xref: /openbmc/linux/arch/powerpc/kvm/book3s_hv.c (revision abfbd895)
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/cpumask.h>
31 #include <linux/spinlock.h>
32 #include <linux/page-flags.h>
33 #include <linux/srcu.h>
34 #include <linux/miscdevice.h>
35 #include <linux/debugfs.h>
36 
37 #include <asm/reg.h>
38 #include <asm/cputable.h>
39 #include <asm/cacheflush.h>
40 #include <asm/tlbflush.h>
41 #include <asm/uaccess.h>
42 #include <asm/io.h>
43 #include <asm/kvm_ppc.h>
44 #include <asm/kvm_book3s.h>
45 #include <asm/mmu_context.h>
46 #include <asm/lppaca.h>
47 #include <asm/processor.h>
48 #include <asm/cputhreads.h>
49 #include <asm/page.h>
50 #include <asm/hvcall.h>
51 #include <asm/switch_to.h>
52 #include <asm/smp.h>
53 #include <asm/dbell.h>
54 #include <linux/gfp.h>
55 #include <linux/vmalloc.h>
56 #include <linux/highmem.h>
57 #include <linux/hugetlb.h>
58 #include <linux/module.h>
59 
60 #include "book3s.h"
61 
62 #define CREATE_TRACE_POINTS
63 #include "trace_hv.h"
64 
65 /* #define EXIT_DEBUG */
66 /* #define EXIT_DEBUG_SIMPLE */
67 /* #define EXIT_DEBUG_INT */
68 
69 /* Used to indicate that a guest page fault needs to be handled */
70 #define RESUME_PAGE_FAULT	(RESUME_GUEST | RESUME_FLAG_ARCH1)
71 
72 /* Used as a "null" value for timebase values */
73 #define TB_NIL	(~(u64)0)
74 
75 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
76 
77 static int dynamic_mt_modes = 6;
78 module_param(dynamic_mt_modes, int, S_IRUGO | S_IWUSR);
79 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
80 static int target_smt_mode;
81 module_param(target_smt_mode, int, S_IRUGO | S_IWUSR);
82 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
83 
84 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
85 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
86 
87 static bool kvmppc_ipi_thread(int cpu)
88 {
89 	/* On POWER8 for IPIs to threads in the same core, use msgsnd */
90 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
91 		preempt_disable();
92 		if (cpu_first_thread_sibling(cpu) ==
93 		    cpu_first_thread_sibling(smp_processor_id())) {
94 			unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
95 			msg |= cpu_thread_in_core(cpu);
96 			smp_mb();
97 			__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
98 			preempt_enable();
99 			return true;
100 		}
101 		preempt_enable();
102 	}
103 
104 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
105 	if (cpu >= 0 && cpu < nr_cpu_ids && paca[cpu].kvm_hstate.xics_phys) {
106 		xics_wake_cpu(cpu);
107 		return true;
108 	}
109 #endif
110 
111 	return false;
112 }
113 
114 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
115 {
116 	int cpu;
117 	wait_queue_head_t *wqp;
118 
119 	wqp = kvm_arch_vcpu_wq(vcpu);
120 	if (waitqueue_active(wqp)) {
121 		wake_up_interruptible(wqp);
122 		++vcpu->stat.halt_wakeup;
123 	}
124 
125 	if (kvmppc_ipi_thread(vcpu->arch.thread_cpu))
126 		return;
127 
128 	/* CPU points to the first thread of the core */
129 	cpu = vcpu->cpu;
130 	if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
131 		smp_send_reschedule(cpu);
132 }
133 
134 /*
135  * We use the vcpu_load/put functions to measure stolen time.
136  * Stolen time is counted as time when either the vcpu is able to
137  * run as part of a virtual core, but the task running the vcore
138  * is preempted or sleeping, or when the vcpu needs something done
139  * in the kernel by the task running the vcpu, but that task is
140  * preempted or sleeping.  Those two things have to be counted
141  * separately, since one of the vcpu tasks will take on the job
142  * of running the core, and the other vcpu tasks in the vcore will
143  * sleep waiting for it to do that, but that sleep shouldn't count
144  * as stolen time.
145  *
146  * Hence we accumulate stolen time when the vcpu can run as part of
147  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
148  * needs its task to do other things in the kernel (for example,
149  * service a page fault) in busy_stolen.  We don't accumulate
150  * stolen time for a vcore when it is inactive, or for a vcpu
151  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
152  * a misnomer; it means that the vcpu task is not executing in
153  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
154  * the kernel.  We don't have any way of dividing up that time
155  * between time that the vcpu is genuinely stopped, time that
156  * the task is actively working on behalf of the vcpu, and time
157  * that the task is preempted, so we don't count any of it as
158  * stolen.
159  *
160  * Updates to busy_stolen are protected by arch.tbacct_lock;
161  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
162  * lock.  The stolen times are measured in units of timebase ticks.
163  * (Note that the != TB_NIL checks below are purely defensive;
164  * they should never fail.)
165  */
166 
167 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
168 {
169 	unsigned long flags;
170 
171 	spin_lock_irqsave(&vc->stoltb_lock, flags);
172 	vc->preempt_tb = mftb();
173 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
174 }
175 
176 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
177 {
178 	unsigned long flags;
179 
180 	spin_lock_irqsave(&vc->stoltb_lock, flags);
181 	if (vc->preempt_tb != TB_NIL) {
182 		vc->stolen_tb += mftb() - vc->preempt_tb;
183 		vc->preempt_tb = TB_NIL;
184 	}
185 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
186 }
187 
188 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
189 {
190 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
191 	unsigned long flags;
192 
193 	/*
194 	 * We can test vc->runner without taking the vcore lock,
195 	 * because only this task ever sets vc->runner to this
196 	 * vcpu, and once it is set to this vcpu, only this task
197 	 * ever sets it to NULL.
198 	 */
199 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
200 		kvmppc_core_end_stolen(vc);
201 
202 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
203 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
204 	    vcpu->arch.busy_preempt != TB_NIL) {
205 		vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
206 		vcpu->arch.busy_preempt = TB_NIL;
207 	}
208 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
209 }
210 
211 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
212 {
213 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
214 	unsigned long flags;
215 
216 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
217 		kvmppc_core_start_stolen(vc);
218 
219 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
220 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
221 		vcpu->arch.busy_preempt = mftb();
222 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
223 }
224 
225 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
226 {
227 	/*
228 	 * Check for illegal transactional state bit combination
229 	 * and if we find it, force the TS field to a safe state.
230 	 */
231 	if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
232 		msr &= ~MSR_TS_MASK;
233 	vcpu->arch.shregs.msr = msr;
234 	kvmppc_end_cede(vcpu);
235 }
236 
237 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
238 {
239 	vcpu->arch.pvr = pvr;
240 }
241 
242 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
243 {
244 	unsigned long pcr = 0;
245 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
246 
247 	if (arch_compat) {
248 		switch (arch_compat) {
249 		case PVR_ARCH_205:
250 			/*
251 			 * If an arch bit is set in PCR, all the defined
252 			 * higher-order arch bits also have to be set.
253 			 */
254 			pcr = PCR_ARCH_206 | PCR_ARCH_205;
255 			break;
256 		case PVR_ARCH_206:
257 		case PVR_ARCH_206p:
258 			pcr = PCR_ARCH_206;
259 			break;
260 		case PVR_ARCH_207:
261 			break;
262 		default:
263 			return -EINVAL;
264 		}
265 
266 		if (!cpu_has_feature(CPU_FTR_ARCH_207S)) {
267 			/* POWER7 can't emulate POWER8 */
268 			if (!(pcr & PCR_ARCH_206))
269 				return -EINVAL;
270 			pcr &= ~PCR_ARCH_206;
271 		}
272 	}
273 
274 	spin_lock(&vc->lock);
275 	vc->arch_compat = arch_compat;
276 	vc->pcr = pcr;
277 	spin_unlock(&vc->lock);
278 
279 	return 0;
280 }
281 
282 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
283 {
284 	int r;
285 
286 	pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
287 	pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
288 	       vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
289 	for (r = 0; r < 16; ++r)
290 		pr_err("r%2d = %.16lx  r%d = %.16lx\n",
291 		       r, kvmppc_get_gpr(vcpu, r),
292 		       r+16, kvmppc_get_gpr(vcpu, r+16));
293 	pr_err("ctr = %.16lx  lr  = %.16lx\n",
294 	       vcpu->arch.ctr, vcpu->arch.lr);
295 	pr_err("srr0 = %.16llx srr1 = %.16llx\n",
296 	       vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
297 	pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
298 	       vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
299 	pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
300 	       vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
301 	pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
302 	       vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
303 	pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
304 	pr_err("fault dar = %.16lx dsisr = %.8x\n",
305 	       vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
306 	pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
307 	for (r = 0; r < vcpu->arch.slb_max; ++r)
308 		pr_err("  ESID = %.16llx VSID = %.16llx\n",
309 		       vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
310 	pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
311 	       vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
312 	       vcpu->arch.last_inst);
313 }
314 
315 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
316 {
317 	int r;
318 	struct kvm_vcpu *v, *ret = NULL;
319 
320 	mutex_lock(&kvm->lock);
321 	kvm_for_each_vcpu(r, v, kvm) {
322 		if (v->vcpu_id == id) {
323 			ret = v;
324 			break;
325 		}
326 	}
327 	mutex_unlock(&kvm->lock);
328 	return ret;
329 }
330 
331 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
332 {
333 	vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
334 	vpa->yield_count = cpu_to_be32(1);
335 }
336 
337 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
338 		   unsigned long addr, unsigned long len)
339 {
340 	/* check address is cacheline aligned */
341 	if (addr & (L1_CACHE_BYTES - 1))
342 		return -EINVAL;
343 	spin_lock(&vcpu->arch.vpa_update_lock);
344 	if (v->next_gpa != addr || v->len != len) {
345 		v->next_gpa = addr;
346 		v->len = addr ? len : 0;
347 		v->update_pending = 1;
348 	}
349 	spin_unlock(&vcpu->arch.vpa_update_lock);
350 	return 0;
351 }
352 
353 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
354 struct reg_vpa {
355 	u32 dummy;
356 	union {
357 		__be16 hword;
358 		__be32 word;
359 	} length;
360 };
361 
362 static int vpa_is_registered(struct kvmppc_vpa *vpap)
363 {
364 	if (vpap->update_pending)
365 		return vpap->next_gpa != 0;
366 	return vpap->pinned_addr != NULL;
367 }
368 
369 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
370 				       unsigned long flags,
371 				       unsigned long vcpuid, unsigned long vpa)
372 {
373 	struct kvm *kvm = vcpu->kvm;
374 	unsigned long len, nb;
375 	void *va;
376 	struct kvm_vcpu *tvcpu;
377 	int err;
378 	int subfunc;
379 	struct kvmppc_vpa *vpap;
380 
381 	tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
382 	if (!tvcpu)
383 		return H_PARAMETER;
384 
385 	subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
386 	if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
387 	    subfunc == H_VPA_REG_SLB) {
388 		/* Registering new area - address must be cache-line aligned */
389 		if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
390 			return H_PARAMETER;
391 
392 		/* convert logical addr to kernel addr and read length */
393 		va = kvmppc_pin_guest_page(kvm, vpa, &nb);
394 		if (va == NULL)
395 			return H_PARAMETER;
396 		if (subfunc == H_VPA_REG_VPA)
397 			len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
398 		else
399 			len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
400 		kvmppc_unpin_guest_page(kvm, va, vpa, false);
401 
402 		/* Check length */
403 		if (len > nb || len < sizeof(struct reg_vpa))
404 			return H_PARAMETER;
405 	} else {
406 		vpa = 0;
407 		len = 0;
408 	}
409 
410 	err = H_PARAMETER;
411 	vpap = NULL;
412 	spin_lock(&tvcpu->arch.vpa_update_lock);
413 
414 	switch (subfunc) {
415 	case H_VPA_REG_VPA:		/* register VPA */
416 		if (len < sizeof(struct lppaca))
417 			break;
418 		vpap = &tvcpu->arch.vpa;
419 		err = 0;
420 		break;
421 
422 	case H_VPA_REG_DTL:		/* register DTL */
423 		if (len < sizeof(struct dtl_entry))
424 			break;
425 		len -= len % sizeof(struct dtl_entry);
426 
427 		/* Check that they have previously registered a VPA */
428 		err = H_RESOURCE;
429 		if (!vpa_is_registered(&tvcpu->arch.vpa))
430 			break;
431 
432 		vpap = &tvcpu->arch.dtl;
433 		err = 0;
434 		break;
435 
436 	case H_VPA_REG_SLB:		/* register SLB shadow buffer */
437 		/* Check that they have previously registered a VPA */
438 		err = H_RESOURCE;
439 		if (!vpa_is_registered(&tvcpu->arch.vpa))
440 			break;
441 
442 		vpap = &tvcpu->arch.slb_shadow;
443 		err = 0;
444 		break;
445 
446 	case H_VPA_DEREG_VPA:		/* deregister VPA */
447 		/* Check they don't still have a DTL or SLB buf registered */
448 		err = H_RESOURCE;
449 		if (vpa_is_registered(&tvcpu->arch.dtl) ||
450 		    vpa_is_registered(&tvcpu->arch.slb_shadow))
451 			break;
452 
453 		vpap = &tvcpu->arch.vpa;
454 		err = 0;
455 		break;
456 
457 	case H_VPA_DEREG_DTL:		/* deregister DTL */
458 		vpap = &tvcpu->arch.dtl;
459 		err = 0;
460 		break;
461 
462 	case H_VPA_DEREG_SLB:		/* deregister SLB shadow buffer */
463 		vpap = &tvcpu->arch.slb_shadow;
464 		err = 0;
465 		break;
466 	}
467 
468 	if (vpap) {
469 		vpap->next_gpa = vpa;
470 		vpap->len = len;
471 		vpap->update_pending = 1;
472 	}
473 
474 	spin_unlock(&tvcpu->arch.vpa_update_lock);
475 
476 	return err;
477 }
478 
479 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
480 {
481 	struct kvm *kvm = vcpu->kvm;
482 	void *va;
483 	unsigned long nb;
484 	unsigned long gpa;
485 
486 	/*
487 	 * We need to pin the page pointed to by vpap->next_gpa,
488 	 * but we can't call kvmppc_pin_guest_page under the lock
489 	 * as it does get_user_pages() and down_read().  So we
490 	 * have to drop the lock, pin the page, then get the lock
491 	 * again and check that a new area didn't get registered
492 	 * in the meantime.
