xref: /linux/arch/powerpc/kvm/book3s_hv.c (revision ebf68996de0ab250c5d520eb2291ab65643e9a1e)
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 #include <asm/hw_breakpoint.h>
78 
79 #include "book3s.h"
80 
81 #define CREATE_TRACE_POINTS
82 #include "trace_hv.h"
83 
84 /* #define EXIT_DEBUG */
85 /* #define EXIT_DEBUG_SIMPLE */
86 /* #define EXIT_DEBUG_INT */
87 
88 /* Used to indicate that a guest page fault needs to be handled */
89 #define RESUME_PAGE_FAULT	(RESUME_GUEST | RESUME_FLAG_ARCH1)
90 /* Used to indicate that a guest passthrough interrupt needs to be handled */
91 #define RESUME_PASSTHROUGH	(RESUME_GUEST | RESUME_FLAG_ARCH2)
92 
93 /* Used as a "null" value for timebase values */
94 #define TB_NIL	(~(u64)0)
95 
96 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
97 
98 static int dynamic_mt_modes = 6;
99 module_param(dynamic_mt_modes, int, 0644);
100 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
101 static int target_smt_mode;
102 module_param(target_smt_mode, int, 0644);
103 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
104 
105 static bool indep_threads_mode = true;
106 module_param(indep_threads_mode, bool, S_IRUGO | S_IWUSR);
107 MODULE_PARM_DESC(indep_threads_mode, "Independent-threads mode (only on POWER9)");
108 
109 static bool one_vm_per_core;
110 module_param(one_vm_per_core, bool, S_IRUGO | S_IWUSR);
111 MODULE_PARM_DESC(one_vm_per_core, "Only run vCPUs from the same VM on a core (requires indep_threads_mode=N)");
112 
113 #ifdef CONFIG_KVM_XICS
114 static struct kernel_param_ops module_param_ops = {
115 	.set = param_set_int,
116 	.get = param_get_int,
117 };
118 
119 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
120 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
121 
122 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
123 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
124 #endif
125 
126 /* If set, guests are allowed to create and control nested guests */
127 static bool nested = true;
128 module_param(nested, bool, S_IRUGO | S_IWUSR);
129 MODULE_PARM_DESC(nested, "Enable nested virtualization (only on POWER9)");
130 
131 static inline bool nesting_enabled(struct kvm *kvm)
132 {
133 	return kvm->arch.nested_enable && kvm_is_radix(kvm);
134 }
135 
136 /* If set, the threads on each CPU core have to be in the same MMU mode */
137 static bool no_mixing_hpt_and_radix;
138 
139 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
140 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
141 
142 /*
143  * RWMR values for POWER8.  These control the rate at which PURR
144  * and SPURR count and should be set according to the number of
145  * online threads in the vcore being run.
146  */
147 #define RWMR_RPA_P8_1THREAD	0x164520C62609AECAUL
148 #define RWMR_RPA_P8_2THREAD	0x7FFF2908450D8DA9UL
149 #define RWMR_RPA_P8_3THREAD	0x164520C62609AECAUL
150 #define RWMR_RPA_P8_4THREAD	0x199A421245058DA9UL
151 #define RWMR_RPA_P8_5THREAD	0x164520C62609AECAUL
152 #define RWMR_RPA_P8_6THREAD	0x164520C62609AECAUL
153 #define RWMR_RPA_P8_7THREAD	0x164520C62609AECAUL
154 #define RWMR_RPA_P8_8THREAD	0x164520C62609AECAUL
155 
156 static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = {
157 	RWMR_RPA_P8_1THREAD,
158 	RWMR_RPA_P8_1THREAD,
159 	RWMR_RPA_P8_2THREAD,
160 	RWMR_RPA_P8_3THREAD,
161 	RWMR_RPA_P8_4THREAD,
162 	RWMR_RPA_P8_5THREAD,
163 	RWMR_RPA_P8_6THREAD,
164 	RWMR_RPA_P8_7THREAD,
165 	RWMR_RPA_P8_8THREAD,
166 };
167 
168 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
169 		int *ip)
170 {
171 	int i = *ip;
172 	struct kvm_vcpu *vcpu;
173 
174 	while (++i < MAX_SMT_THREADS) {
175 		vcpu = READ_ONCE(vc->runnable_threads[i]);
176 		if (vcpu) {
177 			*ip = i;
178 			return vcpu;
179 		}
180 	}
181 	return NULL;
182 }
183 
184 /* Used to traverse the list of runnable threads for a given vcore */
185 #define for_each_runnable_thread(i, vcpu, vc) \
186 	for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
187 
188 static bool kvmppc_ipi_thread(int cpu)
189 {
190 	unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
191 
192 	/* If we're a nested hypervisor, fall back to ordinary IPIs for now */
193 	if (kvmhv_on_pseries())
194 		return false;
195 
196 	/* On POWER9 we can use msgsnd to IPI any cpu */
197 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
198 		msg |= get_hard_smp_processor_id(cpu);
199 		smp_mb();
200 		__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
201 		return true;
202 	}
203 
204 	/* On POWER8 for IPIs to threads in the same core, use msgsnd */
205 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
206 		preempt_disable();
207 		if (cpu_first_thread_sibling(cpu) ==
208 		    cpu_first_thread_sibling(smp_processor_id())) {
209 			msg |= cpu_thread_in_core(cpu);
210 			smp_mb();
211 			__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
212 			preempt_enable();
213 			return true;
214 		}
215 		preempt_enable();
216 	}
217 
218 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
219 	if (cpu >= 0 && cpu < nr_cpu_ids) {
220 		if (paca_ptrs[cpu]->kvm_hstate.xics_phys) {
221 			xics_wake_cpu(cpu);
222 			return true;
223 		}
224 		opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
225 		return true;
226 	}
227 #endif
228 
229 	return false;
230 }
231 
232 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
233 {
234 	int cpu;
235 	struct swait_queue_head *wqp;
236 
237 	wqp = kvm_arch_vcpu_wq(vcpu);
238 	if (swq_has_sleeper(wqp)) {
239 		swake_up_one(wqp);
240 		++vcpu->stat.halt_wakeup;
241 	}
242 
243 	cpu = READ_ONCE(vcpu->arch.thread_cpu);
244 	if (cpu >= 0 && kvmppc_ipi_thread(cpu))
245 		return;
246 
247 	/* CPU points to the first thread of the core */
248 	cpu = vcpu->cpu;
249 	if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
250 		smp_send_reschedule(cpu);
251 }
252 
253 /*
254  * We use the vcpu_load/put functions to measure stolen time.
255  * Stolen time is counted as time when either the vcpu is able to
256  * run as part of a virtual core, but the task running the vcore
257  * is preempted or sleeping, or when the vcpu needs something done
258  * in the kernel by the task running the vcpu, but that task is
259  * preempted or sleeping.  Those two things have to be counted
260  * separately, since one of the vcpu tasks will take on the job
261  * of running the core, and the other vcpu tasks in the vcore will
262  * sleep waiting for it to do that, but that sleep shouldn't count
263  * as stolen time.
264  *
265  * Hence we accumulate stolen time when the vcpu can run as part of
266  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
267  * needs its task to do other things in the kernel (for example,
268  * service a page fault) in busy_stolen.  We don't accumulate
269  * stolen time for a vcore when it is inactive, or for a vcpu
270  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
271  * a misnomer; it means that the vcpu task is not executing in
272  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
273  * the kernel.  We don't have any way of dividing up that time
274  * between time that the vcpu is genuinely stopped, time that
275  * the task is actively working on behalf of the vcpu, and time
276  * that the task is preempted, so we don't count any of it as
277  * stolen.
278  *
279  * Updates to busy_stolen are protected by arch.tbacct_lock;
280  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
281  * lock.  The stolen times are measured in units of timebase ticks.
282  * (Note that the != TB_NIL checks below are purely defensive;
283  * they should never fail.)
284  */
285 
286 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
287 {
288 	unsigned long flags;
289 
290 	spin_lock_irqsave(&vc->stoltb_lock, flags);
291 	vc->preempt_tb = mftb();
292 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
293 }
294 
295 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
296 {
297 	unsigned long flags;
298 
299 	spin_lock_irqsave(&vc->stoltb_lock, flags);
300 	if (vc->preempt_tb != TB_NIL) {
301 		vc->stolen_tb += mftb() - vc->preempt_tb;
302 		vc->preempt_tb = TB_NIL;
303 	}
304 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
305 }
306 
307 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
308 {
309 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
310 	unsigned long flags;
311 
312 	/*
313 	 * We can test vc->runner without taking the vcore lock,
314 	 * because only this task ever sets vc->runner to this
315 	 * vcpu, and once it is set to this vcpu, only this task
316 	 * ever sets it to NULL.
317 	 */
318 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
319 		kvmppc_core_end_stolen(vc);
320 
321 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
322 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
323 	    vcpu->arch.busy_preempt != TB_NIL) {
324 		vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
325 		vcpu->arch.busy_preempt = TB_NIL;
326 	}
327 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
328 }
329 
330 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
331 {
332 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
333 	unsigned long flags;
334 
335 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
336 		kvmppc_core_start_stolen(vc);
337 
338 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
339 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
340 		vcpu->arch.busy_preempt = mftb();
341 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
342 }
343 
344 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
345 {
346 	/*
347 	 * Check for illegal transactional state bit combination
348 	 * and if we find it, force the TS field to a safe state.
349 	 */
350 	if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
351 		msr &= ~MSR_TS_MASK;
352 	vcpu->arch.shregs.msr = msr;
353 	kvmppc_end_cede(vcpu);
354 }
355 
356 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
357 {
358 	vcpu->arch.pvr = pvr;
359 }
360 
361 /* Dummy value used in computing PCR value below */
362 #define PCR_ARCH_300	(PCR_ARCH_207 << 1)
363 
364 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
365 {
366 	unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
367 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
368 
369 	/* We can (emulate) our own architecture version and anything older */
370 	if (cpu_has_feature(CPU_FTR_ARCH_300))
371 		host_pcr_bit = PCR_ARCH_300;
372 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
373 		host_pcr_bit = PCR_ARCH_207;
374 	else if (cpu_has_feature(CPU_FTR_ARCH_206))
375 		host_pcr_bit = PCR_ARCH_206;
376 	else
377 		host_pcr_bit = PCR_ARCH_205;
378 
379 	/* Determine lowest PCR bit needed to run guest in given PVR level */
380 	guest_pcr_bit = host_pcr_bit;
381 	if (arch_compat) {
382 		switch (arch_compat) {
383 		case PVR_ARCH_205:
384 			guest_pcr_bit = PCR_ARCH_205;
385 			break;
386 		case PVR_ARCH_206:
387 		case PVR_ARCH_206p:
388 			guest_pcr_bit = PCR_ARCH_206;
389 			break;
390 		case PVR_ARCH_207:
391 			guest_pcr_bit = PCR_ARCH_207;
392 			break;
393 		case PVR_ARCH_300:
394 			guest_pcr_bit = PCR_ARCH_300;
395 			break;
396 		default:
397 			return -EINVAL;
398 		}
399 	}
400 
401 	/* Check requested PCR bits don't exceed our capabilities */
402 	if (guest_pcr_bit > host_pcr_bit)
403 		return -EINVAL;
404 
405 	spin_lock(&vc->lock);
406 	vc->arch_compat = arch_compat;
407 	/* Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit */
408 	vc->pcr = host_pcr_bit - guest_pcr_bit;
409 	spin_unlock(&vc->lock);
410 
411 	return 0;
412 }
413 
414 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
415 {
416 	int r;
417 
418 	pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
419 	pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
420 	       vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
421 	for (r = 0; r < 16; ++r)
422 		pr_err("r%2d = %.16lx  r%d = %.16lx\n",
423 		       r, kvmppc_get_gpr(vcpu, r),
424 		       r+16, kvmppc_get_gpr(vcpu, r+16));
425 	pr_err("ctr = %.16lx  lr  = %.16lx\n",
426 	       vcpu->arch.regs.ctr, vcpu->arch.regs.link);
427 	pr_err("srr0 = %.16llx srr1 = %.16llx\n",
428 	       vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
429 	pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
430 	       vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
431 	pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
432 	       vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
433 	pr_err("cr = %.8lx  xer = %.16lx  dsisr = %.8x\n",
434 	       vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
435 	pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
436 	pr_err("fault dar = %.16lx dsisr = %.8x\n",
437 	       vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
438 	pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
439 	for (r = 0; r < vcpu->arch.slb_max; ++r)
440 		pr_err("  ESID = %.16llx VSID = %.16llx\n",
441 		       vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
442 	pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
443 	       vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
444 	       vcpu->arch.last_inst);
445 }
446 
447 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
448 {
449 	return kvm_get_vcpu_by_id(kvm, id);
450 }
451 
452 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
453 {
454 	vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
455 	vpa->yield_count = cpu_to_be32(1);
456 }
457 
458 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
459 		   unsigned long addr, unsigned long len)
460 {
461 	/* check address is cacheline aligned */
462 	if (addr & (L1_CACHE_BYTES - 1))
463 		return -EINVAL;
464 	spin_lock(&vcpu->arch.vpa_update_lock);
465 	if (v->next_gpa != addr || v->len != len) {
466 		v->next_gpa = addr;
467 		v->len = addr ? len : 0;
468 		v->update_pending = 1;
469 	}
470 	spin_unlock(&vcpu->arch.vpa_update_lock);
471 	return 0;
472 }
473 
474 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
475 struct reg_vpa {
476 	u32 dummy;
477 	union {
478 		__be16 hword;
479 		__be32 word;
480 	} length;
481 };
482 
483 static int vpa_is_registered(struct kvmppc_vpa *vpap)
484 {
485 	if (vpap->update_pending)
486 		return vpap->next_gpa != 0;
487 	return vpap->pinned_addr != NULL;
488 }
489 
490 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
491 				       unsigned long flags,
492 				       unsigned long vcpuid, unsigned long vpa)
493 {
494 	struct kvm *kvm = vcpu->kvm;
495 	unsigned long len, nb;
496 	void *va;
497 	struct kvm_vcpu *tvcpu;
498 	int err;
499 	int subfunc;
500 	struct kvmppc_vpa *vpap;
501 
502 	tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
503 	if (!tvcpu)
504 		return H_PARAMETER;
505 
506 	subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
507 	if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
508 	    subfunc == H_VPA_REG_SLB) {
509 		/* Registering new area - address must be cache-line aligned */
510 		if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
511 			return H_PARAMETER;
512 
513 		/* convert logical addr to kernel addr and read length */
514 		va = kvmppc_pin_guest_page(kvm, vpa, &nb);
515 		if (va == NULL)
516 			return H_PARAMETER;
517 		if (subfunc == H_VPA_REG_VPA)
518 			len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
519 		else
520 			len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
521 		kvmppc_unpin_guest_page(kvm, va, vpa, false);
522 
523 		/* Check length */
524 		if (len > nb || len < sizeof(struct reg_vpa))
525 			return H_PARAMETER;
526 	} else {
527 		vpa = 0;
528 		len = 0;
529 	}
530 
531 	err = H_PARAMETER;
532 	vpap = NULL;
533 	spin_lock(&tvcpu->arch.vpa_update_lock);
534 
535 	switch (subfunc) {
536 	case H_VPA_REG_VPA:		/* register VPA */
537 		/*
538 		 * The size of our lppaca is 1kB because of the way we align
539 		 * it for the guest to avoid crossing a 4kB boundary. We only
540 		 * use 640 bytes of the structure though, so we should accept
541 		 * clients that set a size of 640.
542 		 */
543 		BUILD_BUG_ON(sizeof(struct lppaca) != 640);
544 		if (len < sizeof(struct lppaca))
545 			break;
546 		vpap = &tvcpu->arch.vpa;
547 		err = 0;
548 		break;
549 
550 	case H_VPA_REG_DTL:		/* register DTL */
551 		if (len < sizeof(struct dtl_entry))
552 			break;
553 		len -= len % sizeof(struct dtl_entry);
554 
555 		/* Check that they have previously registered a VPA */
556 		err = H_RESOURCE;
557 		if (!vpa_is_registered(&tvcpu->arch.vpa))
558 			break;
559 
560 		vpap = &tvcpu->arch.dtl;
561 		err = 0;
562 		break;
563 
564 	case H_VPA_REG_SLB:		/* register SLB shadow buffer */
565 		/* Check that they have previously registered a VPA */
566 		err = H_RESOURCE;
567 		if (!vpa_is_registered(&tvcpu->arch.vpa))
568 			break;
569 
570 		vpap = &tvcpu->arch.slb_shadow;
571 		err = 0;
572 		break;
573 
574 	case H_VPA_DEREG_VPA:		/* deregister VPA */
575 		/* Check they don't still have a DTL or SLB buf registered */
576 		err = H_RESOURCE;
577 		if (vpa_is_registered(&tvcpu->arch.dtl) ||
578 		    vpa_is_registered(&tvcpu->arch.slb_shadow))
579 			break;
580 
581 		vpap = &tvcpu->arch.vpa;
582 		err = 0;
583 		break;
584 
585 	case H_VPA_DEREG_DTL:		/* deregister DTL */
586 		vpap = &tvcpu->arch.dtl;
587 		err = 0;
588 		break;
589 
590 	case H_VPA_DEREG_SLB:		/* deregister SLB shadow buffer */
591 		vpap = &tvcpu->arch.slb_shadow;
592 		err = 0;
593 		break;
594 	}
595 
596 	if (vpap) {
597 		vpap->next_gpa = vpa;
598 		vpap->len = len;
599 		vpap->update_pending = 1;
600 	}
601 
602 	spin_unlock(&tvcpu->arch.vpa_update_lock);
603 
604 	return err;
605 }
606 
607 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
608 {
609 	struct kvm *kvm = vcpu->kvm;
610 	void *va;
611 	unsigned long nb;
612 	unsigned long gpa;
613 
614 	/*
615 	 * We need to pin the page pointed to by vpap->next_gpa,
616 	 * but we can't call kvmppc_pin_guest_page under the lock
617 	 * as it does get_user_pages() and down_read().  So we
618 	 * have to drop the lock, pin the page, then get the lock
619 	 * again and check that a new area didn't get registered
620 	 * in the meantime.
621 	 */
622 	for (;;) {
623 		gpa = vpap->next_gpa;
624 		spin_unlock(&vcpu->arch.vpa_update_lock);
625 		va = NULL;
626 		nb = 0;
627 		if (gpa)
628 			va = kvmppc_pin_guest_page(kvm, gpa, &nb);
629 		spin_lock(&vcpu->arch.vpa_update_lock);
630 		if (gpa == vpap->next_gpa)
631 			break;
632 		/* sigh... unpin that one and try again */
633 		if (va)
634 			kvmppc_unpin_guest_page(kvm, va, gpa, false);
635 	}
636 
637 	vpap->update_pending = 0;
638 	if (va && nb < vpap->len) {
639 		/*
640 		 * If it's now too short, it must be that userspace
641 		 * has changed the mappings underlying guest memory,
642 		 * so unregister the region.
643 		 */
644 		kvmppc_unpin_guest_page(kvm, va, gpa, false);
645 		va = NULL;
646 	}
647 	if (vpap->pinned_addr)
648 		kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
649 					vpap->dirty);
650 	vpap->gpa = gpa;
651 	vpap->pinned_addr = va;
652 	vpap->dirty = false;
653 	if (va)
654 		vpap->pinned_end = va + vpap->len;
655 }
656 
657 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
658 {
659 	if (!(vcpu->arch.vpa.update_pending ||
660 	      vcpu->arch.slb_shadow.update_pending ||
661 	      vcpu->arch.dtl.update_pending))
662 		return;
663 
664 	spin_lock(&vcpu->arch.vpa_update_lock);
665 	if (vcpu->arch.vpa.update_pending) {
666 		kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
667 		if (vcpu->arch.vpa.pinned_addr)
668 			init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
669 	}
670 	if (vcpu->arch.dtl.update_pending) {
671 		kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
672 		vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
673 		vcpu->arch.dtl_index = 0;
674 	}
675 	if (vcpu->arch.slb_shadow.update_pending)
676 		kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
677 	spin_unlock(&vcpu->arch.vpa_update_lock);
678 }
679 
680 /*
681  * Return the accumulated stolen time for the vcore up until `now'.
682  * The caller should hold the vcore lock.
683  */
684 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
685 {
686 	u64 p;
687 	unsigned long flags;
688 
689 	spin_lock_irqsave(&vc->stoltb_lock, flags);
690 	p = vc->stolen_tb;
691 	if (vc->vcore_state != VCORE_INACTIVE &&
692 	    vc->preempt_tb != TB_NIL)
693 		p += now - vc->preempt_tb;
694 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
695 	return p;
696 }
697 
698 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
699 				    struct kvmppc_vcore *vc)
700 {
701 	struct dtl_entry *dt;
702 	struct lppaca *vpa;
703 	unsigned long stolen;
704 	unsigned long core_stolen;
705 	u64 now;
706 	unsigned long flags;
707 
708 	dt = vcpu->arch.dtl_ptr;
709 	vpa = vcpu->arch.vpa.pinned_addr;
710 	now = mftb();
711 	core_stolen = vcore_stolen_time(vc, now);
712 	stolen = core_stolen - vcpu->arch.stolen_logged;
713 	vcpu->arch.stolen_logged = core_stolen;
714 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
715 	stolen += vcpu->arch.busy_stolen;
716 	vcpu->arch.busy_stolen = 0;
717 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
718 	if (!dt || !vpa)
719 		return;
720 	memset(dt, 0, sizeof(struct dtl_entry));
721 	dt->dispatch_reason = 7;
722 	dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
723 	dt->timebase = cpu_to_be64(now + vc->tb_offset);
724 	dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
725 	dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
726 	dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
727 	++dt;
728 	if (dt == vcpu->arch.dtl.pinned_end)
729 		dt = vcpu->arch.dtl.pinned_addr;
730 	vcpu->arch.dtl_ptr = dt;
731 	/* order writing *dt vs. writing vpa->dtl_idx */
732 	smp_wmb();
733 	vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
734 	vcpu->arch.dtl.dirty = true;
735 }
736 
737 /* See if there is a doorbell interrupt pending for a vcpu */
738 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
739 {
740 	int thr;
741 	struct kvmppc_vcore *vc;
742 
743 	if (vcpu->arch.doorbell_request)
744 		return true;
745 	/*
746 	 * Ensure that the read of vcore->dpdes comes after the read
747 	 * of vcpu->doorbell_request.  This barrier matches the
748 	 * smp_wmb() in kvmppc_guest_entry_inject().
