xref: /linux/arch/powerpc/kvm/book3s_hv.c (revision b4db9f840283caca0d904436f187ef56a9126eaa)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
4  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
5  *
6  * Authors:
7  *    Paul Mackerras <paulus@au1.ibm.com>
8  *    Alexander Graf <agraf@suse.de>
9  *    Kevin Wolf <mail@kevin-wolf.de>
10  *
11  * Description: KVM functions specific to running on Book 3S
12  * processors in hypervisor mode (specifically POWER7 and later).
13  *
14  * This file is derived from arch/powerpc/kvm/book3s.c,
15  * by Alexander Graf <agraf@suse.de>.
16  */
17 
18 #include <linux/kvm_host.h>
19 #include <linux/kernel.h>
20 #include <linux/err.h>
21 #include <linux/slab.h>
22 #include <linux/preempt.h>
23 #include <linux/sched/signal.h>
24 #include <linux/sched/stat.h>
25 #include <linux/delay.h>
26 #include <linux/export.h>
27 #include <linux/fs.h>
28 #include <linux/anon_inodes.h>
29 #include <linux/cpu.h>
30 #include <linux/cpumask.h>
31 #include <linux/spinlock.h>
32 #include <linux/page-flags.h>
33 #include <linux/srcu.h>
34 #include <linux/miscdevice.h>
35 #include <linux/debugfs.h>
36 #include <linux/gfp.h>
37 #include <linux/vmalloc.h>
38 #include <linux/highmem.h>
39 #include <linux/hugetlb.h>
40 #include <linux/kvm_irqfd.h>
41 #include <linux/irqbypass.h>
42 #include <linux/module.h>
43 #include <linux/compiler.h>
44 #include <linux/of.h>
45 #include <linux/irqdomain.h>
46 #include <linux/smp.h>
47 
48 #include <asm/ftrace.h>
49 #include <asm/reg.h>
50 #include <asm/ppc-opcode.h>
51 #include <asm/asm-prototypes.h>
52 #include <asm/archrandom.h>
53 #include <asm/debug.h>
54 #include <asm/disassemble.h>
55 #include <asm/cputable.h>
56 #include <asm/cacheflush.h>
57 #include <linux/uaccess.h>
58 #include <asm/interrupt.h>
59 #include <asm/io.h>
60 #include <asm/kvm_ppc.h>
61 #include <asm/kvm_book3s.h>
62 #include <asm/mmu_context.h>
63 #include <asm/lppaca.h>
64 #include <asm/pmc.h>
65 #include <asm/processor.h>
66 #include <asm/cputhreads.h>
67 #include <asm/page.h>
68 #include <asm/hvcall.h>
69 #include <asm/switch_to.h>
70 #include <asm/smp.h>
71 #include <asm/dbell.h>
72 #include <asm/hmi.h>
73 #include <asm/pnv-pci.h>
74 #include <asm/mmu.h>
75 #include <asm/opal.h>
76 #include <asm/xics.h>
77 #include <asm/xive.h>
78 #include <asm/hw_breakpoint.h>
79 #include <asm/kvm_book3s_uvmem.h>
80 #include <asm/ultravisor.h>
81 #include <asm/dtl.h>
82 #include <asm/plpar_wrappers.h>
83 
84 #include <trace/events/ipi.h>
85 
86 #include "book3s.h"
87 #include "book3s_hv.h"
88 
89 #define CREATE_TRACE_POINTS
90 #include "trace_hv.h"
91 
92 /* #define EXIT_DEBUG */
93 /* #define EXIT_DEBUG_SIMPLE */
94 /* #define EXIT_DEBUG_INT */
95 
96 /* Used to indicate that a guest page fault needs to be handled */
97 #define RESUME_PAGE_FAULT	(RESUME_GUEST | RESUME_FLAG_ARCH1)
98 /* Used to indicate that a guest passthrough interrupt needs to be handled */
99 #define RESUME_PASSTHROUGH	(RESUME_GUEST | RESUME_FLAG_ARCH2)
100 
101 /* Used as a "null" value for timebase values */
102 #define TB_NIL	(~(u64)0)
103 
104 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
105 
106 static int dynamic_mt_modes = 6;
107 module_param(dynamic_mt_modes, int, 0644);
108 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
109 static int target_smt_mode;
110 module_param(target_smt_mode, int, 0644);
111 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
112 
113 static bool one_vm_per_core;
114 module_param(one_vm_per_core, bool, S_IRUGO | S_IWUSR);
115 MODULE_PARM_DESC(one_vm_per_core, "Only run vCPUs from the same VM on a core (requires POWER8 or older)");
116 
117 #ifdef CONFIG_KVM_XICS
118 static const struct kernel_param_ops module_param_ops = {
119 	.set = param_set_int,
120 	.get = param_get_int,
121 };
122 
123 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
124 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
125 
126 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
127 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
128 #endif
129 
130 /* If set, guests are allowed to create and control nested guests */
131 static bool nested = true;
132 module_param(nested, bool, S_IRUGO | S_IWUSR);
133 MODULE_PARM_DESC(nested, "Enable nested virtualization (only on POWER9)");
134 
135 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
136 
137 /*
138  * RWMR values for POWER8.  These control the rate at which PURR
139  * and SPURR count and should be set according to the number of
140  * online threads in the vcore being run.
141  */
142 #define RWMR_RPA_P8_1THREAD	0x164520C62609AECAUL
143 #define RWMR_RPA_P8_2THREAD	0x7FFF2908450D8DA9UL
144 #define RWMR_RPA_P8_3THREAD	0x164520C62609AECAUL
145 #define RWMR_RPA_P8_4THREAD	0x199A421245058DA9UL
146 #define RWMR_RPA_P8_5THREAD	0x164520C62609AECAUL
147 #define RWMR_RPA_P8_6THREAD	0x164520C62609AECAUL
148 #define RWMR_RPA_P8_7THREAD	0x164520C62609AECAUL
149 #define RWMR_RPA_P8_8THREAD	0x164520C62609AECAUL
150 
151 static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = {
152 	RWMR_RPA_P8_1THREAD,
153 	RWMR_RPA_P8_1THREAD,
154 	RWMR_RPA_P8_2THREAD,
155 	RWMR_RPA_P8_3THREAD,
156 	RWMR_RPA_P8_4THREAD,
157 	RWMR_RPA_P8_5THREAD,
158 	RWMR_RPA_P8_6THREAD,
159 	RWMR_RPA_P8_7THREAD,
160 	RWMR_RPA_P8_8THREAD,
161 };
162 
163 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
164 		int *ip)
165 {
166 	int i = *ip;
167 	struct kvm_vcpu *vcpu;
168 
169 	while (++i < MAX_SMT_THREADS) {
170 		vcpu = READ_ONCE(vc->runnable_threads[i]);
171 		if (vcpu) {
172 			*ip = i;
173 			return vcpu;
174 		}
175 	}
176 	return NULL;
177 }
178 
179 /* Used to traverse the list of runnable threads for a given vcore */
180 #define for_each_runnable_thread(i, vcpu, vc) \
181 	for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
182 
183 static bool kvmppc_ipi_thread(int cpu)
184 {
185 	unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
186 
187 	/* If we're a nested hypervisor, fall back to ordinary IPIs for now */
188 	if (kvmhv_on_pseries())
189 		return false;
190 
191 	/* On POWER9 we can use msgsnd to IPI any cpu */
192 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
193 		msg |= get_hard_smp_processor_id(cpu);
194 		smp_mb();
195 		__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
196 		return true;
197 	}
198 
199 	/* On POWER8 for IPIs to threads in the same core, use msgsnd */
200 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
201 		preempt_disable();
202 		if (cpu_first_thread_sibling(cpu) ==
203 		    cpu_first_thread_sibling(smp_processor_id())) {
204 			msg |= cpu_thread_in_core(cpu);
205 			smp_mb();
206 			__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
207 			preempt_enable();
208 			return true;
209 		}
210 		preempt_enable();
211 	}
212 
213 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
214 	if (cpu >= 0 && cpu < nr_cpu_ids) {
215 		if (paca_ptrs[cpu]->kvm_hstate.xics_phys) {
216 			xics_wake_cpu(cpu);
217 			return true;
218 		}
219 		opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
220 		return true;
221 	}
222 #endif
223 
224 	return false;
225 }
226 
227 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
228 {
229 	int cpu;
230 	struct rcuwait *waitp;
231 
232 	/*
233 	 * rcuwait_wake_up contains smp_mb() which orders prior stores that
234 	 * create pending work vs below loads of cpu fields. The other side
235 	 * is the barrier in vcpu run that orders setting the cpu fields vs
236 	 * testing for pending work.
237 	 */
238 
239 	waitp = kvm_arch_vcpu_get_wait(vcpu);
240 	if (rcuwait_wake_up(waitp))
241 		++vcpu->stat.generic.halt_wakeup;
242 
243 	cpu = READ_ONCE(vcpu->arch.thread_cpu);
244 	if (cpu >= 0 && kvmppc_ipi_thread(cpu))
245 		return;
246 
247 	/* CPU points to the first thread of the core */
248 	cpu = vcpu->cpu;
249 	if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
250 		smp_send_reschedule(cpu);
251 }
252 
253 /*
254  * We use the vcpu_load/put functions to measure stolen time.
255  *
256  * Stolen time is counted as time when either the vcpu is able to
257  * run as part of a virtual core, but the task running the vcore
258  * is preempted or sleeping, or when the vcpu needs something done
259  * in the kernel by the task running the vcpu, but that task is
260  * preempted or sleeping.  Those two things have to be counted
261  * separately, since one of the vcpu tasks will take on the job
262  * of running the core, and the other vcpu tasks in the vcore will
263  * sleep waiting for it to do that, but that sleep shouldn't count
264  * as stolen time.
265  *
266  * Hence we accumulate stolen time when the vcpu can run as part of
267  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
268  * needs its task to do other things in the kernel (for example,
269  * service a page fault) in busy_stolen.  We don't accumulate
270  * stolen time for a vcore when it is inactive, or for a vcpu
271  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
272  * a misnomer; it means that the vcpu task is not executing in
273  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
274  * the kernel.  We don't have any way of dividing up that time
275  * between time that the vcpu is genuinely stopped, time that
276  * the task is actively working on behalf of the vcpu, and time
277  * that the task is preempted, so we don't count any of it as
278  * stolen.
279  *
280  * Updates to busy_stolen are protected by arch.tbacct_lock;
281  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
282  * lock.  The stolen times are measured in units of timebase ticks.
283  * (Note that the != TB_NIL checks below are purely defensive;
284  * they should never fail.)
285  *
286  * The POWER9 path is simpler, one vcpu per virtual core so the
287  * former case does not exist. If a vcpu is preempted when it is
288  * BUSY_IN_HOST and not ceded or otherwise blocked, then accumulate
289  * the stolen cycles in busy_stolen. RUNNING is not a preemptible
290  * state in the P9 path.
291  */
292 
293 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc, u64 tb)
294 {
295 	unsigned long flags;
296 
297 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
298 
299 	spin_lock_irqsave(&vc->stoltb_lock, flags);
300 	vc->preempt_tb = tb;
301 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
302 }
303 
304 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc, u64 tb)
305 {
306 	unsigned long flags;
307 
308 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
309 
310 	spin_lock_irqsave(&vc->stoltb_lock, flags);
311 	if (vc->preempt_tb != TB_NIL) {
312 		vc->stolen_tb += tb - vc->preempt_tb;
313 		vc->preempt_tb = TB_NIL;
314 	}
315 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
316 }
317 
318 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
319 {
320 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
321 	unsigned long flags;
322 	u64 now;
323 
324 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
325 		if (vcpu->arch.busy_preempt != TB_NIL) {
326 			WARN_ON_ONCE(vcpu->arch.state != KVMPPC_VCPU_BUSY_IN_HOST);
327 			vc->stolen_tb += mftb() - vcpu->arch.busy_preempt;
328 			vcpu->arch.busy_preempt = TB_NIL;
329 		}
330 		return;
331 	}
332 
333 	now = mftb();
334 
335 	/*
336 	 * We can test vc->runner without taking the vcore lock,
337 	 * because only this task ever sets vc->runner to this
338 	 * vcpu, and once it is set to this vcpu, only this task
339 	 * ever sets it to NULL.
340 	 */
341 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
342 		kvmppc_core_end_stolen(vc, now);
343 
344 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
345 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
346 	    vcpu->arch.busy_preempt != TB_NIL) {
347 		vcpu->arch.busy_stolen += now - vcpu->arch.busy_preempt;
348 		vcpu->arch.busy_preempt = TB_NIL;
349 	}
350 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
351 }
352 
353 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
354 {
355 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
356 	unsigned long flags;
357 	u64 now;
358 
359 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
360 		/*
361 		 * In the P9 path, RUNNABLE is not preemptible
362 		 * (nor takes host interrupts)
363 		 */
364 		WARN_ON_ONCE(vcpu->arch.state == KVMPPC_VCPU_RUNNABLE);
365 		/*
366 		 * Account stolen time when preempted while the vcpu task is
367 		 * running in the kernel (but not in qemu, which is INACTIVE).
368 		 */
369 		if (task_is_running(current) &&
370 				vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
371 			vcpu->arch.busy_preempt = mftb();
372 		return;
373 	}
374 
375 	now = mftb();
376 
377 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
378 		kvmppc_core_start_stolen(vc, now);
379 
380 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
381 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
382 		vcpu->arch.busy_preempt = now;
383 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
384 }
385 
386 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
387 {
388 	vcpu->arch.pvr = pvr;
389 }
390 
391 /* Dummy value used in computing PCR value below */
392 #define PCR_ARCH_31    (PCR_ARCH_300 << 1)
393 
394 static inline unsigned long map_pcr_to_cap(unsigned long pcr)
395 {
396 	unsigned long cap = 0;
397 
398 	switch (pcr) {
399 	case PCR_ARCH_300:
400 		cap = H_GUEST_CAP_POWER9;
401 		break;
402 	case PCR_ARCH_31:
403 		cap = H_GUEST_CAP_POWER10;
404 		break;
405 	default:
406 		break;
407 	}
408 
409 	return cap;
410 }
411 
412 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
413 {
414 	unsigned long host_pcr_bit = 0, guest_pcr_bit = 0, cap = 0;
415 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
416 
417 	/* We can (emulate) our own architecture version and anything older */
418 	if (cpu_has_feature(CPU_FTR_ARCH_31))
419 		host_pcr_bit = PCR_ARCH_31;
420 	else if (cpu_has_feature(CPU_FTR_ARCH_300))
421 		host_pcr_bit = PCR_ARCH_300;
422 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
423 		host_pcr_bit = PCR_ARCH_207;
424 	else if (cpu_has_feature(CPU_FTR_ARCH_206))
425 		host_pcr_bit = PCR_ARCH_206;
426 	else
427 		host_pcr_bit = PCR_ARCH_205;
428 
429 	/* Determine lowest PCR bit needed to run guest in given PVR level */
430 	guest_pcr_bit = host_pcr_bit;
431 	if (arch_compat) {
432 		switch (arch_compat) {
433 		case PVR_ARCH_205:
434 			guest_pcr_bit = PCR_ARCH_205;
435 			break;
436 		case PVR_ARCH_206:
437 		case PVR_ARCH_206p:
438 			guest_pcr_bit = PCR_ARCH_206;
439 			break;
440 		case PVR_ARCH_207:
441 			guest_pcr_bit = PCR_ARCH_207;
442 			break;
443 		case PVR_ARCH_300:
444 			guest_pcr_bit = PCR_ARCH_300;
445 			break;
446 		case PVR_ARCH_31:
447 		case PVR_ARCH_31_P11:
448 			guest_pcr_bit = PCR_ARCH_31;
449 			break;
450 		default:
451 			return -EINVAL;
452 		}
453 	}
454 
455 	/* Check requested PCR bits don't exceed our capabilities */
456 	if (guest_pcr_bit > host_pcr_bit)
457 		return -EINVAL;
458 
459 	if (kvmhv_on_pseries() && kvmhv_is_nestedv2()) {
460 		/*
461 		 * 'arch_compat == 0' would mean the guest should default to
462 		 * L1's compatibility. In this case, the guest would pick
463 		 * host's PCR and evaluate the corresponding capabilities.
464 		 */
465 		cap = map_pcr_to_cap(guest_pcr_bit);
466 		if (!(cap & nested_capabilities))
467 			return -EINVAL;
468 	}
469 
470 	spin_lock(&vc->lock);
471 	vc->arch_compat = arch_compat;
472 	kvmhv_nestedv2_mark_dirty(vcpu, KVMPPC_GSID_LOGICAL_PVR);
473 	/*
474 	 * Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit
475 	 * Also set all reserved PCR bits
476 	 */
477 	vc->pcr = (host_pcr_bit - guest_pcr_bit) | PCR_MASK;
478 	spin_unlock(&vc->lock);
479 
480 	return 0;
481 }
482 
483 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
484 {
485 	int r;
486 
487 	pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
488 	pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
489 	       vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
490 	for (r = 0; r < 16; ++r)
491 		pr_err("r%2d = %.16lx  r%d = %.16lx\n",
492 		       r, kvmppc_get_gpr(vcpu, r),
493 		       r+16, kvmppc_get_gpr(vcpu, r+16));
494 	pr_err("ctr = %.16lx  lr  = %.16lx\n",
495 	       vcpu->arch.regs.ctr, vcpu->arch.regs.link);
496 	pr_err("srr0 = %.16llx srr1 = %.16llx\n",
497 	       vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
498 	pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
499 	       vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
500 	pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
501 	       vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
502 	pr_err("cr = %.8lx  xer = %.16lx  dsisr = %.8x\n",
503 	       vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
504 	pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
505 	pr_err("fault dar = %.16lx dsisr = %.8x\n",
506 	       vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
507 	pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
508 	for (r = 0; r < vcpu->arch.slb_max; ++r)
509 		pr_err("  ESID = %.16llx VSID = %.16llx\n",
510 		       vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
511 	pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.16lx\n",
512 	       vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
513 	       vcpu->arch.last_inst);
514 }
515 
516 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
517 {
518 	return kvm_get_vcpu_by_id(kvm, id);
519 }
520 
521 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
522 {
523 	vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
524 	vpa->yield_count = cpu_to_be32(1);
525 }
526 
527 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
528 		   unsigned long addr, unsigned long len)
529 {
530 	/* check address is cacheline aligned */
531 	if (addr & (L1_CACHE_BYTES - 1))
532 		return -EINVAL;
533 	spin_lock(&vcpu->arch.vpa_update_lock);
534 	if (v->next_gpa != addr || v->len != len) {
535 		v->next_gpa = addr;
536 		v->len = addr ? len : 0;
537 		v->update_pending = 1;
538 	}
539 	spin_unlock(&vcpu->arch.vpa_update_lock);
540 	return 0;
541 }
542 
543 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
544 struct reg_vpa {
545 	u32 dummy;
546 	union {
547 		__be16 hword;
548 		__be32 word;
549 	} length;
550 };
551 
552 static int vpa_is_registered(struct kvmppc_vpa *vpap)
553 {
554 	if (vpap->update_pending)
555 		return vpap->next_gpa != 0;
556 	return vpap->pinned_addr != NULL;
557 }
558 
559 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
560 				       unsigned long flags,
561 				       unsigned long vcpuid, unsigned long vpa)
562 {
563 	struct kvm *kvm = vcpu->kvm;
564 	unsigned long len, nb;
565 	void *va;
566 	struct kvm_vcpu *tvcpu;
567 	int err;
568 	int subfunc;
569 	struct kvmppc_vpa *vpap;
570 
571 	tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
572 	if (!tvcpu)
573 		return H_PARAMETER;
574 
575 	subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
576 	if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
577 	    subfunc == H_VPA_REG_SLB) {
578 		/* Registering new area - address must be cache-line aligned */
579 		if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
580 			return H_PARAMETER;
581 
582 		/* convert logical addr to kernel addr and read length */
583 		va = kvmppc_pin_guest_page(kvm, vpa, &nb);
584 		if (va == NULL)
585 			return H_PARAMETER;
586 		if (subfunc == H_VPA_REG_VPA)
587 			len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
588 		else
589 			len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
590 		kvmppc_unpin_guest_page(kvm, va, vpa, false);
591 
592 		/* Check length */
593 		if (len > nb || len < sizeof(struct reg_vpa))
594 			return H_PARAMETER;
595 	} else {
596 		vpa = 0;
597 		len = 0;
598 	}
599 
600 	err = H_PARAMETER;
601 	vpap = NULL;
602 	spin_lock(&tvcpu->arch.vpa_update_lock);
603 
604 	switch (subfunc) {
605 	case H_VPA_REG_VPA:		/* register VPA */
606 		/*
607 		 * The size of our lppaca is 1kB because of the way we align
608 		 * it for the guest to avoid crossing a 4kB boundary. We only
609 		 * use 640 bytes of the structure though, so we should accept
610 		 * clients that set a size of 640.
611 		 */
612 		BUILD_BUG_ON(sizeof(struct lppaca) != 640);
613 		if (len < sizeof(struct lppaca))
614 			break;
615 		vpap = &tvcpu->arch.vpa;
616 		err = 0;
617 		break;
618 
619 	case H_VPA_REG_DTL:		/* register DTL */
620 		if (len < sizeof(struct dtl_entry))
621 			break;
622 		len -= len % sizeof(struct dtl_entry);
623 
624 		/* Check that they have previously registered a VPA */
625 		err = H_RESOURCE;
626 		if (!vpa_is_registered(&tvcpu->arch.vpa))
627 			break;
628 
629 		vpap = &tvcpu->arch.dtl;
630 		err = 0;
631 		break;
632 
633 	case H_VPA_REG_SLB:		/* register SLB shadow buffer */
634 		/* Check that they have previously registered a VPA */
635 		err = H_RESOURCE;
636 		if (!vpa_is_registered(&tvcpu->arch.vpa))
637 			break;
638 
639 		vpap = &tvcpu->arch.slb_shadow;
640 		err = 0;
641 		break;
642 
643 	case H_VPA_DEREG_VPA:		/* deregister VPA */
644 		/* Check they don't still have a DTL or SLB buf registered */
645 		err = H_RESOURCE;
646 		if (vpa_is_registered(&tvcpu->arch.dtl) ||
647 		    vpa_is_registered(&tvcpu->arch.slb_shadow))
648 			break;
649 
650 		vpap = &tvcpu->arch.vpa;
651 		err = 0;
652 		break;
653 
654 	case H_VPA_DEREG_DTL:		/* deregister DTL */
655 		vpap = &tvcpu->arch.dtl;
656 		err = 0;
657 		break;
658 
659 	case H_VPA_DEREG_SLB:		/* deregister SLB shadow buffer */
660 		vpap = &tvcpu->arch.slb_shadow;
661 		err = 0;
662 		break;
663 	}
664 
665 	if (vpap) {
666 		vpap->next_gpa = vpa;
667 		vpap->len = len;
668 		vpap->update_pending = 1;
669 	}
670 
671 	spin_unlock(&tvcpu->arch.vpa_update_lock);
672 
673 	return err;
674 }
675 
676 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap,
677 			       struct kvmppc_vpa *old_vpap)
678 {
679 	struct kvm *kvm = vcpu->kvm;
680 	void *va;
681 	unsigned long nb;
682 	unsigned long gpa;
683 
684 	/*
685 	 * We need to pin the page pointed to by vpap->next_gpa,
686 	 * but we can't call kvmppc_pin_guest_page under the lock
687 	 * as it does get_user_pages() and down_read().  So we
688 	 * have to drop the lock, pin the page, then get the lock
689 	 * again and check that a new area didn't get registered
690 	 * in the meantime.
691 	 */
692 	for (;;) {
693 		gpa = vpap->next_gpa;
694 		spin_unlock(&vcpu->arch.vpa_update_lock);
695 		va = NULL;
696 		nb = 0;
697 		if (gpa)
698 			va = kvmppc_pin_guest_page(kvm, gpa, &nb);
699 		spin_lock(&vcpu->arch.vpa_update_lock);
700 		if (gpa == vpap->next_gpa)
701 			break;
702 		/* sigh... unpin that one and try again */
703 		if (va)
704 			kvmppc_unpin_guest_page(kvm, va, gpa, false);
705 	}
706 
707 	vpap->update_pending = 0;
708 	if (va && nb < vpap->len) {
709 		/*
710 		 * If it's now too short, it must be that userspace
711 		 * has changed the mappings underlying guest memory,
712 		 * so unregister the region.
713 		 */
714 		kvmppc_unpin_guest_page(kvm, va, gpa, false);
715 		va = NULL;
716 	}
717 	*old_vpap = *vpap;
718 
719 	vpap->gpa = gpa;
720 	vpap->pinned_addr = va;
721 	vpap->dirty = false;
722 	if (va)
723 		vpap->pinned_end = va + vpap->len;
724 }
725 
726 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
727 {
728 	struct kvm *kvm = vcpu->kvm;
729 	struct kvmppc_vpa old_vpa = { 0 };
730 
731 	if (!(vcpu->arch.vpa.update_pending ||
732 	      vcpu->arch.slb_shadow.update_pending ||
733 	      vcpu->arch.dtl.update_pending))
734 		return;
735 
736 	spin_lock(&vcpu->arch.vpa_update_lock);
737 	if (vcpu->arch.vpa.update_pending) {
738 		kvmppc_update_vpa(vcpu, &vcpu->arch.vpa, &old_vpa);
739 		if (old_vpa.pinned_addr) {
740 			if (kvmhv_is_nestedv2())
741 				kvmhv_nestedv2_set_vpa(vcpu, ~0ull);
742 			kvmppc_unpin_guest_page(kvm, old_vpa.pinned_addr, old_vpa.gpa,
743 						old_vpa.dirty);
744 		}
745 		if (vcpu->arch.vpa.pinned_addr) {
746 			init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
747 			if (kvmhv_is_nestedv2())
748 				kvmhv_nestedv2_set_vpa(vcpu, __pa(vcpu->arch.vpa.pinned_addr));
749 		}
750 	}
751 	if (vcpu->arch.dtl.update_pending) {
752 		kvmppc_update_vpa(vcpu, &vcpu->arch.dtl, &old_vpa);
753 		if (old_vpa.pinned_addr)
754 			kvmppc_unpin_guest_page(kvm, old_vpa.pinned_addr, old_vpa.gpa,
755 						old_vpa.dirty);
756 		vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
757 		vcpu->arch.dtl_index = 0;
758 	}
759 	if (vcpu->arch.slb_shadow.update_pending) {
760 		kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow, &old_vpa);
761 		if (old_vpa.pinned_addr)
762 			kvmppc_unpin_guest_page(kvm, old_vpa.pinned_addr, old_vpa.gpa,
763 						old_vpa.dirty);
764 	}
765 
766 	spin_unlock(&vcpu->arch.vpa_update_lock);
767 }
768 
769 /*
770  * Return the accumulated stolen time for the vcore up until `now'.
771  * The caller should hold the vcore lock.
772  */
773 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
774 {
775 	u64 p;
776 	unsigned long flags;
777 
778 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
779 
780 	spin_lock_irqsave(&vc->stoltb_lock, flags);
781 	p = vc->stolen_tb;
782 	if (vc->vcore_state != VCORE_INACTIVE &&
783 	    vc->preempt_tb != TB_NIL)
784 		p += now - vc->preempt_tb;
785 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
786 	return p;
787 }
788 
789 static void __kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
790 					struct lppaca *vpa,
791 					unsigned int pcpu, u64 now,
792 					unsigned long stolen)
793 {
794 	struct dtl_entry *dt;
795 
796 	dt = vcpu->arch.dtl_ptr;
797 
798 	if (!dt)
799 		return;
800 
801 	dt->dispatch_reason = 7;
802 	dt->preempt_reason = 0;
803 	dt->processor_id = cpu_to_be16(pcpu + vcpu->arch.ptid);
804 	dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
805 	dt->ready_to_enqueue_time = 0;
806 	dt->waiting_to_ready_time = 0;
807 	dt->timebase = cpu_to_be64(now);
808 	dt->fault_addr = 0;
809 	dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
810 	dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
811 
812 	++dt;
813 	if (dt == vcpu->arch.dtl.pinned_end)
814 		dt = vcpu->arch.dtl.pinned_addr;
815 	vcpu->arch.dtl_ptr = dt;
816 	/* order writing *dt vs. writing vpa->dtl_idx */
817 	smp_wmb();
818 	vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
819 
820 	/* vcpu->arch.dtl.dirty is set by the caller */
821 }
822 
823 static void kvmppc_update_vpa_dispatch(struct kvm_vcpu *vcpu,
824 				       struct kvmppc_vcore *vc)
825 {
826 	struct lppaca *vpa;
827 	unsigned long stolen;
828 	unsigned long core_stolen;
829 	u64 now;
830 	unsigned long flags;
831 
832 	vpa = vcpu->arch.vpa.pinned_addr;
833 	if (!vpa)
834 		return;
835 
836 	now = mftb();
837 
838 	core_stolen = vcore_stolen_time(vc, now);
839 	stolen = core_stolen - vcpu->arch.stolen_logged;
840 	vcpu->arch.stolen_logged = core_stolen;
841 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
842 	stolen += vcpu->arch.busy_stolen;
843 	vcpu->arch.busy_stolen = 0;
844 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
845 
846 	vpa->enqueue_dispatch_tb = cpu_to_be64(be64_to_cpu(vpa->enqueue_dispatch_tb) + stolen);
847 
848 	__kvmppc_create_dtl_entry(vcpu, vpa, vc->pcpu, now + kvmppc_get_tb_offset(vcpu), stolen);
849 
850 	vcpu->arch.vpa.dirty = true;
851 }
852 
853 static void kvmppc_update_vpa_dispatch_p9(struct kvm_vcpu *vcpu,
854 				       struct kvmppc_vcore *vc,
855 				       u64 now)
856 {
857 	struct lppaca *vpa;
858 	unsigned long stolen;
859 	unsigned long stolen_delta;
860 
861 	vpa = vcpu->arch.vpa.pinned_addr;
862 	if (!vpa)
863 		return;
864 
865 	stolen = vc->stolen_tb;
866 	stolen_delta = stolen - vcpu->arch.stolen_logged;
867 	vcpu->arch.stolen_logged = stolen;
868 
869 	vpa->enqueue_dispatch_tb = cpu_to_be64(stolen);
870 
871 	__kvmppc_create_dtl_entry(vcpu, vpa, vc->pcpu, now, stolen_delta);
872 
873 	vcpu->arch.vpa.dirty = true;
874 }
875 
876 /* See if there is a doorbell interrupt pending for a vcpu */
877 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
878 {
879 	int thr;
880 	struct kvmppc_vcore *vc;
881 
882 	if (vcpu->arch.doorbell_request)
883 		return true;
884 	if (cpu_has_feature(CPU_FTR_ARCH_300))
885 		return false;
886 	/*
887 	 * Ensure that the read of vcore->dpdes comes after the read
888 	 * of vcpu->doorbell_request.  This barrier matches the
889 	 * smp_wmb() in kvmppc_guest_entry_inject().
