2 * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
3 * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
6 * Paul Mackerras <paulus@au1.ibm.com>
7 * Alexander Graf <agraf@suse.de>
8 * Kevin Wolf <mail@kevin-wolf.de>
10 * Description: KVM functions specific to running on Book 3S
11 * processors in hypervisor mode (specifically POWER7 and later).
13 * This file is derived from arch/powerpc/kvm/book3s.c,
14 * by Alexander Graf <agraf@suse.de>.
16 * This program is free software; you can redistribute it and/or modify
17 * it under the terms of the GNU General Public License, version 2, as
18 * published by the Free Software Foundation.
21 #include <linux/kvm_host.h>
22 #include <linux/err.h>
23 #include <linux/slab.h>
24 #include <linux/preempt.h>
25 #include <linux/sched.h>
26 #include <linux/delay.h>
27 #include <linux/export.h>
29 #include <linux/anon_inodes.h>
30 #include <linux/cpumask.h>
31 #include <linux/spinlock.h>
32 #include <linux/page-flags.h>
33 #include <linux/srcu.h>
36 #include <asm/cputable.h>
37 #include <asm/cacheflush.h>
38 #include <asm/tlbflush.h>
39 #include <asm/uaccess.h>
41 #include <asm/kvm_ppc.h>
42 #include <asm/kvm_book3s.h>
43 #include <asm/mmu_context.h>
44 #include <asm/lppaca.h>
45 #include <asm/processor.h>
46 #include <asm/cputhreads.h>
48 #include <asm/hvcall.h>
49 #include <asm/switch_to.h>
51 #include <linux/gfp.h>
52 #include <linux/vmalloc.h>
53 #include <linux/highmem.h>
54 #include <linux/hugetlb.h>
56 /* #define EXIT_DEBUG */
57 /* #define EXIT_DEBUG_SIMPLE */
58 /* #define EXIT_DEBUG_INT */
60 /* Used to indicate that a guest page fault needs to be handled */
61 #define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1)
63 /* Used as a "null" value for timebase values */
64 #define TB_NIL (~(u64)0)
66 static void kvmppc_end_cede(struct kvm_vcpu
*vcpu
);
67 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu
*vcpu
);
69 void kvmppc_fast_vcpu_kick(struct kvm_vcpu
*vcpu
)
73 wait_queue_head_t
*wqp
;
75 wqp
= kvm_arch_vcpu_wq(vcpu
);
76 if (waitqueue_active(wqp
)) {
77 wake_up_interruptible(wqp
);
78 ++vcpu
->stat
.halt_wakeup
;
83 /* CPU points to the first thread of the core */
84 if (cpu
!= me
&& cpu
>= 0 && cpu
< nr_cpu_ids
) {
85 int real_cpu
= cpu
+ vcpu
->arch
.ptid
;
86 if (paca
[real_cpu
].kvm_hstate
.xics_phys
)
87 xics_wake_cpu(real_cpu
);
88 else if (cpu_online(cpu
))
89 smp_send_reschedule(cpu
);
95 * We use the vcpu_load/put functions to measure stolen time.
96 * Stolen time is counted as time when either the vcpu is able to
97 * run as part of a virtual core, but the task running the vcore
98 * is preempted or sleeping, or when the vcpu needs something done
99 * in the kernel by the task running the vcpu, but that task is
100 * preempted or sleeping. Those two things have to be counted
101 * separately, since one of the vcpu tasks will take on the job
102 * of running the core, and the other vcpu tasks in the vcore will
103 * sleep waiting for it to do that, but that sleep shouldn't count
106 * Hence we accumulate stolen time when the vcpu can run as part of
107 * a vcore using vc->stolen_tb, and the stolen time when the vcpu
108 * needs its task to do other things in the kernel (for example,
109 * service a page fault) in busy_stolen. We don't accumulate
110 * stolen time for a vcore when it is inactive, or for a vcpu
111 * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of
112 * a misnomer; it means that the vcpu task is not executing in
113 * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
114 * the kernel. We don't have any way of dividing up that time
115 * between time that the vcpu is genuinely stopped, time that
116 * the task is actively working on behalf of the vcpu, and time
117 * that the task is preempted, so we don't count any of it as
120 * Updates to busy_stolen are protected by arch.tbacct_lock;
121 * updates to vc->stolen_tb are protected by the arch.tbacct_lock
122 * of the vcpu that has taken responsibility for running the vcore
123 * (i.e. vc->runner). The stolen times are measured in units of
124 * timebase ticks. (Note that the != TB_NIL checks below are
125 * purely defensive; they should never fail.)
128 void kvmppc_core_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
130 struct kvmppc_vcore
*vc
= vcpu
->arch
.vcore
;
132 spin_lock(&vcpu
->arch
.tbacct_lock
);
133 if (vc
->runner
== vcpu
&& vc
->vcore_state
!= VCORE_INACTIVE
&&
134 vc
->preempt_tb
!= TB_NIL
) {
135 vc
->stolen_tb
+= mftb() - vc
->preempt_tb
;
136 vc
->preempt_tb
= TB_NIL
;
138 if (vcpu
->arch
.state
== KVMPPC_VCPU_BUSY_IN_HOST
&&
139 vcpu
->arch
.busy_preempt
!= TB_NIL
) {
140 vcpu
->arch
.busy_stolen
+= mftb() - vcpu
->arch
.busy_preempt
;
141 vcpu
->arch
.busy_preempt
= TB_NIL
;
143 spin_unlock(&vcpu
->arch
.tbacct_lock
);
146 void kvmppc_core_vcpu_put(struct kvm_vcpu
*vcpu
)
148 struct kvmppc_vcore
*vc
= vcpu
->arch
.vcore
;
150 spin_lock(&vcpu
->arch
.tbacct_lock
);
151 if (vc
->runner
== vcpu
&& vc
->vcore_state
!= VCORE_INACTIVE
)
152 vc
->preempt_tb
= mftb();
153 if (vcpu
->arch
.state
== KVMPPC_VCPU_BUSY_IN_HOST
)
154 vcpu
->arch
.busy_preempt
= mftb();
155 spin_unlock(&vcpu
->arch
.tbacct_lock
);
158 void kvmppc_set_msr(struct kvm_vcpu
*vcpu
, u64 msr
)
160 vcpu
->arch
.shregs
.msr
= msr
;
161 kvmppc_end_cede(vcpu
);
164 void kvmppc_set_pvr(struct kvm_vcpu
*vcpu
, u32 pvr
)
166 vcpu
->arch
.pvr
= pvr
;
169 int kvmppc_set_arch_compat(struct kvm_vcpu
*vcpu
, u32 arch_compat
)
171 unsigned long pcr
= 0;
172 struct kvmppc_vcore
*vc
= vcpu
->arch
.vcore
;
175 if (!cpu_has_feature(CPU_FTR_ARCH_206
))
176 return -EINVAL
; /* 970 has no compat mode support */
178 switch (arch_compat
) {
190 spin_lock(&vc
->lock
);
191 vc
->arch_compat
= arch_compat
;
193 spin_unlock(&vc
->lock
);
198 void kvmppc_dump_regs(struct kvm_vcpu
*vcpu
)
202 pr_err("vcpu %p (%d):\n", vcpu
, vcpu
->vcpu_id
);
203 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
204 vcpu
->arch
.pc
, vcpu
->arch
.shregs
.msr
, vcpu
->arch
.trap
);
205 for (r
= 0; r
< 16; ++r
)
206 pr_err("r%2d = %.16lx r%d = %.16lx\n",
207 r
, kvmppc_get_gpr(vcpu
, r
),
208 r
+16, kvmppc_get_gpr(vcpu
, r
+16));
209 pr_err("ctr = %.16lx lr = %.16lx\n",
210 vcpu
->arch
.ctr
, vcpu
->arch
.lr
);
211 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
212 vcpu
->arch
.shregs
.srr0
, vcpu
->arch
.shregs
.srr1
);
213 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
214 vcpu
