xref: /linux/arch/powerpc/kvm/book3s_hv.c (revision 396042fb2b834a8fbcea9c850dbbd4ae7c7b75a9)
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/kvm_irqfd.h>
40 #include <linux/irqbypass.h>
41 #include <linux/module.h>
42 #include <linux/compiler.h>
43 #include <linux/of.h>
44 #include <linux/irqdomain.h>
45 #include <linux/smp.h>
46 
47 #include <asm/ftrace.h>
48 #include <asm/reg.h>
49 #include <asm/ppc-opcode.h>
50 #include <asm/asm-prototypes.h>
51 #include <asm/archrandom.h>
52 #include <asm/debug.h>
53 #include <asm/disassemble.h>
54 #include <asm/cputable.h>
55 #include <asm/cacheflush.h>
56 #include <linux/uaccess.h>
57 #include <asm/interrupt.h>
58 #include <asm/io.h>
59 #include <asm/kvm_ppc.h>
60 #include <asm/kvm_book3s.h>
61 #include <asm/mmu_context.h>
62 #include <asm/lppaca.h>
63 #include <asm/pmc.h>
64 #include <asm/processor.h>
65 #include <asm/cputhreads.h>
66 #include <asm/page.h>
67 #include <asm/hvcall.h>
68 #include <asm/switch_to.h>
69 #include <asm/smp.h>
70 #include <asm/dbell.h>
71 #include <asm/hmi.h>
72 #include <asm/pnv-pci.h>
73 #include <asm/mmu.h>
74 #include <asm/opal.h>
75 #include <asm/xics.h>
76 #include <asm/xive.h>
77 #include <asm/hw_breakpoint.h>
78 #include <asm/kvm_book3s_uvmem.h>
79 #include <asm/ultravisor.h>
80 #include <asm/dtl.h>
81 #include <asm/plpar_wrappers.h>
82 
83 #include <trace/events/ipi.h>
84 
85 #include "book3s.h"
86 #include "book3s_hv.h"
87 
88 #define CREATE_TRACE_POINTS
89 #include "trace_hv.h"
90 
91 /* #define EXIT_DEBUG */
92 /* #define EXIT_DEBUG_SIMPLE */
93 /* #define EXIT_DEBUG_INT */
94 
95 /* Used to indicate that a guest page fault needs to be handled */
96 #define RESUME_PAGE_FAULT	(RESUME_GUEST | RESUME_FLAG_ARCH1)
97 /* Used to indicate that a guest passthrough interrupt needs to be handled */
98 #define RESUME_PASSTHROUGH	(RESUME_GUEST | RESUME_FLAG_ARCH2)
99 
100 /* Used as a "null" value for timebase values */
101 #define TB_NIL	(~(u64)0)
102 
103 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
104 
105 static int dynamic_mt_modes = 6;
106 module_param(dynamic_mt_modes, int, 0644);
107 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
108 static int target_smt_mode;
109 module_param(target_smt_mode, int, 0644);
110 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
111 
112 static bool one_vm_per_core;
113 module_param(one_vm_per_core, bool, S_IRUGO | S_IWUSR);
114 MODULE_PARM_DESC(one_vm_per_core, "Only run vCPUs from the same VM on a core (requires POWER8 or older)");
115 
116 #ifdef CONFIG_KVM_XICS
117 static const struct kernel_param_ops module_param_ops = {
118 	.set = param_set_int,
119 	.get = param_get_int,
120 };
121 
122 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
123 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
124 
125 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
126 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
127 #endif
128 
129 /* If set, guests are allowed to create and control nested guests */
130 static bool nested = true;
131 module_param(nested, bool, S_IRUGO | S_IWUSR);
132 MODULE_PARM_DESC(nested, "Enable nested virtualization (only on POWER9)");
133 
134 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
135 
136 /*
137  * RWMR values for POWER8.  These control the rate at which PURR
138  * and SPURR count and should be set according to the number of
139  * online threads in the vcore being run.
140  */
141 #define RWMR_RPA_P8_1THREAD	0x164520C62609AECAUL
142 #define RWMR_RPA_P8_2THREAD	0x7FFF2908450D8DA9UL
143 #define RWMR_RPA_P8_3THREAD	0x164520C62609AECAUL
144 #define RWMR_RPA_P8_4THREAD	0x199A421245058DA9UL
145 #define RWMR_RPA_P8_5THREAD	0x164520C62609AECAUL
146 #define RWMR_RPA_P8_6THREAD	0x164520C62609AECAUL
147 #define RWMR_RPA_P8_7THREAD	0x164520C62609AECAUL
148 #define RWMR_RPA_P8_8THREAD	0x164520C62609AECAUL
149 
150 static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = {
151 	RWMR_RPA_P8_1THREAD,
152 	RWMR_RPA_P8_1THREAD,
153 	RWMR_RPA_P8_2THREAD,
154 	RWMR_RPA_P8_3THREAD,
155 	RWMR_RPA_P8_4THREAD,
156 	RWMR_RPA_P8_5THREAD,
157 	RWMR_RPA_P8_6THREAD,
158 	RWMR_RPA_P8_7THREAD,
159 	RWMR_RPA_P8_8THREAD,
160 };
161 
162 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
163 		int *ip)
164 {
165 	int i = *ip;
166 	struct kvm_vcpu *vcpu;
167 
168 	while (++i < MAX_SMT_THREADS) {
169 		vcpu = READ_ONCE(vc->runnable_threads[i]);
170 		if (vcpu) {
171 			*ip = i;
172 			return vcpu;
173 		}
174 	}
175 	return NULL;
176 }
177 
178 /* Used to traverse the list of runnable threads for a given vcore */
179 #define for_each_runnable_thread(i, vcpu, vc) \
180 	for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
181 
182 static bool kvmppc_ipi_thread(int cpu)
183 {
184 	unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
185 
186 	/* If we're a nested hypervisor, fall back to ordinary IPIs for now */
187 	if (kvmhv_on_pseries())
188 		return false;
189 
190 	/* On POWER9 we can use msgsnd to IPI any cpu */
191 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
192 		msg |= get_hard_smp_processor_id(cpu);
193 		smp_mb();
194 		__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
195 		return true;
196 	}
197 
198 	/* On POWER8 for IPIs to threads in the same core, use msgsnd */
199 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
200 		preempt_disable();
201 		if (cpu_first_thread_sibling(cpu) ==
202 		    cpu_first_thread_sibling(smp_processor_id())) {
203 			msg |= cpu_thread_in_core(cpu);
204 			smp_mb();
205 			__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
206 			preempt_enable();
207 			return true;
208 		}
209 		preempt_enable();
210 	}
211 
212 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
213 	if (cpu >= 0 && cpu < nr_cpu_ids) {
214 		if (paca_ptrs[cpu]->kvm_hstate.xics_phys) {
215 			xics_wake_cpu(cpu);
216 			return true;
217 		}
218 		opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
219 		return true;
220 	}
221 #endif
222 
223 	return false;
224 }
225 
226 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
227 {
228 	int cpu;
229 	struct rcuwait *waitp;
230 
231 	/*
232 	 * rcuwait_wake_up contains smp_mb() which orders prior stores that
233 	 * create pending work vs below loads of cpu fields. The other side
234 	 * is the barrier in vcpu run that orders setting the cpu fields vs
235 	 * testing for pending work.
236 	 */
237 
238 	waitp = kvm_arch_vcpu_get_wait(vcpu);
239 	if (rcuwait_wake_up(waitp))
240 		++vcpu->stat.generic.halt_wakeup;
241 
242 	cpu = READ_ONCE(vcpu->arch.thread_cpu);
243 	if (cpu >= 0 && kvmppc_ipi_thread(cpu))
244 		return;
245 
246 	/* CPU points to the first thread of the core */
247 	cpu = vcpu->cpu;
248 	if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
249 		smp_send_reschedule(cpu);
250 }
251 
252 /*
253  * We use the vcpu_load/put functions to measure stolen time.
254  *
255  * Stolen time is counted as time when either the vcpu is able to
256  * run as part of a virtual core, but the task running the vcore
257  * is preempted or sleeping, or when the vcpu needs something done
258  * in the kernel by the task running the vcpu, but that task is
259  * preempted or sleeping.  Those two things have to be counted
260  * separately, since one of the vcpu tasks will take on the job
261  * of running the core, and the other vcpu tasks in the vcore will
262  * sleep waiting for it to do that, but that sleep shouldn't count
263  * as stolen time.
264  *
265  * Hence we accumulate stolen time when the vcpu can run as part of
266  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
267  * needs its task to do other things in the kernel (for example,
268  * service a page fault) in busy_stolen.  We don't accumulate
269  * stolen time for a vcore when it is inactive, or for a vcpu
270  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
271  * a misnomer; it means that the vcpu task is not executing in
272  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
273  * the kernel.  We don't have any way of dividing up that time
274  * between time that the vcpu is genuinely stopped, time that
275  * the task is actively working on behalf of the vcpu, and time
276  * that the task is preempted, so we don't count any of it as
277  * stolen.
278  *
279  * Updates to busy_stolen are protected by arch.tbacct_lock;
280  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
281  * lock.  The stolen times are measured in units of timebase ticks.
282  * (Note that the != TB_NIL checks below are purely defensive;
283  * they should never fail.)
284  *
285  * The POWER9 path is simpler, one vcpu per virtual core so the
286  * former case does not exist. If a vcpu is preempted when it is
287  * BUSY_IN_HOST and not ceded or otherwise blocked, then accumulate
288  * the stolen cycles in busy_stolen. RUNNING is not a preemptible
289  * state in the P9 path.
290  */
291 
292 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc, u64 tb)
293 {
294 	unsigned long flags;
295 
296 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
297 
298 	spin_lock_irqsave(&vc->stoltb_lock, flags);
299 	vc->preempt_tb = tb;
300 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
301 }
302 
303 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc, u64 tb)
304 {
305 	unsigned long flags;
306 
307 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
308 
309 	spin_lock_irqsave(&vc->stoltb_lock, flags);
310 	if (vc->preempt_tb != TB_NIL) {
311 		vc->stolen_tb += tb - vc->preempt_tb;
312 		vc->preempt_tb = TB_NIL;
313 	}
314 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
315 }
316 
317 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
318 {
319 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
320 	unsigned long flags;
321 	u64 now;
322 
323 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
324 		if (vcpu->arch.busy_preempt != TB_NIL) {
325 			WARN_ON_ONCE(vcpu->arch.state != KVMPPC_VCPU_BUSY_IN_HOST);
326 			vc->stolen_tb += mftb() - vcpu->arch.busy_preempt;
327 			vcpu->arch.busy_preempt = TB_NIL;
328 		}
329 		return;
330 	}
331 
332 	now = mftb();
333 
334 	/*
335 	 * We can test vc->runner without taking the vcore lock,
336 	 * because only this task ever sets vc->runner to this
337 	 * vcpu, and once it is set to this vcpu, only this task
338 	 * ever sets it to NULL.
339 	 */
340 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
341 		kvmppc_core_end_stolen(vc, now);
342 
343 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
344 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
345 	    vcpu->arch.busy_preempt != TB_NIL) {
346 		vcpu->arch.busy_stolen += now - vcpu->arch.busy_preempt;
347 		vcpu->arch.busy_preempt = TB_NIL;
348 	}
349 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
350 }
351 
352 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
353 {
354 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
355 	unsigned long flags;
356 	u64 now;
357 
358 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
359 		/*
360 		 * In the P9 path, RUNNABLE is not preemptible
361 		 * (nor takes host interrupts)
362 		 */
363 		WARN_ON_ONCE(vcpu->arch.state == KVMPPC_VCPU_RUNNABLE);
364 		/*
365 		 * Account stolen time when preempted while the vcpu task is
366 		 * running in the kernel (but not in qemu, which is INACTIVE).
367 		 */
368 		if (task_is_running(current) &&
369 				vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
370 			vcpu->arch.busy_preempt = mftb();
371 		return;
372 	}
373 
374 	now = mftb();
375 
376 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
377 		kvmppc_core_start_stolen(vc, now);
378 
379 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
380 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
381 		vcpu->arch.busy_preempt = now;
382 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
383 }
384 
385 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
386 {
387 	vcpu->arch.pvr = pvr;
388 }
389 
390 /* Dummy value used in computing PCR value below */
391 #define PCR_ARCH_31    (PCR_ARCH_300 << 1)
392 
393 static inline unsigned long map_pcr_to_cap(unsigned long pcr)
394 {
395 	unsigned long cap = 0;
396 
397 	switch (pcr) {
398 	case PCR_ARCH_300:
399 		cap = H_GUEST_CAP_POWER9;
400 		break;
401 	case PCR_ARCH_31:
402 		if (cpu_has_feature(CPU_FTR_P11_PVR))
403 			cap = H_GUEST_CAP_POWER11;
404 		else
405 			cap = H_GUEST_CAP_POWER10;
406 		break;
407 	default:
408 		break;
409 	}
410 
411 	return cap;
412 }
413 
414 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
415 {
416 	unsigned long host_pcr_bit = 0, guest_pcr_bit = 0, cap = 0;
417 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
418 
419 	/* We can (emulate) our own architecture version and anything older */
420 	if (cpu_has_feature(CPU_FTR_P11_PVR) || cpu_has_feature(CPU_FTR_ARCH_31))
421 		host_pcr_bit = PCR_ARCH_31;
422 	else if (cpu_has_feature(CPU_FTR_ARCH_300))
423 		host_pcr_bit = PCR_ARCH_300;
424 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
425 		host_pcr_bit = PCR_ARCH_207;
426 	else if (cpu_has_feature(CPU_FTR_ARCH_206))
427 		host_pcr_bit = PCR_ARCH_206;
428 	else
429 		host_pcr_bit = PCR_ARCH_205;
430 
431 	/* Determine lowest PCR bit needed to run guest in given PVR level */
432 	guest_pcr_bit = host_pcr_bit;
433 	if (arch_compat) {
434 		switch (arch_compat) {
435 		case PVR_ARCH_205:
436 			guest_pcr_bit = PCR_ARCH_205;
437 			break;
438 		case PVR_ARCH_206:
439 		case PVR_ARCH_206p:
440 			guest_pcr_bit = PCR_ARCH_206;
441 			break;
442 		case PVR_ARCH_207:
443 			guest_pcr_bit = PCR_ARCH_207;
444 			break;
445 		case PVR_ARCH_300:
446 			guest_pcr_bit = PCR_ARCH_300;
447 			break;
448 		case PVR_ARCH_31:
449 		case PVR_ARCH_31_P11:
450 			guest_pcr_bit = PCR_ARCH_31;
451 			break;
452 		default:
453 			return -EINVAL;
454 		}
455 	}
456 
457 	/* Check requested PCR bits don't exceed our capabilities */
458 	if (guest_pcr_bit > host_pcr_bit)
459 		return -EINVAL;
460 
461 	if (kvmhv_on_pseries() && kvmhv_is_nestedv2()) {
462 		/*
463 		 * 'arch_compat == 0' would mean the guest should default to
464 		 * L1's compatibility. In this case, the guest would pick
465 		 * host's PCR and evaluate the corresponding capabilities.
466 		 */
467 		cap = map_pcr_to_cap(guest_pcr_bit);
468 		if (!(cap & nested_capabilities))
469 			return -EINVAL;
470 	}
471 
472 	spin_lock(&vc->lock);
473 	vc->arch_compat = arch_compat;
474 	kvmhv_nestedv2_mark_dirty(vcpu, KVMPPC_GSID_LOGICAL_PVR);
475 	/*
476 	 * Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit
477 	 * Also set all reserved PCR bits
478 	 */
479 	vc->pcr = (host_pcr_bit - guest_pcr_bit) | PCR_MASK;
480 	spin_unlock(&vc->lock);
481 
482 	return 0;
483 }
484 
485 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
486 {
487 	int r;
488 
489 	pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
490 	pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
491 	       vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
492 	for (r = 0; r < 16; ++r)
493 		pr_err("r%2d = %.16lx  r%d = %.16lx\n",
494 		       r, kvmppc_get_gpr(vcpu, r),
495 		       r+16, kvmppc_get_gpr(vcpu, r+16));
496 	pr_err("ctr = %.16lx  lr  = %.16lx\n",
497 	       vcpu->arch.regs.ctr, vcpu->arch.regs.link);
498 	pr_err("srr0 = %.16llx srr1 = %.16llx\n",
499 	       vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
500 	pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
501 	       vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
502 	pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
503 	       vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
504 	pr_err("cr = %.8lx  xer = %.16lx  dsisr = %.8x\n",
505 	       vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
506 	pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
507 	pr_err("fault dar = %.16lx dsisr = %.8x\n",
508 	       vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
509 	pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
510 	for (r = 0; r < vcpu->arch.slb_max; ++r)
511 		pr_err("  ESID = %.16llx VSID = %.16llx\n",
512 		       vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
513 	pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.16lx\n",
514 	       vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
515 	       vcpu->arch.last_inst);
516 }
517 
518 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
519 {
520 	return kvm_get_vcpu_by_id(kvm, id);
521 }
522 
523 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
524 {
525 	vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
526 	vpa->yield_count = cpu_to_be32(1);
527 }
528 
529 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
530 		   unsigned long addr, unsigned long len)
531 {
532 	/* check address is cacheline aligned */
533 	if (addr & (L1_CACHE_BYTES - 1))
534 		return -EINVAL;
535 	spin_lock(&vcpu->arch.vpa_update_lock);
536 	if (v->next_gpa != addr || v->len != len) {
537 		v->next_gpa = addr;
538 		v->len = addr ? len : 0;
539 		v->update_pending = 1;
540 	}
541 	spin_unlock(&vcpu->arch.vpa_update_lock);
542 	return 0;
543 }
544 
545 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
546 struct reg_vpa {
547 	u32 dummy;
548 	union {
549 		__be16 hword;
550 		__be32 word;
551 	} length;
552 };
553 
554 static int vpa_is_registered(struct kvmppc_vpa *vpap)
555 {
556 	if (vpap->update_pending)
557 		return vpap->next_gpa != 0;
558 	return vpap->pinned_addr != NULL;
559 }
560 
561 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
562 				       unsigned long flags,
563 				       unsigned long vcpuid, unsigned long vpa)
564 {
565 	struct kvm *kvm = vcpu->kvm;
566 	unsigned long len, nb;
567 	void *va;
568 	struct kvm_vcpu *tvcpu;
569 	int err;
570 	int subfunc;
571 	struct kvmppc_vpa *vpap;
572 
573 	tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
574 	if (!tvcpu)
575 		return H_PARAMETER;
576 
577 	subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
578 	if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
579 	    subfunc == H_VPA_REG_SLB) {
580 		/* Registering new area - address must be cache-line aligned */
581 		if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
582 			return H_PARAMETER;
583 
584 		/* convert logical addr to kernel addr and read length */
585 		va = kvmppc_pin_guest_page(kvm, vpa, &nb);
586 		if (va == NULL)
587 			return H_PARAMETER;
588 		if (subfunc == H_VPA_REG_VPA)
589 			len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
590 		else
591 			len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
592 		kvmppc_unpin_guest_page(kvm, va, vpa, false);
593 
594 		/* Check length */
595 		if (len > nb || len < sizeof(struct reg_vpa))
596 			return H_PARAMETER;
597 	} else {
598 		vpa = 0;
599 		len = 0;
600 	}
601 
602 	err = H_PARAMETER;
603 	vpap = NULL;
604 	spin_lock(&tvcpu->arch.vpa_update_lock);
605 
606 	switch (subfunc) {
607 	case H_VPA_REG_VPA:		/* register VPA */
608 		/*
609 		 * The size of our lppaca is 1kB because of the way we align
610 		 * it for the guest to avoid crossing a 4kB boundary. We only
611 		 * use 640 bytes of the structure though, so we should accept
612 		 * clients that set a size of 640.
613 		 */
614 		BUILD_BUG_ON(sizeof(struct lppaca) != 640);
615 		if (len < sizeof(struct lppaca))
616 			break;
617 		vpap = &tvcpu->arch.vpa;
618 		err = 0;
619 		break;
620 
621 	case H_VPA_REG_DTL:		/* register DTL */
622 		if (len < sizeof(struct dtl_entry))
623 			break;
624 		len -= len % sizeof(struct dtl_entry);
625 
626 		/* Check that they have previously registered a VPA */
627 		err = H_RESOURCE;
628 		if (!vpa_is_registered(&tvcpu->arch.vpa))
629 			break;
630 
631 		vpap = &tvcpu->arch.dtl;
632 		err = 0;
633 		break;
634 
635 	case H_VPA_REG_SLB:		/* register SLB shadow buffer */
636 		/* Check that they have previously registered a VPA */
637 		err = H_RESOURCE;
638 		if (!vpa_is_registered(&tvcpu->arch.vpa))
639 			break;
640 
641 		vpap = &tvcpu->arch.slb_shadow;
642 		err = 0;
643 		break;
644 
645 	case H_VPA_DEREG_VPA:		/* deregister VPA */
646 		/* Check they don't still have a DTL or SLB buf registered */
647 		err = H_RESOURCE;
648 		if (vpa_is_registered(&tvcpu->arch.dtl) ||
649 		    vpa_is_registered(&tvcpu->arch.slb_shadow))
650 			break;
651 
652 		vpap = &tvcpu->arch.vpa;
653 		err = 0;
654 		break;
655 
656 	case H_VPA_DEREG_DTL:		/* deregister DTL */
657 		vpap = &tvcpu->arch.dtl;
658 		err = 0;
659 		break;
660 
661 	case H_VPA_DEREG_SLB:		/* deregister SLB shadow buffer */
662 		vpap = &tvcpu->arch.slb_shadow;
663 		err = 0;
664 		break;
665 	}
666 
667 	if (vpap) {
668 		vpap->next_gpa = vpa;
669 		vpap->len = len;
670 		vpap->update_pending = 1;
671 	}
672 
673 	spin_unlock(&tvcpu->arch.vpa_update_lock);
674 
675 	return err;
676 }
677 
678 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap,
679 			       struct kvmppc_vpa *old_vpap)
680 {
681 	struct kvm *kvm = vcpu->kvm;
682 	void *va;
683 	unsigned long nb;
684 	unsigned long gpa;
685 
686 	/*
687 	 * We need to pin the page pointed to by vpap->next_gpa,
688 	 * but we can't call kvmppc_pin_guest_page under the lock
689 	 * as it does get_user_pages() and down_read().  So we
690 	 * have to drop the lock, pin the page, then get the lock
691 	 * again and check that a new area didn't get registered
692 	 * in the meantime.
693 	 */
694 	for (;;) {
695 		gpa = vpap->next_gpa;
696 		spin_unlock(&vcpu->arch.vpa_update_lock);
697 		va = NULL;
698 		nb = 0;
699 		if (gpa)
700 			va = kvmppc_pin_guest_page(kvm, gpa, &nb);
701 		spin_lock(&vcpu->arch.vpa_update_lock);
702 		if (gpa == vpap->next_gpa)
703 			break;
704 		/* sigh... unpin that one and try again */
705 		if (va)
706 			kvmppc_unpin_guest_page(kvm, va, gpa, false);
707 	}
708 
709 	vpap->update_pending = 0;
710 	if (va && nb < vpap->len) {
711 		/*
712 		 * If it's now too short, it must be that userspace
713 		 * has changed the mappings underlying guest memory,
714 		 * so unregister the region.
715 		 */
716 		kvmppc_unpin_guest_page(kvm, va, gpa, false);
717 		va = NULL;
718 	}
719 	*old_vpap = *vpap;
720 
721 	vpap->gpa = gpa;
722 	vpap->pinned_addr = va;
723 	vpap->dirty = false;
724 	if (va)
725 		vpap->pinned_end = va + vpap->len;
726 }
727 
728 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
729 {
730 	struct kvm *kvm = vcpu->kvm;
731 	struct kvmppc_vpa old_vpa = { 0 };
732 
733 	if (!(vcpu->arch.vpa.update_pending ||
734 	      vcpu->arch.slb_shadow.update_pending ||
735 	      vcpu->arch.dtl.update_pending))
736 		return;
737 
738 	spin_lock(&vcpu->arch.vpa_update_lock);
739 	if (vcpu->arch.vpa.update_pending) {
740 		kvmppc_update_vpa(vcpu, &vcpu->arch.vpa, &old_vpa);
741 		if (old_vpa.pinned_addr) {
742 			if (kvmhv_is_nestedv2())
743 				kvmhv_nestedv2_set_vpa(vcpu, ~0ull);
744 			kvmppc_unpin_guest_page(kvm, old_vpa.pinned_addr, old_vpa.gpa,
745 						old_vpa.dirty);
746 		}
747 		if (vcpu->arch.vpa.pinned_addr) {
748 			init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
749 			if (kvmhv_is_nestedv2())
750 				kvmhv_nestedv2_set_vpa(vcpu, __pa(vcpu->arch.vpa.pinned_addr));
751 		}
752 	}
753 	if (vcpu->arch.dtl.update_pending) {
754 		kvmppc_update_vpa(vcpu, &vcpu->arch.dtl, &old_vpa);
755 		if (old_vpa.pinned_addr)
756 			kvmppc_unpin_guest_page(kvm, old_vpa.pinned_addr, old_vpa.gpa,
757 						old_vpa.dirty);
758 		vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
759 		vcpu->arch.dtl_index = 0;
760 	}
761 	if (vcpu->arch.slb_shadow.update_pending) {
762 		kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow, &old_vpa);
763 		if (old_vpa.pinned_addr)
764 			kvmppc_unpin_guest_page(kvm, old_vpa.pinned_addr, old_vpa.gpa,
765 						old_vpa.dirty);
766 	}
767 
768 	spin_unlock(&vcpu->arch.vpa_update_lock);
769 }
770 
771 /*
772  * Return the accumulated stolen time for the vcore up until `now'.
773  * The caller should hold the vcore lock.
774  */
775 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
776 {
777 	u64 p;
778 	unsigned long flags;
779 
780 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
781 
782 	spin_lock_irqsave(&vc->stoltb_lock, flags);
783 	p = vc->stolen_tb;
784 	if (vc->vcore_state != VCORE_INACTIVE &&
785 	    vc->preempt_tb != TB_NIL)
786 		p += now - vc->preempt_tb;
787 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
788 	return p;
789 }
790 
791 static void __kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
792 					struct lppaca *vpa,
793 					unsigned int pcpu, u64 now,
794 					unsigned long stolen)
795 {
796 	struct dtl_entry *dt;
797 
798 	dt = vcpu->arch.dtl_ptr;
799 
800 	if (!dt)
801 		return;
802 
803 	dt->dispatch_reason = 7;
804 	dt->preempt_reason = 0;
805 	dt->processor_id = cpu_to_be16(pcpu + vcpu->arch.ptid);
806 	dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
807 	dt->ready_to_enqueue_time = 0;
808 	dt->waiting_to_ready_time = 0;
809 	dt->timebase = cpu_to_be64(now);
810 	dt->fault_addr = 0;
811 	dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
812 	dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
813 
814 	++dt;
815 	if (dt == vcpu->arch.dtl.pinned_end)
816 		dt = vcpu->arch.dtl.pinned_addr;
817 	vcpu->arch.dtl_ptr = dt;
818 	/* order writing *dt vs. writing vpa->dtl_idx */
819 	smp_wmb();
820 	vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
821 
822 	/* vcpu->arch.dtl.dirty is set by the caller */
823 }
824 
825 static void kvmppc_update_vpa_dispatch(struct kvm_vcpu *vcpu,
826 				       struct kvmppc_vcore *vc)
827 {
828 	struct lppaca *vpa;
829 	unsigned long stolen;
830 	unsigned long core_stolen;
831 	u64 now;
832 	unsigned long flags;
833 
834 	vpa = vcpu->arch.vpa.pinned_addr;
835 	if (!vpa)
836 		return;
837 
838 	now = mftb();
839 
840 	core_stolen = vcore_stolen_time(vc, now);
841 	stolen = core_stolen - vcpu->arch.stolen_logged;
842 	vcpu->arch.stolen_logged = core_stolen;
843 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
844 	stolen += vcpu->arch.busy_stolen;
845 	vcpu->arch.busy_stolen = 0;
846 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
847 
848 	vpa->enqueue_dispatch_tb = cpu_to_be64(be64_to_cpu(vpa->enqueue_dispatch_tb) + stolen);
849 
850 	__kvmppc_create_dtl_entry(vcpu, vpa, vc->pcpu, now + kvmppc_get_tb_offset(vcpu), stolen);
851 
852 	vcpu->arch.vpa.dirty = true;
853 }
854 
855 static void kvmppc_update_vpa_dispatch_p9(struct kvm_vcpu *vcpu,
856 				       struct kvmppc_vcore *vc,
857 				       u64 now)
858 {
859 	struct lppaca *vpa;
860 	unsigned long stolen;
861 	unsigned long stolen_delta;
862 
863 	vpa = vcpu->arch.vpa.pinned_addr;
864 	if (!vpa)
865 		return;
866 
867 	stolen = vc->stolen_tb;
868 	stolen_delta = stolen - vcpu->arch.stolen_logged;
869 	vcpu->arch.stolen_logged = stolen;
870 
871 	vpa->enqueue_dispatch_tb = cpu_to_be64(stolen);
872 
873 	__kvmppc_create_dtl_entry(vcpu, vpa, vc->pcpu, now, stolen_delta);
874 
875 	vcpu->arch.vpa.dirty = true;
876 }
877 
878 /* See if there is a doorbell interrupt pending for a vcpu */
879 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
880 {
881 	int thr;
882 	struct kvmppc_vcore *vc;
883 
884 	if (vcpu->arch.doorbell_request)
885 		return true;
886 	if (cpu_has_feature(CPU_FTR_ARCH_300))
887 		return false;
888 	/*
889 	 * Ensure that the read of vcore->dpdes comes after the read
890 	 * of vcpu->doorbell_request.  This barrier matches the
891 	 * smp_wmb() in kvmppc_guest_entry_inject().
892 	 */
893 	smp_rmb();
894 	vc = vcpu->arch.vcore;
895 	thr = vcpu->vcpu_id - vc->first_vcpuid;
896 	return !!(vc->dpdes & (1 << thr));
897 }
898 
899 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
900 {
901 	if (kvmppc_get_arch_compat(vcpu) >= PVR_ARCH_207)
902 		return true;
903 	if ((!kvmppc_get_arch_compat(vcpu)) &&
904 	    cpu_has_feature(CPU_FTR_ARCH_207S))
905 		return true;
906 	return false;
907 }
908 
909 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
910 			     unsigned long resource, unsigned long value1,
911 			     unsigned long value2)
912 {
913 	switch (resource) {
914 	case H_SET_MODE_RESOURCE_SET_CIABR:
915 		if (!kvmppc_power8_compatible(vcpu))
916 			return H_P2;
917 		if (value2)
918 			return H_P4;
919 		if (mflags)
920 			return H_UNSUPPORTED_FLAG_START;
921 		/* Guests can't breakpoint the hypervisor */
922 		if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
923 			return H_P3;
924 		kvmppc_set_ciabr_hv(vcpu, value1);
925 		return H_SUCCESS;
926 	case H_SET_MODE_RESOURCE_SET_DAWR0:
927 		if (!kvmppc_power8_compatible(vcpu))
928 			return H_P2;
929 		if (!ppc_breakpoint_available())
930 			return H_P2;
931 		if (mflags)
932 			return H_UNSUPPORTED_FLAG_START;
933 		if (value2 & DABRX_HYP)
934 			return H_P4;
935 		kvmppc_set_dawr0_hv(vcpu, value1);
936 		kvmppc_set_dawrx0_hv(vcpu, value2);
937 		return H_SUCCESS;
938 	case H_SET_MODE_RESOURCE_SET_DAWR1:
939 		if (!kvmppc_power8_compatible(vcpu))
940 			return H_P2;
941 		if (!ppc_breakpoint_available())
942 			return H_P2;
943 		if (!cpu_has_feature(CPU_FTR_DAWR1))
944 			return H_P2;
945 		if (!vcpu->kvm->arch.dawr1_enabled)
946 			return H_FUNCTION;
947 		if (mflags)
948 			return H_UNSUPPORTED_FLAG_START;
949 		if (value2 & DABRX_HYP)
950 			return H_P4;
951 		kvmppc_set_dawr1_hv(vcpu, value1);
952 		kvmppc_set_dawrx1_hv(vcpu, value2);
953 		return H_SUCCESS;
954 	case H_SET_MODE_RESOURCE_ADDR_TRANS_MODE:
955 		/*
956 		 * KVM does not support mflags=2 (AIL=2) and AIL=1 is reserved.
