xref: /openbmc/linux/arch/powerpc/kvm/book3s_hv.c (revision 78700c0a)
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/err.h>
23 #include <linux/slab.h>
24 #include <linux/preempt.h>
25 #include <linux/sched.h>
26 #include <linux/delay.h>
27 #include <linux/export.h>
28 #include <linux/fs.h>
29 #include <linux/anon_inodes.h>
30 #include <linux/cpu.h>
31 #include <linux/cpumask.h>
32 #include <linux/spinlock.h>
33 #include <linux/page-flags.h>
34 #include <linux/srcu.h>
35 #include <linux/miscdevice.h>
36 #include <linux/debugfs.h>
37 
38 #include <asm/reg.h>
39 #include <asm/cputable.h>
40 #include <asm/cacheflush.h>
41 #include <asm/tlbflush.h>
42 #include <asm/uaccess.h>
43 #include <asm/io.h>
44 #include <asm/kvm_ppc.h>
45 #include <asm/kvm_book3s.h>
46 #include <asm/mmu_context.h>
47 #include <asm/lppaca.h>
48 #include <asm/processor.h>
49 #include <asm/cputhreads.h>
50 #include <asm/page.h>
51 #include <asm/hvcall.h>
52 #include <asm/switch_to.h>
53 #include <asm/smp.h>
54 #include <asm/dbell.h>
55 #include <linux/gfp.h>
56 #include <linux/vmalloc.h>
57 #include <linux/highmem.h>
58 #include <linux/hugetlb.h>
59 #include <linux/module.h>
60 
61 #include "book3s.h"
62 
63 #define CREATE_TRACE_POINTS
64 #include "trace_hv.h"
65 
66 /* #define EXIT_DEBUG */
67 /* #define EXIT_DEBUG_SIMPLE */
68 /* #define EXIT_DEBUG_INT */
69 
70 /* Used to indicate that a guest page fault needs to be handled */
71 #define RESUME_PAGE_FAULT	(RESUME_GUEST | RESUME_FLAG_ARCH1)
72 
73 /* Used as a "null" value for timebase values */
74 #define TB_NIL	(~(u64)0)
75 
76 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
77 
78 static int dynamic_mt_modes = 6;
79 module_param(dynamic_mt_modes, int, S_IRUGO | S_IWUSR);
80 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
81 static int target_smt_mode;
82 module_param(target_smt_mode, int, S_IRUGO | S_IWUSR);
83 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
84 
85 #ifdef CONFIG_KVM_XICS
86 static struct kernel_param_ops module_param_ops = {
87 	.set = param_set_int,
88 	.get = param_get_int,
89 };
90 
91 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect,
92 							S_IRUGO | S_IWUSR);
93 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
94 #endif
95 
96 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
97 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
98 
99 static bool kvmppc_ipi_thread(int cpu)
100 {
101 	/* On POWER8 for IPIs to threads in the same core, use msgsnd */
102 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
103 		preempt_disable();
104 		if (cpu_first_thread_sibling(cpu) ==
105 		    cpu_first_thread_sibling(smp_processor_id())) {
106 			unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
107 			msg |= cpu_thread_in_core(cpu);
108 			smp_mb();
109 			__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
110 			preempt_enable();
111 			return true;
112 		}
113 		preempt_enable();
114 	}
115 
116 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
117 	if (cpu >= 0 && cpu < nr_cpu_ids && paca[cpu].kvm_hstate.xics_phys) {
118 		xics_wake_cpu(cpu);
119 		return true;
120 	}
121 #endif
122 
123 	return false;
124 }
125 
126 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
127 {
128 	int cpu;
129 	struct swait_queue_head *wqp;
130 
131 	wqp = kvm_arch_vcpu_wq(vcpu);
132 	if (swait_active(wqp)) {
133 		swake_up(wqp);
134 		++vcpu->stat.halt_wakeup;
135 	}
136 
137 	if (kvmppc_ipi_thread(vcpu->arch.thread_cpu))
138 		return;
139 
140 	/* CPU points to the first thread of the core */
141 	cpu = vcpu->cpu;
142 	if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
143 		smp_send_reschedule(cpu);
144 }
145 
146 /*
147  * We use the vcpu_load/put functions to measure stolen time.
148  * Stolen time is counted as time when either the vcpu is able to
149  * run as part of a virtual core, but the task running the vcore
150  * is preempted or sleeping, or when the vcpu needs something done
151  * in the kernel by the task running the vcpu, but that task is
152  * preempted or sleeping.  Those two things have to be counted
153  * separately, since one of the vcpu tasks will take on the job
154  * of running the core, and the other vcpu tasks in the vcore will
155  * sleep waiting for it to do that, but that sleep shouldn't count
156  * as stolen time.
157  *
158  * Hence we accumulate stolen time when the vcpu can run as part of
159  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
160  * needs its task to do other things in the kernel (for example,
161  * service a page fault) in busy_stolen.  We don't accumulate
162  * stolen time for a vcore when it is inactive, or for a vcpu
163  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
164  * a misnomer; it means that the vcpu task is not executing in
165  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
166  * the kernel.  We don't have any way of dividing up that time
167  * between time that the vcpu is genuinely stopped, time that
168  * the task is actively working on behalf of the vcpu, and time
169  * that the task is preempted, so we don't count any of it as
170  * stolen.
171  *
172  * Updates to busy_stolen are protected by arch.tbacct_lock;
173  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
174  * lock.  The stolen times are measured in units of timebase ticks.
175  * (Note that the != TB_NIL checks below are purely defensive;
176  * they should never fail.)
177  */
178 
179 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
180 {
181 	unsigned long flags;
182 
183 	spin_lock_irqsave(&vc->stoltb_lock, flags);
184 	vc->preempt_tb = mftb();
185 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
186 }
187 
188 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
189 {
190 	unsigned long flags;
191 
192 	spin_lock_irqsave(&vc->stoltb_lock, flags);
193 	if (vc->preempt_tb != TB_NIL) {
194 		vc->stolen_tb += mftb() - vc->preempt_tb;
195 		vc->preempt_tb = TB_NIL;
196 	}
197 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
198 }
199 
200 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
201 {
202 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
203 	unsigned long flags;
204 
205 	/*
206 	 * We can test vc->runner without taking the vcore lock,
207 	 * because only this task ever sets vc->runner to this
208 	 * vcpu, and once it is set to this vcpu, only this task
209 	 * ever sets it to NULL.
210 	 */
211 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
212 		kvmppc_core_end_stolen(vc);
213 
214 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
215 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
216 	    vcpu->arch.busy_preempt != TB_NIL) {
217 		vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
218 		vcpu->arch.busy_preempt = TB_NIL;
219 	}
220 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
221 }
222 
223 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
224 {
225 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
226 	unsigned long flags;
227 
228 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
229 		kvmppc_core_start_stolen(vc);
230 
231 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
232 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
233 		vcpu->arch.busy_preempt = mftb();
234 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
235 }
236 
237 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
238 {
239 	/*
240 	 * Check for illegal transactional state bit combination
241 	 * and if we find it, force the TS field to a safe state.
242 	 */
243 	if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
244 		msr &= ~MSR_TS_MASK;
245 	vcpu->arch.shregs.msr = msr;
246 	kvmppc_end_cede(vcpu);
247 }
248 
249 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
250 {
251 	vcpu->arch.pvr = pvr;
252 }
253 
254 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
255 {
256 	unsigned long pcr = 0;
257 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
258 
259 	if (arch_compat) {
260 		switch (arch_compat) {
261 		case PVR_ARCH_205:
262 			/*
263 			 * If an arch bit is set in PCR, all the defined
264 			 * higher-order arch bits also have to be set.
265 			 */
266 			pcr = PCR_ARCH_206 | PCR_ARCH_205;
267 			break;
268 		case PVR_ARCH_206:
269 		case PVR_ARCH_206p:
270 			pcr = PCR_ARCH_206;
271 			break;
272 		case PVR_ARCH_207:
273 			break;
274 		default:
275 			return -EINVAL;
276 		}
277 
278 		if (!cpu_has_feature(CPU_FTR_ARCH_207S)) {
279 			/* POWER7 can't emulate POWER8 */
280 			if (!(pcr & PCR_ARCH_206))
281 				return -EINVAL;
282 			pcr &= ~PCR_ARCH_206;
283 		}
284 	}
285 
286 	spin_lock(&vc->lock);
287 	vc->arch_compat = arch_compat;
288 	vc->pcr = pcr;
289 	spin_unlock(&vc->lock);
290 
291 	return 0;
292 }
293 
294 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
295 {
296 	int r;
297 
298 	pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
299 	pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
300 	       vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
301 	for (r = 0; r < 16; ++r)
302 		pr_err("r%2d = %.16lx  r%d = %.16lx\n",
303 		       r, kvmppc_get_gpr(vcpu, r),
304 		       r+16, kvmppc_get_gpr(vcpu, r+16));
305 	pr_err("ctr = %.16lx  lr  = %.16lx\n",
306 	       vcpu->arch.ctr, vcpu->arch.lr);
307 	pr_err("srr0 = %.16llx srr1 = %.16llx\n",
308 	       vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
309 	pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
310 	       vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
311 	pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
312 	       vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
313 	pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
314 	       vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
315 	pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
316 	pr_err("fault dar = %.16lx dsisr = %.8x\n",
317 	       vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
318 	pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
319 	for (r = 0; r < vcpu->arch.slb_max; ++r)
320 		pr_err("  ESID = %.16llx VSID = %.16llx\n",
321 		       vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
322 	pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
323 	       vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
324 	       vcpu->arch.last_inst);
325 }
326 
327 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
328 {
329 	struct kvm_vcpu *ret;
330 
331 	mutex_lock(&kvm->lock);
332 	ret = kvm_get_vcpu_by_id(kvm, id);
333 	mutex_unlock(&kvm->lock);
334 	return ret;
335 }
336 
337 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
338 {
339 	vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
340 	vpa->yield_count = cpu_to_be32(1);
341 }
342 
343 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
344 		   unsigned long addr, unsigned long len)
345 {
346 	/* check address is cacheline aligned */
347 	if (addr & (L1_CACHE_BYTES - 1))
348 		return -EINVAL;
349 	spin_lock(&vcpu->arch.vpa_update_lock);
350 	if (v->next_gpa != addr || v->len != len) {
351 		v->next_gpa = addr;
352 		v->len = addr ? len : 0;
353 		v->update_pending = 1;
354 	}
355 	spin_unlock(&vcpu->arch.vpa_update_lock);
356 	return 0;
357 }
358 
359 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
360 struct reg_vpa {
361 	u32 dummy;
362 	union {
363 		__be16 hword;
364 		__be32 word;
365 	} length;
366 };
367 
368 static int vpa_is_registered(struct kvmppc_vpa *vpap)
369 {
370 	if (vpap->update_pending)
371 		return vpap->next_gpa != 0;
372 	return vpap->pinned_addr != NULL;
373 }
374 
375 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
376 				       unsigned long flags,
377 				       unsigned long vcpuid, unsigned long vpa)
378 {
379 	struct kvm *kvm = vcpu->kvm;
380 	unsigned long len, nb;
381 	void *va;
382 	struct kvm_vcpu *tvcpu;
383 	int err;
384 	int subfunc;
385 	struct kvmppc_vpa *vpap;
386 
387 	tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
388 	if (!tvcpu)
389 		return H_PARAMETER;
390 
391 	subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
392 	if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
393 	    subfunc == H_VPA_REG_SLB) {
394 		/* Registering new area - address must be cache-line aligned */
395 		if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
396 			return H_PARAMETER;
397 
398 		/* convert logical addr to kernel addr and read length */
399 		va = kvmppc_pin_guest_page(kvm, vpa, &nb);
400 		if (va == NULL)
401 			return H_PARAMETER;
402 		if (subfunc == H_VPA_REG_VPA)
403 			len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
404 		else
405 			len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
406 		kvmppc_unpin_guest_page(kvm, va, vpa, false);
407 
408 		/* Check length */
409 		if (len > nb || len < sizeof(struct reg_vpa))
410 			return H_PARAMETER;
411 	} else {
412 		vpa = 0;
413 		len = 0;
414 	}
415 
416 	err = H_PARAMETER;
417 	vpap = NULL;
418 	spin_lock(&tvcpu->arch.vpa_update_lock);
419 
420 	switch (subfunc) {
421 	case H_VPA_REG_VPA:		/* register VPA */
422 		if (len < sizeof(struct lppaca))
423 			break;
424 		vpap = &tvcpu->arch.vpa;
425 		err = 0;
426 		break;
427 
428 	case H_VPA_REG_DTL:		/* register DTL */
429 		if (len < sizeof(struct dtl_entry))
430 			break;
431 		len -= len % sizeof(struct dtl_entry);
432 
433 		/* Check that they have previously registered a VPA */
434 		err = H_RESOURCE;
435 		if (!vpa_is_registered(&tvcpu->arch.vpa))
436 			break;
437 
438 		vpap = &tvcpu->arch.dtl;
439 		err = 0;
440 		break;
441 
442 	case H_VPA_REG_SLB:		/* register SLB shadow buffer */
443 		/* Check that they have previously registered a VPA */
444 		err = H_RESOURCE;
445 		if (!vpa_is_registered(&tvcpu->arch.vpa))
446 			break;
447 
448 		vpap = &tvcpu->arch.slb_shadow;
449 		err = 0;
450 		break;
451 
452 	case H_VPA_DEREG_VPA:		/* deregister VPA */
453 		/* Check they don't still have a DTL or SLB buf registered */
454 		err = H_RESOURCE;
455 		if (vpa_is_registered(&tvcpu->arch.dtl) ||
456 		    vpa_is_registered(&tvcpu->arch.slb_shadow))
457 			break;
458 
459 		vpap = &tvcpu->arch.vpa;
460 		err = 0;
461 		break;
462 
463 	case H_VPA_DEREG_DTL:		/* deregister DTL */
464 		vpap = &tvcpu->arch.dtl;
465 		err = 0;
466 		break;
467 
468 	case H_VPA_DEREG_SLB:		/* deregister SLB shadow buffer */
469 		vpap = &tvcpu->arch.slb_shadow;
470 		err = 0;
471 		break;
472 	}
473 
474 	if (vpap) {
475 		vpap->next_gpa = vpa;
476 		vpap->len = len;
477 		vpap->update_pending = 1;
478 	}
479 
480 	spin_unlock(&tvcpu->arch.vpa_update_lock);
481 
482 	return err;
483 }
484 
485 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
486 {
487 	struct kvm *kvm = vcpu->kvm;
488 	void *va;
489 	unsigned long nb;
490 	unsigned long gpa;
491 
492 	/*
493 	 * We need to pin the page pointed to by vpap->next_gpa,
494 	 * but we can't call kvmppc_pin_guest_page under the lock
495 	 * as it does get_user_pages() and down_read().  So we
496 	 * have to drop the lock, pin the page, then get the lock
497 	 * again and check that a new area didn't get registered
498 	 * in the meantime.
