projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: PPC: Book3S: Controls for in-kernel sPAPR hypercall handling
[deliverable/linux.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
699a0ea0
PM		 70static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
71
19ccb76a 72static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
32fad281 73static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
19ccb76a 74
3a167bea 75static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
54695c30
BH		 76{
77 int me;
78 int cpu = vcpu->cpu;
79 wait_queue_head_t *wqp;
80
81 wqp = kvm_arch_vcpu_wq(vcpu);
82 if (waitqueue_active(wqp)) {
83 wake_up_interruptible(wqp);
84 ++vcpu->stat.halt_wakeup;
85 }
86
87 me = get_cpu();
88
89 /* CPU points to the first thread of the core */
90 if (cpu != me && cpu >= 0 && cpu < nr_cpu_ids) {
7505258c 91#ifdef CONFIG_PPC_ICP_NATIVE
54695c30
BH		 92 int real_cpu = cpu + vcpu->arch.ptid;
93 if (paca[real_cpu].kvm_hstate.xics_phys)
94 xics_wake_cpu(real_cpu);
48eaef05
AS		 95 else
96#endif
97 if (cpu_online(cpu))
54695c30
BH		 98 smp_send_reschedule(cpu);
99 }
100 put_cpu();
101}
102
c7b67670
PM		 103/*
104 * We use the vcpu_load/put functions to measure stolen time.
105 * Stolen time is counted as time when either the vcpu is able to
106 * run as part of a virtual core, but the task running the vcore
107 * is preempted or sleeping, or when the vcpu needs something done
108 * in the kernel by the task running the vcpu, but that task is
109 * preempted or sleeping. Those two things have to be counted
110 * separately, since one of the vcpu tasks will take on the job
111 * of running the core, and the other vcpu tasks in the vcore will
112 * sleep waiting for it to do that, but that sleep shouldn't count
113 * as stolen time.
114 *
115 * Hence we accumulate stolen time when the vcpu can run as part of
116 * a vcore using vc->stolen_tb, and the stolen time when the vcpu
117 * needs its task to do other things in the kernel (for example,
118 * service a page fault) in busy_stolen. We don't accumulate
119 * stolen time for a vcore when it is inactive, or for a vcpu
120 * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of
121 * a misnomer; it means that the vcpu task is not executing in
122 * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
123 * the kernel. We don't have any way of dividing up that time
124 * between time that the vcpu is genuinely stopped, time that
125 * the task is actively working on behalf of the vcpu, and time
126 * that the task is preempted, so we don't count any of it as
127 * stolen.
128 *
129 * Updates to busy_stolen are protected by arch.tbacct_lock;
130 * updates to vc->stolen_tb are protected by the arch.tbacct_lock
131 * of the vcpu that has taken responsibility for running the vcore
132 * (i.e. vc->runner). The stolen times are measured in units of
133 * timebase ticks. (Note that the != TB_NIL checks below are
134 * purely defensive; they should never fail.)
135 */
136
3a167bea 137static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
de56a948 138{
0456ec4f 139 struct kvmppc_vcore *vc = vcpu->arch.vcore;
bf3d32e1 140 unsigned long flags;
0456ec4f 141
bf3d32e1 142 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
c7b67670
PM		 143 if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE &&
144 vc->preempt_tb != TB_NIL) {
0456ec4f 145 vc->stolen_tb += mftb() - vc->preempt_tb;
c7b67670
PM		 146 vc->preempt_tb = TB_NIL;
147 }
148 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
149 vcpu->arch.busy_preempt != TB_NIL) {
150 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
151 vcpu->arch.busy_preempt = TB_NIL;
152 }
bf3d32e1 153 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
de56a948
PM		 154}
155
3a167bea 156static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
de56a948 157{
0456ec4f 158 struct kvmppc_vcore *vc = vcpu->arch.vcore;
bf3d32e1 159 unsigned long flags;
0456ec4f 160
bf3d32e1 161 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
0456ec4f
PM		 162 if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE)
163 vc->preempt_tb = mftb();
c7b67670
PM		 164 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
165 vcpu->arch.busy_preempt = mftb();
bf3d32e1 166 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
de56a948
PM		 167}
168
3a167bea 169static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
de56a948
PM		 170{
171 vcpu->arch.shregs.msr = msr;
19ccb76a 172 kvmppc_end_cede(vcpu);
de56a948
PM		 173}
174
3a167bea 175void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
de56a948
PM		 176{
177 vcpu->arch.pvr = pvr;
178}
179
388cc6e1
PM		 180int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
181{
182 unsigned long pcr = 0;
183 struct kvmppc_vcore *vc = vcpu->arch.vcore;
184
185 if (arch_compat) {
186 if (!cpu_has_feature(CPU_FTR_ARCH_206))
187 return -EINVAL; /* 970 has no compat mode support */
188
189 switch (arch_compat) {
190 case PVR_ARCH_205:
5557ae0e
PM		 191 /*
192 * If an arch bit is set in PCR, all the defined
193 * higher-order arch bits also have to be set.
194 */
195 pcr = PCR_ARCH_206 | PCR_ARCH_205;
388cc6e1
PM		 196 break;
197 case PVR_ARCH_206:
198 case PVR_ARCH_206p:
5557ae0e
PM		 199 pcr = PCR_ARCH_206;
200 break;
201 case PVR_ARCH_207:
388cc6e1
PM		 202 break;
203 default:
204 return -EINVAL;
205 }
5557ae0e
PM		 206
207 if (!cpu_has_feature(CPU_FTR_ARCH_207S)) {
208 /* POWER7 can't emulate POWER8 */
209 if (!(pcr & PCR_ARCH_206))
210 return -EINVAL;
211 pcr &= ~PCR_ARCH_206;
212 }
388cc6e1
PM		 213 }
214
215 spin_lock(&vc->lock);
216 vc->arch_compat = arch_compat;
217 vc->pcr = pcr;
218 spin_unlock(&vc->lock);
219
220 return 0;
221}
222
de56a948
PM		 223void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
224{
225 int r;
226
227 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
228 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
229 vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
230 for (r = 0; r < 16; ++r)
231 pr_err("r%2d = %.16lx r%d = %.16lx\n",
232 r, kvmppc_get_gpr(vcpu, r),
233 r+16, kvmppc_get_gpr(vcpu, r+16));
234 pr_err("ctr = %.16lx lr = %.16lx\n",
235 vcpu->arch.ctr, vcpu->arch.lr);
236 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
237 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
238 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
239 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
240 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
241 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
242 pr_err("cr = %.8x xer = %.16lx dsisr = %.8x\n",
243 vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
244 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
245 pr_err("fault dar = %.16lx dsisr = %.8x\n",
246 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
247 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
248 for (r = 0; r < vcpu->arch.slb_max; ++r)
249 pr_err(" ESID = %.16llx VSID = %.16llx\n",
250 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
251 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
a0144e2a 252 vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
de56a948
PM		 253 vcpu->arch.last_inst);
254}
255
a8606e20
PM		 256struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
257{
258 int r;
259 struct kvm_vcpu *v, *ret = NULL;
260
261 mutex_lock(&kvm->lock);
262 kvm_for_each_vcpu(r, v, kvm) {
263 if (v->vcpu_id == id) {
264 ret = v;
265 break;
266 }
267 }
268 mutex_unlock(&kvm->lock);
269 return ret;
270}
271
272static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
273{
f13c13a0 274 vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
a8606e20
PM		 275 vpa->yield_count = 1;
276}
277
55b665b0
PM		 278static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
279 unsigned long addr, unsigned long len)
280{
281 /* check address is cacheline aligned */
282 if (addr & (L1_CACHE_BYTES - 1))
283 return -EINVAL;
284 spin_lock(&vcpu->arch.vpa_update_lock);
285 if (v->next_gpa != addr || v->len != len) {
286 v->next_gpa = addr;
287 v->len = addr ? len : 0;
288 v->update_pending = 1;
289 }
290 spin_unlock(&vcpu->arch.vpa_update_lock);
291 return 0;
292}
293
2e25aa5f
PM		 294/* Length for a per-processor buffer is passed in at offset 4 in the buffer */
295struct reg_vpa {
296 u32 dummy;
297 union {
298 u16 hword;
299 u32 word;
300 } length;
301};
302
303static int vpa_is_registered(struct kvmppc_vpa *vpap)
304{
305 if (vpap->update_pending)
306 return vpap->next_gpa != 0;
307 return vpap->pinned_addr != NULL;
308}
309
a8606e20
PM		 310static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
311 unsigned long flags,
312 unsigned long vcpuid, unsigned long vpa)
313{
314 struct kvm *kvm = vcpu->kvm;
93e60249 315 unsigned long len, nb;
a8606e20
PM		 316 void *va;
317 struct kvm_vcpu *tvcpu;
2e25aa5f
PM		 318 int err;
319 int subfunc;
320 struct kvmppc_vpa *vpap;
a8606e20
PM		 321
322 tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
323 if (!tvcpu)
324 return H_PARAMETER;
325
2e25aa5f
PM		 326 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
327 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
328 subfunc == H_VPA_REG_SLB) {
329 /* Registering new area - address must be cache-line aligned */
330 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
a8606e20 331 return H_PARAMETER;
2e25aa5f
PM		 332
333 /* convert logical addr to kernel addr and read length */
93e60249
PM		 334 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
335 if (va == NULL)
b2b2f165 336 return H_PARAMETER;
2e25aa5f
PM		 337 if (subfunc == H_VPA_REG_VPA)
338 len = ((struct reg_vpa *)va)->length.hword;
a8606e20 339 else
2e25aa5f 340 len = ((struct reg_vpa *)va)->length.word;
c35635ef 341 kvmppc_unpin_guest_page(kvm, va, vpa, false);
2e25aa5f
PM		 342
343 /* Check length */
344 if (len > nb || len < sizeof(struct reg_vpa))
345 return H_PARAMETER;
346 } else {
347 vpa = 0;
348 len = 0;
349 }
350
351 err = H_PARAMETER;
352 vpap = NULL;
353 spin_lock(&tvcpu->arch.vpa_update_lock);
354
355 switch (subfunc) {
356 case H_VPA_REG_VPA: /* register VPA */
357 if (len < sizeof(struct lppaca))
a8606e20 358 break;
2e25aa5f
PM		 359 vpap = &tvcpu->arch.vpa;
360 err = 0;
361 break;
362
363 case H_VPA_REG_DTL: /* register DTL */
364 if (len < sizeof(struct dtl_entry))
a8606e20 365 break;
2e25aa5f
PM		 366 len -= len % sizeof(struct dtl_entry);
367
368 /* Check that they have previously registered a VPA */
369 err = H_RESOURCE;
370 if (!vpa_is_registered(&tvcpu->arch.vpa))
a8606e20 371 break;
2e25aa5f
PM		 372
373 vpap = &tvcpu->arch.dtl;
374 err = 0;
375 break;
376
377 case H_VPA_REG_SLB: /* register SLB shadow buffer */
378 /* Check that they have previously registered a VPA */
379 err = H_RESOURCE;
380 if (!vpa_is_registered(&tvcpu->arch.vpa))
a8606e20 381 break;
2e25aa5f
PM		 382
383 vpap = &tvcpu->arch.slb_shadow;
384 err = 0;
385 break;
386
387 case H_VPA_DEREG_VPA: /* deregister VPA */
388 /* Check they don't still have a DTL or SLB buf registered */
389 err = H_RESOURCE;
390 if (vpa_is_registered(&tvcpu->arch.dtl) ||
391 vpa_is_registered(&tvcpu->arch.slb_shadow))
a8606e20 392 break;
2e25aa5f
PM		 393
394 vpap = &tvcpu->arch.vpa;
395 err = 0;
396 break;
397
398 case H_VPA_DEREG_DTL: /* deregister DTL */
399 vpap = &tvcpu->arch.dtl;
400 err = 0;
401 break;
402
403 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */
404 vpap = &tvcpu->arch.slb_shadow;
405 err = 0;
406 break;
407 }
408
409 if (vpap) {
410 vpap->next_gpa = vpa;
411 vpap->len = len;
412 vpap->update_pending = 1;
a8606e20 413 }
93e60249 414
2e25aa5f
PM		 415 spin_unlock(&tvcpu->arch.vpa_update_lock);
416
93e60249 417 return err;
a8606e20
PM		 418}
419
081f323b 420static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
2e25aa5f 421{
081f323b 422 struct kvm *kvm = vcpu->kvm;
2e25aa5f
PM		 423 void *va;
424 unsigned long nb;
081f323b 425 unsigned long gpa;
2e25aa5f 426
081f323b
PM		 427 /*
428 * We need to pin the page pointed to by vpap->next_gpa,
429 * but we can't call kvmppc_pin_guest_page under the lock
430 * as it does get_user_pages() and down_read(). So we
431 * have to drop the lock, pin the page, then get the lock
432 * again and check that a new area didn't get registered
433 * in the meantime.