749 	 */
750 	smp_rmb();
751 	vc = vcpu->arch.vcore;
752 	thr = vcpu->vcpu_id - vc->first_vcpuid;
753 	return !!(vc->dpdes & (1 << thr));
754 }
755 
756 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
757 {
758 	if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
759 		return true;
760 	if ((!vcpu->arch.vcore->arch_compat) &&
761 	    cpu_has_feature(CPU_FTR_ARCH_207S))
762 		return true;
763 	return false;
764 }
765 
766 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
767 			     unsigned long resource, unsigned long value1,
768 			     unsigned long value2)
769 {
770 	switch (resource) {
771 	case H_SET_MODE_RESOURCE_SET_CIABR:
772 		if (!kvmppc_power8_compatible(vcpu))
773 			return H_P2;
774 		if (value2)
775 			return H_P4;
776 		if (mflags)
777 			return H_UNSUPPORTED_FLAG_START;
778 		/* Guests can't breakpoint the hypervisor */
779 		if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
780 			return H_P3;
781 		vcpu->arch.ciabr  = value1;
782 		return H_SUCCESS;
783 	case H_SET_MODE_RESOURCE_SET_DAWR:
784 		if (!kvmppc_power8_compatible(vcpu))
785 			return H_P2;
786 		if (!ppc_breakpoint_available())
787 			return H_P2;
788 		if (mflags)
789 			return H_UNSUPPORTED_FLAG_START;
790 		if (value2 & DABRX_HYP)
791 			return H_P4;
792 		vcpu->arch.dawr  = value1;
793 		vcpu->arch.dawrx = value2;
794 		return H_SUCCESS;
795 	default:
796 		return H_TOO_HARD;
797 	}
798 }
799 
800 /* Copy guest memory in place - must reside within a single memslot */
801 static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from,
802 				  unsigned long len)
803 {
804 	struct kvm_memory_slot *to_memslot = NULL;
805 	struct kvm_memory_slot *from_memslot = NULL;
806 	unsigned long to_addr, from_addr;
807 	int r;
808 
809 	/* Get HPA for from address */
810 	from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT);
811 	if (!from_memslot)
812 		return -EFAULT;
813 	if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages)
814 			     << PAGE_SHIFT))
815 		return -EINVAL;
816 	from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT);
817 	if (kvm_is_error_hva(from_addr))
818 		return -EFAULT;
819 	from_addr |= (from & (PAGE_SIZE - 1));
820 
821 	/* Get HPA for to address */
822 	to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT);
823 	if (!to_memslot)
824 		return -EFAULT;
825 	if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages)
826 			   << PAGE_SHIFT))
827 		return -EINVAL;
828 	to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT);
829 	if (kvm_is_error_hva(to_addr))
830 		return -EFAULT;
831 	to_addr |= (to & (PAGE_SIZE - 1));
832 
833 	/* Perform copy */
834 	r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr,
835 			     len);
836 	if (r)
837 		return -EFAULT;
838 	mark_page_dirty(kvm, to >> PAGE_SHIFT);
839 	return 0;
840 }
841 
842 static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags,
843 			       unsigned long dest, unsigned long src)
844 {
845 	u64 pg_sz = SZ_4K;		/* 4K page size */
846 	u64 pg_mask = SZ_4K - 1;
847 	int ret;
848 
849 	/* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */
850 	if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE |
851 		      H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED))
852 		return H_PARAMETER;
853 
854 	/* dest (and src if copy_page flag set) must be page aligned */
855 	if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask)))
856 		return H_PARAMETER;
857 
858 	/* zero and/or copy the page as determined by the flags */
859 	if (flags & H_COPY_PAGE) {
860 		ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz);
861 		if (ret < 0)
862 			return H_PARAMETER;
863 	} else if (flags & H_ZERO_PAGE) {
864 		ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz);
865 		if (ret < 0)
866 			return H_PARAMETER;
867 	}
868 
869 	/* We can ignore the remaining flags */
870 
871 	return H_SUCCESS;
872 }
873 
874 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
875 {
876 	struct kvmppc_vcore *vcore = target->arch.vcore;
877 
878 	/*
879 	 * We expect to have been called by the real mode handler
880 	 * (kvmppc_rm_h_confer()) which would have directly returned
881 	 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
882 	 * have useful work to do and should not confer) so we don't
883 	 * recheck that here.
884 	 */
885 
886 	spin_lock(&vcore->lock);
887 	if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
888 	    vcore->vcore_state != VCORE_INACTIVE &&
889 	    vcore->runner)
890 		target = vcore->runner;
891 	spin_unlock(&vcore->lock);
892 
893 	return kvm_vcpu_yield_to(target);
894 }
895 
896 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
897 {
898 	int yield_count = 0;
899 	struct lppaca *lppaca;
900 
901 	spin_lock(&vcpu->arch.vpa_update_lock);
902 	lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
903 	if (lppaca)
904 		yield_count = be32_to_cpu(lppaca->yield_count);
905 	spin_unlock(&vcpu->arch.vpa_update_lock);
906 	return yield_count;
907 }
908 
909 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
910 {
911 	unsigned long req = kvmppc_get_gpr(vcpu, 3);
912 	unsigned long target, ret = H_SUCCESS;
913 	int yield_count;
914 	struct kvm_vcpu *tvcpu;
915 	int idx, rc;
916 
917 	if (req <= MAX_HCALL_OPCODE &&
918 	    !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
919 		return RESUME_HOST;
920 
921 	switch (req) {
922 	case H_CEDE:
923 		break;
924 	case H_PROD:
925 		target = kvmppc_get_gpr(vcpu, 4);
926 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
927 		if (!tvcpu) {
928 			ret = H_PARAMETER;
929 			break;
930 		}
931 		tvcpu->arch.prodded = 1;
932 		smp_mb();
933 		if (tvcpu->arch.ceded)
934 			kvmppc_fast_vcpu_kick_hv(tvcpu);
935 		break;
936 	case H_CONFER:
937 		target = kvmppc_get_gpr(vcpu, 4);
938 		if (target == -1)
939 			break;
940 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
941 		if (!tvcpu) {
942 			ret = H_PARAMETER;
943 			break;
944 		}
945 		yield_count = kvmppc_get_gpr(vcpu, 5);
946 		if (kvmppc_get_yield_count(tvcpu) != yield_count)
947 			break;
948 		kvm_arch_vcpu_yield_to(tvcpu);
949 		break;
950 	case H_REGISTER_VPA:
951 		ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
952 					kvmppc_get_gpr(vcpu, 5),
953 					kvmppc_get_gpr(vcpu, 6));
954 		break;
955 	case H_RTAS:
956 		if (list_empty(&vcpu->kvm->arch.rtas_tokens))
957 			return RESUME_HOST;
958 
959 		idx = srcu_read_lock(&vcpu->kvm->srcu);
960 		rc = kvmppc_rtas_hcall(vcpu);
961 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
962 
963 		if (rc == -ENOENT)
964 			return RESUME_HOST;
965 		else if (rc == 0)
966 			break;
967 
968 		/* Send the error out to userspace via KVM_RUN */
969 		return rc;
970 	case H_LOGICAL_CI_LOAD:
971 		ret = kvmppc_h_logical_ci_load(vcpu);
972 		if (ret == H_TOO_HARD)
973 			return RESUME_HOST;
974 		break;
975 	case H_LOGICAL_CI_STORE:
976 		ret = kvmppc_h_logical_ci_store(vcpu);
977 		if (ret == H_TOO_HARD)
978 			return RESUME_HOST;
979 		break;
980 	case H_SET_MODE:
981 		ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
982 					kvmppc_get_gpr(vcpu, 5),
983 					kvmppc_get_gpr(vcpu, 6),
984 					kvmppc_get_gpr(vcpu, 7));
985 		if (ret == H_TOO_HARD)
986 			return RESUME_HOST;
987 		break;
988 	case H_XIRR:
989 	case H_CPPR:
990 	case H_EOI:
991 	case H_IPI:
992 	case H_IPOLL:
993 	case H_XIRR_X:
994 		if (kvmppc_xics_enabled(vcpu)) {
995 			if (xics_on_xive()) {
996 				ret = H_NOT_AVAILABLE;
997 				return RESUME_GUEST;
998 			}
999 			ret = kvmppc_xics_hcall(vcpu, req);
1000 			break;
1001 		}
1002 		return RESUME_HOST;
1003 	case H_SET_DABR:
1004 		ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
1005 		break;
1006 	case H_SET_XDABR:
1007 		ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
1008 						kvmppc_get_gpr(vcpu, 5));
1009 		break;
1010 #ifdef CONFIG_SPAPR_TCE_IOMMU
1011 	case H_GET_TCE:
1012 		ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1013 						kvmppc_get_gpr(vcpu, 5));
1014 		if (ret == H_TOO_HARD)
1015 			return RESUME_HOST;
1016 		break;
1017 	case H_PUT_TCE:
1018 		ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1019 						kvmppc_get_gpr(vcpu, 5),
1020 						kvmppc_get_gpr(vcpu, 6));
1021 		if (ret == H_TOO_HARD)
1022 			return RESUME_HOST;
1023 		break;
1024 	case H_PUT_TCE_INDIRECT:
1025 		ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
1026 						kvmppc_get_gpr(vcpu, 5),
1027 						kvmppc_get_gpr(vcpu, 6),
1028 						kvmppc_get_gpr(vcpu, 7));
1029 		if (ret == H_TOO_HARD)
1030 			return RESUME_HOST;
1031 		break;
1032 	case H_STUFF_TCE:
1033 		ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1034 						kvmppc_get_gpr(vcpu, 5),
1035 						kvmppc_get_gpr(vcpu, 6),
1036 						kvmppc_get_gpr(vcpu, 7));
1037 		if (ret == H_TOO_HARD)
1038 			return RESUME_HOST;
1039 		break;
1040 #endif
1041 	case H_RANDOM:
1042 		if (!powernv_get_random_long(&vcpu->arch.regs.gpr[4]))
1043 			ret = H_HARDWARE;
1044 		break;
1045 
1046 	case H_SET_PARTITION_TABLE:
1047 		ret = H_FUNCTION;
1048 		if (nesting_enabled(vcpu->kvm))
1049 			ret = kvmhv_set_partition_table(vcpu);
1050 		break;
1051 	case H_ENTER_NESTED:
1052 		ret = H_FUNCTION;
1053 		if (!nesting_enabled(vcpu->kvm))
1054 			break;
1055 		ret = kvmhv_enter_nested_guest(vcpu);
1056 		if (ret == H_INTERRUPT) {
1057 			kvmppc_set_gpr(vcpu, 3, 0);
1058 			vcpu->arch.hcall_needed = 0;
1059 			return -EINTR;
1060 		} else if (ret == H_TOO_HARD) {
1061 			kvmppc_set_gpr(vcpu, 3, 0);
1062 			vcpu->arch.hcall_needed = 0;
1063 			return RESUME_HOST;
1064 		}
1065 		break;
1066 	case H_TLB_INVALIDATE:
1067 		ret = H_FUNCTION;
1068 		if (nesting_enabled(vcpu->kvm))
1069 			ret = kvmhv_do_nested_tlbie(vcpu);
1070 		break;
1071 	case H_COPY_TOFROM_GUEST:
1072 		ret = H_FUNCTION;
1073 		if (nesting_enabled(vcpu->kvm))
1074 			ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1075 		break;
1076 	case H_PAGE_INIT:
1077 		ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4),
1078 					 kvmppc_get_gpr(vcpu, 5),
1079 					 kvmppc_get_gpr(vcpu, 6));
1080 		break;
1081 	default:
1082 		return RESUME_HOST;
1083 	}
1084 	kvmppc_set_gpr(vcpu, 3, ret);
1085 	vcpu->arch.hcall_needed = 0;
1086 	return RESUME_GUEST;
1087 }
1088 
1089 /*
1090  * Handle H_CEDE in the nested virtualization case where we haven't
1091  * called the real-mode hcall handlers in book3s_hv_rmhandlers.S.
1092  * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1093  * that the cede logic in kvmppc_run_single_vcpu() works properly.
1094  */
1095 static void kvmppc_nested_cede(struct kvm_vcpu *vcpu)
1096 {
1097 	vcpu->arch.shregs.msr |= MSR_EE;
1098 	vcpu->arch.ceded = 1;
1099 	smp_mb();
1100 	if (vcpu->arch.prodded) {
1101 		vcpu->arch.prodded = 0;
1102 		smp_mb();
1103 		vcpu->arch.ceded = 0;
1104 	}
1105 }
1106 
1107 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1108 {
1109 	switch (cmd) {
1110 	case H_CEDE:
1111 	case H_PROD:
1112 	case H_CONFER:
1113 	case H_REGISTER_VPA:
1114 	case H_SET_MODE:
1115 	case H_LOGICAL_CI_LOAD:
1116 	case H_LOGICAL_CI_STORE:
1117 #ifdef CONFIG_KVM_XICS
1118 	case H_XIRR:
1119 	case H_CPPR:
1120 	case H_EOI:
1121 	case H_IPI:
1122 	case H_IPOLL:
1123 	case H_XIRR_X:
1124 #endif
1125 	case H_PAGE_INIT:
1126 		return 1;
1127 	}
1128 
1129 	/* See if it's in the real-mode table */
1130 	return kvmppc_hcall_impl_hv_realmode(cmd);
1131 }
1132 
1133 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
1134 					struct kvm_vcpu *vcpu)
1135 {
1136 	u32 last_inst;
1137 
1138 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1139 					EMULATE_DONE) {
1140 		/*
1141 		 * Fetch failed, so return to guest and
1142 		 * try executing it again.
1143 		 */
1144 		return RESUME_GUEST;
1145 	}
1146 
1147 	if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
1148 		run->exit_reason = KVM_EXIT_DEBUG;
1149 		run->debug.arch.address = kvmppc_get_pc(vcpu);
1150 		return RESUME_HOST;
1151 	} else {
1152 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1153 		return RESUME_GUEST;
1154 	}
1155 }
1156 
1157 static void do_nothing(void *x)
1158 {
1159 }
1160 
1161 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1162 {
1163 	int thr, cpu, pcpu, nthreads;
1164 	struct kvm_vcpu *v;
1165 	unsigned long dpdes;
1166 
1167 	nthreads = vcpu->kvm->arch.emul_smt_mode;
1168 	dpdes = 0;
1169 	cpu = vcpu->vcpu_id & ~(nthreads - 1);
1170 	for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1171 		v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1172 		if (!v)
1173 			continue;
1174 		/*
1175 		 * If the vcpu is currently running on a physical cpu thread,
1176 		 * interrupt it in order to pull it out of the guest briefly,
1177 		 * which will update its vcore->dpdes value.
1178 		 */
1179 		pcpu = READ_ONCE(v->cpu);
1180 		if (pcpu >= 0)
1181 			smp_call_function_single(pcpu, do_nothing, NULL, 1);
1182 		if (kvmppc_doorbell_pending(v))
1183 			dpdes |= 1 << thr;
1184 	}
1185 	return dpdes;
1186 }
1187 
1188 /*
1189  * On POWER9, emulate doorbell-related instructions in order to
1190  * give the guest the illusion of running on a multi-threaded core.
1191  * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1192  * and mfspr DPDES.
1193  */
1194 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1195 {
1196 	u32 inst, rb, thr;
1197 	unsigned long arg;
1198 	struct kvm *kvm = vcpu->kvm;
1199 	struct kvm_vcpu *tvcpu;
1200 
1201 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1202 		return RESUME_GUEST;
1203 	if (get_op(inst) != 31)
1204 		return EMULATE_FAIL;
1205 	rb = get_rb(inst);
1206 	thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1207 	switch (get_xop(inst)) {
1208 	case OP_31_XOP_MSGSNDP:
1209 		arg = kvmppc_get_gpr(vcpu, rb);
1210 		if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1211 			break;
1212 		arg &= 0x3f;
1213 		if (arg >= kvm->arch.emul_smt_mode)
1214 			break;
1215 		tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1216 		if (!tvcpu)
1217 			break;
1218 		if (!tvcpu->arch.doorbell_request) {
1219 			tvcpu->arch.doorbell_request = 1;
1220 			kvmppc_fast_vcpu_kick_hv(tvcpu);
1221 		}
1222 		break;
1223 	case OP_31_XOP_MSGCLRP:
1224 		arg = kvmppc_get_gpr(vcpu, rb);
1225 		if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1226 			break;
1227 		vcpu->arch.vcore->dpdes = 0;
1228 		vcpu->arch.doorbell_request = 0;
1229 		break;
1230 	case OP_31_XOP_MFSPR:
1231 		switch (get_sprn(inst)) {
1232 		case SPRN_TIR:
1233 			arg = thr;
1234 			break;
1235 		case SPRN_DPDES:
1236 			arg = kvmppc_read_dpdes(vcpu);
1237 			break;
1238 		default:
1239 			return EMULATE_FAIL;
1240 		}
1241 		kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1242 		break;
1243 	default:
1244 		return EMULATE_FAIL;
1245 	}
1246 	kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1247 	return RESUME_GUEST;
1248 }
1249 
1250 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
1251 				 struct task_struct *tsk)
1252 {
1253 	int r = RESUME_HOST;
1254 
1255 	vcpu->stat.sum_exits++;
1256 
1257 	/*
1258 	 * This can happen if an interrupt occurs in the last stages
1259 	 * of guest entry or the first stages of guest exit (i.e. after
1260 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1261 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1262 	 * That can happen due to a bug, or due to a machine check
1263 	 * occurring at just the wrong time.
1264 	 */
1265 	if (vcpu->arch.shregs.msr & MSR_HV) {
1266 		printk(KERN_EMERG "KVM trap in HV mode!\n");
1267 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1268 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1269 			vcpu->arch.shregs.msr);
1270 		kvmppc_dump_regs(vcpu);
1271 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1272 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1273 		return RESUME_HOST;
1274 	}
1275 	run->exit_reason = KVM_EXIT_UNKNOWN;
1276 	run->ready_for_interrupt_injection = 1;
1277 	switch (vcpu->arch.trap) {
1278 	/* We're good on these - the host merely wanted to get our attention */
1279 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1280 		vcpu->stat.dec_exits++;
1281 		r = RESUME_GUEST;
1282 		break;
1283 	case BOOK3S_INTERRUPT_EXTERNAL:
1284 	case BOOK3S_INTERRUPT_H_DOORBELL:
1285 	case BOOK3S_INTERRUPT_H_VIRT:
1286 		vcpu->stat.ext_intr_exits++;
1287 		r = RESUME_GUEST;
1288 		break;
1289 	/* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1290 	case BOOK3S_INTERRUPT_HMI:
1291 	case BOOK3S_INTERRUPT_PERFMON:
1292 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
1293 		r = RESUME_GUEST;
1294 		break;
1295 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
1296 		/* Print the MCE event to host console. */
1297 		machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1298 
1299 		/*
1300 		 * If the guest can do FWNMI, exit to userspace so it can
1301 		 * deliver a FWNMI to the guest.
1302 		 * Otherwise we synthesize a machine check for the guest
1303 		 * so that it knows that the machine check occurred.
1304 		 */
1305 		if (!vcpu->kvm->arch.fwnmi_enabled) {
1306 			ulong flags = vcpu->arch.shregs.msr & 0x083c0000;
1307 			kvmppc_core_queue_machine_check(vcpu, flags);
1308 			r = RESUME_GUEST;
1309 			break;
1310 		}
1311 
1312 		/* Exit to guest with KVM_EXIT_NMI as exit reason */
1313 		run->exit_reason = KVM_EXIT_NMI;
1314 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1315 		/* Clear out the old NMI status from run->flags */
1316 		run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1317 		/* Now set the NMI status */
1318 		if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1319 			run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1320 		else
1321 			run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1322 
1323 		r = RESUME_HOST;
1324 		break;
1325 	case BOOK3S_INTERRUPT_PROGRAM:
1326 	{
1327 		ulong flags;
1328 		/*
1329 		 * Normally program interrupts are delivered directly
1330 		 * to the guest by the hardware, but we can get here
1331 		 * as a result of a hypervisor emulation interrupt
1332 		 * (e40) getting turned into a 700 by BML RTAS.
1333 		 */
1334 		flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1335 		kvmppc_core_queue_program(vcpu, flags);
1336 		r = RESUME_GUEST;
1337 		break;
1338 	}
1339 	case BOOK3S_INTERRUPT_SYSCALL:
1340 	{
1341 		/* hcall - punt to userspace */
1342 		int i;
1343 
1344 		/* hypercall with MSR_PR has already been handled in rmode,
1345 		 * and never reaches here.
1346 		 */
1347 
1348 		run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1349 		for (i = 0; i < 9; ++i)
1350 			run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1351 		run->exit_reason = KVM_EXIT_PAPR_HCALL;
1352 		vcpu->arch.hcall_needed = 1;
1353 		r = RESUME_HOST;
1354 		break;
1355 	}
1356 	/*
1357 	 * We get these next two if the guest accesses a page which it thinks
1358 	 * it has mapped but which is not actually present, either because
1359 	 * it is for an emulated I/O device or because the corresonding
1360 	 * host page has been paged out.  Any other HDSI/HISI interrupts
1361 	 * have been handled already.
1362 	 */
1363 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1364 		r = RESUME_PAGE_FAULT;
1365 		break;
1366 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
1367 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1368 		vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr &
1369 			DSISR_SRR1_MATCH_64S;
1370 		if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1371 			vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1372 		r = RESUME_PAGE_FAULT;
1373 		break;
1374 	/*
1375 	 * This occurs if the guest executes an illegal instruction.
1376 	 * If the guest debug is disabled, generate a program interrupt
1377 	 * to the guest. If guest debug is enabled, we need to check
1378 	 * whether the instruction is a software breakpoint instruction.
1379 	 * Accordingly return to Guest or Host.
1380 	 */
1381 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1382 		if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1383 			vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1384 				swab32(vcpu->arch.emul_inst) :
1385 				vcpu->arch.emul_inst;
1386 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1387 			r = kvmppc_emulate_debug_inst(run, vcpu);
1388 		} else {
1389 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1390 			r = RESUME_GUEST;
1391 		}
1392 		break;
1393 	/*
1394 	 * This occurs if the guest (kernel or userspace), does something that
1395 	 * is prohibited by HFSCR.
1396 	 * On POWER9, this could be a doorbell instruction that we need
1397 	 * to emulate.
1398 	 * Otherwise, we just generate a program interrupt to the guest.
1399 	 */
1400 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1401 		r = EMULATE_FAIL;
1402 		if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1403 		    cpu_has_feature(CPU_FTR_ARCH_300))
1404 			r = kvmppc_emulate_doorbell_instr(vcpu);
1405 		if (r == EMULATE_FAIL) {
1406 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1407 			r = RESUME_GUEST;
1408 		}
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 		r = RESUME_PASSTHROUGH;
1425 		break;
1426 	default:
1427 		kvmppc_dump_regs(vcpu);
1428 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1429 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1430 			vcpu->arch.shregs.msr);
1431 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1432 		r = RESUME_HOST;
1433 		break;
1434 	}
1435 
1436 	return r;
1437 }
1438 
1439 static int kvmppc_handle_nested_exit(struct kvm_run *run, struct kvm_vcpu *vcpu)
1440 {
1441 	int r;
1442 	int srcu_idx;
1443 
1444 	vcpu->stat.sum_exits++;
1445 
1446 	/*
1447 	 * This can happen if an interrupt occurs in the last stages
1448 	 * of guest entry or the first stages of guest exit (i.e. after
1449 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1450 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1451 	 * That can happen due to a bug, or due to a machine check
1452 	 * occurring at just the wrong time.
1453 	 */
1454 	if (vcpu->arch.shregs.msr & MSR_HV) {
1455 		pr_emerg("KVM trap in HV mode while nested!\n");
1456 		pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1457 			 vcpu->arch.trap, kvmppc_get_pc(vcpu),
1458 			 vcpu->arch.shregs.msr);
1459 		kvmppc_dump_regs(vcpu);
1460 		return RESUME_HOST;
1461 	}
1462 	switch (vcpu->arch.trap) {
1463 	/* We're good on these - the host merely wanted to get our attention */
1464 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1465 		vcpu->stat.dec_exits++;
1466 		r = RESUME_GUEST;
1467 		break;
1468 	case BOOK3S_INTERRUPT_EXTERNAL:
1469 		vcpu->stat.ext_intr_exits++;
1470 		r = RESUME_HOST;
1471 		break;
1472 	case BOOK3S_INTERRUPT_H_DOORBELL:
1473 	case BOOK3S_INTERRUPT_H_VIRT:
1474 		vcpu->stat.ext_intr_exits++;
1475 		r = RESUME_GUEST;
1476 		break;
1477 	/* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1478 	case BOOK3S_INTERRUPT_HMI:
1479 	case BOOK3S_INTERRUPT_PERFMON:
1480 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
1481 		r = RESUME_GUEST;
1482 		break;
1483 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
1484 		/* Pass the machine check to the L1 guest */
1485 		r = RESUME_HOST;
1486 		/* Print the MCE event to host console. */
1487 		machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1488 		break;
1489 	/*
1490 	 * We get these next two if the guest accesses a page which it thinks
1491 	 * it has mapped but which is not actually present, either because
1492 	 * it is for an emulated I/O device or because the corresonding
1493 	 * host page has been paged out.
1494 	 */
1495 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1496 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1497 		r = kvmhv_nested_page_fault(run, vcpu);
1498 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1499 		break;
1500 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
1501 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1502 		vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1503 					 DSISR_SRR1_MATCH_64S;
1504 		if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1505 			vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1506 		srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1507 		r = kvmhv_nested_page_fault(run, vcpu);
1508 		srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1509 		break;
1510 
1511 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1512 	case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1513 		/*
1514 		 * This occurs for various TM-related instructions that
1515 		 * we need to emulate on POWER9 DD2.2.  We have already
1516 		 * handled the cases where the guest was in real-suspend
1517 		 * mode and was transitioning to transactional state.