499 	 */
500 	for (;;) {
501 		gpa = vpap->next_gpa;
502 		spin_unlock(&vcpu->arch.vpa_update_lock);
503 		va = NULL;
504 		nb = 0;
505 		if (gpa)
506 			va = kvmppc_pin_guest_page(kvm, gpa, &nb);
507 		spin_lock(&vcpu->arch.vpa_update_lock);
508 		if (gpa == vpap->next_gpa)
509 			break;
510 		/* sigh... unpin that one and try again */
511 		if (va)
512 			kvmppc_unpin_guest_page(kvm, va, gpa, false);
513 	}
514 
515 	vpap->update_pending = 0;
516 	if (va && nb < vpap->len) {
517 		/*
518 		 * If it's now too short, it must be that userspace
519 		 * has changed the mappings underlying guest memory,
520 		 * so unregister the region.
521 		 */
522 		kvmppc_unpin_guest_page(kvm, va, gpa, false);
523 		va = NULL;
524 	}
525 	if (vpap->pinned_addr)
526 		kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
527 					vpap->dirty);
528 	vpap->gpa = gpa;
529 	vpap->pinned_addr = va;
530 	vpap->dirty = false;
531 	if (va)
532 		vpap->pinned_end = va + vpap->len;
533 }
534 
535 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
536 {
537 	if (!(vcpu->arch.vpa.update_pending ||
538 	      vcpu->arch.slb_shadow.update_pending ||
539 	      vcpu->arch.dtl.update_pending))
540 		return;
541 
542 	spin_lock(&vcpu->arch.vpa_update_lock);
543 	if (vcpu->arch.vpa.update_pending) {
544 		kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
545 		if (vcpu->arch.vpa.pinned_addr)
546 			init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
547 	}
548 	if (vcpu->arch.dtl.update_pending) {
549 		kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
550 		vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
551 		vcpu->arch.dtl_index = 0;
552 	}
553 	if (vcpu->arch.slb_shadow.update_pending)
554 		kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
555 	spin_unlock(&vcpu->arch.vpa_update_lock);
556 }
557 
558 /*
559  * Return the accumulated stolen time for the vcore up until `now'.
560  * The caller should hold the vcore lock.
561  */
562 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
563 {
564 	u64 p;
565 	unsigned long flags;
566 
567 	spin_lock_irqsave(&vc->stoltb_lock, flags);
568 	p = vc->stolen_tb;
569 	if (vc->vcore_state != VCORE_INACTIVE &&
570 	    vc->preempt_tb != TB_NIL)
571 		p += now - vc->preempt_tb;
572 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
573 	return p;
574 }
575 
576 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
577 				    struct kvmppc_vcore *vc)
578 {
579 	struct dtl_entry *dt;
580 	struct lppaca *vpa;
581 	unsigned long stolen;
582 	unsigned long core_stolen;
583 	u64 now;
584 
585 	dt = vcpu->arch.dtl_ptr;
586 	vpa = vcpu->arch.vpa.pinned_addr;
587 	now = mftb();
588 	core_stolen = vcore_stolen_time(vc, now);
589 	stolen = core_stolen - vcpu->arch.stolen_logged;
590 	vcpu->arch.stolen_logged = core_stolen;
591 	spin_lock_irq(&vcpu->arch.tbacct_lock);
592 	stolen += vcpu->arch.busy_stolen;
593 	vcpu->arch.busy_stolen = 0;
594 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
595 	if (!dt || !vpa)
596 		return;
597 	memset(dt, 0, sizeof(struct dtl_entry));
598 	dt->dispatch_reason = 7;
599 	dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
600 	dt->timebase = cpu_to_be64(now + vc->tb_offset);
601 	dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
602 	dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
603 	dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
604 	++dt;
605 	if (dt == vcpu->arch.dtl.pinned_end)
606 		dt = vcpu->arch.dtl.pinned_addr;
607 	vcpu->arch.dtl_ptr = dt;
608 	/* order writing *dt vs. writing vpa->dtl_idx */
609 	smp_wmb();
610 	vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
611 	vcpu->arch.dtl.dirty = true;
612 }
613 
614 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
615 {
616 	if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
617 		return true;
618 	if ((!vcpu->arch.vcore->arch_compat) &&
619 	    cpu_has_feature(CPU_FTR_ARCH_207S))
620 		return true;
621 	return false;
622 }
623 
624 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
625 			     unsigned long resource, unsigned long value1,
626 			     unsigned long value2)
627 {
628 	switch (resource) {
629 	case H_SET_MODE_RESOURCE_SET_CIABR:
630 		if (!kvmppc_power8_compatible(vcpu))
631 			return H_P2;
632 		if (value2)
633 			return H_P4;
634 		if (mflags)
635 			return H_UNSUPPORTED_FLAG_START;
636 		/* Guests can't breakpoint the hypervisor */
637 		if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
638 			return H_P3;
639 		vcpu->arch.ciabr  = value1;
640 		return H_SUCCESS;
641 	case H_SET_MODE_RESOURCE_SET_DAWR:
642 		if (!kvmppc_power8_compatible(vcpu))
643 			return H_P2;
644 		if (mflags)
645 			return H_UNSUPPORTED_FLAG_START;
646 		if (value2 & DABRX_HYP)
647 			return H_P4;
648 		vcpu->arch.dawr  = value1;
649 		vcpu->arch.dawrx = value2;
650 		return H_SUCCESS;
651 	default:
652 		return H_TOO_HARD;
653 	}
654 }
655 
656 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
657 {
658 	struct kvmppc_vcore *vcore = target->arch.vcore;
659 
660 	/*
661 	 * We expect to have been called by the real mode handler
662 	 * (kvmppc_rm_h_confer()) which would have directly returned
663 	 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
664 	 * have useful work to do and should not confer) so we don't
665 	 * recheck that here.
666 	 */
667 
668 	spin_lock(&vcore->lock);
669 	if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
670 	    vcore->vcore_state != VCORE_INACTIVE &&
671 	    vcore->runner)
672 		target = vcore->runner;
673 	spin_unlock(&vcore->lock);
674 
675 	return kvm_vcpu_yield_to(target);
676 }
677 
678 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
679 {
680 	int yield_count = 0;
681 	struct lppaca *lppaca;
682 
683 	spin_lock(&vcpu->arch.vpa_update_lock);
684 	lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
685 	if (lppaca)
686 		yield_count = be32_to_cpu(lppaca->yield_count);
687 	spin_unlock(&vcpu->arch.vpa_update_lock);
688 	return yield_count;
689 }
690 
691 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
692 {
693 	unsigned long req = kvmppc_get_gpr(vcpu, 3);
694 	unsigned long target, ret = H_SUCCESS;
695 	int yield_count;
696 	struct kvm_vcpu *tvcpu;
697 	int idx, rc;
698 
699 	if (req <= MAX_HCALL_OPCODE &&
700 	    !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
701 		return RESUME_HOST;
702 
703 	switch (req) {
704 	case H_CEDE:
705 		break;
706 	case H_PROD:
707 		target = kvmppc_get_gpr(vcpu, 4);
708 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
709 		if (!tvcpu) {
710 			ret = H_PARAMETER;
711 			break;
712 		}
713 		tvcpu->arch.prodded = 1;
714 		smp_mb();
715 		if (vcpu->arch.ceded) {
716 			if (swait_active(&vcpu->wq)) {
717 				swake_up(&vcpu->wq);
718 				vcpu->stat.halt_wakeup++;
719 			}
720 		}
721 		break;
722 	case H_CONFER:
723 		target = kvmppc_get_gpr(vcpu, 4);
724 		if (target == -1)
725 			break;
726 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
727 		if (!tvcpu) {
728 			ret = H_PARAMETER;
729 			break;
730 		}
731 		yield_count = kvmppc_get_gpr(vcpu, 5);
732 		if (kvmppc_get_yield_count(tvcpu) != yield_count)
733 			break;
734 		kvm_arch_vcpu_yield_to(tvcpu);
735 		break;
736 	case H_REGISTER_VPA:
737 		ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
738 					kvmppc_get_gpr(vcpu, 5),
739 					kvmppc_get_gpr(vcpu, 6));
740 		break;
741 	case H_RTAS:
742 		if (list_empty(&vcpu->kvm->arch.rtas_tokens))
743 			return RESUME_HOST;
744 
745 		idx = srcu_read_lock(&vcpu->kvm->srcu);
746 		rc = kvmppc_rtas_hcall(vcpu);
747 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
748 
749 		if (rc == -ENOENT)
750 			return RESUME_HOST;
751 		else if (rc == 0)
752 			break;
753 
754 		/* Send the error out to userspace via KVM_RUN */
755 		return rc;
756 	case H_LOGICAL_CI_LOAD:
757 		ret = kvmppc_h_logical_ci_load(vcpu);
758 		if (ret == H_TOO_HARD)
759 			return RESUME_HOST;
760 		break;
761 	case H_LOGICAL_CI_STORE:
762 		ret = kvmppc_h_logical_ci_store(vcpu);
763 		if (ret == H_TOO_HARD)
764 			return RESUME_HOST;
765 		break;
766 	case H_SET_MODE:
767 		ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
768 					kvmppc_get_gpr(vcpu, 5),
769 					kvmppc_get_gpr(vcpu, 6),
770 					kvmppc_get_gpr(vcpu, 7));
771 		if (ret == H_TOO_HARD)
772 			return RESUME_HOST;
773 		break;
774 	case H_XIRR:
775 	case H_CPPR:
776 	case H_EOI:
777 	case H_IPI:
778 	case H_IPOLL:
779 	case H_XIRR_X:
780 		if (kvmppc_xics_enabled(vcpu)) {
781 			ret = kvmppc_xics_hcall(vcpu, req);
782 			break;
783 		}
784 		return RESUME_HOST;
785 	case H_PUT_TCE:
786 		ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
787 						kvmppc_get_gpr(vcpu, 5),
788 						kvmppc_get_gpr(vcpu, 6));
789 		if (ret == H_TOO_HARD)
790 			return RESUME_HOST;
791 		break;
792 	case H_PUT_TCE_INDIRECT:
793 		ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
794 						kvmppc_get_gpr(vcpu, 5),
795 						kvmppc_get_gpr(vcpu, 6),
796 						kvmppc_get_gpr(vcpu, 7));
797 		if (ret == H_TOO_HARD)
798 			return RESUME_HOST;
799 		break;
800 	case H_STUFF_TCE:
801 		ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
802 						kvmppc_get_gpr(vcpu, 5),
803 						kvmppc_get_gpr(vcpu, 6),
804 						kvmppc_get_gpr(vcpu, 7));
805 		if (ret == H_TOO_HARD)
806 			return RESUME_HOST;
807 		break;
808 	default:
809 		return RESUME_HOST;
810 	}
811 	kvmppc_set_gpr(vcpu, 3, ret);
812 	vcpu->arch.hcall_needed = 0;
813 	return RESUME_GUEST;
814 }
815 
816 static int kvmppc_hcall_impl_hv(unsigned long cmd)
817 {
818 	switch (cmd) {
819 	case H_CEDE:
820 	case H_PROD:
821 	case H_CONFER:
822 	case H_REGISTER_VPA:
823 	case H_SET_MODE:
824 	case H_LOGICAL_CI_LOAD:
825 	case H_LOGICAL_CI_STORE:
826 #ifdef CONFIG_KVM_XICS
827 	case H_XIRR:
828 	case H_CPPR:
829 	case H_EOI:
830 	case H_IPI:
831 	case H_IPOLL:
832 	case H_XIRR_X:
833 #endif
834 		return 1;
835 	}
836 
837 	/* See if it's in the real-mode table */
838 	return kvmppc_hcall_impl_hv_realmode(cmd);
839 }
840 
841 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
842 					struct kvm_vcpu *vcpu)
843 {
844 	u32 last_inst;
845 
846 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
847 					EMULATE_DONE) {
848 		/*
849 		 * Fetch failed, so return to guest and
850 		 * try executing it again.
851 		 */
852 		return RESUME_GUEST;
853 	}
854 
855 	if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
856 		run->exit_reason = KVM_EXIT_DEBUG;
857 		run->debug.arch.address = kvmppc_get_pc(vcpu);
858 		return RESUME_HOST;
859 	} else {
860 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
861 		return RESUME_GUEST;
862 	}
863 }
864 
865 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
866 				 struct task_struct *tsk)
867 {
868 	int r = RESUME_HOST;
869 
870 	vcpu->stat.sum_exits++;
871 
872 	/*
873 	 * This can happen if an interrupt occurs in the last stages
874 	 * of guest entry or the first stages of guest exit (i.e. after
875 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
876 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
877 	 * That can happen due to a bug, or due to a machine check
878 	 * occurring at just the wrong time.
879 	 */
880 	if (vcpu->arch.shregs.msr & MSR_HV) {
881 		printk(KERN_EMERG "KVM trap in HV mode!\n");
882 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
883 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
884 			vcpu->arch.shregs.msr);
885 		kvmppc_dump_regs(vcpu);
886 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
887 		run->hw.hardware_exit_reason = vcpu->arch.trap;
888 		return RESUME_HOST;
889 	}
890 	run->exit_reason = KVM_EXIT_UNKNOWN;
891 	run->ready_for_interrupt_injection = 1;
892 	switch (vcpu->arch.trap) {
893 	/* We're good on these - the host merely wanted to get our attention */
894 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
895 		vcpu->stat.dec_exits++;
896 		r = RESUME_GUEST;
897 		break;
898 	case BOOK3S_INTERRUPT_EXTERNAL:
899 	case BOOK3S_INTERRUPT_H_DOORBELL:
900 		vcpu->stat.ext_intr_exits++;
901 		r = RESUME_GUEST;
902 		break;
903 	/* HMI is hypervisor interrupt and host has handled it. Resume guest.*/
904 	case BOOK3S_INTERRUPT_HMI:
905 	case BOOK3S_INTERRUPT_PERFMON:
906 		r = RESUME_GUEST;
907 		break;
908 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
909 		/*
910 		 * Deliver a machine check interrupt to the guest.
911 		 * We have to do this, even if the host has handled the
912 		 * machine check, because machine checks use SRR0/1 and
913 		 * the interrupt might have trashed guest state in them.
914 		 */
915 		kvmppc_book3s_queue_irqprio(vcpu,
916 					    BOOK3S_INTERRUPT_MACHINE_CHECK);
917 		r = RESUME_GUEST;
918 		break;
919 	case BOOK3S_INTERRUPT_PROGRAM:
920 	{
921 		ulong flags;
922 		/*
923 		 * Normally program interrupts are delivered directly
924 		 * to the guest by the hardware, but we can get here
925 		 * as a result of a hypervisor emulation interrupt
926 		 * (e40) getting turned into a 700 by BML RTAS.
