]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - arch/powerpc/kvm/book3s_hv.c
projects / mirror_ubuntu-artful-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: PPC: Book3S HV: Enable hypervisor virtualization interrupts while in guest
[mirror_ubuntu-artful-kernel.git] / arch / powerpc / kvm / book3s_hv.c
CommitLineData
de56a948
PM		 1/*
2 * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
3 * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
4 *
5 * Authors:
6 * Paul Mackerras <paulus@au1.ibm.com>
7 * Alexander Graf <agraf@suse.de>
8 * Kevin Wolf <mail@kevin-wolf.de>
9 *
10 * Description: KVM functions specific to running on Book 3S
11 * processors in hypervisor mode (specifically POWER7 and later).
12 *
13 * This file is derived from arch/powerpc/kvm/book3s.c,
14 * by Alexander Graf <agraf@suse.de>.
15 *
16 * This program is free software; you can redistribute it and/or modify
17 * it under the terms of the GNU General Public License, version 2, as
18 * published by the Free Software Foundation.
19 */
20
21#include <linux/kvm_host.h>
22#include <linux/err.h>
23#include <linux/slab.h>
24#include <linux/preempt.h>
25#include <linux/sched.h>
26#include <linux/delay.h>
66b15db6 27#include <linux/export.h>
de56a948
PM		 28#include <linux/fs.h>
29#include <linux/anon_inodes.h>
07f8ab25 30#include <linux/cpu.h>
de56a948 31#include <linux/cpumask.h>
aa04b4cc
PM		 32#include <linux/spinlock.h>
33#include <linux/page-flags.h>
2c9097e4 34#include <linux/srcu.h>
398a76c6 35#include <linux/miscdevice.h>
e23a808b 36#include <linux/debugfs.h>
de56a948
PM		 37
38#include <asm/reg.h>
39#include <asm/cputable.h>
40#include <asm/cacheflush.h>
41#include <asm/tlbflush.h>
42#include <asm/uaccess.h>
43#include <asm/io.h>
44#include <asm/kvm_ppc.h>
45#include <asm/kvm_book3s.h>
46#include <asm/mmu_context.h>
47#include <asm/lppaca.h>
48#include <asm/processor.h>
371fefd6 49#include <asm/cputhreads.h>
aa04b4cc 50#include <asm/page.h>
de1d9248 51#include <asm/hvcall.h>
ae3a197e 52#include <asm/switch_to.h>
512691d4 53#include <asm/smp.h>
66feed61 54#include <asm/dbell.h>
fd7bacbc 55#include <asm/hmi.h>
c57875f5 56#include <asm/pnv-pci.h>
7a84084c 57#include <asm/mmu.h>
f725758b
PM		 58#include <asm/opal.h>
59#include <asm/xics.h>
de56a948 60#include <linux/gfp.h>
de56a948
PM		 61#include <linux/vmalloc.h>
62#include <linux/highmem.h>
c77162de 63#include <linux/hugetlb.h>
c57875f5
SW		 64#include <linux/kvm_irqfd.h>
65#include <linux/irqbypass.h>
2ba9f0d8 66#include <linux/module.h>
7b5f8272 67#include <linux/compiler.h>
f725758b 68#include <linux/of.h>
de56a948 69
3a167bea
AK		 70#include "book3s.h"
71
3c78f78a
SW		 72#define CREATE_TRACE_POINTS
73#include "trace_hv.h"
74
de56a948
PM		 75/* #define EXIT_DEBUG */
76/* #define EXIT_DEBUG_SIMPLE */
77/* #define EXIT_DEBUG_INT */
78
913d3ff9
PM		 79/* Used to indicate that a guest page fault needs to be handled */
80#define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1)
f7af5209
SW		 81/* Used to indicate that a guest passthrough interrupt needs to be handled */
82#define RESUME_PASSTHROUGH (RESUME_GUEST | RESUME_FLAG_ARCH2)
913d3ff9 83
c7b67670
PM		 84/* Used as a "null" value for timebase values */
85#define TB_NIL (~(u64)0)
86
699a0ea0
PM		 87static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
88
b4deba5c
PM		 89static int dynamic_mt_modes = 6;
90module_param(dynamic_mt_modes, int, S_IRUGO | S_IWUSR);
91MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
ec257165
PM		 92static int target_smt_mode;
93module_param(target_smt_mode, int, S_IRUGO | S_IWUSR);
94MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
9678cdaa 95
520fe9c6
SW		 96#ifdef CONFIG_KVM_XICS
97static struct kernel_param_ops module_param_ops = {
98 .set = param_set_int,
99 .get = param_get_int,
100};
101
644abbb2
SW		 102module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass,
103 S_IRUGO | S_IWUSR);
104MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
105
520fe9c6
SW		 106module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect,
107 S_IRUGO | S_IWUSR);
108MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
109#endif
110
0cda69dd
SJS		 111/* Maximum halt poll interval defaults to KVM_HALT_POLL_NS_DEFAULT */
112static unsigned int halt_poll_max_ns = KVM_HALT_POLL_NS_DEFAULT;
113module_param(halt_poll_max_ns, uint, S_IRUGO | S_IWUSR);
114MODULE_PARM_DESC(halt_poll_max_ns, "Maximum halt poll time in ns");
115
116/* Factor by which the vcore halt poll interval is grown, default is to double
117 */
118static unsigned int halt_poll_ns_grow = 2;
119module_param(halt_poll_ns_grow, int, S_IRUGO);
120MODULE_PARM_DESC(halt_poll_ns_grow, "Factor halt poll time is grown by");
121
122/* Factor by which the vcore halt poll interval is shrunk, default is to reset
123 */
124static unsigned int halt_poll_ns_shrink;
125module_param(halt_poll_ns_shrink, int, S_IRUGO);
126MODULE_PARM_DESC(halt_poll_ns_shrink, "Factor halt poll time is shrunk by");
127
19ccb76a 128static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
32fad281 129static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
19ccb76a 130
7b5f8272
SJS		 131static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
132 int *ip)
133{
134 int i = *ip;
135 struct kvm_vcpu *vcpu;
136
137 while (++i < MAX_SMT_THREADS) {
138 vcpu = READ_ONCE(vc->runnable_threads[i]);
139 if (vcpu) {
140 *ip = i;
141 return vcpu;
142 }
143 }
144 return NULL;
145}
146
147/* Used to traverse the list of runnable threads for a given vcore */
148#define for_each_runnable_thread(i, vcpu, vc) \
149 for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
150
66feed61
PM		 151static bool kvmppc_ipi_thread(int cpu)
152{
1704a81c
PM		 153 unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
154
155 /* On POWER9 we can use msgsnd to IPI any cpu */
156 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
157 msg |= get_hard_smp_processor_id(cpu);
158 smp_mb();
159 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
160 return true;
161 }
162
66feed61
PM		 163 /* On POWER8 for IPIs to threads in the same core, use msgsnd */
164 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
165 preempt_disable();
166 if (cpu_first_thread_sibling(cpu) ==
167 cpu_first_thread_sibling(smp_processor_id())) {
66feed61
PM		 168 msg |= cpu_thread_in_core(cpu);
169 smp_mb();
170 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
171 preempt_enable();
172 return true;
173 }
174 preempt_enable();
175 }
176
177#if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
f725758b
PM		 178 if (cpu >= 0 && cpu < nr_cpu_ids) {
179 if (paca[cpu].kvm_hstate.xics_phys) {
180 xics_wake_cpu(cpu);
181 return true;
182 }
183 opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
66feed61
PM		 184 return true;
185 }
186#endif
187
188 return false;
189}
190
3a167bea 191static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
54695c30 192{
ec257165 193 int cpu;
8577370f 194 struct swait_queue_head *wqp;
54695c30
BH		 195
196 wqp = kvm_arch_vcpu_wq(vcpu);
8577370f
MT		 197 if (swait_active(wqp)) {
198 swake_up(wqp);
54695c30
BH		 199 ++vcpu->stat.halt_wakeup;
200 }
201
ec257165 202 if (kvmppc_ipi_thread(vcpu->arch.thread_cpu))
66feed61 203 return;
54695c30
BH		 204
205 /* CPU points to the first thread of the core */
ec257165 206 cpu = vcpu->cpu;
66feed61
PM		 207 if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
208 smp_send_reschedule(cpu);
54695c30
BH		 209}
210
c7b67670
PM		 211/*
212 * We use the vcpu_load/put functions to measure stolen time.
213 * Stolen time is counted as time when either the vcpu is able to
214 * run as part of a virtual core, but the task running the vcore
215 * is preempted or sleeping, or when the vcpu needs something done
216 * in the kernel by the task running the vcpu, but that task is
217 * preempted or sleeping. Those two things have to be counted
218 * separately, since one of the vcpu tasks will take on the job
219 * of running the core, and the other vcpu tasks in the vcore will
220 * sleep waiting for it to do that, but that sleep shouldn't count
221 * as stolen time.
222 *
223 * Hence we accumulate stolen time when the vcpu can run as part of
224 * a vcore using vc->stolen_tb, and the stolen time when the vcpu
225 * needs its task to do other things in the kernel (for example,
226 * service a page fault) in busy_stolen. We don't accumulate
227 * stolen time for a vcore when it is inactive, or for a vcpu
228 * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of
229 * a misnomer; it means that the vcpu task is not executing in
230 * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
231 * the kernel. We don't have any way of dividing up that time
232 * between time that the vcpu is genuinely stopped, time that
233 * the task is actively working on behalf of the vcpu, and time
234 * that the task is preempted, so we don't count any of it as
235 * stolen.
236 *
237 * Updates to busy_stolen are protected by arch.tbacct_lock;
2711e248
PM		 238 * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
239 * lock. The stolen times are measured in units of timebase ticks.
240 * (Note that the != TB_NIL checks below are purely defensive;
241 * they should never fail.)
c7b67670
PM		 242 */
243
ec257165
PM		 244static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
245{
246 unsigned long flags;
247
248 spin_lock_irqsave(&vc->stoltb_lock, flags);
249 vc->preempt_tb = mftb();
250 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
251}
252
253static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
254{
255 unsigned long flags;
256
257 spin_lock_irqsave(&vc->stoltb_lock, flags);
258 if (vc->preempt_tb != TB_NIL) {
259 vc->stolen_tb += mftb() - vc->preempt_tb;
260 vc->preempt_tb = TB_NIL;
261 }
262 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
263}
264
3a167bea 265static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
de56a948 266{
0456ec4f 267 struct kvmppc_vcore *vc = vcpu->arch.vcore;
bf3d32e1 268 unsigned long flags;
0456ec4f 269
2711e248
PM		 270 /*
271 * We can test vc->runner without taking the vcore lock,
272 * because only this task ever sets vc->runner to this
273 * vcpu, and once it is set to this vcpu, only this task
274 * ever sets it to NULL.
275 */
ec257165
PM		 276 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
277 kvmppc_core_end_stolen(vc);
278
2711e248 279 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
c7b67670
PM		 280 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
281 vcpu->arch.busy_preempt != TB_NIL) {
282 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
283 vcpu->arch.busy_preempt = TB_NIL;
284 }
bf3d32e1 285 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
de56a948
PM		 286}
287
3a167bea 288static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
de56a948 289{
0456ec4f 290 struct kvmppc_vcore *vc = vcpu->arch.vcore;
bf3d32e1 291 unsigned long flags;
0456ec4f 292
ec257165
PM		 293 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
294 kvmppc_core_start_stolen(vc);
295
2711e248 296 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
c7b67670
PM		 297 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
298 vcpu->arch.busy_preempt = mftb();
bf3d32e1 299 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
de56a948
PM		 300}
301
3a167bea 302static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
de56a948 303{
c20875a3
PM		 304 /*
305 * Check for illegal transactional state bit combination
306 * and if we find it, force the TS field to a safe state.
307 */
308 if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
309 msr &= ~MSR_TS_MASK;
de56a948 310 vcpu->arch.shregs.msr = msr;
19ccb76a 311 kvmppc_end_cede(vcpu);
de56a948
PM		 312}
313
5358a963 314static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
de56a948
PM		 315{
316 vcpu->arch.pvr = pvr;
317}
318
5358a963 319static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
388cc6e1
PM		 320{
321 unsigned long pcr = 0;
322 struct kvmppc_vcore *vc = vcpu->arch.vcore;
323
324 if (arch_compat) {
388cc6e1
PM		 325 switch (arch_compat) {
326 case PVR_ARCH_205:
5557ae0e
PM		 327 /*
328 * If an arch bit is set in PCR, all the defined
329 * higher-order arch bits also have to be set.
330 */
331 pcr = PCR_ARCH_206 | PCR_ARCH_205;
388cc6e1
PM		 332 break;
333 case PVR_ARCH_206:
334 case PVR_ARCH_206p:
5557ae0e
PM		 335 pcr = PCR_ARCH_206;
336 break;
337 case PVR_ARCH_207:
388cc6e1
PM		 338 break;
339 default:
340 return -EINVAL;
341 }
5557ae0e
PM		 342
343 if (!cpu_has_feature(CPU_FTR_ARCH_207S)) {
344 /* POWER7 can't emulate POWER8 */
345 if (!(pcr & PCR_ARCH_206))
346 return -EINVAL;
347 pcr &= ~PCR_ARCH_206;
348 }
388cc6e1
PM		 349 }
350
351 spin_lock(&vc->lock);
352 vc->arch_compat = arch_compat;
353 vc->pcr = pcr;
354 spin_unlock(&vc->lock);
355
356 return 0;
357}
358
5358a963 359static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
de56a948
PM		 360{
361 int r;
362
363 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
364 pr_err("pc = %.16lx msr = %.16llx trap = %x\n",
365 vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
366 for (r = 0; r < 16; ++r)
367 pr_err("r%2d = %.16lx r%d = %.16lx\n",
368 r, kvmppc_get_gpr(vcpu, r),
369 r+16, kvmppc_get_gpr(vcpu, r+16));
370 pr_err("ctr = %.16lx lr = %.16lx\n",
371 vcpu->arch.ctr, vcpu->arch.lr);
372 pr_err("srr0 = %.16llx srr1 = %.16llx\n",
373 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
374 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
375 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
376 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
377 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
378 pr_err("cr = %.8x xer = %.16lx dsisr = %.8x\n",
379 vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
380 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
381 pr_err("fault dar = %.16lx dsisr = %.8x\n",
382 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
383 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
384 for (r = 0; r < vcpu->arch.slb_max; ++r)
385 pr_err(" ESID = %.16llx VSID = %.16llx\n",
386 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
387 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
a0144e2a 388 vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
de56a948
PM		 389 vcpu->arch.last_inst);
390}
391
5358a963 392static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
a8606e20 393{
e09fefde 394 struct kvm_vcpu *ret;
a8606e20
PM		 395
396 mutex_lock(&kvm->lock);
e09fefde 397 ret = kvm_get_vcpu_by_id(kvm, id);
a8606e20
PM		 398 mutex_unlock(&kvm->lock);
399 return ret;
400}
401
402static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
403{
f13c13a0 404 vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
02407552 405 vpa->yield_count = cpu_to_be32(1);
a8606e20
PM		 406}
407
55b665b0
PM		 408static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
409 unsigned long addr, unsigned long len)
410{
411 /* check address is cacheline aligned */
412 if (addr & (L1_CACHE_BYTES - 1))
413 return -EINVAL;
414 spin_lock(&vcpu->arch.vpa_update_lock);
415 if (v->next_gpa != addr || v->len != len) {
416 v->next_gpa = addr;
417 v->len = addr ? len : 0;
418 v->update_pending = 1;
419 }
420 spin_unlock(&vcpu->arch.vpa_update_lock);
421 return 0;
422}
423
2e25aa5f
PM		 424/* Length for a per-processor buffer is passed in at offset 4 in the buffer */
425struct reg_vpa {
426 u32 dummy;
427 union {
02407552
AG		 428 __be16 hword;
429 __be32 word;
2e25aa5f
PM		 430 } length;
431};
432
433static int vpa_is_registered(struct kvmppc_vpa *vpap)
434{
435 if (vpap->update_pending)
436 return vpap->next_gpa != 0;
437 return vpap->pinned_addr != NULL;
438}
439
a8606e20
PM		 440static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
441 unsigned long flags,
442 unsigned long vcpuid, unsigned long vpa)
443{
444 struct kvm *kvm = vcpu->kvm;
93e60249 445 unsigned long len, nb;
a8606e20
PM		 446 void *va;
447 struct kvm_vcpu *tvcpu;
2e25aa5f
PM		 448 int err;
449 int subfunc;
450 struct kvmppc_vpa *vpap;
a8606e20
PM		 451
452 tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
453 if (!tvcpu)
454 return H_PARAMETER;
455
2e25aa5f
PM		 456 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
457 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
458 subfunc == H_VPA_REG_SLB) {
459 /* Registering new area - address must be cache-line aligned */
460 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
a8606e20 461 return H_PARAMETER;
2e25aa5f
PM		 462
463 /* convert logical addr to kernel addr and read length */
93e60249
PM		 464 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
465 if (va == NULL)
b2b2f165 466 return H_PARAMETER;
2e25aa5f 467 if (subfunc == H_VPA_REG_VPA)
02407552 468 len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
a8606e20 469 else
02407552 470 len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
c35635ef 471 kvmppc_unpin_guest_page(kvm, va, vpa, false);
2e25aa5f
PM		 472
473 /* Check length */
474 if (len > nb || len < sizeof(struct reg_vpa))
475 return H_PARAMETER;
476 } else {
477 vpa = 0;
478 len = 0;
479 }
480
481 err = H_PARAMETER;
482 vpap = NULL;
483 spin_lock(&tvcpu->arch.vpa_update_lock);
484
485 switch (subfunc) {
486 case H_VPA_REG_VPA: /* register VPA */
487 if (len < sizeof(struct lppaca))
a8606e20 488 break;
2e25aa5f
PM		 489 vpap = &tvcpu->arch.vpa;
490 err = 0;
491 break;
492
493 case H_VPA_REG_DTL: /* register DTL */
494 if (len < sizeof(struct dtl_entry))
a8606e20 495 break;
2e25aa5f
PM		 496 len -= len % sizeof(struct dtl_entry);
497
498 /* Check that they have previously registered a VPA */
499 err = H_RESOURCE;
500 if (!vpa_is_registered(&tvcpu->arch.vpa))
a8606e20 501 break;
2e25aa5f
PM		 502
503 vpap = &tvcpu->arch.dtl;
504 err = 0;
505 break;
506
507 case H_VPA_REG_SLB: /* register SLB shadow buffer */
508 /* Check that they have previously registered a VPA */
509 err = H_RESOURCE;
510 if (!vpa_is_registered(&tvcpu->arch.vpa))
a8606e20 511 break;
2e25aa5f
PM		 512
513 vpap = &tvcpu->arch.slb_shadow;
514 err = 0;
515 break;
516
517 case H_VPA_DEREG_VPA: /* deregister VPA */
518 /* Check they don't still have a DTL or SLB buf registered */
519 err = H_RESOURCE;
520 if (vpa_is_registered(&tvcpu->arch.dtl) ||
521 vpa_is_registered(&tvcpu->arch.slb_shadow))
a8606e20 522 break;
2e25aa5f
PM		 523
524 vpap = &tvcpu->arch.vpa;
525 err = 0;
526 break;
527
528 case H_VPA_DEREG_DTL: /* deregister DTL */
529 vpap = &tvcpu->arch.dtl;
530 err = 0;
531 break;
532
533 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */
534 vpap = &tvcpu->arch.slb_shadow;
535 err = 0;
536 break;
537 }
538
539 if (vpap) {
540 vpap->next_gpa = vpa;
541 vpap->len = len;
542 vpap->update_pending = 1;
a8606e20 543 }
93e60249 544
2e25aa5f
PM		 545 spin_unlock(&tvcpu->arch.vpa_update_lock);
546
93e60249 547 return err;
a8606e20
PM		 548}
549
081f323b 550static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
2e25aa5f 551{
081f323b 552 struct kvm *kvm = vcpu->kvm;
2e25aa5f
PM		 553 void *va;
554 unsigned long nb;
081f323b 555 unsigned long gpa;
2e25aa5f 556
081f323b
PM		 557 /*
558 * We need to pin the page pointed to by vpap->next_gpa,
559 * but we can't call kvmppc_pin_guest_page under the lock
560 * as it does get_user_pages() and down_read(). So we
561 * have to drop the lock, pin the page, then get the lock
562 * again and check that a new area didn't get registered
563 * in the meantime.