493 	 */
494 	for (;;) {
495 		gpa = vpap->next_gpa;
496 		spin_unlock(&vcpu->arch.vpa_update_lock);
497 		va = NULL;
498 		nb = 0;
499 		if (gpa)
500 			va = kvmppc_pin_guest_page(kvm, gpa, &nb);
501 		spin_lock(&vcpu->arch.vpa_update_lock);
502 		if (gpa == vpap->next_gpa)
503 			break;
504 		/* sigh... unpin that one and try again */
505 		if (va)
506 			kvmppc_unpin_guest_page(kvm, va, gpa, false);
507 	}
508 
509 	vpap->update_pending = 0;
510 	if (va && nb < vpap->len) {
511 		/*
512 		 * If it's now too short, it must be that userspace
513 		 * has changed the mappings underlying guest memory,
514 		 * so unregister the region.
515 		 */
516 		kvmppc_unpin_guest_page(kvm, va, gpa, false);
517 		va = NULL;
518 	}
519 	if (vpap->pinned_addr)
520 		kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
521 					vpap->dirty);
522 	vpap->gpa = gpa;
523 	vpap->pinned_addr = va;
524 	vpap->dirty = false;
525 	if (va)
526 		vpap->pinned_end = va + vpap->len;
527 }
528 
529 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
530 {
531 	if (!(vcpu->arch.vpa.update_pending ||
532 	      vcpu->arch.slb_shadow.update_pending ||
533 	      vcpu->arch.dtl.update_pending))
534 		return;
535 
536 	spin_lock(&vcpu->arch.vpa_update_lock);
537 	if (vcpu->arch.vpa.update_pending) {
538 		kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
539 		if (vcpu->arch.vpa.pinned_addr)
540 			init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
541 	}
542 	if (vcpu->arch.dtl.update_pending) {
543 		kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
544 		vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
545 		vcpu->arch.dtl_index = 0;
546 	}
547 	if (vcpu->arch.slb_shadow.update_pending)
548 		kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
549 	spin_unlock(&vcpu->arch.vpa_update_lock);
550 }
551 
552 /*
553  * Return the accumulated stolen time for the vcore up until `now'.
554  * The caller should hold the vcore lock.
555  */
556 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
557 {
558 	u64 p;
559 	unsigned long flags;
560 
561 	spin_lock_irqsave(&vc->stoltb_lock, flags);
562 	p = vc->stolen_tb;
563 	if (vc->vcore_state != VCORE_INACTIVE &&
564 	    vc->preempt_tb != TB_NIL)
565 		p += now - vc->preempt_tb;
566 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
567 	return p;
568 }
569 
570 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
571 				    struct kvmppc_vcore *vc)
572 {
573 	struct dtl_entry *dt;
574 	struct lppaca *vpa;
575 	unsigned long stolen;
576 	unsigned long core_stolen;
577 	u64 now;
578 
579 	dt = vcpu->arch.dtl_ptr;
580 	vpa = vcpu->arch.vpa.pinned_addr;
581 	now = mftb();
582 	core_stolen = vcore_stolen_time(vc, now);
583 	stolen = core_stolen - vcpu->arch.stolen_logged;
584 	vcpu->arch.stolen_logged = core_stolen;
585 	spin_lock_irq(&vcpu->arch.tbacct_lock);
586 	stolen += vcpu->arch.busy_stolen;
587 	vcpu->arch.busy_stolen = 0;
588 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
589 	if (!dt || !vpa)
590 		return;
591 	memset(dt, 0, sizeof(struct dtl_entry));
592 	dt->dispatch_reason = 7;
593 	dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
594 	dt->timebase = cpu_to_be64(now + vc->tb_offset);
595 	dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
596 	dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
597 	dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
598 	++dt;
599 	if (dt == vcpu->arch.dtl.pinned_end)
600 		dt = vcpu->arch.dtl.pinned_addr;
601 	vcpu->arch.dtl_ptr = dt;
602 	/* order writing *dt vs. writing vpa->dtl_idx */
603 	smp_wmb();
604 	vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
605 	vcpu->arch.dtl.dirty = true;
606 }
607 
608 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
609 {
610 	if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
611 		return true;
612 	if ((!vcpu->arch.vcore->arch_compat) &&
613 	    cpu_has_feature(CPU_FTR_ARCH_207S))
614 		return true;
615 	return false;
616 }
617 
618 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
619 			     unsigned long resource, unsigned long value1,
620 			     unsigned long value2)
621 {
622 	switch (resource) {
623 	case H_SET_MODE_RESOURCE_SET_CIABR:
624 		if (!kvmppc_power8_compatible(vcpu))
625 			return H_P2;
626 		if (value2)
627 			return H_P4;
628 		if (mflags)
629 			return H_UNSUPPORTED_FLAG_START;
630 		/* Guests can't breakpoint the hypervisor */
631 		if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
632 			return H_P3;
633 		vcpu->arch.ciabr  = value1;
634 		return H_SUCCESS;
635 	case H_SET_MODE_RESOURCE_SET_DAWR:
636 		if (!kvmppc_power8_compatible(vcpu))
637 			return H_P2;
638 		if (mflags)
639 			return H_UNSUPPORTED_FLAG_START;
640 		if (value2 & DABRX_HYP)
641 			return H_P4;
642 		vcpu->arch.dawr  = value1;
643 		vcpu->arch.dawrx = value2;
644 		return H_SUCCESS;
645 	default:
646 		return H_TOO_HARD;
647 	}
648 }
649 
650 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
651 {
652 	struct kvmppc_vcore *vcore = target->arch.vcore;
653 
654 	/*
655 	 * We expect to have been called by the real mode handler
656 	 * (kvmppc_rm_h_confer()) which would have directly returned
657 	 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
658 	 * have useful work to do and should not confer) so we don't
659 	 * recheck that here.
660 	 */
661 
662 	spin_lock(&vcore->lock);
663 	if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
664 	    vcore->vcore_state != VCORE_INACTIVE &&
665 	    vcore->runner)
666 		target = vcore->runner;
667 	spin_unlock(&vcore->lock);
668 
669 	return kvm_vcpu_yield_to(target);
670 }
671 
672 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
673 {
674 	int yield_count = 0;
675 	struct lppaca *lppaca;
676 
677 	spin_lock(&vcpu->arch.vpa_update_lock);
678 	lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
679 	if (lppaca)
680 		yield_count = be32_to_cpu(lppaca->yield_count);
681 	spin_unlock(&vcpu->arch.vpa_update_lock);
682 	return yield_count;
683 }
684 
685 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
686 {
687 	unsigned long req = kvmppc_get_gpr(vcpu, 3);
688 	unsigned long target, ret = H_SUCCESS;
689 	int yield_count;
690 	struct kvm_vcpu *tvcpu;
691 	int idx, rc;
692 
693 	if (req <= MAX_HCALL_OPCODE &&
694 	    !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
695 		return RESUME_HOST;
696 
697 	switch (req) {
698 	case H_CEDE:
699 		break;
700 	case H_PROD:
701 		target = kvmppc_get_gpr(vcpu, 4);
702 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
703 		if (!tvcpu) {
704 			ret = H_PARAMETER;
705 			break;
706 		}
707 		tvcpu->arch.prodded = 1;
708 		smp_mb();
709 		if (vcpu->arch.ceded) {
710 			if (waitqueue_active(&vcpu->wq)) {
711 				wake_up_interruptible(&vcpu->wq);
712 				vcpu->stat.halt_wakeup++;
713 			}
714 		}
715 		break;
716 	case H_CONFER:
717 		target = kvmppc_get_gpr(vcpu, 4);
718 		if (target == -1)
719 			break;
720 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
721 		if (!tvcpu) {
722 			ret = H_PARAMETER;
723 			break;
724 		}
725 		yield_count = kvmppc_get_gpr(vcpu, 5);
726 		if (kvmppc_get_yield_count(tvcpu) != yield_count)
727 			break;
728 		kvm_arch_vcpu_yield_to(tvcpu);
729 		break;
730 	case H_REGISTER_VPA:
731 		ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
732 					kvmppc_get_gpr(vcpu, 5),
733 					kvmppc_get_gpr(vcpu, 6));
734 		break;
735 	case H_RTAS:
736 		if (list_empty(&vcpu->kvm->arch.rtas_tokens))
737 			return RESUME_HOST;
738 
739 		idx = srcu_read_lock(&vcpu->kvm->srcu);
740 		rc = kvmppc_rtas_hcall(vcpu);
741 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
742 
743 		if (rc == -ENOENT)
744 			return RESUME_HOST;
745 		else if (rc == 0)
746 			break;
747 
748 		/* Send the error out to userspace via KVM_RUN */
749 		return rc;
750 	case H_LOGICAL_CI_LOAD:
751 		ret = kvmppc_h_logical_ci_load(vcpu);
752 		if (ret == H_TOO_HARD)
753 			return RESUME_HOST;
754 		break;
755 	case H_LOGICAL_CI_STORE:
756 		ret = kvmppc_h_logical_ci_store(vcpu);
757 		if (ret == H_TOO_HARD)
758 			return RESUME_HOST;
759 		break;
760 	case H_SET_MODE:
761 		ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
762 					kvmppc_get_gpr(vcpu, 5),
763 					kvmppc_get_gpr(vcpu, 6),
764 					kvmppc_get_gpr(vcpu, 7));
765 		if (ret == H_TOO_HARD)
766 			return RESUME_HOST;
767 		break;
768 	case H_XIRR:
769 	case H_CPPR:
770 	case H_EOI:
771 	case H_IPI:
772 	case H_IPOLL:
773 	case H_XIRR_X:
774 		if (kvmppc_xics_enabled(vcpu)) {
775 			ret = kvmppc_xics_hcall(vcpu, req);
776 			break;
777 		} /* fallthrough */
778 	default:
779 		return RESUME_HOST;
780 	}
781 	kvmppc_set_gpr(vcpu, 3, ret);
782 	vcpu->arch.hcall_needed = 0;
783 	return RESUME_GUEST;
784 }
785 
786 static int kvmppc_hcall_impl_hv(unsigned long cmd)
787 {
788 	switch (cmd) {
789 	case H_CEDE:
790 	case H_PROD:
791 	case H_CONFER:
792 	case H_REGISTER_VPA:
793 	case H_SET_MODE:
794 	case H_LOGICAL_CI_LOAD:
795 	case H_LOGICAL_CI_STORE:
796 #ifdef CONFIG_KVM_XICS
797 	case H_XIRR:
798 	case H_CPPR:
799 	case H_EOI:
800 	case H_IPI:
801 	case H_IPOLL:
802 	case H_XIRR_X:
803 #endif
804 		return 1;
805 	}
806 
807 	/* See if it's in the real-mode table */
808 	return kvmppc_hcall_impl_hv_realmode(cmd);
809 }
810 
811 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
812 					struct kvm_vcpu *vcpu)
813 {
814 	u32 last_inst;
815 
816 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
817 					EMULATE_DONE) {
818 		/*
819 		 * Fetch failed, so return to guest and
820 		 * try executing it again.
821 		 */
822 		return RESUME_GUEST;
823 	}
824 
825 	if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
826 		run->exit_reason = KVM_EXIT_DEBUG;
827 		run->debug.arch.address = kvmppc_get_pc(vcpu);
828 		return RESUME_HOST;
829 	} else {
830 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
831 		return RESUME_GUEST;
832 	}
833 }
834 
835 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
836 				 struct task_struct *tsk)
837 {
838 	int r = RESUME_HOST;
839 
840 	vcpu->stat.sum_exits++;
841 
842 	run->exit_reason = KVM_EXIT_UNKNOWN;
843 	run->ready_for_interrupt_injection = 1;
844 	switch (vcpu->arch.trap) {
845 	/* We're good on these - the host merely wanted to get our attention */
846 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
847 		vcpu->stat.dec_exits++;
848 		r = RESUME_GUEST;
849 		break;
850 	case BOOK3S_INTERRUPT_EXTERNAL:
851 	case BOOK3S_INTERRUPT_H_DOORBELL:
852 		vcpu->stat.ext_intr_exits++;
853 		r = RESUME_GUEST;
854 		break;
855 	/* HMI is hypervisor interrupt and host has handled it. Resume guest.*/
856 	case BOOK3S_INTERRUPT_HMI:
857 	case BOOK3S_INTERRUPT_PERFMON:
858 		r = RESUME_GUEST;
859 		break;
860 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
861 		/*
862 		 * Deliver a machine check interrupt to the guest.
863 		 * We have to do this, even if the host has handled the
864 		 * machine check, because machine checks use SRR0/1 and
865 		 * the interrupt might have trashed guest state in them.
866 		 */
867 		kvmppc_book3s_queue_irqprio(vcpu,
868 					    BOOK3S_INTERRUPT_MACHINE_CHECK);
869 		r = RESUME_GUEST;
870 		break;
871 	case BOOK3S_INTERRUPT_PROGRAM:
872 	{
873 		ulong flags;
874 		/*
875 		 * Normally program interrupts are delivered directly
876 		 * to the guest by the hardware, but we can get here
877 		 * as a result of a hypervisor emulation interrupt
878 		 * (e40) getting turned into a 700 by BML RTAS.
879 		 */
880 		flags = vcpu->arch.shregs.msr & 0x1f0000ull;
881 		kvmppc_core_queue_program(vcpu, flags);
882 		r = RESUME_GUEST;
883 		break;
884 	}
885 	case BOOK3S_INTERRUPT_SYSCALL:
886 	{
887 		/* hcall - punt to userspace */
888 		int i;
889 
890 		/* hypercall with MSR_PR has already been handled in rmode,
891 		 * and never reaches here.
