xref: /linux/arch/powerpc/kvm/book3s_hv.c (revision c0c914eca7f251c70facc37dfebeaf176601918d)
1 /*
2  * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
3  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
4  *
5  * Authors:
6  *    Paul Mackerras <paulus@au1.ibm.com>
7  *    Alexander Graf <agraf@suse.de>
8  *    Kevin Wolf <mail@kevin-wolf.de>
9  *
10  * Description: KVM functions specific to running on Book 3S
11  * processors in hypervisor mode (specifically POWER7 and later).
12  *
13  * This file is derived from arch/powerpc/kvm/book3s.c,
14  * by Alexander Graf <agraf@suse.de>.
15  *
16  * This program is free software; you can redistribute it and/or modify
17  * it under the terms of the GNU General Public License, version 2, as
18  * published by the Free Software Foundation.
19  */
20 
21 #include <linux/kvm_host.h>
22 #include <linux/err.h>
23 #include <linux/slab.h>
24 #include <linux/preempt.h>
25 #include <linux/sched.h>
26 #include <linux/delay.h>
27 #include <linux/export.h>
28 #include <linux/fs.h>
29 #include <linux/anon_inodes.h>
30 #include <linux/cpumask.h>
31 #include <linux/spinlock.h>
32 #include <linux/page-flags.h>
33 #include <linux/srcu.h>
34 #include <linux/miscdevice.h>
35 #include <linux/debugfs.h>
36 
37 #include <asm/reg.h>
38 #include <asm/cputable.h>
39 #include <asm/cacheflush.h>
40 #include <asm/tlbflush.h>
41 #include <asm/uaccess.h>
42 #include <asm/io.h>
43 #include <asm/kvm_ppc.h>
44 #include <asm/kvm_book3s.h>
45 #include <asm/mmu_context.h>
46 #include <asm/lppaca.h>
47 #include <asm/processor.h>
48 #include <asm/cputhreads.h>
49 #include <asm/page.h>
50 #include <asm/hvcall.h>
51 #include <asm/switch_to.h>
52 #include <asm/smp.h>
53 #include <asm/dbell.h>
54 #include <linux/gfp.h>
55 #include <linux/vmalloc.h>
56 #include <linux/highmem.h>
57 #include <linux/hugetlb.h>
58 #include <linux/module.h>
59 
60 #include "book3s.h"
61 
62 #define CREATE_TRACE_POINTS
63 #include "trace_hv.h"
64 
65 /* #define EXIT_DEBUG */
66 /* #define EXIT_DEBUG_SIMPLE */
67 /* #define EXIT_DEBUG_INT */
68 
69 /* Used to indicate that a guest page fault needs to be handled */
70 #define RESUME_PAGE_FAULT	(RESUME_GUEST | RESUME_FLAG_ARCH1)
71 
72 /* Used as a "null" value for timebase values */
73 #define TB_NIL	(~(u64)0)
74 
75 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
76 
77 static int dynamic_mt_modes = 6;
78 module_param(dynamic_mt_modes, int, S_IRUGO | S_IWUSR);
79 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
80 static int target_smt_mode;
81 module_param(target_smt_mode, int, S_IRUGO | S_IWUSR);
82 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
83 
84 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
85 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
86 
87 static bool kvmppc_ipi_thread(int cpu)
88 {
89 	/* On POWER8 for IPIs to threads in the same core, use msgsnd */
90 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
91 		preempt_disable();
92 		if (cpu_first_thread_sibling(cpu) ==
93 		    cpu_first_thread_sibling(smp_processor_id())) {
94 			unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
95 			msg |= cpu_thread_in_core(cpu);
96 			smp_mb();
97 			__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
98 			preempt_enable();
99 			return true;
100 		}
101 		preempt_enable();
102 	}
103 
104 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
105 	if (cpu >= 0 && cpu < nr_cpu_ids && paca[cpu].kvm_hstate.xics_phys) {
106 		xics_wake_cpu(cpu);
107 		return true;
108 	}
109 #endif
110 
111 	return false;
112 }
113 
114 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
115 {
116 	int cpu;
117 	wait_queue_head_t *wqp;
118 
119 	wqp = kvm_arch_vcpu_wq(vcpu);
120 	if (waitqueue_active(wqp)) {
121 		wake_up_interruptible(wqp);
122 		++vcpu->stat.halt_wakeup;
123 	}
124 
125 	if (kvmppc_ipi_thread(vcpu->arch.thread_cpu))
126 		return;
127 
128 	/* CPU points to the first thread of the core */
129 	cpu = vcpu->cpu;
130 	if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
131 		smp_send_reschedule(cpu);
132 }
133 
134 /*
135  * We use the vcpu_load/put functions to measure stolen time.
136  * Stolen time is counted as time when either the vcpu is able to
137  * run as part of a virtual core, but the task running the vcore
138  * is preempted or sleeping, or when the vcpu needs something done
139  * in the kernel by the task running the vcpu, but that task is
140  * preempted or sleeping.  Those two things have to be counted
141  * separately, since one of the vcpu tasks will take on the job
142  * of running the core, and the other vcpu tasks in the vcore will
143  * sleep waiting for it to do that, but that sleep shouldn't count
144  * as stolen time.
145  *
146  * Hence we accumulate stolen time when the vcpu can run as part of
147  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
148  * needs its task to do other things in the kernel (for example,
149  * service a page fault) in busy_stolen.  We don't accumulate
150  * stolen time for a vcore when it is inactive, or for a vcpu
151  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
152  * a misnomer; it means that the vcpu task is not executing in
153  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
154  * the kernel.  We don't have any way of dividing up that time
155  * between time that the vcpu is genuinely stopped, time that
156  * the task is actively working on behalf of the vcpu, and time
157  * that the task is preempted, so we don't count any of it as
158  * stolen.
