projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: PPC: Book3S HV: Improve real-mode handling of external interrupts
[GitHub/mt8127/android_kernel_alcatel_ttab.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>
de56a948
PM		 34
35#include <asm/reg.h>
36#include <asm/cputable.h>
37#include <asm/cacheflush.h>
38#include <asm/tlbflush.h>
39#include <asm/uaccess.h>
40#include <asm/io.h>
41#include <asm/kvm_ppc.h>
42#include <asm/kvm_book3s.h>
43#include <asm/mmu_context.h>
44#include <asm/lppaca.h>
45#include <asm/processor.h>
371fefd6 46#include <asm/cputhreads.h>
aa04b4cc 47#include <asm/page.h>
de1d9248 48#include <asm/hvcall.h>
ae3a197e 49#include <asm/switch_to.h>
512691d4 50#include <asm/smp.h>
de56a948 51#include <linux/gfp.h>
de56a948
PM		 52#include <linux/vmalloc.h>
53#include <linux/highmem.h>
c77162de 54#include <linux/hugetlb.h>
de56a948
PM		 55
56/* #define EXIT_DEBUG */
57/* #define EXIT_DEBUG_SIMPLE */
58/* #define EXIT_DEBUG_INT */
59
913d3ff9
PM		 60/* Used to indicate that a guest page fault needs to be handled */
61#define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1)
62
c7b67670
PM		 63/* Used as a "null" value for timebase values */
64#define TB_NIL (~(u64)0)
65
19ccb76a 66static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
32fad281 67static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
19ccb76a 68
54695c30
BH		 69void kvmppc_fast_vcpu_kick(struct kvm_vcpu *vcpu)
70{
71 int me;
72 int cpu = vcpu->cpu;
73 wait_queue_head_t *wqp;
74
75 wqp = kvm_arch_vcpu_wq(vcpu);
76 if (waitqueue_active(wqp)) {
77 wake_up_interruptible(wqp);
78 ++vcpu->stat.halt_wakeup;
79 }
80
81 me = get_cpu();
82
83 /* CPU points to the first thread of the core */
84 if (cpu != me && cpu >= 0 && cpu < nr_cpu_ids) {
85 int real_cpu = cpu + vcpu->arch.ptid;
86 if (paca[real_cpu].kvm_hstate.xics_phys)
87 xics_wake_cpu(real_cpu);
88 else if (cpu_online(cpu))
89 smp_send_reschedule(cpu);
90 }
91 put_cpu();
92}
93
c7b67670
PM		 94/*
95 * We use the vcpu_load/put functions to measure stolen time.
96 * Stolen time is counted as time when either the vcpu is able to
97 * run as part of a virtual core, but the task running the vcore
98 * is preempted or sleeping, or when the vcpu needs something done
99 * in the kernel by the task running the vcpu, but that task is
100 * preempted or sleeping. Those two things have to be counted
101 * separately, since one of the vcpu tasks will take on the job
102 * of running the core, and the other vcpu tasks in the vcore will
103 * sleep waiting for it to do that, but that sleep shouldn't count
104 * as stolen time.
105 *
106 * Hence we accumulate stolen time when the vcpu can run as part of
107 * a vcore using vc->stolen_tb, and the stolen time when the vcpu
108 * needs its task to do other things in the kernel (for example,
109 * service a page fault) in busy_stolen. We don't accumulate
110 * stolen time for a vcore when it is inactive, or for a vcpu
111 * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of
112 * a misnomer; it means that the vcpu task is not executing in
113 * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
114 * the kernel. We don't have any way of dividing up that time
115 * between time that the vcpu is genuinely stopped, time that
116 * the task is actively working on behalf of the vcpu, and time
117 * that the task is preempted, so we don't count any of it as
118 * stolen.
119 *
120 * Updates to busy_stolen are protected by arch.tbacct_lock;
121 * updates to vc->stolen_tb are protected by the arch.tbacct_lock
122 * of the vcpu that has taken responsibility for running the vcore
123 * (i.e. vc->runner). The stolen times are measured in units of
124 * timebase ticks. (Note that the != TB_NIL checks below are
125 * purely defensive; they should never fail.)
126 */
127
de56a948
PM		 128void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
129{
0456ec4f
PM		 130 struct kvmppc_vcore *vc = vcpu->arch.vcore;
131
c7b67670
PM		 132 spin_lock(&vcpu->arch.tbacct_lock);
133 if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE &&
134 vc->preempt_tb != TB_NIL) {
0456ec4f 135 vc->stolen_tb += mftb() - vc->preempt_tb;
c7b67670
PM		 136 vc->preempt_tb = TB_NIL;
137 }
138 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
139 vcpu->arch.busy_preempt != TB_NIL) {
140 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
141 vcpu->arch.busy_preempt = TB_NIL;
142 }
143 spin_unlock(&vcpu->arch.tbacct_lock);
de56a948
PM		 144}
145
146void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
147{
0456ec4f
PM		 148 struct kvmppc_vcore *vc = vcpu->arch.vcore;
149
c7b67670 150 spin_lock(&vcpu->arch.tbacct_lock);
0456ec4f
PM		 151 if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE)
152 vc->preempt_tb = mftb();
c7b67670
PM		 153 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
154 vcpu->arch.busy_preempt = mftb();
155 spin_unlock(&vcpu->arch.tbacct_lock);
de56a948
PM		 156}
157
de56a948
PM		 158void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
159{
160 vcpu->arch.shregs.msr = msr;
19ccb76a 161 kvmppc_end_cede(vcpu);
de56a948
PM		 162}
163
164void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
165{
166 vcpu->arch.pvr = pvr;
167}
168
169void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
170{
171 int r;
172
173 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
174 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
175 vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
176 for (r = 0; r < 16; ++r)
177 pr_err("r%2d = %.16lx r%d = %.16lx\n",
178 r, kvmppc_get_gpr(vcpu, r),
179 r+16, kvmppc_get_gpr(vcpu, r+16));
180 pr_err("ctr = %.16lx lr = %.16lx\n",
181 vcpu->arch.ctr, vcpu->arch.lr);
182 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
183 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
184 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
185 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
186 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
187 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
188 pr_err("cr = %.8x xer = %.16lx dsisr = %.8x\n",
189 vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
190 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
191 pr_err("fault dar = %.16lx dsisr = %.8x\n",
192 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
193 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
194 for (r = 0; r < vcpu->arch.slb_max; ++r)
195 pr_err(" ESID = %.16llx VSID = %.16llx\n",
196 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
197 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
aa04b4cc 198 vcpu->kvm->arch.lpcr, vcpu->kvm->arch.sdr1,
de56a948
PM		 199 vcpu->arch.last_inst);
200}
201
a8606e20
PM		 202struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
203{
204 int r;
205 struct kvm_vcpu *v, *ret = NULL;
206
207 mutex_lock(&kvm->lock);
208 kvm_for_each_vcpu(r, v, kvm) {
209 if (v->vcpu_id == id) {
210 ret = v;
211 break;
212 }
213 }
214 mutex_unlock(&kvm->lock);
215 return ret;
216}
217
218static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
219{
220 vpa->shared_proc = 1;
221 vpa->yield_count = 1;
222}
223
55b665b0
PM		 224static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
225 unsigned long addr, unsigned long len)
226{
227 /* check address is cacheline aligned */
228 if (addr & (L1_CACHE_BYTES - 1))
229 return -EINVAL;
230 spin_lock(&vcpu->arch.vpa_update_lock);
231 if (v->next_gpa != addr || v->len != len) {
232 v->next_gpa = addr;
233 v->len = addr ? len : 0;
234 v->update_pending = 1;
235 }
236 spin_unlock(&vcpu->arch.vpa_update_lock);
237 return 0;
238}
239
2e25aa5f
PM		 240/* Length for a per-processor buffer is passed in at offset 4 in the buffer */
241struct reg_vpa {
242 u32 dummy;
243 union {
244 u16 hword;
245 u32 word;
246 } length;
247};
248
249static int vpa_is_registered(struct kvmppc_vpa *vpap)
250{
251 if (vpap->update_pending)
252 return vpap->next_gpa != 0;
253 return vpap->pinned_addr != NULL;
254}
255
a8606e20
PM		 256static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
257 unsigned long flags,
258 unsigned long vcpuid, unsigned long vpa)
259{
260 struct kvm *kvm = vcpu->kvm;
93e60249 261 unsigned long len, nb;
a8606e20
PM		 262 void *va;
263 struct kvm_vcpu *tvcpu;
2e25aa5f
PM		 264 int err;
265 int subfunc;
266 struct kvmppc_vpa *vpap;
a8606e20
PM		 267
268 tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
269 if (!tvcpu)
270 return H_PARAMETER;
271
2e25aa5f
PM		 272 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
273 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
274 subfunc == H_VPA_REG_SLB) {
275 /* Registering new area - address must be cache-line aligned */
276 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
a8606e20 277 return H_PARAMETER;
2e25aa5f
PM		 278
279 /* convert logical addr to kernel addr and read length */
93e60249
PM		 280 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
281 if (va == NULL)
b2b2f165 282 return H_PARAMETER;
2e25aa5f
PM		 283 if (subfunc == H_VPA_REG_VPA)
284 len = ((struct reg_vpa *)va)->length.hword;
a8606e20 285 else
2e25aa5f 286 len = ((struct reg_vpa *)va)->length.word;
c35635ef 287 kvmppc_unpin_guest_page(kvm, va, vpa, false);
2e25aa5f
PM		 288
289 /* Check length */
