]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - arch/powerpc/kvm/book3s_hv.c
projects / mirror_ubuntu-artful-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: PPC: Book3S HV: Handle guest using doorbells for IPIs
[mirror_ubuntu-artful-kernel.git] / arch / powerpc / kvm / book3s_hv.c
CommitLineData
de56a948
PM		 1/*
2 * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
3 * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
4 *
5 * Authors:
6 * Paul Mackerras <paulus@au1.ibm.com>
7 * Alexander Graf <agraf@suse.de>
8 * Kevin Wolf <mail@kevin-wolf.de>
9 *
10 * Description: KVM functions specific to running on Book 3S
11 * processors in hypervisor mode (specifically POWER7 and later).
12 *
13 * This file is derived from arch/powerpc/kvm/book3s.c,
14 * by Alexander Graf <agraf@suse.de>.
15 *
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.
19 */
20
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>
66b15db6 27#include <linux/export.h>
de56a948
PM		 28#include <linux/fs.h>
29#include <linux/anon_inodes.h>
30#include <linux/cpumask.h>
aa04b4cc
PM		 31#include <linux/spinlock.h>
32#include <linux/page-flags.h>
2c9097e4 33#include <linux/srcu.h>
398a76c6 34#include <linux/miscdevice.h>
de56a948
PM		 35
36#include <asm/reg.h>
37#include <asm/cputable.h>
38#include <asm/cacheflush.h>
39#include <asm/tlbflush.h>
40#include <asm/uaccess.h>
41#include <asm/io.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>
371fefd6 47#include <asm/cputhreads.h>
aa04b4cc 48#include <asm/page.h>
de1d9248 49#include <asm/hvcall.h>
ae3a197e 50#include <asm/switch_to.h>
512691d4 51#include <asm/smp.h>
de56a948 52#include <linux/gfp.h>
de56a948
PM		 53#include <linux/vmalloc.h>
54#include <linux/highmem.h>
c77162de 55#include <linux/hugetlb.h>
2ba9f0d8 56#include <linux/module.h>
de56a948 57
3a167bea
AK		 58#include "book3s.h"
59
de56a948
PM		 60/* #define EXIT_DEBUG */
61/* #define EXIT_DEBUG_SIMPLE */
62/* #define EXIT_DEBUG_INT */
63
913d3ff9
PM		 64/* Used to indicate that a guest page fault needs to be handled */
65#define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1)
66
c7b67670
PM		 67/* Used as a "null" value for timebase values */
68#define TB_NIL (~(u64)0)
69
19ccb76a 70static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
32fad281 71static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
19ccb76a 72
3a167bea 73static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
54695c30
BH		 74{
75 int me;
76 int cpu = vcpu->cpu;
77 wait_queue_head_t *wqp;
78
79 wqp = kvm_arch_vcpu_wq(vcpu);
80 if (waitqueue_active(wqp)) {
81 wake_up_interruptible(wqp);
82 ++vcpu->stat.halt_wakeup;
83 }
84
85 me = get_cpu();
86
87 /* CPU points to the first thread of the core */
88 if (cpu != me && cpu >= 0 && cpu < nr_cpu_ids) {
48eaef05 89#ifdef CONFIG_KVM_XICS
54695c30
BH		 90 int real_cpu = cpu + vcpu->arch.ptid;
91 if (paca[real_cpu].kvm_hstate.xics_phys)
92 xics_wake_cpu(real_cpu);
48eaef05
AS		 93 else
94#endif
95 if (cpu_online(cpu))
54695c30
BH		 96 smp_send_reschedule(cpu);
97 }
98 put_cpu();
99}
100
c7b67670
PM		 101/*
102 * We use the vcpu_load/put functions to measure stolen time.
103 * Stolen time is counted as time when either the vcpu is able to
104 * run as part of a virtual core, but the task running the vcore
105 * is preempted or sleeping, or when the vcpu needs something done
106 * in the kernel by the task running the vcpu, but that task is
107 * preempted or sleeping. Those two things have to be counted
108 * separately, since one of the vcpu tasks will take on the job
109 * of running the core, and the other vcpu tasks in the vcore will
110 * sleep waiting for it to do that, but that sleep shouldn't count
111 * as stolen time.
112 *
113 * Hence we accumulate stolen time when the vcpu can run as part of
114 * a vcore using vc->stolen_tb, and the stolen time when the vcpu
115 * needs its task to do other things in the kernel (for example,
116 * service a page fault) in busy_stolen. We don't accumulate
117 * stolen time for a vcore when it is inactive, or for a vcpu
118 * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of
119 * a misnomer; it means that the vcpu task is not executing in
120 * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
121 * the kernel. We don't have any way of dividing up that time
122 * between time that the vcpu is genuinely stopped, time that
123 * the task is actively working on behalf of the vcpu, and time
124 * that the task is preempted, so we don't count any of it as
125 * stolen.
126 *
127 * Updates to busy_stolen are protected by arch.tbacct_lock;
128 * updates to vc->stolen_tb are protected by the arch.tbacct_lock
129 * of the vcpu that has taken responsibility for running the vcore
130 * (i.e. vc->runner). The stolen times are measured in units of
131 * timebase ticks. (Note that the != TB_NIL checks below are
132 * purely defensive; they should never fail.)
133 */
134
3a167bea 135static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
de56a948 136{
0456ec4f
PM		 137 struct kvmppc_vcore *vc = vcpu->arch.vcore;
138
c7b67670
PM		 139 spin_lock(&vcpu->arch.tbacct_lock);
140 if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE &&
141 vc->preempt_tb != TB_NIL) {
0456ec4f 142 vc->stolen_tb += mftb() - vc->preempt_tb;
c7b67670
PM		 143 vc->preempt_tb = TB_NIL;
144 }
145 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
146 vcpu->arch.busy_preempt != TB_NIL) {
147 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
148 vcpu->arch.busy_preempt = TB_NIL;
149 }
150 spin_unlock(&vcpu->arch.tbacct_lock);
de56a948
PM		 151}
152
3a167bea 153static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
de56a948 154{
0456ec4f
PM		 155 struct kvmppc_vcore *vc = vcpu->arch.vcore;
156
c7b67670 157 spin_lock(&vcpu->arch.tbacct_lock);
0456ec4f
PM		 158 if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE)
159 vc->preempt_tb = mftb();
c7b67670
PM		 160 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
161 vcpu->arch.busy_preempt = mftb();
162 spin_unlock(&vcpu->arch.tbacct_lock);
de56a948
PM		 163}
164
3a167bea 165static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
de56a948
PM		 166{
167 vcpu->arch.shregs.msr = msr;
19ccb76a 168 kvmppc_end_cede(vcpu);
de56a948
PM		 169}
170
3a167bea 171void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
de56a948
PM		 172{
173 vcpu->arch.pvr = pvr;
174}
175
388cc6e1
PM		 176int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
177{
178 unsigned long pcr = 0;
179 struct kvmppc_vcore *vc = vcpu->arch.vcore;
180
181 if (arch_compat) {
182 if (!cpu_has_feature(CPU_FTR_ARCH_206))
183 return -EINVAL; /* 970 has no compat mode support */
184
185 switch (arch_compat) {
186 case PVR_ARCH_205:
5557ae0e
PM		 187 /*
188 * If an arch bit is set in PCR, all the defined
189 * higher-order arch bits also have to be set.
190 */
191 pcr = PCR_ARCH_206 | PCR_ARCH_205;
388cc6e1
PM		 192 break;
193 case PVR_ARCH_206:
194 case PVR_ARCH_206p:
5557ae0e
PM		 195 pcr = PCR_ARCH_206;
196 break;
197 case PVR_ARCH_207:
388cc6e1
PM		 198 break;
199 default:
200 return -EINVAL;
201 }
5557ae0e
PM		 202
203 if (!cpu_has_feature(CPU_FTR_ARCH_207S)) {
204 /* POWER7 can't emulate POWER8 */
205 if (!(pcr & PCR_ARCH_206))
206 return -EINVAL;
207 pcr &= ~PCR_ARCH_206;
208 }
388cc6e1
PM		 209 }
210
211 spin_lock(&vc->lock);
212 vc->arch_compat = arch_compat;
213 vc->pcr = pcr;
214 spin_unlock(&vc->lock);
215
216 return 0;
217}
218
de56a948
PM		 219void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
220{
221 int r;
222
223 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
224 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
225 vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
226 for (r = 0; r < 16; ++r)
227 pr_err("r%2d = %.16lx r%d = %.16lx\n",
228 r, kvmppc_get_gpr(vcpu, r),
229 r+16, kvmppc_get_gpr(vcpu, r+16));
230 pr_err("ctr = %.16lx lr = %.16lx\n",
231 vcpu->arch.ctr, vcpu->arch.lr);
232 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
233 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
234 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
235 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
236 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
237 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
238 pr_err("cr = %.8x xer = %.16lx dsisr = %.8x\n",
239 vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
240 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
241 pr_err("fault dar = %.16lx dsisr = %.8x\n",
242 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
243 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
244 for (r = 0; r < vcpu->arch.slb_max; ++r)
245 pr_err(" ESID = %.16llx VSID = %.16llx\n",
246 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
247 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
a0144e2a 248 vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
de56a948
PM		 249 vcpu->arch.last_inst);
250}
251
a8606e20
PM		 252struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
253{
254 int r;
255 struct kvm_vcpu *v, *ret = NULL;
256
257 mutex_lock(&kvm->lock);
258 kvm_for_each_vcpu(r, v, kvm) {
259 if (v->vcpu_id == id) {
260 ret = v;
261 break;
262 }
263 }
264 mutex_unlock(&kvm->lock);
265 return ret;
266}
267
268static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
269{
f13c13a0 270 vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
a8606e20
PM		 271 vpa->yield_count = 1;
272}
273
55b665b0
PM		 274static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
275 unsigned long addr, unsigned long len)
276{
277 /* check address is cacheline aligned */
278 if (addr & (L1_CACHE_BYTES - 1))
279 return -EINVAL;
280 spin_lock(&vcpu->arch.vpa_update_lock);
281 if (v->next_gpa != addr || v->len != len) {
282 v->next_gpa = addr;
283 v->len = addr ? len : 0;
284 v->update_pending = 1;
285 }
286 spin_unlock(&vcpu->arch.vpa_update_lock);
287 return 0;
288}
289
2e25aa5f
PM		 290/* Length for a per-processor buffer is passed in at offset 4 in the buffer */
291struct reg_vpa {
292 u32 dummy;
293 union {
294 u16 hword;
295 u32 word;
296 } length;
297};
298
299static int vpa_is_registered(struct kvmppc_vpa *vpap)
300{
301 if (vpap->update_pending)
302 return vpap->next_gpa != 0;
303 return vpap->pinned_addr != NULL;
304}
305
a8606e20
PM		 306static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
307 unsigned long flags,
308 unsigned long vcpuid, unsigned long vpa)
309{
310 struct kvm *kvm = vcpu->kvm;
93e60249 311 unsigned long len, nb;
a8606e20
PM		 312 void *va;
313 struct kvm_vcpu *tvcpu;
2e25aa5f