159  *
160  * Updates to busy_stolen are protected by arch.tbacct_lock;
161  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
162  * lock.  The stolen times are measured in units of timebase ticks.
163  * (Note that the != TB_NIL checks below are purely defensive;
164  * they should never fail.)
165  */
166 
167 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
168 {
169 	unsigned long flags;
170 
171 	spin_lock_irqsave(&vc->stoltb_lock, flags);
172 	vc->preempt_tb = mftb();
173 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
174 }
175 
176 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
177 {
178 	unsigned long flags;
179 
180 	spin_lock_irqsave(&vc->stoltb_lock, flags);
181 	if (vc->preempt_tb != TB_NIL) {
182 		vc->stolen_tb += mftb() - vc->preempt_tb;
183 		vc->preempt_tb = TB_NIL;
184 	}
185 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
186 }
187 
188 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
189 {
190 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
191 	unsigned long flags;
192 
193 	/*
194 	 * We can test vc->runner without taking the vcore lock,
195 	 * because only this task ever sets vc->runner to this
196 	 * vcpu, and once it is set to this vcpu, only this task
197 	 * ever sets it to NULL.
198 	 */
199 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
200 		kvmppc_core_end_stolen(vc);
201 
202 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
203 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
204 	    vcpu->arch.busy_preempt != TB_NIL) {
205 		vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
206 		vcpu->arch.busy_preempt = TB_NIL;
207 	}
208 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
209 }
210 
211 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
212 {
213 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
214 	unsigned long flags;
215 
216 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
217 		kvmppc_core_start_stolen(vc);
218 
219 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
220 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
221 		vcpu->arch.busy_preempt = mftb();
222 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
223 }
224 
225 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
226 {
227 	/*
228 	 * Check for illegal transactional state bit combination
229 	 * and if we find it, force the TS field to a safe state.
230 	 */
231 	if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
232 		msr &= ~MSR_TS_MASK;
233 	vcpu->arch.shregs.msr = msr;
234 	kvmppc_end_cede(vcpu);
235 }
236 
237 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
238 {
239 	vcpu->arch.pvr = pvr;
240 }
241 
242 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
243 {
244 	unsigned long pcr = 0;
245 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
246 
247 	if (arch_compat) {
248 		switch (arch_compat) {
249 		case PVR_ARCH_205:
250 			/*
251 			 * If an arch bit is set in PCR, all the defined
252 			 * higher-order arch bits also have to be set.
253 			 */
254 			pcr = PCR_ARCH_206 | PCR_ARCH_205;
255 			break;
256 		case PVR_ARCH_206:
257 		case PVR_ARCH_206p:
258 			pcr = PCR_ARCH_206;
259 			break;
260 		case PVR_ARCH_207:
261 			break;
262 		default:
263 			return -EINVAL;
264 		}
265 
266 		if (!cpu_has_feature(CPU_FTR_ARCH_207S)) {
267 			/* POWER7 can't emulate POWER8 */
268 			if (!(pcr & PCR_ARCH_206))
269 				return -EINVAL;
270 			pcr &= ~PCR_ARCH_206;
271 		}
272 	}
273 
274 	spin_lock(&vc->lock);
275 	vc->arch_compat = arch_compat;
276 	vc->pcr = pcr;
277 	spin_unlock(&vc->lock);
278 
279 	return 0;
280 }
281 
282 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
283 {
284 	int r;
285 
286 	pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
287 	pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
288 	       vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
289 	for (r = 0; r < 16; ++r)
290 		pr_err("r%2d = %.16lx  r%d = %.16lx\n",
291 		       r, kvmppc_get_gpr(vcpu, r),
292 		       r+16, kvmppc_get_gpr(vcpu, r+16));
293 	pr_err("ctr = %.16lx  lr  = %.16lx\n",
294 	       vcpu->arch.ctr, vcpu->arch.lr);
295 	pr_err("srr0 = %.16llx srr1 = %.16llx\n",
296 	       vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
297 	pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
298 	       vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
299 	pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
300 	       vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
301 	pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
302 	       vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
303 	pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
304 	pr_err("fault dar = %.16lx dsisr = %.8x\n",
305 	       vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
306 	pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
307 	for (r = 0; r < vcpu->arch.slb_max; ++r)
308 		pr_err("  ESID = %.16llx VSID = %.16llx\n",
309 		       vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
310 	pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
311 	       vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
312 	       vcpu->arch.last_inst);
313 }
314 
315 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
316 {
317 	struct kvm_vcpu *ret;
318 
