xref: /openbmc/linux/arch/powerpc/kvm/book3s_hv.c (revision 3db55767)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
4  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
5  *
6  * Authors:
7  *    Paul Mackerras <paulus@au1.ibm.com>
8  *    Alexander Graf <agraf@suse.de>
9  *    Kevin Wolf <mail@kevin-wolf.de>
10  *
11  * Description: KVM functions specific to running on Book 3S
12  * processors in hypervisor mode (specifically POWER7 and later).
13  *
14  * This file is derived from arch/powerpc/kvm/book3s.c,
15  * by Alexander Graf <agraf@suse.de>.
16  */
17 
18 #include <linux/kvm_host.h>
19 #include <linux/kernel.h>
20 #include <linux/err.h>
21 #include <linux/slab.h>
22 #include <linux/preempt.h>
23 #include <linux/sched/signal.h>
24 #include <linux/sched/stat.h>
25 #include <linux/delay.h>
26 #include <linux/export.h>
27 #include <linux/fs.h>
28 #include <linux/anon_inodes.h>
29 #include <linux/cpu.h>
30 #include <linux/cpumask.h>
31 #include <linux/spinlock.h>
32 #include <linux/page-flags.h>
33 #include <linux/srcu.h>
34 #include <linux/miscdevice.h>
35 #include <linux/debugfs.h>
36 #include <linux/gfp.h>
37 #include <linux/vmalloc.h>
38 #include <linux/highmem.h>
39 #include <linux/hugetlb.h>
40 #include <linux/kvm_irqfd.h>
41 #include <linux/irqbypass.h>
42 #include <linux/module.h>
43 #include <linux/compiler.h>
44 #include <linux/of.h>
45 #include <linux/irqdomain.h>
46 #include <linux/smp.h>
47 
48 #include <asm/ftrace.h>
49 #include <asm/reg.h>
50 #include <asm/ppc-opcode.h>
51 #include <asm/asm-prototypes.h>
52 #include <asm/archrandom.h>
53 #include <asm/debug.h>
54 #include <asm/disassemble.h>
55 #include <asm/cputable.h>
56 #include <asm/cacheflush.h>
57 #include <linux/uaccess.h>
58 #include <asm/interrupt.h>
59 #include <asm/io.h>
60 #include <asm/kvm_ppc.h>
61 #include <asm/kvm_book3s.h>
62 #include <asm/mmu_context.h>
63 #include <asm/lppaca.h>
64 #include <asm/pmc.h>
65 #include <asm/processor.h>
66 #include <asm/cputhreads.h>
67 #include <asm/page.h>
68 #include <asm/hvcall.h>
69 #include <asm/switch_to.h>
70 #include <asm/smp.h>
71 #include <asm/dbell.h>
72 #include <asm/hmi.h>
73 #include <asm/pnv-pci.h>
74 #include <asm/mmu.h>
75 #include <asm/opal.h>
76 #include <asm/xics.h>
77 #include <asm/xive.h>
78 #include <asm/hw_breakpoint.h>
79 #include <asm/kvm_book3s_uvmem.h>
80 #include <asm/ultravisor.h>
81 #include <asm/dtl.h>
82 #include <asm/plpar_wrappers.h>
83 
84 #include <trace/events/ipi.h>
85 
86 #include "book3s.h"
87 #include "book3s_hv.h"
88 
89 #define CREATE_TRACE_POINTS
90 #include "trace_hv.h"
91 
92 /* #define EXIT_DEBUG */
93 /* #define EXIT_DEBUG_SIMPLE */
94 /* #define EXIT_DEBUG_INT */
95 
96 /* Used to indicate that a guest page fault needs to be handled */
97 #define RESUME_PAGE_FAULT	(RESUME_GUEST | RESUME_FLAG_ARCH1)
98 /* Used to indicate that a guest passthrough interrupt needs to be handled */
99 #define RESUME_PASSTHROUGH	(RESUME_GUEST | RESUME_FLAG_ARCH2)
100 
101 /* Used as a "null" value for timebase values */
102 #define TB_NIL	(~(u64)0)
103 
104 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
105 
106 static int dynamic_mt_modes = 6;
107 module_param(dynamic_mt_modes, int, 0644);
108 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
109 static int target_smt_mode;
110 module_param(target_smt_mode, int, 0644);
111 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
112 
113 static bool one_vm_per_core;
114 module_param(one_vm_per_core, bool, S_IRUGO | S_IWUSR);
115 MODULE_PARM_DESC(one_vm_per_core, "Only run vCPUs from the same VM on a core (requires POWER8 or older)");
116 
117 #ifdef CONFIG_KVM_XICS
118 static const struct kernel_param_ops module_param_ops = {
119 	.set = param_set_int,
120 	.get = param_get_int,
121 };
122 
123 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
124 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
125 
126 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
127 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
128 #endif
129 
130 /* If set, guests are allowed to create and control nested guests */
131 static bool nested = true;
132 module_param(nested, bool, S_IRUGO | S_IWUSR);
133 MODULE_PARM_DESC(nested, "Enable nested virtualization (only on POWER9)");
134 
135 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
136 
137 /*
138  * RWMR values for POWER8.  These control the rate at which PURR
139  * and SPURR count and should be set according to the number of
140  * online threads in the vcore being run.
141  */
142 #define RWMR_RPA_P8_1THREAD	0x164520C62609AECAUL
143 #define RWMR_RPA_P8_2THREAD	0x7FFF2908450D8DA9UL
144 #define RWMR_RPA_P8_3THREAD	0x164520C62609AECAUL
145 #define RWMR_RPA_P8_4THREAD	0x199A421245058DA9UL
146 #define RWMR_RPA_P8_5THREAD	0x164520C62609AECAUL
147 #define RWMR_RPA_P8_6THREAD	0x164520C62609AECAUL
148 #define RWMR_RPA_P8_7THREAD	0x164520C62609AECAUL
149 #define RWMR_RPA_P8_8THREAD	0x164520C62609AECAUL
150 
151 static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = {
152 	RWMR_RPA_P8_1THREAD,
153 	RWMR_RPA_P8_1THREAD,
154 	RWMR_RPA_P8_2THREAD,
155 	RWMR_RPA_P8_3THREAD,
156 	RWMR_RPA_P8_4THREAD,
157 	RWMR_RPA_P8_5THREAD,
158 	RWMR_RPA_P8_6THREAD,
159 	RWMR_RPA_P8_7THREAD,
160 	RWMR_RPA_P8_8THREAD,
161 };
162 
163 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
164 		int *ip)
165 {
166 	int i = *ip;
167 	struct kvm_vcpu *vcpu;
168 
169 	while (++i < MAX_SMT_THREADS) {
170 		vcpu = READ_ONCE(vc->runnable_threads[i]);
171 		if (vcpu) {
172 			*ip = i;
173 			return vcpu;
174 		}
175 	}
176 	return NULL;
177 }
178 
179 /* Used to traverse the list of runnable threads for a given vcore */
180 #define for_each_runnable_thread(i, vcpu, vc) \
181 	for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
182 
183 static bool kvmppc_ipi_thread(int cpu)
184 {
185 	unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
186 
187 	/* If we're a nested hypervisor, fall back to ordinary IPIs for now */
188 	if (kvmhv_on_pseries())
189 		return false;
190 
191 	/* On POWER9 we can use msgsnd to IPI any cpu */
192 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
193 		msg |= get_hard_smp_processor_id(cpu);
194 		smp_mb();
195 		__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
196 		return true;
197 	}
198 
199 	/* On POWER8 for IPIs to threads in the same core, use msgsnd */
200 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
201 		preempt_disable();
202 		if (cpu_first_thread_sibling(cpu) ==
203 		    cpu_first_thread_sibling(smp_processor_id())) {
204 			msg |= cpu_thread_in_core(cpu);
205 			smp_mb();
206 			__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
207 			preempt_enable();
208 			return true;
209 		}
210 		preempt_enable();
211 	}
212 
213 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
214 	if (cpu >= 0 && cpu < nr_cpu_ids) {
215 		if (paca_ptrs[cpu]->kvm_hstate.xics_phys) {
216 			xics_wake_cpu(cpu);
217 			return true;
218 		}
219 		opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
220 		return true;
221 	}
222 #endif
223 
224 	return false;
225 }
226 
227 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
228 {
229 	int cpu;
230 	struct rcuwait *waitp;
231 
232 	/*
233 	 * rcuwait_wake_up contains smp_mb() which orders prior stores that
234 	 * create pending work vs below loads of cpu fields. The other side
235 	 * is the barrier in vcpu run that orders setting the cpu fields vs
236 	 * testing for pending work.
237 	 */
238 
239 	waitp = kvm_arch_vcpu_get_wait(vcpu);
240 	if (rcuwait_wake_up(waitp))
241 		++vcpu->stat.generic.halt_wakeup;
242 
243 	cpu = READ_ONCE(vcpu->arch.thread_cpu);
244 	if (cpu >= 0 && kvmppc_ipi_thread(cpu))
245 		return;
246 
247 	/* CPU points to the first thread of the core */
248 	cpu = vcpu->cpu;
249 	if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
250 		smp_send_reschedule(cpu);
251 }
252 
253 /*
254  * We use the vcpu_load/put functions to measure stolen time.
255  *
256  * Stolen time is counted as time when either the vcpu is able to
257  * run as part of a virtual core, but the task running the vcore
258  * is preempted or sleeping, or when the vcpu needs something done
259  * in the kernel by the task running the vcpu, but that task is
260  * preempted or sleeping.  Those two things have to be counted
261  * separately, since one of the vcpu tasks will take on the job
262  * of running the core, and the other vcpu tasks in the vcore will
263  * sleep waiting for it to do that, but that sleep shouldn't count
264  * as stolen time.
265  *
266  * Hence we accumulate stolen time when the vcpu can run as part of
267  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
268  * needs its task to do other things in the kernel (for example,
269  * service a page fault) in busy_stolen.  We don't accumulate
270  * stolen time for a vcore when it is inactive, or for a vcpu
271  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
272  * a misnomer; it means that the vcpu task is not executing in
273  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
274  * the kernel.  We don't have any way of dividing up that time
275  * between time that the vcpu is genuinely stopped, time that
276  * the task is actively working on behalf of the vcpu, and time
277  * that the task is preempted, so we don't count any of it as
278  * stolen.
279  *
280  * Updates to busy_stolen are protected by arch.tbacct_lock;
281  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
282  * lock.  The stolen times are measured in units of timebase ticks.
283  * (Note that the != TB_NIL checks below are purely defensive;
284  * they should never fail.)
285  *
286  * The POWER9 path is simpler, one vcpu per virtual core so the
287  * former case does not exist. If a vcpu is preempted when it is
288  * BUSY_IN_HOST and not ceded or otherwise blocked, then accumulate
289  * the stolen cycles in busy_stolen. RUNNING is not a preemptible
290  * state in the P9 path.
291  */
292 
293 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc, u64 tb)
294 {
295 	unsigned long flags;
296 
297 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
298 
299 	spin_lock_irqsave(&vc->stoltb_lock, flags);
300 	vc->preempt_tb = tb;
301 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
302 }
303 
304 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc, u64 tb)
305 {
306 	unsigned long flags;
307 
308 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
309 
310 	spin_lock_irqsave(&vc->stoltb_lock, flags);
311 	if (vc->preempt_tb != TB_NIL) {
312 		vc->stolen_tb += tb - vc->preempt_tb;
313 		vc->preempt_tb = TB_NIL;
314 	}
315 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
316 }
317 
318 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
319 {
320 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
321 	unsigned long flags;
322 	u64 now;
323 
324 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
325 		if (vcpu->arch.busy_preempt != TB_NIL) {
326 			WARN_ON_ONCE(vcpu->arch.state != KVMPPC_VCPU_BUSY_IN_HOST);
327 			vc->stolen_tb += mftb() - vcpu->arch.busy_preempt;
328 			vcpu->arch.busy_preempt = TB_NIL;
329 		}
330 		return;
331 	}
332 
333 	now = mftb();
334 
335 	/*
336 	 * We can test vc->runner without taking the vcore lock,
337 	 * because only this task ever sets vc->runner to this
338 	 * vcpu, and once it is set to this vcpu, only this task
339 	 * ever sets it to NULL.
340 	 */
341 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
342 		kvmppc_core_end_stolen(vc, now);
343 
344 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
345 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
346 	    vcpu->arch.busy_preempt != TB_NIL) {
347 		vcpu->arch.busy_stolen += now - vcpu->arch.busy_preempt;
348 		vcpu->arch.busy_preempt = TB_NIL;
349 	}
350 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
351 }
352 
353 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
354 {
355 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
356 	unsigned long flags;
357 	u64 now;
358 
359 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
360 		/*
361 		 * In the P9 path, RUNNABLE is not preemptible
362 		 * (nor takes host interrupts)
363 		 */
364 		WARN_ON_ONCE(vcpu->arch.state == KVMPPC_VCPU_RUNNABLE);
365 		/*
366 		 * Account stolen time when preempted while the vcpu task is
367 		 * running in the kernel (but not in qemu, which is INACTIVE).
368 		 */
369 		if (task_is_running(current) &&
370 				vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
371 			vcpu->arch.busy_preempt = mftb();
372 		return;
373 	}
374 
375 	now = mftb();
376 
377 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
378 		kvmppc_core_start_stolen(vc, now);
379 
380 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
381 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
382 		vcpu->arch.busy_preempt = now;
383 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
384 }
385 
386 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
387 {
388 	vcpu->arch.pvr = pvr;
389 }
390 
391 /* Dummy value used in computing PCR value below */
392 #define PCR_ARCH_31    (PCR_ARCH_300 << 1)
393 
394 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
395 {
396 	unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
397 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
398 
399 	/* We can (emulate) our own architecture version and anything older */
400 	if (cpu_has_feature(CPU_FTR_ARCH_31))
401 		host_pcr_bit = PCR_ARCH_31;
402 	else if (cpu_has_feature(CPU_FTR_ARCH_300))
403 		host_pcr_bit = PCR_ARCH_300;
404 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
405 		host_pcr_bit = PCR_ARCH_207;
406 	else if (cpu_has_feature(CPU_FTR_ARCH_206))
407 		host_pcr_bit = PCR_ARCH_206;
408 	else
409 		host_pcr_bit = PCR_ARCH_205;
410 
411 	/* Determine lowest PCR bit needed to run guest in given PVR level */
412 	guest_pcr_bit = host_pcr_bit;
413 	if (arch_compat) {
414 		switch (arch_compat) {
415 		case PVR_ARCH_205:
416 			guest_pcr_bit = PCR_ARCH_205;
417 			break;
418 		case PVR_ARCH_206:
419 		case PVR_ARCH_206p:
420 			guest_pcr_bit = PCR_ARCH_206;
421 			break;
422 		case PVR_ARCH_207:
423 			guest_pcr_bit = PCR_ARCH_207;
424 			break;
425 		case PVR_ARCH_300:
426 			guest_pcr_bit = PCR_ARCH_300;
427 			break;
428 		case PVR_ARCH_31:
429 			guest_pcr_bit = PCR_ARCH_31;
430 			break;
431 		default:
432 			return -EINVAL;
433 		}
434 	}
435 
436 	/* Check requested PCR bits don't exceed our capabilities */
437 	if (guest_pcr_bit > host_pcr_bit)
438 		return -EINVAL;
439 
440 	spin_lock(&vc->lock);
441 	vc->arch_compat = arch_compat;
442 	/*
443 	 * Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit
444 	 * Also set all reserved PCR bits
445 	 */
446 	vc->pcr = (host_pcr_bit - guest_pcr_bit) | PCR_MASK;
447 	spin_unlock(&vc->lock);
448 
449 	return 0;
450 }
451 
452 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
453 {
454 	int r;
455 
456 	pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
457 	pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
458 	       vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
459 	for (r = 0; r < 16; ++r)
460 		pr_err("r%2d = %.16lx  r%d = %.16lx\n",
461 		       r, kvmppc_get_gpr(vcpu, r),
462 		       r+16, kvmppc_get_gpr(vcpu, r+16));
463 	pr_err("ctr = %.16lx  lr  = %.16lx\n",
464 	       vcpu->arch.regs.ctr, vcpu->arch.regs.link);
465 	pr_err("srr0 = %.16llx srr1 = %.16llx\n",
466 	       vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
467 	pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
468 	       vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
469 	pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
470 	       vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
471 	pr_err("cr = %.8lx  xer = %.16lx  dsisr = %.8x\n",
472 	       vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
473 	pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
474 	pr_err("fault dar = %.16lx dsisr = %.8x\n",
475 	       vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
476 	pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
477 	for (r = 0; r < vcpu->arch.slb_max; ++r)
478 		pr_err("  ESID = %.16llx VSID = %.16llx\n",
479 		       vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
480 	pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.16lx\n",
481 	       vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
482 	       vcpu->arch.last_inst);
483 }
484 
485 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
486 {
487 	return kvm_get_vcpu_by_id(kvm, id);
488 }
489 
490 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
491 {
492 	vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
493 	vpa->yield_count = cpu_to_be32(1);
494 }
495 
496 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
497 		   unsigned long addr, unsigned long len)
498 {
499 	/* check address is cacheline aligned */
500 	if (addr & (L1_CACHE_BYTES - 1))
501 		return -EINVAL;
502 	spin_lock(&vcpu->arch.vpa_update_lock);
503 	if (v->next_gpa != addr || v->len != len) {
504 		v->next_gpa = addr;
505 		v->len = addr ? len : 0;
506 		v->update_pending = 1;
507 	}
508 	spin_unlock(&vcpu->arch.vpa_update_lock);
509 	return 0;
510 }
511 
512 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
513 struct reg_vpa {
514 	u32 dummy;
515 	union {
516 		__be16 hword;
517 		__be32 word;
518 	} length;
519 };
520 
521 static int vpa_is_registered(struct kvmppc_vpa *vpap)
522 {
523 	if (vpap->update_pending)
524 		return vpap->next_gpa != 0;
525 	return vpap->pinned_addr != NULL;
526 }
527 
528 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
529 				       unsigned long flags,
530 				       unsigned long vcpuid, unsigned long vpa)
531 {
532 	struct kvm *kvm = vcpu->kvm;
533 	unsigned long len, nb;
534 	void *va;
535 	struct kvm_vcpu *tvcpu;
536 	int err;
537 	int subfunc;
538 	struct kvmppc_vpa *vpap;
539 
540 	tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
541 	if (!tvcpu)
542 		return H_PARAMETER;
543 
544 	subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
545 	if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
546 	    subfunc == H_VPA_REG_SLB) {
547 		/* Registering new area - address must be cache-line aligned */
548 		if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
549 			return H_PARAMETER;
550 
551 		/* convert logical addr to kernel addr and read length */
552 		va = kvmppc_pin_guest_page(kvm, vpa, &nb);
553 		if (va == NULL)
554 			return H_PARAMETER;
555 		if (subfunc == H_VPA_REG_VPA)
556 			len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
557 		else
558 			len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
559 		kvmppc_unpin_guest_page(kvm, va, vpa, false);
560 
561 		/* Check length */
562 		if (len > nb || len < sizeof(struct reg_vpa))
563 			return H_PARAMETER;
564 	} else {
565 		vpa = 0;
566 		len = 0;
567 	}
568 
569 	err = H_PARAMETER;
570 	vpap = NULL;
571 	spin_lock(&tvcpu->arch.vpa_update_lock);
572 
573 	switch (subfunc) {
574 	case H_VPA_REG_VPA:		/* register VPA */
575 		/*
576 		 * The size of our lppaca is 1kB because of the way we align
577 		 * it for the guest to avoid crossing a 4kB boundary. We only
578 		 * use 640 bytes of the structure though, so we should accept
579 		 * clients that set a size of 640.
580 		 */
581 		BUILD_BUG_ON(sizeof(struct lppaca) != 640);
582 		if (len < sizeof(struct lppaca))
583 			break;
584 		vpap = &tvcpu->arch.vpa;
585 		err = 0;
586 		break;
587 
588 	case H_VPA_REG_DTL:		/* register DTL */
589 		if (len < sizeof(struct dtl_entry))
590 			break;
591 		len -= len % sizeof(struct dtl_entry);
592 
593 		/* Check that they have previously registered a VPA */
594 		err = H_RESOURCE;
595 		if (!vpa_is_registered(&tvcpu->arch.vpa))
596 			break;
597 
598 		vpap = &tvcpu->arch.dtl;
599 		err = 0;
600 		break;
601 
602 	case H_VPA_REG_SLB:		/* register SLB shadow buffer */
603 		/* Check that they have previously registered a VPA */
604 		err = H_RESOURCE;
605 		if (!vpa_is_registered(&tvcpu->arch.vpa))
606 			break;
607 
608 		vpap = &tvcpu->arch.slb_shadow;
609 		err = 0;
610 		break;
611 
612 	case H_VPA_DEREG_VPA:		/* deregister VPA */
613 		/* Check they don't still have a DTL or SLB buf registered */
614 		err = H_RESOURCE;
615 		if (vpa_is_registered(&tvcpu->arch.dtl) ||
616 		    vpa_is_registered(&tvcpu->arch.slb_shadow))
617 			break;
618 
619 		vpap = &tvcpu->arch.vpa;
620 		err = 0;
621 		break;
622 
623 	case H_VPA_DEREG_DTL:		/* deregister DTL */
624 		vpap = &tvcpu->arch.dtl;
625 		err = 0;
626 		break;
627 
628 	case H_VPA_DEREG_SLB:		/* deregister SLB shadow buffer */
629 		vpap = &tvcpu->arch.slb_shadow;
630 		err = 0;
631 		break;
632 	}
633 
634 	if (vpap) {
635 		vpap->next_gpa = vpa;
636 		vpap->len = len;
637 		vpap->update_pending = 1;
638 	}
639 
640 	spin_unlock(&tvcpu->arch.vpa_update_lock);
641 
642 	return err;
643 }
644 
645 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
646 {
647 	struct kvm *kvm = vcpu->kvm;
648 	void *va;
649 	unsigned long nb;
650 	unsigned long gpa;
651 
652 	/*
653 	 * We need to pin the page pointed to by vpap->next_gpa,
654 	 * but we can't call kvmppc_pin_guest_page under the lock
655 	 * as it does get_user_pages() and down_read().  So we
656 	 * have to drop the lock, pin the page, then get the lock
657 	 * again and check that a new area didn't get registered
658 	 * in the meantime.
659 	 */
660 	for (;;) {
661 		gpa = vpap->next_gpa;
662 		spin_unlock(&vcpu->arch.vpa_update_lock);
663 		va = NULL;
664 		nb = 0;
665 		if (gpa)
666 			va = kvmppc_pin_guest_page(kvm, gpa, &nb);
667 		spin_lock(&vcpu->arch.vpa_update_lock);
668 		if (gpa == vpap->next_gpa)
669 			break;
670 		/* sigh... unpin that one and try again */
671 		if (va)
672 			kvmppc_unpin_guest_page(kvm, va, gpa, false);
673 	}
674 
675 	vpap->update_pending = 0;
676 	if (va && nb < vpap->len) {
677 		/*
678 		 * If it's now too short, it must be that userspace
679 		 * has changed the mappings underlying guest memory,
680 		 * so unregister the region.
681 		 */
682 		kvmppc_unpin_guest_page(kvm, va, gpa, false);
683 		va = NULL;
684 	}
685 	if (vpap->pinned_addr)
686 		kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
687 					vpap->dirty);
688 	vpap->gpa = gpa;
689 	vpap->pinned_addr = va;
690 	vpap->dirty = false;
691 	if (va)
692 		vpap->pinned_end = va + vpap->len;
693 }
694 
695 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
696 {
697 	if (!(vcpu->arch.vpa.update_pending ||
698 	      vcpu->arch.slb_shadow.update_pending ||
699 	      vcpu->arch.dtl.update_pending))
700 		return;
701 
702 	spin_lock(&vcpu->arch.vpa_update_lock);
703 	if (vcpu->arch.vpa.update_pending) {
704 		kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
705 		if (vcpu->arch.vpa.pinned_addr)
706 			init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
707 	}
708 	if (vcpu->arch.dtl.update_pending) {
709 		kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
710 		vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
711 		vcpu->arch.dtl_index = 0;
712 	}
713 	if (vcpu->arch.slb_shadow.update_pending)
714 		kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
715 	spin_unlock(&vcpu->arch.vpa_update_lock);
716 }
717 
718 /*
719  * Return the accumulated stolen time for the vcore up until `now'.
720  * The caller should hold the vcore lock.
721  */
722 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
723 {
724 	u64 p;
725 	unsigned long flags;
726 
727 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
728 
729 	spin_lock_irqsave(&vc->stoltb_lock, flags);
730 	p = vc->stolen_tb;
731 	if (vc->vcore_state != VCORE_INACTIVE &&
732 	    vc->preempt_tb != TB_NIL)
733 		p += now - vc->preempt_tb;
734 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
735 	return p;
736 }
737 
738 static void __kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
739 					struct lppaca *vpa,
740 					unsigned int pcpu, u64 now,
741 					unsigned long stolen)
742 {
743 	struct dtl_entry *dt;
744 
745 	dt = vcpu->arch.dtl_ptr;
746 
747 	if (!dt)
748 		return;
749 
750 	dt->dispatch_reason = 7;
751 	dt->preempt_reason = 0;
752 	dt->processor_id = cpu_to_be16(pcpu + vcpu->arch.ptid);
753 	dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
754 	dt->ready_to_enqueue_time = 0;
755 	dt->waiting_to_ready_time = 0;
756 	dt->timebase = cpu_to_be64(now);
757 	dt->fault_addr = 0;
758 	dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
759 	dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
760 
761 	++dt;
762 	if (dt == vcpu->arch.dtl.pinned_end)
763 		dt = vcpu->arch.dtl.pinned_addr;
764 	vcpu->arch.dtl_ptr = dt;
765 	/* order writing *dt vs. writing vpa->dtl_idx */
766 	smp_wmb();
767 	vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
768 
769 	/* vcpu->arch.dtl.dirty is set by the caller */
770 }
771 
772 static void kvmppc_update_vpa_dispatch(struct kvm_vcpu *vcpu,
773 				       struct kvmppc_vcore *vc)
774 {
775 	struct lppaca *vpa;
776 	unsigned long stolen;
777 	unsigned long core_stolen;
778 	u64 now;
779 	unsigned long flags;
780 
781 	vpa = vcpu->arch.vpa.pinned_addr;
782 	if (!vpa)
783 		return;
784 
785 	now = mftb();
786 
787 	core_stolen = vcore_stolen_time(vc, now);
788 	stolen = core_stolen - vcpu->arch.stolen_logged;
789 	vcpu->arch.stolen_logged = core_stolen;
790 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
791 	stolen += vcpu->arch.busy_stolen;
792 	vcpu->arch.busy_stolen = 0;
793 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
794 
795 	vpa->enqueue_dispatch_tb = cpu_to_be64(be64_to_cpu(vpa->enqueue_dispatch_tb) + stolen);
796 
797 	__kvmppc_create_dtl_entry(vcpu, vpa, vc->pcpu, now + vc->tb_offset, stolen);
798 
799 	vcpu->arch.vpa.dirty = true;
800 }
801 
802 static void kvmppc_update_vpa_dispatch_p9(struct kvm_vcpu *vcpu,
803 				       struct kvmppc_vcore *vc,
804 				       u64 now)
805 {
806 	struct lppaca *vpa;
807 	unsigned long stolen;
808 	unsigned long stolen_delta;
809 
810 	vpa = vcpu->arch.vpa.pinned_addr;
811 	if (!vpa)
812 		return;
813 
814 	stolen = vc->stolen_tb;
815 	stolen_delta = stolen - vcpu->arch.stolen_logged;
816 	vcpu->arch.stolen_logged = stolen;
817 
818 	vpa->enqueue_dispatch_tb = cpu_to_be64(stolen);
819 
820 	__kvmppc_create_dtl_entry(vcpu, vpa, vc->pcpu, now, stolen_delta);
821 
822 	vcpu->arch.vpa.dirty = true;
823 }
824 
825 /* See if there is a doorbell interrupt pending for a vcpu */
826 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
827 {
828 	int thr;
829 	struct kvmppc_vcore *vc;
830 
831 	if (vcpu->arch.doorbell_request)
832 		return true;
833 	if (cpu_has_feature(CPU_FTR_ARCH_300))
834 		return false;
835 	/*
836 	 * Ensure that the read of vcore->dpdes comes after the read
837 	 * of vcpu->doorbell_request.  This barrier matches the
838 	 * smp_wmb() in kvmppc_guest_entry_inject().