564 */
565 for (;;) {
566 gpa = vpap->next_gpa;
567 spin_unlock(&vcpu->arch.vpa_update_lock);
568 va = NULL;
569 nb = 0;
570 if (gpa)
c35635ef 571 va = kvmppc_pin_guest_page(kvm, gpa, &nb);
081f323b
PM		 572 spin_lock(&vcpu->arch.vpa_update_lock);
573 if (gpa == vpap->next_gpa)
574 break;
575 /* sigh... unpin that one and try again */
576 if (va)
c35635ef 577 kvmppc_unpin_guest_page(kvm, va, gpa, false);
081f323b
PM		 578 }
579
580 vpap->update_pending = 0;
581 if (va && nb < vpap->len) {
582 /*
583 * If it's now too short, it must be that userspace
584 * has changed the mappings underlying guest memory,
585 * so unregister the region.
586 */
c35635ef 587 kvmppc_unpin_guest_page(kvm, va, gpa, false);
081f323b 588 va = NULL;
2e25aa5f
PM		 589 }
590 if (vpap->pinned_addr)
c35635ef
PM		 591 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
592 vpap->dirty);
593 vpap->gpa = gpa;
2e25aa5f 594 vpap->pinned_addr = va;
c35635ef 595 vpap->dirty = false;
2e25aa5f
PM		 596 if (va)
597 vpap->pinned_end = va + vpap->len;
598}
599
600static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
601{
2f12f034
PM		 602 if (!(vcpu->arch.vpa.update_pending ||
603 vcpu->arch.slb_shadow.update_pending ||
604 vcpu->arch.dtl.update_pending))
605 return;
606
2e25aa5f
PM		 607 spin_lock(&vcpu->arch.vpa_update_lock);
608 if (vcpu->arch.vpa.update_pending) {
081f323b 609 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
55b665b0
PM		 610 if (vcpu->arch.vpa.pinned_addr)
611 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
2e25aa5f
PM		 612 }
613 if (vcpu->arch.dtl.update_pending) {
081f323b 614 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
2e25aa5f
PM		 615 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
616 vcpu->arch.dtl_index = 0;
617 }
618 if (vcpu->arch.slb_shadow.update_pending)
081f323b 619 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
2e25aa5f
PM		 620 spin_unlock(&vcpu->arch.vpa_update_lock);
621}
622
c7b67670
PM		 623/*
624 * Return the accumulated stolen time for the vcore up until `now'.
625 * The caller should hold the vcore lock.
626 */
627static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
628{
629 u64 p;
2711e248 630 unsigned long flags;
c7b67670 631
2711e248
PM		 632 spin_lock_irqsave(&vc->stoltb_lock, flags);
633 p = vc->stolen_tb;
c7b67670 634 if (vc->vcore_state != VCORE_INACTIVE &&
2711e248
PM		 635 vc->preempt_tb != TB_NIL)
636 p += now - vc->preempt_tb;
637 spin_unlock_irqrestore(&vc->stoltb_lock, flags);
c7b67670
PM		 638 return p;
639}
640
0456ec4f
PM		 641static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
642 struct kvmppc_vcore *vc)
643{
644 struct dtl_entry *dt;
645 struct lppaca *vpa;
c7b67670
PM		 646 unsigned long stolen;
647 unsigned long core_stolen;
648 u64 now;
0456ec4f
PM		 649
650 dt = vcpu->arch.dtl_ptr;
651 vpa = vcpu->arch.vpa.pinned_addr;
c7b67670
PM		 652 now = mftb();
653 core_stolen = vcore_stolen_time(vc, now);
654 stolen = core_stolen - vcpu->arch.stolen_logged;
655 vcpu->arch.stolen_logged = core_stolen;
bf3d32e1 656 spin_lock_irq(&vcpu->arch.tbacct_lock);
c7b67670
PM		 657 stolen += vcpu->arch.busy_stolen;
658 vcpu->arch.busy_stolen = 0;
bf3d32e1 659 spin_unlock_irq(&vcpu->arch.tbacct_lock);
0456ec4f
PM		 660 if (!dt || !vpa)
661 return;
662 memset(dt, 0, sizeof(struct dtl_entry));
663 dt->dispatch_reason = 7;
02407552
AG		 664 dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
665 dt->timebase = cpu_to_be64(now + vc->tb_offset);
666 dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
667 dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
668 dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
0456ec4f
PM		 669 ++dt;
670 if (dt == vcpu->arch.dtl.pinned_end)
671 dt = vcpu->arch.dtl.pinned_addr;
672 vcpu->arch.dtl_ptr = dt;
673 /* order writing *dt vs. writing vpa->dtl_idx */
674 smp_wmb();
02407552 675 vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
c35635ef 676 vcpu->arch.dtl.dirty = true;
0456ec4f
PM		 677}
678
9642382e
MN		 679static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
680{
681 if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
682 return true;
683 if ((!vcpu->arch.vcore->arch_compat) &&
684 cpu_has_feature(CPU_FTR_ARCH_207S))
685 return true;
686 return false;
687}
688
689static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
690 unsigned long resource, unsigned long value1,
691 unsigned long value2)
692{
693 switch (resource) {
694 case H_SET_MODE_RESOURCE_SET_CIABR:
695 if (!kvmppc_power8_compatible(vcpu))
696 return H_P2;
697 if (value2)
698 return H_P4;
699 if (mflags)
700 return H_UNSUPPORTED_FLAG_START;
701 /* Guests can't breakpoint the hypervisor */
702 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
703 return H_P3;
704 vcpu->arch.ciabr = value1;
705 return H_SUCCESS;
706 case H_SET_MODE_RESOURCE_SET_DAWR:
707 if (!kvmppc_power8_compatible(vcpu))
708 return H_P2;
709 if (mflags)
710 return H_UNSUPPORTED_FLAG_START;
711 if (value2 & DABRX_HYP)
712 return H_P4;
713 vcpu->arch.dawr = value1;
714 vcpu->arch.dawrx = value2;
715 return H_SUCCESS;
716 default:
717 return H_TOO_HARD;
718 }
719}
720
90fd09f8
SB		 721static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
722{
723 struct kvmppc_vcore *vcore = target->arch.vcore;
724
725 /*
726 * We expect to have been called by the real mode handler
727 * (kvmppc_rm_h_confer()) which would have directly returned
728 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
729 * have useful work to do and should not confer) so we don't
730 * recheck that here.
731 */
732
733 spin_lock(&vcore->lock);
734 if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
ec257165
PM		 735 vcore->vcore_state != VCORE_INACTIVE &&
736 vcore->runner)
90fd09f8
SB		 737 target = vcore->runner;
738 spin_unlock(&vcore->lock);
739
740 return kvm_vcpu_yield_to(target);
741}
742
743static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
744{
745 int yield_count = 0;
746 struct lppaca *lppaca;
747
748 spin_lock(&vcpu->arch.vpa_update_lock);
749 lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
750 if (lppaca)
ecb6d618 751 yield_count = be32_to_cpu(lppaca->yield_count);
90fd09f8
SB		 752 spin_unlock(&vcpu->arch.vpa_update_lock);
753 return yield_count;
754}
755
a8606e20
PM		 756int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
757{
758 unsigned long req = kvmppc_get_gpr(vcpu, 3);
759 unsigned long target, ret = H_SUCCESS;
90fd09f8 760 int yield_count;
a8606e20 761 struct kvm_vcpu *tvcpu;
8e591cb7 762 int idx, rc;
a8606e20 763
699a0ea0
PM		 764 if (req <= MAX_HCALL_OPCODE &&
765 !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
766 return RESUME_HOST;
767
a8606e20
PM		 768 switch (req) {
769 case H_CEDE:
a8606e20
PM		 770 break;
771 case H_PROD:
772 target = kvmppc_get_gpr(vcpu, 4);
773 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
774 if (!tvcpu) {
775 ret = H_PARAMETER;
776 break;
777 }
778 tvcpu->arch.prodded = 1;
779 smp_mb();
780 if (vcpu->arch.ceded) {
8577370f
MT		 781 if (swait_active(&vcpu->wq)) {
782 swake_up(&vcpu->wq);
a8606e20
PM		 783 vcpu->stat.halt_wakeup++;
784 }
785 }
786 break;
787 case H_CONFER:
42d7604d
PM		 788 target = kvmppc_get_gpr(vcpu, 4);
789 if (target == -1)
790 break;
791 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
792 if (!tvcpu) {
793 ret = H_PARAMETER;
794 break;
795 }
90fd09f8
SB		 796 yield_count = kvmppc_get_gpr(vcpu, 5);
797 if (kvmppc_get_yield_count(tvcpu) != yield_count)
798 break;
799 kvm_arch_vcpu_yield_to(tvcpu);
a8606e20
PM		 800 break;
801 case H_REGISTER_VPA:
802 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
803 kvmppc_get_gpr(vcpu, 5),
804 kvmppc_get_gpr(vcpu, 6));
805 break;
8e591cb7
ME		 806 case H_RTAS:
807 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
808 return RESUME_HOST;
809
c9438092 810 idx = srcu_read_lock(&vcpu->kvm->srcu);
8e591cb7 811 rc = kvmppc_rtas_hcall(vcpu);
c9438092 812 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8e591cb7
ME		 813
814 if (rc == -ENOENT)
815 return RESUME_HOST;
816 else if (rc == 0)
817 break;
818
819 /* Send the error out to userspace via KVM_RUN */
820 return rc;
99342cf8
DG		 821 case H_LOGICAL_CI_LOAD:
822 ret = kvmppc_h_logical_ci_load(vcpu);
823 if (ret == H_TOO_HARD)
824 return RESUME_HOST;
825 break;
826 case H_LOGICAL_CI_STORE:
827 ret = kvmppc_h_logical_ci_store(vcpu);
828 if (ret == H_TOO_HARD)
829 return RESUME_HOST;
830 break;
9642382e
MN		 831 case H_SET_MODE:
832 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
833 kvmppc_get_gpr(vcpu, 5),
834 kvmppc_get_gpr(vcpu, 6),
835 kvmppc_get_gpr(vcpu, 7));
836 if (ret == H_TOO_HARD)
837 return RESUME_HOST;
838 break;
bc5ad3f3
BH		 839 case H_XIRR:
840 case H_CPPR:
841 case H_EOI:
842 case H_IPI:
8e44ddc3
PM		 843 case H_IPOLL:
844 case H_XIRR_X:
bc5ad3f3
BH		 845 if (kvmppc_xics_enabled(vcpu)) {
846 ret = kvmppc_xics_hcall(vcpu, req);
847 break;
d3695aa4
AK		 848 }
849 return RESUME_HOST;
850 case H_PUT_TCE:
851 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
852 kvmppc_get_gpr(vcpu, 5),
853 kvmppc_get_gpr(vcpu, 6));
854 if (ret == H_TOO_HARD)
855 return RESUME_HOST;
856 break;
857 case H_PUT_TCE_INDIRECT:
858 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
859 kvmppc_get_gpr(vcpu, 5),
860 kvmppc_get_gpr(vcpu, 6),
861 kvmppc_get_gpr(vcpu, 7));
862 if (ret == H_TOO_HARD)
863 return RESUME_HOST;
864 break;
865 case H_STUFF_TCE:
866 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
867 kvmppc_get_gpr(vcpu, 5),
868 kvmppc_get_gpr(vcpu, 6),
869 kvmppc_get_gpr(vcpu, 7));
870 if (ret == H_TOO_HARD)
871 return RESUME_HOST;
872 break;
a8606e20
PM		 873 default:
874 return RESUME_HOST;
875 }
876 kvmppc_set_gpr(vcpu, 3, ret);
877 vcpu->arch.hcall_needed = 0;
878 return RESUME_GUEST;
879}
880
ae2113a4
PM		 881static int kvmppc_hcall_impl_hv(unsigned long cmd)
882{
883 switch (cmd) {
884 case H_CEDE:
885 case H_PROD:
886 case H_CONFER:
887 case H_REGISTER_VPA:
9642382e 888 case H_SET_MODE:
99342cf8
DG		 889 case H_LOGICAL_CI_LOAD:
890 case H_LOGICAL_CI_STORE:
ae2113a4
PM		 891#ifdef CONFIG_KVM_XICS
892 case H_XIRR:
893 case H_CPPR:
894 case H_EOI:
895 case H_IPI:
896 case H_IPOLL:
897 case H_XIRR_X:
898#endif
899 return 1;
900 }
901
902 /* See if it's in the real-mode table */
903 return kvmppc_hcall_impl_hv_realmode(cmd);
904}
905
a59c1d9e
MS		 906static int kvmppc_emulate_debug_inst(struct kvm_run *run,
907 struct kvm_vcpu *vcpu)
908{
909 u32 last_inst;
910
911 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
912 EMULATE_DONE) {
913 /*
914 * Fetch failed, so return to guest and
915 * try executing it again.
916 */
917 return RESUME_GUEST;
918 }
919
920 if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
921 run->exit_reason = KVM_EXIT_DEBUG;
922 run->debug.arch.address = kvmppc_get_pc(vcpu);
923 return RESUME_HOST;
924 } else {
925 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
926 return RESUME_GUEST;
927 }
928}
929
3a167bea
AK		 930static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
931 struct task_struct *tsk)
de56a948
PM		 932{
933 int r = RESUME_HOST;
934
935 vcpu->stat.sum_exits++;
936
1c9e3d51
PM		 937 /*
938 * This can happen if an interrupt occurs in the last stages
939 * of guest entry or the first stages of guest exit (i.e. after
940 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
941 * and before setting it to KVM_GUEST_MODE_HOST_HV).
942 * That can happen due to a bug, or due to a machine check
943 * occurring at just the wrong time.
944 */
945 if (vcpu->arch.shregs.msr & MSR_HV) {
946 printk(KERN_EMERG "KVM trap in HV mode!\n");
947 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
948 vcpu->arch.trap, kvmppc_get_pc(vcpu),
949 vcpu->arch.shregs.msr);
950 kvmppc_dump_regs(vcpu);
951 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
952 run->hw.hardware_exit_reason = vcpu->arch.trap;
953 return RESUME_HOST;
954 }
de56a948
PM		 955 run->exit_reason = KVM_EXIT_UNKNOWN;
956 run->ready_for_interrupt_injection = 1;
957 switch (vcpu->arch.trap) {
958 /* We're good on these - the host merely wanted to get our attention */
959 case BOOK3S_INTERRUPT_HV_DECREMENTER:
960 vcpu->stat.dec_exits++;
961 r = RESUME_GUEST;
962 break;
963 case BOOK3S_INTERRUPT_EXTERNAL:
5d00f66b 964 case BOOK3S_INTERRUPT_H_DOORBELL:
84f7139c 965 case BOOK3S_INTERRUPT_H_VIRT:
de56a948
PM		 966 vcpu->stat.ext_intr_exits++;
967 r = RESUME_GUEST;
968 break;
dee6f24c
MS		 969 /* HMI is hypervisor interrupt and host has handled it. Resume guest.*/
970 case BOOK3S_INTERRUPT_HMI:
de56a948
PM		 971 case BOOK3S_INTERRUPT_PERFMON:
972 r = RESUME_GUEST;
973 break;
b4072df4
PM		 974 case BOOK3S_INTERRUPT_MACHINE_CHECK:
975 /*
976 * Deliver a machine check interrupt to the guest.
