2 * hosting zSeries kernel virtual machines
4 * Copyright IBM Corp. 2008, 2009
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License (version 2 only)
8 * as published by the Free Software Foundation.
10 * Author(s): Carsten Otte <cotte@de.ibm.com>
11 * Christian Borntraeger <borntraeger@de.ibm.com>
12 * Heiko Carstens <heiko.carstens@de.ibm.com>
13 * Christian Ehrhardt <ehrhardt@de.ibm.com>
14 * Jason J. Herne <jjherne@us.ibm.com>
17 #include <linux/compiler.h>
18 #include <linux/err.h>
20 #include <linux/hrtimer.h>
21 #include <linux/init.h>
22 #include <linux/kvm.h>
23 #include <linux/kvm_host.h>
24 #include <linux/module.h>
25 #include <linux/slab.h>
26 #include <linux/timer.h>
27 #include <asm/asm-offsets.h>
28 #include <asm/lowcore.h>
29 #include <asm/pgtable.h>
31 #include <asm/switch_to.h>
32 #include <asm/facility.h>
37 #define CREATE_TRACE_POINTS
39 #include "trace-s390.h"
41 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
43 struct kvm_stats_debugfs_item debugfs_entries
[] = {
44 { "userspace_handled", VCPU_STAT(exit_userspace
) },
45 { "exit_null", VCPU_STAT(exit_null
) },
46 { "exit_validity", VCPU_STAT(exit_validity
) },
47 { "exit_stop_request", VCPU_STAT(exit_stop_request
) },
48 { "exit_external_request", VCPU_STAT(exit_external_request
) },
49 { "exit_external_interrupt", VCPU_STAT(exit_external_interrupt
) },
50 { "exit_instruction", VCPU_STAT(exit_instruction
) },
51 { "exit_program_interruption", VCPU_STAT(exit_program_interruption
) },
52 { "exit_instr_and_program_int", VCPU_STAT(exit_instr_and_program
) },
53 { "halt_wakeup", VCPU_STAT(halt_wakeup
) },
54 { "instruction_lctlg", VCPU_STAT(instruction_lctlg
) },
55 { "instruction_lctl", VCPU_STAT(instruction_lctl
) },
56 { "instruction_stctl", VCPU_STAT(instruction_stctl
) },
57 { "instruction_stctg", VCPU_STAT(instruction_stctg
) },
58 { "deliver_emergency_signal", VCPU_STAT(deliver_emergency_signal
) },
59 { "deliver_external_call", VCPU_STAT(deliver_external_call
) },
60 { "deliver_service_signal", VCPU_STAT(deliver_service_signal
) },
61 { "deliver_virtio_interrupt", VCPU_STAT(deliver_virtio_interrupt
) },
62 { "deliver_stop_signal", VCPU_STAT(deliver_stop_signal
) },
63 { "deliver_prefix_signal", VCPU_STAT(deliver_prefix_signal
) },
64 { "deliver_restart_signal", VCPU_STAT(deliver_restart_signal
) },
65 { "deliver_program_interruption", VCPU_STAT(deliver_program_int
) },
66 { "exit_wait_state", VCPU_STAT(exit_wait_state
) },
67 { "instruction_pfmf", VCPU_STAT(instruction_pfmf
) },
68 { "instruction_stidp", VCPU_STAT(instruction_stidp
) },
69 { "instruction_spx", VCPU_STAT(instruction_spx
) },
70 { "instruction_stpx", VCPU_STAT(instruction_stpx
) },
71 { "instruction_stap", VCPU_STAT(instruction_stap
) },
72 { "instruction_storage_key", VCPU_STAT(instruction_storage_key
) },
73 { "instruction_ipte_interlock", VCPU_STAT(instruction_ipte_interlock
) },
74 { "instruction_stsch", VCPU_STAT(instruction_stsch
) },
75 { "instruction_chsc", VCPU_STAT(instruction_chsc
) },
76 { "instruction_essa", VCPU_STAT(instruction_essa
) },
77 { "instruction_stsi", VCPU_STAT(instruction_stsi
) },
78 { "instruction_stfl", VCPU_STAT(instruction_stfl
) },
79 { "instruction_tprot", VCPU_STAT(instruction_tprot
) },
80 { "instruction_sigp_sense", VCPU_STAT(instruction_sigp_sense
) },
81 { "instruction_sigp_sense_running", VCPU_STAT(instruction_sigp_sense_running
) },
82 { "instruction_sigp_external_call", VCPU_STAT(instruction_sigp_external_call
) },
83 { "instruction_sigp_emergency", VCPU_STAT(instruction_sigp_emergency
) },
84 { "instruction_sigp_cond_emergency", VCPU_STAT(instruction_sigp_cond_emergency
) },
85 { "instruction_sigp_start", VCPU_STAT(instruction_sigp_start
) },
86 { "instruction_sigp_stop", VCPU_STAT(instruction_sigp_stop
) },
87 { "instruction_sigp_stop_store_status", VCPU_STAT(instruction_sigp_stop_store_status
) },
88 { "instruction_sigp_store_status", VCPU_STAT(instruction_sigp_store_status
) },
89 { "instruction_sigp_set_arch", VCPU_STAT(instruction_sigp_arch
) },
90 { "instruction_sigp_set_prefix", VCPU_STAT(instruction_sigp_prefix
) },
91 { "instruction_sigp_restart", VCPU_STAT(instruction_sigp_restart
) },
92 { "instruction_sigp_cpu_reset", VCPU_STAT(instruction_sigp_cpu_reset
) },
93 { "instruction_sigp_init_cpu_reset", VCPU_STAT(instruction_sigp_init_cpu_reset
) },
94 { "instruction_sigp_unknown", VCPU_STAT(instruction_sigp_unknown
) },
95 { "diagnose_10", VCPU_STAT(diagnose_10
) },
96 { "diagnose_44", VCPU_STAT(diagnose_44
) },
97 { "diagnose_9c", VCPU_STAT(diagnose_9c
) },
101 unsigned long *vfacilities
;
102 static struct gmap_notifier gmap_notifier
;
104 /* test availability of vfacility */
105 int test_vfacility(unsigned long nr
)
107 return __test_facility(nr
, (void *) vfacilities
);
110 /* Section: not file related */
111 int kvm_arch_hardware_enable(void)
113 /* every s390 is virtualization enabled ;-) */
117 static void kvm_gmap_notifier(struct gmap
*gmap
, unsigned long address
);
119 int kvm_arch_hardware_setup(void)
121 gmap_notifier
.notifier_call
= kvm_gmap_notifier
;
122 gmap_register_ipte_notifier(&gmap_notifier
);
126 void kvm_arch_hardware_unsetup(void)
128 gmap_unregister_ipte_notifier(&gmap_notifier
);
131 int kvm_arch_init(void *opaque
)
133 /* Register floating interrupt controller interface. */
134 return kvm_register_device_ops(&kvm_flic_ops
, KVM_DEV_TYPE_FLIC
);
137 /* Section: device related */
138 long kvm_arch_dev_ioctl(struct file
*filp
,
139 unsigned int ioctl
, unsigned long arg
)
141 if (ioctl
== KVM_S390_ENABLE_SIE
)
142 return s390_enable_sie();
146 int kvm_vm_ioctl_check_extension(struct kvm
*kvm
, long ext
)
151 case KVM_CAP_S390_PSW
:
152 case KVM_CAP_S390_GMAP
:
153 case KVM_CAP_SYNC_MMU
:
154 #ifdef CONFIG_KVM_S390_UCONTROL
155 case KVM_CAP_S390_UCONTROL
:
157 case KVM_CAP_ASYNC_PF
:
158 case KVM_CAP_SYNC_REGS
:
159 case KVM_CAP_ONE_REG
:
160 case KVM_CAP_ENABLE_CAP
:
161 case KVM_CAP_S390_CSS_SUPPORT
:
163 case KVM_CAP_IOEVENTFD
:
164 case KVM_CAP_DEVICE_CTRL
:
165 case KVM_CAP_ENABLE_CAP_VM
:
166 case KVM_CAP_S390_IRQCHIP
:
167 case KVM_CAP_VM_ATTRIBUTES
:
168 case KVM_CAP_MP_STATE
:
171 case KVM_CAP_NR_VCPUS
:
172 case KVM_CAP_MAX_VCPUS
:
175 case KVM_CAP_NR_MEMSLOTS
:
176 r
= KVM_USER_MEM_SLOTS
;
178 case KVM_CAP_S390_COW
:
179 r
= MACHINE_HAS_ESOP
;
187 static void kvm_s390_sync_dirty_log(struct kvm
*kvm
,
188 struct kvm_memory_slot
*memslot
)
190 gfn_t cur_gfn
, last_gfn
;
191 unsigned long address
;
192 struct gmap
*gmap
= kvm
->arch
.gmap
;
194 down_read(&gmap
->mm
->mmap_sem
);
195 /* Loop over all guest pages */
196 last_gfn
= memslot
->base_gfn
+ memslot
->npages
;
197 for (cur_gfn
= memslot
->base_gfn
; cur_gfn
<= last_gfn
; cur_gfn
++) {
198 address
= gfn_to_hva_memslot(memslot
, cur_gfn
);
200 if (gmap_test_and_clear_dirty(address
, gmap
))
201 mark_page_dirty(kvm
, cur_gfn
);
203 up_read(&gmap
->mm
->mmap_sem
);
206 /* Section: vm related */
208 * Get (and clear) the dirty memory log for a memory slot.
