xref: /openbmc/linux/arch/powerpc/kvm/book3s_hv.c (revision 65417d9f)
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/reg.h>
50 #include <asm/ppc-opcode.h>
51 #include <asm/asm-prototypes.h>
52 #include <asm/disassemble.h>
53 #include <asm/cputable.h>
54 #include <asm/cacheflush.h>
55 #include <asm/tlbflush.h>
56 #include <linux/uaccess.h>
57 #include <asm/io.h>
58 #include <asm/kvm_ppc.h>
59 #include <asm/kvm_book3s.h>
60 #include <asm/mmu_context.h>
61 #include <asm/lppaca.h>
62 #include <asm/processor.h>
63 #include <asm/cputhreads.h>
64 #include <asm/page.h>
65 #include <asm/hvcall.h>
66 #include <asm/switch_to.h>
67 #include <asm/smp.h>
68 #include <asm/dbell.h>
69 #include <asm/hmi.h>
70 #include <asm/pnv-pci.h>
71 #include <asm/mmu.h>
72 #include <asm/opal.h>
73 #include <asm/xics.h>
74 #include <asm/xive.h>
75 
76 #include "book3s.h"
77 
78 #define CREATE_TRACE_POINTS
79 #include "trace_hv.h"
80 
81 /* #define EXIT_DEBUG */
82 /* #define EXIT_DEBUG_SIMPLE */
83 /* #define EXIT_DEBUG_INT */
84 
85 /* Used to indicate that a guest page fault needs to be handled */
86 #define RESUME_PAGE_FAULT	(RESUME_GUEST | RESUME_FLAG_ARCH1)
87 /* Used to indicate that a guest passthrough interrupt needs to be handled */
88 #define RESUME_PASSTHROUGH	(RESUME_GUEST | RESUME_FLAG_ARCH2)
89 
90 /* Used as a "null" value for timebase values */
91 #define TB_NIL	(~(u64)0)
92 
93 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
94 
95 static int dynamic_mt_modes = 6;
96 module_param(dynamic_mt_modes, int, S_IRUGO | S_IWUSR);
97 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
98 static int target_smt_mode;
99 module_param(target_smt_mode, int, S_IRUGO | S_IWUSR);
100 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
101 
102 static bool indep_threads_mode = true;
103 module_param(indep_threads_mode, bool, S_IRUGO | S_IWUSR);
104 MODULE_PARM_DESC(indep_threads_mode, "Independent-threads mode (only on POWER9)");
105 
106 #ifdef CONFIG_KVM_XICS
107 static struct kernel_param_ops module_param_ops = {
108 	.set = param_set_int,
109 	.get = param_get_int,
110 };
111 
112 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass,
113 							S_IRUGO | S_IWUSR);
114 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
115 
116 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect,
117 							S_IRUGO | S_IWUSR);
118 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
119 #endif
120 
121 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
122 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
123 
124 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
125 		int *ip)
126 {
127 	int i = *ip;
128 	struct kvm_vcpu *vcpu;
129 
130 	while (++i < MAX_SMT_THREADS) {
131 		vcpu = READ_ONCE(vc->runnable_threads[i]);
132 		if (vcpu) {
133 			*ip = i;
134 			return vcpu;
135 		}
136 	}
137 	return NULL;
138 }
139 
140 /* Used to traverse the list of runnable threads for a given vcore */
141 #define for_each_runnable_thread(i, vcpu, vc) \
142 	for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
143 
144 static bool kvmppc_ipi_thread(int cpu)
145 {
146 	unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
147 
148 	/* On POWER9 we can use msgsnd to IPI any cpu */
149 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
150 		msg |= get_hard_smp_processor_id(cpu);
151 		smp_mb();
152 		__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
153 		return true;
154 	}
155 
156 	/* On POWER8 for IPIs to threads in the same core, use msgsnd */
157 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
158 		preempt_disable();
159 		if (cpu_first_thread_sibling(cpu) ==
160 		    cpu_first_thread_sibling(smp_processor_id())) {
161 			msg |= cpu_thread_in_core(cpu);
162 			smp_mb();
163 			__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
164 			preempt_enable();
165 			return true;
166 		}
167 		preempt_enable();
168 	}
169 
170 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
171 	if (cpu >= 0 && cpu < nr_cpu_ids) {
172 		if (paca[cpu].kvm_hstate.xics_phys) {
173 			xics_wake_cpu(cpu);
174 			return true;
175 		}
176 		opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
177 		return true;
178 	}
179 #endif
180 
181 	return false;
182 }
183 
184 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
185 {
186 	int cpu;
187 	struct swait_queue_head *wqp;
188 
189 	wqp = kvm_arch_vcpu_wq(vcpu);
190 	if (swq_has_sleeper(wqp)) {
191 		swake_up(wqp);
192 		++vcpu->stat.halt_wakeup;
193 	}
194 
195 	cpu = READ_ONCE(vcpu->arch.thread_cpu);
196 	if (cpu >= 0 && kvmppc_ipi_thread(cpu))
197 		return;
198 
199 	/* CPU points to the first thread of the core */
200 	cpu = vcpu->cpu;
201 	if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
202 		smp_send_reschedule(cpu);
203 }
204 
205 /*
206  * We use the vcpu_load/put functions to measure stolen time.
207  * Stolen time is counted as time when either the vcpu is able to
208  * run as part of a virtual core, but the task running the vcore
209  * is preempted or sleeping, or when the vcpu needs something done
210  * in the kernel by the task running the vcpu, but that task is
211  * preempted or sleeping.  Those two things have to be counted
212  * separately, since one of the vcpu tasks will take on the job
213  * of running the core, and the other vcpu tasks in the vcore will
214  * sleep waiting for it to do that, but that sleep shouldn't count
215  * as stolen time.
216  *
217  * Hence we accumulate stolen time when the vcpu can run as part of
218  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
219  * needs its task to do other things in the kernel (for example,
220  * service a page fault) in busy_stolen.  We don't accumulate
221  * stolen time for a vcore when it is inactive, or for a vcpu
222  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
223  * a misnomer; it means that the vcpu task is not executing in
224  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
225  * the kernel.  We don't have any way of dividing up that time
226  * between time that the vcpu is genuinely stopped, time that
227  * the task is actively working on behalf of the vcpu, and time
228  * that the task is preempted, so we don't count any of it as
229  * stolen.
230  *
231  * Updates to busy_stolen are protected by arch.tbacct_lock;
232  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
233  * lock.  The stolen times are measured in units of timebase ticks.
234  * (Note that the != TB_NIL checks below are purely defensive;
235  * they should never fail.)
236  */
237 
238 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
239 {
240 	unsigned long flags;
241 
242 	spin_lock_irqsave(&vc->stoltb_lock, flags);
243 	vc->preempt_tb = mftb();
244 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
245 }
246 
247 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
248 {
249 	unsigned long flags;
250 
251 	spin_lock_irqsave(&vc->stoltb_lock, flags);
252 	if (vc->preempt_tb != TB_NIL) {
253 		vc->stolen_tb += mftb() - vc->preempt_tb;
254 		vc->preempt_tb = TB_NIL;
255 	}
256 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
257 }
258 
259 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
260 {
261 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
262 	unsigned long flags;
263 
264 	/*
265 	 * We can test vc->runner without taking the vcore lock,
266 	 * because only this task ever sets vc->runner to this
267 	 * vcpu, and once it is set to this vcpu, only this task
268 	 * ever sets it to NULL.
269 	 */
270 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
271 		kvmppc_core_end_stolen(vc);
272 
273 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
274 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
275 	    vcpu->arch.busy_preempt != TB_NIL) {
276 		vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
277 		vcpu->arch.busy_preempt = TB_NIL;
278 	}
279 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
280 }
281 
282 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
283 {
284 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
285 	unsigned long flags;
286 
287 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
288 		kvmppc_core_start_stolen(vc);
289 
290 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
291 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
292 		vcpu->arch.busy_preempt = mftb();
293 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
294 }
295 
296 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
297 {
298 	/*
299 	 * Check for illegal transactional state bit combination
300 	 * and if we find it, force the TS field to a safe state.
301 	 */
302 	if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
303 		msr &= ~MSR_TS_MASK;
304 	vcpu->arch.shregs.msr = msr;
305 	kvmppc_end_cede(vcpu);
306 }
307 
308 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
309 {
310 	vcpu->arch.pvr = pvr;
311 }
312 
313 /* Dummy value used in computing PCR value below */
314 #define PCR_ARCH_300	(PCR_ARCH_207 << 1)
315 
316 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
317 {
318 	unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
319 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
320 
321 	/* We can (emulate) our own architecture version and anything older */
322 	if (cpu_has_feature(CPU_FTR_ARCH_300))
323 		host_pcr_bit = PCR_ARCH_300;
324 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
325 		host_pcr_bit = PCR_ARCH_207;
326 	else if (cpu_has_feature(CPU_FTR_ARCH_206))
327 		host_pcr_bit = PCR_ARCH_206;
328 	else
329 		host_pcr_bit = PCR_ARCH_205;
330 
331 	/* Determine lowest PCR bit needed to run guest in given PVR level */
332 	guest_pcr_bit = host_pcr_bit;
333 	if (arch_compat) {
334 		switch (arch_compat) {
335 		case PVR_ARCH_205:
336 			guest_pcr_bit = PCR_ARCH_205;
337 			break;
338 		case PVR_ARCH_206:
339 		case PVR_ARCH_206p:
340 			guest_pcr_bit = PCR_ARCH_206;
341 			break;
342 		case PVR_ARCH_207:
343 			guest_pcr_bit = PCR_ARCH_207;
344 			break;
345 		case PVR_ARCH_300:
346 			guest_pcr_bit = PCR_ARCH_300;
347 			break;
348 		default:
349 			return -EINVAL;
350 		}
351 	}
352 
353 	/* Check requested PCR bits don't exceed our capabilities */
354 	if (guest_pcr_bit > host_pcr_bit)
355 		return -EINVAL;
356 
357 	spin_lock(&vc->lock);
358 	vc->arch_compat = arch_compat;
359 	/* Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit */
360 	vc->pcr = host_pcr_bit - guest_pcr_bit;
361 	spin_unlock(&vc->lock);
362 
363 	return 0;
364 }
365 
366 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
367 {
368 	int r;
369 
370 	pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
371 	pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
372 	       vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
373 	for (r = 0; r < 16; ++r)
374 		pr_err("r%2d = %.16lx  r%d = %.16lx\n",
375 		       r, kvmppc_get_gpr(vcpu, r),
376 		       r+16, kvmppc_get_gpr(vcpu, r+16));
377 	pr_err("ctr = %.16lx  lr  = %.16lx\n",
378 	       vcpu->arch.ctr, vcpu->arch.lr);
379 	pr_err("srr0 = %.16llx srr1 = %.16llx\n",
380 	       vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
381 	pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
382 	       vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
383 	pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
384 	       vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
385 	pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
386 	       vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
387 	pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
388 	pr_err("fault dar = %.16lx dsisr = %.8x\n",
389 	       vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
390 	pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
391 	for (r = 0; r < vcpu->arch.slb_max; ++r)
392 		pr_err("  ESID = %.16llx VSID = %.16llx\n",
393 		       vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
394 	pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
395 	       vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
396 	       vcpu->arch.last_inst);
397 }
398 
399 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
400 {
401 	struct kvm_vcpu *ret;
402 
403 	mutex_lock(&kvm->lock);
404 	ret = kvm_get_vcpu_by_id(kvm, id);
405 	mutex_unlock(&kvm->lock);
406 	return ret;
407 }
408 
409 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
410 {
411 	vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
412 	vpa->yield_count = cpu_to_be32(1);
413 }
414 
415 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
416 		   unsigned long addr, unsigned long len)
417 {
418 	/* check address is cacheline aligned */
419 	if (addr & (L1_CACHE_BYTES - 1))
420 		return -EINVAL;
421 	spin_lock(&vcpu->arch.vpa_update_lock);
422 	if (v->next_gpa != addr || v->len != len) {
423 		v->next_gpa = addr;
424 		v->len = addr ? len : 0;
425 		v->update_pending = 1;
426 	}
427 	spin_unlock(&vcpu->arch.vpa_update_lock);
428 	return 0;
429 }
430 
431 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
432 struct reg_vpa {
433 	u32 dummy;
434 	union {
435 		__be16 hword;
436 		__be32 word;
437 	} length;
438 };
439 
440 static int vpa_is_registered(struct kvmppc_vpa *vpap)
441 {
442 	if (vpap->update_pending)
443 		return vpap->next_gpa != 0;
444 	return vpap->pinned_addr != NULL;
445 }
446 
447 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
448 				       unsigned long flags,
449 				       unsigned long vcpuid, unsigned long vpa)
450 {
451 	struct kvm *kvm = vcpu->kvm;
452 	unsigned long len, nb;
453 	void *va;
454 	struct kvm_vcpu *tvcpu;
455 	int err;
456 	int subfunc;
457 	struct kvmppc_vpa *vpap;
458 
459 	tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
460 	if (!tvcpu)
461 		return H_PARAMETER;
462 
463 	subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
464 	if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
465 	    subfunc == H_VPA_REG_SLB) {
466 		/* Registering new area - address must be cache-line aligned */
467 		if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
468 			return H_PARAMETER;
469 
470 		/* convert logical addr to kernel addr and read length */
471 		va = kvmppc_pin_guest_page(kvm, vpa, &nb);
472 		if (va == NULL)
473 			return H_PARAMETER;
474 		if (subfunc == H_VPA_REG_VPA)
475 			len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
476 		else
477 			len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
478 		kvmppc_unpin_guest_page(kvm, va, vpa, false);
479 
480 		/* Check length */
481 		if (len > nb || len < sizeof(struct reg_vpa))
482 			return H_PARAMETER;
483 	} else {
484 		vpa = 0;
485 		len = 0;
486 	}
487 
488 	err = H_PARAMETER;
489 	vpap = NULL;
490 	spin_lock(&tvcpu->arch.vpa_update_lock);
491 
492 	switch (subfunc) {
493 	case H_VPA_REG_VPA:		/* register VPA */
494 		/*
495 		 * The size of our lppaca is 1kB because of the way we align
496 		 * it for the guest to avoid crossing a 4kB boundary. We only
497 		 * use 640 bytes of the structure though, so we should accept
498 		 * clients that set a size of 640.
499 		 */
500 		if (len < 640)
501 			break;
502 		vpap = &tvcpu->arch.vpa;
503 		err = 0;
504 		break;
505 
506 	case H_VPA_REG_DTL:		/* register DTL */
507 		if (len < sizeof(struct dtl_entry))
508 			break;
509 		len -= len % sizeof(struct dtl_entry);
510 
511 		/* Check that they have previously registered a VPA */
512 		err = H_RESOURCE;
513 		if (!vpa_is_registered(&tvcpu->arch.vpa))
514 			break;
515 
516 		vpap = &tvcpu->arch.dtl;
517 		err = 0;
518 		break;
519 
520 	case H_VPA_REG_SLB:		/* register SLB shadow buffer */
521 		/* Check that they have previously registered a VPA */
522 		err = H_RESOURCE;
523 		if (!vpa_is_registered(&tvcpu->arch.vpa))
524 			break;
525 
526 		vpap = &tvcpu->arch.slb_shadow;
527 		err = 0;
528 		break;
529 
530 	case H_VPA_DEREG_VPA:		/* deregister VPA */
531 		/* Check they don't still have a DTL or SLB buf registered */
532 		err = H_RESOURCE;
533 		if (vpa_is_registered(&tvcpu->arch.dtl) ||
534 		    vpa_is_registered(&tvcpu->arch.slb_shadow))
535 			break;
536 
537 		vpap = &tvcpu->arch.vpa;
538 		err = 0;
539 		break;
540 
541 	case H_VPA_DEREG_DTL:		/* deregister DTL */
542 		vpap = &tvcpu->arch.dtl;
543 		err = 0;
544 		break;
545 
546 	case H_VPA_DEREG_SLB:		/* deregister SLB shadow buffer */
547 		vpap = &tvcpu->arch.slb_shadow;
548 		err = 0;
549 		break;
550 	}
551 
552 	if (vpap) {
553 		vpap->next_gpa = vpa;
554 		vpap->len = len;
555 		vpap->update_pending = 1;
556 	}
557 
558 	spin_unlock(&tvcpu->arch.vpa_update_lock);
559 
560 	return err;
561 }
562 
563 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
564 {
565 	struct kvm *kvm = vcpu->kvm;
566 	void *va;
567 	unsigned long nb;
568 	unsigned long gpa;
569 
570 	/*
571 	 * We need to pin the page pointed to by vpap->next_gpa,
572 	 * but we can't call kvmppc_pin_guest_page under the lock
573 	 * as it does get_user_pages() and down_read().  So we
574 	 * have to drop the lock, pin the page, then get the lock
575 	 * again and check that a new area didn't get registered
576 	 * in the meantime.
577 	 */
578 	for (;;) {
579 		gpa = vpap->next_gpa;
580 		spin_unlock(&vcpu->arch.vpa_update_lock);
581 		va = NULL;
582 		nb = 0;
583 		if (gpa)
584 			va = kvmppc_pin_guest_page(kvm, gpa, &nb);
585 		spin_lock(&vcpu->arch.vpa_update_lock);
586 		if (gpa == vpap->next_gpa)
587 			break;
588 		/* sigh... unpin that one and try again */
589 		if (va)
590 			kvmppc_unpin_guest_page(kvm, va, gpa, false);
591 	}
592 
593 	vpap->update_pending = 0;
594 	if (va && nb < vpap->len) {
595 		/*
596 		 * If it's now too short, it must be that userspace
597 		 * has changed the mappings underlying guest memory,
598 		 * so unregister the region.
599 		 */
600 		kvmppc_unpin_guest_page(kvm, va, gpa, false);
601 		va = NULL;
602 	}
603 	if (vpap->pinned_addr)
604 		kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
605 					vpap->dirty);
606 	vpap->gpa = gpa;
607 	vpap->pinned_addr = va;
608 	vpap->dirty = false;
609 	if (va)
610 		vpap->pinned_end = va + vpap->len;
611 }
612 
613 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
614 {
615 	if (!(vcpu->arch.vpa.update_pending ||
616 	      vcpu->arch.slb_shadow.update_pending ||
617 	      vcpu->arch.dtl.update_pending))
618 		return;
619 
620 	spin_lock(&vcpu->arch.vpa_update_lock);
621 	if (vcpu->arch.vpa.update_pending) {
622 		kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
623 		if (vcpu->arch.vpa.pinned_addr)
624 			init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
625 	}
626 	if (vcpu->arch.dtl.update_pending) {
627 		kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
628 		vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
629 		vcpu->arch.dtl_index = 0;
630 	}
631 	if (vcpu->arch.slb_shadow.update_pending)
632 		kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
633 	spin_unlock(&vcpu->arch.vpa_update_lock);
634 }
635 
636 /*
637  * Return the accumulated stolen time for the vcore up until `now'.
638  * The caller should hold the vcore lock.
