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>
35 #include <linux/debugfs.h>
38 #include <asm/cputable.h>
39 #include <asm/cacheflush.h>
40 #include <asm/tlbflush.h>
41 #include <asm/uaccess.h>
43 #include <asm/kvm_ppc.h>
44 #include <asm/kvm_book3s.h>
45 #include <asm/mmu_context.h>
46 #include <asm/lppaca.h>
47 #include <asm/processor.h>
48 #include <asm/cputhreads.h>
50 #include <asm/hvcall.h>
51 #include <asm/switch_to.h>
53 #include <asm/dbell.h>
54 #include <linux/gfp.h>
55 #include <linux/vmalloc.h>
56 #include <linux/highmem.h>
57 #include <linux/hugetlb.h>
58 #include <linux/module.h>
62 #define CREATE_TRACE_POINTS
65 /* #define EXIT_DEBUG */
66 /* #define EXIT_DEBUG_SIMPLE */
67 /* #define EXIT_DEBUG_INT */
69 /* Used to indicate that a guest page fault needs to be handled */
70 #define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1)
72 /* Used as a "null" value for timebase values */
73 #define TB_NIL (~(u64)0)
75 static DECLARE_BITMAP(default_enabled_hcalls
, MAX_HCALL_OPCODE
/4 + 1);
77 static int dynamic_mt_modes
= 6;
78 module_param(dynamic_mt_modes
, int, S_IRUGO
| S_IWUSR
);
79 MODULE_PARM_DESC(dynamic_mt_modes
, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
80 static int target_smt_mode
;
81 module_param(target_smt_mode
, int, S_IRUGO
| S_IWUSR
);
82 MODULE_PARM_DESC(target_smt_mode
, "Target threads per core (0 = max)");
84 static void kvmppc_end_cede(struct kvm_vcpu
*vcpu
);
85 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu
*vcpu
);
87 static bool kvmppc_ipi_thread(int cpu
)
89 /* On POWER8 for IPIs to threads in the same core, use msgsnd */
90 if (cpu_has_feature(CPU_FTR_ARCH_207S
)) {
92 if (cpu_first_thread_sibling(cpu
) ==
93 cpu_first_thread_sibling(smp_processor_id())) {
94 unsigned long msg
= PPC_DBELL_TYPE(PPC_DBELL_SERVER
);
95 msg
|= cpu_thread_in_core(cpu
);
97 __asm__
__volatile__ (PPC_MSGSND(%0) : : "r" (msg
));
104 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
105 if (cpu
>= 0 && cpu
< nr_cpu_ids
&& paca
[cpu
].kvm_hstate
.xics_phys
) {
114 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu
*vcpu
)
117 wait_queue_head_t
*wqp
;
119 wqp
= kvm_arch_vcpu_wq(vcpu
);
120 if (waitqueue_active(wqp
)) {
121 wake_up_interruptible(wqp
);
122 ++vcpu
->stat
.halt_wakeup
;
125 if (kvmppc_ipi_thread(vcpu
->arch
.thread_cpu
))
128 /* CPU points to the first thread of the core */
130 if (cpu
>= 0 && cpu
< nr_cpu_ids
&& cpu_online(cpu
))
131 smp_send_reschedule(cpu
);
135 * We use the vcpu_load/put functions to measure stolen time.
136 * Stolen time is counted as time when either the vcpu is able to
137 * run as part of a virtual core, but the task running the vcore
138 * is preempted or sleeping, or when the vcpu needs something done
139 * in the kernel by the task running the vcpu, but that task is
140 * preempted or sleeping. Those two things have to be counted
141 * separately, since one of the vcpu tasks will take on the job
142 * of running the core, and the other vcpu tasks in the vcore will
143 * sleep waiting for it to do that, but that sleep shouldn't count
146 * Hence we accumulate stolen time when the vcpu can run as part of
147 * a vcore using vc->stolen_tb, and the stolen time when the vcpu
148 * needs its task to do other things in the kernel (for example,
149 * service a page fault) in busy_stolen. We don't accumulate
150 * stolen time for a vcore when it is inactive, or for a vcpu
151 * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of
152 * a misnomer; it means that the vcpu task is not executing in
153 * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
154 * the kernel. We don't have any way of dividing up that time
155 * between time that the vcpu is genuinely stopped, time that
156 * the task is actively working on behalf of the vcpu, and time
157 * that the task is preempted, so we don't count any of it as
160 * Updates to busy_stolen are protected by arch.tbacct_lock;
161 * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
162 * lock. The stolen times are measured in units of timebase ticks.
163 * (Note that the != TB_NIL checks below are purely defensive;
164 * they should never fail.)
167 static void kvmppc_core_start_stolen(struct kvmppc_vcore
*vc
)
171 spin_lock_irqsave(&vc
->stoltb_lock
, flags
);
172 vc
->preempt_tb
= mftb();
173 spin_unlock_irqrestore(&vc
->stoltb_lock
, flags
);
176 static void kvmppc_core_end_stolen(struct kvmppc_vcore
*vc
)
180 spin_lock_irqsave(&vc
->stoltb_lock
, flags
);
181 if (vc
->preempt_tb
!= TB_NIL
) {
182 vc
->stolen_tb
+= mftb() - vc
->preempt_tb
;
183 vc
->preempt_tb
= TB_NIL
;
185 spin_unlock_irqrestore(&vc
->stoltb_lock
, flags
);
188 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu
*vcpu
, int cpu
)
190 struct kvmppc_vcore
*vc
= vcpu
->arch
.vcore
;
194 * We can test vc->runner without taking the vcore lock,
195 * because only this task ever sets vc->runner to this
196 * vcpu, and once it is set to this vcpu, only this task
197 * ever sets it to NULL.
199 if (vc
->runner
== vcpu
&& vc
->vcore_state
>= VCORE_SLEEPING
)
200 kvmppc_core_end_stolen(vc
);
202 spin_lock_irqsave(&vcpu
->arch
.tbacct_lock
, flags
);
203 if (vcpu
->arch
.state
== KVMPPC_VCPU_BUSY_IN_HOST
&&
204 vcpu
->arch
.busy_preempt
!= TB_NIL
) {
205 vcpu
->arch
.busy_stolen
+= mftb() - vcpu
->arch
.busy_preempt
;
206 vcpu
->arch
.busy_preempt
= TB_NIL
;
208 spin_unlock_irqrestore(&vcpu
->arch
.tbacct_lock
, flags
);
211 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu
*vcpu
)
213 struct kvmppc_vcore
*vc
= vcpu
->arch
.vcore
;
216 if (vc
->runner
== vcpu
&& vc
->vcore_state
>= VCORE_SLEEPING
)
217 kvmppc_core_start_stolen(vc
);
219 spin_lock_irqsave(&vcpu
->arch
.tbacct_lock
, flags
);
220 if (vcpu
->arch
.state
== KVMPPC_VCPU_BUSY_IN_HOST
)
221 vcpu
->arch
.busy_preempt
= mftb();
222 spin_unlock_irqrestore(&vcpu
->arch
.tbacct_lock
, flags
);
225 static void kvmppc_set_msr_hv(struct kvm_vcpu
*vcpu
, u64 msr
)
227 vcpu
->arch
.shregs
.msr
= msr
;
228 kvmppc_end_cede(vcpu
);
231 static void kvmppc_set_pvr_hv(struct kvm_vcpu
*vcpu
, u32 pvr
)
233 vcpu
->arch
.pvr
= pvr
;
236 static int kvmppc_set_arch_compat(struct kvm_vcpu
*vcpu
, u32 arch_compat
)
238 unsigned long pcr
= 0;
239 struct kvmppc_vcore
*vc
= vcpu
->arch
.vcore
;
242 switch (arch_compat
) {
245 * If an arch bit is set in PCR, all the defined
246 * higher-order arch bits also have to be set.
248 pcr
= PCR_ARCH_206
| PCR_ARCH_205
;
260 if (!cpu_has_feature(CPU_FTR_ARCH_207S
)) {
261 /* POWER7 can't emulate POWER8 */
262 if (!(pcr
& PCR_ARCH_206
))
264 pcr
&= ~PCR_ARCH_206
;
268 spin_lock(&vc
->lock
);
269 vc
->arch_compat
= arch_compat
;
271 spin_unlock(&vc
->lock
);
276 static void kvmppc_dump_regs(struct kvm_vcpu
*vcpu
)
280 pr_err("vcpu %p (%d):\n", vcpu
, vcpu
->vcpu_id
);
281 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
282 vcpu
->arch
.pc
, vcpu
->arch
.shregs
.msr
, vcpu
->arch
.trap
);
283 for (r
= 0; r
< 16; ++r
)
284 pr_err("r%2d = %.16lx r%d = %.16lx\n",
285 r
, kvmppc_get_gpr(vcpu
, r
),
286 r
+16, kvmppc_get_gpr(vcpu
, r
+16));
287 pr_err("ctr = %.16lx lr = %.16lx\n",
288 vcpu
->arch
.ctr
, vcpu
->arch
.lr
);
289 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
290 vcpu
->arch
.shregs
.srr0
, vcpu
->arch
.shregs
.srr1
);
291 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
292 vcpu
->arch
.shregs
.sprg0
, vcpu
->arch
.shregs
.sprg1
);
293 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
294 vcpu
->arch
.shregs
.sprg2
, vcpu
->arch
.shregs
.sprg3
);
295 pr_err("cr = %.8x xer = %.16lx dsisr = %.8x\n",
296 vcpu
->arch
.cr
, vcpu
->arch
.xer
, vcpu
->arch
.shregs
.dsisr
);
297 pr_err("dar = %.16llx\n", vcpu
->arch
.shregs
.dar
);
298 pr_err("fault dar = %.16lx dsisr = %.8x\n",
299 vcpu
->arch
.fault_dar
, vcpu
->arch
.fault_dsisr
);
300 pr_err("SLB (%d entries):\n", vcpu
->arch
.slb_max
);
301 for (r
= 0; r
< vcpu
->arch
.slb_max
; ++r
)
302 pr_err(" ESID = %.16llx VSID = %.16llx\n",
303 vcpu
->arch
.slb
[r
].orige
, vcpu
->arch
.slb
[r
].origv
);
304 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
305 vcpu
->arch
.vcore
->lpcr
, vcpu
->kvm
->arch
.sdr1
,
306 vcpu
->arch
.last_inst
);
309 static struct kvm_vcpu
*kvmppc_find_vcpu(struct kvm
*kvm
, int id
)
312 struct kvm_vcpu
*v
, *ret
= NULL
;
314 mutex_lock(&kvm
->lock
);
315 kvm_for_each_vcpu(r
, v
, kvm
) {
316 if (v
->vcpu_id
== id
) {
321 mutex_unlock(&kvm
->lock
);
325 static void init_vpa(struct kvm_vcpu
*vcpu
, struct lppaca
*vpa
)
327 vpa
->__old_status
|= LPPACA_OLD_SHARED_PROC
;
328 vpa
->yield_count
= cpu_to_be32(1);
331 static int set_vpa(struct kvm_vcpu
*vcpu
, struct kvmppc_vpa
*v
,
332 unsigned long addr
, unsigned long len
)
334 /* check address is cacheline aligned */
335 if (addr
& (L1_CACHE_BYTES
- 1))
337 spin_lock(&vcpu
->arch
.vpa_update_lock
);
338 if (v
->next_gpa
!= addr
|| v
->len
!= len
) {
340 v
->len
= addr
? len
: 0;
341 v
->update_pending
= 1;
343 spin_unlock(&vcpu
->arch
.vpa_update_lock
);
347 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
356 static int vpa_is_registered(struct kvmppc_vpa
*vpap
)
358 if (vpap
->update_pending
)
359 return vpap
->next_gpa
!= 0;
360 return vpap
->pinned_addr
!= NULL
;
363 static unsigned long do_h_register_vpa(struct kvm_vcpu
*vcpu
,
365 unsigned long vcpuid
, unsigned long vpa
)
367 struct kvm
*kvm
= vcpu
->kvm
;
368 unsigned long len
, nb
;
370 struct kvm_vcpu
*tvcpu
;
373 struct kvmppc_vpa
*vpap
;
375 tvcpu
= kvmppc_find_vcpu(kvm
, vcpuid
);
379 subfunc
= (flags
>> H_VPA_FUNC_SHIFT
) & H_VPA_FUNC_MASK
;
380 if (subfunc
== H_VPA_REG_VPA
|| subfunc
== H_VPA_REG_DTL
||
381 subfunc
== H_VPA_REG_SLB
) {
382 /* Registering new area - address must be cache-line aligned */
383 if ((vpa
& (L1_CACHE_BYTES
- 1)) || !vpa
)
386 /* convert logical addr to kernel addr and read length */
387 va
= kvmppc_pin_guest_page(kvm
, vpa
, &nb
);
390 if (subfunc
== H_VPA_REG_VPA
)
391 len
= be16_to_cpu(((struct reg_vpa
*)va
)->length
.hword
);
393 len
= be32_to_cpu(((struct reg_vpa
*)va
)->length
.word
);
394 kvmppc_unpin_guest_page(kvm
, va
, vpa
, false);
397 if (len
> nb
|| len
< sizeof(struct reg_vpa
))
406 spin_lock(&tvcpu
->arch
.vpa_update_lock
);
409 case H_VPA_REG_VPA
: /* register VPA */
410 if (len
< sizeof(struct lppaca
))
412 vpap
= &tvcpu
->arch
.vpa
;
416 case H_VPA_REG_DTL
: /* register DTL */
417 if (len
< sizeof(struct dtl_entry
))
419 len
-= len
% sizeof(struct dtl_entry
);
421 /* Check that they have previously registered a VPA */
423 if (!vpa_is_registered(&tvcpu
->arch
.vpa
))
426 vpap
= &tvcpu
->arch
.dtl
;
430 case H_VPA_REG_SLB
: /* register SLB shadow buffer */
431 /* Check that they have previously registered a VPA */
433 if (!vpa_is_registered(&tvcpu
->arch
.vpa
))
436 vpap
= &tvcpu
->arch
.slb_shadow
;
440 case H_VPA_DEREG_VPA
: /* deregister VPA */
441 /* Check they don't still have a DTL or SLB buf registered */
443 if (vpa_is_registered(&tvcpu
->arch
.dtl
) ||
444 vpa_is_registered(&tvcpu
->arch
.slb_shadow
))
447 vpap
= &tvcpu
->arch
.vpa
;
451 case H_VPA_DEREG_DTL
: /* deregister DTL */
452 vpap
= &tvcpu
->arch
.dtl
;
456 case H_VPA_DEREG_SLB
: /* deregister SLB shadow buffer */
457 vpap
= &tvcpu
->arch
.slb_shadow
;
463 vpap
->next_gpa
= vpa
;
465 vpap
->update_pending
= 1;
468 spin_unlock(&tvcpu
->arch
.vpa_update_lock
);
473 static void kvmppc_update_vpa(struct kvm_vcpu
*vcpu
, struct kvmppc_vpa
*vpap
)
475 struct kvm
*kvm
= vcpu
->kvm
;
481 * We need to pin the page pointed to by vpap->next_gpa,
482 * but we can't call kvmppc_pin_guest_page under the lock
483 * as it does get_user_pages() and down_read(). So we
484 * have to drop the lock, pin the page, then get the lock
485 * again and check that a new area didn't get registered
489 gpa
= vpap
->next_gpa
;
490 spin_unlock(&vcpu
->arch
.vpa_update_lock
);
494 va
= kvmppc_pin_guest_page(kvm
, gpa
, &nb
);
495 spin_lock(&vcpu
->arch
.vpa_update_lock
);
496 if (gpa
== vpap
->next_gpa
)
498 /* sigh... unpin that one and try again */
500 kvmppc_unpin_guest_page(kvm
, va
, gpa
, false);
503 vpap
->update_pending
= 0;
504 if (va
&& nb
< vpap
->len
) {
506 * If it's now too short, it must be that userspace
507 * has changed the mappings underlying guest memory,
508 * so unregister the region.
