projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: PPC: Remove 440 support
[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
a0144e2a
PM		 866static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr)
867{
868 struct kvmppc_vcore *vc = vcpu->arch.vcore;
869 u64 mask;
870
871 spin_lock(&vc->lock);
d682916a
AB		 872 /*
873 * If ILE (interrupt little-endian) has changed, update the
874 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
875 */
876 if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
877 struct kvm *kvm = vcpu->kvm;
878 struct kvm_vcpu *vcpu;
879 int i;
880
881 mutex_lock(&kvm->lock);
882 kvm_for_each_vcpu(i, vcpu, kvm) {
883 if (vcpu->arch.vcore != vc)
884 continue;
885 if (new_lpcr & LPCR_ILE)
886 vcpu->arch.intr_msr |= MSR_LE;
887 else
888 vcpu->arch.intr_msr &= ~MSR_LE;
889 }
890 mutex_unlock(&kvm->lock);
891 }
892
a0144e2a
PM		 893 /*
894 * Userspace can only modify DPFD (default prefetch depth),
895 * ILE (interrupt little-endian) and TC (translation control).
e0622bd9 896 * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
a0144e2a
PM		 897 */
898 mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
e0622bd9
PM		 899 if (cpu_has_feature(CPU_FTR_ARCH_207S))
900 mask |= LPCR_AIL;
a0144e2a
PM		 901 vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
902 spin_unlock(&vc->lock);
903}
904
3a167bea
AK		 905static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
906 union kvmppc_one_reg *val)
31f3438e 907{
a136a8bd
PM		 908 int r = 0;
909 long int i;
31f3438e 910
a136a8bd 911 switch (id) {
31f3438e 912 case KVM_REG_PPC_HIOR:
a136a8bd
PM		 913 *val = get_reg_val(id, 0);
914 break;
915 case KVM_REG_PPC_DABR:
916 *val = get_reg_val(id, vcpu->arch.dabr);
917 break;
8563bf52
PM		 918 case KVM_REG_PPC_DABRX:
919 *val = get_reg_val(id, vcpu->arch.dabrx);
920 break;
a136a8bd
PM		 921 case KVM_REG_PPC_DSCR:
922 *val = get_reg_val(id, vcpu->arch.dscr);
923 break;
924 case KVM_REG_PPC_PURR:
925 *val = get_reg_val(id, vcpu->arch.purr);
926 break;
927 case KVM_REG_PPC_SPURR:
928 *val = get_reg_val(id, vcpu->arch.spurr);
929 break;
930 case KVM_REG_PPC_AMR:
931 *val = get_reg_val(id, vcpu->arch.amr);
932 break;
933 case KVM_REG_PPC_UAMOR:
934 *val = get_reg_val(id, vcpu->arch.uamor);
935 break;
b005255e 936 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
a136a8bd
PM		 937 i = id - KVM_REG_PPC_MMCR0;
938 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
939 break;
940 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
941 i = id - KVM_REG_PPC_PMC1;
942 *val = get_reg_val(id, vcpu->arch.pmc[i]);
31f3438e 943 break;
b005255e
MN		 944 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
945 i = id - KVM_REG_PPC_SPMC1;
946 *val = get_reg_val(id, vcpu->arch.spmc[i]);
947 break;
14941789
PM		 948 case KVM_REG_PPC_SIAR:
949 *val = get_reg_val(id, vcpu->arch.siar);
950 break;
951 case KVM_REG_PPC_SDAR:
952 *val = get_reg_val(id, vcpu->arch.sdar);
953 break;
b005255e
MN		 954 case KVM_REG_PPC_SIER:
955 *val = get_reg_val(id, vcpu->arch.sier);
a8bd19ef 956 break;
b005255e
MN		 957 case KVM_REG_PPC_IAMR:
958 *val = get_reg_val(id, vcpu->arch.iamr);
959 break;
b005255e
MN		 960 case KVM_REG_PPC_PSPB:
961 *val = get_reg_val(id, vcpu->arch.pspb);
962 break;
b005255e
MN		 963 case KVM_REG_PPC_DPDES:
964 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
965 break;
966 case KVM_REG_PPC_DAWR:
967 *val = get_reg_val(id, vcpu->arch.dawr);
968 break;
969 case KVM_REG_PPC_DAWRX:
970 *val = get_reg_val(id, vcpu->arch.dawrx);
971 break;
972 case KVM_REG_PPC_CIABR:
973 *val = get_reg_val(id, vcpu->arch.ciabr);
974 break;
b005255e
MN		 975 case KVM_REG_PPC_CSIGR:
976 *val = get_reg_val(id, vcpu->arch.csigr);
977 break;
978 case KVM_REG_PPC_TACR:
979 *val = get_reg_val(id, vcpu->arch.tacr);
980 break;
981 case KVM_REG_PPC_TCSCR:
982 *val = get_reg_val(id, vcpu->arch.tcscr);
983 break;
984 case KVM_REG_PPC_PID:
985 *val = get_reg_val(id, vcpu->arch.pid);
986 break;
987 case KVM_REG_PPC_ACOP:
988 *val = get_reg_val(id, vcpu->arch.acop);
989 break;
990 case KVM_REG_PPC_WORT:
991 *val = get_reg_val(id, vcpu->arch.wort);
a8bd19ef 992 break;
55b665b0
PM		 993 case KVM_REG_PPC_VPA_ADDR:
994 spin_lock(&vcpu->arch.vpa_update_lock);
995 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
996 spin_unlock(&vcpu->arch.vpa_update_lock);
997 break;
998 case KVM_REG_PPC_VPA_SLB:
999 spin_lock(&vcpu->arch.vpa_update_lock);
1000 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1001 val->vpaval.length = vcpu->arch.slb_shadow.len;
1002 spin_unlock(&vcpu->arch.vpa_update_lock);
1003 break;
1004 case KVM_REG_PPC_VPA_DTL:
1005 spin_lock(&vcpu->arch.vpa_update_lock);
1006 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1007 val->vpaval.length = vcpu->arch.dtl.len;
1008 spin_unlock(&vcpu->arch.vpa_update_lock);
1009 break;
93b0f4dc
PM		 1010 case KVM_REG_PPC_TB_OFFSET:
1011 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1012 break;
a0144e2a
PM		 1013 case KVM_REG_PPC_LPCR:
1014 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1015 break;
4b8473c9
PM		 1016 case KVM_REG_PPC_PPR:
1017 *val = get_reg_val(id, vcpu->arch.ppr);
1018 break;
a7d80d01
MN		 1019#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1020 case KVM_REG_PPC_TFHAR:
1021 *val = get_reg_val(id, vcpu->arch.tfhar);
1022 break;
1023 case KVM_REG_PPC_TFIAR:
1024 *val = get_reg_val(id, vcpu->arch.tfiar);
1025 break;
1026 case KVM_REG_PPC_TEXASR:
1027 *val = get_reg_val(id, vcpu->arch.texasr);
1028 break;
1029 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1030 i = id - KVM_REG_PPC_TM_GPR0;
1031 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1032 break;
1033 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1034 {
1035 int j;
1036 i = id - KVM_REG_PPC_TM_VSR0;
1037 if (i < 32)
1038 for (j = 0; j < TS_FPRWIDTH; j++)
1039 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1040 else {
1041 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1042 val->vval = vcpu->arch.vr_tm.vr[i-32];
1043 else
1044 r = -ENXIO;
1045 }
1046 break;
1047 }
1048 case KVM_REG_PPC_TM_CR:
1049 *val = get_reg_val(id, vcpu->arch.cr_tm);
1050 break;
1051 case KVM_REG_PPC_TM_LR:
1052 *val = get_reg_val(id, vcpu->arch.lr_tm);
1053 break;
1054 case KVM_REG_PPC_TM_CTR:
1055 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1056 break;
1057 case KVM_REG_PPC_TM_FPSCR:
1058 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1059 break;
1060 case KVM_REG_PPC_TM_AMR:
1061 *val = get_reg_val(id, vcpu->arch.amr_tm);
1062 break;
1063 case KVM_REG_PPC_TM_PPR:
1064 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1065 break;
1066 case KVM_REG_PPC_TM_VRSAVE:
1067 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1068 break;
1069 case KVM_REG_PPC_TM_VSCR:
1070 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1071 *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1072 else
1073 r = -ENXIO;
1074 break;
1075 case KVM_REG_PPC_TM_DSCR:
1076 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1077 break;
1078 case KVM_REG_PPC_TM_TAR:
1079 *val = get_reg_val(id, vcpu->arch.tar_tm);
1080 break;
1081#endif
388cc6e1
PM		 1082 case KVM_REG_PPC_ARCH_COMPAT:
1083 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1084 break;
31f3438e 1085 default:
a136a8bd 1086 r = -EINVAL;
31f3438e
PM		 1087 break;
1088 }
1089
1090 return r;
1091}
1092
3a167bea
AK		 1093static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1094 union kvmppc_one_reg *val)
31f3438e 1095{
a136a8bd
PM		 1096 int r = 0;
1097 long int i;
55b665b0 1098 unsigned long addr, len;
31f3438e 1099
a136a8bd 1100 switch (id) {
31f3438e 1101 case KVM_REG_PPC_HIOR:
31f3438e 1102 /* Only allow this to be set to zero */
a136a8bd 1103 if (set_reg_val(id, *val))
31f3438e
PM		 1104 r = -EINVAL;
1105 break;
a136a8bd
PM		 1106 case KVM_REG_PPC_DABR:
1107 vcpu->arch.dabr = set_reg_val(id, *val);
1108 break;
8563bf52
PM		 1109 case KVM_REG_PPC_DABRX:
1110 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1111 break;
a136a8bd
PM		 1112 case KVM_REG_PPC_DSCR:
1113 vcpu->arch.dscr = set_reg_val(id, *val);
1114 break;
1115 case KVM_REG_PPC_PURR:
1116 vcpu->arch.purr = set_reg_val(id, *val);
1117 break;
1118 case KVM_REG_PPC_SPURR:
1119 vcpu->arch.spurr = set_reg_val(id, *val);
1120 break;
1121 case KVM_REG_PPC_AMR:
1122 vcpu->arch.amr = set_reg_val(id, *val);
1123 break;
1124 case KVM_REG_PPC_UAMOR:
1125 vcpu->arch.uamor = set_reg_val(id, *val);
1126 break;
b005255e 1127 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
a136a8bd
PM		 1128 i = id - KVM_REG_PPC_MMCR0;
1129 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1130 break;
1131 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1132 i = id - KVM_REG_PPC_PMC1;
1133 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1134 break;
b005255e
MN		 1135 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1136 i = id - KVM_REG_PPC_SPMC1;
1137 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1138 break;
14941789
PM		 1139 case KVM_REG_PPC_SIAR:
1140 vcpu->arch.siar = set_reg_val(id, *val);
1141 break;
1142 case KVM_REG_PPC_SDAR:
1143 vcpu->arch.sdar = set_reg_val(id, *val);
1144 break;
b005255e
MN		 1145 case KVM_REG_PPC_SIER:
1146 vcpu->arch.sier = set_reg_val(id, *val);
a8bd19ef 1147 break;
b005255e
MN		 1148 case KVM_REG_PPC_IAMR:
1149 vcpu->arch.iamr = set_reg_val(id, *val);
1150 break;
b005255e
MN		 1151 case KVM_REG_PPC_PSPB:
1152 vcpu->arch.pspb = set_reg_val(id, *val);
1153 break;
b005255e
MN		 1154 case KVM_REG_PPC_DPDES:
1155 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1156 break;
1157 case KVM_REG_PPC_DAWR:
1158 vcpu->arch.dawr = set_reg_val(id, *val);
1159 break;
1160 case KVM_REG_PPC_DAWRX:
1161 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1162 break;
1163 case KVM_REG_PPC_CIABR:
1164 vcpu->arch.ciabr = set_reg_val(id, *val);
1165 /* Don't allow setting breakpoints in hypervisor code */
1166 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1167 vcpu->arch.ciabr &= ~CIABR_PRIV; /* disable */
1168 break;
b005255e
MN		 1169 case KVM_REG_PPC_CSIGR:
1170 vcpu->arch.csigr = set_reg_val(id, *val);
1171 break;
1172 case KVM_REG_PPC_TACR:
1173 vcpu->arch.tacr = set_reg_val(id, *val);
1174 break;
1175 case KVM_REG_PPC_TCSCR:
1176 vcpu->arch.tcscr = set_reg_val(id, *val);
1177 break;
1178 case KVM_REG_PPC_PID:
1179 vcpu->arch.pid = set_reg_val(id, *val);
1180 break;
1181 case KVM_REG_PPC_ACOP:
1182 vcpu->arch.acop = set_reg_val(id, *val);
1183 break;
1184 case KVM_REG_PPC_WORT:
1185 vcpu->arch.wort = set_reg_val(id, *val);
a8bd19ef 1186 break;
55b665b0
PM		 1187 case KVM_REG_PPC_VPA_ADDR:
1188 addr = set_reg_val(id, *val);
1189 r = -EINVAL;
1190 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1191 vcpu->arch.dtl.next_gpa))
1192 break;
1193 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1194 break;
1195 case KVM_REG_PPC_VPA_SLB:
1196 addr = val->vpaval.addr;
1197 len = val->vpaval.length;
1198 r = -EINVAL;
1199 if (addr && !vcpu->arch.vpa.next_gpa)
1200 break;
1201 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1202 break;
1203 case KVM_REG_PPC_VPA_DTL:
1204 addr = val->vpaval.addr;
1205 len = val->vpaval.length;
1206 r = -EINVAL;
9f8c8c78
PM		 1207 if (addr && (len < sizeof(struct dtl_entry) ||
1208 !vcpu->arch.vpa.next_gpa))
55b665b0
PM		 1209 break;
1210 len -= len % sizeof(struct dtl_entry);
1211 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1212 break;
93b0f4dc
PM		 1213 case KVM_REG_PPC_TB_OFFSET:
1214 /* round up to multiple of 2^24 */
1215 vcpu->arch.vcore->tb_offset =
1216 ALIGN(set_reg_val(id, *val), 1UL << 24);
1217 break;
a0144e2a
PM		 1218 case KVM_REG_PPC_LPCR:
1219 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val));
1220 break;
4b8473c9
PM		 1221 case KVM_REG_PPC_PPR:
1222 vcpu->arch.ppr = set_reg_val(id, *val);
1223 break;
a7d80d01
MN		 1224#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1225 case KVM_REG_PPC_TFHAR:
1226 vcpu->arch.tfhar = set_reg_val(id, *val);
1227 break;
1228 case KVM_REG_PPC_TFIAR:
1229 vcpu->arch.tfiar = set_reg_val(id, *val);
1230 break;
1231 case KVM_REG_PPC_TEXASR:
1232 vcpu->arch.texasr = set_reg_val(id, *val);
1233 break;
1234 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1235 i = id - KVM_REG_PPC_TM_GPR0;
1236 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1237 break;
1238 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1239 {
1240 int j;
1241 i = id - KVM_REG_PPC_TM_VSR0;
1242 if (i < 32)
1243 for (j = 0; j < TS_FPRWIDTH; j++)
1244 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1245 else
1246 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1247 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1248 else
1249 r = -ENXIO;
1250 break;
1251 }
1252 case KVM_REG_PPC_TM_CR:
1253 vcpu->arch.cr_tm = set_reg_val(id, *val);
1254 break;
1255 case KVM_REG_PPC_TM_LR:
1256 vcpu->arch.lr_tm = set_reg_val(id, *val);
1257 break;
1258 case KVM_REG_PPC_TM_CTR:
1259 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1260 break;
1261 case KVM_REG_PPC_TM_FPSCR:
1262 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1263 break;
1264 case KVM_REG_PPC_TM_AMR:
1265 vcpu->arch.amr_tm = set_reg_val(id, *val);
1266 break;
1267 case KVM_REG_PPC_TM_PPR:
1268 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1269 break;
1270 case KVM_REG_PPC_TM_VRSAVE:
1271 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1272 break;
1273 case KVM_REG_PPC_TM_VSCR:
1274 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1275 vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1276 else
1277 r = - ENXIO;
1278 break;
1279 case KVM_REG_PPC_TM_DSCR:
1280 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1281 break;
1282 case KVM_REG_PPC_TM_TAR:
1283 vcpu->arch.tar_tm = set_reg_val(id, *val);
1284 break;
1285#endif
388cc6e1
PM		 1286 case KVM_REG_PPC_ARCH_COMPAT:
1287 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1288 break;
31f3438e 1289 default:
a136a8bd 1290 r = -EINVAL;
31f3438e
PM		 1291 break;
1292 }
1293
1294 return r;
1295}
1296
3a167bea
AK		 1297static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1298 unsigned int id)
de56a948
PM		 1299{
1300 struct kvm_vcpu *vcpu;
371fefd6
PM		 1301 int err = -EINVAL;
1302 int core;
1303 struct kvmppc_vcore *vcore;
de56a948 1304
3102f784 1305 core = id / threads_per_subcore;
371fefd6
PM		 1306 if (core >= KVM_MAX_VCORES)
1307 goto out;
1308
1309 err = -ENOMEM;
6b75e6bf 1310 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
de56a948
PM		 1311 if (!vcpu)
1312 goto out;
1313
1314 err = kvm_vcpu_init(vcpu, kvm, id);
