projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Split out struct kvmppc_vcore creation to separate function
[linux-2.6-block.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;
02407552 275 vpa->yield_count = cpu_to_be32(1);
a8606e20
PM		 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 {
02407552
AG		 298 __be16 hword;
299 __be32 word;
2e25aa5f
PM		 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 337 if (subfunc == H_VPA_REG_VPA)
02407552 338 len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
a8606e20 339 else
02407552 340 len = be32_to_cpu(((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;
02407552
AG		 545 dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
546 dt->timebase = cpu_to_be64(now + vc->tb_offset);
547 dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
548 dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
549 dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
0456ec4f
PM		 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();
02407552 556 vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
c35635ef 557 vcpu->arch.dtl.dirty = true;
0456ec4f
PM		 558}
559
9642382e
MN		 560static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
561{
562 if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
563 return true;
564 if ((!vcpu->arch.vcore->arch_compat) &&
565 cpu_has_feature(CPU_FTR_ARCH_207S))
566 return true;
567 return false;
568}
569
570static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
571 unsigned long resource, unsigned long value1,
572 unsigned long value2)
573{
574 switch (resource) {
575 case H_SET_MODE_RESOURCE_SET_CIABR:
576 if (!kvmppc_power8_compatible(vcpu))
577 return H_P2;
578 if (value2)
579 return H_P4;
580 if (mflags)
581 return H_UNSUPPORTED_FLAG_START;
582 /* Guests can't breakpoint the hypervisor */
583 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
584 return H_P3;
585 vcpu->arch.ciabr = value1;
586 return H_SUCCESS;
587 case H_SET_MODE_RESOURCE_SET_DAWR:
588 if (!kvmppc_power8_compatible(vcpu))
589 return H_P2;
590 if (mflags)
591 return H_UNSUPPORTED_FLAG_START;
592 if (value2 & DABRX_HYP)
593 return H_P4;
594 vcpu->arch.dawr = value1;
595 vcpu->arch.dawrx = value2;
596 return H_SUCCESS;
597 default:
598 return H_TOO_HARD;
599 }
600}
601
a8606e20
PM		 602int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
603{
604 unsigned long req = kvmppc_get_gpr(vcpu, 3);
605 unsigned long target, ret = H_SUCCESS;
606 struct kvm_vcpu *tvcpu;
8e591cb7 607 int idx, rc;
a8606e20 608
699a0ea0
PM		 609 if (req <= MAX_HCALL_OPCODE &&
610 !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
611 return RESUME_HOST;
612
a8606e20 613 switch (req) {
c77162de 614 case H_ENTER:
2c9097e4 615 idx = srcu_read_lock(&vcpu->kvm->srcu);
c77162de
PM		 616 ret = kvmppc_virtmode_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
617 kvmppc_get_gpr(vcpu, 5),
618 kvmppc_get_gpr(vcpu, 6),
619 kvmppc_get_gpr(vcpu, 7));
2c9097e4 620 srcu_read_unlock(&vcpu->kvm->srcu, idx);
c77162de 621 break;
a8606e20 622 case H_CEDE:
a8606e20
PM		 623 break;
624 case H_PROD:
625 target = kvmppc_get_gpr(vcpu, 4);
626 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
627 if (!tvcpu) {
628 ret = H_PARAMETER;
629 break;
630 }
631 tvcpu->arch.prodded = 1;
632 smp_mb();
633 if (vcpu->arch.ceded) {
634 if (waitqueue_active(&vcpu->wq)) {
635 wake_up_interruptible(&vcpu->wq);
636 vcpu->stat.halt_wakeup++;
637 }
638 }
639 break;
640 case H_CONFER:
42d7604d
PM		 641 target = kvmppc_get_gpr(vcpu, 4);
642 if (target == -1)
643 break;
644 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
645 if (!tvcpu) {
646 ret = H_PARAMETER;
647 break;
648 }
649 kvm_vcpu_yield_to(tvcpu);
a8606e20
PM		 650 break;
651 case H_REGISTER_VPA:
652 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
653 kvmppc_get_gpr(vcpu, 5),
654 kvmppc_get_gpr(vcpu, 6));
655 break;
8e591cb7
ME		 656 case H_RTAS:
657 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
658 return RESUME_HOST;
659
c9438092 660 idx = srcu_read_lock(&vcpu->kvm->srcu);
8e591cb7 661 rc = kvmppc_rtas_hcall(vcpu);
c9438092 662 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8e591cb7
ME		 663
664 if (rc == -ENOENT)
665 return RESUME_HOST;
666 else if (rc == 0)
667 break;
668
669 /* Send the error out to userspace via KVM_RUN */
670 return rc;
9642382e
MN		 671 case H_SET_MODE:
672 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
673 kvmppc_get_gpr(vcpu, 5),
674 kvmppc_get_gpr(vcpu, 6),
675 kvmppc_get_gpr(vcpu, 7));
676 if (ret == H_TOO_HARD)
677 return RESUME_HOST;
678 break;
bc5ad3f3
BH		 679 case H_XIRR:
680 case H_CPPR:
681 case H_EOI:
682 case H_IPI:
8e44ddc3
PM		 683 case H_IPOLL:
684 case H_XIRR_X:
bc5ad3f3
BH		 685 if (kvmppc_xics_enabled(vcpu)) {
686 ret = kvmppc_xics_hcall(vcpu, req);
687 break;
688 } /* fallthrough */
a8606e20
PM		 689 default:
690 return RESUME_HOST;
691 }
692 kvmppc_set_gpr(vcpu, 3, ret);
693 vcpu->arch.hcall_needed = 0;
694 return RESUME_GUEST;
695}
696
ae2113a4
PM		 697static int kvmppc_hcall_impl_hv(unsigned long cmd)
698{
699 switch (cmd) {
700 case H_CEDE:
701 case H_PROD:
702 case H_CONFER:
703 case H_REGISTER_VPA:
9642382e 704 case H_SET_MODE:
ae2113a4
PM		 705#ifdef CONFIG_KVM_XICS
706 case H_XIRR:
707 case H_CPPR:
708 case H_EOI:
709 case H_IPI:
710 case H_IPOLL:
711 case H_XIRR_X:
712#endif
713 return 1;
714 }
715
716 /* See if it's in the real-mode table */
717 return kvmppc_hcall_impl_hv_realmode(cmd);
718}
719
3a167bea
AK		 720static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
721 struct task_struct *tsk)
de56a948
PM		 722{
723 int r = RESUME_HOST;
724
725 vcpu->stat.sum_exits++;
726
727 run->exit_reason = KVM_EXIT_UNKNOWN;
728 run->ready_for_interrupt_injection = 1;
729 switch (vcpu->arch.trap) {
730 /* We're good on these - the host merely wanted to get our attention */
731 case BOOK3S_INTERRUPT_HV_DECREMENTER:
732 vcpu->stat.dec_exits++;
733 r = RESUME_GUEST;
734 break;
735 case BOOK3S_INTERRUPT_EXTERNAL:
5d00f66b 736 case BOOK3S_INTERRUPT_H_DOORBELL:
de56a948
PM		 737 vcpu->stat.ext_intr_exits++;
738 r = RESUME_GUEST;
739 break;
740 case BOOK3S_INTERRUPT_PERFMON:
741 r = RESUME_GUEST;
742 break;
b4072df4
PM		 743 case BOOK3S_INTERRUPT_MACHINE_CHECK:
744 /*
745 * Deliver a machine check interrupt to the guest.
746 * We have to do this, even if the host has handled the
747 * machine check, because machine checks use SRR0/1 and
748 * the interrupt might have trashed guest state in them.
749 */
750 kvmppc_book3s_queue_irqprio(vcpu,
751 BOOK3S_INTERRUPT_MACHINE_CHECK);
752 r = RESUME_GUEST;
753 break;
de56a948
PM		 754 case BOOK3S_INTERRUPT_PROGRAM:
755 {
756 ulong flags;
757 /*
758 * Normally program interrupts are delivered directly
759 * to the guest by the hardware, but we can get here
760 * as a result of a hypervisor emulation interrupt
761 * (e40) getting turned into a 700 by BML RTAS.
762 */
763 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
764 kvmppc_core_queue_program(vcpu, flags);
765 r = RESUME_GUEST;
766 break;
767 }
768 case BOOK3S_INTERRUPT_SYSCALL:
769 {
770 /* hcall - punt to userspace */
771 int i;
772
27025a60
LPF		 773 /* hypercall with MSR_PR has already been handled in rmode,
774 * and never reaches here.
775 */
776
de56a948
PM		 777 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
778 for (i = 0; i < 9; ++i)
779 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
780 run->exit_reason = KVM_EXIT_PAPR_HCALL;
781 vcpu->arch.hcall_needed = 1;
782 r = RESUME_HOST;
783 break;
784 }
785 /*
342d3db7
PM		 786 * We get these next two if the guest accesses a page which it thinks
787 * it has mapped but which is not actually present, either because
788 * it is for an emulated I/O device or because the corresonding
789 * host page has been paged out. Any other HDSI/HISI interrupts
790 * have been handled already.
de56a948
PM		 791 */
792 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
913d3ff9 793 r = RESUME_PAGE_FAULT;
de56a948
PM		 794 break;
795 case BOOK3S_INTERRUPT_H_INST_STORAGE:
913d3ff9
PM		 796 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
797 vcpu->arch.fault_dsisr = 0;
798 r = RESUME_PAGE_FAULT;
de56a948
PM		 799 break;
800 /*
801 * This occurs if the guest executes an illegal instruction.
802 * We just generate a program interrupt to the guest, since
803 * we don't emulate any guest instructions at this stage.
804 */
805 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
bd3048b8
ME		 806 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
807 r = RESUME_GUEST;
808 break;
809 /*
810 * This occurs if the guest (kernel or userspace), does something that
811 * is prohibited by HFSCR. We just generate a program interrupt to
812 * the guest.