957 		 * Keep this in synch with kvmppc_filter_guest_lpcr_hv.
958 		 */
959 		if (cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG) &&
960 				kvmhv_vcpu_is_radix(vcpu) && mflags == 3)
961 			return H_UNSUPPORTED_FLAG_START;
962 		return H_TOO_HARD;
963 	default:
964 		return H_TOO_HARD;
965 	}
966 }
967 
968 /* Copy guest memory in place - must reside within a single memslot */
969 static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from,
970 				  unsigned long len)
971 {
972 	struct kvm_memory_slot *to_memslot = NULL;
973 	struct kvm_memory_slot *from_memslot = NULL;
974 	unsigned long to_addr, from_addr;
975 	int r;
976 
977 	/* Get HPA for from address */
978 	from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT);
979 	if (!from_memslot)
980 		return -EFAULT;
981 	if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages)
982 			     << PAGE_SHIFT))
983 		return -EINVAL;
984 	from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT);
985 	if (kvm_is_error_hva(from_addr))
986 		return -EFAULT;
987 	from_addr |= (from & (PAGE_SIZE - 1));
988 
989 	/* Get HPA for to address */
990 	to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT);
991 	if (!to_memslot)
992 		return -EFAULT;
993 	if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages)
994 			   << PAGE_SHIFT))
995 		return -EINVAL;
996 	to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT);
997 	if (kvm_is_error_hva(to_addr))
998 		return -EFAULT;
999 	to_addr |= (to & (PAGE_SIZE - 1));
1000 
1001 	/* Perform copy */
1002 	r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr,
1003 			     len);
1004 	if (r)
1005 		return -EFAULT;
1006 	mark_page_dirty(kvm, to >> PAGE_SHIFT);
1007 	return 0;
1008 }
1009 
1010 static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags,
1011 			       unsigned long dest, unsigned long src)
1012 {
1013 	u64 pg_sz = SZ_4K;		/* 4K page size */
1014 	u64 pg_mask = SZ_4K - 1;
1015 	int ret;
1016 
1017 	/* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */
1018 	if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE |
1019 		      H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED))
1020 		return H_PARAMETER;
1021 
1022 	/* dest (and src if copy_page flag set) must be page aligned */
1023 	if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask)))
1024 		return H_PARAMETER;
1025 
1026 	/* zero and/or copy the page as determined by the flags */
1027 	if (flags & H_COPY_PAGE) {
1028 		ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz);
1029 		if (ret < 0)
1030 			return H_PARAMETER;
1031 	} else if (flags & H_ZERO_PAGE) {
1032 		ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz);
1033 		if (ret < 0)
1034 			return H_PARAMETER;
1035 	}
1036 
1037 	/* We can ignore the remaining flags */
1038 
1039 	return H_SUCCESS;
1040 }
1041 
1042 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
1043 {
1044 	struct kvmppc_vcore *vcore = target->arch.vcore;
1045 
1046 	/*
1047 	 * We expect to have been called by the real mode handler
1048 	 * (kvmppc_rm_h_confer()) which would have directly returned
1049 	 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
1050 	 * have useful work to do and should not confer) so we don't
1051 	 * recheck that here.
1052 	 *
1053 	 * In the case of the P9 single vcpu per vcore case, the real
1054 	 * mode handler is not called but no other threads are in the
1055 	 * source vcore.
1056 	 */
1057 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
1058 		spin_lock(&vcore->lock);
1059 		if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
1060 		    vcore->vcore_state != VCORE_INACTIVE &&
1061 		    vcore->runner)
1062 			target = vcore->runner;
1063 		spin_unlock(&vcore->lock);
1064 	}
1065 
1066 	return kvm_vcpu_yield_to(target);
1067 }
1068 
1069 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
1070 {
1071 	int yield_count = 0;
1072 	struct lppaca *lppaca;
1073 
1074 	spin_lock(&vcpu->arch.vpa_update_lock);
1075 	lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
1076 	if (lppaca)
1077 		yield_count = be32_to_cpu(lppaca->yield_count);
1078 	spin_unlock(&vcpu->arch.vpa_update_lock);
1079 	return yield_count;
1080 }
1081 
1082 /*
1083  * H_RPT_INVALIDATE hcall handler for nested guests.
1084  *
1085  * Handles only nested process-scoped invalidation requests in L0.
1086  */
1087 static int kvmppc_nested_h_rpt_invalidate(struct kvm_vcpu *vcpu)
1088 {
1089 	unsigned long type = kvmppc_get_gpr(vcpu, 6);
1090 	unsigned long pid, pg_sizes, start, end;
1091 
1092 	/*
1093 	 * The partition-scoped invalidations aren't handled here in L0.
1094 	 */
1095 	if (type & H_RPTI_TYPE_NESTED)
1096 		return RESUME_HOST;
1097 
1098 	pid = kvmppc_get_gpr(vcpu, 4);
1099 	pg_sizes = kvmppc_get_gpr(vcpu, 7);
1100 	start = kvmppc_get_gpr(vcpu, 8);
1101 	end = kvmppc_get_gpr(vcpu, 9);
1102 
1103 	do_h_rpt_invalidate_prt(pid, vcpu->arch.nested->shadow_lpid,
1104 				type, pg_sizes, start, end);
1105 
1106 	kvmppc_set_gpr(vcpu, 3, H_SUCCESS);
1107 	return RESUME_GUEST;
1108 }
1109 
1110 static long kvmppc_h_rpt_invalidate(struct kvm_vcpu *vcpu,
1111 				    unsigned long id, unsigned long target,
1112 				    unsigned long type, unsigned long pg_sizes,
1113 				    unsigned long start, unsigned long end)
1114 {
1115 	if (!kvm_is_radix(vcpu->kvm))
1116 		return H_UNSUPPORTED;
1117 
1118 	if (end < start)
1119 		return H_P5;
1120 
1121 	/*
1122 	 * Partition-scoped invalidation for nested guests.
1123 	 */
1124 	if (type & H_RPTI_TYPE_NESTED) {
1125 		if (!nesting_enabled(vcpu->kvm))
1126 			return H_FUNCTION;
1127 
1128 		/* Support only cores as target */
1129 		if (target != H_RPTI_TARGET_CMMU)
1130 			return H_P2;
1131 
1132 		return do_h_rpt_invalidate_pat(vcpu, id, type, pg_sizes,
1133 					       start, end);
1134 	}
1135 
1136 	/*
1137 	 * Process-scoped invalidation for L1 guests.
1138 	 */
1139 	do_h_rpt_invalidate_prt(id, vcpu->kvm->arch.lpid,
1140 				type, pg_sizes, start, end);
1141 	return H_SUCCESS;
1142 }
1143 
1144 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
1145 {
1146 	struct kvm *kvm = vcpu->kvm;
1147 	unsigned long req = kvmppc_get_gpr(vcpu, 3);
1148 	unsigned long target, ret = H_SUCCESS;
1149 	int yield_count;
1150 	struct kvm_vcpu *tvcpu;
1151 	int idx, rc;
1152 
1153 	if (req <= MAX_HCALL_OPCODE &&
1154 	    !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
1155 		return RESUME_HOST;
1156 
1157 	switch (req) {
1158 	case H_REMOVE:
1159 		ret = kvmppc_h_remove(vcpu, kvmppc_get_gpr(vcpu, 4),
1160 					kvmppc_get_gpr(vcpu, 5),
1161 					kvmppc_get_gpr(vcpu, 6));
1162 		if (ret == H_TOO_HARD)
1163 			return RESUME_HOST;
1164 		break;
1165 	case H_ENTER:
1166 		ret = kvmppc_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
1167 					kvmppc_get_gpr(vcpu, 5),
1168 					kvmppc_get_gpr(vcpu, 6),
1169 					kvmppc_get_gpr(vcpu, 7));
1170 		if (ret == H_TOO_HARD)
1171 			return RESUME_HOST;
1172 		break;
1173 	case H_READ:
1174 		ret = kvmppc_h_read(vcpu, kvmppc_get_gpr(vcpu, 4),
1175 					kvmppc_get_gpr(vcpu, 5));
1176 		if (ret == H_TOO_HARD)
1177 			return RESUME_HOST;
1178 		break;
1179 	case H_CLEAR_MOD:
1180 		ret = kvmppc_h_clear_mod(vcpu, kvmppc_get_gpr(vcpu, 4),
1181 					kvmppc_get_gpr(vcpu, 5));
1182 		if (ret == H_TOO_HARD)
1183 			return RESUME_HOST;
1184 		break;
1185 	case H_CLEAR_REF:
1186 		ret = kvmppc_h_clear_ref(vcpu, kvmppc_get_gpr(vcpu, 4),
1187 					kvmppc_get_gpr(vcpu, 5));
1188 		if (ret == H_TOO_HARD)
1189 			return RESUME_HOST;
1190 		break;
1191 	case H_PROTECT:
1192 		ret = kvmppc_h_protect(vcpu, kvmppc_get_gpr(vcpu, 4),
1193 					kvmppc_get_gpr(vcpu, 5),
1194 					kvmppc_get_gpr(vcpu, 6));
1195 		if (ret == H_TOO_HARD)
1196 			return RESUME_HOST;
1197 		break;
1198 	case H_BULK_REMOVE:
1199 		ret = kvmppc_h_bulk_remove(vcpu);
1200 		if (ret == H_TOO_HARD)
1201 			return RESUME_HOST;
1202 		break;
1203 
1204 	case H_CEDE:
1205 		break;
1206 	case H_PROD:
1207 		target = kvmppc_get_gpr(vcpu, 4);
1208 		tvcpu = kvmppc_find_vcpu(kvm, target);
1209 		if (!tvcpu) {
1210 			ret = H_PARAMETER;
1211 			break;
1212 		}
1213 		tvcpu->arch.prodded = 1;
1214 		smp_mb(); /* This orders prodded store vs ceded load */
1215 		if (tvcpu->arch.ceded)
1216 			kvmppc_fast_vcpu_kick_hv(tvcpu);
1217 		break;
1218 	case H_CONFER:
1219 		target = kvmppc_get_gpr(vcpu, 4);
1220 		if (target == -1)
1221 			break;
1222 		tvcpu = kvmppc_find_vcpu(kvm, target);
1223 		if (!tvcpu) {
1224 			ret = H_PARAMETER;
1225 			break;
1226 		}
1227 		yield_count = kvmppc_get_gpr(vcpu, 5);
1228 		if (kvmppc_get_yield_count(tvcpu) != yield_count)
1229 			break;
1230 		kvm_arch_vcpu_yield_to(tvcpu);
1231 		break;
1232 	case H_REGISTER_VPA:
1233 		ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
1234 					kvmppc_get_gpr(vcpu, 5),
1235 					kvmppc_get_gpr(vcpu, 6));
1236 		break;
1237 	case H_RTAS:
1238 		if (list_empty(&kvm->arch.rtas_tokens))
1239 			return RESUME_HOST;
1240 
1241 		idx = srcu_read_lock(&kvm->srcu);
1242 		rc = kvmppc_rtas_hcall(vcpu);
1243 		srcu_read_unlock(&kvm->srcu, idx);
1244 
1245 		if (rc == -ENOENT)
1246 			return RESUME_HOST;
1247 		else if (rc == 0)
1248 			break;
1249 
1250 		/* Send the error out to userspace via KVM_RUN */
1251 		return rc;
1252 	case H_LOGICAL_CI_LOAD:
1253 		ret = kvmppc_h_logical_ci_load(vcpu);
1254 		if (ret == H_TOO_HARD)
1255 			return RESUME_HOST;
1256 		break;
1257 	case H_LOGICAL_CI_STORE:
1258 		ret = kvmppc_h_logical_ci_store(vcpu);
1259 		if (ret == H_TOO_HARD)
1260 			return RESUME_HOST;
1261 		break;
1262 	case H_SET_MODE:
1263 		ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
1264 					kvmppc_get_gpr(vcpu, 5),
1265 					kvmppc_get_gpr(vcpu, 6),
1266 					kvmppc_get_gpr(vcpu, 7));
1267 		if (ret == H_TOO_HARD)
1268 			return RESUME_HOST;
1269 		break;
1270 	case H_XIRR:
1271 	case H_CPPR:
1272 	case H_EOI:
1273 	case H_IPI:
1274 	case H_IPOLL:
1275 	case H_XIRR_X:
1276 		if (kvmppc_xics_enabled(vcpu)) {
1277 			if (xics_on_xive()) {
1278 				ret = H_NOT_AVAILABLE;
1279 				return RESUME_GUEST;
1280 			}
1281 			ret = kvmppc_xics_hcall(vcpu, req);
1282 			break;
1283 		}
1284 		return RESUME_HOST;
1285 	case H_SET_DABR:
1286 		ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
1287 		break;
1288 	case H_SET_XDABR:
1289 		ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
1290 						kvmppc_get_gpr(vcpu, 5));
1291 		break;
1292 #ifdef CONFIG_SPAPR_TCE_IOMMU
1293 	case H_GET_TCE:
1294 		ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1295 						kvmppc_get_gpr(vcpu, 5));
1296 		if (ret == H_TOO_HARD)
1297 			return RESUME_HOST;
1298 		break;
1299 	case H_PUT_TCE:
1300 		ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1301 						kvmppc_get_gpr(vcpu, 5),
1302 						kvmppc_get_gpr(vcpu, 6));
1303 		if (ret == H_TOO_HARD)
1304 			return RESUME_HOST;
1305 		break;
1306 	case H_PUT_TCE_INDIRECT:
1307 		ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
1308 						kvmppc_get_gpr(vcpu, 5),
1309 						kvmppc_get_gpr(vcpu, 6),
1310 						kvmppc_get_gpr(vcpu, 7));
1311 		if (ret == H_TOO_HARD)
1312 			return RESUME_HOST;
1313 		break;
1314 	case H_STUFF_TCE:
1315 		ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1316 						kvmppc_get_gpr(vcpu, 5),
1317 						kvmppc_get_gpr(vcpu, 6),
1318 						kvmppc_get_gpr(vcpu, 7));
1319 		if (ret == H_TOO_HARD)
1320 			return RESUME_HOST;
1321 		break;
1322 #endif
1323 	case H_RANDOM: {
1324 		unsigned long rand;
1325 
1326 		if (!arch_get_random_seed_longs(&rand, 1))
1327 			ret = H_HARDWARE;
1328 		kvmppc_set_gpr(vcpu, 4, rand);
1329 		break;
1330 	}
1331 	case H_RPT_INVALIDATE:
1332 		ret = kvmppc_h_rpt_invalidate(vcpu, kvmppc_get_gpr(vcpu, 4),
1333 					      kvmppc_get_gpr(vcpu, 5),
1334 					      kvmppc_get_gpr(vcpu, 6),
1335 					      kvmppc_get_gpr(vcpu, 7),
1336 					      kvmppc_get_gpr(vcpu, 8),
1337 					      kvmppc_get_gpr(vcpu, 9));
1338 		break;
1339 
1340 	case H_SET_PARTITION_TABLE:
1341 		ret = H_FUNCTION;
1342 		if (nesting_enabled(kvm))
1343 			ret = kvmhv_set_partition_table(vcpu);
1344 		break;
1345 	case H_ENTER_NESTED:
1346 		ret = H_FUNCTION;
1347 		if (!nesting_enabled(kvm))
1348 			break;
1349 		ret = kvmhv_enter_nested_guest(vcpu);
1350 		if (ret == H_INTERRUPT) {
1351 			kvmppc_set_gpr(vcpu, 3, 0);
1352 			vcpu->arch.hcall_needed = 0;
1353 			return -EINTR;
1354 		} else if (ret == H_TOO_HARD) {
1355 			kvmppc_set_gpr(vcpu, 3, 0);
1356 			vcpu->arch.hcall_needed = 0;
1357 			return RESUME_HOST;
1358 		}
1359 		break;
1360 	case H_TLB_INVALIDATE:
1361 		ret = H_FUNCTION;
1362 		if (nesting_enabled(kvm))
1363 			ret = kvmhv_do_nested_tlbie(vcpu);
1364 		break;
1365 	case H_COPY_TOFROM_GUEST:
1366 		ret = H_FUNCTION;
1367 		if (nesting_enabled(kvm))
1368 			ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1369 		break;
1370 	case H_PAGE_INIT:
1371 		ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4),
1372 					 kvmppc_get_gpr(vcpu, 5),
1373 					 kvmppc_get_gpr(vcpu, 6));
1374 		break;
1375 	case H_SVM_PAGE_IN:
1376 		ret = H_UNSUPPORTED;
1377 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1378 			ret = kvmppc_h_svm_page_in(kvm,
1379 						   kvmppc_get_gpr(vcpu, 4),
1380 						   kvmppc_get_gpr(vcpu, 5),
1381 						   kvmppc_get_gpr(vcpu, 6));
1382 		break;
1383 	case H_SVM_PAGE_OUT:
1384 		ret = H_UNSUPPORTED;
1385 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1386 			ret = kvmppc_h_svm_page_out(kvm,
1387 						    kvmppc_get_gpr(vcpu, 4),
1388 						    kvmppc_get_gpr(vcpu, 5),
1389 						    kvmppc_get_gpr(vcpu, 6));
1390 		break;
1391 	case H_SVM_INIT_START:
1392 		ret = H_UNSUPPORTED;
1393 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1394 			ret = kvmppc_h_svm_init_start(kvm);
1395 		break;
1396 	case H_SVM_INIT_DONE:
1397 		ret = H_UNSUPPORTED;
1398 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1399 			ret = kvmppc_h_svm_init_done(kvm);
1400 		break;
1401 	case H_SVM_INIT_ABORT:
1402 		/*
1403 		 * Even if that call is made by the Ultravisor, the SSR1 value
1404 		 * is the guest context one, with the secure bit clear as it has
1405 		 * not yet been secured. So we can't check it here.
1406 		 * Instead the kvm->arch.secure_guest flag is checked inside
1407 		 * kvmppc_h_svm_init_abort().
1408 		 */
1409 		ret = kvmppc_h_svm_init_abort(kvm);
1410 		break;
1411 
1412 	default:
1413 		return RESUME_HOST;
1414 	}
1415 	WARN_ON_ONCE(ret == H_TOO_HARD);
1416 	kvmppc_set_gpr(vcpu, 3, ret);
1417 	vcpu->arch.hcall_needed = 0;
1418 	return RESUME_GUEST;
1419 }
1420 
1421 /*
1422  * Handle H_CEDE in the P9 path where we don't call the real-mode hcall
1423  * handlers in book3s_hv_rmhandlers.S.
1424  *
1425  * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1426  * that the cede logic in kvmppc_run_single_vcpu() works properly.
1427  */
1428 static void kvmppc_cede(struct kvm_vcpu *vcpu)
1429 {
1430 	__kvmppc_set_msr_hv(vcpu, __kvmppc_get_msr_hv(vcpu) | MSR_EE);
1431 	vcpu->arch.ceded = 1;
1432 	smp_mb();
1433 	if (vcpu->arch.prodded) {
1434 		vcpu->arch.prodded = 0;
1435 		smp_mb();
1436 		vcpu->arch.ceded = 0;
1437 	}
1438 }
1439 
1440 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1441 {
1442 	switch (cmd) {
1443 	case H_CEDE:
1444 	case H_PROD:
1445 	case H_CONFER:
1446 	case H_REGISTER_VPA:
1447 	case H_SET_MODE:
1448 #ifdef CONFIG_SPAPR_TCE_IOMMU
1449 	case H_GET_TCE:
1450 	case H_PUT_TCE:
1451 	case H_PUT_TCE_INDIRECT:
1452 	case H_STUFF_TCE:
1453 #endif
1454 	case H_LOGICAL_CI_LOAD:
1455 	case H_LOGICAL_CI_STORE:
1456 #ifdef CONFIG_KVM_XICS
1457 	case H_XIRR:
1458 	case H_CPPR:
1459 	case H_EOI:
1460 	case H_IPI:
1461 	case H_IPOLL:
1462 	case H_XIRR_X:
1463 #endif
1464 	case H_PAGE_INIT:
1465 	case H_RPT_INVALIDATE:
1466 		return 1;
1467 	}
1468 
1469 	/* See if it's in the real-mode table */
1470 	return kvmppc_hcall_impl_hv_realmode(cmd);
1471 }
1472 
1473 static int kvmppc_emulate_debug_inst(struct kvm_vcpu *vcpu)
1474 {
1475 	ppc_inst_t last_inst;
1476 
1477 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1478 					EMULATE_DONE) {
1479 		/*
1480 		 * Fetch failed, so return to guest and
1481 		 * try executing it again.
1482 		 */
1483 		return RESUME_GUEST;
1484 	}
1485 
1486 	if (ppc_inst_val(last_inst) == KVMPPC_INST_SW_BREAKPOINT) {
1487 		vcpu->run->exit_reason = KVM_EXIT_DEBUG;
1488 		vcpu->run->debug.arch.address = kvmppc_get_pc(vcpu);
1489 		return RESUME_HOST;
1490 	} else {
1491 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL |
1492 				(kvmppc_get_msr(vcpu) & SRR1_PREFIXED));
1493 		return RESUME_GUEST;
1494 	}
1495 }
1496 
1497 static void do_nothing(void *x)
1498 {
1499 }
1500 
1501 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1502 {
1503 	int thr, cpu, pcpu, nthreads;
1504 	struct kvm_vcpu *v;
1505 	unsigned long dpdes;
1506 
1507 	nthreads = vcpu->kvm->arch.emul_smt_mode;
1508 	dpdes = 0;
1509 	cpu = vcpu->vcpu_id & ~(nthreads - 1);
1510 	for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1511 		v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1512 		if (!v)
1513 			continue;
1514 		/*
1515 		 * If the vcpu is currently running on a physical cpu thread,
1516 		 * interrupt it in order to pull it out of the guest briefly,
1517 		 * which will update its vcore->dpdes value.
1518 		 */
1519 		pcpu = READ_ONCE(v->cpu);
1520 		if (pcpu >= 0)
1521 			smp_call_function_single(pcpu, do_nothing, NULL, 1);
1522 		if (kvmppc_doorbell_pending(v))
1523 			dpdes |= 1 << thr;
1524 	}
1525 	return dpdes;
1526 }
1527 
1528 /*
1529  * On POWER9, emulate doorbell-related instructions in order to
1530  * give the guest the illusion of running on a multi-threaded core.
1531  * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1532  * and mfspr DPDES.
1533  */
1534 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1535 {
1536 	u32 inst, rb, thr;
1537 	unsigned long arg;
1538 	struct kvm *kvm = vcpu->kvm;
1539 	struct kvm_vcpu *tvcpu;
1540 	ppc_inst_t pinst;
1541 
1542 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &pinst) != EMULATE_DONE)
1543 		return RESUME_GUEST;
1544 	inst = ppc_inst_val(pinst);
1545 	if (get_op(inst) != 31)
1546 		return EMULATE_FAIL;
1547 	rb = get_rb(inst);
1548 	thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1549 	switch (get_xop(inst)) {
1550 	case OP_31_XOP_MSGSNDP:
1551 		arg = kvmppc_get_gpr(vcpu, rb);
1552 		if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1553 			break;
1554 		arg &= 0x7f;
1555 		if (arg >= kvm->arch.emul_smt_mode)
1556 			break;
1557 		tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1558 		if (!tvcpu)
1559 			break;
1560 		if (!tvcpu->arch.doorbell_request) {
1561 			tvcpu->arch.doorbell_request = 1;
1562 			kvmppc_fast_vcpu_kick_hv(tvcpu);
1563 		}
1564 		break;
1565 	case OP_31_XOP_MSGCLRP:
1566 		arg = kvmppc_get_gpr(vcpu, rb);
1567 		if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1568 			break;
1569 		vcpu->arch.vcore->dpdes = 0;
1570 		vcpu->arch.doorbell_request = 0;
1571 		break;
1572 	case OP_31_XOP_MFSPR:
1573 		switch (get_sprn(inst)) {
1574 		case SPRN_TIR:
1575 			arg = thr;
1576 			break;
1577 		case SPRN_DPDES:
1578 			arg = kvmppc_read_dpdes(vcpu);
1579 			break;
1580 		default:
1581 			return EMULATE_FAIL;
1582 		}
1583 		kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1584 		break;
1585 	default:
1586 		return EMULATE_FAIL;
1587 	}
1588 	kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1589 	return RESUME_GUEST;
1590 }
1591 
1592 /*
1593  * If the lppaca had pmcregs_in_use clear when we exited the guest, then
1594  * HFSCR_PM is cleared for next entry. If the guest then tries to access
1595  * the PMU SPRs, we get this facility unavailable interrupt. Putting HFSCR_PM
1596  * back in the guest HFSCR will cause the next entry to load the PMU SPRs and
1597  * allow the guest access to continue.
1598  */
1599 static int kvmppc_pmu_unavailable(struct kvm_vcpu *vcpu)
1600 {
1601 	if (!(vcpu->arch.hfscr_permitted & HFSCR_PM))
1602 		return EMULATE_FAIL;
1603 
1604 	kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) | HFSCR_PM);
1605 
1606 	return RESUME_GUEST;
1607 }
1608 
1609 static int kvmppc_ebb_unavailable(struct kvm_vcpu *vcpu)
1610 {
1611 	if (!(vcpu->arch.hfscr_permitted & HFSCR_EBB))
1612 		return EMULATE_FAIL;
1613 
1614 	kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) | HFSCR_EBB);
1615 
1616 	return RESUME_GUEST;
1617 }
1618 
1619 static int kvmppc_tm_unavailable(struct kvm_vcpu *vcpu)
1620 {
1621 	if (!(vcpu->arch.hfscr_permitted & HFSCR_TM))
1622 		return EMULATE_FAIL;
1623 
1624 	kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) | HFSCR_TM);
1625 
1626 	return RESUME_GUEST;
1627 }
1628 
1629 static int kvmppc_handle_exit_hv(struct kvm_vcpu *vcpu,
1630 				 struct task_struct *tsk)
1631 {
1632 	struct kvm_run *run = vcpu->run;
1633 	int r = RESUME_HOST;
1634 
1635 	vcpu->stat.sum_exits++;
1636 
1637 	/*
1638 	 * This can happen if an interrupt occurs in the last stages
1639 	 * of guest entry or the first stages of guest exit (i.e. after
1640 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1641 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1642 	 * That can happen due to a bug, or due to a machine check
1643 	 * occurring at just the wrong time.
1644 	 */
1645 	if (!kvmhv_is_nestedv2() && (__kvmppc_get_msr_hv(vcpu) & MSR_HV)) {
1646 		printk(KERN_EMERG "KVM trap in HV mode!\n");
1647 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1648 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1649 			vcpu->arch.shregs.msr);
1650 		kvmppc_dump_regs(vcpu);
1651 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1652 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1653 		return RESUME_HOST;
1654 	}
1655 	run->exit_reason = KVM_EXIT_UNKNOWN;
1656 	run->ready_for_interrupt_injection = 1;
1657 	switch (vcpu->arch.trap) {
1658 	/* We're good on these - the host merely wanted to get our attention */
1659 	case BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER:
1660 		WARN_ON_ONCE(1); /* Should never happen */
1661 		vcpu->arch.trap = BOOK3S_INTERRUPT_HV_DECREMENTER;
1662 		fallthrough;
1663 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1664 		vcpu->stat.dec_exits++;
1665 		r = RESUME_GUEST;
1666 		break;
1667 	case BOOK3S_INTERRUPT_EXTERNAL:
1668 	case BOOK3S_INTERRUPT_H_DOORBELL:
1669 	case BOOK3S_INTERRUPT_H_VIRT:
1670 		vcpu->stat.ext_intr_exits++;
1671 		r = RESUME_GUEST;
1672 		break;
1673 	/* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1674 	case BOOK3S_INTERRUPT_HMI:
1675 	case BOOK3S_INTERRUPT_PERFMON:
1676 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
1677 		r = RESUME_GUEST;
1678 		break;
1679 	case BOOK3S_INTERRUPT_MACHINE_CHECK: {
1680 		static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
1681 					      DEFAULT_RATELIMIT_BURST);
1682 		/*
1683 		 * Print the MCE event to host console. Ratelimit so the guest
1684 		 * can't flood the host log.
1685 		 */
1686 		if (__ratelimit(&rs))
1687 			machine_check_print_event_info(&vcpu->arch.mce_evt,false, true);
1688 
1689 		/*
1690 		 * If the guest can do FWNMI, exit to userspace so it can
1691 		 * deliver a FWNMI to the guest.
1692 		 * Otherwise we synthesize a machine check for the guest
1693 		 * so that it knows that the machine check occurred.
1694 		 */
1695 		if (!vcpu->kvm->arch.fwnmi_enabled) {
1696 			ulong flags = (__kvmppc_get_msr_hv(vcpu) & 0x083c0000) |
1697 					(kvmppc_get_msr(vcpu) & SRR1_PREFIXED);
1698 			kvmppc_core_queue_machine_check(vcpu, flags);
1699 			r = RESUME_GUEST;
1700 			break;
1701 		}
1702 
1703 		/* Exit to guest with KVM_EXIT_NMI as exit reason */
1704 		run->exit_reason = KVM_EXIT_NMI;
1705 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1706 		/* Clear out the old NMI status from run->flags */
1707 		run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1708 		/* Now set the NMI status */
1709 		if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1710 			run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1711 		else
1712 			run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1713 
1714 		r = RESUME_HOST;
1715 		break;
1716 	}
1717 	case BOOK3S_INTERRUPT_PROGRAM:
1718 	{
1719 		ulong flags;
1720 		/*
1721 		 * Normally program interrupts are delivered directly
1722 		 * to the guest by the hardware, but we can get here
1723 		 * as a result of a hypervisor emulation interrupt
1724 		 * (e40) getting turned into a 700 by BML RTAS.
1725 		 */
1726 		flags = (__kvmppc_get_msr_hv(vcpu) & 0x1f0000ull) |
1727 			(kvmppc_get_msr(vcpu) & SRR1_PREFIXED);
1728 		kvmppc_core_queue_program(vcpu, flags);
1729 		r = RESUME_GUEST;
1730 		break;
1731 	}
1732 	case BOOK3S_INTERRUPT_SYSCALL:
1733 	{
1734 		int i;
1735 
1736 		if (!kvmhv_is_nestedv2() && unlikely(__kvmppc_get_msr_hv(vcpu) & MSR_PR)) {
1737 			/*
1738 			 * Guest userspace executed sc 1. This can only be
1739 			 * reached by the P9 path because the old path
1740 			 * handles this case in realmode hcall handlers.