319 	mutex_lock(&kvm->lock);
320 	ret = kvm_get_vcpu_by_id(kvm, id);
321 	mutex_unlock(&kvm->lock);
322 	return ret;
323 }
324 
325 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
326 {
327 	vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
328 	vpa->yield_count = cpu_to_be32(1);
329 }
330 
331 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
332 		   unsigned long addr, unsigned long len)
333 {
334 	/* check address is cacheline aligned */
335 	if (addr & (L1_CACHE_BYTES - 1))
336 		return -EINVAL;
337 	spin_lock(&vcpu->arch.vpa_update_lock);
338 	if (v->next_gpa != addr || v->len != len) {
339 		v->next_gpa = addr;
340 		v->len = addr ? len : 0;
341 		v->update_pending = 1;
342 	}
343 	spin_unlock(&vcpu->arch.vpa_update_lock);
344 	return 0;
345 }
346 
347 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
348 struct reg_vpa {
349 	u32 dummy;
350 	union {
351 		__be16 hword;
352 		__be32 word;
353 	} length;
354 };
355 
356 static int vpa_is_registered(struct kvmppc_vpa *vpap)
357 {
358 	if (vpap->update_pending)
359 		return vpap->next_gpa != 0;
360 	return vpap->pinned_addr != NULL;
361 }
362 
363 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
364 				       unsigned long flags,
365 				       unsigned long vcpuid, unsigned long vpa)
366 {
367 	struct kvm *kvm = vcpu->kvm;
368 	unsigned long len, nb;
369 	void *va;
370 	struct kvm_vcpu *tvcpu;
371 	int err;
372 	int subfunc;
373 	struct kvmppc_vpa *vpap;
374 
375 	tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
376 	if (!tvcpu)
377 		return H_PARAMETER;
378 
379 	subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
380 	if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
381 	    subfunc == H_VPA_REG_SLB) {
382 		/* Registering new area - address must be cache-line aligned */
383 		if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
384 			return H_PARAMETER;
385 
386 		/* convert logical addr to kernel addr and read length */
387 		va = kvmppc_pin_guest_page(kvm, vpa, &nb);
388 		if (va == NULL)
389 			return H_PARAMETER;
390 		if (subfunc == H_VPA_REG_VPA)
391 			len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
392 		else
393 			len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
394 		kvmppc_unpin_guest_page(kvm, va, vpa, false);
395 
396 		/* Check length */
397 		if (len > nb || len < sizeof(struct reg_vpa))
398 			return H_PARAMETER;
399 	} else {
400 		vpa = 0;
401 		len = 0;
402 	}
403 
404 	err = H_PARAMETER;
405 	vpap = NULL;
406 	spin_lock(&tvcpu->arch.vpa_update_lock);
407 
408 	switch (subfunc) {
409 	case H_VPA_REG_VPA:		/* register VPA */
410 		if (len < sizeof(struct lppaca))
411 			break;
412 		vpap = &tvcpu->arch.vpa;
413 		err = 0;
414 		break;
415 
416 	case H_VPA_REG_DTL:		/* register DTL */
417 		if (len < sizeof(struct dtl_entry))
418 			break;
419 		len -= len % sizeof(struct dtl_entry);
420 
421 		/* Check that they have previously registered a VPA */
422 		err = H_RESOURCE;
423 		if (!vpa_is_registered(&tvcpu->arch.vpa))
424 			break;
425 
426 		vpap = &tvcpu->arch.dtl;
427 		err = 0;
428 		break;
429 
430 	case H_VPA_REG_SLB:		/* register SLB shadow buffer */
431 		/* Check that they have previously registered a VPA */
432 		err = H_RESOURCE;
433 		if (!vpa_is_registered(&tvcpu->arch.vpa))
434 			break;
435 
436 		vpap = &tvcpu->arch.slb_shadow;
437 		err = 0;
438 		break;
439 
440 	case H_VPA_DEREG_VPA:		/* deregister VPA */
441 		/* Check they don't still have a DTL or SLB buf registered */
442 		err = H_RESOURCE;
443 		if (vpa_is_registered(&tvcpu->arch.dtl) ||
444 		    vpa_is_registered(&tvcpu->arch.slb_shadow))
445 			break;
446 
447 		vpap = &tvcpu->arch.vpa;
448 		err = 0;
449 		break;
450 
451 	case H_VPA_DEREG_DTL:		/* deregister DTL */
452 		vpap = &tvcpu->arch.dtl;
453 		err = 0;
454 		break;
455 
456 	case H_VPA_DEREG_SLB:		/* deregister SLB shadow buffer */
457 		vpap = &tvcpu->arch.slb_shadow;
458 		err = 0;
459 		break;
460 	}
461 
462 	if (vpap) {
463 		vpap->next_gpa = vpa;
464 		vpap->len = len;
465 		vpap->update_pending = 1;
466 	}
467 
468 	spin_unlock(&tvcpu->arch.vpa_update_lock);
469 
470 	return err;
471 }
472 
473 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
474 {
475 	struct kvm *kvm = vcpu->kvm;
476 	void *va;
477 	unsigned long nb;
478 	unsigned long gpa;
479 
480 	/*
481 	 * We need to pin the page pointed to by vpap->next_gpa,
482 	 * but we can't call kvmppc_pin_guest_page under the lock
483 	 * as it does get_user_pages() and down_read().  So we
484 	 * have to drop the lock, pin the page, then get the lock
485 	 * again and check that a new area didn't get registered
486 	 * in the meantime.
487 	 */
488 	for (;;) {
489 		gpa = vpap->next_gpa;
490 		spin_unlock(&vcpu->arch.vpa_update_lock);
491 		va = NULL;
492 		nb = 0;
493 		if (gpa)
494 			va = kvmppc_pin_guest_page(kvm, gpa, &nb);
495 		spin_lock(&vcpu->arch.vpa_update_lock);
496 		if (gpa == vpap->next_gpa)
497 			break;
498 		/* sigh... unpin that one and try again */
499 		if (va)
500 			kvmppc_unpin_guest_page(kvm, va, gpa, false);
501 	}
502 
503 	vpap->update_pending = 0;
504 	if (va && nb < vpap->len) {
505 		/*
506 		 * If it's now too short, it must be that userspace
507 		 * has changed the mappings underlying guest memory,
508 		 * so unregister the region.
