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>
34 #include <linux/miscdevice.h>
37 #include <asm/cputable.h>
38 #include <asm/cacheflush.h>
39 #include <asm/tlbflush.h>
40 #include <asm/uaccess.h>
42 #include <asm/kvm_ppc.h>
43 #include <asm/kvm_book3s.h>
44 #include <asm/mmu_context.h>
45 #include <asm/lppaca.h>
46 #include <asm/processor.h>
47 #include <asm/cputhreads.h>
49 #include <asm/hvcall.h>
50 #include <asm/switch_to.h>
52 #include <linux/gfp.h>
53 #include <linux/vmalloc.h>
54 #include <linux/highmem.h>
55 #include <linux/hugetlb.h>
56 #include <linux/module.h>
60 /* #define EXIT_DEBUG */
61 /* #define EXIT_DEBUG_SIMPLE */
62 /* #define EXIT_DEBUG_INT */
64 /* Used to indicate that a guest page fault needs to be handled */
65 #define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1)
67 /* Used as a "null" value for timebase values */
68 #define TB_NIL (~(u64)0)
70 static DECLARE_BITMAP(default_enabled_hcalls
, MAX_HCALL_OPCODE
/4 + 1);
72 static void kvmppc_end_cede(struct kvm_vcpu
*vcpu
);
73 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu
*vcpu
);
75 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu
*vcpu
)
79 wait_queue_head_t
*wqp
;
81 wqp
= kvm_arch_vcpu_wq(vcpu
);
82 if (waitqueue_active(wqp
)) {
83 wake_up_interruptible(wqp
);
84 ++vcpu
->stat
.halt_wakeup
;
89 /* CPU points to the first thread of the core */
90 if (cpu
!= me
&& cpu
>= 0 && cpu
< nr_cpu_ids
) {
91 #ifdef CONFIG_PPC_ICP_NATIVE
92 int real_cpu
= cpu
+ vcpu
->arch
.ptid
;
93 if (paca
[real_cpu
].kvm_hstate
.xics_phys
)
94 xics_wake_cpu(real_cpu
);
98 smp_send_reschedule(cpu
);
104 * We use the vcpu_load/put functions to measure stolen time.
105 * Stolen time is counted as time when either the vcpu is able to
106 * run as part of a virtual core, but the task running the vcore
107 * is preempted or sleeping, or when the vcpu needs something done
108 * in the kernel by the task running the vcpu, but that task is
109 * preempted or sleeping. Those two things have to be counted
110 * separately, since one of the vcpu tasks will take on the job
111 * of running the core, and the other vcpu tasks in the vcore will
112 * sleep waiting for it to do that, but that sleep shouldn't count
115 * Hence we accumulate stolen time when the vcpu can run as part of
116 * a vcore using vc->stolen_tb, and the stolen time when the vcpu
117 * needs its task to do other things in the kernel (for example,
118 * service a page fault) in busy_stolen. We don't accumulate
119 * stolen time for a vcore when it is inactive, or for a vcpu
120 * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of
121 * a misnomer; it means that the vcpu task is not executing in
122 * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
123 * the kernel. We don't have any way of dividing up that time
124 * between time that the vcpu is genuinely stopped, time that
125 * the task is actively working on behalf of the vcpu, and time
126 * that the task is preempted, so we don't count any of it as
129 * Updates to busy_stolen are protected by arch.tbacct_lock;
130 * updates to vc->stolen_tb are protected by the arch.tbacct_lock
131 * of the vcpu that has taken responsibility for running the vcore
132 * (i.e. vc->runner). The stolen times are measured in units of
133 * timebase ticks. (Note that the != TB_NIL checks below are
134 * purely defensive; they should never fail.)
137 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu
*vcpu
, int cpu
)
139 struct kvmppc_vcore
*vc
= vcpu
->arch
.vcore
;
142 spin_lock_irqsave(&vcpu
->arch
.tbacct_lock
, flags
);
143 if (vc
->runner
== vcpu
&& vc
->vcore_state
!= VCORE_INACTIVE
&&
144 vc
->preempt_tb
!= TB_NIL
) {
145 vc
->stolen_tb
+= mftb() - vc
->preempt_tb
;
146 vc
->preempt_tb
= TB_NIL
;
148 if (vcpu
->arch
.state
== KVMPPC_VCPU_BUSY_IN_HOST
&&
149 vcpu
->arch
.busy_preempt
!= TB_NIL
) {
150 vcpu
->arch
.busy_stolen
+= mftb() - vcpu
->arch
.busy_preempt
;
151 vcpu
->arch
.busy_preempt
= TB_NIL
;
153 spin_unlock_irqrestore(&vcpu
->arch
.tbacct_lock
, flags
);
156 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu
*vcpu
)
158 struct kvmppc_vcore
*vc
= vcpu
->arch
.vcore
;
161 spin_lock_irqsave(&vcpu
->arch
.tbacct_lock
, flags
);
162 if (vc
->runner
== vcpu
&& vc
->vcore_state
!= VCORE_INACTIVE
)
163 vc
->preempt_tb
= mftb();
164 if (vcpu
->arch
.state
== KVMPPC_VCPU_BUSY_IN_HOST
)
165 vcpu
->arch
.busy_preempt
= mftb();
166 spin_unlock_irqrestore(&vcpu
->arch
.tbacct_lock
, flags
);
169 static void kvmppc_set_msr_hv(struct kvm_vcpu
*vcpu
, u64 msr
)
171 vcpu
->arch
.shregs
.msr
= msr
;
172 kvmppc_end_cede(vcpu
);
175 void kvmppc_set_pvr_hv(struct kvm_vcpu
*vcpu
, u32 pvr
)
177 vcpu
->arch
.pvr
= pvr
;
180 int kvmppc_set_arch_compat(struct kvm_vcpu
*vcpu
, u32 arch_compat
)
182 unsigned long pcr
= 0;
183 struct kvmppc_vcore
*vc
= vcpu
->arch
.vcore
;
186 if (!cpu_has_feature(CPU_FTR_ARCH_206
))
187 return -EINVAL
; /* 970 has no compat mode support */
189 switch (arch_compat
) {
192 * If an arch bit is set in PCR, all the defined
193 * higher-order arch bits also have to be set.
195 pcr
= PCR_ARCH_206
| PCR_ARCH_205
;
207 if (!cpu_has_feature(CPU_FTR_ARCH_207S
)) {
208 /* POWER7 can't emulate POWER8 */
209 if (!(pcr
& PCR_ARCH_206
))
211 pcr
&= ~PCR_ARCH_206
;
215 spin_lock(&vc
->lock
);
216 vc
->arch_compat
= arch_compat
;
218 spin_unlock(&vc
->lock
);
223 void kvmppc_dump_regs(struct kvm_vcpu
*vcpu
)
227 pr_err("vcpu %p (%d):\n", vcpu
, vcpu
->vcpu_id
);
228 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
229 vcpu
->arch
.pc
, vcpu
->arch
.shregs
.msr
, vcpu
->arch
.trap
);
230 for (r
= 0; r
< 16; ++r
)
231 pr_err("r%2d = %.16lx r%d = %.16lx\n",
232 r
, kvmppc_get_gpr(vcpu
, r
),
233 r
+16, kvmppc_get_gpr(vcpu
, r
+16));
234 pr_err("ctr = %.16lx lr = %.16lx\n",
235 vcpu
->arch
.ctr
, vcpu
->arch
.lr
);
236 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
237 vcpu
->arch
.shregs
.srr0
, vcpu
->arch
.shregs
.srr1
);
238 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
239 vcpu
->arch
.shregs
.sprg0
, vcpu
->arch
.shregs
.sprg1
);
240 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
241 vcpu
->arch
.shregs
.sprg2
, vcpu
->arch
.shregs
.sprg3
);
242 pr_err("cr = %.8x xer = %.16lx dsisr = %.8x\n",
243 vcpu
->arch
.cr
, vcpu
->arch
.xer
, vcpu
->arch
.shregs
.dsisr
);
244 pr_err("dar = %.16llx\n", vcpu
->arch
.shregs
.dar
);
245 pr_err("fault dar = %.16lx dsisr = %.8x\n",
246 vcpu
->arch
.fault_dar
, vcpu
->arch
.fault_dsisr
);
247 pr_err("SLB (%d entries):\n", vcpu
->arch
.slb_max
);
248 for (r
= 0; r
< vcpu
->arch
.slb_max
; ++r
)
249 pr_err(" ESID = %.16llx VSID = %.16llx\n",
250 vcpu
->arch
.slb
[r
].orige
, vcpu
->arch
.slb
[r
].origv
);
251 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
252 vcpu
->arch
.vcore
->lpcr
, vcpu
->kvm
->arch
.sdr1
,
253 vcpu
->arch
.last_inst
);
256 struct kvm_vcpu
*kvmppc_find_vcpu(struct kvm
*kvm
, int id
)
259 struct kvm_vcpu
*v
, *ret
= NULL
;
261 mutex_lock(&kvm
->lock
);
262 kvm_for_each_vcpu(r
, v
, kvm
) {
263 if (v
->vcpu_id
== id
) {
268 mutex_unlock(&kvm
->lock
);
272 static void init_vpa(struct kvm_vcpu
*vcpu
, struct lppaca
*vpa
)
274 vpa
->__old_status
|= LPPACA_OLD_SHARED_PROC
;
275 vpa
->yield_count
= 1;
278 static int set_vpa(struct kvm_vcpu
*vcpu
, struct kvmppc_vpa
*v
,
279 unsigned long addr
, unsigned long len
)
281 /* check address is cacheline aligned */
282 if (addr
& (L1_CACHE_BYTES
- 1))
284 spin_lock(&vcpu
->arch
.vpa_update_lock
);
285 if (v
->next_gpa
!= addr
|| v
->len
!= len
) {
287 v
->len
= addr
? len
: 0;
288 v
->update_pending
= 1;
290 spin_unlock(&vcpu
->arch
.vpa_update_lock
);
294 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
303 static int vpa_is_registered(struct kvmppc_vpa
*vpap
)
305 if (vpap
->update_pending
)
306 return vpap
->next_gpa
!= 0;
307 return vpap
->pinned_addr
!= NULL
;
310 static unsigned long do_h_register_vpa(struct kvm_vcpu
*vcpu
,
312 unsigned long vcpuid
, unsigned long vpa
)
314 struct kvm
*kvm
= vcpu
->kvm
;
315 unsigned long len
, nb
;
317 struct kvm_vcpu
*tvcpu
;
320 struct kvmppc_vpa
*vpap
;
322 tvcpu
= kvmppc_find_vcpu(kvm
, vcpuid
);
326 subfunc
= (flags
>> H_VPA_FUNC_SHIFT
) & H_VPA_FUNC_MASK
;
327 if (subfunc
== H_VPA_REG_VPA
|| subfunc
== H_VPA_REG_DTL
||
328 subfunc
== H_VPA_REG_SLB
) {
329 /* Registering new area - address must be cache-line aligned */
330 if ((vpa
& (L1_CACHE_BYTES
- 1)) || !vpa
)
333 /* convert logical addr to kernel addr and read length */
334 va
= kvmppc_pin_guest_page(kvm
, vpa
, &nb
);
337 if (subfunc
== H_VPA_REG_VPA
)
338 len
= ((struct reg_vpa
*)va
)->length
.hword
;
340 len
= ((struct reg_vpa
*)va
)->length
.word
;
341 kvmppc_unpin_guest_page(kvm
, va
, vpa
, false);
344 if (len
> nb
|| len
< sizeof(struct reg_vpa
))
353 spin_lock(&tvcpu
->arch
.vpa_update_lock
);
356 case H_VPA_REG_VPA
: /* register VPA */
357 if (len
< sizeof(struct lppaca
))
359 vpap
= &tvcpu
->arch
.vpa
;
363 case H_VPA_REG_DTL
: /* register DTL */
364 if (len
< sizeof(struct dtl_entry
))
366 len
-= len
% sizeof(struct dtl_entry
);
368 /* Check that they have previously registered a VPA */
370 if (!vpa_is_registered(&tvcpu
->arch
.vpa
))
373 vpap
= &tvcpu
->arch
.dtl
;
377 case H_VPA_REG_SLB
: /* register SLB shadow buffer */
378 /* Check that they have previously registered a VPA */
380 if (!vpa_is_registered(&tvcpu
->arch
.vpa
))
383 vpap
= &tvcpu
->arch
.slb_shadow
;
387 case H_VPA_DEREG_VPA
: /* deregister VPA */
388 /* Check they don't still have a DTL or SLB buf registered */
390 if (vpa_is_registered(&tvcpu
->arch
.dtl
) ||
391 vpa_is_registered(&tvcpu
->arch
.slb_shadow
))
394 vpap
= &tvcpu
->arch
.vpa
;
398 case H_VPA_DEREG_DTL
: /* deregister DTL */
399 vpap
= &tvcpu
->arch
.dtl
;
403 case H_VPA_DEREG_SLB
: /* deregister SLB shadow buffer */
404 vpap
= &tvcpu
->arch
.slb_shadow
;
410 vpap
->next_gpa
= vpa
;
412 vpap
->update_pending
= 1;
415 spin_unlock(&tvcpu
->arch
.vpa_update_lock
);
420 static void kvmppc_update_vpa(struct kvm_vcpu
*vcpu
, struct kvmppc_vpa
*vpap
)
422 struct kvm
*kvm
= vcpu
->kvm
;
428 * We need to pin the page pointed to by vpap->next_gpa,
429 * but we can't call kvmppc_pin_guest_page under the lock
430 * as it does get_user_pages() and down_read(). So we
431 * have to drop the lock, pin the page, then get the lock
432 * again and check that a new area didn't get registered
436 gpa
= vpap
->next_gpa
;
437 spin_unlock(&vcpu
->arch
.vpa_update_lock
);
441 va
= kvmppc_pin_guest_page(kvm
, gpa
, &nb
);
442 spin_lock(&vcpu
->arch
.vpa_update_lock
);
443 if (gpa
== vpap
->next_gpa
)
445 /* sigh... unpin that one and try again */
447 kvmppc_unpin_guest_page(kvm
, va
, gpa
, false);
450 vpap
->update_pending
= 0;
451 if (va
&& nb
< vpap
->len
) {
453 * If it's now too short, it must be that userspace
454 * has changed the mappings underlying guest memory,
455 * so unregister the region.
