xref: /openbmc/linux/arch/powerpc/kvm/book3s_hv.c (revision ac8b6f14)
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 			vcpu->arch.hcall_needed = 0;
987 			return -EINTR;
988 		}
989 		break;
990 	case H_TLB_INVALIDATE:
991 		ret = H_FUNCTION;
992 		if (nesting_enabled(vcpu->kvm))
993 			ret = kvmhv_do_nested_tlbie(vcpu);
994 		break;
995 
996 	default:
997 		return RESUME_HOST;
998 	}
999 	kvmppc_set_gpr(vcpu, 3, ret);
1000 	vcpu->arch.hcall_needed = 0;
1001 	return RESUME_GUEST;
1002 }
1003 
1004 /*
1005  * Handle H_CEDE in the nested virtualization case where we haven't
1006  * called the real-mode hcall handlers in book3s_hv_rmhandlers.S.
1007  * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1008  * that the cede logic in kvmppc_run_single_vcpu() works properly.
1009  */
1010 static void kvmppc_nested_cede(struct kvm_vcpu *vcpu)
1011 {
1012 	vcpu->arch.shregs.msr |= MSR_EE;
1013 	vcpu->arch.ceded = 1;
1014 	smp_mb();
1015 	if (vcpu->arch.prodded) {
1016 		vcpu->arch.prodded = 0;
1017 		smp_mb();
1018 		vcpu->arch.ceded = 0;
1019 	}
1020 }
1021 
1022 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1023 {
1024 	switch (cmd) {
1025 	case H_CEDE:
1026 	case H_PROD:
1027 	case H_CONFER:
1028 	case H_REGISTER_VPA:
1029 	case H_SET_MODE:
1030 	case H_LOGICAL_CI_LOAD:
1031 	case H_LOGICAL_CI_STORE:
1032 #ifdef CONFIG_KVM_XICS
1033 	case H_XIRR:
1034 	case H_CPPR:
1035 	case H_EOI:
1036 	case H_IPI:
1037 	case H_IPOLL:
1038 	case H_XIRR_X:
1039 #endif
1040 		return 1;
1041 	}
1042 
1043 	/* See if it's in the real-mode table */
1044 	return kvmppc_hcall_impl_hv_realmode(cmd);
1045 }
1046 
1047 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
1048 					struct kvm_vcpu *vcpu)
1049 {
1050 	u32 last_inst;
1051 
1052 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1053 					EMULATE_DONE) {
1054 		/*
1055 		 * Fetch failed, so return to guest and
1056 		 * try executing it again.
1057 		 */
1058 		return RESUME_GUEST;
1059 	}
1060 
1061 	if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
1062 		run->exit_reason = KVM_EXIT_DEBUG;
1063 		run->debug.arch.address = kvmppc_get_pc(vcpu);
1064 		return RESUME_HOST;
1065 	} else {
1066 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1067 		return RESUME_GUEST;
1068 	}
1069 }
1070 
1071 static void do_nothing(void *x)
1072 {
1073 }
1074 
1075 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1076 {
1077 	int thr, cpu, pcpu, nthreads;
1078 	struct kvm_vcpu *v;
1079 	unsigned long dpdes;
1080 
1081 	nthreads = vcpu->kvm->arch.emul_smt_mode;
1082 	dpdes = 0;
1083 	cpu = vcpu->vcpu_id & ~(nthreads - 1);
1084 	for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1085 		v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1086 		if (!v)
1087 			continue;
1088 		/*
1089 		 * If the vcpu is currently running on a physical cpu thread,
1090 		 * interrupt it in order to pull it out of the guest briefly,
1091 		 * which will update its vcore->dpdes value.
1092 		 */
1093 		pcpu = READ_ONCE(v->cpu);
1094 		if (pcpu >= 0)
1095 			smp_call_function_single(pcpu, do_nothing, NULL, 1);
1096 		if (kvmppc_doorbell_pending(v))
1097 			dpdes |= 1 << thr;
1098 	}
1099 	return dpdes;
1100 }
1101 
1102 /*
1103  * On POWER9, emulate doorbell-related instructions in order to
1104  * give the guest the illusion of running on a multi-threaded core.
1105  * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1106  * and mfspr DPDES.
1107  */
1108 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1109 {
1110 	u32 inst, rb, thr;
1111 	unsigned long arg;
1112 	struct kvm *kvm = vcpu->kvm;
1113 	struct kvm_vcpu *tvcpu;
1114 
1115 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1116 		return RESUME_GUEST;
1117 	if (get_op(inst) != 31)
1118 		return EMULATE_FAIL;
1119 	rb = get_rb(inst);
1120 	thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1121 	switch (get_xop(inst)) {
1122 	case OP_31_XOP_MSGSNDP:
1123 		arg = kvmppc_get_gpr(vcpu, rb);
1124 		if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1125 			break;
1126 		arg &= 0x3f;
1127 		if (arg >= kvm->arch.emul_smt_mode)
1128 			break;
1129 		tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1130 		if (!tvcpu)
1131 			break;
1132 		if (!tvcpu->arch.doorbell_request) {
1133 			tvcpu->arch.doorbell_request = 1;
1134 			kvmppc_fast_vcpu_kick_hv(tvcpu);
1135 		}
1136 		break;
1137 	case OP_31_XOP_MSGCLRP:
1138 		arg = kvmppc_get_gpr(vcpu, rb);
1139 		if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1140 			break;
1141 		vcpu->arch.vcore->dpdes = 0;
1142 		vcpu->arch.doorbell_request = 0;
1143 		break;
1144 	case OP_31_XOP_MFSPR:
1145 		switch (get_sprn(inst)) {
1146 		case SPRN_TIR:
1147 			arg = thr;
1148 			break;
1149 		case SPRN_DPDES:
1150 			arg = kvmppc_read_dpdes(vcpu);
1151 			break;
1152 		default:
1153 			return EMULATE_FAIL;
1154 		}
1155 		kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1156 		break;
1157 	default:
1158 		return EMULATE_FAIL;
1159 	}
1160 	kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1161 	return RESUME_GUEST;
1162 }
1163 
1164 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
1165 				 struct task_struct *tsk)
1166 {
1167 	int r = RESUME_HOST;
1168 
1169 	vcpu->stat.sum_exits++;
1170 
1171 	/*
1172 	 * This can happen if an interrupt occurs in the last stages
1173 	 * of guest entry or the first stages of guest exit (i.e. after
1174 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1175 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1176 	 * That can happen due to a bug, or due to a machine check
1177 	 * occurring at just the wrong time.
1178 	 */
1179 	if (vcpu->arch.shregs.msr & MSR_HV) {
1180 		printk(KERN_EMERG "KVM trap in HV mode!\n");
1181 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1182 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1183 			vcpu->arch.shregs.msr);
1184 		kvmppc_dump_regs(vcpu);
1185 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1186 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1187 		return RESUME_HOST;
1188 	}
1189 	run->exit_reason = KVM_EXIT_UNKNOWN;
1190 	run->ready_for_interrupt_injection = 1;
1191 	switch (vcpu->arch.trap) {
1192 	/* We're good on these - the host merely wanted to get our attention */
1193 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1194 		vcpu->stat.dec_exits++;
1195 		r = RESUME_GUEST;
1196 		break;
1197 	case BOOK3S_INTERRUPT_EXTERNAL:
1198 	case BOOK3S_INTERRUPT_H_DOORBELL:
1199 	case BOOK3S_INTERRUPT_H_VIRT:
1200 		vcpu->stat.ext_intr_exits++;
1201 		r = RESUME_GUEST;
1202 		break;
1203 	/* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1204 	case BOOK3S_INTERRUPT_HMI:
1205 	case BOOK3S_INTERRUPT_PERFMON:
1206 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
1207 		r = RESUME_GUEST;
1208 		break;
1209 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
1210 		/* Exit to guest with KVM_EXIT_NMI as exit reason */
1211 		run->exit_reason = KVM_EXIT_NMI;
1212 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1213 		/* Clear out the old NMI status from run->flags */
1214 		run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1215 		/* Now set the NMI status */
1216 		if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1217 			run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1218 		else
1219 			run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1220 
1221 		r = RESUME_HOST;
1222 		/* Print the MCE event to host console. */
1223 		machine_check_print_event_info(&vcpu->arch.mce_evt, false);
1224 		break;
1225 	case BOOK3S_INTERRUPT_PROGRAM:
1226 	{
1227 		ulong flags;
1228 		/*
1229 		 * Normally program interrupts are delivered directly
1230 		 * to the guest by the hardware, but we can get here
1231 		 * as a result of a hypervisor emulation interrupt
1232 		 * (e40) getting turned into a 700 by BML RTAS.
1233 		 */
1234 		flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1235 		kvmppc_core_queue_program(vcpu, flags);
1236 		r = RESUME_GUEST;
1237 		break;
1238 	}
1239 	case BOOK3S_INTERRUPT_SYSCALL:
1240 	{
1241 		/* hcall - punt to userspace */
1242 		int i;
1243 
1244 		/* hypercall with MSR_PR has already been handled in rmode,
1245 		 * and never reaches here.
1246 		 */
1247 
1248 		run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1249 		for (i = 0; i < 9; ++i)
1250 			run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1251 		run->exit_reason = KVM_EXIT_PAPR_HCALL;
1252 		vcpu->arch.hcall_needed = 1;
1253 		r = RESUME_HOST;
1254 		break;
1255 	}
1256 	/*
1257 	 * We get these next two if the guest accesses a page which it thinks
1258 	 * it has mapped but which is not actually present, either because
1259 	 * it is for an emulated I/O device or because the corresonding
1260 	 * host page has been paged out.  Any other HDSI/HISI interrupts
1261 	 * have been handled already.
1262 	 */
1263 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1264 		r = RESUME_PAGE_FAULT;
1265 		break;
1266 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
1267 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1268 		vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr &
1269 			DSISR_SRR1_MATCH_64S;
1270 		if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1271 			vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1272 		r = RESUME_PAGE_FAULT;
1273 		break;
1274 	/*
1275 	 * This occurs if the guest executes an illegal instruction.
1276 	 * If the guest debug is disabled, generate a program interrupt
1277 	 * to the guest. If guest debug is enabled, we need to check
1278 	 * whether the instruction is a software breakpoint instruction.
1279 	 * Accordingly return to Guest or Host.
1280 	 */
1281 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1282 		if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1283 			vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1284 				swab32(vcpu->arch.emul_inst) :
1285 				vcpu->arch.emul_inst;
1286 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1287 			r = kvmppc_emulate_debug_inst(run, vcpu);
1288 		} else {
1289 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1290 			r = RESUME_GUEST;
1291 		}
1292 		break;
1293 	/*
1294 	 * This occurs if the guest (kernel or userspace), does something that
1295 	 * is prohibited by HFSCR.
1296 	 * On POWER9, this could be a doorbell instruction that we need
1297 	 * to emulate.
1298 	 * Otherwise, we just generate a program interrupt to the guest.
1299 	 */
1300 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1301 		r = EMULATE_FAIL;
1302 		if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1303 		    cpu_has_feature(CPU_FTR_ARCH_300))
1304 			r = kvmppc_emulate_doorbell_instr(vcpu);
1305 		if (r == EMULATE_FAIL) {
1306 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1307 			r = RESUME_GUEST;
1308 		}
1309 		break;
1310 
1311 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1312 	case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1313 		/*
1314 		 * This occurs for various TM-related instructions that
1315 		 * we need to emulate on POWER9 DD2.2.  We have already
1316 		 * handled the cases where the guest was in real-suspend
1317 		 * mode and was transitioning to transactional state.
1318 		 */
1319 		r = kvmhv_p9_tm_emulation(vcpu);
1320 		break;
1321 #endif
1322 
1323 	case BOOK3S_INTERRUPT_HV_RM_HARD:
1324 		r = RESUME_PASSTHROUGH;
1325 		break;
1326 	default:
1327 		kvmppc_dump_regs(vcpu);
1328 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1329 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1330 			vcpu->arch.shregs.msr);
1331 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1332 		r = RESUME_HOST;
1333 		break;
1334 	}
1335 
1336 	return r;
1337 }
1338 
1339 static int kvmppc_handle_nested_exit(struct kvm_vcpu *vcpu)
1340 {
1341 	int r;
1342 	int srcu_idx;
1343 
1344 	vcpu->stat.sum_exits++;
1345 
1346 	/*
1347 	 * This can happen if an interrupt occurs in the last stages
1348 	 * of guest entry or the first stages of guest exit (i.e. after
1349 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1350 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1351 	 * That can happen due to a bug, or due to a machine check
1352 	 * occurring at just the wrong time.
1353 	 */
1354 	if (vcpu->arch.shregs.msr & MSR_HV) {
1355 		pr_emerg("KVM trap in HV mode while nested!\n");
1356 		pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1357 			 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1358 			 vcpu->arch.shregs.msr);
1359 		kvmppc_dump_regs(vcpu);
1360 		return RESUME_HOST;
1361 	}
1362 	switch (vcpu->arch.trap) {
1363 	/* We're good on these - the host merely wanted to get our attention */
1364 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1365 		vcpu->stat.dec_exits++;
1366 		r = RESUME_GUEST;
1367 		break;
1368 	case BOOK3S_INTERRUPT_EXTERNAL:
1369 		vcpu->stat.ext_intr_exits++;
1370 		r = RESUME_HOST;
1371 		break;
1372 	case BOOK3S_INTERRUPT_H_DOORBELL:
1373 	case BOOK3S_INTERRUPT_H_VIRT:
1374 		vcpu->stat.ext_intr_exits++;
1375 		r = RESUME_GUEST;
1376 		break;
1377 	/* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1378 	case BOOK3S_INTERRUPT_HMI:
1379 	case BOOK3S_INTERRUPT_PERFMON:
1380 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
1381 		r = RESUME_GUEST;
1382 		break;
1383 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
1384 		/* Pass the machine check to the L1 guest */
1385 		r = RESUME_HOST;
1386 		/* Print the MCE event to host console. */
1387 		machine_check_print_event_info(&vcpu->arch.mce_evt, false);
1388 		break;
1389 	/*
1390 	 * We get these next two if the guest accesses a page which it thinks
1391 	 * it has mapped but which is not actually present, either because
1392 	 * it is for an emulated I/O device or because the corresonding
1393 	 * host page has been paged out.
1394 	 */
1395 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1396 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1397 		r = kvmhv_nested_page_fault(vcpu);
1398 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1399 		break;
1400 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
1401 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1402 		vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1403 					 DSISR_SRR1_MATCH_64S;
1404 		if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1405 			vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1406 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1407 		r = kvmhv_nested_page_fault(vcpu);
1408 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1409 		break;
1410 
1411 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1412 	case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1413 		/*
1414 		 * This occurs for various TM-related instructions that
1415 		 * we need to emulate on POWER9 DD2.2.  We have already
1416 		 * handled the cases where the guest was in real-suspend
1417 		 * mode and was transitioning to transactional state.
1418 		 */
1419 		r = kvmhv_p9_tm_emulation(vcpu);
1420 		break;
1421 #endif
1422 
1423 	case BOOK3S_INTERRUPT_HV_RM_HARD:
1424 		vcpu->arch.trap = 0;
1425 		r = RESUME_GUEST;
1426 		if (!xive_enabled())
1427 			kvmppc_xics_rm_complete(vcpu, 0);
1428 		break;
1429 	default:
1430 		r = RESUME_HOST;
1431 		break;
1432 	}
1433 
1434 	return r;
1435 }
1436 
1437 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1438 					    struct kvm_sregs *sregs)
1439 {
1440 	int i;
1441 
1442 	memset(sregs, 0, sizeof(struct kvm_sregs));
1443 	sregs->pvr = vcpu->arch.pvr;
1444 	for (i = 0; i < vcpu->arch.slb_max; i++) {
1445 		sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1446 		sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1447 	}
1448 
1449 	return 0;
1450 }
1451 
1452 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1453 					    struct kvm_sregs *sregs)
1454 {
1455 	int i, j;
1456 
1457 	/* Only accept the same PVR as the host's, since we can't spoof it */
1458 	if (sregs->pvr != vcpu->arch.pvr)
1459 		return -EINVAL;
1460 
1461 	j = 0;
1462 	for (i = 0; i < vcpu->arch.slb_nr; i++) {
1463 		if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1464 			vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1465 			vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1466 			++j;
1467 		}
1468 	}
1469 	vcpu->arch.slb_max = j;
1470 
1471 	return 0;
1472 }
1473 
1474 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1475 		bool preserve_top32)
1476 {
1477 	struct kvm *kvm = vcpu->kvm;
1478 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
1479 	u64 mask;
1480 
1481 	mutex_lock(&kvm->lock);
1482 	spin_lock(&vc->lock);
1483 	/*
1484 	 * If ILE (interrupt little-endian) has changed, update the
1485 	 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1486 	 */
1487 	if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1488 		struct kvm_vcpu *vcpu;
1489 		int i;
1490 
1491 		kvm_for_each_vcpu(i, vcpu, kvm) {
1492 			if (vcpu->arch.vcore != vc)
1493 				continue;
1494 			if (new_lpcr & LPCR_ILE)
1495 				vcpu->arch.intr_msr |= MSR_LE;
1496 			else
1497 				vcpu->arch.intr_msr &= ~MSR_LE;
1498 		}
1499 	}
1500 
1501 	/*
1502 	 * Userspace can only modify DPFD (default prefetch depth),
1503 	 * ILE (interrupt little-endian) and TC (translation control).
1504 	 * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1505 	 */
1506 	mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1507 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
1508 		mask |= LPCR_AIL;
1509 	/*
1510 	 * On POWER9, allow userspace to enable large decrementer for the
1511 	 * guest, whether or not the host has it enabled.