509 		 */
510 		kvmppc_unpin_guest_page(kvm, va, gpa, false);
511 		va = NULL;
512 	}
513 	if (vpap->pinned_addr)
514 		kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
515 					vpap->dirty);
516 	vpap->gpa = gpa;
517 	vpap->pinned_addr = va;
518 	vpap->dirty = false;
519 	if (va)
520 		vpap->pinned_end = va + vpap->len;
521 }
522 
523 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
524 {
525 	if (!(vcpu->arch.vpa.update_pending ||
526 	      vcpu->arch.slb_shadow.update_pending ||
527 	      vcpu->arch.dtl.update_pending))
528 		return;
529 
530 	spin_lock(&vcpu->arch.vpa_update_lock);
531 	if (vcpu->arch.vpa.update_pending) {
532 		kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
533 		if (vcpu->arch.vpa.pinned_addr)
534 			init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
535 	}
536 	if (vcpu->arch.dtl.update_pending) {
537 		kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
538 		vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
539 		vcpu->arch.dtl_index = 0;
540 	}
541 	if (vcpu->arch.slb_shadow.update_pending)
542 		kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
543 	spin_unlock(&vcpu->arch.vpa_update_lock);
544 }
545 
546 /*
547  * Return the accumulated stolen time for the vcore up until `now'.
548  * The caller should hold the vcore lock.
549  */
550 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
551 {
552 	u64 p;
553 	unsigned long flags;
554 
555 	spin_lock_irqsave(&vc->stoltb_lock, flags);
556 	p = vc->stolen_tb;
557 	if (vc->vcore_state != VCORE_INACTIVE &&
558 	    vc->preempt_tb != TB_NIL)
559 		p += now - vc->preempt_tb;
560 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
561 	return p;
562 }
563 
564 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
565 				    struct kvmppc_vcore *vc)
566 {
567 	struct dtl_entry *dt;
568 	struct lppaca *vpa;
569 	unsigned long stolen;
570 	unsigned long core_stolen;
571 	u64 now;
572 
573 	dt = vcpu->arch.dtl_ptr;
574 	vpa = vcpu->arch.vpa.pinned_addr;
575 	now = mftb();
576 	core_stolen = vcore_stolen_time(vc, now);
577 	stolen = core_stolen - vcpu->arch.stolen_logged;
578 	vcpu->arch.stolen_logged = core_stolen;
579 	spin_lock_irq(&vcpu->arch.tbacct_lock);
580 	stolen += vcpu->arch.busy_stolen;
581 	vcpu->arch.busy_stolen = 0;
582 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
583 	if (!dt || !vpa)
584 		return;
585 	memset(dt, 0, sizeof(struct dtl_entry));
586 	dt->dispatch_reason = 7;
587 	dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
588 	dt->timebase = cpu_to_be64(now + vc->tb_offset);
589 	dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
590 	dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
591 	dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
592 	++dt;
593 	if (dt == vcpu->arch.dtl.pinned_end)
594 		dt = vcpu->arch.dtl.pinned_addr;
595 	vcpu->arch.dtl_ptr = dt;
596 	/* order writing *dt vs. writing vpa->dtl_idx */
597 	smp_wmb();
598 	vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
599 	vcpu->arch.dtl.dirty = true;
600 }
601 
602 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
603 {
604 	if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
605 		return true;
606 	if ((!vcpu->arch.vcore->arch_compat) &&
607 	    cpu_has_feature(CPU_FTR_ARCH_207S))
608 		return true;
609 	return false;
610 }
611 
612 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
613 			     unsigned long resource, unsigned long value1,
614 			     unsigned long value2)
615 {
616 	switch (resource) {
617 	case H_SET_MODE_RESOURCE_SET_CIABR:
618 		if (!kvmppc_power8_compatible(vcpu))
619 			return H_P2;
620 		if (value2)
621 			return H_P4;
622 		if (mflags)
623 			return H_UNSUPPORTED_FLAG_START;
624 		/* Guests can't breakpoint the hypervisor */
625 		if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
626 			return H_P3;
627 		vcpu->arch.ciabr  = value1;
628 		return H_SUCCESS;
629 	case H_SET_MODE_RESOURCE_SET_DAWR:
630 		if (!kvmppc_power8_compatible(vcpu))
631 			return H_P2;
632 		if (mflags)
633 			return H_UNSUPPORTED_FLAG_START;
634 		if (value2 & DABRX_HYP)
635 			return H_P4;
636 		vcpu->arch.dawr  = value1;
637 		vcpu->arch.dawrx = value2;
638 		return H_SUCCESS;
639 	default:
640 		return H_TOO_HARD;
641 	}
642 }
643 
644 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
645 {
646 	struct kvmppc_vcore *vcore = target->arch.vcore;
647 
648 	/*
649 	 * We expect to have been called by the real mode handler
650 	 * (kvmppc_rm_h_confer()) which would have directly returned
651 	 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
652 	 * have useful work to do and should not confer) so we don't
653 	 * recheck that here.