457 kvmppc_unpin_guest_page(kvm
, va
, gpa
, false);
460 if (vpap
->pinned_addr
)
461 kvmppc_unpin_guest_page(kvm
, vpap
->pinned_addr
, vpap
->gpa
,
464 vpap
->pinned_addr
= va
;
467 vpap
->pinned_end
= va
+ vpap
->len
;
470 static void kvmppc_update_vpas(struct kvm_vcpu
*vcpu
)
472 if (!(vcpu
->arch
.vpa
.update_pending
||
473 vcpu
->arch
.slb_shadow
.update_pending
||
474 vcpu
->arch
.dtl
.update_pending
))
477 spin_lock(&vcpu
->arch
.vpa_update_lock
);
478 if (vcpu
->arch
.vpa
.update_pending
) {
479 kvmppc_update_vpa(vcpu
, &vcpu
->arch
.vpa
);
480 if (vcpu
->arch
.vpa
.pinned_addr
)
481 init_vpa(vcpu
, vcpu
->arch
.vpa
.pinned_addr
);
483 if (vcpu
->arch
.dtl
.update_pending
) {
484 kvmppc_update_vpa(vcpu
, &vcpu
->arch
.dtl
);
485 vcpu
->arch
.dtl_ptr
= vcpu
->arch
.dtl
.pinned_addr
;
486 vcpu
->arch
.dtl_index
= 0;
488 if (vcpu
->arch
.slb_shadow
.update_pending
)
489 kvmppc_update_vpa(vcpu
, &vcpu
->arch
.slb_shadow
);
490 spin_unlock(&vcpu
->arch
.vpa_update_lock
);
494 * Return the accumulated stolen time for the vcore up until `now'.
495 * The caller should hold the vcore lock.
497 static u64
vcore_stolen_time(struct kvmppc_vcore
*vc
, u64 now
)
502 * If we are the task running the vcore, then since we hold
503 * the vcore lock, we can't be preempted, so stolen_tb/preempt_tb
504 * can't be updated, so we don't need the tbacct_lock.
505 * If the vcore is inactive, it can't become active (since we
506 * hold the vcore lock), so the vcpu load/put functions won't
507 * update stolen_tb/preempt_tb, and we don't need tbacct_lock.
509 if (vc
->vcore_state
!= VCORE_INACTIVE
&&
510 vc
->runner
->arch
.run_task
!= current
) {
511 spin_lock_irq(&vc
->runner
->arch
.tbacct_lock
);
513 if (vc
->preempt_tb
!= TB_NIL
)
514 p
+= now
- vc
->preempt_tb
;
515 spin_unlock_irq(&vc
->runner
->arch
.tbacct_lock
);
522 static void kvmppc_create_dtl_entry(struct kvm_vcpu
*vcpu
,
523 struct kvmppc_vcore
*vc
)
525 struct dtl_entry
*dt
;
527 unsigned long stolen
;
528 unsigned long core_stolen
;
531 dt
= vcpu
->arch
.dtl_ptr
;
532 vpa
= vcpu
->arch
.vpa
.pinned_addr
;
534 core_stolen
= vcore_stolen_time(vc
, now
);
535 stolen
= core_stolen
- vcpu
->arch
.stolen_logged
;
536 vcpu
->arch
.stolen_logged
= core_stolen
;
537 spin_lock_irq(&vcpu
->arch
.tbacct_lock
);
538 stolen
+= vcpu
->arch
.busy_stolen
;
539 vcpu
->arch
.busy_stolen
= 0;
540 spin_unlock_irq(&vcpu
->arch
.tbacct_lock
);
543 memset(dt
, 0, sizeof(struct dtl_entry
));
544 dt
->dispatch_reason
= 7;
545 dt
->processor_id
= vc
->pcpu
+ vcpu
->arch
.ptid
;
546 dt
->timebase
= now
+ vc
->tb_offset
;
547 dt
->enqueue_to_dispatch_time
= stolen
;
548 dt
->srr0
= kvmppc_get_pc(vcpu
);
549 dt
->srr1
= vcpu
->arch
.shregs
.msr
;
551 if (dt
== vcpu
->arch
.dtl
.pinned_end
)
552 dt
= vcpu
->arch
.dtl
.pinned_addr
;
553 vcpu
->arch
.dtl_ptr
= dt
;
554 /* order writing *dt vs. writing vpa->dtl_idx */
556 vpa
->dtl_idx
= ++vcpu
->arch
.dtl_index
;
557 vcpu
->arch
.dtl
.dirty
= true;
560 static bool kvmppc_power8_compatible(struct kvm_vcpu
*vcpu
)
562 if (vcpu
->arch
.vcore
->arch_compat
>= PVR_ARCH_207
)
564 if ((!vcpu
->arch
.vcore
->arch_compat
) &&
565 cpu_has_feature(CPU_FTR_ARCH_207S
))
570 static int kvmppc_h_set_mode(struct kvm_vcpu
*vcpu
, unsigned long mflags
,
571 unsigned long resource
, unsigned long value1
,
572 unsigned long value2
)
575 case H_SET_MODE_RESOURCE_SET_CIABR
:
576 if (!kvmppc_power8_compatible(vcpu
))
581 return H_UNSUPPORTED_FLAG_START
;
582 /* Guests can't breakpoint the hypervisor */
583 if ((value1
& CIABR_PRIV
) == CIABR_PRIV_HYPER
)
585 vcpu
->arch
.ciabr
= value1
;
587 case H_SET_MODE_RESOURCE_SET_DAWR
:
588 if (!kvmppc_power8_compatible(vcpu
))
591 return H_UNSUPPORTED_FLAG_START
;
592 if (value2
& DABRX_HYP
)
594 vcpu
->arch
.dawr
= value1
;
595 vcpu
->arch
.dawrx
= value2
;
602 int kvmppc_pseries_do_hcall(struct kvm_vcpu
*vcpu
)
604 unsigned long req
= kvmppc_get_gpr(vcpu
, 3);
605 unsigned long target
, ret
= H_SUCCESS
;
606 struct kvm_vcpu
*tvcpu
;
609 if (req
<= MAX_HCALL_OPCODE
&&
610 !test_bit(req
/4, vcpu
->kvm
->arch
.enabled_hcalls
))
615 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
616 ret
= kvmppc_virtmode_h_enter(vcpu
, kvmppc_get_gpr(vcpu
, 4),
617 kvmppc_get_gpr(vcpu
, 5),
618 kvmppc_get_gpr(vcpu
, 6),
619 kvmppc_get_gpr(vcpu
, 7));
620 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
625 target
= kvmppc_get_gpr(vcpu
, 4);
626 tvcpu
= kvmppc_find_vcpu(vcpu
->kvm
, target
);
631 tvcpu
->arch
.prodded
= 1;
633 if (vcpu
->arch
.ceded
) {
634 if (waitqueue_active(&vcpu
->wq
)) {
635 wake_up_interruptible(&vcpu
->wq
);
636 vcpu
->stat
.halt_wakeup
++;
641 target
= kvmppc_get_gpr(vcpu
, 4);
644 tvcpu
= kvmppc_find_vcpu(vcpu
->kvm
, target
);
649 kvm_vcpu_yield_to(tvcpu
);
652 ret
= do_h_register_vpa(vcpu
, kvmppc_get_gpr(vcpu
, 4),
653 kvmppc_get_gpr(vcpu
, 5),
654 kvmppc_get_gpr(vcpu
, 6));
657 if (list_empty(&vcpu
->kvm
->arch
.rtas_tokens
))
660 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
661 rc
= kvmppc_rtas_hcall(vcpu
);
662 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
669 /* Send the error out to userspace via KVM_RUN */
672 ret
= kvmppc_h_set_mode(vcpu
, kvmppc_get_gpr(vcpu
, 4),
673 kvmppc_get_gpr(vcpu
, 5),
674 kvmppc_get_gpr(vcpu
, 6),
675 kvmppc_get_gpr(vcpu
, 7));
676 if (ret
== H_TOO_HARD
)
685 if (kvmppc_xics_enabled(vcpu
)) {
686 ret
= kvmppc_xics_hcall(vcpu
, req
);
692 kvmppc_set_gpr(vcpu
, 3, ret
);
693 vcpu
->arch
.hcall_needed
= 0;
697 static int kvmppc_hcall_impl_hv(unsigned long cmd
)
705 #ifdef CONFIG_KVM_XICS
716 /* See if it's in the real-mode table */
717 return kvmppc_hcall_impl_hv_realmode(cmd
);
720 static int kvmppc_handle_exit_hv(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
,
721 struct task_struct
*tsk
)
725 vcpu
->stat
.sum_exits
++;
727 run
->exit_reason
= KVM_EXIT_UNKNOWN
;
728 run
->ready_for_interrupt_injection
= 1;
729 switch (vcpu
->arch
.trap
) {
730 /* We're good on these - the host merely wanted to get our attention */
731 case BOOK3S_INTERRUPT_HV_DECREMENTER
:
732 vcpu
->stat
.dec_exits
++;
735 case BOOK3S_INTERRUPT_EXTERNAL
:
736 case BOOK3S_INTERRUPT_H_DOORBELL
:
737 vcpu
->stat
.ext_intr_exits
++;
740 case BOOK3S_INTERRUPT_PERFMON
:
743 case BOOK3S_INTERRUPT_MACHINE_CHECK
:
745 * Deliver a machine check interrupt to the guest.