1512 	 */
1513 	if (cpu_has_feature(CPU_FTR_ARCH_300))
1514 		mask |= LPCR_LD;
1515 
1516 	/* Broken 32-bit version of LPCR must not clear top bits */
1517 	if (preserve_top32)
1518 		mask &= 0xFFFFFFFF;
1519 	vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1520 	spin_unlock(&vc->lock);
1521 	mutex_unlock(&kvm->lock);
1522 }
1523 
1524 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1525 				 union kvmppc_one_reg *val)
1526 {
1527 	int r = 0;
1528 	long int i;
1529 
1530 	switch (id) {
1531 	case KVM_REG_PPC_DEBUG_INST:
1532 		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1533 		break;
1534 	case KVM_REG_PPC_HIOR:
1535 		*val = get_reg_val(id, 0);
1536 		break;
1537 	case KVM_REG_PPC_DABR:
1538 		*val = get_reg_val(id, vcpu->arch.dabr);
1539 		break;
1540 	case KVM_REG_PPC_DABRX:
1541 		*val = get_reg_val(id, vcpu->arch.dabrx);
1542 		break;
1543 	case KVM_REG_PPC_DSCR:
1544 		*val = get_reg_val(id, vcpu->arch.dscr);
1545 		break;
1546 	case KVM_REG_PPC_PURR:
1547 		*val = get_reg_val(id, vcpu->arch.purr);
1548 		break;
1549 	case KVM_REG_PPC_SPURR:
1550 		*val = get_reg_val(id, vcpu->arch.spurr);
1551 		break;
1552 	case KVM_REG_PPC_AMR:
1553 		*val = get_reg_val(id, vcpu->arch.amr);
1554 		break;
1555 	case KVM_REG_PPC_UAMOR:
1556 		*val = get_reg_val(id, vcpu->arch.uamor);
1557 		break;
1558 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1559 		i = id - KVM_REG_PPC_MMCR0;
1560 		*val = get_reg_val(id, vcpu->arch.mmcr[i]);
1561 		break;
1562 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1563 		i = id - KVM_REG_PPC_PMC1;
1564 		*val = get_reg_val(id, vcpu->arch.pmc[i]);
1565 		break;
1566 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1567 		i = id - KVM_REG_PPC_SPMC1;
1568 		*val = get_reg_val(id, vcpu->arch.spmc[i]);
1569 		break;
1570 	case KVM_REG_PPC_SIAR:
1571 		*val = get_reg_val(id, vcpu->arch.siar);
1572 		break;
1573 	case KVM_REG_PPC_SDAR:
1574 		*val = get_reg_val(id, vcpu->arch.sdar);
1575 		break;
1576 	case KVM_REG_PPC_SIER:
1577 		*val = get_reg_val(id, vcpu->arch.sier);
1578 		break;
1579 	case KVM_REG_PPC_IAMR:
1580 		*val = get_reg_val(id, vcpu->arch.iamr);
1581 		break;
1582 	case KVM_REG_PPC_PSPB:
1583 		*val = get_reg_val(id, vcpu->arch.pspb);
1584 		break;
1585 	case KVM_REG_PPC_DPDES:
1586 		*val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1587 		break;
1588 	case KVM_REG_PPC_VTB:
1589 		*val = get_reg_val(id, vcpu->arch.vcore->vtb);
1590 		break;
1591 	case KVM_REG_PPC_DAWR:
1592 		*val = get_reg_val(id, vcpu->arch.dawr);
1593 		break;
1594 	case KVM_REG_PPC_DAWRX:
1595 		*val = get_reg_val(id, vcpu->arch.dawrx);
1596 		break;
1597 	case KVM_REG_PPC_CIABR:
1598 		*val = get_reg_val(id, vcpu->arch.ciabr);
1599 		break;
1600 	case KVM_REG_PPC_CSIGR:
1601 		*val = get_reg_val(id, vcpu->arch.csigr);
1602 		break;
1603 	case KVM_REG_PPC_TACR:
1604 		*val = get_reg_val(id, vcpu->arch.tacr);
1605 		break;
1606 	case KVM_REG_PPC_TCSCR:
1607 		*val = get_reg_val(id, vcpu->arch.tcscr);
1608 		break;
1609 	case KVM_REG_PPC_PID:
1610 		*val = get_reg_val(id, vcpu->arch.pid);
1611 		break;
1612 	case KVM_REG_PPC_ACOP:
1613 		*val = get_reg_val(id, vcpu->arch.acop);
1614 		break;
1615 	case KVM_REG_PPC_WORT:
1616 		*val = get_reg_val(id, vcpu->arch.wort);
1617 		break;
1618 	case KVM_REG_PPC_TIDR:
1619 		*val = get_reg_val(id, vcpu->arch.tid);
1620 		break;
1621 	case KVM_REG_PPC_PSSCR:
1622 		*val = get_reg_val(id, vcpu->arch.psscr);
1623 		break;
1624 	case KVM_REG_PPC_VPA_ADDR:
1625 		spin_lock(&vcpu->arch.vpa_update_lock);
1626 		*val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1627 		spin_unlock(&vcpu->arch.vpa_update_lock);
1628 		break;
1629 	case KVM_REG_PPC_VPA_SLB:
1630 		spin_lock(&vcpu->arch.vpa_update_lock);
1631 		val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1632 		val->vpaval.length = vcpu->arch.slb_shadow.len;
1633 		spin_unlock(&vcpu->arch.vpa_update_lock);
1634 		break;
1635 	case KVM_REG_PPC_VPA_DTL:
1636 		spin_lock(&vcpu->arch.vpa_update_lock);
1637 		val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1638 		val->vpaval.length = vcpu->arch.dtl.len;
1639 		spin_unlock(&vcpu->arch.vpa_update_lock);
1640 		break;
1641 	case KVM_REG_PPC_TB_OFFSET:
1642 		*val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1643 		break;
1644 	case KVM_REG_PPC_LPCR:
1645 	case KVM_REG_PPC_LPCR_64:
1646 		*val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1647 		break;
1648 	case KVM_REG_PPC_PPR:
1649 		*val = get_reg_val(id, vcpu->arch.ppr);
1650 		break;
1651 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1652 	case KVM_REG_PPC_TFHAR:
1653 		*val = get_reg_val(id, vcpu->arch.tfhar);
1654 		break;
1655 	case KVM_REG_PPC_TFIAR:
1656 		*val = get_reg_val(id, vcpu->arch.tfiar);
1657 		break;
1658 	case KVM_REG_PPC_TEXASR:
1659 		*val = get_reg_val(id, vcpu->arch.texasr);
1660 		break;
1661 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1662 		i = id - KVM_REG_PPC_TM_GPR0;
1663 		*val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1664 		break;
1665 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1666 	{
1667 		int j;
1668 		i = id - KVM_REG_PPC_TM_VSR0;
1669 		if (i < 32)
1670 			for (j = 0; j < TS_FPRWIDTH; j++)
1671 				val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1672 		else {
1673 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1674 				val->vval = vcpu->arch.vr_tm.vr[i-32];
1675 			else
1676 				r = -ENXIO;
1677 		}
1678 		break;
1679 	}
1680 	case KVM_REG_PPC_TM_CR:
1681 		*val = get_reg_val(id, vcpu->arch.cr_tm);
1682 		break;
1683 	case KVM_REG_PPC_TM_XER:
1684 		*val = get_reg_val(id, vcpu->arch.xer_tm);
1685 		break;
1686 	case KVM_REG_PPC_TM_LR:
1687 		*val = get_reg_val(id, vcpu->arch.lr_tm);
1688 		break;
1689 	case KVM_REG_PPC_TM_CTR:
1690 		*val = get_reg_val(id, vcpu->arch.ctr_tm);
1691 		break;
1692 	case KVM_REG_PPC_TM_FPSCR:
1693 		*val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1694 		break;
1695 	case KVM_REG_PPC_TM_AMR:
1696 		*val = get_reg_val(id, vcpu->arch.amr_tm);
1697 		break;
1698 	case KVM_REG_PPC_TM_PPR:
1699 		*val = get_reg_val(id, vcpu->arch.ppr_tm);
1700 		break;
1701 	case KVM_REG_PPC_TM_VRSAVE:
1702 		*val = get_reg_val(id, vcpu->arch.vrsave_tm);
1703 		break;
1704 	case KVM_REG_PPC_TM_VSCR:
1705 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1706 			*val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1707 		else
1708 			r = -ENXIO;
1709 		break;
1710 	case KVM_REG_PPC_TM_DSCR:
1711 		*val = get_reg_val(id, vcpu->arch.dscr_tm);
1712 		break;
1713 	case KVM_REG_PPC_TM_TAR:
1714 		*val = get_reg_val(id, vcpu->arch.tar_tm);
1715 		break;
1716 #endif
1717 	case KVM_REG_PPC_ARCH_COMPAT:
1718 		*val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1719 		break;
1720 	case KVM_REG_PPC_DEC_EXPIRY:
1721 		*val = get_reg_val(id, vcpu->arch.dec_expires +
1722 				   vcpu->arch.vcore->tb_offset);
1723 		break;
1724 	case KVM_REG_PPC_ONLINE:
1725 		*val = get_reg_val(id, vcpu->arch.online);
1726 		break;
1727 	case KVM_REG_PPC_PTCR:
1728 		*val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
1729 		break;
1730 	default:
1731 		r = -EINVAL;
1732 		break;
1733 	}
1734 
1735 	return r;
1736 }
1737 
1738 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1739 				 union kvmppc_one_reg *val)
1740 {
1741 	int r = 0;
1742 	long int i;
1743 	unsigned long addr, len;
1744 
1745 	switch (id) {
1746 	case KVM_REG_PPC_HIOR:
1747 		/* Only allow this to be set to zero */
1748 		if (set_reg_val(id, *val))
1749 			r = -EINVAL;
1750 		break;
1751 	case KVM_REG_PPC_DABR:
1752 		vcpu->arch.dabr = set_reg_val(id, *val);
1753 		break;
1754 	case KVM_REG_PPC_DABRX:
1755 		vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1756 		break;
1757 	case KVM_REG_PPC_DSCR:
1758 		vcpu->arch.dscr = set_reg_val(id, *val);
1759 		break;
1760 	case KVM_REG_PPC_PURR:
1761 		vcpu->arch.purr = set_reg_val(id, *val);
1762 		break;
1763 	case KVM_REG_PPC_SPURR:
1764 		vcpu->arch.spurr = set_reg_val(id, *val);
1765 		break;
1766 	case KVM_REG_PPC_AMR:
1767 		vcpu->arch.amr = set_reg_val(id, *val);
1768 		break;
1769 	case KVM_REG_PPC_UAMOR:
1770 		vcpu->arch.uamor = set_reg_val(id, *val);
1771 		break;
1772 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1773 		i = id - KVM_REG_PPC_MMCR0;
1774 		vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1775 		break;
1776 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1777 		i = id - KVM_REG_PPC_PMC1;
1778 		vcpu->arch.pmc[i] = set_reg_val(id, *val);
1779 		break;
1780 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1781 		i = id - KVM_REG_PPC_SPMC1;
1782 		vcpu->arch.spmc[i] = set_reg_val(id, *val);
1783 		break;
1784 	case KVM_REG_PPC_SIAR:
1785 		vcpu->arch.siar = set_reg_val(id, *val);
1786 		break;
1787 	case KVM_REG_PPC_SDAR:
1788 		vcpu->arch.sdar = set_reg_val(id, *val);
1789 		break;
1790 	case KVM_REG_PPC_SIER:
1791 		vcpu->arch.sier = set_reg_val(id, *val);
1792 		break;
1793 	case KVM_REG_PPC_IAMR:
1794 		vcpu->arch.iamr = set_reg_val(id, *val);
1795 		break;
1796 	case KVM_REG_PPC_PSPB:
1797 		vcpu->arch.pspb = set_reg_val(id, *val);
1798 		break;
1799 	case KVM_REG_PPC_DPDES:
1800 		vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1801 		break;
1802 	case KVM_REG_PPC_VTB:
1803 		vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1804 		break;
1805 	case KVM_REG_PPC_DAWR:
1806 		vcpu->arch.dawr = set_reg_val(id, *val);
1807 		break;
1808 	case KVM_REG_PPC_DAWRX:
1809 		vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1810 		break;
1811 	case KVM_REG_PPC_CIABR:
1812 		vcpu->arch.ciabr = set_reg_val(id, *val);
1813 		/* Don't allow setting breakpoints in hypervisor code */
1814 		if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1815 			vcpu->arch.ciabr &= ~CIABR_PRIV;	/* disable */
1816 		break;
1817 	case KVM_REG_PPC_CSIGR:
1818 		vcpu->arch.csigr = set_reg_val(id, *val);
1819 		break;
1820 	case KVM_REG_PPC_TACR:
1821 		vcpu->arch.tacr = set_reg_val(id, *val);
1822 		break;
1823 	case KVM_REG_PPC_TCSCR:
1824 		vcpu->arch.tcscr = set_reg_val(id, *val);
1825 		break;
1826 	case KVM_REG_PPC_PID:
1827 		vcpu->arch.pid = set_reg_val(id, *val);
1828 		break;
1829 	case KVM_REG_PPC_ACOP:
1830 		vcpu->arch.acop = set_reg_val(id, *val);
1831 		break;
1832 	case KVM_REG_PPC_WORT:
1833 		vcpu->arch.wort = set_reg_val(id, *val);
1834 		break;
1835 	case KVM_REG_PPC_TIDR:
1836 		vcpu->arch.tid = set_reg_val(id, *val);
1837 		break;
1838 	case KVM_REG_PPC_PSSCR:
1839 		vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1840 		break;
1841 	case KVM_REG_PPC_VPA_ADDR:
1842 		addr = set_reg_val(id, *val);
1843 		r = -EINVAL;
1844 		if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1845 			      vcpu->arch.dtl.next_gpa))
1846 			break;
1847 		r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1848 		break;
1849 	case KVM_REG_PPC_VPA_SLB:
1850 		addr = val->vpaval.addr;
1851 		len = val->vpaval.length;
1852 		r = -EINVAL;
1853 		if (addr && !vcpu->arch.vpa.next_gpa)
1854 			break;
1855 		r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1856 		break;
1857 	case KVM_REG_PPC_VPA_DTL:
1858 		addr = val->vpaval.addr;
1859 		len = val->vpaval.length;
1860 		r = -EINVAL;
1861 		if (addr && (len < sizeof(struct dtl_entry) ||
1862 			     !vcpu->arch.vpa.next_gpa))
1863 			break;
1864 		len -= len % sizeof(struct dtl_entry);
1865 		r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1866 		break;
1867 	case KVM_REG_PPC_TB_OFFSET:
1868 		/* round up to multiple of 2^24 */
1869 		vcpu->arch.vcore->tb_offset =
1870 			ALIGN(set_reg_val(id, *val), 1UL << 24);
1871 		break;
1872 	case KVM_REG_PPC_LPCR:
1873 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1874 		break;
1875 	case KVM_REG_PPC_LPCR_64:
1876 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1877 		break;
1878 	case KVM_REG_PPC_PPR:
1879 		vcpu->arch.ppr = set_reg_val(id, *val);
1880 		break;
1881 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1882 	case KVM_REG_PPC_TFHAR:
1883 		vcpu->arch.tfhar = set_reg_val(id, *val);
1884 		break;
1885 	case KVM_REG_PPC_TFIAR:
1886 		vcpu->arch.tfiar = set_reg_val(id, *val);
1887 		break;
1888 	case KVM_REG_PPC_TEXASR:
1889 		vcpu->arch.texasr = set_reg_val(id, *val);
1890 		break;
1891 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1892 		i = id - KVM_REG_PPC_TM_GPR0;
1893 		vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1894 		break;
1895 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1896 	{
1897 		int j;
1898 		i = id - KVM_REG_PPC_TM_VSR0;
1899 		if (i < 32)
1900 			for (j = 0; j < TS_FPRWIDTH; j++)
1901 				vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1902 		else
1903 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1904 				vcpu->arch.vr_tm.vr[i-32] = val->vval;
1905 			else
1906 				r = -ENXIO;
1907 		break;
1908 	}
1909 	case KVM_REG_PPC_TM_CR:
1910 		vcpu->arch.cr_tm = set_reg_val(id, *val);
1911 		break;
1912 	case KVM_REG_PPC_TM_XER:
1913 		vcpu->arch.xer_tm = set_reg_val(id, *val);
1914 		break;
1915 	case KVM_REG_PPC_TM_LR:
1916 		vcpu->arch.lr_tm = set_reg_val(id, *val);
1917 		break;
1918 	case KVM_REG_PPC_TM_CTR:
1919 		vcpu->arch.ctr_tm = set_reg_val(id, *val);
1920 		break;
1921 	case KVM_REG_PPC_TM_FPSCR:
1922 		vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1923 		break;
1924 	case KVM_REG_PPC_TM_AMR:
1925 		vcpu->arch.amr_tm = set_reg_val(id, *val);
1926 		break;
1927 	case KVM_REG_PPC_TM_PPR:
1928 		vcpu->arch.ppr_tm = set_reg_val(id, *val);
1929 		break;
1930 	case KVM_REG_PPC_TM_VRSAVE:
1931 		vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1932 		break;
1933 	case KVM_REG_PPC_TM_VSCR:
1934 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1935 			vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1936 		else
1937 			r = - ENXIO;
1938 		break;
1939 	case KVM_REG_PPC_TM_DSCR:
1940 		vcpu->arch.dscr_tm = set_reg_val(id, *val);
1941 		break;
1942 	case KVM_REG_PPC_TM_TAR:
1943 		vcpu->arch.tar_tm = set_reg_val(id, *val);
1944 		break;
1945 #endif
1946 	case KVM_REG_PPC_ARCH_COMPAT:
1947 		r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1948 		break;
1949 	case KVM_REG_PPC_DEC_EXPIRY:
1950 		vcpu->arch.dec_expires = set_reg_val(id, *val) -
1951 			vcpu->arch.vcore->tb_offset;
1952 		break;
1953 	case KVM_REG_PPC_ONLINE:
1954 		i = set_reg_val(id, *val);
1955 		if (i && !vcpu->arch.online)
1956 			atomic_inc(&vcpu->arch.vcore->online_count);
1957 		else if (!i && vcpu->arch.online)
1958 			atomic_dec(&vcpu->arch.vcore->online_count);
1959 		vcpu->arch.online = i;
1960 		break;
1961 	case KVM_REG_PPC_PTCR:
1962 		vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
1963 		break;
1964 	default:
1965 		r = -EINVAL;
1966 		break;
1967 	}
1968 
1969 	return r;
1970 }
1971 
1972 /*
1973  * On POWER9, threads are independent and can be in different partitions.
1974  * Therefore we consider each thread to be a subcore.
1975  * There is a restriction that all threads have to be in the same
1976  * MMU mode (radix or HPT), unfortunately, but since we only support
1977  * HPT guests on a HPT host so far, that isn't an impediment yet.