977 * We have to do this, even if the host has handled the
978 * machine check, because machine checks use SRR0/1 and
979 * the interrupt might have trashed guest state in them.
980 */
981 kvmppc_book3s_queue_irqprio(vcpu,
982 BOOK3S_INTERRUPT_MACHINE_CHECK);
983 r = RESUME_GUEST;
984 break;
de56a948
PM		 985 case BOOK3S_INTERRUPT_PROGRAM:
986 {
987 ulong flags;
988 /*
989 * Normally program interrupts are delivered directly
990 * to the guest by the hardware, but we can get here
991 * as a result of a hypervisor emulation interrupt
992 * (e40) getting turned into a 700 by BML RTAS.
993 */
994 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
995 kvmppc_core_queue_program(vcpu, flags);
996 r = RESUME_GUEST;
997 break;
998 }
999 case BOOK3S_INTERRUPT_SYSCALL:
1000 {
1001 /* hcall - punt to userspace */
1002 int i;
1003
27025a60
LPF		 1004 /* hypercall with MSR_PR has already been handled in rmode,
1005 * and never reaches here.
1006 */
1007
de56a948
PM		 1008 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1009 for (i = 0; i < 9; ++i)
1010 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1011 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1012 vcpu->arch.hcall_needed = 1;
1013 r = RESUME_HOST;
1014 break;
1015 }
1016 /*
342d3db7
PM		 1017 * We get these next two if the guest accesses a page which it thinks
1018 * it has mapped but which is not actually present, either because
1019 * it is for an emulated I/O device or because the corresonding
1020 * host page has been paged out. Any other HDSI/HISI interrupts
1021 * have been handled already.
de56a948
PM		 1022 */
1023 case BOOK3S_INTERRUPT_H_DATA_STORAGE:
913d3ff9 1024 r = RESUME_PAGE_FAULT;
de56a948
PM		 1025 break;
1026 case BOOK3S_INTERRUPT_H_INST_STORAGE:
913d3ff9
PM		 1027 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1028 vcpu->arch.fault_dsisr = 0;
1029 r = RESUME_PAGE_FAULT;
de56a948
PM		 1030 break;
1031 /*
1032 * This occurs if the guest executes an illegal instruction.
a59c1d9e
MS		 1033 * If the guest debug is disabled, generate a program interrupt
1034 * to the guest. If guest debug is enabled, we need to check
1035 * whether the instruction is a software breakpoint instruction.
1036 * Accordingly return to Guest or Host.
de56a948
PM		 1037 */
1038 case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
4a157d61
PM		 1039 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1040 vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1041 swab32(vcpu->arch.emul_inst) :
1042 vcpu->arch.emul_inst;
a59c1d9e
MS		 1043 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1044 r = kvmppc_emulate_debug_inst(run, vcpu);
1045 } else {
1046 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1047 r = RESUME_GUEST;
1048 }
bd3048b8
ME		 1049 break;
1050 /*
1051 * This occurs if the guest (kernel or userspace), does something that
1052 * is prohibited by HFSCR. We just generate a program interrupt to
1053 * the guest.
1054 */
1055 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1056 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
de56a948
PM		 1057 r = RESUME_GUEST;
1058 break;
f7af5209
SW		 1059 case BOOK3S_INTERRUPT_HV_RM_HARD:
1060 r = RESUME_PASSTHROUGH;
1061 break;
de56a948
PM		 1062 default:
1063 kvmppc_dump_regs(vcpu);
1064 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1065 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1066 vcpu->arch.shregs.msr);
f3271d4c 1067 run->hw.hardware_exit_reason = vcpu->arch.trap;
de56a948 1068 r = RESUME_HOST;
de56a948
PM		 1069 break;
1070 }
1071
de56a948
PM		 1072 return r;
1073}
1074
3a167bea
AK		 1075static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1076 struct kvm_sregs *sregs)
de56a948
PM		 1077{
1078 int i;
1079
de56a948 1080 memset(sregs, 0, sizeof(struct kvm_sregs));
87916442 1081 sregs->pvr = vcpu->arch.pvr;
de56a948
PM		 1082 for (i = 0; i < vcpu->arch.slb_max; i++) {
1083 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1084 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1085 }
1086
1087 return 0;
1088}
1089
3a167bea
AK		 1090static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1091 struct kvm_sregs *sregs)
de56a948
PM		 1092{
1093 int i, j;
1094
9333e6c4
PM		 1095 /* Only accept the same PVR as the host's, since we can't spoof it */
1096 if (sregs->pvr != vcpu->arch.pvr)
1097 return -EINVAL;
de56a948
PM		 1098
1099 j = 0;
1100 for (i = 0; i < vcpu->arch.slb_nr; i++) {
1101 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1102 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1103 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1104 ++j;
1105 }
1106 }
1107 vcpu->arch.slb_max = j;
1108
1109 return 0;
1110}
1111
a0840240
AK		 1112static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1113 bool preserve_top32)
a0144e2a 1114{
8f902b00 1115 struct kvm *kvm = vcpu->kvm;
a0144e2a
PM		 1116 struct kvmppc_vcore *vc = vcpu->arch.vcore;
1117 u64 mask;
1118
8f902b00 1119 mutex_lock(&kvm->lock);
a0144e2a 1120 spin_lock(&vc->lock);
d682916a
AB		 1121 /*
1122 * If ILE (interrupt little-endian) has changed, update the
1123 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1124 */
1125 if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
d682916a
AB		 1126 struct kvm_vcpu *vcpu;
1127 int i;
1128
d682916a
AB		 1129 kvm_for_each_vcpu(i, vcpu, kvm) {
1130 if (vcpu->arch.vcore != vc)
1131 continue;
1132 if (new_lpcr & LPCR_ILE)
1133 vcpu->arch.intr_msr |= MSR_LE;
1134 else
1135 vcpu->arch.intr_msr &= ~MSR_LE;
1136 }
d682916a
AB		 1137 }
1138
a0144e2a
PM		 1139 /*
1140 * Userspace can only modify DPFD (default prefetch depth),
1141 * ILE (interrupt little-endian) and TC (translation control).
e0622bd9 1142 * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
a0144e2a
PM		 1143 */
1144 mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
e0622bd9
PM		 1145 if (cpu_has_feature(CPU_FTR_ARCH_207S))
1146 mask |= LPCR_AIL;
a0840240
AK		 1147
1148 /* Broken 32-bit version of LPCR must not clear top bits */
1149 if (preserve_top32)
1150 mask &= 0xFFFFFFFF;
a0144e2a
PM		 1151 vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1152 spin_unlock(&vc->lock);
8f902b00 1153 mutex_unlock(&kvm->lock);
a0144e2a
PM		 1154}
1155
3a167bea
AK		 1156static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1157 union kvmppc_one_reg *val)
31f3438e 1158{
a136a8bd
PM		 1159 int r = 0;
1160 long int i;
31f3438e 1161
a136a8bd 1162 switch (id) {
a59c1d9e
MS		 1163 case KVM_REG_PPC_DEBUG_INST:
1164 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1165 break;
31f3438e 1166 case KVM_REG_PPC_HIOR:
a136a8bd
PM		 1167 *val = get_reg_val(id, 0);
1168 break;
1169 case KVM_REG_PPC_DABR:
1170 *val = get_reg_val(id, vcpu->arch.dabr);
1171 break;
8563bf52
PM		 1172 case KVM_REG_PPC_DABRX:
1173 *val = get_reg_val(id, vcpu->arch.dabrx);
1174 break;
a136a8bd
PM		 1175 case KVM_REG_PPC_DSCR:
1176 *val = get_reg_val(id, vcpu->arch.dscr);
1177 break;
1178 case KVM_REG_PPC_PURR:
1179 *val = get_reg_val(id, vcpu->arch.purr);
1180 break;
1181 case KVM_REG_PPC_SPURR:
1182 *val = get_reg_val(id, vcpu->arch.spurr);
1183 break;
1184 case KVM_REG_PPC_AMR:
1185 *val = get_reg_val(id, vcpu->arch.amr);
1186 break;
1187 case KVM_REG_PPC_UAMOR:
1188 *val = get_reg_val(id, vcpu->arch.uamor);
1189 break;
b005255e 1190 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
a136a8bd
PM		 1191 i = id - KVM_REG_PPC_MMCR0;
1192 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1193 break;
1194 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1195 i = id - KVM_REG_PPC_PMC1;
1196 *val = get_reg_val(id, vcpu->arch.pmc[i]);
31f3438e 1197 break;
b005255e
MN		 1198 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1199 i = id - KVM_REG_PPC_SPMC1;
1200 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1201 break;
14941789
PM		 1202 case KVM_REG_PPC_SIAR:
1203 *val = get_reg_val(id, vcpu->arch.siar);
1204 break;
1205 case KVM_REG_PPC_SDAR:
1206 *val = get_reg_val(id, vcpu->arch.sdar);
1207 break;
b005255e
MN		 1208 case KVM_REG_PPC_SIER:
1209 *val = get_reg_val(id, vcpu->arch.sier);
a8bd19ef 1210 break;
b005255e
MN		 1211 case KVM_REG_PPC_IAMR:
1212 *val = get_reg_val(id, vcpu->arch.iamr);
1213 break;
b005255e
MN		 1214 case KVM_REG_PPC_PSPB:
1215 *val = get_reg_val(id, vcpu->arch.pspb);
1216 break;
b005255e
MN		 1217 case KVM_REG_PPC_DPDES:
1218 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1219 break;
88b02cf9
PM		 1220 case KVM_REG_PPC_VTB:
1221 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1222 break;
b005255e
MN		 1223 case KVM_REG_PPC_DAWR:
1224 *val = get_reg_val(id, vcpu->arch.dawr);
1225 break;
1226 case KVM_REG_PPC_DAWRX:
1227 *val = get_reg_val(id, vcpu->arch.dawrx);
1228 break;
1229 case KVM_REG_PPC_CIABR:
1230 *val = get_reg_val(id, vcpu->arch.ciabr);
1231 break;
b005255e
MN		 1232 case KVM_REG_PPC_CSIGR:
1233 *val = get_reg_val(id, vcpu->arch.csigr);
1234 break;
1235 case KVM_REG_PPC_TACR:
1236 *val = get_reg_val(id, vcpu->arch.tacr);
1237 break;
1238 case KVM_REG_PPC_TCSCR:
1239 *val = get_reg_val(id, vcpu->arch.tcscr);
1240 break;
1241 case KVM_REG_PPC_PID:
1242 *val = get_reg_val(id, vcpu->arch.pid);
1243 break;
1244 case KVM_REG_PPC_ACOP:
1245 *val = get_reg_val(id, vcpu->arch.acop);
1246 break;
1247 case KVM_REG_PPC_WORT:
1248 *val = get_reg_val(id, vcpu->arch.wort);
a8bd19ef 1249 break;
e9cf1e08
PM		 1250 case KVM_REG_PPC_TIDR:
1251 *val = get_reg_val(id, vcpu->arch.tid);
1252 break;
1253 case KVM_REG_PPC_PSSCR:
1254 *val = get_reg_val(id, vcpu->arch.psscr);
1255 break;
55b665b0
PM		 1256 case KVM_REG_PPC_VPA_ADDR:
1257 spin_lock(&vcpu->arch.vpa_update_lock);
1258 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1259 spin_unlock(&vcpu->arch.vpa_update_lock);
1260 break;
1261 case KVM_REG_PPC_VPA_SLB:
1262 spin_lock(&vcpu->arch.vpa_update_lock);
1263 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1264 val->vpaval.length = vcpu->arch.slb_shadow.len;
1265 spin_unlock(&vcpu->arch.vpa_update_lock);
1266 break;
1267 case KVM_REG_PPC_VPA_DTL:
1268 spin_lock(&vcpu->arch.vpa_update_lock);
1269 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1270 val->vpaval.length = vcpu->arch.dtl.len;
1271 spin_unlock(&vcpu->arch.vpa_update_lock);
1272 break;
93b0f4dc
PM		 1273 case KVM_REG_PPC_TB_OFFSET:
1274 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1275 break;
a0144e2a 1276 case KVM_REG_PPC_LPCR:
a0840240 1277 case KVM_REG_PPC_LPCR_64:
a0144e2a
PM		 1278 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1279 break;
4b8473c9
PM		 1280 case KVM_REG_PPC_PPR:
1281 *val = get_reg_val(id, vcpu->arch.ppr);
1282 break;
a7d80d01
MN		 1283#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1284 case KVM_REG_PPC_TFHAR:
1285 *val = get_reg_val(id, vcpu->arch.tfhar);
1286 break;
1287 case KVM_REG_PPC_TFIAR:
1288 *val = get_reg_val(id, vcpu->arch.tfiar);
1289 break;
1290 case KVM_REG_PPC_TEXASR:
1291 *val = get_reg_val(id, vcpu->arch.texasr);
1292 break;
1293 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1294 i = id - KVM_REG_PPC_TM_GPR0;
1295 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1296 break;
1297 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1298 {
1299 int j;
1300 i = id - KVM_REG_PPC_TM_VSR0;
1301 if (i < 32)
1302 for (j = 0; j < TS_FPRWIDTH; j++)
1303 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1304 else {
1305 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1306 val->vval = vcpu->arch.vr_tm.vr[i-32];
1307 else
1308 r = -ENXIO;
1309 }
1310 break;
1311 }
1312 case KVM_REG_PPC_TM_CR:
1313 *val = get_reg_val(id, vcpu->arch.cr_tm);
1314 break;
0d808df0
PM		 1315 case KVM_REG_PPC_TM_XER:
1316 *val = get_reg_val(id, vcpu->arch.xer_tm);
1317 break;
a7d80d01
MN		 1318 case KVM_REG_PPC_TM_LR:
1319 *val = get_reg_val(id, vcpu->arch.lr_tm);
1320 break;
1321 case KVM_REG_PPC_TM_CTR:
1322 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1323 break;
1324 case KVM_REG_PPC_TM_FPSCR:
1325 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1326 break;
1327 case KVM_REG_PPC_TM_AMR:
1328 *val = get_reg_val(id, vcpu->arch.amr_tm);
1329 break;
1330 case KVM_REG_PPC_TM_PPR:
1331 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1332 break;
1333 case KVM_REG_PPC_TM_VRSAVE:
1334 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1335 break;
1336 case KVM_REG_PPC_TM_VSCR:
1337 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1338 *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1339 else
1340 r = -ENXIO;
1341 break;
1342 case KVM_REG_PPC_TM_DSCR:
1343 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1344 break;
1345 case KVM_REG_PPC_TM_TAR:
1346 *val = get_reg_val(id, vcpu->arch.tar_tm);
1347 break;
1348#endif
388cc6e1
PM		 1349 case KVM_REG_PPC_ARCH_COMPAT:
1350 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1351 break;
31f3438e 1352 default:
a136a8bd 1353 r = -EINVAL;
31f3438e
PM		 1354 break;
1355 }
1356
1357 return r;
1358}
1359
3a167bea
AK		 1360static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1361 union kvmppc_one_reg *val)
31f3438e 1362{
a136a8bd
PM		 1363 int r = 0;
1364 long int i;
55b665b0 1365 unsigned long addr, len;
31f3438e 1366
a136a8bd 1367 switch (id) {
31f3438e 1368 case KVM_REG_PPC_HIOR:
31f3438e 1369 /* Only allow this to be set to zero */
a136a8bd 1370 if (set_reg_val(id, *val))
31f3438e
PM		 1371 r = -EINVAL;
1372 break;
a136a8bd
PM		 1373 case KVM_REG_PPC_DABR:
1374 vcpu->arch.dabr = set_reg_val(id, *val);
1375 break;
8563bf52
PM		 1376 case KVM_REG_PPC_DABRX:
1377 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1378 break;
a136a8bd
PM		 1379 case KVM_REG_PPC_DSCR:
1380 vcpu->arch.dscr = set_reg_val(id, *val);
1381 break;
1382 case KVM_REG_PPC_PURR:
1383 vcpu->arch.purr = set_reg_val(id, *val);
1384 break;
1385 case KVM_REG_PPC_SPURR:
1386 vcpu->arch.spurr = set_reg_val(id, *val);
1387 break;
1388 case KVM_REG_PPC_AMR:
1389 vcpu->arch.amr = set_reg_val(id, *val);
1390 break;
1391 case KVM_REG_PPC_UAMOR:
1392 vcpu->arch.uamor = set_reg_val(id, *val);
1393 break;
b005255e 1394 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
a136a8bd
PM		 1395 i = id - KVM_REG_PPC_MMCR0;
1396 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1397 break;
1398 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1399 i = id - KVM_REG_PPC_PMC1;
1400 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1401 break;
b005255e
MN		 1402 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1403 i = id - KVM_REG_PPC_SPMC1;
1404 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1405 break;
14941789
PM		 1406 case KVM_REG_PPC_SIAR:
1407 vcpu->arch.siar = set_reg_val(id, *val);
1408 break;
1409 case KVM_REG_PPC_SDAR:
1410 vcpu->arch.sdar = set_reg_val(id, *val);
1411 break;
b005255e
MN		 1412 case KVM_REG_PPC_SIER:
1413 vcpu->arch.sier = set_reg_val(id, *val);
a8bd19ef 1414 break;
b005255e
MN		 1415 case KVM_REG_PPC_IAMR:
1416 vcpu->arch.iamr = set_reg_val(id, *val);
1417 break;
b005255e
MN		 1418 case KVM_REG_PPC_PSPB:
1419 vcpu->arch.pspb = set_reg_val(id, *val);
1420 break;
b005255e
MN		 1421 case KVM_REG_PPC_DPDES:
1422 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1423 break;
88b02cf9
PM		 1424 case KVM_REG_PPC_VTB:
1425 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1426 break;
b005255e
MN		 1427 case KVM_REG_PPC_DAWR:
1428 vcpu->arch.dawr = set_reg_val(id, *val);
1429 break;