746 * We have to do this, even if the host has handled the
747 * machine check, because machine checks use SRR0/1 and
748 * the interrupt might have trashed guest state in them.
750 kvmppc_book3s_queue_irqprio(vcpu
,
751 BOOK3S_INTERRUPT_MACHINE_CHECK
);
754 case BOOK3S_INTERRUPT_PROGRAM
:
758 * Normally program interrupts are delivered directly
759 * to the guest by the hardware, but we can get here
760 * as a result of a hypervisor emulation interrupt
761 * (e40) getting turned into a 700 by BML RTAS.
763 flags
= vcpu
->arch
.shregs
.msr
& 0x1f0000ull
;
764 kvmppc_core_queue_program(vcpu
, flags
);
768 case BOOK3S_INTERRUPT_SYSCALL
:
770 /* hcall - punt to userspace */
773 /* hypercall with MSR_PR has already been handled in rmode,
774 * and never reaches here.
777 run
->papr_hcall
.nr
= kvmppc_get_gpr(vcpu
, 3);
778 for (i
= 0; i
< 9; ++i
)
779 run
->papr_hcall
.args
[i
] = kvmppc_get_gpr(vcpu
, 4 + i
);
780 run
->exit_reason
= KVM_EXIT_PAPR_HCALL
;
781 vcpu
->arch
.hcall_needed
= 1;
786 * We get these next two if the guest accesses a page which it thinks
787 * it has mapped but which is not actually present, either because
788 * it is for an emulated I/O device or because the corresonding
789 * host page has been paged out. Any other HDSI/HISI interrupts
790 * have been handled already.
792 case BOOK3S_INTERRUPT_H_DATA_STORAGE
:
793 r
= RESUME_PAGE_FAULT
;
795 case BOOK3S_INTERRUPT_H_INST_STORAGE
:
796 vcpu
->arch
.fault_dar
= kvmppc_get_pc(vcpu
);
797 vcpu
->arch
.fault_dsisr
= 0;
798 r
= RESUME_PAGE_FAULT
;
801 * This occurs if the guest executes an illegal instruction.
802 * We just generate a program interrupt to the guest, since
803 * we don't emulate any guest instructions at this stage.
805 case BOOK3S_INTERRUPT_H_EMUL_ASSIST
:
806 kvmppc_core_queue_program(vcpu
, SRR1_PROGILL
);
810 * This occurs if the guest (kernel or userspace), does something that
811 * is prohibited by HFSCR. We just generate a program interrupt to
814 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL
:
815 kvmppc_core_queue_program(vcpu
, SRR1_PROGILL
);
819 kvmppc_dump_regs(vcpu
);
820 printk(KERN_EMERG
"trap=0x%x | pc=0x%lx | msr=0x%llx\n",
821 vcpu
->arch
.trap
, kvmppc_get_pc(vcpu
),
822 vcpu
->arch
.shregs
.msr
);
823 run
->hw
.hardware_exit_reason
= vcpu
->arch
.trap
;
831 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu
*vcpu
,
832 struct kvm_sregs
*sregs
)
836 memset(sregs
, 0, sizeof(struct kvm_sregs
));
837 sregs
->pvr
= vcpu
->arch
.pvr
;
838 for (i
= 0; i
< vcpu
->arch
.slb_max
; i
++) {
839 sregs
->u
.s
.ppc64
.slb
[i
].slbe
= vcpu
->arch
.slb
[i
].orige
;
840 sregs
->u
.s
.ppc64
.slb
[i
].slbv
= vcpu
->arch
.slb
[i
].origv
;
846 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu
*vcpu
,
847 struct kvm_sregs
*sregs
)
851 kvmppc_set_pvr_hv(vcpu
, sregs
->pvr
);
854 for (i
= 0; i
< vcpu
->arch
.slb_nr
; i
++) {
855 if (sregs
->u
.s
.ppc64
.slb
[i
].slbe
& SLB_ESID_V
) {
856 vcpu
->arch
.slb
[j
].orige
= sregs
->u
.s
.ppc64
.slb
[i
].slbe
;
857 vcpu
->arch
.slb
[j
].origv
= sregs
->u
.s
.ppc64
.slb
[i
].slbv
;
861 vcpu
->arch
.slb_max
= j
;
866 static void kvmppc_set_lpcr(struct kvm_vcpu
*vcpu
, u64 new_lpcr
)
868 struct kvmppc_vcore
*vc
= vcpu
->arch
.vcore
;
871 spin_lock(&vc
->lock
);
873 * If ILE (interrupt little-endian) has changed, update the
874 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
876 if ((new_lpcr
& LPCR_ILE
) != (vc
->lpcr
& LPCR_ILE
)) {
877 struct kvm
*kvm
= vcpu
->kvm
;
878 struct kvm_vcpu
*vcpu
;
881 mutex_lock(&kvm
->lock
);
882 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
883 if (vcpu
->arch
.vcore
!= vc
)
885 if (new_lpcr
& LPCR_ILE
)
886 vcpu
->arch
.intr_msr
|= MSR_LE
;
888 vcpu
->arch
.intr_msr
&= ~MSR_LE
;
890 mutex_unlock(&kvm
->lock
);
894 * Userspace can only modify DPFD (default prefetch depth),
895 * ILE (interrupt little-endian) and TC (translation control).
896 * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
898 mask
= LPCR_DPFD
| LPCR_ILE
| LPCR_TC
;
899 if (cpu_has_feature(CPU_FTR_ARCH_207S
))
901 vc
->lpcr
= (vc
->lpcr
& ~mask
) | (new_lpcr
& mask
);
902 spin_unlock(&vc
->lock
);
905 static int kvmppc_get_one_reg_hv(struct kvm_vcpu
*vcpu
, u64 id
,
906 union kvmppc_one_reg
*val
)
912 case KVM_REG_PPC_HIOR
:
913 *val
= get_reg_val(id
, 0);
915 case KVM_REG_PPC_DABR
:
916 *val
= get_reg_val(id
, vcpu
->arch
.dabr
);
918 case KVM_REG_PPC_DABRX
:
919 *val
= get_reg_val(id
, vcpu
->arch
.dabrx
);
921 case KVM_REG_PPC_DSCR
:
922 *val
= get_reg_val(id
, vcpu
->arch
.dscr
);
924 case KVM_REG_PPC_PURR
:
925 *val
= get_reg_val(id
, vcpu
->arch
.purr
);
927 case KVM_REG_PPC_SPURR
:
928 *val
= get_reg_val(id
, vcpu
->arch
.spurr
);
930 case KVM_REG_PPC_AMR
:
931 *val
= get_reg_val(id
, vcpu
->arch
.amr
);
933 case KVM_REG_PPC_UAMOR
:
934 *val
= get_reg_val(id
, vcpu
->arch
.uamor
);
936 case KVM_REG_PPC_MMCR0
... KVM_REG_PPC_MMCRS
:
937 i
= id
- KVM_REG_PPC_MMCR0
;
938 *val
= get_reg_val(id
, vcpu
->arch
.mmcr
[i
]);
940 case KVM_REG_PPC_PMC1
... KVM_REG_PPC_PMC8
:
941 i
= id
- KVM_REG_PPC_PMC1
;
942 *val
= get_reg_val(id
, vcpu
->arch
.pmc
[i
]);
944 case KVM_REG_PPC_SPMC1
... KVM_REG_PPC_SPMC2
:
945 i
= id
- KVM_REG_PPC_SPMC1
;
946 *val
= get_reg_val(id
, vcpu
->arch
.spmc
[i
]);
948 case KVM_REG_PPC_SIAR
:
949 *val
= get_reg_val(id
, vcpu
->arch
.siar
);
951 case KVM_REG_PPC_SDAR
:
952 *val
= get_reg_val(id
, vcpu
->arch
.sdar
);
954 case KVM_REG_PPC_SIER
:
955 *val
= get_reg_val(id
, vcpu
->arch
.sier
);
957 case KVM_REG_PPC_IAMR
:
958 *val
= get_reg_val(id
, vcpu
->arch
.iamr
);
960 case KVM_REG_PPC_PSPB
:
961 *val
= get_reg_val(id
, vcpu
->arch
.pspb
);
963 case KVM_REG_PPC_DPDES
:
964 *val
= get_reg_val(id
, vcpu
->arch
.vcore
->dpdes
);
966 case KVM_REG_PPC_DAWR
:
967 *val
= get_reg_val(id
, vcpu
->arch
.dawr
);
969 case KVM_REG_PPC_DAWRX
:
970 *val
= get_reg_val(id
, vcpu
->arch
.dawrx
);
972 case KVM_REG_PPC_CIABR
:
973 *val
= get_reg_val(id
, vcpu
->arch
.ciabr
);
975 case KVM_REG_PPC_CSIGR
:
976 *val
= get_reg_val(id
, vcpu
->arch
.csigr
);
978 case KVM_REG_PPC_TACR
:
979 *val
= get_reg_val(id
, vcpu
->arch
.tacr
);
981 case KVM_REG_PPC_TCSCR
:
982 *val
= get_reg_val(id
, vcpu
->arch
.tcscr
);
984 case KVM_REG_PPC_PID
:
985 *val
= get_reg_val(id
, vcpu
->arch
.pid
);
987 case KVM_REG_PPC_ACOP
:
988 *val
= get_reg_val(id
, vcpu
->arch
.acop
);
990 case KVM_REG_PPC_WORT
:
991 *val
= get_reg_val(id
, vcpu
->arch
.wort
);
993 case KVM_REG_PPC_VPA_ADDR
:
994 spin_lock(&vcpu
->arch
.vpa_update_lock
);
995 *val
= get_reg_val(id
, vcpu
->arch
.vpa
.next_gpa
);
996 spin_unlock(&vcpu
->arch
.vpa_update_lock
);
998 case KVM_REG_PPC_VPA_SLB
:
999 spin_lock(&vcpu
->arch
.vpa_update_lock
);
1000 val
->vpaval
.addr
= vcpu
->arch
.slb_shadow
.next_gpa
;
1001 val
->vpaval
.length
= vcpu
->arch
.slb_shadow
.len
;
1002 spin_unlock(&vcpu
->arch
.vpa_update_lock
);
1004 case KVM_REG_PPC_VPA_DTL
:
1005 spin_lock(&vcpu
->arch
.vpa_update_lock
);
1006 val
->vpaval
.addr
= vcpu
->arch
.dtl
.next_gpa
;
1007 val
->vpaval
.length
= vcpu
->arch
.dtl
.len
;
1008 spin_unlock(&vcpu
->arch
.vpa_update_lock
);
1010 case KVM_REG_PPC_TB_OFFSET
:
1011 *val
= get_reg_val(id
, vcpu
->arch
.vcore
->tb_offset
);
1013 case KVM_REG_PPC_LPCR
:
1014 *val
= get_reg_val(id
, vcpu
->arch
.vcore
->lpcr
);
1016 case KVM_REG_PPC_PPR
:
1017 *val
= get_reg_val(id
, vcpu
->arch
.ppr
);
1019 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1020 case KVM_REG_PPC_TFHAR
:
1021 *val
= get_reg_val(id
, vcpu
->arch
.tfhar
);
1023 case KVM_REG_PPC_TFIAR
:
1024 *val
= get_reg_val(id
, vcpu
->arch
.tfiar
);
1026 case KVM_REG_PPC_TEXASR
:
1027 *val
= get_reg_val(id
, vcpu
->arch
.texasr
);
1029 case KVM_REG_PPC_TM_GPR0
... KVM_REG_PPC_TM_GPR31
:
1030 i
= id
- KVM_REG_PPC_TM_GPR0
;
1031 *val
= get_reg_val(id
, vcpu
->arch
.gpr_tm
[i
]);
1033 case KVM_REG_PPC_TM_VSR0
... KVM_REG_PPC_TM_VSR63
:
1036 i
= id
- KVM_REG_PPC_TM_VSR0
;
1038 for (j
= 0; j
< TS_FPRWIDTH
; j
++)
1039 val
->vsxval
[j
] = vcpu
->arch
.fp_tm
.fpr
[i
][j
];
1041 if (cpu_has_feature(CPU_FTR_ALTIVEC
))
1042 val
->vval
= vcpu
->arch
.vr_tm
.vr
[i
-32];
1048 case KVM_REG_PPC_TM_CR
:
1049 *val
= get_reg_val(id
, vcpu
->arch
.cr_tm
);
1051 case KVM_REG_PPC_TM_LR
:
1052 *val
= get_reg_val(id
, vcpu
->arch
.lr_tm
);
1054 case KVM_REG_PPC_TM_CTR
:
1055 *val
= get_reg_val(id
, vcpu
->arch
.ctr_tm
);
1057 case KVM_REG_PPC_TM_FPSCR
:
1058 *val
= get_reg_val(id
, vcpu
->arch
.fp_tm
.fpscr
);
1060 case KVM_REG_PPC_TM_AMR
:
1061 *val
= get_reg_val(id
, vcpu
->arch
.amr_tm
);
1063 case KVM_REG_PPC_TM_PPR
:
1064 *val
= get_reg_val(id
, vcpu
->arch
.ppr_tm
);
1066 case KVM_REG_PPC_TM_VRSAVE
:
1067 *val
= get_reg_val(id
, vcpu
->arch
.vrsave_tm
);
1069 case KVM_REG_PPC_TM_VSCR
:
1070 if (cpu_has_feature(CPU_FTR_ALTIVEC
))
1071 *val
= get_reg_val(id
, vcpu
->arch
.vr_tm
.vscr
.u
[3]);
1075 case KVM_REG_PPC_TM_DSCR
:
1076 *val
= get_reg_val(id
, vcpu
->arch
.dscr_tm
);
1078 case KVM_REG_PPC_TM_TAR
:
1079 *val
= get_reg_val(id
, vcpu
->arch
.tar_tm
);
1082 case KVM_REG_PPC_ARCH_COMPAT
:
1083 *val
= get_reg_val(id
, vcpu
->arch
.vcore
->arch_compat
);
1093 static int kvmppc_set_one_reg_hv(struct kvm_vcpu
*vcpu
, u64 id
,
1094 union kvmppc_one_reg
*val
)
1098 unsigned long addr
, len
;
1101 case KVM_REG_PPC_HIOR
:
1102 /* Only allow this to be set to zero */
1103 if (set_reg_val(id
, *val
))
1106 case KVM_REG_PPC_DABR
:
1107 vcpu
->arch
.dabr
= set_reg_val(id
, *val
);
1109 case KVM_REG_PPC_DABRX
:
1110 vcpu
->arch
.dabrx
= set_reg_val(id
, *val
) & ~DABRX_HYP
;
1112 case KVM_REG_PPC_DSCR
:
1113 vcpu
->arch
.dscr
= set_reg_val(id
, *val
);
1115 case KVM_REG_PPC_PURR
:
1116 vcpu
->arch
.purr
= set_reg_val(id
, *val
);
1118 case KVM_REG_PPC_SPURR
:
1119 vcpu
->arch
.spurr
= set_reg_val(id
, *val
);
1121 case KVM_REG_PPC_AMR
:
1122 vcpu
->arch
.amr
= set_reg_val(id
, *val
);
1124 case KVM_REG_PPC_UAMOR
:
1125 vcpu
->arch
.uamor
= set_reg_val(id
, *val
);
1127 case KVM_REG_PPC_MMCR0
... KVM_REG_PPC_MMCRS
:
1128 i
= id
- KVM_REG_PPC_MMCR0
;
1129 vcpu
->arch
.mmcr
[i
] = set_reg_val(id
, *val
);
1131 case KVM_REG_PPC_PMC1
... KVM_REG_PPC_PMC8
:
1132 i
= id
- KVM_REG_PPC_PMC1
;
1133 vcpu
->arch
.pmc
[i
] = set_reg_val(id
, *val
);
1135 case KVM_REG_PPC_SPMC1
... KVM_REG_PPC_SPMC2
:
1136 i
= id
- KVM_REG_PPC_SPMC1
;
1137 vcpu
->arch
.spmc
[i
] = set_reg_val(id
, *val
);
1139 case KVM_REG_PPC_SIAR
:
1140 vcpu
->arch
.siar
= set_reg_val(id
, *val
);
1142 case KVM_REG_PPC_SDAR
:
1143 vcpu
->arch
.sdar
= set_reg_val(id
, *val
);
1145 case KVM_REG_PPC_SIER
:
1146 vcpu
->arch
.sier
= set_reg_val(id
, *val
);
1148 case KVM_REG_PPC_IAMR
:
1149 vcpu
->arch
.iamr
= set_reg_val(id
, *val
);
1151 case KVM_REG_PPC_PSPB
:
1152 vcpu
->arch
.pspb
= set_reg_val(id
, *val
);
1154 case KVM_REG_PPC_DPDES
:
1155 vcpu
->arch
.vcore
->dpdes
= set_reg_val(id
, *val
);
1157 case KVM_REG_PPC_DAWR
:
1158 vcpu
->arch
.dawr
= set_reg_val(id
, *val
);
1160 case KVM_REG_PPC_DAWRX
:
1161 vcpu
->arch
.dawrx
= set_reg_val(id
, *val
) & ~DAWRX_HYP
;
1163 case KVM_REG_PPC_CIABR
:
1164 vcpu
->arch
.ciabr
= set_reg_val(id
, *val
);
1165 /* Don't allow setting breakpoints in hypervisor code */
1166 if ((vcpu
->arch
.ciabr
& CIABR_PRIV
) == CIABR_PRIV_HYPER
)
1167 vcpu
->arch
.ciabr
&= ~CIABR_PRIV
; /* disable */
1169 case KVM_REG_PPC_CSIGR
:
1170 vcpu
->arch
.csigr
= set_reg_val(id
, *val
);
1172 case KVM_REG_PPC_TACR
:
1173 vcpu
->arch
.tacr
= set_reg_val(id
, *val
);
1175 case KVM_REG_PPC_TCSCR
:
1176 vcpu
->arch
.tcscr
= set_reg_val(id
, *val
);
1178 case KVM_REG_PPC_PID
:
1179 vcpu
->arch
.pid
= set_reg_val(id
, *val
);
1181 case KVM_REG_PPC_ACOP
:
1182 vcpu
->arch
.acop
= set_reg_val(id
, *val
);
1184 case KVM_REG_PPC_WORT
:
1185 vcpu
->arch
.wort
= set_reg_val(id
, *val
);
1187 case KVM_REG_PPC_VPA_ADDR
:
1188 addr
= set_reg_val(id
, *val
);
1190 if (!addr
&& (vcpu
->arch
.slb_shadow
.next_gpa
||
1191 vcpu
->arch
.dtl
.next_gpa
))
1193 r
= set_vpa(vcpu
, &vcpu
->arch
.vpa
, addr
, sizeof(struct lppaca
));
1195 case KVM_REG_PPC_VPA_SLB
:
1196 addr
= val
->vpaval
.addr
;
1197 len
= val
->vpaval
.length
;
1199 if (addr
&& !vcpu
->arch
.vpa
.next_gpa
)
1201 r
= set_vpa(vcpu
, &vcpu
->arch
.slb_shadow
, addr
, len
);
1203 case KVM_REG_PPC_VPA_DTL
:
1204 addr
= val
->vpaval
.addr
;
1205 len
= val
->vpaval
.length
;
1207 if (addr
&& (len
< sizeof(struct dtl_entry
) ||
1208 !vcpu
->arch
.vpa
.next_gpa
))
1210 len
-= len
% sizeof(struct dtl_entry
);
1211 r
= set_vpa(vcpu
, &vcpu
->arch
.dtl
, addr
, len
);
1213 case KVM_REG_PPC_TB_OFFSET
:
1214 /* round up to multiple of 2^24 */
1215 vcpu
->arch
.vcore
->tb_offset
=
1216 ALIGN(set_reg_val(id
, *val
), 1UL << 24);
1218 case KVM_REG_PPC_LPCR
:
1219 kvmppc_set_lpcr(vcpu
, set_reg_val(id
, *val
));
1221 case KVM_REG_PPC_PPR
:
1222 vcpu
->arch
.ppr
= set_reg_val(id
, *val
);
1224 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1225 case KVM_REG_PPC_TFHAR
:
1226 vcpu
->arch
.tfhar
= set_reg_val(id
, *val
);
1228 case KVM_REG_PPC_TFIAR
:
1229 vcpu
->arch
.tfiar
= set_reg_val(id
, *val
);
1231 case KVM_REG_PPC_TEXASR
:
1232 vcpu
->arch
.texasr
= set_reg_val(id
, *val
);
1234 case KVM_REG_PPC_TM_GPR0
... KVM_REG_PPC_TM_GPR31
:
1235 i
= id
- KVM_REG_PPC_TM_GPR0
;
1236 vcpu
->arch
.gpr_tm
[i
] = set_reg_val(id
, *val
);
1238 case KVM_REG_PPC_TM_VSR0
... KVM_REG_PPC_TM_VSR63
:
1241 i
= id
- KVM_REG_PPC_TM_VSR0
;
1243 for (j
= 0; j
< TS_FPRWIDTH
; j
++)
1244 vcpu
->arch
.fp_tm
.fpr
[i
][j
] = val
->vsxval
[j
];
1246 if (cpu_has_feature(CPU_FTR_ALTIVEC
))
1247 vcpu
->arch
.vr_tm
.vr
[i
-32] = val
->vval
;
1252 case KVM_REG_PPC_TM_CR
:
1253 vcpu
->arch
.cr_tm
= set_reg_val(id
, *val
);
1255 case KVM_REG_PPC_TM_LR
:
1256 vcpu
->arch
.lr_tm
= set_reg_val(id
, *val
);
1258 case KVM_REG_PPC_TM_CTR
:
1259 vcpu
->arch
.ctr_tm
= set_reg_val(id
, *val
);
1261 case KVM_REG_PPC_TM_FPSCR
:
1262 vcpu
->arch
.fp_tm
.fpscr
= set_reg_val(id
, *val
);
1264 case KVM_REG_PPC_TM_AMR
:
1265 vcpu
->arch
.amr_tm
= set_reg_val(id
, *val
);
1267 case KVM_REG_PPC_TM_PPR
:
1268 vcpu
->arch
.ppr_tm
= set_reg_val(id
, *val
);
1270 case KVM_REG_PPC_TM_VRSAVE
:
1271 vcpu
->arch
.vrsave_tm
= set_reg_val(id
, *val
);
1273 case KVM_REG_PPC_TM_VSCR
:
1274 if (cpu_has_feature(CPU_FTR_ALTIVEC
))
1275 vcpu
->arch
.vr
.vscr
.u
[3] = set_reg_val(id
, *val
);
1279 case KVM_REG_PPC_TM_DSCR
:
1280 vcpu
->arch
.dscr_tm
= set_reg_val(id
, *val
);
1282 case KVM_REG_PPC_TM_TAR
:
1283 vcpu
->arch
.tar_tm
= set_reg_val(id
, *val
);
1286 case KVM_REG_PPC_ARCH_COMPAT
:
1287 r
= kvmppc_set_arch_compat(vcpu
, set_reg_val(id
, *val
));
1297 static struct kvm_vcpu
*kvmppc_core_vcpu_create_hv(struct kvm
*kvm
,
1300 struct kvm_vcpu
*vcpu
;
1303 struct kvmppc_vcore
*vcore
;
1305 core
= id
/ threads_per_subcore
;
1306 if (core
>= KVM_MAX_VCORES
)
1310 vcpu
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL
);
1314 err
= kvm_vcpu_init(vcpu
, kvm
, id
);
1318 vcpu
->arch
.shared
= &vcpu
->arch
