xref: /openbmc/linux/arch/powerpc/kvm/book3s_hv.c (revision d4295e12)
1 /*
2  * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
3  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
4  *
5  * Authors:
6  *    Paul Mackerras <paulus@au1.ibm.com>
7  *    Alexander Graf <agraf@suse.de>
8  *    Kevin Wolf <mail@kevin-wolf.de>
9  *
10  * Description: KVM functions specific to running on Book 3S
11  * processors in hypervisor mode (specifically POWER7 and later).
12  *
13  * This file is derived from arch/powerpc/kvm/book3s.c,
14  * by Alexander Graf <agraf@suse.de>.
15  *
16  * This program is free software; you can redistribute it and/or modify
17  * it under the terms of the GNU General Public License, version 2, as
18  * published by the Free Software Foundation.
19  */
20 
21 #include <linux/kvm_host.h>
22 #include <linux/kernel.h>
23 #include <linux/err.h>
24 #include <linux/slab.h>
25 #include <linux/preempt.h>
26 #include <linux/sched/signal.h>
27 #include <linux/sched/stat.h>
28 #include <linux/delay.h>
29 #include <linux/export.h>
30 #include <linux/fs.h>
31 #include <linux/anon_inodes.h>
32 #include <linux/cpu.h>
33 #include <linux/cpumask.h>
34 #include <linux/spinlock.h>
35 #include <linux/page-flags.h>
36 #include <linux/srcu.h>
37 #include <linux/miscdevice.h>
38 #include <linux/debugfs.h>
39 #include <linux/gfp.h>
40 #include <linux/vmalloc.h>
41 #include <linux/highmem.h>
42 #include <linux/hugetlb.h>
43 #include <linux/kvm_irqfd.h>
44 #include <linux/irqbypass.h>
45 #include <linux/module.h>
46 #include <linux/compiler.h>
47 #include <linux/of.h>
48 
49 #include <asm/ftrace.h>
50 #include <asm/reg.h>
51 #include <asm/ppc-opcode.h>
52 #include <asm/asm-prototypes.h>
53 #include <asm/archrandom.h>
54 #include <asm/debug.h>
55 #include <asm/disassemble.h>
56 #include <asm/cputable.h>
57 #include <asm/cacheflush.h>
58 #include <linux/uaccess.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/processor.h>
65 #include <asm/cputhreads.h>
66 #include <asm/page.h>
67 #include <asm/hvcall.h>
68 #include <asm/switch_to.h>
69 #include <asm/smp.h>
70 #include <asm/dbell.h>
71 #include <asm/hmi.h>
72 #include <asm/pnv-pci.h>
73 #include <asm/mmu.h>
74 #include <asm/opal.h>
75 #include <asm/xics.h>
76 #include <asm/xive.h>
77 
78 #include "book3s.h"
79 
80 #define CREATE_TRACE_POINTS
81 #include "trace_hv.h"
82 
83 /* #define EXIT_DEBUG */
84 /* #define EXIT_DEBUG_SIMPLE */
85 /* #define EXIT_DEBUG_INT */
86 
87 /* Used to indicate that a guest page fault needs to be handled */
88 #define RESUME_PAGE_FAULT	(RESUME_GUEST | RESUME_FLAG_ARCH1)
89 /* Used to indicate that a guest passthrough interrupt needs to be handled */
90 #define RESUME_PASSTHROUGH	(RESUME_GUEST | RESUME_FLAG_ARCH2)
91 
92 /* Used as a "null" value for timebase values */
93 #define TB_NIL	(~(u64)0)
94 
95 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
96 
97 static int dynamic_mt_modes = 6;
98 module_param(dynamic_mt_modes, int, 0644);
99 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
100 static int target_smt_mode;
101 module_param(target_smt_mode, int, 0644);
102 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
103 
104 static bool indep_threads_mode = true;
105 module_param(indep_threads_mode, bool, S_IRUGO | S_IWUSR);
106 MODULE_PARM_DESC(indep_threads_mode, "Independent-threads mode (only on POWER9)");
107 
108 static bool one_vm_per_core;
109 module_param(one_vm_per_core, bool, S_IRUGO | S_IWUSR);
110 MODULE_PARM_DESC(one_vm_per_core, "Only run vCPUs from the same VM on a core (requires indep_threads_mode=N)");
111 
112 #ifdef CONFIG_KVM_XICS
113 static struct kernel_param_ops module_param_ops = {
114 	.set = param_set_int,
115 	.get = param_get_int,
116 };
117 
118 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
119 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
120 
121 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
122 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
123 #endif
124 
125 /* If set, guests are allowed to create and control nested guests */
126 static bool nested = true;
127 module_param(nested, bool, S_IRUGO | S_IWUSR);
128 MODULE_PARM_DESC(nested, "Enable nested virtualization (only on POWER9)");
129 
130 static inline bool nesting_enabled(struct kvm *kvm)
131 {
132 	return kvm->arch.nested_enable && kvm_is_radix(kvm);
133 }
134 
135 /* If set, the threads on each CPU core have to be in the same MMU mode */
136 static bool no_mixing_hpt_and_radix;
137 
138 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
139 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
140 
141 /*
142  * RWMR values for POWER8.  These control the rate at which PURR
143  * and SPURR count and should be set according to the number of
144  * online threads in the vcore being run.
145  */
146 #define RWMR_RPA_P8_1THREAD	0x164520C62609AECAUL
147 #define RWMR_RPA_P8_2THREAD	0x7FFF2908450D8DA9UL
148 #define RWMR_RPA_P8_3THREAD	0x164520C62609AECAUL
149 #define RWMR_RPA_P8_4THREAD	0x199A421245058DA9UL
150 #define RWMR_RPA_P8_5THREAD	0x164520C62609AECAUL
151 #define RWMR_RPA_P8_6THREAD	0x164520C62609AECAUL
152 #define RWMR_RPA_P8_7THREAD	0x164520C62609AECAUL
153 #define RWMR_RPA_P8_8THREAD	0x164520C62609AECAUL
154 
155 static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = {
156 	RWMR_RPA_P8_1THREAD,
157 	RWMR_RPA_P8_1THREAD,
158 	RWMR_RPA_P8_2THREAD,
159 	RWMR_RPA_P8_3THREAD,
160 	RWMR_RPA_P8_4THREAD,
161 	RWMR_RPA_P8_5THREAD,
162 	RWMR_RPA_P8_6THREAD,
163 	RWMR_RPA_P8_7THREAD,
164 	RWMR_RPA_P8_8THREAD,
165 };
166 
167 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
168 		int *ip)
169 {
170 	int i = *ip;
171 	struct kvm_vcpu *vcpu;
172 
173 	while (++i < MAX_SMT_THREADS) {
174 		vcpu = READ_ONCE(vc->runnable_threads[i]);
175 		if (vcpu) {
176 			*ip = i;
177 			return vcpu;
178 		}
179 	}
180 	return NULL;
181 }
182 
183 /* Used to traverse the list of runnable threads for a given vcore */
184 #define for_each_runnable_thread(i, vcpu, vc) \
185 	for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
186 
187 static bool kvmppc_ipi_thread(int cpu)
188 {
189 	unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
190 
191 	/* If we're a nested hypervisor, fall back to ordinary IPIs for now */
192 	if (kvmhv_on_pseries())
193 		return false;
194 
195 	/* On POWER9 we can use msgsnd to IPI any cpu */
196 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
197 		msg |= get_hard_smp_processor_id(cpu);
198 		smp_mb();
199 		__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
200 		return true;
201 	}
202 
203 	/* On POWER8 for IPIs to threads in the same core, use msgsnd */
204 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
205 		preempt_disable();
206 		if (cpu_first_thread_sibling(cpu) ==
207 		    cpu_first_thread_sibling(smp_processor_id())) {
208 			msg |= cpu_thread_in_core(cpu);
209 			smp_mb();
210 			__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
211 			preempt_enable();
212 			return true;
213 		}
214 		preempt_enable();
215 	}
216 
217 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
218 	if (cpu >= 0 && cpu < nr_cpu_ids) {
219 		if (paca_ptrs[cpu]->kvm_hstate.xics_phys) {
220 			xics_wake_cpu(cpu);
221 			return true;
222 		}
223 		opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
224 		return true;
225 	}
226 #endif
227 
228 	return false;
229 }
230 
231 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
232 {
233 	int cpu;
234 	struct swait_queue_head *wqp;
235 
236 	wqp = kvm_arch_vcpu_wq(vcpu);
237 	if (swq_has_sleeper(wqp)) {
238 		swake_up_one(wqp);
239 		++vcpu->stat.halt_wakeup;
240 	}
241 
242 	cpu = READ_ONCE(vcpu->arch.thread_cpu);
243 	if (cpu >= 0 && kvmppc_ipi_thread(cpu))
244 		return;
245 
246 	/* CPU points to the first thread of the core */
247 	cpu = vcpu->cpu;
248 	if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
249 		smp_send_reschedule(cpu);
250 }
251 
252 /*
253  * We use the vcpu_load/put functions to measure stolen time.
254  * Stolen time is counted as time when either the vcpu is able to
255  * run as part of a virtual core, but the task running the vcore
256  * is preempted or sleeping, or when the vcpu needs something done
257  * in the kernel by the task running the vcpu, but that task is
258  * preempted or sleeping.  Those two things have to be counted
259  * separately, since one of the vcpu tasks will take on the job
260  * of running the core, and the other vcpu tasks in the vcore will
261  * sleep waiting for it to do that, but that sleep shouldn't count
262  * as stolen time.
263  *
264  * Hence we accumulate stolen time when the vcpu can run as part of
265  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
266  * needs its task to do other things in the kernel (for example,
267  * service a page fault) in busy_stolen.  We don't accumulate
268  * stolen time for a vcore when it is inactive, or for a vcpu
269  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
270  * a misnomer; it means that the vcpu task is not executing in
271  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
272  * the kernel.  We don't have any way of dividing up that time
273  * between time that the vcpu is genuinely stopped, time that
274  * the task is actively working on behalf of the vcpu, and time
275  * that the task is preempted, so we don't count any of it as
276  * stolen.
277  *
278  * Updates to busy_stolen are protected by arch.tbacct_lock;
279  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
280  * lock.  The stolen times are measured in units of timebase ticks.
281  * (Note that the != TB_NIL checks below are purely defensive;
282  * they should never fail.)
283  */
284 
285 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
286 {
287 	unsigned long flags;
288 
289 	spin_lock_irqsave(&vc->stoltb_lock, flags);
290 	vc->preempt_tb = mftb();
291 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
292 }
293 
294 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
295 {
296 	unsigned long flags;
297 
298 	spin_lock_irqsave(&vc->stoltb_lock, flags);
299 	if (vc->preempt_tb != TB_NIL) {
300 		vc->stolen_tb += mftb() - vc->preempt_tb;
301 		vc->preempt_tb = TB_NIL;
302 	}
303 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
304 }
305 
306 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
307 {
308 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
309 	unsigned long flags;
310 
311 	/*
312 	 * We can test vc->runner without taking the vcore lock,
313 	 * because only this task ever sets vc->runner to this
314 	 * vcpu, and once it is set to this vcpu, only this task
315 	 * ever sets it to NULL.
316 	 */
317 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
318 		kvmppc_core_end_stolen(vc);
319 
320 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
321 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
322 	    vcpu->arch.busy_preempt != TB_NIL) {
323 		vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
324 		vcpu->arch.busy_preempt = TB_NIL;
325 	}
326 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
327 }
328 
329 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
330 {
331 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
332 	unsigned long flags;
333 
334 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
335 		kvmppc_core_start_stolen(vc);
336 
337 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
338 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
339 		vcpu->arch.busy_preempt = mftb();
340 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
341 }
342 
343 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
344 {
345 	/*
346 	 * Check for illegal transactional state bit combination
347 	 * and if we find it, force the TS field to a safe state.
348 	 */
349 	if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
350 		msr &= ~MSR_TS_MASK;
351 	vcpu->arch.shregs.msr = msr;
352 	kvmppc_end_cede(vcpu);
353 }
354 
355 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
356 {
357 	vcpu->arch.pvr = pvr;
358 }
359 
360 /* Dummy value used in computing PCR value below */
361 #define PCR_ARCH_300	(PCR_ARCH_207 << 1)
362 
363 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
364 {
365 	unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
366 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
367 
368 	/* We can (emulate) our own architecture version and anything older */
369 	if (cpu_has_feature(CPU_FTR_ARCH_300))
370 		host_pcr_bit = PCR_ARCH_300;
371 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
372 		host_pcr_bit = PCR_ARCH_207;
373 	else if (cpu_has_feature(CPU_FTR_ARCH_206))
374 		host_pcr_bit = PCR_ARCH_206;
375 	else
376 		host_pcr_bit = PCR_ARCH_205;
377 
378 	/* Determine lowest PCR bit needed to run guest in given PVR level */
379 	guest_pcr_bit = host_pcr_bit;
380 	if (arch_compat) {
381 		switch (arch_compat) {
382 		case PVR_ARCH_205:
383 			guest_pcr_bit = PCR_ARCH_205;
384 			break;
385 		case PVR_ARCH_206:
386 		case PVR_ARCH_206p:
387 			guest_pcr_bit = PCR_ARCH_206;
388 			break;
389 		case PVR_ARCH_207:
390 			guest_pcr_bit = PCR_ARCH_207;
391 			break;
392 		case PVR_ARCH_300:
393 			guest_pcr_bit = PCR_ARCH_300;
394 			break;
395 		default:
396 			return -EINVAL;
397 		}
398 	}
399 
400 	/* Check requested PCR bits don't exceed our capabilities */
401 	if (guest_pcr_bit > host_pcr_bit)
402 		return -EINVAL;
403 
404 	spin_lock(&vc->lock);
405 	vc->arch_compat = arch_compat;
406 	/* Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit */
407 	vc->pcr = host_pcr_bit - guest_pcr_bit;
408 	spin_unlock(&vc->lock);
409 
410 	return 0;
411 }
412 
413 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
414 {
415 	int r;
416 
417 	pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
418 	pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
419 	       vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
420 	for (r = 0; r < 16; ++r)
421 		pr_err("r%2d = %.16lx  r%d = %.16lx\n",
422 		       r, kvmppc_get_gpr(vcpu, r),
423 		       r+16, kvmppc_get_gpr(vcpu, r+16));
424 	pr_err("ctr = %.16lx  lr  = %.16lx\n",
425 	       vcpu->arch.regs.ctr, vcpu->arch.regs.link);
426 	pr_err("srr0 = %.16llx srr1 = %.16llx\n",
427 	       vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
428 	pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
429 	       vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
430 	pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
431 	       vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
432 	pr_err("cr = %.8lx  xer = %.16lx  dsisr = %.8x\n",
433 	       vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
434 	pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
435 	pr_err("fault dar = %.16lx dsisr = %.8x\n",
436 	       vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
437 	pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
438 	for (r = 0; r < vcpu->arch.slb_max; ++r)
439 		pr_err("  ESID = %.16llx VSID = %.16llx\n",
440 		       vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
441 	pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
442 	       vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
443 	       vcpu->arch.last_inst);
444 }
445 
446 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
447 {
448 	struct kvm_vcpu *ret;
449 
450 	mutex_lock(&kvm->lock);
451 	ret = kvm_get_vcpu_by_id(kvm, id);
452 	mutex_unlock(&kvm->lock);
453 	return ret;
454 }
455 
456 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
457 {
458 	vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
459 	vpa->yield_count = cpu_to_be32(1);
460 }
461 
462 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
463 		   unsigned long addr, unsigned long len)
464 {
465 	/* check address is cacheline aligned */
466 	if (addr & (L1_CACHE_BYTES - 1))
467 		return -EINVAL;
468 	spin_lock(&vcpu->arch.vpa_update_lock);
469 	if (v->next_gpa != addr || v->len != len) {
470 		v->next_gpa = addr;
471 		v->len = addr ? len : 0;
472 		v->update_pending = 1;
473 	}
474 	spin_unlock(&vcpu->arch.vpa_update_lock);
475 	return 0;
476 }
477 
478 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
479 struct reg_vpa {
480 	u32 dummy;
481 	union {
482 		__be16 hword;
483 		__be32 word;
484 	} length;
485 };
486 
487 static int vpa_is_registered(struct kvmppc_vpa *vpap)
488 {
489 	if (vpap->update_pending)
490 		return vpap->next_gpa != 0;
491 	return vpap->pinned_addr != NULL;
492 }
493 
494 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
495 				       unsigned long flags,
496 				       unsigned long vcpuid, unsigned long vpa)
497 {
498 	struct kvm *kvm = vcpu->kvm;
499 	unsigned long len, nb;
500 	void *va;
501 	struct kvm_vcpu *tvcpu;
502 	int err;
503 	int subfunc;
504 	struct kvmppc_vpa *vpap;
505 
506 	tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
507 	if (!tvcpu)
508 		return H_PARAMETER;
509 
510 	subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
511 	if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
512 	    subfunc == H_VPA_REG_SLB) {
513 		/* Registering new area - address must be cache-line aligned */
514 		if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
515 			return H_PARAMETER;
516 
517 		/* convert logical addr to kernel addr and read length */
518 		va = kvmppc_pin_guest_page(kvm, vpa, &nb);
519 		if (va == NULL)
520 			return H_PARAMETER;
521 		if (subfunc == H_VPA_REG_VPA)
522 			len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
523 		else
524 			len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
525 		kvmppc_unpin_guest_page(kvm, va, vpa, false);
526 
527 		/* Check length */
528 		if (len > nb || len < sizeof(struct reg_vpa))
529 			return H_PARAMETER;
530 	} else {
531 		vpa = 0;
532 		len = 0;
533 	}
534 
535 	err = H_PARAMETER;
536 	vpap = NULL;
537 	spin_lock(&tvcpu->arch.vpa_update_lock);
538 
539 	switch (subfunc) {
540 	case H_VPA_REG_VPA:		/* register VPA */
541 		/*
542 		 * The size of our lppaca is 1kB because of the way we align
543 		 * it for the guest to avoid crossing a 4kB boundary. We only
544 		 * use 640 bytes of the structure though, so we should accept
545 		 * clients that set a size of 640.
546 		 */
547 		BUILD_BUG_ON(sizeof(struct lppaca) != 640);
548 		if (len < sizeof(struct lppaca))
549 			break;
550 		vpap = &tvcpu->arch.vpa;
551 		err = 0;
552 		break;
553 
554 	case H_VPA_REG_DTL:		/* register DTL */
555 		if (len < sizeof(struct dtl_entry))
556 			break;
557 		len -= len % sizeof(struct dtl_entry);
558 
559 		/* Check that they have previously registered a VPA */
560 		err = H_RESOURCE;
561 		if (!vpa_is_registered(&tvcpu->arch.vpa))
562 			break;
563 
564 		vpap = &tvcpu->arch.dtl;
565 		err = 0;
566 		break;
567 
568 	case H_VPA_REG_SLB:		/* register SLB shadow buffer */
569 		/* Check that they have previously registered a VPA */
570 		err = H_RESOURCE;
571 		if (!vpa_is_registered(&tvcpu->arch.vpa))
572 			break;
573 
574 		vpap = &tvcpu->arch.slb_shadow;
575 		err = 0;
576 		break;
577 
578 	case H_VPA_DEREG_VPA:		/* deregister VPA */
579 		/* Check they don't still have a DTL or SLB buf registered */
580 		err = H_RESOURCE;
581 		if (vpa_is_registered(&tvcpu->arch.dtl) ||
582 		    vpa_is_registered(&tvcpu->arch.slb_shadow))
583 			break;
584 
585 		vpap = &tvcpu->arch.vpa;
586 		err = 0;
587 		break;
588 
589 	case H_VPA_DEREG_DTL:		/* deregister DTL */
590 		vpap = &tvcpu->arch.dtl;
591 		err = 0;
592 		break;
593 
594 	case H_VPA_DEREG_SLB:		/* deregister SLB shadow buffer */
595 		vpap = &tvcpu->arch.slb_shadow;
596 		err = 0;
597 		break;
598 	}
599 
600 	if (vpap) {
601 		vpap->next_gpa = vpa;
602 		vpap->len = len;
603 		vpap->update_pending = 1;
604 	}
605 
606 	spin_unlock(&tvcpu->arch.vpa_update_lock);
607 
608 	return err;
609 }
610 
611 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
612 {
613 	struct kvm *kvm = vcpu->kvm;
614 	void *va;
615 	unsigned long nb;
616 	unsigned long gpa;
617 
618 	/*
619 	 * We need to pin the page pointed to by vpap->next_gpa,
620 	 * but we can't call kvmppc_pin_guest_page under the lock
621 	 * as it does get_user_pages() and down_read().  So we
622 	 * have to drop the lock, pin the page, then get the lock
623 	 * again and check that a new area didn't get registered
624 	 * in the meantime.
625 	 */
626 	for (;;) {
627 		gpa = vpap->next_gpa;
628 		spin_unlock(&vcpu->arch.vpa_update_lock);
629 		va = NULL;
630 		nb = 0;
631 		if (gpa)
632 			va = kvmppc_pin_guest_page(kvm, gpa, &nb);
633 		spin_lock(&vcpu->arch.vpa_update_lock);
634 		if (gpa == vpap->next_gpa)
635 			break;
636 		/* sigh... unpin that one and try again */
637 		if (va)
638 			kvmppc_unpin_guest_page(kvm, va, gpa, false);
639 	}
640 
641 	vpap->update_pending = 0;
642 	if (va && nb < vpap->len) {
643 		/*
644 		 * If it's now too short, it must be that userspace
645 		 * has changed the mappings underlying guest memory,
646 		 * so unregister the region.
647 		 */
648 		kvmppc_unpin_guest_page(kvm, va, gpa, false);
649 		va = NULL;
650 	}
651 	if (vpap->pinned_addr)
652 		kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
653 					vpap->dirty);
654 	vpap->gpa = gpa;
655 	vpap->pinned_addr = va;
656 	vpap->dirty = false;
657 	if (va)
658 		vpap->pinned_end = va + vpap->len;
659 }
660 
661 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
662 {
663 	if (!(vcpu->arch.vpa.update_pending ||
664 	      vcpu->arch.slb_shadow.update_pending ||
665 	      vcpu->arch.dtl.update_pending))
666 		return;
667 
668 	spin_lock(&vcpu->arch.vpa_update_lock);
669 	if (vcpu->arch.vpa.update_pending) {
670 		kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
671 		if (vcpu->arch.vpa.pinned_addr)
672 			init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
673 	}
674 	if (vcpu->arch.dtl.update_pending) {
675 		kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
676 		vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
677 		vcpu->arch.dtl_index = 0;
678 	}
679 	if (vcpu->arch.slb_shadow.update_pending)
680 		kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
681 	spin_unlock(&vcpu->arch.vpa_update_lock);
682 }
683 
684 /*
685  * Return the accumulated stolen time for the vcore up until `now'.
686  * The caller should hold the vcore lock.
687  */
688 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
689 {
690 	u64 p;
691 	unsigned long flags;
692 
693 	spin_lock_irqsave(&vc->stoltb_lock, flags);
694 	p = vc->stolen_tb;
695 	if (vc->vcore_state != VCORE_INACTIVE &&
696 	    vc->preempt_tb != TB_NIL)
697 		p += now - vc->preempt_tb;
698 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
699 	return p;
700 }
701 
702 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
703 				    struct kvmppc_vcore *vc)
704 {
705 	struct dtl_entry *dt;
706 	struct lppaca *vpa;
707 	unsigned long stolen;
708 	unsigned long core_stolen;
709 	u64 now;
710 	unsigned long flags;
711 
712 	dt = vcpu->arch.dtl_ptr;
713 	vpa = vcpu->arch.vpa.pinned_addr;
714 	now = mftb();
715 	core_stolen = vcore_stolen_time(vc, now);
716 	stolen = core_stolen - vcpu->arch.stolen_logged;
717 	vcpu->arch.stolen_logged = core_stolen;
718 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
719 	stolen += vcpu->arch.busy_stolen;
720 	vcpu->arch.busy_stolen = 0;
721 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
722 	if (!dt || !vpa)
723 		return;
724 	memset(dt, 0, sizeof(struct dtl_entry));
725 	dt->dispatch_reason = 7;
726 	dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
727 	dt->timebase = cpu_to_be64(now + vc->tb_offset);
728 	dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
729 	dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
730 	dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
731 	++dt;
732 	if (dt == vcpu->arch.dtl.pinned_end)
733 		dt = vcpu->arch.dtl.pinned_addr;
734 	vcpu->arch.dtl_ptr = dt;
735 	/* order writing *dt vs. writing vpa->dtl_idx */
736 	smp_wmb();
737 	vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
738 	vcpu->arch.dtl.dirty = true;
739 }
740 
741 /* See if there is a doorbell interrupt pending for a vcpu */
742 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
743 {
744 	int thr;
745 	struct kvmppc_vcore *vc;
746 
747 	if (vcpu->arch.doorbell_request)
748 		return true;
749 	/*
750 	 * Ensure that the read of vcore->dpdes comes after the read
751 	 * of vcpu->doorbell_request.  This barrier matches the
752 	 * smb_wmb() in kvmppc_guest_entry_inject().