639  */
640 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
641 {
642 	u64 p;
643 	unsigned long flags;
644 
645 	spin_lock_irqsave(&vc->stoltb_lock, flags);
646 	p = vc->stolen_tb;
647 	if (vc->vcore_state != VCORE_INACTIVE &&
648 	    vc->preempt_tb != TB_NIL)
649 		p += now - vc->preempt_tb;
650 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
651 	return p;
652 }
653 
654 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
655 				    struct kvmppc_vcore *vc)
656 {
657 	struct dtl_entry *dt;
658 	struct lppaca *vpa;
659 	unsigned long stolen;
660 	unsigned long core_stolen;
661 	u64 now;
662 	unsigned long flags;
663 
664 	dt = vcpu->arch.dtl_ptr;
665 	vpa = vcpu->arch.vpa.pinned_addr;
666 	now = mftb();
667 	core_stolen = vcore_stolen_time(vc, now);
668 	stolen = core_stolen - vcpu->arch.stolen_logged;
669 	vcpu->arch.stolen_logged = core_stolen;
670 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
671 	stolen += vcpu->arch.busy_stolen;
672 	vcpu->arch.busy_stolen = 0;
673 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
674 	if (!dt || !vpa)
675 		return;
676 	memset(dt, 0, sizeof(struct dtl_entry));
677 	dt->dispatch_reason = 7;
678 	dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
679 	dt->timebase = cpu_to_be64(now + vc->tb_offset);
680 	dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
681 	dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
682 	dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
683 	++dt;
684 	if (dt == vcpu->arch.dtl.pinned_end)
685 		dt = vcpu->arch.dtl.pinned_addr;
686 	vcpu->arch.dtl_ptr = dt;
687 	/* order writing *dt vs. writing vpa->dtl_idx */
688 	smp_wmb();
689 	vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
690 	vcpu->arch.dtl.dirty = true;
691 }
692 
693 /* See if there is a doorbell interrupt pending for a vcpu */
694 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
695 {
696 	int thr;
697 	struct kvmppc_vcore *vc;
698 
699 	if (vcpu->arch.doorbell_request)
700 		return true;
701 	/*
702 	 * Ensure that the read of vcore->dpdes comes after the read
703 	 * of vcpu->doorbell_request.  This barrier matches the
704 	 * lwsync in book3s_hv_rmhandlers.S just before the
705 	 * fast_guest_return label.
706 	 */
707 	smp_rmb();
708 	vc = vcpu->arch.vcore;
709 	thr = vcpu->vcpu_id - vc->first_vcpuid;
710 	return !!(vc->dpdes & (1 << thr));
711 }
712 
713 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
714 {
715 	if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
716 		return true;
717 	if ((!vcpu->arch.vcore->arch_compat) &&
718 	    cpu_has_feature(CPU_FTR_ARCH_207S))
719 		return true;
720 	return false;
721 }
722 
723 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
724 			     unsigned long resource, unsigned long value1,
725 			     unsigned long value2)
726 {
727 	switch (resource) {
728 	case H_SET_MODE_RESOURCE_SET_CIABR:
729 		if (!kvmppc_power8_compatible(vcpu))
730 			return H_P2;
731 		if (value2)
732 			return H_P4;
733 		if (mflags)
734 			return H_UNSUPPORTED_FLAG_START;
735 		/* Guests can't breakpoint the hypervisor */
736 		if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
737 			return H_P3;
738 		vcpu->arch.ciabr  = value1;
739 		return H_SUCCESS;
740 	case H_SET_MODE_RESOURCE_SET_DAWR:
741 		if (!kvmppc_power8_compatible(vcpu))
742 			return H_P2;
743 		if (mflags)
744 			return H_UNSUPPORTED_FLAG_START;
745 		if (value2 & DABRX_HYP)
746 			return H_P4;
747 		vcpu->arch.dawr  = value1;
748 		vcpu->arch.dawrx = value2;
749 		return H_SUCCESS;
750 	default:
751 		return H_TOO_HARD;
752 	}
753 }
754 
755 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
756 {
757 	struct kvmppc_vcore *vcore = target->arch.vcore;
758 
759 	/*
760 	 * We expect to have been called by the real mode handler
761 	 * (kvmppc_rm_h_confer()) which would have directly returned
762 	 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
763 	 * have useful work to do and should not confer) so we don't
764 	 * recheck that here.
765 	 */
766 
767 	spin_lock(&vcore->lock);
768 	if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
769 	    vcore->vcore_state != VCORE_INACTIVE &&
770 	    vcore->runner)
771 		target = vcore->runner;
772 	spin_unlock(&vcore->lock);
773 
774 	return kvm_vcpu_yield_to(target);
775 }
776 
777 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
778 {
779 	int yield_count = 0;
780 	struct lppaca *lppaca;
781 
782 	spin_lock(&vcpu->arch.vpa_update_lock);
783 	lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
784 	if (lppaca)
785 		yield_count = be32_to_cpu(lppaca->yield_count);
786 	spin_unlock(&vcpu->arch.vpa_update_lock);
787 	return yield_count;
788 }
789 
790 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
791 {
792 	unsigned long req = kvmppc_get_gpr(vcpu, 3);
793 	unsigned long target, ret = H_SUCCESS;
794 	int yield_count;
795 	struct kvm_vcpu *tvcpu;
796 	int idx, rc;
797 
798 	if (req <= MAX_HCALL_OPCODE &&
799 	    !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
800 		return RESUME_HOST;
801 
802 	switch (req) {
803 	case H_CEDE:
804 		break;
805 	case H_PROD:
806 		target = kvmppc_get_gpr(vcpu, 4);
807 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
808 		if (!tvcpu) {
809 			ret = H_PARAMETER;
810 			break;
811 		}
812 		tvcpu->arch.prodded = 1;
813 		smp_mb();
814 		if (tvcpu->arch.ceded)
815 			kvmppc_fast_vcpu_kick_hv(tvcpu);
816 		break;
817 	case H_CONFER:
818 		target = kvmppc_get_gpr(vcpu, 4);
819 		if (target == -1)
820 			break;
821 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
822 		if (!tvcpu) {
823 			ret = H_PARAMETER;
824 			break;
825 		}
826 		yield_count = kvmppc_get_gpr(vcpu, 5);
827 		if (kvmppc_get_yield_count(tvcpu) != yield_count)
828 			break;
829 		kvm_arch_vcpu_yield_to(tvcpu);
830 		break;
831 	case H_REGISTER_VPA:
832 		ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
833 					kvmppc_get_gpr(vcpu, 5),
834 					kvmppc_get_gpr(vcpu, 6));
835 		break;
836 	case H_RTAS:
837 		if (list_empty(&vcpu->kvm->arch.rtas_tokens))
838 			return RESUME_HOST;
839 
840 		idx = srcu_read_lock(&vcpu->kvm->srcu);
841 		rc = kvmppc_rtas_hcall(vcpu);
842 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
843 
844 		if (rc == -ENOENT)
845 			return RESUME_HOST;
846 		else if (rc == 0)
847 			break;
848 
849 		/* Send the error out to userspace via KVM_RUN */
850 		return rc;
851 	case H_LOGICAL_CI_LOAD:
852 		ret = kvmppc_h_logical_ci_load(vcpu);
853 		if (ret == H_TOO_HARD)
854 			return RESUME_HOST;
855 		break;
856 	case H_LOGICAL_CI_STORE:
857 		ret = kvmppc_h_logical_ci_store(vcpu);
858 		if (ret == H_TOO_HARD)
859 			return RESUME_HOST;
860 		break;
861 	case H_SET_MODE:
862 		ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
863 					kvmppc_get_gpr(vcpu, 5),
864 					kvmppc_get_gpr(vcpu, 6),
865 					kvmppc_get_gpr(vcpu, 7));
866 		if (ret == H_TOO_HARD)
867 			return RESUME_HOST;
868 		break;
869 	case H_XIRR:
870 	case H_CPPR:
871 	case H_EOI:
872 	case H_IPI:
873 	case H_IPOLL:
874 	case H_XIRR_X:
875 		if (kvmppc_xics_enabled(vcpu)) {
876 			if (xive_enabled()) {
877 				ret = H_NOT_AVAILABLE;
878 				return RESUME_GUEST;
879 			}
880 			ret = kvmppc_xics_hcall(vcpu, req);
881 			break;
882 		}
883 		return RESUME_HOST;
884 	case H_PUT_TCE:
885 		ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
886 						kvmppc_get_gpr(vcpu, 5),
887 						kvmppc_get_gpr(vcpu, 6));
888 		if (ret == H_TOO_HARD)
889 			return RESUME_HOST;
890 		break;
891 	case H_PUT_TCE_INDIRECT:
892 		ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
893 						kvmppc_get_gpr(vcpu, 5),
894 						kvmppc_get_gpr(vcpu, 6),
895 						kvmppc_get_gpr(vcpu, 7));
896 		if (ret == H_TOO_HARD)
897 			return RESUME_HOST;
898 		break;
899 	case H_STUFF_TCE:
900 		ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
901 						kvmppc_get_gpr(vcpu, 5),
902 						kvmppc_get_gpr(vcpu, 6),
903 						kvmppc_get_gpr(vcpu, 7));
904 		if (ret == H_TOO_HARD)
905 			return RESUME_HOST;
906 		break;
907 	default:
908 		return RESUME_HOST;
909 	}
910 	kvmppc_set_gpr(vcpu, 3, ret);
911 	vcpu->arch.hcall_needed = 0;
912 	return RESUME_GUEST;
913 }
914 
915 static int kvmppc_hcall_impl_hv(unsigned long cmd)
916 {
917 	switch (cmd) {
918 	case H_CEDE:
919 	case H_PROD:
920 	case H_CONFER:
921 	case H_REGISTER_VPA:
922 	case H_SET_MODE:
923 	case H_LOGICAL_CI_LOAD:
924 	case H_LOGICAL_CI_STORE:
925 #ifdef CONFIG_KVM_XICS
926 	case H_XIRR:
927 	case H_CPPR:
928 	case H_EOI:
929 	case H_IPI:
930 	case H_IPOLL:
931 	case H_XIRR_X:
932 #endif
933 		return 1;
934 	}
935 
936 	/* See if it's in the real-mode table */
937 	return kvmppc_hcall_impl_hv_realmode(cmd);
938 }
939 
940 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
941 					struct kvm_vcpu *vcpu)
942 {
943 	u32 last_inst;
944 
945 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
946 					EMULATE_DONE) {
947 		/*
948 		 * Fetch failed, so return to guest and
949 		 * try executing it again.
950 		 */
951 		return RESUME_GUEST;
952 	}
953 
954 	if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
955 		run->exit_reason = KVM_EXIT_DEBUG;
956 		run->debug.arch.address = kvmppc_get_pc(vcpu);
957 		return RESUME_HOST;
958 	} else {
959 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
960 		return RESUME_GUEST;
961 	}
962 }
963 
964 static void do_nothing(void *x)
965 {
966 }
967 
968 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
969 {
970 	int thr, cpu, pcpu, nthreads;
971 	struct kvm_vcpu *v;
972 	unsigned long dpdes;
973 
974 	nthreads = vcpu->kvm->arch.emul_smt_mode;
975 	dpdes = 0;
976 	cpu = vcpu->vcpu_id & ~(nthreads - 1);
977 	for (thr = 0; thr < nthreads; ++thr, ++cpu) {
978 		v = kvmppc_find_vcpu(vcpu->kvm, cpu);
979 		if (!v)
980 			continue;
981 		/*
982 		 * If the vcpu is currently running on a physical cpu thread,
983 		 * interrupt it in order to pull it out of the guest briefly,
984 		 * which will update its vcore->dpdes value.
985 		 */
986 		pcpu = READ_ONCE(v->cpu);
987 		if (pcpu >= 0)
988 			smp_call_function_single(pcpu, do_nothing, NULL, 1);
989 		if (kvmppc_doorbell_pending(v))
990 			dpdes |= 1 << thr;
991 	}
992 	return dpdes;
993 }
994 
995 /*
996  * On POWER9, emulate doorbell-related instructions in order to
997  * give the guest the illusion of running on a multi-threaded core.
998  * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
999  * and mfspr DPDES.
1000  */
1001 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1002 {
1003 	u32 inst, rb, thr;
1004 	unsigned long arg;
1005 	struct kvm *kvm = vcpu->kvm;
1006 	struct kvm_vcpu *tvcpu;
1007 
1008 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
1009 		return EMULATE_FAIL;
1010 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1011 		return RESUME_GUEST;
1012 	if (get_op(inst) != 31)
1013 		return EMULATE_FAIL;
1014 	rb = get_rb(inst);
1015 	thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1016 	switch (get_xop(inst)) {
1017 	case OP_31_XOP_MSGSNDP:
1018 		arg = kvmppc_get_gpr(vcpu, rb);
1019 		if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1020 			break;
1021 		arg &= 0x3f;
1022 		if (arg >= kvm->arch.emul_smt_mode)
1023 			break;
1024 		tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1025 		if (!tvcpu)
1026 			break;
1027 		if (!tvcpu->arch.doorbell_request) {
1028 			tvcpu->arch.doorbell_request = 1;
1029 			kvmppc_fast_vcpu_kick_hv(tvcpu);
1030 		}
1031 		break;
1032 	case OP_31_XOP_MSGCLRP:
1033 		arg = kvmppc_get_gpr(vcpu, rb);
1034 		if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1035 			break;
1036 		vcpu->arch.vcore->dpdes = 0;
1037 		vcpu->arch.doorbell_request = 0;
1038 		break;
1039 	case OP_31_XOP_MFSPR:
1040 		switch (get_sprn(inst)) {
1041 		case SPRN_TIR:
1042 			arg = thr;
1043 			break;
1044 		case SPRN_DPDES:
1045 			arg = kvmppc_read_dpdes(vcpu);
1046 			break;
1047 		default:
1048 			return EMULATE_FAIL;
1049 		}
1050 		kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1051 		break;
1052 	default:
1053 		return EMULATE_FAIL;
1054 	}
1055 	kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1056 	return RESUME_GUEST;
1057 }
1058 
1059 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
1060 				 struct task_struct *tsk)
1061 {
1062 	int r = RESUME_HOST;
1063 
1064 	vcpu->stat.sum_exits++;
1065 
1066 	/*
1067 	 * This can happen if an interrupt occurs in the last stages
1068 	 * of guest entry or the first stages of guest exit (i.e. after
1069 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1070 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
1071 	 * That can happen due to a bug, or due to a machine check
1072 	 * occurring at just the wrong time.
1073 	 */
1074 	if (vcpu->arch.shregs.msr & MSR_HV) {
1075 		printk(KERN_EMERG "KVM trap in HV mode!\n");
1076 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1077 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1078 			vcpu->arch.shregs.msr);
1079 		kvmppc_dump_regs(vcpu);
1080 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1081 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1082 		return RESUME_HOST;
1083 	}
1084 	run->exit_reason = KVM_EXIT_UNKNOWN;
1085 	run->ready_for_interrupt_injection = 1;
1086 	switch (vcpu->arch.trap) {
1087 	/* We're good on these - the host merely wanted to get our attention */
1088 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1089 		vcpu->stat.dec_exits++;
1090 		r = RESUME_GUEST;
1091 		break;
1092 	case BOOK3S_INTERRUPT_EXTERNAL:
1093 	case BOOK3S_INTERRUPT_H_DOORBELL:
1094 	case BOOK3S_INTERRUPT_H_VIRT:
1095 		vcpu->stat.ext_intr_exits++;
1096 		r = RESUME_GUEST;
1097 		break;
1098 	/* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1099 	case BOOK3S_INTERRUPT_HMI:
1100 	case BOOK3S_INTERRUPT_PERFMON:
1101 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
1102 		r = RESUME_GUEST;
1103 		break;
1104 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
1105 		/* Exit to guest with KVM_EXIT_NMI as exit reason */
1106 		run->exit_reason = KVM_EXIT_NMI;
1107 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1108 		/* Clear out the old NMI status from run->flags */
1109 		run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1110 		/* Now set the NMI status */
1111 		if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1112 			run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1113 		else
1114 			run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1115 
1116 		r = RESUME_HOST;
1117 		/* Print the MCE event to host console. */
1118 		machine_check_print_event_info(&vcpu->arch.mce_evt, false);
1119 		break;
1120 	case BOOK3S_INTERRUPT_PROGRAM:
1121 	{
1122 		ulong flags;
1123 		/*
1124 		 * Normally program interrupts are delivered directly
1125 		 * to the guest by the hardware, but we can get here
1126 		 * as a result of a hypervisor emulation interrupt
1127 		 * (e40) getting turned into a 700 by BML RTAS.
1128 		 */
1129 		flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1130 		kvmppc_core_queue_program(vcpu, flags);
1131 		r = RESUME_GUEST;
1132 		break;
1133 	}
1134 	case BOOK3S_INTERRUPT_SYSCALL:
1135 	{
1136 		/* hcall - punt to userspace */
1137 		int i;
1138 
1139 		/* hypercall with MSR_PR has already been handled in rmode,
1140 		 * and never reaches here.
1141 		 */
1142 
1143 		run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1144 		for (i = 0; i < 9; ++i)
1145 			run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1146 		run->exit_reason = KVM_EXIT_PAPR_HCALL;
1147 		vcpu->arch.hcall_needed = 1;
1148 		r = RESUME_HOST;
1149 		break;
1150 	}
1151 	/*
1152 	 * We get these next two if the guest accesses a page which it thinks
1153 	 * it has mapped but which is not actually present, either because
1154 	 * it is for an emulated I/O device or because the corresonding
1155 	 * host page has been paged out.  Any other HDSI/HISI interrupts
1156 	 * have been handled already.
1157 	 */
1158 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1159 		r = RESUME_PAGE_FAULT;
1160 		break;
1161 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
1162 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1163 		vcpu->arch.fault_dsisr = 0;
1164 		r = RESUME_PAGE_FAULT;
1165 		break;
1166 	/*
1167 	 * This occurs if the guest executes an illegal instruction.
1168 	 * If the guest debug is disabled, generate a program interrupt
1169 	 * to the guest. If guest debug is enabled, we need to check
1170 	 * whether the instruction is a software breakpoint instruction.
1171 	 * Accordingly return to Guest or Host.
1172 	 */
1173 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1174 		if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1175 			vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1176 				swab32(vcpu->arch.emul_inst) :
1177 				vcpu->arch.emul_inst;
1178 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1179 			r = kvmppc_emulate_debug_inst(run, vcpu);
1180 		} else {
1181 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1182 			r = RESUME_GUEST;
1183 		}
1184 		break;
1185 	/*
1186 	 * This occurs if the guest (kernel or userspace), does something that
1187 	 * is prohibited by HFSCR.
1188 	 * On POWER9, this could be a doorbell instruction that we need
1189 	 * to emulate.
1190 	 * Otherwise, we just generate a program interrupt to the guest.
