]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - arch/powerpc/kvm/book3s_hv.c
projects / mirror_ubuntu-bionic-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: PPC: e500mc: Enhance tlb invalidation condition on vcpu schedule
[mirror_ubuntu-bionic-kernel.git] / arch / powerpc / kvm / book3s_hv.c
CommitLineData
de56a948
PM		 1/*
2 * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
3 * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
4 *
5 * Authors:
6 * Paul Mackerras <paulus@au1.ibm.com>
7 * Alexander Graf <agraf@suse.de>
8 * Kevin Wolf <mail@kevin-wolf.de>
9 *
10 * Description: KVM functions specific to running on Book 3S
11 * processors in hypervisor mode (specifically POWER7 and later).
12 *
13 * This file is derived from arch/powerpc/kvm/book3s.c,
14 * by Alexander Graf <agraf@suse.de>.
15 *
16 * This program is free software; you can redistribute it and/or modify
17 * it under the terms of the GNU General Public License, version 2, as
18 * published by the Free Software Foundation.
19 */
20
21#include <linux/kvm_host.h>
22#include <linux/err.h>
23#include <linux/slab.h>
24#include <linux/preempt.h>
25#include <linux/sched.h>
26#include <linux/delay.h>
66b15db6 27#include <linux/export.h>
de56a948
PM		 28#include <linux/fs.h>
29#include <linux/anon_inodes.h>
30#include <linux/cpumask.h>
aa04b4cc
PM		 31#include <linux/spinlock.h>
32#include <linux/page-flags.h>
2c9097e4 33#include <linux/srcu.h>
398a76c6 34#include <linux/miscdevice.h>
de56a948
PM		 35
36#include <asm/reg.h>
37#include <asm/cputable.h>
38#include <asm/cacheflush.h>
39#include <asm/tlbflush.h>
40#include <asm/uaccess.h>
41#include <asm/io.h>
42#include <asm/kvm_ppc.h>
43#include <asm/kvm_book3s.h>
44#include <asm/mmu_context.h>
45#include <asm/lppaca.h>
46#include <asm/processor.h>
371fefd6 47#include <asm/cputhreads.h>
aa04b4cc 48#include <asm/page.h>
de1d9248 49#include <asm/hvcall.h>
ae3a197e 50#include <asm/switch_to.h>
512691d4 51#include <asm/smp.h>
de56a948 52#include <linux/gfp.h>
de56a948
PM		 53#include <linux/vmalloc.h>
54#include <linux/highmem.h>
c77162de 55#include <linux/hugetlb.h>
2ba9f0d8 56#include <linux/module.h>
de56a948 57
3a167bea
AK		 58#include "book3s.h"
59
de56a948
PM		 60/* #define EXIT_DEBUG */
61/* #define EXIT_DEBUG_SIMPLE */
62/* #define EXIT_DEBUG_INT */
63
913d3ff9
PM		 64/* Used to indicate that a guest page fault needs to be handled */
65#define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1)
66
c7b67670
PM		 67/* Used as a "null" value for timebase values */
68#define TB_NIL (~(u64)0)
69
19ccb76a 70static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
32fad281 71static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
19ccb76a 72
3a167bea 73static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
54695c30
BH		 74{
75 int me;
76 int cpu = vcpu->cpu;
77 wait_queue_head_t *wqp;
78
79 wqp = kvm_arch_vcpu_wq(vcpu);
80 if (waitqueue_active(wqp)) {
81 wake_up_interruptible(wqp);
82 ++vcpu->stat.halt_wakeup;
83 }
84
85 me = get_cpu();
86
87 /* CPU points to the first thread of the core */
88 if (cpu != me && cpu >= 0 && cpu < nr_cpu_ids) {
7505258c 89#ifdef CONFIG_PPC_ICP_NATIVE
54695c30
BH		 90 int real_cpu = cpu + vcpu->arch.ptid;
91 if (paca[real_cpu].kvm_hstate.xics_phys)
92 xics_wake_cpu(real_cpu);
48eaef05
AS		 93 else
94#endif
95 if (cpu_online(cpu))
54695c30
BH		 96 smp_send_reschedule(cpu);
97 }
98 put_cpu();
99}
100
c7b67670
PM		 101/*
102 * We use the vcpu_load/put functions to measure stolen time.
103 * Stolen time is counted as time when either the vcpu is able to
104 * run as part of a virtual core, but the task running the vcore
105 * is preempted or sleeping, or when the vcpu needs something done
106 * in the kernel by the task running the vcpu, but that task is
107 * preempted or sleeping. Those two things have to be counted
108 * separately, since one of the vcpu tasks will take on the job
109 * of running the core, and the other vcpu tasks in the vcore will
110 * sleep waiting for it to do that, but that sleep shouldn't count
111 * as stolen time.
112 *
113 * Hence we accumulate stolen time when the vcpu can run as part of
114 * a vcore using vc->stolen_tb, and the stolen time when the vcpu
115 * needs its task to do other things in the kernel (for example,
116 * service a page fault) in busy_stolen. We don't accumulate
117 * stolen time for a vcore when it is inactive, or for a vcpu
118 * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of
119 * a misnomer; it means that the vcpu task is not executing in
120 * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
121 * the kernel. We don't have any way of dividing up that time
122 * between time that the vcpu is genuinely stopped, time that
123 * the task is actively working on behalf of the vcpu, and time
124 * that the task is preempted, so we don't count any of it as
125 * stolen.
126 *
127 * Updates to busy_stolen are protected by arch.tbacct_lock;
128 * updates to vc->stolen_tb are protected by the arch.tbacct_lock
129 * of the vcpu that has taken responsibility for running the vcore
130 * (i.e. vc->runner). The stolen times are measured in units of
131 * timebase ticks. (Note that the != TB_NIL checks below are
132 * purely defensive; they should never fail.)
133 */
134
3a167bea 135static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
de56a948 136{
0456ec4f 137 struct kvmppc_vcore *vc = vcpu->arch.vcore;
bf3d32e1 138 unsigned long flags;
0456ec4f 139
bf3d32e1 140 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
c7b67670
PM		 141 if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE &&
142 vc->preempt_tb != TB_NIL) {
0456ec4f 143 vc->stolen_tb += mftb() - vc->preempt_tb;
c7b67670
PM		 144 vc->preempt_tb = TB_NIL;
145 }
146 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
147 vcpu->arch.busy_preempt != TB_NIL) {
148 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
149 vcpu->arch.busy_preempt = TB_NIL;
150 }
bf3d32e1 151 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
de56a948
PM		 152}
153
3a167bea 154static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
de56a948 155{
0456ec4f 156 struct kvmppc_vcore *vc = vcpu->arch.vcore;
bf3d32e1 157 unsigned long flags;
0456ec4f 158
bf3d32e1 159 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
0456ec4f
PM		 160 if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE)
161 vc->preempt_tb = mftb();
c7b67670
PM		 162 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
163 vcpu->arch.busy_preempt = mftb();
bf3d32e1 164 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
de56a948
PM		 165}
166
3a167bea 167static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
de56a948
PM		 168{
169 vcpu->arch.shregs.msr = msr;
19ccb76a 170 kvmppc_end_cede(vcpu);
de56a948
PM		 171}
172
3a167bea 173void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
de56a948
PM		 174{
175 vcpu->arch.pvr = pvr;
176}
177
388cc6e1
PM		 178int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
179{
180 unsigned long pcr = 0;
181 struct kvmppc_vcore *vc = vcpu->arch.vcore;
182
183 if (arch_compat) {
184 if (!cpu_has_feature(CPU_FTR_ARCH_206))
185 return -EINVAL; /* 970 has no compat mode support */
186
187 switch (arch_compat) {
188 case PVR_ARCH_205:
5557ae0e
PM		 189 /*
190 * If an arch bit is set in PCR, all the defined
191 * higher-order arch bits also have to be set.
192 */
193 pcr = PCR_ARCH_206 | PCR_ARCH_205;
388cc6e1
PM		 194 break;
195 case PVR_ARCH_206:
196 case PVR_ARCH_206p:
5557ae0e
PM		 197 pcr = PCR_ARCH_206;
198 break;
199 case PVR_ARCH_207:
388cc6e1
PM		 200 break;
201 default:
202 return -EINVAL;
203 }
5557ae0e
PM		 204
205 if (!cpu_has_feature(CPU_FTR_ARCH_207S)) {
206 /* POWER7 can't emulate POWER8 */
207 if (!(pcr & PCR_ARCH_206))
208 return -EINVAL;
209 pcr &= ~PCR_ARCH_206;
210 }
388cc6e1
PM		 211 }
212
213 spin_lock(&vc->lock);
214 vc->arch_compat = arch_compat;
215 vc->pcr = pcr;
216 spin_unlock(&vc->lock);
217
218 return 0;
219}
220
de56a948
PM		 221void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
222{
223 int r;
224
225 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
226 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
227 vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
228 for (r = 0; r < 16; ++r)
229 pr_err("r%2d = %.16lx r%d = %.16lx\n",
230 r, kvmppc_get_gpr(vcpu, r),
231 r+16, kvmppc_get_gpr(vcpu, r+16));
232 pr_err("ctr = %.16lx lr = %.16lx\n",
233 vcpu->arch.ctr, vcpu->arch.lr);
234 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
235 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
236 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
237 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
238 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
239 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
240 pr_err("cr = %.8x xer = %.16lx dsisr = %.8x\n",
241 vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
242 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
243 pr_err("fault dar = %.16lx dsisr = %.8x\n",
244 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
245 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
246 for (r = 0; r < vcpu->arch.slb_max; ++r)
247 pr_err(" ESID = %.16llx VSID = %.16llx\n",
248 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
249 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
a0144e2a 250 vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
de56a948
PM		 251 vcpu->arch.last_inst);
252}
253
a8606e20
PM		 254struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
255{
256 int r;
257 struct kvm_vcpu *v, *ret = NULL;
258
259 mutex_lock(&kvm->lock);
260 kvm_for_each_vcpu(r, v, kvm) {
261 if (v->vcpu_id == id) {
262 ret = v;
263 break;
264 }
265 }
266 mutex_unlock(&kvm->lock);
267 return ret;
268}
269
270static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
271{
f13c13a0 272 vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
a8606e20
PM		 273 vpa->yield_count = 1;
274}
275
55b665b0
PM		 276static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
277 unsigned long addr, unsigned long len)
278{
279 /* check address is cacheline aligned */
280 if (addr & (L1_CACHE_BYTES - 1))
281 return -EINVAL;
282 spin_lock(&vcpu->arch.vpa_update_lock);
283 if (v->next_gpa != addr || v->len != len) {
284 v->next_gpa = addr;
285 v->len = addr ? len : 0;
286 v->update_pending = 1;
287 }
288 spin_unlock(&vcpu->arch.vpa_update_lock);
289 return 0;
290}
291
2e25aa5f
PM		 292/* Length for a per-processor buffer is passed in at offset 4 in the buffer */
293struct reg_vpa {
294 u32 dummy;
295 union {
296 u16 hword;
297 u32 word;
298 } length;
299};
300
301static int vpa_is_registered(struct kvmppc_vpa *vpap)
302{
303 if (vpap->update_pending)
304 return vpap->next_gpa != 0;
305 return vpap->pinned_addr != NULL;
306}
307
a8606e20
PM		 308static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
309 unsigned long flags,
310 unsigned long vcpuid, unsigned long vpa)
311{
312 struct kvm *kvm = vcpu->kvm;