890 	 */
891 	smp_rmb();
892 	vc = vcpu->arch.vcore;
893 	thr = vcpu->vcpu_id - vc->first_vcpuid;
894 	return !!(vc->dpdes & (1 << thr));
895 }
896 
897 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
898 {
899 	if (kvmppc_get_arch_compat(vcpu) >= PVR_ARCH_207)
900 		return true;
901 	if ((!kvmppc_get_arch_compat(vcpu)) &&
902 	    cpu_has_feature(CPU_FTR_ARCH_207S))
903 		return true;
904 	return false;
905 }
906 
907 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
908 			     unsigned long resource, unsigned long value1,
909 			     unsigned long value2)
910 {
911 	switch (resource) {
912 	case H_SET_MODE_RESOURCE_SET_CIABR:
913 		if (!kvmppc_power8_compatible(vcpu))
914 			return H_P2;
915 		if (value2)
916 			return H_P4;
917 		if (mflags)
918 			return H_UNSUPPORTED_FLAG_START;
919 		/* Guests can't breakpoint the hypervisor */
920 		if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
921 			return H_P3;
922 		kvmppc_set_ciabr_hv(vcpu, value1);
923 		return H_SUCCESS;
924 	case H_SET_MODE_RESOURCE_SET_DAWR0:
925 		if (!kvmppc_power8_compatible(vcpu))
926 			return H_P2;
927 		if (!ppc_breakpoint_available())
928 			return H_P2;
929 		if (mflags)
930 			return H_UNSUPPORTED_FLAG_START;
931 		if (value2 & DABRX_HYP)
932 			return H_P4;
933 		kvmppc_set_dawr0_hv(vcpu, value1);
934 		kvmppc_set_dawrx0_hv(vcpu, value2);
935 		return H_SUCCESS;
936 	case H_SET_MODE_RESOURCE_SET_DAWR1:
937 		if (!kvmppc_power8_compatible(vcpu))
938 			return H_P2;
939 		if (!ppc_breakpoint_available())
940 			return H_P2;
941 		if (!cpu_has_feature(CPU_FTR_DAWR1))
942 			return H_P2;
943 		if (!vcpu->kvm->arch.dawr1_enabled)
944 			return H_FUNCTION;
945 		if (mflags)
946 			return H_UNSUPPORTED_FLAG_START;
947 		if (value2 & DABRX_HYP)
948 			return H_P4;
949 		kvmppc_set_dawr1_hv(vcpu, value1);
950 		kvmppc_set_dawrx1_hv(vcpu, value2);
951 		return H_SUCCESS;
952 	case H_SET_MODE_RESOURCE_ADDR_TRANS_MODE:
953 		/*
954 		 * KVM does not support mflags=2 (AIL=2) and AIL=1 is reserved.
955 		 * Keep this in synch with kvmppc_filter_guest_lpcr_hv.
956 		 */
957 		if (cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG) &&
958 				kvmhv_vcpu_is_radix(vcpu) && mflags == 3)
959 			return H_UNSUPPORTED_FLAG_START;
960 		return H_TOO_HARD;
961 	default:
962 		return H_TOO_HARD;
963 	}
964 }
965 
966 /* Copy guest memory in place - must reside within a single memslot */
967 static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from,
968 				  unsigned long len)
969 {
970 	struct kvm_memory_slot *to_memslot = NULL;
971 	struct kvm_memory_slot *from_memslot = NULL;
972 	unsigned long to_addr, from_addr;
973 	int r;
974 
975 	/* Get HPA for from address */
976 	from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT);
977 	if (!from_memslot)
978 		return -EFAULT;
979 	if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages)
980 			     << PAGE_SHIFT))
981 		return -EINVAL;
982 	from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT);
983 	if (kvm_is_error_hva(from_addr))
984 		return -EFAULT;
985 	from_addr |= (from & (PAGE_SIZE - 1));
986 
987 	/* Get HPA for to address */
988 	to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT);
989 	if (!to_memslot)
990 		return -EFAULT;
991 	if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages)
992 			   << PAGE_SHIFT))
993 		return -EINVAL;
994 	to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT);
995 	if (kvm_is_error_hva(to_addr))
996 		return -EFAULT;
997 	to_addr |= (to & (PAGE_SIZE - 1));
998 
999 	/* Perform copy */
1000 	r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr,
1001 			     len);
1002 	if (r)
1003 		return -EFAULT;
1004 	mark_page_dirty(kvm, to >> PAGE_SHIFT);
1005 	return 0;
1006 }
1007 
1008 static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags,
1009 			       unsigned long dest, unsigned long src)
1010 {
1011 	u64 pg_sz = SZ_4K;		/* 4K page size */
1012 	u64 pg_mask = SZ_4K - 1;
1013 	int ret;
1014 
1015 	/* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */
1016 	if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE |
1017 		      H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED))
1018 		return H_PARAMETER;
1019 
1020 	/* dest (and src if copy_page flag set) must be page aligned */
1021 	if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask)))
1022 		return H_PARAMETER;
1023 
1024 	/* zero and/or copy the page as determined by the flags */
1025 	if (flags & H_COPY_PAGE) {
1026 		ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz);
1027 		if (ret < 0)
1028 			return H_PARAMETER;
1029 	} else if (flags & H_ZERO_PAGE) {
1030 		ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz);
1031 		if (ret < 0)
1032 			return H_PARAMETER;
1033 	}
1034 
1035 	/* We can ignore the remaining flags */
1036 
1037 	return H_SUCCESS;
1038 }
1039 
1040 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
1041 {
1042 	struct kvmppc_vcore *vcore = target->arch.vcore;
1043 
1044 	/*
1045 	 * We expect to have been called by the real mode handler
1046 	 * (kvmppc_rm_h_confer()) which would have directly returned
1047 	 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
1048 	 * have useful work to do and should not confer) so we don't
1049 	 * recheck that here.
1050 	 *
1051 	 * In the case of the P9 single vcpu per vcore case, the real
1052 	 * mode handler is not called but no other threads are in the
1053 	 * source vcore.
1054 	 */
1055 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
1056 		spin_lock(&vcore->lock);
1057 		if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
1058 		    vcore->vcore_state != VCORE_INACTIVE &&
1059 		    vcore->runner)
1060 			target = vcore->runner;
1061 		spin_unlock(&vcore->lock);
1062 	}
1063 
1064 	return kvm_vcpu_yield_to(target);
1065 }
1066 
1067 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
1068 {
1069 	int yield_count = 0;
1070 	struct lppaca *lppaca;
1071 
1072 	spin_lock(&vcpu->arch.vpa_update_lock);
1073 	lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
1074 	if (lppaca)
1075 		yield_count = be32_to_cpu(lppaca->yield_count);
1076 	spin_unlock(&vcpu->arch.vpa_update_lock);
1077 	return yield_count;
1078 }
1079 
1080 /*
1081  * H_RPT_INVALIDATE hcall handler for nested guests.
1082  *
1083  * Handles only nested process-scoped invalidation requests in L0.
1084  */
1085 static int kvmppc_nested_h_rpt_invalidate(struct kvm_vcpu *vcpu)
1086 {
1087 	unsigned long type = kvmppc_get_gpr(vcpu, 6);
1088 	unsigned long pid, pg_sizes, start, end;
1089 
1090 	/*
1091 	 * The partition-scoped invalidations aren't handled here in L0.
1092 	 */
1093 	if (type & H_RPTI_TYPE_NESTED)
1094 		return RESUME_HOST;
1095 
1096 	pid = kvmppc_get_gpr(vcpu, 4);
1097 	pg_sizes = kvmppc_get_gpr(vcpu, 7);
1098 	start = kvmppc_get_gpr(vcpu, 8);
1099 	end = kvmppc_get_gpr(vcpu, 9);
1100 
1101 	do_h_rpt_invalidate_prt(pid, vcpu->arch.nested->shadow_lpid,
1102 				type, pg_sizes, start, end);
1103 
1104 	kvmppc_set_gpr(vcpu, 3, H_SUCCESS);
1105 	return RESUME_GUEST;
1106 }
1107 
1108 static long kvmppc_h_rpt_invalidate(struct kvm_vcpu *vcpu,
1109 				    unsigned long id, unsigned long target,
1110 				    unsigned long type, unsigned long pg_sizes,
1111 				    unsigned long start, unsigned long end)
1112 {
1113 	if (!kvm_is_radix(vcpu->kvm))
1114 		return H_UNSUPPORTED;
1115 
1116 	if (end < start)
1117 		return H_P5;
1118 
1119 	/*
1120 	 * Partition-scoped invalidation for nested guests.
1121 	 */
1122 	if (type & H_RPTI_TYPE_NESTED) {
1123 		if (!nesting_enabled(vcpu->kvm))
1124 			return H_FUNCTION;
1125 
1126 		/* Support only cores as target */
1127 		if (target != H_RPTI_TARGET_CMMU)
1128 			return H_P2;
1129 
1130 		return do_h_rpt_invalidate_pat(vcpu, id, type, pg_sizes,
1131 					       start, end);
1132 	}
1133 
1134 	/*
1135 	 * Process-scoped invalidation for L1 guests.
1136 	 */
1137 	do_h_rpt_invalidate_prt(id, vcpu->kvm->arch.lpid,
1138 				type, pg_sizes, start, end);
1139 	return H_SUCCESS;
1140 }
1141 
1142 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
1143 {
1144 	struct kvm *kvm = vcpu->kvm;
1145 	unsigned long req = kvmppc_get_gpr(vcpu, 3);
1146 	unsigned long target, ret = H_SUCCESS;
1147 	int yield_count;
1148 	struct kvm_vcpu *tvcpu;
1149 	int idx, rc;
1150 
1151 	if (req <= MAX_HCALL_OPCODE &&
1152 	    !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
1153 		return RESUME_HOST;
1154 
1155 	switch (req) {
1156 	case H_REMOVE:
1157 		ret = kvmppc_h_remove(vcpu, kvmppc_get_gpr(vcpu, 4),
1158 					kvmppc_get_gpr(vcpu, 5),
1159 					kvmppc_get_gpr(vcpu, 6));
1160 		if (ret == H_TOO_HARD)
1161 			return RESUME_HOST;
1162 		break;
1163 	case H_ENTER:
1164 		ret = kvmppc_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
1165 					kvmppc_get_gpr(vcpu, 5),
1166 					kvmppc_get_gpr(vcpu, 6),
1167 					kvmppc_get_gpr(vcpu, 7));
1168 		if (ret == H_TOO_HARD)
1169 			return RESUME_HOST;
1170 		break;
1171 	case H_READ:
1172 		ret = kvmppc_h_read(vcpu, kvmppc_get_gpr(vcpu, 4),
1173 					kvmppc_get_gpr(vcpu, 5));
1174 		if (ret == H_TOO_HARD)
1175 			return RESUME_HOST;
1176 		break;
1177 	case H_CLEAR_MOD:
1178 		ret = kvmppc_h_clear_mod(vcpu, kvmppc_get_gpr(vcpu, 4),
1179 					kvmppc_get_gpr(vcpu, 5));
1180 		if (ret == H_TOO_HARD)
1181 			return RESUME_HOST;
1182 		break;
1183 	case H_CLEAR_REF:
1184 		ret = kvmppc_h_clear_ref(vcpu, kvmppc_get_gpr(vcpu, 4),
1185 					kvmppc_get_gpr(vcpu, 5));
1186 		if (ret == H_TOO_HARD)
1187 			return RESUME_HOST;
1188 		break;
1189 	case H_PROTECT:
1190 		ret = kvmppc_h_protect(vcpu, kvmppc_get_gpr(vcpu, 4),
1191 					kvmppc_get_gpr(vcpu, 5),
1192 					kvmppc_get_gpr(vcpu, 6));
1193 		if (ret == H_TOO_HARD)
1194 			return RESUME_HOST;
1195 		break;
1196 	case H_BULK_REMOVE:
1197 		ret = kvmppc_h_bulk_remove(vcpu);
1198 		if (ret == H_TOO_HARD)
1199 			return RESUME_HOST;
1200 		break;
1201 
1202 	case H_CEDE:
1203 		break;
1204 	case H_PROD:
1205 		target = kvmppc_get_gpr(vcpu, 4);
1206 		tvcpu = kvmppc_find_vcpu(kvm, target);
1207 		if (!tvcpu) {
1208 			ret = H_PARAMETER;
1209 			break;
1210 		}
1211 		tvcpu->arch.prodded = 1;
1212 		smp_mb(); /* This orders prodded store vs ceded load */
1213 		if (tvcpu->arch.ceded)
1214 			kvmppc_fast_vcpu_kick_hv(tvcpu);
1215 		break;
1216 	case H_CONFER:
1217 		target = kvmppc_get_gpr(vcpu, 4);
1218 		if (target == -1)
1219 			break;
1220 		tvcpu = kvmppc_find_vcpu(kvm, target);
1221 		if (!tvcpu) {
1222 			ret = H_PARAMETER;
1223 			break;
1224 		}
1225 		yield_count = kvmppc_get_gpr(vcpu, 5);
1226 		if (kvmppc_get_yield_count(tvcpu) != yield_count)
1227 			break;
1228 		kvm_arch_vcpu_yield_to(tvcpu);
1229 		break;
1230 	case H_REGISTER_VPA:
1231 		ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
1232 					kvmppc_get_gpr(vcpu, 5),
1233 					kvmppc_get_gpr(vcpu, 6));
1234 		break;
1235 	case H_RTAS:
1236 		if (list_empty(&kvm->arch.rtas_tokens))
1237 			return RESUME_HOST;
1238 
1239 		idx = srcu_read_lock(&kvm->srcu);
1240 		rc = kvmppc_rtas_hcall(vcpu);
1241 		srcu_read_unlock(&kvm->srcu, idx);
1242 
1243 		if (rc == -ENOENT)
1244 			return RESUME_HOST;
1245 		else if (rc == 0)
1246 			break;
1247 
1248 		/* Send the error out to userspace via KVM_RUN */
1249 		return rc;
1250 	case H_LOGICAL_CI_LOAD:
1251 		ret = kvmppc_h_logical_ci_load(vcpu);
1252 		if (ret == H_TOO_HARD)
1253 			return RESUME_HOST;
1254 		break;
1255 	case H_LOGICAL_CI_STORE:
1256 		ret = kvmppc_h_logical_ci_store(vcpu);
1257 		if (ret == H_TOO_HARD)
1258 			return RESUME_HOST;
1259 		break;
1260 	case H_SET_MODE:
1261 		ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
1262 					kvmppc_get_gpr(vcpu, 5),
1263 					kvmppc_get_gpr(vcpu, 6),
1264 					kvmppc_get_gpr(vcpu, 7));
1265 		if (ret == H_TOO_HARD)
1266 			return RESUME_HOST;
1267 		break;
1268 	case H_XIRR:
1269 	case H_CPPR:
1270 	case H_EOI:
1271 	case H_IPI:
1272 	case H_IPOLL:
1273 	case H_XIRR_X:
1274 		if (kvmppc_xics_enabled(vcpu)) {
1275 			if (xics_on_xive()) {
1276 				ret = H_NOT_AVAILABLE;
1277 				return RESUME_GUEST;
1278 			}
1279 			ret = kvmppc_xics_hcall(vcpu, req);
1280 			break;
1281 		}
1282 		return RESUME_HOST;
1283 	case H_SET_DABR:
1284 		ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
1285 		break;
1286 	case H_SET_XDABR:
1287 		ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
1288 						kvmppc_get_gpr(vcpu, 5));
1289 		break;
1290 #ifdef CONFIG_SPAPR_TCE_IOMMU
1291 	case H_GET_TCE:
1292 		ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1293 						kvmppc_get_gpr(vcpu, 5));
1294 		if (ret == H_TOO_HARD)
1295 			return RESUME_HOST;
1296 		break;
1297 	case H_PUT_TCE:
1298 		ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1299 						kvmppc_get_gpr(vcpu, 5),
1300 						kvmppc_get_gpr(vcpu, 6));
1301 		if (ret == H_TOO_HARD)
1302 			return RESUME_HOST;
1303 		break;
1304 	case H_PUT_TCE_INDIRECT:
1305 		ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
1306 						kvmppc_get_gpr(vcpu, 5),
1307 						kvmppc_get_gpr(vcpu, 6),
1308 						kvmppc_get_gpr(vcpu, 7));
1309 		if (ret == H_TOO_HARD)
1310 			return RESUME_HOST;
1311 		break;
1312 	case H_STUFF_TCE:
1313 		ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1314 						kvmppc_get_gpr(vcpu, 5),
1315 						kvmppc_get_gpr(vcpu, 6),
1316 						kvmppc_get_gpr(vcpu, 7));
1317 		if (ret == H_TOO_HARD)
1318 			return RESUME_HOST;
1319 		break;
1320 #endif
1321 	case H_RANDOM: {
1322 		unsigned long rand;
1323 
1324 		if (!arch_get_random_seed_longs(&rand, 1))
1325 			ret = H_HARDWARE;
1326 		kvmppc_set_gpr(vcpu, 4, rand);
1327 		break;
1328 	}
1329 	case H_RPT_INVALIDATE:
1330 		ret = kvmppc_h_rpt_invalidate(vcpu, kvmppc_get_gpr(vcpu, 4),
1331 					      kvmppc_get_gpr(vcpu, 5),
1332 					      kvmppc_get_gpr(vcpu, 6),
1333 					      kvmppc_get_gpr(vcpu, 7),
1334 					      kvmppc_get_gpr(vcpu, 8),
1335 					      kvmppc_get_gpr(vcpu, 9));
1336 		break;
1337 
1338 	case H_SET_PARTITION_TABLE:
1339 		ret = H_FUNCTION;
1340 		if (nesting_enabled(kvm))
1341 			ret = kvmhv_set_partition_table(vcpu);
1342 		break;
1343 	case H_ENTER_NESTED:
1344 		ret = H_FUNCTION;
1345 		if (!nesting_enabled(kvm))
1346 			break;
1347 		ret = kvmhv_enter_nested_guest(vcpu);
1348 		if (ret == H_INTERRUPT) {
1349 			kvmppc_set_gpr(vcpu, 3, 0);
1350 			vcpu->arch.hcall_needed = 0;
1351 			return -EINTR;
1352 		} else if (ret == H_TOO_HARD) {
1353 			kvmppc_set_gpr(vcpu, 3, 0);
1354 			vcpu->arch.hcall_needed = 0;
1355 			return RESUME_HOST;
1356 		}
1357 		break;
1358 	case H_TLB_INVALIDATE:
1359 		ret = H_FUNCTION;
1360 		if (nesting_enabled(kvm))
1361 			ret = kvmhv_do_nested_tlbie(vcpu);
1362 		break;
1363 	case H_COPY_TOFROM_GUEST:
1364 		ret = H_FUNCTION;
1365 		if (nesting_enabled(kvm))
1366 			ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1367 		break;
1368 	case H_PAGE_INIT:
1369 		ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4),
1370 					 kvmppc_get_gpr(vcpu, 5),
1371 					 kvmppc_get_gpr(vcpu, 6));
1372 		break;
1373 	case H_SVM_PAGE_IN:
1374 		ret = H_UNSUPPORTED;
1375 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1376 			ret = kvmppc_h_svm_page_in(kvm,
1377 						   kvmppc_get_gpr(vcpu, 4),
1378 						   kvmppc_get_gpr(vcpu, 5),
1379 						   kvmppc_get_gpr(vcpu, 6));
1380 		break;
1381 	case H_SVM_PAGE_OUT:
1382 		ret = H_UNSUPPORTED;
1383 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1384 			ret = kvmppc_h_svm_page_out(kvm,
1385 						    kvmppc_get_gpr(vcpu, 4),
1386 						    kvmppc_get_gpr(vcpu, 5),
1387 						    kvmppc_get_gpr(vcpu, 6));
1388 		break;
1389 	case H_SVM_INIT_START:
1390 		ret = H_UNSUPPORTED;
1391 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1392 			ret = kvmppc_h_svm_init_start(kvm);
1393 		break;
1394 	case H_SVM_INIT_DONE:
1395 		ret = H_UNSUPPORTED;
1396 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1397 			ret = kvmppc_h_svm_init_done(kvm);
1398 		break;
1399 	case H_SVM_INIT_ABORT:
1400 		/*
1401 		 * Even if that call is made by the Ultravisor, the SSR1 value
1402 		 * is the guest context one, with the secure bit clear as it has
1403 		 * not yet been secured. So we can't check it here.
1404 		 * Instead the kvm->arch.secure_guest flag is checked inside
1405 		 * kvmppc_h_svm_init_abort().
1406 		 */
1407 		ret = kvmppc_h_svm_init_abort(kvm);
1408 		break;
1409 
1410 	default:
1411 		return RESUME_HOST;
1412 	}
1413 	WARN_ON_ONCE(ret == H_TOO_HARD);
1414 	kvmppc_set_gpr(vcpu, 3, ret);
1415 	vcpu->arch.hcall_needed = 0;
1416 	return RESUME_GUEST;
1417 }
1418 
1419 /*
1420  * Handle H_CEDE in the P9 path where we don't call the real-mode hcall
1421  * handlers in book3s_hv_rmhandlers.S.
1422  *
1423  * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1424  * that the cede logic in kvmppc_run_single_vcpu() works properly.
1425  */
1426 static void kvmppc_cede(struct kvm_vcpu *vcpu)
1427 {
1428 	__kvmppc_set_msr_hv(vcpu, __kvmppc_get_msr_hv(vcpu) | MSR_EE);
1429 	vcpu->arch.ceded = 1;
1430 	smp_mb();
1431 	if (vcpu->arch.prodded) {
1432 		vcpu->arch.prodded = 0;
1433 		smp_mb();
1434 		vcpu->arch.ceded = 0;
1435 	}
1436 }
1437 
1438 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1439 {
1440 	switch (cmd) {
1441 	case H_CEDE:
1442 	case H_PROD:
1443 	case H_CONFER:
1444 	case H_REGISTER_VPA:
1445 	case H_SET_MODE:
1446 #ifdef CONFIG_SPAPR_TCE_IOMMU
1447 	case H_GET_TCE:
1448 	case H_PUT_TCE:
1449 	case H_PUT_TCE_INDIRECT:
1450 	case H_STUFF_TCE:
1451 #endif
1452 	case H_LOGICAL_CI_LOAD:
1453 	case H_LOGICAL_CI_STORE:
1454 #ifdef CONFIG_KVM_XICS
1455 	case H_XIRR:
1456 	case H_CPPR:
1457 	case H_EOI:
1458 	case H_IPI:
1459 	case H_IPOLL:
1460 	case H_XIRR_X:
1461 #endif
1462 	case H_PAGE_INIT:
1463 	case H_RPT_INVALIDATE:
1464 		return 1;
1465 	}
1466 
1467 	/* See if it's in the real-mode table */
1468 	return kvmppc_hcall_impl_hv_realmode(cmd);
1469 }
1470 
1471 static int kvmppc_emulate_debug_inst(struct kvm_vcpu *vcpu)
1472 {
1473 	ppc_inst_t last_inst;
1474 
1475 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1476 					EMULATE_DONE) {
1477 		/*
1478 		 * Fetch failed, so return to guest and
1479 		 * try executing it again.
1480 		 */
1481 		return RESUME_GUEST;
1482 	}
1483 
1484 	if (ppc_inst_val(last_inst) == KVMPPC_INST_SW_BREAKPOINT) {
1485 		vcpu->run->exit_reason = KVM_EXIT_DEBUG;
1486 		vcpu->run->debug.arch.address = kvmppc_get_pc(vcpu);
1487 		return RESUME_HOST;
1488 	} else {
1489 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL |
1490 				(kvmppc_get_msr(vcpu) & SRR1_PREFIXED));
1491 		return RESUME_GUEST;
1492 	}
1493 }
1494 
1495 static void do_nothing(void *x)
1496 {
1497 }
1498 
1499 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1500 {
1501 	int thr, cpu, pcpu, nthreads;
1502 	struct kvm_vcpu *v;
1503 	unsigned long dpdes;
1504 
1505 	nthreads = vcpu->kvm->arch.emul_smt_mode;
1506 	dpdes = 0;
1507 	cpu = vcpu->vcpu_id & ~(nthreads - 1);
1508 	for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1509 		v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1510 		if (!v)
1511 			continue;
1512 		/*
1513 		 * If the vcpu is currently running on a physical cpu thread,
1514 		 * interrupt it in order to pull it out of the guest briefly,
1515 		 * which will update its vcore->dpdes value.
1516 		 */
1517 		pcpu = READ_ONCE(v->cpu);
1518 		if (pcpu >= 0)
1519 			smp_call_function_single(pcpu, do_nothing, NULL, 1);
1520 		if (kvmppc_doorbell_pending(v))
1521 			dpdes |= 1 << thr;
1522 	}
1523 	return dpdes;
1524 }
1525 
1526 /*
1527  * On POWER9, emulate doorbell-related instructions in order to
1528  * give the guest the illusion of running on a multi-threaded core.
1529  * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1530  * and mfspr DPDES.
1531  */
1532 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1533 {
1534 	u32 inst, rb, thr;
1535 	unsigned long arg;
1536 	struct kvm *kvm = vcpu->kvm;
1537 	struct kvm_vcpu *tvcpu;
1538 	ppc_inst_t pinst;
1539 
1540 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &pinst) != EMULATE_DONE)
1541 		return RESUME_GUEST;
1542 	inst = ppc_inst_val(pinst);
1543 	if (get_op(inst) != 31)
1544 		return EMULATE_FAIL;
1545 	rb = get_rb(inst);
1546 	thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1547 	switch (get_xop(inst)) {
1548 	case OP_31_XOP_MSGSNDP:
1549 		arg = kvmppc_get_gpr(vcpu, rb);
1550 		if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1551 			break;
1552 		arg &= 0x7f;
1553 		if (arg >= kvm->arch.emul_smt_mode)
1554 			break;
1555 		tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1556 		if (!tvcpu)
1557 			break;
1558 		if (!tvcpu->arch.doorbell_request) {
1559 			tvcpu->arch.doorbell_request = 1;
1560 			kvmppc_fast_vcpu_kick_hv(tvcpu);
1561 		}
1562 		break;
1563 	case OP_31_XOP_MSGCLRP:
1564 		arg = kvmppc_get_gpr(vcpu, rb);
1565 		if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1566 			break;
1567 		vcpu->arch.vcore->dpdes = 0;
1568 		vcpu->arch.doorbell_request = 0;
1569 		break;
1570 	case OP_31_XOP_MFSPR:
1571 		switch (get_sprn(inst)) {
1572 		case SPRN_TIR:
1573 			arg = thr;
1574 			break;
1575 		case SPRN_DPDES:
1576 			arg = kvmppc_read_dpdes(vcpu);
1577 			break;
1578 		default:
1579 			return EMULATE_FAIL;
1580 		}
1581 		kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1582 		break;
1583 	default:
1584 		return EMULATE_FAIL;
1585 	}
1586 	kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1587 	return RESUME_GUEST;
1588 }
1589 
1590 /*
1591  * If the lppaca had pmcregs_in_use clear when we exited the guest, then
1592  * HFSCR_PM is cleared for next entry. If the guest then tries to access
1593  * the PMU SPRs, we get this facility unavailable interrupt. Putting HFSCR_PM
1594  * back in the guest HFSCR will cause the next entry to load the PMU SPRs and
1595  * allow the guest access to continue.
1596  */
1597 static int kvmppc_pmu_unavailable(struct kvm_vcpu *vcpu)
1598 {
1599 	if (!(vcpu->arch.hfscr_permitted & HFSCR_PM))
1600 		return EMULATE_FAIL;
1601 
1602 	kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) | HFSCR_PM);
1603 
1604 	return RESUME_GUEST;
1605 }
1606 
1607 static int kvmppc_ebb_unavailable(struct kvm_vcpu *vcpu)
1608 {
1609 	if (!(vcpu->arch.hfscr_permitted & HFSCR_EBB))
1610 		return EMULATE_FAIL;
1611 
1612 	kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) | HFSCR_EBB);
1613 
1614 	return RESUME_GUEST;
1615 }
1616 
1617 static int kvmppc_tm_unavailable(struct kvm_vcpu *vcpu)
1618 {
1619 	if (!(vcpu->arch.hfscr_permitted & HFSCR_TM))
1620 		return EMULATE_FAIL;
1621 
1622 	kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) | HFSCR_TM);
1623 
1624 	return RESUME_GUEST;
1625 }
1626 
1627 static int kvmppc_handle_exit_hv(struct kvm_vcpu *vcpu,
1628 				 struct task_struct *tsk)
1629 {
1630 	struct kvm_run *run = vcpu->run;
1631 	int r = RESUME_HOST;
1632 
1633 	vcpu->stat.sum_exits++;
1634 
1635 	/*
1636 	 * This can happen if an interrupt occurs in the last stages
1637 	 * of guest entry or the first stages of guest exit (i.e. after
1638 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1639 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1640 	 * That can happen due to a bug, or due to a machine check
1641 	 * occurring at just the wrong time.
1642 	 */
1643 	if (!kvmhv_is_nestedv2() && (__kvmppc_get_msr_hv(vcpu) & MSR_HV)) {
1644 		printk(KERN_EMERG "KVM trap in HV mode!\n");
1645 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1646 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1647 			vcpu->arch.shregs.msr);
1648 		kvmppc_dump_regs(vcpu);
1649 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1650 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1651 		return RESUME_HOST;
1652 	}
1653 	run->exit_reason = KVM_EXIT_UNKNOWN;
1654 	run->ready_for_interrupt_injection = 1;
1655 	switch (vcpu->arch.trap) {
1656 	/* We're good on these - the host merely wanted to get our attention */
1657 	case BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER:
1658 		WARN_ON_ONCE(1); /* Should never happen */
1659 		vcpu->arch.trap = BOOK3S_INTERRUPT_HV_DECREMENTER;
1660 		fallthrough;
1661 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1662 		vcpu->stat.dec_exits++;
1663 		r = RESUME_GUEST;
1664 		break;
1665 	case BOOK3S_INTERRUPT_EXTERNAL:
1666 	case BOOK3S_INTERRUPT_H_DOORBELL:
1667 	case BOOK3S_INTERRUPT_H_VIRT:
1668 		vcpu->stat.ext_intr_exits++;
1669 		r = RESUME_GUEST;
1670 		break;
1671 	/* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1672 	case BOOK3S_INTERRUPT_HMI:
1673 	case BOOK3S_INTERRUPT_PERFMON:
1674 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
1675 		r = RESUME_GUEST;
1676 		break;
1677 	case BOOK3S_INTERRUPT_MACHINE_CHECK: {
1678 		static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
1679 					      DEFAULT_RATELIMIT_BURST);
1680 		/*
1681 		 * Print the MCE event to host console. Ratelimit so the guest
1682 		 * can't flood the host log.