1191 	 */
1192 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1193 		r = EMULATE_FAIL;
1194 		if ((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG)
1195 			r = kvmppc_emulate_doorbell_instr(vcpu);
1196 		if (r == EMULATE_FAIL) {
1197 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1198 			r = RESUME_GUEST;
1199 		}
1200 		break;
1201 	case BOOK3S_INTERRUPT_HV_RM_HARD:
1202 		r = RESUME_PASSTHROUGH;
1203 		break;
1204 	default:
1205 		kvmppc_dump_regs(vcpu);
1206 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1207 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1208 			vcpu->arch.shregs.msr);
1209 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1210 		r = RESUME_HOST;
1211 		break;
1212 	}
1213 
1214 	return r;
1215 }
1216 
1217 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1218 					    struct kvm_sregs *sregs)
1219 {
1220 	int i;
1221 
1222 	memset(sregs, 0, sizeof(struct kvm_sregs));
1223 	sregs->pvr = vcpu->arch.pvr;
1224 	for (i = 0; i < vcpu->arch.slb_max; i++) {
1225 		sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1226 		sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1227 	}
1228 
1229 	return 0;
1230 }
1231 
1232 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1233 					    struct kvm_sregs *sregs)
1234 {
1235 	int i, j;
1236 
1237 	/* Only accept the same PVR as the host's, since we can't spoof it */
1238 	if (sregs->pvr != vcpu->arch.pvr)
1239 		return -EINVAL;
1240 
1241 	j = 0;
1242 	for (i = 0; i < vcpu->arch.slb_nr; i++) {
1243 		if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1244 			vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1245 			vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1246 			++j;
1247 		}
1248 	}
1249 	vcpu->arch.slb_max = j;
1250 
1251 	return 0;
1252 }
1253 
1254 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1255 		bool preserve_top32)
1256 {
1257 	struct kvm *kvm = vcpu->kvm;
1258 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
1259 	u64 mask;
1260 
1261 	mutex_lock(&kvm->lock);
1262 	spin_lock(&vc->lock);
1263 	/*
1264 	 * If ILE (interrupt little-endian) has changed, update the
1265 	 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1266 	 */
1267 	if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1268 		struct kvm_vcpu *vcpu;
1269 		int i;
1270 
1271 		kvm_for_each_vcpu(i, vcpu, kvm) {
1272 			if (vcpu->arch.vcore != vc)
1273 				continue;
1274 			if (new_lpcr & LPCR_ILE)
1275 				vcpu->arch.intr_msr |= MSR_LE;
1276 			else
1277 				vcpu->arch.intr_msr &= ~MSR_LE;
1278 		}
1279 	}
1280 
1281 	/*
1282 	 * Userspace can only modify DPFD (default prefetch depth),
1283 	 * ILE (interrupt little-endian) and TC (translation control).
1284 	 * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1285 	 */
1286 	mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1287 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
1288 		mask |= LPCR_AIL;
1289 	/*
1290 	 * On POWER9, allow userspace to enable large decrementer for the
1291 	 * guest, whether or not the host has it enabled.
1292 	 */
1293 	if (cpu_has_feature(CPU_FTR_ARCH_300))
1294 		mask |= LPCR_LD;
1295 
1296 	/* Broken 32-bit version of LPCR must not clear top bits */
1297 	if (preserve_top32)
1298 		mask &= 0xFFFFFFFF;
1299 	vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1300 	spin_unlock(&vc->lock);
1301 	mutex_unlock(&kvm->lock);
1302 }
1303 
1304 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1305 				 union kvmppc_one_reg *val)
1306 {
1307 	int r = 0;
1308 	long int i;
1309 
1310 	switch (id) {
1311 	case KVM_REG_PPC_DEBUG_INST:
1312 		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1313 		break;
1314 	case KVM_REG_PPC_HIOR:
1315 		*val = get_reg_val(id, 0);
1316 		break;
1317 	case KVM_REG_PPC_DABR:
1318 		*val = get_reg_val(id, vcpu->arch.dabr);
1319 		break;
1320 	case KVM_REG_PPC_DABRX:
1321 		*val = get_reg_val(id, vcpu->arch.dabrx);
1322 		break;
1323 	case KVM_REG_PPC_DSCR:
1324 		*val = get_reg_val(id, vcpu->arch.dscr);
1325 		break;
1326 	case KVM_REG_PPC_PURR:
1327 		*val = get_reg_val(id, vcpu->arch.purr);
1328 		break;
1329 	case KVM_REG_PPC_SPURR:
1330 		*val = get_reg_val(id, vcpu->arch.spurr);
1331 		break;
1332 	case KVM_REG_PPC_AMR:
1333 		*val = get_reg_val(id, vcpu->arch.amr);
1334 		break;
1335 	case KVM_REG_PPC_UAMOR:
1336 		*val = get_reg_val(id, vcpu->arch.uamor);
1337 		break;
1338 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1339 		i = id - KVM_REG_PPC_MMCR0;
1340 		*val = get_reg_val(id, vcpu->arch.mmcr[i]);
1341 		break;
1342 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1343 		i = id - KVM_REG_PPC_PMC1;
1344 		*val = get_reg_val(id, vcpu->arch.pmc[i]);
1345 		break;
1346 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1347 		i = id - KVM_REG_PPC_SPMC1;
1348 		*val = get_reg_val(id, vcpu->arch.spmc[i]);
1349 		break;
1350 	case KVM_REG_PPC_SIAR:
1351 		*val = get_reg_val(id, vcpu->arch.siar);
1352 		break;
1353 	case KVM_REG_PPC_SDAR:
1354 		*val = get_reg_val(id, vcpu->arch.sdar);
1355 		break;
1356 	case KVM_REG_PPC_SIER:
1357 		*val = get_reg_val(id, vcpu->arch.sier);
1358 		break;
1359 	case KVM_REG_PPC_IAMR:
1360 		*val = get_reg_val(id, vcpu->arch.iamr);
1361 		break;
1362 	case KVM_REG_PPC_PSPB:
1363 		*val = get_reg_val(id, vcpu->arch.pspb);
1364 		break;
1365 	case KVM_REG_PPC_DPDES:
1366 		*val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1367 		break;
1368 	case KVM_REG_PPC_VTB:
1369 		*val = get_reg_val(id, vcpu->arch.vcore->vtb);
1370 		break;
1371 	case KVM_REG_PPC_DAWR:
1372 		*val = get_reg_val(id, vcpu->arch.dawr);
1373 		break;
1374 	case KVM_REG_PPC_DAWRX:
1375 		*val = get_reg_val(id, vcpu->arch.dawrx);
1376 		break;
1377 	case KVM_REG_PPC_CIABR:
1378 		*val = get_reg_val(id, vcpu->arch.ciabr);
1379 		break;
1380 	case KVM_REG_PPC_CSIGR:
1381 		*val = get_reg_val(id, vcpu->arch.csigr);
1382 		break;
1383 	case KVM_REG_PPC_TACR:
1384 		*val = get_reg_val(id, vcpu->arch.tacr);
1385 		break;
1386 	case KVM_REG_PPC_TCSCR:
1387 		*val = get_reg_val(id, vcpu->arch.tcscr);
1388 		break;
1389 	case KVM_REG_PPC_PID:
1390 		*val = get_reg_val(id, vcpu->arch.pid);
1391 		break;
1392 	case KVM_REG_PPC_ACOP:
1393 		*val = get_reg_val(id, vcpu->arch.acop);
1394 		break;
1395 	case KVM_REG_PPC_WORT:
1396 		*val = get_reg_val(id, vcpu->arch.wort);
1397 		break;
1398 	case KVM_REG_PPC_TIDR:
1399 		*val = get_reg_val(id, vcpu->arch.tid);
1400 		break;
1401 	case KVM_REG_PPC_PSSCR:
1402 		*val = get_reg_val(id, vcpu->arch.psscr);
1403 		break;
1404 	case KVM_REG_PPC_VPA_ADDR:
1405 		spin_lock(&vcpu->arch.vpa_update_lock);
1406 		*val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1407 		spin_unlock(&vcpu->arch.vpa_update_lock);
1408 		break;
1409 	case KVM_REG_PPC_VPA_SLB:
1410 		spin_lock(&vcpu->arch.vpa_update_lock);
1411 		val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1412 		val->vpaval.length = vcpu->arch.slb_shadow.len;
1413 		spin_unlock(&vcpu->arch.vpa_update_lock);
1414 		break;
1415 	case KVM_REG_PPC_VPA_DTL:
1416 		spin_lock(&vcpu->arch.vpa_update_lock);
1417 		val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1418 		val->vpaval.length = vcpu->arch.dtl.len;
1419 		spin_unlock(&vcpu->arch.vpa_update_lock);
1420 		break;
1421 	case KVM_REG_PPC_TB_OFFSET:
1422 		*val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1423 		break;
1424 	case KVM_REG_PPC_LPCR:
1425 	case KVM_REG_PPC_LPCR_64:
1426 		*val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1427 		break;
1428 	case KVM_REG_PPC_PPR:
1429 		*val = get_reg_val(id, vcpu->arch.ppr);
1430 		break;
1431 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1432 	case KVM_REG_PPC_TFHAR:
1433 		*val = get_reg_val(id, vcpu->arch.tfhar);
1434 		break;
1435 	case KVM_REG_PPC_TFIAR:
1436 		*val = get_reg_val(id, vcpu->arch.tfiar);
1437 		break;
1438 	case KVM_REG_PPC_TEXASR:
1439 		*val = get_reg_val(id, vcpu->arch.texasr);
1440 		break;
1441 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1442 		i = id - KVM_REG_PPC_TM_GPR0;
1443 		*val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1444 		break;
1445 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1446 	{
1447 		int j;
1448 		i = id - KVM_REG_PPC_TM_VSR0;
1449 		if (i < 32)
1450 			for (j = 0; j < TS_FPRWIDTH; j++)
1451 				val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1452 		else {
1453 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1454 				val->vval = vcpu->arch.vr_tm.vr[i-32];
1455 			else
1456 				r = -ENXIO;
1457 		}
1458 		break;
1459 	}
1460 	case KVM_REG_PPC_TM_CR:
1461 		*val = get_reg_val(id, vcpu->arch.cr_tm);
1462 		break;
1463 	case KVM_REG_PPC_TM_XER:
1464 		*val = get_reg_val(id, vcpu->arch.xer_tm);
1465 		break;
1466 	case KVM_REG_PPC_TM_LR:
1467 		*val = get_reg_val(id, vcpu->arch.lr_tm);
1468 		break;
1469 	case KVM_REG_PPC_TM_CTR:
1470 		*val = get_reg_val(id, vcpu->arch.ctr_tm);
1471 		break;
1472 	case KVM_REG_PPC_TM_FPSCR:
1473 		*val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1474 		break;
1475 	case KVM_REG_PPC_TM_AMR:
1476 		*val = get_reg_val(id, vcpu->arch.amr_tm);
1477 		break;
1478 	case KVM_REG_PPC_TM_PPR:
1479 		*val = get_reg_val(id, vcpu->arch.ppr_tm);
1480 		break;
1481 	case KVM_REG_PPC_TM_VRSAVE:
1482 		*val = get_reg_val(id, vcpu->arch.vrsave_tm);
1483 		break;
1484 	case KVM_REG_PPC_TM_VSCR:
1485 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1486 			*val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1487 		else
1488 			r = -ENXIO;
1489 		break;
1490 	case KVM_REG_PPC_TM_DSCR:
1491 		*val = get_reg_val(id, vcpu->arch.dscr_tm);
1492 		break;
1493 	case KVM_REG_PPC_TM_TAR:
1494 		*val = get_reg_val(id, vcpu->arch.tar_tm);
1495 		break;
1496 #endif
1497 	case KVM_REG_PPC_ARCH_COMPAT:
1498 		*val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1499 		break;
1500 	default:
1501 		r = -EINVAL;
1502 		break;
1503 	}
1504 
1505 	return r;
1506 }
1507 
1508 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1509 				 union kvmppc_one_reg *val)
1510 {
1511 	int r = 0;
1512 	long int i;
1513 	unsigned long addr, len;
1514 
1515 	switch (id) {
1516 	case KVM_REG_PPC_HIOR:
1517 		/* Only allow this to be set to zero */
1518 		if (set_reg_val(id, *val))
1519 			r = -EINVAL;
1520 		break;
1521 	case KVM_REG_PPC_DABR:
1522 		vcpu->arch.dabr = set_reg_val(id, *val);
1523 		break;
1524 	case KVM_REG_PPC_DABRX:
1525 		vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1526 		break;
1527 	case KVM_REG_PPC_DSCR:
1528 		vcpu->arch.dscr = set_reg_val(id, *val);
1529 		break;
1530 	case KVM_REG_PPC_PURR:
1531 		vcpu->arch.purr = set_reg_val(id, *val);
1532 		break;
1533 	case KVM_REG_PPC_SPURR:
1534 		vcpu->arch.spurr = set_reg_val(id, *val);
1535 		break;
1536 	case KVM_REG_PPC_AMR:
1537 		vcpu->arch.amr = set_reg_val(id, *val);
1538 		break;
1539 	case KVM_REG_PPC_UAMOR:
1540 		vcpu->arch.uamor = set_reg_val(id, *val);
1541 		break;
1542 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1543 		i = id - KVM_REG_PPC_MMCR0;
1544 		vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1545 		break;
1546 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1547 		i = id - KVM_REG_PPC_PMC1;
1548 		vcpu->arch.pmc[i] = set_reg_val(id, *val);
1549 		break;
1550 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1551 		i = id - KVM_REG_PPC_SPMC1;
1552 		vcpu->arch.spmc[i] = set_reg_val(id, *val);
1553 		break;
1554 	case KVM_REG_PPC_SIAR:
1555 		vcpu->arch.siar = set_reg_val(id, *val);
1556 		break;
1557 	case KVM_REG_PPC_SDAR:
1558 		vcpu->arch.sdar = set_reg_val(id, *val);
1559 		break;
1560 	case KVM_REG_PPC_SIER:
1561 		vcpu->arch.sier = set_reg_val(id, *val);
1562 		break;
1563 	case KVM_REG_PPC_IAMR:
1564 		vcpu->arch.iamr = set_reg_val(id, *val);
1565 		break;
1566 	case KVM_REG_PPC_PSPB:
1567 		vcpu->arch.pspb = set_reg_val(id, *val);
1568 		break;
1569 	case KVM_REG_PPC_DPDES:
1570 		vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1571 		break;
1572 	case KVM_REG_PPC_VTB:
1573 		vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1574 		break;
1575 	case KVM_REG_PPC_DAWR:
1576 		vcpu->arch.dawr = set_reg_val(id, *val);
1577 		break;
1578 	case KVM_REG_PPC_DAWRX:
1579 		vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1580 		break;
1581 	case KVM_REG_PPC_CIABR:
1582 		vcpu->arch.ciabr = set_reg_val(id, *val);
1583 		/* Don't allow setting breakpoints in hypervisor code */
1584 		if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1585 			vcpu->arch.ciabr &= ~CIABR_PRIV;	/* disable */
1586 		break;
1587 	case KVM_REG_PPC_CSIGR:
1588 		vcpu->arch.csigr = set_reg_val(id, *val);
1589 		break;
1590 	case KVM_REG_PPC_TACR:
1591 		vcpu->arch.tacr = set_reg_val(id, *val);
1592 		break;
1593 	case KVM_REG_PPC_TCSCR:
1594 		vcpu->arch.tcscr = set_reg_val(id, *val);
1595 		break;
1596 	case KVM_REG_PPC_PID:
1597 		vcpu->arch.pid = set_reg_val(id, *val);
1598 		break;
1599 	case KVM_REG_PPC_ACOP:
1600 		vcpu->arch.acop = set_reg_val(id, *val);
1601 		break;
1602 	case KVM_REG_PPC_WORT:
1603 		vcpu->arch.wort = set_reg_val(id, *val);
1604 		break;
1605 	case KVM_REG_PPC_TIDR:
1606 		vcpu->arch.tid = set_reg_val(id, *val);
1607 		break;
1608 	case KVM_REG_PPC_PSSCR:
1609 		vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1610 		break;
1611 	case KVM_REG_PPC_VPA_ADDR:
1612 		addr = set_reg_val(id, *val);
1613 		r = -EINVAL;
1614 		if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1615 			      vcpu->arch.dtl.next_gpa))
1616 			break;
1617 		r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1618 		break;
1619 	case KVM_REG_PPC_VPA_SLB:
1620 		addr = val->vpaval.addr;
1621 		len = val->vpaval.length;
1622 		r = -EINVAL;
1623 		if (addr && !vcpu->arch.vpa.next_gpa)
1624 			break;
1625 		r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1626 		break;
1627 	case KVM_REG_PPC_VPA_DTL:
1628 		addr = val->vpaval.addr;
1629 		len = val->vpaval.length;
1630 		r = -EINVAL;
1631 		if (addr && (len < sizeof(struct dtl_entry) ||
1632 			     !vcpu->arch.vpa.next_gpa))
1633 			break;
1634 		len -= len % sizeof(struct dtl_entry);
1635 		r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1636 		break;
1637 	case KVM_REG_PPC_TB_OFFSET:
1638 		/*
1639 		 * POWER9 DD1 has an erratum where writing TBU40 causes
1640 		 * the timebase to lose ticks.  So we don't let the
1641 		 * timebase offset be changed on P9 DD1.  (It is
1642 		 * initialized to zero.)
1643 		 */
1644 		if (cpu_has_feature(CPU_FTR_POWER9_DD1))
1645 			break;
1646 		/* round up to multiple of 2^24 */
1647 		vcpu->arch.vcore->tb_offset =
1648 			ALIGN(set_reg_val(id, *val), 1UL << 24);
1649 		break;
1650 	case KVM_REG_PPC_LPCR:
1651 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1652 		break;
1653 	case KVM_REG_PPC_LPCR_64:
1654 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1655 		break;
1656 	case KVM_REG_PPC_PPR:
1657 		vcpu->arch.ppr = set_reg_val(id, *val);
1658 		break;
1659 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1660 	case KVM_REG_PPC_TFHAR:
1661 		vcpu->arch.tfhar = set_reg_val(id, *val);
1662 		break;
1663 	case KVM_REG_PPC_TFIAR:
1664 		vcpu->arch.tfiar = set_reg_val(id, *val);
1665 		break;
1666 	case KVM_REG_PPC_TEXASR:
1667 		vcpu->arch.texasr = set_reg_val(id, *val);
1668 		break;
1669 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1670 		i = id - KVM_REG_PPC_TM_GPR0;
1671 		vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1672 		break;
1673 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1674 	{
1675 		int j;
1676 		i = id - KVM_REG_PPC_TM_VSR0;
1677 		if (i < 32)
1678 			for (j = 0; j < TS_FPRWIDTH; j++)
1679 				vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1680 		else
1681 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1682 				vcpu->arch.vr_tm.vr[i-32] = val->vval;
1683 			else
1684 				r = -ENXIO;
1685 		break;
1686 	}
1687 	case KVM_REG_PPC_TM_CR:
1688 		vcpu->arch.cr_tm = set_reg_val(id, *val);
1689 		break;
1690 	case KVM_REG_PPC_TM_XER:
1691 		vcpu->arch.xer_tm = set_reg_val(id, *val);
1692 		break;
1693 	case KVM_REG_PPC_TM_LR:
1694 		vcpu->arch.lr_tm = set_reg_val(id, *val);
1695 		break;
1696 	case KVM_REG_PPC_TM_CTR:
1697 		vcpu->arch.ctr_tm = set_reg_val(id, *val);
1698 		break;
1699 	case KVM_REG_PPC_TM_FPSCR:
1700 		vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1701 		break;
1702 	case KVM_REG_PPC_TM_AMR:
1703 		vcpu->arch.amr_tm = set_reg_val(id, *val);
1704 		break;
1705 	case KVM_REG_PPC_TM_PPR:
1706 		vcpu->arch.ppr_tm = set_reg_val(id, *val);
1707 		break;
1708 	case KVM_REG_PPC_TM_VRSAVE:
1709 		vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1710 		break;
1711 	case KVM_REG_PPC_TM_VSCR:
1712 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1713 			vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1714 		else
1715 			r = - ENXIO;
1716 		break;
1717 	case KVM_REG_PPC_TM_DSCR:
1718 		vcpu->arch.dscr_tm = set_reg_val(id, *val);
1719 		break;
1720 	case KVM_REG_PPC_TM_TAR:
1721 		vcpu->arch.tar_tm = set_reg_val(id, *val);
1722 		break;
1723 #endif
1724 	case KVM_REG_PPC_ARCH_COMPAT:
1725 		r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1726 		break;
1727 	default:
1728 		r = -EINVAL;
1729 		break;
1730 	}
1731 
1732 	return r;
1733 }
1734 
1735 /*
1736  * On POWER9, threads are independent and can be in different partitions.