1518 		 */
1519 		r = kvmhv_p9_tm_emulation(vcpu);
1520 		break;
1521 #endif
1522 
1523 	case BOOK3S_INTERRUPT_HV_RM_HARD:
1524 		vcpu->arch.trap = 0;
1525 		r = RESUME_GUEST;
1526 		if (!xics_on_xive())
1527 			kvmppc_xics_rm_complete(vcpu, 0);
1528 		break;
1529 	default:
1530 		r = RESUME_HOST;
1531 		break;
1532 	}
1533 
1534 	return r;
1535 }
1536 
1537 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1538 					    struct kvm_sregs *sregs)
1539 {
1540 	int i;
1541 
1542 	memset(sregs, 0, sizeof(struct kvm_sregs));
1543 	sregs->pvr = vcpu->arch.pvr;
1544 	for (i = 0; i < vcpu->arch.slb_max; i++) {
1545 		sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1546 		sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1547 	}
1548 
1549 	return 0;
1550 }
1551 
1552 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1553 					    struct kvm_sregs *sregs)
1554 {
1555 	int i, j;
1556 
1557 	/* Only accept the same PVR as the host's, since we can't spoof it */
1558 	if (sregs->pvr != vcpu->arch.pvr)
1559 		return -EINVAL;
1560 
1561 	j = 0;
1562 	for (i = 0; i < vcpu->arch.slb_nr; i++) {
1563 		if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1564 			vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1565 			vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1566 			++j;
1567 		}
1568 	}
1569 	vcpu->arch.slb_max = j;
1570 
1571 	return 0;
1572 }
1573 
1574 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1575 		bool preserve_top32)
1576 {
1577 	struct kvm *kvm = vcpu->kvm;
1578 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
1579 	u64 mask;
1580 
1581 	spin_lock(&vc->lock);
1582 	/*
1583 	 * If ILE (interrupt little-endian) has changed, update the
1584 	 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1585 	 */
1586 	if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1587 		struct kvm_vcpu *vcpu;
1588 		int i;
1589 
1590 		kvm_for_each_vcpu(i, vcpu, kvm) {
1591 			if (vcpu->arch.vcore != vc)
1592 				continue;
1593 			if (new_lpcr & LPCR_ILE)
1594 				vcpu->arch.intr_msr |= MSR_LE;
1595 			else
1596 				vcpu->arch.intr_msr &= ~MSR_LE;
1597 		}
1598 	}
1599 
1600 	/*
1601 	 * Userspace can only modify DPFD (default prefetch depth),
1602 	 * ILE (interrupt little-endian) and TC (translation control).
1603 	 * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1604 	 */
1605 	mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1606 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
1607 		mask |= LPCR_AIL;
1608 	/*
1609 	 * On POWER9, allow userspace to enable large decrementer for the
1610 	 * guest, whether or not the host has it enabled.
1611 	 */
1612 	if (cpu_has_feature(CPU_FTR_ARCH_300))
1613 		mask |= LPCR_LD;
1614 
1615 	/* Broken 32-bit version of LPCR must not clear top bits */
1616 	if (preserve_top32)
1617 		mask &= 0xFFFFFFFF;
1618 	vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1619 	spin_unlock(&vc->lock);
1620 }
1621 
1622 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1623 				 union kvmppc_one_reg *val)
1624 {
1625 	int r = 0;
1626 	long int i;
1627 
1628 	switch (id) {
1629 	case KVM_REG_PPC_DEBUG_INST:
1630 		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1631 		break;
1632 	case KVM_REG_PPC_HIOR:
1633 		*val = get_reg_val(id, 0);
1634 		break;
1635 	case KVM_REG_PPC_DABR:
1636 		*val = get_reg_val(id, vcpu->arch.dabr);
1637 		break;
1638 	case KVM_REG_PPC_DABRX:
1639 		*val = get_reg_val(id, vcpu->arch.dabrx);
1640 		break;
1641 	case KVM_REG_PPC_DSCR:
1642 		*val = get_reg_val(id, vcpu->arch.dscr);
1643 		break;
1644 	case KVM_REG_PPC_PURR:
1645 		*val = get_reg_val(id, vcpu->arch.purr);
1646 		break;
1647 	case KVM_REG_PPC_SPURR:
1648 		*val = get_reg_val(id, vcpu->arch.spurr);
1649 		break;
1650 	case KVM_REG_PPC_AMR:
1651 		*val = get_reg_val(id, vcpu->arch.amr);
1652 		break;
1653 	case KVM_REG_PPC_UAMOR:
1654 		*val = get_reg_val(id, vcpu->arch.uamor);
1655 		break;
1656 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1657 		i = id - KVM_REG_PPC_MMCR0;
1658 		*val = get_reg_val(id, vcpu->arch.mmcr[i]);
1659 		break;
1660 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1661 		i = id - KVM_REG_PPC_PMC1;
1662 		*val = get_reg_val(id, vcpu->arch.pmc[i]);
1663 		break;
1664 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1665 		i = id - KVM_REG_PPC_SPMC1;
1666 		*val = get_reg_val(id, vcpu->arch.spmc[i]);
1667 		break;
1668 	case KVM_REG_PPC_SIAR:
1669 		*val = get_reg_val(id, vcpu->arch.siar);
1670 		break;
1671 	case KVM_REG_PPC_SDAR:
1672 		*val = get_reg_val(id, vcpu->arch.sdar);
1673 		break;
1674 	case KVM_REG_PPC_SIER:
1675 		*val = get_reg_val(id, vcpu->arch.sier);
1676 		break;
1677 	case KVM_REG_PPC_IAMR:
1678 		*val = get_reg_val(id, vcpu->arch.iamr);
1679 		break;
1680 	case KVM_REG_PPC_PSPB:
1681 		*val = get_reg_val(id, vcpu->arch.pspb);
1682 		break;
1683 	case KVM_REG_PPC_DPDES:
1684 		*val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1685 		break;
1686 	case KVM_REG_PPC_VTB:
1687 		*val = get_reg_val(id, vcpu->arch.vcore->vtb);
1688 		break;
1689 	case KVM_REG_PPC_DAWR:
1690 		*val = get_reg_val(id, vcpu->arch.dawr);
1691 		break;
1692 	case KVM_REG_PPC_DAWRX:
1693 		*val = get_reg_val(id, vcpu->arch.dawrx);
1694 		break;
1695 	case KVM_REG_PPC_CIABR:
1696 		*val = get_reg_val(id, vcpu->arch.ciabr);
1697 		break;
1698 	case KVM_REG_PPC_CSIGR:
1699 		*val = get_reg_val(id, vcpu->arch.csigr);
1700 		break;
1701 	case KVM_REG_PPC_TACR:
1702 		*val = get_reg_val(id, vcpu->arch.tacr);
1703 		break;
1704 	case KVM_REG_PPC_TCSCR:
1705 		*val = get_reg_val(id, vcpu->arch.tcscr);
1706 		break;
1707 	case KVM_REG_PPC_PID:
1708 		*val = get_reg_val(id, vcpu->arch.pid);
1709 		break;
1710 	case KVM_REG_PPC_ACOP:
1711 		*val = get_reg_val(id, vcpu->arch.acop);
1712 		break;
1713 	case KVM_REG_PPC_WORT:
1714 		*val = get_reg_val(id, vcpu->arch.wort);
1715 		break;
1716 	case KVM_REG_PPC_TIDR:
1717 		*val = get_reg_val(id, vcpu->arch.tid);
1718 		break;
1719 	case KVM_REG_PPC_PSSCR:
1720 		*val = get_reg_val(id, vcpu->arch.psscr);
1721 		break;
1722 	case KVM_REG_PPC_VPA_ADDR:
1723 		spin_lock(&vcpu->arch.vpa_update_lock);
1724 		*val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1725 		spin_unlock(&vcpu->arch.vpa_update_lock);
1726 		break;
1727 	case KVM_REG_PPC_VPA_SLB:
1728 		spin_lock(&vcpu->arch.vpa_update_lock);
1729 		val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1730 		val->vpaval.length = vcpu->arch.slb_shadow.len;
1731 		spin_unlock(&vcpu->arch.vpa_update_lock);
1732 		break;
1733 	case KVM_REG_PPC_VPA_DTL:
1734 		spin_lock(&vcpu->arch.vpa_update_lock);
1735 		val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1736 		val->vpaval.length = vcpu->arch.dtl.len;
1737 		spin_unlock(&vcpu->arch.vpa_update_lock);
1738 		break;
1739 	case KVM_REG_PPC_TB_OFFSET:
1740 		*val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1741 		break;
1742 	case KVM_REG_PPC_LPCR:
1743 	case KVM_REG_PPC_LPCR_64:
1744 		*val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1745 		break;
1746 	case KVM_REG_PPC_PPR:
1747 		*val = get_reg_val(id, vcpu->arch.ppr);
1748 		break;
1749 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1750 	case KVM_REG_PPC_TFHAR:
1751 		*val = get_reg_val(id, vcpu->arch.tfhar);
1752 		break;
1753 	case KVM_REG_PPC_TFIAR:
1754 		*val = get_reg_val(id, vcpu->arch.tfiar);
1755 		break;
1756 	case KVM_REG_PPC_TEXASR:
1757 		*val = get_reg_val(id, vcpu->arch.texasr);
1758 		break;
1759 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1760 		i = id - KVM_REG_PPC_TM_GPR0;
1761 		*val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1762 		break;
1763 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1764 	{
1765 		int j;
1766 		i = id - KVM_REG_PPC_TM_VSR0;
1767 		if (i < 32)
1768 			for (j = 0; j < TS_FPRWIDTH; j++)
1769 				val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1770 		else {
1771 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1772 				val->vval = vcpu->arch.vr_tm.vr[i-32];
1773 			else
1774 				r = -ENXIO;
1775 		}
1776 		break;
1777 	}
1778 	case KVM_REG_PPC_TM_CR:
1779 		*val = get_reg_val(id, vcpu->arch.cr_tm);
1780 		break;
1781 	case KVM_REG_PPC_TM_XER:
1782 		*val = get_reg_val(id, vcpu->arch.xer_tm);
1783 		break;
1784 	case KVM_REG_PPC_TM_LR:
1785 		*val = get_reg_val(id, vcpu->arch.lr_tm);
1786 		break;
1787 	case KVM_REG_PPC_TM_CTR:
1788 		*val = get_reg_val(id, vcpu->arch.ctr_tm);
1789 		break;
1790 	case KVM_REG_PPC_TM_FPSCR:
1791 		*val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1792 		break;
1793 	case KVM_REG_PPC_TM_AMR:
1794 		*val = get_reg_val(id, vcpu->arch.amr_tm);
1795 		break;
1796 	case KVM_REG_PPC_TM_PPR:
1797 		*val = get_reg_val(id, vcpu->arch.ppr_tm);
1798 		break;
1799 	case KVM_REG_PPC_TM_VRSAVE:
1800 		*val = get_reg_val(id, vcpu->arch.vrsave_tm);
1801 		break;
1802 	case KVM_REG_PPC_TM_VSCR:
1803 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1804 			*val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1805 		else
1806 			r = -ENXIO;
1807 		break;
1808 	case KVM_REG_PPC_TM_DSCR:
1809 		*val = get_reg_val(id, vcpu->arch.dscr_tm);
1810 		break;
1811 	case KVM_REG_PPC_TM_TAR:
1812 		*val = get_reg_val(id, vcpu->arch.tar_tm);
1813 		break;
1814 #endif
1815 	case KVM_REG_PPC_ARCH_COMPAT:
1816 		*val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1817 		break;
1818 	case KVM_REG_PPC_DEC_EXPIRY:
1819 		*val = get_reg_val(id, vcpu->arch.dec_expires +
1820 				   vcpu->arch.vcore->tb_offset);
1821 		break;
1822 	case KVM_REG_PPC_ONLINE:
1823 		*val = get_reg_val(id, vcpu->arch.online);
1824 		break;
1825 	case KVM_REG_PPC_PTCR:
1826 		*val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
1827 		break;
1828 	default:
1829 		r = -EINVAL;
1830 		break;
1831 	}
1832 
1833 	return r;
1834 }
1835 
1836 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1837 				 union kvmppc_one_reg *val)
1838 {
1839 	int r = 0;
1840 	long int i;
1841 	unsigned long addr, len;
1842 
1843 	switch (id) {
1844 	case KVM_REG_PPC_HIOR:
1845 		/* Only allow this to be set to zero */
1846 		if (set_reg_val(id, *val))
1847 			r = -EINVAL;
1848 		break;
1849 	case KVM_REG_PPC_DABR:
1850 		vcpu->arch.dabr = set_reg_val(id, *val);
1851 		break;
1852 	case KVM_REG_PPC_DABRX:
1853 		vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1854 		break;
1855 	case KVM_REG_PPC_DSCR:
1856 		vcpu->arch.dscr = set_reg_val(id, *val);
1857 		break;
1858 	case KVM_REG_PPC_PURR:
1859 		vcpu->arch.purr = set_reg_val(id, *val);
1860 		break;
1861 	case KVM_REG_PPC_SPURR:
1862 		vcpu->arch.spurr = set_reg_val(id, *val);
1863 		break;
1864 	case KVM_REG_PPC_AMR:
1865 		vcpu->arch.amr = set_reg_val(id, *val);
1866 		break;
1867 	case KVM_REG_PPC_UAMOR:
1868 		vcpu->arch.uamor = set_reg_val(id, *val);
1869 		break;
1870 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1871 		i = id - KVM_REG_PPC_MMCR0;
1872 		vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1873 		break;
1874 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1875 		i = id - KVM_REG_PPC_PMC1;
1876 		vcpu->arch.pmc[i] = set_reg_val(id, *val);
1877 		break;
1878 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1879 		i = id - KVM_REG_PPC_SPMC1;
1880 		vcpu->arch.spmc[i] = set_reg_val(id, *val);
1881 		break;
1882 	case KVM_REG_PPC_SIAR:
1883 		vcpu->arch.siar = set_reg_val(id, *val);
1884 		break;
1885 	case KVM_REG_PPC_SDAR:
1886 		vcpu->arch.sdar = set_reg_val(id, *val);
1887 		break;
1888 	case KVM_REG_PPC_SIER:
1889 		vcpu->arch.sier = set_reg_val(id, *val);
1890 		break;
1891 	case KVM_REG_PPC_IAMR:
1892 		vcpu->arch.iamr = set_reg_val(id, *val);
1893 		break;
1894 	case KVM_REG_PPC_PSPB:
1895 		vcpu->arch.pspb = set_reg_val(id, *val);
1896 		break;
1897 	case KVM_REG_PPC_DPDES:
1898 		vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1899 		break;
1900 	case KVM_REG_PPC_VTB:
1901 		vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1902 		break;
1903 	case KVM_REG_PPC_DAWR:
1904 		vcpu->arch.dawr = set_reg_val(id, *val);
1905 		break;
1906 	case KVM_REG_PPC_DAWRX:
1907 		vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1908 		break;
1909 	case KVM_REG_PPC_CIABR:
1910 		vcpu->arch.ciabr = set_reg_val(id, *val);
1911 		/* Don't allow setting breakpoints in hypervisor code */
1912 		if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1913 			vcpu->arch.ciabr &= ~CIABR_PRIV;	/* disable */
1914 		break;
1915 	case KVM_REG_PPC_CSIGR:
1916 		vcpu->arch.csigr = set_reg_val(id, *val);
1917 		break;
1918 	case KVM_REG_PPC_TACR:
1919 		vcpu->arch.tacr = set_reg_val(id, *val);
1920 		break;
1921 	case KVM_REG_PPC_TCSCR:
1922 		vcpu->arch.tcscr = set_reg_val(id, *val);
1923 		break;
1924 	case KVM_REG_PPC_PID:
1925 		vcpu->arch.pid = set_reg_val(id, *val);
1926 		break;
1927 	case KVM_REG_PPC_ACOP:
1928 		vcpu->arch.acop = set_reg_val(id, *val);
1929 		break;
1930 	case KVM_REG_PPC_WORT:
1931 		vcpu->arch.wort = set_reg_val(id, *val);
1932 		break;
1933 	case KVM_REG_PPC_TIDR:
1934 		vcpu->arch.tid = set_reg_val(id, *val);
1935 		break;
1936 	case KVM_REG_PPC_PSSCR:
1937 		vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1938 		break;
1939 	case KVM_REG_PPC_VPA_ADDR:
1940 		addr = set_reg_val(id, *val);
1941 		r = -EINVAL;
1942 		if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1943 			      vcpu->arch.dtl.next_gpa))
1944 			break;
1945 		r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1946 		break;
1947 	case KVM_REG_PPC_VPA_SLB:
1948 		addr = val->vpaval.addr;
1949 		len = val->vpaval.length;
1950 		r = -EINVAL;
1951 		if (addr && !vcpu->arch.vpa.next_gpa)
1952 			break;
1953 		r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1954 		break;
1955 	case KVM_REG_PPC_VPA_DTL:
1956 		addr = val->vpaval.addr;
1957 		len = val->vpaval.length;
1958 		r = -EINVAL;
1959 		if (addr && (len < sizeof(struct dtl_entry) ||
1960 			     !vcpu->arch.vpa.next_gpa))
1961 			break;
1962 		len -= len % sizeof(struct dtl_entry);
1963 		r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1964 		break;
1965 	case KVM_REG_PPC_TB_OFFSET:
1966 		/* round up to multiple of 2^24 */
1967 		vcpu->arch.vcore->tb_offset =
1968 			ALIGN(set_reg_val(id, *val), 1UL << 24);
1969 		break;
1970 	case KVM_REG_PPC_LPCR:
1971 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1972 		break;
1973 	case KVM_REG_PPC_LPCR_64:
1974 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1975 		break;
1976 	case KVM_REG_PPC_PPR:
1977 		vcpu->arch.ppr = set_reg_val(id, *val);
1978 		break;
1979 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1980 	case KVM_REG_PPC_TFHAR:
1981 		vcpu->arch.tfhar = set_reg_val(id, *val);
1982 		break;
1983 	case KVM_REG_PPC_TFIAR:
1984 		vcpu->arch.tfiar = set_reg_val(id, *val);
1985 		break;
1986 	case KVM_REG_PPC_TEXASR:
1987 		vcpu->arch.texasr = set_reg_val(id, *val);
1988 		break;
1989 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1990 		i = id - KVM_REG_PPC_TM_GPR0;
1991 		vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1992 		break;
1993 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1994 	{
1995 		int j;
1996 		i = id - KVM_REG_PPC_TM_VSR0;
1997 		if (i < 32)
1998 			for (j = 0; j < TS_FPRWIDTH; j++)
1999 				vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
2000 		else
2001 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
2002 				vcpu->arch.vr_tm.vr[i-32] = val->vval;
2003 			else
2004 				r = -ENXIO;
2005 		break;
2006 	}
2007 	case KVM_REG_PPC_TM_CR:
2008 		vcpu->arch.cr_tm = set_reg_val(id, *val);
2009 		break;
2010 	case KVM_REG_PPC_TM_XER:
2011 		vcpu->arch.xer_tm = set_reg_val(id, *val);
2012 		break;
2013 	case KVM_REG_PPC_TM_LR:
2014 		vcpu->arch.lr_tm = set_reg_val(id, *val);
2015 		break;
2016 	case KVM_REG_PPC_TM_CTR:
2017 		vcpu->arch.ctr_tm = set_reg_val(id, *val);
2018 		break;
2019 	case KVM_REG_PPC_TM_FPSCR:
2020 		vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
2021 		break;
2022 	case KVM_REG_PPC_TM_AMR:
2023 		vcpu->arch.amr_tm = set_reg_val(id, *val);
2024 		break;
2025 	case KVM_REG_PPC_TM_PPR:
2026 		vcpu->arch.ppr_tm = set_reg_val(id, *val);
2027 		break;
2028 	case KVM_REG_PPC_TM_VRSAVE:
2029 		vcpu->arch.vrsave_tm = set_reg_val(id, *val);
2030 		break;
2031 	case KVM_REG_PPC_TM_VSCR:
2032 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
2033 			vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
2034 		else
2035 			r = - ENXIO;
2036 		break;
2037 	case KVM_REG_PPC_TM_DSCR:
2038 		vcpu->arch.dscr_tm = set_reg_val(id, *val);
2039 		break;
2040 	case KVM_REG_PPC_TM_TAR:
2041 		vcpu->arch.tar_tm = set_reg_val(id, *val);
2042 		break;
2043 #endif
2044 	case KVM_REG_PPC_ARCH_COMPAT:
2045 		r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
2046 		break;
2047 	case KVM_REG_PPC_DEC_EXPIRY:
2048 		vcpu->arch.dec_expires = set_reg_val(id, *val) -
2049 			vcpu->arch.vcore->tb_offset;
2050 		break;
2051 	case KVM_REG_PPC_ONLINE:
2052 		i = set_reg_val(id, *val);
2053 		if (i && !vcpu->arch.online)
2054 			atomic_inc(&vcpu->arch.vcore->online_count);
2055 		else if (!i && vcpu->arch.online)
2056 			atomic_dec(&vcpu->arch.vcore->online_count);
2057 		vcpu->arch.online = i;
2058 		break;
2059 	case KVM_REG_PPC_PTCR:
2060 		vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
2061 		break;
2062 	default:
2063 		r = -EINVAL;
2064 		break;
2065 	}
2066 
2067 	return r;
2068 }
2069 
2070 /*
2071  * On POWER9, threads are independent and can be in different partitions.
2072  * Therefore we consider each thread to be a subcore.
2073  * There is a restriction that all threads have to be in the same
2074  * MMU mode (radix or HPT), unfortunately, but since we only support
2075  * HPT guests on a HPT host so far, that isn't an impediment yet.
