]> Git Repo - linux.git/blame - arch/powerpc/kvm/book3s_hv.c
projects / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: PPC: Book3S HV: Send IPI to host core to wake VCPU
[linux.git] / arch / powerpc / kvm / book3s_hv.c
CommitLineData
de56a948
PM		 1/*
2 * Copyright 2011 Paul Mackerras, IBM Corp. <[email protected]>
3 * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
4 *
5 * Authors:
6 * Paul Mackerras <[email protected]>
7 * Alexander Graf <[email protected]>
8 * Kevin Wolf <[email protected]>
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 <[email protected]>.
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>
e23a808b 35#include <linux/debugfs.h>
de56a948
PM		 36
37#include <asm/reg.h>
38#include <asm/cputable.h>
39#include <asm/cacheflush.h>
40#include <asm/tlbflush.h>
41#include <asm/uaccess.h>
42#include <asm/io.h>
43#include <asm/kvm_ppc.h>
44#include <asm/kvm_book3s.h>
45#include <asm/mmu_context.h>
46#include <asm/lppaca.h>
47#include <asm/processor.h>
371fefd6 48#include <asm/cputhreads.h>
aa04b4cc 49#include <asm/page.h>
de1d9248 50#include <asm/hvcall.h>
ae3a197e 51#include <asm/switch_to.h>
512691d4 52#include <asm/smp.h>
66feed61 53#include <asm/dbell.h>
de56a948 54#include <linux/gfp.h>
de56a948
PM		 55#include <linux/vmalloc.h>
56#include <linux/highmem.h>
c77162de 57#include <linux/hugetlb.h>
2ba9f0d8 58#include <linux/module.h>
de56a948 59
3a167bea
AK		 60#include "book3s.h"
61
3c78f78a
SW		 62#define CREATE_TRACE_POINTS
63#include "trace_hv.h"
64
de56a948
PM		 65/* #define EXIT_DEBUG */
66/* #define EXIT_DEBUG_SIMPLE */
67/* #define EXIT_DEBUG_INT */
68
913d3ff9
PM		 69/* Used to indicate that a guest page fault needs to be handled */
70#define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1)
71
c7b67670
PM		 72/* Used as a "null" value for timebase values */
73#define TB_NIL (~(u64)0)
74
699a0ea0
PM		 75static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
76
b4deba5c
PM		 77static int dynamic_mt_modes = 6;
78module_param(dynamic_mt_modes, int, S_IRUGO | S_IWUSR);
79MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
ec257165
PM		 80static int target_smt_mode;
81module_param(target_smt_mode, int, S_IRUGO | S_IWUSR);
82MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
9678cdaa 83
19ccb76a 84static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
32fad281 85static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
19ccb76a 86
66feed61
PM		 87static bool kvmppc_ipi_thread(int cpu)
88{
89 /* On POWER8 for IPIs to threads in the same core, use msgsnd */
90 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
91 preempt_disable();
92 if (cpu_first_thread_sibling(cpu) ==
93 cpu_first_thread_sibling(smp_processor_id())) {
94 unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
95 msg |= cpu_thread_in_core(cpu);
96 smp_mb();
97 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
98 preempt_enable();
99 return true;
100 }
101 preempt_enable();
102 }
103
104#if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
105 if (cpu >= 0 && cpu < nr_cpu_ids && paca[cpu].kvm_hstate.xics_phys) {
106 xics_wake_cpu(cpu);
107 return true;
108 }
109#endif
110
111 return false;
112}
113
3a167bea 114static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
54695c30 115{
ec257165 116 int cpu;
54695c30
BH		 117 wait_queue_head_t *wqp;
118
119 wqp = kvm_arch_vcpu_wq(vcpu);
120 if (waitqueue_active(wqp)) {
121 wake_up_interruptible(wqp);
122 ++vcpu->stat.halt_wakeup;
123 }
124
ec257165 125 if (kvmppc_ipi_thread(vcpu->arch.thread_cpu))
66feed61 126 return;
54695c30
BH		 127
128 /* CPU points to the first thread of the core */
ec257165 129 cpu = vcpu->cpu;
66feed61
PM		 130 if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
131 smp_send_reschedule(cpu);
54695c30
BH		 132}
133
c7b67670
PM		 134/*
135 * We use the vcpu_load/put functions to measure stolen time.
136 * Stolen time is counted as time when either the vcpu is able to
137 * run as part of a virtual core, but the task running the vcore
138 * is preempted or sleeping, or when the vcpu needs something done
139 * in the kernel by the task running the vcpu, but that task is
140 * preempted or sleeping. Those two things have to be counted
141 * separately, since one of the vcpu tasks will take on the job
142 * of running the core, and the other vcpu tasks in the vcore will
143 * sleep waiting for it to do that, but that sleep shouldn't count
144 * as stolen time.
145 *
146 * Hence we accumulate stolen time when the vcpu can run as part of
147 * a vcore using vc->stolen_tb, and the stolen time when the vcpu
148 * needs its task to do other things in the kernel (for example,
149 * service a page fault) in busy_stolen. We don't accumulate
150 * stolen time for a vcore when it is inactive, or for a vcpu
151 * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of
152 * a misnomer; it means that the vcpu task is not executing in
153 * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
154 * the kernel. We don't have any way of dividing up that time
155 * between time that the vcpu is genuinely stopped, time that
156 * the task is actively working on behalf of the vcpu, and time
157 * that the task is preempted, so we don't count any of it as
158 * stolen.
159 *
160 * Updates to busy_stolen are protected by arch.tbacct_lock;
2711e248
PM		 161 * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
162 * lock. The stolen times are measured in units of timebase ticks.
163 * (Note that the != TB_NIL checks below are purely defensive;
164 * they should never fail.)
c7b67670
PM		 165 */
166
ec257165
PM		 167static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
168{
169 unsigned long flags;
170
171 spin_lock_irqsave(&vc->stoltb_lock, flags);
172 vc->preempt_tb = mftb();
173 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
174}
175
176static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
177{
178 unsigned long flags;
179
180 spin_lock_irqsave(&vc->stoltb_lock, flags);
181 if (vc->preempt_tb != TB_NIL) {
182 vc->stolen_tb += mftb() - vc->preempt_tb;
183 vc->preempt_tb = TB_NIL;
184 }
185 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
186}
187
3a167bea 188static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
de56a948 189{
0456ec4f 190 struct kvmppc_vcore *vc = vcpu->arch.vcore;
bf3d32e1 191 unsigned long flags;
0456ec4f 192
2711e248
PM		 193 /*
194 * We can test vc->runner without taking the vcore lock,
195 * because only this task ever sets vc->runner to this
196 * vcpu, and once it is set to this vcpu, only this task
197 * ever sets it to NULL.
198 */
ec257165
PM		 199 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
200 kvmppc_core_end_stolen(vc);
201
2711e248 202 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
c7b67670
PM		 203 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
204 vcpu->arch.busy_preempt != TB_NIL) {
205 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
206 vcpu->arch.busy_preempt = TB_NIL;
207 }
bf3d32e1 208 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
de56a948
PM		 209}
210
3a167bea 211static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
de56a948 212{
0456ec4f 213 struct kvmppc_vcore *vc = vcpu->arch.vcore;
bf3d32e1 214 unsigned long flags;
0456ec4f 215
ec257165
PM		 216 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
217 kvmppc_core_start_stolen(vc);
218
2711e248 219 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
c7b67670
PM		 220 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
221 vcpu->arch.busy_preempt = mftb();
bf3d32e1 222 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
de56a948
PM		 223}
224
3a167bea 225static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
de56a948 226{
c20875a3
PM		 227 /*
228 * Check for illegal transactional state bit combination
229 * and if we find it, force the TS field to a safe state.
230 */
231 if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
232 msr &= ~MSR_TS_MASK;
de56a948 233 vcpu->arch.shregs.msr = msr;
19ccb76a 234 kvmppc_end_cede(vcpu);
de56a948
PM		 235}
236
5358a963 237static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
de56a948
PM		 238{
239 vcpu->arch.pvr = pvr;
240}
241
5358a963 242static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
388cc6e1
PM		 243{
244 unsigned long pcr = 0;
245 struct kvmppc_vcore *vc = vcpu->arch.vcore;
246
247 if (arch_compat) {
388cc6e1
PM		 248 switch (arch_compat) {
249 case PVR_ARCH_205:
5557ae0e
PM		 250 /*
251 * If an arch bit is set in PCR, all the defined
252 * higher-order arch bits also have to be set.
253 */
254 pcr = PCR_ARCH_206 | PCR_ARCH_205;
388cc6e1
PM		 255 break;
256 case PVR_ARCH_206:
257 case PVR_ARCH_206p:
5557ae0e
PM		 258 pcr = PCR_ARCH_206;
259 break;
260 case PVR_ARCH_207:
388cc6e1
PM		 261 break;
262 default:
263 return -EINVAL;
264 }
5557ae0e
PM		 265
266 if (!cpu_has_feature(CPU_FTR_ARCH_207S)) {
267 /* POWER7 can't emulate POWER8 */
268 if (!(pcr & PCR_ARCH_206))
269 return -EINVAL;
270 pcr &= ~PCR_ARCH_206;
271 }
388cc6e1
PM		 272 }
273
274 spin_lock(&vc->lock);
275 vc->arch_compat = arch_compat;
276 vc->pcr = pcr;
277 spin_unlock(&vc->lock);
278
279 return 0;
280}
281
5358a963 282static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
de56a948
PM		 283{
284 int r;
285
286 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
287 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
288 vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
289 for (r = 0; r < 16; ++r)
290 pr_err("r%2d = %.16lx r%d = %.16lx\n",
291 r, kvmppc_get_gpr(vcpu, r),
292 r+16, kvmppc_get_gpr(vcpu, r+16));
293 pr_err("ctr = %.16lx lr = %.16lx\n",
294 vcpu->arch.ctr, vcpu->arch.lr);
295 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
296 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
297 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
298 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
299 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
300 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
301 pr_err("cr = %.8x xer = %.16lx dsisr = %.8x\n",
302 vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
303 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
304 pr_err("fault dar = %.16lx dsisr = %.8x\n",
305 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
306 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
307 for (r = 0; r < vcpu->arch.slb_max; ++r)
308 pr_err(" ESID = %.16llx VSID = %.16llx\n",
309 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
310 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
a0144e2a 311 vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
de56a948
PM		 312 vcpu->arch.last_inst);
313}
314
5358a963 315static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
a8606e20 316{
e09fefde 317 struct kvm_vcpu *ret;
a8606e20
PM		 318
319 mutex_lock(&kvm->lock);
e09fefde 320 ret = kvm_get_vcpu_by_id(kvm, id);
a8606e20
PM		 321 mutex_unlock(&kvm->lock);
322 return ret;
323}
324
325static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
326{
f13c13a0 327 vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
02407552 328 vpa->yield_count = cpu_to_be32(1);
a8606e20
PM		 329}
330
55b665b0
PM		 331static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
332 unsigned long addr, unsigned long len)
333{
334 /* check address is cacheline aligned */
335 if (addr & (L1_CACHE_BYTES - 1))
336 return -EINVAL;
337 spin_lock(&vcpu->arch.vpa_update_lock);
338 if (v->next_gpa != addr || v->len != len) {
339 v->next_gpa = addr;
340 v->len = addr ? len : 0;
341 v->update_pending = 1;
342 }
343 spin_unlock(&vcpu->arch.vpa_update_lock);
344 return 0;
345}
346
2e25aa5f
PM		 347/* Length for a per-processor buffer is passed in at offset 4 in the buffer */
348struct reg_vpa {
349 u32 dummy;
350 union {
02407552
AG		 351 __be16 hword;
352 __be32 word;
2e25aa5f
PM		 353 } length;
354};
355
356static int vpa_is_registered(struct kvmppc_vpa *vpap)
357{
358 if (vpap->update_pending)
359 return vpap->next_gpa != 0;
360 return vpap->pinned_addr != NULL;
361}
362
a8606e20
PM		 363static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
364 unsigned long flags,
365 unsigned long vcpuid, unsigned long vpa)
366{
367 struct kvm *kvm = vcpu->kvm;
93e60249 368 unsigned long len, nb;
a8606e20
PM		 369 void *va;
370 struct kvm_vcpu *tvcpu;
2e25aa5f
PM		 371 int err;
372 int subfunc;
373 struct kvmppc_vpa *vpap;
a8606e20
PM		 374
375 tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
376 if (!tvcpu)
377 return H_PARAMETER;
378
2e25aa5f
PM		 379 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
380 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
381 subfunc == H_VPA_REG_SLB) {
382 /* Registering new area - address must be cache-line aligned */
383 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
a8606e20 384 return H_PARAMETER;
2e25aa5f
PM		 385
386 /* convert logical addr to kernel addr and read length */
93e60249
PM		 387 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
388 if (va == NULL)
b2b2f165 389 return H_PARAMETER;
2e25aa5f 390 if (subfunc == H_VPA_REG_VPA)
02407552 391 len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
a8606e20 392 else
02407552 393 len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
c35635ef 394 kvmppc_unpin_guest_page(kvm, va, vpa, false);
2e25aa5f
PM		 395
396 /* Check length */
397 if (len > nb || len < sizeof(struct reg_vpa))
398 return H_PARAMETER;
399 } else {
400 vpa = 0;
401 len = 0;
402 }
403
404 err = H_PARAMETER;
405 vpap = NULL;
406 spin_lock(&tvcpu->arch.vpa_update_lock);
407
408 switch (subfunc) {
409 case H_VPA_REG_VPA: /* register VPA */
410 if (len < sizeof(struct lppaca))
a8606e20 411 break;
2e25aa5f
PM		 412 vpap = &tvcpu->arch.vpa;
413 err = 0;
414 break;
415
416 case H_VPA_REG_DTL: /* register DTL */
417 if (len < sizeof(struct dtl_entry))
a8606e20 418 break;
2e25aa5f
PM		 419 len -= len % sizeof(struct dtl_entry);
420
421 /* Check that they have previously registered a VPA */
422 err = H_RESOURCE;
423 if (!vpa_is_registered(&tvcpu->arch.vpa))
a8606e20 424 break;
2e25aa5f
PM		 425
426 vpap = &tvcpu->arch.dtl;
427 err = 0;
428 break;
429
430 case H_VPA_REG_SLB: /* register SLB shadow buffer */
431 /* Check that they have previously registered a VPA */
432 err = H_RESOURCE;
433 if (!vpa_is_registered(&tvcpu->arch.vpa))
a8606e20 434 break;
2e25aa5f
PM		 435
436 vpap = &tvcpu->arch.slb_shadow;
437 err = 0;
438 break;
439
440 case H_VPA_DEREG_VPA: /* deregister VPA */
441 /* Check they don't still have a DTL or SLB buf registered */
442 err = H_RESOURCE;
443 if (vpa_is_registered(&tvcpu->arch.dtl) ||
444 vpa_is_registered(&tvcpu->arch.slb_shadow))
a8606e20 445 break;
2e25aa5f
PM		 446
447 vpap = &tvcpu->arch.vpa;
448 err = 0;
449 break;
450
451 case H_VPA_DEREG_DTL: /* deregister DTL */
452 vpap = &tvcpu->arch.dtl;
453 err = 0;
454 break;
455
456 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */
457 vpap = &tvcpu->arch.slb_shadow;
458 err = 0;
459 break;
460 }
461
462 if (vpap) {
463 vpap->next_gpa = vpa;
464 vpap->len = len;
465 vpap->update_pending = 1;
a8606e20 466 }
93e60249 467
2e25aa5f
PM		 468 spin_unlock(&tvcpu->arch.vpa_update_lock);
469
93e60249 470 return err;
a8606e20
PM		 471}
472
081f323b 473static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
2e25aa5f 474{
081f323b 475 struct kvm *kvm = vcpu->kvm;
2e25aa5f
PM		 476 void *va;
477 unsigned long nb;
081f323b 478 unsigned long gpa;
2e25aa5f 479
081f323b
PM		 480 /*
481 * We need to pin the page pointed to by vpap->next_gpa,
482 * but we can't call kvmppc_pin_guest_page under the lock
483 * as it does get_user_pages() and down_read(). So we
484 * have to drop the lock, pin the page, then get the lock
485 * again and check that a new area didn't get registered
486 * in the meantime.