927 		 */
928 		flags = vcpu->arch.shregs.msr & 0x1f0000ull;
929 		kvmppc_core_queue_program(vcpu, flags);
930 		r = RESUME_GUEST;
931 		break;
932 	}
933 	case BOOK3S_INTERRUPT_SYSCALL:
934 	{
935 		/* hcall - punt to userspace */
936 		int i;
937 
938 		/* hypercall with MSR_PR has already been handled in rmode,
939 		 * and never reaches here.
940 		 */
941 
942 		run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
943 		for (i = 0; i < 9; ++i)
944 			run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
945 		run->exit_reason = KVM_EXIT_PAPR_HCALL;
946 		vcpu->arch.hcall_needed = 1;
947 		r = RESUME_HOST;
948 		break;
949 	}
950 	/*
951 	 * We get these next two if the guest accesses a page which it thinks
952 	 * it has mapped but which is not actually present, either because
953 	 * it is for an emulated I/O device or because the corresonding
954 	 * host page has been paged out.  Any other HDSI/HISI interrupts
955 	 * have been handled already.
956 	 */
957 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
958 		r = RESUME_PAGE_FAULT;
959 		break;
960 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
961 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
962 		vcpu->arch.fault_dsisr = 0;
963 		r = RESUME_PAGE_FAULT;
964 		break;
965 	/*
966 	 * This occurs if the guest executes an illegal instruction.
967 	 * If the guest debug is disabled, generate a program interrupt
968 	 * to the guest. If guest debug is enabled, we need to check
969 	 * whether the instruction is a software breakpoint instruction.
970 	 * Accordingly return to Guest or Host.
971 	 */
972 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
973 		if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
974 			vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
975 				swab32(vcpu->arch.emul_inst) :
976 				vcpu->arch.emul_inst;
977 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
978 			r = kvmppc_emulate_debug_inst(run, vcpu);
979 		} else {
980 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
981 			r = RESUME_GUEST;
982 		}
983 		break;
984 	/*
985 	 * This occurs if the guest (kernel or userspace), does something that
986 	 * is prohibited by HFSCR.  We just generate a program interrupt to
987 	 * the guest.
988 	 */
989 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
990 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
991 		r = RESUME_GUEST;
992 		break;
993 	default:
994 		kvmppc_dump_regs(vcpu);
995 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
996 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
997 			vcpu->arch.shregs.msr);
998 		run->hw.hardware_exit_reason = vcpu->arch.trap;
999 		r = RESUME_HOST;
1000 		break;
1001 	}
1002 
1003 	return r;
1004 }
1005 
1006 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1007 					    struct kvm_sregs *sregs)
1008 {
1009 	int i;
1010 
1011 	memset(sregs, 0, sizeof(struct kvm_sregs));
1012 	sregs->pvr = vcpu->arch.pvr;
1013 	for (i = 0; i < vcpu->arch.slb_max; i++) {
1014 		sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1015 		sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1016 	}
1017 
1018 	return 0;
1019 }
1020 
1021 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1022 					    struct kvm_sregs *sregs)
1023 {
1024 	int i, j;
1025 
1026 	/* Only accept the same PVR as the host's, since we can't spoof it */
1027 	if (sregs->pvr != vcpu->arch.pvr)
1028 		return -EINVAL;
1029 
1030 	j = 0;
1031 	for (i = 0; i < vcpu->arch.slb_nr; i++) {
1032 		if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1033 			vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1034 			vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1035 			++j;
1036 		}
1037 	}
1038 	vcpu->arch.slb_max = j;
1039 
1040 	return 0;
1041 }
1042 
1043 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1044 		bool preserve_top32)
1045 {
1046 	struct kvm *kvm = vcpu->kvm;
1047 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
1048 	u64 mask;
1049 
1050 	mutex_lock(&kvm->lock);
1051 	spin_lock(&vc->lock);
1052 	/*
1053 	 * If ILE (interrupt little-endian) has changed, update the
1054 	 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1055 	 */
1056 	if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1057 		struct kvm_vcpu *vcpu;
1058 		int i;
1059 
1060 		kvm_for_each_vcpu(i, vcpu, kvm) {
1061 			if (vcpu->arch.vcore != vc)
1062 				continue;
1063 			if (new_lpcr & LPCR_ILE)
1064 				vcpu->arch.intr_msr |= MSR_LE;
1065 			else
1066 				vcpu->arch.intr_msr &= ~MSR_LE;
1067 		}
1068 	}
1069 
1070 	/*
1071 	 * Userspace can only modify DPFD (default prefetch depth),
1072 	 * ILE (interrupt little-endian) and TC (translation control).
1073 	 * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
1074 	 */
1075 	mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1076 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
1077 		mask |= LPCR_AIL;
1078 
1079 	/* Broken 32-bit version of LPCR must not clear top bits */
1080 	if (preserve_top32)
1081 		mask &= 0xFFFFFFFF;
1082 	vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1083 	spin_unlock(&vc->lock);
1084 	mutex_unlock(&kvm->lock);
1085 }
1086 
1087 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1088 				 union kvmppc_one_reg *val)
1089 {
1090 	int r = 0;
1091 	long int i;
1092 
1093 	switch (id) {
1094 	case KVM_REG_PPC_DEBUG_INST:
1095 		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1096 		break;
1097 	case KVM_REG_PPC_HIOR:
1098 		*val = get_reg_val(id, 0);
1099 		break;
1100 	case KVM_REG_PPC_DABR:
1101 		*val = get_reg_val(id, vcpu->arch.dabr);
1102 		break;
1103 	case KVM_REG_PPC_DABRX:
1104 		*val = get_reg_val(id, vcpu->arch.dabrx);
1105 		break;
1106 	case KVM_REG_PPC_DSCR:
1107 		*val = get_reg_val(id, vcpu->arch.dscr);
1108 		break;
1109 	case KVM_REG_PPC_PURR:
1110 		*val = get_reg_val(id, vcpu->arch.purr);
1111 		break;
1112 	case KVM_REG_PPC_SPURR:
1113 		*val = get_reg_val(id, vcpu->arch.spurr);
1114 		break;
1115 	case KVM_REG_PPC_AMR:
1116 		*val = get_reg_val(id, vcpu->arch.amr);
1117 		break;
1118 	case KVM_REG_PPC_UAMOR:
1119 		*val = get_reg_val(id, vcpu->arch.uamor);
1120 		break;
1121 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1122 		i = id - KVM_REG_PPC_MMCR0;
1123 		*val = get_reg_val(id, vcpu->arch.mmcr[i]);
1124 		break;
1125 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1126 		i = id - KVM_REG_PPC_PMC1;
1127 		*val = get_reg_val(id, vcpu->arch.pmc[i]);
1128 		break;
1129 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1130 		i = id - KVM_REG_PPC_SPMC1;
1131 		*val = get_reg_val(id, vcpu->arch.spmc[i]);
1132 		break;
1133 	case KVM_REG_PPC_SIAR:
1134 		*val = get_reg_val(id, vcpu->arch.siar);
1135 		break;
1136 	case KVM_REG_PPC_SDAR:
1137 		*val = get_reg_val(id, vcpu->arch.sdar);
1138 		break;
1139 	case KVM_REG_PPC_SIER:
1140 		*val = get_reg_val(id, vcpu->arch.sier);
1141 		break;
1142 	case KVM_REG_PPC_IAMR:
1143 		*val = get_reg_val(id, vcpu->arch.iamr);
1144 		break;
1145 	case KVM_REG_PPC_PSPB:
1146 		*val = get_reg_val(id, vcpu->arch.pspb);
1147 		break;
1148 	case KVM_REG_PPC_DPDES:
1149 		*val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1150 		break;
1151 	case KVM_REG_PPC_DAWR:
1152 		*val = get_reg_val(id, vcpu->arch.dawr);
1153 		break;
1154 	case KVM_REG_PPC_DAWRX:
1155 		*val = get_reg_val(id, vcpu->arch.dawrx);
1156 		break;
1157 	case KVM_REG_PPC_CIABR:
1158 		*val = get_reg_val(id, vcpu->arch.ciabr);
1159 		break;
1160 	case KVM_REG_PPC_CSIGR:
1161 		*val = get_reg_val(id, vcpu->arch.csigr);
1162 		break;
1163 	case KVM_REG_PPC_TACR:
1164 		*val = get_reg_val(id, vcpu->arch.tacr);
1165 		break;
1166 	case KVM_REG_PPC_TCSCR:
1167 		*val = get_reg_val(id, vcpu->arch.tcscr);
1168 		break;
1169 	case KVM_REG_PPC_PID:
1170 		*val = get_reg_val(id, vcpu->arch.pid);
1171 		break;
1172 	case KVM_REG_PPC_ACOP:
1173 		*val = get_reg_val(id, vcpu->arch.acop);
1174 		break;
1175 	case KVM_REG_PPC_WORT:
1176 		*val = get_reg_val(id, vcpu->arch.wort);
1177 		break;
1178 	case KVM_REG_PPC_VPA_ADDR:
1179 		spin_lock(&vcpu->arch.vpa_update_lock);
1180 		*val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1181 		spin_unlock(&vcpu->arch.vpa_update_lock);
1182 		break;
1183 	case KVM_REG_PPC_VPA_SLB:
1184 		spin_lock(&vcpu->arch.vpa_update_lock);
1185 		val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1186 		val->vpaval.length = vcpu->arch.slb_shadow.len;
1187 		spin_unlock(&vcpu->arch.vpa_update_lock);
1188 		break;
1189 	case KVM_REG_PPC_VPA_DTL:
1190 		spin_lock(&vcpu->arch.vpa_update_lock);
1191 		val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1192 		val->vpaval.length = vcpu->arch.dtl.len;
1193 		spin_unlock(&vcpu->arch.vpa_update_lock);
1194 		break;
1195 	case KVM_REG_PPC_TB_OFFSET:
1196 		*val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1197 		break;
1198 	case KVM_REG_PPC_LPCR:
1199 	case KVM_REG_PPC_LPCR_64:
1200 		*val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1201 		break;
1202 	case KVM_REG_PPC_PPR:
1203 		*val = get_reg_val(id, vcpu->arch.ppr);
1204 		break;
1205 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1206 	case KVM_REG_PPC_TFHAR:
1207 		*val = get_reg_val(id, vcpu->arch.tfhar);
1208 		break;
1209 	case KVM_REG_PPC_TFIAR:
1210 		*val = get_reg_val(id, vcpu->arch.tfiar);
1211 		break;
1212 	case KVM_REG_PPC_TEXASR:
1213 		*val = get_reg_val(id, vcpu->arch.texasr);
1214 		break;
1215 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1216 		i = id - KVM_REG_PPC_TM_GPR0;
1217 		*val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1218 		break;
1219 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1220 	{
1221 		int j;
1222 		i = id - KVM_REG_PPC_TM_VSR0;
1223 		if (i < 32)
1224 			for (j = 0; j < TS_FPRWIDTH; j++)
1225 				val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1226 		else {
1227 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1228 				val->vval = vcpu->arch.vr_tm.vr[i-32];
1229 			else
1230 				r = -ENXIO;
1231 		}
1232 		break;
1233 	}
1234 	case KVM_REG_PPC_TM_CR:
1235 		*val = get_reg_val(id, vcpu->arch.cr_tm);
1236 		break;
1237 	case KVM_REG_PPC_TM_LR:
1238 		*val = get_reg_val(id, vcpu->arch.lr_tm);
1239 		break;
1240 	case KVM_REG_PPC_TM_CTR:
1241 		*val = get_reg_val(id, vcpu->arch.ctr_tm);
1242 		break;
1243 	case KVM_REG_PPC_TM_FPSCR:
1244 		*val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1245 		break;
1246 	case KVM_REG_PPC_TM_AMR:
1247 		*val = get_reg_val(id, vcpu->arch.amr_tm);
1248 		break;
1249 	case KVM_REG_PPC_TM_PPR:
1250 		*val = get_reg_val(id, vcpu->arch.ppr_tm);
1251 		break;
1252 	case KVM_REG_PPC_TM_VRSAVE:
1253 		*val = get_reg_val(id, vcpu->arch.vrsave_tm);
1254 		break;
1255 	case KVM_REG_PPC_TM_VSCR:
1256 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1257 			*val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1258 		else
1259 			r = -ENXIO;
1260 		break;
1261 	case KVM_REG_PPC_TM_DSCR:
1262 		*val = get_reg_val(id, vcpu->arch.dscr_tm);
1263 		break;
1264 	case KVM_REG_PPC_TM_TAR:
1265 		*val = get_reg_val(id, vcpu->arch.tar_tm);
1266 		break;
1267 #endif
1268 	case KVM_REG_PPC_ARCH_COMPAT:
1269 		*val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1270 		break;
1271 	default:
1272 		r = -EINVAL;
1273 		break;
1274 	}
1275 
1276 	return r;
1277 }
1278 
1279 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1280 				 union kvmppc_one_reg *val)
1281 {
1282 	int r = 0;
1283 	long int i;
1284 	unsigned long addr, len;
1285 
1286 	switch (id) {
1287 	case KVM_REG_PPC_HIOR:
1288 		/* Only allow this to be set to zero */
1289 		if (set_reg_val(id, *val))
1290 			r = -EINVAL;
1291 		break;
1292 	case KVM_REG_PPC_DABR:
1293 		vcpu->arch.dabr = set_reg_val(id, *val);
1294 		break;
1295 	case KVM_REG_PPC_DABRX:
1296 		vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1297 		break;
1298 	case KVM_REG_PPC_DSCR:
1299 		vcpu->arch.dscr = set_reg_val(id, *val);
1300 		break;
1301 	case KVM_REG_PPC_PURR:
1302 		vcpu->arch.purr = set_reg_val(id, *val);
1303 		break;
1304 	case KVM_REG_PPC_SPURR:
1305 		vcpu->arch.spurr = set_reg_val(id, *val);
1306 		break;
1307 	case KVM_REG_PPC_AMR:
1308 		vcpu->arch.amr = set_reg_val(id, *val);
1309 		break;
1310 	case KVM_REG_PPC_UAMOR:
1311 		vcpu->arch.uamor = set_reg_val(id, *val);
1312 		break;
1313 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1314 		i = id - KVM_REG_PPC_MMCR0;
1315 		vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1316 		break;
1317 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1318 		i = id - KVM_REG_PPC_PMC1;
1319 		vcpu->arch.pmc[i] = set_reg_val(id, *val);
1320 		break;
1321 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1322 		i = id - KVM_REG_PPC_SPMC1;
1323 		vcpu->arch.spmc[i] = set_reg_val(id, *val);
1324 		break;
1325 	case KVM_REG_PPC_SIAR:
1326 		vcpu->arch.siar = set_reg_val(id, *val);
1327 		break;
1328 	case KVM_REG_PPC_SDAR:
1329 		vcpu->arch.sdar = set_reg_val(id, *val);
1330 		break;
1331 	case KVM_REG_PPC_SIER:
1332 		vcpu->arch.sier = set_reg_val(id, *val);
1333 		break;
1334 	case KVM_REG_PPC_IAMR:
1335 		vcpu->arch.iamr = set_reg_val(id, *val);
1336 		break;
1337 	case KVM_REG_PPC_PSPB:
1338 		vcpu->arch.pspb = set_reg_val(id, *val);
1339 		break;
1340 	case KVM_REG_PPC_DPDES:
1341 		vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1342 		break;
1343 	case KVM_REG_PPC_DAWR:
1344 		vcpu->arch.dawr = set_reg_val(id, *val);
1345 		break;
1346 	case KVM_REG_PPC_DAWRX:
1347 		vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1348 		break;
1349 	case KVM_REG_PPC_CIABR:
1350 		vcpu->arch.ciabr = set_reg_val(id, *val);
1351 		/* Don't allow setting breakpoints in hypervisor code */
1352 		if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1353 			vcpu->arch.ciabr &= ~CIABR_PRIV;	/* disable */
1354 		break;
1355 	case KVM_REG_PPC_CSIGR:
1356 		vcpu->arch.csigr = set_reg_val(id, *val);