1315 if (err)
1316 goto free_vcpu;
1317
1318 vcpu->arch.shared = &vcpu->arch.shregs;
5deb8e7a
AG		 1319#ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1320 /*
1321 * The shared struct is never shared on HV,
1322 * so we can always use host endianness
1323 */
1324#ifdef __BIG_ENDIAN__
1325 vcpu->arch.shared_big_endian = true;
1326#else
1327 vcpu->arch.shared_big_endian = false;
1328#endif
1329#endif
de56a948
PM		 1330 vcpu->arch.mmcr[0] = MMCR0_FC;
1331 vcpu->arch.ctrl = CTRL_RUNLATCH;
1332 /* default to host PVR, since we can't spoof it */
3a167bea 1333 kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2e25aa5f 1334 spin_lock_init(&vcpu->arch.vpa_update_lock);
c7b67670
PM		 1335 spin_lock_init(&vcpu->arch.tbacct_lock);
1336 vcpu->arch.busy_preempt = TB_NIL;
d682916a 1337 vcpu->arch.intr_msr = MSR_SF | MSR_ME;
de56a948 1338
de56a948
PM		 1339 kvmppc_mmu_book3s_hv_init(vcpu);
1340
8455d79e 1341 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
371fefd6
PM		 1342
1343 init_waitqueue_head(&vcpu->arch.cpu_run);
1344
1345 mutex_lock(&kvm->lock);
1346 vcore = kvm->arch.vcores[core];
1347 if (!vcore) {
1348 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1349 if (vcore) {
1350 INIT_LIST_HEAD(&vcore->runnable_threads);
1351 spin_lock_init(&vcore->lock);
19ccb76a 1352 init_waitqueue_head(&vcore->wq);
c7b67670 1353 vcore->preempt_tb = TB_NIL;
a0144e2a 1354 vcore->lpcr = kvm->arch.lpcr;
3102f784 1355 vcore->first_vcpuid = core * threads_per_subcore;
e0b7ec05 1356 vcore->kvm = kvm;
371fefd6
PM		 1357 }
1358 kvm->arch.vcores[core] = vcore;
1b400ba0 1359 kvm->arch.online_vcores++;
371fefd6
PM		 1360 }
1361 mutex_unlock(&kvm->lock);
1362
1363 if (!vcore)
1364 goto free_vcpu;
1365
1366 spin_lock(&vcore->lock);
1367 ++vcore->num_threads;
371fefd6
PM		 1368 spin_unlock(&vcore->lock);
1369 vcpu->arch.vcore = vcore;
e0b7ec05 1370 vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
371fefd6 1371
af8f38b3
AG		 1372 vcpu->arch.cpu_type = KVM_CPU_3S_64;
1373 kvmppc_sanity_check(vcpu);
1374
de56a948
PM		 1375 return vcpu;
1376
1377free_vcpu:
6b75e6bf 1378 kmem_cache_free(kvm_vcpu_cache, vcpu);
de56a948
PM		 1379out:
1380 return ERR_PTR(err);
1381}
1382
c35635ef
PM		 1383static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1384{
1385 if (vpa->pinned_addr)
1386 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1387 vpa->dirty);
1388}
1389
3a167bea 1390static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
de56a948 1391{
2e25aa5f 1392 spin_lock(&vcpu->arch.vpa_update_lock);
c35635ef
PM		 1393 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1394 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1395 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2e25aa5f 1396 spin_unlock(&vcpu->arch.vpa_update_lock);
de56a948 1397 kvm_vcpu_uninit(vcpu);
6b75e6bf 1398 kmem_cache_free(kvm_vcpu_cache, vcpu);
de56a948
PM		 1399}
1400
3a167bea
AK		 1401static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1402{
1403 /* Indicate we want to get back into the guest */
1404 return 1;
1405}
1406
19ccb76a 1407static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
371fefd6 1408{
19ccb76a 1409 unsigned long dec_nsec, now;
371fefd6 1410
19ccb76a
PM		 1411 now = get_tb();
1412 if (now > vcpu->arch.dec_expires) {
1413 /* decrementer has already gone negative */
1414 kvmppc_core_queue_dec(vcpu);
7e28e60e 1415 kvmppc_core_prepare_to_enter(vcpu);
19ccb76a 1416 return;
371fefd6 1417 }
19ccb76a
PM		 1418 dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1419 / tb_ticks_per_sec;
1420 hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1421 HRTIMER_MODE_REL);
1422 vcpu->arch.timer_running = 1;
371fefd6
PM		 1423}
1424
19ccb76a 1425static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
371fefd6 1426{
19ccb76a
PM		 1427 vcpu->arch.ceded = 0;
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}
1433
e0b7ec05 1434extern void __kvmppc_vcore_entry(void);
de56a948 1435
371fefd6
PM		 1436static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1437 struct kvm_vcpu *vcpu)
de56a948 1438{
c7b67670
PM		 1439 u64 now;
1440
371fefd6
PM		 1441 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1442 return;
bf3d32e1 1443 spin_lock_irq(&vcpu->arch.tbacct_lock);
c7b67670
PM		 1444 now = mftb();
1445 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1446 vcpu->arch.stolen_logged;
1447 vcpu->arch.busy_preempt = now;
1448 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
bf3d32e1 1449 spin_unlock_irq(&vcpu->arch.tbacct_lock);
371fefd6 1450 --vc->n_runnable;
371fefd6
PM		 1451 list_del(&vcpu->arch.run_list);
1452}
1453
f0888f70
PM		 1454static int kvmppc_grab_hwthread(int cpu)
1455{
1456 struct paca_struct *tpaca;
1457 long timeout = 1000;
1458
1459 tpaca = &paca[cpu];
1460
1461 /* Ensure the thread won't go into the kernel if it wakes */
1462 tpaca->kvm_hstate.hwthread_req = 1;
7b444c67 1463 tpaca->kvm_hstate.kvm_vcpu = NULL;
f0888f70
PM		 1464
1465 /*
1466 * If the thread is already executing in the kernel (e.g. handling
1467 * a stray interrupt), wait for it to get back to nap mode.
1468 * The smp_mb() is to ensure that our setting of hwthread_req
1469 * is visible before we look at hwthread_state, so if this
1470 * races with the code at system_reset_pSeries and the thread
1471 * misses our setting of hwthread_req, we are sure to see its
1472 * setting of hwthread_state, and vice versa.
1473 */
1474 smp_mb();
1475 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1476 if (--timeout <= 0) {
1477 pr_err("KVM: couldn't grab cpu %d\n", cpu);
1478 return -EBUSY;
1479 }
1480 udelay(1);
1481 }
1482 return 0;
1483}
1484
1485static void kvmppc_release_hwthread(int cpu)
1486{
1487 struct paca_struct *tpaca;
1488
1489 tpaca = &paca[cpu];
1490 tpaca->kvm_hstate.hwthread_req = 0;
1491 tpaca->kvm_hstate.kvm_vcpu = NULL;
1492}
1493
371fefd6
PM		 1494static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
1495{
1496 int cpu;
1497 struct paca_struct *tpaca;
1498 struct kvmppc_vcore *vc = vcpu->arch.vcore;
1499
19ccb76a
PM		 1500 if (vcpu->arch.timer_running) {
1501 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1502 vcpu->arch.timer_running = 0;
1503 }
371fefd6
PM		 1504 cpu = vc->pcpu + vcpu->arch.ptid;
1505 tpaca = &paca[cpu];
1506 tpaca->kvm_hstate.kvm_vcpu = vcpu;
1507 tpaca->kvm_hstate.kvm_vcore = vc;
e0b7ec05 1508 tpaca->kvm_hstate.ptid = vcpu->arch.ptid;
19ccb76a 1509 vcpu->cpu = vc->pcpu;
371fefd6 1510 smp_wmb();
251da038 1511#if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
e0b7ec05 1512 if (cpu != smp_processor_id()) {
371fefd6 1513 xics_wake_cpu(cpu);
e0b7ec05
PM		 1514 if (vcpu->arch.ptid)
1515 ++vc->n_woken;
de56a948 1516 }
371fefd6
PM		 1517#endif
1518}
de56a948 1519
371fefd6
PM		 1520static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc)
1521{
1522 int i;
1523
1524 HMT_low();
1525 i = 0;
1526 while (vc->nap_count < vc->n_woken) {
1527 if (++i >= 1000000) {
1528 pr_err("kvmppc_wait_for_nap timeout %d %d\n",
1529 vc->nap_count, vc->n_woken);
1530 break;
1531 }
1532 cpu_relax();
1533 }
1534 HMT_medium();
1535}
1536
1537/*
1538 * Check that we are on thread 0 and that any other threads in
7b444c67
PM		 1539 * this core are off-line. Then grab the threads so they can't
1540 * enter the kernel.