892 		 */
893 
894 		run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
895 		for (i = 0; i < 9; ++i)
896 			run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
897 		run->exit_reason = KVM_EXIT_PAPR_HCALL;
898 		vcpu->arch.hcall_needed = 1;
899 		r = RESUME_HOST;
900 		break;
901 	}
902 	/*
903 	 * We get these next two if the guest accesses a page which it thinks
904 	 * it has mapped but which is not actually present, either because
905 	 * it is for an emulated I/O device or because the corresonding
906 	 * host page has been paged out.  Any other HDSI/HISI interrupts
907 	 * have been handled already.
908 	 */
909 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
910 		r = RESUME_PAGE_FAULT;
911 		break;
912 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
913 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
914 		vcpu->arch.fault_dsisr = 0;
915 		r = RESUME_PAGE_FAULT;
916 		break;
917 	/*
918 	 * This occurs if the guest executes an illegal instruction.
919 	 * If the guest debug is disabled, generate a program interrupt
920 	 * to the guest. If guest debug is enabled, we need to check
921 	 * whether the instruction is a software breakpoint instruction.
922 	 * Accordingly return to Guest or Host.
923 	 */
924 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
925 		if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
926 			vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
927 				swab32(vcpu->arch.emul_inst) :
928 				vcpu->arch.emul_inst;
929 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
930 			r = kvmppc_emulate_debug_inst(run, vcpu);
931 		} else {
932 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
933 			r = RESUME_GUEST;
934 		}
935 		break;
936 	/*
937 	 * This occurs if the guest (kernel or userspace), does something that
938 	 * is prohibited by HFSCR.  We just generate a program interrupt to
939 	 * the guest.
940 	 */
941 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
942 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
943 		r = RESUME_GUEST;
944 		break;
945 	default:
946 		kvmppc_dump_regs(vcpu);
947 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
948 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
949 			vcpu->arch.shregs.msr);
950 		run->hw.hardware_exit_reason = vcpu->arch.trap;
951 		r = RESUME_HOST;
952 		break;
953 	}
954 
955 	return r;
956 }
957 
958 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
959 					    struct kvm_sregs *sregs)
960 {
961 	int i;
962 
963 	memset(sregs, 0, sizeof(struct kvm_sregs));
964 	sregs->pvr = vcpu->arch.pvr;
965 	for (i = 0; i < vcpu->arch.slb_max; i++) {
966 		sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
967 		sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
968 	}
969 
970 	return 0;
971 }
972 
973 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
974 					    struct kvm_sregs *sregs)
975 {
976 	int i, j;
977 
978 	/* Only accept the same PVR as the host's, since we can't spoof it */
979 	if (sregs->pvr != vcpu->arch.pvr)
980 		return -EINVAL;
981 
982 	j = 0;
983 	for (i = 0; i < vcpu->arch.slb_nr; i++) {
984 		if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
985 			vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
986 			vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
987 			++j;
988 		}
989 	}
990 	vcpu->arch.slb_max = j;
991 
992 	return 0;
993 }
994 
995 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
996 		bool preserve_top32)
997 {
998 	struct kvm *kvm = vcpu->kvm;
999 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
1000 	u64 mask;
1001 
1002 	mutex_lock(&kvm->lock);
1003 	spin_lock(&vc->lock);
1004 	/*
1005 	 * If ILE (interrupt little-endian) has changed, update the
1006 	 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1007 	 */
1008 	if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1009 		struct kvm_vcpu *vcpu;
1010 		int i;
1011 
1012 		kvm_for_each_vcpu(i, vcpu, kvm) {
1013 			if (vcpu->arch.vcore != vc)
1014 				continue;
1015 			if (new_lpcr & LPCR_ILE)
1016 				vcpu->arch.intr_msr |= MSR_LE;
1017 			else
1018 				vcpu->arch.intr_msr &= ~MSR_LE;
1019 		}
1020 	}
1021 
1022 	/*
1023 	 * Userspace can only modify DPFD (default prefetch depth),
1024 	 * ILE (interrupt little-endian) and TC (translation control).
1025 	 * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
1026 	 */
1027 	mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1028 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
1029 		mask |= LPCR_AIL;
1030 
1031 	/* Broken 32-bit version of LPCR must not clear top bits */
1032 	if (preserve_top32)
1033 		mask &= 0xFFFFFFFF;
1034 	vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1035 	spin_unlock(&vc->lock);
1036 	mutex_unlock(&kvm->lock);
1037 }
1038 
1039 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1040 				 union kvmppc_one_reg *val)
1041 {
1042 	int r = 0;
1043 	long int i;
1044 
1045 	switch (id) {
1046 	case KVM_REG_PPC_DEBUG_INST:
1047 		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1048 		break;
1049 	case KVM_REG_PPC_HIOR:
1050 		*val = get_reg_val(id, 0);
1051 		break;
1052 	case KVM_REG_PPC_DABR:
1053 		*val = get_reg_val(id, vcpu->arch.dabr);
1054 		break;
1055 	case KVM_REG_PPC_DABRX:
1056 		*val = get_reg_val(id, vcpu->arch.dabrx);
1057 		break;
1058 	case KVM_REG_PPC_DSCR:
1059 		*val = get_reg_val(id, vcpu->arch.dscr);
1060 		break;
1061 	case KVM_REG_PPC_PURR:
1062 		*val = get_reg_val(id, vcpu->arch.purr);
1063 		break;
1064 	case KVM_REG_PPC_SPURR:
1065 		*val = get_reg_val(id, vcpu->arch.spurr);
1066 		break;
1067 	case KVM_REG_PPC_AMR:
1068 		*val = get_reg_val(id, vcpu->arch.amr);
1069 		break;
1070 	case KVM_REG_PPC_UAMOR:
1071 		*val = get_reg_val(id, vcpu->arch.uamor);
1072 		break;
1073 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1074 		i = id - KVM_REG_PPC_MMCR0;
1075 		*val = get_reg_val(id, vcpu->arch.mmcr[i]);
1076 		break;
1077 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1078 		i = id - KVM_REG_PPC_PMC1;
1079 		*val = get_reg_val(id, vcpu->arch.pmc[i]);
1080 		break;
1081 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1082 		i = id - KVM_REG_PPC_SPMC1;
1083 		*val = get_reg_val(id, vcpu->arch.spmc[i]);
1084 		break;
1085 	case KVM_REG_PPC_SIAR:
1086 		*val = get_reg_val(id, vcpu->arch.siar);
1087 		break;
1088 	case KVM_REG_PPC_SDAR:
1089 		*val = get_reg_val(id, vcpu->arch.sdar);
1090 		break;
1091 	case KVM_REG_PPC_SIER:
1092 		*val = get_reg_val(id, vcpu->arch.sier);
1093 		break;
1094 	case KVM_REG_PPC_IAMR:
1095 		*val = get_reg_val(id, vcpu->arch.iamr);
1096 		break;
1097 	case KVM_REG_PPC_PSPB:
1098 		*val = get_reg_val(id, vcpu->arch.pspb);
1099 		break;
1100 	case KVM_REG_PPC_DPDES:
1101 		*val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1102 		break;
1103 	case KVM_REG_PPC_DAWR:
1104 		*val = get_reg_val(id, vcpu->arch.dawr);
1105 		break;
1106 	case KVM_REG_PPC_DAWRX:
1107 		*val = get_reg_val(id, vcpu->arch.dawrx);
1108 		break;
1109 	case KVM_REG_PPC_CIABR:
1110 		*val = get_reg_val(id, vcpu->arch.ciabr);
1111 		break;
1112 	case KVM_REG_PPC_CSIGR:
1113 		*val = get_reg_val(id, vcpu->arch.csigr);
1114 		break;
1115 	case KVM_REG_PPC_TACR:
1116 		*val = get_reg_val(id, vcpu->arch.tacr);
1117 		break;
1118 	case KVM_REG_PPC_TCSCR:
1119 		*val = get_reg_val(id, vcpu->arch.tcscr);
1120 		break;
1121 	case KVM_REG_PPC_PID:
1122 		*val = get_reg_val(id, vcpu->arch.pid);
1123 		break;
1124 	case KVM_REG_PPC_ACOP:
1125 		*val = get_reg_val(id, vcpu->arch.acop);
1126 		break;
1127 	case KVM_REG_PPC_WORT:
1128 		*val = get_reg_val(id, vcpu->arch.wort);
1129 		break;
1130 	case KVM_REG_PPC_VPA_ADDR:
1131 		spin_lock(&vcpu->arch.vpa_update_lock);
1132 		*val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1133 		spin_unlock(&vcpu->arch.vpa_update_lock);
1134 		break;
1135 	case KVM_REG_PPC_VPA_SLB:
1136 		spin_lock(&vcpu->arch.vpa_update_lock);
1137 		val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1138 		val->vpaval.length = vcpu->arch.slb_shadow.len;
1139 		spin_unlock(&vcpu->arch.vpa_update_lock);
1140 		break;
1141 	case KVM_REG_PPC_VPA_DTL:
1142 		spin_lock(&vcpu->arch.vpa_update_lock);
1143 		val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1144 		val->vpaval.length = vcpu->arch.dtl.len;
1145 		spin_unlock(&vcpu->arch.vpa_update_lock);
1146 		break;
1147 	case KVM_REG_PPC_TB_OFFSET:
1148 		*val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1149 		break;
1150 	case KVM_REG_PPC_LPCR:
1151 	case KVM_REG_PPC_LPCR_64:
1152 		*val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1153 		break;
1154 	case KVM_REG_PPC_PPR:
1155 		*val = get_reg_val(id, vcpu->arch.ppr);
1156 		break;
1157 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1158 	case KVM_REG_PPC_TFHAR:
1159 		*val = get_reg_val(id, vcpu->arch.tfhar);
1160 		break;
1161 	case KVM_REG_PPC_TFIAR:
1162 		*val = get_reg_val(id, vcpu->arch.tfiar);
1163 		break;
1164 	case KVM_REG_PPC_TEXASR:
1165 		*val = get_reg_val(id, vcpu->arch.texasr);
1166 		break;
1167 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1168 		i = id - KVM_REG_PPC_TM_GPR0;
1169 		*val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1170 		break;
1171 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1172 	{
1173 		int j;
1174 		i = id - KVM_REG_PPC_TM_VSR0;
1175 		if (i < 32)
1176 			for (j = 0; j < TS_FPRWIDTH; j++)
1177 				val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1178 		else {
1179 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1180 				val->vval = vcpu->arch.vr_tm.vr[i-32];
1181 			else
1182 				r = -ENXIO;
1183 		}
1184 		break;
1185 	}
1186 	case KVM_REG_PPC_TM_CR:
1187 		*val = get_reg_val(id, vcpu->arch.cr_tm);
1188 		break;
1189 	case KVM_REG_PPC_TM_LR:
1190 		*val = get_reg_val(id, vcpu->arch.lr_tm);
1191 		break;
1192 	case KVM_REG_PPC_TM_CTR:
1193 		*val = get_reg_val(id, vcpu->arch.ctr_tm);
1194 		break;
1195 	case KVM_REG_PPC_TM_FPSCR:
1196 		*val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1197 		break;
1198 	case KVM_REG_PPC_TM_AMR:
1199 		*val = get_reg_val(id, vcpu->arch.amr_tm);
1200 		break;
1201 	case KVM_REG_PPC_TM_PPR:
1202 		*val = get_reg_val(id, vcpu->arch.ppr_tm);
1203 		break;
1204 	case KVM_REG_PPC_TM_VRSAVE:
1205 		*val = get_reg_val(id, vcpu->arch.vrsave_tm);
1206 		break;
1207 	case KVM_REG_PPC_TM_VSCR:
1208 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1209 			*val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1210 		else
1211 			r = -ENXIO;
1212 		break;
1213 	case KVM_REG_PPC_TM_DSCR:
1214 		*val = get_reg_val(id, vcpu->arch.dscr_tm);
1215 		break;
1216 	case KVM_REG_PPC_TM_TAR:
1217 		*val = get_reg_val(id, vcpu->arch.tar_tm);
1218 		break;
1219 #endif
1220 	case KVM_REG_PPC_ARCH_COMPAT:
1221 		*val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1222 		break;
1223 	default:
1224 		r = -EINVAL;
1225 		break;
1226 	}
1227 
1228 	return r;
1229 }
1230 
1231 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1232 				 union kvmppc_one_reg *val)
1233 {
1234 	int r = 0;
1235 	long int i;
1236 	unsigned long addr, len;
1237 
1238 	switch (id) {
1239 	case KVM_REG_PPC_HIOR:
1240 		/* Only allow this to be set to zero */
1241 		if (set_reg_val(id, *val))
1242 			r = -EINVAL;
1243 		break;
1244 	case KVM_REG_PPC_DABR:
1245 		vcpu->arch.dabr = set_reg_val(id, *val);
1246 		break;
1247 	case KVM_REG_PPC_DABRX:
1248 		vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1249 		break;
1250 	case KVM_REG_PPC_DSCR:
1251 		vcpu->arch.dscr = set_reg_val(id, *val);
1252 		break;
1253 	case KVM_REG_PPC_PURR:
1254 		vcpu->arch.purr = set_reg_val(id, *val);
1255 		break;
1256 	case KVM_REG_PPC_SPURR:
1257 		vcpu->arch.spurr = set_reg_val(id, *val);
1258 		break;
1259 	case KVM_REG_PPC_AMR:
1260 		vcpu->arch.amr = set_reg_val(id, *val);
1261 		break;
1262 	case KVM_REG_PPC_UAMOR:
1263 		vcpu->arch.uamor = set_reg_val(id, *val);
1264 		break;
1265 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1266 		i = id - KVM_REG_PPC_MMCR0;
1267 		vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1268 		break;
1269 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1270 		i = id - KVM_REG_PPC_PMC1;
1271 		vcpu->arch.pmc[i] = set_reg_val(id, *val);
1272 		break;
1273 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1274 		i = id - KVM_REG_PPC_SPMC1;
1275 		vcpu->arch.spmc[i] = set_reg_val(id, *val);
1276 		break;
1277 	case KVM_REG_PPC_SIAR:
1278 		vcpu->arch.siar = set_reg_val(id, *val);
1279 		break;
1280 	case KVM_REG_PPC_SDAR:
1281 		vcpu->arch.sdar = set_reg_val(id, *val);
1282 		break;
1283 	case KVM_REG_PPC_SIER:
1284 		vcpu->arch.sier = set_reg_val(id, *val);
1285 		break;
1286 	case KVM_REG_PPC_IAMR:
1287 		vcpu->arch.iamr = set_reg_val(id, *val);
1288 		break;
1289 	case KVM_REG_PPC_PSPB:
1290 		vcpu->arch.pspb = set_reg_val(id, *val);
1291 		break;
1292 	case KVM_REG_PPC_DPDES:
1293 		vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1294 		break;
1295 	case KVM_REG_PPC_DAWR:
1296 		vcpu->arch.dawr = set_reg_val(id, *val);
1297 		break;
1298 	case KVM_REG_PPC_DAWRX:
1299 		vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1300 		break;
1301 	case KVM_REG_PPC_CIABR:
1302 		vcpu->arch.ciabr = set_reg_val(id, *val);
1303 		/* Don't allow setting breakpoints in hypervisor code */
1304 		if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1305 			vcpu->arch.ciabr &= ~CIABR_PRIV;	/* disable */
1306 		break;
1307 	case KVM_REG_PPC_CSIGR:
1308 		vcpu->arch.csigr = set_reg_val(id, *val);
1309 		break;
1310 	case KVM_REG_PPC_TACR:
1311 		vcpu->arch.tacr = set_reg_val(id, *val);
1312 		break;
1313 	case KVM_REG_PPC_TCSCR:
1314 		vcpu->arch.tcscr = set_reg_val(id, *val);