210 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
,
211 struct kvm_dirty_log
*log
)
215 struct kvm_memory_slot
*memslot
;
218 mutex_lock(&kvm
->slots_lock
);
221 if (log
->slot
>= KVM_USER_MEM_SLOTS
)
224 memslot
= id_to_memslot(kvm
->memslots
, log
->slot
);
226 if (!memslot
->dirty_bitmap
)
229 kvm_s390_sync_dirty_log(kvm
, memslot
);
230 r
= kvm_get_dirty_log(kvm
, log
, &is_dirty
);
234 /* Clear the dirty log */
236 n
= kvm_dirty_bitmap_bytes(memslot
);
237 memset(memslot
->dirty_bitmap
, 0, n
);
241 mutex_unlock(&kvm
->slots_lock
);
245 static int kvm_vm_ioctl_enable_cap(struct kvm
*kvm
, struct kvm_enable_cap
*cap
)
253 case KVM_CAP_S390_IRQCHIP
:
254 kvm
->arch
.use_irqchip
= 1;
264 static int kvm_s390_mem_control(struct kvm
*kvm
, struct kvm_device_attr
*attr
)
268 switch (attr
->attr
) {
269 case KVM_S390_VM_MEM_ENABLE_CMMA
:
271 mutex_lock(&kvm
->lock
);
272 if (atomic_read(&kvm
->online_vcpus
) == 0) {
273 kvm
->arch
.use_cmma
= 1;
276 mutex_unlock(&kvm
->lock
);
278 case KVM_S390_VM_MEM_CLR_CMMA
:
279 mutex_lock(&kvm
->lock
);
280 idx
= srcu_read_lock(&kvm
->srcu
);
281 s390_reset_cmma(kvm
->arch
.gmap
->mm
);
282 srcu_read_unlock(&kvm
->srcu
, idx
);
283 mutex_unlock(&kvm
->lock
);
293 static int kvm_s390_vm_set_attr(struct kvm
*kvm
, struct kvm_device_attr
*attr
)
297 switch (attr
->group
) {
298 case KVM_S390_VM_MEM_CTRL
:
299 ret
= kvm_s390_mem_control(kvm
, attr
);
309 static int kvm_s390_vm_get_attr(struct kvm
*kvm
, struct kvm_device_attr
*attr
)
314 static int kvm_s390_vm_has_attr(struct kvm
*kvm
, struct kvm_device_attr
*attr
)
318 switch (attr
->group
) {
319 case KVM_S390_VM_MEM_CTRL
:
320 switch (attr
->attr
) {
321 case KVM_S390_VM_MEM_ENABLE_CMMA
:
322 case KVM_S390_VM_MEM_CLR_CMMA
:
338 long kvm_arch_vm_ioctl(struct file
*filp
,
339 unsigned int ioctl
, unsigned long arg
)
341 struct kvm
*kvm
= filp
->private_data
;
342 void __user
*argp
= (void __user
*)arg
;
343 struct kvm_device_attr attr
;
347 case KVM_S390_INTERRUPT
: {
348 struct kvm_s390_interrupt s390int
;
351 if (copy_from_user(&s390int
, argp
, sizeof(s390int
)))
353 r
= kvm_s390_inject_vm(kvm
, &s390int
);
356 case KVM_ENABLE_CAP
: {
357 struct kvm_enable_cap cap
;
359 if (copy_from_user(&cap
, argp
, sizeof(cap
)))
361 r
= kvm_vm_ioctl_enable_cap(kvm
, &cap
);
364 case KVM_CREATE_IRQCHIP
: {
365 struct kvm_irq_routing_entry routing
;
368 if (kvm
->arch
.use_irqchip
) {
369 /* Set up dummy routing. */
370 memset(&routing
, 0, sizeof(routing
));
371 kvm_set_irq_routing(kvm
, &routing
, 0, 0);
376 case KVM_SET_DEVICE_ATTR
: {
378 if (copy_from_user(&attr
, (void __user
*)arg
, sizeof(attr
)))
380 r
= kvm_s390_vm_set_attr(kvm
, &attr
);
383 case KVM_GET_DEVICE_ATTR
: {
385 if (copy_from_user(&attr
, (void __user
*)arg
, sizeof(attr
)))
387 r
= kvm_s390_vm_get_attr(kvm
, &attr
);
390 case KVM_HAS_DEVICE_ATTR
: {
392 if (copy_from_user(&attr
, (void __user
*)arg
, sizeof(attr
)))
394 r
= kvm_s390_vm_has_attr(kvm
, &attr
);
404 static int kvm_s390_crypto_init(struct kvm
*kvm
)
406 if (!test_vfacility(76))
409 kvm
->arch
.crypto
.crycb
= kzalloc(sizeof(*kvm
->arch
.crypto
.crycb
),
410 GFP_KERNEL
| GFP_DMA
);
411 if (!kvm
->arch
.crypto
.crycb
)
414 kvm
->arch
.crypto
.crycbd
= (__u32
) (unsigned long) kvm
->arch
.crypto
.crycb
|
420 int kvm_arch_init_vm(struct kvm
*kvm
, unsigned long type
)
424 static unsigned long sca_offset
;
427 #ifdef CONFIG_KVM_S390_UCONTROL
428 if (type
& ~KVM_VM_S390_UCONTROL
)
430 if ((type
& KVM_VM_S390_UCONTROL
) && (!capable(CAP_SYS_ADMIN
)))
437 rc
= s390_enable_sie();
443 kvm
->arch
.sca
= (struct sca_block
*) get_zeroed_page(GFP_KERNEL
);
446 spin_lock(&kvm_lock
);
447 sca_offset
= (sca_offset
+ 16) & 0x7f0;
448 kvm
->arch
.sca
= (struct sca_block
*) ((char *) kvm
->arch
.sca
+ sca_offset
);
449 spin_unlock(&kvm_lock
);
451 sprintf(debug_name
, "kvm-%u", current
->pid
);
453 kvm
->arch
.dbf
= debug_register(debug_name
, 8, 2, 8 * sizeof(long));
457 if (kvm_s390_crypto_init(kvm
) < 0)
460 spin_lock_init(&kvm
->arch
.float_int
.lock
);
461 INIT_LIST_HEAD(&kvm
->arch
.float_int
.list
);
462 init_waitqueue_head(&kvm
->arch
.ipte_wq
);
463 mutex_init(&kvm
->arch
.ipte_mutex
);
465 debug_register_view(kvm
->arch
.dbf
, &debug_sprintf_view
);
466 VM_EVENT(kvm
, 3, "%s", "vm created");
468 if (type
& KVM_VM_S390_UCONTROL
) {
469 kvm
->arch
.gmap
= NULL
;
471 kvm
->arch
.gmap
= gmap_alloc(current
->mm
, (1UL << 44) - 1);
474 kvm
->arch
.gmap
->private = kvm
;
475 kvm
->arch
.gmap
->pfault_enabled
= 0;
478 kvm
->arch
.css_support
= 0;
479 kvm
->arch
.use_irqchip
= 0;
481 spin_lock_init(&kvm
->arch
.start_stop_lock
);
485 kfree(kvm
->arch
.crypto
.crycb
);
487 debug_unregister(kvm
->arch
.dbf
);
489 free_page((unsigned long)(kvm
->arch
.sca
));
494 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
496 VCPU_EVENT(vcpu
, 3, "%s", "free cpu");
497 trace_kvm_s390_destroy_vcpu(vcpu
->vcpu_id
);
498 kvm_s390_clear_local_irqs(vcpu
);
499 kvm_clear_async_pf_completion_queue(vcpu
);
500 if (!kvm_is_ucontrol(vcpu
->kvm
)) {
501 clear_bit(63 - vcpu
->vcpu_id
,
502 (unsigned long *) &vcpu
->kvm
->arch
.sca
->mcn
);
503 if (vcpu
->kvm
->arch
.sca
->cpu
[vcpu
->vcpu_id
].sda
==
504 (__u64
) vcpu
->arch
.sie_block
)
505 vcpu
->kvm
->arch
.sca
->cpu
[vcpu
->vcpu_id
].sda
= 0;
509 if (kvm_is_ucontrol(vcpu
->kvm
))
510 gmap_free(vcpu
->arch
.gmap
);
512 if (kvm_s390_cmma_enabled(vcpu
->kvm
))
513 kvm_s390_vcpu_unsetup_cmma(vcpu
);
514 free_page((unsigned long)(vcpu
->arch
.sie_block
));
516 kvm_vcpu_uninit(vcpu
);
517 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
520 static void kvm_free_vcpus(struct kvm
*kvm
)
523 struct kvm_vcpu
*vcpu
;
525 kvm_for_each_vcpu(i
, vcpu
, kvm
)
526 kvm_arch_vcpu_destroy(vcpu
);