434 */
435 for (;;) {
436 gpa = vpap->next_gpa;
437 spin_unlock(&vcpu->arch.vpa_update_lock);
438 va = NULL;
439 nb = 0;
440 if (gpa)
c35635ef 441 va = kvmppc_pin_guest_page(kvm, gpa, &nb);
081f323b
PM		 442 spin_lock(&vcpu->arch.vpa_update_lock);
443 if (gpa == vpap->next_gpa)
444 break;
445 /* sigh... unpin that one and try again */
446 if (va)
c35635ef 447 kvmppc_unpin_guest_page(kvm, va, gpa, false);
081f323b
PM		 448 }
449
450 vpap->update_pending = 0;
451 if (va && nb < vpap->len) {
452 /*
453 * If it's now too short, it must be that userspace
454 * has changed the mappings underlying guest memory,
455 * so unregister the region.
456 */
c35635ef 457 kvmppc_unpin_guest_page(kvm, va, gpa, false);
081f323b 458 va = NULL;
2e25aa5f
PM		 459 }
460 if (vpap->pinned_addr)
c35635ef
PM		 461 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
462 vpap->dirty);
463 vpap->gpa = gpa;
2e25aa5f 464 vpap->pinned_addr = va;
c35635ef 465 vpap->dirty = false;
2e25aa5f
PM		 466 if (va)
467 vpap->pinned_end = va + vpap->len;
468}
469
470static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
471{
2f12f034
PM		 472 if (!(vcpu->arch.vpa.update_pending ||
473 vcpu->arch.slb_shadow.update_pending ||
474 vcpu->arch.dtl.update_pending))
475 return;
476
2e25aa5f
PM		 477 spin_lock(&vcpu->arch.vpa_update_lock);
478 if (vcpu->arch.vpa.update_pending) {
081f323b 479 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
55b665b0
PM		 480 if (vcpu->arch.vpa.pinned_addr)
481 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
2e25aa5f
PM		 482 }
483 if (vcpu->arch.dtl.update_pending) {
081f323b 484 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
2e25aa5f
PM		 485 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
486 vcpu->arch.dtl_index = 0;
487 }
488 if (vcpu->arch.slb_shadow.update_pending)
081f323b 489 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
2e25aa5f
PM		 490 spin_unlock(&vcpu->arch.vpa_update_lock);
491}
492
c7b67670
PM		 493/*
494 * Return the accumulated stolen time for the vcore up until `now'.
495 * The caller should hold the vcore lock.
496 */
497static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
498{
499 u64 p;
500
501 /*
502 * If we are the task running the vcore, then since we hold
503 * the vcore lock, we can't be preempted, so stolen_tb/preempt_tb
504 * can't be updated, so we don't need the tbacct_lock.
505 * If the vcore is inactive, it can't become active (since we
506 * hold the vcore lock), so the vcpu load/put functions won't
507 * update stolen_tb/preempt_tb, and we don't need tbacct_lock.
508 */
509 if (vc->vcore_state != VCORE_INACTIVE &&
510 vc->runner->arch.run_task != current) {
bf3d32e1 511 spin_lock_irq(&vc->runner->arch.tbacct_lock);
c7b67670
PM		 512 p = vc->stolen_tb;
513 if (vc->preempt_tb != TB_NIL)
514 p += now - vc->preempt_tb;
bf3d32e1 515 spin_unlock_irq(&vc->runner->arch.tbacct_lock);
c7b67670
PM		 516 } else {
517 p = vc->stolen_tb;
518 }
519 return p;
520}
521
0456ec4f
PM		 522static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
523 struct kvmppc_vcore *vc)
524{
525 struct dtl_entry *dt;
526 struct lppaca *vpa;
c7b67670
PM		 527 unsigned long stolen;
528 unsigned long core_stolen;
529 u64 now;
0456ec4f
PM		 530
531 dt = vcpu->arch.dtl_ptr;
532 vpa = vcpu->arch.vpa.pinned_addr;
c7b67670
PM		 533 now = mftb();
534 core_stolen = vcore_stolen_time(vc, now);
535 stolen = core_stolen - vcpu->arch.stolen_logged;
536 vcpu->arch.stolen_logged = core_stolen;
bf3d32e1 537 spin_lock_irq(&vcpu->arch.tbacct_lock);
c7b67670
PM		 538 stolen += vcpu->arch.busy_stolen;
539 vcpu->arch.busy_stolen = 0;
bf3d32e1 540 spin_unlock_irq(&vcpu->arch.tbacct_lock);
0456ec4f
PM		 541 if (!dt || !vpa)
542 return;
543 memset(dt, 0, sizeof(struct dtl_entry));
544 dt->dispatch_reason = 7;
545 dt->processor_id = vc->pcpu + vcpu->arch.ptid;
93b0f4dc 546 dt->timebase = now + vc->tb_offset;
c7b67670 547 dt->enqueue_to_dispatch_time = stolen;
0456ec4f
PM		 548 dt->srr0 = kvmppc_get_pc(vcpu);
549 dt->srr1 = vcpu->arch.shregs.msr;
550 ++dt;
551 if (dt == vcpu->arch.dtl.pinned_end)
552 dt = vcpu->arch.dtl.pinned_addr;
553 vcpu->arch.dtl_ptr = dt;
554 /* order writing *dt vs. writing vpa->dtl_idx */
555 smp_wmb();
556 vpa->dtl_idx = ++vcpu->arch.dtl_index;
c35635ef 557 vcpu->arch.dtl.dirty = true;
0456ec4f
PM		 558}
559
a8606e20
PM		 560int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
561{
562 unsigned long req = kvmppc_get_gpr(vcpu, 3);
563 unsigned long target, ret = H_SUCCESS;
564 struct kvm_vcpu *tvcpu;
8e591cb7 565 int idx, rc;
a8606e20 566
699a0ea0
PM		 567 if (req <= MAX_HCALL_OPCODE &&
568 !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
569 return RESUME_HOST;
570
a8606e20 571 switch (req) {
c77162de 572 case H_ENTER:
2c9097e4 573 idx = srcu_read_lock(&vcpu->kvm->srcu);
c77162de
PM		 574 ret = kvmppc_virtmode_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
575 kvmppc_get_gpr(vcpu, 5),
576 kvmppc_get_gpr(vcpu, 6),
577 kvmppc_get_gpr(vcpu, 7));
2c9097e4 578 srcu_read_unlock(&vcpu->kvm->srcu, idx);
c77162de 579 break;
a8606e20 580 case H_CEDE:
a8606e20
PM		 581 break;
582 case H_PROD:
583 target = kvmppc_get_gpr(vcpu, 4);
584 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
585 if (!tvcpu) {
586 ret = H_PARAMETER;
587 break;
588 }
589 tvcpu->arch.prodded = 1;
590 smp_mb();
591 if (vcpu->arch.ceded) {
592 if (waitqueue_active(&vcpu->wq)) {
593 wake_up_interruptible(&vcpu->wq);
594 vcpu->stat.halt_wakeup++;
595 }
596 }
597 break;
598 case H_CONFER:
42d7604d
PM		 599 target = kvmppc_get_gpr(vcpu, 4);
600 if (target == -1)
601 break;
602 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
603 if (!tvcpu) {
604 ret = H_PARAMETER;
605 break;
606 }
607 kvm_vcpu_yield_to(tvcpu);
a8606e20
PM		 608 break;
609 case H_REGISTER_VPA:
610 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
611 kvmppc_get_gpr(vcpu, 5),
612 kvmppc_get_gpr(vcpu, 6));
613 break;
8e591cb7
ME		 614 case H_RTAS:
615 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
616 return RESUME_HOST;
617
c9438092 618 idx = srcu_read_lock(&vcpu->kvm->srcu);
8e591cb7 619 rc = kvmppc_rtas_hcall(vcpu);
c9438092 620 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8e591cb7
ME		 621
622 if (rc == -ENOENT)
623 return RESUME_HOST;
624 else if (rc == 0)
625 break;
626
627 /* Send the error out to userspace via KVM_RUN */
628 return rc;
bc5ad3f3
BH		 629
630 case H_XIRR:
631 case H_CPPR:
632 case H_EOI:
633 case H_IPI:
8e44ddc3
PM		 634 case H_IPOLL:
635 case H_XIRR_X:
bc5ad3f3
BH		 636 if (kvmppc_xics_enabled(vcpu)) {
637 ret = kvmppc_xics_hcall(vcpu, req);
638 break;
639 } /* fallthrough */
a8606e20
PM		 640 default:
641 return RESUME_HOST;
642 }
643 kvmppc_set_gpr(vcpu, 3, ret);
644 vcpu->arch.hcall_needed = 0;
645 return RESUME_GUEST;
646}
647
3a167bea
AK		 648static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
649 struct task_struct *tsk)
de56a948
PM		 650{
651 int r = RESUME_HOST;
652
653 vcpu->stat.sum_exits++;
654
655 run->exit_reason = KVM_EXIT_UNKNOWN;
656 run->ready_for_interrupt_injection = 1;
657 switch (vcpu->arch.trap) {
658 /* We're good on these - the host merely wanted to get our attention */
659 case BOOK3S_INTERRUPT_HV_DECREMENTER:
660 vcpu->stat.dec_exits++;
661 r = RESUME_GUEST;
662 break;
663 case BOOK3S_INTERRUPT_EXTERNAL:
5d00f66b 664 case BOOK3S_INTERRUPT_H_DOORBELL:
de56a948
PM		 665 vcpu->stat.ext_intr_exits++;
666 r = RESUME_GUEST;
667 break;
668 case BOOK3S_INTERRUPT_PERFMON:
669 r = RESUME_GUEST;
670 break;
b4072df4
PM		 671 case BOOK3S_INTERRUPT_MACHINE_CHECK:
672 /*
673 * Deliver a machine check interrupt to the guest.
674 * We have to do this, even if the host has handled the
675 * machine check, because machine checks use SRR0/1 and
676 * the interrupt might have trashed guest state in them.
677 */
678 kvmppc_book3s_queue_irqprio(vcpu,
679 BOOK3S_INTERRUPT_MACHINE_CHECK);
680 r = RESUME_GUEST;
681 break;
de56a948
PM		 682 case BOOK3S_INTERRUPT_PROGRAM:
683 {
684 ulong flags;
685 /*
686 * Normally program interrupts are delivered directly
687 * to the guest by the hardware, but we can get here
688 * as a result of a hypervisor emulation interrupt
689 * (e40) getting turned into a 700 by BML RTAS.
690 */
691 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
692 kvmppc_core_queue_program(vcpu, flags);
693 r = RESUME_GUEST;
694 break;
695 }
696 case BOOK3S_INTERRUPT_SYSCALL:
697 {
698 /* hcall - punt to userspace */
699 int i;
700
27025a60
LPF		 701 /* hypercall with MSR_PR has already been handled in rmode,
702 * and never reaches here.