PM		 314 int err;
315 int subfunc;
316 struct kvmppc_vpa *vpap;
a8606e20
PM		 317
318 tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
319 if (!tvcpu)
320 return H_PARAMETER;
321
2e25aa5f
PM		 322 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
323 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
324 subfunc == H_VPA_REG_SLB) {
325 /* Registering new area - address must be cache-line aligned */
326 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
a8606e20 327 return H_PARAMETER;
2e25aa5f
PM		 328
329 /* convert logical addr to kernel addr and read length */
93e60249
PM		 330 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
331 if (va == NULL)
b2b2f165 332 return H_PARAMETER;
2e25aa5f
PM		 333 if (subfunc == H_VPA_REG_VPA)
334 len = ((struct reg_vpa *)va)->length.hword;
a8606e20 335 else
2e25aa5f 336 len = ((struct reg_vpa *)va)->length.word;
c35635ef 337 kvmppc_unpin_guest_page(kvm, va, vpa, false);
2e25aa5f
PM		 338
339 /* Check length */
340 if (len > nb || len < sizeof(struct reg_vpa))
341 return H_PARAMETER;
342 } else {
343 vpa = 0;
344 len = 0;
345 }
346
347 err = H_PARAMETER;
348 vpap = NULL;
349 spin_lock(&tvcpu->arch.vpa_update_lock);
350
351 switch (subfunc) {
352 case H_VPA_REG_VPA: /* register VPA */
353 if (len < sizeof(struct lppaca))
a8606e20 354 break;
2e25aa5f
PM		 355 vpap = &tvcpu->arch.vpa;
356 err = 0;
357 break;
358
359 case H_VPA_REG_DTL: /* register DTL */
360 if (len < sizeof(struct dtl_entry))
a8606e20 361 break;
2e25aa5f
PM		 362 len -= len % sizeof(struct dtl_entry);
363
364 /* Check that they have previously registered a VPA */
365 err = H_RESOURCE;
366 if (!vpa_is_registered(&tvcpu->arch.vpa))
a8606e20 367 break;
2e25aa5f
PM		 368
369 vpap = &tvcpu->arch.dtl;
370 err = 0;
371 break;
372
373 case H_VPA_REG_SLB: /* register SLB shadow buffer */
374 /* Check that they have previously registered a VPA */
375 err = H_RESOURCE;
376 if (!vpa_is_registered(&tvcpu->arch.vpa))
a8606e20 377 break;
2e25aa5f
PM		 378
379 vpap = &tvcpu->arch.slb_shadow;
380 err = 0;
381 break;
382
383 case H_VPA_DEREG_VPA: /* deregister VPA */
384 /* Check they don't still have a DTL or SLB buf registered */
385 err = H_RESOURCE;
386 if (vpa_is_registered(&tvcpu->arch.dtl) ||
387 vpa_is_registered(&tvcpu->arch.slb_shadow))
a8606e20 388 break;
2e25aa5f
PM		 389
390 vpap = &tvcpu->arch.vpa;
391 err = 0;
392 break;
393
394 case H_VPA_DEREG_DTL: /* deregister DTL */
395 vpap = &tvcpu->arch.dtl;
396 err = 0;
397 break;
398
399 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */
400 vpap = &tvcpu->arch.slb_shadow;
401 err = 0;
402 break;
403 }
404
405 if (vpap) {
406 vpap->next_gpa = vpa;
407 vpap->len = len;
408 vpap->update_pending = 1;
a8606e20 409 }
93e60249 410
2e25aa5f
PM		 411 spin_unlock(&tvcpu->arch.vpa_update_lock);
412
93e60249 413 return err;
a8606e20
PM		 414}
415
081f323b 416static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
2e25aa5f 417{
081f323b 418 struct kvm *kvm = vcpu->kvm;
2e25aa5f
PM		 419 void *va;
420 unsigned long nb;
081f323b 421 unsigned long gpa;
2e25aa5f 422
081f323b
PM		 423 /*
424 * We need to pin the page pointed to by vpap->next_gpa,
425 * but we can't call kvmppc_pin_guest_page under the lock
426 * as it does get_user_pages() and down_read(). So we
427 * have to drop the lock, pin the page, then get the lock
428 * again and check that a new area didn't get registered
429 * in the meantime.
430 */
431 for (;;) {
432 gpa = vpap->next_gpa;
433 spin_unlock(&vcpu->arch.vpa_update_lock);
434 va = NULL;
435 nb = 0;
436 if (gpa)
c35635ef 437 va = kvmppc_pin_guest_page(kvm, gpa, &nb);
081f323b
PM		 438 spin_lock(&vcpu->arch.vpa_update_lock);
439 if (gpa == vpap->next_gpa)
440 break;
441 /* sigh... unpin that one and try again */
442 if (va)
c35635ef 443 kvmppc_unpin_guest_page(kvm, va, gpa, false);
081f323b
PM		 444 }
445
446 vpap->update_pending = 0;
447 if (va && nb < vpap->len) {
448 /*
449 * If it's now too short, it must be that userspace
450 * has changed the mappings underlying guest memory,
451 * so unregister the region.
452 */
c35635ef 453 kvmppc_unpin_guest_page(kvm, va, gpa, false);
081f323b 454 va = NULL;
2e25aa5f
PM		 455 }
456 if (vpap->pinned_addr)
c35635ef
PM		 457 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
458 vpap->dirty);
459 vpap->gpa = gpa;
2e25aa5f 460 vpap->pinned_addr = va;
c35635ef 461 vpap->dirty = false;
2e25aa5f
PM		 462 if (va)
463 vpap->pinned_end = va + vpap->len;
464}
465
466static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
467{
2f12f034
PM		 468 if (!(vcpu->arch.vpa.update_pending ||
469 vcpu->arch.slb_shadow.update_pending ||
470 vcpu->arch.dtl.update_pending))
471 return;
472
2e25aa5f
PM		 473 spin_lock(&vcpu->arch.vpa_update_lock);
474 if (vcpu->arch.vpa.update_pending) {
081f323b 475 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
55b665b0
PM		 476 if (vcpu->arch.vpa.pinned_addr)
477 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
2e25aa5f
PM		 478 }
479 if (vcpu->arch.dtl.update_pending) {
081f323b 480 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
2e25aa5f
PM		 481 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
482 vcpu->arch.dtl_index = 0;
483 }
484 if (vcpu->arch.slb_shadow.update_pending)
081f323b 485 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
2e25aa5f
PM		 486 spin_unlock(&vcpu->arch.vpa_update_lock);
487}
488
c7b67670
PM		 489/*
490 * Return the accumulated stolen time for the vcore up until `now'.
491 * The caller should hold the vcore lock.
492 */
493static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
494{
495 u64 p;
496
497 /*
498 * If we are the task running the vcore, then since we hold
499 * the vcore lock, we can't be preempted, so stolen_tb/preempt_tb
500 * can't be updated, so we don't need the tbacct_lock.
501 * If the vcore is inactive, it can't become active (since we
502 * hold the vcore lock), so the vcpu load/put functions won't
503 * update stolen_tb/preempt_tb, and we don't need tbacct_lock.
504 */
505 if (vc->vcore_state != VCORE_INACTIVE &&
506 vc->runner->arch.run_task != current) {
507 spin_lock(&vc->runner->arch.tbacct_lock);
508 p = vc->stolen_tb;
509 if (vc->preempt_tb != TB_NIL)
510 p += now - vc->preempt_tb;
511 spin_unlock(&vc->runner->arch.tbacct_lock);
512 } else {
513 p = vc->stolen_tb;
514 }
515 return p;
516}
517
0456ec4f
PM		 518static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
519 struct kvmppc_vcore *vc)
520{
521 struct dtl_entry *dt;
522 struct lppaca *vpa;
c7b67670
PM		 523 unsigned long stolen;
524 unsigned long core_stolen;
525 u64 now;
0456ec4f
PM		 526
527 dt = vcpu->arch.dtl_ptr;
528 vpa = vcpu->arch.vpa.pinned_addr;
c7b67670
PM		 529 now = mftb();
530 core_stolen = vcore_stolen_time(vc, now);
531 stolen = core_stolen - vcpu->arch.stolen_logged;
532 vcpu->arch.stolen_logged = core_stolen;
533 spin_lock(&vcpu->arch.tbacct_lock);
534 stolen += vcpu->arch.busy_stolen;
535 vcpu->arch.busy_stolen = 0;
536 spin_unlock(&vcpu->arch.tbacct_lock);
0456ec4f
PM		 537 if (!dt || !vpa)
538 return;
539 memset(dt, 0, sizeof(struct dtl_entry));
540 dt->dispatch_reason = 7;
541 dt->processor_id = vc->pcpu + vcpu->arch.ptid;
93b0f4dc 542 dt->timebase = now + vc->tb_offset;
c7b67670 543 dt->enqueue_to_dispatch_time = stolen;
0456ec4f
PM		 544 dt->srr0 = kvmppc_get_pc(vcpu);
545 dt->srr1 = vcpu->arch.shregs.msr;
546 ++dt;
547 if (dt == vcpu->arch.dtl.pinned_end)
548 dt = vcpu->arch.dtl.pinned_addr;
549 vcpu->arch.dtl_ptr = dt;
550 /* order writing *dt vs. writing vpa->dtl_idx */
551 smp_wmb();
552 vpa->dtl_idx = ++vcpu->arch.dtl_index;
c35635ef 553 vcpu->arch.dtl.dirty = true;
0456ec4f
PM		 554}
555
a8606e20
PM		 556int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
557{
558 unsigned long req = kvmppc_get_gpr(vcpu, 3);
559 unsigned long target, ret = H_SUCCESS;
560 struct kvm_vcpu *tvcpu;
8e591cb7 561 int idx, rc;
a8606e20
PM		 562
563 switch (req) {
c77162de 564 case H_ENTER:
2c9097e4 565 idx = srcu_read_lock(&vcpu->kvm->srcu);
c77162de
PM		 566 ret = kvmppc_virtmode_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
567 kvmppc_get_gpr(vcpu, 5),
568 kvmppc_get_gpr(vcpu, 6),
569 kvmppc_get_gpr(vcpu, 7));
2c9097e4 570 srcu_read_unlock(&vcpu->kvm->srcu, idx);
c77162de 571 break;
a8606e20 572 case H_CEDE:
a8606e20
PM		 573 break;
574 case H_PROD:
575 target = kvmppc_get_gpr(vcpu, 4);
576 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
577 if (!tvcpu) {
578 ret = H_PARAMETER;
579 break;
580 }
581 tvcpu->arch.prodded = 1;
582 smp_mb();
583 if (vcpu->arch.ceded) {
584 if (waitqueue_active(&vcpu->wq)) {
585 wake_up_interruptible(&vcpu->wq);
586 vcpu->stat.halt_wakeup++;
587 }
588 }
589 break;
590 case H_CONFER:
42d7604d
PM		 591 target = kvmppc_get_gpr(vcpu, 4);
592 if (target == -1)
593 break;
594 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
595 if (!tvcpu) {
596 ret = H_PARAMETER;
597 break;
598 }
599 kvm_vcpu_yield_to(tvcpu);
a8606e20
PM		 600 break;
601 case H_REGISTER_VPA:
602 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
603 kvmppc_get_gpr(vcpu, 5),
604 kvmppc_get_gpr(vcpu, 6));
605 break;
8e591cb7
ME		 606 case H_RTAS:
607 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
608 return RESUME_HOST;
609
610 rc = kvmppc_rtas_hcall(vcpu);
611
612 if (rc == -ENOENT)
613 return RESUME_HOST;
614 else if (rc == 0)
615 break;
616
617 /* Send the error out to userspace via KVM_RUN */
618 return rc;
bc5ad3f3
BH		 619
620 case H_XIRR:
621 case H_CPPR:
622 case H_EOI:
623 case H_IPI:
8e44ddc3
PM		 624 case H_IPOLL:
625 case H_XIRR_X:
bc5ad3f3
BH		 626 if (kvmppc_xics_enabled(vcpu)) {
627 ret = kvmppc_xics_hcall(vcpu, req);
628 break;
629 } /* fallthrough */
a8606e20
PM		 630 default:
631 return RESUME_HOST;
632 }
633 kvmppc_set_gpr(vcpu, 3, ret);
634 vcpu->arch.hcall_needed = 0;
635 return RESUME_GUEST;
636}
637
3a167bea
AK		 638static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
639 struct task_struct *tsk)
de56a948
PM		 640{
641 int r = RESUME_HOST;
642
643 vcpu->stat.sum_exits++;
644
645 run->exit_reason = KVM_EXIT_UNKNOWN;
646 run->ready_for_interrupt_injection = 1;
647 switch (vcpu->arch.trap) {
648 /* We're good on these - the host merely wanted to get our attention */
649 case BOOK3S_INTERRUPT_HV_DECREMENTER:
650 vcpu->stat.dec_exits++;
651 r = RESUME_GUEST;
652 break;
653 case BOOK3S_INTERRUPT_EXTERNAL:
654 vcpu->stat.ext_intr_exits++;
655 r = RESUME_GUEST;
656 break;
657 case BOOK3S_INTERRUPT_PERFMON:
658 r = RESUME_GUEST;
659 break;
b4072df4
PM		 660 case BOOK3S_INTERRUPT_MACHINE_CHECK:
661 /*
662 * Deliver a machine check interrupt to the guest.
