xref: /linux/arch/powerpc/kvm/book3s_hv.c (revision 10accd2e6890b57db8e717e9aee91b791f90fe14)
1 /*
2  * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
3  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
4  *
5  * Authors:
6  *    Paul Mackerras <paulus@au1.ibm.com>
7  *    Alexander Graf <agraf@suse.de>
8  *    Kevin Wolf <mail@kevin-wolf.de>
9  *
10  * Description: KVM functions specific to running on Book 3S
11  * processors in hypervisor mode (specifically POWER7 and later).
12  *
13  * This file is derived from arch/powerpc/kvm/book3s.c,
14  * by Alexander Graf <agraf@suse.de>.
15  *
16  * This program is free software; you can redistribute it and/or modify
17  * it under the terms of the GNU General Public License, version 2, as
18  * published by the Free Software Foundation.
19  */
20 
21 #include <linux/kvm_host.h>
22 #include <linux/err.h>
23 #include <linux/slab.h>
24 #include <linux/preempt.h>
25 #include <linux/sched.h>
26 #include <linux/delay.h>
27 #include <linux/export.h>
28 #include <linux/fs.h>
29 #include <linux/anon_inodes.h>
30 #include <linux/cpu.h>
31 #include <linux/cpumask.h>
32 #include <linux/spinlock.h>
33 #include <linux/page-flags.h>
34 #include <linux/srcu.h>
35 #include <linux/miscdevice.h>
36 #include <linux/debugfs.h>
37 
38 #include <asm/reg.h>
39 #include <asm/cputable.h>
40 #include <asm/cacheflush.h>
41 #include <asm/tlbflush.h>
42 #include <asm/uaccess.h>
43 #include <asm/io.h>
44 #include <asm/kvm_ppc.h>
45 #include <asm/kvm_book3s.h>
46 #include <asm/mmu_context.h>
47 #include <asm/lppaca.h>
48 #include <asm/processor.h>
49 #include <asm/cputhreads.h>
50 #include <asm/page.h>
51 #include <asm/hvcall.h>
52 #include <asm/switch_to.h>
53 #include <asm/smp.h>
54 #include <asm/dbell.h>
55 #include <asm/hmi.h>
56 #include <linux/gfp.h>
57 #include <linux/vmalloc.h>
58 #include <linux/highmem.h>
59 #include <linux/hugetlb.h>
60 #include <linux/module.h>
61 
62 #include "book3s.h"
63 
64 #define CREATE_TRACE_POINTS
65 #include "trace_hv.h"
66 
67 /* #define EXIT_DEBUG */
68 /* #define EXIT_DEBUG_SIMPLE */
69 /* #define EXIT_DEBUG_INT */
70 
71 /* Used to indicate that a guest page fault needs to be handled */
72 #define RESUME_PAGE_FAULT	(RESUME_GUEST | RESUME_FLAG_ARCH1)
73 
74 /* Used as a "null" value for timebase values */
75 #define TB_NIL	(~(u64)0)
76 
77 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
78 
79 static int dynamic_mt_modes = 6;
80 module_param(dynamic_mt_modes, int, S_IRUGO | S_IWUSR);
81 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
82 static int target_smt_mode;
83 module_param(target_smt_mode, int, S_IRUGO | S_IWUSR);
84 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
85 
86 #ifdef CONFIG_KVM_XICS
87 static struct kernel_param_ops module_param_ops = {
88 	.set = param_set_int,
89 	.get = param_get_int,
90 };
91 
92 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect,
93 							S_IRUGO | S_IWUSR);
94 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
95 #endif
96 
97 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
98 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
99 
100 static bool kvmppc_ipi_thread(int cpu)
101 {
102 	/* On POWER8 for IPIs to threads in the same core, use msgsnd */
103 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
104 		preempt_disable();
105 		if (cpu_first_thread_sibling(cpu) ==
106 		    cpu_first_thread_sibling(smp_processor_id())) {
107 			unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
108 			msg |= cpu_thread_in_core(cpu);
109 			smp_mb();
110 			__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
111 			preempt_enable();
112 			return true;
113 		}
114 		preempt_enable();
115 	}
116 
117 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
118 	if (cpu >= 0 && cpu < nr_cpu_ids && paca[cpu].kvm_hstate.xics_phys) {
119 		xics_wake_cpu(cpu);
120 		return true;
121 	}
122 #endif
123 
124 	return false;
125 }
126 
127 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
128 {
129 	int cpu;
130 	struct swait_queue_head *wqp;
131 
132 	wqp = kvm_arch_vcpu_wq(vcpu);
133 	if (swait_active(wqp)) {
134 		swake_up(wqp);
135 		++vcpu->stat.halt_wakeup;
136 	}
137 
138 	if (kvmppc_ipi_thread(vcpu->arch.thread_cpu))
139 		return;
140 
141 	/* CPU points to the first thread of the core */
142 	cpu = vcpu->cpu;
143 	if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
144 		smp_send_reschedule(cpu);
145 }
146 
147 /*
148  * We use the vcpu_load/put functions to measure stolen time.
149  * Stolen time is counted as time when either the vcpu is able to
150  * run as part of a virtual core, but the task running the vcore
151  * is preempted or sleeping, or when the vcpu needs something done
152  * in the kernel by the task running the vcpu, but that task is
153  * preempted or sleeping.  Those two things have to be counted
154  * separately, since one of the vcpu tasks will take on the job
155  * of running the core, and the other vcpu tasks in the vcore will
156  * sleep waiting for it to do that, but that sleep shouldn't count
157  * as stolen time.
158  *
159  * Hence we accumulate stolen time when the vcpu can run as part of
160  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
161  * needs its task to do other things in the kernel (for example,
162  * service a page fault) in busy_stolen.  We don't accumulate
163  * stolen time for a vcore when it is inactive, or for a vcpu
164  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
165  * a misnomer; it means that the vcpu task is not executing in
166  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
167  * the kernel.  We don't have any way of dividing up that time
168  * between time that the vcpu is genuinely stopped, time that
169  * the task is actively working on behalf of the vcpu, and time
170  * that the task is preempted, so we don't count any of it as
171  * stolen.
172  *
173  * Updates to busy_stolen are protected by arch.tbacct_lock;
174  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
175  * lock.  The stolen times are measured in units of timebase ticks.
176  * (Note that the != TB_NIL checks below are purely defensive;
177  * they should never fail.)
178  */
179 
180 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
181 {
182 	unsigned long flags;
183 
184 	spin_lock_irqsave(&vc->stoltb_lock, flags);
185 	vc->preempt_tb = mftb();
186 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
187 }
188 
189 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
190 {
191 	unsigned long flags;
192 
193 	spin_lock_irqsave(&vc->stoltb_lock, flags);
194 	if (vc->preempt_tb != TB_NIL) {
195 		vc->stolen_tb += mftb() - vc->preempt_tb;
196 		vc->preempt_tb = TB_NIL;
197 	}
198 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
199 }
200 
201 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
202 {
203 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
204 	unsigned long flags;
205 
206 	/*
207 	 * We can test vc->runner without taking the vcore lock,
208 	 * because only this task ever sets vc->runner to this
209 	 * vcpu, and once it is set to this vcpu, only this task
210 	 * ever sets it to NULL.
211 	 */
212 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
213 		kvmppc_core_end_stolen(vc);
214 
215 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
216 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
217 	    vcpu->arch.busy_preempt != TB_NIL) {
218 		vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
219 		vcpu->arch.busy_preempt = TB_NIL;
220 	}
221 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
222 }
223 
224 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
225 {
226 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
227 	unsigned long flags;
228 
229 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
230 		kvmppc_core_start_stolen(vc);
231 
232 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
233 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
234 		vcpu->arch.busy_preempt = mftb();
235 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
236 }
237 
238 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
239 {
240 	/*
241 	 * Check for illegal transactional state bit combination
242 	 * and if we find it, force the TS field to a safe state.
243 	 */
244 	if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
245 		msr &= ~MSR_TS_MASK;
246 	vcpu->arch.shregs.msr = msr;
247 	kvmppc_end_cede(vcpu);
248 }
249 
250 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
251 {
252 	vcpu->arch.pvr = pvr;
253 }
254 
255 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
256 {
257 	unsigned long pcr = 0;
258 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
259 
260 	if (arch_compat) {
261 		switch (arch_compat) {
262 		case PVR_ARCH_205:
263 			/*
264 			 * If an arch bit is set in PCR, all the defined
265 			 * higher-order arch bits also have to be set.
266 			 */
267 			pcr = PCR_ARCH_206 | PCR_ARCH_205;
268 			break;
269 		case PVR_ARCH_206:
270 		case PVR_ARCH_206p:
271 			pcr = PCR_ARCH_206;
272 			break;
273 		case PVR_ARCH_207:
274 			break;
275 		default:
276 			return -EINVAL;
277 		}
278 
279 		if (!cpu_has_feature(CPU_FTR_ARCH_207S)) {
280 			/* POWER7 can't emulate POWER8 */
281 			if (!(pcr & PCR_ARCH_206))
282 				return -EINVAL;
283 			pcr &= ~PCR_ARCH_206;
284 		}
285 	}
286 
287 	spin_lock(&vc->lock);
288 	vc->arch_compat = arch_compat;
289 	vc->pcr = pcr;
290 	spin_unlock(&vc->lock);
291 
292 	return 0;
293 }
294 
295 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
296 {
297 	int r;
298 
299 	pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
300 	pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
301 	       vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
302 	for (r = 0; r < 16; ++r)
303 		pr_err("r%2d = %.16lx  r%d = %.16lx\n",
304 		       r, kvmppc_get_gpr(vcpu, r),
305 		       r+16, kvmppc_get_gpr(vcpu, r+16));
306 	pr_err("ctr = %.16lx  lr  = %.16lx\n",
307 	       vcpu->arch.ctr, vcpu->arch.lr);
308 	pr_err("srr0 = %.16llx srr1 = %.16llx\n",
309 	       vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
310 	pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
311 	       vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
312 	pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
313 	       vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
314 	pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
315 	       vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
316 	pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
317 	pr_err("fault dar = %.16lx dsisr = %.8x\n",
318 	       vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
319 	pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
320 	for (r = 0; r < vcpu->arch.slb_max; ++r)
321 		pr_err("  ESID = %.16llx VSID = %.16llx\n",
322 		       vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
323 	pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
324 	       vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
325 	       vcpu->arch.last_inst);
326 }
327 
328 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
329 {
330 	struct kvm_vcpu *ret;
331 
332 	mutex_lock(&kvm->lock);
333 	ret = kvm_get_vcpu_by_id(kvm, id);
334 	mutex_unlock(&kvm->lock);
335 	return ret;
336 }
337 
338 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
339 {
340 	vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
341 	vpa->yield_count = cpu_to_be32(1);
342 }
343 
344 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
345 		   unsigned long addr, unsigned long len)
346 {
347 	/* check address is cacheline aligned */
348 	if (addr & (L1_CACHE_BYTES - 1))
349 		return -EINVAL;
350 	spin_lock(&vcpu->arch.vpa_update_lock);
351 	if (v->next_gpa != addr || v->len != len) {
352 		v->next_gpa = addr;
353 		v->len = addr ? len : 0;
354 		v->update_pending = 1;
355 	}
356 	spin_unlock(&vcpu->arch.vpa_update_lock);
357 	return 0;
358 }
359 
360 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
361 struct reg_vpa {
362 	u32 dummy;
363 	union {
364 		__be16 hword;
365 		__be32 word;
366 	} length;
367 };
368 
369 static int vpa_is_registered(struct kvmppc_vpa *vpap)
370 {
371 	if (vpap->update_pending)
372 		return vpap->next_gpa != 0;
373 	return vpap->pinned_addr != NULL;
374 }
375 
376 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
377 				       unsigned long flags,
378 				       unsigned long vcpuid, unsigned long vpa)
379 {
380 	struct kvm *kvm = vcpu->kvm;
381 	unsigned long len, nb;
382 	void *va;
383 	struct kvm_vcpu *tvcpu;
384 	int err;
385 	int subfunc;
386 	struct kvmppc_vpa *vpap;
387 
388 	tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
389 	if (!tvcpu)
390 		return H_PARAMETER;
391 
392 	subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
393 	if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
394 	    subfunc == H_VPA_REG_SLB) {
395 		/* Registering new area - address must be cache-line aligned */
396 		if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
397 			return H_PARAMETER;
398 
399 		/* convert logical addr to kernel addr and read length */
400 		va = kvmppc_pin_guest_page(kvm, vpa, &nb);
401 		if (va == NULL)
402 			return H_PARAMETER;
403 		if (subfunc == H_VPA_REG_VPA)
404 			len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
405 		else
406 			len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
407 		kvmppc_unpin_guest_page(kvm, va, vpa, false);
408 
409 		/* Check length */
410 		if (len > nb || len < sizeof(struct reg_vpa))
411 			return H_PARAMETER;
412 	} else {
413 		vpa = 0;
414 		len = 0;
415 	}
416 
417 	err = H_PARAMETER;
418 	vpap = NULL;
419 	spin_lock(&tvcpu->arch.vpa_update_lock);
420 
421 	switch (subfunc) {
422 	case H_VPA_REG_VPA:		/* register VPA */
423 		if (len < sizeof(struct lppaca))
424 			break;
425 		vpap = &tvcpu->arch.vpa;
426 		err = 0;
427 		break;
428 
429 	case H_VPA_REG_DTL:		/* register DTL */
430 		if (len < sizeof(struct dtl_entry))
431 			break;
432 		len -= len % sizeof(struct dtl_entry);
433 
434 		/* Check that they have previously registered a VPA */
435 		err = H_RESOURCE;
436 		if (!vpa_is_registered(&tvcpu->arch.vpa))
437 			break;
438 
439 		vpap = &tvcpu->arch.dtl;
440 		err = 0;
441 		break;
442 
443 	case H_VPA_REG_SLB:		/* register SLB shadow buffer */
444 		/* Check that they have previously registered a VPA */
445 		err = H_RESOURCE;
446 		if (!vpa_is_registered(&tvcpu->arch.vpa))
447 			break;
448 
449 		vpap = &tvcpu->arch.slb_shadow;
450 		err = 0;
451 		break;
452 
453 	case H_VPA_DEREG_VPA:		/* deregister VPA */
454 		/* Check they don't still have a DTL or SLB buf registered */
455 		err = H_RESOURCE;
456 		if (vpa_is_registered(&tvcpu->arch.dtl) ||
457 		    vpa_is_registered(&tvcpu->arch.slb_shadow))
458 			break;
459 
460 		vpap = &tvcpu->arch.vpa;
461 		err = 0;
462 		break;
463 
464 	case H_VPA_DEREG_DTL:		/* deregister DTL */
465 		vpap = &tvcpu->arch.dtl;
466 		err = 0;
467 		break;
468 
469 	case H_VPA_DEREG_SLB:		/* deregister SLB shadow buffer */
470 		vpap = &tvcpu->arch.slb_shadow;
471 		err = 0;
472 		break;
473 	}
474 
475 	if (vpap) {
476 		vpap->next_gpa = vpa;
477 		vpap->len = len;
478 		vpap->update_pending = 1;
479 	}
480 
481 	spin_unlock(&tvcpu->arch.vpa_update_lock);
482 
483 	return err;
484 }
485 
486 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
487 {
488 	struct kvm *kvm = vcpu->kvm;
489 	void *va;
490 	unsigned long nb;
491 	unsigned long gpa;
492 
493 	/*
494 	 * We need to pin the page pointed to by vpap->next_gpa,
495 	 * but we can't call kvmppc_pin_guest_page under the lock
496 	 * as it does get_user_pages() and down_read().  So we
497 	 * have to drop the lock, pin the page, then get the lock
498 	 * again and check that a new area didn't get registered
499 	 * in the meantime.