93e60249 313 unsigned long len, nb;
a8606e20
PM		 314 void *va;
315 struct kvm_vcpu *tvcpu;
2e25aa5f
PM		 316 int err;
317 int subfunc;
318 struct kvmppc_vpa *vpap;
a8606e20
PM		 319
320 tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
321 if (!tvcpu)
322 return H_PARAMETER;
323
2e25aa5f
PM		 324 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
325 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
326 subfunc == H_VPA_REG_SLB) {
327 /* Registering new area - address must be cache-line aligned */
328 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
a8606e20 329 return H_PARAMETER;
2e25aa5f
PM		 330
331 /* convert logical addr to kernel addr and read length */
93e60249
PM		 332 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
333 if (va == NULL)
b2b2f165 334 return H_PARAMETER;
2e25aa5f
PM		 335 if (subfunc == H_VPA_REG_VPA)
336 len = ((struct reg_vpa *)va)->length.hword;
a8606e20 337 else
2e25aa5f 338 len = ((struct reg_vpa *)va)->length.word;
c35635ef 339 kvmppc_unpin_guest_page(kvm, va, vpa, false);
2e25aa5f
PM		 340
341 /* Check length */
342 if (len > nb || len < sizeof(struct reg_vpa))
343 return H_PARAMETER;
344 } else {
345 vpa = 0;
346 len = 0;
347 }
348
349 err = H_PARAMETER;
350 vpap = NULL;
351 spin_lock(&tvcpu->arch.vpa_update_lock);
352
353 switch (subfunc) {
354 case H_VPA_REG_VPA: /* register VPA */
355 if (len < sizeof(struct lppaca))
a8606e20 356 break;
2e25aa5f
PM		 357 vpap = &tvcpu->arch.vpa;
358 err = 0;
359 break;
360
361 case H_VPA_REG_DTL: /* register DTL */
362 if (len < sizeof(struct dtl_entry))
a8606e20 363 break;
2e25aa5f
PM		 364 len -= len % sizeof(struct dtl_entry);
365
366 /* Check that they have previously registered a VPA */
367 err = H_RESOURCE;
368 if (!vpa_is_registered(&tvcpu->arch.vpa))
a8606e20 369 break;
2e25aa5f
PM		 370
371 vpap = &tvcpu->arch.dtl;
372 err = 0;
373 break;
374
375 case H_VPA_REG_SLB: /* register SLB shadow buffer */
376 /* Check that they have previously registered a VPA */
377 err = H_RESOURCE;
378 if (!vpa_is_registered(&tvcpu->arch.vpa))
a8606e20 379 break;
2e25aa5f
PM		 380
381 vpap = &tvcpu->arch.slb_shadow;
382 err = 0;
383 break;
384
385 case H_VPA_DEREG_VPA: /* deregister VPA */
386 /* Check they don't still have a DTL or SLB buf registered */
387 err = H_RESOURCE;
388 if (vpa_is_registered(&tvcpu->arch.dtl) ||
389 vpa_is_registered(&tvcpu->arch.slb_shadow))
a8606e20 390 break;
2e25aa5f
PM		 391
392 vpap = &tvcpu->arch.vpa;
393 err = 0;
394 break;
395
396 case H_VPA_DEREG_DTL: /* deregister DTL */
397 vpap = &tvcpu->arch.dtl;
398 err = 0;
399 break;
400
401 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */
402 vpap = &tvcpu->arch.slb_shadow;
403 err = 0;
404 break;
405 }
406
407 if (vpap) {
408 vpap->next_gpa = vpa;
409 vpap->len = len;
410 vpap->update_pending = 1;
a8606e20 411 }
93e60249 412
2e25aa5f
PM		 413 spin_unlock(&tvcpu->arch.vpa_update_lock);
414
93e60249 415 return err;
a8606e20
PM		 416}
417
081f323b 418static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
2e25aa5f 419{
081f323b 420 struct kvm *kvm = vcpu->kvm;
2e25aa5f
PM		 421 void *va;
422 unsigned long nb;
081f323b 423 unsigned long gpa;
2e25aa5f 424
081f323b
PM		 425 /*
426 * We need to pin the page pointed to by vpap->next_gpa,
427 * but we can't call kvmppc_pin_guest_page under the lock
428 * as it does get_user_pages() and down_read(). So we
429 * have to drop the lock, pin the page, then get the lock
430 * again and check that a new area didn't get registered
431 * in the meantime.
432 */
433 for (;;) {
434 gpa = vpap->next_gpa;
435 spin_unlock(&vcpu->arch.vpa_update_lock);
436 va = NULL;
437 nb = 0;
438 if (gpa)
c35635ef 439 va = kvmppc_pin_guest_page(kvm, gpa, &nb);
081f323b
PM		 440 spin_lock(&vcpu->arch.vpa_update_lock);
441 if (gpa == vpap->next_gpa)
442 break;
443 /* sigh... unpin that one and try again */
444 if (va)
c35635ef 445 kvmppc_unpin_guest_page(kvm, va, gpa, false);
081f323b
PM		 446 }
447
448 vpap->update_pending = 0;
449 if (va && nb < vpap->len) {
450 /*
451 * If it's now too short, it must be that userspace
452 * has changed the mappings underlying guest memory,
453 * so unregister the region.
454 */
c35635ef 455 kvmppc_unpin_guest_page(kvm, va, gpa, false);
081f323b 456 va = NULL;
2e25aa5f
PM		 457 }
458 if (vpap->pinned_addr)
c35635ef
PM		 459 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
460 vpap->dirty);
461 vpap->gpa = gpa;
2e25aa5f 462 vpap->pinned_addr = va;
c35635ef 463 vpap->dirty = false;
2e25aa5f
PM		 464 if (va)
465 vpap->pinned_end = va + vpap->len;
466}
467
468static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
469{
2f12f034
PM		 470 if (!(vcpu->arch.vpa.update_pending ||
471 vcpu->arch.slb_shadow.update_pending ||
472 vcpu->arch.dtl.update_pending))
473 return;
474
2e25aa5f
PM		 475 spin_lock(&vcpu->arch.vpa_update_lock);
476 if (vcpu->arch.vpa.update_pending) {
081f323b 477 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
55b665b0
PM		 478 if (vcpu->arch.vpa.pinned_addr)
479 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
2e25aa5f
PM		 480 }
481 if (vcpu->arch.dtl.update_pending) {
081f323b 482 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
2e25aa5f
PM		 483 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
484 vcpu->arch.dtl_index = 0;
485 }
486 if (vcpu->arch.slb_shadow.update_pending)
081f323b 487 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
2e25aa5f
PM		 488 spin_unlock(&vcpu->arch.vpa_update_lock);
489}
490
c7b67670
PM		 491/*
492 * Return the accumulated stolen time for the vcore up until `now'.
493 * The caller should hold the vcore lock.
494 */
495static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
496{
497 u64 p;
498
499 /*
500 * If we are the task running the vcore, then since we hold
501 * the vcore lock, we can't be preempted, so stolen_tb/preempt_tb
502 * can't be updated, so we don't need the tbacct_lock.
503 * If the vcore is inactive, it can't become active (since we
504 * hold the vcore lock), so the vcpu load/put functions won't
505 * update stolen_tb/preempt_tb, and we don't need tbacct_lock.
506 */
507 if (vc->vcore_state != VCORE_INACTIVE &&
508 vc->runner->arch.run_task != current) {
bf3d32e1 509 spin_lock_irq(&vc->runner->arch.tbacct_lock);
c7b67670
PM		 510 p = vc->stolen_tb;
511 if (vc->preempt_tb != TB_NIL)
512 p += now - vc->preempt_tb;
bf3d32e1 513 spin_unlock_irq(&vc->runner->arch.tbacct_lock);
c7b67670
PM		 514 } else {
515 p = vc->stolen_tb;
516 }
517 return p;
518}
519
0456ec4f
PM		 520static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
521 struct kvmppc_vcore *vc)
522{
523 struct dtl_entry *dt;
524 struct lppaca *vpa;
c7b67670
PM		 525 unsigned long stolen;
526 unsigned long core_stolen;
527 u64 now;
0456ec4f
PM		 528
529 dt = vcpu->arch.dtl_ptr;
530 vpa = vcpu->arch.vpa.pinned_addr;
c7b67670
PM		 531 now = mftb();
532 core_stolen = vcore_stolen_time(vc, now);
533 stolen = core_stolen - vcpu->arch.stolen_logged;
534 vcpu->arch.stolen_logged = core_stolen;
bf3d32e1 535 spin_lock_irq(&vcpu->arch.tbacct_lock);
c7b67670
PM		 536 stolen += vcpu->arch.busy_stolen;
537 vcpu->arch.busy_stolen = 0;
bf3d32e1 538 spin_unlock_irq(&vcpu->arch.tbacct_lock);
0456ec4f
PM		 539 if (!dt || !vpa)
540 return;
541 memset(dt, 0, sizeof(struct dtl_entry));
542 dt->dispatch_reason = 7;
543 dt->processor_id = vc->pcpu + vcpu->arch.ptid;
93b0f4dc 544 dt->timebase = now + vc->tb_offset;
c7b67670 545 dt->enqueue_to_dispatch_time = stolen;
0456ec4f
PM		 546 dt->srr0 = kvmppc_get_pc(vcpu);
547 dt->srr1 = vcpu->arch.shregs.msr;
548 ++dt;
549 if (dt == vcpu->arch.dtl.pinned_end)
550 dt = vcpu->arch.dtl.pinned_addr;
551 vcpu->arch.dtl_ptr = dt;
552 /* order writing *dt vs. writing vpa->dtl_idx */
553 smp_wmb();
554 vpa->dtl_idx = ++vcpu->arch.dtl_index;
c35635ef 555 vcpu->arch.dtl.dirty = true;
0456ec4f
PM		 556}
557
a8606e20
PM		 558int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
559{
560 unsigned long req = kvmppc_get_gpr(vcpu, 3);
561 unsigned long target, ret = H_SUCCESS;
562 struct kvm_vcpu *tvcpu;
8e591cb7 563 int idx, rc;
a8606e20
PM		 564
565 switch (req) {
c77162de 566 case H_ENTER:
2c9097e4 567 idx = srcu_read_lock(&vcpu->kvm->srcu);
c77162de
PM		 568 ret = kvmppc_virtmode_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
569 kvmppc_get_gpr(vcpu, 5),
570 kvmppc_get_gpr(vcpu, 6),
571 kvmppc_get_gpr(vcpu, 7));
2c9097e4 572 srcu_read_unlock(&vcpu->kvm->srcu, idx);
c77162de 573 break;
a8606e20 574 case H_CEDE:
a8606e20
PM		 575 break;
576 case H_PROD:
577 target = kvmppc_get_gpr(vcpu, 4);
578 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
579 if (!tvcpu) {
580 ret = H_PARAMETER;
581 break;
582 }
583 tvcpu->arch.prodded = 1;
584 smp_mb();
585 if (vcpu->arch.ceded) {
586 if (waitqueue_active(&vcpu->wq)) {
587 wake_up_interruptible(&vcpu->wq);
588 vcpu->stat.halt_wakeup++;
589 }
590 }
591 break;
592 case H_CONFER:
42d7604d
PM		 593 target = kvmppc_get_gpr(vcpu, 4);
594 if (target == -1)
595 break;
596 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
597 if (!tvcpu) {
598 ret = H_PARAMETER;
599 break;
600 }
601 kvm_vcpu_yield_to(tvcpu);
a8606e20
PM		 602 break;
603 case H_REGISTER_VPA:
604 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
605 kvmppc_get_gpr(vcpu, 5),
606 kvmppc_get_gpr(vcpu, 6));
607 break;
8e591cb7
ME		 608 case H_RTAS:
609 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
610 return RESUME_HOST;
611
c9438092 612 idx = srcu_read_lock(&vcpu->kvm->srcu);
8e591cb7 613 rc = kvmppc_rtas_hcall(vcpu);
c9438092 614 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8e591cb7
ME		 615
616 if (rc == -ENOENT)
617 return RESUME_HOST;
618 else if (rc == 0)
619 break;
620
621 /* Send the error out to userspace via KVM_RUN */
622 return rc;
bc5ad3f3
BH		 623
624 case H_XIRR:
625 case H_CPPR:
626 case H_EOI:
627 case H_IPI:
8e44ddc3
PM		 628 case H_IPOLL:
629 case H_XIRR_X:
bc5ad3f3
BH		 630 if (kvmppc_xics_enabled(vcpu)) {
631 ret = kvmppc_xics_hcall(vcpu, req);
632 break;
633 } /* fallthrough */
a8606e20
PM		 634 default:
635 return RESUME_HOST;
636 }
637 kvmppc_set_gpr(vcpu, 3, ret);
638 vcpu->arch.hcall_needed = 0;
639 return RESUME_GUEST;
640}
641
3a167bea
AK		 642static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
643 struct task_struct *tsk)
de56a948
PM		 644{
645 int r = RESUME_HOST;
646
647 vcpu->stat.sum_exits++;
648
649 run->exit_reason = KVM_EXIT_UNKNOWN;
650 run->ready_for_interrupt_injection = 1;
651 switch (vcpu->arch.trap) {
652 /* We're good on these - the host merely wanted to get our attention */
653 case BOOK3S_INTERRUPT_HV_DECREMENTER:
654 vcpu->stat.dec_exits++;
655 r = RESUME_GUEST;
656 break;
657 case BOOK3S_INTERRUPT_EXTERNAL:
5d00f66b 658 case BOOK3S_INTERRUPT_H_DOORBELL:
de56a948
PM		 659 vcpu->stat.ext_intr_exits++;
660 r = RESUME_GUEST;
661 break;
662 case BOOK3S_INTERRUPT_PERFMON:
663 r = RESUME_GUEST;
664 break;
b4072df4
PM		 665 case BOOK3S_INTERRUPT_MACHINE_CHECK:
666 /*
667 * Deliver a machine check interrupt to the guest.