753 	 */
754 	smp_rmb();
755 	vc = vcpu->arch.vcore;
756 	thr = vcpu->vcpu_id - vc->first_vcpuid;
757 	return !!(vc->dpdes & (1 << thr));
758 }
759 
760 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
761 {
762 	if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
763 		return true;
764 	if ((!vcpu->arch.vcore->arch_compat) &&
765 	    cpu_has_feature(CPU_FTR_ARCH_207S))
766 		return true;
767 	return false;
768 }
769 
770 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
771 			     unsigned long resource, unsigned long value1,
772 			     unsigned long value2)
773 {
774 	switch (resource) {
775 	case H_SET_MODE_RESOURCE_SET_CIABR:
776 		if (!kvmppc_power8_compatible(vcpu))
777 			return H_P2;
778 		if (value2)
779 			return H_P4;
780 		if (mflags)
781 			return H_UNSUPPORTED_FLAG_START;
782 		/* Guests can't breakpoint the hypervisor */
783 		if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
784 			return H_P3;
785 		vcpu->arch.ciabr  = value1;
786 		return H_SUCCESS;
787 	case H_SET_MODE_RESOURCE_SET_DAWR:
788 		if (!kvmppc_power8_compatible(vcpu))
789 			return H_P2;
790 		if (!ppc_breakpoint_available())
791 			return H_P2;
792 		if (mflags)
793 			return H_UNSUPPORTED_FLAG_START;
794 		if (value2 & DABRX_HYP)
795 			return H_P4;
796 		vcpu->arch.dawr  = value1;
797 		vcpu->arch.dawrx = value2;
798 		return H_SUCCESS;
799 	default:
800 		return H_TOO_HARD;
801 	}
802 }
803 
804 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
805 {
806 	struct kvmppc_vcore *vcore = target->arch.vcore;
807 
808 	/*
809 	 * We expect to have been called by the real mode handler
810 	 * (kvmppc_rm_h_confer()) which would have directly returned
811 	 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
812 	 * have useful work to do and should not confer) so we don't
813 	 * recheck that here.
814 	 */
815 
816 	spin_lock(&vcore->lock);
817 	if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
818 	    vcore->vcore_state != VCORE_INACTIVE &&
819 	    vcore->runner)
820 		target = vcore->runner;
821 	spin_unlock(&vcore->lock);
822 
823 	return kvm_vcpu_yield_to(target);
824 }
825 
826 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
827 {
828 	int yield_count = 0;
829 	struct lppaca *lppaca;
830 
831 	spin_lock(&vcpu->arch.vpa_update_lock);
832 	lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
833 	if (lppaca)
834 		yield_count = be32_to_cpu(lppaca->yield_count);
835 	spin_unlock(&vcpu->arch.vpa_update_lock);
836 	return yield_count;
837 }
838 
839 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
840 {
841 	unsigned long req = kvmppc_get_gpr(vcpu, 3);
842 	unsigned long target, ret = H_SUCCESS;
843 	int yield_count;
844 	struct kvm_vcpu *tvcpu;
845 	int idx, rc;
846 
847 	if (req <= MAX_HCALL_OPCODE &&
848 	    !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
849 		return RESUME_HOST;
850 
851 	switch (req) {
852 	case H_CEDE:
853 		break;
854 	case H_PROD:
855 		target = kvmppc_get_gpr(vcpu, 4);
856 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
857 		if (!tvcpu) {
858 			ret = H_PARAMETER;
859 			break;
860 		}
861 		tvcpu->arch.prodded = 1;
862 		smp_mb();
863 		if (tvcpu->arch.ceded)
864 			kvmppc_fast_vcpu_kick_hv(tvcpu);
865 		break;
866 	case H_CONFER:
867 		target = kvmppc_get_gpr(vcpu, 4);
868 		if (target == -1)
869 			break;
870 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
871 		if (!tvcpu) {
872 			ret = H_PARAMETER;
873 			break;
874 		}
875 		yield_count = kvmppc_get_gpr(vcpu, 5);
876 		if (kvmppc_get_yield_count(tvcpu) != yield_count)
877 			break;
878 		kvm_arch_vcpu_yield_to(tvcpu);
879 		break;
880 	case H_REGISTER_VPA:
881 		ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
882 					kvmppc_get_gpr(vcpu, 5),
883 					kvmppc_get_gpr(vcpu, 6));
884 		break;
885 	case H_RTAS:
886 		if (list_empty(&vcpu->kvm->arch.rtas_tokens))
887 			return RESUME_HOST;
888 
889 		idx = srcu_read_lock(&vcpu->kvm->srcu);
890 		rc = kvmppc_rtas_hcall(vcpu);
891 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
892 
893 		if (rc == -ENOENT)
894 			return RESUME_HOST;
895 		else if (rc == 0)
896 			break;
897 
898 		/* Send the error out to userspace via KVM_RUN */
899 		return rc;
900 	case H_LOGICAL_CI_LOAD:
901 		ret = kvmppc_h_logical_ci_load(vcpu);
902 		if (ret == H_TOO_HARD)
903 			return RESUME_HOST;
904 		break;
905 	case H_LOGICAL_CI_STORE:
906 		ret = kvmppc_h_logical_ci_store(vcpu);
907 		if (ret == H_TOO_HARD)
908 			return RESUME_HOST;
909 		break;
910 	case H_SET_MODE:
911 		ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
912 					kvmppc_get_gpr(vcpu, 5),
913 					kvmppc_get_gpr(vcpu, 6),
914 					kvmppc_get_gpr(vcpu, 7));
915 		if (ret == H_TOO_HARD)
916 			return RESUME_HOST;
917 		break;
918 	case H_XIRR:
919 	case H_CPPR:
920 	case H_EOI:
921 	case H_IPI:
922 	case H_IPOLL:
923 	case H_XIRR_X:
924 		if (kvmppc_xics_enabled(vcpu)) {
925 			if (xive_enabled()) {
926 				ret = H_NOT_AVAILABLE;
927 				return RESUME_GUEST;
928 			}
929 			ret = kvmppc_xics_hcall(vcpu, req);
930 			break;
931 		}
932 		return RESUME_HOST;
933 	case H_SET_DABR:
934 		ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
935 		break;
936 	case H_SET_XDABR:
937 		ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
938 						kvmppc_get_gpr(vcpu, 5));
939 		break;
940 	case H_GET_TCE:
941 		ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
942 						kvmppc_get_gpr(vcpu, 5));
943 		if (ret == H_TOO_HARD)
944 			return RESUME_HOST;
945 		break;
946 	case H_PUT_TCE:
947 		ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
948 						kvmppc_get_gpr(vcpu, 5),
949 						kvmppc_get_gpr(vcpu, 6));
950 		if (ret == H_TOO_HARD)
951 			return RESUME_HOST;
952 		break;
953 	case H_PUT_TCE_INDIRECT:
954 		ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
955 						kvmppc_get_gpr(vcpu, 5),
956 						kvmppc_get_gpr(vcpu, 6),
957 						kvmppc_get_gpr(vcpu, 7));
958 		if (ret == H_TOO_HARD)
959 			return RESUME_HOST;
960 		break;
961 	case H_STUFF_TCE:
962 		ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
963 						kvmppc_get_gpr(vcpu, 5),
964 						kvmppc_get_gpr(vcpu, 6),
965 						kvmppc_get_gpr(vcpu, 7));
966 		if (ret == H_TOO_HARD)
967 			return RESUME_HOST;
968 		break;
969 	case H_RANDOM:
970 		if (!powernv_get_random_long(&vcpu->arch.regs.gpr[4]))
971 			ret = H_HARDWARE;
972 		break;
973 
974 	case H_SET_PARTITION_TABLE:
975 		ret = H_FUNCTION;
976 		if (nesting_enabled(vcpu->kvm))
977 			ret = kvmhv_set_partition_table(vcpu);
978 		break;
979 	case H_ENTER_NESTED:
980 		ret = H_FUNCTION;
981 		if (!nesting_enabled(vcpu->kvm))
982 			break;
983 		ret = kvmhv_enter_nested_guest(vcpu);
984 		if (ret == H_INTERRUPT) {
985 			kvmppc_set_gpr(vcpu, 3, 0);
986 			return -EINTR;
987 		}
988 		break;
989 	case H_TLB_INVALIDATE:
990 		ret = H_FUNCTION;
991 		if (nesting_enabled(vcpu->kvm))
992 			ret = kvmhv_do_nested_tlbie(vcpu);
993 		break;
994 
995 	default:
996 		return RESUME_HOST;
997 	}
998 	kvmppc_set_gpr(vcpu, 3, ret);
999 	vcpu->arch.hcall_needed = 0;
1000 	return RESUME_GUEST;
1001 }
1002 
1003 /*
1004  * Handle H_CEDE in the nested virtualization case where we haven't
1005  * called the real-mode hcall handlers in book3s_hv_rmhandlers.S.
1006  * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1007  * that the cede logic in kvmppc_run_single_vcpu() works properly.
1008  */
1009 static void kvmppc_nested_cede(struct kvm_vcpu *vcpu)
1010 {
1011 	vcpu->arch.shregs.msr |= MSR_EE;
1012 	vcpu->arch.ceded = 1;
1013 	smp_mb();
1014 	if (vcpu->arch.prodded) {
1015 		vcpu->arch.prodded = 0;
1016 		smp_mb();
1017 		vcpu->arch.ceded = 0;
1018 	}
1019 }
1020 
1021 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1022 {
1023 	switch (cmd) {
1024 	case H_CEDE:
1025 	case H_PROD:
1026 	case H_CONFER:
1027 	case H_REGISTER_VPA:
1028 	case H_SET_MODE:
1029 	case H_LOGICAL_CI_LOAD:
1030 	case H_LOGICAL_CI_STORE:
1031 #ifdef CONFIG_KVM_XICS
1032 	case H_XIRR:
1033 	case H_CPPR:
1034 	case H_EOI:
1035 	case H_IPI:
1036 	case H_IPOLL:
1037 	case H_XIRR_X:
1038 #endif
1039 		return 1;
1040 	}
1041 
1042 	/* See if it's in the real-mode table */
1043 	return kvmppc_hcall_impl_hv_realmode(cmd);
1044 }
1045 
1046 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
1047 					struct kvm_vcpu *vcpu)
1048 {
1049 	u32 last_inst;
1050 
1051 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1052 					EMULATE_DONE) {
1053 		/*
1054 		 * Fetch failed, so return to guest and
1055 		 * try executing it again.
1056 		 */
1057 		return RESUME_GUEST;
1058 	}
1059 
1060 	if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
1061 		run->exit_reason = KVM_EXIT_DEBUG;
1062 		run->debug.arch.address = kvmppc_get_pc(vcpu);
1063 		return RESUME_HOST;
1064 	} else {
1065 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1066 		return RESUME_GUEST;
1067 	}
1068 }
1069 
1070 static void do_nothing(void *x)
1071 {
1072 }
1073 
1074 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1075 {
1076 	int thr, cpu, pcpu, nthreads;
1077 	struct kvm_vcpu *v;
1078 	unsigned long dpdes;
1079 
1080 	nthreads = vcpu->kvm->arch.emul_smt_mode;
1081 	dpdes = 0;
1082 	cpu = vcpu->vcpu_id & ~(nthreads - 1);
1083 	for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1084 		v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1085 		if (!v)
1086 			continue;
1087 		/*
1088 		 * If the vcpu is currently running on a physical cpu thread,
1089 		 * interrupt it in order to pull it out of the guest briefly,
1090 		 * which will update its vcore->dpdes value.
1091 		 */
1092 		pcpu = READ_ONCE(v->cpu);
1093 		if (pcpu >= 0)
1094 			smp_call_function_single(pcpu, do_nothing, NULL, 1);
1095 		if (kvmppc_doorbell_pending(v))
1096 			dpdes |= 1 << thr;
1097 	}
1098 	return dpdes;
1099 }
1100 
1101 /*
1102  * On POWER9, emulate doorbell-related instructions in order to
1103  * give the guest the illusion of running on a multi-threaded core.
1104  * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1105  * and mfspr DPDES.
1106  */
1107 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1108 {
1109 	u32 inst, rb, thr;
1110 	unsigned long arg;
1111 	struct kvm *kvm = vcpu->kvm;
1112 	struct kvm_vcpu *tvcpu;
1113 
1114 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1115 		return RESUME_GUEST;
1116 	if (get_op(inst) != 31)
1117 		return EMULATE_FAIL;
1118 	rb = get_rb(inst);
1119 	thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1120 	switch (get_xop(inst)) {
1121 	case OP_31_XOP_MSGSNDP:
1122 		arg = kvmppc_get_gpr(vcpu, rb);
1123 		if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1124 			break;
1125 		arg &= 0x3f;
1126 		if (arg >= kvm->arch.emul_smt_mode)
1127 			break;
1128 		tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1129 		if (!tvcpu)
1130 			break;
1131 		if (!tvcpu->arch.doorbell_request) {
1132 			tvcpu->arch.doorbell_request = 1;
1133 			kvmppc_fast_vcpu_kick_hv(tvcpu);
1134 		}
1135 		break;
1136 	case OP_31_XOP_MSGCLRP:
1137 		arg = kvmppc_get_gpr(vcpu, rb);
1138 		if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1139 			break;
1140 		vcpu->arch.vcore->dpdes = 0;
1141 		vcpu->arch.doorbell_request = 0;
1142 		break;
1143 	case OP_31_XOP_MFSPR:
1144 		switch (get_sprn(inst)) {
1145 		case SPRN_TIR:
1146 			arg = thr;
1147 			break;
1148 		case SPRN_DPDES:
1149 			arg = kvmppc_read_dpdes(vcpu);
1150 			break;
1151 		default:
1152 			return EMULATE_FAIL;
1153 		}
1154 		kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1155 		break;
1156 	default:
1157 		return EMULATE_FAIL;
1158 	}
1159 	kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1160 	return RESUME_GUEST;
1161 }
1162 
1163 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
1164 				 struct task_struct *tsk)
1165 {
1166 	int r = RESUME_HOST;
1167 
1168 	vcpu->stat.sum_exits++;
1169 
1170 	/*
1171 	 * This can happen if an interrupt occurs in the last stages
1172 	 * of guest entry or the first stages of guest exit (i.e. after
1173 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1174 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1175 	 * That can happen due to a bug, or due to a machine check
1176 	 * occurring at just the wrong time.
1177 	 */
1178 	if (vcpu->arch.shregs.msr & MSR_HV) {
1179 		printk(KERN_EMERG "KVM trap in HV mode!\n");
1180 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1181 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1182 			vcpu->arch.shregs.msr);
1183 		kvmppc_dump_regs(vcpu);
1184 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1185 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1186 		return RESUME_HOST;
1187 	}
1188 	run->exit_reason = KVM_EXIT_UNKNOWN;
1189 	run->ready_for_interrupt_injection = 1;
1190 	switch (vcpu->arch.trap) {
1191 	/* We're good on these - the host merely wanted to get our attention */
1192 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1193 		vcpu->stat.dec_exits++;
1194 		r = RESUME_GUEST;
1195 		break;
1196 	case BOOK3S_INTERRUPT_EXTERNAL:
1197 	case BOOK3S_INTERRUPT_H_DOORBELL:
1198 	case BOOK3S_INTERRUPT_H_VIRT:
1199 		vcpu->stat.ext_intr_exits++;
1200 		r = RESUME_GUEST;
1201 		break;
1202 	/* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1203 	case BOOK3S_INTERRUPT_HMI:
1204 	case BOOK3S_INTERRUPT_PERFMON:
1205 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
1206 		r = RESUME_GUEST;
1207 		break;
1208 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
1209 		/* Exit to guest with KVM_EXIT_NMI as exit reason */
1210 		run->exit_reason = KVM_EXIT_NMI;
1211 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1212 		/* Clear out the old NMI status from run->flags */
1213 		run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1214 		/* Now set the NMI status */
1215 		if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1216 			run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1217 		else
1218 			run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1219 
1220 		r = RESUME_HOST;
1221 		/* Print the MCE event to host console. */
1222 		machine_check_print_event_info(&vcpu->arch.mce_evt, false);
1223 		break;
1224 	case BOOK3S_INTERRUPT_PROGRAM:
1225 	{
1226 		ulong flags;
1227 		/*
1228 		 * Normally program interrupts are delivered directly
1229 		 * to the guest by the hardware, but we can get here
1230 		 * as a result of a hypervisor emulation interrupt
1231 		 * (e40) getting turned into a 700 by BML RTAS.
1232 		 */
1233 		flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1234 		kvmppc_core_queue_program(vcpu, flags);
1235 		r = RESUME_GUEST;
1236 		break;
1237 	}
1238 	case BOOK3S_INTERRUPT_SYSCALL:
1239 	{
1240 		/* hcall - punt to userspace */
1241 		int i;
1242 
1243 		/* hypercall with MSR_PR has already been handled in rmode,
1244 		 * and never reaches here.
1245 		 */
1246 
1247 		run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1248 		for (i = 0; i < 9; ++i)
1249 			run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1250 		run->exit_reason = KVM_EXIT_PAPR_HCALL;
1251 		vcpu->arch.hcall_needed = 1;
1252 		r = RESUME_HOST;
1253 		break;
1254 	}
1255 	/*
1256 	 * We get these next two if the guest accesses a page which it thinks
1257 	 * it has mapped but which is not actually present, either because
1258 	 * it is for an emulated I/O device or because the corresonding
1259 	 * host page has been paged out.  Any other HDSI/HISI interrupts
1260 	 * have been handled already.
1261 	 */
1262 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1263 		r = RESUME_PAGE_FAULT;
1264 		break;
1265 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
1266 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1267 		vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr &
1268 			DSISR_SRR1_MATCH_64S;
1269 		if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1270 			vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1271 		r = RESUME_PAGE_FAULT;
1272 		break;
1273 	/*
1274 	 * This occurs if the guest executes an illegal instruction.
1275 	 * If the guest debug is disabled, generate a program interrupt
1276 	 * to the guest. If guest debug is enabled, we need to check
1277 	 * whether the instruction is a software breakpoint instruction.
1278 	 * Accordingly return to Guest or Host.
1279 	 */
1280 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1281 		if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1282 			vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1283 				swab32(vcpu->arch.emul_inst) :
1284 				vcpu->arch.emul_inst;
1285 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1286 			r = kvmppc_emulate_debug_inst(run, vcpu);
1287 		} else {
1288 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1289 			r = RESUME_GUEST;
1290 		}
1291 		break;
1292 	/*
1293 	 * This occurs if the guest (kernel or userspace), does something that
1294 	 * is prohibited by HFSCR.
1295 	 * On POWER9, this could be a doorbell instruction that we need
1296 	 * to emulate.
1297 	 * Otherwise, we just generate a program interrupt to the guest.
1298 	 */
1299 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1300 		r = EMULATE_FAIL;
1301 		if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1302 		    cpu_has_feature(CPU_FTR_ARCH_300))
1303 			r = kvmppc_emulate_doorbell_instr(vcpu);
1304 		if (r == EMULATE_FAIL) {
1305 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1306 			r = RESUME_GUEST;
1307 		}
1308 		break;
1309 
1310 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1311 	case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1312 		/*
1313 		 * This occurs for various TM-related instructions that
1314 		 * we need to emulate on POWER9 DD2.2.  We have already
1315 		 * handled the cases where the guest was in real-suspend
1316 		 * mode and was transitioning to transactional state.
1317 		 */
1318 		r = kvmhv_p9_tm_emulation(vcpu);
1319 		break;
1320 #endif
1321 
1322 	case BOOK3S_INTERRUPT_HV_RM_HARD:
1323 		r = RESUME_PASSTHROUGH;
1324 		break;
1325 	default:
1326 		kvmppc_dump_regs(vcpu);
1327 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1328 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1329 			vcpu->arch.shregs.msr);
1330 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1331 		r = RESUME_HOST;
1332 		break;
1333 	}
1334 
1335 	return r;
1336 }
1337 
1338 static int kvmppc_handle_nested_exit(struct kvm_vcpu *vcpu)
1339 {
1340 	int r;
1341 	int srcu_idx;
1342 
1343 	vcpu->stat.sum_exits++;
1344 
1345 	/*
1346 	 * This can happen if an interrupt occurs in the last stages
1347 	 * of guest entry or the first stages of guest exit (i.e. after
1348 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1349 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1350 	 * That can happen due to a bug, or due to a machine check
1351 	 * occurring at just the wrong time.
1352 	 */
1353 	if (vcpu->arch.shregs.msr & MSR_HV) {
1354 		pr_emerg("KVM trap in HV mode while nested!\n");
1355 		pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1356 			 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1357 			 vcpu->arch.shregs.msr);
1358 		kvmppc_dump_regs(vcpu);
1359 		return RESUME_HOST;
1360 	}
1361 	switch (vcpu->arch.trap) {
1362 	/* We're good on these - the host merely wanted to get our attention */
1363 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1364 		vcpu->stat.dec_exits++;
1365 		r = RESUME_GUEST;
1366 		break;
1367 	case BOOK3S_INTERRUPT_EXTERNAL:
1368 		vcpu->stat.ext_intr_exits++;
1369 		r = RESUME_HOST;
1370 		break;
1371 	case BOOK3S_INTERRUPT_H_DOORBELL:
1372 	case BOOK3S_INTERRUPT_H_VIRT:
1373 		vcpu->stat.ext_intr_exits++;
1374 		r = RESUME_GUEST;
1375 		break;
1376 	/* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1377 	case BOOK3S_INTERRUPT_HMI:
1378 	case BOOK3S_INTERRUPT_PERFMON:
1379 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
1380 		r = RESUME_GUEST;
1381 		break;
1382 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
1383 		/* Pass the machine check to the L1 guest */
1384 		r = RESUME_HOST;
1385 		/* Print the MCE event to host console. */
1386 		machine_check_print_event_info(&vcpu->arch.mce_evt, false);
1387 		break;
1388 	/*
1389 	 * We get these next two if the guest accesses a page which it thinks
1390 	 * it has mapped but which is not actually present, either because
1391 	 * it is for an emulated I/O device or because the corresonding
1392 	 * host page has been paged out.
1393 	 */
1394 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1395 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1396 		r = kvmhv_nested_page_fault(vcpu);
1397 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1398 		break;
1399 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
1400 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1401 		vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1402 					 DSISR_SRR1_MATCH_64S;
1403 		if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1404 			vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1405 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1406 		r = kvmhv_nested_page_fault(vcpu);
1407 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1408 		break;
1409 
1410 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1411 	case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1412 		/*
1413 		 * This occurs for various TM-related instructions that
1414 		 * we need to emulate on POWER9 DD2.2.  We have already
1415 		 * handled the cases where the guest was in real-suspend
1416 		 * mode and was transitioning to transactional state.
