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