839 	 */
840 	smp_rmb();
841 	vc = vcpu->arch.vcore;
842 	thr = vcpu->vcpu_id - vc->first_vcpuid;
843 	return !!(vc->dpdes & (1 << thr));
844 }
845 
846 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
847 {
848 	if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
849 		return true;
850 	if ((!vcpu->arch.vcore->arch_compat) &&
851 	    cpu_has_feature(CPU_FTR_ARCH_207S))
852 		return true;
853 	return false;
854 }
855 
856 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
857 			     unsigned long resource, unsigned long value1,
858 			     unsigned long value2)
859 {
860 	switch (resource) {
861 	case H_SET_MODE_RESOURCE_SET_CIABR:
862 		if (!kvmppc_power8_compatible(vcpu))
863 			return H_P2;
864 		if (value2)
865 			return H_P4;
866 		if (mflags)
867 			return H_UNSUPPORTED_FLAG_START;
868 		/* Guests can't breakpoint the hypervisor */
869 		if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
870 			return H_P3;
871 		vcpu->arch.ciabr  = value1;
872 		return H_SUCCESS;
873 	case H_SET_MODE_RESOURCE_SET_DAWR0:
874 		if (!kvmppc_power8_compatible(vcpu))
875 			return H_P2;
876 		if (!ppc_breakpoint_available())
877 			return H_P2;
878 		if (mflags)
879 			return H_UNSUPPORTED_FLAG_START;
880 		if (value2 & DABRX_HYP)
881 			return H_P4;
882 		vcpu->arch.dawr0  = value1;
883 		vcpu->arch.dawrx0 = value2;
884 		return H_SUCCESS;
885 	case H_SET_MODE_RESOURCE_SET_DAWR1:
886 		if (!kvmppc_power8_compatible(vcpu))
887 			return H_P2;
888 		if (!ppc_breakpoint_available())
889 			return H_P2;
890 		if (!cpu_has_feature(CPU_FTR_DAWR1))
891 			return H_P2;
892 		if (!vcpu->kvm->arch.dawr1_enabled)
893 			return H_FUNCTION;
894 		if (mflags)
895 			return H_UNSUPPORTED_FLAG_START;
896 		if (value2 & DABRX_HYP)
897 			return H_P4;
898 		vcpu->arch.dawr1  = value1;
899 		vcpu->arch.dawrx1 = value2;
900 		return H_SUCCESS;
901 	case H_SET_MODE_RESOURCE_ADDR_TRANS_MODE:
902 		/*
903 		 * KVM does not support mflags=2 (AIL=2) and AIL=1 is reserved.
904 		 * Keep this in synch with kvmppc_filter_guest_lpcr_hv.
905 		 */
906 		if (cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG) &&
907 				kvmhv_vcpu_is_radix(vcpu) && mflags == 3)
908 			return H_UNSUPPORTED_FLAG_START;
909 		return H_TOO_HARD;
910 	default:
911 		return H_TOO_HARD;
912 	}
913 }
914 
915 /* Copy guest memory in place - must reside within a single memslot */
916 static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from,
917 				  unsigned long len)
918 {
919 	struct kvm_memory_slot *to_memslot = NULL;
920 	struct kvm_memory_slot *from_memslot = NULL;
921 	unsigned long to_addr, from_addr;
922 	int r;
923 
924 	/* Get HPA for from address */
925 	from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT);
926 	if (!from_memslot)
927 		return -EFAULT;
928 	if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages)
929 			     << PAGE_SHIFT))
930 		return -EINVAL;
931 	from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT);
932 	if (kvm_is_error_hva(from_addr))
933 		return -EFAULT;
934 	from_addr |= (from & (PAGE_SIZE - 1));
935 
936 	/* Get HPA for to address */
937 	to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT);
938 	if (!to_memslot)
939 		return -EFAULT;
940 	if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages)
941 			   << PAGE_SHIFT))
942 		return -EINVAL;
943 	to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT);
944 	if (kvm_is_error_hva(to_addr))
945 		return -EFAULT;
946 	to_addr |= (to & (PAGE_SIZE - 1));
947 
948 	/* Perform copy */
949 	r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr,
950 			     len);
951 	if (r)
952 		return -EFAULT;
953 	mark_page_dirty(kvm, to >> PAGE_SHIFT);
954 	return 0;
955 }
956 
957 static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags,
958 			       unsigned long dest, unsigned long src)
959 {
960 	u64 pg_sz = SZ_4K;		/* 4K page size */
961 	u64 pg_mask = SZ_4K - 1;
962 	int ret;
963 
964 	/* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */
965 	if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE |
966 		      H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED))
967 		return H_PARAMETER;
968 
969 	/* dest (and src if copy_page flag set) must be page aligned */
970 	if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask)))
971 		return H_PARAMETER;
972 
973 	/* zero and/or copy the page as determined by the flags */
974 	if (flags & H_COPY_PAGE) {
975 		ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz);
976 		if (ret < 0)
977 			return H_PARAMETER;
978 	} else if (flags & H_ZERO_PAGE) {
979 		ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz);
980 		if (ret < 0)
981 			return H_PARAMETER;
982 	}
983 
984 	/* We can ignore the remaining flags */
985 
986 	return H_SUCCESS;
987 }
988 
989 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
990 {
991 	struct kvmppc_vcore *vcore = target->arch.vcore;
992 
993 	/*
994 	 * We expect to have been called by the real mode handler
995 	 * (kvmppc_rm_h_confer()) which would have directly returned
996 	 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
997 	 * have useful work to do and should not confer) so we don't
998 	 * recheck that here.
999 	 *
1000 	 * In the case of the P9 single vcpu per vcore case, the real
1001 	 * mode handler is not called but no other threads are in the
1002 	 * source vcore.
1003 	 */
1004 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
1005 		spin_lock(&vcore->lock);
1006 		if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
1007 		    vcore->vcore_state != VCORE_INACTIVE &&
1008 		    vcore->runner)
1009 			target = vcore->runner;
1010 		spin_unlock(&vcore->lock);
1011 	}
1012 
1013 	return kvm_vcpu_yield_to(target);
1014 }
1015 
1016 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
1017 {
1018 	int yield_count = 0;
1019 	struct lppaca *lppaca;
1020 
1021 	spin_lock(&vcpu->arch.vpa_update_lock);
1022 	lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
1023 	if (lppaca)
1024 		yield_count = be32_to_cpu(lppaca->yield_count);
1025 	spin_unlock(&vcpu->arch.vpa_update_lock);
1026 	return yield_count;
1027 }
1028 
1029 /*
1030  * H_RPT_INVALIDATE hcall handler for nested guests.
1031  *
1032  * Handles only nested process-scoped invalidation requests in L0.
1033  */
1034 static int kvmppc_nested_h_rpt_invalidate(struct kvm_vcpu *vcpu)
1035 {
1036 	unsigned long type = kvmppc_get_gpr(vcpu, 6);
1037 	unsigned long pid, pg_sizes, start, end;
1038 
1039 	/*
1040 	 * The partition-scoped invalidations aren't handled here in L0.
1041 	 */
1042 	if (type & H_RPTI_TYPE_NESTED)
1043 		return RESUME_HOST;
1044 
1045 	pid = kvmppc_get_gpr(vcpu, 4);
1046 	pg_sizes = kvmppc_get_gpr(vcpu, 7);
1047 	start = kvmppc_get_gpr(vcpu, 8);
1048 	end = kvmppc_get_gpr(vcpu, 9);
1049 
1050 	do_h_rpt_invalidate_prt(pid, vcpu->arch.nested->shadow_lpid,
1051 				type, pg_sizes, start, end);
1052 
1053 	kvmppc_set_gpr(vcpu, 3, H_SUCCESS);
1054 	return RESUME_GUEST;
1055 }
1056 
1057 static long kvmppc_h_rpt_invalidate(struct kvm_vcpu *vcpu,
1058 				    unsigned long id, unsigned long target,
1059 				    unsigned long type, unsigned long pg_sizes,
1060 				    unsigned long start, unsigned long end)
1061 {
1062 	if (!kvm_is_radix(vcpu->kvm))
1063 		return H_UNSUPPORTED;
1064 
1065 	if (end < start)
1066 		return H_P5;
1067 
1068 	/*
1069 	 * Partition-scoped invalidation for nested guests.
1070 	 */
1071 	if (type & H_RPTI_TYPE_NESTED) {
1072 		if (!nesting_enabled(vcpu->kvm))
1073 			return H_FUNCTION;
1074 
1075 		/* Support only cores as target */
1076 		if (target != H_RPTI_TARGET_CMMU)
1077 			return H_P2;
1078 
1079 		return do_h_rpt_invalidate_pat(vcpu, id, type, pg_sizes,
1080 					       start, end);
1081 	}
1082 
1083 	/*
1084 	 * Process-scoped invalidation for L1 guests.
1085 	 */
1086 	do_h_rpt_invalidate_prt(id, vcpu->kvm->arch.lpid,
1087 				type, pg_sizes, start, end);
1088 	return H_SUCCESS;
1089 }
1090 
1091 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
1092 {
1093 	struct kvm *kvm = vcpu->kvm;
1094 	unsigned long req = kvmppc_get_gpr(vcpu, 3);
1095 	unsigned long target, ret = H_SUCCESS;
1096 	int yield_count;
1097 	struct kvm_vcpu *tvcpu;
1098 	int idx, rc;
1099 
1100 	if (req <= MAX_HCALL_OPCODE &&
1101 	    !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
1102 		return RESUME_HOST;
1103 
1104 	switch (req) {
1105 	case H_REMOVE:
1106 		ret = kvmppc_h_remove(vcpu, kvmppc_get_gpr(vcpu, 4),
1107 					kvmppc_get_gpr(vcpu, 5),
1108 					kvmppc_get_gpr(vcpu, 6));
1109 		if (ret == H_TOO_HARD)
1110 			return RESUME_HOST;
1111 		break;
1112 	case H_ENTER:
1113 		ret = kvmppc_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
1114 					kvmppc_get_gpr(vcpu, 5),
1115 					kvmppc_get_gpr(vcpu, 6),
1116 					kvmppc_get_gpr(vcpu, 7));
1117 		if (ret == H_TOO_HARD)
1118 			return RESUME_HOST;
1119 		break;
1120 	case H_READ:
1121 		ret = kvmppc_h_read(vcpu, kvmppc_get_gpr(vcpu, 4),
1122 					kvmppc_get_gpr(vcpu, 5));
1123 		if (ret == H_TOO_HARD)
1124 			return RESUME_HOST;
1125 		break;
1126 	case H_CLEAR_MOD:
1127 		ret = kvmppc_h_clear_mod(vcpu, kvmppc_get_gpr(vcpu, 4),
1128 					kvmppc_get_gpr(vcpu, 5));
1129 		if (ret == H_TOO_HARD)
1130 			return RESUME_HOST;
1131 		break;
1132 	case H_CLEAR_REF:
1133 		ret = kvmppc_h_clear_ref(vcpu, kvmppc_get_gpr(vcpu, 4),
1134 					kvmppc_get_gpr(vcpu, 5));
1135 		if (ret == H_TOO_HARD)
1136 			return RESUME_HOST;
1137 		break;
1138 	case H_PROTECT:
1139 		ret = kvmppc_h_protect(vcpu, kvmppc_get_gpr(vcpu, 4),
1140 					kvmppc_get_gpr(vcpu, 5),
1141 					kvmppc_get_gpr(vcpu, 6));
1142 		if (ret == H_TOO_HARD)
1143 			return RESUME_HOST;
1144 		break;
1145 	case H_BULK_REMOVE:
1146 		ret = kvmppc_h_bulk_remove(vcpu);
1147 		if (ret == H_TOO_HARD)
1148 			return RESUME_HOST;
1149 		break;
1150 
1151 	case H_CEDE:
1152 		break;
1153 	case H_PROD:
1154 		target = kvmppc_get_gpr(vcpu, 4);
1155 		tvcpu = kvmppc_find_vcpu(kvm, target);
1156 		if (!tvcpu) {
1157 			ret = H_PARAMETER;
1158 			break;
1159 		}
1160 		tvcpu->arch.prodded = 1;
1161 		smp_mb(); /* This orders prodded store vs ceded load */
1162 		if (tvcpu->arch.ceded)
1163 			kvmppc_fast_vcpu_kick_hv(tvcpu);
1164 		break;
1165 	case H_CONFER:
1166 		target = kvmppc_get_gpr(vcpu, 4);
1167 		if (target == -1)
1168 			break;
1169 		tvcpu = kvmppc_find_vcpu(kvm, target);
1170 		if (!tvcpu) {
1171 			ret = H_PARAMETER;
1172 			break;
1173 		}
1174 		yield_count = kvmppc_get_gpr(vcpu, 5);
1175 		if (kvmppc_get_yield_count(tvcpu) != yield_count)
1176 			break;
1177 		kvm_arch_vcpu_yield_to(tvcpu);
1178 		break;
1179 	case H_REGISTER_VPA:
1180 		ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
1181 					kvmppc_get_gpr(vcpu, 5),
1182 					kvmppc_get_gpr(vcpu, 6));
1183 		break;
1184 	case H_RTAS:
1185 		if (list_empty(&kvm->arch.rtas_tokens))
1186 			return RESUME_HOST;
1187 
1188 		idx = srcu_read_lock(&kvm->srcu);
1189 		rc = kvmppc_rtas_hcall(vcpu);
1190 		srcu_read_unlock(&kvm->srcu, idx);
1191 
1192 		if (rc == -ENOENT)
1193 			return RESUME_HOST;
1194 		else if (rc == 0)
1195 			break;
1196 
1197 		/* Send the error out to userspace via KVM_RUN */
1198 		return rc;
1199 	case H_LOGICAL_CI_LOAD:
1200 		ret = kvmppc_h_logical_ci_load(vcpu);
1201 		if (ret == H_TOO_HARD)
1202 			return RESUME_HOST;
1203 		break;
1204 	case H_LOGICAL_CI_STORE:
1205 		ret = kvmppc_h_logical_ci_store(vcpu);
1206 		if (ret == H_TOO_HARD)
1207 			return RESUME_HOST;
1208 		break;
1209 	case H_SET_MODE:
1210 		ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
1211 					kvmppc_get_gpr(vcpu, 5),
1212 					kvmppc_get_gpr(vcpu, 6),
1213 					kvmppc_get_gpr(vcpu, 7));
1214 		if (ret == H_TOO_HARD)
1215 			return RESUME_HOST;
1216 		break;
1217 	case H_XIRR:
1218 	case H_CPPR:
1219 	case H_EOI:
1220 	case H_IPI:
1221 	case H_IPOLL:
1222 	case H_XIRR_X:
1223 		if (kvmppc_xics_enabled(vcpu)) {
1224 			if (xics_on_xive()) {
1225 				ret = H_NOT_AVAILABLE;
1226 				return RESUME_GUEST;
1227 			}
1228 			ret = kvmppc_xics_hcall(vcpu, req);
1229 			break;
1230 		}
1231 		return RESUME_HOST;
1232 	case H_SET_DABR:
1233 		ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
1234 		break;
1235 	case H_SET_XDABR:
1236 		ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
1237 						kvmppc_get_gpr(vcpu, 5));
1238 		break;
1239 #ifdef CONFIG_SPAPR_TCE_IOMMU
1240 	case H_GET_TCE:
1241 		ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1242 						kvmppc_get_gpr(vcpu, 5));
1243 		if (ret == H_TOO_HARD)
1244 			return RESUME_HOST;
1245 		break;
1246 	case H_PUT_TCE:
1247 		ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1248 						kvmppc_get_gpr(vcpu, 5),
1249 						kvmppc_get_gpr(vcpu, 6));
1250 		if (ret == H_TOO_HARD)
1251 			return RESUME_HOST;
1252 		break;
1253 	case H_PUT_TCE_INDIRECT:
1254 		ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
1255 						kvmppc_get_gpr(vcpu, 5),
1256 						kvmppc_get_gpr(vcpu, 6),
1257 						kvmppc_get_gpr(vcpu, 7));
1258 		if (ret == H_TOO_HARD)
1259 			return RESUME_HOST;
1260 		break;
1261 	case H_STUFF_TCE:
1262 		ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1263 						kvmppc_get_gpr(vcpu, 5),
1264 						kvmppc_get_gpr(vcpu, 6),
1265 						kvmppc_get_gpr(vcpu, 7));
1266 		if (ret == H_TOO_HARD)
1267 			return RESUME_HOST;
1268 		break;
1269 #endif
1270 	case H_RANDOM:
1271 		if (!arch_get_random_seed_longs(&vcpu->arch.regs.gpr[4], 1))
1272 			ret = H_HARDWARE;
1273 		break;
1274 	case H_RPT_INVALIDATE:
1275 		ret = kvmppc_h_rpt_invalidate(vcpu, kvmppc_get_gpr(vcpu, 4),
1276 					      kvmppc_get_gpr(vcpu, 5),
1277 					      kvmppc_get_gpr(vcpu, 6),
1278 					      kvmppc_get_gpr(vcpu, 7),
1279 					      kvmppc_get_gpr(vcpu, 8),
1280 					      kvmppc_get_gpr(vcpu, 9));
1281 		break;
1282 
1283 	case H_SET_PARTITION_TABLE:
1284 		ret = H_FUNCTION;
1285 		if (nesting_enabled(kvm))
1286 			ret = kvmhv_set_partition_table(vcpu);
1287 		break;
1288 	case H_ENTER_NESTED:
1289 		ret = H_FUNCTION;
1290 		if (!nesting_enabled(kvm))
1291 			break;
1292 		ret = kvmhv_enter_nested_guest(vcpu);
1293 		if (ret == H_INTERRUPT) {
1294 			kvmppc_set_gpr(vcpu, 3, 0);
1295 			vcpu->arch.hcall_needed = 0;
1296 			return -EINTR;
1297 		} else if (ret == H_TOO_HARD) {
1298 			kvmppc_set_gpr(vcpu, 3, 0);
1299 			vcpu->arch.hcall_needed = 0;
1300 			return RESUME_HOST;
1301 		}
1302 		break;
1303 	case H_TLB_INVALIDATE:
1304 		ret = H_FUNCTION;
1305 		if (nesting_enabled(kvm))
1306 			ret = kvmhv_do_nested_tlbie(vcpu);
1307 		break;
1308 	case H_COPY_TOFROM_GUEST:
1309 		ret = H_FUNCTION;
1310 		if (nesting_enabled(kvm))
1311 			ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1312 		break;
1313 	case H_PAGE_INIT:
1314 		ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4),
1315 					 kvmppc_get_gpr(vcpu, 5),
1316 					 kvmppc_get_gpr(vcpu, 6));
1317 		break;
1318 	case H_SVM_PAGE_IN:
1319 		ret = H_UNSUPPORTED;
1320 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1321 			ret = kvmppc_h_svm_page_in(kvm,
1322 						   kvmppc_get_gpr(vcpu, 4),
1323 						   kvmppc_get_gpr(vcpu, 5),
1324 						   kvmppc_get_gpr(vcpu, 6));
1325 		break;
1326 	case H_SVM_PAGE_OUT:
1327 		ret = H_UNSUPPORTED;
1328 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1329 			ret = kvmppc_h_svm_page_out(kvm,
1330 						    kvmppc_get_gpr(vcpu, 4),
1331 						    kvmppc_get_gpr(vcpu, 5),
1332 						    kvmppc_get_gpr(vcpu, 6));
1333 		break;
1334 	case H_SVM_INIT_START:
1335 		ret = H_UNSUPPORTED;
1336 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1337 			ret = kvmppc_h_svm_init_start(kvm);
1338 		break;
1339 	case H_SVM_INIT_DONE:
1340 		ret = H_UNSUPPORTED;
1341 		if (kvmppc_get_srr1(vcpu) & MSR_S)
1342 			ret = kvmppc_h_svm_init_done(kvm);
1343 		break;
1344 	case H_SVM_INIT_ABORT:
1345 		/*
1346 		 * Even if that call is made by the Ultravisor, the SSR1 value
1347 		 * is the guest context one, with the secure bit clear as it has
1348 		 * not yet been secured. So we can't check it here.
1349 		 * Instead the kvm->arch.secure_guest flag is checked inside
1350 		 * kvmppc_h_svm_init_abort().
1351 		 */
1352 		ret = kvmppc_h_svm_init_abort(kvm);
1353 		break;
1354 
1355 	default:
1356 		return RESUME_HOST;
1357 	}
1358 	WARN_ON_ONCE(ret == H_TOO_HARD);
1359 	kvmppc_set_gpr(vcpu, 3, ret);
1360 	vcpu->arch.hcall_needed = 0;
1361 	return RESUME_GUEST;
1362 }
1363 
1364 /*
1365  * Handle H_CEDE in the P9 path where we don't call the real-mode hcall
1366  * handlers in book3s_hv_rmhandlers.S.
1367  *
1368  * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1369  * that the cede logic in kvmppc_run_single_vcpu() works properly.
1370  */
1371 static void kvmppc_cede(struct kvm_vcpu *vcpu)
1372 {
1373 	vcpu->arch.shregs.msr |= MSR_EE;
1374 	vcpu->arch.ceded = 1;
1375 	smp_mb();
1376 	if (vcpu->arch.prodded) {
1377 		vcpu->arch.prodded = 0;
1378 		smp_mb();
1379 		vcpu->arch.ceded = 0;
1380 	}
1381 }
1382 
1383 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1384 {
1385 	switch (cmd) {
1386 	case H_CEDE:
1387 	case H_PROD:
1388 	case H_CONFER:
1389 	case H_REGISTER_VPA:
1390 	case H_SET_MODE:
1391 #ifdef CONFIG_SPAPR_TCE_IOMMU
1392 	case H_GET_TCE:
1393 	case H_PUT_TCE:
1394 	case H_PUT_TCE_INDIRECT:
1395 	case H_STUFF_TCE:
1396 #endif
1397 	case H_LOGICAL_CI_LOAD:
1398 	case H_LOGICAL_CI_STORE:
1399 #ifdef CONFIG_KVM_XICS
1400 	case H_XIRR:
1401 	case H_CPPR:
1402 	case H_EOI:
1403 	case H_IPI:
1404 	case H_IPOLL:
1405 	case H_XIRR_X:
1406 #endif
1407 	case H_PAGE_INIT:
1408 	case H_RPT_INVALIDATE:
1409 		return 1;
1410 	}
1411 
1412 	/* See if it's in the real-mode table */
1413 	return kvmppc_hcall_impl_hv_realmode(cmd);
1414 }
1415 
1416 static int kvmppc_emulate_debug_inst(struct kvm_vcpu *vcpu)
1417 {
1418 	ppc_inst_t last_inst;
1419 
1420 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1421 					EMULATE_DONE) {
1422 		/*
1423 		 * Fetch failed, so return to guest and
1424 		 * try executing it again.
1425 		 */
1426 		return RESUME_GUEST;
1427 	}
1428 
1429 	if (ppc_inst_val(last_inst) == KVMPPC_INST_SW_BREAKPOINT) {
1430 		vcpu->run->exit_reason = KVM_EXIT_DEBUG;
1431 		vcpu->run->debug.arch.address = kvmppc_get_pc(vcpu);
1432 		return RESUME_HOST;
1433 	} else {
1434 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL |
1435 				(kvmppc_get_msr(vcpu) & SRR1_PREFIXED));
1436 		return RESUME_GUEST;
1437 	}
1438 }
1439 
1440 static void do_nothing(void *x)
1441 {
1442 }
1443 
1444 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1445 {
1446 	int thr, cpu, pcpu, nthreads;
1447 	struct kvm_vcpu *v;
1448 	unsigned long dpdes;
1449 
1450 	nthreads = vcpu->kvm->arch.emul_smt_mode;
1451 	dpdes = 0;
1452 	cpu = vcpu->vcpu_id & ~(nthreads - 1);
1453 	for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1454 		v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1455 		if (!v)
1456 			continue;
1457 		/*
1458 		 * If the vcpu is currently running on a physical cpu thread,
1459 		 * interrupt it in order to pull it out of the guest briefly,
1460 		 * which will update its vcore->dpdes value.
1461 		 */
1462 		pcpu = READ_ONCE(v->cpu);
1463 		if (pcpu >= 0)
1464 			smp_call_function_single(pcpu, do_nothing, NULL, 1);
1465 		if (kvmppc_doorbell_pending(v))
1466 			dpdes |= 1 << thr;
1467 	}
1468 	return dpdes;
1469 }
1470 
1471 /*
1472  * On POWER9, emulate doorbell-related instructions in order to
1473  * give the guest the illusion of running on a multi-threaded core.
1474  * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1475  * and mfspr DPDES.
1476  */
1477 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1478 {
1479 	u32 inst, rb, thr;
1480 	unsigned long arg;
1481 	struct kvm *kvm = vcpu->kvm;
1482 	struct kvm_vcpu *tvcpu;
1483 	ppc_inst_t pinst;
1484 
1485 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &pinst) != EMULATE_DONE)
1486 		return RESUME_GUEST;
1487 	inst = ppc_inst_val(pinst);
1488 	if (get_op(inst) != 31)
1489 		return EMULATE_FAIL;
1490 	rb = get_rb(inst);
1491 	thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1492 	switch (get_xop(inst)) {
1493 	case OP_31_XOP_MSGSNDP:
1494 		arg = kvmppc_get_gpr(vcpu, rb);
1495 		if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1496 			break;
1497 		arg &= 0x7f;
1498 		if (arg >= kvm->arch.emul_smt_mode)
1499 			break;
1500 		tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1501 		if (!tvcpu)
1502 			break;
1503 		if (!tvcpu->arch.doorbell_request) {
1504 			tvcpu->arch.doorbell_request = 1;
1505 			kvmppc_fast_vcpu_kick_hv(tvcpu);
1506 		}
1507 		break;
1508 	case OP_31_XOP_MSGCLRP:
1509 		arg = kvmppc_get_gpr(vcpu, rb);
1510 		if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER)
1511 			break;
1512 		vcpu->arch.vcore->dpdes = 0;
1513 		vcpu->arch.doorbell_request = 0;
1514 		break;
1515 	case OP_31_XOP_MFSPR:
1516 		switch (get_sprn(inst)) {
1517 		case SPRN_TIR:
1518 			arg = thr;
1519 			break;
1520 		case SPRN_DPDES:
1521 			arg = kvmppc_read_dpdes(vcpu);
1522 			break;
1523 		default:
1524 			return EMULATE_FAIL;
1525 		}
1526 		kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1527 		break;
1528 	default:
1529 		return EMULATE_FAIL;
1530 	}
1531 	kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1532 	return RESUME_GUEST;
1533 }
1534 
1535 /*
1536  * If the lppaca had pmcregs_in_use clear when we exited the guest, then
1537  * HFSCR_PM is cleared for next entry. If the guest then tries to access
1538  * the PMU SPRs, we get this facility unavailable interrupt. Putting HFSCR_PM
1539  * back in the guest HFSCR will cause the next entry to load the PMU SPRs and
1540  * allow the guest access to continue.
1541  */
1542 static int kvmppc_pmu_unavailable(struct kvm_vcpu *vcpu)
1543 {
1544 	if (!(vcpu->arch.hfscr_permitted & HFSCR_PM))
1545 		return EMULATE_FAIL;
1546 
1547 	vcpu->arch.hfscr |= HFSCR_PM;
1548 
1549 	return RESUME_GUEST;
1550 }
1551 
1552 static int kvmppc_ebb_unavailable(struct kvm_vcpu *vcpu)
1553 {
1554 	if (!(vcpu->arch.hfscr_permitted & HFSCR_EBB))
1555 		return EMULATE_FAIL;
1556 
1557 	vcpu->arch.hfscr |= HFSCR_EBB;
1558 
1559 	return RESUME_GUEST;
1560 }
1561 
1562 static int kvmppc_tm_unavailable(struct kvm_vcpu *vcpu)
1563 {
1564 	if (!(vcpu->arch.hfscr_permitted & HFSCR_TM))
1565 		return EMULATE_FAIL;
1566 
1567 	vcpu->arch.hfscr |= HFSCR_TM;
1568 
1569 	return RESUME_GUEST;
1570 }
1571 
1572 static int kvmppc_handle_exit_hv(struct kvm_vcpu *vcpu,
1573 				 struct task_struct *tsk)
1574 {
1575 	struct kvm_run *run = vcpu->run;
1576 	int r = RESUME_HOST;
1577 
1578 	vcpu->stat.sum_exits++;
1579 
1580 	/*
1581 	 * This can happen if an interrupt occurs in the last stages
1582 	 * of guest entry or the first stages of guest exit (i.e. after
1583 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1584 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1585 	 * That can happen due to a bug, or due to a machine check
1586 	 * occurring at just the wrong time.
1587 	 */
1588 	if (vcpu->arch.shregs.msr & MSR_HV) {
1589 		printk(KERN_EMERG "KVM trap in HV mode!\n");
1590 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1591 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1592 			vcpu->arch.shregs.msr);
1593 		kvmppc_dump_regs(vcpu);
1594 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1595 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1596 		return RESUME_HOST;
1597 	}
1598 	run->exit_reason = KVM_EXIT_UNKNOWN;
1599 	run->ready_for_interrupt_injection = 1;
1600 	switch (vcpu->arch.trap) {
1601 	/* We're good on these - the host merely wanted to get our attention */
1602 	case BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER:
1603 		WARN_ON_ONCE(1); /* Should never happen */
1604 		vcpu->arch.trap = BOOK3S_INTERRUPT_HV_DECREMENTER;
1605 		fallthrough;
1606 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1607 		vcpu->stat.dec_exits++;
1608 		r = RESUME_GUEST;
1609 		break;
1610 	case BOOK3S_INTERRUPT_EXTERNAL:
1611 	case BOOK3S_INTERRUPT_H_DOORBELL:
1612 	case BOOK3S_INTERRUPT_H_VIRT:
1613 		vcpu->stat.ext_intr_exits++;
1614 		r = RESUME_GUEST;
1615 		break;
1616 	/* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1617 	case BOOK3S_INTERRUPT_HMI:
1618 	case BOOK3S_INTERRUPT_PERFMON:
1619 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
1620 		r = RESUME_GUEST;
1621 		break;
1622 	case BOOK3S_INTERRUPT_MACHINE_CHECK: {
1623 		static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
1624 					      DEFAULT_RATELIMIT_BURST);
1625 		/*
1626 		 * Print the MCE event to host console. Ratelimit so the guest
1627 		 * can't flood the host log.