487 */
488 for (;;) {
489 gpa = vpap->next_gpa;
490 spin_unlock(&vcpu->arch.vpa_update_lock);
491 va = NULL;
492 nb = 0;
493 if (gpa)
c35635ef 494 va = kvmppc_pin_guest_page(kvm, gpa, &nb);
081f323b
PM		 495 spin_lock(&vcpu->arch.vpa_update_lock);
496 if (gpa == vpap->next_gpa)
497 break;
498 /* sigh... unpin that one and try again */
499 if (va)
c35635ef 500 kvmppc_unpin_guest_page(kvm, va, gpa, false);
081f323b
PM		 501 }
502
503 vpap->update_pending = 0;
504 if (va && nb < vpap->len) {
505 /*
506 * If it's now too short, it must be that userspace
507 * has changed the mappings underlying guest memory,
508 * so unregister the region.
509 */
c35635ef 510 kvmppc_unpin_guest_page(kvm, va, gpa, false);
081f323b 511 va = NULL;
2e25aa5f
PM		 512 }
513 if (vpap->pinned_addr)
c35635ef
PM		 514 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
515 vpap->dirty);
516 vpap->gpa = gpa;
2e25aa5f 517 vpap->pinned_addr = va;
c35635ef 518 vpap->dirty = false;
2e25aa5f
PM		 519 if (va)
520 vpap->pinned_end = va + vpap->len;
521}
522
523static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
524{
2f12f034
PM		 525 if (!(vcpu->arch.vpa.update_pending ||
526 vcpu->arch.slb_shadow.update_pending ||
527 vcpu->arch.dtl.update_pending))
528 return;
529
2e25aa5f
PM		 530 spin_lock(&vcpu->arch.vpa_update_lock);
531 if (vcpu->arch.vpa.update_pending) {
081f323b 532 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
55b665b0
PM		 533 if (vcpu->arch.vpa.pinned_addr)
534 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
2e25aa5f
PM		 535 }
536 if (vcpu->arch.dtl.update_pending) {
081f323b 537 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
2e25aa5f
PM		 538 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
539 vcpu->arch.dtl_index = 0;
540 }
541 if (vcpu->arch.slb_shadow.update_pending)
081f323b 542 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
2e25aa5f
PM		 543 spin_unlock(&vcpu->arch.vpa_update_lock);
544}
545
c7b67670
PM		 546/*
547 * Return the accumulated stolen time for the vcore up until `now'.
548 * The caller should hold the vcore lock.
549 */
550static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
551{
552 u64 p;
2711e248 553 unsigned long flags;
c7b67670 554
2711e248
PM		 555 spin_lock_irqsave(&vc->stoltb_lock, flags);
556 p = vc->stolen_tb;
c7b67670 557 if (vc->vcore_state != VCORE_INACTIVE &&
2711e248
PM		 558 vc->preempt_tb != TB_NIL)
559 p += now - vc->preempt_tb;
560 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
c7b67670
PM		 561 return p;
562}
563
0456ec4f
PM		 564static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
565 struct kvmppc_vcore *vc)
566{
567 struct dtl_entry *dt;
568 struct lppaca *vpa;
c7b67670
PM		 569 unsigned long stolen;
570 unsigned long core_stolen;
571 u64 now;
0456ec4f
PM		 572
573 dt = vcpu->arch.dtl_ptr;
574 vpa = vcpu->arch.vpa.pinned_addr;
c7b67670
PM		 575 now = mftb();
576 core_stolen = vcore_stolen_time(vc, now);
577 stolen = core_stolen - vcpu->arch.stolen_logged;
578 vcpu->arch.stolen_logged = core_stolen;
bf3d32e1 579 spin_lock_irq(&vcpu->arch.tbacct_lock);
c7b67670
PM		 580 stolen += vcpu->arch.busy_stolen;
581 vcpu->arch.busy_stolen = 0;
bf3d32e1 582 spin_unlock_irq(&vcpu->arch.tbacct_lock);
0456ec4f
PM		 583 if (!dt || !vpa)
584 return;
585 memset(dt, 0, sizeof(struct dtl_entry));
586 dt->dispatch_reason = 7;
02407552
AG		 587 dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
588 dt->timebase = cpu_to_be64(now + vc->tb_offset);
589 dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
590 dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
591 dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
0456ec4f
PM		 592 ++dt;
593 if (dt == vcpu->arch.dtl.pinned_end)
594 dt = vcpu->arch.dtl.pinned_addr;
595 vcpu->arch.dtl_ptr = dt;
596 /* order writing *dt vs. writing vpa->dtl_idx */
597 smp_wmb();
02407552 598 vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
c35635ef 599 vcpu->arch.dtl.dirty = true;
0456ec4f
PM		 600}
601
9642382e
MN		 602static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
603{
604 if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
605 return true;
606 if ((!vcpu->arch.vcore->arch_compat) &&
607 cpu_has_feature(CPU_FTR_ARCH_207S))
608 return true;
609 return false;
610}
611
612static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
613 unsigned long resource, unsigned long value1,
614 unsigned long value2)
615{
616 switch (resource) {
617 case H_SET_MODE_RESOURCE_SET_CIABR:
618 if (!kvmppc_power8_compatible(vcpu))
619 return H_P2;
620 if (value2)
621 return H_P4;
622 if (mflags)
623 return H_UNSUPPORTED_FLAG_START;
624 /* Guests can't breakpoint the hypervisor */
625 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
626 return H_P3;
627 vcpu->arch.ciabr = value1;
628 return H_SUCCESS;
629 case H_SET_MODE_RESOURCE_SET_DAWR:
630 if (!kvmppc_power8_compatible(vcpu))
631 return H_P2;
632 if (mflags)
633 return H_UNSUPPORTED_FLAG_START;
634 if (value2 & DABRX_HYP)
635 return H_P4;
636 vcpu->arch.dawr = value1;
637 vcpu->arch.dawrx = value2;
638 return H_SUCCESS;
639 default:
640 return H_TOO_HARD;
641 }
642}
643
90fd09f8
SB		 644static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
645{
646 struct kvmppc_vcore *vcore = target->arch.vcore;
647
648 /*
649 * We expect to have been called by the real mode handler
650 * (kvmppc_rm_h_confer()) which would have directly returned
651 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
652 * have useful work to do and should not confer) so we don't
653 * recheck that here.
654 */
655
656 spin_lock(&vcore->lock);
657 if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
ec257165
PM		 658 vcore->vcore_state != VCORE_INACTIVE &&
659 vcore->runner)
90fd09f8
SB		 660 target = vcore->runner;
661 spin_unlock(&vcore->lock);
662
663 return kvm_vcpu_yield_to(target);
664}
665
666static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
667{
668 int yield_count = 0;
669 struct lppaca *lppaca;
670
671 spin_lock(&vcpu->arch.vpa_update_lock);
672 lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
673 if (lppaca)
ecb6d618 674 yield_count = be32_to_cpu(lppaca->yield_count);
90fd09f8
SB		 675 spin_unlock(&vcpu->arch.vpa_update_lock);
676 return yield_count;
677}
678
a8606e20
PM		 679int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
680{
681 unsigned long req = kvmppc_get_gpr(vcpu, 3);
682 unsigned long target, ret = H_SUCCESS;
90fd09f8 683 int yield_count;
a8606e20 684 struct kvm_vcpu *tvcpu;
8e591cb7 685 int idx, rc;
a8606e20 686
699a0ea0
PM		 687 if (req <= MAX_HCALL_OPCODE &&
688 !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
689 return RESUME_HOST;
690
a8606e20
PM		 691 switch (req) {
692 case H_CEDE:
a8606e20
PM		 693 break;
694 case H_PROD:
695 target = kvmppc_get_gpr(vcpu, 4);
696 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
697 if (!tvcpu) {
698 ret = H_PARAMETER;
699 break;
700 }
701 tvcpu->arch.prodded = 1;
702 smp_mb();
703 if (vcpu->arch.ceded) {
704 if (waitqueue_active(&vcpu->wq)) {
705 wake_up_interruptible(&vcpu->wq);
706 vcpu->stat.halt_wakeup++;
707 }
708 }
709 break;
710 case H_CONFER:
42d7604d
PM		 711 target = kvmppc_get_gpr(vcpu, 4);
712 if (target == -1)
713 break;
714 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
715 if (!tvcpu) {
716 ret = H_PARAMETER;
717 break;
718 }
90fd09f8
SB		 719 yield_count = kvmppc_get_gpr(vcpu, 5);
720 if (kvmppc_get_yield_count(tvcpu) != yield_count)
721 break;
722 kvm_arch_vcpu_yield_to(tvcpu);
a8606e20
PM		 723 break;
724 case H_REGISTER_VPA:
725 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
726 kvmppc_get_gpr(vcpu, 5),
727 kvmppc_get_gpr(vcpu, 6));
728 break;
8e591cb7
ME		 729 case H_RTAS:
730 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
731 return RESUME_HOST;
732
c9438092 733 idx = srcu_read_lock(&vcpu->kvm->srcu);
8e591cb7 734 rc = kvmppc_rtas_hcall(vcpu);
c9438092 735 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8e591cb7
ME		 736
737 if (rc == -ENOENT)
738 return RESUME_HOST;
739 else if (rc == 0)
740 break;
741
742 /* Send the error out to userspace via KVM_RUN */
743 return rc;
99342cf8
DG		 744 case H_LOGICAL_CI_LOAD:
745 ret = kvmppc_h_logical_ci_load(vcpu);
746 if (ret == H_TOO_HARD)
747 return RESUME_HOST;
748 break;
749 case H_LOGICAL_CI_STORE:
750 ret = kvmppc_h_logical_ci_store(vcpu);
751 if (ret == H_TOO_HARD)
752 return RESUME_HOST;
753 break;
9642382e
MN		 754 case H_SET_MODE:
755 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
756 kvmppc_get_gpr(vcpu, 5),
757 kvmppc_get_gpr(vcpu, 6),
758 kvmppc_get_gpr(vcpu, 7));
759 if (ret == H_TOO_HARD)
760 return RESUME_HOST;
761 break;
bc5ad3f3
BH		 762 case H_XIRR:
763 case H_CPPR:
764 case H_EOI:
765 case H_IPI:
8e44ddc3
PM		 766 case H_IPOLL:
767 case H_XIRR_X:
bc5ad3f3
BH		 768 if (kvmppc_xics_enabled(vcpu)) {
769 ret = kvmppc_xics_hcall(vcpu, req);
770 break;
d3695aa4
AK		 771 }
772 return RESUME_HOST;
773 case H_PUT_TCE:
774 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
775 kvmppc_get_gpr(vcpu, 5),
776 kvmppc_get_gpr(vcpu, 6));
777 if (ret == H_TOO_HARD)
778 return RESUME_HOST;
779 break;
780 case H_PUT_TCE_INDIRECT:
781 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
782 kvmppc_get_gpr(vcpu, 5),
783 kvmppc_get_gpr(vcpu, 6),
784 kvmppc_get_gpr(vcpu, 7));
785 if (ret == H_TOO_HARD)
786 return RESUME_HOST;
787 break;
788 case H_STUFF_TCE:
789 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
790 kvmppc_get_gpr(vcpu, 5),
791 kvmppc_get_gpr(vcpu, 6),
792 kvmppc_get_gpr(vcpu, 7));
793 if (ret == H_TOO_HARD)
794 return RESUME_HOST;
795 break;
a8606e20
PM		 796 default:
797 return RESUME_HOST;
798 }
799 kvmppc_set_gpr(vcpu, 3, ret);
800 vcpu->arch.hcall_needed = 0;
801 return RESUME_GUEST;
802}
803
ae2113a4
PM		 804static int kvmppc_hcall_impl_hv(unsigned long cmd)
805{
806 switch (cmd) {
807 case H_CEDE:
808 case H_PROD:
809 case H_CONFER:
810 case H_REGISTER_VPA:
9642382e 811 case H_SET_MODE:
99342cf8
DG		 812 case H_LOGICAL_CI_LOAD:
813 case H_LOGICAL_CI_STORE:
ae2113a4
PM		 814#ifdef CONFIG_KVM_XICS
815 case H_XIRR:
816 case H_CPPR:
817 case H_EOI:
818 case H_IPI:
819 case H_IPOLL:
820 case H_XIRR_X:
821#endif
822 return 1;
823 }
824
825 /* See if it's in the real-mode table */
826 return kvmppc_hcall_impl_hv_realmode(cmd);
827}
828
a59c1d9e
MS		 829static int kvmppc_emulate_debug_inst(struct kvm_run *run,
830 struct kvm_vcpu *vcpu)
831{
832 u32 last_inst;
833
834 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
835 EMULATE_DONE) {
836 /*
837 * Fetch failed, so return to guest and
838 * try executing it again.
839 */
840 return RESUME_GUEST;
841 }
842
843 if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
844 run->exit_reason = KVM_EXIT_DEBUG;
845 run->debug.arch.address = kvmppc_get_pc(vcpu);
846 return RESUME_HOST;
847 } else {
848 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
849 return RESUME_GUEST;
850 }
851}
852
3a167bea
AK		 853static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
854 struct task_struct *tsk)
de56a948
PM		 855{
856 int r = RESUME_HOST;
857
858 vcpu->stat.sum_exits++;
859
1c9e3d51
PM		 860 /*
861 * This can happen if an interrupt occurs in the last stages
862 * of guest entry or the first stages of guest exit (i.e. after
863 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
864 * and before setting it to KVM_GUEST_MODE_HOST_HV).
865 * That can happen due to a bug, or due to a machine check
866 * occurring at just the wrong time.
867 */
868 if (vcpu->arch.shregs.msr & MSR_HV) {
869 printk(KERN_EMERG "KVM trap in HV mode!\n");
870 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
871 vcpu->arch.trap, kvmppc_get_pc(vcpu),
872 vcpu->arch.shregs.msr);
873 kvmppc_dump_regs(vcpu);
874 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
875 run->hw.hardware_exit_reason = vcpu->arch.trap;
876 return RESUME_HOST;
877 }
de56a948
PM		 878 run->exit_reason = KVM_EXIT_UNKNOWN;
879 run->ready_for_interrupt_injection = 1;
880 switch (vcpu->arch.trap) {
881 /* We're good on these - the host merely wanted to get our attention */
882 case BOOK3S_INTERRUPT_HV_DECREMENTER:
883 vcpu->stat.dec_exits++;
884 r = RESUME_GUEST;
885 break;
886 case BOOK3S_INTERRUPT_EXTERNAL:
5d00f66b 887 case BOOK3S_INTERRUPT_H_DOORBELL:
de56a948
PM		 888 vcpu->stat.ext_intr_exits++;
889 r = RESUME_GUEST;
890 break;
dee6f24c
MS		 891 /* HMI is hypervisor interrupt and host has handled it. Resume guest.*/
892 case BOOK3S_INTERRUPT_HMI:
de56a948
PM		 893 case BOOK3S_INTERRUPT_PERFMON:
894 r = RESUME_GUEST;
895 break;
b4072df4
PM		 896 case BOOK3S_INTERRUPT_MACHINE_CHECK:
897 /*
898 * Deliver a machine check interrupt to the guest.
899 * We have to do this, even if the host has handled the
900 * machine check, because machine checks use SRR0/1 and
901 * the interrupt might have trashed guest state in them.
902 */
903 kvmppc_book3s_queue_irqprio(vcpu,
904 BOOK3S_INTERRUPT_MACHINE_CHECK);
905 r = RESUME_GUEST;
906 break;
de56a948
PM		 907 case BOOK3S_INTERRUPT_PROGRAM:
908 {
909 ulong flags;
910 /*
911 * Normally program interrupts are delivered directly
912 * to the guest by the hardware, but we can get here
913 * as a result of a hypervisor emulation interrupt
914 * (e40) getting turned into a 700 by BML RTAS.