1737  * Therefore we consider each thread to be a subcore.
1738  * There is a restriction that all threads have to be in the same
1739  * MMU mode (radix or HPT), unfortunately, but since we only support
1740  * HPT guests on a HPT host so far, that isn't an impediment yet.
1741  */
1742 static int threads_per_vcore(struct kvm *kvm)
1743 {
1744 	if (kvm->arch.threads_indep)
1745 		return 1;
1746 	return threads_per_subcore;
1747 }
1748 
1749 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1750 {
1751 	struct kvmppc_vcore *vcore;
1752 
1753 	vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1754 
1755 	if (vcore == NULL)
1756 		return NULL;
1757 
1758 	spin_lock_init(&vcore->lock);
1759 	spin_lock_init(&vcore->stoltb_lock);
1760 	init_swait_queue_head(&vcore->wq);
1761 	vcore->preempt_tb = TB_NIL;
1762 	vcore->lpcr = kvm->arch.lpcr;
1763 	vcore->first_vcpuid = core * kvm->arch.smt_mode;
1764 	vcore->kvm = kvm;
1765 	INIT_LIST_HEAD(&vcore->preempt_list);
1766 
1767 	return vcore;
1768 }
1769 
1770 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1771 static struct debugfs_timings_element {
1772 	const char *name;
1773 	size_t offset;
1774 } timings[] = {
1775 	{"rm_entry",	offsetof(struct kvm_vcpu, arch.rm_entry)},
1776 	{"rm_intr",	offsetof(struct kvm_vcpu, arch.rm_intr)},
1777 	{"rm_exit",	offsetof(struct kvm_vcpu, arch.rm_exit)},
1778 	{"guest",	offsetof(struct kvm_vcpu, arch.guest_time)},
1779 	{"cede",	offsetof(struct kvm_vcpu, arch.cede_time)},
1780 };
1781 
1782 #define N_TIMINGS	(ARRAY_SIZE(timings))
1783 
1784 struct debugfs_timings_state {
1785 	struct kvm_vcpu	*vcpu;
1786 	unsigned int	buflen;
1787 	char		buf[N_TIMINGS * 100];
1788 };
1789 
1790 static int debugfs_timings_open(struct inode *inode, struct file *file)
1791 {
1792 	struct kvm_vcpu *vcpu = inode->i_private;
1793 	struct debugfs_timings_state *p;
1794 
1795 	p = kzalloc(sizeof(*p), GFP_KERNEL);
1796 	if (!p)
1797 		return -ENOMEM;
1798 
1799 	kvm_get_kvm(vcpu->kvm);
1800 	p->vcpu = vcpu;
1801 	file->private_data = p;
1802 
1803 	return nonseekable_open(inode, file);
1804 }
1805 
1806 static int debugfs_timings_release(struct inode *inode, struct file *file)
1807 {
1808 	struct debugfs_timings_state *p = file->private_data;
1809 
1810 	kvm_put_kvm(p->vcpu->kvm);
1811 	kfree(p);
1812 	return 0;
1813 }
1814 
1815 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1816 				    size_t len, loff_t *ppos)
1817 {
1818 	struct debugfs_timings_state *p = file->private_data;
1819 	struct kvm_vcpu *vcpu = p->vcpu;
1820 	char *s, *buf_end;
1821 	struct kvmhv_tb_accumulator tb;
1822 	u64 count;
1823 	loff_t pos;
1824 	ssize_t n;
1825 	int i, loops;
1826 	bool ok;
1827 
1828 	if (!p->buflen) {
1829 		s = p->buf;
1830 		buf_end = s + sizeof(p->buf);
1831 		for (i = 0; i < N_TIMINGS; ++i) {
1832 			struct kvmhv_tb_accumulator *acc;
1833 
1834 			acc = (struct kvmhv_tb_accumulator *)
1835 				((unsigned long)vcpu + timings[i].offset);
1836 			ok = false;
1837 			for (loops = 0; loops < 1000; ++loops) {
1838 				count = acc->seqcount;
1839 				if (!(count & 1)) {
1840 					smp_rmb();
1841 					tb = *acc;
1842 					smp_rmb();
1843 					if (count == acc->seqcount) {
1844 						ok = true;
1845 						break;
1846 					}
1847 				}
1848 				udelay(1);
1849 			}
1850 			if (!ok)
1851 				snprintf(s, buf_end - s, "%s: stuck\n",
1852 					timings[i].name);
1853 			else
1854 				snprintf(s, buf_end - s,
1855 					"%s: %llu %llu %llu %llu\n",
1856 					timings[i].name, count / 2,
1857 					tb_to_ns(tb.tb_total),
1858 					tb_to_ns(tb.tb_min),
1859 					tb_to_ns(tb.tb_max));
1860 			s += strlen(s);
1861 		}
1862 		p->buflen = s - p->buf;
1863 	}
1864 
1865 	pos = *ppos;
1866 	if (pos >= p->buflen)
1867 		return 0;
1868 	if (len > p->buflen - pos)
1869 		len = p->buflen - pos;
1870 	n = copy_to_user(buf, p->buf + pos, len);
1871 	if (n) {
1872 		if (n == len)
1873 			return -EFAULT;
1874 		len -= n;
1875 	}
1876 	*ppos = pos + len;
1877 	return len;
1878 }
1879 
1880 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1881 				     size_t len, loff_t *ppos)
1882 {
1883 	return -EACCES;
1884 }
1885 
1886 static const struct file_operations debugfs_timings_ops = {
1887 	.owner	 = THIS_MODULE,
1888 	.open	 = debugfs_timings_open,
1889 	.release = debugfs_timings_release,
1890 	.read	 = debugfs_timings_read,
1891 	.write	 = debugfs_timings_write,
1892 	.llseek	 = generic_file_llseek,
1893 };
1894 
1895 /* Create a debugfs directory for the vcpu */
1896 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1897 {
1898 	char buf[16];
1899 	struct kvm *kvm = vcpu->kvm;
1900 
1901 	snprintf(buf, sizeof(buf), "vcpu%u", id);
1902 	if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1903 		return;
1904 	vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1905 	if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1906 		return;
1907 	vcpu->arch.debugfs_timings =
1908 		debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1909 				    vcpu, &debugfs_timings_ops);
1910 }
1911 
1912 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1913 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1914 {
1915 }
1916 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1917 
1918 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1919 						   unsigned int id)
1920 {
1921 	struct kvm_vcpu *vcpu;
1922 	int err;
1923 	int core;
1924 	struct kvmppc_vcore *vcore;
1925 
1926 	err = -ENOMEM;
1927 	vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1928 	if (!vcpu)
1929 		goto out;
1930 
1931 	err = kvm_vcpu_init(vcpu, kvm, id);
1932 	if (err)
1933 		goto free_vcpu;
1934 
1935 	vcpu->arch.shared = &vcpu->arch.shregs;
1936 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1937 	/*
1938 	 * The shared struct is never shared on HV,
1939 	 * so we can always use host endianness
1940 	 */
1941 #ifdef __BIG_ENDIAN__
1942 	vcpu->arch.shared_big_endian = true;
1943 #else
1944 	vcpu->arch.shared_big_endian = false;
1945 #endif
1946 #endif
1947 	vcpu->arch.mmcr[0] = MMCR0_FC;
1948 	vcpu->arch.ctrl = CTRL_RUNLATCH;
1949 	/* default to host PVR, since we can't spoof it */
1950 	kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1951 	spin_lock_init(&vcpu->arch.vpa_update_lock);
1952 	spin_lock_init(&vcpu->arch.tbacct_lock);
1953 	vcpu->arch.busy_preempt = TB_NIL;
1954 	vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1955 
1956 	/*
1957 	 * Set the default HFSCR for the guest from the host value.
1958 	 * This value is only used on POWER9.
1959 	 * On POWER9 DD1, TM doesn't work, so we make sure to
1960 	 * prevent the guest from using it.
1961 	 * On POWER9, we want to virtualize the doorbell facility, so we
1962 	 * turn off the HFSCR bit, which causes those instructions to trap.
1963 	 */
1964 	vcpu->arch.hfscr = mfspr(SPRN_HFSCR);
1965 	if (!cpu_has_feature(CPU_FTR_TM))
1966 		vcpu->arch.hfscr &= ~HFSCR_TM;
1967 	if (cpu_has_feature(CPU_FTR_ARCH_300))
1968 		vcpu->arch.hfscr &= ~HFSCR_MSGP;
1969 
1970 	kvmppc_mmu_book3s_hv_init(vcpu);
1971 
1972 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1973 
1974 	init_waitqueue_head(&vcpu->arch.cpu_run);
1975 
1976 	mutex_lock(&kvm->lock);
1977 	vcore = NULL;
1978 	err = -EINVAL;
1979 	core = id / kvm->arch.smt_mode;
1980 	if (core < KVM_MAX_VCORES) {
1981 		vcore = kvm->arch.vcores[core];
1982 		if (!vcore) {
1983 			err = -ENOMEM;
1984 			vcore = kvmppc_vcore_create(kvm, core);
1985 			kvm->arch.vcores[core] = vcore;
1986 			kvm->arch.online_vcores++;
1987 		}
1988 	}
1989 	mutex_unlock(&kvm->lock);
1990 
1991 	if (!vcore)
1992 		goto free_vcpu;
1993 
1994 	spin_lock(&vcore->lock);
1995 	++vcore->num_threads;
1996 	spin_unlock(&vcore->lock);
1997 	vcpu->arch.vcore = vcore;
1998 	vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1999 	vcpu->arch.thread_cpu = -1;
2000 	vcpu->arch.prev_cpu = -1;
2001 
2002 	vcpu->arch.cpu_type = KVM_CPU_3S_64;
2003 	kvmppc_sanity_check(vcpu);
2004 
2005 	debugfs_vcpu_init(vcpu, id);
2006 
2007 	return vcpu;
2008 
2009 free_vcpu:
2010 	kmem_cache_free(kvm_vcpu_cache, vcpu);
2011 out:
2012 	return ERR_PTR(err);
2013 }
2014 
2015 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2016 			      unsigned long flags)
2017 {
2018 	int err;
2019 	int esmt = 0;
2020 
2021 	if (flags)
2022 		return -EINVAL;
2023 	if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2024 		return -EINVAL;
2025 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2026 		/*
2027 		 * On POWER8 (or POWER7), the threading mode is "strict",
2028 		 * so we pack smt_mode vcpus per vcore.
2029 		 */
2030 		if (smt_mode > threads_per_subcore)
2031 			return -EINVAL;
2032 	} else {
2033 		/*
2034 		 * On POWER9, the threading mode is "loose",
2035 		 * so each vcpu gets its own vcore.
2036 		 */
2037 		esmt = smt_mode;
2038 		smt_mode = 1;
2039 	}
2040 	mutex_lock(&kvm->lock);
2041 	err = -EBUSY;
2042 	if (!kvm->arch.online_vcores) {
2043 		kvm->arch.smt_mode = smt_mode;
2044 		kvm->arch.emul_smt_mode = esmt;
2045 		err = 0;
2046 	}
2047 	mutex_unlock(&kvm->lock);
2048 
2049 	return err;
2050 }
2051 
2052 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2053 {
2054 	if (vpa->pinned_addr)
2055 		kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2056 					vpa->dirty);
2057 }
2058 
2059 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2060 {
2061 	spin_lock(&vcpu->arch.vpa_update_lock);
2062 	unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2063 	unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2064 	unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2065 	spin_unlock(&vcpu->arch.vpa_update_lock);
2066 	kvm_vcpu_uninit(vcpu);
2067 	kmem_cache_free(kvm_vcpu_cache, vcpu);
2068 }
2069 
2070 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2071 {
2072 	/* Indicate we want to get back into the guest */
2073 	return 1;
2074 }
2075 
2076 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2077 {
2078 	unsigned long dec_nsec, now;
2079 
2080 	now = get_tb();
2081 	if (now > vcpu->arch.dec_expires) {
2082 		/* decrementer has already gone negative */
2083 		kvmppc_core_queue_dec(vcpu);
2084 		kvmppc_core_prepare_to_enter(vcpu);
2085 		return;
2086 	}
2087 	dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
2088 		   / tb_ticks_per_sec;
2089 	hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2090 	vcpu->arch.timer_running = 1;
2091 }
2092 
2093 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
2094 {
2095 	vcpu->arch.ceded = 0;
2096 	if (vcpu->arch.timer_running) {
2097 		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2098 		vcpu->arch.timer_running = 0;
2099 	}
2100 }
2101 
2102 extern int __kvmppc_vcore_entry(void);
2103 
2104 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2105 				   struct kvm_vcpu *vcpu)
2106 {
2107 	u64 now;
2108 
2109 	if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2110 		return;
2111 	spin_lock_irq(&vcpu->arch.tbacct_lock);
2112 	now = mftb();
2113 	vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2114 		vcpu->arch.stolen_logged;
2115 	vcpu->arch.busy_preempt = now;
2116 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2117 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
2118 	--vc->n_runnable;
2119 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2120 }
2121 
2122 static int kvmppc_grab_hwthread(int cpu)
2123 {
2124 	struct paca_struct *tpaca;
2125 	long timeout = 10000;
2126 
2127 	tpaca = &paca[cpu];
2128 
2129 	/* Ensure the thread won't go into the kernel if it wakes */
2130 	tpaca->kvm_hstate.kvm_vcpu = NULL;
2131 	tpaca->kvm_hstate.kvm_vcore = NULL;
2132 	tpaca->kvm_hstate.napping = 0;
2133 	smp_wmb();
2134 	tpaca->kvm_hstate.hwthread_req = 1;
2135 
2136 	/*
2137 	 * If the thread is already executing in the kernel (e.g. handling
2138 	 * a stray interrupt), wait for it to get back to nap mode.
2139 	 * The smp_mb() is to ensure that our setting of hwthread_req
2140 	 * is visible before we look at hwthread_state, so if this
2141 	 * races with the code at system_reset_pSeries and the thread
2142 	 * misses our setting of hwthread_req, we are sure to see its
2143 	 * setting of hwthread_state, and vice versa.
2144 	 */
2145 	smp_mb();
2146 	while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2147 		if (--timeout <= 0) {
2148 			pr_err("KVM: couldn't grab cpu %d\n", cpu);
2149 			return -EBUSY;
2150 		}
2151 		udelay(1);
2152 	}
2153 	return 0;
2154 }
2155 
2156 static void kvmppc_release_hwthread(int cpu)
2157 {
2158 	struct paca_struct *tpaca;
2159 
2160 	tpaca = &paca[cpu];
2161 	tpaca->kvm_hstate.hwthread_req = 0;
2162 	tpaca->kvm_hstate.kvm_vcpu = NULL;
2163 	tpaca->kvm_hstate.kvm_vcore = NULL;
2164 	tpaca->kvm_hstate.kvm_split_mode = NULL;
2165 }
2166 
2167 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2168 {
2169 	int i;
2170 
2171 	cpu = cpu_first_thread_sibling(cpu);
2172 	cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2173 	/*
2174 	 * Make sure setting of bit in need_tlb_flush precedes
2175 	 * testing of cpu_in_guest bits.  The matching barrier on
2176 	 * the other side is the first smp_mb() in kvmppc_run_core().
2177 	 */
2178 	smp_mb();
2179 	for (i = 0; i < threads_per_core; ++i)
2180 		if (cpumask_test_cpu(cpu + i, &kvm->arch.cpu_in_guest))
2181 			smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2182 }
2183 
2184 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2185 {
2186 	struct kvm *kvm = vcpu->kvm;
2187 
2188 	/*
2189 	 * With radix, the guest can do TLB invalidations itself,
2190 	 * and it could choose to use the local form (tlbiel) if
2191 	 * it is invalidating a translation that has only ever been
2192 	 * used on one vcpu.  However, that doesn't mean it has
2193 	 * only ever been used on one physical cpu, since vcpus
2194 	 * can move around between pcpus.  To cope with this, when
2195 	 * a vcpu moves from one pcpu to another, we need to tell
2196 	 * any vcpus running on the same core as this vcpu previously
2197 	 * ran to flush the TLB.  The TLB is shared between threads,
2198 	 * so we use a single bit in .need_tlb_flush for all 4 threads.
2199 	 */
2200 	if (vcpu->arch.prev_cpu != pcpu) {
2201 		if (vcpu->arch.prev_cpu >= 0 &&
2202 		    cpu_first_thread_sibling(vcpu->arch.prev_cpu) !=
2203 		    cpu_first_thread_sibling(pcpu))
2204 			radix_flush_cpu(kvm, vcpu->arch.prev_cpu, vcpu);
2205 		vcpu->arch.prev_cpu = pcpu;
2206 	}
2207 }
2208 
2209 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2210 {
2211 	int cpu;
2212 	struct paca_struct *tpaca;
2213 	struct kvm *kvm = vc->kvm;
2214 
2215 	cpu = vc->pcpu;
2216 	if (vcpu) {
2217 		if (vcpu->arch.timer_running) {
2218 			hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2219 			vcpu->arch.timer_running = 0;
2220 		}
2221 		cpu += vcpu->arch.ptid;
2222 		vcpu->cpu = vc->pcpu;
2223 		vcpu->arch.thread_cpu = cpu;
2224 		cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2225 	}
2226 	tpaca = &paca[cpu];
2227 	tpaca->kvm_hstate.kvm_vcpu = vcpu;
2228 	tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2229 	/* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2230 	smp_wmb();
2231 	tpaca->kvm_hstate.kvm_vcore = vc;
2232 	if (cpu != smp_processor_id())
2233 		kvmppc_ipi_thread(cpu);
2234 }
2235 
2236 static void kvmppc_wait_for_nap(int n_threads)
2237 {
2238 	int cpu = smp_processor_id();
2239 	int i, loops;
2240 
2241 	if (n_threads <= 1)
2242 		return;
2243 	for (loops = 0; loops < 1000000; ++loops) {
2244 		/*
2245 		 * Check if all threads are finished.
2246 		 * We set the vcore pointer when starting a thread
2247 		 * and the thread clears it when finished, so we look
2248 		 * for any threads that still have a non-NULL vcore ptr.
2249 		 */
2250 		for (i = 1; i < n_threads; ++i)
2251 			if (paca[cpu + i].kvm_hstate.kvm_vcore)
2252 				break;
2253 		if (i == n_threads) {
2254 			HMT_medium();
2255 			return;
2256 		}
2257 		HMT_low();
2258 	}
2259 	HMT_medium();
2260 	for (i = 1; i < n_threads; ++i)
2261 		if (paca[cpu + i].kvm_hstate.kvm_vcore)
2262 			pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2263 }
2264 
2265 /*
2266  * Check that we are on thread 0 and that any other threads in
2267  * this core are off-line.  Then grab the threads so they can't
2268  * enter the kernel.
