xref: /linux/arch/powerpc/kvm/book3s_hv.c (revision 7f71507851fc7764b36a3221839607d3a45c2025)
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 		if (cpu_has_feature(CPU_FTR_P11_PVR))
404 			cap = H_GUEST_CAP_POWER11;
405 		else
406 			cap = H_GUEST_CAP_POWER10;
407 		break;
408 	default:
409 		break;
410 	}
411 
412 	return cap;
413 }
414 
415 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
416 {
417 	unsigned long host_pcr_bit = 0, guest_pcr_bit = 0, cap = 0;
418 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
419 
420 	/* We can (emulate) our own architecture version and anything older */
421 	if (cpu_has_feature(CPU_FTR_P11_PVR) || cpu_has_feature(CPU_FTR_ARCH_31))
422 		host_pcr_bit = PCR_ARCH_31;
423 	else if (cpu_has_feature(CPU_FTR_ARCH_300))
424 		host_pcr_bit = PCR_ARCH_300;
425 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
426 		host_pcr_bit = PCR_ARCH_207;
427 	else if (cpu_has_feature(CPU_FTR_ARCH_206))
428 		host_pcr_bit = PCR_ARCH_206;
429 	else
430 		host_pcr_bit = PCR_ARCH_205;
431 
432 	/* Determine lowest PCR bit needed to run guest in given PVR level */
433 	guest_pcr_bit = host_pcr_bit;
434 	if (arch_compat) {
435 		switch (arch_compat) {
436 		case PVR_ARCH_205:
437 			guest_pcr_bit = PCR_ARCH_205;
438 			break;
439 		case PVR_ARCH_206:
440 		case PVR_ARCH_206p:
441 			guest_pcr_bit = PCR_ARCH_206;
442 			break;
443 		case PVR_ARCH_207:
444 			guest_pcr_bit = PCR_ARCH_207;
445 			break;
446 		case PVR_ARCH_300:
447 			guest_pcr_bit = PCR_ARCH_300;
448 			break;
449 		case PVR_ARCH_31:
450 		case PVR_ARCH_31_P11:
451 			guest_pcr_bit = PCR_ARCH_31;
452 			break;
453 		default:
454 			return -EINVAL;
455 		}
456 	}
457 
458 	/* Check requested PCR bits don't exceed our capabilities */
459 	if (guest_pcr_bit > host_pcr_bit)
460 		return -EINVAL;
461 
462 	if (kvmhv_on_pseries() && kvmhv_is_nestedv2()) {
463 		/*
464 		 * 'arch_compat == 0' would mean the guest should default to
465 		 * L1's compatibility. In this case, the guest would pick
466 		 * host's PCR and evaluate the corresponding capabilities.
467 		 */
468 		cap = map_pcr_to_cap(guest_pcr_bit);
469 		if (!(cap & nested_capabilities))
470 			return -EINVAL;
471 	}
472 
473 	spin_lock(&vc->lock);
474 	vc->arch_compat = arch_compat;
475 	kvmhv_nestedv2_mark_dirty(vcpu, KVMPPC_GSID_LOGICAL_PVR);
476 	/*
477 	 * Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit
478 	 * Also set all reserved PCR bits
479 	 */
480 	vc->pcr = (host_pcr_bit - guest_pcr_bit) | PCR_MASK;
481 	spin_unlock(&vc->lock);
482 
483 	return 0;
484 }
485 
486 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
487 {
488 	int r;
489 
490 	pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
491 	pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
492 	       vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
493 	for (r = 0; r < 16; ++r)
494 		pr_err("r%2d = %.16lx  r%d = %.16lx\n",
495 		       r, kvmppc_get_gpr(vcpu, r),
496 		       r+16, kvmppc_get_gpr(vcpu, r+16));
497 	pr_err("ctr = %.16lx  lr  = %.16lx\n",
498 	       vcpu->arch.regs.ctr, vcpu->arch.regs.link);
499 	pr_err("srr0 = %.16llx srr1 = %.16llx\n",
500 	       vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
501 	pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
502 	       vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
503 	pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
504 	       vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
505 	pr_err("cr = %.8lx  xer = %.16lx  dsisr = %.8x\n",
506 	       vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
507 	pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
508 	pr_err("fault dar = %.16lx dsisr = %.8x\n",
509 	       vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
510 	pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
511 	for (r = 0; r < vcpu->arch.slb_max; ++r)
512 		pr_err("  ESID = %.16llx VSID = %.16llx\n",
513 		       vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
514 	pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.16lx\n",
515 	       vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
516 	       vcpu->arch.last_inst);
517 }
518 
519 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
520 {
521 	return kvm_get_vcpu_by_id(kvm, id);
522 }
523 
524 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
525 {
526 	vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
527 	vpa->yield_count = cpu_to_be32(1);
528 }
529 
530 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
531 		   unsigned long addr, unsigned long len)
532 {
533 	/* check address is cacheline aligned */
534 	if (addr & (L1_CACHE_BYTES - 1))
535 		return -EINVAL;
536 	spin_lock(&vcpu->arch.vpa_update_lock);
537 	if (v->next_gpa != addr || v->len != len) {
538 		v->next_gpa = addr;
539 		v->len = addr ? len : 0;
540 		v->update_pending = 1;
541 	}
542 	spin_unlock(&vcpu->arch.vpa_update_lock);
543 	return 0;
544 }
545 
546 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
547 struct reg_vpa {
548 	u32 dummy;
549 	union {
550 		__be16 hword;
551 		__be32 word;
552 	} length;
553 };
554 
555 static int vpa_is_registered(struct kvmppc_vpa *vpap)
556 {
557 	if (vpap->update_pending)
558 		return vpap->next_gpa != 0;
559 	return vpap->pinned_addr != NULL;
560 }
561 
562 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
563 				       unsigned long flags,
564 				       unsigned long vcpuid, unsigned long vpa)
565 {
566 	struct kvm *kvm = vcpu->kvm;
567 	unsigned long len, nb;
568 	void *va;
569 	struct kvm_vcpu *tvcpu;
570 	int err;
571 	int subfunc;
572 	struct kvmppc_vpa *vpap;
573 
574 	tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
575 	if (!tvcpu)
576 		return H_PARAMETER;
577 
578 	subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
579 	if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
580 	    subfunc == H_VPA_REG_SLB) {
581 		/* Registering new area - address must be cache-line aligned */
582 		if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
583 			return H_PARAMETER;
584 
585 		/* convert logical addr to kernel addr and read length */
586 		va = kvmppc_pin_guest_page(kvm, vpa, &nb);
587 		if (va == NULL)
588 			return H_PARAMETER;
589 		if (subfunc == H_VPA_REG_VPA)
590 			len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
591 		else
592 			len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
593 		kvmppc_unpin_guest_page(kvm, va, vpa, false);
594 
595 		/* Check length */
596 		if (len > nb || len < sizeof(struct reg_vpa))
597 			return H_PARAMETER;
598 	} else {
599 		vpa = 0;
600 		len = 0;
601 	}
602 
603 	err = H_PARAMETER;
604 	vpap = NULL;
605 	spin_lock(&tvcpu->arch.vpa_update_lock);
606 
607 	switch (subfunc) {
608 	case H_VPA_REG_VPA:		/* register VPA */
609 		/*
610 		 * The size of our lppaca is 1kB because of the way we align
611 		 * it for the guest to avoid crossing a 4kB boundary. We only
612 		 * use 640 bytes of the structure though, so we should accept
613 		 * clients that set a size of 640.
614 		 */
615 		BUILD_BUG_ON(sizeof(struct lppaca) != 640);
616 		if (len < sizeof(struct lppaca))
617 			break;
618 		vpap = &tvcpu->arch.vpa;
619 		err = 0;
620 		break;
621 
622 	case H_VPA_REG_DTL:		/* register DTL */
623 		if (len < sizeof(struct dtl_entry))
624 			break;
625 		len -= len % sizeof(struct dtl_entry);
626 
627 		/* Check that they have previously registered a VPA */
628 		err = H_RESOURCE;
629 		if (!vpa_is_registered(&tvcpu->arch.vpa))
630 			break;
631 
632 		vpap = &tvcpu->arch.dtl;
633 		err = 0;
634 		break;
635 
636 	case H_VPA_REG_SLB:		/* register SLB shadow buffer */
637 		/* Check that they have previously registered a VPA */
638 		err = H_RESOURCE;
639 		if (!vpa_is_registered(&tvcpu->arch.vpa))
640 			break;
641 
642 		vpap = &tvcpu->arch.slb_shadow;
643 		err = 0;
644 		break;
645 
646 	case H_VPA_DEREG_VPA:		/* deregister VPA */
647 		/* Check they don't still have a DTL or SLB buf registered */
648 		err = H_RESOURCE;
649 		if (vpa_is_registered(&tvcpu->arch.dtl) ||
650 		    vpa_is_registered(&tvcpu->arch.slb_shadow))
651 			break;
652 
653 		vpap = &tvcpu->arch.vpa;
654 		err = 0;
655 		break;
656 
657 	case H_VPA_DEREG_DTL:		/* deregister DTL */
658 		vpap = &tvcpu->arch.dtl;
659 		err = 0;
660 		break;
661 
662 	case H_VPA_DEREG_SLB:		/* deregister SLB shadow buffer */
663 		vpap = &tvcpu->arch.slb_shadow;
664 		err = 0;
665 		break;
666 	}
667 
668 	if (vpap) {
669 		vpap->next_gpa = vpa;
670 		vpap->len = len;
671 		vpap->update_pending = 1;
672 	}
673 
674 	spin_unlock(&tvcpu->arch.vpa_update_lock);
675 
676 	return err;
677 }
678 
679 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap,
680 			       struct kvmppc_vpa *old_vpap)
681 {
682 	struct kvm *kvm = vcpu->kvm;
683 	void *va;
684 	unsigned long nb;
685 	unsigned long gpa;
686 
687 	/*
688 	 * We need to pin the page pointed to by vpap->next_gpa,
689 	 * but we can't call kvmppc_pin_guest_page under the lock
690 	 * as it does get_user_pages() and down_read().  So we
691 	 * have to drop the lock, pin the page, then get the lock
692 	 * again and check that a new area didn't get registered
693 	 * in the meantime.
694 	 */
695 	for (;;) {
696 		gpa = vpap->next_gpa;
697 		spin_unlock(&vcpu->arch.vpa_update_lock);
698 		va = NULL;
699 		nb = 0;
700 		if (gpa)
701 			va = kvmppc_pin_guest_page(kvm, gpa, &nb);
702 		spin_lock(&vcpu->arch.vpa_update_lock);
703 		if (gpa == vpap->next_gpa)
704 			break;
705 		/* sigh... unpin that one and try again */
706 		if (va)
707 			kvmppc_unpin_guest_page(kvm, va, gpa, false);
708 	}
709 
710 	vpap->update_pending = 0;
711 	if (va && nb < vpap->len) {
712 		/*
713 		 * If it's now too short, it must be that userspace
714 		 * has changed the mappings underlying guest memory,
715 		 * so unregister the region.
716 		 */
717 		kvmppc_unpin_guest_page(kvm, va, gpa, false);
718 		va = NULL;
719 	}
720 	*old_vpap = *vpap;
721 
722 	vpap->gpa = gpa;
723 	vpap->pinned_addr = va;
724 	vpap->dirty = false;
725 	if (va)
726 		vpap->pinned_end = va + vpap->len;
727 }
728 
729 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
730 {
731 	struct kvm *kvm = vcpu->kvm;
732 	struct kvmppc_vpa old_vpa = { 0 };
733 
734 	if (!(vcpu->arch.vpa.update_pending ||
735 	      vcpu->arch.slb_shadow.update_pending ||
736 	      vcpu->arch.dtl.update_pending))
737 		return;
738 
739 	spin_lock(&vcpu->arch.vpa_update_lock);
740 	if (vcpu->arch.vpa.update_pending) {
741 		kvmppc_update_vpa(vcpu, &vcpu->arch.vpa, &old_vpa);
742 		if (old_vpa.pinned_addr) {
743 			if (kvmhv_is_nestedv2())
744 				kvmhv_nestedv2_set_vpa(vcpu, ~0ull);
745 			kvmppc_unpin_guest_page(kvm, old_vpa.pinned_addr, old_vpa.gpa,
746 						old_vpa.dirty);
747 		}
748 		if (vcpu->arch.vpa.pinned_addr) {
749 			init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
750 			if (kvmhv_is_nestedv2())
751 				kvmhv_nestedv2_set_vpa(vcpu, __pa(vcpu->arch.vpa.pinned_addr));
752 		}
753 	}
754 	if (vcpu->arch.dtl.update_pending) {
755 		kvmppc_update_vpa(vcpu, &vcpu->arch.dtl, &old_vpa);
756 		if (old_vpa.pinned_addr)
757 			kvmppc_unpin_guest_page(kvm, old_vpa.pinned_addr, old_vpa.gpa,
758 						old_vpa.dirty);
759 		vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
760 		vcpu->arch.dtl_index = 0;
761 	}
762 	if (vcpu->arch.slb_shadow.update_pending) {
763 		kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow, &old_vpa);
764 		if (old_vpa.pinned_addr)
765 			kvmppc_unpin_guest_page(kvm, old_vpa.pinned_addr, old_vpa.gpa,
766 						old_vpa.dirty);
767 	}
768 
769 	spin_unlock(&vcpu->arch.vpa_update_lock);
770 }
771 
772 /*
773  * Return the accumulated stolen time for the vcore up until `now'.
774  * The caller should hold the vcore lock.
775  */
776 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
777 {
778 	u64 p;
779 	unsigned long flags;
780 
781 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
782 
783 	spin_lock_irqsave(&vc->stoltb_lock, flags);
784 	p = vc->stolen_tb;
785 	if (vc->vcore_state != VCORE_INACTIVE &&
786 	    vc->preempt_tb != TB_NIL)
787 		p += now - vc->preempt_tb;
788 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
789 	return p;
790 }
791 
792 static void __kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
793 					struct lppaca *vpa,
794 					unsigned int pcpu, u64 now,
795 					unsigned long stolen)
796 {
797 	struct dtl_entry *dt;
798 
799 	dt = vcpu->arch.dtl_ptr;
800 
801 	if (!dt)
802 		return;
803 
804 	dt->dispatch_reason = 7;
805 	dt->preempt_reason = 0;
806 	dt->processor_id = cpu_to_be16(pcpu + vcpu->arch.ptid);
807 	dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
808 	dt->ready_to_enqueue_time = 0;
809 	dt->waiting_to_ready_time = 0;
810 	dt->timebase = cpu_to_be64(now);
811 	dt->fault_addr = 0;
812 	dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
813 	dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
814 
815 	++dt;
816 	if (dt == vcpu->arch.dtl.pinned_end)
817 		dt = vcpu->arch.dtl.pinned_addr;
818 	vcpu->arch.dtl_ptr = dt;
819 	/* order writing *dt vs. writing vpa->dtl_idx */
820 	smp_wmb();
821 	vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
822 
823 	/* vcpu->arch.dtl.dirty is set by the caller */
824 }
825 
826 static void kvmppc_update_vpa_dispatch(struct kvm_vcpu *vcpu,
827 				       struct kvmppc_vcore *vc)
828 {
829 	struct lppaca *vpa;
830 	unsigned long stolen;
831 	unsigned long core_stolen;
832 	u64 now;
833 	unsigned long flags;
834 
835 	vpa = vcpu->arch.vpa.pinned_addr;
836 	if (!vpa)
837 		return;
838 
839 	now = mftb();
840 
841 	core_stolen = vcore_stolen_time(vc, now);
842 	stolen = core_stolen - vcpu->arch.stolen_logged;
843 	vcpu->arch.stolen_logged = core_stolen;
844 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
845 	stolen += vcpu->arch.busy_stolen;
846 	vcpu->arch.busy_stolen = 0;
847 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
848 
849 	vpa->enqueue_dispatch_tb = cpu_to_be64(be64_to_cpu(vpa->enqueue_dispatch_tb) + stolen);
850 
851 	__kvmppc_create_dtl_entry(vcpu, vpa, vc->pcpu, now + kvmppc_get_tb_offset(vcpu), stolen);
852 
853 	vcpu->arch.vpa.dirty = true;
854 }
855 
856 static void kvmppc_update_vpa_dispatch_p9(struct kvm_vcpu *vcpu,
857 				       struct kvmppc_vcore *vc,
858 				       u64 now)
859 {
860 	struct lppaca *vpa;
861 	unsigned long stolen;
862 	unsigned long stolen_delta;
863 
864 	vpa = vcpu->arch.vpa.pinned_addr;
865 	if (!vpa)
866 		return;
867 
868 	stolen = vc->stolen_tb;
869 	stolen_delta = stolen - vcpu->arch.stolen_logged;
870 	vcpu->arch.stolen_logged = stolen;
871 
872 	vpa->enqueue_dispatch_tb = cpu_to_be64(stolen);
873 
874 	__kvmppc_create_dtl_entry(vcpu, vpa, vc->pcpu, now, stolen_delta);
875 
876 	vcpu->arch.vpa.dirty = true;
877 }
878 
879 /* See if there is a doorbell interrupt pending for a vcpu */
880 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
881 {
882 	int thr;
883 	struct kvmppc_vcore *vc;
884 
885 	if (vcpu->arch.doorbell_request)
886 		return true;
887 	if (cpu_has_feature(CPU_FTR_ARCH_300))
888 		return false;
889 	/*
890 	 * Ensure that the read of vcore->dpdes comes after the read
891 	 * of vcpu->doorbell_request.  This barrier matches the
892 	 * smp_wmb() in kvmppc_guest_entry_inject().
893 	 */
894 	smp_rmb();
895 	vc = vcpu->arch.vcore;
896 	thr = vcpu->vcpu_id - vc->first_vcpuid;
897 	return !!(vc->dpdes & (1 << thr));
898 }
899 
900 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
901 {
902 	if (kvmppc_get_arch_compat(vcpu) >= PVR_ARCH_207)
903 		return true;
904 	if ((!kvmppc_get_arch_compat(vcpu)) &&
905 	    cpu_has_feature(CPU_FTR_ARCH_207S))
906 		return true;
907 	return false;
908 }
909 
910 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
911 			     unsigned long resource, unsigned long value1,
912 			     unsigned long value2)
913 {
914 	switch (resource) {
915 	case H_SET_MODE_RESOURCE_SET_CIABR:
916 		if (!kvmppc_power8_compatible(vcpu))
917 			return H_P2;
918 		if (value2)
919 			return H_P4;
920 		if (mflags)
921 			return H_UNSUPPORTED_FLAG_START;
922 		/* Guests can't breakpoint the hypervisor */
923 		if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
924 			return H_P3;
925 		kvmppc_set_ciabr_hv(vcpu, value1);
926 		return H_SUCCESS;
927 	case H_SET_MODE_RESOURCE_SET_DAWR0:
928 		if (!kvmppc_power8_compatible(vcpu))
929 			return H_P2;
930 		if (!ppc_breakpoint_available())
931 			return H_P2;
932 		if (mflags)
933 			return H_UNSUPPORTED_FLAG_START;
934 		if (value2 & DABRX_HYP)
935 			return H_P4;
936 		kvmppc_set_dawr0_hv(vcpu, value1);
937 		kvmppc_set_dawrx0_hv(vcpu, value2);
938 		return H_SUCCESS;
939 	case H_SET_MODE_RESOURCE_SET_DAWR1:
940 		if (!kvmppc_power8_compatible(vcpu))
941 			return H_P2;
942 		if (!ppc_breakpoint_available())
943 			return H_P2;
944 		if (!cpu_has_feature(CPU_FTR_DAWR1))
945 			return H_P2;
946 		if (!vcpu->kvm->arch.dawr1_enabled)
947 			return H_FUNCTION;
948 		if (mflags)
949 			return H_UNSUPPORTED_FLAG_START;
950 		if (value2 & DABRX_HYP)
951 			return H_P4;
952 		kvmppc_set_dawr1_hv(vcpu, value1);
953 		kvmppc_set_dawrx1_hv(vcpu, value2);
954 		return H_SUCCESS;
955 	case H_SET_MODE_RESOURCE_ADDR_TRANS_MODE:
956 		/*
957 		 * KVM does not support mflags=2 (AIL=2) and AIL=1 is reserved.
958 		 * Keep this in synch with kvmppc_filter_guest_lpcr_hv.
959 		 */
960 		if (cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG) &&
961 				kvmhv_vcpu_is_radix(vcpu) && mflags == 3)
962 			return H_UNSUPPORTED_FLAG_START;
963 		return H_TOO_HARD;
964 	default:
965 		return H_TOO_HARD;
966 	}
967 }
968 
969 /* Copy guest memory in place - must reside within a single memslot */
970 static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from,
971 				  unsigned long len)
972 {
973 	struct kvm_memory_slot *to_memslot = NULL;
974 	struct kvm_memory_slot *from_memslot = NULL;
975 	unsigned long to_addr, from_addr;
976 	int r;
977 
978 	/* Get HPA for from address */
979 	from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT);
980 	if (!from_memslot)
981 		return -EFAULT;
982 	if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages)
983 			     << PAGE_SHIFT))
984 		return -EINVAL;
985 	from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT);
986 	if (kvm_is_error_hva(from_addr))
987 		return -EFAULT;
988 	from_addr |= (from & (PAGE_SIZE - 1));
989 
990 	/* Get HPA for to address */
991 	to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT);
992 	if (!to_memslot)
993 		return -EFAULT;
994 	if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages)
995 			   << PAGE_SHIFT))
996 		return -EINVAL;
997 	to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT);
998 	if (kvm_is_error_hva(to_addr))
999 		return -EFAULT;
1000 	to_addr |= (to & (PAGE_SIZE - 1));
1001 
1002 	/* Perform copy */
1003 	r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr,
1004 			     len);
1005 	if (r)
1006 		return -EFAULT;
1007 	mark_page_dirty(kvm, to >> PAGE_SHIFT);
1008 	return 0;
1009 }
1010 
1011 static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags,
1012 			       unsigned long dest, unsigned long src)
1013 {
1014 	u64 pg_sz = SZ_4K;		/* 4K page size */
1015 	u64 pg_mask = SZ_4K - 1;
1016 	int ret;
1017 
1018 	/* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */
1019 	if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE |
1020 		      H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED))
1021 		return H_PARAMETER;
1022 
1023 	/* dest (and src if copy_page flag set) must be page aligned */
1024 	if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask)))
1025 		return H_PARAMETER;
1026 
1027 	/* zero and/or copy the page as determined by the flags */
1028 	if (flags & H_COPY_PAGE) {
1029 		ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz);
1030 		if (ret < 0)
1031 			return H_PARAMETER;
1032 	} else if (flags & H_ZERO_PAGE) {
1033 		ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz);
1034 		if (ret < 0)
1035 			return H_PARAMETER;
1036 	}
1037 
1038 	/* We can ignore the remaining flags */
1039 
1040 	return H_SUCCESS;
1041 }
1042 
1043 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
1044 {
1045 	struct kvmppc_vcore *vcore = target->arch.vcore;
1046 
1047 	/*
1048 	 * We expect to have been called by the real mode handler
1049 	 * (kvmppc_rm_h_confer()) which would have directly returned
1050 	 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
1051 	 * have useful work to do and should not confer) so we don't
1052 	 * recheck that here.
1053 	 *
1054 	 * In the case of the P9 single vcpu per vcore case, the real
1055 	 * mode handler is not called but no other threads are in the
1056 	 * source vcore.
1057 	 */
1058 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
1059 		spin_lock(&vcore->lock);
1060 		if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
1061 		    vcore->vcore_state != VCORE_INACTIVE &&
1062 		    vcore->runner)
1063 			target = vcore->runner;
1064 		spin_unlock(&vcore->lock);
1065 	}
1066 
1067 	return kvm_vcpu_yield_to(target);
1068 }
1069 
1070 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
1071 {
1072 	int yield_count = 0;
1073 	struct lppaca *lppaca;
1074 
1075 	spin_lock(&vcpu->arch.vpa_update_lock);
1076 	lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
1077 	if (lppaca)
1078 		yield_count = be32_to_cpu(lppaca->yield_count);
1079 	spin_unlock(&vcpu->arch.vpa_update_lock);
1080 	return yield_count;
1081 }
1082 
1083 /*
1084  * H_RPT_INVALIDATE hcall handler for nested guests.
1085  *
1086  * Handles only nested process-scoped invalidation requests in L0.
1087  */
1088 static int kvmppc_nested_h_rpt_invalidate(struct kvm_vcpu *vcpu)
1089 {
1090 	unsigned long type = kvmppc_get_gpr(vcpu, 6);
1091 	unsigned long pid, pg_sizes, start, end;
1092 
1093 	/*
1094 	 * The partition-scoped invalidations aren't handled here in L0.
1095 	 */
1096 	if (type & H_RPTI_TYPE_NESTED)
1097 		return RESUME_HOST;
1098 
1099 	pid = kvmppc_get_gpr(vcpu, 4);
1100 	pg_sizes = kvmppc_get_gpr(vcpu, 7);
1101 	start = kvmppc_get_gpr(vcpu, 8);
1102 	end = kvmppc_get_gpr(vcpu, 9);
1103 
1104 	do_h_rpt_invalidate_prt(pid, vcpu->arch.nested->shadow_lpid,
1105 				type, pg_sizes, start, end);
1106 
1107 	kvmppc_set_gpr(vcpu, 3, H_SUCCESS);
1108 	return RESUME_GUEST;
1109 }
1110 
1111 static long kvmppc_h_rpt_invalidate(struct kvm_vcpu *vcpu,
1112 				    unsigned long id, unsigned long target,
1113 				    unsigned long type, unsigned long pg_sizes,
1114 				    unsigned long start, unsigned long end)
1115 {
1116 	if (!kvm_is_radix(vcpu->kvm))
1117 		return H_UNSUPPORTED;
1118 
1119 	if (end < start)
1120 		return H_P5;
1121 
1122 	/*
1123 	 * Partition-scoped invalidation for nested guests.
1124 	 */
1125 	if (type & H_RPTI_TYPE_NESTED) {
1126 		if (!nesting_enabled(vcpu->kvm))
1127 			return H_FUNCTION;
1128 
1129 		/* Support only cores as target */
1130 		if (target != H_RPTI_TARGET_CMMU)
1131 			return H_P2;
1132 
1133 		return do_h_rpt_invalidate_pat(vcpu, id, type, pg_sizes,
1134 					       start, end);
1135 	}
1136 
1137 	/*
1138 	 * Process-scoped invalidation for L1 guests.