500 	 */
501 	for (;;) {
502 		gpa = vpap->next_gpa;
503 		spin_unlock(&vcpu->arch.vpa_update_lock);
504 		va = NULL;
505 		nb = 0;
506 		if (gpa)
507 			va = kvmppc_pin_guest_page(kvm, gpa, &nb);
508 		spin_lock(&vcpu->arch.vpa_update_lock);
509 		if (gpa == vpap->next_gpa)
510 			break;
511 		/* sigh... unpin that one and try again */
512 		if (va)
513 			kvmppc_unpin_guest_page(kvm, va, gpa, false);
514 	}
515 
516 	vpap->update_pending = 0;
517 	if (va && nb < vpap->len) {
518 		/*
519 		 * If it's now too short, it must be that userspace
520 		 * has changed the mappings underlying guest memory,
521 		 * so unregister the region.
522 		 */
523 		kvmppc_unpin_guest_page(kvm, va, gpa, false);
524 		va = NULL;
525 	}
526 	if (vpap->pinned_addr)
527 		kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
528 					vpap->dirty);
529 	vpap->gpa = gpa;
530 	vpap->pinned_addr = va;
531 	vpap->dirty = false;
532 	if (va)
533 		vpap->pinned_end = va + vpap->len;
534 }
535 
536 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
537 {
538 	if (!(vcpu->arch.vpa.update_pending ||
539 	      vcpu->arch.slb_shadow.update_pending ||
540 	      vcpu->arch.dtl.update_pending))
541 		return;
542 
543 	spin_lock(&vcpu->arch.vpa_update_lock);
544 	if (vcpu->arch.vpa.update_pending) {
545 		kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
546 		if (vcpu->arch.vpa.pinned_addr)
547 			init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
548 	}
549 	if (vcpu->arch.dtl.update_pending) {
550 		kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
551 		vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
552 		vcpu->arch.dtl_index = 0;
553 	}
554 	if (vcpu->arch.slb_shadow.update_pending)
555 		kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
556 	spin_unlock(&vcpu->arch.vpa_update_lock);
557 }
558 
559 /*
560  * Return the accumulated stolen time for the vcore up until `now'.
561  * The caller should hold the vcore lock.
562  */
563 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
564 {
565 	u64 p;
566 	unsigned long flags;
567 
568 	spin_lock_irqsave(&vc->stoltb_lock, flags);
569 	p = vc->stolen_tb;
570 	if (vc->vcore_state != VCORE_INACTIVE &&
571 	    vc->preempt_tb != TB_NIL)
572 		p += now - vc->preempt_tb;
573 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
574 	return p;
575 }
576 
577 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
578 				    struct kvmppc_vcore *vc)
579 {
580 	struct dtl_entry *dt;
581 	struct lppaca *vpa;
582 	unsigned long stolen;
583 	unsigned long core_stolen;
584 	u64 now;
585 
586 	dt = vcpu->arch.dtl_ptr;
587 	vpa = vcpu->arch.vpa.pinned_addr;
588 	now = mftb();
589 	core_stolen = vcore_stolen_time(vc, now);
590 	stolen = core_stolen - vcpu->arch.stolen_logged;
591 	vcpu->arch.stolen_logged = core_stolen;
592 	spin_lock_irq(&vcpu->arch.tbacct_lock);
593 	stolen += vcpu->arch.busy_stolen;
594 	vcpu->arch.busy_stolen = 0;
595 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
596 	if (!dt || !vpa)
597 		return;
598 	memset(dt, 0, sizeof(struct dtl_entry));
599 	dt->dispatch_reason = 7;
600 	dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
601 	dt->timebase = cpu_to_be64(now + vc->tb_offset);
602 	dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
603 	dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
604 	dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
605 	++dt;
606 	if (dt == vcpu->arch.dtl.pinned_end)
607 		dt = vcpu->arch.dtl.pinned_addr;
608 	vcpu->arch.dtl_ptr = dt;
609 	/* order writing *dt vs. writing vpa->dtl_idx */
610 	smp_wmb();
611 	vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
612 	vcpu->arch.dtl.dirty = true;
613 }
614 
615 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
616 {
617 	if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
618 		return true;
619 	if ((!vcpu->arch.vcore->arch_compat) &&
620 	    cpu_has_feature(CPU_FTR_ARCH_207S))
621 		return true;
622 	return false;
623 }
624 
625 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
626 			     unsigned long resource, unsigned long value1,
627 			     unsigned long value2)
628 {
629 	switch (resource) {
630 	case H_SET_MODE_RESOURCE_SET_CIABR:
631 		if (!kvmppc_power8_compatible(vcpu))
632 			return H_P2;
633 		if (value2)
634 			return H_P4;
635 		if (mflags)
636 			return H_UNSUPPORTED_FLAG_START;
637 		/* Guests can't breakpoint the hypervisor */
638 		if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
639 			return H_P3;
640 		vcpu->arch.ciabr  = value1;
641 		return H_SUCCESS;
642 	case H_SET_MODE_RESOURCE_SET_DAWR:
643 		if (!kvmppc_power8_compatible(vcpu))
644 			return H_P2;
645 		if (mflags)
646 			return H_UNSUPPORTED_FLAG_START;
647 		if (value2 & DABRX_HYP)
648 			return H_P4;
649 		vcpu->arch.dawr  = value1;
650 		vcpu->arch.dawrx = value2;
651 		return H_SUCCESS;
652 	default:
653 		return H_TOO_HARD;
654 	}
655 }
656 
657 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
658 {
659 	struct kvmppc_vcore *vcore = target->arch.vcore;
660 
661 	/*
662 	 * We expect to have been called by the real mode handler
663 	 * (kvmppc_rm_h_confer()) which would have directly returned
664 	 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
665 	 * have useful work to do and should not confer) so we don't
666 	 * recheck that here.
667 	 */
668 
669 	spin_lock(&vcore->lock);
670 	if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
671 	    vcore->vcore_state != VCORE_INACTIVE &&
672 	    vcore->runner)
673 		target = vcore->runner;
674 	spin_unlock(&vcore->lock);
675 
676 	return kvm_vcpu_yield_to(target);
677 }
678 
679 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
680 {
681 	int yield_count = 0;
682 	struct lppaca *lppaca;
683 
684 	spin_lock(&vcpu->arch.vpa_update_lock);
685 	lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
686 	if (lppaca)
687 		yield_count = be32_to_cpu(lppaca->yield_count);
688 	spin_unlock(&vcpu->arch.vpa_update_lock);
689 	return yield_count;
690 }
691 
692 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
693 {
694 	unsigned long req = kvmppc_get_gpr(vcpu, 3);
695 	unsigned long target, ret = H_SUCCESS;
696 	int yield_count;
697 	struct kvm_vcpu *tvcpu;
698 	int idx, rc;
699 
700 	if (req <= MAX_HCALL_OPCODE &&
701 	    !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
702 		return RESUME_HOST;
703 
704 	switch (req) {
705 	case H_CEDE:
706 		break;
707 	case H_PROD:
708 		target = kvmppc_get_gpr(vcpu, 4);
709 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
710 		if (!tvcpu) {
711 			ret = H_PARAMETER;
712 			break;
713 		}
714 		tvcpu->arch.prodded = 1;
715 		smp_mb();
716 		if (vcpu->arch.ceded) {
717 			if (swait_active(&vcpu->wq)) {
718 				swake_up(&vcpu->wq);
719 				vcpu->stat.halt_wakeup++;
720 			}
721 		}
722 		break;
723 	case H_CONFER:
724 		target = kvmppc_get_gpr(vcpu, 4);
725 		if (target == -1)
726 			break;
727 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
728 		if (!tvcpu) {
729 			ret = H_PARAMETER;
730 			break;
731 		}
732 		yield_count = kvmppc_get_gpr(vcpu, 5);
733 		if (kvmppc_get_yield_count(tvcpu) != yield_count)
734 			break;
735 		kvm_arch_vcpu_yield_to(tvcpu);
736 		break;
737 	case H_REGISTER_VPA:
738 		ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
739 					kvmppc_get_gpr(vcpu, 5),
740 					kvmppc_get_gpr(vcpu, 6));
741 		break;
742 	case H_RTAS:
743 		if (list_empty(&vcpu->kvm->arch.rtas_tokens))
744 			return RESUME_HOST;
745 
746 		idx = srcu_read_lock(&vcpu->kvm->srcu);
747 		rc = kvmppc_rtas_hcall(vcpu);
748 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
749 
750 		if (rc == -ENOENT)
751 			return RESUME_HOST;
752 		else if (rc == 0)
753 			break;
754 
755 		/* Send the error out to userspace via KVM_RUN */
756 		return rc;
757 	case H_LOGICAL_CI_LOAD:
758 		ret = kvmppc_h_logical_ci_load(vcpu);
759 		if (ret == H_TOO_HARD)
760 			return RESUME_HOST;
761 		break;
762 	case H_LOGICAL_CI_STORE:
763 		ret = kvmppc_h_logical_ci_store(vcpu);
764 		if (ret == H_TOO_HARD)
765 			return RESUME_HOST;
766 		break;
767 	case H_SET_MODE:
768 		ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
769 					kvmppc_get_gpr(vcpu, 5),
770 					kvmppc_get_gpr(vcpu, 6),
771 					kvmppc_get_gpr(vcpu, 7));
772 		if (ret == H_TOO_HARD)
773 			return RESUME_HOST;
774 		break;
775 	case H_XIRR:
776 	case H_CPPR:
777 	case H_EOI:
778 	case H_IPI:
779 	case H_IPOLL:
780 	case H_XIRR_X:
781 		if (kvmppc_xics_enabled(vcpu)) {
782 			ret = kvmppc_xics_hcall(vcpu, req);
783 			break;
784 		}
785 		return RESUME_HOST;
786 	case H_PUT_TCE:
787 		ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
788 						kvmppc_get_gpr(vcpu, 5),
789 						kvmppc_get_gpr(vcpu, 6));
790 		if (ret == H_TOO_HARD)
791 			return RESUME_HOST;
792 		break;
793 	case H_PUT_TCE_INDIRECT:
794 		ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
795 						kvmppc_get_gpr(vcpu, 5),
796 						kvmppc_get_gpr(vcpu, 6),
797 						kvmppc_get_gpr(vcpu, 7));
798 		if (ret == H_TOO_HARD)
799 			return RESUME_HOST;
800 		break;
801 	case H_STUFF_TCE:
802 		ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
803 						kvmppc_get_gpr(vcpu, 5),
804 						kvmppc_get_gpr(vcpu, 6),
805 						kvmppc_get_gpr(vcpu, 7));
806 		if (ret == H_TOO_HARD)
807 			return RESUME_HOST;
808 		break;
809 	default:
810 		return RESUME_HOST;
811 	}
812 	kvmppc_set_gpr(vcpu, 3, ret);
813 	vcpu->arch.hcall_needed = 0;
814 	return RESUME_GUEST;
815 }
816 
817 static int kvmppc_hcall_impl_hv(unsigned long cmd)
818 {
819 	switch (cmd) {
820 	case H_CEDE:
821 	case H_PROD:
822 	case H_CONFER:
823 	case H_REGISTER_VPA:
824 	case H_SET_MODE:
825 	case H_LOGICAL_CI_LOAD:
826 	case H_LOGICAL_CI_STORE:
827 #ifdef CONFIG_KVM_XICS
828 	case H_XIRR:
829 	case H_CPPR:
830 	case H_EOI:
831 	case H_IPI:
832 	case H_IPOLL:
833 	case H_XIRR_X:
834 #endif
835 		return 1;
836 	}
837 
838 	/* See if it's in the real-mode table */
839 	return kvmppc_hcall_impl_hv_realmode(cmd);
840 }
841 
842 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
843 					struct kvm_vcpu *vcpu)
844 {
845 	u32 last_inst;
846 
847 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
848 					EMULATE_DONE) {
849 		/*
850 		 * Fetch failed, so return to guest and
851 		 * try executing it again.
852 		 */
853 		return RESUME_GUEST;
854 	}
855 
856 	if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
857 		run->exit_reason = KVM_EXIT_DEBUG;
858 		run->debug.arch.address = kvmppc_get_pc(vcpu);
859 		return RESUME_HOST;
860 	} else {
861 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
862 		return RESUME_GUEST;
863 	}
864 }
865 
866 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
867 				 struct task_struct *tsk)
868 {
869 	int r = RESUME_HOST;
870 
871 	vcpu->stat.sum_exits++;
872 
873 	/*
874 	 * This can happen if an interrupt occurs in the last stages
875 	 * of guest entry or the first stages of guest exit (i.e. after
876 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
877 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
878 	 * That can happen due to a bug, or due to a machine check
879 	 * occurring at just the wrong time.
880 	 */
881 	if (vcpu->arch.shregs.msr & MSR_HV) {
882 		printk(KERN_EMERG "KVM trap in HV mode!\n");
883 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
884 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
885 			vcpu->arch.shregs.msr);
886 		kvmppc_dump_regs(vcpu);
887 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
888 		run->hw.hardware_exit_reason = vcpu->arch.trap;
889 		return RESUME_HOST;
890 	}
891 	run->exit_reason = KVM_EXIT_UNKNOWN;
892 	run->ready_for_interrupt_injection = 1;
893 	switch (vcpu->arch.trap) {
894 	/* We're good on these - the host merely wanted to get our attention */
895 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
896 		vcpu->stat.dec_exits++;
897 		r = RESUME_GUEST;
898 		break;
899 	case BOOK3S_INTERRUPT_EXTERNAL:
900 	case BOOK3S_INTERRUPT_H_DOORBELL:
901 		vcpu->stat.ext_intr_exits++;
902 		r = RESUME_GUEST;
903 		break;
904 	/* HMI is hypervisor interrupt and host has handled it. Resume guest.*/
905 	case BOOK3S_INTERRUPT_HMI:
906 	case BOOK3S_INTERRUPT_PERFMON:
907 		r = RESUME_GUEST;
908 		break;
909 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
910 		/*
911 		 * Deliver a machine check interrupt to the guest.
912 		 * We have to do this, even if the host has handled the
913 		 * machine check, because machine checks use SRR0/1 and
914 		 * the interrupt might have trashed guest state in them.
915 		 */
916 		kvmppc_book3s_queue_irqprio(vcpu,
917 					    BOOK3S_INTERRUPT_MACHINE_CHECK);
918 		r = RESUME_GUEST;
919 		break;
920 	case BOOK3S_INTERRUPT_PROGRAM:
921 	{
922 		ulong flags;
923 		/*
924 		 * Normally program interrupts are delivered directly
925 		 * to the guest by the hardware, but we can get here
926 		 * as a result of a hypervisor emulation interrupt
927 		 * (e40) getting turned into a 700 by BML RTAS.