663 * We have to do this, even if the host has handled the
664 * machine check, because machine checks use SRR0/1 and
665 * the interrupt might have trashed guest state in them.
666 */
667 kvmppc_book3s_queue_irqprio(vcpu,
668 BOOK3S_INTERRUPT_MACHINE_CHECK);
669 r = RESUME_GUEST;
670 break;
de56a948
PM		 671 case BOOK3S_INTERRUPT_PROGRAM:
672 {
673 ulong flags;
674 /*
675 * Normally program interrupts are delivered directly
676 * to the guest by the hardware, but we can get here
677 * as a result of a hypervisor emulation interrupt
678 * (e40) getting turned into a 700 by BML RTAS.
679 */
680 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
681 kvmppc_core_queue_program(vcpu, flags);
682 r = RESUME_GUEST;
683 break;
684 }
685 case BOOK3S_INTERRUPT_SYSCALL:
686 {
687 /* hcall - punt to userspace */
688 int i;
689
27025a60
LPF		 690 /* hypercall with MSR_PR has already been handled in rmode,
691 * and never reaches here.
692 */
693
de56a948
PM		 694 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
695 for (i = 0; i < 9; ++i)
696 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
697 run->exit_reason = KVM_EXIT_PAPR_HCALL;
698 vcpu->arch.hcall_needed = 1;
699 r = RESUME_HOST;
700 break;
701 }
702 /*
342d3db7
PM		 703 * We get these next two if the guest accesses a page which it thinks
704 * it has mapped but which is not actually present, either because
705 * it is for an emulated I/O device or because the corresonding
706 * host page has been paged out. Any other HDSI/HISI interrupts
707 * have been handled already.
de56a948
PM		 708 */
709 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
913d3ff9 710 r = RESUME_PAGE_FAULT;
de56a948
PM		 711 break;
712 case BOOK3S_INTERRUPT_H_INST_STORAGE:
913d3ff9
PM		 713 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
714 vcpu->arch.fault_dsisr = 0;
715 r = RESUME_PAGE_FAULT;
de56a948
PM		 716 break;
717 /*
718 * This occurs if the guest executes an illegal instruction.
719 * We just generate a program interrupt to the guest, since
720 * we don't emulate any guest instructions at this stage.
721 */
722 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
bd3048b8
ME		 723 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
724 r = RESUME_GUEST;
725 break;
726 /*
727 * This occurs if the guest (kernel or userspace), does something that
728 * is prohibited by HFSCR. We just generate a program interrupt to
729 * the guest.
730 */
731 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
732 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
de56a948
PM		 733 r = RESUME_GUEST;
734 break;
735 default:
736 kvmppc_dump_regs(vcpu);
737 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
738 vcpu->arch.trap, kvmppc_get_pc(vcpu),
739 vcpu->arch.shregs.msr);
f3271d4c 740 run->hw.hardware_exit_reason = vcpu->arch.trap;
de56a948 741 r = RESUME_HOST;
de56a948
PM		 742 break;
743 }
744
de56a948
PM		 745 return r;
746}
747
3a167bea
AK		 748static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
749 struct kvm_sregs *sregs)
de56a948
PM		 750{
751 int i;
752
de56a948 753 memset(sregs, 0, sizeof(struct kvm_sregs));
87916442 754 sregs->pvr = vcpu->arch.pvr;
de56a948
PM		 755 for (i = 0; i < vcpu->arch.slb_max; i++) {
756 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
757 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
758 }
759
760 return 0;
761}
762
3a167bea
AK		 763static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
764 struct kvm_sregs *sregs)
de56a948
PM		 765{
766 int i, j;
767
3a167bea 768 kvmppc_set_pvr_hv(vcpu, sregs->pvr);
de56a948
PM		 769
770 j = 0;
771 for (i = 0; i < vcpu->arch.slb_nr; i++) {
772 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
773 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
774 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
775 ++j;
776 }
777 }
778 vcpu->arch.slb_max = j;
779
780 return 0;
781}
782
a0144e2a
PM		 783static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr)
784{
785 struct kvmppc_vcore *vc = vcpu->arch.vcore;
786 u64 mask;
787
788 spin_lock(&vc->lock);
789 /*
790 * Userspace can only modify DPFD (default prefetch depth),
791 * ILE (interrupt little-endian) and TC (translation control).
792 */
793 mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
794 vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
795 spin_unlock(&vc->lock);
796}
797
3a167bea
AK		 798static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
799 union kvmppc_one_reg *val)
31f3438e 800{
a136a8bd
PM		 801 int r = 0;
802 long int i;
31f3438e 803
a136a8bd 804 switch (id) {
31f3438e 805 case KVM_REG_PPC_HIOR:
a136a8bd
PM		 806 *val = get_reg_val(id, 0);
807 break;
808 case KVM_REG_PPC_DABR:
809 *val = get_reg_val(id, vcpu->arch.dabr);
810 break;
811 case KVM_REG_PPC_DSCR:
812 *val = get_reg_val(id, vcpu->arch.dscr);
813 break;
814 case KVM_REG_PPC_PURR:
815 *val = get_reg_val(id, vcpu->arch.purr);
816 break;
817 case KVM_REG_PPC_SPURR:
818 *val = get_reg_val(id, vcpu->arch.spurr);
819 break;
820 case KVM_REG_PPC_AMR:
821 *val = get_reg_val(id, vcpu->arch.amr);
822 break;
823 case KVM_REG_PPC_UAMOR:
824 *val = get_reg_val(id, vcpu->arch.uamor);
825 break;
b005255e 826 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
a136a8bd
PM		 827 i = id - KVM_REG_PPC_MMCR0;
828 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
829 break;
830 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
831 i = id - KVM_REG_PPC_PMC1;
832 *val = get_reg_val(id, vcpu->arch.pmc[i]);
31f3438e 833 break;
b005255e
MN		 834 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
835 i = id - KVM_REG_PPC_SPMC1;
836 *val = get_reg_val(id, vcpu->arch.spmc[i]);
837 break;
14941789
PM		 838 case KVM_REG_PPC_SIAR:
839 *val = get_reg_val(id, vcpu->arch.siar);
840 break;
841 case KVM_REG_PPC_SDAR:
842 *val = get_reg_val(id, vcpu->arch.sdar);
843 break;
b005255e
MN		 844 case KVM_REG_PPC_SIER:
845 *val = get_reg_val(id, vcpu->arch.sier);
846 break;
847 case KVM_REG_PPC_IAMR:
848 *val = get_reg_val(id, vcpu->arch.iamr);
849 break;
850 case KVM_REG_PPC_TFHAR:
851 *val = get_reg_val(id, vcpu->arch.tfhar);
852 break;
853 case KVM_REG_PPC_TFIAR:
854 *val = get_reg_val(id, vcpu->arch.tfiar);
855 break;
856 case KVM_REG_PPC_TEXASR:
857 *val = get_reg_val(id, vcpu->arch.texasr);
858 break;
859 case KVM_REG_PPC_FSCR:
860 *val = get_reg_val(id, vcpu->arch.fscr);
861 break;
862 case KVM_REG_PPC_PSPB:
863 *val = get_reg_val(id, vcpu->arch.pspb);
864 break;
865 case KVM_REG_PPC_EBBHR:
866 *val = get_reg_val(id, vcpu->arch.ebbhr);
867 break;
868 case KVM_REG_PPC_EBBRR:
869 *val = get_reg_val(id, vcpu->arch.ebbrr);
870 break;
871 case KVM_REG_PPC_BESCR:
872 *val = get_reg_val(id, vcpu->arch.bescr);
873 break;
874 case KVM_REG_PPC_TAR:
875 *val = get_reg_val(id, vcpu->arch.tar);
876 break;
877 case KVM_REG_PPC_DPDES:
878 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
879 break;
880 case KVM_REG_PPC_DAWR:
881 *val = get_reg_val(id, vcpu->arch.dawr);
882 break;
883 case KVM_REG_PPC_DAWRX:
884 *val = get_reg_val(id, vcpu->arch.dawrx);
885 break;
886 case KVM_REG_PPC_CIABR:
887 *val = get_reg_val(id, vcpu->arch.ciabr);
888 break;
889 case KVM_REG_PPC_IC:
890 *val = get_reg_val(id, vcpu->arch.ic);
891 break;
892 case KVM_REG_PPC_VTB:
893 *val = get_reg_val(id, vcpu->arch.vtb);
894 break;
895 case KVM_REG_PPC_CSIGR:
896 *val = get_reg_val(id, vcpu->arch.csigr);
897 break;
898 case KVM_REG_PPC_TACR:
899 *val = get_reg_val(id, vcpu->arch.tacr);
900 break;
901 case KVM_REG_PPC_TCSCR:
902 *val = get_reg_val(id, vcpu->arch.tcscr);
903 break;
904 case KVM_REG_PPC_PID:
905 *val = get_reg_val(id, vcpu->arch.pid);
906 break;
907 case KVM_REG_PPC_ACOP:
908 *val = get_reg_val(id, vcpu->arch.acop);
909 break;
910 case KVM_REG_PPC_WORT:
911 *val = get_reg_val(id, vcpu->arch.wort);
912 break;
55b665b0
PM		 913 case KVM_REG_PPC_VPA_ADDR:
914 spin_lock(&vcpu->arch.vpa_update_lock);
915 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
916 spin_unlock(&vcpu->arch.vpa_update_lock);
917 break;
918 case KVM_REG_PPC_VPA_SLB:
919 spin_lock(&vcpu->arch.vpa_update_lock);
920 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
921 val->vpaval.length = vcpu->arch.slb_shadow.len;
922 spin_unlock(&vcpu->arch.vpa_update_lock);
923 break;
924 case KVM_REG_PPC_VPA_DTL:
925 spin_lock(&vcpu->arch.vpa_update_lock);
926 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
927 val->vpaval.length = vcpu->arch.dtl.len;
928 spin_unlock(&vcpu->arch.vpa_update_lock);
929 break;
93b0f4dc
PM		 930 case KVM_REG_PPC_TB_OFFSET:
931 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
932 break;
a0144e2a
PM		 933 case KVM_REG_PPC_LPCR:
934 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
935 break;
4b8473c9
PM		 936 case KVM_REG_PPC_PPR:
937 *val = get_reg_val(id, vcpu->arch.ppr);
938 break;
388cc6e1
PM		 939 case KVM_REG_PPC_ARCH_COMPAT:
940 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
941 break;
31f3438e 942 default:
a136a8bd 943 r = -EINVAL;
31f3438e
PM		 944 break;
945 }
946
947 return r;
948}
949
3a167bea
AK		 950static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
951 union kvmppc_one_reg *val)
31f3438e 952{
a136a8bd
PM		 953 int r = 0;
954 long int i;
55b665b0 955 unsigned long addr, len;
31f3438e 956
a136a8bd 957 switch (id) {
31f3438e 958 case KVM_REG_PPC_HIOR:
31f3438e 959 /* Only allow this to be set to zero */
a136a8bd 960 if (set_reg_val(id, *val))
31f3438e
PM		 961 r = -EINVAL;
962 break;
a136a8bd
PM		 963 case KVM_REG_PPC_DABR:
964 vcpu->arch.dabr = set_reg_val(id, *val);
965 break;
966 case KVM_REG_PPC_DSCR:
967 vcpu->arch.dscr = set_reg_val(id, *val);
968 break;
969 case KVM_REG_PPC_PURR:
970 vcpu->arch.purr = set_reg_val(id, *val);
971 break;
972 case KVM_REG_PPC_SPURR:
973 vcpu->arch.spurr = set_reg_val(id, *val);
974 break;
975 case KVM_REG_PPC_AMR:
976 vcpu->arch.amr = set_reg_val(id, *val);
977 break;
978 case KVM_REG_PPC_UAMOR:
979 vcpu->arch.uamor = set_reg_val(id, *val);
980 break;
b005255e 981 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
a136a8bd
PM		 982 i = id - KVM_REG_PPC_MMCR0;
983 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
984 break;
985 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
986 i = id - KVM_REG_PPC_PMC1;
987 vcpu->arch.pmc[i] = set_reg_val(id, *val);
988 break;
b005255e
MN		 989 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
990 i = id - KVM_REG_PPC_SPMC1;
991 vcpu->arch.spmc[i] = set_reg_val(id, *val);
992 break;
14941789
PM		 993 case KVM_REG_PPC_SIAR:
994 vcpu->arch.siar = set_reg_val(id, *val);
995 break;
996 case KVM_REG_PPC_SDAR:
997 vcpu->arch.sdar = set_reg_val(id, *val);
998 break;
b005255e
MN		 999 case KVM_REG_PPC_SIER:
1000 vcpu->arch.sier = set_reg_val(id, *val);
1001 break;
1002 case KVM_REG_PPC_IAMR:
1003 vcpu->arch.iamr = set_reg_val(id, *val);
1004 break;
1005 case KVM_REG_PPC_TFHAR:
1006 vcpu->arch.tfhar = set_reg_val(id, *val);
1007 break;
1008 case KVM_REG_PPC_TFIAR:
1009 vcpu->arch.tfiar = set_reg_val(id, *val);
1010 break;
1011 case KVM_REG_PPC_TEXASR:
1012 vcpu->arch.texasr = set_reg_val(id, *val);
1013 break;
1014 case KVM_REG_PPC_FSCR:
1015 vcpu->arch.fscr = set_reg_val(id, *val);
1016 break;
1017 case KVM_REG_PPC_PSPB:
1018 vcpu->arch.pspb = set_reg_val(id, *val);
1019 break;
1020 case KVM_REG_PPC_EBBHR:
1021 vcpu->arch.ebbhr = set_reg_val(id, *val);
1022 break;
1023 case KVM_REG_PPC_EBBRR:
1024 vcpu->arch.ebbrr = set_reg_val(id, *val);
1025 break;
1026 case KVM_REG_PPC_BESCR:
1027 vcpu->arch.bescr = set_reg_val(id, *val);
1028 break;
1029 case KVM_REG_PPC_TAR:
1030 vcpu->arch.tar = set_reg_val(id, *val);
1031 break;
1032 case KVM_REG_PPC_DPDES:
1033 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1034 break;
1035 case KVM_REG_PPC_DAWR:
1036 vcpu->arch.dawr = set_reg_val(id, *val);
1037 break;
1038 case KVM_REG_PPC_DAWRX:
1039 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1040 break;
1041 case KVM_REG_PPC_CIABR:
1042 vcpu->arch.ciabr = set_reg_val(id, *val);
1043 /* Don't allow setting breakpoints in hypervisor code */
1044 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1045 vcpu->arch.ciabr &= ~CIABR_PRIV; /* disable */
1046 break;
1047 case KVM_REG_PPC_IC:
1048 vcpu->arch.ic = set_reg_val(id, *val);
1049 break;
1050 case KVM_REG_PPC_VTB:
1051 vcpu->arch.vtb = set_reg_val(id, *val);
1052 break;
1053 case KVM_REG_PPC_CSIGR:
1054 vcpu->arch.csigr = set_reg_val(id, *val);
1055 break;
1056 case KVM_REG_PPC_TACR:
1057 vcpu->arch.tacr = set_reg_val(id, *val);
1058 break;
1059 case KVM_REG_PPC_TCSCR:
1060 vcpu->arch.tcscr = set_reg_val(id, *val);
1061 break;
1062 case KVM_REG_PPC_PID:
1063 vcpu->arch.pid = set_reg_val(id, *val);
1064 break;
1065 case KVM_REG_PPC_ACOP:
1066 vcpu->arch.acop = set_reg_val(id, *val);
1067 break;
1068 case KVM_REG_PPC_WORT:
1069 vcpu->arch.wort = set_reg_val(id, *val);
1070 break;
55b665b0
PM		 1071 case KVM_REG_PPC_VPA_ADDR:
1072 addr = set_reg_val(id, *val);
1073 r = -EINVAL;
1074 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1075 vcpu->arch.dtl.next_gpa))
1076 break;
1077 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1078 break;
1079 case KVM_REG_PPC_VPA_SLB:
1080 addr = val->vpaval.addr;
1081 len = val->vpaval.length;
1082 r = -EINVAL;
1083 if (addr && !vcpu->arch.vpa.next_gpa)
1084 break;
1085 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1086 break;
1087 case KVM_REG_PPC_VPA_DTL:
1088 addr = val->vpaval.addr;
1089 len = val->vpaval.length;
1090 r = -EINVAL;
9f8c8c78
PM		 1091 if (addr && (len < sizeof(struct dtl_entry) ||
1092 !vcpu->arch.vpa.next_gpa))
55b665b0
PM		 1093 break;
1094 len -= len % sizeof(struct dtl_entry);
1095 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1096 break;
93b0f4dc
PM		 1097 case KVM_REG_PPC_TB_OFFSET:
1098 /* round up to multiple of 2^24 */
1099 vcpu->arch.vcore->tb_offset =
1100 ALIGN(set_reg_val(id, *val), 1UL << 24);
1101 break;
a0144e2a
PM		 1102 case KVM_REG_PPC_LPCR:
1103 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val));
1104 break;
4b8473c9
PM		 1105 case KVM_REG_PPC_PPR:
1106 vcpu->arch.ppr = set_reg_val(id, *val);
1107 break;
388cc6e1
PM		 1108 case KVM_REG_PPC_ARCH_COMPAT:
1109 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1110 break;
31f3438e 1111 default:
a136a8bd 1112 r = -EINVAL;
31f3438e
PM		 1113 break;
1114 }
1115
1116 return r;
1117}
1118
3a167bea
AK		 1119static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1120 unsigned int id)
de56a948
PM		 1121{
1122 struct kvm_vcpu *vcpu;
371fefd6
PM		 1123 int err = -EINVAL;
1124 int core;
1125 struct kvmppc_vcore *vcore;
de56a948 1126
371fefd6
PM		 1127 core = id / threads_per_core;
1128 if (core >= KVM_MAX_VCORES)
1129 goto out;
1130
1131 err = -ENOMEM;
6b75e6bf 1132 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
de56a948
PM		 1133 if (!vcpu)
1134 goto out;
1135
1136 err = kvm_vcpu_init(vcpu, kvm, id);
1137 if (err)
1138 goto free_vcpu;
1139
1140 vcpu->arch.shared = &vcpu->arch.shregs;
de56a948
PM		 1141 vcpu->arch.mmcr[0] = MMCR0_FC;
1142 vcpu->arch.ctrl = CTRL_RUNLATCH;
1143 /* default to host PVR, since we can't spoof it */
3a167bea 1144 kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2e25aa5f 1145 spin_lock_init(&vcpu->arch.vpa_update_lock);
c7b67670
PM		 1146 spin_lock_init(&vcpu->arch.tbacct_lock);
1147 vcpu->arch.busy_preempt = TB_NIL;
de56a948 1148
de56a948
PM		 1149 kvmppc_mmu_book3s_hv_init(vcpu);
1150
8455d79e 1151 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
371fefd6
PM		 1152
1153 init_waitqueue_head(&vcpu->arch.cpu_run);
1154
1155 mutex_lock(&kvm->lock);
1156 vcore = kvm->arch.vcores[core];
1157 if (!vcore) {
1158 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1159 if (vcore) {
1160 INIT_LIST_HEAD(&vcore->runnable_threads);
1161 spin_lock_init(&vcore->lock);
19ccb76a 1162 init_waitqueue_head(&vcore->wq);
c7b67670 1163 vcore->preempt_tb = TB_NIL;
a0144e2a 1164 vcore->lpcr = kvm->arch.lpcr;
e0b7ec05
PM		 1165 vcore->first_vcpuid = core * threads_per_core;
1166 vcore->kvm = kvm;
371fefd6
PM		 1167 }
1168 kvm->arch.vcores[core] = vcore;
1b400ba0 1169 kvm->arch.online_vcores++;
371fefd6
PM		 1170 }
1171 mutex_unlock(&kvm->lock);
1172
1173 if (!vcore)
1174 goto free_vcpu;
1175
1176 spin_lock(&vcore->lock);
1177 ++vcore->num_threads;
371fefd6
PM		 1178 spin_unlock(&vcore->lock);
1179 vcpu->arch.vcore = vcore;
e0b7ec05 1180 vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
371fefd6 1181
af8f38b3
AG		 1182 vcpu->arch.cpu_type = KVM_CPU_3S_64;
1183 kvmppc_sanity_check(vcpu);
1184
de56a948
PM		 1185 return vcpu;
1186
1187free_vcpu:
6b75e6bf 1188 kmem_cache_free(kvm_vcpu_cache, vcpu);
de56a948
PM		 1189out:
1190 return ERR_PTR(err);
1191}
1192
c35635ef
PM		 1193static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1194{
1195 if (vpa->pinned_addr)
1196 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1197 vpa->dirty);
1198}
1199
3a167bea 1200static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
de56a948 1201{
2e25aa5f 1202 spin_lock(&vcpu->arch.vpa_update_lock);
c35635ef
PM		 1203 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1204 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1205 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2e25aa5f 1206 spin_unlock(&vcpu->arch.vpa_update_lock);
de56a948 1207 kvm_vcpu_uninit(vcpu);
6b75e6bf 1208 kmem_cache_free(kvm_vcpu_cache, vcpu);
de56a948
PM		 1209}
1210
3a167bea
AK		 1211static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1212{
1213 /* Indicate we want to get back into the guest */
1214 return 1;
1215}
1216
19ccb76a 1217static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
371fefd6 1218{
19ccb76a 1219 unsigned long dec_nsec, now;
371fefd6 1220
19ccb76a
PM		 1221 now = get_tb();
1222 if (now > vcpu->arch.dec_expires) {
1223 /* decrementer has already gone negative */
1224 kvmppc_core_queue_dec(vcpu);
7e28e60e 1225 kvmppc_core_prepare_to_enter(vcpu);
19ccb76a 1226 return;
371fefd6 1227 }
19ccb76a
PM		 1228 dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1229 / tb_ticks_per_sec;
1230 hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1231 HRTIMER_MODE_REL);
1232 vcpu->arch.timer_running = 1;
371fefd6
PM		 1233}
1234
19ccb76a 1235static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
371fefd6 1236{
19ccb76a
PM		 1237 vcpu->arch.ceded = 0;
1238 if (vcpu->arch.timer_running) {
1239 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1240 vcpu->arch.timer_running = 0;
1241 }
371fefd6
PM		 1242}
1243
e0b7ec05 1244extern void __kvmppc_vcore_entry(void);
de56a948 1245
371fefd6
PM		 1246static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1247 struct kvm_vcpu *vcpu)
de56a948 1248{
c7b67670
PM		 1249 u64 now;
1250
371fefd6
PM		 1251 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1252 return;
c7b67670
PM		 1253 spin_lock(&vcpu->arch.tbacct_lock);
1254 now = mftb();
1255 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1256 vcpu->arch.stolen_logged;
1257 vcpu->arch.busy_preempt = now;
1258 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1259 spin_unlock(&vcpu->arch.tbacct_lock);
371fefd6 1260 --vc->n_runnable;
371fefd6
PM		 1261 list_del(&vcpu->arch.run_list);
1262}
1263
f0888f70
PM		 1264static int kvmppc_grab_hwthread(int cpu)
1265{
1266 struct paca_struct *tpaca;
1267 long timeout = 1000;
1268
1269 tpaca = &paca[cpu];
1270
1271 /* Ensure the thread won't go into the kernel if it wakes */
1272 tpaca->kvm_hstate.hwthread_req = 1;
7b444c67 1273 tpaca->kvm_hstate.kvm_vcpu = NULL;
f0888f70
PM		 1274
1275 /*
1276 * If the thread is already executing in the kernel (e.g. handling
1277 * a stray interrupt), wait for it to get back to nap mode.