1628 		 */
1629 		if (__ratelimit(&rs))
1630 			machine_check_print_event_info(&vcpu->arch.mce_evt,false, true);
1631 
1632 		/*
1633 		 * If the guest can do FWNMI, exit to userspace so it can
1634 		 * deliver a FWNMI to the guest.
1635 		 * Otherwise we synthesize a machine check for the guest
1636 		 * so that it knows that the machine check occurred.
1637 		 */
1638 		if (!vcpu->kvm->arch.fwnmi_enabled) {
1639 			ulong flags = (vcpu->arch.shregs.msr & 0x083c0000) |
1640 					(kvmppc_get_msr(vcpu) & SRR1_PREFIXED);
1641 			kvmppc_core_queue_machine_check(vcpu, flags);
1642 			r = RESUME_GUEST;
1643 			break;
1644 		}
1645 
1646 		/* Exit to guest with KVM_EXIT_NMI as exit reason */
1647 		run->exit_reason = KVM_EXIT_NMI;
1648 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1649 		/* Clear out the old NMI status from run->flags */
1650 		run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1651 		/* Now set the NMI status */
1652 		if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1653 			run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1654 		else
1655 			run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1656 
1657 		r = RESUME_HOST;
1658 		break;
1659 	}
1660 	case BOOK3S_INTERRUPT_PROGRAM:
1661 	{
1662 		ulong flags;
1663 		/*
1664 		 * Normally program interrupts are delivered directly
1665 		 * to the guest by the hardware, but we can get here
1666 		 * as a result of a hypervisor emulation interrupt
1667 		 * (e40) getting turned into a 700 by BML RTAS.
1668 		 */
1669 		flags = (vcpu->arch.shregs.msr & 0x1f0000ull) |
1670 			(kvmppc_get_msr(vcpu) & SRR1_PREFIXED);
1671 		kvmppc_core_queue_program(vcpu, flags);
1672 		r = RESUME_GUEST;
1673 		break;
1674 	}
1675 	case BOOK3S_INTERRUPT_SYSCALL:
1676 	{
1677 		int i;
1678 
1679 		if (unlikely(vcpu->arch.shregs.msr & MSR_PR)) {
1680 			/*
1681 			 * Guest userspace executed sc 1. This can only be
1682 			 * reached by the P9 path because the old path
1683 			 * handles this case in realmode hcall handlers.
1684 			 */
1685 			if (!kvmhv_vcpu_is_radix(vcpu)) {
1686 				/*
1687 				 * A guest could be running PR KVM, so this
1688 				 * may be a PR KVM hcall. It must be reflected
1689 				 * to the guest kernel as a sc interrupt.
1690 				 */
1691 				kvmppc_core_queue_syscall(vcpu);
1692 			} else {
1693 				/*
1694 				 * Radix guests can not run PR KVM or nested HV
1695 				 * hash guests which might run PR KVM, so this
1696 				 * is always a privilege fault. Send a program
1697 				 * check to guest kernel.
1698 				 */
1699 				kvmppc_core_queue_program(vcpu, SRR1_PROGPRIV);
1700 			}
1701 			r = RESUME_GUEST;
1702 			break;
1703 		}
1704 
1705 		/*
1706 		 * hcall - gather args and set exit_reason. This will next be
1707 		 * handled by kvmppc_pseries_do_hcall which may be able to deal
1708 		 * with it and resume guest, or may punt to userspace.
1709 		 */
1710 		run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1711 		for (i = 0; i < 9; ++i)
1712 			run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1713 		run->exit_reason = KVM_EXIT_PAPR_HCALL;
1714 		vcpu->arch.hcall_needed = 1;
1715 		r = RESUME_HOST;
1716 		break;
1717 	}
1718 	/*
1719 	 * We get these next two if the guest accesses a page which it thinks
1720 	 * it has mapped but which is not actually present, either because
1721 	 * it is for an emulated I/O device or because the corresonding
1722 	 * host page has been paged out.
1723 	 *
1724 	 * Any other HDSI/HISI interrupts have been handled already for P7/8
1725 	 * guests. For POWER9 hash guests not using rmhandlers, basic hash
1726 	 * fault handling is done here.
1727 	 */
1728 	case BOOK3S_INTERRUPT_H_DATA_STORAGE: {
1729 		unsigned long vsid;
1730 		long err;
1731 
1732 		if (cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG) &&
1733 		    unlikely(vcpu->arch.fault_dsisr == HDSISR_CANARY)) {
1734 			r = RESUME_GUEST; /* Just retry if it's the canary */
1735 			break;
1736 		}
1737 
1738 		if (kvm_is_radix(vcpu->kvm) || !cpu_has_feature(CPU_FTR_ARCH_300)) {
1739 			/*
1740 			 * Radix doesn't require anything, and pre-ISAv3.0 hash
1741 			 * already attempted to handle this in rmhandlers. The
1742 			 * hash fault handling below is v3 only (it uses ASDR
1743 			 * via fault_gpa).
1744 			 */
1745 			r = RESUME_PAGE_FAULT;
1746 			break;
1747 		}
1748 
1749 		if (!(vcpu->arch.fault_dsisr & (DSISR_NOHPTE | DSISR_PROTFAULT))) {
1750 			kvmppc_core_queue_data_storage(vcpu,
1751 				kvmppc_get_msr(vcpu) & SRR1_PREFIXED,
1752 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1753 			r = RESUME_GUEST;
1754 			break;
1755 		}
1756 
1757 		if (!(vcpu->arch.shregs.msr & MSR_DR))
1758 			vsid = vcpu->kvm->arch.vrma_slb_v;
1759 		else
1760 			vsid = vcpu->arch.fault_gpa;
1761 
1762 		err = kvmppc_hpte_hv_fault(vcpu, vcpu->arch.fault_dar,
1763 				vsid, vcpu->arch.fault_dsisr, true);
1764 		if (err == 0) {
1765 			r = RESUME_GUEST;
1766 		} else if (err == -1 || err == -2) {
1767 			r = RESUME_PAGE_FAULT;
1768 		} else {
1769 			kvmppc_core_queue_data_storage(vcpu,
1770 				kvmppc_get_msr(vcpu) & SRR1_PREFIXED,
1771 				vcpu->arch.fault_dar, err);
1772 			r = RESUME_GUEST;
1773 		}
1774 		break;
1775 	}
1776 	case BOOK3S_INTERRUPT_H_INST_STORAGE: {
1777 		unsigned long vsid;
1778 		long err;
1779 
1780 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1781 		vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr &
1782 			DSISR_SRR1_MATCH_64S;
1783 		if (kvm_is_radix(vcpu->kvm) || !cpu_has_feature(CPU_FTR_ARCH_300)) {
1784 			/*
1785 			 * Radix doesn't require anything, and pre-ISAv3.0 hash
1786 			 * already attempted to handle this in rmhandlers. The
1787 			 * hash fault handling below is v3 only (it uses ASDR
1788 			 * via fault_gpa).
1789 			 */
1790 			if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1791 				vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1792 			r = RESUME_PAGE_FAULT;
1793 			break;
1794 		}
1795 
1796 		if (!(vcpu->arch.fault_dsisr & SRR1_ISI_NOPT)) {
1797 			kvmppc_core_queue_inst_storage(vcpu,
1798 				vcpu->arch.fault_dsisr |
1799 				(kvmppc_get_msr(vcpu) & SRR1_PREFIXED));
1800 			r = RESUME_GUEST;
1801 			break;
1802 		}
1803 
1804 		if (!(vcpu->arch.shregs.msr & MSR_IR))
1805 			vsid = vcpu->kvm->arch.vrma_slb_v;
1806 		else
1807 			vsid = vcpu->arch.fault_gpa;
1808 
1809 		err = kvmppc_hpte_hv_fault(vcpu, vcpu->arch.fault_dar,
1810 				vsid, vcpu->arch.fault_dsisr, false);
1811 		if (err == 0) {
1812 			r = RESUME_GUEST;
1813 		} else if (err == -1) {
1814 			r = RESUME_PAGE_FAULT;
1815 		} else {
1816 			kvmppc_core_queue_inst_storage(vcpu,
1817 				err | (kvmppc_get_msr(vcpu) & SRR1_PREFIXED));
1818 			r = RESUME_GUEST;
1819 		}
1820 		break;
1821 	}
1822 
1823 	/*
1824 	 * This occurs if the guest executes an illegal instruction.
1825 	 * If the guest debug is disabled, generate a program interrupt
1826 	 * to the guest. If guest debug is enabled, we need to check
1827 	 * whether the instruction is a software breakpoint instruction.
1828 	 * Accordingly return to Guest or Host.
1829 	 */
1830 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1831 		if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1832 			vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1833 				swab32(vcpu->arch.emul_inst) :
1834 				vcpu->arch.emul_inst;
1835 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1836 			r = kvmppc_emulate_debug_inst(vcpu);
1837 		} else {
1838 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL |
1839 				(kvmppc_get_msr(vcpu) & SRR1_PREFIXED));
1840 			r = RESUME_GUEST;
1841 		}
1842 		break;
1843 
1844 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1845 	case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1846 		/*
1847 		 * This occurs for various TM-related instructions that
1848 		 * we need to emulate on POWER9 DD2.2.  We have already
1849 		 * handled the cases where the guest was in real-suspend
1850 		 * mode and was transitioning to transactional state.
1851 		 */
1852 		r = kvmhv_p9_tm_emulation(vcpu);
1853 		if (r != -1)
1854 			break;
1855 		fallthrough; /* go to facility unavailable handler */
1856 #endif
1857 
1858 	/*
1859 	 * This occurs if the guest (kernel or userspace), does something that
1860 	 * is prohibited by HFSCR.
1861 	 * On POWER9, this could be a doorbell instruction that we need
1862 	 * to emulate.
1863 	 * Otherwise, we just generate a program interrupt to the guest.
1864 	 */
1865 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL: {
1866 		u64 cause = vcpu->arch.hfscr >> 56;
1867 
1868 		r = EMULATE_FAIL;
1869 		if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1870 			if (cause == FSCR_MSGP_LG)
1871 				r = kvmppc_emulate_doorbell_instr(vcpu);
1872 			if (cause == FSCR_PM_LG)
1873 				r = kvmppc_pmu_unavailable(vcpu);
1874 			if (cause == FSCR_EBB_LG)
1875 				r = kvmppc_ebb_unavailable(vcpu);
1876 			if (cause == FSCR_TM_LG)
1877 				r = kvmppc_tm_unavailable(vcpu);
1878 		}
1879 		if (r == EMULATE_FAIL) {
1880 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL |
1881 				(kvmppc_get_msr(vcpu) & SRR1_PREFIXED));
1882 			r = RESUME_GUEST;
1883 		}
1884 		break;
1885 	}
1886 
1887 	case BOOK3S_INTERRUPT_HV_RM_HARD:
1888 		r = RESUME_PASSTHROUGH;
1889 		break;
1890 	default:
1891 		kvmppc_dump_regs(vcpu);
1892 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1893 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1894 			vcpu->arch.shregs.msr);
1895 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1896 		r = RESUME_HOST;
1897 		break;
1898 	}
1899 
1900 	return r;
1901 }
1902 
1903 static int kvmppc_handle_nested_exit(struct kvm_vcpu *vcpu)
1904 {
1905 	int r;
1906 	int srcu_idx;
1907 
1908 	vcpu->stat.sum_exits++;
1909 
1910 	/*
1911 	 * This can happen if an interrupt occurs in the last stages
1912 	 * of guest entry or the first stages of guest exit (i.e. after
1913 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1914 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1915 	 * That can happen due to a bug, or due to a machine check
1916 	 * occurring at just the wrong time.
1917 	 */
1918 	if (vcpu->arch.shregs.msr & MSR_HV) {
1919 		pr_emerg("KVM trap in HV mode while nested!\n");
1920 		pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1921 			 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1922 			 vcpu->arch.shregs.msr);
1923 		kvmppc_dump_regs(vcpu);
1924 		return RESUME_HOST;
1925 	}
1926 	switch (vcpu->arch.trap) {
1927 	/* We're good on these - the host merely wanted to get our attention */
1928 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1929 		vcpu->stat.dec_exits++;
1930 		r = RESUME_GUEST;
1931 		break;
1932 	case BOOK3S_INTERRUPT_EXTERNAL:
1933 		vcpu->stat.ext_intr_exits++;
1934 		r = RESUME_HOST;
1935 		break;
1936 	case BOOK3S_INTERRUPT_H_DOORBELL:
1937 	case BOOK3S_INTERRUPT_H_VIRT:
1938 		vcpu->stat.ext_intr_exits++;
1939 		r = RESUME_GUEST;
1940 		break;
1941 	/* These need to go to the nested HV */
1942 	case BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER:
1943 		vcpu->arch.trap = BOOK3S_INTERRUPT_HV_DECREMENTER;
1944 		vcpu->stat.dec_exits++;
1945 		r = RESUME_HOST;
1946 		break;
1947 	/* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1948 	case BOOK3S_INTERRUPT_HMI:
1949 	case BOOK3S_INTERRUPT_PERFMON:
1950 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
1951 		r = RESUME_GUEST;
1952 		break;
1953 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
1954 	{
1955 		static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
1956 					      DEFAULT_RATELIMIT_BURST);
1957 		/* Pass the machine check to the L1 guest */
1958 		r = RESUME_HOST;
1959 		/* Print the MCE event to host console. */
1960 		if (__ratelimit(&rs))
1961 			machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1962 		break;
1963 	}
1964 	/*
1965 	 * We get these next two if the guest accesses a page which it thinks
1966 	 * it has mapped but which is not actually present, either because
1967 	 * it is for an emulated I/O device or because the corresonding
1968 	 * host page has been paged out.
1969 	 */
1970 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1971 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1972 		r = kvmhv_nested_page_fault(vcpu);
1973 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1974 		break;
1975 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
1976 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1977 		vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1978 					 DSISR_SRR1_MATCH_64S;
1979 		if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1980 			vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1981 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1982 		r = kvmhv_nested_page_fault(vcpu);
1983 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1984 		break;
1985 
1986 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1987 	case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1988 		/*
1989 		 * This occurs for various TM-related instructions that
1990 		 * we need to emulate on POWER9 DD2.2.  We have already
1991 		 * handled the cases where the guest was in real-suspend
1992 		 * mode and was transitioning to transactional state.
1993 		 */
1994 		r = kvmhv_p9_tm_emulation(vcpu);
1995 		if (r != -1)
1996 			break;
1997 		fallthrough; /* go to facility unavailable handler */
1998 #endif
1999 
2000 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL: {
2001 		u64 cause = vcpu->arch.hfscr >> 56;
2002 
2003 		/*
2004 		 * Only pass HFU interrupts to the L1 if the facility is
2005 		 * permitted but disabled by the L1's HFSCR, otherwise
2006 		 * the interrupt does not make sense to the L1 so turn
2007 		 * it into a HEAI.
2008 		 */
2009 		if (!(vcpu->arch.hfscr_permitted & (1UL << cause)) ||
2010 				(vcpu->arch.nested_hfscr & (1UL << cause))) {
2011 			ppc_inst_t pinst;
2012 			vcpu->arch.trap = BOOK3S_INTERRUPT_H_EMUL_ASSIST;
2013 
2014 			/*
2015 			 * If the fetch failed, return to guest and
2016 			 * try executing it again.
2017 			 */
2018 			r = kvmppc_get_last_inst(vcpu, INST_GENERIC, &pinst);
2019 			vcpu->arch.emul_inst = ppc_inst_val(pinst);
2020 			if (r != EMULATE_DONE)
2021 				r = RESUME_GUEST;
2022 			else
2023 				r = RESUME_HOST;
2024 		} else {
2025 			r = RESUME_HOST;
2026 		}
2027 
2028 		break;
2029 	}
2030 
2031 	case BOOK3S_INTERRUPT_HV_RM_HARD:
2032 		vcpu->arch.trap = 0;
2033 		r = RESUME_GUEST;
2034 		if (!xics_on_xive())
2035 			kvmppc_xics_rm_complete(vcpu, 0);
2036 		break;
2037 	case BOOK3S_INTERRUPT_SYSCALL:
2038 	{
2039 		unsigned long req = kvmppc_get_gpr(vcpu, 3);
2040 
2041 		/*
2042 		 * The H_RPT_INVALIDATE hcalls issued by nested
2043 		 * guests for process-scoped invalidations when
2044 		 * GTSE=0, are handled here in L0.
2045 		 */
2046 		if (req == H_RPT_INVALIDATE) {
2047 			r = kvmppc_nested_h_rpt_invalidate(vcpu);
2048 			break;
2049 		}
2050 
2051 		r = RESUME_HOST;
2052 		break;
2053 	}
2054 	default:
2055 		r = RESUME_HOST;
2056 		break;
2057 	}
2058 
2059 	return r;
2060 }
2061 
2062 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
2063 					    struct kvm_sregs *sregs)
2064 {
2065 	int i;
2066 
2067 	memset(sregs, 0, sizeof(struct kvm_sregs));
2068 	sregs->pvr = vcpu->arch.pvr;
2069 	for (i = 0; i < vcpu->arch.slb_max; i++) {
2070 		sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
2071 		sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
2072 	}
2073 
2074 	return 0;
2075 }
2076 
2077 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
2078 					    struct kvm_sregs *sregs)
2079 {
2080 	int i, j;
2081 
2082 	/* Only accept the same PVR as the host's, since we can't spoof it */
2083 	if (sregs->pvr != vcpu->arch.pvr)
2084 		return -EINVAL;
2085 
2086 	j = 0;
2087 	for (i = 0; i < vcpu->arch.slb_nr; i++) {
2088 		if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
2089 			vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
2090 			vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
2091 			++j;
2092 		}
2093 	}
2094 	vcpu->arch.slb_max = j;
2095 
2096 	return 0;
2097 }
2098 
2099 /*
2100  * Enforce limits on guest LPCR values based on hardware availability,
2101  * guest configuration, and possibly hypervisor support and security
2102  * concerns.
2103  */
2104 unsigned long kvmppc_filter_lpcr_hv(struct kvm *kvm, unsigned long lpcr)
2105 {
2106 	/* LPCR_TC only applies to HPT guests */
2107 	if (kvm_is_radix(kvm))
2108 		lpcr &= ~LPCR_TC;
2109 
2110 	/* On POWER8 and above, userspace can modify AIL */
2111 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2112 		lpcr &= ~LPCR_AIL;
2113 	if ((lpcr & LPCR_AIL) != LPCR_AIL_3)
2114 		lpcr &= ~LPCR_AIL; /* LPCR[AIL]=1/2 is disallowed */
2115 	/*
2116 	 * On some POWER9s we force AIL off for radix guests to prevent
2117 	 * executing in MSR[HV]=1 mode with the MMU enabled and PIDR set to
2118 	 * guest, which can result in Q0 translations with LPID=0 PID=PIDR to
2119 	 * be cached, which the host TLB management does not expect.
2120 	 */
2121 	if (kvm_is_radix(kvm) && cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG))
2122 		lpcr &= ~LPCR_AIL;
2123 
2124 	/*
2125 	 * On POWER9, allow userspace to enable large decrementer for the
2126 	 * guest, whether or not the host has it enabled.
2127 	 */
2128 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
2129 		lpcr &= ~LPCR_LD;
2130 
2131 	return lpcr;
2132 }
2133 
2134 static void verify_lpcr(struct kvm *kvm, unsigned long lpcr)
2135 {
2136 	if (lpcr != kvmppc_filter_lpcr_hv(kvm, lpcr)) {
2137 		WARN_ONCE(1, "lpcr 0x%lx differs from filtered 0x%lx\n",
2138 			  lpcr, kvmppc_filter_lpcr_hv(kvm, lpcr));
2139 	}
2140 }
2141 
2142 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
2143 		bool preserve_top32)
2144 {
2145 	struct kvm *kvm = vcpu->kvm;
2146 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
2147 	u64 mask;
2148 
2149 	spin_lock(&vc->lock);
2150 
2151 	/*
2152 	 * Userspace can only modify
2153 	 * DPFD (default prefetch depth), ILE (interrupt little-endian),
2154 	 * TC (translation control), AIL (alternate interrupt location),
2155 	 * LD (large decrementer).
2156 	 * These are subject to restrictions from kvmppc_filter_lcpr_hv().
2157 	 */
2158 	mask = LPCR_DPFD | LPCR_ILE | LPCR_TC | LPCR_AIL | LPCR_LD;
2159 
2160 	/* Broken 32-bit version of LPCR must not clear top bits */
2161 	if (preserve_top32)
2162 		mask &= 0xFFFFFFFF;
2163 
2164 	new_lpcr = kvmppc_filter_lpcr_hv(kvm,
2165 			(vc->lpcr & ~mask) | (new_lpcr & mask));
2166 
2167 	/*
2168 	 * If ILE (interrupt little-endian) has changed, update the
2169 	 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
2170 	 */
2171 	if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
2172 		struct kvm_vcpu *vcpu;
2173 		unsigned long i;
2174 
2175 		kvm_for_each_vcpu(i, vcpu, kvm) {
2176 			if (vcpu->arch.vcore != vc)
2177 				continue;
2178 			if (new_lpcr & LPCR_ILE)
2179 				vcpu->arch.intr_msr |= MSR_LE;
2180 			else
2181 				vcpu->arch.intr_msr &= ~MSR_LE;
2182 		}
2183 	}
2184 
2185 	vc->lpcr = new_lpcr;
2186 
2187 	spin_unlock(&vc->lock);
2188 }
2189 
2190 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
2191 				 union kvmppc_one_reg *val)
2192 {
2193 	int r = 0;
2194 	long int i;
2195 
2196 	switch (id) {
2197 	case KVM_REG_PPC_DEBUG_INST:
2198 		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
2199 		break;
2200 	case KVM_REG_PPC_HIOR:
2201 		*val = get_reg_val(id, 0);
2202 		break;
2203 	case KVM_REG_PPC_DABR:
2204 		*val = get_reg_val(id, vcpu->arch.dabr);
2205 		break;
2206 	case KVM_REG_PPC_DABRX:
2207 		*val = get_reg_val(id, vcpu->arch.dabrx);
2208 		break;
2209 	case KVM_REG_PPC_DSCR:
2210 		*val = get_reg_val(id, vcpu->arch.dscr);
2211 		break;
2212 	case KVM_REG_PPC_PURR:
2213 		*val = get_reg_val(id, vcpu->arch.purr);
2214 		break;
2215 	case KVM_REG_PPC_SPURR:
2216 		*val = get_reg_val(id, vcpu->arch.spurr);
2217 		break;
2218 	case KVM_REG_PPC_AMR:
2219 		*val = get_reg_val(id, vcpu->arch.amr);
2220 		break;
2221 	case KVM_REG_PPC_UAMOR:
2222 		*val = get_reg_val(id, vcpu->arch.uamor);
2223 		break;
2224 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
2225 		i = id - KVM_REG_PPC_MMCR0;
2226 		*val = get_reg_val(id, vcpu->arch.mmcr[i]);
2227 		break;
2228 	case KVM_REG_PPC_MMCR2:
2229 		*val = get_reg_val(id, vcpu->arch.mmcr[2]);
2230 		break;
2231 	case KVM_REG_PPC_MMCRA:
2232 		*val = get_reg_val(id, vcpu->arch.mmcra);
2233 		break;
2234 	case KVM_REG_PPC_MMCRS:
2235 		*val = get_reg_val(id, vcpu->arch.mmcrs);
2236 		break;
2237 	case KVM_REG_PPC_MMCR3:
2238 		*val = get_reg_val(id, vcpu->arch.mmcr[3]);
2239 		break;
2240 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
2241 		i = id - KVM_REG_PPC_PMC1;
2242 		*val = get_reg_val(id, vcpu->arch.pmc[i]);
2243 		break;
2244 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
2245 		i = id - KVM_REG_PPC_SPMC1;
2246 		*val = get_reg_val(id, vcpu->arch.spmc[i]);
2247 		break;
2248 	case KVM_REG_PPC_SIAR:
2249 		*val = get_reg_val(id, vcpu->arch.siar);
2250 		break;
2251 	case KVM_REG_PPC_SDAR:
2252 		*val = get_reg_val(id, vcpu->arch.sdar);
2253 		break;
2254 	case KVM_REG_PPC_SIER:
2255 		*val = get_reg_val(id, vcpu->arch.sier[0]);
2256 		break;
2257 	case KVM_REG_PPC_SIER2:
2258 		*val = get_reg_val(id, vcpu->arch.sier[1]);
2259 		break;
2260 	case KVM_REG_PPC_SIER3:
2261 		*val = get_reg_val(id, vcpu->arch.sier[2]);
2262 		break;
2263 	case KVM_REG_PPC_IAMR:
2264 		*val = get_reg_val(id, vcpu->arch.iamr);
2265 		break;
2266 	case KVM_REG_PPC_PSPB:
2267 		*val = get_reg_val(id, vcpu->arch.pspb);
2268 		break;
2269 	case KVM_REG_PPC_DPDES:
2270 		/*
2271 		 * On POWER9, where we are emulating msgsndp etc.,
2272 		 * we return 1 bit for each vcpu, which can come from
2273 		 * either vcore->dpdes or doorbell_request.
2274 		 * On POWER8, doorbell_request is 0.