1741 			 */
1742 			if (!kvmhv_vcpu_is_radix(vcpu)) {
1743 				/*
1744 				 * A guest could be running PR KVM, so this
1745 				 * may be a PR KVM hcall. It must be reflected
1746 				 * to the guest kernel as a sc interrupt.
1747 				 */
1748 				kvmppc_core_queue_syscall(vcpu);
1749 			} else {
1750 				/*
1751 				 * Radix guests can not run PR KVM or nested HV
1752 				 * hash guests which might run PR KVM, so this
1753 				 * is always a privilege fault. Send a program
1754 				 * check to guest kernel.
1755 				 */
1756 				kvmppc_core_queue_program(vcpu, SRR1_PROGPRIV);
1757 			}
1758 			r = RESUME_GUEST;
1759 			break;
1760 		}
1761 
1762 		/*
1763 		 * hcall - gather args and set exit_reason. This will next be
1764 		 * handled by kvmppc_pseries_do_hcall which may be able to deal
1765 		 * with it and resume guest, or may punt to userspace.
1766 		 */
1767 		run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1768 		for (i = 0; i < 9; ++i)
1769 			run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1770 		run->exit_reason = KVM_EXIT_PAPR_HCALL;
1771 		vcpu->arch.hcall_needed = 1;
1772 		r = RESUME_HOST;
1773 		break;
1774 	}
1775 	/*
1776 	 * We get these next two if the guest accesses a page which it thinks
1777 	 * it has mapped but which is not actually present, either because
1778 	 * it is for an emulated I/O device or because the corresonding
1779 	 * host page has been paged out.
1780 	 *
1781 	 * Any other HDSI/HISI interrupts have been handled already for P7/8
1782 	 * guests. For POWER9 hash guests not using rmhandlers, basic hash
1783 	 * fault handling is done here.
1784 	 */
1785 	case BOOK3S_INTERRUPT_H_DATA_STORAGE: {
1786 		unsigned long vsid;
1787 		long err;
1788 
1789 		if (cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG) &&
1790 		    unlikely(vcpu->arch.fault_dsisr == HDSISR_CANARY)) {
1791 			r = RESUME_GUEST; /* Just retry if it's the canary */
1792 			break;
1793 		}
1794 
1795 		if (kvm_is_radix(vcpu->kvm) || !cpu_has_feature(CPU_FTR_ARCH_300)) {
1796 			/*
1797 			 * Radix doesn't require anything, and pre-ISAv3.0 hash
1798 			 * already attempted to handle this in rmhandlers. The
1799 			 * hash fault handling below is v3 only (it uses ASDR
1800 			 * via fault_gpa).
1801 			 */
1802 			r = RESUME_PAGE_FAULT;
1803 			break;
1804 		}
1805 
1806 		if (!(vcpu->arch.fault_dsisr & (DSISR_NOHPTE | DSISR_PROTFAULT))) {
1807 			kvmppc_core_queue_data_storage(vcpu,
1808 				kvmppc_get_msr(vcpu) & SRR1_PREFIXED,
1809 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1810 			r = RESUME_GUEST;
1811 			break;
1812 		}
1813 
1814 		if (!(__kvmppc_get_msr_hv(vcpu) & MSR_DR))
1815 			vsid = vcpu->kvm->arch.vrma_slb_v;
1816 		else
1817 			vsid = vcpu->arch.fault_gpa;
1818 
1819 		err = kvmppc_hpte_hv_fault(vcpu, vcpu->arch.fault_dar,
1820 				vsid, vcpu->arch.fault_dsisr, true);
1821 		if (err == 0) {
1822 			r = RESUME_GUEST;
1823 		} else if (err == -1 || err == -2) {
1824 			r = RESUME_PAGE_FAULT;
1825 		} else {
1826 			kvmppc_core_queue_data_storage(vcpu,
1827 				kvmppc_get_msr(vcpu) & SRR1_PREFIXED,
1828 				vcpu->arch.fault_dar, err);
1829 			r = RESUME_GUEST;
1830 		}
1831 		break;
1832 	}
1833 	case BOOK3S_INTERRUPT_H_INST_STORAGE: {
1834 		unsigned long vsid;
1835 		long err;
1836 
1837 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1838 		vcpu->arch.fault_dsisr = __kvmppc_get_msr_hv(vcpu) &
1839 			DSISR_SRR1_MATCH_64S;
1840 		if (kvm_is_radix(vcpu->kvm) || !cpu_has_feature(CPU_FTR_ARCH_300)) {
1841 			/*
1842 			 * Radix doesn't require anything, and pre-ISAv3.0 hash
1843 			 * already attempted to handle this in rmhandlers. The
1844 			 * hash fault handling below is v3 only (it uses ASDR
1845 			 * via fault_gpa).
1846 			 */
1847 			if (__kvmppc_get_msr_hv(vcpu) & HSRR1_HISI_WRITE)
1848 				vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1849 			r = RESUME_PAGE_FAULT;
1850 			break;
1851 		}
1852 
1853 		if (!(vcpu->arch.fault_dsisr & SRR1_ISI_NOPT)) {
1854 			kvmppc_core_queue_inst_storage(vcpu,
1855 				vcpu->arch.fault_dsisr |
1856 				(kvmppc_get_msr(vcpu) & SRR1_PREFIXED));
1857 			r = RESUME_GUEST;
1858 			break;
1859 		}
1860 
1861 		if (!(__kvmppc_get_msr_hv(vcpu) & MSR_IR))
1862 			vsid = vcpu->kvm->arch.vrma_slb_v;
1863 		else
1864 			vsid = vcpu->arch.fault_gpa;
1865 
1866 		err = kvmppc_hpte_hv_fault(vcpu, vcpu->arch.fault_dar,
1867 				vsid, vcpu->arch.fault_dsisr, false);
1868 		if (err == 0) {
1869 			r = RESUME_GUEST;
1870 		} else if (err == -1) {
1871 			r = RESUME_PAGE_FAULT;
1872 		} else {
1873 			kvmppc_core_queue_inst_storage(vcpu,
1874 				err | (kvmppc_get_msr(vcpu) & SRR1_PREFIXED));
1875 			r = RESUME_GUEST;
1876 		}
1877 		break;
1878 	}
1879 
1880 	/*
1881 	 * This occurs if the guest executes an illegal instruction.
1882 	 * If the guest debug is disabled, generate a program interrupt
1883 	 * to the guest. If guest debug is enabled, we need to check
1884 	 * whether the instruction is a software breakpoint instruction.
1885 	 * Accordingly return to Guest or Host.
1886 	 */
1887 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1888 		if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1889 			vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1890 				swab32(vcpu->arch.emul_inst) :
1891 				vcpu->arch.emul_inst;
1892 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1893 			r = kvmppc_emulate_debug_inst(vcpu);
1894 		} else {
1895 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL |
1896 				(kvmppc_get_msr(vcpu) & SRR1_PREFIXED));
1897 			r = RESUME_GUEST;
1898 		}
1899 		break;
1900 
1901 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1902 	case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1903 		/*
1904 		 * This occurs for various TM-related instructions that
1905 		 * we need to emulate on POWER9 DD2.2.  We have already
1906 		 * handled the cases where the guest was in real-suspend
1907 		 * mode and was transitioning to transactional state.
1908 		 */
1909 		r = kvmhv_p9_tm_emulation(vcpu);
1910 		if (r != -1)
1911 			break;
1912 		fallthrough; /* go to facility unavailable handler */
1913 #endif
1914 
1915 	/*
1916 	 * This occurs if the guest (kernel or userspace), does something that
1917 	 * is prohibited by HFSCR.
1918 	 * On POWER9, this could be a doorbell instruction that we need
1919 	 * to emulate.
1920 	 * Otherwise, we just generate a program interrupt to the guest.
1921 	 */
1922 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL: {
1923 		u64 cause = kvmppc_get_hfscr_hv(vcpu) >> 56;
1924 
1925 		r = EMULATE_FAIL;
1926 		if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1927 			switch (cause) {
1928 			case FSCR_MSGP_LG:
1929 				r = kvmppc_emulate_doorbell_instr(vcpu);
1930 				break;
1931 			case FSCR_PM_LG:
1932 				r = kvmppc_pmu_unavailable(vcpu);
1933 				break;
1934 			case FSCR_EBB_LG:
1935 				r = kvmppc_ebb_unavailable(vcpu);
1936 				break;
1937 			case FSCR_TM_LG:
1938 				r = kvmppc_tm_unavailable(vcpu);
1939 				break;
1940 			default:
1941 				break;
1942 			}
1943 		}
1944 		if (r == EMULATE_FAIL) {
1945 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL |
1946 				(kvmppc_get_msr(vcpu) & SRR1_PREFIXED));
1947 			r = RESUME_GUEST;
1948 		}
1949 		break;
1950 	}
1951 
1952 	case BOOK3S_INTERRUPT_HV_RM_HARD:
1953 		r = RESUME_PASSTHROUGH;
1954 		break;
1955 	default:
1956 		kvmppc_dump_regs(vcpu);
1957 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1958 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1959 			__kvmppc_get_msr_hv(vcpu));
1960 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1961 		r = RESUME_HOST;
1962 		break;
1963 	}
1964 
1965 	return r;
1966 }
1967 
1968 static int kvmppc_handle_nested_exit(struct kvm_vcpu *vcpu)
1969 {
1970 	int r;
1971 	int srcu_idx;
1972 
1973 	vcpu->stat.sum_exits++;
1974 
1975 	/*
1976 	 * This can happen if an interrupt occurs in the last stages
1977 	 * of guest entry or the first stages of guest exit (i.e. after
1978 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1979 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1980 	 * That can happen due to a bug, or due to a machine check
1981 	 * occurring at just the wrong time.
1982 	 */
1983 	if (__kvmppc_get_msr_hv(vcpu) & MSR_HV) {
1984 		pr_emerg("KVM trap in HV mode while nested!\n");
1985 		pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1986 			 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1987 			 __kvmppc_get_msr_hv(vcpu));
1988 		kvmppc_dump_regs(vcpu);
1989 		return RESUME_HOST;
1990 	}
1991 	switch (vcpu->arch.trap) {
1992 	/* We're good on these - the host merely wanted to get our attention */
1993 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1994 		vcpu->stat.dec_exits++;
1995 		r = RESUME_GUEST;
1996 		break;
1997 	case BOOK3S_INTERRUPT_EXTERNAL:
1998 		vcpu->stat.ext_intr_exits++;
1999 		r = RESUME_HOST;
2000 		break;
2001 	case BOOK3S_INTERRUPT_H_DOORBELL:
2002 	case BOOK3S_INTERRUPT_H_VIRT:
2003 		vcpu->stat.ext_intr_exits++;
2004 		r = RESUME_GUEST;
2005 		break;
2006 	/* These need to go to the nested HV */
2007 	case BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER:
2008 		vcpu->arch.trap = BOOK3S_INTERRUPT_HV_DECREMENTER;
2009 		vcpu->stat.dec_exits++;
2010 		r = RESUME_HOST;
2011 		break;
2012 	/* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
2013 	case BOOK3S_INTERRUPT_HMI:
2014 	case BOOK3S_INTERRUPT_PERFMON:
2015 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
2016 		r = RESUME_GUEST;
2017 		break;
2018 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
2019 	{
2020 		static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
2021 					      DEFAULT_RATELIMIT_BURST);
2022 		/* Pass the machine check to the L1 guest */
2023 		r = RESUME_HOST;
2024 		/* Print the MCE event to host console. */
2025 		if (__ratelimit(&rs))
2026 			machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
2027 		break;
2028 	}
2029 	/*
2030 	 * We get these next two if the guest accesses a page which it thinks
2031 	 * it has mapped but which is not actually present, either because
2032 	 * it is for an emulated I/O device or because the corresonding
2033 	 * host page has been paged out.
2034 	 */
2035 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
2036 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2037 		r = kvmhv_nested_page_fault(vcpu);
2038 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2039 		break;
2040 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
2041 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
2042 		vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
2043 					 DSISR_SRR1_MATCH_64S;
2044 		if (__kvmppc_get_msr_hv(vcpu) & HSRR1_HISI_WRITE)
2045 			vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
2046 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2047 		r = kvmhv_nested_page_fault(vcpu);
2048 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2049 		break;
2050 
2051 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2052 	case BOOK3S_INTERRUPT_HV_SOFTPATCH:
2053 		/*
2054 		 * This occurs for various TM-related instructions that
2055 		 * we need to emulate on POWER9 DD2.2.  We have already
2056 		 * handled the cases where the guest was in real-suspend
2057 		 * mode and was transitioning to transactional state.
2058 		 */
2059 		r = kvmhv_p9_tm_emulation(vcpu);
2060 		if (r != -1)
2061 			break;
2062 		fallthrough; /* go to facility unavailable handler */
2063 #endif
2064 
2065 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
2066 		r = RESUME_HOST;
2067 		break;
2068 
2069 	case BOOK3S_INTERRUPT_HV_RM_HARD:
2070 		vcpu->arch.trap = 0;
2071 		r = RESUME_GUEST;
2072 		if (!xics_on_xive())
2073 			kvmppc_xics_rm_complete(vcpu, 0);
2074 		break;
2075 	case BOOK3S_INTERRUPT_SYSCALL:
2076 	{
2077 		unsigned long req = kvmppc_get_gpr(vcpu, 3);
2078 
2079 		/*
2080 		 * The H_RPT_INVALIDATE hcalls issued by nested
2081 		 * guests for process-scoped invalidations when
2082 		 * GTSE=0, are handled here in L0.
2083 		 */
2084 		if (req == H_RPT_INVALIDATE) {
2085 			r = kvmppc_nested_h_rpt_invalidate(vcpu);
2086 			break;
2087 		}
2088 
2089 		r = RESUME_HOST;
2090 		break;
2091 	}
2092 	default:
2093 		r = RESUME_HOST;
2094 		break;
2095 	}
2096 
2097 	return r;
2098 }
2099 
2100 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
2101 					    struct kvm_sregs *sregs)
2102 {
2103 	int i;
2104 
2105 	memset(sregs, 0, sizeof(struct kvm_sregs));
2106 	sregs->pvr = vcpu->arch.pvr;
2107 	for (i = 0; i < vcpu->arch.slb_max; i++) {
2108 		sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
2109 		sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
2110 	}
2111 
2112 	return 0;
2113 }
2114 
2115 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
2116 					    struct kvm_sregs *sregs)
2117 {
2118 	int i, j;
2119 
2120 	/* Only accept the same PVR as the host's, since we can't spoof it */
2121 	if (sregs->pvr != vcpu->arch.pvr)
2122 		return -EINVAL;
2123 
2124 	j = 0;
2125 	for (i = 0; i < vcpu->arch.slb_nr; i++) {
2126 		if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
2127 			vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
2128 			vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
2129 			++j;
2130 		}
2131 	}
2132 	vcpu->arch.slb_max = j;
2133 
2134 	return 0;
2135 }
2136 
2137 /*
2138  * Enforce limits on guest LPCR values based on hardware availability,
2139  * guest configuration, and possibly hypervisor support and security
2140  * concerns.
2141  */
2142 unsigned long kvmppc_filter_lpcr_hv(struct kvm *kvm, unsigned long lpcr)
2143 {
2144 	/* LPCR_TC only applies to HPT guests */
2145 	if (kvm_is_radix(kvm))
2146 		lpcr &= ~LPCR_TC;
2147 
2148 	/* On POWER8 and above, userspace can modify AIL */
2149 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2150 		lpcr &= ~LPCR_AIL;
2151 	if ((lpcr & LPCR_AIL) != LPCR_AIL_3)
2152 		lpcr &= ~LPCR_AIL; /* LPCR[AIL]=1/2 is disallowed */
2153 	/*
2154 	 * On some POWER9s we force AIL off for radix guests to prevent
2155 	 * executing in MSR[HV]=1 mode with the MMU enabled and PIDR set to
2156 	 * guest, which can result in Q0 translations with LPID=0 PID=PIDR to
2157 	 * be cached, which the host TLB management does not expect.
2158 	 */
2159 	if (kvm_is_radix(kvm) && cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG))
2160 		lpcr &= ~LPCR_AIL;
2161 
2162 	/*
2163 	 * On POWER9, allow userspace to enable large decrementer for the
2164 	 * guest, whether or not the host has it enabled.
2165 	 */
2166 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
2167 		lpcr &= ~LPCR_LD;
2168 
2169 	return lpcr;
2170 }
2171 
2172 static void verify_lpcr(struct kvm *kvm, unsigned long lpcr)
2173 {
2174 	if (lpcr != kvmppc_filter_lpcr_hv(kvm, lpcr)) {
2175 		WARN_ONCE(1, "lpcr 0x%lx differs from filtered 0x%lx\n",
2176 			  lpcr, kvmppc_filter_lpcr_hv(kvm, lpcr));
2177 	}
2178 }
2179 
2180 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
2181 		bool preserve_top32)
2182 {
2183 	struct kvm *kvm = vcpu->kvm;
2184 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
2185 	u64 mask;
2186 
2187 	spin_lock(&vc->lock);
2188 
2189 	/*
2190 	 * Userspace can only modify
2191 	 * DPFD (default prefetch depth), ILE (interrupt little-endian),
2192 	 * TC (translation control), AIL (alternate interrupt location),
2193 	 * LD (large decrementer).
2194 	 * These are subject to restrictions from kvmppc_filter_lcpr_hv().
2195 	 */
2196 	mask = LPCR_DPFD | LPCR_ILE | LPCR_TC | LPCR_AIL | LPCR_LD;
2197 
2198 	/* Broken 32-bit version of LPCR must not clear top bits */
2199 	if (preserve_top32)
2200 		mask &= 0xFFFFFFFF;
2201 
2202 	new_lpcr = kvmppc_filter_lpcr_hv(kvm,
2203 			(vc->lpcr & ~mask) | (new_lpcr & mask));
2204 
2205 	/*
2206 	 * If ILE (interrupt little-endian) has changed, update the
2207 	 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
2208 	 */
2209 	if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
2210 		struct kvm_vcpu *vcpu;
2211 		unsigned long i;
2212 
2213 		kvm_for_each_vcpu(i, vcpu, kvm) {
2214 			if (vcpu->arch.vcore != vc)
2215 				continue;
2216 			if (new_lpcr & LPCR_ILE)
2217 				vcpu->arch.intr_msr |= MSR_LE;
2218 			else
2219 				vcpu->arch.intr_msr &= ~MSR_LE;
2220 		}
2221 	}
2222 
2223 	vc->lpcr = new_lpcr;
2224 	kvmhv_nestedv2_mark_dirty(vcpu, KVMPPC_GSID_LPCR);
2225 
2226 	spin_unlock(&vc->lock);
2227 }
2228 
2229 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
2230 				 union kvmppc_one_reg *val)
2231 {
2232 	int r = 0;
2233 	long int i;
2234 
2235 	switch (id) {
2236 	case KVM_REG_PPC_DEBUG_INST:
2237 		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
2238 		break;
2239 	case KVM_REG_PPC_HIOR:
2240 		*val = get_reg_val(id, 0);
2241 		break;
2242 	case KVM_REG_PPC_DABR:
2243 		*val = get_reg_val(id, vcpu->arch.dabr);
2244 		break;
2245 	case KVM_REG_PPC_DABRX:
2246 		*val = get_reg_val(id, vcpu->arch.dabrx);
2247 		break;
2248 	case KVM_REG_PPC_DSCR:
2249 		*val = get_reg_val(id, kvmppc_get_dscr_hv(vcpu));
2250 		break;
2251 	case KVM_REG_PPC_PURR:
2252 		*val = get_reg_val(id, kvmppc_get_purr_hv(vcpu));
2253 		break;
2254 	case KVM_REG_PPC_SPURR:
2255 		*val = get_reg_val(id, kvmppc_get_spurr_hv(vcpu));
2256 		break;
2257 	case KVM_REG_PPC_AMR:
2258 		*val = get_reg_val(id, kvmppc_get_amr_hv(vcpu));
2259 		break;
2260 	case KVM_REG_PPC_UAMOR:
2261 		*val = get_reg_val(id, kvmppc_get_uamor_hv(vcpu));
2262 		break;
2263 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
2264 		i = id - KVM_REG_PPC_MMCR0;
2265 		*val = get_reg_val(id, kvmppc_get_mmcr_hv(vcpu, i));
2266 		break;
2267 	case KVM_REG_PPC_MMCR2:
2268 		*val = get_reg_val(id, kvmppc_get_mmcr_hv(vcpu, 2));
2269 		break;
2270 	case KVM_REG_PPC_MMCRA:
2271 		*val = get_reg_val(id, kvmppc_get_mmcra_hv(vcpu));
2272 		break;
2273 	case KVM_REG_PPC_MMCRS:
2274 		*val = get_reg_val(id, vcpu->arch.mmcrs);
2275 		break;
2276 	case KVM_REG_PPC_MMCR3:
2277 		*val = get_reg_val(id, kvmppc_get_mmcr_hv(vcpu, 3));
2278 		break;
2279 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
2280 		i = id - KVM_REG_PPC_PMC1;
2281 		*val = get_reg_val(id, kvmppc_get_pmc_hv(vcpu, i));
2282 		break;
2283 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
2284 		i = id - KVM_REG_PPC_SPMC1;
2285 		*val = get_reg_val(id, vcpu->arch.spmc[i]);
2286 		break;
2287 	case KVM_REG_PPC_SIAR:
2288 		*val = get_reg_val(id, kvmppc_get_siar_hv(vcpu));
2289 		break;
2290 	case KVM_REG_PPC_SDAR:
2291 		*val = get_reg_val(id, kvmppc_get_sdar_hv(vcpu));
2292 		break;
2293 	case KVM_REG_PPC_SIER:
2294 		*val = get_reg_val(id, kvmppc_get_sier_hv(vcpu, 0));
2295 		break;
2296 	case KVM_REG_PPC_SIER2:
2297 		*val = get_reg_val(id, kvmppc_get_sier_hv(vcpu, 1));
2298 		break;
2299 	case KVM_REG_PPC_SIER3:
2300 		*val = get_reg_val(id, kvmppc_get_sier_hv(vcpu, 2));
2301 		break;
2302 	case KVM_REG_PPC_IAMR:
2303 		*val = get_reg_val(id, kvmppc_get_iamr_hv(vcpu));
2304 		break;
2305 	case KVM_REG_PPC_PSPB:
2306 		*val = get_reg_val(id, kvmppc_get_pspb_hv(vcpu));
2307 		break;
2308 	case KVM_REG_PPC_DPDES:
2309 		/*
2310 		 * On POWER9, where we are emulating msgsndp etc.,
2311 		 * we return 1 bit for each vcpu, which can come from
2312 		 * either vcore->dpdes or doorbell_request.
2313 		 * On POWER8, doorbell_request is 0.
2314 		 */
2315 		if (cpu_has_feature(CPU_FTR_ARCH_300))
2316 			*val = get_reg_val(id, vcpu->arch.doorbell_request);
2317 		else
2318 			*val = get_reg_val(id, vcpu->arch.vcore->dpdes);
2319 		break;
2320 	case KVM_REG_PPC_VTB:
2321 		*val = get_reg_val(id, kvmppc_get_vtb(vcpu));
2322 		break;
2323 	case KVM_REG_PPC_DAWR:
2324 		*val = get_reg_val(id, kvmppc_get_dawr0_hv(vcpu));
2325 		break;
2326 	case KVM_REG_PPC_DAWRX:
2327 		*val = get_reg_val(id, kvmppc_get_dawrx0_hv(vcpu));
2328 		break;
2329 	case KVM_REG_PPC_DAWR1:
2330 		*val = get_reg_val(id, kvmppc_get_dawr1_hv(vcpu));
2331 		break;
2332 	case KVM_REG_PPC_DAWRX1:
2333 		*val = get_reg_val(id, kvmppc_get_dawrx1_hv(vcpu));
2334 		break;
2335 	case KVM_REG_PPC_DEXCR:
2336 		*val = get_reg_val(id, kvmppc_get_dexcr_hv(vcpu));
2337 		break;
2338 	case KVM_REG_PPC_HASHKEYR:
2339 		*val = get_reg_val(id, kvmppc_get_hashkeyr_hv(vcpu));
2340 		break;
2341 	case KVM_REG_PPC_HASHPKEYR:
2342 		*val = get_reg_val(id, kvmppc_get_hashpkeyr_hv(vcpu));
2343 		break;
2344 	case KVM_REG_PPC_CIABR:
2345 		*val = get_reg_val(id, kvmppc_get_ciabr_hv(vcpu));
2346 		break;
2347 	case KVM_REG_PPC_CSIGR:
2348 		*val = get_reg_val(id, vcpu->arch.csigr);
2349 		break;
2350 	case KVM_REG_PPC_TACR:
2351 		*val = get_reg_val(id, vcpu->arch.tacr);
2352 		break;
2353 	case KVM_REG_PPC_TCSCR:
2354 		*val = get_reg_val(id, vcpu->arch.tcscr);
2355 		break;
2356 	case KVM_REG_PPC_PID:
2357 		*val = get_reg_val(id, kvmppc_get_pid(vcpu));
2358 		break;
2359 	case KVM_REG_PPC_ACOP:
2360 		*val = get_reg_val(id, vcpu->arch.acop);
2361 		break;
2362 	case KVM_REG_PPC_WORT:
2363 		*val = get_reg_val(id, kvmppc_get_wort_hv(vcpu));
2364 		break;
2365 	case KVM_REG_PPC_TIDR:
2366 		*val = get_reg_val(id, vcpu->arch.tid);
2367 		break;
2368 	case KVM_REG_PPC_PSSCR:
2369 		*val = get_reg_val(id, vcpu->arch.psscr);
2370 		break;
2371 	case KVM_REG_PPC_VPA_ADDR:
2372 		spin_lock(&vcpu->arch.vpa_update_lock);
2373 		*val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
2374 		spin_unlock(&vcpu->arch.vpa_update_lock);
2375 		break;
2376 	case KVM_REG_PPC_VPA_SLB:
2377 		spin_lock(&vcpu->arch.vpa_update_lock);
2378 		val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
2379 		val->vpaval.length = vcpu->arch.slb_shadow.len;
2380 		spin_unlock(&vcpu->arch.vpa_update_lock);
2381 		break;
2382 	case KVM_REG_PPC_VPA_DTL:
2383 		spin_lock(&vcpu->arch.vpa_update_lock);
2384 		val->vpaval.addr = vcpu->arch.dtl.next_gpa;
2385 		val->vpaval.length = vcpu->arch.dtl.len;
2386 		spin_unlock(&vcpu->arch.vpa_update_lock);
2387 		break;
2388 	case KVM_REG_PPC_TB_OFFSET:
2389 		*val = get_reg_val(id, kvmppc_get_tb_offset(vcpu));
2390 		break;
2391 	case KVM_REG_PPC_LPCR:
2392 	case KVM_REG_PPC_LPCR_64:
2393 		*val = get_reg_val(id, kvmppc_get_lpcr(vcpu));
2394 		break;
2395 	case KVM_REG_PPC_PPR:
2396 		*val = get_reg_val(id, kvmppc_get_ppr_hv(vcpu));
2397 		break;
2398 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2399 	case KVM_REG_PPC_TFHAR:
2400 		*val = get_reg_val(id, vcpu->arch.tfhar);
2401 		break;
2402 	case KVM_REG_PPC_TFIAR:
2403 		*val = get_reg_val(id, vcpu->arch.tfiar);
2404 		break;
2405 	case KVM_REG_PPC_TEXASR:
2406 		*val = get_reg_val(id, vcpu->arch.texasr);
2407 		break;
2408 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2409 		i = id - KVM_REG_PPC_TM_GPR0;
2410 		*val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
2411 		break;
2412 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2413 	{
2414 		int j;
2415 		i = id - KVM_REG_PPC_TM_VSR0;
2416 		if (i < 32)
2417 			for (j = 0; j < TS_FPRWIDTH; j++)
2418 				val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
2419 		else {
2420 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
2421 				val->vval = vcpu->arch.vr_tm.vr[i-32];
2422 			else
2423 				r = -ENXIO;
2424 		}
2425 		break;
2426 	}
2427 	case KVM_REG_PPC_TM_CR:
2428 		*val = get_reg_val(id, vcpu->arch.cr_tm);
2429 		break;
2430 	case KVM_REG_PPC_TM_XER:
2431 		*val = get_reg_val(id, vcpu->arch.xer_tm);
2432 		break;
2433 	case KVM_REG_PPC_TM_LR:
2434 		*val = get_reg_val(id, vcpu->arch.lr_tm);
2435 		break;
2436 	case KVM_REG_PPC_TM_CTR:
2437 		*val = get_reg_val(id, vcpu->arch.ctr_tm);
2438 		break;
2439 	case KVM_REG_PPC_TM_FPSCR:
2440 		*val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
2441 		break;
2442 	case KVM_REG_PPC_TM_AMR:
2443 		*val = get_reg_val(id, vcpu->arch.amr_tm);
2444 		break;
2445 	case KVM_REG_PPC_TM_PPR:
2446 		*val = get_reg_val(id, vcpu->arch.ppr_tm);
2447 		break;
2448 	case KVM_REG_PPC_TM_VRSAVE:
2449 		*val = get_reg_val(id, vcpu->arch.vrsave_tm);
2450 		break;
2451 	case KVM_REG_PPC_TM_VSCR:
2452 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
2453 			*val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
2454 		else
2455 			r = -ENXIO;
2456 		break;
2457 	case KVM_REG_PPC_TM_DSCR:
2458 		*val = get_reg_val(id, vcpu->arch.dscr_tm);
2459 		break;
2460 	case KVM_REG_PPC_TM_TAR:
2461 		*val = get_reg_val(id, vcpu->arch.tar_tm);
2462 		break;
2463 #endif
2464 	case KVM_REG_PPC_ARCH_COMPAT:
2465 		*val = get_reg_val(id, kvmppc_get_arch_compat(vcpu));
2466 		break;
2467 	case KVM_REG_PPC_DEC_EXPIRY:
2468 		*val = get_reg_val(id, kvmppc_get_dec_expires(vcpu));
2469 		break;
2470 	case KVM_REG_PPC_ONLINE:
2471 		*val = get_reg_val(id, vcpu->arch.online);
2472 		break;
2473 	case KVM_REG_PPC_PTCR:
2474 		*val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
2475 		break;
2476 	case KVM_REG_PPC_FSCR:
2477 		*val = get_reg_val(id, kvmppc_get_fscr_hv(vcpu));
2478 		break;
2479 	default:
2480 		r = -EINVAL;
2481 		break;
2482 	}
2483 
2484 	return r;
2485 }
2486 
2487 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
2488 				 union kvmppc_one_reg *val)
2489 {
2490 	int r = 0;
2491 	long int i;
2492 	unsigned long addr, len;
2493 
2494 	switch (id) {
2495 	case KVM_REG_PPC_HIOR:
2496 		/* Only allow this to be set to zero */
2497 		if (set_reg_val(id, *val))
2498 			r = -EINVAL;
2499 		break;
2500 	case KVM_REG_PPC_DABR:
2501 		vcpu->arch.dabr = set_reg_val(id, *val);
2502 		break;
2503 	case KVM_REG_PPC_DABRX:
2504 		vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
2505 		break;
2506 	case KVM_REG_PPC_DSCR:
2507 		kvmppc_set_dscr_hv(vcpu, set_reg_val(id, *val));
2508 		break;
2509 	case KVM_REG_PPC_PURR:
2510 		kvmppc_set_purr_hv(vcpu, set_reg_val(id, *val));
2511 		break;
2512 	case KVM_REG_PPC_SPURR:
2513 		kvmppc_set_spurr_hv(vcpu, set_reg_val(id, *val));
2514 		break;
2515 	case KVM_REG_PPC_AMR:
2516 		kvmppc_set_amr_hv(vcpu, set_reg_val(id, *val));
2517 		break;
2518 	case KVM_REG_PPC_UAMOR:
2519 		kvmppc_set_uamor_hv(vcpu, set_reg_val(id, *val));
2520 		break;
2521 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
2522 		i = id - KVM_REG_PPC_MMCR0;
2523 		kvmppc_set_mmcr_hv(vcpu, i, set_reg_val(id, *val));
2524 		break;
2525 	case KVM_REG_PPC_MMCR2:
2526 		kvmppc_set_mmcr_hv(vcpu, 2, set_reg_val(id, *val));
2527 		break;
2528 	case KVM_REG_PPC_MMCRA:
2529 		kvmppc_set_mmcra_hv(vcpu, set_reg_val(id, *val));
2530 		break;
2531 	case KVM_REG_PPC_MMCRS:
2532 		vcpu->arch.mmcrs = set_reg_val(id, *val);
2533 		break;
2534 	case KVM_REG_PPC_MMCR3:
2535 		kvmppc_set_mmcr_hv(vcpu, 3, set_reg_val(id, *val));
2536 		break;
2537 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
2538 		i = id - KVM_REG_PPC_PMC1;
2539 		kvmppc_set_pmc_hv(vcpu, i, set_reg_val(id, *val));
2540 		break;
2541 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
2542 		i = id - KVM_REG_PPC_SPMC1;
2543 		vcpu->arch.spmc[i] = set_reg_val(id, *val);
2544 		break;
2545 	case KVM_REG_PPC_SIAR:
2546 		kvmppc_set_siar_hv(vcpu, set_reg_val(id, *val));
2547 		break;
2548 	case KVM_REG_PPC_SDAR:
2549 		kvmppc_set_sdar_hv(vcpu, set_reg_val(id, *val));
2550 		break;
2551 	case KVM_REG_PPC_SIER:
2552 		kvmppc_set_sier_hv(vcpu, 0, set_reg_val(id, *val));
2553 		break;
2554 	case KVM_REG_PPC_SIER2:
2555 		kvmppc_set_sier_hv(vcpu, 1, set_reg_val(id, *val));
2556 		break;
2557 	case KVM_REG_PPC_SIER3:
2558 		kvmppc_set_sier_hv(vcpu, 2, set_reg_val(id, *val));
2559 		break;
2560 	case KVM_REG_PPC_IAMR:
2561 		kvmppc_set_iamr_hv(vcpu, set_reg_val(id, *val));
2562 		break;
2563 	case KVM_REG_PPC_PSPB:
2564 		kvmppc_set_pspb_hv(vcpu, set_reg_val(id, *val));
2565 		break;
2566 	case KVM_REG_PPC_DPDES:
2567 		if (cpu_has_feature(CPU_FTR_ARCH_300))
2568 			vcpu->arch.doorbell_request = set_reg_val(id, *val) & 1;
2569 		else
2570 			vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
2571 		break;
2572 	case KVM_REG_PPC_VTB:
2573 		kvmppc_set_vtb(vcpu, set_reg_val(id, *val));
2574 		break;
2575 	case KVM_REG_PPC_DAWR:
2576 		kvmppc_set_dawr0_hv(vcpu, set_reg_val(id, *val));
2577 		break;
2578 	case KVM_REG_PPC_DAWRX:
2579 		kvmppc_set_dawrx0_hv(vcpu, set_reg_val(id, *val) & ~DAWRX_HYP);
2580 		break;
2581 	case KVM_REG_PPC_DAWR1:
2582 		kvmppc_set_dawr1_hv(vcpu, set_reg_val(id, *val));
2583 		break;
2584 	case KVM_REG_PPC_DAWRX1:
2585 		kvmppc_set_dawrx1_hv(vcpu, set_reg_val(id, *val) & ~DAWRX_HYP);
2586 		break;
2587 	case KVM_REG_PPC_DEXCR:
2588 		kvmppc_set_dexcr_hv(vcpu, set_reg_val(id, *val));
2589 		break;
2590 	case KVM_REG_PPC_HASHKEYR:
2591 		kvmppc_set_hashkeyr_hv(vcpu, set_reg_val(id, *val));
2592 		break;
2593 	case KVM_REG_PPC_HASHPKEYR:
2594 		kvmppc_set_hashpkeyr_hv(vcpu, set_reg_val(id, *val));
2595 		break;
2596 	case KVM_REG_PPC_CIABR:
2597 		kvmppc_set_ciabr_hv(vcpu, set_reg_val(id, *val));
2598 		/* Don't allow setting breakpoints in hypervisor code */
2599 		if ((kvmppc_get_ciabr_hv(vcpu) & CIABR_PRIV) == CIABR_PRIV_HYPER)
2600 			kvmppc_set_ciabr_hv(vcpu, kvmppc_get_ciabr_hv(vcpu) & ~CIABR_PRIV);
2601 		break;
2602 	case KVM_REG_PPC_CSIGR:
2603 		vcpu->arch.csigr = set_reg_val(id, *val);
2604 		break;
2605 	case KVM_REG_PPC_TACR:
2606 		vcpu->arch.tacr = set_reg_val(id, *val);
2607 		break;
2608 	case KVM_REG_PPC_TCSCR:
2609 		vcpu->arch.tcscr = set_reg_val(id, *val);
2610 		break;
2611 	case KVM_REG_PPC_PID:
2612 		kvmppc_set_pid(vcpu, set_reg_val(id, *val));
2613 		break;
2614 	case KVM_REG_PPC_ACOP:
2615 		vcpu->arch.acop = set_reg_val(id, *val);
2616 		break;
2617 	case KVM_REG_PPC_WORT:
2618 		kvmppc_set_wort_hv(vcpu, set_reg_val(id, *val));
2619 		break;
2620 	case KVM_REG_PPC_TIDR:
2621 		vcpu->arch.tid = set_reg_val(id, *val);
2622 		break;
2623 	case KVM_REG_PPC_PSSCR:
2624 		vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
2625 		break;
2626 	case KVM_REG_PPC_VPA_ADDR:
2627 		addr = set_reg_val(id, *val);
2628 		r = -EINVAL;
2629 		if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
2630 			      vcpu->arch.dtl.next_gpa))
2631 			break;
2632 		r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
2633 		break;
2634 	case KVM_REG_PPC_VPA_SLB:
2635 		addr = val->vpaval.addr;
2636 		len = val->vpaval.length;
2637 		r = -EINVAL;
2638 		if (addr && !vcpu->arch.vpa.next_gpa)
2639 			break;
2640 		r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
2641 		break;
2642 	case KVM_REG_PPC_VPA_DTL:
2643 		addr = val->vpaval.addr;
2644 		len = val->vpaval.length;
2645 		r = -EINVAL;
2646 		if (addr && (len < sizeof(struct dtl_entry) ||
2647 			     !vcpu->arch.vpa.next_gpa))
2648 			break;
2649 		len -= len % sizeof(struct dtl_entry);
2650 		r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
2651 		break;
2652 	case KVM_REG_PPC_TB_OFFSET:
2653 	{
2654 		/* round up to multiple of 2^24 */
2655 		u64 tb_offset = ALIGN(set_reg_val(id, *val), 1UL << 24);
2656 
2657 		/*
2658 		 * Now that we know the timebase offset, update the
2659 		 * decrementer expiry with a guest timebase value. If
2660 		 * the userspace does not set DEC_EXPIRY, this ensures
2661 		 * a migrated vcpu at least starts with an expired
2662 		 * decrementer, which is better than a large one that
2663 		 * causes a hang.