->arch
.shregs
.sprg0
, vcpu
->arch
.shregs
.sprg1
);
215 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
216 vcpu
->arch
.shregs
.sprg2
, vcpu
->arch
.shregs
.sprg3
);
217 pr_err("cr = %.8x xer = %.16lx dsisr = %.8x\n",
218 vcpu
->arch
.cr
, vcpu
->arch
.xer
, vcpu
->arch
.shregs
.dsisr
);
219 pr_err("dar = %.16llx\n", vcpu
->arch
.shregs
.dar
);
220 pr_err("fault dar = %.16lx dsisr = %.8x\n",
221 vcpu
->arch
.fault_dar
, vcpu
->arch
.fault_dsisr
);
222 pr_err("SLB (%d entries):\n", vcpu
->arch
.slb_max
);
223 for (r
= 0; r
< vcpu
->arch
.slb_max
; ++r
)
224 pr_err(" ESID = %.16llx VSID = %.16llx\n",
225 vcpu
->arch
.slb
[r
].orige
, vcpu
->arch
.slb
[r
].origv
);
226 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
227 vcpu
->arch
.vcore
->lpcr
, vcpu
->kvm
->arch
.sdr1
,
228 vcpu
->arch
.last_inst
);
231 struct kvm_vcpu
*kvmppc_find_vcpu(struct kvm
*kvm
, int id
)
234 struct kvm_vcpu
*v
, *ret
= NULL
;
236 mutex_lock(&kvm
->lock
);
237 kvm_for_each_vcpu(r
, v
, kvm
) {
238 if (v
->vcpu_id
== id
) {
243 mutex_unlock(&kvm
->lock
);
247 static void init_vpa(struct kvm_vcpu
*vcpu
, struct lppaca
*vpa
)
249 vpa
->__old_status
|= LPPACA_OLD_SHARED_PROC
;
250 vpa
->yield_count
= 1;
253 static int set_vpa(struct kvm_vcpu
*vcpu
, struct kvmppc_vpa
*v
,
254 unsigned long addr
, unsigned long len
)
256 /* check address is cacheline aligned */
257 if (addr
& (L1_CACHE_BYTES
- 1))
259 spin_lock(&vcpu
->arch
.vpa_update_lock
);
260 if (v
->next_gpa
!= addr
|| v
->len
!= len
) {
262 v
->len
= addr
? len
: 0;
263 v
->update_pending
= 1;
265 spin_unlock(&vcpu
->arch
.vpa_update_lock
);
269 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
278 static int vpa_is_registered(struct kvmppc_vpa
*vpap
)
280 if (vpap
->update_pending
)
281 return vpap
->next_gpa
!= 0;
282 return vpap
->pinned_addr
!= NULL
;
285 static unsigned long do_h_register_vpa(struct kvm_vcpu
*vcpu
,
287 unsigned long vcpuid
, unsigned long vpa
)
289 struct kvm
*kvm
= vcpu
->kvm
;
290 unsigned long len
, nb
;
292 struct kvm_vcpu
*tvcpu
;
295 struct kvmppc_vpa
*vpap
;
297 tvcpu
= kvmppc_find_vcpu(kvm
, vcpuid
);
301 subfunc
= (flags
>> H_VPA_FUNC_SHIFT
) & H_VPA_FUNC_MASK
;
302 if (subfunc
== H_VPA_REG_VPA
|| subfunc
== H_VPA_REG_DTL
||
303 subfunc
== H_VPA_REG_SLB
) {
304 /* Registering new area - address must be cache-line aligned */
305 if ((vpa
& (L1_CACHE_BYTES
- 1)) || !vpa
)
308 /* convert logical addr to kernel addr and read length */
309 va
= kvmppc_pin_guest_page(kvm
, vpa
, &nb
);
312 if (subfunc
== H_VPA_REG_VPA
)
313 len
= ((struct reg_vpa
*)va
)->length
.hword
;
315 len
= ((struct reg_vpa
*)va
)->length
.word
;
316 kvmppc_unpin_guest_page(kvm
, va
, vpa
, false);
319 if (len
> nb
|| len
< sizeof(struct reg_vpa
))
328 spin_lock(&tvcpu
->arch
.vpa_update_lock
);
331 case H_VPA_REG_VPA
: /* register VPA */
332 if (len
< sizeof(struct lppaca
))
334 vpap
= &tvcpu
->arch
.vpa
;
338 case H_VPA_REG_DTL
: /* register DTL */
339 if (len
< sizeof(struct dtl_entry
))
341 len
-= len
% sizeof(struct dtl_entry
);
343 /* Check that they have previously registered a VPA */
345 if (!vpa_is_registered(&tvcpu
->arch
.vpa
))
348 vpap
= &tvcpu
->arch
.dtl
;
352 case H_VPA_REG_SLB
: /* register SLB shadow buffer */
353 /* Check that they have previously registered a VPA */
355 if (!vpa_is_registered(&tvcpu
->arch
.vpa
))
358 vpap
= &tvcpu
->arch
.slb_shadow
;
362 case H_VPA_DEREG_VPA
: /* deregister VPA */
363 /* Check they don't still have a DTL or SLB buf registered */
365 if (vpa_is_registered(&tvcpu
->arch
.dtl
) ||
366 vpa_is_registered(&tvcpu
->arch
.slb_shadow
))
369 vpap
= &tvcpu
->arch
.vpa
;
373 case H_VPA_DEREG_DTL
: /* deregister DTL */
374 vpap
= &tvcpu
->arch
.dtl
;
378 case H_VPA_DEREG_SLB
: /* deregister SLB shadow buffer */
379 vpap
= &tvcpu
->arch
.slb_shadow
;
385 vpap
->next_gpa
= vpa
;
387 vpap
->update_pending
= 1;
390 spin_unlock(&tvcpu
->arch
.vpa_update_lock
);
395 static void kvmppc_update_vpa(struct kvm_vcpu
*vcpu
, struct kvmppc_vpa
*vpap
)
397 struct kvm
*kvm
= vcpu
->kvm
;
403 * We need to pin the page pointed to by vpap->next_gpa,
404 * but we can't call kvmppc_pin_guest_page under the lock
405 * as it does get_user_pages() and down_read(). So we
406 * have to drop the lock, pin the page, then get the lock
407 * again and check that a new area didn't get registered
411 gpa
= vpap
->next_gpa
;
412 spin_unlock(&vcpu
->arch
.vpa_update_lock
);
416 va
= kvmppc_pin_guest_page(kvm
, gpa
, &nb
);
417 spin_lock(&vcpu
->arch
.vpa_update_lock
);
418 if (gpa
== vpap
->next_gpa
)
420 /* sigh... unpin that one and try again */
422 kvmppc_unpin_guest_page(kvm
, va
, gpa
, false);
425 vpap
->update_pending
= 0;
426 if (va
&& nb
< vpap
->len
) {
428 * If it's now too short, it must be that userspace
429 * has changed the mappings underlying guest memory,
430 * so unregister the region.
432 kvmppc_unpin_guest_page(kvm
, va
, gpa
, false);
435 if (vpap
->pinned_addr
)
436 kvmppc_unpin_guest_page(kvm
, vpap
->pinned_addr
, vpap
->gpa
,
439 vpap
->pinned_addr
= va
;
442 vpap
->pinned_end
= va
+ vpap
->len
;
445 static void kvmppc_update_vpas(struct kvm_vcpu
*vcpu
)
447 if (!(vcpu
->arch
.vpa
.update_pending
||
448 vcpu
->arch
.slb_shadow
.update_pending
||
449 vcpu
->arch
.dtl
.update_pending
))
452 spin_lock(&vcpu
->arch
.vpa_update_lock
);
453 if (vcpu
->arch
.vpa
.update_pending
) {
454 kvmppc_update_vpa(vcpu
, &vcpu
->arch
.vpa
);
455 if (vcpu
->arch
.vpa
.pinned_addr
)
456 init_vpa(vcpu
, vcpu
->arch
.vpa
.pinned_addr
);
458 if (vcpu
->arch
.dtl
.update_pending
) {
459 kvmppc_update_vpa(vcpu
, &vcpu
->arch
.dtl
);
460 vcpu
->arch
.dtl_ptr
= vcpu
->arch
.dtl
.pinned_addr
;
461 vcpu
->arch
.dtl_index
= 0;
463 if (vcpu
->arch
.slb_shadow
.update_pending
)
464 kvmppc_update_vpa(vcpu
, &vcpu
->arch
.slb_shadow
);
465 spin_unlock(&vcpu
->arch
.vpa_update_lock
);
469 * Return the accumulated stolen time for the vcore up until `now'.
470 * The caller should hold the vcore lock.
472 static u64
vcore_stolen_time(struct kvmppc_vcore
*vc
, u64 now
)
477 * If we are the task running the vcore, then since we hold
478 * the vcore lock, we can't be preempted, so stolen_tb/preempt_tb
479 * can't be updated, so we don't need the tbacct_lock.
480 * If the vcore is inactive, it can't become active (since we
481 * hold the vcore lock), so the vcpu load/put functions won't
482 * update stolen_tb/preempt_tb, and we don't need tbacct_lock.