290 if (len > nb || len < sizeof(struct reg_vpa))
291 return H_PARAMETER;
292 } else {
293 vpa = 0;
294 len = 0;
295 }
296
297 err = H_PARAMETER;
298 vpap = NULL;
299 spin_lock(&tvcpu->arch.vpa_update_lock);
300
301 switch (subfunc) {
302 case H_VPA_REG_VPA: /* register VPA */
303 if (len < sizeof(struct lppaca))
a8606e20 304 break;
2e25aa5f
PM		 305 vpap = &tvcpu->arch.vpa;
306 err = 0;
307 break;
308
309 case H_VPA_REG_DTL: /* register DTL */
310 if (len < sizeof(struct dtl_entry))
a8606e20 311 break;
2e25aa5f
PM		 312 len -= len % sizeof(struct dtl_entry);
313
314 /* Check that they have previously registered a VPA */
315 err = H_RESOURCE;
316 if (!vpa_is_registered(&tvcpu->arch.vpa))
a8606e20 317 break;
2e25aa5f
PM		 318
319 vpap = &tvcpu->arch.dtl;
320 err = 0;
321 break;
322
323 case H_VPA_REG_SLB: /* register SLB shadow buffer */
324 /* Check that they have previously registered a VPA */
325 err = H_RESOURCE;
326 if (!vpa_is_registered(&tvcpu->arch.vpa))
a8606e20 327 break;
2e25aa5f
PM		 328
329 vpap = &tvcpu->arch.slb_shadow;
330 err = 0;
331 break;
332
333 case H_VPA_DEREG_VPA: /* deregister VPA */
334 /* Check they don't still have a DTL or SLB buf registered */
335 err = H_RESOURCE;
336 if (vpa_is_registered(&tvcpu->arch.dtl) ||
337 vpa_is_registered(&tvcpu->arch.slb_shadow))
a8606e20 338 break;
2e25aa5f
PM		 339
340 vpap = &tvcpu->arch.vpa;
341 err = 0;
342 break;
343
344 case H_VPA_DEREG_DTL: /* deregister DTL */
345 vpap = &tvcpu->arch.dtl;
346 err = 0;
347 break;
348
349 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */
350 vpap = &tvcpu->arch.slb_shadow;
351 err = 0;
352 break;
353 }
354
355 if (vpap) {
356 vpap->next_gpa = vpa;
357 vpap->len = len;
358 vpap->update_pending = 1;
a8606e20 359 }
93e60249 360
2e25aa5f
PM		 361 spin_unlock(&tvcpu->arch.vpa_update_lock);
362
93e60249 363 return err;
a8606e20
PM		 364}
365
081f323b 366static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
2e25aa5f 367{
081f323b 368 struct kvm *kvm = vcpu->kvm;
2e25aa5f
PM		 369 void *va;
370 unsigned long nb;
081f323b 371 unsigned long gpa;
2e25aa5f 372
081f323b
PM		 373 /*
374 * We need to pin the page pointed to by vpap->next_gpa,
375 * but we can't call kvmppc_pin_guest_page under the lock
376 * as it does get_user_pages() and down_read(). So we
377 * have to drop the lock, pin the page, then get the lock
378 * again and check that a new area didn't get registered
379 * in the meantime.
380 */
381 for (;;) {
382 gpa = vpap->next_gpa;
383 spin_unlock(&vcpu->arch.vpa_update_lock);
384 va = NULL;
385 nb = 0;
386 if (gpa)
c35635ef 387 va = kvmppc_pin_guest_page(kvm, gpa, &nb);
081f323b
PM		 388 spin_lock(&vcpu->arch.vpa_update_lock);
389 if (gpa == vpap->next_gpa)
390 break;
391 /* sigh... unpin that one and try again */
392 if (va)
c35635ef 393 kvmppc_unpin_guest_page(kvm, va, gpa, false);
081f323b
PM		 394 }
395
396 vpap->update_pending = 0;
397 if (va && nb < vpap->len) {
398 /*
399 * If it's now too short, it must be that userspace
400 * has changed the mappings underlying guest memory,
401 * so unregister the region.
402 */
c35635ef 403 kvmppc_unpin_guest_page(kvm, va, gpa, false);
081f323b 404 va = NULL;
2e25aa5f
PM		 405 }
406 if (vpap->pinned_addr)
c35635ef
PM		 407 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
408 vpap->dirty);
409 vpap->gpa = gpa;
2e25aa5f 410 vpap->pinned_addr = va;
c35635ef 411 vpap->dirty = false;
2e25aa5f
PM		 412 if (va)
413 vpap->pinned_end = va + vpap->len;
414}
415
416static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
417{
2f12f034
PM		 418 if (!(vcpu->arch.vpa.update_pending ||
419 vcpu->arch.slb_shadow.update_pending ||
420 vcpu->arch.dtl.update_pending))
421 return;
422
2e25aa5f
PM		 423 spin_lock(&vcpu->arch.vpa_update_lock);
424 if (vcpu->arch.vpa.update_pending) {
081f323b 425 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
55b665b0
PM		 426 if (vcpu->arch.vpa.pinned_addr)
427 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
2e25aa5f
PM		 428 }
429 if (vcpu->arch.dtl.update_pending) {
081f323b 430 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
2e25aa5f
PM		 431 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
432 vcpu->arch.dtl_index = 0;
433 }
434 if (vcpu->arch.slb_shadow.update_pending)
081f323b 435 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
2e25aa5f
PM		 436 spin_unlock(&vcpu->arch.vpa_update_lock);
437}
438
c7b67670
PM		 439/*
440 * Return the accumulated stolen time for the vcore up until `now'.
441 * The caller should hold the vcore lock.
442 */
443static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
444{
445 u64 p;
446
447 /*
448 * If we are the task running the vcore, then since we hold
449 * the vcore lock, we can't be preempted, so stolen_tb/preempt_tb
450 * can't be updated, so we don't need the tbacct_lock.
451 * If the vcore is inactive, it can't become active (since we
452 * hold the vcore lock), so the vcpu load/put functions won't
453 * update stolen_tb/preempt_tb, and we don't need tbacct_lock.
454 */
455 if (vc->vcore_state != VCORE_INACTIVE &&
456 vc->runner->arch.run_task != current) {
457 spin_lock(&vc->runner->arch.tbacct_lock);
458 p = vc->stolen_tb;
459 if (vc->preempt_tb != TB_NIL)
460 p += now - vc->preempt_tb;
461 spin_unlock(&vc->runner->arch.tbacct_lock);
462 } else {
463 p = vc->stolen_tb;
464 }
465 return p;
466}
467
0456ec4f
PM		 468static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
469 struct kvmppc_vcore *vc)
470{
471 struct dtl_entry *dt;
472 struct lppaca *vpa;
c7b67670
PM		 473 unsigned long stolen;
474 unsigned long core_stolen;
475 u64 now;
0456ec4f
PM		 476
477 dt = vcpu->arch.dtl_ptr;
478 vpa = vcpu->arch.vpa.pinned_addr;
c7b67670
PM		 479 now = mftb();
480 core_stolen = vcore_stolen_time(vc, now);
481 stolen = core_stolen - vcpu->arch.stolen_logged;
482 vcpu->arch.stolen_logged = core_stolen;
483 spin_lock(&vcpu->arch.tbacct_lock);
484 stolen += vcpu->arch.busy_stolen;
485 vcpu->arch.busy_stolen = 0;
486 spin_unlock(&vcpu->arch.tbacct_lock);
0456ec4f
PM		 487 if (!dt || !vpa)
488 return;
489 memset(dt, 0, sizeof(struct dtl_entry));
490 dt->dispatch_reason = 7;
491 dt->processor_id = vc->pcpu + vcpu->arch.ptid;
c7b67670
PM		 492 dt->timebase = now;
493 dt->enqueue_to_dispatch_time = stolen;
0456ec4f
PM		 494 dt->srr0 = kvmppc_get_pc(vcpu);
495 dt->srr1 = vcpu->arch.shregs.msr;
496 ++dt;
497 if (dt == vcpu->arch.dtl.pinned_end)
498 dt = vcpu->arch.dtl.pinned_addr;
499 vcpu->arch.dtl_ptr = dt;
500 /* order writing *dt vs. writing vpa->dtl_idx */
501 smp_wmb();
502 vpa->dtl_idx = ++vcpu->arch.dtl_index;
c35635ef 503 vcpu->arch.dtl.dirty = true;
0456ec4f
PM		 504}
505
a8606e20
PM		 506int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
507{
508 unsigned long req = kvmppc_get_gpr(vcpu, 3);
509 unsigned long target, ret = H_SUCCESS;
510 struct kvm_vcpu *tvcpu;
8e591cb7 511 int idx, rc;
a8606e20
PM		 512
513 switch (req) {
c77162de 514 case H_ENTER:
2c9097e4 515 idx = srcu_read_lock(&vcpu->kvm->srcu);
c77162de
PM		 516 ret = kvmppc_virtmode_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
517 kvmppc_get_gpr(vcpu, 5),
518 kvmppc_get_gpr(vcpu, 6),
519 kvmppc_get_gpr(vcpu, 7));
2c9097e4 520 srcu_read_unlock(&vcpu->kvm->srcu, idx);
c77162de 521 break;
a8606e20 522 case H_CEDE:
a8606e20
PM		 523 break;
524 case H_PROD:
525 target = kvmppc_get_gpr(vcpu, 4);
526 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
527 if (!tvcpu) {
528 ret = H_PARAMETER;
529 break;
530 }
531 tvcpu->arch.prodded = 1;
532 smp_mb();
533 if (vcpu->arch.ceded) {
534 if (waitqueue_active(&vcpu->wq)) {
535 wake_up_interruptible(&vcpu->wq);
536 vcpu->stat.halt_wakeup++;
537 }
538 }
539 break;
540 case H_CONFER:
541 break;
542 case H_REGISTER_VPA:
543 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
544 kvmppc_get_gpr(vcpu, 5),
545 kvmppc_get_gpr(vcpu, 6));
546 break;
8e591cb7
ME		 547 case H_RTAS:
548 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
549 return RESUME_HOST;
550
551 rc = kvmppc_rtas_hcall(vcpu);
552
553 if (rc == -ENOENT)
554 return RESUME_HOST;
555 else if (rc == 0)
556 break;
557
558 /* Send the error out to userspace via KVM_RUN */
559 return rc;
bc5ad3f3
BH		 560
561 case H_XIRR:
562 case H_CPPR:
563 case H_EOI:
564 case H_IPI:
565 if (kvmppc_xics_enabled(vcpu)) {
566 ret = kvmppc_xics_hcall(vcpu, req);
567 break;
568 } /* fallthrough */
a8606e20
PM		 569 default:
570 return RESUME_HOST;
571 }
572 kvmppc_set_gpr(vcpu, 3, ret);
573 vcpu->arch.hcall_needed = 0;
574 return RESUME_GUEST;
575}
576
de56a948
PM		 577static int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
578 struct task_struct *tsk)
579{
580 int r = RESUME_HOST;
581
582 vcpu->stat.sum_exits++;
583
584 run->exit_reason = KVM_EXIT_UNKNOWN;
585 run->ready_for_interrupt_injection = 1;
586 switch (vcpu->arch.trap) {
587 /* We're good on these - the host merely wanted to get our attention */
588 case BOOK3S_INTERRUPT_HV_DECREMENTER:
589 vcpu->stat.dec_exits++;
590 r = RESUME_GUEST;
591 break;
592 case BOOK3S_INTERRUPT_EXTERNAL:
593 vcpu->stat.ext_intr_exits++;
594 r = RESUME_GUEST;
595 break;
596 case BOOK3S_INTERRUPT_PERFMON:
597 r = RESUME_GUEST;
598 break;
b4072df4
PM		 599 case BOOK3S_INTERRUPT_MACHINE_CHECK:
600 /*
601 * Deliver a machine check interrupt to the guest.