703 */
704
de56a948
PM		 705 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
706 for (i = 0; i < 9; ++i)
707 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
708 run->exit_reason = KVM_EXIT_PAPR_HCALL;
709 vcpu->arch.hcall_needed = 1;
710 r = RESUME_HOST;
711 break;
712 }
713 /*
342d3db7
PM		 714 * We get these next two if the guest accesses a page which it thinks
715 * it has mapped but which is not actually present, either because
716 * it is for an emulated I/O device or because the corresonding
717 * host page has been paged out. Any other HDSI/HISI interrupts
718 * have been handled already.
de56a948
PM		 719 */
720 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
913d3ff9 721 r = RESUME_PAGE_FAULT;
de56a948
PM		 722 break;
723 case BOOK3S_INTERRUPT_H_INST_STORAGE:
913d3ff9
PM		 724 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
725 vcpu->arch.fault_dsisr = 0;
726 r = RESUME_PAGE_FAULT;
de56a948
PM		 727 break;
728 /*
729 * This occurs if the guest executes an illegal instruction.
730 * We just generate a program interrupt to the guest, since
731 * we don't emulate any guest instructions at this stage.
732 */
733 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
bd3048b8
ME		 734 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
735 r = RESUME_GUEST;
736 break;
737 /*
738 * This occurs if the guest (kernel or userspace), does something that
739 * is prohibited by HFSCR. We just generate a program interrupt to
740 * the guest.
741 */
742 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
743 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
de56a948
PM		 744 r = RESUME_GUEST;
745 break;
746 default:
747 kvmppc_dump_regs(vcpu);
748 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
749 vcpu->arch.trap, kvmppc_get_pc(vcpu),
750 vcpu->arch.shregs.msr);
f3271d4c 751 run->hw.hardware_exit_reason = vcpu->arch.trap;
de56a948 752 r = RESUME_HOST;
de56a948
PM		 753 break;
754 }
755
de56a948
PM		 756 return r;
757}
758
3a167bea
AK		 759static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
760 struct kvm_sregs *sregs)
de56a948
PM		 761{
762 int i;
763
de56a948 764 memset(sregs, 0, sizeof(struct kvm_sregs));
87916442 765 sregs->pvr = vcpu->arch.pvr;
de56a948
PM		 766 for (i = 0; i < vcpu->arch.slb_max; i++) {
767 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
768 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
769 }
770
771 return 0;
772}
773
3a167bea
AK		 774static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
775 struct kvm_sregs *sregs)
de56a948
PM		 776{
777 int i, j;
778
3a167bea 779 kvmppc_set_pvr_hv(vcpu, sregs->pvr);
de56a948
PM		 780
781 j = 0;
782 for (i = 0; i < vcpu->arch.slb_nr; i++) {
783 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
784 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
785 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
786 ++j;
787 }
788 }
789 vcpu->arch.slb_max = j;
790
791 return 0;
792}
793
a0144e2a
PM		 794static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr)
795{
796 struct kvmppc_vcore *vc = vcpu->arch.vcore;
797 u64 mask;
798
799 spin_lock(&vc->lock);
d682916a
AB		 800 /*
801 * If ILE (interrupt little-endian) has changed, update the
802 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
803 */
804 if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
805 struct kvm *kvm = vcpu->kvm;
806 struct kvm_vcpu *vcpu;
807 int i;
808
809 mutex_lock(&kvm->lock);
810 kvm_for_each_vcpu(i, vcpu, kvm) {
811 if (vcpu->arch.vcore != vc)
812 continue;
813 if (new_lpcr & LPCR_ILE)
814 vcpu->arch.intr_msr |= MSR_LE;
815 else
816 vcpu->arch.intr_msr &= ~MSR_LE;
817 }
818 mutex_unlock(&kvm->lock);
819 }
820
a0144e2a
PM		 821 /*
822 * Userspace can only modify DPFD (default prefetch depth),
823 * ILE (interrupt little-endian) and TC (translation control).
e0622bd9 824 * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
a0144e2a
PM		 825 */
826 mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
e0622bd9
PM		 827 if (cpu_has_feature(CPU_FTR_ARCH_207S))
828 mask |= LPCR_AIL;
a0144e2a
PM		 829 vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
830 spin_unlock(&vc->lock);
831}
832
3a167bea
AK		 833static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
834 union kvmppc_one_reg *val)
31f3438e 835{
a136a8bd
PM		 836 int r = 0;
837 long int i;
31f3438e 838
a136a8bd 839 switch (id) {
31f3438e 840 case KVM_REG_PPC_HIOR:
a136a8bd
PM		 841 *val = get_reg_val(id, 0);
842 break;
843 case KVM_REG_PPC_DABR:
844 *val = get_reg_val(id, vcpu->arch.dabr);
845 break;
8563bf52
PM		 846 case KVM_REG_PPC_DABRX:
847 *val = get_reg_val(id, vcpu->arch.dabrx);
848 break;
a136a8bd
PM		 849 case KVM_REG_PPC_DSCR:
850 *val = get_reg_val(id, vcpu->arch.dscr);
851 break;
852 case KVM_REG_PPC_PURR:
853 *val = get_reg_val(id, vcpu->arch.purr);
854 break;
855 case KVM_REG_PPC_SPURR:
856 *val = get_reg_val(id, vcpu->arch.spurr);
857 break;
858 case KVM_REG_PPC_AMR:
859 *val = get_reg_val(id, vcpu->arch.amr);
860 break;
861 case KVM_REG_PPC_UAMOR:
862 *val = get_reg_val(id, vcpu->arch.uamor);
863 break;
b005255e 864 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
a136a8bd
PM		 865 i = id - KVM_REG_PPC_MMCR0;
866 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
867 break;
868 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
869 i = id - KVM_REG_PPC_PMC1;
870 *val = get_reg_val(id, vcpu->arch.pmc[i]);
31f3438e 871 break;
b005255e
MN		 872 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
873 i = id - KVM_REG_PPC_SPMC1;
874 *val = get_reg_val(id, vcpu->arch.spmc[i]);
875 break;
14941789
PM		 876 case KVM_REG_PPC_SIAR:
877 *val = get_reg_val(id, vcpu->arch.siar);
878 break;
879 case KVM_REG_PPC_SDAR:
880 *val = get_reg_val(id, vcpu->arch.sdar);
881 break;
b005255e
MN		 882 case KVM_REG_PPC_SIER:
883 *val = get_reg_val(id, vcpu->arch.sier);
a8bd19ef 884 break;
b005255e
MN		 885 case KVM_REG_PPC_IAMR:
886 *val = get_reg_val(id, vcpu->arch.iamr);
887 break;
b005255e
MN		 888 case KVM_REG_PPC_PSPB:
889 *val = get_reg_val(id, vcpu->arch.pspb);
890 break;
b005255e
MN		 891 case KVM_REG_PPC_DPDES:
892 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
893 break;
894 case KVM_REG_PPC_DAWR:
895 *val = get_reg_val(id, vcpu->arch.dawr);
896 break;
897 case KVM_REG_PPC_DAWRX:
898 *val = get_reg_val(id, vcpu->arch.dawrx);
899 break;
900 case KVM_REG_PPC_CIABR:
901 *val = get_reg_val(id, vcpu->arch.ciabr);
902 break;
b005255e
MN		 903 case KVM_REG_PPC_CSIGR:
904 *val = get_reg_val(id, vcpu->arch.csigr);
905 break;
906 case KVM_REG_PPC_TACR:
907 *val = get_reg_val(id, vcpu->arch.tacr);
908 break;
909 case KVM_REG_PPC_TCSCR:
910 *val = get_reg_val(id, vcpu->arch.tcscr);
911 break;
912 case KVM_REG_PPC_PID:
913 *val = get_reg_val(id, vcpu->arch.pid);
914 break;
915 case KVM_REG_PPC_ACOP:
916 *val = get_reg_val(id, vcpu->arch.acop);
917 break;
918 case KVM_REG_PPC_WORT:
919 *val = get_reg_val(id, vcpu->arch.wort);
a8bd19ef 920 break;
55b665b0
PM		 921 case KVM_REG_PPC_VPA_ADDR:
922 spin_lock(&vcpu->arch.vpa_update_lock);
923 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
924 spin_unlock(&vcpu->arch.vpa_update_lock);
925 break;
926 case KVM_REG_PPC_VPA_SLB:
927 spin_lock(&vcpu->arch.vpa_update_lock);
928 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
929 val->vpaval.length = vcpu->arch.slb_shadow.len;
930 spin_unlock(&vcpu->arch.vpa_update_lock);
931 break;
932 case KVM_REG_PPC_VPA_DTL:
933 spin_lock(&vcpu->arch.vpa_update_lock);
934 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
935 val->vpaval.length = vcpu->arch.dtl.len;
936 spin_unlock(&vcpu->arch.vpa_update_lock);
937 break;
93b0f4dc
PM		 938 case KVM_REG_PPC_TB_OFFSET:
939 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
940 break;
a0144e2a
PM		 941 case KVM_REG_PPC_LPCR:
942 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
943 break;
4b8473c9
PM		 944 case KVM_REG_PPC_PPR:
945 *val = get_reg_val(id, vcpu->arch.ppr);
946 break;
a7d80d01
MN		 947#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
948 case KVM_REG_PPC_TFHAR:
949 *val = get_reg_val(id, vcpu->arch.tfhar);
950 break;
951 case KVM_REG_PPC_TFIAR:
952 *val = get_reg_val(id, vcpu->arch.tfiar);
953 break;
954 case KVM_REG_PPC_TEXASR:
955 *val = get_reg_val(id, vcpu->arch.texasr);
956 break;
957 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
958 i = id - KVM_REG_PPC_TM_GPR0;
959 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
960 break;
961 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
962 {
963 int j;
964 i = id - KVM_REG_PPC_TM_VSR0;
965 if (i < 32)
966 for (j = 0; j < TS_FPRWIDTH; j++)
967 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
968 else {
969 if (cpu_has_feature(CPU_FTR_ALTIVEC))
970 val->vval = vcpu->arch.vr_tm.vr[i-32];
971 else
972 r = -ENXIO;
973 }
974 break;
975 }
976 case KVM_REG_PPC_TM_CR:
977 *val = get_reg_val(id, vcpu->arch.cr_tm);
978 break;
979 case KVM_REG_PPC_TM_LR:
980 *val = get_reg_val(id, vcpu->arch.lr_tm);
981 break;
982 case KVM_REG_PPC_TM_CTR:
983 *val = get_reg_val(id, vcpu->arch.ctr_tm);
984 break;
985 case KVM_REG_PPC_TM_FPSCR:
986 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
987 break;
988 case KVM_REG_PPC_TM_AMR:
989 *val = get_reg_val(id, vcpu->arch.amr_tm);
990 break;
991 case KVM_REG_PPC_TM_PPR:
992 *val = get_reg_val(id, vcpu->arch.ppr_tm);
993 break;
994 case KVM_REG_PPC_TM_VRSAVE:
995 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
996 break;
997 case KVM_REG_PPC_TM_VSCR:
998 if (cpu_has_feature(CPU_FTR_ALTIVEC))
999 *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1000 else
1001 r = -ENXIO;
1002 break;
1003 case KVM_REG_PPC_TM_DSCR:
1004 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1005 break;
1006 case KVM_REG_PPC_TM_TAR:
1007 *val = get_reg_val(id, vcpu->arch.tar_tm);
1008 break;
1009#endif
388cc6e1
PM		 1010 case KVM_REG_PPC_ARCH_COMPAT:
1011 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1012 break;
31f3438e 1013 default:
a136a8bd 1014 r = -EINVAL;
31f3438e
PM		 1015 break;
1016 }
1017
1018 return r;
1019}
1020
3a167bea
AK		 1021static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1022 union kvmppc_one_reg *val)
31f3438e 1023{
a136a8bd
PM		 1024 int r = 0;
1025 long int i;
55b665b0 1026 unsigned long addr, len;
31f3438e 1027
a136a8bd 1028 switch (id) {
31f3438e 1029 case KVM_REG_PPC_HIOR:
31f3438e 1030 /* Only allow this to be set to zero */
a136a8bd 1031 if (set_reg_val(id, *val))
31f3438e
PM		 1032 r = -EINVAL;
1033 break;
a136a8bd
PM		 1034 case KVM_REG_PPC_DABR:
1035 vcpu->arch.dabr = set_reg_val(id, *val);
1036 break;
8563bf52
PM		 1037 case KVM_REG_PPC_DABRX:
1038 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1039 break;
a136a8bd
PM		 1040 case KVM_REG_PPC_DSCR:
1041 vcpu->arch.dscr = set_reg_val(id, *val);
1042 break;
1043 case KVM_REG_PPC_PURR:
1044 vcpu->arch.purr = set_reg_val(id, *val);
1045 break;
1046 case KVM_REG_PPC_SPURR:
1047 vcpu->arch.spurr = set_reg_val(id, *val);
1048 break;
1049 case KVM_REG_PPC_AMR:
1050 vcpu->arch.amr = set_reg_val(id, *val);
1051 break;
1052 case KVM_REG_PPC_UAMOR:
1053 vcpu->arch.uamor = set_reg_val(id, *val);
1054 break;
b005255e 1055 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