668 * We have to do this, even if the host has handled the
669 * machine check, because machine checks use SRR0/1 and
670 * the interrupt might have trashed guest state in them.
671 */
672 kvmppc_book3s_queue_irqprio(vcpu,
673 BOOK3S_INTERRUPT_MACHINE_CHECK);
674 r = RESUME_GUEST;
675 break;
de56a948
PM		 676 case BOOK3S_INTERRUPT_PROGRAM:
677 {
678 ulong flags;
679 /*
680 * Normally program interrupts are delivered directly
681 * to the guest by the hardware, but we can get here
682 * as a result of a hypervisor emulation interrupt
683 * (e40) getting turned into a 700 by BML RTAS.
684 */
685 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
686 kvmppc_core_queue_program(vcpu, flags);
687 r = RESUME_GUEST;
688 break;
689 }
690 case BOOK3S_INTERRUPT_SYSCALL:
691 {
692 /* hcall - punt to userspace */
693 int i;
694
27025a60
LPF		 695 /* hypercall with MSR_PR has already been handled in rmode,
696 * and never reaches here.
697 */
698
de56a948
PM		 699 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
700 for (i = 0; i < 9; ++i)
701 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
702 run->exit_reason = KVM_EXIT_PAPR_HCALL;
703 vcpu->arch.hcall_needed = 1;
704 r = RESUME_HOST;
705 break;
706 }
707 /*
342d3db7
PM		 708 * We get these next two if the guest accesses a page which it thinks
709 * it has mapped but which is not actually present, either because
710 * it is for an emulated I/O device or because the corresonding
711 * host page has been paged out. Any other HDSI/HISI interrupts
712 * have been handled already.
de56a948
PM		 713 */
714 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
913d3ff9 715 r = RESUME_PAGE_FAULT;
de56a948
PM		 716 break;
717 case BOOK3S_INTERRUPT_H_INST_STORAGE:
913d3ff9
PM		 718 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
719 vcpu->arch.fault_dsisr = 0;
720 r = RESUME_PAGE_FAULT;
de56a948
PM		 721 break;
722 /*
723 * This occurs if the guest executes an illegal instruction.
724 * We just generate a program interrupt to the guest, since
725 * we don't emulate any guest instructions at this stage.
726 */
727 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
bd3048b8
ME		 728 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
729 r = RESUME_GUEST;
730 break;
731 /*
732 * This occurs if the guest (kernel or userspace), does something that
733 * is prohibited by HFSCR. We just generate a program interrupt to
734 * the guest.
735 */
736 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
737 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
de56a948
PM		 738 r = RESUME_GUEST;
739 break;
740 default:
741 kvmppc_dump_regs(vcpu);
742 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
743 vcpu->arch.trap, kvmppc_get_pc(vcpu),
744 vcpu->arch.shregs.msr);
f3271d4c 745 run->hw.hardware_exit_reason = vcpu->arch.trap;
de56a948 746 r = RESUME_HOST;
de56a948
PM		 747 break;
748 }
749
de56a948
PM		 750 return r;
751}
752
3a167bea
AK		 753static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
754 struct kvm_sregs *sregs)
de56a948
PM		 755{
756 int i;
757
de56a948 758 memset(sregs, 0, sizeof(struct kvm_sregs));
87916442 759 sregs->pvr = vcpu->arch.pvr;
de56a948
PM		 760 for (i = 0; i < vcpu->arch.slb_max; i++) {
761 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
762 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
763 }
764
765 return 0;
766}
767
3a167bea
AK		 768static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
769 struct kvm_sregs *sregs)
de56a948
PM		 770{
771 int i, j;
772
3a167bea 773 kvmppc_set_pvr_hv(vcpu, sregs->pvr);
de56a948
PM		 774
775 j = 0;
776 for (i = 0; i < vcpu->arch.slb_nr; i++) {
777 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
778 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
779 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
780 ++j;
781 }
782 }
783 vcpu->arch.slb_max = j;
784
785 return 0;
786}
787
a0144e2a
PM		 788static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr)
789{
790 struct kvmppc_vcore *vc = vcpu->arch.vcore;
791 u64 mask;
792
793 spin_lock(&vc->lock);
d682916a
AB		 794 /*
795 * If ILE (interrupt little-endian) has changed, update the
796 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
797 */
798 if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
799 struct kvm *kvm = vcpu->kvm;
800 struct kvm_vcpu *vcpu;
801 int i;
802
803 mutex_lock(&kvm->lock);
804 kvm_for_each_vcpu(i, vcpu, kvm) {
805 if (vcpu->arch.vcore != vc)
806 continue;
807 if (new_lpcr & LPCR_ILE)
808 vcpu->arch.intr_msr |= MSR_LE;
809 else
810 vcpu->arch.intr_msr &= ~MSR_LE;
811 }
812 mutex_unlock(&kvm->lock);
813 }
814
a0144e2a
PM		 815 /*
816 * Userspace can only modify DPFD (default prefetch depth),
817 * ILE (interrupt little-endian) and TC (translation control).
e0622bd9 818 * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
a0144e2a
PM		 819 */
820 mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
e0622bd9
PM		 821 if (cpu_has_feature(CPU_FTR_ARCH_207S))
822 mask |= LPCR_AIL;
a0144e2a
PM		 823 vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
824 spin_unlock(&vc->lock);
825}
826
3a167bea
AK		 827static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
828 union kvmppc_one_reg *val)
31f3438e 829{
a136a8bd
PM		 830 int r = 0;
831 long int i;
31f3438e 832
a136a8bd 833 switch (id) {
31f3438e 834 case KVM_REG_PPC_HIOR:
a136a8bd
PM		 835 *val = get_reg_val(id, 0);
836 break;
837 case KVM_REG_PPC_DABR:
838 *val = get_reg_val(id, vcpu->arch.dabr);
839 break;
8563bf52
PM		 840 case KVM_REG_PPC_DABRX:
841 *val = get_reg_val(id, vcpu->arch.dabrx);
842 break;
a136a8bd
PM		 843 case KVM_REG_PPC_DSCR:
844 *val = get_reg_val(id, vcpu->arch.dscr);
845 break;
846 case KVM_REG_PPC_PURR:
847 *val = get_reg_val(id, vcpu->arch.purr);
848 break;
849 case KVM_REG_PPC_SPURR:
850 *val = get_reg_val(id, vcpu->arch.spurr);
851 break;
852 case KVM_REG_PPC_AMR:
853 *val = get_reg_val(id, vcpu->arch.amr);
854 break;
855 case KVM_REG_PPC_UAMOR:
856 *val = get_reg_val(id, vcpu->arch.uamor);
857 break;
b005255e 858 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
a136a8bd
PM		 859 i = id - KVM_REG_PPC_MMCR0;
860 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
861 break;
862 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
863 i = id - KVM_REG_PPC_PMC1;
864 *val = get_reg_val(id, vcpu->arch.pmc[i]);
31f3438e 865 break;
b005255e
MN		 866 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
867 i = id - KVM_REG_PPC_SPMC1;
868 *val = get_reg_val(id, vcpu->arch.spmc[i]);
869 break;
14941789
PM		 870 case KVM_REG_PPC_SIAR:
871 *val = get_reg_val(id, vcpu->arch.siar);
872 break;
873 case KVM_REG_PPC_SDAR:
874 *val = get_reg_val(id, vcpu->arch.sdar);
875 break;
b005255e
MN		 876 case KVM_REG_PPC_SIER:
877 *val = get_reg_val(id, vcpu->arch.sier);
a8bd19ef 878 break;
b005255e
MN		 879 case KVM_REG_PPC_IAMR:
880 *val = get_reg_val(id, vcpu->arch.iamr);
881 break;
b005255e
MN		 882 case KVM_REG_PPC_PSPB:
883 *val = get_reg_val(id, vcpu->arch.pspb);
884 break;
b005255e
MN		 885 case KVM_REG_PPC_DPDES:
886 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
887 break;
888 case KVM_REG_PPC_DAWR:
889 *val = get_reg_val(id, vcpu->arch.dawr);
890 break;
891 case KVM_REG_PPC_DAWRX:
892 *val = get_reg_val(id, vcpu->arch.dawrx);
893 break;
894 case KVM_REG_PPC_CIABR:
895 *val = get_reg_val(id, vcpu->arch.ciabr);
896 break;
b005255e
MN		 897 case KVM_REG_PPC_CSIGR:
898 *val = get_reg_val(id, vcpu->arch.csigr);
899 break;
900 case KVM_REG_PPC_TACR:
901 *val = get_reg_val(id, vcpu->arch.tacr);
902 break;
903 case KVM_REG_PPC_TCSCR:
904 *val = get_reg_val(id, vcpu->arch.tcscr);
905 break;
906 case KVM_REG_PPC_PID:
907 *val = get_reg_val(id, vcpu->arch.pid);
908 break;
909 case KVM_REG_PPC_ACOP:
910 *val = get_reg_val(id, vcpu->arch.acop);
911 break;
912 case KVM_REG_PPC_WORT:
913 *val = get_reg_val(id, vcpu->arch.wort);
a8bd19ef 914 break;
55b665b0
PM		 915 case KVM_REG_PPC_VPA_ADDR:
916 spin_lock(&vcpu->arch.vpa_update_lock);
917 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
918 spin_unlock(&vcpu->arch.vpa_update_lock);
919 break;
920 case KVM_REG_PPC_VPA_SLB:
921 spin_lock(&vcpu->arch.vpa_update_lock);
922 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
923 val->vpaval.length = vcpu->arch.slb_shadow.len;
924 spin_unlock(&vcpu->arch.vpa_update_lock);
925 break;
926 case KVM_REG_PPC_VPA_DTL:
927 spin_lock(&vcpu->arch.vpa_update_lock);
928 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
929 val->vpaval.length = vcpu->arch.dtl.len;
930 spin_unlock(&vcpu->arch.vpa_update_lock);
931 break;
93b0f4dc
PM		 932 case KVM_REG_PPC_TB_OFFSET:
933 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
934 break;
a0144e2a
PM		 935 case KVM_REG_PPC_LPCR:
936 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
937 break;
4b8473c9
PM		 938 case KVM_REG_PPC_PPR:
939 *val = get_reg_val(id, vcpu->arch.ppr);
940 break;
a7d80d01
MN		 941#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
942 case KVM_REG_PPC_TFHAR:
943 *val = get_reg_val(id, vcpu->arch.tfhar);
944 break;
945 case KVM_REG_PPC_TFIAR:
946 *val = get_reg_val(id, vcpu->arch.tfiar);
947 break;
948 case KVM_REG_PPC_TEXASR:
949 *val = get_reg_val(id, vcpu->arch.texasr);
950 break;
951 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
952 i = id - KVM_REG_PPC_TM_GPR0;
953 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