1139 	 */
1140 	do_h_rpt_invalidate_prt(id, vcpu->kvm->arch.lpid,
1141 				type, pg_sizes, start, end);
1142 	return H_SUCCESS;
1143 }
1144 
1145 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
1146 {
1147 	struct kvm *kvm = vcpu->kvm;
1148 	unsigned long req = kvmppc_get_gpr(vcpu, 3);
1149 	unsigned long target, ret = H_SUCCESS;
1150 	int yield_count;
1151 	struct kvm_vcpu *tvcpu;
1152 	int idx, rc;
1153 
1154 	if (req <= MAX_HCALL_OPCODE &&
1155 	    !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
1156 		return RESUME_HOST;
1157 
1158 	switch (req) {
1159 	case H_REMOVE:
1160 		ret = kvmppc_h_remove(vcpu, kvmppc_get_gpr(vcpu, 4),
1161 					kvmppc_get_gpr(vcpu, 5),
1162 					kvmppc_get_gpr(vcpu, 6));
1163 		if (ret == H_TOO_HARD)
1164 			return RESUME_HOST;
1165 		break;
1166 	case H_ENTER:
1167 		ret = kvmppc_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
1168 					kvmppc_get_gpr(vcpu, 5),
1169 					kvmppc_get_gpr(vcpu, 6),
1170 					kvmppc_get_gpr(vcpu, 7));
1171 		if (ret == H_TOO_HARD)
1172 			return RESUME_HOST;
1173 		break;
1174 	case H_READ:
1175 		ret = kvmppc_h_read(vcpu, kvmppc_get_gpr(vcpu, 4),
1176 					kvmppc_get_gpr(vcpu, 5));
1177 		if (ret == H_TOO_HARD)
1178 			return RESUME_HOST;
1179 		break;
1180 	case H_CLEAR_MOD:
1181 		ret = kvmppc_h_clear_mod(vcpu, kvmppc_get_gpr(vcpu, 4),
1182 					kvmppc_get_gpr(vcpu, 5));
1183 		if (ret == H_TOO_HARD)
1184 			return RESUME_HOST;
1185 		break;
1186 	case H_CLEAR_REF:
1187 		ret = kvmppc_h_clear_ref(vcpu, kvmppc_get_gpr(vcpu, 4),
1188 					kvmppc_get_gpr(vcpu, 5));
1189 		if (ret == H_TOO_HARD)
1190 			return RESUME_HOST;
1191 		break;
1192 	case H_PROTECT:
1193 		ret = kvmppc_h_protect(vcpu, kvmppc_get_gpr(vcpu, 4),
1194 					kvmppc_get_gpr(vcpu, 5),
1195 					kvmppc_get_gpr(vcpu, 6));
1196 		if (ret == H_TOO_HARD)
1197 			return RESUME_HOST;
1198 		break;
1199 	case H_BULK_REMOVE:
1200 		ret = kvmppc_h_bulk_remove(vcpu);
1201 		if (ret == H_TOO_HARD)
1202 			return RESUME_HOST;
1203 		break;
1204 
1205 	case H_CEDE:
1206 		break;
1207 	case H_PROD:
1208 		target = kvmppc_get_gpr(vcpu, 4);
1209 		tvcpu = kvmppc_find_vcpu(kvm, target);
1210 		if (!tvcpu) {
1211 			ret = H_PARAMETER;
1212 			break;
1213 		}
1214 		tvcpu->arch.prodded = 1;
1215 		smp_mb(); /* This orders prodded store vs ceded load */
1216 		if (tvcpu->arch.ceded)
1217 			kvmppc_fast_vcpu_kick_hv(tvcpu);
1218 		break;
1219 	case H_CONFER:
1220 		target = kvmppc_get_gpr(vcpu, 4);
1221 		if (target == -1)
1222 			break;
1223 		tvcpu = kvmppc_find_vcpu(kvm, target);
1224 		if (!tvcpu) {
1225 			ret = H_PARAMETER;
1226 			break;
1227 		}
1228 		yield_count = kvmppc_get_gpr(vcpu, 5);
1229 		if (kvmppc_get_yield_count(tvcpu) != yield_count)
1230 			break;
1231 		kvm_arch_vcpu_yield_to(tvcpu);
1232 		break;
1233 	case H_REGISTER_VPA:
1234 		ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
1235 					kvmppc_get_gpr(vcpu, 5),
1236 					kvmppc_get_gpr(vcpu, 6));
1237 		break;
1238 	case H_RTAS:
1239 		if (list_empty(&kvm->arch.rtas_tokens))
1240 			return RESUME_HOST;
1241 
1242 		idx = srcu_read_lock(&kvm->srcu);
1243 		rc = kvmppc_rtas_hcall(vcpu);
1244 		srcu_read_unlock(&kvm->srcu, idx);
1245 
1246 		if (rc == -ENOENT)
1247 			return RESUME_HOST;
1248 		else if (rc == 0)
1249 			break;
1250 
1251 		/* Send the error out to userspace via KVM_RUN */
1252 		return rc;
1253 	case H_LOGICAL_CI_LOAD:
1254 		ret = kvmppc_h_logical_ci_load(vcpu);
1255 		if (ret == H_TOO_HARD)
1256 			return RESUME_HOST;
1257 		break;
1258 	case H_LOGICAL_CI_STORE:
1259 		ret = kvmppc_h_logical_ci_store(vcpu);
1260 		if (ret == H_TOO_HARD)
1261 			return RESUME_HOST;
1262 		break;
1263 	case H_SET_MODE:
1264 		ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
1265 					kvmppc_get_gpr(vcpu, 5),
1266 					kvmppc_get_gpr(vcpu, 6),
1267 					kvmppc_get_gpr(vcpu, 7));
1268 		if (ret == H_TOO_HARD)
1269 			return RESUME_HOST;
1270 		break;
1271 	case H_XIRR:
1272 	case H_CPPR:
1273 	case H_EOI:
1274 	case H_IPI:
1275 	case H_IPOLL:
1276 	case H_XIRR_X:
1277 		if (kvmppc_xics_enabled(vcpu)) {
1278 			if (xics_on_xive()) {
1279 				ret = H_NOT_AVAILABLE;
1280 				return RESUME_GUEST;
1281 			}
1282 			ret = kvmppc_xics_hcall(vcpu, req);
1283 			break;
1284 		}
1285 		return RESUME_HOST;
1286 	case H_SET_DABR:
1287 		ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
1288 		break;
1289 	case H_SET_XDABR:
1290 		ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
1291 						kvmppc_get_gpr(vcpu, 5));
1292 		break;
1293 #ifdef CONFIG_SPAPR_TCE_IOMMU
1294 	case H_GET_TCE:
1295 		ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1296 						kvmppc_get_gpr(vcpu, 5));
1297 		if (ret == H_TOO_HARD)
1298 			return RESUME_HOST;
1299 		break;
1300 	case H_PUT_TCE:
1301 		ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1302 						kvmppc_get_gpr(vcpu, 5),
1303 						kvmppc_get_gpr(vcpu, 6));
1304 		if (ret == H_TOO_HARD)
1305 			return RESUME_HOST;
1306 		break;
1307 	case H_PUT_TCE_INDIRECT:
1308 		ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
1309 						kvmppc_get_gpr(vcpu, 5),
1310 						kvmppc_get_gpr(vcpu, 6),
1311 						kvmppc_get_gpr(vcpu, 7));
1312 		if (ret == H_TOO_HARD)
1313 			return RESUME_HOST;
1314 		break;
1315 	case H_STUFF_TCE:
1316 		ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1317 						kvmppc_get_gpr(vcpu, 5),
1318 						kvmppc_get_gpr(vcpu, 6),
1319 						kvmppc_get_gpr(vcpu, 7));
1320 		if (ret == H_TOO_HARD)
1321 			return RESUME_HOST;
1322 		break;
1323 #endif
1324 	case H_RANDOM: {
1325 		unsigned long rand;
1326 
1327 		if (!arch_get_random_seed_longs(&rand, 1))
1328 			ret = H_HARDWARE;
1329 		kvmppc_set_gpr(vcpu, 4, rand);
1330 		break;
1331 	}
1332 	case H_RPT_INVALIDATE:
1333 		ret = kvmppc_h_rpt_invalidate(vcpu, kvmppc_get_gpr(vcpu, 4),
1334 					      kvmppc_get_gpr(vcpu, 5),
1335 					      kvmppc_get_gpr(vcpu, 6),
1336 					      kvmppc_get_gpr(vcpu, 7),
1337 					      kvmppc_get_gpr(vcpu, 8),
1338 					      kvmppc_get_gpr(vcpu, 9));
1339 		break;
1340 
1341 	case H_SET_PARTITION_TABLE:
1342 		ret = H_FUNCTION;
1343 		if (nesting_enabled(kvm))
1344 			ret = kvmhv_set_partition_table(vcpu);
1345 		break;
1346 	case H_ENTER_NESTED:
1347 		ret = H_FUNCTION;
1348 		if (!nesting_enabled(kvm))
1349 			break;
1350 		ret = kvmhv_enter_nested_guest(vcpu);
1351 		if (ret == H_INTERRUPT) {
1352 			kvmppc_set_gpr(vcpu, 3, 0);
1353 			vcpu->arch.hcall_needed = 0;
1354 			return -EINTR;
1355 		} else if (ret == H_TOO_HARD) {
1356 			kvmppc_set_gpr(vcpu, 3, 0);
1357 			vcpu->arch.hcall_needed = 0;
1358 			return RESUME_HOST;
1359 		}
1360 		break;
1361 	case H_TLB_INVALIDATE:
1362 		ret = H_FUNCTION;
1363 		if (nesting_enabled(kvm))
1364 			ret = kvmhv_do_nested_tlbie(vcpu);
1365 		break;
1366 	case H_COPY_TOFROM_GUEST:
1367 		ret = H_FUNCTION;
1368 		if (nesting_enabled(kvm))
1369 			ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1370 		break;
1371 	case H_PAGE_INIT:
1372 		ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4),
1373 					 kvmppc_get_gpr(vcpu, 5),
1374 					 kvmppc_get_gpr(vcpu, 6));
1375 		break;
1376 	case H_SVM_PAGE_IN:
1377 		ret = H_UNSUPPORTED;
1378 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1379 			ret = kvmppc_h_svm_page_in(kvm,
1380 						   kvmppc_get_gpr(vcpu, 4),
1381 						   kvmppc_get_gpr(vcpu, 5),
1382 						   kvmppc_get_gpr(vcpu, 6));
1383 		break;
1384 	case H_SVM_PAGE_OUT:
1385 		ret = H_UNSUPPORTED;
1386 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1387 			ret = kvmppc_h_svm_page_out(kvm,
1388 						    kvmppc_get_gpr(vcpu, 4),
1389 						    kvmppc_get_gpr(vcpu, 5),
1390 						    kvmppc_get_gpr(vcpu, 6));
1391 		break;
1392 	case H_SVM_INIT_START:
1393 		ret = H_UNSUPPORTED;
1394 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1395 			ret = kvmppc_h_svm_init_start(kvm);
1396 		break;
1397 	case H_SVM_INIT_DONE:
1398 		ret = H_UNSUPPORTED;
1399 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1400 			ret = kvmppc_h_svm_init_done(kvm);
1401 		break;
1402 	case H_SVM_INIT_ABORT:
1403 		/*
1404 		 * Even if that call is made by the Ultravisor, the SSR1 value
1405 		 * is the guest context one, with the secure bit clear as it has
1406 		 * not yet been secured. So we can't check it here.
1407 		 * Instead the kvm->arch.secure_guest flag is checked inside
1408 		 * kvmppc_h_svm_init_abort().
1409 		 */
1410 		ret = kvmppc_h_svm_init_abort(kvm);
1411 		break;
1412 
1413 	default:
1414 		return RESUME_HOST;
1415 	}
1416 	WARN_ON_ONCE(ret == H_TOO_HARD);
1417 	kvmppc_set_gpr(vcpu, 3, ret);
1418 	vcpu->arch.hcall_needed = 0;
1419 	return RESUME_GUEST;
1420 }
1421 
1422 /*
1423  * Handle H_CEDE in the P9 path where we don't call the real-mode hcall
1424  * handlers in book3s_hv_rmhandlers.S.
1425  *
1426  * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1427  * that the cede logic in kvmppc_run_single_vcpu() works properly.
1428  */
1429 static void kvmppc_cede(struct kvm_vcpu *vcpu)
1430 {
1431 	__kvmppc_set_msr_hv(vcpu, __kvmppc_get_msr_hv(vcpu) | MSR_EE);
1432 	vcpu->arch.ceded = 1;
1433 	smp_mb();
1434 	if (vcpu->arch.prodded) {
1435 		vcpu->arch.prodded = 0;
1436 		smp_mb();
1437 		vcpu->arch.ceded = 0;
1438 	}
1439 }
1440 
1441 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1442 {
1443 	switch (cmd) {
1444 	case H_CEDE:
1445 	case H_PROD:
1446 	case H_CONFER:
1447 	case H_REGISTER_VPA:
1448 	case H_SET_MODE:
1449 #ifdef CONFIG_SPAPR_TCE_IOMMU
1450 	case H_GET_TCE:
1451 	case H_PUT_TCE:
1452 	case H_PUT_TCE_INDIRECT:
1453 	case H_STUFF_TCE:
1454 #endif
1455 	case H_LOGICAL_CI_LOAD:
1456 	case H_LOGICAL_CI_STORE:
1457 #ifdef CONFIG_KVM_XICS
1458 	case H_XIRR:
1459 	case H_CPPR:
1460 	case H_EOI:
1461 	case H_IPI:
1462 	case H_IPOLL:
1463 	case H_XIRR_X:
1464 #endif
1465 	case H_PAGE_INIT:
1466 	case H_RPT_INVALIDATE:
1467 		return 1;
1468 	}
1469 
1470 	/* See if it's in the real-mode table */
1471 	return kvmppc_hcall_impl_hv_realmode(cmd);
1472 }
1473 
1474 static int kvmppc_emulate_debug_inst(struct kvm_vcpu *vcpu)
1475 {
1476 	ppc_inst_t last_inst;
1477 
1478 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1479 					EMULATE_DONE) {
1480 		/*
1481 		 * Fetch failed, so return to guest and
1482 		 * try executing it again.
1483 		 */
1484 		return RESUME_GUEST;
1485 	}
1486 
1487 	if (ppc_inst_val(last_inst) == KVMPPC_INST_SW_BREAKPOINT) {
1488 		vcpu->run->exit_reason = KVM_EXIT_DEBUG;
1489 		vcpu->run->debug.arch.address = kvmppc_get_pc(vcpu);
1490 		return RESUME_HOST;
1491 	} else {
1492 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL |
1493 				(kvmppc_get_msr(vcpu) & SRR1_PREFIXED));
1494 		return RESUME_GUEST;
1495 	}
1496 }
1497 
1498 static void do_nothing(void *x)
1499 {
1500 }
1501 
1502 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1503 {
1504 	int thr, cpu, pcpu, nthreads;
1505 	struct kvm_vcpu *v;
1506 	unsigned long dpdes;
1507 
1508 	nthreads = vcpu->kvm->arch.emul_smt_mode;
1509 	dpdes = 0;
1510 	cpu = vcpu->vcpu_id & ~(nthreads - 1);
1511 	for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1512 		v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1513 		if (!v)
1514 			continue;
1515 		/*
1516 		 * If the vcpu is currently running on a physical cpu thread,
1517 		 * interrupt it in order to pull it out of the guest briefly,
1518 		 * which will update its vcore->dpdes value.
1519 		 */
1520 		pcpu = READ_ONCE(v->cpu);
1521 		if (pcpu >= 0)
1522 			smp_call_function_single(pcpu, do_nothing, NULL, 1);
1523 		if (kvmppc_doorbell_pending(v))
1524 			dpdes |= 1 << thr;
1525 	}
1526 	return dpdes;
1527 }
1528 
1529 /*
1530  * On POWER9, emulate doorbell-related instructions in order to
1531  * give the guest the illusion of running on a multi-threaded core.
1532  * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1533  * and mfspr DPDES.
1534  */
1535 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1536 {
1537 	u32 inst, rb, thr;
1538 	unsigned long arg;
1539 	struct kvm *kvm = vcpu->kvm;
1540 	struct kvm_vcpu *tvcpu;
1541 	ppc_inst_t pinst;
1542 
1543 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &pinst) != EMULATE_DONE)
1544 		return RESUME_GUEST;
1545 	inst = ppc_inst_val(pinst);
1546 	if (get_op(inst) != 31)
1547 		return EMULATE_FAIL;
1548 	rb = get_rb(inst);
1549 	thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1550 	switch (get_xop(inst)) {
1551 	case OP_31_XOP_MSGSNDP:
1552 		arg = kvmppc_get_gpr(vcpu, rb);
1553 		if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1554 			break;
1555 		arg &= 0x7f;
1556 		if (arg >= kvm->arch.emul_smt_mode)
1557 			break;
1558 		tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1559 		if (!tvcpu)
1560 			break;
1561 		if (!tvcpu->arch.doorbell_request) {
1562 			tvcpu->arch.doorbell_request = 1;
1563 			kvmppc_fast_vcpu_kick_hv(tvcpu);
1564 		}
1565 		break;
1566 	case OP_31_XOP_MSGCLRP:
1567 		arg = kvmppc_get_gpr(vcpu, rb);
1568 		if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1569 			break;
1570 		vcpu->arch.vcore->dpdes = 0;
1571 		vcpu->arch.doorbell_request = 0;
1572 		break;
1573 	case OP_31_XOP_MFSPR:
1574 		switch (get_sprn(inst)) {
1575 		case SPRN_TIR:
1576 			arg = thr;
1577 			break;
1578 		case SPRN_DPDES:
1579 			arg = kvmppc_read_dpdes(vcpu);
1580 			break;
1581 		default:
1582 			return EMULATE_FAIL;
1583 		}
1584 		kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1585 		break;
1586 	default:
1587 		return EMULATE_FAIL;
1588 	}
1589 	kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1590 	return RESUME_GUEST;
1591 }
1592 
1593 /*
1594  * If the lppaca had pmcregs_in_use clear when we exited the guest, then
1595  * HFSCR_PM is cleared for next entry. If the guest then tries to access
1596  * the PMU SPRs, we get this facility unavailable interrupt. Putting HFSCR_PM
1597  * back in the guest HFSCR will cause the next entry to load the PMU SPRs and
1598  * allow the guest access to continue.
1599  */
1600 static int kvmppc_pmu_unavailable(struct kvm_vcpu *vcpu)
1601 {
1602 	if (!(vcpu->arch.hfscr_permitted & HFSCR_PM))
1603 		return EMULATE_FAIL;
1604 
1605 	kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) | HFSCR_PM);
1606 
1607 	return RESUME_GUEST;
1608 }
1609 
1610 static int kvmppc_ebb_unavailable(struct kvm_vcpu *vcpu)
1611 {
1612 	if (!(vcpu->arch.hfscr_permitted & HFSCR_EBB))
1613 		return EMULATE_FAIL;
1614 
1615 	kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) | HFSCR_EBB);
1616 
1617 	return RESUME_GUEST;
1618 }
1619 
1620 static int kvmppc_tm_unavailable(struct kvm_vcpu *vcpu)
1621 {
1622 	if (!(vcpu->arch.hfscr_permitted & HFSCR_TM))
1623 		return EMULATE_FAIL;
1624 
1625 	kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) | HFSCR_TM);
1626 
1627 	return RESUME_GUEST;
1628 }
1629 
1630 static int kvmppc_handle_exit_hv(struct kvm_vcpu *vcpu,
1631 				 struct task_struct *tsk)
1632 {
1633 	struct kvm_run *run = vcpu->run;
1634 	int r = RESUME_HOST;
1635 
1636 	vcpu->stat.sum_exits++;
1637 
1638 	/*
1639 	 * This can happen if an interrupt occurs in the last stages
1640 	 * of guest entry or the first stages of guest exit (i.e. after
1641 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1642 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1643 	 * That can happen due to a bug, or due to a machine check
1644 	 * occurring at just the wrong time.
1645 	 */
1646 	if (!kvmhv_is_nestedv2() && (__kvmppc_get_msr_hv(vcpu) & MSR_HV)) {
1647 		printk(KERN_EMERG "KVM trap in HV mode!\n");
1648 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1649 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1650 			vcpu->arch.shregs.msr);
1651 		kvmppc_dump_regs(vcpu);
1652 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1653 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1654 		return RESUME_HOST;
1655 	}
1656 	run->exit_reason = KVM_EXIT_UNKNOWN;
1657 	run->ready_for_interrupt_injection = 1;
1658 	switch (vcpu->arch.trap) {
1659 	/* We're good on these - the host merely wanted to get our attention */
1660 	case BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER:
1661 		WARN_ON_ONCE(1); /* Should never happen */
1662 		vcpu->arch.trap = BOOK3S_INTERRUPT_HV_DECREMENTER;
1663 		fallthrough;
1664 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1665 		vcpu->stat.dec_exits++;
1666 		r = RESUME_GUEST;
1667 		break;
1668 	case BOOK3S_INTERRUPT_EXTERNAL:
1669 	case BOOK3S_INTERRUPT_H_DOORBELL:
1670 	case BOOK3S_INTERRUPT_H_VIRT:
1671 		vcpu->stat.ext_intr_exits++;
1672 		r = RESUME_GUEST;
1673 		break;
1674 	/* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1675 	case BOOK3S_INTERRUPT_HMI:
1676 	case BOOK3S_INTERRUPT_PERFMON:
1677 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
1678 		r = RESUME_GUEST;
1679 		break;
1680 	case BOOK3S_INTERRUPT_MACHINE_CHECK: {
1681 		static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
1682 					      DEFAULT_RATELIMIT_BURST);
1683 		/*
1684 		 * Print the MCE event to host console. Ratelimit so the guest
1685 		 * can't flood the host log.
1686 		 */
1687 		if (__ratelimit(&rs))
1688 			machine_check_print_event_info(&vcpu->arch.mce_evt,false, true);
1689 
1690 		/*
1691 		 * If the guest can do FWNMI, exit to userspace so it can
1692 		 * deliver a FWNMI to the guest.
1693 		 * Otherwise we synthesize a machine check for the guest
1694 		 * so that it knows that the machine check occurred.
1695 		 */
1696 		if (!vcpu->kvm->arch.fwnmi_enabled) {
1697 			ulong flags = (__kvmppc_get_msr_hv(vcpu) & 0x083c0000) |
1698 					(kvmppc_get_msr(vcpu) & SRR1_PREFIXED);
1699 			kvmppc_core_queue_machine_check(vcpu, flags);
1700 			r = RESUME_GUEST;
1701 			break;
1702 		}
1703 
1704 		/* Exit to guest with KVM_EXIT_NMI as exit reason */
1705 		run->exit_reason = KVM_EXIT_NMI;
1706 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1707 		/* Clear out the old NMI status from run->flags */
1708 		run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1709 		/* Now set the NMI status */
1710 		if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1711 			run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1712 		else
1713 			run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1714 
1715 		r = RESUME_HOST;
1716 		break;
1717 	}
1718 	case BOOK3S_INTERRUPT_PROGRAM:
1719 	{
1720 		ulong flags;
1721 		/*
1722 		 * Normally program interrupts are delivered directly
1723 		 * to the guest by the hardware, but we can get here
1724 		 * as a result of a hypervisor emulation interrupt
1725 		 * (e40) getting turned into a 700 by BML RTAS.
1726 		 */
1727 		flags = (__kvmppc_get_msr_hv(vcpu) & 0x1f0000ull) |
1728 			(kvmppc_get_msr(vcpu) & SRR1_PREFIXED);
1729 		kvmppc_core_queue_program(vcpu, flags);
1730 		r = RESUME_GUEST;
1731 		break;
1732 	}
1733 	case BOOK3S_INTERRUPT_SYSCALL:
1734 	{
1735 		int i;
1736 
1737 		if (!kvmhv_is_nestedv2() && unlikely(__kvmppc_get_msr_hv(vcpu) & MSR_PR)) {
1738 			/*
1739 			 * Guest userspace executed sc 1. This can only be
1740 			 * reached by the P9 path because the old path
1741 			 * handles this case in realmode hcall handlers.
1742 			 */
1743 			if (!kvmhv_vcpu_is_radix(vcpu)) {
1744 				/*
1745 				 * A guest could be running PR KVM, so this
1746 				 * may be a PR KVM hcall. It must be reflected
1747 				 * to the guest kernel as a sc interrupt.
1748 				 */
1749 				kvmppc_core_queue_syscall(vcpu);
1750 			} else {
1751 				/*
1752 				 * Radix guests can not run PR KVM or nested HV
1753 				 * hash guests which might run PR KVM, so this
1754 				 * is always a privilege fault. Send a program
1755 				 * check to guest kernel.
1756 				 */
1757 				kvmppc_core_queue_program(vcpu, SRR1_PROGPRIV);
1758 			}
1759 			r = RESUME_GUEST;
1760 			break;
1761 		}
1762 
1763 		/*
1764 		 * hcall - gather args and set exit_reason. This will next be
1765 		 * handled by kvmppc_pseries_do_hcall which may be able to deal
1766 		 * with it and resume guest, or may punt to userspace.
1767 		 */
1768 		run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1769 		for (i = 0; i < 9; ++i)
1770 			run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1771 		run->exit_reason = KVM_EXIT_PAPR_HCALL;
1772 		vcpu->arch.hcall_needed = 1;
1773 		r = RESUME_HOST;
1774 		break;
1775 	}
1776 	/*
1777 	 * We get these next two if the guest accesses a page which it thinks
1778 	 * it has mapped but which is not actually present, either because
1779 	 * it is for an emulated I/O device or because the corresonding
1780 	 * host page has been paged out.
1781 	 *
1782 	 * Any other HDSI/HISI interrupts have been handled already for P7/8
1783 	 * guests. For POWER9 hash guests not using rmhandlers, basic hash
1784 	 * fault handling is done here.
1785 	 */
1786 	case BOOK3S_INTERRUPT_H_DATA_STORAGE: {
1787 		unsigned long vsid;
1788 		long err;
1789 
1790 		if (cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG) &&
1791 		    unlikely(vcpu->arch.fault_dsisr == HDSISR_CANARY)) {
1792 			r = RESUME_GUEST; /* Just retry if it's the canary */
1793 			break;
1794 		}
1795 
1796 		if (kvm_is_radix(vcpu->kvm) || !cpu_has_feature(CPU_FTR_ARCH_300)) {
1797 			/*
1798 			 * Radix doesn't require anything, and pre-ISAv3.0 hash
1799 			 * already attempted to handle this in rmhandlers. The
1800 			 * hash fault handling below is v3 only (it uses ASDR
1801 			 * via fault_gpa).
1802 			 */
1803 			r = RESUME_PAGE_FAULT;
1804 			break;
1805 		}
1806 
1807 		if (!(vcpu->arch.fault_dsisr & (DSISR_NOHPTE | DSISR_PROTFAULT))) {
1808 			kvmppc_core_queue_data_storage(vcpu,
1809 				kvmppc_get_msr(vcpu) & SRR1_PREFIXED,
1810 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1811 			r = RESUME_GUEST;
1812 			break;
1813 		}
1814 
1815 		if (!(__kvmppc_get_msr_hv(vcpu) & MSR_DR))
1816 			vsid = vcpu->kvm->arch.vrma_slb_v;
1817 		else
1818 			vsid = vcpu->arch.fault_gpa;
1819 
1820 		err = kvmppc_hpte_hv_fault(vcpu, vcpu->arch.fault_dar,
1821 				vsid, vcpu->arch.fault_dsisr, true);
1822 		if (err == 0) {
1823 			r = RESUME_GUEST;
1824 		} else if (err == -1 || err == -2) {
1825 			r = RESUME_PAGE_FAULT;
1826 		} else {
1827 			kvmppc_core_queue_data_storage(vcpu,
1828 				kvmppc_get_msr(vcpu) & SRR1_PREFIXED,
1829 				vcpu->arch.fault_dar, err);
1830 			r = RESUME_GUEST;
1831 		}
1832 		break;
1833 	}
1834 	case BOOK3S_INTERRUPT_H_INST_STORAGE: {
1835 		unsigned long vsid;
1836 		long err;
1837 
1838 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1839 		vcpu->arch.fault_dsisr = __kvmppc_get_msr_hv(vcpu) &
1840 			DSISR_SRR1_MATCH_64S;
1841 		if (kvm_is_radix(vcpu->kvm) || !cpu_has_feature(CPU_FTR_ARCH_300)) {
1842 			/*
1843 			 * Radix doesn't require anything, and pre-ISAv3.0 hash
1844 			 * already attempted to handle this in rmhandlers. The
1845 			 * hash fault handling below is v3 only (it uses ASDR
1846 			 * via fault_gpa).
1847 			 */
1848 			if (__kvmppc_get_msr_hv(vcpu) & HSRR1_HISI_WRITE)
1849 				vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1850 			r = RESUME_PAGE_FAULT;
1851 			break;
1852 		}
1853 
1854 		if (!(vcpu->arch.fault_dsisr & SRR1_ISI_NOPT)) {
1855 			kvmppc_core_queue_inst_storage(vcpu,
1856 				vcpu->arch.fault_dsisr |
1857 				(kvmppc_get_msr(vcpu) & SRR1_PREFIXED));
1858 			r = RESUME_GUEST;
1859 			break;
1860 		}
1861 
1862 		if (!(__kvmppc_get_msr_hv(vcpu) & MSR_IR))
1863 			vsid = vcpu->kvm->arch.vrma_slb_v;
1864 		else
1865 			vsid = vcpu->arch.fault_gpa;
1866 
1867 		err = kvmppc_hpte_hv_fault(vcpu, vcpu->arch.fault_dar,
1868 				vsid, vcpu->arch.fault_dsisr, false);
1869 		if (err == 0) {
1870 			r = RESUME_GUEST;
1871 		} else if (err == -1) {
1872 			r = RESUME_PAGE_FAULT;
1873 		} else {
1874 			kvmppc_core_queue_inst_storage(vcpu,
1875 				err | (kvmppc_get_msr(vcpu) & SRR1_PREFIXED));
1876 			r = RESUME_GUEST;
1877 		}
1878 		break;
1879 	}
1880 
1881 	/*
1882 	 * This occurs if the guest executes an illegal instruction.
1883 	 * If the guest debug is disabled, generate a program interrupt
1884 	 * to the guest. If guest debug is enabled, we need to check
1885 	 * whether the instruction is a software breakpoint instruction.
1886 	 * Accordingly return to Guest or Host.
1887 	 */
1888 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1889 		if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1890 			vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1891 				swab32(vcpu->arch.emul_inst) :
1892 				vcpu->arch.emul_inst;
1893 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1894 			r = kvmppc_emulate_debug_inst(vcpu);
1895 		} else {
1896 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL |
1897 				(kvmppc_get_msr(vcpu) & SRR1_PREFIXED));
1898 			r = RESUME_GUEST;
1899 		}
1900 		break;
1901 
1902 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1903 	case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1904 		/*
1905 		 * This occurs for various TM-related instructions that
1906 		 * we need to emulate on POWER9 DD2.2.  We have already
1907 		 * handled the cases where the guest was in real-suspend
1908 		 * mode and was transitioning to transactional state.
1909 		 */
1910 		r = kvmhv_p9_tm_emulation(vcpu);
1911 		if (r != -1)
1912 			break;
1913 		fallthrough; /* go to facility unavailable handler */
1914 #endif
1915 
1916 	/*
1917 	 * This occurs if the guest (kernel or userspace), does something that
1918 	 * is prohibited by HFSCR.
1919 	 * On POWER9, this could be a doorbell instruction that we need
1920 	 * to emulate.
1921 	 * Otherwise, we just generate a program interrupt to the guest.
1922 	 */
1923 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL: {
1924 		u64 cause = kvmppc_get_hfscr_hv(vcpu) >> 56;
1925 
1926 		r = EMULATE_FAIL;
1927 		if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1928 			switch (cause) {
1929 			case FSCR_MSGP_LG:
1930 				r = kvmppc_emulate_doorbell_instr(vcpu);
1931 				break;
1932 			case FSCR_PM_LG:
1933 				r = kvmppc_pmu_unavailable(vcpu);
1934 				break;
1935 			case FSCR_EBB_LG:
1936 				r = kvmppc_ebb_unavailable(vcpu);
1937 				break;
1938 			case FSCR_TM_LG:
1939 				r = kvmppc_tm_unavailable(vcpu);
1940 				break;
1941 			default:
1942 				break;
1943 			}
1944 		}
1945 		if (r == EMULATE_FAIL) {
1946 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL |
1947 				(kvmppc_get_msr(vcpu) & SRR1_PREFIXED));
1948 			r = RESUME_GUEST;
1949 		}
1950 		break;
1951 	}
1952 
1953 	case BOOK3S_INTERRUPT_HV_RM_HARD:
1954 		r = RESUME_PASSTHROUGH;
1955 		break;
1956 	default:
1957 		kvmppc_dump_regs(vcpu);
1958 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1959 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1960 			__kvmppc_get_msr_hv(vcpu));
1961 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1962 		r = RESUME_HOST;
1963 		break;
1964 	}
1965 
1966 	return r;
1967 }
1968 
1969 static int kvmppc_handle_nested_exit(struct kvm_vcpu *vcpu)
1970 {
1971 	int r;
1972 	int srcu_idx;
1973 
1974 	vcpu->stat.sum_exits++;
1975 
1976 	/*
1977 	 * This can happen if an interrupt occurs in the last stages
1978 	 * of guest entry or the first stages of guest exit (i.e. after
1979 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1980 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1981 	 * That can happen due to a bug, or due to a machine check
1982 	 * occurring at just the wrong time.