2269  */
2270 static int on_primary_thread(void)
2271 {
2272 	int cpu = smp_processor_id();
2273 	int thr;
2274 
2275 	/* Are we on a primary subcore? */
2276 	if (cpu_thread_in_subcore(cpu))
2277 		return 0;
2278 
2279 	thr = 0;
2280 	while (++thr < threads_per_subcore)
2281 		if (cpu_online(cpu + thr))
2282 			return 0;
2283 
2284 	/* Grab all hw threads so they can't go into the kernel */
2285 	for (thr = 1; thr < threads_per_subcore; ++thr) {
2286 		if (kvmppc_grab_hwthread(cpu + thr)) {
2287 			/* Couldn't grab one; let the others go */
2288 			do {
2289 				kvmppc_release_hwthread(cpu + thr);
2290 			} while (--thr > 0);
2291 			return 0;
2292 		}
2293 	}
2294 	return 1;
2295 }
2296 
2297 /*
2298  * A list of virtual cores for each physical CPU.
2299  * These are vcores that could run but their runner VCPU tasks are
2300  * (or may be) preempted.
2301  */
2302 struct preempted_vcore_list {
2303 	struct list_head	list;
2304 	spinlock_t		lock;
2305 };
2306 
2307 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2308 
2309 static void init_vcore_lists(void)
2310 {
2311 	int cpu;
2312 
2313 	for_each_possible_cpu(cpu) {
2314 		struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2315 		spin_lock_init(&lp->lock);
2316 		INIT_LIST_HEAD(&lp->list);
2317 	}
2318 }
2319 
2320 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2321 {
2322 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2323 
2324 	vc->vcore_state = VCORE_PREEMPT;
2325 	vc->pcpu = smp_processor_id();
2326 	if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2327 		spin_lock(&lp->lock);
2328 		list_add_tail(&vc->preempt_list, &lp->list);
2329 		spin_unlock(&lp->lock);
2330 	}
2331 
2332 	/* Start accumulating stolen time */
2333 	kvmppc_core_start_stolen(vc);
2334 }
2335 
2336 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2337 {
2338 	struct preempted_vcore_list *lp;
2339 
2340 	kvmppc_core_end_stolen(vc);
2341 	if (!list_empty(&vc->preempt_list)) {
2342 		lp = &per_cpu(preempted_vcores, vc->pcpu);
2343 		spin_lock(&lp->lock);
2344 		list_del_init(&vc->preempt_list);
2345 		spin_unlock(&lp->lock);
2346 	}
2347 	vc->vcore_state = VCORE_INACTIVE;
2348 }
2349 
2350 /*
2351  * This stores information about the virtual cores currently
2352  * assigned to a physical core.
2353  */
2354 struct core_info {
2355 	int		n_subcores;
2356 	int		max_subcore_threads;
2357 	int		total_threads;
2358 	int		subcore_threads[MAX_SUBCORES];
2359 	struct kvmppc_vcore *vc[MAX_SUBCORES];
2360 };
2361 
2362 /*
2363  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2364  * respectively in 2-way micro-threading (split-core) mode on POWER8.
2365  */
2366 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2367 
2368 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2369 {
2370 	memset(cip, 0, sizeof(*cip));
2371 	cip->n_subcores = 1;
2372 	cip->max_subcore_threads = vc->num_threads;
2373 	cip->total_threads = vc->num_threads;
2374 	cip->subcore_threads[0] = vc->num_threads;
2375 	cip->vc[0] = vc;
2376 }
2377 
2378 static bool subcore_config_ok(int n_subcores, int n_threads)
2379 {
2380 	/*
2381 	 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way split-core
2382 	 * mode, with one thread per subcore.
2383 	 */
2384 	if (cpu_has_feature(CPU_FTR_ARCH_300))
2385 		return n_subcores <= 4 && n_threads == 1;
2386 
2387 	/* On POWER8, can only dynamically split if unsplit to begin with */
2388 	if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2389 		return false;
2390 	if (n_subcores > MAX_SUBCORES)
2391 		return false;
2392 	if (n_subcores > 1) {
2393 		if (!(dynamic_mt_modes & 2))
2394 			n_subcores = 4;
2395 		if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2396 			return false;
2397 	}
2398 
2399 	return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2400 }
2401 
2402 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2403 {
2404 	vc->entry_exit_map = 0;
2405 	vc->in_guest = 0;
2406 	vc->napping_threads = 0;
2407 	vc->conferring_threads = 0;
2408 }
2409 
2410 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2411 {
2412 	int n_threads = vc->num_threads;
2413 	int sub;
2414 
2415 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2416 		return false;
2417 
2418 	/* POWER9 currently requires all threads to be in the same MMU mode */
2419 	if (cpu_has_feature(CPU_FTR_ARCH_300) &&
2420 	    kvm_is_radix(vc->kvm) != kvm_is_radix(cip->vc[0]->kvm))
2421 		return false;
2422 
2423 	if (n_threads < cip->max_subcore_threads)
2424 		n_threads = cip->max_subcore_threads;
2425 	if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2426 		return false;
2427 	cip->max_subcore_threads = n_threads;
2428 
2429 	sub = cip->n_subcores;
2430 	++cip->n_subcores;
2431 	cip->total_threads += vc->num_threads;
2432 	cip->subcore_threads[sub] = vc->num_threads;
2433 	cip->vc[sub] = vc;
2434 	init_vcore_to_run(vc);
2435 	list_del_init(&vc->preempt_list);
2436 
2437 	return true;
2438 }
2439 
2440 /*
2441  * Work out whether it is possible to piggyback the execution of
2442  * vcore *pvc onto the execution of the other vcores described in *cip.
2443  */
2444 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2445 			  int target_threads)
2446 {
2447 	if (cip->total_threads + pvc->num_threads > target_threads)
2448 		return false;
2449 
2450 	return can_dynamic_split(pvc, cip);
2451 }
2452 
2453 static void prepare_threads(struct kvmppc_vcore *vc)
2454 {
2455 	int i;
2456 	struct kvm_vcpu *vcpu;
2457 
2458 	for_each_runnable_thread(i, vcpu, vc) {
2459 		if (signal_pending(vcpu->arch.run_task))
2460 			vcpu->arch.ret = -EINTR;
2461 		else if (vcpu->arch.vpa.update_pending ||
2462 			 vcpu->arch.slb_shadow.update_pending ||
2463 			 vcpu->arch.dtl.update_pending)
2464 			vcpu->arch.ret = RESUME_GUEST;
2465 		else
2466 			continue;
2467 		kvmppc_remove_runnable(vc, vcpu);
2468 		wake_up(&vcpu->arch.cpu_run);
2469 	}
2470 }
2471 
2472 static void collect_piggybacks(struct core_info *cip, int target_threads)
2473 {
2474 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2475 	struct kvmppc_vcore *pvc, *vcnext;
2476 
2477 	spin_lock(&lp->lock);
2478 	list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2479 		if (!spin_trylock(&pvc->lock))
2480 			continue;
2481 		prepare_threads(pvc);
2482 		if (!pvc->n_runnable) {
2483 			list_del_init(&pvc->preempt_list);
2484 			if (pvc->runner == NULL) {
2485 				pvc->vcore_state = VCORE_INACTIVE;
2486 				kvmppc_core_end_stolen(pvc);
2487 			}
2488 			spin_unlock(&pvc->lock);
2489 			continue;
2490 		}
2491 		if (!can_piggyback(pvc, cip, target_threads)) {
2492 			spin_unlock(&pvc->lock);
2493 			continue;
2494 		}
2495 		kvmppc_core_end_stolen(pvc);
2496 		pvc->vcore_state = VCORE_PIGGYBACK;
2497 		if (cip->total_threads >= target_threads)
2498 			break;
2499 	}
2500 	spin_unlock(&lp->lock);
2501 }
2502 
2503 static bool recheck_signals(struct core_info *cip)
2504 {
2505 	int sub, i;
2506 	struct kvm_vcpu *vcpu;
2507 
2508 	for (sub = 0; sub < cip->n_subcores; ++sub)
2509 		for_each_runnable_thread(i, vcpu, cip->vc[sub])
2510 			if (signal_pending(vcpu->arch.run_task))
2511 				return true;
2512 	return false;
2513 }
2514 
2515 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2516 {
2517 	int still_running = 0, i;
2518 	u64 now;
2519 	long ret;
2520 	struct kvm_vcpu *vcpu;
2521 
2522 	spin_lock(&vc->lock);
2523 	now = get_tb();
2524 	for_each_runnable_thread(i, vcpu, vc) {
2525 		/* cancel pending dec exception if dec is positive */
2526 		if (now < vcpu->arch.dec_expires &&
2527 		    kvmppc_core_pending_dec(vcpu))
2528 			kvmppc_core_dequeue_dec(vcpu);
2529 
2530 		trace_kvm_guest_exit(vcpu);
2531 
2532 		ret = RESUME_GUEST;
2533 		if (vcpu->arch.trap)
2534 			ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2535 						    vcpu->arch.run_task);
2536 
2537 		vcpu->arch.ret = ret;
2538 		vcpu->arch.trap = 0;
2539 
2540 		if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2541 			if (vcpu->arch.pending_exceptions)
2542 				kvmppc_core_prepare_to_enter(vcpu);
2543 			if (vcpu->arch.ceded)
2544 				kvmppc_set_timer(vcpu);
2545 			else
2546 				++still_running;
2547 		} else {
2548 			kvmppc_remove_runnable(vc, vcpu);
2549 			wake_up(&vcpu->arch.cpu_run);
2550 		}
2551 	}
2552 	if (!is_master) {
2553 		if (still_running > 0) {
2554 			kvmppc_vcore_preempt(vc);
2555 		} else if (vc->runner) {
2556 			vc->vcore_state = VCORE_PREEMPT;
2557 			kvmppc_core_start_stolen(vc);
2558 		} else {
2559 			vc->vcore_state = VCORE_INACTIVE;
2560 		}
2561 		if (vc->n_runnable > 0 && vc->runner == NULL) {
2562 			/* make sure there's a candidate runner awake */
2563 			i = -1;
2564 			vcpu = next_runnable_thread(vc, &i);
2565 			wake_up(&vcpu->arch.cpu_run);
2566 		}
2567 	}
2568 	spin_unlock(&vc->lock);
2569 }
2570 
2571 /*
2572  * Clear core from the list of active host cores as we are about to
2573  * enter the guest. Only do this if it is the primary thread of the
2574  * core (not if a subcore) that is entering the guest.
2575  */
2576 static inline int kvmppc_clear_host_core(unsigned int cpu)
2577 {
2578 	int core;
2579 
2580 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2581 		return 0;
2582 	/*
2583 	 * Memory barrier can be omitted here as we will do a smp_wmb()
2584 	 * later in kvmppc_start_thread and we need ensure that state is
2585 	 * visible to other CPUs only after we enter guest.
2586 	 */
2587 	core = cpu >> threads_shift;
2588 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2589 	return 0;
2590 }
2591 
2592 /*
2593  * Advertise this core as an active host core since we exited the guest
2594  * Only need to do this if it is the primary thread of the core that is
2595  * exiting.
2596  */
2597 static inline int kvmppc_set_host_core(unsigned int cpu)
2598 {
2599 	int core;
2600 
2601 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2602 		return 0;
2603 
2604 	/*
2605 	 * Memory barrier can be omitted here because we do a spin_unlock
2606 	 * immediately after this which provides the memory barrier.
2607 	 */
2608 	core = cpu >> threads_shift;
2609 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2610 	return 0;
2611 }
2612 
2613 static void set_irq_happened(int trap)
2614 {
2615 	switch (trap) {
2616 	case BOOK3S_INTERRUPT_EXTERNAL:
2617 		local_paca->irq_happened |= PACA_IRQ_EE;
2618 		break;
2619 	case BOOK3S_INTERRUPT_H_DOORBELL:
2620 		local_paca->irq_happened |= PACA_IRQ_DBELL;
2621 		break;
2622 	case BOOK3S_INTERRUPT_HMI:
2623 		local_paca->irq_happened |= PACA_IRQ_HMI;
2624 		break;
2625 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
2626 		replay_system_reset();
2627 		break;
2628 	}
2629 }
2630 
2631 /*
2632  * Run a set of guest threads on a physical core.
2633  * Called with vc->lock held.
2634  */
2635 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2636 {
2637 	struct kvm_vcpu *vcpu;
2638 	int i;
2639 	int srcu_idx;
2640 	struct core_info core_info;
2641 	struct kvmppc_vcore *pvc;
2642 	struct kvm_split_mode split_info, *sip;
2643 	int split, subcore_size, active;
2644 	int sub;
2645 	bool thr0_done;
2646 	unsigned long cmd_bit, stat_bit;
2647 	int pcpu, thr;
2648 	int target_threads;
2649 	int controlled_threads;
2650 	int trap;
2651 	bool is_power8;
2652 	bool hpt_on_radix;
2653 
2654 	/*
2655 	 * Remove from the list any threads that have a signal pending
2656 	 * or need a VPA update done
2657 	 */
2658 	prepare_threads(vc);
2659 
2660 	/* if the runner is no longer runnable, let the caller pick a new one */
2661 	if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2662 		return;
2663 
2664 	/*
2665 	 * Initialize *vc.
2666 	 */
2667 	init_vcore_to_run(vc);
2668 	vc->preempt_tb = TB_NIL;
2669 
2670 	/*
2671 	 * Number of threads that we will be controlling: the same as
2672 	 * the number of threads per subcore, except on POWER9,
2673 	 * where it's 1 because the threads are (mostly) independent.
2674 	 */
2675 	controlled_threads = threads_per_vcore(vc->kvm);
2676 
2677 	/*
2678 	 * Make sure we are running on primary threads, and that secondary
2679 	 * threads are offline.  Also check if the number of threads in this
2680 	 * guest are greater than the current system threads per guest.
2681 	 * On POWER9, we need to be not in independent-threads mode if
2682 	 * this is a HPT guest on a radix host.
2683 	 */
2684 	hpt_on_radix = radix_enabled() && !kvm_is_radix(vc->kvm);
2685 	if (((controlled_threads > 1) &&
2686 	     ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) ||
2687 	    (hpt_on_radix && vc->kvm->arch.threads_indep)) {
2688 		for_each_runnable_thread(i, vcpu, vc) {
2689 			vcpu->arch.ret = -EBUSY;
2690 			kvmppc_remove_runnable(vc, vcpu);
2691 			wake_up(&vcpu->arch.cpu_run);
2692 		}
2693 		goto out;
2694 	}
2695 
2696 	/*
2697 	 * See if we could run any other vcores on the physical core
2698 	 * along with this one.
2699 	 */
2700 	init_core_info(&core_info, vc);
2701 	pcpu = smp_processor_id();
2702 	target_threads = controlled_threads;
2703 	if (target_smt_mode && target_smt_mode < target_threads)
2704 		target_threads = target_smt_mode;
2705 	if (vc->num_threads < target_threads)
2706 		collect_piggybacks(&core_info, target_threads);
2707 
2708 	/*
2709 	 * On radix, arrange for TLB flushing if necessary.
2710 	 * This has to be done before disabling interrupts since
2711 	 * it uses smp_call_function().
2712 	 */
2713 	pcpu = smp_processor_id();
2714 	if (kvm_is_radix(vc->kvm)) {
2715 		for (sub = 0; sub < core_info.n_subcores; ++sub)
2716 			for_each_runnable_thread(i, vcpu, core_info.vc[sub])
2717 				kvmppc_prepare_radix_vcpu(vcpu, pcpu);
2718 	}
2719 
2720 	/*
2721 	 * Hard-disable interrupts, and check resched flag and signals.
2722 	 * If we need to reschedule or deliver a signal, clean up
2723 	 * and return without going into the guest(s).
2724 	 * If the mmu_ready flag has been cleared, don't go into the
2725 	 * guest because that means a HPT resize operation is in progress.
2726 	 */
2727 	local_irq_disable();
2728 	hard_irq_disable();
2729 	if (lazy_irq_pending() || need_resched() ||
2730 	    recheck_signals(&core_info) || !vc->kvm->arch.mmu_ready) {
2731 		local_irq_enable();
2732 		vc->vcore_state = VCORE_INACTIVE;
2733 		/* Unlock all except the primary vcore */
2734 		for (sub = 1; sub < core_info.n_subcores; ++sub) {
2735 			pvc = core_info.vc[sub];
2736 			/* Put back on to the preempted vcores list */
2737 			kvmppc_vcore_preempt(pvc);
2738 			spin_unlock(&pvc->lock);
2739 		}
2740 		for (i = 0; i < controlled_threads; ++i)
2741 			kvmppc_release_hwthread(pcpu + i);
2742 		return;
2743 	}
2744 
2745 	kvmppc_clear_host_core(pcpu);
2746 
2747 	/* Decide on micro-threading (split-core) mode */
2748 	subcore_size = threads_per_subcore;
2749 	cmd_bit = stat_bit = 0;
2750 	split = core_info.n_subcores;
2751 	sip = NULL;
2752 	is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
2753 		&& !cpu_has_feature(CPU_FTR_ARCH_300);
2754 
2755 	if (split > 1 || hpt_on_radix) {
2756 		sip = &split_info;
2757 		memset(&split_info, 0, sizeof(split_info));
2758 		for (sub = 0; sub < core_info.n_subcores; ++sub)
2759 			split_info.vc[sub] = core_info.vc[sub];
2760 
2761 		if (is_power8) {
2762 			if (split == 2 && (dynamic_mt_modes & 2)) {
2763 				cmd_bit = HID0_POWER8_1TO2LPAR;
2764 				stat_bit = HID0_POWER8_2LPARMODE;
2765 			} else {
2766 				split = 4;
2767 				cmd_bit = HID0_POWER8_1TO4LPAR;
2768 				stat_bit = HID0_POWER8_4LPARMODE;
2769 			}
2770 			subcore_size = MAX_SMT_THREADS / split;
2771 			split_info.rpr = mfspr(SPRN_RPR);
2772 			split_info.pmmar = mfspr(SPRN_PMMAR);
2773 			split_info.ldbar = mfspr(SPRN_LDBAR);
2774 			split_info.subcore_size = subcore_size;
2775 		} else {
2776 			split_info.subcore_size = 1;
2777 			if (hpt_on_radix) {
2778 				/* Use the split_info for LPCR/LPIDR changes */
2779 				split_info.lpcr_req = vc->lpcr;
2780 				split_info.lpidr_req = vc->kvm->arch.lpid;
2781 				split_info.host_lpcr = vc->kvm->arch.host_lpcr;
2782 				split_info.do_set = 1;
2783 			}
2784 		}
2785 
2786 		/* order writes to split_info before kvm_split_mode pointer */
2787 		smp_wmb();
2788 	}
2789 
2790 	for (thr = 0; thr < controlled_threads; ++thr) {
2791 		paca[pcpu + thr].kvm_hstate.tid = thr;
2792 		paca[pcpu + thr].kvm_hstate.napping = 0;
2793 		paca[pcpu + thr].kvm_hstate.kvm_split_mode = sip;
2794 	}
2795 
2796 	/* Initiate micro-threading (split-core) on POWER8 if required */
2797 	if (cmd_bit) {
2798 		unsigned long hid0 = mfspr(SPRN_HID0);
2799 
2800 		hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2801 		mb();
2802 		mtspr(SPRN_HID0, hid0);
2803 		isync();
2804 		for (;;) {
2805 			hid0 = mfspr(SPRN_HID0);
2806 			if (hid0 & stat_bit)
2807 				break;
2808 			cpu_relax();
2809 		}
2810 	}
2811 
2812 	/* Start all the threads */
2813 	active = 0;
2814 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
2815 		thr = is_power8 ? subcore_thread_map[sub] : sub;
2816 		thr0_done = false;
2817 		active |= 1 << thr;
2818 		pvc = core_info.vc[sub];
2819 		pvc->pcpu = pcpu + thr;
2820 		for_each_runnable_thread(i, vcpu, pvc) {
2821 			kvmppc_start_thread(vcpu, pvc);
2822 			kvmppc_create_dtl_entry(vcpu, pvc);
2823 			trace_kvm_guest_enter(vcpu);
2824 			if (!vcpu->arch.ptid)
2825 				thr0_done = true;
2826 			active |= 1 << (thr + vcpu->arch.ptid);
2827 		}
2828 		/*
2829 		 * We need to start the first thread of each subcore
2830 		 * even if it doesn't have a vcpu.