654 	 */
655 
656 	spin_lock(&vcore->lock);
657 	if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
658 	    vcore->vcore_state != VCORE_INACTIVE &&
659 	    vcore->runner)
660 		target = vcore->runner;
661 	spin_unlock(&vcore->lock);
662 
663 	return kvm_vcpu_yield_to(target);
664 }
665 
666 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
667 {
668 	int yield_count = 0;
669 	struct lppaca *lppaca;
670 
671 	spin_lock(&vcpu->arch.vpa_update_lock);
672 	lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
673 	if (lppaca)
674 		yield_count = be32_to_cpu(lppaca->yield_count);
675 	spin_unlock(&vcpu->arch.vpa_update_lock);
676 	return yield_count;
677 }
678 
679 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
680 {
681 	unsigned long req = kvmppc_get_gpr(vcpu, 3);
682 	unsigned long target, ret = H_SUCCESS;
683 	int yield_count;
684 	struct kvm_vcpu *tvcpu;
685 	int idx, rc;
686 
687 	if (req <= MAX_HCALL_OPCODE &&
688 	    !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
689 		return RESUME_HOST;
690 
691 	switch (req) {
692 	case H_CEDE:
693 		break;
694 	case H_PROD:
695 		target = kvmppc_get_gpr(vcpu, 4);
696 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
697 		if (!tvcpu) {
698 			ret = H_PARAMETER;
699 			break;
700 		}
701 		tvcpu->arch.prodded = 1;
702 		smp_mb();
703 		if (vcpu->arch.ceded) {
704 			if (waitqueue_active(&vcpu->wq)) {
705 				wake_up_interruptible(&vcpu->wq);
706 				vcpu->stat.halt_wakeup++;
707 			}
708 		}
709 		break;
710 	case H_CONFER:
711 		target = kvmppc_get_gpr(vcpu, 4);
712 		if (target == -1)
713 			break;
714 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
715 		if (!tvcpu) {
716 			ret = H_PARAMETER;
717 			break;
718 		}
719 		yield_count = kvmppc_get_gpr(vcpu, 5);
720 		if (kvmppc_get_yield_count(tvcpu) != yield_count)
721 			break;
722 		kvm_arch_vcpu_yield_to(tvcpu);
723 		break;
724 	case H_REGISTER_VPA:
725 		ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
726 					kvmppc_get_gpr(vcpu, 5),
727 					kvmppc_get_gpr(vcpu, 6));
728 		break;
729 	case H_RTAS:
730 		if (list_empty(&vcpu->kvm->arch.rtas_tokens))
731 			return RESUME_HOST;
732 
733 		idx = srcu_read_lock(&vcpu->kvm->srcu);
734 		rc = kvmppc_rtas_hcall(vcpu);
735 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
736 
737 		if (rc == -ENOENT)
738 			return RESUME_HOST;
739 		else if (rc == 0)
740 			break;
741 
742 		/* Send the error out to userspace via KVM_RUN */
743 		return rc;
744 	case H_LOGICAL_CI_LOAD:
745 		ret = kvmppc_h_logical_ci_load(vcpu);
746 		if (ret == H_TOO_HARD)
747 			return RESUME_HOST;
748 		break;
749 	case H_LOGICAL_CI_STORE:
750 		ret = kvmppc_h_logical_ci_store(vcpu);
751 		if (ret == H_TOO_HARD)
752 			return RESUME_HOST;
753 		break;
754 	case H_SET_MODE:
755 		ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
756 					kvmppc_get_gpr(vcpu, 5),
757 					kvmppc_get_gpr(vcpu, 6),
758 					kvmppc_get_gpr(vcpu, 7));
759 		if (ret == H_TOO_HARD)
760 			return RESUME_HOST;
761 		break;
762 	case H_XIRR:
763 	case H_CPPR:
764 	case H_EOI:
765 	case H_IPI:
766 	case H_IPOLL:
767 	case H_XIRR_X:
768 		if (kvmppc_xics_enabled(vcpu)) {
769 			ret = kvmppc_xics_hcall(vcpu, req);
770 			break;
771 		} /* fallthrough */
772 	default:
773 		return RESUME_HOST;
774 	}
775 	kvmppc_set_gpr(vcpu, 3, ret);
776 	vcpu->arch.hcall_needed = 0;
777 	return RESUME_GUEST;
778 }
779 
780 static int kvmppc_hcall_impl_hv(unsigned long cmd)
781 {
782 	switch (cmd) {
783 	case H_CEDE:
784 	case H_PROD:
785 	case H_CONFER:
786 	case H_REGISTER_VPA:
787 	case H_SET_MODE:
788 	case H_LOGICAL_CI_LOAD:
789 	case H_LOGICAL_CI_STORE:
790 #ifdef CONFIG_KVM_XICS
791 	case H_XIRR:
792 	case H_CPPR:
793 	case H_EOI:
794 	case H_IPI:
795 	case H_IPOLL:
796 	case H_XIRR_X:
797 #endif
798 		return 1;
799 	}
800 
801 	/* See if it's in the real-mode table */
802 	return kvmppc_hcall_impl_hv_realmode(cmd);
803 }
804 
805 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
806 					struct kvm_vcpu *vcpu)
807 {
808 	u32 last_inst;
809 
810 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
811 					EMULATE_DONE) {
812 		/*
813 		 * Fetch failed, so return to guest and
814 		 * try executing it again.
815 		 */
816 		return RESUME_GUEST;
817 	}
818 
819 	if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
820 		run->exit_reason = KVM_EXIT_DEBUG;
821 		run->debug.arch.address = kvmppc_get_pc(vcpu);
822 		return RESUME_HOST;
823 	} else {
824 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
825 		return RESUME_GUEST;
826 	}
827 }
828 
829 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
830 				 struct task_struct *tsk)
831 {
832 	int r = RESUME_HOST;
833 
834 	vcpu->stat.sum_exits++;
835 
836 	/*
837 	 * This can happen if an interrupt occurs in the last stages
838 	 * of guest entry or the first stages of guest exit (i.e. after
839 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
840 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
841 	 * That can happen due to a bug, or due to a machine check
842 	 * occurring at just the wrong time.