1430 case KVM_REG_PPC_DAWRX:
1431 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1432 break;
1433 case KVM_REG_PPC_CIABR:
1434 vcpu->arch.ciabr = set_reg_val(id, *val);
1435 /* Don't allow setting breakpoints in hypervisor code */
1436 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1437 vcpu->arch.ciabr &= ~CIABR_PRIV; /* disable */
1438 break;
b005255e
MN		 1439 case KVM_REG_PPC_CSIGR:
1440 vcpu->arch.csigr = set_reg_val(id, *val);
1441 break;
1442 case KVM_REG_PPC_TACR:
1443 vcpu->arch.tacr = set_reg_val(id, *val);
1444 break;
1445 case KVM_REG_PPC_TCSCR:
1446 vcpu->arch.tcscr = set_reg_val(id, *val);
1447 break;
1448 case KVM_REG_PPC_PID:
1449 vcpu->arch.pid = set_reg_val(id, *val);
1450 break;
1451 case KVM_REG_PPC_ACOP:
1452 vcpu->arch.acop = set_reg_val(id, *val);
1453 break;
1454 case KVM_REG_PPC_WORT:
1455 vcpu->arch.wort = set_reg_val(id, *val);
a8bd19ef 1456 break;
e9cf1e08
PM		 1457 case KVM_REG_PPC_TIDR:
1458 vcpu->arch.tid = set_reg_val(id, *val);
1459 break;
1460 case KVM_REG_PPC_PSSCR:
1461 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1462 break;
55b665b0
PM		 1463 case KVM_REG_PPC_VPA_ADDR:
1464 addr = set_reg_val(id, *val);
1465 r = -EINVAL;
1466 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1467 vcpu->arch.dtl.next_gpa))
1468 break;
1469 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1470 break;
1471 case KVM_REG_PPC_VPA_SLB:
1472 addr = val->vpaval.addr;
1473 len = val->vpaval.length;
1474 r = -EINVAL;
1475 if (addr && !vcpu->arch.vpa.next_gpa)
1476 break;
1477 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1478 break;
1479 case KVM_REG_PPC_VPA_DTL:
1480 addr = val->vpaval.addr;
1481 len = val->vpaval.length;
1482 r = -EINVAL;
9f8c8c78
PM		 1483 if (addr && (len < sizeof(struct dtl_entry) ||
1484 !vcpu->arch.vpa.next_gpa))
55b665b0
PM		 1485 break;
1486 len -= len % sizeof(struct dtl_entry);
1487 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1488 break;
93b0f4dc
PM		 1489 case KVM_REG_PPC_TB_OFFSET:
1490 /* round up to multiple of 2^24 */
1491 vcpu->arch.vcore->tb_offset =
1492 ALIGN(set_reg_val(id, *val), 1UL << 24);
1493 break;
a0144e2a 1494 case KVM_REG_PPC_LPCR:
a0840240
AK		 1495 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1496 break;
1497 case KVM_REG_PPC_LPCR_64:
1498 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
a0144e2a 1499 break;
4b8473c9
PM		 1500 case KVM_REG_PPC_PPR:
1501 vcpu->arch.ppr = set_reg_val(id, *val);
1502 break;
a7d80d01
MN		 1503#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1504 case KVM_REG_PPC_TFHAR:
1505 vcpu->arch.tfhar = set_reg_val(id, *val);
1506 break;
1507 case KVM_REG_PPC_TFIAR:
1508 vcpu->arch.tfiar = set_reg_val(id, *val);
1509 break;
1510 case KVM_REG_PPC_TEXASR:
1511 vcpu->arch.texasr = set_reg_val(id, *val);
1512 break;
1513 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1514 i = id - KVM_REG_PPC_TM_GPR0;
1515 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1516 break;
1517 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1518 {
1519 int j;
1520 i = id - KVM_REG_PPC_TM_VSR0;
1521 if (i < 32)
1522 for (j = 0; j < TS_FPRWIDTH; j++)
1523 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1524 else
1525 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1526 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1527 else
1528 r = -ENXIO;
1529 break;
1530 }
1531 case KVM_REG_PPC_TM_CR:
1532 vcpu->arch.cr_tm = set_reg_val(id, *val);
1533 break;
0d808df0
PM		 1534 case KVM_REG_PPC_TM_XER:
1535 vcpu->arch.xer_tm = set_reg_val(id, *val);
1536 break;
a7d80d01
MN		 1537 case KVM_REG_PPC_TM_LR:
1538 vcpu->arch.lr_tm = set_reg_val(id, *val);
1539 break;
1540 case KVM_REG_PPC_TM_CTR:
1541 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1542 break;
1543 case KVM_REG_PPC_TM_FPSCR:
1544 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1545 break;
1546 case KVM_REG_PPC_TM_AMR:
1547 vcpu->arch.amr_tm = set_reg_val(id, *val);
1548 break;
1549 case KVM_REG_PPC_TM_PPR:
1550 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1551 break;
1552 case KVM_REG_PPC_TM_VRSAVE:
1553 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1554 break;
1555 case KVM_REG_PPC_TM_VSCR:
1556 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1557 vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1558 else
1559 r = - ENXIO;
1560 break;
1561 case KVM_REG_PPC_TM_DSCR:
1562 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1563 break;
1564 case KVM_REG_PPC_TM_TAR:
1565 vcpu->arch.tar_tm = set_reg_val(id, *val);
1566 break;
1567#endif
388cc6e1
PM		 1568 case KVM_REG_PPC_ARCH_COMPAT:
1569 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1570 break;
31f3438e 1571 default:
a136a8bd 1572 r = -EINVAL;
31f3438e
PM		 1573 break;
1574 }
1575
1576 return r;
1577}
1578
de9bdd1a
SS		 1579static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1580{
1581 struct kvmppc_vcore *vcore;
1582
1583 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1584
1585 if (vcore == NULL)
1586 return NULL;
1587
de9bdd1a 1588 spin_lock_init(&vcore->lock);
2711e248 1589 spin_lock_init(&vcore->stoltb_lock);
8577370f 1590 init_swait_queue_head(&vcore->wq);
de9bdd1a
SS		 1591 vcore->preempt_tb = TB_NIL;
1592 vcore->lpcr = kvm->arch.lpcr;
1593 vcore->first_vcpuid = core * threads_per_subcore;
1594 vcore->kvm = kvm;
ec257165 1595 INIT_LIST_HEAD(&vcore->preempt_list);
de9bdd1a
SS		 1596
1597 return vcore;
1598}
1599
b6c295df
PM		 1600#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1601static struct debugfs_timings_element {
1602 const char *name;
1603 size_t offset;
1604} timings[] = {
1605 {"rm_entry", offsetof(struct kvm_vcpu, arch.rm_entry)},
1606 {"rm_intr", offsetof(struct kvm_vcpu, arch.rm_intr)},
1607 {"rm_exit", offsetof(struct kvm_vcpu, arch.rm_exit)},
1608 {"guest", offsetof(struct kvm_vcpu, arch.guest_time)},
1609 {"cede", offsetof(struct kvm_vcpu, arch.cede_time)},
1610};
1611
1612#define N_TIMINGS (sizeof(timings) / sizeof(timings[0]))
1613
1614struct debugfs_timings_state {
1615 struct kvm_vcpu *vcpu;
1616 unsigned int buflen;
1617 char buf[N_TIMINGS * 100];
1618};
1619
1620static int debugfs_timings_open(struct inode *inode, struct file *file)
1621{
1622 struct kvm_vcpu *vcpu = inode->i_private;
1623 struct debugfs_timings_state *p;
1624
1625 p = kzalloc(sizeof(*p), GFP_KERNEL);
1626 if (!p)
1627 return -ENOMEM;
1628
1629 kvm_get_kvm(vcpu->kvm);
1630 p->vcpu = vcpu;
1631 file->private_data = p;
1632
1633 return nonseekable_open(inode, file);
1634}
1635
1636static int debugfs_timings_release(struct inode *inode, struct file *file)
1637{
1638 struct debugfs_timings_state *p = file->private_data;
1639
1640 kvm_put_kvm(p->vcpu->kvm);
1641 kfree(p);
1642 return 0;
1643}
1644
1645static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1646 size_t len, loff_t *ppos)
1647{
1648 struct debugfs_timings_state *p = file->private_data;
1649 struct kvm_vcpu *vcpu = p->vcpu;
1650 char *s, *buf_end;
1651 struct kvmhv_tb_accumulator tb;
1652 u64 count;
1653 loff_t pos;
1654 ssize_t n;
1655 int i, loops;
1656 bool ok;
1657
1658 if (!p->buflen) {
1659 s = p->buf;
1660 buf_end = s + sizeof(p->buf);
1661 for (i = 0; i < N_TIMINGS; ++i) {
1662 struct kvmhv_tb_accumulator *acc;
1663
1664 acc = (struct kvmhv_tb_accumulator *)
1665 ((unsigned long)vcpu + timings[i].offset);
1666 ok = false;
1667 for (loops = 0; loops < 1000; ++loops) {
1668 count = acc->seqcount;
1669 if (!(count & 1)) {
1670 smp_rmb();
1671 tb = *acc;
1672 smp_rmb();
1673 if (count == acc->seqcount) {
1674 ok = true;
1675 break;
1676 }
1677 }
1678 udelay(1);
1679 }
1680 if (!ok)
1681 snprintf(s, buf_end - s, "%s: stuck\n",
1682 timings[i].name);
1683 else
1684 snprintf(s, buf_end - s,
1685 "%s: %llu %llu %llu %llu\n",
1686 timings[i].name, count / 2,
1687 tb_to_ns(tb.tb_total),
1688 tb_to_ns(tb.tb_min),
1689 tb_to_ns(tb.tb_max));
1690 s += strlen(s);
1691 }
1692 p->buflen = s - p->buf;
1693 }
1694
1695 pos = *ppos;
1696 if (pos >= p->buflen)
1697 return 0;
1698 if (len > p->buflen - pos)
1699 len = p->buflen - pos;
1700 n = copy_to_user(buf, p->buf + pos, len);
1701 if (n) {
1702 if (n == len)
1703 return -EFAULT;
1704 len -= n;
1705 }
1706 *ppos = pos + len;
1707 return len;
1708}
1709
1710static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1711 size_t len, loff_t *ppos)
1712{
1713 return -EACCES;
1714}
1715
1716static const struct file_operations debugfs_timings_ops = {
1717 .owner = THIS_MODULE,
1718 .open = debugfs_timings_open,
1719 .release = debugfs_timings_release,
1720 .read = debugfs_timings_read,
1721 .write = debugfs_timings_write,
1722 .llseek = generic_file_llseek,
1723};
1724
1725/* Create a debugfs directory for the vcpu */
1726static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1727{
1728 char buf[16];
1729 struct kvm *kvm = vcpu->kvm;
1730
1731 snprintf(buf, sizeof(buf), "vcpu%u", id);
1732 if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1733 return;
1734 vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1735 if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1736 return;
1737 vcpu->arch.debugfs_timings =
1738 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1739 vcpu, &debugfs_timings_ops);
1740}
1741
1742#else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1743static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1744{
1745}
1746#endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1747
3a167bea
AK		 1748static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1749 unsigned int id)
de56a948
PM		 1750{
1751 struct kvm_vcpu *vcpu;
371fefd6
PM		 1752 int err = -EINVAL;
1753 int core;
1754 struct kvmppc_vcore *vcore;
de56a948 1755
3102f784 1756 core = id / threads_per_subcore;
371fefd6
PM		 1757 if (core >= KVM_MAX_VCORES)
1758 goto out;
1759
1760 err = -ENOMEM;
6b75e6bf 1761 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
de56a948
PM		 1762 if (!vcpu)
1763 goto out;
1764
1765 err = kvm_vcpu_init(vcpu, kvm, id);
1766 if (err)
1767 goto free_vcpu;
1768
1769 vcpu->arch.shared = &vcpu->arch.shregs;
5deb8e7a
AG		 1770#ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1771 /*
1772 * The shared struct is never shared on HV,
1773 * so we can always use host endianness
1774 */
1775#ifdef __BIG_ENDIAN__
1776 vcpu->arch.shared_big_endian = true;
1777#else
1778 vcpu->arch.shared_big_endian = false;
1779#endif
1780#endif
de56a948
PM		 1781 vcpu->arch.mmcr[0] = MMCR0_FC;
1782 vcpu->arch.ctrl = CTRL_RUNLATCH;
1783 /* default to host PVR, since we can't spoof it */
3a167bea 1784 kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2e25aa5f 1785 spin_lock_init(&vcpu->arch.vpa_update_lock);
c7b67670
PM		 1786 spin_lock_init(&vcpu->arch.tbacct_lock);
1787 vcpu->arch.busy_preempt = TB_NIL;
d682916a 1788 vcpu->arch.intr_msr = MSR_SF | MSR_ME;
de56a948 1789
de56a948
PM		 1790 kvmppc_mmu_book3s_hv_init(vcpu);
1791
8455d79e 1792 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
371fefd6
PM		 1793
1794 init_waitqueue_head(&vcpu->arch.cpu_run);
1795
1796 mutex_lock(&kvm->lock);
1797 vcore = kvm->arch.vcores[core];
1798 if (!vcore) {
de9bdd1a 1799 vcore = kvmppc_vcore_create(kvm, core);
371fefd6 1800 kvm->arch.vcores[core] = vcore;
1b400ba0 1801 kvm->arch.online_vcores++;
371fefd6
PM		 1802 }
1803 mutex_unlock(&kvm->lock);
1804
1805 if (!vcore)
1806 goto free_vcpu;
1807
1808 spin_lock(&vcore->lock);
1809 ++vcore->num_threads;
371fefd6
PM		 1810 spin_unlock(&vcore->lock);
1811 vcpu->arch.vcore = vcore;
e0b7ec05 1812 vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
ec257165 1813 vcpu->arch.thread_cpu = -1;
371fefd6 1814
af8f38b3
AG		 1815 vcpu->arch.cpu_type = KVM_CPU_3S_64;
1816 kvmppc_sanity_check(vcpu);
1817
b6c295df
PM		 1818 debugfs_vcpu_init(vcpu, id);
1819
de56a948
PM		 1820 return vcpu;
1821
1822free_vcpu:
6b75e6bf 1823 kmem_cache_free(kvm_vcpu_cache, vcpu);
de56a948
PM		 1824out:
1825 return ERR_PTR(err);
1826}
1827
c35635ef
PM		 1828static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1829{
1830 if (vpa->pinned_addr)
1831 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1832 vpa->dirty);
1833}
1834
3a167bea 1835static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
de56a948 1836{
2e25aa5f 1837 spin_lock(&vcpu->arch.vpa_update_lock);
c35635ef
PM		 1838 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1839 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1840 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2e25aa5f 1841 spin_unlock(&vcpu->arch.vpa_update_lock);
de56a948 1842 kvm_vcpu_uninit(vcpu);
6b75e6bf 1843 kmem_cache_free(kvm_vcpu_cache, vcpu);
de56a948
PM		 1844}
1845
3a167bea
AK		 1846static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1847{
1848 /* Indicate we want to get back into the guest */
1849 return 1;
1850}
1851
19ccb76a 1852static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
371fefd6 1853{
19ccb76a 1854 unsigned long dec_nsec, now;
371fefd6 1855
19ccb76a
PM		 1856 now = get_tb();
1857 if (now > vcpu->arch.dec_expires) {
1858 /* decrementer has already gone negative */
1859 kvmppc_core_queue_dec(vcpu);
7e28e60e 1860 kvmppc_core_prepare_to_enter(vcpu);
19ccb76a 1861 return;
371fefd6 1862 }
19ccb76a
PM		 1863 dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1864 / tb_ticks_per_sec;
1865 hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1866 HRTIMER_MODE_REL);
1867 vcpu->arch.timer_running = 1;
371fefd6
PM		 1868}
1869
19ccb76a 1870static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
371fefd6 1871{
19ccb76a
PM		 1872 vcpu->arch.ceded = 0;
1873 if (vcpu->arch.timer_running) {
1874 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1875 vcpu->arch.timer_running = 0;
1876 }
371fefd6
PM		 1877}
1878
e0b7ec05 1879extern void __kvmppc_vcore_entry(void);
de56a948 1880
371fefd6
PM		 1881static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1882 struct kvm_vcpu *vcpu)
de56a948 1883{
c7b67670
PM		 1884 u64 now;
1885
371fefd6
PM		 1886 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1887 return;
bf3d32e1 1888 spin_lock_irq(&vcpu->arch.tbacct_lock);
c7b67670
PM		 1889 now = mftb();
1890 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1891 vcpu->arch.stolen_logged;
1892 vcpu->arch.busy_preempt = now;
1893 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
bf3d32e1 1894 spin_unlock_irq(&vcpu->arch.tbacct_lock);
371fefd6 1895 --vc->n_runnable;
7b5f8272 1896 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
371fefd6
PM		 1897}
1898
f0888f70
PM		 1899static int kvmppc_grab_hwthread(int cpu)
1900{
1901 struct paca_struct *tpaca;
b754c739 1902 long timeout = 10000;
f0888f70
PM		 1903
1904 tpaca = &paca[cpu];
1905
1906 /* Ensure the thread won't go into the kernel if it wakes */
7b444c67 1907 tpaca->kvm_hstate.kvm_vcpu = NULL;
b4deba5c 1908 tpaca->kvm_hstate.kvm_vcore = NULL;
5d5b99cd
PM		 1909 tpaca->kvm_hstate.napping = 0;
1910 smp_wmb();
1911 tpaca->kvm_hstate.hwthread_req = 1;
f0888f70
PM		 1912
1913 /*
1914 * If the thread is already executing in the kernel (e.g. handling
1915 * a stray interrupt), wait for it to get back to nap mode.
1916 * The smp_mb() is to ensure that our setting of hwthread_req
1917 * is visible before we look at hwthread_state, so if this
1918 * races with the code at system_reset_pSeries and the thread
1919 * misses our setting of hwthread_req, we are sure to see its
1920 * setting of hwthread_state, and vice versa.