a136a8bd
PM		 1056 i = id - KVM_REG_PPC_MMCR0;
1057 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1058 break;
1059 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1060 i = id - KVM_REG_PPC_PMC1;
1061 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1062 break;
b005255e
MN		 1063 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1064 i = id - KVM_REG_PPC_SPMC1;
1065 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1066 break;
14941789
PM		 1067 case KVM_REG_PPC_SIAR:
1068 vcpu->arch.siar = set_reg_val(id, *val);
1069 break;
1070 case KVM_REG_PPC_SDAR:
1071 vcpu->arch.sdar = set_reg_val(id, *val);
1072 break;
b005255e
MN		 1073 case KVM_REG_PPC_SIER:
1074 vcpu->arch.sier = set_reg_val(id, *val);
a8bd19ef 1075 break;
b005255e
MN		 1076 case KVM_REG_PPC_IAMR:
1077 vcpu->arch.iamr = set_reg_val(id, *val);
1078 break;
b005255e
MN		 1079 case KVM_REG_PPC_PSPB:
1080 vcpu->arch.pspb = set_reg_val(id, *val);
1081 break;
b005255e
MN		 1082 case KVM_REG_PPC_DPDES:
1083 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1084 break;
1085 case KVM_REG_PPC_DAWR:
1086 vcpu->arch.dawr = set_reg_val(id, *val);
1087 break;
1088 case KVM_REG_PPC_DAWRX:
1089 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1090 break;
1091 case KVM_REG_PPC_CIABR:
1092 vcpu->arch.ciabr = set_reg_val(id, *val);
1093 /* Don't allow setting breakpoints in hypervisor code */
1094 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1095 vcpu->arch.ciabr &= ~CIABR_PRIV; /* disable */
1096 break;
b005255e
MN		 1097 case KVM_REG_PPC_CSIGR:
1098 vcpu->arch.csigr = set_reg_val(id, *val);
1099 break;
1100 case KVM_REG_PPC_TACR:
1101 vcpu->arch.tacr = set_reg_val(id, *val);
1102 break;
1103 case KVM_REG_PPC_TCSCR:
1104 vcpu->arch.tcscr = set_reg_val(id, *val);
1105 break;
1106 case KVM_REG_PPC_PID:
1107 vcpu->arch.pid = set_reg_val(id, *val);
1108 break;
1109 case KVM_REG_PPC_ACOP:
1110 vcpu->arch.acop = set_reg_val(id, *val);
1111 break;
1112 case KVM_REG_PPC_WORT:
1113 vcpu->arch.wort = set_reg_val(id, *val);
a8bd19ef 1114 break;
55b665b0
PM		 1115 case KVM_REG_PPC_VPA_ADDR:
1116 addr = set_reg_val(id, *val);
1117 r = -EINVAL;
1118 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1119 vcpu->arch.dtl.next_gpa))
1120 break;
1121 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1122 break;
1123 case KVM_REG_PPC_VPA_SLB:
1124 addr = val->vpaval.addr;
1125 len = val->vpaval.length;
1126 r = -EINVAL;
1127 if (addr && !vcpu->arch.vpa.next_gpa)
1128 break;
1129 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1130 break;
1131 case KVM_REG_PPC_VPA_DTL:
1132 addr = val->vpaval.addr;
1133 len = val->vpaval.length;
1134 r = -EINVAL;
9f8c8c78
PM		 1135 if (addr && (len < sizeof(struct dtl_entry) ||
1136 !vcpu->arch.vpa.next_gpa))
55b665b0
PM		 1137 break;
1138 len -= len % sizeof(struct dtl_entry);
1139 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1140 break;
93b0f4dc
PM		 1141 case KVM_REG_PPC_TB_OFFSET:
1142 /* round up to multiple of 2^24 */
1143 vcpu->arch.vcore->tb_offset =
1144 ALIGN(set_reg_val(id, *val), 1UL << 24);
1145 break;
a0144e2a
PM		 1146 case KVM_REG_PPC_LPCR:
1147 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val));
1148 break;
4b8473c9
PM		 1149 case KVM_REG_PPC_PPR:
1150 vcpu->arch.ppr = set_reg_val(id, *val);
1151 break;
a7d80d01
MN		 1152#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1153 case KVM_REG_PPC_TFHAR:
1154 vcpu->arch.tfhar = set_reg_val(id, *val);
1155 break;
1156 case KVM_REG_PPC_TFIAR:
1157 vcpu->arch.tfiar = set_reg_val(id, *val);
1158 break;
1159 case KVM_REG_PPC_TEXASR:
1160 vcpu->arch.texasr = set_reg_val(id, *val);
1161 break;
1162 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1163 i = id - KVM_REG_PPC_TM_GPR0;
1164 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1165 break;
1166 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1167 {
1168 int j;
1169 i = id - KVM_REG_PPC_TM_VSR0;
1170 if (i < 32)
1171 for (j = 0; j < TS_FPRWIDTH; j++)
1172 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1173 else
1174 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1175 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1176 else
1177 r = -ENXIO;
1178 break;
1179 }
1180 case KVM_REG_PPC_TM_CR:
1181 vcpu->arch.cr_tm = set_reg_val(id, *val);
1182 break;
1183 case KVM_REG_PPC_TM_LR:
1184 vcpu->arch.lr_tm = set_reg_val(id, *val);
1185 break;
1186 case KVM_REG_PPC_TM_CTR:
1187 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1188 break;
1189 case KVM_REG_PPC_TM_FPSCR:
1190 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1191 break;
1192 case KVM_REG_PPC_TM_AMR:
1193 vcpu->arch.amr_tm = set_reg_val(id, *val);
1194 break;
1195 case KVM_REG_PPC_TM_PPR:
1196 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1197 break;
1198 case KVM_REG_PPC_TM_VRSAVE:
1199 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1200 break;
1201 case KVM_REG_PPC_TM_VSCR:
1202 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1203 vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1204 else
1205 r = - ENXIO;
1206 break;
1207 case KVM_REG_PPC_TM_DSCR:
1208 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1209 break;
1210 case KVM_REG_PPC_TM_TAR:
1211 vcpu->arch.tar_tm = set_reg_val(id, *val);
1212 break;
1213#endif
388cc6e1
PM		 1214 case KVM_REG_PPC_ARCH_COMPAT:
1215 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1216 break;
31f3438e 1217 default:
a136a8bd 1218 r = -EINVAL;
31f3438e
PM		 1219 break;
1220 }
1221
1222 return r;
1223}
1224
3a167bea
AK		 1225static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1226 unsigned int id)
de56a948
PM		 1227{
1228 struct kvm_vcpu *vcpu;
371fefd6
PM		 1229 int err = -EINVAL;
1230 int core;
1231 struct kvmppc_vcore *vcore;
de56a948 1232
3102f784 1233 core = id / threads_per_subcore;
371fefd6
PM		 1234 if (core >= KVM_MAX_VCORES)
1235 goto out;
1236
1237 err = -ENOMEM;
6b75e6bf 1238 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
de56a948
PM		 1239 if (!vcpu)
1240 goto out;
1241
1242 err = kvm_vcpu_init(vcpu, kvm, id);
1243 if (err)
1244 goto free_vcpu;
1245
1246 vcpu->arch.shared = &vcpu->arch.shregs;
5deb8e7a
AG		 1247#ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1248 /*
1249 * The shared struct is never shared on HV,
1250 * so we can always use host endianness
1251 */
1252#ifdef __BIG_ENDIAN__
1253 vcpu->arch.shared_big_endian = true;
1254#else
1255 vcpu->arch.shared_big_endian = false;
1256#endif
1257#endif
de56a948
PM		 1258 vcpu->arch.mmcr[0] = MMCR0_FC;
1259 vcpu->arch.ctrl = CTRL_RUNLATCH;
1260 /* default to host PVR, since we can't spoof it */
3a167bea 1261 kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2e25aa5f 1262 spin_lock_init(&vcpu->arch.vpa_update_lock);
c7b67670
PM		 1263 spin_lock_init(&vcpu->arch.tbacct_lock);
1264 vcpu->arch.busy_preempt = TB_NIL;
d682916a 1265 vcpu->arch.intr_msr = MSR_SF | MSR_ME;
de56a948 1266
de56a948
PM		 1267 kvmppc_mmu_book3s_hv_init(vcpu);
1268
8455d79e 1269 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
371fefd6
PM		 1270
1271 init_waitqueue_head(&vcpu->arch.cpu_run);
1272
1273 mutex_lock(&kvm->lock);
1274 vcore = kvm->arch.vcores[core];
1275 if (!vcore) {
1276 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1277 if (vcore) {
1278 INIT_LIST_HEAD(&vcore->runnable_threads);
1279 spin_lock_init(&vcore->lock);
19ccb76a 1280 init_waitqueue_head(&vcore->wq);
c7b67670 1281 vcore->preempt_tb = TB_NIL;
a0144e2a 1282 vcore->lpcr = kvm->arch.lpcr;
3102f784 1283 vcore->first_vcpuid = core * threads_per_subcore;
e0b7ec05 1284 vcore->kvm = kvm;
371fefd6
PM		 1285 }
1286 kvm->arch.vcores[core] = vcore;
1b400ba0 1287 kvm->arch.online_vcores++;
371fefd6
PM		 1288 }
1289 mutex_unlock(&kvm->lock);
1290
1291 if (!vcore)
1292 goto free_vcpu;
1293
1294 spin_lock(&vcore->lock);
1295 ++vcore->num_threads;
371fefd6
PM		 1296 spin_unlock(&vcore->lock);
1297 vcpu->arch.vcore = vcore;
e0b7ec05 1298 vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
371fefd6 1299
af8f38b3
AG		 1300 vcpu->arch.cpu_type = KVM_CPU_3S_64;
1301 kvmppc_sanity_check(vcpu);
1302
de56a948
PM		 1303 return vcpu;
1304
1305free_vcpu:
6b75e6bf 1306 kmem_cache_free(kvm_vcpu_cache, vcpu);
de56a948
PM		 1307out:
1308 return ERR_PTR(err);
1309}
1310
c35635ef
PM		 1311static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1312{
1313 if (vpa->pinned_addr)
1314 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1315 vpa->dirty);
1316}
1317
3a167bea 1318static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
de56a948 1319{
2e25aa5f 1320 spin_lock(&vcpu->arch.vpa_update_lock);
c35635ef
PM		 1321 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1322 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1323 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2e25aa5f 1324 spin_unlock(&vcpu->arch.vpa_update_lock);
de56a948 1325 kvm_vcpu_uninit(vcpu);
6b75e6bf 1326 kmem_cache_free(kvm_vcpu_cache, vcpu);
de56a948
PM		 1327}
1328
3a167bea
AK		 1329static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1330{
1331 /* Indicate we want to get back into the guest */
1332 return 1;
1333}
1334
19ccb76a 1335static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
371fefd6 1336{
19ccb76a 1337 unsigned long dec_nsec, now;
371fefd6 1338
19ccb76a
PM		 1339 now = get_tb();
1340 if (now > vcpu->arch.dec_expires) {
1341 /* decrementer has already gone negative */
1342 kvmppc_core_queue_dec(vcpu);
7e28e60e 1343 kvmppc_core_prepare_to_enter(vcpu);
19ccb76a 1344 return;
371fefd6 1345 }
19ccb76a
PM		 1346 dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1347 / tb_ticks_per_sec;
1348 hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1349 HRTIMER_MODE_REL);
1350 vcpu->arch.timer_running = 1;
371fefd6
PM		 1351}
1352
19ccb76a 1353static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
371fefd6 1354{
19ccb76a
PM		 1355 vcpu->arch.ceded = 0;
1356 if (vcpu->arch.timer_running) {
1357 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1358 vcpu->arch.timer_running = 0;
1359 }
371fefd6
PM		 1360}
1361
e0b7ec05 1362extern void __kvmppc_vcore_entry(void);
de56a948 1363
371fefd6
PM		 1364static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1365 struct kvm_vcpu *vcpu)
de56a948 1366{
c7b67670
PM		 1367 u64 now;
1368
371fefd6
PM		 1369 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1370 return;
bf3d32e1 1371 spin_lock_irq(&vcpu->arch.tbacct_lock);
c7b67670
PM		 1372 now = mftb();
1373 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1374 vcpu->arch.stolen_logged;
1375 vcpu->arch.busy_preempt = now;
1376 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
bf3d32e1 1377 spin_unlock_irq(&vcpu->arch.tbacct_lock);
371fefd6 1378 --vc->n_runnable;
371fefd6
PM		 1379 list_del(&vcpu->arch.run_list);
1380}
1381
f0888f70
PM		 1382static int kvmppc_grab_hwthread(int cpu)
1383{
1384 struct paca_struct *tpaca;
1385 long timeout = 1000;
1386
1387 tpaca = &paca[cpu];
1388
1389 /* Ensure the thread won't go into the kernel if it wakes */
1390 tpaca->kvm_hstate.hwthread_req = 1;
7b444c67 1391 tpaca->kvm_hstate.kvm_vcpu = NULL;
f0888f70
PM		 1392
1393 /*
1394 * If the thread is already executing in the kernel (e.g. handling
1395 * a stray interrupt), wait for it to get back to nap mode.