915 */
916 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
917 kvmppc_core_queue_program(vcpu, flags);
918 r = RESUME_GUEST;
919 break;
920 }
921 case BOOK3S_INTERRUPT_SYSCALL:
922 {
923 /* hcall - punt to userspace */
924 int i;
925
27025a60
LPF		 926 /* hypercall with MSR_PR has already been handled in rmode,
927 * and never reaches here.
928 */
929
de56a948
PM		 930 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
931 for (i = 0; i < 9; ++i)
932 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
933 run->exit_reason = KVM_EXIT_PAPR_HCALL;
934 vcpu->arch.hcall_needed = 1;
935 r = RESUME_HOST;
936 break;
937 }
938 /*
342d3db7
PM		 939 * We get these next two if the guest accesses a page which it thinks
940 * it has mapped but which is not actually present, either because
941 * it is for an emulated I/O device or because the corresonding
942 * host page has been paged out. Any other HDSI/HISI interrupts
943 * have been handled already.
de56a948
PM		 944 */
945 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
913d3ff9 946 r = RESUME_PAGE_FAULT;
de56a948
PM		 947 break;
948 case BOOK3S_INTERRUPT_H_INST_STORAGE:
913d3ff9
PM		 949 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
950 vcpu->arch.fault_dsisr = 0;
951 r = RESUME_PAGE_FAULT;
de56a948
PM		 952 break;
953 /*
954 * This occurs if the guest executes an illegal instruction.
a59c1d9e
MS		 955 * If the guest debug is disabled, generate a program interrupt
956 * to the guest. If guest debug is enabled, we need to check
957 * whether the instruction is a software breakpoint instruction.
958 * Accordingly return to Guest or Host.
de56a948
PM		 959 */
960 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
4a157d61
PM		 961 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
962 vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
963 swab32(vcpu->arch.emul_inst) :
964 vcpu->arch.emul_inst;
a59c1d9e
MS		 965 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
966 r = kvmppc_emulate_debug_inst(run, vcpu);
967 } else {
968 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
969 r = RESUME_GUEST;
970 }
bd3048b8
ME		 971 break;
972 /*
973 * This occurs if the guest (kernel or userspace), does something that
974 * is prohibited by HFSCR. We just generate a program interrupt to
975 * the guest.
976 */
977 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
978 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
de56a948
PM		 979 r = RESUME_GUEST;
980 break;
981 default:
982 kvmppc_dump_regs(vcpu);
983 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
984 vcpu->arch.trap, kvmppc_get_pc(vcpu),
985 vcpu->arch.shregs.msr);
f3271d4c 986 run->hw.hardware_exit_reason = vcpu->arch.trap;
de56a948 987 r = RESUME_HOST;
de56a948
PM		 988 break;
989 }
990
de56a948
PM		 991 return r;
992}
993
3a167bea
AK		 994static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
995 struct kvm_sregs *sregs)
de56a948
PM		 996{
997 int i;
998
de56a948 999 memset(sregs, 0, sizeof(struct kvm_sregs));
87916442 1000 sregs->pvr = vcpu->arch.pvr;
de56a948
PM		 1001 for (i = 0; i < vcpu->arch.slb_max; i++) {
1002 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1003 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1004 }
1005
1006 return 0;
1007}
1008
3a167bea
AK		 1009static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1010 struct kvm_sregs *sregs)
de56a948
PM		 1011{
1012 int i, j;
1013
9333e6c4
PM		 1014 /* Only accept the same PVR as the host's, since we can't spoof it */
1015 if (sregs->pvr != vcpu->arch.pvr)
1016 return -EINVAL;
de56a948
PM		 1017
1018 j = 0;
1019 for (i = 0; i < vcpu->arch.slb_nr; i++) {
1020 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1021 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1022 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1023 ++j;
1024 }
1025 }
1026 vcpu->arch.slb_max = j;
1027
1028 return 0;
1029}
1030
a0840240
AK		 1031static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1032 bool preserve_top32)
a0144e2a 1033{
8f902b00 1034 struct kvm *kvm = vcpu->kvm;
a0144e2a
PM		 1035 struct kvmppc_vcore *vc = vcpu->arch.vcore;
1036 u64 mask;
1037
8f902b00 1038 mutex_lock(&kvm->lock);
a0144e2a 1039 spin_lock(&vc->lock);
d682916a
AB		 1040 /*
1041 * If ILE (interrupt little-endian) has changed, update the
1042 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1043 */
1044 if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
d682916a
AB		 1045 struct kvm_vcpu *vcpu;
1046 int i;
1047
d682916a
AB		 1048 kvm_for_each_vcpu(i, vcpu, kvm) {
1049 if (vcpu->arch.vcore != vc)
1050 continue;
1051 if (new_lpcr & LPCR_ILE)
1052 vcpu->arch.intr_msr |= MSR_LE;
1053 else
1054 vcpu->arch.intr_msr &= ~MSR_LE;
1055 }
d682916a
AB		 1056 }
1057
a0144e2a
PM		 1058 /*
1059 * Userspace can only modify DPFD (default prefetch depth),
1060 * ILE (interrupt little-endian) and TC (translation control).
e0622bd9 1061 * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
a0144e2a
PM		 1062 */
1063 mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
e0622bd9
PM		 1064 if (cpu_has_feature(CPU_FTR_ARCH_207S))
1065 mask |= LPCR_AIL;
a0840240
AK		 1066
1067 /* Broken 32-bit version of LPCR must not clear top bits */
1068 if (preserve_top32)
1069 mask &= 0xFFFFFFFF;
a0144e2a
PM		 1070 vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1071 spin_unlock(&vc->lock);
8f902b00 1072 mutex_unlock(&kvm->lock);
a0144e2a
PM		 1073}
1074
3a167bea
AK		 1075static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1076 union kvmppc_one_reg *val)
31f3438e 1077{
a136a8bd
PM		 1078 int r = 0;
1079 long int i;
31f3438e 1080
a136a8bd 1081 switch (id) {
a59c1d9e
MS		 1082 case KVM_REG_PPC_DEBUG_INST:
1083 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1084 break;
31f3438e 1085 case KVM_REG_PPC_HIOR:
a136a8bd
PM		 1086 *val = get_reg_val(id, 0);
1087 break;
1088 case KVM_REG_PPC_DABR:
1089 *val = get_reg_val(id, vcpu->arch.dabr);
1090 break;
8563bf52
PM		 1091 case KVM_REG_PPC_DABRX:
1092 *val = get_reg_val(id, vcpu->arch.dabrx);
1093 break;
a136a8bd
PM		 1094 case KVM_REG_PPC_DSCR:
1095 *val = get_reg_val(id, vcpu->arch.dscr);
1096 break;
1097 case KVM_REG_PPC_PURR:
1098 *val = get_reg_val(id, vcpu->arch.purr);
1099 break;
1100 case KVM_REG_PPC_SPURR:
1101 *val = get_reg_val(id, vcpu->arch.spurr);
1102 break;
1103 case KVM_REG_PPC_AMR:
1104 *val = get_reg_val(id, vcpu->arch.amr);
1105 break;
1106 case KVM_REG_PPC_UAMOR:
1107 *val = get_reg_val(id, vcpu->arch.uamor);
1108 break;
b005255e 1109 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
a136a8bd
PM		 1110 i = id - KVM_REG_PPC_MMCR0;
1111 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1112 break;
1113 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1114 i = id - KVM_REG_PPC_PMC1;
1115 *val = get_reg_val(id, vcpu->arch.pmc[i]);
31f3438e 1116 break;
b005255e
MN		 1117 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1118 i = id - KVM_REG_PPC_SPMC1;
1119 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1120 break;
14941789
PM		 1121 case KVM_REG_PPC_SIAR:
1122 *val = get_reg_val(id, vcpu->arch.siar);
1123 break;
1124 case KVM_REG_PPC_SDAR:
1125 *val = get_reg_val(id, vcpu->arch.sdar);
1126 break;
b005255e
MN		 1127 case KVM_REG_PPC_SIER:
1128 *val = get_reg_val(id, vcpu->arch.sier);
a8bd19ef 1129 break;
b005255e
MN		 1130 case KVM_REG_PPC_IAMR:
1131 *val = get_reg_val(id, vcpu->arch.iamr);
1132 break;
b005255e
MN		 1133 case KVM_REG_PPC_PSPB:
1134 *val = get_reg_val(id, vcpu->arch.pspb);
1135 break;
b005255e
MN		 1136 case KVM_REG_PPC_DPDES:
1137 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1138 break;
1139 case KVM_REG_PPC_DAWR:
1140 *val = get_reg_val(id, vcpu->arch.dawr);
1141 break;
1142 case KVM_REG_PPC_DAWRX:
1143 *val = get_reg_val(id, vcpu->arch.dawrx);
1144 break;
1145 case KVM_REG_PPC_CIABR:
1146 *val = get_reg_val(id, vcpu->arch.ciabr);
1147 break;
b005255e
MN		 1148 case KVM_REG_PPC_CSIGR:
1149 *val = get_reg_val(id, vcpu->arch.csigr);
1150 break;
1151 case KVM_REG_PPC_TACR:
1152 *val = get_reg_val(id, vcpu->arch.tacr);
1153 break;
1154 case KVM_REG_PPC_TCSCR:
1155 *val = get_reg_val(id, vcpu->arch.tcscr);
1156 break;
1157 case KVM_REG_PPC_PID:
1158 *val = get_reg_val(id, vcpu->arch.pid);
1159 break;
1160 case KVM_REG_PPC_ACOP:
1161 *val = get_reg_val(id, vcpu->arch.acop);
1162 break;
1163 case KVM_REG_PPC_WORT:
1164 *val = get_reg_val(id, vcpu->arch.wort);
a8bd19ef 1165 break;
55b665b0
PM		 1166 case KVM_REG_PPC_VPA_ADDR:
1167 spin_lock(&vcpu->arch.vpa_update_lock);
1168 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1169 spin_unlock(&vcpu->arch.vpa_update_lock);
1170 break;
1171 case KVM_REG_PPC_VPA_SLB:
1172 spin_lock(&vcpu->arch.vpa_update_lock);
1173 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1174 val->vpaval.length = vcpu->arch.slb_shadow.len;
1175 spin_unlock(&vcpu->arch.vpa_update_lock);
1176 break;
1177 case KVM_REG_PPC_VPA_DTL:
1178 spin_lock(&vcpu->arch.vpa_update_lock);
1179 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1180 val->vpaval.length = vcpu->arch.dtl.len;
1181 spin_unlock(&vcpu->arch.vpa_update_lock);
1182 break;
93b0f4dc
PM		 1183 case KVM_REG_PPC_TB_OFFSET:
1184 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1185 break;
a0144e2a 1186 case KVM_REG_PPC_LPCR:
a0840240 1187 case KVM_REG_PPC_LPCR_64:
a0144e2a
PM		 1188 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1189 break;
4b8473c9
PM		 1190 case KVM_REG_PPC_PPR:
1191 *val = get_reg_val(id, vcpu->arch.ppr);
1192 break;
a7d80d01
MN		 1193#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1194 case KVM_REG_PPC_TFHAR:
1195 *val = get_reg_val(id, vcpu->arch.tfhar);
1196 break;
1197 case KVM_REG_PPC_TFIAR:
1198 *val = get_reg_val(id, vcpu->arch.tfiar);
1199 break;
1200 case KVM_REG_PPC_TEXASR:
1201 *val = get_reg_val(id, vcpu->arch.texasr);
1202 break;
1203 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1204 i = id - KVM_REG_PPC_TM_GPR0;
1205 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1206 break;
1207 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1208 {
1209 int j;
1210 i = id - KVM_REG_PPC_TM_VSR0;
1211 if (i < 32)
1212 for (j = 0; j < TS_FPRWIDTH; j++)
1213 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1214 else {
1215 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1216 val->vval = vcpu->arch.vr_tm.vr[i-32];
1217 else
1218 r = -ENXIO;
1219 }
1220 break;
1221 }
1222 case KVM_REG_PPC_TM_CR:
1223 *val = get_reg_val(id, vcpu->arch.cr_tm);
1224 break;
1225 case KVM_REG_PPC_TM_LR:
1226 *val = get_reg_val(id, vcpu->arch.lr_tm);
1227 break;
1228 case KVM_REG_PPC_TM_CTR:
1229 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1230 break;
1231 case KVM_REG_PPC_TM_FPSCR:
1232 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1233 break;
1234 case KVM_REG_PPC_TM_AMR:
1235 *val = get_reg_val(id, vcpu->arch.amr_tm);
1236 break;
1237 case KVM_REG_PPC_TM_PPR:
1238 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1239 break;
1240 case KVM_REG_PPC_TM_VRSAVE:
1241 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1242 break;
1243 case KVM_REG_PPC_TM_VSCR:
1244 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1245 *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1246 else
1247 r = -ENXIO;
1248 break;
1249 case KVM_REG_PPC_TM_DSCR:
1250 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1251 break;
1252 case KVM_REG_PPC_TM_TAR:
1253 *val = get_reg_val(id, vcpu->arch.tar_tm);
1254 break;
1255#endif
388cc6e1
PM		 1256 case KVM_REG_PPC_ARCH_COMPAT:
1257 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1258 break;
31f3438e 1259 default:
a136a8bd 1260 r = -EINVAL;
31f3438e
PM		 1261 break;
1262 }
1263
1264 return r;
1265}
1266
3a167bea
AK		 1267static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1268 union kvmppc_one_reg *val)
31f3438e 1269{
a136a8bd
PM		 1270 int r = 0;
1271 long int i;
55b665b0 1272 unsigned long addr, len;
31f3438e 1273
a136a8bd 1274 switch (id) {
31f3438e 1275 case KVM_REG_PPC_HIOR:
31f3438e 1276 /* Only allow this to be set to zero */
a136a8bd 1277 if (set_reg_val(id, *val))
31f3438e
PM		 1278 r = -EINVAL;
1279 break;
a136a8bd
PM		 1280 case KVM_REG_PPC_DABR:
1281 vcpu->arch.dabr = set_reg_val(id, *val);
1282 break;
8563bf52
PM		 1283 case KVM_REG_PPC_DABRX:
1284 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1285 break;
a136a8bd
PM		 1286 case KVM_REG_PPC_DSCR:
1287 vcpu->arch.dscr = set_reg_val(id, *val);
1288 break;
1289 case KVM_REG_PPC_PURR:
1290 vcpu->arch.purr = set_reg_val(id, *val);
1291 break;
1292 case KVM_REG_PPC_SPURR:
1293 vcpu->arch.spurr = set_reg_val(id, *val);
1294 break;
1295 case KVM_REG_PPC_AMR:
1296 vcpu->arch.amr = set_reg_val(id, *val);
1297 break;
1298 case KVM_REG_PPC_UAMOR:
1299 vcpu->arch.uamor = set_reg_val(id, *val);
1300 break;
b005255e 1301 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
a136a8bd
PM		 1302 i = id - KVM_REG_PPC_MMCR0;
1303 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1304 break;
1305 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1306 i = id - KVM_REG_PPC_PMC1;
1307 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1308 break;
b005255e
MN		 1309 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1310 i = id - KVM_REG_PPC_SPMC1;
1311 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1312 break;
14941789
PM		 1313 case KVM_REG_PPC_SIAR:
1314 vcpu->arch.siar = set_reg_val(id, *val);
1315 break;
1316 case KVM_REG_PPC_SDAR:
1317 vcpu->arch.sdar = set_reg_val(id, *val);
1318 break;
b005255e
MN		 1319 case KVM_REG_PPC_SIER:
1320 vcpu->arch.sier = set_reg_val(id, *val);
a8bd19ef 1321 break;
b005255e
MN		 1322 case KVM_REG_PPC_IAMR:
1323 vcpu->arch.iamr = set_reg_val(id, *val);
1324 break;
b005255e
MN		 1325 case KVM_REG_PPC_PSPB:
1326 vcpu->arch.pspb = set_reg_val(id, *val);
1327 break;
b005255e
MN		 1328 case KVM_REG_PPC_DPDES:
1329 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1330 break;
1331 case KVM_REG_PPC_DAWR:
1332 vcpu->arch.dawr = set_reg_val(id, *val);
1333 break;
1334 case KVM_REG_PPC_DAWRX:
1335 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1336 break;
1337 case KVM_REG_PPC_CIABR:
1338 vcpu->arch.ciabr = set_reg_val(id, *val);
1339 /* Don't allow setting breakpoints in hypervisor code */
1340 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1341 vcpu->arch.ciabr &= ~CIABR_PRIV; /* disable */
1342 break;
b005255e