1983 	 */
1984 	if (__kvmppc_get_msr_hv(vcpu) & MSR_HV) {
1985 		pr_emerg("KVM trap in HV mode while nested!\n");
1986 		pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1987 			 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1988 			 __kvmppc_get_msr_hv(vcpu));
1989 		kvmppc_dump_regs(vcpu);
1990 		return RESUME_HOST;
1991 	}
1992 	switch (vcpu->arch.trap) {
1993 	/* We're good on these - the host merely wanted to get our attention */
1994 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1995 		vcpu->stat.dec_exits++;
1996 		r = RESUME_GUEST;
1997 		break;
1998 	case BOOK3S_INTERRUPT_EXTERNAL:
1999 		vcpu->stat.ext_intr_exits++;
2000 		r = RESUME_HOST;
2001 		break;
2002 	case BOOK3S_INTERRUPT_H_DOORBELL:
2003 	case BOOK3S_INTERRUPT_H_VIRT:
2004 		vcpu->stat.ext_intr_exits++;
2005 		r = RESUME_GUEST;
2006 		break;
2007 	/* These need to go to the nested HV */
2008 	case BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER:
2009 		vcpu->arch.trap = BOOK3S_INTERRUPT_HV_DECREMENTER;
2010 		vcpu->stat.dec_exits++;
2011 		r = RESUME_HOST;
2012 		break;
2013 	/* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
2014 	case BOOK3S_INTERRUPT_HMI:
2015 	case BOOK3S_INTERRUPT_PERFMON:
2016 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
2017 		r = RESUME_GUEST;
2018 		break;
2019 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
2020 	{
2021 		static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
2022 					      DEFAULT_RATELIMIT_BURST);
2023 		/* Pass the machine check to the L1 guest */
2024 		r = RESUME_HOST;
2025 		/* Print the MCE event to host console. */
2026 		if (__ratelimit(&rs))
2027 			machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
2028 		break;
2029 	}
2030 	/*
2031 	 * We get these next two if the guest accesses a page which it thinks
2032 	 * it has mapped but which is not actually present, either because
2033 	 * it is for an emulated I/O device or because the corresonding
2034 	 * host page has been paged out.
2035 	 */
2036 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
2037 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2038 		r = kvmhv_nested_page_fault(vcpu);
2039 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2040 		break;
2041 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
2042 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
2043 		vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
2044 					 DSISR_SRR1_MATCH_64S;
2045 		if (__kvmppc_get_msr_hv(vcpu) & HSRR1_HISI_WRITE)
2046 			vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
2047 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2048 		r = kvmhv_nested_page_fault(vcpu);
2049 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2050 		break;
2051 
2052 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2053 	case BOOK3S_INTERRUPT_HV_SOFTPATCH:
2054 		/*
2055 		 * This occurs for various TM-related instructions that
2056 		 * we need to emulate on POWER9 DD2.2.  We have already
2057 		 * handled the cases where the guest was in real-suspend
2058 		 * mode and was transitioning to transactional state.
2059 		 */
2060 		r = kvmhv_p9_tm_emulation(vcpu);
2061 		if (r != -1)
2062 			break;
2063 		fallthrough; /* go to facility unavailable handler */
2064 #endif
2065 
2066 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
2067 		r = RESUME_HOST;
2068 		break;
2069 
2070 	case BOOK3S_INTERRUPT_HV_RM_HARD:
2071 		vcpu->arch.trap = 0;
2072 		r = RESUME_GUEST;
2073 		if (!xics_on_xive())
2074 			kvmppc_xics_rm_complete(vcpu, 0);
2075 		break;
2076 	case BOOK3S_INTERRUPT_SYSCALL:
2077 	{
2078 		unsigned long req = kvmppc_get_gpr(vcpu, 3);
2079 
2080 		/*
2081 		 * The H_RPT_INVALIDATE hcalls issued by nested
2082 		 * guests for process-scoped invalidations when
2083 		 * GTSE=0, are handled here in L0.
2084 		 */
2085 		if (req == H_RPT_INVALIDATE) {
2086 			r = kvmppc_nested_h_rpt_invalidate(vcpu);
2087 			break;
2088 		}
2089 
2090 		r = RESUME_HOST;
2091 		break;
2092 	}
2093 	default:
2094 		r = RESUME_HOST;
2095 		break;
2096 	}
2097 
2098 	return r;
2099 }
2100 
2101 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
2102 					    struct kvm_sregs *sregs)
2103 {
2104 	int i;
2105 
2106 	memset(sregs, 0, sizeof(struct kvm_sregs));
2107 	sregs->pvr = vcpu->arch.pvr;
2108 	for (i = 0; i < vcpu->arch.slb_max; i++) {
2109 		sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
2110 		sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
2111 	}
2112 
2113 	return 0;
2114 }
2115 
2116 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
2117 					    struct kvm_sregs *sregs)
2118 {
2119 	int i, j;
2120 
2121 	/* Only accept the same PVR as the host's, since we can't spoof it */
2122 	if (sregs->pvr != vcpu->arch.pvr)
2123 		return -EINVAL;
2124 
2125 	j = 0;
2126 	for (i = 0; i < vcpu->arch.slb_nr; i++) {
2127 		if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
2128 			vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
2129 			vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
2130 			++j;
2131 		}
2132 	}
2133 	vcpu->arch.slb_max = j;
2134 
2135 	return 0;
2136 }
2137 
2138 /*
2139  * Enforce limits on guest LPCR values based on hardware availability,
2140  * guest configuration, and possibly hypervisor support and security
2141  * concerns.
2142  */
2143 unsigned long kvmppc_filter_lpcr_hv(struct kvm *kvm, unsigned long lpcr)
2144 {
2145 	/* LPCR_TC only applies to HPT guests */
2146 	if (kvm_is_radix(kvm))
2147 		lpcr &= ~LPCR_TC;
2148 
2149 	/* On POWER8 and above, userspace can modify AIL */
2150 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2151 		lpcr &= ~LPCR_AIL;
2152 	if ((lpcr & LPCR_AIL) != LPCR_AIL_3)
2153 		lpcr &= ~LPCR_AIL; /* LPCR[AIL]=1/2 is disallowed */
2154 	/*
2155 	 * On some POWER9s we force AIL off for radix guests to prevent
2156 	 * executing in MSR[HV]=1 mode with the MMU enabled and PIDR set to
2157 	 * guest, which can result in Q0 translations with LPID=0 PID=PIDR to
2158 	 * be cached, which the host TLB management does not expect.
2159 	 */
2160 	if (kvm_is_radix(kvm) && cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG))
2161 		lpcr &= ~LPCR_AIL;
2162 
2163 	/*
2164 	 * On POWER9, allow userspace to enable large decrementer for the
2165 	 * guest, whether or not the host has it enabled.
2166 	 */
2167 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
2168 		lpcr &= ~LPCR_LD;
2169 
2170 	return lpcr;
2171 }
2172 
2173 static void verify_lpcr(struct kvm *kvm, unsigned long lpcr)
2174 {
2175 	if (lpcr != kvmppc_filter_lpcr_hv(kvm, lpcr)) {
2176 		WARN_ONCE(1, "lpcr 0x%lx differs from filtered 0x%lx\n",
2177 			  lpcr, kvmppc_filter_lpcr_hv(kvm, lpcr));
2178 	}
2179 }
2180 
2181 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
2182 		bool preserve_top32)
2183 {
2184 	struct kvm *kvm = vcpu->kvm;
2185 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
2186 	u64 mask;
2187 
2188 	spin_lock(&vc->lock);
2189 
2190 	/*
2191 	 * Userspace can only modify
2192 	 * DPFD (default prefetch depth), ILE (interrupt little-endian),
2193 	 * TC (translation control), AIL (alternate interrupt location),
2194 	 * LD (large decrementer).
2195 	 * These are subject to restrictions from kvmppc_filter_lcpr_hv().
2196 	 */
2197 	mask = LPCR_DPFD | LPCR_ILE | LPCR_TC | LPCR_AIL | LPCR_LD;
2198 
2199 	/* Broken 32-bit version of LPCR must not clear top bits */
2200 	if (preserve_top32)
2201 		mask &= 0xFFFFFFFF;
2202 
2203 	new_lpcr = kvmppc_filter_lpcr_hv(kvm,
2204 			(vc->lpcr & ~mask) | (new_lpcr & mask));
2205 
2206 	/*
2207 	 * If ILE (interrupt little-endian) has changed, update the
2208 	 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
2209 	 */
2210 	if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
2211 		struct kvm_vcpu *vcpu;
2212 		unsigned long i;
2213 
2214 		kvm_for_each_vcpu(i, vcpu, kvm) {
2215 			if (vcpu->arch.vcore != vc)
2216 				continue;
2217 			if (new_lpcr & LPCR_ILE)
2218 				vcpu->arch.intr_msr |= MSR_LE;
2219 			else
2220 				vcpu->arch.intr_msr &= ~MSR_LE;
2221 		}
2222 	}
2223 
2224 	vc->lpcr = new_lpcr;
2225 	kvmhv_nestedv2_mark_dirty(vcpu, KVMPPC_GSID_LPCR);
2226 
2227 	spin_unlock(&vc->lock);
2228 }
2229 
2230 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
2231 				 union kvmppc_one_reg *val)
2232 {
2233 	int r = 0;
2234 	long int i;
2235 
2236 	switch (id) {
2237 	case KVM_REG_PPC_DEBUG_INST:
2238 		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
2239 		break;
2240 	case KVM_REG_PPC_HIOR:
2241 		*val = get_reg_val(id, 0);
2242 		break;
2243 	case KVM_REG_PPC_DABR:
2244 		*val = get_reg_val(id, vcpu->arch.dabr);
2245 		break;
2246 	case KVM_REG_PPC_DABRX:
2247 		*val = get_reg_val(id, vcpu->arch.dabrx);
2248 		break;
2249 	case KVM_REG_PPC_DSCR:
2250 		*val = get_reg_val(id, kvmppc_get_dscr_hv(vcpu));
2251 		break;
2252 	case KVM_REG_PPC_PURR:
2253 		*val = get_reg_val(id, kvmppc_get_purr_hv(vcpu));
2254 		break;
2255 	case KVM_REG_PPC_SPURR:
2256 		*val = get_reg_val(id, kvmppc_get_spurr_hv(vcpu));
2257 		break;
2258 	case KVM_REG_PPC_AMR:
2259 		*val = get_reg_val(id, kvmppc_get_amr_hv(vcpu));
2260 		break;
2261 	case KVM_REG_PPC_UAMOR:
2262 		*val = get_reg_val(id, kvmppc_get_uamor_hv(vcpu));
2263 		break;
2264 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
2265 		i = id - KVM_REG_PPC_MMCR0;
2266 		*val = get_reg_val(id, kvmppc_get_mmcr_hv(vcpu, i));
2267 		break;
2268 	case KVM_REG_PPC_MMCR2:
2269 		*val = get_reg_val(id, kvmppc_get_mmcr_hv(vcpu, 2));
2270 		break;
2271 	case KVM_REG_PPC_MMCRA:
2272 		*val = get_reg_val(id, kvmppc_get_mmcra_hv(vcpu));
2273 		break;
2274 	case KVM_REG_PPC_MMCRS:
2275 		*val = get_reg_val(id, vcpu->arch.mmcrs);
2276 		break;
2277 	case KVM_REG_PPC_MMCR3:
2278 		*val = get_reg_val(id, kvmppc_get_mmcr_hv(vcpu, 3));
2279 		break;
2280 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
2281 		i = id - KVM_REG_PPC_PMC1;
2282 		*val = get_reg_val(id, kvmppc_get_pmc_hv(vcpu, i));
2283 		break;
2284 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
2285 		i = id - KVM_REG_PPC_SPMC1;
2286 		*val = get_reg_val(id, vcpu->arch.spmc[i]);
2287 		break;
2288 	case KVM_REG_PPC_SIAR:
2289 		*val = get_reg_val(id, kvmppc_get_siar_hv(vcpu));
2290 		break;
2291 	case KVM_REG_PPC_SDAR:
2292 		*val = get_reg_val(id, kvmppc_get_sdar_hv(vcpu));
2293 		break;
2294 	case KVM_REG_PPC_SIER:
2295 		*val = get_reg_val(id, kvmppc_get_sier_hv(vcpu, 0));
2296 		break;
2297 	case KVM_REG_PPC_SIER2:
2298 		*val = get_reg_val(id, kvmppc_get_sier_hv(vcpu, 1));
2299 		break;
2300 	case KVM_REG_PPC_SIER3:
2301 		*val = get_reg_val(id, kvmppc_get_sier_hv(vcpu, 2));
2302 		break;
2303 	case KVM_REG_PPC_IAMR:
2304 		*val = get_reg_val(id, kvmppc_get_iamr_hv(vcpu));
2305 		break;
2306 	case KVM_REG_PPC_PSPB:
2307 		*val = get_reg_val(id, kvmppc_get_pspb_hv(vcpu));
2308 		break;
2309 	case KVM_REG_PPC_DPDES:
2310 		/*
2311 		 * On POWER9, where we are emulating msgsndp etc.,
2312 		 * we return 1 bit for each vcpu, which can come from
2313 		 * either vcore->dpdes or doorbell_request.
2314 		 * On POWER8, doorbell_request is 0.
2315 		 */
2316 		if (cpu_has_feature(CPU_FTR_ARCH_300))
2317 			*val = get_reg_val(id, vcpu->arch.doorbell_request);
2318 		else
2319 			*val = get_reg_val(id, vcpu->arch.vcore->dpdes);
2320 		break;
2321 	case KVM_REG_PPC_VTB:
2322 		*val = get_reg_val(id, kvmppc_get_vtb(vcpu));
2323 		break;
2324 	case KVM_REG_PPC_DAWR:
2325 		*val = get_reg_val(id, kvmppc_get_dawr0_hv(vcpu));
2326 		break;
2327 	case KVM_REG_PPC_DAWRX:
2328 		*val = get_reg_val(id, kvmppc_get_dawrx0_hv(vcpu));
2329 		break;
2330 	case KVM_REG_PPC_DAWR1:
2331 		*val = get_reg_val(id, kvmppc_get_dawr1_hv(vcpu));
2332 		break;
2333 	case KVM_REG_PPC_DAWRX1:
2334 		*val = get_reg_val(id, kvmppc_get_dawrx1_hv(vcpu));
2335 		break;
2336 	case KVM_REG_PPC_DEXCR:
2337 		*val = get_reg_val(id, kvmppc_get_dexcr_hv(vcpu));
2338 		break;
2339 	case KVM_REG_PPC_HASHKEYR:
2340 		*val = get_reg_val(id, kvmppc_get_hashkeyr_hv(vcpu));
2341 		break;
2342 	case KVM_REG_PPC_HASHPKEYR:
2343 		*val = get_reg_val(id, kvmppc_get_hashpkeyr_hv(vcpu));
2344 		break;
2345 	case KVM_REG_PPC_CIABR:
2346 		*val = get_reg_val(id, kvmppc_get_ciabr_hv(vcpu));
2347 		break;
2348 	case KVM_REG_PPC_CSIGR:
2349 		*val = get_reg_val(id, vcpu->arch.csigr);
2350 		break;
2351 	case KVM_REG_PPC_TACR:
2352 		*val = get_reg_val(id, vcpu->arch.tacr);
2353 		break;
2354 	case KVM_REG_PPC_TCSCR:
2355 		*val = get_reg_val(id, vcpu->arch.tcscr);
2356 		break;
2357 	case KVM_REG_PPC_PID:
2358 		*val = get_reg_val(id, kvmppc_get_pid(vcpu));
2359 		break;
2360 	case KVM_REG_PPC_ACOP:
2361 		*val = get_reg_val(id, vcpu->arch.acop);
2362 		break;
2363 	case KVM_REG_PPC_WORT:
2364 		*val = get_reg_val(id, kvmppc_get_wort_hv(vcpu));
2365 		break;
2366 	case KVM_REG_PPC_TIDR:
2367 		*val = get_reg_val(id, vcpu->arch.tid);
2368 		break;
2369 	case KVM_REG_PPC_PSSCR:
2370 		*val = get_reg_val(id, vcpu->arch.psscr);
2371 		break;
2372 	case KVM_REG_PPC_VPA_ADDR:
2373 		spin_lock(&vcpu->arch.vpa_update_lock);
2374 		*val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
2375 		spin_unlock(&vcpu->arch.vpa_update_lock);
2376 		break;
2377 	case KVM_REG_PPC_VPA_SLB:
2378 		spin_lock(&vcpu->arch.vpa_update_lock);
2379 		val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
2380 		val->vpaval.length = vcpu->arch.slb_shadow.len;
2381 		spin_unlock(&vcpu->arch.vpa_update_lock);
2382 		break;
2383 	case KVM_REG_PPC_VPA_DTL:
2384 		spin_lock(&vcpu->arch.vpa_update_lock);
2385 		val->vpaval.addr = vcpu->arch.dtl.next_gpa;
2386 		val->vpaval.length = vcpu->arch.dtl.len;
2387 		spin_unlock(&vcpu->arch.vpa_update_lock);
2388 		break;
2389 	case KVM_REG_PPC_TB_OFFSET:
2390 		*val = get_reg_val(id, kvmppc_get_tb_offset(vcpu));
2391 		break;
2392 	case KVM_REG_PPC_LPCR:
2393 	case KVM_REG_PPC_LPCR_64:
2394 		*val = get_reg_val(id, kvmppc_get_lpcr(vcpu));
2395 		break;
2396 	case KVM_REG_PPC_PPR:
2397 		*val = get_reg_val(id, kvmppc_get_ppr_hv(vcpu));
2398 		break;
2399 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2400 	case KVM_REG_PPC_TFHAR:
2401 		*val = get_reg_val(id, vcpu->arch.tfhar);
2402 		break;
2403 	case KVM_REG_PPC_TFIAR:
2404 		*val = get_reg_val(id, vcpu->arch.tfiar);
2405 		break;
2406 	case KVM_REG_PPC_TEXASR:
2407 		*val = get_reg_val(id, vcpu->arch.texasr);
2408 		break;
2409 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2410 		i = id - KVM_REG_PPC_TM_GPR0;
2411 		*val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
2412 		break;
2413 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2414 	{
2415 		int j;
2416 		i = id - KVM_REG_PPC_TM_VSR0;
2417 		if (i < 32)
2418 			for (j = 0; j < TS_FPRWIDTH; j++)
2419 				val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
2420 		else {
2421 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
2422 				val->vval = vcpu->arch.vr_tm.vr[i-32];
2423 			else
2424 				r = -ENXIO;
2425 		}
2426 		break;
2427 	}
2428 	case KVM_REG_PPC_TM_CR:
2429 		*val = get_reg_val(id, vcpu->arch.cr_tm);
2430 		break;
2431 	case KVM_REG_PPC_TM_XER:
2432 		*val = get_reg_val(id, vcpu->arch.xer_tm);
2433 		break;
2434 	case KVM_REG_PPC_TM_LR:
2435 		*val = get_reg_val(id, vcpu->arch.lr_tm);
2436 		break;
2437 	case KVM_REG_PPC_TM_CTR:
2438 		*val = get_reg_val(id, vcpu->arch.ctr_tm);
2439 		break;
2440 	case KVM_REG_PPC_TM_FPSCR:
2441 		*val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
2442 		break;
2443 	case KVM_REG_PPC_TM_AMR:
2444 		*val = get_reg_val(id, vcpu->arch.amr_tm);
2445 		break;
2446 	case KVM_REG_PPC_TM_PPR:
2447 		*val = get_reg_val(id, vcpu->arch.ppr_tm);
2448 		break;
2449 	case KVM_REG_PPC_TM_VRSAVE:
2450 		*val = get_reg_val(id, vcpu->arch.vrsave_tm);
2451 		break;
2452 	case KVM_REG_PPC_TM_VSCR:
2453 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
2454 			*val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
2455 		else
2456 			r = -ENXIO;
2457 		break;
2458 	case KVM_REG_PPC_TM_DSCR:
2459 		*val = get_reg_val(id, vcpu->arch.dscr_tm);
2460 		break;
2461 	case KVM_REG_PPC_TM_TAR:
2462 		*val = get_reg_val(id, vcpu->arch.tar_tm);
2463 		break;
2464 #endif
2465 	case KVM_REG_PPC_ARCH_COMPAT:
2466 		*val = get_reg_val(id, kvmppc_get_arch_compat(vcpu));
2467 		break;
2468 	case KVM_REG_PPC_DEC_EXPIRY:
2469 		*val = get_reg_val(id, kvmppc_get_dec_expires(vcpu));
2470 		break;
2471 	case KVM_REG_PPC_ONLINE:
2472 		*val = get_reg_val(id, vcpu->arch.online);
2473 		break;
2474 	case KVM_REG_PPC_PTCR:
2475 		*val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
2476 		break;
2477 	case KVM_REG_PPC_FSCR:
2478 		*val = get_reg_val(id, kvmppc_get_fscr_hv(vcpu));
2479 		break;
2480 	default:
2481 		r = -EINVAL;
2482 		break;
2483 	}
2484 
2485 	return r;
2486 }
2487 
2488 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
2489 				 union kvmppc_one_reg *val)
2490 {
2491 	int r = 0;
2492 	long int i;
2493 	unsigned long addr, len;
2494 
2495 	switch (id) {
2496 	case KVM_REG_PPC_HIOR:
2497 		/* Only allow this to be set to zero */
2498 		if (set_reg_val(id, *val))
2499 			r = -EINVAL;
2500 		break;
2501 	case KVM_REG_PPC_DABR:
2502 		vcpu->arch.dabr = set_reg_val(id, *val);
2503 		break;
2504 	case KVM_REG_PPC_DABRX:
2505 		vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
2506 		break;
2507 	case KVM_REG_PPC_DSCR:
2508 		kvmppc_set_dscr_hv(vcpu, set_reg_val(id, *val));
2509 		break;
2510 	case KVM_REG_PPC_PURR:
2511 		kvmppc_set_purr_hv(vcpu, set_reg_val(id, *val));
2512 		break;
2513 	case KVM_REG_PPC_SPURR:
2514 		kvmppc_set_spurr_hv(vcpu, set_reg_val(id, *val));
2515 		break;
2516 	case KVM_REG_PPC_AMR:
2517 		kvmppc_set_amr_hv(vcpu, set_reg_val(id, *val));
2518 		break;
2519 	case KVM_REG_PPC_UAMOR:
2520 		kvmppc_set_uamor_hv(vcpu, set_reg_val(id, *val));
2521 		break;
2522 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
2523 		i = id - KVM_REG_PPC_MMCR0;
2524 		kvmppc_set_mmcr_hv(vcpu, i, set_reg_val(id, *val));
2525 		break;
2526 	case KVM_REG_PPC_MMCR2:
2527 		kvmppc_set_mmcr_hv(vcpu, 2, set_reg_val(id, *val));
2528 		break;
2529 	case KVM_REG_PPC_MMCRA:
2530 		kvmppc_set_mmcra_hv(vcpu, set_reg_val(id, *val));
2531 		break;
2532 	case KVM_REG_PPC_MMCRS:
2533 		vcpu->arch.mmcrs = set_reg_val(id, *val);
2534 		break;
2535 	case KVM_REG_PPC_MMCR3:
2536 		kvmppc_set_mmcr_hv(vcpu, 3, set_reg_val(id, *val));
2537 		break;
2538 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
2539 		i = id - KVM_REG_PPC_PMC1;
2540 		kvmppc_set_pmc_hv(vcpu, i, set_reg_val(id, *val));
2541 		break;
2542 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
2543 		i = id - KVM_REG_PPC_SPMC1;
2544 		vcpu->arch.spmc[i] = set_reg_val(id, *val);
2545 		break;
2546 	case KVM_REG_PPC_SIAR:
2547 		kvmppc_set_siar_hv(vcpu, set_reg_val(id, *val));
2548 		break;
2549 	case KVM_REG_PPC_SDAR:
2550 		kvmppc_set_sdar_hv(vcpu, set_reg_val(id, *val));
2551 		break;
2552 	case KVM_REG_PPC_SIER:
2553 		kvmppc_set_sier_hv(vcpu, 0, set_reg_val(id, *val));
2554 		break;
2555 	case KVM_REG_PPC_SIER2:
2556 		kvmppc_set_sier_hv(vcpu, 1, set_reg_val(id, *val));
2557 		break;
2558 	case KVM_REG_PPC_SIER3:
2559 		kvmppc_set_sier_hv(vcpu, 2, set_reg_val(id, *val));
2560 		break;
2561 	case KVM_REG_PPC_IAMR:
2562 		kvmppc_set_iamr_hv(vcpu, set_reg_val(id, *val));
2563 		break;
2564 	case KVM_REG_PPC_PSPB:
2565 		kvmppc_set_pspb_hv(vcpu, set_reg_val(id, *val));
2566 		break;
2567 	case KVM_REG_PPC_DPDES:
2568 		if (cpu_has_feature(CPU_FTR_ARCH_300))
2569 			vcpu->arch.doorbell_request = set_reg_val(id, *val) & 1;
2570 		else
2571 			vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
2572 		break;
2573 	case KVM_REG_PPC_VTB:
2574 		kvmppc_set_vtb(vcpu, set_reg_val(id, *val));
2575 		break;
2576 	case KVM_REG_PPC_DAWR:
2577 		kvmppc_set_dawr0_hv(vcpu, set_reg_val(id, *val));
2578 		break;
2579 	case KVM_REG_PPC_DAWRX:
2580 		kvmppc_set_dawrx0_hv(vcpu, set_reg_val(id, *val) & ~DAWRX_HYP);
2581 		break;
2582 	case KVM_REG_PPC_DAWR1:
2583 		kvmppc_set_dawr1_hv(vcpu, set_reg_val(id, *val));
2584 		break;
2585 	case KVM_REG_PPC_DAWRX1:
2586 		kvmppc_set_dawrx1_hv(vcpu, set_reg_val(id, *val) & ~DAWRX_HYP);
2587 		break;
2588 	case KVM_REG_PPC_DEXCR:
2589 		kvmppc_set_dexcr_hv(vcpu, set_reg_val(id, *val));
2590 		break;
2591 	case KVM_REG_PPC_HASHKEYR:
2592 		kvmppc_set_hashkeyr_hv(vcpu, set_reg_val(id, *val));
2593 		break;
2594 	case KVM_REG_PPC_HASHPKEYR:
2595 		kvmppc_set_hashpkeyr_hv(vcpu, set_reg_val(id, *val));
2596 		break;
2597 	case KVM_REG_PPC_CIABR:
2598 		kvmppc_set_ciabr_hv(vcpu, set_reg_val(id, *val));
2599 		/* Don't allow setting breakpoints in hypervisor code */
2600 		if ((kvmppc_get_ciabr_hv(vcpu) & CIABR_PRIV) == CIABR_PRIV_HYPER)
2601 			kvmppc_set_ciabr_hv(vcpu, kvmppc_get_ciabr_hv(vcpu) & ~CIABR_PRIV);
2602 		break;
2603 	case KVM_REG_PPC_CSIGR:
2604 		vcpu->arch.csigr = set_reg_val(id, *val);
2605 		break;
2606 	case KVM_REG_PPC_TACR:
2607 		vcpu->arch.tacr = set_reg_val(id, *val);
2608 		break;
2609 	case KVM_REG_PPC_TCSCR:
2610 		vcpu->arch.tcscr = set_reg_val(id, *val);
2611 		break;
2612 	case KVM_REG_PPC_PID:
2613 		kvmppc_set_pid(vcpu, set_reg_val(id, *val));
2614 		break;
2615 	case KVM_REG_PPC_ACOP:
2616 		vcpu->arch.acop = set_reg_val(id, *val);
2617 		break;
2618 	case KVM_REG_PPC_WORT:
2619 		kvmppc_set_wort_hv(vcpu, set_reg_val(id, *val));
2620 		break;
2621 	case KVM_REG_PPC_TIDR:
2622 		vcpu->arch.tid = set_reg_val(id, *val);
2623 		break;
2624 	case KVM_REG_PPC_PSSCR:
2625 		vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
2626 		break;
2627 	case KVM_REG_PPC_VPA_ADDR:
2628 		addr = set_reg_val(id, *val);
2629 		r = -EINVAL;
2630 		if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
2631 			      vcpu->arch.dtl.next_gpa))
2632 			break;
2633 		r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
2634 		break;
2635 	case KVM_REG_PPC_VPA_SLB:
2636 		addr = val->vpaval.addr;
2637 		len = val->vpaval.length;
2638 		r = -EINVAL;
2639 		if (addr && !vcpu->arch.vpa.next_gpa)
2640 			break;
2641 		r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
2642 		break;
2643 	case KVM_REG_PPC_VPA_DTL:
2644 		addr = val->vpaval.addr;
2645 		len = val->vpaval.length;
2646 		r = -EINVAL;
2647 		if (addr && (len < sizeof(struct dtl_entry) ||
2648 			     !vcpu->arch.vpa.next_gpa))
2649 			break;
2650 		len -= len % sizeof(struct dtl_entry);
2651 		r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
2652 		break;
2653 	case KVM_REG_PPC_TB_OFFSET:
2654 	{
2655 		/* round up to multiple of 2^24 */
2656 		u64 tb_offset = ALIGN(set_reg_val(id, *val), 1UL << 24);
2657 
2658 		/*
2659 		 * Now that we know the timebase offset, update the
2660 		 * decrementer expiry with a guest timebase value. If
2661 		 * the userspace does not set DEC_EXPIRY, this ensures
2662 		 * a migrated vcpu at least starts with an expired
2663 		 * decrementer, which is better than a large one that
2664 		 * causes a hang.