813 */
814 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
815 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
de56a948
PM		 816 r = RESUME_GUEST;
817 break;
818 default:
819 kvmppc_dump_regs(vcpu);
820 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
821 vcpu->arch.trap, kvmppc_get_pc(vcpu),
822 vcpu->arch.shregs.msr);
f3271d4c 823 run->hw.hardware_exit_reason = vcpu->arch.trap;
de56a948 824 r = RESUME_HOST;
de56a948
PM		 825 break;
826 }
827
de56a948
PM		 828 return r;
829}
830
3a167bea
AK		 831static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
832 struct kvm_sregs *sregs)
de56a948
PM		 833{
834 int i;
835
de56a948 836 memset(sregs, 0, sizeof(struct kvm_sregs));
87916442 837 sregs->pvr = vcpu->arch.pvr;
de56a948
PM		 838 for (i = 0; i < vcpu->arch.slb_max; i++) {
839 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
840 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
841 }
842
843 return 0;
844}
845
3a167bea
AK		 846static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
847 struct kvm_sregs *sregs)
de56a948
PM		 848{
849 int i, j;
850
3a167bea 851 kvmppc_set_pvr_hv(vcpu, sregs->pvr);
de56a948
PM		 852
853 j = 0;
854 for (i = 0; i < vcpu->arch.slb_nr; i++) {
855 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
856 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
857 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
858 ++j;
859 }
860 }
861 vcpu->arch.slb_max = j;
862
863 return 0;
864}
865
a0840240
AK		 866static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
867 bool preserve_top32)
a0144e2a
PM		 868{
869 struct kvmppc_vcore *vc = vcpu->arch.vcore;
870 u64 mask;
871
872 spin_lock(&vc->lock);
d682916a
AB		 873 /*
874 * If ILE (interrupt little-endian) has changed, update the
875 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
876 */
877 if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
878 struct kvm *kvm = vcpu->kvm;
879 struct kvm_vcpu *vcpu;
880 int i;
881
882 mutex_lock(&kvm->lock);
883 kvm_for_each_vcpu(i, vcpu, kvm) {
884 if (vcpu->arch.vcore != vc)
885 continue;
886 if (new_lpcr & LPCR_ILE)
887 vcpu->arch.intr_msr |= MSR_LE;
888 else
889 vcpu->arch.intr_msr &= ~MSR_LE;
890 }
891 mutex_unlock(&kvm->lock);
892 }
893
a0144e2a
PM		 894 /*
895 * Userspace can only modify DPFD (default prefetch depth),
896 * ILE (interrupt little-endian) and TC (translation control).
e0622bd9 897 * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
a0144e2a
PM		 898 */
899 mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
e0622bd9
PM		 900 if (cpu_has_feature(CPU_FTR_ARCH_207S))
901 mask |= LPCR_AIL;
a0840240
AK		 902
903 /* Broken 32-bit version of LPCR must not clear top bits */
904 if (preserve_top32)
905 mask &= 0xFFFFFFFF;
a0144e2a
PM		 906 vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
907 spin_unlock(&vc->lock);
908}
909
3a167bea
AK		 910static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
911 union kvmppc_one_reg *val)
31f3438e 912{
a136a8bd
PM		 913 int r = 0;
914 long int i;
31f3438e 915
a136a8bd 916 switch (id) {
31f3438e 917 case KVM_REG_PPC_HIOR:
a136a8bd
PM		 918 *val = get_reg_val(id, 0);
919 break;
920 case KVM_REG_PPC_DABR:
921 *val = get_reg_val(id, vcpu->arch.dabr);
922 break;
8563bf52
PM		 923 case KVM_REG_PPC_DABRX:
924 *val = get_reg_val(id, vcpu->arch.dabrx);
925 break;
a136a8bd
PM		 926 case KVM_REG_PPC_DSCR:
927 *val = get_reg_val(id, vcpu->arch.dscr);
928 break;
929 case KVM_REG_PPC_PURR:
930 *val = get_reg_val(id, vcpu->arch.purr);
931 break;
932 case KVM_REG_PPC_SPURR:
933 *val = get_reg_val(id, vcpu->arch.spurr);
934 break;
935 case KVM_REG_PPC_AMR:
936 *val = get_reg_val(id, vcpu->arch.amr);
937 break;
938 case KVM_REG_PPC_UAMOR:
939 *val = get_reg_val(id, vcpu->arch.uamor);
940 break;
b005255e 941 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
a136a8bd
PM		 942 i = id - KVM_REG_PPC_MMCR0;
943 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
944 break;
945 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
946 i = id - KVM_REG_PPC_PMC1;
947 *val = get_reg_val(id, vcpu->arch.pmc[i]);
31f3438e 948 break;
b005255e
MN		 949 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
950 i = id - KVM_REG_PPC_SPMC1;
951 *val = get_reg_val(id, vcpu->arch.spmc[i]);
952 break;
14941789
PM		 953 case KVM_REG_PPC_SIAR:
954 *val = get_reg_val(id, vcpu->arch.siar);
955 break;
956 case KVM_REG_PPC_SDAR:
957 *val = get_reg_val(id, vcpu->arch.sdar);
958 break;
b005255e
MN		 959 case KVM_REG_PPC_SIER:
960 *val = get_reg_val(id, vcpu->arch.sier);
a8bd19ef 961 break;
b005255e
MN		 962 case KVM_REG_PPC_IAMR:
963 *val = get_reg_val(id, vcpu->arch.iamr);
964 break;
b005255e
MN		 965 case KVM_REG_PPC_PSPB:
966 *val = get_reg_val(id, vcpu->arch.pspb);
967 break;
b005255e
MN		 968 case KVM_REG_PPC_DPDES:
969 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
970 break;
971 case KVM_REG_PPC_DAWR:
972 *val = get_reg_val(id, vcpu->arch.dawr);
973 break;
974 case KVM_REG_PPC_DAWRX:
975 *val = get_reg_val(id, vcpu->arch.dawrx);
976 break;
977 case KVM_REG_PPC_CIABR:
978 *val = get_reg_val(id, vcpu->arch.ciabr);
979 break;
b005255e
MN		 980 case KVM_REG_PPC_CSIGR:
981 *val = get_reg_val(id, vcpu->arch.csigr);
982 break;
983 case KVM_REG_PPC_TACR:
984 *val = get_reg_val(id, vcpu->arch.tacr);
985 break;
986 case KVM_REG_PPC_TCSCR:
987 *val = get_reg_val(id, vcpu->arch.tcscr);
988 break;
989 case KVM_REG_PPC_PID:
990 *val = get_reg_val(id, vcpu->arch.pid);
991 break;
992 case KVM_REG_PPC_ACOP:
993 *val = get_reg_val(id, vcpu->arch.acop);
994 break;
995 case KVM_REG_PPC_WORT:
996 *val = get_reg_val(id, vcpu->arch.wort);
a8bd19ef 997 break;
55b665b0
PM		 998 case KVM_REG_PPC_VPA_ADDR:
999 spin_lock(&vcpu->arch.vpa_update_lock);
1000 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1001 spin_unlock(&vcpu->arch.vpa_update_lock);
1002 break;
1003 case KVM_REG_PPC_VPA_SLB:
1004 spin_lock(&vcpu->arch.vpa_update_lock);
1005 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1006 val->vpaval.length = vcpu->arch.slb_shadow.len;
1007 spin_unlock(&vcpu->arch.vpa_update_lock);
1008 break;
1009 case KVM_REG_PPC_VPA_DTL:
1010 spin_lock(&vcpu->arch.vpa_update_lock);
1011 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1012 val->vpaval.length = vcpu->arch.dtl.len;
1013 spin_unlock(&vcpu->arch.vpa_update_lock);
1014 break;
93b0f4dc
PM		 1015 case KVM_REG_PPC_TB_OFFSET:
1016 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1017 break;
a0144e2a 1018 case KVM_REG_PPC_LPCR:
a0840240 1019 case KVM_REG_PPC_LPCR_64:
a0144e2a
PM		 1020 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1021 break;
4b8473c9
PM		 1022 case KVM_REG_PPC_PPR:
1023 *val = get_reg_val(id, vcpu->arch.ppr);
1024 break;
a7d80d01
MN		 1025#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1026 case KVM_REG_PPC_TFHAR:
1027 *val = get_reg_val(id, vcpu->arch.tfhar);
1028 break;
1029 case KVM_REG_PPC_TFIAR:
1030 *val = get_reg_val(id, vcpu->arch.tfiar);
1031 break;
1032 case KVM_REG_PPC_TEXASR:
1033 *val = get_reg_val(id, vcpu->arch.texasr);
1034 break;
1035 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1036 i = id - KVM_REG_PPC_TM_GPR0;
1037 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1038 break;
1039 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1040 {
1041 int j;
1042 i = id - KVM_REG_PPC_TM_VSR0;
1043 if (i < 32)
1044 for (j = 0; j < TS_FPRWIDTH; j++)
1045 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1046 else {
1047 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1048 val->vval = vcpu->arch.vr_tm.vr[i-32];
1049 else
1050 r = -ENXIO;
1051 }
1052 break;
1053 }
1054 case KVM_REG_PPC_TM_CR:
1055 *val = get_reg_val(id, vcpu->arch.cr_tm);
1056 break;
1057 case KVM_REG_PPC_TM_LR:
1058 *val = get_reg_val(id, vcpu->arch.lr_tm);
1059 break;
1060 case KVM_REG_PPC_TM_CTR:
1061 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1062 break;
1063 case KVM_REG_PPC_TM_FPSCR:
1064 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1065 break;
1066 case KVM_REG_PPC_TM_AMR:
1067 *val = get_reg_val(id, vcpu->arch.amr_tm);
1068 break;
1069 case KVM_REG_PPC_TM_PPR:
1070 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1071 break;
1072 case KVM_REG_PPC_TM_VRSAVE:
1073 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1074 break;
1075 case KVM_REG_PPC_TM_VSCR:
1076 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1077 *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1078 else
1079 r = -ENXIO;
1080 break;
1081 case KVM_REG_PPC_TM_DSCR:
1082 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1083 break;
1084 case KVM_REG_PPC_TM_TAR:
1085 *val = get_reg_val(id, vcpu->arch.tar_tm);
1086 break;
1087#endif
388cc6e1
PM		 1088 case KVM_REG_PPC_ARCH_COMPAT:
1089 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1090 break;
31f3438e 1091 default:
a136a8bd 1092 r = -EINVAL;
31f3438e
PM		 1093 break;
1094 }
1095
1096 return r;
1097}
1098
3a167bea
AK		 1099static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1100 union kvmppc_one_reg *val)
31f3438e 1101{
a136a8bd
PM		 1102 int r = 0;
1103 long int i;
55b665b0 1104 unsigned long addr, len;
31f3438e 1105
a136a8bd 1106 switch (id) {
31f3438e 1107 case KVM_REG_PPC_HIOR:
31f3438e 1108 /* Only allow this to be set to zero */
a136a8bd 1109 if (set_reg_val(id, *val))
31f3438e
PM		 1110 r = -EINVAL;
1111 break;
a136a8bd
PM		 1112 case KVM_REG_PPC_DABR:
1113 vcpu->arch.dabr = set_reg_val(id, *val);
1114 break;
8563bf52
PM		 1115 case KVM_REG_PPC_DABRX:
1116 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1117 break;
a136a8bd
PM		 1118 case KVM_REG_PPC_DSCR:
1119 vcpu->arch.dscr = set_reg_val(id, *val);
1120 break;
1121 case KVM_REG_PPC_PURR:
1122 vcpu->arch.purr = set_reg_val(id, *val);
1123 break;
1124 case KVM_REG_PPC_SPURR:
1125 vcpu->arch.spurr = set_reg_val(id, *val);
1126 break;
1127 case KVM_REG_PPC_AMR:
1128 vcpu->arch.amr = set_reg_val(id, *val);
1129 break;
1130 case KVM_REG_PPC_UAMOR:
1131 vcpu->arch.uamor = set_reg_val(id, *val);
1132 break;
b005255e 1133 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
a136a8bd
PM		 1134 i = id - KVM_REG_PPC_MMCR0;
1135 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1136 break;
1137 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1138 i = id - KVM_REG_PPC_PMC1;
1139 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1140 break;
b005255e
MN		 1141 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1142 i = id - KVM_REG_PPC_SPMC1;
1143 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1144 break;
14941789
PM		 1145 case KVM_REG_PPC_SIAR:
1146 vcpu->arch.siar = set_reg_val(id, *val);
1147 break;
1148 case KVM_REG_PPC_SDAR:
1149 vcpu->arch.sdar = set_reg_val(id, *val);
1150 break;
b005255e
MN		 1151 case KVM_REG_PPC_SIER:
1152 vcpu->arch.sier = set_reg_val(id, *val);
a8bd19ef 1153 break;
b005255e
MN		 1154 case KVM_REG_PPC_IAMR:
1155 vcpu->arch.iamr = set_reg_val(id, *val);
1156 break;
b005255e
MN		 1157 case KVM_REG_PPC_PSPB:
1158 vcpu->arch.pspb = set_reg_val(id, *val);
1159 break;
b005255e
MN		 1160 case KVM_REG_PPC_DPDES:
1161 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1162 break;
1163 case KVM_REG_PPC_DAWR:
1164 vcpu->arch.dawr = set_reg_val(id, *val);
1165 break;
1166 case KVM_REG_PPC_DAWRX:
1167 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1168 break;
1169 case KVM_REG_PPC_CIABR:
1170 vcpu->arch.ciabr = set_reg_val(id, *val);
1171 /* Don't allow setting breakpoints in hypervisor code */
1172 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1173 vcpu->arch.ciabr &= ~CIABR_PRIV; /* disable */
1174 break;
b005255e
MN		 1175 case KVM_REG_PPC_CSIGR:
1176 vcpu->arch.csigr = set_reg_val(id, *val);
1177 break;
1178 case KVM_REG_PPC_TACR:
1179 vcpu->arch.tacr = set_reg_val(id, *val);
1180 break;
1181 case KVM_REG_PPC_TCSCR:
1182 vcpu->arch.tcscr = set_reg_val(id, *val);
1183 break;
1184 case KVM_REG_PPC_PID:
1185 vcpu->arch.pid = set_reg_val(id, *val);
1186 break;
1187 case KVM_REG_PPC_ACOP:
1188 vcpu->arch.acop = set_reg_val(id, *val);
1189 break;
1190 case KVM_REG_PPC_WORT:
1191 vcpu->arch.wort = set_reg_val(id, *val);
a8bd19ef 1192 break;
55b665b0
PM		 1193 case KVM_REG_PPC_VPA_ADDR:
1194 addr = set_reg_val(id, *val);
1195 r = -EINVAL;
1196 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1197 vcpu->arch.dtl.next_gpa))
1198 break;
1199 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1200 break;
1201 case KVM_REG_PPC_VPA_SLB:
1202 addr = val->vpaval.addr;
1203 len = val->vpaval.length;
1204 r = -EINVAL;
1205 if (addr && !vcpu->arch.vpa.next_gpa)
1206 break;
1207 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1208 break;
1209 case KVM_REG_PPC_VPA_DTL:
1210 addr = val->vpaval.addr;
1211 len = val->vpaval.length;
1212 r = -EINVAL;
9f8c8c78
PM		 1213 if (addr && (len < sizeof(struct dtl_entry) ||
1214 !vcpu->arch.vpa.next_gpa))
55b665b0
PM		 1215 break;
1216 len -= len % sizeof(struct dtl_entry);
1217 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1218 break;
93b0f4dc
PM		 1219 case KVM_REG_PPC_TB_OFFSET:
1220 /* round up to multiple of 2^24 */
1221 vcpu->arch.vcore->tb_offset =
1222 ALIGN(set_reg_val(id, *val), 1UL << 24);
1223 break;
a0144e2a 1224 case KVM_REG_PPC_LPCR:
a0840240
AK		 1225 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1226 break;
1227 case KVM_REG_PPC_LPCR_64:
1228 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
a0144e2a 1229 break;
4b8473c9
PM		 1230 case KVM_REG_PPC_PPR:
1231 vcpu->arch.ppr = set_reg_val(id, *val);
1232 break;
a7d80d01
MN		 1233#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1234 case KVM_REG_PPC_TFHAR:
1235 vcpu->arch.tfhar = set_reg_val(id, *val);
1236 break;
1237 case KVM_REG_PPC_TFIAR:
1238 vcpu->arch.tfiar = set_reg_val(id, *val);
1239 break;
1240 case KVM_REG_PPC_TEXASR:
1241 vcpu->arch.texasr = set_reg_val(id, *val);
1242 break;
1243 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1244 i = id - KVM_REG_PPC_TM_GPR0;
1245 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1246 break;
1247 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1248 {
1249 int j;
1250 i = id - KVM_REG_PPC_TM_VSR0;
1251 if (i < 32)
1252 for (j = 0; j < TS_FPRWIDTH; j++)
1253 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1254 else
1255 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1256 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1257 else
1258 r = -ENXIO;
1259 break;
1260 }
1261 case KVM_REG_PPC_TM_CR:
1262 vcpu->arch.cr_tm = set_reg_val(id, *val);
1263 break;
1264 case KVM_REG_PPC_TM_LR:
1265 vcpu->arch.lr_tm = set_reg_val(id, *val);
1266 break;
1267 case KVM_REG_PPC_TM_CTR:
1268 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1269 break;
1270 case KVM_REG_PPC_TM_FPSCR:
1271 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1272 break;
1273 case KVM_REG_PPC_TM_AMR:
1274 vcpu->arch.amr_tm = set_reg_val(id, *val);
1275 break;
1276 case KVM_REG_PPC_TM_PPR:
1277 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1278 break;
1279 case KVM_REG_PPC_TM_VRSAVE:
1280 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1281 break;
1282 case KVM_REG_PPC_TM_VSCR:
1283 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1284 vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1285 else
1286 r = - ENXIO;
1287 break;
1288 case KVM_REG_PPC_TM_DSCR:
1289 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1290 break;
1291 case KVM_REG_PPC_TM_TAR:
1292 vcpu->arch.tar_tm = set_reg_val(id, *val);
1293 break;
1294#endif
388cc6e1
PM		 1295 case KVM_REG_PPC_ARCH_COMPAT:
1296 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1297 break;
31f3438e 1298 default:
a136a8bd 1299 r = -EINVAL;
31f3438e
PM		 1300 break;
1301 }
1302
1303 return r;
1304}
1305
de9bdd1a
SS		 1306static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1307{
1308 struct kvmppc_vcore *vcore;
1309
1310 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1311
1312 if (vcore == NULL)
1313 return NULL;
1314
1315 INIT_LIST_HEAD(&vcore->runnable_threads);
1316 spin_lock_init(&vcore->lock);
1317 init_waitqueue_head(&vcore->wq);
1318 vcore->preempt_tb = TB_NIL;
1319 vcore->lpcr = kvm->arch.lpcr;
1320 vcore->first_vcpuid = core * threads_per_subcore;
1321 vcore->kvm = kvm;
1322
1323 return vcore;
1324}
1325
3a167bea
AK		 1326static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1327 unsigned int id)
de56a948
PM		 1328{
1329 struct kvm_vcpu *vcpu;
371fefd6
PM		 1330 int err = -EINVAL;
1331 int core;
1332 struct kvmppc_vcore *vcore;
de56a948 1333
3102f784 1334 core = id / threads_per_subcore;
371fefd6
PM		 1335 if (core >= KVM_MAX_VCORES)
1336 goto out;
1337
1338 err = -ENOMEM;
6b75e6bf 1339 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
de56a948
PM		 1340 if (!vcpu)
1341 goto out;
1342
1343 err = kvm_vcpu_init(vcpu, kvm, id);
1344 if (err)
1345 goto free_vcpu;
1346
1347 vcpu->arch.shared = &vcpu->arch.shregs;
5deb8e7a
AG		 1348#ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1349 /*
1350 * The shared struct is never shared on HV,
1351 * so we can always use host endianness
1352 */