MN		 1343 case KVM_REG_PPC_CSIGR:
1344 vcpu->arch.csigr = set_reg_val(id, *val);
1345 break;
1346 case KVM_REG_PPC_TACR:
1347 vcpu->arch.tacr = set_reg_val(id, *val);
1348 break;
1349 case KVM_REG_PPC_TCSCR:
1350 vcpu->arch.tcscr = set_reg_val(id, *val);
1351 break;
1352 case KVM_REG_PPC_PID:
1353 vcpu->arch.pid = set_reg_val(id, *val);
1354 break;
1355 case KVM_REG_PPC_ACOP:
1356 vcpu->arch.acop = set_reg_val(id, *val);
1357 break;
1358 case KVM_REG_PPC_WORT:
1359 vcpu->arch.wort = set_reg_val(id, *val);
a8bd19ef 1360 break;
55b665b0
PM		 1361 case KVM_REG_PPC_VPA_ADDR:
1362 addr = set_reg_val(id, *val);
1363 r = -EINVAL;
1364 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1365 vcpu->arch.dtl.next_gpa))
1366 break;
1367 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1368 break;
1369 case KVM_REG_PPC_VPA_SLB:
1370 addr = val->vpaval.addr;
1371 len = val->vpaval.length;
1372 r = -EINVAL;
1373 if (addr && !vcpu->arch.vpa.next_gpa)
1374 break;
1375 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1376 break;
1377 case KVM_REG_PPC_VPA_DTL:
1378 addr = val->vpaval.addr;
1379 len = val->vpaval.length;
1380 r = -EINVAL;
9f8c8c78
PM		 1381 if (addr && (len < sizeof(struct dtl_entry) ||
1382 !vcpu->arch.vpa.next_gpa))
55b665b0
PM		 1383 break;
1384 len -= len % sizeof(struct dtl_entry);
1385 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1386 break;
93b0f4dc
PM		 1387 case KVM_REG_PPC_TB_OFFSET:
1388 /* round up to multiple of 2^24 */
1389 vcpu->arch.vcore->tb_offset =
1390 ALIGN(set_reg_val(id, *val), 1UL << 24);
1391 break;
a0144e2a 1392 case KVM_REG_PPC_LPCR:
a0840240
AK		 1393 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1394 break;
1395 case KVM_REG_PPC_LPCR_64:
1396 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
a0144e2a 1397 break;
4b8473c9
PM		 1398 case KVM_REG_PPC_PPR:
1399 vcpu->arch.ppr = set_reg_val(id, *val);
1400 break;
a7d80d01
MN		 1401#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1402 case KVM_REG_PPC_TFHAR:
1403 vcpu->arch.tfhar = set_reg_val(id, *val);
1404 break;
1405 case KVM_REG_PPC_TFIAR:
1406 vcpu->arch.tfiar = set_reg_val(id, *val);
1407 break;
1408 case KVM_REG_PPC_TEXASR:
1409 vcpu->arch.texasr = set_reg_val(id, *val);
1410 break;
1411 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1412 i = id - KVM_REG_PPC_TM_GPR0;
1413 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1414 break;
1415 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1416 {
1417 int j;
1418 i = id - KVM_REG_PPC_TM_VSR0;
1419 if (i < 32)
1420 for (j = 0; j < TS_FPRWIDTH; j++)
1421 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1422 else
1423 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1424 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1425 else
1426 r = -ENXIO;
1427 break;
1428 }
1429 case KVM_REG_PPC_TM_CR:
1430 vcpu->arch.cr_tm = set_reg_val(id, *val);
1431 break;
1432 case KVM_REG_PPC_TM_LR:
1433 vcpu->arch.lr_tm = set_reg_val(id, *val);
1434 break;
1435 case KVM_REG_PPC_TM_CTR:
1436 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1437 break;
1438 case KVM_REG_PPC_TM_FPSCR:
1439 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1440 break;
1441 case KVM_REG_PPC_TM_AMR:
1442 vcpu->arch.amr_tm = set_reg_val(id, *val);
1443 break;
1444 case KVM_REG_PPC_TM_PPR:
1445 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1446 break;
1447 case KVM_REG_PPC_TM_VRSAVE:
1448 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1449 break;
1450 case KVM_REG_PPC_TM_VSCR:
1451 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1452 vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1453 else
1454 r = - ENXIO;
1455 break;
1456 case KVM_REG_PPC_TM_DSCR:
1457 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1458 break;
1459 case KVM_REG_PPC_TM_TAR:
1460 vcpu->arch.tar_tm = set_reg_val(id, *val);
1461 break;
1462#endif
388cc6e1
PM		 1463 case KVM_REG_PPC_ARCH_COMPAT:
1464 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1465 break;
31f3438e 1466 default:
a136a8bd 1467 r = -EINVAL;
31f3438e
PM		 1468 break;
1469 }
1470
1471 return r;
1472}
1473
de9bdd1a
SS		 1474static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1475{
1476 struct kvmppc_vcore *vcore;
1477
1478 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1479
1480 if (vcore == NULL)
1481 return NULL;
1482
1483 INIT_LIST_HEAD(&vcore->runnable_threads);
1484 spin_lock_init(&vcore->lock);
2711e248 1485 spin_lock_init(&vcore->stoltb_lock);
de9bdd1a
SS		 1486 init_waitqueue_head(&vcore->wq);
1487 vcore->preempt_tb = TB_NIL;
1488 vcore->lpcr = kvm->arch.lpcr;
1489 vcore->first_vcpuid = core * threads_per_subcore;
1490 vcore->kvm = kvm;
ec257165 1491 INIT_LIST_HEAD(&vcore->preempt_list);
de9bdd1a
SS		 1492
1493 return vcore;
1494}
1495
b6c295df
PM		 1496#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1497static struct debugfs_timings_element {
1498 const char *name;
1499 size_t offset;
1500} timings[] = {
1501 {"rm_entry", offsetof(struct kvm_vcpu, arch.rm_entry)},
1502 {"rm_intr", offsetof(struct kvm_vcpu, arch.rm_intr)},
1503 {"rm_exit", offsetof(struct kvm_vcpu, arch.rm_exit)},
1504 {"guest", offsetof(struct kvm_vcpu, arch.guest_time)},
1505 {"cede", offsetof(struct kvm_vcpu, arch.cede_time)},
1506};
1507
1508#define N_TIMINGS (sizeof(timings) / sizeof(timings[0]))
1509
1510struct debugfs_timings_state {
1511 struct kvm_vcpu *vcpu;
1512 unsigned int buflen;
1513 char buf[N_TIMINGS * 100];
1514};
1515
1516static int debugfs_timings_open(struct inode *inode, struct file *file)
1517{
1518 struct kvm_vcpu *vcpu = inode->i_private;
1519 struct debugfs_timings_state *p;
1520
1521 p = kzalloc(sizeof(*p), GFP_KERNEL);
1522 if (!p)
1523 return -ENOMEM;
1524
1525 kvm_get_kvm(vcpu->kvm);
1526 p->vcpu = vcpu;
1527 file->private_data = p;
1528
1529 return nonseekable_open(inode, file);
1530}
1531
1532static int debugfs_timings_release(struct inode *inode, struct file *file)
1533{
1534 struct debugfs_timings_state *p = file->private_data;
1535
1536 kvm_put_kvm(p->vcpu->kvm);
1537 kfree(p);
1538 return 0;
1539}
1540
1541static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1542 size_t len, loff_t *ppos)
1543{
1544 struct debugfs_timings_state *p = file->private_data;
1545 struct kvm_vcpu *vcpu = p->vcpu;
1546 char *s, *buf_end;
1547 struct kvmhv_tb_accumulator tb;
1548 u64 count;
1549 loff_t pos;
1550 ssize_t n;
1551 int i, loops;
1552 bool ok;
1553
1554 if (!p->buflen) {
1555 s = p->buf;
1556 buf_end = s + sizeof(p->buf);
1557 for (i = 0; i < N_TIMINGS; ++i) {
1558 struct kvmhv_tb_accumulator *acc;
1559
1560 acc = (struct kvmhv_tb_accumulator *)
1561 ((unsigned long)vcpu + timings[i].offset);
1562 ok = false;
1563 for (loops = 0; loops < 1000; ++loops) {
1564 count = acc->seqcount;
1565 if (!(count & 1)) {
1566 smp_rmb();
1567 tb = *acc;
1568 smp_rmb();
1569 if (count == acc->seqcount) {
1570 ok = true;
1571 break;
1572 }
1573 }
1574 udelay(1);
1575 }
1576 if (!ok)
1577 snprintf(s, buf_end - s, "%s: stuck\n",
1578 timings[i].name);
1579 else
1580 snprintf(s, buf_end - s,
1581 "%s: %llu %llu %llu %llu\n",
1582 timings[i].name, count / 2,
1583 tb_to_ns(tb.tb_total),
1584 tb_to_ns(tb.tb_min),
1585 tb_to_ns(tb.tb_max));
1586 s += strlen(s);
1587 }
1588 p->buflen = s - p->buf;
1589 }
1590
1591 pos = *ppos;
1592 if (pos >= p->buflen)
1593 return 0;
1594 if (len > p->buflen - pos)
1595 len = p->buflen - pos;
1596 n = copy_to_user(buf, p->buf + pos, len);
1597 if (n) {
1598 if (n == len)
1599 return -EFAULT;
1600 len -= n;
1601 }
1602 *ppos = pos + len;
1603 return len;
1604}
1605
1606static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1607 size_t len, loff_t *ppos)
1608{
1609 return -EACCES;
1610}
1611
1612static const struct file_operations debugfs_timings_ops = {
1613 .owner = THIS_MODULE,
1614 .open = debugfs_timings_open,
1615 .release = debugfs_timings_release,
1616 .read = debugfs_timings_read,
1617 .write = debugfs_timings_write,
1618 .llseek = generic_file_llseek,
1619};
1620
1621/* Create a debugfs directory for the vcpu */
1622static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1623{
1624 char buf[16];
1625 struct kvm *kvm = vcpu->kvm;
1626
1627 snprintf(buf, sizeof(buf), "vcpu%u", id);
1628 if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1629 return;
1630 vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1631 if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1632 return;
1633 vcpu->arch.debugfs_timings =
1634 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1635 vcpu, &debugfs_timings_ops);
1636}
1637
1638#else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1639static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1640{
1641}
1642#endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1643
3a167bea
AK		 1644static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1645 unsigned int id)
de56a948
PM		 1646{
1647 struct kvm_vcpu *vcpu;
371fefd6
PM		 1648 int err = -EINVAL;
1649 int core;
1650 struct kvmppc_vcore *vcore;
de56a948 1651
3102f784 1652 core = id / threads_per_subcore;
371fefd6
PM		 1653 if (core >= KVM_MAX_VCORES)
1654 goto out;
1655
1656 err = -ENOMEM;
6b75e6bf 1657 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
de56a948
PM		 1658 if (!vcpu)
1659 goto out;
1660
1661 err = kvm_vcpu_init(vcpu, kvm, id);
1662 if (err)
1663 goto free_vcpu;
1664
1665 vcpu->arch.shared = &vcpu->arch.shregs;
5deb8e7a
AG		 1666#ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1667 /*
1668 * The shared struct is never shared on HV,
1669 * so we can always use host endianness
1670 */
1671#ifdef __BIG_ENDIAN__
1672 vcpu->arch.shared_big_endian = true;
1673#else
1674 vcpu->arch.shared_big_endian = false;
1675#endif
1676#endif
de56a948
PM		 1677 vcpu->arch.mmcr[0] = MMCR0_FC;
1678 vcpu->arch.ctrl = CTRL_RUNLATCH;
1679 /* default to host PVR, since we can't spoof it */
3a167bea 1680 kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2e25aa5f 1681 spin_lock_init(&vcpu->arch.vpa_update_lock);
c7b67670
PM		 1682 spin_lock_init(&vcpu->arch.tbacct_lock);
1683 vcpu->arch.busy_preempt = TB_NIL;
d682916a 1684 vcpu->arch.intr_msr = MSR_SF | MSR_ME;
de56a948 1685
de56a948
PM		 1686 kvmppc_mmu_book3s_hv_init(vcpu);
1687
8455d79e 1688 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
371fefd6
PM		 1689
1690 init_waitqueue_head(&vcpu->arch.cpu_run);
1691
1692 mutex_lock(&kvm->lock);
1693 vcore = kvm->arch.vcores[core];
1694 if (!vcore) {
de9bdd1a 1695 vcore = kvmppc_vcore_create(kvm, core);
371fefd6 1696 kvm->arch.vcores[core] = vcore;
1b400ba0 1697 kvm->arch.online_vcores++;
371fefd6
PM		 1698 }
1699 mutex_unlock(&kvm->lock);
1700
1701 if (!vcore)
1702 goto free_vcpu;
1703
1704 spin_lock(&vcore->lock);
1705 ++vcore->num_threads;
371fefd6
PM		 1706 spin_unlock(&vcore->lock);
1707 vcpu->arch.vcore = vcore;
e0b7ec05 1708 vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
ec257165 1709 vcpu->arch.thread_cpu = -1;
371fefd6 1710
af8f38b3
AG		 1711 vcpu->arch.cpu_type = KVM_CPU_3S_64;
1712 kvmppc_sanity_check(vcpu);
1713
b6c295df
PM		 1714 debugfs_vcpu_init(vcpu, id);
1715
de56a948
PM		 1716 return vcpu;
1717
1718free_vcpu:
6b75e6bf 1719 kmem_cache_free(kvm_vcpu_cache, vcpu);
de56a948
PM		 1720out:
1721 return ERR_PTR(err);
1722}
1723
c35635ef
PM		 1724static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1725{
1726 if (vpa->pinned_addr)
1727 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1728 vpa->dirty);
1729}
1730
3a167bea 1731static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
de56a948 1732{
2e25aa5f 1733 spin_lock(&vcpu->arch.vpa_update_lock);
c35635ef
PM		 1734 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1735 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1736 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2e25aa5f 1737 spin_unlock(&vcpu->arch.vpa_update_lock);
de56a948 1738 kvm_vcpu_uninit(vcpu);
6b75e6bf 1739 kmem_cache_free(kvm_vcpu_cache, vcpu);
de56a948
PM		 1740}
1741
3a167bea
AK		 1742static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1743{
1744 /* Indicate we want to get back into the guest */
1745 return 1;
1746}
1747
19ccb76a 1748static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
371fefd6 1749{
19ccb76a 1750 unsigned long dec_nsec, now;
371fefd6 1751
19ccb76a
PM		 1752 now = get_tb();
1753 if (now > vcpu->arch.dec_expires) {
1754 /* decrementer has already gone negative */
1755 kvmppc_core_queue_dec(vcpu);
7e28e60e 1756 kvmppc_core_prepare_to_enter(vcpu);
19ccb76a 1757 return;
371fefd6 1758 }
19ccb76a
PM		 1759 dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1760 / tb_ticks_per_sec;
1761 hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1762 HRTIMER_MODE_REL);
1763 vcpu->arch.timer_running = 1;
371fefd6
PM		 1764}
1765
19ccb76a 1766static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
371fefd6 1767{
19ccb76a
PM		 1768 vcpu->arch.ceded = 0;
1769 if (vcpu->arch.timer_running) {
1770 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1771 vcpu->arch.timer_running = 0;
1772 }
371fefd6
PM		 1773}
1774
e0b7ec05 1775extern void __kvmppc_vcore_entry(void);
de56a948 1776
371fefd6
PM		 1777static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1778 struct kvm_vcpu *vcpu)
de56a948 1779{
c7b67670
PM		 1780 u64 now;
1781
371fefd6
PM		 1782 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1783 return;
bf3d32e1 1784 spin_lock_irq(&vcpu->arch.tbacct_lock);
c7b67670
PM		 1785 now = mftb();
1786 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1787 vcpu->arch.stolen_logged;
1788 vcpu->arch.busy_preempt = now;
1789 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
bf3d32e1 1790 spin_unlock_irq(&vcpu->arch.tbacct_lock);
371fefd6 1791 --vc->n_runnable;
371fefd6
PM		 1792 list_del(&vcpu->arch.run_list);
1793}
1794
f0888f70
PM		 1795static int kvmppc_grab_hwthread(int cpu)
1796{
1797 struct paca_struct *tpaca;
b754c739 1798 long timeout = 10000;
f0888f70
PM		 1799
1800 tpaca = &paca[cpu];
1801
1802 /* Ensure the thread won't go into the kernel if it wakes */
7b444c67 1803 tpaca->kvm_hstate.kvm_vcpu = NULL;
b4deba5c 1804 tpaca->kvm_hstate.kvm_vcore = NULL;
5d5b99cd
PM		 1805 tpaca->kvm_hstate.napping = 0;
1806 smp_wmb();
1807 tpaca->kvm_hstate.hwthread_req = 1;
f0888f70
PM		 1808
1809 /*
1810 * If the thread is already executing in the kernel (e.g. handling
1811 * a stray interrupt), wait for it to get back to nap mode.