510 kvmppc_unpin_guest_page(kvm
, va
, gpa
, false);
513 if (vpap
->pinned_addr
)
514 kvmppc_unpin_guest_page(kvm
, vpap
->pinned_addr
, vpap
->gpa
,
517 vpap
->pinned_addr
= va
;
520 vpap
->pinned_end
= va
+ vpap
->len
;
523 static void kvmppc_update_vpas(struct kvm_vcpu
*vcpu
)
525 if (!(vcpu
->arch
.vpa
.update_pending
||
526 vcpu
->arch
.slb_shadow
.update_pending
||
527 vcpu
->arch
.dtl
.update_pending
))
530 spin_lock(&vcpu
->arch
.vpa_update_lock
);
531 if (vcpu
->arch
.vpa
.update_pending
) {
532 kvmppc_update_vpa(vcpu
, &vcpu
->arch
.vpa
);
533 if (vcpu
->arch
.vpa
.pinned_addr
)
534 init_vpa(vcpu
, vcpu
->arch
.vpa
.pinned_addr
);
536 if (vcpu
->arch
.dtl
.update_pending
) {
537 kvmppc_update_vpa(vcpu
, &vcpu
->arch
.dtl
);
538 vcpu
->arch
.dtl_ptr
= vcpu
->arch
.dtl
.pinned_addr
;
539 vcpu
->arch
.dtl_index
= 0;
541 if (vcpu
->arch
.slb_shadow
.update_pending
)
542 kvmppc_update_vpa(vcpu
, &vcpu
->arch
.slb_shadow
);
543 spin_unlock(&vcpu
->arch
.vpa_update_lock
);
547 * Return the accumulated stolen time for the vcore up until `now'.
548 * The caller should hold the vcore lock.
550 static u64
vcore_stolen_time(struct kvmppc_vcore
*vc
, u64 now
)
555 spin_lock_irqsave(&vc
->stoltb_lock
, flags
);
557 if (vc
->vcore_state
!= VCORE_INACTIVE
&&
558 vc
->preempt_tb
!= TB_NIL
)
559 p
+= now
- vc
->preempt_tb
;
560 spin_unlock_irqrestore(&vc
->stoltb_lock
, flags
);
564 static void kvmppc_create_dtl_entry(struct kvm_vcpu
*vcpu
,
565 struct kvmppc_vcore
*vc
)
567 struct dtl_entry
*dt
;
569 unsigned long stolen
;
570 unsigned long core_stolen
;
573 dt
= vcpu
->arch
.dtl_ptr
;
574 vpa
= vcpu
->arch
.vpa
.pinned_addr
;
576 core_stolen
= vcore_stolen_time(vc
, now
);
577 stolen
= core_stolen
- vcpu
->arch
.stolen_logged
;
578 vcpu
->arch
.stolen_logged
= core_stolen
;
579 spin_lock_irq(&vcpu
->arch
.tbacct_lock
);
580 stolen
+= vcpu
->arch
.busy_stolen
;
581 vcpu
->arch
.busy_stolen
= 0;
582 spin_unlock_irq(&vcpu
->arch
.tbacct_lock
);
585 memset(dt
, 0, sizeof(struct dtl_entry
));
586 dt
->dispatch_reason
= 7;
587 dt
->processor_id
= cpu_to_be16(vc
->pcpu
+ vcpu
->arch
.ptid
);
588 dt
->timebase
= cpu_to_be64(now
+ vc
->tb_offset
);
589 dt
->enqueue_to_dispatch_time
= cpu_to_be32(stolen
);
590 dt
->srr0
= cpu_to_be64(kvmppc_get_pc(vcpu
));
591 dt
->srr1
= cpu_to_be64(vcpu
->arch
.shregs
.msr
);
593 if (dt
== vcpu
->arch
.dtl
.pinned_end
)
594 dt
= vcpu
->arch
.dtl
.pinned_addr
;
595 vcpu
->arch
.dtl_ptr
= dt
;
596 /* order writing *dt vs. writing vpa->dtl_idx */
598 vpa
->dtl_idx
= cpu_to_be64(++vcpu
->arch
.dtl_index
);
599 vcpu
->arch
.dtl
.dirty
= true;
602 static bool kvmppc_power8_compatible(struct kvm_vcpu
*vcpu
)
604 if (vcpu
->arch
.vcore
->arch_compat
>= PVR_ARCH_207
)
606 if ((!vcpu
->arch
.vcore
->arch_compat
) &&
607 cpu_has_feature(CPU_FTR_ARCH_207S
))
612 static int kvmppc_h_set_mode(struct kvm_vcpu
*vcpu
, unsigned long mflags
,
613 unsigned long resource
, unsigned long value1
,
614 unsigned long value2
)
617 case H_SET_MODE_RESOURCE_SET_CIABR
:
618 if (!kvmppc_power8_compatible(vcpu
))
623 return H_UNSUPPORTED_FLAG_START
;
624 /* Guests can't breakpoint the hypervisor */
625 if ((value1
& CIABR_PRIV
) == CIABR_PRIV_HYPER
)
627 vcpu
->arch
.ciabr
= value1
;
629 case H_SET_MODE_RESOURCE_SET_DAWR
:
630 if (!kvmppc_power8_compatible(vcpu
))
633 return H_UNSUPPORTED_FLAG_START
;
634 if (value2
& DABRX_HYP
)
636 vcpu
->arch
.dawr
= value1
;
637 vcpu
->arch
.dawrx
= value2
;
644 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu
*target
)
646 struct kvmppc_vcore
*vcore
= target
->arch
.vcore
;
649 * We expect to have been called by the real mode handler
650 * (kvmppc_rm_h_confer()) which would have directly returned
651 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
652 * have useful work to do and should not confer) so we don't
656 spin_lock(&vcore
->lock
);
657 if (target
->arch
.state
== KVMPPC_VCPU_RUNNABLE
&&
658 vcore
->vcore_state
!= VCORE_INACTIVE
&&
660 target
= vcore
->runner
;
661 spin_unlock(&vcore
->lock
);
663 return kvm_vcpu_yield_to(target
);
666 static int kvmppc_get_yield_count(struct kvm_vcpu
*vcpu
)
669 struct lppaca
*lppaca
;
671 spin_lock(&vcpu
->arch
.vpa_update_lock
);
672 lppaca
= (struct lppaca
*)vcpu
->arch
.vpa
.pinned_addr
;
674 yield_count
= be32_to_cpu(lppaca
->yield_count
);
675 spin_unlock(&vcpu
->arch
.vpa_update_lock
);
679 int kvmppc_pseries_do_hcall(struct kvm_vcpu
*vcpu
)
681 unsigned long req
= kvmppc_get_gpr(vcpu
, 3);
682 unsigned long target
, ret
= H_SUCCESS
;
684 struct kvm_vcpu
*tvcpu
;
687 if (req
<= MAX_HCALL_OPCODE
&&
688 !test_bit(req
/4, vcpu
->kvm
->arch
.enabled_hcalls
))
695 target
= kvmppc_get_gpr(vcpu
, 4);
696 tvcpu
= kvmppc_find_vcpu(vcpu
->kvm
, target
);
701 tvcpu
->arch
.prodded
= 1;
703 if (vcpu
->arch
.ceded
) {
704 if (waitqueue_active(&vcpu
->wq
)) {
705 wake_up_interruptible(&vcpu
->wq
);
706 vcpu
->stat
.halt_wakeup
++;
711 target
= kvmppc_get_gpr(vcpu
, 4);
714 tvcpu
= kvmppc_find_vcpu(vcpu
->kvm
, target
);
719 yield_count
= kvmppc_get_gpr(vcpu
, 5);
720 if (kvmppc_get_yield_count(tvcpu
) != yield_count
)
722 kvm_arch_vcpu_yield_to(tvcpu
);
725 ret
= do_h_register_vpa(vcpu
, kvmppc_get_gpr(vcpu
, 4),
726 kvmppc_get_gpr(vcpu
, 5),
727 kvmppc_get_gpr(vcpu
, 6));
730 if (list_empty(&vcpu
->kvm
->arch
.rtas_tokens
))
733 idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
734 rc
= kvmppc_rtas_hcall(vcpu
);
735 srcu_read_unlock(&vcpu
->kvm
->srcu
, idx
);
742 /* Send the error out to userspace via KVM_RUN */
744 case H_LOGICAL_CI_LOAD
:
745 ret
= kvmppc_h_logical_ci_load(vcpu
);
746 if (ret
== H_TOO_HARD
)
749 case H_LOGICAL_CI_STORE
:
750 ret
= kvmppc_h_logical_ci_store(vcpu
);
751 if (ret
== H_TOO_HARD
)
755 ret
= kvmppc_h_set_mode(vcpu
, kvmppc_get_gpr(vcpu
, 4),
756 kvmppc_get_gpr(vcpu
, 5),
757 kvmppc_get_gpr(vcpu
, 6),
758 kvmppc_get_gpr(vcpu
, 7));
759 if (ret
== H_TOO_HARD
)
768 if (kvmppc_xics_enabled(vcpu
)) {
769 ret
= kvmppc_xics_hcall(vcpu
, req
);
775 kvmppc_set_gpr(vcpu
, 3, ret
);
776 vcpu
->arch
.hcall_needed
= 0;
780 static int kvmppc_hcall_impl_hv(unsigned long cmd
)
788 case H_LOGICAL_CI_LOAD
:
789 case H_LOGICAL_CI_STORE
:
790 #ifdef CONFIG_KVM_XICS
801 /* See if it's in the real-mode table */
802 return kvmppc_hcall_impl_hv_realmode(cmd
);
805 static int kvmppc_emulate_debug_inst(struct kvm_run
*run
,
806 struct kvm_vcpu
*vcpu
)
810 if (kvmppc_get_last_inst(vcpu
, INST_GENERIC
, &last_inst
) !=
813 * Fetch failed, so return to guest and
814 * try executing it again.
819 if (last_inst
== KVMPPC_INST_SW_BREAKPOINT
) {
820 run
->exit_reason
= KVM_EXIT_DEBUG
;
821 run
->debug
.arch
.address
= kvmppc_get_pc(vcpu
);
824 kvmppc_core_queue_program(vcpu
, SRR1_PROGILL
);
829 static int kvmppc_handle_exit_hv(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
,
830 struct task_struct
*tsk
)
834 vcpu
->stat
.sum_exits
++;
836 run
->exit_reason
= KVM_EXIT_UNKNOWN
;
837 run
->ready_for_interrupt_injection
= 1;
838 switch (vcpu
->arch
.trap
) {
839 /* We're good on these - the host merely wanted to get our attention */
840 case BOOK3S_INTERRUPT_HV_DECREMENTER
:
841 vcpu
->stat
.dec_exits
++;
844 case BOOK3S_INTERRUPT_EXTERNAL
:
845 case BOOK3S_INTERRUPT_H_DOORBELL
:
846 vcpu
->stat
.ext_intr_exits
++;
849 /* HMI is hypervisor interrupt and host has handled it. Resume guest.*/
850 case BOOK3S_INTERRUPT_HMI
:
851 case BOOK3S_INTERRUPT_PERFMON
:
854 case BOOK3S_INTERRUPT_MACHINE_CHECK
:
856 * Deliver a machine check interrupt to the guest.
857 * We have to do this, even if the host has handled the
858 * machine check, because machine checks use SRR0/1 and
859 * the interrupt might have trashed guest state in them.
861 kvmppc_book3s_queue_irqprio(vcpu
,
862 BOOK3S_INTERRUPT_MACHINE_CHECK
);
865 case BOOK3S_INTERRUPT_PROGRAM
:
869 * Normally program interrupts are delivered directly
870 * to the guest by the hardware, but we can get here
871 * as a result of a hypervisor emulation interrupt
872 * (e40) getting turned into a 700 by BML RTAS.
874 flags
= vcpu
->arch
.shregs
.msr
& 0x1f0000ull
;
875 kvmppc_core_queue_program(vcpu
, flags
);
879 case BOOK3S_INTERRUPT_SYSCALL
:
881 /* hcall - punt to userspace */
884 /* hypercall with MSR_PR has already been handled in rmode,
885 * and never reaches here.
888 run
->papr_hcall
.nr
= kvmppc_get_gpr(vcpu
, 3);
889 for (i
= 0; i
< 9; ++i
)
890 run
->papr_hcall
.args
[i
] = kvmppc_get_gpr(vcpu
, 4 + i
);
891 run
->exit_reason
= KVM_EXIT_PAPR_HCALL
;
892 vcpu
->arch
.hcall_needed
= 1;
897 * We get these next two if the guest accesses a page which it thinks
898 * it has mapped but which is not actually present, either because
899 * it is for an emulated I/O device or because the corresonding
900 * host page has been paged out. Any other HDSI/HISI interrupts
901 * have been handled already.