528 mutex_lock(&kvm
->lock
);
529 for (i
= 0; i
< atomic_read(&kvm
->online_vcpus
); i
++)
530 kvm
->vcpus
[i
] = NULL
;
532 atomic_set(&kvm
->online_vcpus
, 0);
533 mutex_unlock(&kvm
->lock
);
536 void kvm_arch_destroy_vm(struct kvm
*kvm
)
539 free_page((unsigned long)(kvm
->arch
.sca
));
540 debug_unregister(kvm
->arch
.dbf
);
541 kfree(kvm
->arch
.crypto
.crycb
);
542 if (!kvm_is_ucontrol(kvm
))
543 gmap_free(kvm
->arch
.gmap
);
544 kvm_s390_destroy_adapters(kvm
);
545 kvm_s390_clear_float_irqs(kvm
);
548 /* Section: vcpu related */
549 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
551 vcpu
->arch
.pfault_token
= KVM_S390_PFAULT_TOKEN_INVALID
;
552 kvm_clear_async_pf_completion_queue(vcpu
);
553 if (kvm_is_ucontrol(vcpu
->kvm
)) {
554 vcpu
->arch
.gmap
= gmap_alloc(current
->mm
, -1UL);
555 if (!vcpu
->arch
.gmap
)
557 vcpu
->arch
.gmap
->private = vcpu
->kvm
;
561 vcpu
->arch
.gmap
= vcpu
->kvm
->arch
.gmap
;
562 vcpu
->run
->kvm_valid_regs
= KVM_SYNC_PREFIX
|
571 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
573 save_fp_ctl(&vcpu
->arch
.host_fpregs
.fpc
);
574 save_fp_regs(vcpu
->arch
.host_fpregs
.fprs
);
575 save_access_regs(vcpu
->arch
.host_acrs
);
576 restore_fp_ctl(&vcpu
->arch
.guest_fpregs
.fpc
);
577 restore_fp_regs(vcpu
->arch
.guest_fpregs
.fprs
);
578 restore_access_regs(vcpu
->run
->s
.regs
.acrs
);
579 gmap_enable(vcpu
->arch
.gmap
);
580 atomic_set_mask(CPUSTAT_RUNNING
, &vcpu
->arch
.sie_block
->cpuflags
);
583 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
585 atomic_clear_mask(CPUSTAT_RUNNING
, &vcpu
->arch
.sie_block
->cpuflags
);
586 gmap_disable(vcpu
->arch
.gmap
);
587 save_fp_ctl(&vcpu
->arch
.guest_fpregs
.fpc
);
588 save_fp_regs(vcpu
->arch
.guest_fpregs
.fprs
);
589 save_access_regs(vcpu
->run
->s
.regs
.acrs
);
590 restore_fp_ctl(&vcpu
->arch
.host_fpregs
.fpc
);
591 restore_fp_regs(vcpu
->arch
.host_fpregs
.fprs
);
592 restore_access_regs(vcpu
->arch
.host_acrs
);
595 static void kvm_s390_vcpu_initial_reset(struct kvm_vcpu
*vcpu
)
597 /* this equals initial cpu reset in pop, but we don't switch to ESA */
598 vcpu
->arch
.sie_block
->gpsw
.mask
= 0UL;
599 vcpu
->arch
.sie_block
->gpsw
.addr
= 0UL;
600 kvm_s390_set_prefix(vcpu
, 0);
601 vcpu
->arch
.sie_block
->cputm
= 0UL;
602 vcpu
->arch
.sie_block
->ckc
= 0UL;
603 vcpu
->arch
.sie_block
->todpr
= 0;
604 memset(vcpu
->arch
.sie_block
->gcr
, 0, 16 * sizeof(__u64
));
605 vcpu
->arch
.sie_block
->gcr
[0] = 0xE0UL
;
606 vcpu
->arch
.sie_block
->gcr
[14] = 0xC2000000UL
;
607 vcpu
->arch
.guest_fpregs
.fpc
= 0;
608 asm volatile("lfpc %0" : : "Q" (vcpu
->arch
.guest_fpregs
.fpc
));
609 vcpu
->arch
.sie_block
->gbea
= 1;
610 vcpu
->arch
.sie_block
->pp
= 0;
611 vcpu
->arch
.pfault_token
= KVM_S390_PFAULT_TOKEN_INVALID
;
612 kvm_clear_async_pf_completion_queue(vcpu
);
613 if (!kvm_s390_user_cpu_state_ctrl(vcpu
->kvm
))
614 kvm_s390_vcpu_stop(vcpu
);
615 kvm_s390_clear_local_irqs(vcpu
);
618 int kvm_arch_vcpu_postcreate(struct kvm_vcpu
*vcpu
)
623 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu
*vcpu
)
625 if (!test_vfacility(76))
628 vcpu
->arch
.sie_block
->crycbd
= vcpu
->kvm
->arch
.crypto
.crycbd
;
631 void kvm_s390_vcpu_unsetup_cmma(struct kvm_vcpu
*vcpu
)
633 free_page(vcpu
->arch
.sie_block
->cbrlo
);
634 vcpu
->arch
.sie_block
->cbrlo
= 0;
637 int kvm_s390_vcpu_setup_cmma(struct kvm_vcpu
*vcpu
)
639 vcpu
->arch
.sie_block
->cbrlo
= get_zeroed_page(GFP_KERNEL
);
640 if (!vcpu
->arch
.sie_block
->cbrlo
)
643 vcpu
->arch
.sie_block
->ecb2
|= 0x80;
644 vcpu
->arch
.sie_block
->ecb2
&= ~0x08;
648 int kvm_arch_vcpu_setup(struct kvm_vcpu
*vcpu
)
652 atomic_set(&vcpu
->arch
.sie_block
->cpuflags
, CPUSTAT_ZARCH
|
656 vcpu
->arch
.sie_block
->ecb
= 6;
657 if (test_vfacility(50) && test_vfacility(73))
658 vcpu
->arch
.sie_block
->ecb
|= 0x10;
660 vcpu
->arch
.sie_block
->ecb2
= 8;
661 vcpu
->arch
.sie_block
->eca
= 0xD1002000U
;
663 vcpu
->arch
.sie_block
->eca
|= 1;
664 vcpu
->arch
.sie_block
->fac
= (int) (long) vfacilities
;
665 vcpu
->arch
.sie_block
->ictl
|= ICTL_ISKE
| ICTL_SSKE
| ICTL_RRBE
|
668 if (kvm_s390_cmma_enabled(vcpu
->kvm
)) {
669 rc
= kvm_s390_vcpu_setup_cmma(vcpu
);
673 hrtimer_init(&vcpu
->arch
.ckc_timer
, CLOCK_REALTIME
, HRTIMER_MODE_ABS
);
674 vcpu
->arch
.ckc_timer
.function
= kvm_s390_idle_wakeup
;
675 get_cpu_id(&vcpu
->arch
.cpu_id
);
676 vcpu
->arch
.cpu_id
.version
= 0xff;
678 kvm_s390_vcpu_crypto_setup(vcpu
);
683 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
,
686 struct kvm_vcpu
*vcpu
;
687 struct sie_page
*sie_page
;
690 if (id
>= KVM_MAX_VCPUS
)
695 vcpu
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL
);
699 sie_page
= (struct sie_page
*) get_zeroed_page(GFP_KERNEL
);
703 vcpu
->arch
.sie_block
= &sie_page
->sie_block
;
704 vcpu
->arch
.sie_block
->itdba
= (unsigned long) &sie_page
->itdb
;
706 vcpu
->arch
.sie_block
->icpua
= id
;
707 if (!kvm_is_ucontrol(kvm
)) {
708 if (!kvm
->arch
.sca
) {
712 if (!kvm
->arch
.sca
->cpu
[id
].sda
)
713 kvm
->arch
.sca
->cpu
[id
].sda
=
714 (__u64
) vcpu
->arch
.sie_block
;
715 vcpu
->arch
.sie_block
->scaoh
=
716 (__u32
)(((__u64
)kvm
->arch
.sca
) >> 32);
717 vcpu
->arch
.sie_block
->scaol
= (__u32
)(__u64
)kvm
->arch
.sca
;
718 set_bit(63 - id
, (unsigned long *) &kvm
->arch
.sca
->mcn
);
721 spin_lock_init(&vcpu
->arch
.local_int
.lock
);
722 vcpu
->arch
.local_int
.float_int
= &kvm
->arch
.float_int
;
723 vcpu
->arch
.local_int
.wq
= &vcpu
->wq
;
724 vcpu
->arch
.local_int
.cpuflags
= &vcpu
->arch
.sie_block
->cpuflags
;
726 rc
= kvm_vcpu_init(vcpu
, kvm
, id
);
728 goto out_free_sie_block
;
729 VM_EVENT(kvm
, 3, "create cpu %d at %p, sie block at %p", id
, vcpu
,
730 vcpu
->arch
.sie_block
);
731 trace_kvm_s390_create_vcpu(id
, vcpu
, vcpu