1978  */
1979 static int threads_per_vcore(struct kvm *kvm)
1980 {
1981 	if (kvm->arch.threads_indep)
1982 		return 1;
1983 	return threads_per_subcore;
1984 }
1985 
1986 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
1987 {
1988 	struct kvmppc_vcore *vcore;
1989 
1990 	vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1991 
1992 	if (vcore == NULL)
1993 		return NULL;
1994 
1995 	spin_lock_init(&vcore->lock);
1996 	spin_lock_init(&vcore->stoltb_lock);
1997 	init_swait_queue_head(&vcore->wq);
1998 	vcore->preempt_tb = TB_NIL;
1999 	vcore->lpcr = kvm->arch.lpcr;
2000 	vcore->first_vcpuid = id;
2001 	vcore->kvm = kvm;
2002 	INIT_LIST_HEAD(&vcore->preempt_list);
2003 
2004 	return vcore;
2005 }
2006 
2007 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2008 static struct debugfs_timings_element {
2009 	const char *name;
2010 	size_t offset;
2011 } timings[] = {
2012 	{"rm_entry",	offsetof(struct kvm_vcpu, arch.rm_entry)},
2013 	{"rm_intr",	offsetof(struct kvm_vcpu, arch.rm_intr)},
2014 	{"rm_exit",	offsetof(struct kvm_vcpu, arch.rm_exit)},
2015 	{"guest",	offsetof(struct kvm_vcpu, arch.guest_time)},
2016 	{"cede",	offsetof(struct kvm_vcpu, arch.cede_time)},
2017 };
2018 
2019 #define N_TIMINGS	(ARRAY_SIZE(timings))
2020 
2021 struct debugfs_timings_state {
2022 	struct kvm_vcpu	*vcpu;
2023 	unsigned int	buflen;
2024 	char		buf[N_TIMINGS * 100];
2025 };
2026 
2027 static int debugfs_timings_open(struct inode *inode, struct file *file)
2028 {
2029 	struct kvm_vcpu *vcpu = inode->i_private;
2030 	struct debugfs_timings_state *p;
2031 
2032 	p = kzalloc(sizeof(*p), GFP_KERNEL);
2033 	if (!p)
2034 		return -ENOMEM;
2035 
2036 	kvm_get_kvm(vcpu->kvm);
2037 	p->vcpu = vcpu;
2038 	file->private_data = p;
2039 
2040 	return nonseekable_open(inode, file);
2041 }
2042 
2043 static int debugfs_timings_release(struct inode *inode, struct file *file)
2044 {
2045 	struct debugfs_timings_state *p = file->private_data;
2046 
2047 	kvm_put_kvm(p->vcpu->kvm);
2048 	kfree(p);
2049 	return 0;
2050 }
2051 
2052 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2053 				    size_t len, loff_t *ppos)
2054 {
2055 	struct debugfs_timings_state *p = file->private_data;
2056 	struct kvm_vcpu *vcpu = p->vcpu;
2057 	char *s, *buf_end;
2058 	struct kvmhv_tb_accumulator tb;
2059 	u64 count;
2060 	loff_t pos;
2061 	ssize_t n;
2062 	int i, loops;
2063 	bool ok;
2064 
2065 	if (!p->buflen) {
2066 		s = p->buf;
2067 		buf_end = s + sizeof(p->buf);
2068 		for (i = 0; i < N_TIMINGS; ++i) {
2069 			struct kvmhv_tb_accumulator *acc;
2070 
2071 			acc = (struct kvmhv_tb_accumulator *)
2072 				((unsigned long)vcpu + timings[i].offset);
2073 			ok = false;
2074 			for (loops = 0; loops < 1000; ++loops) {
2075 				count = acc->seqcount;
2076 				if (!(count & 1)) {
2077 					smp_rmb();
2078 					tb = *acc;
2079 					smp_rmb();
2080 					if (count == acc->seqcount) {
2081 						ok = true;
2082 						break;
2083 					}
2084 				}
2085 				udelay(1);
2086 			}
2087 			if (!ok)
2088 				snprintf(s, buf_end - s, "%s: stuck\n",
2089 					timings[i].name);
2090 			else
2091 				snprintf(s, buf_end - s,
2092 					"%s: %llu %llu %llu %llu\n",
2093 					timings[i].name, count / 2,
2094 					tb_to_ns(tb.tb_total),
2095 					tb_to_ns(tb.tb_min),
2096 					tb_to_ns(tb.tb_max));
2097 			s += strlen(s);
2098 		}
2099 		p->buflen = s - p->buf;
2100 	}
2101 
2102 	pos = *ppos;
2103 	if (pos >= p->buflen)
2104 		return 0;
2105 	if (len > p->buflen - pos)
2106 		len = p->buflen - pos;
2107 	n = copy_to_user(buf, p->buf + pos, len);
2108 	if (n) {
2109 		if (n == len)
2110 			return -EFAULT;
2111 		len -= n;
2112 	}
2113 	*ppos = pos + len;
2114 	return len;
2115 }
2116 
2117 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2118 				     size_t len, loff_t *ppos)
2119 {
2120 	return -EACCES;
2121 }
2122 
2123 static const struct file_operations debugfs_timings_ops = {
2124 	.owner	 = THIS_MODULE,
2125 	.open	 = debugfs_timings_open,
2126 	.release = debugfs_timings_release,
2127 	.read	 = debugfs_timings_read,
2128 	.write	 = debugfs_timings_write,
2129 	.llseek	 = generic_file_llseek,
2130 };
2131 
2132 /* Create a debugfs directory for the vcpu */
2133 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2134 {
2135 	char buf[16];
2136 	struct kvm *kvm = vcpu->kvm;
2137 
2138 	snprintf(buf, sizeof(buf), "vcpu%u", id);
2139 	if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
2140 		return;
2141 	vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
2142 	if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
2143 		return;
2144 	vcpu->arch.debugfs_timings =
2145 		debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
2146 				    vcpu, &debugfs_timings_ops);
2147 }
2148 
2149 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2150 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2151 {
2152 }
2153 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2154 
2155 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
2156 						   unsigned int id)
2157 {
2158 	struct kvm_vcpu *vcpu;
2159 	int err;
2160 	int core;
2161 	struct kvmppc_vcore *vcore;
2162 
2163 	err = -ENOMEM;
2164 	vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2165 	if (!vcpu)
2166 		goto out;
2167 
2168 	err = kvm_vcpu_init(vcpu, kvm, id);
2169 	if (err)
2170 		goto free_vcpu;
2171 
2172 	vcpu->arch.shared = &vcpu->arch.shregs;
2173 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2174 	/*
2175 	 * The shared struct is never shared on HV,
2176 	 * so we can always use host endianness
2177 	 */
2178 #ifdef __BIG_ENDIAN__
2179 	vcpu->arch.shared_big_endian = true;
2180 #else
2181 	vcpu->arch.shared_big_endian = false;
2182 #endif
2183 #endif
2184 	vcpu->arch.mmcr[0] = MMCR0_FC;
2185 	vcpu->arch.ctrl = CTRL_RUNLATCH;
2186 	/* default to host PVR, since we can't spoof it */
2187 	kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2188 	spin_lock_init(&vcpu->arch.vpa_update_lock);
2189 	spin_lock_init(&vcpu->arch.tbacct_lock);
2190 	vcpu->arch.busy_preempt = TB_NIL;
2191 	vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2192 
2193 	/*
2194 	 * Set the default HFSCR for the guest from the host value.
2195 	 * This value is only used on POWER9.
2196 	 * On POWER9, we want to virtualize the doorbell facility, so we
2197 	 * don't set the HFSCR_MSGP bit, and that causes those instructions
2198 	 * to trap and then we emulate them.
2199 	 */
2200 	vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2201 		HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP;
2202 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
2203 		vcpu->arch.hfscr &= mfspr(SPRN_HFSCR);
2204 		if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2205 			vcpu->arch.hfscr |= HFSCR_TM;
2206 	}
2207 	if (cpu_has_feature(CPU_FTR_TM_COMP))
2208 		vcpu->arch.hfscr |= HFSCR_TM;
2209 
2210 	kvmppc_mmu_book3s_hv_init(vcpu);
2211 
2212 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2213 
2214 	init_waitqueue_head(&vcpu->arch.cpu_run);
2215 
2216 	mutex_lock(&kvm->lock);
2217 	vcore = NULL;
2218 	err = -EINVAL;
2219 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2220 		if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2221 			pr_devel("KVM: VCPU ID too high\n");
2222 			core = KVM_MAX_VCORES;
2223 		} else {
2224 			BUG_ON(kvm->arch.smt_mode != 1);
2225 			core = kvmppc_pack_vcpu_id(kvm, id);
2226 		}
2227 	} else {
2228 		core = id / kvm->arch.smt_mode;
2229 	}
2230 	if (core < KVM_MAX_VCORES) {
2231 		vcore = kvm->arch.vcores[core];
2232 		if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2233 			pr_devel("KVM: collision on id %u", id);
2234 			vcore = NULL;
2235 		} else if (!vcore) {
2236 			err = -ENOMEM;
2237 			vcore = kvmppc_vcore_create(kvm,
2238 					id & ~(kvm->arch.smt_mode - 1));
2239 			kvm->arch.vcores[core] = vcore;
2240 			kvm->arch.online_vcores++;
2241 		}
2242 	}
2243 	mutex_unlock(&kvm->lock);
2244 
2245 	if (!vcore)
2246 		goto free_vcpu;
2247 
2248 	spin_lock(&vcore->lock);
2249 	++vcore->num_threads;
2250 	spin_unlock(&vcore->lock);
2251 	vcpu->arch.vcore = vcore;
2252 	vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2253 	vcpu->arch.thread_cpu = -1;
2254 	vcpu->arch.prev_cpu = -1;
2255 
2256 	vcpu->arch.cpu_type = KVM_CPU_3S_64;
2257 	kvmppc_sanity_check(vcpu);
2258 
2259 	debugfs_vcpu_init(vcpu, id);
2260 
2261 	return vcpu;
2262 
2263 free_vcpu:
2264 	kmem_cache_free(kvm_vcpu_cache, vcpu);
2265 out:
2266 	return ERR_PTR(err);
2267 }
2268 
2269 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2270 			      unsigned long flags)
2271 {
2272 	int err;
2273 	int esmt = 0;
2274 
2275 	if (flags)
2276 		return -EINVAL;
2277 	if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2278 		return -EINVAL;
2279 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2280 		/*
2281 		 * On POWER8 (or POWER7), the threading mode is "strict",
2282 		 * so we pack smt_mode vcpus per vcore.
2283 		 */
2284 		if (smt_mode > threads_per_subcore)
2285 			return -EINVAL;
2286 	} else {
2287 		/*
2288 		 * On POWER9, the threading mode is "loose",
2289 		 * so each vcpu gets its own vcore.
2290 		 */
2291 		esmt = smt_mode;
2292 		smt_mode = 1;
2293 	}
2294 	mutex_lock(&kvm->lock);
2295 	err = -EBUSY;
2296 	if (!kvm->arch.online_vcores) {
2297 		kvm->arch.smt_mode = smt_mode;
2298 		kvm->arch.emul_smt_mode = esmt;
2299 		err = 0;
2300 	}
2301 	mutex_unlock(&kvm->lock);
2302 
2303 	return err;
2304 }
2305 
2306 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2307 {
2308 	if (vpa->pinned_addr)
2309 		kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2310 					vpa->dirty);
2311 }
2312 
2313 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2314 {
2315 	spin_lock(&vcpu->arch.vpa_update_lock);
2316 	unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2317 	unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2318 	unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2319 	spin_unlock(&vcpu->arch.vpa_update_lock);
2320 	kvm_vcpu_uninit(vcpu);
2321 	kmem_cache_free(kvm_vcpu_cache, vcpu);
2322 }
2323 
2324 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2325 {
2326 	/* Indicate we want to get back into the guest */
2327 	return 1;
2328 }
2329 
2330 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2331 {
2332 	unsigned long dec_nsec, now;
2333 
2334 	now = get_tb();
2335 	if (now > vcpu->arch.dec_expires) {
2336 		/* decrementer has already gone negative */
2337 		kvmppc_core_queue_dec(vcpu);
2338 		kvmppc_core_prepare_to_enter(vcpu);
2339 		return;
2340 	}
2341 	dec_nsec = tb_to_ns(vcpu->arch.dec_expires - now);
2342 	hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2343 	vcpu->arch.timer_running = 1;
2344 }
2345 
2346 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
2347 {
2348 	vcpu->arch.ceded = 0;
2349 	if (vcpu->arch.timer_running) {
2350 		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2351 		vcpu->arch.timer_running = 0;
2352 	}
2353 }
2354 
2355 extern int __kvmppc_vcore_entry(void);
2356 
2357 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2358 				   struct kvm_vcpu *vcpu)
2359 {
2360 	u64 now;
2361 
2362 	if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2363 		return;
2364 	spin_lock_irq(&vcpu->arch.tbacct_lock);
2365 	now = mftb();
2366 	vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2367 		vcpu->arch.stolen_logged;
2368 	vcpu->arch.busy_preempt = now;
2369 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2370 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
2371 	--vc->n_runnable;
2372 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2373 }
2374 
2375 static int kvmppc_grab_hwthread(int cpu)
2376 {
2377 	struct paca_struct *tpaca;
2378 	long timeout = 10000;
2379 
2380 	tpaca = paca_ptrs[cpu];
2381 
2382 	/* Ensure the thread won't go into the kernel if it wakes */
2383 	tpaca->kvm_hstate.kvm_vcpu = NULL;
2384 	tpaca->kvm_hstate.kvm_vcore = NULL;
2385 	tpaca->kvm_hstate.napping = 0;
2386 	smp_wmb();
2387 	tpaca->kvm_hstate.hwthread_req = 1;
2388 
2389 	/*
2390 	 * If the thread is already executing in the kernel (e.g. handling
2391 	 * a stray interrupt), wait for it to get back to nap mode.
2392 	 * The smp_mb() is to ensure that our setting of hwthread_req
2393 	 * is visible before we look at hwthread_state, so if this
2394 	 * races with the code at system_reset_pSeries and the thread
2395 	 * misses our setting of hwthread_req, we are sure to see its
2396 	 * setting of hwthread_state, and vice versa.
2397 	 */
2398 	smp_mb();
2399 	while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2400 		if (--timeout <= 0) {
2401 			pr_err("KVM: couldn't grab cpu %d\n", cpu);
2402 			return -EBUSY;
2403 		}
2404 		udelay(1);
2405 	}
2406 	return 0;
2407 }
2408 
2409 static void kvmppc_release_hwthread(int cpu)
2410 {
2411 	struct paca_struct *tpaca;
2412 
2413 	tpaca = paca_ptrs[cpu];
2414 	tpaca->kvm_hstate.hwthread_req = 0;
2415 	tpaca->kvm_hstate.kvm_vcpu = NULL;
2416 	tpaca->kvm_hstate.kvm_vcore = NULL;
2417 	tpaca->kvm_hstate.kvm_split_mode = NULL;
2418 }
2419 
2420 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2421 {
2422 	struct kvm_nested_guest *nested = vcpu->arch.nested;
2423 	cpumask_t *cpu_in_guest;
2424 	int i;
2425 
2426 	cpu = cpu_first_thread_sibling(cpu);
2427 	if (nested) {
2428 		cpumask_set_cpu(cpu, &nested->need_tlb_flush);
2429 		cpu_in_guest = &nested->cpu_in_guest;
2430 	} else {
2431 		cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2432 		cpu_in_guest = &kvm->arch.cpu_in_guest;
2433 	}
2434 	/*
2435 	 * Make sure setting of bit in need_tlb_flush precedes
2436 	 * testing of cpu_in_guest bits.  The matching barrier on
2437 	 * the other side is the first smp_mb() in kvmppc_run_core().
2438 	 */
2439 	smp_mb();
2440 	for (i = 0; i < threads_per_core; ++i)
2441 		if (cpumask_test_cpu(cpu + i, cpu_in_guest))
2442 			smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2443 }
2444 
2445 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2446 {
2447 	struct kvm_nested_guest *nested = vcpu->arch.nested;
2448 	struct kvm *kvm = vcpu->kvm;
2449 	int prev_cpu;
2450 
2451 	if (!cpu_has_feature(CPU_FTR_HVMODE))
2452 		return;
2453 
2454 	if (nested)
2455 		prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
2456 	else
2457 		prev_cpu = vcpu->arch.prev_cpu;
2458 
2459 	/*
2460 	 * With radix, the guest can do TLB invalidations itself,
2461 	 * and it could choose to use the local form (tlbiel) if
2462 	 * it is invalidating a translation that has only ever been
2463 	 * used on one vcpu.  However, that doesn't mean it has
2464 	 * only ever been used on one physical cpu, since vcpus
2465 	 * can move around between pcpus.  To cope with this, when
2466 	 * a vcpu moves from one pcpu to another, we need to tell
2467 	 * any vcpus running on the same core as this vcpu previously
2468 	 * ran to flush the TLB.  The TLB is shared between threads,
2469 	 * so we use a single bit in .need_tlb_flush for all 4 threads.
2470 	 */
2471 	if (prev_cpu != pcpu) {
2472 		if (prev_cpu >= 0 &&
2473 		    cpu_first_thread_sibling(prev_cpu) !=
2474 		    cpu_first_thread_sibling(pcpu))
2475 			radix_flush_cpu(kvm, prev_cpu, vcpu);
2476 		if (nested)
2477 			nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
2478 		else
2479 			vcpu->arch.prev_cpu = pcpu;
2480 	}
2481 }
2482 
2483 static void kvmppc_radix_check_need_tlb_flush(struct kvm *kvm, int pcpu,
2484 					      struct kvm_nested_guest *nested)
2485 {
2486 	cpumask_t *need_tlb_flush;
2487 	int lpid;
2488 
2489 	if (!cpu_has_feature(CPU_FTR_HVMODE))
2490 		return;
2491 
2492 	if (cpu_has_feature(CPU_FTR_ARCH_300))
2493 		pcpu &= ~0x3UL;
2494 
2495 	if (nested) {
2496 		lpid = nested->shadow_lpid;
2497 		need_tlb_flush = &nested->need_tlb_flush;
2498 	} else {
2499 		lpid = kvm->arch.lpid;
2500 		need_tlb_flush = &kvm->arch.need_tlb_flush;
2501 	}
2502 
2503 	mtspr(SPRN_LPID, lpid);
2504 	isync();
2505 	smp_mb();
2506 
2507 	if (cpumask_test_cpu(pcpu, need_tlb_flush)) {
2508 		radix__local_flush_tlb_lpid_guest(lpid);
2509 		/* Clear the bit after the TLB flush */
2510 		cpumask_clear_cpu(pcpu, need_tlb_flush);
2511 	}
2512 }
2513 
2514 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2515 {
2516 	int cpu;
2517 	struct paca_struct *tpaca;
2518 	struct kvm *kvm = vc->kvm;
2519 
2520 	cpu = vc->pcpu;
2521 	if (vcpu) {
2522 		if (vcpu->arch.timer_running) {
2523 			hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2524 			vcpu->arch.timer_running = 0;
2525 		}
2526 		cpu += vcpu->arch.ptid;
2527 		vcpu->cpu = vc->pcpu;
2528 		vcpu->arch.thread_cpu = cpu;
2529 		cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2530 	}
2531 	tpaca = paca_ptrs[cpu];
2532 	tpaca->kvm_hstate.kvm_vcpu = vcpu;
2533 	tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2534 	tpaca->kvm_hstate.fake_suspend = 0;
2535 	/* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2536 	smp_wmb();
2537 	tpaca->kvm_hstate.kvm_vcore = vc;
2538 	if (cpu != smp_processor_id())
2539 		kvmppc_ipi_thread(cpu);
2540 }
2541 
2542 static void kvmppc_wait_for_nap(int n_threads)
2543 {
2544 	int cpu = smp_processor_id();
2545 	int i, loops;
2546 
2547 	if (n_threads <= 1)
2548 		return;
2549 	for (loops = 0; loops < 1000000; ++loops) {
2550 		/*
2551 		 * Check if all threads are finished.
2552 		 * We set the vcore pointer when starting a thread
2553 		 * and the thread clears it when finished, so we look
2554 		 * for any threads that still have a non-NULL vcore ptr.
2555 		 */
2556 		for (i = 1; i < n_threads; ++i)
2557 			if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2558 				break;
2559 		if (i == n_threads) {
2560 			HMT_medium();
2561 			return;
2562 		}
2563 		HMT_low();
2564 	}
2565 	HMT_medium();
2566 	for (i = 1; i < n_threads; ++i)
2567 		if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2568 			pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2569 }
2570 
2571 /*
2572  * Check that we are on thread 0 and that any other threads in
2573  * this core are off-line.  Then grab the threads so they can't
2574  * enter the kernel.
2575  */
2576 static int on_primary_thread(void)
2577 {
2578 	int cpu = smp_processor_id();
2579 	int thr;
2580 
2581 	/* Are we on a primary subcore? */
2582 	if (cpu_thread_in_subcore(cpu))
2583 		return 0;
2584 
2585 	thr = 0;
2586 	while (++thr < threads_per_subcore)
2587 		if (cpu_online(cpu + thr))
2588 			return 0;
2589 
2590 	/* Grab all hw threads so they can't go into the kernel */
2591 	for (thr = 1; thr < threads_per_subcore; ++thr) {
2592 		if (kvmppc_grab_hwthread(cpu + thr)) {
2593 			/* Couldn't grab one; let the others go */
2594 			do {
2595 				kvmppc_release_hwthread(cpu + thr);
2596 			} while (--thr > 0);
2597 			return 0;
2598 		}
2599 	}
2600 	return 1;
2601 }
2602 
2603 /*
2604  * A list of virtual cores for each physical CPU.
2605  * These are vcores that could run but their runner VCPU tasks are
2606  * (or may be) preempted.
2607  */
2608 struct preempted_vcore_list {
2609 	struct list_head	list;
2610 	spinlock_t		lock;
2611 };
2612 
2613 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2614 
2615 static void init_vcore_lists(void)
2616 {
2617 	int cpu;
2618 
2619 	for_each_possible_cpu(cpu) {
2620 		struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2621 		spin_lock_init(&lp->lock);
2622 		INIT_LIST_HEAD(&lp->list);
2623 	}
2624 }
2625 
2626 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2627 {
2628 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2629 
2630 	vc->vcore_state = VCORE_PREEMPT;
2631 	vc->pcpu = smp_processor_id();
2632 	if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2633 		spin_lock(&lp->lock);
2634 		list_add_tail(&vc->preempt_list, &lp->list);
2635 		spin_unlock(&lp->lock);
2636 	}
2637 
2638 	/* Start accumulating stolen time */
2639 	kvmppc_core_start_stolen(vc);
2640 }
2641 
2642 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2643 {
2644 	struct preempted_vcore_list *lp;
2645 
2646 	kvmppc_core_end_stolen(vc);
2647 	if (!list_empty(&vc->preempt_list)) {
2648 		lp = &per_cpu(preempted_vcores, vc->pcpu);
2649 		spin_lock(&lp->lock);
2650 		list_del_init(&vc->preempt_list);
2651 		spin_unlock(&lp->lock);
2652 	}
2653 	vc->vcore_state = VCORE_INACTIVE;
2654 }
2655 
2656 /*
2657  * This stores information about the virtual cores currently
2658  * assigned to a physical core.