2275 		 */
2276 		if (cpu_has_feature(CPU_FTR_ARCH_300))
2277 			*val = get_reg_val(id, vcpu->arch.doorbell_request);
2278 		else
2279 			*val = get_reg_val(id, vcpu->arch.vcore->dpdes);
2280 		break;
2281 	case KVM_REG_PPC_VTB:
2282 		*val = get_reg_val(id, vcpu->arch.vcore->vtb);
2283 		break;
2284 	case KVM_REG_PPC_DAWR:
2285 		*val = get_reg_val(id, vcpu->arch.dawr0);
2286 		break;
2287 	case KVM_REG_PPC_DAWRX:
2288 		*val = get_reg_val(id, vcpu->arch.dawrx0);
2289 		break;
2290 	case KVM_REG_PPC_DAWR1:
2291 		*val = get_reg_val(id, vcpu->arch.dawr1);
2292 		break;
2293 	case KVM_REG_PPC_DAWRX1:
2294 		*val = get_reg_val(id, vcpu->arch.dawrx1);
2295 		break;
2296 	case KVM_REG_PPC_CIABR:
2297 		*val = get_reg_val(id, vcpu->arch.ciabr);
2298 		break;
2299 	case KVM_REG_PPC_CSIGR:
2300 		*val = get_reg_val(id, vcpu->arch.csigr);
2301 		break;
2302 	case KVM_REG_PPC_TACR:
2303 		*val = get_reg_val(id, vcpu->arch.tacr);
2304 		break;
2305 	case KVM_REG_PPC_TCSCR:
2306 		*val = get_reg_val(id, vcpu->arch.tcscr);
2307 		break;
2308 	case KVM_REG_PPC_PID:
2309 		*val = get_reg_val(id, vcpu->arch.pid);
2310 		break;
2311 	case KVM_REG_PPC_ACOP:
2312 		*val = get_reg_val(id, vcpu->arch.acop);
2313 		break;
2314 	case KVM_REG_PPC_WORT:
2315 		*val = get_reg_val(id, vcpu->arch.wort);
2316 		break;
2317 	case KVM_REG_PPC_TIDR:
2318 		*val = get_reg_val(id, vcpu->arch.tid);
2319 		break;
2320 	case KVM_REG_PPC_PSSCR:
2321 		*val = get_reg_val(id, vcpu->arch.psscr);
2322 		break;
2323 	case KVM_REG_PPC_VPA_ADDR:
2324 		spin_lock(&vcpu->arch.vpa_update_lock);
2325 		*val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
2326 		spin_unlock(&vcpu->arch.vpa_update_lock);
2327 		break;
2328 	case KVM_REG_PPC_VPA_SLB:
2329 		spin_lock(&vcpu->arch.vpa_update_lock);
2330 		val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
2331 		val->vpaval.length = vcpu->arch.slb_shadow.len;
2332 		spin_unlock(&vcpu->arch.vpa_update_lock);
2333 		break;
2334 	case KVM_REG_PPC_VPA_DTL:
2335 		spin_lock(&vcpu->arch.vpa_update_lock);
2336 		val->vpaval.addr = vcpu->arch.dtl.next_gpa;
2337 		val->vpaval.length = vcpu->arch.dtl.len;
2338 		spin_unlock(&vcpu->arch.vpa_update_lock);
2339 		break;
2340 	case KVM_REG_PPC_TB_OFFSET:
2341 		*val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
2342 		break;
2343 	case KVM_REG_PPC_LPCR:
2344 	case KVM_REG_PPC_LPCR_64:
2345 		*val = get_reg_val(id, vcpu->arch.vcore->lpcr);
2346 		break;
2347 	case KVM_REG_PPC_PPR:
2348 		*val = get_reg_val(id, vcpu->arch.ppr);
2349 		break;
2350 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2351 	case KVM_REG_PPC_TFHAR:
2352 		*val = get_reg_val(id, vcpu->arch.tfhar);
2353 		break;
2354 	case KVM_REG_PPC_TFIAR:
2355 		*val = get_reg_val(id, vcpu->arch.tfiar);
2356 		break;
2357 	case KVM_REG_PPC_TEXASR:
2358 		*val = get_reg_val(id, vcpu->arch.texasr);
2359 		break;
2360 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2361 		i = id - KVM_REG_PPC_TM_GPR0;
2362 		*val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
2363 		break;
2364 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2365 	{
2366 		int j;
2367 		i = id - KVM_REG_PPC_TM_VSR0;
2368 		if (i < 32)
2369 			for (j = 0; j < TS_FPRWIDTH; j++)
2370 				val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
2371 		else {
2372 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
2373 				val->vval = vcpu->arch.vr_tm.vr[i-32];
2374 			else
2375 				r = -ENXIO;
2376 		}
2377 		break;
2378 	}
2379 	case KVM_REG_PPC_TM_CR:
2380 		*val = get_reg_val(id, vcpu->arch.cr_tm);
2381 		break;
2382 	case KVM_REG_PPC_TM_XER:
2383 		*val = get_reg_val(id, vcpu->arch.xer_tm);
2384 		break;
2385 	case KVM_REG_PPC_TM_LR:
2386 		*val = get_reg_val(id, vcpu->arch.lr_tm);
2387 		break;
2388 	case KVM_REG_PPC_TM_CTR:
2389 		*val = get_reg_val(id, vcpu->arch.ctr_tm);
2390 		break;
2391 	case KVM_REG_PPC_TM_FPSCR:
2392 		*val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
2393 		break;
2394 	case KVM_REG_PPC_TM_AMR:
2395 		*val = get_reg_val(id, vcpu->arch.amr_tm);
2396 		break;
2397 	case KVM_REG_PPC_TM_PPR:
2398 		*val = get_reg_val(id, vcpu->arch.ppr_tm);
2399 		break;
2400 	case KVM_REG_PPC_TM_VRSAVE:
2401 		*val = get_reg_val(id, vcpu->arch.vrsave_tm);
2402 		break;
2403 	case KVM_REG_PPC_TM_VSCR:
2404 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
2405 			*val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
2406 		else
2407 			r = -ENXIO;
2408 		break;
2409 	case KVM_REG_PPC_TM_DSCR:
2410 		*val = get_reg_val(id, vcpu->arch.dscr_tm);
2411 		break;
2412 	case KVM_REG_PPC_TM_TAR:
2413 		*val = get_reg_val(id, vcpu->arch.tar_tm);
2414 		break;
2415 #endif
2416 	case KVM_REG_PPC_ARCH_COMPAT:
2417 		*val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
2418 		break;
2419 	case KVM_REG_PPC_DEC_EXPIRY:
2420 		*val = get_reg_val(id, vcpu->arch.dec_expires);
2421 		break;
2422 	case KVM_REG_PPC_ONLINE:
2423 		*val = get_reg_val(id, vcpu->arch.online);
2424 		break;
2425 	case KVM_REG_PPC_PTCR:
2426 		*val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
2427 		break;
2428 	default:
2429 		r = -EINVAL;
2430 		break;
2431 	}
2432 
2433 	return r;
2434 }
2435 
2436 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
2437 				 union kvmppc_one_reg *val)
2438 {
2439 	int r = 0;
2440 	long int i;
2441 	unsigned long addr, len;
2442 
2443 	switch (id) {
2444 	case KVM_REG_PPC_HIOR:
2445 		/* Only allow this to be set to zero */
2446 		if (set_reg_val(id, *val))
2447 			r = -EINVAL;
2448 		break;
2449 	case KVM_REG_PPC_DABR:
2450 		vcpu->arch.dabr = set_reg_val(id, *val);
2451 		break;
2452 	case KVM_REG_PPC_DABRX:
2453 		vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
2454 		break;
2455 	case KVM_REG_PPC_DSCR:
2456 		vcpu->arch.dscr = set_reg_val(id, *val);
2457 		break;
2458 	case KVM_REG_PPC_PURR:
2459 		vcpu->arch.purr = set_reg_val(id, *val);
2460 		break;
2461 	case KVM_REG_PPC_SPURR:
2462 		vcpu->arch.spurr = set_reg_val(id, *val);
2463 		break;
2464 	case KVM_REG_PPC_AMR:
2465 		vcpu->arch.amr = set_reg_val(id, *val);
2466 		break;
2467 	case KVM_REG_PPC_UAMOR:
2468 		vcpu->arch.uamor = set_reg_val(id, *val);
2469 		break;
2470 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1:
2471 		i = id - KVM_REG_PPC_MMCR0;
2472 		vcpu->arch.mmcr[i] = set_reg_val(id, *val);
2473 		break;
2474 	case KVM_REG_PPC_MMCR2:
2475 		vcpu->arch.mmcr[2] = set_reg_val(id, *val);
2476 		break;
2477 	case KVM_REG_PPC_MMCRA:
2478 		vcpu->arch.mmcra = set_reg_val(id, *val);
2479 		break;
2480 	case KVM_REG_PPC_MMCRS:
2481 		vcpu->arch.mmcrs = set_reg_val(id, *val);
2482 		break;
2483 	case KVM_REG_PPC_MMCR3:
2484 		*val = get_reg_val(id, vcpu->arch.mmcr[3]);
2485 		break;
2486 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
2487 		i = id - KVM_REG_PPC_PMC1;
2488 		vcpu->arch.pmc[i] = set_reg_val(id, *val);
2489 		break;
2490 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
2491 		i = id - KVM_REG_PPC_SPMC1;
2492 		vcpu->arch.spmc[i] = set_reg_val(id, *val);
2493 		break;
2494 	case KVM_REG_PPC_SIAR:
2495 		vcpu->arch.siar = set_reg_val(id, *val);
2496 		break;
2497 	case KVM_REG_PPC_SDAR:
2498 		vcpu->arch.sdar = set_reg_val(id, *val);
2499 		break;
2500 	case KVM_REG_PPC_SIER:
2501 		vcpu->arch.sier[0] = set_reg_val(id, *val);
2502 		break;
2503 	case KVM_REG_PPC_SIER2:
2504 		vcpu->arch.sier[1] = set_reg_val(id, *val);
2505 		break;
2506 	case KVM_REG_PPC_SIER3:
2507 		vcpu->arch.sier[2] = set_reg_val(id, *val);
2508 		break;
2509 	case KVM_REG_PPC_IAMR:
2510 		vcpu->arch.iamr = set_reg_val(id, *val);
2511 		break;
2512 	case KVM_REG_PPC_PSPB:
2513 		vcpu->arch.pspb = set_reg_val(id, *val);
2514 		break;
2515 	case KVM_REG_PPC_DPDES:
2516 		if (cpu_has_feature(CPU_FTR_ARCH_300))
2517 			vcpu->arch.doorbell_request = set_reg_val(id, *val) & 1;
2518 		else
2519 			vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
2520 		break;
2521 	case KVM_REG_PPC_VTB:
2522 		vcpu->arch.vcore->vtb = set_reg_val(id, *val);
2523 		break;
2524 	case KVM_REG_PPC_DAWR:
2525 		vcpu->arch.dawr0 = set_reg_val(id, *val);
2526 		break;
2527 	case KVM_REG_PPC_DAWRX:
2528 		vcpu->arch.dawrx0 = set_reg_val(id, *val) & ~DAWRX_HYP;
2529 		break;
2530 	case KVM_REG_PPC_DAWR1:
2531 		vcpu->arch.dawr1 = set_reg_val(id, *val);
2532 		break;
2533 	case KVM_REG_PPC_DAWRX1:
2534 		vcpu->arch.dawrx1 = set_reg_val(id, *val) & ~DAWRX_HYP;
2535 		break;
2536 	case KVM_REG_PPC_CIABR:
2537 		vcpu->arch.ciabr = set_reg_val(id, *val);
2538 		/* Don't allow setting breakpoints in hypervisor code */
2539 		if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
2540 			vcpu->arch.ciabr &= ~CIABR_PRIV;	/* disable */
2541 		break;
2542 	case KVM_REG_PPC_CSIGR:
2543 		vcpu->arch.csigr = set_reg_val(id, *val);
2544 		break;
2545 	case KVM_REG_PPC_TACR:
2546 		vcpu->arch.tacr = set_reg_val(id, *val);
2547 		break;
2548 	case KVM_REG_PPC_TCSCR:
2549 		vcpu->arch.tcscr = set_reg_val(id, *val);
2550 		break;
2551 	case KVM_REG_PPC_PID:
2552 		vcpu->arch.pid = set_reg_val(id, *val);
2553 		break;
2554 	case KVM_REG_PPC_ACOP:
2555 		vcpu->arch.acop = set_reg_val(id, *val);
2556 		break;
2557 	case KVM_REG_PPC_WORT:
2558 		vcpu->arch.wort = set_reg_val(id, *val);
2559 		break;
2560 	case KVM_REG_PPC_TIDR:
2561 		vcpu->arch.tid = set_reg_val(id, *val);
2562 		break;
2563 	case KVM_REG_PPC_PSSCR:
2564 		vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
2565 		break;
2566 	case KVM_REG_PPC_VPA_ADDR:
2567 		addr = set_reg_val(id, *val);
2568 		r = -EINVAL;
2569 		if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
2570 			      vcpu->arch.dtl.next_gpa))
2571 			break;
2572 		r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
2573 		break;
2574 	case KVM_REG_PPC_VPA_SLB:
2575 		addr = val->vpaval.addr;
2576 		len = val->vpaval.length;
2577 		r = -EINVAL;
2578 		if (addr && !vcpu->arch.vpa.next_gpa)
2579 			break;
2580 		r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
2581 		break;
2582 	case KVM_REG_PPC_VPA_DTL:
2583 		addr = val->vpaval.addr;
2584 		len = val->vpaval.length;
2585 		r = -EINVAL;
2586 		if (addr && (len < sizeof(struct dtl_entry) ||
2587 			     !vcpu->arch.vpa.next_gpa))
2588 			break;
2589 		len -= len % sizeof(struct dtl_entry);
2590 		r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
2591 		break;
2592 	case KVM_REG_PPC_TB_OFFSET:
2593 	{
2594 		/* round up to multiple of 2^24 */
2595 		u64 tb_offset = ALIGN(set_reg_val(id, *val), 1UL << 24);
2596 
2597 		/*
2598 		 * Now that we know the timebase offset, update the
2599 		 * decrementer expiry with a guest timebase value. If
2600 		 * the userspace does not set DEC_EXPIRY, this ensures
2601 		 * a migrated vcpu at least starts with an expired
2602 		 * decrementer, which is better than a large one that
2603 		 * causes a hang.
2604 		 */
2605 		if (!vcpu->arch.dec_expires && tb_offset)
2606 			vcpu->arch.dec_expires = get_tb() + tb_offset;
2607 
2608 		vcpu->arch.vcore->tb_offset = tb_offset;
2609 		break;
2610 	}
2611 	case KVM_REG_PPC_LPCR:
2612 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
2613 		break;
2614 	case KVM_REG_PPC_LPCR_64:
2615 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
2616 		break;
2617 	case KVM_REG_PPC_PPR:
2618 		vcpu->arch.ppr = set_reg_val(id, *val);
2619 		break;
2620 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2621 	case KVM_REG_PPC_TFHAR:
2622 		vcpu->arch.tfhar = set_reg_val(id, *val);
2623 		break;
2624 	case KVM_REG_PPC_TFIAR:
2625 		vcpu->arch.tfiar = set_reg_val(id, *val);
2626 		break;
2627 	case KVM_REG_PPC_TEXASR:
2628 		vcpu->arch.texasr = set_reg_val(id, *val);
2629 		break;
2630 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2631 		i = id - KVM_REG_PPC_TM_GPR0;
2632 		vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
2633 		break;
2634 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2635 	{
2636 		int j;
2637 		i = id - KVM_REG_PPC_TM_VSR0;
2638 		if (i < 32)
2639 			for (j = 0; j < TS_FPRWIDTH; j++)
2640 				vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
2641 		else
2642 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
2643 				vcpu->arch.vr_tm.vr[i-32] = val->vval;
2644 			else
2645 				r = -ENXIO;
2646 		break;
2647 	}
2648 	case KVM_REG_PPC_TM_CR:
2649 		vcpu->arch.cr_tm = set_reg_val(id, *val);
2650 		break;
2651 	case KVM_REG_PPC_TM_XER:
2652 		vcpu->arch.xer_tm = set_reg_val(id, *val);
2653 		break;
2654 	case KVM_REG_PPC_TM_LR:
2655 		vcpu->arch.lr_tm = set_reg_val(id, *val);
2656 		break;
2657 	case KVM_REG_PPC_TM_CTR:
2658 		vcpu->arch.ctr_tm = set_reg_val(id, *val);
2659 		break;
2660 	case KVM_REG_PPC_TM_FPSCR:
2661 		vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
2662 		break;
2663 	case KVM_REG_PPC_TM_AMR:
2664 		vcpu->arch.amr_tm = set_reg_val(id, *val);
2665 		break;
2666 	case KVM_REG_PPC_TM_PPR:
2667 		vcpu->arch.ppr_tm = set_reg_val(id, *val);
2668 		break;
2669 	case KVM_REG_PPC_TM_VRSAVE:
2670 		vcpu->arch.vrsave_tm = set_reg_val(id, *val);
2671 		break;
2672 	case KVM_REG_PPC_TM_VSCR:
2673 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
2674 			vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
2675 		else
2676 			r = - ENXIO;
2677 		break;
2678 	case KVM_REG_PPC_TM_DSCR:
2679 		vcpu->arch.dscr_tm = set_reg_val(id, *val);
2680 		break;
2681 	case KVM_REG_PPC_TM_TAR:
2682 		vcpu->arch.tar_tm = set_reg_val(id, *val);
2683 		break;
2684 #endif
2685 	case KVM_REG_PPC_ARCH_COMPAT:
2686 		r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
2687 		break;
2688 	case KVM_REG_PPC_DEC_EXPIRY:
2689 		vcpu->arch.dec_expires = set_reg_val(id, *val);
2690 		break;
2691 	case KVM_REG_PPC_ONLINE:
2692 		i = set_reg_val(id, *val);
2693 		if (i && !vcpu->arch.online)
2694 			atomic_inc(&vcpu->arch.vcore->online_count);
2695 		else if (!i && vcpu->arch.online)
2696 			atomic_dec(&vcpu->arch.vcore->online_count);
2697 		vcpu->arch.online = i;
2698 		break;
2699 	case KVM_REG_PPC_PTCR:
2700 		vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
2701 		break;
2702 	default:
2703 		r = -EINVAL;
2704 		break;
2705 	}
2706 
2707 	return r;
2708 }
2709 
2710 /*
2711  * On POWER9, threads are independent and can be in different partitions.
2712  * Therefore we consider each thread to be a subcore.
2713  * There is a restriction that all threads have to be in the same
2714  * MMU mode (radix or HPT), unfortunately, but since we only support
2715  * HPT guests on a HPT host so far, that isn't an impediment yet.
2716  */
2717 static int threads_per_vcore(struct kvm *kvm)
2718 {
2719 	if (cpu_has_feature(CPU_FTR_ARCH_300))
2720 		return 1;
2721 	return threads_per_subcore;
2722 }
2723 
2724 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
2725 {
2726 	struct kvmppc_vcore *vcore;
2727 
2728 	vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
2729 
2730 	if (vcore == NULL)
2731 		return NULL;
2732 
2733 	spin_lock_init(&vcore->lock);
2734 	spin_lock_init(&vcore->stoltb_lock);
2735 	rcuwait_init(&vcore->wait);
2736 	vcore->preempt_tb = TB_NIL;
2737 	vcore->lpcr = kvm->arch.lpcr;
2738 	vcore->first_vcpuid = id;
2739 	vcore->kvm = kvm;
2740 	INIT_LIST_HEAD(&vcore->preempt_list);
2741 
2742 	return vcore;
2743 }
2744 
2745 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2746 static struct debugfs_timings_element {
2747 	const char *name;
2748 	size_t offset;
2749 } timings[] = {
2750 #ifdef CONFIG_KVM_BOOK3S_HV_P9_TIMING
2751 	{"vcpu_entry",	offsetof(struct kvm_vcpu, arch.vcpu_entry)},
2752 	{"guest_entry",	offsetof(struct kvm_vcpu, arch.guest_entry)},
2753 	{"in_guest",	offsetof(struct kvm_vcpu, arch.in_guest)},
2754 	{"guest_exit",	offsetof(struct kvm_vcpu, arch.guest_exit)},
2755 	{"vcpu_exit",	offsetof(struct kvm_vcpu, arch.vcpu_exit)},
2756 	{"hypercall",	offsetof(struct kvm_vcpu, arch.hcall)},
2757 	{"page_fault",	offsetof(struct kvm_vcpu, arch.pg_fault)},
2758 #else
2759 	{"rm_entry",	offsetof(struct kvm_vcpu, arch.rm_entry)},
2760 	{"rm_intr",	offsetof(struct kvm_vcpu, arch.rm_intr)},
2761 	{"rm_exit",	offsetof(struct kvm_vcpu, arch.rm_exit)},
2762 	{"guest",	offsetof(struct kvm_vcpu, arch.guest_time)},
2763 	{"cede",	offsetof(struct kvm_vcpu, arch.cede_time)},
2764 #endif
2765 };
2766 
2767 #define N_TIMINGS	(ARRAY_SIZE(timings))
2768 
2769 struct debugfs_timings_state {
2770 	struct kvm_vcpu	*vcpu;
2771 	unsigned int	buflen;
2772 	char		buf[N_TIMINGS * 100];
2773 };
2774 
2775 static int debugfs_timings_open(struct inode *inode, struct file *file)
2776 {
2777 	struct kvm_vcpu *vcpu = inode->i_private;
2778 	struct debugfs_timings_state *p;
2779 
2780 	p = kzalloc(sizeof(*p), GFP_KERNEL);
2781 	if (!p)
2782 		return -ENOMEM;
2783 
2784 	kvm_get_kvm(vcpu->kvm);
2785 	p->vcpu = vcpu;
2786 	file->private_data = p;
2787 
2788 	return nonseekable_open(inode, file);
2789 }
2790 
2791 static int debugfs_timings_release(struct inode *inode, struct file *file)
2792 {
2793 	struct debugfs_timings_state *p = file->private_data;
2794 
2795 	kvm_put_kvm(p->vcpu->kvm);
2796 	kfree(p);
2797 	return 0;
2798 }
2799 
2800 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2801 				    size_t len, loff_t *ppos)
2802 {
2803 	struct debugfs_timings_state *p = file->private_data;
2804 	struct kvm_vcpu *vcpu = p->vcpu;
2805 	char *s, *buf_end;
2806 	struct kvmhv_tb_accumulator tb;
2807 	u64 count;
2808 	loff_t pos;
2809 	ssize_t n;
2810 	int i, loops;
2811 	bool ok;
2812 
2813 	if (!p->buflen) {
2814 		s = p->buf;
2815 		buf_end = s + sizeof(p->buf);
2816 		for (i = 0; i < N_TIMINGS; ++i) {
2817 			struct kvmhv_tb_accumulator *acc;
2818 
2819 			acc = (struct kvmhv_tb_accumulator *)
2820 				((unsigned long)vcpu + timings[i].offset);
2821 			ok = false;
2822 			for (loops = 0; loops < 1000; ++loops) {
2823 				count = acc->seqcount;
2824 				if (!(count & 1)) {
2825 					smp_rmb();
2826 					tb = *acc;
2827 					smp_rmb();
2828 					if (count == acc->seqcount) {
2829 						ok = true;
2830 						break;
2831 					}
2832 				}
2833 				udelay(1);
2834 			}
2835 			if (!ok)
2836 				snprintf(s, buf_end - s, "%s: stuck\n",
2837 					timings[i].name);
2838 			else
2839 				snprintf(s, buf_end - s,
2840 					"%s: %llu %llu %llu %llu\n",
2841 					timings[i].name, count / 2,
2842 					tb_to_ns(tb.tb_total),
2843 					tb_to_ns(tb.tb_min),
2844 					tb_to_ns(tb.tb_max));
2845 			s += strlen(s);
2846 		}
2847 		p->buflen = s - p->buf;
2848 	}
2849 
2850 	pos = *ppos;
2851 	if (pos >= p->buflen)
2852 		return 0;
2853 	if (len > p->buflen - pos)
2854 		len = p->buflen - pos;
2855 	n = copy_to_user(buf, p->buf + pos, len);
2856 	if (n) {
2857 		if (n == len)
2858 			return -EFAULT;
2859 		len -= n;
2860 	}
2861 	*ppos = pos + len;
2862 	return len;
2863 }
2864 
2865 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2866 				     size_t len, loff_t *ppos)
2867 {
2868 	return -EACCES;
2869 }
2870 
2871 static const struct file_operations debugfs_timings_ops = {
2872 	.owner	 = THIS_MODULE,
2873 	.open	 = debugfs_timings_open,
2874 	.release = debugfs_timings_release,
2875 	.read	 = debugfs_timings_read,
2876 	.write	 = debugfs_timings_write,
2877 	.llseek	 = generic_file_llseek,
2878 };
2879 
2880 /* Create a debugfs directory for the vcpu */
2881 static int kvmppc_arch_create_vcpu_debugfs_hv(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry)
2882 {
2883 	if (cpu_has_feature(CPU_FTR_ARCH_300) == IS_ENABLED(CONFIG_KVM_BOOK3S_HV_P9_TIMING))
2884 		debugfs_create_file("timings", 0444, debugfs_dentry, vcpu,
2885 				    &debugfs_timings_ops);
2886 	return 0;
2887 }
2888 
2889 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2890 static int kvmppc_arch_create_vcpu_debugfs_hv(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry)
2891 {
2892 	return 0;
2893 }
2894 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2895 
2896 static int kvmppc_core_vcpu_create_hv(struct kvm_vcpu *vcpu)
2897 {
2898 	int err;
2899 	int core;
2900 	struct kvmppc_vcore *vcore;
2901 	struct kvm *kvm;
2902 	unsigned int id;
2903 
2904 	kvm = vcpu->kvm;
2905 	id = vcpu->vcpu_id;
2906 
2907 	vcpu->arch.shared = &vcpu->arch.shregs;
2908 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2909 	/*
2910 	 * The shared struct is never shared on HV,
2911 	 * so we can always use host endianness
2912 	 */
2913 #ifdef __BIG_ENDIAN__
2914 	vcpu->arch.shared_big_endian = true;
2915 #else
2916 	vcpu->arch.shared_big_endian = false;
2917 #endif
2918 #endif
2919 	vcpu->arch.mmcr[0] = MMCR0_FC;
2920 	if (cpu_has_feature(CPU_FTR_ARCH_31)) {
2921 		vcpu->arch.mmcr[0] |= MMCR0_PMCCEXT;
2922 		vcpu->arch.mmcra = MMCRA_BHRB_DISABLE;
2923 	}
2924 
2925 	vcpu->arch.ctrl = CTRL_RUNLATCH;
2926 	/* default to host PVR, since we can't spoof it */
2927 	kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2928 	spin_lock_init(&vcpu->arch.vpa_update_lock);
2929 	spin_lock_init(&vcpu->arch.tbacct_lock);
2930 	vcpu->arch.busy_preempt = TB_NIL;
2931 	vcpu->arch.shregs.msr = MSR_ME;
2932 	vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2933 
2934 	/*
2935 	 * Set the default HFSCR for the guest from the host value.
2936 	 * This value is only used on POWER9 and later.
2937 	 * On >= POWER9, we want to virtualize the doorbell facility, so we
2938 	 * don't set the HFSCR_MSGP bit, and that causes those instructions
2939 	 * to trap and then we emulate them.
2940 	 */
2941 	vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2942 		HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP;
2943 
2944 	/* On POWER10 and later, allow prefixed instructions */
2945 	if (cpu_has_feature(CPU_FTR_ARCH_31))
2946 		vcpu->arch.hfscr |= HFSCR_PREFIX;
2947 
2948 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
2949 		vcpu->arch.hfscr &= mfspr(SPRN_HFSCR);
2950 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2951 		if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2952 			vcpu->arch.hfscr |= HFSCR_TM;
2953 #endif
2954 	}
2955 	if (cpu_has_feature(CPU_FTR_TM_COMP))
2956 		vcpu->arch.hfscr |= HFSCR_TM;
2957 
2958 	vcpu->arch.hfscr_permitted = vcpu->arch.hfscr;
2959 
2960 	/*
2961 	 * PM, EBB, TM are demand-faulted so start with it clear.
2962 	 */
2963 	vcpu->arch.hfscr &= ~(HFSCR_PM | HFSCR_EBB | HFSCR_TM);
2964 
2965 	kvmppc_mmu_book3s_hv_init(vcpu);
2966 
2967 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2968 
2969 	init_waitqueue_head(&vcpu->arch.cpu_run);
2970 
2971 	mutex_lock(&kvm->lock);
2972 	vcore = NULL;
2973 	err = -EINVAL;
2974 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2975 		if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2976 			pr_devel("KVM: VCPU ID too high\n");
2977 			core = KVM_MAX_VCORES;
2978 		} else {
2979 			BUG_ON(kvm->arch.smt_mode != 1);
2980 			core = kvmppc_pack_vcpu_id(kvm, id);
2981 		}
2982 	} else {
2983 		core = id / kvm->arch.smt_mode;
2984 	}
2985 	if (core < KVM_MAX_VCORES) {
2986 		vcore = kvm->arch.vcores[core];
2987 		if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2988 			pr_devel("KVM: collision on id %u", id);
2989 			vcore = NULL;
2990 		} else if (!vcore) {
2991 			/*
2992 			 * Take mmu_setup_lock for mutual exclusion
2993 			 * with kvmppc_update_lpcr().
2994 			 */
2995 			err = -ENOMEM;
2996 			vcore = kvmppc_vcore_create(kvm,
2997 					id & ~(kvm->arch.smt_mode - 1));
2998 			mutex_lock(&kvm->arch.mmu_setup_lock);
2999 			kvm->arch.vcores[core] = vcore;
3000 			kvm->arch.online_vcores++;
3001 			mutex_unlock(&kvm->arch.mmu_setup_lock);
3002 		}
3003 	}
3004 	mutex_unlock(&kvm->lock);
3005 
3006 	if (!vcore)
3007 		return err;
3008 
3009 	spin_lock(&vcore->lock);
3010 	++vcore->num_threads;
3011 	spin_unlock(&vcore->lock);
3012 	vcpu->arch.vcore = vcore;
3013 	vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
3014 	vcpu->arch.thread_cpu = -1;
3015 	vcpu->arch.prev_cpu = -1;
3016 
3017 	vcpu->arch.cpu_type = KVM_CPU_3S_64;
3018 	kvmppc_sanity_check(vcpu);
3019 
3020 	return 0;
3021 }
3022 
3023 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
3024 			      unsigned long flags)
3025 {
3026 	int err;
3027 	int esmt = 0;
3028 
3029 	if (flags)
3030 		return -EINVAL;
3031 	if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
3032 		return -EINVAL;
3033 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
3034 		/*
3035 		 * On POWER8 (or POWER7), the threading mode is "strict",
3036 		 * so we pack smt_mode vcpus per vcore.
3037 		 */
3038 		if (smt_mode > threads_per_subcore)
3039 			return -EINVAL;
3040 	} else {
3041 		/*
3042 		 * On POWER9, the threading mode is "loose",
3043 		 * so each vcpu gets its own vcore.
3044 		 */
3045 		esmt = smt_mode;
3046 		smt_mode = 1;
3047 	}
3048 	mutex_lock(&kvm->lock);
3049 	err = -EBUSY;
3050 	if (!kvm->arch.online_vcores) {
3051 		kvm->arch.smt_mode = smt_mode;
3052 		kvm->arch.emul_smt_mode = esmt;
3053 		err = 0;
3054 	}
3055 	mutex_unlock(&kvm->lock);
3056 
3057 	return err;
3058 }
3059 
3060 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
3061 {
3062 	if (vpa->pinned_addr)
3063 		kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
3064 					vpa->dirty);
3065 }
3066 
3067 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
3068 {
3069 	spin_lock(&vcpu->arch.vpa_update_lock);
3070 	unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
3071 	unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
3072 	unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
3073 	spin_unlock(&vcpu->arch.vpa_update_lock);
3074 }
3075 
3076 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
3077 {
3078 	/* Indicate we want to get back into the guest */
3079 	return 1;
3080 }
3081 
3082 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
3083 {
3084 	unsigned long dec_nsec, now;
3085 
3086 	now = get_tb();
3087 	if (now > kvmppc_dec_expires_host_tb(vcpu)) {
3088 		/* decrementer has already gone negative */
3089 		kvmppc_core_queue_dec(vcpu);
3090 		kvmppc_core_prepare_to_enter(vcpu);
3091 		return;
3092 	}
3093 	dec_nsec = tb_to_ns(kvmppc_dec_expires_host_tb(vcpu) - now);
3094 	hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
3095 	vcpu->arch.timer_running = 1;
3096 }
3097 
3098 extern int __kvmppc_vcore_entry(void);
3099 
3100 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
3101 				   struct kvm_vcpu *vcpu, u64 tb)
3102 {
3103 	u64 now;
3104 
3105 	if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3106 		return;
3107 	spin_lock_irq(&vcpu->arch.tbacct_lock);
3108 	now = tb;
3109 	vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
3110 		vcpu->arch.stolen_logged;
3111 	vcpu->arch.busy_preempt = now;
3112 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
3113 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
3114 	--vc->n_runnable;
3115 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
3116 }
3117 
3118 static int kvmppc_grab_hwthread(int cpu)
3119 {
3120 	struct paca_struct *tpaca;
3121 	long timeout = 10000;
3122 
3123 	tpaca = paca_ptrs[cpu];
3124 
3125 	/* Ensure the thread won't go into the kernel if it wakes */
3126 	tpaca->kvm_hstate.kvm_vcpu = NULL;
3127 	tpaca->kvm_hstate.kvm_vcore = NULL;
3128 	tpaca->kvm_hstate.napping = 0;
3129 	smp_wmb();
3130 	tpaca->kvm_hstate.hwthread_req = 1;
3131 
3132 	/*
3133 	 * If the thread is already executing in the kernel (e.g. handling
3134 	 * a stray interrupt), wait for it to get back to nap mode.