928 		 */
929 		flags = vcpu->arch.shregs.msr & 0x1f0000ull;
930 		kvmppc_core_queue_program(vcpu, flags);
931 		r = RESUME_GUEST;
932 		break;
933 	}
934 	case BOOK3S_INTERRUPT_SYSCALL:
935 	{
936 		/* hcall - punt to userspace */
937 		int i;
938 
939 		/* hypercall with MSR_PR has already been handled in rmode,
940 		 * and never reaches here.
941 		 */
942 
943 		run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
944 		for (i = 0; i < 9; ++i)
945 			run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
946 		run->exit_reason = KVM_EXIT_PAPR_HCALL;
947 		vcpu->arch.hcall_needed = 1;
948 		r = RESUME_HOST;
949 		break;
950 	}
951 	/*
952 	 * We get these next two if the guest accesses a page which it thinks
953 	 * it has mapped but which is not actually present, either because
954 	 * it is for an emulated I/O device or because the corresonding
955 	 * host page has been paged out.  Any other HDSI/HISI interrupts
956 	 * have been handled already.
957 	 */
958 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
959 		r = RESUME_PAGE_FAULT;
960 		break;
961 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
962 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
963 		vcpu->arch.fault_dsisr = 0;
964 		r = RESUME_PAGE_FAULT;
965 		break;
966 	/*
967 	 * This occurs if the guest executes an illegal instruction.
968 	 * If the guest debug is disabled, generate a program interrupt
969 	 * to the guest. If guest debug is enabled, we need to check
970 	 * whether the instruction is a software breakpoint instruction.
971 	 * Accordingly return to Guest or Host.
972 	 */
973 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
974 		if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
975 			vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
976 				swab32(vcpu->arch.emul_inst) :
977 				vcpu->arch.emul_inst;
978 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
979 			r = kvmppc_emulate_debug_inst(run, vcpu);
980 		} else {
981 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
982 			r = RESUME_GUEST;
983 		}
984 		break;
985 	/*
986 	 * This occurs if the guest (kernel or userspace), does something that
987 	 * is prohibited by HFSCR.  We just generate a program interrupt to
988 	 * the guest.
989 	 */
990 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
991 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
992 		r = RESUME_GUEST;
993 		break;
994 	default:
995 		kvmppc_dump_regs(vcpu);
996 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
997 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
998 			vcpu->arch.shregs.msr);
999 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1000 		r = RESUME_HOST;
1001 		break;
1002 	}
1003 
1004 	return r;
1005 }
1006 
1007 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1008 					    struct kvm_sregs *sregs)
1009 {
1010 	int i;
1011 
1012 	memset(sregs, 0, sizeof(struct kvm_sregs));
1013 	sregs->pvr = vcpu->arch.pvr;
1014 	for (i = 0; i < vcpu->arch.slb_max; i++) {
1015 		sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1016 		sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1017 	}
1018 
1019 	return 0;
1020 }
1021 
1022 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1023 					    struct kvm_sregs *sregs)
1024 {
1025 	int i, j;
1026 
1027 	/* Only accept the same PVR as the host's, since we can't spoof it */
1028 	if (sregs->pvr != vcpu->arch.pvr)
1029 		return -EINVAL;
1030 
1031 	j = 0;
1032 	for (i = 0; i < vcpu->arch.slb_nr; i++) {
1033 		if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1034 			vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1035 			vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1036 			++j;
1037 		}
1038 	}
1039 	vcpu->arch.slb_max = j;
1040 
1041 	return 0;
1042 }
1043 
1044 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1045 		bool preserve_top32)
1046 {
1047 	struct kvm *kvm = vcpu->kvm;
1048 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
1049 	u64 mask;
1050 
1051 	mutex_lock(&kvm->lock);
1052 	spin_lock(&vc->lock);
1053 	/*
1054 	 * If ILE (interrupt little-endian) has changed, update the
1055 	 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1056 	 */
1057 	if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1058 		struct kvm_vcpu *vcpu;
1059 		int i;
1060 
1061 		kvm_for_each_vcpu(i, vcpu, kvm) {
1062 			if (vcpu->arch.vcore != vc)
1063 				continue;
1064 			if (new_lpcr & LPCR_ILE)
1065 				vcpu->arch.intr_msr |= MSR_LE;
1066 			else
1067 				vcpu->arch.intr_msr &= ~MSR_LE;
1068 		}
1069 	}
1070 
1071 	/*
1072 	 * Userspace can only modify DPFD (default prefetch depth),
1073 	 * ILE (interrupt little-endian) and TC (translation control).
1074 	 * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
1075 	 */
1076 	mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1077 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
1078 		mask |= LPCR_AIL;
1079 
1080 	/* Broken 32-bit version of LPCR must not clear top bits */
1081 	if (preserve_top32)
1082 		mask &= 0xFFFFFFFF;
1083 	vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1084 	spin_unlock(&vc->lock);
1085 	mutex_unlock(&kvm->lock);
1086 }
1087 
1088 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1089 				 union kvmppc_one_reg *val)
1090 {
1091 	int r = 0;
1092 	long int i;
1093 
1094 	switch (id) {
1095 	case KVM_REG_PPC_DEBUG_INST:
1096 		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1097 		break;
1098 	case KVM_REG_PPC_HIOR:
1099 		*val = get_reg_val(id, 0);
1100 		break;
1101 	case KVM_REG_PPC_DABR:
1102 		*val = get_reg_val(id, vcpu->arch.dabr);
1103 		break;
1104 	case KVM_REG_PPC_DABRX:
1105 		*val = get_reg_val(id, vcpu->arch.dabrx);
1106 		break;
1107 	case KVM_REG_PPC_DSCR:
1108 		*val = get_reg_val(id, vcpu->arch.dscr);
1109 		break;
1110 	case KVM_REG_PPC_PURR:
1111 		*val = get_reg_val(id, vcpu->arch.purr);
1112 		break;
1113 	case KVM_REG_PPC_SPURR:
1114 		*val = get_reg_val(id, vcpu->arch.spurr);
1115 		break;
1116 	case KVM_REG_PPC_AMR:
1117 		*val = get_reg_val(id, vcpu->arch.amr);
1118 		break;
1119 	case KVM_REG_PPC_UAMOR:
1120 		*val = get_reg_val(id, vcpu->arch.uamor);
1121 		break;
1122 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1123 		i = id - KVM_REG_PPC_MMCR0;
1124 		*val = get_reg_val(id, vcpu->arch.mmcr[i]);
1125 		break;
1126 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1127 		i = id - KVM_REG_PPC_PMC1;
1128 		*val = get_reg_val(id, vcpu->arch.pmc[i]);
1129 		break;
1130 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1131 		i = id - KVM_REG_PPC_SPMC1;
1132 		*val = get_reg_val(id, vcpu->arch.spmc[i]);
1133 		break;
1134 	case KVM_REG_PPC_SIAR:
1135 		*val = get_reg_val(id, vcpu->arch.siar);
1136 		break;
1137 	case KVM_REG_PPC_SDAR:
1138 		*val = get_reg_val(id, vcpu->arch.sdar);
1139 		break;
1140 	case KVM_REG_PPC_SIER:
1141 		*val = get_reg_val(id, vcpu->arch.sier);
1142 		break;
1143 	case KVM_REG_PPC_IAMR:
1144 		*val = get_reg_val(id, vcpu->arch.iamr);
1145 		break;
1146 	case KVM_REG_PPC_PSPB:
1147 		*val = get_reg_val(id, vcpu->arch.pspb);
1148 		break;
1149 	case KVM_REG_PPC_DPDES:
1150 		*val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1151 		break;
1152 	case KVM_REG_PPC_DAWR:
1153 		*val = get_reg_val(id, vcpu->arch.dawr);
1154 		break;
1155 	case KVM_REG_PPC_DAWRX:
1156 		*val = get_reg_val(id, vcpu->arch.dawrx);
1157 		break;
1158 	case KVM_REG_PPC_CIABR:
1159 		*val = get_reg_val(id, vcpu->arch.ciabr);
1160 		break;
1161 	case KVM_REG_PPC_CSIGR:
1162 		*val = get_reg_val(id, vcpu->arch.csigr);
1163 		break;
1164 	case KVM_REG_PPC_TACR:
1165 		*val = get_reg_val(id, vcpu->arch.tacr);
1166 		break;
1167 	case KVM_REG_PPC_TCSCR:
1168 		*val = get_reg_val(id, vcpu->arch.tcscr);
1169 		break;
1170 	case KVM_REG_PPC_PID:
1171 		*val = get_reg_val(id, vcpu->arch.pid);
1172 		break;
1173 	case KVM_REG_PPC_ACOP:
1174 		*val = get_reg_val(id, vcpu->arch.acop);
1175 		break;
1176 	case KVM_REG_PPC_WORT:
1177 		*val = get_reg_val(id, vcpu->arch.wort);
1178 		break;
1179 	case KVM_REG_PPC_VPA_ADDR:
1180 		spin_lock(&vcpu->arch.vpa_update_lock);
1181 		*val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1182 		spin_unlock(&vcpu->arch.vpa_update_lock);
1183 		break;
1184 	case KVM_REG_PPC_VPA_SLB:
1185 		spin_lock(&vcpu->arch.vpa_update_lock);
1186 		val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1187 		val->vpaval.length = vcpu->arch.slb_shadow.len;
1188 		spin_unlock(&vcpu->arch.vpa_update_lock);
1189 		break;
1190 	case KVM_REG_PPC_VPA_DTL:
1191 		spin_lock(&vcpu->arch.vpa_update_lock);
1192 		val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1193 		val->vpaval.length = vcpu->arch.dtl.len;
1194 		spin_unlock(&vcpu->arch.vpa_update_lock);
1195 		break;
1196 	case KVM_REG_PPC_TB_OFFSET:
1197 		*val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1198 		break;
1199 	case KVM_REG_PPC_LPCR:
1200 	case KVM_REG_PPC_LPCR_64:
1201 		*val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1202 		break;
1203 	case KVM_REG_PPC_PPR:
1204 		*val = get_reg_val(id, vcpu->arch.ppr);
1205 		break;
1206 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1207 	case KVM_REG_PPC_TFHAR:
1208 		*val = get_reg_val(id, vcpu->arch.tfhar);
1209 		break;
1210 	case KVM_REG_PPC_TFIAR:
1211 		*val = get_reg_val(id, vcpu->arch.tfiar);
1212 		break;
1213 	case KVM_REG_PPC_TEXASR:
1214 		*val = get_reg_val(id, vcpu->arch.texasr);
1215 		break;
1216 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1217 		i = id - KVM_REG_PPC_TM_GPR0;
1218 		*val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1219 		break;
1220 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1221 	{
1222 		int j;
1223 		i = id - KVM_REG_PPC_TM_VSR0;
1224 		if (i < 32)
1225 			for (j = 0; j < TS_FPRWIDTH; j++)
1226 				val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1227 		else {
1228 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1229 				val->vval = vcpu->arch.vr_tm.vr[i-32];
1230 			else
1231 				r = -ENXIO;
1232 		}
1233 		break;
1234 	}
1235 	case KVM_REG_PPC_TM_CR:
1236 		*val = get_reg_val(id, vcpu->arch.cr_tm);
1237 		break;
1238 	case KVM_REG_PPC_TM_LR:
1239 		*val = get_reg_val(id, vcpu->arch.lr_tm);
1240 		break;
1241 	case KVM_REG_PPC_TM_CTR:
1242 		*val = get_reg_val(id, vcpu->arch.ctr_tm);
1243 		break;
1244 	case KVM_REG_PPC_TM_FPSCR:
1245 		*val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1246 		break;
1247 	case KVM_REG_PPC_TM_AMR:
1248 		*val = get_reg_val(id, vcpu->arch.amr_tm);
1249 		break;
1250 	case KVM_REG_PPC_TM_PPR:
1251 		*val = get_reg_val(id, vcpu->arch.ppr_tm);
1252 		break;
1253 	case KVM_REG_PPC_TM_VRSAVE:
1254 		*val = get_reg_val(id, vcpu->arch.vrsave_tm);
1255 		break;
1256 	case KVM_REG_PPC_TM_VSCR:
1257 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1258 			*val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1259 		else
1260 			r = -ENXIO;
1261 		break;
1262 	case KVM_REG_PPC_TM_DSCR:
1263 		*val = get_reg_val(id, vcpu->arch.dscr_tm);
1264 		break;
1265 	case KVM_REG_PPC_TM_TAR:
1266 		*val = get_reg_val(id, vcpu->arch.tar_tm);
1267 		break;
1268 #endif
1269 	case KVM_REG_PPC_ARCH_COMPAT:
1270 		*val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1271 		break;
1272 	default:
1273 		r = -EINVAL;
1274 		break;
1275 	}
1276 
1277 	return r;
1278 }
1279 
1280 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1281 				 union kvmppc_one_reg *val)
1282 {
1283 	int r = 0;
1284 	long int i;
1285 	unsigned long addr, len;
1286 
1287 	switch (id) {
1288 	case KVM_REG_PPC_HIOR:
1289 		/* Only allow this to be set to zero */
1290 		if (set_reg_val(id, *val))
1291 			r = -EINVAL;
1292 		break;
1293 	case KVM_REG_PPC_DABR:
1294 		vcpu->arch.dabr = set_reg_val(id, *val);
1295 		break;
1296 	case KVM_REG_PPC_DABRX:
1297 		vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1298 		break;
1299 	case KVM_REG_PPC_DSCR:
1300 		vcpu->arch.dscr = set_reg_val(id, *val);
1301 		break;
1302 	case KVM_REG_PPC_PURR:
1303 		vcpu->arch.purr = set_reg_val(id, *val);
1304 		break;
1305 	case KVM_REG_PPC_SPURR:
1306 		vcpu->arch.spurr = set_reg_val(id, *val);
1307 		break;
1308 	case KVM_REG_PPC_AMR:
1309 		vcpu->arch.amr = set_reg_val(id, *val);
1310 		break;
1311 	case KVM_REG_PPC_UAMOR:
1312 		vcpu->arch.uamor = set_reg_val(id, *val);
1313 		break;
1314 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1315 		i = id - KVM_REG_PPC_MMCR0;
1316 		vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1317 		break;
1318 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1319 		i = id - KVM_REG_PPC_PMC1;
1320 		vcpu->arch.pmc[i] = set_reg_val(id, *val);
1321 		break;
1322 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1323 		i = id - KVM_REG_PPC_SPMC1;
1324 		vcpu->arch.spmc[i] = set_reg_val(id, *val);
1325 		break;
1326 	case KVM_REG_PPC_SIAR:
1327 		vcpu->arch.siar = set_reg_val(id, *val);
1328 		break;
1329 	case KVM_REG_PPC_SDAR:
1330 		vcpu->arch.sdar = set_reg_val(id, *val);
1331 		break;
1332 	case KVM_REG_PPC_SIER:
1333 		vcpu->arch.sier = set_reg_val(id, *val);
1334 		break;
1335 	case KVM_REG_PPC_IAMR:
1336 		vcpu->arch.iamr = set_reg_val(id, *val);
1337 		break;
1338 	case KVM_REG_PPC_PSPB:
1339 		vcpu->arch.pspb = set_reg_val(id, *val);
1340 		break;
1341 	case KVM_REG_PPC_DPDES:
1342 		vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1343 		break;
1344 	case KVM_REG_PPC_DAWR:
1345 		vcpu->arch.dawr = set_reg_val(id, *val);
1346 		break;
1347 	case KVM_REG_PPC_DAWRX:
1348 		vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1349 		break;
1350 	case KVM_REG_PPC_CIABR:
1351 		vcpu->arch.ciabr = set_reg_val(id, *val);
1352 		/* Don't allow setting breakpoints in hypervisor code */
1353 		if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1354 			vcpu->arch.ciabr &= ~CIABR_PRIV;	/* disable */
1355 		break;
1356 	case KVM_REG_PPC_CSIGR:
1357 		vcpu->arch.csigr = set_reg_val(id, *val);