1353#ifdef __BIG_ENDIAN__
1354 vcpu->arch.shared_big_endian = true;
1355#else
1356 vcpu->arch.shared_big_endian = false;
1357#endif
1358#endif
de56a948
PM		 1359 vcpu->arch.mmcr[0] = MMCR0_FC;
1360 vcpu->arch.ctrl = CTRL_RUNLATCH;
1361 /* default to host PVR, since we can't spoof it */
3a167bea 1362 kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2e25aa5f 1363 spin_lock_init(&vcpu->arch.vpa_update_lock);
c7b67670
PM		 1364 spin_lock_init(&vcpu->arch.tbacct_lock);
1365 vcpu->arch.busy_preempt = TB_NIL;
d682916a 1366 vcpu->arch.intr_msr = MSR_SF | MSR_ME;
de56a948 1367
de56a948
PM		 1368 kvmppc_mmu_book3s_hv_init(vcpu);
1369
8455d79e 1370 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
371fefd6
PM		 1371
1372 init_waitqueue_head(&vcpu->arch.cpu_run);
1373
1374 mutex_lock(&kvm->lock);
1375 vcore = kvm->arch.vcores[core];
1376 if (!vcore) {
de9bdd1a 1377 vcore = kvmppc_vcore_create(kvm, core);
371fefd6 1378 kvm->arch.vcores[core] = vcore;
1b400ba0 1379 kvm->arch.online_vcores++;
371fefd6
PM		 1380 }
1381 mutex_unlock(&kvm->lock);
1382
1383 if (!vcore)
1384 goto free_vcpu;
1385
1386 spin_lock(&vcore->lock);
1387 ++vcore->num_threads;
371fefd6
PM		 1388 spin_unlock(&vcore->lock);
1389 vcpu->arch.vcore = vcore;
e0b7ec05 1390 vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
371fefd6 1391
af8f38b3
AG		 1392 vcpu->arch.cpu_type = KVM_CPU_3S_64;
1393 kvmppc_sanity_check(vcpu);
1394
de56a948
PM		 1395 return vcpu;
1396
1397free_vcpu:
6b75e6bf 1398 kmem_cache_free(kvm_vcpu_cache, vcpu);
de56a948
PM		 1399out:
1400 return ERR_PTR(err);
1401}
1402
c35635ef
PM		 1403static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1404{
1405 if (vpa->pinned_addr)
1406 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1407 vpa->dirty);
1408}
1409
3a167bea 1410static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
de56a948 1411{
2e25aa5f 1412 spin_lock(&vcpu->arch.vpa_update_lock);
c35635ef
PM		 1413 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1414 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1415 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2e25aa5f 1416 spin_unlock(&vcpu->arch.vpa_update_lock);
de56a948 1417 kvm_vcpu_uninit(vcpu);
6b75e6bf 1418 kmem_cache_free(kvm_vcpu_cache, vcpu);
de56a948
PM		 1419}
1420
3a167bea
AK		 1421static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1422{
1423 /* Indicate we want to get back into the guest */
1424 return 1;
1425}
1426
19ccb76a 1427static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
371fefd6 1428{
19ccb76a 1429 unsigned long dec_nsec, now;
371fefd6 1430
19ccb76a
PM		 1431 now = get_tb();
1432 if (now > vcpu->arch.dec_expires) {
1433 /* decrementer has already gone negative */
1434 kvmppc_core_queue_dec(vcpu);
7e28e60e 1435 kvmppc_core_prepare_to_enter(vcpu);
19ccb76a 1436 return;
371fefd6 1437 }
19ccb76a
PM		 1438 dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1439 / tb_ticks_per_sec;
1440 hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1441 HRTIMER_MODE_REL);
1442 vcpu->arch.timer_running = 1;
371fefd6
PM		 1443}
1444
19ccb76a 1445static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
371fefd6 1446{
19ccb76a
PM		 1447 vcpu->arch.ceded = 0;
1448 if (vcpu->arch.timer_running) {
1449 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1450 vcpu->arch.timer_running = 0;
1451 }
371fefd6
PM		 1452}
1453
e0b7ec05 1454extern void __kvmppc_vcore_entry(void);
de56a948 1455
371fefd6
PM		 1456static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1457 struct kvm_vcpu *vcpu)
de56a948 1458{
c7b67670
PM		 1459 u64 now;
1460
371fefd6
PM		 1461 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1462 return;
bf3d32e1 1463 spin_lock_irq(&vcpu->arch.tbacct_lock);
c7b67670
PM		 1464 now = mftb();
1465 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1466 vcpu->arch.stolen_logged;
1467 vcpu->arch.busy_preempt = now;
1468 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
bf3d32e1 1469 spin_unlock_irq(&vcpu->arch.tbacct_lock);
371fefd6 1470 --vc->n_runnable;
371fefd6
PM		 1471 list_del(&vcpu->arch.run_list);
1472}
1473
f0888f70
PM		 1474static int kvmppc_grab_hwthread(int cpu)
1475{
1476 struct paca_struct *tpaca;
1477 long timeout = 1000;
1478
1479 tpaca = &paca[cpu];
1480
1481 /* Ensure the thread won't go into the kernel if it wakes */
1482 tpaca->kvm_hstate.hwthread_req = 1;
7b444c67 1483 tpaca->kvm_hstate.kvm_vcpu = NULL;
f0888f70
PM		 1484
1485 /*
1486 * If the thread is already executing in the kernel (e.g. handling
1487 * a stray interrupt), wait for it to get back to nap mode.
1488 * The smp_mb() is to ensure that our setting of hwthread_req
1489 * is visible before we look at hwthread_state, so if this
1490 * races with the code at system_reset_pSeries and the thread
1491 * misses our setting of hwthread_req, we are sure to see its
1492 * setting of hwthread_state, and vice versa.
1493 */
1494 smp_mb();
1495 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1496 if (--timeout <= 0) {
1497 pr_err("KVM: couldn't grab cpu %d\n", cpu);
1498 return -EBUSY;
1499 }
1500 udelay(1);
1501 }
1502 return 0;
1503}
1504
1505static void kvmppc_release_hwthread(int cpu)
1506{
1507 struct paca_struct *tpaca;
1508
1509 tpaca = &paca[cpu];
1510 tpaca->kvm_hstate.hwthread_req = 0;
1511 tpaca->kvm_hstate.kvm_vcpu = NULL;
1512}
1513
371fefd6
PM		 1514static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
1515{
1516 int cpu;
1517 struct paca_struct *tpaca;
1518 struct kvmppc_vcore *vc = vcpu->arch.vcore;
1519
19ccb76a
PM		 1520 if (vcpu->arch.timer_running) {
1521 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1522 vcpu->arch.timer_running = 0;
1523 }
371fefd6
PM		 1524 cpu = vc->pcpu + vcpu->arch.ptid;
1525 tpaca = &paca[cpu];
1526 tpaca->kvm_hstate.kvm_vcpu = vcpu;
1527 tpaca->kvm_hstate.kvm_vcore = vc;
e0b7ec05 1528 tpaca->kvm_hstate.ptid = vcpu->arch.ptid;
19ccb76a 1529 vcpu->cpu = vc->pcpu;
371fefd6 1530 smp_wmb();
251da038 1531#if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
e0b7ec05 1532 if (cpu != smp_processor_id()) {
371fefd6 1533 xics_wake_cpu(cpu);
e0b7ec05
PM		 1534 if (vcpu->arch.ptid)
1535 ++vc->n_woken;
de56a948 1536 }
371fefd6
PM		 1537#endif
1538}
de56a948 1539
371fefd6
PM		 1540static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc)
1541{
1542 int i;
1543
1544 HMT_low();
1545 i = 0;
1546 while (vc->nap_count < vc->n_woken) {
1547 if (++i >= 1000000) {
1548 pr_err("kvmppc_wait_for_nap timeout %d %d\n",
1549 vc->nap_count, vc->n_woken);
1550 break;
1551 }
1552 cpu_relax();
1553 }
1554 HMT_medium();
1555}
1556
1557/*
1558 * Check that we are on thread 0 and that any other threads in
7b444c67
PM		 1559 * this core are off-line. Then grab the threads so they can't
1560 * enter the kernel.
371fefd6
PM		 1561 */
1562static int on_primary_thread(void)
1563{
1564 int cpu = smp_processor_id();
3102f784 1565 int thr;
371fefd6 1566
3102f784
ME		 1567 /* Are we on a primary subcore? */
1568 if (cpu_thread_in_subcore(cpu))
371fefd6 1569 return 0;
3102f784
ME		 1570
1571 thr = 0;
1572 while (++thr < threads_per_subcore)
371fefd6
PM		 1573 if (cpu_online(cpu + thr))
1574 return 0;
7b444c67
PM		 1575
1576 /* Grab all hw threads so they can't go into the kernel */
3102f784 1577 for (thr = 1; thr < threads_per_subcore; ++thr) {
7b444c67
PM		 1578 if (kvmppc_grab_hwthread(cpu + thr)) {
1579 /* Couldn't grab one; let the others go */
1580 do {
1581 kvmppc_release_hwthread(cpu + thr);
1582 } while (--thr > 0);
1583 return 0;
1584 }
1585 }
371fefd6
PM		 1586 return 1;
1587}
1588
1589/*
1590 * Run a set of guest threads on a physical core.
1591 * Called with vc->lock held.
1592 */
913d3ff9 1593static void kvmppc_run_core(struct kvmppc_vcore *vc)
371fefd6 1594{
e0b7ec05 1595 struct kvm_vcpu *vcpu, *vnext;
371fefd6
PM		 1596 long ret;
1597 u64 now;
e0b7ec05 1598 int i, need_vpa_update;
2c9097e4 1599 int srcu_idx;
913d3ff9 1600 struct kvm_vcpu *vcpus_to_update[threads_per_core];
371fefd6
PM		 1601
1602 /* don't start if any threads have a signal pending */
081f323b
PM		 1603 need_vpa_update = 0;
1604 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
371fefd6 1605 if (signal_pending(vcpu->arch.run_task))
913d3ff9
PM		 1606 return;
1607 if (vcpu->arch.vpa.update_pending ||
1608 vcpu->arch.slb_shadow.update_pending ||
1609 vcpu->arch.dtl.update_pending)
1610 vcpus_to_update[need_vpa_update++] = vcpu;
081f323b
PM		 1611 }
1612
1613 /*
1614 * Initialize *vc, in particular vc->vcore_state, so we can
1615 * drop the vcore lock if necessary.