371fefd6
PM		 1541 */
1542static int on_primary_thread(void)
1543{
1544 int cpu = smp_processor_id();
3102f784 1545 int thr;
371fefd6 1546
3102f784
ME		 1547 /* Are we on a primary subcore? */
1548 if (cpu_thread_in_subcore(cpu))
371fefd6 1549 return 0;
3102f784
ME		 1550
1551 thr = 0;
1552 while (++thr < threads_per_subcore)
371fefd6
PM		 1553 if (cpu_online(cpu + thr))
1554 return 0;
7b444c67
PM		 1555
1556 /* Grab all hw threads so they can't go into the kernel */
3102f784 1557 for (thr = 1; thr < threads_per_subcore; ++thr) {
7b444c67
PM		 1558 if (kvmppc_grab_hwthread(cpu + thr)) {
1559 /* Couldn't grab one; let the others go */
1560 do {
1561 kvmppc_release_hwthread(cpu + thr);
1562 } while (--thr > 0);
1563 return 0;
1564 }
1565 }
371fefd6
PM		 1566 return 1;
1567}
1568
1569/*
1570 * Run a set of guest threads on a physical core.
1571 * Called with vc->lock held.
1572 */
913d3ff9 1573static void kvmppc_run_core(struct kvmppc_vcore *vc)
371fefd6 1574{
e0b7ec05 1575 struct kvm_vcpu *vcpu, *vnext;
371fefd6
PM		 1576 long ret;
1577 u64 now;
e0b7ec05 1578 int i, need_vpa_update;
2c9097e4 1579 int srcu_idx;
913d3ff9 1580 struct kvm_vcpu *vcpus_to_update[threads_per_core];
371fefd6
PM		 1581
1582 /* don't start if any threads have a signal pending */
081f323b
PM		 1583 need_vpa_update = 0;
1584 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
371fefd6 1585 if (signal_pending(vcpu->arch.run_task))
913d3ff9
PM		 1586 return;
1587 if (vcpu->arch.vpa.update_pending ||
1588 vcpu->arch.slb_shadow.update_pending ||
1589 vcpu->arch.dtl.update_pending)
1590 vcpus_to_update[need_vpa_update++] = vcpu;
081f323b
PM		 1591 }
1592
1593 /*
1594 * Initialize *vc, in particular vc->vcore_state, so we can
1595 * drop the vcore lock if necessary.
1596 */
1597 vc->n_woken = 0;
1598 vc->nap_count = 0;
1599 vc->entry_exit_count = 0;
2f12f034 1600 vc->vcore_state = VCORE_STARTING;
081f323b
PM		 1601 vc->in_guest = 0;
1602 vc->napping_threads = 0;
1603
1604 /*
1605 * Updating any of the vpas requires calling kvmppc_pin_guest_page,
1606 * which can't be called with any spinlocks held.
1607 */
1608 if (need_vpa_update) {
1609 spin_unlock(&vc->lock);
913d3ff9
PM		 1610 for (i = 0; i < need_vpa_update; ++i)
1611 kvmppc_update_vpas(vcpus_to_update[i]);
081f323b
PM		 1612 spin_lock(&vc->lock);
1613 }
de56a948 1614
7b444c67 1615 /*
3102f784
ME		 1616 * Make sure we are running on primary threads, and that secondary
1617 * threads are offline. Also check if the number of threads in this
1618 * guest are greater than the current system threads per guest.
7b444c67 1619 */
3102f784
ME		 1620 if ((threads_per_core > 1) &&
1621 ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
7b444c67
PM		 1622 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1623 vcpu->arch.ret = -EBUSY;
1624 goto out;
1625 }
1626
3102f784 1627
371fefd6 1628 vc->pcpu = smp_processor_id();
2e25aa5f 1629 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
371fefd6 1630 kvmppc_start_thread(vcpu);
0456ec4f 1631 kvmppc_create_dtl_entry(vcpu, vc);
2e25aa5f 1632 }
371fefd6 1633
e0b7ec05
PM		 1634 /* Set this explicitly in case thread 0 doesn't have a vcpu */
1635 get_paca()->kvm_hstate.kvm_vcore = vc;
1636 get_paca()->kvm_hstate.ptid = 0;
1637
2f12f034 1638 vc->vcore_state = VCORE_RUNNING;
19ccb76a 1639 preempt_disable();
371fefd6 1640 spin_unlock(&vc->lock);
de56a948 1641
371fefd6 1642 kvm_guest_enter();
2c9097e4 1643
e0b7ec05 1644 srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2c9097e4 1645
e0b7ec05 1646 __kvmppc_vcore_entry();
de56a948 1647
371fefd6 1648 spin_lock(&vc->lock);
19ccb76a
PM		 1649 /* disable sending of IPIs on virtual external irqs */
1650 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1651 vcpu->cpu = -1;
1652 /* wait for secondary threads to finish writing their state to memory */
371fefd6
PM		 1653 if (vc->nap_count < vc->n_woken)
1654 kvmppc_wait_for_nap(vc);
3102f784 1655 for (i = 0; i < threads_per_subcore; ++i)
2f12f034 1656 kvmppc_release_hwthread(vc->pcpu + i);
371fefd6 1657 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
19ccb76a 1658 vc->vcore_state = VCORE_EXITING;
371fefd6
PM		 1659 spin_unlock(&vc->lock);
1660
e0b7ec05 1661 srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2c9097e4 1662
371fefd6
PM		 1663 /* make sure updates to secondary vcpu structs are visible now */
1664 smp_mb();
de56a948
PM		 1665 kvm_guest_exit();
1666
1667 preempt_enable();
c08ac06a 1668 cond_resched();
de56a948 1669
913d3ff9 1670 spin_lock(&vc->lock);
de56a948 1671 now = get_tb();
371fefd6
PM		 1672 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1673 /* cancel pending dec exception if dec is positive */
1674 if (now < vcpu->arch.dec_expires &&
1675 kvmppc_core_pending_dec(vcpu))
1676 kvmppc_core_dequeue_dec(vcpu);
19ccb76a
PM		 1677
1678 ret = RESUME_GUEST;
1679 if (vcpu->arch.trap)
3a167bea
AK		 1680 ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
1681 vcpu->arch.run_task);
19ccb76a 1682
371fefd6
PM		 1683 vcpu->arch.ret = ret;
1684 vcpu->arch.trap = 0;
19ccb76a
PM		 1685
1686 if (vcpu->arch.ceded) {
e59d24e6 1687 if (!is_kvmppc_resume_guest(ret))
19ccb76a
PM		 1688 kvmppc_end_cede(vcpu);
1689 else
1690 kvmppc_set_timer(vcpu);
1691 }
371fefd6 1692 }
de56a948
PM		 1693
1694 out:
19ccb76a 1695 vc->vcore_state = VCORE_INACTIVE;
371fefd6
PM		 1696 list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1697 arch.run_list) {
e59d24e6 1698 if (!is_kvmppc_resume_guest(vcpu->arch.ret)) {
371fefd6
PM		 1699 kvmppc_remove_runnable(vc, vcpu);
1700 wake_up(&vcpu->arch.cpu_run);
1701 }
1702 }
371fefd6
PM		 1703}
1704
19ccb76a
PM		 1705/*
1706 * Wait for some other vcpu thread to execute us, and
1707 * wake us up when we need to handle something in the host.
1708 */
1709static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
371fefd6 1710{
371fefd6
PM		 1711 DEFINE_WAIT(wait);
1712
19ccb76a
PM		 1713 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
1714 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
1715 schedule();
1716 finish_wait(&vcpu->arch.cpu_run, &wait);
1717}
1718
1719/*
1720 * All the vcpus in this vcore are idle, so wait for a decrementer
1721 * or external interrupt to one of the vcpus. vc->lock is held.