1683 		 */
1684 		if (__ratelimit(&rs))
1685 			machine_check_print_event_info(&vcpu->arch.mce_evt,false, true);
1686 
1687 		/*
1688 		 * If the guest can do FWNMI, exit to userspace so it can
1689 		 * deliver a FWNMI to the guest.
1690 		 * Otherwise we synthesize a machine check for the guest
1691 		 * so that it knows that the machine check occurred.
1692 		 */
1693 		if (!vcpu->kvm->arch.fwnmi_enabled) {
1694 			ulong flags = (__kvmppc_get_msr_hv(vcpu) & 0x083c0000) |
1695 					(kvmppc_get_msr(vcpu) & SRR1_PREFIXED);
1696 			kvmppc_core_queue_machine_check(vcpu, flags);
1697 			r = RESUME_GUEST;
1698 			break;
1699 		}
1700 
1701 		/* Exit to guest with KVM_EXIT_NMI as exit reason */
1702 		run->exit_reason = KVM_EXIT_NMI;
1703 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1704 		/* Clear out the old NMI status from run->flags */
1705 		run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1706 		/* Now set the NMI status */
1707 		if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1708 			run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1709 		else
1710 			run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1711 
1712 		r = RESUME_HOST;
1713 		break;
1714 	}
1715 	case BOOK3S_INTERRUPT_PROGRAM:
1716 	{
1717 		ulong flags;
1718 		/*
1719 		 * Normally program interrupts are delivered directly
1720 		 * to the guest by the hardware, but we can get here
1721 		 * as a result of a hypervisor emulation interrupt
1722 		 * (e40) getting turned into a 700 by BML RTAS.
1723 		 */
1724 		flags = (__kvmppc_get_msr_hv(vcpu) & 0x1f0000ull) |
1725 			(kvmppc_get_msr(vcpu) & SRR1_PREFIXED);
1726 		kvmppc_core_queue_program(vcpu, flags);
1727 		r = RESUME_GUEST;
1728 		break;
1729 	}
1730 	case BOOK3S_INTERRUPT_SYSCALL:
1731 	{
1732 		int i;
1733 
1734 		if (!kvmhv_is_nestedv2() && unlikely(__kvmppc_get_msr_hv(vcpu) & MSR_PR)) {
1735 			/*
1736 			 * Guest userspace executed sc 1. This can only be
1737 			 * reached by the P9 path because the old path
1738 			 * handles this case in realmode hcall handlers.
1739 			 */
1740 			if (!kvmhv_vcpu_is_radix(vcpu)) {
1741 				/*
1742 				 * A guest could be running PR KVM, so this
1743 				 * may be a PR KVM hcall. It must be reflected
1744 				 * to the guest kernel as a sc interrupt.
1745 				 */
1746 				kvmppc_core_queue_syscall(vcpu);
1747 			} else {
1748 				/*
1749 				 * Radix guests can not run PR KVM or nested HV
1750 				 * hash guests which might run PR KVM, so this
1751 				 * is always a privilege fault. Send a program
1752 				 * check to guest kernel.
1753 				 */
1754 				kvmppc_core_queue_program(vcpu, SRR1_PROGPRIV);
1755 			}
1756 			r = RESUME_GUEST;
1757 			break;
1758 		}
1759 
1760 		/*
1761 		 * hcall - gather args and set exit_reason. This will next be
1762 		 * handled by kvmppc_pseries_do_hcall which may be able to deal
1763 		 * with it and resume guest, or may punt to userspace.
1764 		 */
1765 		run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1766 		for (i = 0; i < 9; ++i)
1767 			run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1768 		run->exit_reason = KVM_EXIT_PAPR_HCALL;
1769 		vcpu->arch.hcall_needed = 1;
1770 		r = RESUME_HOST;
1771 		break;
1772 	}
1773 	/*
1774 	 * We get these next two if the guest accesses a page which it thinks
1775 	 * it has mapped but which is not actually present, either because
1776 	 * it is for an emulated I/O device or because the corresonding
1777 	 * host page has been paged out.
1778 	 *
1779 	 * Any other HDSI/HISI interrupts have been handled already for P7/8
1780 	 * guests. For POWER9 hash guests not using rmhandlers, basic hash
1781 	 * fault handling is done here.
1782 	 */
1783 	case BOOK3S_INTERRUPT_H_DATA_STORAGE: {
1784 		unsigned long vsid;
1785 		long err;
1786 
1787 		if (cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG) &&
1788 		    unlikely(vcpu->arch.fault_dsisr == HDSISR_CANARY)) {
1789 			r = RESUME_GUEST; /* Just retry if it's the canary */
1790 			break;
1791 		}
1792 
1793 		if (kvm_is_radix(vcpu->kvm) || !cpu_has_feature(CPU_FTR_ARCH_300)) {
1794 			/*
1795 			 * Radix doesn't require anything, and pre-ISAv3.0 hash
1796 			 * already attempted to handle this in rmhandlers. The
1797 			 * hash fault handling below is v3 only (it uses ASDR
1798 			 * via fault_gpa).
1799 			 */
1800 			r = RESUME_PAGE_FAULT;
1801 			break;
1802 		}
1803 
1804 		if (!(vcpu->arch.fault_dsisr & (DSISR_NOHPTE | DSISR_PROTFAULT))) {
1805 			kvmppc_core_queue_data_storage(vcpu,
1806 				kvmppc_get_msr(vcpu) & SRR1_PREFIXED,
1807 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1808 			r = RESUME_GUEST;
1809 			break;
1810 		}
1811 
1812 		if (!(__kvmppc_get_msr_hv(vcpu) & MSR_DR))
1813 			vsid = vcpu->kvm->arch.vrma_slb_v;
1814 		else
1815 			vsid = vcpu->arch.fault_gpa;
1816 
1817 		err = kvmppc_hpte_hv_fault(vcpu, vcpu->arch.fault_dar,
1818 				vsid, vcpu->arch.fault_dsisr, true);
1819 		if (err == 0) {
1820 			r = RESUME_GUEST;
1821 		} else if (err == -1 || err == -2) {
1822 			r = RESUME_PAGE_FAULT;
1823 		} else {
1824 			kvmppc_core_queue_data_storage(vcpu,
1825 				kvmppc_get_msr(vcpu) & SRR1_PREFIXED,
1826 				vcpu->arch.fault_dar, err);
1827 			r = RESUME_GUEST;
1828 		}
1829 		break;
1830 	}
1831 	case BOOK3S_INTERRUPT_H_INST_STORAGE: {
1832 		unsigned long vsid;
1833 		long err;
1834 
1835 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1836 		vcpu->arch.fault_dsisr = __kvmppc_get_msr_hv(vcpu) &
1837 			DSISR_SRR1_MATCH_64S;
1838 		if (kvm_is_radix(vcpu->kvm) || !cpu_has_feature(CPU_FTR_ARCH_300)) {
1839 			/*
1840 			 * Radix doesn't require anything, and pre-ISAv3.0 hash
1841 			 * already attempted to handle this in rmhandlers. The
1842 			 * hash fault handling below is v3 only (it uses ASDR
1843 			 * via fault_gpa).
1844 			 */
1845 			if (__kvmppc_get_msr_hv(vcpu) & HSRR1_HISI_WRITE)
1846 				vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1847 			r = RESUME_PAGE_FAULT;
1848 			break;
1849 		}
1850 
1851 		if (!(vcpu->arch.fault_dsisr & SRR1_ISI_NOPT)) {
1852 			kvmppc_core_queue_inst_storage(vcpu,
1853 				vcpu->arch.fault_dsisr |
1854 				(kvmppc_get_msr(vcpu) & SRR1_PREFIXED));
1855 			r = RESUME_GUEST;
1856 			break;
1857 		}
1858 
1859 		if (!(__kvmppc_get_msr_hv(vcpu) & MSR_IR))
1860 			vsid = vcpu->kvm->arch.vrma_slb_v;
1861 		else
1862 			vsid = vcpu->arch.fault_gpa;
1863 
1864 		err = kvmppc_hpte_hv_fault(vcpu, vcpu->arch.fault_dar,
1865 				vsid, vcpu->arch.fault_dsisr, false);
1866 		if (err == 0) {
1867 			r = RESUME_GUEST;
1868 		} else if (err == -1) {
1869 			r = RESUME_PAGE_FAULT;
1870 		} else {
1871 			kvmppc_core_queue_inst_storage(vcpu,
1872 				err | (kvmppc_get_msr(vcpu) & SRR1_PREFIXED));
1873 			r = RESUME_GUEST;
1874 		}
1875 		break;
1876 	}
1877 
1878 	/*
1879 	 * This occurs if the guest executes an illegal instruction.
1880 	 * If the guest debug is disabled, generate a program interrupt
1881 	 * to the guest. If guest debug is enabled, we need to check
1882 	 * whether the instruction is a software breakpoint instruction.
1883 	 * Accordingly return to Guest or Host.
1884 	 */
1885 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1886 		if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1887 			vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1888 				swab32(vcpu->arch.emul_inst) :
1889 				vcpu->arch.emul_inst;
1890 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1891 			r = kvmppc_emulate_debug_inst(vcpu);
1892 		} else {
1893 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL |
1894 				(kvmppc_get_msr(vcpu) & SRR1_PREFIXED));
1895 			r = RESUME_GUEST;
1896 		}
1897 		break;
1898 
1899 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1900 	case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1901 		/*
1902 		 * This occurs for various TM-related instructions that
1903 		 * we need to emulate on POWER9 DD2.2.  We have already
1904 		 * handled the cases where the guest was in real-suspend
1905 		 * mode and was transitioning to transactional state.
1906 		 */
1907 		r = kvmhv_p9_tm_emulation(vcpu);
1908 		if (r != -1)
1909 			break;
1910 		fallthrough; /* go to facility unavailable handler */
1911 #endif
1912 
1913 	/*
1914 	 * This occurs if the guest (kernel or userspace), does something that
1915 	 * is prohibited by HFSCR.
1916 	 * On POWER9, this could be a doorbell instruction that we need
1917 	 * to emulate.
1918 	 * Otherwise, we just generate a program interrupt to the guest.
1919 	 */
1920 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL: {
1921 		u64 cause = kvmppc_get_hfscr_hv(vcpu) >> 56;
1922 
1923 		r = EMULATE_FAIL;
1924 		if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1925 			if (cause == FSCR_MSGP_LG)
1926 				r = kvmppc_emulate_doorbell_instr(vcpu);
1927 			if (cause == FSCR_PM_LG)
1928 				r = kvmppc_pmu_unavailable(vcpu);
1929 			if (cause == FSCR_EBB_LG)
1930 				r = kvmppc_ebb_unavailable(vcpu);
1931 			if (cause == FSCR_TM_LG)
1932 				r = kvmppc_tm_unavailable(vcpu);
1933 		}
1934 		if (r == EMULATE_FAIL) {
1935 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL |
1936 				(kvmppc_get_msr(vcpu) & SRR1_PREFIXED));
1937 			r = RESUME_GUEST;
1938 		}
1939 		break;
1940 	}
1941 
1942 	case BOOK3S_INTERRUPT_HV_RM_HARD:
1943 		r = RESUME_PASSTHROUGH;
1944 		break;
1945 	default:
1946 		kvmppc_dump_regs(vcpu);
1947 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1948 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1949 			__kvmppc_get_msr_hv(vcpu));
1950 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1951 		r = RESUME_HOST;
1952 		break;
1953 	}
1954 
1955 	return r;
1956 }
1957 
1958 static int kvmppc_handle_nested_exit(struct kvm_vcpu *vcpu)
1959 {
1960 	int r;
1961 	int srcu_idx;
1962 
1963 	vcpu->stat.sum_exits++;
1964 
1965 	/*
1966 	 * This can happen if an interrupt occurs in the last stages
1967 	 * of guest entry or the first stages of guest exit (i.e. after
1968 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1969 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1970 	 * That can happen due to a bug, or due to a machine check
1971 	 * occurring at just the wrong time.
1972 	 */
1973 	if (__kvmppc_get_msr_hv(vcpu) & MSR_HV) {
1974 		pr_emerg("KVM trap in HV mode while nested!\n");
1975 		pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1976 			 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1977 			 __kvmppc_get_msr_hv(vcpu));
1978 		kvmppc_dump_regs(vcpu);
1979 		return RESUME_HOST;
1980 	}
1981 	switch (vcpu->arch.trap) {
1982 	/* We're good on these - the host merely wanted to get our attention */
1983 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1984 		vcpu->stat.dec_exits++;
1985 		r = RESUME_GUEST;
1986 		break;
1987 	case BOOK3S_INTERRUPT_EXTERNAL:
1988 		vcpu->stat.ext_intr_exits++;
1989 		r = RESUME_HOST;
1990 		break;
1991 	case BOOK3S_INTERRUPT_H_DOORBELL:
1992 	case BOOK3S_INTERRUPT_H_VIRT:
1993 		vcpu->stat.ext_intr_exits++;
1994 		r = RESUME_GUEST;
1995 		break;
1996 	/* These need to go to the nested HV */
1997 	case BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER:
1998 		vcpu->arch.trap = BOOK3S_INTERRUPT_HV_DECREMENTER;
1999 		vcpu->stat.dec_exits++;
2000 		r = RESUME_HOST;
2001 		break;
2002 	/* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
2003 	case BOOK3S_INTERRUPT_HMI:
2004 	case BOOK3S_INTERRUPT_PERFMON:
2005 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
2006 		r = RESUME_GUEST;
2007 		break;
2008 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
2009 	{
2010 		static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
2011 					      DEFAULT_RATELIMIT_BURST);
2012 		/* Pass the machine check to the L1 guest */
2013 		r = RESUME_HOST;
2014 		/* Print the MCE event to host console. */
2015 		if (__ratelimit(&rs))
2016 			machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
2017 		break;
2018 	}
2019 	/*
2020 	 * We get these next two if the guest accesses a page which it thinks
2021 	 * it has mapped but which is not actually present, either because
2022 	 * it is for an emulated I/O device or because the corresonding
2023 	 * host page has been paged out.
2024 	 */
2025 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
2026 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2027 		r = kvmhv_nested_page_fault(vcpu);
2028 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2029 		break;
2030 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
2031 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
2032 		vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
2033 					 DSISR_SRR1_MATCH_64S;
2034 		if (__kvmppc_get_msr_hv(vcpu) & HSRR1_HISI_WRITE)
2035 			vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
2036 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2037 		r = kvmhv_nested_page_fault(vcpu);
2038 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2039 		break;
2040 
2041 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2042 	case BOOK3S_INTERRUPT_HV_SOFTPATCH:
2043 		/*
2044 		 * This occurs for various TM-related instructions that
2045 		 * we need to emulate on POWER9 DD2.2.  We have already
2046 		 * handled the cases where the guest was in real-suspend
2047 		 * mode and was transitioning to transactional state.
2048 		 */
2049 		r = kvmhv_p9_tm_emulation(vcpu);
2050 		if (r != -1)
2051 			break;
2052 		fallthrough; /* go to facility unavailable handler */
2053 #endif
2054 
2055 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL: {
2056 		u64 cause = vcpu->arch.hfscr >> 56;
2057 
2058 		/*
2059 		 * Only pass HFU interrupts to the L1 if the facility is
2060 		 * permitted but disabled by the L1's HFSCR, otherwise
2061 		 * the interrupt does not make sense to the L1 so turn
2062 		 * it into a HEAI.
2063 		 */
2064 		if (!(vcpu->arch.hfscr_permitted & (1UL << cause)) ||
2065 				(vcpu->arch.nested_hfscr & (1UL << cause))) {
2066 			ppc_inst_t pinst;
2067 			vcpu->arch.trap = BOOK3S_INTERRUPT_H_EMUL_ASSIST;
2068 
2069 			/*
2070 			 * If the fetch failed, return to guest and
2071 			 * try executing it again.
2072 			 */
2073 			r = kvmppc_get_last_inst(vcpu, INST_GENERIC, &pinst);
2074 			vcpu->arch.emul_inst = ppc_inst_val(pinst);
2075 			if (r != EMULATE_DONE)
2076 				r = RESUME_GUEST;
2077 			else
2078 				r = RESUME_HOST;
2079 		} else {
2080 			r = RESUME_HOST;
2081 		}
2082 
2083 		break;
2084 	}
2085 
2086 	case BOOK3S_INTERRUPT_HV_RM_HARD:
2087 		vcpu->arch.trap = 0;
2088 		r = RESUME_GUEST;
2089 		if (!xics_on_xive())
2090 			kvmppc_xics_rm_complete(vcpu, 0);
2091 		break;
2092 	case BOOK3S_INTERRUPT_SYSCALL:
2093 	{
2094 		unsigned long req = kvmppc_get_gpr(vcpu, 3);
2095 
2096 		/*
2097 		 * The H_RPT_INVALIDATE hcalls issued by nested
2098 		 * guests for process-scoped invalidations when
2099 		 * GTSE=0, are handled here in L0.
2100 		 */
2101 		if (req == H_RPT_INVALIDATE) {
2102 			r = kvmppc_nested_h_rpt_invalidate(vcpu);
2103 			break;
2104 		}
2105 
2106 		r = RESUME_HOST;
2107 		break;
2108 	}
2109 	default:
2110 		r = RESUME_HOST;
2111 		break;
2112 	}
2113 
2114 	return r;
2115 }
2116 
2117 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
2118 					    struct kvm_sregs *sregs)
2119 {
2120 	int i;
2121 
2122 	memset(sregs, 0, sizeof(struct kvm_sregs));
2123 	sregs->pvr = vcpu->arch.pvr;
2124 	for (i = 0; i < vcpu->arch.slb_max; i++) {
2125 		sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
2126 		sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
2127 	}
2128 
2129 	return 0;
2130 }
2131 
2132 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
2133 					    struct kvm_sregs *sregs)
2134 {
2135 	int i, j;
2136 
2137 	/* Only accept the same PVR as the host's, since we can't spoof it */
2138 	if (sregs->pvr != vcpu->arch.pvr)
2139 		return -EINVAL;
2140 
2141 	j = 0;
2142 	for (i = 0; i < vcpu->arch.slb_nr; i++) {
2143 		if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
2144 			vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
2145 			vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
2146 			++j;
2147 		}
2148 	}
2149 	vcpu->arch.slb_max = j;
2150 
2151 	return 0;
2152 }
2153 
2154 /*
2155  * Enforce limits on guest LPCR values based on hardware availability,
2156  * guest configuration, and possibly hypervisor support and security
2157  * concerns.
2158  */
2159 unsigned long kvmppc_filter_lpcr_hv(struct kvm *kvm, unsigned long lpcr)
2160 {
2161 	/* LPCR_TC only applies to HPT guests */
2162 	if (kvm_is_radix(kvm))
2163 		lpcr &= ~LPCR_TC;
2164 
2165 	/* On POWER8 and above, userspace can modify AIL */
2166 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2167 		lpcr &= ~LPCR_AIL;
2168 	if ((lpcr & LPCR_AIL) != LPCR_AIL_3)
2169 		lpcr &= ~LPCR_AIL; /* LPCR[AIL]=1/2 is disallowed */
2170 	/*
2171 	 * On some POWER9s we force AIL off for radix guests to prevent
2172 	 * executing in MSR[HV]=1 mode with the MMU enabled and PIDR set to
2173 	 * guest, which can result in Q0 translations with LPID=0 PID=PIDR to
2174 	 * be cached, which the host TLB management does not expect.
2175 	 */
2176 	if (kvm_is_radix(kvm) && cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG))
2177 		lpcr &= ~LPCR_AIL;
2178 
2179 	/*
2180 	 * On POWER9, allow userspace to enable large decrementer for the
2181 	 * guest, whether or not the host has it enabled.
2182 	 */
2183 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
2184 		lpcr &= ~LPCR_LD;
2185 
2186 	return lpcr;
2187 }
2188 
2189 static void verify_lpcr(struct kvm *kvm, unsigned long lpcr)
2190 {
2191 	if (lpcr != kvmppc_filter_lpcr_hv(kvm, lpcr)) {
2192 		WARN_ONCE(1, "lpcr 0x%lx differs from filtered 0x%lx\n",
2193 			  lpcr, kvmppc_filter_lpcr_hv(kvm, lpcr));
2194 	}
2195 }
2196 
2197 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
2198 		bool preserve_top32)
2199 {
2200 	struct kvm *kvm = vcpu->kvm;
2201 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
2202 	u64 mask;
2203 
2204 	spin_lock(&vc->lock);
2205 
2206 	/*
2207 	 * Userspace can only modify
2208 	 * DPFD (default prefetch depth), ILE (interrupt little-endian),
2209 	 * TC (translation control), AIL (alternate interrupt location),
2210 	 * LD (large decrementer).
2211 	 * These are subject to restrictions from kvmppc_filter_lcpr_hv().
2212 	 */
2213 	mask = LPCR_DPFD | LPCR_ILE | LPCR_TC | LPCR_AIL | LPCR_LD;
2214 
2215 	/* Broken 32-bit version of LPCR must not clear top bits */
2216 	if (preserve_top32)
2217 		mask &= 0xFFFFFFFF;
2218 
2219 	new_lpcr = kvmppc_filter_lpcr_hv(kvm,
2220 			(vc->lpcr & ~mask) | (new_lpcr & mask));
2221 
2222 	/*
2223 	 * If ILE (interrupt little-endian) has changed, update the
2224 	 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
2225 	 */
2226 	if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
2227 		struct kvm_vcpu *vcpu;
2228 		unsigned long i;
2229 
2230 		kvm_for_each_vcpu(i, vcpu, kvm) {
2231 			if (vcpu->arch.vcore != vc)
2232 				continue;
2233 			if (new_lpcr & LPCR_ILE)
2234 				vcpu->arch.intr_msr |= MSR_LE;
2235 			else
2236 				vcpu->arch.intr_msr &= ~MSR_LE;
2237 		}
2238 	}
2239 
2240 	vc->lpcr = new_lpcr;
2241 	kvmhv_nestedv2_mark_dirty(vcpu, KVMPPC_GSID_LPCR);
2242 
2243 	spin_unlock(&vc->lock);
2244 }
2245 
2246 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
2247 				 union kvmppc_one_reg *val)
2248 {
2249 	int r = 0;
2250 	long int i;
2251 
2252 	switch (id) {
2253 	case KVM_REG_PPC_DEBUG_INST:
2254 		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
2255 		break;
2256 	case KVM_REG_PPC_HIOR:
2257 		*val = get_reg_val(id, 0);
2258 		break;
2259 	case KVM_REG_PPC_DABR:
2260 		*val = get_reg_val(id, vcpu->arch.dabr);
2261 		break;
2262 	case KVM_REG_PPC_DABRX:
2263 		*val = get_reg_val(id, vcpu->arch.dabrx);
2264 		break;
2265 	case KVM_REG_PPC_DSCR:
2266 		*val = get_reg_val(id, kvmppc_get_dscr_hv(vcpu));
2267 		break;
2268 	case KVM_REG_PPC_PURR:
2269 		*val = get_reg_val(id, kvmppc_get_purr_hv(vcpu));
2270 		break;
2271 	case KVM_REG_PPC_SPURR:
2272 		*val = get_reg_val(id, kvmppc_get_spurr_hv(vcpu));
2273 		break;
2274 	case KVM_REG_PPC_AMR:
2275 		*val = get_reg_val(id, kvmppc_get_amr_hv(vcpu));
2276 		break;
2277 	case KVM_REG_PPC_UAMOR:
2278 		*val = get_reg_val(id, kvmppc_get_uamor_hv(vcpu));
2279 		break;
2280 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
2281 		i = id - KVM_REG_PPC_MMCR0;
2282 		*val = get_reg_val(id, kvmppc_get_mmcr_hv(vcpu, i));
2283 		break;
2284 	case KVM_REG_PPC_MMCR2:
2285 		*val = get_reg_val(id, kvmppc_get_mmcr_hv(vcpu, 2));
2286 		break;
2287 	case KVM_REG_PPC_MMCRA:
2288 		*val = get_reg_val(id, kvmppc_get_mmcra_hv(vcpu));
2289 		break;
2290 	case KVM_REG_PPC_MMCRS:
2291 		*val = get_reg_val(id, vcpu->arch.mmcrs);
2292 		break;
2293 	case KVM_REG_PPC_MMCR3:
2294 		*val = get_reg_val(id, kvmppc_get_mmcr_hv(vcpu, 3));
2295 		break;
2296 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
2297 		i = id - KVM_REG_PPC_PMC1;
2298 		*val = get_reg_val(id, kvmppc_get_pmc_hv(vcpu, i));
2299 		break;
2300 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
2301 		i = id - KVM_REG_PPC_SPMC1;
2302 		*val = get_reg_val(id, vcpu->arch.spmc[i]);
2303 		break;
2304 	case KVM_REG_PPC_SIAR:
2305 		*val = get_reg_val(id, kvmppc_get_siar_hv(vcpu));
2306 		break;
2307 	case KVM_REG_PPC_SDAR:
2308 		*val = get_reg_val(id, kvmppc_get_siar_hv(vcpu));
2309 		break;
2310 	case KVM_REG_PPC_SIER:
2311 		*val = get_reg_val(id, kvmppc_get_sier_hv(vcpu, 0));
2312 		break;
2313 	case KVM_REG_PPC_SIER2:
2314 		*val = get_reg_val(id, kvmppc_get_sier_hv(vcpu, 1));
2315 		break;
2316 	case KVM_REG_PPC_SIER3:
2317 		*val = get_reg_val(id, kvmppc_get_sier_hv(vcpu, 2));
2318 		break;
2319 	case KVM_REG_PPC_IAMR:
2320 		*val = get_reg_val(id, kvmppc_get_iamr_hv(vcpu));
2321 		break;
2322 	case KVM_REG_PPC_PSPB:
2323 		*val = get_reg_val(id, kvmppc_get_pspb_hv(vcpu));
2324 		break;
2325 	case KVM_REG_PPC_DPDES:
2326 		/*
2327 		 * On POWER9, where we are emulating msgsndp etc.,
2328 		 * we return 1 bit for each vcpu, which can come from
2329 		 * either vcore->dpdes or doorbell_request.
2330 		 * On POWER8, doorbell_request is 0.