2664 		 */
2665 		kvmppc_set_tb_offset(vcpu, tb_offset);
2666 		if (!kvmppc_get_dec_expires(vcpu) && tb_offset)
2667 			kvmppc_set_dec_expires(vcpu, get_tb() + tb_offset);
2668 
2669 		kvmppc_set_tb_offset(vcpu, tb_offset);
2670 		break;
2671 	}
2672 	case KVM_REG_PPC_LPCR:
2673 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
2674 		break;
2675 	case KVM_REG_PPC_LPCR_64:
2676 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
2677 		break;
2678 	case KVM_REG_PPC_PPR:
2679 		kvmppc_set_ppr_hv(vcpu, set_reg_val(id, *val));
2680 		break;
2681 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2682 	case KVM_REG_PPC_TFHAR:
2683 		vcpu->arch.tfhar = set_reg_val(id, *val);
2684 		break;
2685 	case KVM_REG_PPC_TFIAR:
2686 		vcpu->arch.tfiar = set_reg_val(id, *val);
2687 		break;
2688 	case KVM_REG_PPC_TEXASR:
2689 		vcpu->arch.texasr = set_reg_val(id, *val);
2690 		break;
2691 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2692 		i = id - KVM_REG_PPC_TM_GPR0;
2693 		vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
2694 		break;
2695 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2696 	{
2697 		int j;
2698 		i = id - KVM_REG_PPC_TM_VSR0;
2699 		if (i < 32)
2700 			for (j = 0; j < TS_FPRWIDTH; j++)
2701 				vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
2702 		else
2703 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
2704 				vcpu->arch.vr_tm.vr[i-32] = val->vval;
2705 			else
2706 				r = -ENXIO;
2707 		break;
2708 	}
2709 	case KVM_REG_PPC_TM_CR:
2710 		vcpu->arch.cr_tm = set_reg_val(id, *val);
2711 		break;
2712 	case KVM_REG_PPC_TM_XER:
2713 		vcpu->arch.xer_tm = set_reg_val(id, *val);
2714 		break;
2715 	case KVM_REG_PPC_TM_LR:
2716 		vcpu->arch.lr_tm = set_reg_val(id, *val);
2717 		break;
2718 	case KVM_REG_PPC_TM_CTR:
2719 		vcpu->arch.ctr_tm = set_reg_val(id, *val);
2720 		break;
2721 	case KVM_REG_PPC_TM_FPSCR:
2722 		vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
2723 		break;
2724 	case KVM_REG_PPC_TM_AMR:
2725 		vcpu->arch.amr_tm = set_reg_val(id, *val);
2726 		break;
2727 	case KVM_REG_PPC_TM_PPR:
2728 		vcpu->arch.ppr_tm = set_reg_val(id, *val);
2729 		break;
2730 	case KVM_REG_PPC_TM_VRSAVE:
2731 		vcpu->arch.vrsave_tm = set_reg_val(id, *val);
2732 		break;
2733 	case KVM_REG_PPC_TM_VSCR:
2734 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
2735 			vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
2736 		else
2737 			r = - ENXIO;
2738 		break;
2739 	case KVM_REG_PPC_TM_DSCR:
2740 		vcpu->arch.dscr_tm = set_reg_val(id, *val);
2741 		break;
2742 	case KVM_REG_PPC_TM_TAR:
2743 		vcpu->arch.tar_tm = set_reg_val(id, *val);
2744 		break;
2745 #endif
2746 	case KVM_REG_PPC_ARCH_COMPAT:
2747 		r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
2748 		break;
2749 	case KVM_REG_PPC_DEC_EXPIRY:
2750 		kvmppc_set_dec_expires(vcpu, set_reg_val(id, *val));
2751 		break;
2752 	case KVM_REG_PPC_ONLINE:
2753 		i = set_reg_val(id, *val);
2754 		if (i && !vcpu->arch.online)
2755 			atomic_inc(&vcpu->arch.vcore->online_count);
2756 		else if (!i && vcpu->arch.online)
2757 			atomic_dec(&vcpu->arch.vcore->online_count);
2758 		vcpu->arch.online = i;
2759 		break;
2760 	case KVM_REG_PPC_PTCR:
2761 		vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
2762 		break;
2763 	case KVM_REG_PPC_FSCR:
2764 		kvmppc_set_fscr_hv(vcpu, set_reg_val(id, *val));
2765 		break;
2766 	default:
2767 		r = -EINVAL;
2768 		break;
2769 	}
2770 
2771 	return r;
2772 }
2773 
2774 /*
2775  * On POWER9, threads are independent and can be in different partitions.
2776  * Therefore we consider each thread to be a subcore.
2777  * There is a restriction that all threads have to be in the same
2778  * MMU mode (radix or HPT), unfortunately, but since we only support
2779  * HPT guests on a HPT host so far, that isn't an impediment yet.
2780  */
2781 static int threads_per_vcore(struct kvm *kvm)
2782 {
2783 	if (cpu_has_feature(CPU_FTR_ARCH_300))
2784 		return 1;
2785 	return threads_per_subcore;
2786 }
2787 
2788 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
2789 {
2790 	struct kvmppc_vcore *vcore;
2791 
2792 	vcore = kzalloc_obj(struct kvmppc_vcore);
2793 
2794 	if (vcore == NULL)
2795 		return NULL;
2796 
2797 	spin_lock_init(&vcore->lock);
2798 	spin_lock_init(&vcore->stoltb_lock);
2799 	rcuwait_init(&vcore->wait);
2800 	vcore->preempt_tb = TB_NIL;
2801 	vcore->lpcr = kvm->arch.lpcr;
2802 	vcore->first_vcpuid = id;
2803 	vcore->kvm = kvm;
2804 	INIT_LIST_HEAD(&vcore->preempt_list);
2805 
2806 	return vcore;
2807 }
2808 
2809 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2810 static struct debugfs_timings_element {
2811 	const char *name;
2812 	size_t offset;
2813 } timings[] = {
2814 #ifdef CONFIG_KVM_BOOK3S_HV_P9_TIMING
2815 	{"vcpu_entry",	offsetof(struct kvm_vcpu, arch.vcpu_entry)},
2816 	{"guest_entry",	offsetof(struct kvm_vcpu, arch.guest_entry)},
2817 	{"in_guest",	offsetof(struct kvm_vcpu, arch.in_guest)},
2818 	{"guest_exit",	offsetof(struct kvm_vcpu, arch.guest_exit)},
2819 	{"vcpu_exit",	offsetof(struct kvm_vcpu, arch.vcpu_exit)},
2820 	{"hypercall",	offsetof(struct kvm_vcpu, arch.hcall)},
2821 	{"page_fault",	offsetof(struct kvm_vcpu, arch.pg_fault)},
2822 #else
2823 	{"rm_entry",	offsetof(struct kvm_vcpu, arch.rm_entry)},
2824 	{"rm_intr",	offsetof(struct kvm_vcpu, arch.rm_intr)},
2825 	{"rm_exit",	offsetof(struct kvm_vcpu, arch.rm_exit)},
2826 	{"guest",	offsetof(struct kvm_vcpu, arch.guest_time)},
2827 	{"cede",	offsetof(struct kvm_vcpu, arch.cede_time)},
2828 #endif
2829 };
2830 
2831 #define N_TIMINGS	(ARRAY_SIZE(timings))
2832 
2833 struct debugfs_timings_state {
2834 	struct kvm_vcpu	*vcpu;
2835 	unsigned int	buflen;
2836 	char		buf[N_TIMINGS * 100];
2837 };
2838 
2839 static int debugfs_timings_open(struct inode *inode, struct file *file)
2840 {
2841 	struct kvm_vcpu *vcpu = inode->i_private;
2842 	struct debugfs_timings_state *p;
2843 
2844 	p = kzalloc_obj(*p);
2845 	if (!p)
2846 		return -ENOMEM;
2847 
2848 	kvm_get_kvm(vcpu->kvm);
2849 	p->vcpu = vcpu;
2850 	file->private_data = p;
2851 
2852 	return nonseekable_open(inode, file);
2853 }
2854 
2855 static int debugfs_timings_release(struct inode *inode, struct file *file)
2856 {
2857 	struct debugfs_timings_state *p = file->private_data;
2858 
2859 	kvm_put_kvm(p->vcpu->kvm);
2860 	kfree(p);
2861 	return 0;
2862 }
2863 
2864 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2865 				    size_t len, loff_t *ppos)
2866 {
2867 	struct debugfs_timings_state *p = file->private_data;
2868 	struct kvm_vcpu *vcpu = p->vcpu;
2869 	char *s, *buf_end;
2870 	struct kvmhv_tb_accumulator tb;
2871 	u64 count;
2872 	loff_t pos;
2873 	ssize_t n;
2874 	int i, loops;
2875 	bool ok;
2876 
2877 	if (!p->buflen) {
2878 		s = p->buf;
2879 		buf_end = s + sizeof(p->buf);
2880 		for (i = 0; i < N_TIMINGS; ++i) {
2881 			struct kvmhv_tb_accumulator *acc;
2882 
2883 			acc = (struct kvmhv_tb_accumulator *)
2884 				((unsigned long)vcpu + timings[i].offset);
2885 			ok = false;
2886 			for (loops = 0; loops < 1000; ++loops) {
2887 				count = acc->seqcount;
2888 				if (!(count & 1)) {
2889 					smp_rmb();
2890 					tb = *acc;
2891 					smp_rmb();
2892 					if (count == acc->seqcount) {
2893 						ok = true;
2894 						break;
2895 					}
2896 				}
2897 				udelay(1);
2898 			}
2899 			if (!ok)
2900 				snprintf(s, buf_end - s, "%s: stuck\n",
2901 					timings[i].name);
2902 			else
2903 				snprintf(s, buf_end - s,
2904 					"%s: %llu %llu %llu %llu\n",
2905 					timings[i].name, count / 2,
2906 					tb_to_ns(tb.tb_total),
2907 					tb_to_ns(tb.tb_min),
2908 					tb_to_ns(tb.tb_max));
2909 			s += strlen(s);
2910 		}
2911 		p->buflen = s - p->buf;
2912 	}
2913 
2914 	pos = *ppos;
2915 	if (pos >= p->buflen)
2916 		return 0;
2917 	if (len > p->buflen - pos)
2918 		len = p->buflen - pos;
2919 	n = copy_to_user(buf, p->buf + pos, len);
2920 	if (n) {
2921 		if (n == len)
2922 			return -EFAULT;
2923 		len -= n;
2924 	}
2925 	*ppos = pos + len;
2926 	return len;
2927 }
2928 
2929 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2930 				     size_t len, loff_t *ppos)
2931 {
2932 	return -EACCES;
2933 }
2934 
2935 static const struct file_operations debugfs_timings_ops = {
2936 	.owner	 = THIS_MODULE,
2937 	.open	 = debugfs_timings_open,
2938 	.release = debugfs_timings_release,
2939 	.read	 = debugfs_timings_read,
2940 	.write	 = debugfs_timings_write,
2941 	.llseek	 = generic_file_llseek,
2942 };
2943 
2944 /* Create a debugfs directory for the vcpu */
2945 static int kvmppc_arch_create_vcpu_debugfs_hv(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry)
2946 {
2947 	if (cpu_has_feature(CPU_FTR_ARCH_300) == IS_ENABLED(CONFIG_KVM_BOOK3S_HV_P9_TIMING))
2948 		debugfs_create_file("timings", 0444, debugfs_dentry, vcpu,
2949 				    &debugfs_timings_ops);
2950 	return 0;
2951 }
2952 
2953 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2954 static int kvmppc_arch_create_vcpu_debugfs_hv(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry)
2955 {
2956 	return 0;
2957 }
2958 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2959 
2960 static int kvmppc_core_vcpu_create_hv(struct kvm_vcpu *vcpu)
2961 {
2962 	int err;
2963 	int core;
2964 	struct kvmppc_vcore *vcore;
2965 	struct kvm *kvm;
2966 	unsigned int id;
2967 
2968 	kvm = vcpu->kvm;
2969 	id = vcpu->vcpu_id;
2970 
2971 	vcpu->arch.shared = &vcpu->arch.shregs;
2972 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2973 	/*
2974 	 * The shared struct is never shared on HV,
2975 	 * so we can always use host endianness
2976 	 */
2977 #ifdef __BIG_ENDIAN__
2978 	vcpu->arch.shared_big_endian = true;
2979 #else
2980 	vcpu->arch.shared_big_endian = false;
2981 #endif
2982 #endif
2983 
2984 	if (kvmhv_is_nestedv2()) {
2985 		err = kvmhv_nestedv2_vcpu_create(vcpu, &vcpu->arch.nestedv2_io);
2986 		if (err < 0)
2987 			return err;
2988 	}
2989 
2990 	kvmppc_set_mmcr_hv(vcpu, 0, MMCR0_FC);
2991 	if (cpu_has_feature(CPU_FTR_ARCH_31)) {
2992 		kvmppc_set_mmcr_hv(vcpu, 0, kvmppc_get_mmcr_hv(vcpu, 0) | MMCR0_PMCCEXT);
2993 		kvmppc_set_mmcra_hv(vcpu, MMCRA_BHRB_DISABLE);
2994 	}
2995 
2996 	kvmppc_set_ctrl_hv(vcpu, CTRL_RUNLATCH);
2997 	/* default to host PVR, since we can't spoof it */
2998 	kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2999 	spin_lock_init(&vcpu->arch.vpa_update_lock);
3000 	spin_lock_init(&vcpu->arch.tbacct_lock);
3001 	vcpu->arch.busy_preempt = TB_NIL;
3002 	__kvmppc_set_msr_hv(vcpu, MSR_ME);
3003 	vcpu->arch.intr_msr = MSR_SF | MSR_ME;
3004 
3005 	/*
3006 	 * Set the default HFSCR for the guest from the host value.
3007 	 * This value is only used on POWER9 and later.
3008 	 * On >= POWER9, we want to virtualize the doorbell facility, so we
3009 	 * don't set the HFSCR_MSGP bit, and that causes those instructions
3010 	 * to trap and then we emulate them.
3011 	 */
3012 	kvmppc_set_hfscr_hv(vcpu, HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
3013 			    HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP);
3014 
3015 	/* On POWER10 and later, allow prefixed instructions */
3016 	if (cpu_has_feature(CPU_FTR_ARCH_31))
3017 		kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) | HFSCR_PREFIX);
3018 
3019 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
3020 		kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) & mfspr(SPRN_HFSCR));
3021 
3022 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
3023 		if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3024 			kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) | HFSCR_TM);
3025 #endif
3026 	}
3027 	if (cpu_has_feature(CPU_FTR_TM_COMP))
3028 		vcpu->arch.hfscr |= HFSCR_TM;
3029 
3030 	vcpu->arch.hfscr_permitted = kvmppc_get_hfscr_hv(vcpu);
3031 
3032 	/*
3033 	 * PM, EBB, TM are demand-faulted so start with it clear.
3034 	 */
3035 	kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) & ~(HFSCR_PM | HFSCR_EBB | HFSCR_TM));
3036 
3037 	kvmppc_mmu_book3s_hv_init(vcpu);
3038 
3039 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
3040 
3041 	init_waitqueue_head(&vcpu->arch.cpu_run);
3042 
3043 	mutex_lock(&kvm->lock);
3044 	vcore = NULL;
3045 	err = -EINVAL;
3046 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
3047 		if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
3048 			pr_devel("KVM: VCPU ID too high\n");
3049 			core = KVM_MAX_VCORES;
3050 		} else {
3051 			BUG_ON(kvm->arch.smt_mode != 1);
3052 			core = kvmppc_pack_vcpu_id(kvm, id);
3053 		}
3054 	} else {
3055 		core = id / kvm->arch.smt_mode;
3056 	}
3057 	if (core < KVM_MAX_VCORES) {
3058 		vcore = kvm->arch.vcores[core];
3059 		if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
3060 			pr_devel("KVM: collision on id %u", id);
3061 			vcore = NULL;
3062 		} else if (!vcore) {
3063 			/*
3064 			 * Take mmu_setup_lock for mutual exclusion
3065 			 * with kvmppc_update_lpcr().
3066 			 */
3067 			err = -ENOMEM;
3068 			vcore = kvmppc_vcore_create(kvm,
3069 					id & ~(kvm->arch.smt_mode - 1));
3070 			mutex_lock(&kvm->arch.mmu_setup_lock);
3071 			kvm->arch.vcores[core] = vcore;
3072 			kvm->arch.online_vcores++;
3073 			mutex_unlock(&kvm->arch.mmu_setup_lock);
3074 		}
3075 	}
3076 	mutex_unlock(&kvm->lock);
3077 
3078 	if (!vcore)
3079 		return err;
3080 
3081 	spin_lock(&vcore->lock);
3082 	++vcore->num_threads;
3083 	spin_unlock(&vcore->lock);
3084 	vcpu->arch.vcore = vcore;
3085 	vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
3086 	vcpu->arch.thread_cpu = -1;
3087 	vcpu->arch.prev_cpu = -1;
3088 
3089 	vcpu->arch.cpu_type = KVM_CPU_3S_64;
3090 	kvmppc_sanity_check(vcpu);
3091 
3092 	return 0;
3093 }
3094 
3095 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
3096 			      unsigned long flags)
3097 {
3098 	int err;
3099 	int esmt = 0;
3100 
3101 	if (flags)
3102 		return -EINVAL;
3103 	if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
3104 		return -EINVAL;
3105 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
3106 		/*
3107 		 * On POWER8 (or POWER7), the threading mode is "strict",
3108 		 * so we pack smt_mode vcpus per vcore.
3109 		 */
3110 		if (smt_mode > threads_per_subcore)
3111 			return -EINVAL;
3112 	} else {
3113 		/*
3114 		 * On POWER9, the threading mode is "loose",
3115 		 * so each vcpu gets its own vcore.
3116 		 */
3117 		esmt = smt_mode;
3118 		smt_mode = 1;
3119 	}
3120 	mutex_lock(&kvm->lock);
3121 	err = -EBUSY;
3122 	if (!kvm->arch.online_vcores) {
3123 		kvm->arch.smt_mode = smt_mode;
3124 		kvm->arch.emul_smt_mode = esmt;
3125 		err = 0;
3126 	}
3127 	mutex_unlock(&kvm->lock);
3128 
3129 	return err;
3130 }
3131 
3132 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
3133 {
3134 	if (vpa->pinned_addr)
3135 		kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
3136 					vpa->dirty);
3137 }
3138 
3139 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
3140 {
3141 	spin_lock(&vcpu->arch.vpa_update_lock);
3142 	unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
3143 	unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
3144 	unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
3145 	spin_unlock(&vcpu->arch.vpa_update_lock);
3146 	if (kvmhv_is_nestedv2())
3147 		kvmhv_nestedv2_vcpu_free(vcpu, &vcpu->arch.nestedv2_io);
3148 }
3149 
3150 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
3151 {
3152 	/* Indicate we want to get back into the guest */
3153 	return 1;
3154 }
3155 
3156 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
3157 {
3158 	unsigned long dec_nsec, now;
3159 
3160 	now = get_tb();
3161 	if (now > kvmppc_dec_expires_host_tb(vcpu)) {
3162 		/* decrementer has already gone negative */
3163 		kvmppc_core_queue_dec(vcpu);
3164 		kvmppc_core_prepare_to_enter(vcpu);
3165 		return;
3166 	}
3167 	dec_nsec = tb_to_ns(kvmppc_dec_expires_host_tb(vcpu) - now);
3168 	hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
3169 	vcpu->arch.timer_running = 1;
3170 }
3171 
3172 extern int __kvmppc_vcore_entry(void);
3173 
3174 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
3175 				   struct kvm_vcpu *vcpu, u64 tb)
3176 {
3177 	u64 now;
3178 
3179 	if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3180 		return;
3181 	spin_lock_irq(&vcpu->arch.tbacct_lock);
3182 	now = tb;
3183 	vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
3184 		vcpu->arch.stolen_logged;
3185 	vcpu->arch.busy_preempt = now;
3186 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
3187 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
3188 	--vc->n_runnable;
3189 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
3190 }
3191 
3192 static int kvmppc_grab_hwthread(int cpu)
3193 {
3194 	struct paca_struct *tpaca;
3195 	long timeout = 10000;
3196 
3197 	tpaca = paca_ptrs[cpu];
3198 
3199 	/* Ensure the thread won't go into the kernel if it wakes */
3200 	tpaca->kvm_hstate.kvm_vcpu = NULL;
3201 	tpaca->kvm_hstate.kvm_vcore = NULL;
3202 	tpaca->kvm_hstate.napping = 0;
3203 	smp_wmb();
3204 	tpaca->kvm_hstate.hwthread_req = 1;
3205 
3206 	/*
3207 	 * If the thread is already executing in the kernel (e.g. handling
3208 	 * a stray interrupt), wait for it to get back to nap mode.
3209 	 * The smp_mb() is to ensure that our setting of hwthread_req
3210 	 * is visible before we look at hwthread_state, so if this
3211 	 * races with the code at system_reset_pSeries and the thread
3212 	 * misses our setting of hwthread_req, we are sure to see its
3213 	 * setting of hwthread_state, and vice versa.
3214 	 */
3215 	smp_mb();
3216 	while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
3217 		if (--timeout <= 0) {
3218 			pr_err("KVM: couldn't grab cpu %d\n", cpu);
3219 			return -EBUSY;
3220 		}
3221 		udelay(1);
3222 	}
3223 	return 0;
3224 }
3225 
3226 static void kvmppc_release_hwthread(int cpu)
3227 {
3228 	struct paca_struct *tpaca;
3229 
3230 	tpaca = paca_ptrs[cpu];
3231 	tpaca->kvm_hstate.hwthread_req = 0;
3232 	tpaca->kvm_hstate.kvm_vcpu = NULL;
3233 	tpaca->kvm_hstate.kvm_vcore = NULL;
3234 	tpaca->kvm_hstate.kvm_split_mode = NULL;
3235 }
3236 
3237 static DEFINE_PER_CPU(struct kvm *, cpu_in_guest);
3238 
3239 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
3240 {
3241 	struct kvm_nested_guest *nested = vcpu->arch.nested;
3242 	cpumask_t *need_tlb_flush;
3243 	int i;
3244 
3245 	if (nested)
3246 		need_tlb_flush = &nested->need_tlb_flush;
3247 	else
3248 		need_tlb_flush = &kvm->arch.need_tlb_flush;
3249 
3250 	cpu = cpu_first_tlb_thread_sibling(cpu);
3251 	for (i = cpu; i <= cpu_last_tlb_thread_sibling(cpu);
3252 					i += cpu_tlb_thread_sibling_step())
3253 		cpumask_set_cpu(i, need_tlb_flush);
3254 
3255 	/*
3256 	 * Make sure setting of bit in need_tlb_flush precedes testing of
3257 	 * cpu_in_guest. The matching barrier on the other side is hwsync
3258 	 * when switching to guest MMU mode, which happens between
3259 	 * cpu_in_guest being set to the guest kvm, and need_tlb_flush bit
3260 	 * being tested.
3261 	 */
3262 	smp_mb();
3263 
3264 	for (i = cpu; i <= cpu_last_tlb_thread_sibling(cpu);
3265 					i += cpu_tlb_thread_sibling_step()) {
3266 		struct kvm *running = *per_cpu_ptr(&cpu_in_guest, i);
3267 
3268 		if (running == kvm)
3269 			smp_call_function_single(i, do_nothing, NULL, 1);
3270 	}
3271 }
3272 
3273 static void do_migrate_away_vcpu(void *arg)
3274 {
3275 	struct kvm_vcpu *vcpu = arg;
3276 	struct kvm *kvm = vcpu->kvm;
3277 
3278 	/*
3279 	 * If the guest has GTSE, it may execute tlbie, so do a eieio; tlbsync;
3280 	 * ptesync sequence on the old CPU before migrating to a new one, in
3281 	 * case we interrupted the guest between a tlbie ; eieio ;
3282 	 * tlbsync; ptesync sequence.
3283 	 *
3284 	 * Otherwise, ptesync is sufficient for ordering tlbiel sequences.