954 break;
955 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
956 {
957 int j;
958 i = id - KVM_REG_PPC_TM_VSR0;
959 if (i < 32)
960 for (j = 0; j < TS_FPRWIDTH; j++)
961 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
962 else {
963 if (cpu_has_feature(CPU_FTR_ALTIVEC))
964 val->vval = vcpu->arch.vr_tm.vr[i-32];
965 else
966 r = -ENXIO;
967 }
968 break;
969 }
970 case KVM_REG_PPC_TM_CR:
971 *val = get_reg_val(id, vcpu->arch.cr_tm);
972 break;
973 case KVM_REG_PPC_TM_LR:
974 *val = get_reg_val(id, vcpu->arch.lr_tm);
975 break;
976 case KVM_REG_PPC_TM_CTR:
977 *val = get_reg_val(id, vcpu->arch.ctr_tm);
978 break;
979 case KVM_REG_PPC_TM_FPSCR:
980 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
981 break;
982 case KVM_REG_PPC_TM_AMR:
983 *val = get_reg_val(id, vcpu->arch.amr_tm);
984 break;
985 case KVM_REG_PPC_TM_PPR:
986 *val = get_reg_val(id, vcpu->arch.ppr_tm);
987 break;
988 case KVM_REG_PPC_TM_VRSAVE:
989 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
990 break;
991 case KVM_REG_PPC_TM_VSCR:
992 if (cpu_has_feature(CPU_FTR_ALTIVEC))
993 *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
994 else
995 r = -ENXIO;
996 break;
997 case KVM_REG_PPC_TM_DSCR:
998 *val = get_reg_val(id, vcpu->arch.dscr_tm);
999 break;
1000 case KVM_REG_PPC_TM_TAR:
1001 *val = get_reg_val(id, vcpu->arch.tar_tm);
1002 break;
1003#endif
388cc6e1
PM		 1004 case KVM_REG_PPC_ARCH_COMPAT:
1005 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1006 break;
31f3438e 1007 default:
a136a8bd 1008 r = -EINVAL;
31f3438e
PM		 1009 break;
1010 }
1011
1012 return r;
1013}
1014
3a167bea
AK		 1015static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1016 union kvmppc_one_reg *val)
31f3438e 1017{
a136a8bd
PM		 1018 int r = 0;
1019 long int i;
55b665b0 1020 unsigned long addr, len;
31f3438e 1021
a136a8bd 1022 switch (id) {
31f3438e 1023 case KVM_REG_PPC_HIOR:
31f3438e 1024 /* Only allow this to be set to zero */
a136a8bd 1025 if (set_reg_val(id, *val))
31f3438e
PM		 1026 r = -EINVAL;
1027 break;
a136a8bd
PM		 1028 case KVM_REG_PPC_DABR:
1029 vcpu->arch.dabr = set_reg_val(id, *val);
1030 break;
8563bf52
PM		 1031 case KVM_REG_PPC_DABRX:
1032 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1033 break;
a136a8bd
PM		 1034 case KVM_REG_PPC_DSCR:
1035 vcpu->arch.dscr = set_reg_val(id, *val);
1036 break;
1037 case KVM_REG_PPC_PURR:
1038 vcpu->arch.purr = set_reg_val(id, *val);
1039 break;
1040 case KVM_REG_PPC_SPURR:
1041 vcpu->arch.spurr = set_reg_val(id, *val);
1042 break;
1043 case KVM_REG_PPC_AMR:
1044 vcpu->arch.amr = set_reg_val(id, *val);
1045 break;
1046 case KVM_REG_PPC_UAMOR:
1047 vcpu->arch.uamor = set_reg_val(id, *val);
1048 break;
b005255e 1049 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
a136a8bd
PM		 1050 i = id - KVM_REG_PPC_MMCR0;
1051 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1052 break;
1053 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1054 i = id - KVM_REG_PPC_PMC1;
1055 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1056 break;
b005255e
MN		 1057 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1058 i = id - KVM_REG_PPC_SPMC1;
1059 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1060 break;
14941789
PM		 1061 case KVM_REG_PPC_SIAR:
1062 vcpu->arch.siar = set_reg_val(id, *val);
1063 break;
1064 case KVM_REG_PPC_SDAR:
1065 vcpu->arch.sdar = set_reg_val(id, *val);
1066 break;
b005255e
MN		 1067 case KVM_REG_PPC_SIER:
1068 vcpu->arch.sier = set_reg_val(id, *val);
a8bd19ef 1069 break;
b005255e
MN		 1070 case KVM_REG_PPC_IAMR:
1071 vcpu->arch.iamr = set_reg_val(id, *val);
1072 break;
b005255e
MN		 1073 case KVM_REG_PPC_PSPB:
1074 vcpu->arch.pspb = set_reg_val(id, *val);
1075 break;
b005255e
MN		 1076 case KVM_REG_PPC_DPDES:
1077 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1078 break;
1079 case KVM_REG_PPC_DAWR:
1080 vcpu->arch.dawr = set_reg_val(id, *val);
1081 break;
1082 case KVM_REG_PPC_DAWRX:
1083 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1084 break;
1085 case KVM_REG_PPC_CIABR:
1086 vcpu->arch.ciabr = set_reg_val(id, *val);
1087 /* Don't allow setting breakpoints in hypervisor code */
1088 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1089 vcpu->arch.ciabr &= ~CIABR_PRIV; /* disable */
1090 break;
b005255e
MN		 1091 case KVM_REG_PPC_CSIGR:
1092 vcpu->arch.csigr = set_reg_val(id, *val);
1093 break;
1094 case KVM_REG_PPC_TACR:
1095 vcpu->arch.tacr = set_reg_val(id, *val);
1096 break;
1097 case KVM_REG_PPC_TCSCR:
1098 vcpu->arch.tcscr = set_reg_val(id, *val);
1099 break;
1100 case KVM_REG_PPC_PID:
1101 vcpu->arch.pid = set_reg_val(id, *val);
1102 break;
1103 case KVM_REG_PPC_ACOP:
1104 vcpu->arch.acop = set_reg_val(id, *val);
1105 break;
1106 case KVM_REG_PPC_WORT:
1107 vcpu->arch.wort = set_reg_val(id, *val);
a8bd19ef 1108 break;
55b665b0
PM		 1109 case KVM_REG_PPC_VPA_ADDR:
1110 addr = set_reg_val(id, *val);
1111 r = -EINVAL;
1112 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1113 vcpu->arch.dtl.next_gpa))
1114 break;
1115 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1116 break;
1117 case KVM_REG_PPC_VPA_SLB:
1118 addr = val->vpaval.addr;
1119 len = val->vpaval.length;
1120 r = -EINVAL;
1121 if (addr && !vcpu->arch.vpa.next_gpa)
1122 break;
1123 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1124 break;
1125 case KVM_REG_PPC_VPA_DTL:
1126 addr = val->vpaval.addr;
1127 len = val->vpaval.length;
1128 r = -EINVAL;
9f8c8c78
PM		 1129 if (addr && (len < sizeof(struct dtl_entry) ||
1130 !vcpu->arch.vpa.next_gpa))
55b665b0
PM		 1131 break;
1132 len -= len % sizeof(struct dtl_entry);
1133 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1134 break;
93b0f4dc
PM		 1135 case KVM_REG_PPC_TB_OFFSET:
1136 /* round up to multiple of 2^24 */
1137 vcpu->arch.vcore->tb_offset =
1138 ALIGN(set_reg_val(id, *val), 1UL << 24);
1139 break;
a0144e2a
PM		 1140 case KVM_REG_PPC_LPCR:
1141 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val));
1142 break;
4b8473c9
PM		 1143 case KVM_REG_PPC_PPR:
1144 vcpu->arch.ppr = set_reg_val(id, *val);
1145 break;
a7d80d01
MN		 1146#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1147 case KVM_REG_PPC_TFHAR:
1148 vcpu->arch.tfhar = set_reg_val(id, *val);
1149 break;
1150 case KVM_REG_PPC_TFIAR:
1151 vcpu->arch.tfiar = set_reg_val(id, *val);
1152 break;
1153 case KVM_REG_PPC_TEXASR:
1154 vcpu->arch.texasr = set_reg_val(id, *val);
1155 break;
1156 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1157 i = id - KVM_REG_PPC_TM_GPR0;
1158 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1159 break;
1160 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1161 {
1162 int j;
1163 i = id - KVM_REG_PPC_TM_VSR0;
1164 if (i < 32)
1165 for (j = 0; j < TS_FPRWIDTH; j++)
1166 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1167 else
1168 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1169 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1170 else
1171 r = -ENXIO;
1172 break;
1173 }
1174 case KVM_REG_PPC_TM_CR:
1175 vcpu->arch.cr_tm = set_reg_val(id, *val);
1176 break;
1177 case KVM_REG_PPC_TM_LR:
1178 vcpu->arch.lr_tm = set_reg_val(id, *val);
1179 break;
1180 case KVM_REG_PPC_TM_CTR:
1181 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1182 break;
1183 case KVM_REG_PPC_TM_FPSCR:
1184 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1185 break;
1186 case KVM_REG_PPC_TM_AMR:
1187 vcpu->arch.amr_tm = set_reg_val(id, *val);
1188 break;
1189 case KVM_REG_PPC_TM_PPR:
1190 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1191 break;
1192 case KVM_REG_PPC_TM_VRSAVE:
1193 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1194 break;
1195 case KVM_REG_PPC_TM_VSCR:
1196 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1197 vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1198 else
1199 r = - ENXIO;
1200 break;
1201 case KVM_REG_PPC_TM_DSCR:
1202 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1203 break;
1204 case KVM_REG_PPC_TM_TAR:
1205 vcpu->arch.tar_tm = set_reg_val(id, *val);
1206 break;
1207#endif
388cc6e1
PM		 1208 case KVM_REG_PPC_ARCH_COMPAT:
1209 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1210 break;
31f3438e 1211 default:
a136a8bd 1212 r = -EINVAL;
31f3438e
PM		 1213 break;
1214 }
1215
1216 return r;
1217}
1218
3a167bea
AK		 1219static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1220 unsigned int id)
de56a948
PM		 1221{
1222 struct kvm_vcpu *vcpu;
371fefd6
PM		 1223 int err = -EINVAL;
1224 int core;
1225 struct kvmppc_vcore *vcore;
de56a948 1226
3102f784 1227 core = id / threads_per_subcore;
371fefd6
PM		 1228 if (core >= KVM_MAX_VCORES)
1229 goto out;
1230
1231 err = -ENOMEM;
6b75e6bf 1232 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
de56a948
PM		 1233 if (!vcpu)
1234 goto out;
1235
1236 err = kvm_vcpu_init(vcpu, kvm, id);
1237 if (err)
1238 goto free_vcpu;
1239
1240 vcpu->arch.shared = &vcpu->arch.shregs;
5deb8e7a
AG		 1241#ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1242 /*
1243 * The shared struct is never shared on HV,
1244 * so we can always use host endianness
1245 */
1246#ifdef __BIG_ENDIAN__
1247 vcpu->arch.shared_big_endian = true;
1248#else
1249 vcpu->arch.shared_big_endian = false;
1250#endif
1251#endif
de56a948
PM		 1252 vcpu->arch.mmcr[0] = MMCR0_FC;
1253 vcpu->arch.ctrl = CTRL_RUNLATCH;
1254 /* default to host PVR, since we can't spoof it */