->arch
.sie_block
);
735 free_page((unsigned long)(vcpu
->arch
.sie_block
));
737 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
742 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*vcpu
)
744 return kvm_cpu_has_interrupt(vcpu
);
747 void s390_vcpu_block(struct kvm_vcpu
*vcpu
)
749 atomic_set_mask(PROG_BLOCK_SIE
, &vcpu
->arch
.sie_block
->prog20
);
752 void s390_vcpu_unblock(struct kvm_vcpu
*vcpu
)
754 atomic_clear_mask(PROG_BLOCK_SIE
, &vcpu
->arch
.sie_block
->prog20
);
758 * Kick a guest cpu out of SIE and wait until SIE is not running.
759 * If the CPU is not running (e.g. waiting as idle) the function will
760 * return immediately. */
761 void exit_sie(struct kvm_vcpu
*vcpu
)
763 atomic_set_mask(CPUSTAT_STOP_INT
, &vcpu
->arch
.sie_block
->cpuflags
);
764 while (vcpu
->arch
.sie_block
->prog0c
& PROG_IN_SIE
)
768 /* Kick a guest cpu out of SIE and prevent SIE-reentry */
769 void exit_sie_sync(struct kvm_vcpu
*vcpu
)
771 s390_vcpu_block(vcpu
);
775 static void kvm_gmap_notifier(struct gmap
*gmap
, unsigned long address
)
778 struct kvm
*kvm
= gmap
->private;
779 struct kvm_vcpu
*vcpu
;
781 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
782 /* match against both prefix pages */
783 if (kvm_s390_get_prefix(vcpu
) == (address
& ~0x1000UL
)) {
784 VCPU_EVENT(vcpu
, 2, "gmap notifier for %lx", address
);
785 kvm_make_request(KVM_REQ_MMU_RELOAD
, vcpu
);
791 int kvm_arch_vcpu_should_kick(struct kvm_vcpu
*vcpu
)
793 /* kvm common code refers to this, but never calls it */
798 static int kvm_arch_vcpu_ioctl_get_one_reg(struct kvm_vcpu
*vcpu
,
799 struct kvm_one_reg
*reg
)
804 case KVM_REG_S390_TODPR
:
805 r
= put_user(vcpu
->arch
.sie_block
->todpr
,
806 (u32 __user
*)reg
->addr
);
808 case KVM_REG_S390_EPOCHDIFF
:
809 r
= put_user(vcpu
->arch
.sie_block
->epoch
,
810 (u64 __user
*)reg
->addr
);
812 case KVM_REG_S390_CPU_TIMER
:
813 r
= put_user(vcpu
->arch
.sie_block
->cputm
,
814 (u64 __user
*)reg
->addr
);
816 case KVM_REG_S390_CLOCK_COMP
:
817 r
= put_user(vcpu
->arch
.sie_block
->ckc
,
818 (u64 __user
*)reg
->addr
);
820 case KVM_REG_S390_PFTOKEN
:
821 r
= put_user(vcpu
->arch
.pfault_token
,
822 (u64 __user
*)reg
->addr
);
824 case KVM_REG_S390_PFCOMPARE
:
825 r
= put_user(vcpu
->arch
.pfault_compare
,
826 (u64 __user
*)reg
->addr
);
828 case KVM_REG_S390_PFSELECT
:
829 r
= put_user(vcpu
->arch
.pfault_select
,
830 (u64 __user
*)reg
->addr
);
832 case KVM_REG_S390_PP
:
833 r
= put_user(vcpu
->arch
.sie_block
->pp
,
834 (u64 __user
*)reg
->addr
);
836 case KVM_REG_S390_GBEA
:
837 r
= put_user(vcpu
->arch
.sie_block
->gbea
,
838 (u64 __user
*)reg
->addr
);
847 static int kvm_arch_vcpu_ioctl_set_one_reg(struct kvm_vcpu
*vcpu
,
848 struct kvm_one_reg
*reg
)
853 case KVM_REG_S390_TODPR
:
854 r
= get_user(vcpu
->arch
.sie_block
->todpr
,
855 (u32 __user
*)reg
->addr
);
857 case KVM_REG_S390_EPOCHDIFF
:
858 r
= get_user(vcpu
->arch
.sie_block
->epoch
,
859 (u64 __user
*)reg
->addr
);
861 case KVM_REG_S390_CPU_TIMER
:
862 r
= get_user(vcpu
->arch
.sie_block
->cputm
,
863 (u64 __user
*)reg
->addr
);
865 case KVM_REG_S390_CLOCK_COMP
:
866 r
= get_user(vcpu
->arch
.sie_block
->ckc
,
867 (u64 __user
*)reg
->addr
);
869 case KVM_REG_S390_PFTOKEN
:
870 r
= get_user(vcpu
->arch
.pfault_token
,
871 (u64 __user
*)reg
->addr
);
873 case KVM_REG_S390_PFCOMPARE
:
874 r
= get_user(vcpu
->arch
.pfault_compare
,
875 (u64 __user
*)reg
->addr
);
877 case KVM_REG_S390_PFSELECT
:
878 r
= get_user(vcpu
->arch
.pfault_select
,
879 (u64 __user
*)reg
->addr
);
881 case KVM_REG_S390_PP
:
882 r
= get_user(vcpu
->arch
.sie_block
->pp
,
883 (u64 __user
*)reg
->addr
);
885 case KVM_REG_S390_GBEA
:
886 r
= get_user(vcpu
->arch
.sie_block
->gbea
,
887 (u64 __user
*)reg
->addr
);
896 static int kvm_arch_vcpu_ioctl_initial_reset(struct kvm_vcpu
*vcpu
)
898 kvm_s390_vcpu_initial_reset(vcpu
);
902 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
904 memcpy(&vcpu
->run
->s
.regs
.gprs
, ®s
->gprs
, sizeof(regs
->gprs
));
908 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
910 memcpy(®s
->gprs
, &vcpu
->run
->s
.regs
.gprs
, sizeof(regs
->gprs
));
914 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu
*vcpu
,
915 struct kvm_sregs
*sregs
)
917 memcpy(&vcpu
->run
->s
.regs
.acrs
, &sregs
->acrs
, sizeof(sregs
->acrs
));
918 memcpy(&vcpu
->arch
.sie_block
->gcr
, &sregs
->crs
, sizeof(sregs
->crs
));
919 restore_access_regs(vcpu
->run
->s
.regs
.acrs
);
923 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu
*vcpu
,
924 struct kvm_sregs
*sregs
)
926 memcpy(&sregs
->acrs
, &vcpu
->run
->s
.regs
.acrs
, sizeof(sregs
->acrs
));
927 memcpy(&sregs
->crs
, &vcpu
->arch
.sie_block
->gcr
, sizeof(sregs
->crs
));
931 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
933 if (test_fp_ctl(fpu
->fpc
))
935 memcpy(&vcpu
->arch
.guest_fpregs
.fprs
, &fpu
->fprs
, sizeof(fpu
->fprs
));
936 vcpu
->arch
.guest_fpregs
.fpc
= fpu
->fpc
;
937 restore_fp_ctl(&vcpu
->arch
.guest_fpregs
.fpc
);
938 restore_fp_regs(vcpu
->arch
.guest_fpregs
.fprs
);
942 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
944 memcpy(&fpu
->fprs
, &vcpu
->arch
.guest_fpregs
.fprs
, sizeof(fpu
->fprs
));
945 fpu
->fpc
= vcpu
->arch
.guest_fpregs
.fpc
;
949 static int kvm_arch_vcpu_ioctl_set_initial_psw(struct kvm_vcpu
*vcpu
, psw_t psw
)
953 if (!is_vcpu_stopped(vcpu
))
956 vcpu
->run
->psw_mask
= psw
.mask
;
957 vcpu
->run
->psw_addr
= psw
.addr
;
962 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu
*vcpu
,
963 struct kvm_translation
*tr
)
965 return -EINVAL
; /* not implemented yet */
968 #define VALID_GUESTDBG_FLAGS (KVM_GUESTDBG_SINGLESTEP | \
969 KVM_GUESTDBG_USE_HW_BP | \
972 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu
*vcpu
,
973 struct kvm_guest_debug