1616 */
1617 vc->n_woken = 0;
1618 vc->nap_count = 0;
1619 vc->entry_exit_count = 0;
2f12f034 1620 vc->vcore_state = VCORE_STARTING;
081f323b
PM		 1621 vc->in_guest = 0;
1622 vc->napping_threads = 0;
1623
1624 /*
1625 * Updating any of the vpas requires calling kvmppc_pin_guest_page,
1626 * which can't be called with any spinlocks held.
1627 */
1628 if (need_vpa_update) {
1629 spin_unlock(&vc->lock);
913d3ff9
PM		 1630 for (i = 0; i < need_vpa_update; ++i)
1631 kvmppc_update_vpas(vcpus_to_update[i]);
081f323b
PM		 1632 spin_lock(&vc->lock);
1633 }
de56a948 1634
7b444c67 1635 /*
3102f784
ME		 1636 * Make sure we are running on primary threads, and that secondary
1637 * threads are offline. Also check if the number of threads in this
1638 * guest are greater than the current system threads per guest.
7b444c67 1639 */
3102f784
ME		 1640 if ((threads_per_core > 1) &&
1641 ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
7b444c67
PM		 1642 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1643 vcpu->arch.ret = -EBUSY;
1644 goto out;
1645 }
1646
3102f784 1647
371fefd6 1648 vc->pcpu = smp_processor_id();
2e25aa5f 1649 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
371fefd6 1650 kvmppc_start_thread(vcpu);
0456ec4f 1651 kvmppc_create_dtl_entry(vcpu, vc);
2e25aa5f 1652 }
371fefd6 1653
e0b7ec05
PM		 1654 /* Set this explicitly in case thread 0 doesn't have a vcpu */
1655 get_paca()->kvm_hstate.kvm_vcore = vc;
1656 get_paca()->kvm_hstate.ptid = 0;
1657
2f12f034 1658 vc->vcore_state = VCORE_RUNNING;
19ccb76a 1659 preempt_disable();
371fefd6 1660 spin_unlock(&vc->lock);
de56a948 1661
371fefd6 1662 kvm_guest_enter();
2c9097e4 1663
e0b7ec05 1664 srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2c9097e4 1665
e0b7ec05 1666 __kvmppc_vcore_entry();
de56a948 1667
371fefd6 1668 spin_lock(&vc->lock);
19ccb76a
PM		 1669 /* disable sending of IPIs on virtual external irqs */
1670 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1671 vcpu->cpu = -1;
1672 /* wait for secondary threads to finish writing their state to memory */
371fefd6
PM		 1673 if (vc->nap_count < vc->n_woken)
1674 kvmppc_wait_for_nap(vc);
3102f784 1675 for (i = 0; i < threads_per_subcore; ++i)
2f12f034 1676 kvmppc_release_hwthread(vc->pcpu + i);
371fefd6 1677 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
19ccb76a 1678 vc->vcore_state = VCORE_EXITING;
371fefd6
PM		 1679 spin_unlock(&vc->lock);
1680
e0b7ec05 1681 srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2c9097e4 1682
371fefd6
PM		 1683 /* make sure updates to secondary vcpu structs are visible now */
1684 smp_mb();
de56a948
PM		 1685 kvm_guest_exit();
1686
1687 preempt_enable();
c08ac06a 1688 cond_resched();
de56a948 1689
913d3ff9 1690 spin_lock(&vc->lock);
de56a948 1691 now = get_tb();
371fefd6
PM		 1692 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1693 /* cancel pending dec exception if dec is positive */
1694 if (now < vcpu->arch.dec_expires &&
1695 kvmppc_core_pending_dec(vcpu))
1696 kvmppc_core_dequeue_dec(vcpu);
19ccb76a
PM		 1697
1698 ret = RESUME_GUEST;
1699 if (vcpu->arch.trap)
3a167bea
AK		 1700 ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
1701 vcpu->arch.run_task);
19ccb76a 1702
371fefd6
PM		 1703 vcpu->arch.ret = ret;
1704 vcpu->arch.trap = 0;
19ccb76a
PM		 1705
1706 if (vcpu->arch.ceded) {
e59d24e6 1707 if (!is_kvmppc_resume_guest(ret))
19ccb76a
PM		 1708 kvmppc_end_cede(vcpu);
1709 else
1710 kvmppc_set_timer(vcpu);
1711 }
371fefd6 1712 }
de56a948
PM		 1713
1714 out:
19ccb76a 1715 vc->vcore_state = VCORE_INACTIVE;
371fefd6
PM		 1716 list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1717 arch.run_list) {
e59d24e6 1718 if (!is_kvmppc_resume_guest(vcpu->arch.ret)) {
371fefd6
PM		 1719 kvmppc_remove_runnable(vc, vcpu);
1720 wake_up(&vcpu->arch.cpu_run);
1721 }
1722 }
371fefd6
PM		 1723}
1724
19ccb76a
PM		 1725/*
1726 * Wait for some other vcpu thread to execute us, and
1727 * wake us up when we need to handle something in the host.
1728 */
1729static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
371fefd6 1730{
371fefd6
PM		 1731 DEFINE_WAIT(wait);
1732
19ccb76a
PM		 1733 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
1734 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
1735 schedule();
1736 finish_wait(&vcpu->arch.cpu_run, &wait);
1737}
1738
1739/*
1740 * All the vcpus in this vcore are idle, so wait for a decrementer
1741 * or external interrupt to one of the vcpus. vc->lock is held.
1742 */
1743static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
1744{
1745 DEFINE_WAIT(wait);
19ccb76a
PM		 1746
1747 prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
1748 vc->vcore_state = VCORE_SLEEPING;
1749 spin_unlock(&vc->lock);
913d3ff9 1750 schedule();
19ccb76a
PM		 1751 finish_wait(&vc->wq, &wait);
1752 spin_lock(&vc->lock);
1753 vc->vcore_state = VCORE_INACTIVE;
1754}
371fefd6 1755
19ccb76a
PM		 1756static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1757{
1758 int n_ceded;
19ccb76a
PM		 1759 struct kvmppc_vcore *vc;
1760 struct kvm_vcpu *v, *vn;
9e368f29 1761
371fefd6
PM		 1762 kvm_run->exit_reason = 0;
1763 vcpu->arch.ret = RESUME_GUEST;
1764 vcpu->arch.trap = 0;
2f12f034 1765 kvmppc_update_vpas(vcpu);
371fefd6 1766
371fefd6
PM		 1767 /*
1768 * Synchronize with other threads in this virtual core
1769 */
1770 vc = vcpu->arch.vcore;
1771 spin_lock(&vc->lock);
19ccb76a 1772 vcpu->arch.ceded = 0;
371fefd6
PM		 1773 vcpu->arch.run_task = current;
1774 vcpu->arch.kvm_run = kvm_run;
c7b67670 1775 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
19ccb76a 1776 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
c7b67670 1777 vcpu->arch.busy_preempt = TB_NIL;
371fefd6
PM		 1778 list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
1779 ++vc->n_runnable;
1780
19ccb76a
PM		 1781 /*
1782 * This happens the first time this is called for a vcpu.
1783 * If the vcore is already running, we may be able to start
1784 * this thread straight away and have it join in.
1785 */
8455d79e 1786 if (!signal_pending(current)) {
19ccb76a
PM		 1787 if (vc->vcore_state == VCORE_RUNNING &&
1788 VCORE_EXIT_COUNT(vc) == 0) {
2f12f034 1789 kvmppc_create_dtl_entry(vcpu, vc);
19ccb76a 1790 kvmppc_start_thread(vcpu);
8455d79e
PM		 1791 } else if (vc->vcore_state == VCORE_SLEEPING) {
1792 wake_up(&vc->wq);
371fefd6
PM		 1793 }
1794
8455d79e 1795 }
371fefd6 1796
19ccb76a
PM		 1797 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1798 !signal_pending(current)) {
8455d79e 1799 if (vc->vcore_state != VCORE_INACTIVE) {
19ccb76a
PM		 1800 spin_unlock(&vc->lock);
1801 kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
1802 spin_lock(&vc->lock);
1803 continue;
1804 }
19ccb76a
PM		 1805 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
1806 arch.run_list) {
7e28e60e 1807 kvmppc_core_prepare_to_enter(v);
19ccb76a
PM		 1808 if (signal_pending(v->arch.run_task)) {
1809 kvmppc_remove_runnable(vc, v);
1810 v->stat.signal_exits++;
1811 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
1812 v->arch.ret = -EINTR;
1813 wake_up(&v->arch.cpu_run);
1814 }
1815 }
8455d79e
PM		 1816 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1817 break;
1818 vc->runner = vcpu;
1819 n_ceded = 0;
4619ac88 1820 list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
8455d79e
PM		 1821 if (!v->arch.pending_exceptions)
1822 n_ceded += v->arch.ceded;
4619ac88
PM		 1823 else
1824 v->arch.ceded = 0;
1825 }
8455d79e
PM		 1826 if (n_ceded == vc->n_runnable)
1827 kvmppc_vcore_blocked(vc);
1828 else
1829 kvmppc_run_core(vc);
0456ec4f 1830 vc->runner = NULL;
19ccb76a 1831 }
371fefd6 1832
8455d79e
PM		 1833 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1834 (vc->vcore_state == VCORE_RUNNING ||
1835 vc->vcore_state == VCORE_EXITING)) {
1836 spin_unlock(&vc->lock);
1837 kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
1838 spin_lock(&vc->lock);
1839 }
1840
1841 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
1842 kvmppc_remove_runnable(vc, vcpu);
1843 vcpu->stat.signal_exits++;
1844 kvm_run->exit_reason = KVM_EXIT_INTR;
1845 vcpu->arch.ret = -EINTR;
1846 }
1847
1848 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
1849 /* Wake up some vcpu to run the core */
1850 v = list_first_entry(&vc->runnable_threads,
1851 struct kvm_vcpu, arch.run_list);
1852 wake_up(&v->arch.cpu_run);
371fefd6
PM		 1853 }
1854
371fefd6 1855 spin_unlock(&vc->lock);
371fefd6 1856 return vcpu->arch.ret;
de56a948
PM		 1857}
1858
3a167bea 1859static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
a8606e20
PM		 1860{
1861 int r;
913d3ff9 1862 int srcu_idx;
a8606e20 1863
af8f38b3
AG		 1864 if (!vcpu->arch.sane) {
1865 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1866 return -EINVAL;
1867 }
1868
25051b5a
SW		 1869 kvmppc_core_prepare_to_enter(vcpu);
1870
19ccb76a
PM		 1871 /* No need to go into the guest when all we'll do is come back out */
1872 if (signal_pending(current)) {
1873 run->exit_reason = KVM_EXIT_INTR;
1874 return -EINTR;
1875 }
1876
32fad281
PM		 1877 atomic_inc(&vcpu->kvm->arch.vcpus_running);
1878 /* Order vcpus_running vs. rma_setup_done, see kvmppc_alloc_reset_hpt */
1879 smp_mb();
1880
1881 /* On the first time here, set up HTAB and VRMA or RMA */
c77162de 1882 if (!vcpu->kvm->arch.rma_setup_done) {
32fad281 1883 r = kvmppc_hv_setup_htab_rma(vcpu);
c77162de 1884 if (r)
32fad281 1885 goto out;
c77162de 1886 }
19ccb76a
PM		 1887
1888 flush_fp_to_thread(current);
1889 flush_altivec_to_thread(current);
1890 flush_vsx_to_thread(current);
1891 vcpu->arch.wqp = &vcpu->arch.vcore->wq;
342d3db7 1892 vcpu->arch.pgdir = current->mm->pgd;
c7b67670 1893 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
19ccb76a 1894
a8606e20
PM		 1895 do {
1896 r = kvmppc_run_vcpu(run, vcpu);
1897
1898 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
1899 !(vcpu->arch.shregs.msr & MSR_PR)) {
1900 r = kvmppc_pseries_do_hcall(vcpu);
7e28e60e 1901 kvmppc_core_prepare_to_enter(vcpu);
913d3ff9
PM		 1902 } else if (r == RESUME_PAGE_FAULT) {
1903 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1904 r = kvmppc_book3s_hv_page_fault(run, vcpu,
1905 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1906 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
a8606e20 1907 }
e59d24e6 1908 } while (is_kvmppc_resume_guest(r));
32fad281
PM		 1909
1910 out:
c7b67670 1911 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
32fad281 1912 atomic_dec(&vcpu->kvm->arch.vcpus_running);
a8606e20
PM		 1913 return r;
1914}
1915
54738c09 1916
aa04b4cc 1917/* Work out RMLS (real mode limit selector) field value for a given RMA size.