2665 		 */
2666 		kvmppc_set_tb_offset(vcpu, tb_offset);
2667 		if (!kvmppc_get_dec_expires(vcpu) && tb_offset)
2668 			kvmppc_set_dec_expires(vcpu, get_tb() + tb_offset);
2669 
2670 		kvmppc_set_tb_offset(vcpu, tb_offset);
2671 		break;
2672 	}
2673 	case KVM_REG_PPC_LPCR:
2674 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
2675 		break;
2676 	case KVM_REG_PPC_LPCR_64:
2677 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
2678 		break;
2679 	case KVM_REG_PPC_PPR:
2680 		kvmppc_set_ppr_hv(vcpu, set_reg_val(id, *val));
2681 		break;
2682 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2683 	case KVM_REG_PPC_TFHAR:
2684 		vcpu->arch.tfhar = set_reg_val(id, *val);
2685 		break;
2686 	case KVM_REG_PPC_TFIAR:
2687 		vcpu->arch.tfiar = set_reg_val(id, *val);
2688 		break;
2689 	case KVM_REG_PPC_TEXASR:
2690 		vcpu->arch.texasr = set_reg_val(id, *val);
2691 		break;
2692 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2693 		i = id - KVM_REG_PPC_TM_GPR0;
2694 		vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
2695 		break;
2696 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2697 	{
2698 		int j;
2699 		i = id - KVM_REG_PPC_TM_VSR0;
2700 		if (i < 32)
2701 			for (j = 0; j < TS_FPRWIDTH; j++)
2702 				vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
2703 		else
2704 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
2705 				vcpu->arch.vr_tm.vr[i-32] = val->vval;
2706 			else
2707 				r = -ENXIO;
2708 		break;
2709 	}
2710 	case KVM_REG_PPC_TM_CR:
2711 		vcpu->arch.cr_tm = set_reg_val(id, *val);
2712 		break;
2713 	case KVM_REG_PPC_TM_XER:
2714 		vcpu->arch.xer_tm = set_reg_val(id, *val);
2715 		break;
2716 	case KVM_REG_PPC_TM_LR:
2717 		vcpu->arch.lr_tm = set_reg_val(id, *val);
2718 		break;
2719 	case KVM_REG_PPC_TM_CTR:
2720 		vcpu->arch.ctr_tm = set_reg_val(id, *val);
2721 		break;
2722 	case KVM_REG_PPC_TM_FPSCR:
2723 		vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
2724 		break;
2725 	case KVM_REG_PPC_TM_AMR:
2726 		vcpu->arch.amr_tm = set_reg_val(id, *val);
2727 		break;
2728 	case KVM_REG_PPC_TM_PPR:
2729 		vcpu->arch.ppr_tm = set_reg_val(id, *val);
2730 		break;
2731 	case KVM_REG_PPC_TM_VRSAVE:
2732 		vcpu->arch.vrsave_tm = set_reg_val(id, *val);
2733 		break;
2734 	case KVM_REG_PPC_TM_VSCR:
2735 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
2736 			vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
2737 		else
2738 			r = - ENXIO;
2739 		break;
2740 	case KVM_REG_PPC_TM_DSCR:
2741 		vcpu->arch.dscr_tm = set_reg_val(id, *val);
2742 		break;
2743 	case KVM_REG_PPC_TM_TAR:
2744 		vcpu->arch.tar_tm = set_reg_val(id, *val);
2745 		break;
2746 #endif
2747 	case KVM_REG_PPC_ARCH_COMPAT:
2748 		r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
2749 		break;
2750 	case KVM_REG_PPC_DEC_EXPIRY:
2751 		kvmppc_set_dec_expires(vcpu, set_reg_val(id, *val));
2752 		break;
2753 	case KVM_REG_PPC_ONLINE:
2754 		i = set_reg_val(id, *val);
2755 		if (i && !vcpu->arch.online)
2756 			atomic_inc(&vcpu->arch.vcore->online_count);
2757 		else if (!i && vcpu->arch.online)
2758 			atomic_dec(&vcpu->arch.vcore->online_count);
2759 		vcpu->arch.online = i;
2760 		break;
2761 	case KVM_REG_PPC_PTCR:
2762 		vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
2763 		break;
2764 	case KVM_REG_PPC_FSCR:
2765 		kvmppc_set_fscr_hv(vcpu, set_reg_val(id, *val));
2766 		break;
2767 	default:
2768 		r = -EINVAL;
2769 		break;
2770 	}
2771 
2772 	return r;
2773 }
2774 
2775 /*
2776  * On POWER9, threads are independent and can be in different partitions.
2777  * Therefore we consider each thread to be a subcore.
2778  * There is a restriction that all threads have to be in the same
2779  * MMU mode (radix or HPT), unfortunately, but since we only support
2780  * HPT guests on a HPT host so far, that isn't an impediment yet.
2781  */
2782 static int threads_per_vcore(struct kvm *kvm)
2783 {
2784 	if (cpu_has_feature(CPU_FTR_ARCH_300))
2785 		return 1;
2786 	return threads_per_subcore;
2787 }
2788 
2789 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
2790 {
2791 	struct kvmppc_vcore *vcore;
2792 
2793 	vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
2794 
2795 	if (vcore == NULL)
2796 		return NULL;
2797 
2798 	spin_lock_init(&vcore->lock);
2799 	spin_lock_init(&vcore->stoltb_lock);
2800 	rcuwait_init(&vcore->wait);
2801 	vcore->preempt_tb = TB_NIL;
2802 	vcore->lpcr = kvm->arch.lpcr;
2803 	vcore->first_vcpuid = id;
2804 	vcore->kvm = kvm;
2805 	INIT_LIST_HEAD(&vcore->preempt_list);
2806 
2807 	return vcore;
2808 }
2809 
2810 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2811 static struct debugfs_timings_element {
2812 	const char *name;
2813 	size_t offset;
2814 } timings[] = {
2815 #ifdef CONFIG_KVM_BOOK3S_HV_P9_TIMING
2816 	{"vcpu_entry",	offsetof(struct kvm_vcpu, arch.vcpu_entry)},
2817 	{"guest_entry",	offsetof(struct kvm_vcpu, arch.guest_entry)},
2818 	{"in_guest",	offsetof(struct kvm_vcpu, arch.in_guest)},
2819 	{"guest_exit",	offsetof(struct kvm_vcpu, arch.guest_exit)},
2820 	{"vcpu_exit",	offsetof(struct kvm_vcpu, arch.vcpu_exit)},
2821 	{"hypercall",	offsetof(struct kvm_vcpu, arch.hcall)},
2822 	{"page_fault",	offsetof(struct kvm_vcpu, arch.pg_fault)},
2823 #else
2824 	{"rm_entry",	offsetof(struct kvm_vcpu, arch.rm_entry)},
2825 	{"rm_intr",	offsetof(struct kvm_vcpu, arch.rm_intr)},
2826 	{"rm_exit",	offsetof(struct kvm_vcpu, arch.rm_exit)},
2827 	{"guest",	offsetof(struct kvm_vcpu, arch.guest_time)},
2828 	{"cede",	offsetof(struct kvm_vcpu, arch.cede_time)},
2829 #endif
2830 };
2831 
2832 #define N_TIMINGS	(ARRAY_SIZE(timings))
2833 
2834 struct debugfs_timings_state {
2835 	struct kvm_vcpu	*vcpu;
2836 	unsigned int	buflen;
2837 	char		buf[N_TIMINGS * 100];
2838 };
2839 
2840 static int debugfs_timings_open(struct inode *inode, struct file *file)
2841 {
2842 	struct kvm_vcpu *vcpu = inode->i_private;
2843 	struct debugfs_timings_state *p;
2844 
2845 	p = kzalloc(sizeof(*p), GFP_KERNEL);
2846 	if (!p)
2847 		return -ENOMEM;
2848 
2849 	kvm_get_kvm(vcpu->kvm);
2850 	p->vcpu = vcpu;
2851 	file->private_data = p;
2852 
2853 	return nonseekable_open(inode, file);
2854 }
2855 
2856 static int debugfs_timings_release(struct inode *inode, struct file *file)
2857 {
2858 	struct debugfs_timings_state *p = file->private_data;
2859 
2860 	kvm_put_kvm(p->vcpu->kvm);
2861 	kfree(p);
2862 	return 0;
2863 }
2864 
2865 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2866 				    size_t len, loff_t *ppos)
2867 {
2868 	struct debugfs_timings_state *p = file->private_data;
2869 	struct kvm_vcpu *vcpu = p->vcpu;
2870 	char *s, *buf_end;
2871 	struct kvmhv_tb_accumulator tb;
2872 	u64 count;
2873 	loff_t pos;
2874 	ssize_t n;
2875 	int i, loops;
2876 	bool ok;
2877 
2878 	if (!p->buflen) {
2879 		s = p->buf;
2880 		buf_end = s + sizeof(p->buf);
2881 		for (i = 0; i < N_TIMINGS; ++i) {
2882 			struct kvmhv_tb_accumulator *acc;
2883 
2884 			acc = (struct kvmhv_tb_accumulator *)
2885 				((unsigned long)vcpu + timings[i].offset);
2886 			ok = false;
2887 			for (loops = 0; loops < 1000; ++loops) {
2888 				count = acc->seqcount;
2889 				if (!(count & 1)) {
2890 					smp_rmb();
2891 					tb = *acc;
2892 					smp_rmb();
2893 					if (count == acc->seqcount) {
2894 						ok = true;
2895 						break;
2896 					}
2897 				}
2898 				udelay(1);
2899 			}
2900 			if (!ok)
2901 				snprintf(s, buf_end - s, "%s: stuck\n",
2902 					timings[i].name);
2903 			else
2904 				snprintf(s, buf_end - s,
2905 					"%s: %llu %llu %llu %llu\n",
2906 					timings[i].name, count / 2,
2907 					tb_to_ns(tb.tb_total),
2908 					tb_to_ns(tb.tb_min),
2909 					tb_to_ns(tb.tb_max));
2910 			s += strlen(s);
2911 		}
2912 		p->buflen = s - p->buf;
2913 	}
2914 
2915 	pos = *ppos;
2916 	if (pos >= p->buflen)
2917 		return 0;
2918 	if (len > p->buflen - pos)
2919 		len = p->buflen - pos;
2920 	n = copy_to_user(buf, p->buf + pos, len);
2921 	if (n) {
2922 		if (n == len)
2923 			return -EFAULT;
2924 		len -= n;
2925 	}
2926 	*ppos = pos + len;
2927 	return len;
2928 }
2929 
2930 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2931 				     size_t len, loff_t *ppos)
2932 {
2933 	return -EACCES;
2934 }
2935 
2936 static const struct file_operations debugfs_timings_ops = {
2937 	.owner	 = THIS_MODULE,
2938 	.open	 = debugfs_timings_open,
2939 	.release = debugfs_timings_release,
2940 	.read	 = debugfs_timings_read,
2941 	.write	 = debugfs_timings_write,
2942 	.llseek	 = generic_file_llseek,
2943 };
2944 
2945 /* Create a debugfs directory for the vcpu */
2946 static int kvmppc_arch_create_vcpu_debugfs_hv(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry)
2947 {
2948 	if (cpu_has_feature(CPU_FTR_ARCH_300) == IS_ENABLED(CONFIG_KVM_BOOK3S_HV_P9_TIMING))
2949 		debugfs_create_file("timings", 0444, debugfs_dentry, vcpu,
2950 				    &debugfs_timings_ops);
2951 	return 0;
2952 }
2953 
2954 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2955 static int kvmppc_arch_create_vcpu_debugfs_hv(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry)
2956 {
2957 	return 0;
2958 }
2959 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2960 
2961 static int kvmppc_core_vcpu_create_hv(struct kvm_vcpu *vcpu)
2962 {
2963 	int err;
2964 	int core;
2965 	struct kvmppc_vcore *vcore;
2966 	struct kvm *kvm;
2967 	unsigned int id;
2968 
2969 	kvm = vcpu->kvm;
2970 	id = vcpu->vcpu_id;
2971 
2972 	vcpu->arch.shared = &vcpu->arch.shregs;
2973 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2974 	/*
2975 	 * The shared struct is never shared on HV,
2976 	 * so we can always use host endianness
2977 	 */
2978 #ifdef __BIG_ENDIAN__
2979 	vcpu->arch.shared_big_endian = true;
2980 #else
2981 	vcpu->arch.shared_big_endian = false;
2982 #endif
2983 #endif
2984 
2985 	if (kvmhv_is_nestedv2()) {
2986 		err = kvmhv_nestedv2_vcpu_create(vcpu, &vcpu->arch.nestedv2_io);
2987 		if (err < 0)
2988 			return err;
2989 	}
2990 
2991 	kvmppc_set_mmcr_hv(vcpu, 0, MMCR0_FC);
2992 	if (cpu_has_feature(CPU_FTR_ARCH_31)) {
2993 		kvmppc_set_mmcr_hv(vcpu, 0, kvmppc_get_mmcr_hv(vcpu, 0) | MMCR0_PMCCEXT);
2994 		kvmppc_set_mmcra_hv(vcpu, MMCRA_BHRB_DISABLE);
2995 	}
2996 
2997 	kvmppc_set_ctrl_hv(vcpu, CTRL_RUNLATCH);
2998 	/* default to host PVR, since we can't spoof it */
2999 	kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
3000 	spin_lock_init(&vcpu->arch.vpa_update_lock);
3001 	spin_lock_init(&vcpu->arch.tbacct_lock);
3002 	vcpu->arch.busy_preempt = TB_NIL;
3003 	__kvmppc_set_msr_hv(vcpu, MSR_ME);
3004 	vcpu->arch.intr_msr = MSR_SF | MSR_ME;
3005 
3006 	/*
3007 	 * Set the default HFSCR for the guest from the host value.
3008 	 * This value is only used on POWER9 and later.
3009 	 * On >= POWER9, we want to virtualize the doorbell facility, so we
3010 	 * don't set the HFSCR_MSGP bit, and that causes those instructions
3011 	 * to trap and then we emulate them.
3012 	 */
3013 	kvmppc_set_hfscr_hv(vcpu, HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
3014 			    HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP);
3015 
3016 	/* On POWER10 and later, allow prefixed instructions */
3017 	if (cpu_has_feature(CPU_FTR_ARCH_31))
3018 		kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) | HFSCR_PREFIX);
3019 
3020 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
3021 		kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) & mfspr(SPRN_HFSCR));
3022 
3023 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
3024 		if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3025 			kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) | HFSCR_TM);
3026 #endif
3027 	}
3028 	if (cpu_has_feature(CPU_FTR_TM_COMP))
3029 		vcpu->arch.hfscr |= HFSCR_TM;
3030 
3031 	vcpu->arch.hfscr_permitted = kvmppc_get_hfscr_hv(vcpu);
3032 
3033 	/*
3034 	 * PM, EBB, TM are demand-faulted so start with it clear.
3035 	 */
3036 	kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) & ~(HFSCR_PM | HFSCR_EBB | HFSCR_TM));
3037 
3038 	kvmppc_mmu_book3s_hv_init(vcpu);
3039 
3040 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
3041 
3042 	init_waitqueue_head(&vcpu->arch.cpu_run);
3043 
3044 	mutex_lock(&kvm->lock);
3045 	vcore = NULL;
3046 	err = -EINVAL;
3047 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
3048 		if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
3049 			pr_devel("KVM: VCPU ID too high\n");
3050 			core = KVM_MAX_VCORES;
3051 		} else {
3052 			BUG_ON(kvm->arch.smt_mode != 1);
3053 			core = kvmppc_pack_vcpu_id(kvm, id);
3054 		}
3055 	} else {
3056 		core = id / kvm->arch.smt_mode;
3057 	}
3058 	if (core < KVM_MAX_VCORES) {
3059 		vcore = kvm->arch.vcores[core];
3060 		if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
3061 			pr_devel("KVM: collision on id %u", id);
3062 			vcore = NULL;
3063 		} else if (!vcore) {
3064 			/*
3065 			 * Take mmu_setup_lock for mutual exclusion
3066 			 * with kvmppc_update_lpcr().
3067 			 */
3068 			err = -ENOMEM;
3069 			vcore = kvmppc_vcore_create(kvm,
3070 					id & ~(kvm->arch.smt_mode - 1));
3071 			mutex_lock(&kvm->arch.mmu_setup_lock);
3072 			kvm->arch.vcores[core] = vcore;
3073 			kvm->arch.online_vcores++;
3074 			mutex_unlock(&kvm->arch.mmu_setup_lock);
3075 		}
3076 	}
3077 	mutex_unlock(&kvm->lock);
3078 
3079 	if (!vcore)
3080 		return err;
3081 
3082 	spin_lock(&vcore->lock);
3083 	++vcore->num_threads;
3084 	spin_unlock(&vcore->lock);
3085 	vcpu->arch.vcore = vcore;
3086 	vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
3087 	vcpu->arch.thread_cpu = -1;
3088 	vcpu->arch.prev_cpu = -1;
3089 
3090 	vcpu->arch.cpu_type = KVM_CPU_3S_64;
3091 	kvmppc_sanity_check(vcpu);
3092 
3093 	return 0;
3094 }
3095 
3096 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
3097 			      unsigned long flags)
3098 {
3099 	int err;
3100 	int esmt = 0;
3101 
3102 	if (flags)
3103 		return -EINVAL;
3104 	if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
3105 		return -EINVAL;
3106 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
3107 		/*
3108 		 * On POWER8 (or POWER7), the threading mode is "strict",
3109 		 * so we pack smt_mode vcpus per vcore.
3110 		 */
3111 		if (smt_mode > threads_per_subcore)
3112 			return -EINVAL;
3113 	} else {
3114 		/*
3115 		 * On POWER9, the threading mode is "loose",
3116 		 * so each vcpu gets its own vcore.
3117 		 */
3118 		esmt = smt_mode;
3119 		smt_mode = 1;
3120 	}
3121 	mutex_lock(&kvm->lock);
3122 	err = -EBUSY;
3123 	if (!kvm->arch.online_vcores) {
3124 		kvm->arch.smt_mode = smt_mode;
3125 		kvm->arch.emul_smt_mode = esmt;
3126 		err = 0;
3127 	}
3128 	mutex_unlock(&kvm->lock);
3129 
3130 	return err;
3131 }
3132 
3133 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
3134 {
3135 	if (vpa->pinned_addr)
3136 		kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
3137 					vpa->dirty);
3138 }
3139 
3140 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
3141 {
3142 	spin_lock(&vcpu->arch.vpa_update_lock);
3143 	unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
3144 	unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
3145 	unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
3146 	spin_unlock(&vcpu->arch.vpa_update_lock);
3147 	if (kvmhv_is_nestedv2())
3148 		kvmhv_nestedv2_vcpu_free(vcpu, &vcpu->arch.nestedv2_io);
3149 }
3150 
3151 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
3152 {
3153 	/* Indicate we want to get back into the guest */
3154 	return 1;
3155 }
3156 
3157 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
3158 {
3159 	unsigned long dec_nsec, now;
3160 
3161 	now = get_tb();
3162 	if (now > kvmppc_dec_expires_host_tb(vcpu)) {
3163 		/* decrementer has already gone negative */
3164 		kvmppc_core_queue_dec(vcpu);
3165 		kvmppc_core_prepare_to_enter(vcpu);
3166 		return;
3167 	}
3168 	dec_nsec = tb_to_ns(kvmppc_dec_expires_host_tb(vcpu) - now);
3169 	hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
3170 	vcpu->arch.timer_running = 1;
3171 }
3172 
3173 extern int __kvmppc_vcore_entry(void);
3174 
3175 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
3176 				   struct kvm_vcpu *vcpu, u64 tb)
3177 {
3178 	u64 now;
3179 
3180 	if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3181 		return;
3182 	spin_lock_irq(&vcpu->arch.tbacct_lock);
3183 	now = tb;
3184 	vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
3185 		vcpu->arch.stolen_logged;
3186 	vcpu->arch.busy_preempt = now;
3187 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
3188 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
3189 	--vc->n_runnable;
3190 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
3191 }
3192 
3193 static int kvmppc_grab_hwthread(int cpu)
3194 {
3195 	struct paca_struct *tpaca;
3196 	long timeout = 10000;
3197 
3198 	tpaca = paca_ptrs[cpu];
3199 
3200 	/* Ensure the thread won't go into the kernel if it wakes */
3201 	tpaca->kvm_hstate.kvm_vcpu = NULL;
3202 	tpaca->kvm_hstate.kvm_vcore = NULL;
3203 	tpaca->kvm_hstate.napping = 0;
3204 	smp_wmb();
3205 	tpaca->kvm_hstate.hwthread_req = 1;
3206 
3207 	/*
3208 	 * If the thread is already executing in the kernel (e.g. handling
3209 	 * a stray interrupt), wait for it to get back to nap mode.
3210 	 * The smp_mb() is to ensure that our setting of hwthread_req
3211 	 * is visible before we look at hwthread_state, so if this
3212 	 * races with the code at system_reset_pSeries and the thread
3213 	 * misses our setting of hwthread_req, we are sure to see its
3214 	 * setting of hwthread_state, and vice versa.
3215 	 */
3216 	smp_mb();
3217 	while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
3218 		if (--timeout <= 0) {
3219 			pr_err("KVM: couldn't grab cpu %d\n", cpu);
3220 			return -EBUSY;
3221 		}
3222 		udelay(1);
3223 	}
3224 	return 0;
3225 }
3226 
3227 static void kvmppc_release_hwthread(int cpu)
3228 {
3229 	struct paca_struct *tpaca;
3230 
3231 	tpaca = paca_ptrs[cpu];
3232 	tpaca->kvm_hstate.hwthread_req = 0;
3233 	tpaca->kvm_hstate.kvm_vcpu = NULL;
3234 	tpaca->kvm_hstate.kvm_vcore = NULL;
3235 	tpaca->kvm_hstate.kvm_split_mode = NULL;
3236 }
3237 
3238 static DEFINE_PER_CPU(struct kvm *, cpu_in_guest);
3239 
3240 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
3241 {
3242 	struct kvm_nested_guest *nested = vcpu->arch.nested;
3243 	cpumask_t *need_tlb_flush;
3244 	int i;
3245 
3246 	if (nested)
3247 		need_tlb_flush = &nested->need_tlb_flush;
3248 	else
3249 		need_tlb_flush = &kvm->arch.need_tlb_flush;
3250 
3251 	cpu = cpu_first_tlb_thread_sibling(cpu);
3252 	for (i = cpu; i <= cpu_last_tlb_thread_sibling(cpu);
3253 					i += cpu_tlb_thread_sibling_step())
3254 		cpumask_set_cpu(i, need_tlb_flush);
3255 
3256 	/*
3257 	 * Make sure setting of bit in need_tlb_flush precedes testing of
3258 	 * cpu_in_guest. The matching barrier on the other side is hwsync
3259 	 * when switching to guest MMU mode, which happens between
3260 	 * cpu_in_guest being set to the guest kvm, and need_tlb_flush bit
3261 	 * being tested.
3262 	 */
3263 	smp_mb();
3264 
3265 	for (i = cpu; i <= cpu_last_tlb_thread_sibling(cpu);
3266 					i += cpu_tlb_thread_sibling_step()) {
3267 		struct kvm *running = *per_cpu_ptr(&cpu_in_guest, i);
3268 
3269 		if (running == kvm)
3270 			smp_call_function_single(i, do_nothing, NULL, 1);
3271 	}
3272 }
3273 
3274 static void do_migrate_away_vcpu(void *arg)
3275 {
3276 	struct kvm_vcpu *vcpu = arg;
3277 	struct kvm *kvm = vcpu->kvm;
3278 
3279 	/*
3280 	 * If the guest has GTSE, it may execute tlbie, so do a eieio; tlbsync;
3281 	 * ptesync sequence on the old CPU before migrating to a new one, in
3282 	 * case we interrupted the guest between a tlbie ; eieio ;
3283 	 * tlbsync; ptesync sequence.
3284 	 *
3285 	 * Otherwise, ptesync is sufficient for ordering tlbiel sequences.
3286 	 */
3287 	if (kvm->arch.lpcr & LPCR_GTSE)
3288 		asm volatile("eieio; tlbsync; ptesync");
3289 	else
3290 		asm volatile("ptesync");
3291 }
3292 
3293 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
3294 {
3295 	struct kvm_nested_guest *nested = vcpu->arch.nested;
3296 	struct kvm *kvm = vcpu->kvm;
3297 	int prev_cpu;
3298 
3299 	if (!cpu_has_feature(CPU_FTR_HVMODE))
3300 		return;
3301 
3302 	if (nested)
3303 		prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
3304 	else
3305 		prev_cpu = vcpu->arch.prev_cpu;
3306 
3307 	/*
3308 	 * With radix, the guest can do TLB invalidations itself,
3309 	 * and it could choose to use the local form (tlbiel) if
3310 	 * it is invalidating a translation that has only ever been
3311 	 * used on one vcpu.  However, that doesn't mean it has
3312 	 * only ever been used on one physical cpu, since vcpus
3313 	 * can move around between pcpus.  To cope with this, when
3314 	 * a vcpu moves from one pcpu to another, we need to tell
3315 	 * any vcpus running on the same core as this vcpu previously
3316 	 * ran to flush the TLB.
3317 	 */
3318 	if (prev_cpu != pcpu) {
3319 		if (prev_cpu >= 0) {
3320 			if (cpu_first_tlb_thread_sibling(prev_cpu) !=
3321 			    cpu_first_tlb_thread_sibling(pcpu))
3322 				radix_flush_cpu(kvm, prev_cpu, vcpu);
3323 
3324 			smp_call_function_single(prev_cpu,
3325 					do_migrate_away_vcpu, vcpu, 1);
3326 		}
3327 		if (nested)
3328 			nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
3329 		else
3330 			vcpu->arch.prev_cpu = pcpu;
3331 	}
3332 }
3333 
3334 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
3335 {
3336 	int cpu;
3337 	struct paca_struct *tpaca;
3338 
3339 	cpu = vc->pcpu;
3340 	if (vcpu) {
3341 		if (vcpu->arch.timer_running) {
3342 			hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
3343 			vcpu->arch.timer_running = 0;
3344 		}
3345 		cpu += vcpu->arch.ptid;
3346 		vcpu->cpu = vc->pcpu;
3347 		vcpu->arch.thread_cpu = cpu;
3348 	}
3349 	tpaca = paca_ptrs[cpu];
3350 	tpaca->kvm_hstate.kvm_vcpu = vcpu;
3351 	tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
3352 	tpaca->kvm_hstate.fake_suspend = 0;
3353 	/* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
3354 	smp_wmb();
3355 	tpaca->kvm_hstate.kvm_vcore = vc;
3356 	if (cpu != smp_processor_id())
3357 		kvmppc_ipi_thread(cpu);
3358 }
3359 
3360 static void kvmppc_wait_for_nap(int n_threads)
3361 {
3362 	int cpu = smp_processor_id();
3363 	int i, loops;
3364 
3365 	if (n_threads <= 1)
3366 		return;
3367 	for (loops = 0; loops < 1000000; ++loops) {
3368 		/*
3369 		 * Check if all threads are finished.
3370 		 * We set the vcore pointer when starting a thread
3371 		 * and the thread clears it when finished, so we look
3372 		 * for any threads that still have a non-NULL vcore ptr.
3373 		 */
3374 		for (i = 1; i < n_threads; ++i)
3375 			if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
3376 				break;
3377 		if (i == n_threads) {
3378 			HMT_medium();
3379 			return;
3380 		}
3381 		HMT_low();
3382 	}
3383 	HMT_medium();
3384 	for (i = 1; i < n_threads; ++i)
3385 		if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
3386 			pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
3387 }
3388 
3389 /*
3390  * Check that we are on thread 0 and that any other threads in
3391  * this core are off-line.  Then grab the threads so they can't
3392  * enter the kernel.
3393  */
3394 static int on_primary_thread(void)
3395 {
3396 	int cpu = smp_processor_id();
3397 	int thr;
3398 
3399 	/* Are we on a primary subcore? */
3400 	if (cpu_thread_in_subcore(cpu))
3401 		return 0;
3402 
3403 	thr = 0;
3404 	while (++thr < threads_per_subcore)
3405 		if (cpu_online(cpu + thr))
3406 			return 0;
3407 
3408 	/* Grab all hw threads so they can't go into the kernel */
3409 	for (thr = 1; thr < threads_per_subcore; ++thr) {
3410 		if (kvmppc_grab_hwthread(cpu + thr)) {
3411 			/* Couldn't grab one; let the others go */
3412 			do {
3413 				kvmppc_release_hwthread(cpu + thr);
3414 			} while (--thr > 0);
3415 			return 0;
3416 		}
3417 	}
3418 	return 1;
3419 }
3420 
3421 /*
3422  * A list of virtual cores for each physical CPU.