1315 		break;
1316 	case KVM_REG_PPC_PID:
1317 		vcpu->arch.pid = set_reg_val(id, *val);
1318 		break;
1319 	case KVM_REG_PPC_ACOP:
1320 		vcpu->arch.acop = set_reg_val(id, *val);
1321 		break;
1322 	case KVM_REG_PPC_WORT:
1323 		vcpu->arch.wort = set_reg_val(id, *val);
1324 		break;
1325 	case KVM_REG_PPC_VPA_ADDR:
1326 		addr = set_reg_val(id, *val);
1327 		r = -EINVAL;
1328 		if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1329 			      vcpu->arch.dtl.next_gpa))
1330 			break;
1331 		r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1332 		break;
1333 	case KVM_REG_PPC_VPA_SLB:
1334 		addr = val->vpaval.addr;
1335 		len = val->vpaval.length;
1336 		r = -EINVAL;
1337 		if (addr && !vcpu->arch.vpa.next_gpa)
1338 			break;
1339 		r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1340 		break;
1341 	case KVM_REG_PPC_VPA_DTL:
1342 		addr = val->vpaval.addr;
1343 		len = val->vpaval.length;
1344 		r = -EINVAL;
1345 		if (addr && (len < sizeof(struct dtl_entry) ||
1346 			     !vcpu->arch.vpa.next_gpa))
1347 			break;
1348 		len -= len % sizeof(struct dtl_entry);
1349 		r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1350 		break;
1351 	case KVM_REG_PPC_TB_OFFSET:
1352 		/* round up to multiple of 2^24 */
1353 		vcpu->arch.vcore->tb_offset =
1354 			ALIGN(set_reg_val(id, *val), 1UL << 24);
1355 		break;
1356 	case KVM_REG_PPC_LPCR:
1357 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1358 		break;
1359 	case KVM_REG_PPC_LPCR_64:
1360 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1361 		break;
1362 	case KVM_REG_PPC_PPR:
1363 		vcpu->arch.ppr = set_reg_val(id, *val);
1364 		break;
1365 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1366 	case KVM_REG_PPC_TFHAR:
1367 		vcpu->arch.tfhar = set_reg_val(id, *val);
1368 		break;
1369 	case KVM_REG_PPC_TFIAR:
1370 		vcpu->arch.tfiar = set_reg_val(id, *val);
1371 		break;
1372 	case KVM_REG_PPC_TEXASR:
1373 		vcpu->arch.texasr = set_reg_val(id, *val);
1374 		break;
1375 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1376 		i = id - KVM_REG_PPC_TM_GPR0;
1377 		vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1378 		break;
1379 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1380 	{
1381 		int j;
1382 		i = id - KVM_REG_PPC_TM_VSR0;
1383 		if (i < 32)
1384 			for (j = 0; j < TS_FPRWIDTH; j++)
1385 				vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1386 		else
1387 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1388 				vcpu->arch.vr_tm.vr[i-32] = val->vval;
1389 			else
1390 				r = -ENXIO;
1391 		break;
1392 	}
1393 	case KVM_REG_PPC_TM_CR:
1394 		vcpu->arch.cr_tm = set_reg_val(id, *val);
1395 		break;
1396 	case KVM_REG_PPC_TM_LR:
1397 		vcpu->arch.lr_tm = set_reg_val(id, *val);
1398 		break;
1399 	case KVM_REG_PPC_TM_CTR:
1400 		vcpu->arch.ctr_tm = set_reg_val(id, *val);
1401 		break;
1402 	case KVM_REG_PPC_TM_FPSCR:
1403 		vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1404 		break;
1405 	case KVM_REG_PPC_TM_AMR:
1406 		vcpu->arch.amr_tm = set_reg_val(id, *val);
1407 		break;
1408 	case KVM_REG_PPC_TM_PPR:
1409 		vcpu->arch.ppr_tm = set_reg_val(id, *val);
1410 		break;
1411 	case KVM_REG_PPC_TM_VRSAVE:
1412 		vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1413 		break;
1414 	case KVM_REG_PPC_TM_VSCR:
1415 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1416 			vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1417 		else
1418 			r = - ENXIO;
1419 		break;
1420 	case KVM_REG_PPC_TM_DSCR:
1421 		vcpu->arch.dscr_tm = set_reg_val(id, *val);
1422 		break;
1423 	case KVM_REG_PPC_TM_TAR:
1424 		vcpu->arch.tar_tm = set_reg_val(id, *val);
1425 		break;
1426 #endif
1427 	case KVM_REG_PPC_ARCH_COMPAT:
1428 		r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1429 		break;
1430 	default:
1431 		r = -EINVAL;
1432 		break;
1433 	}
1434 
1435 	return r;
1436 }
1437 
1438 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1439 {
1440 	struct kvmppc_vcore *vcore;
1441 
1442 	vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1443 
1444 	if (vcore == NULL)
1445 		return NULL;
1446 
1447 	INIT_LIST_HEAD(&vcore->runnable_threads);
1448 	spin_lock_init(&vcore->lock);
1449 	spin_lock_init(&vcore->stoltb_lock);
1450 	init_waitqueue_head(&vcore->wq);
1451 	vcore->preempt_tb = TB_NIL;
1452 	vcore->lpcr = kvm->arch.lpcr;
1453 	vcore->first_vcpuid = core * threads_per_subcore;
1454 	vcore->kvm = kvm;
1455 	INIT_LIST_HEAD(&vcore->preempt_list);
1456 
1457 	return vcore;
1458 }
1459 
1460 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1461 static struct debugfs_timings_element {
1462 	const char *name;
1463 	size_t offset;
1464 } timings[] = {
1465 	{"rm_entry",	offsetof(struct kvm_vcpu, arch.rm_entry)},
1466 	{"rm_intr",	offsetof(struct kvm_vcpu, arch.rm_intr)},
1467 	{"rm_exit",	offsetof(struct kvm_vcpu, arch.rm_exit)},
1468 	{"guest",	offsetof(struct kvm_vcpu, arch.guest_time)},
1469 	{"cede",	offsetof(struct kvm_vcpu, arch.cede_time)},
1470 };
1471 
1472 #define N_TIMINGS	(sizeof(timings) / sizeof(timings[0]))
1473 
1474 struct debugfs_timings_state {
1475 	struct kvm_vcpu	*vcpu;
1476 	unsigned int	buflen;
1477 	char		buf[N_TIMINGS * 100];
1478 };
1479 
1480 static int debugfs_timings_open(struct inode *inode, struct file *file)
1481 {
1482 	struct kvm_vcpu *vcpu = inode->i_private;
1483 	struct debugfs_timings_state *p;
1484 
1485 	p = kzalloc(sizeof(*p), GFP_KERNEL);
1486 	if (!p)
1487 		return -ENOMEM;
1488 
1489 	kvm_get_kvm(vcpu->kvm);
1490 	p->vcpu = vcpu;
1491 	file->private_data = p;
1492 
1493 	return nonseekable_open(inode, file);
1494 }
1495 
1496 static int debugfs_timings_release(struct inode *inode, struct file *file)
1497 {
1498 	struct debugfs_timings_state *p = file->private_data;
1499 
1500 	kvm_put_kvm(p->vcpu->kvm);
1501 	kfree(p);
1502 	return 0;
1503 }
1504 
1505 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1506 				    size_t len, loff_t *ppos)
1507 {
1508 	struct debugfs_timings_state *p = file->private_data;
1509 	struct kvm_vcpu *vcpu = p->vcpu;
1510 	char *s, *buf_end;
1511 	struct kvmhv_tb_accumulator tb;
1512 	u64 count;
1513 	loff_t pos;
1514 	ssize_t n;
1515 	int i, loops;
1516 	bool ok;
1517 
1518 	if (!p->buflen) {
1519 		s = p->buf;
1520 		buf_end = s + sizeof(p->buf);
1521 		for (i = 0; i < N_TIMINGS; ++i) {
1522 			struct kvmhv_tb_accumulator *acc;
1523 
1524 			acc = (struct kvmhv_tb_accumulator *)
1525 				((unsigned long)vcpu + timings[i].offset);
1526 			ok = false;
1527 			for (loops = 0; loops < 1000; ++loops) {
1528 				count = acc->seqcount;
1529 				if (!(count & 1)) {
1530 					smp_rmb();
1531 					tb = *acc;
1532 					smp_rmb();
1533 					if (count == acc->seqcount) {
1534 						ok = true;
1535 						break;
1536 					}
1537 				}
1538 				udelay(1);
1539 			}
1540 			if (!ok)
1541 				snprintf(s, buf_end - s, "%s: stuck\n",
1542 					timings[i].name);
1543 			else
1544 				snprintf(s, buf_end - s,
1545 					"%s: %llu %llu %llu %llu\n",
1546 					timings[i].name, count / 2,
1547 					tb_to_ns(tb.tb_total),
1548 					tb_to_ns(tb.tb_min),
1549 					tb_to_ns(tb.tb_max));
1550 			s += strlen(s);
1551 		}
1552 		p->buflen = s - p->buf;
1553 	}
1554 
1555 	pos = *ppos;
1556 	if (pos >= p->buflen)
1557 		return 0;
1558 	if (len > p->buflen - pos)
1559 		len = p->buflen - pos;
1560 	n = copy_to_user(buf, p->buf + pos, len);
1561 	if (n) {
1562 		if (n == len)
1563 			return -EFAULT;
1564 		len -= n;
1565 	}
1566 	*ppos = pos + len;
1567 	return len;
1568 }
1569 
1570 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1571 				     size_t len, loff_t *ppos)
1572 {
1573 	return -EACCES;
1574 }
1575 
1576 static const struct file_operations debugfs_timings_ops = {
1577 	.owner	 = THIS_MODULE,
1578 	.open	 = debugfs_timings_open,
1579 	.release = debugfs_timings_release,
1580 	.read	 = debugfs_timings_read,
1581 	.write	 = debugfs_timings_write,
1582 	.llseek	 = generic_file_llseek,
1583 };
1584 
1585 /* Create a debugfs directory for the vcpu */
1586 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1587 {
1588 	char buf[16];
1589 	struct kvm *kvm = vcpu->kvm;
1590 
1591 	snprintf(buf, sizeof(buf), "vcpu%u", id);
1592 	if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1593 		return;
1594 	vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1595 	if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1596 		return;
1597 	vcpu->arch.debugfs_timings =
1598 		debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1599 				    vcpu, &debugfs_timings_ops);
1600 }
1601 
1602 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1603 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1604 {
1605 }
1606 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1607 
1608 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1609 						   unsigned int id)
1610 {
1611 	struct kvm_vcpu *vcpu;
1612 	int err = -EINVAL;
1613 	int core;
1614 	struct kvmppc_vcore *vcore;
1615 
1616 	core = id / threads_per_subcore;
1617 	if (core >= KVM_MAX_VCORES)
1618 		goto out;
1619 
1620 	err = -ENOMEM;
1621 	vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1622 	if (!vcpu)
1623 		goto out;
1624 
1625 	err = kvm_vcpu_init(vcpu, kvm, id);
1626 	if (err)
1627 		goto free_vcpu;
1628 
1629 	vcpu->arch.shared = &vcpu->arch.shregs;
1630 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1631 	/*
1632 	 * The shared struct is never shared on HV,
1633 	 * so we can always use host endianness
1634 	 */
1635 #ifdef __BIG_ENDIAN__
1636 	vcpu->arch.shared_big_endian = true;
1637 #else
1638 	vcpu->arch.shared_big_endian = false;
1639 #endif
1640 #endif
1641 	vcpu->arch.mmcr[0] = MMCR0_FC;
1642 	vcpu->arch.ctrl = CTRL_RUNLATCH;
1643 	/* default to host PVR, since we can't spoof it */
1644 	kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1645 	spin_lock_init(&vcpu->arch.vpa_update_lock);
1646 	spin_lock_init(&vcpu->arch.tbacct_lock);
1647 	vcpu->arch.busy_preempt = TB_NIL;
1648 	vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1649 
1650 	kvmppc_mmu_book3s_hv_init(vcpu);
1651 
1652 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1653 
1654 	init_waitqueue_head(&vcpu->arch.cpu_run);
1655 
1656 	mutex_lock(&kvm->lock);
1657 	vcore = kvm->arch.vcores[core];
1658 	if (!vcore) {
1659 		vcore = kvmppc_vcore_create(kvm, core);
1660 		kvm->arch.vcores[core] = vcore;
1661 		kvm->arch.online_vcores++;
1662 	}
1663 	mutex_unlock(&kvm->lock);
1664 
1665 	if (!vcore)
1666 		goto free_vcpu;
1667 
1668 	spin_lock(&vcore->lock);
1669 	++vcore->num_threads;
1670 	spin_unlock(&vcore->lock);
1671 	vcpu->arch.vcore = vcore;
1672 	vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1673 	vcpu->arch.thread_cpu = -1;
1674 
1675 	vcpu->arch.cpu_type = KVM_CPU_3S_64;
1676 	kvmppc_sanity_check(vcpu);
1677 
1678 	debugfs_vcpu_init(vcpu, id);
1679 
1680 	return vcpu;
1681 
1682 free_vcpu:
1683 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1684 out:
1685 	return ERR_PTR(err);
1686 }
1687 
1688 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1689 {
1690 	if (vpa->pinned_addr)
1691 		kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1692 					vpa->dirty);
1693 }
1694 
1695 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
1696 {
1697 	spin_lock(&vcpu->arch.vpa_update_lock);
1698 	unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1699 	unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1700 	unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1701 	spin_unlock(&vcpu->arch.vpa_update_lock);
1702 	kvm_vcpu_uninit(vcpu);
1703 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1704 }
1705 
1706 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1707 {
1708 	/* Indicate we want to get back into the guest */
1709 	return 1;
1710 }
1711 
1712 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1713 {
1714 	unsigned long dec_nsec, now;
1715 
1716 	now = get_tb();
1717 	if (now > vcpu->arch.dec_expires) {
1718 		/* decrementer has already gone negative */
1719 		kvmppc_core_queue_dec(vcpu);
1720 		kvmppc_core_prepare_to_enter(vcpu);
1721 		return;
1722 	}
1723 	dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1724 		   / tb_ticks_per_sec;
1725 	hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1726 		      HRTIMER_MODE_REL);
1727 	vcpu->arch.timer_running = 1;
1728 }
1729 
1730 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1731 {
1732 	vcpu->arch.ceded = 0;
1733 	if (vcpu->arch.timer_running) {
1734 		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1735 		vcpu->arch.timer_running = 0;
1736 	}
1737 }
1738 
1739 extern void __kvmppc_vcore_entry(void);
1740 
1741 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1742 				   struct kvm_vcpu *vcpu)
1743 {
1744 	u64 now;
1745 
1746 	if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1747 		return;
1748 	spin_lock_irq(&vcpu->arch.tbacct_lock);
1749 	now = mftb();
1750 	vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1751 		vcpu->arch.stolen_logged;
1752 	vcpu->arch.busy_preempt = now;
1753 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1754 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
1755 	--vc->n_runnable;
1756 	list_del(&vcpu->arch.run_list);
1757 }
1758 
1759 static int kvmppc_grab_hwthread(int cpu)
1760 {
1761 	struct paca_struct *tpaca;
1762 	long timeout = 10000;
1763 
1764 	tpaca = &paca[cpu];
1765 
1766 	/* Ensure the thread won't go into the kernel if it wakes */
1767 	tpaca->kvm_hstate.kvm_vcpu = NULL;
1768 	tpaca->kvm_hstate.kvm_vcore = NULL;
1769 	tpaca->kvm_hstate.napping = 0;
1770 	smp_wmb();
1771 	tpaca->kvm_hstate.hwthread_req = 1;
1772 
1773 	/*
1774 	 * If the thread is already executing in the kernel (e.g. handling
1775 	 * a stray interrupt), wait for it to get back to nap mode.