602 * We have to do this, even if the host has handled the
603 * machine check, because machine checks use SRR0/1 and
604 * the interrupt might have trashed guest state in them.
605 */
606 kvmppc_book3s_queue_irqprio(vcpu,
607 BOOK3S_INTERRUPT_MACHINE_CHECK);
608 r = RESUME_GUEST;
609 break;
de56a948
PM		 610 case BOOK3S_INTERRUPT_PROGRAM:
611 {
612 ulong flags;
613 /*
614 * Normally program interrupts are delivered directly
615 * to the guest by the hardware, but we can get here
616 * as a result of a hypervisor emulation interrupt
617 * (e40) getting turned into a 700 by BML RTAS.
618 */
619 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
620 kvmppc_core_queue_program(vcpu, flags);
621 r = RESUME_GUEST;
622 break;
623 }
624 case BOOK3S_INTERRUPT_SYSCALL:
625 {
626 /* hcall - punt to userspace */
627 int i;
628
629 if (vcpu->arch.shregs.msr & MSR_PR) {
630 /* sc 1 from userspace - reflect to guest syscall */
631 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_SYSCALL);
632 r = RESUME_GUEST;
633 break;
634 }
635 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
636 for (i = 0; i < 9; ++i)
637 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
638 run->exit_reason = KVM_EXIT_PAPR_HCALL;
639 vcpu->arch.hcall_needed = 1;
640 r = RESUME_HOST;
641 break;
642 }
643 /*
342d3db7
PM		 644 * We get these next two if the guest accesses a page which it thinks
645 * it has mapped but which is not actually present, either because
646 * it is for an emulated I/O device or because the corresonding
647 * host page has been paged out. Any other HDSI/HISI interrupts
648 * have been handled already.
de56a948
PM		 649 */
650 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
913d3ff9 651 r = RESUME_PAGE_FAULT;
de56a948
PM		 652 break;
653 case BOOK3S_INTERRUPT_H_INST_STORAGE:
913d3ff9
PM		 654 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
655 vcpu->arch.fault_dsisr = 0;
656 r = RESUME_PAGE_FAULT;
de56a948
PM		 657 break;
658 /*
659 * This occurs if the guest executes an illegal instruction.
660 * We just generate a program interrupt to the guest, since
661 * we don't emulate any guest instructions at this stage.
662 */
663 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
664 kvmppc_core_queue_program(vcpu, 0x80000);
665 r = RESUME_GUEST;
666 break;
667 default:
668 kvmppc_dump_regs(vcpu);
669 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
670 vcpu->arch.trap, kvmppc_get_pc(vcpu),
671 vcpu->arch.shregs.msr);
672 r = RESUME_HOST;
673 BUG();
674 break;
675 }
676
de56a948
PM		 677 return r;
678}
679
680int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
681 struct kvm_sregs *sregs)
682{
683 int i;
684
685 sregs->pvr = vcpu->arch.pvr;
686
687 memset(sregs, 0, sizeof(struct kvm_sregs));
688 for (i = 0; i < vcpu->arch.slb_max; i++) {
689 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
690 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
691 }
692
693 return 0;
694}
695
696int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
697 struct kvm_sregs *sregs)
698{
699 int i, j;
700
701 kvmppc_set_pvr(vcpu, sregs->pvr);
702
703 j = 0;
704 for (i = 0; i < vcpu->arch.slb_nr; i++) {
705 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
706 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
707 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
708 ++j;
709 }
710 }
711 vcpu->arch.slb_max = j;
712
713 return 0;
714}
715
a136a8bd 716int kvmppc_get_one_reg(struct kvm_vcpu *vcpu, u64 id, union kvmppc_one_reg *val)
31f3438e 717{
a136a8bd
PM		 718 int r = 0;
719 long int i;
31f3438e 720
a136a8bd 721 switch (id) {
31f3438e 722 case KVM_REG_PPC_HIOR:
a136a8bd
PM		 723 *val = get_reg_val(id, 0);
724 break;
725 case KVM_REG_PPC_DABR:
726 *val = get_reg_val(id, vcpu->arch.dabr);
727 break;
728 case KVM_REG_PPC_DSCR:
729 *val = get_reg_val(id, vcpu->arch.dscr);
730 break;
731 case KVM_REG_PPC_PURR:
732 *val = get_reg_val(id, vcpu->arch.purr);
733 break;
734 case KVM_REG_PPC_SPURR:
735 *val = get_reg_val(id, vcpu->arch.spurr);
736 break;
737 case KVM_REG_PPC_AMR:
738 *val = get_reg_val(id, vcpu->arch.amr);
739 break;
740 case KVM_REG_PPC_UAMOR:
741 *val = get_reg_val(id, vcpu->arch.uamor);
742 break;
743 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRA:
744 i = id - KVM_REG_PPC_MMCR0;
745 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
746 break;
747 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
748 i = id - KVM_REG_PPC_PMC1;
749 *val = get_reg_val(id, vcpu->arch.pmc[i]);
31f3438e 750 break;
a8bd19ef
PM		 751#ifdef CONFIG_VSX
752 case KVM_REG_PPC_FPR0 ... KVM_REG_PPC_FPR31:
753 if (cpu_has_feature(CPU_FTR_VSX)) {
754 /* VSX => FP reg i is stored in arch.vsr[2*i] */
755 long int i = id - KVM_REG_PPC_FPR0;
756 *val = get_reg_val(id, vcpu->arch.vsr[2 * i]);
757 } else {
758 /* let generic code handle it */
759 r = -EINVAL;
760 }
761 break;
762 case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31:
763 if (cpu_has_feature(CPU_FTR_VSX)) {
764 long int i = id - KVM_REG_PPC_VSR0;
765 val->vsxval[0] = vcpu->arch.vsr[2 * i];
766 val->vsxval[1] = vcpu->arch.vsr[2 * i + 1];
767 } else {
768 r = -ENXIO;
769 }
770 break;
771#endif /* CONFIG_VSX */
55b665b0
PM		 772 case KVM_REG_PPC_VPA_ADDR:
773 spin_lock(&vcpu->arch.vpa_update_lock);
774 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
775 spin_unlock(&vcpu->arch.vpa_update_lock);
776 break;
777 case KVM_REG_PPC_VPA_SLB:
778 spin_lock(&vcpu->arch.vpa_update_lock);
779 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
780 val->vpaval.length = vcpu->arch.slb_shadow.len;
781 spin_unlock(&vcpu->arch.vpa_update_lock);
782 break;
783 case KVM_REG_PPC_VPA_DTL:
784 spin_lock(&vcpu->arch.vpa_update_lock);
785 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
786 val->vpaval.length = vcpu->arch.dtl.len;
787 spin_unlock(&vcpu->arch.vpa_update_lock);
788 break;
31f3438e 789 default:
a136a8bd 790 r = -EINVAL;
31f3438e
PM		 791 break;
792 }
793
794 return r;
795}
796
a136a8bd 797int kvmppc_set_one_reg(struct kvm_vcpu *vcpu, u64 id, union kvmppc_one_reg *val)
31f3438e 798{
a136a8bd
PM		 799 int r = 0;
800 long int i;
55b665b0 801 unsigned long addr, len;
31f3438e 802
a136a8bd 803 switch (id) {
31f3438e 804 case KVM_REG_PPC_HIOR:
31f3438e 805 /* Only allow this to be set to zero */
a136a8bd 806 if (set_reg_val(id, *val))
31f3438e
PM		 807 r = -EINVAL;
808 break;
a136a8bd
PM		 809 case KVM_REG_PPC_DABR:
810 vcpu->arch.dabr = set_reg_val(id, *val);
811 break;
812 case KVM_REG_PPC_DSCR:
813 vcpu->arch.dscr = set_reg_val(id, *val);
814 break;
815 case KVM_REG_PPC_PURR:
816 vcpu->arch.purr = set_reg_val(id, *val);
817 break;
818 case KVM_REG_PPC_SPURR:
819 vcpu->arch.spurr = set_reg_val(id, *val);
820 break;
821 case KVM_REG_PPC_AMR:
822 vcpu->arch.amr = set_reg_val(id, *val);
823 break;
824 case KVM_REG_PPC_UAMOR:
825 vcpu->arch.uamor = set_reg_val(id, *val);
826 break;