1396 * The smp_mb() is to ensure that our setting of hwthread_req
1397 * is visible before we look at hwthread_state, so if this
1398 * races with the code at system_reset_pSeries and the thread
1399 * misses our setting of hwthread_req, we are sure to see its
1400 * setting of hwthread_state, and vice versa.
1401 */
1402 smp_mb();
1403 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1404 if (--timeout <= 0) {
1405 pr_err("KVM: couldn't grab cpu %d\n", cpu);
1406 return -EBUSY;
1407 }
1408 udelay(1);
1409 }
1410 return 0;
1411}
1412
1413static void kvmppc_release_hwthread(int cpu)
1414{
1415 struct paca_struct *tpaca;
1416
1417 tpaca = &paca[cpu];
1418 tpaca->kvm_hstate.hwthread_req = 0;
1419 tpaca->kvm_hstate.kvm_vcpu = NULL;
1420}
1421
371fefd6
PM		 1422static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
1423{
1424 int cpu;
1425 struct paca_struct *tpaca;
1426 struct kvmppc_vcore *vc = vcpu->arch.vcore;
1427
19ccb76a
PM		 1428 if (vcpu->arch.timer_running) {
1429 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1430 vcpu->arch.timer_running = 0;
1431 }
371fefd6
PM		 1432 cpu = vc->pcpu + vcpu->arch.ptid;
1433 tpaca = &paca[cpu];
1434 tpaca->kvm_hstate.kvm_vcpu = vcpu;
1435 tpaca->kvm_hstate.kvm_vcore = vc;
e0b7ec05 1436 tpaca->kvm_hstate.ptid = vcpu->arch.ptid;
19ccb76a 1437 vcpu->cpu = vc->pcpu;
371fefd6 1438 smp_wmb();
251da038 1439#if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
e0b7ec05 1440 if (cpu != smp_processor_id()) {
371fefd6 1441 xics_wake_cpu(cpu);
e0b7ec05
PM		 1442 if (vcpu->arch.ptid)
1443 ++vc->n_woken;
de56a948 1444 }
371fefd6
PM		 1445#endif
1446}
de56a948 1447
371fefd6
PM		 1448static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc)
1449{
1450 int i;
1451
1452 HMT_low();
1453 i = 0;
1454 while (vc->nap_count < vc->n_woken) {
1455 if (++i >= 1000000) {
1456 pr_err("kvmppc_wait_for_nap timeout %d %d\n",
1457 vc->nap_count, vc->n_woken);
1458 break;
1459 }
1460 cpu_relax();
1461 }
1462 HMT_medium();
1463}
1464
1465/*
1466 * Check that we are on thread 0 and that any other threads in
7b444c67
PM		 1467 * this core are off-line. Then grab the threads so they can't
1468 * enter the kernel.
371fefd6
PM		 1469 */
1470static int on_primary_thread(void)
1471{
1472 int cpu = smp_processor_id();
3102f784 1473 int thr;
371fefd6 1474
3102f784
ME		 1475 /* Are we on a primary subcore? */
1476 if (cpu_thread_in_subcore(cpu))
371fefd6 1477 return 0;
3102f784
ME		 1478
1479 thr = 0;
1480 while (++thr < threads_per_subcore)
371fefd6
PM		 1481 if (cpu_online(cpu + thr))
1482 return 0;
7b444c67
PM		 1483
1484 /* Grab all hw threads so they can't go into the kernel */
3102f784 1485 for (thr = 1; thr < threads_per_subcore; ++thr) {
7b444c67
PM		 1486 if (kvmppc_grab_hwthread(cpu + thr)) {
1487 /* Couldn't grab one; let the others go */
1488 do {
1489 kvmppc_release_hwthread(cpu + thr);
1490 } while (--thr > 0);
1491 return 0;
1492 }
1493 }
371fefd6
PM		 1494 return 1;
1495}
1496
1497/*
1498 * Run a set of guest threads on a physical core.
1499 * Called with vc->lock held.
1500 */
913d3ff9 1501static void kvmppc_run_core(struct kvmppc_vcore *vc)
371fefd6 1502{
e0b7ec05 1503 struct kvm_vcpu *vcpu, *vnext;
371fefd6
PM		 1504 long ret;
1505 u64 now;
e0b7ec05 1506 int i, need_vpa_update;
2c9097e4 1507 int srcu_idx;
913d3ff9 1508 struct kvm_vcpu *vcpus_to_update[threads_per_core];
371fefd6
PM		 1509
1510 /* don't start if any threads have a signal pending */
081f323b
PM		 1511 need_vpa_update = 0;
1512 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
371fefd6 1513 if (signal_pending(vcpu->arch.run_task))
913d3ff9
PM		 1514 return;
1515 if (vcpu->arch.vpa.update_pending ||
1516 vcpu->arch.slb_shadow.update_pending ||
1517 vcpu->arch.dtl.update_pending)
1518 vcpus_to_update[need_vpa_update++] = vcpu;
081f323b
PM		 1519 }
1520
1521 /*
1522 * Initialize *vc, in particular vc->vcore_state, so we can
1523 * drop the vcore lock if necessary.
1524 */
1525 vc->n_woken = 0;
1526 vc->nap_count = 0;
1527 vc->entry_exit_count = 0;
2f12f034 1528 vc->vcore_state = VCORE_STARTING;
081f323b
PM		 1529 vc->in_guest = 0;
1530 vc->napping_threads = 0;
1531
1532 /*
1533 * Updating any of the vpas requires calling kvmppc_pin_guest_page,
1534 * which can't be called with any spinlocks held.
1535 */
1536 if (need_vpa_update) {
1537 spin_unlock(&vc->lock);
913d3ff9
PM		 1538 for (i = 0; i < need_vpa_update; ++i)
1539 kvmppc_update_vpas(vcpus_to_update[i]);
081f323b
PM		 1540 spin_lock(&vc->lock);
1541 }
de56a948 1542
7b444c67 1543 /*
3102f784
ME		 1544 * Make sure we are running on primary threads, and that secondary
1545 * threads are offline. Also check if the number of threads in this
1546 * guest are greater than the current system threads per guest.
7b444c67 1547 */
3102f784
ME		 1548 if ((threads_per_core > 1) &&
1549 ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
7b444c67
PM		 1550 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1551 vcpu->arch.ret = -EBUSY;
1552 goto out;
1553 }
1554
3102f784 1555
371fefd6 1556 vc->pcpu = smp_processor_id();
2e25aa5f 1557 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
371fefd6 1558 kvmppc_start_thread(vcpu);
0456ec4f 1559 kvmppc_create_dtl_entry(vcpu, vc);
2e25aa5f 1560 }
371fefd6 1561
e0b7ec05
PM		 1562 /* Set this explicitly in case thread 0 doesn't have a vcpu */
1563 get_paca()->kvm_hstate.kvm_vcore = vc;
1564 get_paca()->kvm_hstate.ptid = 0;
1565
2f12f034 1566 vc->vcore_state = VCORE_RUNNING;
19ccb76a 1567 preempt_disable();
371fefd6 1568 spin_unlock(&vc->lock);
de56a948 1569
371fefd6 1570 kvm_guest_enter();
2c9097e4 1571
e0b7ec05 1572 srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2c9097e4 1573
e0b7ec05 1574 __kvmppc_vcore_entry();
de56a948 1575
371fefd6 1576 spin_lock(&vc->lock);
19ccb76a
PM		 1577 /* disable sending of IPIs on virtual external irqs */
1578 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1579 vcpu->cpu = -1;
1580 /* wait for secondary threads to finish writing their state to memory */
371fefd6
PM		 1581 if (vc->nap_count < vc->n_woken)
1582 kvmppc_wait_for_nap(vc);
3102f784 1583 for (i = 0; i < threads_per_subcore; ++i)
2f12f034 1584 kvmppc_release_hwthread(vc->pcpu + i);
371fefd6 1585 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
19ccb76a 1586 vc->vcore_state = VCORE_EXITING;
371fefd6
PM		 1587 spin_unlock(&vc->lock);
1588
e0b7ec05 1589 srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2c9097e4 1590
371fefd6
PM		 1591 /* make sure updates to secondary vcpu structs are visible now */
1592 smp_mb();
de56a948
PM		 1593 kvm_guest_exit();
1594
1595 preempt_enable();
c08ac06a 1596 cond_resched();
de56a948 1597
913d3ff9 1598 spin_lock(&vc->lock);
de56a948 1599 now = get_tb();
371fefd6
PM		 1600 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1601 /* cancel pending dec exception if dec is positive */
1602 if (now < vcpu->arch.dec_expires &&
1603 kvmppc_core_pending_dec(vcpu))
1604 kvmppc_core_dequeue_dec(vcpu);
19ccb76a
PM		 1605
1606 ret = RESUME_GUEST;
1607 if (vcpu->arch.trap)
3a167bea
AK		 1608 ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
1609 vcpu->arch.run_task);
19ccb76a 1610
371fefd6
PM		 1611 vcpu->arch.ret = ret;
1612 vcpu->arch.trap = 0;
19ccb76a
PM		 1613
1614 if (vcpu->arch.ceded) {
e59d24e6 1615 if (!is_kvmppc_resume_guest(ret))
19ccb76a
PM		 1616 kvmppc_end_cede(vcpu);
1617 else
1618 kvmppc_set_timer(vcpu);
1619 }
371fefd6 1620 }
de56a948
PM		 1621
1622 out:
19ccb76a 1623 vc->vcore_state = VCORE_INACTIVE;
371fefd6
PM		 1624 list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1625 arch.run_list) {
e59d24e6 1626 if (!is_kvmppc_resume_guest(vcpu->arch.ret)) {
371fefd6
PM		 1627 kvmppc_remove_runnable(vc, vcpu);
1628 wake_up(&vcpu->arch.cpu_run);
1629 }
1630 }
371fefd6
PM		 1631}
1632
19ccb76a
PM		 1633/*
1634 * Wait for some other vcpu thread to execute us, and
1635 * wake us up when we need to handle something in the host.