484 if (vc
->vcore_state
!= VCORE_INACTIVE
&&
485 vc
->runner
->arch
.run_task
!= current
) {
486 spin_lock(&vc
->runner
->arch
.tbacct_lock
);
488 if (vc
->preempt_tb
!= TB_NIL
)
489 p
+= now
- vc
->preempt_tb
;
490 spin_unlock(&vc
->runner
->arch
.tbacct_lock
);
497 static void kvmppc_create_dtl_entry(struct kvm_vcpu
*vcpu
,
498 struct kvmppc_vcore
*vc
)
500 struct dtl_entry
*dt
;
502 unsigned long stolen
;
503 unsigned long core_stolen
;
506 dt
= vcpu
->arch
.dtl_ptr
;
507 vpa
= vcpu
->arch
.vpa
.pinned_addr
;
509 core_stolen
= vcore_stolen_time(vc
, now
);
510 stolen
= core_stolen
- vcpu
->arch
.stolen_logged
;
511 vcpu
->arch
.stolen_logged
= core_stolen
;
512 spin_lock(&vcpu
->arch
.tbacct_lock
);
513 stolen
+= vcpu
->arch
.busy_stolen
;
514 vcpu
->arch
.busy_stolen
= 0;
515 spin_unlock(&vcpu
->arch
.tbacct_lock
);
518 memset(dt
, 0, sizeof(struct dtl_entry
));
519 dt
->dispatch_reason
= 7;
520 dt
->processor_id
= vc
->pcpu
+ vcpu
->arch
.ptid
;
521 dt
->timebase
= now
+ vc
->tb_offset
;
522 dt
->enqueue_to_dispatch_time
= stolen
;
523 dt
->srr0
= kvmppc_get_pc(vcpu
);
524 dt
->srr1
= vcpu
->arch
.shregs
.msr
;
526 if (dt
== vcpu
->arch
.dtl
.pinned_end
)
527 dt
= vcpu
->arch
.dtl
.pinned_addr
;
528 vcpu
->arch
.dtl_ptr
= dt
;
529 /* order writing *dt vs. writing vpa->dtl_idx */
531 vpa
->dtl_idx
= ++vcpu
->arch
.dtl_index
;
532 vcpu
->arch
.dtl
.dirty
= true;
535 int kvmppc_pseries_do_hcall(struct kvm_vcpu
*vcpu
)
537 unsigned long req
= kvmppc_get_gpr(vcpu
, 3);
538 unsigned long target
, ret
= H_SUCCESS
;
539 struct kvm_vcpu
*tvcpu
;
544 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
545 ret
= kvmppc_virtmode_h_enter(vcpu
, kvmppc_get_gpr(vcpu
, 4),
546 kvmppc_get_gpr(vcpu
, 5),
547 kvmppc_get_gpr(vcpu
, 6),
548 kvmppc_get_gpr(vcpu
, 7));
549 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
554 target
= kvmppc_get_gpr(vcpu
, 4);
555 tvcpu
= kvmppc_find_vcpu(vcpu
->kvm
, target
);
560 tvcpu
->arch
.prodded
= 1;
562 if (vcpu
->arch
.ceded
) {
563 if (waitqueue_active(&vcpu
->wq
)) {
564 wake_up_interruptible(&vcpu
->wq
);
565 vcpu
->stat
.halt_wakeup
++;
570 target
= kvmppc_get_gpr(vcpu
, 4);
573 tvcpu
= kvmppc_find_vcpu(vcpu
->kvm
, target
);
578 kvm_vcpu_yield_to(tvcpu
);
581 ret
= do_h_register_vpa(vcpu
, kvmppc_get_gpr(vcpu
, 4),
582 kvmppc_get_gpr(vcpu
, 5),
583 kvmppc_get_gpr(vcpu
, 6));
586 if (list_empty(&vcpu
->kvm
->arch
.rtas_tokens
))
589 rc
= kvmppc_rtas_hcall(vcpu
);
596 /* Send the error out to userspace via KVM_RUN */
605 if (kvmppc_xics_enabled(vcpu
)) {
606 ret
= kvmppc_xics_hcall(vcpu
, req
);
612 kvmppc_set_gpr(vcpu
, 3, ret
);
613 vcpu
->arch
.hcall_needed
= 0;
617 static int kvmppc_handle_exit(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
,
618 struct task_struct
*tsk
)
622 vcpu
->stat
.sum_exits
++;
624 run
->exit_reason
= KVM_EXIT_UNKNOWN
;
625 run
->ready_for_interrupt_injection
= 1;
626 switch (vcpu
->arch
.trap
) {
627 /* We're good on these - the host merely wanted to get our attention */
628 case BOOK3S_INTERRUPT_HV_DECREMENTER
:
629 vcpu
->stat
.dec_exits
++;
632 case BOOK3S_INTERRUPT_EXTERNAL
:
633 vcpu
->stat
.ext_intr_exits
++;
636 case BOOK3S_INTERRUPT_PERFMON
:
639 case BOOK3S_INTERRUPT_MACHINE_CHECK
:
641 * Deliver a machine check interrupt to the guest.
642 * We have to do this, even if the host has handled the
643 * machine check, because machine checks use SRR0/1 and
644 * the interrupt might have trashed guest state in them.
646 kvmppc_book3s_queue_irqprio(vcpu
,
647 BOOK3S_INTERRUPT_MACHINE_CHECK
);
650 case BOOK3S_INTERRUPT_PROGRAM
:
654 * Normally program interrupts are delivered directly
655 * to the guest by the hardware, but we can get here
656 * as a result of a hypervisor emulation interrupt
657 * (e40) getting turned into a 700 by BML RTAS.
659 flags
= vcpu
->arch
.shregs
.msr
& 0x1f0000ull
;
660 kvmppc_core_queue_program(vcpu
, flags
);
664 case BOOK3S_INTERRUPT_SYSCALL
:
666 /* hcall - punt to userspace */
669 if (vcpu
->arch
.shregs
.msr
& MSR_PR
) {
670 /* sc 1 from userspace - reflect to guest syscall */
671 kvmppc_book3s_queue_irqprio(vcpu
, BOOK3S_INTERRUPT_SYSCALL
);
675 run
->papr_hcall
.nr
= kvmppc_get_gpr(vcpu
, 3);
676 for (i
= 0; i
< 9; ++i
)
677 run
->papr_hcall
.args
[i
] = kvmppc_get_gpr(vcpu
, 4 + i
);
678 run
->exit_reason
= KVM_EXIT_PAPR_HCALL
;
679 vcpu
->arch
.hcall_needed
= 1;
684 * We get these next two if the guest accesses a page which it thinks
685 * it has mapped but which is not actually present, either because
686 * it is for an emulated I/O device or because the corresonding
687 * host page has been paged out. Any other HDSI/HISI interrupts
688 * have been handled already.
690 case BOOK3S_INTERRUPT_H_DATA_STORAGE
:
691 r
= RESUME_PAGE_FAULT
;
693 case BOOK3S_INTERRUPT_H_INST_STORAGE
:
694 vcpu
->arch
.fault_dar
= kvmppc_get_pc(vcpu
);
695 vcpu
->arch
.fault_dsisr
= 0;
696 r
= RESUME_PAGE_FAULT
;
699 * This occurs if the guest executes an illegal instruction.
700 * We just generate a program interrupt to the guest, since
701 * we don't emulate any guest instructions at this stage.
703 case BOOK3S_INTERRUPT_H_EMUL_ASSIST
:
704 kvmppc_core_queue_program(vcpu
, 0x80000);
708 kvmppc_dump_regs(vcpu
);
709 printk(KERN_EMERG
"trap=0x%x | pc=0x%lx | msr=0x%llx\n",
710 vcpu
->arch
.trap
, kvmppc_get_pc(vcpu
),
711 vcpu
->arch
.shregs
.msr
);
712 run
->hw
.hardware_exit_reason
= vcpu
->arch
.trap
;
720 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu
*vcpu
,
721 struct kvm_sregs
*sregs
)
725 memset(sregs
, 0, sizeof(struct kvm_sregs
));
726 sregs
->pvr
= vcpu
->arch
.pvr
;
727 for (i
= 0; i
< vcpu
->arch
.slb_max
; i
++) {
728 sregs
->u
.s
.ppc64
.slb
[i
].slbe
= vcpu
->arch
.slb
[i
].orige
;
729 sregs
->u
.s
.ppc64
.slb
[i
].slbv
= vcpu
->arch
.slb
[i
].origv
;
735 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu
*vcpu
,
736 struct kvm_sregs
*sregs
)
740 kvmppc_set_pvr(vcpu
, sregs
->pvr
);
743 for (i
= 0; i
< vcpu
->arch
.slb_nr
; i
++) {
744 if (sregs
->u
.s
.ppc64
.slb
[i
].slbe
& SLB_ESID_V
) {
745 vcpu
->arch
.slb
[j
].orige
= sregs
->u
.s
.ppc64
.slb
[i
].slbe
;
746 vcpu
->arch
.slb
[j
].origv
= sregs
->u
.s
.ppc64
.slb
[i
].slbv
;
750 vcpu
->arch
.slb_max
= j
;
755 static void kvmppc_set_lpcr(struct kvm_vcpu
*vcpu
, u64 new_lpcr
)
757 struct kvmppc_vcore
*vc
= vcpu
->arch
.vcore
;
760 spin_lock(&vc
->lock
);
762 * Userspace can only modify DPFD (default prefetch depth),
763 * ILE (interrupt little-endian) and TC (translation control).