1358 		break;
1359 	case KVM_REG_PPC_TACR:
1360 		vcpu->arch.tacr = set_reg_val(id, *val);
1361 		break;
1362 	case KVM_REG_PPC_TCSCR:
1363 		vcpu->arch.tcscr = set_reg_val(id, *val);
1364 		break;
1365 	case KVM_REG_PPC_PID:
1366 		vcpu->arch.pid = set_reg_val(id, *val);
1367 		break;
1368 	case KVM_REG_PPC_ACOP:
1369 		vcpu->arch.acop = set_reg_val(id, *val);
1370 		break;
1371 	case KVM_REG_PPC_WORT:
1372 		vcpu->arch.wort = set_reg_val(id, *val);
1373 		break;
1374 	case KVM_REG_PPC_VPA_ADDR:
1375 		addr = set_reg_val(id, *val);
1376 		r = -EINVAL;
1377 		if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1378 			      vcpu->arch.dtl.next_gpa))
1379 			break;
1380 		r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1381 		break;
1382 	case KVM_REG_PPC_VPA_SLB:
1383 		addr = val->vpaval.addr;
1384 		len = val->vpaval.length;
1385 		r = -EINVAL;
1386 		if (addr && !vcpu->arch.vpa.next_gpa)
1387 			break;
1388 		r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1389 		break;
1390 	case KVM_REG_PPC_VPA_DTL:
1391 		addr = val->vpaval.addr;
1392 		len = val->vpaval.length;
1393 		r = -EINVAL;
1394 		if (addr && (len < sizeof(struct dtl_entry) ||
1395 			     !vcpu->arch.vpa.next_gpa))
1396 			break;
1397 		len -= len % sizeof(struct dtl_entry);
1398 		r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1399 		break;
1400 	case KVM_REG_PPC_TB_OFFSET:
1401 		/* round up to multiple of 2^24 */
1402 		vcpu->arch.vcore->tb_offset =
1403 			ALIGN(set_reg_val(id, *val), 1UL << 24);
1404 		break;
1405 	case KVM_REG_PPC_LPCR:
1406 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1407 		break;
1408 	case KVM_REG_PPC_LPCR_64:
1409 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1410 		break;
1411 	case KVM_REG_PPC_PPR:
1412 		vcpu->arch.ppr = set_reg_val(id, *val);
1413 		break;
1414 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1415 	case KVM_REG_PPC_TFHAR:
1416 		vcpu->arch.tfhar = set_reg_val(id, *val);
1417 		break;
1418 	case KVM_REG_PPC_TFIAR:
1419 		vcpu->arch.tfiar = set_reg_val(id, *val);
1420 		break;
1421 	case KVM_REG_PPC_TEXASR:
1422 		vcpu->arch.texasr = set_reg_val(id, *val);
1423 		break;
1424 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1425 		i = id - KVM_REG_PPC_TM_GPR0;
1426 		vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1427 		break;
1428 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1429 	{
1430 		int j;
1431 		i = id - KVM_REG_PPC_TM_VSR0;
1432 		if (i < 32)
1433 			for (j = 0; j < TS_FPRWIDTH; j++)
1434 				vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1435 		else
1436 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1437 				vcpu->arch.vr_tm.vr[i-32] = val->vval;
1438 			else
1439 				r = -ENXIO;
1440 		break;
1441 	}
1442 	case KVM_REG_PPC_TM_CR:
1443 		vcpu->arch.cr_tm = set_reg_val(id, *val);
1444 		break;
1445 	case KVM_REG_PPC_TM_LR:
1446 		vcpu->arch.lr_tm = set_reg_val(id, *val);
1447 		break;
1448 	case KVM_REG_PPC_TM_CTR:
1449 		vcpu->arch.ctr_tm = set_reg_val(id, *val);
1450 		break;
1451 	case KVM_REG_PPC_TM_FPSCR:
1452 		vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1453 		break;
1454 	case KVM_REG_PPC_TM_AMR:
1455 		vcpu->arch.amr_tm = set_reg_val(id, *val);
1456 		break;
1457 	case KVM_REG_PPC_TM_PPR:
1458 		vcpu->arch.ppr_tm = set_reg_val(id, *val);
1459 		break;
1460 	case KVM_REG_PPC_TM_VRSAVE:
1461 		vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1462 		break;
1463 	case KVM_REG_PPC_TM_VSCR:
1464 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1465 			vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1466 		else
1467 			r = - ENXIO;
1468 		break;
1469 	case KVM_REG_PPC_TM_DSCR:
1470 		vcpu->arch.dscr_tm = set_reg_val(id, *val);
1471 		break;
1472 	case KVM_REG_PPC_TM_TAR:
1473 		vcpu->arch.tar_tm = set_reg_val(id, *val);
1474 		break;
1475 #endif
1476 	case KVM_REG_PPC_ARCH_COMPAT:
1477 		r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1478 		break;
1479 	default:
1480 		r = -EINVAL;
1481 		break;
1482 	}
1483 
1484 	return r;
1485 }
1486 
1487 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1488 {
1489 	struct kvmppc_vcore *vcore;
1490 
1491 	vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1492 
1493 	if (vcore == NULL)
1494 		return NULL;
1495 
1496 	INIT_LIST_HEAD(&vcore->runnable_threads);
1497 	spin_lock_init(&vcore->lock);
1498 	spin_lock_init(&vcore->stoltb_lock);
1499 	init_swait_queue_head(&vcore->wq);
1500 	vcore->preempt_tb = TB_NIL;
1501 	vcore->lpcr = kvm->arch.lpcr;
1502 	vcore->first_vcpuid = core * threads_per_subcore;
1503 	vcore->kvm = kvm;
1504 	INIT_LIST_HEAD(&vcore->preempt_list);
1505 
1506 	return vcore;
1507 }
1508 
1509 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1510 static struct debugfs_timings_element {
1511 	const char *name;
1512 	size_t offset;
1513 } timings[] = {
1514 	{"rm_entry",	offsetof(struct kvm_vcpu, arch.rm_entry)},
1515 	{"rm_intr",	offsetof(struct kvm_vcpu, arch.rm_intr)},
1516 	{"rm_exit",	offsetof(struct kvm_vcpu, arch.rm_exit)},
1517 	{"guest",	offsetof(struct kvm_vcpu, arch.guest_time)},
1518 	{"cede",	offsetof(struct kvm_vcpu, arch.cede_time)},
1519 };
1520 
1521 #define N_TIMINGS	(sizeof(timings) / sizeof(timings[0]))
1522 
1523 struct debugfs_timings_state {
1524 	struct kvm_vcpu	*vcpu;
1525 	unsigned int	buflen;
1526 	char		buf[N_TIMINGS * 100];
1527 };
1528 
1529 static int debugfs_timings_open(struct inode *inode, struct file *file)
1530 {
1531 	struct kvm_vcpu *vcpu = inode->i_private;
1532 	struct debugfs_timings_state *p;
1533 
1534 	p = kzalloc(sizeof(*p), GFP_KERNEL);
1535 	if (!p)
1536 		return -ENOMEM;
1537 
1538 	kvm_get_kvm(vcpu->kvm);
1539 	p->vcpu = vcpu;
1540 	file->private_data = p;
1541 
1542 	return nonseekable_open(inode, file);
1543 }
1544 
1545 static int debugfs_timings_release(struct inode *inode, struct file *file)
1546 {
1547 	struct debugfs_timings_state *p = file->private_data;
1548 
1549 	kvm_put_kvm(p->vcpu->kvm);
1550 	kfree(p);
1551 	return 0;
1552 }
1553 
1554 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1555 				    size_t len, loff_t *ppos)
1556 {
1557 	struct debugfs_timings_state *p = file->private_data;
1558 	struct kvm_vcpu *vcpu = p->vcpu;
1559 	char *s, *buf_end;
1560 	struct kvmhv_tb_accumulator tb;
1561 	u64 count;
1562 	loff_t pos;
1563 	ssize_t n;
1564 	int i, loops;
1565 	bool ok;
1566 
1567 	if (!p->buflen) {
1568 		s = p->buf;
1569 		buf_end = s + sizeof(p->buf);
1570 		for (i = 0; i < N_TIMINGS; ++i) {
1571 			struct kvmhv_tb_accumulator *acc;
1572 
1573 			acc = (struct kvmhv_tb_accumulator *)
1574 				((unsigned long)vcpu + timings[i].offset);
1575 			ok = false;
1576 			for (loops = 0; loops < 1000; ++loops) {
1577 				count = acc->seqcount;
1578 				if (!(count & 1)) {
1579 					smp_rmb();
1580 					tb = *acc;
1581 					smp_rmb();
1582 					if (count == acc->seqcount) {
1583 						ok = true;
1584 						break;
1585 					}
1586 				}
1587 				udelay(1);
1588 			}
1589 			if (!ok)
1590 				snprintf(s, buf_end - s, "%s: stuck\n",
1591 					timings[i].name);
1592 			else
1593 				snprintf(s, buf_end - s,
1594 					"%s: %llu %llu %llu %llu\n",
1595 					timings[i].name, count / 2,
1596 					tb_to_ns(tb.tb_total),
1597 					tb_to_ns(tb.tb_min),
1598 					tb_to_ns(tb.tb_max));
1599 			s += strlen(s);
1600 		}
1601 		p->buflen = s - p->buf;
1602 	}
1603 
1604 	pos = *ppos;
1605 	if (pos >= p->buflen)
1606 		return 0;
1607 	if (len > p->buflen - pos)
1608 		len = p->buflen - pos;
1609 	n = copy_to_user(buf, p->buf + pos, len);
1610 	if (n) {
1611 		if (n == len)
1612 			return -EFAULT;
1613 		len -= n;
1614 	}
1615 	*ppos = pos + len;
1616 	return len;
1617 }
1618 
1619 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1620 				     size_t len, loff_t *ppos)
1621 {
1622 	return -EACCES;
1623 }
1624 
1625 static const struct file_operations debugfs_timings_ops = {
1626 	.owner	 = THIS_MODULE,
1627 	.open	 = debugfs_timings_open,
1628 	.release = debugfs_timings_release,
1629 	.read	 = debugfs_timings_read,
1630 	.write	 = debugfs_timings_write,
1631 	.llseek	 = generic_file_llseek,
1632 };
1633 
1634 /* Create a debugfs directory for the vcpu */
1635 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1636 {
1637 	char buf[16];
1638 	struct kvm *kvm = vcpu->kvm;
1639 
1640 	snprintf(buf, sizeof(buf), "vcpu%u", id);
1641 	if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1642 		return;
1643 	vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1644 	if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1645 		return;
1646 	vcpu->arch.debugfs_timings =
1647 		debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1648 				    vcpu, &debugfs_timings_ops);
1649 }
1650 
1651 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1652 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1653 {
1654 }
1655 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1656 
1657 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1658 						   unsigned int id)
1659 {
1660 	struct kvm_vcpu *vcpu;
1661 	int err = -EINVAL;
1662 	int core;
1663 	struct kvmppc_vcore *vcore;
1664 
1665 	core = id / threads_per_subcore;
1666 	if (core >= KVM_MAX_VCORES)
1667 		goto out;
1668 
1669 	err = -ENOMEM;
1670 	vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1671 	if (!vcpu)
1672 		goto out;
1673 
1674 	err = kvm_vcpu_init(vcpu, kvm, id);
1675 	if (err)
1676 		goto free_vcpu;
1677 
1678 	vcpu->arch.shared = &vcpu->arch.shregs;
1679 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1680 	/*
1681 	 * The shared struct is never shared on HV,
1682 	 * so we can always use host endianness
1683 	 */
1684 #ifdef __BIG_ENDIAN__
1685 	vcpu->arch.shared_big_endian = true;
1686 #else
1687 	vcpu->arch.shared_big_endian = false;
1688 #endif
1689 #endif
1690 	vcpu->arch.mmcr[0] = MMCR0_FC;
1691 	vcpu->arch.ctrl = CTRL_RUNLATCH;
1692 	/* default to host PVR, since we can't spoof it */
1693 	kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1694 	spin_lock_init(&vcpu->arch.vpa_update_lock);
1695 	spin_lock_init(&vcpu->arch.tbacct_lock);
1696 	vcpu->arch.busy_preempt = TB_NIL;
1697 	vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1698 
1699 	kvmppc_mmu_book3s_hv_init(vcpu);
1700 
1701 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1702 
1703 	init_waitqueue_head(&vcpu->arch.cpu_run);
1704 
1705 	mutex_lock(&kvm->lock);
1706 	vcore = kvm->arch.vcores[core];
1707 	if (!vcore) {
1708 		vcore = kvmppc_vcore_create(kvm, core);
1709 		kvm->arch.vcores[core] = vcore;
1710 		kvm->arch.online_vcores++;
1711 	}
1712 	mutex_unlock(&kvm->lock);
1713 
1714 	if (!vcore)
1715 		goto free_vcpu;
1716 
1717 	spin_lock(&vcore->lock);
1718 	++vcore->num_threads;
1719 	spin_unlock(&vcore->lock);
1720 	vcpu->arch.vcore = vcore;
1721 	vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1722 	vcpu->arch.thread_cpu = -1;
1723 
1724 	vcpu->arch.cpu_type = KVM_CPU_3S_64;
1725 	kvmppc_sanity_check(vcpu);
1726 
1727 	debugfs_vcpu_init(vcpu, id);
1728 
1729 	return vcpu;
1730 
1731 free_vcpu:
1732 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1733 out:
1734 	return ERR_PTR(err);
1735 }
1736 
1737 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1738 {
1739 	if (vpa->pinned_addr)
1740 		kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1741 					vpa->dirty);
1742 }
1743 
1744 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
1745 {
1746 	spin_lock(&vcpu->arch.vpa_update_lock);
1747 	unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1748 	unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1749 	unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1750 	spin_unlock(&vcpu->arch.vpa_update_lock);
1751 	kvm_vcpu_uninit(vcpu);
1752 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1753 }
1754 
1755 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1756 {
1757 	/* Indicate we want to get back into the guest */
1758 	return 1;
1759 }
1760 
1761 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1762 {
1763 	unsigned long dec_nsec, now;
1764 
1765 	now = get_tb();
1766 	if (now > vcpu->arch.dec_expires) {
1767 		/* decrementer has already gone negative */
1768 		kvmppc_core_queue_dec(vcpu);
1769 		kvmppc_core_prepare_to_enter(vcpu);
1770 		return;
1771 	}
1772 	dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1773 		   / tb_ticks_per_sec;
1774 	hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1775 		      HRTIMER_MODE_REL);
1776 	vcpu->arch.timer_running = 1;
1777 }
1778 
1779 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1780 {
1781 	vcpu->arch.ceded = 0;
1782 	if (vcpu->arch.timer_running) {
1783 		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1784 		vcpu->arch.timer_running = 0;
1785 	}
1786 }
1787 
1788 extern void __kvmppc_vcore_entry(void);
1789 
1790 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1791 				   struct kvm_vcpu *vcpu)
1792 {
1793 	u64 now;
1794 
1795 	if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1796 		return;
1797 	spin_lock_irq(&vcpu->arch.tbacct_lock);
1798 	now = mftb();
1799 	vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1800 		vcpu->arch.stolen_logged;
1801 	vcpu->arch.busy_preempt = now;
1802 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1803 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
1804 	--vc->n_runnable;
1805 	list_del(&vcpu->arch.run_list);
1806 }
1807 
1808 static int kvmppc_grab_hwthread(int cpu)
1809 {
1810 	struct paca_struct *tpaca;
1811 	long timeout = 10000;
1812 
1813 	tpaca = &paca[cpu];
1814 
1815 	/* Ensure the thread won't go into the kernel if it wakes */
1816 	tpaca->kvm_hstate.kvm_vcpu = NULL;
1817 	tpaca->kvm_hstate.kvm_vcore = NULL;
1818 	tpaca->kvm_hstate.napping = 0;
1819 	smp_wmb();
1820 	tpaca->kvm_hstate.hwthread_req = 1;
1821 
1822 	/*
1823 	 * If the thread is already executing in the kernel (e.g. handling
1824 	 * a stray interrupt), wait for it to get back to nap mode.