1722 */
1723static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
1724{
1725 DEFINE_WAIT(wait);
19ccb76a
PM		 1726
1727 prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
1728 vc->vcore_state = VCORE_SLEEPING;
1729 spin_unlock(&vc->lock);
913d3ff9 1730 schedule();
19ccb76a
PM		 1731 finish_wait(&vc->wq, &wait);
1732 spin_lock(&vc->lock);
1733 vc->vcore_state = VCORE_INACTIVE;
1734}
371fefd6 1735
19ccb76a
PM		 1736static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1737{
1738 int n_ceded;
19ccb76a
PM		 1739 struct kvmppc_vcore *vc;
1740 struct kvm_vcpu *v, *vn;
9e368f29 1741
371fefd6
PM		 1742 kvm_run->exit_reason = 0;
1743 vcpu->arch.ret = RESUME_GUEST;
1744 vcpu->arch.trap = 0;
2f12f034 1745 kvmppc_update_vpas(vcpu);
371fefd6 1746
371fefd6
PM		 1747 /*
1748 * Synchronize with other threads in this virtual core
1749 */
1750 vc = vcpu->arch.vcore;
1751 spin_lock(&vc->lock);
19ccb76a 1752 vcpu->arch.ceded = 0;
371fefd6
PM		 1753 vcpu->arch.run_task = current;
1754 vcpu->arch.kvm_run = kvm_run;
c7b67670 1755 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
19ccb76a 1756 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
c7b67670 1757 vcpu->arch.busy_preempt = TB_NIL;
371fefd6
PM		 1758 list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
1759 ++vc->n_runnable;
1760
19ccb76a
PM		 1761 /*
1762 * This happens the first time this is called for a vcpu.
1763 * If the vcore is already running, we may be able to start
1764 * this thread straight away and have it join in.
1765 */
8455d79e 1766 if (!signal_pending(current)) {
19ccb76a
PM		 1767 if (vc->vcore_state == VCORE_RUNNING &&
1768 VCORE_EXIT_COUNT(vc) == 0) {
2f12f034 1769 kvmppc_create_dtl_entry(vcpu, vc);
19ccb76a 1770 kvmppc_start_thread(vcpu);
8455d79e
PM		 1771 } else if (vc->vcore_state == VCORE_SLEEPING) {
1772 wake_up(&vc->wq);
371fefd6
PM		 1773 }
1774
8455d79e 1775 }
371fefd6 1776
19ccb76a
PM		 1777 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1778 !signal_pending(current)) {
8455d79e 1779 if (vc->vcore_state != VCORE_INACTIVE) {
19ccb76a
PM		 1780 spin_unlock(&vc->lock);
1781 kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
1782 spin_lock(&vc->lock);
1783 continue;
1784 }
19ccb76a
PM		 1785 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
1786 arch.run_list) {
7e28e60e 1787 kvmppc_core_prepare_to_enter(v);
19ccb76a
PM		 1788 if (signal_pending(v->arch.run_task)) {
1789 kvmppc_remove_runnable(vc, v);
1790 v->stat.signal_exits++;
1791 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
1792 v->arch.ret = -EINTR;
1793 wake_up(&v->arch.cpu_run);
1794 }
1795 }
8455d79e
PM		 1796 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1797 break;
1798 vc->runner = vcpu;
1799 n_ceded = 0;
4619ac88 1800 list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
8455d79e
PM		 1801 if (!v->arch.pending_exceptions)
1802 n_ceded += v->arch.ceded;
4619ac88
PM		 1803 else
1804 v->arch.ceded = 0;
1805 }
8455d79e
PM		 1806 if (n_ceded == vc->n_runnable)
1807 kvmppc_vcore_blocked(vc);
1808 else
1809 kvmppc_run_core(vc);
0456ec4f 1810 vc->runner = NULL;
19ccb76a 1811 }
371fefd6 1812
8455d79e
PM		 1813 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1814 (vc->vcore_state == VCORE_RUNNING ||
1815 vc->vcore_state == VCORE_EXITING)) {
1816 spin_unlock(&vc->lock);
1817 kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
1818 spin_lock(&vc->lock);
1819 }
1820
1821 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
1822 kvmppc_remove_runnable(vc, vcpu);
1823 vcpu->stat.signal_exits++;
1824 kvm_run->exit_reason = KVM_EXIT_INTR;
1825 vcpu->arch.ret = -EINTR;
1826 }
1827
1828 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
1829 /* Wake up some vcpu to run the core */
1830 v = list_first_entry(&vc->runnable_threads,
1831 struct kvm_vcpu, arch.run_list);
1832 wake_up(&v->arch.cpu_run);
371fefd6
PM		 1833 }
1834
371fefd6 1835 spin_unlock(&vc->lock);
371fefd6 1836 return vcpu->arch.ret;
de56a948
PM		 1837}
1838
3a167bea 1839static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
a8606e20
PM		 1840{
1841 int r;
913d3ff9 1842 int srcu_idx;
a8606e20 1843
af8f38b3
AG		 1844 if (!vcpu->arch.sane) {
1845 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1846 return -EINVAL;
1847 }
1848
25051b5a
SW		 1849 kvmppc_core_prepare_to_enter(vcpu);
1850
19ccb76a
PM		 1851 /* No need to go into the guest when all we'll do is come back out */
1852 if (signal_pending(current)) {
1853 run->exit_reason = KVM_EXIT_INTR;
1854 return -EINTR;
1855 }
1856
32fad281
PM		 1857 atomic_inc(&vcpu->kvm->arch.vcpus_running);
1858 /* Order vcpus_running vs. rma_setup_done, see kvmppc_alloc_reset_hpt */
1859 smp_mb();
1860
1861 /* On the first time here, set up HTAB and VRMA or RMA */
c77162de 1862 if (!vcpu->kvm->arch.rma_setup_done) {
32fad281 1863 r = kvmppc_hv_setup_htab_rma(vcpu);
c77162de 1864 if (r)
32fad281 1865 goto out;
c77162de 1866 }
19ccb76a
PM		 1867
1868 flush_fp_to_thread(current);
1869 flush_altivec_to_thread(current);
1870 flush_vsx_to_thread(current);
1871 vcpu->arch.wqp = &vcpu->arch.vcore->wq;
342d3db7 1872 vcpu->arch.pgdir = current->mm->pgd;
c7b67670 1873 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
19ccb76a 1874
a8606e20
PM		 1875 do {
1876 r = kvmppc_run_vcpu(run, vcpu);
1877
1878 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
1879 !(vcpu->arch.shregs.msr & MSR_PR)) {
1880 r = kvmppc_pseries_do_hcall(vcpu);
7e28e60e 1881 kvmppc_core_prepare_to_enter(vcpu);
913d3ff9
PM		 1882 } else if (r == RESUME_PAGE_FAULT) {
1883 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1884 r = kvmppc_book3s_hv_page_fault(run, vcpu,
1885 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1886 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
a8606e20 1887 }
e59d24e6 1888 } while (is_kvmppc_resume_guest(r));
32fad281
PM		 1889
1890 out:
c7b67670 1891 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
32fad281 1892 atomic_dec(&vcpu->kvm->arch.vcpus_running);
a8606e20
PM		 1893 return r;
1894}
1895
54738c09 1896
aa04b4cc 1897/* Work out RMLS (real mode limit selector) field value for a given RMA size.