2659  */
2660 struct core_info {
2661 	int		n_subcores;
2662 	int		max_subcore_threads;
2663 	int		total_threads;
2664 	int		subcore_threads[MAX_SUBCORES];
2665 	struct kvmppc_vcore *vc[MAX_SUBCORES];
2666 };
2667 
2668 /*
2669  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2670  * respectively in 2-way micro-threading (split-core) mode on POWER8.
2671  */
2672 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2673 
2674 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2675 {
2676 	memset(cip, 0, sizeof(*cip));
2677 	cip->n_subcores = 1;
2678 	cip->max_subcore_threads = vc->num_threads;
2679 	cip->total_threads = vc->num_threads;
2680 	cip->subcore_threads[0] = vc->num_threads;
2681 	cip->vc[0] = vc;
2682 }
2683 
2684 static bool subcore_config_ok(int n_subcores, int n_threads)
2685 {
2686 	/*
2687 	 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
2688 	 * split-core mode, with one thread per subcore.
2689 	 */
2690 	if (cpu_has_feature(CPU_FTR_ARCH_300))
2691 		return n_subcores <= 4 && n_threads == 1;
2692 
2693 	/* On POWER8, can only dynamically split if unsplit to begin with */
2694 	if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2695 		return false;
2696 	if (n_subcores > MAX_SUBCORES)
2697 		return false;
2698 	if (n_subcores > 1) {
2699 		if (!(dynamic_mt_modes & 2))
2700 			n_subcores = 4;
2701 		if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2702 			return false;
2703 	}
2704 
2705 	return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2706 }
2707 
2708 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2709 {
2710 	vc->entry_exit_map = 0;
2711 	vc->in_guest = 0;
2712 	vc->napping_threads = 0;
2713 	vc->conferring_threads = 0;
2714 	vc->tb_offset_applied = 0;
2715 }
2716 
2717 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2718 {
2719 	int n_threads = vc->num_threads;
2720 	int sub;
2721 
2722 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2723 		return false;
2724 
2725 	/* In one_vm_per_core mode, require all vcores to be from the same vm */
2726 	if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
2727 		return false;
2728 
2729 	/* Some POWER9 chips require all threads to be in the same MMU mode */
2730 	if (no_mixing_hpt_and_radix &&
2731 	    kvm_is_radix(vc->kvm) != kvm_is_radix(cip->vc[0]->kvm))
2732 		return false;
2733 
2734 	if (n_threads < cip->max_subcore_threads)
2735 		n_threads = cip->max_subcore_threads;
2736 	if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2737 		return false;
2738 	cip->max_subcore_threads = n_threads;
2739 
2740 	sub = cip->n_subcores;
2741 	++cip->n_subcores;
2742 	cip->total_threads += vc->num_threads;
2743 	cip->subcore_threads[sub] = vc->num_threads;
2744 	cip->vc[sub] = vc;
2745 	init_vcore_to_run(vc);
2746 	list_del_init(&vc->preempt_list);
2747 
2748 	return true;
2749 }
2750 
2751 /*
2752  * Work out whether it is possible to piggyback the execution of
2753  * vcore *pvc onto the execution of the other vcores described in *cip.
2754  */
2755 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2756 			  int target_threads)
2757 {
2758 	if (cip->total_threads + pvc->num_threads > target_threads)
2759 		return false;
2760 
2761 	return can_dynamic_split(pvc, cip);
2762 }
2763 
2764 static void prepare_threads(struct kvmppc_vcore *vc)
2765 {
2766 	int i;
2767 	struct kvm_vcpu *vcpu;
2768 
2769 	for_each_runnable_thread(i, vcpu, vc) {
2770 		if (signal_pending(vcpu->arch.run_task))
2771 			vcpu->arch.ret = -EINTR;
2772 		else if (vcpu->arch.vpa.update_pending ||
2773 			 vcpu->arch.slb_shadow.update_pending ||
2774 			 vcpu->arch.dtl.update_pending)
2775 			vcpu->arch.ret = RESUME_GUEST;
2776 		else
2777 			continue;
2778 		kvmppc_remove_runnable(vc, vcpu);
2779 		wake_up(&vcpu->arch.cpu_run);
2780 	}
2781 }
2782 
2783 static void collect_piggybacks(struct core_info *cip, int target_threads)
2784 {
2785 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2786 	struct kvmppc_vcore *pvc, *vcnext;
2787 
2788 	spin_lock(&lp->lock);
2789 	list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2790 		if (!spin_trylock(&pvc->lock))
2791 			continue;
2792 		prepare_threads(pvc);
2793 		if (!pvc->n_runnable) {
2794 			list_del_init(&pvc->preempt_list);
2795 			if (pvc->runner == NULL) {
2796 				pvc->vcore_state = VCORE_INACTIVE;
2797 				kvmppc_core_end_stolen(pvc);
2798 			}
2799 			spin_unlock(&pvc->lock);
2800 			continue;
2801 		}
2802 		if (!can_piggyback(pvc, cip, target_threads)) {
2803 			spin_unlock(&pvc->lock);
2804 			continue;
2805 		}
2806 		kvmppc_core_end_stolen(pvc);
2807 		pvc->vcore_state = VCORE_PIGGYBACK;
2808 		if (cip->total_threads >= target_threads)
2809 			break;
2810 	}
2811 	spin_unlock(&lp->lock);
2812 }
2813 
2814 static bool recheck_signals(struct core_info *cip)
2815 {
2816 	int sub, i;
2817 	struct kvm_vcpu *vcpu;
2818 
2819 	for (sub = 0; sub < cip->n_subcores; ++sub)
2820 		for_each_runnable_thread(i, vcpu, cip->vc[sub])
2821 			if (signal_pending(vcpu->arch.run_task))
2822 				return true;
2823 	return false;
2824 }
2825 
2826 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2827 {
2828 	int still_running = 0, i;
2829 	u64 now;
2830 	long ret;
2831 	struct kvm_vcpu *vcpu;
2832 
2833 	spin_lock(&vc->lock);
2834 	now = get_tb();
2835 	for_each_runnable_thread(i, vcpu, vc) {
2836 		/*
2837 		 * It's safe to unlock the vcore in the loop here, because
2838 		 * for_each_runnable_thread() is safe against removal of
2839 		 * the vcpu, and the vcore state is VCORE_EXITING here,
2840 		 * so any vcpus becoming runnable will have their arch.trap
2841 		 * set to zero and can't actually run in the guest.
2842 		 */
2843 		spin_unlock(&vc->lock);
2844 		/* cancel pending dec exception if dec is positive */
2845 		if (now < vcpu->arch.dec_expires &&
2846 		    kvmppc_core_pending_dec(vcpu))
2847 			kvmppc_core_dequeue_dec(vcpu);
2848 
2849 		trace_kvm_guest_exit(vcpu);
2850 
2851 		ret = RESUME_GUEST;
2852 		if (vcpu->arch.trap)
2853 			ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2854 						    vcpu->arch.run_task);
2855 
2856 		vcpu->arch.ret = ret;
2857 		vcpu->arch.trap = 0;
2858 
2859 		spin_lock(&vc->lock);
2860 		if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2861 			if (vcpu->arch.pending_exceptions)
2862 				kvmppc_core_prepare_to_enter(vcpu);
2863 			if (vcpu->arch.ceded)
2864 				kvmppc_set_timer(vcpu);
2865 			else
2866 				++still_running;
2867 		} else {
2868 			kvmppc_remove_runnable(vc, vcpu);
2869 			wake_up(&vcpu->arch.cpu_run);
2870 		}
2871 	}
2872 	if (!is_master) {
2873 		if (still_running > 0) {
2874 			kvmppc_vcore_preempt(vc);
2875 		} else if (vc->runner) {
2876 			vc->vcore_state = VCORE_PREEMPT;
2877 			kvmppc_core_start_stolen(vc);
2878 		} else {
2879 			vc->vcore_state = VCORE_INACTIVE;
2880 		}
2881 		if (vc->n_runnable > 0 && vc->runner == NULL) {
2882 			/* make sure there's a candidate runner awake */
2883 			i = -1;
2884 			vcpu = next_runnable_thread(vc, &i);
2885 			wake_up(&vcpu->arch.cpu_run);
2886 		}
2887 	}
2888 	spin_unlock(&vc->lock);
2889 }
2890 
2891 /*
2892  * Clear core from the list of active host cores as we are about to
2893  * enter the guest. Only do this if it is the primary thread of the
2894  * core (not if a subcore) that is entering the guest.
2895  */
2896 static inline int kvmppc_clear_host_core(unsigned int cpu)
2897 {
2898 	int core;
2899 
2900 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2901 		return 0;
2902 	/*
2903 	 * Memory barrier can be omitted here as we will do a smp_wmb()
2904 	 * later in kvmppc_start_thread and we need ensure that state is
2905 	 * visible to other CPUs only after we enter guest.
2906 	 */
2907 	core = cpu >> threads_shift;
2908 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2909 	return 0;
2910 }
2911 
2912 /*
2913  * Advertise this core as an active host core since we exited the guest
2914  * Only need to do this if it is the primary thread of the core that is
2915  * exiting.
2916  */
2917 static inline int kvmppc_set_host_core(unsigned int cpu)
2918 {
2919 	int core;
2920 
2921 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2922 		return 0;
2923 
2924 	/*
2925 	 * Memory barrier can be omitted here because we do a spin_unlock
2926 	 * immediately after this which provides the memory barrier.
2927 	 */
2928 	core = cpu >> threads_shift;
2929 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2930 	return 0;
2931 }
2932 
2933 static void set_irq_happened(int trap)
2934 {
2935 	switch (trap) {
2936 	case BOOK3S_INTERRUPT_EXTERNAL:
2937 		local_paca->irq_happened |= PACA_IRQ_EE;
2938 		break;
2939 	case BOOK3S_INTERRUPT_H_DOORBELL:
2940 		local_paca->irq_happened |= PACA_IRQ_DBELL;
2941 		break;
2942 	case BOOK3S_INTERRUPT_HMI:
2943 		local_paca->irq_happened |= PACA_IRQ_HMI;
2944 		break;
2945 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
2946 		replay_system_reset();
2947 		break;
2948 	}
2949 }
2950 
2951 /*
2952  * Run a set of guest threads on a physical core.
2953  * Called with vc->lock held.
2954  */
2955 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2956 {
2957 	struct kvm_vcpu *vcpu;
2958 	int i;
2959 	int srcu_idx;
2960 	struct core_info core_info;
2961 	struct kvmppc_vcore *pvc;
2962 	struct kvm_split_mode split_info, *sip;
2963 	int split, subcore_size, active;
2964 	int sub;
2965 	bool thr0_done;
2966 	unsigned long cmd_bit, stat_bit;
2967 	int pcpu, thr;
2968 	int target_threads;
2969 	int controlled_threads;
2970 	int trap;
2971 	bool is_power8;
2972 	bool hpt_on_radix;
2973 
2974 	/*
2975 	 * Remove from the list any threads that have a signal pending
2976 	 * or need a VPA update done
2977 	 */
2978 	prepare_threads(vc);
2979 
2980 	/* if the runner is no longer runnable, let the caller pick a new one */
2981 	if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2982 		return;
2983 
2984 	/*
2985 	 * Initialize *vc.
2986 	 */
2987 	init_vcore_to_run(vc);
2988 	vc->preempt_tb = TB_NIL;
2989 
2990 	/*
2991 	 * Number of threads that we will be controlling: the same as
2992 	 * the number of threads per subcore, except on POWER9,
2993 	 * where it's 1 because the threads are (mostly) independent.
2994 	 */
2995 	controlled_threads = threads_per_vcore(vc->kvm);
2996 
2997 	/*
2998 	 * Make sure we are running on primary threads, and that secondary
2999 	 * threads are offline.  Also check if the number of threads in this
3000 	 * guest are greater than the current system threads per guest.
3001 	 * On POWER9, we need to be not in independent-threads mode if
3002 	 * this is a HPT guest on a radix host machine where the
3003 	 * CPU threads may not be in different MMU modes.
3004 	 */
3005 	hpt_on_radix = no_mixing_hpt_and_radix && radix_enabled() &&
3006 		!kvm_is_radix(vc->kvm);
3007 	if (((controlled_threads > 1) &&
3008 	     ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) ||
3009 	    (hpt_on_radix && vc->kvm->arch.threads_indep)) {
3010 		for_each_runnable_thread(i, vcpu, vc) {
3011 			vcpu->arch.ret = -EBUSY;
3012 			kvmppc_remove_runnable(vc, vcpu);
3013 			wake_up(&vcpu->arch.cpu_run);
3014 		}
3015 		goto out;
3016 	}
3017 
3018 	/*
3019 	 * See if we could run any other vcores on the physical core
3020 	 * along with this one.
3021 	 */
3022 	init_core_info(&core_info, vc);
3023 	pcpu = smp_processor_id();
3024 	target_threads = controlled_threads;
3025 	if (target_smt_mode && target_smt_mode < target_threads)
3026 		target_threads = target_smt_mode;
3027 	if (vc->num_threads < target_threads)
3028 		collect_piggybacks(&core_info, target_threads);
3029 
3030 	/*
3031 	 * On radix, arrange for TLB flushing if necessary.
3032 	 * This has to be done before disabling interrupts since
3033 	 * it uses smp_call_function().
3034 	 */
3035 	pcpu = smp_processor_id();
3036 	if (kvm_is_radix(vc->kvm)) {
3037 		for (sub = 0; sub < core_info.n_subcores; ++sub)
3038 			for_each_runnable_thread(i, vcpu, core_info.vc[sub])
3039 				kvmppc_prepare_radix_vcpu(vcpu, pcpu);
3040 	}
3041 
3042 	/*
3043 	 * Hard-disable interrupts, and check resched flag and signals.
3044 	 * If we need to reschedule or deliver a signal, clean up
3045 	 * and return without going into the guest(s).
3046 	 * If the mmu_ready flag has been cleared, don't go into the
3047 	 * guest because that means a HPT resize operation is in progress.
3048 	 */
3049 	local_irq_disable();
3050 	hard_irq_disable();
3051 	if (lazy_irq_pending() || need_resched() ||
3052 	    recheck_signals(&core_info) || !vc->kvm->arch.mmu_ready) {
3053 		local_irq_enable();
3054 		vc->vcore_state = VCORE_INACTIVE;
3055 		/* Unlock all except the primary vcore */
3056 		for (sub = 1; sub < core_info.n_subcores; ++sub) {
3057 			pvc = core_info.vc[sub];
3058 			/* Put back on to the preempted vcores list */
3059 			kvmppc_vcore_preempt(pvc);
3060 			spin_unlock(&pvc->lock);
3061 		}
3062 		for (i = 0; i < controlled_threads; ++i)
3063 			kvmppc_release_hwthread(pcpu + i);
3064 		return;
3065 	}
3066 
3067 	kvmppc_clear_host_core(pcpu);
3068 
3069 	/* Decide on micro-threading (split-core) mode */
3070 	subcore_size = threads_per_subcore;
3071 	cmd_bit = stat_bit = 0;
3072 	split = core_info.n_subcores;
3073 	sip = NULL;
3074 	is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
3075 		&& !cpu_has_feature(CPU_FTR_ARCH_300);
3076 
3077 	if (split > 1 || hpt_on_radix) {
3078 		sip = &split_info;
3079 		memset(&split_info, 0, sizeof(split_info));
3080 		for (sub = 0; sub < core_info.n_subcores; ++sub)
3081 			split_info.vc[sub] = core_info.vc[sub];
3082 
3083 		if (is_power8) {
3084 			if (split == 2 && (dynamic_mt_modes & 2)) {
3085 				cmd_bit = HID0_POWER8_1TO2LPAR;
3086 				stat_bit = HID0_POWER8_2LPARMODE;
3087 			} else {
3088 				split = 4;
3089 				cmd_bit = HID0_POWER8_1TO4LPAR;
3090 				stat_bit = HID0_POWER8_4LPARMODE;
3091 			}
3092 			subcore_size = MAX_SMT_THREADS / split;
3093 			split_info.rpr = mfspr(SPRN_RPR);
3094 			split_info.pmmar = mfspr(SPRN_PMMAR);
3095 			split_info.ldbar = mfspr(SPRN_LDBAR);
3096 			split_info.subcore_size = subcore_size;
3097 		} else {
3098 			split_info.subcore_size = 1;
3099 			if (hpt_on_radix) {
3100 				/* Use the split_info for LPCR/LPIDR changes */
3101 				split_info.lpcr_req = vc->lpcr;
3102 				split_info.lpidr_req = vc->kvm->arch.lpid;
3103 				split_info.host_lpcr = vc->kvm->arch.host_lpcr;
3104 				split_info.do_set = 1;
3105 			}
3106 		}
3107 
3108 		/* order writes to split_info before kvm_split_mode pointer */
3109 		smp_wmb();
3110 	}
3111 
3112 	for (thr = 0; thr < controlled_threads; ++thr) {
3113 		struct paca_struct *paca = paca_ptrs[pcpu + thr];
3114 
3115 		paca->kvm_hstate.tid = thr;
3116 		paca->kvm_hstate.napping = 0;
3117 		paca->kvm_hstate.kvm_split_mode = sip;
3118 	}
3119 
3120 	/* Initiate micro-threading (split-core) on POWER8 if required */
3121 	if (cmd_bit) {
3122 		unsigned long hid0 = mfspr(SPRN_HID0);
3123 
3124 		hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3125 		mb();
3126 		mtspr(SPRN_HID0, hid0);
3127 		isync();
3128 		for (;;) {
3129 			hid0 = mfspr(SPRN_HID0);
3130 			if (hid0 & stat_bit)
3131 				break;
3132 			cpu_relax();
3133 		}
3134 	}
3135 
3136 	/*
3137 	 * On POWER8, set RWMR register.
3138 	 * Since it only affects PURR and SPURR, it doesn't affect
3139 	 * the host, so we don't save/restore the host value.
3140 	 */
3141 	if (is_power8) {
3142 		unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3143 		int n_online = atomic_read(&vc->online_count);
3144 
3145 		/*
3146 		 * Use the 8-thread value if we're doing split-core
3147 		 * or if the vcore's online count looks bogus.
3148 		 */
3149 		if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3150 		    n_online >= 1 && n_online <= MAX_SMT_THREADS)
3151 			rwmr_val = p8_rwmr_values[n_online];
3152 		mtspr(SPRN_RWMR, rwmr_val);
3153 	}
3154 
3155 	/* Start all the threads */
3156 	active = 0;
3157 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
3158 		thr = is_power8 ? subcore_thread_map[sub] : sub;
3159 		thr0_done = false;
3160 		active |= 1 << thr;
3161 		pvc = core_info.vc[sub];
3162 		pvc->pcpu = pcpu + thr;
3163 		for_each_runnable_thread(i, vcpu, pvc) {
3164 			kvmppc_start_thread(vcpu, pvc);
3165 			kvmppc_create_dtl_entry(vcpu, pvc);
3166 			trace_kvm_guest_enter(vcpu);
3167 			if (!vcpu->arch.ptid)
3168 				thr0_done = true;
3169 			active |= 1 << (thr + vcpu->arch.ptid);
3170 		}
3171 		/*
3172 		 * We need to start the first thread of each subcore
3173 		 * even if it doesn't have a vcpu.
3174 		 */
3175 		if (!thr0_done)
3176 			kvmppc_start_thread(NULL, pvc);
3177 	}
3178 
3179 	/*
3180 	 * Ensure that split_info.do_nap is set after setting
3181 	 * the vcore pointer in the PACA of the secondaries.
3182 	 */
3183 	smp_mb();
3184 
3185 	/*
3186 	 * When doing micro-threading, poke the inactive threads as well.
3187 	 * This gets them to the nap instruction after kvm_do_nap,
3188 	 * which reduces the time taken to unsplit later.
3189 	 * For POWER9 HPT guest on radix host, we need all the secondary
3190 	 * threads woken up so they can do the LPCR/LPIDR change.