765 mask
= LPCR_DPFD
| LPCR_ILE
| LPCR_TC
;
766 vc
->lpcr
= (vc
->lpcr
& ~mask
) | (new_lpcr
& mask
);
767 spin_unlock(&vc
->lock
);
770 int kvmppc_get_one_reg(struct kvm_vcpu
*vcpu
, u64 id
, union kvmppc_one_reg
*val
)
776 case KVM_REG_PPC_HIOR
:
777 *val
= get_reg_val(id
, 0);
779 case KVM_REG_PPC_DABR
:
780 *val
= get_reg_val(id
, vcpu
->arch
.dabr
);
782 case KVM_REG_PPC_DSCR
:
783 *val
= get_reg_val(id
, vcpu
->arch
.dscr
);
785 case KVM_REG_PPC_PURR
:
786 *val
= get_reg_val(id
, vcpu
->arch
.purr
);
788 case KVM_REG_PPC_SPURR
:
789 *val
= get_reg_val(id
, vcpu
->arch
.spurr
);
791 case KVM_REG_PPC_AMR
:
792 *val
= get_reg_val(id
, vcpu
->arch
.amr
);
794 case KVM_REG_PPC_UAMOR
:
795 *val
= get_reg_val(id
, vcpu
->arch
.uamor
);
797 case KVM_REG_PPC_MMCR0
... KVM_REG_PPC_MMCRA
:
798 i
= id
- KVM_REG_PPC_MMCR0
;
799 *val
= get_reg_val(id
, vcpu
->arch
.mmcr
[i
]);
801 case KVM_REG_PPC_PMC1
... KVM_REG_PPC_PMC8
:
802 i
= id
- KVM_REG_PPC_PMC1
;
803 *val
= get_reg_val(id
, vcpu
->arch
.pmc
[i
]);
805 case KVM_REG_PPC_SIAR
:
806 *val
= get_reg_val(id
, vcpu
->arch
.siar
);
808 case KVM_REG_PPC_SDAR
:
809 *val
= get_reg_val(id
, vcpu
->arch
.sdar
);
812 case KVM_REG_PPC_FPR0
... KVM_REG_PPC_FPR31
:
813 if (cpu_has_feature(CPU_FTR_VSX
)) {
814 /* VSX => FP reg i is stored in arch.vsr[2*i] */
815 long int i
= id
- KVM_REG_PPC_FPR0
;
816 *val
= get_reg_val(id
, vcpu
->arch
.vsr
[2 * i
]);
818 /* let generic code handle it */
822 case KVM_REG_PPC_VSR0
... KVM_REG_PPC_VSR31
:
823 if (cpu_has_feature(CPU_FTR_VSX
)) {
824 long int i
= id
- KVM_REG_PPC_VSR0
;
825 val
->vsxval
[0] = vcpu
->arch
.vsr
[2 * i
];
826 val
->vsxval
[1] = vcpu
->arch
.vsr
[2 * i
+ 1];
831 #endif /* CONFIG_VSX */
832 case KVM_REG_PPC_VPA_ADDR
:
833 spin_lock(&vcpu
->arch
.vpa_update_lock
);
834 *val
= get_reg_val(id
, vcpu
->arch
.vpa
.next_gpa
);
835 spin_unlock(&vcpu
->arch
.vpa_update_lock
);
837 case KVM_REG_PPC_VPA_SLB
:
838 spin_lock(&vcpu
->arch
.vpa_update_lock
);
839 val
->vpaval
.addr
= vcpu
->arch
.slb_shadow
.next_gpa
;
840 val
->vpaval
.length
= vcpu
->arch
.slb_shadow
.len
;
841 spin_unlock(&vcpu
->arch
.vpa_update_lock
);
843 case KVM_REG_PPC_VPA_DTL
:
844 spin_lock(&vcpu
->arch
.vpa_update_lock
);
845 val
->vpaval
.addr
= vcpu
->arch
.dtl
.next_gpa
;
846 val
->vpaval
.length
= vcpu
->arch
.dtl
.len
;
847 spin_unlock(&vcpu
->arch
.vpa_update_lock
);
849 case KVM_REG_PPC_TB_OFFSET
:
850 *val
= get_reg_val(id
, vcpu
->arch
.vcore
->tb_offset
);
852 case KVM_REG_PPC_LPCR
:
853 *val
= get_reg_val(id
, vcpu
->arch
.vcore
->lpcr
);
855 case KVM_REG_PPC_PPR
:
856 *val
= get_reg_val(id
, vcpu
->arch
.ppr
);
858 case KVM_REG_PPC_ARCH_COMPAT
:
859 *val
= get_reg_val(id
, vcpu
->arch
.vcore
->arch_compat
);
869 int kvmppc_set_one_reg(struct kvm_vcpu
*vcpu
, u64 id
, union kvmppc_one_reg
*val
)
873 unsigned long addr
, len
;
876 case KVM_REG_PPC_HIOR
:
877 /* Only allow this to be set to zero */
878 if (set_reg_val(id
, *val
))
881 case KVM_REG_PPC_DABR
:
882 vcpu
->arch
.dabr
= set_reg_val(id
, *val
);
884 case KVM_REG_PPC_DSCR
:
885 vcpu
->arch
.dscr
= set_reg_val(id
, *val
);
887 case KVM_REG_PPC_PURR
:
888 vcpu
->arch
.purr
= set_reg_val(id
, *val
);
890 case KVM_REG_PPC_SPURR
:
891 vcpu
->arch
.spurr
= set_reg_val(id
, *val
);
893 case KVM_REG_PPC_AMR
:
894 vcpu
->arch
.amr
= set_reg_val(id
, *val
);
896 case KVM_REG_PPC_UAMOR
:
897 vcpu
->arch
.uamor
= set_reg_val(id
, *val
);
899 case KVM_REG_PPC_MMCR0
... KVM_REG_PPC_MMCRA
:
900 i
= id
- KVM_REG_PPC_MMCR0
;
901 vcpu
->arch
.mmcr
[i
] = set_reg_val(id
, *val
);
903 case KVM_REG_PPC_PMC1
... KVM_REG_PPC_PMC8
:
904 i
= id
- KVM_REG_PPC_PMC1
;
905 vcpu
->arch
.pmc
[i
] = set_reg_val(id
, *val
);
907 case KVM_REG_PPC_SIAR
:
908 vcpu
->arch
.siar
= set_reg_val(id
, *val
);
910 case KVM_REG_PPC_SDAR
:
911 vcpu
->arch
.sdar
= set_reg_val(id
, *val
);
914 case KVM_REG_PPC_FPR0
... KVM_REG_PPC_FPR31
:
915 if (cpu_has_feature(CPU_FTR_VSX
)) {
916 /* VSX => FP reg i is stored in arch.vsr[2*i] */
917 long int i
= id
- KVM_REG_PPC_FPR0
;
918 vcpu
->arch
.vsr
[2 * i
] = set_reg_val(id
, *val
);
920 /* let generic code handle it */
924 case KVM_REG_PPC_VSR0
... KVM_REG_PPC_VSR31
:
925 if (cpu_has_feature(CPU_FTR_VSX
)) {
926 long int i
= id
- KVM_REG_PPC_VSR0
;
927 vcpu
->arch
.vsr
[2 * i
] = val
->vsxval
[0];
928 vcpu
->arch
.vsr
[2 * i
+ 1] = val
->vsxval
[1];
933 #endif /* CONFIG_VSX */
934 case KVM_REG_PPC_VPA_ADDR
:
935 addr
= set_reg_val(id
, *val
);
937 if (!addr
&& (vcpu
->arch
.slb_shadow
.next_gpa
||
938 vcpu
->arch
.dtl
.next_gpa
))
940 r
= set_vpa(vcpu
, &vcpu
->arch
.vpa
, addr
, sizeof(struct lppaca
));
942 case KVM_REG_PPC_VPA_SLB
:
943 addr
= val
->vpaval
.addr
;
944 len
= val
->vpaval
.length
;
946 if (addr
&& !vcpu
->arch
.vpa
.next_gpa
)
948 r
= set_vpa(vcpu
, &vcpu
->arch
.slb_shadow
, addr
, len
);
950 case KVM_REG_PPC_VPA_DTL
:
951 addr
= val
->vpaval
.addr
;
952 len
= val
->vpaval
.length
;
954 if (addr
&& (len
< sizeof(struct dtl_entry
) ||
955 !vcpu
->arch
.vpa
.next_gpa
))
957 len
-= len
% sizeof(struct dtl_entry
);
958 r
= set_vpa(vcpu
, &vcpu
->arch
.dtl
, addr
, len
);
960 case KVM_REG_PPC_TB_OFFSET
:
961 /* round up to multiple of 2^24 */
962 vcpu
->arch
.vcore
->tb_offset
=
963 ALIGN(set_reg_val(id
, *val
), 1UL << 24);
965 case KVM_REG_PPC_LPCR
:
966 kvmppc_set_lpcr(vcpu
, set_reg_val(id
, *val
));
968 case KVM_REG_PPC_PPR
:
969 vcpu
->arch
.ppr
= set_reg_val(id
, *val
);
971 case KVM_REG_PPC_ARCH_COMPAT
:
972 r
= kvmppc_set_arch_compat(vcpu
, set_reg_val(id
, *val
));
982 int kvmppc_core_check_processor_compat(void)
984 if (cpu_has_feature(CPU_FTR_HVMODE
))
989 struct kvm_vcpu
*kvmppc_core_vcpu_create(struct kvm
*kvm
, unsigned int id
)
991 struct kvm_vcpu
*vcpu
;
994 struct kvmppc_vcore
*vcore
;
996 core
= id
/ threads_per_core
;
997 if (core
>= KVM_MAX_VCORES
)
1001 vcpu
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL
);
1005 err
= kvm_vcpu_init(vcpu
, kvm
, id
);
1009 vcpu
->arch
.shared
= &vcpu
->arch
.shregs
;
1010 vcpu
->arch
.mmcr
[0] = MMCR0_FC
;
1011 vcpu
->arch
.ctrl
= CTRL_RUNLATCH
;
1012 /* default to host PVR, since we can't spoof it */
1013 vcpu
->arch
.pvr
= mfspr(SPRN_PVR
);
1014 kvmppc_set_pvr(vcpu
, vcpu
->arch
.pvr
);
1015 spin_lock_init(&vcpu
->arch
.vpa_update_lock
);
1016 spin_lock_init(&vcpu
->arch
.tbacct_lock
);
1017 vcpu
->arch
.busy_preempt
= TB_NIL
;
1019 kvmppc_mmu_book3s_hv_init(vcpu
);
1021 vcpu
->arch
.state
= KVMPPC_VCPU_NOTREADY
;
1023 init_waitqueue_head(&vcpu
->arch
.cpu_run
);
1025 mutex_lock(&kvm
->lock
);
1026 vcore
= kvm
->arch
.vcores
[core
];
1028 vcore