2331 		 */
2332 		if (cpu_has_feature(CPU_FTR_ARCH_300))
2333 			*val = get_reg_val(id, vcpu->arch.doorbell_request);
2334 		else
2335 			*val = get_reg_val(id, vcpu->arch.vcore->dpdes);
2336 		break;
2337 	case KVM_REG_PPC_VTB:
2338 		*val = get_reg_val(id, kvmppc_get_vtb(vcpu));
2339 		break;
2340 	case KVM_REG_PPC_DAWR:
2341 		*val = get_reg_val(id, kvmppc_get_dawr0_hv(vcpu));
2342 		break;
2343 	case KVM_REG_PPC_DAWRX:
2344 		*val = get_reg_val(id, kvmppc_get_dawrx0_hv(vcpu));
2345 		break;
2346 	case KVM_REG_PPC_DAWR1:
2347 		*val = get_reg_val(id, kvmppc_get_dawr1_hv(vcpu));
2348 		break;
2349 	case KVM_REG_PPC_DAWRX1:
2350 		*val = get_reg_val(id, kvmppc_get_dawrx1_hv(vcpu));
2351 		break;
2352 	case KVM_REG_PPC_CIABR:
2353 		*val = get_reg_val(id, kvmppc_get_ciabr_hv(vcpu));
2354 		break;
2355 	case KVM_REG_PPC_CSIGR:
2356 		*val = get_reg_val(id, vcpu->arch.csigr);
2357 		break;
2358 	case KVM_REG_PPC_TACR:
2359 		*val = get_reg_val(id, vcpu->arch.tacr);
2360 		break;
2361 	case KVM_REG_PPC_TCSCR:
2362 		*val = get_reg_val(id, vcpu->arch.tcscr);
2363 		break;
2364 	case KVM_REG_PPC_PID:
2365 		*val = get_reg_val(id, kvmppc_get_pid(vcpu));
2366 		break;
2367 	case KVM_REG_PPC_ACOP:
2368 		*val = get_reg_val(id, vcpu->arch.acop);
2369 		break;
2370 	case KVM_REG_PPC_WORT:
2371 		*val = get_reg_val(id, kvmppc_get_wort_hv(vcpu));
2372 		break;
2373 	case KVM_REG_PPC_TIDR:
2374 		*val = get_reg_val(id, vcpu->arch.tid);
2375 		break;
2376 	case KVM_REG_PPC_PSSCR:
2377 		*val = get_reg_val(id, vcpu->arch.psscr);
2378 		break;
2379 	case KVM_REG_PPC_VPA_ADDR:
2380 		spin_lock(&vcpu->arch.vpa_update_lock);
2381 		*val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
2382 		spin_unlock(&vcpu->arch.vpa_update_lock);
2383 		break;
2384 	case KVM_REG_PPC_VPA_SLB:
2385 		spin_lock(&vcpu->arch.vpa_update_lock);
2386 		val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
2387 		val->vpaval.length = vcpu->arch.slb_shadow.len;
2388 		spin_unlock(&vcpu->arch.vpa_update_lock);
2389 		break;
2390 	case KVM_REG_PPC_VPA_DTL:
2391 		spin_lock(&vcpu->arch.vpa_update_lock);
2392 		val->vpaval.addr = vcpu->arch.dtl.next_gpa;
2393 		val->vpaval.length = vcpu->arch.dtl.len;
2394 		spin_unlock(&vcpu->arch.vpa_update_lock);
2395 		break;
2396 	case KVM_REG_PPC_TB_OFFSET:
2397 		*val = get_reg_val(id, kvmppc_get_tb_offset(vcpu));
2398 		break;
2399 	case KVM_REG_PPC_LPCR:
2400 	case KVM_REG_PPC_LPCR_64:
2401 		*val = get_reg_val(id, kvmppc_get_lpcr(vcpu));
2402 		break;
2403 	case KVM_REG_PPC_PPR:
2404 		*val = get_reg_val(id, kvmppc_get_ppr_hv(vcpu));
2405 		break;
2406 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2407 	case KVM_REG_PPC_TFHAR:
2408 		*val = get_reg_val(id, vcpu->arch.tfhar);
2409 		break;
2410 	case KVM_REG_PPC_TFIAR:
2411 		*val = get_reg_val(id, vcpu->arch.tfiar);
2412 		break;
2413 	case KVM_REG_PPC_TEXASR:
2414 		*val = get_reg_val(id, vcpu->arch.texasr);
2415 		break;
2416 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2417 		i = id - KVM_REG_PPC_TM_GPR0;
2418 		*val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
2419 		break;
2420 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2421 	{
2422 		int j;
2423 		i = id - KVM_REG_PPC_TM_VSR0;
2424 		if (i < 32)
2425 			for (j = 0; j < TS_FPRWIDTH; j++)
2426 				val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
2427 		else {
2428 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
2429 				val->vval = vcpu->arch.vr_tm.vr[i-32];
2430 			else
2431 				r = -ENXIO;
2432 		}
2433 		break;
2434 	}
2435 	case KVM_REG_PPC_TM_CR:
2436 		*val = get_reg_val(id, vcpu->arch.cr_tm);
2437 		break;
2438 	case KVM_REG_PPC_TM_XER:
2439 		*val = get_reg_val(id, vcpu->arch.xer_tm);
2440 		break;
2441 	case KVM_REG_PPC_TM_LR:
2442 		*val = get_reg_val(id, vcpu->arch.lr_tm);
2443 		break;
2444 	case KVM_REG_PPC_TM_CTR:
2445 		*val = get_reg_val(id, vcpu->arch.ctr_tm);
2446 		break;
2447 	case KVM_REG_PPC_TM_FPSCR:
2448 		*val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
2449 		break;
2450 	case KVM_REG_PPC_TM_AMR:
2451 		*val = get_reg_val(id, vcpu->arch.amr_tm);
2452 		break;
2453 	case KVM_REG_PPC_TM_PPR:
2454 		*val = get_reg_val(id, vcpu->arch.ppr_tm);
2455 		break;
2456 	case KVM_REG_PPC_TM_VRSAVE:
2457 		*val = get_reg_val(id, vcpu->arch.vrsave_tm);
2458 		break;
2459 	case KVM_REG_PPC_TM_VSCR:
2460 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
2461 			*val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
2462 		else
2463 			r = -ENXIO;
2464 		break;
2465 	case KVM_REG_PPC_TM_DSCR:
2466 		*val = get_reg_val(id, vcpu->arch.dscr_tm);
2467 		break;
2468 	case KVM_REG_PPC_TM_TAR:
2469 		*val = get_reg_val(id, vcpu->arch.tar_tm);
2470 		break;
2471 #endif
2472 	case KVM_REG_PPC_ARCH_COMPAT:
2473 		*val = get_reg_val(id, kvmppc_get_arch_compat(vcpu));
2474 		break;
2475 	case KVM_REG_PPC_DEC_EXPIRY:
2476 		*val = get_reg_val(id, kvmppc_get_dec_expires(vcpu));
2477 		break;
2478 	case KVM_REG_PPC_ONLINE:
2479 		*val = get_reg_val(id, vcpu->arch.online);
2480 		break;
2481 	case KVM_REG_PPC_PTCR:
2482 		*val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
2483 		break;
2484 	case KVM_REG_PPC_FSCR:
2485 		*val = get_reg_val(id, kvmppc_get_fscr_hv(vcpu));
2486 		break;
2487 	default:
2488 		r = -EINVAL;
2489 		break;
2490 	}
2491 
2492 	return r;
2493 }
2494 
2495 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
2496 				 union kvmppc_one_reg *val)
2497 {
2498 	int r = 0;
2499 	long int i;
2500 	unsigned long addr, len;
2501 
2502 	switch (id) {
2503 	case KVM_REG_PPC_HIOR:
2504 		/* Only allow this to be set to zero */
2505 		if (set_reg_val(id, *val))
2506 			r = -EINVAL;
2507 		break;
2508 	case KVM_REG_PPC_DABR:
2509 		vcpu->arch.dabr = set_reg_val(id, *val);
2510 		break;
2511 	case KVM_REG_PPC_DABRX:
2512 		vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
2513 		break;
2514 	case KVM_REG_PPC_DSCR:
2515 		kvmppc_set_dscr_hv(vcpu, set_reg_val(id, *val));
2516 		break;
2517 	case KVM_REG_PPC_PURR:
2518 		kvmppc_set_purr_hv(vcpu, set_reg_val(id, *val));
2519 		break;
2520 	case KVM_REG_PPC_SPURR:
2521 		kvmppc_set_spurr_hv(vcpu, set_reg_val(id, *val));
2522 		break;
2523 	case KVM_REG_PPC_AMR:
2524 		kvmppc_set_amr_hv(vcpu, set_reg_val(id, *val));
2525 		break;
2526 	case KVM_REG_PPC_UAMOR:
2527 		kvmppc_set_uamor_hv(vcpu, set_reg_val(id, *val));
2528 		break;
2529 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
2530 		i = id - KVM_REG_PPC_MMCR0;
2531 		kvmppc_set_mmcr_hv(vcpu, i, set_reg_val(id, *val));
2532 		break;
2533 	case KVM_REG_PPC_MMCR2:
2534 		kvmppc_set_mmcr_hv(vcpu, 2, set_reg_val(id, *val));
2535 		break;
2536 	case KVM_REG_PPC_MMCRA:
2537 		kvmppc_set_mmcra_hv(vcpu, set_reg_val(id, *val));
2538 		break;
2539 	case KVM_REG_PPC_MMCRS:
2540 		vcpu->arch.mmcrs = set_reg_val(id, *val);
2541 		break;
2542 	case KVM_REG_PPC_MMCR3:
2543 		*val = get_reg_val(id, vcpu->arch.mmcr[3]);
2544 		break;
2545 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
2546 		i = id - KVM_REG_PPC_PMC1;
2547 		kvmppc_set_pmc_hv(vcpu, i, set_reg_val(id, *val));
2548 		break;
2549 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
2550 		i = id - KVM_REG_PPC_SPMC1;
2551 		vcpu->arch.spmc[i] = set_reg_val(id, *val);
2552 		break;
2553 	case KVM_REG_PPC_SIAR:
2554 		kvmppc_set_siar_hv(vcpu, set_reg_val(id, *val));
2555 		break;
2556 	case KVM_REG_PPC_SDAR:
2557 		kvmppc_set_sdar_hv(vcpu, set_reg_val(id, *val));
2558 		break;
2559 	case KVM_REG_PPC_SIER:
2560 		kvmppc_set_sier_hv(vcpu, 0, set_reg_val(id, *val));
2561 		break;
2562 	case KVM_REG_PPC_SIER2:
2563 		kvmppc_set_sier_hv(vcpu, 1, set_reg_val(id, *val));
2564 		break;
2565 	case KVM_REG_PPC_SIER3:
2566 		kvmppc_set_sier_hv(vcpu, 2, set_reg_val(id, *val));
2567 		break;
2568 	case KVM_REG_PPC_IAMR:
2569 		kvmppc_set_iamr_hv(vcpu, set_reg_val(id, *val));
2570 		break;
2571 	case KVM_REG_PPC_PSPB:
2572 		kvmppc_set_pspb_hv(vcpu, set_reg_val(id, *val));
2573 		break;
2574 	case KVM_REG_PPC_DPDES:
2575 		if (cpu_has_feature(CPU_FTR_ARCH_300))
2576 			vcpu->arch.doorbell_request = set_reg_val(id, *val) & 1;
2577 		else
2578 			vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
2579 		break;
2580 	case KVM_REG_PPC_VTB:
2581 		kvmppc_set_vtb(vcpu, set_reg_val(id, *val));
2582 		break;
2583 	case KVM_REG_PPC_DAWR:
2584 		kvmppc_set_dawr0_hv(vcpu, set_reg_val(id, *val));
2585 		break;
2586 	case KVM_REG_PPC_DAWRX:
2587 		kvmppc_set_dawrx0_hv(vcpu, set_reg_val(id, *val) & ~DAWRX_HYP);
2588 		break;
2589 	case KVM_REG_PPC_DAWR1:
2590 		kvmppc_set_dawr1_hv(vcpu, set_reg_val(id, *val));
2591 		break;
2592 	case KVM_REG_PPC_DAWRX1:
2593 		kvmppc_set_dawrx1_hv(vcpu, set_reg_val(id, *val) & ~DAWRX_HYP);
2594 		break;
2595 	case KVM_REG_PPC_CIABR:
2596 		kvmppc_set_ciabr_hv(vcpu, set_reg_val(id, *val));
2597 		/* Don't allow setting breakpoints in hypervisor code */
2598 		if ((kvmppc_get_ciabr_hv(vcpu) & CIABR_PRIV) == CIABR_PRIV_HYPER)
2599 			kvmppc_set_ciabr_hv(vcpu, kvmppc_get_ciabr_hv(vcpu) & ~CIABR_PRIV);
2600 		break;
2601 	case KVM_REG_PPC_CSIGR:
2602 		vcpu->arch.csigr = set_reg_val(id, *val);
2603 		break;
2604 	case KVM_REG_PPC_TACR:
2605 		vcpu->arch.tacr = set_reg_val(id, *val);
2606 		break;
2607 	case KVM_REG_PPC_TCSCR:
2608 		vcpu->arch.tcscr = set_reg_val(id, *val);
2609 		break;
2610 	case KVM_REG_PPC_PID:
2611 		kvmppc_set_pid(vcpu, set_reg_val(id, *val));
2612 		break;
2613 	case KVM_REG_PPC_ACOP:
2614 		vcpu->arch.acop = set_reg_val(id, *val);
2615 		break;
2616 	case KVM_REG_PPC_WORT:
2617 		kvmppc_set_wort_hv(vcpu, set_reg_val(id, *val));
2618 		break;
2619 	case KVM_REG_PPC_TIDR:
2620 		vcpu->arch.tid = set_reg_val(id, *val);
2621 		break;
2622 	case KVM_REG_PPC_PSSCR:
2623 		vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
2624 		break;
2625 	case KVM_REG_PPC_VPA_ADDR:
2626 		addr = set_reg_val(id, *val);
2627 		r = -EINVAL;
2628 		if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
2629 			      vcpu->arch.dtl.next_gpa))
2630 			break;
2631 		r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
2632 		break;
2633 	case KVM_REG_PPC_VPA_SLB:
2634 		addr = val->vpaval.addr;
2635 		len = val->vpaval.length;
2636 		r = -EINVAL;
2637 		if (addr && !vcpu->arch.vpa.next_gpa)
2638 			break;
2639 		r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
2640 		break;
2641 	case KVM_REG_PPC_VPA_DTL:
2642 		addr = val->vpaval.addr;
2643 		len = val->vpaval.length;
2644 		r = -EINVAL;
2645 		if (addr && (len < sizeof(struct dtl_entry) ||
2646 			     !vcpu->arch.vpa.next_gpa))
2647 			break;
2648 		len -= len % sizeof(struct dtl_entry);
2649 		r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
2650 		break;
2651 	case KVM_REG_PPC_TB_OFFSET:
2652 	{
2653 		/* round up to multiple of 2^24 */
2654 		u64 tb_offset = ALIGN(set_reg_val(id, *val), 1UL << 24);
2655 
2656 		/*
2657 		 * Now that we know the timebase offset, update the
2658 		 * decrementer expiry with a guest timebase value. If
2659 		 * the userspace does not set DEC_EXPIRY, this ensures
2660 		 * a migrated vcpu at least starts with an expired
2661 		 * decrementer, which is better than a large one that
2662 		 * causes a hang.
2663 		 */
2664 		kvmppc_set_tb_offset(vcpu, tb_offset);
2665 		if (!kvmppc_get_dec_expires(vcpu) && tb_offset)
2666 			kvmppc_set_dec_expires(vcpu, get_tb() + tb_offset);
2667 
2668 		kvmppc_set_tb_offset(vcpu, tb_offset);
2669 		break;
2670 	}
2671 	case KVM_REG_PPC_LPCR:
2672 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
2673 		break;
2674 	case KVM_REG_PPC_LPCR_64:
2675 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
2676 		break;
2677 	case KVM_REG_PPC_PPR:
2678 		kvmppc_set_ppr_hv(vcpu, set_reg_val(id, *val));
2679 		break;
2680 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2681 	case KVM_REG_PPC_TFHAR:
2682 		vcpu->arch.tfhar = set_reg_val(id, *val);
2683 		break;
2684 	case KVM_REG_PPC_TFIAR:
2685 		vcpu->arch.tfiar = set_reg_val(id, *val);
2686 		break;
2687 	case KVM_REG_PPC_TEXASR:
2688 		vcpu->arch.texasr = set_reg_val(id, *val);
2689 		break;
2690 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2691 		i = id - KVM_REG_PPC_TM_GPR0;
2692 		vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
2693 		break;
2694 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2695 	{
2696 		int j;
2697 		i = id - KVM_REG_PPC_TM_VSR0;
2698 		if (i < 32)
2699 			for (j = 0; j < TS_FPRWIDTH; j++)
2700 				vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
2701 		else
2702 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
2703 				vcpu->arch.vr_tm.vr[i-32] = val->vval;
2704 			else
2705 				r = -ENXIO;
2706 		break;
2707 	}
2708 	case KVM_REG_PPC_TM_CR:
2709 		vcpu->arch.cr_tm = set_reg_val(id, *val);
2710 		break;
2711 	case KVM_REG_PPC_TM_XER:
2712 		vcpu->arch.xer_tm = set_reg_val(id, *val);
2713 		break;
2714 	case KVM_REG_PPC_TM_LR:
2715 		vcpu->arch.lr_tm = set_reg_val(id, *val);
2716 		break;
2717 	case KVM_REG_PPC_TM_CTR:
2718 		vcpu->arch.ctr_tm = set_reg_val(id, *val);
2719 		break;
2720 	case KVM_REG_PPC_TM_FPSCR:
2721 		vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
2722 		break;
2723 	case KVM_REG_PPC_TM_AMR:
2724 		vcpu->arch.amr_tm = set_reg_val(id, *val);
2725 		break;
2726 	case KVM_REG_PPC_TM_PPR:
2727 		vcpu->arch.ppr_tm = set_reg_val(id, *val);
2728 		break;
2729 	case KVM_REG_PPC_TM_VRSAVE:
2730 		vcpu->arch.vrsave_tm = set_reg_val(id, *val);
2731 		break;
2732 	case KVM_REG_PPC_TM_VSCR:
2733 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
2734 			vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
2735 		else
2736 			r = - ENXIO;
2737 		break;
2738 	case KVM_REG_PPC_TM_DSCR:
2739 		vcpu->arch.dscr_tm = set_reg_val(id, *val);
2740 		break;
2741 	case KVM_REG_PPC_TM_TAR:
2742 		vcpu->arch.tar_tm = set_reg_val(id, *val);
2743 		break;
2744 #endif
2745 	case KVM_REG_PPC_ARCH_COMPAT:
2746 		r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
2747 		break;
2748 	case KVM_REG_PPC_DEC_EXPIRY:
2749 		kvmppc_set_dec_expires(vcpu, set_reg_val(id, *val));
2750 		break;
2751 	case KVM_REG_PPC_ONLINE:
2752 		i = set_reg_val(id, *val);
2753 		if (i && !vcpu->arch.online)
2754 			atomic_inc(&vcpu->arch.vcore->online_count);
2755 		else if (!i && vcpu->arch.online)
2756 			atomic_dec(&vcpu->arch.vcore->online_count);
2757 		vcpu->arch.online = i;
2758 		break;
2759 	case KVM_REG_PPC_PTCR:
2760 		vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
2761 		break;
2762 	case KVM_REG_PPC_FSCR:
2763 		kvmppc_set_fscr_hv(vcpu, set_reg_val(id, *val));
2764 		break;
2765 	default:
2766 		r = -EINVAL;
2767 		break;
2768 	}
2769 
2770 	return r;
2771 }
2772 
2773 /*
2774  * On POWER9, threads are independent and can be in different partitions.
2775  * Therefore we consider each thread to be a subcore.
2776  * There is a restriction that all threads have to be in the same
2777  * MMU mode (radix or HPT), unfortunately, but since we only support
2778  * HPT guests on a HPT host so far, that isn't an impediment yet.
2779  */
2780 static int threads_per_vcore(struct kvm *kvm)
2781 {
2782 	if (cpu_has_feature(CPU_FTR_ARCH_300))
2783 		return 1;
2784 	return threads_per_subcore;
2785 }
2786 
2787 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
2788 {
2789 	struct kvmppc_vcore *vcore;
2790 
2791 	vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
2792 
2793 	if (vcore == NULL)
2794 		return NULL;
2795 
2796 	spin_lock_init(&vcore->lock);
2797 	spin_lock_init(&vcore->stoltb_lock);
2798 	rcuwait_init(&vcore->wait);
2799 	vcore->preempt_tb = TB_NIL;
2800 	vcore->lpcr = kvm->arch.lpcr;
2801 	vcore->first_vcpuid = id;
2802 	vcore->kvm = kvm;
2803 	INIT_LIST_HEAD(&vcore->preempt_list);
2804 
2805 	return vcore;
2806 }
2807 
2808 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2809 static struct debugfs_timings_element {
2810 	const char *name;
2811 	size_t offset;
2812 } timings[] = {
2813 #ifdef CONFIG_KVM_BOOK3S_HV_P9_TIMING
2814 	{"vcpu_entry",	offsetof(struct kvm_vcpu, arch.vcpu_entry)},
2815 	{"guest_entry",	offsetof(struct kvm_vcpu, arch.guest_entry)},
2816 	{"in_guest",	offsetof(struct kvm_vcpu, arch.in_guest)},
2817 	{"guest_exit",	offsetof(struct kvm_vcpu, arch.guest_exit)},
2818 	{"vcpu_exit",	offsetof(struct kvm_vcpu, arch.vcpu_exit)},
2819 	{"hypercall",	offsetof(struct kvm_vcpu, arch.hcall)},
2820 	{"page_fault",	offsetof(struct kvm_vcpu, arch.pg_fault)},
2821 #else
2822 	{"rm_entry",	offsetof(struct kvm_vcpu, arch.rm_entry)},
2823 	{"rm_intr",	offsetof(struct kvm_vcpu, arch.rm_intr)},
2824 	{"rm_exit",	offsetof(struct kvm_vcpu, arch.rm_exit)},
2825 	{"guest",	offsetof(struct kvm_vcpu, arch.guest_time)},
2826 	{"cede",	offsetof(struct kvm_vcpu, arch.cede_time)},
2827 #endif
2828 };
2829 
2830 #define N_TIMINGS	(ARRAY_SIZE(timings))
2831 
2832 struct debugfs_timings_state {
2833 	struct kvm_vcpu	*vcpu;
2834 	unsigned int	buflen;
2835 	char		buf[N_TIMINGS * 100];
2836 };
2837 
2838 static int debugfs_timings_open(struct inode *inode, struct file *file)
2839 {
2840 	struct kvm_vcpu *vcpu = inode->i_private;
2841 	struct debugfs_timings_state *p;
2842 
2843 	p = kzalloc(sizeof(*p), GFP_KERNEL);
2844 	if (!p)
2845 		return -ENOMEM;
2846 
2847 	kvm_get_kvm(vcpu->kvm);
2848 	p->vcpu = vcpu;
2849 	file->private_data = p;
2850 
2851 	return nonseekable_open(inode, file);
2852 }
2853 
2854 static int debugfs_timings_release(struct inode *inode, struct file *file)
2855 {
2856 	struct debugfs_timings_state *p = file->private_data;
2857 
2858 	kvm_put_kvm(p->vcpu->kvm);
2859 	kfree(p);
2860 	return 0;
2861 }
2862 
2863 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2864 				    size_t len, loff_t *ppos)
2865 {
2866 	struct debugfs_timings_state *p = file->private_data;
2867 	struct kvm_vcpu *vcpu = p->vcpu;
2868 	char *s, *buf_end;
2869 	struct kvmhv_tb_accumulator tb;
2870 	u64 count;
2871 	loff_t pos;
2872 	ssize_t n;
2873 	int i, loops;
2874 	bool ok;
2875 
2876 	if (!p->buflen) {
2877 		s = p->buf;
2878 		buf_end = s + sizeof(p->buf);
2879 		for (i = 0; i < N_TIMINGS; ++i) {
2880 			struct kvmhv_tb_accumulator *acc;
2881 
2882 			acc = (struct kvmhv_tb_accumulator *)
2883 				((unsigned long)vcpu + timings[i].offset);
2884 			ok = false;
2885 			for (loops = 0; loops < 1000; ++loops) {
2886 				count = acc->seqcount;
2887 				if (!(count & 1)) {
2888 					smp_rmb();
2889 					tb = *acc;
2890 					smp_rmb();
2891 					if (count == acc->seqcount) {
2892 						ok = true;
2893 						break;
2894 					}
2895 				}
2896 				udelay(1);
2897 			}
2898 			if (!ok)
2899 				snprintf(s, buf_end - s, "%s: stuck\n",
2900 					timings[i].name);
2901 			else
2902 				snprintf(s, buf_end - s,
2903 					"%s: %llu %llu %llu %llu\n",
2904 					timings[i].name, count / 2,
2905 					tb_to_ns(tb.tb_total),
2906 					tb_to_ns(tb.tb_min),
2907 					tb_to_ns(tb.tb_max));
2908 			s += strlen(s);
2909 		}
2910 		p->buflen = s - p->buf;
2911 	}
2912 
2913 	pos = *ppos;
2914 	if (pos >= p->buflen)
2915 		return 0;
2916 	if (len > p->buflen - pos)
2917 		len = p->buflen - pos;
2918 	n = copy_to_user(buf, p->buf + pos, len);
2919 	if (n) {
2920 		if (n == len)
2921 			return -EFAULT;
2922 		len -= n;
2923 	}
2924 	*ppos = pos + len;
2925 	return len;
2926 }
2927 
2928 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2929 				     size_t len, loff_t *ppos)
2930 {
2931 	return -EACCES;
2932 }
2933 
2934 static const struct file_operations debugfs_timings_ops = {
2935 	.owner	 = THIS_MODULE,
2936 	.open	 = debugfs_timings_open,
2937 	.release = debugfs_timings_release,
2938 	.read	 = debugfs_timings_read,
2939 	.write	 = debugfs_timings_write,
2940 	.llseek	 = generic_file_llseek,
2941 };
2942 
2943 /* Create a debugfs directory for the vcpu */
2944 static int kvmppc_arch_create_vcpu_debugfs_hv(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry)
2945 {
2946 	if (cpu_has_feature(CPU_FTR_ARCH_300) == IS_ENABLED(CONFIG_KVM_BOOK3S_HV_P9_TIMING))
2947 		debugfs_create_file("timings", 0444, debugfs_dentry, vcpu,
2948 				    &debugfs_timings_ops);
2949 	return 0;
2950 }
2951 
2952 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2953 static int kvmppc_arch_create_vcpu_debugfs_hv(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry)
2954 {
2955 	return 0;
2956 }
2957 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2958 
2959 static int kvmppc_core_vcpu_create_hv(struct kvm_vcpu *vcpu)
2960 {
2961 	int err;
2962 	int core;
2963 	struct kvmppc_vcore *vcore;
2964 	struct kvm *kvm;
2965 	unsigned int id;
2966 
2967 	kvm = vcpu->kvm;
2968 	id = vcpu->vcpu_id;
2969 
2970 	vcpu->arch.shared = &vcpu->arch.shregs;
2971 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2972 	/*
2973 	 * The shared struct is never shared on HV,
2974 	 * so we can always use host endianness
2975 	 */
2976 #ifdef __BIG_ENDIAN__
2977 	vcpu->arch.shared_big_endian = true;
2978 #else
2979 	vcpu->arch.shared_big_endian = false;
2980 #endif
2981 #endif
2982 
2983 	if (kvmhv_is_nestedv2()) {
2984 		err = kvmhv_nestedv2_vcpu_create(vcpu, &vcpu->arch.nestedv2_io);
2985 		if (err < 0)
2986 			return err;
2987 	}
2988 
2989 	kvmppc_set_mmcr_hv(vcpu, 0, MMCR0_FC);
2990 	if (cpu_has_feature(CPU_FTR_ARCH_31)) {
2991 		kvmppc_set_mmcr_hv(vcpu, 0, kvmppc_get_mmcr_hv(vcpu, 0) | MMCR0_PMCCEXT);
2992 		kvmppc_set_mmcra_hv(vcpu, MMCRA_BHRB_DISABLE);
2993 	}
2994 
2995 	kvmppc_set_ctrl_hv(vcpu, CTRL_RUNLATCH);
2996 	/* default to host PVR, since we can't spoof it */
2997 	kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2998 	spin_lock_init(&vcpu->arch.vpa_update_lock);
2999 	spin_lock_init(&vcpu->arch.tbacct_lock);
3000 	vcpu->arch.busy_preempt = TB_NIL;
3001 	__kvmppc_set_msr_hv(vcpu, MSR_ME);
3002 	vcpu->arch.intr_msr = MSR_SF | MSR_ME;
3003 
3004 	/*
3005 	 * Set the default HFSCR for the guest from the host value.
3006 	 * This value is only used on POWER9 and later.
3007 	 * On >= POWER9, we want to virtualize the doorbell facility, so we
3008 	 * don't set the HFSCR_MSGP bit, and that causes those instructions
3009 	 * to trap and then we emulate them.
3010 	 */
3011 	kvmppc_set_hfscr_hv(vcpu, HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
3012 			    HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP);
3013 
3014 	/* On POWER10 and later, allow prefixed instructions */
3015 	if (cpu_has_feature(CPU_FTR_ARCH_31))
3016 		kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) | HFSCR_PREFIX);
3017 
3018 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
3019 		kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) & mfspr(SPRN_HFSCR));
3020 
3021 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
3022 		if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3023 			kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) | HFSCR_TM);
3024 #endif
3025 	}
3026 	if (cpu_has_feature(CPU_FTR_TM_COMP))
3027 		vcpu->arch.hfscr |= HFSCR_TM;
3028 
3029 	vcpu->arch.hfscr_permitted = kvmppc_get_hfscr_hv(vcpu);
3030 
3031 	/*
3032 	 * PM, EBB, TM are demand-faulted so start with it clear.
3033 	 */
3034 	kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) & ~(HFSCR_PM | HFSCR_EBB | HFSCR_TM));
3035 
3036 	kvmppc_mmu_book3s_hv_init(vcpu);
3037 
3038 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
3039 
3040 	init_waitqueue_head(&vcpu->arch.cpu_run);
3041 
3042 	mutex_lock(&kvm->lock);
3043 	vcore = NULL;
3044 	err = -EINVAL;
3045 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
3046 		if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
3047 			pr_devel("KVM: VCPU ID too high\n");
3048 			core = KVM_MAX_VCORES;
3049 		} else {
3050 			BUG_ON(kvm->arch.smt_mode != 1);
3051 			core = kvmppc_pack_vcpu_id(kvm, id);
3052 		}
3053 	} else {
3054 		core = id / kvm->arch.smt_mode;
3055 	}
3056 	if (core < KVM_MAX_VCORES) {
3057 		vcore = kvm->arch.vcores[core];
3058 		if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
3059 			pr_devel("KVM: collision on id %u", id);
3060 			vcore = NULL;
3061 		} else if (!vcore) {
3062 			/*
3063 			 * Take mmu_setup_lock for mutual exclusion
3064 			 * with kvmppc_update_lpcr().
3065 			 */
3066 			err = -ENOMEM;
3067 			vcore = kvmppc_vcore_create(kvm,
3068 					id & ~(kvm->arch.smt_mode - 1));
3069 			mutex_lock(&kvm->arch.mmu_setup_lock);
3070 			kvm->arch.vcores[core] = vcore;
3071 			kvm->arch.online_vcores++;
3072 			mutex_unlock(&kvm->arch.mmu_setup_lock);
3073 		}
3074 	}
3075 	mutex_unlock(&kvm->lock);
3076 
3077 	if (!vcore)
3078 		return err;
3079 
3080 	spin_lock(&vcore->lock);
3081 	++vcore->num_threads;
3082 	spin_unlock(&vcore->lock);
3083 	vcpu->arch.vcore = vcore;
3084 	vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
3085 	vcpu->arch.thread_cpu = -1;
3086 	vcpu->arch.prev_cpu = -1;
3087 
3088 	vcpu->arch.cpu_type = KVM_CPU_3S_64;
3089 	kvmppc_sanity_check(vcpu);
3090 
3091 	return 0;
3092 }
3093 
3094 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
3095 			      unsigned long flags)
3096 {
3097 	int err;
3098 	int esmt = 0;
3099 
3100 	if (flags)
3101 		return -EINVAL;
3102 	if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
3103 		return -EINVAL;
3104 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
3105 		/*
3106 		 * On POWER8 (or POWER7), the threading mode is "strict",
3107 		 * so we pack smt_mode vcpus per vcore.
3108 		 */
3109 		if (smt_mode > threads_per_subcore)
3110 			return -EINVAL;
3111 	} else {
3112 		/*
3113 		 * On POWER9, the threading mode is "loose",
3114 		 * so each vcpu gets its own vcore.
3115 		 */
3116 		esmt = smt_mode;
3117 		smt_mode = 1;
3118 	}
3119 	mutex_lock(&kvm->lock);
3120 	err = -EBUSY;
3121 	if (!kvm->arch.online_vcores) {
3122 		kvm->arch.smt_mode = smt_mode;
3123 		kvm->arch.emul_smt_mode = esmt;
3124 		err = 0;
3125 	}
3126 	mutex_unlock(&kvm->lock);
3127 
3128 	return err;
3129 }
3130 
3131 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
3132 {
3133 	if (vpa->pinned_addr)
3134 		kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
3135 					vpa->dirty);
3136 }
3137 
3138 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
3139 {
3140 	spin_lock(&vcpu->arch.vpa_update_lock);
3141 	unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
3142 	unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
3143 	unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
3144 	spin_unlock(&vcpu->arch.vpa_update_lock);
3145 	if (kvmhv_is_nestedv2())
3146 		kvmhv_nestedv2_vcpu_free(vcpu, &vcpu->arch.nestedv2_io);
3147 }
3148 
3149 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
3150 {
3151 	/* Indicate we want to get back into the guest */
3152 	return 1;
3153 }
3154 
3155 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
3156 {
3157 	unsigned long dec_nsec, now;
3158 
3159 	now = get_tb();
3160 	if (now > kvmppc_dec_expires_host_tb(vcpu)) {
3161 		/* decrementer has already gone negative */
3162 		kvmppc_core_queue_dec(vcpu);
3163 		kvmppc_core_prepare_to_enter(vcpu);
3164 		return;
3165 	}
3166 	dec_nsec = tb_to_ns(kvmppc_dec_expires_host_tb(vcpu) - now);
3167 	hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
3168 	vcpu->arch.timer_running = 1;
3169 }
3170 
3171 extern int __kvmppc_vcore_entry(void);
3172 
3173 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
3174 				   struct kvm_vcpu *vcpu, u64 tb)
3175 {
3176 	u64 now;
3177 
3178 	if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3179 		return;
3180 	spin_lock_irq(&vcpu->arch.tbacct_lock);
3181 	now = tb;
3182 	vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
3183 		vcpu->arch.stolen_logged;
3184 	vcpu->arch.busy_preempt = now;
3185 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
3186 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
3187 	--vc->n_runnable;
3188 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
3189 }
3190 
3191 static int kvmppc_grab_hwthread(int cpu)
3192 {
3193 	struct paca_struct *tpaca;
3194 	long timeout = 10000;
3195 
3196 	tpaca = paca_ptrs[cpu];
3197 
3198 	/* Ensure the thread won't go into the kernel if it wakes */
3199 	tpaca->kvm_hstate.kvm_vcpu = NULL;
3200 	tpaca->kvm_hstate.kvm_vcore = NULL;
3201 	tpaca->kvm_hstate.napping = 0;
3202 	smp_wmb();
3203 	tpaca->kvm_hstate.hwthread_req = 1;
3204 
3205 	/*
3206 	 * If the thread is already executing in the kernel (e.g. handling
3207 	 * a stray interrupt), wait for it to get back to nap mode.