3191 	 */
3192 	if (cmd_bit || hpt_on_radix) {
3193 		split_info.do_nap = 1;	/* ask secondaries to nap when done */
3194 		for (thr = 1; thr < threads_per_subcore; ++thr)
3195 			if (!(active & (1 << thr)))
3196 				kvmppc_ipi_thread(pcpu + thr);
3197 	}
3198 
3199 	vc->vcore_state = VCORE_RUNNING;
3200 	preempt_disable();
3201 
3202 	trace_kvmppc_run_core(vc, 0);
3203 
3204 	for (sub = 0; sub < core_info.n_subcores; ++sub)
3205 		spin_unlock(&core_info.vc[sub]->lock);
3206 
3207 	if (kvm_is_radix(vc->kvm)) {
3208 		/*
3209 		 * Do we need to flush the process scoped TLB for the LPAR?
3210 		 *
3211 		 * On POWER9, individual threads can come in here, but the
3212 		 * TLB is shared between the 4 threads in a core, hence
3213 		 * invalidating on one thread invalidates for all.
3214 		 * Thus we make all 4 threads use the same bit here.
3215 		 *
3216 		 * Hash must be flushed in realmode in order to use tlbiel.
3217 		 */
3218 		kvmppc_radix_check_need_tlb_flush(vc->kvm, pcpu, NULL);
3219 	}
3220 
3221 	/*
3222 	 * Interrupts will be enabled once we get into the guest,
3223 	 * so tell lockdep that we're about to enable interrupts.
3224 	 */
3225 	trace_hardirqs_on();
3226 
3227 	guest_enter_irqoff();
3228 
3229 	srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3230 
3231 	this_cpu_disable_ftrace();
3232 
3233 	trap = __kvmppc_vcore_entry();
3234 
3235 	this_cpu_enable_ftrace();
3236 
3237 	srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3238 
3239 	trace_hardirqs_off();
3240 	set_irq_happened(trap);
3241 
3242 	spin_lock(&vc->lock);
3243 	/* prevent other vcpu threads from doing kvmppc_start_thread() now */
3244 	vc->vcore_state = VCORE_EXITING;
3245 
3246 	/* wait for secondary threads to finish writing their state to memory */
3247 	kvmppc_wait_for_nap(controlled_threads);
3248 
3249 	/* Return to whole-core mode if we split the core earlier */
3250 	if (cmd_bit) {
3251 		unsigned long hid0 = mfspr(SPRN_HID0);
3252 		unsigned long loops = 0;
3253 
3254 		hid0 &= ~HID0_POWER8_DYNLPARDIS;
3255 		stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3256 		mb();
3257 		mtspr(SPRN_HID0, hid0);
3258 		isync();
3259 		for (;;) {
3260 			hid0 = mfspr(SPRN_HID0);
3261 			if (!(hid0 & stat_bit))
3262 				break;
3263 			cpu_relax();
3264 			++loops;
3265 		}
3266 	} else if (hpt_on_radix) {
3267 		/* Wait for all threads to have seen final sync */
3268 		for (thr = 1; thr < controlled_threads; ++thr) {
3269 			struct paca_struct *paca = paca_ptrs[pcpu + thr];
3270 
3271 			while (paca->kvm_hstate.kvm_split_mode) {
3272 				HMT_low();
3273 				barrier();
3274 			}
3275 			HMT_medium();
3276 		}
3277 	}
3278 	split_info.do_nap = 0;
3279 
3280 	kvmppc_set_host_core(pcpu);
3281 
3282 	local_irq_enable();
3283 	guest_exit();
3284 
3285 	/* Let secondaries go back to the offline loop */
3286 	for (i = 0; i < controlled_threads; ++i) {
3287 		kvmppc_release_hwthread(pcpu + i);
3288 		if (sip && sip->napped[i])
3289 			kvmppc_ipi_thread(pcpu + i);
3290 		cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
3291 	}
3292 
3293 	spin_unlock(&vc->lock);
3294 
3295 	/* make sure updates to secondary vcpu structs are visible now */
3296 	smp_mb();
3297 
3298 	preempt_enable();
3299 
3300 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
3301 		pvc = core_info.vc[sub];
3302 		post_guest_process(pvc, pvc == vc);
3303 	}
3304 
3305 	spin_lock(&vc->lock);
3306 
3307  out:
3308 	vc->vcore_state = VCORE_INACTIVE;
3309 	trace_kvmppc_run_core(vc, 1);
3310 }
3311 
3312 /*
3313  * Load up hypervisor-mode registers on P9.
3314  */
3315 static int kvmhv_load_hv_regs_and_go(struct kvm_vcpu *vcpu, u64 time_limit,
3316 				     unsigned long lpcr)
3317 {
3318 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
3319 	s64 hdec;
3320 	u64 tb, purr, spurr;
3321 	int trap;
3322 	unsigned long host_hfscr = mfspr(SPRN_HFSCR);
3323 	unsigned long host_ciabr = mfspr(SPRN_CIABR);
3324 	unsigned long host_dawr = mfspr(SPRN_DAWR);
3325 	unsigned long host_dawrx = mfspr(SPRN_DAWRX);
3326 	unsigned long host_psscr = mfspr(SPRN_PSSCR);
3327 	unsigned long host_pidr = mfspr(SPRN_PID);
3328 
3329 	hdec = time_limit - mftb();
3330 	if (hdec < 0)
3331 		return BOOK3S_INTERRUPT_HV_DECREMENTER;
3332 	mtspr(SPRN_HDEC, hdec);
3333 
3334 	if (vc->tb_offset) {
3335 		u64 new_tb = mftb() + vc->tb_offset;
3336 		mtspr(SPRN_TBU40, new_tb);
3337 		tb = mftb();
3338 		if ((tb & 0xffffff) < (new_tb & 0xffffff))
3339 			mtspr(SPRN_TBU40, new_tb + 0x1000000);
3340 		vc->tb_offset_applied = vc->tb_offset;
3341 	}
3342 
3343 	if (vc->pcr)
3344 		mtspr(SPRN_PCR, vc->pcr);
3345 	mtspr(SPRN_DPDES, vc->dpdes);
3346 	mtspr(SPRN_VTB, vc->vtb);
3347 
3348 	local_paca->kvm_hstate.host_purr = mfspr(SPRN_PURR);
3349 	local_paca->kvm_hstate.host_spurr = mfspr(SPRN_SPURR);
3350 	mtspr(SPRN_PURR, vcpu->arch.purr);
3351 	mtspr(SPRN_SPURR, vcpu->arch.spurr);
3352 
3353 	if (cpu_has_feature(CPU_FTR_DAWR)) {
3354 		mtspr(SPRN_DAWR, vcpu->arch.dawr);
3355 		mtspr(SPRN_DAWRX, vcpu->arch.dawrx);
3356 	}
3357 	mtspr(SPRN_CIABR, vcpu->arch.ciabr);
3358 	mtspr(SPRN_IC, vcpu->arch.ic);
3359 	mtspr(SPRN_PID, vcpu->arch.pid);
3360 
3361 	mtspr(SPRN_PSSCR, vcpu->arch.psscr | PSSCR_EC |
3362 	      (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3363 
3364 	mtspr(SPRN_HFSCR, vcpu->arch.hfscr);
3365 
3366 	mtspr(SPRN_SPRG0, vcpu->arch.shregs.sprg0);
3367 	mtspr(SPRN_SPRG1, vcpu->arch.shregs.sprg1);
3368 	mtspr(SPRN_SPRG2, vcpu->arch.shregs.sprg2);
3369 	mtspr(SPRN_SPRG3, vcpu->arch.shregs.sprg3);
3370 
3371 	mtspr(SPRN_AMOR, ~0UL);
3372 
3373 	mtspr(SPRN_LPCR, lpcr);
3374 	isync();
3375 
3376 	kvmppc_xive_push_vcpu(vcpu);
3377 
3378 	mtspr(SPRN_SRR0, vcpu->arch.shregs.srr0);
3379 	mtspr(SPRN_SRR1, vcpu->arch.shregs.srr1);
3380 
3381 	trap = __kvmhv_vcpu_entry_p9(vcpu);
3382 
3383 	/* Advance host PURR/SPURR by the amount used by guest */
3384 	purr = mfspr(SPRN_PURR);
3385 	spurr = mfspr(SPRN_SPURR);
3386 	mtspr(SPRN_PURR, local_paca->kvm_hstate.host_purr +
3387 	      purr - vcpu->arch.purr);
3388 	mtspr(SPRN_SPURR, local_paca->kvm_hstate.host_spurr +
3389 	      spurr - vcpu->arch.spurr);
3390 	vcpu->arch.purr = purr;
3391 	vcpu->arch.spurr = spurr;
3392 
3393 	vcpu->arch.ic = mfspr(SPRN_IC);
3394 	vcpu->arch.pid = mfspr(SPRN_PID);
3395 	vcpu->arch.psscr = mfspr(SPRN_PSSCR) & PSSCR_GUEST_VIS;
3396 
3397 	vcpu->arch.shregs.sprg0 = mfspr(SPRN_SPRG0);
3398 	vcpu->arch.shregs.sprg1 = mfspr(SPRN_SPRG1);
3399 	vcpu->arch.shregs.sprg2 = mfspr(SPRN_SPRG2);
3400 	vcpu->arch.shregs.sprg3 = mfspr(SPRN_SPRG3);
3401 
3402 	mtspr(SPRN_PSSCR, host_psscr);
3403 	mtspr(SPRN_HFSCR, host_hfscr);
3404 	mtspr(SPRN_CIABR, host_ciabr);
3405 	mtspr(SPRN_DAWR, host_dawr);
3406 	mtspr(SPRN_DAWRX, host_dawrx);
3407 	mtspr(SPRN_PID, host_pidr);
3408 
3409 	/*
3410 	 * Since this is radix, do a eieio; tlbsync; ptesync sequence in
3411 	 * case we interrupted the guest between a tlbie and a ptesync.
3412 	 */
3413 	asm volatile("eieio; tlbsync; ptesync");
3414 
3415 	mtspr(SPRN_LPID, vcpu->kvm->arch.host_lpid);	/* restore host LPID */
3416 	isync();
3417 
3418 	vc->dpdes = mfspr(SPRN_DPDES);
3419 	vc->vtb = mfspr(SPRN_VTB);
3420 	mtspr(SPRN_DPDES, 0);
3421 	if (vc->pcr)
3422 		mtspr(SPRN_PCR, 0);
3423 
3424 	if (vc->tb_offset_applied) {
3425 		u64 new_tb = mftb() - vc->tb_offset_applied;
3426 		mtspr(SPRN_TBU40, new_tb);
3427 		tb = mftb();
3428 		if ((tb & 0xffffff) < (new_tb & 0xffffff))
3429 			mtspr(SPRN_TBU40, new_tb + 0x1000000);
3430 		vc->tb_offset_applied = 0;
3431 	}
3432 
3433 	mtspr(SPRN_HDEC, 0x7fffffff);
3434 	mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr);
3435 
3436 	return trap;
3437 }
3438 
3439 /*
3440  * Virtual-mode guest entry for POWER9 and later when the host and
3441  * guest are both using the radix MMU.  The LPIDR has already been set.
3442  */
3443 int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
3444 			 unsigned long lpcr)
3445 {
3446 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
3447 	unsigned long host_dscr = mfspr(SPRN_DSCR);
3448 	unsigned long host_tidr = mfspr(SPRN_TIDR);
3449 	unsigned long host_iamr = mfspr(SPRN_IAMR);
3450 	s64 dec;
3451 	u64 tb;
3452 	int trap, save_pmu;
3453 
3454 	dec = mfspr(SPRN_DEC);
3455 	tb = mftb();
3456 	if (dec < 512)
3457 		return BOOK3S_INTERRUPT_HV_DECREMENTER;
3458 	local_paca->kvm_hstate.dec_expires = dec + tb;
3459 	if (local_paca->kvm_hstate.dec_expires < time_limit)
3460 		time_limit = local_paca->kvm_hstate.dec_expires;
3461 
3462 	vcpu->arch.ceded = 0;
3463 
3464 	kvmhv_save_host_pmu();		/* saves it to PACA kvm_hstate */
3465 
3466 	kvmppc_subcore_enter_guest();
3467 
3468 	vc->entry_exit_map = 1;
3469 	vc->in_guest = 1;
3470 
3471 	if (vcpu->arch.vpa.pinned_addr) {
3472 		struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3473 		u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3474 		lp->yield_count = cpu_to_be32(yield_count);
3475 		vcpu->arch.vpa.dirty = 1;
3476 	}
3477 
3478 	if (cpu_has_feature(CPU_FTR_TM) ||
3479 	    cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3480 		kvmppc_restore_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3481 
3482 	kvmhv_load_guest_pmu(vcpu);
3483 
3484 	msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3485 	load_fp_state(&vcpu->arch.fp);
3486 #ifdef CONFIG_ALTIVEC
3487 	load_vr_state(&vcpu->arch.vr);
3488 #endif
3489 
3490 	mtspr(SPRN_DSCR, vcpu->arch.dscr);
3491 	mtspr(SPRN_IAMR, vcpu->arch.iamr);
3492 	mtspr(SPRN_PSPB, vcpu->arch.pspb);
3493 	mtspr(SPRN_FSCR, vcpu->arch.fscr);
3494 	mtspr(SPRN_TAR, vcpu->arch.tar);
3495 	mtspr(SPRN_EBBHR, vcpu->arch.ebbhr);
3496 	mtspr(SPRN_EBBRR, vcpu->arch.ebbrr);
3497 	mtspr(SPRN_BESCR, vcpu->arch.bescr);
3498 	mtspr(SPRN_WORT, vcpu->arch.wort);
3499 	mtspr(SPRN_TIDR, vcpu->arch.tid);
3500 	mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
3501 	mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
3502 	mtspr(SPRN_AMR, vcpu->arch.amr);
3503 	mtspr(SPRN_UAMOR, vcpu->arch.uamor);
3504 
3505 	if (!(vcpu->arch.ctrl & 1))
3506 		mtspr(SPRN_CTRLT, mfspr(SPRN_CTRLF) & ~1);
3507 
3508 	mtspr(SPRN_DEC, vcpu->arch.dec_expires - mftb());
3509 
3510 	if (kvmhv_on_pseries()) {
3511 		/* call our hypervisor to load up HV regs and go */
3512 		struct hv_guest_state hvregs;
3513 
3514 		kvmhv_save_hv_regs(vcpu, &hvregs);
3515 		hvregs.lpcr = lpcr;
3516 		vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
3517 		hvregs.version = HV_GUEST_STATE_VERSION;
3518 		if (vcpu->arch.nested) {
3519 			hvregs.lpid = vcpu->arch.nested->shadow_lpid;
3520 			hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
3521 		} else {
3522 			hvregs.lpid = vcpu->kvm->arch.lpid;
3523 			hvregs.vcpu_token = vcpu->vcpu_id;
3524 		}
3525 		hvregs.hdec_expiry = time_limit;
3526 		trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
3527 					  __pa(&vcpu->arch.regs));
3528 		kvmhv_restore_hv_return_state(vcpu, &hvregs);
3529 		vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
3530 		vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
3531 		vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
3532 
3533 		/* H_CEDE has to be handled now, not later */
3534 		if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
3535 		    kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
3536 			kvmppc_nested_cede(vcpu);
3537 			trap = 0;
3538 		}
3539 	} else {
3540 		trap = kvmhv_load_hv_regs_and_go(vcpu, time_limit, lpcr);
3541 	}
3542 
3543 	vcpu->arch.slb_max = 0;
3544 	dec = mfspr(SPRN_DEC);
3545 	tb = mftb();
3546 	vcpu->arch.dec_expires = dec + tb;
3547 	vcpu->cpu = -1;
3548 	vcpu->arch.thread_cpu = -1;
3549 	vcpu->arch.ctrl = mfspr(SPRN_CTRLF);
3550 
3551 	vcpu->arch.iamr = mfspr(SPRN_IAMR);
3552 	vcpu->arch.pspb = mfspr(SPRN_PSPB);
3553 	vcpu->arch.fscr = mfspr(SPRN_FSCR);
3554 	vcpu->arch.tar = mfspr(SPRN_TAR);
3555 	vcpu->arch.ebbhr = mfspr(SPRN_EBBHR);
3556 	vcpu->arch.ebbrr = mfspr(SPRN_EBBRR);
3557 	vcpu->arch.bescr = mfspr(SPRN_BESCR);
3558 	vcpu->arch.wort = mfspr(SPRN_WORT);
3559 	vcpu->arch.tid = mfspr(SPRN_TIDR);
3560 	vcpu->arch.amr = mfspr(SPRN_AMR);
3561 	vcpu->arch.uamor = mfspr(SPRN_UAMOR);
3562 	vcpu->arch.dscr = mfspr(SPRN_DSCR);
3563 
3564 	mtspr(SPRN_PSPB, 0);
3565 	mtspr(SPRN_WORT, 0);
3566 	mtspr(SPRN_AMR, 0);
3567 	mtspr(SPRN_UAMOR, 0);
3568 	mtspr(SPRN_DSCR, host_dscr);
3569 	mtspr(SPRN_TIDR, host_tidr);
3570 	mtspr(SPRN_IAMR, host_iamr);
3571 	mtspr(SPRN_PSPB, 0);
3572 
3573 	msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3574 	store_fp_state(&vcpu->arch.fp);
3575 #ifdef CONFIG_ALTIVEC
3576 	store_vr_state(&vcpu->arch.vr);
3577 #endif
3578 
3579 	if (cpu_has_feature(CPU_FTR_TM) ||
3580 	    cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3581 		kvmppc_save_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3582 
3583 	save_pmu = 1;
3584 	if (vcpu->arch.vpa.pinned_addr) {
3585 		struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3586 		u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3587 		lp->yield_count = cpu_to_be32(yield_count);
3588 		vcpu->arch.vpa.dirty = 1;
3589 		save_pmu = lp->pmcregs_in_use;
3590 	}
3591 
3592 	kvmhv_save_guest_pmu(vcpu, save_pmu);
3593 
3594 	vc->entry_exit_map = 0x101;
3595 	vc->in_guest = 0;
3596 
3597 	mtspr(SPRN_DEC, local_paca->kvm_hstate.dec_expires - mftb());
3598 
3599 	kvmhv_load_host_pmu();
3600 
3601 	kvmppc_subcore_exit_guest();
3602 
3603 	return trap;
3604 }
3605 
3606 /*
3607  * Wait for some other vcpu thread to execute us, and
3608  * wake us up when we need to handle something in the host.
3609  */
3610 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
3611 				 struct kvm_vcpu *vcpu, int wait_state)
3612 {
3613 	DEFINE_WAIT(wait);
3614 
3615 	prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
3616 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3617 		spin_unlock(&vc->lock);
3618 		schedule();
3619 		spin_lock(&vc->lock);
3620 	}
3621 	finish_wait(&vcpu->arch.cpu_run, &wait);
3622 }
3623 
3624 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
3625 {
3626 	/* 10us base */
3627 	if (vc->halt_poll_ns == 0 && halt_poll_ns_grow)
3628 		vc->halt_poll_ns = 10000;
3629 	else
3630 		vc->halt_poll_ns *= halt_poll_ns_grow;
3631 }
3632 
3633 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
3634 {
3635 	if (halt_poll_ns_shrink == 0)
3636 		vc->halt_poll_ns = 0;
3637 	else
3638 		vc->halt_poll_ns /= halt_poll_ns_shrink;
3639 }
3640 
3641 #ifdef CONFIG_KVM_XICS
3642 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3643 {
3644 	if (!xive_enabled())
3645 		return false;
3646 	return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
3647 		vcpu->arch.xive_saved_state.cppr;
3648 }
3649 #else
3650 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3651 {
3652 	return false;
3653 }
3654 #endif /* CONFIG_KVM_XICS */
3655 
3656 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3657 {
3658 	if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3659 	    kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3660 		return true;
3661 
3662 	return false;
3663 }
3664 
3665 /*
3666  * Check to see if any of the runnable vcpus on the vcore have pending
3667  * exceptions or are no longer ceded
3668  */
3669 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3670 {
3671 	struct kvm_vcpu *vcpu;
3672 	int i;
3673 
3674 	for_each_runnable_thread(i, vcpu, vc) {
3675 		if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3676 			return 1;
3677 	}
3678 
3679 	return 0;
3680 }
3681 
3682 /*
3683  * All the vcpus in this vcore are idle, so wait for a decrementer
3684  * or external interrupt to one of the vcpus.  vc->lock is held.