= kzalloc(sizeof(struct kvmppc_vcore
), GFP_KERNEL
);
1030 INIT_LIST_HEAD(&vcore
->runnable_threads
);
1031 spin_lock_init(&vcore
->lock
);
1032 init_waitqueue_head(&vcore
->wq
);
1033 vcore
->preempt_tb
= TB_NIL
;
1034 vcore
->lpcr
= kvm
->arch
.lpcr
;
1036 kvm
->arch
.vcores
[core
] = vcore
;
1037 kvm
->arch
.online_vcores
++;
1039 mutex_unlock(&kvm
->lock
);
1044 spin_lock(&vcore
->lock
);
1045 ++vcore
->num_threads
;
1046 spin_unlock(&vcore
->lock
);
1047 vcpu
->arch
.vcore
= vcore
;
1049 vcpu
->arch
.cpu_type
= KVM_CPU_3S_64
;
1050 kvmppc_sanity_check(vcpu
);
1055 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
1057 return ERR_PTR(err
);
1060 static void unpin_vpa(struct kvm
*kvm
, struct kvmppc_vpa
*vpa
)
1062 if (vpa
->pinned_addr
)
1063 kvmppc_unpin_guest_page(kvm
, vpa
->pinned_addr
, vpa
->gpa
,
1067 void kvmppc_core_vcpu_free(struct kvm_vcpu
*vcpu
)
1069 spin_lock(&vcpu
->arch
.vpa_update_lock
);
1070 unpin_vpa(vcpu
->kvm
, &vcpu
->arch
.dtl
);
1071 unpin_vpa(vcpu
->kvm
, &vcpu
->arch
.slb_shadow
);
1072 unpin_vpa(vcpu
->kvm
, &vcpu
->arch
.vpa
);
1073 spin_unlock(&vcpu
->arch
.vpa_update_lock
);
1074 kvm_vcpu_uninit(vcpu
);
1075 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
1078 static void kvmppc_set_timer(struct kvm_vcpu
*vcpu
)
1080 unsigned long dec_nsec
, now
;
1083 if (now
> vcpu
->arch
.dec_expires
) {
1084 /* decrementer has already gone negative */
1085 kvmppc_core_queue_dec(vcpu
);
1086 kvmppc_core_prepare_to_enter(vcpu
);
1089 dec_nsec
= (vcpu
->arch
.dec_expires
- now
) * NSEC_PER_SEC
1091 hrtimer_start(&vcpu
->arch
.dec_timer
, ktime_set(0, dec_nsec
),
1093 vcpu
->arch
.timer_running
= 1;
1096 static void kvmppc_end_cede(struct kvm_vcpu
*vcpu
)
1098 vcpu
->arch
.ceded
= 0;
1099 if (vcpu
->arch
.timer_running
) {
1100 hrtimer_try_to_cancel(&vcpu
->arch
.dec_timer
);
1101 vcpu
->arch
.timer_running
= 0;
1105 extern int __kvmppc_vcore_entry(struct kvm_run
*kvm_run
, struct kvm_vcpu
*vcpu
);
1107 static void kvmppc_remove_runnable(struct kvmppc_vcore
*vc
,
1108 struct kvm_vcpu
*vcpu
)
1112 if (vcpu
->arch
.state
!= KVMPPC_VCPU_RUNNABLE
)
1114 spin_lock(&vcpu
->arch
.tbacct_lock
);
1116 vcpu
->arch
.busy_stolen
+= vcore_stolen_time(vc
, now
) -
1117 vcpu
->arch
.stolen_logged
;
1118 vcpu
->arch
.busy_preempt
= now
;
1119 vcpu
->arch
.state
= KVMPPC_VCPU_BUSY_IN_HOST
;
1120 spin_unlock(&vcpu
->arch
.tbacct_lock
);
1122 list_del(&vcpu
->arch
.run_list
);
1125 static int kvmppc_grab_hwthread(int cpu
)
1127 struct paca_struct
*tpaca
;
1128 long timeout
= 1000;
1132 /* Ensure the thread won't go into the kernel if it wakes */
1133 tpaca
->kvm_hstate
.hwthread_req
= 1;
1134 tpaca
->kvm_hstate
.kvm_vcpu
= NULL
;
1137 * If the thread is already executing in the kernel (e.g. handling
1138 * a stray interrupt), wait for it to get back to nap mode.
1139 * The smp_mb() is to ensure that our setting of hwthread_req
1140 * is visible before we look at hwthread_state, so if this
1141 * races with the code at system_reset_pSeries and the thread
1142 * misses our setting of hwthread_req, we are sure to see its
1143 * setting of hwthread_state, and vice versa.
1146 while (tpaca
->kvm_hstate
.hwthread_state
== KVM_HWTHREAD_IN_KERNEL
) {
1147 if (--timeout
<= 0) {
1148 pr_err("KVM: couldn't grab cpu %d\n", cpu
);
1156 static void kvmppc_release_hwthread(int cpu
)
1158 struct paca_struct
*tpaca
;
1161 tpaca
->kvm_hstate
.hwthread_req
= 0;
1162 tpaca
->kvm_hstate
.kvm_vcpu
= NULL
;
1165 static void kvmppc_start_thread(struct kvm_vcpu
*vcpu
)
1168 struct paca_struct
*tpaca
;
1169 struct kvmppc_vcore
*vc
= vcpu
->arch
.vcore
;
1171 if (vcpu
->arch
.timer_running
) {
1172 hrtimer_try_to_cancel(&vcpu
->arch
.dec_timer
);
1173 vcpu
->arch
.timer_running
= 0;
1175 cpu
= vc
->pcpu
+ vcpu
->arch
.ptid
;
1177 tpaca
->kvm_hstate
.kvm_vcpu
= vcpu
;
1178 tpaca
->kvm_hstate
.kvm_vcore
= vc
;
1179 tpaca
->kvm_hstate
.napping
= 0;
1180 vcpu
->cpu
= vc
->pcpu
;
1182 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
1183 if (vcpu
->arch
.ptid
) {
1190 static void kvmppc_wait_for_nap(struct kvmppc_vcore
*vc
)
1196 while (vc
->nap_count
< vc
->n_woken
) {
1197 if (++i
>= 1000000) {
1198 pr_err("kvmppc_wait_for_nap timeout %d %d\n",
1199 vc
->nap_count
, vc
->n_woken
);
1208 * Check that we are on thread 0 and that any other threads in
1209 * this core are off-line. Then grab the threads so they can't
1212 static int on_primary_thread(void)
1214 int cpu
= smp_processor_id();
1215 int thr
= cpu_thread_in_core(cpu
);
1219 while (++thr
< threads_per_core
)
1220 if (cpu_online(cpu
+ thr
))
1223 /* Grab all hw threads so they can't go into the kernel */
1224 for (thr
= 1; thr
< threads_per_core
; ++thr
) {
1225 if (kvmppc_grab_hwthread(cpu
+ thr
)) {
1226 /* Couldn't grab one; let the others go */
1228 kvmppc_release_hwthread(cpu
+ thr
);
1229 } while (--thr
> 0);
1237 * Run a set of guest threads on a physical core.
1238 * Called with vc->lock held.
1240 static void kvmppc_run_core(struct kvmppc_vcore
*vc
)
1242 struct kvm_vcpu
*vcpu
, *vcpu0
, *vnext
;
1245 int ptid
, i
, need_vpa_update
;
1247 struct kvm_vcpu
*vcpus_to_update
[threads_per_core
];
1249 /* don't start if any threads have a signal pending */
1250 need_vpa_update
= 0;
1251 list_for_each_entry(vcpu
, &vc
->runnable_threads
, arch
.run_list
) {
1252 if (signal_pending(vcpu
->arch
.run_task
))
1254 if (vcpu
->arch
.vpa
.update_pending
||
1255 vcpu
->arch
.slb_shadow
.update_pending
||
1256 vcpu
->arch
.dtl
.update_pending
)
1257 vcpus_to_update
[need_vpa_update
++] = vcpu
;
1261 * Initialize *vc, in particular vc->vcore_state, so we can
1262 * drop the vcore lock if necessary.
1266 vc
->entry_exit_count
= 0;
1267 vc
->vcore_state
= VCORE_STARTING
;
1269 vc
->napping_threads
= 0;
1272 * Updating any of the vpas requires calling kvmppc_pin_guest_page,
1273 * which can't be called with any spinlocks held.
1275 if (need_vpa_update
) {
1276 spin_unlock(&vc
->lock
);
1277 for (i
= 0; i
< need_vpa_update
; ++i
)
1278 kvmppc_update_vpas(vcpus_to_update
[i
]);
1279 spin_lock(&vc
->lock
);
1283 * Assign physical thread IDs, first to non-ceded vcpus
1284 * and then to ceded ones.
1288 list_for_each_entry(vcpu
, &vc
->runnable_threads
, arch
.run_list
) {
1289 if (!vcpu
->arch
.ceded
) {
1292 vcpu
->arch
.ptid
= ptid
++;
1296 goto out
; /* nothing to run; should never happen */
1297 list_for_each_entry(vcpu
, &vc
->runnable_threads
, arch
.run_list
)
1298 if (vcpu
->arch
.ceded
)
1299 vcpu
->arch
.ptid
= ptid
++;
1302 * Make sure we are running on thread 0, and that
1303 * secondary threads are offline.