3135 	 * The smp_mb() is to ensure that our setting of hwthread_req
3136 	 * is visible before we look at hwthread_state, so if this
3137 	 * races with the code at system_reset_pSeries and the thread
3138 	 * misses our setting of hwthread_req, we are sure to see its
3139 	 * setting of hwthread_state, and vice versa.
3140 	 */
3141 	smp_mb();
3142 	while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
3143 		if (--timeout <= 0) {
3144 			pr_err("KVM: couldn't grab cpu %d\n", cpu);
3145 			return -EBUSY;
3146 		}
3147 		udelay(1);
3148 	}
3149 	return 0;
3150 }
3151 
3152 static void kvmppc_release_hwthread(int cpu)
3153 {
3154 	struct paca_struct *tpaca;
3155 
3156 	tpaca = paca_ptrs[cpu];
3157 	tpaca->kvm_hstate.hwthread_req = 0;
3158 	tpaca->kvm_hstate.kvm_vcpu = NULL;
3159 	tpaca->kvm_hstate.kvm_vcore = NULL;
3160 	tpaca->kvm_hstate.kvm_split_mode = NULL;
3161 }
3162 
3163 static DEFINE_PER_CPU(struct kvm *, cpu_in_guest);
3164 
3165 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
3166 {
3167 	struct kvm_nested_guest *nested = vcpu->arch.nested;
3168 	cpumask_t *need_tlb_flush;
3169 	int i;
3170 
3171 	if (nested)
3172 		need_tlb_flush = &nested->need_tlb_flush;
3173 	else
3174 		need_tlb_flush = &kvm->arch.need_tlb_flush;
3175 
3176 	cpu = cpu_first_tlb_thread_sibling(cpu);
3177 	for (i = cpu; i <= cpu_last_tlb_thread_sibling(cpu);
3178 					i += cpu_tlb_thread_sibling_step())
3179 		cpumask_set_cpu(i, need_tlb_flush);
3180 
3181 	/*
3182 	 * Make sure setting of bit in need_tlb_flush precedes testing of
3183 	 * cpu_in_guest. The matching barrier on the other side is hwsync
3184 	 * when switching to guest MMU mode, which happens between
3185 	 * cpu_in_guest being set to the guest kvm, and need_tlb_flush bit
3186 	 * being tested.
3187 	 */
3188 	smp_mb();
3189 
3190 	for (i = cpu; i <= cpu_last_tlb_thread_sibling(cpu);
3191 					i += cpu_tlb_thread_sibling_step()) {
3192 		struct kvm *running = *per_cpu_ptr(&cpu_in_guest, i);
3193 
3194 		if (running == kvm)
3195 			smp_call_function_single(i, do_nothing, NULL, 1);
3196 	}
3197 }
3198 
3199 static void do_migrate_away_vcpu(void *arg)
3200 {
3201 	struct kvm_vcpu *vcpu = arg;
3202 	struct kvm *kvm = vcpu->kvm;
3203 
3204 	/*
3205 	 * If the guest has GTSE, it may execute tlbie, so do a eieio; tlbsync;
3206 	 * ptesync sequence on the old CPU before migrating to a new one, in
3207 	 * case we interrupted the guest between a tlbie ; eieio ;
3208 	 * tlbsync; ptesync sequence.
3209 	 *
3210 	 * Otherwise, ptesync is sufficient for ordering tlbiel sequences.
3211 	 */
3212 	if (kvm->arch.lpcr & LPCR_GTSE)
3213 		asm volatile("eieio; tlbsync; ptesync");
3214 	else
3215 		asm volatile("ptesync");
3216 }
3217 
3218 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
3219 {
3220 	struct kvm_nested_guest *nested = vcpu->arch.nested;
3221 	struct kvm *kvm = vcpu->kvm;
3222 	int prev_cpu;
3223 
3224 	if (!cpu_has_feature(CPU_FTR_HVMODE))
3225 		return;
3226 
3227 	if (nested)
3228 		prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
3229 	else
3230 		prev_cpu = vcpu->arch.prev_cpu;
3231 
3232 	/*
3233 	 * With radix, the guest can do TLB invalidations itself,
3234 	 * and it could choose to use the local form (tlbiel) if
3235 	 * it is invalidating a translation that has only ever been
3236 	 * used on one vcpu.  However, that doesn't mean it has
3237 	 * only ever been used on one physical cpu, since vcpus
3238 	 * can move around between pcpus.  To cope with this, when
3239 	 * a vcpu moves from one pcpu to another, we need to tell
3240 	 * any vcpus running on the same core as this vcpu previously
3241 	 * ran to flush the TLB.
3242 	 */
3243 	if (prev_cpu != pcpu) {
3244 		if (prev_cpu >= 0) {
3245 			if (cpu_first_tlb_thread_sibling(prev_cpu) !=
3246 			    cpu_first_tlb_thread_sibling(pcpu))
3247 				radix_flush_cpu(kvm, prev_cpu, vcpu);
3248 
3249 			smp_call_function_single(prev_cpu,
3250 					do_migrate_away_vcpu, vcpu, 1);
3251 		}
3252 		if (nested)
3253 			nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
3254 		else
3255 			vcpu->arch.prev_cpu = pcpu;
3256 	}
3257 }
3258 
3259 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
3260 {
3261 	int cpu;
3262 	struct paca_struct *tpaca;
3263 
3264 	cpu = vc->pcpu;
3265 	if (vcpu) {
3266 		if (vcpu->arch.timer_running) {
3267 			hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
3268 			vcpu->arch.timer_running = 0;
3269 		}
3270 		cpu += vcpu->arch.ptid;
3271 		vcpu->cpu = vc->pcpu;
3272 		vcpu->arch.thread_cpu = cpu;
3273 	}
3274 	tpaca = paca_ptrs[cpu];
3275 	tpaca->kvm_hstate.kvm_vcpu = vcpu;
3276 	tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
3277 	tpaca->kvm_hstate.fake_suspend = 0;
3278 	/* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
3279 	smp_wmb();
3280 	tpaca->kvm_hstate.kvm_vcore = vc;
3281 	if (cpu != smp_processor_id())
3282 		kvmppc_ipi_thread(cpu);
3283 }
3284 
3285 static void kvmppc_wait_for_nap(int n_threads)
3286 {
3287 	int cpu = smp_processor_id();
3288 	int i, loops;
3289 
3290 	if (n_threads <= 1)
3291 		return;
3292 	for (loops = 0; loops < 1000000; ++loops) {
3293 		/*
3294 		 * Check if all threads are finished.
3295 		 * We set the vcore pointer when starting a thread
3296 		 * and the thread clears it when finished, so we look
3297 		 * for any threads that still have a non-NULL vcore ptr.
3298 		 */
3299 		for (i = 1; i < n_threads; ++i)
3300 			if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
3301 				break;
3302 		if (i == n_threads) {
3303 			HMT_medium();
3304 			return;
3305 		}
3306 		HMT_low();
3307 	}
3308 	HMT_medium();
3309 	for (i = 1; i < n_threads; ++i)
3310 		if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
3311 			pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
3312 }
3313 
3314 /*
3315  * Check that we are on thread 0 and that any other threads in
3316  * this core are off-line.  Then grab the threads so they can't
3317  * enter the kernel.
3318  */
3319 static int on_primary_thread(void)
3320 {
3321 	int cpu = smp_processor_id();
3322 	int thr;
3323 
3324 	/* Are we on a primary subcore? */
3325 	if (cpu_thread_in_subcore(cpu))
3326 		return 0;
3327 
3328 	thr = 0;
3329 	while (++thr < threads_per_subcore)
3330 		if (cpu_online(cpu + thr))
3331 			return 0;
3332 
3333 	/* Grab all hw threads so they can't go into the kernel */
3334 	for (thr = 1; thr < threads_per_subcore; ++thr) {
3335 		if (kvmppc_grab_hwthread(cpu + thr)) {
3336 			/* Couldn't grab one; let the others go */
3337 			do {
3338 				kvmppc_release_hwthread(cpu + thr);
3339 			} while (--thr > 0);
3340 			return 0;
3341 		}
3342 	}
3343 	return 1;
3344 }
3345 
3346 /*
3347  * A list of virtual cores for each physical CPU.
3348  * These are vcores that could run but their runner VCPU tasks are
3349  * (or may be) preempted.
3350  */
3351 struct preempted_vcore_list {
3352 	struct list_head	list;
3353 	spinlock_t		lock;
3354 };
3355 
3356 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
3357 
3358 static void init_vcore_lists(void)
3359 {
3360 	int cpu;
3361 
3362 	for_each_possible_cpu(cpu) {
3363 		struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
3364 		spin_lock_init(&lp->lock);
3365 		INIT_LIST_HEAD(&lp->list);
3366 	}
3367 }
3368 
3369 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
3370 {
3371 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
3372 
3373 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
3374 
3375 	vc->vcore_state = VCORE_PREEMPT;
3376 	vc->pcpu = smp_processor_id();
3377 	if (vc->num_threads < threads_per_vcore(vc->kvm)) {
3378 		spin_lock(&lp->lock);
3379 		list_add_tail(&vc->preempt_list, &lp->list);
3380 		spin_unlock(&lp->lock);
3381 	}
3382 
3383 	/* Start accumulating stolen time */
3384 	kvmppc_core_start_stolen(vc, mftb());
3385 }
3386 
3387 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
3388 {
3389 	struct preempted_vcore_list *lp;
3390 
3391 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
3392 
3393 	kvmppc_core_end_stolen(vc, mftb());
3394 	if (!list_empty(&vc->preempt_list)) {
3395 		lp = &per_cpu(preempted_vcores, vc->pcpu);
3396 		spin_lock(&lp->lock);
3397 		list_del_init(&vc->preempt_list);
3398 		spin_unlock(&lp->lock);
3399 	}
3400 	vc->vcore_state = VCORE_INACTIVE;
3401 }
3402 
3403 /*
3404  * This stores information about the virtual cores currently
3405  * assigned to a physical core.
3406  */
3407 struct core_info {
3408 	int		n_subcores;
3409 	int		max_subcore_threads;
3410 	int		total_threads;
3411 	int		subcore_threads[MAX_SUBCORES];
3412 	struct kvmppc_vcore *vc[MAX_SUBCORES];
3413 };
3414 
3415 /*
3416  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
3417  * respectively in 2-way micro-threading (split-core) mode on POWER8.
3418  */
3419 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
3420 
3421 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
3422 {
3423 	memset(cip, 0, sizeof(*cip));
3424 	cip->n_subcores = 1;
3425 	cip->max_subcore_threads = vc->num_threads;
3426 	cip->total_threads = vc->num_threads;
3427 	cip->subcore_threads[0] = vc->num_threads;
3428 	cip->vc[0] = vc;
3429 }
3430 
3431 static bool subcore_config_ok(int n_subcores, int n_threads)
3432 {
3433 	/*
3434 	 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
3435 	 * split-core mode, with one thread per subcore.
3436 	 */
3437 	if (cpu_has_feature(CPU_FTR_ARCH_300))
3438 		return n_subcores <= 4 && n_threads == 1;
3439 
3440 	/* On POWER8, can only dynamically split if unsplit to begin with */
3441 	if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
3442 		return false;
3443 	if (n_subcores > MAX_SUBCORES)
3444 		return false;
3445 	if (n_subcores > 1) {
3446 		if (!(dynamic_mt_modes & 2))
3447 			n_subcores = 4;
3448 		if (n_subcores > 2 && !(dynamic_mt_modes & 4))
3449 			return false;
3450 	}
3451 
3452 	return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
3453 }
3454 
3455 static void init_vcore_to_run(struct kvmppc_vcore *vc)
3456 {
3457 	vc->entry_exit_map = 0;
3458 	vc->in_guest = 0;
3459 	vc->napping_threads = 0;
3460 	vc->conferring_threads = 0;
3461 	vc->tb_offset_applied = 0;
3462 }
3463 
3464 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
3465 {
3466 	int n_threads = vc->num_threads;
3467 	int sub;
3468 
3469 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
3470 		return false;
3471 
3472 	/* In one_vm_per_core mode, require all vcores to be from the same vm */
3473 	if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
3474 		return false;
3475 
3476 	if (n_threads < cip->max_subcore_threads)
3477 		n_threads = cip->max_subcore_threads;
3478 	if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
3479 		return false;
3480 	cip->max_subcore_threads = n_threads;
3481 
3482 	sub = cip->n_subcores;
3483 	++cip->n_subcores;
3484 	cip->total_threads += vc->num_threads;
3485 	cip->subcore_threads[sub] = vc->num_threads;
3486 	cip->vc[sub] = vc;
3487 	init_vcore_to_run(vc);
3488 	list_del_init(&vc->preempt_list);
3489 
3490 	return true;
3491 }
3492 
3493 /*
3494  * Work out whether it is possible to piggyback the execution of
3495  * vcore *pvc onto the execution of the other vcores described in *cip.
3496  */
3497 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
3498 			  int target_threads)
3499 {
3500 	if (cip->total_threads + pvc->num_threads > target_threads)
3501 		return false;
3502 
3503 	return can_dynamic_split(pvc, cip);
3504 }
3505 
3506 static void prepare_threads(struct kvmppc_vcore *vc)
3507 {
3508 	int i;
3509 	struct kvm_vcpu *vcpu;
3510 
3511 	for_each_runnable_thread(i, vcpu, vc) {
3512 		if (signal_pending(vcpu->arch.run_task))
3513 			vcpu->arch.ret = -EINTR;
3514 		else if (vcpu->arch.vpa.update_pending ||
3515 			 vcpu->arch.slb_shadow.update_pending ||
3516 			 vcpu->arch.dtl.update_pending)
3517 			vcpu->arch.ret = RESUME_GUEST;
3518 		else
3519 			continue;
3520 		kvmppc_remove_runnable(vc, vcpu, mftb());
3521 		wake_up(&vcpu->arch.cpu_run);
3522 	}
3523 }
3524 
3525 static void collect_piggybacks(struct core_info *cip, int target_threads)
3526 {
3527 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
3528 	struct kvmppc_vcore *pvc, *vcnext;
3529 
3530 	spin_lock(&lp->lock);
3531 	list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
3532 		if (!spin_trylock(&pvc->lock))
3533 			continue;
3534 		prepare_threads(pvc);
3535 		if (!pvc->n_runnable || !pvc->kvm->arch.mmu_ready) {
3536 			list_del_init(&pvc->preempt_list);
3537 			if (pvc->runner == NULL) {
3538 				pvc->vcore_state = VCORE_INACTIVE;
3539 				kvmppc_core_end_stolen(pvc, mftb());
3540 			}
3541 			spin_unlock(&pvc->lock);
3542 			continue;
3543 		}
3544 		if (!can_piggyback(pvc, cip, target_threads)) {
3545 			spin_unlock(&pvc->lock);
3546 			continue;
3547 		}
3548 		kvmppc_core_end_stolen(pvc, mftb());
3549 		pvc->vcore_state = VCORE_PIGGYBACK;
3550 		if (cip->total_threads >= target_threads)
3551 			break;
3552 	}
3553 	spin_unlock(&lp->lock);
3554 }
3555 
3556 static bool recheck_signals_and_mmu(struct core_info *cip)
3557 {
3558 	int sub, i;
3559 	struct kvm_vcpu *vcpu;
3560 	struct kvmppc_vcore *vc;
3561 
3562 	for (sub = 0; sub < cip->n_subcores; ++sub) {
3563 		vc = cip->vc[sub];
3564 		if (!vc->kvm->arch.mmu_ready)
3565 			return true;
3566 		for_each_runnable_thread(i, vcpu, vc)
3567 			if (signal_pending(vcpu->arch.run_task))
3568 				return true;
3569 	}
3570 	return false;
3571 }
3572 
3573 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
3574 {
3575 	int still_running = 0, i;
3576 	u64 now;
3577 	long ret;
3578 	struct kvm_vcpu *vcpu;
3579 
3580 	spin_lock(&vc->lock);
3581 	now = get_tb();
3582 	for_each_runnable_thread(i, vcpu, vc) {
3583 		/*
3584 		 * It's safe to unlock the vcore in the loop here, because
3585 		 * for_each_runnable_thread() is safe against removal of
3586 		 * the vcpu, and the vcore state is VCORE_EXITING here,
3587 		 * so any vcpus becoming runnable will have their arch.trap
3588 		 * set to zero and can't actually run in the guest.
3589 		 */
3590 		spin_unlock(&vc->lock);
3591 		/* cancel pending dec exception if dec is positive */
3592 		if (now < kvmppc_dec_expires_host_tb(vcpu) &&
3593 		    kvmppc_core_pending_dec(vcpu))
3594 			kvmppc_core_dequeue_dec(vcpu);
3595 
3596 		trace_kvm_guest_exit(vcpu);
3597 
3598 		ret = RESUME_GUEST;
3599 		if (vcpu->arch.trap)
3600 			ret = kvmppc_handle_exit_hv(vcpu,
3601 						    vcpu->arch.run_task);
3602 
3603 		vcpu->arch.ret = ret;
3604 		vcpu->arch.trap = 0;
3605 
3606 		spin_lock(&vc->lock);
3607 		if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
3608 			if (vcpu->arch.pending_exceptions)
3609 				kvmppc_core_prepare_to_enter(vcpu);
3610 			if (vcpu->arch.ceded)
3611 				kvmppc_set_timer(vcpu);
3612 			else
3613 				++still_running;
3614 		} else {
3615 			kvmppc_remove_runnable(vc, vcpu, mftb());
3616 			wake_up(&vcpu->arch.cpu_run);
3617 		}
3618 	}
3619 	if (!is_master) {
3620 		if (still_running > 0) {
3621 			kvmppc_vcore_preempt(vc);
3622 		} else if (vc->runner) {
3623 			vc->vcore_state = VCORE_PREEMPT;
3624 			kvmppc_core_start_stolen(vc, mftb());
3625 		} else {
3626 			vc->vcore_state = VCORE_INACTIVE;
3627 		}
3628 		if (vc->n_runnable > 0 && vc->runner == NULL) {
3629 			/* make sure there's a candidate runner awake */
3630 			i = -1;
3631 			vcpu = next_runnable_thread(vc, &i);
3632 			wake_up(&vcpu->arch.cpu_run);
3633 		}
3634 	}
3635 	spin_unlock(&vc->lock);
3636 }
3637 
3638 /*
3639  * Clear core from the list of active host cores as we are about to
3640  * enter the guest. Only do this if it is the primary thread of the
3641  * core (not if a subcore) that is entering the guest.
3642  */
3643 static inline int kvmppc_clear_host_core(unsigned int cpu)
3644 {
3645 	int core;
3646 
3647 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3648 		return 0;
3649 	/*
3650 	 * Memory barrier can be omitted here as we will do a smp_wmb()
3651 	 * later in kvmppc_start_thread and we need ensure that state is
3652 	 * visible to other CPUs only after we enter guest.
3653 	 */
3654 	core = cpu >> threads_shift;
3655 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
3656 	return 0;
3657 }
3658 
3659 /*
3660  * Advertise this core as an active host core since we exited the guest
3661  * Only need to do this if it is the primary thread of the core that is
3662  * exiting.
3663  */
3664 static inline int kvmppc_set_host_core(unsigned int cpu)
3665 {
3666 	int core;
3667 
3668 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3669 		return 0;
3670 
3671 	/*
3672 	 * Memory barrier can be omitted here because we do a spin_unlock
3673 	 * immediately after this which provides the memory barrier.
3674 	 */
3675 	core = cpu >> threads_shift;
3676 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
3677 	return 0;
3678 }
3679 
3680 static void set_irq_happened(int trap)
3681 {
3682 	switch (trap) {
3683 	case BOOK3S_INTERRUPT_EXTERNAL:
3684 		local_paca->irq_happened |= PACA_IRQ_EE;
3685 		break;
3686 	case BOOK3S_INTERRUPT_H_DOORBELL:
3687 		local_paca->irq_happened |= PACA_IRQ_DBELL;
3688 		break;
3689 	case BOOK3S_INTERRUPT_HMI:
3690 		local_paca->irq_happened |= PACA_IRQ_HMI;
3691 		break;
3692 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
3693 		replay_system_reset();
3694 		break;
3695 	}
3696 }
3697 
3698 /*
3699  * Run a set of guest threads on a physical core.
3700  * Called with vc->lock held.
3701  */
3702 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
3703 {
3704 	struct kvm_vcpu *vcpu;
3705 	int i;
3706 	int srcu_idx;
3707 	struct core_info core_info;
3708 	struct kvmppc_vcore *pvc;
3709 	struct kvm_split_mode split_info, *sip;
3710 	int split, subcore_size, active;
3711 	int sub;
3712 	bool thr0_done;
3713 	unsigned long cmd_bit, stat_bit;
3714 	int pcpu, thr;
3715 	int target_threads;
3716 	int controlled_threads;
3717 	int trap;
3718 	bool is_power8;
3719 
3720 	if (WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300)))
3721 		return;
3722 
3723 	/*
3724 	 * Remove from the list any threads that have a signal pending
3725 	 * or need a VPA update done
3726 	 */
3727 	prepare_threads(vc);
3728 
3729 	/* if the runner is no longer runnable, let the caller pick a new one */
3730 	if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3731 		return;
3732 
3733 	/*
3734 	 * Initialize *vc.
3735 	 */
3736 	init_vcore_to_run(vc);
3737 	vc->preempt_tb = TB_NIL;
3738 
3739 	/*
3740 	 * Number of threads that we will be controlling: the same as
3741 	 * the number of threads per subcore, except on POWER9,
3742 	 * where it's 1 because the threads are (mostly) independent.
3743 	 */
3744 	controlled_threads = threads_per_vcore(vc->kvm);
3745 
3746 	/*
3747 	 * Make sure we are running on primary threads, and that secondary
3748 	 * threads are offline.  Also check if the number of threads in this
3749 	 * guest are greater than the current system threads per guest.
3750 	 */
3751 	if ((controlled_threads > 1) &&
3752 	    ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
3753 		for_each_runnable_thread(i, vcpu, vc) {
3754 			vcpu->arch.ret = -EBUSY;
3755 			kvmppc_remove_runnable(vc, vcpu, mftb());
3756 			wake_up(&vcpu->arch.cpu_run);
3757 		}
3758 		goto out;
3759 	}
3760 
3761 	/*
3762 	 * See if we could run any other vcores on the physical core
3763 	 * along with this one.
3764 	 */
3765 	init_core_info(&core_info, vc);
3766 	pcpu = smp_processor_id();
3767 	target_threads = controlled_threads;
3768 	if (target_smt_mode && target_smt_mode < target_threads)
3769 		target_threads = target_smt_mode;
3770 	if (vc->num_threads < target_threads)
3771 		collect_piggybacks(&core_info, target_threads);
3772 
3773 	/*
3774 	 * Hard-disable interrupts, and check resched flag and signals.
3775 	 * If we need to reschedule or deliver a signal, clean up
3776 	 * and return without going into the guest(s).
3777 	 * If the mmu_ready flag has been cleared, don't go into the
3778 	 * guest because that means a HPT resize operation is in progress.
3779 	 */
3780 	local_irq_disable();
3781 	hard_irq_disable();
3782 	if (lazy_irq_pending() || need_resched() ||
3783 	    recheck_signals_and_mmu(&core_info)) {
3784 		local_irq_enable();
3785 		vc->vcore_state = VCORE_INACTIVE;
3786 		/* Unlock all except the primary vcore */
3787 		for (sub = 1; sub < core_info.n_subcores; ++sub) {
3788 			pvc = core_info.vc[sub];
3789 			/* Put back on to the preempted vcores list */
3790 			kvmppc_vcore_preempt(pvc);
3791 			spin_unlock(&pvc->lock);
3792 		}
3793 		for (i = 0; i < controlled_threads; ++i)
3794 			kvmppc_release_hwthread(pcpu + i);
3795 		return;
3796 	}
3797 
3798 	kvmppc_clear_host_core(pcpu);
3799 
3800 	/* Decide on micro-threading (split-core) mode */
3801 	subcore_size = threads_per_subcore;
3802 	cmd_bit = stat_bit = 0;
3803 	split = core_info.n_subcores;
3804 	sip = NULL;
3805 	is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S);
3806 
3807 	if (split > 1) {
3808 		sip = &split_info;
3809 		memset(&split_info, 0, sizeof(split_info));
3810 		for (sub = 0; sub < core_info.n_subcores; ++sub)
3811 			split_info.vc[sub] = core_info.vc[sub];
3812 
3813 		if (is_power8) {
3814 			if (split == 2 && (dynamic_mt_modes & 2)) {
3815 				cmd_bit = HID0_POWER8_1TO2LPAR;
3816 				stat_bit = HID0_POWER8_2LPARMODE;
3817 			} else {
3818 				split = 4;
3819 				cmd_bit = HID0_POWER8_1TO4LPAR;
3820 				stat_bit = HID0_POWER8_4LPARMODE;
3821 			}
3822 			subcore_size = MAX_SMT_THREADS / split;
3823 			split_info.rpr = mfspr(SPRN_RPR);
3824 			split_info.pmmar = mfspr(SPRN_PMMAR);
3825 			split_info.ldbar = mfspr(SPRN_LDBAR);
3826 			split_info.subcore_size = subcore_size;
3827 		} else {
3828 			split_info.subcore_size = 1;
3829 		}
3830 
3831 		/* order writes to split_info before kvm_split_mode pointer */
3832 		smp_wmb();
3833 	}
3834 
3835 	for (thr = 0; thr < controlled_threads; ++thr) {
3836 		struct paca_struct *paca = paca_ptrs[pcpu + thr];
3837 
3838 		paca->kvm_hstate.napping = 0;
3839 		paca->kvm_hstate.kvm_split_mode = sip;
3840 	}
3841 
3842 	/* Initiate micro-threading (split-core) on POWER8 if required */
3843 	if (cmd_bit) {
3844 		unsigned long hid0 = mfspr(SPRN_HID0);
3845 
3846 		hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3847 		mb();
3848 		mtspr(SPRN_HID0, hid0);
3849 		isync();
3850 		for (;;) {
3851 			hid0 = mfspr(SPRN_HID0);
3852 			if (hid0 & stat_bit)
3853 				break;
3854 			cpu_relax();
3855 		}
3856 	}
3857 
3858 	/*
3859 	 * On POWER8, set RWMR register.
3860 	 * Since it only affects PURR and SPURR, it doesn't affect
3861 	 * the host, so we don't save/restore the host value.
3862 	 */
3863 	if (is_power8) {
3864 		unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3865 		int n_online = atomic_read(&vc->online_count);
3866 
3867 		/*
3868 		 * Use the 8-thread value if we're doing split-core
3869 		 * or if the vcore's online count looks bogus.
3870 		 */
3871 		if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3872 		    n_online >= 1 && n_online <= MAX_SMT_THREADS)
3873 			rwmr_val = p8_rwmr_values[n_online];
3874 		mtspr(SPRN_RWMR, rwmr_val);
3875 	}
3876 
3877 	/* Start all the threads */
3878 	active = 0;
3879 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
3880 		thr = is_power8 ? subcore_thread_map[sub] : sub;
3881 		thr0_done = false;
3882 		active |= 1 << thr;
3883 		pvc = core_info.vc[sub];
3884 		pvc->pcpu = pcpu + thr;
3885 		for_each_runnable_thread(i, vcpu, pvc) {
3886 			/*
3887 			 * XXX: is kvmppc_start_thread called too late here?
3888 			 * It updates vcpu->cpu and vcpu->arch.thread_cpu
3889 			 * which are used by kvmppc_fast_vcpu_kick_hv(), but
3890 			 * kick is called after new exceptions become available
3891 			 * and exceptions are checked earlier than here, by
3892 			 * kvmppc_core_prepare_to_enter.
3893 			 */
3894 			kvmppc_start_thread(vcpu, pvc);
3895 			kvmppc_update_vpa_dispatch(vcpu, pvc);
3896 			trace_kvm_guest_enter(vcpu);
3897 			if (!vcpu->arch.ptid)
3898 				thr0_done = true;
3899 			active |= 1 << (thr + vcpu->arch.ptid);
3900 		}
3901 		/*
3902 		 * We need to start the first thread of each subcore
3903 		 * even if it doesn't have a vcpu.