1636 */
1637static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
371fefd6 1638{
371fefd6
PM		 1639 DEFINE_WAIT(wait);
1640
19ccb76a
PM		 1641 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
1642 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
1643 schedule();
1644 finish_wait(&vcpu->arch.cpu_run, &wait);
1645}
1646
1647/*
1648 * All the vcpus in this vcore are idle, so wait for a decrementer
1649 * or external interrupt to one of the vcpus. vc->lock is held.
1650 */
1651static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
1652{
1653 DEFINE_WAIT(wait);
19ccb76a
PM		 1654
1655 prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
1656 vc->vcore_state = VCORE_SLEEPING;
1657 spin_unlock(&vc->lock);
913d3ff9 1658 schedule();
19ccb76a
PM		 1659 finish_wait(&vc->wq, &wait);
1660 spin_lock(&vc->lock);
1661 vc->vcore_state = VCORE_INACTIVE;
1662}
371fefd6 1663
19ccb76a
PM		 1664static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1665{
1666 int n_ceded;
19ccb76a
PM		 1667 struct kvmppc_vcore *vc;
1668 struct kvm_vcpu *v, *vn;
9e368f29 1669
371fefd6
PM		 1670 kvm_run->exit_reason = 0;
1671 vcpu->arch.ret = RESUME_GUEST;
1672 vcpu->arch.trap = 0;
2f12f034 1673 kvmppc_update_vpas(vcpu);
371fefd6 1674
371fefd6
PM		 1675 /*
1676 * Synchronize with other threads in this virtual core
1677 */
1678 vc = vcpu->arch.vcore;
1679 spin_lock(&vc->lock);
19ccb76a 1680 vcpu->arch.ceded = 0;
371fefd6
PM		 1681 vcpu->arch.run_task = current;
1682 vcpu->arch.kvm_run = kvm_run;
c7b67670 1683 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
19ccb76a 1684 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
c7b67670 1685 vcpu->arch.busy_preempt = TB_NIL;
371fefd6
PM		 1686 list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
1687 ++vc->n_runnable;
1688
19ccb76a
PM		 1689 /*
1690 * This happens the first time this is called for a vcpu.
1691 * If the vcore is already running, we may be able to start
1692 * this thread straight away and have it join in.
1693 */
8455d79e 1694 if (!signal_pending(current)) {
19ccb76a
PM		 1695 if (vc->vcore_state == VCORE_RUNNING &&
1696 VCORE_EXIT_COUNT(vc) == 0) {
2f12f034 1697 kvmppc_create_dtl_entry(vcpu, vc);
19ccb76a 1698 kvmppc_start_thread(vcpu);
8455d79e
PM		 1699 } else if (vc->vcore_state == VCORE_SLEEPING) {
1700 wake_up(&vc->wq);
371fefd6
PM		 1701 }
1702
8455d79e 1703 }
371fefd6 1704
19ccb76a
PM		 1705 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1706 !signal_pending(current)) {
8455d79e 1707 if (vc->vcore_state != VCORE_INACTIVE) {
19ccb76a
PM		 1708 spin_unlock(&vc->lock);
1709 kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
1710 spin_lock(&vc->lock);
1711 continue;
1712 }
19ccb76a
PM		 1713 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
1714 arch.run_list) {
7e28e60e 1715 kvmppc_core_prepare_to_enter(v);
19ccb76a
PM		 1716 if (signal_pending(v->arch.run_task)) {
1717 kvmppc_remove_runnable(vc, v);
1718 v->stat.signal_exits++;
1719 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
1720 v->arch.ret = -EINTR;
1721 wake_up(&v->arch.cpu_run);
1722 }
1723 }
8455d79e
PM		 1724 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1725 break;
1726 vc->runner = vcpu;
1727 n_ceded = 0;
4619ac88 1728 list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
8455d79e
PM		 1729 if (!v->arch.pending_exceptions)
1730 n_ceded += v->arch.ceded;
4619ac88
PM		 1731 else
1732 v->arch.ceded = 0;
1733 }
8455d79e
PM		 1734 if (n_ceded == vc->n_runnable)
1735 kvmppc_vcore_blocked(vc);
1736 else
1737 kvmppc_run_core(vc);
0456ec4f 1738 vc->runner = NULL;
19ccb76a 1739 }
371fefd6 1740
8455d79e
PM		 1741 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1742 (vc->vcore_state == VCORE_RUNNING ||
1743 vc->vcore_state == VCORE_EXITING)) {
1744 spin_unlock(&vc->lock);
1745 kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
1746 spin_lock(&vc->lock);
1747 }
1748
1749 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
1750 kvmppc_remove_runnable(vc, vcpu);
1751 vcpu->stat.signal_exits++;
1752 kvm_run->exit_reason = KVM_EXIT_INTR;
1753 vcpu->arch.ret = -EINTR;
1754 }
1755
1756 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
1757 /* Wake up some vcpu to run the core */
1758 v = list_first_entry(&vc->runnable_threads,
1759 struct kvm_vcpu, arch.run_list);
1760 wake_up(&v->arch.cpu_run);
371fefd6
PM		 1761 }
1762
371fefd6 1763 spin_unlock(&vc->lock);
371fefd6 1764 return vcpu->arch.ret;
de56a948
PM		 1765}
1766
3a167bea 1767static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
a8606e20
PM		 1768{
1769 int r;
913d3ff9 1770 int srcu_idx;
a8606e20 1771
af8f38b3
AG		 1772 if (!vcpu->arch.sane) {
1773 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1774 return -EINVAL;
1775 }
1776
25051b5a
SW		 1777 kvmppc_core_prepare_to_enter(vcpu);
1778
19ccb76a
PM		 1779 /* No need to go into the guest when all we'll do is come back out */
1780 if (signal_pending(current)) {
1781 run->exit_reason = KVM_EXIT_INTR;
1782 return -EINTR;
1783 }
1784
32fad281
PM		 1785 atomic_inc(&vcpu->kvm->arch.vcpus_running);
1786 /* Order vcpus_running vs. rma_setup_done, see kvmppc_alloc_reset_hpt */
1787 smp_mb();
1788
1789 /* On the first time here, set up HTAB and VRMA or RMA */
c77162de 1790 if (!vcpu->kvm->arch.rma_setup_done) {
32fad281 1791 r = kvmppc_hv_setup_htab_rma(vcpu);
c77162de 1792 if (r)
32fad281 1793 goto out;
c77162de 1794 }
19ccb76a
PM		 1795
1796 flush_fp_to_thread(current);
1797 flush_altivec_to_thread(current);
1798 flush_vsx_to_thread(current);
1799 vcpu->arch.wqp = &vcpu->arch.vcore->wq;
342d3db7 1800 vcpu->arch.pgdir = current->mm->pgd;
c7b67670 1801 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
19ccb76a 1802
a8606e20
PM		 1803 do {
1804 r = kvmppc_run_vcpu(run, vcpu);
1805
1806 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
1807 !(vcpu->arch.shregs.msr & MSR_PR)) {
1808 r = kvmppc_pseries_do_hcall(vcpu);
7e28e60e 1809 kvmppc_core_prepare_to_enter(vcpu);
913d3ff9
PM		 1810 } else if (r == RESUME_PAGE_FAULT) {
1811 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1812 r = kvmppc_book3s_hv_page_fault(run, vcpu,
1813 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1814 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
a8606e20 1815 }
e59d24e6 1816 } while (is_kvmppc_resume_guest(r));
32fad281
PM		 1817
1818 out:
c7b67670 1819 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
32fad281 1820 atomic_dec(&vcpu->kvm->arch.vcpus_running);
a8606e20
PM		 1821 return r;
1822}
1823
54738c09 1824
aa04b4cc 1825/* Work out RMLS (real mode limit selector) field value for a given RMA size.
9e368f29 1826 Assumes POWER7 or PPC970. */
aa04b4cc
PM		 1827static inline int lpcr_rmls(unsigned long rma_size)
1828{
1829 switch (rma_size) {
1830 case 32ul << 20: /* 32 MB */
9e368f29
PM		 1831 if (cpu_has_feature(CPU_FTR_ARCH_206))
1832 return 8; /* only supported on POWER7 */
1833 return -1;
aa04b4cc
PM		 1834 case 64ul << 20: /* 64 MB */
1835 return 3;
1836 case 128ul << 20: /* 128 MB */
1837 return 7;
1838 case 256ul << 20: /* 256 MB */
1839 return 4;
1840 case 1ul << 30: /* 1 GB */
1841 return 2;
1842 case 16ul << 30: /* 16 GB */
1843 return 1;
1844 case 256ul << 30: /* 256 GB */
1845 return 0;
1846 default:
1847 return -1;
1848 }
1849}
1850
1851static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1852{
aa04b4cc 1853 struct page *page;
6c45b810 1854 struct kvm_rma_info *ri = vma->vm_file->private_data;
aa04b4cc 1855
6c45b810 1856 if (vmf->pgoff >= kvm_rma_pages)
aa04b4cc
PM		 1857 return VM_FAULT_SIGBUS;
1858
1859 page = pfn_to_page(ri->base_pfn + vmf->pgoff);
1860 get_page(page);
1861 vmf->page = page;
1862 return 0;
1863}
1864
1865static const struct vm_operations_struct kvm_rma_vm_ops = {
1866 .fault = kvm_rma_fault,
1867};
1868
1869static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma)
1870{
314e51b9 1871 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
aa04b4cc
PM		 1872 vma->vm_ops = &kvm_rma_vm_ops;
1873 return 0;
1874}
1875
1876static int kvm_rma_release(struct inode *inode, struct file *filp)
1877{
6c45b810 1878 struct kvm_rma_info *ri = filp->private_data;
aa04b4cc
PM		 1879
1880 kvm_release_rma(ri);
1881 return 0;
1882}
1883
75ef9de1 1884static const struct file_operations kvm_rma_fops = {
aa04b4cc
PM		 1885 .mmap = kvm_rma_mmap,
1886 .release = kvm_rma_release,
1887};
1888
3a167bea
AK		 1889static long kvm_vm_ioctl_allocate_rma(struct kvm *kvm,
1890 struct kvm_allocate_rma *ret)
aa04b4cc 1891{
aa04b4cc 1892 long fd;
6c45b810
AK		 1893 struct kvm_rma_info *ri;
1894 /*
1895 * Only do this on PPC970 in HV mode
1896 */
1897 if (!cpu_has_feature(CPU_FTR_HVMODE) ||
1898 !cpu_has_feature(CPU_FTR_ARCH_201))
1899 return -EINVAL;
1900
1901 if (!kvm_rma_pages)
1902 return -EINVAL;
aa04b4cc
PM		 1903
1904 ri = kvm_alloc_rma();
1905 if (!ri)
1906 return -ENOMEM;
1907
2f84d5ea 1908 fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR | O_CLOEXEC);
aa04b4cc
PM		 1909 if (fd < 0)
1910 kvm_release_rma(ri);
1911
6c45b810 1912 ret->rma_size = kvm_rma_pages << PAGE_SHIFT;
aa04b4cc
PM		 1913 return fd;
1914}
1915
5b74716e
BH		 1916static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
1917 int linux_psize)
1918{
1919 struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
1920
1921 if (!def->shift)
1922 return;
1923 (*sps)->page_shift = def->shift;
1924 (*sps)->slb_enc = def->sllp;
1925 (*sps)->enc[0].page_shift = def->shift;
b1022fbd
AK		 1926 /*
1927 * Only return base page encoding. We don't want to return
1928 * all the supporting pte_enc, because our H_ENTER doesn't
1929 * support MPSS yet. Once they do, we can start passing all
1930 * support pte_enc here
1931 */
1932 (*sps)->enc[0].pte_enc = def->penc[linux_psize];
1f365bb0
AK		 1933 /*
1934 * Add 16MB MPSS support if host supports it
1935 */
1936 if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
1937 (*sps)->enc[1].page_shift = 24;
1938 (*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
1939 }
5b74716e
BH		 1940 (*sps)++;
1941}
1942
3a167bea
AK		 1943static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
1944 struct kvm_ppc_smmu_info *info)
5b74716e
BH		 1945{
1946 struct kvm_ppc_one_seg_page_size *sps;
1947
1948 info->flags = KVM_PPC_PAGE_SIZES_REAL;
1949 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1950 info->flags |= KVM_PPC_1T_SEGMENTS;
1951 info->slb_size = mmu_slb_size;
1952
1953 /* We only support these sizes for now, and no muti-size segments */
1954 sps = &info->sps[0];
1955 kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
1956 kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
1957 kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
1958
1959 return 0;
1960}
1961
82ed3616
PM		 1962/*
1963 * Get (and clear) the dirty memory log for a memory slot.