3423  * These are vcores that could run but their runner VCPU tasks are
3424  * (or may be) preempted.
3425  */
3426 struct preempted_vcore_list {
3427 	struct list_head	list;
3428 	spinlock_t		lock;
3429 };
3430 
3431 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
3432 
3433 static void init_vcore_lists(void)
3434 {
3435 	int cpu;
3436 
3437 	for_each_possible_cpu(cpu) {
3438 		struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
3439 		spin_lock_init(&lp->lock);
3440 		INIT_LIST_HEAD(&lp->list);
3441 	}
3442 }
3443 
3444 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
3445 {
3446 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
3447 
3448 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
3449 
3450 	vc->vcore_state = VCORE_PREEMPT;
3451 	vc->pcpu = smp_processor_id();
3452 	if (vc->num_threads < threads_per_vcore(vc->kvm)) {
3453 		spin_lock(&lp->lock);
3454 		list_add_tail(&vc->preempt_list, &lp->list);
3455 		spin_unlock(&lp->lock);
3456 	}
3457 
3458 	/* Start accumulating stolen time */
3459 	kvmppc_core_start_stolen(vc, mftb());
3460 }
3461 
3462 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
3463 {
3464 	struct preempted_vcore_list *lp;
3465 
3466 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
3467 
3468 	kvmppc_core_end_stolen(vc, mftb());
3469 	if (!list_empty(&vc->preempt_list)) {
3470 		lp = &per_cpu(preempted_vcores, vc->pcpu);
3471 		spin_lock(&lp->lock);
3472 		list_del_init(&vc->preempt_list);
3473 		spin_unlock(&lp->lock);
3474 	}
3475 	vc->vcore_state = VCORE_INACTIVE;
3476 }
3477 
3478 /*
3479  * This stores information about the virtual cores currently
3480  * assigned to a physical core.
3481  */
3482 struct core_info {
3483 	int		n_subcores;
3484 	int		max_subcore_threads;
3485 	int		total_threads;
3486 	int		subcore_threads[MAX_SUBCORES];
3487 	struct kvmppc_vcore *vc[MAX_SUBCORES];
3488 };
3489 
3490 /*
3491  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
3492  * respectively in 2-way micro-threading (split-core) mode on POWER8.
3493  */
3494 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
3495 
3496 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
3497 {
3498 	memset(cip, 0, sizeof(*cip));
3499 	cip->n_subcores = 1;
3500 	cip->max_subcore_threads = vc->num_threads;
3501 	cip->total_threads = vc->num_threads;
3502 	cip->subcore_threads[0] = vc->num_threads;
3503 	cip->vc[0] = vc;
3504 }
3505 
3506 static bool subcore_config_ok(int n_subcores, int n_threads)
3507 {
3508 	/*
3509 	 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
3510 	 * split-core mode, with one thread per subcore.
3511 	 */
3512 	if (cpu_has_feature(CPU_FTR_ARCH_300))
3513 		return n_subcores <= 4 && n_threads == 1;
3514 
3515 	/* On POWER8, can only dynamically split if unsplit to begin with */
3516 	if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
3517 		return false;
3518 	if (n_subcores > MAX_SUBCORES)
3519 		return false;
3520 	if (n_subcores > 1) {
3521 		if (!(dynamic_mt_modes & 2))
3522 			n_subcores = 4;
3523 		if (n_subcores > 2 && !(dynamic_mt_modes & 4))
3524 			return false;
3525 	}
3526 
3527 	return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
3528 }
3529 
3530 static void init_vcore_to_run(struct kvmppc_vcore *vc)
3531 {
3532 	vc->entry_exit_map = 0;
3533 	vc->in_guest = 0;
3534 	vc->napping_threads = 0;
3535 	vc->conferring_threads = 0;
3536 	vc->tb_offset_applied = 0;
3537 }
3538 
3539 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
3540 {
3541 	int n_threads = vc->num_threads;
3542 	int sub;
3543 
3544 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
3545 		return false;
3546 
3547 	/* In one_vm_per_core mode, require all vcores to be from the same vm */
3548 	if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
3549 		return false;
3550 
3551 	if (n_threads < cip->max_subcore_threads)
3552 		n_threads = cip->max_subcore_threads;
3553 	if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
3554 		return false;
3555 	cip->max_subcore_threads = n_threads;
3556 
3557 	sub = cip->n_subcores;
3558 	++cip->n_subcores;
3559 	cip->total_threads += vc->num_threads;
3560 	cip->subcore_threads[sub] = vc->num_threads;
3561 	cip->vc[sub] = vc;
3562 	init_vcore_to_run(vc);
3563 	list_del_init(&vc->preempt_list);
3564 
3565 	return true;
3566 }
3567 
3568 /*
3569  * Work out whether it is possible to piggyback the execution of
3570  * vcore *pvc onto the execution of the other vcores described in *cip.
3571  */
3572 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
3573 			  int target_threads)
3574 {
3575 	if (cip->total_threads + pvc->num_threads > target_threads)
3576 		return false;
3577 
3578 	return can_dynamic_split(pvc, cip);
3579 }
3580 
3581 static void prepare_threads(struct kvmppc_vcore *vc)
3582 {
3583 	int i;
3584 	struct kvm_vcpu *vcpu;
3585 
3586 	for_each_runnable_thread(i, vcpu, vc) {
3587 		if (signal_pending(vcpu->arch.run_task))
3588 			vcpu->arch.ret = -EINTR;
3589 		else if (vcpu->arch.vpa.update_pending ||
3590 			 vcpu->arch.slb_shadow.update_pending ||
3591 			 vcpu->arch.dtl.update_pending)
3592 			vcpu->arch.ret = RESUME_GUEST;
3593 		else
3594 			continue;
3595 		kvmppc_remove_runnable(vc, vcpu, mftb());
3596 		wake_up(&vcpu->arch.cpu_run);
3597 	}
3598 }
3599 
3600 static void collect_piggybacks(struct core_info *cip, int target_threads)
3601 {
3602 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
3603 	struct kvmppc_vcore *pvc, *vcnext;
3604 
3605 	spin_lock(&lp->lock);
3606 	list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
3607 		if (!spin_trylock(&pvc->lock))
3608 			continue;
3609 		prepare_threads(pvc);
3610 		if (!pvc->n_runnable || !pvc->kvm->arch.mmu_ready) {
3611 			list_del_init(&pvc->preempt_list);
3612 			if (pvc->runner == NULL) {
3613 				pvc->vcore_state = VCORE_INACTIVE;
3614 				kvmppc_core_end_stolen(pvc, mftb());
3615 			}
3616 			spin_unlock(&pvc->lock);
3617 			continue;
3618 		}
3619 		if (!can_piggyback(pvc, cip, target_threads)) {
3620 			spin_unlock(&pvc->lock);
3621 			continue;
3622 		}
3623 		kvmppc_core_end_stolen(pvc, mftb());
3624 		pvc->vcore_state = VCORE_PIGGYBACK;
3625 		if (cip->total_threads >= target_threads)
3626 			break;
3627 	}
3628 	spin_unlock(&lp->lock);
3629 }
3630 
3631 static bool recheck_signals_and_mmu(struct core_info *cip)
3632 {
3633 	int sub, i;
3634 	struct kvm_vcpu *vcpu;
3635 	struct kvmppc_vcore *vc;
3636 
3637 	for (sub = 0; sub < cip->n_subcores; ++sub) {
3638 		vc = cip->vc[sub];
3639 		if (!vc->kvm->arch.mmu_ready)
3640 			return true;
3641 		for_each_runnable_thread(i, vcpu, vc)
3642 			if (signal_pending(vcpu->arch.run_task))
3643 				return true;
3644 	}
3645 	return false;
3646 }
3647 
3648 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
3649 {
3650 	int still_running = 0, i;
3651 	u64 now;
3652 	long ret;
3653 	struct kvm_vcpu *vcpu;
3654 
3655 	spin_lock(&vc->lock);
3656 	now = get_tb();
3657 	for_each_runnable_thread(i, vcpu, vc) {
3658 		/*
3659 		 * It's safe to unlock the vcore in the loop here, because
3660 		 * for_each_runnable_thread() is safe against removal of
3661 		 * the vcpu, and the vcore state is VCORE_EXITING here,
3662 		 * so any vcpus becoming runnable will have their arch.trap
3663 		 * set to zero and can't actually run in the guest.
3664 		 */
3665 		spin_unlock(&vc->lock);
3666 		/* cancel pending dec exception if dec is positive */
3667 		if (now < kvmppc_dec_expires_host_tb(vcpu) &&
3668 		    kvmppc_core_pending_dec(vcpu))
3669 			kvmppc_core_dequeue_dec(vcpu);
3670 
3671 		trace_kvm_guest_exit(vcpu);
3672 
3673 		ret = RESUME_GUEST;
3674 		if (vcpu->arch.trap)
3675 			ret = kvmppc_handle_exit_hv(vcpu,
3676 						    vcpu->arch.run_task);
3677 
3678 		vcpu->arch.ret = ret;
3679 		vcpu->arch.trap = 0;
3680 
3681 		spin_lock(&vc->lock);
3682 		if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
3683 			if (vcpu->arch.pending_exceptions)
3684 				kvmppc_core_prepare_to_enter(vcpu);
3685 			if (vcpu->arch.ceded)
3686 				kvmppc_set_timer(vcpu);
3687 			else
3688 				++still_running;
3689 		} else {
3690 			kvmppc_remove_runnable(vc, vcpu, mftb());
3691 			wake_up(&vcpu->arch.cpu_run);
3692 		}
3693 	}
3694 	if (!is_master) {
3695 		if (still_running > 0) {
3696 			kvmppc_vcore_preempt(vc);
3697 		} else if (vc->runner) {
3698 			vc->vcore_state = VCORE_PREEMPT;
3699 			kvmppc_core_start_stolen(vc, mftb());
3700 		} else {
3701 			vc->vcore_state = VCORE_INACTIVE;
3702 		}
3703 		if (vc->n_runnable > 0 && vc->runner == NULL) {
3704 			/* make sure there's a candidate runner awake */
3705 			i = -1;
3706 			vcpu = next_runnable_thread(vc, &i);
3707 			wake_up(&vcpu->arch.cpu_run);
3708 		}
3709 	}
3710 	spin_unlock(&vc->lock);
3711 }
3712 
3713 /*
3714  * Clear core from the list of active host cores as we are about to
3715  * enter the guest. Only do this if it is the primary thread of the
3716  * core (not if a subcore) that is entering the guest.
3717  */
3718 static inline int kvmppc_clear_host_core(unsigned int cpu)
3719 {
3720 	int core;
3721 
3722 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3723 		return 0;
3724 	/*
3725 	 * Memory barrier can be omitted here as we will do a smp_wmb()
3726 	 * later in kvmppc_start_thread and we need ensure that state is
3727 	 * visible to other CPUs only after we enter guest.
3728 	 */
3729 	core = cpu >> threads_shift;
3730 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
3731 	return 0;
3732 }
3733 
3734 /*
3735  * Advertise this core as an active host core since we exited the guest
3736  * Only need to do this if it is the primary thread of the core that is
3737  * exiting.
3738  */
3739 static inline int kvmppc_set_host_core(unsigned int cpu)
3740 {
3741 	int core;
3742 
3743 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3744 		return 0;
3745 
3746 	/*
3747 	 * Memory barrier can be omitted here because we do a spin_unlock
3748 	 * immediately after this which provides the memory barrier.
3749 	 */
3750 	core = cpu >> threads_shift;
3751 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
3752 	return 0;
3753 }
3754 
3755 static void set_irq_happened(int trap)
3756 {
3757 	switch (trap) {
3758 	case BOOK3S_INTERRUPT_EXTERNAL:
3759 		local_paca->irq_happened |= PACA_IRQ_EE;
3760 		break;
3761 	case BOOK3S_INTERRUPT_H_DOORBELL:
3762 		local_paca->irq_happened |= PACA_IRQ_DBELL;
3763 		break;
3764 	case BOOK3S_INTERRUPT_HMI:
3765 		local_paca->irq_happened |= PACA_IRQ_HMI;
3766 		break;
3767 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
3768 		replay_system_reset();
3769 		break;
3770 	}
3771 }
3772 
3773 /*
3774  * Run a set of guest threads on a physical core.
3775  * Called with vc->lock held.
3776  */
3777 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
3778 {
3779 	struct kvm_vcpu *vcpu;
3780 	int i;
3781 	int srcu_idx;
3782 	struct core_info core_info;
3783 	struct kvmppc_vcore *pvc;
3784 	struct kvm_split_mode split_info, *sip;
3785 	int split, subcore_size, active;
3786 	int sub;
3787 	bool thr0_done;
3788 	unsigned long cmd_bit, stat_bit;
3789 	int pcpu, thr;
3790 	int target_threads;
3791 	int controlled_threads;
3792 	int trap;
3793 	bool is_power8;
3794 
3795 	if (WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300)))
3796 		return;
3797 
3798 	/*
3799 	 * Remove from the list any threads that have a signal pending
3800 	 * or need a VPA update done
3801 	 */
3802 	prepare_threads(vc);
3803 
3804 	/* if the runner is no longer runnable, let the caller pick a new one */
3805 	if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3806 		return;
3807 
3808 	/*
3809 	 * Initialize *vc.
3810 	 */
3811 	init_vcore_to_run(vc);
3812 	vc->preempt_tb = TB_NIL;
3813 
3814 	/*
3815 	 * Number of threads that we will be controlling: the same as
3816 	 * the number of threads per subcore, except on POWER9,
3817 	 * where it's 1 because the threads are (mostly) independent.
3818 	 */
3819 	controlled_threads = threads_per_vcore(vc->kvm);
3820 
3821 	/*
3822 	 * Make sure we are running on primary threads, and that secondary
3823 	 * threads are offline.  Also check if the number of threads in this
3824 	 * guest are greater than the current system threads per guest.
3825 	 */
3826 	if ((controlled_threads > 1) &&
3827 	    ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
3828 		for_each_runnable_thread(i, vcpu, vc) {
3829 			vcpu->arch.ret = -EBUSY;
3830 			kvmppc_remove_runnable(vc, vcpu, mftb());
3831 			wake_up(&vcpu->arch.cpu_run);
3832 		}
3833 		goto out;
3834 	}
3835 
3836 	/*
3837 	 * See if we could run any other vcores on the physical core
3838 	 * along with this one.
3839 	 */
3840 	init_core_info(&core_info, vc);
3841 	pcpu = smp_processor_id();
3842 	target_threads = controlled_threads;
3843 	if (target_smt_mode && target_smt_mode < target_threads)
3844 		target_threads = target_smt_mode;
3845 	if (vc->num_threads < target_threads)
3846 		collect_piggybacks(&core_info, target_threads);
3847 
3848 	/*
3849 	 * Hard-disable interrupts, and check resched flag and signals.
3850 	 * If we need to reschedule or deliver a signal, clean up
3851 	 * and return without going into the guest(s).
3852 	 * If the mmu_ready flag has been cleared, don't go into the
3853 	 * guest because that means a HPT resize operation is in progress.
3854 	 */
3855 	local_irq_disable();
3856 	hard_irq_disable();
3857 	if (lazy_irq_pending() || need_resched() ||
3858 	    recheck_signals_and_mmu(&core_info)) {
3859 		local_irq_enable();
3860 		vc->vcore_state = VCORE_INACTIVE;
3861 		/* Unlock all except the primary vcore */
3862 		for (sub = 1; sub < core_info.n_subcores; ++sub) {
3863 			pvc = core_info.vc[sub];
3864 			/* Put back on to the preempted vcores list */
3865 			kvmppc_vcore_preempt(pvc);
3866 			spin_unlock(&pvc->lock);
3867 		}
3868 		for (i = 0; i < controlled_threads; ++i)
3869 			kvmppc_release_hwthread(pcpu + i);
3870 		return;
3871 	}
3872 
3873 	kvmppc_clear_host_core(pcpu);
3874 
3875 	/* Decide on micro-threading (split-core) mode */
3876 	subcore_size = threads_per_subcore;
3877 	cmd_bit = stat_bit = 0;
3878 	split = core_info.n_subcores;
3879 	sip = NULL;
3880 	is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S);
3881 
3882 	if (split > 1) {
3883 		sip = &split_info;
3884 		memset(&split_info, 0, sizeof(split_info));
3885 		for (sub = 0; sub < core_info.n_subcores; ++sub)
3886 			split_info.vc[sub] = core_info.vc[sub];
3887 
3888 		if (is_power8) {
3889 			if (split == 2 && (dynamic_mt_modes & 2)) {
3890 				cmd_bit = HID0_POWER8_1TO2LPAR;
3891 				stat_bit = HID0_POWER8_2LPARMODE;
3892 			} else {
3893 				split = 4;
3894 				cmd_bit = HID0_POWER8_1TO4LPAR;
3895 				stat_bit = HID0_POWER8_4LPARMODE;
3896 			}
3897 			subcore_size = MAX_SMT_THREADS / split;
3898 			split_info.rpr = mfspr(SPRN_RPR);
3899 			split_info.pmmar = mfspr(SPRN_PMMAR);
3900 			split_info.ldbar = mfspr(SPRN_LDBAR);
3901 			split_info.subcore_size = subcore_size;
3902 		} else {
3903 			split_info.subcore_size = 1;
3904 		}
3905 
3906 		/* order writes to split_info before kvm_split_mode pointer */
3907 		smp_wmb();
3908 	}
3909 
3910 	for (thr = 0; thr < controlled_threads; ++thr) {
3911 		struct paca_struct *paca = paca_ptrs[pcpu + thr];
3912 
3913 		paca->kvm_hstate.napping = 0;
3914 		paca->kvm_hstate.kvm_split_mode = sip;
3915 	}
3916 
3917 	/* Initiate micro-threading (split-core) on POWER8 if required */
3918 	if (cmd_bit) {
3919 		unsigned long hid0 = mfspr(SPRN_HID0);
3920 
3921 		hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3922 		mb();
3923 		mtspr(SPRN_HID0, hid0);
3924 		isync();
3925 		for (;;) {
3926 			hid0 = mfspr(SPRN_HID0);
3927 			if (hid0 & stat_bit)
3928 				break;
3929 			cpu_relax();
3930 		}
3931 	}
3932 
3933 	/*
3934 	 * On POWER8, set RWMR register.
3935 	 * Since it only affects PURR and SPURR, it doesn't affect
3936 	 * the host, so we don't save/restore the host value.
3937 	 */
3938 	if (is_power8) {
3939 		unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3940 		int n_online = atomic_read(&vc->online_count);
3941 
3942 		/*
3943 		 * Use the 8-thread value if we're doing split-core
3944 		 * or if the vcore's online count looks bogus.
3945 		 */
3946 		if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3947 		    n_online >= 1 && n_online <= MAX_SMT_THREADS)
3948 			rwmr_val = p8_rwmr_values[n_online];
3949 		mtspr(SPRN_RWMR, rwmr_val);
3950 	}
3951 
3952 	/* Start all the threads */
3953 	active = 0;
3954 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
3955 		thr = is_power8 ? subcore_thread_map[sub] : sub;
3956 		thr0_done = false;
3957 		active |= 1 << thr;
3958 		pvc = core_info.vc[sub];
3959 		pvc->pcpu = pcpu + thr;
3960 		for_each_runnable_thread(i, vcpu, pvc) {
3961 			/*
3962 			 * XXX: is kvmppc_start_thread called too late here?
3963 			 * It updates vcpu->cpu and vcpu->arch.thread_cpu
3964 			 * which are used by kvmppc_fast_vcpu_kick_hv(), but
3965 			 * kick is called after new exceptions become available
3966 			 * and exceptions are checked earlier than here, by
3967 			 * kvmppc_core_prepare_to_enter.
3968 			 */
3969 			kvmppc_start_thread(vcpu, pvc);
3970 			kvmppc_update_vpa_dispatch(vcpu, pvc);
3971 			trace_kvm_guest_enter(vcpu);
3972 			if (!vcpu->arch.ptid)
3973 				thr0_done = true;
3974 			active |= 1 << (thr + vcpu->arch.ptid);
3975 		}
3976 		/*
3977 		 * We need to start the first thread of each subcore
3978 		 * even if it doesn't have a vcpu.
3979 		 */
3980 		if (!thr0_done)
3981 			kvmppc_start_thread(NULL, pvc);
3982 	}
3983 
3984 	/*
3985 	 * Ensure that split_info.do_nap is set after setting
3986 	 * the vcore pointer in the PACA of the secondaries.
3987 	 */
3988 	smp_mb();
3989 
3990 	/*
3991 	 * When doing micro-threading, poke the inactive threads as well.
3992 	 * This gets them to the nap instruction after kvm_do_nap,
3993 	 * which reduces the time taken to unsplit later.
3994 	 */
3995 	if (cmd_bit) {
3996 		split_info.do_nap = 1;	/* ask secondaries to nap when done */
3997 		for (thr = 1; thr < threads_per_subcore; ++thr)
3998 			if (!(active & (1 << thr)))
3999 				kvmppc_ipi_thread(pcpu + thr);
4000 	}
4001 
4002 	vc->vcore_state = VCORE_RUNNING;
4003 	preempt_disable();
4004 
4005 	trace_kvmppc_run_core(vc, 0);
4006 
4007 	for (sub = 0; sub < core_info.n_subcores; ++sub)
4008 		spin_unlock(&core_info.vc[sub]->lock);
4009 
4010 	guest_timing_enter_irqoff();
4011 
4012 	srcu_idx = srcu_read_lock(&vc->kvm->srcu);
4013 
4014 	guest_state_enter_irqoff();
4015 	this_cpu_disable_ftrace();
4016 
4017 	trap = __kvmppc_vcore_entry();
4018 
4019 	this_cpu_enable_ftrace();
4020 	guest_state_exit_irqoff();
4021 
4022 	srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
4023 
4024 	set_irq_happened(trap);
4025 
4026 	spin_lock(&vc->lock);
4027 	/* prevent other vcpu threads from doing kvmppc_start_thread() now */
4028 	vc->vcore_state = VCORE_EXITING;
4029 
4030 	/* wait for secondary threads to finish writing their state to memory */
4031 	kvmppc_wait_for_nap(controlled_threads);
4032 
4033 	/* Return to whole-core mode if we split the core earlier */
4034 	if (cmd_bit) {
4035 		unsigned long hid0 = mfspr(SPRN_HID0);
4036 
4037 		hid0 &= ~HID0_POWER8_DYNLPARDIS;
4038 		stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
4039 		mb();
4040 		mtspr(SPRN_HID0, hid0);
4041 		isync();
4042 		for (;;) {
4043 			hid0 = mfspr(SPRN_HID0);
4044 			if (!(hid0 & stat_bit))
4045 				break;
4046 			cpu_relax();
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 /* Helper functions for reading L2's stats from L1's VPA */
4112 #ifdef CONFIG_PPC_PSERIES
4113 static DEFINE_PER_CPU(u64, l1_to_l2_cs);
4114 static DEFINE_PER_CPU(u64, l2_to_l1_cs);
4115 static DEFINE_PER_CPU(u64, l2_runtime_agg);
4116 
4117 int kvmhv_get_l2_counters_status(void)
4118 {
4119 	return firmware_has_feature(FW_FEATURE_LPAR) &&
4120 		get_lppaca()->l2_counters_enable;
4121 }
4122 
4123 void kvmhv_set_l2_counters_status(int cpu, bool status)
4124 {
4125 	if (!firmware_has_feature(FW_FEATURE_LPAR))
4126 		return;
4127 	if (status)
4128 		lppaca_of(cpu).l2_counters_enable = 1;
4129 	else
4130 		lppaca_of(cpu).l2_counters_enable = 0;
4131 }
4132 EXPORT_SYMBOL(kvmhv_set_l2_counters_status);
4133 
4134 int kvmhv_counters_tracepoint_regfunc(void)
4135 {
4136 	int cpu;
4137 
4138 	for_each_present_cpu(cpu) {
4139 		kvmhv_set_l2_counters_status(cpu, true);
4140 	}
4141 	return 0;
4142 }
4143 
4144 void kvmhv_counters_tracepoint_unregfunc(void)
4145 {
4146 	int cpu;
4147 
4148 	for_each_present_cpu(cpu) {
4149 		kvmhv_set_l2_counters_status(cpu, false);
4150 	}
4151 }
4152 
4153 static void do_trace_nested_cs_time(struct kvm_vcpu *vcpu)
4154 {
4155 	struct lppaca *lp = get_lppaca();
4156 	u64 l1_to_l2_ns, l2_to_l1_ns, l2_runtime_ns;
4157 	u64 *l1_to_l2_cs_ptr = this_cpu_ptr(&l1_to_l2_cs);
4158 	u64 *l2_to_l1_cs_ptr = this_cpu_ptr(&l2_to_l1_cs);
4159 	u64 *l2_runtime_agg_ptr = this_cpu_ptr(&l2_runtime_agg);
4160 
4161 	l1_to_l2_ns = tb_to_ns(be64_to_cpu(lp->l1_to_l2_cs_tb));
4162 	l2_to_l1_ns = tb_to_ns(be64_to_cpu(lp->l2_to_l1_cs_tb));
4163 	l2_runtime_ns = tb_to_ns(be64_to_cpu(lp->l2_runtime_tb));
4164 	trace_kvmppc_vcpu_stats(vcpu, l1_to_l2_ns - *l1_to_l2_cs_ptr,
4165 					l2_to_l1_ns - *l2_to_l1_cs_ptr,
4166 					l2_runtime_ns - *l2_runtime_agg_ptr);
4167 	*l1_to_l2_cs_ptr = l1_to_l2_ns;
4168 	*l2_to_l1_cs_ptr = l2_to_l1_ns;
4169 	*l2_runtime_agg_ptr = l2_runtime_ns;
4170 	vcpu->arch.l1_to_l2_cs = l1_to_l2_ns;
4171 	vcpu->arch.l2_to_l1_cs = l2_to_l1_ns;
4172 	vcpu->arch.l2_runtime_agg = l2_runtime_ns;
4173 }
4174 
4175 u64 kvmhv_get_l1_to_l2_cs_time(void)
4176 {
4177 	return tb_to_ns(be64_to_cpu(get_lppaca()->l1_to_l2_cs_tb));
4178 }
4179 EXPORT_SYMBOL(kvmhv_get_l1_to_l2_cs_time);
4180 
4181 u64 kvmhv_get_l2_to_l1_cs_time(void)
4182 {
4183 	return tb_to_ns(be64_to_cpu(get_lppaca()->l2_to_l1_cs_tb));
4184 }
4185 EXPORT_SYMBOL(kvmhv_get_l2_to_l1_cs_time);
4186 
4187 u64 kvmhv_get_l2_runtime_agg(void)
4188 {
4189 	return tb_to_ns(be64_to_cpu(get_lppaca()->l2_runtime_tb));
4190 }
4191 EXPORT_SYMBOL(kvmhv_get_l2_runtime_agg);
4192 
4193 u64 kvmhv_get_l1_to_l2_cs_time_vcpu(void)
4194 {
4195 	struct kvm_vcpu *vcpu;
4196 	struct kvm_vcpu_arch *arch;
4197 
4198 	vcpu = local_paca->kvm_hstate.kvm_vcpu;
4199 	if (vcpu) {
4200 		arch = &vcpu->arch;
4201 		return arch->l1_to_l2_cs;
4202 	} else {
4203 		return 0;
4204 	}
4205 }
4206 EXPORT_SYMBOL(kvmhv_get_l1_to_l2_cs_time_vcpu);
4207 
4208 u64 kvmhv_get_l2_to_l1_cs_time_vcpu(void)
4209 {
4210 	struct kvm_vcpu *vcpu;
4211 	struct kvm_vcpu_arch *arch;
4212 
4213 	vcpu = local_paca->kvm_hstate.kvm_vcpu;
4214 	if (vcpu) {
4215 		arch = &vcpu->arch;
4216 		return arch->l2_to_l1_cs;
4217 	} else {
4218 		return 0;
4219 	}
4220 }
4221 EXPORT_SYMBOL(kvmhv_get_l2_to_l1_cs_time_vcpu);
4222 
4223 u64 kvmhv_get_l2_runtime_agg_vcpu(void)
4224 {
4225 	struct kvm_vcpu *vcpu;
4226 	struct kvm_vcpu_arch *arch;
4227 
4228 	vcpu = local_paca->kvm_hstate.kvm_vcpu;
4229 	if (vcpu) {
4230 		arch = &vcpu->arch;
4231 		return arch->l2_runtime_agg;
4232 	} else {
4233 		return 0;
4234 	}
4235 }
4236 EXPORT_SYMBOL(kvmhv_get_l2_runtime_agg_vcpu);
4237 
4238 #else
4239 int kvmhv_get_l2_counters_status(void)
4240 {
4241 	return 0;
4242 }
4243 
4244 static void do_trace_nested_cs_time(struct kvm_vcpu *vcpu)
4245 {
4246 }
4247 #endif
4248 
4249 static int kvmhv_vcpu_entry_nestedv2(struct kvm_vcpu *vcpu, u64 time_limit,
4250 				     unsigned long lpcr, u64 *tb)
4251 {
4252 	struct kvmhv_nestedv2_io *io;
4253 	unsigned long msr, i;
4254 	int trap;
4255 	long rc;
4256 
4257 	if (vcpu->arch.doorbell_request) {
4258 		vcpu->arch.doorbell_request = 0;
4259 		kvmppc_set_dpdes(vcpu, 1);
4260 	}
4261 
4262 	io = &vcpu->arch.nestedv2_io;
4263 
4264 	msr = mfmsr();
4265 	kvmppc_msr_hard_disable_set_facilities(vcpu, msr);
4266 	if (lazy_irq_pending())
4267 		return 0;
4268 
4269 	rc = kvmhv_nestedv2_flush_vcpu(vcpu, time_limit);
4270 	if (rc < 0)
4271 		return -EINVAL;
4272 
4273 	kvmppc_gse_put_u64(io->vcpu_run_input, KVMPPC_GSID_LPCR, lpcr);
4274 
4275 	accumulate_time(vcpu, &vcpu->arch.in_guest);
4276 	rc = plpar_guest_run_vcpu(0, vcpu->kvm->arch.lpid, vcpu->vcpu_id,
4277 				  &trap, &i);
4278 
4279 	if (rc != H_SUCCESS) {
4280 		pr_err("KVM Guest Run VCPU hcall failed\n");
4281 		if (rc == H_INVALID_ELEMENT_ID)
4282 			pr_err("KVM: Guest Run VCPU invalid element id at %ld\n", i);
4283 		else if (rc == H_INVALID_ELEMENT_SIZE)
4284 			pr_err("KVM: Guest Run VCPU invalid element size at %ld\n", i);
4285 		else if (rc == H_INVALID_ELEMENT_VALUE)
4286 			pr_err("KVM: Guest Run VCPU invalid element value at %ld\n", i);
4287 		return -EINVAL;
4288 	}
4289 	accumulate_time(vcpu, &vcpu->arch.guest_exit);
4290 
4291 	*tb = mftb();
4292 	kvmppc_gsm_reset(io->vcpu_message);
4293 	kvmppc_gsm_reset(io->vcore_message);
4294 	kvmppc_gsbm_zero(&io->valids);
4295 
4296 	rc = kvmhv_nestedv2_parse_output(vcpu);
4297 	if (rc < 0)
4298 		return -EINVAL;
4299 
4300 	timer_rearm_host_dec(*tb);
4301 
4302 	/* Record context switch and guest_run_time data */
4303 	if (kvmhv_get_l2_counters_status())
4304 		do_trace_nested_cs_time(vcpu);
4305 
4306 	return trap;
4307 }
4308 
4309 /* call our hypervisor to load up HV regs and go */
4310 static int kvmhv_vcpu_entry_p9_nested(struct kvm_vcpu *vcpu, u64 time_limit, unsigned long lpcr, u64 *tb)
4311 {
4312 	unsigned long host_psscr;
4313 	unsigned long msr;
4314 	struct hv_guest_state hvregs;
4315 	struct p9_host_os_sprs host_os_sprs;
4316 	s64 dec;
4317 	int trap;
4318 
4319 	msr = mfmsr();
4320 
4321 	save_p9_host_os_sprs(&host_os_sprs);
4322 
4323 	/*
4324 	 * We need to save and restore the guest visible part of the
4325 	 * psscr (i.e. using SPRN_PSSCR_PR) since the hypervisor
4326 	 * doesn't do this for us. Note only required if pseries since
4327 	 * this is done in kvmhv_vcpu_entry_p9() below otherwise.