903 case BOOK3S_INTERRUPT_H_DATA_STORAGE
:
904 r
= RESUME_PAGE_FAULT
;
906 case BOOK3S_INTERRUPT_H_INST_STORAGE
:
907 vcpu
->arch
.fault_dar
= kvmppc_get_pc(vcpu
);
908 vcpu
->arch
.fault_dsisr
= 0;
909 r
= RESUME_PAGE_FAULT
;
912 * This occurs if the guest executes an illegal instruction.
913 * If the guest debug is disabled, generate a program interrupt
914 * to the guest. If guest debug is enabled, we need to check
915 * whether the instruction is a software breakpoint instruction.
916 * Accordingly return to Guest or Host.
918 case BOOK3S_INTERRUPT_H_EMUL_ASSIST
:
919 if (vcpu
->arch
.emul_inst
!= KVM_INST_FETCH_FAILED
)
920 vcpu
->arch
.last_inst
= kvmppc_need_byteswap(vcpu
) ?
921 swab32(vcpu
->arch
.emul_inst
) :
922 vcpu
->arch
.emul_inst
;
923 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_SW_BP
) {
924 r
= kvmppc_emulate_debug_inst(run
, vcpu
);
926 kvmppc_core_queue_program(vcpu
, SRR1_PROGILL
);
931 * This occurs if the guest (kernel or userspace), does something that
932 * is prohibited by HFSCR. We just generate a program interrupt to
935 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL
:
936 kvmppc_core_queue_program(vcpu
, SRR1_PROGILL
);
940 kvmppc_dump_regs(vcpu
);
941 printk(KERN_EMERG
"trap=0x%x | pc=0x%lx | msr=0x%llx\n",
942 vcpu
->arch
.trap
, kvmppc_get_pc(vcpu
),
943 vcpu
->arch
.shregs
.msr
);
944 run
->hw
.hardware_exit_reason
= vcpu
->arch
.trap
;
952 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu
*vcpu
,
953 struct kvm_sregs
*sregs
)
957 memset(sregs
, 0, sizeof(struct kvm_sregs
));
958 sregs
->pvr
= vcpu
->arch
.pvr
;
959 for (i
= 0; i
< vcpu
->arch
.slb_max
; i
++) {
960 sregs
->u
.s
.ppc64
.slb
[i
].slbe
= vcpu
->arch
.slb
[i
].orige
;
961 sregs
->u
.s
.ppc64
.slb
[i
].slbv
= vcpu
->arch
.slb
[i
].origv
;
967 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu
*vcpu
,
968 struct kvm_sregs
*sregs
)
972 /* Only accept the same PVR as the host's, since we can't spoof it */
973 if (sregs
->pvr
!= vcpu
->arch
.pvr
)
977 for (i
= 0; i
< vcpu
->arch
.slb_nr
; i
++) {
978 if (sregs
->u
.s
.ppc64
.slb
[i
].slbe
& SLB_ESID_V
) {
979 vcpu
->arch
.slb
[j
].orige
= sregs
->u
.s
.ppc64
.slb
[i
].slbe
;
980 vcpu
->arch
.slb
[j
].origv
= sregs
->u
.s
.ppc64
.slb
[i
].slbv
;
984 vcpu
->arch
.slb_max
= j
;
989 static void kvmppc_set_lpcr(struct kvm_vcpu
*vcpu
, u64 new_lpcr
,
992 struct kvm
*kvm
= vcpu
->kvm
;
993 struct kvmppc_vcore
*vc
= vcpu
->arch
.vcore
;
996 mutex_lock(&kvm
->lock
);
997 spin_lock(&vc
->lock
);
999 * If ILE (interrupt little-endian) has changed, update the
1000 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1002 if ((new_lpcr
& LPCR_ILE
) != (vc
->lpcr
& LPCR_ILE
)) {
1003 struct kvm_vcpu
*vcpu
;
1006 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
1007 if (vcpu
->arch
.vcore
!= vc
)
1009 if (new_lpcr
& LPCR_ILE
)
1010 vcpu
->arch
.intr_msr
|= MSR_LE
;
1012 vcpu
->arch
.intr_msr
&= ~MSR_LE
;
1017 * Userspace can only modify DPFD (default prefetch depth),
1018 * ILE (interrupt little-endian) and TC (translation control).
1019 * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
1021 mask
= LPCR_DPFD
| LPCR_ILE
| LPCR_TC
;
1022 if (cpu_has_feature(CPU_FTR_ARCH_207S
))
1025 /* Broken 32-bit version of LPCR must not clear top bits */
1028 vc
->lpcr
= (vc
->lpcr
& ~mask
) | (new_lpcr
& mask
);
1029 spin_unlock(&vc
->lock
);
1030 mutex_unlock(&kvm
->lock
);
1033 static int kvmppc_get_one_reg_hv(struct kvm_vcpu
*vcpu
, u64 id
,
1034 union kvmppc_one_reg
*val
)
1040 case KVM_REG_PPC_DEBUG_INST
:
1041 *val
= get_reg_val(id
, KVMPPC_INST_SW_BREAKPOINT
);
1043 case KVM_REG_PPC_HIOR
:
1044 *val
= get_reg_val(id
, 0);
1046 case KVM_REG_PPC_DABR
:
1047 *val
= get_reg_val(id
, vcpu
->arch
.dabr
);
1049 case KVM_REG_PPC_DABRX
:
1050 *val
= get_reg_val(id
, vcpu
->arch
.dabrx
);
1052 case KVM_REG_PPC_DSCR
:
1053 *val
= get_reg_val(id
, vcpu
->arch
.dscr
);
1055 case KVM_REG_PPC_PURR
:
1056 *val
= get_reg_val(id
, vcpu
->arch
.purr
);
1058 case KVM_REG_PPC_SPURR
:
1059 *val
= get_reg_val(id
, vcpu
->arch
.spurr
);
1061 case KVM_REG_PPC_AMR
:
1062 *val
= get_reg_val(id
, vcpu
->arch
.amr
);
1064 case KVM_REG_PPC_UAMOR
:
1065 *val
= get_reg_val(id
, vcpu
->arch
.uamor
);
1067 case KVM_REG_PPC_MMCR0
... KVM_REG_PPC_MMCRS
:
1068 i
= id
- KVM_REG_PPC_MMCR0
;
1069 *val
= get_reg_val(id
, vcpu
->arch
.mmcr
[i
]);
1071 case KVM_REG_PPC_PMC1
... KVM_REG_PPC_PMC8
:
1072 i
= id
- KVM_REG_PPC_PMC1
;
1073 *val
= get_reg_val(id
, vcpu
->arch
.pmc
[i
]);
1075 case KVM_REG_PPC_SPMC1
... KVM_REG_PPC_SPMC2
:
1076 i
= id
- KVM_REG_PPC_SPMC1
;
1077 *val
= get_reg_val(id
, vcpu
->arch
.spmc
[i
]);
1079 case KVM_REG_PPC_SIAR
:
1080 *val
= get_reg_val(id
, vcpu
->arch
.siar
);
1082 case KVM_REG_PPC_SDAR
:
1083 *val
= get_reg_val(id
, vcpu
->arch
.sdar
);
1085 case KVM_REG_PPC_SIER
:
1086 *val
= get_reg_val(id
, vcpu
->arch
.sier
);
1088 case KVM_REG_PPC_IAMR
:
1089 *val
= get_reg_val(id
, vcpu
->arch
.iamr
);
1091 case KVM_REG_PPC_PSPB
:
1092 *val
= get_reg_val(id
, vcpu
->arch
.pspb
);
1094 case KVM_REG_PPC_DPDES
:
1095 *val
= get_reg_val(id
, vcpu
->arch
.vcore
->dpdes
);
1097 case KVM_REG_PPC_DAWR
:
1098 *val
= get_reg_val(id
, vcpu
->arch
.dawr
);
1100 case KVM_REG_PPC_DAWRX
:
1101 *val
= get_reg_val(id
, vcpu
->arch
.dawrx
);
1103 case KVM_REG_PPC_CIABR
:
1104 *val
= get_reg_val(id
, vcpu
->arch
.ciabr
);
1106 case KVM_REG_PPC_CSIGR
:
1107 *val
= get_reg_val(id
, vcpu
->arch
.csigr
);
1109 case KVM_REG_PPC_TACR
:
1110 *val
= get_reg_val(id
, vcpu
->arch
.tacr
);
1112 case KVM_REG_PPC_TCSCR
:
1113 *val
= get_reg_val(id
, vcpu
->arch
.tcscr
);
1115 case KVM_REG_PPC_PID
:
1116 *val
= get_reg_val(id
, vcpu
->arch
.pid
);
1118 case KVM_REG_PPC_ACOP
:
1119 *val
= get_reg_val(id
, vcpu
->arch
.acop
);
1121 case KVM_REG_PPC_WORT
:
1122 *val
= get_reg_val(id
, vcpu
->arch
.wort
);
1124 case KVM_REG_PPC_VPA_ADDR
:
1125 spin_lock(&vcpu
->arch
.vpa_update_lock
);
1126 *val
= get_reg_val(id
, vcpu
->arch
.vpa
.next_gpa
);
1127 spin_unlock(&vcpu
->arch
.vpa_update_lock
);
1129 case KVM_REG_PPC_VPA_SLB
:
1130 spin_lock(&vcpu
->arch
.vpa_update_lock
);
1131 val
->vpaval
.addr
= vcpu
->arch
.slb_shadow
.next_gpa
;
1132 val
->vpaval
.length
= vcpu
->arch
.slb_shadow
.len
;
1133 spin_unlock(&vcpu
->arch
.vpa_update_lock
);
1135 case KVM_REG_PPC_VPA_DTL
:
1136 spin_lock(&vcpu
->arch
.vpa_update_lock
);
1137 val
->vpaval
.addr
= vcpu
->arch
.dtl
.next_gpa
;
1138 val
->vpaval
.length
= vcpu
->arch
.dtl
.len
;
1139 spin_unlock(&vcpu
->arch
.vpa_update_lock
);
1141 case KVM_REG_PPC_TB_OFFSET
:
1142 *val
= get_reg_val(id
, vcpu
->arch
.vcore
->tb_offset
);
1144 case KVM_REG_PPC_LPCR
:
1145 case KVM_REG_PPC_LPCR_64
:
1146 *val
= get_reg_val(id
, vcpu
->arch
.vcore
->lpcr
);
1148 case KVM_REG_PPC_PPR
:
1149 *val
= get_reg_val(id
, vcpu
->arch
.ppr
);
1151 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1152 case KVM_REG_PPC_TFHAR
:
1153 *val
= get_reg_val(id
, vcpu
->arch
.tfhar
);
1155 case KVM_REG_PPC_TFIAR
:
1156 *val
= get_reg_val(id
, vcpu
->arch
.tfiar
);
1158 case KVM_REG_PPC_TEXASR
:
1159 *val
= get_reg_val(id
, vcpu
->arch
.texasr
);
1161 case KVM_REG_PPC_TM_GPR0
... KVM_REG_PPC_TM_GPR31
:
1162 i
= id
- KVM_REG_PPC_TM_GPR0
;
1163 *val
= get_reg_val(id
, vcpu
->arch
.gpr_tm
[i
]);
1165 case KVM_REG_PPC_TM_VSR0
... KVM_REG_PPC_TM_VSR63
:
1168 i
= id
- KVM_REG_PPC_TM_VSR0
;
1170 for (j
= 0; j
< TS_FPRWIDTH
; j
++)
1171 val
->vsxval
[j
] = vcpu
->arch
.fp_tm
.fpr
[i
][j
];
1173 if (cpu_has_feature(CPU_FTR_ALTIVEC
))
1174 val
->vval
= vcpu
->arch
.vr_tm
.vr
[i
-32];
1180 case KVM_REG_PPC_TM_CR
:
1181 *val
= get_reg_val(id
, vcpu
->arch
.cr_tm
);
1183 case KVM_REG_PPC_TM_LR
:
1184 *val
= get_reg_val(id
, vcpu
->arch
.lr_tm
);
1186 case KVM_REG_PPC_TM_CTR
:
1187 *val
= get_reg_val(id
, vcpu
->arch
.ctr_tm
);
1189 case KVM_REG_PPC_TM_FPSCR
:
1190 *val
= get_reg_val(id
, vcpu
->arch
.fp_tm
.fpscr
);
1192 case KVM_REG_PPC_TM_AMR
:
1193 *val
= get_reg_val(id
, vcpu
->arch
.amr_tm
);
1195 case KVM_REG_PPC_TM_PPR
:
1196 *val
= get_reg_val(id
, vcpu
->arch
.ppr_tm
);
1198 case KVM_REG_PPC_TM_VRSAVE
:
1199 *val
= get_reg_val(id
, vcpu
->arch
.vrsave_tm
);
1201 case KVM_REG_PPC_TM_VSCR
:
1202 if (cpu_has_feature(CPU_FTR_ALTIVEC
))
1203 *val
= get_reg_val(id
, vcpu
->arch
.vr_tm
.vscr
.u
[3]);
1207 case KVM_REG_PPC_TM_DSCR
:
1208 *val
= get_reg_val(id
, vcpu
->arch
.dscr_tm
);
1210 case KVM_REG_PPC_TM_TAR
:
1211 *val
= get_reg_val(id
, vcpu
->arch
.tar_tm
);
1214 case KVM_REG_PPC_ARCH_COMPAT
:
1215 *val
= get_reg_val(id
, vcpu
->arch
.vcore
->arch_compat
);
1225 static int kvmppc_set_one_reg_hv(struct kvm_vcpu
*vcpu
, u64 id
,
1226 union kvmppc_one_reg
*val
)
1230 unsigned long addr
, len
;
1233 case KVM_REG_PPC_HIOR
:
1234 /* Only allow this to be set to zero */
1235 if (set_reg_val(id
, *val
))
1238 case KVM_REG_PPC_DABR
:
1239 vcpu
->arch
.dabr
= set_reg_val(id
, *val
);
1241 case KVM_REG_PPC_DABRX
:
1242 vcpu
->arch
.dabrx
= set_reg_val(id
, *val
) & ~DABRX_HYP
;
1244 case KVM_REG_PPC_DSCR
:
1245 vcpu
->arch
.dscr
= set_reg_val(id
, *val
);
1247 case KVM_REG_PPC_PURR
:
1248 vcpu
->arch
.purr
= set_reg_val(id
, *val
);
1250 case KVM_REG_PPC_SPURR
:
1251 vcpu
->arch
.spurr
= set_reg_val(id
, *val
);
1253 case KVM_REG_PPC_AMR
:
1254 vcpu
->arch
.amr
= set_reg_val(id
, *val
);
1256 case KVM_REG_PPC_UAMOR
:
1257 vcpu
->arch
.uamor
= set_reg_val(id
, *val
);
1259 case KVM_REG_PPC_MMCR0
... KVM_REG_PPC_MMCRS
:
1260 i
= id
- KVM_REG_PPC_MMCR0
;
1261 vcpu
->arch