1417 		 */
1418 		r = kvmhv_p9_tm_emulation(vcpu);
1419 		break;
1420 #endif
1421 
1422 	case BOOK3S_INTERRUPT_HV_RM_HARD:
1423 		vcpu->arch.trap = 0;
1424 		r = RESUME_GUEST;
1425 		if (!xive_enabled())
1426 			kvmppc_xics_rm_complete(vcpu, 0);
1427 		break;
1428 	default:
1429 		r = RESUME_HOST;
1430 		break;
1431 	}
1432 
1433 	return r;
1434 }
1435 
1436 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1437 					    struct kvm_sregs *sregs)
1438 {
1439 	int i;
1440 
1441 	memset(sregs, 0, sizeof(struct kvm_sregs));
1442 	sregs->pvr = vcpu->arch.pvr;
1443 	for (i = 0; i < vcpu->arch.slb_max; i++) {
1444 		sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1445 		sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1446 	}
1447 
1448 	return 0;
1449 }
1450 
1451 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1452 					    struct kvm_sregs *sregs)
1453 {
1454 	int i, j;
1455 
1456 	/* Only accept the same PVR as the host's, since we can't spoof it */
1457 	if (sregs->pvr != vcpu->arch.pvr)
1458 		return -EINVAL;
1459 
1460 	j = 0;
1461 	for (i = 0; i < vcpu->arch.slb_nr; i++) {
1462 		if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1463 			vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1464 			vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1465 			++j;
1466 		}
1467 	}
1468 	vcpu->arch.slb_max = j;
1469 
1470 	return 0;
1471 }
1472 
1473 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1474 		bool preserve_top32)
1475 {
1476 	struct kvm *kvm = vcpu->kvm;
1477 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
1478 	u64 mask;
1479 
1480 	mutex_lock(&kvm->lock);
1481 	spin_lock(&vc->lock);
1482 	/*
1483 	 * If ILE (interrupt little-endian) has changed, update the
1484 	 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1485 	 */
1486 	if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1487 		struct kvm_vcpu *vcpu;
1488 		int i;
1489 
1490 		kvm_for_each_vcpu(i, vcpu, kvm) {
1491 			if (vcpu->arch.vcore != vc)
1492 				continue;
1493 			if (new_lpcr & LPCR_ILE)
1494 				vcpu->arch.intr_msr |= MSR_LE;
1495 			else
1496 				vcpu->arch.intr_msr &= ~MSR_LE;
1497 		}
1498 	}
1499 
1500 	/*
1501 	 * Userspace can only modify DPFD (default prefetch depth),
1502 	 * ILE (interrupt little-endian) and TC (translation control).
1503 	 * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1504 	 */
1505 	mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1506 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
1507 		mask |= LPCR_AIL;
1508 	/*
1509 	 * On POWER9, allow userspace to enable large decrementer for the
1510 	 * guest, whether or not the host has it enabled.
1511 	 */
1512 	if (cpu_has_feature(CPU_FTR_ARCH_300))
1513 		mask |= LPCR_LD;
1514 
1515 	/* Broken 32-bit version of LPCR must not clear top bits */
1516 	if (preserve_top32)
1517 		mask &= 0xFFFFFFFF;
1518 	vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1519 	spin_unlock(&vc->lock);
1520 	mutex_unlock(&kvm->lock);
1521 }
1522 
1523 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1524 				 union kvmppc_one_reg *val)
1525 {
1526 	int r = 0;
1527 	long int i;
1528 
1529 	switch (id) {
1530 	case KVM_REG_PPC_DEBUG_INST:
1531 		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1532 		break;
1533 	case KVM_REG_PPC_HIOR:
1534 		*val = get_reg_val(id, 0);
1535 		break;
1536 	case KVM_REG_PPC_DABR:
1537 		*val = get_reg_val(id, vcpu->arch.dabr);
1538 		break;
1539 	case KVM_REG_PPC_DABRX:
1540 		*val = get_reg_val(id, vcpu->arch.dabrx);
1541 		break;
1542 	case KVM_REG_PPC_DSCR:
1543 		*val = get_reg_val(id, vcpu->arch.dscr);
1544 		break;
1545 	case KVM_REG_PPC_PURR:
1546 		*val = get_reg_val(id, vcpu->arch.purr);
1547 		break;
1548 	case KVM_REG_PPC_SPURR:
1549 		*val = get_reg_val(id, vcpu->arch.spurr);
1550 		break;
1551 	case KVM_REG_PPC_AMR:
1552 		*val = get_reg_val(id, vcpu->arch.amr);
1553 		break;
1554 	case KVM_REG_PPC_UAMOR:
1555 		*val = get_reg_val(id, vcpu->arch.uamor);
1556 		break;
1557 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1558 		i = id - KVM_REG_PPC_MMCR0;
1559 		*val = get_reg_val(id, vcpu->arch.mmcr[i]);
1560 		break;
1561 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1562 		i = id - KVM_REG_PPC_PMC1;
1563 		*val = get_reg_val(id, vcpu->arch.pmc[i]);
1564 		break;
1565 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1566 		i = id - KVM_REG_PPC_SPMC1;
1567 		*val = get_reg_val(id, vcpu->arch.spmc[i]);
1568 		break;
1569 	case KVM_REG_PPC_SIAR:
1570 		*val = get_reg_val(id, vcpu->arch.siar);
1571 		break;
1572 	case KVM_REG_PPC_SDAR:
1573 		*val = get_reg_val(id, vcpu->arch.sdar);
1574 		break;
1575 	case KVM_REG_PPC_SIER:
1576 		*val = get_reg_val(id, vcpu->arch.sier);
1577 		break;
1578 	case KVM_REG_PPC_IAMR:
1579 		*val = get_reg_val(id, vcpu->arch.iamr);
1580 		break;
1581 	case KVM_REG_PPC_PSPB:
1582 		*val = get_reg_val(id, vcpu->arch.pspb);
1583 		break;
1584 	case KVM_REG_PPC_DPDES:
1585 		*val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1586 		break;
1587 	case KVM_REG_PPC_VTB:
1588 		*val = get_reg_val(id, vcpu->arch.vcore->vtb);
1589 		break;
1590 	case KVM_REG_PPC_DAWR:
1591 		*val = get_reg_val(id, vcpu->arch.dawr);
1592 		break;
1593 	case KVM_REG_PPC_DAWRX:
1594 		*val = get_reg_val(id, vcpu->arch.dawrx);
1595 		break;
1596 	case KVM_REG_PPC_CIABR:
1597 		*val = get_reg_val(id, vcpu->arch.ciabr);
1598 		break;
1599 	case KVM_REG_PPC_CSIGR:
1600 		*val = get_reg_val(id, vcpu->arch.csigr);
1601 		break;
1602 	case KVM_REG_PPC_TACR:
1603 		*val = get_reg_val(id, vcpu->arch.tacr);
1604 		break;
1605 	case KVM_REG_PPC_TCSCR:
1606 		*val = get_reg_val(id, vcpu->arch.tcscr);
1607 		break;
1608 	case KVM_REG_PPC_PID:
1609 		*val = get_reg_val(id, vcpu->arch.pid);
1610 		break;
1611 	case KVM_REG_PPC_ACOP:
1612 		*val = get_reg_val(id, vcpu->arch.acop);
1613 		break;
1614 	case KVM_REG_PPC_WORT:
1615 		*val = get_reg_val(id, vcpu->arch.wort);
1616 		break;
1617 	case KVM_REG_PPC_TIDR:
1618 		*val = get_reg_val(id, vcpu->arch.tid);
1619 		break;
1620 	case KVM_REG_PPC_PSSCR:
1621 		*val = get_reg_val(id, vcpu->arch.psscr);
1622 		break;
1623 	case KVM_REG_PPC_VPA_ADDR:
1624 		spin_lock(&vcpu->arch.vpa_update_lock);
1625 		*val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1626 		spin_unlock(&vcpu->arch.vpa_update_lock);
1627 		break;
1628 	case KVM_REG_PPC_VPA_SLB:
1629 		spin_lock(&vcpu->arch.vpa_update_lock);
1630 		val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1631 		val->vpaval.length = vcpu->arch.slb_shadow.len;
1632 		spin_unlock(&vcpu->arch.vpa_update_lock);
1633 		break;
1634 	case KVM_REG_PPC_VPA_DTL:
1635 		spin_lock(&vcpu->arch.vpa_update_lock);
1636 		val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1637 		val->vpaval.length = vcpu->arch.dtl.len;
1638 		spin_unlock(&vcpu->arch.vpa_update_lock);
1639 		break;
1640 	case KVM_REG_PPC_TB_OFFSET:
1641 		*val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1642 		break;
1643 	case KVM_REG_PPC_LPCR:
1644 	case KVM_REG_PPC_LPCR_64:
1645 		*val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1646 		break;
1647 	case KVM_REG_PPC_PPR:
1648 		*val = get_reg_val(id, vcpu->arch.ppr);
1649 		break;
1650 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1651 	case KVM_REG_PPC_TFHAR:
1652 		*val = get_reg_val(id, vcpu->arch.tfhar);
1653 		break;
1654 	case KVM_REG_PPC_TFIAR:
1655 		*val = get_reg_val(id, vcpu->arch.tfiar);
1656 		break;
1657 	case KVM_REG_PPC_TEXASR:
1658 		*val = get_reg_val(id, vcpu->arch.texasr);
1659 		break;
1660 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1661 		i = id - KVM_REG_PPC_TM_GPR0;
1662 		*val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1663 		break;
1664 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1665 	{
1666 		int j;
1667 		i = id - KVM_REG_PPC_TM_VSR0;
1668 		if (i < 32)
1669 			for (j = 0; j < TS_FPRWIDTH; j++)
1670 				val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1671 		else {
1672 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1673 				val->vval = vcpu->arch.vr_tm.vr[i-32];
1674 			else
1675 				r = -ENXIO;
1676 		}
1677 		break;
1678 	}
1679 	case KVM_REG_PPC_TM_CR:
1680 		*val = get_reg_val(id, vcpu->arch.cr_tm);
1681 		break;
1682 	case KVM_REG_PPC_TM_XER:
1683 		*val = get_reg_val(id, vcpu->arch.xer_tm);
1684 		break;
1685 	case KVM_REG_PPC_TM_LR:
1686 		*val = get_reg_val(id, vcpu->arch.lr_tm);
1687 		break;
1688 	case KVM_REG_PPC_TM_CTR:
1689 		*val = get_reg_val(id, vcpu->arch.ctr_tm);
1690 		break;
1691 	case KVM_REG_PPC_TM_FPSCR:
1692 		*val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1693 		break;
1694 	case KVM_REG_PPC_TM_AMR:
1695 		*val = get_reg_val(id, vcpu->arch.amr_tm);
1696 		break;
1697 	case KVM_REG_PPC_TM_PPR:
1698 		*val = get_reg_val(id, vcpu->arch.ppr_tm);
1699 		break;
1700 	case KVM_REG_PPC_TM_VRSAVE:
1701 		*val = get_reg_val(id, vcpu->arch.vrsave_tm);
1702 		break;
1703 	case KVM_REG_PPC_TM_VSCR:
1704 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1705 			*val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1706 		else
1707 			r = -ENXIO;
1708 		break;
1709 	case KVM_REG_PPC_TM_DSCR:
1710 		*val = get_reg_val(id, vcpu->arch.dscr_tm);
1711 		break;
1712 	case KVM_REG_PPC_TM_TAR:
1713 		*val = get_reg_val(id, vcpu->arch.tar_tm);
1714 		break;
1715 #endif
1716 	case KVM_REG_PPC_ARCH_COMPAT:
1717 		*val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1718 		break;
1719 	case KVM_REG_PPC_DEC_EXPIRY:
1720 		*val = get_reg_val(id, vcpu->arch.dec_expires +
1721 				   vcpu->arch.vcore->tb_offset);
1722 		break;
1723 	case KVM_REG_PPC_ONLINE:
1724 		*val = get_reg_val(id, vcpu->arch.online);
1725 		break;
1726 	case KVM_REG_PPC_PTCR:
1727 		*val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
1728 		break;
1729 	default:
1730 		r = -EINVAL;
1731 		break;
1732 	}
1733 
1734 	return r;
1735 }
1736 
1737 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1738 				 union kvmppc_one_reg *val)
1739 {
1740 	int r = 0;
1741 	long int i;
1742 	unsigned long addr, len;
1743 
1744 	switch (id) {
1745 	case KVM_REG_PPC_HIOR:
1746 		/* Only allow this to be set to zero */
1747 		if (set_reg_val(id, *val))
1748 			r = -EINVAL;
1749 		break;
1750 	case KVM_REG_PPC_DABR:
1751 		vcpu->arch.dabr = set_reg_val(id, *val);
1752 		break;
1753 	case KVM_REG_PPC_DABRX:
1754 		vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1755 		break;
1756 	case KVM_REG_PPC_DSCR:
1757 		vcpu->arch.dscr = set_reg_val(id, *val);
1758 		break;
1759 	case KVM_REG_PPC_PURR:
1760 		vcpu->arch.purr = set_reg_val(id, *val);
1761 		break;
1762 	case KVM_REG_PPC_SPURR:
1763 		vcpu->arch.spurr = set_reg_val(id, *val);
1764 		break;
1765 	case KVM_REG_PPC_AMR:
1766 		vcpu->arch.amr = set_reg_val(id, *val);
1767 		break;
1768 	case KVM_REG_PPC_UAMOR:
1769 		vcpu->arch.uamor = set_reg_val(id, *val);
1770 		break;
1771 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1772 		i = id - KVM_REG_PPC_MMCR0;
1773 		vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1774 		break;
1775 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1776 		i = id - KVM_REG_PPC_PMC1;
1777 		vcpu->arch.pmc[i] = set_reg_val(id, *val);
1778 		break;
1779 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1780 		i = id - KVM_REG_PPC_SPMC1;
1781 		vcpu->arch.spmc[i] = set_reg_val(id, *val);
1782 		break;
1783 	case KVM_REG_PPC_SIAR:
1784 		vcpu->arch.siar = set_reg_val(id, *val);
1785 		break;
1786 	case KVM_REG_PPC_SDAR:
1787 		vcpu->arch.sdar = set_reg_val(id, *val);
1788 		break;
1789 	case KVM_REG_PPC_SIER:
1790 		vcpu->arch.sier = set_reg_val(id, *val);
1791 		break;
1792 	case KVM_REG_PPC_IAMR:
1793 		vcpu->arch.iamr = set_reg_val(id, *val);
1794 		break;
1795 	case KVM_REG_PPC_PSPB:
1796 		vcpu->arch.pspb = set_reg_val(id, *val);
1797 		break;
1798 	case KVM_REG_PPC_DPDES:
1799 		vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1800 		break;
1801 	case KVM_REG_PPC_VTB:
1802 		vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1803 		break;
1804 	case KVM_REG_PPC_DAWR:
1805 		vcpu->arch.dawr = set_reg_val(id, *val);
1806 		break;
1807 	case KVM_REG_PPC_DAWRX:
1808 		vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1809 		break;
1810 	case KVM_REG_PPC_CIABR:
1811 		vcpu->arch.ciabr = set_reg_val(id, *val);
1812 		/* Don't allow setting breakpoints in hypervisor code */
1813 		if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1814 			vcpu->arch.ciabr &= ~CIABR_PRIV;	/* disable */
1815 		break;
1816 	case KVM_REG_PPC_CSIGR:
1817 		vcpu->arch.csigr = set_reg_val(id, *val);
1818 		break;
1819 	case KVM_REG_PPC_TACR:
1820 		vcpu->arch.tacr = set_reg_val(id, *val);
1821 		break;
1822 	case KVM_REG_PPC_TCSCR:
1823 		vcpu->arch.tcscr = set_reg_val(id, *val);
1824 		break;
1825 	case KVM_REG_PPC_PID:
1826 		vcpu->arch.pid = set_reg_val(id, *val);
1827 		break;
1828 	case KVM_REG_PPC_ACOP:
1829 		vcpu->arch.acop = set_reg_val(id, *val);
1830 		break;
1831 	case KVM_REG_PPC_WORT:
1832 		vcpu->arch.wort = set_reg_val(id, *val);
1833 		break;
1834 	case KVM_REG_PPC_TIDR:
1835 		vcpu->arch.tid = set_reg_val(id, *val);
1836 		break;
1837 	case KVM_REG_PPC_PSSCR:
1838 		vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1839 		break;
1840 	case KVM_REG_PPC_VPA_ADDR:
1841 		addr = set_reg_val(id, *val);
1842 		r = -EINVAL;
1843 		if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1844 			      vcpu->arch.dtl.next_gpa))
1845 			break;
1846 		r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1847 		break;
1848 	case KVM_REG_PPC_VPA_SLB:
1849 		addr = val->vpaval.addr;
1850 		len = val->vpaval.length;
1851 		r = -EINVAL;
1852 		if (addr && !vcpu->arch.vpa.next_gpa)
1853 			break;
1854 		r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1855 		break;
1856 	case KVM_REG_PPC_VPA_DTL:
1857 		addr = val->vpaval.addr;
1858 		len = val->vpaval.length;
1859 		r = -EINVAL;
1860 		if (addr && (len < sizeof(struct dtl_entry) ||
1861 			     !vcpu->arch.vpa.next_gpa))
1862 			break;
1863 		len -= len % sizeof(struct dtl_entry);
1864 		r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1865 		break;
1866 	case KVM_REG_PPC_TB_OFFSET:
1867 		/* round up to multiple of 2^24 */
1868 		vcpu->arch.vcore->tb_offset =
1869 			ALIGN(set_reg_val(id, *val), 1UL << 24);
1870 		break;
1871 	case KVM_REG_PPC_LPCR:
1872 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1873 		break;
1874 	case KVM_REG_PPC_LPCR_64:
1875 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1876 		break;
1877 	case KVM_REG_PPC_PPR:
1878 		vcpu->arch.ppr = set_reg_val(id, *val);
1879 		break;
1880 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1881 	case KVM_REG_PPC_TFHAR:
1882 		vcpu->arch.tfhar = set_reg_val(id, *val);
1883 		break;
1884 	case KVM_REG_PPC_TFIAR:
1885 		vcpu->arch.tfiar = set_reg_val(id, *val);
1886 		break;
1887 	case KVM_REG_PPC_TEXASR:
1888 		vcpu->arch.texasr = set_reg_val(id, *val);
1889 		break;
1890 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1891 		i = id - KVM_REG_PPC_TM_GPR0;
1892 		vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1893 		break;
1894 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1895 	{
1896 		int j;
1897 		i = id - KVM_REG_PPC_TM_VSR0;
1898 		if (i < 32)
1899 			for (j = 0; j < TS_FPRWIDTH; j++)
1900 				vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1901 		else
1902 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1903 				vcpu->arch.vr_tm.vr[i-32] = val->vval;
1904 			else
1905 				r = -ENXIO;
1906 		break;
1907 	}
1908 	case KVM_REG_PPC_TM_CR:
1909 		vcpu->arch.cr_tm = set_reg_val(id, *val);
1910 		break;
1911 	case KVM_REG_PPC_TM_XER:
1912 		vcpu->arch.xer_tm = set_reg_val(id, *val);
1913 		break;
1914 	case KVM_REG_PPC_TM_LR:
1915 		vcpu->arch.lr_tm = set_reg_val(id, *val);
1916 		break;
1917 	case KVM_REG_PPC_TM_CTR:
1918 		vcpu->arch.ctr_tm = set_reg_val(id, *val);
1919 		break;
1920 	case KVM_REG_PPC_TM_FPSCR:
1921 		vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1922 		break;
1923 	case KVM_REG_PPC_TM_AMR:
1924 		vcpu->arch.amr_tm = set_reg_val(id, *val);
1925 		break;
1926 	case KVM_REG_PPC_TM_PPR:
1927 		vcpu->arch.ppr_tm = set_reg_val(id, *val);
1928 		break;
1929 	case KVM_REG_PPC_TM_VRSAVE:
1930 		vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1931 		break;
1932 	case KVM_REG_PPC_TM_VSCR:
1933 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1934 			vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1935 		else
1936 			r = - ENXIO;
1937 		break;
1938 	case KVM_REG_PPC_TM_DSCR:
1939 		vcpu->arch.dscr_tm = set_reg_val(id, *val);
1940 		break;
1941 	case KVM_REG_PPC_TM_TAR:
1942 		vcpu->arch.tar_tm = set_reg_val(id, *val);
1943 		break;
1944 #endif
1945 	case KVM_REG_PPC_ARCH_COMPAT:
1946 		r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1947 		break;
1948 	case KVM_REG_PPC_DEC_EXPIRY:
1949 		vcpu->arch.dec_expires = set_reg_val(id, *val) -
1950 			vcpu->arch.vcore->tb_offset;
1951 		break;
1952 	case KVM_REG_PPC_ONLINE:
1953 		i = set_reg_val(id, *val);
1954 		if (i && !vcpu->arch.online)
1955 			atomic_inc(&vcpu->arch.vcore->online_count);
1956 		else if (!i && vcpu->arch.online)
1957 			atomic_dec(&vcpu->arch.vcore->online_count);
1958 		vcpu->arch.online = i;
1959 		break;
1960 	case KVM_REG_PPC_PTCR:
1961 		vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
1962 		break;
1963 	default:
1964 		r = -EINVAL;
1965 		break;
1966 	}
1967 
1968 	return r;
1969 }
1970 
1971 /*
1972  * On POWER9, threads are independent and can be in different partitions.
1973  * Therefore we consider each thread to be a subcore.
1974  * There is a restriction that all threads have to be in the same
1975  * MMU mode (radix or HPT), unfortunately, but since we only support
1976  * HPT guests on a HPT host so far, that isn't an impediment yet.