843 	 */
844 	if (vcpu->arch.shregs.msr & MSR_HV) {
845 		printk(KERN_EMERG "KVM trap in HV mode!\n");
846 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
847 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
848 			vcpu->arch.shregs.msr);
849 		kvmppc_dump_regs(vcpu);
850 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
851 		run->hw.hardware_exit_reason = vcpu->arch.trap;
852 		return RESUME_HOST;
853 	}
854 	run->exit_reason = KVM_EXIT_UNKNOWN;
855 	run->ready_for_interrupt_injection = 1;
856 	switch (vcpu->arch.trap) {
857 	/* We're good on these - the host merely wanted to get our attention */
858 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
859 		vcpu->stat.dec_exits++;
860 		r = RESUME_GUEST;
861 		break;
862 	case BOOK3S_INTERRUPT_EXTERNAL:
863 	case BOOK3S_INTERRUPT_H_DOORBELL:
864 		vcpu->stat.ext_intr_exits++;
865 		r = RESUME_GUEST;
866 		break;
867 	/* HMI is hypervisor interrupt and host has handled it. Resume guest.*/
868 	case BOOK3S_INTERRUPT_HMI:
869 	case BOOK3S_INTERRUPT_PERFMON:
870 		r = RESUME_GUEST;
871 		break;
872 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
873 		/*
874 		 * Deliver a machine check interrupt to the guest.
875 		 * We have to do this, even if the host has handled the
876 		 * machine check, because machine checks use SRR0/1 and
877 		 * the interrupt might have trashed guest state in them.
878 		 */
879 		kvmppc_book3s_queue_irqprio(vcpu,
880 					    BOOK3S_INTERRUPT_MACHINE_CHECK);
881 		r = RESUME_GUEST;
882 		break;
883 	case BOOK3S_INTERRUPT_PROGRAM:
884 	{
885 		ulong flags;
886 		/*
887 		 * Normally program interrupts are delivered directly
888 		 * to the guest by the hardware, but we can get here
889 		 * as a result of a hypervisor emulation interrupt
890 		 * (e40) getting turned into a 700 by BML RTAS.
891 		 */
892 		flags = vcpu->arch.shregs.msr & 0x1f0000ull;
893 		kvmppc_core_queue_program(vcpu, flags);
894 		r = RESUME_GUEST;
895 		break;
896 	}
897 	case BOOK3S_INTERRUPT_SYSCALL:
898 	{
899 		/* hcall - punt to userspace */
900 		int i;
901 
902 		/* hypercall with MSR_PR has already been handled in rmode,
903 		 * and never reaches here.
904 		 */
905 
906 		run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
907 		for (i = 0; i < 9; ++i)
908 			run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
909 		run->exit_reason = KVM_EXIT_PAPR_HCALL;
910 		vcpu->arch.hcall_needed = 1;
911 		r = RESUME_HOST;
912 		break;
913 	}
914 	/*
915 	 * We get these next two if the guest accesses a page which it thinks
916 	 * it has mapped but which is not actually present, either because
917 	 * it is for an emulated I/O device or because the corresonding
918 	 * host page has been paged out.  Any other HDSI/HISI interrupts
919 	 * have been handled already.
920 	 */
921 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
922 		r = RESUME_PAGE_FAULT;
923 		break;
924 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
925 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
926 		vcpu->arch.fault_dsisr = 0;
927 		r = RESUME_PAGE_FAULT;
928 		break;
929 	/*
930 	 * This occurs if the guest executes an illegal instruction.
931 	 * If the guest debug is disabled, generate a program interrupt
932 	 * to the guest. If guest debug is enabled, we need to check
933 	 * whether the instruction is a software breakpoint instruction.
934 	 * Accordingly return to Guest or Host.
935 	 */
936 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
937 		if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
938 			vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
939 				swab32(vcpu->arch.emul_inst) :
940 				vcpu->arch.emul_inst;
941 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
942 			r = kvmppc_emulate_debug_inst(run, vcpu);
943 		} else {
944 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
945 			r = RESUME_GUEST;
946 		}
947 		break;
948 	/*
949 	 * This occurs if the guest (kernel or userspace), does something that
950 	 * is prohibited by HFSCR.  We just generate a program interrupt to
951 	 * the guest.
952 	 */
953 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
954 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
955 		r = RESUME_GUEST;
956 		break;
957 	default:
958 		kvmppc_dump_regs(vcpu);
959 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
960 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
961 			vcpu->arch.shregs.msr);
962 		run->hw.hardware_exit_reason = vcpu->arch.trap;
963 		r = RESUME_HOST;
964 		break;
965 	}
966 
967 	return r;
968 }
969 
970 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
971 					    struct kvm_sregs *sregs)
972 {
973 	int i;
974 
975 	memset(sregs, 0, sizeof(struct kvm_sregs));
976 	sregs->pvr = vcpu->arch.pvr;
977 	for (i = 0; i < vcpu->arch.slb_max; i++) {
978 		sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
979 		sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
980 	}
981 
982 	return 0;
983 }
984 
985 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
986 					    struct kvm_sregs *sregs)
987 {
988 	int i, j;
989 
990 	/* Only accept the same PVR as the host's, since we can't spoof it */
991 	if (sregs->pvr != vcpu->arch.pvr)
992 		return -EINVAL;
993 
994 	j = 0;
995 	for (i = 0; i < vcpu->arch.slb_nr; i++) {
996 		if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
997 			vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
998 			vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
999 			++j;
1000 		}
1001 	}
1002 	vcpu->arch.slb_max = j;
1003 
1004 	return 0;
1005 }
1006 
1007 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1008 		bool preserve_top32)
1009 {
1010 	struct kvm *kvm = vcpu->kvm;
1011 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
1012 	u64 mask;
1013 
1014 	mutex_lock(&kvm->lock);
1015 	spin_lock(&vc->lock);
1016 	/*
1017 	 * If ILE (interrupt little-endian) has changed, update the
1018 	 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1019 	 */
1020 	if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1021 		struct kvm_vcpu *vcpu;
1022 		int i;
1023 
1024 		kvm_for_each_vcpu(i, vcpu, kvm) {
1025 			if (vcpu->arch.vcore != vc)
1026 				continue;
1027 			if (new_lpcr & LPCR_ILE)
1028 				vcpu->arch.intr_msr |= MSR_LE;
1029 			else
1030 				vcpu->arch.intr_msr &= ~MSR_LE;
1031 		}
1032 	}
1033 
1034 	/*
1035 	 * Userspace can only modify DPFD (default prefetch depth),
1036 	 * ILE (interrupt little-endian) and TC (translation control).