.shregs
;
1319 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1321 * The shared struct is never shared on HV,
1322 * so we can always use host endianness
1324 #ifdef __BIG_ENDIAN__
1325 vcpu
->arch
.shared_big_endian
= true;
1327 vcpu
->arch
.shared_big_endian
= false;
1330 vcpu
->arch
.mmcr
[0] = MMCR0_FC
;
1331 vcpu
->arch
.ctrl
= CTRL_RUNLATCH
;
1332 /* default to host PVR, since we can't spoof it */
1333 kvmppc_set_pvr_hv(vcpu
, mfspr(SPRN_PVR
));
1334 spin_lock_init(&vcpu
->arch
.vpa_update_lock
);
1335 spin_lock_init(&vcpu
->arch
.tbacct_lock
);
1336 vcpu
->arch
.busy_preempt
= TB_NIL
;
1337 vcpu
->arch
.intr_msr
= MSR_SF
| MSR_ME
;
1339 kvmppc_mmu_book3s_hv_init(vcpu
);
1341 vcpu
->arch
.state
= KVMPPC_VCPU_NOTREADY
;
1343 init_waitqueue_head(&vcpu
->arch
.cpu_run
);
1345 mutex_lock(&kvm
->lock
);
1346 vcore
= kvm
->arch
.vcores
[core
];
1348 vcore
= kzalloc(sizeof(struct kvmppc_vcore
), GFP_KERNEL
);
1350 INIT_LIST_HEAD(&vcore
->runnable_threads
);
1351 spin_lock_init(&vcore
->lock
);
1352 init_waitqueue_head(&vcore
->wq
);
1353 vcore
->preempt_tb
= TB_NIL
;
1354 vcore
->lpcr
= kvm
->arch
.lpcr
;
1355 vcore
->first_vcpuid
= core
* threads_per_subcore
;
1358 kvm
->arch
.vcores
[core
] = vcore
;
1359 kvm
->arch
.online_vcores
++;
1361 mutex_unlock(&kvm
->lock
);
1366 spin_lock(&vcore
->lock
);
1367 ++vcore
->num_threads
;
1368 spin_unlock(&vcore
->lock
);
1369 vcpu
->arch
.vcore
= vcore
;
1370 vcpu
->arch
.ptid
= vcpu
->vcpu_id
- vcore
->first_vcpuid
;
1372 vcpu
->arch
.cpu_type
= KVM_CPU_3S_64
;
1373 kvmppc_sanity_check(vcpu
);
1378 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
1380 return ERR_PTR(err
);
1383 static void unpin_vpa(struct kvm
*kvm
, struct kvmppc_vpa
*vpa
)
1385 if (vpa
->pinned_addr
)
1386 kvmppc_unpin_guest_page(kvm
, vpa
->pinned_addr
, vpa
->gpa
,
1390 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu
*vcpu
)
1392 spin_lock(&vcpu
->arch
.vpa_update_lock
);
1393 unpin_vpa(vcpu
->kvm
, &vcpu
->arch
.dtl
);
1394 unpin_vpa(vcpu
->kvm
, &vcpu
->arch
.slb_shadow
);
1395 unpin_vpa(vcpu
->kvm
, &vcpu
->arch
.vpa
);
1396 spin_unlock(&vcpu
->arch
.vpa_update_lock
);
1397 kvm_vcpu_uninit(vcpu
);
1398 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
1401 static int kvmppc_core_check_requests_hv(struct kvm_vcpu
*vcpu
)
1403 /* Indicate we want to get back into the guest */
1407 static void kvmppc_set_timer(struct kvm_vcpu
*vcpu
)
1409 unsigned long dec_nsec
, now
;
1412 if (now
> vcpu
->arch
.dec_expires
) {
1413 /* decrementer has already gone negative */
1414 kvmppc_core_queue_dec(vcpu
);
1415 kvmppc_core_prepare_to_enter(vcpu
);
1418 dec_nsec
= (vcpu
->arch
.dec_expires
- now
) * NSEC_PER_SEC
1420 hrtimer_start(&vcpu
->arch
.dec_timer
, ktime_set(0, dec_nsec
),
1422 vcpu
->arch
.timer_running
= 1;
1425 static void kvmppc_end_cede(struct kvm_vcpu
*vcpu
)
1427 vcpu
->arch
.ceded
= 0;
1428 if (vcpu
->arch
.timer_running
) {
1429 hrtimer_try_to_cancel(&vcpu
->arch
.dec_timer
);
1430 vcpu
->arch
.timer_running
= 0;
1434 extern void __kvmppc_vcore_entry(void);
1436 static void kvmppc_remove_runnable(struct kvmppc_vcore
*vc
,
1437 struct kvm_vcpu
*vcpu
)
1441 if (vcpu
->arch
.state
!= KVMPPC_VCPU_RUNNABLE
)
1443 spin_lock_irq(&vcpu
->arch
.tbacct_lock
);
1445 vcpu
->arch
.busy_stolen
+= vcore_stolen_time(vc
, now
) -
1446 vcpu
->arch
.stolen_logged
;
1447 vcpu
->arch
.busy_preempt
= now
;
1448 vcpu
->arch
.state
= KVMPPC_VCPU_BUSY_IN_HOST
;
1449 spin_unlock_irq(&vcpu
->arch
.tbacct_lock
);
1451 list_del(&vcpu
->arch
.run_list
);
1454 static int kvmppc_grab_hwthread(int cpu
)
1456 struct paca_struct
*tpaca
;
1457 long timeout
= 1000;
1461 /* Ensure the thread won't go into the kernel if it wakes */
1462 tpaca
->kvm_hstate
.hwthread_req
= 1;
1463 tpaca
->kvm_hstate
.kvm_vcpu
= NULL
;
1466 * If the thread is already executing in the kernel (e.g. handling
1467 * a stray interrupt), wait for it to get back to nap mode.
1468 * The smp_mb() is to ensure that our setting of hwthread_req
1469 * is visible before we look at hwthread_state, so if this
1470 * races with the code at system_reset_pSeries and the thread
1471 * misses our setting of hwthread_req, we are sure to see its
1472 * setting of hwthread_state, and vice versa.
1475 while (tpaca
->kvm_hstate
.hwthread_state
== KVM_HWTHREAD_IN_KERNEL
) {
1476 if (--timeout
<= 0) {
1477 pr_err("KVM: couldn't grab cpu %d\n", cpu
);
1485 static void kvmppc_release_hwthread(int cpu
)
1487 struct paca_struct
*tpaca
;
1490 tpaca
->kvm_hstate
.hwthread_req
= 0;
1491 tpaca
->kvm_hstate
.kvm_vcpu
= NULL
;
1494 static void kvmppc_start_thread(struct kvm_vcpu
*vcpu
)
1497 struct paca_struct
*tpaca
;
1498 struct kvmppc_vcore
*vc
= vcpu
->arch
.vcore
;
1500 if (vcpu
->arch
.timer_running
) {
1501 hrtimer_try_to_cancel(&vcpu
->arch
.dec_timer
);
1502 vcpu
->arch
.timer_running
= 0;
1504 cpu
= vc
->pcpu
+ vcpu
->arch
.ptid
;
1506 tpaca
->kvm_hstate
.kvm_vcpu
= vcpu
;
1507 tpaca
->kvm_hstate
.kvm_vcore
= vc
;
1508 tpaca
->kvm_hstate
.ptid
= vcpu
->arch
.ptid
;
1509 vcpu
->cpu
= vc
->pcpu
;
1511 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
1512 if (cpu
!= smp_processor_id()) {
1514 if (vcpu
->arch
.ptid
)
1520 static void kvmppc_wait_for_nap(struct kvmppc_vcore
*vc
)
1526 while (vc
->nap_count
< vc
->n_woken
) {
1527 if (++i
>= 1000000) {
1528 pr_err("kvmppc_wait_for_nap timeout %d %d\n",
1529 vc
->nap_count
, vc
->n_woken
);
1538 * Check that we are on thread 0 and that any other threads in
1539 * this core are off-line. Then grab the threads so they can't
1542 static int on_primary_thread(void)
1544 int cpu
= smp_processor_id();
1547 /* Are we on a primary subcore? */
1548 if (cpu_thread_in_subcore(cpu
))
1552 while (++thr
< threads_per_subcore
)
1553 if (cpu_online(cpu
+ thr
))
1556 /* Grab all hw threads so they can't go into the kernel */
1557 for (thr
= 1; thr
< threads_per_subcore
; ++thr
) {
1558 if (kvmppc_grab_hwthread(cpu
+ thr
)) {
1559 /* Couldn't grab one; let the others go */
1561 kvmppc_release_hwthread(cpu
+ thr
);
1562 } while (--thr
> 0);
1570 * Run a set of guest threads on a physical core.
1571 * Called with vc->lock held.
1573 static void kvmppc_run_core(struct kvmppc_vcore
*vc
)
1575 struct kvm_vcpu
*vcpu
, *vnext
;
1578 int i
, need_vpa_update
;
1580 struct kvm_vcpu
*vcpus_to_update
[threads_per_core
];
1582 /* don't start if any threads have a signal pending */
1583 need_vpa_update
= 0;
1584 list_for_each_entry(vcpu
, &vc
->runnable_threads
, arch
.run_list
) {
1585 if (signal_pending(vcpu
->arch
.run_task
))
1587 if (vcpu
->arch
.vpa
.update_pending
||
1588 vcpu
->arch
.slb_shadow
.update_pending
||
1589 vcpu
->arch
.dtl
.update_pending
)
1590 vcpus_to_update
[need_vpa_update
++] = vcpu
;
1594 * Initialize *vc, in particular vc->vcore_state, so we can
1595 * drop the vcore lock if necessary.
1599 vc
->entry_exit_count
= 0;
1600 vc
->vcore_state
= VCORE_STARTING
;
1602 vc
->napping_threads
= 0;
1605 * Updating any of the vpas requires calling kvmppc_pin_guest_page,
1606 * which can't be called with any spinlocks held.
1608 if (need_vpa_update
) {
1609 spin_unlock(&vc
->lock
);
1610 for (i
= 0; i
< need_vpa_update
; ++i
)
1611 kvmppc_update_vpas(vcpus_to_update
[i
]);
1612 spin_lock(&vc
->lock
);
1616 * Make sure we are running on primary threads, and that secondary
1617 * threads are offline. Also check if the number of threads in this
1618 * guest are greater than the current system threads per guest.