1977  */
1978 static int threads_per_vcore(struct kvm *kvm)
1979 {
1980 	if (kvm->arch.threads_indep)
1981 		return 1;
1982 	return threads_per_subcore;
1983 }
1984 
1985 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
1986 {
1987 	struct kvmppc_vcore *vcore;
1988 
1989 	vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1990 
1991 	if (vcore == NULL)
1992 		return NULL;
1993 
1994 	spin_lock_init(&vcore->lock);
1995 	spin_lock_init(&vcore->stoltb_lock);
1996 	init_swait_queue_head(&vcore->wq);
1997 	vcore->preempt_tb = TB_NIL;
1998 	vcore->lpcr = kvm->arch.lpcr;
1999 	vcore->first_vcpuid = id;
2000 	vcore->kvm = kvm;
2001 	INIT_LIST_HEAD(&vcore->preempt_list);
2002 
2003 	return vcore;
2004 }
2005 
2006 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2007 static struct debugfs_timings_element {
2008 	const char *name;
2009 	size_t offset;
2010 } timings[] = {
2011 	{"rm_entry",	offsetof(struct kvm_vcpu, arch.rm_entry)},
2012 	{"rm_intr",	offsetof(struct kvm_vcpu, arch.rm_intr)},
2013 	{"rm_exit",	offsetof(struct kvm_vcpu, arch.rm_exit)},
2014 	{"guest",	offsetof(struct kvm_vcpu, arch.guest_time)},
2015 	{"cede",	offsetof(struct kvm_vcpu, arch.cede_time)},
2016 };
2017 
2018 #define N_TIMINGS	(ARRAY_SIZE(timings))
2019 
2020 struct debugfs_timings_state {
2021 	struct kvm_vcpu	*vcpu;
2022 	unsigned int	buflen;
2023 	char		buf[N_TIMINGS * 100];
2024 };
2025 
2026 static int debugfs_timings_open(struct inode *inode, struct file *file)
2027 {
2028 	struct kvm_vcpu *vcpu = inode->i_private;
2029 	struct debugfs_timings_state *p;
2030 
2031 	p = kzalloc(sizeof(*p), GFP_KERNEL);
2032 	if (!p)
2033 		return -ENOMEM;
2034 
2035 	kvm_get_kvm(vcpu->kvm);
2036 	p->vcpu = vcpu;
2037 	file->private_data = p;
2038 
2039 	return nonseekable_open(inode, file);
2040 }
2041 
2042 static int debugfs_timings_release(struct inode *inode, struct file *file)
2043 {
2044 	struct debugfs_timings_state *p = file->private_data;
2045 
2046 	kvm_put_kvm(p->vcpu->kvm);
2047 	kfree(p);
2048 	return 0;
2049 }
2050 
2051 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2052 				    size_t len, loff_t *ppos)
2053 {
2054 	struct debugfs_timings_state *p = file->private_data;
2055 	struct kvm_vcpu *vcpu = p->vcpu;
2056 	char *s, *buf_end;
2057 	struct kvmhv_tb_accumulator tb;
2058 	u64 count;
2059 	loff_t pos;
2060 	ssize_t n;
2061 	int i, loops;
2062 	bool ok;
2063 
2064 	if (!p->buflen) {
2065 		s = p->buf;
2066 		buf_end = s + sizeof(p->buf);
2067 		for (i = 0; i < N_TIMINGS; ++i) {
2068 			struct kvmhv_tb_accumulator *acc;
2069 
2070 			acc = (struct kvmhv_tb_accumulator *)
2071 				((unsigned long)vcpu + timings[i].offset);
2072 			ok = false;
2073 			for (loops = 0; loops < 1000; ++loops) {
2074 				count = acc->seqcount;
2075 				if (!(count & 1)) {
2076 					smp_rmb();
2077 					tb = *acc;
2078 					smp_rmb();
2079 					if (count == acc->seqcount) {
2080 						ok = true;
2081 						break;
2082 					}
2083 				}
2084 				udelay(1);
2085 			}
2086 			if (!ok)
2087 				snprintf(s, buf_end - s, "%s: stuck\n",
2088 					timings[i].name);
2089 			else
2090 				snprintf(s, buf_end - s,
2091 					"%s: %llu %llu %llu %llu\n",
2092 					timings[i].name, count / 2,
2093 					tb_to_ns(tb.tb_total),
2094 					tb_to_ns(tb.tb_min),
2095 					tb_to_ns(tb.tb_max));
2096 			s += strlen(s);
2097 		}
2098 		p->buflen = s - p->buf;
2099 	}
2100 
2101 	pos = *ppos;
2102 	if (pos >= p->buflen)
2103 		return 0;
2104 	if (len > p->buflen - pos)
2105 		len = p->buflen - pos;
2106 	n = copy_to_user(buf, p->buf + pos, len);
2107 	if (n) {
2108 		if (n == len)
2109 			return -EFAULT;
2110 		len -= n;
2111 	}
2112 	*ppos = pos + len;
2113 	return len;
2114 }
2115 
2116 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2117 				     size_t len, loff_t *ppos)
2118 {
2119 	return -EACCES;
2120 }
2121 
2122 static const struct file_operations debugfs_timings_ops = {
2123 	.owner	 = THIS_MODULE,
2124 	.open	 = debugfs_timings_open,
2125 	.release = debugfs_timings_release,
2126 	.read	 = debugfs_timings_read,
2127 	.write	 = debugfs_timings_write,
2128 	.llseek	 = generic_file_llseek,
2129 };
2130 
2131 /* Create a debugfs directory for the vcpu */
2132 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2133 {
2134 	char buf[16];
2135 	struct kvm *kvm = vcpu->kvm;
2136 
2137 	snprintf(buf, sizeof(buf), "vcpu%u", id);
2138 	if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
2139 		return;
2140 	vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
2141 	if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
2142 		return;
2143 	vcpu->arch.debugfs_timings =
2144 		debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
2145 				    vcpu, &debugfs_timings_ops);
2146 }
2147 
2148 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2149 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2150 {
2151 }
2152 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2153 
2154 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
2155 						   unsigned int id)
2156 {
2157 	struct kvm_vcpu *vcpu;
2158 	int err;
2159 	int core;
2160 	struct kvmppc_vcore *vcore;
2161 
2162 	err = -ENOMEM;
2163 	vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2164 	if (!vcpu)
2165 		goto out;
2166 
2167 	err = kvm_vcpu_init(vcpu, kvm, id);
2168 	if (err)
2169 		goto free_vcpu;
2170 
2171 	vcpu->arch.shared = &vcpu->arch.shregs;
2172 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2173 	/*
2174 	 * The shared struct is never shared on HV,
2175 	 * so we can always use host endianness
2176 	 */
2177 #ifdef __BIG_ENDIAN__
2178 	vcpu->arch.shared_big_endian = true;
2179 #else
2180 	vcpu->arch.shared_big_endian = false;
2181 #endif
2182 #endif
2183 	vcpu->arch.mmcr[0] = MMCR0_FC;
2184 	vcpu->arch.ctrl = CTRL_RUNLATCH;
2185 	/* default to host PVR, since we can't spoof it */
2186 	kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2187 	spin_lock_init(&vcpu->arch.vpa_update_lock);
2188 	spin_lock_init(&vcpu->arch.tbacct_lock);
2189 	vcpu->arch.busy_preempt = TB_NIL;
2190 	vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2191 
2192 	/*
2193 	 * Set the default HFSCR for the guest from the host value.
2194 	 * This value is only used on POWER9.
2195 	 * On POWER9, we want to virtualize the doorbell facility, so we
2196 	 * don't set the HFSCR_MSGP bit, and that causes those instructions
2197 	 * to trap and then we emulate them.
2198 	 */
2199 	vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2200 		HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP;
2201 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
2202 		vcpu->arch.hfscr &= mfspr(SPRN_HFSCR);
2203 		if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2204 			vcpu->arch.hfscr |= HFSCR_TM;
2205 	}
2206 	if (cpu_has_feature(CPU_FTR_TM_COMP))
2207 		vcpu->arch.hfscr |= HFSCR_TM;
2208 
2209 	kvmppc_mmu_book3s_hv_init(vcpu);
2210 
2211 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2212 
2213 	init_waitqueue_head(&vcpu->arch.cpu_run);
2214 
2215 	mutex_lock(&kvm->lock);
2216 	vcore = NULL;
2217 	err = -EINVAL;
2218 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2219 		if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2220 			pr_devel("KVM: VCPU ID too high\n");
2221 			core = KVM_MAX_VCORES;
2222 		} else {
2223 			BUG_ON(kvm->arch.smt_mode != 1);
2224 			core = kvmppc_pack_vcpu_id(kvm, id);
2225 		}
2226 	} else {
2227 		core = id / kvm->arch.smt_mode;
2228 	}
2229 	if (core < KVM_MAX_VCORES) {
2230 		vcore = kvm->arch.vcores[core];
2231 		if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2232 			pr_devel("KVM: collision on id %u", id);
2233 			vcore = NULL;
2234 		} else if (!vcore) {
2235 			err = -ENOMEM;
2236 			vcore = kvmppc_vcore_create(kvm,
2237 					id & ~(kvm->arch.smt_mode - 1));
2238 			kvm->arch.vcores[core] = vcore;
2239 			kvm->arch.online_vcores++;
2240 		}
2241 	}
2242 	mutex_unlock(&kvm->lock);
2243 
2244 	if (!vcore)
2245 		goto free_vcpu;
2246 
2247 	spin_lock(&vcore->lock);
2248 	++vcore->num_threads;
2249 	spin_unlock(&vcore->lock);
2250 	vcpu->arch.vcore = vcore;
2251 	vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2252 	vcpu->arch.thread_cpu = -1;
2253 	vcpu->arch.prev_cpu = -1;
2254 
2255 	vcpu->arch.cpu_type = KVM_CPU_3S_64;
2256 	kvmppc_sanity_check(vcpu);
2257 
2258 	debugfs_vcpu_init(vcpu, id);
2259 
2260 	return vcpu;
2261 
2262 free_vcpu:
2263 	kmem_cache_free(kvm_vcpu_cache, vcpu);
2264 out:
2265 	return ERR_PTR(err);
2266 }
2267 
2268 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2269 			      unsigned long flags)
2270 {
2271 	int err;
2272 	int esmt = 0;
2273 
2274 	if (flags)
2275 		return -EINVAL;
2276 	if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2277 		return -EINVAL;
2278 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2279 		/*
2280 		 * On POWER8 (or POWER7), the threading mode is "strict",
2281 		 * so we pack smt_mode vcpus per vcore.
2282 		 */
2283 		if (smt_mode > threads_per_subcore)
2284 			return -EINVAL;
2285 	} else {
2286 		/*
2287 		 * On POWER9, the threading mode is "loose",
2288 		 * so each vcpu gets its own vcore.
2289 		 */
2290 		esmt = smt_mode;
2291 		smt_mode = 1;
2292 	}
2293 	mutex_lock(&kvm->lock);
2294 	err = -EBUSY;
2295 	if (!kvm->arch.online_vcores) {
2296 		kvm->arch.smt_mode = smt_mode;
2297 		kvm->arch.emul_smt_mode = esmt;
2298 		err = 0;
2299 	}
2300 	mutex_unlock(&kvm->lock);
2301 
2302 	return err;
2303 }
2304 
2305 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2306 {
2307 	if (vpa->pinned_addr)
2308 		kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2309 					vpa->dirty);
2310 }
2311 
2312 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2313 {
2314 	spin_lock(&vcpu->arch.vpa_update_lock);
2315 	unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2316 	unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2317 	unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2318 	spin_unlock(&vcpu->arch.vpa_update_lock);
2319 	kvm_vcpu_uninit(vcpu);
2320 	kmem_cache_free(kvm_vcpu_cache, vcpu);
2321 }
2322 
2323 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2324 {
2325 	/* Indicate we want to get back into the guest */
2326 	return 1;
2327 }
2328 
2329 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2330 {
2331 	unsigned long dec_nsec, now;
2332 
2333 	now = get_tb();
2334 	if (now > vcpu->arch.dec_expires) {
2335 		/* decrementer has already gone negative */
2336 		kvmppc_core_queue_dec(vcpu);
2337 		kvmppc_core_prepare_to_enter(vcpu);
2338 		return;
2339 	}
2340 	dec_nsec = tb_to_ns(vcpu->arch.dec_expires - now);
2341 	hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2342 	vcpu->arch.timer_running = 1;
2343 }
2344 
2345 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
2346 {
2347 	vcpu->arch.ceded = 0;
2348 	if (vcpu->arch.timer_running) {
2349 		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2350 		vcpu->arch.timer_running = 0;
2351 	}
2352 }
2353 
2354 extern int __kvmppc_vcore_entry(void);
2355 
2356 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2357 				   struct kvm_vcpu *vcpu)
2358 {
2359 	u64 now;
2360 
2361 	if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2362 		return;
2363 	spin_lock_irq(&vcpu->arch.tbacct_lock);
2364 	now = mftb();
2365 	vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2366 		vcpu->arch.stolen_logged;
2367 	vcpu->arch.busy_preempt = now;
2368 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2369 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
2370 	--vc->n_runnable;
2371 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2372 }
2373 
2374 static int kvmppc_grab_hwthread(int cpu)
2375 {
2376 	struct paca_struct *tpaca;
2377 	long timeout = 10000;
2378 
2379 	tpaca = paca_ptrs[cpu];
2380 
2381 	/* Ensure the thread won't go into the kernel if it wakes */
2382 	tpaca->kvm_hstate.kvm_vcpu = NULL;
2383 	tpaca->kvm_hstate.kvm_vcore = NULL;
2384 	tpaca->kvm_hstate.napping = 0;
2385 	smp_wmb();
2386 	tpaca->kvm_hstate.hwthread_req = 1;
2387 
2388 	/*
2389 	 * If the thread is already executing in the kernel (e.g. handling
2390 	 * a stray interrupt), wait for it to get back to nap mode.
2391 	 * The smp_mb() is to ensure that our setting of hwthread_req
2392 	 * is visible before we look at hwthread_state, so if this
2393 	 * races with the code at system_reset_pSeries and the thread
2394 	 * misses our setting of hwthread_req, we are sure to see its
2395 	 * setting of hwthread_state, and vice versa.
2396 	 */
2397 	smp_mb();
2398 	while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2399 		if (--timeout <= 0) {
2400 			pr_err("KVM: couldn't grab cpu %d\n", cpu);
2401 			return -EBUSY;
2402 		}
2403 		udelay(1);
2404 	}
2405 	return 0;
2406 }
2407 
2408 static void kvmppc_release_hwthread(int cpu)
2409 {
2410 	struct paca_struct *tpaca;
2411 
2412 	tpaca = paca_ptrs[cpu];
2413 	tpaca->kvm_hstate.hwthread_req = 0;
2414 	tpaca->kvm_hstate.kvm_vcpu = NULL;
2415 	tpaca->kvm_hstate.kvm_vcore = NULL;
2416 	tpaca->kvm_hstate.kvm_split_mode = NULL;
2417 }
2418 
2419 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2420 {
2421 	struct kvm_nested_guest *nested = vcpu->arch.nested;
2422 	cpumask_t *cpu_in_guest;
2423 	int i;
2424 
2425 	cpu = cpu_first_thread_sibling(cpu);
2426 	if (nested) {
2427 		cpumask_set_cpu(cpu, &nested->need_tlb_flush);
2428 		cpu_in_guest = &nested->cpu_in_guest;
2429 	} else {
2430 		cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2431 		cpu_in_guest = &kvm->arch.cpu_in_guest;
2432 	}
2433 	/*
2434 	 * Make sure setting of bit in need_tlb_flush precedes
2435 	 * testing of cpu_in_guest bits.  The matching barrier on
2436 	 * the other side is the first smp_mb() in kvmppc_run_core().
2437 	 */
2438 	smp_mb();
2439 	for (i = 0; i < threads_per_core; ++i)
2440 		if (cpumask_test_cpu(cpu + i, cpu_in_guest))
2441 			smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2442 }
2443 
2444 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2445 {
2446 	struct kvm_nested_guest *nested = vcpu->arch.nested;
2447 	struct kvm *kvm = vcpu->kvm;
2448 	int prev_cpu;
2449 
2450 	if (!cpu_has_feature(CPU_FTR_HVMODE))
2451 		return;
2452 
2453 	if (nested)
2454 		prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
2455 	else
2456 		prev_cpu = vcpu->arch.prev_cpu;
2457 
2458 	/*
2459 	 * With radix, the guest can do TLB invalidations itself,
2460 	 * and it could choose to use the local form (tlbiel) if
2461 	 * it is invalidating a translation that has only ever been
2462 	 * used on one vcpu.  However, that doesn't mean it has
2463 	 * only ever been used on one physical cpu, since vcpus
2464 	 * can move around between pcpus.  To cope with this, when
2465 	 * a vcpu moves from one pcpu to another, we need to tell
2466 	 * any vcpus running on the same core as this vcpu previously
2467 	 * ran to flush the TLB.  The TLB is shared between threads,
2468 	 * so we use a single bit in .need_tlb_flush for all 4 threads.
2469 	 */
2470 	if (prev_cpu != pcpu) {
2471 		if (prev_cpu >= 0 &&
2472 		    cpu_first_thread_sibling(prev_cpu) !=
2473 		    cpu_first_thread_sibling(pcpu))
2474 			radix_flush_cpu(kvm, prev_cpu, vcpu);
2475 		if (nested)
2476 			nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
2477 		else
2478 			vcpu->arch.prev_cpu = pcpu;
2479 	}
2480 }
2481 
2482 static void kvmppc_radix_check_need_tlb_flush(struct kvm *kvm, int pcpu,
2483 					      struct kvm_nested_guest *nested)
2484 {
2485 	cpumask_t *need_tlb_flush;
2486 	int lpid;
2487 
2488 	if (!cpu_has_feature(CPU_FTR_HVMODE))
2489 		return;
2490 
2491 	if (cpu_has_feature(CPU_FTR_ARCH_300))
2492 		pcpu &= ~0x3UL;
2493 
2494 	if (nested) {
2495 		lpid = nested->shadow_lpid;
2496 		need_tlb_flush = &nested->need_tlb_flush;
2497 	} else {
2498 		lpid = kvm->arch.lpid;
2499 		need_tlb_flush = &kvm->arch.need_tlb_flush;
2500 	}
2501 
2502 	mtspr(SPRN_LPID, lpid);
2503 	isync();
2504 	smp_mb();
2505 
2506 	if (cpumask_test_cpu(pcpu, need_tlb_flush)) {
2507 		radix__local_flush_tlb_lpid_guest(lpid);
2508 		/* Clear the bit after the TLB flush */
2509 		cpumask_clear_cpu(pcpu, need_tlb_flush);
2510 	}
2511 }
2512 
2513 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2514 {
2515 	int cpu;
2516 	struct paca_struct *tpaca;
2517 	struct kvm *kvm = vc->kvm;
2518 
2519 	cpu = vc->pcpu;
2520 	if (vcpu) {
2521 		if (vcpu->arch.timer_running) {
2522 			hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2523 			vcpu->arch.timer_running = 0;
2524 		}
2525 		cpu += vcpu->arch.ptid;
2526 		vcpu->cpu = vc->pcpu;
2527 		vcpu->arch.thread_cpu = cpu;
2528 		cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2529 	}
2530 	tpaca = paca_ptrs[cpu];
2531 	tpaca->kvm_hstate.kvm_vcpu = vcpu;
2532 	tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2533 	tpaca->kvm_hstate.fake_suspend = 0;
2534 	/* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2535 	smp_wmb();
2536 	tpaca->kvm_hstate.kvm_vcore = vc;
2537 	if (cpu != smp_processor_id())
2538 		kvmppc_ipi_thread(cpu);
2539 }
2540 
2541 static void kvmppc_wait_for_nap(int n_threads)
2542 {
2543 	int cpu = smp_processor_id();
2544 	int i, loops;
2545 
2546 	if (n_threads <= 1)
2547 		return;
2548 	for (loops = 0; loops < 1000000; ++loops) {
2549 		/*
2550 		 * Check if all threads are finished.
2551 		 * We set the vcore pointer when starting a thread
2552 		 * and the thread clears it when finished, so we look
2553 		 * for any threads that still have a non-NULL vcore ptr.
2554 		 */
2555 		for (i = 1; i < n_threads; ++i)
2556 			if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2557 				break;
2558 		if (i == n_threads) {
2559 			HMT_medium();
2560 			return;
2561 		}
2562 		HMT_low();
2563 	}
2564 	HMT_medium();
2565 	for (i = 1; i < n_threads; ++i)
2566 		if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2567 			pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2568 }
2569 
2570 /*
2571  * Check that we are on thread 0 and that any other threads in
2572  * this core are off-line.  Then grab the threads so they can't
2573  * enter the kernel.
2574  */
2575 static int on_primary_thread(void)
2576 {
2577 	int cpu = smp_processor_id();
2578 	int thr;
2579 
2580 	/* Are we on a primary subcore? */
2581 	if (cpu_thread_in_subcore(cpu))
2582 		return 0;
2583 
2584 	thr = 0;
2585 	while (++thr < threads_per_subcore)
2586 		if (cpu_online(cpu + thr))
2587 			return 0;
2588 
2589 	/* Grab all hw threads so they can't go into the kernel */
2590 	for (thr = 1; thr < threads_per_subcore; ++thr) {
2591 		if (kvmppc_grab_hwthread(cpu + thr)) {
2592 			/* Couldn't grab one; let the others go */
2593 			do {
2594 				kvmppc_release_hwthread(cpu + thr);
2595 			} while (--thr > 0);
2596 			return 0;
2597 		}
2598 	}
2599 	return 1;
2600 }
2601 
2602 /*
2603  * A list of virtual cores for each physical CPU.
2604  * These are vcores that could run but their runner VCPU tasks are
2605  * (or may be) preempted.
2606  */
2607 struct preempted_vcore_list {
2608 	struct list_head	list;
2609 	spinlock_t		lock;
2610 };
2611 
2612 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2613 
2614 static void init_vcore_lists(void)
2615 {
2616 	int cpu;
2617 
2618 	for_each_possible_cpu(cpu) {
2619 		struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2620 		spin_lock_init(&lp->lock);
2621 		INIT_LIST_HEAD(&lp->list);
2622 	}
2623 }
2624 
2625 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2626 {
2627 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2628 
2629 	vc->vcore_state = VCORE_PREEMPT;
2630 	vc->pcpu = smp_processor_id();
2631 	if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2632 		spin_lock(&lp->lock);
2633 		list_add_tail(&vc->preempt_list, &lp->list);
2634 		spin_unlock(&lp->lock);
2635 	}
2636 
2637 	/* Start accumulating stolen time */
2638 	kvmppc_core_start_stolen(vc);
2639 }
2640 
2641 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2642 {
2643 	struct preempted_vcore_list *lp;
2644 
2645 	kvmppc_core_end_stolen(vc);
2646 	if (!list_empty(&vc->preempt_list)) {
2647 		lp = &per_cpu(preempted_vcores, vc->pcpu);
2648 		spin_lock(&lp->lock);
2649 		list_del_init(&vc->preempt_list);
2650 		spin_unlock(&lp->lock);
2651 	}
2652 	vc->vcore_state = VCORE_INACTIVE;
2653 }
2654 
2655 /*
2656  * This stores information about the virtual cores currently
2657  * assigned to a physical core.
2658  */
2659 struct core_info {
2660 	int		n_subcores;
2661 	int		max_subcore_threads;
2662 	int		total_threads;
2663 	int		subcore_threads[MAX_SUBCORES];
2664 	struct kvmppc_vcore *vc[MAX_SUBCORES];
2665 };
2666 
2667 /*
2668  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2669  * respectively in 2-way micro-threading (split-core) mode on POWER8.
2670  */
2671 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2672 
2673 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2674 {
2675 	memset(cip, 0, sizeof(*cip));
2676 	cip->n_subcores = 1;
2677 	cip->max_subcore_threads = vc->num_threads;
2678 	cip->total_threads = vc->num_threads;
2679 	cip->subcore_threads[0] = vc->num_threads;
2680 	cip->vc[0] = vc;
2681 }
2682 
2683 static bool subcore_config_ok(int n_subcores, int n_threads)
2684 {
2685 	/*
2686 	 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
2687 	 * split-core mode, with one thread per subcore.
2688 	 */
2689 	if (cpu_has_feature(CPU_FTR_ARCH_300))
2690 		return n_subcores <= 4 && n_threads == 1;
2691 
2692 	/* On POWER8, can only dynamically split if unsplit to begin with */
2693 	if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2694 		return false;
2695 	if (n_subcores > MAX_SUBCORES)
2696 		return false;
2697 	if (n_subcores > 1) {
2698 		if (!(dynamic_mt_modes & 2))
2699 			n_subcores = 4;
2700 		if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2701 			return false;
2702 	}
2703 
2704 	return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2705 }
2706 
2707 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2708 {
2709 	vc->entry_exit_map = 0;
2710 	vc->in_guest = 0;
2711 	vc->napping_threads = 0;
2712 	vc->conferring_threads = 0;
2713 	vc->tb_offset_applied = 0;
2714 }
2715 
2716 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2717 {
2718 	int n_threads = vc->num_threads;
2719 	int sub;
2720 
2721 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2722 		return false;
2723 
2724 	/* In one_vm_per_core mode, require all vcores to be from the same vm */
2725 	if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
2726 		return false;
2727 
2728 	/* Some POWER9 chips require all threads to be in the same MMU mode */
2729 	if (no_mixing_hpt_and_radix &&
2730 	    kvm_is_radix(vc->kvm) != kvm_is_radix(cip->vc[0]->kvm))
2731 		return false;
2732 
2733 	if (n_threads < cip->max_subcore_threads)
2734 		n_threads = cip->max_subcore_threads;
2735 	if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2736 		return false;
2737 	cip->max_subcore_threads = n_threads;
2738 
2739 	sub = cip->n_subcores;
2740 	++cip->n_subcores;
2741 	cip->total_threads += vc->num_threads;
2742 	cip->subcore_threads[sub] = vc->num_threads;
2743 	cip->vc[sub] = vc;
2744 	init_vcore_to_run(vc);
2745 	list_del_init(&vc->preempt_list);
2746 
2747 	return true;
2748 }
2749 
2750 /*
2751  * Work out whether it is possible to piggyback the execution of
2752  * vcore *pvc onto the execution of the other vcores described in *cip.