2831 		 */
2832 		if (!thr0_done)
2833 			kvmppc_start_thread(NULL, pvc);
2834 		thr += pvc->num_threads;
2835 	}
2836 
2837 	/*
2838 	 * Ensure that split_info.do_nap is set after setting
2839 	 * the vcore pointer in the PACA of the secondaries.
2840 	 */
2841 	smp_mb();
2842 
2843 	/*
2844 	 * When doing micro-threading, poke the inactive threads as well.
2845 	 * This gets them to the nap instruction after kvm_do_nap,
2846 	 * which reduces the time taken to unsplit later.
2847 	 * For POWER9 HPT guest on radix host, we need all the secondary
2848 	 * threads woken up so they can do the LPCR/LPIDR change.
2849 	 */
2850 	if (cmd_bit || hpt_on_radix) {
2851 		split_info.do_nap = 1;	/* ask secondaries to nap when done */
2852 		for (thr = 1; thr < threads_per_subcore; ++thr)
2853 			if (!(active & (1 << thr)))
2854 				kvmppc_ipi_thread(pcpu + thr);
2855 	}
2856 
2857 	vc->vcore_state = VCORE_RUNNING;
2858 	preempt_disable();
2859 
2860 	trace_kvmppc_run_core(vc, 0);
2861 
2862 	for (sub = 0; sub < core_info.n_subcores; ++sub)
2863 		spin_unlock(&core_info.vc[sub]->lock);
2864 
2865 	/*
2866 	 * Interrupts will be enabled once we get into the guest,
2867 	 * so tell lockdep that we're about to enable interrupts.
2868 	 */
2869 	trace_hardirqs_on();
2870 
2871 	guest_enter();
2872 
2873 	srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2874 
2875 	trap = __kvmppc_vcore_entry();
2876 
2877 	srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2878 
2879 	guest_exit();
2880 
2881 	trace_hardirqs_off();
2882 	set_irq_happened(trap);
2883 
2884 	spin_lock(&vc->lock);
2885 	/* prevent other vcpu threads from doing kvmppc_start_thread() now */
2886 	vc->vcore_state = VCORE_EXITING;
2887 
2888 	/* wait for secondary threads to finish writing their state to memory */
2889 	kvmppc_wait_for_nap(controlled_threads);
2890 
2891 	/* Return to whole-core mode if we split the core earlier */
2892 	if (cmd_bit) {
2893 		unsigned long hid0 = mfspr(SPRN_HID0);
2894 		unsigned long loops = 0;
2895 
2896 		hid0 &= ~HID0_POWER8_DYNLPARDIS;
2897 		stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
2898 		mb();
2899 		mtspr(SPRN_HID0, hid0);
2900 		isync();
2901 		for (;;) {
2902 			hid0 = mfspr(SPRN_HID0);
2903 			if (!(hid0 & stat_bit))
2904 				break;
2905 			cpu_relax();
2906 			++loops;
2907 		}
2908 	} else if (hpt_on_radix) {
2909 		/* Wait for all threads to have seen final sync */
2910 		for (thr = 1; thr < controlled_threads; ++thr) {
2911 			while (paca[pcpu + thr].kvm_hstate.kvm_split_mode) {
2912 				HMT_low();
2913 				barrier();
2914 			}
2915 			HMT_medium();
2916 		}
2917 	}
2918 	split_info.do_nap = 0;
2919 
2920 	kvmppc_set_host_core(pcpu);
2921 
2922 	local_irq_enable();
2923 
2924 	/* Let secondaries go back to the offline loop */
2925 	for (i = 0; i < controlled_threads; ++i) {
2926 		kvmppc_release_hwthread(pcpu + i);
2927 		if (sip && sip->napped[i])
2928 			kvmppc_ipi_thread(pcpu + i);
2929 		cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
2930 	}
2931 
2932 	spin_unlock(&vc->lock);
2933 
2934 	/* make sure updates to secondary vcpu structs are visible now */
2935 	smp_mb();
2936 
2937 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
2938 		pvc = core_info.vc[sub];
2939 		post_guest_process(pvc, pvc == vc);
2940 	}
2941 
2942 	spin_lock(&vc->lock);
2943 	preempt_enable();
2944 
2945  out:
2946 	vc->vcore_state = VCORE_INACTIVE;
2947 	trace_kvmppc_run_core(vc, 1);
2948 }
2949 
2950 /*
2951  * Wait for some other vcpu thread to execute us, and
2952  * wake us up when we need to handle something in the host.
2953  */
2954 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
2955 				 struct kvm_vcpu *vcpu, int wait_state)
2956 {
2957 	DEFINE_WAIT(wait);
2958 
2959 	prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
2960 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2961 		spin_unlock(&vc->lock);
2962 		schedule();
2963 		spin_lock(&vc->lock);
2964 	}
2965 	finish_wait(&vcpu->arch.cpu_run, &wait);
2966 }
2967 
2968 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
2969 {
2970 	/* 10us base */
2971 	if (vc->halt_poll_ns == 0 && halt_poll_ns_grow)
2972 		vc->halt_poll_ns = 10000;
2973 	else
2974 		vc->halt_poll_ns *= halt_poll_ns_grow;
2975 }
2976 
2977 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
2978 {
2979 	if (halt_poll_ns_shrink == 0)
2980 		vc->halt_poll_ns = 0;
2981 	else
2982 		vc->halt_poll_ns /= halt_poll_ns_shrink;
2983 }
2984 
2985 #ifdef CONFIG_KVM_XICS
2986 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
2987 {
2988 	if (!xive_enabled())
2989 		return false;
2990 	return vcpu->arch.xive_saved_state.pipr <
2991 		vcpu->arch.xive_saved_state.cppr;
2992 }
2993 #else
2994 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
2995 {
2996 	return false;
2997 }
2998 #endif /* CONFIG_KVM_XICS */
2999 
3000 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3001 {
3002 	if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3003 	    kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3004 		return true;
3005 
3006 	return false;
3007 }
3008 
3009 /*
3010  * Check to see if any of the runnable vcpus on the vcore have pending
3011  * exceptions or are no longer ceded
3012  */
3013 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3014 {
3015 	struct kvm_vcpu *vcpu;
3016 	int i;
3017 
3018 	for_each_runnable_thread(i, vcpu, vc) {
3019 		if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3020 			return 1;
3021 	}
3022 
3023 	return 0;
3024 }
3025 
3026 /*
3027  * All the vcpus in this vcore are idle, so wait for a decrementer
3028  * or external interrupt to one of the vcpus.  vc->lock is held.
3029  */
3030 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3031 {
3032 	ktime_t cur, start_poll, start_wait;
3033 	int do_sleep = 1;
3034 	u64 block_ns;
3035 	DECLARE_SWAITQUEUE(wait);
3036 
3037 	/* Poll for pending exceptions and ceded state */
3038 	cur = start_poll = ktime_get();
3039 	if (vc->halt_poll_ns) {
3040 		ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3041 		++vc->runner->stat.halt_attempted_poll;
3042 
3043 		vc->vcore_state = VCORE_POLLING;
3044 		spin_unlock(&vc->lock);
3045 
3046 		do {
3047 			if (kvmppc_vcore_check_block(vc)) {
3048 				do_sleep = 0;
3049 				break;
3050 			}
3051 			cur = ktime_get();
3052 		} while (single_task_running() && ktime_before(cur, stop));
3053 
3054 		spin_lock(&vc->lock);
3055 		vc->vcore_state = VCORE_INACTIVE;
3056 
3057 		if (!do_sleep) {
3058 			++vc->runner->stat.halt_successful_poll;
3059 			goto out;
3060 		}
3061 	}
3062 
3063 	prepare_to_swait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
3064 
3065 	if (kvmppc_vcore_check_block(vc)) {
3066 		finish_swait(&vc->wq, &wait);
3067 		do_sleep = 0;
3068 		/* If we polled, count this as a successful poll */
3069 		if (vc->halt_poll_ns)
3070 			++vc->runner->stat.halt_successful_poll;
3071 		goto out;
3072 	}
3073 
3074 	start_wait = ktime_get();
3075 
3076 	vc->vcore_state = VCORE_SLEEPING;
3077 	trace_kvmppc_vcore_blocked(vc, 0);
3078 	spin_unlock(&vc->lock);
3079 	schedule();
3080 	finish_swait(&vc->wq, &wait);
3081 	spin_lock(&vc->lock);
3082 	vc->vcore_state = VCORE_INACTIVE;
3083 	trace_kvmppc_vcore_blocked(vc, 1);
3084 	++vc->runner->stat.halt_successful_wait;
3085 
3086 	cur = ktime_get();
3087 
3088 out:
3089 	block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3090 
3091 	/* Attribute wait time */
3092 	if (do_sleep) {
3093 		vc->runner->stat.halt_wait_ns +=
3094 			ktime_to_ns(cur) - ktime_to_ns(start_wait);
3095 		/* Attribute failed poll time */
3096 		if (vc->halt_poll_ns)
3097 			vc->runner->stat.halt_poll_fail_ns +=
3098 				ktime_to_ns(start_wait) -
3099 				ktime_to_ns(start_poll);
3100 	} else {
3101 		/* Attribute successful poll time */
3102 		if (vc->halt_poll_ns)
3103 			vc->runner->stat.halt_poll_success_ns +=
3104 				ktime_to_ns(cur) -
3105 				ktime_to_ns(start_poll);
3106 	}
3107 
3108 	/* Adjust poll time */
3109 	if (halt_poll_ns) {
3110 		if (block_ns <= vc->halt_poll_ns)
3111 			;
3112 		/* We slept and blocked for longer than the max halt time */
3113 		else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3114 			shrink_halt_poll_ns(vc);
3115 		/* We slept and our poll time is too small */
3116 		else if (vc->halt_poll_ns < halt_poll_ns &&
3117 				block_ns < halt_poll_ns)
3118 			grow_halt_poll_ns(vc);
3119 		if (vc->halt_poll_ns > halt_poll_ns)
3120 			vc->halt_poll_ns = halt_poll_ns;
3121 	} else
3122 		vc->halt_poll_ns = 0;
3123 
3124 	trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3125 }
3126 
3127 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
3128 {
3129 	int r = 0;
3130 	struct kvm *kvm = vcpu->kvm;
3131 
3132 	mutex_lock(&kvm->lock);
3133 	if (!kvm->arch.mmu_ready) {
3134 		if (!kvm_is_radix(kvm))
3135 			r = kvmppc_hv_setup_htab_rma(vcpu);
3136 		if (!r) {
3137 			if (cpu_has_feature(CPU_FTR_ARCH_300))
3138 				kvmppc_setup_partition_table(kvm);
3139 			kvm->arch.mmu_ready = 1;
3140 		}
3141 	}
3142 	mutex_unlock(&kvm->lock);
3143 	return r;
3144 }
3145 
3146 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3147 {
3148 	int n_ceded, i, r;
3149 	struct kvmppc_vcore *vc;
3150 	struct kvm_vcpu *v;
3151 
3152 	trace_kvmppc_run_vcpu_enter(vcpu);
3153 
3154 	kvm_run->exit_reason = 0;
3155 	vcpu->arch.ret = RESUME_GUEST;
3156 	vcpu->arch.trap = 0;
3157 	kvmppc_update_vpas(vcpu);
3158 
3159 	/*
3160 	 * Synchronize with other threads in this virtual core
3161 	 */
3162 	vc = vcpu->arch.vcore;
3163 	spin_lock(&vc->lock);
3164 	vcpu->arch.ceded = 0;
3165 	vcpu->arch.run_task = current;
3166 	vcpu->arch.kvm_run = kvm_run;
3167 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3168 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3169 	vcpu->arch.busy_preempt = TB_NIL;
3170 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3171 	++vc->n_runnable;
3172 
3173 	/*
3174 	 * This happens the first time this is called for a vcpu.
3175 	 * If the vcore is already running, we may be able to start
3176 	 * this thread straight away and have it join in.
3177 	 */
3178 	if (!signal_pending(current)) {
3179 		if (vc->vcore_state == VCORE_PIGGYBACK) {
3180 			if (spin_trylock(&vc->lock)) {
3181 				if (vc->vcore_state == VCORE_RUNNING &&
3182 				    !VCORE_IS_EXITING(vc)) {
3183 					kvmppc_create_dtl_entry(vcpu, vc);
3184 					kvmppc_start_thread(vcpu, vc);
3185 					trace_kvm_guest_enter(vcpu);
3186 				}
3187 				spin_unlock(&vc->lock);
3188 			}
3189 		} else if (vc->vcore_state == VCORE_RUNNING &&
3190 			   !VCORE_IS_EXITING(vc)) {
3191 			kvmppc_create_dtl_entry(vcpu, vc);
3192 			kvmppc_start_thread(vcpu, vc);
3193 			trace_kvm_guest_enter(vcpu);
3194 		} else if (vc->vcore_state == VCORE_SLEEPING) {
3195 			swake_up(&vc->wq);
3196 		}
3197 
3198 	}
3199 
3200 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3201 	       !signal_pending(current)) {
3202 		/* See if the MMU is ready to go */
3203 		if (!vcpu->kvm->arch.mmu_ready) {
3204 			spin_unlock(&vc->lock);
3205 			r = kvmhv_setup_mmu(vcpu);
3206 			spin_lock(&vc->lock);
3207 			if (r) {
3208 				kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3209 				kvm_run->fail_entry.
3210 					hardware_entry_failure_reason = 0;
3211 				vcpu->arch.ret = r;
3212 				break;
3213 			}
3214 		}
3215 
3216 		if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3217 			kvmppc_vcore_end_preempt(vc);
3218 
3219 		if (vc->vcore_state != VCORE_INACTIVE) {
3220 			kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
3221 			continue;
3222 		}
3223 		for_each_runnable_thread(i, v, vc) {
3224 			kvmppc_core_prepare_to_enter(v);
3225 			if (signal_pending(v->arch.run_task)) {
3226 				kvmppc_remove_runnable(vc, v);
3227 				v->stat.signal_exits++;
3228 				v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
3229 				v->arch.ret = -EINTR;
3230 				wake_up(&v->arch.cpu_run);
3231 			}
3232 		}
3233 		if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3234 			break;
3235 		n_ceded = 0;
3236 		for_each_runnable_thread(i, v, vc) {
3237 			if (!kvmppc_vcpu_woken(v))
3238 				n_ceded += v->arch.ceded;
3239 			else
3240 				v->arch.ceded = 0;
3241 		}
3242 		vc->runner = vcpu;
3243 		if (n_ceded == vc->n_runnable) {
3244 			kvmppc_vcore_blocked(vc);
3245 		} else if (need_resched()) {
3246 			kvmppc_vcore_preempt(vc);
3247 			/* Let something else run */
3248 			cond_resched_lock(&vc->lock);
3249 			if (vc->vcore_state == VCORE_PREEMPT)
3250 				kvmppc_vcore_end_preempt(vc);
3251 		} else {
3252 			kvmppc_run_core(vc);
3253 		}
3254 		vc->runner = NULL;
3255 	}
3256 
3257 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3258 	       (vc->vcore_state == VCORE_RUNNING ||
3259 		vc->vcore_state == VCORE_EXITING ||
3260 		vc->vcore_state == VCORE_PIGGYBACK))
3261 		kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
3262 
3263 	if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3264 		kvmppc_vcore_end_preempt(vc);
3265 
3266 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3267 		kvmppc_remove_runnable(vc, vcpu);
3268 		vcpu->stat.signal_exits++;
3269 		kvm_run->exit_reason = KVM_EXIT_INTR;
3270 		vcpu->arch.ret = -EINTR;
3271 	}
3272 
3273 	if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
3274 		/* Wake up some vcpu to run the core */
3275 		i = -1;
3276 		v = next_runnable_thread(vc, &i);
3277 		wake_up(&v->arch.cpu_run);
3278 	}
3279 
3280 	trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
3281 	spin_unlock(&vc->lock);
3282 	return vcpu->arch.ret;
3283 }
3284 
3285 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
3286 {
3287 	int r;
3288 	int srcu_idx;
3289 	unsigned long ebb_regs[3] = {};	/* shut up GCC */
3290 	unsigned long user_tar = 0;
3291 	unsigned int user_vrsave;
3292 	struct kvm *kvm;
3293 
3294 	if (!vcpu->arch.sane) {
3295 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3296 		return -EINVAL;
3297 	}
3298 
3299 	/*
3300 	 * Don't allow entry with a suspended transaction, because
3301 	 * the guest entry/exit code will lose it.
3302 	 * If the guest has TM enabled, save away their TM-related SPRs
3303 	 * (they will get restored by the TM unavailable interrupt).