3a167bea 1255 kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2e25aa5f 1256 spin_lock_init(&vcpu->arch.vpa_update_lock);
c7b67670
PM		 1257 spin_lock_init(&vcpu->arch.tbacct_lock);
1258 vcpu->arch.busy_preempt = TB_NIL;
d682916a 1259 vcpu->arch.intr_msr = MSR_SF | MSR_ME;
de56a948 1260
de56a948
PM		 1261 kvmppc_mmu_book3s_hv_init(vcpu);
1262
8455d79e 1263 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
371fefd6
PM		 1264
1265 init_waitqueue_head(&vcpu->arch.cpu_run);
1266
1267 mutex_lock(&kvm->lock);
1268 vcore = kvm->arch.vcores[core];
1269 if (!vcore) {
1270 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1271 if (vcore) {
1272 INIT_LIST_HEAD(&vcore->runnable_threads);
1273 spin_lock_init(&vcore->lock);
19ccb76a 1274 init_waitqueue_head(&vcore->wq);
c7b67670 1275 vcore->preempt_tb = TB_NIL;
a0144e2a 1276 vcore->lpcr = kvm->arch.lpcr;
3102f784 1277 vcore->first_vcpuid = core * threads_per_subcore;
e0b7ec05 1278 vcore->kvm = kvm;
371fefd6
PM		 1279 }
1280 kvm->arch.vcores[core] = vcore;
1b400ba0 1281 kvm->arch.online_vcores++;
371fefd6
PM		 1282 }
1283 mutex_unlock(&kvm->lock);
1284
1285 if (!vcore)
1286 goto free_vcpu;
1287
1288 spin_lock(&vcore->lock);
1289 ++vcore->num_threads;
371fefd6
PM		 1290 spin_unlock(&vcore->lock);
1291 vcpu->arch.vcore = vcore;
e0b7ec05 1292 vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
371fefd6 1293
af8f38b3
AG		 1294 vcpu->arch.cpu_type = KVM_CPU_3S_64;
1295 kvmppc_sanity_check(vcpu);
1296
de56a948
PM		 1297 return vcpu;
1298
1299free_vcpu:
6b75e6bf 1300 kmem_cache_free(kvm_vcpu_cache, vcpu);
de56a948
PM		 1301out:
1302 return ERR_PTR(err);
1303}
1304
c35635ef
PM		 1305static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1306{
1307 if (vpa->pinned_addr)
1308 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1309 vpa->dirty);
1310}
1311
3a167bea 1312static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
de56a948 1313{
2e25aa5f 1314 spin_lock(&vcpu->arch.vpa_update_lock);
c35635ef
PM		 1315 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1316 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1317 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2e25aa5f 1318 spin_unlock(&vcpu->arch.vpa_update_lock);
de56a948 1319 kvm_vcpu_uninit(vcpu);
6b75e6bf 1320 kmem_cache_free(kvm_vcpu_cache, vcpu);
de56a948
PM		 1321}
1322
3a167bea
AK		 1323static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1324{
1325 /* Indicate we want to get back into the guest */
1326 return 1;
1327}
1328
19ccb76a 1329static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
371fefd6 1330{
19ccb76a 1331 unsigned long dec_nsec, now;
371fefd6 1332
19ccb76a
PM		 1333 now = get_tb();
1334 if (now > vcpu->arch.dec_expires) {
1335 /* decrementer has already gone negative */
1336 kvmppc_core_queue_dec(vcpu);
7e28e60e 1337 kvmppc_core_prepare_to_enter(vcpu);
19ccb76a 1338 return;
371fefd6 1339 }
19ccb76a
PM		 1340 dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1341 / tb_ticks_per_sec;
1342 hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1343 HRTIMER_MODE_REL);
1344 vcpu->arch.timer_running = 1;
371fefd6
PM		 1345}
1346
19ccb76a 1347static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
371fefd6 1348{
19ccb76a
PM		 1349 vcpu->arch.ceded = 0;
1350 if (vcpu->arch.timer_running) {
1351 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1352 vcpu->arch.timer_running = 0;
1353 }
371fefd6
PM		 1354}
1355
e0b7ec05 1356extern void __kvmppc_vcore_entry(void);
de56a948 1357
371fefd6
PM		 1358static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1359 struct kvm_vcpu *vcpu)
de56a948 1360{
c7b67670
PM		 1361 u64 now;
1362
371fefd6
PM		 1363 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1364 return;
bf3d32e1 1365 spin_lock_irq(&vcpu->arch.tbacct_lock);
c7b67670
PM		 1366 now = mftb();
1367 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1368 vcpu->arch.stolen_logged;
1369 vcpu->arch.busy_preempt = now;
1370 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
bf3d32e1 1371 spin_unlock_irq(&vcpu->arch.tbacct_lock);
371fefd6 1372 --vc->n_runnable;
371fefd6
PM		 1373 list_del(&vcpu->arch.run_list);
1374}
1375
f0888f70
PM		 1376static int kvmppc_grab_hwthread(int cpu)
1377{
1378 struct paca_struct *tpaca;
1379 long timeout = 1000;
1380
1381 tpaca = &paca[cpu];
1382
1383 /* Ensure the thread won't go into the kernel if it wakes */
1384 tpaca->kvm_hstate.hwthread_req = 1;
7b444c67 1385 tpaca->kvm_hstate.kvm_vcpu = NULL;
f0888f70
PM		 1386
1387 /*
1388 * If the thread is already executing in the kernel (e.g. handling
1389 * a stray interrupt), wait for it to get back to nap mode.
1390 * The smp_mb() is to ensure that our setting of hwthread_req
1391 * is visible before we look at hwthread_state, so if this
1392 * races with the code at system_reset_pSeries and the thread
1393 * misses our setting of hwthread_req, we are sure to see its
1394 * setting of hwthread_state, and vice versa.
1395 */
1396 smp_mb();
1397 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1398 if (--timeout <= 0) {
1399 pr_err("KVM: couldn't grab cpu %d\n", cpu);
1400 return -EBUSY;
1401 }
1402 udelay(1);
1403 }
1404 return 0;
1405}
1406
1407static void kvmppc_release_hwthread(int cpu)
1408{
1409 struct paca_struct *tpaca;
1410
1411 tpaca = &paca[cpu];
1412 tpaca->kvm_hstate.hwthread_req = 0;
1413 tpaca->kvm_hstate.kvm_vcpu = NULL;
1414}
1415
371fefd6
PM		 1416static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
1417{
1418 int cpu;
1419 struct paca_struct *tpaca;
1420 struct kvmppc_vcore *vc = vcpu->arch.vcore;
1421
19ccb76a
PM		 1422 if (vcpu->arch.timer_running) {
1423 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1424 vcpu->arch.timer_running = 0;
1425 }
371fefd6
PM		 1426 cpu = vc->pcpu + vcpu->arch.ptid;
1427 tpaca = &paca[cpu];
1428 tpaca->kvm_hstate.kvm_vcpu = vcpu;
1429 tpaca->kvm_hstate.kvm_vcore = vc;
e0b7ec05 1430 tpaca->kvm_hstate.ptid = vcpu->arch.ptid;
19ccb76a 1431 vcpu->cpu = vc->pcpu;
371fefd6 1432 smp_wmb();
251da038 1433#if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
e0b7ec05 1434 if (cpu != smp_processor_id()) {
371fefd6 1435 xics_wake_cpu(cpu);
e0b7ec05
PM		 1436 if (vcpu->arch.ptid)
1437 ++vc->n_woken;
de56a948 1438 }
371fefd6
PM		 1439#endif
1440}
de56a948 1441
371fefd6
PM		 1442static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc)
1443{
1444 int i;
1445
1446 HMT_low();
1447 i = 0;
1448 while (vc->nap_count < vc->n_woken) {
1449 if (++i >= 1000000) {
1450 pr_err("kvmppc_wait_for_nap timeout %d %d\n",
1451 vc->nap_count, vc->n_woken);
1452 break;
1453 }
1454 cpu_relax();
1455 }
1456 HMT_medium();
1457}
1458
1459/*
1460 * Check that we are on thread 0 and that any other threads in
7b444c67
PM		 1461 * this core are off-line. Then grab the threads so they can't
1462 * enter the kernel.
371fefd6
PM		 1463 */
1464static int on_primary_thread(void)
1465{
1466 int cpu = smp_processor_id();
3102f784 1467 int thr;
371fefd6 1468
3102f784
ME		 1469 /* Are we on a primary subcore? */
1470 if (cpu_thread_in_subcore(cpu))
371fefd6 1471 return 0;
3102f784
ME		 1472
1473 thr = 0;
1474 while (++thr < threads_per_subcore)
371fefd6
PM		 1475 if (cpu_online(cpu + thr))
1476 return 0;
7b444c67
PM		 1477
1478 /* Grab all hw threads so they can't go into the kernel */
3102f784 1479 for (thr = 1; thr < threads_per_subcore; ++thr) {
7b444c67
PM		 1480 if (kvmppc_grab_hwthread(cpu + thr)) {
1481 /* Couldn't grab one; let the others go */
1482 do {
1483 kvmppc_release_hwthread(cpu + thr);
1484 } while (--thr > 0);
1485 return 0;
1486 }
1487 }
371fefd6
PM		 1488 return 1;
1489}
1490
1491/*
1492 * Run a set of guest threads on a physical core.
1493 * Called with vc->lock held.
1494 */
913d3ff9 1495static void kvmppc_run_core(struct kvmppc_vcore *vc)
371fefd6 1496{
e0b7ec05 1497 struct kvm_vcpu *vcpu, *vnext;
371fefd6
PM		 1498 long ret;
1499 u64 now;
e0b7ec05 1500 int i, need_vpa_update;
2c9097e4 1501 int srcu_idx;
913d3ff9 1502 struct kvm_vcpu *vcpus_to_update[threads_per_core];
371fefd6
PM		 1503
1504 /* don't start if any threads have a signal pending */
081f323b
PM		 1505 need_vpa_update = 0;
1506 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
371fefd6 1507 if (signal_pending(vcpu->arch.run_task))
913d3ff9
PM		 1508 return;
1509 if (vcpu->arch.vpa.update_pending ||
1510 vcpu->arch.slb_shadow.update_pending ||
1511 vcpu->arch.dtl.update_pending)
1512 vcpus_to_update[need_vpa_update++] = vcpu;
081f323b
PM		 1513 }
1514
1515 /*
1516 * Initialize *vc, in particular vc->vcore_state, so we can
1517 * drop the vcore lock if necessary.
1518 */
1519 vc->n_woken = 0;
1520 vc->nap_count = 0;
1521 vc->entry_exit_count = 0;
2f12f034 1522 vc->vcore_state = VCORE_STARTING;
081f323b
PM		 1523 vc->in_guest = 0;
1524 vc->napping_threads = 0;
1525
1526 /*
1527 * Updating any of the vpas requires calling kvmppc_pin_guest_page,
1528 * which can't be called with any spinlocks held.
1529 */
1530 if (need_vpa_update) {
1531 spin_unlock(&vc->lock);
913d3ff9
PM		 1532 for (i = 0; i < need_vpa_update; ++i)
1533 kvmppc_update_vpas(vcpus_to_update[i]);
081f323b
PM		 1534 spin_lock(&vc->lock);
1535 }
de56a948 1536
7b444c67 1537 /*
3102f784
ME		 1538 * Make sure we are running on primary threads, and that secondary
1539 * threads are offline. Also check if the number of threads in this
1540 * guest are greater than the current system threads per guest.