3685  */
3686 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3687 {
3688 	ktime_t cur, start_poll, start_wait;
3689 	int do_sleep = 1;
3690 	u64 block_ns;
3691 	DECLARE_SWAITQUEUE(wait);
3692 
3693 	/* Poll for pending exceptions and ceded state */
3694 	cur = start_poll = ktime_get();
3695 	if (vc->halt_poll_ns) {
3696 		ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3697 		++vc->runner->stat.halt_attempted_poll;
3698 
3699 		vc->vcore_state = VCORE_POLLING;
3700 		spin_unlock(&vc->lock);
3701 
3702 		do {
3703 			if (kvmppc_vcore_check_block(vc)) {
3704 				do_sleep = 0;
3705 				break;
3706 			}
3707 			cur = ktime_get();
3708 		} while (single_task_running() && ktime_before(cur, stop));
3709 
3710 		spin_lock(&vc->lock);
3711 		vc->vcore_state = VCORE_INACTIVE;
3712 
3713 		if (!do_sleep) {
3714 			++vc->runner->stat.halt_successful_poll;
3715 			goto out;
3716 		}
3717 	}
3718 
3719 	prepare_to_swait_exclusive(&vc->wq, &wait, TASK_INTERRUPTIBLE);
3720 
3721 	if (kvmppc_vcore_check_block(vc)) {
3722 		finish_swait(&vc->wq, &wait);
3723 		do_sleep = 0;
3724 		/* If we polled, count this as a successful poll */
3725 		if (vc->halt_poll_ns)
3726 			++vc->runner->stat.halt_successful_poll;
3727 		goto out;
3728 	}
3729 
3730 	start_wait = ktime_get();
3731 
3732 	vc->vcore_state = VCORE_SLEEPING;
3733 	trace_kvmppc_vcore_blocked(vc, 0);
3734 	spin_unlock(&vc->lock);
3735 	schedule();
3736 	finish_swait(&vc->wq, &wait);
3737 	spin_lock(&vc->lock);
3738 	vc->vcore_state = VCORE_INACTIVE;
3739 	trace_kvmppc_vcore_blocked(vc, 1);
3740 	++vc->runner->stat.halt_successful_wait;
3741 
3742 	cur = ktime_get();
3743 
3744 out:
3745 	block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3746 
3747 	/* Attribute wait time */
3748 	if (do_sleep) {
3749 		vc->runner->stat.halt_wait_ns +=
3750 			ktime_to_ns(cur) - ktime_to_ns(start_wait);
3751 		/* Attribute failed poll time */
3752 		if (vc->halt_poll_ns)
3753 			vc->runner->stat.halt_poll_fail_ns +=
3754 				ktime_to_ns(start_wait) -
3755 				ktime_to_ns(start_poll);
3756 	} else {
3757 		/* Attribute successful poll time */
3758 		if (vc->halt_poll_ns)
3759 			vc->runner->stat.halt_poll_success_ns +=
3760 				ktime_to_ns(cur) -
3761 				ktime_to_ns(start_poll);
3762 	}
3763 
3764 	/* Adjust poll time */
3765 	if (halt_poll_ns) {
3766 		if (block_ns <= vc->halt_poll_ns)
3767 			;
3768 		/* We slept and blocked for longer than the max halt time */
3769 		else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3770 			shrink_halt_poll_ns(vc);
3771 		/* We slept and our poll time is too small */
3772 		else if (vc->halt_poll_ns < halt_poll_ns &&
3773 				block_ns < halt_poll_ns)
3774 			grow_halt_poll_ns(vc);
3775 		if (vc->halt_poll_ns > halt_poll_ns)
3776 			vc->halt_poll_ns = halt_poll_ns;
3777 	} else
3778 		vc->halt_poll_ns = 0;
3779 
3780 	trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3781 }
3782 
3783 /*
3784  * This never fails for a radix guest, as none of the operations it does
3785  * for a radix guest can fail or have a way to report failure.
3786  * kvmhv_run_single_vcpu() relies on this fact.
3787  */
3788 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
3789 {
3790 	int r = 0;
3791 	struct kvm *kvm = vcpu->kvm;
3792 
3793 	mutex_lock(&kvm->lock);
3794 	if (!kvm->arch.mmu_ready) {
3795 		if (!kvm_is_radix(kvm))
3796 			r = kvmppc_hv_setup_htab_rma(vcpu);
3797 		if (!r) {
3798 			if (cpu_has_feature(CPU_FTR_ARCH_300))
3799 				kvmppc_setup_partition_table(kvm);
3800 			kvm->arch.mmu_ready = 1;
3801 		}
3802 	}
3803 	mutex_unlock(&kvm->lock);
3804 	return r;
3805 }
3806 
3807 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3808 {
3809 	int n_ceded, i, r;
3810 	struct kvmppc_vcore *vc;
3811 	struct kvm_vcpu *v;
3812 
3813 	trace_kvmppc_run_vcpu_enter(vcpu);
3814 
3815 	kvm_run->exit_reason = 0;
3816 	vcpu->arch.ret = RESUME_GUEST;
3817 	vcpu->arch.trap = 0;
3818 	kvmppc_update_vpas(vcpu);
3819 
3820 	/*
3821 	 * Synchronize with other threads in this virtual core
3822 	 */
3823 	vc = vcpu->arch.vcore;
3824 	spin_lock(&vc->lock);
3825 	vcpu->arch.ceded = 0;
3826 	vcpu->arch.run_task = current;
3827 	vcpu->arch.kvm_run = kvm_run;
3828 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3829 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3830 	vcpu->arch.busy_preempt = TB_NIL;
3831 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3832 	++vc->n_runnable;
3833 
3834 	/*
3835 	 * This happens the first time this is called for a vcpu.
3836 	 * If the vcore is already running, we may be able to start
3837 	 * this thread straight away and have it join in.
3838 	 */
3839 	if (!signal_pending(current)) {
3840 		if ((vc->vcore_state == VCORE_PIGGYBACK ||
3841 		     vc->vcore_state == VCORE_RUNNING) &&
3842 			   !VCORE_IS_EXITING(vc)) {
3843 			kvmppc_create_dtl_entry(vcpu, vc);
3844 			kvmppc_start_thread(vcpu, vc);
3845 			trace_kvm_guest_enter(vcpu);
3846 		} else if (vc->vcore_state == VCORE_SLEEPING) {
3847 			swake_up_one(&vc->wq);
3848 		}
3849 
3850 	}
3851 
3852 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3853 	       !signal_pending(current)) {
3854 		/* See if the MMU is ready to go */
3855 		if (!vcpu->kvm->arch.mmu_ready) {
3856 			spin_unlock(&vc->lock);
3857 			r = kvmhv_setup_mmu(vcpu);
3858 			spin_lock(&vc->lock);
3859 			if (r) {
3860 				kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3861 				kvm_run->fail_entry.
3862 					hardware_entry_failure_reason = 0;
3863 				vcpu->arch.ret = r;
3864 				break;
3865 			}
3866 		}
3867 
3868 		if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3869 			kvmppc_vcore_end_preempt(vc);
3870 
3871 		if (vc->vcore_state != VCORE_INACTIVE) {
3872 			kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
3873 			continue;
3874 		}
3875 		for_each_runnable_thread(i, v, vc) {
3876 			kvmppc_core_prepare_to_enter(v);
3877 			if (signal_pending(v->arch.run_task)) {
3878 				kvmppc_remove_runnable(vc, v);
3879 				v->stat.signal_exits++;
3880 				v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
3881 				v->arch.ret = -EINTR;
3882 				wake_up(&v->arch.cpu_run);
3883 			}
3884 		}
3885 		if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3886 			break;
3887 		n_ceded = 0;
3888 		for_each_runnable_thread(i, v, vc) {
3889 			if (!kvmppc_vcpu_woken(v))
3890 				n_ceded += v->arch.ceded;
3891 			else
3892 				v->arch.ceded = 0;
3893 		}
3894 		vc->runner = vcpu;
3895 		if (n_ceded == vc->n_runnable) {
3896 			kvmppc_vcore_blocked(vc);
3897 		} else if (need_resched()) {
3898 			kvmppc_vcore_preempt(vc);
3899 			/* Let something else run */
3900 			cond_resched_lock(&vc->lock);
3901 			if (vc->vcore_state == VCORE_PREEMPT)
3902 				kvmppc_vcore_end_preempt(vc);
3903 		} else {
3904 			kvmppc_run_core(vc);
3905 		}
3906 		vc->runner = NULL;
3907 	}
3908 
3909 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3910 	       (vc->vcore_state == VCORE_RUNNING ||
3911 		vc->vcore_state == VCORE_EXITING ||
3912 		vc->vcore_state == VCORE_PIGGYBACK))
3913 		kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
3914 
3915 	if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3916 		kvmppc_vcore_end_preempt(vc);
3917 
3918 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3919 		kvmppc_remove_runnable(vc, vcpu);
3920 		vcpu->stat.signal_exits++;
3921 		kvm_run->exit_reason = KVM_EXIT_INTR;
3922 		vcpu->arch.ret = -EINTR;
3923 	}
3924 
3925 	if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
3926 		/* Wake up some vcpu to run the core */
3927 		i = -1;
3928 		v = next_runnable_thread(vc, &i);
3929 		wake_up(&v->arch.cpu_run);
3930 	}
3931 
3932 	trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
3933 	spin_unlock(&vc->lock);
3934 	return vcpu->arch.ret;
3935 }
3936 
3937 int kvmhv_run_single_vcpu(struct kvm_run *kvm_run,
3938 			  struct kvm_vcpu *vcpu, u64 time_limit,
3939 			  unsigned long lpcr)
3940 {
3941 	int trap, r, pcpu;
3942 	int srcu_idx;
3943 	struct kvmppc_vcore *vc;
3944 	struct kvm *kvm = vcpu->kvm;
3945 	struct kvm_nested_guest *nested = vcpu->arch.nested;
3946 
3947 	trace_kvmppc_run_vcpu_enter(vcpu);
3948 
3949 	kvm_run->exit_reason = 0;
3950 	vcpu->arch.ret = RESUME_GUEST;
3951 	vcpu->arch.trap = 0;
3952 
3953 	vc = vcpu->arch.vcore;
3954 	vcpu->arch.ceded = 0;
3955 	vcpu->arch.run_task = current;
3956 	vcpu->arch.kvm_run = kvm_run;
3957 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3958 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3959 	vcpu->arch.busy_preempt = TB_NIL;
3960 	vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
3961 	vc->runnable_threads[0] = vcpu;
3962 	vc->n_runnable = 1;
3963 	vc->runner = vcpu;
3964 
3965 	/* See if the MMU is ready to go */
3966 	if (!kvm->arch.mmu_ready)
3967 		kvmhv_setup_mmu(vcpu);
3968 
3969 	if (need_resched())
3970 		cond_resched();
3971 
3972 	kvmppc_update_vpas(vcpu);
3973 
3974 	init_vcore_to_run(vc);
3975 	vc->preempt_tb = TB_NIL;
3976 
3977 	preempt_disable();
3978 	pcpu = smp_processor_id();
3979 	vc->pcpu = pcpu;
3980 	kvmppc_prepare_radix_vcpu(vcpu, pcpu);
3981 
3982 	local_irq_disable();
3983 	hard_irq_disable();
3984 	if (signal_pending(current))
3985 		goto sigpend;
3986 	if (lazy_irq_pending() || need_resched() || !kvm->arch.mmu_ready)
3987 		goto out;
3988 
3989 	if (!nested) {
3990 		kvmppc_core_prepare_to_enter(vcpu);
3991 		if (vcpu->arch.doorbell_request) {
3992 			vc->dpdes = 1;
3993 			smp_wmb();
3994 			vcpu->arch.doorbell_request = 0;
3995 		}
3996 		if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
3997 			     &vcpu->arch.pending_exceptions))
3998 			lpcr |= LPCR_MER;
3999 	} else if (vcpu->arch.pending_exceptions ||
4000 		   vcpu->arch.doorbell_request ||
4001 		   xive_interrupt_pending(vcpu)) {
4002 		vcpu->arch.ret = RESUME_HOST;
4003 		goto out;
4004 	}
4005 
4006 	kvmppc_clear_host_core(pcpu);
4007 
4008 	local_paca->kvm_hstate.tid = 0;
4009 	local_paca->kvm_hstate.napping = 0;
4010 	local_paca->kvm_hstate.kvm_split_mode = NULL;
4011 	kvmppc_start_thread(vcpu, vc);
4012 	kvmppc_create_dtl_entry(vcpu, vc);
4013 	trace_kvm_guest_enter(vcpu);
4014 
4015 	vc->vcore_state = VCORE_RUNNING;
4016 	trace_kvmppc_run_core(vc, 0);
4017 
4018 	if (cpu_has_feature(CPU_FTR_HVMODE))
4019 		kvmppc_radix_check_need_tlb_flush(kvm, pcpu, nested);
4020 
4021 	trace_hardirqs_on();
4022 	guest_enter_irqoff();
4023 
4024 	srcu_idx = srcu_read_lock(&kvm->srcu);
4025 
4026 	this_cpu_disable_ftrace();
4027 
4028 	trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr);
4029 	vcpu->arch.trap = trap;
4030 
4031 	this_cpu_enable_ftrace();
4032 
4033 	srcu_read_unlock(&kvm->srcu, srcu_idx);
4034 
4035 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
4036 		mtspr(SPRN_LPID, kvm->arch.host_lpid);
4037 		isync();
4038 	}
4039 
4040 	trace_hardirqs_off();
4041 	set_irq_happened(trap);
4042 
4043 	kvmppc_set_host_core(pcpu);
4044 
4045 	local_irq_enable();
4046 	guest_exit();
4047 
4048 	cpumask_clear_cpu(pcpu, &kvm->arch.cpu_in_guest);
4049 
4050 	preempt_enable();
4051 
4052 	/* cancel pending decrementer exception if DEC is now positive */
4053 	if (get_tb() < vcpu->arch.dec_expires && kvmppc_core_pending_dec(vcpu))
4054 		kvmppc_core_dequeue_dec(vcpu);
4055 
4056 	trace_kvm_guest_exit(vcpu);
4057 	r = RESUME_GUEST;
4058 	if (trap) {
4059 		if (!nested)
4060 			r = kvmppc_handle_exit_hv(kvm_run, vcpu, current);
4061 		else
4062 			r = kvmppc_handle_nested_exit(vcpu);
4063 	}
4064 	vcpu->arch.ret = r;
4065 
4066 	if (is_kvmppc_resume_guest(r) && vcpu->arch.ceded &&
4067 	    !kvmppc_vcpu_woken(vcpu)) {
4068 		kvmppc_set_timer(vcpu);
4069 		while (vcpu->arch.ceded && !kvmppc_vcpu_woken(vcpu)) {
4070 			if (signal_pending(current)) {
4071 				vcpu->stat.signal_exits++;
4072 				kvm_run->exit_reason = KVM_EXIT_INTR;
4073 				vcpu->arch.ret = -EINTR;
4074 				break;
4075 			}
4076 			spin_lock(&vc->lock);
4077 			kvmppc_vcore_blocked(vc);
4078 			spin_unlock(&vc->lock);
4079 		}
4080 	}
4081 	vcpu->arch.ceded = 0;
4082 
4083 	vc->vcore_state = VCORE_INACTIVE;
4084 	trace_kvmppc_run_core(vc, 1);
4085 
4086  done:
4087 	kvmppc_remove_runnable(vc, vcpu);
4088 	trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
4089 
4090 	return vcpu->arch.ret;
4091 
4092  sigpend:
4093 	vcpu->stat.signal_exits++;
4094 	kvm_run->exit_reason = KVM_EXIT_INTR;
4095 	vcpu->arch.ret = -EINTR;
4096  out:
4097 	local_irq_enable();
4098 	preempt_enable();
4099 	goto done;
4100 }
4101 
4102 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
4103 {
4104 	int r;
4105 	int srcu_idx;
4106 	unsigned long ebb_regs[3] = {};	/* shut up GCC */
4107 	unsigned long user_tar = 0;
4108 	unsigned int user_vrsave;
4109 	struct kvm *kvm;
4110 
4111 	if (!vcpu->arch.sane) {
4112 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4113 		return -EINVAL;
4114 	}
4115 
4116 	/*
4117 	 * Don't allow entry with a suspended transaction, because
4118 	 * the guest entry/exit code will lose it.
4119 	 * If the guest has TM enabled, save away their TM-related SPRs
4120 	 * (they will get restored by the TM unavailable interrupt).
4121 	 */
4122 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4123 	if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4124 	    (current->thread.regs->msr & MSR_TM)) {
4125 		if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4126 			run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4127 			run->fail_entry.hardware_entry_failure_reason = 0;
4128 			return -EINVAL;
4129 		}
4130 		/* Enable TM so we can read the TM SPRs */
4131 		mtmsr(mfmsr() | MSR_TM);
4132 		current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
4133 		current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
4134 		current->thread.tm_texasr = mfspr(SPRN_TEXASR);
4135 		current->thread.regs->msr &= ~MSR_TM;
4136 	}
4137 #endif
4138 
4139 	/*
4140 	 * Force online to 1 for the sake of old userspace which doesn't
4141 	 * set it.
4142 	 */
4143 	if (!vcpu->arch.online) {
4144 		atomic_inc(&vcpu->arch.vcore->online_count);
4145 		vcpu->arch.online = 1;
4146 	}
4147 
4148 	kvmppc_core_prepare_to_enter(vcpu);
4149 
4150 	/* No need to go into the guest when all we'll do is come back out */
4151 	if (signal_pending(current)) {
4152 		run->exit_reason = KVM_EXIT_INTR;
4153 		return -EINTR;
4154 	}
4155 
4156 	kvm = vcpu->kvm;
4157 	atomic_inc(&kvm->arch.vcpus_running);
4158 	/* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4159 	smp_mb();
4160 
4161 	flush_all_to_thread(current);
4162 
4163 	/* Save userspace EBB and other register values */
4164 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4165 		ebb_regs[0] = mfspr(SPRN_EBBHR);
4166 		ebb_regs[1] = mfspr(SPRN_EBBRR);
4167 		ebb_regs[2] = mfspr(SPRN_BESCR);
4168 		user_tar = mfspr(SPRN_TAR);
4169 	}
4170 	user_vrsave = mfspr(SPRN_VRSAVE);
4171 
4172 	vcpu->arch.wqp = &vcpu->arch.vcore->wq;
4173 	vcpu->arch.pgdir = current->mm->pgd;
4174 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4175 
4176 	do {
4177 		/*
4178 		 * The early POWER9 chips that can't mix radix and HPT threads
4179 		 * on the same core also need the workaround for the problem
4180 		 * where the TLB would prefetch entries in the guest exit path
4181 		 * for radix guests using the guest PIDR value and LPID 0.
4182 		 * The workaround is in the old path (kvmppc_run_vcpu())
4183 		 * but not the new path (kvmhv_run_single_vcpu()).