1305 if (threads_per_core
> 1 && !on_primary_thread()) {
1306 list_for_each_entry(vcpu
, &vc
->runnable_threads
, arch
.run_list
)
1307 vcpu
->arch
.ret
= -EBUSY
;
1311 vc
->pcpu
= smp_processor_id();
1312 list_for_each_entry(vcpu
, &vc
->runnable_threads
, arch
.run_list
) {
1313 kvmppc_start_thread(vcpu
);
1314 kvmppc_create_dtl_entry(vcpu
, vc
);
1317 vc
->vcore_state
= VCORE_RUNNING
;
1319 spin_unlock(&vc
->lock
);
1323 srcu_idx
= srcu_read_lock(&vcpu0
->kvm
->srcu
);
1325 __kvmppc_vcore_entry(NULL
, vcpu0
);
1327 spin_lock(&vc
->lock
);
1328 /* disable sending of IPIs on virtual external irqs */
1329 list_for_each_entry(vcpu
, &vc
->runnable_threads
, arch
.run_list
)
1331 /* wait for secondary threads to finish writing their state to memory */
1332 if (vc
->nap_count
< vc
->n_woken
)
1333 kvmppc_wait_for_nap(vc
);
1334 for (i
= 0; i
< threads_per_core
; ++i
)
1335 kvmppc_release_hwthread(vc
->pcpu
+ i
);
1336 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
1337 vc
->vcore_state
= VCORE_EXITING
;
1338 spin_unlock(&vc
->lock
);
1340 srcu_read_unlock(&vcpu0
->kvm
->srcu
, srcu_idx
);
1342 /* make sure updates to secondary vcpu structs are visible now */
1349 spin_lock(&vc
->lock
);
1351 list_for_each_entry(vcpu
, &vc
->runnable_threads
, arch
.run_list
) {
1352 /* cancel pending dec exception if dec is positive */
1353 if (now
< vcpu
->arch
.dec_expires
&&
1354 kvmppc_core_pending_dec(vcpu
))
1355 kvmppc_core_dequeue_dec(vcpu
);
1358 if (vcpu
->arch
.trap
)
1359 ret
= kvmppc_handle_exit(vcpu
->arch
.kvm_run
, vcpu
,
1360 vcpu
->arch
.run_task
);
1362 vcpu
->arch
.ret
= ret
;
1363 vcpu
->arch
.trap
= 0;
1365 if (vcpu
->arch
.ceded
) {
1366 if (ret
!= RESUME_GUEST
)
1367 kvmppc_end_cede(vcpu
);
1369 kvmppc_set_timer(vcpu
);
1374 vc
->vcore_state
= VCORE_INACTIVE
;
1375 list_for_each_entry_safe(vcpu
, vnext
, &vc
->runnable_threads
,
1377 if (vcpu
->arch
.ret
!= RESUME_GUEST
) {
1378 kvmppc_remove_runnable(vc
, vcpu
);
1379 wake_up(&vcpu
->arch
.cpu_run
);
1385 * Wait for some other vcpu thread to execute us, and
1386 * wake us up when we need to handle something in the host.
1388 static void kvmppc_wait_for_exec(struct kvm_vcpu
*vcpu
, int wait_state
)
1392 prepare_to_wait(&vcpu
->arch
.cpu_run
, &wait
, wait_state
);
1393 if (vcpu
->arch
.state
== KVMPPC_VCPU_RUNNABLE
)
1395 finish_wait(&vcpu
->arch
.cpu_run
, &wait
);
1399 * All the vcpus in this vcore are idle, so wait for a decrementer
1400 * or external interrupt to one of the vcpus. vc->lock is held.
1402 static void kvmppc_vcore_blocked(struct kvmppc_vcore
*vc
)
1406 prepare_to_wait(&vc
->wq
, &wait
, TASK_INTERRUPTIBLE
);
1407 vc
->vcore_state
= VCORE_SLEEPING
;
1408 spin_unlock(&vc
->lock
);
1410 finish_wait(&vc
->wq
, &wait
);
1411 spin_lock(&vc
->lock
);
1412 vc
->vcore_state
= VCORE_INACTIVE
;
1415 static int kvmppc_run_vcpu(struct kvm_run
*kvm_run
, struct kvm_vcpu
*vcpu
)
1418 struct kvmppc_vcore
*vc
;
1419 struct kvm_vcpu
*v
, *vn
;
1421 kvm_run
->exit_reason
= 0;
1422 vcpu
->arch
.ret
= RESUME_GUEST
;
1423 vcpu
->arch
.trap
= 0;
1424 kvmppc_update_vpas(vcpu
);
1427 * Synchronize with other threads in this virtual core
1429 vc
= vcpu
->arch
.vcore
;
1430 spin_lock(&vc
->lock
);
1431 vcpu
->arch
.ceded
= 0;
1432 vcpu
->arch
.run_task
= current
;
1433 vcpu
->arch
.kvm_run
= kvm_run
;
1434 vcpu
->arch
.stolen_logged
= vcore_stolen_time(vc
, mftb());
1435 vcpu
->arch
.state
= KVMPPC_VCPU_RUNNABLE
;
1436 vcpu
->arch
.busy_preempt
= TB_NIL
;
1437 list_add_tail(&vcpu
->arch
.run_list
, &vc
->runnable_threads
);
1441 * This happens the first time this is called for a vcpu.
1442 * If the vcore is already running, we may be able to start
1443 * this thread straight away and have it join in.
1445 if (!signal_pending(current
)) {
1446 if (vc
->vcore_state
== VCORE_RUNNING
&&
1447 VCORE_EXIT_COUNT(vc
) == 0) {
1448 vcpu
->arch
.ptid
= vc
->n_runnable
- 1;
1449 kvmppc_create_dtl_entry(vcpu
, vc
);
1450 kvmppc_start_thread(vcpu
);
1451 } else if (vc
->vcore_state
== VCORE_SLEEPING
) {
1457 while (vcpu
->arch
.state
== KVMPPC_VCPU_RUNNABLE
&&
1458 !signal_pending(current
)) {
1459 if (vc
->vcore_state
!= VCORE_INACTIVE
) {
1460 spin_unlock(&vc
->lock
);
1461 kvmppc_wait_for_exec(vcpu
, TASK_INTERRUPTIBLE
);
1462 spin_lock(&vc
->lock
);
1465 list_for_each_entry_safe(v
, vn
, &vc
->runnable_threads
,
1467 kvmppc_core_prepare_to_enter(v
);
1468 if (signal_pending(v
->arch
.run_task
)) {
1469 kvmppc_remove_runnable(vc
, v
);
1470 v
->stat
.signal_exits
++;
1471 v
->arch
.kvm_run
->exit_reason
= KVM_EXIT_INTR
;
1472 v
->arch
.ret
= -EINTR
;
1473 wake_up(&v
->arch
.cpu_run
);
1476 if (!vc
->n_runnable
|| vcpu
->arch
.state
!= KVMPPC_VCPU_RUNNABLE
)
1480 list_for_each_entry(v
, &vc
->runnable_threads
, arch
.run_list
) {
1481 if (!v
->arch
.pending_exceptions
)
1482 n_ceded
+= v
->arch
.ceded
;
1486 if (n_ceded
== vc
->n_runnable
)
1487 kvmppc_vcore_blocked(vc
);
1489 kvmppc_run_core(vc
);
1493 while (vcpu
->arch
.state
== KVMPPC_VCPU_RUNNABLE
&&
1494 (vc
->vcore_state
== VCORE_RUNNING
||
1495 vc
->vcore_state
== VCORE_EXITING
)) {
1496 spin_unlock(&vc
->lock
);
1497 kvmppc_wait_for_exec(vcpu
, TASK_UNINTERRUPTIBLE
);
1498 spin_lock(&vc
->lock
);
1501 if (vcpu
->arch
.state
== KVMPPC_VCPU_RUNNABLE
) {
1502 kvmppc_remove_runnable(vc
, vcpu
);
1503 vcpu
->stat
.signal_exits
++;
1504 kvm_run
->exit_reason
= KVM_EXIT_INTR
;
1505 vcpu
->arch
.ret
= -EINTR
;
1508 if (vc
->n_runnable
&& vc
->vcore_state
== VCORE_INACTIVE
) {
1509 /* Wake up some vcpu to run the core */
1510 v
= list_first_entry(&vc
->runnable_threads
,
1511 struct kvm_vcpu
, arch
.run_list
);
1512 wake_up(&v
->arch
.cpu_run
);
1515 spin_unlock(&vc
->lock
);
1516 return vcpu
->arch
.ret
;
1519 int kvmppc_vcpu_run(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
)
1524 if (!vcpu
->arch
.sane
) {
1525 run
->exit_reason
= KVM_EXIT_INTERNAL_ERROR
;
1529 kvmppc_core_prepare_to_enter(vcpu
);
1531 /* No need to go into the guest when all we'll do is come back out */
1532 if (signal_pending(current
)) {
1533 run
->exit_reason
= KVM_EXIT_INTR
;
1537 atomic_inc(&vcpu
->kvm
->arch
.vcpus_running
);
1538 /* Order vcpus_running vs. rma_setup_done, see kvmppc_alloc_reset_hpt */
1541 /* On the first time here, set up HTAB and VRMA or RMA */
1542 if (!vcpu
->kvm
->arch
.rma_setup_done
) {
1543 r
= kvmppc_hv_setup_htab_rma(vcpu
);
1548 flush_fp_to_thread(current
);
1549 flush_altivec_to_thread(current
);
1550 flush_vsx_to_thread(current
);
1551 vcpu
->arch
.wqp
= &vcpu
->arch
.vcore
->wq
;
1552 vcpu
->arch
.pgdir
= current
->mm
->pgd
;
1553 vcpu
->arch
.state
= KVMPPC_VCPU_BUSY_IN_HOST
;
1556 r
= kvmppc_run_vcpu(run
, vcpu
);
1558 if (run
->exit_reason
== KVM_EXIT_PAPR_HCALL
&&
1559 !(vcpu
->arch
.shregs
.msr
& MSR_PR
)) {
1560 r
= kvmppc_pseries_do_hcall(vcpu
);
1561 kvmppc_core_prepare_to_enter(vcpu
);
1562 } else if (r
== RESUME_PAGE_FAULT
) {
1563 srcu_idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
1564 r
= kvmppc_book3s_hv_page_fault(run
, vcpu
,
1565 vcpu
->arch
.fault_dar
, vcpu
->arch
.fault_dsisr
);
1566 srcu_read_unlock(&vcpu
->kvm
->srcu
, srcu_idx
);
1568 } while (r
== RESUME_GUEST
);
1571 vcpu
->arch
.state
= KVMPPC_VCPU_NOTREADY
;
1572 atomic_dec(&vcpu
->kvm
->arch
.vcpus_running
);
1577 /* Work out RMLS (real mode limit selector) field value for a given RMA size.