.mmcr
[i
] = set_reg_val(id
, *val
);
1263 case KVM_REG_PPC_PMC1
... KVM_REG_PPC_PMC8
:
1264 i
= id
- KVM_REG_PPC_PMC1
;
1265 vcpu
->arch
.pmc
[i
] = set_reg_val(id
, *val
);
1267 case KVM_REG_PPC_SPMC1
... KVM_REG_PPC_SPMC2
:
1268 i
= id
- KVM_REG_PPC_SPMC1
;
1269 vcpu
->arch
.spmc
[i
] = set_reg_val(id
, *val
);
1271 case KVM_REG_PPC_SIAR
:
1272 vcpu
->arch
.siar
= set_reg_val(id
, *val
);
1274 case KVM_REG_PPC_SDAR
:
1275 vcpu
->arch
.sdar
= set_reg_val(id
, *val
);
1277 case KVM_REG_PPC_SIER
:
1278 vcpu
->arch
.sier
= set_reg_val(id
, *val
);
1280 case KVM_REG_PPC_IAMR
:
1281 vcpu
->arch
.iamr
= set_reg_val(id
, *val
);
1283 case KVM_REG_PPC_PSPB
:
1284 vcpu
->arch
.pspb
= set_reg_val(id
, *val
);
1286 case KVM_REG_PPC_DPDES
:
1287 vcpu
->arch
.vcore
->dpdes
= set_reg_val(id
, *val
);
1289 case KVM_REG_PPC_DAWR
:
1290 vcpu
->arch
.dawr
= set_reg_val(id
, *val
);
1292 case KVM_REG_PPC_DAWRX
:
1293 vcpu
->arch
.dawrx
= set_reg_val(id
, *val
) & ~DAWRX_HYP
;
1295 case KVM_REG_PPC_CIABR
:
1296 vcpu
->arch
.ciabr
= set_reg_val(id
, *val
);
1297 /* Don't allow setting breakpoints in hypervisor code */
1298 if ((vcpu
->arch
.ciabr
& CIABR_PRIV
) == CIABR_PRIV_HYPER
)
1299 vcpu
->arch
.ciabr
&= ~CIABR_PRIV
; /* disable */
1301 case KVM_REG_PPC_CSIGR
:
1302 vcpu
->arch
.csigr
= set_reg_val(id
, *val
);
1304 case KVM_REG_PPC_TACR
:
1305 vcpu
->arch
.tacr
= set_reg_val(id
, *val
);
1307 case KVM_REG_PPC_TCSCR
:
1308 vcpu
->arch
.tcscr
= set_reg_val(id
, *val
);
1310 case KVM_REG_PPC_PID
:
1311 vcpu
->arch
.pid
= set_reg_val(id
, *val
);
1313 case KVM_REG_PPC_ACOP
:
1314 vcpu
->arch
.acop
= set_reg_val(id
, *val
);
1316 case KVM_REG_PPC_WORT
:
1317 vcpu
->arch
.wort
= set_reg_val(id
, *val
);
1319 case KVM_REG_PPC_VPA_ADDR
:
1320 addr
= set_reg_val(id
, *val
);
1322 if (!addr
&& (vcpu
->arch
.slb_shadow
.next_gpa
||
1323 vcpu
->arch
.dtl
.next_gpa
))
1325 r
= set_vpa(vcpu
, &vcpu
->arch
.vpa
, addr
, sizeof(struct lppaca
));
1327 case KVM_REG_PPC_VPA_SLB
:
1328 addr
= val
->vpaval
.addr
;
1329 len
= val
->vpaval
.length
;
1331 if (addr
&& !vcpu
->arch
.vpa
.next_gpa
)
1333 r
= set_vpa(vcpu
, &vcpu
->arch
.slb_shadow
, addr
, len
);
1335 case KVM_REG_PPC_VPA_DTL
:
1336 addr
= val
->vpaval
.addr
;
1337 len
= val
->vpaval
.length
;
1339 if (addr
&& (len
< sizeof(struct dtl_entry
) ||
1340 !vcpu
->arch
.vpa
.next_gpa
))
1342 len
-= len
% sizeof(struct dtl_entry
);
1343 r
= set_vpa(vcpu
, &vcpu
->arch
.dtl
, addr
, len
);
1345 case KVM_REG_PPC_TB_OFFSET
:
1346 /* round up to multiple of 2^24 */
1347 vcpu
->arch
.vcore
->tb_offset
=
1348 ALIGN(set_reg_val(id
, *val
), 1UL << 24);
1350 case KVM_REG_PPC_LPCR
:
1351 kvmppc_set_lpcr(vcpu
, set_reg_val(id
, *val
), true);
1353 case KVM_REG_PPC_LPCR_64
:
1354 kvmppc_set_lpcr(vcpu
, set_reg_val(id
, *val
), false);
1356 case KVM_REG_PPC_PPR
:
1357 vcpu
->arch
.ppr
= set_reg_val(id
, *val
);
1359 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1360 case KVM_REG_PPC_TFHAR
:
1361 vcpu
->arch
.tfhar
= set_reg_val(id
, *val
);
1363 case KVM_REG_PPC_TFIAR
:
1364 vcpu
->arch
.tfiar
= set_reg_val(id
, *val
);
1366 case KVM_REG_PPC_TEXASR
:
1367 vcpu
->arch
.texasr
= set_reg_val(id
, *val
);
1369 case KVM_REG_PPC_TM_GPR0
... KVM_REG_PPC_TM_GPR31
:
1370 i
= id
- KVM_REG_PPC_TM_GPR0
;
1371 vcpu
->arch
.gpr_tm
[i
] = set_reg_val(id
, *val
);
1373 case KVM_REG_PPC_TM_VSR0
... KVM_REG_PPC_TM_VSR63
:
1376 i
= id
- KVM_REG_PPC_TM_VSR0
;
1378 for (j
= 0; j
< TS_FPRWIDTH
; j
++)
1379 vcpu
->arch
.fp_tm
.fpr
[i
][j
] = val
->vsxval
[j
];
1381 if (cpu_has_feature(CPU_FTR_ALTIVEC
))
1382 vcpu
->arch
.vr_tm
.vr
[i
-32] = val
->vval
;
1387 case KVM_REG_PPC_TM_CR
:
1388 vcpu
->arch
.cr_tm
= set_reg_val(id
, *val
);
1390 case KVM_REG_PPC_TM_LR
:
1391 vcpu
->arch
.lr_tm
= set_reg_val(id
, *val
);
1393 case KVM_REG_PPC_TM_CTR
:
1394 vcpu
->arch
.ctr_tm
= set_reg_val(id
, *val
);
1396 case KVM_REG_PPC_TM_FPSCR
:
1397 vcpu
->arch
.fp_tm
.fpscr
= set_reg_val(id
, *val
);
1399 case KVM_REG_PPC_TM_AMR
:
1400 vcpu
->arch
.amr_tm
= set_reg_val(id
, *val
);
1402 case KVM_REG_PPC_TM_PPR
:
1403 vcpu
->arch
.ppr_tm
= set_reg_val(id
, *val
);
1405 case KVM_REG_PPC_TM_VRSAVE
:
1406 vcpu
->arch
.vrsave_tm
= set_reg_val(id
, *val
);
1408 case KVM_REG_PPC_TM_VSCR
:
1409 if (cpu_has_feature(CPU_FTR_ALTIVEC
))
1410 vcpu
->arch
.vr
.vscr
.u
[3] = set_reg_val(id
, *val
);
1414 case KVM_REG_PPC_TM_DSCR
:
1415 vcpu
->arch
.dscr_tm
= set_reg_val(id
, *val
);
1417 case KVM_REG_PPC_TM_TAR
:
1418 vcpu
->arch
.tar_tm
= set_reg_val(id
, *val
);
1421 case KVM_REG_PPC_ARCH_COMPAT
:
1422 r
= kvmppc_set_arch_compat(vcpu
, set_reg_val(id
, *val
));
1432 static struct kvmppc_vcore
*kvmppc_vcore_create(struct kvm
*kvm
, int core
)
1434 struct kvmppc_vcore
*vcore
;
1436 vcore
= kzalloc(sizeof(struct kvmppc_vcore
), GFP_KERNEL
);
1441 INIT_LIST_HEAD(&vcore
->runnable_threads
);
1442 spin_lock_init(&vcore
->lock
);
1443 spin_lock_init(&vcore
->stoltb_lock
);
1444 init_waitqueue_head(&vcore
->wq
);
1445 vcore
->preempt_tb
= TB_NIL
;
1446 vcore
->lpcr
= kvm
->arch
.lpcr
;
1447 vcore
->first_vcpuid
= core
* threads_per_subcore
;
1449 INIT_LIST_HEAD(&vcore
->preempt_list
);
1454 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1455 static struct debugfs_timings_element
{
1459 {"rm_entry", offsetof(struct kvm_vcpu
, arch
.rm_entry
)},
1460 {"rm_intr", offsetof(struct kvm_vcpu
, arch
.rm_intr
)},
1461 {"rm_exit", offsetof(struct kvm_vcpu
, arch
.rm_exit
)},
1462 {"guest", offsetof(struct kvm_vcpu
, arch
.guest_time
)},
1463 {"cede", offsetof(struct kvm_vcpu
, arch
.cede_time
)},
1466 #define N_TIMINGS (sizeof(timings) / sizeof(timings[0]))
1468 struct debugfs_timings_state
{
1469 struct kvm_vcpu
*vcpu
;
1470 unsigned int buflen
;
1471 char buf
[N_TIMINGS
* 100];
1474 static int debugfs_timings_open(struct inode
*inode
, struct file
*file
)
1476 struct kvm_vcpu
*vcpu
= inode
->i_private
;
1477 struct debugfs_timings_state
*p
;
1479 p
= kzalloc(sizeof(*p
), GFP_KERNEL
);
1483 kvm_get_kvm(vcpu
->kvm
);
1485 file
->private_data
= p
;
1487 return nonseekable_open(inode
, file
);
1490 static int debugfs_timings_release(struct inode
*inode
, struct file
*file
)
1492 struct debugfs_timings_state
*p
= file
->private_data
;
1494 kvm_put_kvm(p
->vcpu
->kvm
);
1499 static ssize_t
debugfs_timings_read(struct file
*file
, char __user
*buf
,
1500 size_t len
, loff_t
*ppos
)
1502 struct debugfs_timings_state
*p
= file
->private_data
;
1503 struct kvm_vcpu
*vcpu
= p
->vcpu
;
1505 struct kvmhv_tb_accumulator tb
;
1514 buf_end
= s
+ sizeof(p
->buf
);
1515 for (i
= 0; i
< N_TIMINGS
; ++i
) {
1516 struct kvmhv_tb_accumulator
*acc
;
1518 acc
= (struct kvmhv_tb_accumulator
*)
1519 ((unsigned long)vcpu
+ timings
[i
].offset
);
1521 for (loops
= 0; loops
< 1000; ++loops
) {
1522 count
= acc
->seqcount
;
1527 if (count
== acc
->seqcount
) {
1535 snprintf(s
, buf_end
- s
, "%s: stuck\n",
1538 snprintf(s
, buf_end
- s
,
1539 "%s: %llu %llu %llu %llu\n",
1540 timings
[i
].name
, count
/ 2,
1541 tb_to_ns(tb
.tb_total
),
1542 tb_to_ns(tb
.tb_min
),
1543 tb_to_ns(tb
.tb_max
));
1546 p
->buflen
= s
- p
->buf
;
1550 if (pos
>= p
->buflen
)
1552 if (len
> p
->buflen
- pos
)
1553 len
= p
->buflen
- pos
;
1554 n
= copy_to_user(buf
, p
->buf
+ pos
, len
);
1564 static ssize_t
debugfs_timings_write(struct file
*file
, const char __user
*buf
,
1565 size_t len
, loff_t
*ppos
)
1570 static const struct file_operations debugfs_timings_ops
= {
1571 .owner
= THIS_MODULE
,
1572 .open
= debugfs_timings_open
,
1573 .release
= debugfs_timings_release
,
1574 .read
= debugfs_timings_read
,
1575 .write
= debugfs_timings_write
,
1576 .llseek
= generic_file_llseek
,
1579 /* Create a debugfs directory for the vcpu */
1580 static void debugfs_vcpu_init(struct kvm_vcpu
*vcpu
, unsigned int id
)
1583 struct kvm
*kvm
= vcpu
->kvm
;
1585 snprintf(buf
, sizeof(buf
), "vcpu%u", id
);
1586 if (IS_ERR_OR_NULL(kvm
->arch
.debugfs_dir
))
1588 vcpu
->arch
.debugfs_dir
= debugfs_create_dir(buf
, kvm
->arch
.debugfs_dir
);
1589 if (IS_ERR_OR_NULL(vcpu
->arch
.debugfs_dir
))
1591 vcpu
->arch
.debugfs_timings
=
1592 debugfs_create_file("timings", 0444, vcpu
->arch
.debugfs_dir
,
1593 vcpu
, &debugfs_timings_ops
);
1596 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1597 static void debugfs_vcpu_init(struct kvm_vcpu
*vcpu
, unsigned int id
)
1600 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1602 static struct kvm_vcpu
*kvmppc_core_vcpu_create_hv(struct kvm
*kvm
,
1605 struct kvm_vcpu
*vcpu
;
1608 struct kvmppc_vcore
*vcore
;
1610 core
= id
/ threads_per_subcore
;
1611 if (core
>= KVM_MAX_VCORES
)
1615 vcpu
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL
);
1619 err
= kvm_vcpu_init(vcpu
, kvm
, id
);
1623 vcpu
->arch
.shared
= &vcpu
->arch
.shregs
;
1624 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1626 * The shared struct is never shared on HV,
1627 * so we can always use host endianness
1629 #ifdef __BIG_ENDIAN__
1630 vcpu
->arch
.shared_big_endian
= true;
1632 vcpu
->arch
.shared_big_endian
= false;
1635 vcpu
->arch
.mmcr
[0] = MMCR0_FC
;
1636 vcpu
->arch
.ctrl
= CTRL_RUNLATCH
;
1637 /* default to host PVR, since we can't spoof it */
1638 kvmppc_set_pvr_hv(vcpu
, mfspr(SPRN_PVR
));
1639 spin_lock_init(&vcpu
->arch
.vpa_update_lock
);
1640 spin_lock_init(&vcpu
->arch
.tbacct_lock
);
1641 vcpu
->arch
.busy_preempt
= TB_NIL
;
1642 vcpu
->arch
.intr_msr
= MSR_SF
| MSR_ME
;
1644 kvmppc_mmu_book3s_hv_init(vcpu
);
1646 vcpu
->arch
.state