1812 * The smp_mb() is to ensure that our setting of hwthread_req
1813 * is visible before we look at hwthread_state, so if this
1814 * races with the code at system_reset_pSeries and the thread
1815 * misses our setting of hwthread_req, we are sure to see its
1816 * setting of hwthread_state, and vice versa.
1817 */
1818 smp_mb();
1819 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1820 if (--timeout <= 0) {
1821 pr_err("KVM: couldn't grab cpu %d\n", cpu);
1822 return -EBUSY;
1823 }
1824 udelay(1);
1825 }
1826 return 0;
1827}
1828
1829static void kvmppc_release_hwthread(int cpu)
1830{
1831 struct paca_struct *tpaca;
1832
1833 tpaca = &paca[cpu];
1834 tpaca->kvm_hstate.hwthread_req = 0;
1835 tpaca->kvm_hstate.kvm_vcpu = NULL;
b4deba5c
PM		 1836 tpaca->kvm_hstate.kvm_vcore = NULL;
1837 tpaca->kvm_hstate.kvm_split_mode = NULL;
f0888f70
PM		 1838}
1839
b4deba5c 1840static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
371fefd6
PM		 1841{
1842 int cpu;
1843 struct paca_struct *tpaca;
ec257165 1844 struct kvmppc_vcore *mvc = vc->master_vcore;
371fefd6 1845
b4deba5c
PM		 1846 cpu = vc->pcpu;
1847 if (vcpu) {
1848 if (vcpu->arch.timer_running) {
1849 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1850 vcpu->arch.timer_running = 0;
1851 }
1852 cpu += vcpu->arch.ptid;
1853 vcpu->cpu = mvc->pcpu;
1854 vcpu->arch.thread_cpu = cpu;
19ccb76a 1855 }
371fefd6 1856 tpaca = &paca[cpu];
5d5b99cd 1857 tpaca->kvm_hstate.kvm_vcpu = vcpu;
ec257165 1858 tpaca->kvm_hstate.ptid = cpu - mvc->pcpu;
ec257165 1859 /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
371fefd6 1860 smp_wmb();
b4deba5c 1861 tpaca->kvm_hstate.kvm_vcore = mvc;
5d5b99cd 1862 if (cpu != smp_processor_id())
66feed61 1863 kvmppc_ipi_thread(cpu);
371fefd6 1864}
de56a948 1865
5d5b99cd 1866static void kvmppc_wait_for_nap(void)
371fefd6 1867{
5d5b99cd
PM		 1868 int cpu = smp_processor_id();
1869 int i, loops;
371fefd6 1870
5d5b99cd
PM		 1871 for (loops = 0; loops < 1000000; ++loops) {
1872 /*
1873 * Check if all threads are finished.
b4deba5c 1874 * We set the vcore pointer when starting a thread
5d5b99cd 1875 * and the thread clears it when finished, so we look
b4deba5c 1876 * for any threads that still have a non-NULL vcore ptr.
5d5b99cd
PM		 1877 */
1878 for (i = 1; i < threads_per_subcore; ++i)
b4deba5c 1879 if (paca[cpu + i].kvm_hstate.kvm_vcore)
5d5b99cd
PM		 1880 break;
1881 if (i == threads_per_subcore) {
1882 HMT_medium();
1883 return;
371fefd6 1884 }
5d5b99cd 1885 HMT_low();
371fefd6
PM		 1886 }
1887 HMT_medium();
5d5b99cd 1888 for (i = 1; i < threads_per_subcore; ++i)
b4deba5c 1889 if (paca[cpu + i].kvm_hstate.kvm_vcore)
5d5b99cd 1890 pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
371fefd6
PM		 1891}
1892
1893/*
1894 * Check that we are on thread 0 and that any other threads in
7b444c67
PM		 1895 * this core are off-line. Then grab the threads so they can't
1896 * enter the kernel.
371fefd6
PM		 1897 */
1898static int on_primary_thread(void)
1899{
1900 int cpu = smp_processor_id();
3102f784 1901 int thr;
371fefd6 1902
3102f784
ME		 1903 /* Are we on a primary subcore? */
1904 if (cpu_thread_in_subcore(cpu))
371fefd6 1905 return 0;
3102f784
ME		 1906
1907 thr = 0;
1908 while (++thr < threads_per_subcore)
371fefd6
PM		 1909 if (cpu_online(cpu + thr))
1910 return 0;
7b444c67
PM		 1911
1912 /* Grab all hw threads so they can't go into the kernel */
3102f784 1913 for (thr = 1; thr < threads_per_subcore; ++thr) {
7b444c67
PM		 1914 if (kvmppc_grab_hwthread(cpu + thr)) {
1915 /* Couldn't grab one; let the others go */
1916 do {
1917 kvmppc_release_hwthread(cpu + thr);
1918 } while (--thr > 0);
1919 return 0;
1920 }
1921 }
371fefd6
PM		 1922 return 1;
1923}
1924
ec257165
PM		 1925/*
1926 * A list of virtual cores for each physical CPU.
1927 * These are vcores that could run but their runner VCPU tasks are
1928 * (or may be) preempted.
1929 */
1930struct preempted_vcore_list {
1931 struct list_head list;
1932 spinlock_t lock;
1933};
1934
1935static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
1936
1937static void init_vcore_lists(void)
1938{
1939 int cpu;
1940
1941 for_each_possible_cpu(cpu) {
1942 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
1943 spin_lock_init(&lp->lock);
1944 INIT_LIST_HEAD(&lp->list);
1945 }
1946}
1947
1948static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
1949{
1950 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
1951
1952 vc->vcore_state = VCORE_PREEMPT;
1953 vc->pcpu = smp_processor_id();
1954 if (vc->num_threads < threads_per_subcore) {
1955 spin_lock(&lp->lock);
1956 list_add_tail(&vc->preempt_list, &lp->list);
1957 spin_unlock(&lp->lock);
1958 }
1959
1960 /* Start accumulating stolen time */
1961 kvmppc_core_start_stolen(vc);
1962}
1963
1964static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
1965{
402813fe 1966 struct preempted_vcore_list *lp;
ec257165
PM		 1967
1968 kvmppc_core_end_stolen(vc);
1969 if (!list_empty(&vc->preempt_list)) {
402813fe 1970 lp = &per_cpu(preempted_vcores, vc->pcpu);
ec257165
PM		 1971 spin_lock(&lp->lock);
1972 list_del_init(&vc->preempt_list);
1973 spin_unlock(&lp->lock);
1974 }
1975 vc->vcore_state = VCORE_INACTIVE;
1976}
1977
b4deba5c
PM		 1978/*
1979 * This stores information about the virtual cores currently
1980 * assigned to a physical core.
1981 */
ec257165 1982struct core_info {
b4deba5c
PM		 1983 int n_subcores;
1984 int max_subcore_threads;
ec257165 1985 int total_threads;
b4deba5c
PM		 1986 int subcore_threads[MAX_SUBCORES];
1987 struct kvm *subcore_vm[MAX_SUBCORES];
1988 struct list_head vcs[MAX_SUBCORES];
ec257165
PM		 1989};
1990
b4deba5c
PM		 1991/*
1992 * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
1993 * respectively in 2-way micro-threading (split-core) mode.
1994 */
1995static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
1996
ec257165
PM		 1997static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
1998{
b4deba5c
PM		 1999 int sub;
2000
ec257165 2001 memset(cip, 0, sizeof(*cip));
b4deba5c
PM		 2002 cip->n_subcores = 1;
2003 cip->max_subcore_threads = vc->num_threads;
ec257165 2004 cip->total_threads = vc->num_threads;
b4deba5c
PM		 2005 cip->subcore_threads[0] = vc->num_threads;
2006 cip->subcore_vm[0] = vc->kvm;
2007 for (sub = 0; sub < MAX_SUBCORES; ++sub)
2008 INIT_LIST_HEAD(&cip->vcs[sub]);
2009 list_add_tail(&vc->preempt_list, &cip->vcs[0]);
2010}
2011
2012static bool subcore_config_ok(int n_subcores, int n_threads)
2013{
2014 /* Can only dynamically split if unsplit to begin with */
2015 if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2016 return false;
2017 if (n_subcores > MAX_SUBCORES)
2018 return false;
2019 if (n_subcores > 1) {
2020 if (!(dynamic_mt_modes & 2))
2021 n_subcores = 4;
2022 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2023 return false;
2024 }
2025
2026 return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
ec257165
PM		 2027}
2028
2029static void init_master_vcore(struct kvmppc_vcore *vc)
2030{
2031 vc->master_vcore = vc;
2032 vc->entry_exit_map = 0;
2033 vc->in_guest = 0;
2034 vc->napping_threads = 0;
2035 vc->conferring_threads = 0;
2036}
2037
2038/*
b4deba5c
PM		 2039 * See if the existing subcores can be split into 3 (or fewer) subcores
2040 * of at most two threads each, so we can fit in another vcore. This
2041 * assumes there are at most two subcores and at most 6 threads in total.
ec257165 2042 */
b4deba5c
PM		 2043static bool can_split_piggybacked_subcores(struct core_info *cip)
2044{
2045 int sub, new_sub;
2046 int large_sub = -1;
2047 int thr;
2048 int n_subcores = cip->n_subcores;
2049 struct kvmppc_vcore *vc, *vcnext;
2050 struct kvmppc_vcore *master_vc = NULL;
2051
2052 for (sub = 0; sub < cip->n_subcores; ++sub) {
2053 if (cip->subcore_threads[sub] <= 2)
2054 continue;
2055 if (large_sub >= 0)
2056 return false;
2057 large_sub = sub;
2058 vc = list_first_entry(&cip->vcs[sub], struct kvmppc_vcore,
2059 preempt_list);
2060 if (vc->num_threads > 2)
2061 return false;
2062 n_subcores += (cip->subcore_threads[sub] - 1) >> 1;
2063 }
f74f2e2e 2064 if (large_sub < 0 || !subcore_config_ok(n_subcores + 1, 2))
b4deba5c
PM		 2065 return false;
2066
2067 /*
2068 * Seems feasible, so go through and move vcores to new subcores.
2069 * Note that when we have two or more vcores in one subcore,
2070 * all those vcores must have only one thread each.
2071 */
2072 new_sub = cip->n_subcores;
2073 thr = 0;
2074 sub = large_sub;
2075 list_for_each_entry_safe(vc, vcnext, &cip->vcs[sub], preempt_list) {
2076 if (thr >= 2) {
2077 list_del(&vc->preempt_list);
2078 list_add_tail(&vc->preempt_list, &cip->vcs[new_sub]);
2079 /* vc->num_threads must be 1 */
2080 if (++cip->subcore_threads[new_sub] == 1) {
2081 cip->subcore_vm[new_sub] = vc->kvm;
2082 init_master_vcore(vc);
2083 master_vc = vc;
2084 ++cip->n_subcores;
2085 } else {
2086 vc->master_vcore = master_vc;
2087 ++new_sub;
2088 }
2089 }
2090 thr += vc->num_threads;
2091 }
2092 cip->subcore_threads[large_sub] = 2;
2093 cip->max_subcore_threads = 2;
2094
2095 return true;
2096}
2097
2098static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2099{
2100 int n_threads = vc->num_threads;
2101 int sub;
2102
2103 if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2104 return false;
2105
2106 if (n_threads < cip->max_subcore_threads)
2107 n_threads = cip->max_subcore_threads;
2108 if (subcore_config_ok(cip->n_subcores + 1, n_threads)) {
2109 cip->max_subcore_threads = n_threads;
2110 } else if (cip->n_subcores <= 2 && cip->total_threads <= 6 &&
2111 vc->num_threads <= 2) {
2112 /*
2113 * We may be able to fit another subcore in by
2114 * splitting an existing subcore with 3 or 4
2115 * threads into two 2-thread subcores, or one
2116 * with 5 or 6 threads into three subcores.
2117 * We can only do this if those subcores have
2118 * piggybacked virtual cores.
2119 */
2120 if (!can_split_piggybacked_subcores(cip))
2121 return false;
2122 } else {
2123 return false;
2124 }
2125
2126 sub = cip->n_subcores;
2127 ++cip->n_subcores;
2128 cip->total_threads += vc->num_threads;
2129 cip->subcore_threads[sub] = vc->num_threads;
2130 cip->subcore_vm[sub] = vc->kvm;
2131 init_master_vcore(vc);
2132 list_del(&vc->preempt_list);
2133 list_add_tail(&vc->preempt_list, &cip->vcs[sub]);
2134
2135 return true;
2136}
2137
2138static bool can_piggyback_subcore(struct kvmppc_vcore *pvc,
2139 struct core_info *cip, int sub)
ec257165
PM		 2140{
2141 struct kvmppc_vcore *vc;
b4deba5c 2142 int n_thr;
ec257165 2143
b4deba5c
PM		 2144 vc = list_first_entry(&cip->vcs[sub], struct kvmppc_vcore,
2145 preempt_list);
ec257165
PM		 2146
2147 /* require same VM and same per-core reg values */
2148 if (pvc->kvm != vc->kvm ||
2149 pvc->tb_offset != vc->tb_offset ||
2150 pvc->pcr != vc->pcr ||
2151 pvc->lpcr != vc->lpcr)
2152 return false;
2153
2154 /* P8 guest with > 1 thread per core would see wrong TIR value */
2155 if (cpu_has_feature(CPU_FTR_ARCH_207S) &&
2156 (vc->num_threads > 1 || pvc->num_threads > 1))
2157 return false;
2158
b4deba5c
PM		 2159 n_thr = cip->subcore_threads[sub] + pvc->num_threads;
2160 if (n_thr > cip->max_subcore_threads) {
2161 if (!subcore_config_ok(cip->n_subcores, n_thr))
2162 return false;
2163 cip->max_subcore_threads = n_thr;
2164 }
ec257165
PM		 2165
2166 cip->total_threads += pvc->num_threads;
b4deba5c 2167 cip->subcore_threads[sub] = n_thr;
ec257165
PM		 2168 pvc->master_vcore = vc;
2169 list_del(&pvc->preempt_list);
b4deba5c 2170 list_add_tail(&pvc->preempt_list, &cip->vcs[sub]);
ec257165
PM		 2171
2172 return true;
2173}
2174
b4deba5c
PM		 2175/*
2176 * Work out whether it is possible to piggyback the execution of
2177 * vcore *pvc onto the execution of the other vcores described in *cip.
