_check_extension(kvm_state, KVM_CAP_IOEVENTFD)) {
1502 kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
1504 return kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
1507 int kvm_s390_get_memslot_count(KVMState *s)
1509 return kvm_check_extension(s, KVM_CAP_NR_MEMSLOTS);
1512 int kvm_s390_set_cpu_state(S390CPU *cpu, uint8_t cpu_state)
1514 struct kvm_mp_state mp_state = {};
1517 /* the kvm part might not have been initialized yet */
1518 if (CPU(cpu)->kvm_state == NULL) {
1522 switch (cpu_state) {
1523 case CPU_STATE_STOPPED:
1524 mp_state.mp_state = KVM_MP_STATE_STOPPED;
1526 case CPU_STATE_CHECK_STOP:
1527 mp_state.mp_state = KVM_MP_STATE_CHECK_STOP;
1529 case CPU_STATE_OPERATING:
1530 mp_state.mp_state = KVM_MP_STATE_OPERATING;
1532 case CPU_STATE_LOAD:
1533 mp_state.mp_state = KVM_MP_STATE_LOAD;
1536 error_report("Requested CPU state is not a valid S390 CPU state: %u",
1541 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
1543 trace_kvm_failed_cpu_state_set(CPU(cpu)->cpu_index, cpu_state,
1550 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
1551 uint64_t address, uint32_t data)
1553 S390PCIBusDevice *pbdev;
1554 uint32_t fid = data >> ZPCI_MSI_VEC_BITS;
1555 uint32_t vec = data & ZPCI_MSI_VEC_MASK;
1557 pbdev = s390_pci_find_dev_by_fid(fid);
1559 DPRINTF("add_msi_route no dev\n");
1563 pbdev->routes.adapter.ind_offset = vec;
1565 route->type = KVM_IRQ_ROUTING_S390_ADAPTER;
1567 route->u.adapter.summary_addr = pbdev->routes.adapter.summary_addr;
1568 route->u.adapter.ind_addr = pbdev->routes.adapter.ind_addr;
1569 route->u.adapter.summary_offset = pbdev->routes.adapter.summary_offset;
1570 route->u.adapter.ind_offset = pbdev->routes.adapter.ind_offset;
1571 route->u.adapter.adapter_id = pbdev->routes.adapter.adapter_id;