3285 	 */
3286 	if (kvm->arch.lpcr & LPCR_GTSE)
3287 		asm volatile("eieio; tlbsync; ptesync");
3288 	else
3289 		asm volatile("ptesync");
3290 }
3291 
3292 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
3293 {
3294 	struct kvm_nested_guest *nested = vcpu->arch.nested;
3295 	struct kvm *kvm = vcpu->kvm;
3296 	int prev_cpu;
3297 
3298 	if (!cpu_has_feature(CPU_FTR_HVMODE))
3299 		return;
3300 
3301 	if (nested)
3302 		prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
3303 	else
3304 		prev_cpu = vcpu->arch.prev_cpu;
3305 
3306 	/*
3307 	 * With radix, the guest can do TLB invalidations itself,
3308 	 * and it could choose to use the local form (tlbiel) if
3309 	 * it is invalidating a translation that has only ever been
3310 	 * used on one vcpu.  However, that doesn't mean it has
3311 	 * only ever been used on one physical cpu, since vcpus
3312 	 * can move around between pcpus.  To cope with this, when
3313 	 * a vcpu moves from one pcpu to another, we need to tell
3314 	 * any vcpus running on the same core as this vcpu previously
3315 	 * ran to flush the TLB.
3316 	 */
3317 	if (prev_cpu != pcpu) {
3318 		if (prev_cpu >= 0) {
3319 			if (cpu_first_tlb_thread_sibling(prev_cpu) !=
3320 			    cpu_first_tlb_thread_sibling(pcpu))
3321 				radix_flush_cpu(kvm, prev_cpu, vcpu);
3322 
3323 			smp_call_function_single(prev_cpu,
3324 					do_migrate_away_vcpu, vcpu, 1);
3325 		}
3326 		if (nested)
3327 			nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
3328 		else
3329 			vcpu->arch.prev_cpu = pcpu;
3330 	}
3331 }
3332 
3333 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
3334 {
3335 	int cpu;
3336 	struct paca_struct *tpaca;
3337 
3338 	cpu = vc->pcpu;
3339 	if (vcpu) {
3340 		if (vcpu->arch.timer_running) {
3341 			hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
3342 			vcpu->arch.timer_running = 0;
3343 		}
3344 		cpu += vcpu->arch.ptid;
3345 		vcpu->cpu = vc->pcpu;
3346 		vcpu->arch.thread_cpu = cpu;
3347 	}
3348 	tpaca = paca_ptrs[cpu];
3349 	tpaca->kvm_hstate.kvm_vcpu = vcpu;
3350 	tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
3351 	tpaca->kvm_hstate.fake_suspend = 0;
3352 	/* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
3353 	smp_wmb();
3354 	tpaca->kvm_hstate.kvm_vcore = vc;
3355 	if (cpu != smp_processor_id())
3356 		kvmppc_ipi_thread(cpu);
3357 }
3358 
3359 static void kvmppc_wait_for_nap(int n_threads)
3360 {
3361 	int cpu = smp_processor_id();
3362 	int i, loops;
3363 
3364 	if (n_threads <= 1)
3365 		return;
3366 	for (loops = 0; loops < 1000000; ++loops) {
3367 		/*
3368 		 * Check if all threads are finished.
3369 		 * We set the vcore pointer when starting a thread
3370 		 * and the thread clears it when finished, so we look
3371 		 * for any threads that still have a non-NULL vcore ptr.
3372 		 */
3373 		for (i = 1; i < n_threads; ++i)
3374 			if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
3375 				break;
3376 		if (i == n_threads) {
3377 			HMT_medium();
3378 			return;
3379 		}
3380 		HMT_low();
3381 	}
3382 	HMT_medium();
3383 	for (i = 1; i < n_threads; ++i)
3384 		if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
3385 			pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
3386 }
3387 
3388 /*
3389  * Check that we are on thread 0 and that any other threads in
3390  * this core are off-line.  Then grab the threads so they can't
3391  * enter the kernel.
3392  */
3393 static int on_primary_thread(void)
3394 {
3395 	int cpu = smp_processor_id();
3396 	int thr;
3397 
3398 	/* Are we on a primary subcore? */
3399 	if (cpu_thread_in_subcore(cpu))
3400 		return 0;
3401 
3402 	thr = 0;
3403 	while (++thr < threads_per_subcore)
3404 		if (cpu_online(cpu + thr))
3405 			return 0;
3406 
3407 	/* Grab all hw threads so they can't go into the kernel */
3408 	for (thr = 1; thr < threads_per_subcore; ++thr) {
3409 		if (kvmppc_grab_hwthread(cpu + thr)) {
3410 			/* Couldn't grab one; let the others go */
3411 			do {
3412 				kvmppc_release_hwthread(cpu + thr);
3413 			} while (--thr > 0);
3414 			return 0;
3415 		}
3416 	}
3417 	return 1;
3418 }
3419 
3420 /*
3421  * A list of virtual cores for each physical CPU.
3422  * These are vcores that could run but their runner VCPU tasks are
3423  * (or may be) preempted.
3424  */
3425 struct preempted_vcore_list {
3426 	struct list_head	list;
3427 	spinlock_t		lock;
3428 };
3429 
3430 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
3431 
3432 static void init_vcore_lists(void)
3433 {
3434 	int cpu;
3435 
3436 	for_each_possible_cpu(cpu) {
3437 		struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
3438 		spin_lock_init(&lp->lock);
3439 		INIT_LIST_HEAD(&lp->list);
3440 	}
3441 }
3442 
3443 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
3444 {
3445 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
3446 
3447 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
3448 
3449 	vc->vcore_state = VCORE_PREEMPT;
3450 	vc->pcpu = smp_processor_id();
3451 	if (vc->num_threads < threads_per_vcore(vc->kvm)) {
3452 		spin_lock(&lp->lock);
3453 		list_add_tail(&vc->preempt_list, &lp->list);
3454 		spin_unlock(&lp->lock);
3455 	}
3456 
3457 	/* Start accumulating stolen time */
3458 	kvmppc_core_start_stolen(vc, mftb());
3459 }
3460 
3461 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
3462 {
3463 	struct preempted_vcore_list *lp;
3464 
3465 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
3466 
3467 	kvmppc_core_end_stolen(vc, mftb());
3468 	if (!list_empty(&vc->preempt_list)) {
3469 		lp = &per_cpu(preempted_vcores, vc->pcpu);
3470 		spin_lock(&lp->lock);
3471 		list_del_init(&vc->preempt_list);
3472 		spin_unlock(&lp->lock);
3473 	}
3474 	vc->vcore_state = VCORE_INACTIVE;
3475 }
3476 
3477 /*
3478  * This stores information about the virtual cores currently
3479  * assigned to a physical core.
3480  */
3481 struct core_info {
3482 	int		n_subcores;
3483 	int		max_subcore_threads;
3484 	int		total_threads;
3485 	int		subcore_threads[MAX_SUBCORES];
3486 	struct kvmppc_vcore *vc[MAX_SUBCORES];
3487 };
3488 
3489 /*
3490  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
3491  * respectively in 2-way micro-threading (split-core) mode on POWER8.
3492  */
3493 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
3494 
3495 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
3496 {
3497 	memset(cip, 0, sizeof(*cip));
3498 	cip->n_subcores = 1;
3499 	cip->max_subcore_threads = vc->num_threads;
3500 	cip->total_threads = vc->num_threads;
3501 	cip->subcore_threads[0] = vc->num_threads;
3502 	cip->vc[0] = vc;
3503 }
3504 
3505 static bool subcore_config_ok(int n_subcores, int n_threads)
3506 {
3507 	/*
3508 	 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
3509 	 * split-core mode, with one thread per subcore.
3510 	 */
3511 	if (cpu_has_feature(CPU_FTR_ARCH_300))
3512 		return n_subcores <= 4 && n_threads == 1;
3513 
3514 	/* On POWER8, can only dynamically split if unsplit to begin with */
3515 	if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
3516 		return false;
3517 	if (n_subcores > MAX_SUBCORES)
3518 		return false;
3519 	if (n_subcores > 1) {
3520 		if (!(dynamic_mt_modes & 2))
3521 			n_subcores = 4;
3522 		if (n_subcores > 2 && !(dynamic_mt_modes & 4))
3523 			return false;
3524 	}
3525 
3526 	return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
3527 }
3528 
3529 static void init_vcore_to_run(struct kvmppc_vcore *vc)
3530 {
3531 	vc->entry_exit_map = 0;
3532 	vc->in_guest = 0;
3533 	vc->napping_threads = 0;
3534 	vc->conferring_threads = 0;
3535 	vc->tb_offset_applied = 0;
3536 }
3537 
3538 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
3539 {
3540 	int n_threads = vc->num_threads;
3541 	int sub;
3542 
3543 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
3544 		return false;
3545 
3546 	/* In one_vm_per_core mode, require all vcores to be from the same vm */
3547 	if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
3548 		return false;
3549 
3550 	if (n_threads < cip->max_subcore_threads)
3551 		n_threads = cip->max_subcore_threads;
3552 	if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
3553 		return false;
3554 	cip->max_subcore_threads = n_threads;
3555 
3556 	sub = cip->n_subcores;
3557 	++cip->n_subcores;
3558 	cip->total_threads += vc->num_threads;
3559 	cip->subcore_threads[sub] = vc->num_threads;
3560 	cip->vc[sub] = vc;
3561 	init_vcore_to_run(vc);
3562 	list_del_init(&vc->preempt_list);
3563 
3564 	return true;
3565 }
3566 
3567 /*
3568  * Work out whether it is possible to piggyback the execution of
3569  * vcore *pvc onto the execution of the other vcores described in *cip.
3570  */
3571 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
3572 			  int target_threads)
3573 {
3574 	if (cip->total_threads + pvc->num_threads > target_threads)
3575 		return false;
3576 
3577 	return can_dynamic_split(pvc, cip);
3578 }
3579 
3580 static void prepare_threads(struct kvmppc_vcore *vc)
3581 {
3582 	int i;
3583 	struct kvm_vcpu *vcpu;
3584 
3585 	for_each_runnable_thread(i, vcpu, vc) {
3586 		if (signal_pending(vcpu->arch.run_task))
3587 			vcpu->arch.ret = -EINTR;
3588 		else if (vcpu->arch.vpa.update_pending ||
3589 			 vcpu->arch.slb_shadow.update_pending ||
3590 			 vcpu->arch.dtl.update_pending)
3591 			vcpu->arch.ret = RESUME_GUEST;
3592 		else
3593 			continue;
3594 		kvmppc_remove_runnable(vc, vcpu, mftb());
3595 		wake_up(&vcpu->arch.cpu_run);
3596 	}
3597 }
3598 
3599 static void collect_piggybacks(struct core_info *cip, int target_threads)
3600 {
3601 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
3602 	struct kvmppc_vcore *pvc, *vcnext;
3603 
3604 	spin_lock(&lp->lock);
3605 	list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
3606 		if (!spin_trylock(&pvc->lock))
3607 			continue;
3608 		prepare_threads(pvc);
3609 		if (!pvc->n_runnable || !pvc->kvm->arch.mmu_ready) {
3610 			list_del_init(&pvc->preempt_list);
3611 			if (pvc->runner == NULL) {
3612 				pvc->vcore_state = VCORE_INACTIVE;
3613 				kvmppc_core_end_stolen(pvc, mftb());
3614 			}
3615 			spin_unlock(&pvc->lock);
3616 			continue;
3617 		}
3618 		if (!can_piggyback(pvc, cip, target_threads)) {
3619 			spin_unlock(&pvc->lock);
3620 			continue;
3621 		}
3622 		kvmppc_core_end_stolen(pvc, mftb());
3623 		pvc->vcore_state = VCORE_PIGGYBACK;
3624 		if (cip->total_threads >= target_threads)
3625 			break;
3626 	}
3627 	spin_unlock(&lp->lock);
3628 }
3629 
3630 static bool recheck_signals_and_mmu(struct core_info *cip)
3631 {
3632 	int sub, i;
3633 	struct kvm_vcpu *vcpu;
3634 	struct kvmppc_vcore *vc;
3635 
3636 	for (sub = 0; sub < cip->n_subcores; ++sub) {
3637 		vc = cip->vc[sub];
3638 		if (!vc->kvm->arch.mmu_ready)
3639 			return true;
3640 		for_each_runnable_thread(i, vcpu, vc)
3641 			if (signal_pending(vcpu->arch.run_task))
3642 				return true;
3643 	}
3644 	return false;
3645 }
3646 
3647 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
3648 {
3649 	int still_running = 0, i;
3650 	u64 now;
3651 	long ret;
3652 	struct kvm_vcpu *vcpu;
3653 
3654 	spin_lock(&vc->lock);
3655 	now = get_tb();
3656 	for_each_runnable_thread(i, vcpu, vc) {
3657 		/*
3658 		 * It's safe to unlock the vcore in the loop here, because
3659 		 * for_each_runnable_thread() is safe against removal of
3660 		 * the vcpu, and the vcore state is VCORE_EXITING here,
3661 		 * so any vcpus becoming runnable will have their arch.trap
3662 		 * set to zero and can't actually run in the guest.
3663 		 */
3664 		spin_unlock(&vc->lock);
3665 		/* cancel pending dec exception if dec is positive */
3666 		if (now < kvmppc_dec_expires_host_tb(vcpu) &&
3667 		    kvmppc_core_pending_dec(vcpu))
3668 			kvmppc_core_dequeue_dec(vcpu);
3669 
3670 		trace_kvm_guest_exit(vcpu);
3671 
3672 		ret = RESUME_GUEST;
3673 		if (vcpu->arch.trap)
3674 			ret = kvmppc_handle_exit_hv(vcpu,
3675 						    vcpu->arch.run_task);
3676 
3677 		vcpu->arch.ret = ret;
3678 		vcpu->arch.trap = 0;
3679 
3680 		spin_lock(&vc->lock);
3681 		if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
3682 			if (vcpu->arch.pending_exceptions)
3683 				kvmppc_core_prepare_to_enter(vcpu);
3684 			if (vcpu->arch.ceded)
3685 				kvmppc_set_timer(vcpu);
3686 			else
3687 				++still_running;
3688 		} else {
3689 			kvmppc_remove_runnable(vc, vcpu, mftb());
3690 			wake_up(&vcpu->arch.cpu_run);
3691 		}
3692 	}
3693 	if (!is_master) {
3694 		if (still_running > 0) {
3695 			kvmppc_vcore_preempt(vc);
3696 		} else if (vc->runner) {
3697 			vc->vcore_state = VCORE_PREEMPT;
3698 			kvmppc_core_start_stolen(vc, mftb());
3699 		} else {
3700 			vc->vcore_state = VCORE_INACTIVE;
3701 		}
3702 		if (vc->n_runnable > 0 && vc->runner == NULL) {
3703 			/* make sure there's a candidate runner awake */
3704 			i = -1;
3705 			vcpu = next_runnable_thread(vc, &i);
3706 			wake_up(&vcpu->arch.cpu_run);
3707 		}
3708 	}
3709 	spin_unlock(&vc->lock);
3710 }
3711 
3712 /*
3713  * Clear core from the list of active host cores as we are about to
3714  * enter the guest. Only do this if it is the primary thread of the
3715  * core (not if a subcore) that is entering the guest.
3716  */
3717 static inline int kvmppc_clear_host_core(unsigned int cpu)
3718 {
3719 	int core;
3720 
3721 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3722 		return 0;
3723 	/*
3724 	 * Memory barrier can be omitted here as we will do a smp_wmb()
3725 	 * later in kvmppc_start_thread and we need ensure that state is
3726 	 * visible to other CPUs only after we enter guest.
3727 	 */
3728 	core = cpu >> threads_shift;
3729 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
3730 	return 0;
3731 }
3732 
3733 /*
3734  * Advertise this core as an active host core since we exited the guest
3735  * Only need to do this if it is the primary thread of the core that is
3736  * exiting.
3737  */
3738 static inline int kvmppc_set_host_core(unsigned int cpu)
3739 {
3740 	int core;
3741 
3742 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3743 		return 0;
3744 
3745 	/*
3746 	 * Memory barrier can be omitted here because we do a spin_unlock
3747 	 * immediately after this which provides the memory barrier.
3748 	 */
3749 	core = cpu >> threads_shift;
3750 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
3751 	return 0;
3752 }
3753 
3754 static void set_irq_happened(int trap)
3755 {
3756 	switch (trap) {
3757 	case BOOK3S_INTERRUPT_EXTERNAL:
3758 		local_paca->irq_happened |= PACA_IRQ_EE;
3759 		break;
3760 	case BOOK3S_INTERRUPT_H_DOORBELL:
3761 		local_paca->irq_happened |= PACA_IRQ_DBELL;
3762 		break;
3763 	case BOOK3S_INTERRUPT_HMI:
3764 		local_paca->irq_happened |= PACA_IRQ_HMI;
3765 		break;
3766 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
3767 		replay_system_reset();
3768 		break;
3769 	}
3770 }
3771 
3772 /*
3773  * Run a set of guest threads on a physical core.
3774  * Called with vc->lock held.
3775  */
3776 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
3777 {
3778 	struct kvm_vcpu *vcpu;
3779 	int i;
3780 	int srcu_idx;
3781 	struct core_info core_info;
3782 	struct kvmppc_vcore *pvc;
3783 	struct kvm_split_mode split_info, *sip;
3784 	int split, subcore_size, active;
3785 	int sub;
3786 	bool thr0_done;
3787 	unsigned long cmd_bit, stat_bit;
3788 	int pcpu, thr;
3789 	int target_threads;
3790 	int controlled_threads;
3791 	int trap;
3792 	bool is_power8;
3793 
3794 	if (WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300)))
3795 		return;
3796 
3797 	/*
3798 	 * Remove from the list any threads that have a signal pending
3799 	 * or need a VPA update done
3800 	 */
3801 	prepare_threads(vc);
3802 
3803 	/* if the runner is no longer runnable, let the caller pick a new one */
3804 	if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3805 		return;
3806 
3807 	/*
3808 	 * Initialize *vc.
3809 	 */
3810 	init_vcore_to_run(vc);
3811 	vc->preempt_tb = TB_NIL;
3812 
3813 	/*
3814 	 * Number of threads that we will be controlling: the same as
3815 	 * the number of threads per subcore, except on POWER9,
3816 	 * where it's 1 because the threads are (mostly) independent.
3817 	 */
3818 	controlled_threads = threads_per_vcore(vc->kvm);
3819 
3820 	/*
3821 	 * Make sure we are running on primary threads, and that secondary
3822 	 * threads are offline.  Also check if the number of threads in this
3823 	 * guest are greater than the current system threads per guest.
3824 	 */
3825 	if ((controlled_threads > 1) &&
3826 	    ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
3827 		for_each_runnable_thread(i, vcpu, vc) {
3828 			vcpu->arch.ret = -EBUSY;
3829 			kvmppc_remove_runnable(vc, vcpu, mftb());
3830 			wake_up(&vcpu->arch.cpu_run);
3831 		}
3832 		goto out;
3833 	}
3834 
3835 	/*
3836 	 * See if we could run any other vcores on the physical core
3837 	 * along with this one.
3838 	 */
3839 	init_core_info(&core_info, vc);
3840 	pcpu = smp_processor_id();
3841 	target_threads = controlled_threads;
3842 	if (target_smt_mode && target_smt_mode < target_threads)
3843 		target_threads = target_smt_mode;
3844 	if (vc->num_threads < target_threads)
3845 		collect_piggybacks(&core_info, target_threads);
3846 
3847 	/*
3848 	 * Hard-disable interrupts, and check resched flag and signals.
3849 	 * If we need to reschedule or deliver a signal, clean up
3850 	 * and return without going into the guest(s).
3851 	 * If the mmu_ready flag has been cleared, don't go into the
3852 	 * guest because that means a HPT resize operation is in progress.
3853 	 */
3854 	local_irq_disable();
3855 	hard_irq_disable();
3856 	if (lazy_irq_pending() || need_resched() ||
3857 	    recheck_signals_and_mmu(&core_info)) {
3858 		local_irq_enable();
3859 		vc->vcore_state = VCORE_INACTIVE;
3860 		/* Unlock all except the primary vcore */
3861 		for (sub = 1; sub < core_info.n_subcores; ++sub) {
3862 			pvc = core_info.vc[sub];
3863 			/* Put back on to the preempted vcores list */
3864 			kvmppc_vcore_preempt(pvc);
3865 			spin_unlock(&pvc->lock);
3866 		}
3867 		for (i = 0; i < controlled_threads; ++i)
3868 			kvmppc_release_hwthread(pcpu + i);
3869 		return;
3870 	}
3871 
3872 	kvmppc_clear_host_core(pcpu);
3873 
3874 	/* Decide on micro-threading (split-core) mode */
3875 	subcore_size = threads_per_subcore;
3876 	cmd_bit = stat_bit = 0;
3877 	split = core_info.n_subcores;
3878 	sip = NULL;
3879 	is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S);
3880 
3881 	if (split > 1) {
3882 		sip = &split_info;
3883 		memset(&split_info, 0, sizeof(split_info));
3884 		for (sub = 0; sub < core_info.n_subcores; ++sub)
3885 			split_info.vc[sub] = core_info.vc[sub];
3886 
3887 		if (is_power8) {
3888 			if (split == 2 && (dynamic_mt_modes & 2)) {
3889 				cmd_bit = HID0_POWER8_1TO2LPAR;
3890 				stat_bit = HID0_POWER8_2LPARMODE;
3891 			} else {
3892 				split = 4;
3893 				cmd_bit = HID0_POWER8_1TO4LPAR;
3894 				stat_bit = HID0_POWER8_4LPARMODE;
3895 			}
3896 			subcore_size = MAX_SMT_THREADS / split;
3897 			split_info.rpr = mfspr(SPRN_RPR);
3898 			split_info.pmmar = mfspr(SPRN_PMMAR);
3899 			split_info.ldbar = mfspr(SPRN_LDBAR);
3900 			split_info.subcore_size = subcore_size;
3901 		} else {
3902 			split_info.subcore_size = 1;
3903 		}
3904 
3905 		/* order writes to split_info before kvm_split_mode pointer */
3906 		smp_wmb();
3907 	}
3908 
3909 	for (thr = 0; thr < controlled_threads; ++thr) {
3910 		struct paca_struct *paca = paca_ptrs[pcpu + thr];
3911 
3912 		paca->kvm_hstate.napping = 0;
3913 		paca->kvm_hstate.kvm_split_mode = sip;
3914 	}
3915 
3916 	/* Initiate micro-threading (split-core) on POWER8 if required */
3917 	if (cmd_bit) {
3918 		unsigned long hid0 = mfspr(SPRN_HID0);
3919 
3920 		hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3921 		mb();
3922 		mtspr(SPRN_HID0, hid0);
3923 		isync();
3924 		for (;;) {
3925 			hid0 = mfspr(SPRN_HID0);
3926 			if (hid0 & stat_bit)
3927 				break;
3928 			cpu_relax();
3929 		}
3930 	}
3931 
3932 	/*
3933 	 * On POWER8, set RWMR register.
3934 	 * Since it only affects PURR and SPURR, it doesn't affect
3935 	 * the host, so we don't save/restore the host value.
3936 	 */
3937 	if (is_power8) {
3938 		unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3939 		int n_online = atomic_read(&vc->online_count);
3940 
3941 		/*
3942 		 * Use the 8-thread value if we're doing split-core
3943 		 * or if the vcore's online count looks bogus.
3944 		 */
3945 		if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3946 		    n_online >= 1 && n_online <= MAX_SMT_THREADS)
3947 			rwmr_val = p8_rwmr_values[n_online];
3948 		mtspr(SPRN_RWMR, rwmr_val);
3949 	}
3950 
3951 	/* Start all the threads */
3952 	active = 0;
3953 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
3954 		thr = is_power8 ? subcore_thread_map[sub] : sub;
3955 		thr0_done = false;
3956 		active |= 1 << thr;
3957 		pvc = core_info.vc[sub];
3958 		pvc->pcpu = pcpu + thr;
3959 		for_each_runnable_thread(i, vcpu, pvc) {
3960 			/*
3961 			 * XXX: is kvmppc_start_thread called too late here?
3962 			 * It updates vcpu->cpu and vcpu->arch.thread_cpu
3963 			 * which are used by kvmppc_fast_vcpu_kick_hv(), but
3964 			 * kick is called after new exceptions become available
3965 			 * and exceptions are checked earlier than here, by
3966 			 * kvmppc_core_prepare_to_enter.
3967 			 */
3968 			kvmppc_start_thread(vcpu, pvc);
3969 			kvmppc_update_vpa_dispatch(vcpu, pvc);
3970 			trace_kvm_guest_enter(vcpu);
3971 			if (!vcpu->arch.ptid)
3972 				thr0_done = true;
3973 			active |= 1 << (thr + vcpu->arch.ptid);
3974 		}
3975 		/*
3976 		 * We need to start the first thread of each subcore
3977 		 * even if it doesn't have a vcpu.
3978 		 */
3979 		if (!thr0_done)
3980 			kvmppc_start_thread(NULL, pvc);
3981 	}
3982 
3983 	/*
3984 	 * Ensure that split_info.do_nap is set after setting
3985 	 * the vcore pointer in the PACA of the secondaries.
3986 	 */
3987 	smp_mb();
3988 
3989 	/*
3990 	 * When doing micro-threading, poke the inactive threads as well.
3991 	 * This gets them to the nap instruction after kvm_do_nap,
3992 	 * which reduces the time taken to unsplit later.
3993 	 */
3994 	if (cmd_bit) {
3995 		split_info.do_nap = 1;	/* ask secondaries to nap when done */
3996 		for (thr = 1; thr < threads_per_subcore; ++thr)
3997 			if (!(active & (1 << thr)))
3998 				kvmppc_ipi_thread(pcpu + thr);
3999 	}
4000 
4001 	vc->vcore_state = VCORE_RUNNING;
4002 	preempt_disable();
4003 
4004 	trace_kvmppc_run_core(vc, 0);
4005 
4006 	for (sub = 0; sub < core_info.n_subcores; ++sub)
4007 		spin_unlock(&core_info.vc[sub]->lock);
4008 
4009 	guest_timing_enter_irqoff();
4010 
4011 	srcu_idx = srcu_read_lock(&vc->kvm->srcu);
4012 
4013 	guest_state_enter_irqoff();
4014 	this_cpu_disable_ftrace();
4015 
4016 	trap = __kvmppc_vcore_entry();
4017 
4018 	this_cpu_enable_ftrace();
4019 	guest_state_exit_irqoff();
4020 
4021 	srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
4022 
4023 	set_irq_happened(trap);
4024 
4025 	spin_lock(&vc->lock);
4026 	/* prevent other vcpu threads from doing kvmppc_start_thread() now */
4027 	vc->vcore_state = VCORE_EXITING;
4028 
4029 	/* wait for secondary threads to finish writing their state to memory */
4030 	kvmppc_wait_for_nap(controlled_threads);
4031 
4032 	/* Return to whole-core mode if we split the core earlier */
4033 	if (cmd_bit) {
4034 		unsigned long hid0 = mfspr(SPRN_HID0);
4035 
4036 		hid0 &= ~HID0_POWER8_DYNLPARDIS;
4037 		stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
4038 		mb();
4039 		mtspr(SPRN_HID0, hid0);
4040 		isync();
4041 		for (;;) {
4042 			hid0 = mfspr(SPRN_HID0);
4043 			if (!(hid0 & stat_bit))
4044 				break;
4045 			cpu_relax();
4046 		}
4047 		split_info.do_nap = 0;
4048 	}
4049 
4050 	kvmppc_set_host_core(pcpu);
4051 
4052 	if (!vtime_accounting_enabled_this_cpu()) {
4053 		local_irq_enable();
4054 		/*
4055 		 * Service IRQs here before guest_timing_exit_irqoff() so any
4056 		 * ticks that occurred while running the guest are accounted to
4057 		 * the guest. If vtime accounting is enabled, accounting uses
4058 		 * TB rather than ticks, so it can be done without enabling
4059 		 * interrupts here, which has the problem that it accounts
4060 		 * interrupt processing overhead to the host.