2753  */
2754 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2755 			  int target_threads)
2756 {
2757 	if (cip->total_threads + pvc->num_threads > target_threads)
2758 		return false;
2759 
2760 	return can_dynamic_split(pvc, cip);
2761 }
2762 
2763 static void prepare_threads(struct kvmppc_vcore *vc)
2764 {
2765 	int i;
2766 	struct kvm_vcpu *vcpu;
2767 
2768 	for_each_runnable_thread(i, vcpu, vc) {
2769 		if (signal_pending(vcpu->arch.run_task))
2770 			vcpu->arch.ret = -EINTR;
2771 		else if (vcpu->arch.vpa.update_pending ||
2772 			 vcpu->arch.slb_shadow.update_pending ||
2773 			 vcpu->arch.dtl.update_pending)
2774 			vcpu->arch.ret = RESUME_GUEST;
2775 		else
2776 			continue;
2777 		kvmppc_remove_runnable(vc, vcpu);
2778 		wake_up(&vcpu->arch.cpu_run);
2779 	}
2780 }
2781 
2782 static void collect_piggybacks(struct core_info *cip, int target_threads)
2783 {
2784 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2785 	struct kvmppc_vcore *pvc, *vcnext;
2786 
2787 	spin_lock(&lp->lock);
2788 	list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2789 		if (!spin_trylock(&pvc->lock))
2790 			continue;
2791 		prepare_threads(pvc);
2792 		if (!pvc->n_runnable) {
2793 			list_del_init(&pvc->preempt_list);
2794 			if (pvc->runner == NULL) {
2795 				pvc->vcore_state = VCORE_INACTIVE;
2796 				kvmppc_core_end_stolen(pvc);
2797 			}
2798 			spin_unlock(&pvc->lock);
2799 			continue;
2800 		}
2801 		if (!can_piggyback(pvc, cip, target_threads)) {
2802 			spin_unlock(&pvc->lock);
2803 			continue;
2804 		}
2805 		kvmppc_core_end_stolen(pvc);
2806 		pvc->vcore_state = VCORE_PIGGYBACK;
2807 		if (cip->total_threads >= target_threads)
2808 			break;
2809 	}
2810 	spin_unlock(&lp->lock);
2811 }
2812 
2813 static bool recheck_signals(struct core_info *cip)
2814 {
2815 	int sub, i;
2816 	struct kvm_vcpu *vcpu;
2817 
2818 	for (sub = 0; sub < cip->n_subcores; ++sub)
2819 		for_each_runnable_thread(i, vcpu, cip->vc[sub])
2820 			if (signal_pending(vcpu->arch.run_task))
2821 				return true;
2822 	return false;
2823 }
2824 
2825 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2826 {
2827 	int still_running = 0, i;
2828 	u64 now;
2829 	long ret;
2830 	struct kvm_vcpu *vcpu;
2831 
2832 	spin_lock(&vc->lock);
2833 	now = get_tb();
2834 	for_each_runnable_thread(i, vcpu, vc) {
2835 		/*
2836 		 * It's safe to unlock the vcore in the loop here, because
2837 		 * for_each_runnable_thread() is safe against removal of
2838 		 * the vcpu, and the vcore state is VCORE_EXITING here,
2839 		 * so any vcpus becoming runnable will have their arch.trap
2840 		 * set to zero and can't actually run in the guest.
2841 		 */
2842 		spin_unlock(&vc->lock);
2843 		/* cancel pending dec exception if dec is positive */
2844 		if (now < vcpu->arch.dec_expires &&
2845 		    kvmppc_core_pending_dec(vcpu))
2846 			kvmppc_core_dequeue_dec(vcpu);
2847 
2848 		trace_kvm_guest_exit(vcpu);
2849 
2850 		ret = RESUME_GUEST;
2851 		if (vcpu->arch.trap)
2852 			ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2853 						    vcpu->arch.run_task);
2854 
2855 		vcpu->arch.ret = ret;
2856 		vcpu->arch.trap = 0;
2857 
2858 		spin_lock(&vc->lock);
2859 		if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2860 			if (vcpu->arch.pending_exceptions)
2861 				kvmppc_core_prepare_to_enter(vcpu);
2862 			if (vcpu->arch.ceded)
2863 				kvmppc_set_timer(vcpu);
2864 			else
2865 				++still_running;
2866 		} else {
2867 			kvmppc_remove_runnable(vc, vcpu);
2868 			wake_up(&vcpu->arch.cpu_run);
2869 		}
2870 	}
2871 	if (!is_master) {
2872 		if (still_running > 0) {
2873 			kvmppc_vcore_preempt(vc);
2874 		} else if (vc->runner) {
2875 			vc->vcore_state = VCORE_PREEMPT;
2876 			kvmppc_core_start_stolen(vc);
2877 		} else {
2878 			vc->vcore_state = VCORE_INACTIVE;
2879 		}
2880 		if (vc->n_runnable > 0 && vc->runner == NULL) {
2881 			/* make sure there's a candidate runner awake */
2882 			i = -1;
2883 			vcpu = next_runnable_thread(vc, &i);
2884 			wake_up(&vcpu->arch.cpu_run);
2885 		}
2886 	}
2887 	spin_unlock(&vc->lock);
2888 }
2889 
2890 /*
2891  * Clear core from the list of active host cores as we are about to
2892  * enter the guest. Only do this if it is the primary thread of the
2893  * core (not if a subcore) that is entering the guest.
2894  */
2895 static inline int kvmppc_clear_host_core(unsigned int cpu)
2896 {
2897 	int core;
2898 
2899 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2900 		return 0;
2901 	/*
2902 	 * Memory barrier can be omitted here as we will do a smp_wmb()
2903 	 * later in kvmppc_start_thread and we need ensure that state is
2904 	 * visible to other CPUs only after we enter guest.
2905 	 */
2906 	core = cpu >> threads_shift;
2907 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2908 	return 0;
2909 }
2910 
2911 /*
2912  * Advertise this core as an active host core since we exited the guest
2913  * Only need to do this if it is the primary thread of the core that is
2914  * exiting.
2915  */
2916 static inline int kvmppc_set_host_core(unsigned int cpu)
2917 {
2918 	int core;
2919 
2920 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2921 		return 0;
2922 
2923 	/*
2924 	 * Memory barrier can be omitted here because we do a spin_unlock
2925 	 * immediately after this which provides the memory barrier.
2926 	 */
2927 	core = cpu >> threads_shift;
2928 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2929 	return 0;
2930 }
2931 
2932 static void set_irq_happened(int trap)
2933 {
2934 	switch (trap) {
2935 	case BOOK3S_INTERRUPT_EXTERNAL:
2936 		local_paca->irq_happened |= PACA_IRQ_EE;
2937 		break;
2938 	case BOOK3S_INTERRUPT_H_DOORBELL:
2939 		local_paca->irq_happened |= PACA_IRQ_DBELL;
2940 		break;
2941 	case BOOK3S_INTERRUPT_HMI:
2942 		local_paca->irq_happened |= PACA_IRQ_HMI;
2943 		break;
2944 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
2945 		replay_system_reset();
2946 		break;
2947 	}
2948 }
2949 
2950 /*
2951  * Run a set of guest threads on a physical core.
2952  * Called with vc->lock held.
2953  */
2954 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2955 {
2956 	struct kvm_vcpu *vcpu;
2957 	int i;
2958 	int srcu_idx;
2959 	struct core_info core_info;
2960 	struct kvmppc_vcore *pvc;
2961 	struct kvm_split_mode split_info, *sip;
2962 	int split, subcore_size, active;
2963 	int sub;
2964 	bool thr0_done;
2965 	unsigned long cmd_bit, stat_bit;
2966 	int pcpu, thr;
2967 	int target_threads;
2968 	int controlled_threads;
2969 	int trap;
2970 	bool is_power8;
2971 	bool hpt_on_radix;
2972 
2973 	/*
2974 	 * Remove from the list any threads that have a signal pending
2975 	 * or need a VPA update done
2976 	 */
2977 	prepare_threads(vc);
2978 
2979 	/* if the runner is no longer runnable, let the caller pick a new one */
2980 	if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2981 		return;
2982 
2983 	/*
2984 	 * Initialize *vc.
2985 	 */
2986 	init_vcore_to_run(vc);
2987 	vc->preempt_tb = TB_NIL;
2988 
2989 	/*
2990 	 * Number of threads that we will be controlling: the same as
2991 	 * the number of threads per subcore, except on POWER9,
2992 	 * where it's 1 because the threads are (mostly) independent.
2993 	 */
2994 	controlled_threads = threads_per_vcore(vc->kvm);
2995 
2996 	/*
2997 	 * Make sure we are running on primary threads, and that secondary
2998 	 * threads are offline.  Also check if the number of threads in this
2999 	 * guest are greater than the current system threads per guest.
3000 	 * On POWER9, we need to be not in independent-threads mode if
3001 	 * this is a HPT guest on a radix host machine where the
3002 	 * CPU threads may not be in different MMU modes.
3003 	 */
3004 	hpt_on_radix = no_mixing_hpt_and_radix && radix_enabled() &&
3005 		!kvm_is_radix(vc->kvm);
3006 	if (((controlled_threads > 1) &&
3007 	     ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) ||
3008 	    (hpt_on_radix && vc->kvm->arch.threads_indep)) {
3009 		for_each_runnable_thread(i, vcpu, vc) {
3010 			vcpu->arch.ret = -EBUSY;
3011 			kvmppc_remove_runnable(vc, vcpu);
3012 			wake_up(&vcpu->arch.cpu_run);
3013 		}
3014 		goto out;
3015 	}
3016 
3017 	/*
3018 	 * See if we could run any other vcores on the physical core
3019 	 * along with this one.
3020 	 */
3021 	init_core_info(&core_info, vc);
3022 	pcpu = smp_processor_id();
3023 	target_threads = controlled_threads;
3024 	if (target_smt_mode && target_smt_mode < target_threads)
3025 		target_threads = target_smt_mode;
3026 	if (vc->num_threads < target_threads)
3027 		collect_piggybacks(&core_info, target_threads);
3028 
3029 	/*
3030 	 * On radix, arrange for TLB flushing if necessary.
3031 	 * This has to be done before disabling interrupts since
3032 	 * it uses smp_call_function().
3033 	 */
3034 	pcpu = smp_processor_id();
3035 	if (kvm_is_radix(vc->kvm)) {
3036 		for (sub = 0; sub < core_info.n_subcores; ++sub)
3037 			for_each_runnable_thread(i, vcpu, core_info.vc[sub])
3038 				kvmppc_prepare_radix_vcpu(vcpu, pcpu);
3039 	}
3040 
3041 	/*
3042 	 * Hard-disable interrupts, and check resched flag and signals.
3043 	 * If we need to reschedule or deliver a signal, clean up
3044 	 * and return without going into the guest(s).
3045 	 * If the mmu_ready flag has been cleared, don't go into the
3046 	 * guest because that means a HPT resize operation is in progress.
3047 	 */
3048 	local_irq_disable();
3049 	hard_irq_disable();
3050 	if (lazy_irq_pending() || need_resched() ||
3051 	    recheck_signals(&core_info) || !vc->kvm->arch.mmu_ready) {
3052 		local_irq_enable();
3053 		vc->vcore_state = VCORE_INACTIVE;
3054 		/* Unlock all except the primary vcore */
3055 		for (sub = 1; sub < core_info.n_subcores; ++sub) {
3056 			pvc = core_info.vc[sub];
3057 			/* Put back on to the preempted vcores list */
3058 			kvmppc_vcore_preempt(pvc);
3059 			spin_unlock(&pvc->lock);
3060 		}
3061 		for (i = 0; i < controlled_threads; ++i)
3062 			kvmppc_release_hwthread(pcpu + i);
3063 		return;
3064 	}
3065 
3066 	kvmppc_clear_host_core(pcpu);
3067 
3068 	/* Decide on micro-threading (split-core) mode */
3069 	subcore_size = threads_per_subcore;
3070 	cmd_bit = stat_bit = 0;
3071 	split = core_info.n_subcores;
3072 	sip = NULL;
3073 	is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
3074 		&& !cpu_has_feature(CPU_FTR_ARCH_300);
3075 
3076 	if (split > 1 || hpt_on_radix) {
3077 		sip = &split_info;
3078 		memset(&split_info, 0, sizeof(split_info));
3079 		for (sub = 0; sub < core_info.n_subcores; ++sub)
3080 			split_info.vc[sub] = core_info.vc[sub];
3081 
3082 		if (is_power8) {
3083 			if (split == 2 && (dynamic_mt_modes & 2)) {
3084 				cmd_bit = HID0_POWER8_1TO2LPAR;
3085 				stat_bit = HID0_POWER8_2LPARMODE;
3086 			} else {
3087 				split = 4;
3088 				cmd_bit = HID0_POWER8_1TO4LPAR;
3089 				stat_bit = HID0_POWER8_4LPARMODE;
3090 			}
3091 			subcore_size = MAX_SMT_THREADS / split;
3092 			split_info.rpr = mfspr(SPRN_RPR);
3093 			split_info.pmmar = mfspr(SPRN_PMMAR);
3094 			split_info.ldbar = mfspr(SPRN_LDBAR);
3095 			split_info.subcore_size = subcore_size;
3096 		} else {
3097 			split_info.subcore_size = 1;
3098 			if (hpt_on_radix) {
3099 				/* Use the split_info for LPCR/LPIDR changes */
3100 				split_info.lpcr_req = vc->lpcr;
3101 				split_info.lpidr_req = vc->kvm->arch.lpid;
3102 				split_info.host_lpcr = vc->kvm->arch.host_lpcr;
3103 				split_info.do_set = 1;
3104 			}
3105 		}
3106 
3107 		/* order writes to split_info before kvm_split_mode pointer */
3108 		smp_wmb();
3109 	}
3110 
3111 	for (thr = 0; thr < controlled_threads; ++thr) {
3112 		struct paca_struct *paca = paca_ptrs[pcpu + thr];
3113 
3114 		paca->kvm_hstate.tid = thr;
3115 		paca->kvm_hstate.napping = 0;
3116 		paca->kvm_hstate.kvm_split_mode = sip;
3117 	}
3118 
3119 	/* Initiate micro-threading (split-core) on POWER8 if required */
3120 	if (cmd_bit) {
3121 		unsigned long hid0 = mfspr(SPRN_HID0);
3122 
3123 		hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3124 		mb();
3125 		mtspr(SPRN_HID0, hid0);
3126 		isync();
3127 		for (;;) {
3128 			hid0 = mfspr(SPRN_HID0);
3129 			if (hid0 & stat_bit)
3130 				break;
3131 			cpu_relax();
3132 		}
3133 	}
3134 
3135 	/*
3136 	 * On POWER8, set RWMR register.
3137 	 * Since it only affects PURR and SPURR, it doesn't affect
3138 	 * the host, so we don't save/restore the host value.
3139 	 */
3140 	if (is_power8) {
3141 		unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3142 		int n_online = atomic_read(&vc->online_count);
3143 
3144 		/*
3145 		 * Use the 8-thread value if we're doing split-core
3146 		 * or if the vcore's online count looks bogus.
3147 		 */
3148 		if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3149 		    n_online >= 1 && n_online <= MAX_SMT_THREADS)
3150 			rwmr_val = p8_rwmr_values[n_online];
3151 		mtspr(SPRN_RWMR, rwmr_val);
3152 	}
3153 
3154 	/* Start all the threads */
3155 	active = 0;
3156 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
3157 		thr = is_power8 ? subcore_thread_map[sub] : sub;
3158 		thr0_done = false;
3159 		active |= 1 << thr;
3160 		pvc = core_info.vc[sub];
3161 		pvc->pcpu = pcpu + thr;
3162 		for_each_runnable_thread(i, vcpu, pvc) {
3163 			kvmppc_start_thread(vcpu, pvc);
3164 			kvmppc_create_dtl_entry(vcpu, pvc);
3165 			trace_kvm_guest_enter(vcpu);
3166 			if (!vcpu->arch.ptid)
3167 				thr0_done = true;
3168 			active |= 1 << (thr + vcpu->arch.ptid);
3169 		}
3170 		/*
3171 		 * We need to start the first thread of each subcore
3172 		 * even if it doesn't have a vcpu.
3173 		 */
3174 		if (!thr0_done)
3175 			kvmppc_start_thread(NULL, pvc);
3176 	}
3177 
3178 	/*
3179 	 * Ensure that split_info.do_nap is set after setting
3180 	 * the vcore pointer in the PACA of the secondaries.
3181 	 */
3182 	smp_mb();
3183 
3184 	/*
3185 	 * When doing micro-threading, poke the inactive threads as well.
3186 	 * This gets them to the nap instruction after kvm_do_nap,
3187 	 * which reduces the time taken to unsplit later.
3188 	 * For POWER9 HPT guest on radix host, we need all the secondary
3189 	 * threads woken up so they can do the LPCR/LPIDR change.
3190 	 */
3191 	if (cmd_bit || hpt_on_radix) {
3192 		split_info.do_nap = 1;	/* ask secondaries to nap when done */
3193 		for (thr = 1; thr < threads_per_subcore; ++thr)
3194 			if (!(active & (1 << thr)))
3195 				kvmppc_ipi_thread(pcpu + thr);
3196 	}
3197 
3198 	vc->vcore_state = VCORE_RUNNING;
3199 	preempt_disable();
3200 
3201 	trace_kvmppc_run_core(vc, 0);
3202 
3203 	for (sub = 0; sub < core_info.n_subcores; ++sub)
3204 		spin_unlock(&core_info.vc[sub]->lock);
3205 
3206 	if (kvm_is_radix(vc->kvm)) {
3207 		/*
3208 		 * Do we need to flush the process scoped TLB for the LPAR?
3209 		 *
3210 		 * On POWER9, individual threads can come in here, but the
3211 		 * TLB is shared between the 4 threads in a core, hence
3212 		 * invalidating on one thread invalidates for all.
3213 		 * Thus we make all 4 threads use the same bit here.
3214 		 *
3215 		 * Hash must be flushed in realmode in order to use tlbiel.
3216 		 */
3217 		kvmppc_radix_check_need_tlb_flush(vc->kvm, pcpu, NULL);
3218 	}
3219 
3220 	/*
3221 	 * Interrupts will be enabled once we get into the guest,
3222 	 * so tell lockdep that we're about to enable interrupts.
3223 	 */
3224 	trace_hardirqs_on();
3225 
3226 	guest_enter_irqoff();
3227 
3228 	srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3229 
3230 	this_cpu_disable_ftrace();
3231 
3232 	trap = __kvmppc_vcore_entry();
3233 
3234 	this_cpu_enable_ftrace();
3235 
3236 	srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3237 
3238 	trace_hardirqs_off();
3239 	set_irq_happened(trap);
3240 
3241 	spin_lock(&vc->lock);
3242 	/* prevent other vcpu threads from doing kvmppc_start_thread() now */
3243 	vc->vcore_state = VCORE_EXITING;
3244 
3245 	/* wait for secondary threads to finish writing their state to memory */
3246 	kvmppc_wait_for_nap(controlled_threads);
3247 
3248 	/* Return to whole-core mode if we split the core earlier */
3249 	if (cmd_bit) {
3250 		unsigned long hid0 = mfspr(SPRN_HID0);
3251 		unsigned long loops = 0;
3252 
3253 		hid0 &= ~HID0_POWER8_DYNLPARDIS;
3254 		stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3255 		mb();
3256 		mtspr(SPRN_HID0, hid0);
3257 		isync();
3258 		for (;;) {
3259 			hid0 = mfspr(SPRN_HID0);
3260 			if (!(hid0 & stat_bit))
3261 				break;
3262 			cpu_relax();
3263 			++loops;
3264 		}
3265 	} else if (hpt_on_radix) {
3266 		/* Wait for all threads to have seen final sync */
3267 		for (thr = 1; thr < controlled_threads; ++thr) {
3268 			struct paca_struct *paca = paca_ptrs[pcpu + thr];
3269 
3270 			while (paca->kvm_hstate.kvm_split_mode) {
3271 				HMT_low();
3272 				barrier();
3273 			}
3274 			HMT_medium();
3275 		}
3276 	}
3277 	split_info.do_nap = 0;
3278 
3279 	kvmppc_set_host_core(pcpu);
3280 
3281 	local_irq_enable();
3282 	guest_exit();
3283 
3284 	/* Let secondaries go back to the offline loop */
3285 	for (i = 0; i < controlled_threads; ++i) {
3286 		kvmppc_release_hwthread(pcpu + i);
3287 		if (sip && sip->napped[i])
3288 			kvmppc_ipi_thread(pcpu + i);
3289 		cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
3290 	}
3291 
3292 	spin_unlock(&vc->lock);
3293 
3294 	/* make sure updates to secondary vcpu structs are visible now */
3295 	smp_mb();
3296 
3297 	preempt_enable();
3298 
3299 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
3300 		pvc = core_info.vc[sub];
3301 		post_guest_process(pvc, pvc == vc);
3302 	}
3303 
3304 	spin_lock(&vc->lock);
3305 
3306  out:
3307 	vc->vcore_state = VCORE_INACTIVE;
3308 	trace_kvmppc_run_core(vc, 1);
3309 }
3310 
3311 /*
3312  * Load up hypervisor-mode registers on P9.
3313  */
3314 static int kvmhv_load_hv_regs_and_go(struct kvm_vcpu *vcpu, u64 time_limit,
3315 				     unsigned long lpcr)
3316 {
3317 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
3318 	s64 hdec;
3319 	u64 tb, purr, spurr;
3320 	int trap;
3321 	unsigned long host_hfscr = mfspr(SPRN_HFSCR);
3322 	unsigned long host_ciabr = mfspr(SPRN_CIABR);
3323 	unsigned long host_dawr = mfspr(SPRN_DAWR);
3324 	unsigned long host_dawrx = mfspr(SPRN_DAWRX);
3325 	unsigned long host_psscr = mfspr(SPRN_PSSCR);
3326 	unsigned long host_pidr = mfspr(SPRN_PID);
3327 
3328 	hdec = time_limit - mftb();
3329 	if (hdec < 0)
3330 		return BOOK3S_INTERRUPT_HV_DECREMENTER;
3331 	mtspr(SPRN_HDEC, hdec);
3332 
3333 	if (vc->tb_offset) {
3334 		u64 new_tb = mftb() + vc->tb_offset;
3335 		mtspr(SPRN_TBU40, new_tb);
3336 		tb = mftb();
3337 		if ((tb & 0xffffff) < (new_tb & 0xffffff))
3338 			mtspr(SPRN_TBU40, new_tb + 0x1000000);
3339 		vc->tb_offset_applied = vc->tb_offset;
3340 	}
3341 
3342 	if (vc->pcr)
3343 		mtspr(SPRN_PCR, vc->pcr);
3344 	mtspr(SPRN_DPDES, vc->dpdes);
3345 	mtspr(SPRN_VTB, vc->vtb);
3346 
3347 	local_paca->kvm_hstate.host_purr = mfspr(SPRN_PURR);
3348 	local_paca->kvm_hstate.host_spurr = mfspr(SPRN_SPURR);
3349 	mtspr(SPRN_PURR, vcpu->arch.purr);
3350 	mtspr(SPRN_SPURR, vcpu->arch.spurr);
3351 
3352 	if (cpu_has_feature(CPU_FTR_DAWR)) {
3353 		mtspr(SPRN_DAWR, vcpu->arch.dawr);
3354 		mtspr(SPRN_DAWRX, vcpu->arch.dawrx);
3355 	}
3356 	mtspr(SPRN_CIABR, vcpu->arch.ciabr);
3357 	mtspr(SPRN_IC, vcpu->arch.ic);
3358 	mtspr(SPRN_PID, vcpu->arch.pid);
3359 
3360 	mtspr(SPRN_PSSCR, vcpu->arch.psscr | PSSCR_EC |
3361 	      (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3362 
3363 	mtspr(SPRN_HFSCR, vcpu->arch.hfscr);
3364 
3365 	mtspr(SPRN_SPRG0, vcpu->arch.shregs.sprg0);
3366 	mtspr(SPRN_SPRG1, vcpu->arch.shregs.sprg1);
3367 	mtspr(SPRN_SPRG2, vcpu->arch.shregs.sprg2);
3368 	mtspr(SPRN_SPRG3, vcpu->arch.shregs.sprg3);
3369 
3370 	mtspr(SPRN_AMOR, ~0UL);
3371 
3372 	mtspr(SPRN_LPCR, lpcr);
3373 	isync();
3374 
3375 	kvmppc_xive_push_vcpu(vcpu);
3376 
3377 	mtspr(SPRN_SRR0, vcpu->arch.shregs.srr0);
3378 	mtspr(SPRN_SRR1, vcpu->arch.shregs.srr1);
3379 
3380 	trap = __kvmhv_vcpu_entry_p9(vcpu);
3381 
3382 	/* Advance host PURR/SPURR by the amount used by guest */
3383 	purr = mfspr(SPRN_PURR);
3384 	spurr = mfspr(SPRN_SPURR);
3385 	mtspr(SPRN_PURR, local_paca->kvm_hstate.host_purr +
3386 	      purr - vcpu->arch.purr);
3387 	mtspr(SPRN_SPURR, local_paca->kvm_hstate.host_spurr +
3388 	      spurr - vcpu->arch.spurr);
3389 	vcpu->arch.purr = purr;
3390 	vcpu->arch.spurr = spurr;
3391 
3392 	vcpu->arch.ic = mfspr(SPRN_IC);
3393 	vcpu->arch.pid = mfspr(SPRN_PID);
3394 	vcpu->arch.psscr = mfspr(SPRN_PSSCR) & PSSCR_GUEST_VIS;
3395 
3396 	vcpu->arch.shregs.sprg0 = mfspr(SPRN_SPRG0);
3397 	vcpu->arch.shregs.sprg1 = mfspr(SPRN_SPRG1);
3398 	vcpu->arch.shregs.sprg2 = mfspr(SPRN_SPRG2);
3399 	vcpu->arch.shregs.sprg3 = mfspr(SPRN_SPRG3);
3400 
3401 	mtspr(SPRN_PSSCR, host_psscr);
3402 	mtspr(SPRN_HFSCR, host_hfscr);
3403 	mtspr(SPRN_CIABR, host_ciabr);
3404 	mtspr(SPRN_DAWR, host_dawr);
3405 	mtspr(SPRN_DAWRX, host_dawrx);
3406 	mtspr(SPRN_PID, host_pidr);
3407 
3408 	/*
3409 	 * Since this is radix, do a eieio; tlbsync; ptesync sequence in
3410 	 * case we interrupted the guest between a tlbie and a ptesync.