4328 	 */
4329 	host_psscr = mfspr(SPRN_PSSCR_PR);
4330 
4331 	kvmppc_msr_hard_disable_set_facilities(vcpu, msr);
4332 	if (lazy_irq_pending())
4333 		return 0;
4334 
4335 	if (unlikely(load_vcpu_state(vcpu, &host_os_sprs)))
4336 		msr = mfmsr(); /* TM restore can update msr */
4337 
4338 	if (vcpu->arch.psscr != host_psscr)
4339 		mtspr(SPRN_PSSCR_PR, vcpu->arch.psscr);
4340 
4341 	kvmhv_save_hv_regs(vcpu, &hvregs);
4342 	hvregs.lpcr = lpcr;
4343 	hvregs.amor = ~0;
4344 	vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
4345 	hvregs.version = HV_GUEST_STATE_VERSION;
4346 	if (vcpu->arch.nested) {
4347 		hvregs.lpid = vcpu->arch.nested->shadow_lpid;
4348 		hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
4349 	} else {
4350 		hvregs.lpid = vcpu->kvm->arch.lpid;
4351 		hvregs.vcpu_token = vcpu->vcpu_id;
4352 	}
4353 	hvregs.hdec_expiry = time_limit;
4354 
4355 	/*
4356 	 * hvregs has the doorbell status, so zero it here which
4357 	 * enables us to receive doorbells when H_ENTER_NESTED is
4358 	 * in progress for this vCPU
4359 	 */
4360 
4361 	if (vcpu->arch.doorbell_request)
4362 		vcpu->arch.doorbell_request = 0;
4363 
4364 	/*
4365 	 * When setting DEC, we must always deal with irq_work_raise
4366 	 * via NMI vs setting DEC. The problem occurs right as we
4367 	 * switch into guest mode if a NMI hits and sets pending work
4368 	 * and sets DEC, then that will apply to the guest and not
4369 	 * bring us back to the host.
4370 	 *
4371 	 * irq_work_raise could check a flag (or possibly LPCR[HDICE]
4372 	 * for example) and set HDEC to 1? That wouldn't solve the
4373 	 * nested hv case which needs to abort the hcall or zero the
4374 	 * time limit.
4375 	 *
4376 	 * XXX: Another day's problem.
4377 	 */
4378 	mtspr(SPRN_DEC, kvmppc_dec_expires_host_tb(vcpu) - *tb);
4379 
4380 	mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
4381 	mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
4382 	switch_pmu_to_guest(vcpu, &host_os_sprs);
4383 	accumulate_time(vcpu, &vcpu->arch.in_guest);
4384 	trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
4385 				  __pa(&vcpu->arch.regs));
4386 	accumulate_time(vcpu, &vcpu->arch.guest_exit);
4387 	kvmhv_restore_hv_return_state(vcpu, &hvregs);
4388 	switch_pmu_to_host(vcpu, &host_os_sprs);
4389 	vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
4390 	vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
4391 	vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
4392 	vcpu->arch.psscr = mfspr(SPRN_PSSCR_PR);
4393 
4394 	store_vcpu_state(vcpu);
4395 
4396 	dec = mfspr(SPRN_DEC);
4397 	if (!(lpcr & LPCR_LD)) /* Sign extend if not using large decrementer */
4398 		dec = (s32) dec;
4399 	*tb = mftb();
4400 	vcpu->arch.dec_expires = dec + (*tb + kvmppc_get_tb_offset(vcpu));
4401 
4402 	timer_rearm_host_dec(*tb);
4403 
4404 	restore_p9_host_os_sprs(vcpu, &host_os_sprs);
4405 	if (vcpu->arch.psscr != host_psscr)
4406 		mtspr(SPRN_PSSCR_PR, host_psscr);
4407 
4408 	return trap;
4409 }
4410 
4411 /*
4412  * Guest entry for POWER9 and later CPUs.
4413  */
4414 static int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
4415 			 unsigned long lpcr, u64 *tb)
4416 {
4417 	struct kvm *kvm = vcpu->kvm;
4418 	struct kvm_nested_guest *nested = vcpu->arch.nested;
4419 	u64 next_timer;
4420 	int trap;
4421 
4422 	next_timer = timer_get_next_tb();
4423 	if (*tb >= next_timer)
4424 		return BOOK3S_INTERRUPT_HV_DECREMENTER;
4425 	if (next_timer < time_limit)
4426 		time_limit = next_timer;
4427 	else if (*tb >= time_limit) /* nested time limit */
4428 		return BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER;
4429 
4430 	vcpu->arch.ceded = 0;
4431 
4432 	vcpu_vpa_increment_dispatch(vcpu);
4433 
4434 	if (kvmhv_on_pseries()) {
4435 		if (kvmhv_is_nestedv1())
4436 			trap = kvmhv_vcpu_entry_p9_nested(vcpu, time_limit, lpcr, tb);
4437 		else
4438 			trap = kvmhv_vcpu_entry_nestedv2(vcpu, time_limit, lpcr, tb);
4439 
4440 		/* H_CEDE has to be handled now, not later */
4441 		if (trap == BOOK3S_INTERRUPT_SYSCALL && !nested &&
4442 		    kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
4443 			kvmppc_cede(vcpu);
4444 			kvmppc_set_gpr(vcpu, 3, 0);
4445 			trap = 0;
4446 		}
4447 
4448 	} else if (nested) {
4449 		__this_cpu_write(cpu_in_guest, kvm);
4450 		trap = kvmhv_vcpu_entry_p9(vcpu, time_limit, lpcr, tb);
4451 		__this_cpu_write(cpu_in_guest, NULL);
4452 
4453 	} else {
4454 		kvmppc_xive_push_vcpu(vcpu);
4455 
4456 		__this_cpu_write(cpu_in_guest, kvm);
4457 		trap = kvmhv_vcpu_entry_p9(vcpu, time_limit, lpcr, tb);
4458 		__this_cpu_write(cpu_in_guest, NULL);
4459 
4460 		if (trap == BOOK3S_INTERRUPT_SYSCALL &&
4461 		    !(__kvmppc_get_msr_hv(vcpu) & MSR_PR)) {
4462 			unsigned long req = kvmppc_get_gpr(vcpu, 3);
4463 
4464 			/*
4465 			 * XIVE rearm and XICS hcalls must be handled
4466 			 * before xive context is pulled (is this
4467 			 * true?)
4468 			 */
4469 			if (req == H_CEDE) {
4470 				/* H_CEDE has to be handled now */
4471 				kvmppc_cede(vcpu);
4472 				if (!kvmppc_xive_rearm_escalation(vcpu)) {
4473 					/*
4474 					 * Pending escalation so abort
4475 					 * the cede.
4476 					 */
4477 					vcpu->arch.ceded = 0;
4478 				}
4479 				kvmppc_set_gpr(vcpu, 3, 0);
4480 				trap = 0;
4481 
4482 			} else if (req == H_ENTER_NESTED) {
4483 				/*
4484 				 * L2 should not run with the L1
4485 				 * context so rearm and pull it.
4486 				 */
4487 				if (!kvmppc_xive_rearm_escalation(vcpu)) {
4488 					/*
4489 					 * Pending escalation so abort
4490 					 * H_ENTER_NESTED.
4491 					 */
4492 					kvmppc_set_gpr(vcpu, 3, 0);
4493 					trap = 0;
4494 				}
4495 
4496 			} else if (hcall_is_xics(req)) {
4497 				int ret;
4498 
4499 				ret = kvmppc_xive_xics_hcall(vcpu, req);
4500 				if (ret != H_TOO_HARD) {
4501 					kvmppc_set_gpr(vcpu, 3, ret);
4502 					trap = 0;
4503 				}
4504 			}
4505 		}
4506 		kvmppc_xive_pull_vcpu(vcpu);
4507 
4508 		if (kvm_is_radix(kvm))
4509 			vcpu->arch.slb_max = 0;
4510 	}
4511 
4512 	vcpu_vpa_increment_dispatch(vcpu);
4513 
4514 	return trap;
4515 }
4516 
4517 /*
4518  * Wait for some other vcpu thread to execute us, and
4519  * wake us up when we need to handle something in the host.
4520  */
4521 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
4522 				 struct kvm_vcpu *vcpu, int wait_state)
4523 {
4524 	DEFINE_WAIT(wait);
4525 
4526 	prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
4527 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4528 		spin_unlock(&vc->lock);
4529 		schedule();
4530 		spin_lock(&vc->lock);
4531 	}
4532 	finish_wait(&vcpu->arch.cpu_run, &wait);
4533 }
4534 
4535 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
4536 {
4537 	if (!halt_poll_ns_grow)
4538 		return;
4539 
4540 	vc->halt_poll_ns *= halt_poll_ns_grow;
4541 	if (vc->halt_poll_ns < halt_poll_ns_grow_start)
4542 		vc->halt_poll_ns = halt_poll_ns_grow_start;
4543 }
4544 
4545 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
4546 {
4547 	if (halt_poll_ns_shrink == 0)
4548 		vc->halt_poll_ns = 0;
4549 	else
4550 		vc->halt_poll_ns /= halt_poll_ns_shrink;
4551 }
4552 
4553 #ifdef CONFIG_KVM_XICS
4554 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
4555 {
4556 	if (!xics_on_xive())
4557 		return false;
4558 	return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
4559 		vcpu->arch.xive_saved_state.cppr;
4560 }
4561 #else
4562 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
4563 {
4564 	return false;
4565 }
4566 #endif /* CONFIG_KVM_XICS */
4567 
4568 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
4569 {
4570 	if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
4571 	    kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
4572 		return true;
4573 
4574 	return false;
4575 }
4576 
4577 static bool kvmppc_vcpu_check_block(struct kvm_vcpu *vcpu)
4578 {
4579 	if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
4580 		return true;
4581 	return false;
4582 }
4583 
4584 /*
4585  * Check to see if any of the runnable vcpus on the vcore have pending
4586  * exceptions or are no longer ceded
4587  */
4588 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
4589 {
4590 	struct kvm_vcpu *vcpu;
4591 	int i;
4592 
4593 	for_each_runnable_thread(i, vcpu, vc) {
4594 		if (kvmppc_vcpu_check_block(vcpu))
4595 			return 1;
4596 	}
4597 
4598 	return 0;
4599 }
4600 
4601 /*
4602  * All the vcpus in this vcore are idle, so wait for a decrementer
4603  * or external interrupt to one of the vcpus.  vc->lock is held.
4604  */
4605 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
4606 {
4607 	ktime_t cur, start_poll, start_wait;
4608 	int do_sleep = 1;
4609 	u64 block_ns;
4610 
4611 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
4612 
4613 	/* Poll for pending exceptions and ceded state */
4614 	cur = start_poll = ktime_get();
4615 	if (vc->halt_poll_ns) {
4616 		ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
4617 		++vc->runner->stat.generic.halt_attempted_poll;
4618 
4619 		vc->vcore_state = VCORE_POLLING;
4620 		spin_unlock(&vc->lock);
4621 
4622 		do {
4623 			if (kvmppc_vcore_check_block(vc)) {
4624 				do_sleep = 0;
4625 				break;
4626 			}
4627 			cur = ktime_get();
4628 		} while (kvm_vcpu_can_poll(cur, stop));
4629 
4630 		spin_lock(&vc->lock);
4631 		vc->vcore_state = VCORE_INACTIVE;
4632 
4633 		if (!do_sleep) {
4634 			++vc->runner->stat.generic.halt_successful_poll;
4635 			goto out;
4636 		}
4637 	}
4638 
4639 	prepare_to_rcuwait(&vc->wait);
4640 	set_current_state(TASK_INTERRUPTIBLE);
4641 	if (kvmppc_vcore_check_block(vc)) {
4642 		finish_rcuwait(&vc->wait);
4643 		do_sleep = 0;
4644 		/* If we polled, count this as a successful poll */
4645 		if (vc->halt_poll_ns)
4646 			++vc->runner->stat.generic.halt_successful_poll;
4647 		goto out;
4648 	}
4649 
4650 	start_wait = ktime_get();
4651 
4652 	vc->vcore_state = VCORE_SLEEPING;
4653 	trace_kvmppc_vcore_blocked(vc->runner, 0);
4654 	spin_unlock(&vc->lock);
4655 	schedule();
4656 	finish_rcuwait(&vc->wait);
4657 	spin_lock(&vc->lock);
4658 	vc->vcore_state = VCORE_INACTIVE;
4659 	trace_kvmppc_vcore_blocked(vc->runner, 1);
4660 	++vc->runner->stat.halt_successful_wait;
4661 
4662 	cur = ktime_get();
4663 
4664 out:
4665 	block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
4666 
4667 	/* Attribute wait time */
4668 	if (do_sleep) {
4669 		vc->runner->stat.generic.halt_wait_ns +=
4670 			ktime_to_ns(cur) - ktime_to_ns(start_wait);
4671 		KVM_STATS_LOG_HIST_UPDATE(
4672 				vc->runner->stat.generic.halt_wait_hist,
4673 				ktime_to_ns(cur) - ktime_to_ns(start_wait));
4674 		/* Attribute failed poll time */
4675 		if (vc->halt_poll_ns) {
4676 			vc->runner->stat.generic.halt_poll_fail_ns +=
4677 				ktime_to_ns(start_wait) -
4678 				ktime_to_ns(start_poll);
4679 			KVM_STATS_LOG_HIST_UPDATE(
4680 				vc->runner->stat.generic.halt_poll_fail_hist,
4681 				ktime_to_ns(start_wait) -
4682 				ktime_to_ns(start_poll));
4683 		}
4684 	} else {
4685 		/* Attribute successful poll time */
4686 		if (vc->halt_poll_ns) {
4687 			vc->runner->stat.generic.halt_poll_success_ns +=
4688 				ktime_to_ns(cur) -
4689 				ktime_to_ns(start_poll);
4690 			KVM_STATS_LOG_HIST_UPDATE(
4691 				vc->runner->stat.generic.halt_poll_success_hist,
4692 				ktime_to_ns(cur) - ktime_to_ns(start_poll));
4693 		}
4694 	}
4695 
4696 	/* Adjust poll time */
4697 	if (halt_poll_ns) {
4698 		if (block_ns <= vc->halt_poll_ns)
4699 			;
4700 		/* We slept and blocked for longer than the max halt time */
4701 		else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
4702 			shrink_halt_poll_ns(vc);
4703 		/* We slept and our poll time is too small */
4704 		else if (vc->halt_poll_ns < halt_poll_ns &&
4705 				block_ns < halt_poll_ns)
4706 			grow_halt_poll_ns(vc);
4707 		if (vc->halt_poll_ns > halt_poll_ns)
4708 			vc->halt_poll_ns = halt_poll_ns;
4709 	} else
4710 		vc->halt_poll_ns = 0;
4711 
4712 	trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
4713 }
4714 
4715 /*
4716  * This never fails for a radix guest, as none of the operations it does
4717  * for a radix guest can fail or have a way to report failure.
4718  */
4719 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
4720 {
4721 	int r = 0;
4722 	struct kvm *kvm = vcpu->kvm;
4723 
4724 	mutex_lock(&kvm->arch.mmu_setup_lock);
4725 	if (!kvm->arch.mmu_ready) {
4726 		if (!kvm_is_radix(kvm))
4727 			r = kvmppc_hv_setup_htab_rma(vcpu);
4728 		if (!r) {
4729 			if (cpu_has_feature(CPU_FTR_ARCH_300))
4730 				kvmppc_setup_partition_table(kvm);
4731 			kvm->arch.mmu_ready = 1;
4732 		}
4733 	}
4734 	mutex_unlock(&kvm->arch.mmu_setup_lock);
4735 	return r;
4736 }
4737 
4738 static int kvmppc_run_vcpu(struct kvm_vcpu *vcpu)
4739 {
4740 	struct kvm_run *run = vcpu->run;
4741 	int n_ceded, i, r;
4742 	struct kvmppc_vcore *vc;
4743 	struct kvm_vcpu *v;
4744 
4745 	trace_kvmppc_run_vcpu_enter(vcpu);
4746 
4747 	run->exit_reason = 0;
4748 	vcpu->arch.ret = RESUME_GUEST;
4749 	vcpu->arch.trap = 0;
4750 	kvmppc_update_vpas(vcpu);
4751 
4752 	/*
4753 	 * Synchronize with other threads in this virtual core
4754 	 */
4755 	vc = vcpu->arch.vcore;
4756 	spin_lock(&vc->lock);
4757 	vcpu->arch.ceded = 0;
4758 	vcpu->arch.run_task = current;
4759 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4760 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4761 	vcpu->arch.busy_preempt = TB_NIL;
4762 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
4763 	++vc->n_runnable;
4764 
4765 	/*
4766 	 * This happens the first time this is called for a vcpu.
4767 	 * If the vcore is already running, we may be able to start
4768 	 * this thread straight away and have it join in.
4769 	 */
4770 	if (!signal_pending(current)) {
4771 		if ((vc->vcore_state == VCORE_PIGGYBACK ||
4772 		     vc->vcore_state == VCORE_RUNNING) &&
4773 			   !VCORE_IS_EXITING(vc)) {
4774 			kvmppc_update_vpa_dispatch(vcpu, vc);
4775 			kvmppc_start_thread(vcpu, vc);
4776 			trace_kvm_guest_enter(vcpu);
4777 		} else if (vc->vcore_state == VCORE_SLEEPING) {
4778 		        rcuwait_wake_up(&vc->wait);
4779 		}
4780 
4781 	}
4782 
4783 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4784 	       !signal_pending(current)) {
4785 		/* See if the MMU is ready to go */
4786 		if (!vcpu->kvm->arch.mmu_ready) {
4787 			spin_unlock(&vc->lock);
4788 			r = kvmhv_setup_mmu(vcpu);
4789 			spin_lock(&vc->lock);
4790 			if (r) {
4791 				run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4792 				run->fail_entry.
4793 					hardware_entry_failure_reason = 0;
4794 				vcpu->arch.ret = r;
4795 				break;
4796 			}
4797 		}
4798 
4799 		if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4800 			kvmppc_vcore_end_preempt(vc);
4801 
4802 		if (vc->vcore_state != VCORE_INACTIVE) {
4803 			kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
4804 			continue;
4805 		}
4806 		for_each_runnable_thread(i, v, vc) {
4807 			kvmppc_core_prepare_to_enter(v);
4808 			if (signal_pending(v->arch.run_task)) {
4809 				kvmppc_remove_runnable(vc, v, mftb());
4810 				v->stat.signal_exits++;
4811 				v->run->exit_reason = KVM_EXIT_INTR;
4812 				v->arch.ret = -EINTR;
4813 				wake_up(&v->arch.cpu_run);
4814 			}
4815 		}
4816 		if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
4817 			break;
4818 		n_ceded = 0;
4819 		for_each_runnable_thread(i, v, vc) {
4820 			if (!kvmppc_vcpu_woken(v))
4821 				n_ceded += v->arch.ceded;
4822 			else
4823 				v->arch.ceded = 0;
4824 		}
4825 		vc->runner = vcpu;
4826 		if (n_ceded == vc->n_runnable) {
4827 			kvmppc_vcore_blocked(vc);
4828 		} else if (need_resched()) {
4829 			kvmppc_vcore_preempt(vc);
4830 			/* Let something else run */
4831 			cond_resched_lock(&vc->lock);
4832 			if (vc->vcore_state == VCORE_PREEMPT)
4833 				kvmppc_vcore_end_preempt(vc);
4834 		} else {
4835 			kvmppc_run_core(vc);
4836 		}
4837 		vc->runner = NULL;
4838 	}
4839 
4840 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4841 	       (vc->vcore_state == VCORE_RUNNING ||
4842 		vc->vcore_state == VCORE_EXITING ||
4843 		vc->vcore_state == VCORE_PIGGYBACK))
4844 		kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
4845 
4846 	if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4847 		kvmppc_vcore_end_preempt(vc);
4848 
4849 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4850 		kvmppc_remove_runnable(vc, vcpu, mftb());
4851 		vcpu->stat.signal_exits++;
4852 		run->exit_reason = KVM_EXIT_INTR;
4853 		vcpu->arch.ret = -EINTR;
4854 	}
4855 
4856 	if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
4857 		/* Wake up some vcpu to run the core */
4858 		i = -1;
4859 		v = next_runnable_thread(vc, &i);
4860 		wake_up(&v->arch.cpu_run);
4861 	}
4862 
4863 	trace_kvmppc_run_vcpu_exit(vcpu);
4864 	spin_unlock(&vc->lock);
4865 	return vcpu->arch.ret;
4866 }
4867 
4868 int kvmhv_run_single_vcpu(struct kvm_vcpu *vcpu, u64 time_limit,
4869 			  unsigned long lpcr)
4870 {
4871 	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
4872 	struct kvm_run *run = vcpu->run;
4873 	int trap, r, pcpu;
4874 	int srcu_idx;
4875 	struct kvmppc_vcore *vc;
4876 	struct kvm *kvm = vcpu->kvm;
4877 	struct kvm_nested_guest *nested = vcpu->arch.nested;
4878 	unsigned long flags;
4879 	u64 tb;
4880 
4881 	trace_kvmppc_run_vcpu_enter(vcpu);
4882 
4883 	run->exit_reason = 0;
4884 	vcpu->arch.ret = RESUME_GUEST;
4885 	vcpu->arch.trap = 0;
4886 
4887 	vc = vcpu->arch.vcore;
4888 	vcpu->arch.ceded = 0;
4889 	vcpu->arch.run_task = current;
4890 	vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
4891 
4892 	/* See if the MMU is ready to go */
4893 	if (unlikely(!kvm->arch.mmu_ready)) {
4894 		r = kvmhv_setup_mmu(vcpu);
4895 		if (r) {
4896 			run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4897 			run->fail_entry.hardware_entry_failure_reason = 0;
4898 			vcpu->arch.ret = r;
4899 			return r;
4900 		}
4901 	}
4902 
4903 	if (need_resched())
4904 		cond_resched();
4905 
4906 	kvmppc_update_vpas(vcpu);
4907 
4908 	preempt_disable();
4909 	pcpu = smp_processor_id();
4910 	if (kvm_is_radix(kvm))
4911 		kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4912 
4913 	/* flags save not required, but irq_pmu has no disable/enable API */
4914 	powerpc_local_irq_pmu_save(flags);
4915 
4916 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4917 
4918 	if (signal_pending(current))
4919 		goto sigpend;
4920 	if (need_resched() || !kvm->arch.mmu_ready)
4921 		goto out;
4922 
4923 	vcpu->cpu = pcpu;
4924 	vcpu->arch.thread_cpu = pcpu;
4925 	vc->pcpu = pcpu;
4926 	local_paca->kvm_hstate.kvm_vcpu = vcpu;
4927 	local_paca->kvm_hstate.ptid = 0;
4928 	local_paca->kvm_hstate.fake_suspend = 0;
4929 
4930 	/*
4931 	 * Orders set cpu/thread_cpu vs testing for pending interrupts and
4932 	 * doorbells below. The other side is when these fields are set vs
4933 	 * kvmppc_fast_vcpu_kick_hv reading the cpu/thread_cpu fields to
4934 	 * kick a vCPU to notice the pending interrupt.
4935 	 */
4936 	smp_mb();
4937 
4938 	if (!nested) {
4939 		kvmppc_core_prepare_to_enter(vcpu);
4940 		if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4941 			     &vcpu->arch.pending_exceptions) ||
4942 		    xive_interrupt_pending(vcpu)) {
4943 			/*
4944 			 * For nested HV, don't synthesize but always pass MER,
4945 			 * the L0 will be able to optimise that more
4946 			 * effectively than manipulating registers directly.
4947 			 */
4948 			if (!kvmhv_on_pseries() && (__kvmppc_get_msr_hv(vcpu) & MSR_EE))
4949 				kvmppc_inject_interrupt_hv(vcpu,
4950 							   BOOK3S_INTERRUPT_EXTERNAL, 0);
4951 			else
4952 				lpcr |= LPCR_MER;
4953 		} else {
4954 			/*
4955 			 * L1's copy of L2's LPCR (vcpu->arch.vcore->lpcr) can get its MER bit
4956 			 * unexpectedly set - for e.g. during NMI handling when all register
4957 			 * states are synchronized from L0 to L1. L1 needs to inform L0 about
4958 			 * MER=1 only when there are pending external interrupts.
4959 			 * In the above if check, MER bit is set if there are pending
4960 			 * external interrupts. Hence, explicity mask off MER bit
4961 			 * here as otherwise it may generate spurious interrupts in L2 KVM
4962 			 * causing an endless loop, which results in L2 guest getting hung.
4963 			 */
4964 			lpcr &= ~LPCR_MER;
4965 		}
4966 	} else if (vcpu->arch.pending_exceptions ||
4967 		   xive_interrupt_pending(vcpu)) {
4968 		vcpu->arch.ret = RESUME_HOST;
4969 		goto out;
4970 	}
4971 
4972 	if (vcpu->arch.timer_running) {
4973 		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
4974 		vcpu->arch.timer_running = 0;
4975 	}
4976 
4977 	tb = mftb();
4978 
4979 	kvmppc_update_vpa_dispatch_p9(vcpu, vc, tb + kvmppc_get_tb_offset(vcpu));
4980 
4981 	trace_kvm_guest_enter(vcpu);
4982 
4983 	guest_timing_enter_irqoff();
4984 
4985 	srcu_idx = srcu_read_lock(&kvm->srcu);
4986 
4987 	guest_state_enter_irqoff();
4988 	this_cpu_disable_ftrace();
4989 
4990 	trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr, &tb);
4991 	vcpu->arch.trap = trap;
4992 
4993 	this_cpu_enable_ftrace();
4994 	guest_state_exit_irqoff();
4995 
4996 	srcu_read_unlock(&kvm->srcu, srcu_idx);
4997 
4998 	set_irq_happened(trap);
4999 
5000 	vcpu->cpu = -1;
5001 	vcpu->arch.thread_cpu = -1;
5002 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
5003 
5004 	if (!vtime_accounting_enabled_this_cpu()) {
5005 		powerpc_local_irq_pmu_restore(flags);
5006 		/*
5007 		 * Service IRQs here before guest_timing_exit_irqoff() so any
5008 		 * ticks that occurred while running the guest are accounted to
5009 		 * the guest. If vtime accounting is enabled, accounting uses
5010 		 * TB rather than ticks, so it can be done without enabling
5011 		 * interrupts here, which has the problem that it accounts
5012 		 * interrupt processing overhead to the host.