9e368f29 1898 Assumes POWER7 or PPC970. */
aa04b4cc
PM		 1899static inline int lpcr_rmls(unsigned long rma_size)
1900{
1901 switch (rma_size) {
1902 case 32ul << 20: /* 32 MB */
9e368f29
PM		 1903 if (cpu_has_feature(CPU_FTR_ARCH_206))
1904 return 8; /* only supported on POWER7 */
1905 return -1;
aa04b4cc
PM		 1906 case 64ul << 20: /* 64 MB */
1907 return 3;
1908 case 128ul << 20: /* 128 MB */
1909 return 7;
1910 case 256ul << 20: /* 256 MB */
1911 return 4;
1912 case 1ul << 30: /* 1 GB */
1913 return 2;
1914 case 16ul << 30: /* 16 GB */
1915 return 1;
1916 case 256ul << 30: /* 256 GB */
1917 return 0;
1918 default:
1919 return -1;
1920 }
1921}
1922
1923static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1924{
aa04b4cc 1925 struct page *page;
6c45b810 1926 struct kvm_rma_info *ri = vma->vm_file->private_data;
aa04b4cc 1927
6c45b810 1928 if (vmf->pgoff >= kvm_rma_pages)
aa04b4cc
PM		 1929 return VM_FAULT_SIGBUS;
1930
1931 page = pfn_to_page(ri->base_pfn + vmf->pgoff);
1932 get_page(page);
1933 vmf->page = page;
1934 return 0;
1935}
1936
1937static const struct vm_operations_struct kvm_rma_vm_ops = {
1938 .fault = kvm_rma_fault,
1939};
1940
1941static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma)
1942{
314e51b9 1943 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
aa04b4cc
PM		 1944 vma->vm_ops = &kvm_rma_vm_ops;
1945 return 0;
1946}
1947
1948static int kvm_rma_release(struct inode *inode, struct file *filp)
1949{
6c45b810 1950 struct kvm_rma_info *ri = filp->private_data;
aa04b4cc
PM		 1951
1952 kvm_release_rma(ri);
1953 return 0;
1954}
1955
75ef9de1 1956static const struct file_operations kvm_rma_fops = {
aa04b4cc
PM		 1957 .mmap = kvm_rma_mmap,
1958 .release = kvm_rma_release,
1959};
1960
3a167bea
AK		 1961static long kvm_vm_ioctl_allocate_rma(struct kvm *kvm,
1962 struct kvm_allocate_rma *ret)
aa04b4cc 1963{
aa04b4cc 1964 long fd;
6c45b810
AK		 1965 struct kvm_rma_info *ri;
1966 /*
1967 * Only do this on PPC970 in HV mode
1968 */
1969 if (!cpu_has_feature(CPU_FTR_HVMODE) ||
1970 !cpu_has_feature(CPU_FTR_ARCH_201))
1971 return -EINVAL;
1972
1973 if (!kvm_rma_pages)
1974 return -EINVAL;
aa04b4cc
PM		 1975
1976 ri = kvm_alloc_rma();
1977 if (!ri)
1978 return -ENOMEM;
1979
2f84d5ea 1980 fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR | O_CLOEXEC);
aa04b4cc
PM		 1981 if (fd < 0)
1982 kvm_release_rma(ri);
1983
6c45b810 1984 ret->rma_size = kvm_rma_pages << PAGE_SHIFT;
aa04b4cc
PM		 1985 return fd;
1986}
1987
5b74716e
BH		 1988static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
1989 int linux_psize)
1990{
1991 struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
1992
1993 if (!def->shift)
1994 return;
1995 (*sps)->page_shift = def->shift;
1996 (*sps)->slb_enc = def->sllp;
1997 (*sps)->enc[0].page_shift = def->shift;
b1022fbd
AK		 1998 /*
1999 * Only return base page encoding. We don't want to return
2000 * all the supporting pte_enc, because our H_ENTER doesn't
2001 * support MPSS yet. Once they do, we can start passing all
2002 * support pte_enc here
2003 */
2004 (*sps)->enc[0].pte_enc = def->penc[linux_psize];
1f365bb0
AK		 2005 /*
2006 * Add 16MB MPSS support if host supports it
2007 */
2008 if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
2009 (*sps)->enc[1].page_shift = 24;
2010 (*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
2011 }
5b74716e
BH		 2012 (*sps)++;
2013}
2014
3a167bea
AK		 2015static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
2016 struct kvm_ppc_smmu_info *info)
5b74716e
BH		 2017{
2018 struct kvm_ppc_one_seg_page_size *sps;
2019
2020 info->flags = KVM_PPC_PAGE_SIZES_REAL;
2021 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
2022 info->flags |= KVM_PPC_1T_SEGMENTS;
2023 info->slb_size = mmu_slb_size;
2024
2025 /* We only support these sizes for now, and no muti-size segments */
2026 sps = &info->sps[0];
2027 kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
2028 kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
2029 kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
2030
2031 return 0;
2032}
2033
82ed3616
PM		 2034/*
2035 * Get (and clear) the dirty memory log for a memory slot.
2036 */
3a167bea
AK		 2037static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
2038 struct kvm_dirty_log *log)
82ed3616
PM		 2039{
2040 struct kvm_memory_slot *memslot;
2041 int r;
2042 unsigned long n;
2043
2044 mutex_lock(&kvm->slots_lock);
2045
2046 r = -EINVAL;
bbacc0c1 2047 if (log->slot >= KVM_USER_MEM_SLOTS)
82ed3616
PM		 2048 goto out;
2049
2050 memslot = id_to_memslot(kvm->memslots, log->slot);
2051 r = -ENOENT;
2052 if (!memslot->dirty_bitmap)
2053 goto out;
2054
2055 n = kvm_dirty_bitmap_bytes(memslot);
2056 memset(memslot->dirty_bitmap, 0, n);
2057
dfe49dbd 2058 r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
82ed3616
PM		 2059 if (r)
2060 goto out;
2061
2062 r = -EFAULT;
2063 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
2064 goto out;
2065
2066 r = 0;
2067out:
2068 mutex_unlock(&kvm->slots_lock);
2069 return r;
2070}
2071
a66b48c3 2072static void unpin_slot(struct kvm_memory_slot *memslot)
de56a948 2073{
a66b48c3
PM		 2074 unsigned long *physp;
2075 unsigned long j, npages, pfn;
2076 struct page *page;
aa04b4cc 2077
a66b48c3
PM		 2078 physp = memslot->arch.slot_phys;
2079 npages = memslot->npages;
2080 if (!physp)
2081 return;
2082 for (j = 0; j < npages; j++) {
2083 if (!(physp[j] & KVMPPC_GOT_PAGE))
2084 continue;
2085 pfn = physp[j] >> PAGE_SHIFT;
2086 page = pfn_to_page(pfn);
2087 SetPageDirty(page);
2088 put_page(page);
2089 }
2090}
2091
3a167bea
AK		 2092static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
2093 struct kvm_memory_slot *dont)
a66b48c3
PM		 2094{
2095 if (!dont || free->arch.rmap != dont->arch.rmap) {
2096 vfree(free->arch.rmap);
2097 free->arch.rmap = NULL;
b2b2f165 2098 }
a66b48c3
PM		 2099 if (!dont || free->arch.slot_phys != dont->arch.slot_phys) {
2100 unpin_slot(free);
2101 vfree(free->arch.slot_phys);
2102 free->arch.slot_phys = NULL;
2103 }
2104}
2105
3a167bea
AK		 2106static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
2107 unsigned long npages)
a66b48c3
PM		 2108{
2109 slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
2110 if (!slot->arch.rmap)
2111 return -ENOMEM;
2112 slot->arch.slot_phys = NULL;
aa04b4cc 2113
c77162de
PM		 2114 return 0;
2115}
aa04b4cc 2116
3a167bea
AK		 2117static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
2118 struct kvm_memory_slot *memslot,
2119 struct kvm_userspace_memory_region *mem)
c77162de 2120{
a66b48c3 2121 unsigned long *phys;
c77162de 2122
a66b48c3
PM		 2123 /* Allocate a slot_phys array if needed */
2124 phys = memslot->arch.slot_phys;
2125 if (!kvm->arch.using_mmu_notifiers && !phys && memslot->npages) {
2126 phys = vzalloc(memslot->npages * sizeof(unsigned long));
2127 if (!phys)
2128 return -ENOMEM;
2129 memslot->arch.slot_phys = phys;
aa04b4cc 2130 }
a66b48c3
PM		 2131
2132 return 0;
c77162de
PM		 2133}
2134
3a167bea
AK		 2135static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
2136 struct kvm_userspace_memory_region *mem,
2137 const struct kvm_memory_slot *old)
c77162de 2138{
dfe49dbd
PM		 2139 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
2140 struct kvm_memory_slot *memslot;
2141
8482644a 2142 if (npages && old->npages) {
dfe49dbd
PM		 2143 /*
2144 * If modifying a memslot, reset all the rmap dirty bits.
2145 * If this is a new memslot, we don't need to do anything
2146 * since the rmap array starts out as all zeroes,
2147 * i.e. no pages are dirty.
2148 */
2149 memslot = id_to_memslot(kvm->memslots, mem->slot);
2150 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
2151 }
c77162de
PM		 2152}
2153
a0144e2a
PM		 2154/*
2155 * Update LPCR values in kvm->arch and in vcores.
2156 * Caller must hold kvm->lock.