*dbg
)
977 vcpu
->guest_debug
= 0;
978 kvm_s390_clear_bp_data(vcpu
);
980 if (dbg
->control
& ~VALID_GUESTDBG_FLAGS
)
983 if (dbg
->control
& KVM_GUESTDBG_ENABLE
) {
984 vcpu
->guest_debug
= dbg
->control
;
985 /* enforce guest PER */
986 atomic_set_mask(CPUSTAT_P
, &vcpu
->arch
.sie_block
->cpuflags
);
988 if (dbg
->control
& KVM_GUESTDBG_USE_HW_BP
)
989 rc
= kvm_s390_import_bp_data(vcpu
, dbg
);
991 atomic_clear_mask(CPUSTAT_P
, &vcpu
->arch
.sie_block
->cpuflags
);
992 vcpu
->arch
.guestdbg
.last_bp
= 0;
996 vcpu
->guest_debug
= 0;
997 kvm_s390_clear_bp_data(vcpu
);
998 atomic_clear_mask(CPUSTAT_P
, &vcpu
->arch
.sie_block
->cpuflags
);
1004 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
1005 struct kvm_mp_state
*mp_state
)
1007 /* CHECK_STOP and LOAD are not supported yet */
1008 return is_vcpu_stopped(vcpu
) ? KVM_MP_STATE_STOPPED
:
1009 KVM_MP_STATE_OPERATING
;
1012 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
1013 struct kvm_mp_state
*mp_state
)
1017 /* user space knows about this interface - let it control the state */
1018 vcpu
->kvm
->arch
.user_cpu_state_ctrl
= 1;
1020 switch (mp_state
->mp_state
) {
1021 case KVM_MP_STATE_STOPPED
:
1022 kvm_s390_vcpu_stop(vcpu
);
1024 case KVM_MP_STATE_OPERATING
:
1025 kvm_s390_vcpu_start(vcpu
);
1027 case KVM_MP_STATE_LOAD
:
1028 case KVM_MP_STATE_CHECK_STOP
:
1029 /* fall through - CHECK_STOP and LOAD are not supported yet */
1037 bool kvm_s390_cmma_enabled(struct kvm
*kvm
)
1039 if (!MACHINE_IS_LPAR
)
1041 /* only enable for z10 and later */
1042 if (!MACHINE_HAS_EDAT1
)
1044 if (!kvm
->arch
.use_cmma
)
1049 static bool ibs_enabled(struct kvm_vcpu
*vcpu
)
1051 return atomic_read(&vcpu
->arch
.sie_block
->cpuflags
) & CPUSTAT_IBS
;
1054 static int kvm_s390_handle_requests(struct kvm_vcpu
*vcpu
)
1057 s390_vcpu_unblock(vcpu
);
1059 * We use MMU_RELOAD just to re-arm the ipte notifier for the
1060 * guest prefix page. gmap_ipte_notify will wait on the ptl lock.
1061 * This ensures that the ipte instruction for this request has
1062 * already finished. We might race against a second unmapper that
1063 * wants to set the blocking bit. Lets just retry the request loop.
1065 if (kvm_check_request(KVM_REQ_MMU_RELOAD
, vcpu
)) {
1067 rc
= gmap_ipte_notify(vcpu
->arch
.gmap
,
1068 kvm_s390_get_prefix(vcpu
),
1075 if (kvm_check_request(KVM_REQ_TLB_FLUSH
, vcpu
)) {
1076 vcpu
->arch
.sie_block
->ihcpu
= 0xffff;
1080 if (kvm_check_request(KVM_REQ_ENABLE_IBS
, vcpu
)) {
1081 if (!ibs_enabled(vcpu
)) {
1082 trace_kvm_s390_enable_disable_ibs(vcpu
->vcpu_id
, 1);
1083 atomic_set_mask(CPUSTAT_IBS
,
1084 &vcpu
->arch
.sie_block
->cpuflags
);
1089 if (kvm_check_request(KVM_REQ_DISABLE_IBS
, vcpu
)) {
1090 if (ibs_enabled(vcpu
)) {
1091 trace_kvm_s390_enable_disable_ibs(vcpu
->vcpu_id
, 0);
1092 atomic_clear_mask(CPUSTAT_IBS
,
1093 &vcpu
->arch
.sie_block
->cpuflags
);
1098 /* nothing to do, just clear the request */
1099 clear_bit(KVM_REQ_UNHALT
, &vcpu
->requests
);
1105 * kvm_arch_fault_in_page - fault-in guest page if necessary
1106 * @vcpu: The corresponding virtual cpu
1107 * @gpa: Guest physical address
1108 * @writable: Whether the page should be writable or not
1110 * Make sure that a guest page has been faulted-in on the host.
1112 * Return: Zero on success, negative error code otherwise.
1114 long kvm_arch_fault_in_page(struct kvm_vcpu
*vcpu
, gpa_t gpa
, int writable
)
1116 return gmap_fault(vcpu
->arch
.gmap
, gpa
,
1117 writable
? FAULT_FLAG_WRITE
: 0);
1120 static void __kvm_inject_pfault_token(struct kvm_vcpu
*vcpu
, bool start_token
,
1121 unsigned long token
)
1123 struct kvm_s390_interrupt inti
;
1124 struct kvm_s390_irq irq
;
1127 irq
.u
.ext
.ext_params2
= token
;
1128 irq
.type
= KVM_S390_INT_PFAULT_INIT
;
1129 WARN_ON_ONCE(kvm_s390_inject_vcpu(vcpu
, &irq
));
1131 inti
.type
= KVM_S390_INT_PFAULT_DONE
;
1132 inti
.parm64
= token
;
1133 WARN_ON_ONCE(kvm_s390_inject_vm(vcpu
->kvm
, &inti
));
1137 void kvm_arch_async_page_not_present(struct kvm_vcpu
*vcpu
,
1138 struct kvm_async_pf
*work
)
1140 trace_kvm_s390_pfault_init(vcpu
, work
->arch
.pfault_token
);
1141 __kvm_inject_pfault_token(vcpu
, true, work
->arch
.pfault_token
);
1144 void kvm_arch_async_page_present(struct kvm_vcpu
*vcpu
,
1145 struct kvm_async_pf
*work
)
1147 trace_kvm_s390_pfault_done(vcpu
, work
->arch
.pfault_token
);
1148 __kvm_inject_pfault_token(vcpu
, false, work
->arch
.pfault_token
);
1151 void kvm_arch_async_page_ready(struct kvm_vcpu
*vcpu
,
1152 struct kvm_async_pf
*work
)
1154 /* s390 will always inject the page directly */
1157 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu
*vcpu
)
1160 * s390 will always inject the page directly,
1161 * but we still want check_async_completion to cleanup
1166 static int kvm_arch_setup_async_pf(struct kvm_vcpu
*vcpu
)
1169 struct kvm_arch_async_pf arch
;
1172 if (vcpu
->arch
.pfault_token
== KVM_S390_PFAULT_TOKEN_INVALID
)
1174 if ((vcpu
->arch
.sie_block
->gpsw
.mask
& vcpu
->arch
.pfault_select
) !=
1175 vcpu
->arch
.pfault_compare
)
1177 if (psw_extint_disabled(vcpu
))
1179 if (kvm_cpu_has_interrupt(vcpu
))
1181 if (!(vcpu
->arch
.sie_block
->gcr
[0] & 0x200ul
))
1183 if (!vcpu
->arch
.gmap
->pfault_enabled
)
1186 hva
= gfn_to_hva(vcpu
->kvm
, gpa_to_gfn(current
->thread
.gmap_addr
));
1187 hva
+= current
->thread
.gmap_addr
& ~PAGE_MASK
;
1188 if (read_guest_real(vcpu
, vcpu
->arch
.pfault_token
, &arch
.pfault_token
, 8))
1191 rc
= kvm_setup_async_pf(vcpu
, current
->thread
.gmap_addr
, hva
, &arch
);
1195 static int vcpu_pre_run(struct kvm_vcpu
*vcpu
)
1200 * On s390 notifications for arriving pages will be delivered directly
1201 * to the guest but the house keeping for completed pfaults is
1202 * handled outside the worker.