1278 * The smp_mb() is to ensure that our setting of hwthread_req
1279 * is visible before we look at hwthread_state, so if this
1280 * races with the code at system_reset_pSeries and the thread
1281 * misses our setting of hwthread_req, we are sure to see its
1282 * setting of hwthread_state, and vice versa.
1283 */
1284 smp_mb();
1285 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1286 if (--timeout <= 0) {
1287 pr_err("KVM: couldn't grab cpu %d\n", cpu);
1288 return -EBUSY;
1289 }
1290 udelay(1);
1291 }
1292 return 0;
1293}
1294
1295static void kvmppc_release_hwthread(int cpu)
1296{
1297 struct paca_struct *tpaca;
1298
1299 tpaca = &paca[cpu];
1300 tpaca->kvm_hstate.hwthread_req = 0;
1301 tpaca->kvm_hstate.kvm_vcpu = NULL;
1302}
1303
371fefd6
PM		 1304static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
1305{
1306 int cpu;
1307 struct paca_struct *tpaca;
1308 struct kvmppc_vcore *vc = vcpu->arch.vcore;
1309
19ccb76a
PM		 1310 if (vcpu->arch.timer_running) {
1311 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1312 vcpu->arch.timer_running = 0;
1313 }
371fefd6
PM		 1314 cpu = vc->pcpu + vcpu->arch.ptid;
1315 tpaca = &paca[cpu];
1316 tpaca->kvm_hstate.kvm_vcpu = vcpu;
1317 tpaca->kvm_hstate.kvm_vcore = vc;
e0b7ec05 1318 tpaca->kvm_hstate.ptid = vcpu->arch.ptid;
19ccb76a 1319 vcpu->cpu = vc->pcpu;
371fefd6 1320 smp_wmb();
251da038 1321#if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
e0b7ec05 1322 if (cpu != smp_processor_id()) {
48eaef05 1323#ifdef CONFIG_KVM_XICS
371fefd6 1324 xics_wake_cpu(cpu);
48eaef05 1325#endif
e0b7ec05
PM		 1326 if (vcpu->arch.ptid)
1327 ++vc->n_woken;
de56a948 1328 }
371fefd6
PM		 1329#endif
1330}
de56a948 1331
371fefd6
PM		 1332static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc)
1333{
1334 int i;
1335
1336 HMT_low();
1337 i = 0;
1338 while (vc->nap_count < vc->n_woken) {
1339 if (++i >= 1000000) {
1340 pr_err("kvmppc_wait_for_nap timeout %d %d\n",
1341 vc->nap_count, vc->n_woken);
1342 break;
1343 }
1344 cpu_relax();
1345 }
1346 HMT_medium();
1347}
1348
1349/*
1350 * Check that we are on thread 0 and that any other threads in
7b444c67
PM		 1351 * this core are off-line. Then grab the threads so they can't
1352 * enter the kernel.
371fefd6
PM		 1353 */
1354static int on_primary_thread(void)
1355{
1356 int cpu = smp_processor_id();
1357 int thr = cpu_thread_in_core(cpu);
1358
1359 if (thr)
1360 return 0;
1361 while (++thr < threads_per_core)
1362 if (cpu_online(cpu + thr))
1363 return 0;
7b444c67
PM		 1364
1365 /* Grab all hw threads so they can't go into the kernel */
1366 for (thr = 1; thr < threads_per_core; ++thr) {
1367 if (kvmppc_grab_hwthread(cpu + thr)) {
1368 /* Couldn't grab one; let the others go */
1369 do {
1370 kvmppc_release_hwthread(cpu + thr);
1371 } while (--thr > 0);
1372 return 0;
1373 }
1374 }
371fefd6
PM		 1375 return 1;
1376}
1377
1378/*
1379 * Run a set of guest threads on a physical core.
1380 * Called with vc->lock held.
1381 */
913d3ff9 1382static void kvmppc_run_core(struct kvmppc_vcore *vc)
371fefd6 1383{
e0b7ec05 1384 struct kvm_vcpu *vcpu, *vnext;
371fefd6
PM		 1385 long ret;
1386 u64 now;
e0b7ec05 1387 int i, need_vpa_update;
2c9097e4 1388 int srcu_idx;
913d3ff9 1389 struct kvm_vcpu *vcpus_to_update[threads_per_core];
371fefd6
PM		 1390
1391 /* don't start if any threads have a signal pending */
081f323b
PM		 1392 need_vpa_update = 0;
1393 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
371fefd6 1394 if (signal_pending(vcpu->arch.run_task))
913d3ff9
PM		 1395 return;
1396 if (vcpu->arch.vpa.update_pending ||
1397 vcpu->arch.slb_shadow.update_pending ||
1398 vcpu->arch.dtl.update_pending)
1399 vcpus_to_update[need_vpa_update++] = vcpu;
081f323b
PM		 1400 }
1401
1402 /*
1403 * Initialize *vc, in particular vc->vcore_state, so we can
1404 * drop the vcore lock if necessary.
1405 */
1406 vc->n_woken = 0;
1407 vc->nap_count = 0;
1408 vc->entry_exit_count = 0;
2f12f034 1409 vc->vcore_state = VCORE_STARTING;
081f323b
PM		 1410 vc->in_guest = 0;
1411 vc->napping_threads = 0;
1412
1413 /*
1414 * Updating any of the vpas requires calling kvmppc_pin_guest_page,
1415 * which can't be called with any spinlocks held.
1416 */
1417 if (need_vpa_update) {
1418 spin_unlock(&vc->lock);
913d3ff9
PM		 1419 for (i = 0; i < need_vpa_update; ++i)
1420 kvmppc_update_vpas(vcpus_to_update[i]);
081f323b
PM		 1421 spin_lock(&vc->lock);
1422 }
de56a948 1423
7b444c67
PM		 1424 /*
1425 * Make sure we are running on thread 0, and that
1426 * secondary threads are offline.
1427 */
1428 if (threads_per_core > 1 && !on_primary_thread()) {
1429 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1430 vcpu->arch.ret = -EBUSY;
1431 goto out;
1432 }
1433
371fefd6 1434 vc->pcpu = smp_processor_id();
2e25aa5f 1435 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
371fefd6 1436 kvmppc_start_thread(vcpu);
0456ec4f 1437 kvmppc_create_dtl_entry(vcpu, vc);
2e25aa5f 1438 }
371fefd6 1439
e0b7ec05
PM		 1440 /* Set this explicitly in case thread 0 doesn't have a vcpu */
1441 get_paca()->kvm_hstate.kvm_vcore = vc;
1442 get_paca()->kvm_hstate.ptid = 0;
1443
2f12f034 1444 vc->vcore_state = VCORE_RUNNING;
19ccb76a 1445 preempt_disable();
371fefd6 1446 spin_unlock(&vc->lock);
de56a948 1447
371fefd6 1448 kvm_guest_enter();
2c9097e4 1449
e0b7ec05 1450 srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2c9097e4 1451
e0b7ec05 1452 __kvmppc_vcore_entry();
de56a948 1453
371fefd6 1454 spin_lock(&vc->lock);
19ccb76a
PM		 1455 /* disable sending of IPIs on virtual external irqs */
1456 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1457 vcpu->cpu = -1;
1458 /* wait for secondary threads to finish writing their state to memory */
371fefd6
PM		 1459 if (vc->nap_count < vc->n_woken)
1460 kvmppc_wait_for_nap(vc);
2f12f034
PM		 1461 for (i = 0; i < threads_per_core; ++i)
1462 kvmppc_release_hwthread(vc->pcpu + i);
371fefd6 1463 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
19ccb76a 1464 vc->vcore_state = VCORE_EXITING;
371fefd6
PM		 1465 spin_unlock(&vc->lock);
1466
e0b7ec05 1467 srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2c9097e4 1468
371fefd6
PM		 1469 /* make sure updates to secondary vcpu structs are visible now */
1470 smp_mb();
de56a948
PM		 1471 kvm_guest_exit();
1472
1473 preempt_enable();
1474 kvm_resched(vcpu);
1475
913d3ff9 1476 spin_lock(&vc->lock);
de56a948 1477 now = get_tb();
371fefd6
PM		 1478 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1479 /* cancel pending dec exception if dec is positive */
1480 if (now < vcpu->arch.dec_expires &&
1481 kvmppc_core_pending_dec(vcpu))
1482 kvmppc_core_dequeue_dec(vcpu);
19ccb76a
PM		 1483
1484 ret = RESUME_GUEST;
1485 if (vcpu->arch.trap)
3a167bea
AK		 1486 ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
1487 vcpu->arch.run_task);
19ccb76a 1488
371fefd6
PM		 1489 vcpu->arch.ret = ret;
1490 vcpu->arch.trap = 0;
19ccb76a
PM		 1491
1492 if (vcpu->arch.ceded) {
1493 if (ret != RESUME_GUEST)
1494 kvmppc_end_cede(vcpu);
1495 else
1496 kvmppc_set_timer(vcpu);
1497 }
371fefd6 1498 }
de56a948
PM		 1499
1500 out:
19ccb76a 1501 vc->vcore_state = VCORE_INACTIVE;
371fefd6
PM		 1502 list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1503 arch.run_list) {
1504 if (vcpu->arch.ret != RESUME_GUEST) {
1505 kvmppc_remove_runnable(vc, vcpu);
1506 wake_up(&vcpu->arch.cpu_run);
1507 }
1508 }
371fefd6
PM		 1509}
1510
19ccb76a
PM		 1511/*
1512 * Wait for some other vcpu thread to execute us, and
1513 * wake us up when we need to handle something in the host.