827 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRA:
828 i = id - KVM_REG_PPC_MMCR0;
829 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
830 break;
831 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
832 i = id - KVM_REG_PPC_PMC1;
833 vcpu->arch.pmc[i] = set_reg_val(id, *val);
834 break;
a8bd19ef
PM		 835#ifdef CONFIG_VSX
836 case KVM_REG_PPC_FPR0 ... KVM_REG_PPC_FPR31:
837 if (cpu_has_feature(CPU_FTR_VSX)) {
838 /* VSX => FP reg i is stored in arch.vsr[2*i] */
839 long int i = id - KVM_REG_PPC_FPR0;
840 vcpu->arch.vsr[2 * i] = set_reg_val(id, *val);
841 } else {
842 /* let generic code handle it */
843 r = -EINVAL;
844 }
845 break;
846 case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31:
847 if (cpu_has_feature(CPU_FTR_VSX)) {
848 long int i = id - KVM_REG_PPC_VSR0;
849 vcpu->arch.vsr[2 * i] = val->vsxval[0];
850 vcpu->arch.vsr[2 * i + 1] = val->vsxval[1];
851 } else {
852 r = -ENXIO;
853 }
854 break;
855#endif /* CONFIG_VSX */
55b665b0
PM		 856 case KVM_REG_PPC_VPA_ADDR:
857 addr = set_reg_val(id, *val);
858 r = -EINVAL;
859 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
860 vcpu->arch.dtl.next_gpa))
861 break;
862 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
863 break;
864 case KVM_REG_PPC_VPA_SLB:
865 addr = val->vpaval.addr;
866 len = val->vpaval.length;
867 r = -EINVAL;
868 if (addr && !vcpu->arch.vpa.next_gpa)
869 break;
870 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
871 break;
872 case KVM_REG_PPC_VPA_DTL:
873 addr = val->vpaval.addr;
874 len = val->vpaval.length;
875 r = -EINVAL;
9f8c8c78
PM		 876 if (addr && (len < sizeof(struct dtl_entry) ||
877 !vcpu->arch.vpa.next_gpa))
55b665b0
PM		 878 break;
879 len -= len % sizeof(struct dtl_entry);
880 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
881 break;
31f3438e 882 default:
a136a8bd 883 r = -EINVAL;
31f3438e
PM		 884 break;
885 }
886
887 return r;
888}
889
de56a948
PM		 890int kvmppc_core_check_processor_compat(void)
891{
9e368f29 892 if (cpu_has_feature(CPU_FTR_HVMODE))
de56a948
PM		 893 return 0;
894 return -EIO;
895}
896
897struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
898{
899 struct kvm_vcpu *vcpu;
371fefd6
PM		 900 int err = -EINVAL;
901 int core;
902 struct kvmppc_vcore *vcore;
de56a948 903
371fefd6
PM		 904 core = id / threads_per_core;
905 if (core >= KVM_MAX_VCORES)
906 goto out;
907
908 err = -ENOMEM;
6b75e6bf 909 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
de56a948
PM		 910 if (!vcpu)
911 goto out;
912
913 err = kvm_vcpu_init(vcpu, kvm, id);
914 if (err)
915 goto free_vcpu;
916
917 vcpu->arch.shared = &vcpu->arch.shregs;
de56a948
PM		 918 vcpu->arch.mmcr[0] = MMCR0_FC;
919 vcpu->arch.ctrl = CTRL_RUNLATCH;
920 /* default to host PVR, since we can't spoof it */
921 vcpu->arch.pvr = mfspr(SPRN_PVR);
922 kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
2e25aa5f 923 spin_lock_init(&vcpu->arch.vpa_update_lock);
c7b67670
PM		 924 spin_lock_init(&vcpu->arch.tbacct_lock);
925 vcpu->arch.busy_preempt = TB_NIL;
de56a948 926
de56a948
PM		 927 kvmppc_mmu_book3s_hv_init(vcpu);
928
8455d79e 929 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
371fefd6
PM		 930
931 init_waitqueue_head(&vcpu->arch.cpu_run);
932
933 mutex_lock(&kvm->lock);
934 vcore = kvm->arch.vcores[core];
935 if (!vcore) {
936 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
937 if (vcore) {
938 INIT_LIST_HEAD(&vcore->runnable_threads);
939 spin_lock_init(&vcore->lock);
19ccb76a 940 init_waitqueue_head(&vcore->wq);
c7b67670 941 vcore->preempt_tb = TB_NIL;
371fefd6
PM		 942 }
943 kvm->arch.vcores[core] = vcore;
1b400ba0 944 kvm->arch.online_vcores++;
371fefd6
PM		 945 }
946 mutex_unlock(&kvm->lock);
947
948 if (!vcore)
949 goto free_vcpu;
950
951 spin_lock(&vcore->lock);
952 ++vcore->num_threads;
371fefd6
PM		 953 spin_unlock(&vcore->lock);
954 vcpu->arch.vcore = vcore;
955
af8f38b3
AG		 956 vcpu->arch.cpu_type = KVM_CPU_3S_64;
957 kvmppc_sanity_check(vcpu);
958
de56a948
PM		 959 return vcpu;
960
961free_vcpu:
6b75e6bf 962 kmem_cache_free(kvm_vcpu_cache, vcpu);
de56a948
PM		 963out:
964 return ERR_PTR(err);
965}
966
c35635ef
PM		 967static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
968{
969 if (vpa->pinned_addr)
970 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
971 vpa->dirty);
972}
973
de56a948
PM		 974void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
975{
2e25aa5f 976 spin_lock(&vcpu->arch.vpa_update_lock);
c35635ef
PM		 977 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
978 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
979 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2e25aa5f 980 spin_unlock(&vcpu->arch.vpa_update_lock);
de56a948 981 kvm_vcpu_uninit(vcpu);
6b75e6bf 982 kmem_cache_free(kvm_vcpu_cache, vcpu);
de56a948
PM		 983}
984
19ccb76a 985static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
371fefd6 986{
19ccb76a 987 unsigned long dec_nsec, now;
371fefd6 988
19ccb76a
PM		 989 now = get_tb();
990 if (now > vcpu->arch.dec_expires) {
991 /* decrementer has already gone negative */
992 kvmppc_core_queue_dec(vcpu);
7e28e60e 993 kvmppc_core_prepare_to_enter(vcpu);
19ccb76a 994 return;
371fefd6 995 }
19ccb76a
PM		 996 dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
997 / tb_ticks_per_sec;
998 hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
999 HRTIMER_MODE_REL);
1000 vcpu->arch.timer_running = 1;
371fefd6
PM		 1001}
1002
19ccb76a 1003static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
371fefd6 1004{
19ccb76a
PM		 1005 vcpu->arch.ceded = 0;
1006 if (vcpu->arch.timer_running) {
1007 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1008 vcpu->arch.timer_running = 0;
1009 }
371fefd6
PM		 1010}
1011
de56a948
PM		 1012extern int __kvmppc_vcore_entry(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu);
1013
371fefd6
PM		 1014static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1015 struct kvm_vcpu *vcpu)
de56a948 1016{
c7b67670
PM		 1017 u64 now;
1018
371fefd6
PM		 1019 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1020 return;
c7b67670
PM		 1021 spin_lock(&vcpu->arch.tbacct_lock);
1022 now = mftb();
1023 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1024 vcpu->arch.stolen_logged;
1025 vcpu->arch.busy_preempt = now;
1026 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1027 spin_unlock(&vcpu->arch.tbacct_lock);
371fefd6 1028 --vc->n_runnable;
371fefd6
PM		 1029 list_del(&vcpu->arch.run_list);
1030}
1031
f0888f70
PM		 1032static int kvmppc_grab_hwthread(int cpu)
1033{
1034 struct paca_struct *tpaca;
1035 long timeout = 1000;
1036
1037 tpaca = &paca[cpu];
1038
1039 /* Ensure the thread won't go into the kernel if it wakes */
1040 tpaca->kvm_hstate.hwthread_req = 1;
7b444c67 1041 tpaca->kvm_hstate.kvm_vcpu = NULL;
f0888f70
PM		 1042
1043 /*
1044 * If the thread is already executing in the kernel (e.g. handling
1045 * a stray interrupt), wait for it to get back to nap mode.