1620 if ((threads_per_core
> 1) &&
1621 ((vc
->num_threads
> threads_per_subcore
) || !on_primary_thread())) {
1622 list_for_each_entry(vcpu
, &vc
->runnable_threads
, arch
.run_list
)
1623 vcpu
->arch
.ret
= -EBUSY
;
1628 vc
->pcpu
= smp_processor_id();
1629 list_for_each_entry(vcpu
, &vc
->runnable_threads
, arch
.run_list
) {
1630 kvmppc_start_thread(vcpu
);
1631 kvmppc_create_dtl_entry(vcpu
, vc
);
1634 /* Set this explicitly in case thread 0 doesn't have a vcpu */
1635 get_paca()->kvm_hstate
.kvm_vcore
= vc
;
1636 get_paca()->kvm_hstate
.ptid
= 0;
1638 vc
->vcore_state
= VCORE_RUNNING
;
1640 spin_unlock(&vc
->lock
);
1644 srcu_idx
= srcu_read_lock(&vc
->kvm
->srcu
);
1646 __kvmppc_vcore_entry();
1648 spin_lock(&vc
->lock
);
1649 /* disable sending of IPIs on virtual external irqs */
1650 list_for_each_entry(vcpu
, &vc
->runnable_threads
, arch
.run_list
)
1652 /* wait for secondary threads to finish writing their state to memory */
1653 if (vc
->nap_count
< vc
->n_woken
)
1654 kvmppc_wait_for_nap(vc
);
1655 for (i
= 0; i
< threads_per_subcore
; ++i
)
1656 kvmppc_release_hwthread(vc
->pcpu
+ i
);
1657 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
1658 vc
->vcore_state
= VCORE_EXITING
;
1659 spin_unlock(&vc
->lock
);
1661 srcu_read_unlock(&vc
->kvm
->srcu
, srcu_idx
);
1663 /* make sure updates to secondary vcpu structs are visible now */
1670 spin_lock(&vc
->lock
);
1672 list_for_each_entry(vcpu
, &vc
->runnable_threads
, arch
.run_list
) {
1673 /* cancel pending dec exception if dec is positive */
1674 if (now
< vcpu
->arch
.dec_expires
&&
1675 kvmppc_core_pending_dec(vcpu
))
1676 kvmppc_core_dequeue_dec(vcpu
);
1679 if (vcpu
->arch
.trap
)
1680 ret
= kvmppc_handle_exit_hv(vcpu
->arch
.kvm_run
, vcpu
,
1681 vcpu
->arch
.run_task
);
1683 vcpu
->arch
.ret
= ret
;
1684 vcpu
->arch
.trap
= 0;
1686 if (vcpu
->arch
.ceded
) {
1687 if (!is_kvmppc_resume_guest(ret
))
1688 kvmppc_end_cede(vcpu
);
1690 kvmppc_set_timer(vcpu
);
1695 vc
->vcore_state
= VCORE_INACTIVE
;
1696 list_for_each_entry_safe(vcpu
, vnext
, &vc
->runnable_threads
,
1698 if (!is_kvmppc_resume_guest(vcpu
->arch
.ret
)) {
1699 kvmppc_remove_runnable(vc
, vcpu
);
1700 wake_up(&vcpu
->arch
.cpu_run
);
1706 * Wait for some other vcpu thread to execute us, and
1707 * wake us up when we need to handle something in the host.
1709 static void kvmppc_wait_for_exec(struct kvm_vcpu
*vcpu
, int wait_state
)
1713 prepare_to_wait(&vcpu
->arch
.cpu_run
, &wait
, wait_state
);
1714 if (vcpu
->arch
.state
== KVMPPC_VCPU_RUNNABLE
)
1716 finish_wait(&vcpu
->arch
.cpu_run
, &wait
);
1720 * All the vcpus in this vcore are idle, so wait for a decrementer
1721 * or external interrupt to one of the vcpus. vc->lock is held.
1723 static void kvmppc_vcore_blocked(struct kvmppc_vcore
*vc
)
1727 prepare_to_wait(&vc
->wq
, &wait
, TASK_INTERRUPTIBLE
);
1728 vc
->vcore_state
= VCORE_SLEEPING
;
1729 spin_unlock(&vc
->lock
);
1731 finish_wait(&vc
->wq
, &wait
);
1732 spin_lock(&vc
->lock
);
1733 vc
->vcore_state
= VCORE_INACTIVE
;
1736 static int kvmppc_run_vcpu(struct kvm_run
*kvm_run
, struct kvm_vcpu
*vcpu
)
1739 struct kvmppc_vcore
*vc
;
1740 struct kvm_vcpu
*v
, *vn
;
1742 kvm_run
->exit_reason
= 0;
1743 vcpu
->arch
.ret
= RESUME_GUEST
;
1744 vcpu
->arch
.trap
= 0;
1745 kvmppc_update_vpas(vcpu
);
1748 * Synchronize with other threads in this virtual core
1750 vc
= vcpu
->arch
.vcore
;
1751 spin_lock(&vc
->lock
);
1752 vcpu
->arch
.ceded
= 0;
1753 vcpu
->arch
.run_task
= current
;
1754 vcpu
->arch
.kvm_run
= kvm_run
;
1755 vcpu
->arch
.stolen_logged
= vcore_stolen_time(vc
, mftb());
1756 vcpu
->arch
.state
= KVMPPC_VCPU_RUNNABLE
;
1757 vcpu
->arch
.busy_preempt
= TB_NIL
;
1758 list_add_tail(&vcpu
->arch
.run_list
, &vc
->runnable_threads
);
1762 * This happens the first time this is called for a vcpu.
1763 * If the vcore is already running, we may be able to start
1764 * this thread straight away and have it join in.
1766 if (!signal_pending(current
)) {
1767 if (vc
->vcore_state
== VCORE_RUNNING
&&
1768 VCORE_EXIT_COUNT(vc
) == 0) {
1769 kvmppc_create_dtl_entry(vcpu
, vc
);
1770 kvmppc_start_thread(vcpu
);
1771 } else if (vc
->vcore_state
== VCORE_SLEEPING
) {
1777 while (vcpu
->arch
.state
== KVMPPC_VCPU_RUNNABLE
&&
1778 !signal_pending(current
)) {
1779 if (vc
->vcore_state
!= VCORE_INACTIVE
) {
1780 spin_unlock(&vc
->lock
);
1781 kvmppc_wait_for_exec(vcpu
, TASK_INTERRUPTIBLE
);
1782 spin_lock(&vc
->lock
);
1785 list_for_each_entry_safe(v
, vn
, &vc
->runnable_threads
,
1787 kvmppc_core_prepare_to_enter(v
);
1788 if (signal_pending(v
->arch
.run_task
)) {
1789 kvmppc_remove_runnable(vc
, v
);
1790 v
->stat
.signal_exits
++;
1791 v
->arch
.kvm_run
->exit_reason
= KVM_EXIT_INTR
;
1792 v
->arch
.ret
= -EINTR
;
1793 wake_up(&v
->arch
.cpu_run
);
1796 if (!vc
->n_runnable
|| vcpu
->arch
.state
!= KVMPPC_VCPU_RUNNABLE
)
1800 list_for_each_entry(v
, &vc
->runnable_threads
, arch
.run_list
) {
1801 if (!v
->arch
.pending_exceptions
)
1802 n_ceded
+= v
->arch
.ceded
;
1806 if (n_ceded
== vc
->n_runnable
)
1807 kvmppc_vcore_blocked(vc
);
1809 kvmppc_run_core(vc
);
1813 while (vcpu
->arch
.state
== KVMPPC_VCPU_RUNNABLE
&&
1814 (vc
->vcore_state
== VCORE_RUNNING
||
1815 vc
->vcore_state
== VCORE_EXITING
)) {
1816 spin_unlock(&vc
->lock
);
1817 kvmppc_wait_for_exec(vcpu
, TASK_UNINTERRUPTIBLE
);
1818 spin_lock(&vc
->lock
);
1821 if (vcpu
->arch
.state
== KVMPPC_VCPU_RUNNABLE
) {
1822 kvmppc_remove_runnable(vc
, vcpu
);
1823 vcpu
->stat
.signal_exits
++;
1824 kvm_run
->exit_reason
= KVM_EXIT_INTR
;
1825 vcpu
->arch
.ret
= -EINTR
;
1828 if (vc
->n_runnable
&& vc
->vcore_state
== VCORE_INACTIVE
) {
1829 /* Wake up some vcpu to run the core */
1830 v
= list_first_entry(&vc
->runnable_threads
,
1831 struct kvm_vcpu
, arch
.run_list
);
1832 wake_up(&v
->arch
.cpu_run
);
1835 spin_unlock(&vc
->lock
);
1836 return vcpu
->arch
.ret
;
1839 static int kvmppc_vcpu_run_hv(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
)
1844 if (!vcpu
->arch
.sane
) {
1845 run
->exit_reason
= KVM_EXIT_INTERNAL_ERROR
;
1849 kvmppc_core_prepare_to_enter(vcpu
);
1851 /* No need to go into the guest when all we'll do is come back out */
1852 if (signal_pending(current
)) {
1853 run
->exit_reason
= KVM_EXIT_INTR
;
1857 atomic_inc(&vcpu
->kvm
->arch
.vcpus_running
);
1858 /* Order vcpus_running vs. rma_setup_done, see kvmppc_alloc_reset_hpt */
1861 /* On the first time here, set up HTAB and VRMA or RMA */
1862 if (!vcpu
->kvm
->arch
.rma_setup_done
) {
1863 r
= kvmppc_hv_setup_htab_rma(vcpu
);
1868 flush_fp_to_thread(current
);
1869 flush_altivec_to_thread(current
);
1870 flush_vsx_to_thread(current
);
1871 vcpu
->arch
.wqp
= &vcpu
->arch
.vcore
->wq
;
1872 vcpu
->arch
.pgdir
= current
->mm
->pgd
;
1873 vcpu
->arch
.state
= KVMPPC_VCPU_BUSY_IN_HOST
;
1876 r
= kvmppc_run_vcpu(run
, vcpu
);
1878 if (run
->exit_reason
== KVM_EXIT_PAPR_HCALL
&&
1879 !(vcpu
->arch
.shregs
.msr
& MSR_PR
)) {
1880 r
= kvmppc_pseries_do_hcall(vcpu
);
1881 kvmppc_core_prepare_to_enter(vcpu
);
1882 } else if (r
== RESUME_PAGE_FAULT
) {
1883 srcu_idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
1884 r
= kvmppc_book3s_hv_page_fault(run
, vcpu
,
1885 vcpu
->arch
.fault_dar
, vcpu
->arch
.fault_dsisr
);
1886 srcu_read_unlock(&vcpu
->kvm
->srcu
, srcu_idx
);
1888 } while (is_kvmppc_resume_guest(r
));
1891 vcpu
->arch
.state
= KVMPPC_VCPU_NOTREADY
;
1892 atomic_dec(&vcpu
->kvm
->arch
.vcpus_running
);
1897 /* Work out RMLS (real mode limit selector) field value for a given RMA size.