9e368f29 1918 Assumes POWER7 or PPC970. */
aa04b4cc
PM		 1919static inline int lpcr_rmls(unsigned long rma_size)
1920{
1921 switch (rma_size) {
1922 case 32ul << 20: /* 32 MB */
9e368f29
PM		 1923 if (cpu_has_feature(CPU_FTR_ARCH_206))
1924 return 8; /* only supported on POWER7 */
1925 return -1;
aa04b4cc
PM		 1926 case 64ul << 20: /* 64 MB */
1927 return 3;
1928 case 128ul << 20: /* 128 MB */
1929 return 7;
1930 case 256ul << 20: /* 256 MB */
1931 return 4;
1932 case 1ul << 30: /* 1 GB */
1933 return 2;
1934 case 16ul << 30: /* 16 GB */
1935 return 1;
1936 case 256ul << 30: /* 256 GB */
1937 return 0;
1938 default:
1939 return -1;
1940 }
1941}
1942
1943static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1944{
aa04b4cc 1945 struct page *page;
6c45b810 1946 struct kvm_rma_info *ri = vma->vm_file->private_data;
aa04b4cc 1947
6c45b810 1948 if (vmf->pgoff >= kvm_rma_pages)
aa04b4cc
PM		 1949 return VM_FAULT_SIGBUS;
1950
1951 page = pfn_to_page(ri->base_pfn + vmf->pgoff);
1952 get_page(page);
1953 vmf->page = page;
1954 return 0;
1955}
1956
1957static const struct vm_operations_struct kvm_rma_vm_ops = {
1958 .fault = kvm_rma_fault,
1959};
1960
1961static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma)
1962{
314e51b9 1963 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
aa04b4cc
PM		 1964 vma->vm_ops = &kvm_rma_vm_ops;
1965 return 0;
1966}
1967
1968static int kvm_rma_release(struct inode *inode, struct file *filp)
1969{
6c45b810 1970 struct kvm_rma_info *ri = filp->private_data;
aa04b4cc
PM		 1971
1972 kvm_release_rma(ri);
1973 return 0;
1974}
1975
75ef9de1 1976static const struct file_operations kvm_rma_fops = {
aa04b4cc
PM		 1977 .mmap = kvm_rma_mmap,
1978 .release = kvm_rma_release,
1979};
1980
3a167bea
AK		 1981static long kvm_vm_ioctl_allocate_rma(struct kvm *kvm,
1982 struct kvm_allocate_rma *ret)
aa04b4cc 1983{
aa04b4cc 1984 long fd;
6c45b810
AK		 1985 struct kvm_rma_info *ri;
1986 /*
1987 * Only do this on PPC970 in HV mode
1988 */
1989 if (!cpu_has_feature(CPU_FTR_HVMODE) ||
1990 !cpu_has_feature(CPU_FTR_ARCH_201))
1991 return -EINVAL;
1992
1993 if (!kvm_rma_pages)
1994 return -EINVAL;
aa04b4cc
PM		 1995
1996 ri = kvm_alloc_rma();
1997 if (!ri)
1998 return -ENOMEM;
1999
2f84d5ea 2000 fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR | O_CLOEXEC);
aa04b4cc
PM		 2001 if (fd < 0)
2002 kvm_release_rma(ri);
2003
6c45b810 2004 ret->rma_size = kvm_rma_pages << PAGE_SHIFT;
aa04b4cc
PM		 2005 return fd;
2006}
2007
5b74716e
BH		 2008static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
2009 int linux_psize)
2010{
2011 struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
2012
2013 if (!def->shift)
2014 return;
2015 (*sps)->page_shift = def->shift;
2016 (*sps)->slb_enc = def->sllp;
2017 (*sps)->enc[0].page_shift = def->shift;
b1022fbd
AK		 2018 /*
2019 * Only return base page encoding. We don't want to return
2020 * all the supporting pte_enc, because our H_ENTER doesn't
2021 * support MPSS yet. Once they do, we can start passing all
2022 * support pte_enc here
2023 */
2024 (*sps)->enc[0].pte_enc = def->penc[linux_psize];
1f365bb0
AK		 2025 /*
2026 * Add 16MB MPSS support if host supports it
2027 */
2028 if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
2029 (*sps)->enc[1].page_shift = 24;
2030 (*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
2031 }
5b74716e
BH		 2032 (*sps)++;
2033}
2034
3a167bea
AK		 2035static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
2036 struct kvm_ppc_smmu_info *info)
5b74716e
BH		 2037{
2038 struct kvm_ppc_one_seg_page_size *sps;
2039
2040 info->flags = KVM_PPC_PAGE_SIZES_REAL;
2041 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
2042 info->flags |= KVM_PPC_1T_SEGMENTS;
2043 info->slb_size = mmu_slb_size;
2044
2045 /* We only support these sizes for now, and no muti-size segments */
2046 sps = &info->sps[0];
2047 kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
2048 kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
2049 kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
2050
2051 return 0;
2052}
2053
82ed3616
PM		 2054/*
2055 * Get (and clear) the dirty memory log for a memory slot.
2056 */
3a167bea
AK		 2057static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
2058 struct kvm_dirty_log *log)
82ed3616
PM		 2059{
2060 struct kvm_memory_slot *memslot;
2061 int r;
2062 unsigned long n;
2063
2064 mutex_lock(&kvm->slots_lock);
2065
2066 r = -EINVAL;
bbacc0c1 2067 if (log->slot >= KVM_USER_MEM_SLOTS)
82ed3616
PM		 2068 goto out;
2069
2070 memslot = id_to_memslot(kvm->memslots, log->slot);
2071 r = -ENOENT;
2072 if (!memslot->dirty_bitmap)
2073 goto out;
2074
2075 n = kvm_dirty_bitmap_bytes(memslot);
2076 memset(memslot->dirty_bitmap, 0, n);
2077
dfe49dbd 2078 r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
82ed3616
PM		 2079 if (r)
2080 goto out;
2081
2082 r = -EFAULT;
2083 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
2084 goto out;
2085
2086 r = 0;
2087out:
2088 mutex_unlock(&kvm->slots_lock);
2089 return r;
2090}
2091
a66b48c3 2092static void unpin_slot(struct kvm_memory_slot *memslot)
de56a948 2093{
a66b48c3
PM		 2094 unsigned long *physp;
2095 unsigned long j, npages, pfn;
2096 struct page *page;
aa04b4cc 2097
a66b48c3
PM		 2098 physp = memslot->arch.slot_phys;
2099 npages = memslot->npages;
2100 if (!physp)
2101 return;
2102 for (j = 0; j < npages; j++) {
2103 if (!(physp[j] & KVMPPC_GOT_PAGE))
2104 continue;
2105 pfn = physp[j] >> PAGE_SHIFT;
2106 page = pfn_to_page(pfn);
2107 SetPageDirty(page);
2108 put_page(page);
2109 }
2110}
2111
3a167bea
AK		 2112static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
2113 struct kvm_memory_slot *dont)
a66b48c3
PM		 2114{
2115 if (!dont || free->arch.rmap != dont->arch.rmap) {
2116 vfree(free->arch.rmap);
2117 free->arch.rmap = NULL;
b2b2f165 2118 }
a66b48c3
PM		 2119 if (!dont || free->arch.slot_phys != dont->arch.slot_phys) {
2120 unpin_slot(free);
2121 vfree(free->arch.slot_phys);
2122 free->arch.slot_phys = NULL;
2123 }
2124}
2125
3a167bea
AK		 2126static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
2127 unsigned long npages)
a66b48c3
PM		 2128{
2129 slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
2130 if (!slot->arch.rmap)
2131 return -ENOMEM;
2132 slot->arch.slot_phys = NULL;
aa04b4cc 2133
c77162de
PM		 2134 return 0;
2135}
aa04b4cc 2136
3a167bea
AK		 2137static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
2138 struct kvm_memory_slot *memslot,
2139 struct kvm_userspace_memory_region *mem)
c77162de 2140{
a66b48c3 2141 unsigned long *phys;
c77162de 2142
a66b48c3
PM		 2143 /* Allocate a slot_phys array if needed */
2144 phys = memslot->arch.slot_phys;
2145 if (!kvm->arch.using_mmu_notifiers && !phys && memslot->npages) {
2146 phys = vzalloc(memslot->npages * sizeof(unsigned long));
2147 if (!phys)
2148 return -ENOMEM;
2149 memslot->arch.slot_phys = phys;
aa04b4cc 2150 }
a66b48c3
PM		 2151
2152 return 0;
c77162de
PM		 2153}
2154
3a167bea
AK		 2155static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
2156 struct kvm_userspace_memory_region *mem,
2157 const struct kvm_memory_slot *old)
c77162de 2158{
dfe49dbd
PM		 2159 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
2160 struct kvm_memory_slot *memslot;
2161
8482644a 2162 if (npages && old->npages) {
dfe49dbd
PM		 2163 /*
2164 * If modifying a memslot, reset all the rmap dirty bits.