1776 	 * The smp_mb() is to ensure that our setting of hwthread_req
1777 	 * is visible before we look at hwthread_state, so if this
1778 	 * races with the code at system_reset_pSeries and the thread
1779 	 * misses our setting of hwthread_req, we are sure to see its
1780 	 * setting of hwthread_state, and vice versa.
1781 	 */
1782 	smp_mb();
1783 	while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1784 		if (--timeout <= 0) {
1785 			pr_err("KVM: couldn't grab cpu %d\n", cpu);
1786 			return -EBUSY;
1787 		}
1788 		udelay(1);
1789 	}
1790 	return 0;
1791 }
1792 
1793 static void kvmppc_release_hwthread(int cpu)
1794 {
1795 	struct paca_struct *tpaca;
1796 
1797 	tpaca = &paca[cpu];
1798 	tpaca->kvm_hstate.hwthread_req = 0;
1799 	tpaca->kvm_hstate.kvm_vcpu = NULL;
1800 	tpaca->kvm_hstate.kvm_vcore = NULL;
1801 	tpaca->kvm_hstate.kvm_split_mode = NULL;
1802 }
1803 
1804 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
1805 {
1806 	int cpu;
1807 	struct paca_struct *tpaca;
1808 	struct kvmppc_vcore *mvc = vc->master_vcore;
1809 
1810 	cpu = vc->pcpu;
1811 	if (vcpu) {
1812 		if (vcpu->arch.timer_running) {
1813 			hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1814 			vcpu->arch.timer_running = 0;
1815 		}
1816 		cpu += vcpu->arch.ptid;
1817 		vcpu->cpu = mvc->pcpu;
1818 		vcpu->arch.thread_cpu = cpu;
1819 	}
1820 	tpaca = &paca[cpu];
1821 	tpaca->kvm_hstate.kvm_vcpu = vcpu;
1822 	tpaca->kvm_hstate.ptid = cpu - mvc->pcpu;
1823 	/* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
1824 	smp_wmb();
1825 	tpaca->kvm_hstate.kvm_vcore = mvc;
1826 	if (cpu != smp_processor_id())
1827 		kvmppc_ipi_thread(cpu);
1828 }
1829 
1830 static void kvmppc_wait_for_nap(void)
1831 {
1832 	int cpu = smp_processor_id();
1833 	int i, loops;
1834 
1835 	for (loops = 0; loops < 1000000; ++loops) {
1836 		/*
1837 		 * Check if all threads are finished.
1838 		 * We set the vcore pointer when starting a thread
1839 		 * and the thread clears it when finished, so we look
1840 		 * for any threads that still have a non-NULL vcore ptr.
1841 		 */
1842 		for (i = 1; i < threads_per_subcore; ++i)
1843 			if (paca[cpu + i].kvm_hstate.kvm_vcore)
1844 				break;
1845 		if (i == threads_per_subcore) {
1846 			HMT_medium();
1847 			return;
1848 		}
1849 		HMT_low();
1850 	}
1851 	HMT_medium();
1852 	for (i = 1; i < threads_per_subcore; ++i)
1853 		if (paca[cpu + i].kvm_hstate.kvm_vcore)
1854 			pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
1855 }
1856 
1857 /*
1858  * Check that we are on thread 0 and that any other threads in
1859  * this core are off-line.  Then grab the threads so they can't
1860  * enter the kernel.
1861  */
1862 static int on_primary_thread(void)
1863 {
1864 	int cpu = smp_processor_id();
1865 	int thr;
1866 
1867 	/* Are we on a primary subcore? */
1868 	if (cpu_thread_in_subcore(cpu))
1869 		return 0;
1870 
1871 	thr = 0;
1872 	while (++thr < threads_per_subcore)
1873 		if (cpu_online(cpu + thr))
1874 			return 0;
1875 
1876 	/* Grab all hw threads so they can't go into the kernel */
1877 	for (thr = 1; thr < threads_per_subcore; ++thr) {
1878 		if (kvmppc_grab_hwthread(cpu + thr)) {
1879 			/* Couldn't grab one; let the others go */
1880 			do {
1881 				kvmppc_release_hwthread(cpu + thr);
1882 			} while (--thr > 0);
1883 			return 0;
1884 		}
1885 	}
1886 	return 1;
1887 }
1888 
1889 /*
1890  * A list of virtual cores for each physical CPU.
1891  * These are vcores that could run but their runner VCPU tasks are
1892  * (or may be) preempted.
1893  */
1894 struct preempted_vcore_list {
1895 	struct list_head	list;
1896 	spinlock_t		lock;
1897 };
1898 
1899 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
1900 
1901 static void init_vcore_lists(void)
1902 {
1903 	int cpu;
1904 
1905 	for_each_possible_cpu(cpu) {
1906 		struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
1907 		spin_lock_init(&lp->lock);
1908 		INIT_LIST_HEAD(&lp->list);
1909 	}
1910 }
1911 
1912 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
1913 {
1914 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
1915 
1916 	vc->vcore_state = VCORE_PREEMPT;
1917 	vc->pcpu = smp_processor_id();
1918 	if (vc->num_threads < threads_per_subcore) {
1919 		spin_lock(&lp->lock);
1920 		list_add_tail(&vc->preempt_list, &lp->list);
1921 		spin_unlock(&lp->lock);
1922 	}
1923 
1924 	/* Start accumulating stolen time */
1925 	kvmppc_core_start_stolen(vc);
1926 }
1927 
1928 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
1929 {
1930 	struct preempted_vcore_list *lp;
1931 
1932 	kvmppc_core_end_stolen(vc);
1933 	if (!list_empty(&vc->preempt_list)) {
1934 		lp = &per_cpu(preempted_vcores, vc->pcpu);
1935 		spin_lock(&lp->lock);
1936 		list_del_init(&vc->preempt_list);
1937 		spin_unlock(&lp->lock);
1938 	}
1939 	vc->vcore_state = VCORE_INACTIVE;
1940 }
1941 
1942 /*
1943  * This stores information about the virtual cores currently
1944  * assigned to a physical core.
1945  */
1946 struct core_info {
1947 	int		n_subcores;
1948 	int		max_subcore_threads;
1949 	int		total_threads;
1950 	int		subcore_threads[MAX_SUBCORES];
1951 	struct kvm	*subcore_vm[MAX_SUBCORES];
1952 	struct list_head vcs[MAX_SUBCORES];
1953 };
1954 
1955 /*
1956  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
1957  * respectively in 2-way micro-threading (split-core) mode.
1958  */
1959 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
1960 
1961 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
1962 {
1963 	int sub;
1964 
1965 	memset(cip, 0, sizeof(*cip));
1966 	cip->n_subcores = 1;
1967 	cip->max_subcore_threads = vc->num_threads;
1968 	cip->total_threads = vc->num_threads;
1969 	cip->subcore_threads[0] = vc->num_threads;
1970 	cip->subcore_vm[0] = vc->kvm;
1971 	for (sub = 0; sub < MAX_SUBCORES; ++sub)
1972 		INIT_LIST_HEAD(&cip->vcs[sub]);
1973 	list_add_tail(&vc->preempt_list, &cip->vcs[0]);
1974 }
1975 
1976 static bool subcore_config_ok(int n_subcores, int n_threads)
1977 {
1978 	/* Can only dynamically split if unsplit to begin with */
1979 	if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
1980 		return false;
1981 	if (n_subcores > MAX_SUBCORES)
1982 		return false;
1983 	if (n_subcores > 1) {
1984 		if (!(dynamic_mt_modes & 2))
1985 			n_subcores = 4;
1986 		if (n_subcores > 2 && !(dynamic_mt_modes & 4))
1987 			return false;
1988 	}
1989 
1990 	return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
1991 }
1992 
1993 static void init_master_vcore(struct kvmppc_vcore *vc)
1994 {
1995 	vc->master_vcore = vc;
1996 	vc->entry_exit_map = 0;
1997 	vc->in_guest = 0;
1998 	vc->napping_threads = 0;
1999 	vc->conferring_threads = 0;
2000 }
2001 
2002 /*
2003  * See if the existing subcores can be split into 3 (or fewer) subcores
2004  * of at most two threads each, so we can fit in another vcore.  This
2005  * assumes there are at most two subcores and at most 6 threads in total.
2006  */
2007 static bool can_split_piggybacked_subcores(struct core_info *cip)
2008 {
2009 	int sub, new_sub;
2010 	int large_sub = -1;
2011 	int thr;
2012 	int n_subcores = cip->n_subcores;
2013 	struct kvmppc_vcore *vc, *vcnext;
2014 	struct kvmppc_vcore *master_vc = NULL;
2015 
2016 	for (sub = 0; sub < cip->n_subcores; ++sub) {
2017 		if (cip->subcore_threads[sub] <= 2)
2018 			continue;
2019 		if (large_sub >= 0)
2020 			return false;
2021 		large_sub = sub;
2022 		vc = list_first_entry(&cip->vcs[sub], struct kvmppc_vcore,
2023 				      preempt_list);
2024 		if (vc->num_threads > 2)
2025 			return false;
2026 		n_subcores += (cip->subcore_threads[sub] - 1) >> 1;
2027 	}
2028 	if (large_sub < 0 || !subcore_config_ok(n_subcores + 1, 2))
2029 		return false;
2030 
2031 	/*
2032 	 * Seems feasible, so go through and move vcores to new subcores.
2033 	 * Note that when we have two or more vcores in one subcore,
2034 	 * all those vcores must have only one thread each.
2035 	 */
2036 	new_sub = cip->n_subcores;
2037 	thr = 0;
2038 	sub = large_sub;
2039 	list_for_each_entry_safe(vc, vcnext, &cip->vcs[sub], preempt_list) {
2040 		if (thr >= 2) {
2041 			list_del(&vc->preempt_list);
2042 			list_add_tail(&vc->preempt_list, &cip->vcs[new_sub]);
2043 			/* vc->num_threads must be 1 */
2044 			if (++cip->subcore_threads[new_sub] == 1) {
2045 				cip->subcore_vm[new_sub] = vc->kvm;
2046 				init_master_vcore(vc);
2047 				master_vc = vc;
2048 				++cip->n_subcores;
2049 			} else {
2050 				vc->master_vcore = master_vc;
2051 				++new_sub;
2052 			}
2053 		}
2054 		thr += vc->num_threads;
2055 	}
2056 	cip->subcore_threads[large_sub] = 2;
2057 	cip->max_subcore_threads = 2;
2058 
2059 	return true;
2060 }
2061 
2062 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2063 {
2064 	int n_threads = vc->num_threads;
2065 	int sub;
2066 
2067 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2068 		return false;
2069 
2070 	if (n_threads < cip->max_subcore_threads)
2071 		n_threads = cip->max_subcore_threads;
2072 	if (subcore_config_ok(cip->n_subcores + 1, n_threads)) {
2073 		cip->max_subcore_threads = n_threads;
2074 	} else if (cip->n_subcores <= 2 && cip->total_threads <= 6 &&
2075 		   vc->num_threads <= 2) {
2076 		/*
2077 		 * We may be able to fit another subcore in by
2078 		 * splitting an existing subcore with 3 or 4
2079 		 * threads into two 2-thread subcores, or one
2080 		 * with 5 or 6 threads into three subcores.