1964 */
3a167bea
AK		 1965static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
1966 struct kvm_dirty_log *log)
82ed3616
PM		 1967{
1968 struct kvm_memory_slot *memslot;
1969 int r;
1970 unsigned long n;
1971
1972 mutex_lock(&kvm->slots_lock);
1973
1974 r = -EINVAL;
bbacc0c1 1975 if (log->slot >= KVM_USER_MEM_SLOTS)
82ed3616
PM		 1976 goto out;
1977
1978 memslot = id_to_memslot(kvm->memslots, log->slot);
1979 r = -ENOENT;
1980 if (!memslot->dirty_bitmap)
1981 goto out;
1982
1983 n = kvm_dirty_bitmap_bytes(memslot);
1984 memset(memslot->dirty_bitmap, 0, n);
1985
dfe49dbd 1986 r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
82ed3616
PM		 1987 if (r)
1988 goto out;
1989
1990 r = -EFAULT;
1991 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1992 goto out;
1993
1994 r = 0;
1995out:
1996 mutex_unlock(&kvm->slots_lock);
1997 return r;
1998}
1999
a66b48c3 2000static void unpin_slot(struct kvm_memory_slot *memslot)
de56a948 2001{
a66b48c3
PM		 2002 unsigned long *physp;
2003 unsigned long j, npages, pfn;
2004 struct page *page;
aa04b4cc 2005
a66b48c3
PM		 2006 physp = memslot->arch.slot_phys;
2007 npages = memslot->npages;
2008 if (!physp)
2009 return;
2010 for (j = 0; j < npages; j++) {
2011 if (!(physp[j] & KVMPPC_GOT_PAGE))
2012 continue;
2013 pfn = physp[j] >> PAGE_SHIFT;
2014 page = pfn_to_page(pfn);
2015 SetPageDirty(page);
2016 put_page(page);
2017 }
2018}
2019
3a167bea
AK		 2020static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
2021 struct kvm_memory_slot *dont)
a66b48c3
PM		 2022{
2023 if (!dont || free->arch.rmap != dont->arch.rmap) {
2024 vfree(free->arch.rmap);
2025 free->arch.rmap = NULL;
b2b2f165 2026 }
a66b48c3
PM		 2027 if (!dont || free->arch.slot_phys != dont->arch.slot_phys) {
2028 unpin_slot(free);
2029 vfree(free->arch.slot_phys);
2030 free->arch.slot_phys = NULL;
2031 }
2032}
2033
3a167bea
AK		 2034static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
2035 unsigned long npages)
a66b48c3
PM		 2036{
2037 slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
2038 if (!slot->arch.rmap)
2039 return -ENOMEM;
2040 slot->arch.slot_phys = NULL;
aa04b4cc 2041
c77162de
PM		 2042 return 0;
2043}
aa04b4cc 2044
3a167bea
AK		 2045static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
2046 struct kvm_memory_slot *memslot,
2047 struct kvm_userspace_memory_region *mem)
c77162de 2048{
a66b48c3 2049 unsigned long *phys;
c77162de 2050
a66b48c3
PM		 2051 /* Allocate a slot_phys array if needed */
2052 phys = memslot->arch.slot_phys;
2053 if (!kvm->arch.using_mmu_notifiers && !phys && memslot->npages) {
2054 phys = vzalloc(memslot->npages * sizeof(unsigned long));
2055 if (!phys)
2056 return -ENOMEM;
2057 memslot->arch.slot_phys = phys;
aa04b4cc 2058 }
a66b48c3
PM		 2059
2060 return 0;
c77162de
PM		 2061}
2062
3a167bea
AK		 2063static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
2064 struct kvm_userspace_memory_region *mem,
2065 const struct kvm_memory_slot *old)
c77162de 2066{
dfe49dbd
PM		 2067 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
2068 struct kvm_memory_slot *memslot;
2069
8482644a 2070 if (npages && old->npages) {
dfe49dbd
PM		 2071 /*
2072 * If modifying a memslot, reset all the rmap dirty bits.
2073 * If this is a new memslot, we don't need to do anything
2074 * since the rmap array starts out as all zeroes,
2075 * i.e. no pages are dirty.
2076 */
2077 memslot = id_to_memslot(kvm->memslots, mem->slot);
2078 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
2079 }
c77162de
PM		 2080}
2081
a0144e2a
PM		 2082/*
2083 * Update LPCR values in kvm->arch and in vcores.
2084 * Caller must hold kvm->lock.
2085 */
2086void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
2087{
2088 long int i;
2089 u32 cores_done = 0;
2090
2091 if ((kvm->arch.lpcr & mask) == lpcr)
2092 return;
2093
2094 kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
2095
2096 for (i = 0; i < KVM_MAX_VCORES; ++i) {
2097 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
2098 if (!vc)
2099 continue;
2100 spin_lock(&vc->lock);
2101 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
2102 spin_unlock(&vc->lock);
2103 if (++cores_done >= kvm->arch.online_vcores)
2104 break;
2105 }
2106}
2107
3a167bea
AK		 2108static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
2109{
2110 return;
2111}
2112
32fad281 2113static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
c77162de
PM		 2114{
2115 int err = 0;
2116 struct kvm *kvm = vcpu->kvm;
6c45b810 2117 struct kvm_rma_info *ri = NULL;
c77162de
PM		 2118 unsigned long hva;
2119 struct kvm_memory_slot *memslot;
2120 struct vm_area_struct *vma;
a0144e2a
PM		 2121 unsigned long lpcr = 0, senc;
2122 unsigned long lpcr_mask = 0;
c77162de
PM		 2123 unsigned long psize, porder;
2124 unsigned long rma_size;
2125 unsigned long rmls;
2126 unsigned long *physp;
da9d1d7f 2127 unsigned long i, npages;
2c9097e4 2128 int srcu_idx;
c77162de
PM		 2129
2130 mutex_lock(&kvm->lock);
2131 if (kvm->arch.rma_setup_done)
2132 goto out; /* another vcpu beat us to it */
aa04b4cc 2133
32fad281
PM		 2134 /* Allocate hashed page table (if not done already) and reset it */
2135 if (!kvm->arch.hpt_virt) {
2136 err = kvmppc_alloc_hpt(kvm, NULL);
2137 if (err) {
2138 pr_err("KVM: Couldn't alloc HPT\n");
2139 goto out;
2140 }
2141 }
2142
c77162de 2143 /* Look up the memslot for guest physical address 0 */
2c9097e4 2144 srcu_idx = srcu_read_lock(&kvm->srcu);
c77162de 2145 memslot = gfn_to_memslot(kvm, 0);
aa04b4cc 2146
c77162de
PM		 2147 /* We must have some memory at 0 by now */
2148 err = -EINVAL;
2149 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
2c9097e4 2150 goto out_srcu;
c77162de
PM		 2151
2152 /* Look up the VMA for the start of this memory slot */
2153 hva = memslot->userspace_addr;
2154 down_read(&current->mm->mmap_sem);
2155 vma = find_vma(current->mm, hva);
2156 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
2157 goto up_out;
2158
2159 psize = vma_kernel_pagesize(vma);
da9d1d7f 2160 porder = __ilog2(psize);
c77162de
PM		 2161
2162 /* Is this one of our preallocated RMAs? */
2163 if (vma->vm_file && vma->vm_file->f_op == &kvm_rma_fops &&
2164 hva == vma->vm_start)
2165 ri = vma->vm_file->private_data;
2166
2167 up_read(&current->mm->mmap_sem);
2168
2169 if (!ri) {
2170 /* On POWER7, use VRMA; on PPC970, give up */
2171 err = -EPERM;
2172 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2173 pr_err("KVM: CPU requires an RMO\n");
2c9097e4 2174 goto out_srcu;
c77162de
PM		 2175 }
2176
da9d1d7f
PM		 2177 /* We can handle 4k, 64k or 16M pages in the VRMA */
2178 err = -EINVAL;
2179 if (!(psize == 0x1000 || psize == 0x10000 ||
2180 psize == 0x1000000))
2c9097e4 2181 goto out_srcu;
da9d1d7f 2182
c77162de 2183 /* Update VRMASD field in the LPCR */
da9d1d7f 2184 senc = slb_pgsize_encoding(psize);
697d3899
PM		 2185 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
2186 (VRMA_VSID << SLB_VSID_SHIFT_1T);
a0144e2a
PM		 2187 lpcr_mask = LPCR_VRMASD;
2188 /* the -4 is to account for senc values starting at 0x10 */
2189 lpcr = senc << (LPCR_VRMASD_SH - 4);
c77162de
PM		 2190
2191 /* Create HPTEs in the hash page table for the VRMA */
da9d1d7f 2192 kvmppc_map_vrma(vcpu, memslot, porder);
c77162de
PM		 2193
2194 } else {
2195 /* Set up to use an RMO region */
6c45b810 2196 rma_size = kvm_rma_pages;
c77162de
PM		 2197 if (rma_size > memslot->npages)
2198 rma_size = memslot->npages;
2199 rma_size <<= PAGE_SHIFT;
aa04b4cc 2200 rmls = lpcr_rmls(rma_size);
c77162de 2201 err = -EINVAL;
5d226ae5 2202 if ((long)rmls < 0) {
c77162de 2203 pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size);
2c9097e4 2204 goto out_srcu;
aa04b4cc
PM		 2205 }
2206 atomic_inc(&ri->use_count);
2207 kvm->arch.rma = ri;
9e368f29
PM		 2208
2209 /* Update LPCR and RMOR */
9e368f29
PM		 2210 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2211 /* PPC970; insert RMLS value (split field) in HID4 */
a0144e2a
PM		 2212 lpcr_mask = (1ul << HID4_RMLS0_SH) |
2213 (3ul << HID4_RMLS2_SH) | HID4_RMOR;
2214 lpcr = ((rmls >> 2) << HID4_RMLS0_SH) |
9e368f29
PM		 2215 ((rmls & 3) << HID4_RMLS2_SH);
2216 /* RMOR is also in HID4 */
2217 lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff)
2218 << HID4_RMOR_SH;
2219 } else {
2220 /* POWER7 */
a0144e2a
PM		 2221 lpcr_mask = LPCR_VPM0 | LPCR_VRMA_L | LPCR_RMLS;
2222 lpcr = rmls << LPCR_RMLS_SH;
6c45b810 2223 kvm->arch.rmor = ri->base_pfn << PAGE_SHIFT;
9e368f29 2224 }
c77162de 2225 pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n",
aa04b4cc 2226 ri->base_pfn << PAGE_SHIFT, rma_size, lpcr);
aa04b4cc 2227
c77162de 2228 /* Initialize phys addrs of pages in RMO */
6c45b810 2229 npages = kvm_rma_pages;
da9d1d7f 2230 porder = __ilog2(npages);
a66b48c3
PM		 2231 physp = memslot->arch.slot_phys;
2232 if (physp) {
2233 if (npages > memslot->npages)
2234 npages = memslot->npages;
2235 spin_lock(&kvm->arch.slot_phys_lock);
2236 for (i = 0; i < npages; ++i)
2237 physp[i] = ((ri->base_pfn + i) << PAGE_SHIFT) +
2238 porder;
2239 spin_unlock(&kvm->arch.slot_phys_lock);
2240 }
aa04b4cc
PM		 2241 }
2242
a0144e2a
PM		 2243 kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
2244
c77162de
PM		 2245 /* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */
2246 smp_wmb();
2247 kvm->arch.rma_setup_done = 1;
2248 err = 0;
2c9097e4
PM		 2249 out_srcu:
2250 srcu_read_unlock(&kvm->srcu, srcu_idx);
c77162de
PM		 2251 out:
2252 mutex_unlock(&kvm->lock);
2253 return err;
b2b2f165 2254
c77162de
PM		 2255 up_out:
2256 up_read(&current->mm->mmap_sem);
505d6421 2257 goto out_srcu;
de56a948
PM		 2258}
2259
3a167bea 2260static int kvmppc_core_init_vm_hv(struct kvm *kvm)
de56a948 2261{
32fad281 2262 unsigned long lpcr, lpid;
de56a948 2263
32fad281
PM		 2264 /* Allocate the guest's logical partition ID */
2265
2266 lpid = kvmppc_alloc_lpid();
5d226ae5 2267 if ((long)lpid < 0)
32fad281
PM		 2268 return -ENOMEM;
2269 kvm->arch.lpid = lpid;
de56a948 2270
1b400ba0
PM		 2271 /*
2272 * Since we don't flush the TLB when tearing down a VM,
2273 * and this lpid might have previously been used,
2274 * make sure we flush on each core before running the new VM.