2076  */
2077 static int threads_per_vcore(struct kvm *kvm)
2078 {
2079 	if (kvm->arch.threads_indep)
2080 		return 1;
2081 	return threads_per_subcore;
2082 }
2083 
2084 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
2085 {
2086 	struct kvmppc_vcore *vcore;
2087 
2088 	vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
2089 
2090 	if (vcore == NULL)
2091 		return NULL;
2092 
2093 	spin_lock_init(&vcore->lock);
2094 	spin_lock_init(&vcore->stoltb_lock);
2095 	init_swait_queue_head(&vcore->wq);
2096 	vcore->preempt_tb = TB_NIL;
2097 	vcore->lpcr = kvm->arch.lpcr;
2098 	vcore->first_vcpuid = id;
2099 	vcore->kvm = kvm;
2100 	INIT_LIST_HEAD(&vcore->preempt_list);
2101 
2102 	return vcore;
2103 }
2104 
2105 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2106 static struct debugfs_timings_element {
2107 	const char *name;
2108 	size_t offset;
2109 } timings[] = {
2110 	{"rm_entry",	offsetof(struct kvm_vcpu, arch.rm_entry)},
2111 	{"rm_intr",	offsetof(struct kvm_vcpu, arch.rm_intr)},
2112 	{"rm_exit",	offsetof(struct kvm_vcpu, arch.rm_exit)},
2113 	{"guest",	offsetof(struct kvm_vcpu, arch.guest_time)},
2114 	{"cede",	offsetof(struct kvm_vcpu, arch.cede_time)},
2115 };
2116 
2117 #define N_TIMINGS	(ARRAY_SIZE(timings))
2118 
2119 struct debugfs_timings_state {
2120 	struct kvm_vcpu	*vcpu;
2121 	unsigned int	buflen;
2122 	char		buf[N_TIMINGS * 100];
2123 };
2124 
2125 static int debugfs_timings_open(struct inode *inode, struct file *file)
2126 {
2127 	struct kvm_vcpu *vcpu = inode->i_private;
2128 	struct debugfs_timings_state *p;
2129 
2130 	p = kzalloc(sizeof(*p), GFP_KERNEL);
2131 	if (!p)
2132 		return -ENOMEM;
2133 
2134 	kvm_get_kvm(vcpu->kvm);
2135 	p->vcpu = vcpu;
2136 	file->private_data = p;
2137 
2138 	return nonseekable_open(inode, file);
2139 }
2140 
2141 static int debugfs_timings_release(struct inode *inode, struct file *file)
2142 {
2143 	struct debugfs_timings_state *p = file->private_data;
2144 
2145 	kvm_put_kvm(p->vcpu->kvm);
2146 	kfree(p);
2147 	return 0;
2148 }
2149 
2150 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2151 				    size_t len, loff_t *ppos)
2152 {
2153 	struct debugfs_timings_state *p = file->private_data;
2154 	struct kvm_vcpu *vcpu = p->vcpu;
2155 	char *s, *buf_end;
2156 	struct kvmhv_tb_accumulator tb;
2157 	u64 count;
2158 	loff_t pos;
2159 	ssize_t n;
2160 	int i, loops;
2161 	bool ok;
2162 
2163 	if (!p->buflen) {
2164 		s = p->buf;
2165 		buf_end = s + sizeof(p->buf);
2166 		for (i = 0; i < N_TIMINGS; ++i) {
2167 			struct kvmhv_tb_accumulator *acc;
2168 
2169 			acc = (struct kvmhv_tb_accumulator *)
2170 				((unsigned long)vcpu + timings[i].offset);
2171 			ok = false;
2172 			for (loops = 0; loops < 1000; ++loops) {
2173 				count = acc->seqcount;
2174 				if (!(count & 1)) {
2175 					smp_rmb();
2176 					tb = *acc;
2177 					smp_rmb();
2178 					if (count == acc->seqcount) {
2179 						ok = true;
2180 						break;
2181 					}
2182 				}
2183 				udelay(1);
2184 			}
2185 			if (!ok)
2186 				snprintf(s, buf_end - s, "%s: stuck\n",
2187 					timings[i].name);
2188 			else
2189 				snprintf(s, buf_end - s,
2190 					"%s: %llu %llu %llu %llu\n",
2191 					timings[i].name, count / 2,
2192 					tb_to_ns(tb.tb_total),
2193 					tb_to_ns(tb.tb_min),
2194 					tb_to_ns(tb.tb_max));
2195 			s += strlen(s);
2196 		}
2197 		p->buflen = s - p->buf;
2198 	}
2199 
2200 	pos = *ppos;
2201 	if (pos >= p->buflen)
2202 		return 0;
2203 	if (len > p->buflen - pos)
2204 		len = p->buflen - pos;
2205 	n = copy_to_user(buf, p->buf + pos, len);
2206 	if (n) {
2207 		if (n == len)
2208 			return -EFAULT;
2209 		len -= n;
2210 	}
2211 	*ppos = pos + len;
2212 	return len;
2213 }
2214 
2215 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2216 				     size_t len, loff_t *ppos)
2217 {
2218 	return -EACCES;
2219 }
2220 
2221 static const struct file_operations debugfs_timings_ops = {
2222 	.owner	 = THIS_MODULE,
2223 	.open	 = debugfs_timings_open,
2224 	.release = debugfs_timings_release,
2225 	.read	 = debugfs_timings_read,
2226 	.write	 = debugfs_timings_write,
2227 	.llseek	 = generic_file_llseek,
2228 };
2229 
2230 /* Create a debugfs directory for the vcpu */
2231 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2232 {
2233 	char buf[16];
2234 	struct kvm *kvm = vcpu->kvm;
2235 
2236 	snprintf(buf, sizeof(buf), "vcpu%u", id);
2237 	if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
2238 		return;
2239 	vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
2240 	if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
2241 		return;
2242 	vcpu->arch.debugfs_timings =
2243 		debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
2244 				    vcpu, &debugfs_timings_ops);
2245 }
2246 
2247 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2248 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2249 {
2250 }
2251 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2252 
2253 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
2254 						   unsigned int id)
2255 {
2256 	struct kvm_vcpu *vcpu;
2257 	int err;
2258 	int core;
2259 	struct kvmppc_vcore *vcore;
2260 
2261 	err = -ENOMEM;
2262 	vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2263 	if (!vcpu)
2264 		goto out;
2265 
2266 	err = kvm_vcpu_init(vcpu, kvm, id);
2267 	if (err)
2268 		goto free_vcpu;
2269 
2270 	vcpu->arch.shared = &vcpu->arch.shregs;
2271 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2272 	/*
2273 	 * The shared struct is never shared on HV,
2274 	 * so we can always use host endianness
2275 	 */
2276 #ifdef __BIG_ENDIAN__
2277 	vcpu->arch.shared_big_endian = true;
2278 #else
2279 	vcpu->arch.shared_big_endian = false;
2280 #endif
2281 #endif
2282 	vcpu->arch.mmcr[0] = MMCR0_FC;
2283 	vcpu->arch.ctrl = CTRL_RUNLATCH;
2284 	/* default to host PVR, since we can't spoof it */
2285 	kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2286 	spin_lock_init(&vcpu->arch.vpa_update_lock);
2287 	spin_lock_init(&vcpu->arch.tbacct_lock);
2288 	vcpu->arch.busy_preempt = TB_NIL;
2289 	vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2290 
2291 	/*
2292 	 * Set the default HFSCR for the guest from the host value.
2293 	 * This value is only used on POWER9.
2294 	 * On POWER9, we want to virtualize the doorbell facility, so we
2295 	 * don't set the HFSCR_MSGP bit, and that causes those instructions
2296 	 * to trap and then we emulate them.
2297 	 */
2298 	vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2299 		HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP;
2300 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
2301 		vcpu->arch.hfscr &= mfspr(SPRN_HFSCR);
2302 		if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2303 			vcpu->arch.hfscr |= HFSCR_TM;
2304 	}
2305 	if (cpu_has_feature(CPU_FTR_TM_COMP))
2306 		vcpu->arch.hfscr |= HFSCR_TM;
2307 
2308 	kvmppc_mmu_book3s_hv_init(vcpu);
2309 
2310 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2311 
2312 	init_waitqueue_head(&vcpu->arch.cpu_run);
2313 
2314 	mutex_lock(&kvm->lock);
2315 	vcore = NULL;
2316 	err = -EINVAL;
2317 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2318 		if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2319 			pr_devel("KVM: VCPU ID too high\n");
2320 			core = KVM_MAX_VCORES;
2321 		} else {
2322 			BUG_ON(kvm->arch.smt_mode != 1);
2323 			core = kvmppc_pack_vcpu_id(kvm, id);
2324 		}
2325 	} else {
2326 		core = id / kvm->arch.smt_mode;
2327 	}
2328 	if (core < KVM_MAX_VCORES) {
2329 		vcore = kvm->arch.vcores[core];
2330 		if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2331 			pr_devel("KVM: collision on id %u", id);
2332 			vcore = NULL;
2333 		} else if (!vcore) {
2334 			/*
2335 			 * Take mmu_setup_lock for mutual exclusion
2336 			 * with kvmppc_update_lpcr().
2337 			 */
2338 			err = -ENOMEM;
2339 			vcore = kvmppc_vcore_create(kvm,
2340 					id & ~(kvm->arch.smt_mode - 1));
2341 			mutex_lock(&kvm->arch.mmu_setup_lock);
2342 			kvm->arch.vcores[core] = vcore;
2343 			kvm->arch.online_vcores++;
2344 			mutex_unlock(&kvm->arch.mmu_setup_lock);
2345 		}
2346 	}
2347 	mutex_unlock(&kvm->lock);
2348 
2349 	if (!vcore)
2350 		goto free_vcpu;
2351 
2352 	spin_lock(&vcore->lock);
2353 	++vcore->num_threads;
2354 	spin_unlock(&vcore->lock);
2355 	vcpu->arch.vcore = vcore;
2356 	vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2357 	vcpu->arch.thread_cpu = -1;
2358 	vcpu->arch.prev_cpu = -1;
2359 
2360 	vcpu->arch.cpu_type = KVM_CPU_3S_64;
2361 	kvmppc_sanity_check(vcpu);
2362 
2363 	debugfs_vcpu_init(vcpu, id);
2364 
2365 	return vcpu;
2366 
2367 free_vcpu:
2368 	kmem_cache_free(kvm_vcpu_cache, vcpu);
2369 out:
2370 	return ERR_PTR(err);
2371 }
2372 
2373 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2374 			      unsigned long flags)
2375 {
2376 	int err;
2377 	int esmt = 0;
2378 
2379 	if (flags)
2380 		return -EINVAL;
2381 	if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2382 		return -EINVAL;
2383 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2384 		/*
2385 		 * On POWER8 (or POWER7), the threading mode is "strict",
2386 		 * so we pack smt_mode vcpus per vcore.
2387 		 */
2388 		if (smt_mode > threads_per_subcore)
2389 			return -EINVAL;
2390 	} else {
2391 		/*
2392 		 * On POWER9, the threading mode is "loose",
2393 		 * so each vcpu gets its own vcore.
2394 		 */
2395 		esmt = smt_mode;
2396 		smt_mode = 1;
2397 	}
2398 	mutex_lock(&kvm->lock);
2399 	err = -EBUSY;
2400 	if (!kvm->arch.online_vcores) {
2401 		kvm->arch.smt_mode = smt_mode;
2402 		kvm->arch.emul_smt_mode = esmt;
2403 		err = 0;
2404 	}
2405 	mutex_unlock(&kvm->lock);
2406 
2407 	return err;
2408 }
2409 
2410 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2411 {
2412 	if (vpa->pinned_addr)
2413 		kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2414 					vpa->dirty);
2415 }
2416 
2417 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2418 {
2419 	spin_lock(&vcpu->arch.vpa_update_lock);
2420 	unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2421 	unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2422 	unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2423 	spin_unlock(&vcpu->arch.vpa_update_lock);
2424 	kvm_vcpu_uninit(vcpu);
2425 	kmem_cache_free(kvm_vcpu_cache, vcpu);
2426 }
2427 
2428 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2429 {
2430 	/* Indicate we want to get back into the guest */
2431 	return 1;
2432 }
2433 
2434 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2435 {
2436 	unsigned long dec_nsec, now;
2437 
2438 	now = get_tb();
2439 	if (now > vcpu->arch.dec_expires) {
2440 		/* decrementer has already gone negative */
2441 		kvmppc_core_queue_dec(vcpu);
2442 		kvmppc_core_prepare_to_enter(vcpu);
2443 		return;
2444 	}
2445 	dec_nsec = tb_to_ns(vcpu->arch.dec_expires - now);
2446 	hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2447 	vcpu->arch.timer_running = 1;
2448 }
2449 
2450 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
2451 {
2452 	vcpu->arch.ceded = 0;
2453 	if (vcpu->arch.timer_running) {
2454 		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2455 		vcpu->arch.timer_running = 0;
2456 	}
2457 }
2458 
2459 extern int __kvmppc_vcore_entry(void);
2460 
2461 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2462 				   struct kvm_vcpu *vcpu)
2463 {
2464 	u64 now;
2465 
2466 	if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2467 		return;
2468 	spin_lock_irq(&vcpu->arch.tbacct_lock);
2469 	now = mftb();
2470 	vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2471 		vcpu->arch.stolen_logged;
2472 	vcpu->arch.busy_preempt = now;
2473 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2474 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
2475 	--vc->n_runnable;
2476 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2477 }
2478 
2479 static int kvmppc_grab_hwthread(int cpu)
2480 {
2481 	struct paca_struct *tpaca;
2482 	long timeout = 10000;
2483 
2484 	tpaca = paca_ptrs[cpu];
2485 
2486 	/* Ensure the thread won't go into the kernel if it wakes */
2487 	tpaca->kvm_hstate.kvm_vcpu = NULL;
2488 	tpaca->kvm_hstate.kvm_vcore = NULL;
2489 	tpaca->kvm_hstate.napping = 0;
2490 	smp_wmb();
2491 	tpaca->kvm_hstate.hwthread_req = 1;
2492 
2493 	/*
2494 	 * If the thread is already executing in the kernel (e.g. handling
2495 	 * a stray interrupt), wait for it to get back to nap mode.
2496 	 * The smp_mb() is to ensure that our setting of hwthread_req
2497 	 * is visible before we look at hwthread_state, so if this
2498 	 * races with the code at system_reset_pSeries and the thread
2499 	 * misses our setting of hwthread_req, we are sure to see its
2500 	 * setting of hwthread_state, and vice versa.
2501 	 */
2502 	smp_mb();
2503 	while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2504 		if (--timeout <= 0) {
2505 			pr_err("KVM: couldn't grab cpu %d\n", cpu);
2506 			return -EBUSY;
2507 		}
2508 		udelay(1);
2509 	}
2510 	return 0;
2511 }
2512 
2513 static void kvmppc_release_hwthread(int cpu)
2514 {
2515 	struct paca_struct *tpaca;
2516 
2517 	tpaca = paca_ptrs[cpu];
2518 	tpaca->kvm_hstate.hwthread_req = 0;
2519 	tpaca->kvm_hstate.kvm_vcpu = NULL;
2520 	tpaca->kvm_hstate.kvm_vcore = NULL;
2521 	tpaca->kvm_hstate.kvm_split_mode = NULL;
2522 }
2523 
2524 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2525 {
2526 	struct kvm_nested_guest *nested = vcpu->arch.nested;
2527 	cpumask_t *cpu_in_guest;
2528 	int i;
2529 
2530 	cpu = cpu_first_thread_sibling(cpu);
2531 	if (nested) {
2532 		cpumask_set_cpu(cpu, &nested->need_tlb_flush);
2533 		cpu_in_guest = &nested->cpu_in_guest;
2534 	} else {
2535 		cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2536 		cpu_in_guest = &kvm->arch.cpu_in_guest;
2537 	}
2538 	/*
2539 	 * Make sure setting of bit in need_tlb_flush precedes
2540 	 * testing of cpu_in_guest bits.  The matching barrier on
2541 	 * the other side is the first smp_mb() in kvmppc_run_core().
2542 	 */
2543 	smp_mb();
2544 	for (i = 0; i < threads_per_core; ++i)
2545 		if (cpumask_test_cpu(cpu + i, cpu_in_guest))
2546 			smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2547 }
2548 
2549 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2550 {
2551 	struct kvm_nested_guest *nested = vcpu->arch.nested;
2552 	struct kvm *kvm = vcpu->kvm;
2553 	int prev_cpu;
2554 
2555 	if (!cpu_has_feature(CPU_FTR_HVMODE))
2556 		return;
2557 
2558 	if (nested)
2559 		prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
2560 	else
2561 		prev_cpu = vcpu->arch.prev_cpu;
2562 
2563 	/*
2564 	 * With radix, the guest can do TLB invalidations itself,
2565 	 * and it could choose to use the local form (tlbiel) if
2566 	 * it is invalidating a translation that has only ever been
2567 	 * used on one vcpu.  However, that doesn't mean it has
2568 	 * only ever been used on one physical cpu, since vcpus
2569 	 * can move around between pcpus.  To cope with this, when
2570 	 * a vcpu moves from one pcpu to another, we need to tell
2571 	 * any vcpus running on the same core as this vcpu previously
2572 	 * ran to flush the TLB.  The TLB is shared between threads,
2573 	 * so we use a single bit in .need_tlb_flush for all 4 threads.
2574 	 */
2575 	if (prev_cpu != pcpu) {
2576 		if (prev_cpu >= 0 &&
2577 		    cpu_first_thread_sibling(prev_cpu) !=
2578 		    cpu_first_thread_sibling(pcpu))
2579 			radix_flush_cpu(kvm, prev_cpu, vcpu);
2580 		if (nested)
2581 			nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
2582 		else
2583 			vcpu->arch.prev_cpu = pcpu;
2584 	}
2585 }
2586 
2587 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2588 {
2589 	int cpu;
2590 	struct paca_struct *tpaca;
2591 	struct kvm *kvm = vc->kvm;
2592 
2593 	cpu = vc->pcpu;
2594 	if (vcpu) {
2595 		if (vcpu->arch.timer_running) {
2596 			hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2597 			vcpu->arch.timer_running = 0;
2598 		}
2599 		cpu += vcpu->arch.ptid;
2600 		vcpu->cpu = vc->pcpu;
2601 		vcpu->arch.thread_cpu = cpu;
2602 		cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2603 	}
2604 	tpaca = paca_ptrs[cpu];
2605 	tpaca->kvm_hstate.kvm_vcpu = vcpu;
2606 	tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2607 	tpaca->kvm_hstate.fake_suspend = 0;
2608 	/* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2609 	smp_wmb();
2610 	tpaca->kvm_hstate.kvm_vcore = vc;
2611 	if (cpu != smp_processor_id())
2612 		kvmppc_ipi_thread(cpu);
2613 }
2614 
2615 static void kvmppc_wait_for_nap(int n_threads)
2616 {
2617 	int cpu = smp_processor_id();
2618 	int i, loops;
2619 
2620 	if (n_threads <= 1)
2621 		return;
2622 	for (loops = 0; loops < 1000000; ++loops) {
2623 		/*
2624 		 * Check if all threads are finished.
2625 		 * We set the vcore pointer when starting a thread
2626 		 * and the thread clears it when finished, so we look
2627 		 * for any threads that still have a non-NULL vcore ptr.
2628 		 */
2629 		for (i = 1; i < n_threads; ++i)
2630 			if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2631 				break;
2632 		if (i == n_threads) {
2633 			HMT_medium();
2634 			return;
2635 		}
2636 		HMT_low();
2637 	}
2638 	HMT_medium();
2639 	for (i = 1; i < n_threads; ++i)
2640 		if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2641 			pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2642 }
2643 
2644 /*
2645  * Check that we are on thread 0 and that any other threads in
2646  * this core are off-line.  Then grab the threads so they can't
2647  * enter the kernel.
2648  */
2649 static int on_primary_thread(void)
2650 {
2651 	int cpu = smp_processor_id();
2652 	int thr;
2653 
2654 	/* Are we on a primary subcore? */
2655 	if (cpu_thread_in_subcore(cpu))
2656 		return 0;
2657 
2658 	thr = 0;
2659 	while (++thr < threads_per_subcore)
2660 		if (cpu_online(cpu + thr))
2661 			return 0;
2662 
2663 	/* Grab all hw threads so they can't go into the kernel */
2664 	for (thr = 1; thr < threads_per_subcore; ++thr) {
2665 		if (kvmppc_grab_hwthread(cpu + thr)) {
2666 			/* Couldn't grab one; let the others go */
2667 			do {
2668 				kvmppc_release_hwthread(cpu + thr);
2669 			} while (--thr > 0);
2670 			return 0;
2671 		}
2672 	}
2673 	return 1;
2674 }
2675 
2676 /*
2677  * A list of virtual cores for each physical CPU.
2678  * These are vcores that could run but their runner VCPU tasks are
2679  * (or may be) preempted.
2680  */
2681 struct preempted_vcore_list {
2682 	struct list_head	list;
2683 	spinlock_t		lock;
2684 };
2685 
2686 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2687 
2688 static void init_vcore_lists(void)
2689 {
2690 	int cpu;
2691 
2692 	for_each_possible_cpu(cpu) {
2693 		struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2694 		spin_lock_init(&lp->lock);
2695 		INIT_LIST_HEAD(&lp->list);
2696 	}
2697 }
2698 
2699 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2700 {
2701 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2702 
2703 	vc->vcore_state = VCORE_PREEMPT;
2704 	vc->pcpu = smp_processor_id();
2705 	if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2706 		spin_lock(&lp->lock);
2707 		list_add_tail(&vc->preempt_list, &lp->list);
2708 		spin_unlock(&lp->lock);
2709 	}
2710 
2711 	/* Start accumulating stolen time */
2712 	kvmppc_core_start_stolen(vc);
2713 }
2714 
2715 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2716 {
2717 	struct preempted_vcore_list *lp;
2718 
2719 	kvmppc_core_end_stolen(vc);
2720 	if (!list_empty(&vc->preempt_list)) {
2721 		lp = &per_cpu(preempted_vcores, vc->pcpu);
2722 		spin_lock(&lp->lock);
2723 		list_del_init(&vc->preempt_list);
2724 		spin_unlock(&lp->lock);
2725 	}
2726 	vc->vcore_state = VCORE_INACTIVE;
2727 }
2728 
2729 /*
2730  * This stores information about the virtual cores currently
2731  * assigned to a physical core.
2732  */
2733 struct core_info {
2734 	int		n_subcores;
2735 	int		max_subcore_threads;
2736 	int		total_threads;
2737 	int		subcore_threads[MAX_SUBCORES];
2738 	struct kvmppc_vcore *vc[MAX_SUBCORES];
2739 };
2740 
2741 /*
2742  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2743  * respectively in 2-way micro-threading (split-core) mode on POWER8.
2744  */
2745 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2746 
2747 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2748 {
2749 	memset(cip, 0, sizeof(*cip));
2750 	cip->n_subcores = 1;
2751 	cip->max_subcore_threads = vc->num_threads;
2752 	cip->total_threads = vc->num_threads;
2753 	cip->subcore_threads[0] = vc->num_threads;
2754 	cip->vc[0] = vc;
2755 }
2756 
2757 static bool subcore_config_ok(int n_subcores, int n_threads)
2758 {
2759 	/*
2760 	 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
2761 	 * split-core mode, with one thread per subcore.
2762 	 */
2763 	if (cpu_has_feature(CPU_FTR_ARCH_300))
2764 		return n_subcores <= 4 && n_threads == 1;
2765 
2766 	/* On POWER8, can only dynamically split if unsplit to begin with */
2767 	if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2768 		return false;
2769 	if (n_subcores > MAX_SUBCORES)
2770 		return false;
2771 	if (n_subcores > 1) {
2772 		if (!(dynamic_mt_modes & 2))
2773 			n_subcores = 4;
2774 		if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2775 			return false;
2776 	}
2777 
2778 	return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2779 }
2780 
2781 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2782 {
2783 	vc->entry_exit_map = 0;
2784 	vc->in_guest = 0;
2785 	vc->napping_threads = 0;
2786 	vc->conferring_threads = 0;
2787 	vc->tb_offset_applied = 0;
2788 }
2789 
2790 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2791 {
2792 	int n_threads = vc->num_threads;
2793 	int sub;
2794 
2795 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2796 		return false;
2797 
2798 	/* In one_vm_per_core mode, require all vcores to be from the same vm */
2799 	if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
2800 		return false;
2801 
2802 	/* Some POWER9 chips require all threads to be in the same MMU mode */
2803 	if (no_mixing_hpt_and_radix &&
2804 	    kvm_is_radix(vc->kvm) != kvm_is_radix(cip->vc[0]->kvm))
2805 		return false;
2806 
2807 	if (n_threads < cip->max_subcore_threads)
2808 		n_threads = cip->max_subcore_threads;
2809 	if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2810 		return false;
2811 	cip->max_subcore_threads = n_threads;
2812 
2813 	sub = cip->n_subcores;
2814 	++cip->n_subcores;
2815 	cip->total_threads += vc->num_threads;
2816 	cip->subcore_threads[sub] = vc->num_threads;
2817 	cip->vc[sub] = vc;
2818 	init_vcore_to_run(vc);
2819 	list_del_init(&vc->preempt_list);
2820 
2821 	return true;
2822 }
2823 
2824 /*
2825  * Work out whether it is possible to piggyback the execution of
2826  * vcore *pvc onto the execution of the other vcores described in *cip.
2827  */
2828 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2829 			  int target_threads)
2830 {
2831 	if (cip->total_threads + pvc->num_threads > target_threads)
2832 		return false;
2833 
2834 	return can_dynamic_split(pvc, cip);
2835 }
2836 
2837 static void prepare_threads(struct kvmppc_vcore *vc)
2838 {
2839 	int i;
2840 	struct kvm_vcpu *vcpu;
2841 
2842 	for_each_runnable_thread(i, vcpu, vc) {
2843 		if (signal_pending(vcpu->arch.run_task))
2844 			vcpu->arch.ret = -EINTR;
2845 		else if (vcpu->arch.vpa.update_pending ||
2846 			 vcpu->arch.slb_shadow.update_pending ||
2847 			 vcpu->arch.dtl.update_pending)
2848 			vcpu->arch.ret = RESUME_GUEST;
2849 		else
2850 			continue;
2851 		kvmppc_remove_runnable(vc, vcpu);
2852 		wake_up(&vcpu->arch.cpu_run);
2853 	}
2854 }
2855 
2856 static void collect_piggybacks(struct core_info *cip, int target_threads)
2857 {
2858 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2859 	struct kvmppc_vcore *pvc, *vcnext;
2860 
2861 	spin_lock(&lp->lock);
2862 	list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2863 		if (!spin_trylock(&pvc->lock))
2864 			continue;
2865 		prepare_threads(pvc);
2866 		if (!pvc->n_runnable) {
2867 			list_del_init(&pvc->preempt_list);
2868 			if (pvc->runner == NULL) {
2869 				pvc->vcore_state = VCORE_INACTIVE;
2870 				kvmppc_core_end_stolen(pvc);
2871 			}
2872 			spin_unlock(&pvc->lock);
2873 			continue;
2874 		}
2875 		if (!can_piggyback(pvc, cip, target_threads)) {
2876 			spin_unlock(&pvc->lock);
2877 			continue;
2878 		}
2879 		kvmppc_core_end_stolen(pvc);
2880 		pvc->vcore_state = VCORE_PIGGYBACK;
2881 		if (cip->total_threads >= target_threads)
2882 			break;
2883 	}
2884 	spin_unlock(&lp->lock);
2885 }
2886 
2887 static bool recheck_signals(struct core_info *cip)
2888 {
2889 	int sub, i;
2890 	struct kvm_vcpu *vcpu;
2891 
2892 	for (sub = 0; sub < cip->n_subcores; ++sub)
2893 		for_each_runnable_thread(i, vcpu, cip->vc[sub])
2894 			if (signal_pending(vcpu->arch.run_task))
2895 				return true;
2896 	return false;
2897 }
2898 
2899 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2900 {
2901 	int still_running = 0, i;
2902 	u64 now;
2903 	long ret;
2904 	struct kvm_vcpu *vcpu;
2905 
2906 	spin_lock(&vc->lock);
2907 	now = get_tb();
2908 	for_each_runnable_thread(i, vcpu, vc) {
2909 		/*
2910 		 * It's safe to unlock the vcore in the loop here, because
2911 		 * for_each_runnable_thread() is safe against removal of
2912 		 * the vcpu, and the vcore state is VCORE_EXITING here,
2913 		 * so any vcpus becoming runnable will have their arch.trap
2914 		 * set to zero and can't actually run in the guest.