= KVMPPC_VCPU_NOTREADY
;
1648 init_waitqueue_head(&vcpu
->arch
.cpu_run
);
1650 mutex_lock(&kvm
->lock
);
1651 vcore
= kvm
->arch
.vcores
[core
];
1653 vcore
= kvmppc_vcore_create(kvm
, core
);
1654 kvm
->arch
.vcores
[core
] = vcore
;
1655 kvm
->arch
.online_vcores
++;
1657 mutex_unlock(&kvm
->lock
);
1662 spin_lock(&vcore
->lock
);
1663 ++vcore
->num_threads
;
1664 spin_unlock(&vcore
->lock
);
1665 vcpu
->arch
.vcore
= vcore
;
1666 vcpu
->arch
.ptid
= vcpu
->vcpu_id
- vcore
->first_vcpuid
;
1667 vcpu
->arch
.thread_cpu
= -1;
1669 vcpu
->arch
.cpu_type
= KVM_CPU_3S_64
;
1670 kvmppc_sanity_check(vcpu
);
1672 debugfs_vcpu_init(vcpu
, id
);
1677 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
1679 return ERR_PTR(err
);
1682 static void unpin_vpa(struct kvm
*kvm
, struct kvmppc_vpa
*vpa
)
1684 if (vpa
->pinned_addr
)
1685 kvmppc_unpin_guest_page(kvm
, vpa
->pinned_addr
, vpa
->gpa
,
1689 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu
*vcpu
)
1691 spin_lock(&vcpu
->arch
.vpa_update_lock
);
1692 unpin_vpa(vcpu
->kvm
, &vcpu
->arch
.dtl
);
1693 unpin_vpa(vcpu
->kvm
, &vcpu
->arch
.slb_shadow
);
1694 unpin_vpa(vcpu
->kvm
, &vcpu
->arch
.vpa
);
1695 spin_unlock(&vcpu
->arch
.vpa_update_lock
);
1696 kvm_vcpu_uninit(vcpu
);
1697 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
1700 static int kvmppc_core_check_requests_hv(struct kvm_vcpu
*vcpu
)
1702 /* Indicate we want to get back into the guest */
1706 static void kvmppc_set_timer(struct kvm_vcpu
*vcpu
)
1708 unsigned long dec_nsec
, now
;
1711 if (now
> vcpu
->arch
.dec_expires
) {
1712 /* decrementer has already gone negative */
1713 kvmppc_core_queue_dec(vcpu
);
1714 kvmppc_core_prepare_to_enter(vcpu
);
1717 dec_nsec
= (vcpu
->arch
.dec_expires
- now
) * NSEC_PER_SEC
1719 hrtimer_start(&vcpu
->arch
.dec_timer
, ktime_set(0, dec_nsec
),
1721 vcpu
->arch
.timer_running
= 1;
1724 static void kvmppc_end_cede(struct kvm_vcpu
*vcpu
)
1726 vcpu
->arch
.ceded
= 0;
1727 if (vcpu
->arch
.timer_running
) {
1728 hrtimer_try_to_cancel(&vcpu
->arch
.dec_timer
);
1729 vcpu
->arch
.timer_running
= 0;
1733 extern void __kvmppc_vcore_entry(void);
1735 static void kvmppc_remove_runnable(struct kvmppc_vcore
*vc
,
1736 struct kvm_vcpu
*vcpu
)
1740 if (vcpu
->arch
.state
!= KVMPPC_VCPU_RUNNABLE
)
1742 spin_lock_irq(&vcpu
->arch
.tbacct_lock
);
1744 vcpu
->arch
.busy_stolen
+= vcore_stolen_time(vc
, now
) -
1745 vcpu
->arch
.stolen_logged
;
1746 vcpu
->arch
.busy_preempt
= now
;
1747 vcpu
->arch
.state
= KVMPPC_VCPU_BUSY_IN_HOST
;
1748 spin_unlock_irq(&vcpu
->arch
.tbacct_lock
);
1750 list_del(&vcpu
->arch
.run_list
);
1753 static int kvmppc_grab_hwthread(int cpu
)
1755 struct paca_struct
*tpaca
;
1756 long timeout
= 10000;
1760 /* Ensure the thread won't go into the kernel if it wakes */
1761 tpaca
->kvm_hstate
.kvm_vcpu
= NULL
;
1762 tpaca
->kvm_hstate
.kvm_vcore
= NULL
;
1763 tpaca
->kvm_hstate
.napping
= 0;
1765 tpaca
->kvm_hstate
.hwthread_req
= 1;
1768 * If the thread is already executing in the kernel (e.g. handling
1769 * a stray interrupt), wait for it to get back to nap mode.
1770 * The smp_mb() is to ensure that our setting of hwthread_req
1771 * is visible before we look at hwthread_state, so if this
1772 * races with the code at system_reset_pSeries and the thread
1773 * misses our setting of hwthread_req, we are sure to see its
1774 * setting of hwthread_state, and vice versa.
1777 while (tpaca
->kvm_hstate
.hwthread_state
== KVM_HWTHREAD_IN_KERNEL
) {
1778 if (--timeout
<= 0) {
1779 pr_err("KVM: couldn't grab cpu %d\n", cpu
);
1787 static void kvmppc_release_hwthread(int cpu
)
1789 struct paca_struct
*tpaca
;
1792 tpaca
->kvm_hstate
.hwthread_req
= 0;
1793 tpaca
->kvm_hstate
.kvm_vcpu
= NULL
;
1794 tpaca
->kvm_hstate
.kvm_vcore
= NULL
;
1795 tpaca
->kvm_hstate
.kvm_split_mode
= NULL
;
1798 static void kvmppc_start_thread(struct kvm_vcpu
*vcpu
, struct kvmppc_vcore
*vc
)
1801 struct paca_struct
*tpaca
;
1802 struct kvmppc_vcore
*mvc
= vc
->master_vcore
;
1806 if (vcpu
->arch
.timer_running
) {
1807 hrtimer_try_to_cancel(&vcpu
->arch
.dec_timer
);
1808 vcpu
->arch
.timer_running
= 0;
1810 cpu
+= vcpu
->arch
.ptid
;
1811 vcpu
->cpu
= mvc
->pcpu
;
1812 vcpu
->arch
.thread_cpu
= cpu
;
1815 tpaca
->kvm_hstate
.kvm_vcpu
= vcpu
;
1816 tpaca
->kvm_hstate
.ptid
= cpu
- mvc
->pcpu
;
1817 /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
1819 tpaca
->kvm_hstate
.kvm_vcore
= mvc
;
1820 if (cpu
!= smp_processor_id())
1821 kvmppc_ipi_thread(cpu
);
1824 static void kvmppc_wait_for_nap(void)
1826 int cpu
= smp_processor_id();
1829 for (loops
= 0; loops
< 1000000; ++loops
) {
1831 * Check if all threads are finished.
1832 * We set the vcore pointer when starting a thread
1833 * and the thread clears it when finished, so we look
1834 * for any threads that still have a non-NULL vcore ptr.
1836 for (i
= 1; i
< threads_per_subcore
; ++i
)
1837 if (paca
[cpu
+ i
].kvm_hstate
.kvm_vcore
)
1839 if (i
== threads_per_subcore
) {
1846 for (i
= 1; i
< threads_per_subcore
; ++i
)
1847 if (paca
[cpu
+ i
].kvm_hstate
.kvm_vcore
)
1848 pr_err("KVM: CPU %d seems to be stuck\n", cpu
+ i
);
1852 * Check that we are on thread 0 and that any other threads in
1853 * this core are off-line. Then grab the threads so they can't
1856 static int on_primary_thread(void)
1858 int cpu
= smp_processor_id();
1861 /* Are we on a primary subcore? */
1862 if (cpu_thread_in_subcore(cpu
))
1866 while (++thr
< threads_per_subcore
)
1867 if (cpu_online(cpu
+ thr
))
1870 /* Grab all hw threads so they can't go into the kernel */
1871 for (thr
= 1; thr
< threads_per_subcore
; ++thr
) {
1872 if (kvmppc_grab_hwthread(cpu
+ thr
)) {
1873 /* Couldn't grab one; let the others go */
1875 kvmppc_release_hwthread(cpu
+ thr
);
1876 } while (--thr
> 0);
1884 * A list of virtual cores for each physical CPU.
1885 * These are vcores that could run but their runner VCPU tasks are
1886 * (or may be) preempted.
1888 struct preempted_vcore_list
{
1889 struct list_head list
;
1893 static DEFINE_PER_CPU(struct preempted_vcore_list
, preempted_vcores
);
1895 static void init_vcore_lists(void)
1899 for_each_possible_cpu(cpu
) {
1900 struct preempted_vcore_list
*lp
= &per_cpu(preempted_vcores
, cpu
);
1901 spin_lock_init(&lp
->lock
);
1902 INIT_LIST_HEAD(&lp
->list
);
1906 static void kvmppc_vcore_preempt(struct kvmppc_vcore
*vc
)
1908 struct preempted_vcore_list
*lp
= this_cpu_ptr(&preempted_vcores
);
1910 vc
->vcore_state
= VCORE_PREEMPT
;
1911 vc
->pcpu
= smp_processor_id();
1912 if (vc
->num_threads
< threads_per_subcore
) {
1913 spin_lock(&lp
->lock
);
1914 list_add_tail(&vc
->preempt_list
, &lp
->list
);
1915 spin_unlock(&lp
->lock
);
1918 /* Start accumulating stolen time */
1919 kvmppc_core_start_stolen(vc
);
1922 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore
*vc
)
1924 struct preempted_vcore_list
*lp
;
1926 kvmppc_core_end_stolen(vc
);
1927 if (!list_empty(&vc
->preempt_list
)) {
1928 lp
= &per_cpu(preempted_vcores
, vc
->pcpu
);
1929 spin_lock(&lp
->lock
);
1930 list_del_init(&vc
->preempt_list
);
1931 spin_unlock(&lp
->lock
);
1933 vc
->vcore_state
= VCORE_INACTIVE
;
1937 * This stores information about the virtual cores currently
1938 * assigned to a physical core.
1942 int max_subcore_threads
;
1944 int subcore_threads
[MAX_SUBCORES
];
1945 struct kvm
*subcore_vm
[MAX_SUBCORES
];
1946 struct list_head vcs
[MAX_SUBCORES
];
1950 * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
1951 * respectively in 2-way micro-threading (split-core) mode.
1953 static int subcore_thread_map
[MAX_SUBCORES
] = { 0, 4, 2, 6 };
1955 static void init_core_info(struct core_info
*cip
, struct kvmppc_vcore
*vc
)
1959 memset(cip
, 0, sizeof(*cip
));
1960 cip
->n_subcores
= 1;
1961 cip
->max_subcore_threads
= vc
->num_threads
;
1962 cip
->total_threads
= vc
->num_threads
;
1963 cip
->subcore_threads
[0] = vc
->num_threads
;
1964 cip
->subcore_vm
[0] = vc
->kvm
;
1965 for (sub
= 0; sub
< MAX_SUBCORES
; ++sub
)
1966 INIT_LIST_HEAD(&cip
->vcs
[sub
]);
1967 list_add_tail(&vc
->preempt_list
, &cip
->vcs
[0]);
1970 static bool subcore_config_ok(int n_subcores
, int n_threads
)
1972 /* Can only dynamically split if unsplit to begin with */
1973 if (n_subcores
> 1 && threads_per_subcore
< MAX_SMT_THREADS
)
1975 if (n_subcores
> MAX_SUBCORES
)
1977 if (n_subcores
> 1) {
1978 if (!(dynamic_mt_modes
& 2))
1980 if (n_subcores
> 2 && !(dynamic_mt_modes
& 4))
1984 return n_subcores
* roundup_pow_of_two(n_threads
) <= MAX_SMT_THREADS
;
1987 static void init_master_vcore(struct kvmppc_vcore
*vc
)
1989 vc
->master_vcore
= vc
;
1990 vc
->entry_exit_map
= 0;
1992 vc
->napping_threads
= 0;
1993 vc
->conferring_threads
= 0;
1997 * See if the existing subcores can be split into 3 (or fewer) subcores
1998 * of at most two threads each, so we can fit in another vcore. This
1999 * assumes there are at most two subcores and at most 6 threads in total.
2001 static bool can_split_piggybacked_subcores(struct core_info
*cip
)
2006 int n_subcores
= cip
->n_subcores
;
2007 struct kvmppc_vcore
*vc
, *vcnext
;
2008 struct kvmppc_vcore
*master_vc
= NULL
;
2010 for (sub
= 0; sub
< cip
->n_subcores
; ++sub
) {
2011 if (cip
->subcore_threads
[sub
] <= 2)
2016 vc
= list_first_entry(&cip
->vcs
[sub
], struct kvmppc_vcore
,
2018 if (vc
->num_threads
> 2)
2020 n_subcores
+= (cip
->subcore_threads
[sub
] - 1) >> 1;
2022 if (n_subcores
> 3 || large_sub
< 0)
2026 * Seems feasible, so go through and move vcores to new subcores.