1046 * The smp_mb() is to ensure that our setting of hwthread_req
1047 * is visible before we look at hwthread_state, so if this
1048 * races with the code at system_reset_pSeries and the thread
1049 * misses our setting of hwthread_req, we are sure to see its
1050 * setting of hwthread_state, and vice versa.
1051 */
1052 smp_mb();
1053 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1054 if (--timeout <= 0) {
1055 pr_err("KVM: couldn't grab cpu %d\n", cpu);
1056 return -EBUSY;
1057 }
1058 udelay(1);
1059 }
1060 return 0;
1061}
1062
1063static void kvmppc_release_hwthread(int cpu)
1064{
1065 struct paca_struct *tpaca;
1066
1067 tpaca = &paca[cpu];
1068 tpaca->kvm_hstate.hwthread_req = 0;
1069 tpaca->kvm_hstate.kvm_vcpu = NULL;
1070}
1071
371fefd6
PM		 1072static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
1073{
1074 int cpu;
1075 struct paca_struct *tpaca;
1076 struct kvmppc_vcore *vc = vcpu->arch.vcore;
1077
19ccb76a
PM		 1078 if (vcpu->arch.timer_running) {
1079 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1080 vcpu->arch.timer_running = 0;
1081 }
371fefd6
PM		 1082 cpu = vc->pcpu + vcpu->arch.ptid;
1083 tpaca = &paca[cpu];
1084 tpaca->kvm_hstate.kvm_vcpu = vcpu;
1085 tpaca->kvm_hstate.kvm_vcore = vc;
19ccb76a
PM		 1086 tpaca->kvm_hstate.napping = 0;
1087 vcpu->cpu = vc->pcpu;
371fefd6 1088 smp_wmb();
251da038 1089#if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
371fefd6 1090 if (vcpu->arch.ptid) {
371fefd6
PM		 1091 xics_wake_cpu(cpu);
1092 ++vc->n_woken;
de56a948 1093 }
371fefd6
PM		 1094#endif
1095}
de56a948 1096
371fefd6
PM		 1097static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc)
1098{
1099 int i;
1100
1101 HMT_low();
1102 i = 0;
1103 while (vc->nap_count < vc->n_woken) {
1104 if (++i >= 1000000) {
1105 pr_err("kvmppc_wait_for_nap timeout %d %d\n",
1106 vc->nap_count, vc->n_woken);
1107 break;
1108 }
1109 cpu_relax();
1110 }
1111 HMT_medium();
1112}
1113
1114/*
1115 * Check that we are on thread 0 and that any other threads in
7b444c67
PM		 1116 * this core are off-line. Then grab the threads so they can't
1117 * enter the kernel.
371fefd6
PM		 1118 */
1119static int on_primary_thread(void)
1120{
1121 int cpu = smp_processor_id();
1122 int thr = cpu_thread_in_core(cpu);
1123
1124 if (thr)
1125 return 0;
1126 while (++thr < threads_per_core)
1127 if (cpu_online(cpu + thr))
1128 return 0;
7b444c67
PM		 1129
1130 /* Grab all hw threads so they can't go into the kernel */
1131 for (thr = 1; thr < threads_per_core; ++thr) {
1132 if (kvmppc_grab_hwthread(cpu + thr)) {
1133 /* Couldn't grab one; let the others go */
1134 do {
1135 kvmppc_release_hwthread(cpu + thr);
1136 } while (--thr > 0);
1137 return 0;
1138 }
1139 }
371fefd6
PM		 1140 return 1;
1141}
1142
1143/*
1144 * Run a set of guest threads on a physical core.
1145 * Called with vc->lock held.
1146 */
913d3ff9 1147static void kvmppc_run_core(struct kvmppc_vcore *vc)
371fefd6 1148{
19ccb76a 1149 struct kvm_vcpu *vcpu, *vcpu0, *vnext;
371fefd6
PM		 1150 long ret;
1151 u64 now;
081f323b 1152 int ptid, i, need_vpa_update;
2c9097e4 1153 int srcu_idx;
913d3ff9 1154 struct kvm_vcpu *vcpus_to_update[threads_per_core];
371fefd6
PM		 1155
1156 /* don't start if any threads have a signal pending */
081f323b
PM		 1157 need_vpa_update = 0;
1158 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
371fefd6 1159 if (signal_pending(vcpu->arch.run_task))
913d3ff9
PM		 1160 return;
1161 if (vcpu->arch.vpa.update_pending ||
1162 vcpu->arch.slb_shadow.update_pending ||
1163 vcpu->arch.dtl.update_pending)
1164 vcpus_to_update[need_vpa_update++] = vcpu;
081f323b
PM		 1165 }
1166
1167 /*
1168 * Initialize *vc, in particular vc->vcore_state, so we can
1169 * drop the vcore lock if necessary.
1170 */
1171 vc->n_woken = 0;
1172 vc->nap_count = 0;
1173 vc->entry_exit_count = 0;
2f12f034 1174 vc->vcore_state = VCORE_STARTING;
081f323b
PM		 1175 vc->in_guest = 0;
1176 vc->napping_threads = 0;
1177
1178 /*
1179 * Updating any of the vpas requires calling kvmppc_pin_guest_page,
1180 * which can't be called with any spinlocks held.
1181 */
1182 if (need_vpa_update) {
1183 spin_unlock(&vc->lock);
913d3ff9
PM		 1184 for (i = 0; i < need_vpa_update; ++i)
1185 kvmppc_update_vpas(vcpus_to_update[i]);
081f323b
PM		 1186 spin_lock(&vc->lock);
1187 }
de56a948 1188
19ccb76a
PM		 1189 /*
1190 * Assign physical thread IDs, first to non-ceded vcpus
1191 * and then to ceded ones.
1192 */
1193 ptid = 0;
1194 vcpu0 = NULL;
1195 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1196 if (!vcpu->arch.ceded) {
1197 if (!ptid)
1198 vcpu0 = vcpu;
1199 vcpu->arch.ptid = ptid++;
1200 }
1201 }
c7b67670
PM		 1202 if (!vcpu0)
1203 goto out; /* nothing to run; should never happen */
19ccb76a
PM		 1204 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1205 if (vcpu->arch.ceded)
1206 vcpu->arch.ptid = ptid++;
1207
7b444c67
PM		 1208 /*
1209 * Make sure we are running on thread 0, and that
1210 * secondary threads are offline.
1211 */
1212 if (threads_per_core > 1 && !on_primary_thread()) {
1213 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1214 vcpu->arch.ret = -EBUSY;
1215 goto out;
1216 }
1217
371fefd6 1218 vc->pcpu = smp_processor_id();
2e25aa5f 1219 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
371fefd6 1220 kvmppc_start_thread(vcpu);
0456ec4f 1221 kvmppc_create_dtl_entry(vcpu, vc);
2e25aa5f 1222 }
371fefd6 1223
2f12f034 1224 vc->vcore_state = VCORE_RUNNING;
19ccb76a 1225 preempt_disable();
371fefd6 1226 spin_unlock(&vc->lock);
de56a948 1227
371fefd6 1228 kvm_guest_enter();
2c9097e4
PM		 1229
1230 srcu_idx = srcu_read_lock(&vcpu0->kvm->srcu);
1231
19ccb76a 1232 __kvmppc_vcore_entry(NULL, vcpu0);
de56a948 1233
371fefd6 1234 spin_lock(&vc->lock);
19ccb76a
PM		 1235 /* disable sending of IPIs on virtual external irqs */
1236 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1237 vcpu->cpu = -1;
1238 /* wait for secondary threads to finish writing their state to memory */
371fefd6
PM		 1239 if (vc->nap_count < vc->n_woken)
1240 kvmppc_wait_for_nap(vc);
2f12f034
PM		 1241 for (i = 0; i < threads_per_core; ++i)
1242 kvmppc_release_hwthread(vc->pcpu + i);
371fefd6 1243 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
19ccb76a 1244 vc->vcore_state = VCORE_EXITING;
371fefd6
PM		 1245 spin_unlock(&vc->lock);
1246
2c9097e4
PM		 1247 srcu_read_unlock(&vcpu0->kvm->srcu, srcu_idx);
1248
371fefd6
PM		 1249 /* make sure updates to secondary vcpu structs are visible now */
1250 smp_mb();
de56a948
PM		 1251 kvm_guest_exit();
1252
1253 preempt_enable();
1254 kvm_resched(vcpu);
1255
913d3ff9 1256 spin_lock(&vc->lock);
de56a948 1257 now = get_tb();
371fefd6
PM		 1258 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1259 /* cancel pending dec exception if dec is positive */
1260 if (now < vcpu->arch.dec_expires &&
1261 kvmppc_core_pending_dec(vcpu))
1262 kvmppc_core_dequeue_dec(vcpu);
19ccb76a
PM		 1263
1264 ret = RESUME_GUEST;
1265 if (vcpu->arch.trap)
1266 ret = kvmppc_handle_exit(vcpu->arch.kvm_run, vcpu,
1267 vcpu->arch.run_task);
1268
371fefd6
PM		 1269 vcpu->arch.ret = ret;
1270 vcpu->arch.trap = 0;
19ccb76a
PM		 1271
1272 if (vcpu->arch.ceded) {
1273 if (ret != RESUME_GUEST)
1274 kvmppc_end_cede(vcpu);
1275 else
1276 kvmppc_set_timer(vcpu);
1277 }
371fefd6 1278 }
de56a948
PM		 1279
1280 out:
19ccb76a 1281 vc->vcore_state = VCORE_INACTIVE;
371fefd6
PM		 1282 list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1283 arch.run_list) {
1284 if (vcpu->arch.ret != RESUME_GUEST) {
1285 kvmppc_remove_runnable(vc, vcpu);
1286 wake_up(&vcpu->arch.cpu_run);
1287 }
1288 }
371fefd6
PM		 1289}
1290
19ccb76a
PM		 1291/*
1292 * Wait for some other vcpu thread to execute us, and
1293 * wake us up when we need to handle something in the host.