1514 */
1515static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
371fefd6 1516{
371fefd6
PM		 1517 DEFINE_WAIT(wait);
1518
19ccb76a
PM		 1519 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
1520 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
1521 schedule();
1522 finish_wait(&vcpu->arch.cpu_run, &wait);
1523}
1524
1525/*
1526 * All the vcpus in this vcore are idle, so wait for a decrementer
1527 * or external interrupt to one of the vcpus. vc->lock is held.
1528 */
1529static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
1530{
1531 DEFINE_WAIT(wait);
19ccb76a
PM		 1532
1533 prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
1534 vc->vcore_state = VCORE_SLEEPING;
1535 spin_unlock(&vc->lock);
913d3ff9 1536 schedule();
19ccb76a
PM		 1537 finish_wait(&vc->wq, &wait);
1538 spin_lock(&vc->lock);
1539 vc->vcore_state = VCORE_INACTIVE;
1540}
371fefd6 1541
19ccb76a
PM		 1542static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1543{
1544 int n_ceded;
19ccb76a
PM		 1545 struct kvmppc_vcore *vc;
1546 struct kvm_vcpu *v, *vn;
9e368f29 1547
371fefd6
PM		 1548 kvm_run->exit_reason = 0;
1549 vcpu->arch.ret = RESUME_GUEST;
1550 vcpu->arch.trap = 0;
2f12f034 1551 kvmppc_update_vpas(vcpu);
371fefd6 1552
371fefd6
PM		 1553 /*
1554 * Synchronize with other threads in this virtual core
1555 */
1556 vc = vcpu->arch.vcore;
1557 spin_lock(&vc->lock);
19ccb76a 1558 vcpu->arch.ceded = 0;
371fefd6
PM		 1559 vcpu->arch.run_task = current;
1560 vcpu->arch.kvm_run = kvm_run;
c7b67670 1561 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
19ccb76a 1562 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
c7b67670 1563 vcpu->arch.busy_preempt = TB_NIL;
371fefd6
PM		 1564 list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
1565 ++vc->n_runnable;
1566
19ccb76a
PM		 1567 /*
1568 * This happens the first time this is called for a vcpu.
1569 * If the vcore is already running, we may be able to start
1570 * this thread straight away and have it join in.
1571 */
8455d79e 1572 if (!signal_pending(current)) {
19ccb76a
PM		 1573 if (vc->vcore_state == VCORE_RUNNING &&
1574 VCORE_EXIT_COUNT(vc) == 0) {
2f12f034 1575 kvmppc_create_dtl_entry(vcpu, vc);
19ccb76a 1576 kvmppc_start_thread(vcpu);
8455d79e
PM		 1577 } else if (vc->vcore_state == VCORE_SLEEPING) {
1578 wake_up(&vc->wq);
371fefd6
PM		 1579 }
1580
8455d79e 1581 }
371fefd6 1582
19ccb76a
PM		 1583 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1584 !signal_pending(current)) {
8455d79e 1585 if (vc->vcore_state != VCORE_INACTIVE) {
19ccb76a
PM		 1586 spin_unlock(&vc->lock);
1587 kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
1588 spin_lock(&vc->lock);
1589 continue;
1590 }
19ccb76a
PM		 1591 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
1592 arch.run_list) {
7e28e60e 1593 kvmppc_core_prepare_to_enter(v);
19ccb76a
PM		 1594 if (signal_pending(v->arch.run_task)) {
1595 kvmppc_remove_runnable(vc, v);
1596 v->stat.signal_exits++;
1597 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
1598 v->arch.ret = -EINTR;
1599 wake_up(&v->arch.cpu_run);
1600 }
1601 }
8455d79e
PM		 1602 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1603 break;
1604 vc->runner = vcpu;
1605 n_ceded = 0;
4619ac88 1606 list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
8455d79e
PM		 1607 if (!v->arch.pending_exceptions)
1608 n_ceded += v->arch.ceded;
4619ac88
PM		 1609 else
1610 v->arch.ceded = 0;
1611 }
8455d79e
PM		 1612 if (n_ceded == vc->n_runnable)
1613 kvmppc_vcore_blocked(vc);
1614 else
1615 kvmppc_run_core(vc);
0456ec4f 1616 vc->runner = NULL;
19ccb76a 1617 }
371fefd6 1618
8455d79e
PM		 1619 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1620 (vc->vcore_state == VCORE_RUNNING ||
1621 vc->vcore_state == VCORE_EXITING)) {
1622 spin_unlock(&vc->lock);
1623 kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
1624 spin_lock(&vc->lock);
1625 }
1626
1627 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
1628 kvmppc_remove_runnable(vc, vcpu);
1629 vcpu->stat.signal_exits++;
1630 kvm_run->exit_reason = KVM_EXIT_INTR;
1631 vcpu->arch.ret = -EINTR;
1632 }
1633
1634 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
1635 /* Wake up some vcpu to run the core */
1636 v = list_first_entry(&vc->runnable_threads,
1637 struct kvm_vcpu, arch.run_list);
1638 wake_up(&v->arch.cpu_run);
371fefd6
PM		 1639 }
1640
371fefd6 1641 spin_unlock(&vc->lock);
371fefd6 1642 return vcpu->arch.ret;
de56a948
PM		 1643}
1644
3a167bea 1645static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
a8606e20
PM		 1646{
1647 int r;
913d3ff9 1648 int srcu_idx;
a8606e20 1649
af8f38b3
AG		 1650 if (!vcpu->arch.sane) {
1651 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1652 return -EINVAL;
1653 }
1654
25051b5a
SW		 1655 kvmppc_core_prepare_to_enter(vcpu);
1656
19ccb76a
PM		 1657 /* No need to go into the guest when all we'll do is come back out */
1658 if (signal_pending(current)) {
1659 run->exit_reason = KVM_EXIT_INTR;
1660 return -EINTR;
1661 }
1662
32fad281
PM		 1663 atomic_inc(&vcpu->kvm->arch.vcpus_running);
1664 /* Order vcpus_running vs. rma_setup_done, see kvmppc_alloc_reset_hpt */
1665 smp_mb();
1666
1667 /* On the first time here, set up HTAB and VRMA or RMA */
c77162de 1668 if (!vcpu->kvm->arch.rma_setup_done) {
32fad281 1669 r = kvmppc_hv_setup_htab_rma(vcpu);
c77162de 1670 if (r)
32fad281 1671 goto out;
c77162de 1672 }
19ccb76a
PM		 1673
1674 flush_fp_to_thread(current);
1675 flush_altivec_to_thread(current);
1676 flush_vsx_to_thread(current);
1677 vcpu->arch.wqp = &vcpu->arch.vcore->wq;
342d3db7 1678 vcpu->arch.pgdir = current->mm->pgd;
c7b67670 1679 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
19ccb76a 1680
a8606e20
PM		 1681 do {
1682 r = kvmppc_run_vcpu(run, vcpu);
1683
1684 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
1685 !(vcpu->arch.shregs.msr & MSR_PR)) {
1686 r = kvmppc_pseries_do_hcall(vcpu);
7e28e60e 1687 kvmppc_core_prepare_to_enter(vcpu);
913d3ff9
PM		 1688 } else if (r == RESUME_PAGE_FAULT) {
1689 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1690 r = kvmppc_book3s_hv_page_fault(run, vcpu,
1691 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1692 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
a8606e20
PM		 1693 }
1694 } while (r == RESUME_GUEST);
32fad281
PM		 1695
1696 out:
c7b67670 1697 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
32fad281 1698 atomic_dec(&vcpu->kvm->arch.vcpus_running);
a8606e20
PM		 1699 return r;
1700}
1701
54738c09 1702
aa04b4cc 1703/* Work out RMLS (real mode limit selector) field value for a given RMA size.
9e368f29 1704 Assumes POWER7 or PPC970. */
aa04b4cc
PM		 1705static inline int lpcr_rmls(unsigned long rma_size)
1706{
1707 switch (rma_size) {
1708 case 32ul << 20: /* 32 MB */
9e368f29
PM		 1709 if (cpu_has_feature(CPU_FTR_ARCH_206))
1710 return 8; /* only supported on POWER7 */
1711 return -1;
aa04b4cc
PM		 1712 case 64ul << 20: /* 64 MB */
1713 return 3;
1714 case 128ul << 20: /* 128 MB */
1715 return 7;
1716 case 256ul << 20: /* 256 MB */
1717 return 4;
1718 case 1ul << 30: /* 1 GB */
1719 return 2;
1720 case 16ul << 30: /* 16 GB */
1721 return 1;
1722 case 256ul << 30: /* 256 GB */
1723 return 0;
1724 default:
1725 return -1;
1726 }
1727}
1728
1729static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1730{
aa04b4cc 1731 struct page *page;
6c45b810 1732 struct kvm_rma_info *ri = vma->vm_file->private_data;
aa04b4cc 1733
6c45b810 1734 if (vmf->pgoff >= kvm_rma_pages)
aa04b4cc
PM		 1735 return VM_FAULT_SIGBUS;
1736
1737 page = pfn_to_page(ri->base_pfn + vmf->pgoff);
1738 get_page(page);
1739 vmf->page = page;
1740 return 0;
1741}
1742
1743static const struct vm_operations_struct kvm_rma_vm_ops = {
1744 .fault = kvm_rma_fault,
1745};
1746
1747static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma)
1748{
314e51b9 1749 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
aa04b4cc
PM		 1750 vma->vm_ops = &kvm_rma_vm_ops;
1751 return 0;
1752}
1753
1754static int kvm_rma_release(struct inode *inode, struct file *filp)
1755{
6c45b810 1756 struct kvm_rma_info *ri = filp->private_data;
aa04b4cc
PM		 1757
1758 kvm_release_rma(ri);
1759 return 0;
1760}
1761
75ef9de1 1762static const struct file_operations kvm_rma_fops = {
aa04b4cc
PM		 1763 .mmap = kvm_rma_mmap,
1764 .release = kvm_rma_release,
1765};
1766
3a167bea
AK		 1767static long kvm_vm_ioctl_allocate_rma(struct kvm *kvm,
1768 struct kvm_allocate_rma *ret)
aa04b4cc 1769{
aa04b4cc 1770 long fd;
6c45b810
AK		 1771 struct kvm_rma_info *ri;
1772 /*
1773 * Only do this on PPC970 in HV mode
1774 */
1775 if (!cpu_has_feature(CPU_FTR_HVMODE) ||
1776 !cpu_has_feature(CPU_FTR_ARCH_201))
1777 return -EINVAL;
1778
1779 if (!kvm_rma_pages)
1780 return -EINVAL;
aa04b4cc
PM		 1781
1782 ri = kvm_alloc_rma();
1783 if (!ri)
1784 return -ENOMEM;
1785
2f84d5ea 1786 fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR | O_CLOEXEC);
aa04b4cc
PM		 1787 if (fd < 0)
1788 kvm_release_rma(ri);
1789
6c45b810 1790 ret->rma_size = kvm_rma_pages << PAGE_SHIFT;
aa04b4cc
PM		 1791 return fd;
1792}
1793
5b74716e
BH		 1794static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
1795 int linux_psize)
1796{
1797 struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
1798
1799 if (!def->shift)
1800 return;
1801 (*sps)->page_shift = def->shift;
1802 (*sps)->slb_enc = def->sllp;
1803 (*sps)->enc[0].page_shift = def->shift;
b1022fbd
AK		 1804 /*
1805 * Only return base page encoding. We don't want to return
1806 * all the supporting pte_enc, because our H_ENTER doesn't
1807 * support MPSS yet. Once they do, we can start passing all
1808 * support pte_enc here
1809 */
1810 (*sps)->enc[0].pte_enc = def->penc[linux_psize];
5b74716e
BH		 1811 (*sps)++;
1812}
1813
3a167bea
AK		 1814static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
1815 struct kvm_ppc_smmu_info *info)
5b74716e
BH		 1816{
1817 struct kvm_ppc_one_seg_page_size *sps;
1818
1819 info->flags = KVM_PPC_PAGE_SIZES_REAL;
1820 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1821 info->flags |= KVM_PPC_1T_SEGMENTS;
1822 info->slb_size = mmu_slb_size;
1823
1824 /* We only support these sizes for now, and no muti-size segments */
1825 sps = &info->sps[0];
1826 kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
1827 kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
1828 kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
1829
1830 return 0;
1831}
1832
82ed3616
PM		 1833/*
1834 * Get (and clear) the dirty memory log for a memory slot.