1921 */
1922 smp_mb();
1923 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1924 if (--timeout <= 0) {
1925 pr_err("KVM: couldn't grab cpu %d\n", cpu);
1926 return -EBUSY;
1927 }
1928 udelay(1);
1929 }
1930 return 0;
1931}
1932
1933static void kvmppc_release_hwthread(int cpu)
1934{
1935 struct paca_struct *tpaca;
1936
1937 tpaca = &paca[cpu];
1938 tpaca->kvm_hstate.hwthread_req = 0;
1939 tpaca->kvm_hstate.kvm_vcpu = NULL;
b4deba5c
PM		 1940 tpaca->kvm_hstate.kvm_vcore = NULL;
1941 tpaca->kvm_hstate.kvm_split_mode = NULL;
f0888f70
PM		 1942}
1943
b4deba5c 1944static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
371fefd6
PM		 1945{
1946 int cpu;
1947 struct paca_struct *tpaca;
ec257165 1948 struct kvmppc_vcore *mvc = vc->master_vcore;
371fefd6 1949
b4deba5c
PM		 1950 cpu = vc->pcpu;
1951 if (vcpu) {
1952 if (vcpu->arch.timer_running) {
1953 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1954 vcpu->arch.timer_running = 0;
1955 }
1956 cpu += vcpu->arch.ptid;
1957 vcpu->cpu = mvc->pcpu;
1958 vcpu->arch.thread_cpu = cpu;
19ccb76a 1959 }
371fefd6 1960 tpaca = &paca[cpu];
5d5b99cd 1961 tpaca->kvm_hstate.kvm_vcpu = vcpu;
ec257165 1962 tpaca->kvm_hstate.ptid = cpu - mvc->pcpu;
ec257165 1963 /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
371fefd6 1964 smp_wmb();
b4deba5c 1965 tpaca->kvm_hstate.kvm_vcore = mvc;
5d5b99cd 1966 if (cpu != smp_processor_id())
66feed61 1967 kvmppc_ipi_thread(cpu);
371fefd6 1968}
de56a948 1969
5d5b99cd 1970static void kvmppc_wait_for_nap(void)
371fefd6 1971{
5d5b99cd
PM		 1972 int cpu = smp_processor_id();
1973 int i, loops;
371fefd6 1974
5d5b99cd
PM		 1975 for (loops = 0; loops < 1000000; ++loops) {
1976 /*
1977 * Check if all threads are finished.
b4deba5c 1978 * We set the vcore pointer when starting a thread
5d5b99cd 1979 * and the thread clears it when finished, so we look
b4deba5c 1980 * for any threads that still have a non-NULL vcore ptr.
5d5b99cd
PM		 1981 */
1982 for (i = 1; i < threads_per_subcore; ++i)
b4deba5c 1983 if (paca[cpu + i].kvm_hstate.kvm_vcore)
5d5b99cd
PM		 1984 break;
1985 if (i == threads_per_subcore) {
1986 HMT_medium();
1987 return;
371fefd6 1988 }
5d5b99cd 1989 HMT_low();
371fefd6
PM		 1990 }
1991 HMT_medium();
5d5b99cd 1992 for (i = 1; i < threads_per_subcore; ++i)
b4deba5c 1993 if (paca[cpu + i].kvm_hstate.kvm_vcore)
5d5b99cd 1994 pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
371fefd6
PM		 1995}
1996
1997/*
1998 * Check that we are on thread 0 and that any other threads in
7b444c67
PM		 1999 * this core are off-line. Then grab the threads so they can't
2000 * enter the kernel.
371fefd6
PM		 2001 */
2002static int on_primary_thread(void)
2003{
2004 int cpu = smp_processor_id();
3102f784 2005 int thr;
371fefd6 2006
3102f784
ME		 2007 /* Are we on a primary subcore? */
2008 if (cpu_thread_in_subcore(cpu))
371fefd6 2009 return 0;
3102f784
ME		 2010
2011 thr = 0;
2012 while (++thr < threads_per_subcore)
371fefd6
PM		 2013 if (cpu_online(cpu + thr))
2014 return 0;
7b444c67
PM		 2015
2016 /* Grab all hw threads so they can't go into the kernel */
3102f784 2017 for (thr = 1; thr < threads_per_subcore; ++thr) {
7b444c67
PM		 2018 if (kvmppc_grab_hwthread(cpu + thr)) {
2019 /* Couldn't grab one; let the others go */
2020 do {
2021 kvmppc_release_hwthread(cpu + thr);
2022 } while (--thr > 0);
2023 return 0;
2024 }
2025 }
371fefd6
PM		 2026 return 1;
2027}
2028
ec257165
PM		 2029/*
2030 * A list of virtual cores for each physical CPU.
2031 * These are vcores that could run but their runner VCPU tasks are
2032 * (or may be) preempted.
2033 */
2034struct preempted_vcore_list {
2035 struct list_head list;
2036 spinlock_t lock;
2037};
2038
2039static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2040
2041static void init_vcore_lists(void)
2042{
2043 int cpu;
2044
2045 for_each_possible_cpu(cpu) {
2046 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2047 spin_lock_init(&lp->lock);
2048 INIT_LIST_HEAD(&lp->list);
2049 }
2050}
2051
2052static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2053{
2054 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2055
2056 vc->vcore_state = VCORE_PREEMPT;
2057 vc->pcpu = smp_processor_id();
2058 if (vc->num_threads < threads_per_subcore) {
2059 spin_lock(&lp->lock);
2060 list_add_tail(&vc->preempt_list, &lp->list);
2061 spin_unlock(&lp->lock);
2062 }
2063
2064 /* Start accumulating stolen time */
2065 kvmppc_core_start_stolen(vc);
2066}
2067
2068static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2069{
402813fe 2070 struct preempted_vcore_list *lp;
ec257165
PM		 2071
2072 kvmppc_core_end_stolen(vc);
2073 if (!list_empty(&vc->preempt_list)) {
402813fe 2074 lp = &per_cpu(preempted_vcores, vc->pcpu);
ec257165
PM		 2075 spin_lock(&lp->lock);
2076 list_del_init(&vc->preempt_list);
2077 spin_unlock(&lp->lock);
2078 }
2079 vc->vcore_state = VCORE_INACTIVE;
2080}
2081
b4deba5c
PM		 2082/*
2083 * This stores information about the virtual cores currently
2084 * assigned to a physical core.
2085 */
ec257165 2086struct core_info {
b4deba5c
PM		 2087 int n_subcores;
2088 int max_subcore_threads;
ec257165 2089 int total_threads;
b4deba5c
PM		 2090 int subcore_threads[MAX_SUBCORES];
2091 struct kvm *subcore_vm[MAX_SUBCORES];
2092 struct list_head vcs[MAX_SUBCORES];
ec257165
PM		 2093};
2094
b4deba5c
PM		 2095/*
2096 * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2097 * respectively in 2-way micro-threading (split-core) mode.
2098 */
2099static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2100
ec257165
PM		 2101static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2102{
b4deba5c
PM		 2103 int sub;
2104
ec257165 2105 memset(cip, 0, sizeof(*cip));
b4deba5c
PM		 2106 cip->n_subcores = 1;
2107 cip->max_subcore_threads = vc->num_threads;
ec257165 2108 cip->total_threads = vc->num_threads;
b4deba5c
PM		 2109 cip->subcore_threads[0] = vc->num_threads;
2110 cip->subcore_vm[0] = vc->kvm;
2111 for (sub = 0; sub < MAX_SUBCORES; ++sub)
2112 INIT_LIST_HEAD(&cip->vcs[sub]);
2113 list_add_tail(&vc->preempt_list, &cip->vcs[0]);
2114}
2115
2116static bool subcore_config_ok(int n_subcores, int n_threads)
2117{
2118 /* Can only dynamically split if unsplit to begin with */
2119 if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2120 return false;
2121 if (n_subcores > MAX_SUBCORES)
2122 return false;
2123 if (n_subcores > 1) {
2124 if (!(dynamic_mt_modes & 2))
2125 n_subcores = 4;
2126 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2127 return false;
2128 }
2129
2130 return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
ec257165
PM		 2131}
2132
2133static void init_master_vcore(struct kvmppc_vcore *vc)
2134{
2135 vc->master_vcore = vc;
2136 vc->entry_exit_map = 0;
2137 vc->in_guest = 0;
2138 vc->napping_threads = 0;
2139 vc->conferring_threads = 0;
2140}
2141
b4deba5c
PM		 2142static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2143{
2144 int n_threads = vc->num_threads;
2145 int sub;
2146
2147 if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2148 return false;
2149
2150 if (n_threads < cip->max_subcore_threads)
2151 n_threads = cip->max_subcore_threads;
b009031f 2152 if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
b4deba5c 2153 return false;
b009031f 2154 cip->max_subcore_threads = n_threads;
b4deba5c
PM		 2155
2156 sub = cip->n_subcores;
2157 ++cip->n_subcores;
2158 cip->total_threads += vc->num_threads;
2159 cip->subcore_threads[sub] = vc->num_threads;
2160 cip->subcore_vm[sub] = vc->kvm;
2161 init_master_vcore(vc);
28d057c8 2162 list_move_tail(&vc->preempt_list, &cip->vcs[sub]);
b4deba5c
PM		 2163
2164 return true;
2165}
2166
b4deba5c
PM		 2167/*
2168 * Work out whether it is possible to piggyback the execution of
2169 * vcore *pvc onto the execution of the other vcores described in *cip.
2170 */
2171static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2172 int target_threads)
2173{
b4deba5c
PM		 2174 if (cip->total_threads + pvc->num_threads > target_threads)
2175 return false;
b4deba5c 2176
b009031f 2177 return can_dynamic_split(pvc, cip);
b4deba5c
PM		 2178}
2179
d911f0be
PM		 2180static void prepare_threads(struct kvmppc_vcore *vc)
2181{
7b5f8272
SJS		 2182 int i;
2183 struct kvm_vcpu *vcpu;
d911f0be 2184
7b5f8272 2185 for_each_runnable_thread(i, vcpu, vc) {
d911f0be
PM		 2186 if (signal_pending(vcpu->arch.run_task))
2187 vcpu->arch.ret = -EINTR;
2188 else if (vcpu->arch.vpa.update_pending ||
2189 vcpu->arch.slb_shadow.update_pending ||
2190 vcpu->arch.dtl.update_pending)
2191 vcpu->arch.ret = RESUME_GUEST;
2192 else
2193 continue;
2194 kvmppc_remove_runnable(vc, vcpu);
2195 wake_up(&vcpu->arch.cpu_run);
2196 }
2197}
2198
ec257165
PM		 2199static void collect_piggybacks(struct core_info *cip, int target_threads)
2200{
2201 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2202 struct kvmppc_vcore *pvc, *vcnext;
2203
2204 spin_lock(&lp->lock);
2205 list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2206 if (!spin_trylock(&pvc->lock))
2207 continue;
2208 prepare_threads(pvc);
2209 if (!pvc->n_runnable) {
2210 list_del_init(&pvc->preempt_list);
2211 if (pvc->runner == NULL) {
2212 pvc->vcore_state = VCORE_INACTIVE;
2213 kvmppc_core_end_stolen(pvc);
2214 }
2215 spin_unlock(&pvc->lock);
2216 continue;
2217 }
2218 if (!can_piggyback(pvc, cip, target_threads)) {
2219 spin_unlock(&pvc->lock);
2220 continue;
2221 }
2222 kvmppc_core_end_stolen(pvc);
2223 pvc->vcore_state = VCORE_PIGGYBACK;
2224 if (cip->total_threads >= target_threads)
2225 break;
2226 }
2227 spin_unlock(&lp->lock);
2228}
2229
2230static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
25fedfca 2231{
7b5f8272 2232 int still_running = 0, i;
25fedfca
PM		 2233 u64 now;
2234 long ret;
7b5f8272 2235 struct kvm_vcpu *vcpu;
25fedfca 2236
ec257165 2237 spin_lock(&vc->lock);
25fedfca 2238 now = get_tb();
7b5f8272 2239 for_each_runnable_thread(i, vcpu, vc) {
25fedfca
PM		 2240 /* cancel pending dec exception if dec is positive */
2241 if (now < vcpu->arch.dec_expires &&
2242 kvmppc_core_pending_dec(vcpu))
2243 kvmppc_core_dequeue_dec(vcpu);
2244
2245 trace_kvm_guest_exit(vcpu);
2246
2247 ret = RESUME_GUEST;
2248 if (vcpu->arch.trap)
2249 ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2250 vcpu->arch.run_task);
2251
2252 vcpu->arch.ret = ret;
2253 vcpu->arch.trap = 0;
2254
ec257165
PM		 2255 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2256 if (vcpu->arch.pending_exceptions)
2257 kvmppc_core_prepare_to_enter(vcpu);
2258 if (vcpu->arch.ceded)
25fedfca 2259 kvmppc_set_timer(vcpu);
ec257165
PM		 2260 else
2261 ++still_running;
2262 } else {
25fedfca
PM		 2263 kvmppc_remove_runnable(vc, vcpu);
2264 wake_up(&vcpu->arch.cpu_run);
2265 }
2266 }
ec257165
PM		 2267 list_del_init(&vc->preempt_list);
2268 if (!is_master) {
563a1e93 2269 if (still_running > 0) {
ec257165 2270 kvmppc_vcore_preempt(vc);
563a1e93
PM		 2271 } else if (vc->runner) {
2272 vc->vcore_state = VCORE_PREEMPT;
2273 kvmppc_core_start_stolen(vc);
2274 } else {
2275 vc->vcore_state = VCORE_INACTIVE;
2276 }
ec257165
PM		 2277 if (vc->n_runnable > 0 && vc->runner == NULL) {
2278 /* make sure there's a candidate runner awake */
7b5f8272
SJS		 2279 i = -1;
2280 vcpu = next_runnable_thread(vc, &i);
ec257165
PM		 2281 wake_up(&vcpu->arch.cpu_run);
2282 }
2283 }
2284 spin_unlock(&vc->lock);
25fedfca
PM		 2285}
2286
b8e6a87c
SW		 2287/*
2288 * Clear core from the list of active host cores as we are about to
2289 * enter the guest. Only do this if it is the primary thread of the
2290 * core (not if a subcore) that is entering the guest.
2291 */
2292static inline void kvmppc_clear_host_core(int cpu)
2293{
2294 int core;
2295
2296 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2297 return;
2298 /*
2299 * Memory barrier can be omitted here as we will do a smp_wmb()
2300 * later in kvmppc_start_thread and we need ensure that state is
2301 * visible to other CPUs only after we enter guest.
2302 */
2303 core = cpu >> threads_shift;
2304 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2305}
2306
2307/*
2308 * Advertise this core as an active host core since we exited the guest
2309 * Only need to do this if it is the primary thread of the core that is
2310 * exiting.
2311 */
2312static inline void kvmppc_set_host_core(int cpu)
2313{
2314 int core;
2315
2316 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2317 return;
2318
2319 /*
2320 * Memory barrier can be omitted here because we do a spin_unlock
2321 * immediately after this which provides the memory barrier.
2322 */
2323 core = cpu >> threads_shift;
2324 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2325}
2326
371fefd6
PM		 2327/*
2328 * Run a set of guest threads on a physical core.
2329 * Called with vc->lock held.
2330 */
66feed61 2331static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
371fefd6 2332{
7b5f8272 2333 struct kvm_vcpu *vcpu;
d911f0be 2334 int i;
2c9097e4 2335 int srcu_idx;
ec257165
PM		 2336 struct core_info core_info;
2337 struct kvmppc_vcore *pvc, *vcnext;
b4deba5c
PM		 2338 struct kvm_split_mode split_info, *sip;
2339 int split, subcore_size, active;
2340 int sub;
2341 bool thr0_done;
2342 unsigned long cmd_bit, stat_bit;
ec257165
PM		 2343 int pcpu, thr;
2344 int target_threads;
371fefd6 2345
d911f0be
PM		 2346 /*
2347 * Remove from the list any threads that have a signal pending
2348 * or need a VPA update done
2349 */
2350 prepare_threads(vc);
2351
2352 /* if the runner is no longer runnable, let the caller pick a new one */
2353 if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2354 return;
081f323b
PM		 2355
2356 /*
d911f0be 2357 * Initialize *vc.
081f323b 2358 */
ec257165 2359 init_master_vcore(vc);
2711e248 2360 vc->preempt_tb = TB_NIL;
081f323b 2361
7b444c67 2362 /*
3102f784
ME		 2363 * Make sure we are running on primary threads, and that secondary
2364 * threads are offline. Also check if the number of threads in this
2365 * guest are greater than the current system threads per guest.
7b444c67 2366 */
3102f784
ME		 2367 if ((threads_per_core > 1) &&
2368 ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
7b5f8272 2369 for_each_runnable_thread(i, vcpu, vc) {
7b444c67 2370 vcpu->arch.ret = -EBUSY;
25fedfca
PM		 2371 kvmppc_remove_runnable(vc, vcpu);
2372 wake_up(&vcpu->arch.cpu_run);
2373 }
7b444c67
PM		 2374 goto out;
2375 }
2376
ec257165
PM		 2377 /*
2378 * See if we could run any other vcores on the physical core
2379 * along with this one.