1357 		break;
1358 	case KVM_REG_PPC_TACR:
1359 		vcpu->arch.tacr = set_reg_val(id, *val);
1360 		break;
1361 	case KVM_REG_PPC_TCSCR:
1362 		vcpu->arch.tcscr = set_reg_val(id, *val);
1363 		break;
1364 	case KVM_REG_PPC_PID:
1365 		vcpu->arch.pid = set_reg_val(id, *val);
1366 		break;
1367 	case KVM_REG_PPC_ACOP:
1368 		vcpu->arch.acop = set_reg_val(id, *val);
1369 		break;
1370 	case KVM_REG_PPC_WORT:
1371 		vcpu->arch.wort = set_reg_val(id, *val);
1372 		break;
1373 	case KVM_REG_PPC_VPA_ADDR:
1374 		addr = set_reg_val(id, *val);
1375 		r = -EINVAL;
1376 		if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1377 			      vcpu->arch.dtl.next_gpa))
1378 			break;
1379 		r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1380 		break;
1381 	case KVM_REG_PPC_VPA_SLB:
1382 		addr = val->vpaval.addr;
1383 		len = val->vpaval.length;
1384 		r = -EINVAL;
1385 		if (addr && !vcpu->arch.vpa.next_gpa)
1386 			break;
1387 		r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1388 		break;
1389 	case KVM_REG_PPC_VPA_DTL:
1390 		addr = val->vpaval.addr;
1391 		len = val->vpaval.length;
1392 		r = -EINVAL;
1393 		if (addr && (len < sizeof(struct dtl_entry) ||
1394 			     !vcpu->arch.vpa.next_gpa))
1395 			break;
1396 		len -= len % sizeof(struct dtl_entry);
1397 		r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1398 		break;
1399 	case KVM_REG_PPC_TB_OFFSET:
1400 		/* round up to multiple of 2^24 */
1401 		vcpu->arch.vcore->tb_offset =
1402 			ALIGN(set_reg_val(id, *val), 1UL << 24);
1403 		break;
1404 	case KVM_REG_PPC_LPCR:
1405 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1406 		break;
1407 	case KVM_REG_PPC_LPCR_64:
1408 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1409 		break;
1410 	case KVM_REG_PPC_PPR:
1411 		vcpu->arch.ppr = set_reg_val(id, *val);
1412 		break;
1413 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1414 	case KVM_REG_PPC_TFHAR:
1415 		vcpu->arch.tfhar = set_reg_val(id, *val);
1416 		break;
1417 	case KVM_REG_PPC_TFIAR:
1418 		vcpu->arch.tfiar = set_reg_val(id, *val);
1419 		break;
1420 	case KVM_REG_PPC_TEXASR:
1421 		vcpu->arch.texasr = set_reg_val(id, *val);
1422 		break;
1423 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1424 		i = id - KVM_REG_PPC_TM_GPR0;
1425 		vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1426 		break;
1427 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1428 	{
1429 		int j;
1430 		i = id - KVM_REG_PPC_TM_VSR0;
1431 		if (i < 32)
1432 			for (j = 0; j < TS_FPRWIDTH; j++)
1433 				vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1434 		else
1435 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1436 				vcpu->arch.vr_tm.vr[i-32] = val->vval;
1437 			else
1438 				r = -ENXIO;
1439 		break;
1440 	}
1441 	case KVM_REG_PPC_TM_CR:
1442 		vcpu->arch.cr_tm = set_reg_val(id, *val);
1443 		break;
1444 	case KVM_REG_PPC_TM_LR:
1445 		vcpu->arch.lr_tm = set_reg_val(id, *val);
1446 		break;
1447 	case KVM_REG_PPC_TM_CTR:
1448 		vcpu->arch.ctr_tm = set_reg_val(id, *val);
1449 		break;
1450 	case KVM_REG_PPC_TM_FPSCR:
1451 		vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1452 		break;
1453 	case KVM_REG_PPC_TM_AMR:
1454 		vcpu->arch.amr_tm = set_reg_val(id, *val);
1455 		break;
1456 	case KVM_REG_PPC_TM_PPR:
1457 		vcpu->arch.ppr_tm = set_reg_val(id, *val);
1458 		break;
1459 	case KVM_REG_PPC_TM_VRSAVE:
1460 		vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1461 		break;
1462 	case KVM_REG_PPC_TM_VSCR:
1463 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1464 			vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1465 		else
1466 			r = - ENXIO;
1467 		break;
1468 	case KVM_REG_PPC_TM_DSCR:
1469 		vcpu->arch.dscr_tm = set_reg_val(id, *val);
1470 		break;
1471 	case KVM_REG_PPC_TM_TAR:
1472 		vcpu->arch.tar_tm = set_reg_val(id, *val);
1473 		break;
1474 #endif
1475 	case KVM_REG_PPC_ARCH_COMPAT:
1476 		r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1477 		break;
1478 	default:
1479 		r = -EINVAL;
1480 		break;
1481 	}
1482 
1483 	return r;
1484 }
1485 
1486 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1487 {
1488 	struct kvmppc_vcore *vcore;
1489 
1490 	vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1491 
1492 	if (vcore == NULL)
1493 		return NULL;
1494 
1495 	INIT_LIST_HEAD(&vcore->runnable_threads);
1496 	spin_lock_init(&vcore->lock);
1497 	spin_lock_init(&vcore->stoltb_lock);
1498 	init_swait_queue_head(&vcore->wq);
1499 	vcore->preempt_tb = TB_NIL;
1500 	vcore->lpcr = kvm->arch.lpcr;
1501 	vcore->first_vcpuid = core * threads_per_subcore;
1502 	vcore->kvm = kvm;
1503 	INIT_LIST_HEAD(&vcore->preempt_list);
1504 
1505 	return vcore;
1506 }
1507 
1508 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1509 static struct debugfs_timings_element {
1510 	const char *name;
1511 	size_t offset;
1512 } timings[] = {
1513 	{"rm_entry",	offsetof(struct kvm_vcpu, arch.rm_entry)},
1514 	{"rm_intr",	offsetof(struct kvm_vcpu, arch.rm_intr)},
1515 	{"rm_exit",	offsetof(struct kvm_vcpu, arch.rm_exit)},
1516 	{"guest",	offsetof(struct kvm_vcpu, arch.guest_time)},
1517 	{"cede",	offsetof(struct kvm_vcpu, arch.cede_time)},
1518 };
1519 
1520 #define N_TIMINGS	(sizeof(timings) / sizeof(timings[0]))
1521 
1522 struct debugfs_timings_state {
1523 	struct kvm_vcpu	*vcpu;
1524 	unsigned int	buflen;
1525 	char		buf[N_TIMINGS * 100];
1526 };
1527 
1528 static int debugfs_timings_open(struct inode *inode, struct file *file)
1529 {
1530 	struct kvm_vcpu *vcpu = inode->i_private;
1531 	struct debugfs_timings_state *p;
1532 
1533 	p = kzalloc(sizeof(*p), GFP_KERNEL);
1534 	if (!p)
1535 		return -ENOMEM;
1536 
1537 	kvm_get_kvm(vcpu->kvm);
1538 	p->vcpu = vcpu;
1539 	file->private_data = p;
1540 
1541 	return nonseekable_open(inode, file);
1542 }
1543 
1544 static int debugfs_timings_release(struct inode *inode, struct file *file)
1545 {
1546 	struct debugfs_timings_state *p = file->private_data;
1547 
1548 	kvm_put_kvm(p->vcpu->kvm);
1549 	kfree(p);
1550 	return 0;
1551 }
1552 
1553 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1554 				    size_t len, loff_t *ppos)
1555 {
1556 	struct debugfs_timings_state *p = file->private_data;
1557 	struct kvm_vcpu *vcpu = p->vcpu;
1558 	char *s, *buf_end;
1559 	struct kvmhv_tb_accumulator tb;
1560 	u64 count;
1561 	loff_t pos;
1562 	ssize_t n;
1563 	int i, loops;
1564 	bool ok;
1565 
1566 	if (!p->buflen) {
1567 		s = p->buf;
1568 		buf_end = s + sizeof(p->buf);
1569 		for (i = 0; i < N_TIMINGS; ++i) {
1570 			struct kvmhv_tb_accumulator *acc;
1571 
1572 			acc = (struct kvmhv_tb_accumulator *)
1573 				((unsigned long)vcpu + timings[i].offset);
1574 			ok = false;
1575 			for (loops = 0; loops < 1000; ++loops) {
1576 				count = acc->seqcount;
1577 				if (!(count & 1)) {
1578 					smp_rmb();
1579 					tb = *acc;
1580 					smp_rmb();
1581 					if (count == acc->seqcount) {
1582 						ok = true;
1583 						break;
1584 					}
1585 				}
1586 				udelay(1);
1587 			}
1588 			if (!ok)
1589 				snprintf(s, buf_end - s, "%s: stuck\n",
1590 					timings[i].name);
1591 			else
1592 				snprintf(s, buf_end - s,
1593 					"%s: %llu %llu %llu %llu\n",
1594 					timings[i].name, count / 2,
1595 					tb_to_ns(tb.tb_total),
1596 					tb_to_ns(tb.tb_min),
1597 					tb_to_ns(tb.tb_max));
1598 			s += strlen(s);
1599 		}
1600 		p->buflen = s - p->buf;
1601 	}
1602 
1603 	pos = *ppos;
1604 	if (pos >= p->buflen)
1605 		return 0;
1606 	if (len > p->buflen - pos)
1607 		len = p->buflen - pos;
1608 	n = copy_to_user(buf, p->buf + pos, len);
1609 	if (n) {
1610 		if (n == len)
1611 			return -EFAULT;
1612 		len -= n;
1613 	}
1614 	*ppos = pos + len;
1615 	return len;
1616 }
1617 
1618 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1619 				     size_t len, loff_t *ppos)
1620 {
1621 	return -EACCES;
1622 }
1623 
1624 static const struct file_operations debugfs_timings_ops = {
1625 	.owner	 = THIS_MODULE,
1626 	.open	 = debugfs_timings_open,
1627 	.release = debugfs_timings_release,
1628 	.read	 = debugfs_timings_read,
1629 	.write	 = debugfs_timings_write,
1630 	.llseek	 = generic_file_llseek,
1631 };
1632 
1633 /* Create a debugfs directory for the vcpu */
1634 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1635 {
1636 	char buf[16];
1637 	struct kvm *kvm = vcpu->kvm;
1638 
1639 	snprintf(buf, sizeof(buf), "vcpu%u", id);
1640 	if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1641 		return;
1642 	vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1643 	if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1644 		return;
1645 	vcpu->arch.debugfs_timings =
1646 		debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1647 				    vcpu, &debugfs_timings_ops);
1648 }
1649 
1650 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1651 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1652 {
1653 }
1654 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1655 
1656 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1657 						   unsigned int id)
1658 {
1659 	struct kvm_vcpu *vcpu;
1660 	int err = -EINVAL;
1661 	int core;
1662 	struct kvmppc_vcore *vcore;
1663 
1664 	core = id / threads_per_subcore;
1665 	if (core >= KVM_MAX_VCORES)
1666 		goto out;
1667 
1668 	err = -ENOMEM;
1669 	vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1670 	if (!vcpu)
1671 		goto out;
1672 
1673 	err = kvm_vcpu_init(vcpu, kvm, id);
1674 	if (err)
1675 		goto free_vcpu;
1676 
1677 	vcpu->arch.shared = &vcpu->arch.shregs;
1678 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1679 	/*
1680 	 * The shared struct is never shared on HV,
1681 	 * so we can always use host endianness
1682 	 */
1683 #ifdef __BIG_ENDIAN__
1684 	vcpu->arch.shared_big_endian = true;
1685 #else
1686 	vcpu->arch.shared_big_endian = false;
1687 #endif
1688 #endif
1689 	vcpu->arch.mmcr[0] = MMCR0_FC;
1690 	vcpu->arch.ctrl = CTRL_RUNLATCH;
1691 	/* default to host PVR, since we can't spoof it */
1692 	kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1693 	spin_lock_init(&vcpu->arch.vpa_update_lock);
1694 	spin_lock_init(&vcpu->arch.tbacct_lock);
1695 	vcpu->arch.busy_preempt = TB_NIL;
1696 	vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1697 
1698 	kvmppc_mmu_book3s_hv_init(vcpu);
1699 
1700 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1701 
1702 	init_waitqueue_head(&vcpu->arch.cpu_run);
1703 
1704 	mutex_lock(&kvm->lock);
1705 	vcore = kvm->arch.vcores[core];
1706 	if (!vcore) {
1707 		vcore = kvmppc_vcore_create(kvm, core);
1708 		kvm->arch.vcores[core] = vcore;
1709 		kvm->arch.online_vcores++;
1710 	}
1711 	mutex_unlock(&kvm->lock);
1712 
1713 	if (!vcore)
1714 		goto free_vcpu;
1715 
1716 	spin_lock(&vcore->lock);
1717 	++vcore->num_threads;
1718 	spin_unlock(&vcore->lock);
1719 	vcpu->arch.vcore = vcore;
1720 	vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1721 	vcpu->arch.thread_cpu = -1;
1722 
1723 	vcpu->arch.cpu_type = KVM_CPU_3S_64;
1724 	kvmppc_sanity_check(vcpu);
1725 
1726 	debugfs_vcpu_init(vcpu, id);
1727 
1728 	return vcpu;
1729 
1730 free_vcpu:
1731 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1732 out:
1733 	return ERR_PTR(err);
1734 }
1735 
1736 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1737 {
1738 	if (vpa->pinned_addr)
1739 		kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1740 					vpa->dirty);
1741 }
1742 
1743 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
1744 {
1745 	spin_lock(&vcpu->arch.vpa_update_lock);
1746 	unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1747 	unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1748 	unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1749 	spin_unlock(&vcpu->arch.vpa_update_lock);
1750 	kvm_vcpu_uninit(vcpu);
1751 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1752 }
1753 
1754 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1755 {
1756 	/* Indicate we want to get back into the guest */
1757 	return 1;
1758 }
1759 
1760 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1761 {
1762 	unsigned long dec_nsec, now;
1763 
1764 	now = get_tb();
1765 	if (now > vcpu->arch.dec_expires) {
1766 		/* decrementer has already gone negative */
1767 		kvmppc_core_queue_dec(vcpu);
1768 		kvmppc_core_prepare_to_enter(vcpu);
1769 		return;
1770 	}
1771 	dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1772 		   / tb_ticks_per_sec;
1773 	hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1774 		      HRTIMER_MODE_REL);
1775 	vcpu->arch.timer_running = 1;
1776 }
1777 
1778 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1779 {
1780 	vcpu->arch.ceded = 0;
1781 	if (vcpu->arch.timer_running) {
1782 		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1783 		vcpu->arch.timer_running = 0;
1784 	}
1785 }
1786 
1787 extern void __kvmppc_vcore_entry(void);
1788 
1789 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1790 				   struct kvm_vcpu *vcpu)
1791 {
1792 	u64 now;
1793 
1794 	if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1795 		return;
1796 	spin_lock_irq(&vcpu->arch.tbacct_lock);
1797 	now = mftb();
1798 	vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1799 		vcpu->arch.stolen_logged;
1800 	vcpu->arch.busy_preempt = now;
1801 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1802 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
1803 	--vc->n_runnable;
1804 	list_del(&vcpu->arch.run_list);
1805 }
1806 
1807 static int kvmppc_grab_hwthread(int cpu)
1808 {
1809 	struct paca_struct *tpaca;
1810 	long timeout = 10000;
1811 
1812 	tpaca = &paca[cpu];
1813 
1814 	/* Ensure the thread won't go into the kernel if it wakes */
1815 	tpaca->kvm_hstate.kvm_vcpu = NULL;
1816 	tpaca->kvm_hstate.kvm_vcore = NULL;
1817 	tpaca->kvm_hstate.napping = 0;
1818 	smp_wmb();
1819 	tpaca->kvm_hstate.hwthread_req = 1;
1820 
1821 	/*
1822 	 * If the thread is already executing in the kernel (e.g. handling
1823 	 * a stray interrupt), wait for it to get back to nap mode.