1898 Assumes POWER7 or PPC970. */
1899 static inline int lpcr_rmls(unsigned long rma_size
)
1902 case 32ul << 20: /* 32 MB */
1903 if (cpu_has_feature(CPU_FTR_ARCH_206
))
1904 return 8; /* only supported on POWER7 */
1906 case 64ul << 20: /* 64 MB */
1908 case 128ul << 20: /* 128 MB */
1910 case 256ul << 20: /* 256 MB */
1912 case 1ul << 30: /* 1 GB */
1914 case 16ul << 30: /* 16 GB */
1916 case 256ul << 30: /* 256 GB */
1923 static int kvm_rma_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1926 struct kvm_rma_info
*ri
= vma
->vm_file
->private_data
;
1928 if (vmf
->pgoff
>= kvm_rma_pages
)
1929 return VM_FAULT_SIGBUS
;
1931 page
= pfn_to_page(ri
->base_pfn
+ vmf
->pgoff
);
1937 static const struct vm_operations_struct kvm_rma_vm_ops
= {
1938 .fault
= kvm_rma_fault
,
1941 static int kvm_rma_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1943 vma
->vm_flags
|= VM_DONTEXPAND
| VM_DONTDUMP
;
1944 vma
->vm_ops
= &kvm_rma_vm_ops
;
1948 static int kvm_rma_release(struct inode
*inode
, struct file
*filp
)
1950 struct kvm_rma_info
*ri
= filp
->private_data
;
1952 kvm_release_rma(ri
);
1956 static const struct file_operations kvm_rma_fops
= {
1957 .mmap
= kvm_rma_mmap
,
1958 .release
= kvm_rma_release
,
1961 static long kvm_vm_ioctl_allocate_rma(struct kvm
*kvm
,
1962 struct kvm_allocate_rma
*ret
)
1965 struct kvm_rma_info
*ri
;
1967 * Only do this on PPC970 in HV mode
1969 if (!cpu_has_feature(CPU_FTR_HVMODE
) ||
1970 !cpu_has_feature(CPU_FTR_ARCH_201
))
1976 ri
= kvm_alloc_rma();
1980 fd
= anon_inode_getfd("kvm-rma", &kvm_rma_fops
, ri
, O_RDWR
| O_CLOEXEC
);
1982 kvm_release_rma(ri
);
1984 ret
->rma_size
= kvm_rma_pages
<< PAGE_SHIFT
;
1988 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size
**sps
,
1991 struct mmu_psize_def
*def
= &mmu_psize_defs
[linux_psize
];
1995 (*sps
)->page_shift
= def
->shift
;
1996 (*sps
)->slb_enc
= def
->sllp
;
1997 (*sps
)->enc
[0].page_shift
= def
->shift
;
1999 * Only return base page encoding. We don't want to return
2000 * all the supporting pte_enc, because our H_ENTER doesn't
2001 * support MPSS yet. Once they do, we can start passing all
2002 * support pte_enc here
2004 (*sps
)->enc
[0].pte_enc
= def
->penc
[linux_psize
];
2006 * Add 16MB MPSS support if host supports it
2008 if (linux_psize
!= MMU_PAGE_16M
&& def
->penc
[MMU_PAGE_16M
] != -1) {
2009 (*sps
)->enc
[1].page_shift
= 24;
2010 (*sps
)->enc
[1].pte_enc
= def
->penc
[MMU_PAGE_16M
];
2015 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm
*kvm
,
2016 struct kvm_ppc_smmu_info
*info
)
2018 struct kvm_ppc_one_seg_page_size
*sps
;
2020 info
->flags
= KVM_PPC_PAGE_SIZES_REAL
;
2021 if (mmu_has_feature(MMU_FTR_1T_SEGMENT
))
2022 info
->flags
|= KVM_PPC_1T_SEGMENTS
;
2023 info
->slb_size
= mmu_slb_size
;
2025 /* We only support these sizes for now, and no muti-size segments */
2026 sps
= &info
->sps
[0];
2027 kvmppc_add_seg_page_size(&sps
, MMU_PAGE_4K
);
2028 kvmppc_add_seg_page_size(&sps
, MMU_PAGE_64K
);
2029 kvmppc_add_seg_page_size(&sps
, MMU_PAGE_16M
);
2035 * Get (and clear) the dirty memory log for a memory slot.
2037 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm
*kvm
,
2038 struct kvm_dirty_log
*log
)
2040 struct kvm_memory_slot
*memslot
;
2044 mutex_lock(&kvm
->slots_lock
);
2047 if (log
->slot
>= KVM_USER_MEM_SLOTS
)
2050 memslot
= id_to_memslot(kvm
->memslots
, log
->slot
);
2052 if (!memslot
->dirty_bitmap
)
2055 n
= kvm_dirty_bitmap_bytes(memslot
);
2056 memset(memslot
->dirty_bitmap
, 0, n
);
2058 r
= kvmppc_hv_get_dirty_log(kvm
, memslot
, memslot
->dirty_bitmap
);
2063 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
2068 mutex_unlock(&kvm
->slots_lock
);
2072 static void unpin_slot(struct kvm_memory_slot
*memslot
)
2074 unsigned long *physp
;
2075 unsigned long j
, npages
, pfn
;
2078 physp
= memslot
->arch
.slot_phys
;
2079 npages
= memslot
->npages
;
2082 for (j
= 0; j
< npages
; j
++) {
2083 if (!(physp
[j
] & KVMPPC_GOT_PAGE
))
2085 pfn
= physp
[j
] >> PAGE_SHIFT
;
2086 page
= pfn_to_page(pfn
);
2092 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot
*free
,
2093 struct kvm_memory_slot
*dont
)
2095 if (!dont
|| free
->arch
.rmap
!= dont
->arch
.rmap
) {
2096 vfree(free
->arch
.rmap
);
2097 free
->arch
.rmap
= NULL
;
2099 if (!dont
|| free
->arch
.slot_phys
!= dont
->arch
.slot_phys
) {
2101 vfree(free
->arch
.slot_phys
);
2102 free
->arch
.slot_phys
= NULL
;
2106 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot
*slot
,
2107 unsigned long npages
)
2109 slot
->arch
.rmap
= vzalloc(npages
* sizeof(*slot
->arch
.rmap
));
2110 if (!slot
->arch
.rmap
)
2112 slot
->arch
.slot_phys
= NULL
;
2117 static int kvmppc_core_prepare_memory_region_hv(struct kvm
*kvm
,
2118 struct kvm_memory_slot
*memslot
,
2119 struct kvm_userspace_memory_region
*mem
)
2121 unsigned long *phys
;
2123 /* Allocate a slot_phys array if needed */
2124 phys
= memslot
->arch
.slot_phys
;
2125 if (!kvm
->arch
.using_mmu_notifiers
&& !phys
&& memslot
->npages
) {
2126 phys
= vzalloc(memslot
->npages
* sizeof(unsigned long));
2129 memslot
->arch
.slot_phys
= phys
;
2135 static void kvmppc_core_commit_memory_region_hv(struct kvm
*kvm
,
2136 struct kvm_userspace_memory_region
*mem
,
2137 const struct kvm_memory_slot
*old
)
2139 unsigned long npages
= mem
->memory_size
>> PAGE_SHIFT
;
2140 struct kvm_memory_slot
*memslot
;
2142 if (npages
&& old
->npages
) {
2144 * If modifying a memslot, reset all the rmap dirty bits.
2145 * If this is a new memslot, we don't need to do anything
2146 * since the rmap array starts out as all zeroes,
2147 * i.e. no pages are dirty.
2149 memslot
= id_to_memslot(kvm
->memslots
, mem
->slot
);
2150 kvmppc_hv_get_dirty_log(kvm
, memslot
, NULL
);
2155 * Update LPCR values in kvm->arch and in vcores.
2156 * Caller must hold kvm->lock.
2158 void kvmppc_update_lpcr(struct kvm
*kvm
, unsigned long lpcr
, unsigned long mask
)
2163 if ((kvm
->arch
.lpcr
& mask
) == lpcr
)
2166 kvm
->arch
.lpcr
= (kvm
->arch
.lpcr
& ~mask
) | lpcr
;
2168 for (i
= 0; i
< KVM_MAX_VCORES
; ++i
) {
2169 struct kvmppc_vcore
*vc
= kvm
->arch
.vcores
[i
];
2172 spin_lock(&vc
->lock
);
2173 vc
->lpcr
= (vc
->lpcr
& ~mask
) | lpcr
;
2174 spin_unlock(&vc
->lock
);
2175 if (++cores_done
>= kvm
->arch
.online_vcores
)
2180 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu
*vcpu
)
2185 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu
*vcpu
)
2188 struct kvm
*kvm
= vcpu
->kvm
;
2189 struct kvm_rma_info
*ri
= NULL
;
2191 struct kvm_memory_slot
*memslot
;
2192 struct vm_area_struct
*vma
;
2193 unsigned long lpcr
= 0, senc
;
2194 unsigned long lpcr_mask
= 0;
2195 unsigned long psize
, porder
;
2196 unsigned long rma_size
;
2198 unsigned long *physp
;
2199 unsigned long i
, npages
;
2202 mutex_lock(&kvm
->lock
);
2203 if (kvm
->arch
.rma_setup_done
)
2204 goto out
; /* another vcpu beat us to it */
2206 /* Allocate hashed page table (if not done already) and reset it */
2207 if (!kvm
->arch
.hpt_virt
) {
2208 err
= kvmppc_alloc_hpt(kvm
, NULL
);
2210 pr_err("KVM: Couldn't alloc HPT\n");
2215 /* Look up the memslot for guest physical address 0 */
2216 srcu_idx
= srcu_read_lock(&kvm
->srcu
);
2217 memslot
= gfn_to_memslot(kvm
, 0);
2219 /* We must have some memory at 0 by now */
2221 if (!memslot
|| (memslot
->flags
& KVM_MEMSLOT_INVALID
))
2224 /* Look up the VMA for the start of this memory slot */
2225 hva
= memslot
->userspace_addr
;
2226 down_read(¤t
->mm
->mmap_sem
);
2227 vma
= find_vma(current
->mm
, hva
);
2228 if (!vma
|| vma
->vm_start
> hva
|| (vma
->vm_flags
& VM_IO
))
2231 psize
= vma_kernel_pagesize(vma
);
2232 porder
= __ilog2(psize
);
2234 /* Is this one of our preallocated RMAs? */
2235 if (vma
->vm_file
&& vma
->vm_file
->f_op
== &kvm_rma_fops
&&
2236 hva
== vma
->vm_start
)
2237 ri
= vma
->vm_file
->private_data
;
2239 up_read(¤t
->mm
->mmap_sem
);
2242 /* On POWER7, use VRMA; on PPC970, give up */
2244 if (cpu_has_feature(CPU_FTR_ARCH_201
)) {
2245 pr_err("KVM: CPU requires an RMO\n");
2249 /* We can handle 4k, 64k or 16M pages in the VRMA */
2251 if (!(psize
== 0x1000 || psize
== 0x10000 ||
2252 psize
== 0x1000000))
2255 /* Update VRMASD field in the LPCR */
2256 senc
= slb_pgsize_encoding(psize
);
2257 kvm
->arch
.vrma_slb_v
= senc
| SLB_VSID_B_1T
|
2258 (VRMA_VSID
<< SLB_VSID_SHIFT_1T
);
2259 lpcr_mask
= LPCR_VRMASD
;
2260 /* the -4 is to account for senc values starting at 0x10 */
2261 lpcr
= senc
<< (LPCR_VRMASD_SH
- 4);
2263 /* Create HPTEs in the hash page table for the VRMA */
2264 kvmppc_map_vrma(vcpu
, memslot
, porder
);
2267 /* Set up to use an RMO region */
2268 rma_size
= kvm_rma_pages
;
2269 if (rma_size
> memslot
->npages
)
2270 rma_size
= memslot
->npages
;
2271 rma_size
<<= PAGE_SHIFT
;
2272 rmls
= lpcr_rmls(rma_size
);
2274 if ((long)rmls
< 0) {
2275 pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size
);
2278 atomic_inc(&ri
->use_count
);
2281 /* Update LPCR and RMOR */
2282 if (cpu_has_feature(CPU_FTR_ARCH_201
)) {
2283 /* PPC970; insert RMLS value (split field) in HID4 */
2284 lpcr_mask
= (1ul << HID4_RMLS0_SH
) |
2285 (3ul << HID4_RMLS2_SH
) | HID4_RMOR
;
2286 lpcr
= ((rmls
>> 2) << HID4_RMLS0_SH
) |
2287 ((rmls
& 3) << HID4_RMLS2_SH
);
2288 /* RMOR is also in HID4 */
2289 lpcr
|= ((ri
->base_pfn
>> (26 - PAGE_SHIFT
)) & 0xffff)
2293 lpcr_mask
= LPCR_VPM0
| LPCR_VRMA_L
| LPCR_RMLS
;
2294 lpcr
= rmls
<< LPCR_RMLS_SH
;
2295 kvm
->arch
.rmor
= ri
->base_pfn
<< PAGE_SHIFT
;
2297 pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n",
2298 ri
->base_pfn
<< PAGE_SHIFT
, rma_size
, lpcr
);
2300 /* Initialize phys addrs of pages in RMO */
2301 npages
= kvm_rma_pages
;
2302 porder
= __ilog2(npages
);
2303 physp
= memslot
->arch
.slot_phys
;
2305 if (npages
> memslot
->npages
)
2306 npages
= memslot
->npages
;
2307 spin_lock(&kvm
->arch
.slot_phys_lock
);
2308 for (i
= 0; i
< npages
; ++i
)
2309 physp
[i
] = ((ri
->base_pfn
+ i
) << PAGE_SHIFT
) +
2311 spin_unlock(&kvm
->arch
.slot_phys_lock
);
2315 kvmppc_update_lpcr(kvm
, lpcr
, lpcr_mask
);
2317 /* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */
2319 kvm
->arch
.rma_setup_done
= 1;
2322 srcu_read_unlock(&kvm
->srcu
, srcu_idx
);
2324 mutex_unlock(&kvm
->lock
);
2328 up_read(¤t
->mm
->mmap_sem
);
2332 static int kvmppc_core_init_vm_hv(struct kvm
*kvm
)
2334 unsigned long lpcr
, lpid
;
2336 /* Allocate the guest's logical partition ID */
2338 lpid
= kvmppc_alloc_lpid();
2341 kvm
->arch
.lpid
= lpid
;
2344 * Since we don't flush the TLB when tearing down a VM,
2345 * and this lpid might have previously been used,
2346 * make sure we flush on each core before running the new VM.