3411 	 */
3412 	asm volatile("eieio; tlbsync; ptesync");
3413 
3414 	mtspr(SPRN_LPID, vcpu->kvm->arch.host_lpid);	/* restore host LPID */
3415 	isync();
3416 
3417 	vc->dpdes = mfspr(SPRN_DPDES);
3418 	vc->vtb = mfspr(SPRN_VTB);
3419 	mtspr(SPRN_DPDES, 0);
3420 	if (vc->pcr)
3421 		mtspr(SPRN_PCR, 0);
3422 
3423 	if (vc->tb_offset_applied) {
3424 		u64 new_tb = mftb() - vc->tb_offset_applied;
3425 		mtspr(SPRN_TBU40, new_tb);
3426 		tb = mftb();
3427 		if ((tb & 0xffffff) < (new_tb & 0xffffff))
3428 			mtspr(SPRN_TBU40, new_tb + 0x1000000);
3429 		vc->tb_offset_applied = 0;
3430 	}
3431 
3432 	mtspr(SPRN_HDEC, 0x7fffffff);
3433 	mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr);
3434 
3435 	return trap;
3436 }
3437 
3438 /*
3439  * Virtual-mode guest entry for POWER9 and later when the host and
3440  * guest are both using the radix MMU.  The LPIDR has already been set.
3441  */
3442 int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
3443 			 unsigned long lpcr)
3444 {
3445 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
3446 	unsigned long host_dscr = mfspr(SPRN_DSCR);
3447 	unsigned long host_tidr = mfspr(SPRN_TIDR);
3448 	unsigned long host_iamr = mfspr(SPRN_IAMR);
3449 	s64 dec;
3450 	u64 tb;
3451 	int trap, save_pmu;
3452 
3453 	dec = mfspr(SPRN_DEC);
3454 	tb = mftb();
3455 	if (dec < 512)
3456 		return BOOK3S_INTERRUPT_HV_DECREMENTER;
3457 	local_paca->kvm_hstate.dec_expires = dec + tb;
3458 	if (local_paca->kvm_hstate.dec_expires < time_limit)
3459 		time_limit = local_paca->kvm_hstate.dec_expires;
3460 
3461 	vcpu->arch.ceded = 0;
3462 
3463 	kvmhv_save_host_pmu();		/* saves it to PACA kvm_hstate */
3464 
3465 	kvmppc_subcore_enter_guest();
3466 
3467 	vc->entry_exit_map = 1;
3468 	vc->in_guest = 1;
3469 
3470 	if (vcpu->arch.vpa.pinned_addr) {
3471 		struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3472 		u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3473 		lp->yield_count = cpu_to_be32(yield_count);
3474 		vcpu->arch.vpa.dirty = 1;
3475 	}
3476 
3477 	if (cpu_has_feature(CPU_FTR_TM) ||
3478 	    cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3479 		kvmppc_restore_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3480 
3481 	kvmhv_load_guest_pmu(vcpu);
3482 
3483 	msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3484 	load_fp_state(&vcpu->arch.fp);
3485 #ifdef CONFIG_ALTIVEC
3486 	load_vr_state(&vcpu->arch.vr);
3487 #endif
3488 
3489 	mtspr(SPRN_DSCR, vcpu->arch.dscr);
3490 	mtspr(SPRN_IAMR, vcpu->arch.iamr);
3491 	mtspr(SPRN_PSPB, vcpu->arch.pspb);
3492 	mtspr(SPRN_FSCR, vcpu->arch.fscr);
3493 	mtspr(SPRN_TAR, vcpu->arch.tar);
3494 	mtspr(SPRN_EBBHR, vcpu->arch.ebbhr);
3495 	mtspr(SPRN_EBBRR, vcpu->arch.ebbrr);
3496 	mtspr(SPRN_BESCR, vcpu->arch.bescr);
3497 	mtspr(SPRN_WORT, vcpu->arch.wort);
3498 	mtspr(SPRN_TIDR, vcpu->arch.tid);
3499 	mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
3500 	mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
3501 	mtspr(SPRN_AMR, vcpu->arch.amr);
3502 	mtspr(SPRN_UAMOR, vcpu->arch.uamor);
3503 
3504 	if (!(vcpu->arch.ctrl & 1))
3505 		mtspr(SPRN_CTRLT, mfspr(SPRN_CTRLF) & ~1);
3506 
3507 	mtspr(SPRN_DEC, vcpu->arch.dec_expires - mftb());
3508 
3509 	if (kvmhv_on_pseries()) {
3510 		/* call our hypervisor to load up HV regs and go */
3511 		struct hv_guest_state hvregs;
3512 
3513 		kvmhv_save_hv_regs(vcpu, &hvregs);
3514 		hvregs.lpcr = lpcr;
3515 		vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
3516 		hvregs.version = HV_GUEST_STATE_VERSION;
3517 		if (vcpu->arch.nested) {
3518 			hvregs.lpid = vcpu->arch.nested->shadow_lpid;
3519 			hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
3520 		} else {
3521 			hvregs.lpid = vcpu->kvm->arch.lpid;
3522 			hvregs.vcpu_token = vcpu->vcpu_id;
3523 		}
3524 		hvregs.hdec_expiry = time_limit;
3525 		trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
3526 					  __pa(&vcpu->arch.regs));
3527 		kvmhv_restore_hv_return_state(vcpu, &hvregs);
3528 		vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
3529 		vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
3530 		vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
3531 
3532 		/* H_CEDE has to be handled now, not later */
3533 		if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
3534 		    kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
3535 			kvmppc_nested_cede(vcpu);
3536 			trap = 0;
3537 		}
3538 	} else {
3539 		trap = kvmhv_load_hv_regs_and_go(vcpu, time_limit, lpcr);
3540 	}
3541 
3542 	vcpu->arch.slb_max = 0;
3543 	dec = mfspr(SPRN_DEC);
3544 	tb = mftb();
3545 	vcpu->arch.dec_expires = dec + tb;
3546 	vcpu->cpu = -1;
3547 	vcpu->arch.thread_cpu = -1;
3548 	vcpu->arch.ctrl = mfspr(SPRN_CTRLF);
3549 
3550 	vcpu->arch.iamr = mfspr(SPRN_IAMR);
3551 	vcpu->arch.pspb = mfspr(SPRN_PSPB);
3552 	vcpu->arch.fscr = mfspr(SPRN_FSCR);
3553 	vcpu->arch.tar = mfspr(SPRN_TAR);
3554 	vcpu->arch.ebbhr = mfspr(SPRN_EBBHR);
3555 	vcpu->arch.ebbrr = mfspr(SPRN_EBBRR);
3556 	vcpu->arch.bescr = mfspr(SPRN_BESCR);
3557 	vcpu->arch.wort = mfspr(SPRN_WORT);
3558 	vcpu->arch.tid = mfspr(SPRN_TIDR);
3559 	vcpu->arch.amr = mfspr(SPRN_AMR);
3560 	vcpu->arch.uamor = mfspr(SPRN_UAMOR);
3561 	vcpu->arch.dscr = mfspr(SPRN_DSCR);
3562 
3563 	mtspr(SPRN_PSPB, 0);
3564 	mtspr(SPRN_WORT, 0);
3565 	mtspr(SPRN_AMR, 0);
3566 	mtspr(SPRN_UAMOR, 0);
3567 	mtspr(SPRN_DSCR, host_dscr);
3568 	mtspr(SPRN_TIDR, host_tidr);
3569 	mtspr(SPRN_IAMR, host_iamr);
3570 	mtspr(SPRN_PSPB, 0);
3571 
3572 	msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3573 	store_fp_state(&vcpu->arch.fp);
3574 #ifdef CONFIG_ALTIVEC
3575 	store_vr_state(&vcpu->arch.vr);
3576 #endif
3577 
3578 	if (cpu_has_feature(CPU_FTR_TM) ||
3579 	    cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3580 		kvmppc_save_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3581 
3582 	save_pmu = 1;
3583 	if (vcpu->arch.vpa.pinned_addr) {
3584 		struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3585 		u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3586 		lp->yield_count = cpu_to_be32(yield_count);
3587 		vcpu->arch.vpa.dirty = 1;
3588 		save_pmu = lp->pmcregs_in_use;
3589 	}
3590 
3591 	kvmhv_save_guest_pmu(vcpu, save_pmu);
3592 
3593 	vc->entry_exit_map = 0x101;
3594 	vc->in_guest = 0;
3595 
3596 	mtspr(SPRN_DEC, local_paca->kvm_hstate.dec_expires - mftb());
3597 
3598 	kvmhv_load_host_pmu();
3599 
3600 	kvmppc_subcore_exit_guest();
3601 
3602 	return trap;
3603 }
3604 
3605 /*
3606  * Wait for some other vcpu thread to execute us, and
3607  * wake us up when we need to handle something in the host.
3608  */
3609 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
3610 				 struct kvm_vcpu *vcpu, int wait_state)
3611 {
3612 	DEFINE_WAIT(wait);
3613 
3614 	prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
3615 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3616 		spin_unlock(&vc->lock);
3617 		schedule();
3618 		spin_lock(&vc->lock);
3619 	}
3620 	finish_wait(&vcpu->arch.cpu_run, &wait);
3621 }
3622 
3623 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
3624 {
3625 	/* 10us base */
3626 	if (vc->halt_poll_ns == 0 && halt_poll_ns_grow)
3627 		vc->halt_poll_ns = 10000;
3628 	else
3629 		vc->halt_poll_ns *= halt_poll_ns_grow;
3630 }
3631 
3632 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
3633 {
3634 	if (halt_poll_ns_shrink == 0)
3635 		vc->halt_poll_ns = 0;
3636 	else
3637 		vc->halt_poll_ns /= halt_poll_ns_shrink;
3638 }
3639 
3640 #ifdef CONFIG_KVM_XICS
3641 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3642 {
3643 	if (!xive_enabled())
3644 		return false;
3645 	return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
3646 		vcpu->arch.xive_saved_state.cppr;
3647 }
3648 #else
3649 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3650 {
3651 	return false;
3652 }
3653 #endif /* CONFIG_KVM_XICS */
3654 
3655 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3656 {
3657 	if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3658 	    kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3659 		return true;
3660 
3661 	return false;
3662 }
3663 
3664 /*
3665  * Check to see if any of the runnable vcpus on the vcore have pending
3666  * exceptions or are no longer ceded
3667  */
3668 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3669 {
3670 	struct kvm_vcpu *vcpu;
3671 	int i;
3672 
3673 	for_each_runnable_thread(i, vcpu, vc) {
3674 		if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3675 			return 1;
3676 	}
3677 
3678 	return 0;
3679 }
3680 
3681 /*
3682  * All the vcpus in this vcore are idle, so wait for a decrementer
3683  * or external interrupt to one of the vcpus.  vc->lock is held.
3684  */
3685 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3686 {
3687 	ktime_t cur, start_poll, start_wait;
3688 	int do_sleep = 1;
3689 	u64 block_ns;
3690 	DECLARE_SWAITQUEUE(wait);
3691 
3692 	/* Poll for pending exceptions and ceded state */
3693 	cur = start_poll = ktime_get();
3694 	if (vc->halt_poll_ns) {
3695 		ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3696 		++vc->runner->stat.halt_attempted_poll;
3697 
3698 		vc->vcore_state = VCORE_POLLING;
3699 		spin_unlock(&vc->lock);
3700 
3701 		do {
3702 			if (kvmppc_vcore_check_block(vc)) {
3703 				do_sleep = 0;
3704 				break;
3705 			}
3706 			cur = ktime_get();
3707 		} while (single_task_running() && ktime_before(cur, stop));
3708 
3709 		spin_lock(&vc->lock);
3710 		vc->vcore_state = VCORE_INACTIVE;
3711 
3712 		if (!do_sleep) {
3713 			++vc->runner->stat.halt_successful_poll;
3714 			goto out;
3715 		}
3716 	}
3717 
3718 	prepare_to_swait_exclusive(&vc->wq, &wait, TASK_INTERRUPTIBLE);
3719 
3720 	if (kvmppc_vcore_check_block(vc)) {
3721 		finish_swait(&vc->wq, &wait);
3722 		do_sleep = 0;
3723 		/* If we polled, count this as a successful poll */
3724 		if (vc->halt_poll_ns)
3725 			++vc->runner->stat.halt_successful_poll;
3726 		goto out;
3727 	}
3728 
3729 	start_wait = ktime_get();
3730 
3731 	vc->vcore_state = VCORE_SLEEPING;
3732 	trace_kvmppc_vcore_blocked(vc, 0);
3733 	spin_unlock(&vc->lock);
3734 	schedule();
3735 	finish_swait(&vc->wq, &wait);
3736 	spin_lock(&vc->lock);
3737 	vc->vcore_state = VCORE_INACTIVE;
3738 	trace_kvmppc_vcore_blocked(vc, 1);
3739 	++vc->runner->stat.halt_successful_wait;
3740 
3741 	cur = ktime_get();
3742 
3743 out:
3744 	block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3745 
3746 	/* Attribute wait time */
3747 	if (do_sleep) {
3748 		vc->runner->stat.halt_wait_ns +=
3749 			ktime_to_ns(cur) - ktime_to_ns(start_wait);
3750 		/* Attribute failed poll time */
3751 		if (vc->halt_poll_ns)
3752 			vc->runner->stat.halt_poll_fail_ns +=
3753 				ktime_to_ns(start_wait) -
3754 				ktime_to_ns(start_poll);
3755 	} else {
3756 		/* Attribute successful poll time */
3757 		if (vc->halt_poll_ns)
3758 			vc->runner->stat.halt_poll_success_ns +=
3759 				ktime_to_ns(cur) -
3760 				ktime_to_ns(start_poll);
3761 	}
3762 
3763 	/* Adjust poll time */
3764 	if (halt_poll_ns) {
3765 		if (block_ns <= vc->halt_poll_ns)
3766 			;
3767 		/* We slept and blocked for longer than the max halt time */
3768 		else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3769 			shrink_halt_poll_ns(vc);
3770 		/* We slept and our poll time is too small */
3771 		else if (vc->halt_poll_ns < halt_poll_ns &&
3772 				block_ns < halt_poll_ns)
3773 			grow_halt_poll_ns(vc);
3774 		if (vc->halt_poll_ns > halt_poll_ns)
3775 			vc->halt_poll_ns = halt_poll_ns;
3776 	} else
3777 		vc->halt_poll_ns = 0;
3778 
3779 	trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3780 }
3781 
3782 /*
3783  * This never fails for a radix guest, as none of the operations it does
3784  * for a radix guest can fail or have a way to report failure.
3785  * kvmhv_run_single_vcpu() relies on this fact.
3786  */
3787 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
3788 {
3789 	int r = 0;
3790 	struct kvm *kvm = vcpu->kvm;
3791 
3792 	mutex_lock(&kvm->lock);
3793 	if (!kvm->arch.mmu_ready) {
3794 		if (!kvm_is_radix(kvm))
3795 			r = kvmppc_hv_setup_htab_rma(vcpu);
3796 		if (!r) {
3797 			if (cpu_has_feature(CPU_FTR_ARCH_300))
3798 				kvmppc_setup_partition_table(kvm);
3799 			kvm->arch.mmu_ready = 1;
3800 		}
3801 	}
3802 	mutex_unlock(&kvm->lock);
3803 	return r;
3804 }
3805 
3806 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3807 {
3808 	int n_ceded, i, r;
3809 	struct kvmppc_vcore *vc;
3810 	struct kvm_vcpu *v;
3811 
3812 	trace_kvmppc_run_vcpu_enter(vcpu);
3813 
3814 	kvm_run->exit_reason = 0;
3815 	vcpu->arch.ret = RESUME_GUEST;
3816 	vcpu->arch.trap = 0;
3817 	kvmppc_update_vpas(vcpu);
3818 
3819 	/*
3820 	 * Synchronize with other threads in this virtual core
3821 	 */
3822 	vc = vcpu->arch.vcore;
3823 	spin_lock(&vc->lock);
3824 	vcpu->arch.ceded = 0;
3825 	vcpu->arch.run_task = current;
3826 	vcpu->arch.kvm_run = kvm_run;
3827 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3828 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3829 	vcpu->arch.busy_preempt = TB_NIL;
3830 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3831 	++vc->n_runnable;
3832 
3833 	/*
3834 	 * This happens the first time this is called for a vcpu.
3835 	 * If the vcore is already running, we may be able to start
3836 	 * this thread straight away and have it join in.
3837 	 */
3838 	if (!signal_pending(current)) {
3839 		if ((vc->vcore_state == VCORE_PIGGYBACK ||
3840 		     vc->vcore_state == VCORE_RUNNING) &&
3841 			   !VCORE_IS_EXITING(vc)) {
3842 			kvmppc_create_dtl_entry(vcpu, vc);
3843 			kvmppc_start_thread(vcpu, vc);
3844 			trace_kvm_guest_enter(vcpu);
3845 		} else if (vc->vcore_state == VCORE_SLEEPING) {
3846 			swake_up_one(&vc->wq);
3847 		}
3848 
3849 	}
3850 
3851 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3852 	       !signal_pending(current)) {
3853 		/* See if the MMU is ready to go */
3854 		if (!vcpu->kvm->arch.mmu_ready) {
3855 			spin_unlock(&vc->lock);
3856 			r = kvmhv_setup_mmu(vcpu);
3857 			spin_lock(&vc->lock);
3858 			if (r) {
3859 				kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3860 				kvm_run->fail_entry.