3208 	 * The smp_mb() is to ensure that our setting of hwthread_req
3209 	 * is visible before we look at hwthread_state, so if this
3210 	 * races with the code at system_reset_pSeries and the thread
3211 	 * misses our setting of hwthread_req, we are sure to see its
3212 	 * setting of hwthread_state, and vice versa.
3213 	 */
3214 	smp_mb();
3215 	while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
3216 		if (--timeout <= 0) {
3217 			pr_err("KVM: couldn't grab cpu %d\n", cpu);
3218 			return -EBUSY;
3219 		}
3220 		udelay(1);
3221 	}
3222 	return 0;
3223 }
3224 
3225 static void kvmppc_release_hwthread(int cpu)
3226 {
3227 	struct paca_struct *tpaca;
3228 
3229 	tpaca = paca_ptrs[cpu];
3230 	tpaca->kvm_hstate.hwthread_req = 0;
3231 	tpaca->kvm_hstate.kvm_vcpu = NULL;
3232 	tpaca->kvm_hstate.kvm_vcore = NULL;
3233 	tpaca->kvm_hstate.kvm_split_mode = NULL;
3234 }
3235 
3236 static DEFINE_PER_CPU(struct kvm *, cpu_in_guest);
3237 
3238 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
3239 {
3240 	struct kvm_nested_guest *nested = vcpu->arch.nested;
3241 	cpumask_t *need_tlb_flush;
3242 	int i;
3243 
3244 	if (nested)
3245 		need_tlb_flush = &nested->need_tlb_flush;
3246 	else
3247 		need_tlb_flush = &kvm->arch.need_tlb_flush;
3248 
3249 	cpu = cpu_first_tlb_thread_sibling(cpu);
3250 	for (i = cpu; i <= cpu_last_tlb_thread_sibling(cpu);
3251 					i += cpu_tlb_thread_sibling_step())
3252 		cpumask_set_cpu(i, need_tlb_flush);
3253 
3254 	/*
3255 	 * Make sure setting of bit in need_tlb_flush precedes testing of
3256 	 * cpu_in_guest. The matching barrier on the other side is hwsync
3257 	 * when switching to guest MMU mode, which happens between
3258 	 * cpu_in_guest being set to the guest kvm, and need_tlb_flush bit
3259 	 * being tested.
3260 	 */
3261 	smp_mb();
3262 
3263 	for (i = cpu; i <= cpu_last_tlb_thread_sibling(cpu);
3264 					i += cpu_tlb_thread_sibling_step()) {
3265 		struct kvm *running = *per_cpu_ptr(&cpu_in_guest, i);
3266 
3267 		if (running == kvm)
3268 			smp_call_function_single(i, do_nothing, NULL, 1);
3269 	}
3270 }
3271 
3272 static void do_migrate_away_vcpu(void *arg)
3273 {
3274 	struct kvm_vcpu *vcpu = arg;
3275 	struct kvm *kvm = vcpu->kvm;
3276 
3277 	/*
3278 	 * If the guest has GTSE, it may execute tlbie, so do a eieio; tlbsync;
3279 	 * ptesync sequence on the old CPU before migrating to a new one, in
3280 	 * case we interrupted the guest between a tlbie ; eieio ;
3281 	 * tlbsync; ptesync sequence.
3282 	 *
3283 	 * Otherwise, ptesync is sufficient for ordering tlbiel sequences.
3284 	 */
3285 	if (kvm->arch.lpcr & LPCR_GTSE)
3286 		asm volatile("eieio; tlbsync; ptesync");
3287 	else
3288 		asm volatile("ptesync");
3289 }
3290 
3291 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
3292 {
3293 	struct kvm_nested_guest *nested = vcpu->arch.nested;
3294 	struct kvm *kvm = vcpu->kvm;
3295 	int prev_cpu;
3296 
3297 	if (!cpu_has_feature(CPU_FTR_HVMODE))
3298 		return;
3299 
3300 	if (nested)
3301 		prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
3302 	else
3303 		prev_cpu = vcpu->arch.prev_cpu;
3304 
3305 	/*
3306 	 * With radix, the guest can do TLB invalidations itself,
3307 	 * and it could choose to use the local form (tlbiel) if
3308 	 * it is invalidating a translation that has only ever been
3309 	 * used on one vcpu.  However, that doesn't mean it has
3310 	 * only ever been used on one physical cpu, since vcpus
3311 	 * can move around between pcpus.  To cope with this, when
3312 	 * a vcpu moves from one pcpu to another, we need to tell
3313 	 * any vcpus running on the same core as this vcpu previously
3314 	 * ran to flush the TLB.
3315 	 */
3316 	if (prev_cpu != pcpu) {
3317 		if (prev_cpu >= 0) {
3318 			if (cpu_first_tlb_thread_sibling(prev_cpu) !=
3319 			    cpu_first_tlb_thread_sibling(pcpu))
3320 				radix_flush_cpu(kvm, prev_cpu, vcpu);
3321 
3322 			smp_call_function_single(prev_cpu,
3323 					do_migrate_away_vcpu, vcpu, 1);
3324 		}
3325 		if (nested)
3326 			nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
3327 		else
3328 			vcpu->arch.prev_cpu = pcpu;
3329 	}
3330 }
3331 
3332 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
3333 {
3334 	int cpu;
3335 	struct paca_struct *tpaca;
3336 
3337 	cpu = vc->pcpu;
3338 	if (vcpu) {
3339 		if (vcpu->arch.timer_running) {
3340 			hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
3341 			vcpu->arch.timer_running = 0;
3342 		}
3343 		cpu += vcpu->arch.ptid;
3344 		vcpu->cpu = vc->pcpu;
3345 		vcpu->arch.thread_cpu = cpu;
3346 	}
3347 	tpaca = paca_ptrs[cpu];
3348 	tpaca->kvm_hstate.kvm_vcpu = vcpu;
3349 	tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
3350 	tpaca->kvm_hstate.fake_suspend = 0;
3351 	/* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
3352 	smp_wmb();
3353 	tpaca->kvm_hstate.kvm_vcore = vc;
3354 	if (cpu != smp_processor_id())
3355 		kvmppc_ipi_thread(cpu);
3356 }
3357 
3358 static void kvmppc_wait_for_nap(int n_threads)
3359 {
3360 	int cpu = smp_processor_id();
3361 	int i, loops;
3362 
3363 	if (n_threads <= 1)
3364 		return;
3365 	for (loops = 0; loops < 1000000; ++loops) {
3366 		/*
3367 		 * Check if all threads are finished.
3368 		 * We set the vcore pointer when starting a thread
3369 		 * and the thread clears it when finished, so we look
3370 		 * for any threads that still have a non-NULL vcore ptr.
3371 		 */
3372 		for (i = 1; i < n_threads; ++i)
3373 			if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
3374 				break;
3375 		if (i == n_threads) {
3376 			HMT_medium();
3377 			return;
3378 		}
3379 		HMT_low();
3380 	}
3381 	HMT_medium();
3382 	for (i = 1; i < n_threads; ++i)
3383 		if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
3384 			pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
3385 }
3386 
3387 /*
3388  * Check that we are on thread 0 and that any other threads in
3389  * this core are off-line.  Then grab the threads so they can't
3390  * enter the kernel.
3391  */
3392 static int on_primary_thread(void)
3393 {
3394 	int cpu = smp_processor_id();
3395 	int thr;
3396 
3397 	/* Are we on a primary subcore? */
3398 	if (cpu_thread_in_subcore(cpu))
3399 		return 0;
3400 
3401 	thr = 0;
3402 	while (++thr < threads_per_subcore)
3403 		if (cpu_online(cpu + thr))
3404 			return 0;
3405 
3406 	/* Grab all hw threads so they can't go into the kernel */
3407 	for (thr = 1; thr < threads_per_subcore; ++thr) {
3408 		if (kvmppc_grab_hwthread(cpu + thr)) {
3409 			/* Couldn't grab one; let the others go */
3410 			do {
3411 				kvmppc_release_hwthread(cpu + thr);
3412 			} while (--thr > 0);
3413 			return 0;
3414 		}
3415 	}
3416 	return 1;
3417 }
3418 
3419 /*
3420  * A list of virtual cores for each physical CPU.
3421  * These are vcores that could run but their runner VCPU tasks are
3422  * (or may be) preempted.
3423  */
3424 struct preempted_vcore_list {
3425 	struct list_head	list;
3426 	spinlock_t		lock;
3427 };
3428 
3429 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
3430 
3431 static void init_vcore_lists(void)
3432 {
3433 	int cpu;
3434 
3435 	for_each_possible_cpu(cpu) {
3436 		struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
3437 		spin_lock_init(&lp->lock);
3438 		INIT_LIST_HEAD(&lp->list);
3439 	}
3440 }
3441 
3442 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
3443 {
3444 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
3445 
3446 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
3447 
3448 	vc->vcore_state = VCORE_PREEMPT;
3449 	vc->pcpu = smp_processor_id();
3450 	if (vc->num_threads < threads_per_vcore(vc->kvm)) {
3451 		spin_lock(&lp->lock);
3452 		list_add_tail(&vc->preempt_list, &lp->list);
3453 		spin_unlock(&lp->lock);
3454 	}
3455 
3456 	/* Start accumulating stolen time */
3457 	kvmppc_core_start_stolen(vc, mftb());
3458 }
3459 
3460 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
3461 {
3462 	struct preempted_vcore_list *lp;
3463 
3464 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
3465 
3466 	kvmppc_core_end_stolen(vc, mftb());
3467 	if (!list_empty(&vc->preempt_list)) {
3468 		lp = &per_cpu(preempted_vcores, vc->pcpu);
3469 		spin_lock(&lp->lock);
3470 		list_del_init(&vc->preempt_list);
3471 		spin_unlock(&lp->lock);
3472 	}
3473 	vc->vcore_state = VCORE_INACTIVE;
3474 }
3475 
3476 /*
3477  * This stores information about the virtual cores currently
3478  * assigned to a physical core.
3479  */
3480 struct core_info {
3481 	int		n_subcores;
3482 	int		max_subcore_threads;
3483 	int		total_threads;
3484 	int		subcore_threads[MAX_SUBCORES];
3485 	struct kvmppc_vcore *vc[MAX_SUBCORES];
3486 };
3487 
3488 /*
3489  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
3490  * respectively in 2-way micro-threading (split-core) mode on POWER8.
3491  */
3492 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
3493 
3494 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
3495 {
3496 	memset(cip, 0, sizeof(*cip));
3497 	cip->n_subcores = 1;
3498 	cip->max_subcore_threads = vc->num_threads;
3499 	cip->total_threads = vc->num_threads;
3500 	cip->subcore_threads[0] = vc->num_threads;
3501 	cip->vc[0] = vc;
3502 }
3503 
3504 static bool subcore_config_ok(int n_subcores, int n_threads)
3505 {
3506 	/*
3507 	 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
3508 	 * split-core mode, with one thread per subcore.
3509 	 */
3510 	if (cpu_has_feature(CPU_FTR_ARCH_300))
3511 		return n_subcores <= 4 && n_threads == 1;
3512 
3513 	/* On POWER8, can only dynamically split if unsplit to begin with */
3514 	if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
3515 		return false;
3516 	if (n_subcores > MAX_SUBCORES)
3517 		return false;
3518 	if (n_subcores > 1) {
3519 		if (!(dynamic_mt_modes & 2))
3520 			n_subcores = 4;
3521 		if (n_subcores > 2 && !(dynamic_mt_modes & 4))
3522 			return false;
3523 	}
3524 
3525 	return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
3526 }
3527 
3528 static void init_vcore_to_run(struct kvmppc_vcore *vc)
3529 {
3530 	vc->entry_exit_map = 0;
3531 	vc->in_guest = 0;
3532 	vc->napping_threads = 0;
3533 	vc->conferring_threads = 0;
3534 	vc->tb_offset_applied = 0;
3535 }
3536 
3537 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
3538 {
3539 	int n_threads = vc->num_threads;
3540 	int sub;
3541 
3542 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
3543 		return false;
3544 
3545 	/* In one_vm_per_core mode, require all vcores to be from the same vm */
3546 	if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
3547 		return false;
3548 
3549 	if (n_threads < cip->max_subcore_threads)
3550 		n_threads = cip->max_subcore_threads;
3551 	if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
3552 		return false;
3553 	cip->max_subcore_threads = n_threads;
3554 
3555 	sub = cip->n_subcores;
3556 	++cip->n_subcores;
3557 	cip->total_threads += vc->num_threads;
3558 	cip->subcore_threads[sub] = vc->num_threads;
3559 	cip->vc[sub] = vc;
3560 	init_vcore_to_run(vc);
3561 	list_del_init(&vc->preempt_list);
3562 
3563 	return true;
3564 }
3565 
3566 /*
3567  * Work out whether it is possible to piggyback the execution of
3568  * vcore *pvc onto the execution of the other vcores described in *cip.
3569  */
3570 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
3571 			  int target_threads)
3572 {
3573 	if (cip->total_threads + pvc->num_threads > target_threads)
3574 		return false;
3575 
3576 	return can_dynamic_split(pvc, cip);
3577 }
3578 
3579 static void prepare_threads(struct kvmppc_vcore *vc)
3580 {
3581 	int i;
3582 	struct kvm_vcpu *vcpu;
3583 
3584 	for_each_runnable_thread(i, vcpu, vc) {
3585 		if (signal_pending(vcpu->arch.run_task))
3586 			vcpu->arch.ret = -EINTR;
3587 		else if (vcpu->arch.vpa.update_pending ||
3588 			 vcpu->arch.slb_shadow.update_pending ||
3589 			 vcpu->arch.dtl.update_pending)
3590 			vcpu->arch.ret = RESUME_GUEST;
3591 		else
3592 			continue;
3593 		kvmppc_remove_runnable(vc, vcpu, mftb());
3594 		wake_up(&vcpu->arch.cpu_run);
3595 	}
3596 }
3597 
3598 static void collect_piggybacks(struct core_info *cip, int target_threads)
3599 {
3600 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
3601 	struct kvmppc_vcore *pvc, *vcnext;
3602 
3603 	spin_lock(&lp->lock);
3604 	list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
3605 		if (!spin_trylock(&pvc->lock))
3606 			continue;
3607 		prepare_threads(pvc);
3608 		if (!pvc->n_runnable || !pvc->kvm->arch.mmu_ready) {
3609 			list_del_init(&pvc->preempt_list);
3610 			if (pvc->runner == NULL) {
3611 				pvc->vcore_state = VCORE_INACTIVE;
3612 				kvmppc_core_end_stolen(pvc, mftb());
3613 			}
3614 			spin_unlock(&pvc->lock);
3615 			continue;
3616 		}
3617 		if (!can_piggyback(pvc, cip, target_threads)) {
3618 			spin_unlock(&pvc->lock);
3619 			continue;
3620 		}
3621 		kvmppc_core_end_stolen(pvc, mftb());
3622 		pvc->vcore_state = VCORE_PIGGYBACK;
3623 		if (cip->total_threads >= target_threads)
3624 			break;
3625 	}
3626 	spin_unlock(&lp->lock);
3627 }
3628 
3629 static bool recheck_signals_and_mmu(struct core_info *cip)
3630 {
3631 	int sub, i;
3632 	struct kvm_vcpu *vcpu;
3633 	struct kvmppc_vcore *vc;
3634 
3635 	for (sub = 0; sub < cip->n_subcores; ++sub) {
3636 		vc = cip->vc[sub];
3637 		if (!vc->kvm->arch.mmu_ready)
3638 			return true;
3639 		for_each_runnable_thread(i, vcpu, vc)
3640 			if (signal_pending(vcpu->arch.run_task))
3641 				return true;
3642 	}
3643 	return false;
3644 }
3645 
3646 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
3647 {
3648 	int still_running = 0, i;
3649 	u64 now;
3650 	long ret;
3651 	struct kvm_vcpu *vcpu;
3652 
3653 	spin_lock(&vc->lock);
3654 	now = get_tb();
3655 	for_each_runnable_thread(i, vcpu, vc) {
3656 		/*
3657 		 * It's safe to unlock the vcore in the loop here, because
3658 		 * for_each_runnable_thread() is safe against removal of
3659 		 * the vcpu, and the vcore state is VCORE_EXITING here,
3660 		 * so any vcpus becoming runnable will have their arch.trap
3661 		 * set to zero and can't actually run in the guest.
3662 		 */
3663 		spin_unlock(&vc->lock);
3664 		/* cancel pending dec exception if dec is positive */
3665 		if (now < kvmppc_dec_expires_host_tb(vcpu) &&
3666 		    kvmppc_core_pending_dec(vcpu))
3667 			kvmppc_core_dequeue_dec(vcpu);
3668 
3669 		trace_kvm_guest_exit(vcpu);
3670 
3671 		ret = RESUME_GUEST;
3672 		if (vcpu->arch.trap)
3673 			ret = kvmppc_handle_exit_hv(vcpu,
3674 						    vcpu->arch.run_task);
3675 
3676 		vcpu->arch.ret = ret;
3677 		vcpu->arch.trap = 0;
3678 
3679 		spin_lock(&vc->lock);
3680 		if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
3681 			if (vcpu->arch.pending_exceptions)
3682 				kvmppc_core_prepare_to_enter(vcpu);
3683 			if (vcpu->arch.ceded)
3684 				kvmppc_set_timer(vcpu);
3685 			else
3686 				++still_running;
3687 		} else {
3688 			kvmppc_remove_runnable(vc, vcpu, mftb());
3689 			wake_up(&vcpu->arch.cpu_run);
3690 		}
3691 	}
3692 	if (!is_master) {
3693 		if (still_running > 0) {
3694 			kvmppc_vcore_preempt(vc);
3695 		} else if (vc->runner) {
3696 			vc->vcore_state = VCORE_PREEMPT;
3697 			kvmppc_core_start_stolen(vc, mftb());
3698 		} else {
3699 			vc->vcore_state = VCORE_INACTIVE;
3700 		}
3701 		if (vc->n_runnable > 0 && vc->runner == NULL) {
3702 			/* make sure there's a candidate runner awake */
3703 			i = -1;
3704 			vcpu = next_runnable_thread(vc, &i);
3705 			wake_up(&vcpu->arch.cpu_run);
3706 		}
3707 	}
3708 	spin_unlock(&vc->lock);
3709 }
3710 
3711 /*
3712  * Clear core from the list of active host cores as we are about to
3713  * enter the guest. Only do this if it is the primary thread of the
3714  * core (not if a subcore) that is entering the guest.
3715  */
3716 static inline int kvmppc_clear_host_core(unsigned int cpu)
3717 {
3718 	int core;
3719 
3720 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3721 		return 0;
3722 	/*
3723 	 * Memory barrier can be omitted here as we will do a smp_wmb()
3724 	 * later in kvmppc_start_thread and we need ensure that state is
3725 	 * visible to other CPUs only after we enter guest.
3726 	 */
3727 	core = cpu >> threads_shift;
3728 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
3729 	return 0;
3730 }
3731 
3732 /*
3733  * Advertise this core as an active host core since we exited the guest
3734  * Only need to do this if it is the primary thread of the core that is
3735  * exiting.
3736  */
3737 static inline int kvmppc_set_host_core(unsigned int cpu)
3738 {
3739 	int core;
3740 
3741 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3742 		return 0;
3743 
3744 	/*
3745 	 * Memory barrier can be omitted here because we do a spin_unlock
3746 	 * immediately after this which provides the memory barrier.
3747 	 */
3748 	core = cpu >> threads_shift;
3749 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
3750 	return 0;
3751 }
3752 
3753 static void set_irq_happened(int trap)
3754 {
3755 	switch (trap) {
3756 	case BOOK3S_INTERRUPT_EXTERNAL:
3757 		local_paca->irq_happened |= PACA_IRQ_EE;
3758 		break;
3759 	case BOOK3S_INTERRUPT_H_DOORBELL:
3760 		local_paca->irq_happened |= PACA_IRQ_DBELL;
3761 		break;
3762 	case BOOK3S_INTERRUPT_HMI:
3763 		local_paca->irq_happened |= PACA_IRQ_HMI;
3764 		break;
3765 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
3766 		replay_system_reset();
3767 		break;
3768 	}
3769 }
3770 
3771 /*
3772  * Run a set of guest threads on a physical core.
3773  * Called with vc->lock held.
3774  */
3775 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
3776 {
3777 	struct kvm_vcpu *vcpu;
3778 	int i;
3779 	int srcu_idx;
3780 	struct core_info core_info;
3781 	struct kvmppc_vcore *pvc;
3782 	struct kvm_split_mode split_info, *sip;
3783 	int split, subcore_size, active;
3784 	int sub;
3785 	bool thr0_done;
3786 	unsigned long cmd_bit, stat_bit;
3787 	int pcpu, thr;
3788 	int target_threads;
3789 	int controlled_threads;
3790 	int trap;
3791 	bool is_power8;
3792 
3793 	if (WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300)))
3794 		return;
3795 
3796 	/*
3797 	 * Remove from the list any threads that have a signal pending
3798 	 * or need a VPA update done
3799 	 */
3800 	prepare_threads(vc);
3801 
3802 	/* if the runner is no longer runnable, let the caller pick a new one */
3803 	if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3804 		return;
3805 
3806 	/*
3807 	 * Initialize *vc.
3808 	 */
3809 	init_vcore_to_run(vc);
3810 	vc->preempt_tb = TB_NIL;
3811 
3812 	/*
3813 	 * Number of threads that we will be controlling: the same as
3814 	 * the number of threads per subcore, except on POWER9,
3815 	 * where it's 1 because the threads are (mostly) independent.
3816 	 */
3817 	controlled_threads = threads_per_vcore(vc->kvm);
3818 
3819 	/*
3820 	 * Make sure we are running on primary threads, and that secondary
3821 	 * threads are offline.  Also check if the number of threads in this
3822 	 * guest are greater than the current system threads per guest.
3823 	 */
3824 	if ((controlled_threads > 1) &&
3825 	    ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
3826 		for_each_runnable_thread(i, vcpu, vc) {
3827 			vcpu->arch.ret = -EBUSY;
3828 			kvmppc_remove_runnable(vc, vcpu, mftb());
3829 			wake_up(&vcpu->arch.cpu_run);
3830 		}
3831 		goto out;
3832 	}
3833 
3834 	/*
3835 	 * See if we could run any other vcores on the physical core
3836 	 * along with this one.
3837 	 */
3838 	init_core_info(&core_info, vc);
3839 	pcpu = smp_processor_id();
3840 	target_threads = controlled_threads;
3841 	if (target_smt_mode && target_smt_mode < target_threads)
3842 		target_threads = target_smt_mode;
3843 	if (vc->num_threads < target_threads)
3844 		collect_piggybacks(&core_info, target_threads);
3845 
3846 	/*
3847 	 * Hard-disable interrupts, and check resched flag and signals.
3848 	 * If we need to reschedule or deliver a signal, clean up
3849 	 * and return without going into the guest(s).
3850 	 * If the mmu_ready flag has been cleared, don't go into the
3851 	 * guest because that means a HPT resize operation is in progress.
3852 	 */
3853 	local_irq_disable();
3854 	hard_irq_disable();
3855 	if (lazy_irq_pending() || need_resched() ||
3856 	    recheck_signals_and_mmu(&core_info)) {
3857 		local_irq_enable();
3858 		vc->vcore_state = VCORE_INACTIVE;
3859 		/* Unlock all except the primary vcore */
3860 		for (sub = 1; sub < core_info.n_subcores; ++sub) {
3861 			pvc = core_info.vc[sub];
3862 			/* Put back on to the preempted vcores list */
3863 			kvmppc_vcore_preempt(pvc);
3864 			spin_unlock(&pvc->lock);
3865 		}
3866 		for (i = 0; i < controlled_threads; ++i)
3867 			kvmppc_release_hwthread(pcpu + i);
3868 		return;
3869 	}
3870 
3871 	kvmppc_clear_host_core(pcpu);
3872 
3873 	/* Decide on micro-threading (split-core) mode */
3874 	subcore_size = threads_per_subcore;
3875 	cmd_bit = stat_bit = 0;
3876 	split = core_info.n_subcores;
3877 	sip = NULL;
3878 	is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S);
3879 
3880 	if (split > 1) {
3881 		sip = &split_info;
3882 		memset(&split_info, 0, sizeof(split_info));
3883 		for (sub = 0; sub < core_info.n_subcores; ++sub)
3884 			split_info.vc[sub] = core_info.vc[sub];
3885 
3886 		if (is_power8) {
3887 			if (split == 2 && (dynamic_mt_modes & 2)) {
3888 				cmd_bit = HID0_POWER8_1TO2LPAR;
3889 				stat_bit = HID0_POWER8_2LPARMODE;
3890 			} else {
3891 				split = 4;
3892 				cmd_bit = HID0_POWER8_1TO4LPAR;
3893 				stat_bit = HID0_POWER8_4LPARMODE;
3894 			}
3895 			subcore_size = MAX_SMT_THREADS / split;
3896 			split_info.rpr = mfspr(SPRN_RPR);
3897 			split_info.pmmar = mfspr(SPRN_PMMAR);
3898 			split_info.ldbar = mfspr(SPRN_LDBAR);
3899 			split_info.subcore_size = subcore_size;
3900 		} else {
3901 			split_info.subcore_size = 1;
3902 		}
3903 
3904 		/* order writes to split_info before kvm_split_mode pointer */
3905 		smp_wmb();
3906 	}
3907 
3908 	for (thr = 0; thr < controlled_threads; ++thr) {
3909 		struct paca_struct *paca = paca_ptrs[pcpu + thr];
3910 
3911 		paca->kvm_hstate.napping = 0;
3912 		paca->kvm_hstate.kvm_split_mode = sip;
3913 	}
3914 
3915 	/* Initiate micro-threading (split-core) on POWER8 if required */
3916 	if (cmd_bit) {
3917 		unsigned long hid0 = mfspr(SPRN_HID0);
3918 
3919 		hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3920 		mb();
3921 		mtspr(SPRN_HID0, hid0);
3922 		isync();
3923 		for (;;) {
3924 			hid0 = mfspr(SPRN_HID0);
3925 			if (hid0 & stat_bit)
3926 				break;
3927 			cpu_relax();
3928 		}
3929 	}
3930 
3931 	/*
3932 	 * On POWER8, set RWMR register.
3933 	 * Since it only affects PURR and SPURR, it doesn't affect
3934 	 * the host, so we don't save/restore the host value.
3935 	 */
3936 	if (is_power8) {
3937 		unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3938 		int n_online = atomic_read(&vc->online_count);
3939 
3940 		/*
3941 		 * Use the 8-thread value if we're doing split-core
3942 		 * or if the vcore's online count looks bogus.
3943 		 */
3944 		if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3945 		    n_online >= 1 && n_online <= MAX_SMT_THREADS)
3946 			rwmr_val = p8_rwmr_values[n_online];
3947 		mtspr(SPRN_RWMR, rwmr_val);
3948 	}
3949 
3950 	/* Start all the threads */
3951 	active = 0;
3952 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
3953 		thr = is_power8 ? subcore_thread_map[sub] : sub;
3954 		thr0_done = false;
3955 		active |= 1 << thr;
3956 		pvc = core_info.vc[sub];
3957 		pvc->pcpu = pcpu + thr;
3958 		for_each_runnable_thread(i, vcpu, pvc) {
3959 			/*
3960 			 * XXX: is kvmppc_start_thread called too late here?
3961 			 * It updates vcpu->cpu and vcpu->arch.thread_cpu
3962 			 * which are used by kvmppc_fast_vcpu_kick_hv(), but
3963 			 * kick is called after new exceptions become available
3964 			 * and exceptions are checked earlier than here, by
3965 			 * kvmppc_core_prepare_to_enter.
3966 			 */
3967 			kvmppc_start_thread(vcpu, pvc);
3968 			kvmppc_update_vpa_dispatch(vcpu, pvc);
3969 			trace_kvm_guest_enter(vcpu);
3970 			if (!vcpu->arch.ptid)
3971 				thr0_done = true;
3972 			active |= 1 << (thr + vcpu->arch.ptid);
3973 		}
3974 		/*
3975 		 * We need to start the first thread of each subcore
3976 		 * even if it doesn't have a vcpu.
3977 		 */
3978 		if (!thr0_done)
3979 			kvmppc_start_thread(NULL, pvc);
3980 	}
3981 
3982 	/*
3983 	 * Ensure that split_info.do_nap is set after setting
3984 	 * the vcore pointer in the PACA of the secondaries.
3985 	 */
3986 	smp_mb();
3987 
3988 	/*
3989 	 * When doing micro-threading, poke the inactive threads as well.
3990 	 * This gets them to the nap instruction after kvm_do_nap,
3991 	 * which reduces the time taken to unsplit later.