1204 kvm_check_async_pf_completion(vcpu
);
1206 memcpy(&vcpu
->arch
.sie_block
->gg14
, &vcpu
->run
->s
.regs
.gprs
[14], 16);
1211 if (test_cpu_flag(CIF_MCCK_PENDING
))
1214 if (!kvm_is_ucontrol(vcpu
->kvm
)) {
1215 rc
= kvm_s390_deliver_pending_interrupts(vcpu
);
1220 rc
= kvm_s390_handle_requests(vcpu
);
1224 if (guestdbg_enabled(vcpu
)) {
1225 kvm_s390_backup_guest_per_regs(vcpu
);
1226 kvm_s390_patch_guest_per_regs(vcpu
);
1229 vcpu
->arch
.sie_block
->icptcode
= 0;
1230 cpuflags
= atomic_read(&vcpu
->arch
.sie_block
->cpuflags
);
1231 VCPU_EVENT(vcpu
, 6, "entering sie flags %x", cpuflags
);
1232 trace_kvm_s390_sie_enter(vcpu
, cpuflags
);
1237 static int vcpu_post_run(struct kvm_vcpu
*vcpu
, int exit_reason
)
1241 VCPU_EVENT(vcpu
, 6, "exit sie icptcode %d",
1242 vcpu
->arch
.sie_block
->icptcode
);
1243 trace_kvm_s390_sie_exit(vcpu
, vcpu
->arch
.sie_block
->icptcode
);
1245 if (guestdbg_enabled(vcpu
))
1246 kvm_s390_restore_guest_per_regs(vcpu
);
1248 if (exit_reason
>= 0) {
1250 } else if (kvm_is_ucontrol(vcpu
->kvm
)) {
1251 vcpu
->run
->exit_reason
= KVM_EXIT_S390_UCONTROL
;
1252 vcpu
->run
->s390_ucontrol
.trans_exc_code
=
1253 current
->thread
.gmap_addr
;
1254 vcpu
->run
->s390_ucontrol
.pgm_code
= 0x10;
1257 } else if (current
->thread
.gmap_pfault
) {
1258 trace_kvm_s390_major_guest_pfault(vcpu
);
1259 current
->thread
.gmap_pfault
= 0;
1260 if (kvm_arch_setup_async_pf(vcpu
)) {
1263 gpa_t gpa
= current
->thread
.gmap_addr
;
1264 rc
= kvm_arch_fault_in_page(vcpu
, gpa
, 1);
1269 VCPU_EVENT(vcpu
, 3, "%s", "fault in sie instruction");
1270 trace_kvm_s390_sie_fault(vcpu
);
1271 rc
= kvm_s390_inject_program_int(vcpu
, PGM_ADDRESSING
);
1274 memcpy(&vcpu
->run
->s
.regs
.gprs
[14], &vcpu
->arch
.sie_block
->gg14
, 16);
1277 if (kvm_is_ucontrol(vcpu
->kvm
))
1278 /* Don't exit for host interrupts. */
1279 rc
= vcpu
->arch
.sie_block
->icptcode
? -EOPNOTSUPP
: 0;
1281 rc
= kvm_handle_sie_intercept(vcpu
);
1287 static int __vcpu_run(struct kvm_vcpu
*vcpu
)
1289 int rc
, exit_reason
;
1292 * We try to hold kvm->srcu during most of vcpu_run (except when run-
1293 * ning the guest), so that memslots (and other stuff) are protected
1295 vcpu
->srcu_idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
1298 rc
= vcpu_pre_run(vcpu
);
1302 srcu_read_unlock(&vcpu
->kvm
->srcu
, vcpu
->srcu_idx
);
1304 * As PF_VCPU will be used in fault handler, between
1305 * guest_enter and guest_exit should be no uaccess.
1310 exit_reason
= sie64a(vcpu
->arch
.sie_block
,
1311 vcpu
->run
->s
.regs
.gprs
);
1313 vcpu
->srcu_idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
1315 rc
= vcpu_post_run(vcpu
, exit_reason
);
1316 } while (!signal_pending(current
) && !guestdbg_exit_pending(vcpu
) && !rc
);
1318 srcu_read_unlock(&vcpu
->kvm
->srcu
, vcpu
->srcu_idx
);
1322 static void sync_regs(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1324 vcpu
->arch
.sie_block
->gpsw
.mask
= kvm_run
->psw_mask
;
1325 vcpu
->arch
.sie_block
->gpsw
.addr
= kvm_run
->psw_addr
;
1326 if (kvm_run
->kvm_dirty_regs
& KVM_SYNC_PREFIX
)
1327 kvm_s390_set_prefix(vcpu
, kvm_run
->s
.regs
.prefix
);
1328 if (kvm_run
->kvm_dirty_regs
& KVM_SYNC_CRS
) {
1329 memcpy(&vcpu
->arch
.sie_block
->gcr
, &kvm_run
->s
.regs
.crs
, 128);
1330 /* some control register changes require a tlb flush */
1331 kvm_make_request(KVM_REQ_TLB_FLUSH
, vcpu
);
1333 if (kvm_run
->kvm_dirty_regs
& KVM_SYNC_ARCH0
) {
1334 vcpu
->arch
.sie_block
->cputm
= kvm_run
->s
.regs
.cputm
;
1335 vcpu
->arch
.sie_block
->ckc
= kvm_run
->s
.regs
.ckc
;
1336 vcpu
->arch
.sie_block
->todpr
= kvm_run
->s
.regs
.todpr
;
1337 vcpu
->arch
.sie_block
->pp
= kvm_run
->s
.regs
.pp
;
1338 vcpu
->arch
.sie_block
->gbea
= kvm_run
->s
.regs
.gbea
;
1340 if (kvm_run
->kvm_dirty_regs
& KVM_SYNC_PFAULT
) {
1341 vcpu
->arch
.pfault_token
= kvm_run
->s
.regs
.pft
;
1342 vcpu
->arch
.pfault_select
= kvm_run
->s
.regs
.pfs
;
1343 vcpu
->arch
.pfault_compare
= kvm_run
->s
.regs
.pfc
;
1345 kvm_run
->kvm_dirty_regs
= 0;
1348 static void store_regs(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1350 kvm_run
->psw_mask
= vcpu
->arch
.sie_block
->gpsw
.mask
;
1351 kvm_run
->psw_addr
= vcpu
->arch
.sie_block
->gpsw
.addr
;
1352 kvm_run
->s
.regs
.prefix
= kvm_s390_get_prefix(vcpu
);
1353 memcpy(&kvm_run
->s
.regs
.crs
, &vcpu
->arch
.sie_block
->gcr
, 128);
1354 kvm_run
->s
.regs
.cputm
= vcpu
->arch
.sie_block
->cputm
;
1355 kvm_run
->s
.regs
.ckc
= vcpu
->arch
.sie_block
->ckc
;
1356 kvm_run
->s
.regs
.todpr
= vcpu
->arch
.sie_block
->todpr
;
1357 kvm_run
->s
.regs
.pp
= vcpu
->arch
.sie_block
->pp
;
1358 kvm_run
->s
.regs
.gbea
= vcpu
->arch
.sie_block
->gbea
;
1359 kvm_run
->s
.regs
.pft
= vcpu
->arch
.pfault_token
;
1360 kvm_run
->s
.regs
.pfs
= vcpu
->arch
.pfault_select
;
1361 kvm_run
->s
.regs
.pfc
= vcpu
->arch
.pfault_compare
;
1364 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
1369 if (guestdbg_exit_pending(vcpu
)) {
1370 kvm_s390_prepare_debug_exit(vcpu
);
1374 if (vcpu
->sigset_active
)