3861 					hardware_entry_failure_reason = 0;
3862 				vcpu->arch.ret = r;
3863 				break;
3864 			}
3865 		}
3866 
3867 		if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3868 			kvmppc_vcore_end_preempt(vc);
3869 
3870 		if (vc->vcore_state != VCORE_INACTIVE) {
3871 			kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
3872 			continue;
3873 		}
3874 		for_each_runnable_thread(i, v, vc) {
3875 			kvmppc_core_prepare_to_enter(v);
3876 			if (signal_pending(v->arch.run_task)) {
3877 				kvmppc_remove_runnable(vc, v);
3878 				v->stat.signal_exits++;
3879 				v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
3880 				v->arch.ret = -EINTR;
3881 				wake_up(&v->arch.cpu_run);
3882 			}
3883 		}
3884 		if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3885 			break;
3886 		n_ceded = 0;
3887 		for_each_runnable_thread(i, v, vc) {
3888 			if (!kvmppc_vcpu_woken(v))
3889 				n_ceded += v->arch.ceded;
3890 			else
3891 				v->arch.ceded = 0;
3892 		}
3893 		vc->runner = vcpu;
3894 		if (n_ceded == vc->n_runnable) {
3895 			kvmppc_vcore_blocked(vc);
3896 		} else if (need_resched()) {
3897 			kvmppc_vcore_preempt(vc);
3898 			/* Let something else run */
3899 			cond_resched_lock(&vc->lock);
3900 			if (vc->vcore_state == VCORE_PREEMPT)
3901 				kvmppc_vcore_end_preempt(vc);
3902 		} else {
3903 			kvmppc_run_core(vc);
3904 		}
3905 		vc->runner = NULL;
3906 	}
3907 
3908 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3909 	       (vc->vcore_state == VCORE_RUNNING ||
3910 		vc->vcore_state == VCORE_EXITING ||
3911 		vc->vcore_state == VCORE_PIGGYBACK))
3912 		kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
3913 
3914 	if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3915 		kvmppc_vcore_end_preempt(vc);
3916 
3917 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3918 		kvmppc_remove_runnable(vc, vcpu);
3919 		vcpu->stat.signal_exits++;
3920 		kvm_run->exit_reason = KVM_EXIT_INTR;
3921 		vcpu->arch.ret = -EINTR;
3922 	}
3923 
3924 	if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
3925 		/* Wake up some vcpu to run the core */
3926 		i = -1;
3927 		v = next_runnable_thread(vc, &i);
3928 		wake_up(&v->arch.cpu_run);
3929 	}
3930 
3931 	trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
3932 	spin_unlock(&vc->lock);
3933 	return vcpu->arch.ret;
3934 }
3935 
3936 int kvmhv_run_single_vcpu(struct kvm_run *kvm_run,
3937 			  struct kvm_vcpu *vcpu, u64 time_limit,
3938 			  unsigned long lpcr)
3939 {
3940 	int trap, r, pcpu;
3941 	int srcu_idx;
3942 	struct kvmppc_vcore *vc;
3943 	struct kvm *kvm = vcpu->kvm;
3944 	struct kvm_nested_guest *nested = vcpu->arch.nested;
3945 
3946 	trace_kvmppc_run_vcpu_enter(vcpu);
3947 
3948 	kvm_run->exit_reason = 0;
3949 	vcpu->arch.ret = RESUME_GUEST;
3950 	vcpu->arch.trap = 0;
3951 
3952 	vc = vcpu->arch.vcore;
3953 	vcpu->arch.ceded = 0;
3954 	vcpu->arch.run_task = current;
3955 	vcpu->arch.kvm_run = kvm_run;
3956 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3957 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3958 	vcpu->arch.busy_preempt = TB_NIL;
3959 	vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
3960 	vc->runnable_threads[0] = vcpu;
3961 	vc->n_runnable = 1;
3962 	vc->runner = vcpu;
3963 
3964 	/* See if the MMU is ready to go */
3965 	if (!kvm->arch.mmu_ready)
3966 		kvmhv_setup_mmu(vcpu);
3967 
3968 	if (need_resched())
3969 		cond_resched();
3970 
3971 	kvmppc_update_vpas(vcpu);
3972 
3973 	init_vcore_to_run(vc);
3974 	vc->preempt_tb = TB_NIL;
3975 
3976 	preempt_disable();
3977 	pcpu = smp_processor_id();
3978 	vc->pcpu = pcpu;
3979 	kvmppc_prepare_radix_vcpu(vcpu, pcpu);
3980 
3981 	local_irq_disable();
3982 	hard_irq_disable();
3983 	if (signal_pending(current))
3984 		goto sigpend;
3985 	if (lazy_irq_pending() || need_resched() || !kvm->arch.mmu_ready)
3986 		goto out;
3987 
3988 	if (!nested) {
3989 		kvmppc_core_prepare_to_enter(vcpu);
3990 		if (vcpu->arch.doorbell_request) {
3991 			vc->dpdes = 1;
3992 			smp_wmb();
3993 			vcpu->arch.doorbell_request = 0;
3994 		}
3995 		if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
3996 			     &vcpu->arch.pending_exceptions))
3997 			lpcr |= LPCR_MER;
3998 	} else if (vcpu->arch.pending_exceptions ||
3999 		   vcpu->arch.doorbell_request ||
4000 		   xive_interrupt_pending(vcpu)) {
4001 		vcpu->arch.ret = RESUME_HOST;
4002 		goto out;
4003 	}
4004 
4005 	kvmppc_clear_host_core(pcpu);
4006 
4007 	local_paca->kvm_hstate.tid = 0;
4008 	local_paca->kvm_hstate.napping = 0;
4009 	local_paca->kvm_hstate.kvm_split_mode = NULL;
4010 	kvmppc_start_thread(vcpu, vc);
4011 	kvmppc_create_dtl_entry(vcpu, vc);
4012 	trace_kvm_guest_enter(vcpu);
4013 
4014 	vc->vcore_state = VCORE_RUNNING;
4015 	trace_kvmppc_run_core(vc, 0);
4016 
4017 	if (cpu_has_feature(CPU_FTR_HVMODE))
4018 		kvmppc_radix_check_need_tlb_flush(kvm, pcpu, nested);
4019 
4020 	trace_hardirqs_on();
4021 	guest_enter_irqoff();
4022 
4023 	srcu_idx = srcu_read_lock(&kvm->srcu);
4024 
4025 	this_cpu_disable_ftrace();
4026 
4027 	trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr);
4028 	vcpu->arch.trap = trap;
4029 
4030 	this_cpu_enable_ftrace();
4031 
4032 	srcu_read_unlock(&kvm->srcu, srcu_idx);
4033 
4034 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
4035 		mtspr(SPRN_LPID, kvm->arch.host_lpid);
4036 		isync();
4037 	}
4038 
4039 	trace_hardirqs_off();
4040 	set_irq_happened(trap);
4041 
4042 	kvmppc_set_host_core(pcpu);
4043 
4044 	local_irq_enable();
4045 	guest_exit();
4046 
4047 	cpumask_clear_cpu(pcpu, &kvm->arch.cpu_in_guest);
4048 
4049 	preempt_enable();
4050 
4051 	/* cancel pending decrementer exception if DEC is now positive */
4052 	if (get_tb() < vcpu->arch.dec_expires && kvmppc_core_pending_dec(vcpu))
4053 		kvmppc_core_dequeue_dec(vcpu);
4054 
4055 	trace_kvm_guest_exit(vcpu);
4056 	r = RESUME_GUEST;
4057 	if (trap) {
4058 		if (!nested)
4059 			r = kvmppc_handle_exit_hv(kvm_run, vcpu, current);
4060 		else
4061 			r = kvmppc_handle_nested_exit(vcpu);
4062 	}
4063 	vcpu->arch.ret = r;
4064 
4065 	if (is_kvmppc_resume_guest(r) && vcpu->arch.ceded &&
4066 	    !kvmppc_vcpu_woken(vcpu)) {
4067 		kvmppc_set_timer(vcpu);
4068 		while (vcpu->arch.ceded && !kvmppc_vcpu_woken(vcpu)) {
4069 			if (signal_pending(current)) {
4070 				vcpu->stat.signal_exits++;
4071 				kvm_run->exit_reason = KVM_EXIT_INTR;
4072 				vcpu->arch.ret = -EINTR;
4073 				break;
4074 			}
4075 			spin_lock(&vc->lock);
4076 			kvmppc_vcore_blocked(vc);
4077 			spin_unlock(&vc->lock);
4078 		}
4079 	}
4080 	vcpu->arch.ceded = 0;
4081 
4082 	vc->vcore_state = VCORE_INACTIVE;
4083 	trace_kvmppc_run_core(vc, 1);
4084 
4085  done:
4086 	kvmppc_remove_runnable(vc, vcpu);
4087 	trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
4088 
4089 	return vcpu->arch.ret;
4090 
4091  sigpend:
4092 	vcpu->stat.signal_exits++;
4093 	kvm_run->exit_reason = KVM_EXIT_INTR;
4094 	vcpu->arch.ret = -EINTR;
4095  out:
4096 	local_irq_enable();
4097 	preempt_enable();
4098 	goto done;
4099 }
4100 
4101 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
4102 {
4103 	int r;
4104 	int srcu_idx;
4105 	unsigned long ebb_regs[3] = {};	/* shut up GCC */
4106 	unsigned long user_tar = 0;
4107 	unsigned int user_vrsave;
4108 	struct kvm *kvm;
4109 
4110 	if (!vcpu->arch.sane) {
4111 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4112 		return -EINVAL;
4113 	}
4114 
4115 	/*
4116 	 * Don't allow entry with a suspended transaction, because
4117 	 * the guest entry/exit code will lose it.
4118 	 * If the guest has TM enabled, save away their TM-related SPRs
4119 	 * (they will get restored by the TM unavailable interrupt).
4120 	 */
4121 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4122 	if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4123 	    (current->thread.regs->msr & MSR_TM)) {
4124 		if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4125 			run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4126 			run->fail_entry.hardware_entry_failure_reason = 0;
4127 			return -EINVAL;
4128 		}
4129 		/* Enable TM so we can read the TM SPRs */
4130 		mtmsr(mfmsr() | MSR_TM);
4131 		current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
4132 		current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
4133 		current->thread.tm_texasr = mfspr(SPRN_TEXASR);
4134 		current->thread.regs->msr &= ~MSR_TM;
4135 	}
4136 #endif
4137 
4138 	/*
4139 	 * Force online to 1 for the sake of old userspace which doesn't
4140 	 * set it.
4141 	 */
4142 	if (!vcpu->arch.online) {
4143 		atomic_inc(&vcpu->arch.vcore->online_count);
4144 		vcpu->arch.online = 1;
4145 	}
4146 
4147 	kvmppc_core_prepare_to_enter(vcpu);
4148 
4149 	/* No need to go into the guest when all we'll do is come back out */
4150 	if (signal_pending(current)) {
4151 		run->exit_reason = KVM_EXIT_INTR;
4152 		return -EINTR;
4153 	}
4154 
4155 	kvm = vcpu->kvm;
4156 	atomic_inc(&kvm->arch.vcpus_running);
4157 	/* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4158 	smp_mb();
4159 
4160 	flush_all_to_thread(current);
4161 
4162 	/* Save userspace EBB and other register values */
4163 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4164 		ebb_regs[0] = mfspr(SPRN_EBBHR);
4165 		ebb_regs[1] = mfspr(SPRN_EBBRR);
4166 		ebb_regs[2] = mfspr(SPRN_BESCR);
4167 		user_tar = mfspr(SPRN_TAR);
4168 	}
4169 	user_vrsave = mfspr(SPRN_VRSAVE);
4170 
4171 	vcpu->arch.wqp = &vcpu->arch.vcore->wq;
4172 	vcpu->arch.pgdir = current->mm->pgd;
4173 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4174 
4175 	do {
4176 		/*
4177 		 * The early POWER9 chips that can't mix radix and HPT threads
4178 		 * on the same core also need the workaround for the problem
4179 		 * where the TLB would prefetch entries in the guest exit path
4180 		 * for radix guests using the guest PIDR value and LPID 0.
4181 		 * The workaround is in the old path (kvmppc_run_vcpu())
4182 		 * but not the new path (kvmhv_run_single_vcpu()).
4183 		 */
4184 		if (kvm->arch.threads_indep && kvm_is_radix(kvm) &&
4185 		    !no_mixing_hpt_and_radix)
4186 			r = kvmhv_run_single_vcpu(run, vcpu, ~(u64)0,
4187 						  vcpu->arch.vcore->lpcr);
4188 		else
4189 			r = kvmppc_run_vcpu(run, vcpu);
4190 
4191 		if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
4192 		    !(vcpu->arch.shregs.msr & MSR_PR)) {
4193 			trace_kvm_hcall_enter(vcpu);
4194 			r = kvmppc_pseries_do_hcall(vcpu);
4195 			trace_kvm_hcall_exit(vcpu, r);
4196 			kvmppc_core_prepare_to_enter(vcpu);
4197 		} else if (r == RESUME_PAGE_FAULT) {
4198 			srcu_idx = srcu_read_lock(&kvm->srcu);
4199 			r = kvmppc_book3s_hv_page_fault(run, vcpu,
4200 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4201 			srcu_read_unlock(&kvm->srcu, srcu_idx);
4202 		} else if (r == RESUME_PASSTHROUGH) {
4203 			if (WARN_ON(xive_enabled()))
4204 				r = H_SUCCESS;
4205 			else
4206 				r = kvmppc_xics_rm_complete(vcpu, 0);
4207 		}
4208 	} while (is_kvmppc_resume_guest(r));
4209 
4210 	/* Restore userspace EBB and other register values */
4211 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4212 		mtspr(SPRN_EBBHR, ebb_regs[0]);
4213 		mtspr(SPRN_EBBRR, ebb_regs[1]);
4214 		mtspr(SPRN_BESCR, ebb_regs[2]);
4215 		mtspr(SPRN_TAR, user_tar);
4216 		mtspr(SPRN_FSCR, current->thread.fscr);
4217 	}
4218 	mtspr(SPRN_VRSAVE, user_vrsave);
4219 
4220 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4221 	atomic_dec(&kvm->arch.vcpus_running);
4222 	return r;
4223 }
4224 
4225 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4226 				     int shift, int sllp)
4227 {
4228 	(*sps)->page_shift = shift;
4229 	(*sps)->slb_enc = sllp;
4230 	(*sps)->enc[0].page_shift = shift;
4231 	(*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4232 	/*
4233 	 * Add 16MB MPSS support (may get filtered out by userspace)
4234 	 */
4235 	if (shift != 24) {
4236 		int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4237 		if (penc != -1) {
4238 			(*sps)->enc[1].page_shift = 24;
4239 			(*sps)->enc[1].pte_enc = penc;
4240 		}
4241 	}
4242 	(*sps)++;
4243 }
4244 
4245 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4246 					 struct kvm_ppc_smmu_info *info)
4247 {
4248 	struct kvm_ppc_one_seg_page_size *sps;
4249 
4250 	/*
4251 	 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4252 	 * POWER7 doesn't support keys for instruction accesses,
4253 	 * POWER8 and POWER9 do.
4254 	 */
4255 	info->data_keys = 32;
4256 	info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4257 
4258 	/* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4259 	info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4260 	info->slb_size = 32;
4261 
4262 	/* We only support these sizes for now, and no muti-size segments */
4263 	sps = &info->sps[0];
4264 	kvmppc_add_seg_page_size(&sps, 12, 0);
4265 	kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4266 	kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4267 
4268 	/* If running as a nested hypervisor, we don't support HPT guests */
4269 	if (kvmhv_on_pseries())
4270 		info->flags |= KVM_PPC_NO_HASH;
4271 
4272 	return 0;
4273 }
4274 
4275 /*
4276  * Get (and clear) the dirty memory log for a memory slot.
4277  */
4278 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4279 					 struct kvm_dirty_log *log)
4280 {
4281 	struct kvm_memslots *slots;
4282 	struct kvm_memory_slot *memslot;
4283 	int i, r;
4284 	unsigned long n;
4285 	unsigned long *buf, *p;
4286 	struct kvm_vcpu *vcpu;
4287 
4288 	mutex_lock(&kvm->slots_lock);
4289 
4290 	r = -EINVAL;
4291 	if (log->slot >= KVM_USER_MEM_SLOTS)
4292 		goto out;
4293 
4294 	slots = kvm_memslots(kvm);
4295 	memslot = id_to_memslot(slots, log->slot);
4296 	r = -ENOENT;
4297 	if (!memslot->dirty_bitmap)
4298 		goto out;
4299 
4300 	/*
4301 	 * Use second half of bitmap area because both HPT and radix
4302 	 * accumulate bits in the first half.
4303 	 */
4304 	n = kvm_dirty_bitmap_bytes(memslot);
4305 	buf = memslot->dirty_bitmap + n / sizeof(long);
4306 	memset(buf, 0, n);
4307 
4308 	if (kvm_is_radix(kvm))
4309 		r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4310 	else
4311 		r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4312 	if (r)
4313 		goto out;
4314 
4315 	/*
4316 	 * We accumulate dirty bits in the first half of the
4317 	 * memslot's dirty_bitmap area, for when pages are paged
4318 	 * out or modified by the host directly.  Pick up these
4319 	 * bits and add them to the map.
4320 	 */
4321 	p = memslot->dirty_bitmap;
4322 	for (i = 0; i < n / sizeof(long); ++i)
4323 		buf[i] |= xchg(&p[i], 0);
4324 
4325 	/* Harvest dirty bits from VPA and DTL updates */
4326 	/* Note: we never modify the SLB shadow buffer areas */
4327 	kvm_for_each_vcpu(i, vcpu, kvm) {
4328 		spin_lock(&vcpu->arch.vpa_update_lock);
4329 		kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4330 		kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4331 		spin_unlock(&vcpu->arch.vpa_update_lock);
4332 	}
4333 
4334 	r = -EFAULT;
4335 	if (copy_to_user(log->dirty_bitmap, buf, n))
4336 		goto out;
4337 
4338 	r = 0;
4339 out:
4340 	mutex_unlock(&kvm->slots_lock);
4341 	return r;
4342 }
4343 
4344 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
4345 					struct kvm_memory_slot *dont)
4346 {
4347 	if (!dont || free->arch.rmap != dont->arch.rmap) {
4348 		vfree(free->arch.rmap);
4349 		free->arch.rmap = NULL;
4350 	}
4351 }
4352 
4353 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
4354 					 unsigned long npages)
4355 {
4356 	slot->arch.rmap = vzalloc(array_size(npages, sizeof(*slot->arch.rmap)));
4357 	if (!slot->arch.rmap)
4358 		return -ENOMEM;
4359 
4360 	return 0;
4361 }
4362 
4363 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4364 					struct kvm_memory_slot *memslot,
4365 					const struct kvm_userspace_memory_region *mem)
4366 {
4367 	return 0;
4368 }
4369 
4370 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
4371 				const struct kvm_userspace_memory_region *mem,
4372 				const struct kvm_memory_slot *old,
4373 				const struct kvm_memory_slot *new)
4374 {
4375 	unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4376 
4377 	/*
4378 	 * If we are making a new memslot, it might make
4379 	 * some address that was previously cached as emulated
4380 	 * MMIO be no longer emulated MMIO, so invalidate
4381 	 * all the caches of emulated MMIO translations.
4382 	 */
4383 	if (npages)
4384 		atomic64_inc(&kvm->arch.mmio_update);
4385 }
4386 
4387 /*
4388  * Update LPCR values in kvm->arch and in vcores.
4389  * Caller must hold kvm->lock.
4390  */
4391 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
4392 {
4393 	long int i;
4394 	u32 cores_done = 0;
4395 
4396 	if ((kvm->arch.lpcr & mask) == lpcr)
4397 		return;
4398 
4399 	kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
4400 
4401 	for (i = 0; i < KVM_MAX_VCORES; ++i) {
4402 		struct kvmppc_vcore *vc = kvm->arch.vcores[i];
4403 		if (!vc)
4404 			continue;
4405 		spin_lock(&vc->lock);
4406 		vc->lpcr = (vc->lpcr & ~mask) | lpcr;
4407 		spin_unlock(&vc->lock);
4408 		if (++cores_done >= kvm->arch.online_vcores)
4409 			break;
4410 	}
4411 }
4412 
4413 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
4414 {
4415 	return;
4416 }
4417 
4418 void kvmppc_setup_partition_table(struct kvm *kvm)
4419 {
4420 	unsigned long dw0, dw1;
4421 
4422 	if (!kvm_is_radix(kvm)) {
4423 		/* PS field - page size for VRMA */
4424 		dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
4425 			((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
4426 		/* HTABSIZE and HTABORG fields */
4427 		dw0 |= kvm->arch.sdr1;
4428 
4429 		/* Second dword as set by userspace */
4430 		dw1 = kvm->arch.process_table;
4431 	} else {
4432 		dw0 = PATB_HR | radix__get_tree_size() |
4433 			__pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
4434 		dw1 = PATB_GR | kvm->arch.process_table;
4435 	}
4436 	kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
4437 }
4438 
4439 /*
4440  * Set up HPT (hashed page table) and RMA (real-mode area).
4441  * Must be called with kvm->lock held.
4442  */
4443 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
4444 {
4445 	int err = 0;
4446 	struct kvm *kvm = vcpu->kvm;
4447 	unsigned long hva;
4448 	struct kvm_memory_slot *memslot;
4449 	struct vm_area_struct *vma;
4450 	unsigned long lpcr = 0, senc;
4451 	unsigned long psize, porder;
4452 	int srcu_idx;
4453 
4454 	/* Allocate hashed page table (if not done already) and reset it */
4455 	if (!kvm->arch.hpt.virt) {
4456 		int order = KVM_DEFAULT_HPT_ORDER;
4457 		struct kvm_hpt_info info;
4458 
4459 		err = kvmppc_allocate_hpt(&info, order);
4460 		/* If we get here, it means userspace didn't specify a
4461 		 * size explicitly.  So, try successively smaller
4462 		 * sizes if the default failed. */
4463 		while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
4464 			err  = kvmppc_allocate_hpt(&info, order);
4465 
4466 		if (err < 0) {
4467 			pr_err("KVM: Couldn't alloc HPT\n");
4468 			goto out;
4469 		}
4470 
4471 		kvmppc_set_hpt(kvm, &info);
4472 	}
4473 
4474 	/* Look up the memslot for guest physical address 0 */
4475 	srcu_idx = srcu_read_lock(&kvm->srcu);
4476 	memslot = gfn_to_memslot(kvm, 0);
4477 
4478 	/* We must have some memory at 0 by now */
4479 	err = -EINVAL;
4480 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
4481 		goto out_srcu;
4482 
4483 	/* Look up the VMA for the start of this memory slot */
4484 	hva = memslot->userspace_addr;
4485 	down_read(&current->mm->mmap_sem);
4486 	vma = find_vma(current->mm, hva);
4487 	if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
4488 		goto up_out;
4489 
4490 	psize = vma_kernel_pagesize(vma);
4491 
4492 	up_read(&current->mm->mmap_sem);
4493 
4494 	/* We can handle 4k, 64k or 16M pages in the VRMA */
4495 	if (psize >= 0x1000000)
4496 		psize = 0x1000000;
4497 	else if (psize >= 0x10000)
4498 		psize = 0x10000;
4499 	else
4500 		psize = 0x1000;
4501 	porder = __ilog2(psize);
4502 
4503 	senc = slb_pgsize_encoding(psize);
4504 	kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
4505 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
4506 	/* Create HPTEs in the hash page table for the VRMA */
4507 	kvmppc_map_vrma(vcpu, memslot, porder);
4508 
4509 	/* Update VRMASD field in the LPCR */
4510 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
4511 		/* the -4 is to account for senc values starting at 0x10 */
4512 		lpcr = senc << (LPCR_VRMASD_SH - 4);
4513 		kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
4514 	}
4515 
4516 	/* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
4517 	smp_wmb();
4518 	err = 0;
4519  out_srcu:
4520 	srcu_read_unlock(&kvm->srcu, srcu_idx);
4521  out:
4522 	return err;
4523 
4524  up_out:
4525 	up_read(&current->mm->mmap_sem);
4526 	goto out_srcu;
4527 }
4528 
4529 /* Must be called with kvm->lock held and mmu_ready = 0 and no vcpus running */
4530 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
4531 {
4532 	if (nesting_enabled(kvm))
4533 		kvmhv_release_all_nested(kvm);
4534 	kvmppc_free_radix(kvm);
4535 	kvmppc_update_lpcr(kvm, LPCR_VPM1,
4536 			   LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4537 	kvmppc_rmap_reset(kvm);
4538 	kvm->arch.radix = 0;
4539 	kvm->arch.process_table = 0;
4540 	return 0;
4541 }
4542 
4543 /* Must be called with kvm->lock held and mmu_ready = 0 and no vcpus running */
4544 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
4545 {
4546 	int err;
4547 
4548 	err = kvmppc_init_vm_radix(kvm);
4549 	if (err)
4550 		return err;
4551 
4552 	kvmppc_free_hpt(&kvm->arch.hpt);
4553 	kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
4554 			   LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4555 	kvmppc_rmap_reset(kvm);
4556 	kvm->arch.radix = 1;
4557 	return 0;
4558 }
4559 
4560 #ifdef CONFIG_KVM_XICS
4561 /*
4562  * Allocate a per-core structure for managing state about which cores are
4563  * running in the host versus the guest and for exchanging data between
4564  * real mode KVM and CPU running in the host.
4565  * This is only done for the first VM.
4566  * The allocated structure stays even if all VMs have stopped.
4567  * It is only freed when the kvm-hv module is unloaded.
4568  * It's OK for this routine to fail, we just don't support host
4569  * core operations like redirecting H_IPI wakeups.
4570  */
4571 void kvmppc_alloc_host_rm_ops(void)
4572 {
4573 	struct kvmppc_host_rm_ops *ops;
4574 	unsigned long l_ops;
4575 	int cpu, core;
4576 	int size;
4577 
4578 	/* Not the first time here ? */
4579 	if (kvmppc_host_rm_ops_hv != NULL)
4580 		return;
4581 
4582 	ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
4583 	if (!ops)
4584 		return;
4585 
4586 	size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
4587 	ops->rm_core = kzalloc(size, GFP_KERNEL);
4588 
4589 	if (!ops->rm_core) {
4590 		kfree(ops);
4591 		return;
4592 	}
4593 
4594 	cpus_read_lock();
4595 
4596 	for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
4597 		if (!cpu_online(cpu))
4598 			continue;
4599 
4600 		core = cpu >> threads_shift;
4601 		ops->rm_core[core].rm_state.in_host = 1;
4602 	}
4603 
4604 	ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
4605 
4606 	/*
4607 	 * Make the contents of the kvmppc_host_rm_ops structure visible
4608 	 * to other CPUs before we assign it to the global variable.
4609 	 * Do an atomic assignment (no locks used here), but if someone
4610 	 * beats us to it, just free our copy and return.
4611 	 */
4612 	smp_wmb();
4613 	l_ops = (unsigned long) ops;
4614 
4615 	if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
4616 		cpus_read_unlock();
4617 		kfree(ops->rm_core);
4618 		kfree(ops);
4619 		return;
4620 	}
4621 
4622 	cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
4623 					     "ppc/kvm_book3s:prepare",
4624 					     kvmppc_set_host_core,
4625 					     kvmppc_clear_host_core);
4626 	cpus_read_unlock();
4627 }
4628 
4629 void kvmppc_free_host_rm_ops(void)
4630 {
4631 	if (kvmppc_host_rm_ops_hv) {
4632 		cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
4633 		kfree(kvmppc_host_rm_ops_hv->rm_core);
4634 		kfree(kvmppc_host_rm_ops_hv);
4635 		kvmppc_host_rm_ops_hv = NULL;
4636 	}
4637 }
4638 #endif
4639 
4640 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
4641 {
4642 	unsigned long lpcr, lpid;
4643 	char buf[32];
4644 	int ret;
4645 
4646 	/* Allocate the guest's logical partition ID */
4647 
4648 	lpid = kvmppc_alloc_lpid();
4649 	if ((long)lpid < 0)
4650 		return -ENOMEM;
4651 	kvm->arch.lpid = lpid;
4652 
4653 	kvmppc_alloc_host_rm_ops();
4654 
4655 	kvmhv_vm_nested_init(kvm);
4656 
4657 	/*
4658 	 * Since we don't flush the TLB when tearing down a VM,
4659 	 * and this lpid might have previously been used,
4660 	 * make sure we flush on each core before running the new VM.
4661 	 * On POWER9, the tlbie in mmu_partition_table_set_entry()
4662 	 * does this flush for us.