5013 		 */
5014 		powerpc_local_irq_pmu_save(flags);
5015 	}
5016 	guest_timing_exit_irqoff();
5017 
5018 	powerpc_local_irq_pmu_restore(flags);
5019 
5020 	preempt_enable();
5021 
5022 	/*
5023 	 * cancel pending decrementer exception if DEC is now positive, or if
5024 	 * entering a nested guest in which case the decrementer is now owned
5025 	 * by L2 and the L1 decrementer is provided in hdec_expires
5026 	 */
5027 	if (kvmppc_core_pending_dec(vcpu) &&
5028 			((tb < kvmppc_dec_expires_host_tb(vcpu)) ||
5029 			 (trap == BOOK3S_INTERRUPT_SYSCALL &&
5030 			  kvmppc_get_gpr(vcpu, 3) == H_ENTER_NESTED)))
5031 		kvmppc_core_dequeue_dec(vcpu);
5032 
5033 	trace_kvm_guest_exit(vcpu);
5034 	r = RESUME_GUEST;
5035 	if (trap) {
5036 		if (!nested)
5037 			r = kvmppc_handle_exit_hv(vcpu, current);
5038 		else
5039 			r = kvmppc_handle_nested_exit(vcpu);
5040 	}
5041 	vcpu->arch.ret = r;
5042 
5043 	if (is_kvmppc_resume_guest(r) && !kvmppc_vcpu_check_block(vcpu)) {
5044 		kvmppc_set_timer(vcpu);
5045 
5046 		prepare_to_rcuwait(wait);
5047 		for (;;) {
5048 			set_current_state(TASK_INTERRUPTIBLE);
5049 			if (signal_pending(current)) {
5050 				vcpu->stat.signal_exits++;
5051 				run->exit_reason = KVM_EXIT_INTR;
5052 				vcpu->arch.ret = -EINTR;
5053 				break;
5054 			}
5055 
5056 			if (kvmppc_vcpu_check_block(vcpu))
5057 				break;
5058 
5059 			trace_kvmppc_vcore_blocked(vcpu, 0);
5060 			schedule();
5061 			trace_kvmppc_vcore_blocked(vcpu, 1);
5062 		}
5063 		finish_rcuwait(wait);
5064 	}
5065 	vcpu->arch.ceded = 0;
5066 
5067  done:
5068 	trace_kvmppc_run_vcpu_exit(vcpu);
5069 
5070 	return vcpu->arch.ret;
5071 
5072  sigpend:
5073 	vcpu->stat.signal_exits++;
5074 	run->exit_reason = KVM_EXIT_INTR;
5075 	vcpu->arch.ret = -EINTR;
5076  out:
5077 	vcpu->cpu = -1;
5078 	vcpu->arch.thread_cpu = -1;
5079 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
5080 	powerpc_local_irq_pmu_restore(flags);
5081 	preempt_enable();
5082 	goto done;
5083 }
5084 
5085 static int kvmppc_vcpu_run_hv(struct kvm_vcpu *vcpu)
5086 {
5087 	struct kvm_run *run = vcpu->run;
5088 	int r;
5089 	int srcu_idx;
5090 	struct kvm *kvm;
5091 	unsigned long msr;
5092 
5093 	start_timing(vcpu, &vcpu->arch.vcpu_entry);
5094 
5095 	if (!vcpu->arch.sane) {
5096 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5097 		return -EINVAL;
5098 	}
5099 
5100 	/* No need to go into the guest when all we'll do is come back out */
5101 	if (signal_pending(current)) {
5102 		run->exit_reason = KVM_EXIT_INTR;
5103 		return -EINTR;
5104 	}
5105 
5106 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
5107 	/*
5108 	 * Don't allow entry with a suspended transaction, because
5109 	 * the guest entry/exit code will lose it.
5110 	 */
5111 	if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
5112 	    (current->thread.regs->msr & MSR_TM)) {
5113 		if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
5114 			run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5115 			run->fail_entry.hardware_entry_failure_reason = 0;
5116 			return -EINVAL;
5117 		}
5118 	}
5119 #endif
5120 
5121 	/*
5122 	 * Force online to 1 for the sake of old userspace which doesn't
5123 	 * set it.
5124 	 */
5125 	if (!vcpu->arch.online) {
5126 		atomic_inc(&vcpu->arch.vcore->online_count);
5127 		vcpu->arch.online = 1;
5128 	}
5129 
5130 	kvmppc_core_prepare_to_enter(vcpu);
5131 
5132 	kvm = vcpu->kvm;
5133 	atomic_inc(&kvm->arch.vcpus_running);
5134 	/* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
5135 	smp_mb();
5136 
5137 	msr = 0;
5138 	if (IS_ENABLED(CONFIG_PPC_FPU))
5139 		msr |= MSR_FP;
5140 	if (cpu_has_feature(CPU_FTR_ALTIVEC))
5141 		msr |= MSR_VEC;
5142 	if (cpu_has_feature(CPU_FTR_VSX))
5143 		msr |= MSR_VSX;
5144 	if ((cpu_has_feature(CPU_FTR_TM) ||
5145 	    cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST)) &&
5146 			(kvmppc_get_hfscr_hv(vcpu) & HFSCR_TM))
5147 		msr |= MSR_TM;
5148 	msr = msr_check_and_set(msr);
5149 
5150 	kvmppc_save_user_regs();
5151 
5152 	kvmppc_save_current_sprs();
5153 
5154 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5155 		vcpu->arch.waitp = &vcpu->arch.vcore->wait;
5156 	vcpu->arch.pgdir = kvm->mm->pgd;
5157 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
5158 
5159 	do {
5160 		accumulate_time(vcpu, &vcpu->arch.guest_entry);
5161 		if (cpu_has_feature(CPU_FTR_ARCH_300))
5162 			r = kvmhv_run_single_vcpu(vcpu, ~(u64)0,
5163 						  vcpu->arch.vcore->lpcr);
5164 		else
5165 			r = kvmppc_run_vcpu(vcpu);
5166 
5167 		if (run->exit_reason == KVM_EXIT_PAPR_HCALL) {
5168 			accumulate_time(vcpu, &vcpu->arch.hcall);
5169 
5170 			if (!kvmhv_is_nestedv2() && WARN_ON_ONCE(__kvmppc_get_msr_hv(vcpu) & MSR_PR)) {
5171 				/*
5172 				 * These should have been caught reflected
5173 				 * into the guest by now. Final sanity check:
5174 				 * don't allow userspace to execute hcalls in
5175 				 * the hypervisor.
5176 				 */
5177 				r = RESUME_GUEST;
5178 				continue;
5179 			}
5180 			trace_kvm_hcall_enter(vcpu);
5181 			r = kvmppc_pseries_do_hcall(vcpu);
5182 			trace_kvm_hcall_exit(vcpu, r);
5183 			kvmppc_core_prepare_to_enter(vcpu);
5184 		} else if (r == RESUME_PAGE_FAULT) {
5185 			accumulate_time(vcpu, &vcpu->arch.pg_fault);
5186 			srcu_idx = srcu_read_lock(&kvm->srcu);
5187 			r = kvmppc_book3s_hv_page_fault(vcpu,
5188 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
5189 			srcu_read_unlock(&kvm->srcu, srcu_idx);
5190 		} else if (r == RESUME_PASSTHROUGH) {
5191 			if (WARN_ON(xics_on_xive()))
5192 				r = H_SUCCESS;
5193 			else
5194 				r = kvmppc_xics_rm_complete(vcpu, 0);
5195 		}
5196 	} while (is_kvmppc_resume_guest(r));
5197 	accumulate_time(vcpu, &vcpu->arch.vcpu_exit);
5198 
5199 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
5200 	atomic_dec(&kvm->arch.vcpus_running);
5201 
5202 	srr_regs_clobbered();
5203 
5204 	end_timing(vcpu);
5205 
5206 	return r;
5207 }
5208 
5209 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
5210 				     int shift, int sllp)
5211 {
5212 	(*sps)->page_shift = shift;
5213 	(*sps)->slb_enc = sllp;
5214 	(*sps)->enc[0].page_shift = shift;
5215 	(*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
5216 	/*
5217 	 * Add 16MB MPSS support (may get filtered out by userspace)
5218 	 */
5219 	if (shift != 24) {
5220 		int penc = kvmppc_pgsize_lp_encoding(shift, 24);
5221 		if (penc != -1) {
5222 			(*sps)->enc[1].page_shift = 24;
5223 			(*sps)->enc[1].pte_enc = penc;
5224 		}
5225 	}
5226 	(*sps)++;
5227 }
5228 
5229 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
5230 					 struct kvm_ppc_smmu_info *info)
5231 {
5232 	struct kvm_ppc_one_seg_page_size *sps;
5233 
5234 	/*
5235 	 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
5236 	 * POWER7 doesn't support keys for instruction accesses,
5237 	 * POWER8 and POWER9 do.
5238 	 */
5239 	info->data_keys = 32;
5240 	info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
5241 
5242 	/* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
5243 	info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
5244 	info->slb_size = 32;
5245 
5246 	/* We only support these sizes for now, and no muti-size segments */
5247 	sps = &info->sps[0];
5248 	kvmppc_add_seg_page_size(&sps, 12, 0);
5249 	kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
5250 	kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
5251 
5252 	/* If running as a nested hypervisor, we don't support HPT guests */
5253 	if (kvmhv_on_pseries())
5254 		info->flags |= KVM_PPC_NO_HASH;
5255 
5256 	return 0;
5257 }
5258 
5259 /*
5260  * Get (and clear) the dirty memory log for a memory slot.
5261  */
5262 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
5263 					 struct kvm_dirty_log *log)
5264 {
5265 	struct kvm_memslots *slots;
5266 	struct kvm_memory_slot *memslot;
5267 	int r;
5268 	unsigned long n, i;
5269 	unsigned long *buf, *p;
5270 	struct kvm_vcpu *vcpu;
5271 
5272 	mutex_lock(&kvm->slots_lock);
5273 
5274 	r = -EINVAL;
5275 	if (log->slot >= KVM_USER_MEM_SLOTS)
5276 		goto out;
5277 
5278 	slots = kvm_memslots(kvm);
5279 	memslot = id_to_memslot(slots, log->slot);
5280 	r = -ENOENT;
5281 	if (!memslot || !memslot->dirty_bitmap)
5282 		goto out;
5283 
5284 	/*
5285 	 * Use second half of bitmap area because both HPT and radix
5286 	 * accumulate bits in the first half.
5287 	 */
5288 	n = kvm_dirty_bitmap_bytes(memslot);
5289 	buf = memslot->dirty_bitmap + n / sizeof(long);
5290 	memset(buf, 0, n);
5291 
5292 	if (kvm_is_radix(kvm))
5293 		r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
5294 	else
5295 		r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
5296 	if (r)
5297 		goto out;
5298 
5299 	/*
5300 	 * We accumulate dirty bits in the first half of the
5301 	 * memslot's dirty_bitmap area, for when pages are paged
5302 	 * out or modified by the host directly.  Pick up these
5303 	 * bits and add them to the map.
5304 	 */
5305 	p = memslot->dirty_bitmap;
5306 	for (i = 0; i < n / sizeof(long); ++i)
5307 		buf[i] |= xchg(&p[i], 0);
5308 
5309 	/* Harvest dirty bits from VPA and DTL updates */
5310 	/* Note: we never modify the SLB shadow buffer areas */
5311 	kvm_for_each_vcpu(i, vcpu, kvm) {
5312 		spin_lock(&vcpu->arch.vpa_update_lock);
5313 		kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
5314 		kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
5315 		spin_unlock(&vcpu->arch.vpa_update_lock);
5316 	}
5317 
5318 	r = -EFAULT;
5319 	if (copy_to_user(log->dirty_bitmap, buf, n))
5320 		goto out;
5321 
5322 	r = 0;
5323 out:
5324 	mutex_unlock(&kvm->slots_lock);
5325 	return r;
5326 }
5327 
5328 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *slot)
5329 {
5330 	vfree(slot->arch.rmap);
5331 	slot->arch.rmap = NULL;
5332 }
5333 
5334 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
5335 				const struct kvm_memory_slot *old,
5336 				struct kvm_memory_slot *new,
5337 				enum kvm_mr_change change)
5338 {
5339 	if (change == KVM_MR_CREATE) {
5340 		unsigned long size = array_size(new->npages, sizeof(*new->arch.rmap));
5341 
5342 		if ((size >> PAGE_SHIFT) > totalram_pages())
5343 			return -ENOMEM;
5344 
5345 		new->arch.rmap = vzalloc(size);
5346 		if (!new->arch.rmap)
5347 			return -ENOMEM;
5348 	} else if (change != KVM_MR_DELETE) {
5349 		new->arch.rmap = old->arch.rmap;
5350 	}
5351 
5352 	return 0;
5353 }
5354 
5355 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
5356 				struct kvm_memory_slot *old,
5357 				const struct kvm_memory_slot *new,
5358 				enum kvm_mr_change change)
5359 {
5360 	/*
5361 	 * If we are creating or modifying a memslot, it might make
5362 	 * some address that was previously cached as emulated
5363 	 * MMIO be no longer emulated MMIO, so invalidate
5364 	 * all the caches of emulated MMIO translations.
5365 	 */
5366 	if (change != KVM_MR_DELETE)
5367 		atomic64_inc(&kvm->arch.mmio_update);
5368 
5369 	/*
5370 	 * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
5371 	 * have already called kvm_arch_flush_shadow_memslot() to
5372 	 * flush shadow mappings.  For KVM_MR_CREATE we have no
5373 	 * previous mappings.  So the only case to handle is
5374 	 * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
5375 	 * has been changed.
5376 	 * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
5377 	 * to get rid of any THP PTEs in the partition-scoped page tables
5378 	 * so we can track dirtiness at the page level; we flush when
5379 	 * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
5380 	 * using THP PTEs.
5381 	 */
5382 	if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
5383 	    ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
5384 		kvmppc_radix_flush_memslot(kvm, old);
5385 	/*
5386 	 * If UV hasn't yet called H_SVM_INIT_START, don't register memslots.
5387 	 */
5388 	if (!kvm->arch.secure_guest)
5389 		return;
5390 
5391 	switch (change) {
5392 	case KVM_MR_CREATE:
5393 		/*
5394 		 * @TODO kvmppc_uvmem_memslot_create() can fail and
5395 		 * return error. Fix this.
5396 		 */
5397 		kvmppc_uvmem_memslot_create(kvm, new);
5398 		break;
5399 	case KVM_MR_DELETE:
5400 		kvmppc_uvmem_memslot_delete(kvm, old);
5401 		break;
5402 	default:
5403 		/* TODO: Handle KVM_MR_MOVE */
5404 		break;
5405 	}
5406 }
5407 
5408 /*
5409  * Update LPCR values in kvm->arch and in vcores.
5410  * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion
5411  * of kvm->arch.lpcr update).
5412  */
5413 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
5414 {
5415 	long int i;
5416 	u32 cores_done = 0;
5417 
5418 	if ((kvm->arch.lpcr & mask) == lpcr)
5419 		return;
5420 
5421 	kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
5422 
5423 	for (i = 0; i < KVM_MAX_VCORES; ++i) {
5424 		struct kvmppc_vcore *vc = kvm->arch.vcores[i];
5425 		if (!vc)
5426 			continue;
5427 
5428 		spin_lock(&vc->lock);
5429 		vc->lpcr = (vc->lpcr & ~mask) | lpcr;
5430 		verify_lpcr(kvm, vc->lpcr);
5431 		spin_unlock(&vc->lock);
5432 		if (++cores_done >= kvm->arch.online_vcores)
5433 			break;
5434 	}
5435 
5436 	if (kvmhv_is_nestedv2()) {
5437 		struct kvm_vcpu *vcpu;
5438 
5439 		kvm_for_each_vcpu(i, vcpu, kvm) {
5440 			kvmhv_nestedv2_mark_dirty(vcpu, KVMPPC_GSID_LPCR);
5441 		}
5442 	}
5443 }
5444 
5445 void kvmppc_setup_partition_table(struct kvm *kvm)
5446 {
5447 	unsigned long dw0, dw1;
5448 
5449 	if (!kvm_is_radix(kvm)) {
5450 		/* PS field - page size for VRMA */
5451 		dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
5452 			((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
5453 		/* HTABSIZE and HTABORG fields */
5454 		dw0 |= kvm->arch.sdr1;
5455 
5456 		/* Second dword as set by userspace */
5457 		dw1 = kvm->arch.process_table;
5458 	} else {
5459 		dw0 = PATB_HR | radix__get_tree_size() |
5460 			__pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
5461 		dw1 = PATB_GR | kvm->arch.process_table;
5462 	}
5463 	kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
5464 }
5465 
5466 /*
5467  * Set up HPT (hashed page table) and RMA (real-mode area).
5468  * Must be called with kvm->arch.mmu_setup_lock held.
5469  */
5470 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
5471 {
5472 	int err = 0;
5473 	struct kvm *kvm = vcpu->kvm;
5474 	unsigned long hva;
5475 	struct kvm_memory_slot *memslot;
5476 	struct vm_area_struct *vma;
5477 	unsigned long lpcr = 0, senc;
5478 	unsigned long psize, porder;
5479 	int srcu_idx;
5480 
5481 	/* Allocate hashed page table (if not done already) and reset it */
5482 	if (!kvm->arch.hpt.virt) {
5483 		int order = KVM_DEFAULT_HPT_ORDER;
5484 		struct kvm_hpt_info info;
5485 
5486 		err = kvmppc_allocate_hpt(&info, order);
5487 		/* If we get here, it means userspace didn't specify a
5488 		 * size explicitly.  So, try successively smaller
5489 		 * sizes if the default failed. */
5490 		while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
5491 			err  = kvmppc_allocate_hpt(&info, order);
5492 
5493 		if (err < 0) {
5494 			pr_err("KVM: Couldn't alloc HPT\n");
5495 			goto out;
5496 		}
5497 
5498 		kvmppc_set_hpt(kvm, &info);
5499 	}
5500 
5501 	/* Look up the memslot for guest physical address 0 */
5502 	srcu_idx = srcu_read_lock(&kvm->srcu);
5503 	memslot = gfn_to_memslot(kvm, 0);
5504 
5505 	/* We must have some memory at 0 by now */
5506 	err = -EINVAL;
5507 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
5508 		goto out_srcu;
5509 
5510 	/* Look up the VMA for the start of this memory slot */
5511 	hva = memslot->userspace_addr;
5512 	mmap_read_lock(kvm->mm);
5513 	vma = vma_lookup(kvm->mm, hva);
5514 	if (!vma || (vma->vm_flags & VM_IO))
5515 		goto up_out;
5516 
5517 	psize = vma_kernel_pagesize(vma);
5518 
5519 	mmap_read_unlock(kvm->mm);
5520 
5521 	/* We can handle 4k, 64k or 16M pages in the VRMA */
5522 	if (psize >= 0x1000000)
5523 		psize = 0x1000000;
5524 	else if (psize >= 0x10000)
5525 		psize = 0x10000;
5526 	else
5527 		psize = 0x1000;
5528 	porder = __ilog2(psize);
5529 
5530 	senc = slb_pgsize_encoding(psize);
5531 	kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
5532 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
5533 	/* Create HPTEs in the hash page table for the VRMA */
5534 	kvmppc_map_vrma(vcpu, memslot, porder);
5535 
5536 	/* Update VRMASD field in the LPCR */
5537 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
5538 		/* the -4 is to account for senc values starting at 0x10 */
5539 		lpcr = senc << (LPCR_VRMASD_SH - 4);
5540 		kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
5541 	}
5542 
5543 	/* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
5544 	smp_wmb();
5545 	err = 0;
5546  out_srcu:
5547 	srcu_read_unlock(&kvm->srcu, srcu_idx);
5548  out:
5549 	return err;
5550 
5551  up_out:
5552 	mmap_read_unlock(kvm->mm);
5553 	goto out_srcu;
5554 }
5555 
5556 /*
5557  * Must be called with kvm->arch.mmu_setup_lock held and
5558  * mmu_ready = 0 and no vcpus running.
5559  */
5560 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
5561 {
5562 	unsigned long lpcr, lpcr_mask;
5563 
5564 	if (nesting_enabled(kvm))
5565 		kvmhv_release_all_nested(kvm);
5566 	kvmppc_rmap_reset(kvm);
5567 	kvm->arch.process_table = 0;
5568 	/* Mutual exclusion with kvm_unmap_gfn_range etc. */
5569 	spin_lock(&kvm->mmu_lock);
5570 	kvm->arch.radix = 0;
5571 	spin_unlock(&kvm->mmu_lock);
5572 	kvmppc_free_radix(kvm);
5573 
5574 	lpcr = LPCR_VPM1;
5575 	lpcr_mask = LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5576 	if (cpu_has_feature(CPU_FTR_ARCH_31))
5577 		lpcr_mask |= LPCR_HAIL;
5578 	kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
5579 
5580 	return 0;
5581 }
5582 
5583 /*
5584  * Must be called with kvm->arch.mmu_setup_lock held and
5585  * mmu_ready = 0 and no vcpus running.
5586  */
5587 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
5588 {
5589 	unsigned long lpcr, lpcr_mask;
5590 	int err;
5591 
5592 	err = kvmppc_init_vm_radix(kvm);
5593 	if (err)
5594 		return err;
5595 	kvmppc_rmap_reset(kvm);
5596 	/* Mutual exclusion with kvm_unmap_gfn_range etc. */
5597 	spin_lock(&kvm->mmu_lock);
5598 	kvm->arch.radix = 1;
5599 	spin_unlock(&kvm->mmu_lock);
5600 	kvmppc_free_hpt(&kvm->arch.hpt);
5601 
5602 	lpcr = LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5603 	lpcr_mask = LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5604 	if (cpu_has_feature(CPU_FTR_ARCH_31)) {
5605 		lpcr_mask |= LPCR_HAIL;
5606 		if (cpu_has_feature(CPU_FTR_HVMODE) &&
5607 				(kvm->arch.host_lpcr & LPCR_HAIL))
5608 			lpcr |= LPCR_HAIL;
5609 	}
5610 	kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
5611 
5612 	return 0;
5613 }
5614 
5615 #ifdef CONFIG_KVM_XICS
5616 /*
5617  * Allocate a per-core structure for managing state about which cores are
5618  * running in the host versus the guest and for exchanging data between
5619  * real mode KVM and CPU running in the host.
5620  * This is only done for the first VM.
5621  * The allocated structure stays even if all VMs have stopped.
5622  * It is only freed when the kvm-hv module is unloaded.
5623  * It's OK for this routine to fail, we just don't support host
5624  * core operations like redirecting H_IPI wakeups.
5625  */
5626 void kvmppc_alloc_host_rm_ops(void)
5627 {
5628 	struct kvmppc_host_rm_ops *ops;
5629 	unsigned long l_ops;
5630 	int cpu, core;
5631 	int size;
5632 
5633 	if (cpu_has_feature(CPU_FTR_ARCH_300))
5634 		return;
5635 
5636 	/* Not the first time here ? */
5637 	if (kvmppc_host_rm_ops_hv != NULL)
5638 		return;
5639 
5640 	ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
5641 	if (!ops)
5642 		return;
5643 
5644 	size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
5645 	ops->rm_core = kzalloc(size, GFP_KERNEL);
5646 
5647 	if (!ops->rm_core) {
5648 		kfree(ops);
5649 		return;
5650 	}
5651 
5652 	cpus_read_lock();
5653 
5654 	for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
5655 		if (!cpu_online(cpu))
5656 			continue;
5657 
5658 		core = cpu >> threads_shift;
5659 		ops->rm_core[core].rm_state.in_host = 1;
5660 	}
5661 
5662 	ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
5663 
5664 	/*
5665 	 * Make the contents of the kvmppc_host_rm_ops structure visible
5666 	 * to other CPUs before we assign it to the global variable.
5667 	 * Do an atomic assignment (no locks used here), but if someone
5668 	 * beats us to it, just free our copy and return.
5669 	 */
5670 	smp_wmb();
5671 	l_ops = (unsigned long) ops;
5672 
5673 	if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
5674 		cpus_read_unlock();
5675 		kfree(ops->rm_core);
5676 		kfree(ops);
5677 		return;
5678 	}
5679 
5680 	cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
5681 					     "ppc/kvm_book3s:prepare",
5682 					     kvmppc_set_host_core,
5683 					     kvmppc_clear_host_core);
5684 	cpus_read_unlock();
5685 }
5686 
5687 void kvmppc_free_host_rm_ops(void)
5688 {
5689 	if (kvmppc_host_rm_ops_hv) {
5690 		cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
5691 		kfree(kvmppc_host_rm_ops_hv->rm_core);
5692 		kfree(kvmppc_host_rm_ops_hv);
5693 		kvmppc_host_rm_ops_hv = NULL;
5694 	}
5695 }
5696 #endif
5697 
5698 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
5699 {
5700 	unsigned long lpcr, lpid;
5701 	int ret;
5702 
5703 	mutex_init(&kvm->arch.uvmem_lock);
5704 	INIT_LIST_HEAD(&kvm->arch.uvmem_pfns);
5705 	mutex_init(&kvm->arch.mmu_setup_lock);
5706 
5707 	/* Allocate the guest's logical partition ID */
5708 
5709 	if (!kvmhv_is_nestedv2()) {
5710 		lpid = kvmppc_alloc_lpid();
5711 		if ((long)lpid < 0)
5712 			return -ENOMEM;
5713 		kvm->arch.lpid = lpid;
5714 	}
5715 
5716 	kvmppc_alloc_host_rm_ops();
5717 
5718 	kvmhv_vm_nested_init(kvm);
5719 
5720 	if (kvmhv_is_nestedv2()) {
5721 		long rc;
5722 		unsigned long guest_id;
5723 
5724 		rc = plpar_guest_create(0, &guest_id);
5725 
5726 		if (rc != H_SUCCESS)
5727 			pr_err("KVM: Create Guest hcall failed, rc=%ld\n", rc);
5728 
5729 		switch (rc) {
5730 		case H_PARAMETER:
5731 		case H_FUNCTION:
5732 		case H_STATE:
5733 			return -EINVAL;
5734 		case H_NOT_ENOUGH_RESOURCES:
5735 		case H_ABORTED:
5736 			return -ENOMEM;
5737 		case H_AUTHORITY:
5738 			return -EPERM;
5739 		case H_NOT_AVAILABLE:
5740 			return -EBUSY;
5741 		}
5742 		kvm->arch.lpid = guest_id;
5743 	}
5744 
5745 
5746 	/*
5747 	 * Since we don't flush the TLB when tearing down a VM,
5748 	 * and this lpid might have previously been used,
5749 	 * make sure we flush on each core before running the new VM.
5750 	 * On POWER9, the tlbie in mmu_partition_table_set_entry()
5751 	 * does this flush for us.
5752 	 */
5753 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5754 		cpumask_setall(&kvm->arch.need_tlb_flush);
5755 
5756 	/* Start out with the default set of hcalls enabled */
5757 	memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
5758 	       sizeof(kvm->arch.enabled_hcalls));
5759 
5760 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5761 		kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
5762 
5763 	/* Init LPCR for virtual RMA mode */
5764 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
5765 		kvm->arch.host_lpid = mfspr(SPRN_LPID);
5766 		kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
5767 		lpcr &= LPCR_PECE | LPCR_LPES;
5768 	} else {
5769 		/*
5770 		 * The L2 LPES mode will be set by the L0 according to whether
5771 		 * or not it needs to take external interrupts in HV mode.
5772 		 */
5773 		lpcr = 0;
5774 	}
5775 	lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
5776 		LPCR_VPM0 | LPCR_VPM1;
5777 	kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
5778 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
5779 	/* On POWER8 turn on online bit to enable PURR/SPURR */
5780 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
5781 		lpcr |= LPCR_ONL;
5782 	/*
5783 	 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
5784 	 * Set HVICE bit to enable hypervisor virtualization interrupts.
5785 	 * Set HEIC to prevent OS interrupts to go to hypervisor (should
5786 	 * be unnecessary but better safe than sorry in case we re-enable
5787 	 * EE in HV mode with this LPCR still set)
5788 	 */
5789 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5790 		lpcr &= ~LPCR_VPM0;
5791 		lpcr |= LPCR_HVICE | LPCR_HEIC;
5792 
5793 		/*
5794 		 * If xive is enabled, we route 0x500 interrupts directly
5795 		 * to the guest.
5796 		 */
5797 		if (xics_on_xive())
5798 			lpcr |= LPCR_LPES;
5799 	}
5800 
5801 	/*
5802 	 * If the host uses radix, the guest starts out as radix.
5803 	 */
5804 	if (radix_enabled()) {
5805 		kvm->arch.radix = 1;
5806 		kvm->arch.mmu_ready = 1;
5807 		lpcr &= ~LPCR_VPM1;
5808 		lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5809 		if (cpu_has_feature(CPU_FTR_HVMODE) &&
5810 		    cpu_has_feature(CPU_FTR_ARCH_31) &&
5811 		    (kvm->arch.host_lpcr & LPCR_HAIL))
5812 			lpcr |= LPCR_HAIL;
5813 		ret = kvmppc_init_vm_radix(kvm);
5814 		if (ret) {
5815 			if (kvmhv_is_nestedv2())
5816 				plpar_guest_delete(0, kvm->arch.lpid);
5817 			else
5818 				kvmppc_free_lpid(kvm->arch.lpid);
5819 			return ret;
5820 		}
5821 		kvmppc_setup_partition_table(kvm);
5822 	}
5823 
5824 	verify_lpcr(kvm, lpcr);
5825 	kvm->arch.lpcr = lpcr;
5826 
5827 	/* Initialization for future HPT resizes */
5828 	kvm->arch.resize_hpt = NULL;
5829 
5830 	/*
5831 	 * Work out how many sets the TLB has, for the use of
5832 	 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
5833 	 */
5834 	if (cpu_has_feature(CPU_FTR_ARCH_31)) {
5835 		/*
5836 		 * P10 will flush all the congruence class with a single tlbiel
5837 		 */
5838 		kvm->arch.tlb_sets = 1;
5839 	} else if (radix_enabled())
5840 		kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;	/* 128 */
5841 	else if (cpu_has_feature(CPU_FTR_ARCH_300))
5842 		kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;	/* 256 */
5843 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
5844 		kvm->arch.tlb_sets = POWER8_TLB_SETS;		/* 512 */
5845 	else
5846 		kvm->arch.tlb_sets = POWER7_TLB_SETS;		/* 128 */
5847 
5848 	/*
5849 	 * Track that we now have a HV mode VM active. This blocks secondary
5850 	 * CPU threads from coming online.