1294 */
1295static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
371fefd6 1296{
371fefd6
PM		 1297 DEFINE_WAIT(wait);
1298
19ccb76a
PM		 1299 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
1300 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
1301 schedule();
1302 finish_wait(&vcpu->arch.cpu_run, &wait);
1303}
1304
1305/*
1306 * All the vcpus in this vcore are idle, so wait for a decrementer
1307 * or external interrupt to one of the vcpus. vc->lock is held.
1308 */
1309static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
1310{
1311 DEFINE_WAIT(wait);
19ccb76a
PM		 1312
1313 prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
1314 vc->vcore_state = VCORE_SLEEPING;
1315 spin_unlock(&vc->lock);
913d3ff9 1316 schedule();
19ccb76a
PM		 1317 finish_wait(&vc->wq, &wait);
1318 spin_lock(&vc->lock);
1319 vc->vcore_state = VCORE_INACTIVE;
1320}
371fefd6 1321
19ccb76a
PM		 1322static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1323{
1324 int n_ceded;
19ccb76a
PM		 1325 struct kvmppc_vcore *vc;
1326 struct kvm_vcpu *v, *vn;
9e368f29 1327
371fefd6
PM		 1328 kvm_run->exit_reason = 0;
1329 vcpu->arch.ret = RESUME_GUEST;
1330 vcpu->arch.trap = 0;
2f12f034 1331 kvmppc_update_vpas(vcpu);
371fefd6 1332
371fefd6
PM		 1333 /*
1334 * Synchronize with other threads in this virtual core
1335 */
1336 vc = vcpu->arch.vcore;
1337 spin_lock(&vc->lock);
19ccb76a 1338 vcpu->arch.ceded = 0;
371fefd6
PM		 1339 vcpu->arch.run_task = current;
1340 vcpu->arch.kvm_run = kvm_run;
c7b67670 1341 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
19ccb76a 1342 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
c7b67670 1343 vcpu->arch.busy_preempt = TB_NIL;
371fefd6
PM		 1344 list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
1345 ++vc->n_runnable;
1346
19ccb76a
PM		 1347 /*
1348 * This happens the first time this is called for a vcpu.
1349 * If the vcore is already running, we may be able to start
1350 * this thread straight away and have it join in.
1351 */
8455d79e 1352 if (!signal_pending(current)) {
19ccb76a
PM		 1353 if (vc->vcore_state == VCORE_RUNNING &&
1354 VCORE_EXIT_COUNT(vc) == 0) {
1355 vcpu->arch.ptid = vc->n_runnable - 1;
2f12f034 1356 kvmppc_create_dtl_entry(vcpu, vc);
19ccb76a 1357 kvmppc_start_thread(vcpu);
8455d79e
PM		 1358 } else if (vc->vcore_state == VCORE_SLEEPING) {
1359 wake_up(&vc->wq);
371fefd6
PM		 1360 }
1361
8455d79e 1362 }
371fefd6 1363
19ccb76a
PM		 1364 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1365 !signal_pending(current)) {
8455d79e 1366 if (vc->vcore_state != VCORE_INACTIVE) {
19ccb76a
PM		 1367 spin_unlock(&vc->lock);
1368 kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
1369 spin_lock(&vc->lock);
1370 continue;
1371 }
19ccb76a
PM		 1372 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
1373 arch.run_list) {
7e28e60e 1374 kvmppc_core_prepare_to_enter(v);
19ccb76a
PM		 1375 if (signal_pending(v->arch.run_task)) {
1376 kvmppc_remove_runnable(vc, v);
1377 v->stat.signal_exits++;
1378 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
1379 v->arch.ret = -EINTR;
1380 wake_up(&v->arch.cpu_run);
1381 }
1382 }
8455d79e
PM		 1383 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1384 break;
1385 vc->runner = vcpu;
1386 n_ceded = 0;
4619ac88 1387 list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
8455d79e
PM		 1388 if (!v->arch.pending_exceptions)
1389 n_ceded += v->arch.ceded;
4619ac88
PM		 1390 else
1391 v->arch.ceded = 0;
1392 }
8455d79e
PM		 1393 if (n_ceded == vc->n_runnable)
1394 kvmppc_vcore_blocked(vc);
1395 else
1396 kvmppc_run_core(vc);
0456ec4f 1397 vc->runner = NULL;
19ccb76a 1398 }
371fefd6 1399
8455d79e
PM		 1400 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1401 (vc->vcore_state == VCORE_RUNNING ||
1402 vc->vcore_state == VCORE_EXITING)) {
1403 spin_unlock(&vc->lock);
1404 kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
1405 spin_lock(&vc->lock);
1406 }
1407
1408 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
1409 kvmppc_remove_runnable(vc, vcpu);
1410 vcpu->stat.signal_exits++;
1411 kvm_run->exit_reason = KVM_EXIT_INTR;
1412 vcpu->arch.ret = -EINTR;
1413 }
1414
1415 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
1416 /* Wake up some vcpu to run the core */
1417 v = list_first_entry(&vc->runnable_threads,
1418 struct kvm_vcpu, arch.run_list);
1419 wake_up(&v->arch.cpu_run);
371fefd6
PM		 1420 }
1421
371fefd6 1422 spin_unlock(&vc->lock);
371fefd6 1423 return vcpu->arch.ret;
de56a948
PM		 1424}
1425
a8606e20
PM		 1426int kvmppc_vcpu_run(struct kvm_run *run, struct kvm_vcpu *vcpu)
1427{
1428 int r;
913d3ff9 1429 int srcu_idx;
a8606e20 1430
af8f38b3
AG		 1431 if (!vcpu->arch.sane) {
1432 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1433 return -EINVAL;
1434 }
1435
25051b5a
SW		 1436 kvmppc_core_prepare_to_enter(vcpu);
1437
19ccb76a
PM		 1438 /* No need to go into the guest when all we'll do is come back out */
1439 if (signal_pending(current)) {
1440 run->exit_reason = KVM_EXIT_INTR;
1441 return -EINTR;
1442 }
1443
32fad281
PM		 1444 atomic_inc(&vcpu->kvm->arch.vcpus_running);
1445 /* Order vcpus_running vs. rma_setup_done, see kvmppc_alloc_reset_hpt */
1446 smp_mb();
1447
1448 /* On the first time here, set up HTAB and VRMA or RMA */
c77162de 1449 if (!vcpu->kvm->arch.rma_setup_done) {
32fad281 1450 r = kvmppc_hv_setup_htab_rma(vcpu);
c77162de 1451 if (r)
32fad281 1452 goto out;
c77162de 1453 }
19ccb76a
PM		 1454
1455 flush_fp_to_thread(current);
1456 flush_altivec_to_thread(current);
1457 flush_vsx_to_thread(current);
1458 vcpu->arch.wqp = &vcpu->arch.vcore->wq;
342d3db7 1459 vcpu->arch.pgdir = current->mm->pgd;
c7b67670 1460 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
19ccb76a 1461
a8606e20
PM		 1462 do {
1463 r = kvmppc_run_vcpu(run, vcpu);
1464
1465 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
1466 !(vcpu->arch.shregs.msr & MSR_PR)) {
1467 r = kvmppc_pseries_do_hcall(vcpu);
7e28e60e 1468 kvmppc_core_prepare_to_enter(vcpu);
913d3ff9
PM		 1469 } else if (r == RESUME_PAGE_FAULT) {
1470 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1471 r = kvmppc_book3s_hv_page_fault(run, vcpu,
1472 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1473 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
a8606e20
PM		 1474 }
1475 } while (r == RESUME_GUEST);
32fad281
PM		 1476
1477 out:
c7b67670 1478 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
32fad281 1479 atomic_dec(&vcpu->kvm->arch.vcpus_running);
a8606e20
PM		 1480 return r;
1481}
1482
54738c09 1483
aa04b4cc 1484/* Work out RMLS (real mode limit selector) field value for a given RMA size.