2915 		 */
2916 		spin_unlock(&vc->lock);
2917 		/* cancel pending dec exception if dec is positive */
2918 		if (now < vcpu->arch.dec_expires &&
2919 		    kvmppc_core_pending_dec(vcpu))
2920 			kvmppc_core_dequeue_dec(vcpu);
2921 
2922 		trace_kvm_guest_exit(vcpu);
2923 
2924 		ret = RESUME_GUEST;
2925 		if (vcpu->arch.trap)
2926 			ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2927 						    vcpu->arch.run_task);
2928 
2929 		vcpu->arch.ret = ret;
2930 		vcpu->arch.trap = 0;
2931 
2932 		spin_lock(&vc->lock);
2933 		if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2934 			if (vcpu->arch.pending_exceptions)
2935 				kvmppc_core_prepare_to_enter(vcpu);
2936 			if (vcpu->arch.ceded)
2937 				kvmppc_set_timer(vcpu);
2938 			else
2939 				++still_running;
2940 		} else {
2941 			kvmppc_remove_runnable(vc, vcpu);
2942 			wake_up(&vcpu->arch.cpu_run);
2943 		}
2944 	}
2945 	if (!is_master) {
2946 		if (still_running > 0) {
2947 			kvmppc_vcore_preempt(vc);
2948 		} else if (vc->runner) {
2949 			vc->vcore_state = VCORE_PREEMPT;
2950 			kvmppc_core_start_stolen(vc);
2951 		} else {
2952 			vc->vcore_state = VCORE_INACTIVE;
2953 		}
2954 		if (vc->n_runnable > 0 && vc->runner == NULL) {
2955 			/* make sure there's a candidate runner awake */
2956 			i = -1;
2957 			vcpu = next_runnable_thread(vc, &i);
2958 			wake_up(&vcpu->arch.cpu_run);
2959 		}
2960 	}
2961 	spin_unlock(&vc->lock);
2962 }
2963 
2964 /*
2965  * Clear core from the list of active host cores as we are about to
2966  * enter the guest. Only do this if it is the primary thread of the
2967  * core (not if a subcore) that is entering the guest.
2968  */
2969 static inline int kvmppc_clear_host_core(unsigned int cpu)
2970 {
2971 	int core;
2972 
2973 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2974 		return 0;
2975 	/*
2976 	 * Memory barrier can be omitted here as we will do a smp_wmb()
2977 	 * later in kvmppc_start_thread and we need ensure that state is
2978 	 * visible to other CPUs only after we enter guest.
2979 	 */
2980 	core = cpu >> threads_shift;
2981 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2982 	return 0;
2983 }
2984 
2985 /*
2986  * Advertise this core as an active host core since we exited the guest
2987  * Only need to do this if it is the primary thread of the core that is
2988  * exiting.
2989  */
2990 static inline int kvmppc_set_host_core(unsigned int cpu)
2991 {
2992 	int core;
2993 
2994 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2995 		return 0;
2996 
2997 	/*
2998 	 * Memory barrier can be omitted here because we do a spin_unlock
2999 	 * immediately after this which provides the memory barrier.
3000 	 */
3001 	core = cpu >> threads_shift;
3002 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
3003 	return 0;
3004 }
3005 
3006 static void set_irq_happened(int trap)
3007 {
3008 	switch (trap) {
3009 	case BOOK3S_INTERRUPT_EXTERNAL:
3010 		local_paca->irq_happened |= PACA_IRQ_EE;
3011 		break;
3012 	case BOOK3S_INTERRUPT_H_DOORBELL:
3013 		local_paca->irq_happened |= PACA_IRQ_DBELL;
3014 		break;
3015 	case BOOK3S_INTERRUPT_HMI:
3016 		local_paca->irq_happened |= PACA_IRQ_HMI;
3017 		break;
3018 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
3019 		replay_system_reset();
3020 		break;
3021 	}
3022 }
3023 
3024 /*
3025  * Run a set of guest threads on a physical core.
3026  * Called with vc->lock held.
3027  */
3028 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
3029 {
3030 	struct kvm_vcpu *vcpu;
3031 	int i;
3032 	int srcu_idx;
3033 	struct core_info core_info;
3034 	struct kvmppc_vcore *pvc;
3035 	struct kvm_split_mode split_info, *sip;
3036 	int split, subcore_size, active;
3037 	int sub;
3038 	bool thr0_done;
3039 	unsigned long cmd_bit, stat_bit;
3040 	int pcpu, thr;
3041 	int target_threads;
3042 	int controlled_threads;
3043 	int trap;
3044 	bool is_power8;
3045 	bool hpt_on_radix;
3046 
3047 	/*
3048 	 * Remove from the list any threads that have a signal pending
3049 	 * or need a VPA update done
3050 	 */
3051 	prepare_threads(vc);
3052 
3053 	/* if the runner is no longer runnable, let the caller pick a new one */
3054 	if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3055 		return;
3056 
3057 	/*
3058 	 * Initialize *vc.
3059 	 */
3060 	init_vcore_to_run(vc);
3061 	vc->preempt_tb = TB_NIL;
3062 
3063 	/*
3064 	 * Number of threads that we will be controlling: the same as
3065 	 * the number of threads per subcore, except on POWER9,
3066 	 * where it's 1 because the threads are (mostly) independent.
3067 	 */
3068 	controlled_threads = threads_per_vcore(vc->kvm);
3069 
3070 	/*
3071 	 * Make sure we are running on primary threads, and that secondary
3072 	 * threads are offline.  Also check if the number of threads in this
3073 	 * guest are greater than the current system threads per guest.
3074 	 * On POWER9, we need to be not in independent-threads mode if
3075 	 * this is a HPT guest on a radix host machine where the
3076 	 * CPU threads may not be in different MMU modes.
3077 	 */
3078 	hpt_on_radix = no_mixing_hpt_and_radix && radix_enabled() &&
3079 		!kvm_is_radix(vc->kvm);
3080 	if (((controlled_threads > 1) &&
3081 	     ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) ||
3082 	    (hpt_on_radix && vc->kvm->arch.threads_indep)) {
3083 		for_each_runnable_thread(i, vcpu, vc) {
3084 			vcpu->arch.ret = -EBUSY;
3085 			kvmppc_remove_runnable(vc, vcpu);
3086 			wake_up(&vcpu->arch.cpu_run);
3087 		}
3088 		goto out;
3089 	}
3090 
3091 	/*
3092 	 * See if we could run any other vcores on the physical core
3093 	 * along with this one.
3094 	 */
3095 	init_core_info(&core_info, vc);
3096 	pcpu = smp_processor_id();
3097 	target_threads = controlled_threads;
3098 	if (target_smt_mode && target_smt_mode < target_threads)
3099 		target_threads = target_smt_mode;
3100 	if (vc->num_threads < target_threads)
3101 		collect_piggybacks(&core_info, target_threads);
3102 
3103 	/*
3104 	 * On radix, arrange for TLB flushing if necessary.
3105 	 * This has to be done before disabling interrupts since
3106 	 * it uses smp_call_function().
3107 	 */
3108 	pcpu = smp_processor_id();
3109 	if (kvm_is_radix(vc->kvm)) {
3110 		for (sub = 0; sub < core_info.n_subcores; ++sub)
3111 			for_each_runnable_thread(i, vcpu, core_info.vc[sub])
3112 				kvmppc_prepare_radix_vcpu(vcpu, pcpu);
3113 	}
3114 
3115 	/*
3116 	 * Hard-disable interrupts, and check resched flag and signals.
3117 	 * If we need to reschedule or deliver a signal, clean up
3118 	 * and return without going into the guest(s).
3119 	 * If the mmu_ready flag has been cleared, don't go into the
3120 	 * guest because that means a HPT resize operation is in progress.
3121 	 */
3122 	local_irq_disable();
3123 	hard_irq_disable();
3124 	if (lazy_irq_pending() || need_resched() ||
3125 	    recheck_signals(&core_info) || !vc->kvm->arch.mmu_ready) {
3126 		local_irq_enable();
3127 		vc->vcore_state = VCORE_INACTIVE;
3128 		/* Unlock all except the primary vcore */
3129 		for (sub = 1; sub < core_info.n_subcores; ++sub) {
3130 			pvc = core_info.vc[sub];
3131 			/* Put back on to the preempted vcores list */
3132 			kvmppc_vcore_preempt(pvc);
3133 			spin_unlock(&pvc->lock);
3134 		}
3135 		for (i = 0; i < controlled_threads; ++i)
3136 			kvmppc_release_hwthread(pcpu + i);
3137 		return;
3138 	}
3139 
3140 	kvmppc_clear_host_core(pcpu);
3141 
3142 	/* Decide on micro-threading (split-core) mode */
3143 	subcore_size = threads_per_subcore;
3144 	cmd_bit = stat_bit = 0;
3145 	split = core_info.n_subcores;
3146 	sip = NULL;
3147 	is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
3148 		&& !cpu_has_feature(CPU_FTR_ARCH_300);
3149 
3150 	if (split > 1 || hpt_on_radix) {
3151 		sip = &split_info;
3152 		memset(&split_info, 0, sizeof(split_info));
3153 		for (sub = 0; sub < core_info.n_subcores; ++sub)
3154 			split_info.vc[sub] = core_info.vc[sub];
3155 
3156 		if (is_power8) {
3157 			if (split == 2 && (dynamic_mt_modes & 2)) {
3158 				cmd_bit = HID0_POWER8_1TO2LPAR;
3159 				stat_bit = HID0_POWER8_2LPARMODE;
3160 			} else {
3161 				split = 4;
3162 				cmd_bit = HID0_POWER8_1TO4LPAR;
3163 				stat_bit = HID0_POWER8_4LPARMODE;
3164 			}
3165 			subcore_size = MAX_SMT_THREADS / split;
3166 			split_info.rpr = mfspr(SPRN_RPR);
3167 			split_info.pmmar = mfspr(SPRN_PMMAR);
3168 			split_info.ldbar = mfspr(SPRN_LDBAR);
3169 			split_info.subcore_size = subcore_size;
3170 		} else {
3171 			split_info.subcore_size = 1;
3172 			if (hpt_on_radix) {
3173 				/* Use the split_info for LPCR/LPIDR changes */
3174 				split_info.lpcr_req = vc->lpcr;
3175 				split_info.lpidr_req = vc->kvm->arch.lpid;
3176 				split_info.host_lpcr = vc->kvm->arch.host_lpcr;
3177 				split_info.do_set = 1;
3178 			}
3179 		}
3180 
3181 		/* order writes to split_info before kvm_split_mode pointer */
3182 		smp_wmb();
3183 	}
3184 
3185 	for (thr = 0; thr < controlled_threads; ++thr) {
3186 		struct paca_struct *paca = paca_ptrs[pcpu + thr];
3187 
3188 		paca->kvm_hstate.tid = thr;
3189 		paca->kvm_hstate.napping = 0;
3190 		paca->kvm_hstate.kvm_split_mode = sip;
3191 	}
3192 
3193 	/* Initiate micro-threading (split-core) on POWER8 if required */
3194 	if (cmd_bit) {
3195 		unsigned long hid0 = mfspr(SPRN_HID0);
3196 
3197 		hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3198 		mb();
3199 		mtspr(SPRN_HID0, hid0);
3200 		isync();
3201 		for (;;) {
3202 			hid0 = mfspr(SPRN_HID0);
3203 			if (hid0 & stat_bit)
3204 				break;
3205 			cpu_relax();
3206 		}
3207 	}
3208 
3209 	/*
3210 	 * On POWER8, set RWMR register.
3211 	 * Since it only affects PURR and SPURR, it doesn't affect
3212 	 * the host, so we don't save/restore the host value.
3213 	 */
3214 	if (is_power8) {
3215 		unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3216 		int n_online = atomic_read(&vc->online_count);
3217 
3218 		/*
3219 		 * Use the 8-thread value if we're doing split-core
3220 		 * or if the vcore's online count looks bogus.
3221 		 */
3222 		if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3223 		    n_online >= 1 && n_online <= MAX_SMT_THREADS)
3224 			rwmr_val = p8_rwmr_values[n_online];
3225 		mtspr(SPRN_RWMR, rwmr_val);
3226 	}
3227 
3228 	/* Start all the threads */
3229 	active = 0;
3230 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
3231 		thr = is_power8 ? subcore_thread_map[sub] : sub;
3232 		thr0_done = false;
3233 		active |= 1 << thr;
3234 		pvc = core_info.vc[sub];
3235 		pvc->pcpu = pcpu + thr;
3236 		for_each_runnable_thread(i, vcpu, pvc) {
3237 			kvmppc_start_thread(vcpu, pvc);
3238 			kvmppc_create_dtl_entry(vcpu, pvc);
3239 			trace_kvm_guest_enter(vcpu);
3240 			if (!vcpu->arch.ptid)
3241 				thr0_done = true;
3242 			active |= 1 << (thr + vcpu->arch.ptid);
3243 		}
3244 		/*
3245 		 * We need to start the first thread of each subcore
3246 		 * even if it doesn't have a vcpu.
3247 		 */
3248 		if (!thr0_done)
3249 			kvmppc_start_thread(NULL, pvc);
3250 	}
3251 
3252 	/*
3253 	 * Ensure that split_info.do_nap is set after setting
3254 	 * the vcore pointer in the PACA of the secondaries.
3255 	 */
3256 	smp_mb();
3257 
3258 	/*
3259 	 * When doing micro-threading, poke the inactive threads as well.
3260 	 * This gets them to the nap instruction after kvm_do_nap,
3261 	 * which reduces the time taken to unsplit later.
3262 	 * For POWER9 HPT guest on radix host, we need all the secondary
3263 	 * threads woken up so they can do the LPCR/LPIDR change.
3264 	 */
3265 	if (cmd_bit || hpt_on_radix) {
3266 		split_info.do_nap = 1;	/* ask secondaries to nap when done */
3267 		for (thr = 1; thr < threads_per_subcore; ++thr)
3268 			if (!(active & (1 << thr)))
3269 				kvmppc_ipi_thread(pcpu + thr);
3270 	}
3271 
3272 	vc->vcore_state = VCORE_RUNNING;
3273 	preempt_disable();
3274 
3275 	trace_kvmppc_run_core(vc, 0);
3276 
3277 	for (sub = 0; sub < core_info.n_subcores; ++sub)
3278 		spin_unlock(&core_info.vc[sub]->lock);
3279 
3280 	guest_enter_irqoff();
3281 
3282 	srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3283 
3284 	this_cpu_disable_ftrace();
3285 
3286 	/*
3287 	 * Interrupts will be enabled once we get into the guest,
3288 	 * so tell lockdep that we're about to enable interrupts.
3289 	 */
3290 	trace_hardirqs_on();
3291 
3292 	trap = __kvmppc_vcore_entry();
3293 
3294 	trace_hardirqs_off();
3295 
3296 	this_cpu_enable_ftrace();
3297 
3298 	srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3299 
3300 	set_irq_happened(trap);
3301 
3302 	spin_lock(&vc->lock);
3303 	/* prevent other vcpu threads from doing kvmppc_start_thread() now */
3304 	vc->vcore_state = VCORE_EXITING;
3305 
3306 	/* wait for secondary threads to finish writing their state to memory */
3307 	kvmppc_wait_for_nap(controlled_threads);
3308 
3309 	/* Return to whole-core mode if we split the core earlier */
3310 	if (cmd_bit) {
3311 		unsigned long hid0 = mfspr(SPRN_HID0);
3312 		unsigned long loops = 0;
3313 
3314 		hid0 &= ~HID0_POWER8_DYNLPARDIS;
3315 		stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3316 		mb();
3317 		mtspr(SPRN_HID0, hid0);
3318 		isync();
3319 		for (;;) {
3320 			hid0 = mfspr(SPRN_HID0);
3321 			if (!(hid0 & stat_bit))
3322 				break;
3323 			cpu_relax();
3324 			++loops;
3325 		}
3326 	} else if (hpt_on_radix) {
3327 		/* Wait for all threads to have seen final sync */
3328 		for (thr = 1; thr < controlled_threads; ++thr) {
3329 			struct paca_struct *paca = paca_ptrs[pcpu + thr];
3330 
3331 			while (paca->kvm_hstate.kvm_split_mode) {
3332 				HMT_low();
3333 				barrier();
3334 			}
3335 			HMT_medium();
3336 		}
3337 	}
3338 	split_info.do_nap = 0;
3339 
3340 	kvmppc_set_host_core(pcpu);
3341 
3342 	local_irq_enable();
3343 	guest_exit();
3344 
3345 	/* Let secondaries go back to the offline loop */
3346 	for (i = 0; i < controlled_threads; ++i) {
3347 		kvmppc_release_hwthread(pcpu + i);
3348 		if (sip && sip->napped[i])
3349 			kvmppc_ipi_thread(pcpu + i);
3350 		cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
3351 	}
3352 
3353 	spin_unlock(&vc->lock);
3354 
3355 	/* make sure updates to secondary vcpu structs are visible now */
3356 	smp_mb();
3357 
3358 	preempt_enable();
3359 
3360 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
3361 		pvc = core_info.vc[sub];
3362 		post_guest_process(pvc, pvc == vc);
3363 	}
3364 
3365 	spin_lock(&vc->lock);
3366 
3367  out:
3368 	vc->vcore_state = VCORE_INACTIVE;
3369 	trace_kvmppc_run_core(vc, 1);
3370 }
3371 
3372 /*
3373  * Load up hypervisor-mode registers on P9.
3374  */
3375 static int kvmhv_load_hv_regs_and_go(struct kvm_vcpu *vcpu, u64 time_limit,
3376 				     unsigned long lpcr)
3377 {
3378 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
3379 	s64 hdec;
3380 	u64 tb, purr, spurr;
3381 	int trap;
3382 	unsigned long host_hfscr = mfspr(SPRN_HFSCR);
3383 	unsigned long host_ciabr = mfspr(SPRN_CIABR);
3384 	unsigned long host_dawr = mfspr(SPRN_DAWR);
3385 	unsigned long host_dawrx = mfspr(SPRN_DAWRX);
3386 	unsigned long host_psscr = mfspr(SPRN_PSSCR);
3387 	unsigned long host_pidr = mfspr(SPRN_PID);
3388 
3389 	hdec = time_limit - mftb();
3390 	if (hdec < 0)
3391 		return BOOK3S_INTERRUPT_HV_DECREMENTER;
3392 	mtspr(SPRN_HDEC, hdec);
3393 
3394 	if (vc->tb_offset) {
3395 		u64 new_tb = mftb() + vc->tb_offset;
3396 		mtspr(SPRN_TBU40, new_tb);
3397 		tb = mftb();
3398 		if ((tb & 0xffffff) < (new_tb & 0xffffff))
3399 			mtspr(SPRN_TBU40, new_tb + 0x1000000);
3400 		vc->tb_offset_applied = vc->tb_offset;
3401 	}
3402 
3403 	if (vc->pcr)
3404 		mtspr(SPRN_PCR, vc->pcr);
3405 	mtspr(SPRN_DPDES, vc->dpdes);
3406 	mtspr(SPRN_VTB, vc->vtb);
3407 
3408 	local_paca->kvm_hstate.host_purr = mfspr(SPRN_PURR);
3409 	local_paca->kvm_hstate.host_spurr = mfspr(SPRN_SPURR);
3410 	mtspr(SPRN_PURR, vcpu->arch.purr);
3411 	mtspr(SPRN_SPURR, vcpu->arch.spurr);
3412 
3413 	if (dawr_enabled()) {
3414 		mtspr(SPRN_DAWR, vcpu->arch.dawr);
3415 		mtspr(SPRN_DAWRX, vcpu->arch.dawrx);
3416 	}
3417 	mtspr(SPRN_CIABR, vcpu->arch.ciabr);
3418 	mtspr(SPRN_IC, vcpu->arch.ic);
3419 	mtspr(SPRN_PID, vcpu->arch.pid);
3420 
3421 	mtspr(SPRN_PSSCR, vcpu->arch.psscr | PSSCR_EC |
3422 	      (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3423 
3424 	mtspr(SPRN_HFSCR, vcpu->arch.hfscr);
3425 
3426 	mtspr(SPRN_SPRG0, vcpu->arch.shregs.sprg0);
3427 	mtspr(SPRN_SPRG1, vcpu->arch.shregs.sprg1);
3428 	mtspr(SPRN_SPRG2, vcpu->arch.shregs.sprg2);
3429 	mtspr(SPRN_SPRG3, vcpu->arch.shregs.sprg3);
3430 
3431 	mtspr(SPRN_AMOR, ~0UL);
3432 
3433 	mtspr(SPRN_LPCR, lpcr);
3434 	isync();
3435 
3436 	kvmppc_xive_push_vcpu(vcpu);
3437 
3438 	mtspr(SPRN_SRR0, vcpu->arch.shregs.srr0);
3439 	mtspr(SPRN_SRR1, vcpu->arch.shregs.srr1);
3440 
3441 	trap = __kvmhv_vcpu_entry_p9(vcpu);
3442 
3443 	/* Advance host PURR/SPURR by the amount used by guest */
3444 	purr = mfspr(SPRN_PURR);
3445 	spurr = mfspr(SPRN_SPURR);
3446 	mtspr(SPRN_PURR, local_paca->kvm_hstate.host_purr +
3447 	      purr - vcpu->arch.purr);
3448 	mtspr(SPRN_SPURR, local_paca->kvm_hstate.host_spurr +
3449 	      spurr - vcpu->arch.spurr);
3450 	vcpu->arch.purr = purr;
3451 	vcpu->arch.spurr = spurr;
3452 
3453 	vcpu->arch.ic = mfspr(SPRN_IC);
3454 	vcpu->arch.pid = mfspr(SPRN_PID);
3455 	vcpu->arch.psscr = mfspr(SPRN_PSSCR) & PSSCR_GUEST_VIS;
3456 
3457 	vcpu->arch.shregs.sprg0 = mfspr(SPRN_SPRG0);
3458 	vcpu->arch.shregs.sprg1 = mfspr(SPRN_SPRG1);
3459 	vcpu->arch.shregs.sprg2 = mfspr(SPRN_SPRG2);
3460 	vcpu->arch.shregs.sprg3 = mfspr(SPRN_SPRG3);
3461 
3462 	/* Preserve PSSCR[FAKE_SUSPEND] until we've called kvmppc_save_tm_hv */
3463 	mtspr(SPRN_PSSCR, host_psscr |
3464 	      (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3465 	mtspr(SPRN_HFSCR, host_hfscr);
3466 	mtspr(SPRN_CIABR, host_ciabr);
3467 	mtspr(SPRN_DAWR, host_dawr);
3468 	mtspr(SPRN_DAWRX, host_dawrx);
3469 	mtspr(SPRN_PID, host_pidr);
3470 
3471 	/*
3472 	 * Since this is radix, do a eieio; tlbsync; ptesync sequence in
3473 	 * case we interrupted the guest between a tlbie and a ptesync.