2165 * If this is a new memslot, we don't need to do anything
2166 * since the rmap array starts out as all zeroes,
2167 * i.e. no pages are dirty.
2168 */
2169 memslot = id_to_memslot(kvm->memslots, mem->slot);
2170 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
2171 }
c77162de
PM		 2172}
2173
a0144e2a
PM		 2174/*
2175 * Update LPCR values in kvm->arch and in vcores.
2176 * Caller must hold kvm->lock.
2177 */
2178void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
2179{
2180 long int i;
2181 u32 cores_done = 0;
2182
2183 if ((kvm->arch.lpcr & mask) == lpcr)
2184 return;
2185
2186 kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
2187
2188 for (i = 0; i < KVM_MAX_VCORES; ++i) {
2189 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
2190 if (!vc)
2191 continue;
2192 spin_lock(&vc->lock);
2193 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
2194 spin_unlock(&vc->lock);
2195 if (++cores_done >= kvm->arch.online_vcores)
2196 break;
2197 }
2198}
2199
3a167bea
AK		 2200static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
2201{
2202 return;
2203}
2204
32fad281 2205static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
c77162de
PM		 2206{
2207 int err = 0;
2208 struct kvm *kvm = vcpu->kvm;
6c45b810 2209 struct kvm_rma_info *ri = NULL;
c77162de
PM		 2210 unsigned long hva;
2211 struct kvm_memory_slot *memslot;
2212 struct vm_area_struct *vma;
a0144e2a
PM		 2213 unsigned long lpcr = 0, senc;
2214 unsigned long lpcr_mask = 0;
c77162de
PM		 2215 unsigned long psize, porder;
2216 unsigned long rma_size;
2217 unsigned long rmls;
2218 unsigned long *physp;
da9d1d7f 2219 unsigned long i, npages;
2c9097e4 2220 int srcu_idx;
c77162de
PM		 2221
2222 mutex_lock(&kvm->lock);
2223 if (kvm->arch.rma_setup_done)
2224 goto out; /* another vcpu beat us to it */
aa04b4cc 2225
32fad281
PM		 2226 /* Allocate hashed page table (if not done already) and reset it */
2227 if (!kvm->arch.hpt_virt) {
2228 err = kvmppc_alloc_hpt(kvm, NULL);
2229 if (err) {
2230 pr_err("KVM: Couldn't alloc HPT\n");
2231 goto out;
2232 }
2233 }
2234
c77162de 2235 /* Look up the memslot for guest physical address 0 */
2c9097e4 2236 srcu_idx = srcu_read_lock(&kvm->srcu);
c77162de 2237 memslot = gfn_to_memslot(kvm, 0);
aa04b4cc 2238
c77162de
PM		 2239 /* We must have some memory at 0 by now */
2240 err = -EINVAL;
2241 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
2c9097e4 2242 goto out_srcu;
c77162de
PM		 2243
2244 /* Look up the VMA for the start of this memory slot */
2245 hva = memslot->userspace_addr;
2246 down_read(&current->mm->mmap_sem);
2247 vma = find_vma(current->mm, hva);
2248 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
2249 goto up_out;
2250
2251 psize = vma_kernel_pagesize(vma);
da9d1d7f 2252 porder = __ilog2(psize);
c77162de
PM		 2253
2254 /* Is this one of our preallocated RMAs? */
2255 if (vma->vm_file && vma->vm_file->f_op == &kvm_rma_fops &&
2256 hva == vma->vm_start)
2257 ri = vma->vm_file->private_data;
2258
2259 up_read(&current->mm->mmap_sem);
2260
2261 if (!ri) {
2262 /* On POWER7, use VRMA; on PPC970, give up */
2263 err = -EPERM;
2264 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2265 pr_err("KVM: CPU requires an RMO\n");
2c9097e4 2266 goto out_srcu;
c77162de
PM		 2267 }
2268
da9d1d7f
PM		 2269 /* We can handle 4k, 64k or 16M pages in the VRMA */
2270 err = -EINVAL;
2271 if (!(psize == 0x1000 || psize == 0x10000 ||
2272 psize == 0x1000000))
2c9097e4 2273 goto out_srcu;
da9d1d7f 2274
c77162de 2275 /* Update VRMASD field in the LPCR */
da9d1d7f 2276 senc = slb_pgsize_encoding(psize);
697d3899
PM		 2277 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
2278 (VRMA_VSID << SLB_VSID_SHIFT_1T);
a0144e2a
PM		 2279 lpcr_mask = LPCR_VRMASD;
2280 /* the -4 is to account for senc values starting at 0x10 */
2281 lpcr = senc << (LPCR_VRMASD_SH - 4);
c77162de
PM		 2282
2283 /* Create HPTEs in the hash page table for the VRMA */
da9d1d7f 2284 kvmppc_map_vrma(vcpu, memslot, porder);
c77162de
PM		 2285
2286 } else {
2287 /* Set up to use an RMO region */
6c45b810 2288 rma_size = kvm_rma_pages;
c77162de
PM		 2289 if (rma_size > memslot->npages)
2290 rma_size = memslot->npages;
2291 rma_size <<= PAGE_SHIFT;
aa04b4cc 2292 rmls = lpcr_rmls(rma_size);
c77162de 2293 err = -EINVAL;
5d226ae5 2294 if ((long)rmls < 0) {
c77162de 2295 pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size);
2c9097e4 2296 goto out_srcu;
aa04b4cc
PM		 2297 }
2298 atomic_inc(&ri->use_count);
2299 kvm->arch.rma = ri;
9e368f29
PM		 2300
2301 /* Update LPCR and RMOR */
9e368f29
PM		 2302 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2303 /* PPC970; insert RMLS value (split field) in HID4 */
a0144e2a
PM		 2304 lpcr_mask = (1ul << HID4_RMLS0_SH) |
2305 (3ul << HID4_RMLS2_SH) | HID4_RMOR;
2306 lpcr = ((rmls >> 2) << HID4_RMLS0_SH) |
9e368f29
PM		 2307 ((rmls & 3) << HID4_RMLS2_SH);
2308 /* RMOR is also in HID4 */
2309 lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff)
2310 << HID4_RMOR_SH;
2311 } else {
2312 /* POWER7 */
a0144e2a
PM		 2313 lpcr_mask = LPCR_VPM0 | LPCR_VRMA_L | LPCR_RMLS;
2314 lpcr = rmls << LPCR_RMLS_SH;
6c45b810 2315 kvm->arch.rmor = ri->base_pfn << PAGE_SHIFT;
9e368f29 2316 }
c77162de 2317 pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n",
aa04b4cc 2318 ri->base_pfn << PAGE_SHIFT, rma_size, lpcr);
aa04b4cc 2319
c77162de 2320 /* Initialize phys addrs of pages in RMO */
6c45b810 2321 npages = kvm_rma_pages;
da9d1d7f 2322 porder = __ilog2(npages);
a66b48c3
PM		 2323 physp = memslot->arch.slot_phys;
2324 if (physp) {
2325 if (npages > memslot->npages)
2326 npages = memslot->npages;
2327 spin_lock(&kvm->arch.slot_phys_lock);
2328 for (i = 0; i < npages; ++i)
2329 physp[i] = ((ri->base_pfn + i) << PAGE_SHIFT) +
2330 porder;
2331 spin_unlock(&kvm->arch.slot_phys_lock);
2332 }
aa04b4cc
PM		 2333 }
2334
a0144e2a
PM		 2335 kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
2336
c77162de
PM		 2337 /* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */
2338 smp_wmb();
2339 kvm->arch.rma_setup_done = 1;
2340 err = 0;
2c9097e4
PM		 2341 out_srcu:
2342 srcu_read_unlock(&kvm->srcu, srcu_idx);
c77162de
PM		 2343 out:
2344 mutex_unlock(&kvm->lock);
2345 return err;
b2b2f165 2346
c77162de
PM		 2347 up_out:
2348 up_read(&current->mm->mmap_sem);
505d6421 2349 goto out_srcu;
de56a948
PM		 2350}
2351
3a167bea 2352static int kvmppc_core_init_vm_hv(struct kvm *kvm)
de56a948 2353{
32fad281 2354 unsigned long lpcr, lpid;
de56a948 2355
32fad281
PM		 2356 /* Allocate the guest's logical partition ID */
2357
2358 lpid = kvmppc_alloc_lpid();
5d226ae5 2359 if ((long)lpid < 0)
32fad281
PM		 2360 return -ENOMEM;
2361 kvm->arch.lpid = lpid;
de56a948 2362
1b400ba0
PM		 2363 /*
2364 * Since we don't flush the TLB when tearing down a VM,
2365 * and this lpid might have previously been used,
2366 * make sure we flush on each core before running the new VM.