2081 		 * We can only do this if those subcores have
2082 		 * piggybacked virtual cores.
2083 		 */
2084 		if (!can_split_piggybacked_subcores(cip))
2085 			return false;
2086 	} else {
2087 		return false;
2088 	}
2089 
2090 	sub = cip->n_subcores;
2091 	++cip->n_subcores;
2092 	cip->total_threads += vc->num_threads;
2093 	cip->subcore_threads[sub] = vc->num_threads;
2094 	cip->subcore_vm[sub] = vc->kvm;
2095 	init_master_vcore(vc);
2096 	list_del(&vc->preempt_list);
2097 	list_add_tail(&vc->preempt_list, &cip->vcs[sub]);
2098 
2099 	return true;
2100 }
2101 
2102 static bool can_piggyback_subcore(struct kvmppc_vcore *pvc,
2103 				  struct core_info *cip, int sub)
2104 {
2105 	struct kvmppc_vcore *vc;
2106 	int n_thr;
2107 
2108 	vc = list_first_entry(&cip->vcs[sub], struct kvmppc_vcore,
2109 			      preempt_list);
2110 
2111 	/* require same VM and same per-core reg values */
2112 	if (pvc->kvm != vc->kvm ||
2113 	    pvc->tb_offset != vc->tb_offset ||
2114 	    pvc->pcr != vc->pcr ||
2115 	    pvc->lpcr != vc->lpcr)
2116 		return false;
2117 
2118 	/* P8 guest with > 1 thread per core would see wrong TIR value */
2119 	if (cpu_has_feature(CPU_FTR_ARCH_207S) &&
2120 	    (vc->num_threads > 1 || pvc->num_threads > 1))
2121 		return false;
2122 
2123 	n_thr = cip->subcore_threads[sub] + pvc->num_threads;
2124 	if (n_thr > cip->max_subcore_threads) {
2125 		if (!subcore_config_ok(cip->n_subcores, n_thr))
2126 			return false;
2127 		cip->max_subcore_threads = n_thr;
2128 	}
2129 
2130 	cip->total_threads += pvc->num_threads;
2131 	cip->subcore_threads[sub] = n_thr;
2132 	pvc->master_vcore = vc;
2133 	list_del(&pvc->preempt_list);
2134 	list_add_tail(&pvc->preempt_list, &cip->vcs[sub]);
2135 
2136 	return true;
2137 }
2138 
2139 /*
2140  * Work out whether it is possible to piggyback the execution of
2141  * vcore *pvc onto the execution of the other vcores described in *cip.
2142  */
2143 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2144 			  int target_threads)
2145 {
2146 	int sub;
2147 
2148 	if (cip->total_threads + pvc->num_threads > target_threads)
2149 		return false;
2150 	for (sub = 0; sub < cip->n_subcores; ++sub)
2151 		if (cip->subcore_threads[sub] &&
2152 		    can_piggyback_subcore(pvc, cip, sub))
2153 			return true;
2154 
2155 	if (can_dynamic_split(pvc, cip))
2156 		return true;
2157 
2158 	return false;
2159 }
2160 
2161 static void prepare_threads(struct kvmppc_vcore *vc)
2162 {
2163 	struct kvm_vcpu *vcpu, *vnext;
2164 
2165 	list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2166 				 arch.run_list) {
2167 		if (signal_pending(vcpu->arch.run_task))
2168 			vcpu->arch.ret = -EINTR;
2169 		else if (vcpu->arch.vpa.update_pending ||
2170 			 vcpu->arch.slb_shadow.update_pending ||
2171 			 vcpu->arch.dtl.update_pending)
2172 			vcpu->arch.ret = RESUME_GUEST;
2173 		else
2174 			continue;
2175 		kvmppc_remove_runnable(vc, vcpu);
2176 		wake_up(&vcpu->arch.cpu_run);
2177 	}
2178 }
2179 
2180 static void collect_piggybacks(struct core_info *cip, int target_threads)
2181 {
2182 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2183 	struct kvmppc_vcore *pvc, *vcnext;
2184 
2185 	spin_lock(&lp->lock);
2186 	list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2187 		if (!spin_trylock(&pvc->lock))
2188 			continue;
2189 		prepare_threads(pvc);
2190 		if (!pvc->n_runnable) {
2191 			list_del_init(&pvc->preempt_list);
2192 			if (pvc->runner == NULL) {
2193 				pvc->vcore_state = VCORE_INACTIVE;
2194 				kvmppc_core_end_stolen(pvc);
2195 			}
2196 			spin_unlock(&pvc->lock);
2197 			continue;
2198 		}
2199 		if (!can_piggyback(pvc, cip, target_threads)) {
2200 			spin_unlock(&pvc->lock);
2201 			continue;
2202 		}
2203 		kvmppc_core_end_stolen(pvc);
2204 		pvc->vcore_state = VCORE_PIGGYBACK;
2205 		if (cip->total_threads >= target_threads)
2206 			break;
2207 	}
2208 	spin_unlock(&lp->lock);
2209 }
2210 
2211 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2212 {
2213 	int still_running = 0;
2214 	u64 now;
2215 	long ret;
2216 	struct kvm_vcpu *vcpu, *vnext;
2217 
2218 	spin_lock(&vc->lock);
2219 	now = get_tb();
2220 	list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2221 				 arch.run_list) {
2222 		/* cancel pending dec exception if dec is positive */
2223 		if (now < vcpu->arch.dec_expires &&
2224 		    kvmppc_core_pending_dec(vcpu))
2225 			kvmppc_core_dequeue_dec(vcpu);
2226 
2227 		trace_kvm_guest_exit(vcpu);
2228 
2229 		ret = RESUME_GUEST;
2230 		if (vcpu->arch.trap)
2231 			ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2232 						    vcpu->arch.run_task);
2233 
2234 		vcpu->arch.ret = ret;
2235 		vcpu->arch.trap = 0;
2236 
2237 		if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2238 			if (vcpu->arch.pending_exceptions)
2239 				kvmppc_core_prepare_to_enter(vcpu);
2240 			if (vcpu->arch.ceded)
2241 				kvmppc_set_timer(vcpu);
2242 			else
2243 				++still_running;
2244 		} else {
2245 			kvmppc_remove_runnable(vc, vcpu);
2246 			wake_up(&vcpu->arch.cpu_run);
2247 		}
2248 	}
2249 	list_del_init(&vc->preempt_list);
2250 	if (!is_master) {
2251 		if (still_running > 0) {
2252 			kvmppc_vcore_preempt(vc);
2253 		} else if (vc->runner) {
2254 			vc->vcore_state = VCORE_PREEMPT;
2255 			kvmppc_core_start_stolen(vc);
2256 		} else {
2257 			vc->vcore_state = VCORE_INACTIVE;
2258 		}
2259 		if (vc->n_runnable > 0 && vc->runner == NULL) {
2260 			/* make sure there's a candidate runner awake */
2261 			vcpu = list_first_entry(&vc->runnable_threads,
2262 						struct kvm_vcpu, arch.run_list);
2263 			wake_up(&vcpu->arch.cpu_run);
2264 		}
2265 	}
2266 	spin_unlock(&vc->lock);
2267 }
2268 
2269 /*
2270  * Run a set of guest threads on a physical core.
2271  * Called with vc->lock held.
2272  */
2273 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2274 {
2275 	struct kvm_vcpu *vcpu, *vnext;
2276 	int i;
2277 	int srcu_idx;
2278 	struct core_info core_info;
2279 	struct kvmppc_vcore *pvc, *vcnext;
2280 	struct kvm_split_mode split_info, *sip;
2281 	int split, subcore_size, active;
2282 	int sub;
2283 	bool thr0_done;
2284 	unsigned long cmd_bit, stat_bit;
2285 	int pcpu, thr;
2286 	int target_threads;
2287 
2288 	/*
2289 	 * Remove from the list any threads that have a signal pending
2290 	 * or need a VPA update done
2291 	 */
2292 	prepare_threads(vc);
2293 
2294 	/* if the runner is no longer runnable, let the caller pick a new one */
2295 	if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2296 		return;
2297 
2298 	/*
2299 	 * Initialize *vc.
2300 	 */
2301 	init_master_vcore(vc);
2302 	vc->preempt_tb = TB_NIL;
2303 
2304 	/*
2305 	 * Make sure we are running on primary threads, and that secondary
2306 	 * threads are offline.  Also check if the number of threads in this
2307 	 * guest are greater than the current system threads per guest.
2308 	 */
2309 	if ((threads_per_core > 1) &&
2310 	    ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
2311 		list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2312 					 arch.run_list) {
2313 			vcpu->arch.ret = -EBUSY;
2314 			kvmppc_remove_runnable(vc, vcpu);
2315 			wake_up(&vcpu->arch.cpu_run);
2316 		}
2317 		goto out;
2318 	}
2319 
2320 	/*
2321 	 * See if we could run any other vcores on the physical core
2322 	 * along with this one.
2323 	 */
2324 	init_core_info(&core_info, vc);
2325 	pcpu = smp_processor_id();
2326 	target_threads = threads_per_subcore;
2327 	if (target_smt_mode && target_smt_mode < target_threads)
2328 		target_threads = target_smt_mode;
2329 	if (vc->num_threads < target_threads)
2330 		collect_piggybacks(&core_info, target_threads);
2331 
2332 	/* Decide on micro-threading (split-core) mode */
2333 	subcore_size = threads_per_subcore;
2334 	cmd_bit = stat_bit = 0;
2335 	split = core_info.n_subcores;
2336 	sip = NULL;
2337 	if (split > 1) {
2338 		/* threads_per_subcore must be MAX_SMT_THREADS (8) here */
2339 		if (split == 2 && (dynamic_mt_modes & 2)) {
2340 			cmd_bit = HID0_POWER8_1TO2LPAR;
2341 			stat_bit = HID0_POWER8_2LPARMODE;
2342 		} else {
2343 			split = 4;
2344 			cmd_bit = HID0_POWER8_1TO4LPAR;
2345 			stat_bit = HID0_POWER8_4LPARMODE;
2346 		}
2347 		subcore_size = MAX_SMT_THREADS / split;
2348 		sip = &split_info;
2349 		memset(&split_info, 0, sizeof(split_info));
2350 		split_info.rpr = mfspr(SPRN_RPR);
2351 		split_info.pmmar = mfspr(SPRN_PMMAR);
2352 		split_info.ldbar = mfspr(SPRN_LDBAR);
2353 		split_info.subcore_size = subcore_size;
2354 		for (sub = 0; sub < core_info.n_subcores; ++sub)
2355 			split_info.master_vcs[sub] =
2356 				list_first_entry(&core_info.vcs[sub],
2357 					struct kvmppc_vcore, preempt_list);
2358 		/* order writes to split_info before kvm_split_mode pointer */
2359 		smp_wmb();
2360 	}
2361 	pcpu = smp_processor_id();
2362 	for (thr = 0; thr < threads_per_subcore; ++thr)
2363 		paca[pcpu + thr].kvm_hstate.kvm_split_mode = sip;
2364 
2365 	/* Initiate micro-threading (split-core) if required */
2366 	if (cmd_bit) {
2367 		unsigned long hid0 = mfspr(SPRN_HID0);
2368 
2369 		hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2370 		mb();
2371 		mtspr(SPRN_HID0, hid0);
2372 		isync();
2373 		for (;;) {
2374 			hid0 = mfspr(SPRN_HID0);
2375 			if (hid0 & stat_bit)
2376 				break;
2377 			cpu_relax();
2378 		}
2379 	}
2380 
2381 	/* Start all the threads */
2382 	active = 0;
2383 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
2384 		thr = subcore_thread_map[sub];
2385 		thr0_done = false;
2386 		active |= 1 << thr;
2387 		list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list) {
2388 			pvc->pcpu = pcpu + thr;
2389 			list_for_each_entry(vcpu, &pvc->runnable_threads,
2390 					    arch.run_list) {
2391 				kvmppc_start_thread(vcpu, pvc);
2392 				kvmppc_create_dtl_entry(vcpu, pvc);
2393 				trace_kvm_guest_enter(vcpu);
2394 				if (!vcpu->arch.ptid)
2395 					thr0_done = true;
2396 				active |= 1 << (thr + vcpu->arch.ptid);
2397 			}
2398 			/*
2399 			 * We need to start the first thread of each subcore
2400 			 * even if it doesn't have a vcpu.
2401 			 */
2402 			if (pvc->master_vcore == pvc && !thr0_done)
2403 				kvmppc_start_thread(NULL, pvc);
2404 			thr += pvc->num_threads;
2405 		}
2406 	}
2407 
2408 	/*
2409 	 * Ensure that split_info.do_nap is set after setting
2410 	 * the vcore pointer in the PACA of the secondaries.
2411 	 */
2412 	smp_mb();
2413 	if (cmd_bit)
2414 		split_info.do_nap = 1;	/* ask secondaries to nap when done */
2415 
2416 	/*
2417 	 * When doing micro-threading, poke the inactive threads as well.
2418 	 * This gets them to the nap instruction after kvm_do_nap,
2419 	 * which reduces the time taken to unsplit later.