4184 		 */
4185 		if (kvm->arch.threads_indep && kvm_is_radix(kvm) &&
4186 		    !no_mixing_hpt_and_radix)
4187 			r = kvmhv_run_single_vcpu(run, vcpu, ~(u64)0,
4188 						  vcpu->arch.vcore->lpcr);
4189 		else
4190 			r = kvmppc_run_vcpu(run, vcpu);
4191 
4192 		if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
4193 		    !(vcpu->arch.shregs.msr & MSR_PR)) {
4194 			trace_kvm_hcall_enter(vcpu);
4195 			r = kvmppc_pseries_do_hcall(vcpu);
4196 			trace_kvm_hcall_exit(vcpu, r);
4197 			kvmppc_core_prepare_to_enter(vcpu);
4198 		} else if (r == RESUME_PAGE_FAULT) {
4199 			srcu_idx = srcu_read_lock(&kvm->srcu);
4200 			r = kvmppc_book3s_hv_page_fault(run, vcpu,
4201 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4202 			srcu_read_unlock(&kvm->srcu, srcu_idx);
4203 		} else if (r == RESUME_PASSTHROUGH) {
4204 			if (WARN_ON(xive_enabled()))
4205 				r = H_SUCCESS;
4206 			else
4207 				r = kvmppc_xics_rm_complete(vcpu, 0);
4208 		}
4209 	} while (is_kvmppc_resume_guest(r));
4210 
4211 	/* Restore userspace EBB and other register values */
4212 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4213 		mtspr(SPRN_EBBHR, ebb_regs[0]);
4214 		mtspr(SPRN_EBBRR, ebb_regs[1]);
4215 		mtspr(SPRN_BESCR, ebb_regs[2]);
4216 		mtspr(SPRN_TAR, user_tar);
4217 		mtspr(SPRN_FSCR, current->thread.fscr);
4218 	}
4219 	mtspr(SPRN_VRSAVE, user_vrsave);
4220 
4221 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4222 	atomic_dec(&kvm->arch.vcpus_running);
4223 	return r;
4224 }
4225 
4226 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4227 				     int shift, int sllp)
4228 {
4229 	(*sps)->page_shift = shift;
4230 	(*sps)->slb_enc = sllp;
4231 	(*sps)->enc[0].page_shift = shift;
4232 	(*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4233 	/*
4234 	 * Add 16MB MPSS support (may get filtered out by userspace)
4235 	 */
4236 	if (shift != 24) {
4237 		int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4238 		if (penc != -1) {
4239 			(*sps)->enc[1].page_shift = 24;
4240 			(*sps)->enc[1].pte_enc = penc;
4241 		}
4242 	}
4243 	(*sps)++;
4244 }
4245 
4246 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4247 					 struct kvm_ppc_smmu_info *info)
4248 {
4249 	struct kvm_ppc_one_seg_page_size *sps;
4250 
4251 	/*
4252 	 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4253 	 * POWER7 doesn't support keys for instruction accesses,
4254 	 * POWER8 and POWER9 do.
4255 	 */
4256 	info->data_keys = 32;
4257 	info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4258 
4259 	/* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4260 	info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4261 	info->slb_size = 32;
4262 
4263 	/* We only support these sizes for now, and no muti-size segments */
4264 	sps = &info->sps[0];
4265 	kvmppc_add_seg_page_size(&sps, 12, 0);
4266 	kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4267 	kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4268 
4269 	/* If running as a nested hypervisor, we don't support HPT guests */
4270 	if (kvmhv_on_pseries())
4271 		info->flags |= KVM_PPC_NO_HASH;
4272 
4273 	return 0;
4274 }
4275 
4276 /*
4277  * Get (and clear) the dirty memory log for a memory slot.
4278  */
4279 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4280 					 struct kvm_dirty_log *log)
4281 {
4282 	struct kvm_memslots *slots;
4283 	struct kvm_memory_slot *memslot;
4284 	int i, r;
4285 	unsigned long n;
4286 	unsigned long *buf, *p;
4287 	struct kvm_vcpu *vcpu;
4288 
4289 	mutex_lock(&kvm->slots_lock);
4290 
4291 	r = -EINVAL;
4292 	if (log->slot >= KVM_USER_MEM_SLOTS)
4293 		goto out;
4294 
4295 	slots = kvm_memslots(kvm);
4296 	memslot = id_to_memslot(slots, log->slot);
4297 	r = -ENOENT;
4298 	if (!memslot->dirty_bitmap)
4299 		goto out;
4300 
4301 	/*
4302 	 * Use second half of bitmap area because both HPT and radix
4303 	 * accumulate bits in the first half.
4304 	 */
4305 	n = kvm_dirty_bitmap_bytes(memslot);
4306 	buf = memslot->dirty_bitmap + n / sizeof(long);
4307 	memset(buf, 0, n);
4308 
4309 	if (kvm_is_radix(kvm))
4310 		r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4311 	else
4312 		r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4313 	if (r)
4314 		goto out;
4315 
4316 	/*
4317 	 * We accumulate dirty bits in the first half of the
4318 	 * memslot's dirty_bitmap area, for when pages are paged
4319 	 * out or modified by the host directly.  Pick up these
4320 	 * bits and add them to the map.
4321 	 */
4322 	p = memslot->dirty_bitmap;
4323 	for (i = 0; i < n / sizeof(long); ++i)
4324 		buf[i] |= xchg(&p[i], 0);
4325 
4326 	/* Harvest dirty bits from VPA and DTL updates */
4327 	/* Note: we never modify the SLB shadow buffer areas */
4328 	kvm_for_each_vcpu(i, vcpu, kvm) {
4329 		spin_lock(&vcpu->arch.vpa_update_lock);
4330 		kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4331 		kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4332 		spin_unlock(&vcpu->arch.vpa_update_lock);
4333 	}
4334 
4335 	r = -EFAULT;
4336 	if (copy_to_user(log->dirty_bitmap, buf, n))
4337 		goto out;
4338 
4339 	r = 0;
4340 out:
4341 	mutex_unlock(&kvm->slots_lock);
4342 	return r;
4343 }
4344 
4345 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
4346 					struct kvm_memory_slot *dont)
4347 {
4348 	if (!dont || free->arch.rmap != dont->arch.rmap) {
4349 		vfree(free->arch.rmap);
4350 		free->arch.rmap = NULL;
4351 	}
4352 }
4353 
4354 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
4355 					 unsigned long npages)
4356 {
4357 	slot->arch.rmap = vzalloc(array_size(npages, sizeof(*slot->arch.rmap)));
4358 	if (!slot->arch.rmap)
4359 		return -ENOMEM;
4360 
4361 	return 0;
4362 }
4363 
4364 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4365 					struct kvm_memory_slot *memslot,
4366 					const struct kvm_userspace_memory_region *mem)
4367 {
4368 	return 0;
4369 }
4370 
4371 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
4372 				const struct kvm_userspace_memory_region *mem,
4373 				const struct kvm_memory_slot *old,
4374 				const struct kvm_memory_slot *new)
4375 {
4376 	unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4377 
4378 	/*
4379 	 * If we are making a new memslot, it might make
4380 	 * some address that was previously cached as emulated
4381 	 * MMIO be no longer emulated MMIO, so invalidate
4382 	 * all the caches of emulated MMIO translations.
4383 	 */
4384 	if (npages)
4385 		atomic64_inc(&kvm->arch.mmio_update);
4386 }
4387 
4388 /*
4389  * Update LPCR values in kvm->arch and in vcores.
4390  * Caller must hold kvm->lock.
4391  */
4392 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
4393 {
4394 	long int i;
4395 	u32 cores_done = 0;
4396 
4397 	if ((kvm->arch.lpcr & mask) == lpcr)
4398 		return;
4399 
4400 	kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
4401 
4402 	for (i = 0; i < KVM_MAX_VCORES; ++i) {
4403 		struct kvmppc_vcore *vc = kvm->arch.vcores[i];
4404 		if (!vc)
4405 			continue;
4406 		spin_lock(&vc->lock);
4407 		vc->lpcr = (vc->lpcr & ~mask) | lpcr;
4408 		spin_unlock(&vc->lock);
4409 		if (++cores_done >= kvm->arch.online_vcores)
4410 			break;
4411 	}
4412 }
4413 
4414 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
4415 {
4416 	return;
4417 }
4418 
4419 void kvmppc_setup_partition_table(struct kvm *kvm)
4420 {
4421 	unsigned long dw0, dw1;
4422 
4423 	if (!kvm_is_radix(kvm)) {
4424 		/* PS field - page size for VRMA */
4425 		dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
4426 			((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
4427 		/* HTABSIZE and HTABORG fields */
4428 		dw0 |= kvm->arch.sdr1;
4429 
4430 		/* Second dword as set by userspace */
4431 		dw1 = kvm->arch.process_table;
4432 	} else {
4433 		dw0 = PATB_HR | radix__get_tree_size() |
4434 			__pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
4435 		dw1 = PATB_GR | kvm->arch.process_table;
4436 	}
4437 	kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
4438 }
4439 
4440 /*
4441  * Set up HPT (hashed page table) and RMA (real-mode area).
4442  * Must be called with kvm->lock held.
4443  */
4444 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
4445 {
4446 	int err = 0;
4447 	struct kvm *kvm = vcpu->kvm;
4448 	unsigned long hva;
4449 	struct kvm_memory_slot *memslot;
4450 	struct vm_area_struct *vma;
4451 	unsigned long lpcr = 0, senc;
4452 	unsigned long psize, porder;
4453 	int srcu_idx;
4454 
4455 	/* Allocate hashed page table (if not done already) and reset it */
4456 	if (!kvm->arch.hpt.virt) {
4457 		int order = KVM_DEFAULT_HPT_ORDER;
4458 		struct kvm_hpt_info info;
4459 
4460 		err = kvmppc_allocate_hpt(&info, order);
4461 		/* If we get here, it means userspace didn't specify a
4462 		 * size explicitly.  So, try successively smaller
4463 		 * sizes if the default failed. */
4464 		while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
4465 			err  = kvmppc_allocate_hpt(&info, order);
4466 
4467 		if (err < 0) {
4468 			pr_err("KVM: Couldn't alloc HPT\n");
4469 			goto out;
4470 		}
4471 
4472 		kvmppc_set_hpt(kvm, &info);
4473 	}
4474 
4475 	/* Look up the memslot for guest physical address 0 */
4476 	srcu_idx = srcu_read_lock(&kvm->srcu);
4477 	memslot = gfn_to_memslot(kvm, 0);
4478 
4479 	/* We must have some memory at 0 by now */
4480 	err = -EINVAL;
4481 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
4482 		goto out_srcu;
4483 
4484 	/* Look up the VMA for the start of this memory slot */
4485 	hva = memslot->userspace_addr;
4486 	down_read(&current->mm->mmap_sem);
4487 	vma = find_vma(current->mm, hva);
4488 	if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
4489 		goto up_out;
4490 
4491 	psize = vma_kernel_pagesize(vma);
4492 
4493 	up_read(&current->mm->mmap_sem);
4494 
4495 	/* We can handle 4k, 64k or 16M pages in the VRMA */
4496 	if (psize >= 0x1000000)
4497 		psize = 0x1000000;
4498 	else if (psize >= 0x10000)
4499 		psize = 0x10000;
4500 	else
4501 		psize = 0x1000;
4502 	porder = __ilog2(psize);
4503 
4504 	senc = slb_pgsize_encoding(psize);
4505 	kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
4506 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
4507 	/* Create HPTEs in the hash page table for the VRMA */
4508 	kvmppc_map_vrma(vcpu, memslot, porder);
4509 
4510 	/* Update VRMASD field in the LPCR */
4511 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
4512 		/* the -4 is to account for senc values starting at 0x10 */
4513 		lpcr = senc << (LPCR_VRMASD_SH - 4);
4514 		kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
4515 	}
4516 
4517 	/* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
4518 	smp_wmb();
4519 	err = 0;
4520  out_srcu:
4521 	srcu_read_unlock(&kvm->srcu, srcu_idx);
4522  out:
4523 	return err;
4524 
4525  up_out:
4526 	up_read(&current->mm->mmap_sem);
4527 	goto out_srcu;
4528 }
4529 
4530 /* Must be called with kvm->lock held and mmu_ready = 0 and no vcpus running */
4531 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
4532 {
4533 	if (nesting_enabled(kvm))
4534 		kvmhv_release_all_nested(kvm);
4535 	kvmppc_free_radix(kvm);
4536 	kvmppc_update_lpcr(kvm, LPCR_VPM1,
4537 			   LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4538 	kvmppc_rmap_reset(kvm);
4539 	kvm->arch.radix = 0;
4540 	kvm->arch.process_table = 0;
4541 	return 0;
4542 }
4543 
4544 /* Must be called with kvm->lock held and mmu_ready = 0 and no vcpus running */
4545 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
4546 {
4547 	int err;
4548 
4549 	err = kvmppc_init_vm_radix(kvm);
4550 	if (err)
4551 		return err;
4552 
4553 	kvmppc_free_hpt(&kvm->arch.hpt);
4554 	kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
4555 			   LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4556 	kvmppc_rmap_reset(kvm);
4557 	kvm->arch.radix = 1;
4558 	return 0;
4559 }
4560 
4561 #ifdef CONFIG_KVM_XICS
4562 /*
4563  * Allocate a per-core structure for managing state about which cores are
4564  * running in the host versus the guest and for exchanging data between
4565  * real mode KVM and CPU running in the host.
4566  * This is only done for the first VM.
4567  * The allocated structure stays even if all VMs have stopped.
4568  * It is only freed when the kvm-hv module is unloaded.
4569  * It's OK for this routine to fail, we just don't support host
4570  * core operations like redirecting H_IPI wakeups.
4571  */
4572 void kvmppc_alloc_host_rm_ops(void)
4573 {
4574 	struct kvmppc_host_rm_ops *ops;
4575 	unsigned long l_ops;
4576 	int cpu, core;
4577 	int size;
4578 
4579 	/* Not the first time here ? */
4580 	if (kvmppc_host_rm_ops_hv != NULL)
4581 		return;
4582 
4583 	ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
4584 	if (!ops)
4585 		return;
4586 
4587 	size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
4588 	ops->rm_core = kzalloc(size, GFP_KERNEL);
4589 
4590 	if (!ops->rm_core) {
4591 		kfree(ops);
4592 		return;
4593 	}
4594 
4595 	cpus_read_lock();
4596 
4597 	for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
4598 		if (!cpu_online(cpu))
4599 			continue;
4600 
4601 		core = cpu >> threads_shift;
4602 		ops->rm_core[core].rm_state.in_host = 1;
4603 	}
4604 
4605 	ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
4606 
4607 	/*
4608 	 * Make the contents of the kvmppc_host_rm_ops structure visible
4609 	 * to other CPUs before we assign it to the global variable.
4610 	 * Do an atomic assignment (no locks used here), but if someone
4611 	 * beats us to it, just free our copy and return.
4612 	 */
4613 	smp_wmb();
4614 	l_ops = (unsigned long) ops;
4615 
4616 	if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
4617 		cpus_read_unlock();
4618 		kfree(ops->rm_core);
4619 		kfree(ops);
4620 		return;
4621 	}
4622 
4623 	cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
4624 					     "ppc/kvm_book3s:prepare",
4625 					     kvmppc_set_host_core,
4626 					     kvmppc_clear_host_core);
4627 	cpus_read_unlock();
4628 }
4629 
4630 void kvmppc_free_host_rm_ops(void)
4631 {
4632 	if (kvmppc_host_rm_ops_hv) {
4633 		cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
4634 		kfree(kvmppc_host_rm_ops_hv->rm_core);
4635 		kfree(kvmppc_host_rm_ops_hv);
4636 		kvmppc_host_rm_ops_hv = NULL;
4637 	}
4638 }
4639 #endif
4640 
4641 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
4642 {
4643 	unsigned long lpcr, lpid;
4644 	char buf[32];
4645 	int ret;
4646 
4647 	/* Allocate the guest's logical partition ID */
4648 
4649 	lpid = kvmppc_alloc_lpid();
4650 	if ((long)lpid < 0)
4651 		return -ENOMEM;
4652 	kvm->arch.lpid = lpid;
4653 
4654 	kvmppc_alloc_host_rm_ops();
4655 
4656 	kvmhv_vm_nested_init(kvm);
4657 
4658 	/*
4659 	 * Since we don't flush the TLB when tearing down a VM,
4660 	 * and this lpid might have previously been used,
4661 	 * make sure we flush on each core before running the new VM.
4662 	 * On POWER9, the tlbie in mmu_partition_table_set_entry()
4663 	 * does this flush for us.
4664 	 */
4665 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
4666 		cpumask_setall(&kvm->arch.need_tlb_flush);
4667 
4668 	/* Start out with the default set of hcalls enabled */
4669 	memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
4670 	       sizeof(kvm->arch.enabled_hcalls));
4671 
4672 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
4673 		kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
4674 
4675 	/* Init LPCR for virtual RMA mode */
4676 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
4677 		kvm->arch.host_lpid = mfspr(SPRN_LPID);
4678 		kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
4679 		lpcr &= LPCR_PECE | LPCR_LPES;
4680 	} else {
4681 		lpcr = 0;
4682 	}
4683 	lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
4684 		LPCR_VPM0 | LPCR_VPM1;
4685 	kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
4686 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
4687 	/* On POWER8 turn on online bit to enable PURR/SPURR */
4688 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
4689 		lpcr |= LPCR_ONL;
4690 	/*
4691 	 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
4692 	 * Set HVICE bit to enable hypervisor virtualization interrupts.
4693 	 * Set HEIC to prevent OS interrupts to go to hypervisor (should
4694 	 * be unnecessary but better safe than sorry in case we re-enable
4695 	 * EE in HV mode with this LPCR still set)
4696 	 */
4697 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4698 		lpcr &= ~LPCR_VPM0;
4699 		lpcr |= LPCR_HVICE | LPCR_HEIC;
4700 
4701 		/*
4702 		 * If xive is enabled, we route 0x500 interrupts directly
4703 		 * to the guest.
4704 		 */
4705 		if (xive_enabled())
4706 			lpcr |= LPCR_LPES;
4707 	}
4708 
4709 	/*
4710 	 * If the host uses radix, the guest starts out as radix.
4711 	 */
4712 	if (radix_enabled()) {
4713 		kvm->arch.radix = 1;
4714 		kvm->arch.mmu_ready = 1;
4715 		lpcr &= ~LPCR_VPM1;
4716 		lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
4717 		ret = kvmppc_init_vm_radix(kvm);
4718 		if (ret) {
4719 			kvmppc_free_lpid(kvm->arch.lpid);
4720 			return ret;
4721 		}
4722 		kvmppc_setup_partition_table(kvm);
4723 	}
4724 
4725 	kvm->arch.lpcr = lpcr;
4726 
4727 	/* Initialization for future HPT resizes */
4728 	kvm->arch.resize_hpt = NULL;
4729 
4730 	/*
4731 	 * Work out how many sets the TLB has, for the use of
4732 	 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
4733 	 */
4734 	if (radix_enabled())
4735 		kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;	/* 128 */
4736 	else if (cpu_has_feature(CPU_FTR_ARCH_300))
4737 		kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;	/* 256 */
4738 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
4739 		kvm->arch.tlb_sets = POWER8_TLB_SETS;		/* 512 */
4740 	else
4741 		kvm->arch.tlb_sets = POWER7_TLB_SETS;		/* 128 */
4742 
4743 	/*
4744 	 * Track that we now have a HV mode VM active. This blocks secondary
4745 	 * CPU threads from coming online.
4746 	 * On POWER9, we only need to do this if the "indep_threads_mode"
4747 	 * module parameter has been set to N.
4748 	 */
4749 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4750 		if (!indep_threads_mode && !cpu_has_feature(CPU_FTR_HVMODE)) {
4751 			pr_warn("KVM: Ignoring indep_threads_mode=N in nested hypervisor\n");
4752 			kvm->arch.threads_indep = true;
4753 		} else {
4754 			kvm->arch.threads_indep = indep_threads_mode;
4755 		}
4756 	}
4757 	if (!kvm->arch.threads_indep)
4758 		kvm_hv_vm_activated();
4759 
4760 	/*
4761 	 * Initialize smt_mode depending on processor.
4762 	 * POWER8 and earlier have to use "strict" threading, where
4763 	 * all vCPUs in a vcore have to run on the same (sub)core,
4764 	 * whereas on POWER9 the threads can each run a different
4765 	 * guest.