4061 		 */
4062 		local_irq_disable();
4063 	}
4064 	guest_timing_exit_irqoff();
4065 
4066 	local_irq_enable();
4067 
4068 	/* Let secondaries go back to the offline loop */
4069 	for (i = 0; i < controlled_threads; ++i) {
4070 		kvmppc_release_hwthread(pcpu + i);
4071 		if (sip && sip->napped[i])
4072 			kvmppc_ipi_thread(pcpu + i);
4073 	}
4074 
4075 	spin_unlock(&vc->lock);
4076 
4077 	/* make sure updates to secondary vcpu structs are visible now */
4078 	smp_mb();
4079 
4080 	preempt_enable();
4081 
4082 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
4083 		pvc = core_info.vc[sub];
4084 		post_guest_process(pvc, pvc == vc);
4085 	}
4086 
4087 	spin_lock(&vc->lock);
4088 
4089  out:
4090 	vc->vcore_state = VCORE_INACTIVE;
4091 	trace_kvmppc_run_core(vc, 1);
4092 }
4093 
4094 static inline bool hcall_is_xics(unsigned long req)
4095 {
4096 	return req == H_EOI || req == H_CPPR || req == H_IPI ||
4097 		req == H_IPOLL || req == H_XIRR || req == H_XIRR_X;
4098 }
4099 
4100 static void vcpu_vpa_increment_dispatch(struct kvm_vcpu *vcpu)
4101 {
4102 	struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
4103 	if (lp) {
4104 		u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
4105 		lp->yield_count = cpu_to_be32(yield_count);
4106 		vcpu->arch.vpa.dirty = 1;
4107 	}
4108 }
4109 
4110 /* Helper functions for reading L2's stats from L1's VPA */
4111 #ifdef CONFIG_PPC_PSERIES
4112 static DEFINE_PER_CPU(u64, l1_to_l2_cs);
4113 static DEFINE_PER_CPU(u64, l2_to_l1_cs);
4114 static DEFINE_PER_CPU(u64, l2_runtime_agg);
4115 
4116 int kvmhv_get_l2_counters_status(void)
4117 {
4118 	return firmware_has_feature(FW_FEATURE_LPAR) &&
4119 		get_lppaca()->l2_counters_enable;
4120 }
4121 
4122 void kvmhv_set_l2_counters_status(int cpu, bool status)
4123 {
4124 	if (!firmware_has_feature(FW_FEATURE_LPAR))
4125 		return;
4126 	if (status)
4127 		lppaca_of(cpu).l2_counters_enable = 1;
4128 	else
4129 		lppaca_of(cpu).l2_counters_enable = 0;
4130 }
4131 EXPORT_SYMBOL(kvmhv_set_l2_counters_status);
4132 
4133 int kvmhv_counters_tracepoint_regfunc(void)
4134 {
4135 	int cpu;
4136 
4137 	for_each_present_cpu(cpu) {
4138 		kvmhv_set_l2_counters_status(cpu, true);
4139 	}
4140 	return 0;
4141 }
4142 
4143 void kvmhv_counters_tracepoint_unregfunc(void)
4144 {
4145 	int cpu;
4146 
4147 	for_each_present_cpu(cpu) {
4148 		kvmhv_set_l2_counters_status(cpu, false);
4149 	}
4150 }
4151 
4152 static void do_trace_nested_cs_time(struct kvm_vcpu *vcpu)
4153 {
4154 	struct lppaca *lp = get_lppaca();
4155 	u64 l1_to_l2_ns, l2_to_l1_ns, l2_runtime_ns;
4156 	u64 *l1_to_l2_cs_ptr = this_cpu_ptr(&l1_to_l2_cs);
4157 	u64 *l2_to_l1_cs_ptr = this_cpu_ptr(&l2_to_l1_cs);
4158 	u64 *l2_runtime_agg_ptr = this_cpu_ptr(&l2_runtime_agg);
4159 
4160 	l1_to_l2_ns = tb_to_ns(be64_to_cpu(lp->l1_to_l2_cs_tb));
4161 	l2_to_l1_ns = tb_to_ns(be64_to_cpu(lp->l2_to_l1_cs_tb));
4162 	l2_runtime_ns = tb_to_ns(be64_to_cpu(lp->l2_runtime_tb));
4163 	trace_kvmppc_vcpu_stats(vcpu, l1_to_l2_ns - *l1_to_l2_cs_ptr,
4164 					l2_to_l1_ns - *l2_to_l1_cs_ptr,
4165 					l2_runtime_ns - *l2_runtime_agg_ptr);
4166 	*l1_to_l2_cs_ptr = l1_to_l2_ns;
4167 	*l2_to_l1_cs_ptr = l2_to_l1_ns;
4168 	*l2_runtime_agg_ptr = l2_runtime_ns;
4169 	vcpu->arch.l1_to_l2_cs = l1_to_l2_ns;
4170 	vcpu->arch.l2_to_l1_cs = l2_to_l1_ns;
4171 	vcpu->arch.l2_runtime_agg = l2_runtime_ns;
4172 }
4173 
4174 u64 kvmhv_get_l1_to_l2_cs_time(void)
4175 {
4176 	return tb_to_ns(be64_to_cpu(get_lppaca()->l1_to_l2_cs_tb));
4177 }
4178 EXPORT_SYMBOL(kvmhv_get_l1_to_l2_cs_time);
4179 
4180 u64 kvmhv_get_l2_to_l1_cs_time(void)
4181 {
4182 	return tb_to_ns(be64_to_cpu(get_lppaca()->l2_to_l1_cs_tb));
4183 }
4184 EXPORT_SYMBOL(kvmhv_get_l2_to_l1_cs_time);
4185 
4186 u64 kvmhv_get_l2_runtime_agg(void)
4187 {
4188 	return tb_to_ns(be64_to_cpu(get_lppaca()->l2_runtime_tb));
4189 }
4190 EXPORT_SYMBOL(kvmhv_get_l2_runtime_agg);
4191 
4192 u64 kvmhv_get_l1_to_l2_cs_time_vcpu(void)
4193 {
4194 	struct kvm_vcpu *vcpu;
4195 	struct kvm_vcpu_arch *arch;
4196 
4197 	vcpu = local_paca->kvm_hstate.kvm_vcpu;
4198 	if (vcpu) {
4199 		arch = &vcpu->arch;
4200 		return arch->l1_to_l2_cs;
4201 	} else {
4202 		return 0;
4203 	}
4204 }
4205 EXPORT_SYMBOL(kvmhv_get_l1_to_l2_cs_time_vcpu);
4206 
4207 u64 kvmhv_get_l2_to_l1_cs_time_vcpu(void)
4208 {
4209 	struct kvm_vcpu *vcpu;
4210 	struct kvm_vcpu_arch *arch;
4211 
4212 	vcpu = local_paca->kvm_hstate.kvm_vcpu;
4213 	if (vcpu) {
4214 		arch = &vcpu->arch;
4215 		return arch->l2_to_l1_cs;
4216 	} else {
4217 		return 0;
4218 	}
4219 }
4220 EXPORT_SYMBOL(kvmhv_get_l2_to_l1_cs_time_vcpu);
4221 
4222 u64 kvmhv_get_l2_runtime_agg_vcpu(void)
4223 {
4224 	struct kvm_vcpu *vcpu;
4225 	struct kvm_vcpu_arch *arch;
4226 
4227 	vcpu = local_paca->kvm_hstate.kvm_vcpu;
4228 	if (vcpu) {
4229 		arch = &vcpu->arch;
4230 		return arch->l2_runtime_agg;
4231 	} else {
4232 		return 0;
4233 	}
4234 }
4235 EXPORT_SYMBOL(kvmhv_get_l2_runtime_agg_vcpu);
4236 
4237 #else
4238 int kvmhv_get_l2_counters_status(void)
4239 {
4240 	return 0;
4241 }
4242 
4243 static void do_trace_nested_cs_time(struct kvm_vcpu *vcpu)
4244 {
4245 }
4246 #endif
4247 
4248 static int kvmhv_vcpu_entry_nestedv2(struct kvm_vcpu *vcpu, u64 time_limit,
4249 				     unsigned long lpcr, u64 *tb)
4250 {
4251 	struct kvmhv_nestedv2_io *io;
4252 	unsigned long msr, i;
4253 	int trap;
4254 	long rc;
4255 
4256 	if (vcpu->arch.doorbell_request) {
4257 		vcpu->arch.doorbell_request = 0;
4258 		kvmppc_set_dpdes(vcpu, 1);
4259 	}
4260 
4261 	io = &vcpu->arch.nestedv2_io;
4262 
4263 	msr = mfmsr();
4264 	kvmppc_msr_hard_disable_set_facilities(vcpu, msr);
4265 	if (lazy_irq_pending())
4266 		return 0;
4267 
4268 	rc = kvmhv_nestedv2_flush_vcpu(vcpu, time_limit);
4269 	if (rc < 0)
4270 		return -EINVAL;
4271 
4272 	kvmppc_gse_put_u64(io->vcpu_run_input, KVMPPC_GSID_LPCR, lpcr);
4273 
4274 	accumulate_time(vcpu, &vcpu->arch.in_guest);
4275 	rc = plpar_guest_run_vcpu(0, vcpu->kvm->arch.lpid, vcpu->vcpu_id,
4276 				  &trap, &i);
4277 
4278 	if (rc != H_SUCCESS) {
4279 		pr_err("KVM Guest Run VCPU hcall failed\n");
4280 		if (rc == H_INVALID_ELEMENT_ID)
4281 			pr_err("KVM: Guest Run VCPU invalid element id at %ld\n", i);
4282 		else if (rc == H_INVALID_ELEMENT_SIZE)
4283 			pr_err("KVM: Guest Run VCPU invalid element size at %ld\n", i);
4284 		else if (rc == H_INVALID_ELEMENT_VALUE)
4285 			pr_err("KVM: Guest Run VCPU invalid element value at %ld\n", i);
4286 		return -EINVAL;
4287 	}
4288 	accumulate_time(vcpu, &vcpu->arch.guest_exit);
4289 
4290 	*tb = mftb();
4291 	kvmppc_gsm_reset(io->vcpu_message);
4292 	kvmppc_gsm_reset(io->vcore_message);
4293 	kvmppc_gsbm_zero(&io->valids);
4294 
4295 	rc = kvmhv_nestedv2_parse_output(vcpu);
4296 	if (rc < 0)
4297 		return -EINVAL;
4298 
4299 	timer_rearm_host_dec(*tb);
4300 
4301 	/* Record context switch and guest_run_time data */
4302 	if (kvmhv_get_l2_counters_status())
4303 		do_trace_nested_cs_time(vcpu);
4304 
4305 	return trap;
4306 }
4307 
4308 /* call our hypervisor to load up HV regs and go */
4309 static int kvmhv_vcpu_entry_p9_nested(struct kvm_vcpu *vcpu, u64 time_limit, unsigned long lpcr, u64 *tb)
4310 {
4311 	unsigned long host_psscr;
4312 	unsigned long msr;
4313 	struct hv_guest_state hvregs;
4314 	struct p9_host_os_sprs host_os_sprs;
4315 	s64 dec;
4316 	int trap;
4317 
4318 	msr = mfmsr();
4319 
4320 	save_p9_host_os_sprs(&host_os_sprs);
4321 
4322 	/*
4323 	 * We need to save and restore the guest visible part of the
4324 	 * psscr (i.e. using SPRN_PSSCR_PR) since the hypervisor
4325 	 * doesn't do this for us. Note only required if pseries since
4326 	 * this is done in kvmhv_vcpu_entry_p9() below otherwise.
4327 	 */
4328 	host_psscr = mfspr(SPRN_PSSCR_PR);
4329 
4330 	kvmppc_msr_hard_disable_set_facilities(vcpu, msr);
4331 	if (lazy_irq_pending())
4332 		return 0;
4333 
4334 	if (unlikely(load_vcpu_state(vcpu, &host_os_sprs)))
4335 		msr = mfmsr(); /* TM restore can update msr */
4336 
4337 	if (vcpu->arch.psscr != host_psscr)
4338 		mtspr(SPRN_PSSCR_PR, vcpu->arch.psscr);
4339 
4340 	kvmhv_save_hv_regs(vcpu, &hvregs);
4341 	hvregs.lpcr = lpcr;
4342 	hvregs.amor = ~0;
4343 	vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
4344 	hvregs.version = HV_GUEST_STATE_VERSION;
4345 	if (vcpu->arch.nested) {
4346 		hvregs.lpid = vcpu->arch.nested->shadow_lpid;
4347 		hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
4348 	} else {
4349 		hvregs.lpid = vcpu->kvm->arch.lpid;
4350 		hvregs.vcpu_token = vcpu->vcpu_id;
4351 	}
4352 	hvregs.hdec_expiry = time_limit;
4353 
4354 	/*
4355 	 * hvregs has the doorbell status, so zero it here which
4356 	 * enables us to receive doorbells when H_ENTER_NESTED is
4357 	 * in progress for this vCPU
4358 	 */
4359 
4360 	if (vcpu->arch.doorbell_request)
4361 		vcpu->arch.doorbell_request = 0;
4362 
4363 	/*
4364 	 * When setting DEC, we must always deal with irq_work_raise
4365 	 * via NMI vs setting DEC. The problem occurs right as we
4366 	 * switch into guest mode if a NMI hits and sets pending work
4367 	 * and sets DEC, then that will apply to the guest and not
4368 	 * bring us back to the host.
4369 	 *
4370 	 * irq_work_raise could check a flag (or possibly LPCR[HDICE]
4371 	 * for example) and set HDEC to 1? That wouldn't solve the
4372 	 * nested hv case which needs to abort the hcall or zero the
4373 	 * time limit.
4374 	 *
4375 	 * XXX: Another day's problem.
4376 	 */
4377 	mtspr(SPRN_DEC, kvmppc_dec_expires_host_tb(vcpu) - *tb);
4378 
4379 	mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
4380 	mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
4381 	switch_pmu_to_guest(vcpu, &host_os_sprs);
4382 	accumulate_time(vcpu, &vcpu->arch.in_guest);
4383 	trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
4384 				  __pa(&vcpu->arch.regs));
4385 	accumulate_time(vcpu, &vcpu->arch.guest_exit);
4386 	kvmhv_restore_hv_return_state(vcpu, &hvregs);
4387 	switch_pmu_to_host(vcpu, &host_os_sprs);
4388 	vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
4389 	vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
4390 	vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
4391 	vcpu->arch.psscr = mfspr(SPRN_PSSCR_PR);
4392 
4393 	store_vcpu_state(vcpu);
4394 
4395 	dec = mfspr(SPRN_DEC);
4396 	if (!(lpcr & LPCR_LD)) /* Sign extend if not using large decrementer */
4397 		dec = (s32) dec;
4398 	*tb = mftb();
4399 	vcpu->arch.dec_expires = dec + (*tb + kvmppc_get_tb_offset(vcpu));
4400 
4401 	timer_rearm_host_dec(*tb);
4402 
4403 	restore_p9_host_os_sprs(vcpu, &host_os_sprs);
4404 	if (vcpu->arch.psscr != host_psscr)
4405 		mtspr(SPRN_PSSCR_PR, host_psscr);
4406 
4407 	return trap;
4408 }
4409 
4410 /*
4411  * Guest entry for POWER9 and later CPUs.
4412  */
4413 static int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
4414 			 unsigned long lpcr, u64 *tb)
4415 {
4416 	struct kvm *kvm = vcpu->kvm;
4417 	struct kvm_nested_guest *nested = vcpu->arch.nested;
4418 	u64 next_timer;
4419 	int trap;
4420 
4421 	next_timer = timer_get_next_tb();
4422 	if (*tb >= next_timer)
4423 		return BOOK3S_INTERRUPT_HV_DECREMENTER;
4424 	if (next_timer < time_limit)
4425 		time_limit = next_timer;
4426 	else if (*tb >= time_limit) /* nested time limit */
4427 		return BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER;
4428 
4429 	vcpu->arch.ceded = 0;
4430 
4431 	vcpu_vpa_increment_dispatch(vcpu);
4432 
4433 	if (kvmhv_on_pseries()) {
4434 		if (kvmhv_is_nestedv1())
4435 			trap = kvmhv_vcpu_entry_p9_nested(vcpu, time_limit, lpcr, tb);
4436 		else
4437 			trap = kvmhv_vcpu_entry_nestedv2(vcpu, time_limit, lpcr, tb);
4438 
4439 		/* H_CEDE has to be handled now, not later */
4440 		if (trap == BOOK3S_INTERRUPT_SYSCALL && !nested &&
4441 		    kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
4442 			kvmppc_cede(vcpu);
4443 			kvmppc_set_gpr(vcpu, 3, 0);
4444 			trap = 0;
4445 		}
4446 
4447 	} else if (nested) {
4448 		__this_cpu_write(cpu_in_guest, kvm);
4449 		trap = kvmhv_vcpu_entry_p9(vcpu, time_limit, lpcr, tb);
4450 		__this_cpu_write(cpu_in_guest, NULL);
4451 
4452 	} else {
4453 		kvmppc_xive_push_vcpu(vcpu);
4454 
4455 		__this_cpu_write(cpu_in_guest, kvm);
4456 		trap = kvmhv_vcpu_entry_p9(vcpu, time_limit, lpcr, tb);
4457 		__this_cpu_write(cpu_in_guest, NULL);
4458 
4459 		if (trap == BOOK3S_INTERRUPT_SYSCALL &&
4460 		    !(__kvmppc_get_msr_hv(vcpu) & MSR_PR)) {
4461 			unsigned long req = kvmppc_get_gpr(vcpu, 3);
4462 
4463 			/*
4464 			 * XIVE rearm and XICS hcalls must be handled
4465 			 * before xive context is pulled (is this
4466 			 * true?)
4467 			 */
4468 			if (req == H_CEDE) {
4469 				/* H_CEDE has to be handled now */
4470 				kvmppc_cede(vcpu);
4471 				if (!kvmppc_xive_rearm_escalation(vcpu)) {
4472 					/*
4473 					 * Pending escalation so abort
4474 					 * the cede.
4475 					 */
4476 					vcpu->arch.ceded = 0;
4477 				}
4478 				kvmppc_set_gpr(vcpu, 3, 0);
4479 				trap = 0;
4480 
4481 			} else if (req == H_ENTER_NESTED) {
4482 				/*
4483 				 * L2 should not run with the L1
4484 				 * context so rearm and pull it.
4485 				 */
4486 				if (!kvmppc_xive_rearm_escalation(vcpu)) {
4487 					/*
4488 					 * Pending escalation so abort
4489 					 * H_ENTER_NESTED.
4490 					 */
4491 					kvmppc_set_gpr(vcpu, 3, 0);
4492 					trap = 0;
4493 				}
4494 
4495 			} else if (hcall_is_xics(req)) {
4496 				int ret;
4497 
4498 				ret = kvmppc_xive_xics_hcall(vcpu, req);
4499 				if (ret != H_TOO_HARD) {
4500 					kvmppc_set_gpr(vcpu, 3, ret);
4501 					trap = 0;
4502 				}
4503 			}
4504 		}
4505 		kvmppc_xive_pull_vcpu(vcpu);
4506 
4507 		if (kvm_is_radix(kvm))
4508 			vcpu->arch.slb_max = 0;
4509 	}
4510 
4511 	vcpu_vpa_increment_dispatch(vcpu);
4512 
4513 	return trap;
4514 }
4515 
4516 /*
4517  * Wait for some other vcpu thread to execute us, and
4518  * wake us up when we need to handle something in the host.
4519  */
4520 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
4521 				 struct kvm_vcpu *vcpu, int wait_state)
4522 {
4523 	DEFINE_WAIT(wait);
4524 
4525 	prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
4526 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4527 		spin_unlock(&vc->lock);
4528 		schedule();
4529 		spin_lock(&vc->lock);
4530 	}
4531 	finish_wait(&vcpu->arch.cpu_run, &wait);
4532 }
4533 
4534 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
4535 {
4536 	if (!halt_poll_ns_grow)
4537 		return;
4538 
4539 	vc->halt_poll_ns *= halt_poll_ns_grow;
4540 	if (vc->halt_poll_ns < halt_poll_ns_grow_start)
4541 		vc->halt_poll_ns = halt_poll_ns_grow_start;
4542 }
4543 
4544 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
4545 {
4546 	if (halt_poll_ns_shrink == 0)
4547 		vc->halt_poll_ns = 0;
4548 	else
4549 		vc->halt_poll_ns /= halt_poll_ns_shrink;
4550 }
4551 
4552 #ifdef CONFIG_KVM_XICS
4553 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
4554 {
4555 	if (!xics_on_xive())
4556 		return false;
4557 	return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
4558 		vcpu->arch.xive_saved_state.cppr;
4559 }
4560 #else
4561 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
4562 {
4563 	return false;
4564 }
4565 #endif /* CONFIG_KVM_XICS */
4566 
4567 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
4568 {
4569 	if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
4570 	    kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
4571 		return true;
4572 
4573 	return false;
4574 }
4575 
4576 static bool kvmppc_vcpu_check_block(struct kvm_vcpu *vcpu)
4577 {
4578 	if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
4579 		return true;
4580 	return false;
4581 }
4582 
4583 /*
4584  * Check to see if any of the runnable vcpus on the vcore have pending
4585  * exceptions or are no longer ceded
4586  */
4587 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
4588 {
4589 	struct kvm_vcpu *vcpu;
4590 	int i;
4591 
4592 	for_each_runnable_thread(i, vcpu, vc) {
4593 		if (kvmppc_vcpu_check_block(vcpu))
4594 			return 1;
4595 	}
4596 
4597 	return 0;
4598 }
4599 
4600 /*
4601  * All the vcpus in this vcore are idle, so wait for a decrementer
4602  * or external interrupt to one of the vcpus.  vc->lock is held.
4603  */
4604 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
4605 {
4606 	ktime_t cur, start_poll, start_wait;
4607 	int do_sleep = 1;
4608 	u64 block_ns;
4609 
4610 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
4611 
4612 	/* Poll for pending exceptions and ceded state */
4613 	cur = start_poll = ktime_get();
4614 	if (vc->halt_poll_ns) {
4615 		ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
4616 		++vc->runner->stat.generic.halt_attempted_poll;
4617 
4618 		vc->vcore_state = VCORE_POLLING;
4619 		spin_unlock(&vc->lock);
4620 
4621 		do {
4622 			if (kvmppc_vcore_check_block(vc)) {
4623 				do_sleep = 0;
4624 				break;
4625 			}
4626 			cur = ktime_get();
4627 		} while (kvm_vcpu_can_poll(cur, stop));
4628 
4629 		spin_lock(&vc->lock);
4630 		vc->vcore_state = VCORE_INACTIVE;
4631 
4632 		if (!do_sleep) {
4633 			++vc->runner->stat.generic.halt_successful_poll;
4634 			goto out;
4635 		}
4636 	}
4637 
4638 	prepare_to_rcuwait(&vc->wait);
4639 	set_current_state(TASK_INTERRUPTIBLE);
4640 	if (kvmppc_vcore_check_block(vc)) {
4641 		finish_rcuwait(&vc->wait);
4642 		do_sleep = 0;
4643 		/* If we polled, count this as a successful poll */
4644 		if (vc->halt_poll_ns)
4645 			++vc->runner->stat.generic.halt_successful_poll;
4646 		goto out;
4647 	}
4648 
4649 	start_wait = ktime_get();
4650 
4651 	vc->vcore_state = VCORE_SLEEPING;
4652 	trace_kvmppc_vcore_blocked(vc->runner, 0);
4653 	spin_unlock(&vc->lock);
4654 	schedule();
4655 	finish_rcuwait(&vc->wait);
4656 	spin_lock(&vc->lock);
4657 	vc->vcore_state = VCORE_INACTIVE;
4658 	trace_kvmppc_vcore_blocked(vc->runner, 1);
4659 	++vc->runner->stat.halt_successful_wait;
4660 
4661 	cur = ktime_get();
4662 
4663 out:
4664 	block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
4665 
4666 	/* Attribute wait time */
4667 	if (do_sleep) {
4668 		vc->runner->stat.generic.halt_wait_ns +=
4669 			ktime_to_ns(cur) - ktime_to_ns(start_wait);
4670 		KVM_STATS_LOG_HIST_UPDATE(
4671 				vc->runner->stat.generic.halt_wait_hist,
4672 				ktime_to_ns(cur) - ktime_to_ns(start_wait));
4673 		/* Attribute failed poll time */
4674 		if (vc->halt_poll_ns) {
4675 			vc->runner->stat.generic.halt_poll_fail_ns +=
4676 				ktime_to_ns(start_wait) -
4677 				ktime_to_ns(start_poll);
4678 			KVM_STATS_LOG_HIST_UPDATE(
4679 				vc->runner->stat.generic.halt_poll_fail_hist,
4680 				ktime_to_ns(start_wait) -
4681 				ktime_to_ns(start_poll));
4682 		}
4683 	} else {
4684 		/* Attribute successful poll time */
4685 		if (vc->halt_poll_ns) {
4686 			vc->runner->stat.generic.halt_poll_success_ns +=
4687 				ktime_to_ns(cur) -
4688 				ktime_to_ns(start_poll);
4689 			KVM_STATS_LOG_HIST_UPDATE(
4690 				vc->runner->stat.generic.halt_poll_success_hist,
4691 				ktime_to_ns(cur) - ktime_to_ns(start_poll));
4692 		}
4693 	}
4694 
4695 	/* Adjust poll time */
4696 	if (halt_poll_ns) {
4697 		if (block_ns <= vc->halt_poll_ns)
4698 			;
4699 		/* We slept and blocked for longer than the max halt time */
4700 		else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
4701 			shrink_halt_poll_ns(vc);
4702 		/* We slept and our poll time is too small */
4703 		else if (vc->halt_poll_ns < halt_poll_ns &&
4704 				block_ns < halt_poll_ns)
4705 			grow_halt_poll_ns(vc);
4706 		if (vc->halt_poll_ns > halt_poll_ns)
4707 			vc->halt_poll_ns = halt_poll_ns;
4708 	} else
4709 		vc->halt_poll_ns = 0;
4710 
4711 	trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
4712 }
4713 
4714 /*
4715  * This never fails for a radix guest, as none of the operations it does
4716  * for a radix guest can fail or have a way to report failure.
4717  */
4718 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
4719 {
4720 	int r = 0;
4721 	struct kvm *kvm = vcpu->kvm;
4722 
4723 	mutex_lock(&kvm->arch.mmu_setup_lock);
4724 	if (!kvm->arch.mmu_ready) {
4725 		if (!kvm_is_radix(kvm))
4726 			r = kvmppc_hv_setup_htab_rma(vcpu);
4727 		if (!r) {
4728 			if (cpu_has_feature(CPU_FTR_ARCH_300))
4729 				kvmppc_setup_partition_table(kvm);
4730 			kvm->arch.mmu_ready = 1;
4731 		}
4732 	}
4733 	mutex_unlock(&kvm->arch.mmu_setup_lock);
4734 	return r;
4735 }
4736 
4737 static int kvmppc_run_vcpu(struct kvm_vcpu *vcpu)
4738 {
4739 	struct kvm_run *run = vcpu->run;
4740 	int n_ceded, i, r;
4741 	struct kvmppc_vcore *vc;
4742 	struct kvm_vcpu *v;
4743 
4744 	trace_kvmppc_run_vcpu_enter(vcpu);
4745 
4746 	run->exit_reason = 0;
4747 	vcpu->arch.ret = RESUME_GUEST;
4748 	vcpu->arch.trap = 0;
4749 	kvmppc_update_vpas(vcpu);
4750 
4751 	/*
4752 	 * Synchronize with other threads in this virtual core
4753 	 */
4754 	vc = vcpu->arch.vcore;
4755 	spin_lock(&vc->lock);
4756 	vcpu->arch.ceded = 0;
4757 	vcpu->arch.run_task = current;
4758 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4759 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4760 	vcpu->arch.busy_preempt = TB_NIL;
4761 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
4762 	++vc->n_runnable;
4763 
4764 	/*
4765 	 * This happens the first time this is called for a vcpu.
4766 	 * If the vcore is already running, we may be able to start
4767 	 * this thread straight away and have it join in.
4768 	 */
4769 	if (!signal_pending(current)) {
4770 		if ((vc->vcore_state == VCORE_PIGGYBACK ||
4771 		     vc->vcore_state == VCORE_RUNNING) &&
4772 			   !VCORE_IS_EXITING(vc)) {
4773 			kvmppc_update_vpa_dispatch(vcpu, vc);
4774 			kvmppc_start_thread(vcpu, vc);
4775 			trace_kvm_guest_enter(vcpu);
4776 		} else if (vc->vcore_state == VCORE_SLEEPING) {
4777 		        rcuwait_wake_up(&vc->wait);
4778 		}
4779 
4780 	}
4781 
4782 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4783 	       !signal_pending(current)) {
4784 		/* See if the MMU is ready to go */
4785 		if (!vcpu->kvm->arch.mmu_ready) {
4786 			spin_unlock(&vc->lock);
4787 			r = kvmhv_setup_mmu(vcpu);
4788 			spin_lock(&vc->lock);
4789 			if (r) {
4790 				run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4791 				run->fail_entry.
4792 					hardware_entry_failure_reason = 0;
4793 				vcpu->arch.ret = r;
4794 				break;
4795 			}
4796 		}
4797 
4798 		if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4799 			kvmppc_vcore_end_preempt(vc);
4800 
4801 		if (vc->vcore_state != VCORE_INACTIVE) {
4802 			kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
4803 			continue;
4804 		}
4805 		for_each_runnable_thread(i, v, vc) {
4806 			kvmppc_core_prepare_to_enter(v);
4807 			if (signal_pending(v->arch.run_task)) {
4808 				kvmppc_remove_runnable(vc, v, mftb());
4809 				v->stat.signal_exits++;
4810 				v->run->exit_reason = KVM_EXIT_INTR;
4811 				v->arch.ret = -EINTR;
4812 				wake_up(&v->arch.cpu_run);
4813 			}
4814 		}
4815 		if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
4816 			break;
4817 		n_ceded = 0;
4818 		for_each_runnable_thread(i, v, vc) {
4819 			if (!kvmppc_vcpu_woken(v))
4820 				n_ceded += v->arch.ceded;
4821 			else
4822 				v->arch.ceded = 0;
4823 		}
4824 		vc->runner = vcpu;
4825 		if (n_ceded == vc->n_runnable) {
4826 			kvmppc_vcore_blocked(vc);
4827 		} else if (need_resched()) {
4828 			kvmppc_vcore_preempt(vc);
4829 			/* Let something else run */
4830 			cond_resched_lock(&vc->lock);
4831 			if (vc->vcore_state == VCORE_PREEMPT)
4832 				kvmppc_vcore_end_preempt(vc);
4833 		} else {
4834 			kvmppc_run_core(vc);
4835 		}
4836 		vc->runner = NULL;
4837 	}
4838 
4839 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4840 	       (vc->vcore_state == VCORE_RUNNING ||
4841 		vc->vcore_state == VCORE_EXITING ||
4842 		vc->vcore_state == VCORE_PIGGYBACK))
4843 		kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
4844 
4845 	if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4846 		kvmppc_vcore_end_preempt(vc);
4847 
4848 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4849 		kvmppc_remove_runnable(vc, vcpu, mftb());
4850 		vcpu->stat.signal_exits++;
4851 		run->exit_reason = KVM_EXIT_INTR;
4852 		vcpu->arch.ret = -EINTR;
4853 	}
4854 
4855 	if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
4856 		/* Wake up some vcpu to run the core */
4857 		i = -1;
4858 		v = next_runnable_thread(vc, &i);
4859 		wake_up(&v->arch.cpu_run);
4860 	}
4861 
4862 	trace_kvmppc_run_vcpu_exit(vcpu);
4863 	spin_unlock(&vc->lock);
4864 	return vcpu->arch.ret;
4865 }
4866 
4867 int kvmhv_run_single_vcpu(struct kvm_vcpu *vcpu, u64 time_limit,
4868 			  unsigned long lpcr)
4869 {
4870 	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
4871 	struct kvm_run *run = vcpu->run;
4872 	int trap, r, pcpu;
4873 	int srcu_idx;
4874 	struct kvmppc_vcore *vc;
4875 	struct kvm *kvm = vcpu->kvm;
4876 	struct kvm_nested_guest *nested = vcpu->arch.nested;
4877 	unsigned long flags;
4878 	u64 tb;
4879 
4880 	trace_kvmppc_run_vcpu_enter(vcpu);
4881 
4882 	run->exit_reason = 0;
4883 	vcpu->arch.ret = RESUME_GUEST;
4884 	vcpu->arch.trap = 0;
4885 
4886 	vc = vcpu->arch.vcore;
4887 	vcpu->arch.ceded = 0;
4888 	vcpu->arch.run_task = current;
4889 	vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
4890 
4891 	/* See if the MMU is ready to go */
4892 	if (unlikely(!kvm->arch.mmu_ready)) {
4893 		r = kvmhv_setup_mmu(vcpu);
4894 		if (r) {
4895 			run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4896 			run->fail_entry.hardware_entry_failure_reason = 0;
4897 			vcpu->arch.ret = r;
4898 			return r;
4899 		}
4900 	}
4901 
4902 	if (need_resched())
4903 		cond_resched();
4904 
4905 	kvmppc_update_vpas(vcpu);
4906 
4907 	preempt_disable();
4908 	pcpu = smp_processor_id();
4909 	if (kvm_is_radix(kvm))
4910 		kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4911 
4912 	/* flags save not required, but irq_pmu has no disable/enable API */
4913 	powerpc_local_irq_pmu_save(flags);
4914 
4915 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4916 
4917 	if (signal_pending(current))
4918 		goto sigpend;
4919 	if (need_resched() || !kvm->arch.mmu_ready)
4920 		goto out;
4921 
4922 	vcpu->cpu = pcpu;
4923 	vcpu->arch.thread_cpu = pcpu;
4924 	vc->pcpu = pcpu;
4925 	local_paca->kvm_hstate.kvm_vcpu = vcpu;
4926 	local_paca->kvm_hstate.ptid = 0;
4927 	local_paca->kvm_hstate.fake_suspend = 0;
4928 
4929 	/*
4930 	 * Orders set cpu/thread_cpu vs testing for pending interrupts and
4931 	 * doorbells below. The other side is when these fields are set vs
4932 	 * kvmppc_fast_vcpu_kick_hv reading the cpu/thread_cpu fields to
4933 	 * kick a vCPU to notice the pending interrupt.
4934 	 */
4935 	smp_mb();
4936 
4937 	if (!nested) {
4938 		kvmppc_core_prepare_to_enter(vcpu);
4939 		if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4940 			     &vcpu->arch.pending_exceptions) ||
4941 		    xive_interrupt_pending(vcpu)) {
4942 			/*
4943 			 * For nested HV, don't synthesize but always pass MER,
4944 			 * the L0 will be able to optimise that more
4945 			 * effectively than manipulating registers directly.