5851 	 */
5852 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5853 		kvm_hv_vm_activated();
5854 
5855 	/*
5856 	 * Initialize smt_mode depending on processor.
5857 	 * POWER8 and earlier have to use "strict" threading, where
5858 	 * all vCPUs in a vcore have to run on the same (sub)core,
5859 	 * whereas on POWER9 the threads can each run a different
5860 	 * guest.
5861 	 */
5862 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5863 		kvm->arch.smt_mode = threads_per_subcore;
5864 	else
5865 		kvm->arch.smt_mode = 1;
5866 	kvm->arch.emul_smt_mode = 1;
5867 
5868 	return 0;
5869 }
5870 
5871 static int kvmppc_arch_create_vm_debugfs_hv(struct kvm *kvm)
5872 {
5873 	kvmppc_mmu_debugfs_init(kvm);
5874 	if (radix_enabled())
5875 		kvmhv_radix_debugfs_init(kvm);
5876 	return 0;
5877 }
5878 
5879 static void kvmppc_free_vcores(struct kvm *kvm)
5880 {
5881 	long int i;
5882 
5883 	for (i = 0; i < KVM_MAX_VCORES; ++i)
5884 		kfree(kvm->arch.vcores[i]);
5885 	kvm->arch.online_vcores = 0;
5886 }
5887 
5888 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
5889 {
5890 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5891 		kvm_hv_vm_deactivated();
5892 
5893 	kvmppc_free_vcores(kvm);
5894 
5895 
5896 	if (kvm_is_radix(kvm))
5897 		kvmppc_free_radix(kvm);
5898 	else
5899 		kvmppc_free_hpt(&kvm->arch.hpt);
5900 
5901 	/* Perform global invalidation and return lpid to the pool */
5902 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5903 		if (nesting_enabled(kvm))
5904 			kvmhv_release_all_nested(kvm);
5905 		kvm->arch.process_table = 0;
5906 		if (kvm->arch.secure_guest)
5907 			uv_svm_terminate(kvm->arch.lpid);
5908 		if (!kvmhv_is_nestedv2())
5909 			kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
5910 	}
5911 
5912 	if (kvmhv_is_nestedv2()) {
5913 		kvmhv_flush_lpid(kvm->arch.lpid);
5914 		plpar_guest_delete(0, kvm->arch.lpid);
5915 	} else {
5916 		kvmppc_free_lpid(kvm->arch.lpid);
5917 	}
5918 
5919 	kvmppc_free_pimap(kvm);
5920 }
5921 
5922 /* We don't need to emulate any privileged instructions or dcbz */
5923 static int kvmppc_core_emulate_op_hv(struct kvm_vcpu *vcpu,
5924 				     unsigned int inst, int *advance)
5925 {
5926 	return EMULATE_FAIL;
5927 }
5928 
5929 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
5930 					ulong spr_val)
5931 {
5932 	return EMULATE_FAIL;
5933 }
5934 
5935 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
5936 					ulong *spr_val)
5937 {
5938 	return EMULATE_FAIL;
5939 }
5940 
5941 static int kvmppc_core_check_processor_compat_hv(void)
5942 {
5943 	if (cpu_has_feature(CPU_FTR_HVMODE) &&
5944 	    cpu_has_feature(CPU_FTR_ARCH_206))
5945 		return 0;
5946 
5947 	/* POWER9 in radix mode is capable of being a nested hypervisor. */
5948 	if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
5949 		return 0;
5950 
5951 	return -EIO;
5952 }
5953 
5954 #ifdef CONFIG_KVM_XICS
5955 
5956 void kvmppc_free_pimap(struct kvm *kvm)
5957 {
5958 	kfree(kvm->arch.pimap);
5959 }
5960 
5961 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
5962 {
5963 	return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
5964 }
5965 
5966 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5967 {
5968 	struct irq_desc *desc;
5969 	struct kvmppc_irq_map *irq_map;
5970 	struct kvmppc_passthru_irqmap *pimap;
5971 	struct irq_chip *chip;
5972 	int i, rc = 0;
5973 	struct irq_data *host_data;
5974 
5975 	if (!kvm_irq_bypass)
5976 		return 1;
5977 
5978 	desc = irq_to_desc(host_irq);
5979 	if (!desc)
5980 		return -EIO;
5981 
5982 	mutex_lock(&kvm->lock);
5983 
5984 	pimap = kvm->arch.pimap;
5985 	if (pimap == NULL) {
5986 		/* First call, allocate structure to hold IRQ map */
5987 		pimap = kvmppc_alloc_pimap();
5988 		if (pimap == NULL) {
5989 			mutex_unlock(&kvm->lock);
5990 			return -ENOMEM;
5991 		}
5992 		kvm->arch.pimap = pimap;
5993 	}
5994 
5995 	/*
5996 	 * For now, we only support interrupts for which the EOI operation
5997 	 * is an OPAL call followed by a write to XIRR, since that's
5998 	 * what our real-mode EOI code does, or a XIVE interrupt
5999 	 */
6000 	chip = irq_data_get_irq_chip(&desc->irq_data);
6001 	if (!chip || !is_pnv_opal_msi(chip)) {
6002 		pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
6003 			host_irq, guest_gsi);
6004 		mutex_unlock(&kvm->lock);
6005 		return -ENOENT;
6006 	}
6007 
6008 	/*
6009 	 * See if we already have an entry for this guest IRQ number.
6010 	 * If it's mapped to a hardware IRQ number, that's an error,
6011 	 * otherwise re-use this entry.
6012 	 */
6013 	for (i = 0; i < pimap->n_mapped; i++) {
6014 		if (guest_gsi == pimap->mapped[i].v_hwirq) {
6015 			if (pimap->mapped[i].r_hwirq) {
6016 				mutex_unlock(&kvm->lock);
6017 				return -EINVAL;
6018 			}
6019 			break;
6020 		}
6021 	}
6022 
6023 	if (i == KVMPPC_PIRQ_MAPPED) {
6024 		mutex_unlock(&kvm->lock);
6025 		return -EAGAIN;		/* table is full */
6026 	}
6027 
6028 	irq_map = &pimap->mapped[i];
6029 
6030 	irq_map->v_hwirq = guest_gsi;
6031 	irq_map->desc = desc;
6032 
6033 	/*
6034 	 * Order the above two stores before the next to serialize with
6035 	 * the KVM real mode handler.
6036 	 */
6037 	smp_wmb();
6038 
6039 	/*
6040 	 * The 'host_irq' number is mapped in the PCI-MSI domain but
6041 	 * the underlying calls, which will EOI the interrupt in real
6042 	 * mode, need an HW IRQ number mapped in the XICS IRQ domain.
6043 	 */
6044 	host_data = irq_domain_get_irq_data(irq_get_default_host(), host_irq);
6045 	irq_map->r_hwirq = (unsigned int)irqd_to_hwirq(host_data);
6046 
6047 	if (i == pimap->n_mapped)
6048 		pimap->n_mapped++;
6049 
6050 	if (xics_on_xive())
6051 		rc = kvmppc_xive_set_mapped(kvm, guest_gsi, host_irq);
6052 	else
6053 		kvmppc_xics_set_mapped(kvm, guest_gsi, irq_map->r_hwirq);
6054 	if (rc)
6055 		irq_map->r_hwirq = 0;
6056 
6057 	mutex_unlock(&kvm->lock);
6058 
6059 	return 0;
6060 }
6061 
6062 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
6063 {
6064 	struct irq_desc *desc;
6065 	struct kvmppc_passthru_irqmap *pimap;
6066 	int i, rc = 0;
6067 
6068 	if (!kvm_irq_bypass)
6069 		return 0;
6070 
6071 	desc = irq_to_desc(host_irq);
6072 	if (!desc)
6073 		return -EIO;
6074 
6075 	mutex_lock(&kvm->lock);
6076 	if (!kvm->arch.pimap)
6077 		goto unlock;
6078 
6079 	pimap = kvm->arch.pimap;
6080 
6081 	for (i = 0; i < pimap->n_mapped; i++) {
6082 		if (guest_gsi == pimap->mapped[i].v_hwirq)
6083 			break;
6084 	}
6085 
6086 	if (i == pimap->n_mapped) {
6087 		mutex_unlock(&kvm->lock);
6088 		return -ENODEV;
6089 	}
6090 
6091 	if (xics_on_xive())
6092 		rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, host_irq);
6093 	else
6094 		kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
6095 
6096 	/* invalidate the entry (what to do on error from the above ?) */
6097 	pimap->mapped[i].r_hwirq = 0;
6098 
6099 	/*
6100 	 * We don't free this structure even when the count goes to
6101 	 * zero. The structure is freed when we destroy the VM.
6102 	 */
6103  unlock:
6104 	mutex_unlock(&kvm->lock);
6105 	return rc;
6106 }
6107 
6108 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
6109 					     struct irq_bypass_producer *prod)
6110 {
6111 	int ret = 0;
6112 	struct kvm_kernel_irqfd *irqfd =
6113 		container_of(cons, struct kvm_kernel_irqfd, consumer);
6114 
6115 	irqfd->producer = prod;
6116 
6117 	ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
6118 	if (ret)
6119 		pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
6120 			prod->irq, irqfd->gsi, ret);
6121 
6122 	return ret;
6123 }
6124 
6125 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
6126 					      struct irq_bypass_producer *prod)
6127 {
6128 	int ret;
6129 	struct kvm_kernel_irqfd *irqfd =
6130 		container_of(cons, struct kvm_kernel_irqfd, consumer);
6131 
6132 	irqfd->producer = NULL;
6133 
6134 	/*
6135 	 * When producer of consumer is unregistered, we change back to
6136 	 * default external interrupt handling mode - KVM real mode
6137 	 * will switch back to host.
6138 	 */
6139 	ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
6140 	if (ret)
6141 		pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
6142 			prod->irq, irqfd->gsi, ret);
6143 }
6144 #endif
6145 
6146 static int kvm_arch_vm_ioctl_hv(struct file *filp,
6147 				unsigned int ioctl, unsigned long arg)
6148 {
6149 	struct kvm *kvm __maybe_unused = filp->private_data;
6150 	void __user *argp = (void __user *)arg;
6151 	int r;
6152 
6153 	switch (ioctl) {
6154 
6155 	case KVM_PPC_ALLOCATE_HTAB: {
6156 		u32 htab_order;
6157 
6158 		/* If we're a nested hypervisor, we currently only support radix */
6159 		if (kvmhv_on_pseries()) {
6160 			r = -EOPNOTSUPP;
6161 			break;
6162 		}
6163 
6164 		r = -EFAULT;
6165 		if (get_user(htab_order, (u32 __user *)argp))
6166 			break;
6167 		r = kvmppc_alloc_reset_hpt(kvm, htab_order);
6168 		if (r)
6169 			break;
6170 		r = 0;
6171 		break;
6172 	}
6173 
6174 	case KVM_PPC_GET_HTAB_FD: {
6175 		struct kvm_get_htab_fd ghf;
6176 
6177 		r = -EFAULT;
6178 		if (copy_from_user(&ghf, argp, sizeof(ghf)))
6179 			break;
6180 		r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
6181 		break;
6182 	}
6183 
6184 	case KVM_PPC_RESIZE_HPT_PREPARE: {
6185 		struct kvm_ppc_resize_hpt rhpt;
6186 
6187 		r = -EFAULT;
6188 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
6189 			break;
6190 
6191 		r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
6192 		break;
6193 	}
6194 
6195 	case KVM_PPC_RESIZE_HPT_COMMIT: {
6196 		struct kvm_ppc_resize_hpt rhpt;
6197 
6198 		r = -EFAULT;
6199 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
6200 			break;
6201 
6202 		r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
6203 		break;
6204 	}
6205 
6206 	default:
6207 		r = -ENOTTY;
6208 	}
6209 
6210 	return r;
6211 }
6212 
6213 /*
6214  * List of hcall numbers to enable by default.
6215  * For compatibility with old userspace, we enable by default
6216  * all hcalls that were implemented before the hcall-enabling
6217  * facility was added.  Note this list should not include H_RTAS.
6218  */
6219 static unsigned int default_hcall_list[] = {
6220 	H_REMOVE,
6221 	H_ENTER,
6222 	H_READ,
6223 	H_PROTECT,
6224 	H_BULK_REMOVE,
6225 #ifdef CONFIG_SPAPR_TCE_IOMMU
6226 	H_GET_TCE,
6227 	H_PUT_TCE,
6228 #endif
6229 	H_SET_DABR,
6230 	H_SET_XDABR,
6231 	H_CEDE,
6232 	H_PROD,
6233 	H_CONFER,
6234 	H_REGISTER_VPA,
6235 #ifdef CONFIG_KVM_XICS
6236 	H_EOI,
6237 	H_CPPR,
6238 	H_IPI,
6239 	H_IPOLL,
6240 	H_XIRR,
6241 	H_XIRR_X,
6242 #endif
6243 	0
6244 };
6245 
6246 static void init_default_hcalls(void)
6247 {
6248 	int i;
6249 	unsigned int hcall;
6250 
6251 	for (i = 0; default_hcall_list[i]; ++i) {
6252 		hcall = default_hcall_list[i];
6253 		WARN_ON(!kvmppc_hcall_impl_hv(hcall));
6254 		__set_bit(hcall / 4, default_enabled_hcalls);
6255 	}
6256 }
6257 
6258 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
6259 {
6260 	unsigned long lpcr;
6261 	int radix;
6262 	int err;
6263 
6264 	/* If not on a POWER9, reject it */
6265 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
6266 		return -ENODEV;
6267 
6268 	/* If any unknown flags set, reject it */
6269 	if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
6270 		return -EINVAL;
6271 
6272 	/* GR (guest radix) bit in process_table field must match */
6273 	radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
6274 	if (!!(cfg->process_table & PATB_GR) != radix)
6275 		return -EINVAL;
6276 
6277 	/* Process table size field must be reasonable, i.e. <= 24 */
6278 	if ((cfg->process_table & PRTS_MASK) > 24)
6279 		return -EINVAL;
6280 
6281 	/* We can change a guest to/from radix now, if the host is radix */
6282 	if (radix && !radix_enabled())
6283 		return -EINVAL;
6284 
6285 	/* If we're a nested hypervisor, we currently only support radix */
6286 	if (kvmhv_on_pseries() && !radix)
6287 		return -EINVAL;
6288 
6289 	mutex_lock(&kvm->arch.mmu_setup_lock);
6290 	if (radix != kvm_is_radix(kvm)) {
6291 		if (kvm->arch.mmu_ready) {
6292 			kvm->arch.mmu_ready = 0;
6293 			/* order mmu_ready vs. vcpus_running */
6294 			smp_mb();
6295 			if (atomic_read(&kvm->arch.vcpus_running)) {
6296 				kvm->arch.mmu_ready = 1;
6297 				err = -EBUSY;
6298 				goto out_unlock;
6299 			}
6300 		}
6301 		if (radix)
6302 			err = kvmppc_switch_mmu_to_radix(kvm);
6303 		else
6304 			err = kvmppc_switch_mmu_to_hpt(kvm);
6305 		if (err)
6306 			goto out_unlock;
6307 	}
6308 
6309 	kvm->arch.process_table = cfg->process_table;
6310 	kvmppc_setup_partition_table(kvm);
6311 
6312 	lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
6313 	kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
6314 	err = 0;
6315 
6316  out_unlock:
6317 	mutex_unlock(&kvm->arch.mmu_setup_lock);
6318 	return err;
6319 }
6320 
6321 static int kvmhv_enable_nested(struct kvm *kvm)
6322 {
6323 	if (!nested)
6324 		return -EPERM;
6325 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
6326 		return -ENODEV;
6327 	if (!radix_enabled())
6328 		return -ENODEV;
6329 	if (kvmhv_is_nestedv2())
6330 		return -ENODEV;
6331 
6332 	/* kvm == NULL means the caller is testing if the capability exists */
6333 	if (kvm)
6334 		kvm->arch.nested_enable = true;
6335 	return 0;
6336 }
6337 
6338 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
6339 				 int size)
6340 {
6341 	int rc = -EINVAL;
6342 
6343 	if (kvmhv_vcpu_is_radix(vcpu)) {
6344 		rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
6345 
6346 		if (rc > 0)
6347 			rc = -EINVAL;
6348 	}
6349 
6350 	/* For now quadrants are the only way to access nested guest memory */
6351 	if (rc && vcpu->arch.nested)
6352 		rc = -EAGAIN;
6353 
6354 	return rc;
6355 }
6356 
6357 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
6358 				int size)
6359 {
6360 	int rc = -EINVAL;
6361 
6362 	if (kvmhv_vcpu_is_radix(vcpu)) {
6363 		rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
6364 
6365 		if (rc > 0)
6366 			rc = -EINVAL;
6367 	}
6368 
6369 	/* For now quadrants are the only way to access nested guest memory */
6370 	if (rc && vcpu->arch.nested)
6371 		rc = -EAGAIN;
6372 
6373 	return rc;
6374 }
6375 
6376 static void unpin_vpa_reset(struct kvm *kvm, struct kvmppc_vpa *vpa)
6377 {
6378 	unpin_vpa(kvm, vpa);
6379 	vpa->gpa = 0;
6380 	vpa->pinned_addr = NULL;
6381 	vpa->dirty = false;
6382 	vpa->update_pending = 0;
6383 }
6384 
6385 /*
6386  * Enable a guest to become a secure VM, or test whether
6387  * that could be enabled.
6388  * Called when the KVM_CAP_PPC_SECURE_GUEST capability is
6389  * tested (kvm == NULL) or enabled (kvm != NULL).
6390  */
6391 static int kvmhv_enable_svm(struct kvm *kvm)
6392 {
6393 	if (!kvmppc_uvmem_available())
6394 		return -EINVAL;
6395 	if (kvm)
6396 		kvm->arch.svm_enabled = 1;
6397 	return 0;
6398 }
6399 
6400 /*
6401  *  IOCTL handler to turn off secure mode of guest
6402  *
6403  * - Release all device pages
6404  * - Issue ucall to terminate the guest on the UV side
6405  * - Unpin the VPA pages.
6406  * - Reinit the partition scoped page tables
6407  */
6408 static int kvmhv_svm_off(struct kvm *kvm)
6409 {
6410 	struct kvm_vcpu *vcpu;
6411 	int mmu_was_ready;
6412 	int srcu_idx;
6413 	int ret = 0;
6414 	unsigned long i;
6415 
6416 	if (!(kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START))
6417 		return ret;
6418 
6419 	mutex_lock(&kvm->arch.mmu_setup_lock);
6420 	mmu_was_ready = kvm->arch.mmu_ready;
6421 	if (kvm->arch.mmu_ready) {
6422 		kvm->arch.mmu_ready = 0;
6423 		/* order mmu_ready vs. vcpus_running */
6424 		smp_mb();
6425 		if (atomic_read(&kvm->arch.vcpus_running)) {
6426 			kvm->arch.mmu_ready = 1;
6427 			ret = -EBUSY;
6428 			goto out;
6429 		}
6430 	}
6431 
6432 	srcu_idx = srcu_read_lock(&kvm->srcu);
6433 	for (i = 0; i < kvm_arch_nr_memslot_as_ids(kvm); i++) {
6434 		struct kvm_memory_slot *memslot;
6435 		struct kvm_memslots *slots = __kvm_memslots(kvm, i);
6436 		int bkt;
6437 
6438 		if (!slots)
6439 			continue;
6440 
6441 		kvm_for_each_memslot(memslot, bkt, slots) {
6442 			kvmppc_uvmem_drop_pages(memslot, kvm, true);
6443 			uv_unregister_mem_slot(kvm->arch.lpid, memslot->id);
6444 		}
6445 	}
6446 	srcu_read_unlock(&kvm->srcu, srcu_idx);
6447 
6448 	ret = uv_svm_terminate(kvm->arch.lpid);
6449 	if (ret != U_SUCCESS) {
6450 		ret = -EINVAL;
6451 		goto out;
6452 	}
6453 
6454 	/*
6455 	 * When secure guest is reset, all the guest pages are sent
6456 	 * to UV via UV_PAGE_IN before the non-boot vcpus get a
6457 	 * chance to run and unpin their VPA pages. Unpinning of all
6458 	 * VPA pages is done here explicitly so that VPA pages
6459 	 * can be migrated to the secure side.
6460 	 *
6461 	 * This is required to for the secure SMP guest to reboot
6462 	 * correctly.
6463 	 */
6464 	kvm_for_each_vcpu(i, vcpu, kvm) {
6465 		spin_lock(&vcpu->arch.vpa_update_lock);
6466 		unpin_vpa_reset(kvm, &vcpu->arch.dtl);
6467 		unpin_vpa_reset(kvm, &vcpu->arch.slb_shadow);
6468 		unpin_vpa_reset(kvm, &vcpu->arch.vpa);
6469 		spin_unlock(&vcpu->arch.vpa_update_lock);
6470 	}
6471 
6472 	kvmppc_setup_partition_table(kvm);
6473 	kvm->arch.secure_guest = 0;
6474 	kvm->arch.mmu_ready = mmu_was_ready;
6475 out:
6476 	mutex_unlock(&kvm->arch.mmu_setup_lock);
6477 	return ret;
6478 }
6479 
6480 static int kvmhv_enable_dawr1(struct kvm *kvm)
6481 {
6482 	if (!cpu_has_feature(CPU_FTR_DAWR1))
6483 		return -ENODEV;
6484 
6485 	/* kvm == NULL means the caller is testing if the capability exists */
6486 	if (kvm)
6487 		kvm->arch.dawr1_enabled = true;
6488 	return 0;
6489 }
6490 
6491 static bool kvmppc_hash_v3_possible(void)
6492 {
6493 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
6494 		return false;
6495 
6496 	if (!cpu_has_feature(CPU_FTR_HVMODE))
6497 		return false;
6498 
6499 	/*
6500 	 * POWER9 chips before version 2.02 can't have some threads in
6501 	 * HPT mode and some in radix mode on the same core.
6502 	 */
6503 	if (radix_enabled()) {
6504 		unsigned int pvr = mfspr(SPRN_PVR);
6505 		if ((pvr >> 16) == PVR_POWER9 &&
6506 		    (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
6507 		     ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
6508 			return false;
6509 	}
6510 
6511 	return true;
6512 }
6513 
6514 static struct kvmppc_ops kvm_ops_hv = {
6515 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
6516 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
6517 	.get_one_reg = kvmppc_get_one_reg_hv,
6518 	.set_one_reg = kvmppc_set_one_reg_hv,
6519 	.vcpu_load   = kvmppc_core_vcpu_load_hv,
6520 	.vcpu_put    = kvmppc_core_vcpu_put_hv,
6521 	.inject_interrupt = kvmppc_inject_interrupt_hv,
6522 	.set_msr     = kvmppc_set_msr_hv,
6523 	.vcpu_run    = kvmppc_vcpu_run_hv,
6524 	.vcpu_create = kvmppc_core_vcpu_create_hv,
6525 	.vcpu_free   = kvmppc_core_vcpu_free_hv,
6526 	.check_requests = kvmppc_core_check_requests_hv,
6527 	.get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
6528 	.flush_memslot  = kvmppc_core_flush_memslot_hv,
6529 	.prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
6530 	.commit_memory_region  = kvmppc_core_commit_memory_region_hv,
6531 	.unmap_gfn_range = kvm_unmap_gfn_range_hv,
6532 	.age_gfn = kvm_age_gfn_hv,
6533 	.test_age_gfn = kvm_test_age_gfn_hv,
6534 	.free_memslot = kvmppc_core_free_memslot_hv,
6535 	.init_vm =  kvmppc_core_init_vm_hv,
6536 	.destroy_vm = kvmppc_core_destroy_vm_hv,
6537 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
6538 	.emulate_op = kvmppc_core_emulate_op_hv,
6539 	.emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
6540 	.emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
6541 	.fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
6542 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
6543 	.hcall_implemented = kvmppc_hcall_impl_hv,
6544 #ifdef CONFIG_KVM_XICS
6545 	.irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
6546 	.irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
6547 #endif
6548 	.configure_mmu = kvmhv_configure_mmu,
6549 	.get_rmmu_info = kvmhv_get_rmmu_info,
6550 	.set_smt_mode = kvmhv_set_smt_mode,
6551 	.enable_nested = kvmhv_enable_nested,
6552 	.load_from_eaddr = kvmhv_load_from_eaddr,
6553 	.store_to_eaddr = kvmhv_store_to_eaddr,
6554 	.enable_svm = kvmhv_enable_svm,
6555 	.svm_off = kvmhv_svm_off,
6556 	.enable_dawr1 = kvmhv_enable_dawr1,
6557 	.hash_v3_possible = kvmppc_hash_v3_possible,
6558 	.create_vcpu_debugfs = kvmppc_arch_create_vcpu_debugfs_hv,
6559 	.create_vm_debugfs = kvmppc_arch_create_vm_debugfs_hv,
6560 };
6561 
6562 static int kvm_init_subcore_bitmap(void)
6563 {
6564 	int i, j;
6565 	int nr_cores = cpu_nr_cores();
6566 	struct sibling_subcore_state *sibling_subcore_state;
6567 
6568 	for (i = 0; i < nr_cores; i++) {
6569 		int first_cpu = i * threads_per_core;
6570 		int node = cpu_to_node(first_cpu);
6571 
6572 		/* Ignore if it is already allocated. */
6573 		if (paca_ptrs[first_cpu]->sibling_subcore_state)
6574 			continue;
6575 
6576 		sibling_subcore_state =
6577 			kzalloc_node(sizeof(struct sibling_subcore_state),
6578 							GFP_KERNEL, node);
6579 		if (!sibling_subcore_state)
6580 			return -ENOMEM;
6581 
6582 
6583 		for (j = 0; j < threads_per_core; j++) {
6584 			int cpu = first_cpu + j;
6585 
6586 			paca_ptrs[cpu]->sibling_subcore_state =
6587 						sibling_subcore_state;
6588 		}
6589 	}
6590 	return 0;
6591 }
6592 
6593 static int kvmppc_radix_possible(void)
6594 {
6595 	return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
6596 }
6597 
6598 static int kvmppc_book3s_init_hv(void)
6599 {
6600 	int r;
6601 
6602 	if (!tlbie_capable) {
6603 		pr_err("KVM-HV: Host does not support TLBIE\n");
6604 		return -ENODEV;
6605 	}
6606 
6607 	/*
6608 	 * FIXME!! Do we need to check on all cpus ?
6609 	 */
6610 	r = kvmppc_core_check_processor_compat_hv();
6611 	if (r < 0)
6612 		return -ENODEV;
6613 
6614 	r = kvmhv_nested_init();
6615 	if (r)
6616 		return r;
6617 
6618 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
6619 		r = kvm_init_subcore_bitmap();
6620 		if (r)
6621 			goto err;
6622 	}
6623 
6624 	/*
6625 	 * We need a way of accessing the XICS interrupt controller,
6626 	 * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
6627 	 * indirectly, via OPAL.
6628 	 */
6629 #ifdef CONFIG_SMP
6630 	if (!xics_on_xive() && !kvmhv_on_pseries() &&
6631 	    !local_paca->kvm_hstate.xics_phys) {
6632 		struct device_node *np;
6633 
6634 		np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
6635 		if (!np) {
6636 			pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
6637 			r = -ENODEV;
6638 			goto err;
6639 		}
6640 		/* presence of intc confirmed - node can be dropped again */
6641 		of_node_put(np);
6642 	}
6643 #endif
6644 
6645 	init_default_hcalls();
6646 
6647 	init_vcore_lists();
6648 
6649 	r = kvmppc_mmu_hv_init();
6650 	if (r)
6651 		goto err;
6652 
6653 	if (kvmppc_radix_possible()) {
6654 		r = kvmppc_radix_init();
6655 		if (r)
6656 			goto err;
6657 	}
6658 
6659 	r = kvmppc_uvmem_init();
6660 	if (r < 0) {
6661 		pr_err("KVM-HV: kvmppc_uvmem_init failed %d\n", r);
6662 		return r;
6663 	}
6664 
6665 	kvm_ops_hv.owner = THIS_MODULE;
6666 	kvmppc_hv_ops = &kvm_ops_hv;
6667 
6668 	return 0;
6669 
6670 err:
6671 	kvmhv_nested_exit();
6672 	kvmppc_radix_exit();
6673 
6674 	return r;
6675 }
6676 
6677 static void kvmppc_book3s_exit_hv(void)
6678 {
6679 	kvmppc_uvmem_free();
6680 	kvmppc_free_host_rm_ops();
6681 	if (kvmppc_radix_possible())
6682 		kvmppc_radix_exit();
6683 	kvmppc_hv_ops = NULL;
6684 	kvmhv_nested_exit();
6685 }
6686 
6687 module_init(kvmppc_book3s_init_hv);
6688 module_exit(kvmppc_book3s_exit_hv);
6689 MODULE_DESCRIPTION("KVM on Book3S (POWER8 and later) in hypervisor mode");
6690 MODULE_LICENSE("GPL");
6691 MODULE_ALIAS_MISCDEV(KVM_MINOR);
6692 MODULE_ALIAS("devname:kvm");
6693