1824 	 * The smp_mb() is to ensure that our setting of hwthread_req
1825 	 * is visible before we look at hwthread_state, so if this
1826 	 * races with the code at system_reset_pSeries and the thread
1827 	 * misses our setting of hwthread_req, we are sure to see its
1828 	 * setting of hwthread_state, and vice versa.
1829 	 */
1830 	smp_mb();
1831 	while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1832 		if (--timeout <= 0) {
1833 			pr_err("KVM: couldn't grab cpu %d\n", cpu);
1834 			return -EBUSY;
1835 		}
1836 		udelay(1);
1837 	}
1838 	return 0;
1839 }
1840 
1841 static void kvmppc_release_hwthread(int cpu)
1842 {
1843 	struct paca_struct *tpaca;
1844 
1845 	tpaca = &paca[cpu];
1846 	tpaca->kvm_hstate.hwthread_req = 0;
1847 	tpaca->kvm_hstate.kvm_vcpu = NULL;
1848 	tpaca->kvm_hstate.kvm_vcore = NULL;
1849 	tpaca->kvm_hstate.kvm_split_mode = NULL;
1850 }
1851 
1852 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
1853 {
1854 	int cpu;
1855 	struct paca_struct *tpaca;
1856 	struct kvmppc_vcore *mvc = vc->master_vcore;
1857 
1858 	cpu = vc->pcpu;
1859 	if (vcpu) {
1860 		if (vcpu->arch.timer_running) {
1861 			hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1862 			vcpu->arch.timer_running = 0;
1863 		}
1864 		cpu += vcpu->arch.ptid;
1865 		vcpu->cpu = mvc->pcpu;
1866 		vcpu->arch.thread_cpu = cpu;
1867 	}
1868 	tpaca = &paca[cpu];
1869 	tpaca->kvm_hstate.kvm_vcpu = vcpu;
1870 	tpaca->kvm_hstate.ptid = cpu - mvc->pcpu;
1871 	/* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
1872 	smp_wmb();
1873 	tpaca->kvm_hstate.kvm_vcore = mvc;
1874 	if (cpu != smp_processor_id())
1875 		kvmppc_ipi_thread(cpu);
1876 }
1877 
1878 static void kvmppc_wait_for_nap(void)
1879 {
1880 	int cpu = smp_processor_id();
1881 	int i, loops;
1882 
1883 	for (loops = 0; loops < 1000000; ++loops) {
1884 		/*
1885 		 * Check if all threads are finished.
1886 		 * We set the vcore pointer when starting a thread
1887 		 * and the thread clears it when finished, so we look
1888 		 * for any threads that still have a non-NULL vcore ptr.
1889 		 */
1890 		for (i = 1; i < threads_per_subcore; ++i)
1891 			if (paca[cpu + i].kvm_hstate.kvm_vcore)
1892 				break;
1893 		if (i == threads_per_subcore) {
1894 			HMT_medium();
1895 			return;
1896 		}
1897 		HMT_low();
1898 	}
1899 	HMT_medium();
1900 	for (i = 1; i < threads_per_subcore; ++i)
1901 		if (paca[cpu + i].kvm_hstate.kvm_vcore)
1902 			pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
1903 }
1904 
1905 /*
1906  * Check that we are on thread 0 and that any other threads in
1907  * this core are off-line.  Then grab the threads so they can't
1908  * enter the kernel.
1909  */
1910 static int on_primary_thread(void)
1911 {
1912 	int cpu = smp_processor_id();
1913 	int thr;
1914 
1915 	/* Are we on a primary subcore? */
1916 	if (cpu_thread_in_subcore(cpu))
1917 		return 0;
1918 
1919 	thr = 0;
1920 	while (++thr < threads_per_subcore)
1921 		if (cpu_online(cpu + thr))
1922 			return 0;
1923 
1924 	/* Grab all hw threads so they can't go into the kernel */
1925 	for (thr = 1; thr < threads_per_subcore; ++thr) {
1926 		if (kvmppc_grab_hwthread(cpu + thr)) {
1927 			/* Couldn't grab one; let the others go */
1928 			do {
1929 				kvmppc_release_hwthread(cpu + thr);
1930 			} while (--thr > 0);
1931 			return 0;
1932 		}
1933 	}
1934 	return 1;
1935 }
1936 
1937 /*
1938  * A list of virtual cores for each physical CPU.
1939  * These are vcores that could run but their runner VCPU tasks are
1940  * (or may be) preempted.
1941  */
1942 struct preempted_vcore_list {
1943 	struct list_head	list;
1944 	spinlock_t		lock;
1945 };
1946 
1947 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
1948 
1949 static void init_vcore_lists(void)
1950 {
1951 	int cpu;
1952 
1953 	for_each_possible_cpu(cpu) {
1954 		struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
1955 		spin_lock_init(&lp->lock);
1956 		INIT_LIST_HEAD(&lp->list);
1957 	}
1958 }
1959 
1960 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
1961 {
1962 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
1963 
1964 	vc->vcore_state = VCORE_PREEMPT;
1965 	vc->pcpu = smp_processor_id();
1966 	if (vc->num_threads < threads_per_subcore) {
1967 		spin_lock(&lp->lock);
1968 		list_add_tail(&vc->preempt_list, &lp->list);
1969 		spin_unlock(&lp->lock);
1970 	}
1971 
1972 	/* Start accumulating stolen time */
1973 	kvmppc_core_start_stolen(vc);
1974 }
1975 
1976 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
1977 {
1978 	struct preempted_vcore_list *lp;
1979 
1980 	kvmppc_core_end_stolen(vc);
1981 	if (!list_empty(&vc->preempt_list)) {
1982 		lp = &per_cpu(preempted_vcores, vc->pcpu);
1983 		spin_lock(&lp->lock);
1984 		list_del_init(&vc->preempt_list);
1985 		spin_unlock(&lp->lock);
1986 	}
1987 	vc->vcore_state = VCORE_INACTIVE;
1988 }
1989 
1990 /*
1991  * This stores information about the virtual cores currently
1992  * assigned to a physical core.
1993  */
1994 struct core_info {
1995 	int		n_subcores;
1996 	int		max_subcore_threads;
1997 	int		total_threads;
1998 	int		subcore_threads[MAX_SUBCORES];
1999 	struct kvm	*subcore_vm[MAX_SUBCORES];
2000 	struct list_head vcs[MAX_SUBCORES];
2001 };
2002 
2003 /*
2004  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2005  * respectively in 2-way micro-threading (split-core) mode.
2006  */
2007 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2008 
2009 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2010 {
2011 	int sub;
2012 
2013 	memset(cip, 0, sizeof(*cip));
2014 	cip->n_subcores = 1;
2015 	cip->max_subcore_threads = vc->num_threads;
2016 	cip->total_threads = vc->num_threads;
2017 	cip->subcore_threads[0] = vc->num_threads;
2018 	cip->subcore_vm[0] = vc->kvm;
2019 	for (sub = 0; sub < MAX_SUBCORES; ++sub)
2020 		INIT_LIST_HEAD(&cip->vcs[sub]);
2021 	list_add_tail(&vc->preempt_list, &cip->vcs[0]);
2022 }
2023 
2024 static bool subcore_config_ok(int n_subcores, int n_threads)
2025 {
2026 	/* Can only dynamically split if unsplit to begin with */
2027 	if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2028 		return false;
2029 	if (n_subcores > MAX_SUBCORES)
2030 		return false;
2031 	if (n_subcores > 1) {
2032 		if (!(dynamic_mt_modes & 2))
2033 			n_subcores = 4;
2034 		if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2035 			return false;
2036 	}
2037 
2038 	return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2039 }
2040 
2041 static void init_master_vcore(struct kvmppc_vcore *vc)
2042 {
2043 	vc->master_vcore = vc;
2044 	vc->entry_exit_map = 0;
2045 	vc->in_guest = 0;
2046 	vc->napping_threads = 0;
2047 	vc->conferring_threads = 0;
2048 }
2049 
2050 /*
2051  * See if the existing subcores can be split into 3 (or fewer) subcores
2052  * of at most two threads each, so we can fit in another vcore.  This
2053  * assumes there are at most two subcores and at most 6 threads in total.
2054  */
2055 static bool can_split_piggybacked_subcores(struct core_info *cip)
2056 {
2057 	int sub, new_sub;
2058 	int large_sub = -1;
2059 	int thr;
2060 	int n_subcores = cip->n_subcores;
2061 	struct kvmppc_vcore *vc, *vcnext;
2062 	struct kvmppc_vcore *master_vc = NULL;
2063 
2064 	for (sub = 0; sub < cip->n_subcores; ++sub) {
2065 		if (cip->subcore_threads[sub] <= 2)
2066 			continue;
2067 		if (large_sub >= 0)
2068 			return false;
2069 		large_sub = sub;
2070 		vc = list_first_entry(&cip->vcs[sub], struct kvmppc_vcore,
2071 				      preempt_list);
2072 		if (vc->num_threads > 2)
2073 			return false;
2074 		n_subcores += (cip->subcore_threads[sub] - 1) >> 1;
2075 	}
2076 	if (large_sub < 0 || !subcore_config_ok(n_subcores + 1, 2))
2077 		return false;
2078 
2079 	/*
2080 	 * Seems feasible, so go through and move vcores to new subcores.
2081 	 * Note that when we have two or more vcores in one subcore,
2082 	 * all those vcores must have only one thread each.
2083 	 */
2084 	new_sub = cip->n_subcores;
2085 	thr = 0;
2086 	sub = large_sub;
2087 	list_for_each_entry_safe(vc, vcnext, &cip->vcs[sub], preempt_list) {
2088 		if (thr >= 2) {
2089 			list_del(&vc->preempt_list);
2090 			list_add_tail(&vc->preempt_list, &cip->vcs[new_sub]);
2091 			/* vc->num_threads must be 1 */
2092 			if (++cip->subcore_threads[new_sub] == 1) {
2093 				cip->subcore_vm[new_sub] = vc->kvm;
2094 				init_master_vcore(vc);
2095 				master_vc = vc;
2096 				++cip->n_subcores;
2097 			} else {
2098 				vc->master_vcore = master_vc;
2099 				++new_sub;
2100 			}
2101 		}
2102 		thr += vc->num_threads;
2103 	}
2104 	cip->subcore_threads[large_sub] = 2;
2105 	cip->max_subcore_threads = 2;
2106 
2107 	return true;
2108 }
2109 
2110 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2111 {
2112 	int n_threads = vc->num_threads;
2113 	int sub;
2114 
2115 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2116 		return false;
2117 
2118 	if (n_threads < cip->max_subcore_threads)
2119 		n_threads = cip->max_subcore_threads;
2120 	if (subcore_config_ok(cip->n_subcores + 1, n_threads)) {
2121 		cip->max_subcore_threads = n_threads;
2122 	} else if (cip->n_subcores <= 2 && cip->total_threads <= 6 &&
2123 		   vc->num_threads <= 2) {
2124 		/*
2125 		 * We may be able to fit another subcore in by
2126 		 * splitting an existing subcore with 3 or 4
2127 		 * threads into two 2-thread subcores, or one
2128 		 * with 5 or 6 threads into three subcores.
2129 		 * We can only do this if those subcores have
2130 		 * piggybacked virtual cores.
2131 		 */
2132 		if (!can_split_piggybacked_subcores(cip))
2133 			return false;
2134 	} else {
2135 		return false;
2136 	}
2137 
2138 	sub = cip->n_subcores;
2139 	++cip->n_subcores;
2140 	cip->total_threads += vc->num_threads;
2141 	cip->subcore_threads[sub] = vc->num_threads;
2142 	cip->subcore_vm[sub] = vc->kvm;
2143 	init_master_vcore(vc);
2144 	list_del(&vc->preempt_list);
2145 	list_add_tail(&vc->preempt_list, &cip->vcs[sub]);
2146 
2147 	return true;
2148 }
2149 
2150 static bool can_piggyback_subcore(struct kvmppc_vcore *pvc,
2151 				  struct core_info *cip, int sub)
2152 {
2153 	struct kvmppc_vcore *vc;
2154 	int n_thr;
2155 
2156 	vc = list_first_entry(&cip->vcs[sub], struct kvmppc_vcore,
2157 			      preempt_list);
2158 
2159 	/* require same VM and same per-core reg values */
2160 	if (pvc->kvm != vc->kvm ||
2161 	    pvc->tb_offset != vc->tb_offset ||
2162 	    pvc->pcr != vc->pcr ||
2163 	    pvc->lpcr != vc->lpcr)
2164 		return false;
2165 
2166 	/* P8 guest with > 1 thread per core would see wrong TIR value */
2167 	if (cpu_has_feature(CPU_FTR_ARCH_207S) &&
2168 	    (vc->num_threads > 1 || pvc->num_threads > 1))
2169 		return false;
2170 
2171 	n_thr = cip->subcore_threads[sub] + pvc->num_threads;
2172 	if (n_thr > cip->max_subcore_threads) {
2173 		if (!subcore_config_ok(cip->n_subcores, n_thr))
2174 			return false;
2175 		cip->max_subcore_threads = n_thr;
2176 	}
2177 
2178 	cip->total_threads += pvc->num_threads;
2179 	cip->subcore_threads[sub] = n_thr;
2180 	pvc->master_vcore = vc;
2181 	list_del(&pvc->preempt_list);
2182 	list_add_tail(&pvc->preempt_list, &cip->vcs[sub]);
2183 
2184 	return true;
2185 }
2186 
2187 /*
2188  * Work out whether it is possible to piggyback the execution of
2189  * vcore *pvc onto the execution of the other vcores described in *cip.