2178 */
2179static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2180 int target_threads)
2181{
2182 int sub;
2183
2184 if (cip->total_threads + pvc->num_threads > target_threads)
2185 return false;
2186 for (sub = 0; sub < cip->n_subcores; ++sub)
2187 if (cip->subcore_threads[sub] &&
2188 can_piggyback_subcore(pvc, cip, sub))
2189 return true;
2190
2191 if (can_dynamic_split(pvc, cip))
2192 return true;
2193
2194 return false;
2195}
2196
d911f0be
PM		 2197static void prepare_threads(struct kvmppc_vcore *vc)
2198{
2199 struct kvm_vcpu *vcpu, *vnext;
2200
2201 list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2202 arch.run_list) {
2203 if (signal_pending(vcpu->arch.run_task))
2204 vcpu->arch.ret = -EINTR;
2205 else if (vcpu->arch.vpa.update_pending ||
2206 vcpu->arch.slb_shadow.update_pending ||
2207 vcpu->arch.dtl.update_pending)
2208 vcpu->arch.ret = RESUME_GUEST;
2209 else
2210 continue;
2211 kvmppc_remove_runnable(vc, vcpu);
2212 wake_up(&vcpu->arch.cpu_run);
2213 }
2214}
2215
ec257165
PM		 2216static void collect_piggybacks(struct core_info *cip, int target_threads)
2217{
2218 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2219 struct kvmppc_vcore *pvc, *vcnext;
2220
2221 spin_lock(&lp->lock);
2222 list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2223 if (!spin_trylock(&pvc->lock))
2224 continue;
2225 prepare_threads(pvc);
2226 if (!pvc->n_runnable) {
2227 list_del_init(&pvc->preempt_list);
2228 if (pvc->runner == NULL) {
2229 pvc->vcore_state = VCORE_INACTIVE;
2230 kvmppc_core_end_stolen(pvc);
2231 }
2232 spin_unlock(&pvc->lock);
2233 continue;
2234 }
2235 if (!can_piggyback(pvc, cip, target_threads)) {
2236 spin_unlock(&pvc->lock);
2237 continue;
2238 }
2239 kvmppc_core_end_stolen(pvc);
2240 pvc->vcore_state = VCORE_PIGGYBACK;
2241 if (cip->total_threads >= target_threads)
2242 break;
2243 }
2244 spin_unlock(&lp->lock);
2245}
2246
2247static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
25fedfca 2248{
ec257165 2249 int still_running = 0;
25fedfca
PM		 2250 u64 now;
2251 long ret;
2252 struct kvm_vcpu *vcpu, *vnext;
2253
ec257165 2254 spin_lock(&vc->lock);
25fedfca
PM		 2255 now = get_tb();
2256 list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2257 arch.run_list) {
2258 /* cancel pending dec exception if dec is positive */
2259 if (now < vcpu->arch.dec_expires &&
2260 kvmppc_core_pending_dec(vcpu))
2261 kvmppc_core_dequeue_dec(vcpu);
2262
2263 trace_kvm_guest_exit(vcpu);
2264
2265 ret = RESUME_GUEST;
2266 if (vcpu->arch.trap)
2267 ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2268 vcpu->arch.run_task);
2269
2270 vcpu->arch.ret = ret;
2271 vcpu->arch.trap = 0;
2272
ec257165
PM		 2273 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2274 if (vcpu->arch.pending_exceptions)
2275 kvmppc_core_prepare_to_enter(vcpu);
2276 if (vcpu->arch.ceded)
25fedfca 2277 kvmppc_set_timer(vcpu);
ec257165
PM		 2278 else
2279 ++still_running;
2280 } else {
25fedfca
PM		 2281 kvmppc_remove_runnable(vc, vcpu);
2282 wake_up(&vcpu->arch.cpu_run);
2283 }
2284 }
ec257165
PM		 2285 list_del_init(&vc->preempt_list);
2286 if (!is_master) {
563a1e93 2287 if (still_running > 0) {
ec257165 2288 kvmppc_vcore_preempt(vc);
563a1e93
PM		 2289 } else if (vc->runner) {
2290 vc->vcore_state = VCORE_PREEMPT;
2291 kvmppc_core_start_stolen(vc);
2292 } else {
2293 vc->vcore_state = VCORE_INACTIVE;
2294 }
ec257165
PM		 2295 if (vc->n_runnable > 0 && vc->runner == NULL) {
2296 /* make sure there's a candidate runner awake */
2297 vcpu = list_first_entry(&vc->runnable_threads,
2298 struct kvm_vcpu, arch.run_list);
2299 wake_up(&vcpu->arch.cpu_run);
2300 }
2301 }
2302 spin_unlock(&vc->lock);
25fedfca
PM		 2303}
2304
b8e6a87c
SW		 2305/*
2306 * Clear core from the list of active host cores as we are about to
2307 * enter the guest. Only do this if it is the primary thread of the
2308 * core (not if a subcore) that is entering the guest.
2309 */
2310static inline void kvmppc_clear_host_core(int cpu)
2311{
2312 int core;
2313
2314 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2315 return;
2316 /*
2317 * Memory barrier can be omitted here as we will do a smp_wmb()
2318 * later in kvmppc_start_thread and we need ensure that state is
2319 * visible to other CPUs only after we enter guest.
2320 */
2321 core = cpu >> threads_shift;
2322 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2323}
2324
2325/*
2326 * Advertise this core as an active host core since we exited the guest
2327 * Only need to do this if it is the primary thread of the core that is
2328 * exiting.
2329 */
2330static inline void kvmppc_set_host_core(int cpu)
2331{
2332 int core;
2333
2334 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2335 return;
2336
2337 /*
2338 * Memory barrier can be omitted here because we do a spin_unlock
2339 * immediately after this which provides the memory barrier.
2340 */
2341 core = cpu >> threads_shift;
2342 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2343}
2344
371fefd6
PM		 2345/*
2346 * Run a set of guest threads on a physical core.
2347 * Called with vc->lock held.
2348 */
66feed61 2349static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
371fefd6 2350{
17d48901 2351 struct kvm_vcpu *vcpu, *vnext;
d911f0be 2352 int i;
2c9097e4 2353 int srcu_idx;
ec257165
PM		 2354 struct core_info core_info;
2355 struct kvmppc_vcore *pvc, *vcnext;
b4deba5c
PM		 2356 struct kvm_split_mode split_info, *sip;
2357 int split, subcore_size, active;
2358 int sub;
2359 bool thr0_done;
2360 unsigned long cmd_bit, stat_bit;
ec257165
PM		 2361 int pcpu, thr;
2362 int target_threads;
371fefd6 2363
d911f0be
PM		 2364 /*
2365 * Remove from the list any threads that have a signal pending
2366 * or need a VPA update done
2367 */
2368 prepare_threads(vc);
2369
2370 /* if the runner is no longer runnable, let the caller pick a new one */
2371 if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2372 return;
081f323b
PM		 2373
2374 /*
d911f0be 2375 * Initialize *vc.
081f323b 2376 */
ec257165 2377 init_master_vcore(vc);
2711e248 2378 vc->preempt_tb = TB_NIL;
081f323b 2379
7b444c67 2380 /*
3102f784
ME		 2381 * Make sure we are running on primary threads, and that secondary
2382 * threads are offline. Also check if the number of threads in this
2383 * guest are greater than the current system threads per guest.
7b444c67 2384 */
3102f784
ME		 2385 if ((threads_per_core > 1) &&
2386 ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
17d48901
PM		 2387 list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2388 arch.run_list) {
7b444c67 2389 vcpu->arch.ret = -EBUSY;
25fedfca
PM		 2390 kvmppc_remove_runnable(vc, vcpu);
2391 wake_up(&vcpu->arch.cpu_run);
2392 }
7b444c67
PM		 2393 goto out;
2394 }
2395
ec257165
PM		 2396 /*
2397 * See if we could run any other vcores on the physical core
2398 * along with this one.
2399 */
2400 init_core_info(&core_info, vc);
2401 pcpu = smp_processor_id();
2402 target_threads = threads_per_subcore;
2403 if (target_smt_mode && target_smt_mode < target_threads)
2404 target_threads = target_smt_mode;
2405 if (vc->num_threads < target_threads)
2406 collect_piggybacks(&core_info, target_threads);
3102f784 2407
b4deba5c
PM		 2408 /* Decide on micro-threading (split-core) mode */
2409 subcore_size = threads_per_subcore;
2410 cmd_bit = stat_bit = 0;
2411 split = core_info.n_subcores;
2412 sip = NULL;
2413 if (split > 1) {
2414 /* threads_per_subcore must be MAX_SMT_THREADS (8) here */
2415 if (split == 2 && (dynamic_mt_modes & 2)) {
2416 cmd_bit = HID0_POWER8_1TO2LPAR;
2417 stat_bit = HID0_POWER8_2LPARMODE;
2418 } else {
2419 split = 4;
2420 cmd_bit = HID0_POWER8_1TO4LPAR;
2421 stat_bit = HID0_POWER8_4LPARMODE;
2422 }
2423 subcore_size = MAX_SMT_THREADS / split;
2424 sip = &split_info;
2425 memset(&split_info, 0, sizeof(split_info));
2426 split_info.rpr = mfspr(SPRN_RPR);
2427 split_info.pmmar = mfspr(SPRN_PMMAR);
2428 split_info.ldbar = mfspr(SPRN_LDBAR);
2429 split_info.subcore_size = subcore_size;
2430 for (sub = 0; sub < core_info.n_subcores; ++sub)
2431 split_info.master_vcs[sub] =
2432 list_first_entry(&core_info.vcs[sub],
2433 struct kvmppc_vcore, preempt_list);
2434 /* order writes to split_info before kvm_split_mode pointer */
2435 smp_wmb();
2436 }
2437 pcpu = smp_processor_id();
2438 for (thr = 0; thr < threads_per_subcore; ++thr)
2439 paca[pcpu + thr].kvm_hstate.kvm_split_mode = sip;
2440
2441 /* Initiate micro-threading (split-core) if required */
2442 if (cmd_bit) {
2443 unsigned long hid0 = mfspr(SPRN_HID0);
2444
2445 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2446 mb();
2447 mtspr(SPRN_HID0, hid0);
2448 isync();
2449 for (;;) {
2450 hid0 = mfspr(SPRN_HID0);
2451 if (hid0 & stat_bit)
2452 break;
2453 cpu_relax();
ec257165 2454 }
2e25aa5f 2455 }
3102f784 2456
b8e6a87c
SW		 2457 kvmppc_clear_host_core(pcpu);
2458
b4deba5c
PM		 2459 /* Start all the threads */
2460 active = 0;
2461 for (sub = 0; sub < core_info.n_subcores; ++sub) {
2462 thr = subcore_thread_map[sub];
2463 thr0_done = false;
2464 active |= 1 << thr;
2465 list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list) {
2466 pvc->pcpu = pcpu + thr;
2467 list_for_each_entry(vcpu, &pvc->runnable_threads,
2468 arch.run_list) {
2469 kvmppc_start_thread(vcpu, pvc);
2470 kvmppc_create_dtl_entry(vcpu, pvc);
2471 trace_kvm_guest_enter(vcpu);
2472 if (!vcpu->arch.ptid)
2473 thr0_done = true;
2474 active |= 1 << (thr + vcpu->arch.ptid);
2475 }
2476 /*
2477 * We need to start the first thread of each subcore
2478 * even if it doesn't have a vcpu.
2479 */
2480 if (pvc->master_vcore == pvc && !thr0_done)
2481 kvmppc_start_thread(NULL, pvc);
2482 thr += pvc->num_threads;
2483 }
2e25aa5f 2484 }
371fefd6 2485
7f235328
GS		 2486 /*
2487 * Ensure that split_info.do_nap is set after setting
2488 * the vcore pointer in the PACA of the secondaries.
2489 */
2490 smp_mb();
2491 if (cmd_bit)
2492 split_info.do_nap = 1; /* ask secondaries to nap when done */
2493
b4deba5c
PM		 2494 /*
2495 * When doing micro-threading, poke the inactive threads as well.
2496 * This gets them to the nap instruction after kvm_do_nap,
2497 * which reduces the time taken to unsplit later.
2498 */
2499 if (split > 1)
2500 for (thr = 1; thr < threads_per_subcore; ++thr)
2501 if (!(active & (1 << thr)))
2502 kvmppc_ipi_thread(pcpu + thr);
e0b7ec05 2503
2f12f034 2504 vc->vcore_state = VCORE_RUNNING;
19ccb76a 2505 preempt_disable();
3c78f78a
SW		 2506
2507 trace_kvmppc_run_core(vc, 0);
2508
b4deba5c
PM		 2509 for (sub = 0; sub < core_info.n_subcores; ++sub)
2510 list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list)
2511 spin_unlock(&pvc->lock);
de56a948 2512
371fefd6 2513 kvm_guest_enter();
2c9097e4 2514
e0b7ec05 2515 srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2c9097e4 2516
e0b7ec05 2517 __kvmppc_vcore_entry();
de56a948 2518
ec257165
PM		 2519 srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2520
2521 spin_lock(&vc->lock);
371fefd6 2522 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
19ccb76a 2523 vc->vcore_state = VCORE_EXITING;
371fefd6 2524
19ccb76a 2525 /* wait for secondary threads to finish writing their state to memory */
5d5b99cd 2526 kvmppc_wait_for_nap();
b4deba5c
PM		 2527
2528 /* Return to whole-core mode if we split the core earlier */
2529 if (split > 1) {
2530 unsigned long hid0 = mfspr(SPRN_HID0);
2531 unsigned long loops = 0;
2532
2533 hid0 &= ~HID0_POWER8_DYNLPARDIS;
2534 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
2535 mb();
2536 mtspr(SPRN_HID0, hid0);
2537 isync();
2538 for (;;) {
2539 hid0 = mfspr(SPRN_HID0);
2540 if (!(hid0 & stat_bit))
2541 break;
2542 cpu_relax();
2543 ++loops;
2544 }
2545 split_info.do_nap = 0;
2546 }
2547
2548 /* Let secondaries go back to the offline loop */
2549 for (i = 0; i < threads_per_subcore; ++i) {
2550 kvmppc_release_hwthread(pcpu + i);
2551 if (sip && sip->napped[i])
2552 kvmppc_ipi_thread(pcpu + i);
2553 }
2554
b8e6a87c
SW		 2555 kvmppc_set_host_core(pcpu);
2556
371fefd6 2557 spin_unlock(&vc->lock);
2c9097e4 2558
371fefd6
PM		 2559 /* make sure updates to secondary vcpu structs are visible now */
2560 smp_mb();
de56a948
PM		 2561 kvm_guest_exit();
2562
b4deba5c
PM		 2563 for (sub = 0; sub < core_info.n_subcores; ++sub)
2564 list_for_each_entry_safe(pvc, vcnext, &core_info.vcs[sub],
2565 preempt_list)
2566 post_guest_process(pvc, pvc == vc);
de56a948 2567
913d3ff9 2568 spin_lock(&vc->lock);
ec257165 2569 preempt_enable();
de56a948
PM		 2570
2571 out:
19ccb76a 2572 vc->vcore_state = VCORE_INACTIVE;
3c78f78a 2573 trace_kvmppc_run_core(vc, 1);
371fefd6
PM		 2574}
2575
19ccb76a
PM		 2576/*
2577 * Wait for some other vcpu thread to execute us, and
2578 * wake us up when we need to handle something in the host.
2579 */
ec257165
PM		 2580static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
2581 struct kvm_vcpu *vcpu, int wait_state)
371fefd6 2582{
371fefd6
PM		 2583 DEFINE_WAIT(wait);
2584
19ccb76a 2585 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
ec257165
PM		 2586 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2587 spin_unlock(&vc->lock);
19ccb76a 2588 schedule();
ec257165
PM		 2589 spin_lock(&vc->lock);
2590 }
19ccb76a
PM		 2591 finish_wait(&vcpu->arch.cpu_run, &wait);
2592}
2593
2594/*
2595 * All the vcpus in this vcore are idle, so wait for a decrementer
2596 * or external interrupt to one of the vcpus. vc->lock is held.