4946 			 */
4947 			if (!kvmhv_on_pseries() && (__kvmppc_get_msr_hv(vcpu) & MSR_EE))
4948 				kvmppc_inject_interrupt_hv(vcpu,
4949 							   BOOK3S_INTERRUPT_EXTERNAL, 0);
4950 			else
4951 				lpcr |= LPCR_MER;
4952 		} else {
4953 			/*
4954 			 * L1's copy of L2's LPCR (vcpu->arch.vcore->lpcr) can get its MER bit
4955 			 * unexpectedly set - for e.g. during NMI handling when all register
4956 			 * states are synchronized from L0 to L1. L1 needs to inform L0 about
4957 			 * MER=1 only when there are pending external interrupts.
4958 			 * In the above if check, MER bit is set if there are pending
4959 			 * external interrupts. Hence, explicitly mask off MER bit
4960 			 * here as otherwise it may generate spurious interrupts in L2 KVM
4961 			 * causing an endless loop, which results in L2 guest getting hung.
4962 			 */
4963 			lpcr &= ~LPCR_MER;
4964 		}
4965 	} else if (vcpu->arch.pending_exceptions ||
4966 		   xive_interrupt_pending(vcpu)) {
4967 		vcpu->arch.ret = RESUME_HOST;
4968 		goto out;
4969 	}
4970 
4971 	if (vcpu->arch.timer_running) {
4972 		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
4973 		vcpu->arch.timer_running = 0;
4974 	}
4975 
4976 	tb = mftb();
4977 
4978 	kvmppc_update_vpa_dispatch_p9(vcpu, vc, tb + kvmppc_get_tb_offset(vcpu));
4979 
4980 	trace_kvm_guest_enter(vcpu);
4981 
4982 	guest_timing_enter_irqoff();
4983 
4984 	srcu_idx = srcu_read_lock(&kvm->srcu);
4985 
4986 	guest_state_enter_irqoff();
4987 	this_cpu_disable_ftrace();
4988 
4989 	trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr, &tb);
4990 	vcpu->arch.trap = trap;
4991 
4992 	this_cpu_enable_ftrace();
4993 	guest_state_exit_irqoff();
4994 
4995 	srcu_read_unlock(&kvm->srcu, srcu_idx);
4996 
4997 	set_irq_happened(trap);
4998 
4999 	vcpu->cpu = -1;
5000 	vcpu->arch.thread_cpu = -1;
5001 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
5002 
5003 	if (!vtime_accounting_enabled_this_cpu()) {
5004 		powerpc_local_irq_pmu_restore(flags);
5005 		/*
5006 		 * Service IRQs here before guest_timing_exit_irqoff() so any
5007 		 * ticks that occurred while running the guest are accounted to
5008 		 * the guest. If vtime accounting is enabled, accounting uses
5009 		 * TB rather than ticks, so it can be done without enabling
5010 		 * interrupts here, which has the problem that it accounts
5011 		 * interrupt processing overhead to the host.
5012 		 */
5013 		powerpc_local_irq_pmu_save(flags);
5014 	}
5015 	guest_timing_exit_irqoff();
5016 
5017 	powerpc_local_irq_pmu_restore(flags);
5018 
5019 	preempt_enable();
5020 
5021 	/*
5022 	 * cancel pending decrementer exception if DEC is now positive, or if
5023 	 * entering a nested guest in which case the decrementer is now owned
5024 	 * by L2 and the L1 decrementer is provided in hdec_expires
5025 	 */
5026 	if (kvmppc_core_pending_dec(vcpu) &&
5027 			((tb < kvmppc_dec_expires_host_tb(vcpu)) ||
5028 			 (trap == BOOK3S_INTERRUPT_SYSCALL &&
5029 			  kvmppc_get_gpr(vcpu, 3) == H_ENTER_NESTED)))
5030 		kvmppc_core_dequeue_dec(vcpu);
5031 
5032 	trace_kvm_guest_exit(vcpu);
5033 	r = RESUME_GUEST;
5034 	if (trap) {
5035 		if (!nested)
5036 			r = kvmppc_handle_exit_hv(vcpu, current);
5037 		else
5038 			r = kvmppc_handle_nested_exit(vcpu);
5039 	}
5040 	vcpu->arch.ret = r;
5041 
5042 	if (is_kvmppc_resume_guest(r) && !kvmppc_vcpu_check_block(vcpu)) {
5043 		kvmppc_set_timer(vcpu);
5044 
5045 		prepare_to_rcuwait(wait);
5046 		for (;;) {
5047 			set_current_state(TASK_INTERRUPTIBLE);
5048 			if (signal_pending(current)) {
5049 				vcpu->stat.signal_exits++;
5050 				run->exit_reason = KVM_EXIT_INTR;
5051 				vcpu->arch.ret = -EINTR;
5052 				break;
5053 			}
5054 
5055 			if (kvmppc_vcpu_check_block(vcpu))
5056 				break;
5057 
5058 			trace_kvmppc_vcore_blocked(vcpu, 0);
5059 			schedule();
5060 			trace_kvmppc_vcore_blocked(vcpu, 1);
5061 		}
5062 		finish_rcuwait(wait);
5063 	}
5064 	vcpu->arch.ceded = 0;
5065 
5066  done:
5067 	trace_kvmppc_run_vcpu_exit(vcpu);
5068 
5069 	return vcpu->arch.ret;
5070 
5071  sigpend:
5072 	vcpu->stat.signal_exits++;
5073 	run->exit_reason = KVM_EXIT_INTR;
5074 	vcpu->arch.ret = -EINTR;
5075  out:
5076 	vcpu->cpu = -1;
5077 	vcpu->arch.thread_cpu = -1;
5078 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
5079 	powerpc_local_irq_pmu_restore(flags);
5080 	preempt_enable();
5081 	goto done;
5082 }
5083 
5084 static int kvmppc_vcpu_run_hv(struct kvm_vcpu *vcpu)
5085 {
5086 	struct kvm_run *run = vcpu->run;
5087 	int r;
5088 	int srcu_idx;
5089 	struct kvm *kvm;
5090 	unsigned long msr;
5091 
5092 	start_timing(vcpu, &vcpu->arch.vcpu_entry);
5093 
5094 	if (!vcpu->arch.sane) {
5095 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5096 		return -EINVAL;
5097 	}
5098 
5099 	/* No need to go into the guest when all we'll do is come back out */
5100 	if (signal_pending(current)) {
5101 		run->exit_reason = KVM_EXIT_INTR;
5102 		return -EINTR;
5103 	}
5104 
5105 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
5106 	/*
5107 	 * Don't allow entry with a suspended transaction, because
5108 	 * the guest entry/exit code will lose it.
5109 	 */
5110 	if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
5111 	    (current->thread.regs->msr & MSR_TM)) {
5112 		if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
5113 			run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5114 			run->fail_entry.hardware_entry_failure_reason = 0;
5115 			return -EINVAL;
5116 		}
5117 	}
5118 #endif
5119 
5120 	/*
5121 	 * Force online to 1 for the sake of old userspace which doesn't
5122 	 * set it.
5123 	 */
5124 	if (!vcpu->arch.online) {
5125 		atomic_inc(&vcpu->arch.vcore->online_count);
5126 		vcpu->arch.online = 1;
5127 	}
5128 
5129 	kvmppc_core_prepare_to_enter(vcpu);
5130 
5131 	kvm = vcpu->kvm;
5132 	atomic_inc(&kvm->arch.vcpus_running);
5133 	/* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
5134 	smp_mb();
5135 
5136 	msr = 0;
5137 	if (IS_ENABLED(CONFIG_PPC_FPU))
5138 		msr |= MSR_FP;
5139 	if (cpu_has_feature(CPU_FTR_ALTIVEC))
5140 		msr |= MSR_VEC;
5141 	if (cpu_has_feature(CPU_FTR_VSX))
5142 		msr |= MSR_VSX;
5143 	if ((cpu_has_feature(CPU_FTR_TM) ||
5144 	    cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST)) &&
5145 			(kvmppc_get_hfscr_hv(vcpu) & HFSCR_TM))
5146 		msr |= MSR_TM;
5147 	msr = msr_check_and_set(msr);
5148 
5149 	kvmppc_save_user_regs();
5150 
5151 	kvmppc_save_current_sprs();
5152 
5153 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5154 		vcpu->arch.waitp = &vcpu->arch.vcore->wait;
5155 	vcpu->arch.pgdir = kvm->mm->pgd;
5156 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
5157 
5158 	do {
5159 		accumulate_time(vcpu, &vcpu->arch.guest_entry);
5160 		if (cpu_has_feature(CPU_FTR_ARCH_300))
5161 			r = kvmhv_run_single_vcpu(vcpu, ~(u64)0,
5162 						  vcpu->arch.vcore->lpcr);
5163 		else
5164 			r = kvmppc_run_vcpu(vcpu);
5165 
5166 		if (run->exit_reason == KVM_EXIT_PAPR_HCALL) {
5167 			accumulate_time(vcpu, &vcpu->arch.hcall);
5168 
5169 			if (!kvmhv_is_nestedv2() && WARN_ON_ONCE(__kvmppc_get_msr_hv(vcpu) & MSR_PR)) {
5170 				/*
5171 				 * These should have been caught reflected
5172 				 * into the guest by now. Final sanity check:
5173 				 * don't allow userspace to execute hcalls in
5174 				 * the hypervisor.
5175 				 */
5176 				r = RESUME_GUEST;
5177 				continue;
5178 			}
5179 			trace_kvm_hcall_enter(vcpu);
5180 			r = kvmppc_pseries_do_hcall(vcpu);
5181 			trace_kvm_hcall_exit(vcpu, r);
5182 			kvmppc_core_prepare_to_enter(vcpu);
5183 		} else if (r == RESUME_PAGE_FAULT) {
5184 			accumulate_time(vcpu, &vcpu->arch.pg_fault);
5185 			srcu_idx = srcu_read_lock(&kvm->srcu);
5186 			r = kvmppc_book3s_hv_page_fault(vcpu,
5187 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
5188 			srcu_read_unlock(&kvm->srcu, srcu_idx);
5189 		} else if (r == RESUME_PASSTHROUGH) {
5190 			if (WARN_ON(xics_on_xive()))
5191 				r = H_SUCCESS;
5192 			else
5193 				r = kvmppc_xics_rm_complete(vcpu, 0);
5194 		}
5195 	} while (is_kvmppc_resume_guest(r));
5196 	accumulate_time(vcpu, &vcpu->arch.vcpu_exit);
5197 
5198 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
5199 	atomic_dec(&kvm->arch.vcpus_running);
5200 
5201 	srr_regs_clobbered();
5202 
5203 	end_timing(vcpu);
5204 
5205 	return r;
5206 }
5207 
5208 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
5209 				     int shift, int sllp)
5210 {
5211 	(*sps)->page_shift = shift;
5212 	(*sps)->slb_enc = sllp;
5213 	(*sps)->enc[0].page_shift = shift;
5214 	(*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
5215 	/*
5216 	 * Add 16MB MPSS support (may get filtered out by userspace)
5217 	 */
5218 	if (shift != 24) {
5219 		int penc = kvmppc_pgsize_lp_encoding(shift, 24);
5220 		if (penc != -1) {
5221 			(*sps)->enc[1].page_shift = 24;
5222 			(*sps)->enc[1].pte_enc = penc;
5223 		}
5224 	}
5225 	(*sps)++;
5226 }
5227 
5228 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
5229 					 struct kvm_ppc_smmu_info *info)
5230 {
5231 	struct kvm_ppc_one_seg_page_size *sps;
5232 
5233 	/*
5234 	 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
5235 	 * POWER7 doesn't support keys for instruction accesses,
5236 	 * POWER8 and POWER9 do.
5237 	 */
5238 	info->data_keys = 32;
5239 	info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
5240 
5241 	/* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
5242 	info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
5243 	info->slb_size = 32;
5244 
5245 	/* We only support these sizes for now, and no muti-size segments */
5246 	sps = &info->sps[0];
5247 	kvmppc_add_seg_page_size(&sps, 12, 0);
5248 	kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
5249 	kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
5250 
5251 	/* If running as a nested hypervisor, we don't support HPT guests */
5252 	if (kvmhv_on_pseries())
5253 		info->flags |= KVM_PPC_NO_HASH;
5254 
5255 	return 0;
5256 }
5257 
5258 /*
5259  * Get (and clear) the dirty memory log for a memory slot.
5260  */
5261 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
5262 					 struct kvm_dirty_log *log)
5263 {
5264 	struct kvm_memslots *slots;
5265 	struct kvm_memory_slot *memslot;
5266 	int r;
5267 	unsigned long n, i;
5268 	unsigned long *buf, *p;
5269 	struct kvm_vcpu *vcpu;
5270 
5271 	mutex_lock(&kvm->slots_lock);
5272 
5273 	r = -EINVAL;
5274 	if (log->slot >= KVM_USER_MEM_SLOTS)
5275 		goto out;
5276 
5277 	slots = kvm_memslots(kvm);
5278 	memslot = id_to_memslot(slots, log->slot);
5279 	r = -ENOENT;
5280 	if (!memslot || !memslot->dirty_bitmap)
5281 		goto out;
5282 
5283 	/*
5284 	 * Use second half of bitmap area because both HPT and radix
5285 	 * accumulate bits in the first half.
5286 	 */
5287 	n = kvm_dirty_bitmap_bytes(memslot);
5288 	buf = memslot->dirty_bitmap + n / sizeof(long);
5289 	memset(buf, 0, n);
5290 
5291 	if (kvm_is_radix(kvm))
5292 		r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
5293 	else
5294 		r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
5295 	if (r)
5296 		goto out;
5297 
5298 	/*
5299 	 * We accumulate dirty bits in the first half of the
5300 	 * memslot's dirty_bitmap area, for when pages are paged
5301 	 * out or modified by the host directly.  Pick up these
5302 	 * bits and add them to the map.
5303 	 */
5304 	p = memslot->dirty_bitmap;
5305 	for (i = 0; i < n / sizeof(long); ++i)
5306 		buf[i] |= xchg(&p[i], 0);
5307 
5308 	/* Harvest dirty bits from VPA and DTL updates */
5309 	/* Note: we never modify the SLB shadow buffer areas */
5310 	kvm_for_each_vcpu(i, vcpu, kvm) {
5311 		spin_lock(&vcpu->arch.vpa_update_lock);
5312 		kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
5313 		kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
5314 		spin_unlock(&vcpu->arch.vpa_update_lock);
5315 	}
5316 
5317 	r = -EFAULT;
5318 	if (copy_to_user(log->dirty_bitmap, buf, n))
5319 		goto out;
5320 
5321 	r = 0;
5322 out:
5323 	mutex_unlock(&kvm->slots_lock);
5324 	return r;
5325 }
5326 
5327 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *slot)
5328 {
5329 	vfree(slot->arch.rmap);
5330 	slot->arch.rmap = NULL;
5331 }
5332 
5333 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
5334 				const struct kvm_memory_slot *old,
5335 				struct kvm_memory_slot *new,
5336 				enum kvm_mr_change change)
5337 {
5338 	if (change == KVM_MR_CREATE) {
5339 		unsigned long size = array_size(new->npages, sizeof(*new->arch.rmap));
5340 
5341 		if ((size >> PAGE_SHIFT) > totalram_pages())
5342 			return -ENOMEM;
5343 
5344 		new->arch.rmap = vzalloc(size);
5345 		if (!new->arch.rmap)
5346 			return -ENOMEM;
5347 	} else if (change != KVM_MR_DELETE) {
5348 		new->arch.rmap = old->arch.rmap;
5349 	}
5350 
5351 	return 0;
5352 }
5353 
5354 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
5355 				struct kvm_memory_slot *old,
5356 				const struct kvm_memory_slot *new,
5357 				enum kvm_mr_change change)
5358 {
5359 	/*
5360 	 * If we are creating or modifying a memslot, it might make
5361 	 * some address that was previously cached as emulated
5362 	 * MMIO be no longer emulated MMIO, so invalidate
5363 	 * all the caches of emulated MMIO translations.
5364 	 */
5365 	if (change != KVM_MR_DELETE)
5366 		atomic64_inc(&kvm->arch.mmio_update);
5367 
5368 	/*
5369 	 * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
5370 	 * have already called kvm_arch_flush_shadow_memslot() to
5371 	 * flush shadow mappings.  For KVM_MR_CREATE we have no
5372 	 * previous mappings.  So the only case to handle is
5373 	 * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
5374 	 * has been changed.
5375 	 * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
5376 	 * to get rid of any THP PTEs in the partition-scoped page tables
5377 	 * so we can track dirtiness at the page level; we flush when
5378 	 * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
5379 	 * using THP PTEs.
5380 	 */
5381 	if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
5382 	    ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
5383 		kvmppc_radix_flush_memslot(kvm, old);
5384 	/*
5385 	 * If UV hasn't yet called H_SVM_INIT_START, don't register memslots.
5386 	 */
5387 	if (!kvm->arch.secure_guest)
5388 		return;
5389 
5390 	switch (change) {
5391 	case KVM_MR_CREATE:
5392 		/*
5393 		 * @TODO kvmppc_uvmem_memslot_create() can fail and
5394 		 * return error. Fix this.
5395 		 */
5396 		kvmppc_uvmem_memslot_create(kvm, new);
5397 		break;
5398 	case KVM_MR_DELETE:
5399 		kvmppc_uvmem_memslot_delete(kvm, old);
5400 		break;
5401 	default:
5402 		/* TODO: Handle KVM_MR_MOVE */
5403 		break;
5404 	}
5405 }
5406 
5407 /*
5408  * Update LPCR values in kvm->arch and in vcores.
5409  * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion
5410  * of kvm->arch.lpcr update).
5411  */
5412 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
5413 {
5414 	long int i;
5415 	u32 cores_done = 0;
5416 
5417 	if ((kvm->arch.lpcr & mask) == lpcr)
5418 		return;
5419 
5420 	kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
5421 
5422 	for (i = 0; i < KVM_MAX_VCORES; ++i) {
5423 		struct kvmppc_vcore *vc = kvm->arch.vcores[i];
5424 		if (!vc)
5425 			continue;
5426 
5427 		spin_lock(&vc->lock);
5428 		vc->lpcr = (vc->lpcr & ~mask) | lpcr;
5429 		verify_lpcr(kvm, vc->lpcr);
5430 		spin_unlock(&vc->lock);
5431 		if (++cores_done >= kvm->arch.online_vcores)
5432 			break;
5433 	}
5434 
5435 	if (kvmhv_is_nestedv2()) {
5436 		struct kvm_vcpu *vcpu;
5437 
5438 		kvm_for_each_vcpu(i, vcpu, kvm) {
5439 			kvmhv_nestedv2_mark_dirty(vcpu, KVMPPC_GSID_LPCR);
5440 		}
5441 	}
5442 }
5443 
5444 void kvmppc_setup_partition_table(struct kvm *kvm)
5445 {
5446 	unsigned long dw0, dw1;
5447 
5448 	if (!kvm_is_radix(kvm)) {
5449 		/* PS field - page size for VRMA */
5450 		dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
5451 			((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
5452 		/* HTABSIZE and HTABORG fields */
5453 		dw0 |= kvm->arch.sdr1;
5454 
5455 		/* Second dword as set by userspace */
5456 		dw1 = kvm->arch.process_table;
5457 	} else {
5458 		dw0 = PATB_HR | radix__get_tree_size() |
5459 			__pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
5460 		dw1 = PATB_GR | kvm->arch.process_table;
5461 	}
5462 	kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
5463 }
5464 
5465 /*
5466  * Set up HPT (hashed page table) and RMA (real-mode area).
5467  * Must be called with kvm->arch.mmu_setup_lock held.
5468  */
5469 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
5470 {
5471 	int err = 0;
5472 	struct kvm *kvm = vcpu->kvm;
5473 	unsigned long hva;
5474 	struct kvm_memory_slot *memslot;
5475 	struct vm_area_struct *vma;
5476 	unsigned long lpcr = 0, senc;
5477 	unsigned long psize, porder;
5478 	int srcu_idx;
5479 
5480 	/* Allocate hashed page table (if not done already) and reset it */
5481 	if (!kvm->arch.hpt.virt) {
5482 		int order = KVM_DEFAULT_HPT_ORDER;
5483 		struct kvm_hpt_info info;
5484 
5485 		err = kvmppc_allocate_hpt(&info, order);
5486 		/* If we get here, it means userspace didn't specify a
5487 		 * size explicitly.  So, try successively smaller
5488 		 * sizes if the default failed. */
5489 		while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
5490 			err  = kvmppc_allocate_hpt(&info, order);
5491 
5492 		if (err < 0) {
5493 			pr_err("KVM: Couldn't alloc HPT\n");
5494 			goto out;
5495 		}
5496 
5497 		kvmppc_set_hpt(kvm, &info);
5498 	}
5499 
5500 	/* Look up the memslot for guest physical address 0 */
5501 	srcu_idx = srcu_read_lock(&kvm->srcu);
5502 	memslot = gfn_to_memslot(kvm, 0);
5503 
5504 	/* We must have some memory at 0 by now */
5505 	err = -EINVAL;
5506 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
5507 		goto out_srcu;
5508 
5509 	/* Look up the VMA for the start of this memory slot */
5510 	hva = memslot->userspace_addr;
5511 	mmap_read_lock(kvm->mm);
5512 	vma = vma_lookup(kvm->mm, hva);
5513 	if (!vma || (vma->vm_flags & VM_IO))
5514 		goto up_out;
5515 
5516 	psize = vma_kernel_pagesize(vma);
5517 
5518 	mmap_read_unlock(kvm->mm);
5519 
5520 	/* We can handle 4k, 64k or 16M pages in the VRMA */
5521 	if (psize >= 0x1000000)
5522 		psize = 0x1000000;
5523 	else if (psize >= 0x10000)
5524 		psize = 0x10000;
5525 	else
5526 		psize = 0x1000;
5527 	porder = __ilog2(psize);
5528 
5529 	senc = slb_pgsize_encoding(psize);
5530 	kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
5531 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
5532 	/* Create HPTEs in the hash page table for the VRMA */
5533 	kvmppc_map_vrma(vcpu, memslot, porder);
5534 
5535 	/* Update VRMASD field in the LPCR */
5536 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
5537 		/* the -4 is to account for senc values starting at 0x10 */
5538 		lpcr = senc << (LPCR_VRMASD_SH - 4);
5539 		kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
5540 	}
5541 
5542 	/* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
5543 	smp_wmb();
5544 	err = 0;
5545  out_srcu:
5546 	srcu_read_unlock(&kvm->srcu, srcu_idx);
5547  out:
5548 	return err;
5549 
5550  up_out:
5551 	mmap_read_unlock(kvm->mm);
5552 	goto out_srcu;
5553 }
5554 
5555 /*
5556  * Must be called with kvm->arch.mmu_setup_lock held and
5557  * mmu_ready = 0 and no vcpus running.
5558  */
5559 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
5560 {
5561 	unsigned long lpcr, lpcr_mask;
5562 
5563 	if (nesting_enabled(kvm))
5564 		kvmhv_release_all_nested(kvm);
5565 	kvmppc_rmap_reset(kvm);
5566 	kvm->arch.process_table = 0;
5567 	/* Mutual exclusion with kvm_unmap_gfn_range etc. */
5568 	spin_lock(&kvm->mmu_lock);
5569 	kvm->arch.radix = 0;
5570 	spin_unlock(&kvm->mmu_lock);
5571 	kvmppc_free_radix(kvm);
5572 
5573 	lpcr = LPCR_VPM1;
5574 	lpcr_mask = LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5575 	if (cpu_has_feature(CPU_FTR_ARCH_31))
5576 		lpcr_mask |= LPCR_HAIL;
5577 	kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
5578 
5579 	return 0;
5580 }
5581 
5582 /*
5583  * Must be called with kvm->arch.mmu_setup_lock held and
5584  * mmu_ready = 0 and no vcpus running.
5585  */
5586 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
5587 {
5588 	unsigned long lpcr, lpcr_mask;
5589 	int err;
5590 
5591 	err = kvmppc_init_vm_radix(kvm);
5592 	if (err)
5593 		return err;
5594 	kvmppc_rmap_reset(kvm);
5595 	/* Mutual exclusion with kvm_unmap_gfn_range etc. */
5596 	spin_lock(&kvm->mmu_lock);
5597 	kvm->arch.radix = 1;
5598 	spin_unlock(&kvm->mmu_lock);
5599 	kvmppc_free_hpt(&kvm->arch.hpt);
5600 
5601 	lpcr = LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5602 	lpcr_mask = LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5603 	if (cpu_has_feature(CPU_FTR_ARCH_31)) {
5604 		lpcr_mask |= LPCR_HAIL;
5605 		if (cpu_has_feature(CPU_FTR_HVMODE) &&
5606 				(kvm->arch.host_lpcr & LPCR_HAIL))
5607 			lpcr |= LPCR_HAIL;
5608 	}
5609 	kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
5610 
5611 	return 0;
5612 }
5613 
5614 #ifdef CONFIG_KVM_XICS
5615 /*
5616  * Allocate a per-core structure for managing state about which cores are
5617  * running in the host versus the guest and for exchanging data between
5618  * real mode KVM and CPU running in the host.
5619  * This is only done for the first VM.
5620  * The allocated structure stays even if all VMs have stopped.
5621  * It is only freed when the kvm-hv module is unloaded.
5622  * It's OK for this routine to fail, we just don't support host
5623  * core operations like redirecting H_IPI wakeups.
5624  */
5625 void kvmppc_alloc_host_rm_ops(void)
5626 {
5627 	struct kvmppc_host_rm_ops *ops;
5628 	unsigned long l_ops;
5629 	int cpu, core;
5630 	int size;
5631 
5632 	if (cpu_has_feature(CPU_FTR_ARCH_300))
5633 		return;
5634 
5635 	/* Not the first time here ? */
5636 	if (kvmppc_host_rm_ops_hv != NULL)
5637 		return;
5638 
5639 	ops = kzalloc_obj(struct kvmppc_host_rm_ops);
5640 	if (!ops)
5641 		return;
5642 
5643 	size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
5644 	ops->rm_core = kzalloc(size, GFP_KERNEL);
5645 
5646 	if (!ops->rm_core) {
5647 		kfree(ops);
5648 		return;
5649 	}
5650 
5651 	cpus_read_lock();
5652 
5653 	for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
5654 		if (!cpu_online(cpu))
5655 			continue;
5656 
5657 		core = cpu >> threads_shift;
5658 		ops->rm_core[core].rm_state.in_host = 1;
5659 	}
5660 
5661 	ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
5662 
5663 	/*
5664 	 * Make the contents of the kvmppc_host_rm_ops structure visible
5665 	 * to other CPUs before we assign it to the global variable.
5666 	 * Do an atomic assignment (no locks used here), but if someone
5667 	 * beats us to it, just free our copy and return.
5668 	 */
5669 	smp_wmb();
5670 	l_ops = (unsigned long) ops;
5671 
5672 	if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
5673 		cpus_read_unlock();
5674 		kfree(ops->rm_core);
5675 		kfree(ops);
5676 		return;
5677 	}
5678 
5679 	cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
5680 					     "ppc/kvm_book3s:prepare",
5681 					     kvmppc_set_host_core,
5682 					     kvmppc_clear_host_core);
5683 	cpus_read_unlock();
5684 }
5685 
5686 void kvmppc_free_host_rm_ops(void)
5687 {
5688 	if (kvmppc_host_rm_ops_hv) {
5689 		cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
5690 		kfree(kvmppc_host_rm_ops_hv->rm_core);
5691 		kfree(kvmppc_host_rm_ops_hv);
5692 		kvmppc_host_rm_ops_hv = NULL;
5693 	}
5694 }
5695 #endif
5696 
5697 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
5698 {
5699 	unsigned long lpcr, lpid;
5700 	int ret;
5701 
5702 	mutex_init(&kvm->arch.uvmem_lock);
5703 	INIT_LIST_HEAD(&kvm->arch.uvmem_pfns);
5704 	mutex_init(&kvm->arch.mmu_setup_lock);
5705 
5706 	/* Allocate the guest's logical partition ID */
5707 
5708 	if (!kvmhv_is_nestedv2()) {
5709 		lpid = kvmppc_alloc_lpid();
5710 		if ((long)lpid < 0)
5711 			return -ENOMEM;
5712 		kvm->arch.lpid = lpid;
5713 	}
5714 
5715 	kvmppc_alloc_host_rm_ops();
5716 
5717 	kvmhv_vm_nested_init(kvm);
5718 
5719 	if (kvmhv_is_nestedv2()) {
5720 		long rc;
5721 		unsigned long guest_id;
5722 
5723 		rc = plpar_guest_create(0, &guest_id);
5724 
5725 		if (rc != H_SUCCESS)
5726 			pr_err("KVM: Create Guest hcall failed, rc=%ld\n", rc);
5727 
5728 		switch (rc) {
5729 		case H_PARAMETER:
5730 		case H_FUNCTION:
5731 		case H_STATE:
5732 			return -EINVAL;
5733 		case H_NOT_ENOUGH_RESOURCES:
5734 		case H_ABORTED:
5735 			return -ENOMEM;
5736 		case H_AUTHORITY:
5737 			return -EPERM;
5738 		case H_NOT_AVAILABLE:
5739 			return -EBUSY;
5740 		}
5741 		kvm->arch.lpid = guest_id;
5742 	}
5743 
5744 
5745 	/*
5746 	 * Since we don't flush the TLB when tearing down a VM,
5747 	 * and this lpid might have previously been used,
5748 	 * make sure we flush on each core before running the new VM.
5749 	 * On POWER9, the tlbie in mmu_partition_table_set_entry()
5750 	 * does this flush for us.
5751 	 */
5752 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5753 		cpumask_setall(&kvm->arch.need_tlb_flush);
5754 
5755 	/* Start out with the default set of hcalls enabled */
5756 	memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
5757 	       sizeof(kvm->arch.enabled_hcalls));
5758 
5759 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5760 		kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
5761 
5762 	/* Init LPCR for virtual RMA mode */
5763 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
5764 		kvm->arch.host_lpid = mfspr(SPRN_LPID);
5765 		kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
5766 		lpcr &= LPCR_PECE | LPCR_LPES;
5767 	} else {
5768 		/*
5769 		 * The L2 LPES mode will be set by the L0 according to whether
5770 		 * or not it needs to take external interrupts in HV mode.
5771 		 */
5772 		lpcr = 0;
5773 	}
5774 	lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
5775 		LPCR_VPM0 | LPCR_VPM1;
5776 	kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
5777 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
5778 	/* On POWER8 turn on online bit to enable PURR/SPURR */
5779 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
5780 		lpcr |= LPCR_ONL;
5781 	/*
5782 	 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
5783 	 * Set HVICE bit to enable hypervisor virtualization interrupts.
5784 	 * Set HEIC to prevent OS interrupts to go to hypervisor (should
5785 	 * be unnecessary but better safe than sorry in case we re-enable
5786 	 * EE in HV mode with this LPCR still set)
5787 	 */
5788 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5789 		lpcr &= ~LPCR_VPM0;
5790 		lpcr |= LPCR_HVICE | LPCR_HEIC;
5791 
5792 		/*
5793 		 * If xive is enabled, we route 0x500 interrupts directly
5794 		 * to the guest.
5795 		 */
5796 		if (xics_on_xive())
5797 			lpcr |= LPCR_LPES;
5798 	}
5799 
5800 	/*
5801 	 * If the host uses radix, the guest starts out as radix.