3992 	 */
3993 	if (cmd_bit) {
3994 		split_info.do_nap = 1;	/* ask secondaries to nap when done */
3995 		for (thr = 1; thr < threads_per_subcore; ++thr)
3996 			if (!(active & (1 << thr)))
3997 				kvmppc_ipi_thread(pcpu + thr);
3998 	}
3999 
4000 	vc->vcore_state = VCORE_RUNNING;
4001 	preempt_disable();
4002 
4003 	trace_kvmppc_run_core(vc, 0);
4004 
4005 	for (sub = 0; sub < core_info.n_subcores; ++sub)
4006 		spin_unlock(&core_info.vc[sub]->lock);
4007 
4008 	guest_timing_enter_irqoff();
4009 
4010 	srcu_idx = srcu_read_lock(&vc->kvm->srcu);
4011 
4012 	guest_state_enter_irqoff();
4013 	this_cpu_disable_ftrace();
4014 
4015 	trap = __kvmppc_vcore_entry();
4016 
4017 	this_cpu_enable_ftrace();
4018 	guest_state_exit_irqoff();
4019 
4020 	srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
4021 
4022 	set_irq_happened(trap);
4023 
4024 	spin_lock(&vc->lock);
4025 	/* prevent other vcpu threads from doing kvmppc_start_thread() now */
4026 	vc->vcore_state = VCORE_EXITING;
4027 
4028 	/* wait for secondary threads to finish writing their state to memory */
4029 	kvmppc_wait_for_nap(controlled_threads);
4030 
4031 	/* Return to whole-core mode if we split the core earlier */
4032 	if (cmd_bit) {
4033 		unsigned long hid0 = mfspr(SPRN_HID0);
4034 		unsigned long loops = 0;
4035 
4036 		hid0 &= ~HID0_POWER8_DYNLPARDIS;
4037 		stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
4038 		mb();
4039 		mtspr(SPRN_HID0, hid0);
4040 		isync();
4041 		for (;;) {
4042 			hid0 = mfspr(SPRN_HID0);
4043 			if (!(hid0 & stat_bit))
4044 				break;
4045 			cpu_relax();
4046 			++loops;
4047 		}
4048 		split_info.do_nap = 0;
4049 	}
4050 
4051 	kvmppc_set_host_core(pcpu);
4052 
4053 	if (!vtime_accounting_enabled_this_cpu()) {
4054 		local_irq_enable();
4055 		/*
4056 		 * Service IRQs here before guest_timing_exit_irqoff() so any
4057 		 * ticks that occurred while running the guest are accounted to
4058 		 * the guest. If vtime accounting is enabled, accounting uses
4059 		 * TB rather than ticks, so it can be done without enabling
4060 		 * interrupts here, which has the problem that it accounts
4061 		 * interrupt processing overhead to the host.
4062 		 */
4063 		local_irq_disable();
4064 	}
4065 	guest_timing_exit_irqoff();
4066 
4067 	local_irq_enable();
4068 
4069 	/* Let secondaries go back to the offline loop */
4070 	for (i = 0; i < controlled_threads; ++i) {
4071 		kvmppc_release_hwthread(pcpu + i);
4072 		if (sip && sip->napped[i])
4073 			kvmppc_ipi_thread(pcpu + i);
4074 	}
4075 
4076 	spin_unlock(&vc->lock);
4077 
4078 	/* make sure updates to secondary vcpu structs are visible now */
4079 	smp_mb();
4080 
4081 	preempt_enable();
4082 
4083 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
4084 		pvc = core_info.vc[sub];
4085 		post_guest_process(pvc, pvc == vc);
4086 	}
4087 
4088 	spin_lock(&vc->lock);
4089 
4090  out:
4091 	vc->vcore_state = VCORE_INACTIVE;
4092 	trace_kvmppc_run_core(vc, 1);
4093 }
4094 
4095 static inline bool hcall_is_xics(unsigned long req)
4096 {
4097 	return req == H_EOI || req == H_CPPR || req == H_IPI ||
4098 		req == H_IPOLL || req == H_XIRR || req == H_XIRR_X;
4099 }
4100 
4101 static void vcpu_vpa_increment_dispatch(struct kvm_vcpu *vcpu)
4102 {
4103 	struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
4104 	if (lp) {
4105 		u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
4106 		lp->yield_count = cpu_to_be32(yield_count);
4107 		vcpu->arch.vpa.dirty = 1;
4108 	}
4109 }
4110 
4111 static int kvmhv_vcpu_entry_nestedv2(struct kvm_vcpu *vcpu, u64 time_limit,
4112 				     unsigned long lpcr, u64 *tb)
4113 {
4114 	struct kvmhv_nestedv2_io *io;
4115 	unsigned long msr, i;
4116 	int trap;
4117 	long rc;
4118 
4119 	io = &vcpu->arch.nestedv2_io;
4120 
4121 	msr = mfmsr();
4122 	kvmppc_msr_hard_disable_set_facilities(vcpu, msr);
4123 	if (lazy_irq_pending())
4124 		return 0;
4125 
4126 	rc = kvmhv_nestedv2_flush_vcpu(vcpu, time_limit);
4127 	if (rc < 0)
4128 		return -EINVAL;
4129 
4130 	kvmppc_gse_put_u64(io->vcpu_run_input, KVMPPC_GSID_LPCR, lpcr);
4131 
4132 	accumulate_time(vcpu, &vcpu->arch.in_guest);
4133 	rc = plpar_guest_run_vcpu(0, vcpu->kvm->arch.lpid, vcpu->vcpu_id,
4134 				  &trap, &i);
4135 
4136 	if (rc != H_SUCCESS) {
4137 		pr_err("KVM Guest Run VCPU hcall failed\n");
4138 		if (rc == H_INVALID_ELEMENT_ID)
4139 			pr_err("KVM: Guest Run VCPU invalid element id at %ld\n", i);
4140 		else if (rc == H_INVALID_ELEMENT_SIZE)
4141 			pr_err("KVM: Guest Run VCPU invalid element size at %ld\n", i);
4142 		else if (rc == H_INVALID_ELEMENT_VALUE)
4143 			pr_err("KVM: Guest Run VCPU invalid element value at %ld\n", i);
4144 		return -EINVAL;
4145 	}
4146 	accumulate_time(vcpu, &vcpu->arch.guest_exit);
4147 
4148 	*tb = mftb();
4149 	kvmppc_gsm_reset(io->vcpu_message);
4150 	kvmppc_gsm_reset(io->vcore_message);
4151 	kvmppc_gsbm_zero(&io->valids);
4152 
4153 	rc = kvmhv_nestedv2_parse_output(vcpu);
4154 	if (rc < 0)
4155 		return -EINVAL;
4156 
4157 	timer_rearm_host_dec(*tb);
4158 
4159 	return trap;
4160 }
4161 
4162 /* call our hypervisor to load up HV regs and go */
4163 static int kvmhv_vcpu_entry_p9_nested(struct kvm_vcpu *vcpu, u64 time_limit, unsigned long lpcr, u64 *tb)
4164 {
4165 	unsigned long host_psscr;
4166 	unsigned long msr;
4167 	struct hv_guest_state hvregs;
4168 	struct p9_host_os_sprs host_os_sprs;
4169 	s64 dec;
4170 	int trap;
4171 
4172 	msr = mfmsr();
4173 
4174 	save_p9_host_os_sprs(&host_os_sprs);
4175 
4176 	/*
4177 	 * We need to save and restore the guest visible part of the
4178 	 * psscr (i.e. using SPRN_PSSCR_PR) since the hypervisor
4179 	 * doesn't do this for us. Note only required if pseries since
4180 	 * this is done in kvmhv_vcpu_entry_p9() below otherwise.
4181 	 */
4182 	host_psscr = mfspr(SPRN_PSSCR_PR);
4183 
4184 	kvmppc_msr_hard_disable_set_facilities(vcpu, msr);
4185 	if (lazy_irq_pending())
4186 		return 0;
4187 
4188 	if (unlikely(load_vcpu_state(vcpu, &host_os_sprs)))
4189 		msr = mfmsr(); /* TM restore can update msr */
4190 
4191 	if (vcpu->arch.psscr != host_psscr)
4192 		mtspr(SPRN_PSSCR_PR, vcpu->arch.psscr);
4193 
4194 	kvmhv_save_hv_regs(vcpu, &hvregs);
4195 	hvregs.lpcr = lpcr;
4196 	hvregs.amor = ~0;
4197 	vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
4198 	hvregs.version = HV_GUEST_STATE_VERSION;
4199 	if (vcpu->arch.nested) {
4200 		hvregs.lpid = vcpu->arch.nested->shadow_lpid;
4201 		hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
4202 	} else {
4203 		hvregs.lpid = vcpu->kvm->arch.lpid;
4204 		hvregs.vcpu_token = vcpu->vcpu_id;
4205 	}
4206 	hvregs.hdec_expiry = time_limit;
4207 
4208 	/*
4209 	 * When setting DEC, we must always deal with irq_work_raise
4210 	 * via NMI vs setting DEC. The problem occurs right as we
4211 	 * switch into guest mode if a NMI hits and sets pending work
4212 	 * and sets DEC, then that will apply to the guest and not
4213 	 * bring us back to the host.
4214 	 *
4215 	 * irq_work_raise could check a flag (or possibly LPCR[HDICE]
4216 	 * for example) and set HDEC to 1? That wouldn't solve the
4217 	 * nested hv case which needs to abort the hcall or zero the
4218 	 * time limit.
4219 	 *
4220 	 * XXX: Another day's problem.
4221 	 */
4222 	mtspr(SPRN_DEC, kvmppc_dec_expires_host_tb(vcpu) - *tb);
4223 
4224 	mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
4225 	mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
4226 	switch_pmu_to_guest(vcpu, &host_os_sprs);
4227 	accumulate_time(vcpu, &vcpu->arch.in_guest);
4228 	trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
4229 				  __pa(&vcpu->arch.regs));
4230 	accumulate_time(vcpu, &vcpu->arch.guest_exit);
4231 	kvmhv_restore_hv_return_state(vcpu, &hvregs);
4232 	switch_pmu_to_host(vcpu, &host_os_sprs);
4233 	vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
4234 	vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
4235 	vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
4236 	vcpu->arch.psscr = mfspr(SPRN_PSSCR_PR);
4237 
4238 	store_vcpu_state(vcpu);
4239 
4240 	dec = mfspr(SPRN_DEC);
4241 	if (!(lpcr & LPCR_LD)) /* Sign extend if not using large decrementer */
4242 		dec = (s32) dec;
4243 	*tb = mftb();
4244 	vcpu->arch.dec_expires = dec + (*tb + kvmppc_get_tb_offset(vcpu));
4245 
4246 	timer_rearm_host_dec(*tb);
4247 
4248 	restore_p9_host_os_sprs(vcpu, &host_os_sprs);
4249 	if (vcpu->arch.psscr != host_psscr)
4250 		mtspr(SPRN_PSSCR_PR, host_psscr);
4251 
4252 	return trap;
4253 }
4254 
4255 /*
4256  * Guest entry for POWER9 and later CPUs.
4257  */
4258 static int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
4259 			 unsigned long lpcr, u64 *tb)
4260 {
4261 	struct kvm *kvm = vcpu->kvm;
4262 	struct kvm_nested_guest *nested = vcpu->arch.nested;
4263 	u64 next_timer;
4264 	int trap;
4265 
4266 	next_timer = timer_get_next_tb();
4267 	if (*tb >= next_timer)
4268 		return BOOK3S_INTERRUPT_HV_DECREMENTER;
4269 	if (next_timer < time_limit)
4270 		time_limit = next_timer;
4271 	else if (*tb >= time_limit) /* nested time limit */
4272 		return BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER;
4273 
4274 	vcpu->arch.ceded = 0;
4275 
4276 	vcpu_vpa_increment_dispatch(vcpu);
4277 
4278 	if (kvmhv_on_pseries()) {
4279 		if (kvmhv_is_nestedv1())
4280 			trap = kvmhv_vcpu_entry_p9_nested(vcpu, time_limit, lpcr, tb);
4281 		else
4282 			trap = kvmhv_vcpu_entry_nestedv2(vcpu, time_limit, lpcr, tb);
4283 
4284 		/* H_CEDE has to be handled now, not later */
4285 		if (trap == BOOK3S_INTERRUPT_SYSCALL && !nested &&
4286 		    kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
4287 			kvmppc_cede(vcpu);
4288 			kvmppc_set_gpr(vcpu, 3, 0);
4289 			trap = 0;
4290 		}
4291 
4292 	} else if (nested) {
4293 		__this_cpu_write(cpu_in_guest, kvm);
4294 		trap = kvmhv_vcpu_entry_p9(vcpu, time_limit, lpcr, tb);
4295 		__this_cpu_write(cpu_in_guest, NULL);
4296 
4297 	} else {
4298 		kvmppc_xive_push_vcpu(vcpu);
4299 
4300 		__this_cpu_write(cpu_in_guest, kvm);
4301 		trap = kvmhv_vcpu_entry_p9(vcpu, time_limit, lpcr, tb);
4302 		__this_cpu_write(cpu_in_guest, NULL);
4303 
4304 		if (trap == BOOK3S_INTERRUPT_SYSCALL &&
4305 		    !(__kvmppc_get_msr_hv(vcpu) & MSR_PR)) {
4306 			unsigned long req = kvmppc_get_gpr(vcpu, 3);
4307 
4308 			/*
4309 			 * XIVE rearm and XICS hcalls must be handled
4310 			 * before xive context is pulled (is this
4311 			 * true?)
4312 			 */
4313 			if (req == H_CEDE) {
4314 				/* H_CEDE has to be handled now */
4315 				kvmppc_cede(vcpu);
4316 				if (!kvmppc_xive_rearm_escalation(vcpu)) {
4317 					/*
4318 					 * Pending escalation so abort
4319 					 * the cede.
4320 					 */
4321 					vcpu->arch.ceded = 0;
4322 				}
4323 				kvmppc_set_gpr(vcpu, 3, 0);
4324 				trap = 0;
4325 
4326 			} else if (req == H_ENTER_NESTED) {
4327 				/*
4328 				 * L2 should not run with the L1
4329 				 * context so rearm and pull it.
4330 				 */
4331 				if (!kvmppc_xive_rearm_escalation(vcpu)) {
4332 					/*
4333 					 * Pending escalation so abort
4334 					 * H_ENTER_NESTED.
4335 					 */
4336 					kvmppc_set_gpr(vcpu, 3, 0);
4337 					trap = 0;
4338 				}
4339 
4340 			} else if (hcall_is_xics(req)) {
4341 				int ret;
4342 
4343 				ret = kvmppc_xive_xics_hcall(vcpu, req);
4344 				if (ret != H_TOO_HARD) {
4345 					kvmppc_set_gpr(vcpu, 3, ret);
4346 					trap = 0;
4347 				}
4348 			}
4349 		}
4350 		kvmppc_xive_pull_vcpu(vcpu);
4351 
4352 		if (kvm_is_radix(kvm))
4353 			vcpu->arch.slb_max = 0;
4354 	}
4355 
4356 	vcpu_vpa_increment_dispatch(vcpu);
4357 
4358 	return trap;
4359 }
4360 
4361 /*
4362  * Wait for some other vcpu thread to execute us, and
4363  * wake us up when we need to handle something in the host.
4364  */
4365 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
4366 				 struct kvm_vcpu *vcpu, int wait_state)
4367 {
4368 	DEFINE_WAIT(wait);
4369 
4370 	prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
4371 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4372 		spin_unlock(&vc->lock);
4373 		schedule();
4374 		spin_lock(&vc->lock);
4375 	}
4376 	finish_wait(&vcpu->arch.cpu_run, &wait);
4377 }
4378 
4379 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
4380 {
4381 	if (!halt_poll_ns_grow)
4382 		return;
4383 
4384 	vc->halt_poll_ns *= halt_poll_ns_grow;
4385 	if (vc->halt_poll_ns < halt_poll_ns_grow_start)
4386 		vc->halt_poll_ns = halt_poll_ns_grow_start;
4387 }
4388 
4389 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
4390 {
4391 	if (halt_poll_ns_shrink == 0)
4392 		vc->halt_poll_ns = 0;
4393 	else
4394 		vc->halt_poll_ns /= halt_poll_ns_shrink;
4395 }
4396 
4397 #ifdef CONFIG_KVM_XICS
4398 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
4399 {
4400 	if (!xics_on_xive())
4401 		return false;
4402 	return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
4403 		vcpu->arch.xive_saved_state.cppr;
4404 }
4405 #else
4406 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
4407 {
4408 	return false;
4409 }
4410 #endif /* CONFIG_KVM_XICS */
4411 
4412 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
4413 {
4414 	if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
4415 	    kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
4416 		return true;
4417 
4418 	return false;
4419 }
4420 
4421 static bool kvmppc_vcpu_check_block(struct kvm_vcpu *vcpu)
4422 {
4423 	if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
4424 		return true;
4425 	return false;
4426 }
4427 
4428 /*
4429  * Check to see if any of the runnable vcpus on the vcore have pending
4430  * exceptions or are no longer ceded
4431  */
4432 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
4433 {
4434 	struct kvm_vcpu *vcpu;
4435 	int i;
4436 
4437 	for_each_runnable_thread(i, vcpu, vc) {
4438 		if (kvmppc_vcpu_check_block(vcpu))
4439 			return 1;
4440 	}
4441 
4442 	return 0;
4443 }
4444 
4445 /*
4446  * All the vcpus in this vcore are idle, so wait for a decrementer
4447  * or external interrupt to one of the vcpus.  vc->lock is held.
4448  */
4449 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
4450 {
4451 	ktime_t cur, start_poll, start_wait;
4452 	int do_sleep = 1;
4453 	u64 block_ns;
4454 
4455 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
4456 
4457 	/* Poll for pending exceptions and ceded state */
4458 	cur = start_poll = ktime_get();
4459 	if (vc->halt_poll_ns) {
4460 		ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
4461 		++vc->runner->stat.generic.halt_attempted_poll;
4462 
4463 		vc->vcore_state = VCORE_POLLING;
4464 		spin_unlock(&vc->lock);
4465 
4466 		do {
4467 			if (kvmppc_vcore_check_block(vc)) {
4468 				do_sleep = 0;
4469 				break;
4470 			}
4471 			cur = ktime_get();
4472 		} while (kvm_vcpu_can_poll(cur, stop));
4473 
4474 		spin_lock(&vc->lock);
4475 		vc->vcore_state = VCORE_INACTIVE;
4476 
4477 		if (!do_sleep) {
4478 			++vc->runner->stat.generic.halt_successful_poll;
4479 			goto out;
4480 		}
4481 	}
4482 
4483 	prepare_to_rcuwait(&vc->wait);
4484 	set_current_state(TASK_INTERRUPTIBLE);
4485 	if (kvmppc_vcore_check_block(vc)) {
4486 		finish_rcuwait(&vc->wait);
4487 		do_sleep = 0;
4488 		/* If we polled, count this as a successful poll */
4489 		if (vc->halt_poll_ns)
4490 			++vc->runner->stat.generic.halt_successful_poll;
4491 		goto out;
4492 	}
4493 
4494 	start_wait = ktime_get();
4495 
4496 	vc->vcore_state = VCORE_SLEEPING;
4497 	trace_kvmppc_vcore_blocked(vc->runner, 0);
4498 	spin_unlock(&vc->lock);
4499 	schedule();
4500 	finish_rcuwait(&vc->wait);
4501 	spin_lock(&vc->lock);
4502 	vc->vcore_state = VCORE_INACTIVE;
4503 	trace_kvmppc_vcore_blocked(vc->runner, 1);
4504 	++vc->runner->stat.halt_successful_wait;
4505 
4506 	cur = ktime_get();
4507 
4508 out:
4509 	block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
4510 
4511 	/* Attribute wait time */
4512 	if (do_sleep) {
4513 		vc->runner->stat.generic.halt_wait_ns +=
4514 			ktime_to_ns(cur) - ktime_to_ns(start_wait);
4515 		KVM_STATS_LOG_HIST_UPDATE(
4516 				vc->runner->stat.generic.halt_wait_hist,
4517 				ktime_to_ns(cur) - ktime_to_ns(start_wait));
4518 		/* Attribute failed poll time */
4519 		if (vc->halt_poll_ns) {
4520 			vc->runner->stat.generic.halt_poll_fail_ns +=
4521 				ktime_to_ns(start_wait) -
4522 				ktime_to_ns(start_poll);
4523 			KVM_STATS_LOG_HIST_UPDATE(
4524 				vc->runner->stat.generic.halt_poll_fail_hist,
4525 				ktime_to_ns(start_wait) -
4526 				ktime_to_ns(start_poll));
4527 		}
4528 	} else {
4529 		/* Attribute successful poll time */
4530 		if (vc->halt_poll_ns) {
4531 			vc->runner->stat.generic.halt_poll_success_ns +=
4532 				ktime_to_ns(cur) -
4533 				ktime_to_ns(start_poll);
4534 			KVM_STATS_LOG_HIST_UPDATE(
4535 				vc->runner->stat.generic.halt_poll_success_hist,
4536 				ktime_to_ns(cur) - ktime_to_ns(start_poll));
4537 		}
4538 	}
4539 
4540 	/* Adjust poll time */
4541 	if (halt_poll_ns) {
4542 		if (block_ns <= vc->halt_poll_ns)
4543 			;
4544 		/* We slept and blocked for longer than the max halt time */
4545 		else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
4546 			shrink_halt_poll_ns(vc);
4547 		/* We slept and our poll time is too small */
4548 		else if (vc->halt_poll_ns < halt_poll_ns &&
4549 				block_ns < halt_poll_ns)
4550 			grow_halt_poll_ns(vc);
4551 		if (vc->halt_poll_ns > halt_poll_ns)
4552 			vc->halt_poll_ns = halt_poll_ns;
4553 	} else
4554 		vc->halt_poll_ns = 0;
4555 
4556 	trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
4557 }
4558 
4559 /*
4560  * This never fails for a radix guest, as none of the operations it does
4561  * for a radix guest can fail or have a way to report failure.
4562  */
4563 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
4564 {
4565 	int r = 0;
4566 	struct kvm *kvm = vcpu->kvm;
4567 
4568 	mutex_lock(&kvm->arch.mmu_setup_lock);
4569 	if (!kvm->arch.mmu_ready) {
4570 		if (!kvm_is_radix(kvm))
4571 			r = kvmppc_hv_setup_htab_rma(vcpu);
4572 		if (!r) {
4573 			if (cpu_has_feature(CPU_FTR_ARCH_300))
4574 				kvmppc_setup_partition_table(kvm);
4575 			kvm->arch.mmu_ready = 1;
4576 		}
4577 	}
4578 	mutex_unlock(&kvm->arch.mmu_setup_lock);
4579 	return r;
4580 }
4581 
4582 static int kvmppc_run_vcpu(struct kvm_vcpu *vcpu)
4583 {
4584 	struct kvm_run *run = vcpu->run;
4585 	int n_ceded, i, r;
4586 	struct kvmppc_vcore *vc;
4587 	struct kvm_vcpu *v;
4588 
4589 	trace_kvmppc_run_vcpu_enter(vcpu);
4590 
4591 	run->exit_reason = 0;
4592 	vcpu->arch.ret = RESUME_GUEST;
4593 	vcpu->arch.trap = 0;
4594 	kvmppc_update_vpas(vcpu);
4595 
4596 	/*
4597 	 * Synchronize with other threads in this virtual core
4598 	 */
4599 	vc = vcpu->arch.vcore;
4600 	spin_lock(&vc->lock);
4601 	vcpu->arch.ceded = 0;
4602 	vcpu->arch.run_task = current;
4603 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4604 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4605 	vcpu->arch.busy_preempt = TB_NIL;
4606 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
4607 	++vc->n_runnable;
4608 
4609 	/*
4610 	 * This happens the first time this is called for a vcpu.
4611 	 * If the vcore is already running, we may be able to start
4612 	 * this thread straight away and have it join in.
4613 	 */
4614 	if (!signal_pending(current)) {
4615 		if ((vc->vcore_state == VCORE_PIGGYBACK ||
4616 		     vc->vcore_state == VCORE_RUNNING) &&
4617 			   !VCORE_IS_EXITING(vc)) {
4618 			kvmppc_update_vpa_dispatch(vcpu, vc);
4619 			kvmppc_start_thread(vcpu, vc);
4620 			trace_kvm_guest_enter(vcpu);
4621 		} else if (vc->vcore_state == VCORE_SLEEPING) {
4622 		        rcuwait_wake_up(&vc->wait);
4623 		}
4624 
4625 	}
4626 
4627 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4628 	       !signal_pending(current)) {
4629 		/* See if the MMU is ready to go */
4630 		if (!vcpu->kvm->arch.mmu_ready) {
4631 			spin_unlock(&vc->lock);
4632 			r = kvmhv_setup_mmu(vcpu);
4633 			spin_lock(&vc->lock);
4634 			if (r) {
4635 				run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4636 				run->fail_entry.
4637 					hardware_entry_failure_reason = 0;
4638 				vcpu->arch.ret = r;
4639 				break;
4640 			}
4641 		}
4642 
4643 		if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4644 			kvmppc_vcore_end_preempt(vc);
4645 
4646 		if (vc->vcore_state != VCORE_INACTIVE) {
4647 			kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
4648 			continue;
4649 		}
4650 		for_each_runnable_thread(i, v, vc) {
4651 			kvmppc_core_prepare_to_enter(v);
4652 			if (signal_pending(v->arch.run_task)) {
4653 				kvmppc_remove_runnable(vc, v, mftb());
4654 				v->stat.signal_exits++;
4655 				v->run->exit_reason = KVM_EXIT_INTR;
4656 				v->arch.ret = -EINTR;
4657 				wake_up(&v->arch.cpu_run);
4658 			}
4659 		}
4660 		if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
4661 			break;
4662 		n_ceded = 0;
4663 		for_each_runnable_thread(i, v, vc) {
4664 			if (!kvmppc_vcpu_woken(v))
4665 				n_ceded += v->arch.ceded;
4666 			else
4667 				v->arch.ceded = 0;
4668 		}
4669 		vc->runner = vcpu;
4670 		if (n_ceded == vc->n_runnable) {
4671 			kvmppc_vcore_blocked(vc);
4672 		} else if (need_resched()) {
4673 			kvmppc_vcore_preempt(vc);
4674 			/* Let something else run */
4675 			cond_resched_lock(&vc->lock);
4676 			if (vc->vcore_state == VCORE_PREEMPT)
4677 				kvmppc_vcore_end_preempt(vc);
4678 		} else {
4679 			kvmppc_run_core(vc);
4680 		}
4681 		vc->runner = NULL;
4682 	}
4683 
4684 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4685 	       (vc->vcore_state == VCORE_RUNNING ||
4686 		vc->vcore_state == VCORE_EXITING ||
4687 		vc->vcore_state == VCORE_PIGGYBACK))
4688 		kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
4689 
4690 	if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4691 		kvmppc_vcore_end_preempt(vc);
4692 
4693 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4694 		kvmppc_remove_runnable(vc, vcpu, mftb());
4695 		vcpu->stat.signal_exits++;
4696 		run->exit_reason = KVM_EXIT_INTR;
4697 		vcpu->arch.ret = -EINTR;
4698 	}
4699 
4700 	if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
4701 		/* Wake up some vcpu to run the core */
4702 		i = -1;
4703 		v = next_runnable_thread(vc, &i);
4704 		wake_up(&v->arch.cpu_run);
4705 	}
4706 
4707 	trace_kvmppc_run_vcpu_exit(vcpu);
4708 	spin_unlock(&vc->lock);
4709 	return vcpu->arch.ret;
4710 }
4711 
4712 int kvmhv_run_single_vcpu(struct kvm_vcpu *vcpu, u64 time_limit,
4713 			  unsigned long lpcr)
4714 {
4715 	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
4716 	struct kvm_run *run = vcpu->run;
4717 	int trap, r, pcpu;
4718 	int srcu_idx;
4719 	struct kvmppc_vcore *vc;
4720 	struct kvm *kvm = vcpu->kvm;
4721 	struct kvm_nested_guest *nested = vcpu->arch.nested;
4722 	unsigned long flags;
4723 	u64 tb;
4724 
4725 	trace_kvmppc_run_vcpu_enter(vcpu);
4726 
4727 	run->exit_reason = 0;
4728 	vcpu->arch.ret = RESUME_GUEST;
4729 	vcpu->arch.trap = 0;
4730 
4731 	vc = vcpu->arch.vcore;
4732 	vcpu->arch.ceded = 0;
4733 	vcpu->arch.run_task = current;
4734 	vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
4735 
4736 	/* See if the MMU is ready to go */
4737 	if (unlikely(!kvm->arch.mmu_ready)) {
4738 		r = kvmhv_setup_mmu(vcpu);
4739 		if (r) {
4740 			run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4741 			run->fail_entry.hardware_entry_failure_reason = 0;
4742 			vcpu->arch.ret = r;
4743 			return r;
4744 		}
4745 	}
4746 
4747 	if (need_resched())
4748 		cond_resched();
4749 
4750 	kvmppc_update_vpas(vcpu);
4751 
4752 	preempt_disable();
4753 	pcpu = smp_processor_id();
4754 	if (kvm_is_radix(kvm))
4755 		kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4756 
4757 	/* flags save not required, but irq_pmu has no disable/enable API */
4758 	powerpc_local_irq_pmu_save(flags);
4759 
4760 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4761 
4762 	if (signal_pending(current))
4763 		goto sigpend;
4764 	if (need_resched() || !kvm->arch.mmu_ready)
4765 		goto out;
4766 
4767 	vcpu->cpu = pcpu;
4768 	vcpu->arch.thread_cpu = pcpu;
4769 	vc->pcpu = pcpu;
4770 	local_paca->kvm_hstate.kvm_vcpu = vcpu;
4771 	local_paca->kvm_hstate.ptid = 0;
4772 	local_paca->kvm_hstate.fake_suspend = 0;
4773 
4774 	/*
4775 	 * Orders set cpu/thread_cpu vs testing for pending interrupts and
4776 	 * doorbells below. The other side is when these fields are set vs
4777 	 * kvmppc_fast_vcpu_kick_hv reading the cpu/thread_cpu fields to
4778 	 * kick a vCPU to notice the pending interrupt.
4779 	 */
4780 	smp_mb();
4781 
4782 	if (!nested) {
4783 		kvmppc_core_prepare_to_enter(vcpu);
4784 		if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4785 			     &vcpu->arch.pending_exceptions) ||
4786 		    xive_interrupt_pending(vcpu)) {
4787 			/*
4788 			 * For nested HV, don't synthesize but always pass MER,
4789 			 * the L0 will be able to optimise that more
4790 			 * effectively than manipulating registers directly.