1578 Assumes POWER7 or PPC970. */
1579 static inline int lpcr_rmls(unsigned long rma_size
)
1582 case 32ul << 20: /* 32 MB */
1583 if (cpu_has_feature(CPU_FTR_ARCH_206
))
1584 return 8; /* only supported on POWER7 */
1586 case 64ul << 20: /* 64 MB */
1588 case 128ul << 20: /* 128 MB */
1590 case 256ul << 20: /* 256 MB */
1592 case 1ul << 30: /* 1 GB */
1594 case 16ul << 30: /* 16 GB */
1596 case 256ul << 30: /* 256 GB */
1603 static int kvm_rma_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1606 struct kvm_rma_info
*ri
= vma
->vm_file
->private_data
;
1608 if (vmf
->pgoff
>= kvm_rma_pages
)
1609 return VM_FAULT_SIGBUS
;
1611 page
= pfn_to_page(ri
->base_pfn
+ vmf
->pgoff
);
1617 static const struct vm_operations_struct kvm_rma_vm_ops
= {
1618 .fault
= kvm_rma_fault
,
1621 static int kvm_rma_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1623 vma
->vm_flags
|= VM_DONTEXPAND
| VM_DONTDUMP
;
1624 vma
->vm_ops
= &kvm_rma_vm_ops
;
1628 static int kvm_rma_release(struct inode
*inode
, struct file
*filp
)
1630 struct kvm_rma_info
*ri
= filp
->private_data
;
1632 kvm_release_rma(ri
);
1636 static const struct file_operations kvm_rma_fops
= {
1637 .mmap
= kvm_rma_mmap
,
1638 .release
= kvm_rma_release
,
1641 long kvm_vm_ioctl_allocate_rma(struct kvm
*kvm
, struct kvm_allocate_rma
*ret
)
1644 struct kvm_rma_info
*ri
;
1646 * Only do this on PPC970 in HV mode
1648 if (!cpu_has_feature(CPU_FTR_HVMODE
) ||
1649 !cpu_has_feature(CPU_FTR_ARCH_201
))
1655 ri
= kvm_alloc_rma();
1659 fd
= anon_inode_getfd("kvm-rma", &kvm_rma_fops
, ri
, O_RDWR
| O_CLOEXEC
);
1661 kvm_release_rma(ri
);
1663 ret
->rma_size
= kvm_rma_pages
<< PAGE_SHIFT
;
1667 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size
**sps
,
1670 struct mmu_psize_def
*def
= &mmu_psize_defs
[linux_psize
];
1674 (*sps
)->page_shift
= def
->shift
;
1675 (*sps
)->slb_enc
= def
->sllp
;
1676 (*sps
)->enc
[0].page_shift
= def
->shift
;
1678 * Only return base page encoding. We don't want to return
1679 * all the supporting pte_enc, because our H_ENTER doesn't
1680 * support MPSS yet. Once they do, we can start passing all
1681 * support pte_enc here
1683 (*sps
)->enc
[0].pte_enc
= def
->penc
[linux_psize
];
1687 int kvm_vm_ioctl_get_smmu_info(struct kvm
*kvm
, struct kvm_ppc_smmu_info
*info
)
1689 struct kvm_ppc_one_seg_page_size
*sps
;
1691 info
->flags
= KVM_PPC_PAGE_SIZES_REAL
;
1692 if (mmu_has_feature(MMU_FTR_1T_SEGMENT
))
1693 info
->flags
|= KVM_PPC_1T_SEGMENTS
;
1694 info
->slb_size
= mmu_slb_size
;
1696 /* We only support these sizes for now, and no muti-size segments */
1697 sps
= &info
->sps
[0];
1698 kvmppc_add_seg_page_size(&sps
, MMU_PAGE_4K
);
1699 kvmppc_add_seg_page_size(&sps
, MMU_PAGE_64K
);
1700 kvmppc_add_seg_page_size(&sps
, MMU_PAGE_16M
);
1706 * Get (and clear) the dirty memory log for a memory slot.
1708 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
, struct kvm_dirty_log
*log
)
1710 struct kvm_memory_slot
*memslot
;
1714 mutex_lock(&kvm
->slots_lock
);
1717 if (log
->slot
>= KVM_USER_MEM_SLOTS
)
1720 memslot
= id_to_memslot(kvm
->memslots
, log
->slot
);
1722 if (!memslot
->dirty_bitmap
)
1725 n
= kvm_dirty_bitmap_bytes(memslot
);
1726 memset(memslot
->dirty_bitmap
, 0, n
);
1728 r
= kvmppc_hv_get_dirty_log(kvm
, memslot
, memslot
->dirty_bitmap
);
1733 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
1738 mutex_unlock(&kvm
->slots_lock
);
1742 static void unpin_slot(struct kvm_memory_slot
*memslot
)
1744 unsigned long *physp
;
1745 unsigned long j
, npages
, pfn
;
1748 physp
= memslot
->arch
.slot_phys
;
1749 npages
= memslot
->npages
;
1752 for (j
= 0; j
< npages
; j
++) {
1753 if (!(physp
[j
] & KVMPPC_GOT_PAGE
))
1755 pfn
= physp
[j
] >> PAGE_SHIFT
;
1756 page
= pfn_to_page(pfn
);
1762 void kvmppc_core_free_memslot(struct kvm_memory_slot
*free
,
1763 struct kvm_memory_slot
*dont
)
1765 if (!dont
|| free
->arch
.rmap
!= dont
->arch
.rmap
) {
1766 vfree(free
->arch
.rmap
);
1767 free
->arch
.rmap
= NULL
;
1769 if (!dont
|| free
->arch
.slot_phys
!= dont
->arch
.slot_phys
) {
1771 vfree(free
->arch
.slot_phys
);
1772 free
->arch
.slot_phys
= NULL
;
1776 int kvmppc_core_create_memslot(struct kvm_memory_slot
*slot
,
1777 unsigned long npages
)
1779 slot
->arch
.rmap
= vzalloc(npages
* sizeof(*slot
->arch
.rmap
));
1780 if (!slot
->arch
.rmap
)
1782 slot
->arch
.slot_phys
= NULL
;
1787 int kvmppc_core_prepare_memory_region(struct kvm
*kvm
,
1788 struct kvm_memory_slot
*memslot
,
1789 struct kvm_userspace_memory_region
*mem
)
1791 unsigned long *phys
;
1793 /* Allocate a slot_phys array if needed */
1794 phys
= memslot
->arch
.slot_phys
;
1795 if (!kvm
->arch
.using_mmu_notifiers
&& !phys
&& memslot
->npages
) {
1796 phys
= vzalloc(memslot
->npages
* sizeof(unsigned long));
1799 memslot
->arch
.slot_phys
= phys
;
1805 void kvmppc_core_commit_memory_region(struct kvm
*kvm
,
1806 struct kvm_userspace_memory_region
*mem
,
1807 const struct kvm_memory_slot
*old
)
1809 unsigned long npages
= mem
->memory_size
>> PAGE_SHIFT
;
1810 struct kvm_memory_slot
*memslot
;
1812 if (npages
&& old
->npages
) {
1814 * If modifying a memslot, reset all the rmap dirty bits.
1815 * If this is a new memslot, we don't need to do anything
1816 * since the rmap array starts out as all zeroes,
1817 * i.e. no pages are dirty.
1819 memslot
= id_to_memslot(kvm
->memslots
, mem
->slot
);
1820 kvmppc_hv_get_dirty_log(kvm
, memslot
, NULL
);
1825 * Update LPCR values in kvm->arch and in vcores.
1826 * Caller must hold kvm->lock.