5802 	 */
5803 	if (radix_enabled()) {
5804 		kvm->arch.radix = 1;
5805 		kvm->arch.mmu_ready = 1;
5806 		lpcr &= ~LPCR_VPM1;
5807 		lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5808 		if (cpu_has_feature(CPU_FTR_HVMODE) &&
5809 		    cpu_has_feature(CPU_FTR_ARCH_31) &&
5810 		    (kvm->arch.host_lpcr & LPCR_HAIL))
5811 			lpcr |= LPCR_HAIL;
5812 		ret = kvmppc_init_vm_radix(kvm);
5813 		if (ret) {
5814 			if (kvmhv_is_nestedv2())
5815 				plpar_guest_delete(0, kvm->arch.lpid);
5816 			else
5817 				kvmppc_free_lpid(kvm->arch.lpid);
5818 			return ret;
5819 		}
5820 		kvmppc_setup_partition_table(kvm);
5821 	}
5822 
5823 	verify_lpcr(kvm, lpcr);
5824 	kvm->arch.lpcr = lpcr;
5825 
5826 	/* Initialization for future HPT resizes */
5827 	kvm->arch.resize_hpt = NULL;
5828 
5829 	/*
5830 	 * Work out how many sets the TLB has, for the use of
5831 	 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
5832 	 */
5833 	if (cpu_has_feature(CPU_FTR_ARCH_31)) {
5834 		/*
5835 		 * P10 will flush all the congruence class with a single tlbiel
5836 		 */
5837 		kvm->arch.tlb_sets = 1;
5838 	} else if (radix_enabled())
5839 		kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;	/* 128 */
5840 	else if (cpu_has_feature(CPU_FTR_ARCH_300))
5841 		kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;	/* 256 */
5842 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
5843 		kvm->arch.tlb_sets = POWER8_TLB_SETS;		/* 512 */
5844 	else
5845 		kvm->arch.tlb_sets = POWER7_TLB_SETS;		/* 128 */
5846 
5847 	/*
5848 	 * Track that we now have a HV mode VM active. This blocks secondary
5849 	 * CPU threads from coming online.
5850 	 */
5851 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5852 		kvm_hv_vm_activated();
5853 
5854 	/*
5855 	 * Initialize smt_mode depending on processor.
5856 	 * POWER8 and earlier have to use "strict" threading, where
5857 	 * all vCPUs in a vcore have to run on the same (sub)core,
5858 	 * whereas on POWER9 the threads can each run a different
5859 	 * guest.
5860 	 */
5861 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5862 		kvm->arch.smt_mode = threads_per_subcore;
5863 	else
5864 		kvm->arch.smt_mode = 1;
5865 	kvm->arch.emul_smt_mode = 1;
5866 
5867 	return 0;
5868 }
5869 
5870 static int kvmppc_arch_create_vm_debugfs_hv(struct kvm *kvm)
5871 {
5872 	kvmppc_mmu_debugfs_init(kvm);
5873 	if (radix_enabled())
5874 		kvmhv_radix_debugfs_init(kvm);
5875 	return 0;
5876 }
5877 
5878 static void kvmppc_free_vcores(struct kvm *kvm)
5879 {
5880 	long int i;
5881 
5882 	for (i = 0; i < KVM_MAX_VCORES; ++i)
5883 		kfree(kvm->arch.vcores[i]);
5884 	kvm->arch.online_vcores = 0;
5885 }
5886 
5887 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
5888 {
5889 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5890 		kvm_hv_vm_deactivated();
5891 
5892 	kvmppc_free_vcores(kvm);
5893 
5894 
5895 	if (kvm_is_radix(kvm))
5896 		kvmppc_free_radix(kvm);
5897 	else
5898 		kvmppc_free_hpt(&kvm->arch.hpt);
5899 
5900 	/* Perform global invalidation and return lpid to the pool */
5901 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5902 		if (nesting_enabled(kvm))
5903 			kvmhv_release_all_nested(kvm);
5904 		kvm->arch.process_table = 0;
5905 		if (kvm->arch.secure_guest)
5906 			uv_svm_terminate(kvm->arch.lpid);
5907 		if (!kvmhv_is_nestedv2())
5908 			kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
5909 	}
5910 
5911 	if (kvmhv_is_nestedv2()) {
5912 		kvmhv_flush_lpid(kvm->arch.lpid);
5913 		plpar_guest_delete(0, kvm->arch.lpid);
5914 	} else {
5915 		kvmppc_free_lpid(kvm->arch.lpid);
5916 	}
5917 
5918 	kvmppc_free_pimap(kvm);
5919 }
5920 
5921 /* We don't need to emulate any privileged instructions or dcbz */
5922 static int kvmppc_core_emulate_op_hv(struct kvm_vcpu *vcpu,
5923 				     unsigned int inst, int *advance)
5924 {
5925 	return EMULATE_FAIL;
5926 }
5927 
5928 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
5929 					ulong spr_val)
5930 {
5931 	return EMULATE_FAIL;
5932 }
5933 
5934 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
5935 					ulong *spr_val)
5936 {
5937 	return EMULATE_FAIL;
5938 }
5939 
5940 static int kvmppc_core_check_processor_compat_hv(void)
5941 {
5942 	if (cpu_has_feature(CPU_FTR_HVMODE) &&
5943 	    cpu_has_feature(CPU_FTR_ARCH_206))
5944 		return 0;
5945 
5946 	/* POWER9 in radix mode is capable of being a nested hypervisor. */
5947 	if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
5948 		return 0;
5949 
5950 	return -EIO;
5951 }
5952 
5953 #ifdef CONFIG_KVM_XICS
5954 
5955 void kvmppc_free_pimap(struct kvm *kvm)
5956 {
5957 	kfree(kvm->arch.pimap);
5958 }
5959 
5960 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
5961 {
5962 	return kzalloc_obj(struct kvmppc_passthru_irqmap);
5963 }
5964 
5965 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5966 {
5967 	struct irq_desc *desc;
5968 	struct kvmppc_irq_map *irq_map;
5969 	struct kvmppc_passthru_irqmap *pimap;
5970 	struct irq_chip *chip;
5971 	int i, rc = 0;
5972 	struct irq_data *host_data;
5973 
5974 	if (!kvm_irq_bypass)
5975 		return 1;
5976 
5977 	desc = irq_to_desc(host_irq);
5978 	if (!desc)
5979 		return -EIO;
5980 
5981 	mutex_lock(&kvm->lock);
5982 
5983 	pimap = kvm->arch.pimap;
5984 	if (pimap == NULL) {
5985 		/* First call, allocate structure to hold IRQ map */
5986 		pimap = kvmppc_alloc_pimap();
5987 		if (pimap == NULL) {
5988 			mutex_unlock(&kvm->lock);
5989 			return -ENOMEM;
5990 		}
5991 		kvm->arch.pimap = pimap;
5992 	}
5993 
5994 	/*
5995 	 * For now, we only support interrupts for which the EOI operation
5996 	 * is an OPAL call followed by a write to XIRR, since that's
5997 	 * what our real-mode EOI code does, or a XIVE interrupt
5998 	 */
5999 	chip = irq_data_get_irq_chip(&desc->irq_data);
6000 	if (!chip || !is_pnv_opal_msi(chip)) {
6001 		pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
6002 			host_irq, guest_gsi);
6003 		mutex_unlock(&kvm->lock);
6004 		return -ENOENT;
6005 	}
6006 
6007 	/*
6008 	 * See if we already have an entry for this guest IRQ number.
6009 	 * If it's mapped to a hardware IRQ number, that's an error,
6010 	 * otherwise re-use this entry.
6011 	 */
6012 	for (i = 0; i < pimap->n_mapped; i++) {
6013 		if (guest_gsi == pimap->mapped[i].v_hwirq) {
6014 			if (pimap->mapped[i].r_hwirq) {
6015 				mutex_unlock(&kvm->lock);
6016 				return -EINVAL;
6017 			}
6018 			break;
6019 		}
6020 	}
6021 
6022 	if (i == KVMPPC_PIRQ_MAPPED) {
6023 		mutex_unlock(&kvm->lock);
6024 		return -EAGAIN;		/* table is full */
6025 	}
6026 
6027 	irq_map = &pimap->mapped[i];
6028 
6029 	irq_map->v_hwirq = guest_gsi;
6030 	irq_map->desc = desc;
6031 
6032 	/*
6033 	 * Order the above two stores before the next to serialize with
6034 	 * the KVM real mode handler.
6035 	 */
6036 	smp_wmb();
6037 
6038 	/*
6039 	 * The 'host_irq' number is mapped in the PCI-MSI domain but
6040 	 * the underlying calls, which will EOI the interrupt in real
6041 	 * mode, need an HW IRQ number mapped in the XICS IRQ domain.
6042 	 */
6043 	host_data = irq_domain_get_irq_data(irq_get_default_domain(), host_irq);
6044 	irq_map->r_hwirq = (unsigned int)irqd_to_hwirq(host_data);
6045 
6046 	if (i == pimap->n_mapped)
6047 		pimap->n_mapped++;
6048 
6049 	if (xics_on_xive())
6050 		rc = kvmppc_xive_set_mapped(kvm, guest_gsi, host_irq);
6051 	else
6052 		kvmppc_xics_set_mapped(kvm, guest_gsi, irq_map->r_hwirq);
6053 	if (rc)
6054 		irq_map->r_hwirq = 0;
6055 
6056 	mutex_unlock(&kvm->lock);
6057 
6058 	return 0;
6059 }
6060 
6061 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
6062 {
6063 	struct irq_desc *desc;
6064 	struct kvmppc_passthru_irqmap *pimap;
6065 	int i, rc = 0;
6066 
6067 	if (!kvm_irq_bypass)
6068 		return 0;
6069 
6070 	desc = irq_to_desc(host_irq);
6071 	if (!desc)
6072 		return -EIO;
6073 
6074 	mutex_lock(&kvm->lock);
6075 	if (!kvm->arch.pimap)
6076 		goto unlock;
6077 
6078 	pimap = kvm->arch.pimap;
6079 
6080 	for (i = 0; i < pimap->n_mapped; i++) {
6081 		if (guest_gsi == pimap->mapped[i].v_hwirq)
6082 			break;
6083 	}
6084 
6085 	if (i == pimap->n_mapped) {
6086 		mutex_unlock(&kvm->lock);
6087 		return -ENODEV;
6088 	}
6089 
6090 	if (xics_on_xive())
6091 		rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, host_irq);
6092 	else
6093 		kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
6094 
6095 	/* invalidate the entry (what to do on error from the above ?) */
6096 	pimap->mapped[i].r_hwirq = 0;
6097 
6098 	/*
6099 	 * We don't free this structure even when the count goes to
6100 	 * zero. The structure is freed when we destroy the VM.
6101 	 */
6102  unlock:
6103 	mutex_unlock(&kvm->lock);
6104 	return rc;
6105 }
6106 
6107 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
6108 					     struct irq_bypass_producer *prod)
6109 {
6110 	int ret = 0;
6111 	struct kvm_kernel_irqfd *irqfd =
6112 		container_of(cons, struct kvm_kernel_irqfd, consumer);
6113 
6114 	irqfd->producer = prod;
6115 
6116 	ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
6117 	if (ret)
6118 		pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
6119 			prod->irq, irqfd->gsi, ret);
6120 
6121 	return ret;
6122 }
6123 
6124 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
6125 					      struct irq_bypass_producer *prod)
6126 {
6127 	int ret;
6128 	struct kvm_kernel_irqfd *irqfd =
6129 		container_of(cons, struct kvm_kernel_irqfd, consumer);
6130 
6131 	irqfd->producer = NULL;
6132 
6133 	/*
6134 	 * When producer of consumer is unregistered, we change back to
6135 	 * default external interrupt handling mode - KVM real mode
6136 	 * will switch back to host.
6137 	 */
6138 	ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
6139 	if (ret)
6140 		pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
6141 			prod->irq, irqfd->gsi, ret);
6142 }
6143 #endif
6144 
6145 static int kvm_arch_vm_ioctl_hv(struct file *filp,
6146 				unsigned int ioctl, unsigned long arg)
6147 {
6148 	struct kvm *kvm __maybe_unused = filp->private_data;
6149 	void __user *argp = (void __user *)arg;
6150 	int r;
6151 
6152 	switch (ioctl) {
6153 
6154 	case KVM_PPC_ALLOCATE_HTAB: {
6155 		u32 htab_order;
6156 
6157 		/* If we're a nested hypervisor, we currently only support radix */
6158 		if (kvmhv_on_pseries()) {
6159 			r = -EOPNOTSUPP;
6160 			break;
6161 		}
6162 
6163 		r = -EFAULT;
6164 		if (get_user(htab_order, (u32 __user *)argp))
6165 			break;
6166 		r = kvmppc_alloc_reset_hpt(kvm, htab_order);
6167 		if (r)
6168 			break;
6169 		r = 0;
6170 		break;
6171 	}
6172 
6173 	case KVM_PPC_GET_HTAB_FD: {
6174 		struct kvm_get_htab_fd ghf;
6175 
6176 		r = -EFAULT;
6177 		if (copy_from_user(&ghf, argp, sizeof(ghf)))
6178 			break;
6179 		r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
6180 		break;
6181 	}
6182 
6183 	case KVM_PPC_RESIZE_HPT_PREPARE: {
6184 		struct kvm_ppc_resize_hpt rhpt;
6185 
6186 		r = -EFAULT;
6187 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
6188 			break;
6189 
6190 		r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
6191 		break;
6192 	}
6193 
6194 	case KVM_PPC_RESIZE_HPT_COMMIT: {
6195 		struct kvm_ppc_resize_hpt rhpt;
6196 
6197 		r = -EFAULT;
6198 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
6199 			break;
6200 
6201 		r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
6202 		break;
6203 	}
6204 
6205 	default:
6206 		r = -ENOTTY;
6207 	}
6208 
6209 	return r;
6210 }
6211 
6212 /*
6213  * List of hcall numbers to enable by default.
6214  * For compatibility with old userspace, we enable by default
6215  * all hcalls that were implemented before the hcall-enabling
6216  * facility was added.  Note this list should not include H_RTAS.
6217  */
6218 static unsigned int default_hcall_list[] = {
6219 	H_REMOVE,
6220 	H_ENTER,
6221 	H_READ,
6222 	H_PROTECT,
6223 	H_BULK_REMOVE,
6224 #ifdef CONFIG_SPAPR_TCE_IOMMU
6225 	H_GET_TCE,
6226 	H_PUT_TCE,
6227 #endif
6228 	H_SET_DABR,
6229 	H_SET_XDABR,
6230 	H_CEDE,
6231 	H_PROD,
6232 	H_CONFER,
6233 	H_REGISTER_VPA,
6234 #ifdef CONFIG_KVM_XICS
6235 	H_EOI,
6236 	H_CPPR,
6237 	H_IPI,
6238 	H_IPOLL,
6239 	H_XIRR,
6240 	H_XIRR_X,
6241 #endif
6242 	0
6243 };
6244 
6245 static void init_default_hcalls(void)
6246 {
6247 	int i;
6248 	unsigned int hcall;
6249 
6250 	for (i = 0; default_hcall_list[i]; ++i) {
6251 		hcall = default_hcall_list[i];
6252 		WARN_ON(!kvmppc_hcall_impl_hv(hcall));
6253 		__set_bit(hcall / 4, default_enabled_hcalls);
6254 	}
6255 }
6256 
6257 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
6258 {
6259 	unsigned long lpcr;
6260 	int radix;
6261 	int err;
6262 
6263 	/* If not on a POWER9, reject it */
6264 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
6265 		return -ENODEV;
6266 
6267 	/* If any unknown flags set, reject it */
6268 	if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
6269 		return -EINVAL;
6270 
6271 	/* GR (guest radix) bit in process_table field must match */
6272 	radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
6273 	if (!!(cfg->process_table & PATB_GR) != radix)
6274 		return -EINVAL;
6275 
6276 	/* Process table size field must be reasonable, i.e. <= 24 */
6277 	if ((cfg->process_table & PRTS_MASK) > 24)
6278 		return -EINVAL;
6279 
6280 	/* We can change a guest to/from radix now, if the host is radix */
6281 	if (radix && !radix_enabled())
6282 		return -EINVAL;
6283 
6284 	/* If we're a nested hypervisor, we currently only support radix */
6285 	if (kvmhv_on_pseries() && !radix)
6286 		return -EINVAL;
6287 
6288 	mutex_lock(&kvm->arch.mmu_setup_lock);
6289 	if (radix != kvm_is_radix(kvm)) {
6290 		if (kvm->arch.mmu_ready) {
6291 			kvm->arch.mmu_ready = 0;
6292 			/* order mmu_ready vs. vcpus_running */
6293 			smp_mb();
6294 			if (atomic_read(&kvm->arch.vcpus_running)) {
6295 				kvm->arch.mmu_ready = 1;
6296 				err = -EBUSY;
6297 				goto out_unlock;
6298 			}
6299 		}
6300 		if (radix)
6301 			err = kvmppc_switch_mmu_to_radix(kvm);
6302 		else
6303 			err = kvmppc_switch_mmu_to_hpt(kvm);
6304 		if (err)
6305 			goto out_unlock;
6306 	}
6307 
6308 	kvm->arch.process_table = cfg->process_table;
6309 	kvmppc_setup_partition_table(kvm);
6310 
6311 	lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
6312 	kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
6313 	err = 0;
6314 
6315  out_unlock:
6316 	mutex_unlock(&kvm->arch.mmu_setup_lock);
6317 	return err;
6318 }
6319 
6320 static int kvmhv_enable_nested(struct kvm *kvm)
6321 {
6322 	if (!nested)
6323 		return -EPERM;
6324 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
6325 		return -ENODEV;
6326 	if (!radix_enabled())
6327 		return -ENODEV;
6328 	if (kvmhv_is_nestedv2())
6329 		return -ENODEV;
6330 
6331 	/* kvm == NULL means the caller is testing if the capability exists */
6332 	if (kvm)
6333 		kvm->arch.nested_enable = true;
6334 	return 0;
6335 }
6336 
6337 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
6338 				 int size)
6339 {
6340 	int rc = -EINVAL;
6341 
6342 	if (kvmhv_vcpu_is_radix(vcpu)) {
6343 		rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
6344 
6345 		if (rc > 0)
6346 			rc = -EINVAL;
6347 	}
6348 
6349 	/* For now quadrants are the only way to access nested guest memory */
6350 	if (rc && vcpu->arch.nested)
6351 		rc = -EAGAIN;
6352 
6353 	return rc;
6354 }
6355 
6356 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
6357 				int size)
6358 {
6359 	int rc = -EINVAL;
6360 
6361 	if (kvmhv_vcpu_is_radix(vcpu)) {
6362 		rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
6363 
6364 		if (rc > 0)
6365 			rc = -EINVAL;
6366 	}
6367 
6368 	/* For now quadrants are the only way to access nested guest memory */
6369 	if (rc && vcpu->arch.nested)
6370 		rc = -EAGAIN;
6371 
6372 	return rc;
6373 }
6374 
6375 static void unpin_vpa_reset(struct kvm *kvm, struct kvmppc_vpa *vpa)
6376 {
6377 	unpin_vpa(kvm, vpa);
6378 	vpa->gpa = 0;
6379 	vpa->pinned_addr = NULL;
6380 	vpa->dirty = false;
6381 	vpa->update_pending = 0;
6382 }
6383 
6384 /*
6385  * Enable a guest to become a secure VM, or test whether
6386  * that could be enabled.
6387  * Called when the KVM_CAP_PPC_SECURE_GUEST capability is
6388  * tested (kvm == NULL) or enabled (kvm != NULL).
6389  */
6390 static int kvmhv_enable_svm(struct kvm *kvm)
6391 {
6392 	if (!kvmppc_uvmem_available())
6393 		return -EINVAL;
6394 	if (kvm)
6395 		kvm->arch.svm_enabled = 1;
6396 	return 0;
6397 }
6398 
6399 /*
6400  *  IOCTL handler to turn off secure mode of guest
6401  *
6402  * - Release all device pages
6403  * - Issue ucall to terminate the guest on the UV side
6404  * - Unpin the VPA pages.
6405  * - Reinit the partition scoped page tables
6406  */
6407 static int kvmhv_svm_off(struct kvm *kvm)
6408 {
6409 	struct kvm_vcpu *vcpu;
6410 	int mmu_was_ready;
6411 	int srcu_idx;
6412 	int ret = 0;
6413 	unsigned long i;
6414 
6415 	if (!(kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START))
6416 		return ret;
6417 
6418 	mutex_lock(&kvm->arch.mmu_setup_lock);
6419 	mmu_was_ready = kvm->arch.mmu_ready;
6420 	if (kvm->arch.mmu_ready) {
6421 		kvm->arch.mmu_ready = 0;
6422 		/* order mmu_ready vs. vcpus_running */
6423 		smp_mb();
6424 		if (atomic_read(&kvm->arch.vcpus_running)) {
6425 			kvm->arch.mmu_ready = 1;
6426 			ret = -EBUSY;
6427 			goto out;
6428 		}
6429 	}
6430 
6431 	srcu_idx = srcu_read_lock(&kvm->srcu);
6432 	for (i = 0; i < kvm_arch_nr_memslot_as_ids(kvm); i++) {
6433 		struct kvm_memory_slot *memslot;
6434 		struct kvm_memslots *slots = __kvm_memslots(kvm, i);
6435 		int bkt;
6436 
6437 		if (!slots)
6438 			continue;
6439 
6440 		kvm_for_each_memslot(memslot, bkt, slots) {
6441 			kvmppc_uvmem_drop_pages(memslot, kvm, true);
6442 			uv_unregister_mem_slot(kvm->arch.lpid, memslot->id);
6443 		}
6444 	}
6445 	srcu_read_unlock(&kvm->srcu, srcu_idx);
6446 
6447 	ret = uv_svm_terminate(kvm->arch.lpid);
6448 	if (ret != U_SUCCESS) {
6449 		ret = -EINVAL;
6450 		goto out;
6451 	}
6452 
6453 	/*
6454 	 * When secure guest is reset, all the guest pages are sent
6455 	 * to UV via UV_PAGE_IN before the non-boot vcpus get a
6456 	 * chance to run and unpin their VPA pages. Unpinning of all
6457 	 * VPA pages is done here explicitly so that VPA pages
6458 	 * can be migrated to the secure side.
6459 	 *
6460 	 * This is required to for the secure SMP guest to reboot
6461 	 * correctly.
6462 	 */
6463 	kvm_for_each_vcpu(i, vcpu, kvm) {
6464 		spin_lock(&vcpu->arch.vpa_update_lock);
6465 		unpin_vpa_reset(kvm, &vcpu->arch.dtl);
6466 		unpin_vpa_reset(kvm, &vcpu->arch.slb_shadow);
6467 		unpin_vpa_reset(kvm, &vcpu->arch.vpa);
6468 		spin_unlock(&vcpu->arch.vpa_update_lock);
6469 	}
6470 
6471 	kvmppc_setup_partition_table(kvm);
6472 	kvm->arch.secure_guest = 0;
6473 	kvm->arch.mmu_ready = mmu_was_ready;
6474 out:
6475 	mutex_unlock(&kvm->arch.mmu_setup_lock);
6476 	return ret;
6477 }
6478 
6479 static int kvmhv_enable_dawr1(struct kvm *kvm)
6480 {
6481 	if (!cpu_has_feature(CPU_FTR_DAWR1))
6482 		return -ENODEV;
6483 
6484 	/* kvm == NULL means the caller is testing if the capability exists */
6485 	if (kvm)
6486 		kvm->arch.dawr1_enabled = true;
6487 	return 0;
6488 }
6489 
6490 static bool kvmppc_hash_v3_possible(void)
6491 {
6492 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
6493 		return false;
6494 
6495 	if (!cpu_has_feature(CPU_FTR_HVMODE))
6496 		return false;
6497 
6498 	/*
6499 	 * POWER9 chips before version 2.02 can't have some threads in
6500 	 * HPT mode and some in radix mode on the same core.
6501 	 */
6502 	if (radix_enabled()) {
6503 		unsigned int pvr = mfspr(SPRN_PVR);
6504 		if ((pvr >> 16) == PVR_POWER9 &&
6505 		    (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
6506 		     ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
6507 			return false;
6508 	}
6509 
6510 	return true;
6511 }
6512 
6513 static struct kvmppc_ops kvm_ops_hv = {
6514 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
6515 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
6516 	.get_one_reg = kvmppc_get_one_reg_hv,
6517 	.set_one_reg = kvmppc_set_one_reg_hv,
6518 	.vcpu_load   = kvmppc_core_vcpu_load_hv,
6519 	.vcpu_put    = kvmppc_core_vcpu_put_hv,
6520 	.inject_interrupt = kvmppc_inject_interrupt_hv,
6521 	.set_msr     = kvmppc_set_msr_hv,
6522 	.vcpu_run    = kvmppc_vcpu_run_hv,
6523 	.vcpu_create = kvmppc_core_vcpu_create_hv,
6524 	.vcpu_free   = kvmppc_core_vcpu_free_hv,
6525 	.check_requests = kvmppc_core_check_requests_hv,
6526 	.get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
6527 	.flush_memslot  = kvmppc_core_flush_memslot_hv,
6528 	.prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
6529 	.commit_memory_region  = kvmppc_core_commit_memory_region_hv,
6530 	.unmap_gfn_range = kvm_unmap_gfn_range_hv,
6531 	.age_gfn = kvm_age_gfn_hv,
6532 	.test_age_gfn = kvm_test_age_gfn_hv,
6533 	.free_memslot = kvmppc_core_free_memslot_hv,
6534 	.init_vm =  kvmppc_core_init_vm_hv,
6535 	.destroy_vm = kvmppc_core_destroy_vm_hv,
6536 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
6537 	.emulate_op = kvmppc_core_emulate_op_hv,
6538 	.emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
6539 	.emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
6540 	.fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
6541 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
6542 	.hcall_implemented = kvmppc_hcall_impl_hv,
6543 	.configure_mmu = kvmhv_configure_mmu,
6544 	.get_rmmu_info = kvmhv_get_rmmu_info,
6545 	.set_smt_mode = kvmhv_set_smt_mode,
6546 	.enable_nested = kvmhv_enable_nested,
6547 	.load_from_eaddr = kvmhv_load_from_eaddr,
6548 	.store_to_eaddr = kvmhv_store_to_eaddr,
6549 	.enable_svm = kvmhv_enable_svm,
6550 	.svm_off = kvmhv_svm_off,
6551 	.enable_dawr1 = kvmhv_enable_dawr1,
6552 	.hash_v3_possible = kvmppc_hash_v3_possible,
6553 	.create_vcpu_debugfs = kvmppc_arch_create_vcpu_debugfs_hv,
6554 	.create_vm_debugfs = kvmppc_arch_create_vm_debugfs_hv,
6555 };
6556 
6557 static int kvm_init_subcore_bitmap(void)
6558 {
6559 	int i, j;
6560 	int nr_cores = cpu_nr_cores();
6561 	struct sibling_subcore_state *sibling_subcore_state;
6562 
6563 	for (i = 0; i < nr_cores; i++) {
6564 		int first_cpu = i * threads_per_core;
6565 		int node = cpu_to_node(first_cpu);
6566 
6567 		/* Ignore if it is already allocated. */
6568 		if (paca_ptrs[first_cpu]->sibling_subcore_state)
6569 			continue;
6570 
6571 		sibling_subcore_state =
6572 			kzalloc_node(sizeof(struct sibling_subcore_state),
6573 							GFP_KERNEL, node);
6574 		if (!sibling_subcore_state)
6575 			return -ENOMEM;
6576 
6577 
6578 		for (j = 0; j < threads_per_core; j++) {
6579 			int cpu = first_cpu + j;
6580 
6581 			paca_ptrs[cpu]->sibling_subcore_state =
6582 						sibling_subcore_state;
6583 		}
6584 	}
6585 	return 0;
6586 }
6587 
6588 static int kvmppc_radix_possible(void)
6589 {
6590 	return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
6591 }
6592 
6593 static int kvmppc_book3s_init_hv(void)
6594 {
6595 	int r;
6596 
6597 	if (!tlbie_capable) {
6598 		pr_err("KVM-HV: Host does not support TLBIE\n");
6599 		return -ENODEV;
6600 	}
6601 
6602 	/*
6603 	 * FIXME!! Do we need to check on all cpus ?
6604 	 */
6605 	r = kvmppc_core_check_processor_compat_hv();
6606 	if (r < 0)
6607 		return -ENODEV;
6608 
6609 	r = kvmhv_nested_init();
6610 	if (r)
6611 		return r;
6612 
6613 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
6614 		r = kvm_init_subcore_bitmap();
6615 		if (r)
6616 			goto err;
6617 	}
6618 
6619 	/*
6620 	 * We need a way of accessing the XICS interrupt controller,
6621 	 * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
6622 	 * indirectly, via OPAL.
6623 	 */
6624 #ifdef CONFIG_SMP
6625 	if (!xics_on_xive() && !kvmhv_on_pseries() &&
6626 	    !local_paca->kvm_hstate.xics_phys) {
6627 		struct device_node *np;
6628 
6629 		np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
6630 		if (!np) {
6631 			pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
6632 			r = -ENODEV;
6633 			goto err;
6634 		}
6635 		/* presence of intc confirmed - node can be dropped again */
6636 		of_node_put(np);
6637 	}
6638 #endif
6639 
6640 	init_default_hcalls();
6641 
6642 	init_vcore_lists();
6643 
6644 	r = kvmppc_mmu_hv_init();
6645 	if (r)
6646 		goto err;
6647 
6648 	if (kvmppc_radix_possible()) {
6649 		r = kvmppc_radix_init();
6650 		if (r)
6651 			goto err;
6652 	}
6653 
6654 	r = kvmppc_uvmem_init();
6655 	if (r < 0) {
6656 		pr_err("KVM-HV: kvmppc_uvmem_init failed %d\n", r);
6657 		return r;
6658 	}
6659 
6660 #if defined(CONFIG_KVM_XICS)
6661 	/*
6662 	 * IRQ bypass is supported only for interrupts whose EOI operations are
6663 	 * handled via OPAL calls. Therefore, register IRQ bypass handlers
6664 	 * exclusively for PowerNV KVM when booted with 'xive=off', indicating
6665 	 * the use of the emulated XICS interrupt controller.
6666 	 */
6667 	if (!kvmhv_on_pseries()) {
6668 		pr_info("KVM-HV: Enabling IRQ bypass\n");
6669 		kvm_ops_hv.irq_bypass_add_producer =
6670 			kvmppc_irq_bypass_add_producer_hv;
6671 		kvm_ops_hv.irq_bypass_del_producer =
6672 			kvmppc_irq_bypass_del_producer_hv;
6673 	}
6674 #endif
6675 
6676 	kvm_ops_hv.owner = THIS_MODULE;
6677 	kvmppc_hv_ops = &kvm_ops_hv;
6678 
6679 	return 0;
6680 
6681 err:
6682 	kvmhv_nested_exit();
6683 	kvmppc_radix_exit();
6684 
6685 	return r;
6686 }
6687 
6688 static void kvmppc_book3s_exit_hv(void)
6689 {
6690 	kvmppc_uvmem_free();
6691 	kvmppc_free_host_rm_ops();
6692 	if (kvmppc_radix_possible())
6693 		kvmppc_radix_exit();
6694 	kvmppc_hv_ops = NULL;
6695 	kvmhv_nested_exit();
6696 }
6697 
6698 module_init(kvmppc_book3s_init_hv);
6699 module_exit(kvmppc_book3s_exit_hv);
6700 MODULE_DESCRIPTION("KVM on Book3S (POWER8 and later) in hypervisor mode");
6701 MODULE_LICENSE("GPL");
6702 MODULE_ALIAS_MISCDEV(KVM_MINOR);
6703 MODULE_ALIAS("devname:kvm");
6704