1825 	 * The smp_mb() is to ensure that our setting of hwthread_req
1826 	 * is visible before we look at hwthread_state, so if this
1827 	 * races with the code at system_reset_pSeries and the thread
1828 	 * misses our setting of hwthread_req, we are sure to see its
1829 	 * setting of hwthread_state, and vice versa.
1830 	 */
1831 	smp_mb();
1832 	while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1833 		if (--timeout <= 0) {
1834 			pr_err("KVM: couldn't grab cpu %d\n", cpu);
1835 			return -EBUSY;
1836 		}
1837 		udelay(1);
1838 	}
1839 	return 0;
1840 }
1841 
1842 static void kvmppc_release_hwthread(int cpu)
1843 {
1844 	struct paca_struct *tpaca;
1845 
1846 	tpaca = &paca[cpu];
1847 	tpaca->kvm_hstate.hwthread_req = 0;
1848 	tpaca->kvm_hstate.kvm_vcpu = NULL;
1849 	tpaca->kvm_hstate.kvm_vcore = NULL;
1850 	tpaca->kvm_hstate.kvm_split_mode = NULL;
1851 }
1852 
1853 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
1854 {
1855 	int cpu;
1856 	struct paca_struct *tpaca;
1857 	struct kvmppc_vcore *mvc = vc->master_vcore;
1858 
1859 	cpu = vc->pcpu;
1860 	if (vcpu) {
1861 		if (vcpu->arch.timer_running) {
1862 			hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1863 			vcpu->arch.timer_running = 0;
1864 		}
1865 		cpu += vcpu->arch.ptid;
1866 		vcpu->cpu = mvc->pcpu;
1867 		vcpu->arch.thread_cpu = cpu;
1868 	}
1869 	tpaca = &paca[cpu];
1870 	tpaca->kvm_hstate.kvm_vcpu = vcpu;
1871 	tpaca->kvm_hstate.ptid = cpu - mvc->pcpu;
1872 	/* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
1873 	smp_wmb();
1874 	tpaca->kvm_hstate.kvm_vcore = mvc;
1875 	if (cpu != smp_processor_id())
1876 		kvmppc_ipi_thread(cpu);
1877 }
1878 
1879 static void kvmppc_wait_for_nap(void)
1880 {
1881 	int cpu = smp_processor_id();
1882 	int i, loops;
1883 
1884 	for (loops = 0; loops < 1000000; ++loops) {
1885 		/*
1886 		 * Check if all threads are finished.
1887 		 * We set the vcore pointer when starting a thread
1888 		 * and the thread clears it when finished, so we look
1889 		 * for any threads that still have a non-NULL vcore ptr.
1890 		 */
1891 		for (i = 1; i < threads_per_subcore; ++i)
1892 			if (paca[cpu + i].kvm_hstate.kvm_vcore)
1893 				break;
1894 		if (i == threads_per_subcore) {
1895 			HMT_medium();
1896 			return;
1897 		}
1898 		HMT_low();
1899 	}
1900 	HMT_medium();
1901 	for (i = 1; i < threads_per_subcore; ++i)
1902 		if (paca[cpu + i].kvm_hstate.kvm_vcore)
1903 			pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
1904 }
1905 
1906 /*
1907  * Check that we are on thread 0 and that any other threads in
1908  * this core are off-line.  Then grab the threads so they can't
1909  * enter the kernel.
1910  */
1911 static int on_primary_thread(void)
1912 {
1913 	int cpu = smp_processor_id();
1914 	int thr;
1915 
1916 	/* Are we on a primary subcore? */
1917 	if (cpu_thread_in_subcore(cpu))
1918 		return 0;
1919 
1920 	thr = 0;
1921 	while (++thr < threads_per_subcore)
1922 		if (cpu_online(cpu + thr))
1923 			return 0;
1924 
1925 	/* Grab all hw threads so they can't go into the kernel */
1926 	for (thr = 1; thr < threads_per_subcore; ++thr) {
1927 		if (kvmppc_grab_hwthread(cpu + thr)) {
1928 			/* Couldn't grab one; let the others go */
1929 			do {
1930 				kvmppc_release_hwthread(cpu + thr);
1931 			} while (--thr > 0);
1932 			return 0;
1933 		}
1934 	}
1935 	return 1;
1936 }
1937 
1938 /*
1939  * A list of virtual cores for each physical CPU.
1940  * These are vcores that could run but their runner VCPU tasks are
1941  * (or may be) preempted.
1942  */
1943 struct preempted_vcore_list {
1944 	struct list_head	list;
1945 	spinlock_t		lock;
1946 };
1947 
1948 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
1949 
1950 static void init_vcore_lists(void)
1951 {
1952 	int cpu;
1953 
1954 	for_each_possible_cpu(cpu) {
1955 		struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
1956 		spin_lock_init(&lp->lock);
1957 		INIT_LIST_HEAD(&lp->list);
1958 	}
1959 }
1960 
1961 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
1962 {
1963 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
1964 
1965 	vc->vcore_state = VCORE_PREEMPT;
1966 	vc->pcpu = smp_processor_id();
1967 	if (vc->num_threads < threads_per_subcore) {
1968 		spin_lock(&lp->lock);
1969 		list_add_tail(&vc->preempt_list, &lp->list);
1970 		spin_unlock(&lp->lock);
1971 	}
1972 
1973 	/* Start accumulating stolen time */
1974 	kvmppc_core_start_stolen(vc);
1975 }
1976 
1977 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
1978 {
1979 	struct preempted_vcore_list *lp;
1980 
1981 	kvmppc_core_end_stolen(vc);
1982 	if (!list_empty(&vc->preempt_list)) {
1983 		lp = &per_cpu(preempted_vcores, vc->pcpu);
1984 		spin_lock(&lp->lock);
1985 		list_del_init(&vc->preempt_list);
1986 		spin_unlock(&lp->lock);
1987 	}
1988 	vc->vcore_state = VCORE_INACTIVE;
1989 }
1990 
1991 /*
1992  * This stores information about the virtual cores currently
1993  * assigned to a physical core.
1994  */
1995 struct core_info {
1996 	int		n_subcores;
1997 	int		max_subcore_threads;
1998 	int		total_threads;
1999 	int		subcore_threads[MAX_SUBCORES];
2000 	struct kvm	*subcore_vm[MAX_SUBCORES];
2001 	struct list_head vcs[MAX_SUBCORES];
2002 };
2003 
2004 /*
2005  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2006  * respectively in 2-way micro-threading (split-core) mode.
2007  */
2008 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2009 
2010 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2011 {
2012 	int sub;
2013 
2014 	memset(cip, 0, sizeof(*cip));
2015 	cip->n_subcores = 1;
2016 	cip->max_subcore_threads = vc->num_threads;
2017 	cip->total_threads = vc->num_threads;
2018 	cip->subcore_threads[0] = vc->num_threads;
2019 	cip->subcore_vm[0] = vc->kvm;
2020 	for (sub = 0; sub < MAX_SUBCORES; ++sub)
2021 		INIT_LIST_HEAD(&cip->vcs[sub]);
2022 	list_add_tail(&vc->preempt_list, &cip->vcs[0]);
2023 }
2024 
2025 static bool subcore_config_ok(int n_subcores, int n_threads)
2026 {
2027 	/* Can only dynamically split if unsplit to begin with */
2028 	if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2029 		return false;
2030 	if (n_subcores > MAX_SUBCORES)
2031 		return false;
2032 	if (n_subcores > 1) {
2033 		if (!(dynamic_mt_modes & 2))
2034 			n_subcores = 4;
2035 		if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2036 			return false;
2037 	}
2038 
2039 	return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2040 }
2041 
2042 static void init_master_vcore(struct kvmppc_vcore *vc)
2043 {
2044 	vc->master_vcore = vc;
2045 	vc->entry_exit_map = 0;
2046 	vc->in_guest = 0;
2047 	vc->napping_threads = 0;
2048 	vc->conferring_threads = 0;
2049 }
2050 
2051 /*
2052  * See if the existing subcores can be split into 3 (or fewer) subcores
2053  * of at most two threads each, so we can fit in another vcore.  This
2054  * assumes there are at most two subcores and at most 6 threads in total.
2055  */
2056 static bool can_split_piggybacked_subcores(struct core_info *cip)
2057 {
2058 	int sub, new_sub;
2059 	int large_sub = -1;
2060 	int thr;
2061 	int n_subcores = cip->n_subcores;
2062 	struct kvmppc_vcore *vc, *vcnext;
2063 	struct kvmppc_vcore *master_vc = NULL;
2064 
2065 	for (sub = 0; sub < cip->n_subcores; ++sub) {
2066 		if (cip->subcore_threads[sub] <= 2)
2067 			continue;
2068 		if (large_sub >= 0)
2069 			return false;
2070 		large_sub = sub;
2071 		vc = list_first_entry(&cip->vcs[sub], struct kvmppc_vcore,
2072 				      preempt_list);
2073 		if (vc->num_threads > 2)
2074 			return false;
2075 		n_subcores += (cip->subcore_threads[sub] - 1) >> 1;
2076 	}
2077 	if (large_sub < 0 || !subcore_config_ok(n_subcores + 1, 2))
2078 		return false;
2079 
2080 	/*
2081 	 * Seems feasible, so go through and move vcores to new subcores.
2082 	 * Note that when we have two or more vcores in one subcore,
2083 	 * all those vcores must have only one thread each.
2084 	 */
2085 	new_sub = cip->n_subcores;
2086 	thr = 0;
2087 	sub = large_sub;
2088 	list_for_each_entry_safe(vc, vcnext, &cip->vcs[sub], preempt_list) {
2089 		if (thr >= 2) {
2090 			list_del(&vc->preempt_list);
2091 			list_add_tail(&vc->preempt_list, &cip->vcs[new_sub]);
2092 			/* vc->num_threads must be 1 */
2093 			if (++cip->subcore_threads[new_sub] == 1) {
2094 				cip->subcore_vm[new_sub] = vc->kvm;
2095 				init_master_vcore(vc);
2096 				master_vc = vc;
2097 				++cip->n_subcores;
2098 			} else {
2099 				vc->master_vcore = master_vc;
2100 				++new_sub;
2101 			}
2102 		}
2103 		thr += vc->num_threads;
2104 	}
2105 	cip->subcore_threads[large_sub] = 2;
2106 	cip->max_subcore_threads = 2;
2107 
2108 	return true;
2109 }
2110 
2111 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2112 {
2113 	int n_threads = vc->num_threads;
2114 	int sub;
2115 
2116 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2117 		return false;
2118 
2119 	if (n_threads < cip->max_subcore_threads)
2120 		n_threads = cip->max_subcore_threads;
2121 	if (subcore_config_ok(cip->n_subcores + 1, n_threads)) {
2122 		cip->max_subcore_threads = n_threads;
2123 	} else if (cip->n_subcores <= 2 && cip->total_threads <= 6 &&
2124 		   vc->num_threads <= 2) {
2125 		/*
2126 		 * We may be able to fit another subcore in by
2127 		 * splitting an existing subcore with 3 or 4
2128 		 * threads into two 2-thread subcores, or one
2129 		 * with 5 or 6 threads into three subcores.
2130 		 * We can only do this if those subcores have
2131 		 * piggybacked virtual cores.
2132 		 */
2133 		if (!can_split_piggybacked_subcores(cip))
2134 			return false;
2135 	} else {
2136 		return false;
2137 	}
2138 
2139 	sub = cip->n_subcores;
2140 	++cip->n_subcores;
2141 	cip->total_threads += vc->num_threads;
2142 	cip->subcore_threads[sub] = vc->num_threads;
2143 	cip->subcore_vm[sub] = vc->kvm;
2144 	init_master_vcore(vc);
2145 	list_del(&vc->preempt_list);
2146 	list_add_tail(&vc->preempt_list, &cip->vcs[sub]);
2147 
2148 	return true;
2149 }
2150 
2151 static bool can_piggyback_subcore(struct kvmppc_vcore *pvc,
2152 				  struct core_info *cip, int sub)
2153 {
2154 	struct kvmppc_vcore *vc;
2155 	int n_thr;
2156 
2157 	vc = list_first_entry(&cip->vcs[sub], struct kvmppc_vcore,
2158 			      preempt_list);
2159 
2160 	/* require same VM and same per-core reg values */
2161 	if (pvc->kvm != vc->kvm ||
2162 	    pvc->tb_offset != vc->tb_offset ||
2163 	    pvc->pcr != vc->pcr ||
2164 	    pvc->lpcr != vc->lpcr)
2165 		return false;
2166 
2167 	/* P8 guest with > 1 thread per core would see wrong TIR value */
2168 	if (cpu_has_feature(CPU_FTR_ARCH_207S) &&
2169 	    (vc->num_threads > 1 || pvc->num_threads > 1))
2170 		return false;
2171 
2172 	n_thr = cip->subcore_threads[sub] + pvc->num_threads;
2173 	if (n_thr > cip->max_subcore_threads) {
2174 		if (!subcore_config_ok(cip->n_subcores, n_thr))
2175 			return false;
2176 		cip->max_subcore_threads = n_thr;
2177 	}
2178 
2179 	cip->total_threads += pvc->num_threads;
2180 	cip->subcore_threads[sub] = n_thr;
2181 	pvc->master_vcore = vc;
2182 	list_del(&pvc->preempt_list);
2183 	list_add_tail(&pvc->preempt_list, &cip->vcs[sub]);
2184 
2185 	return true;
2186 }
2187 
2188 /*
2189  * Work out whether it is possible to piggyback the execution of
2190  * vcore *pvc onto the execution of the other vcores described in *cip.