1375 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
1377 if (!kvm_s390_user_cpu_state_ctrl(vcpu
->kvm
)) {
1378 kvm_s390_vcpu_start(vcpu
);
1379 } else if (is_vcpu_stopped(vcpu
)) {
1380 pr_err_ratelimited("kvm-s390: can't run stopped vcpu %d\n",
1385 sync_regs(vcpu
, kvm_run
);
1388 rc
= __vcpu_run(vcpu
);
1390 if (signal_pending(current
) && !rc
) {
1391 kvm_run
->exit_reason
= KVM_EXIT_INTR
;
1395 if (guestdbg_exit_pending(vcpu
) && !rc
) {
1396 kvm_s390_prepare_debug_exit(vcpu
);
1400 if (rc
== -EOPNOTSUPP
) {
1401 /* intercept cannot be handled in-kernel, prepare kvm-run */
1402 kvm_run
->exit_reason
= KVM_EXIT_S390_SIEIC
;
1403 kvm_run
->s390_sieic
.icptcode
= vcpu
->arch
.sie_block
->icptcode
;
1404 kvm_run
->s390_sieic
.ipa
= vcpu
->arch
.sie_block
->ipa
;
1405 kvm_run
->s390_sieic
.ipb
= vcpu
->arch
.sie_block
->ipb
;
1409 if (rc
== -EREMOTE
) {
1410 /* intercept was handled, but userspace support is needed
1411 * kvm_run has been prepared by the handler */
1415 store_regs(vcpu
, kvm_run
);
1417 if (vcpu
->sigset_active
)
1418 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
1420 vcpu
->stat
.exit_userspace
++;
1425 * store status at address
1426 * we use have two special cases:
1427 * KVM_S390_STORE_STATUS_NOADDR: -> 0x1200 on 64 bit
1428 * KVM_S390_STORE_STATUS_PREFIXED: -> prefix
1430 int kvm_s390_store_status_unloaded(struct kvm_vcpu
*vcpu
, unsigned long gpa
)
1432 unsigned char archmode
= 1;
1437 if (gpa
== KVM_S390_STORE_STATUS_NOADDR
) {
1438 if (write_guest_abs(vcpu
, 163, &archmode
, 1))
1440 gpa
= SAVE_AREA_BASE
;
1441 } else if (gpa
== KVM_S390_STORE_STATUS_PREFIXED
) {
1442 if (write_guest_real(vcpu
, 163, &archmode
, 1))
1444 gpa
= kvm_s390_real_to_abs(vcpu
, SAVE_AREA_BASE
);
1446 rc
= write_guest_abs(vcpu
, gpa
+ offsetof(struct save_area
, fp_regs
),
1447 vcpu
->arch
.guest_fpregs
.fprs
, 128);
1448 rc
|= write_guest_abs(vcpu
, gpa
+ offsetof(struct save_area
, gp_regs
),
1449 vcpu
->run
->s
.regs
.gprs
, 128);
1450 rc
|= write_guest_abs(vcpu
, gpa
+ offsetof(struct save_area
, psw
),
1451 &vcpu
->arch
.sie_block
->gpsw
, 16);
1452 px
= kvm_s390_get_prefix(vcpu
);
1453 rc
|= write_guest_abs(vcpu
, gpa
+ offsetof(struct save_area
, pref_reg
),
1455 rc
|= write_guest_abs(vcpu
,
1456 gpa
+ offsetof(struct save_area
, fp_ctrl_reg
),
1457 &vcpu
->arch
.guest_fpregs
.fpc
, 4);
1458 rc
|= write_guest_abs(vcpu
, gpa
+ offsetof(struct save_area
, tod_reg
),
1459 &vcpu
->arch
.sie_block
->todpr
, 4);
1460 rc
|= write_guest_abs(vcpu
, gpa
+ offsetof(struct save_area
, timer
),
1461 &vcpu
->arch
.sie_block
->cputm
, 8);
1462 clkcomp
= vcpu
->arch
.sie_block
->ckc
>> 8;
1463 rc
|= write_guest_abs(vcpu
, gpa
+ offsetof(struct save_area
, clk_cmp
),
1465 rc
|= write_guest_abs(vcpu
, gpa
+ offsetof(struct save_area
, acc_regs
),
1466 &vcpu
->run
->s
.regs
.acrs
, 64);
1467 rc
|= write_guest_abs(vcpu
, gpa
+ offsetof(struct save_area
, ctrl_regs
),
1468 &vcpu
->arch
.sie_block
->gcr
, 128);
1469 return rc
? -EFAULT
: 0;
1472 int kvm_s390_vcpu_store_status(struct kvm_vcpu
*vcpu
, unsigned long addr
)
1475 * The guest FPRS and ACRS are in the host FPRS/ACRS due to the lazy
1476 * copying in vcpu load/put. Lets update our copies before we save
1477 * it into the save area
1479 save_fp_ctl(&vcpu
->arch
.guest_fpregs
.fpc
);
1480 save_fp_regs(vcpu
->arch
.guest_fpregs
.fprs
);
1481 save_access_regs(vcpu
->run
->s
.regs
.acrs
);
1483 return kvm_s390_store_status_unloaded(vcpu
, addr
);
1486 static void __disable_ibs_on_vcpu(struct kvm_vcpu
*vcpu
)
1488 kvm_check_request(KVM_REQ_ENABLE_IBS
, vcpu
);
1489 kvm_make_request(KVM_REQ_DISABLE_IBS
, vcpu
);
1490 exit_sie_sync(vcpu
);
1493 static void __disable_ibs_on_all_vcpus(struct kvm
*kvm
)
1496 struct kvm_vcpu
*vcpu
;
1498 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
1499 __disable_ibs_on_vcpu(vcpu
);
1503 static void __enable_ibs_on_vcpu(struct kvm_vcpu
*vcpu
)
1505 kvm_check_request(KVM_REQ_DISABLE_IBS
, vcpu
);
1506 kvm_make_request(KVM_REQ_ENABLE_IBS
, vcpu
);
1507 exit_sie_sync(vcpu
);
1510 void kvm_s390_vcpu_start(struct kvm_vcpu
*vcpu
)
1512 int i
, online_vcpus
, started_vcpus
= 0;
1514 if (!is_vcpu_stopped(vcpu
))
1517 trace_kvm_s390_vcpu_start_stop(vcpu
->vcpu_id
, 1);
1518 /* Only one cpu at a time may enter/leave the STOPPED state. */
1519 spin_lock(&vcpu
->kvm
->arch
.start_stop_lock
);
1520 online_vcpus
= atomic_read(&vcpu
->kvm
->online_vcpus
);
1522 for (i
= 0; i
< online_vcpus
; i
++) {
1523 if (!is_vcpu_stopped(vcpu
->kvm
->vcpus
[i
]))
1527 if (started_vcpus
== 0) {
1528 /* we're the only active VCPU -> speed it up */
1529 __enable_ibs_on_vcpu(vcpu
);
1530 } else if (started_vcpus
== 1) {
1532 * As we are starting a second VCPU, we have to disable
1533 * the IBS facility on all VCPUs to remove potentially
1534 * oustanding ENABLE requests.
1536 __disable_ibs_on_all_vcpus(vcpu
->kvm
);
1539 atomic_clear_mask(CPUSTAT_STOPPED
, &vcpu
->arch
.sie_block
->cpuflags
);
1541 * Another VCPU might have used IBS while we were offline.
1542 * Let's play safe and flush the VCPU at startup.