7b444c67 1541 */
3102f784
ME		 1542 if ((threads_per_core > 1) &&
1543 ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
7b444c67
PM		 1544 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1545 vcpu->arch.ret = -EBUSY;
1546 goto out;
1547 }
1548
3102f784 1549
371fefd6 1550 vc->pcpu = smp_processor_id();
2e25aa5f 1551 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
371fefd6 1552 kvmppc_start_thread(vcpu);
0456ec4f 1553 kvmppc_create_dtl_entry(vcpu, vc);
2e25aa5f 1554 }
371fefd6 1555
e0b7ec05
PM		 1556 /* Set this explicitly in case thread 0 doesn't have a vcpu */
1557 get_paca()->kvm_hstate.kvm_vcore = vc;
1558 get_paca()->kvm_hstate.ptid = 0;
1559
2f12f034 1560 vc->vcore_state = VCORE_RUNNING;
19ccb76a 1561 preempt_disable();
371fefd6 1562 spin_unlock(&vc->lock);
de56a948 1563
371fefd6 1564 kvm_guest_enter();
2c9097e4 1565
e0b7ec05 1566 srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2c9097e4 1567
e0b7ec05 1568 __kvmppc_vcore_entry();
de56a948 1569
371fefd6 1570 spin_lock(&vc->lock);
19ccb76a
PM		 1571 /* disable sending of IPIs on virtual external irqs */
1572 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1573 vcpu->cpu = -1;
1574 /* wait for secondary threads to finish writing their state to memory */
371fefd6
PM		 1575 if (vc->nap_count < vc->n_woken)
1576 kvmppc_wait_for_nap(vc);
3102f784 1577 for (i = 0; i < threads_per_subcore; ++i)
2f12f034 1578 kvmppc_release_hwthread(vc->pcpu + i);
371fefd6 1579 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
19ccb76a 1580 vc->vcore_state = VCORE_EXITING;
371fefd6
PM		 1581 spin_unlock(&vc->lock);
1582
e0b7ec05 1583 srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2c9097e4 1584
371fefd6
PM		 1585 /* make sure updates to secondary vcpu structs are visible now */
1586 smp_mb();
de56a948
PM		 1587 kvm_guest_exit();
1588
1589 preempt_enable();
c08ac06a 1590 cond_resched();
de56a948 1591
913d3ff9 1592 spin_lock(&vc->lock);
de56a948 1593 now = get_tb();
371fefd6
PM		 1594 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1595 /* cancel pending dec exception if dec is positive */
1596 if (now < vcpu->arch.dec_expires &&
1597 kvmppc_core_pending_dec(vcpu))
1598 kvmppc_core_dequeue_dec(vcpu);
19ccb76a
PM		 1599
1600 ret = RESUME_GUEST;
1601 if (vcpu->arch.trap)
3a167bea
AK		 1602 ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
1603 vcpu->arch.run_task);
19ccb76a 1604
371fefd6
PM		 1605 vcpu->arch.ret = ret;
1606 vcpu->arch.trap = 0;
19ccb76a
PM		 1607
1608 if (vcpu->arch.ceded) {
e59d24e6 1609 if (!is_kvmppc_resume_guest(ret))
19ccb76a
PM		 1610 kvmppc_end_cede(vcpu);
1611 else
1612 kvmppc_set_timer(vcpu);
1613 }
371fefd6 1614 }
de56a948
PM		 1615
1616 out:
19ccb76a 1617 vc->vcore_state = VCORE_INACTIVE;
371fefd6
PM		 1618 list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1619 arch.run_list) {
e59d24e6 1620 if (!is_kvmppc_resume_guest(vcpu->arch.ret)) {
371fefd6
PM		 1621 kvmppc_remove_runnable(vc, vcpu);
1622 wake_up(&vcpu->arch.cpu_run);
1623 }
1624 }
371fefd6
PM		 1625}
1626
19ccb76a
PM		 1627/*
1628 * Wait for some other vcpu thread to execute us, and
1629 * wake us up when we need to handle something in the host.
1630 */
1631static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
371fefd6 1632{
371fefd6
PM		 1633 DEFINE_WAIT(wait);
1634
19ccb76a
PM		 1635 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
1636 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
1637 schedule();
1638 finish_wait(&vcpu->arch.cpu_run, &wait);
1639}
1640
1641/*
1642 * All the vcpus in this vcore are idle, so wait for a decrementer
1643 * or external interrupt to one of the vcpus. vc->lock is held.
1644 */
1645static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
1646{
1647 DEFINE_WAIT(wait);
19ccb76a
PM		 1648
1649 prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
1650 vc->vcore_state = VCORE_SLEEPING;
1651 spin_unlock(&vc->lock);
913d3ff9 1652 schedule();
19ccb76a
PM		 1653 finish_wait(&vc->wq, &wait);
1654 spin_lock(&vc->lock);
1655 vc->vcore_state = VCORE_INACTIVE;
1656}
371fefd6 1657
19ccb76a
PM		 1658static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1659{
1660 int n_ceded;
19ccb76a
PM		 1661 struct kvmppc_vcore *vc;
1662 struct kvm_vcpu *v, *vn;
9e368f29 1663
371fefd6
PM		 1664 kvm_run->exit_reason = 0;
1665 vcpu->arch.ret = RESUME_GUEST;
1666 vcpu->arch.trap = 0;
2f12f034 1667 kvmppc_update_vpas(vcpu);
371fefd6 1668
371fefd6
PM		 1669 /*
1670 * Synchronize with other threads in this virtual core
1671 */
1672 vc = vcpu->arch.vcore;
1673 spin_lock(&vc->lock);
19ccb76a 1674 vcpu->arch.ceded = 0;
371fefd6
PM		 1675 vcpu->arch.run_task = current;
1676 vcpu->arch.kvm_run = kvm_run;
c7b67670 1677 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
19ccb76a 1678 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
c7b67670 1679 vcpu->arch.busy_preempt = TB_NIL;
371fefd6
PM		 1680 list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
1681 ++vc->n_runnable;
1682
19ccb76a
PM		 1683 /*
1684 * This happens the first time this is called for a vcpu.
1685 * If the vcore is already running, we may be able to start
1686 * this thread straight away and have it join in.
1687 */
8455d79e 1688 if (!signal_pending(current)) {
19ccb76a
PM		 1689 if (vc->vcore_state == VCORE_RUNNING &&
1690 VCORE_EXIT_COUNT(vc) == 0) {
2f12f034 1691 kvmppc_create_dtl_entry(vcpu, vc);
19ccb76a 1692 kvmppc_start_thread(vcpu);
8455d79e
PM		 1693 } else if (vc->vcore_state == VCORE_SLEEPING) {
1694 wake_up(&vc->wq);
371fefd6
PM		 1695 }
1696
8455d79e 1697 }
371fefd6 1698
19ccb76a
PM		 1699 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1700 !signal_pending(current)) {
8455d79e 1701 if (vc->vcore_state != VCORE_INACTIVE) {
19ccb76a
PM		 1702 spin_unlock(&vc->lock);
1703 kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
1704 spin_lock(&vc->lock);
1705 continue;
1706 }
19ccb76a
PM		 1707 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
1708 arch.run_list) {
7e28e60e 1709 kvmppc_core_prepare_to_enter(v);
19ccb76a
PM		 1710 if (signal_pending(v->arch.run_task)) {
1711 kvmppc_remove_runnable(vc, v);
1712 v->stat.signal_exits++;
1713 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
1714 v->arch.ret = -EINTR;
1715 wake_up(&v->arch.cpu_run);
1716 }
1717 }
8455d79e
PM		 1718 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1719 break;
1720 vc->runner = vcpu;
1721 n_ceded = 0;
4619ac88 1722 list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
8455d79e
PM		 1723 if (!v->arch.pending_exceptions)
1724 n_ceded += v->arch.ceded;
4619ac88
PM		 1725 else
1726 v->arch.ceded = 0;
1727 }
8455d79e
PM		 1728 if (n_ceded == vc->n_runnable)
1729 kvmppc_vcore_blocked(vc);
1730 else
1731 kvmppc_run_core(vc);
0456ec4f 1732 vc->runner = NULL;
19ccb76a 1733 }
371fefd6 1734
8455d79e
PM		 1735 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1736 (vc->vcore_state == VCORE_RUNNING ||
1737 vc->vcore_state == VCORE_EXITING)) {
1738 spin_unlock(&vc->lock);
1739 kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
1740 spin_lock(&vc->lock);
1741 }
1742
1743 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
1744 kvmppc_remove_runnable(vc, vcpu);
1745 vcpu->stat.signal_exits++;
1746 kvm_run->exit_reason = KVM_EXIT_INTR;
1747 vcpu->arch.ret = -EINTR;
1748 }
1749
1750 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
1751 /* Wake up some vcpu to run the core */
1752 v = list_first_entry(&vc->runnable_threads,
1753 struct kvm_vcpu, arch.run_list);
1754 wake_up(&v->arch.cpu_run);
371fefd6
PM		 1755 }
1756
371fefd6 1757 spin_unlock(&vc->lock);
371fefd6 1758 return vcpu->arch.ret;
de56a948
PM		 1759}
1760
3a167bea 1761static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
a8606e20
PM		 1762{
1763 int r;
913d3ff9 1764 int srcu_idx;
a8606e20 1765
af8f38b3
AG		 1766 if (!vcpu->arch.sane) {
1767 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1768 return -EINVAL;
1769 }
1770
25051b5a
SW		 1771 kvmppc_core_prepare_to_enter(vcpu);
1772
19ccb76a
PM		 1773 /* No need to go into the guest when all we'll do is come back out */
1774 if (signal_pending(current)) {
1775 run->exit_reason = KVM_EXIT_INTR;
1776 return -EINTR;
1777 }
1778
32fad281
PM		 1779 atomic_inc(&vcpu->kvm->arch.vcpus_running);
1780 /* Order vcpus_running vs. rma_setup_done, see kvmppc_alloc_reset_hpt */
1781 smp_mb();
1782
1783 /* On the first time here, set up HTAB and VRMA or RMA */
c77162de 1784 if (!vcpu->kvm->arch.rma_setup_done) {
32fad281 1785 r = kvmppc_hv_setup_htab_rma(vcpu);
c77162de 1786 if (r)
32fad281 1787 goto out;
c77162de 1788 }
19ccb76a
PM		 1789
1790 flush_fp_to_thread(current);
1791 flush_altivec_to_thread(current);
1792 flush_vsx_to_thread(current);
1793 vcpu->arch.wqp = &vcpu->arch.vcore->wq;
342d3db7 1794 vcpu->arch.pgdir = current->mm->pgd;
c7b67670 1795 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
19ccb76a 1796
a8606e20
PM		 1797 do {
1798 r = kvmppc_run_vcpu(run, vcpu);
1799
1800 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
1801 !(vcpu->arch.shregs.msr & MSR_PR)) {
1802 r = kvmppc_pseries_do_hcall(vcpu);
7e28e60e 1803 kvmppc_core_prepare_to_enter(vcpu);
913d3ff9
PM		 1804 } else if (r == RESUME_PAGE_FAULT) {
1805 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1806 r = kvmppc_book3s_hv_page_fault(run, vcpu,
1807 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1808 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
a8606e20 1809 }
e59d24e6 1810 } while (is_kvmppc_resume_guest(r));
32fad281
PM		 1811
1812 out:
c7b67670 1813 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
32fad281 1814 atomic_dec(&vcpu->kvm->arch.vcpus_running);
a8606e20
PM		 1815 return r;
1816}
1817
54738c09 1818
aa04b4cc 1819/* Work out RMLS (real mode limit selector) field value for a given RMA size.