3304 	 */
3305 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
3306 	if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
3307 	    (current->thread.regs->msr & MSR_TM)) {
3308 		if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
3309 			run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3310 			run->fail_entry.hardware_entry_failure_reason = 0;
3311 			return -EINVAL;
3312 		}
3313 		/* Enable TM so we can read the TM SPRs */
3314 		mtmsr(mfmsr() | MSR_TM);
3315 		current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
3316 		current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
3317 		current->thread.tm_texasr = mfspr(SPRN_TEXASR);
3318 		current->thread.regs->msr &= ~MSR_TM;
3319 	}
3320 #endif
3321 
3322 	kvmppc_core_prepare_to_enter(vcpu);
3323 
3324 	/* No need to go into the guest when all we'll do is come back out */
3325 	if (signal_pending(current)) {
3326 		run->exit_reason = KVM_EXIT_INTR;
3327 		return -EINTR;
3328 	}
3329 
3330 	kvm = vcpu->kvm;
3331 	atomic_inc(&kvm->arch.vcpus_running);
3332 	/* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
3333 	smp_mb();
3334 
3335 	flush_all_to_thread(current);
3336 
3337 	/* Save userspace EBB and other register values */
3338 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
3339 		ebb_regs[0] = mfspr(SPRN_EBBHR);
3340 		ebb_regs[1] = mfspr(SPRN_EBBRR);
3341 		ebb_regs[2] = mfspr(SPRN_BESCR);
3342 		user_tar = mfspr(SPRN_TAR);
3343 	}
3344 	user_vrsave = mfspr(SPRN_VRSAVE);
3345 
3346 	vcpu->arch.wqp = &vcpu->arch.vcore->wq;
3347 	vcpu->arch.pgdir = current->mm->pgd;
3348 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
3349 
3350 	do {
3351 		r = kvmppc_run_vcpu(run, vcpu);
3352 
3353 		if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
3354 		    !(vcpu->arch.shregs.msr & MSR_PR)) {
3355 			trace_kvm_hcall_enter(vcpu);
3356 			r = kvmppc_pseries_do_hcall(vcpu);
3357 			trace_kvm_hcall_exit(vcpu, r);
3358 			kvmppc_core_prepare_to_enter(vcpu);
3359 		} else if (r == RESUME_PAGE_FAULT) {
3360 			srcu_idx = srcu_read_lock(&kvm->srcu);
3361 			r = kvmppc_book3s_hv_page_fault(run, vcpu,
3362 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
3363 			srcu_read_unlock(&kvm->srcu, srcu_idx);
3364 		} else if (r == RESUME_PASSTHROUGH) {
3365 			if (WARN_ON(xive_enabled()))
3366 				r = H_SUCCESS;
3367 			else
3368 				r = kvmppc_xics_rm_complete(vcpu, 0);
3369 		}
3370 	} while (is_kvmppc_resume_guest(r));
3371 
3372 	/* Restore userspace EBB and other register values */
3373 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
3374 		mtspr(SPRN_EBBHR, ebb_regs[0]);
3375 		mtspr(SPRN_EBBRR, ebb_regs[1]);
3376 		mtspr(SPRN_BESCR, ebb_regs[2]);
3377 		mtspr(SPRN_TAR, user_tar);
3378 		mtspr(SPRN_FSCR, current->thread.fscr);
3379 	}
3380 	mtspr(SPRN_VRSAVE, user_vrsave);
3381 
3382 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
3383 	atomic_dec(&kvm->arch.vcpus_running);
3384 	return r;
3385 }
3386 
3387 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
3388 				     int shift, int sllp)
3389 {
3390 	(*sps)->page_shift = shift;
3391 	(*sps)->slb_enc = sllp;
3392 	(*sps)->enc[0].page_shift = shift;
3393 	(*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
3394 	/*
3395 	 * Add 16MB MPSS support (may get filtered out by userspace)
3396 	 */
3397 	if (shift != 24) {
3398 		int penc = kvmppc_pgsize_lp_encoding(shift, 24);
3399 		if (penc != -1) {
3400 			(*sps)->enc[1].page_shift = 24;
3401 			(*sps)->enc[1].pte_enc = penc;
3402 		}
3403 	}
3404 	(*sps)++;
3405 }
3406 
3407 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
3408 					 struct kvm_ppc_smmu_info *info)
3409 {
3410 	struct kvm_ppc_one_seg_page_size *sps;
3411 
3412 	/*
3413 	 * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
3414 	 * POWER7 doesn't support keys for instruction accesses,
3415 	 * POWER8 and POWER9 do.
3416 	 */
3417 	info->data_keys = 32;
3418 	info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
3419 
3420 	/* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
3421 	info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
3422 	info->slb_size = 32;
3423 
3424 	/* We only support these sizes for now, and no muti-size segments */
3425 	sps = &info->sps[0];
3426 	kvmppc_add_seg_page_size(&sps, 12, 0);
3427 	kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
3428 	kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
3429 
3430 	return 0;
3431 }
3432 
3433 /*
3434  * Get (and clear) the dirty memory log for a memory slot.
3435  */
3436 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
3437 					 struct kvm_dirty_log *log)
3438 {
3439 	struct kvm_memslots *slots;
3440 	struct kvm_memory_slot *memslot;
3441 	int i, r;
3442 	unsigned long n;
3443 	unsigned long *buf, *p;
3444 	struct kvm_vcpu *vcpu;
3445 
3446 	mutex_lock(&kvm->slots_lock);
3447 
3448 	r = -EINVAL;
3449 	if (log->slot >= KVM_USER_MEM_SLOTS)
3450 		goto out;
3451 
3452 	slots = kvm_memslots(kvm);
3453 	memslot = id_to_memslot(slots, log->slot);
3454 	r = -ENOENT;
3455 	if (!memslot->dirty_bitmap)
3456 		goto out;
3457 
3458 	/*
3459 	 * Use second half of bitmap area because both HPT and radix
3460 	 * accumulate bits in the first half.
3461 	 */
3462 	n = kvm_dirty_bitmap_bytes(memslot);
3463 	buf = memslot->dirty_bitmap + n / sizeof(long);
3464 	memset(buf, 0, n);
3465 
3466 	if (kvm_is_radix(kvm))
3467 		r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
3468 	else
3469 		r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
3470 	if (r)
3471 		goto out;
3472 
3473 	/*
3474 	 * We accumulate dirty bits in the first half of the
3475 	 * memslot's dirty_bitmap area, for when pages are paged
3476 	 * out or modified by the host directly.  Pick up these
3477 	 * bits and add them to the map.
3478 	 */
3479 	p = memslot->dirty_bitmap;
3480 	for (i = 0; i < n / sizeof(long); ++i)
3481 		buf[i] |= xchg(&p[i], 0);
3482 
3483 	/* Harvest dirty bits from VPA and DTL updates */
3484 	/* Note: we never modify the SLB shadow buffer areas */
3485 	kvm_for_each_vcpu(i, vcpu, kvm) {
3486 		spin_lock(&vcpu->arch.vpa_update_lock);
3487 		kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
3488 		kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
3489 		spin_unlock(&vcpu->arch.vpa_update_lock);
3490 	}
3491 
3492 	r = -EFAULT;
3493 	if (copy_to_user(log->dirty_bitmap, buf, n))
3494 		goto out;
3495 
3496 	r = 0;
3497 out:
3498 	mutex_unlock(&kvm->slots_lock);
3499 	return r;
3500 }
3501 
3502 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
3503 					struct kvm_memory_slot *dont)
3504 {
3505 	if (!dont || free->arch.rmap != dont->arch.rmap) {
3506 		vfree(free->arch.rmap);
3507 		free->arch.rmap = NULL;
3508 	}
3509 }
3510 
3511 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
3512 					 unsigned long npages)
3513 {
3514 	slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
3515 	if (!slot->arch.rmap)
3516 		return -ENOMEM;
3517 
3518 	return 0;
3519 }
3520 
3521 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
3522 					struct kvm_memory_slot *memslot,
3523 					const struct kvm_userspace_memory_region *mem)
3524 {
3525 	return 0;
3526 }
3527 
3528 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
3529 				const struct kvm_userspace_memory_region *mem,
3530 				const struct kvm_memory_slot *old,
3531 				const struct kvm_memory_slot *new)
3532 {
3533 	unsigned long npages = mem->memory_size >> PAGE_SHIFT;
3534 
3535 	/*
3536 	 * If we are making a new memslot, it might make
3537 	 * some address that was previously cached as emulated
3538 	 * MMIO be no longer emulated MMIO, so invalidate
3539 	 * all the caches of emulated MMIO translations.
3540 	 */
3541 	if (npages)
3542 		atomic64_inc(&kvm->arch.mmio_update);
3543 }
3544 
3545 /*
3546  * Update LPCR values in kvm->arch and in vcores.
3547  * Caller must hold kvm->lock.
3548  */
3549 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
3550 {
3551 	long int i;
3552 	u32 cores_done = 0;
3553 
3554 	if ((kvm->arch.lpcr & mask) == lpcr)
3555 		return;
3556 
3557 	kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
3558 
3559 	for (i = 0; i < KVM_MAX_VCORES; ++i) {
3560 		struct kvmppc_vcore *vc = kvm->arch.vcores[i];
3561 		if (!vc)
3562 			continue;
3563 		spin_lock(&vc->lock);
3564 		vc->lpcr = (vc->lpcr & ~mask) | lpcr;
3565 		spin_unlock(&vc->lock);
3566 		if (++cores_done >= kvm->arch.online_vcores)
3567 			break;
3568 	}
3569 }
3570 
3571 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
3572 {
3573 	return;
3574 }
3575 
3576 void kvmppc_setup_partition_table(struct kvm *kvm)
3577 {
3578 	unsigned long dw0, dw1;
3579 
3580 	if (!kvm_is_radix(kvm)) {
3581 		/* PS field - page size for VRMA */
3582 		dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
3583 			((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
3584 		/* HTABSIZE and HTABORG fields */
3585 		dw0 |= kvm->arch.sdr1;
3586 
3587 		/* Second dword as set by userspace */
3588 		dw1 = kvm->arch.process_table;
3589 	} else {
3590 		dw0 = PATB_HR | radix__get_tree_size() |
3591 			__pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
3592 		dw1 = PATB_GR | kvm->arch.process_table;
3593 	}
3594 
3595 	mmu_partition_table_set_entry(kvm->arch.lpid, dw0, dw1);
3596 }
3597 
3598 /*
3599  * Set up HPT (hashed page table) and RMA (real-mode area).
3600  * Must be called with kvm->lock held.
3601  */
3602 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
3603 {
3604 	int err = 0;
3605 	struct kvm *kvm = vcpu->kvm;
3606 	unsigned long hva;
3607 	struct kvm_memory_slot *memslot;
3608 	struct vm_area_struct *vma;
3609 	unsigned long lpcr = 0, senc;
3610 	unsigned long psize, porder;
3611 	int srcu_idx;
3612 
3613 	/* Allocate hashed page table (if not done already) and reset it */
3614 	if (!kvm->arch.hpt.virt) {
3615 		int order = KVM_DEFAULT_HPT_ORDER;
3616 		struct kvm_hpt_info info;
3617 
3618 		err = kvmppc_allocate_hpt(&info, order);
3619 		/* If we get here, it means userspace didn't specify a
3620 		 * size explicitly.  So, try successively smaller
3621 		 * sizes if the default failed. */
3622 		while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
3623 			err  = kvmppc_allocate_hpt(&info, order);
3624 
3625 		if (err < 0) {
3626 			pr_err("KVM: Couldn't alloc HPT\n");
3627 			goto out;
3628 		}
3629 
3630 		kvmppc_set_hpt(kvm, &info);
3631 	}
3632 
3633 	/* Look up the memslot for guest physical address 0 */
3634 	srcu_idx = srcu_read_lock(&kvm->srcu);
3635 	memslot = gfn_to_memslot(kvm, 0);
3636 
3637 	/* We must have some memory at 0 by now */
3638 	err = -EINVAL;
3639 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
3640 		goto out_srcu;
3641 
3642 	/* Look up the VMA for the start of this memory slot */
3643 	hva = memslot->userspace_addr;
3644 	down_read(&current->mm->mmap_sem);
3645 	vma = find_vma(current->mm, hva);
3646 	if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
3647 		goto up_out;
3648 
3649 	psize = vma_kernel_pagesize(vma);
3650 	porder = __ilog2(psize);
3651 
3652 	up_read(&current->mm->mmap_sem);
3653 
3654 	/* We can handle 4k, 64k or 16M pages in the VRMA */
3655 	err = -EINVAL;
3656 	if (!(psize == 0x1000 || psize == 0x10000 ||
3657 	      psize == 0x1000000))
3658 		goto out_srcu;
3659 
3660 	senc = slb_pgsize_encoding(psize);
3661 	kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
3662 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
3663 	/* Create HPTEs in the hash page table for the VRMA */
3664 	kvmppc_map_vrma(vcpu, memslot, porder);
3665 
3666 	/* Update VRMASD field in the LPCR */
3667 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
3668 		/* the -4 is to account for senc values starting at 0x10 */
3669 		lpcr = senc << (LPCR_VRMASD_SH - 4);
3670 		kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
3671 	}
3672 
3673 	/* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
3674 	smp_wmb();
3675 	err = 0;
3676  out_srcu:
3677 	srcu_read_unlock(&kvm->srcu, srcu_idx);
3678  out:
3679 	return err;
3680 
3681  up_out:
3682 	up_read(&current->mm->mmap_sem);
3683 	goto out_srcu;
3684 }
3685 
3686 /* Must be called with kvm->lock held and mmu_ready = 0 and no vcpus running */
3687 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
3688 {
3689 	kvmppc_free_radix(kvm);
3690 	kvmppc_update_lpcr(kvm, LPCR_VPM1,
3691 			   LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
3692 	kvmppc_rmap_reset(kvm);
3693 	kvm->arch.radix = 0;
3694 	kvm->arch.process_table = 0;
3695 	return 0;
3696 }
3697 
3698 /* Must be called with kvm->lock held and mmu_ready = 0 and no vcpus running */
3699 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
3700 {
3701 	int err;
3702 
3703 	err = kvmppc_init_vm_radix(kvm);
3704 	if (err)
3705 		return err;
3706 
3707 	kvmppc_free_hpt(&kvm->arch.hpt);
3708 	kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
3709 			   LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
3710 	kvm->arch.radix = 1;
3711 	return 0;
3712 }
3713 
3714 #ifdef CONFIG_KVM_XICS
3715 /*
3716  * Allocate a per-core structure for managing state about which cores are
3717  * running in the host versus the guest and for exchanging data between
3718  * real mode KVM and CPU running in the host.
3719  * This is only done for the first VM.
3720  * The allocated structure stays even if all VMs have stopped.
3721  * It is only freed when the kvm-hv module is unloaded.
3722  * It's OK for this routine to fail, we just don't support host
3723  * core operations like redirecting H_IPI wakeups.
3724  */
3725 void kvmppc_alloc_host_rm_ops(void)
3726 {
3727 	struct kvmppc_host_rm_ops *ops;
3728 	unsigned long l_ops;
3729 	int cpu, core;
3730 	int size;
3731 
3732 	/* Not the first time here ? */
3733 	if (kvmppc_host_rm_ops_hv != NULL)
3734 		return;
3735 
3736 	ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
3737 	if (!ops)
3738 		return;
3739 
3740 	size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
3741 	ops->rm_core = kzalloc(size, GFP_KERNEL);
3742 
3743 	if (!ops->rm_core) {
3744 		kfree(ops);
3745 		return;
3746 	}
3747 
3748 	cpus_read_lock();
3749 
3750 	for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
3751 		if (!cpu_online(cpu))
3752 			continue;
3753 
3754 		core = cpu >> threads_shift;
3755 		ops->rm_core[core].rm_state.in_host = 1;
3756 	}
3757 
3758 	ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
3759 
3760 	/*
3761 	 * Make the contents of the kvmppc_host_rm_ops structure visible
3762 	 * to other CPUs before we assign it to the global variable.
3763 	 * Do an atomic assignment (no locks used here), but if someone
3764 	 * beats us to it, just free our copy and return.
3765 	 */
3766 	smp_wmb();
3767 	l_ops = (unsigned long) ops;
3768 
3769 	if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
3770 		cpus_read_unlock();
3771 		kfree(ops->rm_core);
3772 		kfree(ops);
3773 		return;
3774 	}
3775 
3776 	cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
3777 					     "ppc/kvm_book3s:prepare",
3778 					     kvmppc_set_host_core,
3779 					     kvmppc_clear_host_core);
3780 	cpus_read_unlock();
3781 }
3782 
3783 void kvmppc_free_host_rm_ops(void)
3784 {
3785 	if (kvmppc_host_rm_ops_hv) {
3786 		cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
3787 		kfree(kvmppc_host_rm_ops_hv->rm_core);
3788 		kfree(kvmppc_host_rm_ops_hv);
3789 		kvmppc_host_rm_ops_hv = NULL;
3790 	}
3791 }
3792 #endif
3793 
3794 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
3795 {
3796 	unsigned long lpcr, lpid;
3797 	char buf[32];
3798 	int ret;
3799 
3800 	/* Allocate the guest's logical partition ID */
3801 
3802 	lpid = kvmppc_alloc_lpid();
3803 	if ((long)lpid < 0)
3804 		return -ENOMEM;
3805 	kvm->arch.lpid = lpid;
3806 
3807 	kvmppc_alloc_host_rm_ops();
3808 
3809 	/*
3810 	 * Since we don't flush the TLB when tearing down a VM,
3811 	 * and this lpid might have previously been used,
3812 	 * make sure we flush on each core before running the new VM.
3813 	 * On POWER9, the tlbie in mmu_partition_table_set_entry()
3814 	 * does this flush for us.
3815 	 */
3816 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
3817 		cpumask_setall(&kvm->arch.need_tlb_flush);
3818 
3819 	/* Start out with the default set of hcalls enabled */
3820 	memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
3821 	       sizeof(kvm->arch.enabled_hcalls));
3822 
3823 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
3824 		kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
3825 
3826 	/* Init LPCR for virtual RMA mode */
3827 	kvm->arch.host_lpid = mfspr(SPRN_LPID);
3828 	kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
3829 	lpcr &= LPCR_PECE | LPCR_LPES;
3830 	lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3831 		LPCR_VPM0 | LPCR_VPM1;
3832 	kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3833 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
3834 	/* On POWER8 turn on online bit to enable PURR/SPURR */
3835 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
3836 		lpcr |= LPCR_ONL;
3837 	/*
3838 	 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
3839 	 * Set HVICE bit to enable hypervisor virtualization interrupts.
3840 	 * Set HEIC to prevent OS interrupts to go to hypervisor (should
3841 	 * be unnecessary but better safe than sorry in case we re-enable
3842 	 * EE in HV mode with this LPCR still set)
3843 	 */
3844 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
3845 		lpcr &= ~LPCR_VPM0;
3846 		lpcr |= LPCR_HVICE | LPCR_HEIC;
3847 
3848 		/*
3849 		 * If xive is enabled, we route 0x500 interrupts directly
3850 		 * to the guest.
3851 		 */
3852 		if (xive_enabled())
3853 			lpcr |= LPCR_LPES;
3854 	}
3855 
3856 	/*
3857 	 * If the host uses radix, the guest starts out as radix.
3858 	 */
3859 	if (radix_enabled()) {
3860 		kvm->arch.radix = 1;
3861 		kvm->arch.mmu_ready = 1;
3862 		lpcr &= ~LPCR_VPM1;
3863 		lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
3864 		ret = kvmppc_init_vm_radix(kvm);
3865 		if (ret) {
3866 			kvmppc_free_lpid(kvm->arch.lpid);
3867 			return ret;
3868 		}
3869 		kvmppc_setup_partition_table(kvm);
3870 	}
3871 
3872 	kvm->arch.lpcr = lpcr;
3873 
3874 	/* Initialization for future HPT resizes */
3875 	kvm->arch.resize_hpt = NULL;
3876 
3877 	/*
3878 	 * Work out how many sets the TLB has, for the use of
3879 	 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
3880 	 */
3881 	if (radix_enabled())
3882 		kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;	/* 128 */
3883 	else if (cpu_has_feature(CPU_FTR_ARCH_300))
3884 		kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;	/* 256 */
3885 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
3886 		kvm->arch.tlb_sets = POWER8_TLB_SETS;		/* 512 */
3887 	else
3888 		kvm->arch.tlb_sets = POWER7_TLB_SETS;		/* 128 */
3889 
3890 	/*
3891 	 * Track that we now have a HV mode VM active. This blocks secondary
3892 	 * CPU threads from coming online.