2190  */
2191 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2192 			  int target_threads)
2193 {
2194 	int sub;
2195 
2196 	if (cip->total_threads + pvc->num_threads > target_threads)
2197 		return false;
2198 	for (sub = 0; sub < cip->n_subcores; ++sub)
2199 		if (cip->subcore_threads[sub] &&
2200 		    can_piggyback_subcore(pvc, cip, sub))
2201 			return true;
2202 
2203 	if (can_dynamic_split(pvc, cip))
2204 		return true;
2205 
2206 	return false;
2207 }
2208 
2209 static void prepare_threads(struct kvmppc_vcore *vc)
2210 {
2211 	struct kvm_vcpu *vcpu, *vnext;
2212 
2213 	list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2214 				 arch.run_list) {
2215 		if (signal_pending(vcpu->arch.run_task))
2216 			vcpu->arch.ret = -EINTR;
2217 		else if (vcpu->arch.vpa.update_pending ||
2218 			 vcpu->arch.slb_shadow.update_pending ||
2219 			 vcpu->arch.dtl.update_pending)
2220 			vcpu->arch.ret = RESUME_GUEST;
2221 		else
2222 			continue;
2223 		kvmppc_remove_runnable(vc, vcpu);
2224 		wake_up(&vcpu->arch.cpu_run);
2225 	}
2226 }
2227 
2228 static void collect_piggybacks(struct core_info *cip, int target_threads)
2229 {
2230 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2231 	struct kvmppc_vcore *pvc, *vcnext;
2232 
2233 	spin_lock(&lp->lock);
2234 	list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2235 		if (!spin_trylock(&pvc->lock))
2236 			continue;
2237 		prepare_threads(pvc);
2238 		if (!pvc->n_runnable) {
2239 			list_del_init(&pvc->preempt_list);
2240 			if (pvc->runner == NULL) {
2241 				pvc->vcore_state = VCORE_INACTIVE;
2242 				kvmppc_core_end_stolen(pvc);
2243 			}
2244 			spin_unlock(&pvc->lock);
2245 			continue;
2246 		}
2247 		if (!can_piggyback(pvc, cip, target_threads)) {
2248 			spin_unlock(&pvc->lock);
2249 			continue;
2250 		}
2251 		kvmppc_core_end_stolen(pvc);
2252 		pvc->vcore_state = VCORE_PIGGYBACK;
2253 		if (cip->total_threads >= target_threads)
2254 			break;
2255 	}
2256 	spin_unlock(&lp->lock);
2257 }
2258 
2259 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2260 {
2261 	int still_running = 0;
2262 	u64 now;
2263 	long ret;
2264 	struct kvm_vcpu *vcpu, *vnext;
2265 
2266 	spin_lock(&vc->lock);
2267 	now = get_tb();
2268 	list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2269 				 arch.run_list) {
2270 		/* cancel pending dec exception if dec is positive */
2271 		if (now < vcpu->arch.dec_expires &&
2272 		    kvmppc_core_pending_dec(vcpu))
2273 			kvmppc_core_dequeue_dec(vcpu);
2274 
2275 		trace_kvm_guest_exit(vcpu);
2276 
2277 		ret = RESUME_GUEST;
2278 		if (vcpu->arch.trap)
2279 			ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2280 						    vcpu->arch.run_task);
2281 
2282 		vcpu->arch.ret = ret;
2283 		vcpu->arch.trap = 0;
2284 
2285 		if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2286 			if (vcpu->arch.pending_exceptions)
2287 				kvmppc_core_prepare_to_enter(vcpu);
2288 			if (vcpu->arch.ceded)
2289 				kvmppc_set_timer(vcpu);
2290 			else
2291 				++still_running;
2292 		} else {
2293 			kvmppc_remove_runnable(vc, vcpu);
2294 			wake_up(&vcpu->arch.cpu_run);
2295 		}
2296 	}
2297 	list_del_init(&vc->preempt_list);
2298 	if (!is_master) {
2299 		if (still_running > 0) {
2300 			kvmppc_vcore_preempt(vc);
2301 		} else if (vc->runner) {
2302 			vc->vcore_state = VCORE_PREEMPT;
2303 			kvmppc_core_start_stolen(vc);
2304 		} else {
2305 			vc->vcore_state = VCORE_INACTIVE;
2306 		}
2307 		if (vc->n_runnable > 0 && vc->runner == NULL) {
2308 			/* make sure there's a candidate runner awake */
2309 			vcpu = list_first_entry(&vc->runnable_threads,
2310 						struct kvm_vcpu, arch.run_list);
2311 			wake_up(&vcpu->arch.cpu_run);
2312 		}
2313 	}
2314 	spin_unlock(&vc->lock);
2315 }
2316 
2317 /*
2318  * Clear core from the list of active host cores as we are about to
2319  * enter the guest. Only do this if it is the primary thread of the
2320  * core (not if a subcore) that is entering the guest.
2321  */
2322 static inline void kvmppc_clear_host_core(int cpu)
2323 {
2324 	int core;
2325 
2326 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2327 		return;
2328 	/*
2329 	 * Memory barrier can be omitted here as we will do a smp_wmb()
2330 	 * later in kvmppc_start_thread and we need ensure that state is
2331 	 * visible to other CPUs only after we enter guest.
2332 	 */
2333 	core = cpu >> threads_shift;
2334 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2335 }
2336 
2337 /*
2338  * Advertise this core as an active host core since we exited the guest
2339  * Only need to do this if it is the primary thread of the core that is
2340  * exiting.
2341  */
2342 static inline void kvmppc_set_host_core(int cpu)
2343 {
2344 	int core;
2345 
2346 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2347 		return;
2348 
2349 	/*
2350 	 * Memory barrier can be omitted here because we do a spin_unlock
2351 	 * immediately after this which provides the memory barrier.
2352 	 */
2353 	core = cpu >> threads_shift;
2354 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2355 }
2356 
2357 /*
2358  * Run a set of guest threads on a physical core.
2359  * Called with vc->lock held.
2360  */
2361 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2362 {
2363 	struct kvm_vcpu *vcpu, *vnext;
2364 	int i;
2365 	int srcu_idx;
2366 	struct core_info core_info;
2367 	struct kvmppc_vcore *pvc, *vcnext;
2368 	struct kvm_split_mode split_info, *sip;
2369 	int split, subcore_size, active;
2370 	int sub;
2371 	bool thr0_done;
2372 	unsigned long cmd_bit, stat_bit;
2373 	int pcpu, thr;
2374 	int target_threads;
2375 
2376 	/*
2377 	 * Remove from the list any threads that have a signal pending
2378 	 * or need a VPA update done
2379 	 */
2380 	prepare_threads(vc);
2381 
2382 	/* if the runner is no longer runnable, let the caller pick a new one */
2383 	if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2384 		return;
2385 
2386 	/*
2387 	 * Initialize *vc.
2388 	 */
2389 	init_master_vcore(vc);
2390 	vc->preempt_tb = TB_NIL;
2391 
2392 	/*
2393 	 * Make sure we are running on primary threads, and that secondary
2394 	 * threads are offline.  Also check if the number of threads in this
2395 	 * guest are greater than the current system threads per guest.
2396 	 */
2397 	if ((threads_per_core > 1) &&
2398 	    ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
2399 		list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2400 					 arch.run_list) {
2401 			vcpu->arch.ret = -EBUSY;
2402 			kvmppc_remove_runnable(vc, vcpu);
2403 			wake_up(&vcpu->arch.cpu_run);
2404 		}
2405 		goto out;
2406 	}
2407 
2408 	/*
2409 	 * See if we could run any other vcores on the physical core
2410 	 * along with this one.
2411 	 */
2412 	init_core_info(&core_info, vc);
2413 	pcpu = smp_processor_id();
2414 	target_threads = threads_per_subcore;
2415 	if (target_smt_mode && target_smt_mode < target_threads)
2416 		target_threads = target_smt_mode;
2417 	if (vc->num_threads < target_threads)
2418 		collect_piggybacks(&core_info, target_threads);
2419 
2420 	/* Decide on micro-threading (split-core) mode */
2421 	subcore_size = threads_per_subcore;
2422 	cmd_bit = stat_bit = 0;
2423 	split = core_info.n_subcores;
2424 	sip = NULL;
2425 	if (split > 1) {
2426 		/* threads_per_subcore must be MAX_SMT_THREADS (8) here */
2427 		if (split == 2 && (dynamic_mt_modes & 2)) {
2428 			cmd_bit = HID0_POWER8_1TO2LPAR;
2429 			stat_bit = HID0_POWER8_2LPARMODE;
2430 		} else {
2431 			split = 4;
2432 			cmd_bit = HID0_POWER8_1TO4LPAR;
2433 			stat_bit = HID0_POWER8_4LPARMODE;
2434 		}
2435 		subcore_size = MAX_SMT_THREADS / split;
2436 		sip = &split_info;
2437 		memset(&split_info, 0, sizeof(split_info));
2438 		split_info.rpr = mfspr(SPRN_RPR);
2439 		split_info.pmmar = mfspr(SPRN_PMMAR);
2440 		split_info.ldbar = mfspr(SPRN_LDBAR);
2441 		split_info.subcore_size = subcore_size;
2442 		for (sub = 0; sub < core_info.n_subcores; ++sub)
2443 			split_info.master_vcs[sub] =
2444 				list_first_entry(&core_info.vcs[sub],
2445 					struct kvmppc_vcore, preempt_list);
2446 		/* order writes to split_info before kvm_split_mode pointer */
2447 		smp_wmb();
2448 	}
2449 	pcpu = smp_processor_id();
2450 	for (thr = 0; thr < threads_per_subcore; ++thr)
2451 		paca[pcpu + thr].kvm_hstate.kvm_split_mode = sip;
2452 
2453 	/* Initiate micro-threading (split-core) if required */
2454 	if (cmd_bit) {
2455 		unsigned long hid0 = mfspr(SPRN_HID0);
2456 
2457 		hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2458 		mb();
2459 		mtspr(SPRN_HID0, hid0);
2460 		isync();
2461 		for (;;) {
2462 			hid0 = mfspr(SPRN_HID0);
2463 			if (hid0 & stat_bit)
2464 				break;
2465 			cpu_relax();
2466 		}
2467 	}
2468 
2469 	kvmppc_clear_host_core(pcpu);
2470 
2471 	/* Start all the threads */
2472 	active = 0;
2473 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
2474 		thr = subcore_thread_map[sub];
2475 		thr0_done = false;
2476 		active |= 1 << thr;
2477 		list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list) {
2478 			pvc->pcpu = pcpu + thr;
2479 			list_for_each_entry(vcpu, &pvc->runnable_threads,
2480 					    arch.run_list) {
2481 				kvmppc_start_thread(vcpu, pvc);
2482 				kvmppc_create_dtl_entry(vcpu, pvc);
2483 				trace_kvm_guest_enter(vcpu);
2484 				if (!vcpu->arch.ptid)
2485 					thr0_done = true;
2486 				active |= 1 << (thr + vcpu->arch.ptid);
2487 			}
2488 			/*
2489 			 * We need to start the first thread of each subcore
2490 			 * even if it doesn't have a vcpu.
2491 			 */
2492 			if (pvc->master_vcore == pvc && !thr0_done)
2493 				kvmppc_start_thread(NULL, pvc);
2494 			thr += pvc->num_threads;
2495 		}
2496 	}
2497 
2498 	/*
2499 	 * Ensure that split_info.do_nap is set after setting
2500 	 * the vcore pointer in the PACA of the secondaries.
2501 	 */
2502 	smp_mb();
2503 	if (cmd_bit)
2504 		split_info.do_nap = 1;	/* ask secondaries to nap when done */
2505 
2506 	/*
2507 	 * When doing micro-threading, poke the inactive threads as well.
2508 	 * This gets them to the nap instruction after kvm_do_nap,
2509 	 * which reduces the time taken to unsplit later.
2510 	 */
2511 	if (split > 1)
2512 		for (thr = 1; thr < threads_per_subcore; ++thr)
2513 			if (!(active & (1 << thr)))
2514 				kvmppc_ipi_thread(pcpu + thr);
2515 
2516 	vc->vcore_state = VCORE_RUNNING;
2517 	preempt_disable();
2518 
2519 	trace_kvmppc_run_core(vc, 0);
2520 
2521 	for (sub = 0; sub < core_info.n_subcores; ++sub)
2522 		list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list)
2523 			spin_unlock(&pvc->lock);
2524 
2525 	kvm_guest_enter();
2526 
2527 	srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2528 
2529 	__kvmppc_vcore_entry();
2530 
2531 	srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2532 
2533 	spin_lock(&vc->lock);
2534 	/* prevent other vcpu threads from doing kvmppc_start_thread() now */
2535 	vc->vcore_state = VCORE_EXITING;
2536 
2537 	/* wait for secondary threads to finish writing their state to memory */
2538 	kvmppc_wait_for_nap();
2539 
2540 	/* Return to whole-core mode if we split the core earlier */
2541 	if (split > 1) {
2542 		unsigned long hid0 = mfspr(SPRN_HID0);
2543 		unsigned long loops = 0;
2544 
2545 		hid0 &= ~HID0_POWER8_DYNLPARDIS;
2546 		stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
2547 		mb();
2548 		mtspr(SPRN_HID0, hid0);
2549 		isync();
2550 		for (;;) {
2551 			hid0 = mfspr(SPRN_HID0);
2552 			if (!(hid0 & stat_bit))
2553 				break;
2554 			cpu_relax();
2555 			++loops;
2556 		}
2557 		split_info.do_nap = 0;
2558 	}
2559 
2560 	/* Let secondaries go back to the offline loop */
2561 	for (i = 0; i < threads_per_subcore; ++i) {
2562 		kvmppc_release_hwthread(pcpu + i);
2563 		if (sip && sip->napped[i])
2564 			kvmppc_ipi_thread(pcpu + i);
2565 	}
2566 
2567 	kvmppc_set_host_core(pcpu);
2568 
2569 	spin_unlock(&vc->lock);
2570 
2571 	/* make sure updates to secondary vcpu structs are visible now */
2572 	smp_mb();
2573 	kvm_guest_exit();
2574 
2575 	for (sub = 0; sub < core_info.n_subcores; ++sub)
2576 		list_for_each_entry_safe(pvc, vcnext, &core_info.vcs[sub],
2577 					 preempt_list)
2578 			post_guest_process(pvc, pvc == vc);
2579 
2580 	spin_lock(&vc->lock);
2581 	preempt_enable();
2582 
2583  out:
2584 	vc->vcore_state = VCORE_INACTIVE;
2585 	trace_kvmppc_run_core(vc, 1);
2586 }
2587 
2588 /*
2589  * Wait for some other vcpu thread to execute us, and
2590  * wake us up when we need to handle something in the host.