2380 */
2381 init_core_info(&core_info, vc);
2382 pcpu = smp_processor_id();
2383 target_threads = threads_per_subcore;
2384 if (target_smt_mode && target_smt_mode < target_threads)
2385 target_threads = target_smt_mode;
2386 if (vc->num_threads < target_threads)
2387 collect_piggybacks(&core_info, target_threads);
3102f784 2388
b4deba5c
PM		 2389 /* Decide on micro-threading (split-core) mode */
2390 subcore_size = threads_per_subcore;
2391 cmd_bit = stat_bit = 0;
2392 split = core_info.n_subcores;
2393 sip = NULL;
2394 if (split > 1) {
2395 /* threads_per_subcore must be MAX_SMT_THREADS (8) here */
2396 if (split == 2 && (dynamic_mt_modes & 2)) {
2397 cmd_bit = HID0_POWER8_1TO2LPAR;
2398 stat_bit = HID0_POWER8_2LPARMODE;
2399 } else {
2400 split = 4;
2401 cmd_bit = HID0_POWER8_1TO4LPAR;
2402 stat_bit = HID0_POWER8_4LPARMODE;
2403 }
2404 subcore_size = MAX_SMT_THREADS / split;
2405 sip = &split_info;
2406 memset(&split_info, 0, sizeof(split_info));
2407 split_info.rpr = mfspr(SPRN_RPR);
2408 split_info.pmmar = mfspr(SPRN_PMMAR);
2409 split_info.ldbar = mfspr(SPRN_LDBAR);
2410 split_info.subcore_size = subcore_size;
2411 for (sub = 0; sub < core_info.n_subcores; ++sub)
2412 split_info.master_vcs[sub] =
2413 list_first_entry(&core_info.vcs[sub],
2414 struct kvmppc_vcore, preempt_list);
2415 /* order writes to split_info before kvm_split_mode pointer */
2416 smp_wmb();
2417 }
2418 pcpu = smp_processor_id();
2419 for (thr = 0; thr < threads_per_subcore; ++thr)
2420 paca[pcpu + thr].kvm_hstate.kvm_split_mode = sip;
2421
2422 /* Initiate micro-threading (split-core) if required */
2423 if (cmd_bit) {
2424 unsigned long hid0 = mfspr(SPRN_HID0);
2425
2426 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2427 mb();
2428 mtspr(SPRN_HID0, hid0);
2429 isync();
2430 for (;;) {
2431 hid0 = mfspr(SPRN_HID0);
2432 if (hid0 & stat_bit)
2433 break;
2434 cpu_relax();
ec257165 2435 }
2e25aa5f 2436 }
3102f784 2437
b8e6a87c
SW		 2438 kvmppc_clear_host_core(pcpu);
2439
b4deba5c
PM		 2440 /* Start all the threads */
2441 active = 0;
2442 for (sub = 0; sub < core_info.n_subcores; ++sub) {
2443 thr = subcore_thread_map[sub];
2444 thr0_done = false;
2445 active |= 1 << thr;
2446 list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list) {
2447 pvc->pcpu = pcpu + thr;
7b5f8272 2448 for_each_runnable_thread(i, vcpu, pvc) {
b4deba5c
PM		 2449 kvmppc_start_thread(vcpu, pvc);
2450 kvmppc_create_dtl_entry(vcpu, pvc);
2451 trace_kvm_guest_enter(vcpu);
2452 if (!vcpu->arch.ptid)
2453 thr0_done = true;
2454 active |= 1 << (thr + vcpu->arch.ptid);
2455 }
2456 /*
2457 * We need to start the first thread of each subcore
2458 * even if it doesn't have a vcpu.
2459 */
2460 if (pvc->master_vcore == pvc && !thr0_done)
2461 kvmppc_start_thread(NULL, pvc);
2462 thr += pvc->num_threads;
2463 }
2e25aa5f 2464 }
371fefd6 2465
7f235328
GS		 2466 /*
2467 * Ensure that split_info.do_nap is set after setting
2468 * the vcore pointer in the PACA of the secondaries.
2469 */
2470 smp_mb();
2471 if (cmd_bit)
2472 split_info.do_nap = 1; /* ask secondaries to nap when done */
2473
b4deba5c
PM		 2474 /*
2475 * When doing micro-threading, poke the inactive threads as well.
2476 * This gets them to the nap instruction after kvm_do_nap,
2477 * which reduces the time taken to unsplit later.
2478 */
2479 if (split > 1)
2480 for (thr = 1; thr < threads_per_subcore; ++thr)
2481 if (!(active & (1 << thr)))
2482 kvmppc_ipi_thread(pcpu + thr);
e0b7ec05 2483
2f12f034 2484 vc->vcore_state = VCORE_RUNNING;
19ccb76a 2485 preempt_disable();
3c78f78a
SW		 2486
2487 trace_kvmppc_run_core(vc, 0);
2488
b4deba5c
PM		 2489 for (sub = 0; sub < core_info.n_subcores; ++sub)
2490 list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list)
2491 spin_unlock(&pvc->lock);
de56a948 2492
6edaa530 2493 guest_enter();
2c9097e4 2494
e0b7ec05 2495 srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2c9097e4 2496
e0b7ec05 2497 __kvmppc_vcore_entry();
de56a948 2498
ec257165
PM		 2499 srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2500
2501 spin_lock(&vc->lock);
371fefd6 2502 /* prevent other vcpu threads from doing kvmppc_start_thread() now */
19ccb76a 2503 vc->vcore_state = VCORE_EXITING;
371fefd6 2504
19ccb76a 2505 /* wait for secondary threads to finish writing their state to memory */
5d5b99cd 2506 kvmppc_wait_for_nap();
b4deba5c
PM		 2507
2508 /* Return to whole-core mode if we split the core earlier */
2509 if (split > 1) {
2510 unsigned long hid0 = mfspr(SPRN_HID0);
2511 unsigned long loops = 0;
2512
2513 hid0 &= ~HID0_POWER8_DYNLPARDIS;
2514 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
2515 mb();
2516 mtspr(SPRN_HID0, hid0);
2517 isync();
2518 for (;;) {
2519 hid0 = mfspr(SPRN_HID0);
2520 if (!(hid0 & stat_bit))
2521 break;
2522 cpu_relax();
2523 ++loops;
2524 }
2525 split_info.do_nap = 0;
2526 }
2527
2528 /* Let secondaries go back to the offline loop */
2529 for (i = 0; i < threads_per_subcore; ++i) {
2530 kvmppc_release_hwthread(pcpu + i);
2531 if (sip && sip->napped[i])
2532 kvmppc_ipi_thread(pcpu + i);
2533 }
2534
b8e6a87c
SW		 2535 kvmppc_set_host_core(pcpu);
2536
371fefd6 2537 spin_unlock(&vc->lock);
2c9097e4 2538
371fefd6
PM		 2539 /* make sure updates to secondary vcpu structs are visible now */
2540 smp_mb();
6edaa530 2541 guest_exit();
de56a948 2542
b4deba5c
PM		 2543 for (sub = 0; sub < core_info.n_subcores; ++sub)
2544 list_for_each_entry_safe(pvc, vcnext, &core_info.vcs[sub],
2545 preempt_list)
2546 post_guest_process(pvc, pvc == vc);
de56a948 2547
913d3ff9 2548 spin_lock(&vc->lock);
ec257165 2549 preempt_enable();
de56a948
PM		 2550
2551 out:
19ccb76a 2552 vc->vcore_state = VCORE_INACTIVE;
3c78f78a 2553 trace_kvmppc_run_core(vc, 1);
371fefd6
PM		 2554}
2555
19ccb76a
PM		 2556/*
2557 * Wait for some other vcpu thread to execute us, and
2558 * wake us up when we need to handle something in the host.
2559 */
ec257165
PM		 2560static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
2561 struct kvm_vcpu *vcpu, int wait_state)
371fefd6 2562{
371fefd6
PM		 2563 DEFINE_WAIT(wait);
2564
19ccb76a 2565 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
ec257165
PM		 2566 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2567 spin_unlock(&vc->lock);
19ccb76a 2568 schedule();
ec257165
PM		 2569 spin_lock(&vc->lock);
2570 }
19ccb76a
PM		 2571 finish_wait(&vcpu->arch.cpu_run, &wait);
2572}
2573
0cda69dd
SJS		 2574static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
2575{
2576 /* 10us base */
2577 if (vc->halt_poll_ns == 0 && halt_poll_ns_grow)
2578 vc->halt_poll_ns = 10000;
2579 else
2580 vc->halt_poll_ns *= halt_poll_ns_grow;
2581
2582 if (vc->halt_poll_ns > halt_poll_max_ns)
2583 vc->halt_poll_ns = halt_poll_max_ns;
2584}
2585
2586static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
2587{
2588 if (halt_poll_ns_shrink == 0)
2589 vc->halt_poll_ns = 0;
2590 else
2591 vc->halt_poll_ns /= halt_poll_ns_shrink;
2592}
2593
2594/* Check to see if any of the runnable vcpus on the vcore have pending
2595 * exceptions or are no longer ceded
2596 */
2597static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
2598{
2599 struct kvm_vcpu *vcpu;
2600 int i;
2601
2602 for_each_runnable_thread(i, vcpu, vc) {
2603 if (vcpu->arch.pending_exceptions || !vcpu->arch.ceded)
2604 return 1;
2605 }
2606
2607 return 0;
2608}
2609
19ccb76a
PM		 2610/*
2611 * All the vcpus in this vcore are idle, so wait for a decrementer
2612 * or external interrupt to one of the vcpus. vc->lock is held.
2613 */
2614static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
2615{
2a27f514 2616 ktime_t cur, start_poll, start_wait;
0cda69dd 2617 int do_sleep = 1;
0cda69dd 2618 u64 block_ns;
8577370f 2619 DECLARE_SWAITQUEUE(wait);
1bc5d59c 2620
0cda69dd 2621 /* Poll for pending exceptions and ceded state */
2a27f514 2622 cur = start_poll = ktime_get();
0cda69dd 2623 if (vc->halt_poll_ns) {
2a27f514
SJS		 2624 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
2625 ++vc->runner->stat.halt_attempted_poll;
1bc5d59c 2626
0cda69dd
SJS		 2627 vc->vcore_state = VCORE_POLLING;
2628 spin_unlock(&vc->lock);
2629
2630 do {
2631 if (kvmppc_vcore_check_block(vc)) {
2632 do_sleep = 0;
2633 break;
2634 }
2635 cur = ktime_get();
2636 } while (single_task_running() && ktime_before(cur, stop));
2637
2638 spin_lock(&vc->lock);
2639 vc->vcore_state = VCORE_INACTIVE;
2640
2a27f514
SJS		 2641 if (!do_sleep) {
2642 ++vc->runner->stat.halt_successful_poll;
0cda69dd 2643 goto out;
2a27f514 2644 }
1bc5d59c
SW		 2645 }
2646
0cda69dd
SJS		 2647 prepare_to_swait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
2648
2649 if (kvmppc_vcore_check_block(vc)) {
8577370f 2650 finish_swait(&vc->wq, &wait);
0cda69dd 2651 do_sleep = 0;
2a27f514
SJS		 2652 /* If we polled, count this as a successful poll */
2653 if (vc->halt_poll_ns)
2654 ++vc->runner->stat.halt_successful_poll;
0cda69dd 2655 goto out;
1bc5d59c
SW		 2656 }
2657
2a27f514
SJS		 2658 start_wait = ktime_get();
2659
19ccb76a 2660 vc->vcore_state = VCORE_SLEEPING;
3c78f78a 2661 trace_kvmppc_vcore_blocked(vc, 0);
19ccb76a 2662 spin_unlock(&vc->lock);
913d3ff9 2663 schedule();
8577370f 2664 finish_swait(&vc->wq, &wait);
19ccb76a
PM		 2665 spin_lock(&vc->lock);
2666 vc->vcore_state = VCORE_INACTIVE;
3c78f78a 2667 trace_kvmppc_vcore_blocked(vc, 1);
2a27f514 2668 ++vc->runner->stat.halt_successful_wait;
0cda69dd
SJS		 2669
2670 cur = ktime_get();
2671
2672out:
2a27f514
SJS		 2673 block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
2674
2675 /* Attribute wait time */
2676 if (do_sleep) {
2677 vc->runner->stat.halt_wait_ns +=
2678 ktime_to_ns(cur) - ktime_to_ns(start_wait);
2679 /* Attribute failed poll time */
2680 if (vc->halt_poll_ns)
2681 vc->runner->stat.halt_poll_fail_ns +=
2682 ktime_to_ns(start_wait) -
2683 ktime_to_ns(start_poll);
2684 } else {
2685 /* Attribute successful poll time */
2686 if (vc->halt_poll_ns)
2687 vc->runner->stat.halt_poll_success_ns +=
2688 ktime_to_ns(cur) -
2689 ktime_to_ns(start_poll);
2690 }
0cda69dd
SJS		 2691
2692 /* Adjust poll time */
2693 if (halt_poll_max_ns) {
2694 if (block_ns <= vc->halt_poll_ns)
2695 ;
2696 /* We slept and blocked for longer than the max halt time */
2697 else if (vc->halt_poll_ns && block_ns > halt_poll_max_ns)
2698 shrink_halt_poll_ns(vc);
2699 /* We slept and our poll time is too small */
2700 else if (vc->halt_poll_ns < halt_poll_max_ns &&
2701 block_ns < halt_poll_max_ns)
2702 grow_halt_poll_ns(vc);
2703 } else
2704 vc->halt_poll_ns = 0;
2705
2706 trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
19ccb76a 2707}
371fefd6 2708
19ccb76a
PM		 2709static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2710{
7b5f8272 2711 int n_ceded, i;
19ccb76a 2712 struct kvmppc_vcore *vc;
7b5f8272 2713 struct kvm_vcpu *v;
9e368f29 2714
3c78f78a
SW		 2715 trace_kvmppc_run_vcpu_enter(vcpu);
2716
371fefd6
PM		 2717 kvm_run->exit_reason = 0;
2718 vcpu->arch.ret = RESUME_GUEST;
2719 vcpu->arch.trap = 0;
2f12f034 2720 kvmppc_update_vpas(vcpu);
371fefd6 2721
371fefd6
PM		 2722 /*
2723 * Synchronize with other threads in this virtual core
2724 */
2725 vc = vcpu->arch.vcore;
2726 spin_lock(&vc->lock);
19ccb76a 2727 vcpu->arch.ceded = 0;
371fefd6
PM		 2728 vcpu->arch.run_task = current;
2729 vcpu->arch.kvm_run = kvm_run;
c7b67670 2730 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
19ccb76a 2731 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
c7b67670 2732 vcpu->arch.busy_preempt = TB_NIL;
7b5f8272 2733 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
371fefd6
PM		 2734 ++vc->n_runnable;
2735
19ccb76a
PM		 2736 /*
2737 * This happens the first time this is called for a vcpu.
2738 * If the vcore is already running, we may be able to start
2739 * this thread straight away and have it join in.