2597 */
2598static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
2599{
1bc5d59c
SW		 2600 struct kvm_vcpu *vcpu;
2601 int do_sleep = 1;
2602
19ccb76a 2603 DEFINE_WAIT(wait);
19ccb76a
PM		 2604
2605 prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
1bc5d59c
SW		 2606
2607 /*
2608 * Check one last time for pending exceptions and ceded state after
2609 * we put ourselves on the wait queue
2610 */
2611 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
2612 if (vcpu->arch.pending_exceptions || !vcpu->arch.ceded) {
2613 do_sleep = 0;
2614 break;
2615 }
2616 }
2617
2618 if (!do_sleep) {
2619 finish_wait(&vc->wq, &wait);
2620 return;
2621 }
2622
19ccb76a 2623 vc->vcore_state = VCORE_SLEEPING;
3c78f78a 2624 trace_kvmppc_vcore_blocked(vc, 0);
19ccb76a 2625 spin_unlock(&vc->lock);
913d3ff9 2626 schedule();
19ccb76a
PM		 2627 finish_wait(&vc->wq, &wait);
2628 spin_lock(&vc->lock);
2629 vc->vcore_state = VCORE_INACTIVE;
3c78f78a 2630 trace_kvmppc_vcore_blocked(vc, 1);
19ccb76a 2631}
371fefd6 2632
19ccb76a
PM		 2633static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2634{
2635 int n_ceded;
19ccb76a
PM		 2636 struct kvmppc_vcore *vc;
2637 struct kvm_vcpu *v, *vn;
9e368f29 2638
3c78f78a
SW		 2639 trace_kvmppc_run_vcpu_enter(vcpu);
2640
371fefd6
PM		 2641 kvm_run->exit_reason = 0;
2642 vcpu->arch.ret = RESUME_GUEST;
2643 vcpu->arch.trap = 0;
2f12f034 2644 kvmppc_update_vpas(vcpu);
371fefd6 2645
371fefd6
PM		 2646 /*
2647 * Synchronize with other threads in this virtual core
2648 */
2649 vc = vcpu->arch.vcore;
2650 spin_lock(&vc->lock);
19ccb76a 2651 vcpu->arch.ceded = 0;
371fefd6
PM		 2652 vcpu->arch.run_task = current;
2653 vcpu->arch.kvm_run = kvm_run;
c7b67670 2654 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
19ccb76a 2655 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
c7b67670 2656 vcpu->arch.busy_preempt = TB_NIL;
371fefd6
PM		 2657 list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
2658 ++vc->n_runnable;
2659
19ccb76a
PM		 2660 /*
2661 * This happens the first time this is called for a vcpu.
2662 * If the vcore is already running, we may be able to start
2663 * this thread straight away and have it join in.
2664 */
8455d79e 2665 if (!signal_pending(current)) {
ec257165
PM		 2666 if (vc->vcore_state == VCORE_PIGGYBACK) {
2667 struct kvmppc_vcore *mvc = vc->master_vcore;
2668 if (spin_trylock(&mvc->lock)) {
2669 if (mvc->vcore_state == VCORE_RUNNING &&
2670 !VCORE_IS_EXITING(mvc)) {
2671 kvmppc_create_dtl_entry(vcpu, vc);
b4deba5c 2672 kvmppc_start_thread(vcpu, vc);
ec257165
PM		 2673 trace_kvm_guest_enter(vcpu);
2674 }
2675 spin_unlock(&mvc->lock);
2676 }
2677 } else if (vc->vcore_state == VCORE_RUNNING &&
2678 !VCORE_IS_EXITING(vc)) {
2f12f034 2679 kvmppc_create_dtl_entry(vcpu, vc);
b4deba5c 2680 kvmppc_start_thread(vcpu, vc);
3c78f78a 2681 trace_kvm_guest_enter(vcpu);
8455d79e
PM		 2682 } else if (vc->vcore_state == VCORE_SLEEPING) {
2683 wake_up(&vc->wq);
371fefd6
PM		 2684 }
2685
8455d79e 2686 }
371fefd6 2687
19ccb76a
PM		 2688 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2689 !signal_pending(current)) {
ec257165
PM		 2690 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2691 kvmppc_vcore_end_preempt(vc);
2692
8455d79e 2693 if (vc->vcore_state != VCORE_INACTIVE) {
ec257165 2694 kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
19ccb76a
PM		 2695 continue;
2696 }
19ccb76a
PM		 2697 list_for_each_entry_safe(v, vn, &vc->runnable_threads,
2698 arch.run_list) {
7e28e60e 2699 kvmppc_core_prepare_to_enter(v);
19ccb76a
PM		 2700 if (signal_pending(v->arch.run_task)) {
2701 kvmppc_remove_runnable(vc, v);
2702 v->stat.signal_exits++;
2703 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
2704 v->arch.ret = -EINTR;
2705 wake_up(&v->arch.cpu_run);
2706 }
2707 }
8455d79e
PM		 2708 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2709 break;
8455d79e 2710 n_ceded = 0;
4619ac88 2711 list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
8455d79e
PM		 2712 if (!v->arch.pending_exceptions)
2713 n_ceded += v->arch.ceded;
4619ac88
PM		 2714 else
2715 v->arch.ceded = 0;
2716 }
25fedfca
PM		 2717 vc->runner = vcpu;
2718 if (n_ceded == vc->n_runnable) {
8455d79e 2719 kvmppc_vcore_blocked(vc);
c56dadf3 2720 } else if (need_resched()) {
ec257165 2721 kvmppc_vcore_preempt(vc);
25fedfca
PM		 2722 /* Let something else run */
2723 cond_resched_lock(&vc->lock);
ec257165
PM		 2724 if (vc->vcore_state == VCORE_PREEMPT)
2725 kvmppc_vcore_end_preempt(vc);
25fedfca 2726 } else {
8455d79e 2727 kvmppc_run_core(vc);
25fedfca 2728 }
0456ec4f 2729 vc->runner = NULL;
19ccb76a 2730 }
371fefd6 2731
8455d79e
PM		 2732 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2733 (vc->vcore_state == VCORE_RUNNING ||
5fc3e64f
PM		 2734 vc->vcore_state == VCORE_EXITING ||
2735 vc->vcore_state == VCORE_PIGGYBACK))
ec257165 2736 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
8455d79e 2737
5fc3e64f
PM		 2738 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2739 kvmppc_vcore_end_preempt(vc);
2740
8455d79e
PM		 2741 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2742 kvmppc_remove_runnable(vc, vcpu);
2743 vcpu->stat.signal_exits++;
2744 kvm_run->exit_reason = KVM_EXIT_INTR;
2745 vcpu->arch.ret = -EINTR;
2746 }
2747
2748 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
2749 /* Wake up some vcpu to run the core */
2750 v = list_first_entry(&vc->runnable_threads,
2751 struct kvm_vcpu, arch.run_list);
2752 wake_up(&v->arch.cpu_run);
371fefd6
PM		 2753 }
2754
3c78f78a 2755 trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
371fefd6 2756 spin_unlock(&vc->lock);
371fefd6 2757 return vcpu->arch.ret;
de56a948
PM		 2758}
2759
3a167bea 2760static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
a8606e20
PM		 2761{
2762 int r;
913d3ff9 2763 int srcu_idx;
a8606e20 2764
af8f38b3
AG		 2765 if (!vcpu->arch.sane) {
2766 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2767 return -EINVAL;
2768 }
2769
25051b5a
SW		 2770 kvmppc_core_prepare_to_enter(vcpu);
2771
19ccb76a
PM		 2772 /* No need to go into the guest when all we'll do is come back out */
2773 if (signal_pending(current)) {
2774 run->exit_reason = KVM_EXIT_INTR;
2775 return -EINTR;
2776 }
2777
32fad281 2778 atomic_inc(&vcpu->kvm->arch.vcpus_running);
31037eca 2779 /* Order vcpus_running vs. hpte_setup_done, see kvmppc_alloc_reset_hpt */
32fad281
PM		 2780 smp_mb();
2781
c17b98cf 2782 /* On the first time here, set up HTAB and VRMA */
31037eca 2783 if (!vcpu->kvm->arch.hpte_setup_done) {
32fad281 2784 r = kvmppc_hv_setup_htab_rma(vcpu);
c77162de 2785 if (r)
32fad281 2786 goto out;
c77162de 2787 }
19ccb76a 2788
579e633e
AB		 2789 flush_all_to_thread(current);
2790
19ccb76a 2791 vcpu->arch.wqp = &vcpu->arch.vcore->wq;
342d3db7 2792 vcpu->arch.pgdir = current->mm->pgd;
c7b67670 2793 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
19ccb76a 2794
a8606e20
PM		 2795 do {
2796 r = kvmppc_run_vcpu(run, vcpu);
2797
2798 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
2799 !(vcpu->arch.shregs.msr & MSR_PR)) {
3c78f78a 2800 trace_kvm_hcall_enter(vcpu);
a8606e20 2801 r = kvmppc_pseries_do_hcall(vcpu);
3c78f78a 2802 trace_kvm_hcall_exit(vcpu, r);
7e28e60e 2803 kvmppc_core_prepare_to_enter(vcpu);
913d3ff9
PM		 2804 } else if (r == RESUME_PAGE_FAULT) {
2805 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2806 r = kvmppc_book3s_hv_page_fault(run, vcpu,
2807 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
2808 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
a8606e20 2809 }
e59d24e6 2810 } while (is_kvmppc_resume_guest(r));
32fad281
PM		 2811
2812 out:
c7b67670 2813 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
32fad281 2814 atomic_dec(&vcpu->kvm->arch.vcpus_running);
a8606e20
PM		 2815 return r;
2816}
2817
5b74716e
BH		 2818static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
2819 int linux_psize)
2820{
2821 struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
2822
2823 if (!def->shift)
2824 return;
2825 (*sps)->page_shift = def->shift;
2826 (*sps)->slb_enc = def->sllp;
2827 (*sps)->enc[0].page_shift = def->shift;
b1022fbd 2828 (*sps)->enc[0].pte_enc = def->penc[linux_psize];
1f365bb0
AK		 2829 /*
2830 * Add 16MB MPSS support if host supports it
2831 */
2832 if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
2833 (*sps)->enc[1].page_shift = 24;
2834 (*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
2835 }
5b74716e
BH		 2836 (*sps)++;
2837}
2838
3a167bea
AK		 2839static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
2840 struct kvm_ppc_smmu_info *info)
5b74716e
BH		 2841{
2842 struct kvm_ppc_one_seg_page_size *sps;
2843
2844 info->flags = KVM_PPC_PAGE_SIZES_REAL;
2845 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
2846 info->flags |= KVM_PPC_1T_SEGMENTS;
2847 info->slb_size = mmu_slb_size;
2848
2849 /* We only support these sizes for now, and no muti-size segments */
2850 sps = &info->sps[0];
2851 kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
2852 kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
2853 kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
2854
2855 return 0;
2856}
2857
82ed3616
PM		 2858/*
2859 * Get (and clear) the dirty memory log for a memory slot.
2860 */
3a167bea
AK		 2861static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
2862 struct kvm_dirty_log *log)
82ed3616 2863{
9f6b8029 2864 struct kvm_memslots *slots;
82ed3616
PM		 2865 struct kvm_memory_slot *memslot;
2866 int r;
2867 unsigned long n;
2868
2869 mutex_lock(&kvm->slots_lock);
2870
2871 r = -EINVAL;
bbacc0c1 2872 if (log->slot >= KVM_USER_MEM_SLOTS)
82ed3616
PM		 2873 goto out;
2874
9f6b8029
PB		 2875 slots = kvm_memslots(kvm);
2876 memslot = id_to_memslot(slots, log->slot);
82ed3616
PM		 2877 r = -ENOENT;
2878 if (!memslot->dirty_bitmap)
2879 goto out;
2880
2881 n = kvm_dirty_bitmap_bytes(memslot);
2882 memset(memslot->dirty_bitmap, 0, n);
2883
dfe49dbd 2884 r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
82ed3616
PM		 2885 if (r)
2886 goto out;
2887
2888 r = -EFAULT;
2889 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
2890 goto out;
2891
2892 r = 0;
2893out:
2894 mutex_unlock(&kvm->slots_lock);
2895 return r;
2896}
2897
3a167bea
AK		 2898static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
2899 struct kvm_memory_slot *dont)
a66b48c3
PM		 2900{
2901 if (!dont || free->arch.rmap != dont->arch.rmap) {
2902 vfree(free->arch.rmap);
2903 free->arch.rmap = NULL;
b2b2f165 2904 }
a66b48c3
PM		 2905}
2906
3a167bea
AK		 2907static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
2908 unsigned long npages)
a66b48c3
PM		 2909{
2910 slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
2911 if (!slot->arch.rmap)
2912 return -ENOMEM;
aa04b4cc 2913
c77162de
PM		 2914 return 0;
2915}
aa04b4cc 2916
3a167bea
AK		 2917static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
2918 struct kvm_memory_slot *memslot,
09170a49 2919 const struct kvm_userspace_memory_region *mem)
c77162de 2920{
a66b48c3 2921 return 0;
c77162de
PM		 2922}
2923
3a167bea 2924static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
09170a49 2925 const struct kvm_userspace_memory_region *mem,
f36f3f28
PB		 2926 const struct kvm_memory_slot *old,
2927 const struct kvm_memory_slot *new)
c77162de 2928{
dfe49dbd 2929 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
9f6b8029 2930 struct kvm_memslots *slots;
dfe49dbd
PM		 2931 struct kvm_memory_slot *memslot;
2932
8482644a 2933 if (npages && old->npages) {
dfe49dbd
PM		 2934 /*
2935 * If modifying a memslot, reset all the rmap dirty bits.
2936 * If this is a new memslot, we don't need to do anything
2937 * since the rmap array starts out as all zeroes,
2938 * i.e. no pages are dirty.
2939 */
9f6b8029
PB		 2940 slots = kvm_memslots(kvm);
2941 memslot = id_to_memslot(slots, mem->slot);
dfe49dbd
PM		 2942 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
2943 }
c77162de
PM		 2944}
2945
a0144e2a
PM		 2946/*
2947 * Update LPCR values in kvm->arch and in vcores.
2948 * Caller must hold kvm->lock.