1828 void kvmppc_update_lpcr(struct kvm
*kvm
, unsigned long lpcr
, unsigned long mask
)
1833 if ((kvm
->arch
.lpcr
& mask
) == lpcr
)
1836 kvm
->arch
.lpcr
= (kvm
->arch
.lpcr
& ~mask
) | lpcr
;
1838 for (i
= 0; i
< KVM_MAX_VCORES
; ++i
) {
1839 struct kvmppc_vcore
*vc
= kvm
->arch
.vcores
[i
];
1842 spin_lock(&vc
->lock
);
1843 vc
->lpcr
= (vc
->lpcr
& ~mask
) | lpcr
;
1844 spin_unlock(&vc
->lock
);
1845 if (++cores_done
>= kvm
->arch
.online_vcores
)
1850 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu
*vcpu
)
1853 struct kvm
*kvm
= vcpu
->kvm
;
1854 struct kvm_rma_info
*ri
= NULL
;
1856 struct kvm_memory_slot
*memslot
;
1857 struct vm_area_struct
*vma
;
1858 unsigned long lpcr
= 0, senc
;
1859 unsigned long lpcr_mask
= 0;
1860 unsigned long psize
, porder
;
1861 unsigned long rma_size
;
1863 unsigned long *physp
;
1864 unsigned long i
, npages
;
1867 mutex_lock(&kvm
->lock
);
1868 if (kvm
->arch
.rma_setup_done
)
1869 goto out
; /* another vcpu beat us to it */
1871 /* Allocate hashed page table (if not done already) and reset it */
1872 if (!kvm
->arch
.hpt_virt
) {
1873 err
= kvmppc_alloc_hpt(kvm
, NULL
);
1875 pr_err("KVM: Couldn't alloc HPT\n");
1880 /* Look up the memslot for guest physical address 0 */
1881 srcu_idx
= srcu_read_lock(&kvm
->srcu
);
1882 memslot
= gfn_to_memslot(kvm
, 0);
1884 /* We must have some memory at 0 by now */
1886 if (!memslot
|| (memslot
->flags
& KVM_MEMSLOT_INVALID
))
1889 /* Look up the VMA for the start of this memory slot */
1890 hva
= memslot
->userspace_addr
;
1891 down_read(¤t
->mm
->mmap_sem
);
1892 vma
= find_vma(current
->mm
, hva
);
1893 if (!vma
|| vma
->vm_start
> hva
|| (vma
->vm_flags
& VM_IO
))
1896 psize
= vma_kernel_pagesize(vma
);
1897 porder
= __ilog2(psize
);
1899 /* Is this one of our preallocated RMAs? */
1900 if (vma
->vm_file
&& vma
->vm_file
->f_op
== &kvm_rma_fops
&&
1901 hva
== vma
->vm_start
)
1902 ri
= vma
->vm_file
->private_data
;
1904 up_read(¤t
->mm
->mmap_sem
);
1907 /* On POWER7, use VRMA; on PPC970, give up */
1909 if (cpu_has_feature(CPU_FTR_ARCH_201
)) {
1910 pr_err("KVM: CPU requires an RMO\n");
1914 /* We can handle 4k, 64k or 16M pages in the VRMA */
1916 if (!(psize
== 0x1000 || psize
== 0x10000 ||
1917 psize
== 0x1000000))
1920 /* Update VRMASD field in the LPCR */
1921 senc
= slb_pgsize_encoding(psize
);
1922 kvm
->arch
.vrma_slb_v
= senc
| SLB_VSID_B_1T
|
1923 (VRMA_VSID
<< SLB_VSID_SHIFT_1T
);
1924 lpcr_mask
= LPCR_VRMASD
;
1925 /* the -4 is to account for senc values starting at 0x10 */
1926 lpcr
= senc
<< (LPCR_VRMASD_SH
- 4);
1928 /* Create HPTEs in the hash page table for the VRMA */
1929 kvmppc_map_vrma(vcpu
, memslot
, porder
);
1932 /* Set up to use an RMO region */
1933 rma_size
= kvm_rma_pages
;
1934 if (rma_size
> memslot
->npages
)
1935 rma_size
= memslot
->npages
;
1936 rma_size
<<= PAGE_SHIFT
;
1937 rmls
= lpcr_rmls(rma_size
);
1939 if ((long)rmls
< 0) {
1940 pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size
);
1943 atomic_inc(&ri
->use_count
);
1946 /* Update LPCR and RMOR */
1947 if (cpu_has_feature(CPU_FTR_ARCH_201
)) {
1948 /* PPC970; insert RMLS value (split field) in HID4 */
1949 lpcr_mask
= (1ul << HID4_RMLS0_SH
) |
1950 (3ul << HID4_RMLS2_SH
) | HID4_RMOR
;
1951 lpcr
= ((rmls
>> 2) << HID4_RMLS0_SH
) |
1952 ((rmls
& 3) << HID4_RMLS2_SH
);
1953 /* RMOR is also in HID4 */
1954 lpcr
|= ((ri
->base_pfn
>> (26 - PAGE_SHIFT
)) & 0xffff)
1958 lpcr_mask
= LPCR_VPM0
| LPCR_VRMA_L
| LPCR_RMLS
;
1959 lpcr
= rmls
<< LPCR_RMLS_SH
;
1960 kvm
->arch
.rmor
= ri
->base_pfn
<< PAGE_SHIFT
;
1962 pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n",
1963 ri
->base_pfn
<< PAGE_SHIFT
, rma_size
, lpcr
);
1965 /* Initialize phys addrs of pages in RMO */
1966 npages
= kvm_rma_pages
;
1967 porder
= __ilog2(npages
);
1968 physp
= memslot
->arch
.slot_phys
;
1970 if (npages
> memslot
->npages
)
1971 npages
= memslot
->npages
;
1972 spin_lock(&kvm
->arch
.slot_phys_lock
);
1973 for (i
= 0; i
< npages
; ++i
)
1974 physp
[i
] = ((ri
->base_pfn
+ i
) << PAGE_SHIFT
) +
1976 spin_unlock(&kvm
->arch
.slot_phys_lock
);
1980 kvmppc_update_lpcr(kvm
, lpcr
, lpcr_mask
);
1982 /* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */
1984 kvm
->arch
.rma_setup_done
= 1;
1987 srcu_read_unlock(&kvm
->srcu
, srcu_idx
);
1989 mutex_unlock(&kvm
->lock
);
1993 up_read(¤t
->mm
->mmap_sem
);
1997 int kvmppc_core_init_vm(struct kvm
*kvm
)
1999 unsigned long lpcr
, lpid
;
2001 /* Allocate the guest's logical partition ID */
2003 lpid
= kvmppc_alloc_lpid();
2006 kvm
->arch
.lpid
= lpid
;
2009 * Since we don't flush the TLB when tearing down a VM,
2010 * and this lpid might have previously been used,
2011 * make sure we flush on each core before running the new VM.
2013 cpumask_setall(&kvm
->arch
.need_tlb_flush
);
2015 INIT_LIST_HEAD(&kvm
->arch
.spapr_tce_tables
);
2016 INIT_LIST_HEAD(&kvm
->arch
.rtas_tokens
);
2018 kvm
->arch
.rma
= NULL
;
2020 kvm
->arch
.host_sdr1
= mfspr(SPRN_SDR1
);
2022 if (cpu_has_feature(CPU_FTR_ARCH_201
)) {
2023 /* PPC970; HID4 is effectively the LPCR */
2024 kvm
->arch
.host_lpid
= 0;
2025 kvm
->arch
.host_lpcr
= lpcr
= mfspr(SPRN_HID4
);
2026 lpcr
&= ~((3 << HID4_LPID1_SH
) | (0xful
<< HID4_LPID5_SH
));
2027 lpcr
|= ((lpid
>> 4) << HID4_LPID1_SH
) |
2028 ((lpid
& 0xf) << HID4_LPID5_SH
);
2030 /* POWER7; init LPCR for virtual RMA mode */
2031 kvm
->arch
.host_lpid
= mfspr(SPRN_LPID
);
2032 kvm
->arch
.host_lpcr
= lpcr
= mfspr(SPRN_LPCR
);
2033 lpcr
&= LPCR_PECE
| LPCR_LPES
;
2034 lpcr
|= (4UL << LPCR_DPFD_SH
) | LPCR_HDICE
|
2035 LPCR_VPM0
| LPCR_VPM1
;
2036 kvm
->arch
.vrma_slb_v
= SLB_VSID_B_1T
|
2037 (VRMA_VSID
<< SLB_VSID_SHIFT_1T
);
2039 kvm
->arch
.lpcr
= lpcr
;
2041 kvm
->arch
.using_mmu_notifiers
= !!cpu_has_feature(CPU_FTR_ARCH_206
);
2042 spin_lock_init(&kvm
->arch
.slot_phys_lock
);
2045 * Don't allow secondary CPU threads to come online
2046 * while any KVM VMs exist.
2048 inhibit_secondary_onlining();
2053 static void kvmppc_free_vcores(struct kvm
*kvm
)
2057 for (i
= 0; i
< KVM_MAX_VCORES
; ++i
)
2058 kfree(kvm
->arch
.vcores
[i
]);
2059 kvm
->arch
.online_vcores
= 0;
2062 void kvmppc_core_destroy_vm(struct kvm
*kvm
)
2064 uninhibit_secondary_onlining();
2066 kvmppc_free_vcores(kvm
);
2067 if (kvm
->arch
.rma
) {
2068 kvm_release_rma(kvm
->arch
.rma
);
2069 kvm
->arch
.rma
= NULL
;
2072 kvmppc_rtas_tokens_free(kvm
);
2074 kvmppc_free_hpt(kvm
);
2075 WARN_ON(!list_empty(&kvm
->arch
.spapr_tce_tables
));
2078 /* These are stubs for now */
2079 void kvmppc_mmu_pte_pflush(struct kvm_vcpu
*vcpu
, ulong pa_start
, ulong pa_end
)
2083 /* We don't need to emulate any privileged instructions or dcbz */
2084 int kvmppc_core_emulate_op(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
,
2085 unsigned int inst
, int *advance
)
2087 return EMULATE_FAIL
;
2090 int kvmppc_core_emulate_mtspr(struct kvm_vcpu
*vcpu
, int sprn
, ulong spr_val
)
2092 return EMULATE_FAIL
;
2095 int kvmppc_core_emulate_mfspr(struct kvm_vcpu
*vcpu
, int sprn
, ulong
*spr_val
)
2097 return EMULATE_FAIL
;
2100 static int kvmppc_book3s_hv_init(void)
2104 r
= kvm_init(NULL
, sizeof(struct kvm_vcpu
), 0, THIS_MODULE
);
2109 r
= kvmppc_mmu_hv_init();
2114 static void kvmppc_book3s_hv_exit(void)
2119 module_init(kvmppc_book3s_hv_init
);
2120 module_exit(kvmppc_book3s_hv_exit
);