2027 * Note that when we have two or more vcores in one subcore,
2028 * all those vcores must have only one thread each.
2030 new_sub
= cip
->n_subcores
;
2033 list_for_each_entry_safe(vc
, vcnext
, &cip
->vcs
[sub
], preempt_list
) {
2035 list_del(&vc
->preempt_list
);
2036 list_add_tail(&vc
->preempt_list
, &cip
->vcs
[new_sub
]);
2037 /* vc->num_threads must be 1 */
2038 if (++cip
->subcore_threads
[new_sub
] == 1) {
2039 cip
->subcore_vm
[new_sub
] = vc
->kvm
;
2040 init_master_vcore(vc
);
2044 vc
->master_vcore
= master_vc
;
2048 thr
+= vc
->num_threads
;
2050 cip
->subcore_threads
[large_sub
] = 2;
2051 cip
->max_subcore_threads
= 2;
2056 static bool can_dynamic_split(struct kvmppc_vcore
*vc
, struct core_info
*cip
)
2058 int n_threads
= vc
->num_threads
;
2061 if (!cpu_has_feature(CPU_FTR_ARCH_207S
))
2064 if (n_threads
< cip
->max_subcore_threads
)
2065 n_threads
= cip
->max_subcore_threads
;
2066 if (subcore_config_ok(cip
->n_subcores
+ 1, n_threads
)) {
2067 cip
->max_subcore_threads
= n_threads
;
2068 } else if (cip
->n_subcores
<= 2 && cip
->total_threads
<= 6 &&
2069 vc
->num_threads
<= 2) {
2071 * We may be able to fit another subcore in by
2072 * splitting an existing subcore with 3 or 4
2073 * threads into two 2-thread subcores, or one
2074 * with 5 or 6 threads into three subcores.
2075 * We can only do this if those subcores have
2076 * piggybacked virtual cores.
2078 if (!can_split_piggybacked_subcores(cip
))
2084 sub
= cip
->n_subcores
;
2086 cip
->total_threads
+= vc
->num_threads
;
2087 cip
->subcore_threads
[sub
] = vc
->num_threads
;
2088 cip
->subcore_vm
[sub
] = vc
->kvm
;
2089 init_master_vcore(vc
);
2090 list_del(&vc
->preempt_list
);
2091 list_add_tail(&vc
->preempt_list
, &cip
->vcs
[sub
]);
2096 static bool can_piggyback_subcore(struct kvmppc_vcore
*pvc
,
2097 struct core_info
*cip
, int sub
)
2099 struct kvmppc_vcore
*vc
;
2102 vc
= list_first_entry(&cip
->vcs
[sub
], struct kvmppc_vcore
,
2105 /* require same VM and same per-core reg values */
2106 if (pvc
->kvm
!= vc
->kvm
||
2107 pvc
->tb_offset
!= vc
->tb_offset
||
2108 pvc
->pcr
!= vc
->pcr
||
2109 pvc
->lpcr
!= vc
->lpcr
)
2112 /* P8 guest with > 1 thread per core would see wrong TIR value */
2113 if (cpu_has_feature(CPU_FTR_ARCH_207S
) &&
2114 (vc
->num_threads
> 1 || pvc
->num_threads
> 1))
2117 n_thr
= cip
->subcore_threads
[sub
] + pvc
->num_threads
;
2118 if (n_thr
> cip
->max_subcore_threads
) {
2119 if (!subcore_config_ok(cip
->n_subcores
, n_thr
))
2121 cip
->max_subcore_threads
= n_thr
;
2124 cip
->total_threads
+= pvc
->num_threads
;
2125 cip
->subcore_threads
[sub
] = n_thr
;
2126 pvc
->master_vcore
= vc
;
2127 list_del(&pvc
->preempt_list
);
2128 list_add_tail(&pvc
->preempt_list
, &cip
->vcs
[sub
]);
2134 * Work out whether it is possible to piggyback the execution of
2135 * vcore *pvc onto the execution of the other vcores described in *cip.
2137 static bool can_piggyback(struct kvmppc_vcore
*pvc
, struct core_info
*cip
,
2142 if (cip
->total_threads
+ pvc
->num_threads
> target_threads
)
2144 for (sub
= 0; sub
< cip
->n_subcores
; ++sub
)
2145 if (cip
->subcore_threads
[sub
] &&
2146 can_piggyback_subcore(pvc
, cip
, sub
))
2149 if (can_dynamic_split(pvc
, cip
))
2155 static void prepare_threads(struct kvmppc_vcore
*vc
)
2157 struct kvm_vcpu
*vcpu
, *vnext
;
2159 list_for_each_entry_safe(vcpu
, vnext
, &vc
->runnable_threads
,
2161 if (signal_pending(vcpu
->arch
.run_task
))
2162 vcpu
->arch
.ret
= -EINTR
;
2163 else if (vcpu
->arch
.vpa
.update_pending
||
2164 vcpu
->arch
.slb_shadow
.update_pending
||
2165 vcpu
->arch
.dtl
.update_pending
)
2166 vcpu
->arch
.ret
= RESUME_GUEST
;
2169 kvmppc_remove_runnable(vc
, vcpu
);
2170 wake_up(&vcpu
->arch
.cpu_run
);
2174 static void collect_piggybacks(struct core_info
*cip
, int target_threads
)
2176 struct preempted_vcore_list
*lp
= this_cpu_ptr(&preempted_vcores
);
2177 struct kvmppc_vcore
*pvc
, *vcnext
;
2179 spin_lock(&lp
->lock
);
2180 list_for_each_entry_safe(pvc
, vcnext
, &lp
->list
, preempt_list
) {
2181 if (!spin_trylock(&pvc
->lock
))
2183 prepare_threads(pvc
);
2184 if (!pvc
->n_runnable
) {
2185 list_del_init(&pvc
->preempt_list
);
2186 if (pvc
->runner
== NULL
) {
2187 pvc
->vcore_state
= VCORE_INACTIVE
;
2188 kvmppc_core_end_stolen(pvc
);
2190 spin_unlock(&pvc
->lock
);
2193 if (!can_piggyback(pvc
, cip
, target_threads
)) {
2194 spin_unlock(&pvc
->lock
);
2197 kvmppc_core_end_stolen(pvc
);
2198 pvc
->vcore_state
= VCORE_PIGGYBACK
;
2199 if (cip
->total_threads
>= target_threads
)
2202 spin_unlock(&lp
->lock
);
2205 static void post_guest_process(struct kvmppc_vcore
*vc
, bool is_master
)
2207 int still_running
= 0;
2210 struct kvm_vcpu
*vcpu
, *vnext
;
2212 spin_lock(&vc
->lock
);
2214 list_for_each_entry_safe(vcpu
, vnext
, &vc
->runnable_threads
,
2216 /* cancel pending dec exception if dec is positive */
2217 if (now
< vcpu
->arch
.dec_expires
&&
2218 kvmppc_core_pending_dec(vcpu
))
2219 kvmppc_core_dequeue_dec(vcpu
);
2221 trace_kvm_guest_exit(vcpu
);
2224 if (vcpu
->arch
.trap
)
2225 ret
= kvmppc_handle_exit_hv(vcpu
->arch
.kvm_run
, vcpu
,
2226 vcpu
->arch
.run_task
);
2228 vcpu
->arch
.ret
= ret
;
2229 vcpu
->arch
.trap
= 0;
2231 if (is_kvmppc_resume_guest(vcpu
->arch
.ret
)) {
2232 if (vcpu
->arch
.pending_exceptions
)
2233 kvmppc_core_prepare_to_enter(vcpu
);
2234 if (vcpu
->arch
.ceded
)
2235 kvmppc_set_timer(vcpu
);
2239 kvmppc_remove_runnable(vc
, vcpu
);
2240 wake_up(&vcpu
->arch
.cpu_run
);
2243 list_del_init(&vc
->preempt_list
);
2245 if (still_running
> 0) {
2246 kvmppc_vcore_preempt(vc
);
2247 } else if (vc
->runner
) {
2248 vc
->vcore_state
= VCORE_PREEMPT
;
2249 kvmppc_core_start_stolen(vc
);
2251 vc
->vcore_state
= VCORE_INACTIVE
;
2253 if (vc
->n_runnable
> 0 && vc
->runner
== NULL
) {
2254 /* make sure there's a candidate runner awake */
2255 vcpu
= list_first_entry(&vc
->runnable_threads
,
2256 struct kvm_vcpu
, arch
.run_list
);
2257 wake_up(&vcpu
->arch
.cpu_run
);
2260 spin_unlock(&vc
->lock
);
2264 * Run a set of guest threads on a physical core.
2265 * Called with vc->lock held.
2267 static noinline
void kvmppc_run_core(struct kvmppc_vcore
*vc
)
2269 struct kvm_vcpu
*vcpu
, *vnext
;
2272 struct core_info core_info
;
2273 struct kvmppc_vcore
*pvc
, *vcnext
;
2274 struct kvm_split_mode split_info
, *sip
;
2275 int split
, subcore_size
, active
;
2278 unsigned long cmd_bit
, stat_bit
;
2283 * Remove from the list any threads that have a signal pending
2284 * or need a VPA update done
2286 prepare_threads(vc
);
2288 /* if the runner is no longer runnable, let the caller pick a new one */
2289 if (vc
->runner
->arch
.state
!= KVMPPC_VCPU_RUNNABLE
)
2295 init_master_vcore(vc
);
2296 vc
->preempt_tb
= TB_NIL
;
2299 * Make sure we are running on primary threads, and that secondary
2300 * threads are offline. Also check if the number of threads in this
2301 * guest are greater than the current system threads per guest.
2303 if ((threads_per_core
> 1) &&
2304 ((vc
->num_threads
> threads_per_subcore
) || !on_primary_thread())) {
2305 list_for_each_entry_safe(vcpu
, vnext
, &vc
->runnable_threads
,
2307 vcpu
->arch
.ret
= -EBUSY
;
2308 kvmppc_remove_runnable(vc
, vcpu
);
2309 wake_up(&vcpu
->arch
.cpu_run
);
2315 * See if we could run any other vcores on the physical core
2316 * along with this one.
2318 init_core_info(&core_info
, vc
);
2319 pcpu
= smp_processor_id();
2320 target_threads
= threads_per_subcore
;
2321 if (target_smt_mode
&& target_smt_mode
< target_threads
)
2322 target_threads
= target_smt_mode
;
2323 if (vc
->num_threads
< target_threads
)
2324 collect_piggybacks(&core_info
, target_threads
);
2326 /* Decide on micro-threading (split-core) mode */
2327 subcore_size
= threads_per_subcore
;
2328 cmd_bit
= stat_bit
= 0;
2329 split
= core_info
.n_subcores
;
2332 /* threads_per_subcore must be MAX_SMT_THREADS (8) here */
2333 if (split
== 2 && (dynamic_mt_modes
& 2)) {
2334 cmd_bit
= HID0_POWER8_1TO2LPAR
;
2335 stat_bit
= HID0_POWER8_2LPARMODE
;
2338 cmd_bit
= HID0_POWER8_1TO4LPAR
;
2339 stat_bit
= HID0_POWER8_4LPARMODE
;
2341 subcore_size
= MAX_SMT_THREADS
/ split
;
2343 memset(&split_info
, 0, sizeof(split_info
));
2344 split_info
.rpr
= mfspr(SPRN_RPR
);
2345 split_info
.pmmar
= mfspr(SPRN_PMMAR
);
2346 split_info
.ldbar
= mfspr(SPRN_LDBAR
);
2347 split_info
.subcore_size
= subcore_size
;
2348 for (sub
= 0; sub
< core_info
.n_subcores
; ++sub
)
2349 split_info
.master_vcs
[sub
] =
2350 list_first_entry(&core_info
.vcs
[sub
],
2351 struct kvmppc_vcore
, preempt_list
);
2352 /* order writes to split_info before kvm_split_mode pointer */
2355 pcpu
= smp_processor_id();
2356 for (thr
= 0; thr
< threads_per_subcore
; ++thr
)
2357 paca
[pcpu
+ thr
].kvm_hstate
.kvm_split_mode
= sip
;
2359 /* Initiate micro-threading (split-core) if required */
2361 unsigned long hid0
= mfspr(SPRN_HID0
);
2363 hid0
|= cmd_bit
| HID0_POWER8_DYNLPARDIS
;
2365 mtspr(SPRN_HID0
, hid0
);
2368 hid0
= mfspr(SPRN_HID0
);
2369 if (hid0
& stat_bit
)
2375 /* Start all the threads */
2377 for (sub
= 0; sub
< core_info
.n_subcores
; ++sub
) {
2378 thr
= subcore_thread_map
[sub
];
2381 list_for_each_entry(pvc
, &core_info
.vcs
[sub
], preempt_list
) {
2382 pvc
->pcpu
= pcpu
+ thr
;
2383 list_for_each_entry(vcpu
, &pvc
->runnable_threads
,
2385 kvmppc_start_thread(vcpu
, pvc
);
2386 kvmppc_create_dtl_entry(vcpu
, pvc
);
2387 trace_kvm_guest_enter(vcpu
);
2388 if (!vcpu
->arch
.ptid
)
2390 active
|= 1 << (thr
+ vcpu
->arch
.ptid
);
2393 * We need to start the first thread of each subcore
2394 * even if it doesn't have a vcpu.
2396 if (pvc
->master_vcore
== pvc
&& !thr0_done
)
2397 kvmppc_start_thread(NULL
, pvc
);
2398 thr
+= pvc
->num_threads
;
2403 * Ensure that split_info.do_nap is set after setting
2404 * the vcore pointer in the PACA of the secondaries.
2408 split_info
.do_nap
= 1; /* ask secondaries to nap when done */
2411 * When doing micro-threading, poke the inactive threads as well.
2412 * This gets them to the nap instruction after kvm_do_nap,
2413 * which reduces the time taken to unsplit later.