2420 	 */
2421 	if (split > 1)
2422 		for (thr = 1; thr < threads_per_subcore; ++thr)
2423 			if (!(active & (1 << thr)))
2424 				kvmppc_ipi_thread(pcpu + thr);
2425 
2426 	vc->vcore_state = VCORE_RUNNING;
2427 	preempt_disable();
2428 
2429 	trace_kvmppc_run_core(vc, 0);
2430 
2431 	for (sub = 0; sub < core_info.n_subcores; ++sub)
2432 		list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list)
2433 			spin_unlock(&pvc->lock);
2434 
2435 	kvm_guest_enter();
2436 
2437 	srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2438 
2439 	__kvmppc_vcore_entry();
2440 
2441 	srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2442 
2443 	spin_lock(&vc->lock);
2444 	/* prevent other vcpu threads from doing kvmppc_start_thread() now */
2445 	vc->vcore_state = VCORE_EXITING;
2446 
2447 	/* wait for secondary threads to finish writing their state to memory */
2448 	kvmppc_wait_for_nap();
2449 
2450 	/* Return to whole-core mode if we split the core earlier */
2451 	if (split > 1) {
2452 		unsigned long hid0 = mfspr(SPRN_HID0);
2453 		unsigned long loops = 0;
2454 
2455 		hid0 &= ~HID0_POWER8_DYNLPARDIS;
2456 		stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
2457 		mb();
2458 		mtspr(SPRN_HID0, hid0);
2459 		isync();
2460 		for (;;) {
2461 			hid0 = mfspr(SPRN_HID0);
2462 			if (!(hid0 & stat_bit))
2463 				break;
2464 			cpu_relax();
2465 			++loops;
2466 		}
2467 		split_info.do_nap = 0;
2468 	}
2469 
2470 	/* Let secondaries go back to the offline loop */
2471 	for (i = 0; i < threads_per_subcore; ++i) {
2472 		kvmppc_release_hwthread(pcpu + i);
2473 		if (sip && sip->napped[i])
2474 			kvmppc_ipi_thread(pcpu + i);
2475 	}
2476 
2477 	spin_unlock(&vc->lock);
2478 
2479 	/* make sure updates to secondary vcpu structs are visible now */
2480 	smp_mb();
2481 	kvm_guest_exit();
2482 
2483 	for (sub = 0; sub < core_info.n_subcores; ++sub)
2484 		list_for_each_entry_safe(pvc, vcnext, &core_info.vcs[sub],
2485 					 preempt_list)
2486 			post_guest_process(pvc, pvc == vc);
2487 
2488 	spin_lock(&vc->lock);
2489 	preempt_enable();
2490 
2491  out:
2492 	vc->vcore_state = VCORE_INACTIVE;
2493 	trace_kvmppc_run_core(vc, 1);
2494 }
2495 
2496 /*
2497  * Wait for some other vcpu thread to execute us, and
2498  * wake us up when we need to handle something in the host.
2499  */
2500 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
2501 				 struct kvm_vcpu *vcpu, int wait_state)
2502 {
2503 	DEFINE_WAIT(wait);
2504 
2505 	prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
2506 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2507 		spin_unlock(&vc->lock);
2508 		schedule();
2509 		spin_lock(&vc->lock);
2510 	}
2511 	finish_wait(&vcpu->arch.cpu_run, &wait);
2512 }
2513 
2514 /*
2515  * All the vcpus in this vcore are idle, so wait for a decrementer
2516  * or external interrupt to one of the vcpus.  vc->lock is held.
2517  */
2518 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
2519 {
2520 	struct kvm_vcpu *vcpu;
2521 	int do_sleep = 1;
2522 
2523 	DEFINE_WAIT(wait);
2524 
2525 	prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
2526 
2527 	/*
2528 	 * Check one last time for pending exceptions and ceded state after
2529 	 * we put ourselves on the wait queue
2530 	 */
2531 	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
2532 		if (vcpu->arch.pending_exceptions || !vcpu->arch.ceded) {
2533 			do_sleep = 0;
2534 			break;
2535 		}
2536 	}
2537 
2538 	if (!do_sleep) {
2539 		finish_wait(&vc->wq, &wait);
2540 		return;
2541 	}
2542 
2543 	vc->vcore_state = VCORE_SLEEPING;
2544 	trace_kvmppc_vcore_blocked(vc, 0);
2545 	spin_unlock(&vc->lock);
2546 	schedule();
2547 	finish_wait(&vc->wq, &wait);
2548 	spin_lock(&vc->lock);
2549 	vc->vcore_state = VCORE_INACTIVE;
2550 	trace_kvmppc_vcore_blocked(vc, 1);
2551 }
2552 
2553 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2554 {
2555 	int n_ceded;
2556 	struct kvmppc_vcore *vc;
2557 	struct kvm_vcpu *v, *vn;
2558 
2559 	trace_kvmppc_run_vcpu_enter(vcpu);
2560 
2561 	kvm_run->exit_reason = 0;
2562 	vcpu->arch.ret = RESUME_GUEST;
2563 	vcpu->arch.trap = 0;
2564 	kvmppc_update_vpas(vcpu);
2565 
2566 	/*
2567 	 * Synchronize with other threads in this virtual core
2568 	 */
2569 	vc = vcpu->arch.vcore;
2570 	spin_lock(&vc->lock);
2571 	vcpu->arch.ceded = 0;
2572 	vcpu->arch.run_task = current;
2573 	vcpu->arch.kvm_run = kvm_run;
2574 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
2575 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
2576 	vcpu->arch.busy_preempt = TB_NIL;
2577 	list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
2578 	++vc->n_runnable;
2579 
2580 	/*
2581 	 * This happens the first time this is called for a vcpu.
2582 	 * If the vcore is already running, we may be able to start
2583 	 * this thread straight away and have it join in.
2584 	 */
2585 	if (!signal_pending(current)) {
2586 		if (vc->vcore_state == VCORE_PIGGYBACK) {
2587 			struct kvmppc_vcore *mvc = vc->master_vcore;
2588 			if (spin_trylock(&mvc->lock)) {
2589 				if (mvc->vcore_state == VCORE_RUNNING &&
2590 				    !VCORE_IS_EXITING(mvc)) {
2591 					kvmppc_create_dtl_entry(vcpu, vc);
2592 					kvmppc_start_thread(vcpu, vc);
2593 					trace_kvm_guest_enter(vcpu);
2594 				}
2595 				spin_unlock(&mvc->lock);
2596 			}
2597 		} else if (vc->vcore_state == VCORE_RUNNING &&
2598 			   !VCORE_IS_EXITING(vc)) {
2599 			kvmppc_create_dtl_entry(vcpu, vc);
2600 			kvmppc_start_thread(vcpu, vc);
2601 			trace_kvm_guest_enter(vcpu);
2602 		} else if (vc->vcore_state == VCORE_SLEEPING) {
2603 			wake_up(&vc->wq);
2604 		}
2605 
2606 	}
2607 
2608 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2609 	       !signal_pending(current)) {
2610 		if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2611 			kvmppc_vcore_end_preempt(vc);
2612 
2613 		if (vc->vcore_state != VCORE_INACTIVE) {
2614 			kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
2615 			continue;
2616 		}
2617 		list_for_each_entry_safe(v, vn, &vc->runnable_threads,
2618 					 arch.run_list) {
2619 			kvmppc_core_prepare_to_enter(v);
2620 			if (signal_pending(v->arch.run_task)) {
2621 				kvmppc_remove_runnable(vc, v);
2622 				v->stat.signal_exits++;
2623 				v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
2624 				v->arch.ret = -EINTR;
2625 				wake_up(&v->arch.cpu_run);
2626 			}
2627 		}
2628 		if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2629 			break;
2630 		n_ceded = 0;
2631 		list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
2632 			if (!v->arch.pending_exceptions)
2633 				n_ceded += v->arch.ceded;
2634 			else
2635 				v->arch.ceded = 0;
2636 		}
2637 		vc->runner = vcpu;
2638 		if (n_ceded == vc->n_runnable) {
2639 			kvmppc_vcore_blocked(vc);
2640 		} else if (need_resched()) {
2641 			kvmppc_vcore_preempt(vc);
2642 			/* Let something else run */
2643 			cond_resched_lock(&vc->lock);
2644 			if (vc->vcore_state == VCORE_PREEMPT)
2645 				kvmppc_vcore_end_preempt(vc);
2646 		} else {
2647 			kvmppc_run_core(vc);
2648 		}
2649 		vc->runner = NULL;
2650 	}
2651 
2652 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2653 	       (vc->vcore_state == VCORE_RUNNING ||
2654 		vc->vcore_state == VCORE_EXITING ||
2655 		vc->vcore_state == VCORE_PIGGYBACK))
2656 		kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
2657 
2658 	if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2659 		kvmppc_vcore_end_preempt(vc);
2660 
2661 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2662 		kvmppc_remove_runnable(vc, vcpu);
2663 		vcpu->stat.signal_exits++;
2664 		kvm_run->exit_reason = KVM_EXIT_INTR;
2665 		vcpu->arch.ret = -EINTR;
2666 	}
2667 
2668 	if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
2669 		/* Wake up some vcpu to run the core */
2670 		v = list_first_entry(&vc->runnable_threads,
2671 				     struct kvm_vcpu, arch.run_list);
2672 		wake_up(&v->arch.cpu_run);
2673 	}
2674 
2675 	trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
2676 	spin_unlock(&vc->lock);
2677 	return vcpu->arch.ret;
2678 }
2679 
2680 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
2681 {
2682 	int r;
2683 	int srcu_idx;
2684 
2685 	if (!vcpu->arch.sane) {
2686 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2687 		return -EINVAL;
2688 	}
2689 
2690 	kvmppc_core_prepare_to_enter(vcpu);
2691 
2692 	/* No need to go into the guest when all we'll do is come back out */
2693 	if (signal_pending(current)) {
2694 		run->exit_reason = KVM_EXIT_INTR;
2695 		return -EINTR;
2696 	}
2697 
2698 	atomic_inc(&vcpu->kvm->arch.vcpus_running);
2699 	/* Order vcpus_running vs. hpte_setup_done, see kvmppc_alloc_reset_hpt */
2700 	smp_mb();
2701 
2702 	/* On the first time here, set up HTAB and VRMA */
2703 	if (!vcpu->kvm->arch.hpte_setup_done) {
2704 		r = kvmppc_hv_setup_htab_rma(vcpu);
2705 		if (r)
2706 			goto out;
2707 	}
2708 
2709 	flush_fp_to_thread(current);
2710 	flush_altivec_to_thread(current);
2711 	flush_vsx_to_thread(current);
2712 	vcpu->arch.wqp = &vcpu->arch.vcore->wq;
2713 	vcpu->arch.pgdir = current->mm->pgd;
2714 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2715 
2716 	do {
2717 		r = kvmppc_run_vcpu(run, vcpu);
2718 
2719 		if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
2720 		    !(vcpu->arch.shregs.msr & MSR_PR)) {
2721 			trace_kvm_hcall_enter(vcpu);
2722 			r = kvmppc_pseries_do_hcall(vcpu);
2723 			trace_kvm_hcall_exit(vcpu, r);
2724 			kvmppc_core_prepare_to_enter(vcpu);
2725 		} else if (r == RESUME_PAGE_FAULT) {
2726 			srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2727 			r = kvmppc_book3s_hv_page_fault(run, vcpu,
2728 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
2729 			srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2730 		}
2731 	} while (is_kvmppc_resume_guest(r));
2732 
2733  out:
2734 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2735 	atomic_dec(&vcpu->kvm->arch.vcpus_running);
2736 	return r;
2737 }
2738 
2739 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
2740 				     int linux_psize)
2741 {
2742 	struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
2743 
2744 	if (!def->shift)
2745 		return;
2746 	(*sps)->page_shift = def->shift;
2747 	(*sps)->slb_enc = def->sllp;
2748 	(*sps)->enc[0].page_shift = def->shift;
2749 	(*sps)->enc[0].pte_enc = def->penc[linux_psize];
2750 	/*
2751 	 * Add 16MB MPSS support if host supports it
2752 	 */
2753 	if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
2754 		(*sps)->enc[1].page_shift = 24;
2755 		(*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
2756 	}
2757 	(*sps)++;
2758 }
2759 
2760 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
2761 					 struct kvm_ppc_smmu_info *info)
2762 {
2763 	struct kvm_ppc_one_seg_page_size *sps;
2764 
2765 	info->flags = KVM_PPC_PAGE_SIZES_REAL;
2766 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
2767 		info->flags |= KVM_PPC_1T_SEGMENTS;
2768 	info->slb_size = mmu_slb_size;
2769 
2770 	/* We only support these sizes for now, and no muti-size segments */
2771 	sps = &info->sps[0];
2772 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
2773 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
2774 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
2775 
2776 	return 0;
2777 }
2778 
2779 /*
2780  * Get (and clear) the dirty memory log for a memory slot.
2781  */
2782 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
2783 					 struct kvm_dirty_log *log)
2784 {
2785 	struct kvm_memslots *slots;
2786 	struct kvm_memory_slot *memslot;
2787 	int r;
2788 	unsigned long n;
2789 
2790 	mutex_lock(&kvm->slots_lock);
2791 
2792 	r = -EINVAL;
2793 	if (log->slot >= KVM_USER_MEM_SLOTS)
2794 		goto out;
2795 
2796 	slots = kvm_memslots(kvm);
2797 	memslot = id_to_memslot(slots, log->slot);
2798 	r = -ENOENT;
2799 	if (!memslot->dirty_bitmap)
2800 		goto out;
2801 
2802 	n = kvm_dirty_bitmap_bytes(memslot);
2803 	memset(memslot->dirty_bitmap, 0, n);
2804 
2805 	r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
2806 	if (r)
2807 		goto out;
2808 
2809 	r = -EFAULT;
2810 	if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
2811 		goto out;
2812 
2813 	r = 0;
2814 out:
2815 	mutex_unlock(&kvm->slots_lock);
2816 	return r;
2817 }
2818 
2819 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
2820 					struct kvm_memory_slot *dont)
2821 {
2822 	if (!dont || free->arch.rmap != dont->arch.rmap) {
2823 		vfree(free->arch.rmap);
2824 		free->arch.rmap = NULL;
2825 	}
2826 }
2827 
2828 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
2829 					 unsigned long npages)
2830 {
2831 	slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
2832 	if (!slot->arch.rmap)
2833 		return -ENOMEM;
2834 
2835 	return 0;
2836 }
2837 
2838 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
2839 					struct kvm_memory_slot *memslot,
2840 					const struct kvm_userspace_memory_region *mem)
2841 {
2842 	return 0;
2843 }
2844 
2845 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
2846 				const struct kvm_userspace_memory_region *mem,
2847 				const struct kvm_memory_slot *old,
2848 				const struct kvm_memory_slot *new)
2849 {
2850 	unsigned long npages = mem->memory_size >> PAGE_SHIFT;
2851 	struct kvm_memslots *slots;
2852 	struct kvm_memory_slot *memslot;
2853 
2854 	if (npages && old->npages) {
2855 		/*
2856 		 * If modifying a memslot, reset all the rmap dirty bits.