2157 */
2158void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
2159{
2160 long int i;
2161 u32 cores_done = 0;
2162
2163 if ((kvm->arch.lpcr & mask) == lpcr)
2164 return;
2165
2166 kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
2167
2168 for (i = 0; i < KVM_MAX_VCORES; ++i) {
2169 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
2170 if (!vc)
2171 continue;
2172 spin_lock(&vc->lock);
2173 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
2174 spin_unlock(&vc->lock);
2175 if (++cores_done >= kvm->arch.online_vcores)
2176 break;
2177 }
2178}
2179
3a167bea
AK		 2180static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
2181{
2182 return;
2183}
2184
32fad281 2185static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
c77162de
PM		 2186{
2187 int err = 0;
2188 struct kvm *kvm = vcpu->kvm;
6c45b810 2189 struct kvm_rma_info *ri = NULL;
c77162de
PM		 2190 unsigned long hva;
2191 struct kvm_memory_slot *memslot;
2192 struct vm_area_struct *vma;
a0144e2a
PM		 2193 unsigned long lpcr = 0, senc;
2194 unsigned long lpcr_mask = 0;
c77162de
PM		 2195 unsigned long psize, porder;
2196 unsigned long rma_size;
2197 unsigned long rmls;
2198 unsigned long *physp;
da9d1d7f 2199 unsigned long i, npages;
2c9097e4 2200 int srcu_idx;
c77162de
PM		 2201
2202 mutex_lock(&kvm->lock);
2203 if (kvm->arch.rma_setup_done)
2204 goto out; /* another vcpu beat us to it */
aa04b4cc 2205
32fad281
PM		 2206 /* Allocate hashed page table (if not done already) and reset it */
2207 if (!kvm->arch.hpt_virt) {
2208 err = kvmppc_alloc_hpt(kvm, NULL);
2209 if (err) {
2210 pr_err("KVM: Couldn't alloc HPT\n");
2211 goto out;
2212 }
2213 }
2214
c77162de 2215 /* Look up the memslot for guest physical address 0 */
2c9097e4 2216 srcu_idx = srcu_read_lock(&kvm->srcu);
c77162de 2217 memslot = gfn_to_memslot(kvm, 0);
aa04b4cc 2218
c77162de
PM		 2219 /* We must have some memory at 0 by now */
2220 err = -EINVAL;
2221 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
2c9097e4 2222 goto out_srcu;
c77162de
PM		 2223
2224 /* Look up the VMA for the start of this memory slot */
2225 hva = memslot->userspace_addr;
2226 down_read(&current->mm->mmap_sem);
2227 vma = find_vma(current->mm, hva);
2228 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
2229 goto up_out;
2230
2231 psize = vma_kernel_pagesize(vma);
da9d1d7f 2232 porder = __ilog2(psize);
c77162de
PM		 2233
2234 /* Is this one of our preallocated RMAs? */
2235 if (vma->vm_file && vma->vm_file->f_op == &kvm_rma_fops &&
2236 hva == vma->vm_start)
2237 ri = vma->vm_file->private_data;
2238
2239 up_read(&current->mm->mmap_sem);
2240
2241 if (!ri) {
2242 /* On POWER7, use VRMA; on PPC970, give up */
2243 err = -EPERM;
2244 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2245 pr_err("KVM: CPU requires an RMO\n");
2c9097e4 2246 goto out_srcu;
c77162de
PM		 2247 }
2248
da9d1d7f
PM		 2249 /* We can handle 4k, 64k or 16M pages in the VRMA */
2250 err = -EINVAL;
2251 if (!(psize == 0x1000 || psize == 0x10000 ||
2252 psize == 0x1000000))
2c9097e4 2253 goto out_srcu;
da9d1d7f 2254
c77162de 2255 /* Update VRMASD field in the LPCR */
da9d1d7f 2256 senc = slb_pgsize_encoding(psize);
697d3899
PM		 2257 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
2258 (VRMA_VSID << SLB_VSID_SHIFT_1T);
a0144e2a
PM		 2259 lpcr_mask = LPCR_VRMASD;
2260 /* the -4 is to account for senc values starting at 0x10 */
2261 lpcr = senc << (LPCR_VRMASD_SH - 4);
c77162de
PM		 2262
2263 /* Create HPTEs in the hash page table for the VRMA */
da9d1d7f 2264 kvmppc_map_vrma(vcpu, memslot, porder);
c77162de
PM		 2265
2266 } else {
2267 /* Set up to use an RMO region */
6c45b810 2268 rma_size = kvm_rma_pages;
c77162de
PM		 2269 if (rma_size > memslot->npages)
2270 rma_size = memslot->npages;
2271 rma_size <<= PAGE_SHIFT;
aa04b4cc 2272 rmls = lpcr_rmls(rma_size);
c77162de 2273 err = -EINVAL;
5d226ae5 2274 if ((long)rmls < 0) {
c77162de 2275 pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size);
2c9097e4 2276 goto out_srcu;
aa04b4cc
PM		 2277 }
2278 atomic_inc(&ri->use_count);
2279 kvm->arch.rma = ri;
9e368f29
PM		 2280
2281 /* Update LPCR and RMOR */
9e368f29
PM		 2282 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2283 /* PPC970; insert RMLS value (split field) in HID4 */
a0144e2a
PM		 2284 lpcr_mask = (1ul << HID4_RMLS0_SH) |
2285 (3ul << HID4_RMLS2_SH) | HID4_RMOR;
2286 lpcr = ((rmls >> 2) << HID4_RMLS0_SH) |
9e368f29
PM		 2287 ((rmls & 3) << HID4_RMLS2_SH);
2288 /* RMOR is also in HID4 */
2289 lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff)
2290 << HID4_RMOR_SH;
2291 } else {
2292 /* POWER7 */
a0144e2a
PM		 2293 lpcr_mask = LPCR_VPM0 | LPCR_VRMA_L | LPCR_RMLS;
2294 lpcr = rmls << LPCR_RMLS_SH;
6c45b810 2295 kvm->arch.rmor = ri->base_pfn << PAGE_SHIFT;
9e368f29 2296 }
c77162de 2297 pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n",
aa04b4cc 2298 ri->base_pfn << PAGE_SHIFT, rma_size, lpcr);
aa04b4cc 2299
c77162de 2300 /* Initialize phys addrs of pages in RMO */
6c45b810 2301 npages = kvm_rma_pages;
da9d1d7f 2302 porder = __ilog2(npages);
a66b48c3
PM		 2303 physp = memslot->arch.slot_phys;
2304 if (physp) {
2305 if (npages > memslot->npages)
2306 npages = memslot->npages;
2307 spin_lock(&kvm->arch.slot_phys_lock);
2308 for (i = 0; i < npages; ++i)
2309 physp[i] = ((ri->base_pfn + i) << PAGE_SHIFT) +
2310 porder;
2311 spin_unlock(&kvm->arch.slot_phys_lock);
2312 }
aa04b4cc
PM		 2313 }
2314
a0144e2a
PM		 2315 kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
2316
c77162de
PM		 2317 /* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */
2318 smp_wmb();
2319 kvm->arch.rma_setup_done = 1;
2320 err = 0;
2c9097e4
PM		 2321 out_srcu:
2322 srcu_read_unlock(&kvm->srcu, srcu_idx);
c77162de
PM		 2323 out:
2324 mutex_unlock(&kvm->lock);
2325 return err;
b2b2f165 2326
c77162de
PM		 2327 up_out:
2328 up_read(&current->mm->mmap_sem);
505d6421 2329 goto out_srcu;
de56a948
PM		 2330}
2331
3a167bea 2332static int kvmppc_core_init_vm_hv(struct kvm *kvm)
de56a948 2333{
32fad281 2334 unsigned long lpcr, lpid;
de56a948 2335
32fad281
PM		 2336 /* Allocate the guest's logical partition ID */
2337
2338 lpid = kvmppc_alloc_lpid();
5d226ae5 2339 if ((long)lpid < 0)
32fad281
PM		 2340 return -ENOMEM;
2341 kvm->arch.lpid = lpid;
de56a948 2342
1b400ba0
PM		 2343 /*
2344 * Since we don't flush the TLB when tearing down a VM,
2345 * and this lpid might have previously been used,
2346 * make sure we flush on each core before running the new VM.
2347 */
2348 cpumask_setall(&kvm->arch.need_tlb_flush);
2349
699a0ea0
PM		 2350 /* Start out with the default set of hcalls enabled */
2351 memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
2352 sizeof(kvm->arch.enabled_hcalls));
2353
aa04b4cc 2354 kvm->arch.rma = NULL;
aa04b4cc 2355
9e368f29 2356 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
aa04b4cc 2357
9e368f29
PM		 2358 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2359 /* PPC970; HID4 is effectively the LPCR */
9e368f29
PM		 2360 kvm->arch.host_lpid = 0;
2361 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4);
2362 lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH));
2363 lpcr |= ((lpid >> 4) << HID4_LPID1_SH) |
2364 ((lpid & 0xf) << HID4_LPID5_SH);
2365 } else {
2366 /* POWER7; init LPCR for virtual RMA mode */
2367 kvm->arch.host_lpid = mfspr(SPRN_LPID);
2368 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
2369 lpcr &= LPCR_PECE | LPCR_LPES;
2370 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
697d3899
PM		 2371 LPCR_VPM0 | LPCR_VPM1;
2372 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
2373 (VRMA_VSID << SLB_VSID_SHIFT_1T);
e0622bd9
PM		 2374 /* On POWER8 turn on online bit to enable PURR/SPURR */
2375 if (cpu_has_feature(CPU_FTR_ARCH_207S))
2376 lpcr |= LPCR_ONL;
9e368f29
PM		 2377 }
2378 kvm->arch.lpcr = lpcr;
aa04b4cc 2379
342d3db7 2380 kvm->arch.using_mmu_notifiers = !!cpu_has_feature(CPU_FTR_ARCH_206);
c77162de 2381 spin_lock_init(&kvm->arch.slot_phys_lock);
512691d4
PM		 2382
2383 /*
441c19c8
ME		 2384 * Track that we now have a HV mode VM active. This blocks secondary
2385 * CPU threads from coming online.