2191  */
2192 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2193 			  int target_threads)
2194 {
2195 	int sub;
2196 
2197 	if (cip->total_threads + pvc->num_threads > target_threads)
2198 		return false;
2199 	for (sub = 0; sub < cip->n_subcores; ++sub)
2200 		if (cip->subcore_threads[sub] &&
2201 		    can_piggyback_subcore(pvc, cip, sub))
2202 			return true;
2203 
2204 	if (can_dynamic_split(pvc, cip))
2205 		return true;
2206 
2207 	return false;
2208 }
2209 
2210 static void prepare_threads(struct kvmppc_vcore *vc)
2211 {
2212 	struct kvm_vcpu *vcpu, *vnext;
2213 
2214 	list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2215 				 arch.run_list) {
2216 		if (signal_pending(vcpu->arch.run_task))
2217 			vcpu->arch.ret = -EINTR;
2218 		else if (vcpu->arch.vpa.update_pending ||
2219 			 vcpu->arch.slb_shadow.update_pending ||
2220 			 vcpu->arch.dtl.update_pending)
2221 			vcpu->arch.ret = RESUME_GUEST;
2222 		else
2223 			continue;
2224 		kvmppc_remove_runnable(vc, vcpu);
2225 		wake_up(&vcpu->arch.cpu_run);
2226 	}
2227 }
2228 
2229 static void collect_piggybacks(struct core_info *cip, int target_threads)
2230 {
2231 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2232 	struct kvmppc_vcore *pvc, *vcnext;
2233 
2234 	spin_lock(&lp->lock);
2235 	list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2236 		if (!spin_trylock(&pvc->lock))
2237 			continue;
2238 		prepare_threads(pvc);
2239 		if (!pvc->n_runnable) {
2240 			list_del_init(&pvc->preempt_list);
2241 			if (pvc->runner == NULL) {
2242 				pvc->vcore_state = VCORE_INACTIVE;
2243 				kvmppc_core_end_stolen(pvc);
2244 			}
2245 			spin_unlock(&pvc->lock);
2246 			continue;
2247 		}
2248 		if (!can_piggyback(pvc, cip, target_threads)) {
2249 			spin_unlock(&pvc->lock);
2250 			continue;
2251 		}
2252 		kvmppc_core_end_stolen(pvc);
2253 		pvc->vcore_state = VCORE_PIGGYBACK;
2254 		if (cip->total_threads >= target_threads)
2255 			break;
2256 	}
2257 	spin_unlock(&lp->lock);
2258 }
2259 
2260 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2261 {
2262 	int still_running = 0;
2263 	u64 now;
2264 	long ret;
2265 	struct kvm_vcpu *vcpu, *vnext;
2266 
2267 	spin_lock(&vc->lock);
2268 	now = get_tb();
2269 	list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2270 				 arch.run_list) {
2271 		/* cancel pending dec exception if dec is positive */
2272 		if (now < vcpu->arch.dec_expires &&
2273 		    kvmppc_core_pending_dec(vcpu))
2274 			kvmppc_core_dequeue_dec(vcpu);
2275 
2276 		trace_kvm_guest_exit(vcpu);
2277 
2278 		ret = RESUME_GUEST;
2279 		if (vcpu->arch.trap)
2280 			ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2281 						    vcpu->arch.run_task);
2282 
2283 		vcpu->arch.ret = ret;
2284 		vcpu->arch.trap = 0;
2285 
2286 		if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2287 			if (vcpu->arch.pending_exceptions)
2288 				kvmppc_core_prepare_to_enter(vcpu);
2289 			if (vcpu->arch.ceded)
2290 				kvmppc_set_timer(vcpu);
2291 			else
2292 				++still_running;
2293 		} else {
2294 			kvmppc_remove_runnable(vc, vcpu);
2295 			wake_up(&vcpu->arch.cpu_run);
2296 		}
2297 	}
2298 	list_del_init(&vc->preempt_list);
2299 	if (!is_master) {
2300 		if (still_running > 0) {
2301 			kvmppc_vcore_preempt(vc);
2302 		} else if (vc->runner) {
2303 			vc->vcore_state = VCORE_PREEMPT;
2304 			kvmppc_core_start_stolen(vc);
2305 		} else {
2306 			vc->vcore_state = VCORE_INACTIVE;
2307 		}
2308 		if (vc->n_runnable > 0 && vc->runner == NULL) {
2309 			/* make sure there's a candidate runner awake */
2310 			vcpu = list_first_entry(&vc->runnable_threads,
2311 						struct kvm_vcpu, arch.run_list);
2312 			wake_up(&vcpu->arch.cpu_run);
2313 		}
2314 	}
2315 	spin_unlock(&vc->lock);
2316 }
2317 
2318 /*
2319  * Clear core from the list of active host cores as we are about to
2320  * enter the guest. Only do this if it is the primary thread of the
2321  * core (not if a subcore) that is entering the guest.
2322  */
2323 static inline void kvmppc_clear_host_core(int cpu)
2324 {
2325 	int core;
2326 
2327 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2328 		return;
2329 	/*
2330 	 * Memory barrier can be omitted here as we will do a smp_wmb()
2331 	 * later in kvmppc_start_thread and we need ensure that state is
2332 	 * visible to other CPUs only after we enter guest.
2333 	 */
2334 	core = cpu >> threads_shift;
2335 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2336 }
2337 
2338 /*
2339  * Advertise this core as an active host core since we exited the guest
2340  * Only need to do this if it is the primary thread of the core that is
2341  * exiting.
2342  */
2343 static inline void kvmppc_set_host_core(int cpu)
2344 {
2345 	int core;
2346 
2347 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2348 		return;
2349 
2350 	/*
2351 	 * Memory barrier can be omitted here because we do a spin_unlock
2352 	 * immediately after this which provides the memory barrier.
2353 	 */
2354 	core = cpu >> threads_shift;
2355 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2356 }
2357 
2358 /*
2359  * Run a set of guest threads on a physical core.
2360  * Called with vc->lock held.
2361  */
2362 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2363 {
2364 	struct kvm_vcpu *vcpu, *vnext;
2365 	int i;
2366 	int srcu_idx;
2367 	struct core_info core_info;
2368 	struct kvmppc_vcore *pvc, *vcnext;
2369 	struct kvm_split_mode split_info, *sip;
2370 	int split, subcore_size, active;
2371 	int sub;
2372 	bool thr0_done;
2373 	unsigned long cmd_bit, stat_bit;
2374 	int pcpu, thr;
2375 	int target_threads;
2376 
2377 	/*
2378 	 * Remove from the list any threads that have a signal pending
2379 	 * or need a VPA update done
2380 	 */
2381 	prepare_threads(vc);
2382 
2383 	/* if the runner is no longer runnable, let the caller pick a new one */
2384 	if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2385 		return;
2386 
2387 	/*
2388 	 * Initialize *vc.
2389 	 */
2390 	init_master_vcore(vc);
2391 	vc->preempt_tb = TB_NIL;
2392 
2393 	/*
2394 	 * Make sure we are running on primary threads, and that secondary
2395 	 * threads are offline.  Also check if the number of threads in this
2396 	 * guest are greater than the current system threads per guest.
2397 	 */
2398 	if ((threads_per_core > 1) &&
2399 	    ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
2400 		list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2401 					 arch.run_list) {
2402 			vcpu->arch.ret = -EBUSY;
2403 			kvmppc_remove_runnable(vc, vcpu);
2404 			wake_up(&vcpu->arch.cpu_run);
2405 		}
2406 		goto out;
2407 	}
2408 
2409 	/*
2410 	 * See if we could run any other vcores on the physical core
2411 	 * along with this one.
2412 	 */
2413 	init_core_info(&core_info, vc);
2414 	pcpu = smp_processor_id();
2415 	target_threads = threads_per_subcore;
2416 	if (target_smt_mode && target_smt_mode < target_threads)
2417 		target_threads = target_smt_mode;
2418 	if (vc->num_threads < target_threads)
2419 		collect_piggybacks(&core_info, target_threads);
2420 
2421 	/* Decide on micro-threading (split-core) mode */
2422 	subcore_size = threads_per_subcore;
2423 	cmd_bit = stat_bit = 0;
2424 	split = core_info.n_subcores;
2425 	sip = NULL;
2426 	if (split > 1) {
2427 		/* threads_per_subcore must be MAX_SMT_THREADS (8) here */
2428 		if (split == 2 && (dynamic_mt_modes & 2)) {
2429 			cmd_bit = HID0_POWER8_1TO2LPAR;
2430 			stat_bit = HID0_POWER8_2LPARMODE;
2431 		} else {
2432 			split = 4;
2433 			cmd_bit = HID0_POWER8_1TO4LPAR;
2434 			stat_bit = HID0_POWER8_4LPARMODE;
2435 		}
2436 		subcore_size = MAX_SMT_THREADS / split;
2437 		sip = &split_info;
2438 		memset(&split_info, 0, sizeof(split_info));
2439 		split_info.rpr = mfspr(SPRN_RPR);
2440 		split_info.pmmar = mfspr(SPRN_PMMAR);
2441 		split_info.ldbar = mfspr(SPRN_LDBAR);
2442 		split_info.subcore_size = subcore_size;
2443 		for (sub = 0; sub < core_info.n_subcores; ++sub)
2444 			split_info.master_vcs[sub] =
2445 				list_first_entry(&core_info.vcs[sub],
2446 					struct kvmppc_vcore, preempt_list);
2447 		/* order writes to split_info before kvm_split_mode pointer */
2448 		smp_wmb();
2449 	}
2450 	pcpu = smp_processor_id();
2451 	for (thr = 0; thr < threads_per_subcore; ++thr)
2452 		paca[pcpu + thr].kvm_hstate.kvm_split_mode = sip;
2453 
2454 	/* Initiate micro-threading (split-core) if required */
2455 	if (cmd_bit) {
2456 		unsigned long hid0 = mfspr(SPRN_HID0);
2457 
2458 		hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2459 		mb();
2460 		mtspr(SPRN_HID0, hid0);
2461 		isync();
2462 		for (;;) {
2463 			hid0 = mfspr(SPRN_HID0);
2464 			if (hid0 & stat_bit)
2465 				break;
2466 			cpu_relax();
2467 		}
2468 	}
2469 
2470 	kvmppc_clear_host_core(pcpu);
2471 
2472 	/* Start all the threads */
2473 	active = 0;
2474 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
2475 		thr = subcore_thread_map[sub];
2476 		thr0_done = false;
2477 		active |= 1 << thr;
2478 		list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list) {
2479 			pvc->pcpu = pcpu + thr;
2480 			list_for_each_entry(vcpu, &pvc->runnable_threads,
2481 					    arch.run_list) {
2482 				kvmppc_start_thread(vcpu, pvc);
2483 				kvmppc_create_dtl_entry(vcpu, pvc);
2484 				trace_kvm_guest_enter(vcpu);
2485 				if (!vcpu->arch.ptid)
2486 					thr0_done = true;
2487 				active |= 1 << (thr + vcpu->arch.ptid);
2488 			}
2489 			/*
2490 			 * We need to start the first thread of each subcore
2491 			 * even if it doesn't have a vcpu.
2492 			 */
2493 			if (pvc->master_vcore == pvc && !thr0_done)
2494 				kvmppc_start_thread(NULL, pvc);
2495 			thr += pvc->num_threads;
2496 		}
2497 	}
2498 
2499 	/*
2500 	 * Ensure that split_info.do_nap is set after setting
2501 	 * the vcore pointer in the PACA of the secondaries.
2502 	 */
2503 	smp_mb();
2504 	if (cmd_bit)
2505 		split_info.do_nap = 1;	/* ask secondaries to nap when done */
2506 
2507 	/*
2508 	 * When doing micro-threading, poke the inactive threads as well.
2509 	 * This gets them to the nap instruction after kvm_do_nap,
2510 	 * which reduces the time taken to unsplit later.
2511 	 */
2512 	if (split > 1)
2513 		for (thr = 1; thr < threads_per_subcore; ++thr)
2514 			if (!(active & (1 << thr)))
2515 				kvmppc_ipi_thread(pcpu + thr);
2516 
2517 	vc->vcore_state = VCORE_RUNNING;
2518 	preempt_disable();
2519 
2520 	trace_kvmppc_run_core(vc, 0);
2521 
2522 	for (sub = 0; sub < core_info.n_subcores; ++sub)
2523 		list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list)
2524 			spin_unlock(&pvc->lock);
2525 
2526 	guest_enter();
2527 
2528 	srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2529 
2530 	__kvmppc_vcore_entry();
2531 
2532 	srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2533 
2534 	spin_lock(&vc->lock);
2535 	/* prevent other vcpu threads from doing kvmppc_start_thread() now */
2536 	vc->vcore_state = VCORE_EXITING;
2537 
2538 	/* wait for secondary threads to finish writing their state to memory */
2539 	kvmppc_wait_for_nap();
2540 
2541 	/* Return to whole-core mode if we split the core earlier */
2542 	if (split > 1) {
2543 		unsigned long hid0 = mfspr(SPRN_HID0);
2544 		unsigned long loops = 0;
2545 
2546 		hid0 &= ~HID0_POWER8_DYNLPARDIS;
2547 		stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
2548 		mb();
2549 		mtspr(SPRN_HID0, hid0);
2550 		isync();
2551 		for (;;) {
2552 			hid0 = mfspr(SPRN_HID0);
2553 			if (!(hid0 & stat_bit))
2554 				break;
2555 			cpu_relax();
2556 			++loops;
2557 		}
2558 		split_info.do_nap = 0;
2559 	}
2560 
2561 	/* Let secondaries go back to the offline loop */
2562 	for (i = 0; i < threads_per_subcore; ++i) {
2563 		kvmppc_release_hwthread(pcpu + i);
2564 		if (sip && sip->napped[i])
2565 			kvmppc_ipi_thread(pcpu + i);
2566 	}
2567 
2568 	kvmppc_set_host_core(pcpu);
2569 
2570 	spin_unlock(&vc->lock);
2571 
2572 	/* make sure updates to secondary vcpu structs are visible now */
2573 	smp_mb();
2574 	guest_exit();
2575 
2576 	for (sub = 0; sub < core_info.n_subcores; ++sub)
2577 		list_for_each_entry_safe(pvc, vcnext, &core_info.vcs[sub],
2578 					 preempt_list)
2579 			post_guest_process(pvc, pvc == vc);
2580 
2581 	spin_lock(&vc->lock);
2582 	preempt_enable();
2583 
2584  out:
2585 	vc->vcore_state = VCORE_INACTIVE;
2586 	trace_kvmppc_run_core(vc, 1);
2587 }
2588 
2589 /*
2590  * Wait for some other vcpu thread to execute us, and
2591  * wake us up when we need to handle something in the host.