2740 */
8455d79e 2741 if (!signal_pending(current)) {
ec257165
PM		 2742 if (vc->vcore_state == VCORE_PIGGYBACK) {
2743 struct kvmppc_vcore *mvc = vc->master_vcore;
2744 if (spin_trylock(&mvc->lock)) {
2745 if (mvc->vcore_state == VCORE_RUNNING &&
2746 !VCORE_IS_EXITING(mvc)) {
2747 kvmppc_create_dtl_entry(vcpu, vc);
b4deba5c 2748 kvmppc_start_thread(vcpu, vc);
ec257165
PM		 2749 trace_kvm_guest_enter(vcpu);
2750 }
2751 spin_unlock(&mvc->lock);
2752 }
2753 } else if (vc->vcore_state == VCORE_RUNNING &&
2754 !VCORE_IS_EXITING(vc)) {
2f12f034 2755 kvmppc_create_dtl_entry(vcpu, vc);
b4deba5c 2756 kvmppc_start_thread(vcpu, vc);
3c78f78a 2757 trace_kvm_guest_enter(vcpu);
8455d79e 2758 } else if (vc->vcore_state == VCORE_SLEEPING) {
8577370f 2759 swake_up(&vc->wq);
371fefd6
PM		 2760 }
2761
8455d79e 2762 }
371fefd6 2763
19ccb76a
PM		 2764 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2765 !signal_pending(current)) {
ec257165
PM		 2766 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2767 kvmppc_vcore_end_preempt(vc);
2768
8455d79e 2769 if (vc->vcore_state != VCORE_INACTIVE) {
ec257165 2770 kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
19ccb76a
PM		 2771 continue;
2772 }
7b5f8272 2773 for_each_runnable_thread(i, v, vc) {
7e28e60e 2774 kvmppc_core_prepare_to_enter(v);
19ccb76a
PM		 2775 if (signal_pending(v->arch.run_task)) {
2776 kvmppc_remove_runnable(vc, v);
2777 v->stat.signal_exits++;
2778 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
2779 v->arch.ret = -EINTR;
2780 wake_up(&v->arch.cpu_run);
2781 }
2782 }
8455d79e
PM		 2783 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2784 break;
8455d79e 2785 n_ceded = 0;
7b5f8272 2786 for_each_runnable_thread(i, v, vc) {
8455d79e
PM		 2787 if (!v->arch.pending_exceptions)
2788 n_ceded += v->arch.ceded;
4619ac88
PM		 2789 else
2790 v->arch.ceded = 0;
2791 }
25fedfca
PM		 2792 vc->runner = vcpu;
2793 if (n_ceded == vc->n_runnable) {
8455d79e 2794 kvmppc_vcore_blocked(vc);
c56dadf3 2795 } else if (need_resched()) {
ec257165 2796 kvmppc_vcore_preempt(vc);
25fedfca
PM		 2797 /* Let something else run */
2798 cond_resched_lock(&vc->lock);
ec257165
PM		 2799 if (vc->vcore_state == VCORE_PREEMPT)
2800 kvmppc_vcore_end_preempt(vc);
25fedfca 2801 } else {
8455d79e 2802 kvmppc_run_core(vc);
25fedfca 2803 }
0456ec4f 2804 vc->runner = NULL;
19ccb76a 2805 }
371fefd6 2806
8455d79e
PM		 2807 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2808 (vc->vcore_state == VCORE_RUNNING ||
5fc3e64f
PM		 2809 vc->vcore_state == VCORE_EXITING ||
2810 vc->vcore_state == VCORE_PIGGYBACK))
ec257165 2811 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
8455d79e 2812
5fc3e64f
PM		 2813 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2814 kvmppc_vcore_end_preempt(vc);
2815
8455d79e
PM		 2816 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2817 kvmppc_remove_runnable(vc, vcpu);
2818 vcpu->stat.signal_exits++;
2819 kvm_run->exit_reason = KVM_EXIT_INTR;
2820 vcpu->arch.ret = -EINTR;
2821 }
2822
2823 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
2824 /* Wake up some vcpu to run the core */
7b5f8272
SJS		 2825 i = -1;
2826 v = next_runnable_thread(vc, &i);
8455d79e 2827 wake_up(&v->arch.cpu_run);
371fefd6
PM		 2828 }
2829
3c78f78a 2830 trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
371fefd6 2831 spin_unlock(&vc->lock);
371fefd6 2832 return vcpu->arch.ret;
de56a948
PM		 2833}
2834
3a167bea 2835static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
a8606e20
PM		 2836{
2837 int r;
913d3ff9 2838 int srcu_idx;
a8606e20 2839
af8f38b3
AG		 2840 if (!vcpu->arch.sane) {
2841 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2842 return -EINVAL;
2843 }
2844
25051b5a
SW		 2845 kvmppc_core_prepare_to_enter(vcpu);
2846
19ccb76a
PM		 2847 /* No need to go into the guest when all we'll do is come back out */
2848 if (signal_pending(current)) {
2849 run->exit_reason = KVM_EXIT_INTR;
2850 return -EINTR;
2851 }
2852
32fad281 2853 atomic_inc(&vcpu->kvm->arch.vcpus_running);
31037eca 2854 /* Order vcpus_running vs. hpte_setup_done, see kvmppc_alloc_reset_hpt */
32fad281
PM		 2855 smp_mb();
2856
c17b98cf 2857 /* On the first time here, set up HTAB and VRMA */
31037eca 2858 if (!vcpu->kvm->arch.hpte_setup_done) {
32fad281 2859 r = kvmppc_hv_setup_htab_rma(vcpu);
c77162de 2860 if (r)
32fad281 2861 goto out;
c77162de 2862 }
19ccb76a 2863
579e633e
AB		 2864 flush_all_to_thread(current);
2865
19ccb76a 2866 vcpu->arch.wqp = &vcpu->arch.vcore->wq;
342d3db7 2867 vcpu->arch.pgdir = current->mm->pgd;
c7b67670 2868 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
19ccb76a 2869
a8606e20
PM		 2870 do {
2871 r = kvmppc_run_vcpu(run, vcpu);
2872
2873 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
2874 !(vcpu->arch.shregs.msr & MSR_PR)) {
3c78f78a 2875 trace_kvm_hcall_enter(vcpu);
a8606e20 2876 r = kvmppc_pseries_do_hcall(vcpu);
3c78f78a 2877 trace_kvm_hcall_exit(vcpu, r);
7e28e60e 2878 kvmppc_core_prepare_to_enter(vcpu);
913d3ff9
PM		 2879 } else if (r == RESUME_PAGE_FAULT) {
2880 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2881 r = kvmppc_book3s_hv_page_fault(run, vcpu,
2882 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
2883 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
f7af5209
SW		 2884 } else if (r == RESUME_PASSTHROUGH)
2885 r = kvmppc_xics_rm_complete(vcpu, 0);
e59d24e6 2886 } while (is_kvmppc_resume_guest(r));
32fad281
PM		 2887
2888 out:
c7b67670 2889 vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
32fad281 2890 atomic_dec(&vcpu->kvm->arch.vcpus_running);
a8606e20
PM		 2891 return r;
2892}
2893
5b74716e
BH		 2894static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
2895 int linux_psize)
2896{
2897 struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
2898
2899 if (!def->shift)
2900 return;
2901 (*sps)->page_shift = def->shift;
2902 (*sps)->slb_enc = def->sllp;
2903 (*sps)->enc[0].page_shift = def->shift;
b1022fbd 2904 (*sps)->enc[0].pte_enc = def->penc[linux_psize];
1f365bb0
AK		 2905 /*
2906 * Add 16MB MPSS support if host supports it
2907 */
2908 if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
2909 (*sps)->enc[1].page_shift = 24;
2910 (*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
2911 }
5b74716e
BH		 2912 (*sps)++;
2913}
2914
3a167bea
AK		 2915static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
2916 struct kvm_ppc_smmu_info *info)
5b74716e
BH		 2917{
2918 struct kvm_ppc_one_seg_page_size *sps;
2919
2920 info->flags = KVM_PPC_PAGE_SIZES_REAL;
2921 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
2922 info->flags |= KVM_PPC_1T_SEGMENTS;
2923 info->slb_size = mmu_slb_size;
2924
2925 /* We only support these sizes for now, and no muti-size segments */
2926 sps = &info->sps[0];
2927 kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
2928 kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
2929 kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
2930
2931 return 0;
2932}
2933
82ed3616
PM		 2934/*
2935 * Get (and clear) the dirty memory log for a memory slot.
2936 */
3a167bea
AK		 2937static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
2938 struct kvm_dirty_log *log)
82ed3616 2939{
9f6b8029 2940 struct kvm_memslots *slots;
82ed3616
PM		 2941 struct kvm_memory_slot *memslot;
2942 int r;
2943 unsigned long n;
2944
2945 mutex_lock(&kvm->slots_lock);
2946
2947 r = -EINVAL;
bbacc0c1 2948 if (log->slot >= KVM_USER_MEM_SLOTS)
82ed3616
PM		 2949 goto out;
2950
9f6b8029
PB		 2951 slots = kvm_memslots(kvm);
2952 memslot = id_to_memslot(slots, log->slot);
82ed3616
PM		 2953 r = -ENOENT;
2954 if (!memslot->dirty_bitmap)
2955 goto out;
2956
2957 n = kvm_dirty_bitmap_bytes(memslot);
2958 memset(memslot->dirty_bitmap, 0, n);
2959
dfe49dbd 2960 r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
82ed3616
PM		 2961 if (r)
2962 goto out;
2963
2964 r = -EFAULT;
2965 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
2966 goto out;
2967
2968 r = 0;
2969out:
2970 mutex_unlock(&kvm->slots_lock);
2971 return r;
2972}
2973
3a167bea
AK		 2974static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
2975 struct kvm_memory_slot *dont)
a66b48c3
PM		 2976{
2977 if (!dont || free->arch.rmap != dont->arch.rmap) {
2978 vfree(free->arch.rmap);
2979 free->arch.rmap = NULL;
b2b2f165 2980 }
a66b48c3
PM		 2981}
2982
3a167bea
AK		 2983static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
2984 unsigned long npages)
a66b48c3
PM		 2985{
2986 slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
2987 if (!slot->arch.rmap)
2988 return -ENOMEM;
aa04b4cc 2989
c77162de
PM		 2990 return 0;
2991}
aa04b4cc 2992
3a167bea
AK		 2993static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
2994 struct kvm_memory_slot *memslot,
09170a49 2995 const struct kvm_userspace_memory_region *mem)
c77162de 2996{
a66b48c3 2997 return 0;
c77162de
PM		 2998}
2999
3a167bea 3000static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
09170a49 3001 const struct kvm_userspace_memory_region *mem,
f36f3f28
PB		 3002 const struct kvm_memory_slot *old,
3003 const struct kvm_memory_slot *new)
c77162de 3004{
dfe49dbd 3005 unsigned long npages = mem->memory_size >> PAGE_SHIFT;
9f6b8029 3006 struct kvm_memslots *slots;
dfe49dbd
PM		 3007 struct kvm_memory_slot *memslot;
3008
a56ee9f8
YX		 3009 /*
3010 * If we are making a new memslot, it might make
3011 * some address that was previously cached as emulated
3012 * MMIO be no longer emulated MMIO, so invalidate
3013 * all the caches of emulated MMIO translations.
3014 */
3015 if (npages)
3016 atomic64_inc(&kvm->arch.mmio_update);
3017
8482644a 3018 if (npages && old->npages) {
dfe49dbd
PM		 3019 /*
3020 * If modifying a memslot, reset all the rmap dirty bits.
3021 * If this is a new memslot, we don't need to do anything
3022 * since the rmap array starts out as all zeroes,
3023 * i.e. no pages are dirty.
3024 */
9f6b8029
PB		 3025 slots = kvm_memslots(kvm);
3026 memslot = id_to_memslot(slots, mem->slot);
dfe49dbd
PM		 3027 kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
3028 }
c77162de
PM		 3029}
3030
a0144e2a
PM		 3031/*
3032 * Update LPCR values in kvm->arch and in vcores.
3033 * Caller must hold kvm->lock.
3034 */
3035void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
3036{
3037 long int i;
3038 u32 cores_done = 0;
3039
3040 if ((kvm->arch.lpcr & mask) == lpcr)
3041 return;
3042
3043 kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
3044
3045 for (i = 0; i < KVM_MAX_VCORES; ++i) {
3046 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
3047 if (!vc)
3048 continue;
3049 spin_lock(&vc->lock);
3050 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
3051 spin_unlock(&vc->lock);
3052 if (++cores_done >= kvm->arch.online_vcores)
3053 break;
3054 }
3055}
3056
3a167bea
AK		 3057static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
3058{
3059 return;
3060}
3061
7a84084c
PM		 3062static void kvmppc_setup_partition_table(struct kvm *kvm)
3063{
3064 unsigned long dw0, dw1;
3065
3066 /* PS field - page size for VRMA */
3067 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
3068 ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
3069 /* HTABSIZE and HTABORG fields */
3070 dw0 |= kvm->arch.sdr1;
3071
3072 /* Second dword has GR=0; other fields are unused since UPRT=0 */
3073 dw1 = 0;
3074
3075 mmu_partition_table_set_entry(kvm->arch.lpid, dw0, dw1);
3076}
3077
32fad281 3078static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
c77162de
PM		 3079{
3080 int err = 0;
3081 struct kvm *kvm = vcpu->kvm;
c77162de
PM		 3082 unsigned long hva;
3083 struct kvm_memory_slot *memslot;
3084 struct vm_area_struct *vma;
a0144e2a 3085 unsigned long lpcr = 0, senc;
c77162de 3086 unsigned long psize, porder;
2c9097e4 3087 int srcu_idx;
c77162de
PM		 3088
3089 mutex_lock(&kvm->lock);
31037eca 3090 if (kvm->arch.hpte_setup_done)
c77162de 3091 goto out; /* another vcpu beat us to it */
aa04b4cc 3092
32fad281
PM		 3093 /* Allocate hashed page table (if not done already) and reset it */
3094 if (!kvm->arch.hpt_virt) {
3095 err = kvmppc_alloc_hpt(kvm, NULL);
3096 if (err) {
3097 pr_err("KVM: Couldn't alloc HPT\n");
3098 goto out;
3099 }
3100 }
3101
c77162de 3102 /* Look up the memslot for guest physical address 0 */
2c9097e4 3103 srcu_idx = srcu_read_lock(&kvm->srcu);
c77162de 3104 memslot = gfn_to_memslot(kvm, 0);
aa04b4cc 3105
c77162de
PM		 3106 /* We must have some memory at 0 by now */
3107 err = -EINVAL;
3108 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
2c9097e4 3109 goto out_srcu;
c77162de
PM		 3110
3111 /* Look up the VMA for the start of this memory slot */
3112 hva = memslot->userspace_addr;
3113 down_read(&current->mm->mmap_sem);
3114 vma = find_vma(current->mm, hva);
3115 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
3116 goto up_out;
3117
3118 psize = vma_kernel_pagesize(vma);
da9d1d7f 3119 porder = __ilog2(psize);
c77162de 3120
c77162de
PM		 3121 up_read(&current->mm->mmap_sem);
3122
c17b98cf
PM		 3123 /* We can handle 4k, 64k or 16M pages in the VRMA */
3124 err = -EINVAL;
3125 if (!(psize == 0x1000 || psize == 0x10000 ||
3126 psize == 0x1000000))
3127 goto out_srcu;
c77162de 3128
c17b98cf
PM		 3129 senc = slb_pgsize_encoding(psize);
3130 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
3131 (VRMA_VSID << SLB_VSID_SHIFT_1T);
c17b98cf
PM		 3132 /* Create HPTEs in the hash page table for the VRMA */
3133 kvmppc_map_vrma(vcpu, memslot, porder);
aa04b4cc 3134
7a84084c
PM		 3135 /* Update VRMASD field in the LPCR */
3136 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
3137 /* the -4 is to account for senc values starting at 0x10 */
3138 lpcr = senc << (LPCR_VRMASD_SH - 4);
3139 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
3140 } else {
3141 kvmppc_setup_partition_table(kvm);
3142 }
a0144e2a 3143
31037eca 3144 /* Order updates to kvm->arch.lpcr etc. vs. hpte_setup_done */
c77162de 3145 smp_wmb();
31037eca 3146 kvm->arch.hpte_setup_done = 1;
c77162de 3147 err = 0;
2c9097e4
PM		 3148 out_srcu:
3149 srcu_read_unlock(&kvm->srcu, srcu_idx);
c77162de
PM		 3150 out:
3151 mutex_unlock(&kvm->lock);
3152 return err;
b2b2f165 3153
c77162de
PM		 3154 up_out:
3155 up_read(&current->mm->mmap_sem);
505d6421 3156 goto out_srcu;
de56a948
PM		 3157}
3158
79b6c247 3159#ifdef CONFIG_KVM_XICS
6f3bb809
SW		 3160static int kvmppc_cpu_notify(struct notifier_block *self, unsigned long action,
3161 void *hcpu)
3162{
3163 unsigned long cpu = (long)hcpu;
3164
3165 switch (action) {
3166 case CPU_UP_PREPARE:
3167 case CPU_UP_PREPARE_FROZEN:
3168 kvmppc_set_host_core(cpu);
3169 break;
3170
3171#ifdef CONFIG_HOTPLUG_CPU
3172 case CPU_DEAD:
3173 case CPU_DEAD_FROZEN:
3174 case CPU_UP_CANCELED:
3175 case CPU_UP_CANCELED_FROZEN:
3176 kvmppc_clear_host_core(cpu);
3177 break;
3178#endif
3179 default:
3180 break;
3181 }
3182
3183 return NOTIFY_OK;
3184}
3185
3186static struct notifier_block kvmppc_cpu_notifier = {
3187 .notifier_call = kvmppc_cpu_notify,
3188};
3189
79b6c247
SW		 3190/*
3191 * Allocate a per-core structure for managing state about which cores are
3192 * running in the host versus the guest and for exchanging data between
3193 * real mode KVM and CPU running in the host.
3194 * This is only done for the first VM.
3195 * The allocated structure stays even if all VMs have stopped.
3196 * It is only freed when the kvm-hv module is unloaded.
3197 * It's OK for this routine to fail, we just don't support host
3198 * core operations like redirecting H_IPI wakeups.
3199 */
3200void kvmppc_alloc_host_rm_ops(void)
3201{
3202 struct kvmppc_host_rm_ops *ops;
3203 unsigned long l_ops;
3204 int cpu, core;
3205 int size;
3206
3207 /* Not the first time here ? */
3208 if (kvmppc_host_rm_ops_hv != NULL)
3209 return;
3210
3211 ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
3212 if (!ops)
3213 return;
3214
3215 size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
3216 ops->rm_core = kzalloc(size, GFP_KERNEL);
3217
3218 if (!ops->rm_core) {
3219 kfree(ops);
3220 return;
3221 }
3222
6f3bb809
SW		 3223 get_online_cpus();
3224
79b6c247
SW		 3225 for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
3226 if (!cpu_online(cpu))
3227 continue;
3228
3229 core = cpu >> threads_shift;
3230 ops->rm_core[core].rm_state.in_host = 1;
3231 }
3232
0c2a6606
SW		 3233 ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
3234
79b6c247
SW		 3235 /*
3236 * Make the contents of the kvmppc_host_rm_ops structure visible
3237 * to other CPUs before we assign it to the global variable.
3238 * Do an atomic assignment (no locks used here), but if someone
3239 * beats us to it, just free our copy and return.
3240 */
3241 smp_wmb();
3242 l_ops = (unsigned long) ops;
3243
3244 if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
6f3bb809 3245 put_online_cpus();
79b6c247
SW		 3246 kfree(ops->rm_core);
3247 kfree(ops);
6f3bb809 3248 return;
79b6c247 3249 }
6f3bb809
SW		 3250
3251 register_cpu_notifier(&kvmppc_cpu_notifier);
3252
3253 put_online_cpus();
79b6c247
SW		 3254}
3255
3256void kvmppc_free_host_rm_ops(void)
3257{
3258 if (kvmppc_host_rm_ops_hv) {
6f3bb809 3259 unregister_cpu_notifier(&kvmppc_cpu_notifier);
79b6c247
SW		 3260 kfree(kvmppc_host_rm_ops_hv->rm_core);
3261 kfree(kvmppc_host_rm_ops_hv);
3262 kvmppc_host_rm_ops_hv = NULL;
3263 }
3264}
3265#endif
3266
3a167bea 3267static int kvmppc_core_init_vm_hv(struct kvm *kvm)
de56a948 3268{
32fad281 3269 unsigned long lpcr, lpid;
e23a808b 3270 char buf[32];
de56a948 3271
32fad281
PM		 3272 /* Allocate the guest's logical partition ID */
3273
3274 lpid = kvmppc_alloc_lpid();
5d226ae5 3275 if ((long)lpid < 0)
32fad281
PM		 3276 return -ENOMEM;
3277 kvm->arch.lpid = lpid;
de56a948 3278
79b6c247
SW		 3279 kvmppc_alloc_host_rm_ops();
3280
1b400ba0
PM		 3281 /*
3282 * Since we don't flush the TLB when tearing down a VM,
3283 * and this lpid might have previously been used,
3284 * make sure we flush on each core before running the new VM.
7c5b06ca
PM		 3285 * On POWER9, the tlbie in mmu_partition_table_set_entry()
3286 * does this flush for us.