2591  */
2592 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
2593 				 struct kvm_vcpu *vcpu, int wait_state)
2594 {
2595 	DEFINE_WAIT(wait);
2596 
2597 	prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
2598 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2599 		spin_unlock(&vc->lock);
2600 		schedule();
2601 		spin_lock(&vc->lock);
2602 	}
2603 	finish_wait(&vcpu->arch.cpu_run, &wait);
2604 }
2605 
2606 /*
2607  * All the vcpus in this vcore are idle, so wait for a decrementer
2608  * or external interrupt to one of the vcpus.  vc->lock is held.
2609  */
2610 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
2611 {
2612 	struct kvm_vcpu *vcpu;
2613 	int do_sleep = 1;
2614 	DECLARE_SWAITQUEUE(wait);
2615 
2616 	prepare_to_swait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
2617 
2618 	/*
2619 	 * Check one last time for pending exceptions and ceded state after
2620 	 * we put ourselves on the wait queue
2621 	 */
2622 	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
2623 		if (vcpu->arch.pending_exceptions || !vcpu->arch.ceded) {
2624 			do_sleep = 0;
2625 			break;
2626 		}
2627 	}
2628 
2629 	if (!do_sleep) {
2630 		finish_swait(&vc->wq, &wait);
2631 		return;
2632 	}
2633 
2634 	vc->vcore_state = VCORE_SLEEPING;
2635 	trace_kvmppc_vcore_blocked(vc, 0);
2636 	spin_unlock(&vc->lock);
2637 	schedule();
2638 	finish_swait(&vc->wq, &wait);
2639 	spin_lock(&vc->lock);
2640 	vc->vcore_state = VCORE_INACTIVE;
2641 	trace_kvmppc_vcore_blocked(vc, 1);
2642 }
2643 
2644 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2645 {
2646 	int n_ceded;
2647 	struct kvmppc_vcore *vc;
2648 	struct kvm_vcpu *v, *vn;
2649 
2650 	trace_kvmppc_run_vcpu_enter(vcpu);
2651 
2652 	kvm_run->exit_reason = 0;
2653 	vcpu->arch.ret = RESUME_GUEST;
2654 	vcpu->arch.trap = 0;
2655 	kvmppc_update_vpas(vcpu);
2656 
2657 	/*
2658 	 * Synchronize with other threads in this virtual core
2659 	 */
2660 	vc = vcpu->arch.vcore;
2661 	spin_lock(&vc->lock);
2662 	vcpu->arch.ceded = 0;
2663 	vcpu->arch.run_task = current;
2664 	vcpu->arch.kvm_run = kvm_run;
2665 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
2666 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
2667 	vcpu->arch.busy_preempt = TB_NIL;
2668 	list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
2669 	++vc->n_runnable;
2670 
2671 	/*
2672 	 * This happens the first time this is called for a vcpu.
2673 	 * If the vcore is already running, we may be able to start
2674 	 * this thread straight away and have it join in.
2675 	 */
2676 	if (!signal_pending(current)) {
2677 		if (vc->vcore_state == VCORE_PIGGYBACK) {
2678 			struct kvmppc_vcore *mvc = vc->master_vcore;
2679 			if (spin_trylock(&mvc->lock)) {
2680 				if (mvc->vcore_state == VCORE_RUNNING &&
2681 				    !VCORE_IS_EXITING(mvc)) {
2682 					kvmppc_create_dtl_entry(vcpu, vc);
2683 					kvmppc_start_thread(vcpu, vc);
2684 					trace_kvm_guest_enter(vcpu);
2685 				}
2686 				spin_unlock(&mvc->lock);
2687 			}
2688 		} else if (vc->vcore_state == VCORE_RUNNING &&
2689 			   !VCORE_IS_EXITING(vc)) {
2690 			kvmppc_create_dtl_entry(vcpu, vc);
2691 			kvmppc_start_thread(vcpu, vc);
2692 			trace_kvm_guest_enter(vcpu);
2693 		} else if (vc->vcore_state == VCORE_SLEEPING) {
2694 			swake_up(&vc->wq);
2695 		}
2696 
2697 	}
2698 
2699 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2700 	       !signal_pending(current)) {
2701 		if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2702 			kvmppc_vcore_end_preempt(vc);
2703 
2704 		if (vc->vcore_state != VCORE_INACTIVE) {
2705 			kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
2706 			continue;
2707 		}
2708 		list_for_each_entry_safe(v, vn, &vc->runnable_threads,
2709 					 arch.run_list) {
2710 			kvmppc_core_prepare_to_enter(v);
2711 			if (signal_pending(v->arch.run_task)) {
2712 				kvmppc_remove_runnable(vc, v);
2713 				v->stat.signal_exits++;
2714 				v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
2715 				v->arch.ret = -EINTR;
2716 				wake_up(&v->arch.cpu_run);
2717 			}
2718 		}
2719 		if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2720 			break;
2721 		n_ceded = 0;
2722 		list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
2723 			if (!v->arch.pending_exceptions)
2724 				n_ceded += v->arch.ceded;
2725 			else
2726 				v->arch.ceded = 0;
2727 		}
2728 		vc->runner = vcpu;
2729 		if (n_ceded == vc->n_runnable) {
2730 			kvmppc_vcore_blocked(vc);
2731 		} else if (need_resched()) {
2732 			kvmppc_vcore_preempt(vc);
2733 			/* Let something else run */
2734 			cond_resched_lock(&vc->lock);
2735 			if (vc->vcore_state == VCORE_PREEMPT)
2736 				kvmppc_vcore_end_preempt(vc);
2737 		} else {
2738 			kvmppc_run_core(vc);
2739 		}
2740 		vc->runner = NULL;
2741 	}
2742 
2743 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2744 	       (vc->vcore_state == VCORE_RUNNING ||
2745 		vc->vcore_state == VCORE_EXITING ||
2746 		vc->vcore_state == VCORE_PIGGYBACK))
2747 		kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
2748 
2749 	if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2750 		kvmppc_vcore_end_preempt(vc);
2751 
2752 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2753 		kvmppc_remove_runnable(vc, vcpu);
2754 		vcpu->stat.signal_exits++;
2755 		kvm_run->exit_reason = KVM_EXIT_INTR;
2756 		vcpu->arch.ret = -EINTR;
2757 	}
2758 
2759 	if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
2760 		/* Wake up some vcpu to run the core */
2761 		v = list_first_entry(&vc->runnable_threads,
2762 				     struct kvm_vcpu, arch.run_list);
2763 		wake_up(&v->arch.cpu_run);
2764 	}
2765 
2766 	trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
2767 	spin_unlock(&vc->lock);
2768 	return vcpu->arch.ret;
2769 }
2770 
2771 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
2772 {
2773 	int r;
2774 	int srcu_idx;
2775 
2776 	if (!vcpu->arch.sane) {
2777 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2778 		return -EINVAL;
2779 	}
2780 
2781 	kvmppc_core_prepare_to_enter(vcpu);
2782 
2783 	/* No need to go into the guest when all we'll do is come back out */
2784 	if (signal_pending(current)) {
2785 		run->exit_reason = KVM_EXIT_INTR;
2786 		return -EINTR;
2787 	}
2788 
2789 	atomic_inc(&vcpu->kvm->arch.vcpus_running);
2790 	/* Order vcpus_running vs. hpte_setup_done, see kvmppc_alloc_reset_hpt */
2791 	smp_mb();
2792 
2793 	/* On the first time here, set up HTAB and VRMA */
2794 	if (!vcpu->kvm->arch.hpte_setup_done) {
2795 		r = kvmppc_hv_setup_htab_rma(vcpu);
2796 		if (r)
2797 			goto out;
2798 	}
2799 
2800 	flush_all_to_thread(current);
2801 
2802 	vcpu->arch.wqp = &vcpu->arch.vcore->wq;
2803 	vcpu->arch.pgdir = current->mm->pgd;
2804 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2805 
2806 	do {
2807 		r = kvmppc_run_vcpu(run, vcpu);
2808 
2809 		if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
2810 		    !(vcpu->arch.shregs.msr & MSR_PR)) {
2811 			trace_kvm_hcall_enter(vcpu);
2812 			r = kvmppc_pseries_do_hcall(vcpu);
2813 			trace_kvm_hcall_exit(vcpu, r);
2814 			kvmppc_core_prepare_to_enter(vcpu);
2815 		} else if (r == RESUME_PAGE_FAULT) {
2816 			srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2817 			r = kvmppc_book3s_hv_page_fault(run, vcpu,
2818 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
2819 			srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2820 		}
2821 	} while (is_kvmppc_resume_guest(r));
2822 
2823  out:
2824 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2825 	atomic_dec(&vcpu->kvm->arch.vcpus_running);
2826 	return r;
2827 }
2828 
2829 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
2830 				     int linux_psize)
2831 {
2832 	struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
2833 
2834 	if (!def->shift)
2835 		return;
2836 	(*sps)->page_shift = def->shift;
2837 	(*sps)->slb_enc = def->sllp;
2838 	(*sps)->enc[0].page_shift = def->shift;
2839 	(*sps)->enc[0].pte_enc = def->penc[linux_psize];
2840 	/*
2841 	 * Add 16MB MPSS support if host supports it
2842 	 */
2843 	if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
2844 		(*sps)->enc[1].page_shift = 24;
2845 		(*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
2846 	}
2847 	(*sps)++;
2848 }
2849 
2850 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
2851 					 struct kvm_ppc_smmu_info *info)
2852 {
2853 	struct kvm_ppc_one_seg_page_size *sps;
2854 
2855 	info->flags = KVM_PPC_PAGE_SIZES_REAL;
2856 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
2857 		info->flags |= KVM_PPC_1T_SEGMENTS;
2858 	info->slb_size = mmu_slb_size;
2859 
2860 	/* We only support these sizes for now, and no muti-size segments */
2861 	sps = &info->sps[0];
2862 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
2863 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
2864 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
2865 
2866 	return 0;
2867 }
2868 
2869 /*
2870  * Get (and clear) the dirty memory log for a memory slot.
2871  */
2872 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
2873 					 struct kvm_dirty_log *log)
2874 {
2875 	struct kvm_memslots *slots;
2876 	struct kvm_memory_slot *memslot;
2877 	int r;
2878 	unsigned long n;
2879 
2880 	mutex_lock(&kvm->slots_lock);
2881 
2882 	r = -EINVAL;
2883 	if (log->slot >= KVM_USER_MEM_SLOTS)
2884 		goto out;
2885 
2886 	slots = kvm_memslots(kvm);
2887 	memslot = id_to_memslot(slots, log->slot);
2888 	r = -ENOENT;
2889 	if (!memslot->dirty_bitmap)
2890 		goto out;
2891 
2892 	n = kvm_dirty_bitmap_bytes(memslot);
2893 	memset(memslot->dirty_bitmap, 0, n);
2894 
2895 	r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
2896 	if (r)
2897 		goto out;
2898 
2899 	r = -EFAULT;
2900 	if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
2901 		goto out;
2902 
2903 	r = 0;
2904 out:
2905 	mutex_unlock(&kvm->slots_lock);
2906 	return r;
2907 }
2908 
2909 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
2910 					struct kvm_memory_slot *dont)
2911 {
2912 	if (!dont || free->arch.rmap != dont->arch.rmap) {
2913 		vfree(free->arch.rmap);
2914 		free->arch.rmap = NULL;
2915 	}
2916 }
2917 
2918 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
2919 					 unsigned long npages)
2920 {
2921 	slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
2922 	if (!slot->arch.rmap)
2923 		return -ENOMEM;
2924 
2925 	return 0;
2926 }
2927 
2928 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
2929 					struct kvm_memory_slot *memslot,
2930 					const struct kvm_userspace_memory_region *mem)
2931 {
2932 	return 0;
2933 }
2934 
2935 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
2936 				const struct kvm_userspace_memory_region *mem,
2937 				const struct kvm_memory_slot *old,
2938 				const struct kvm_memory_slot *new)
2939 {
2940 	unsigned long npages = mem->memory_size >> PAGE_SHIFT;
2941 	struct kvm_memslots *slots;
2942 	struct kvm_memory_slot *memslot;
2943 
2944 	if (npages && old->npages) {
2945 		/*
2946 		 * If modifying a memslot, reset all the rmap dirty bits.
2947 		 * If this is a new memslot, we don't need to do anything
2948 		 * since the rmap array starts out as all zeroes,
2949 		 * i.e. no pages are dirty.
2950 		 */
2951 		slots = kvm_memslots(kvm);
2952 		memslot = id_to_memslot(slots, mem->slot);
2953 		kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
2954 	}
2955 }
2956 
2957 /*
2958  * Update LPCR values in kvm->arch and in vcores.
2959  * Caller must hold kvm->lock.