2275 */
2276 cpumask_setall(&kvm->arch.need_tlb_flush);
2277
699a0ea0
PM		 2278 /* Start out with the default set of hcalls enabled */
2279 memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
2280 sizeof(kvm->arch.enabled_hcalls));
2281
aa04b4cc 2282 kvm->arch.rma = NULL;
aa04b4cc 2283
9e368f29 2284 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
aa04b4cc 2285
9e368f29
PM		 2286 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2287 /* PPC970; HID4 is effectively the LPCR */
9e368f29
PM		 2288 kvm->arch.host_lpid = 0;
2289 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4);
2290 lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH));
2291 lpcr |= ((lpid >> 4) << HID4_LPID1_SH) |
2292 ((lpid & 0xf) << HID4_LPID5_SH);
2293 } else {
2294 /* POWER7; init LPCR for virtual RMA mode */
2295 kvm->arch.host_lpid = mfspr(SPRN_LPID);
2296 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
2297 lpcr &= LPCR_PECE | LPCR_LPES;
2298 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
697d3899
PM		 2299 LPCR_VPM0 | LPCR_VPM1;
2300 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
2301 (VRMA_VSID << SLB_VSID_SHIFT_1T);
e0622bd9
PM		 2302 /* On POWER8 turn on online bit to enable PURR/SPURR */
2303 if (cpu_has_feature(CPU_FTR_ARCH_207S))
2304 lpcr |= LPCR_ONL;
9e368f29
PM		 2305 }
2306 kvm->arch.lpcr = lpcr;
aa04b4cc 2307
342d3db7 2308 kvm->arch.using_mmu_notifiers = !!cpu_has_feature(CPU_FTR_ARCH_206);
c77162de 2309 spin_lock_init(&kvm->arch.slot_phys_lock);
512691d4
PM		 2310
2311 /*
441c19c8
ME		 2312 * Track that we now have a HV mode VM active. This blocks secondary
2313 * CPU threads from coming online.
512691d4 2314 */
441c19c8 2315 kvm_hv_vm_activated();
512691d4 2316
54738c09 2317 return 0;
de56a948
PM		 2318}
2319
f1378b1c
PM		 2320static void kvmppc_free_vcores(struct kvm *kvm)
2321{
2322 long int i;
2323
2324 for (i = 0; i < KVM_MAX_VCORES; ++i)
2325 kfree(kvm->arch.vcores[i]);
2326 kvm->arch.online_vcores = 0;
2327}
2328
3a167bea 2329static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
de56a948 2330{
441c19c8 2331 kvm_hv_vm_deactivated();
512691d4 2332
f1378b1c 2333 kvmppc_free_vcores(kvm);
aa04b4cc
PM		 2334 if (kvm->arch.rma) {
2335 kvm_release_rma(kvm->arch.rma);
2336 kvm->arch.rma = NULL;
2337 }
2338
de56a948
PM		 2339 kvmppc_free_hpt(kvm);
2340}
2341
3a167bea
AK		 2342/* We don't need to emulate any privileged instructions or dcbz */
2343static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
2344 unsigned int inst, int *advance)
de56a948 2345{
3a167bea 2346 return EMULATE_FAIL;
de56a948
PM		 2347}
2348
3a167bea
AK		 2349static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
2350 ulong spr_val)
de56a948
PM		 2351{
2352 return EMULATE_FAIL;
2353}
2354
3a167bea
AK		 2355static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
2356 ulong *spr_val)
de56a948
PM		 2357{
2358 return EMULATE_FAIL;
2359}
2360
3a167bea 2361static int kvmppc_core_check_processor_compat_hv(void)
de56a948 2362{
3a167bea
AK		 2363 if (!cpu_has_feature(CPU_FTR_HVMODE))
2364 return -EIO;
2365 return 0;
de56a948
PM		 2366}
2367
3a167bea
AK		 2368static long kvm_arch_vm_ioctl_hv(struct file *filp,
2369 unsigned int ioctl, unsigned long arg)
2370{
2371 struct kvm *kvm __maybe_unused = filp->private_data;
2372 void __user *argp = (void __user *)arg;
2373 long r;
2374
2375 switch (ioctl) {
2376
2377 case KVM_ALLOCATE_RMA: {
2378 struct kvm_allocate_rma rma;
2379 struct kvm *kvm = filp->private_data;
2380
2381 r = kvm_vm_ioctl_allocate_rma(kvm, &rma);
2382 if (r >= 0 && copy_to_user(argp, &rma, sizeof(rma)))
2383 r = -EFAULT;
2384 break;
2385 }
2386
2387 case KVM_PPC_ALLOCATE_HTAB: {
2388 u32 htab_order;
2389
2390 r = -EFAULT;
2391 if (get_user(htab_order, (u32 __user *)argp))
2392 break;
2393 r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
2394 if (r)
2395 break;
2396 r = -EFAULT;
2397 if (put_user(htab_order, (u32 __user *)argp))
2398 break;
2399 r = 0;
2400 break;
2401 }
2402
2403 case KVM_PPC_GET_HTAB_FD: {
2404 struct kvm_get_htab_fd ghf;
2405
2406 r = -EFAULT;
2407 if (copy_from_user(&ghf, argp, sizeof(ghf)))
2408 break;
2409 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
2410 break;
2411 }
2412
2413 default:
2414 r = -ENOTTY;
2415 }
2416
2417 return r;
2418}
2419
699a0ea0
PM		 2420/*
2421 * List of hcall numbers to enable by default.
2422 * For compatibility with old userspace, we enable by default
2423 * all hcalls that were implemented before the hcall-enabling
2424 * facility was added. Note this list should not include H_RTAS.
2425 */
2426static unsigned int default_hcall_list[] = {
2427 H_REMOVE,
2428 H_ENTER,
2429 H_READ,
2430 H_PROTECT,
2431 H_BULK_REMOVE,
2432 H_GET_TCE,
2433 H_PUT_TCE,
2434 H_SET_DABR,
2435 H_SET_XDABR,
2436 H_CEDE,
2437 H_PROD,
2438 H_CONFER,
2439 H_REGISTER_VPA,
2440#ifdef CONFIG_KVM_XICS
2441 H_EOI,
2442 H_CPPR,
2443 H_IPI,
2444 H_IPOLL,
2445 H_XIRR,
2446 H_XIRR_X,
2447#endif
2448 0
2449};
2450
2451static void init_default_hcalls(void)
2452{
2453 int i;
2454
2455 for (i = 0; default_hcall_list[i]; ++i)
2456 __set_bit(default_hcall_list[i] / 4, default_enabled_hcalls);
2457}
2458
cbbc58d4 2459static struct kvmppc_ops kvm_ops_hv = {
3a167bea
AK		 2460 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
2461 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
2462 .get_one_reg = kvmppc_get_one_reg_hv,
2463 .set_one_reg = kvmppc_set_one_reg_hv,
2464 .vcpu_load = kvmppc_core_vcpu_load_hv,
2465 .vcpu_put = kvmppc_core_vcpu_put_hv,
2466 .set_msr = kvmppc_set_msr_hv,
2467 .vcpu_run = kvmppc_vcpu_run_hv,
2468 .vcpu_create = kvmppc_core_vcpu_create_hv,
2469 .vcpu_free = kvmppc_core_vcpu_free_hv,
2470 .check_requests = kvmppc_core_check_requests_hv,
2471 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv,
2472 .flush_memslot = kvmppc_core_flush_memslot_hv,
2473 .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
2474 .commit_memory_region = kvmppc_core_commit_memory_region_hv,
2475 .unmap_hva = kvm_unmap_hva_hv,
2476 .unmap_hva_range = kvm_unmap_hva_range_hv,
2477 .age_hva = kvm_age_hva_hv,
2478 .test_age_hva = kvm_test_age_hva_hv,
2479 .set_spte_hva = kvm_set_spte_hva_hv,
2480 .mmu_destroy = kvmppc_mmu_destroy_hv,
2481 .free_memslot = kvmppc_core_free_memslot_hv,
2482 .create_memslot = kvmppc_core_create_memslot_hv,
2483 .init_vm = kvmppc_core_init_vm_hv,
2484 .destroy_vm = kvmppc_core_destroy_vm_hv,
3a167bea
AK		 2485 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
2486 .emulate_op = kvmppc_core_emulate_op_hv,
2487 .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
2488 .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
2489 .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
2490 .arch_vm_ioctl = kvm_arch_vm_ioctl_hv,
2491};
2492
2493static int kvmppc_book3s_init_hv(void)
de56a948
PM		 2494{
2495 int r;
cbbc58d4
AK		 2496 /*
2497 * FIXME!! Do we need to check on all cpus ?
2498 */
2499 r = kvmppc_core_check_processor_compat_hv();
2500 if (r < 0)
739e2425 2501 return -ENODEV;
de56a948 2502
cbbc58d4
AK		 2503 kvm_ops_hv.owner = THIS_MODULE;
2504 kvmppc_hv_ops = &kvm_ops_hv;
de56a948 2505
699a0ea0
PM		 2506 init_default_hcalls();
2507
cbbc58d4 2508 r = kvmppc_mmu_hv_init();
de56a948
PM		 2509 return r;
2510}
2511
3a167bea 2512static void kvmppc_book3s_exit_hv(void)
de56a948 2513{
cbbc58d4 2514 kvmppc_hv_ops = NULL;
de56a948
PM		 2515}
2516
3a167bea
AK		 2517module_init(kvmppc_book3s_init_hv);
2518module_exit(kvmppc_book3s_exit_hv);
2ba9f0d8 2519MODULE_LICENSE("GPL");
398a76c6
AG		 2520MODULE_ALIAS_MISCDEV(KVM_MINOR);
2521MODULE_ALIAS("devname:kvm");
Linux kernel - Deliverables
This page took 0.357771 seconds and 5 git commands to generate.