2592  */
2593 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
2594 				 struct kvm_vcpu *vcpu, int wait_state)
2595 {
2596 	DEFINE_WAIT(wait);
2597 
2598 	prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
2599 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2600 		spin_unlock(&vc->lock);
2601 		schedule();
2602 		spin_lock(&vc->lock);
2603 	}
2604 	finish_wait(&vcpu->arch.cpu_run, &wait);
2605 }
2606 
2607 /*
2608  * All the vcpus in this vcore are idle, so wait for a decrementer
2609  * or external interrupt to one of the vcpus.  vc->lock is held.
2610  */
2611 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
2612 {
2613 	struct kvm_vcpu *vcpu;
2614 	int do_sleep = 1;
2615 	DECLARE_SWAITQUEUE(wait);
2616 
2617 	prepare_to_swait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
2618 
2619 	/*
2620 	 * Check one last time for pending exceptions and ceded state after
2621 	 * we put ourselves on the wait queue
2622 	 */
2623 	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
2624 		if (vcpu->arch.pending_exceptions || !vcpu->arch.ceded) {
2625 			do_sleep = 0;
2626 			break;
2627 		}
2628 	}
2629 
2630 	if (!do_sleep) {
2631 		finish_swait(&vc->wq, &wait);
2632 		return;
2633 	}
2634 
2635 	vc->vcore_state = VCORE_SLEEPING;
2636 	trace_kvmppc_vcore_blocked(vc, 0);
2637 	spin_unlock(&vc->lock);
2638 	schedule();
2639 	finish_swait(&vc->wq, &wait);
2640 	spin_lock(&vc->lock);
2641 	vc->vcore_state = VCORE_INACTIVE;
2642 	trace_kvmppc_vcore_blocked(vc, 1);
2643 }
2644 
2645 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2646 {
2647 	int n_ceded;
2648 	struct kvmppc_vcore *vc;
2649 	struct kvm_vcpu *v, *vn;
2650 
2651 	trace_kvmppc_run_vcpu_enter(vcpu);
2652 
2653 	kvm_run->exit_reason = 0;
2654 	vcpu->arch.ret = RESUME_GUEST;
2655 	vcpu->arch.trap = 0;
2656 	kvmppc_update_vpas(vcpu);
2657 
2658 	/*
2659 	 * Synchronize with other threads in this virtual core
2660 	 */
2661 	vc = vcpu->arch.vcore;
2662 	spin_lock(&vc->lock);
2663 	vcpu->arch.ceded = 0;
2664 	vcpu->arch.run_task = current;
2665 	vcpu->arch.kvm_run = kvm_run;
2666 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
2667 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
2668 	vcpu->arch.busy_preempt = TB_NIL;
2669 	list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
2670 	++vc->n_runnable;
2671 
2672 	/*
2673 	 * This happens the first time this is called for a vcpu.
2674 	 * If the vcore is already running, we may be able to start
2675 	 * this thread straight away and have it join in.
2676 	 */
2677 	if (!signal_pending(current)) {
2678 		if (vc->vcore_state == VCORE_PIGGYBACK) {
2679 			struct kvmppc_vcore *mvc = vc->master_vcore;
2680 			if (spin_trylock(&mvc->lock)) {
2681 				if (mvc->vcore_state == VCORE_RUNNING &&
2682 				    !VCORE_IS_EXITING(mvc)) {
2683 					kvmppc_create_dtl_entry(vcpu, vc);
2684 					kvmppc_start_thread(vcpu, vc);
2685 					trace_kvm_guest_enter(vcpu);
2686 				}
2687 				spin_unlock(&mvc->lock);
2688 			}
2689 		} else if (vc->vcore_state == VCORE_RUNNING &&
2690 			   !VCORE_IS_EXITING(vc)) {
2691 			kvmppc_create_dtl_entry(vcpu, vc);
2692 			kvmppc_start_thread(vcpu, vc);
2693 			trace_kvm_guest_enter(vcpu);
2694 		} else if (vc->vcore_state == VCORE_SLEEPING) {
2695 			swake_up(&vc->wq);
2696 		}
2697 
2698 	}
2699 
2700 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2701 	       !signal_pending(current)) {
2702 		if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2703 			kvmppc_vcore_end_preempt(vc);
2704 
2705 		if (vc->vcore_state != VCORE_INACTIVE) {
2706 			kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
2707 			continue;
2708 		}
2709 		list_for_each_entry_safe(v, vn, &vc->runnable_threads,
2710 					 arch.run_list) {
2711 			kvmppc_core_prepare_to_enter(v);
2712 			if (signal_pending(v->arch.run_task)) {
2713 				kvmppc_remove_runnable(vc, v);
2714 				v->stat.signal_exits++;
2715 				v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
2716 				v->arch.ret = -EINTR;
2717 				wake_up(&v->arch.cpu_run);
2718 			}
2719 		}
2720 		if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2721 			break;
2722 		n_ceded = 0;
2723 		list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
2724 			if (!v->arch.pending_exceptions)
2725 				n_ceded += v->arch.ceded;
2726 			else
2727 				v->arch.ceded = 0;
2728 		}
2729 		vc->runner = vcpu;
2730 		if (n_ceded == vc->n_runnable) {
2731 			kvmppc_vcore_blocked(vc);
2732 		} else if (need_resched()) {
2733 			kvmppc_vcore_preempt(vc);
2734 			/* Let something else run */
2735 			cond_resched_lock(&vc->lock);
2736 			if (vc->vcore_state == VCORE_PREEMPT)
2737 				kvmppc_vcore_end_preempt(vc);
2738 		} else {
2739 			kvmppc_run_core(vc);
2740 		}
2741 		vc->runner = NULL;
2742 	}
2743 
2744 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2745 	       (vc->vcore_state == VCORE_RUNNING ||
2746 		vc->vcore_state == VCORE_EXITING ||
2747 		vc->vcore_state == VCORE_PIGGYBACK))
2748 		kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
2749 
2750 	if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2751 		kvmppc_vcore_end_preempt(vc);
2752 
2753 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2754 		kvmppc_remove_runnable(vc, vcpu);
2755 		vcpu->stat.signal_exits++;
2756 		kvm_run->exit_reason = KVM_EXIT_INTR;
2757 		vcpu->arch.ret = -EINTR;
2758 	}
2759 
2760 	if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
2761 		/* Wake up some vcpu to run the core */
2762 		v = list_first_entry(&vc->runnable_threads,
2763 				     struct kvm_vcpu, arch.run_list);
2764 		wake_up(&v->arch.cpu_run);
2765 	}
2766 
2767 	trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
2768 	spin_unlock(&vc->lock);
2769 	return vcpu->arch.ret;
2770 }
2771 
2772 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
2773 {
2774 	int r;
2775 	int srcu_idx;
2776 
2777 	if (!vcpu->arch.sane) {
2778 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2779 		return -EINVAL;
2780 	}
2781 
2782 	kvmppc_core_prepare_to_enter(vcpu);
2783 
2784 	/* No need to go into the guest when all we'll do is come back out */
2785 	if (signal_pending(current)) {
2786 		run->exit_reason = KVM_EXIT_INTR;
2787 		return -EINTR;
2788 	}
2789 
2790 	atomic_inc(&vcpu->kvm->arch.vcpus_running);
2791 	/* Order vcpus_running vs. hpte_setup_done, see kvmppc_alloc_reset_hpt */
2792 	smp_mb();
2793 
2794 	/* On the first time here, set up HTAB and VRMA */
2795 	if (!vcpu->kvm->arch.hpte_setup_done) {
2796 		r = kvmppc_hv_setup_htab_rma(vcpu);
2797 		if (r)
2798 			goto out;
2799 	}
2800 
2801 	flush_all_to_thread(current);
2802 
2803 	vcpu->arch.wqp = &vcpu->arch.vcore->wq;
2804 	vcpu->arch.pgdir = current->mm->pgd;
2805 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2806 
2807 	do {
2808 		r = kvmppc_run_vcpu(run, vcpu);
2809 
2810 		if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
2811 		    !(vcpu->arch.shregs.msr & MSR_PR)) {
2812 			trace_kvm_hcall_enter(vcpu);
2813 			r = kvmppc_pseries_do_hcall(vcpu);
2814 			trace_kvm_hcall_exit(vcpu, r);
2815 			kvmppc_core_prepare_to_enter(vcpu);
2816 		} else if (r == RESUME_PAGE_FAULT) {
2817 			srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2818 			r = kvmppc_book3s_hv_page_fault(run, vcpu,
2819 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
2820 			srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2821 		}
2822 	} while (is_kvmppc_resume_guest(r));
2823 
2824  out:
2825 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2826 	atomic_dec(&vcpu->kvm->arch.vcpus_running);
2827 	return r;
2828 }
2829 
2830 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
2831 				     int linux_psize)
2832 {
2833 	struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
2834 
2835 	if (!def->shift)
2836 		return;
2837 	(*sps)->page_shift = def->shift;
2838 	(*sps)->slb_enc = def->sllp;
2839 	(*sps)->enc[0].page_shift = def->shift;
2840 	(*sps)->enc[0].pte_enc = def->penc[linux_psize];
2841 	/*
2842 	 * Add 16MB MPSS support if host supports it
2843 	 */
2844 	if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
2845 		(*sps)->enc[1].page_shift = 24;
2846 		(*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
2847 	}
2848 	(*sps)++;
2849 }
2850 
2851 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
2852 					 struct kvm_ppc_smmu_info *info)
2853 {
2854 	struct kvm_ppc_one_seg_page_size *sps;
2855 
2856 	info->flags = KVM_PPC_PAGE_SIZES_REAL;
2857 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
2858 		info->flags |= KVM_PPC_1T_SEGMENTS;
2859 	info->slb_size = mmu_slb_size;
2860 
2861 	/* We only support these sizes for now, and no muti-size segments */
2862 	sps = &info->sps[0];
2863 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
2864 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
2865 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
2866 
2867 	return 0;
2868 }
2869 
2870 /*
2871  * Get (and clear) the dirty memory log for a memory slot.
2872  */
2873 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
2874 					 struct kvm_dirty_log *log)
2875 {
2876 	struct kvm_memslots *slots;
2877 	struct kvm_memory_slot *memslot;
2878 	int r;
2879 	unsigned long n;
2880 
2881 	mutex_lock(&kvm->slots_lock);
2882 
2883 	r = -EINVAL;
2884 	if (log->slot >= KVM_USER_MEM_SLOTS)
2885 		goto out;
2886 
2887 	slots = kvm_memslots(kvm);
2888 	memslot = id_to_memslot(slots, log->slot);
2889 	r = -ENOENT;
2890 	if (!memslot->dirty_bitmap)
2891 		goto out;
2892 
2893 	n = kvm_dirty_bitmap_bytes(memslot);
2894 	memset(memslot->dirty_bitmap, 0, n);
2895 
2896 	r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
2897 	if (r)
2898 		goto out;
2899 
2900 	r = -EFAULT;
2901 	if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
2902 		goto out;
2903 
2904 	r = 0;
2905 out:
2906 	mutex_unlock(&kvm->slots_lock);
2907 	return r;
2908 }
2909 
2910 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
2911 					struct kvm_memory_slot *dont)
2912 {
2913 	if (!dont || free->arch.rmap != dont->arch.rmap) {
2914 		vfree(free->arch.rmap);
2915 		free->arch.rmap = NULL;
2916 	}
2917 }
2918 
2919 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
2920 					 unsigned long npages)
2921 {
2922 	slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
2923 	if (!slot->arch.rmap)
2924 		return -ENOMEM;
2925 
2926 	return 0;
2927 }
2928 
2929 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
2930 					struct kvm_memory_slot *memslot,
2931 					const struct kvm_userspace_memory_region *mem)
2932 {
2933 	return 0;
2934 }
2935 
2936 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
2937 				const struct kvm_userspace_memory_region *mem,
2938 				const struct kvm_memory_slot *old,
2939 				const struct kvm_memory_slot *new)
2940 {
2941 	unsigned long npages = mem->memory_size >> PAGE_SHIFT;
2942 	struct kvm_memslots *slots;
2943 	struct kvm_memory_slot *memslot;
2944 
2945 	if (npages && old->npages) {
2946 		/*
2947 		 * If modifying a memslot, reset all the rmap dirty bits.
2948 		 * If this is a new memslot, we don't need to do anything
2949 		 * since the rmap array starts out as all zeroes,
2950 		 * i.e. no pages are dirty.
2951 		 */
2952 		slots = kvm_memslots(kvm);
2953 		memslot = id_to_memslot(slots, mem->slot);
2954 		kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
2955 	}
2956 }
2957 
2958 /*
2959  * Update LPCR values in kvm->arch and in vcores.
2960  * Caller must hold kvm->lock.
2961  */
2962 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
2963 {
2964 	long int i;
2965 	u32 cores_done = 0;
2966 
2967 	if ((kvm->arch.lpcr & mask) == lpcr)
2968 		return;
2969 
2970 	kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
2971 
2972 	for (i = 0; i < KVM_MAX_VCORES; ++i) {
2973 		struct kvmppc_vcore *vc = kvm->arch.vcores[i];
2974 		if (!vc)
2975 			continue;
2976 		spin_lock(&vc->lock);
2977 		vc->lpcr = (vc->lpcr & ~mask) | lpcr;
2978 		spin_unlock(&vc->lock);
2979 		if (++cores_done >= kvm->arch.online_vcores)
2980 			break;
2981 	}
2982 }
2983 
2984 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
2985 {
2986 	return;
2987 }
2988 
2989 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
2990 {
2991 	int err = 0;
2992 	struct kvm *kvm = vcpu->kvm;
2993 	unsigned long hva;
2994 	struct kvm_memory_slot *memslot;
2995 	struct vm_area_struct *vma;
2996 	unsigned long lpcr = 0, senc;
2997 	unsigned long psize, porder;
2998 	int srcu_idx;
2999 
3000 	mutex_lock(&kvm->lock);
3001 	if (kvm->arch.hpte_setup_done)
3002 		goto out;	/* another vcpu beat us to it */
3003 
3004 	/* Allocate hashed page table (if not done already) and reset it */
3005 	if (!kvm->arch.hpt_virt) {
3006 		err = kvmppc_alloc_hpt(kvm, NULL);
3007 		if (err) {
3008 			pr_err("KVM: Couldn't alloc HPT\n");
3009 			goto out;
3010 		}
3011 	}
3012 
3013 	/* Look up the memslot for guest physical address 0 */
3014 	srcu_idx = srcu_read_lock(&kvm->srcu);
3015 	memslot = gfn_to_memslot(kvm, 0);
3016 
3017 	/* We must have some memory at 0 by now */
3018 	err = -EINVAL;
3019 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
3020 		goto out_srcu;
3021 
3022 	/* Look up the VMA for the start of this memory slot */
3023 	hva = memslot->userspace_addr;
3024 	down_read(&current->mm->mmap_sem);
3025 	vma = find_vma(current->mm, hva);
3026 	if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
3027 		goto up_out;
3028 
3029 	psize = vma_kernel_pagesize(vma);
3030 	porder = __ilog2(psize);
3031 
3032 	up_read(&current->mm->mmap_sem);
3033 
3034 	/* We can handle 4k, 64k or 16M pages in the VRMA */
3035 	err = -EINVAL;
3036 	if (!(psize == 0x1000 || psize == 0x10000 ||
3037 	      psize == 0x1000000))
3038 		goto out_srcu;
3039 
3040 	/* Update VRMASD field in the LPCR */
3041 	senc = slb_pgsize_encoding(psize);
3042 	kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
3043 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
3044 	/* the -4 is to account for senc values starting at 0x10 */
3045 	lpcr = senc << (LPCR_VRMASD_SH - 4);
3046 
3047 	/* Create HPTEs in the hash page table for the VRMA */
3048 	kvmppc_map_vrma(vcpu, memslot, porder);
3049 
3050 	kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
3051 
3052 	/* Order updates to kvm->arch.lpcr etc. vs. hpte_setup_done */
3053 	smp_wmb();
3054 	kvm->arch.hpte_setup_done = 1;
3055 	err = 0;
3056  out_srcu:
3057 	srcu_read_unlock(&kvm->srcu, srcu_idx);
3058  out:
3059 	mutex_unlock(&kvm->lock);
3060 	return err;
3061 
3062  up_out:
3063 	up_read(&current->mm->mmap_sem);
3064 	goto out_srcu;
3065 }
3066 
3067 #ifdef CONFIG_KVM_XICS
3068 static int kvmppc_cpu_notify(struct notifier_block *self, unsigned long action,
3069 			void *hcpu)
3070 {
3071 	unsigned long cpu = (long)hcpu;
3072 
3073 	switch (action) {
3074 	case CPU_UP_PREPARE:
3075 	case CPU_UP_PREPARE_FROZEN:
3076 		kvmppc_set_host_core(cpu);
3077 		break;
3078 
3079 #ifdef CONFIG_HOTPLUG_CPU
3080 	case CPU_DEAD:
3081 	case CPU_DEAD_FROZEN:
3082 	case CPU_UP_CANCELED:
3083 	case CPU_UP_CANCELED_FROZEN:
3084 		kvmppc_clear_host_core(cpu);
3085 		break;
3086 #endif
3087 	default:
3088 		break;
3089 	}
3090 
3091 	return NOTIFY_OK;
3092 }
3093 
3094 static struct notifier_block kvmppc_cpu_notifier = {
3095 	    .notifier_call = kvmppc_cpu_notify,
3096 };
3097 
3098 /*
3099  * Allocate a per-core structure for managing state about which cores are
3100  * running in the host versus the guest and for exchanging data between
3101  * real mode KVM and CPU running in the host.