3904 		 */
3905 		if (!thr0_done)
3906 			kvmppc_start_thread(NULL, pvc);
3907 	}
3908 
3909 	/*
3910 	 * Ensure that split_info.do_nap is set after setting
3911 	 * the vcore pointer in the PACA of the secondaries.
3912 	 */
3913 	smp_mb();
3914 
3915 	/*
3916 	 * When doing micro-threading, poke the inactive threads as well.
3917 	 * This gets them to the nap instruction after kvm_do_nap,
3918 	 * which reduces the time taken to unsplit later.
3919 	 */
3920 	if (cmd_bit) {
3921 		split_info.do_nap = 1;	/* ask secondaries to nap when done */
3922 		for (thr = 1; thr < threads_per_subcore; ++thr)
3923 			if (!(active & (1 << thr)))
3924 				kvmppc_ipi_thread(pcpu + thr);
3925 	}
3926 
3927 	vc->vcore_state = VCORE_RUNNING;
3928 	preempt_disable();
3929 
3930 	trace_kvmppc_run_core(vc, 0);
3931 
3932 	for (sub = 0; sub < core_info.n_subcores; ++sub)
3933 		spin_unlock(&core_info.vc[sub]->lock);
3934 
3935 	guest_timing_enter_irqoff();
3936 
3937 	srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3938 
3939 	guest_state_enter_irqoff();
3940 	this_cpu_disable_ftrace();
3941 
3942 	trap = __kvmppc_vcore_entry();
3943 
3944 	this_cpu_enable_ftrace();
3945 	guest_state_exit_irqoff();
3946 
3947 	srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3948 
3949 	set_irq_happened(trap);
3950 
3951 	spin_lock(&vc->lock);
3952 	/* prevent other vcpu threads from doing kvmppc_start_thread() now */
3953 	vc->vcore_state = VCORE_EXITING;
3954 
3955 	/* wait for secondary threads to finish writing their state to memory */
3956 	kvmppc_wait_for_nap(controlled_threads);
3957 
3958 	/* Return to whole-core mode if we split the core earlier */
3959 	if (cmd_bit) {
3960 		unsigned long hid0 = mfspr(SPRN_HID0);
3961 		unsigned long loops = 0;
3962 
3963 		hid0 &= ~HID0_POWER8_DYNLPARDIS;
3964 		stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3965 		mb();
3966 		mtspr(SPRN_HID0, hid0);
3967 		isync();
3968 		for (;;) {
3969 			hid0 = mfspr(SPRN_HID0);
3970 			if (!(hid0 & stat_bit))
3971 				break;
3972 			cpu_relax();
3973 			++loops;
3974 		}
3975 		split_info.do_nap = 0;
3976 	}
3977 
3978 	kvmppc_set_host_core(pcpu);
3979 
3980 	if (!vtime_accounting_enabled_this_cpu()) {
3981 		local_irq_enable();
3982 		/*
3983 		 * Service IRQs here before guest_timing_exit_irqoff() so any
3984 		 * ticks that occurred while running the guest are accounted to
3985 		 * the guest. If vtime accounting is enabled, accounting uses
3986 		 * TB rather than ticks, so it can be done without enabling
3987 		 * interrupts here, which has the problem that it accounts
3988 		 * interrupt processing overhead to the host.
3989 		 */
3990 		local_irq_disable();
3991 	}
3992 	guest_timing_exit_irqoff();
3993 
3994 	local_irq_enable();
3995 
3996 	/* Let secondaries go back to the offline loop */
3997 	for (i = 0; i < controlled_threads; ++i) {
3998 		kvmppc_release_hwthread(pcpu + i);
3999 		if (sip && sip->napped[i])
4000 			kvmppc_ipi_thread(pcpu + i);
4001 	}
4002 
4003 	spin_unlock(&vc->lock);
4004 
4005 	/* make sure updates to secondary vcpu structs are visible now */
4006 	smp_mb();
4007 
4008 	preempt_enable();
4009 
4010 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
4011 		pvc = core_info.vc[sub];
4012 		post_guest_process(pvc, pvc == vc);
4013 	}
4014 
4015 	spin_lock(&vc->lock);
4016 
4017  out:
4018 	vc->vcore_state = VCORE_INACTIVE;
4019 	trace_kvmppc_run_core(vc, 1);
4020 }
4021 
4022 static inline bool hcall_is_xics(unsigned long req)
4023 {
4024 	return req == H_EOI || req == H_CPPR || req == H_IPI ||
4025 		req == H_IPOLL || req == H_XIRR || req == H_XIRR_X;
4026 }
4027 
4028 static void vcpu_vpa_increment_dispatch(struct kvm_vcpu *vcpu)
4029 {
4030 	struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
4031 	if (lp) {
4032 		u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
4033 		lp->yield_count = cpu_to_be32(yield_count);
4034 		vcpu->arch.vpa.dirty = 1;
4035 	}
4036 }
4037 
4038 /* call our hypervisor to load up HV regs and go */
4039 static int kvmhv_vcpu_entry_p9_nested(struct kvm_vcpu *vcpu, u64 time_limit, unsigned long lpcr, u64 *tb)
4040 {
4041 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
4042 	unsigned long host_psscr;
4043 	unsigned long msr;
4044 	struct hv_guest_state hvregs;
4045 	struct p9_host_os_sprs host_os_sprs;
4046 	s64 dec;
4047 	int trap;
4048 
4049 	msr = mfmsr();
4050 
4051 	save_p9_host_os_sprs(&host_os_sprs);
4052 
4053 	/*
4054 	 * We need to save and restore the guest visible part of the
4055 	 * psscr (i.e. using SPRN_PSSCR_PR) since the hypervisor
4056 	 * doesn't do this for us. Note only required if pseries since
4057 	 * this is done in kvmhv_vcpu_entry_p9() below otherwise.
4058 	 */
4059 	host_psscr = mfspr(SPRN_PSSCR_PR);
4060 
4061 	kvmppc_msr_hard_disable_set_facilities(vcpu, msr);
4062 	if (lazy_irq_pending())
4063 		return 0;
4064 
4065 	if (unlikely(load_vcpu_state(vcpu, &host_os_sprs)))
4066 		msr = mfmsr(); /* TM restore can update msr */
4067 
4068 	if (vcpu->arch.psscr != host_psscr)
4069 		mtspr(SPRN_PSSCR_PR, vcpu->arch.psscr);
4070 
4071 	kvmhv_save_hv_regs(vcpu, &hvregs);
4072 	hvregs.lpcr = lpcr;
4073 	hvregs.amor = ~0;
4074 	vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
4075 	hvregs.version = HV_GUEST_STATE_VERSION;
4076 	if (vcpu->arch.nested) {
4077 		hvregs.lpid = vcpu->arch.nested->shadow_lpid;
4078 		hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
4079 	} else {
4080 		hvregs.lpid = vcpu->kvm->arch.lpid;
4081 		hvregs.vcpu_token = vcpu->vcpu_id;
4082 	}
4083 	hvregs.hdec_expiry = time_limit;
4084 
4085 	/*
4086 	 * When setting DEC, we must always deal with irq_work_raise
4087 	 * via NMI vs setting DEC. The problem occurs right as we
4088 	 * switch into guest mode if a NMI hits and sets pending work
4089 	 * and sets DEC, then that will apply to the guest and not
4090 	 * bring us back to the host.
4091 	 *
4092 	 * irq_work_raise could check a flag (or possibly LPCR[HDICE]
4093 	 * for example) and set HDEC to 1? That wouldn't solve the
4094 	 * nested hv case which needs to abort the hcall or zero the
4095 	 * time limit.
4096 	 *
4097 	 * XXX: Another day's problem.
4098 	 */
4099 	mtspr(SPRN_DEC, kvmppc_dec_expires_host_tb(vcpu) - *tb);
4100 
4101 	mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
4102 	mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
4103 	switch_pmu_to_guest(vcpu, &host_os_sprs);
4104 	accumulate_time(vcpu, &vcpu->arch.in_guest);
4105 	trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
4106 				  __pa(&vcpu->arch.regs));
4107 	accumulate_time(vcpu, &vcpu->arch.guest_exit);
4108 	kvmhv_restore_hv_return_state(vcpu, &hvregs);
4109 	switch_pmu_to_host(vcpu, &host_os_sprs);
4110 	vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
4111 	vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
4112 	vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
4113 	vcpu->arch.psscr = mfspr(SPRN_PSSCR_PR);
4114 
4115 	store_vcpu_state(vcpu);
4116 
4117 	dec = mfspr(SPRN_DEC);
4118 	if (!(lpcr & LPCR_LD)) /* Sign extend if not using large decrementer */
4119 		dec = (s32) dec;
4120 	*tb = mftb();
4121 	vcpu->arch.dec_expires = dec + (*tb + vc->tb_offset);
4122 
4123 	timer_rearm_host_dec(*tb);
4124 
4125 	restore_p9_host_os_sprs(vcpu, &host_os_sprs);
4126 	if (vcpu->arch.psscr != host_psscr)
4127 		mtspr(SPRN_PSSCR_PR, host_psscr);
4128 
4129 	return trap;
4130 }
4131 
4132 /*
4133  * Guest entry for POWER9 and later CPUs.
4134  */
4135 static int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
4136 			 unsigned long lpcr, u64 *tb)
4137 {
4138 	struct kvm *kvm = vcpu->kvm;
4139 	struct kvm_nested_guest *nested = vcpu->arch.nested;
4140 	u64 next_timer;
4141 	int trap;
4142 
4143 	next_timer = timer_get_next_tb();
4144 	if (*tb >= next_timer)
4145 		return BOOK3S_INTERRUPT_HV_DECREMENTER;
4146 	if (next_timer < time_limit)
4147 		time_limit = next_timer;
4148 	else if (*tb >= time_limit) /* nested time limit */
4149 		return BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER;
4150 
4151 	vcpu->arch.ceded = 0;
4152 
4153 	vcpu_vpa_increment_dispatch(vcpu);
4154 
4155 	if (kvmhv_on_pseries()) {
4156 		trap = kvmhv_vcpu_entry_p9_nested(vcpu, time_limit, lpcr, tb);
4157 
4158 		/* H_CEDE has to be handled now, not later */
4159 		if (trap == BOOK3S_INTERRUPT_SYSCALL && !nested &&
4160 		    kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
4161 			kvmppc_cede(vcpu);
4162 			kvmppc_set_gpr(vcpu, 3, 0);
4163 			trap = 0;
4164 		}
4165 
4166 	} else if (nested) {
4167 		__this_cpu_write(cpu_in_guest, kvm);
4168 		trap = kvmhv_vcpu_entry_p9(vcpu, time_limit, lpcr, tb);
4169 		__this_cpu_write(cpu_in_guest, NULL);
4170 
4171 	} else {
4172 		kvmppc_xive_push_vcpu(vcpu);
4173 
4174 		__this_cpu_write(cpu_in_guest, kvm);
4175 		trap = kvmhv_vcpu_entry_p9(vcpu, time_limit, lpcr, tb);
4176 		__this_cpu_write(cpu_in_guest, NULL);
4177 
4178 		if (trap == BOOK3S_INTERRUPT_SYSCALL &&
4179 		    !(vcpu->arch.shregs.msr & MSR_PR)) {
4180 			unsigned long req = kvmppc_get_gpr(vcpu, 3);
4181 
4182 			/*
4183 			 * XIVE rearm and XICS hcalls must be handled
4184 			 * before xive context is pulled (is this
4185 			 * true?)
4186 			 */
4187 			if (req == H_CEDE) {
4188 				/* H_CEDE has to be handled now */
4189 				kvmppc_cede(vcpu);
4190 				if (!kvmppc_xive_rearm_escalation(vcpu)) {
4191 					/*
4192 					 * Pending escalation so abort
4193 					 * the cede.
4194 					 */
4195 					vcpu->arch.ceded = 0;
4196 				}
4197 				kvmppc_set_gpr(vcpu, 3, 0);
4198 				trap = 0;
4199 
4200 			} else if (req == H_ENTER_NESTED) {
4201 				/*
4202 				 * L2 should not run with the L1
4203 				 * context so rearm and pull it.
4204 				 */
4205 				if (!kvmppc_xive_rearm_escalation(vcpu)) {
4206 					/*
4207 					 * Pending escalation so abort
4208 					 * H_ENTER_NESTED.
4209 					 */
4210 					kvmppc_set_gpr(vcpu, 3, 0);
4211 					trap = 0;
4212 				}
4213 
4214 			} else if (hcall_is_xics(req)) {
4215 				int ret;
4216 
4217 				ret = kvmppc_xive_xics_hcall(vcpu, req);
4218 				if (ret != H_TOO_HARD) {
4219 					kvmppc_set_gpr(vcpu, 3, ret);
4220 					trap = 0;
4221 				}
4222 			}
4223 		}
4224 		kvmppc_xive_pull_vcpu(vcpu);
4225 
4226 		if (kvm_is_radix(kvm))
4227 			vcpu->arch.slb_max = 0;
4228 	}
4229 
4230 	vcpu_vpa_increment_dispatch(vcpu);
4231 
4232 	return trap;
4233 }
4234 
4235 /*
4236  * Wait for some other vcpu thread to execute us, and
4237  * wake us up when we need to handle something in the host.
4238  */
4239 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
4240 				 struct kvm_vcpu *vcpu, int wait_state)
4241 {
4242 	DEFINE_WAIT(wait);
4243 
4244 	prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
4245 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4246 		spin_unlock(&vc->lock);
4247 		schedule();
4248 		spin_lock(&vc->lock);
4249 	}
4250 	finish_wait(&vcpu->arch.cpu_run, &wait);
4251 }
4252 
4253 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
4254 {
4255 	if (!halt_poll_ns_grow)
4256 		return;
4257 
4258 	vc->halt_poll_ns *= halt_poll_ns_grow;
4259 	if (vc->halt_poll_ns < halt_poll_ns_grow_start)
4260 		vc->halt_poll_ns = halt_poll_ns_grow_start;
4261 }
4262 
4263 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
4264 {
4265 	if (halt_poll_ns_shrink == 0)
4266 		vc->halt_poll_ns = 0;
4267 	else
4268 		vc->halt_poll_ns /= halt_poll_ns_shrink;
4269 }
4270 
4271 #ifdef CONFIG_KVM_XICS
4272 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
4273 {
4274 	if (!xics_on_xive())
4275 		return false;
4276 	return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
4277 		vcpu->arch.xive_saved_state.cppr;
4278 }
4279 #else
4280 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
4281 {
4282 	return false;
4283 }
4284 #endif /* CONFIG_KVM_XICS */
4285 
4286 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
4287 {
4288 	if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
4289 	    kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
4290 		return true;
4291 
4292 	return false;
4293 }
4294 
4295 static bool kvmppc_vcpu_check_block(struct kvm_vcpu *vcpu)
4296 {
4297 	if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
4298 		return true;
4299 	return false;
4300 }
4301 
4302 /*
4303  * Check to see if any of the runnable vcpus on the vcore have pending
4304  * exceptions or are no longer ceded
4305  */
4306 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
4307 {
4308 	struct kvm_vcpu *vcpu;
4309 	int i;
4310 
4311 	for_each_runnable_thread(i, vcpu, vc) {
4312 		if (kvmppc_vcpu_check_block(vcpu))
4313 			return 1;
4314 	}
4315 
4316 	return 0;
4317 }
4318 
4319 /*
4320  * All the vcpus in this vcore are idle, so wait for a decrementer
4321  * or external interrupt to one of the vcpus.  vc->lock is held.
4322  */
4323 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
4324 {
4325 	ktime_t cur, start_poll, start_wait;
4326 	int do_sleep = 1;
4327 	u64 block_ns;
4328 
4329 	WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
4330 
4331 	/* Poll for pending exceptions and ceded state */
4332 	cur = start_poll = ktime_get();
4333 	if (vc->halt_poll_ns) {
4334 		ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
4335 		++vc->runner->stat.generic.halt_attempted_poll;
4336 
4337 		vc->vcore_state = VCORE_POLLING;
4338 		spin_unlock(&vc->lock);
4339 
4340 		do {
4341 			if (kvmppc_vcore_check_block(vc)) {
4342 				do_sleep = 0;
4343 				break;
4344 			}
4345 			cur = ktime_get();
4346 		} while (kvm_vcpu_can_poll(cur, stop));
4347 
4348 		spin_lock(&vc->lock);
4349 		vc->vcore_state = VCORE_INACTIVE;
4350 
4351 		if (!do_sleep) {
4352 			++vc->runner->stat.generic.halt_successful_poll;
4353 			goto out;
4354 		}
4355 	}
4356 
4357 	prepare_to_rcuwait(&vc->wait);
4358 	set_current_state(TASK_INTERRUPTIBLE);
4359 	if (kvmppc_vcore_check_block(vc)) {
4360 		finish_rcuwait(&vc->wait);
4361 		do_sleep = 0;
4362 		/* If we polled, count this as a successful poll */
4363 		if (vc->halt_poll_ns)
4364 			++vc->runner->stat.generic.halt_successful_poll;
4365 		goto out;
4366 	}
4367 
4368 	start_wait = ktime_get();
4369 
4370 	vc->vcore_state = VCORE_SLEEPING;
4371 	trace_kvmppc_vcore_blocked(vc->runner, 0);
4372 	spin_unlock(&vc->lock);
4373 	schedule();
4374 	finish_rcuwait(&vc->wait);
4375 	spin_lock(&vc->lock);
4376 	vc->vcore_state = VCORE_INACTIVE;
4377 	trace_kvmppc_vcore_blocked(vc->runner, 1);
4378 	++vc->runner->stat.halt_successful_wait;
4379 
4380 	cur = ktime_get();
4381 
4382 out:
4383 	block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
4384 
4385 	/* Attribute wait time */
4386 	if (do_sleep) {
4387 		vc->runner->stat.generic.halt_wait_ns +=
4388 			ktime_to_ns(cur) - ktime_to_ns(start_wait);
4389 		KVM_STATS_LOG_HIST_UPDATE(
4390 				vc->runner->stat.generic.halt_wait_hist,
4391 				ktime_to_ns(cur) - ktime_to_ns(start_wait));
4392 		/* Attribute failed poll time */
4393 		if (vc->halt_poll_ns) {
4394 			vc->runner->stat.generic.halt_poll_fail_ns +=
4395 				ktime_to_ns(start_wait) -
4396 				ktime_to_ns(start_poll);
4397 			KVM_STATS_LOG_HIST_UPDATE(
4398 				vc->runner->stat.generic.halt_poll_fail_hist,
4399 				ktime_to_ns(start_wait) -
4400 				ktime_to_ns(start_poll));
4401 		}
4402 	} else {
4403 		/* Attribute successful poll time */
4404 		if (vc->halt_poll_ns) {
4405 			vc->runner->stat.generic.halt_poll_success_ns +=
4406 				ktime_to_ns(cur) -
4407 				ktime_to_ns(start_poll);
4408 			KVM_STATS_LOG_HIST_UPDATE(
4409 				vc->runner->stat.generic.halt_poll_success_hist,
4410 				ktime_to_ns(cur) - ktime_to_ns(start_poll));
4411 		}
4412 	}
4413 
4414 	/* Adjust poll time */
4415 	if (halt_poll_ns) {
4416 		if (block_ns <= vc->halt_poll_ns)
4417 			;
4418 		/* We slept and blocked for longer than the max halt time */
4419 		else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
4420 			shrink_halt_poll_ns(vc);
4421 		/* We slept and our poll time is too small */
4422 		else if (vc->halt_poll_ns < halt_poll_ns &&
4423 				block_ns < halt_poll_ns)
4424 			grow_halt_poll_ns(vc);
4425 		if (vc->halt_poll_ns > halt_poll_ns)
4426 			vc->halt_poll_ns = halt_poll_ns;
4427 	} else
4428 		vc->halt_poll_ns = 0;
4429 
4430 	trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
4431 }
4432 
4433 /*
4434  * This never fails for a radix guest, as none of the operations it does
4435  * for a radix guest can fail or have a way to report failure.
4436  */
4437 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
4438 {
4439 	int r = 0;
4440 	struct kvm *kvm = vcpu->kvm;
4441 
4442 	mutex_lock(&kvm->arch.mmu_setup_lock);
4443 	if (!kvm->arch.mmu_ready) {
4444 		if (!kvm_is_radix(kvm))
4445 			r = kvmppc_hv_setup_htab_rma(vcpu);
4446 		if (!r) {
4447 			if (cpu_has_feature(CPU_FTR_ARCH_300))
4448 				kvmppc_setup_partition_table(kvm);
4449 			kvm->arch.mmu_ready = 1;
4450 		}
4451 	}
4452 	mutex_unlock(&kvm->arch.mmu_setup_lock);
4453 	return r;
4454 }
4455 
4456 static int kvmppc_run_vcpu(struct kvm_vcpu *vcpu)
4457 {
4458 	struct kvm_run *run = vcpu->run;
4459 	int n_ceded, i, r;
4460 	struct kvmppc_vcore *vc;
4461 	struct kvm_vcpu *v;
4462 
4463 	trace_kvmppc_run_vcpu_enter(vcpu);
4464 
4465 	run->exit_reason = 0;
4466 	vcpu->arch.ret = RESUME_GUEST;
4467 	vcpu->arch.trap = 0;
4468 	kvmppc_update_vpas(vcpu);
4469 
4470 	/*
4471 	 * Synchronize with other threads in this virtual core
4472 	 */
4473 	vc = vcpu->arch.vcore;
4474 	spin_lock(&vc->lock);
4475 	vcpu->arch.ceded = 0;
4476 	vcpu->arch.run_task = current;
4477 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4478 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4479 	vcpu->arch.busy_preempt = TB_NIL;
4480 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
4481 	++vc->n_runnable;
4482 
4483 	/*
4484 	 * This happens the first time this is called for a vcpu.
4485 	 * If the vcore is already running, we may be able to start
4486 	 * this thread straight away and have it join in.
4487 	 */
4488 	if (!signal_pending(current)) {
4489 		if ((vc->vcore_state == VCORE_PIGGYBACK ||
4490 		     vc->vcore_state == VCORE_RUNNING) &&
4491 			   !VCORE_IS_EXITING(vc)) {
4492 			kvmppc_update_vpa_dispatch(vcpu, vc);
4493 			kvmppc_start_thread(vcpu, vc);
4494 			trace_kvm_guest_enter(vcpu);
4495 		} else if (vc->vcore_state == VCORE_SLEEPING) {
4496 		        rcuwait_wake_up(&vc->wait);
4497 		}
4498 
4499 	}
4500 
4501 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4502 	       !signal_pending(current)) {
4503 		/* See if the MMU is ready to go */
4504 		if (!vcpu->kvm->arch.mmu_ready) {
4505 			spin_unlock(&vc->lock);
4506 			r = kvmhv_setup_mmu(vcpu);
4507 			spin_lock(&vc->lock);
4508 			if (r) {
4509 				run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4510 				run->fail_entry.
4511 					hardware_entry_failure_reason = 0;
4512 				vcpu->arch.ret = r;
4513 				break;
4514 			}
4515 		}
4516 
4517 		if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4518 			kvmppc_vcore_end_preempt(vc);
4519 
4520 		if (vc->vcore_state != VCORE_INACTIVE) {
4521 			kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
4522 			continue;
4523 		}
4524 		for_each_runnable_thread(i, v, vc) {
4525 			kvmppc_core_prepare_to_enter(v);
4526 			if (signal_pending(v->arch.run_task)) {
4527 				kvmppc_remove_runnable(vc, v, mftb());
4528 				v->stat.signal_exits++;
4529 				v->run->exit_reason = KVM_EXIT_INTR;
4530 				v->arch.ret = -EINTR;
4531 				wake_up(&v->arch.cpu_run);
4532 			}
4533 		}
4534 		if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
4535 			break;
4536 		n_ceded = 0;
4537 		for_each_runnable_thread(i, v, vc) {
4538 			if (!kvmppc_vcpu_woken(v))
4539 				n_ceded += v->arch.ceded;
4540 			else
4541 				v->arch.ceded = 0;
4542 		}
4543 		vc->runner = vcpu;
4544 		if (n_ceded == vc->n_runnable) {
4545 			kvmppc_vcore_blocked(vc);
4546 		} else if (need_resched()) {
4547 			kvmppc_vcore_preempt(vc);
4548 			/* Let something else run */
4549 			cond_resched_lock(&vc->lock);
4550 			if (vc->vcore_state == VCORE_PREEMPT)
4551 				kvmppc_vcore_end_preempt(vc);
4552 		} else {
4553 			kvmppc_run_core(vc);
4554 		}
4555 		vc->runner = NULL;
4556 	}
4557 
4558 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4559 	       (vc->vcore_state == VCORE_RUNNING ||
4560 		vc->vcore_state == VCORE_EXITING ||
4561 		vc->vcore_state == VCORE_PIGGYBACK))
4562 		kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
4563 
4564 	if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4565 		kvmppc_vcore_end_preempt(vc);
4566 
4567 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4568 		kvmppc_remove_runnable(vc, vcpu, mftb());
4569 		vcpu->stat.signal_exits++;
4570 		run->exit_reason = KVM_EXIT_INTR;
4571 		vcpu->arch.ret = -EINTR;
4572 	}
4573 
4574 	if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
4575 		/* Wake up some vcpu to run the core */
4576 		i = -1;
4577 		v = next_runnable_thread(vc, &i);
4578 		wake_up(&v->arch.cpu_run);
4579 	}
4580 
4581 	trace_kvmppc_run_vcpu_exit(vcpu);
4582 	spin_unlock(&vc->lock);
4583 	return vcpu->arch.ret;
4584 }
4585 
4586 int kvmhv_run_single_vcpu(struct kvm_vcpu *vcpu, u64 time_limit,
4587 			  unsigned long lpcr)
4588 {
4589 	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
4590 	struct kvm_run *run = vcpu->run;
4591 	int trap, r, pcpu;
4592 	int srcu_idx;
4593 	struct kvmppc_vcore *vc;
4594 	struct kvm *kvm = vcpu->kvm;
4595 	struct kvm_nested_guest *nested = vcpu->arch.nested;
4596 	unsigned long flags;
4597 	u64 tb;
4598 
4599 	trace_kvmppc_run_vcpu_enter(vcpu);
4600 
4601 	run->exit_reason = 0;
4602 	vcpu->arch.ret = RESUME_GUEST;
4603 	vcpu->arch.trap = 0;
4604 
4605 	vc = vcpu->arch.vcore;
4606 	vcpu->arch.ceded = 0;
4607 	vcpu->arch.run_task = current;
4608 	vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
4609 
4610 	/* See if the MMU is ready to go */
4611 	if (unlikely(!kvm->arch.mmu_ready)) {
4612 		r = kvmhv_setup_mmu(vcpu);
4613 		if (r) {
4614 			run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4615 			run->fail_entry.hardware_entry_failure_reason = 0;
4616 			vcpu->arch.ret = r;
4617 			return r;
4618 		}
4619 	}
4620 
4621 	if (need_resched())
4622 		cond_resched();
4623 
4624 	kvmppc_update_vpas(vcpu);
4625 
4626 	preempt_disable();
4627 	pcpu = smp_processor_id();
4628 	if (kvm_is_radix(kvm))
4629 		kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4630 
4631 	/* flags save not required, but irq_pmu has no disable/enable API */
4632 	powerpc_local_irq_pmu_save(flags);
4633 
4634 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4635 
4636 	if (signal_pending(current))
4637 		goto sigpend;
4638 	if (need_resched() || !kvm->arch.mmu_ready)
4639 		goto out;
4640 
4641 	vcpu->cpu = pcpu;
4642 	vcpu->arch.thread_cpu = pcpu;
4643 	vc->pcpu = pcpu;
4644 	local_paca->kvm_hstate.kvm_vcpu = vcpu;
4645 	local_paca->kvm_hstate.ptid = 0;
4646 	local_paca->kvm_hstate.fake_suspend = 0;
4647 
4648 	/*
4649 	 * Orders set cpu/thread_cpu vs testing for pending interrupts and
4650 	 * doorbells below. The other side is when these fields are set vs
4651 	 * kvmppc_fast_vcpu_kick_hv reading the cpu/thread_cpu fields to
4652 	 * kick a vCPU to notice the pending interrupt.