9e368f29 1820 Assumes POWER7 or PPC970. */
aa04b4cc
PM		 1821static inline int lpcr_rmls(unsigned long rma_size)
1822{
1823 switch (rma_size) {
1824 case 32ul << 20: /* 32 MB */
9e368f29
PM		 1825 if (cpu_has_feature(CPU_FTR_ARCH_206))
1826 return 8; /* only supported on POWER7 */
1827 return -1;
aa04b4cc
PM		 1828 case 64ul << 20: /* 64 MB */
1829 return 3;
1830 case 128ul << 20: /* 128 MB */
1831 return 7;
1832 case 256ul << 20: /* 256 MB */
1833 return 4;
1834 case 1ul << 30: /* 1 GB */
1835 return 2;
1836 case 16ul << 30: /* 16 GB */
1837 return 1;
1838 case 256ul << 30: /* 256 GB */
1839 return 0;
1840 default:
1841 return -1;
1842 }
1843}
1844
1845static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1846{
aa04b4cc 1847 struct page *page;
6c45b810 1848 struct kvm_rma_info *ri = vma->vm_file->private_data;
aa04b4cc 1849
6c45b810 1850 if (vmf->pgoff >= kvm_rma_pages)
aa04b4cc
PM		 1851 return VM_FAULT_SIGBUS;
1852
1853 page = pfn_to_page(ri->base_pfn + vmf->pgoff);
1854 get_page(page);
1855 vmf->page = page;
1856 return 0;
1857}
1858
1859static const struct vm_operations_struct kvm_rma_vm_ops = {
1860 .fault = kvm_rma_fault,
1861};
1862
1863static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma)
1864{
314e51b9 1865 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
aa04b4cc
PM		 1866 vma->vm_ops = &kvm_rma_vm_ops;
1867 return 0;
1868}
1869
1870static int kvm_rma_release(struct inode *inode, struct file *filp)
1871{
6c45b810 1872 struct kvm_rma_info *ri = filp->private_data;
aa04b4cc
PM		 1873
1874 kvm_release_rma(ri);
1875 return 0;
1876}
1877
75ef9de1 1878static const struct file_operations kvm_rma_fops = {
aa04b4cc
PM		 1879 .mmap = kvm_rma_mmap,
1880 .release = kvm_rma_release,
1881};
1882
3a167bea
AK		 1883static long kvm_vm_ioctl_allocate_rma(struct kvm *kvm,
1884 struct kvm_allocate_rma *ret)
aa04b4cc 1885{
aa04b4cc 1886 long fd;
6c45b810
AK		 1887 struct kvm_rma_info *ri;
1888 /*
1889 * Only do this on PPC970 in HV mode
1890 */
1891 if (!cpu_has_feature(CPU_FTR_HVMODE) ||
1892 !cpu_has_feature(CPU_FTR_ARCH_201))
1893 return -EINVAL;
1894
1895 if (!kvm_rma_pages)
1896 return -EINVAL;
aa04b4cc
PM		 1897
1898 ri = kvm_alloc_rma();
1899 if (!ri)
1900 return -ENOMEM;
1901
2f84d5ea 1902 fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR | O_CLOEXEC);
aa04b4cc
PM		 1903 if (fd < 0)
1904 kvm_release_rma(ri);
1905
6c45b810 1906 ret->rma_size = kvm_rma_pages << PAGE_SHIFT;
aa04b4cc
PM		 1907 return fd;
1908}
1909
5b74716e
BH		 1910static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
1911 int linux_psize)
1912{
1913 struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
1914
1915 if (!def->shift)
1916 return;
1917 (*sps)->page_shift = def->shift;
1918 (*sps)->slb_enc = def->sllp;
1919 (*sps)->enc[0].page_shift = def->shift;
b1022fbd
AK		 1920 /*
1921 * Only return base page encoding. We don't want to return
1922 * all the supporting pte_enc, because our H_ENTER doesn't
1923 * support MPSS yet. Once they do, we can start passing all
1924 * support pte_enc here
1925 */
1926 (*sps)->enc[0].pte_enc = def->penc[linux_psize];
1f365bb0
AK		 1927 /*
1928 * Add 16MB MPSS support if host supports it
1929 */
1930 if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
1931 (*sps)->enc[1].page_shift = 24;
1932 (*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
1933 }
5b74716e
BH		 1934 (*sps)++;
1935}
1936
3a167bea
AK		 1937static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
1938 struct kvm_ppc_smmu_info *info)
5b74716e
BH		 1939{
1940 struct kvm_ppc_one_seg_page_size *sps;
1941
1942 info->flags = KVM_PPC_PAGE_SIZES_REAL;
1943 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1944 info->flags |= KVM_PPC_1T_SEGMENTS;
1945 info->slb_size = mmu_slb_size;
1946
1947 /* We only support these sizes for now, and no muti-size segments */
1948 sps = &info->sps[0];
1949 kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
1950 kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
1951 kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
1952
1953 return 0;
1954}
1955
82ed3616
PM		 1956/*
1957 * Get (and clear) the dirty memory log for a memory slot.
1958 */
3a167bea
AK		 1959static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
1960 struct kvm_dirty_log *log)
82ed3616
PM		 1961{
1962 struct kvm_memory_slot *memslot;
1963 int r;
1964 unsigned long n;
1965
1966 mutex_lock(&kvm->slots_lock);
1967
1968 r = -EINVAL;
bbacc0c1 1969 if (log->slot >= KVM_USER_MEM_SLOTS)
82ed3616
PM		 1970 goto out;
1971
1972 memslot = id_to_memslot(kvm->memslots, log->slot);
1973 r = -ENOENT;
1974 if (!memslot->dirty_bitmap)
1975 goto out;
1976
1977 n = kvm_dirty_bitmap_bytes(memslot);
1978 memset(memslot->dirty_bitmap, 0, n);
1979
dfe49dbd 1980 r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
82ed3616
PM		 1981 if (r)
1982 goto out;
1983
1984 r = -EFAULT;
1985 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1986 goto out;
1987
1988 r = 0;
1989out:
1990 mutex_unlock(&kvm->slots_lock);
1991 return r;
1992}
1993
a66b48c3 1994static void unpin_slot(struct kvm_memory_slot *memslot)
de56a948 1995{
a66b48c3
PM		 1996 unsigned long *physp;
1997 unsigned long j, npages, pfn;
1998 struct page *page;
aa04b4cc 1999
a66b48c3
PM		 2000 physp = memslot->arch.slot_phys;
2001 npages = memslot->npages;
2002 if (!physp)
2003 return;
2004 for (j = 0; j < npages; j++) {
2005 if (!(physp[j] & KVMPPC_GOT_PAGE))
2006 continue;
2007 pfn = physp[j] >> PAGE_SHIFT;
2008 page = pfn_to_page(pfn);
2009 SetPageDirty(page);
2010 put_page(page);
2011 }
2012}
2013
3a167bea
AK		 2014static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
2015 struct kvm_memory_slot *dont)
a66b48c3
PM		 2016{
2017 if (!dont || free->arch.rmap != dont->arch.rmap) {
2018 vfree(free->arch.rmap);
2019 free->arch.rmap = NULL;
b2b2f165 2020 }
a66b48c3
PM		 2021 if (!dont || free->arch.slot_phys != dont->arch.slot_phys) {
2022 unpin_slot(free);
2023 vfree(free->arch.slot_phys);
2024 free->arch.slot_phys = NULL;
2025 }
2026}
2027
3a167bea
AK		 2028static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
2029 unsigned long npages)
a66b48c3
PM		 2030{
2031 slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
2032 if (!slot->arch.rmap)
2033 return -ENOMEM;
2034 slot->arch.slot_phys = NULL;
aa04b4cc 2035
c77162de
PM		 2036 return 0;
2037}
aa04b4cc 2038
3a167bea
AK		 2039static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
2040 struct kvm_memory_slot *memslot,
2041 struct kvm_userspace_memory_region *mem)
c77162de 2042{
a66b48c3 2043 unsigned long *phys;
c77162de 2044
a66b48c3
PM		 2045 /* Allocate a slot_phys array if needed */
2046 phys = memslot->arch.slot_phys;
2047 if (!kvm->arch.using_mmu_notifiers && !phys && memslot->npages) {
2048 phys = vzalloc(memslot->npages * sizeof(unsigned long));
2049 if (!phys)
2050 return -ENOMEM;
2051 memslot->arch.slot_phys = phys;
aa04b4cc 2052 }
a66b48c3
PM		 2053
2054 return 0;
c77162de
PM		 2055}
2056
3a167bea
AK		 2057static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
2058 struct kvm_userspace_memory_region *mem,
2059 const struct kvm_memory_slot *old)
c77162de 2060{
dfe49dbd
PM		 2061 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
2062 struct kvm_memory_slot *memslot;
2063
8482644a 2064 if (npages && old->npages) {
dfe49dbd
PM		 2065 /*
2066 * If modifying a memslot, reset all the rmap dirty bits.
2067 * If this is a new memslot, we don't need to do anything
2068 * since the rmap array starts out as all zeroes,
2069 * i.e. no pages are dirty.
2070 */
2071 memslot = id_to_memslot(kvm->memslots, mem->slot);
2072 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
2073 }
c77162de
PM		 2074}
2075
a0144e2a
PM		 2076/*
2077 * Update LPCR values in kvm->arch and in vcores.
2078 * Caller must hold kvm->lock.
2079 */
2080void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
2081{
2082 long int i;
2083 u32 cores_done = 0;
2084
2085 if ((kvm->arch.lpcr & mask) == lpcr)
2086 return;
2087
2088 kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
2089
2090 for (i = 0; i < KVM_MAX_VCORES; ++i) {
2091 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
2092 if (!vc)
2093 continue;
2094 spin_lock(&vc->lock);
2095 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
2096 spin_unlock(&vc->lock);
2097 if (++cores_done >= kvm->arch.online_vcores)
2098 break;
2099 }
2100}
2101
3a167bea
AK		 2102static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
2103{
2104 return;
2105}
2106
32fad281 2107static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
c77162de
PM		 2108{
2109 int err = 0;
2110 struct kvm *kvm = vcpu->kvm;
6c45b810 2111 struct kvm_rma_info *ri = NULL;
c77162de
PM		 2112 unsigned long hva;
2113 struct kvm_memory_slot *memslot;
2114 struct vm_area_struct *vma;
a0144e2a
PM		 2115 unsigned long lpcr = 0, senc;
2116 unsigned long lpcr_mask = 0;
c77162de
PM		 2117 unsigned long psize, porder;
2118 unsigned long rma_size;
2119 unsigned long rmls;
2120 unsigned long *physp;
da9d1d7f 2121 unsigned long i, npages;
2c9097e4 2122 int srcu_idx;
c77162de
PM		 2123
2124 mutex_lock(&kvm->lock);
2125 if (kvm->arch.rma_setup_done)
2126 goto out; /* another vcpu beat us to it */
aa04b4cc 2127
32fad281
PM		 2128 /* Allocate hashed page table (if not done already) and reset it */
2129 if (!kvm->arch.hpt_virt) {
2130 err = kvmppc_alloc_hpt(kvm, NULL);
2131 if (err) {
2132 pr_err("KVM: Couldn't alloc HPT\n");
2133 goto out;
2134 }
2135 }
2136
c77162de 2137 /* Look up the memslot for guest physical address 0 */
2c9097e4 2138 srcu_idx = srcu_read_lock(&kvm->srcu);
c77162de 2139 memslot = gfn_to_memslot(kvm, 0);
aa04b4cc 2140
c77162de
PM		 2141 /* We must have some memory at 0 by now */
2142 err = -EINVAL;
2143 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
2c9097e4 2144 goto out_srcu;
c77162de
PM		 2145
2146 /* Look up the VMA for the start of this memory slot */
2147 hva = memslot->userspace_addr;
2148 down_read(&current->mm->mmap_sem);
2149 vma = find_vma(current->mm, hva);
2150 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
2151 goto up_out;
2152
2153 psize = vma_kernel_pagesize(vma);
da9d1d7f 2154 porder = __ilog2(psize);
c77162de
PM		 2155
2156 /* Is this one of our preallocated RMAs? */
2157 if (vma->vm_file && vma->vm_file->f_op == &kvm_rma_fops &&
2158 hva == vma->vm_start)
2159 ri = vma->vm_file->private_data;
2160
2161 up_read(&current->mm->mmap_sem);