1b400ba0 3287 */
7c5b06ca
PM		 3288 if (!cpu_has_feature(CPU_FTR_ARCH_300))
3289 cpumask_setall(&kvm->arch.need_tlb_flush);
1b400ba0 3290
699a0ea0
PM		 3291 /* Start out with the default set of hcalls enabled */
3292 memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
3293 sizeof(kvm->arch.enabled_hcalls));
3294
7a84084c
PM		 3295 if (!cpu_has_feature(CPU_FTR_ARCH_300))
3296 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
aa04b4cc 3297
c17b98cf
PM		 3298 /* Init LPCR for virtual RMA mode */
3299 kvm->arch.host_lpid = mfspr(SPRN_LPID);
3300 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
3301 lpcr &= LPCR_PECE | LPCR_LPES;
3302 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3303 LPCR_VPM0 | LPCR_VPM1;
3304 kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3305 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3306 /* On POWER8 turn on online bit to enable PURR/SPURR */
3307 if (cpu_has_feature(CPU_FTR_ARCH_207S))
3308 lpcr |= LPCR_ONL;
84f7139c
PM		 3309 /*
3310 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
3311 * Set HVICE bit to enable hypervisor virtualization interrupts.
3312 */
3313 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
7a84084c 3314 lpcr &= ~LPCR_VPM0;
84f7139c
PM		 3315 lpcr |= LPCR_HVICE;
3316 }
3317
9e368f29 3318 kvm->arch.lpcr = lpcr;
aa04b4cc 3319
7c5b06ca
PM		 3320 /*
3321 * Work out how many sets the TLB has, for the use of
3322 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
3323 */
3324 if (cpu_has_feature(CPU_FTR_ARCH_300))
3325 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH; /* 256 */
3326 else if (cpu_has_feature(CPU_FTR_ARCH_207S))
3327 kvm->arch.tlb_sets = POWER8_TLB_SETS; /* 512 */
3328 else
3329 kvm->arch.tlb_sets = POWER7_TLB_SETS; /* 128 */
3330
512691d4 3331 /*
441c19c8
ME		 3332 * Track that we now have a HV mode VM active. This blocks secondary
3333 * CPU threads from coming online.
512691d4 3334 */
441c19c8 3335 kvm_hv_vm_activated();
512691d4 3336
e23a808b
PM		 3337 /*
3338 * Create a debugfs directory for the VM
3339 */
3340 snprintf(buf, sizeof(buf), "vm%d", current->pid);
3341 kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
3342 if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
3343 kvmppc_mmu_debugfs_init(kvm);
3344
54738c09 3345 return 0;
de56a948
PM		 3346}
3347
f1378b1c
PM		 3348static void kvmppc_free_vcores(struct kvm *kvm)
3349{
3350 long int i;
3351
23316316 3352 for (i = 0; i < KVM_MAX_VCORES; ++i)
f1378b1c
PM		 3353 kfree(kvm->arch.vcores[i]);
3354 kvm->arch.online_vcores = 0;
3355}
3356
3a167bea 3357static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
de56a948 3358{
e23a808b
PM		 3359 debugfs_remove_recursive(kvm->arch.debugfs_dir);
3360
441c19c8 3361 kvm_hv_vm_deactivated();
512691d4 3362
f1378b1c 3363 kvmppc_free_vcores(kvm);
aa04b4cc 3364
de56a948 3365 kvmppc_free_hpt(kvm);
c57875f5
SW		 3366
3367 kvmppc_free_pimap(kvm);
de56a948
PM		 3368}
3369
3a167bea
AK		 3370/* We don't need to emulate any privileged instructions or dcbz */
3371static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
3372 unsigned int inst, int *advance)
de56a948 3373{
3a167bea 3374 return EMULATE_FAIL;
de56a948
PM		 3375}
3376
3a167bea
AK		 3377static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
3378 ulong spr_val)
de56a948
PM		 3379{
3380 return EMULATE_FAIL;
3381}
3382
3a167bea
AK		 3383static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
3384 ulong *spr_val)
de56a948
PM		 3385{
3386 return EMULATE_FAIL;
3387}
3388
3a167bea 3389static int kvmppc_core_check_processor_compat_hv(void)
de56a948 3390{
c17b98cf
PM		 3391 if (!cpu_has_feature(CPU_FTR_HVMODE) ||
3392 !cpu_has_feature(CPU_FTR_ARCH_206))
3a167bea 3393 return -EIO;
50de596d
AK		 3394 /*
3395 * Disable KVM for Power9, untill the required bits merged.
3396 */
3397 if (cpu_has_feature(CPU_FTR_ARCH_300))
3398 return -EIO;
3399
3a167bea 3400 return 0;
de56a948
PM		 3401}
3402
8daaafc8
SW		 3403#ifdef CONFIG_KVM_XICS
3404
3405void kvmppc_free_pimap(struct kvm *kvm)
3406{
3407 kfree(kvm->arch.pimap);
3408}
3409
c57875f5 3410static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
8daaafc8
SW		 3411{
3412 return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
3413}
c57875f5
SW		 3414
3415static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
3416{
3417 struct irq_desc *desc;
3418 struct kvmppc_irq_map *irq_map;
3419 struct kvmppc_passthru_irqmap *pimap;
3420 struct irq_chip *chip;
3421 int i;
3422
644abbb2
SW		 3423 if (!kvm_irq_bypass)
3424 return 1;
3425
c57875f5
SW		 3426 desc = irq_to_desc(host_irq);
3427 if (!desc)
3428 return -EIO;
3429
3430 mutex_lock(&kvm->lock);
3431
3432 pimap = kvm->arch.pimap;
3433 if (pimap == NULL) {
3434 /* First call, allocate structure to hold IRQ map */
3435 pimap = kvmppc_alloc_pimap();
3436 if (pimap == NULL) {
3437 mutex_unlock(&kvm->lock);
3438 return -ENOMEM;
3439 }
3440 kvm->arch.pimap = pimap;
3441 }
3442
3443 /*
3444 * For now, we only support interrupts for which the EOI operation
3445 * is an OPAL call followed by a write to XIRR, since that's
3446 * what our real-mode EOI code does.
3447 */
3448 chip = irq_data_get_irq_chip(&desc->irq_data);
3449 if (!chip || !is_pnv_opal_msi(chip)) {
3450 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
3451 host_irq, guest_gsi);
3452 mutex_unlock(&kvm->lock);
3453 return -ENOENT;
3454 }
3455
3456 /*
3457 * See if we already have an entry for this guest IRQ number.
3458 * If it's mapped to a hardware IRQ number, that's an error,
3459 * otherwise re-use this entry.
3460 */
3461 for (i = 0; i < pimap->n_mapped; i++) {
3462 if (guest_gsi == pimap->mapped[i].v_hwirq) {
3463 if (pimap->mapped[i].r_hwirq) {
3464 mutex_unlock(&kvm->lock);
3465 return -EINVAL;
3466 }
3467 break;
3468 }
3469 }
3470
3471 if (i == KVMPPC_PIRQ_MAPPED) {
3472 mutex_unlock(&kvm->lock);
3473 return -EAGAIN; /* table is full */
3474 }
3475
3476 irq_map = &pimap->mapped[i];
3477
3478 irq_map->v_hwirq = guest_gsi;
c57875f5
SW		 3479 irq_map->desc = desc;
3480
e3c13e56
SW		 3481 /*
3482 * Order the above two stores before the next to serialize with
3483 * the KVM real mode handler.
3484 */
3485 smp_wmb();
3486 irq_map->r_hwirq = desc->irq_data.hwirq;
3487
c57875f5
SW		 3488 if (i == pimap->n_mapped)
3489 pimap->n_mapped++;
3490
5d375199
PM		 3491 kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
3492
c57875f5
SW		 3493 mutex_unlock(&kvm->lock);
3494
3495 return 0;
3496}
3497
3498static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
3499{
3500 struct irq_desc *desc;
3501 struct kvmppc_passthru_irqmap *pimap;
3502 int i;
3503
644abbb2
SW		 3504 if (!kvm_irq_bypass)
3505 return 0;
3506
c57875f5
SW		 3507 desc = irq_to_desc(host_irq);
3508 if (!desc)
3509 return -EIO;
3510
3511 mutex_lock(&kvm->lock);
3512
3513 if (kvm->arch.pimap == NULL) {
3514 mutex_unlock(&kvm->lock);
3515 return 0;
3516 }
3517 pimap = kvm->arch.pimap;
3518
3519 for (i = 0; i < pimap->n_mapped; i++) {
3520 if (guest_gsi == pimap->mapped[i].v_hwirq)
3521 break;
3522 }
3523
3524 if (i == pimap->n_mapped) {
3525 mutex_unlock(&kvm->lock);
3526 return -ENODEV;
3527 }
3528
5d375199
PM		 3529 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
3530
c57875f5
SW		 3531 /* invalidate the entry */
3532 pimap->mapped[i].r_hwirq = 0;
3533
3534 /*
3535 * We don't free this structure even when the count goes to
3536 * zero. The structure is freed when we destroy the VM.
3537 */
3538
3539 mutex_unlock(&kvm->lock);
3540 return 0;
3541}
3542
3543static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
3544 struct irq_bypass_producer *prod)
3545{
3546 int ret = 0;
3547 struct kvm_kernel_irqfd *irqfd =
3548 container_of(cons, struct kvm_kernel_irqfd, consumer);
3549
3550 irqfd->producer = prod;
3551
3552 ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
3553 if (ret)
3554 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
3555 prod->irq, irqfd->gsi, ret);
3556
3557 return ret;
3558}
3559
3560static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
3561 struct irq_bypass_producer *prod)
3562{
3563 int ret;
3564 struct kvm_kernel_irqfd *irqfd =
3565 container_of(cons, struct kvm_kernel_irqfd, consumer);
3566
3567 irqfd->producer = NULL;
3568
3569 /*
3570 * When producer of consumer is unregistered, we change back to
3571 * default external interrupt handling mode - KVM real mode
3572 * will switch back to host.
3573 */
3574 ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
3575 if (ret)
3576 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
3577 prod->irq, irqfd->gsi, ret);
3578}
8daaafc8
SW		 3579#endif
3580
3a167bea
AK		 3581static long kvm_arch_vm_ioctl_hv(struct file *filp,
3582 unsigned int ioctl, unsigned long arg)
3583{
3584 struct kvm *kvm __maybe_unused = filp->private_data;
3585 void __user *argp = (void __user *)arg;
3586 long r;
3587
3588 switch (ioctl) {
3589
3a167bea
AK		 3590 case KVM_PPC_ALLOCATE_HTAB: {
3591 u32 htab_order;
3592
3593 r = -EFAULT;
3594 if (get_user(htab_order, (u32 __user *)argp))
3595 break;
3596 r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
3597 if (r)
3598 break;
3599 r = -EFAULT;
3600 if (put_user(htab_order, (u32 __user *)argp))
3601 break;
3602 r = 0;
3603 break;
3604 }
3605
3606 case KVM_PPC_GET_HTAB_FD: {
3607 struct kvm_get_htab_fd ghf;
3608
3609 r = -EFAULT;
3610 if (copy_from_user(&ghf, argp, sizeof(ghf)))
3611 break;
3612 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
3613 break;
3614 }
3615
3616 default:
3617 r = -ENOTTY;
3618 }
3619
3620 return r;
3621}
3622
699a0ea0
PM		 3623/*
3624 * List of hcall numbers to enable by default.
3625 * For compatibility with old userspace, we enable by default
3626 * all hcalls that were implemented before the hcall-enabling
3627 * facility was added. Note this list should not include H_RTAS.
3628 */
3629static unsigned int default_hcall_list[] = {
3630 H_REMOVE,
3631 H_ENTER,
3632 H_READ,
3633 H_PROTECT,
3634 H_BULK_REMOVE,
3635 H_GET_TCE,
3636 H_PUT_TCE,
3637 H_SET_DABR,
3638 H_SET_XDABR,
3639 H_CEDE,
3640 H_PROD,
3641 H_CONFER,
3642 H_REGISTER_VPA,
3643#ifdef CONFIG_KVM_XICS
3644 H_EOI,
3645 H_CPPR,
3646 H_IPI,
3647 H_IPOLL,
3648 H_XIRR,
3649 H_XIRR_X,
3650#endif
3651 0
3652};
3653
3654static void init_default_hcalls(void)
3655{
3656 int i;
ae2113a4 3657 unsigned int hcall;
699a0ea0 3658
ae2113a4
PM		 3659 for (i = 0; default_hcall_list[i]; ++i) {
3660 hcall = default_hcall_list[i];
3661 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
3662 __set_bit(hcall / 4, default_enabled_hcalls);
3663 }
699a0ea0
PM		 3664}
3665
cbbc58d4 3666static struct kvmppc_ops kvm_ops_hv = {
3a167bea
AK		 3667 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
3668 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
3669 .get_one_reg = kvmppc_get_one_reg_hv,
3670 .set_one_reg = kvmppc_set_one_reg_hv,
3671 .vcpu_load = kvmppc_core_vcpu_load_hv,
3672 .vcpu_put = kvmppc_core_vcpu_put_hv,
3673 .set_msr = kvmppc_set_msr_hv,
3674 .vcpu_run = kvmppc_vcpu_run_hv,
3675 .vcpu_create = kvmppc_core_vcpu_create_hv,
3676 .vcpu_free = kvmppc_core_vcpu_free_hv,
3677 .check_requests = kvmppc_core_check_requests_hv,
3678 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv,
3679 .flush_memslot = kvmppc_core_flush_memslot_hv,
3680 .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
3681 .commit_memory_region = kvmppc_core_commit_memory_region_hv,
3682 .unmap_hva = kvm_unmap_hva_hv,
3683 .unmap_hva_range = kvm_unmap_hva_range_hv,
3684 .age_hva = kvm_age_hva_hv,
3685 .test_age_hva = kvm_test_age_hva_hv,
3686 .set_spte_hva = kvm_set_spte_hva_hv,
3687 .mmu_destroy = kvmppc_mmu_destroy_hv,
3688 .free_memslot = kvmppc_core_free_memslot_hv,
3689 .create_memslot = kvmppc_core_create_memslot_hv,
3690 .init_vm = kvmppc_core_init_vm_hv,
3691 .destroy_vm = kvmppc_core_destroy_vm_hv,
3a167bea
AK		 3692 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
3693 .emulate_op = kvmppc_core_emulate_op_hv,
3694 .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
3695 .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
3696 .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
3697 .arch_vm_ioctl = kvm_arch_vm_ioctl_hv,
ae2113a4 3698 .hcall_implemented = kvmppc_hcall_impl_hv,
c57875f5
SW		 3699#ifdef CONFIG_KVM_XICS
3700 .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
3701 .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
3702#endif
3a167bea
AK		 3703};
3704
fd7bacbc
MS		 3705static int kvm_init_subcore_bitmap(void)
3706{
3707 int i, j;
3708 int nr_cores = cpu_nr_cores();
3709 struct sibling_subcore_state *sibling_subcore_state;
3710
3711 for (i = 0; i < nr_cores; i++) {
3712 int first_cpu = i * threads_per_core;
3713 int node = cpu_to_node(first_cpu);
3714
3715 /* Ignore if it is already allocated. */
3716 if (paca[first_cpu].sibling_subcore_state)
3717 continue;
3718
3719 sibling_subcore_state =
3720 kmalloc_node(sizeof(struct sibling_subcore_state),
3721 GFP_KERNEL, node);
3722 if (!sibling_subcore_state)
3723 return -ENOMEM;
3724
3725 memset(sibling_subcore_state, 0,
3726 sizeof(struct sibling_subcore_state));
3727
3728 for (j = 0; j < threads_per_core; j++) {
3729 int cpu = first_cpu + j;
3730
3731 paca[cpu].sibling_subcore_state = sibling_subcore_state;
3732 }
3733 }
3734 return 0;
3735}
3736
3a167bea 3737static int kvmppc_book3s_init_hv(void)
de56a948
PM		 3738{
3739 int r;
cbbc58d4
AK		 3740 /*
3741 * FIXME!! Do we need to check on all cpus ?
3742 */
3743 r = kvmppc_core_check_processor_compat_hv();
3744 if (r < 0)
739e2425 3745 return -ENODEV;
de56a948 3746
fd7bacbc
MS		 3747 r = kvm_init_subcore_bitmap();
3748 if (r)
3749 return r;
3750
f725758b
PM		 3751 /*
3752 * We need a way of accessing the XICS interrupt controller,
3753 * either directly, via paca[cpu].kvm_hstate.xics_phys, or
3754 * indirectly, via OPAL.
3755 */
3756#ifdef CONFIG_SMP
3757 if (!get_paca()->kvm_hstate.xics_phys) {
3758 struct device_node *np;
3759
3760 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
3761 if (!np) {
3762 pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
3763 return -ENODEV;
3764 }
3765 }
3766#endif
3767
cbbc58d4
AK		 3768 kvm_ops_hv.owner = THIS_MODULE;
3769 kvmppc_hv_ops = &kvm_ops_hv;
de56a948 3770
699a0ea0
PM		 3771 init_default_hcalls();
3772
ec257165
PM		 3773 init_vcore_lists();
3774
cbbc58d4 3775 r = kvmppc_mmu_hv_init();
de56a948
PM		 3776 return r;
3777}
3778
3a167bea 3779static void kvmppc_book3s_exit_hv(void)
de56a948 3780{
79b6c247 3781 kvmppc_free_host_rm_ops();
cbbc58d4 3782 kvmppc_hv_ops = NULL;
de56a948
PM		 3783}
3784
3a167bea
AK		 3785module_init(kvmppc_book3s_init_hv);
3786module_exit(kvmppc_book3s_exit_hv);
2ba9f0d8 3787MODULE_LICENSE("GPL");
398a76c6
AG		 3788MODULE_ALIAS_MISCDEV(KVM_MINOR);
3789MODULE_ALIAS("devname:kvm");
7c5b06ca 3790
Ubuntu Artful kernel mirror