1835 */
3a167bea
AK		 1836static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
1837 struct kvm_dirty_log *log)
82ed3616
PM		 1838{
1839 struct kvm_memory_slot *memslot;
1840 int r;
1841 unsigned long n;
1842
1843 mutex_lock(&kvm->slots_lock);
1844
1845 r = -EINVAL;
bbacc0c1 1846 if (log->slot >= KVM_USER_MEM_SLOTS)
82ed3616
PM		 1847 goto out;
1848
1849 memslot = id_to_memslot(kvm->memslots, log->slot);
1850 r = -ENOENT;
1851 if (!memslot->dirty_bitmap)
1852 goto out;
1853
1854 n = kvm_dirty_bitmap_bytes(memslot);
1855 memset(memslot->dirty_bitmap, 0, n);
1856
dfe49dbd 1857 r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
82ed3616
PM		 1858 if (r)
1859 goto out;
1860
1861 r = -EFAULT;
1862 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1863 goto out;
1864
1865 r = 0;
1866out:
1867 mutex_unlock(&kvm->slots_lock);
1868 return r;
1869}
1870
a66b48c3 1871static void unpin_slot(struct kvm_memory_slot *memslot)
de56a948 1872{
a66b48c3
PM		 1873 unsigned long *physp;
1874 unsigned long j, npages, pfn;
1875 struct page *page;
aa04b4cc 1876
a66b48c3
PM		 1877 physp = memslot->arch.slot_phys;
1878 npages = memslot->npages;
1879 if (!physp)
1880 return;
1881 for (j = 0; j < npages; j++) {
1882 if (!(physp[j] & KVMPPC_GOT_PAGE))
1883 continue;
1884 pfn = physp[j] >> PAGE_SHIFT;
1885 page = pfn_to_page(pfn);
1886 SetPageDirty(page);
1887 put_page(page);
1888 }
1889}
1890
3a167bea
AK		 1891static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
1892 struct kvm_memory_slot *dont)
a66b48c3
PM		 1893{
1894 if (!dont || free->arch.rmap != dont->arch.rmap) {
1895 vfree(free->arch.rmap);
1896 free->arch.rmap = NULL;
b2b2f165 1897 }
a66b48c3
PM		 1898 if (!dont || free->arch.slot_phys != dont->arch.slot_phys) {
1899 unpin_slot(free);
1900 vfree(free->arch.slot_phys);
1901 free->arch.slot_phys = NULL;
1902 }
1903}
1904
3a167bea
AK		 1905static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
1906 unsigned long npages)
a66b48c3
PM		 1907{
1908 slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
1909 if (!slot->arch.rmap)
1910 return -ENOMEM;
1911 slot->arch.slot_phys = NULL;
aa04b4cc 1912
c77162de
PM		 1913 return 0;
1914}
aa04b4cc 1915
3a167bea
AK		 1916static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
1917 struct kvm_memory_slot *memslot,
1918 struct kvm_userspace_memory_region *mem)
c77162de 1919{
a66b48c3 1920 unsigned long *phys;
c77162de 1921
a66b48c3
PM		 1922 /* Allocate a slot_phys array if needed */
1923 phys = memslot->arch.slot_phys;
1924 if (!kvm->arch.using_mmu_notifiers && !phys && memslot->npages) {
1925 phys = vzalloc(memslot->npages * sizeof(unsigned long));
1926 if (!phys)
1927 return -ENOMEM;
1928 memslot->arch.slot_phys = phys;
aa04b4cc 1929 }
a66b48c3
PM		 1930
1931 return 0;
c77162de
PM		 1932}
1933
3a167bea
AK		 1934static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
1935 struct kvm_userspace_memory_region *mem,
1936 const struct kvm_memory_slot *old)
c77162de 1937{
dfe49dbd
PM		 1938 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
1939 struct kvm_memory_slot *memslot;
1940
8482644a 1941 if (npages && old->npages) {
dfe49dbd
PM		 1942 /*
1943 * If modifying a memslot, reset all the rmap dirty bits.
1944 * If this is a new memslot, we don't need to do anything
1945 * since the rmap array starts out as all zeroes,
1946 * i.e. no pages are dirty.
1947 */
1948 memslot = id_to_memslot(kvm->memslots, mem->slot);
1949 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
1950 }
c77162de
PM		 1951}
1952
a0144e2a
PM		 1953/*
1954 * Update LPCR values in kvm->arch and in vcores.
1955 * Caller must hold kvm->lock.
1956 */
1957void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
1958{
1959 long int i;
1960 u32 cores_done = 0;
1961
1962 if ((kvm->arch.lpcr & mask) == lpcr)
1963 return;
1964
1965 kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
1966
1967 for (i = 0; i < KVM_MAX_VCORES; ++i) {
1968 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
1969 if (!vc)
1970 continue;
1971 spin_lock(&vc->lock);
1972 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
1973 spin_unlock(&vc->lock);
1974 if (++cores_done >= kvm->arch.online_vcores)
1975 break;
1976 }
1977}
1978
3a167bea
AK		 1979static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
1980{
1981 return;
1982}
1983
32fad281 1984static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
c77162de
PM		 1985{
1986 int err = 0;
1987 struct kvm *kvm = vcpu->kvm;
6c45b810 1988 struct kvm_rma_info *ri = NULL;
c77162de
PM		 1989 unsigned long hva;
1990 struct kvm_memory_slot *memslot;
1991 struct vm_area_struct *vma;
a0144e2a
PM		 1992 unsigned long lpcr = 0, senc;
1993 unsigned long lpcr_mask = 0;
c77162de
PM		 1994 unsigned long psize, porder;
1995 unsigned long rma_size;
1996 unsigned long rmls;
1997 unsigned long *physp;
da9d1d7f 1998 unsigned long i, npages;
2c9097e4 1999 int srcu_idx;
c77162de
PM		 2000
2001 mutex_lock(&kvm->lock);
2002 if (kvm->arch.rma_setup_done)
2003 goto out; /* another vcpu beat us to it */
aa04b4cc 2004
32fad281
PM		 2005 /* Allocate hashed page table (if not done already) and reset it */
2006 if (!kvm->arch.hpt_virt) {
2007 err = kvmppc_alloc_hpt(kvm, NULL);
2008 if (err) {
2009 pr_err("KVM: Couldn't alloc HPT\n");
2010 goto out;
2011 }
2012 }
2013
c77162de 2014 /* Look up the memslot for guest physical address 0 */
2c9097e4 2015 srcu_idx = srcu_read_lock(&kvm->srcu);
c77162de 2016 memslot = gfn_to_memslot(kvm, 0);
aa04b4cc 2017
c77162de
PM		 2018 /* We must have some memory at 0 by now */
2019 err = -EINVAL;
2020 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
2c9097e4 2021 goto out_srcu;
c77162de
PM		 2022
2023 /* Look up the VMA for the start of this memory slot */
2024 hva = memslot->userspace_addr;
2025 down_read(&current->mm->mmap_sem);
2026 vma = find_vma(current->mm, hva);
2027 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
2028 goto up_out;
2029
2030 psize = vma_kernel_pagesize(vma);
da9d1d7f 2031 porder = __ilog2(psize);
c77162de
PM		 2032
2033 /* Is this one of our preallocated RMAs? */
2034 if (vma->vm_file && vma->vm_file->f_op == &kvm_rma_fops &&
2035 hva == vma->vm_start)
2036 ri = vma->vm_file->private_data;
2037
2038 up_read(&current->mm->mmap_sem);
2039
2040 if (!ri) {
2041 /* On POWER7, use VRMA; on PPC970, give up */
2042 err = -EPERM;
2043 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2044 pr_err("KVM: CPU requires an RMO\n");
2c9097e4 2045 goto out_srcu;
c77162de
PM		 2046 }
2047
da9d1d7f
PM		 2048 /* We can handle 4k, 64k or 16M pages in the VRMA */
2049 err = -EINVAL;
2050 if (!(psize == 0x1000 || psize == 0x10000 ||
2051 psize == 0x1000000))
2c9097e4 2052 goto out_srcu;
da9d1d7f 2053
c77162de 2054 /* Update VRMASD field in the LPCR */
da9d1d7f 2055 senc = slb_pgsize_encoding(psize);
697d3899
PM		 2056 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
2057 (VRMA_VSID << SLB_VSID_SHIFT_1T);
a0144e2a
PM		 2058 lpcr_mask = LPCR_VRMASD;
2059 /* the -4 is to account for senc values starting at 0x10 */
2060 lpcr = senc << (LPCR_VRMASD_SH - 4);
c77162de
PM		 2061
2062 /* Create HPTEs in the hash page table for the VRMA */
da9d1d7f 2063 kvmppc_map_vrma(vcpu, memslot, porder);
c77162de
PM		 2064
2065 } else {
2066 /* Set up to use an RMO region */
6c45b810 2067 rma_size = kvm_rma_pages;
c77162de
PM		 2068 if (rma_size > memslot->npages)
2069 rma_size = memslot->npages;
2070 rma_size <<= PAGE_SHIFT;
aa04b4cc 2071 rmls = lpcr_rmls(rma_size);
c77162de 2072 err = -EINVAL;
5d226ae5 2073 if ((long)rmls < 0) {
c77162de 2074 pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size);
2c9097e4 2075 goto out_srcu;
aa04b4cc
PM		 2076 }
2077 atomic_inc(&ri->use_count);
2078 kvm->arch.rma = ri;
9e368f29
PM		 2079
2080 /* Update LPCR and RMOR */
9e368f29
PM		 2081 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2082 /* PPC970; insert RMLS value (split field) in HID4 */
a0144e2a
PM		 2083 lpcr_mask = (1ul << HID4_RMLS0_SH) |
2084 (3ul << HID4_RMLS2_SH) | HID4_RMOR;
2085 lpcr = ((rmls >> 2) << HID4_RMLS0_SH) |
9e368f29
PM		 2086 ((rmls & 3) << HID4_RMLS2_SH);
2087 /* RMOR is also in HID4 */
2088 lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff)
2089 << HID4_RMOR_SH;
2090 } else {
2091 /* POWER7 */
a0144e2a
PM		 2092 lpcr_mask = LPCR_VPM0 | LPCR_VRMA_L | LPCR_RMLS;
2093 lpcr = rmls << LPCR_RMLS_SH;
6c45b810 2094 kvm->arch.rmor = ri->base_pfn << PAGE_SHIFT;
9e368f29 2095 }
c77162de 2096 pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n",
aa04b4cc 2097 ri->base_pfn << PAGE_SHIFT, rma_size, lpcr);
aa04b4cc 2098
c77162de 2099 /* Initialize phys addrs of pages in RMO */
6c45b810 2100 npages = kvm_rma_pages;
da9d1d7f 2101 porder = __ilog2(npages);
a66b48c3
PM		 2102 physp = memslot->arch.slot_phys;
2103 if (physp) {
2104 if (npages > memslot->npages)
2105 npages = memslot->npages;
2106 spin_lock(&kvm->arch.slot_phys_lock);
2107 for (i = 0; i < npages; ++i)
2108 physp[i] = ((ri->base_pfn + i) << PAGE_SHIFT) +
2109 porder;
2110 spin_unlock(&kvm->arch.slot_phys_lock);
2111 }
aa04b4cc
PM		 2112 }
2113
a0144e2a
PM		 2114 kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
2115
c77162de
PM		 2116 /* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */
2117 smp_wmb();
2118 kvm->arch.rma_setup_done = 1;
2119 err = 0;
2c9097e4
PM		 2120 out_srcu:
2121 srcu_read_unlock(&kvm->srcu, srcu_idx);
c77162de
PM		 2122 out:
2123 mutex_unlock(&kvm->lock);
2124 return err;
b2b2f165 2125
c77162de
PM		 2126 up_out:
2127 up_read(&current->mm->mmap_sem);
505d6421 2128 goto out_srcu;
de56a948
PM		 2129}
2130
3a167bea 2131static int kvmppc_core_init_vm_hv(struct kvm *kvm)
de56a948 2132{
32fad281 2133 unsigned long lpcr, lpid;
de56a948 2134
32fad281
PM		 2135 /* Allocate the guest's logical partition ID */
2136
2137 lpid = kvmppc_alloc_lpid();
5d226ae5 2138 if ((long)lpid < 0)
32fad281
PM		 2139 return -ENOMEM;
2140 kvm->arch.lpid = lpid;
de56a948 2141
1b400ba0
PM		 2142 /*
2143 * Since we don't flush the TLB when tearing down a VM,
2144 * and this lpid might have previously been used,
2145 * make sure we flush on each core before running the new VM.