4791 			 */
4792 			if (!kvmhv_on_pseries() && (__kvmppc_get_msr_hv(vcpu) & MSR_EE))
4793 				kvmppc_inject_interrupt_hv(vcpu,
4794 							   BOOK3S_INTERRUPT_EXTERNAL, 0);
4795 			else
4796 				lpcr |= LPCR_MER;
4797 		}
4798 	} else if (vcpu->arch.pending_exceptions ||
4799 		   vcpu->arch.doorbell_request ||
4800 		   xive_interrupt_pending(vcpu)) {
4801 		vcpu->arch.ret = RESUME_HOST;
4802 		goto out;
4803 	}
4804 
4805 	if (vcpu->arch.timer_running) {
4806 		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
4807 		vcpu->arch.timer_running = 0;
4808 	}
4809 
4810 	tb = mftb();
4811 
4812 	kvmppc_update_vpa_dispatch_p9(vcpu, vc, tb + kvmppc_get_tb_offset(vcpu));
4813 
4814 	trace_kvm_guest_enter(vcpu);
4815 
4816 	guest_timing_enter_irqoff();
4817 
4818 	srcu_idx = srcu_read_lock(&kvm->srcu);
4819 
4820 	guest_state_enter_irqoff();
4821 	this_cpu_disable_ftrace();
4822 
4823 	trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr, &tb);
4824 	vcpu->arch.trap = trap;
4825 
4826 	this_cpu_enable_ftrace();
4827 	guest_state_exit_irqoff();
4828 
4829 	srcu_read_unlock(&kvm->srcu, srcu_idx);
4830 
4831 	set_irq_happened(trap);
4832 
4833 	vcpu->cpu = -1;
4834 	vcpu->arch.thread_cpu = -1;
4835 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4836 
4837 	if (!vtime_accounting_enabled_this_cpu()) {
4838 		powerpc_local_irq_pmu_restore(flags);
4839 		/*
4840 		 * Service IRQs here before guest_timing_exit_irqoff() so any
4841 		 * ticks that occurred while running the guest are accounted to
4842 		 * the guest. If vtime accounting is enabled, accounting uses
4843 		 * TB rather than ticks, so it can be done without enabling
4844 		 * interrupts here, which has the problem that it accounts
4845 		 * interrupt processing overhead to the host.
4846 		 */
4847 		powerpc_local_irq_pmu_save(flags);
4848 	}
4849 	guest_timing_exit_irqoff();
4850 
4851 	powerpc_local_irq_pmu_restore(flags);
4852 
4853 	preempt_enable();
4854 
4855 	/*
4856 	 * cancel pending decrementer exception if DEC is now positive, or if
4857 	 * entering a nested guest in which case the decrementer is now owned
4858 	 * by L2 and the L1 decrementer is provided in hdec_expires
4859 	 */
4860 	if (!kvmhv_is_nestedv2() && kvmppc_core_pending_dec(vcpu) &&
4861 			((tb < kvmppc_dec_expires_host_tb(vcpu)) ||
4862 			 (trap == BOOK3S_INTERRUPT_SYSCALL &&
4863 			  kvmppc_get_gpr(vcpu, 3) == H_ENTER_NESTED)))
4864 		kvmppc_core_dequeue_dec(vcpu);
4865 
4866 	trace_kvm_guest_exit(vcpu);
4867 	r = RESUME_GUEST;
4868 	if (trap) {
4869 		if (!nested)
4870 			r = kvmppc_handle_exit_hv(vcpu, current);
4871 		else
4872 			r = kvmppc_handle_nested_exit(vcpu);
4873 	}
4874 	vcpu->arch.ret = r;
4875 
4876 	if (is_kvmppc_resume_guest(r) && !kvmppc_vcpu_check_block(vcpu)) {
4877 		kvmppc_set_timer(vcpu);
4878 
4879 		prepare_to_rcuwait(wait);
4880 		for (;;) {
4881 			set_current_state(TASK_INTERRUPTIBLE);
4882 			if (signal_pending(current)) {
4883 				vcpu->stat.signal_exits++;
4884 				run->exit_reason = KVM_EXIT_INTR;
4885 				vcpu->arch.ret = -EINTR;
4886 				break;
4887 			}
4888 
4889 			if (kvmppc_vcpu_check_block(vcpu))
4890 				break;
4891 
4892 			trace_kvmppc_vcore_blocked(vcpu, 0);
4893 			schedule();
4894 			trace_kvmppc_vcore_blocked(vcpu, 1);
4895 		}
4896 		finish_rcuwait(wait);
4897 	}
4898 	vcpu->arch.ceded = 0;
4899 
4900  done:
4901 	trace_kvmppc_run_vcpu_exit(vcpu);
4902 
4903 	return vcpu->arch.ret;
4904 
4905  sigpend:
4906 	vcpu->stat.signal_exits++;
4907 	run->exit_reason = KVM_EXIT_INTR;
4908 	vcpu->arch.ret = -EINTR;
4909  out:
4910 	vcpu->cpu = -1;
4911 	vcpu->arch.thread_cpu = -1;
4912 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4913 	powerpc_local_irq_pmu_restore(flags);
4914 	preempt_enable();
4915 	goto done;
4916 }
4917 
4918 static int kvmppc_vcpu_run_hv(struct kvm_vcpu *vcpu)
4919 {
4920 	struct kvm_run *run = vcpu->run;
4921 	int r;
4922 	int srcu_idx;
4923 	struct kvm *kvm;
4924 	unsigned long msr;
4925 
4926 	start_timing(vcpu, &vcpu->arch.vcpu_entry);
4927 
4928 	if (!vcpu->arch.sane) {
4929 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4930 		return -EINVAL;
4931 	}
4932 
4933 	/* No need to go into the guest when all we'll do is come back out */
4934 	if (signal_pending(current)) {
4935 		run->exit_reason = KVM_EXIT_INTR;
4936 		return -EINTR;
4937 	}
4938 
4939 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4940 	/*
4941 	 * Don't allow entry with a suspended transaction, because
4942 	 * the guest entry/exit code will lose it.
4943 	 */
4944 	if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4945 	    (current->thread.regs->msr & MSR_TM)) {
4946 		if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4947 			run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4948 			run->fail_entry.hardware_entry_failure_reason = 0;
4949 			return -EINVAL;
4950 		}
4951 	}
4952 #endif
4953 
4954 	/*
4955 	 * Force online to 1 for the sake of old userspace which doesn't
4956 	 * set it.
4957 	 */
4958 	if (!vcpu->arch.online) {
4959 		atomic_inc(&vcpu->arch.vcore->online_count);
4960 		vcpu->arch.online = 1;
4961 	}
4962 
4963 	kvmppc_core_prepare_to_enter(vcpu);
4964 
4965 	kvm = vcpu->kvm;
4966 	atomic_inc(&kvm->arch.vcpus_running);
4967 	/* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4968 	smp_mb();
4969 
4970 	msr = 0;
4971 	if (IS_ENABLED(CONFIG_PPC_FPU))
4972 		msr |= MSR_FP;
4973 	if (cpu_has_feature(CPU_FTR_ALTIVEC))
4974 		msr |= MSR_VEC;
4975 	if (cpu_has_feature(CPU_FTR_VSX))
4976 		msr |= MSR_VSX;
4977 	if ((cpu_has_feature(CPU_FTR_TM) ||
4978 	    cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST)) &&
4979 			(kvmppc_get_hfscr_hv(vcpu) & HFSCR_TM))
4980 		msr |= MSR_TM;
4981 	msr = msr_check_and_set(msr);
4982 
4983 	kvmppc_save_user_regs();
4984 
4985 	kvmppc_save_current_sprs();
4986 
4987 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
4988 		vcpu->arch.waitp = &vcpu->arch.vcore->wait;
4989 	vcpu->arch.pgdir = kvm->mm->pgd;
4990 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4991 
4992 	do {
4993 		accumulate_time(vcpu, &vcpu->arch.guest_entry);
4994 		if (cpu_has_feature(CPU_FTR_ARCH_300))
4995 			r = kvmhv_run_single_vcpu(vcpu, ~(u64)0,
4996 						  vcpu->arch.vcore->lpcr);
4997 		else
4998 			r = kvmppc_run_vcpu(vcpu);
4999 
5000 		if (run->exit_reason == KVM_EXIT_PAPR_HCALL) {
5001 			accumulate_time(vcpu, &vcpu->arch.hcall);
5002 
5003 			if (!kvmhv_is_nestedv2() && WARN_ON_ONCE(__kvmppc_get_msr_hv(vcpu) & MSR_PR)) {
5004 				/*
5005 				 * These should have been caught reflected
5006 				 * into the guest by now. Final sanity check:
5007 				 * don't allow userspace to execute hcalls in
5008 				 * the hypervisor.
5009 				 */
5010 				r = RESUME_GUEST;
5011 				continue;
5012 			}
5013 			trace_kvm_hcall_enter(vcpu);
5014 			r = kvmppc_pseries_do_hcall(vcpu);
5015 			trace_kvm_hcall_exit(vcpu, r);
5016 			kvmppc_core_prepare_to_enter(vcpu);
5017 		} else if (r == RESUME_PAGE_FAULT) {
5018 			accumulate_time(vcpu, &vcpu->arch.pg_fault);
5019 			srcu_idx = srcu_read_lock(&kvm->srcu);
5020 			r = kvmppc_book3s_hv_page_fault(vcpu,
5021 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
5022 			srcu_read_unlock(&kvm->srcu, srcu_idx);
5023 		} else if (r == RESUME_PASSTHROUGH) {
5024 			if (WARN_ON(xics_on_xive()))
5025 				r = H_SUCCESS;
5026 			else
5027 				r = kvmppc_xics_rm_complete(vcpu, 0);
5028 		}
5029 	} while (is_kvmppc_resume_guest(r));
5030 	accumulate_time(vcpu, &vcpu->arch.vcpu_exit);
5031 
5032 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
5033 	atomic_dec(&kvm->arch.vcpus_running);
5034 
5035 	srr_regs_clobbered();
5036 
5037 	end_timing(vcpu);
5038 
5039 	return r;
5040 }
5041 
5042 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
5043 				     int shift, int sllp)
5044 {
5045 	(*sps)->page_shift = shift;
5046 	(*sps)->slb_enc = sllp;
5047 	(*sps)->enc[0].page_shift = shift;
5048 	(*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
5049 	/*
5050 	 * Add 16MB MPSS support (may get filtered out by userspace)
5051 	 */
5052 	if (shift != 24) {
5053 		int penc = kvmppc_pgsize_lp_encoding(shift, 24);
5054 		if (penc != -1) {
5055 			(*sps)->enc[1].page_shift = 24;
5056 			(*sps)->enc[1].pte_enc = penc;
5057 		}
5058 	}
5059 	(*sps)++;
5060 }
5061 
5062 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
5063 					 struct kvm_ppc_smmu_info *info)
5064 {
5065 	struct kvm_ppc_one_seg_page_size *sps;
5066 
5067 	/*
5068 	 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
5069 	 * POWER7 doesn't support keys for instruction accesses,
5070 	 * POWER8 and POWER9 do.
5071 	 */
5072 	info->data_keys = 32;
5073 	info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
5074 
5075 	/* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
5076 	info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
5077 	info->slb_size = 32;
5078 
5079 	/* We only support these sizes for now, and no muti-size segments */
5080 	sps = &info->sps[0];
5081 	kvmppc_add_seg_page_size(&sps, 12, 0);
5082 	kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
5083 	kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
5084 
5085 	/* If running as a nested hypervisor, we don't support HPT guests */
5086 	if (kvmhv_on_pseries())
5087 		info->flags |= KVM_PPC_NO_HASH;
5088 
5089 	return 0;
5090 }
5091 
5092 /*
5093  * Get (and clear) the dirty memory log for a memory slot.
5094  */
5095 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
5096 					 struct kvm_dirty_log *log)
5097 {
5098 	struct kvm_memslots *slots;
5099 	struct kvm_memory_slot *memslot;
5100 	int r;
5101 	unsigned long n, i;
5102 	unsigned long *buf, *p;
5103 	struct kvm_vcpu *vcpu;
5104 
5105 	mutex_lock(&kvm->slots_lock);
5106 
5107 	r = -EINVAL;
5108 	if (log->slot >= KVM_USER_MEM_SLOTS)
5109 		goto out;
5110 
5111 	slots = kvm_memslots(kvm);
5112 	memslot = id_to_memslot(slots, log->slot);
5113 	r = -ENOENT;
5114 	if (!memslot || !memslot->dirty_bitmap)
5115 		goto out;
5116 
5117 	/*
5118 	 * Use second half of bitmap area because both HPT and radix
5119 	 * accumulate bits in the first half.
5120 	 */
5121 	n = kvm_dirty_bitmap_bytes(memslot);
5122 	buf = memslot->dirty_bitmap + n / sizeof(long);
5123 	memset(buf, 0, n);
5124 
5125 	if (kvm_is_radix(kvm))
5126 		r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
5127 	else
5128 		r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
5129 	if (r)
5130 		goto out;
5131 
5132 	/*
5133 	 * We accumulate dirty bits in the first half of the
5134 	 * memslot's dirty_bitmap area, for when pages are paged
5135 	 * out or modified by the host directly.  Pick up these
5136 	 * bits and add them to the map.
5137 	 */
5138 	p = memslot->dirty_bitmap;
5139 	for (i = 0; i < n / sizeof(long); ++i)
5140 		buf[i] |= xchg(&p[i], 0);
5141 
5142 	/* Harvest dirty bits from VPA and DTL updates */
5143 	/* Note: we never modify the SLB shadow buffer areas */
5144 	kvm_for_each_vcpu(i, vcpu, kvm) {
5145 		spin_lock(&vcpu->arch.vpa_update_lock);
5146 		kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
5147 		kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
5148 		spin_unlock(&vcpu->arch.vpa_update_lock);
5149 	}
5150 
5151 	r = -EFAULT;
5152 	if (copy_to_user(log->dirty_bitmap, buf, n))
5153 		goto out;
5154 
5155 	r = 0;
5156 out:
5157 	mutex_unlock(&kvm->slots_lock);
5158 	return r;
5159 }
5160 
5161 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *slot)
5162 {
5163 	vfree(slot->arch.rmap);
5164 	slot->arch.rmap = NULL;
5165 }
5166 
5167 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
5168 				const struct kvm_memory_slot *old,
5169 				struct kvm_memory_slot *new,
5170 				enum kvm_mr_change change)
5171 {
5172 	if (change == KVM_MR_CREATE) {
5173 		unsigned long size = array_size(new->npages, sizeof(*new->arch.rmap));
5174 
5175 		if ((size >> PAGE_SHIFT) > totalram_pages())
5176 			return -ENOMEM;
5177 
5178 		new->arch.rmap = vzalloc(size);
5179 		if (!new->arch.rmap)
5180 			return -ENOMEM;
5181 	} else if (change != KVM_MR_DELETE) {
5182 		new->arch.rmap = old->arch.rmap;
5183 	}
5184 
5185 	return 0;
5186 }
5187 
5188 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
5189 				struct kvm_memory_slot *old,
5190 				const struct kvm_memory_slot *new,
5191 				enum kvm_mr_change change)
5192 {
5193 	/*
5194 	 * If we are creating or modifying a memslot, it might make
5195 	 * some address that was previously cached as emulated
5196 	 * MMIO be no longer emulated MMIO, so invalidate
5197 	 * all the caches of emulated MMIO translations.
5198 	 */
5199 	if (change != KVM_MR_DELETE)
5200 		atomic64_inc(&kvm->arch.mmio_update);
5201 
5202 	/*
5203 	 * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
5204 	 * have already called kvm_arch_flush_shadow_memslot() to
5205 	 * flush shadow mappings.  For KVM_MR_CREATE we have no
5206 	 * previous mappings.  So the only case to handle is
5207 	 * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
5208 	 * has been changed.
5209 	 * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
5210 	 * to get rid of any THP PTEs in the partition-scoped page tables
5211 	 * so we can track dirtiness at the page level; we flush when
5212 	 * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
5213 	 * using THP PTEs.
5214 	 */
5215 	if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
5216 	    ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
5217 		kvmppc_radix_flush_memslot(kvm, old);
5218 	/*
5219 	 * If UV hasn't yet called H_SVM_INIT_START, don't register memslots.
5220 	 */
5221 	if (!kvm->arch.secure_guest)
5222 		return;
5223 
5224 	switch (change) {
5225 	case KVM_MR_CREATE:
5226 		/*
5227 		 * @TODO kvmppc_uvmem_memslot_create() can fail and
5228 		 * return error. Fix this.
5229 		 */
5230 		kvmppc_uvmem_memslot_create(kvm, new);
5231 		break;
5232 	case KVM_MR_DELETE:
5233 		kvmppc_uvmem_memslot_delete(kvm, old);
5234 		break;
5235 	default:
5236 		/* TODO: Handle KVM_MR_MOVE */
5237 		break;
5238 	}
5239 }
5240 
5241 /*
5242  * Update LPCR values in kvm->arch and in vcores.
5243  * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion
5244  * of kvm->arch.lpcr update).
5245  */
5246 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
5247 {
5248 	long int i;
5249 	u32 cores_done = 0;
5250 
5251 	if ((kvm->arch.lpcr & mask) == lpcr)
5252 		return;
5253 
5254 	kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
5255 
5256 	for (i = 0; i < KVM_MAX_VCORES; ++i) {
5257 		struct kvmppc_vcore *vc = kvm->arch.vcores[i];
5258 		if (!vc)
5259 			continue;
5260 
5261 		spin_lock(&vc->lock);
5262 		vc->lpcr = (vc->lpcr & ~mask) | lpcr;
5263 		verify_lpcr(kvm, vc->lpcr);
5264 		spin_unlock(&vc->lock);
5265 		if (++cores_done >= kvm->arch.online_vcores)
5266 			break;
5267 	}
5268 
5269 	if (kvmhv_is_nestedv2()) {
5270 		struct kvm_vcpu *vcpu;
5271 
5272 		kvm_for_each_vcpu(i, vcpu, kvm) {
5273 			kvmhv_nestedv2_mark_dirty(vcpu, KVMPPC_GSID_LPCR);
5274 		}
5275 	}
5276 }
5277 
5278 void kvmppc_setup_partition_table(struct kvm *kvm)
5279 {
5280 	unsigned long dw0, dw1;
5281 
5282 	if (!kvm_is_radix(kvm)) {
5283 		/* PS field - page size for VRMA */
5284 		dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
5285 			((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
5286 		/* HTABSIZE and HTABORG fields */
5287 		dw0 |= kvm->arch.sdr1;
5288 
5289 		/* Second dword as set by userspace */
5290 		dw1 = kvm->arch.process_table;
5291 	} else {
5292 		dw0 = PATB_HR | radix__get_tree_size() |
5293 			__pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
5294 		dw1 = PATB_GR | kvm->arch.process_table;
5295 	}
5296 	kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
5297 }
5298 
5299 /*
5300  * Set up HPT (hashed page table) and RMA (real-mode area).
5301  * Must be called with kvm->arch.mmu_setup_lock held.
5302  */
5303 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
5304 {
5305 	int err = 0;
5306 	struct kvm *kvm = vcpu->kvm;
5307 	unsigned long hva;
5308 	struct kvm_memory_slot *memslot;
5309 	struct vm_area_struct *vma;
5310 	unsigned long lpcr = 0, senc;
5311 	unsigned long psize, porder;
5312 	int srcu_idx;
5313 
5314 	/* Allocate hashed page table (if not done already) and reset it */
5315 	if (!kvm->arch.hpt.virt) {
5316 		int order = KVM_DEFAULT_HPT_ORDER;
5317 		struct kvm_hpt_info info;
5318 
5319 		err = kvmppc_allocate_hpt(&info, order);
5320 		/* If we get here, it means userspace didn't specify a
5321 		 * size explicitly.  So, try successively smaller
5322 		 * sizes if the default failed. */
5323 		while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
5324 			err  = kvmppc_allocate_hpt(&info, order);
5325 
5326 		if (err < 0) {
5327 			pr_err("KVM: Couldn't alloc HPT\n");
5328 			goto out;
5329 		}
5330 
5331 		kvmppc_set_hpt(kvm, &info);
5332 	}
5333 
5334 	/* Look up the memslot for guest physical address 0 */
5335 	srcu_idx = srcu_read_lock(&kvm->srcu);
5336 	memslot = gfn_to_memslot(kvm, 0);
5337 
5338 	/* We must have some memory at 0 by now */
5339 	err = -EINVAL;
5340 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
5341 		goto out_srcu;
5342 
5343 	/* Look up the VMA for the start of this memory slot */
5344 	hva = memslot->userspace_addr;
5345 	mmap_read_lock(kvm->mm);
5346 	vma = vma_lookup(kvm->mm, hva);
5347 	if (!vma || (vma->vm_flags & VM_IO))
5348 		goto up_out;
5349 
5350 	psize = vma_kernel_pagesize(vma);
5351 
5352 	mmap_read_unlock(kvm->mm);
5353 
5354 	/* We can handle 4k, 64k or 16M pages in the VRMA */
5355 	if (psize >= 0x1000000)
5356 		psize = 0x1000000;
5357 	else if (psize >= 0x10000)
5358 		psize = 0x10000;
5359 	else
5360 		psize = 0x1000;
5361 	porder = __ilog2(psize);
5362 
5363 	senc = slb_pgsize_encoding(psize);
5364 	kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
5365 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
5366 	/* Create HPTEs in the hash page table for the VRMA */
5367 	kvmppc_map_vrma(vcpu, memslot, porder);
5368 
5369 	/* Update VRMASD field in the LPCR */
5370 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
5371 		/* the -4 is to account for senc values starting at 0x10 */
5372 		lpcr = senc << (LPCR_VRMASD_SH - 4);
5373 		kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
5374 	}
5375 
5376 	/* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
5377 	smp_wmb();
5378 	err = 0;
5379  out_srcu:
5380 	srcu_read_unlock(&kvm->srcu, srcu_idx);
5381  out:
5382 	return err;
5383 
5384  up_out:
5385 	mmap_read_unlock(kvm->mm);
5386 	goto out_srcu;
5387 }
5388 
5389 /*
5390  * Must be called with kvm->arch.mmu_setup_lock held and
5391  * mmu_ready = 0 and no vcpus running.
5392  */
5393 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
5394 {
5395 	unsigned long lpcr, lpcr_mask;
5396 
5397 	if (nesting_enabled(kvm))
5398 		kvmhv_release_all_nested(kvm);
5399 	kvmppc_rmap_reset(kvm);
5400 	kvm->arch.process_table = 0;
5401 	/* Mutual exclusion with kvm_unmap_gfn_range etc. */
5402 	spin_lock(&kvm->mmu_lock);
5403 	kvm->arch.radix = 0;
5404 	spin_unlock(&kvm->mmu_lock);
5405 	kvmppc_free_radix(kvm);
5406 
5407 	lpcr = LPCR_VPM1;
5408 	lpcr_mask = LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5409 	if (cpu_has_feature(CPU_FTR_ARCH_31))
5410 		lpcr_mask |= LPCR_HAIL;
5411 	kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
5412 
5413 	return 0;
5414 }
5415 
5416 /*
5417  * Must be called with kvm->arch.mmu_setup_lock held and
5418  * mmu_ready = 0 and no vcpus running.
5419  */
5420 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
5421 {
5422 	unsigned long lpcr, lpcr_mask;
5423 	int err;
5424 
5425 	err = kvmppc_init_vm_radix(kvm);
5426 	if (err)
5427 		return err;
5428 	kvmppc_rmap_reset(kvm);
5429 	/* Mutual exclusion with kvm_unmap_gfn_range etc. */
5430 	spin_lock(&kvm->mmu_lock);
5431 	kvm->arch.radix = 1;
5432 	spin_unlock(&kvm->mmu_lock);
5433 	kvmppc_free_hpt(&kvm->arch.hpt);
5434 
5435 	lpcr = LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5436 	lpcr_mask = LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5437 	if (cpu_has_feature(CPU_FTR_ARCH_31)) {
5438 		lpcr_mask |= LPCR_HAIL;
5439 		if (cpu_has_feature(CPU_FTR_HVMODE) &&
5440 				(kvm->arch.host_lpcr & LPCR_HAIL))
5441 			lpcr |= LPCR_HAIL;
5442 	}
5443 	kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
5444 
5445 	return 0;
5446 }
5447 
5448 #ifdef CONFIG_KVM_XICS
5449 /*
5450  * Allocate a per-core structure for managing state about which cores are
5451  * running in the host versus the guest and for exchanging data between
5452  * real mode KVM and CPU running in the host.
5453  * This is only done for the first VM.
5454  * The allocated structure stays even if all VMs have stopped.
5455  * It is only freed when the kvm-hv module is unloaded.
5456  * It's OK for this routine to fail, we just don't support host
5457  * core operations like redirecting H_IPI wakeups.
5458  */
5459 void kvmppc_alloc_host_rm_ops(void)
5460 {
5461 	struct kvmppc_host_rm_ops *ops;
5462 	unsigned long l_ops;
5463 	int cpu, core;
5464 	int size;
5465 
5466 	if (cpu_has_feature(CPU_FTR_ARCH_300))
5467 		return;
5468 
5469 	/* Not the first time here ? */
5470 	if (kvmppc_host_rm_ops_hv != NULL)
5471 		return;
5472 
5473 	ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
5474 	if (!ops)
5475 		return;
5476 
5477 	size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
5478 	ops->rm_core = kzalloc(size, GFP_KERNEL);
5479 
5480 	if (!ops->rm_core) {
5481 		kfree(ops);
5482 		return;
5483 	}
5484 
5485 	cpus_read_lock();
5486 
5487 	for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
5488 		if (!cpu_online(cpu))
5489 			continue;
5490 
5491 		core = cpu >> threads_shift;
5492 		ops->rm_core[core].rm_state.in_host = 1;
5493 	}
5494 
5495 	ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
5496 
5497 	/*
5498 	 * Make the contents of the kvmppc_host_rm_ops structure visible
5499 	 * to other CPUs before we assign it to the global variable.
5500 	 * Do an atomic assignment (no locks used here), but if someone
5501 	 * beats us to it, just free our copy and return.
5502 	 */
5503 	smp_wmb();
5504 	l_ops = (unsigned long) ops;
5505 
5506 	if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
5507 		cpus_read_unlock();
5508 		kfree(ops->rm_core);
5509 		kfree(ops);
5510 		return;
5511 	}
5512 
5513 	cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
5514 					     "ppc/kvm_book3s:prepare",
5515 					     kvmppc_set_host_core,
5516 					     kvmppc_clear_host_core);
5517 	cpus_read_unlock();
5518 }
5519 
5520 void kvmppc_free_host_rm_ops(void)
5521 {
5522 	if (kvmppc_host_rm_ops_hv) {
5523 		cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
5524 		kfree(kvmppc_host_rm_ops_hv->rm_core);
5525 		kfree(kvmppc_host_rm_ops_hv);
5526 		kvmppc_host_rm_ops_hv = NULL;
5527 	}
5528 }
5529 #endif
5530 
5531 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
5532 {
5533 	unsigned long lpcr, lpid;
5534 	int ret;
5535 
5536 	mutex_init(&kvm->arch.uvmem_lock);
5537 	INIT_LIST_HEAD(&kvm->arch.uvmem_pfns);
5538 	mutex_init(&kvm->arch.mmu_setup_lock);
5539 
5540 	/* Allocate the guest's logical partition ID */
5541 
5542 	if (!kvmhv_is_nestedv2()) {
5543 		lpid = kvmppc_alloc_lpid();
5544 		if ((long)lpid < 0)
5545 			return -ENOMEM;
5546 		kvm->arch.lpid = lpid;
5547 	}
5548 
5549 	kvmppc_alloc_host_rm_ops();
5550 
5551 	kvmhv_vm_nested_init(kvm);
5552 
5553 	if (kvmhv_is_nestedv2()) {
5554 		long rc;
5555 		unsigned long guest_id;
5556 
5557 		rc = plpar_guest_create(0, &guest_id);
5558 
5559 		if (rc != H_SUCCESS)
5560 			pr_err("KVM: Create Guest hcall failed, rc=%ld\n", rc);
5561 
5562 		switch (rc) {
5563 		case H_PARAMETER:
5564 		case H_FUNCTION:
5565 		case H_STATE:
5566 			return -EINVAL;
5567 		case H_NOT_ENOUGH_RESOURCES:
5568 		case H_ABORTED:
5569 			return -ENOMEM;
5570 		case H_AUTHORITY:
5571 			return -EPERM;
5572 		case H_NOT_AVAILABLE:
5573 			return -EBUSY;
5574 		}
5575 		kvm->arch.lpid = guest_id;
5576 	}
5577 
5578 
5579 	/*
5580 	 * Since we don't flush the TLB when tearing down a VM,
5581 	 * and this lpid might have previously been used,
5582 	 * make sure we flush on each core before running the new VM.
5583 	 * On POWER9, the tlbie in mmu_partition_table_set_entry()
5584 	 * does this flush for us.
5585 	 */
5586 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5587 		cpumask_setall(&kvm->arch.need_tlb_flush);
5588 
5589 	/* Start out with the default set of hcalls enabled */
5590 	memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
5591 	       sizeof(kvm->arch.enabled_hcalls));
5592 
5593 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5594 		kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
5595 
5596 	/* Init LPCR for virtual RMA mode */
5597 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
5598 		kvm->arch.host_lpid = mfspr(SPRN_LPID);
5599 		kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
5600 		lpcr &= LPCR_PECE | LPCR_LPES;
5601 	} else {
5602 		/*
5603 		 * The L2 LPES mode will be set by the L0 according to whether
5604 		 * or not it needs to take external interrupts in HV mode.
5605 		 */
5606 		lpcr = 0;
5607 	}
5608 	lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
5609 		LPCR_VPM0 | LPCR_VPM1;
5610 	kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
5611 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
5612 	/* On POWER8 turn on online bit to enable PURR/SPURR */
5613 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
5614 		lpcr |= LPCR_ONL;
5615 	/*
5616 	 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
5617 	 * Set HVICE bit to enable hypervisor virtualization interrupts.
5618 	 * Set HEIC to prevent OS interrupts to go to hypervisor (should
5619 	 * be unnecessary but better safe than sorry in case we re-enable
5620 	 * EE in HV mode with this LPCR still set)
5621 	 */
5622 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5623 		lpcr &= ~LPCR_VPM0;
5624 		lpcr |= LPCR_HVICE | LPCR_HEIC;
5625 
5626 		/*
5627 		 * If xive is enabled, we route 0x500 interrupts directly
5628 		 * to the guest.
5629 		 */
5630 		if (xics_on_xive())
5631 			lpcr |= LPCR_LPES;
5632 	}
5633 
5634 	/*
5635 	 * If the host uses radix, the guest starts out as radix.