2960  */
2961 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
2962 {
2963 	long int i;
2964 	u32 cores_done = 0;
2965 
2966 	if ((kvm->arch.lpcr & mask) == lpcr)
2967 		return;
2968 
2969 	kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
2970 
2971 	for (i = 0; i < KVM_MAX_VCORES; ++i) {
2972 		struct kvmppc_vcore *vc = kvm->arch.vcores[i];
2973 		if (!vc)
2974 			continue;
2975 		spin_lock(&vc->lock);
2976 		vc->lpcr = (vc->lpcr & ~mask) | lpcr;
2977 		spin_unlock(&vc->lock);
2978 		if (++cores_done >= kvm->arch.online_vcores)
2979 			break;
2980 	}
2981 }
2982 
2983 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
2984 {
2985 	return;
2986 }
2987 
2988 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
2989 {
2990 	int err = 0;
2991 	struct kvm *kvm = vcpu->kvm;
2992 	unsigned long hva;
2993 	struct kvm_memory_slot *memslot;
2994 	struct vm_area_struct *vma;
2995 	unsigned long lpcr = 0, senc;
2996 	unsigned long psize, porder;
2997 	int srcu_idx;
2998 
2999 	mutex_lock(&kvm->lock);
3000 	if (kvm->arch.hpte_setup_done)
3001 		goto out;	/* another vcpu beat us to it */
3002 
3003 	/* Allocate hashed page table (if not done already) and reset it */
3004 	if (!kvm->arch.hpt_virt) {
3005 		err = kvmppc_alloc_hpt(kvm, NULL);
3006 		if (err) {
3007 			pr_err("KVM: Couldn't alloc HPT\n");
3008 			goto out;
3009 		}
3010 	}
3011 
3012 	/* Look up the memslot for guest physical address 0 */
3013 	srcu_idx = srcu_read_lock(&kvm->srcu);
3014 	memslot = gfn_to_memslot(kvm, 0);
3015 
3016 	/* We must have some memory at 0 by now */
3017 	err = -EINVAL;
3018 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
3019 		goto out_srcu;
3020 
3021 	/* Look up the VMA for the start of this memory slot */
3022 	hva = memslot->userspace_addr;
3023 	down_read(&current->mm->mmap_sem);
3024 	vma = find_vma(current->mm, hva);
3025 	if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
3026 		goto up_out;
3027 
3028 	psize = vma_kernel_pagesize(vma);
3029 	porder = __ilog2(psize);
3030 
3031 	up_read(&current->mm->mmap_sem);
3032 
3033 	/* We can handle 4k, 64k or 16M pages in the VRMA */
3034 	err = -EINVAL;
3035 	if (!(psize == 0x1000 || psize == 0x10000 ||
3036 	      psize == 0x1000000))
3037 		goto out_srcu;
3038 
3039 	/* Update VRMASD field in the LPCR */
3040 	senc = slb_pgsize_encoding(psize);
3041 	kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
3042 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
3043 	/* the -4 is to account for senc values starting at 0x10 */
3044 	lpcr = senc << (LPCR_VRMASD_SH - 4);
3045 
3046 	/* Create HPTEs in the hash page table for the VRMA */
3047 	kvmppc_map_vrma(vcpu, memslot, porder);
3048 
3049 	kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
3050 
3051 	/* Order updates to kvm->arch.lpcr etc. vs. hpte_setup_done */
3052 	smp_wmb();
3053 	kvm->arch.hpte_setup_done = 1;
3054 	err = 0;
3055  out_srcu:
3056 	srcu_read_unlock(&kvm->srcu, srcu_idx);
3057  out:
3058 	mutex_unlock(&kvm->lock);
3059 	return err;
3060 
3061  up_out:
3062 	up_read(&current->mm->mmap_sem);
3063 	goto out_srcu;
3064 }
3065 
3066 #ifdef CONFIG_KVM_XICS
3067 static int kvmppc_cpu_notify(struct notifier_block *self, unsigned long action,
3068 			void *hcpu)
3069 {
3070 	unsigned long cpu = (long)hcpu;
3071 
3072 	switch (action) {
3073 	case CPU_UP_PREPARE:
3074 	case CPU_UP_PREPARE_FROZEN:
3075 		kvmppc_set_host_core(cpu);
3076 		break;
3077 
3078 #ifdef CONFIG_HOTPLUG_CPU
3079 	case CPU_DEAD:
3080 	case CPU_DEAD_FROZEN:
3081 	case CPU_UP_CANCELED:
3082 	case CPU_UP_CANCELED_FROZEN:
3083 		kvmppc_clear_host_core(cpu);
3084 		break;
3085 #endif
3086 	default:
3087 		break;
3088 	}
3089 
3090 	return NOTIFY_OK;
3091 }
3092 
3093 static struct notifier_block kvmppc_cpu_notifier = {
3094 	    .notifier_call = kvmppc_cpu_notify,
3095 };
3096 
3097 /*
3098  * Allocate a per-core structure for managing state about which cores are
3099  * running in the host versus the guest and for exchanging data between
3100  * real mode KVM and CPU running in the host.
3101  * This is only done for the first VM.
3102  * The allocated structure stays even if all VMs have stopped.
3103  * It is only freed when the kvm-hv module is unloaded.
3104  * It's OK for this routine to fail, we just don't support host
3105  * core operations like redirecting H_IPI wakeups.
3106  */
3107 void kvmppc_alloc_host_rm_ops(void)
3108 {
3109 	struct kvmppc_host_rm_ops *ops;
3110 	unsigned long l_ops;
3111 	int cpu, core;
3112 	int size;
3113 
3114 	/* Not the first time here ? */
3115 	if (kvmppc_host_rm_ops_hv != NULL)
3116 		return;
3117 
3118 	ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
3119 	if (!ops)
3120 		return;
3121 
3122 	size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
3123 	ops->rm_core = kzalloc(size, GFP_KERNEL);
3124 
3125 	if (!ops->rm_core) {
3126 		kfree(ops);
3127 		return;
3128 	}
3129 
3130 	get_online_cpus();
3131 
3132 	for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
3133 		if (!cpu_online(cpu))
3134 			continue;
3135 
3136 		core = cpu >> threads_shift;
3137 		ops->rm_core[core].rm_state.in_host = 1;
3138 	}
3139 
3140 	ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
3141 
3142 	/*
3143 	 * Make the contents of the kvmppc_host_rm_ops structure visible
3144 	 * to other CPUs before we assign it to the global variable.
3145 	 * Do an atomic assignment (no locks used here), but if someone
3146 	 * beats us to it, just free our copy and return.
3147 	 */
3148 	smp_wmb();
3149 	l_ops = (unsigned long) ops;
3150 
3151 	if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
3152 		put_online_cpus();
3153 		kfree(ops->rm_core);
3154 		kfree(ops);
3155 		return;
3156 	}
3157 
3158 	register_cpu_notifier(&kvmppc_cpu_notifier);
3159 
3160 	put_online_cpus();
3161 }
3162 
3163 void kvmppc_free_host_rm_ops(void)
3164 {
3165 	if (kvmppc_host_rm_ops_hv) {
3166 		unregister_cpu_notifier(&kvmppc_cpu_notifier);
3167 		kfree(kvmppc_host_rm_ops_hv->rm_core);
3168 		kfree(kvmppc_host_rm_ops_hv);
3169 		kvmppc_host_rm_ops_hv = NULL;
3170 	}
3171 }
3172 #endif
3173 
3174 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
3175 {
3176 	unsigned long lpcr, lpid;
3177 	char buf[32];
3178 
3179 	/* Allocate the guest's logical partition ID */
3180 
3181 	lpid = kvmppc_alloc_lpid();
3182 	if ((long)lpid < 0)
3183 		return -ENOMEM;
3184 	kvm->arch.lpid = lpid;
3185 
3186 	kvmppc_alloc_host_rm_ops();
3187 
3188 	/*
3189 	 * Since we don't flush the TLB when tearing down a VM,
3190 	 * and this lpid might have previously been used,
3191 	 * make sure we flush on each core before running the new VM.
3192 	 */
3193 	cpumask_setall(&kvm->arch.need_tlb_flush);
3194 
3195 	/* Start out with the default set of hcalls enabled */
3196 	memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
3197 	       sizeof(kvm->arch.enabled_hcalls));
3198 
3199 	kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
3200 
3201 	/* Init LPCR for virtual RMA mode */
3202 	kvm->arch.host_lpid = mfspr(SPRN_LPID);
3203 	kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
3204 	lpcr &= LPCR_PECE | LPCR_LPES;
3205 	lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3206 		LPCR_VPM0 | LPCR_VPM1;
3207 	kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3208 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
3209 	/* On POWER8 turn on online bit to enable PURR/SPURR */
3210 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
3211 		lpcr |= LPCR_ONL;
3212 	kvm->arch.lpcr = lpcr;
3213 
3214 	/*
3215 	 * Track that we now have a HV mode VM active. This blocks secondary
3216 	 * CPU threads from coming online.
3217 	 */
3218 	kvm_hv_vm_activated();
3219 
3220 	/*
3221 	 * Create a debugfs directory for the VM
3222 	 */
3223 	snprintf(buf, sizeof(buf), "vm%d", current->pid);
3224 	kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
3225 	if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
3226 		kvmppc_mmu_debugfs_init(kvm);
3227 
3228 	return 0;
3229 }
3230 
3231 static void kvmppc_free_vcores(struct kvm *kvm)
3232 {
3233 	long int i;
3234 
3235 	for (i = 0; i < KVM_MAX_VCORES; ++i)
3236 		kfree(kvm->arch.vcores[i]);
3237 	kvm->arch.online_vcores = 0;
3238 }
3239 
3240 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
3241 {
3242 	debugfs_remove_recursive(kvm->arch.debugfs_dir);
3243 
3244 	kvm_hv_vm_deactivated();
3245 
3246 	kvmppc_free_vcores(kvm);
3247 
3248 	kvmppc_free_hpt(kvm);
3249 }
3250 
3251 /* We don't need to emulate any privileged instructions or dcbz */
3252 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
3253 				     unsigned int inst, int *advance)
3254 {
3255 	return EMULATE_FAIL;
3256 }
3257 
3258 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
3259 					ulong spr_val)
3260 {
3261 	return EMULATE_FAIL;
3262 }
3263 
3264 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
3265 					ulong *spr_val)
3266 {
3267 	return EMULATE_FAIL;
3268 }
3269 
3270 static int kvmppc_core_check_processor_compat_hv(void)
3271 {
3272 	if (!cpu_has_feature(CPU_FTR_HVMODE) ||
3273 	    !cpu_has_feature(CPU_FTR_ARCH_206))
3274 		return -EIO;
3275 	/*
3276 	 * Disable KVM for Power9, untill the required bits merged.
3277 	 */
3278 	if (cpu_has_feature(CPU_FTR_ARCH_300))
3279 		return -EIO;
3280 
3281 	return 0;
3282 }
3283 
3284 static long kvm_arch_vm_ioctl_hv(struct file *filp,
3285 				 unsigned int ioctl, unsigned long arg)
3286 {
3287 	struct kvm *kvm __maybe_unused = filp->private_data;
3288 	void __user *argp = (void __user *)arg;
3289 	long r;
3290 
3291 	switch (ioctl) {
3292 
3293 	case KVM_PPC_ALLOCATE_HTAB: {
3294 		u32 htab_order;
3295 
3296 		r = -EFAULT;
3297 		if (get_user(htab_order, (u32 __user *)argp))
3298 			break;
3299 		r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
3300 		if (r)
3301 			break;
3302 		r = -EFAULT;
3303 		if (put_user(htab_order, (u32 __user *)argp))
3304 			break;
3305 		r = 0;
3306 		break;
3307 	}
3308 
3309 	case KVM_PPC_GET_HTAB_FD: {
3310 		struct kvm_get_htab_fd ghf;
3311 
3312 		r = -EFAULT;
3313 		if (copy_from_user(&ghf, argp, sizeof(ghf)))
3314 			break;
3315 		r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
3316 		break;
3317 	}
3318 
3319 	default:
3320 		r = -ENOTTY;
3321 	}
3322 
3323 	return r;
3324 }
3325 
3326 /*
3327  * List of hcall numbers to enable by default.
3328  * For compatibility with old userspace, we enable by default
3329  * all hcalls that were implemented before the hcall-enabling
3330  * facility was added.  Note this list should not include H_RTAS.
3331  */
3332 static unsigned int default_hcall_list[] = {
3333 	H_REMOVE,
3334 	H_ENTER,
3335 	H_READ,
3336 	H_PROTECT,
3337 	H_BULK_REMOVE,
3338 	H_GET_TCE,
3339 	H_PUT_TCE,
3340 	H_SET_DABR,
3341 	H_SET_XDABR,
3342 	H_CEDE,
3343 	H_PROD,
3344 	H_CONFER,
3345 	H_REGISTER_VPA,
3346 #ifdef CONFIG_KVM_XICS
3347 	H_EOI,
3348 	H_CPPR,
3349 	H_IPI,
3350 	H_IPOLL,
3351 	H_XIRR,
3352 	H_XIRR_X,
3353 #endif
3354 	0
3355 };
3356 
3357 static void init_default_hcalls(void)
3358 {
3359 	int i;
3360 	unsigned int hcall;
3361 
3362 	for (i = 0; default_hcall_list[i]; ++i) {
3363 		hcall = default_hcall_list[i];
3364 		WARN_ON(!kvmppc_hcall_impl_hv(hcall));
3365 		__set_bit(hcall / 4, default_enabled_hcalls);
3366 	}
3367 }
3368 
3369 static struct kvmppc_ops kvm_ops_hv = {
3370 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
3371 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
3372 	.get_one_reg = kvmppc_get_one_reg_hv,
3373 	.set_one_reg = kvmppc_set_one_reg_hv,
3374 	.vcpu_load   = kvmppc_core_vcpu_load_hv,
3375 	.vcpu_put    = kvmppc_core_vcpu_put_hv,
3376 	.set_msr     = kvmppc_set_msr_hv,
3377 	.vcpu_run    = kvmppc_vcpu_run_hv,
3378 	.vcpu_create = kvmppc_core_vcpu_create_hv,
3379 	.vcpu_free   = kvmppc_core_vcpu_free_hv,
3380 	.check_requests = kvmppc_core_check_requests_hv,
3381 	.get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
3382 	.flush_memslot  = kvmppc_core_flush_memslot_hv,
3383 	.prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
3384 	.commit_memory_region  = kvmppc_core_commit_memory_region_hv,
3385 	.unmap_hva = kvm_unmap_hva_hv,
3386 	.unmap_hva_range = kvm_unmap_hva_range_hv,
3387 	.age_hva  = kvm_age_hva_hv,
3388 	.test_age_hva = kvm_test_age_hva_hv,
3389 	.set_spte_hva = kvm_set_spte_hva_hv,
3390 	.mmu_destroy  = kvmppc_mmu_destroy_hv,
3391 	.free_memslot = kvmppc_core_free_memslot_hv,
3392 	.create_memslot = kvmppc_core_create_memslot_hv,
3393 	.init_vm =  kvmppc_core_init_vm_hv,
3394 	.destroy_vm = kvmppc_core_destroy_vm_hv,
3395 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
3396 	.emulate_op = kvmppc_core_emulate_op_hv,
3397 	.emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
3398 	.emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
3399 	.fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
3400 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
3401 	.hcall_implemented = kvmppc_hcall_impl_hv,
3402 };
3403 
3404 static int kvmppc_book3s_init_hv(void)
3405 {
3406 	int r;
3407 	/*
3408 	 * FIXME!! Do we need to check on all cpus ?
3409 	 */
3410 	r = kvmppc_core_check_processor_compat_hv();
3411 	if (r < 0)
3412 		return -ENODEV;
3413 
3414 	kvm_ops_hv.owner = THIS_MODULE;
3415 	kvmppc_hv_ops = &kvm_ops_hv;
3416 
3417 	init_default_hcalls();
3418 
3419 	init_vcore_lists();
3420 
3421 	r = kvmppc_mmu_hv_init();
3422 	return r;
3423 }
3424 
3425 static void kvmppc_book3s_exit_hv(void)
3426 {
3427 	kvmppc_free_host_rm_ops();
3428 	kvmppc_hv_ops = NULL;
3429 }
3430 
3431 module_init(kvmppc_book3s_init_hv);
3432 module_exit(kvmppc_book3s_exit_hv);
3433 MODULE_LICENSE("GPL");
3434 MODULE_ALIAS_MISCDEV(KVM_MINOR);
3435 MODULE_ALIAS("devname:kvm");
3436