4653 	 */
4654 	smp_mb();
4655 
4656 	if (!nested) {
4657 		kvmppc_core_prepare_to_enter(vcpu);
4658 		if (vcpu->arch.shregs.msr & MSR_EE) {
4659 			if (xive_interrupt_pending(vcpu))
4660 				kvmppc_inject_interrupt_hv(vcpu,
4661 						BOOK3S_INTERRUPT_EXTERNAL, 0);
4662 		} else if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4663 			     &vcpu->arch.pending_exceptions)) {
4664 			lpcr |= LPCR_MER;
4665 		}
4666 	} else if (vcpu->arch.pending_exceptions ||
4667 		   vcpu->arch.doorbell_request ||
4668 		   xive_interrupt_pending(vcpu)) {
4669 		vcpu->arch.ret = RESUME_HOST;
4670 		goto out;
4671 	}
4672 
4673 	if (vcpu->arch.timer_running) {
4674 		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
4675 		vcpu->arch.timer_running = 0;
4676 	}
4677 
4678 	tb = mftb();
4679 
4680 	kvmppc_update_vpa_dispatch_p9(vcpu, vc, tb + vc->tb_offset);
4681 
4682 	trace_kvm_guest_enter(vcpu);
4683 
4684 	guest_timing_enter_irqoff();
4685 
4686 	srcu_idx = srcu_read_lock(&kvm->srcu);
4687 
4688 	guest_state_enter_irqoff();
4689 	this_cpu_disable_ftrace();
4690 
4691 	trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr, &tb);
4692 	vcpu->arch.trap = trap;
4693 
4694 	this_cpu_enable_ftrace();
4695 	guest_state_exit_irqoff();
4696 
4697 	srcu_read_unlock(&kvm->srcu, srcu_idx);
4698 
4699 	set_irq_happened(trap);
4700 
4701 	vcpu->cpu = -1;
4702 	vcpu->arch.thread_cpu = -1;
4703 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4704 
4705 	if (!vtime_accounting_enabled_this_cpu()) {
4706 		powerpc_local_irq_pmu_restore(flags);
4707 		/*
4708 		 * Service IRQs here before guest_timing_exit_irqoff() so any
4709 		 * ticks that occurred while running the guest are accounted to
4710 		 * the guest. If vtime accounting is enabled, accounting uses
4711 		 * TB rather than ticks, so it can be done without enabling
4712 		 * interrupts here, which has the problem that it accounts
4713 		 * interrupt processing overhead to the host.
4714 		 */
4715 		powerpc_local_irq_pmu_save(flags);
4716 	}
4717 	guest_timing_exit_irqoff();
4718 
4719 	powerpc_local_irq_pmu_restore(flags);
4720 
4721 	preempt_enable();
4722 
4723 	/*
4724 	 * cancel pending decrementer exception if DEC is now positive, or if
4725 	 * entering a nested guest in which case the decrementer is now owned
4726 	 * by L2 and the L1 decrementer is provided in hdec_expires
4727 	 */
4728 	if (kvmppc_core_pending_dec(vcpu) &&
4729 			((tb < kvmppc_dec_expires_host_tb(vcpu)) ||
4730 			 (trap == BOOK3S_INTERRUPT_SYSCALL &&
4731 			  kvmppc_get_gpr(vcpu, 3) == H_ENTER_NESTED)))
4732 		kvmppc_core_dequeue_dec(vcpu);
4733 
4734 	trace_kvm_guest_exit(vcpu);
4735 	r = RESUME_GUEST;
4736 	if (trap) {
4737 		if (!nested)
4738 			r = kvmppc_handle_exit_hv(vcpu, current);
4739 		else
4740 			r = kvmppc_handle_nested_exit(vcpu);
4741 	}
4742 	vcpu->arch.ret = r;
4743 
4744 	if (is_kvmppc_resume_guest(r) && !kvmppc_vcpu_check_block(vcpu)) {
4745 		kvmppc_set_timer(vcpu);
4746 
4747 		prepare_to_rcuwait(wait);
4748 		for (;;) {
4749 			set_current_state(TASK_INTERRUPTIBLE);
4750 			if (signal_pending(current)) {
4751 				vcpu->stat.signal_exits++;
4752 				run->exit_reason = KVM_EXIT_INTR;
4753 				vcpu->arch.ret = -EINTR;
4754 				break;
4755 			}
4756 
4757 			if (kvmppc_vcpu_check_block(vcpu))
4758 				break;
4759 
4760 			trace_kvmppc_vcore_blocked(vcpu, 0);
4761 			schedule();
4762 			trace_kvmppc_vcore_blocked(vcpu, 1);
4763 		}
4764 		finish_rcuwait(wait);
4765 	}
4766 	vcpu->arch.ceded = 0;
4767 
4768  done:
4769 	trace_kvmppc_run_vcpu_exit(vcpu);
4770 
4771 	return vcpu->arch.ret;
4772 
4773  sigpend:
4774 	vcpu->stat.signal_exits++;
4775 	run->exit_reason = KVM_EXIT_INTR;
4776 	vcpu->arch.ret = -EINTR;
4777  out:
4778 	vcpu->cpu = -1;
4779 	vcpu->arch.thread_cpu = -1;
4780 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4781 	powerpc_local_irq_pmu_restore(flags);
4782 	preempt_enable();
4783 	goto done;
4784 }
4785 
4786 static int kvmppc_vcpu_run_hv(struct kvm_vcpu *vcpu)
4787 {
4788 	struct kvm_run *run = vcpu->run;
4789 	int r;
4790 	int srcu_idx;
4791 	struct kvm *kvm;
4792 	unsigned long msr;
4793 
4794 	start_timing(vcpu, &vcpu->arch.vcpu_entry);
4795 
4796 	if (!vcpu->arch.sane) {
4797 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4798 		return -EINVAL;
4799 	}
4800 
4801 	/* No need to go into the guest when all we'll do is come back out */
4802 	if (signal_pending(current)) {
4803 		run->exit_reason = KVM_EXIT_INTR;
4804 		return -EINTR;
4805 	}
4806 
4807 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4808 	/*
4809 	 * Don't allow entry with a suspended transaction, because
4810 	 * the guest entry/exit code will lose it.
4811 	 */
4812 	if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4813 	    (current->thread.regs->msr & MSR_TM)) {
4814 		if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4815 			run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4816 			run->fail_entry.hardware_entry_failure_reason = 0;
4817 			return -EINVAL;
4818 		}
4819 	}
4820 #endif
4821 
4822 	/*
4823 	 * Force online to 1 for the sake of old userspace which doesn't
4824 	 * set it.
4825 	 */
4826 	if (!vcpu->arch.online) {
4827 		atomic_inc(&vcpu->arch.vcore->online_count);
4828 		vcpu->arch.online = 1;
4829 	}
4830 
4831 	kvmppc_core_prepare_to_enter(vcpu);
4832 
4833 	kvm = vcpu->kvm;
4834 	atomic_inc(&kvm->arch.vcpus_running);
4835 	/* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4836 	smp_mb();
4837 
4838 	msr = 0;
4839 	if (IS_ENABLED(CONFIG_PPC_FPU))
4840 		msr |= MSR_FP;
4841 	if (cpu_has_feature(CPU_FTR_ALTIVEC))
4842 		msr |= MSR_VEC;
4843 	if (cpu_has_feature(CPU_FTR_VSX))
4844 		msr |= MSR_VSX;
4845 	if ((cpu_has_feature(CPU_FTR_TM) ||
4846 	    cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST)) &&
4847 			(vcpu->arch.hfscr & HFSCR_TM))
4848 		msr |= MSR_TM;
4849 	msr = msr_check_and_set(msr);
4850 
4851 	kvmppc_save_user_regs();
4852 
4853 	kvmppc_save_current_sprs();
4854 
4855 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
4856 		vcpu->arch.waitp = &vcpu->arch.vcore->wait;
4857 	vcpu->arch.pgdir = kvm->mm->pgd;
4858 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4859 
4860 	do {
4861 		accumulate_time(vcpu, &vcpu->arch.guest_entry);
4862 		if (cpu_has_feature(CPU_FTR_ARCH_300))
4863 			r = kvmhv_run_single_vcpu(vcpu, ~(u64)0,
4864 						  vcpu->arch.vcore->lpcr);
4865 		else
4866 			r = kvmppc_run_vcpu(vcpu);
4867 
4868 		if (run->exit_reason == KVM_EXIT_PAPR_HCALL) {
4869 			accumulate_time(vcpu, &vcpu->arch.hcall);
4870 
4871 			if (WARN_ON_ONCE(vcpu->arch.shregs.msr & MSR_PR)) {
4872 				/*
4873 				 * These should have been caught reflected
4874 				 * into the guest by now. Final sanity check:
4875 				 * don't allow userspace to execute hcalls in
4876 				 * the hypervisor.
4877 				 */
4878 				r = RESUME_GUEST;
4879 				continue;
4880 			}
4881 			trace_kvm_hcall_enter(vcpu);
4882 			r = kvmppc_pseries_do_hcall(vcpu);
4883 			trace_kvm_hcall_exit(vcpu, r);
4884 			kvmppc_core_prepare_to_enter(vcpu);
4885 		} else if (r == RESUME_PAGE_FAULT) {
4886 			accumulate_time(vcpu, &vcpu->arch.pg_fault);
4887 			srcu_idx = srcu_read_lock(&kvm->srcu);
4888 			r = kvmppc_book3s_hv_page_fault(vcpu,
4889 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4890 			srcu_read_unlock(&kvm->srcu, srcu_idx);
4891 		} else if (r == RESUME_PASSTHROUGH) {
4892 			if (WARN_ON(xics_on_xive()))
4893 				r = H_SUCCESS;
4894 			else
4895 				r = kvmppc_xics_rm_complete(vcpu, 0);
4896 		}
4897 	} while (is_kvmppc_resume_guest(r));
4898 	accumulate_time(vcpu, &vcpu->arch.vcpu_exit);
4899 
4900 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4901 	atomic_dec(&kvm->arch.vcpus_running);
4902 
4903 	srr_regs_clobbered();
4904 
4905 	end_timing(vcpu);
4906 
4907 	return r;
4908 }
4909 
4910 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4911 				     int shift, int sllp)
4912 {
4913 	(*sps)->page_shift = shift;
4914 	(*sps)->slb_enc = sllp;
4915 	(*sps)->enc[0].page_shift = shift;
4916 	(*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4917 	/*
4918 	 * Add 16MB MPSS support (may get filtered out by userspace)
4919 	 */
4920 	if (shift != 24) {
4921 		int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4922 		if (penc != -1) {
4923 			(*sps)->enc[1].page_shift = 24;
4924 			(*sps)->enc[1].pte_enc = penc;
4925 		}
4926 	}
4927 	(*sps)++;
4928 }
4929 
4930 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4931 					 struct kvm_ppc_smmu_info *info)
4932 {
4933 	struct kvm_ppc_one_seg_page_size *sps;
4934 
4935 	/*
4936 	 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4937 	 * POWER7 doesn't support keys for instruction accesses,
4938 	 * POWER8 and POWER9 do.
4939 	 */
4940 	info->data_keys = 32;
4941 	info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4942 
4943 	/* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4944 	info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4945 	info->slb_size = 32;
4946 
4947 	/* We only support these sizes for now, and no muti-size segments */
4948 	sps = &info->sps[0];
4949 	kvmppc_add_seg_page_size(&sps, 12, 0);
4950 	kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4951 	kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4952 
4953 	/* If running as a nested hypervisor, we don't support HPT guests */
4954 	if (kvmhv_on_pseries())
4955 		info->flags |= KVM_PPC_NO_HASH;
4956 
4957 	return 0;
4958 }
4959 
4960 /*
4961  * Get (and clear) the dirty memory log for a memory slot.
4962  */
4963 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4964 					 struct kvm_dirty_log *log)
4965 {
4966 	struct kvm_memslots *slots;
4967 	struct kvm_memory_slot *memslot;
4968 	int r;
4969 	unsigned long n, i;
4970 	unsigned long *buf, *p;
4971 	struct kvm_vcpu *vcpu;
4972 
4973 	mutex_lock(&kvm->slots_lock);
4974 
4975 	r = -EINVAL;
4976 	if (log->slot >= KVM_USER_MEM_SLOTS)
4977 		goto out;
4978 
4979 	slots = kvm_memslots(kvm);
4980 	memslot = id_to_memslot(slots, log->slot);
4981 	r = -ENOENT;
4982 	if (!memslot || !memslot->dirty_bitmap)
4983 		goto out;
4984 
4985 	/*
4986 	 * Use second half of bitmap area because both HPT and radix
4987 	 * accumulate bits in the first half.
4988 	 */
4989 	n = kvm_dirty_bitmap_bytes(memslot);
4990 	buf = memslot->dirty_bitmap + n / sizeof(long);
4991 	memset(buf, 0, n);
4992 
4993 	if (kvm_is_radix(kvm))
4994 		r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4995 	else
4996 		r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4997 	if (r)
4998 		goto out;
4999 
5000 	/*
5001 	 * We accumulate dirty bits in the first half of the
5002 	 * memslot's dirty_bitmap area, for when pages are paged
5003 	 * out or modified by the host directly.  Pick up these
5004 	 * bits and add them to the map.
5005 	 */
5006 	p = memslot->dirty_bitmap;
5007 	for (i = 0; i < n / sizeof(long); ++i)
5008 		buf[i] |= xchg(&p[i], 0);
5009 
5010 	/* Harvest dirty bits from VPA and DTL updates */
5011 	/* Note: we never modify the SLB shadow buffer areas */
5012 	kvm_for_each_vcpu(i, vcpu, kvm) {
5013 		spin_lock(&vcpu->arch.vpa_update_lock);
5014 		kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
5015 		kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
5016 		spin_unlock(&vcpu->arch.vpa_update_lock);
5017 	}
5018 
5019 	r = -EFAULT;
5020 	if (copy_to_user(log->dirty_bitmap, buf, n))
5021 		goto out;
5022 
5023 	r = 0;
5024 out:
5025 	mutex_unlock(&kvm->slots_lock);
5026 	return r;
5027 }
5028 
5029 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *slot)
5030 {
5031 	vfree(slot->arch.rmap);
5032 	slot->arch.rmap = NULL;
5033 }
5034 
5035 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
5036 				const struct kvm_memory_slot *old,
5037 				struct kvm_memory_slot *new,
5038 				enum kvm_mr_change change)
5039 {
5040 	if (change == KVM_MR_CREATE) {
5041 		unsigned long size = array_size(new->npages, sizeof(*new->arch.rmap));
5042 
5043 		if ((size >> PAGE_SHIFT) > totalram_pages())
5044 			return -ENOMEM;
5045 
5046 		new->arch.rmap = vzalloc(size);
5047 		if (!new->arch.rmap)
5048 			return -ENOMEM;
5049 	} else if (change != KVM_MR_DELETE) {
5050 		new->arch.rmap = old->arch.rmap;
5051 	}
5052 
5053 	return 0;
5054 }
5055 
5056 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
5057 				struct kvm_memory_slot *old,
5058 				const struct kvm_memory_slot *new,
5059 				enum kvm_mr_change change)
5060 {
5061 	/*
5062 	 * If we are creating or modifying a memslot, it might make
5063 	 * some address that was previously cached as emulated
5064 	 * MMIO be no longer emulated MMIO, so invalidate
5065 	 * all the caches of emulated MMIO translations.
5066 	 */
5067 	if (change != KVM_MR_DELETE)
5068 		atomic64_inc(&kvm->arch.mmio_update);
5069 
5070 	/*
5071 	 * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
5072 	 * have already called kvm_arch_flush_shadow_memslot() to
5073 	 * flush shadow mappings.  For KVM_MR_CREATE we have no
5074 	 * previous mappings.  So the only case to handle is
5075 	 * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
5076 	 * has been changed.
5077 	 * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
5078 	 * to get rid of any THP PTEs in the partition-scoped page tables
5079 	 * so we can track dirtiness at the page level; we flush when
5080 	 * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
5081 	 * using THP PTEs.
5082 	 */
5083 	if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
5084 	    ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
5085 		kvmppc_radix_flush_memslot(kvm, old);
5086 	/*
5087 	 * If UV hasn't yet called H_SVM_INIT_START, don't register memslots.
5088 	 */
5089 	if (!kvm->arch.secure_guest)
5090 		return;
5091 
5092 	switch (change) {
5093 	case KVM_MR_CREATE:
5094 		/*
5095 		 * @TODO kvmppc_uvmem_memslot_create() can fail and
5096 		 * return error. Fix this.
5097 		 */
5098 		kvmppc_uvmem_memslot_create(kvm, new);
5099 		break;
5100 	case KVM_MR_DELETE:
5101 		kvmppc_uvmem_memslot_delete(kvm, old);
5102 		break;
5103 	default:
5104 		/* TODO: Handle KVM_MR_MOVE */
5105 		break;
5106 	}
5107 }
5108 
5109 /*
5110  * Update LPCR values in kvm->arch and in vcores.
5111  * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion
5112  * of kvm->arch.lpcr update).
5113  */
5114 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
5115 {
5116 	long int i;
5117 	u32 cores_done = 0;
5118 
5119 	if ((kvm->arch.lpcr & mask) == lpcr)
5120 		return;
5121 
5122 	kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
5123 
5124 	for (i = 0; i < KVM_MAX_VCORES; ++i) {
5125 		struct kvmppc_vcore *vc = kvm->arch.vcores[i];
5126 		if (!vc)
5127 			continue;
5128 
5129 		spin_lock(&vc->lock);
5130 		vc->lpcr = (vc->lpcr & ~mask) | lpcr;
5131 		verify_lpcr(kvm, vc->lpcr);
5132 		spin_unlock(&vc->lock);
5133 		if (++cores_done >= kvm->arch.online_vcores)
5134 			break;
5135 	}
5136 }
5137 
5138 void kvmppc_setup_partition_table(struct kvm *kvm)
5139 {
5140 	unsigned long dw0, dw1;
5141 
5142 	if (!kvm_is_radix(kvm)) {
5143 		/* PS field - page size for VRMA */
5144 		dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
5145 			((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
5146 		/* HTABSIZE and HTABORG fields */
5147 		dw0 |= kvm->arch.sdr1;
5148 
5149 		/* Second dword as set by userspace */
5150 		dw1 = kvm->arch.process_table;
5151 	} else {
5152 		dw0 = PATB_HR | radix__get_tree_size() |
5153 			__pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
5154 		dw1 = PATB_GR | kvm->arch.process_table;
5155 	}
5156 	kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
5157 }
5158 
5159 /*
5160  * Set up HPT (hashed page table) and RMA (real-mode area).
5161  * Must be called with kvm->arch.mmu_setup_lock held.
5162  */
5163 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
5164 {
5165 	int err = 0;
5166 	struct kvm *kvm = vcpu->kvm;
5167 	unsigned long hva;
5168 	struct kvm_memory_slot *memslot;
5169 	struct vm_area_struct *vma;
5170 	unsigned long lpcr = 0, senc;
5171 	unsigned long psize, porder;
5172 	int srcu_idx;
5173 
5174 	/* Allocate hashed page table (if not done already) and reset it */
5175 	if (!kvm->arch.hpt.virt) {
5176 		int order = KVM_DEFAULT_HPT_ORDER;
5177 		struct kvm_hpt_info info;
5178 
5179 		err = kvmppc_allocate_hpt(&info, order);
5180 		/* If we get here, it means userspace didn't specify a
5181 		 * size explicitly.  So, try successively smaller
5182 		 * sizes if the default failed. */
5183 		while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
5184 			err  = kvmppc_allocate_hpt(&info, order);
5185 
5186 		if (err < 0) {
5187 			pr_err("KVM: Couldn't alloc HPT\n");
5188 			goto out;
5189 		}
5190 
5191 		kvmppc_set_hpt(kvm, &info);
5192 	}
5193 
5194 	/* Look up the memslot for guest physical address 0 */
5195 	srcu_idx = srcu_read_lock(&kvm->srcu);
5196 	memslot = gfn_to_memslot(kvm, 0);
5197 
5198 	/* We must have some memory at 0 by now */
5199 	err = -EINVAL;
5200 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
5201 		goto out_srcu;
5202 
5203 	/* Look up the VMA for the start of this memory slot */
5204 	hva = memslot->userspace_addr;
5205 	mmap_read_lock(kvm->mm);
5206 	vma = vma_lookup(kvm->mm, hva);
5207 	if (!vma || (vma->vm_flags & VM_IO))
5208 		goto up_out;
5209 
5210 	psize = vma_kernel_pagesize(vma);
5211 
5212 	mmap_read_unlock(kvm->mm);
5213 
5214 	/* We can handle 4k, 64k or 16M pages in the VRMA */
5215 	if (psize >= 0x1000000)
5216 		psize = 0x1000000;
5217 	else if (psize >= 0x10000)
5218 		psize = 0x10000;
5219 	else
5220 		psize = 0x1000;
5221 	porder = __ilog2(psize);
5222 
5223 	senc = slb_pgsize_encoding(psize);
5224 	kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
5225 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
5226 	/* Create HPTEs in the hash page table for the VRMA */
5227 	kvmppc_map_vrma(vcpu, memslot, porder);
5228 
5229 	/* Update VRMASD field in the LPCR */
5230 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
5231 		/* the -4 is to account for senc values starting at 0x10 */
5232 		lpcr = senc << (LPCR_VRMASD_SH - 4);
5233 		kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
5234 	}
5235 
5236 	/* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
5237 	smp_wmb();
5238 	err = 0;
5239  out_srcu:
5240 	srcu_read_unlock(&kvm->srcu, srcu_idx);
5241  out:
5242 	return err;
5243 
5244  up_out:
5245 	mmap_read_unlock(kvm->mm);
5246 	goto out_srcu;
5247 }
5248 
5249 /*
5250  * Must be called with kvm->arch.mmu_setup_lock held and
5251  * mmu_ready = 0 and no vcpus running.
5252  */
5253 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
5254 {
5255 	unsigned long lpcr, lpcr_mask;
5256 
5257 	if (nesting_enabled(kvm))
5258 		kvmhv_release_all_nested(kvm);
5259 	kvmppc_rmap_reset(kvm);
5260 	kvm->arch.process_table = 0;
5261 	/* Mutual exclusion with kvm_unmap_gfn_range etc. */
5262 	spin_lock(&kvm->mmu_lock);
5263 	kvm->arch.radix = 0;
5264 	spin_unlock(&kvm->mmu_lock);
5265 	kvmppc_free_radix(kvm);
5266 
5267 	lpcr = LPCR_VPM1;
5268 	lpcr_mask = LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5269 	if (cpu_has_feature(CPU_FTR_ARCH_31))
5270 		lpcr_mask |= LPCR_HAIL;
5271 	kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
5272 
5273 	return 0;
5274 }
5275 
5276 /*
5277  * Must be called with kvm->arch.mmu_setup_lock held and
5278  * mmu_ready = 0 and no vcpus running.
5279  */
5280 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
5281 {
5282 	unsigned long lpcr, lpcr_mask;
5283 	int err;
5284 
5285 	err = kvmppc_init_vm_radix(kvm);
5286 	if (err)
5287 		return err;
5288 	kvmppc_rmap_reset(kvm);
5289 	/* Mutual exclusion with kvm_unmap_gfn_range etc. */
5290 	spin_lock(&kvm->mmu_lock);
5291 	kvm->arch.radix = 1;
5292 	spin_unlock(&kvm->mmu_lock);
5293 	kvmppc_free_hpt(&kvm->arch.hpt);
5294 
5295 	lpcr = LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5296 	lpcr_mask = LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5297 	if (cpu_has_feature(CPU_FTR_ARCH_31)) {
5298 		lpcr_mask |= LPCR_HAIL;
5299 		if (cpu_has_feature(CPU_FTR_HVMODE) &&
5300 				(kvm->arch.host_lpcr & LPCR_HAIL))
5301 			lpcr |= LPCR_HAIL;
5302 	}
5303 	kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
5304 
5305 	return 0;
5306 }
5307 
5308 #ifdef CONFIG_KVM_XICS
5309 /*
5310  * Allocate a per-core structure for managing state about which cores are
5311  * running in the host versus the guest and for exchanging data between
5312  * real mode KVM and CPU running in the host.
5313  * This is only done for the first VM.
5314  * The allocated structure stays even if all VMs have stopped.
5315  * It is only freed when the kvm-hv module is unloaded.
5316  * It's OK for this routine to fail, we just don't support host
5317  * core operations like redirecting H_IPI wakeups.
5318  */
5319 void kvmppc_alloc_host_rm_ops(void)
5320 {
5321 	struct kvmppc_host_rm_ops *ops;
5322 	unsigned long l_ops;
5323 	int cpu, core;
5324 	int size;
5325 
5326 	if (cpu_has_feature(CPU_FTR_ARCH_300))
5327 		return;
5328 
5329 	/* Not the first time here ? */
5330 	if (kvmppc_host_rm_ops_hv != NULL)
5331 		return;
5332 
5333 	ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
5334 	if (!ops)
5335 		return;
5336 
5337 	size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
5338 	ops->rm_core = kzalloc(size, GFP_KERNEL);
5339 
5340 	if (!ops->rm_core) {
5341 		kfree(ops);
5342 		return;
5343 	}
5344 
5345 	cpus_read_lock();
5346 
5347 	for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
5348 		if (!cpu_online(cpu))
5349 			continue;
5350 
5351 		core = cpu >> threads_shift;
5352 		ops->rm_core[core].rm_state.in_host = 1;
5353 	}
5354 
5355 	ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
5356 
5357 	/*
5358 	 * Make the contents of the kvmppc_host_rm_ops structure visible
5359 	 * to other CPUs before we assign it to the global variable.
5360 	 * Do an atomic assignment (no locks used here), but if someone
5361 	 * beats us to it, just free our copy and return.
5362 	 */
5363 	smp_wmb();
5364 	l_ops = (unsigned long) ops;
5365 
5366 	if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
5367 		cpus_read_unlock();
5368 		kfree(ops->rm_core);
5369 		kfree(ops);
5370 		return;
5371 	}
5372 
5373 	cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
5374 					     "ppc/kvm_book3s:prepare",
5375 					     kvmppc_set_host_core,
5376 					     kvmppc_clear_host_core);
5377 	cpus_read_unlock();
5378 }
5379 
5380 void kvmppc_free_host_rm_ops(void)
5381 {
5382 	if (kvmppc_host_rm_ops_hv) {
5383 		cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
5384 		kfree(kvmppc_host_rm_ops_hv->rm_core);
5385 		kfree(kvmppc_host_rm_ops_hv);
5386 		kvmppc_host_rm_ops_hv = NULL;
5387 	}
5388 }
5389 #endif
5390 
5391 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
5392 {
5393 	unsigned long lpcr, lpid;
5394 	int ret;
5395 
5396 	mutex_init(&kvm->arch.uvmem_lock);
5397 	INIT_LIST_HEAD(&kvm->arch.uvmem_pfns);
5398 	mutex_init(&kvm->arch.mmu_setup_lock);
5399 
5400 	/* Allocate the guest's logical partition ID */
5401 
5402 	lpid = kvmppc_alloc_lpid();
5403 	if ((long)lpid < 0)
5404 		return -ENOMEM;
5405 	kvm->arch.lpid = lpid;
5406 
5407 	kvmppc_alloc_host_rm_ops();
5408 
5409 	kvmhv_vm_nested_init(kvm);
5410 
5411 	/*
5412 	 * Since we don't flush the TLB when tearing down a VM,
5413 	 * and this lpid might have previously been used,
5414 	 * make sure we flush on each core before running the new VM.
5415 	 * On POWER9, the tlbie in mmu_partition_table_set_entry()
5416 	 * does this flush for us.
5417 	 */
5418 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5419 		cpumask_setall(&kvm->arch.need_tlb_flush);
5420 
5421 	/* Start out with the default set of hcalls enabled */
5422 	memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
5423 	       sizeof(kvm->arch.enabled_hcalls));
5424 
5425 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5426 		kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
5427 
5428 	/* Init LPCR for virtual RMA mode */
5429 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
5430 		kvm->arch.host_lpid = mfspr(SPRN_LPID);
5431 		kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
5432 		lpcr &= LPCR_PECE | LPCR_LPES;
5433 	} else {
5434 		/*
5435 		 * The L2 LPES mode will be set by the L0 according to whether
5436 		 * or not it needs to take external interrupts in HV mode.
5437 		 */
5438 		lpcr = 0;
5439 	}
5440 	lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
5441 		LPCR_VPM0 | LPCR_VPM1;
5442 	kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
5443 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
5444 	/* On POWER8 turn on online bit to enable PURR/SPURR */
5445 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
5446 		lpcr |= LPCR_ONL;
5447 	/*
5448 	 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
5449 	 * Set HVICE bit to enable hypervisor virtualization interrupts.
5450 	 * Set HEIC to prevent OS interrupts to go to hypervisor (should
5451 	 * be unnecessary but better safe than sorry in case we re-enable
5452 	 * EE in HV mode with this LPCR still set)
5453 	 */
5454 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5455 		lpcr &= ~LPCR_VPM0;
5456 		lpcr |= LPCR_HVICE | LPCR_HEIC;
5457 
5458 		/*
5459 		 * If xive is enabled, we route 0x500 interrupts directly
5460 		 * to the guest.
5461 		 */
5462 		if (xics_on_xive())
5463 			lpcr |= LPCR_LPES;
5464 	}
5465 
5466 	/*
5467 	 * If the host uses radix, the guest starts out as radix.
5468 	 */
5469 	if (radix_enabled()) {
5470 		kvm->arch.radix = 1;
5471 		kvm->arch.mmu_ready = 1;
5472 		lpcr &= ~LPCR_VPM1;
5473 		lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
5474 		if (cpu_has_feature(CPU_FTR_HVMODE) &&
5475 		    cpu_has_feature(CPU_FTR_ARCH_31) &&
5476 		    (kvm->arch.host_lpcr & LPCR_HAIL))
5477 			lpcr |= LPCR_HAIL;
5478 		ret = kvmppc_init_vm_radix(kvm);
5479 		if (ret) {
5480 			kvmppc_free_lpid(kvm->arch.lpid);
5481 			return ret;
5482 		}
5483 		kvmppc_setup_partition_table(kvm);
5484 	}
5485 
5486 	verify_lpcr(kvm, lpcr);
5487 	kvm->arch.lpcr = lpcr;
5488 
5489 	/* Initialization for future HPT resizes */
5490 	kvm->arch.resize_hpt = NULL;
5491 
5492 	/*
5493 	 * Work out how many sets the TLB has, for the use of
5494 	 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
5495 	 */
5496 	if (cpu_has_feature(CPU_FTR_ARCH_31)) {
5497 		/*
5498 		 * P10 will flush all the congruence class with a single tlbiel
5499 		 */
5500 		kvm->arch.tlb_sets = 1;
5501 	} else if (radix_enabled())
5502 		kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;	/* 128 */
5503 	else if (cpu_has_feature(CPU_FTR_ARCH_300))
5504 		kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;	/* 256 */
5505 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
5506 		kvm->arch.tlb_sets = POWER8_TLB_SETS;		/* 512 */
5507 	else
5508 		kvm->arch.tlb_sets = POWER7_TLB_SETS;		/* 128 */
5509 
5510 	/*
5511 	 * Track that we now have a HV mode VM active. This blocks secondary
5512 	 * CPU threads from coming online.