3474 	 */
3475 	asm volatile("eieio; tlbsync; ptesync");
3476 
3477 	mtspr(SPRN_LPID, vcpu->kvm->arch.host_lpid);	/* restore host LPID */
3478 	isync();
3479 
3480 	vc->dpdes = mfspr(SPRN_DPDES);
3481 	vc->vtb = mfspr(SPRN_VTB);
3482 	mtspr(SPRN_DPDES, 0);
3483 	if (vc->pcr)
3484 		mtspr(SPRN_PCR, 0);
3485 
3486 	if (vc->tb_offset_applied) {
3487 		u64 new_tb = mftb() - vc->tb_offset_applied;
3488 		mtspr(SPRN_TBU40, new_tb);
3489 		tb = mftb();
3490 		if ((tb & 0xffffff) < (new_tb & 0xffffff))
3491 			mtspr(SPRN_TBU40, new_tb + 0x1000000);
3492 		vc->tb_offset_applied = 0;
3493 	}
3494 
3495 	mtspr(SPRN_HDEC, 0x7fffffff);
3496 	mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr);
3497 
3498 	return trap;
3499 }
3500 
3501 /*
3502  * Virtual-mode guest entry for POWER9 and later when the host and
3503  * guest are both using the radix MMU.  The LPIDR has already been set.
3504  */
3505 int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
3506 			 unsigned long lpcr)
3507 {
3508 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
3509 	unsigned long host_dscr = mfspr(SPRN_DSCR);
3510 	unsigned long host_tidr = mfspr(SPRN_TIDR);
3511 	unsigned long host_iamr = mfspr(SPRN_IAMR);
3512 	unsigned long host_amr = mfspr(SPRN_AMR);
3513 	s64 dec;
3514 	u64 tb;
3515 	int trap, save_pmu;
3516 
3517 	dec = mfspr(SPRN_DEC);
3518 	tb = mftb();
3519 	if (dec < 512)
3520 		return BOOK3S_INTERRUPT_HV_DECREMENTER;
3521 	local_paca->kvm_hstate.dec_expires = dec + tb;
3522 	if (local_paca->kvm_hstate.dec_expires < time_limit)
3523 		time_limit = local_paca->kvm_hstate.dec_expires;
3524 
3525 	vcpu->arch.ceded = 0;
3526 
3527 	kvmhv_save_host_pmu();		/* saves it to PACA kvm_hstate */
3528 
3529 	kvmppc_subcore_enter_guest();
3530 
3531 	vc->entry_exit_map = 1;
3532 	vc->in_guest = 1;
3533 
3534 	if (vcpu->arch.vpa.pinned_addr) {
3535 		struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3536 		u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3537 		lp->yield_count = cpu_to_be32(yield_count);
3538 		vcpu->arch.vpa.dirty = 1;
3539 	}
3540 
3541 	if (cpu_has_feature(CPU_FTR_TM) ||
3542 	    cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3543 		kvmppc_restore_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3544 
3545 	kvmhv_load_guest_pmu(vcpu);
3546 
3547 	msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3548 	load_fp_state(&vcpu->arch.fp);
3549 #ifdef CONFIG_ALTIVEC
3550 	load_vr_state(&vcpu->arch.vr);
3551 #endif
3552 	mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
3553 
3554 	mtspr(SPRN_DSCR, vcpu->arch.dscr);
3555 	mtspr(SPRN_IAMR, vcpu->arch.iamr);
3556 	mtspr(SPRN_PSPB, vcpu->arch.pspb);
3557 	mtspr(SPRN_FSCR, vcpu->arch.fscr);
3558 	mtspr(SPRN_TAR, vcpu->arch.tar);
3559 	mtspr(SPRN_EBBHR, vcpu->arch.ebbhr);
3560 	mtspr(SPRN_EBBRR, vcpu->arch.ebbrr);
3561 	mtspr(SPRN_BESCR, vcpu->arch.bescr);
3562 	mtspr(SPRN_WORT, vcpu->arch.wort);
3563 	mtspr(SPRN_TIDR, vcpu->arch.tid);
3564 	mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
3565 	mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
3566 	mtspr(SPRN_AMR, vcpu->arch.amr);
3567 	mtspr(SPRN_UAMOR, vcpu->arch.uamor);
3568 
3569 	if (!(vcpu->arch.ctrl & 1))
3570 		mtspr(SPRN_CTRLT, mfspr(SPRN_CTRLF) & ~1);
3571 
3572 	mtspr(SPRN_DEC, vcpu->arch.dec_expires - mftb());
3573 
3574 	if (kvmhv_on_pseries()) {
3575 		/* call our hypervisor to load up HV regs and go */
3576 		struct hv_guest_state hvregs;
3577 
3578 		kvmhv_save_hv_regs(vcpu, &hvregs);
3579 		hvregs.lpcr = lpcr;
3580 		vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
3581 		hvregs.version = HV_GUEST_STATE_VERSION;
3582 		if (vcpu->arch.nested) {
3583 			hvregs.lpid = vcpu->arch.nested->shadow_lpid;
3584 			hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
3585 		} else {
3586 			hvregs.lpid = vcpu->kvm->arch.lpid;
3587 			hvregs.vcpu_token = vcpu->vcpu_id;
3588 		}
3589 		hvregs.hdec_expiry = time_limit;
3590 		trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
3591 					  __pa(&vcpu->arch.regs));
3592 		kvmhv_restore_hv_return_state(vcpu, &hvregs);
3593 		vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
3594 		vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
3595 		vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
3596 
3597 		/* H_CEDE has to be handled now, not later */
3598 		if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
3599 		    kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
3600 			kvmppc_nested_cede(vcpu);
3601 			trap = 0;
3602 		}
3603 	} else {
3604 		trap = kvmhv_load_hv_regs_and_go(vcpu, time_limit, lpcr);
3605 	}
3606 
3607 	vcpu->arch.slb_max = 0;
3608 	dec = mfspr(SPRN_DEC);
3609 	tb = mftb();
3610 	vcpu->arch.dec_expires = dec + tb;
3611 	vcpu->cpu = -1;
3612 	vcpu->arch.thread_cpu = -1;
3613 	vcpu->arch.ctrl = mfspr(SPRN_CTRLF);
3614 
3615 	vcpu->arch.iamr = mfspr(SPRN_IAMR);
3616 	vcpu->arch.pspb = mfspr(SPRN_PSPB);
3617 	vcpu->arch.fscr = mfspr(SPRN_FSCR);
3618 	vcpu->arch.tar = mfspr(SPRN_TAR);
3619 	vcpu->arch.ebbhr = mfspr(SPRN_EBBHR);
3620 	vcpu->arch.ebbrr = mfspr(SPRN_EBBRR);
3621 	vcpu->arch.bescr = mfspr(SPRN_BESCR);
3622 	vcpu->arch.wort = mfspr(SPRN_WORT);
3623 	vcpu->arch.tid = mfspr(SPRN_TIDR);
3624 	vcpu->arch.amr = mfspr(SPRN_AMR);
3625 	vcpu->arch.uamor = mfspr(SPRN_UAMOR);
3626 	vcpu->arch.dscr = mfspr(SPRN_DSCR);
3627 
3628 	mtspr(SPRN_PSPB, 0);
3629 	mtspr(SPRN_WORT, 0);
3630 	mtspr(SPRN_UAMOR, 0);
3631 	mtspr(SPRN_DSCR, host_dscr);
3632 	mtspr(SPRN_TIDR, host_tidr);
3633 	mtspr(SPRN_IAMR, host_iamr);
3634 	mtspr(SPRN_PSPB, 0);
3635 
3636 	if (host_amr != vcpu->arch.amr)
3637 		mtspr(SPRN_AMR, host_amr);
3638 
3639 	msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3640 	store_fp_state(&vcpu->arch.fp);
3641 #ifdef CONFIG_ALTIVEC
3642 	store_vr_state(&vcpu->arch.vr);
3643 #endif
3644 	vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
3645 
3646 	if (cpu_has_feature(CPU_FTR_TM) ||
3647 	    cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3648 		kvmppc_save_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3649 
3650 	save_pmu = 1;
3651 	if (vcpu->arch.vpa.pinned_addr) {
3652 		struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3653 		u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3654 		lp->yield_count = cpu_to_be32(yield_count);
3655 		vcpu->arch.vpa.dirty = 1;
3656 		save_pmu = lp->pmcregs_in_use;
3657 	}
3658 
3659 	kvmhv_save_guest_pmu(vcpu, save_pmu);
3660 
3661 	vc->entry_exit_map = 0x101;
3662 	vc->in_guest = 0;
3663 
3664 	mtspr(SPRN_DEC, local_paca->kvm_hstate.dec_expires - mftb());
3665 	mtspr(SPRN_SPRG_VDSO_WRITE, local_paca->sprg_vdso);
3666 
3667 	kvmhv_load_host_pmu();
3668 
3669 	kvmppc_subcore_exit_guest();
3670 
3671 	return trap;
3672 }
3673 
3674 /*
3675  * Wait for some other vcpu thread to execute us, and
3676  * wake us up when we need to handle something in the host.
3677  */
3678 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
3679 				 struct kvm_vcpu *vcpu, int wait_state)
3680 {
3681 	DEFINE_WAIT(wait);
3682 
3683 	prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
3684 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3685 		spin_unlock(&vc->lock);
3686 		schedule();
3687 		spin_lock(&vc->lock);
3688 	}
3689 	finish_wait(&vcpu->arch.cpu_run, &wait);
3690 }
3691 
3692 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
3693 {
3694 	if (!halt_poll_ns_grow)
3695 		return;
3696 
3697 	vc->halt_poll_ns *= halt_poll_ns_grow;
3698 	if (vc->halt_poll_ns < halt_poll_ns_grow_start)
3699 		vc->halt_poll_ns = halt_poll_ns_grow_start;
3700 }
3701 
3702 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
3703 {
3704 	if (halt_poll_ns_shrink == 0)
3705 		vc->halt_poll_ns = 0;
3706 	else
3707 		vc->halt_poll_ns /= halt_poll_ns_shrink;
3708 }
3709 
3710 #ifdef CONFIG_KVM_XICS
3711 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3712 {
3713 	if (!xics_on_xive())
3714 		return false;
3715 	return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
3716 		vcpu->arch.xive_saved_state.cppr;
3717 }
3718 #else
3719 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3720 {
3721 	return false;
3722 }
3723 #endif /* CONFIG_KVM_XICS */
3724 
3725 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3726 {
3727 	if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3728 	    kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3729 		return true;
3730 
3731 	return false;
3732 }
3733 
3734 /*
3735  * Check to see if any of the runnable vcpus on the vcore have pending
3736  * exceptions or are no longer ceded
3737  */
3738 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3739 {
3740 	struct kvm_vcpu *vcpu;
3741 	int i;
3742 
3743 	for_each_runnable_thread(i, vcpu, vc) {
3744 		if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3745 			return 1;
3746 	}
3747 
3748 	return 0;
3749 }
3750 
3751 /*
3752  * All the vcpus in this vcore are idle, so wait for a decrementer
3753  * or external interrupt to one of the vcpus.  vc->lock is held.
3754  */
3755 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3756 {
3757 	ktime_t cur, start_poll, start_wait;
3758 	int do_sleep = 1;
3759 	u64 block_ns;
3760 	DECLARE_SWAITQUEUE(wait);
3761 
3762 	/* Poll for pending exceptions and ceded state */
3763 	cur = start_poll = ktime_get();
3764 	if (vc->halt_poll_ns) {
3765 		ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3766 		++vc->runner->stat.halt_attempted_poll;
3767 
3768 		vc->vcore_state = VCORE_POLLING;
3769 		spin_unlock(&vc->lock);
3770 
3771 		do {
3772 			if (kvmppc_vcore_check_block(vc)) {
3773 				do_sleep = 0;
3774 				break;
3775 			}
3776 			cur = ktime_get();
3777 		} while (single_task_running() && ktime_before(cur, stop));
3778 
3779 		spin_lock(&vc->lock);
3780 		vc->vcore_state = VCORE_INACTIVE;
3781 
3782 		if (!do_sleep) {
3783 			++vc->runner->stat.halt_successful_poll;
3784 			goto out;
3785 		}
3786 	}
3787 
3788 	prepare_to_swait_exclusive(&vc->wq, &wait, TASK_INTERRUPTIBLE);
3789 
3790 	if (kvmppc_vcore_check_block(vc)) {
3791 		finish_swait(&vc->wq, &wait);
3792 		do_sleep = 0;
3793 		/* If we polled, count this as a successful poll */
3794 		if (vc->halt_poll_ns)
3795 			++vc->runner->stat.halt_successful_poll;
3796 		goto out;
3797 	}
3798 
3799 	start_wait = ktime_get();
3800 
3801 	vc->vcore_state = VCORE_SLEEPING;
3802 	trace_kvmppc_vcore_blocked(vc, 0);
3803 	spin_unlock(&vc->lock);
3804 	schedule();
3805 	finish_swait(&vc->wq, &wait);
3806 	spin_lock(&vc->lock);
3807 	vc->vcore_state = VCORE_INACTIVE;
3808 	trace_kvmppc_vcore_blocked(vc, 1);
3809 	++vc->runner->stat.halt_successful_wait;
3810 
3811 	cur = ktime_get();
3812 
3813 out:
3814 	block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3815 
3816 	/* Attribute wait time */
3817 	if (do_sleep) {
3818 		vc->runner->stat.halt_wait_ns +=
3819 			ktime_to_ns(cur) - ktime_to_ns(start_wait);
3820 		/* Attribute failed poll time */
3821 		if (vc->halt_poll_ns)
3822 			vc->runner->stat.halt_poll_fail_ns +=
3823 				ktime_to_ns(start_wait) -
3824 				ktime_to_ns(start_poll);
3825 	} else {
3826 		/* Attribute successful poll time */
3827 		if (vc->halt_poll_ns)
3828 			vc->runner->stat.halt_poll_success_ns +=
3829 				ktime_to_ns(cur) -
3830 				ktime_to_ns(start_poll);
3831 	}
3832 
3833 	/* Adjust poll time */
3834 	if (halt_poll_ns) {
3835 		if (block_ns <= vc->halt_poll_ns)
3836 			;
3837 		/* We slept and blocked for longer than the max halt time */
3838 		else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3839 			shrink_halt_poll_ns(vc);
3840 		/* We slept and our poll time is too small */
3841 		else if (vc->halt_poll_ns < halt_poll_ns &&
3842 				block_ns < halt_poll_ns)
3843 			grow_halt_poll_ns(vc);
3844 		if (vc->halt_poll_ns > halt_poll_ns)
3845 			vc->halt_poll_ns = halt_poll_ns;
3846 	} else
3847 		vc->halt_poll_ns = 0;
3848 
3849 	trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3850 }
3851 
3852 /*
3853  * This never fails for a radix guest, as none of the operations it does
3854  * for a radix guest can fail or have a way to report failure.
3855  * kvmhv_run_single_vcpu() relies on this fact.
3856  */
3857 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
3858 {
3859 	int r = 0;
3860 	struct kvm *kvm = vcpu->kvm;
3861 
3862 	mutex_lock(&kvm->arch.mmu_setup_lock);
3863 	if (!kvm->arch.mmu_ready) {
3864 		if (!kvm_is_radix(kvm))
3865 			r = kvmppc_hv_setup_htab_rma(vcpu);
3866 		if (!r) {
3867 			if (cpu_has_feature(CPU_FTR_ARCH_300))
3868 				kvmppc_setup_partition_table(kvm);
3869 			kvm->arch.mmu_ready = 1;
3870 		}
3871 	}
3872 	mutex_unlock(&kvm->arch.mmu_setup_lock);
3873 	return r;
3874 }
3875 
3876 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3877 {
3878 	int n_ceded, i, r;
3879 	struct kvmppc_vcore *vc;
3880 	struct kvm_vcpu *v;
3881 
3882 	trace_kvmppc_run_vcpu_enter(vcpu);
3883 
3884 	kvm_run->exit_reason = 0;
3885 	vcpu->arch.ret = RESUME_GUEST;
3886 	vcpu->arch.trap = 0;
3887 	kvmppc_update_vpas(vcpu);
3888 
3889 	/*
3890 	 * Synchronize with other threads in this virtual core
3891 	 */
3892 	vc = vcpu->arch.vcore;
3893 	spin_lock(&vc->lock);
3894 	vcpu->arch.ceded = 0;
3895 	vcpu->arch.run_task = current;
3896 	vcpu->arch.kvm_run = kvm_run;
3897 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3898 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3899 	vcpu->arch.busy_preempt = TB_NIL;
3900 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3901 	++vc->n_runnable;
3902 
3903 	/*
3904 	 * This happens the first time this is called for a vcpu.
3905 	 * If the vcore is already running, we may be able to start
3906 	 * this thread straight away and have it join in.
3907 	 */
3908 	if (!signal_pending(current)) {
3909 		if ((vc->vcore_state == VCORE_PIGGYBACK ||
3910 		     vc->vcore_state == VCORE_RUNNING) &&
3911 			   !VCORE_IS_EXITING(vc)) {
3912 			kvmppc_create_dtl_entry(vcpu, vc);
3913 			kvmppc_start_thread(vcpu, vc);
3914 			trace_kvm_guest_enter(vcpu);
3915 		} else if (vc->vcore_state == VCORE_SLEEPING) {
3916 			swake_up_one(&vc->wq);
3917 		}
3918 
3919 	}
3920 
3921 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3922 	       !signal_pending(current)) {
3923 		/* See if the MMU is ready to go */
3924 		if (!vcpu->kvm->arch.mmu_ready) {
3925 			spin_unlock(&vc->lock);
3926 			r = kvmhv_setup_mmu(vcpu);
3927 			spin_lock(&vc->lock);
3928 			if (r) {
3929 				kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3930 				kvm_run->fail_entry.
3931 					hardware_entry_failure_reason = 0;
3932 				vcpu->arch.ret = r;
3933 				break;
3934 			}
3935 		}
3936 
3937 		if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3938 			kvmppc_vcore_end_preempt(vc);
3939 
3940 		if (vc->vcore_state != VCORE_INACTIVE) {
3941 			kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
3942 			continue;
3943 		}
3944 		for_each_runnable_thread(i, v, vc) {
3945 			kvmppc_core_prepare_to_enter(v);
3946 			if (signal_pending(v->arch.run_task)) {
3947 				kvmppc_remove_runnable(vc, v);
3948 				v->stat.signal_exits++;
3949 				v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
3950 				v->arch.ret = -EINTR;
3951 				wake_up(&v->arch.cpu_run);
3952 			}
3953 		}
3954 		if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3955 			break;
3956 		n_ceded = 0;
3957 		for_each_runnable_thread(i, v, vc) {
3958 			if (!kvmppc_vcpu_woken(v))
3959 				n_ceded += v->arch.ceded;
3960 			else
3961 				v->arch.ceded = 0;
3962 		}
3963 		vc->runner = vcpu;
3964 		if (n_ceded == vc->n_runnable) {
3965 			kvmppc_vcore_blocked(vc);
3966 		} else if (need_resched()) {
3967 			kvmppc_vcore_preempt(vc);
3968 			/* Let something else run */
3969 			cond_resched_lock(&vc->lock);
3970 			if (vc->vcore_state == VCORE_PREEMPT)
3971 				kvmppc_vcore_end_preempt(vc);
3972 		} else {
3973 			kvmppc_run_core(vc);
3974 		}
3975 		vc->runner = NULL;
3976 	}
3977 
3978 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3979 	       (vc->vcore_state == VCORE_RUNNING ||
3980 		vc->vcore_state == VCORE_EXITING ||
3981 		vc->vcore_state == VCORE_PIGGYBACK))
3982 		kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
3983 
3984 	if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3985 		kvmppc_vcore_end_preempt(vc);
3986 
3987 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3988 		kvmppc_remove_runnable(vc, vcpu);
3989 		vcpu->stat.signal_exits++;
3990 		kvm_run->exit_reason = KVM_EXIT_INTR;
3991 		vcpu->arch.ret = -EINTR;
3992 	}
3993 
3994 	if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
3995 		/* Wake up some vcpu to run the core */
3996 		i = -1;
3997 		v = next_runnable_thread(vc, &i);
3998 		wake_up(&v->arch.cpu_run);
3999 	}
4000 
4001 	trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
4002 	spin_unlock(&vc->lock);
4003 	return vcpu->arch.ret;
4004 }
4005 
4006 int kvmhv_run_single_vcpu(struct kvm_run *kvm_run,
4007 			  struct kvm_vcpu *vcpu, u64 time_limit,
4008 			  unsigned long lpcr)
4009 {
4010 	int trap, r, pcpu;
4011 	int srcu_idx, lpid;
4012 	struct kvmppc_vcore *vc;
4013 	struct kvm *kvm = vcpu->kvm;
4014 	struct kvm_nested_guest *nested = vcpu->arch.nested;
4015 
4016 	trace_kvmppc_run_vcpu_enter(vcpu);
4017 
4018 	kvm_run->exit_reason = 0;
4019 	vcpu->arch.ret = RESUME_GUEST;
4020 	vcpu->arch.trap = 0;
4021 
4022 	vc = vcpu->arch.vcore;
4023 	vcpu->arch.ceded = 0;
4024 	vcpu->arch.run_task = current;
4025 	vcpu->arch.kvm_run = kvm_run;
4026 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4027 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4028 	vcpu->arch.busy_preempt = TB_NIL;
4029 	vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
4030 	vc->runnable_threads[0] = vcpu;
4031 	vc->n_runnable = 1;
4032 	vc->runner = vcpu;
4033 
4034 	/* See if the MMU is ready to go */
4035 	if (!kvm->arch.mmu_ready)
4036 		kvmhv_setup_mmu(vcpu);
4037 
4038 	if (need_resched())
4039 		cond_resched();
4040 
4041 	kvmppc_update_vpas(vcpu);
4042 
4043 	init_vcore_to_run(vc);
4044 	vc->preempt_tb = TB_NIL;
4045 
4046 	preempt_disable();
4047 	pcpu = smp_processor_id();
4048 	vc->pcpu = pcpu;
4049 	kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4050 
4051 	local_irq_disable();
4052 	hard_irq_disable();
4053 	if (signal_pending(current))
4054 		goto sigpend;
4055 	if (lazy_irq_pending() || need_resched() || !kvm->arch.mmu_ready)
4056 		goto out;
4057 
4058 	if (!nested) {
4059 		kvmppc_core_prepare_to_enter(vcpu);
4060 		if (vcpu->arch.doorbell_request) {
4061 			vc->dpdes = 1;
4062 			smp_wmb();
4063 			vcpu->arch.doorbell_request = 0;
4064 		}
4065 		if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4066 			     &vcpu->arch.pending_exceptions))
4067 			lpcr |= LPCR_MER;
4068 	} else if (vcpu->arch.pending_exceptions ||
4069 		   vcpu->arch.doorbell_request ||
4070 		   xive_interrupt_pending(vcpu)) {
4071 		vcpu->arch.ret = RESUME_HOST;
4072 		goto out;
4073 	}
4074 
4075 	kvmppc_clear_host_core(pcpu);
4076 
4077 	local_paca->kvm_hstate.tid = 0;
4078 	local_paca->kvm_hstate.napping = 0;
4079 	local_paca->kvm_hstate.kvm_split_mode = NULL;
4080 	kvmppc_start_thread(vcpu, vc);
4081 	kvmppc_create_dtl_entry(vcpu, vc);
4082 	trace_kvm_guest_enter(vcpu);
4083 
4084 	vc->vcore_state = VCORE_RUNNING;
4085 	trace_kvmppc_run_core(vc, 0);
4086 
4087 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
4088 		lpid = nested ? nested->shadow_lpid : kvm->arch.lpid;
4089 		mtspr(SPRN_LPID, lpid);
4090 		isync();
4091 		kvmppc_check_need_tlb_flush(kvm, pcpu, nested);
4092 	}
4093 
4094 	guest_enter_irqoff();
4095 
4096 	srcu_idx = srcu_read_lock(&kvm->srcu);
4097 
4098 	this_cpu_disable_ftrace();
4099 
4100 	/* Tell lockdep that we're about to enable interrupts */
4101 	trace_hardirqs_on();
4102 
4103 	trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr);
4104 	vcpu->arch.trap = trap;
4105 
4106 	trace_hardirqs_off();
4107 
4108 	this_cpu_enable_ftrace();
4109 
4110 	srcu_read_unlock(&kvm->srcu, srcu_idx);
4111 
4112 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
4113 		mtspr(SPRN_LPID, kvm->arch.host_lpid);
4114 		isync();
4115 	}
4116 
4117 	set_irq_happened(trap);
4118 
4119 	kvmppc_set_host_core(pcpu);
4120 
4121 	local_irq_enable();
4122 	guest_exit();
4123 
4124 	cpumask_clear_cpu(pcpu, &kvm->arch.cpu_in_guest);
4125 
4126 	preempt_enable();
4127 
4128 	/* cancel pending decrementer exception if DEC is now positive */
4129 	if (get_tb() < vcpu->arch.dec_expires && kvmppc_core_pending_dec(vcpu))
4130 		kvmppc_core_dequeue_dec(vcpu);
4131 
4132 	trace_kvm_guest_exit(vcpu);
4133 	r = RESUME_GUEST;
4134 	if (trap) {
4135 		if (!nested)
4136 			r = kvmppc_handle_exit_hv(kvm_run, vcpu, current);
4137 		else
4138 			r = kvmppc_handle_nested_exit(kvm_run, vcpu);
4139 	}
4140 	vcpu->arch.ret = r;
4141 
4142 	if (is_kvmppc_resume_guest(r) && vcpu->arch.ceded &&
4143 	    !kvmppc_vcpu_woken(vcpu)) {
4144 		kvmppc_set_timer(vcpu);
4145 		while (vcpu->arch.ceded && !kvmppc_vcpu_woken(vcpu)) {
4146 			if (signal_pending(current)) {
4147 				vcpu->stat.signal_exits++;
4148 				kvm_run->exit_reason = KVM_EXIT_INTR;
4149 				vcpu->arch.ret = -EINTR;
4150 				break;
4151 			}
4152 			spin_lock(&vc->lock);
4153 			kvmppc_vcore_blocked(vc);
4154 			spin_unlock(&vc->lock);
4155 		}
4156 	}
4157 	vcpu->arch.ceded = 0;
4158 
4159 	vc->vcore_state = VCORE_INACTIVE;
4160 	trace_kvmppc_run_core(vc, 1);
4161 
4162  done:
4163 	kvmppc_remove_runnable(vc, vcpu);
4164 	trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
4165 
4166 	return vcpu->arch.ret;
4167 
4168  sigpend:
4169 	vcpu->stat.signal_exits++;
4170 	kvm_run->exit_reason = KVM_EXIT_INTR;
4171 	vcpu->arch.ret = -EINTR;
4172  out:
4173 	local_irq_enable();
4174 	preempt_enable();
4175 	goto done;
4176 }
4177 
4178 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
4179 {
4180 	int r;
4181 	int srcu_idx;
4182 	unsigned long ebb_regs[3] = {};	/* shut up GCC */
4183 	unsigned long user_tar = 0;
4184 	unsigned int user_vrsave;
4185 	struct kvm *kvm;
4186 
4187 	if (!vcpu->arch.sane) {
4188 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4189 		return -EINVAL;
4190 	}
4191 
4192 	/*
4193 	 * Don't allow entry with a suspended transaction, because
4194 	 * the guest entry/exit code will lose it.