3893 	 * On POWER9, we only need to do this if the "indep_threads_mode"
3894 	 * module parameter has been set to N.
3895 	 */
3896 	if (cpu_has_feature(CPU_FTR_ARCH_300))
3897 		kvm->arch.threads_indep = indep_threads_mode;
3898 	if (!kvm->arch.threads_indep)
3899 		kvm_hv_vm_activated();
3900 
3901 	/*
3902 	 * Initialize smt_mode depending on processor.
3903 	 * POWER8 and earlier have to use "strict" threading, where
3904 	 * all vCPUs in a vcore have to run on the same (sub)core,
3905 	 * whereas on POWER9 the threads can each run a different
3906 	 * guest.
3907 	 */
3908 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
3909 		kvm->arch.smt_mode = threads_per_subcore;
3910 	else
3911 		kvm->arch.smt_mode = 1;
3912 	kvm->arch.emul_smt_mode = 1;
3913 
3914 	/*
3915 	 * Create a debugfs directory for the VM
3916 	 */
3917 	snprintf(buf, sizeof(buf), "vm%d", current->pid);
3918 	kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
3919 	if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
3920 		kvmppc_mmu_debugfs_init(kvm);
3921 
3922 	return 0;
3923 }
3924 
3925 static void kvmppc_free_vcores(struct kvm *kvm)
3926 {
3927 	long int i;
3928 
3929 	for (i = 0; i < KVM_MAX_VCORES; ++i)
3930 		kfree(kvm->arch.vcores[i]);
3931 	kvm->arch.online_vcores = 0;
3932 }
3933 
3934 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
3935 {
3936 	debugfs_remove_recursive(kvm->arch.debugfs_dir);
3937 
3938 	if (!kvm->arch.threads_indep)
3939 		kvm_hv_vm_deactivated();
3940 
3941 	kvmppc_free_vcores(kvm);
3942 
3943 	kvmppc_free_lpid(kvm->arch.lpid);
3944 
3945 	if (kvm_is_radix(kvm))
3946 		kvmppc_free_radix(kvm);
3947 	else
3948 		kvmppc_free_hpt(&kvm->arch.hpt);
3949 
3950 	kvmppc_free_pimap(kvm);
3951 }
3952 
3953 /* We don't need to emulate any privileged instructions or dcbz */
3954 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
3955 				     unsigned int inst, int *advance)
3956 {
3957 	return EMULATE_FAIL;
3958 }
3959 
3960 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
3961 					ulong spr_val)
3962 {
3963 	return EMULATE_FAIL;
3964 }
3965 
3966 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
3967 					ulong *spr_val)
3968 {
3969 	return EMULATE_FAIL;
3970 }
3971 
3972 static int kvmppc_core_check_processor_compat_hv(void)
3973 {
3974 	if (!cpu_has_feature(CPU_FTR_HVMODE) ||
3975 	    !cpu_has_feature(CPU_FTR_ARCH_206))
3976 		return -EIO;
3977 
3978 	return 0;
3979 }
3980 
3981 #ifdef CONFIG_KVM_XICS
3982 
3983 void kvmppc_free_pimap(struct kvm *kvm)
3984 {
3985 	kfree(kvm->arch.pimap);
3986 }
3987 
3988 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
3989 {
3990 	return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
3991 }
3992 
3993 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
3994 {
3995 	struct irq_desc *desc;
3996 	struct kvmppc_irq_map *irq_map;
3997 	struct kvmppc_passthru_irqmap *pimap;
3998 	struct irq_chip *chip;
3999 	int i, rc = 0;
4000 
4001 	if (!kvm_irq_bypass)
4002 		return 1;
4003 
4004 	desc = irq_to_desc(host_irq);
4005 	if (!desc)
4006 		return -EIO;
4007 
4008 	mutex_lock(&kvm->lock);
4009 
4010 	pimap = kvm->arch.pimap;
4011 	if (pimap == NULL) {
4012 		/* First call, allocate structure to hold IRQ map */
4013 		pimap = kvmppc_alloc_pimap();
4014 		if (pimap == NULL) {
4015 			mutex_unlock(&kvm->lock);
4016 			return -ENOMEM;
4017 		}
4018 		kvm->arch.pimap = pimap;
4019 	}
4020 
4021 	/*
4022 	 * For now, we only support interrupts for which the EOI operation
4023 	 * is an OPAL call followed by a write to XIRR, since that's
4024 	 * what our real-mode EOI code does, or a XIVE interrupt
4025 	 */
4026 	chip = irq_data_get_irq_chip(&desc->irq_data);
4027 	if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
4028 		pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
4029 			host_irq, guest_gsi);
4030 		mutex_unlock(&kvm->lock);
4031 		return -ENOENT;
4032 	}
4033 
4034 	/*
4035 	 * See if we already have an entry for this guest IRQ number.
4036 	 * If it's mapped to a hardware IRQ number, that's an error,
4037 	 * otherwise re-use this entry.
4038 	 */
4039 	for (i = 0; i < pimap->n_mapped; i++) {
4040 		if (guest_gsi == pimap->mapped[i].v_hwirq) {
4041 			if (pimap->mapped[i].r_hwirq) {
4042 				mutex_unlock(&kvm->lock);
4043 				return -EINVAL;
4044 			}
4045 			break;
4046 		}
4047 	}
4048 
4049 	if (i == KVMPPC_PIRQ_MAPPED) {
4050 		mutex_unlock(&kvm->lock);
4051 		return -EAGAIN;		/* table is full */
4052 	}
4053 
4054 	irq_map = &pimap->mapped[i];
4055 
4056 	irq_map->v_hwirq = guest_gsi;
4057 	irq_map->desc = desc;
4058 
4059 	/*
4060 	 * Order the above two stores before the next to serialize with
4061 	 * the KVM real mode handler.
4062 	 */
4063 	smp_wmb();
4064 	irq_map->r_hwirq = desc->irq_data.hwirq;
4065 
4066 	if (i == pimap->n_mapped)
4067 		pimap->n_mapped++;
4068 
4069 	if (xive_enabled())
4070 		rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
4071 	else
4072 		kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
4073 	if (rc)
4074 		irq_map->r_hwirq = 0;
4075 
4076 	mutex_unlock(&kvm->lock);
4077 
4078 	return 0;
4079 }
4080 
4081 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4082 {
4083 	struct irq_desc *desc;
4084 	struct kvmppc_passthru_irqmap *pimap;
4085 	int i, rc = 0;
4086 
4087 	if (!kvm_irq_bypass)
4088 		return 0;
4089 
4090 	desc = irq_to_desc(host_irq);
4091 	if (!desc)
4092 		return -EIO;
4093 
4094 	mutex_lock(&kvm->lock);
4095 	if (!kvm->arch.pimap)
4096 		goto unlock;
4097 
4098 	pimap = kvm->arch.pimap;
4099 
4100 	for (i = 0; i < pimap->n_mapped; i++) {
4101 		if (guest_gsi == pimap->mapped[i].v_hwirq)
4102 			break;
4103 	}
4104 
4105 	if (i == pimap->n_mapped) {
4106 		mutex_unlock(&kvm->lock);
4107 		return -ENODEV;
4108 	}
4109 
4110 	if (xive_enabled())
4111 		rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
4112 	else
4113 		kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
4114 
4115 	/* invalidate the entry (what do do on error from the above ?) */
4116 	pimap->mapped[i].r_hwirq = 0;
4117 
4118 	/*
4119 	 * We don't free this structure even when the count goes to
4120 	 * zero. The structure is freed when we destroy the VM.
4121 	 */
4122  unlock:
4123 	mutex_unlock(&kvm->lock);
4124 	return rc;
4125 }
4126 
4127 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
4128 					     struct irq_bypass_producer *prod)
4129 {
4130 	int ret = 0;
4131 	struct kvm_kernel_irqfd *irqfd =
4132 		container_of(cons, struct kvm_kernel_irqfd, consumer);
4133 
4134 	irqfd->producer = prod;
4135 
4136 	ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
4137 	if (ret)
4138 		pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
4139 			prod->irq, irqfd->gsi, ret);
4140 
4141 	return ret;
4142 }
4143 
4144 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
4145 					      struct irq_bypass_producer *prod)
4146 {
4147 	int ret;
4148 	struct kvm_kernel_irqfd *irqfd =
4149 		container_of(cons, struct kvm_kernel_irqfd, consumer);
4150 
4151 	irqfd->producer = NULL;
4152 
4153 	/*
4154 	 * When producer of consumer is unregistered, we change back to
4155 	 * default external interrupt handling mode - KVM real mode
4156 	 * will switch back to host.
4157 	 */
4158 	ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
4159 	if (ret)
4160 		pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
4161 			prod->irq, irqfd->gsi, ret);
4162 }
4163 #endif
4164 
4165 static long kvm_arch_vm_ioctl_hv(struct file *filp,
4166 				 unsigned int ioctl, unsigned long arg)
4167 {
4168 	struct kvm *kvm __maybe_unused = filp->private_data;
4169 	void __user *argp = (void __user *)arg;
4170 	long r;
4171 
4172 	switch (ioctl) {
4173 
4174 	case KVM_PPC_ALLOCATE_HTAB: {
4175 		u32 htab_order;
4176 
4177 		r = -EFAULT;
4178 		if (get_user(htab_order, (u32 __user *)argp))
4179 			break;
4180 		r = kvmppc_alloc_reset_hpt(kvm, htab_order);
4181 		if (r)
4182 			break;
4183 		r = 0;
4184 		break;
4185 	}
4186 
4187 	case KVM_PPC_GET_HTAB_FD: {
4188 		struct kvm_get_htab_fd ghf;
4189 
4190 		r = -EFAULT;
4191 		if (copy_from_user(&ghf, argp, sizeof(ghf)))
4192 			break;
4193 		r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
4194 		break;
4195 	}
4196 
4197 	case KVM_PPC_RESIZE_HPT_PREPARE: {
4198 		struct kvm_ppc_resize_hpt rhpt;
4199 
4200 		r = -EFAULT;
4201 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
4202 			break;
4203 
4204 		r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
4205 		break;
4206 	}
4207 
4208 	case KVM_PPC_RESIZE_HPT_COMMIT: {
4209 		struct kvm_ppc_resize_hpt rhpt;
4210 
4211 		r = -EFAULT;
4212 		if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
4213 			break;
4214 
4215 		r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
4216 		break;
4217 	}
4218 
4219 	default:
4220 		r = -ENOTTY;
4221 	}
4222 
4223 	return r;
4224 }
4225 
4226 /*
4227  * List of hcall numbers to enable by default.
4228  * For compatibility with old userspace, we enable by default
4229  * all hcalls that were implemented before the hcall-enabling
4230  * facility was added.  Note this list should not include H_RTAS.
4231  */
4232 static unsigned int default_hcall_list[] = {
4233 	H_REMOVE,
4234 	H_ENTER,
4235 	H_READ,
4236 	H_PROTECT,
4237 	H_BULK_REMOVE,
4238 	H_GET_TCE,
4239 	H_PUT_TCE,
4240 	H_SET_DABR,
4241 	H_SET_XDABR,
4242 	H_CEDE,
4243 	H_PROD,
4244 	H_CONFER,
4245 	H_REGISTER_VPA,
4246 #ifdef CONFIG_KVM_XICS
4247 	H_EOI,
4248 	H_CPPR,
4249 	H_IPI,
4250 	H_IPOLL,
4251 	H_XIRR,
4252 	H_XIRR_X,
4253 #endif
4254 	0
4255 };
4256 
4257 static void init_default_hcalls(void)
4258 {
4259 	int i;
4260 	unsigned int hcall;
4261 
4262 	for (i = 0; default_hcall_list[i]; ++i) {
4263 		hcall = default_hcall_list[i];
4264 		WARN_ON(!kvmppc_hcall_impl_hv(hcall));
4265 		__set_bit(hcall / 4, default_enabled_hcalls);
4266 	}
4267 }
4268 
4269 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
4270 {
4271 	unsigned long lpcr;
4272 	int radix;
4273 	int err;
4274 
4275 	/* If not on a POWER9, reject it */
4276 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
4277 		return -ENODEV;
4278 
4279 	/* If any unknown flags set, reject it */
4280 	if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
4281 		return -EINVAL;
4282 
4283 	/* GR (guest radix) bit in process_table field must match */
4284 	radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
4285 	if (!!(cfg->process_table & PATB_GR) != radix)
4286 		return -EINVAL;
4287 
4288 	/* Process table size field must be reasonable, i.e. <= 24 */
4289 	if ((cfg->process_table & PRTS_MASK) > 24)
4290 		return -EINVAL;
4291 
4292 	/* We can change a guest to/from radix now, if the host is radix */
4293 	if (radix && !radix_enabled())
4294 		return -EINVAL;
4295 
4296 	mutex_lock(&kvm->lock);
4297 	if (radix != kvm_is_radix(kvm)) {
4298 		if (kvm->arch.mmu_ready) {
4299 			kvm->arch.mmu_ready = 0;
4300 			/* order mmu_ready vs. vcpus_running */
4301 			smp_mb();
4302 			if (atomic_read(&kvm->arch.vcpus_running)) {
4303 				kvm->arch.mmu_ready = 1;
4304 				err = -EBUSY;
4305 				goto out_unlock;
4306 			}
4307 		}
4308 		if (radix)
4309 			err = kvmppc_switch_mmu_to_radix(kvm);
4310 		else
4311 			err = kvmppc_switch_mmu_to_hpt(kvm);
4312 		if (err)
4313 			goto out_unlock;
4314 	}
4315 
4316 	kvm->arch.process_table = cfg->process_table;
4317 	kvmppc_setup_partition_table(kvm);
4318 
4319 	lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
4320 	kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
4321 	err = 0;
4322 
4323  out_unlock:
4324 	mutex_unlock(&kvm->lock);
4325 	return err;
4326 }
4327 
4328 static struct kvmppc_ops kvm_ops_hv = {
4329 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
4330 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
4331 	.get_one_reg = kvmppc_get_one_reg_hv,
4332 	.set_one_reg = kvmppc_set_one_reg_hv,
4333 	.vcpu_load   = kvmppc_core_vcpu_load_hv,
4334 	.vcpu_put    = kvmppc_core_vcpu_put_hv,
4335 	.set_msr     = kvmppc_set_msr_hv,
4336 	.vcpu_run    = kvmppc_vcpu_run_hv,
4337 	.vcpu_create = kvmppc_core_vcpu_create_hv,
4338 	.vcpu_free   = kvmppc_core_vcpu_free_hv,
4339 	.check_requests = kvmppc_core_check_requests_hv,
4340 	.get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
4341 	.flush_memslot  = kvmppc_core_flush_memslot_hv,
4342 	.prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
4343 	.commit_memory_region  = kvmppc_core_commit_memory_region_hv,
4344 	.unmap_hva = kvm_unmap_hva_hv,
4345 	.unmap_hva_range = kvm_unmap_hva_range_hv,
4346 	.age_hva  = kvm_age_hva_hv,
4347 	.test_age_hva = kvm_test_age_hva_hv,
4348 	.set_spte_hva = kvm_set_spte_hva_hv,
4349 	.mmu_destroy  = kvmppc_mmu_destroy_hv,
4350 	.free_memslot = kvmppc_core_free_memslot_hv,
4351 	.create_memslot = kvmppc_core_create_memslot_hv,
4352 	.init_vm =  kvmppc_core_init_vm_hv,
4353 	.destroy_vm = kvmppc_core_destroy_vm_hv,
4354 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
4355 	.emulate_op = kvmppc_core_emulate_op_hv,
4356 	.emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
4357 	.emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
4358 	.fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
4359 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
4360 	.hcall_implemented = kvmppc_hcall_impl_hv,
4361 #ifdef CONFIG_KVM_XICS
4362 	.irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
4363 	.irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
4364 #endif
4365 	.configure_mmu = kvmhv_configure_mmu,
4366 	.get_rmmu_info = kvmhv_get_rmmu_info,
4367 	.set_smt_mode = kvmhv_set_smt_mode,
4368 };
4369 
4370 static int kvm_init_subcore_bitmap(void)
4371 {
4372 	int i, j;
4373 	int nr_cores = cpu_nr_cores();
4374 	struct sibling_subcore_state *sibling_subcore_state;
4375 
4376 	for (i = 0; i < nr_cores; i++) {
4377 		int first_cpu = i * threads_per_core;
4378 		int node = cpu_to_node(first_cpu);
4379 
4380 		/* Ignore if it is already allocated. */
4381 		if (paca[first_cpu].sibling_subcore_state)
4382 			continue;
4383 
4384 		sibling_subcore_state =
4385 			kmalloc_node(sizeof(struct sibling_subcore_state),
4386 							GFP_KERNEL, node);
4387 		if (!sibling_subcore_state)
4388 			return -ENOMEM;
4389 
4390 		memset(sibling_subcore_state, 0,
4391 				sizeof(struct sibling_subcore_state));
4392 
4393 		for (j = 0; j < threads_per_core; j++) {
4394 			int cpu = first_cpu + j;
4395 
4396 			paca[cpu].sibling_subcore_state = sibling_subcore_state;
4397 		}
4398 	}
4399 	return 0;
4400 }
4401 
4402 static int kvmppc_radix_possible(void)
4403 {
4404 	return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
4405 }
4406 
4407 static int kvmppc_book3s_init_hv(void)
4408 {
4409 	int r;
4410 	/*
4411 	 * FIXME!! Do we need to check on all cpus ?
4412 	 */
4413 	r = kvmppc_core_check_processor_compat_hv();
4414 	if (r < 0)
4415 		return -ENODEV;
4416 
4417 	r = kvm_init_subcore_bitmap();
4418 	if (r)
4419 		return r;
4420 
4421 	/*
4422 	 * We need a way of accessing the XICS interrupt controller,
4423 	 * either directly, via paca[cpu].kvm_hstate.xics_phys, or
4424 	 * indirectly, via OPAL.
4425 	 */
4426 #ifdef CONFIG_SMP
4427 	if (!xive_enabled() && !local_paca->kvm_hstate.xics_phys) {
4428 		struct device_node *np;
4429 
4430 		np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
4431 		if (!np) {
4432 			pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
4433 			return -ENODEV;
4434 		}
4435 	}
4436 #endif
4437 
4438 	kvm_ops_hv.owner = THIS_MODULE;
4439 	kvmppc_hv_ops = &kvm_ops_hv;
4440 
4441 	init_default_hcalls();
4442 
4443 	init_vcore_lists();
4444 
4445 	r = kvmppc_mmu_hv_init();
4446 	if (r)
4447 		return r;
4448 
4449 	if (kvmppc_radix_possible())
4450 		r = kvmppc_radix_init();
4451 	return r;
4452 }
4453 
4454 static void kvmppc_book3s_exit_hv(void)
4455 {
4456 	kvmppc_free_host_rm_ops();
4457 	if (kvmppc_radix_possible())
4458 		kvmppc_radix_exit();
4459 	kvmppc_hv_ops = NULL;
4460 }
4461 
4462 module_init(kvmppc_book3s_init_hv);
4463 module_exit(kvmppc_book3s_exit_hv);
4464 MODULE_LICENSE("GPL");
4465 MODULE_ALIAS_MISCDEV(KVM_MINOR);
4466 MODULE_ALIAS("devname:kvm");
4467