1544 kvm_make_request(KVM_REQ_TLB_FLUSH
, vcpu
);
1545 spin_unlock(&vcpu
->kvm
->arch
.start_stop_lock
);
1549 void kvm_s390_vcpu_stop(struct kvm_vcpu
*vcpu
)
1551 int i
, online_vcpus
, started_vcpus
= 0;
1552 struct kvm_vcpu
*started_vcpu
= NULL
;
1554 if (is_vcpu_stopped(vcpu
))
1557 trace_kvm_s390_vcpu_start_stop(vcpu
->vcpu_id
, 0);
1558 /* Only one cpu at a time may enter/leave the STOPPED state. */
1559 spin_lock(&vcpu
->kvm
->arch
.start_stop_lock
);
1560 online_vcpus
= atomic_read(&vcpu
->kvm
->online_vcpus
);
1562 /* Need to lock access to action_bits to avoid a SIGP race condition */
1563 spin_lock(&vcpu
->arch
.local_int
.lock
);
1564 atomic_set_mask(CPUSTAT_STOPPED
, &vcpu
->arch
.sie_block
->cpuflags
);
1566 /* SIGP STOP and SIGP STOP AND STORE STATUS has been fully processed */
1567 vcpu
->arch
.local_int
.action_bits
&=
1568 ~(ACTION_STOP_ON_STOP
| ACTION_STORE_ON_STOP
);
1569 spin_unlock(&vcpu
->arch
.local_int
.lock
);
1571 __disable_ibs_on_vcpu(vcpu
);
1573 for (i
= 0; i
< online_vcpus
; i
++) {
1574 if (!is_vcpu_stopped(vcpu
->kvm
->vcpus
[i
])) {
1576 started_vcpu
= vcpu
->kvm
->vcpus
[i
];
1580 if (started_vcpus
== 1) {
1582 * As we only have one VCPU left, we want to enable the
1583 * IBS facility for that VCPU to speed it up.
1585 __enable_ibs_on_vcpu(started_vcpu
);
1588 spin_unlock(&vcpu
->kvm
->arch
.start_stop_lock
);
1592 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu
*vcpu
,
1593 struct kvm_enable_cap
*cap
)
1601 case KVM_CAP_S390_CSS_SUPPORT
:
1602 if (!vcpu
->kvm
->arch
.css_support
) {
1603 vcpu
->kvm
->arch
.css_support
= 1;
1604 trace_kvm_s390_enable_css(vcpu
->kvm
);
1615 long kvm_arch_vcpu_ioctl(struct file
*filp
,
1616 unsigned int ioctl
, unsigned long arg
)
1618 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1619 void __user
*argp
= (void __user
*)arg
;
1624 case KVM_S390_INTERRUPT
: {
1625 struct kvm_s390_interrupt s390int
;
1626 struct kvm_s390_irq s390irq
;
1629 if (copy_from_user(&s390int
, argp
, sizeof(s390int
)))
1631 if (s390int_to_s390irq(&s390int
, &s390irq
))
1633 r
= kvm_s390_inject_vcpu(vcpu
, &s390irq
);
1636 case KVM_S390_STORE_STATUS
:
1637 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
1638 r
= kvm_s390_vcpu_store_status(vcpu
, arg
);
1639 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
1641 case KVM_S390_SET_INITIAL_PSW
: {
1645 if (copy_from_user(&psw
, argp
, sizeof(psw
)))
1647 r
= kvm_arch_vcpu_ioctl_set_initial_psw(vcpu
, psw
);
1650 case KVM_S390_INITIAL_RESET
:
1651 r
= kvm_arch_vcpu_ioctl_initial_reset(vcpu
);
1653 case KVM_SET_ONE_REG
:
1654 case KVM_GET_ONE_REG
: {
1655 struct kvm_one_reg reg
;
1657 if (copy_from_user(®
, argp
, sizeof(reg
)))
1659 if (ioctl
== KVM_SET_ONE_REG
)
1660 r
= kvm_arch_vcpu_ioctl_set_one_reg(vcpu
, ®
);
1662 r
= kvm_arch_vcpu_ioctl_get_one_reg(vcpu
, ®
);
1665 #ifdef CONFIG_KVM_S390_UCONTROL
1666 case KVM_S390_UCAS_MAP
: {
1667 struct kvm_s390_ucas_mapping ucasmap
;
1669 if (copy_from_user(&ucasmap
, argp
, sizeof(ucasmap
))) {
1674 if (!kvm_is_ucontrol(vcpu
->kvm
)) {
1679 r
= gmap_map_segment(vcpu
->arch
.gmap
, ucasmap
.user_addr
,
1680 ucasmap
.vcpu_addr
, ucasmap
.length
);
1683 case KVM_S390_UCAS_UNMAP
: {
1684 struct kvm_s390_ucas_mapping ucasmap
;
1686 if (copy_from_user(&ucasmap
, argp
, sizeof(ucasmap
))) {
1691 if (!kvm_is_ucontrol(vcpu
->kvm
)) {
1696 r
= gmap_unmap_segment(vcpu
->arch
.gmap
, ucasmap
.vcpu_addr
,
1701 case KVM_S390_VCPU_FAULT
: {
1702 r
= gmap_fault(vcpu
->arch
.gmap
, arg
, 0);
1705 case KVM_ENABLE_CAP
:
1707 struct kvm_enable_cap cap
;
1709 if (copy_from_user(&cap
, argp
, sizeof(cap
)))
1711 r
= kvm_vcpu_ioctl_enable_cap(vcpu
, &cap
);
1720 int kvm_arch_vcpu_fault(struct kvm_vcpu
*vcpu
, struct vm_fault
*vmf
)
1722 #ifdef CONFIG_KVM_S390_UCONTROL
1723 if ((vmf
->pgoff
== KVM_S390_SIE_PAGE_OFFSET
)
1724 && (kvm_is_ucontrol(vcpu
->kvm
))) {
1725 vmf
->page
= virt_to_page(vcpu
->arch
.sie_block
);
1726 get_page(vmf
->page
);
1730 return VM_FAULT_SIGBUS
;
1733 int kvm_arch_create_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*slot
,
1734 unsigned long npages
)
1739 /* Section: memory related */
1740 int kvm_arch_prepare_memory_region(struct kvm
*kvm
,
1741 struct kvm_memory_slot
*memslot
,
1742 struct kvm_userspace_memory_region
*mem
,
1743 enum kvm_mr_change change
)
1745 /* A few sanity checks. We can have memory slots which have to be
1746 located/ended at a segment boundary (1MB). The memory in userland is
1747 ok to be fragmented into various different vmas. It is okay to mmap()
1748 and munmap() stuff in this slot after doing this call at any time */
1750 if (mem
->userspace_addr
& 0xffffful
)
1753 if (mem
->memory_size
& 0xffffful
)
1759 void kvm_arch_commit_memory_region(struct kvm
*kvm
,
1760 struct kvm_userspace_memory_region
*mem
,
1761 const struct kvm_memory_slot
*old
,
1762 enum kvm_mr_change change
)
1766 /* If the basics of the memslot do not change, we do not want
1767 * to update the gmap. Every update causes several unnecessary
1768 * segment translation exceptions. This is usually handled just
1769 * fine by the normal fault handler + gmap, but it will also
1770 * cause faults on the prefix page of running guest CPUs.
1772 if (old
->userspace_addr
== mem
->userspace_addr
&&
1773 old
->base_gfn
* PAGE_SIZE
== mem
->guest_phys_addr
&&
1774 old
->npages
* PAGE_SIZE
== mem
->memory_size
)
1777 rc
= gmap_map_segment(kvm
->arch
.gmap
, mem
->userspace_addr
,
1778 mem
->guest_phys_addr
, mem
->memory_size
);
1780 printk(KERN_WARNING
"kvm-s390: failed to commit memory region\n");
1784 static int __init
kvm_s390_init(void)
1787 ret
= kvm_init(NULL
, sizeof(struct kvm_vcpu
), 0, THIS_MODULE
);
1792 * guests can ask for up to 255+1 double words, we need a full page
1793 * to hold the maximum amount of facilities. On the other hand, we
1794 * only set facilities that are known to work in KVM.
1796 vfacilities
= (unsigned long *) get_zeroed_page(GFP_KERNEL
|GFP_DMA
);
1801 memcpy(vfacilities
, S390_lowcore
.stfle_fac_list
, 16);
1802 vfacilities
[0] &= 0xff82fffbf47c2000UL
;
1803 vfacilities
[1] &= 0x005c000000000000UL
;
1807 static void __exit
kvm_s390_exit(void)
1809 free_page((unsigned long) vfacilities
);
1813 module_init(kvm_s390_init
);
1814 module_exit(kvm_s390_exit
);
1817 * Enable autoloading of the kvm module.
1818 * Note that we add the module alias here instead of virt/kvm/kvm_main.c
1819 * since x86 takes a different approach.
1821 #include <linux/miscdevice.h>
1822 MODULE_ALIAS_MISCDEV(KVM_MINOR
);
1823 MODULE_ALIAS("devname:kvm");