2162
2163 if (!ri) {
2164 /* On POWER7, use VRMA; on PPC970, give up */
2165 err = -EPERM;
2166 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2167 pr_err("KVM: CPU requires an RMO\n");
2c9097e4 2168 goto out_srcu;
c77162de
PM		 2169 }
2170
da9d1d7f
PM		 2171 /* We can handle 4k, 64k or 16M pages in the VRMA */
2172 err = -EINVAL;
2173 if (!(psize == 0x1000 || psize == 0x10000 ||
2174 psize == 0x1000000))
2c9097e4 2175 goto out_srcu;
da9d1d7f 2176
c77162de 2177 /* Update VRMASD field in the LPCR */
da9d1d7f 2178 senc = slb_pgsize_encoding(psize);
697d3899
PM		 2179 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
2180 (VRMA_VSID << SLB_VSID_SHIFT_1T);
a0144e2a
PM		 2181 lpcr_mask = LPCR_VRMASD;
2182 /* the -4 is to account for senc values starting at 0x10 */
2183 lpcr = senc << (LPCR_VRMASD_SH - 4);
c77162de
PM		 2184
2185 /* Create HPTEs in the hash page table for the VRMA */
da9d1d7f 2186 kvmppc_map_vrma(vcpu, memslot, porder);
c77162de
PM		 2187
2188 } else {
2189 /* Set up to use an RMO region */
6c45b810 2190 rma_size = kvm_rma_pages;
c77162de
PM		 2191 if (rma_size > memslot->npages)
2192 rma_size = memslot->npages;
2193 rma_size <<= PAGE_SHIFT;
aa04b4cc 2194 rmls = lpcr_rmls(rma_size);
c77162de 2195 err = -EINVAL;
5d226ae5 2196 if ((long)rmls < 0) {
c77162de 2197 pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size);
2c9097e4 2198 goto out_srcu;
aa04b4cc
PM		 2199 }
2200 atomic_inc(&ri->use_count);
2201 kvm->arch.rma = ri;
9e368f29
PM		 2202
2203 /* Update LPCR and RMOR */
9e368f29
PM		 2204 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2205 /* PPC970; insert RMLS value (split field) in HID4 */
a0144e2a
PM		 2206 lpcr_mask = (1ul << HID4_RMLS0_SH) |
2207 (3ul << HID4_RMLS2_SH) | HID4_RMOR;
2208 lpcr = ((rmls >> 2) << HID4_RMLS0_SH) |
9e368f29
PM		 2209 ((rmls & 3) << HID4_RMLS2_SH);
2210 /* RMOR is also in HID4 */
2211 lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff)
2212 << HID4_RMOR_SH;
2213 } else {
2214 /* POWER7 */
a0144e2a
PM		 2215 lpcr_mask = LPCR_VPM0 | LPCR_VRMA_L | LPCR_RMLS;
2216 lpcr = rmls << LPCR_RMLS_SH;
6c45b810 2217 kvm->arch.rmor = ri->base_pfn << PAGE_SHIFT;
9e368f29 2218 }
c77162de 2219 pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n",
aa04b4cc 2220 ri->base_pfn << PAGE_SHIFT, rma_size, lpcr);
aa04b4cc 2221
c77162de 2222 /* Initialize phys addrs of pages in RMO */
6c45b810 2223 npages = kvm_rma_pages;
da9d1d7f 2224 porder = __ilog2(npages);
a66b48c3
PM		 2225 physp = memslot->arch.slot_phys;
2226 if (physp) {
2227 if (npages > memslot->npages)
2228 npages = memslot->npages;
2229 spin_lock(&kvm->arch.slot_phys_lock);
2230 for (i = 0; i < npages; ++i)
2231 physp[i] = ((ri->base_pfn + i) << PAGE_SHIFT) +
2232 porder;
2233 spin_unlock(&kvm->arch.slot_phys_lock);
2234 }
aa04b4cc
PM		 2235 }
2236
a0144e2a
PM		 2237 kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
2238
c77162de
PM		 2239 /* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */
2240 smp_wmb();
2241 kvm->arch.rma_setup_done = 1;
2242 err = 0;
2c9097e4
PM		 2243 out_srcu:
2244 srcu_read_unlock(&kvm->srcu, srcu_idx);
c77162de
PM		 2245 out:
2246 mutex_unlock(&kvm->lock);
2247 return err;
b2b2f165 2248
c77162de
PM		 2249 up_out:
2250 up_read(&current->mm->mmap_sem);
505d6421 2251 goto out_srcu;
de56a948
PM		 2252}
2253
3a167bea 2254static int kvmppc_core_init_vm_hv(struct kvm *kvm)
de56a948 2255{
32fad281 2256 unsigned long lpcr, lpid;
de56a948 2257
32fad281
PM		 2258 /* Allocate the guest's logical partition ID */
2259
2260 lpid = kvmppc_alloc_lpid();
5d226ae5 2261 if ((long)lpid < 0)
32fad281
PM		 2262 return -ENOMEM;
2263 kvm->arch.lpid = lpid;
de56a948 2264
1b400ba0
PM		 2265 /*
2266 * Since we don't flush the TLB when tearing down a VM,
2267 * and this lpid might have previously been used,
2268 * make sure we flush on each core before running the new VM.
2269 */
2270 cpumask_setall(&kvm->arch.need_tlb_flush);
2271
aa04b4cc 2272 kvm->arch.rma = NULL;
aa04b4cc 2273
9e368f29 2274 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
aa04b4cc 2275
9e368f29
PM		 2276 if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2277 /* PPC970; HID4 is effectively the LPCR */
9e368f29
PM		 2278 kvm->arch.host_lpid = 0;
2279 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4);
2280 lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH));
2281 lpcr |= ((lpid >> 4) << HID4_LPID1_SH) |
2282 ((lpid & 0xf) << HID4_LPID5_SH);
2283 } else {
2284 /* POWER7; init LPCR for virtual RMA mode */
2285 kvm->arch.host_lpid = mfspr(SPRN_LPID);
2286 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
2287 lpcr &= LPCR_PECE | LPCR_LPES;
2288 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
697d3899
PM		 2289 LPCR_VPM0 | LPCR_VPM1;
2290 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
2291 (VRMA_VSID << SLB_VSID_SHIFT_1T);
e0622bd9
PM		 2292 /* On POWER8 turn on online bit to enable PURR/SPURR */
2293 if (cpu_has_feature(CPU_FTR_ARCH_207S))
2294 lpcr |= LPCR_ONL;
9e368f29
PM		 2295 }
2296 kvm->arch.lpcr = lpcr;
aa04b4cc 2297
342d3db7 2298 kvm->arch.using_mmu_notifiers = !!cpu_has_feature(CPU_FTR_ARCH_206);
c77162de 2299 spin_lock_init(&kvm->arch.slot_phys_lock);
512691d4
PM		 2300
2301 /*
441c19c8
ME		 2302 * Track that we now have a HV mode VM active. This blocks secondary
2303 * CPU threads from coming online.
512691d4 2304 */
441c19c8 2305 kvm_hv_vm_activated();
512691d4 2306
54738c09 2307 return 0;
de56a948
PM		 2308}
2309
f1378b1c
PM		 2310static void kvmppc_free_vcores(struct kvm *kvm)
2311{
2312 long int i;
2313
2314 for (i = 0; i < KVM_MAX_VCORES; ++i)
2315 kfree(kvm->arch.vcores[i]);
2316 kvm->arch.online_vcores = 0;
2317}
2318
3a167bea 2319static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
de56a948 2320{
441c19c8 2321 kvm_hv_vm_deactivated();
512691d4 2322
f1378b1c 2323 kvmppc_free_vcores(kvm);
aa04b4cc
PM		 2324 if (kvm->arch.rma) {
2325 kvm_release_rma(kvm->arch.rma);
2326 kvm->arch.rma = NULL;
2327 }
2328
de56a948
PM		 2329 kvmppc_free_hpt(kvm);
2330}
2331
3a167bea
AK		 2332/* We don't need to emulate any privileged instructions or dcbz */
2333static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
2334 unsigned int inst, int *advance)
de56a948 2335{
3a167bea 2336 return EMULATE_FAIL;
de56a948
PM		 2337}
2338
3a167bea
AK		 2339static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
2340 ulong spr_val)
de56a948
PM		 2341{
2342 return EMULATE_FAIL;
2343}
2344
3a167bea
AK		 2345static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
2346 ulong *spr_val)
de56a948
PM		 2347{
2348 return EMULATE_FAIL;
2349}
2350
3a167bea 2351static int kvmppc_core_check_processor_compat_hv(void)
de56a948 2352{
3a167bea
AK		 2353 if (!cpu_has_feature(CPU_FTR_HVMODE))
2354 return -EIO;
2355 return 0;
de56a948
PM		 2356}
2357
3a167bea
AK		 2358static long kvm_arch_vm_ioctl_hv(struct file *filp,
2359 unsigned int ioctl, unsigned long arg)
2360{
2361 struct kvm *kvm __maybe_unused = filp->private_data;
2362 void __user *argp = (void __user *)arg;
2363 long r;
2364
2365 switch (ioctl) {
2366
2367 case KVM_ALLOCATE_RMA: {
2368 struct kvm_allocate_rma rma;
2369 struct kvm *kvm = filp->private_data;
2370
2371 r = kvm_vm_ioctl_allocate_rma(kvm, &rma);
2372 if (r >= 0 && copy_to_user(argp, &rma, sizeof(rma)))
2373 r = -EFAULT;
2374 break;
2375 }
2376
2377 case KVM_PPC_ALLOCATE_HTAB: {
2378 u32 htab_order;
2379
2380 r = -EFAULT;
2381 if (get_user(htab_order, (u32 __user *)argp))
2382 break;
2383 r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
2384 if (r)
2385 break;
2386 r = -EFAULT;
2387 if (put_user(htab_order, (u32 __user *)argp))
2388 break;
2389 r = 0;
2390 break;
2391 }
2392
2393 case KVM_PPC_GET_HTAB_FD: {
2394 struct kvm_get_htab_fd ghf;
2395
2396 r = -EFAULT;
2397 if (copy_from_user(&ghf, argp, sizeof(ghf)))
2398 break;
2399 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
2400 break;
2401 }
2402
2403 default:
2404 r = -ENOTTY;
2405 }
2406
2407 return r;
2408}
2409
cbbc58d4 2410static struct kvmppc_ops kvm_ops_hv = {
3a167bea
AK		 2411 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
2412 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
2413 .get_one_reg = kvmppc_get_one_reg_hv,
2414 .set_one_reg = kvmppc_set_one_reg_hv,
2415 .vcpu_load = kvmppc_core_vcpu_load_hv,
2416 .vcpu_put = kvmppc_core_vcpu_put_hv,
2417 .set_msr = kvmppc_set_msr_hv,
2418 .vcpu_run = kvmppc_vcpu_run_hv,
2419 .vcpu_create = kvmppc_core_vcpu_create_hv,
2420 .vcpu_free = kvmppc_core_vcpu_free_hv,
2421 .check_requests = kvmppc_core_check_requests_hv,
2422 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv,
2423 .flush_memslot = kvmppc_core_flush_memslot_hv,
2424 .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
2425 .commit_memory_region = kvmppc_core_commit_memory_region_hv,
2426 .unmap_hva = kvm_unmap_hva_hv,
2427 .unmap_hva_range = kvm_unmap_hva_range_hv,
2428 .age_hva = kvm_age_hva_hv,
2429 .test_age_hva = kvm_test_age_hva_hv,
2430 .set_spte_hva = kvm_set_spte_hva_hv,
2431 .mmu_destroy = kvmppc_mmu_destroy_hv,
2432 .free_memslot = kvmppc_core_free_memslot_hv,
2433 .create_memslot = kvmppc_core_create_memslot_hv,
2434 .init_vm = kvmppc_core_init_vm_hv,
2435 .destroy_vm = kvmppc_core_destroy_vm_hv,
3a167bea
AK		 2436 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
2437 .emulate_op = kvmppc_core_emulate_op_hv,
2438 .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
2439 .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
2440 .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
2441 .arch_vm_ioctl = kvm_arch_vm_ioctl_hv,
2442};
2443
2444static int kvmppc_book3s_init_hv(void)
de56a948
PM		 2445{
2446 int r;
cbbc58d4
AK		 2447 /*
2448 * FIXME!! Do we need to check on all cpus ?
2449 */
2450 r = kvmppc_core_check_processor_compat_hv();
2451 if (r < 0)
739e2425 2452 return -ENODEV;
de56a948 2453
cbbc58d4
AK		 2454 kvm_ops_hv.owner = THIS_MODULE;
2455 kvmppc_hv_ops = &kvm_ops_hv;
de56a948 2456
cbbc58d4 2457 r = kvmppc_mmu_hv_init();
de56a948
PM		 2458 return r;
2459}
2460
3a167bea 2461static void kvmppc_book3s_exit_hv(void)
de56a948 2462{
cbbc58d4 2463 kvmppc_hv_ops = NULL;
de56a948
PM		 2464}
2465
3a167bea
AK		 2466module_init(kvmppc_book3s_init_hv);
2467module_exit(kvmppc_book3s_exit_hv);
2ba9f0d8 2468MODULE_LICENSE("GPL");
398a76c6
AG		 2469MODULE_ALIAS_MISCDEV(KVM_MINOR);
2470MODULE_ALIAS("devname:kvm");
ubuntu-bionic-kernel mirror