3102  * This is only done for the first VM.
3103  * The allocated structure stays even if all VMs have stopped.
3104  * It is only freed when the kvm-hv module is unloaded.
3105  * It's OK for this routine to fail, we just don't support host
3106  * core operations like redirecting H_IPI wakeups.
3107  */
3108 void kvmppc_alloc_host_rm_ops(void)
3109 {
3110 	struct kvmppc_host_rm_ops *ops;
3111 	unsigned long l_ops;
3112 	int cpu, core;
3113 	int size;
3114 
3115 	/* Not the first time here ? */
3116 	if (kvmppc_host_rm_ops_hv != NULL)
3117 		return;
3118 
3119 	ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
3120 	if (!ops)
3121 		return;
3122 
3123 	size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
3124 	ops->rm_core = kzalloc(size, GFP_KERNEL);
3125 
3126 	if (!ops->rm_core) {
3127 		kfree(ops);
3128 		return;
3129 	}
3130 
3131 	get_online_cpus();
3132 
3133 	for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
3134 		if (!cpu_online(cpu))
3135 			continue;
3136 
3137 		core = cpu >> threads_shift;
3138 		ops->rm_core[core].rm_state.in_host = 1;
3139 	}
3140 
3141 	ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
3142 
3143 	/*
3144 	 * Make the contents of the kvmppc_host_rm_ops structure visible
3145 	 * to other CPUs before we assign it to the global variable.
3146 	 * Do an atomic assignment (no locks used here), but if someone
3147 	 * beats us to it, just free our copy and return.
3148 	 */
3149 	smp_wmb();
3150 	l_ops = (unsigned long) ops;
3151 
3152 	if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
3153 		put_online_cpus();
3154 		kfree(ops->rm_core);
3155 		kfree(ops);
3156 		return;
3157 	}
3158 
3159 	register_cpu_notifier(&kvmppc_cpu_notifier);
3160 
3161 	put_online_cpus();
3162 }
3163 
3164 void kvmppc_free_host_rm_ops(void)
3165 {
3166 	if (kvmppc_host_rm_ops_hv) {
3167 		unregister_cpu_notifier(&kvmppc_cpu_notifier);
3168 		kfree(kvmppc_host_rm_ops_hv->rm_core);
3169 		kfree(kvmppc_host_rm_ops_hv);
3170 		kvmppc_host_rm_ops_hv = NULL;
3171 	}
3172 }
3173 #endif
3174 
3175 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
3176 {
3177 	unsigned long lpcr, lpid;
3178 	char buf[32];
3179 
3180 	/* Allocate the guest's logical partition ID */
3181 
3182 	lpid = kvmppc_alloc_lpid();
3183 	if ((long)lpid < 0)
3184 		return -ENOMEM;
3185 	kvm->arch.lpid = lpid;
3186 
3187 	kvmppc_alloc_host_rm_ops();
3188 
3189 	/*
3190 	 * Since we don't flush the TLB when tearing down a VM,
3191 	 * and this lpid might have previously been used,
3192 	 * make sure we flush on each core before running the new VM.
3193 	 */
3194 	cpumask_setall(&kvm->arch.need_tlb_flush);
3195 
3196 	/* Start out with the default set of hcalls enabled */
3197 	memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
3198 	       sizeof(kvm->arch.enabled_hcalls));
3199 
3200 	kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
3201 
3202 	/* Init LPCR for virtual RMA mode */
3203 	kvm->arch.host_lpid = mfspr(SPRN_LPID);
3204 	kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
3205 	lpcr &= LPCR_PECE | LPCR_LPES;
3206 	lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3207 		LPCR_VPM0 | LPCR_VPM1;
3208 	kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3209 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
3210 	/* On POWER8 turn on online bit to enable PURR/SPURR */
3211 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
3212 		lpcr |= LPCR_ONL;
3213 	kvm->arch.lpcr = lpcr;
3214 
3215 	/*
3216 	 * Track that we now have a HV mode VM active. This blocks secondary
3217 	 * CPU threads from coming online.
3218 	 */
3219 	kvm_hv_vm_activated();
3220 
3221 	/*
3222 	 * Create a debugfs directory for the VM
3223 	 */
3224 	snprintf(buf, sizeof(buf), "vm%d", current->pid);
3225 	kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
3226 	if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
3227 		kvmppc_mmu_debugfs_init(kvm);
3228 
3229 	return 0;
3230 }
3231 
3232 static void kvmppc_free_vcores(struct kvm *kvm)
3233 {
3234 	long int i;
3235 
3236 	for (i = 0; i < KVM_MAX_VCORES; ++i)
3237 		kfree(kvm->arch.vcores[i]);
3238 	kvm->arch.online_vcores = 0;
3239 }
3240 
3241 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
3242 {
3243 	debugfs_remove_recursive(kvm->arch.debugfs_dir);
3244 
3245 	kvm_hv_vm_deactivated();
3246 
3247 	kvmppc_free_vcores(kvm);
3248 
3249 	kvmppc_free_hpt(kvm);
3250 }
3251 
3252 /* We don't need to emulate any privileged instructions or dcbz */
3253 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
3254 				     unsigned int inst, int *advance)
3255 {
3256 	return EMULATE_FAIL;
3257 }
3258 
3259 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
3260 					ulong spr_val)
3261 {
3262 	return EMULATE_FAIL;
3263 }
3264 
3265 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
3266 					ulong *spr_val)
3267 {
3268 	return EMULATE_FAIL;
3269 }
3270 
3271 static int kvmppc_core_check_processor_compat_hv(void)
3272 {
3273 	if (!cpu_has_feature(CPU_FTR_HVMODE) ||
3274 	    !cpu_has_feature(CPU_FTR_ARCH_206))
3275 		return -EIO;
3276 	/*
3277 	 * Disable KVM for Power9, untill the required bits merged.
3278 	 */
3279 	if (cpu_has_feature(CPU_FTR_ARCH_300))
3280 		return -EIO;
3281 
3282 	return 0;
3283 }
3284 
3285 static long kvm_arch_vm_ioctl_hv(struct file *filp,
3286 				 unsigned int ioctl, unsigned long arg)
3287 {
3288 	struct kvm *kvm __maybe_unused = filp->private_data;
3289 	void __user *argp = (void __user *)arg;
3290 	long r;
3291 
3292 	switch (ioctl) {
3293 
3294 	case KVM_PPC_ALLOCATE_HTAB: {
3295 		u32 htab_order;
3296 
3297 		r = -EFAULT;
3298 		if (get_user(htab_order, (u32 __user *)argp))
3299 			break;
3300 		r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
3301 		if (r)
3302 			break;
3303 		r = -EFAULT;
3304 		if (put_user(htab_order, (u32 __user *)argp))
3305 			break;
3306 		r = 0;
3307 		break;
3308 	}
3309 
3310 	case KVM_PPC_GET_HTAB_FD: {
3311 		struct kvm_get_htab_fd ghf;
3312 
3313 		r = -EFAULT;
3314 		if (copy_from_user(&ghf, argp, sizeof(ghf)))
3315 			break;
3316 		r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
3317 		break;
3318 	}
3319 
3320 	default:
3321 		r = -ENOTTY;
3322 	}
3323 
3324 	return r;
3325 }
3326 
3327 /*
3328  * List of hcall numbers to enable by default.
3329  * For compatibility with old userspace, we enable by default
3330  * all hcalls that were implemented before the hcall-enabling
3331  * facility was added.  Note this list should not include H_RTAS.
3332  */
3333 static unsigned int default_hcall_list[] = {
3334 	H_REMOVE,
3335 	H_ENTER,
3336 	H_READ,
3337 	H_PROTECT,
3338 	H_BULK_REMOVE,
3339 	H_GET_TCE,
3340 	H_PUT_TCE,
3341 	H_SET_DABR,
3342 	H_SET_XDABR,
3343 	H_CEDE,
3344 	H_PROD,
3345 	H_CONFER,
3346 	H_REGISTER_VPA,
3347 #ifdef CONFIG_KVM_XICS
3348 	H_EOI,
3349 	H_CPPR,
3350 	H_IPI,
3351 	H_IPOLL,
3352 	H_XIRR,
3353 	H_XIRR_X,
3354 #endif
3355 	0
3356 };
3357 
3358 static void init_default_hcalls(void)
3359 {
3360 	int i;
3361 	unsigned int hcall;
3362 
3363 	for (i = 0; default_hcall_list[i]; ++i) {
3364 		hcall = default_hcall_list[i];
3365 		WARN_ON(!kvmppc_hcall_impl_hv(hcall));
3366 		__set_bit(hcall / 4, default_enabled_hcalls);
3367 	}
3368 }
3369 
3370 static struct kvmppc_ops kvm_ops_hv = {
3371 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
3372 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
3373 	.get_one_reg = kvmppc_get_one_reg_hv,
3374 	.set_one_reg = kvmppc_set_one_reg_hv,
3375 	.vcpu_load   = kvmppc_core_vcpu_load_hv,
3376 	.vcpu_put    = kvmppc_core_vcpu_put_hv,
3377 	.set_msr     = kvmppc_set_msr_hv,
3378 	.vcpu_run    = kvmppc_vcpu_run_hv,
3379 	.vcpu_create = kvmppc_core_vcpu_create_hv,
3380 	.vcpu_free   = kvmppc_core_vcpu_free_hv,
3381 	.check_requests = kvmppc_core_check_requests_hv,
3382 	.get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
3383 	.flush_memslot  = kvmppc_core_flush_memslot_hv,
3384 	.prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
3385 	.commit_memory_region  = kvmppc_core_commit_memory_region_hv,
3386 	.unmap_hva = kvm_unmap_hva_hv,
3387 	.unmap_hva_range = kvm_unmap_hva_range_hv,
3388 	.age_hva  = kvm_age_hva_hv,
3389 	.test_age_hva = kvm_test_age_hva_hv,
3390 	.set_spte_hva = kvm_set_spte_hva_hv,
3391 	.mmu_destroy  = kvmppc_mmu_destroy_hv,
3392 	.free_memslot = kvmppc_core_free_memslot_hv,
3393 	.create_memslot = kvmppc_core_create_memslot_hv,
3394 	.init_vm =  kvmppc_core_init_vm_hv,
3395 	.destroy_vm = kvmppc_core_destroy_vm_hv,
3396 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
3397 	.emulate_op = kvmppc_core_emulate_op_hv,
3398 	.emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
3399 	.emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
3400 	.fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
3401 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
3402 	.hcall_implemented = kvmppc_hcall_impl_hv,
3403 };
3404 
3405 static int kvm_init_subcore_bitmap(void)
3406 {
3407 	int i, j;
3408 	int nr_cores = cpu_nr_cores();
3409 	struct sibling_subcore_state *sibling_subcore_state;
3410 
3411 	for (i = 0; i < nr_cores; i++) {
3412 		int first_cpu = i * threads_per_core;
3413 		int node = cpu_to_node(first_cpu);
3414 
3415 		/* Ignore if it is already allocated. */
3416 		if (paca[first_cpu].sibling_subcore_state)
3417 			continue;
3418 
3419 		sibling_subcore_state =
3420 			kmalloc_node(sizeof(struct sibling_subcore_state),
3421 							GFP_KERNEL, node);
3422 		if (!sibling_subcore_state)
3423 			return -ENOMEM;
3424 
3425 		memset(sibling_subcore_state, 0,
3426 				sizeof(struct sibling_subcore_state));
3427 
3428 		for (j = 0; j < threads_per_core; j++) {
3429 			int cpu = first_cpu + j;
3430 
3431 			paca[cpu].sibling_subcore_state = sibling_subcore_state;
3432 		}
3433 	}
3434 	return 0;
3435 }
3436 
3437 static int kvmppc_book3s_init_hv(void)
3438 {
3439 	int r;
3440 	/*
3441 	 * FIXME!! Do we need to check on all cpus ?
3442 	 */
3443 	r = kvmppc_core_check_processor_compat_hv();
3444 	if (r < 0)
3445 		return -ENODEV;
3446 
3447 	r = kvm_init_subcore_bitmap();
3448 	if (r)
3449 		return r;
3450 
3451 	kvm_ops_hv.owner = THIS_MODULE;
3452 	kvmppc_hv_ops = &kvm_ops_hv;
3453 
3454 	init_default_hcalls();
3455 
3456 	init_vcore_lists();
3457 
3458 	r = kvmppc_mmu_hv_init();
3459 	return r;
3460 }
3461 
3462 static void kvmppc_book3s_exit_hv(void)
3463 {
3464 	kvmppc_free_host_rm_ops();
3465 	kvmppc_hv_ops = NULL;
3466 }
3467 
3468 module_init(kvmppc_book3s_init_hv);
3469 module_exit(kvmppc_book3s_exit_hv);
3470 MODULE_LICENSE("GPL");
3471 MODULE_ALIAS_MISCDEV(KVM_MINOR);
3472 MODULE_ALIAS("devname:kvm");
3473