2146 */
2147 cpumask_setall(&kvm->arch.need_tlb_flush);
2148
aa04b4cc 2149 kvm->arch.rma = NULL;
aa04b4cc 2150
9e368f29 2151 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
aa04b4cc 2152
9e368f29
PM		 2153 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2154 /* PPC970; HID4 is effectively the LPCR */
9e368f29
PM		 2155 kvm->arch.host_lpid = 0;
2156 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4);
2157 lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH));
2158 lpcr |= ((lpid >> 4) << HID4_LPID1_SH) |
2159 ((lpid & 0xf) << HID4_LPID5_SH);
2160 } else {
2161 /* POWER7; init LPCR for virtual RMA mode */
2162 kvm->arch.host_lpid = mfspr(SPRN_LPID);
2163 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
2164 lpcr &= LPCR_PECE | LPCR_LPES;
2165 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
697d3899
PM		 2166 LPCR_VPM0 | LPCR_VPM1;
2167 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
2168 (VRMA_VSID << SLB_VSID_SHIFT_1T);
9e368f29
PM		 2169 }
2170 kvm->arch.lpcr = lpcr;
aa04b4cc 2171
342d3db7 2172 kvm->arch.using_mmu_notifiers = !!cpu_has_feature(CPU_FTR_ARCH_206);
c77162de 2173 spin_lock_init(&kvm->arch.slot_phys_lock);
512691d4
PM		 2174
2175 /*
2176 * Don't allow secondary CPU threads to come online
2177 * while any KVM VMs exist.
2178 */
2179 inhibit_secondary_onlining();
2180
54738c09 2181 return 0;
de56a948
PM		 2182}
2183
f1378b1c
PM		 2184static void kvmppc_free_vcores(struct kvm *kvm)
2185{
2186 long int i;
2187
2188 for (i = 0; i < KVM_MAX_VCORES; ++i)
2189 kfree(kvm->arch.vcores[i]);
2190 kvm->arch.online_vcores = 0;
2191}
2192
3a167bea 2193static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
de56a948 2194{
512691d4
PM		 2195 uninhibit_secondary_onlining();
2196
f1378b1c 2197 kvmppc_free_vcores(kvm);
aa04b4cc
PM		 2198 if (kvm->arch.rma) {
2199 kvm_release_rma(kvm->arch.rma);
2200 kvm->arch.rma = NULL;
2201 }
2202
de56a948
PM		 2203 kvmppc_free_hpt(kvm);
2204}
2205
3a167bea
AK		 2206/* We don't need to emulate any privileged instructions or dcbz */
2207static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
2208 unsigned int inst, int *advance)
de56a948 2209{
3a167bea 2210 return EMULATE_FAIL;
de56a948
PM		 2211}
2212
3a167bea
AK		 2213static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
2214 ulong spr_val)
de56a948
PM		 2215{
2216 return EMULATE_FAIL;
2217}
2218
3a167bea
AK		 2219static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
2220 ulong *spr_val)
de56a948
PM		 2221{
2222 return EMULATE_FAIL;
2223}
2224
3a167bea 2225static int kvmppc_core_check_processor_compat_hv(void)
de56a948 2226{
3a167bea
AK		 2227 if (!cpu_has_feature(CPU_FTR_HVMODE))
2228 return -EIO;
2229 return 0;
de56a948
PM		 2230}
2231
3a167bea
AK		 2232static long kvm_arch_vm_ioctl_hv(struct file *filp,
2233 unsigned int ioctl, unsigned long arg)
2234{
2235 struct kvm *kvm __maybe_unused = filp->private_data;
2236 void __user *argp = (void __user *)arg;
2237 long r;
2238
2239 switch (ioctl) {
2240
2241 case KVM_ALLOCATE_RMA: {
2242 struct kvm_allocate_rma rma;
2243 struct kvm *kvm = filp->private_data;
2244
2245 r = kvm_vm_ioctl_allocate_rma(kvm, &rma);
2246 if (r >= 0 && copy_to_user(argp, &rma, sizeof(rma)))
2247 r = -EFAULT;
2248 break;
2249 }
2250
2251 case KVM_PPC_ALLOCATE_HTAB: {
2252 u32 htab_order;
2253
2254 r = -EFAULT;
2255 if (get_user(htab_order, (u32 __user *)argp))
2256 break;
2257 r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
2258 if (r)
2259 break;
2260 r = -EFAULT;
2261 if (put_user(htab_order, (u32 __user *)argp))
2262 break;
2263 r = 0;
2264 break;
2265 }
2266
2267 case KVM_PPC_GET_HTAB_FD: {
2268 struct kvm_get_htab_fd ghf;
2269
2270 r = -EFAULT;
2271 if (copy_from_user(&ghf, argp, sizeof(ghf)))
2272 break;
2273 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
2274 break;
2275 }
2276
2277 default:
2278 r = -ENOTTY;
2279 }
2280
2281 return r;
2282}
2283
cbbc58d4 2284static struct kvmppc_ops kvm_ops_hv = {
3a167bea
AK		 2285 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
2286 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
2287 .get_one_reg = kvmppc_get_one_reg_hv,
2288 .set_one_reg = kvmppc_set_one_reg_hv,
2289 .vcpu_load = kvmppc_core_vcpu_load_hv,
2290 .vcpu_put = kvmppc_core_vcpu_put_hv,
2291 .set_msr = kvmppc_set_msr_hv,
2292 .vcpu_run = kvmppc_vcpu_run_hv,
2293 .vcpu_create = kvmppc_core_vcpu_create_hv,
2294 .vcpu_free = kvmppc_core_vcpu_free_hv,
2295 .check_requests = kvmppc_core_check_requests_hv,
2296 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv,
2297 .flush_memslot = kvmppc_core_flush_memslot_hv,
2298 .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
2299 .commit_memory_region = kvmppc_core_commit_memory_region_hv,
2300 .unmap_hva = kvm_unmap_hva_hv,
2301 .unmap_hva_range = kvm_unmap_hva_range_hv,
2302 .age_hva = kvm_age_hva_hv,
2303 .test_age_hva = kvm_test_age_hva_hv,
2304 .set_spte_hva = kvm_set_spte_hva_hv,
2305 .mmu_destroy = kvmppc_mmu_destroy_hv,
2306 .free_memslot = kvmppc_core_free_memslot_hv,
2307 .create_memslot = kvmppc_core_create_memslot_hv,
2308 .init_vm = kvmppc_core_init_vm_hv,
2309 .destroy_vm = kvmppc_core_destroy_vm_hv,
3a167bea
AK		 2310 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
2311 .emulate_op = kvmppc_core_emulate_op_hv,
2312 .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
2313 .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
2314 .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
2315 .arch_vm_ioctl = kvm_arch_vm_ioctl_hv,
2316};
2317
2318static int kvmppc_book3s_init_hv(void)
de56a948
PM		 2319{
2320 int r;
cbbc58d4
AK		 2321 /*
2322 * FIXME!! Do we need to check on all cpus ?
2323 */
2324 r = kvmppc_core_check_processor_compat_hv();
2325 if (r < 0)
de56a948
PM		 2326 return r;
2327
cbbc58d4
AK		 2328 kvm_ops_hv.owner = THIS_MODULE;
2329 kvmppc_hv_ops = &kvm_ops_hv;
de56a948 2330
cbbc58d4 2331 r = kvmppc_mmu_hv_init();
de56a948
PM		 2332 return r;
2333}
2334
3a167bea 2335static void kvmppc_book3s_exit_hv(void)
de56a948 2336{
cbbc58d4 2337 kvmppc_hv_ops = NULL;
de56a948
PM		 2338}
2339
3a167bea
AK		 2340module_init(kvmppc_book3s_init_hv);
2341module_exit(kvmppc_book3s_exit_hv);
2ba9f0d8 2342MODULE_LICENSE("GPL");
398a76c6
AG		 2343MODULE_ALIAS_MISCDEV(KVM_MINOR);
2344MODULE_ALIAS("devname:kvm");
Ubuntu Artful kernel mirror