4766 	 */
4767 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
4768 		kvm->arch.smt_mode = threads_per_subcore;
4769 	else
4770 		kvm->arch.smt_mode = 1;
4771 	kvm->arch.emul_smt_mode = 1;
4772 
4773 	/*
4774 	 * Create a debugfs directory for the VM
4775 	 */
4776 	snprintf(buf, sizeof(buf), "vm%d", current->pid);
4777 	kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
4778 	kvmppc_mmu_debugfs_init(kvm);
4779 	if (radix_enabled())
4780 		kvmhv_radix_debugfs_init(kvm);
4781 
4782 	return 0;
4783 }
4784 
4785 static void kvmppc_free_vcores(struct kvm *kvm)
4786 {
4787 	long int i;
4788 
4789 	for (i = 0; i < KVM_MAX_VCORES; ++i)
4790 		kfree(kvm->arch.vcores[i]);
4791 	kvm->arch.online_vcores = 0;
4792 }
4793 
4794 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
4795 {
4796 	debugfs_remove_recursive(kvm->arch.debugfs_dir);
4797 
4798 	if (!kvm->arch.threads_indep)
4799 		kvm_hv_vm_deactivated();
4800 
4801 	kvmppc_free_vcores(kvm);
4802 
4803 
4804 	if (kvm_is_radix(kvm))
4805 		kvmppc_free_radix(kvm);
4806 	else
4807 		kvmppc_free_hpt(&kvm->arch.hpt);
4808 
4809 	/* Perform global invalidation and return lpid to the pool */
4810 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4811 		if (nesting_enabled(kvm))
4812 			kvmhv_release_all_nested(kvm);
4813 		kvm->arch.process_table = 0;
4814 		kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
4815 	}
4816 	kvmppc_free_lpid(kvm->arch.lpid);
4817 
4818 	kvmppc_free_pimap(kvm);
4819 }
4820 
4821 /* We don't need to emulate any privileged instructions or dcbz */
4822 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
4823 				     unsigned int inst, int *advance)
4824 {
4825 	return EMULATE_FAIL;
4826 }
4827 
4828 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
4829 					ulong spr_val)
4830 {
4831 	return EMULATE_FAIL;
4832 }
4833 
4834 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
4835 					ulong *spr_val)
4836 {
4837 	return EMULATE_FAIL;
4838 }
4839 
4840 static int kvmppc_core_check_processor_compat_hv(void)
4841 {
4842 	if (cpu_has_feature(CPU_FTR_HVMODE) &&
4843 	    cpu_has_feature(CPU_FTR_ARCH_206))
4844 		return 0;
4845 
4846 	/* POWER9 in radix mode is capable of being a nested hypervisor. */
4847 	if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
4848 		return 0;
4849 
4850 	return -EIO;
4851 }
4852 
4853 #ifdef CONFIG_KVM_XICS
4854 
4855 void kvmppc_free_pimap(struct kvm *kvm)
4856 {
4857 	kfree(kvm->arch.pimap);
4858 }
4859 
4860 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
4861 {
4862 	return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
4863 }
4864 
4865 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4866 {
4867 	struct irq_desc *desc;
4868 	struct kvmppc_irq_map *irq_map;
4869 	struct kvmppc_passthru_irqmap *pimap;
4870 	struct irq_chip *chip;
4871 	int i, rc = 0;
4872 
4873 	if (!kvm_irq_bypass)
4874 		return 1;
4875 
4876 	desc = irq_to_desc(host_irq);
4877 	if (!desc)
4878 		return -EIO;
4879 
4880 	mutex_lock(&kvm->lock);
4881 
4882 	pimap = kvm->arch.pimap;
4883 	if (pimap == NULL) {
4884 		/* First call, allocate structure to hold IRQ map */
4885 		pimap = kvmppc_alloc_pimap();
4886 		if (pimap == NULL) {
4887 			mutex_unlock(&kvm->lock);
4888 			return -ENOMEM;
4889 		}
4890 		kvm->arch.pimap = pimap;
4891 	}
4892 
4893 	/*
4894 	 * For now, we only support interrupts for which the EOI operation
4895 	 * is an OPAL call followed by a write to XIRR, since that's
4896 	 * what our real-mode EOI code does, or a XIVE interrupt
4897 	 */
4898 	chip = irq_data_get_irq_chip(&desc->irq_data);
4899 	if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
4900 		pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
4901 			host_irq, guest_gsi);
4902 		mutex_unlock(&kvm->lock);
4903 		return -ENOENT;
4904 	}
4905 
4906 	/*
4907 	 * See if we already have an entry for this guest IRQ number.
4908 	 * If it's mapped to a hardware IRQ number, that's an error,
4909 	 * otherwise re-use this entry.
4910 	 */
4911 	for (i = 0; i < pimap->n_mapped; i++) {
4912 		if (guest_gsi == pimap->mapped[i].v_hwirq) {
4913 			if (pimap->mapped[i].r_hwirq) {
4914 				mutex_unlock(&kvm->lock);
4915 				return -EINVAL;
4916 			}
4917 			break;
4918 		}
4919 	}
4920 
4921 	if (i == KVMPPC_PIRQ_MAPPED) {
4922 		mutex_unlock(&kvm->lock);
4923 		return -EAGAIN;		/* table is full */
4924 	}
4925 
4926 	irq_map = &pimap->mapped[i];
4927 
4928 	irq_map->v_hwirq = guest_gsi;
4929 	irq_map->desc = desc;
4930 
4931 	/*
4932 	 * Order the above two stores before the next to serialize with
4933 	 * the KVM real mode handler.
4934 	 */
4935 	smp_wmb();
4936 	irq_map->r_hwirq = desc->irq_data.hwirq;
4937 
4938 	if (i == pimap->n_mapped)
4939 		pimap->n_mapped++;
4940 
4941 	if (xive_enabled())
4942 		rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
4943 	else
4944 		kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
4945 	if (rc)
4946 		irq_map->r_hwirq = 0;
4947 
4948 	mutex_unlock(&kvm->lock);
4949 
4950 	return 0;
4951 }
4952 
4953 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4954 {
4955 	struct irq_desc *desc;
4956 	struct kvmppc_passthru_irqmap *pimap;
4957 	int i, rc = 0;
4958 
4959 	if (!kvm_irq_bypass)
4960 		return 0;
4961 
4962 	desc = irq_to_desc(host_irq);
4963 	if (!desc)
4964 		return -EIO;
4965 
4966 	mutex_lock(&kvm->lock);
4967 	if (!kvm->arch.pimap)
4968 		goto unlock;
4969 
4970 	pimap = kvm->arch.pimap;
4971 
4972 	for (i = 0; i < pimap->n_mapped; i++) {
4973 		if (guest_gsi == pimap->mapped[i].v_hwirq)
4974 			break;
4975 	}
4976 
4977 	if (i == pimap->n_mapped) {
4978 		mutex_unlock(&kvm->lock);
4979 		return -ENODEV;
4980 	}
4981 
4982 	if (xive_enabled())
4983 		rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
4984 	else
4985 		kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
4986 
4987 	/* invalidate the entry (what do do on error from the above ?) */
4988 	pimap->mapped[i].r_hwirq = 0;
4989 
4990 	/*
4991 	 * We don't free this structure even when the count goes to
4992 	 * zero. The structure is freed when we destroy the VM.
4993 	 */
4994  unlock:
4995 	mutex_unlock(&kvm->lock);
4996 	return rc;
4997 }
4998 
4999 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5000 					     struct irq_bypass_producer *prod)
5001 {
5002 	int ret = 0;
5003 	struct kvm_kernel_irqfd *irqfd =
5004 		container_of(cons, struct kvm_kernel_irqfd, consumer);
5005 
5006 	irqfd->producer = prod;
5007 
5008 	ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5009 	if (ret)
5010 		pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5011 			prod->irq, irqfd->gsi, ret);
5012 
5013 	return ret;
5014 }
5015 
5016 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5017 					      struct irq_bypass_producer *prod)
5018 {
5019 	int ret;
5020 	struct kvm_kernel_irqfd *irqfd =
5021 		container_of(cons, struct kvm_kernel_irqfd, consumer);
5022 
5023 	irqfd->producer = NULL;
5024 
5025 	/*
5026 	 * When producer of consumer is unregistered, we change back to
5027 	 * default external interrupt handling mode - KVM real mode
5028 	 * will switch back to host.
5029 	 */
5030 	ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5031 	if (ret)
5032 		pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5033 			prod->irq, irqfd->gsi, ret);
5034 }
5035 #endif
5036 
5037 static long kvm_arch_vm_ioctl_hv(struct file *filp,
5038 				 unsigned int ioctl, unsigned long arg)
5039 {
5040 	struct kvm *kvm __maybe_unused = filp->private_data;
5041 	void __user *argp = (void __user *)arg;
5042 	long r;
5043 
5044 	switch (ioctl) {
5045 
5046 	case KVM_PPC_ALLOCATE_HTAB: {
5047 		u32 htab_order;
5048 
5049 		r = -EFAULT;
5050 		if (get_user(htab_order, (u32 __user *)argp))
5051 			break;
5052 		r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5053 		if (r)
5054 			break;
5055 		r = 0;
5056 		break;
5057 	}
5058 
5059 	case KVM_PPC_GET_HTAB_FD: {
5060 		struct kvm_get_htab_fd ghf;
5061 
5062 		r = -EFAULT;
5063 		if (copy_from_user(&ghf, argp, sizeof(ghf)))
5064 			break;
5065 		r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5066 		break;
5067 	}
5068 
5069 	case KVM_PPC_RESIZE_HPT_PREPARE: {
5070 		struct kvm_ppc_resize_hpt rhpt;
5071 
5072 		r = -EFAULT;
5073 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5074 			break;
5075 
5076 		r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5077 		break;
5078 	}
5079 
5080 	case KVM_PPC_RESIZE_HPT_COMMIT: {
5081 		struct kvm_ppc_resize_hpt rhpt;
5082 
5083 		r = -EFAULT;
5084 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5085 			break;
5086 
5087 		r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5088 		break;
5089 	}
5090 
5091 	default:
5092 		r = -ENOTTY;
5093 	}
5094 
5095 	return r;
5096 }
5097 
5098 /*
5099  * List of hcall numbers to enable by default.
5100  * For compatibility with old userspace, we enable by default
5101  * all hcalls that were implemented before the hcall-enabling
5102  * facility was added.  Note this list should not include H_RTAS.
5103  */
5104 static unsigned int default_hcall_list[] = {
5105 	H_REMOVE,
5106 	H_ENTER,
5107 	H_READ,
5108 	H_PROTECT,
5109 	H_BULK_REMOVE,
5110 	H_GET_TCE,
5111 	H_PUT_TCE,
5112 	H_SET_DABR,
5113 	H_SET_XDABR,
5114 	H_CEDE,
5115 	H_PROD,
5116 	H_CONFER,
5117 	H_REGISTER_VPA,
5118 #ifdef CONFIG_KVM_XICS
5119 	H_EOI,
5120 	H_CPPR,
5121 	H_IPI,
5122 	H_IPOLL,
5123 	H_XIRR,
5124 	H_XIRR_X,
5125 #endif
5126 	0
5127 };
5128 
5129 static void init_default_hcalls(void)
5130 {
5131 	int i;
5132 	unsigned int hcall;
5133 
5134 	for (i = 0; default_hcall_list[i]; ++i) {
5135 		hcall = default_hcall_list[i];
5136 		WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5137 		__set_bit(hcall / 4, default_enabled_hcalls);
5138 	}
5139 }
5140 
5141 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5142 {
5143 	unsigned long lpcr;
5144 	int radix;
5145 	int err;
5146 
5147 	/* If not on a POWER9, reject it */
5148 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5149 		return -ENODEV;
5150 
5151 	/* If any unknown flags set, reject it */
5152 	if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5153 		return -EINVAL;
5154 
5155 	/* GR (guest radix) bit in process_table field must match */
5156 	radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5157 	if (!!(cfg->process_table & PATB_GR) != radix)
5158 		return -EINVAL;
5159 
5160 	/* Process table size field must be reasonable, i.e. <= 24 */
5161 	if ((cfg->process_table & PRTS_MASK) > 24)
5162 		return -EINVAL;
5163 
5164 	/* We can change a guest to/from radix now, if the host is radix */
5165 	if (radix && !radix_enabled())
5166 		return -EINVAL;
5167 
5168 	/* If we're a nested hypervisor, we currently only support radix */
5169 	if (kvmhv_on_pseries() && !radix)
5170 		return -EINVAL;
5171 
5172 	mutex_lock(&kvm->lock);
5173 	if (radix != kvm_is_radix(kvm)) {
5174 		if (kvm->arch.mmu_ready) {
5175 			kvm->arch.mmu_ready = 0;
5176 			/* order mmu_ready vs. vcpus_running */
5177 			smp_mb();
5178 			if (atomic_read(&kvm->arch.vcpus_running)) {
5179 				kvm->arch.mmu_ready = 1;
5180 				err = -EBUSY;
5181 				goto out_unlock;
5182 			}
5183 		}
5184 		if (radix)
5185 			err = kvmppc_switch_mmu_to_radix(kvm);
5186 		else
5187 			err = kvmppc_switch_mmu_to_hpt(kvm);
5188 		if (err)
5189 			goto out_unlock;
5190 	}
5191 
5192 	kvm->arch.process_table = cfg->process_table;
5193 	kvmppc_setup_partition_table(kvm);
5194 
5195 	lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5196 	kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5197 	err = 0;
5198 
5199  out_unlock:
5200 	mutex_unlock(&kvm->lock);
5201 	return err;
5202 }
5203 
5204 static int kvmhv_enable_nested(struct kvm *kvm)
5205 {
5206 	if (!nested)
5207 		return -EPERM;
5208 	if (!cpu_has_feature(CPU_FTR_ARCH_300) || no_mixing_hpt_and_radix)
5209 		return -ENODEV;
5210 
5211 	/* kvm == NULL means the caller is testing if the capability exists */
5212 	if (kvm)
5213 		kvm->arch.nested_enable = true;
5214 	return 0;
5215 }
5216 
5217 static struct kvmppc_ops kvm_ops_hv = {
5218 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
5219 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
5220 	.get_one_reg = kvmppc_get_one_reg_hv,
5221 	.set_one_reg = kvmppc_set_one_reg_hv,
5222 	.vcpu_load   = kvmppc_core_vcpu_load_hv,
5223 	.vcpu_put    = kvmppc_core_vcpu_put_hv,
5224 	.set_msr     = kvmppc_set_msr_hv,
5225 	.vcpu_run    = kvmppc_vcpu_run_hv,
5226 	.vcpu_create = kvmppc_core_vcpu_create_hv,
5227 	.vcpu_free   = kvmppc_core_vcpu_free_hv,
5228 	.check_requests = kvmppc_core_check_requests_hv,
5229 	.get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
5230 	.flush_memslot  = kvmppc_core_flush_memslot_hv,
5231 	.prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
5232 	.commit_memory_region  = kvmppc_core_commit_memory_region_hv,
5233 	.unmap_hva_range = kvm_unmap_hva_range_hv,
5234 	.age_hva  = kvm_age_hva_hv,
5235 	.test_age_hva = kvm_test_age_hva_hv,
5236 	.set_spte_hva = kvm_set_spte_hva_hv,
5237 	.mmu_destroy  = kvmppc_mmu_destroy_hv,
5238 	.free_memslot = kvmppc_core_free_memslot_hv,
5239 	.create_memslot = kvmppc_core_create_memslot_hv,
5240 	.init_vm =  kvmppc_core_init_vm_hv,
5241 	.destroy_vm = kvmppc_core_destroy_vm_hv,
5242 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
5243 	.emulate_op = kvmppc_core_emulate_op_hv,
5244 	.emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
5245 	.emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
5246 	.fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
5247 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
5248 	.hcall_implemented = kvmppc_hcall_impl_hv,
5249 #ifdef CONFIG_KVM_XICS
5250 	.irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
5251 	.irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
5252 #endif
5253 	.configure_mmu = kvmhv_configure_mmu,
5254 	.get_rmmu_info = kvmhv_get_rmmu_info,
5255 	.set_smt_mode = kvmhv_set_smt_mode,
5256 	.enable_nested = kvmhv_enable_nested,
5257 };
5258 
5259 static int kvm_init_subcore_bitmap(void)
5260 {
5261 	int i, j;
5262 	int nr_cores = cpu_nr_cores();
5263 	struct sibling_subcore_state *sibling_subcore_state;
5264 
5265 	for (i = 0; i < nr_cores; i++) {
5266 		int first_cpu = i * threads_per_core;
5267 		int node = cpu_to_node(first_cpu);
5268 
5269 		/* Ignore if it is already allocated. */
5270 		if (paca_ptrs[first_cpu]->sibling_subcore_state)
5271 			continue;
5272 
5273 		sibling_subcore_state =
5274 			kmalloc_node(sizeof(struct sibling_subcore_state),
5275 							GFP_KERNEL, node);
5276 		if (!sibling_subcore_state)
5277 			return -ENOMEM;
5278 
5279 		memset(sibling_subcore_state, 0,
5280 				sizeof(struct sibling_subcore_state));
5281 
5282 		for (j = 0; j < threads_per_core; j++) {
5283 			int cpu = first_cpu + j;
5284 
5285 			paca_ptrs[cpu]->sibling_subcore_state =
5286 						sibling_subcore_state;
5287 		}
5288 	}
5289 	return 0;
5290 }
5291 
5292 static int kvmppc_radix_possible(void)
5293 {
5294 	return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
5295 }
5296 
5297 static int kvmppc_book3s_init_hv(void)
5298 {
5299 	int r;
5300 	/*
5301 	 * FIXME!! Do we need to check on all cpus ?
5302 	 */
5303 	r = kvmppc_core_check_processor_compat_hv();
5304 	if (r < 0)
5305 		return -ENODEV;
5306 
5307 	r = kvmhv_nested_init();
5308 	if (r)
5309 		return r;
5310 
5311 	r = kvm_init_subcore_bitmap();
5312 	if (r)
5313 		return r;
5314 
5315 	/*
5316 	 * We need a way of accessing the XICS interrupt controller,
5317 	 * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
5318 	 * indirectly, via OPAL.
5319 	 */
5320 #ifdef CONFIG_SMP
5321 	if (!xive_enabled() && !kvmhv_on_pseries() &&
5322 	    !local_paca->kvm_hstate.xics_phys) {
5323 		struct device_node *np;
5324 
5325 		np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
5326 		if (!np) {
5327 			pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
5328 			return -ENODEV;
5329 		}
5330 		/* presence of intc confirmed - node can be dropped again */
5331 		of_node_put(np);
5332 	}
5333 #endif
5334 
5335 	kvm_ops_hv.owner = THIS_MODULE;
5336 	kvmppc_hv_ops = &kvm_ops_hv;
5337 
5338 	init_default_hcalls();
5339 
5340 	init_vcore_lists();
5341 
5342 	r = kvmppc_mmu_hv_init();
5343 	if (r)
5344 		return r;
5345 
5346 	if (kvmppc_radix_possible())
5347 		r = kvmppc_radix_init();
5348 
5349 	/*
5350 	 * POWER9 chips before version 2.02 can't have some threads in
5351 	 * HPT mode and some in radix mode on the same core.
5352 	 */
5353 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5354 		unsigned int pvr = mfspr(SPRN_PVR);
5355 		if ((pvr >> 16) == PVR_POWER9 &&
5356 		    (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
5357 		     ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
5358 			no_mixing_hpt_and_radix = true;
5359 	}
5360 
5361 	return r;
5362 }
5363 
5364 static void kvmppc_book3s_exit_hv(void)
5365 {
5366 	kvmppc_free_host_rm_ops();
5367 	if (kvmppc_radix_possible())
5368 		kvmppc_radix_exit();
5369 	kvmppc_hv_ops = NULL;
5370 	kvmhv_nested_exit();
5371 }
5372 
5373 module_init(kvmppc_book3s_init_hv);
5374 module_exit(kvmppc_book3s_exit_hv);
5375 MODULE_LICENSE("GPL");
5376 MODULE_ALIAS_MISCDEV(KVM_MINOR);
5377 MODULE_ALIAS("devname:kvm");
5378