5513 	 */
5514 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5515 		kvm_hv_vm_activated();
5516 
5517 	/*
5518 	 * Initialize smt_mode depending on processor.
5519 	 * POWER8 and earlier have to use "strict" threading, where
5520 	 * all vCPUs in a vcore have to run on the same (sub)core,
5521 	 * whereas on POWER9 the threads can each run a different
5522 	 * guest.
5523 	 */
5524 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5525 		kvm->arch.smt_mode = threads_per_subcore;
5526 	else
5527 		kvm->arch.smt_mode = 1;
5528 	kvm->arch.emul_smt_mode = 1;
5529 
5530 	return 0;
5531 }
5532 
5533 static int kvmppc_arch_create_vm_debugfs_hv(struct kvm *kvm)
5534 {
5535 	kvmppc_mmu_debugfs_init(kvm);
5536 	if (radix_enabled())
5537 		kvmhv_radix_debugfs_init(kvm);
5538 	return 0;
5539 }
5540 
5541 static void kvmppc_free_vcores(struct kvm *kvm)
5542 {
5543 	long int i;
5544 
5545 	for (i = 0; i < KVM_MAX_VCORES; ++i)
5546 		kfree(kvm->arch.vcores[i]);
5547 	kvm->arch.online_vcores = 0;
5548 }
5549 
5550 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
5551 {
5552 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5553 		kvm_hv_vm_deactivated();
5554 
5555 	kvmppc_free_vcores(kvm);
5556 
5557 
5558 	if (kvm_is_radix(kvm))
5559 		kvmppc_free_radix(kvm);
5560 	else
5561 		kvmppc_free_hpt(&kvm->arch.hpt);
5562 
5563 	/* Perform global invalidation and return lpid to the pool */
5564 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5565 		if (nesting_enabled(kvm))
5566 			kvmhv_release_all_nested(kvm);
5567 		kvm->arch.process_table = 0;
5568 		if (kvm->arch.secure_guest)
5569 			uv_svm_terminate(kvm->arch.lpid);
5570 		kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
5571 	}
5572 
5573 	kvmppc_free_lpid(kvm->arch.lpid);
5574 
5575 	kvmppc_free_pimap(kvm);
5576 }
5577 
5578 /* We don't need to emulate any privileged instructions or dcbz */
5579 static int kvmppc_core_emulate_op_hv(struct kvm_vcpu *vcpu,
5580 				     unsigned int inst, int *advance)
5581 {
5582 	return EMULATE_FAIL;
5583 }
5584 
5585 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
5586 					ulong spr_val)
5587 {
5588 	return EMULATE_FAIL;
5589 }
5590 
5591 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
5592 					ulong *spr_val)
5593 {
5594 	return EMULATE_FAIL;
5595 }
5596 
5597 static int kvmppc_core_check_processor_compat_hv(void)
5598 {
5599 	if (cpu_has_feature(CPU_FTR_HVMODE) &&
5600 	    cpu_has_feature(CPU_FTR_ARCH_206))
5601 		return 0;
5602 
5603 	/* POWER9 in radix mode is capable of being a nested hypervisor. */
5604 	if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
5605 		return 0;
5606 
5607 	return -EIO;
5608 }
5609 
5610 #ifdef CONFIG_KVM_XICS
5611 
5612 void kvmppc_free_pimap(struct kvm *kvm)
5613 {
5614 	kfree(kvm->arch.pimap);
5615 }
5616 
5617 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
5618 {
5619 	return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
5620 }
5621 
5622 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5623 {
5624 	struct irq_desc *desc;
5625 	struct kvmppc_irq_map *irq_map;
5626 	struct kvmppc_passthru_irqmap *pimap;
5627 	struct irq_chip *chip;
5628 	int i, rc = 0;
5629 	struct irq_data *host_data;
5630 
5631 	if (!kvm_irq_bypass)
5632 		return 1;
5633 
5634 	desc = irq_to_desc(host_irq);
5635 	if (!desc)
5636 		return -EIO;
5637 
5638 	mutex_lock(&kvm->lock);
5639 
5640 	pimap = kvm->arch.pimap;
5641 	if (pimap == NULL) {
5642 		/* First call, allocate structure to hold IRQ map */
5643 		pimap = kvmppc_alloc_pimap();
5644 		if (pimap == NULL) {
5645 			mutex_unlock(&kvm->lock);
5646 			return -ENOMEM;
5647 		}
5648 		kvm->arch.pimap = pimap;
5649 	}
5650 
5651 	/*
5652 	 * For now, we only support interrupts for which the EOI operation
5653 	 * is an OPAL call followed by a write to XIRR, since that's
5654 	 * what our real-mode EOI code does, or a XIVE interrupt
5655 	 */
5656 	chip = irq_data_get_irq_chip(&desc->irq_data);
5657 	if (!chip || !is_pnv_opal_msi(chip)) {
5658 		pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
5659 			host_irq, guest_gsi);
5660 		mutex_unlock(&kvm->lock);
5661 		return -ENOENT;
5662 	}
5663 
5664 	/*
5665 	 * See if we already have an entry for this guest IRQ number.
5666 	 * If it's mapped to a hardware IRQ number, that's an error,
5667 	 * otherwise re-use this entry.
5668 	 */
5669 	for (i = 0; i < pimap->n_mapped; i++) {
5670 		if (guest_gsi == pimap->mapped[i].v_hwirq) {
5671 			if (pimap->mapped[i].r_hwirq) {
5672 				mutex_unlock(&kvm->lock);
5673 				return -EINVAL;
5674 			}
5675 			break;
5676 		}
5677 	}
5678 
5679 	if (i == KVMPPC_PIRQ_MAPPED) {
5680 		mutex_unlock(&kvm->lock);
5681 		return -EAGAIN;		/* table is full */
5682 	}
5683 
5684 	irq_map = &pimap->mapped[i];
5685 
5686 	irq_map->v_hwirq = guest_gsi;
5687 	irq_map->desc = desc;
5688 
5689 	/*
5690 	 * Order the above two stores before the next to serialize with
5691 	 * the KVM real mode handler.
5692 	 */
5693 	smp_wmb();
5694 
5695 	/*
5696 	 * The 'host_irq' number is mapped in the PCI-MSI domain but
5697 	 * the underlying calls, which will EOI the interrupt in real
5698 	 * mode, need an HW IRQ number mapped in the XICS IRQ domain.
5699 	 */
5700 	host_data = irq_domain_get_irq_data(irq_get_default_host(), host_irq);
5701 	irq_map->r_hwirq = (unsigned int)irqd_to_hwirq(host_data);
5702 
5703 	if (i == pimap->n_mapped)
5704 		pimap->n_mapped++;
5705 
5706 	if (xics_on_xive())
5707 		rc = kvmppc_xive_set_mapped(kvm, guest_gsi, host_irq);
5708 	else
5709 		kvmppc_xics_set_mapped(kvm, guest_gsi, irq_map->r_hwirq);
5710 	if (rc)
5711 		irq_map->r_hwirq = 0;
5712 
5713 	mutex_unlock(&kvm->lock);
5714 
5715 	return 0;
5716 }
5717 
5718 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5719 {
5720 	struct irq_desc *desc;
5721 	struct kvmppc_passthru_irqmap *pimap;
5722 	int i, rc = 0;
5723 
5724 	if (!kvm_irq_bypass)
5725 		return 0;
5726 
5727 	desc = irq_to_desc(host_irq);
5728 	if (!desc)
5729 		return -EIO;
5730 
5731 	mutex_lock(&kvm->lock);
5732 	if (!kvm->arch.pimap)
5733 		goto unlock;
5734 
5735 	pimap = kvm->arch.pimap;
5736 
5737 	for (i = 0; i < pimap->n_mapped; i++) {
5738 		if (guest_gsi == pimap->mapped[i].v_hwirq)
5739 			break;
5740 	}
5741 
5742 	if (i == pimap->n_mapped) {
5743 		mutex_unlock(&kvm->lock);
5744 		return -ENODEV;
5745 	}
5746 
5747 	if (xics_on_xive())
5748 		rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, host_irq);
5749 	else
5750 		kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
5751 
5752 	/* invalidate the entry (what to do on error from the above ?) */
5753 	pimap->mapped[i].r_hwirq = 0;
5754 
5755 	/*
5756 	 * We don't free this structure even when the count goes to
5757 	 * zero. The structure is freed when we destroy the VM.
5758 	 */
5759  unlock:
5760 	mutex_unlock(&kvm->lock);
5761 	return rc;
5762 }
5763 
5764 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5765 					     struct irq_bypass_producer *prod)
5766 {
5767 	int ret = 0;
5768 	struct kvm_kernel_irqfd *irqfd =
5769 		container_of(cons, struct kvm_kernel_irqfd, consumer);
5770 
5771 	irqfd->producer = prod;
5772 
5773 	ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5774 	if (ret)
5775 		pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5776 			prod->irq, irqfd->gsi, ret);
5777 
5778 	return ret;
5779 }
5780 
5781 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5782 					      struct irq_bypass_producer *prod)
5783 {
5784 	int ret;
5785 	struct kvm_kernel_irqfd *irqfd =
5786 		container_of(cons, struct kvm_kernel_irqfd, consumer);
5787 
5788 	irqfd->producer = NULL;
5789 
5790 	/*
5791 	 * When producer of consumer is unregistered, we change back to
5792 	 * default external interrupt handling mode - KVM real mode
5793 	 * will switch back to host.
5794 	 */
5795 	ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5796 	if (ret)
5797 		pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5798 			prod->irq, irqfd->gsi, ret);
5799 }
5800 #endif
5801 
5802 static int kvm_arch_vm_ioctl_hv(struct file *filp,
5803 				unsigned int ioctl, unsigned long arg)
5804 {
5805 	struct kvm *kvm __maybe_unused = filp->private_data;
5806 	void __user *argp = (void __user *)arg;
5807 	int r;
5808 
5809 	switch (ioctl) {
5810 
5811 	case KVM_PPC_ALLOCATE_HTAB: {
5812 		u32 htab_order;
5813 
5814 		/* If we're a nested hypervisor, we currently only support radix */
5815 		if (kvmhv_on_pseries()) {
5816 			r = -EOPNOTSUPP;
5817 			break;
5818 		}
5819 
5820 		r = -EFAULT;
5821 		if (get_user(htab_order, (u32 __user *)argp))
5822 			break;
5823 		r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5824 		if (r)
5825 			break;
5826 		r = 0;
5827 		break;
5828 	}
5829 
5830 	case KVM_PPC_GET_HTAB_FD: {
5831 		struct kvm_get_htab_fd ghf;
5832 
5833 		r = -EFAULT;
5834 		if (copy_from_user(&ghf, argp, sizeof(ghf)))
5835 			break;
5836 		r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5837 		break;
5838 	}
5839 
5840 	case KVM_PPC_RESIZE_HPT_PREPARE: {
5841 		struct kvm_ppc_resize_hpt rhpt;
5842 
5843 		r = -EFAULT;
5844 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5845 			break;
5846 
5847 		r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5848 		break;
5849 	}
5850 
5851 	case KVM_PPC_RESIZE_HPT_COMMIT: {
5852 		struct kvm_ppc_resize_hpt rhpt;
5853 
5854 		r = -EFAULT;
5855 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5856 			break;
5857 
5858 		r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5859 		break;
5860 	}
5861 
5862 	default:
5863 		r = -ENOTTY;
5864 	}
5865 
5866 	return r;
5867 }
5868 
5869 /*
5870  * List of hcall numbers to enable by default.
5871  * For compatibility with old userspace, we enable by default
5872  * all hcalls that were implemented before the hcall-enabling
5873  * facility was added.  Note this list should not include H_RTAS.
5874  */
5875 static unsigned int default_hcall_list[] = {
5876 	H_REMOVE,
5877 	H_ENTER,
5878 	H_READ,
5879 	H_PROTECT,
5880 	H_BULK_REMOVE,
5881 #ifdef CONFIG_SPAPR_TCE_IOMMU
5882 	H_GET_TCE,
5883 	H_PUT_TCE,
5884 #endif
5885 	H_SET_DABR,
5886 	H_SET_XDABR,
5887 	H_CEDE,
5888 	H_PROD,
5889 	H_CONFER,
5890 	H_REGISTER_VPA,
5891 #ifdef CONFIG_KVM_XICS
5892 	H_EOI,
5893 	H_CPPR,
5894 	H_IPI,
5895 	H_IPOLL,
5896 	H_XIRR,
5897 	H_XIRR_X,
5898 #endif
5899 	0
5900 };
5901 
5902 static void init_default_hcalls(void)
5903 {
5904 	int i;
5905 	unsigned int hcall;
5906 
5907 	for (i = 0; default_hcall_list[i]; ++i) {
5908 		hcall = default_hcall_list[i];
5909 		WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5910 		__set_bit(hcall / 4, default_enabled_hcalls);
5911 	}
5912 }
5913 
5914 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5915 {
5916 	unsigned long lpcr;
5917 	int radix;
5918 	int err;
5919 
5920 	/* If not on a POWER9, reject it */
5921 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5922 		return -ENODEV;
5923 
5924 	/* If any unknown flags set, reject it */
5925 	if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5926 		return -EINVAL;
5927 
5928 	/* GR (guest radix) bit in process_table field must match */
5929 	radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5930 	if (!!(cfg->process_table & PATB_GR) != radix)
5931 		return -EINVAL;
5932 
5933 	/* Process table size field must be reasonable, i.e. <= 24 */
5934 	if ((cfg->process_table & PRTS_MASK) > 24)
5935 		return -EINVAL;
5936 
5937 	/* We can change a guest to/from radix now, if the host is radix */
5938 	if (radix && !radix_enabled())
5939 		return -EINVAL;
5940 
5941 	/* If we're a nested hypervisor, we currently only support radix */
5942 	if (kvmhv_on_pseries() && !radix)
5943 		return -EINVAL;
5944 
5945 	mutex_lock(&kvm->arch.mmu_setup_lock);
5946 	if (radix != kvm_is_radix(kvm)) {
5947 		if (kvm->arch.mmu_ready) {
5948 			kvm->arch.mmu_ready = 0;
5949 			/* order mmu_ready vs. vcpus_running */
5950 			smp_mb();
5951 			if (atomic_read(&kvm->arch.vcpus_running)) {
5952 				kvm->arch.mmu_ready = 1;
5953 				err = -EBUSY;
5954 				goto out_unlock;
5955 			}
5956 		}
5957 		if (radix)
5958 			err = kvmppc_switch_mmu_to_radix(kvm);
5959 		else
5960 			err = kvmppc_switch_mmu_to_hpt(kvm);
5961 		if (err)
5962 			goto out_unlock;
5963 	}
5964 
5965 	kvm->arch.process_table = cfg->process_table;
5966 	kvmppc_setup_partition_table(kvm);
5967 
5968 	lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5969 	kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5970 	err = 0;
5971 
5972  out_unlock:
5973 	mutex_unlock(&kvm->arch.mmu_setup_lock);
5974 	return err;
5975 }
5976 
5977 static int kvmhv_enable_nested(struct kvm *kvm)
5978 {
5979 	if (!nested)
5980 		return -EPERM;
5981 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5982 		return -ENODEV;
5983 	if (!radix_enabled())
5984 		return -ENODEV;
5985 
5986 	/* kvm == NULL means the caller is testing if the capability exists */
5987 	if (kvm)
5988 		kvm->arch.nested_enable = true;
5989 	return 0;
5990 }
5991 
5992 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5993 				 int size)
5994 {
5995 	int rc = -EINVAL;
5996 
5997 	if (kvmhv_vcpu_is_radix(vcpu)) {
5998 		rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
5999 
6000 		if (rc > 0)
6001 			rc = -EINVAL;
6002 	}
6003 
6004 	/* For now quadrants are the only way to access nested guest memory */
6005 	if (rc && vcpu->arch.nested)
6006 		rc = -EAGAIN;
6007 
6008 	return rc;
6009 }
6010 
6011 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
6012 				int size)
6013 {
6014 	int rc = -EINVAL;
6015 
6016 	if (kvmhv_vcpu_is_radix(vcpu)) {
6017 		rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
6018 
6019 		if (rc > 0)
6020 			rc = -EINVAL;
6021 	}
6022 
6023 	/* For now quadrants are the only way to access nested guest memory */
6024 	if (rc && vcpu->arch.nested)
6025 		rc = -EAGAIN;
6026 
6027 	return rc;
6028 }
6029 
6030 static void unpin_vpa_reset(struct kvm *kvm, struct kvmppc_vpa *vpa)
6031 {
6032 	unpin_vpa(kvm, vpa);
6033 	vpa->gpa = 0;
6034 	vpa->pinned_addr = NULL;
6035 	vpa->dirty = false;
6036 	vpa->update_pending = 0;
6037 }
6038 
6039 /*
6040  * Enable a guest to become a secure VM, or test whether
6041  * that could be enabled.
6042  * Called when the KVM_CAP_PPC_SECURE_GUEST capability is
6043  * tested (kvm == NULL) or enabled (kvm != NULL).
6044  */
6045 static int kvmhv_enable_svm(struct kvm *kvm)
6046 {
6047 	if (!kvmppc_uvmem_available())
6048 		return -EINVAL;
6049 	if (kvm)
6050 		kvm->arch.svm_enabled = 1;
6051 	return 0;
6052 }
6053 
6054 /*
6055  *  IOCTL handler to turn off secure mode of guest
6056  *
6057  * - Release all device pages
6058  * - Issue ucall to terminate the guest on the UV side
6059  * - Unpin the VPA pages.
6060  * - Reinit the partition scoped page tables
6061  */
6062 static int kvmhv_svm_off(struct kvm *kvm)
6063 {
6064 	struct kvm_vcpu *vcpu;
6065 	int mmu_was_ready;
6066 	int srcu_idx;
6067 	int ret = 0;
6068 	unsigned long i;
6069 
6070 	if (!(kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START))
6071 		return ret;
6072 
6073 	mutex_lock(&kvm->arch.mmu_setup_lock);
6074 	mmu_was_ready = kvm->arch.mmu_ready;
6075 	if (kvm->arch.mmu_ready) {
6076 		kvm->arch.mmu_ready = 0;
6077 		/* order mmu_ready vs. vcpus_running */
6078 		smp_mb();
6079 		if (atomic_read(&kvm->arch.vcpus_running)) {
6080 			kvm->arch.mmu_ready = 1;
6081 			ret = -EBUSY;
6082 			goto out;
6083 		}
6084 	}
6085 
6086 	srcu_idx = srcu_read_lock(&kvm->srcu);
6087 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
6088 		struct kvm_memory_slot *memslot;
6089 		struct kvm_memslots *slots = __kvm_memslots(kvm, i);
6090 		int bkt;
6091 
6092 		if (!slots)
6093 			continue;
6094 
6095 		kvm_for_each_memslot(memslot, bkt, slots) {
6096 			kvmppc_uvmem_drop_pages(memslot, kvm, true);
6097 			uv_unregister_mem_slot(kvm->arch.lpid, memslot->id);
6098 		}
6099 	}
6100 	srcu_read_unlock(&kvm->srcu, srcu_idx);
6101 
6102 	ret = uv_svm_terminate(kvm->arch.lpid);
6103 	if (ret != U_SUCCESS) {
6104 		ret = -EINVAL;
6105 		goto out;
6106 	}
6107 
6108 	/*
6109 	 * When secure guest is reset, all the guest pages are sent
6110 	 * to UV via UV_PAGE_IN before the non-boot vcpus get a
6111 	 * chance to run and unpin their VPA pages. Unpinning of all
6112 	 * VPA pages is done here explicitly so that VPA pages
6113 	 * can be migrated to the secure side.
6114 	 *
6115 	 * This is required to for the secure SMP guest to reboot
6116 	 * correctly.
6117 	 */
6118 	kvm_for_each_vcpu(i, vcpu, kvm) {
6119 		spin_lock(&vcpu->arch.vpa_update_lock);
6120 		unpin_vpa_reset(kvm, &vcpu->arch.dtl);
6121 		unpin_vpa_reset(kvm, &vcpu->arch.slb_shadow);
6122 		unpin_vpa_reset(kvm, &vcpu->arch.vpa);
6123 		spin_unlock(&vcpu->arch.vpa_update_lock);
6124 	}
6125 
6126 	kvmppc_setup_partition_table(kvm);
6127 	kvm->arch.secure_guest = 0;
6128 	kvm->arch.mmu_ready = mmu_was_ready;
6129 out:
6130 	mutex_unlock(&kvm->arch.mmu_setup_lock);
6131 	return ret;
6132 }
6133 
6134 static int kvmhv_enable_dawr1(struct kvm *kvm)
6135 {
6136 	if (!cpu_has_feature(CPU_FTR_DAWR1))
6137 		return -ENODEV;
6138 
6139 	/* kvm == NULL means the caller is testing if the capability exists */
6140 	if (kvm)
6141 		kvm->arch.dawr1_enabled = true;
6142 	return 0;
6143 }
6144 
6145 static bool kvmppc_hash_v3_possible(void)
6146 {
6147 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
6148 		return false;
6149 
6150 	if (!cpu_has_feature(CPU_FTR_HVMODE))
6151 		return false;
6152 
6153 	/*
6154 	 * POWER9 chips before version 2.02 can't have some threads in
6155 	 * HPT mode and some in radix mode on the same core.
6156 	 */
6157 	if (radix_enabled()) {
6158 		unsigned int pvr = mfspr(SPRN_PVR);
6159 		if ((pvr >> 16) == PVR_POWER9 &&
6160 		    (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
6161 		     ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
6162 			return false;
6163 	}
6164 
6165 	return true;
6166 }
6167 
6168 static struct kvmppc_ops kvm_ops_hv = {
6169 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
6170 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
6171 	.get_one_reg = kvmppc_get_one_reg_hv,
6172 	.set_one_reg = kvmppc_set_one_reg_hv,
6173 	.vcpu_load   = kvmppc_core_vcpu_load_hv,
6174 	.vcpu_put    = kvmppc_core_vcpu_put_hv,
6175 	.inject_interrupt = kvmppc_inject_interrupt_hv,
6176 	.set_msr     = kvmppc_set_msr_hv,
6177 	.vcpu_run    = kvmppc_vcpu_run_hv,
6178 	.vcpu_create = kvmppc_core_vcpu_create_hv,
6179 	.vcpu_free   = kvmppc_core_vcpu_free_hv,
6180 	.check_requests = kvmppc_core_check_requests_hv,
6181 	.get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
6182 	.flush_memslot  = kvmppc_core_flush_memslot_hv,
6183 	.prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
6184 	.commit_memory_region  = kvmppc_core_commit_memory_region_hv,
6185 	.unmap_gfn_range = kvm_unmap_gfn_range_hv,
6186 	.age_gfn = kvm_age_gfn_hv,
6187 	.test_age_gfn = kvm_test_age_gfn_hv,
6188 	.set_spte_gfn = kvm_set_spte_gfn_hv,
6189 	.free_memslot = kvmppc_core_free_memslot_hv,
6190 	.init_vm =  kvmppc_core_init_vm_hv,
6191 	.destroy_vm = kvmppc_core_destroy_vm_hv,
6192 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
6193 	.emulate_op = kvmppc_core_emulate_op_hv,
6194 	.emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
6195 	.emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
6196 	.fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
6197 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
6198 	.hcall_implemented = kvmppc_hcall_impl_hv,
6199 #ifdef CONFIG_KVM_XICS
6200 	.irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
6201 	.irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
6202 #endif
6203 	.configure_mmu = kvmhv_configure_mmu,
6204 	.get_rmmu_info = kvmhv_get_rmmu_info,
6205 	.set_smt_mode = kvmhv_set_smt_mode,
6206 	.enable_nested = kvmhv_enable_nested,
6207 	.load_from_eaddr = kvmhv_load_from_eaddr,
6208 	.store_to_eaddr = kvmhv_store_to_eaddr,
6209 	.enable_svm = kvmhv_enable_svm,
6210 	.svm_off = kvmhv_svm_off,
6211 	.enable_dawr1 = kvmhv_enable_dawr1,
6212 	.hash_v3_possible = kvmppc_hash_v3_possible,
6213 	.create_vcpu_debugfs = kvmppc_arch_create_vcpu_debugfs_hv,
6214 	.create_vm_debugfs = kvmppc_arch_create_vm_debugfs_hv,
6215 };
6216 
6217 static int kvm_init_subcore_bitmap(void)
6218 {
6219 	int i, j;
6220 	int nr_cores = cpu_nr_cores();
6221 	struct sibling_subcore_state *sibling_subcore_state;
6222 
6223 	for (i = 0; i < nr_cores; i++) {
6224 		int first_cpu = i * threads_per_core;
6225 		int node = cpu_to_node(first_cpu);
6226 
6227 		/* Ignore if it is already allocated. */
6228 		if (paca_ptrs[first_cpu]->sibling_subcore_state)
6229 			continue;
6230 
6231 		sibling_subcore_state =
6232 			kzalloc_node(sizeof(struct sibling_subcore_state),
6233 							GFP_KERNEL, node);
6234 		if (!sibling_subcore_state)
6235 			return -ENOMEM;
6236 
6237 
6238 		for (j = 0; j < threads_per_core; j++) {
6239 			int cpu = first_cpu + j;
6240 
6241 			paca_ptrs[cpu]->sibling_subcore_state =
6242 						sibling_subcore_state;
6243 		}
6244 	}
6245 	return 0;
6246 }
6247 
6248 static int kvmppc_radix_possible(void)
6249 {
6250 	return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
6251 }
6252 
6253 static int kvmppc_book3s_init_hv(void)
6254 {
6255 	int r;
6256 
6257 	if (!tlbie_capable) {
6258 		pr_err("KVM-HV: Host does not support TLBIE\n");
6259 		return -ENODEV;
6260 	}
6261 
6262 	/*
6263 	 * FIXME!! Do we need to check on all cpus ?
6264 	 */
6265 	r = kvmppc_core_check_processor_compat_hv();
6266 	if (r < 0)
6267 		return -ENODEV;
6268 
6269 	r = kvmhv_nested_init();
6270 	if (r)
6271 		return r;
6272 
6273 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
6274 		r = kvm_init_subcore_bitmap();
6275 		if (r)
6276 			goto err;
6277 	}
6278 
6279 	/*
6280 	 * We need a way of accessing the XICS interrupt controller,
6281 	 * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
6282 	 * indirectly, via OPAL.
6283 	 */
6284 #ifdef CONFIG_SMP
6285 	if (!xics_on_xive() && !kvmhv_on_pseries() &&
6286 	    !local_paca->kvm_hstate.xics_phys) {
6287 		struct device_node *np;
6288 
6289 		np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
6290 		if (!np) {
6291 			pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
6292 			r = -ENODEV;
6293 			goto err;
6294 		}
6295 		/* presence of intc confirmed - node can be dropped again */
6296 		of_node_put(np);
6297 	}
6298 #endif
6299 
6300 	init_default_hcalls();
6301 
6302 	init_vcore_lists();
6303 
6304 	r = kvmppc_mmu_hv_init();
6305 	if (r)
6306 		goto err;
6307 
6308 	if (kvmppc_radix_possible()) {
6309 		r = kvmppc_radix_init();
6310 		if (r)
6311 			goto err;
6312 	}
6313 
6314 	r = kvmppc_uvmem_init();
6315 	if (r < 0) {
6316 		pr_err("KVM-HV: kvmppc_uvmem_init failed %d\n", r);
6317 		return r;
6318 	}
6319 
6320 	kvm_ops_hv.owner = THIS_MODULE;
6321 	kvmppc_hv_ops = &kvm_ops_hv;
6322 
6323 	return 0;
6324 
6325 err:
6326 	kvmhv_nested_exit();
6327 	kvmppc_radix_exit();
6328 
6329 	return r;
6330 }
6331 
6332 static void kvmppc_book3s_exit_hv(void)
6333 {
6334 	kvmppc_uvmem_free();
6335 	kvmppc_free_host_rm_ops();
6336 	if (kvmppc_radix_possible())
6337 		kvmppc_radix_exit();
6338 	kvmppc_hv_ops = NULL;
6339 	kvmhv_nested_exit();
6340 }
6341 
6342 module_init(kvmppc_book3s_init_hv);
6343 module_exit(kvmppc_book3s_exit_hv);
6344 MODULE_LICENSE("GPL");
6345 MODULE_ALIAS_MISCDEV(KVM_MINOR);
6346 MODULE_ALIAS("devname:kvm");
6347