4663 	 */
4664 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
4665 		cpumask_setall(&kvm->arch.need_tlb_flush);
4666 
4667 	/* Start out with the default set of hcalls enabled */
4668 	memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
4669 	       sizeof(kvm->arch.enabled_hcalls));
4670 
4671 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
4672 		kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
4673 
4674 	/* Init LPCR for virtual RMA mode */
4675 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
4676 		kvm->arch.host_lpid = mfspr(SPRN_LPID);
4677 		kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
4678 		lpcr &= LPCR_PECE | LPCR_LPES;
4679 	} else {
4680 		lpcr = 0;
4681 	}
4682 	lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
4683 		LPCR_VPM0 | LPCR_VPM1;
4684 	kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
4685 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
4686 	/* On POWER8 turn on online bit to enable PURR/SPURR */
4687 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
4688 		lpcr |= LPCR_ONL;
4689 	/*
4690 	 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
4691 	 * Set HVICE bit to enable hypervisor virtualization interrupts.
4692 	 * Set HEIC to prevent OS interrupts to go to hypervisor (should
4693 	 * be unnecessary but better safe than sorry in case we re-enable
4694 	 * EE in HV mode with this LPCR still set)
4695 	 */
4696 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4697 		lpcr &= ~LPCR_VPM0;
4698 		lpcr |= LPCR_HVICE | LPCR_HEIC;
4699 
4700 		/*
4701 		 * If xive is enabled, we route 0x500 interrupts directly
4702 		 * to the guest.
4703 		 */
4704 		if (xive_enabled())
4705 			lpcr |= LPCR_LPES;
4706 	}
4707 
4708 	/*
4709 	 * If the host uses radix, the guest starts out as radix.
4710 	 */
4711 	if (radix_enabled()) {
4712 		kvm->arch.radix = 1;
4713 		kvm->arch.mmu_ready = 1;
4714 		lpcr &= ~LPCR_VPM1;
4715 		lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
4716 		ret = kvmppc_init_vm_radix(kvm);
4717 		if (ret) {
4718 			kvmppc_free_lpid(kvm->arch.lpid);
4719 			return ret;
4720 		}
4721 		kvmppc_setup_partition_table(kvm);
4722 	}
4723 
4724 	kvm->arch.lpcr = lpcr;
4725 
4726 	/* Initialization for future HPT resizes */
4727 	kvm->arch.resize_hpt = NULL;
4728 
4729 	/*
4730 	 * Work out how many sets the TLB has, for the use of
4731 	 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
4732 	 */
4733 	if (radix_enabled())
4734 		kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;	/* 128 */
4735 	else if (cpu_has_feature(CPU_FTR_ARCH_300))
4736 		kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;	/* 256 */
4737 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
4738 		kvm->arch.tlb_sets = POWER8_TLB_SETS;		/* 512 */
4739 	else
4740 		kvm->arch.tlb_sets = POWER7_TLB_SETS;		/* 128 */
4741 
4742 	/*
4743 	 * Track that we now have a HV mode VM active. This blocks secondary
4744 	 * CPU threads from coming online.
4745 	 * On POWER9, we only need to do this if the "indep_threads_mode"
4746 	 * module parameter has been set to N.
4747 	 */
4748 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4749 		if (!indep_threads_mode && !cpu_has_feature(CPU_FTR_HVMODE)) {
4750 			pr_warn("KVM: Ignoring indep_threads_mode=N in nested hypervisor\n");
4751 			kvm->arch.threads_indep = true;
4752 		} else {
4753 			kvm->arch.threads_indep = indep_threads_mode;
4754 		}
4755 	}
4756 	if (!kvm->arch.threads_indep)
4757 		kvm_hv_vm_activated();
4758 
4759 	/*
4760 	 * Initialize smt_mode depending on processor.
4761 	 * POWER8 and earlier have to use "strict" threading, where
4762 	 * all vCPUs in a vcore have to run on the same (sub)core,
4763 	 * whereas on POWER9 the threads can each run a different
4764 	 * guest.
4765 	 */
4766 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
4767 		kvm->arch.smt_mode = threads_per_subcore;
4768 	else
4769 		kvm->arch.smt_mode = 1;
4770 	kvm->arch.emul_smt_mode = 1;
4771 
4772 	/*
4773 	 * Create a debugfs directory for the VM
4774 	 */
4775 	snprintf(buf, sizeof(buf), "vm%d", current->pid);
4776 	kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
4777 	kvmppc_mmu_debugfs_init(kvm);
4778 	if (radix_enabled())
4779 		kvmhv_radix_debugfs_init(kvm);
4780 
4781 	return 0;
4782 }
4783 
4784 static void kvmppc_free_vcores(struct kvm *kvm)
4785 {
4786 	long int i;
4787 
4788 	for (i = 0; i < KVM_MAX_VCORES; ++i)
4789 		kfree(kvm->arch.vcores[i]);
4790 	kvm->arch.online_vcores = 0;
4791 }
4792 
4793 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
4794 {
4795 	debugfs_remove_recursive(kvm->arch.debugfs_dir);
4796 
4797 	if (!kvm->arch.threads_indep)
4798 		kvm_hv_vm_deactivated();
4799 
4800 	kvmppc_free_vcores(kvm);
4801 
4802 
4803 	if (kvm_is_radix(kvm))
4804 		kvmppc_free_radix(kvm);
4805 	else
4806 		kvmppc_free_hpt(&kvm->arch.hpt);
4807 
4808 	/* Perform global invalidation and return lpid to the pool */
4809 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4810 		if (nesting_enabled(kvm))
4811 			kvmhv_release_all_nested(kvm);
4812 		kvm->arch.process_table = 0;
4813 		kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
4814 	}
4815 	kvmppc_free_lpid(kvm->arch.lpid);
4816 
4817 	kvmppc_free_pimap(kvm);
4818 }
4819 
4820 /* We don't need to emulate any privileged instructions or dcbz */
4821 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
4822 				     unsigned int inst, int *advance)
4823 {
4824 	return EMULATE_FAIL;
4825 }
4826 
4827 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
4828 					ulong spr_val)
4829 {
4830 	return EMULATE_FAIL;
4831 }
4832 
4833 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
4834 					ulong *spr_val)
4835 {
4836 	return EMULATE_FAIL;
4837 }
4838 
4839 static int kvmppc_core_check_processor_compat_hv(void)
4840 {
4841 	if (cpu_has_feature(CPU_FTR_HVMODE) &&
4842 	    cpu_has_feature(CPU_FTR_ARCH_206))
4843 		return 0;
4844 
4845 	/* POWER9 in radix mode is capable of being a nested hypervisor. */
4846 	if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
4847 		return 0;
4848 
4849 	return -EIO;
4850 }
4851 
4852 #ifdef CONFIG_KVM_XICS
4853 
4854 void kvmppc_free_pimap(struct kvm *kvm)
4855 {
4856 	kfree(kvm->arch.pimap);
4857 }
4858 
4859 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
4860 {
4861 	return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
4862 }
4863 
4864 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4865 {
4866 	struct irq_desc *desc;
4867 	struct kvmppc_irq_map *irq_map;
4868 	struct kvmppc_passthru_irqmap *pimap;
4869 	struct irq_chip *chip;
4870 	int i, rc = 0;
4871 
4872 	if (!kvm_irq_bypass)
4873 		return 1;
4874 
4875 	desc = irq_to_desc(host_irq);
4876 	if (!desc)
4877 		return -EIO;
4878 
4879 	mutex_lock(&kvm->lock);
4880 
4881 	pimap = kvm->arch.pimap;
4882 	if (pimap == NULL) {
4883 		/* First call, allocate structure to hold IRQ map */
4884 		pimap = kvmppc_alloc_pimap();
4885 		if (pimap == NULL) {
4886 			mutex_unlock(&kvm->lock);
4887 			return -ENOMEM;
4888 		}
4889 		kvm->arch.pimap = pimap;
4890 	}
4891 
4892 	/*
4893 	 * For now, we only support interrupts for which the EOI operation
4894 	 * is an OPAL call followed by a write to XIRR, since that's
4895 	 * what our real-mode EOI code does, or a XIVE interrupt
4896 	 */
4897 	chip = irq_data_get_irq_chip(&desc->irq_data);
4898 	if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
4899 		pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
4900 			host_irq, guest_gsi);
4901 		mutex_unlock(&kvm->lock);
4902 		return -ENOENT;
4903 	}
4904 
4905 	/*
4906 	 * See if we already have an entry for this guest IRQ number.
4907 	 * If it's mapped to a hardware IRQ number, that's an error,
4908 	 * otherwise re-use this entry.
4909 	 */
4910 	for (i = 0; i < pimap->n_mapped; i++) {
4911 		if (guest_gsi == pimap->mapped[i].v_hwirq) {
4912 			if (pimap->mapped[i].r_hwirq) {
4913 				mutex_unlock(&kvm->lock);
4914 				return -EINVAL;
4915 			}
4916 			break;
4917 		}
4918 	}
4919 
4920 	if (i == KVMPPC_PIRQ_MAPPED) {
4921 		mutex_unlock(&kvm->lock);
4922 		return -EAGAIN;		/* table is full */
4923 	}
4924 
4925 	irq_map = &pimap->mapped[i];
4926 
4927 	irq_map->v_hwirq = guest_gsi;
4928 	irq_map->desc = desc;
4929 
4930 	/*
4931 	 * Order the above two stores before the next to serialize with
4932 	 * the KVM real mode handler.
4933 	 */
4934 	smp_wmb();
4935 	irq_map->r_hwirq = desc->irq_data.hwirq;
4936 
4937 	if (i == pimap->n_mapped)
4938 		pimap->n_mapped++;
4939 
4940 	if (xive_enabled())
4941 		rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
4942 	else
4943 		kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
4944 	if (rc)
4945 		irq_map->r_hwirq = 0;
4946 
4947 	mutex_unlock(&kvm->lock);
4948 
4949 	return 0;
4950 }
4951 
4952 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4953 {
4954 	struct irq_desc *desc;
4955 	struct kvmppc_passthru_irqmap *pimap;
4956 	int i, rc = 0;
4957 
4958 	if (!kvm_irq_bypass)
4959 		return 0;
4960 
4961 	desc = irq_to_desc(host_irq);
4962 	if (!desc)
4963 		return -EIO;
4964 
4965 	mutex_lock(&kvm->lock);
4966 	if (!kvm->arch.pimap)
4967 		goto unlock;
4968 
4969 	pimap = kvm->arch.pimap;
4970 
4971 	for (i = 0; i < pimap->n_mapped; i++) {
4972 		if (guest_gsi == pimap->mapped[i].v_hwirq)
4973 			break;
4974 	}
4975 
4976 	if (i == pimap->n_mapped) {
4977 		mutex_unlock(&kvm->lock);
4978 		return -ENODEV;
4979 	}
4980 
4981 	if (xive_enabled())
4982 		rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
4983 	else
4984 		kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
4985 
4986 	/* invalidate the entry (what do do on error from the above ?) */
4987 	pimap->mapped[i].r_hwirq = 0;
4988 
4989 	/*
4990 	 * We don't free this structure even when the count goes to
4991 	 * zero. The structure is freed when we destroy the VM.
4992 	 */
4993  unlock:
4994 	mutex_unlock(&kvm->lock);
4995 	return rc;
4996 }
4997 
4998 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
4999 					     struct irq_bypass_producer *prod)
5000 {
5001 	int ret = 0;
5002 	struct kvm_kernel_irqfd *irqfd =
5003 		container_of(cons, struct kvm_kernel_irqfd, consumer);
5004 
5005 	irqfd->producer = prod;
5006 
5007 	ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5008 	if (ret)
5009 		pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5010 			prod->irq, irqfd->gsi, ret);
5011 
5012 	return ret;
5013 }
5014 
5015 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5016 					      struct irq_bypass_producer *prod)
5017 {
5018 	int ret;
5019 	struct kvm_kernel_irqfd *irqfd =
5020 		container_of(cons, struct kvm_kernel_irqfd, consumer);
5021 
5022 	irqfd->producer = NULL;
5023 
5024 	/*
5025 	 * When producer of consumer is unregistered, we change back to
5026 	 * default external interrupt handling mode - KVM real mode
5027 	 * will switch back to host.
5028 	 */
5029 	ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5030 	if (ret)
5031 		pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5032 			prod->irq, irqfd->gsi, ret);
5033 }
5034 #endif
5035 
5036 static long kvm_arch_vm_ioctl_hv(struct file *filp,
5037 				 unsigned int ioctl, unsigned long arg)
5038 {
5039 	struct kvm *kvm __maybe_unused = filp->private_data;
5040 	void __user *argp = (void __user *)arg;
5041 	long r;
5042 
5043 	switch (ioctl) {
5044 
5045 	case KVM_PPC_ALLOCATE_HTAB: {
5046 		u32 htab_order;
5047 
5048 		r = -EFAULT;
5049 		if (get_user(htab_order, (u32 __user *)argp))
5050 			break;
5051 		r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5052 		if (r)
5053 			break;
5054 		r = 0;
5055 		break;
5056 	}
5057 
5058 	case KVM_PPC_GET_HTAB_FD: {
5059 		struct kvm_get_htab_fd ghf;
5060 
5061 		r = -EFAULT;
5062 		if (copy_from_user(&ghf, argp, sizeof(ghf)))
5063 			break;
5064 		r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5065 		break;
5066 	}
5067 
5068 	case KVM_PPC_RESIZE_HPT_PREPARE: {
5069 		struct kvm_ppc_resize_hpt rhpt;
5070 
5071 		r = -EFAULT;
5072 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5073 			break;
5074 
5075 		r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5076 		break;
5077 	}
5078 
5079 	case KVM_PPC_RESIZE_HPT_COMMIT: {
5080 		struct kvm_ppc_resize_hpt rhpt;
5081 
5082 		r = -EFAULT;
5083 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5084 			break;
5085 
5086 		r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5087 		break;
5088 	}
5089 
5090 	default:
5091 		r = -ENOTTY;
5092 	}
5093 
5094 	return r;
5095 }
5096 
5097 /*
5098  * List of hcall numbers to enable by default.
5099  * For compatibility with old userspace, we enable by default
5100  * all hcalls that were implemented before the hcall-enabling
5101  * facility was added.  Note this list should not include H_RTAS.
5102  */
5103 static unsigned int default_hcall_list[] = {
5104 	H_REMOVE,
5105 	H_ENTER,
5106 	H_READ,
5107 	H_PROTECT,
5108 	H_BULK_REMOVE,
5109 	H_GET_TCE,
5110 	H_PUT_TCE,
5111 	H_SET_DABR,
5112 	H_SET_XDABR,
5113 	H_CEDE,
5114 	H_PROD,
5115 	H_CONFER,
5116 	H_REGISTER_VPA,
5117 #ifdef CONFIG_KVM_XICS
5118 	H_EOI,
5119 	H_CPPR,
5120 	H_IPI,
5121 	H_IPOLL,
5122 	H_XIRR,
5123 	H_XIRR_X,
5124 #endif
5125 	0
5126 };
5127 
5128 static void init_default_hcalls(void)
5129 {
5130 	int i;
5131 	unsigned int hcall;
5132 
5133 	for (i = 0; default_hcall_list[i]; ++i) {
5134 		hcall = default_hcall_list[i];
5135 		WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5136 		__set_bit(hcall / 4, default_enabled_hcalls);
5137 	}
5138 }
5139 
5140 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5141 {
5142 	unsigned long lpcr;
5143 	int radix;
5144 	int err;
5145 
5146 	/* If not on a POWER9, reject it */
5147 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5148 		return -ENODEV;
5149 
5150 	/* If any unknown flags set, reject it */
5151 	if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5152 		return -EINVAL;
5153 
5154 	/* GR (guest radix) bit in process_table field must match */
5155 	radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5156 	if (!!(cfg->process_table & PATB_GR) != radix)
5157 		return -EINVAL;
5158 
5159 	/* Process table size field must be reasonable, i.e. <= 24 */
5160 	if ((cfg->process_table & PRTS_MASK) > 24)
5161 		return -EINVAL;
5162 
5163 	/* We can change a guest to/from radix now, if the host is radix */
5164 	if (radix && !radix_enabled())
5165 		return -EINVAL;
5166 
5167 	/* If we're a nested hypervisor, we currently only support radix */
5168 	if (kvmhv_on_pseries() && !radix)
5169 		return -EINVAL;
5170 
5171 	mutex_lock(&kvm->lock);
5172 	if (radix != kvm_is_radix(kvm)) {
5173 		if (kvm->arch.mmu_ready) {
5174 			kvm->arch.mmu_ready = 0;
5175 			/* order mmu_ready vs. vcpus_running */
5176 			smp_mb();
5177 			if (atomic_read(&kvm->arch.vcpus_running)) {
5178 				kvm->arch.mmu_ready = 1;
5179 				err = -EBUSY;
5180 				goto out_unlock;
5181 			}
5182 		}
5183 		if (radix)
5184 			err = kvmppc_switch_mmu_to_radix(kvm);
5185 		else
5186 			err = kvmppc_switch_mmu_to_hpt(kvm);
5187 		if (err)
5188 			goto out_unlock;
5189 	}
5190 
5191 	kvm->arch.process_table = cfg->process_table;
5192 	kvmppc_setup_partition_table(kvm);
5193 
5194 	lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5195 	kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5196 	err = 0;
5197 
5198  out_unlock:
5199 	mutex_unlock(&kvm->lock);
5200 	return err;
5201 }
5202 
5203 static int kvmhv_enable_nested(struct kvm *kvm)
5204 {
5205 	if (!nested)
5206 		return -EPERM;
5207 	if (!cpu_has_feature(CPU_FTR_ARCH_300) || no_mixing_hpt_and_radix)
5208 		return -ENODEV;
5209 
5210 	/* kvm == NULL means the caller is testing if the capability exists */
5211 	if (kvm)
5212 		kvm->arch.nested_enable = true;
5213 	return 0;
5214 }
5215 
5216 static struct kvmppc_ops kvm_ops_hv = {
5217 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
5218 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
5219 	.get_one_reg = kvmppc_get_one_reg_hv,
5220 	.set_one_reg = kvmppc_set_one_reg_hv,
5221 	.vcpu_load   = kvmppc_core_vcpu_load_hv,
5222 	.vcpu_put    = kvmppc_core_vcpu_put_hv,
5223 	.set_msr     = kvmppc_set_msr_hv,
5224 	.vcpu_run    = kvmppc_vcpu_run_hv,
5225 	.vcpu_create = kvmppc_core_vcpu_create_hv,
5226 	.vcpu_free   = kvmppc_core_vcpu_free_hv,
5227 	.check_requests = kvmppc_core_check_requests_hv,
5228 	.get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
5229 	.flush_memslot  = kvmppc_core_flush_memslot_hv,
5230 	.prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
5231 	.commit_memory_region  = kvmppc_core_commit_memory_region_hv,
5232 	.unmap_hva_range = kvm_unmap_hva_range_hv,
5233 	.age_hva  = kvm_age_hva_hv,
5234 	.test_age_hva = kvm_test_age_hva_hv,
5235 	.set_spte_hva = kvm_set_spte_hva_hv,
5236 	.mmu_destroy  = kvmppc_mmu_destroy_hv,
5237 	.free_memslot = kvmppc_core_free_memslot_hv,
5238 	.create_memslot = kvmppc_core_create_memslot_hv,
5239 	.init_vm =  kvmppc_core_init_vm_hv,
5240 	.destroy_vm = kvmppc_core_destroy_vm_hv,
5241 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
5242 	.emulate_op = kvmppc_core_emulate_op_hv,
5243 	.emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
5244 	.emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
5245 	.fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
5246 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
5247 	.hcall_implemented = kvmppc_hcall_impl_hv,
5248 #ifdef CONFIG_KVM_XICS
5249 	.irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
5250 	.irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
5251 #endif
5252 	.configure_mmu = kvmhv_configure_mmu,
5253 	.get_rmmu_info = kvmhv_get_rmmu_info,
5254 	.set_smt_mode = kvmhv_set_smt_mode,
5255 	.enable_nested = kvmhv_enable_nested,
5256 };
5257 
5258 static int kvm_init_subcore_bitmap(void)
5259 {
5260 	int i, j;
5261 	int nr_cores = cpu_nr_cores();
5262 	struct sibling_subcore_state *sibling_subcore_state;
5263 
5264 	for (i = 0; i < nr_cores; i++) {
5265 		int first_cpu = i * threads_per_core;
5266 		int node = cpu_to_node(first_cpu);
5267 
5268 		/* Ignore if it is already allocated. */
5269 		if (paca_ptrs[first_cpu]->sibling_subcore_state)
5270 			continue;
5271 
5272 		sibling_subcore_state =
5273 			kmalloc_node(sizeof(struct sibling_subcore_state),
5274 							GFP_KERNEL, node);
5275 		if (!sibling_subcore_state)
5276 			return -ENOMEM;
5277 
5278 		memset(sibling_subcore_state, 0,
5279 				sizeof(struct sibling_subcore_state));
5280 
5281 		for (j = 0; j < threads_per_core; j++) {
5282 			int cpu = first_cpu + j;
5283 
5284 			paca_ptrs[cpu]->sibling_subcore_state =
5285 						sibling_subcore_state;
5286 		}
5287 	}
5288 	return 0;
5289 }
5290 
5291 static int kvmppc_radix_possible(void)
5292 {
5293 	return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
5294 }
5295 
5296 static int kvmppc_book3s_init_hv(void)
5297 {
5298 	int r;
5299 	/*
5300 	 * FIXME!! Do we need to check on all cpus ?
5301 	 */
5302 	r = kvmppc_core_check_processor_compat_hv();
5303 	if (r < 0)
5304 		return -ENODEV;
5305 
5306 	r = kvmhv_nested_init();
5307 	if (r)
5308 		return r;
5309 
5310 	r = kvm_init_subcore_bitmap();
5311 	if (r)
5312 		return r;
5313 
5314 	/*
5315 	 * We need a way of accessing the XICS interrupt controller,
5316 	 * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
5317 	 * indirectly, via OPAL.
5318 	 */
5319 #ifdef CONFIG_SMP
5320 	if (!xive_enabled() && !kvmhv_on_pseries() &&
5321 	    !local_paca->kvm_hstate.xics_phys) {
5322 		struct device_node *np;
5323 
5324 		np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
5325 		if (!np) {
5326 			pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
5327 			return -ENODEV;
5328 		}
5329 		/* presence of intc confirmed - node can be dropped again */
5330 		of_node_put(np);
5331 	}
5332 #endif
5333 
5334 	kvm_ops_hv.owner = THIS_MODULE;
5335 	kvmppc_hv_ops = &kvm_ops_hv;
5336 
5337 	init_default_hcalls();
5338 
5339 	init_vcore_lists();
5340 
5341 	r = kvmppc_mmu_hv_init();
5342 	if (r)
5343 		return r;
5344 
5345 	if (kvmppc_radix_possible())
5346 		r = kvmppc_radix_init();
5347 
5348 	/*
5349 	 * POWER9 chips before version 2.02 can't have some threads in
5350 	 * HPT mode and some in radix mode on the same core.
5351 	 */
5352 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5353 		unsigned int pvr = mfspr(SPRN_PVR);
5354 		if ((pvr >> 16) == PVR_POWER9 &&
5355 		    (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
5356 		     ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
5357 			no_mixing_hpt_and_radix = true;
5358 	}
5359 
5360 	return r;
5361 }
5362 
5363 static void kvmppc_book3s_exit_hv(void)
5364 {
5365 	kvmppc_free_host_rm_ops();
5366 	if (kvmppc_radix_possible())
5367 		kvmppc_radix_exit();
5368 	kvmppc_hv_ops = NULL;
5369 	kvmhv_nested_exit();
5370 }
5371 
5372 module_init(kvmppc_book3s_init_hv);
5373 module_exit(kvmppc_book3s_exit_hv);
5374 MODULE_LICENSE("GPL");
5375 MODULE_ALIAS_MISCDEV(KVM_MINOR);
5376 MODULE_ALIAS("devname:kvm");
5377