4195 	 * If the guest has TM enabled, save away their TM-related SPRs
4196 	 * (they will get restored by the TM unavailable interrupt).
4197 	 */
4198 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4199 	if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4200 	    (current->thread.regs->msr & MSR_TM)) {
4201 		if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4202 			run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4203 			run->fail_entry.hardware_entry_failure_reason = 0;
4204 			return -EINVAL;
4205 		}
4206 		/* Enable TM so we can read the TM SPRs */
4207 		mtmsr(mfmsr() | MSR_TM);
4208 		current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
4209 		current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
4210 		current->thread.tm_texasr = mfspr(SPRN_TEXASR);
4211 		current->thread.regs->msr &= ~MSR_TM;
4212 	}
4213 #endif
4214 
4215 	/*
4216 	 * Force online to 1 for the sake of old userspace which doesn't
4217 	 * set it.
4218 	 */
4219 	if (!vcpu->arch.online) {
4220 		atomic_inc(&vcpu->arch.vcore->online_count);
4221 		vcpu->arch.online = 1;
4222 	}
4223 
4224 	kvmppc_core_prepare_to_enter(vcpu);
4225 
4226 	/* No need to go into the guest when all we'll do is come back out */
4227 	if (signal_pending(current)) {
4228 		run->exit_reason = KVM_EXIT_INTR;
4229 		return -EINTR;
4230 	}
4231 
4232 	kvm = vcpu->kvm;
4233 	atomic_inc(&kvm->arch.vcpus_running);
4234 	/* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4235 	smp_mb();
4236 
4237 	flush_all_to_thread(current);
4238 
4239 	/* Save userspace EBB and other register values */
4240 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4241 		ebb_regs[0] = mfspr(SPRN_EBBHR);
4242 		ebb_regs[1] = mfspr(SPRN_EBBRR);
4243 		ebb_regs[2] = mfspr(SPRN_BESCR);
4244 		user_tar = mfspr(SPRN_TAR);
4245 	}
4246 	user_vrsave = mfspr(SPRN_VRSAVE);
4247 
4248 	vcpu->arch.wqp = &vcpu->arch.vcore->wq;
4249 	vcpu->arch.pgdir = current->mm->pgd;
4250 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4251 
4252 	do {
4253 		/*
4254 		 * The early POWER9 chips that can't mix radix and HPT threads
4255 		 * on the same core also need the workaround for the problem
4256 		 * where the TLB would prefetch entries in the guest exit path
4257 		 * for radix guests using the guest PIDR value and LPID 0.
4258 		 * The workaround is in the old path (kvmppc_run_vcpu())
4259 		 * but not the new path (kvmhv_run_single_vcpu()).
4260 		 */
4261 		if (kvm->arch.threads_indep && kvm_is_radix(kvm) &&
4262 		    !no_mixing_hpt_and_radix)
4263 			r = kvmhv_run_single_vcpu(run, vcpu, ~(u64)0,
4264 						  vcpu->arch.vcore->lpcr);
4265 		else
4266 			r = kvmppc_run_vcpu(run, vcpu);
4267 
4268 		if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
4269 		    !(vcpu->arch.shregs.msr & MSR_PR)) {
4270 			trace_kvm_hcall_enter(vcpu);
4271 			r = kvmppc_pseries_do_hcall(vcpu);
4272 			trace_kvm_hcall_exit(vcpu, r);
4273 			kvmppc_core_prepare_to_enter(vcpu);
4274 		} else if (r == RESUME_PAGE_FAULT) {
4275 			srcu_idx = srcu_read_lock(&kvm->srcu);
4276 			r = kvmppc_book3s_hv_page_fault(run, vcpu,
4277 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4278 			srcu_read_unlock(&kvm->srcu, srcu_idx);
4279 		} else if (r == RESUME_PASSTHROUGH) {
4280 			if (WARN_ON(xics_on_xive()))
4281 				r = H_SUCCESS;
4282 			else
4283 				r = kvmppc_xics_rm_complete(vcpu, 0);
4284 		}
4285 	} while (is_kvmppc_resume_guest(r));
4286 
4287 	/* Restore userspace EBB and other register values */
4288 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4289 		mtspr(SPRN_EBBHR, ebb_regs[0]);
4290 		mtspr(SPRN_EBBRR, ebb_regs[1]);
4291 		mtspr(SPRN_BESCR, ebb_regs[2]);
4292 		mtspr(SPRN_TAR, user_tar);
4293 		mtspr(SPRN_FSCR, current->thread.fscr);
4294 	}
4295 	mtspr(SPRN_VRSAVE, user_vrsave);
4296 
4297 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4298 	atomic_dec(&kvm->arch.vcpus_running);
4299 	return r;
4300 }
4301 
4302 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4303 				     int shift, int sllp)
4304 {
4305 	(*sps)->page_shift = shift;
4306 	(*sps)->slb_enc = sllp;
4307 	(*sps)->enc[0].page_shift = shift;
4308 	(*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4309 	/*
4310 	 * Add 16MB MPSS support (may get filtered out by userspace)
4311 	 */
4312 	if (shift != 24) {
4313 		int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4314 		if (penc != -1) {
4315 			(*sps)->enc[1].page_shift = 24;
4316 			(*sps)->enc[1].pte_enc = penc;
4317 		}
4318 	}
4319 	(*sps)++;
4320 }
4321 
4322 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4323 					 struct kvm_ppc_smmu_info *info)
4324 {
4325 	struct kvm_ppc_one_seg_page_size *sps;
4326 
4327 	/*
4328 	 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4329 	 * POWER7 doesn't support keys for instruction accesses,
4330 	 * POWER8 and POWER9 do.
4331 	 */
4332 	info->data_keys = 32;
4333 	info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4334 
4335 	/* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4336 	info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4337 	info->slb_size = 32;
4338 
4339 	/* We only support these sizes for now, and no muti-size segments */
4340 	sps = &info->sps[0];
4341 	kvmppc_add_seg_page_size(&sps, 12, 0);
4342 	kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4343 	kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4344 
4345 	/* If running as a nested hypervisor, we don't support HPT guests */
4346 	if (kvmhv_on_pseries())
4347 		info->flags |= KVM_PPC_NO_HASH;
4348 
4349 	return 0;
4350 }
4351 
4352 /*
4353  * Get (and clear) the dirty memory log for a memory slot.
4354  */
4355 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4356 					 struct kvm_dirty_log *log)
4357 {
4358 	struct kvm_memslots *slots;
4359 	struct kvm_memory_slot *memslot;
4360 	int i, r;
4361 	unsigned long n;
4362 	unsigned long *buf, *p;
4363 	struct kvm_vcpu *vcpu;
4364 
4365 	mutex_lock(&kvm->slots_lock);
4366 
4367 	r = -EINVAL;
4368 	if (log->slot >= KVM_USER_MEM_SLOTS)
4369 		goto out;
4370 
4371 	slots = kvm_memslots(kvm);
4372 	memslot = id_to_memslot(slots, log->slot);
4373 	r = -ENOENT;
4374 	if (!memslot->dirty_bitmap)
4375 		goto out;
4376 
4377 	/*
4378 	 * Use second half of bitmap area because both HPT and radix
4379 	 * accumulate bits in the first half.
4380 	 */
4381 	n = kvm_dirty_bitmap_bytes(memslot);
4382 	buf = memslot->dirty_bitmap + n / sizeof(long);
4383 	memset(buf, 0, n);
4384 
4385 	if (kvm_is_radix(kvm))
4386 		r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4387 	else
4388 		r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4389 	if (r)
4390 		goto out;
4391 
4392 	/*
4393 	 * We accumulate dirty bits in the first half of the
4394 	 * memslot's dirty_bitmap area, for when pages are paged
4395 	 * out or modified by the host directly.  Pick up these
4396 	 * bits and add them to the map.
4397 	 */
4398 	p = memslot->dirty_bitmap;
4399 	for (i = 0; i < n / sizeof(long); ++i)
4400 		buf[i] |= xchg(&p[i], 0);
4401 
4402 	/* Harvest dirty bits from VPA and DTL updates */
4403 	/* Note: we never modify the SLB shadow buffer areas */
4404 	kvm_for_each_vcpu(i, vcpu, kvm) {
4405 		spin_lock(&vcpu->arch.vpa_update_lock);
4406 		kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4407 		kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4408 		spin_unlock(&vcpu->arch.vpa_update_lock);
4409 	}
4410 
4411 	r = -EFAULT;
4412 	if (copy_to_user(log->dirty_bitmap, buf, n))
4413 		goto out;
4414 
4415 	r = 0;
4416 out:
4417 	mutex_unlock(&kvm->slots_lock);
4418 	return r;
4419 }
4420 
4421 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
4422 					struct kvm_memory_slot *dont)
4423 {
4424 	if (!dont || free->arch.rmap != dont->arch.rmap) {
4425 		vfree(free->arch.rmap);
4426 		free->arch.rmap = NULL;
4427 	}
4428 }
4429 
4430 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
4431 					 unsigned long npages)
4432 {
4433 	slot->arch.rmap = vzalloc(array_size(npages, sizeof(*slot->arch.rmap)));
4434 	if (!slot->arch.rmap)
4435 		return -ENOMEM;
4436 
4437 	return 0;
4438 }
4439 
4440 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4441 					struct kvm_memory_slot *memslot,
4442 					const struct kvm_userspace_memory_region *mem)
4443 {
4444 	return 0;
4445 }
4446 
4447 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
4448 				const struct kvm_userspace_memory_region *mem,
4449 				const struct kvm_memory_slot *old,
4450 				const struct kvm_memory_slot *new,
4451 				enum kvm_mr_change change)
4452 {
4453 	unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4454 
4455 	/*
4456 	 * If we are making a new memslot, it might make
4457 	 * some address that was previously cached as emulated
4458 	 * MMIO be no longer emulated MMIO, so invalidate
4459 	 * all the caches of emulated MMIO translations.
4460 	 */
4461 	if (npages)
4462 		atomic64_inc(&kvm->arch.mmio_update);
4463 
4464 	/*
4465 	 * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
4466 	 * have already called kvm_arch_flush_shadow_memslot() to
4467 	 * flush shadow mappings.  For KVM_MR_CREATE we have no
4468 	 * previous mappings.  So the only case to handle is
4469 	 * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
4470 	 * has been changed.
4471 	 * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
4472 	 * to get rid of any THP PTEs in the partition-scoped page tables
4473 	 * so we can track dirtiness at the page level; we flush when
4474 	 * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
4475 	 * using THP PTEs.
4476 	 */
4477 	if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
4478 	    ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
4479 		kvmppc_radix_flush_memslot(kvm, old);
4480 }
4481 
4482 /*
4483  * Update LPCR values in kvm->arch and in vcores.
4484  * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion
4485  * of kvm->arch.lpcr update).
4486  */
4487 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
4488 {
4489 	long int i;
4490 	u32 cores_done = 0;
4491 
4492 	if ((kvm->arch.lpcr & mask) == lpcr)
4493 		return;
4494 
4495 	kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
4496 
4497 	for (i = 0; i < KVM_MAX_VCORES; ++i) {
4498 		struct kvmppc_vcore *vc = kvm->arch.vcores[i];
4499 		if (!vc)
4500 			continue;
4501 		spin_lock(&vc->lock);
4502 		vc->lpcr = (vc->lpcr & ~mask) | lpcr;
4503 		spin_unlock(&vc->lock);
4504 		if (++cores_done >= kvm->arch.online_vcores)
4505 			break;
4506 	}
4507 }
4508 
4509 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
4510 {
4511 	return;
4512 }
4513 
4514 void kvmppc_setup_partition_table(struct kvm *kvm)
4515 {
4516 	unsigned long dw0, dw1;
4517 
4518 	if (!kvm_is_radix(kvm)) {
4519 		/* PS field - page size for VRMA */
4520 		dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
4521 			((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
4522 		/* HTABSIZE and HTABORG fields */
4523 		dw0 |= kvm->arch.sdr1;
4524 
4525 		/* Second dword as set by userspace */
4526 		dw1 = kvm->arch.process_table;
4527 	} else {
4528 		dw0 = PATB_HR | radix__get_tree_size() |
4529 			__pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
4530 		dw1 = PATB_GR | kvm->arch.process_table;
4531 	}
4532 	kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
4533 }
4534 
4535 /*
4536  * Set up HPT (hashed page table) and RMA (real-mode area).
4537  * Must be called with kvm->arch.mmu_setup_lock held.
4538  */
4539 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
4540 {
4541 	int err = 0;
4542 	struct kvm *kvm = vcpu->kvm;
4543 	unsigned long hva;
4544 	struct kvm_memory_slot *memslot;
4545 	struct vm_area_struct *vma;
4546 	unsigned long lpcr = 0, senc;
4547 	unsigned long psize, porder;
4548 	int srcu_idx;
4549 
4550 	/* Allocate hashed page table (if not done already) and reset it */
4551 	if (!kvm->arch.hpt.virt) {
4552 		int order = KVM_DEFAULT_HPT_ORDER;
4553 		struct kvm_hpt_info info;
4554 
4555 		err = kvmppc_allocate_hpt(&info, order);
4556 		/* If we get here, it means userspace didn't specify a
4557 		 * size explicitly.  So, try successively smaller
4558 		 * sizes if the default failed. */
4559 		while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
4560 			err  = kvmppc_allocate_hpt(&info, order);
4561 
4562 		if (err < 0) {
4563 			pr_err("KVM: Couldn't alloc HPT\n");
4564 			goto out;
4565 		}
4566 
4567 		kvmppc_set_hpt(kvm, &info);
4568 	}
4569 
4570 	/* Look up the memslot for guest physical address 0 */
4571 	srcu_idx = srcu_read_lock(&kvm->srcu);
4572 	memslot = gfn_to_memslot(kvm, 0);
4573 
4574 	/* We must have some memory at 0 by now */
4575 	err = -EINVAL;
4576 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
4577 		goto out_srcu;
4578 
4579 	/* Look up the VMA for the start of this memory slot */
4580 	hva = memslot->userspace_addr;
4581 	down_read(&current->mm->mmap_sem);
4582 	vma = find_vma(current->mm, hva);
4583 	if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
4584 		goto up_out;
4585 
4586 	psize = vma_kernel_pagesize(vma);
4587 
4588 	up_read(&current->mm->mmap_sem);
4589 
4590 	/* We can handle 4k, 64k or 16M pages in the VRMA */
4591 	if (psize >= 0x1000000)
4592 		psize = 0x1000000;
4593 	else if (psize >= 0x10000)
4594 		psize = 0x10000;
4595 	else
4596 		psize = 0x1000;
4597 	porder = __ilog2(psize);
4598 
4599 	senc = slb_pgsize_encoding(psize);
4600 	kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
4601 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
4602 	/* Create HPTEs in the hash page table for the VRMA */
4603 	kvmppc_map_vrma(vcpu, memslot, porder);
4604 
4605 	/* Update VRMASD field in the LPCR */
4606 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
4607 		/* the -4 is to account for senc values starting at 0x10 */
4608 		lpcr = senc << (LPCR_VRMASD_SH - 4);
4609 		kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
4610 	}
4611 
4612 	/* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
4613 	smp_wmb();
4614 	err = 0;
4615  out_srcu:
4616 	srcu_read_unlock(&kvm->srcu, srcu_idx);
4617  out:
4618 	return err;
4619 
4620  up_out:
4621 	up_read(&current->mm->mmap_sem);
4622 	goto out_srcu;
4623 }
4624 
4625 /*
4626  * Must be called with kvm->arch.mmu_setup_lock held and
4627  * mmu_ready = 0 and no vcpus running.
4628  */
4629 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
4630 {
4631 	if (nesting_enabled(kvm))
4632 		kvmhv_release_all_nested(kvm);
4633 	kvmppc_rmap_reset(kvm);
4634 	kvm->arch.process_table = 0;
4635 	/* Mutual exclusion with kvm_unmap_hva_range etc. */
4636 	spin_lock(&kvm->mmu_lock);
4637 	kvm->arch.radix = 0;
4638 	spin_unlock(&kvm->mmu_lock);
4639 	kvmppc_free_radix(kvm);
4640 	kvmppc_update_lpcr(kvm, LPCR_VPM1,
4641 			   LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4642 	return 0;
4643 }
4644 
4645 /*
4646  * Must be called with kvm->arch.mmu_setup_lock held and
4647  * mmu_ready = 0 and no vcpus running.
4648  */
4649 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
4650 {
4651 	int err;
4652 
4653 	err = kvmppc_init_vm_radix(kvm);
4654 	if (err)
4655 		return err;
4656 	kvmppc_rmap_reset(kvm);
4657 	/* Mutual exclusion with kvm_unmap_hva_range etc. */
4658 	spin_lock(&kvm->mmu_lock);
4659 	kvm->arch.radix = 1;
4660 	spin_unlock(&kvm->mmu_lock);
4661 	kvmppc_free_hpt(&kvm->arch.hpt);
4662 	kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
4663 			   LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4664 	return 0;
4665 }
4666 
4667 #ifdef CONFIG_KVM_XICS
4668 /*
4669  * Allocate a per-core structure for managing state about which cores are
4670  * running in the host versus the guest and for exchanging data between
4671  * real mode KVM and CPU running in the host.
4672  * This is only done for the first VM.
4673  * The allocated structure stays even if all VMs have stopped.
4674  * It is only freed when the kvm-hv module is unloaded.
4675  * It's OK for this routine to fail, we just don't support host
4676  * core operations like redirecting H_IPI wakeups.
4677  */
4678 void kvmppc_alloc_host_rm_ops(void)
4679 {
4680 	struct kvmppc_host_rm_ops *ops;
4681 	unsigned long l_ops;
4682 	int cpu, core;
4683 	int size;
4684 
4685 	/* Not the first time here ? */
4686 	if (kvmppc_host_rm_ops_hv != NULL)
4687 		return;
4688 
4689 	ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
4690 	if (!ops)
4691 		return;
4692 
4693 	size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
4694 	ops->rm_core = kzalloc(size, GFP_KERNEL);
4695 
4696 	if (!ops->rm_core) {
4697 		kfree(ops);
4698 		return;
4699 	}
4700 
4701 	cpus_read_lock();
4702 
4703 	for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
4704 		if (!cpu_online(cpu))
4705 			continue;
4706 
4707 		core = cpu >> threads_shift;
4708 		ops->rm_core[core].rm_state.in_host = 1;
4709 	}
4710 
4711 	ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
4712 
4713 	/*
4714 	 * Make the contents of the kvmppc_host_rm_ops structure visible
4715 	 * to other CPUs before we assign it to the global variable.
4716 	 * Do an atomic assignment (no locks used here), but if someone
4717 	 * beats us to it, just free our copy and return.
4718 	 */
4719 	smp_wmb();
4720 	l_ops = (unsigned long) ops;
4721 
4722 	if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
4723 		cpus_read_unlock();
4724 		kfree(ops->rm_core);
4725 		kfree(ops);
4726 		return;
4727 	}
4728 
4729 	cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
4730 					     "ppc/kvm_book3s:prepare",
4731 					     kvmppc_set_host_core,
4732 					     kvmppc_clear_host_core);
4733 	cpus_read_unlock();
4734 }
4735 
4736 void kvmppc_free_host_rm_ops(void)
4737 {
4738 	if (kvmppc_host_rm_ops_hv) {
4739 		cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
4740 		kfree(kvmppc_host_rm_ops_hv->rm_core);
4741 		kfree(kvmppc_host_rm_ops_hv);
4742 		kvmppc_host_rm_ops_hv = NULL;
4743 	}
4744 }
4745 #endif
4746 
4747 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
4748 {
4749 	unsigned long lpcr, lpid;
4750 	char buf[32];
4751 	int ret;
4752 
4753 	mutex_init(&kvm->arch.mmu_setup_lock);
4754 
4755 	/* Allocate the guest's logical partition ID */
4756 
4757 	lpid = kvmppc_alloc_lpid();
4758 	if ((long)lpid < 0)
4759 		return -ENOMEM;
4760 	kvm->arch.lpid = lpid;
4761 
4762 	kvmppc_alloc_host_rm_ops();
4763 
4764 	kvmhv_vm_nested_init(kvm);
4765 
4766 	/*
4767 	 * Since we don't flush the TLB when tearing down a VM,
4768 	 * and this lpid might have previously been used,
4769 	 * make sure we flush on each core before running the new VM.
4770 	 * On POWER9, the tlbie in mmu_partition_table_set_entry()
4771 	 * does this flush for us.
4772 	 */
4773 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
4774 		cpumask_setall(&kvm->arch.need_tlb_flush);
4775 
4776 	/* Start out with the default set of hcalls enabled */
4777 	memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
4778 	       sizeof(kvm->arch.enabled_hcalls));
4779 
4780 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
4781 		kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
4782 
4783 	/* Init LPCR for virtual RMA mode */
4784 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
4785 		kvm->arch.host_lpid = mfspr(SPRN_LPID);
4786 		kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
4787 		lpcr &= LPCR_PECE | LPCR_LPES;
4788 	} else {
4789 		lpcr = 0;
4790 	}
4791 	lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
4792 		LPCR_VPM0 | LPCR_VPM1;
4793 	kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
4794 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
4795 	/* On POWER8 turn on online bit to enable PURR/SPURR */
4796 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
4797 		lpcr |= LPCR_ONL;
4798 	/*
4799 	 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
4800 	 * Set HVICE bit to enable hypervisor virtualization interrupts.