512691d4 2386 */
441c19c8 2387 kvm_hv_vm_activated();
512691d4 2388
54738c09 2389 return 0;
de56a948
PM		 2390}
2391
f1378b1c
PM		 2392static void kvmppc_free_vcores(struct kvm *kvm)
2393{
2394 long int i;
2395
2396 for (i = 0; i < KVM_MAX_VCORES; ++i)
2397 kfree(kvm->arch.vcores[i]);
2398 kvm->arch.online_vcores = 0;
2399}
2400
3a167bea 2401static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
de56a948 2402{
441c19c8 2403 kvm_hv_vm_deactivated();
512691d4 2404
f1378b1c 2405 kvmppc_free_vcores(kvm);
aa04b4cc
PM		 2406 if (kvm->arch.rma) {
2407 kvm_release_rma(kvm->arch.rma);
2408 kvm->arch.rma = NULL;
2409 }
2410
de56a948
PM		 2411 kvmppc_free_hpt(kvm);
2412}
2413
3a167bea
AK		 2414/* We don't need to emulate any privileged instructions or dcbz */
2415static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
2416 unsigned int inst, int *advance)
de56a948 2417{
3a167bea 2418 return EMULATE_FAIL;
de56a948
PM		 2419}
2420
3a167bea
AK		 2421static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
2422 ulong spr_val)
de56a948
PM		 2423{
2424 return EMULATE_FAIL;
2425}
2426
3a167bea
AK		 2427static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
2428 ulong *spr_val)
de56a948
PM		 2429{
2430 return EMULATE_FAIL;
2431}
2432
3a167bea 2433static int kvmppc_core_check_processor_compat_hv(void)
de56a948 2434{
3a167bea
AK		 2435 if (!cpu_has_feature(CPU_FTR_HVMODE))
2436 return -EIO;
2437 return 0;
de56a948
PM		 2438}
2439
3a167bea
AK		 2440static long kvm_arch_vm_ioctl_hv(struct file *filp,
2441 unsigned int ioctl, unsigned long arg)
2442{
2443 struct kvm *kvm __maybe_unused = filp->private_data;
2444 void __user *argp = (void __user *)arg;
2445 long r;
2446
2447 switch (ioctl) {
2448
2449 case KVM_ALLOCATE_RMA: {
2450 struct kvm_allocate_rma rma;
2451 struct kvm *kvm = filp->private_data;
2452
2453 r = kvm_vm_ioctl_allocate_rma(kvm, &rma);
2454 if (r >= 0 && copy_to_user(argp, &rma, sizeof(rma)))
2455 r = -EFAULT;
2456 break;
2457 }
2458
2459 case KVM_PPC_ALLOCATE_HTAB: {
2460 u32 htab_order;
2461
2462 r = -EFAULT;
2463 if (get_user(htab_order, (u32 __user *)argp))
2464 break;
2465 r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
2466 if (r)
2467 break;
2468 r = -EFAULT;
2469 if (put_user(htab_order, (u32 __user *)argp))
2470 break;
2471 r = 0;
2472 break;
2473 }
2474
2475 case KVM_PPC_GET_HTAB_FD: {
2476 struct kvm_get_htab_fd ghf;
2477
2478 r = -EFAULT;
2479 if (copy_from_user(&ghf, argp, sizeof(ghf)))
2480 break;
2481 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
2482 break;
2483 }
2484
2485 default:
2486 r = -ENOTTY;
2487 }
2488
2489 return r;
2490}
2491
699a0ea0
PM		 2492/*
2493 * List of hcall numbers to enable by default.
2494 * For compatibility with old userspace, we enable by default
2495 * all hcalls that were implemented before the hcall-enabling
2496 * facility was added. Note this list should not include H_RTAS.
2497 */
2498static unsigned int default_hcall_list[] = {
2499 H_REMOVE,
2500 H_ENTER,
2501 H_READ,
2502 H_PROTECT,
2503 H_BULK_REMOVE,
2504 H_GET_TCE,
2505 H_PUT_TCE,
2506 H_SET_DABR,
2507 H_SET_XDABR,
2508 H_CEDE,
2509 H_PROD,
2510 H_CONFER,
2511 H_REGISTER_VPA,
2512#ifdef CONFIG_KVM_XICS
2513 H_EOI,
2514 H_CPPR,
2515 H_IPI,
2516 H_IPOLL,
2517 H_XIRR,
2518 H_XIRR_X,
2519#endif
2520 0
2521};
2522
2523static void init_default_hcalls(void)
2524{
2525 int i;
ae2113a4 2526 unsigned int hcall;
699a0ea0 2527
ae2113a4
PM		 2528 for (i = 0; default_hcall_list[i]; ++i) {
2529 hcall = default_hcall_list[i];
2530 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
2531 __set_bit(hcall / 4, default_enabled_hcalls);
2532 }
699a0ea0
PM		 2533}
2534
cbbc58d4 2535static struct kvmppc_ops kvm_ops_hv = {
3a167bea
AK		 2536 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
2537 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
2538 .get_one_reg = kvmppc_get_one_reg_hv,
2539 .set_one_reg = kvmppc_set_one_reg_hv,
2540 .vcpu_load = kvmppc_core_vcpu_load_hv,
2541 .vcpu_put = kvmppc_core_vcpu_put_hv,
2542 .set_msr = kvmppc_set_msr_hv,
2543 .vcpu_run = kvmppc_vcpu_run_hv,
2544 .vcpu_create = kvmppc_core_vcpu_create_hv,
2545 .vcpu_free = kvmppc_core_vcpu_free_hv,
2546 .check_requests = kvmppc_core_check_requests_hv,
2547 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv,
2548 .flush_memslot = kvmppc_core_flush_memslot_hv,
2549 .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
2550 .commit_memory_region = kvmppc_core_commit_memory_region_hv,
2551 .unmap_hva = kvm_unmap_hva_hv,
2552 .unmap_hva_range = kvm_unmap_hva_range_hv,
2553 .age_hva = kvm_age_hva_hv,
2554 .test_age_hva = kvm_test_age_hva_hv,
2555 .set_spte_hva = kvm_set_spte_hva_hv,
2556 .mmu_destroy = kvmppc_mmu_destroy_hv,
2557 .free_memslot = kvmppc_core_free_memslot_hv,
2558 .create_memslot = kvmppc_core_create_memslot_hv,
2559 .init_vm = kvmppc_core_init_vm_hv,
2560 .destroy_vm = kvmppc_core_destroy_vm_hv,
3a167bea
AK		 2561 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
2562 .emulate_op = kvmppc_core_emulate_op_hv,
2563 .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
2564 .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
2565 .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
2566 .arch_vm_ioctl = kvm_arch_vm_ioctl_hv,
ae2113a4 2567 .hcall_implemented = kvmppc_hcall_impl_hv,
3a167bea
AK		 2568};
2569
2570static int kvmppc_book3s_init_hv(void)
de56a948
PM		 2571{
2572 int r;
cbbc58d4
AK		 2573 /*
2574 * FIXME!! Do we need to check on all cpus ?
2575 */
2576 r = kvmppc_core_check_processor_compat_hv();
2577 if (r < 0)
739e2425 2578 return -ENODEV;
de56a948 2579
cbbc58d4
AK		 2580 kvm_ops_hv.owner = THIS_MODULE;
2581 kvmppc_hv_ops = &kvm_ops_hv;
de56a948 2582
699a0ea0
PM		 2583 init_default_hcalls();
2584
cbbc58d4 2585 r = kvmppc_mmu_hv_init();
de56a948
PM		 2586 return r;
2587}
2588
3a167bea 2589static void kvmppc_book3s_exit_hv(void)
de56a948 2590{
cbbc58d4 2591 kvmppc_hv_ops = NULL;
de56a948
PM		 2592}
2593
3a167bea
AK		 2594module_init(kvmppc_book3s_init_hv);
2595module_exit(kvmppc_book3s_exit_hv);
2ba9f0d8 2596MODULE_LICENSE("GPL");
398a76c6
AG		 2597MODULE_ALIAS_MISCDEV(KVM_MINOR);
2598MODULE_ALIAS("devname:kvm");
Linux 6.x block layer and io_uring tree(s)