2367 */
2368 cpumask_setall(&kvm->arch.need_tlb_flush);
2369
699a0ea0
PM		 2370 /* Start out with the default set of hcalls enabled */
2371 memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
2372 sizeof(kvm->arch.enabled_hcalls));
2373
aa04b4cc 2374 kvm->arch.rma = NULL;
aa04b4cc 2375
9e368f29 2376 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
aa04b4cc 2377
9e368f29
PM		 2378 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2379 /* PPC970; HID4 is effectively the LPCR */
9e368f29
PM		 2380 kvm->arch.host_lpid = 0;
2381 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4);
2382 lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH));
2383 lpcr |= ((lpid >> 4) << HID4_LPID1_SH) |
2384 ((lpid & 0xf) << HID4_LPID5_SH);
2385 } else {
2386 /* POWER7; init LPCR for virtual RMA mode */
2387 kvm->arch.host_lpid = mfspr(SPRN_LPID);
2388 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
2389 lpcr &= LPCR_PECE | LPCR_LPES;
2390 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
697d3899
PM		 2391 LPCR_VPM0 | LPCR_VPM1;
2392 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
2393 (VRMA_VSID << SLB_VSID_SHIFT_1T);
e0622bd9
PM		 2394 /* On POWER8 turn on online bit to enable PURR/SPURR */
2395 if (cpu_has_feature(CPU_FTR_ARCH_207S))
2396 lpcr |= LPCR_ONL;
9e368f29
PM		 2397 }
2398 kvm->arch.lpcr = lpcr;
aa04b4cc 2399
342d3db7 2400 kvm->arch.using_mmu_notifiers = !!cpu_has_feature(CPU_FTR_ARCH_206);
c77162de 2401 spin_lock_init(&kvm->arch.slot_phys_lock);
512691d4
PM		 2402
2403 /*
441c19c8
ME		 2404 * Track that we now have a HV mode VM active. This blocks secondary
2405 * CPU threads from coming online.
512691d4 2406 */
441c19c8 2407 kvm_hv_vm_activated();
512691d4 2408
54738c09 2409 return 0;
de56a948
PM		 2410}
2411
f1378b1c
PM		 2412static void kvmppc_free_vcores(struct kvm *kvm)
2413{
2414 long int i;
2415
2416 for (i = 0; i < KVM_MAX_VCORES; ++i)
2417 kfree(kvm->arch.vcores[i]);
2418 kvm->arch.online_vcores = 0;
2419}
2420
3a167bea 2421static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
de56a948 2422{
441c19c8 2423 kvm_hv_vm_deactivated();
512691d4 2424
f1378b1c 2425 kvmppc_free_vcores(kvm);
aa04b4cc
PM		 2426 if (kvm->arch.rma) {
2427 kvm_release_rma(kvm->arch.rma);
2428 kvm->arch.rma = NULL;
2429 }
2430
de56a948
PM		 2431 kvmppc_free_hpt(kvm);
2432}
2433
3a167bea
AK		 2434/* We don't need to emulate any privileged instructions or dcbz */
2435static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
2436 unsigned int inst, int *advance)
de56a948 2437{
3a167bea 2438 return EMULATE_FAIL;
de56a948
PM		 2439}
2440
3a167bea
AK		 2441static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
2442 ulong spr_val)
de56a948
PM		 2443{
2444 return EMULATE_FAIL;
2445}
2446
3a167bea
AK		 2447static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
2448 ulong *spr_val)
de56a948
PM		 2449{
2450 return EMULATE_FAIL;
2451}
2452
3a167bea 2453static int kvmppc_core_check_processor_compat_hv(void)
de56a948 2454{
3a167bea
AK		 2455 if (!cpu_has_feature(CPU_FTR_HVMODE))
2456 return -EIO;
2457 return 0;
de56a948
PM		 2458}
2459
3a167bea
AK		 2460static long kvm_arch_vm_ioctl_hv(struct file *filp,
2461 unsigned int ioctl, unsigned long arg)
2462{
2463 struct kvm *kvm __maybe_unused = filp->private_data;
2464 void __user *argp = (void __user *)arg;
2465 long r;
2466
2467 switch (ioctl) {
2468
2469 case KVM_ALLOCATE_RMA: {
2470 struct kvm_allocate_rma rma;
2471 struct kvm *kvm = filp->private_data;
2472
2473 r = kvm_vm_ioctl_allocate_rma(kvm, &rma);
2474 if (r >= 0 && copy_to_user(argp, &rma, sizeof(rma)))
2475 r = -EFAULT;
2476 break;
2477 }
2478
2479 case KVM_PPC_ALLOCATE_HTAB: {
2480 u32 htab_order;
2481
2482 r = -EFAULT;
2483 if (get_user(htab_order, (u32 __user *)argp))
2484 break;
2485 r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
2486 if (r)
2487 break;
2488 r = -EFAULT;
2489 if (put_user(htab_order, (u32 __user *)argp))
2490 break;
2491 r = 0;
2492 break;
2493 }
2494
2495 case KVM_PPC_GET_HTAB_FD: {
2496 struct kvm_get_htab_fd ghf;
2497
2498 r = -EFAULT;
2499 if (copy_from_user(&ghf, argp, sizeof(ghf)))
2500 break;
2501 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
2502 break;
2503 }
2504
2505 default:
2506 r = -ENOTTY;
2507 }
2508
2509 return r;
2510}
2511
699a0ea0
PM		 2512/*
2513 * List of hcall numbers to enable by default.
2514 * For compatibility with old userspace, we enable by default
2515 * all hcalls that were implemented before the hcall-enabling
2516 * facility was added. Note this list should not include H_RTAS.
2517 */
2518static unsigned int default_hcall_list[] = {
2519 H_REMOVE,
2520 H_ENTER,
2521 H_READ,
2522 H_PROTECT,
2523 H_BULK_REMOVE,
2524 H_GET_TCE,
2525 H_PUT_TCE,
2526 H_SET_DABR,
2527 H_SET_XDABR,
2528 H_CEDE,
2529 H_PROD,
2530 H_CONFER,
2531 H_REGISTER_VPA,
2532#ifdef CONFIG_KVM_XICS
2533 H_EOI,
2534 H_CPPR,
2535 H_IPI,
2536 H_IPOLL,
2537 H_XIRR,
2538 H_XIRR_X,
2539#endif
2540 0
2541};
2542
2543static void init_default_hcalls(void)
2544{
2545 int i;
ae2113a4 2546 unsigned int hcall;
699a0ea0 2547
ae2113a4
PM		 2548 for (i = 0; default_hcall_list[i]; ++i) {
2549 hcall = default_hcall_list[i];
2550 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
2551 __set_bit(hcall / 4, default_enabled_hcalls);
2552 }
699a0ea0
PM		 2553}
2554
cbbc58d4 2555static struct kvmppc_ops kvm_ops_hv = {
3a167bea
AK		 2556 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
2557 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
2558 .get_one_reg = kvmppc_get_one_reg_hv,
2559 .set_one_reg = kvmppc_set_one_reg_hv,
2560 .vcpu_load = kvmppc_core_vcpu_load_hv,
2561 .vcpu_put = kvmppc_core_vcpu_put_hv,
2562 .set_msr = kvmppc_set_msr_hv,
2563 .vcpu_run = kvmppc_vcpu_run_hv,
2564 .vcpu_create = kvmppc_core_vcpu_create_hv,
2565 .vcpu_free = kvmppc_core_vcpu_free_hv,
2566 .check_requests = kvmppc_core_check_requests_hv,
2567 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv,
2568 .flush_memslot = kvmppc_core_flush_memslot_hv,
2569 .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
2570 .commit_memory_region = kvmppc_core_commit_memory_region_hv,
2571 .unmap_hva = kvm_unmap_hva_hv,
2572 .unmap_hva_range = kvm_unmap_hva_range_hv,
2573 .age_hva = kvm_age_hva_hv,
2574 .test_age_hva = kvm_test_age_hva_hv,
2575 .set_spte_hva = kvm_set_spte_hva_hv,
2576 .mmu_destroy = kvmppc_mmu_destroy_hv,
2577 .free_memslot = kvmppc_core_free_memslot_hv,
2578 .create_memslot = kvmppc_core_create_memslot_hv,
2579 .init_vm = kvmppc_core_init_vm_hv,
2580 .destroy_vm = kvmppc_core_destroy_vm_hv,
3a167bea
AK		 2581 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
2582 .emulate_op = kvmppc_core_emulate_op_hv,
2583 .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
2584 .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
2585 .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
2586 .arch_vm_ioctl = kvm_arch_vm_ioctl_hv,
ae2113a4 2587 .hcall_implemented = kvmppc_hcall_impl_hv,
3a167bea
AK		 2588};
2589
2590static int kvmppc_book3s_init_hv(void)
de56a948
PM		 2591{
2592 int r;
cbbc58d4
AK		 2593 /*
2594 * FIXME!! Do we need to check on all cpus ?
2595 */
2596 r = kvmppc_core_check_processor_compat_hv();
2597 if (r < 0)
739e2425 2598 return -ENODEV;
de56a948 2599
cbbc58d4
AK		 2600 kvm_ops_hv.owner = THIS_MODULE;
2601 kvmppc_hv_ops = &kvm_ops_hv;
de56a948 2602
699a0ea0
PM		 2603 init_default_hcalls();
2604
cbbc58d4 2605 r = kvmppc_mmu_hv_init();
de56a948
PM		 2606 return r;
2607}
2608
3a167bea 2609static void kvmppc_book3s_exit_hv(void)
de56a948 2610{
cbbc58d4 2611 kvmppc_hv_ops = NULL;
de56a948
PM		 2612}
2613
3a167bea
AK		 2614module_init(kvmppc_book3s_init_hv);
2615module_exit(kvmppc_book3s_exit_hv);
2ba9f0d8 2616MODULE_LICENSE("GPL");
398a76c6
AG		 2617MODULE_ALIAS_MISCDEV(KVM_MINOR);
2618MODULE_ALIAS("devname:kvm");
Linux 6.x block layer and io_uring tree(s)