2949 */
2950void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
2951{
2952 long int i;
2953 u32 cores_done = 0;
2954
2955 if ((kvm->arch.lpcr & mask) == lpcr)
2956 return;
2957
2958 kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
2959
2960 for (i = 0; i < KVM_MAX_VCORES; ++i) {
2961 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
2962 if (!vc)
2963 continue;
2964 spin_lock(&vc->lock);
2965 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
2966 spin_unlock(&vc->lock);
2967 if (++cores_done >= kvm->arch.online_vcores)
2968 break;
2969 }
2970}
2971
3a167bea
AK		 2972static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
2973{
2974 return;
2975}
2976
32fad281 2977static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
c77162de
PM		 2978{
2979 int err = 0;
2980 struct kvm *kvm = vcpu->kvm;
c77162de
PM		 2981 unsigned long hva;
2982 struct kvm_memory_slot *memslot;
2983 struct vm_area_struct *vma;
a0144e2a 2984 unsigned long lpcr = 0, senc;
c77162de 2985 unsigned long psize, porder;
2c9097e4 2986 int srcu_idx;
c77162de
PM		 2987
2988 mutex_lock(&kvm->lock);
31037eca 2989 if (kvm->arch.hpte_setup_done)
c77162de 2990 goto out; /* another vcpu beat us to it */
aa04b4cc 2991
32fad281
PM		 2992 /* Allocate hashed page table (if not done already) and reset it */
2993 if (!kvm->arch.hpt_virt) {
2994 err = kvmppc_alloc_hpt(kvm, NULL);
2995 if (err) {
2996 pr_err("KVM: Couldn't alloc HPT\n");
2997 goto out;
2998 }
2999 }
3000
c77162de 3001 /* Look up the memslot for guest physical address 0 */
2c9097e4 3002 srcu_idx = srcu_read_lock(&kvm->srcu);
c77162de 3003 memslot = gfn_to_memslot(kvm, 0);
aa04b4cc 3004
c77162de
PM		 3005 /* We must have some memory at 0 by now */
3006 err = -EINVAL;
3007 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
2c9097e4 3008 goto out_srcu;
c77162de
PM		 3009
3010 /* Look up the VMA for the start of this memory slot */
3011 hva = memslot->userspace_addr;
3012 down_read(&current->mm->mmap_sem);
3013 vma = find_vma(current->mm, hva);
3014 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
3015 goto up_out;
3016
3017 psize = vma_kernel_pagesize(vma);
da9d1d7f 3018 porder = __ilog2(psize);
c77162de 3019
c77162de
PM		 3020 up_read(&current->mm->mmap_sem);
3021
c17b98cf
PM		 3022 /* We can handle 4k, 64k or 16M pages in the VRMA */
3023 err = -EINVAL;
3024 if (!(psize == 0x1000 || psize == 0x10000 ||
3025 psize == 0x1000000))
3026 goto out_srcu;
c77162de 3027
c17b98cf
PM		 3028 /* Update VRMASD field in the LPCR */
3029 senc = slb_pgsize_encoding(psize);
3030 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
3031 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3032 /* the -4 is to account for senc values starting at 0x10 */
3033 lpcr = senc << (LPCR_VRMASD_SH - 4);
c77162de 3034
c17b98cf
PM		 3035 /* Create HPTEs in the hash page table for the VRMA */
3036 kvmppc_map_vrma(vcpu, memslot, porder);
aa04b4cc 3037
c17b98cf 3038 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
a0144e2a 3039
31037eca 3040 /* Order updates to kvm->arch.lpcr etc. vs. hpte_setup_done */
c77162de 3041 smp_wmb();
31037eca 3042 kvm->arch.hpte_setup_done = 1;
c77162de 3043 err = 0;
2c9097e4
PM		 3044 out_srcu:
3045 srcu_read_unlock(&kvm->srcu, srcu_idx);
c77162de
PM		 3046 out:
3047 mutex_unlock(&kvm->lock);
3048 return err;
b2b2f165 3049
c77162de
PM		 3050 up_out:
3051 up_read(&current->mm->mmap_sem);
505d6421 3052 goto out_srcu;
de56a948
PM		 3053}
3054
79b6c247 3055#ifdef CONFIG_KVM_XICS
6f3bb809
SW		 3056static int kvmppc_cpu_notify(struct notifier_block *self, unsigned long action,
3057 void *hcpu)
3058{
3059 unsigned long cpu = (long)hcpu;
3060
3061 switch (action) {
3062 case CPU_UP_PREPARE:
3063 case CPU_UP_PREPARE_FROZEN:
3064 kvmppc_set_host_core(cpu);
3065 break;
3066
3067#ifdef CONFIG_HOTPLUG_CPU
3068 case CPU_DEAD:
3069 case CPU_DEAD_FROZEN:
3070 case CPU_UP_CANCELED:
3071 case CPU_UP_CANCELED_FROZEN:
3072 kvmppc_clear_host_core(cpu);
3073 break;
3074#endif
3075 default:
3076 break;
3077 }
3078
3079 return NOTIFY_OK;
3080}
3081
3082static struct notifier_block kvmppc_cpu_notifier = {
3083 .notifier_call = kvmppc_cpu_notify,
3084};
3085
79b6c247
SW		 3086/*
3087 * Allocate a per-core structure for managing state about which cores are
3088 * running in the host versus the guest and for exchanging data between
3089 * real mode KVM and CPU running in the host.
3090 * This is only done for the first VM.
3091 * The allocated structure stays even if all VMs have stopped.
3092 * It is only freed when the kvm-hv module is unloaded.
3093 * It's OK for this routine to fail, we just don't support host
3094 * core operations like redirecting H_IPI wakeups.
3095 */
3096void kvmppc_alloc_host_rm_ops(void)
3097{
3098 struct kvmppc_host_rm_ops *ops;
3099 unsigned long l_ops;
3100 int cpu, core;
3101 int size;
3102
3103 /* Not the first time here ? */
3104 if (kvmppc_host_rm_ops_hv != NULL)
3105 return;
3106
3107 ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
3108 if (!ops)
3109 return;
3110
3111 size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
3112 ops->rm_core = kzalloc(size, GFP_KERNEL);
3113
3114 if (!ops->rm_core) {
3115 kfree(ops);
3116 return;
3117 }
3118
6f3bb809
SW		 3119 get_online_cpus();
3120
79b6c247
SW		 3121 for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
3122 if (!cpu_online(cpu))
3123 continue;
3124
3125 core = cpu >> threads_shift;
3126 ops->rm_core[core].rm_state.in_host = 1;
3127 }
3128
0c2a6606
SW		 3129 ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
3130
79b6c247
SW		 3131 /*
3132 * Make the contents of the kvmppc_host_rm_ops structure visible
3133 * to other CPUs before we assign it to the global variable.
3134 * Do an atomic assignment (no locks used here), but if someone
3135 * beats us to it, just free our copy and return.
3136 */
3137 smp_wmb();
3138 l_ops = (unsigned long) ops;
3139
3140 if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
6f3bb809 3141 put_online_cpus();
79b6c247
SW		 3142 kfree(ops->rm_core);
3143 kfree(ops);
6f3bb809 3144 return;
79b6c247 3145 }
6f3bb809
SW		 3146
3147 register_cpu_notifier(&kvmppc_cpu_notifier);
3148
3149 put_online_cpus();
79b6c247
SW		 3150}
3151
3152void kvmppc_free_host_rm_ops(void)
3153{
3154 if (kvmppc_host_rm_ops_hv) {
6f3bb809 3155 unregister_cpu_notifier(&kvmppc_cpu_notifier);
79b6c247
SW		 3156 kfree(kvmppc_host_rm_ops_hv->rm_core);
3157 kfree(kvmppc_host_rm_ops_hv);
3158 kvmppc_host_rm_ops_hv = NULL;
3159 }
3160}
3161#endif
3162
3a167bea 3163static int kvmppc_core_init_vm_hv(struct kvm *kvm)
de56a948 3164{
32fad281 3165 unsigned long lpcr, lpid;
e23a808b 3166 char buf[32];
de56a948 3167
32fad281
PM		 3168 /* Allocate the guest's logical partition ID */
3169
3170 lpid = kvmppc_alloc_lpid();
5d226ae5 3171 if ((long)lpid < 0)
32fad281
PM		 3172 return -ENOMEM;
3173 kvm->arch.lpid = lpid;
de56a948 3174
79b6c247
SW		 3175 kvmppc_alloc_host_rm_ops();
3176
1b400ba0
PM		 3177 /*
3178 * Since we don't flush the TLB when tearing down a VM,
3179 * and this lpid might have previously been used,
3180 * make sure we flush on each core before running the new VM.
3181 */
3182 cpumask_setall(&kvm->arch.need_tlb_flush);
3183
699a0ea0
PM		 3184 /* Start out with the default set of hcalls enabled */
3185 memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
3186 sizeof(kvm->arch.enabled_hcalls));
3187
9e368f29 3188 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
aa04b4cc 3189
c17b98cf
PM		 3190 /* Init LPCR for virtual RMA mode */
3191 kvm->arch.host_lpid = mfspr(SPRN_LPID);
3192 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
3193 lpcr &= LPCR_PECE | LPCR_LPES;
3194 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3195 LPCR_VPM0 | LPCR_VPM1;
3196 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3197 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3198 /* On POWER8 turn on online bit to enable PURR/SPURR */
3199 if (cpu_has_feature(CPU_FTR_ARCH_207S))
3200 lpcr |= LPCR_ONL;
9e368f29 3201 kvm->arch.lpcr = lpcr;
aa04b4cc 3202
512691d4 3203 /*
441c19c8
ME		 3204 * Track that we now have a HV mode VM active. This blocks secondary
3205 * CPU threads from coming online.
512691d4 3206 */
441c19c8 3207 kvm_hv_vm_activated();
512691d4 3208
e23a808b
PM		 3209 /*
3210 * Create a debugfs directory for the VM
3211 */
3212 snprintf(buf, sizeof(buf), "vm%d", current->pid);
3213 kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
3214 if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
3215 kvmppc_mmu_debugfs_init(kvm);
3216
54738c09 3217 return 0;
de56a948
PM		 3218}
3219
f1378b1c
PM		 3220static void kvmppc_free_vcores(struct kvm *kvm)
3221{
3222 long int i;
3223
23316316 3224 for (i = 0; i < KVM_MAX_VCORES; ++i)
f1378b1c
PM		 3225 kfree(kvm->arch.vcores[i]);
3226 kvm->arch.online_vcores = 0;
3227}
3228
3a167bea 3229static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
de56a948 3230{
e23a808b
PM		 3231 debugfs_remove_recursive(kvm->arch.debugfs_dir);
3232
441c19c8 3233 kvm_hv_vm_deactivated();
512691d4 3234
f1378b1c 3235 kvmppc_free_vcores(kvm);
aa04b4cc 3236
de56a948
PM		 3237 kvmppc_free_hpt(kvm);
3238}
3239
3a167bea
AK		 3240/* We don't need to emulate any privileged instructions or dcbz */
3241static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
3242 unsigned int inst, int *advance)
de56a948 3243{
3a167bea 3244 return EMULATE_FAIL;
de56a948
PM		 3245}
3246
3a167bea
AK		 3247static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
3248 ulong spr_val)
de56a948
PM		 3249{
3250 return EMULATE_FAIL;
3251}
3252
3a167bea
AK		 3253static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
3254 ulong *spr_val)
de56a948
PM		 3255{
3256 return EMULATE_FAIL;
3257}
3258
3a167bea 3259static int kvmppc_core_check_processor_compat_hv(void)
de56a948 3260{
c17b98cf
PM		 3261 if (!cpu_has_feature(CPU_FTR_HVMODE) ||
3262 !cpu_has_feature(CPU_FTR_ARCH_206))
3a167bea
AK		 3263 return -EIO;
3264 return 0;
de56a948
PM		 3265}
3266
3a167bea
AK		 3267static long kvm_arch_vm_ioctl_hv(struct file *filp,
3268 unsigned int ioctl, unsigned long arg)
3269{
3270 struct kvm *kvm __maybe_unused = filp->private_data;
3271 void __user *argp = (void __user *)arg;
3272 long r;
3273
3274 switch (ioctl) {
3275
3a167bea
AK		 3276 case KVM_PPC_ALLOCATE_HTAB: {
3277 u32 htab_order;
3278
3279 r = -EFAULT;
3280 if (get_user(htab_order, (u32 __user *)argp))
3281 break;
3282 r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
3283 if (r)
3284 break;
3285 r = -EFAULT;
3286 if (put_user(htab_order, (u32 __user *)argp))
3287 break;
3288 r = 0;
3289 break;
3290 }
3291
3292 case KVM_PPC_GET_HTAB_FD: {
3293 struct kvm_get_htab_fd ghf;
3294
3295 r = -EFAULT;
3296 if (copy_from_user(&ghf, argp, sizeof(ghf)))
3297 break;
3298 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
3299 break;
3300 }
3301
3302 default:
3303 r = -ENOTTY;
3304 }
3305
3306 return r;
3307}
3308
699a0ea0
PM		 3309/*
3310 * List of hcall numbers to enable by default.
3311 * For compatibility with old userspace, we enable by default
3312 * all hcalls that were implemented before the hcall-enabling
3313 * facility was added. Note this list should not include H_RTAS.
3314 */
3315static unsigned int default_hcall_list[] = {
3316 H_REMOVE,
3317 H_ENTER,
3318 H_READ,
3319 H_PROTECT,
3320 H_BULK_REMOVE,
3321 H_GET_TCE,
3322 H_PUT_TCE,
3323 H_SET_DABR,
3324 H_SET_XDABR,
3325 H_CEDE,
3326 H_PROD,
3327 H_CONFER,
3328 H_REGISTER_VPA,
3329#ifdef CONFIG_KVM_XICS
3330 H_EOI,
3331 H_CPPR,
3332 H_IPI,
3333 H_IPOLL,
3334 H_XIRR,
3335 H_XIRR_X,
3336#endif
3337 0
3338};
3339
3340static void init_default_hcalls(void)
3341{
3342 int i;
ae2113a4 3343 unsigned int hcall;
699a0ea0 3344
ae2113a4
PM		 3345 for (i = 0; default_hcall_list[i]; ++i) {
3346 hcall = default_hcall_list[i];
3347 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
3348 __set_bit(hcall / 4, default_enabled_hcalls);
3349 }
699a0ea0
PM		 3350}
3351
cbbc58d4 3352static struct kvmppc_ops kvm_ops_hv = {
3a167bea
AK		 3353 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
3354 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
3355 .get_one_reg = kvmppc_get_one_reg_hv,
3356 .set_one_reg = kvmppc_set_one_reg_hv,
3357 .vcpu_load = kvmppc_core_vcpu_load_hv,
3358 .vcpu_put = kvmppc_core_vcpu_put_hv,
3359 .set_msr = kvmppc_set_msr_hv,
3360 .vcpu_run = kvmppc_vcpu_run_hv,
3361 .vcpu_create = kvmppc_core_vcpu_create_hv,
3362 .vcpu_free = kvmppc_core_vcpu_free_hv,
3363 .check_requests = kvmppc_core_check_requests_hv,
3364 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv,
3365 .flush_memslot = kvmppc_core_flush_memslot_hv,
3366 .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
3367 .commit_memory_region = kvmppc_core_commit_memory_region_hv,
3368 .unmap_hva = kvm_unmap_hva_hv,
3369 .unmap_hva_range = kvm_unmap_hva_range_hv,
3370 .age_hva = kvm_age_hva_hv,
3371 .test_age_hva = kvm_test_age_hva_hv,
3372 .set_spte_hva = kvm_set_spte_hva_hv,
3373 .mmu_destroy = kvmppc_mmu_destroy_hv,
3374 .free_memslot = kvmppc_core_free_memslot_hv,
3375 .create_memslot = kvmppc_core_create_memslot_hv,
3376 .init_vm = kvmppc_core_init_vm_hv,
3377 .destroy_vm = kvmppc_core_destroy_vm_hv,
3a167bea
AK		 3378 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
3379 .emulate_op = kvmppc_core_emulate_op_hv,
3380 .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
3381 .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
3382 .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
3383 .arch_vm_ioctl = kvm_arch_vm_ioctl_hv,
ae2113a4 3384 .hcall_implemented = kvmppc_hcall_impl_hv,
3a167bea
AK		 3385};
3386
3387static int kvmppc_book3s_init_hv(void)
de56a948
PM		 3388{
3389 int r;
cbbc58d4
AK		 3390 /*
3391 * FIXME!! Do we need to check on all cpus ?
3392 */
3393 r = kvmppc_core_check_processor_compat_hv();
3394 if (r < 0)
739e2425 3395 return -ENODEV;
de56a948 3396
cbbc58d4
AK		 3397 kvm_ops_hv.owner = THIS_MODULE;
3398 kvmppc_hv_ops = &kvm_ops_hv;
de56a948 3399
699a0ea0
PM		 3400 init_default_hcalls();
3401
ec257165
PM		 3402 init_vcore_lists();
3403
cbbc58d4 3404 r = kvmppc_mmu_hv_init();
de56a948
PM		 3405 return r;
3406}
3407
3a167bea 3408static void kvmppc_book3s_exit_hv(void)
de56a948 3409{
79b6c247 3410 kvmppc_free_host_rm_ops();
cbbc58d4 3411 kvmppc_hv_ops = NULL;
de56a948
PM		 3412}
3413
3a167bea
AK		 3414module_init(kvmppc_book3s_init_hv);
3415module_exit(kvmppc_book3s_exit_hv);
2ba9f0d8 3416MODULE_LICENSE("GPL");
398a76c6
AG		 3417MODULE_ALIAS_MISCDEV(KVM_MINOR);
3418MODULE_ALIAS("devname:kvm");
Empty description
This page took 0.795476 seconds and 4 git commands to generate.