4801 	 * Set HEIC to prevent OS interrupts to go to hypervisor (should
4802 	 * be unnecessary but better safe than sorry in case we re-enable
4803 	 * EE in HV mode with this LPCR still set)
4804 	 */
4805 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4806 		lpcr &= ~LPCR_VPM0;
4807 		lpcr |= LPCR_HVICE | LPCR_HEIC;
4808 
4809 		/*
4810 		 * If xive is enabled, we route 0x500 interrupts directly
4811 		 * to the guest.
4812 		 */
4813 		if (xics_on_xive())
4814 			lpcr |= LPCR_LPES;
4815 	}
4816 
4817 	/*
4818 	 * If the host uses radix, the guest starts out as radix.
4819 	 */
4820 	if (radix_enabled()) {
4821 		kvm->arch.radix = 1;
4822 		kvm->arch.mmu_ready = 1;
4823 		lpcr &= ~LPCR_VPM1;
4824 		lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
4825 		ret = kvmppc_init_vm_radix(kvm);
4826 		if (ret) {
4827 			kvmppc_free_lpid(kvm->arch.lpid);
4828 			return ret;
4829 		}
4830 		kvmppc_setup_partition_table(kvm);
4831 	}
4832 
4833 	kvm->arch.lpcr = lpcr;
4834 
4835 	/* Initialization for future HPT resizes */
4836 	kvm->arch.resize_hpt = NULL;
4837 
4838 	/*
4839 	 * Work out how many sets the TLB has, for the use of
4840 	 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
4841 	 */
4842 	if (radix_enabled())
4843 		kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;	/* 128 */
4844 	else if (cpu_has_feature(CPU_FTR_ARCH_300))
4845 		kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;	/* 256 */
4846 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
4847 		kvm->arch.tlb_sets = POWER8_TLB_SETS;		/* 512 */
4848 	else
4849 		kvm->arch.tlb_sets = POWER7_TLB_SETS;		/* 128 */
4850 
4851 	/*
4852 	 * Track that we now have a HV mode VM active. This blocks secondary
4853 	 * CPU threads from coming online.
4854 	 * On POWER9, we only need to do this if the "indep_threads_mode"
4855 	 * module parameter has been set to N.
4856 	 */
4857 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4858 		if (!indep_threads_mode && !cpu_has_feature(CPU_FTR_HVMODE)) {
4859 			pr_warn("KVM: Ignoring indep_threads_mode=N in nested hypervisor\n");
4860 			kvm->arch.threads_indep = true;
4861 		} else {
4862 			kvm->arch.threads_indep = indep_threads_mode;
4863 		}
4864 	}
4865 	if (!kvm->arch.threads_indep)
4866 		kvm_hv_vm_activated();
4867 
4868 	/*
4869 	 * Initialize smt_mode depending on processor.
4870 	 * POWER8 and earlier have to use "strict" threading, where
4871 	 * all vCPUs in a vcore have to run on the same (sub)core,
4872 	 * whereas on POWER9 the threads can each run a different
4873 	 * guest.
4874 	 */
4875 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
4876 		kvm->arch.smt_mode = threads_per_subcore;
4877 	else
4878 		kvm->arch.smt_mode = 1;
4879 	kvm->arch.emul_smt_mode = 1;
4880 
4881 	/*
4882 	 * Create a debugfs directory for the VM
4883 	 */
4884 	snprintf(buf, sizeof(buf), "vm%d", current->pid);
4885 	kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
4886 	kvmppc_mmu_debugfs_init(kvm);
4887 	if (radix_enabled())
4888 		kvmhv_radix_debugfs_init(kvm);
4889 
4890 	return 0;
4891 }
4892 
4893 static void kvmppc_free_vcores(struct kvm *kvm)
4894 {
4895 	long int i;
4896 
4897 	for (i = 0; i < KVM_MAX_VCORES; ++i)
4898 		kfree(kvm->arch.vcores[i]);
4899 	kvm->arch.online_vcores = 0;
4900 }
4901 
4902 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
4903 {
4904 	debugfs_remove_recursive(kvm->arch.debugfs_dir);
4905 
4906 	if (!kvm->arch.threads_indep)
4907 		kvm_hv_vm_deactivated();
4908 
4909 	kvmppc_free_vcores(kvm);
4910 
4911 
4912 	if (kvm_is_radix(kvm))
4913 		kvmppc_free_radix(kvm);
4914 	else
4915 		kvmppc_free_hpt(&kvm->arch.hpt);
4916 
4917 	/* Perform global invalidation and return lpid to the pool */
4918 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4919 		if (nesting_enabled(kvm))
4920 			kvmhv_release_all_nested(kvm);
4921 		kvm->arch.process_table = 0;
4922 		kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
4923 	}
4924 	kvmppc_free_lpid(kvm->arch.lpid);
4925 
4926 	kvmppc_free_pimap(kvm);
4927 }
4928 
4929 /* We don't need to emulate any privileged instructions or dcbz */
4930 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
4931 				     unsigned int inst, int *advance)
4932 {
4933 	return EMULATE_FAIL;
4934 }
4935 
4936 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
4937 					ulong spr_val)
4938 {
4939 	return EMULATE_FAIL;
4940 }
4941 
4942 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
4943 					ulong *spr_val)
4944 {
4945 	return EMULATE_FAIL;
4946 }
4947 
4948 static int kvmppc_core_check_processor_compat_hv(void)
4949 {
4950 	if (cpu_has_feature(CPU_FTR_HVMODE) &&
4951 	    cpu_has_feature(CPU_FTR_ARCH_206))
4952 		return 0;
4953 
4954 	/* POWER9 in radix mode is capable of being a nested hypervisor. */
4955 	if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
4956 		return 0;
4957 
4958 	return -EIO;
4959 }
4960 
4961 #ifdef CONFIG_KVM_XICS
4962 
4963 void kvmppc_free_pimap(struct kvm *kvm)
4964 {
4965 	kfree(kvm->arch.pimap);
4966 }
4967 
4968 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
4969 {
4970 	return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
4971 }
4972 
4973 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4974 {
4975 	struct irq_desc *desc;
4976 	struct kvmppc_irq_map *irq_map;
4977 	struct kvmppc_passthru_irqmap *pimap;
4978 	struct irq_chip *chip;
4979 	int i, rc = 0;
4980 
4981 	if (!kvm_irq_bypass)
4982 		return 1;
4983 
4984 	desc = irq_to_desc(host_irq);
4985 	if (!desc)
4986 		return -EIO;
4987 
4988 	mutex_lock(&kvm->lock);
4989 
4990 	pimap = kvm->arch.pimap;
4991 	if (pimap == NULL) {
4992 		/* First call, allocate structure to hold IRQ map */
4993 		pimap = kvmppc_alloc_pimap();
4994 		if (pimap == NULL) {
4995 			mutex_unlock(&kvm->lock);
4996 			return -ENOMEM;
4997 		}
4998 		kvm->arch.pimap = pimap;
4999 	}
5000 
5001 	/*
5002 	 * For now, we only support interrupts for which the EOI operation
5003 	 * is an OPAL call followed by a write to XIRR, since that's
5004 	 * what our real-mode EOI code does, or a XIVE interrupt
5005 	 */
5006 	chip = irq_data_get_irq_chip(&desc->irq_data);
5007 	if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
5008 		pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
5009 			host_irq, guest_gsi);
5010 		mutex_unlock(&kvm->lock);
5011 		return -ENOENT;
5012 	}
5013 
5014 	/*
5015 	 * See if we already have an entry for this guest IRQ number.
5016 	 * If it's mapped to a hardware IRQ number, that's an error,
5017 	 * otherwise re-use this entry.
5018 	 */
5019 	for (i = 0; i < pimap->n_mapped; i++) {
5020 		if (guest_gsi == pimap->mapped[i].v_hwirq) {
5021 			if (pimap->mapped[i].r_hwirq) {
5022 				mutex_unlock(&kvm->lock);
5023 				return -EINVAL;
5024 			}
5025 			break;
5026 		}
5027 	}
5028 
5029 	if (i == KVMPPC_PIRQ_MAPPED) {
5030 		mutex_unlock(&kvm->lock);
5031 		return -EAGAIN;		/* table is full */
5032 	}
5033 
5034 	irq_map = &pimap->mapped[i];
5035 
5036 	irq_map->v_hwirq = guest_gsi;
5037 	irq_map->desc = desc;
5038 
5039 	/*
5040 	 * Order the above two stores before the next to serialize with
5041 	 * the KVM real mode handler.
5042 	 */
5043 	smp_wmb();
5044 	irq_map->r_hwirq = desc->irq_data.hwirq;
5045 
5046 	if (i == pimap->n_mapped)
5047 		pimap->n_mapped++;
5048 
5049 	if (xics_on_xive())
5050 		rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
5051 	else
5052 		kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
5053 	if (rc)
5054 		irq_map->r_hwirq = 0;
5055 
5056 	mutex_unlock(&kvm->lock);
5057 
5058 	return 0;
5059 }
5060 
5061 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5062 {
5063 	struct irq_desc *desc;
5064 	struct kvmppc_passthru_irqmap *pimap;
5065 	int i, rc = 0;
5066 
5067 	if (!kvm_irq_bypass)
5068 		return 0;
5069 
5070 	desc = irq_to_desc(host_irq);
5071 	if (!desc)
5072 		return -EIO;
5073 
5074 	mutex_lock(&kvm->lock);
5075 	if (!kvm->arch.pimap)
5076 		goto unlock;
5077 
5078 	pimap = kvm->arch.pimap;
5079 
5080 	for (i = 0; i < pimap->n_mapped; i++) {
5081 		if (guest_gsi == pimap->mapped[i].v_hwirq)
5082 			break;
5083 	}
5084 
5085 	if (i == pimap->n_mapped) {
5086 		mutex_unlock(&kvm->lock);
5087 		return -ENODEV;
5088 	}
5089 
5090 	if (xics_on_xive())
5091 		rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
5092 	else
5093 		kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
5094 
5095 	/* invalidate the entry (what do do on error from the above ?) */
5096 	pimap->mapped[i].r_hwirq = 0;
5097 
5098 	/*
5099 	 * We don't free this structure even when the count goes to
5100 	 * zero. The structure is freed when we destroy the VM.
5101 	 */
5102  unlock:
5103 	mutex_unlock(&kvm->lock);
5104 	return rc;
5105 }
5106 
5107 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5108 					     struct irq_bypass_producer *prod)
5109 {
5110 	int ret = 0;
5111 	struct kvm_kernel_irqfd *irqfd =
5112 		container_of(cons, struct kvm_kernel_irqfd, consumer);
5113 
5114 	irqfd->producer = prod;
5115 
5116 	ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5117 	if (ret)
5118 		pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5119 			prod->irq, irqfd->gsi, ret);
5120 
5121 	return ret;
5122 }
5123 
5124 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5125 					      struct irq_bypass_producer *prod)
5126 {
5127 	int ret;
5128 	struct kvm_kernel_irqfd *irqfd =
5129 		container_of(cons, struct kvm_kernel_irqfd, consumer);
5130 
5131 	irqfd->producer = NULL;
5132 
5133 	/*
5134 	 * When producer of consumer is unregistered, we change back to
5135 	 * default external interrupt handling mode - KVM real mode
5136 	 * will switch back to host.
5137 	 */
5138 	ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5139 	if (ret)
5140 		pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5141 			prod->irq, irqfd->gsi, ret);
5142 }
5143 #endif
5144 
5145 static long kvm_arch_vm_ioctl_hv(struct file *filp,
5146 				 unsigned int ioctl, unsigned long arg)
5147 {
5148 	struct kvm *kvm __maybe_unused = filp->private_data;
5149 	void __user *argp = (void __user *)arg;
5150 	long r;
5151 
5152 	switch (ioctl) {
5153 
5154 	case KVM_PPC_ALLOCATE_HTAB: {
5155 		u32 htab_order;
5156 
5157 		r = -EFAULT;
5158 		if (get_user(htab_order, (u32 __user *)argp))
5159 			break;
5160 		r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5161 		if (r)
5162 			break;
5163 		r = 0;
5164 		break;
5165 	}
5166 
5167 	case KVM_PPC_GET_HTAB_FD: {
5168 		struct kvm_get_htab_fd ghf;
5169 
5170 		r = -EFAULT;
5171 		if (copy_from_user(&ghf, argp, sizeof(ghf)))
5172 			break;
5173 		r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5174 		break;
5175 	}
5176 
5177 	case KVM_PPC_RESIZE_HPT_PREPARE: {
5178 		struct kvm_ppc_resize_hpt rhpt;
5179 
5180 		r = -EFAULT;
5181 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5182 			break;
5183 
5184 		r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5185 		break;
5186 	}
5187 
5188 	case KVM_PPC_RESIZE_HPT_COMMIT: {
5189 		struct kvm_ppc_resize_hpt rhpt;
5190 
5191 		r = -EFAULT;
5192 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5193 			break;
5194 
5195 		r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5196 		break;
5197 	}
5198 
5199 	default:
5200 		r = -ENOTTY;
5201 	}
5202 
5203 	return r;
5204 }
5205 
5206 /*
5207  * List of hcall numbers to enable by default.
5208  * For compatibility with old userspace, we enable by default
5209  * all hcalls that were implemented before the hcall-enabling
5210  * facility was added.  Note this list should not include H_RTAS.
5211  */
5212 static unsigned int default_hcall_list[] = {
5213 	H_REMOVE,
5214 	H_ENTER,
5215 	H_READ,
5216 	H_PROTECT,
5217 	H_BULK_REMOVE,
5218 	H_GET_TCE,
5219 	H_PUT_TCE,
5220 	H_SET_DABR,
5221 	H_SET_XDABR,
5222 	H_CEDE,
5223 	H_PROD,
5224 	H_CONFER,
5225 	H_REGISTER_VPA,
5226 #ifdef CONFIG_KVM_XICS
5227 	H_EOI,
5228 	H_CPPR,
5229 	H_IPI,
5230 	H_IPOLL,
5231 	H_XIRR,
5232 	H_XIRR_X,
5233 #endif
5234 	0
5235 };
5236 
5237 static void init_default_hcalls(void)
5238 {
5239 	int i;
5240 	unsigned int hcall;
5241 
5242 	for (i = 0; default_hcall_list[i]; ++i) {
5243 		hcall = default_hcall_list[i];
5244 		WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5245 		__set_bit(hcall / 4, default_enabled_hcalls);
5246 	}
5247 }
5248 
5249 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5250 {
5251 	unsigned long lpcr;
5252 	int radix;
5253 	int err;
5254 
5255 	/* If not on a POWER9, reject it */
5256 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
5257 		return -ENODEV;
5258 
5259 	/* If any unknown flags set, reject it */
5260 	if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5261 		return -EINVAL;
5262 
5263 	/* GR (guest radix) bit in process_table field must match */
5264 	radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5265 	if (!!(cfg->process_table & PATB_GR) != radix)
5266 		return -EINVAL;
5267 
5268 	/* Process table size field must be reasonable, i.e. <= 24 */
5269 	if ((cfg->process_table & PRTS_MASK) > 24)
5270 		return -EINVAL;
5271 
5272 	/* We can change a guest to/from radix now, if the host is radix */
5273 	if (radix && !radix_enabled())
5274 		return -EINVAL;
5275 
5276 	/* If we're a nested hypervisor, we currently only support radix */
5277 	if (kvmhv_on_pseries() && !radix)
5278 		return -EINVAL;
5279 
5280 	mutex_lock(&kvm->arch.mmu_setup_lock);
5281 	if (radix != kvm_is_radix(kvm)) {
5282 		if (kvm->arch.mmu_ready) {
5283 			kvm->arch.mmu_ready = 0;
5284 			/* order mmu_ready vs. vcpus_running */
5285 			smp_mb();
5286 			if (atomic_read(&kvm->arch.vcpus_running)) {
5287 				kvm->arch.mmu_ready = 1;
5288 				err = -EBUSY;
5289 				goto out_unlock;
5290 			}
5291 		}
5292 		if (radix)
5293 			err = kvmppc_switch_mmu_to_radix(kvm);
5294 		else
5295 			err = kvmppc_switch_mmu_to_hpt(kvm);
5296 		if (err)
5297 			goto out_unlock;
5298 	}
5299 
5300 	kvm->arch.process_table = cfg->process_table;
5301 	kvmppc_setup_partition_table(kvm);
5302 
5303 	lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5304 	kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5305 	err = 0;
5306 
5307  out_unlock:
5308 	mutex_unlock(&kvm->arch.mmu_setup_lock);
5309 	return err;
5310 }
5311 
5312 static int kvmhv_enable_nested(struct kvm *kvm)
5313 {
5314 	if (!nested)
5315 		return -EPERM;
5316 	if (!cpu_has_feature(CPU_FTR_ARCH_300) || no_mixing_hpt_and_radix)
5317 		return -ENODEV;
5318 
5319 	/* kvm == NULL means the caller is testing if the capability exists */
5320 	if (kvm)
5321 		kvm->arch.nested_enable = true;
5322 	return 0;
5323 }
5324 
5325 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5326 				 int size)
5327 {
5328 	int rc = -EINVAL;
5329 
5330 	if (kvmhv_vcpu_is_radix(vcpu)) {
5331 		rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
5332 
5333 		if (rc > 0)
5334 			rc = -EINVAL;
5335 	}
5336 
5337 	/* For now quadrants are the only way to access nested guest memory */
5338 	if (rc && vcpu->arch.nested)
5339 		rc = -EAGAIN;
5340 
5341 	return rc;
5342 }
5343 
5344 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5345 				int size)
5346 {
5347 	int rc = -EINVAL;
5348 
5349 	if (kvmhv_vcpu_is_radix(vcpu)) {
5350 		rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
5351 
5352 		if (rc > 0)
5353 			rc = -EINVAL;
5354 	}
5355 
5356 	/* For now quadrants are the only way to access nested guest memory */
5357 	if (rc && vcpu->arch.nested)
5358 		rc = -EAGAIN;
5359 
5360 	return rc;
5361 }
5362 
5363 static struct kvmppc_ops kvm_ops_hv = {
5364 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
5365 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
5366 	.get_one_reg = kvmppc_get_one_reg_hv,
5367 	.set_one_reg = kvmppc_set_one_reg_hv,
5368 	.vcpu_load   = kvmppc_core_vcpu_load_hv,
5369 	.vcpu_put    = kvmppc_core_vcpu_put_hv,
5370 	.set_msr     = kvmppc_set_msr_hv,
5371 	.vcpu_run    = kvmppc_vcpu_run_hv,
5372 	.vcpu_create = kvmppc_core_vcpu_create_hv,
5373 	.vcpu_free   = kvmppc_core_vcpu_free_hv,
5374 	.check_requests = kvmppc_core_check_requests_hv,
5375 	.get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
5376 	.flush_memslot  = kvmppc_core_flush_memslot_hv,
5377 	.prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
5378 	.commit_memory_region  = kvmppc_core_commit_memory_region_hv,
5379 	.unmap_hva_range = kvm_unmap_hva_range_hv,
5380 	.age_hva  = kvm_age_hva_hv,
5381 	.test_age_hva = kvm_test_age_hva_hv,
5382 	.set_spte_hva = kvm_set_spte_hva_hv,
5383 	.mmu_destroy  = kvmppc_mmu_destroy_hv,
5384 	.free_memslot = kvmppc_core_free_memslot_hv,
5385 	.create_memslot = kvmppc_core_create_memslot_hv,
5386 	.init_vm =  kvmppc_core_init_vm_hv,
5387 	.destroy_vm = kvmppc_core_destroy_vm_hv,
5388 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
5389 	.emulate_op = kvmppc_core_emulate_op_hv,
5390 	.emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
5391 	.emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
5392 	.fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
5393 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
5394 	.hcall_implemented = kvmppc_hcall_impl_hv,
5395 #ifdef CONFIG_KVM_XICS
5396 	.irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
5397 	.irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
5398 #endif
5399 	.configure_mmu = kvmhv_configure_mmu,
5400 	.get_rmmu_info = kvmhv_get_rmmu_info,
5401 	.set_smt_mode = kvmhv_set_smt_mode,
5402 	.enable_nested = kvmhv_enable_nested,
5403 	.load_from_eaddr = kvmhv_load_from_eaddr,
5404 	.store_to_eaddr = kvmhv_store_to_eaddr,
5405 };
5406 
5407 static int kvm_init_subcore_bitmap(void)
5408 {
5409 	int i, j;
5410 	int nr_cores = cpu_nr_cores();
5411 	struct sibling_subcore_state *sibling_subcore_state;
5412 
5413 	for (i = 0; i < nr_cores; i++) {
5414 		int first_cpu = i * threads_per_core;
5415 		int node = cpu_to_node(first_cpu);
5416 
5417 		/* Ignore if it is already allocated. */
5418 		if (paca_ptrs[first_cpu]->sibling_subcore_state)
5419 			continue;
5420 
5421 		sibling_subcore_state =
5422 			kzalloc_node(sizeof(struct sibling_subcore_state),
5423 							GFP_KERNEL, node);
5424 		if (!sibling_subcore_state)
5425 			return -ENOMEM;
5426 
5427 
5428 		for (j = 0; j < threads_per_core; j++) {
5429 			int cpu = first_cpu + j;
5430 
5431 			paca_ptrs[cpu]->sibling_subcore_state =
5432 						sibling_subcore_state;
5433 		}
5434 	}
5435 	return 0;
5436 }
5437 
5438 static int kvmppc_radix_possible(void)
5439 {
5440 	return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
5441 }
5442 
5443 static int kvmppc_book3s_init_hv(void)
5444 {
5445 	int r;
5446 	/*
5447 	 * FIXME!! Do we need to check on all cpus ?
5448 	 */
5449 	r = kvmppc_core_check_processor_compat_hv();
5450 	if (r < 0)
5451 		return -ENODEV;
5452 
5453 	r = kvmhv_nested_init();
5454 	if (r)
5455 		return r;
5456 
5457 	r = kvm_init_subcore_bitmap();
5458 	if (r)
5459 		return r;
5460 
5461 	/*
5462 	 * We need a way of accessing the XICS interrupt controller,
5463 	 * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
5464 	 * indirectly, via OPAL.
5465 	 */
5466 #ifdef CONFIG_SMP
5467 	if (!xics_on_xive() && !kvmhv_on_pseries() &&
5468 	    !local_paca->kvm_hstate.xics_phys) {
5469 		struct device_node *np;
5470 
5471 		np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
5472 		if (!np) {
5473 			pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
5474 			return -ENODEV;
5475 		}
5476 		/* presence of intc confirmed - node can be dropped again */
5477 		of_node_put(np);
5478 	}
5479 #endif
5480 
5481 	kvm_ops_hv.owner = THIS_MODULE;
5482 	kvmppc_hv_ops = &kvm_ops_hv;
5483 
5484 	init_default_hcalls();
5485 
5486 	init_vcore_lists();
5487 
5488 	r = kvmppc_mmu_hv_init();
5489 	if (r)
5490 		return r;
5491 
5492 	if (kvmppc_radix_possible())
5493 		r = kvmppc_radix_init();
5494 
5495 	/*
5496 	 * POWER9 chips before version 2.02 can't have some threads in
5497 	 * HPT mode and some in radix mode on the same core.
5498 	 */
5499 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5500 		unsigned int pvr = mfspr(SPRN_PVR);
5501 		if ((pvr >> 16) == PVR_POWER9 &&
5502 		    (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
5503 		     ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
5504 			no_mixing_hpt_and_radix = true;
5505 	}
5506 
5507 	return r;
5508 }
5509 
5510 static void kvmppc_book3s_exit_hv(void)
5511 {
5512 	kvmppc_free_host_rm_ops();
5513 	if (kvmppc_radix_possible())
5514 		kvmppc_radix_exit();
5515 	kvmppc_hv_ops = NULL;
5516 	kvmhv_nested_exit();
5517 }
5518 
5519 module_init(kvmppc_book3s_init_hv);
5520 module_exit(kvmppc_book3s_exit_hv);
5521 MODULE_LICENSE("GPL");
5522 MODULE_ALIAS_MISCDEV(KVM_MINOR);
5523 MODULE_ALIAS("devname:kvm");
5524