2416 for (thr
= 1; thr
< threads_per_subcore
; ++thr
)
2417 if (!(active
& (1 << thr
)))
2418 kvmppc_ipi_thread(pcpu
+ thr
);
2420 vc
->vcore_state
= VCORE_RUNNING
;
2423 trace_kvmppc_run_core(vc
, 0);
2425 for (sub
= 0; sub
< core_info
.n_subcores
; ++sub
)
2426 list_for_each_entry(pvc
, &core_info
.vcs
[sub
], preempt_list
)
2427 spin_unlock(&pvc
->lock
);
2431 srcu_idx
= srcu_read_lock(&vc
->kvm
->srcu
);
2433 __kvmppc_vcore_entry();
2435 srcu_read_unlock(&vc
->kvm
->srcu
, srcu_idx
);
2437 spin_lock(&vc
->lock
);
2438 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
2439 vc
->vcore_state
= VCORE_EXITING
;
2441 /* wait for secondary threads to finish writing their state to memory */
2442 kvmppc_wait_for_nap();
2444 /* Return to whole-core mode if we split the core earlier */
2446 unsigned long hid0
= mfspr(SPRN_HID0
);
2447 unsigned long loops
= 0;
2449 hid0
&= ~HID0_POWER8_DYNLPARDIS
;
2450 stat_bit
= HID0_POWER8_2LPARMODE
| HID0_POWER8_4LPARMODE
;
2452 mtspr(SPRN_HID0
, hid0
);
2455 hid0
= mfspr(SPRN_HID0
);
2456 if (!(hid0
& stat_bit
))
2461 split_info
.do_nap
= 0;
2464 /* Let secondaries go back to the offline loop */
2465 for (i
= 0; i
< threads_per_subcore
; ++i
) {
2466 kvmppc_release_hwthread(pcpu
+ i
);
2467 if (sip
&& sip
->napped
[i
])
2468 kvmppc_ipi_thread(pcpu
+ i
);
2471 spin_unlock(&vc
->lock
);
2473 /* make sure updates to secondary vcpu structs are visible now */
2477 for (sub
= 0; sub
< core_info
.n_subcores
; ++sub
)
2478 list_for_each_entry_safe(pvc
, vcnext
, &core_info
.vcs
[sub
],
2480 post_guest_process(pvc
, pvc
== vc
);
2482 spin_lock(&vc
->lock
);
2486 vc
->vcore_state
= VCORE_INACTIVE
;
2487 trace_kvmppc_run_core(vc
, 1);
2491 * Wait for some other vcpu thread to execute us, and
2492 * wake us up when we need to handle something in the host.
2494 static void kvmppc_wait_for_exec(struct kvmppc_vcore
*vc
,
2495 struct kvm_vcpu
*vcpu
, int wait_state
)
2499 prepare_to_wait(&vcpu
->arch
.cpu_run
, &wait
, wait_state
);
2500 if (vcpu
->arch
.state
== KVMPPC_VCPU_RUNNABLE
) {
2501 spin_unlock(&vc
->lock
);
2503 spin_lock(&vc
->lock
);
2505 finish_wait(&vcpu
->arch
.cpu_run
, &wait
);
2509 * All the vcpus in this vcore are idle, so wait for a decrementer
2510 * or external interrupt to one of the vcpus. vc->lock is held.
2512 static void kvmppc_vcore_blocked(struct kvmppc_vcore
*vc
)
2514 struct kvm_vcpu
*vcpu
;
2519 prepare_to_wait(&vc
->wq
, &wait
, TASK_INTERRUPTIBLE
);
2522 * Check one last time for pending exceptions and ceded state after
2523 * we put ourselves on the wait queue
2525 list_for_each_entry(vcpu
, &vc
->runnable_threads
, arch
.run_list
) {
2526 if (vcpu
->arch
.pending_exceptions
|| !vcpu
->arch
.ceded
) {
2533 finish_wait(&vc
->wq
, &wait
);
2537 vc
->vcore_state
= VCORE_SLEEPING
;
2538 trace_kvmppc_vcore_blocked(vc
, 0);
2539 spin_unlock(&vc
->lock
);
2541 finish_wait(&vc
->wq
, &wait
);
2542 spin_lock(&vc
->lock
);
2543 vc
->vcore_state
= VCORE_INACTIVE
;
2544 trace_kvmppc_vcore_blocked(vc
, 1);
2547 static int kvmppc_run_vcpu(struct kvm_run
*kvm_run
, struct kvm_vcpu
*vcpu
)
2550 struct kvmppc_vcore
*vc
;
2551 struct kvm_vcpu
*v
, *vn
;
2553 trace_kvmppc_run_vcpu_enter(vcpu
);
2555 kvm_run
->exit_reason
= 0;
2556 vcpu
->arch
.ret
= RESUME_GUEST
;
2557 vcpu
->arch
.trap
= 0;
2558 kvmppc_update_vpas(vcpu
);
2561 * Synchronize with other threads in this virtual core
2563 vc
= vcpu
->arch
.vcore
;
2564 spin_lock(&vc
->lock
);
2565 vcpu
->arch
.ceded
= 0;
2566 vcpu
->arch
.run_task
= current
;
2567 vcpu
->arch
.kvm_run
= kvm_run
;
2568 vcpu
->arch
.stolen_logged
= vcore_stolen_time(vc
, mftb());
2569 vcpu
->arch
.state
= KVMPPC_VCPU_RUNNABLE
;
2570 vcpu
->arch
.busy_preempt
= TB_NIL
;
2571 list_add_tail(&vcpu
->arch
.run_list
, &vc
->runnable_threads
);
2575 * This happens the first time this is called for a vcpu.
2576 * If the vcore is already running, we may be able to start
2577 * this thread straight away and have it join in.
2579 if (!signal_pending(current
)) {
2580 if (vc
->vcore_state
== VCORE_PIGGYBACK
) {
2581 struct kvmppc_vcore
*mvc
= vc
->master_vcore
;
2582 if (spin_trylock(&mvc
->lock
)) {
2583 if (mvc
->vcore_state
== VCORE_RUNNING
&&
2584 !VCORE_IS_EXITING(mvc
)) {
2585 kvmppc_create_dtl_entry(vcpu
, vc
);
2586 kvmppc_start_thread(vcpu
, vc
);
2587 trace_kvm_guest_enter(vcpu
);
2589 spin_unlock(&mvc
->lock
);
2591 } else if (vc
->vcore_state
== VCORE_RUNNING
&&
2592 !VCORE_IS_EXITING(vc
)) {
2593 kvmppc_create_dtl_entry(vcpu
, vc
);
2594 kvmppc_start_thread(vcpu
, vc
);
2595 trace_kvm_guest_enter(vcpu
);
2596 } else if (vc
->vcore_state
== VCORE_SLEEPING
) {
2602 while (vcpu
->arch
.state
== KVMPPC_VCPU_RUNNABLE
&&
2603 !signal_pending(current
)) {
2604 if (vc
->vcore_state
== VCORE_PREEMPT
&& vc
->runner
== NULL
)
2605 kvmppc_vcore_end_preempt(vc
);
2607 if (vc
->vcore_state
!= VCORE_INACTIVE
) {
2608 kvmppc_wait_for_exec(vc
, vcpu
, TASK_INTERRUPTIBLE
);
2611 list_for_each_entry_safe(v
, vn
, &vc
->runnable_threads
,
2613 kvmppc_core_prepare_to_enter(v
);
2614 if (signal_pending(v
->arch
.run_task
)) {
2615 kvmppc_remove_runnable(vc
, v
);
2616 v
->stat
.signal_exits
++;
2617 v
->arch
.kvm_run
->exit_reason
= KVM_EXIT_INTR
;
2618 v
->arch
.ret
= -EINTR
;
2619 wake_up(&v
->arch
.cpu_run
);
2622 if (!vc
->n_runnable
|| vcpu
->arch
.state
!= KVMPPC_VCPU_RUNNABLE
)
2625 list_for_each_entry(v
, &vc
->runnable_threads
, arch
.run_list
) {
2626 if (!v
->arch
.pending_exceptions
)
2627 n_ceded
+= v
->arch
.ceded
;
2632 if (n_ceded
== vc
->n_runnable
) {
2633 kvmppc_vcore_blocked(vc
);
2634 } else if (need_resched()) {
2635 kvmppc_vcore_preempt(vc
);
2636 /* Let something else run */
2637 cond_resched_lock(&vc
->lock
);
2638 if (vc
->vcore_state
== VCORE_PREEMPT
)
2639 kvmppc_vcore_end_preempt(vc
);
2641 kvmppc_run_core(vc
);
2646 while (vcpu
->arch
.state
== KVMPPC_VCPU_RUNNABLE
&&
2647 (vc
->vcore_state
== VCORE_RUNNING
||
2648 vc
->vcore_state
== VCORE_EXITING
||
2649 vc
->vcore_state
== VCORE_PIGGYBACK
))
2650 kvmppc_wait_for_exec(vc
, vcpu
, TASK_UNINTERRUPTIBLE
);
2652 if (vc
->vcore_state
== VCORE_PREEMPT
&& vc
->runner
== NULL
)
2653 kvmppc_vcore_end_preempt(vc
);
2655 if (vcpu
->arch
.state
== KVMPPC_VCPU_RUNNABLE
) {
2656 kvmppc_remove_runnable(vc
, vcpu
);
2657 vcpu
->stat
.signal_exits
++;
2658 kvm_run
->exit_reason
= KVM_EXIT_INTR
;
2659 vcpu
->arch
.ret
= -EINTR
;
2662 if (vc
->n_runnable
&& vc
->vcore_state
== VCORE_INACTIVE
) {
2663 /* Wake up some vcpu to run the core */
2664 v
= list_first_entry(&vc
->runnable_threads
,
2665 struct kvm_vcpu
, arch
.run_list
);
2666 wake_up(&v
->arch
.cpu_run
);
2669 trace_kvmppc_run_vcpu_exit(vcpu
, kvm_run
);
2670 spin_unlock(&vc
->lock
);
2671 return vcpu
->arch
.ret
;
2674 static int kvmppc_vcpu_run_hv(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
)
2679 if (!vcpu
->arch
.sane
) {
2680 run
->exit_reason
= KVM_EXIT_INTERNAL_ERROR
;
2684 kvmppc_core_prepare_to_enter(vcpu
);
2686 /* No need to go into the guest when all we'll do is come back out */
2687 if (signal_pending(current
)) {
2688 run
->exit_reason
= KVM_EXIT_INTR
;
2692 atomic_inc(&vcpu
->kvm
->arch
.vcpus_running
);
2693 /* Order vcpus_running vs. hpte_setup_done, see kvmppc_alloc_reset_hpt */
2696 /* On the first time here, set up HTAB and VRMA */
2697 if (!vcpu
->kvm
->arch
.hpte_setup_done
) {
2698 r
= kvmppc_hv_setup_htab_rma(vcpu
);
2703 flush_fp_to_thread(current
);
2704 flush_altivec_to_thread(current
);
2705 flush_vsx_to_thread(current
);
2706 vcpu
->arch
.wqp
= &vcpu
->arch
.vcore
->wq
;
2707 vcpu
->arch
.pgdir
= current
->mm
->pgd
;
2708 vcpu
->arch
.state
= KVMPPC_VCPU_BUSY_IN_HOST
;
2711 r
= kvmppc_run_vcpu(run
, vcpu
);
2713 if (run
->exit_reason
== KVM_EXIT_PAPR_HCALL
&&
2714 !(vcpu
->arch
.shregs
.msr
& MSR_PR
)) {
2715 trace_kvm_hcall_enter(vcpu
);
2716 r
= kvmppc_pseries_do_hcall(vcpu
);
2717 trace_kvm_hcall_exit(vcpu
, r
);
2718 kvmppc_core_prepare_to_enter(vcpu
);
2719 } else if (r
== RESUME_PAGE_FAULT
) {
2720 srcu_idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
2721 r
= kvmppc_book3s_hv_page_fault(run
, vcpu
,
2722 vcpu
->arch
.fault_dar
, vcpu
->arch
.fault_dsisr
);
2723 srcu_read_unlock(&vcpu
->kvm
->srcu
, srcu_idx
);
2725 } while (is_kvmppc_resume_guest(r
));
2728 vcpu
->arch
.state
= KVMPPC_VCPU_NOTREADY
;
2729 atomic_dec(&vcpu
->kvm
->arch
.vcpus_running
);
2733 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size
**sps
,
2736 struct mmu_psize_def
*def
= &mmu_psize_defs
[linux_psize
];
2740 (*sps
)->page_shift
= def
->shift
;
2741 (*sps
)->slb_enc
= def
->sllp
;
2742 (*sps
)->enc
[0].page_shift
= def
->shift
;
2743 (*sps
)->enc
[0].pte_enc
= def
->penc
[linux_psize
];
2745 * Add 16MB MPSS support if host supports it
2747 if (linux_psize
!= MMU_PAGE_16M
&& def
->penc
[MMU_PAGE_16M
] != -1) {
2748 (*sps
)->enc
[1].page_shift
= 24;
2749 (*sps
)->enc
[1].pte_enc
= def
->penc
[MMU_PAGE_16M
];
2754 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm
*kvm
,
2755 struct kvm_ppc_smmu_info
*info
)
2757 struct kvm_ppc_one_seg_page_size
*sps
;
2759 info
->flags
= KVM_PPC_PAGE_SIZES_REAL
;
2760 if (mmu_has_feature(MMU_FTR_1T_SEGMENT
))
2761 info
->flags
|= KVM_PPC_1T_SEGMENTS
;
2762 info
->slb_size
= mmu_slb_size
;
2764 /* We only support these sizes for now, and no muti-size segments */
2765 sps
= &info
->sps
[0];
2766 kvmppc_add_seg_page_size(&sps
, MMU_PAGE_4K
);
2767 kvmppc_add_seg_page_size(&sps
, MMU_PAGE_64K
);
2768 kvmppc_add_seg_page_size(&sps
, MMU_PAGE_16M
);
2774 * Get (and clear) the dirty memory log for a memory slot.