2857 		 * If this is a new memslot, we don't need to do anything
2858 		 * since the rmap array starts out as all zeroes,
2859 		 * i.e. no pages are dirty.
2860 		 */
2861 		slots = kvm_memslots(kvm);
2862 		memslot = id_to_memslot(slots, mem->slot);
2863 		kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
2864 	}
2865 }
2866 
2867 /*
2868  * Update LPCR values in kvm->arch and in vcores.
2869  * Caller must hold kvm->lock.
2870  */
2871 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
2872 {
2873 	long int i;
2874 	u32 cores_done = 0;
2875 
2876 	if ((kvm->arch.lpcr & mask) == lpcr)
2877 		return;
2878 
2879 	kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
2880 
2881 	for (i = 0; i < KVM_MAX_VCORES; ++i) {
2882 		struct kvmppc_vcore *vc = kvm->arch.vcores[i];
2883 		if (!vc)
2884 			continue;
2885 		spin_lock(&vc->lock);
2886 		vc->lpcr = (vc->lpcr & ~mask) | lpcr;
2887 		spin_unlock(&vc->lock);
2888 		if (++cores_done >= kvm->arch.online_vcores)
2889 			break;
2890 	}
2891 }
2892 
2893 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
2894 {
2895 	return;
2896 }
2897 
2898 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
2899 {
2900 	int err = 0;
2901 	struct kvm *kvm = vcpu->kvm;
2902 	unsigned long hva;
2903 	struct kvm_memory_slot *memslot;
2904 	struct vm_area_struct *vma;
2905 	unsigned long lpcr = 0, senc;
2906 	unsigned long psize, porder;
2907 	int srcu_idx;
2908 
2909 	mutex_lock(&kvm->lock);
2910 	if (kvm->arch.hpte_setup_done)
2911 		goto out;	/* another vcpu beat us to it */
2912 
2913 	/* Allocate hashed page table (if not done already) and reset it */
2914 	if (!kvm->arch.hpt_virt) {
2915 		err = kvmppc_alloc_hpt(kvm, NULL);
2916 		if (err) {
2917 			pr_err("KVM: Couldn't alloc HPT\n");
2918 			goto out;
2919 		}
2920 	}
2921 
2922 	/* Look up the memslot for guest physical address 0 */
2923 	srcu_idx = srcu_read_lock(&kvm->srcu);
2924 	memslot = gfn_to_memslot(kvm, 0);
2925 
2926 	/* We must have some memory at 0 by now */
2927 	err = -EINVAL;
2928 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
2929 		goto out_srcu;
2930 
2931 	/* Look up the VMA for the start of this memory slot */
2932 	hva = memslot->userspace_addr;
2933 	down_read(&current->mm->mmap_sem);
2934 	vma = find_vma(current->mm, hva);
2935 	if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
2936 		goto up_out;
2937 
2938 	psize = vma_kernel_pagesize(vma);
2939 	porder = __ilog2(psize);
2940 
2941 	up_read(&current->mm->mmap_sem);
2942 
2943 	/* We can handle 4k, 64k or 16M pages in the VRMA */
2944 	err = -EINVAL;
2945 	if (!(psize == 0x1000 || psize == 0x10000 ||
2946 	      psize == 0x1000000))
2947 		goto out_srcu;
2948 
2949 	/* Update VRMASD field in the LPCR */
2950 	senc = slb_pgsize_encoding(psize);
2951 	kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
2952 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
2953 	/* the -4 is to account for senc values starting at 0x10 */
2954 	lpcr = senc << (LPCR_VRMASD_SH - 4);
2955 
2956 	/* Create HPTEs in the hash page table for the VRMA */
2957 	kvmppc_map_vrma(vcpu, memslot, porder);
2958 
2959 	kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
2960 
2961 	/* Order updates to kvm->arch.lpcr etc. vs. hpte_setup_done */
2962 	smp_wmb();
2963 	kvm->arch.hpte_setup_done = 1;
2964 	err = 0;
2965  out_srcu:
2966 	srcu_read_unlock(&kvm->srcu, srcu_idx);
2967  out:
2968 	mutex_unlock(&kvm->lock);
2969 	return err;
2970 
2971  up_out:
2972 	up_read(&current->mm->mmap_sem);
2973 	goto out_srcu;
2974 }
2975 
2976 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
2977 {
2978 	unsigned long lpcr, lpid;
2979 	char buf[32];
2980 
2981 	/* Allocate the guest's logical partition ID */
2982 
2983 	lpid = kvmppc_alloc_lpid();
2984 	if ((long)lpid < 0)
2985 		return -ENOMEM;
2986 	kvm->arch.lpid = lpid;
2987 
2988 	/*
2989 	 * Since we don't flush the TLB when tearing down a VM,
2990 	 * and this lpid might have previously been used,
2991 	 * make sure we flush on each core before running the new VM.
2992 	 */
2993 	cpumask_setall(&kvm->arch.need_tlb_flush);
2994 
2995 	/* Start out with the default set of hcalls enabled */
2996 	memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
2997 	       sizeof(kvm->arch.enabled_hcalls));
2998 
2999 	kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
3000 
3001 	/* Init LPCR for virtual RMA mode */
3002 	kvm->arch.host_lpid = mfspr(SPRN_LPID);
3003 	kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
3004 	lpcr &= LPCR_PECE | LPCR_LPES;
3005 	lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3006 		LPCR_VPM0 | LPCR_VPM1;
3007 	kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3008 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
3009 	/* On POWER8 turn on online bit to enable PURR/SPURR */
3010 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
3011 		lpcr |= LPCR_ONL;
3012 	kvm->arch.lpcr = lpcr;
3013 
3014 	/*
3015 	 * Track that we now have a HV mode VM active. This blocks secondary
3016 	 * CPU threads from coming online.
3017 	 */
3018 	kvm_hv_vm_activated();
3019 
3020 	/*
3021 	 * Create a debugfs directory for the VM
3022 	 */
3023 	snprintf(buf, sizeof(buf), "vm%d", current->pid);
3024 	kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
3025 	if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
3026 		kvmppc_mmu_debugfs_init(kvm);
3027 
3028 	return 0;
3029 }
3030 
3031 static void kvmppc_free_vcores(struct kvm *kvm)
3032 {
3033 	long int i;
3034 
3035 	for (i = 0; i < KVM_MAX_VCORES; ++i)
3036 		kfree(kvm->arch.vcores[i]);
3037 	kvm->arch.online_vcores = 0;
3038 }
3039 
3040 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
3041 {
3042 	debugfs_remove_recursive(kvm->arch.debugfs_dir);
3043 
3044 	kvm_hv_vm_deactivated();
3045 
3046 	kvmppc_free_vcores(kvm);
3047 
3048 	kvmppc_free_hpt(kvm);
3049 }
3050 
3051 /* We don't need to emulate any privileged instructions or dcbz */
3052 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
3053 				     unsigned int inst, int *advance)
3054 {
3055 	return EMULATE_FAIL;
3056 }
3057 
3058 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
3059 					ulong spr_val)
3060 {
3061 	return EMULATE_FAIL;
3062 }
3063 
3064 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
3065 					ulong *spr_val)
3066 {
3067 	return EMULATE_FAIL;
3068 }
3069 
3070 static int kvmppc_core_check_processor_compat_hv(void)
3071 {
3072 	if (!cpu_has_feature(CPU_FTR_HVMODE) ||
3073 	    !cpu_has_feature(CPU_FTR_ARCH_206))
3074 		return -EIO;
3075 	return 0;
3076 }
3077 
3078 static long kvm_arch_vm_ioctl_hv(struct file *filp,
3079 				 unsigned int ioctl, unsigned long arg)
3080 {
3081 	struct kvm *kvm __maybe_unused = filp->private_data;
3082 	void __user *argp = (void __user *)arg;
3083 	long r;
3084 
3085 	switch (ioctl) {
3086 
3087 	case KVM_PPC_ALLOCATE_HTAB: {
3088 		u32 htab_order;
3089 
3090 		r = -EFAULT;
3091 		if (get_user(htab_order, (u32 __user *)argp))
3092 			break;
3093 		r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
3094 		if (r)
3095 			break;
3096 		r = -EFAULT;
3097 		if (put_user(htab_order, (u32 __user *)argp))
3098 			break;
3099 		r = 0;
3100 		break;
3101 	}
3102 
3103 	case KVM_PPC_GET_HTAB_FD: {
3104 		struct kvm_get_htab_fd ghf;
3105 
3106 		r = -EFAULT;
3107 		if (copy_from_user(&ghf, argp, sizeof(ghf)))
3108 			break;
3109 		r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
3110 		break;
3111 	}
3112 
3113 	default:
3114 		r = -ENOTTY;
3115 	}
3116 
3117 	return r;
3118 }
3119 
3120 /*
3121  * List of hcall numbers to enable by default.
3122  * For compatibility with old userspace, we enable by default
3123  * all hcalls that were implemented before the hcall-enabling
3124  * facility was added.  Note this list should not include H_RTAS.
3125  */
3126 static unsigned int default_hcall_list[] = {
3127 	H_REMOVE,
3128 	H_ENTER,
3129 	H_READ,
3130 	H_PROTECT,
3131 	H_BULK_REMOVE,
3132 	H_GET_TCE,
3133 	H_PUT_TCE,
3134 	H_SET_DABR,
3135 	H_SET_XDABR,
3136 	H_CEDE,
3137 	H_PROD,
3138 	H_CONFER,
3139 	H_REGISTER_VPA,
3140 #ifdef CONFIG_KVM_XICS
3141 	H_EOI,
3142 	H_CPPR,
3143 	H_IPI,
3144 	H_IPOLL,
3145 	H_XIRR,
3146 	H_XIRR_X,
3147 #endif
3148 	0
3149 };
3150 
3151 static void init_default_hcalls(void)
3152 {
3153 	int i;
3154 	unsigned int hcall;
3155 
3156 	for (i = 0; default_hcall_list[i]; ++i) {
3157 		hcall = default_hcall_list[i];
3158 		WARN_ON(!kvmppc_hcall_impl_hv(hcall));
3159 		__set_bit(hcall / 4, default_enabled_hcalls);
3160 	}
3161 }
3162 
3163 static struct kvmppc_ops kvm_ops_hv = {
3164 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
3165 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
3166 	.get_one_reg = kvmppc_get_one_reg_hv,
3167 	.set_one_reg = kvmppc_set_one_reg_hv,
3168 	.vcpu_load   = kvmppc_core_vcpu_load_hv,
3169 	.vcpu_put    = kvmppc_core_vcpu_put_hv,
3170 	.set_msr     = kvmppc_set_msr_hv,
3171 	.vcpu_run    = kvmppc_vcpu_run_hv,
3172 	.vcpu_create = kvmppc_core_vcpu_create_hv,
3173 	.vcpu_free   = kvmppc_core_vcpu_free_hv,
3174 	.check_requests = kvmppc_core_check_requests_hv,
3175 	.get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
3176 	.flush_memslot  = kvmppc_core_flush_memslot_hv,
3177 	.prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
3178 	.commit_memory_region  = kvmppc_core_commit_memory_region_hv,
3179 	.unmap_hva = kvm_unmap_hva_hv,
3180 	.unmap_hva_range = kvm_unmap_hva_range_hv,
3181 	.age_hva  = kvm_age_hva_hv,
3182 	.test_age_hva = kvm_test_age_hva_hv,
3183 	.set_spte_hva = kvm_set_spte_hva_hv,
3184 	.mmu_destroy  = kvmppc_mmu_destroy_hv,
3185 	.free_memslot = kvmppc_core_free_memslot_hv,
3186 	.create_memslot = kvmppc_core_create_memslot_hv,
3187 	.init_vm =  kvmppc_core_init_vm_hv,
3188 	.destroy_vm = kvmppc_core_destroy_vm_hv,
3189 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
3190 	.emulate_op = kvmppc_core_emulate_op_hv,
3191 	.emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
3192 	.emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
3193 	.fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
3194 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
3195 	.hcall_implemented = kvmppc_hcall_impl_hv,
3196 };
3197 
3198 static int kvmppc_book3s_init_hv(void)
3199 {
3200 	int r;
3201 	/*
3202 	 * FIXME!! Do we need to check on all cpus ?
3203 	 */
3204 	r = kvmppc_core_check_processor_compat_hv();
3205 	if (r < 0)
3206 		return -ENODEV;
3207 
3208 	kvm_ops_hv.owner = THIS_MODULE;
3209 	kvmppc_hv_ops = &kvm_ops_hv;
3210 
3211 	init_default_hcalls();
3212 
3213 	init_vcore_lists();
3214 
3215 	r = kvmppc_mmu_hv_init();
3216 	return r;
3217 }
3218 
3219 static void kvmppc_book3s_exit_hv(void)
3220 {
3221 	kvmppc_hv_ops = NULL;
3222 }
3223 
3224 module_init(kvmppc_book3s_init_hv);
3225 module_exit(kvmppc_book3s_exit_hv);
3226 MODULE_LICENSE("GPL");
3227 MODULE_ALIAS_MISCDEV(KVM_MINOR);
3228 MODULE_ALIAS("devname:kvm");
3229