2348 cpumask_setall(&kvm
->arch
.need_tlb_flush
);
2350 /* Start out with the default set of hcalls enabled */
2351 memcpy(kvm
->arch
.enabled_hcalls
, default_enabled_hcalls
,
2352 sizeof(kvm
->arch
.enabled_hcalls
));
2354 kvm
->arch
.rma
= NULL
;
2356 kvm
->arch
.host_sdr1
= mfspr(SPRN_SDR1
);
2358 if (cpu_has_feature(CPU_FTR_ARCH_201
)) {
2359 /* PPC970; HID4 is effectively the LPCR */
2360 kvm
->arch
.host_lpid
= 0;
2361 kvm
->arch
.host_lpcr
= lpcr
= mfspr(SPRN_HID4
);
2362 lpcr
&= ~((3 << HID4_LPID1_SH
) | (0xful
<< HID4_LPID5_SH
));
2363 lpcr
|= ((lpid
>> 4) << HID4_LPID1_SH
) |
2364 ((lpid
& 0xf) << HID4_LPID5_SH
);
2366 /* POWER7; init LPCR for virtual RMA mode */
2367 kvm
->arch
.host_lpid
= mfspr(SPRN_LPID
);
2368 kvm
->arch
.host_lpcr
= lpcr
= mfspr(SPRN_LPCR
);
2369 lpcr
&= LPCR_PECE
| LPCR_LPES
;
2370 lpcr
|= (4UL << LPCR_DPFD_SH
) | LPCR_HDICE
|
2371 LPCR_VPM0
| LPCR_VPM1
;
2372 kvm
->arch
.vrma_slb_v
= SLB_VSID_B_1T
|
2373 (VRMA_VSID
<< SLB_VSID_SHIFT_1T
);
2374 /* On POWER8 turn on online bit to enable PURR/SPURR */
2375 if (cpu_has_feature(CPU_FTR_ARCH_207S
))
2378 kvm
->arch
.lpcr
= lpcr
;
2380 kvm
->arch
.using_mmu_notifiers
= !!cpu_has_feature(CPU_FTR_ARCH_206
);
2381 spin_lock_init(&kvm
->arch
.slot_phys_lock
);
2384 * Track that we now have a HV mode VM active. This blocks secondary
2385 * CPU threads from coming online.
2387 kvm_hv_vm_activated();
2392 static void kvmppc_free_vcores(struct kvm
*kvm
)
2396 for (i
= 0; i
< KVM_MAX_VCORES
; ++i
)
2397 kfree(kvm
->arch
.vcores
[i
]);
2398 kvm
->arch
.online_vcores
= 0;
2401 static void kvmppc_core_destroy_vm_hv(struct kvm
*kvm
)
2403 kvm_hv_vm_deactivated();
2405 kvmppc_free_vcores(kvm
);
2406 if (kvm
->arch
.rma
) {
2407 kvm_release_rma(kvm
->arch
.rma
);
2408 kvm
->arch
.rma
= NULL
;
2411 kvmppc_free_hpt(kvm
);
2414 /* We don't need to emulate any privileged instructions or dcbz */
2415 static int kvmppc_core_emulate_op_hv(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
,
2416 unsigned int inst
, int *advance
)
2418 return EMULATE_FAIL
;
2421 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu
*vcpu
, int sprn
,
2424 return EMULATE_FAIL
;
2427 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu
*vcpu
, int sprn
,
2430 return EMULATE_FAIL
;
2433 static int kvmppc_core_check_processor_compat_hv(void)
2435 if (!cpu_has_feature(CPU_FTR_HVMODE
))
2440 static long kvm_arch_vm_ioctl_hv(struct file
*filp
,
2441 unsigned int ioctl
, unsigned long arg
)
2443 struct kvm
*kvm __maybe_unused
= filp
->private_data
;
2444 void __user
*argp
= (void __user
*)arg
;
2449 case KVM_ALLOCATE_RMA
: {
2450 struct kvm_allocate_rma rma
;
2451 struct kvm
*kvm
= filp
->private_data
;
2453 r
= kvm_vm_ioctl_allocate_rma(kvm
, &rma
);
2454 if (r
>= 0 && copy_to_user(argp
, &rma
, sizeof(rma
)))
2459 case KVM_PPC_ALLOCATE_HTAB
: {
2463 if (get_user(htab_order
, (u32 __user
*)argp
))
2465 r
= kvmppc_alloc_reset_hpt(kvm
, &htab_order
);
2469 if (put_user(htab_order
, (u32 __user
*)argp
))
2475 case KVM_PPC_GET_HTAB_FD
: {
2476 struct kvm_get_htab_fd ghf
;
2479 if (copy_from_user(&ghf
, argp
, sizeof(ghf
)))
2481 r
= kvm_vm_ioctl_get_htab_fd(kvm
, &ghf
);
2493 * List of hcall numbers to enable by default.
2494 * For compatibility with old userspace, we enable by default
2495 * all hcalls that were implemented before the hcall-enabling
2496 * facility was added. Note this list should not include H_RTAS.
2498 static unsigned int default_hcall_list
[] = {
2512 #ifdef CONFIG_KVM_XICS
2523 static void init_default_hcalls(void)
2528 for (i
= 0; default_hcall_list
[i
]; ++i
) {
2529 hcall
= default_hcall_list
[i
];
2530 WARN_ON(!kvmppc_hcall_impl_hv(hcall
));
2531 __set_bit(hcall
/ 4, default_enabled_hcalls
);
2535 static struct kvmppc_ops kvm_ops_hv
= {
2536 .get_sregs
= kvm_arch_vcpu_ioctl_get_sregs_hv
,
2537 .set_sregs
= kvm_arch_vcpu_ioctl_set_sregs_hv
,
2538 .get_one_reg
= kvmppc_get_one_reg_hv
,
2539 .set_one_reg
= kvmppc_set_one_reg_hv
,
2540 .vcpu_load
= kvmppc_core_vcpu_load_hv
,
2541 .vcpu_put
= kvmppc_core_vcpu_put_hv
,
2542 .set_msr
= kvmppc_set_msr_hv
,
2543 .vcpu_run
= kvmppc_vcpu_run_hv
,
2544 .vcpu_create
= kvmppc_core_vcpu_create_hv
,
2545 .vcpu_free
= kvmppc_core_vcpu_free_hv
,
2546 .check_requests
= kvmppc_core_check_requests_hv
,
2547 .get_dirty_log
= kvm_vm_ioctl_get_dirty_log_hv
,
2548 .flush_memslot
= kvmppc_core_flush_memslot_hv
,
2549 .prepare_memory_region
= kvmppc_core_prepare_memory_region_hv
,
2550 .commit_memory_region
= kvmppc_core_commit_memory_region_hv
,
2551 .unmap_hva
= kvm_unmap_hva_hv
,
2552 .unmap_hva_range
= kvm_unmap_hva_range_hv
,
2553 .age_hva
= kvm_age_hva_hv
,
2554 .test_age_hva
= kvm_test_age_hva_hv
,
2555 .set_spte_hva
= kvm_set_spte_hva_hv
,
2556 .mmu_destroy
= kvmppc_mmu_destroy_hv
,
2557 .free_memslot
= kvmppc_core_free_memslot_hv
,
2558 .create_memslot
= kvmppc_core_create_memslot_hv
,
2559 .init_vm
= kvmppc_core_init_vm_hv
,
2560 .destroy_vm
= kvmppc_core_destroy_vm_hv
,
2561 .get_smmu_info
= kvm_vm_ioctl_get_smmu_info_hv
,
2562 .emulate_op
= kvmppc_core_emulate_op_hv
,
2563 .emulate_mtspr
= kvmppc_core_emulate_mtspr_hv
,
2564 .emulate_mfspr
= kvmppc_core_emulate_mfspr_hv
,
2565 .fast_vcpu_kick
= kvmppc_fast_vcpu_kick_hv
,
2566 .arch_vm_ioctl
= kvm_arch_vm_ioctl_hv
,
2567 .hcall_implemented
= kvmppc_hcall_impl_hv
,
2570 static int kvmppc_book3s_init_hv(void)
2574 * FIXME!! Do we need to check on all cpus ?
2576 r
= kvmppc_core_check_processor_compat_hv();
2580 kvm_ops_hv
.owner
= THIS_MODULE
;
2581 kvmppc_hv_ops
= &kvm_ops_hv
;
2583 init_default_hcalls();
2585 r
= kvmppc_mmu_hv_init();
2589 static void kvmppc_book3s_exit_hv(void)
2591 kvmppc_hv_ops
= NULL
;
2594 module_init(kvmppc_book3s_init_hv
);
2595 module_exit(kvmppc_book3s_exit_hv
);
2596 MODULE_LICENSE("GPL");
2597 MODULE_ALIAS_MISCDEV(KVM_MINOR
);
2598 MODULE_ALIAS("devname:kvm");