9e368f29 1485 Assumes POWER7 or PPC970. */
aa04b4cc
PM		 1486static inline int lpcr_rmls(unsigned long rma_size)
1487{
1488 switch (rma_size) {
1489 case 32ul << 20: /* 32 MB */
9e368f29
PM		 1490 if (cpu_has_feature(CPU_FTR_ARCH_206))
1491 return 8; /* only supported on POWER7 */
1492 return -1;
aa04b4cc
PM		 1493 case 64ul << 20: /* 64 MB */
1494 return 3;
1495 case 128ul << 20: /* 128 MB */
1496 return 7;
1497 case 256ul << 20: /* 256 MB */
1498 return 4;
1499 case 1ul << 30: /* 1 GB */
1500 return 2;
1501 case 16ul << 30: /* 16 GB */
1502 return 1;
1503 case 256ul << 30: /* 256 GB */
1504 return 0;
1505 default:
1506 return -1;
1507 }
1508}
1509
1510static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1511{
b4e70611 1512 struct kvmppc_linear_info *ri = vma->vm_file->private_data;
aa04b4cc
PM		 1513 struct page *page;
1514
1515 if (vmf->pgoff >= ri->npages)
1516 return VM_FAULT_SIGBUS;
1517
1518 page = pfn_to_page(ri->base_pfn + vmf->pgoff);
1519 get_page(page);
1520 vmf->page = page;
1521 return 0;
1522}
1523
1524static const struct vm_operations_struct kvm_rma_vm_ops = {
1525 .fault = kvm_rma_fault,
1526};
1527
1528static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma)
1529{
314e51b9 1530 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
aa04b4cc
PM		 1531 vma->vm_ops = &kvm_rma_vm_ops;
1532 return 0;
1533}
1534
1535static int kvm_rma_release(struct inode *inode, struct file *filp)
1536{
b4e70611 1537 struct kvmppc_linear_info *ri = filp->private_data;
aa04b4cc
PM		 1538
1539 kvm_release_rma(ri);
1540 return 0;
1541}
1542
1543static struct file_operations kvm_rma_fops = {
1544 .mmap = kvm_rma_mmap,
1545 .release = kvm_rma_release,
1546};
1547
1548long kvm_vm_ioctl_allocate_rma(struct kvm *kvm, struct kvm_allocate_rma *ret)
1549{
b4e70611 1550 struct kvmppc_linear_info *ri;
aa04b4cc
PM		 1551 long fd;
1552
1553 ri = kvm_alloc_rma();
1554 if (!ri)
1555 return -ENOMEM;
1556
1557 fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR);
1558 if (fd < 0)
1559 kvm_release_rma(ri);
1560
1561 ret->rma_size = ri->npages << PAGE_SHIFT;
1562 return fd;
1563}
1564
5b74716e
BH		 1565static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
1566 int linux_psize)
1567{
1568 struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
1569
1570 if (!def->shift)
1571 return;
1572 (*sps)->page_shift = def->shift;
1573 (*sps)->slb_enc = def->sllp;
1574 (*sps)->enc[0].page_shift = def->shift;
1575 (*sps)->enc[0].pte_enc = def->penc;
1576 (*sps)++;
1577}
1578
1579int kvm_vm_ioctl_get_smmu_info(struct kvm *kvm, struct kvm_ppc_smmu_info *info)
1580{
1581 struct kvm_ppc_one_seg_page_size *sps;
1582
1583 info->flags = KVM_PPC_PAGE_SIZES_REAL;
1584 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1585 info->flags |= KVM_PPC_1T_SEGMENTS;
1586 info->slb_size = mmu_slb_size;
1587
1588 /* We only support these sizes for now, and no muti-size segments */
1589 sps = &info->sps[0];
1590 kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
1591 kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
1592 kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
1593
1594 return 0;
1595}
1596
82ed3616
PM		 1597/*
1598 * Get (and clear) the dirty memory log for a memory slot.
1599 */
1600int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1601{
1602 struct kvm_memory_slot *memslot;
1603 int r;
1604 unsigned long n;
1605
1606 mutex_lock(&kvm->slots_lock);
1607
1608 r = -EINVAL;
bbacc0c1 1609 if (log->slot >= KVM_USER_MEM_SLOTS)
82ed3616
PM		 1610 goto out;
1611
1612 memslot = id_to_memslot(kvm->memslots, log->slot);
1613 r = -ENOENT;
1614 if (!memslot->dirty_bitmap)
1615 goto out;
1616
1617 n = kvm_dirty_bitmap_bytes(memslot);
1618 memset(memslot->dirty_bitmap, 0, n);
1619
dfe49dbd 1620 r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
82ed3616
PM		 1621 if (r)
1622 goto out;
1623
1624 r = -EFAULT;
1625 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1626 goto out;
1627
1628 r = 0;
1629out:
1630 mutex_unlock(&kvm->slots_lock);
1631 return r;
1632}
1633
a66b48c3 1634static void unpin_slot(struct kvm_memory_slot *memslot)
de56a948 1635{
a66b48c3
PM		 1636 unsigned long *physp;
1637 unsigned long j, npages, pfn;
1638 struct page *page;
aa04b4cc 1639
a66b48c3
PM		 1640 physp = memslot->arch.slot_phys;
1641 npages = memslot->npages;
1642 if (!physp)
1643 return;
1644 for (j = 0; j < npages; j++) {
1645 if (!(physp[j] & KVMPPC_GOT_PAGE))
1646 continue;
1647 pfn = physp[j] >> PAGE_SHIFT;
1648 page = pfn_to_page(pfn);
1649 SetPageDirty(page);
1650 put_page(page);
1651 }
1652}
1653
1654void kvmppc_core_free_memslot(struct kvm_memory_slot *free,
1655 struct kvm_memory_slot *dont)
1656{
1657 if (!dont || free->arch.rmap != dont->arch.rmap) {
1658 vfree(free->arch.rmap);
1659 free->arch.rmap = NULL;
b2b2f165 1660 }
a66b48c3
PM		 1661 if (!dont || free->arch.slot_phys != dont->arch.slot_phys) {
1662 unpin_slot(free);
1663 vfree(free->arch.slot_phys);
1664 free->arch.slot_phys = NULL;
1665 }
1666}
1667
1668int kvmppc_core_create_memslot(struct kvm_memory_slot *slot,
1669 unsigned long npages)
1670{
1671 slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
1672 if (!slot->arch.rmap)
1673 return -ENOMEM;
1674 slot->arch.slot_phys = NULL;
aa04b4cc 1675
c77162de
PM		 1676 return 0;
1677}
aa04b4cc 1678
a66b48c3
PM		 1679int kvmppc_core_prepare_memory_region(struct kvm *kvm,
1680 struct kvm_memory_slot *memslot,
1681 struct kvm_userspace_memory_region *mem)
c77162de 1682{
a66b48c3 1683 unsigned long *phys;
c77162de 1684
a66b48c3
PM		 1685 /* Allocate a slot_phys array if needed */
1686 phys = memslot->arch.slot_phys;
1687 if (!kvm->arch.using_mmu_notifiers && !phys && memslot->npages) {
1688 phys = vzalloc(memslot->npages * sizeof(unsigned long));
1689 if (!phys)
1690 return -ENOMEM;
1691 memslot->arch.slot_phys = phys;
aa04b4cc 1692 }
a66b48c3
PM		 1693
1694 return 0;
c77162de
PM		 1695}
1696
1697void kvmppc_core_commit_memory_region(struct kvm *kvm,
dfe49dbd 1698 struct kvm_userspace_memory_region *mem,
8482644a 1699 const struct kvm_memory_slot *old)
c77162de 1700{
dfe49dbd
PM		 1701 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
1702 struct kvm_memory_slot *memslot;
1703
8482644a 1704 if (npages && old->npages) {
dfe49dbd
PM		 1705 /*
1706 * If modifying a memslot, reset all the rmap dirty bits.
1707 * If this is a new memslot, we don't need to do anything
1708 * since the rmap array starts out as all zeroes,
1709 * i.e. no pages are dirty.