5636 	 */
5637 	if (radix_enabled()) {
5638 		kvm->arch.radix = 1;
5639 		kvm->arch.mmu_ready = 1;
5640 		lpcr &= ~LPCR_VPM1;
5641 		lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5642 		if (cpu_has_feature(CPU_FTR_HVMODE) &&
5643 		    cpu_has_feature(CPU_FTR_ARCH_31) &&
5644 		    (kvm->arch.host_lpcr & LPCR_HAIL))
5645 			lpcr |= LPCR_HAIL;
5646 		ret = kvmppc_init_vm_radix(kvm);
5647 		if (ret) {
5648 			if (kvmhv_is_nestedv2())
5649 				plpar_guest_delete(0, kvm->arch.lpid);
5650 			else
5651 				kvmppc_free_lpid(kvm->arch.lpid);
5652 			return ret;
5653 		}
5654 		kvmppc_setup_partition_table(kvm);
5655 	}
5656 
5657 	verify_lpcr(kvm, lpcr);
5658 	kvm->arch.lpcr = lpcr;
5659 
5660 	/* Initialization for future HPT resizes */
5661 	kvm->arch.resize_hpt = NULL;
5662 
5663 	/*
5664 	 * Work out how many sets the TLB has, for the use of
5665 	 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
5666 	 */
5667 	if (cpu_has_feature(CPU_FTR_ARCH_31)) {
5668 		/*
5669 		 * P10 will flush all the congruence class with a single tlbiel
5670 		 */
5671 		kvm->arch.tlb_sets = 1;
5672 	} else if (radix_enabled())
5673 		kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;	/* 128 */
5674 	else if (cpu_has_feature(CPU_FTR_ARCH_300))
5675 		kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;	/* 256 */
5676 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
5677 		kvm->arch.tlb_sets = POWER8_TLB_SETS;		/* 512 */
5678 	else
5679 		kvm->arch.tlb_sets = POWER7_TLB_SETS;		/* 128 */
5680 
5681 	/*
5682 	 * Track that we now have a HV mode VM active. This blocks secondary
5683 	 * CPU threads from coming online.
5684 	 */
5685 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5686 		kvm_hv_vm_activated();
5687 
5688 	/*
5689 	 * Initialize smt_mode depending on processor.
5690 	 * POWER8 and earlier have to use "strict" threading, where
5691 	 * all vCPUs in a vcore have to run on the same (sub)core,
5692 	 * whereas on POWER9 the threads can each run a different
5693 	 * guest.
5694 	 */
5695 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5696 		kvm->arch.smt_mode = threads_per_subcore;
5697 	else
5698 		kvm->arch.smt_mode = 1;
5699 	kvm->arch.emul_smt_mode = 1;
5700 
5701 	return 0;
5702 }
5703 
5704 static int kvmppc_arch_create_vm_debugfs_hv(struct kvm *kvm)
5705 {
5706 	kvmppc_mmu_debugfs_init(kvm);
5707 	if (radix_enabled())
5708 		kvmhv_radix_debugfs_init(kvm);
5709 	return 0;
5710 }
5711 
5712 static void kvmppc_free_vcores(struct kvm *kvm)
5713 {
5714 	long int i;
5715 
5716 	for (i = 0; i < KVM_MAX_VCORES; ++i)
5717 		kfree(kvm->arch.vcores[i]);
5718 	kvm->arch.online_vcores = 0;
5719 }
5720 
5721 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
5722 {
5723 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5724 		kvm_hv_vm_deactivated();
5725 
5726 	kvmppc_free_vcores(kvm);
5727 
5728 
5729 	if (kvm_is_radix(kvm))
5730 		kvmppc_free_radix(kvm);
5731 	else
5732 		kvmppc_free_hpt(&kvm->arch.hpt);
5733 
5734 	/* Perform global invalidation and return lpid to the pool */
5735 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5736 		if (nesting_enabled(kvm))
5737 			kvmhv_release_all_nested(kvm);
5738 		kvm->arch.process_table = 0;
5739 		if (kvm->arch.secure_guest)
5740 			uv_svm_terminate(kvm->arch.lpid);
5741 		if (!kvmhv_is_nestedv2())
5742 			kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
5743 	}
5744 
5745 	if (kvmhv_is_nestedv2()) {
5746 		kvmhv_flush_lpid(kvm->arch.lpid);
5747 		plpar_guest_delete(0, kvm->arch.lpid);
5748 	} else {
5749 		kvmppc_free_lpid(kvm->arch.lpid);
5750 	}
5751 
5752 	kvmppc_free_pimap(kvm);
5753 }
5754 
5755 /* We don't need to emulate any privileged instructions or dcbz */
5756 static int kvmppc_core_emulate_op_hv(struct kvm_vcpu *vcpu,
5757 				     unsigned int inst, int *advance)
5758 {
5759 	return EMULATE_FAIL;
5760 }
5761 
5762 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
5763 					ulong spr_val)
5764 {
5765 	return EMULATE_FAIL;
5766 }
5767 
5768 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
5769 					ulong *spr_val)
5770 {
5771 	return EMULATE_FAIL;
5772 }
5773 
5774 static int kvmppc_core_check_processor_compat_hv(void)
5775 {
5776 	if (cpu_has_feature(CPU_FTR_HVMODE) &&
5777 	    cpu_has_feature(CPU_FTR_ARCH_206))
5778 		return 0;
5779 
5780 	/* POWER9 in radix mode is capable of being a nested hypervisor. */
5781 	if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
5782 		return 0;
5783 
5784 	return -EIO;
5785 }
5786 
5787 #ifdef CONFIG_KVM_XICS
5788 
5789 void kvmppc_free_pimap(struct kvm *kvm)
5790 {
5791 	kfree(kvm->arch.pimap);
5792 }
5793 
5794 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
5795 {
5796 	return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
5797 }
5798 
5799 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5800 {
5801 	struct irq_desc *desc;
5802 	struct kvmppc_irq_map *irq_map;
5803 	struct kvmppc_passthru_irqmap *pimap;
5804 	struct irq_chip *chip;
5805 	int i, rc = 0;
5806 	struct irq_data *host_data;
5807 
5808 	if (!kvm_irq_bypass)
5809 		return 1;
5810 
5811 	desc = irq_to_desc(host_irq);
5812 	if (!desc)
5813 		return -EIO;
5814 
5815 	mutex_lock(&kvm->lock);
5816 
5817 	pimap = kvm->arch.pimap;
5818 	if (pimap == NULL) {
5819 		/* First call, allocate structure to hold IRQ map */
5820 		pimap = kvmppc_alloc_pimap();
5821 		if (pimap == NULL) {
5822 			mutex_unlock(&kvm->lock);
5823 			return -ENOMEM;
5824 		}
5825 		kvm->arch.pimap = pimap;
5826 	}
5827 
5828 	/*
5829 	 * For now, we only support interrupts for which the EOI operation
5830 	 * is an OPAL call followed by a write to XIRR, since that's
5831 	 * what our real-mode EOI code does, or a XIVE interrupt
5832 	 */
5833 	chip = irq_data_get_irq_chip(&desc->irq_data);
5834 	if (!chip || !is_pnv_opal_msi(chip)) {
5835 		pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
5836 			host_irq, guest_gsi);
5837 		mutex_unlock(&kvm->lock);
5838 		return -ENOENT;
5839 	}
5840 
5841 	/*
5842 	 * See if we already have an entry for this guest IRQ number.
5843 	 * If it's mapped to a hardware IRQ number, that's an error,
5844 	 * otherwise re-use this entry.
5845 	 */
5846 	for (i = 0; i < pimap->n_mapped; i++) {
5847 		if (guest_gsi == pimap->mapped[i].v_hwirq) {
5848 			if (pimap->mapped[i].r_hwirq) {
5849 				mutex_unlock(&kvm->lock);
5850 				return -EINVAL;
5851 			}
5852 			break;
5853 		}
5854 	}
5855 
5856 	if (i == KVMPPC_PIRQ_MAPPED) {
5857 		mutex_unlock(&kvm->lock);
5858 		return -EAGAIN;		/* table is full */
5859 	}
5860 
5861 	irq_map = &pimap->mapped[i];
5862 
5863 	irq_map->v_hwirq = guest_gsi;
5864 	irq_map->desc = desc;
5865 
5866 	/*
5867 	 * Order the above two stores before the next to serialize with
5868 	 * the KVM real mode handler.
5869 	 */
5870 	smp_wmb();
5871 
5872 	/*
5873 	 * The 'host_irq' number is mapped in the PCI-MSI domain but
5874 	 * the underlying calls, which will EOI the interrupt in real
5875 	 * mode, need an HW IRQ number mapped in the XICS IRQ domain.
5876 	 */
5877 	host_data = irq_domain_get_irq_data(irq_get_default_host(), host_irq);
5878 	irq_map->r_hwirq = (unsigned int)irqd_to_hwirq(host_data);
5879 
5880 	if (i == pimap->n_mapped)
5881 		pimap->n_mapped++;
5882 
5883 	if (xics_on_xive())
5884 		rc = kvmppc_xive_set_mapped(kvm, guest_gsi, host_irq);
5885 	else
5886 		kvmppc_xics_set_mapped(kvm, guest_gsi, irq_map->r_hwirq);
5887 	if (rc)
5888 		irq_map->r_hwirq = 0;
5889 
5890 	mutex_unlock(&kvm->lock);
5891 
5892 	return 0;
5893 }
5894 
5895 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5896 {
5897 	struct irq_desc *desc;
5898 	struct kvmppc_passthru_irqmap *pimap;
5899 	int i, rc = 0;
5900 
5901 	if (!kvm_irq_bypass)
5902 		return 0;
5903 
5904 	desc = irq_to_desc(host_irq);
5905 	if (!desc)
5906 		return -EIO;
5907 
5908 	mutex_lock(&kvm->lock);
5909 	if (!kvm->arch.pimap)
5910 		goto unlock;
5911 
5912 	pimap = kvm->arch.pimap;
5913 
5914 	for (i = 0; i < pimap->n_mapped; i++) {
5915 		if (guest_gsi == pimap->mapped[i].v_hwirq)
5916 			break;
5917 	}
5918 
5919 	if (i == pimap->n_mapped) {
5920 		mutex_unlock(&kvm->lock);
5921 		return -ENODEV;
5922 	}
5923 
5924 	if (xics_on_xive())
5925 		rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, host_irq);
5926 	else
5927 		kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
5928 
5929 	/* invalidate the entry (what to do on error from the above ?) */
5930 	pimap->mapped[i].r_hwirq = 0;
5931 
5932 	/*
5933 	 * We don't free this structure even when the count goes to
5934 	 * zero. The structure is freed when we destroy the VM.
5935 	 */
5936  unlock:
5937 	mutex_unlock(&kvm->lock);
5938 	return rc;
5939 }
5940 
5941 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5942 					     struct irq_bypass_producer *prod)
5943 {
5944 	int ret = 0;
5945 	struct kvm_kernel_irqfd *irqfd =
5946 		container_of(cons, struct kvm_kernel_irqfd, consumer);
5947 
5948 	irqfd->producer = prod;
5949 
5950 	ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5951 	if (ret)
5952 		pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5953 			prod->irq, irqfd->gsi, ret);
5954 
5955 	return ret;
5956 }
5957 
5958 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5959 					      struct irq_bypass_producer *prod)
5960 {
5961 	int ret;
5962 	struct kvm_kernel_irqfd *irqfd =
5963 		container_of(cons, struct kvm_kernel_irqfd, consumer);
5964 
5965 	irqfd->producer = NULL;
5966 
5967 	/*
5968 	 * When producer of consumer is unregistered, we change back to
5969 	 * default external interrupt handling mode - KVM real mode
5970 	 * will switch back to host.
5971 	 */
5972 	ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5973 	if (ret)
5974 		pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5975 			prod->irq, irqfd->gsi, ret);
5976 }
5977 #endif
5978 
5979 static int kvm_arch_vm_ioctl_hv(struct file *filp,
5980 				unsigned int ioctl, unsigned long arg)
5981 {
5982 	struct kvm *kvm __maybe_unused = filp->private_data;
5983 	void __user *argp = (void __user *)arg;
5984 	int r;
5985 
5986 	switch (ioctl) {
5987 
5988 	case KVM_PPC_ALLOCATE_HTAB: {
5989 		u32 htab_order;
5990 
5991 		/* If we're a nested hypervisor, we currently only support radix */
5992 		if (kvmhv_on_pseries()) {
5993 			r = -EOPNOTSUPP;
5994 			break;
5995 		}
5996 
5997 		r = -EFAULT;
5998 		if (get_user(htab_order, (u32 __user *)argp))
5999 			break;
6000 		r = kvmppc_alloc_reset_hpt(kvm, htab_order);
6001 		if (r)
6002 			break;
6003 		r = 0;
6004 		break;
6005 	}
6006 
6007 	case KVM_PPC_GET_HTAB_FD: {
6008 		struct kvm_get_htab_fd ghf;
6009 
6010 		r = -EFAULT;
6011 		if (copy_from_user(&ghf, argp, sizeof(ghf)))
6012 			break;
6013 		r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
6014 		break;
6015 	}
6016 
6017 	case KVM_PPC_RESIZE_HPT_PREPARE: {
6018 		struct kvm_ppc_resize_hpt rhpt;
6019 
6020 		r = -EFAULT;
6021 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
6022 			break;
6023 
6024 		r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
6025 		break;
6026 	}
6027 
6028 	case KVM_PPC_RESIZE_HPT_COMMIT: {
6029 		struct kvm_ppc_resize_hpt rhpt;
6030 
6031 		r = -EFAULT;
6032 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
6033 			break;
6034 
6035 		r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
6036 		break;
6037 	}
6038 
6039 	default:
6040 		r = -ENOTTY;
6041 	}
6042 
6043 	return r;
6044 }
6045 
6046 /*
6047  * List of hcall numbers to enable by default.
6048  * For compatibility with old userspace, we enable by default
6049  * all hcalls that were implemented before the hcall-enabling
6050  * facility was added.  Note this list should not include H_RTAS.
6051  */
6052 static unsigned int default_hcall_list[] = {
6053 	H_REMOVE,
6054 	H_ENTER,
6055 	H_READ,
6056 	H_PROTECT,
6057 	H_BULK_REMOVE,
6058 #ifdef CONFIG_SPAPR_TCE_IOMMU
6059 	H_GET_TCE,
6060 	H_PUT_TCE,
6061 #endif
6062 	H_SET_DABR,
6063 	H_SET_XDABR,
6064 	H_CEDE,
6065 	H_PROD,
6066 	H_CONFER,
6067 	H_REGISTER_VPA,
6068 #ifdef CONFIG_KVM_XICS
6069 	H_EOI,
6070 	H_CPPR,
6071 	H_IPI,
6072 	H_IPOLL,
6073 	H_XIRR,
6074 	H_XIRR_X,
6075 #endif
6076 	0
6077 };
6078 
6079 static void init_default_hcalls(void)
6080 {
6081 	int i;
6082 	unsigned int hcall;
6083 
6084 	for (i = 0; default_hcall_list[i]; ++i) {
6085 		hcall = default_hcall_list[i];
6086 		WARN_ON(!kvmppc_hcall_impl_hv(hcall));
6087 		__set_bit(hcall / 4, default_enabled_hcalls);
6088 	}
6089 }
6090 
6091 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
6092 {
6093 	unsigned long lpcr;
6094 	int radix;
6095 	int err;
6096 
6097 	/* If not on a POWER9, reject it */
6098 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
6099 		return -ENODEV;
6100 
6101 	/* If any unknown flags set, reject it */
6102 	if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
6103 		return -EINVAL;
6104 
6105 	/* GR (guest radix) bit in process_table field must match */
6106 	radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
6107 	if (!!(cfg->process_table & PATB_GR) != radix)
6108 		return -EINVAL;
6109 
6110 	/* Process table size field must be reasonable, i.e. <= 24 */
6111 	if ((cfg->process_table & PRTS_MASK) > 24)
6112 		return -EINVAL;
6113 
6114 	/* We can change a guest to/from radix now, if the host is radix */
6115 	if (radix && !radix_enabled())
6116 		return -EINVAL;
6117 
6118 	/* If we're a nested hypervisor, we currently only support radix */
6119 	if (kvmhv_on_pseries() && !radix)
6120 		return -EINVAL;
6121 
6122 	mutex_lock(&kvm->arch.mmu_setup_lock);
6123 	if (radix != kvm_is_radix(kvm)) {
6124 		if (kvm->arch.mmu_ready) {
6125 			kvm->arch.mmu_ready = 0;
6126 			/* order mmu_ready vs. vcpus_running */
6127 			smp_mb();
6128 			if (atomic_read(&kvm->arch.vcpus_running)) {
6129 				kvm->arch.mmu_ready = 1;
6130 				err = -EBUSY;
6131 				goto out_unlock;
6132 			}
6133 		}
6134 		if (radix)
6135 			err = kvmppc_switch_mmu_to_radix(kvm);
6136 		else
6137 			err = kvmppc_switch_mmu_to_hpt(kvm);
6138 		if (err)
6139 			goto out_unlock;
6140 	}
6141 
6142 	kvm->arch.process_table = cfg->process_table;
6143 	kvmppc_setup_partition_table(kvm);
6144 
6145 	lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
6146 	kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
6147 	err = 0;
6148 
6149  out_unlock:
6150 	mutex_unlock(&kvm->arch.mmu_setup_lock);
6151 	return err;
6152 }
6153 
6154 static int kvmhv_enable_nested(struct kvm *kvm)
6155 {
6156 	if (!nested)
6157 		return -EPERM;
6158 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
6159 		return -ENODEV;
6160 	if (!radix_enabled())
6161 		return -ENODEV;
6162 	if (kvmhv_is_nestedv2())
6163 		return -ENODEV;
6164 
6165 	/* kvm == NULL means the caller is testing if the capability exists */
6166 	if (kvm)
6167 		kvm->arch.nested_enable = true;
6168 	return 0;
6169 }
6170 
6171 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
6172 				 int size)
6173 {
6174 	int rc = -EINVAL;
6175 
6176 	if (kvmhv_vcpu_is_radix(vcpu)) {
6177 		rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
6178 
6179 		if (rc > 0)
6180 			rc = -EINVAL;
6181 	}
6182 
6183 	/* For now quadrants are the only way to access nested guest memory */
6184 	if (rc && vcpu->arch.nested)
6185 		rc = -EAGAIN;
6186 
6187 	return rc;
6188 }
6189 
6190 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
6191 				int size)
6192 {
6193 	int rc = -EINVAL;
6194 
6195 	if (kvmhv_vcpu_is_radix(vcpu)) {
6196 		rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
6197 
6198 		if (rc > 0)
6199 			rc = -EINVAL;
6200 	}
6201 
6202 	/* For now quadrants are the only way to access nested guest memory */
6203 	if (rc && vcpu->arch.nested)
6204 		rc = -EAGAIN;
6205 
6206 	return rc;
6207 }
6208 
6209 static void unpin_vpa_reset(struct kvm *kvm, struct kvmppc_vpa *vpa)
6210 {
6211 	unpin_vpa(kvm, vpa);
6212 	vpa->gpa = 0;
6213 	vpa->pinned_addr = NULL;
6214 	vpa->dirty = false;
6215 	vpa->update_pending = 0;
6216 }
6217 
6218 /*
6219  * Enable a guest to become a secure VM, or test whether
6220  * that could be enabled.
6221  * Called when the KVM_CAP_PPC_SECURE_GUEST capability is
6222  * tested (kvm == NULL) or enabled (kvm != NULL).
6223  */
6224 static int kvmhv_enable_svm(struct kvm *kvm)
6225 {
6226 	if (!kvmppc_uvmem_available())
6227 		return -EINVAL;
6228 	if (kvm)
6229 		kvm->arch.svm_enabled = 1;
6230 	return 0;
6231 }
6232 
6233 /*
6234  *  IOCTL handler to turn off secure mode of guest
6235  *
6236  * - Release all device pages
6237  * - Issue ucall to terminate the guest on the UV side
6238  * - Unpin the VPA pages.
6239  * - Reinit the partition scoped page tables
6240  */
6241 static int kvmhv_svm_off(struct kvm *kvm)
6242 {
6243 	struct kvm_vcpu *vcpu;
6244 	int mmu_was_ready;
6245 	int srcu_idx;
6246 	int ret = 0;
6247 	unsigned long i;
6248 
6249 	if (!(kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START))
6250 		return ret;
6251 
6252 	mutex_lock(&kvm->arch.mmu_setup_lock);
6253 	mmu_was_ready = kvm->arch.mmu_ready;
6254 	if (kvm->arch.mmu_ready) {
6255 		kvm->arch.mmu_ready = 0;
6256 		/* order mmu_ready vs. vcpus_running */
6257 		smp_mb();
6258 		if (atomic_read(&kvm->arch.vcpus_running)) {
6259 			kvm->arch.mmu_ready = 1;
6260 			ret = -EBUSY;
6261 			goto out;
6262 		}
6263 	}
6264 
6265 	srcu_idx = srcu_read_lock(&kvm->srcu);
6266 	for (i = 0; i < kvm_arch_nr_memslot_as_ids(kvm); i++) {
6267 		struct kvm_memory_slot *memslot;
6268 		struct kvm_memslots *slots = __kvm_memslots(kvm, i);
6269 		int bkt;
6270 
6271 		if (!slots)
6272 			continue;
6273 
6274 		kvm_for_each_memslot(memslot, bkt, slots) {
6275 			kvmppc_uvmem_drop_pages(memslot, kvm, true);
6276 			uv_unregister_mem_slot(kvm->arch.lpid, memslot->id);
6277 		}
6278 	}
6279 	srcu_read_unlock(&kvm->srcu, srcu_idx);
6280 
6281 	ret = uv_svm_terminate(kvm->arch.lpid);
6282 	if (ret != U_SUCCESS) {
6283 		ret = -EINVAL;
6284 		goto out;
6285 	}
6286 
6287 	/*
6288 	 * When secure guest is reset, all the guest pages are sent
6289 	 * to UV via UV_PAGE_IN before the non-boot vcpus get a
6290 	 * chance to run and unpin their VPA pages. Unpinning of all
6291 	 * VPA pages is done here explicitly so that VPA pages
6292 	 * can be migrated to the secure side.
6293 	 *
6294 	 * This is required to for the secure SMP guest to reboot
6295 	 * correctly.
6296 	 */
6297 	kvm_for_each_vcpu(i, vcpu, kvm) {
6298 		spin_lock(&vcpu->arch.vpa_update_lock);
6299 		unpin_vpa_reset(kvm, &vcpu->arch.dtl);
6300 		unpin_vpa_reset(kvm, &vcpu->arch.slb_shadow);
6301 		unpin_vpa_reset(kvm, &vcpu->arch.vpa);
6302 		spin_unlock(&vcpu->arch.vpa_update_lock);
6303 	}
6304 
6305 	kvmppc_setup_partition_table(kvm);
6306 	kvm->arch.secure_guest = 0;
6307 	kvm->arch.mmu_ready = mmu_was_ready;
6308 out:
6309 	mutex_unlock(&kvm->arch.mmu_setup_lock);
6310 	return ret;
6311 }
6312 
6313 static int kvmhv_enable_dawr1(struct kvm *kvm)
6314 {
6315 	if (!cpu_has_feature(CPU_FTR_DAWR1))
6316 		return -ENODEV;
6317 
6318 	/* kvm == NULL means the caller is testing if the capability exists */
6319 	if (kvm)
6320 		kvm->arch.dawr1_enabled = true;
6321 	return 0;
6322 }
6323 
6324 static bool kvmppc_hash_v3_possible(void)
6325 {
6326 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
6327 		return false;
6328 
6329 	if (!cpu_has_feature(CPU_FTR_HVMODE))
6330 		return false;
6331 
6332 	/*
6333 	 * POWER9 chips before version 2.02 can't have some threads in
6334 	 * HPT mode and some in radix mode on the same core.
6335 	 */
6336 	if (radix_enabled()) {
6337 		unsigned int pvr = mfspr(SPRN_PVR);
6338 		if ((pvr >> 16) == PVR_POWER9 &&
6339 		    (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
6340 		     ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
6341 			return false;
6342 	}
6343 
6344 	return true;
6345 }
6346 
6347 static struct kvmppc_ops kvm_ops_hv = {
6348 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
6349 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
6350 	.get_one_reg = kvmppc_get_one_reg_hv,
6351 	.set_one_reg = kvmppc_set_one_reg_hv,
6352 	.vcpu_load   = kvmppc_core_vcpu_load_hv,
6353 	.vcpu_put    = kvmppc_core_vcpu_put_hv,
6354 	.inject_interrupt = kvmppc_inject_interrupt_hv,
6355 	.set_msr     = kvmppc_set_msr_hv,
6356 	.vcpu_run    = kvmppc_vcpu_run_hv,
6357 	.vcpu_create = kvmppc_core_vcpu_create_hv,
6358 	.vcpu_free   = kvmppc_core_vcpu_free_hv,
6359 	.check_requests = kvmppc_core_check_requests_hv,
6360 	.get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
6361 	.flush_memslot  = kvmppc_core_flush_memslot_hv,
6362 	.prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
6363 	.commit_memory_region  = kvmppc_core_commit_memory_region_hv,
6364 	.unmap_gfn_range = kvm_unmap_gfn_range_hv,
6365 	.age_gfn = kvm_age_gfn_hv,
6366 	.test_age_gfn = kvm_test_age_gfn_hv,
6367 	.set_spte_gfn = kvm_set_spte_gfn_hv,
6368 	.free_memslot = kvmppc_core_free_memslot_hv,
6369 	.init_vm =  kvmppc_core_init_vm_hv,
6370 	.destroy_vm = kvmppc_core_destroy_vm_hv,
6371 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
6372 	.emulate_op = kvmppc_core_emulate_op_hv,
6373 	.emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
6374 	.emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
6375 	.fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
6376 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
6377 	.hcall_implemented = kvmppc_hcall_impl_hv,
6378 #ifdef CONFIG_KVM_XICS
6379 	.irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
6380 	.irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
6381 #endif
6382 	.configure_mmu = kvmhv_configure_mmu,
6383 	.get_rmmu_info = kvmhv_get_rmmu_info,
6384 	.set_smt_mode = kvmhv_set_smt_mode,
6385 	.enable_nested = kvmhv_enable_nested,
6386 	.load_from_eaddr = kvmhv_load_from_eaddr,
6387 	.store_to_eaddr = kvmhv_store_to_eaddr,
6388 	.enable_svm = kvmhv_enable_svm,
6389 	.svm_off = kvmhv_svm_off,
6390 	.enable_dawr1 = kvmhv_enable_dawr1,
6391 	.hash_v3_possible = kvmppc_hash_v3_possible,
6392 	.create_vcpu_debugfs = kvmppc_arch_create_vcpu_debugfs_hv,
6393 	.create_vm_debugfs = kvmppc_arch_create_vm_debugfs_hv,
6394 };
6395 
6396 static int kvm_init_subcore_bitmap(void)
6397 {
6398 	int i, j;
6399 	int nr_cores = cpu_nr_cores();
6400 	struct sibling_subcore_state *sibling_subcore_state;
6401 
6402 	for (i = 0; i < nr_cores; i++) {
6403 		int first_cpu = i * threads_per_core;
6404 		int node = cpu_to_node(first_cpu);
6405 
6406 		/* Ignore if it is already allocated. */
6407 		if (paca_ptrs[first_cpu]->sibling_subcore_state)
6408 			continue;
6409 
6410 		sibling_subcore_state =
6411 			kzalloc_node(sizeof(struct sibling_subcore_state),
6412 							GFP_KERNEL, node);
6413 		if (!sibling_subcore_state)
6414 			return -ENOMEM;
6415 
6416 
6417 		for (j = 0; j < threads_per_core; j++) {
6418 			int cpu = first_cpu + j;
6419 
6420 			paca_ptrs[cpu]->sibling_subcore_state =
6421 						sibling_subcore_state;
6422 		}
6423 	}
6424 	return 0;
6425 }
6426 
6427 static int kvmppc_radix_possible(void)
6428 {
6429 	return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
6430 }
6431 
6432 static int kvmppc_book3s_init_hv(void)
6433 {
6434 	int r;
6435 
6436 	if (!tlbie_capable) {
6437 		pr_err("KVM-HV: Host does not support TLBIE\n");
6438 		return -ENODEV;
6439 	}
6440 
6441 	/*
6442 	 * FIXME!! Do we need to check on all cpus ?
6443 	 */
6444 	r = kvmppc_core_check_processor_compat_hv();
6445 	if (r < 0)
6446 		return -ENODEV;
6447 
6448 	r = kvmhv_nested_init();
6449 	if (r)
6450 		return r;
6451 
6452 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
6453 		r = kvm_init_subcore_bitmap();
6454 		if (r)
6455 			goto err;
6456 	}
6457 
6458 	/*
6459 	 * We need a way of accessing the XICS interrupt controller,
6460 	 * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
6461 	 * indirectly, via OPAL.
6462 	 */
6463 #ifdef CONFIG_SMP
6464 	if (!xics_on_xive() && !kvmhv_on_pseries() &&
6465 	    !local_paca->kvm_hstate.xics_phys) {
6466 		struct device_node *np;
6467 
6468 		np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
6469 		if (!np) {
6470 			pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
6471 			r = -ENODEV;
6472 			goto err;
6473 		}
6474 		/* presence of intc confirmed - node can be dropped again */
6475 		of_node_put(np);
6476 	}
6477 #endif
6478 
6479 	init_default_hcalls();
6480 
6481 	init_vcore_lists();
6482 
6483 	r = kvmppc_mmu_hv_init();
6484 	if (r)
6485 		goto err;
6486 
6487 	if (kvmppc_radix_possible()) {
6488 		r = kvmppc_radix_init();
6489 		if (r)
6490 			goto err;
6491 	}
6492 
6493 	r = kvmppc_uvmem_init();
6494 	if (r < 0) {
6495 		pr_err("KVM-HV: kvmppc_uvmem_init failed %d\n", r);
6496 		return r;
6497 	}
6498 
6499 	kvm_ops_hv.owner = THIS_MODULE;
6500 	kvmppc_hv_ops = &kvm_ops_hv;
6501 
6502 	return 0;
6503 
6504 err:
6505 	kvmhv_nested_exit();
6506 	kvmppc_radix_exit();
6507 
6508 	return r;
6509 }
6510 
6511 static void kvmppc_book3s_exit_hv(void)
6512 {
6513 	kvmppc_uvmem_free();
6514 	kvmppc_free_host_rm_ops();
6515 	if (kvmppc_radix_possible())
6516 		kvmppc_radix_exit();
6517 	kvmppc_hv_ops = NULL;
6518 	kvmhv_nested_exit();
6519 }
6520 
6521 module_init(kvmppc_book3s_init_hv);
6522 module_exit(kvmppc_book3s_exit_hv);
6523 MODULE_LICENSE("GPL");
6524 MODULE_ALIAS_MISCDEV(KVM_MINOR);
6525 MODULE_ALIAS("devname:kvm");
6526