2776 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm
*kvm
,
2777 struct kvm_dirty_log
*log
)
2779 struct kvm_memslots
*slots
;
2780 struct kvm_memory_slot
*memslot
;
2784 mutex_lock(&kvm
->slots_lock
);
2787 if (log
->slot
>= KVM_USER_MEM_SLOTS
)
2790 slots
= kvm_memslots(kvm
);
2791 memslot
= id_to_memslot(slots
, log
->slot
);
2793 if (!memslot
->dirty_bitmap
)
2796 n
= kvm_dirty_bitmap_bytes(memslot
);
2797 memset(memslot
->dirty_bitmap
, 0, n
);
2799 r
= kvmppc_hv_get_dirty_log(kvm
, memslot
, memslot
->dirty_bitmap
);
2804 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
2809 mutex_unlock(&kvm
->slots_lock
);
2813 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot
*free
,
2814 struct kvm_memory_slot
*dont
)
2816 if (!dont
|| free
->arch
.rmap
!= dont
->arch
.rmap
) {
2817 vfree(free
->arch
.rmap
);
2818 free
->arch
.rmap
= NULL
;
2822 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot
*slot
,
2823 unsigned long npages
)
2825 slot
->arch
.rmap
= vzalloc(npages
* sizeof(*slot
->arch
.rmap
));
2826 if (!slot
->arch
.rmap
)
2832 static int kvmppc_core_prepare_memory_region_hv(struct kvm
*kvm
,
2833 struct kvm_memory_slot
*memslot
,
2834 const struct kvm_userspace_memory_region
*mem
)
2839 static void kvmppc_core_commit_memory_region_hv(struct kvm
*kvm
,
2840 const struct kvm_userspace_memory_region
*mem
,
2841 const struct kvm_memory_slot
*old
,
2842 const struct kvm_memory_slot
*new)
2844 unsigned long npages
= mem
->memory_size
>> PAGE_SHIFT
;
2845 struct kvm_memslots
*slots
;
2846 struct kvm_memory_slot
*memslot
;
2848 if (npages
&& old
->npages
) {
2850 * If modifying a memslot, reset all the rmap dirty bits.
2851 * If this is a new memslot, we don't need to do anything
2852 * since the rmap array starts out as all zeroes,
2853 * i.e. no pages are dirty.
2855 slots
= kvm_memslots(kvm
);
2856 memslot
= id_to_memslot(slots
, mem
->slot
);
2857 kvmppc_hv_get_dirty_log(kvm
, memslot
, NULL
);
2862 * Update LPCR values in kvm->arch and in vcores.
2863 * Caller must hold kvm->lock.
2865 void kvmppc_update_lpcr(struct kvm
*kvm
, unsigned long lpcr
, unsigned long mask
)
2870 if ((kvm
->arch
.lpcr
& mask
) == lpcr
)
2873 kvm
->arch
.lpcr
= (kvm
->arch
.lpcr
& ~mask
) | lpcr
;
2875 for (i
= 0; i
< KVM_MAX_VCORES
; ++i
) {
2876 struct kvmppc_vcore
*vc
= kvm
->arch
.vcores
[i
];
2879 spin_lock(&vc
->lock
);
2880 vc
->lpcr
= (vc
->lpcr
& ~mask
) | lpcr
;
2881 spin_unlock(&vc
->lock
);
2882 if (++cores_done
>= kvm
->arch
.online_vcores
)
2887 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu
*vcpu
)
2892 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu
*vcpu
)
2895 struct kvm
*kvm
= vcpu
->kvm
;
2897 struct kvm_memory_slot
*memslot
;
2898 struct vm_area_struct
*vma
;
2899 unsigned long lpcr
= 0, senc
;
2900 unsigned long psize
, porder
;
2903 mutex_lock(&kvm
->lock
);
2904 if (kvm
->arch
.hpte_setup_done
)
2905 goto out
; /* another vcpu beat us to it */
2907 /* Allocate hashed page table (if not done already) and reset it */
2908 if (!kvm
->arch
.hpt_virt
) {
2909 err
= kvmppc_alloc_hpt(kvm
, NULL
);
2911 pr_err("KVM: Couldn't alloc HPT\n");
2916 /* Look up the memslot for guest physical address 0 */
2917 srcu_idx
= srcu_read_lock(&kvm
->srcu
);
2918 memslot
= gfn_to_memslot(kvm
, 0);
2920 /* We must have some memory at 0 by now */
2922 if (!memslot
|| (memslot
->flags
& KVM_MEMSLOT_INVALID
))
2925 /* Look up the VMA for the start of this memory slot */
2926 hva
= memslot
->userspace_addr
;
2927 down_read(¤t
->mm
->mmap_sem
);
2928 vma
= find_vma(current
->mm
, hva
);
2929 if (!vma
|| vma
->vm_start
> hva
|| (vma
->vm_flags
& VM_IO
))
2932 psize
= vma_kernel_pagesize(vma
);
2933 porder
= __ilog2(psize
);
2935 up_read(¤t
->mm
->mmap_sem
);
2937 /* We can handle 4k, 64k or 16M pages in the VRMA */
2939 if (!(psize
== 0x1000 || psize
== 0x10000 ||
2940 psize
== 0x1000000))
2943 /* Update VRMASD field in the LPCR */
2944 senc
= slb_pgsize_encoding(psize
);
2945 kvm
->arch
.vrma_slb_v
= senc
| SLB_VSID_B_1T
|
2946 (VRMA_VSID
<< SLB_VSID_SHIFT_1T
);
2947 /* the -4 is to account for senc values starting at 0x10 */
2948 lpcr
= senc
<< (LPCR_VRMASD_SH
- 4);
2950 /* Create HPTEs in the hash page table for the VRMA */
2951 kvmppc_map_vrma(vcpu
, memslot
, porder
);
2953 kvmppc_update_lpcr(kvm
, lpcr
, LPCR_VRMASD
);
2955 /* Order updates to kvm->arch.lpcr etc. vs. hpte_setup_done */
2957 kvm
->arch
.hpte_setup_done
= 1;
2960 srcu_read_unlock(&kvm
->srcu
, srcu_idx
);
2962 mutex_unlock(&kvm
->lock
);
2966 up_read(¤t
->mm
->mmap_sem
);
2970 static int kvmppc_core_init_vm_hv(struct kvm
*kvm
)
2972 unsigned long lpcr
, lpid
;
2975 /* Allocate the guest's logical partition ID */
2977 lpid
= kvmppc_alloc_lpid();
2980 kvm
->arch
.lpid
= lpid
;
2983 * Since we don't flush the TLB when tearing down a VM,
2984 * and this lpid might have previously been used,
2985 * make sure we flush on each core before running the new VM.
2987 cpumask_setall(&kvm
->arch
.need_tlb_flush
);
2989 /* Start out with the default set of hcalls enabled */
2990 memcpy(kvm
->arch
.enabled_hcalls
, default_enabled_hcalls
,
2991 sizeof(kvm
->arch
.enabled_hcalls
));
2993 kvm
->arch
.host_sdr1
= mfspr(SPRN_SDR1
);
2995 /* Init LPCR for virtual RMA mode */
2996 kvm
->arch
.host_lpid
= mfspr(SPRN_LPID
);
2997 kvm
->arch
.host_lpcr
= lpcr
= mfspr(SPRN_LPCR
);
2998 lpcr
&= LPCR_PECE
| LPCR_LPES
;
2999 lpcr
|= (4UL << LPCR_DPFD_SH
) | LPCR_HDICE
|
3000 LPCR_VPM0
| LPCR_VPM1
;
3001 kvm
->arch
.vrma_slb_v
= SLB_VSID_B_1T
|
3002 (VRMA_VSID
<< SLB_VSID_SHIFT_1T
);
3003 /* On POWER8 turn on online bit to enable PURR/SPURR */
3004 if (cpu_has_feature(CPU_FTR_ARCH_207S
))
3006 kvm
->arch
.lpcr
= lpcr
;
3009 * Track that we now have a HV mode VM active. This blocks secondary
3010 * CPU threads from coming online.
3012 kvm_hv_vm_activated();
3015 * Create a debugfs directory for the VM
3017 snprintf(buf
, sizeof(buf
), "vm%d", current
->pid
);
3018 kvm
->arch
.debugfs_dir
= debugfs_create_dir(buf
, kvm_debugfs_dir
);
3019 if (!IS_ERR_OR_NULL(kvm
->arch
.debugfs_dir
))
3020 kvmppc_mmu_debugfs_init(kvm
);
3025 static void kvmppc_free_vcores(struct kvm
*kvm
)
3029 for (i
= 0; i
< KVM_MAX_VCORES
; ++i
)
3030 kfree(kvm
->arch
.vcores
[i
]);
3031 kvm
->arch
.online_vcores
= 0;
3034 static void kvmppc_core_destroy_vm_hv(struct kvm
*kvm
)
3036 debugfs_remove_recursive(kvm
->arch
.debugfs_dir
);
3038 kvm_hv_vm_deactivated();
3040 kvmppc_free_vcores(kvm
);
3042 kvmppc_free_hpt(kvm
);
3045 /* We don't need to emulate any privileged instructions or dcbz */
3046 static int kvmppc_core_emulate_op_hv(struct kvm_run
*run
, struct kvm_vcpu
*vcpu
,
3047 unsigned int inst
, int *advance
)
3049 return EMULATE_FAIL
;
3052 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu
*vcpu
, int sprn
,
3055 return EMULATE_FAIL
;
3058 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu
*vcpu
, int sprn
,
3061 return EMULATE_FAIL
;
3064 static int kvmppc_core_check_processor_compat_hv(void)
3066 if (!cpu_has_feature(CPU_FTR_HVMODE
) ||
3067 !cpu_has_feature(CPU_FTR_ARCH_206
))
3072 static long kvm_arch_vm_ioctl_hv(struct file
*filp
,
3073 unsigned int ioctl
, unsigned long arg
)
3075 struct kvm
*kvm __maybe_unused
= filp
->private_data
;
3076 void __user
*argp
= (void __user
*)arg
;
3081 case KVM_PPC_ALLOCATE_HTAB
: {
3085 if (get_user(htab_order
, (u32 __user
*)argp
))
3087 r
= kvmppc_alloc_reset_hpt(kvm
, &htab_order
);
3091 if (put_user(htab_order
, (u32 __user
*)argp
))
3097 case KVM_PPC_GET_HTAB_FD
: {
3098 struct kvm_get_htab_fd ghf
;
3101 if (copy_from_user(&ghf
, argp
, sizeof(ghf
)))
3103 r
= kvm_vm_ioctl_get_htab_fd(kvm
, &ghf
);
3115 * List of hcall numbers to enable by default.
3116 * For compatibility with old userspace, we enable by default
3117 * all hcalls that were implemented before the hcall-enabling
3118 * facility was added. Note this list should not include H_RTAS.
3120 static unsigned int default_hcall_list
[] = {
3134 #ifdef CONFIG_KVM_XICS
3145 static void init_default_hcalls(void)
3150 for (i
= 0; default_hcall_list
[i
]; ++i
) {
3151 hcall
= default_hcall_list
[i
];
3152 WARN_ON(!kvmppc_hcall_impl_hv(hcall
));
3153 __set_bit(hcall
/ 4, default_enabled_hcalls
);
3157 static struct kvmppc_ops kvm_ops_hv
= {
3158 .get_sregs
= kvm_arch_vcpu_ioctl_get_sregs_hv
,
3159 .set_sregs
= kvm_arch_vcpu_ioctl_set_sregs_hv
,
3160 .get_one_reg
= kvmppc_get_one_reg_hv
,
3161 .set_one_reg
= kvmppc_set_one_reg_hv
,
3162 .vcpu_load
= kvmppc_core_vcpu_load_hv
,
3163 .vcpu_put
= kvmppc_core_vcpu_put_hv
,
3164 .set_msr
= kvmppc_set_msr_hv
,
3165 .vcpu_run
= kvmppc_vcpu_run_hv
,
3166 .vcpu_create
= kvmppc_core_vcpu_create_hv
,
3167 .vcpu_free
= kvmppc_core_vcpu_free_hv
,
3168 .check_requests
= kvmppc_core_check_requests_hv
,
3169 .get_dirty_log
= kvm_vm_ioctl_get_dirty_log_hv
,
3170 .flush_memslot
= kvmppc_core_flush_memslot_hv
,
3171 .prepare_memory_region
= kvmppc_core_prepare_memory_region_hv
,
3172 .commit_memory_region
= kvmppc_core_commit_memory_region_hv
,
3173 .unmap_hva
= kvm_unmap_hva_hv
,
3174 .unmap_hva_range
= kvm_unmap_hva_range_hv
,
3175 .age_hva
= kvm_age_hva_hv
,
3176 .test_age_hva
= kvm_test_age_hva_hv
,
3177 .set_spte_hva
= kvm_set_spte_hva_hv
,
3178 .mmu_destroy
= kvmppc_mmu_destroy_hv
,
3179 .free_memslot
= kvmppc_core_free_memslot_hv
,
3180 .create_memslot
= kvmppc_core_create_memslot_hv
,
3181 .init_vm
= kvmppc_core_init_vm_hv
,
3182 .destroy_vm
= kvmppc_core_destroy_vm_hv
,
3183 .get_smmu_info
= kvm_vm_ioctl_get_smmu_info_hv
,
3184 .emulate_op
= kvmppc_core_emulate_op_hv
,
3185 .emulate_mtspr
= kvmppc_core_emulate_mtspr_hv
,
3186 .emulate_mfspr
= kvmppc_core_emulate_mfspr_hv
,
3187 .fast_vcpu_kick
= kvmppc_fast_vcpu_kick_hv
,
3188 .arch_vm_ioctl
= kvm_arch_vm_ioctl_hv
,
3189 .hcall_implemented
= kvmppc_hcall_impl_hv
,
3192 static int kvmppc_book3s_init_hv(void)
3196 * FIXME!! Do we need to check on all cpus ?
3198 r
= kvmppc_core_check_processor_compat_hv();
3202 kvm_ops_hv
.owner
= THIS_MODULE
;
3203 kvmppc_hv_ops
= &kvm_ops_hv
;
3205 init_default_hcalls();
3209 r
= kvmppc_mmu_hv_init();
3213 static void kvmppc_book3s_exit_hv(void)
3215 kvmppc_hv_ops
= NULL
;
3218 module_init(kvmppc_book3s_init_hv
);
3219 module_exit(kvmppc_book3s_exit_hv
);
3220 MODULE_LICENSE("GPL");
3221 MODULE_ALIAS_MISCDEV(KVM_MINOR
);
3222 MODULE_ALIAS("devname:kvm");