1710 */
1711 memslot = id_to_memslot(kvm->memslots, mem->slot);
1712 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
1713 }
c77162de
PM		 1714}
1715
32fad281 1716static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
c77162de
PM		 1717{
1718 int err = 0;
1719 struct kvm *kvm = vcpu->kvm;
b4e70611 1720 struct kvmppc_linear_info *ri = NULL;
c77162de
PM		 1721 unsigned long hva;
1722 struct kvm_memory_slot *memslot;
1723 struct vm_area_struct *vma;
da9d1d7f 1724 unsigned long lpcr, senc;
c77162de
PM		 1725 unsigned long psize, porder;
1726 unsigned long rma_size;
1727 unsigned long rmls;
1728 unsigned long *physp;
da9d1d7f 1729 unsigned long i, npages;
2c9097e4 1730 int srcu_idx;
c77162de
PM		 1731
1732 mutex_lock(&kvm->lock);
1733 if (kvm->arch.rma_setup_done)
1734 goto out; /* another vcpu beat us to it */
aa04b4cc 1735
32fad281
PM		 1736 /* Allocate hashed page table (if not done already) and reset it */
1737 if (!kvm->arch.hpt_virt) {
1738 err = kvmppc_alloc_hpt(kvm, NULL);
1739 if (err) {
1740 pr_err("KVM: Couldn't alloc HPT\n");
1741 goto out;
1742 }
1743 }
1744
c77162de 1745 /* Look up the memslot for guest physical address 0 */
2c9097e4 1746 srcu_idx = srcu_read_lock(&kvm->srcu);
c77162de 1747 memslot = gfn_to_memslot(kvm, 0);
aa04b4cc 1748
c77162de
PM		 1749 /* We must have some memory at 0 by now */
1750 err = -EINVAL;
1751 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
2c9097e4 1752 goto out_srcu;
c77162de
PM		 1753
1754 /* Look up the VMA for the start of this memory slot */
1755 hva = memslot->userspace_addr;
1756 down_read(&current->mm->mmap_sem);
1757 vma = find_vma(current->mm, hva);
1758 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
1759 goto up_out;
1760
1761 psize = vma_kernel_pagesize(vma);
da9d1d7f 1762 porder = __ilog2(psize);
c77162de
PM		 1763
1764 /* Is this one of our preallocated RMAs? */
1765 if (vma->vm_file && vma->vm_file->f_op == &kvm_rma_fops &&
1766 hva == vma->vm_start)
1767 ri = vma->vm_file->private_data;
1768
1769 up_read(&current->mm->mmap_sem);
1770
1771 if (!ri) {
1772 /* On POWER7, use VRMA; on PPC970, give up */
1773 err = -EPERM;
1774 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1775 pr_err("KVM: CPU requires an RMO\n");
2c9097e4 1776 goto out_srcu;
c77162de
PM		 1777 }
1778
da9d1d7f
PM		 1779 /* We can handle 4k, 64k or 16M pages in the VRMA */
1780 err = -EINVAL;
1781 if (!(psize == 0x1000 || psize == 0x10000 ||
1782 psize == 0x1000000))
2c9097e4 1783 goto out_srcu;
da9d1d7f 1784
c77162de 1785 /* Update VRMASD field in the LPCR */
da9d1d7f 1786 senc = slb_pgsize_encoding(psize);
697d3899
PM		 1787 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
1788 (VRMA_VSID << SLB_VSID_SHIFT_1T);
da9d1d7f
PM		 1789 lpcr = kvm->arch.lpcr & ~LPCR_VRMASD;
1790 lpcr |= senc << (LPCR_VRMASD_SH - 4);
c77162de
PM		 1791 kvm->arch.lpcr = lpcr;
1792
1793 /* Create HPTEs in the hash page table for the VRMA */
da9d1d7f 1794 kvmppc_map_vrma(vcpu, memslot, porder);
c77162de
PM		 1795
1796 } else {
1797 /* Set up to use an RMO region */
1798 rma_size = ri->npages;
1799 if (rma_size > memslot->npages)
1800 rma_size = memslot->npages;
1801 rma_size <<= PAGE_SHIFT;
aa04b4cc 1802 rmls = lpcr_rmls(rma_size);
c77162de 1803 err = -EINVAL;
aa04b4cc 1804 if (rmls < 0) {
c77162de 1805 pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size);
2c9097e4 1806 goto out_srcu;
aa04b4cc
PM		 1807 }
1808 atomic_inc(&ri->use_count);
1809 kvm->arch.rma = ri;
9e368f29
PM		 1810
1811 /* Update LPCR and RMOR */
1812 lpcr = kvm->arch.lpcr;
1813 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1814 /* PPC970; insert RMLS value (split field) in HID4 */
1815 lpcr &= ~((1ul << HID4_RMLS0_SH) |
1816 (3ul << HID4_RMLS2_SH));
1817 lpcr |= ((rmls >> 2) << HID4_RMLS0_SH) |
1818 ((rmls & 3) << HID4_RMLS2_SH);
1819 /* RMOR is also in HID4 */
1820 lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff)
1821 << HID4_RMOR_SH;
1822 } else {
1823 /* POWER7 */
1824 lpcr &= ~(LPCR_VPM0 | LPCR_VRMA_L);
1825 lpcr |= rmls << LPCR_RMLS_SH;
1826 kvm->arch.rmor = kvm->arch.rma->base_pfn << PAGE_SHIFT;
1827 }
aa04b4cc 1828 kvm->arch.lpcr = lpcr;
c77162de 1829 pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n",
aa04b4cc 1830 ri->base_pfn << PAGE_SHIFT, rma_size, lpcr);
aa04b4cc 1831
c77162de 1832 /* Initialize phys addrs of pages in RMO */
da9d1d7f
PM		 1833 npages = ri->npages;
1834 porder = __ilog2(npages);
a66b48c3
PM		 1835 physp = memslot->arch.slot_phys;
1836 if (physp) {
1837 if (npages > memslot->npages)
1838 npages = memslot->npages;
1839 spin_lock(&kvm->arch.slot_phys_lock);
1840 for (i = 0; i < npages; ++i)
1841 physp[i] = ((ri->base_pfn + i) << PAGE_SHIFT) +
1842 porder;
1843 spin_unlock(&kvm->arch.slot_phys_lock);
1844 }
aa04b4cc
PM		 1845 }
1846
c77162de
PM		 1847 /* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */
1848 smp_wmb();
1849 kvm->arch.rma_setup_done = 1;
1850 err = 0;
2c9097e4
PM		 1851 out_srcu:
1852 srcu_read_unlock(&kvm->srcu, srcu_idx);
c77162de
PM		 1853 out:
1854 mutex_unlock(&kvm->lock);
1855 return err;
b2b2f165 1856
c77162de
PM		 1857 up_out:
1858 up_read(&current->mm->mmap_sem);
1859 goto out;
de56a948
PM		 1860}
1861
1862int kvmppc_core_init_vm(struct kvm *kvm)
1863{
32fad281 1864 unsigned long lpcr, lpid;
de56a948 1865
32fad281
PM		 1866 /* Allocate the guest's logical partition ID */
1867
1868 lpid = kvmppc_alloc_lpid();
1869 if (lpid < 0)
1870 return -ENOMEM;
1871 kvm->arch.lpid = lpid;
de56a948 1872
1b400ba0
PM		 1873 /*
1874 * Since we don't flush the TLB when tearing down a VM,
1875 * and this lpid might have previously been used,
1876 * make sure we flush on each core before running the new VM.
1877 */
1878 cpumask_setall(&kvm->arch.need_tlb_flush);
1879
54738c09 1880 INIT_LIST_HEAD(&kvm->arch.spapr_tce_tables);
8e591cb7 1881 INIT_LIST_HEAD(&kvm->arch.rtas_tokens);
aa04b4cc 1882
aa04b4cc 1883 kvm->arch.rma = NULL;
aa04b4cc 1884
9e368f29 1885 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
aa04b4cc 1886
9e368f29
PM		 1887 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1888 /* PPC970; HID4 is effectively the LPCR */
9e368f29
PM		 1889 kvm->arch.host_lpid = 0;
1890 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4);
1891 lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH));
1892 lpcr |= ((lpid >> 4) << HID4_LPID1_SH) |
1893 ((lpid & 0xf) << HID4_LPID5_SH);
1894 } else {
1895 /* POWER7; init LPCR for virtual RMA mode */
1896 kvm->arch.host_lpid = mfspr(SPRN_LPID);
1897 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
1898 lpcr &= LPCR_PECE | LPCR_LPES;
1899 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
697d3899
PM		 1900 LPCR_VPM0 | LPCR_VPM1;
1901 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
1902 (VRMA_VSID << SLB_VSID_SHIFT_1T);
9e368f29
PM		 1903 }
1904 kvm->arch.lpcr = lpcr;
aa04b4cc 1905
342d3db7 1906 kvm->arch.using_mmu_notifiers = !!cpu_has_feature(CPU_FTR_ARCH_206);
c77162de 1907 spin_lock_init(&kvm->arch.slot_phys_lock);
512691d4
PM		 1908
1909 /*
1910 * Don't allow secondary CPU threads to come online
1911 * while any KVM VMs exist.
1912 */
1913 inhibit_secondary_onlining();
1914
54738c09 1915 return 0;
de56a948
PM		 1916}
1917
1918void kvmppc_core_destroy_vm(struct kvm *kvm)
1919{
512691d4
PM		 1920 uninhibit_secondary_onlining();
1921
aa04b4cc
PM		 1922 if (kvm->arch.rma) {
1923 kvm_release_rma(kvm->arch.rma);
1924 kvm->arch.rma = NULL;
1925 }
1926
8e591cb7
ME		 1927 kvmppc_rtas_tokens_free(kvm);
1928
de56a948 1929 kvmppc_free_hpt(kvm);
54738c09 1930 WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
de56a948
PM		 1931}
1932
1933/* These are stubs for now */
1934void kvmppc_mmu_pte_pflush(struct kvm_vcpu *vcpu, ulong pa_start, ulong pa_end)
1935{
1936}
1937
1938/* We don't need to emulate any privileged instructions or dcbz */
1939int kvmppc_core_emulate_op(struct kvm_run *run, struct kvm_vcpu *vcpu,
1940 unsigned int inst, int *advance)
1941{
1942 return EMULATE_FAIL;
1943}
1944
54771e62 1945int kvmppc_core_emulate_mtspr(struct kvm_vcpu *vcpu, int sprn, ulong spr_val)
de56a948
PM		 1946{
1947 return EMULATE_FAIL;
1948}
1949
54771e62 1950int kvmppc_core_emulate_mfspr(struct kvm_vcpu *vcpu, int sprn, ulong *spr_val)
de56a948
PM		 1951{
1952 return EMULATE_FAIL;
1953}
1954
1955static int kvmppc_book3s_hv_init(void)
1956{
1957 int r;
1958
1959 r = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1960
1961 if (r)
1962 return r;
1963
1964 r = kvmppc_mmu_hv_init();
1965
1966 return r;
1967}
1968
1969static void kvmppc_book3s_hv_exit(void)
1970{
1971 kvm_exit();
1972}
1973
1974module_init(kvmppc_book3s_hv_init);
1975module_exit(kvmppc_book3s_hv_exit);
kernel source for telekom puls (alcatel/mediatek)