1037 	 * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
1038 	 */
1039 	mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1040 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
1041 		mask |= LPCR_AIL;
1042 
1043 	/* Broken 32-bit version of LPCR must not clear top bits */
1044 	if (preserve_top32)
1045 		mask &= 0xFFFFFFFF;
1046 	vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1047 	spin_unlock(&vc->lock);
1048 	mutex_unlock(&kvm->lock);
1049 }
1050 
1051 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1052 				 union kvmppc_one_reg *val)
1053 {
1054 	int r = 0;
1055 	long int i;
1056 
1057 	switch (id) {
1058 	case KVM_REG_PPC_DEBUG_INST:
1059 		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1060 		break;
1061 	case KVM_REG_PPC_HIOR:
1062 		*val = get_reg_val(id, 0);
1063 		break;
1064 	case KVM_REG_PPC_DABR:
1065 		*val = get_reg_val(id, vcpu->arch.dabr);
1066 		break;
1067 	case KVM_REG_PPC_DABRX:
1068 		*val = get_reg_val(id, vcpu->arch.dabrx);
1069 		break;
1070 	case KVM_REG_PPC_DSCR:
1071 		*val = get_reg_val(id, vcpu->arch.dscr);
1072 		break;
1073 	case KVM_REG_PPC_PURR:
1074 		*val = get_reg_val(id, vcpu->arch.purr);
1075 		break;
1076 	case KVM_REG_PPC_SPURR:
1077 		*val = get_reg_val(id, vcpu->arch.spurr);
1078 		break;
1079 	case KVM_REG_PPC_AMR:
1080 		*val = get_reg_val(id, vcpu->arch.amr);
1081 		break;
1082 	case KVM_REG_PPC_UAMOR:
1083 		*val = get_reg_val(id, vcpu->arch.uamor);
1084 		break;
1085 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1086 		i = id - KVM_REG_PPC_MMCR0;
1087 		*val = get_reg_val(id, vcpu->arch.mmcr[i]);
1088 		break;
1089 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1090 		i = id - KVM_REG_PPC_PMC1;
1091 		*val = get_reg_val(id, vcpu->arch.pmc[i]);
1092 		break;
1093 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1094 		i = id - KVM_REG_PPC_SPMC1;
1095 		*val = get_reg_val(id, vcpu->arch.spmc[i]);
1096 		break;
1097 	case KVM_REG_PPC_SIAR:
1098 		*val = get_reg_val(id, vcpu->arch.siar);
1099 		break;
1100 	case KVM_REG_PPC_SDAR:
1101 		*val = get_reg_val(id, vcpu->arch.sdar);
1102 		break;
1103 	case KVM_REG_PPC_SIER:
1104 		*val = get_reg_val(id, vcpu->arch.sier);
1105 		break;
1106 	case KVM_REG_PPC_IAMR:
1107 		*val = get_reg_val(id, vcpu->arch.iamr);
1108 		break;
1109 	case KVM_REG_PPC_PSPB:
1110 		*val = get_reg_val(id, vcpu->arch.pspb);
1111 		break;
1112 	case KVM_REG_PPC_DPDES:
1113 		*val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1114 		break;
1115 	case KVM_REG_PPC_DAWR:
1116 		*val = get_reg_val(id, vcpu->arch.dawr);
1117 		break;
1118 	case KVM_REG_PPC_DAWRX:
1119 		*val = get_reg_val(id, vcpu->arch.dawrx);
1120 		break;
1121 	case KVM_REG_PPC_CIABR:
1122 		*val = get_reg_val(id, vcpu->arch.ciabr);
1123 		break;
1124 	case KVM_REG_PPC_CSIGR:
1125 		*val = get_reg_val(id, vcpu->arch.csigr);
1126 		break;
1127 	case KVM_REG_PPC_TACR:
1128 		*val = get_reg_val(id, vcpu->arch.tacr);
1129 		break;
1130 	case KVM_REG_PPC_TCSCR:
1131 		*val = get_reg_val(id, vcpu->arch.tcscr);
1132 		break;
1133 	case KVM_REG_PPC_PID:
1134 		*val = get_reg_val(id, vcpu->arch.pid);
1135 		break;
1136 	case KVM_REG_PPC_ACOP:
1137 		*val = get_reg_val(id, vcpu->arch.acop);
1138 		break;
1139 	case KVM_REG_PPC_WORT:
1140 		*val = get_reg_val(id, vcpu->arch.wort);
1141 		break;
1142 	case KVM_REG_PPC_VPA_ADDR:
1143 		spin_lock(&vcpu->arch.vpa_update_lock);
1144 		*val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1145 		spin_unlock(&vcpu->arch.vpa_update_lock);
1146 		break;
1147 	case KVM_REG_PPC_VPA_SLB:
1148 		spin_lock(&vcpu->arch.vpa_update_lock);
1149 		val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1150 		val->vpaval.length = vcpu->arch.slb_shadow.len;
1151 		spin_unlock(&vcpu->arch.vpa_update_lock);
1152 		break;
1153 	case KVM_REG_PPC_VPA_DTL:
1154 		spin_lock(&vcpu->arch.vpa_update_lock);
1155 		val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1156 		val->vpaval.length = vcpu->arch.dtl.len;
1157 		spin_unlock(&vcpu->arch.vpa_update_lock);
1158 		break;
1159 	case KVM_REG_PPC_TB_OFFSET:
1160 		*val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1161 		break;
1162 	case KVM_REG_PPC_LPCR:
1163 	case KVM_REG_PPC_LPCR_64:
1164 		*val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1165 		break;
1166 	case KVM_REG_PPC_PPR:
1167 		*val = get_reg_val(id, vcpu->arch.ppr);
1168 		break;
1169 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1170 	case KVM_REG_PPC_TFHAR:
1171 		*val = get_reg_val(id, vcpu->arch.tfhar);
1172 		break;
1173 	case KVM_REG_PPC_TFIAR:
1174 		*val = get_reg_val(id, vcpu->arch.tfiar);
1175 		break;
1176 	case KVM_REG_PPC_TEXASR:
1177 		*val = get_reg_val(id, vcpu->arch.texasr);
1178 		break;
1179 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1180 		i = id - KVM_REG_PPC_TM_GPR0;
1181 		*val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1182 		break;
1183 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1184 	{
1185 		int j;
1186 		i = id - KVM_REG_PPC_TM_VSR0;
1187 		if (i < 32)
1188 			for (j = 0; j < TS_FPRWIDTH; j++)
1189 				val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1190 		else {
1191 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1192 				val->vval = vcpu->arch.vr_tm.vr[i-32];
1193 			else
1194 				r = -ENXIO;
1195 		}
1196 		break;
1197 	}
1198 	case KVM_REG_PPC_TM_CR:
1199 		*val = get_reg_val(id, vcpu->arch.cr_tm);
1200 		break;
1201 	case KVM_REG_PPC_TM_LR:
1202 		*val = get_reg_val(id, vcpu->arch.lr_tm);
1203 		break;
1204 	case KVM_REG_PPC_TM_CTR:
1205 		*val = get_reg_val(id, vcpu->arch.ctr_tm);
1206 		break;
1207 	case KVM_REG_PPC_TM_FPSCR:
1208 		*val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1209 		break;
1210 	case KVM_REG_PPC_TM_AMR:
1211 		*val = get_reg_val(id, vcpu->arch.amr_tm);
1212 		break;
1213 	case KVM_REG_PPC_TM_PPR:
1214 		*val = get_reg_val(id, vcpu->arch.ppr_tm);
1215 		break;
1216 	case KVM_REG_PPC_TM_VRSAVE:
1217 		*val = get_reg_val(id, vcpu->arch.vrsave_tm);
1218 		break;
1219 	case KVM_REG_PPC_TM_VSCR:
1220 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1221 			*val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1222 		else
1223 			r = -ENXIO;
1224 		break;
1225 	case KVM_REG_PPC_TM_DSCR:
1226 		*val = get_reg_val(id, vcpu->arch.dscr_tm);
1227 		break;
1228 	case KVM_REG_PPC_TM_TAR:
1229 		*val = get_reg_val(id, vcpu->arch.tar_tm);
1230 		break;
1231 #endif
1232 	case KVM_REG_PPC_ARCH_COMPAT:
1233 		*val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1234 		break;
1235 	default:
1236 		r = -EINVAL;
1237 		break;
1238 	}
1239 
1240 	return r;
1241 }
1242 
1243 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1244 				 union kvmppc_one_reg *val)
1245 {
1246 	int r = 0;
1247 	long int i;
1248 	unsigned long addr, len;
1249 
1250 	switch (id) {
1251 	case KVM_REG_PPC_HIOR:
1252 		/* Only allow this to be set to zero */
1253 		if (set_reg_val(id, *val))
1254 			r = -EINVAL;
1255 		break;
1256 	case KVM_REG_PPC_DABR:
1257 		vcpu->arch.dabr = set_reg_val(id, *val);
1258 		break;
1259 	case KVM_REG_PPC_DABRX:
1260 		vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1261 		break;
1262 	case KVM_REG_PPC_DSCR:
1263 		vcpu->arch.dscr = set_reg_val(id, *val);
1264 		break;
1265 	case KVM_REG_PPC_PURR:
1266 		vcpu->arch.purr = set_reg_val(id, *val);
1267 		break;
1268 	case KVM_REG_PPC_SPURR:
1269 		vcpu->arch.spurr = set_reg_val(id, *val);
1270 		break;
1271 	case KVM_REG_PPC_AMR:
1272 		vcpu->arch.amr = set_reg_val(id, *val);
1273 		break;
1274 	case KVM_REG_PPC_UAMOR:
1275 		vcpu->arch.uamor = set_reg_val(id, *val);
1276 		break;
1277 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1278 		i = id - KVM_REG_PPC_MMCR0;
1279 		vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1280 		break;
1281 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1282 		i = id - KVM_REG_PPC_PMC1;
1283 		vcpu->arch.pmc[i] = set_reg_val(id, *val);
1284 		break;
1285 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1286 		i = id - KVM_REG_PPC_SPMC1;
1287 		vcpu->arch.spmc[i] = set_reg_val(id, *val);
1288 		break;
1289 	case KVM_REG_PPC_SIAR:
1290 		vcpu->arch.siar = set_reg_val(id, *val);
1291 		break;
1292 	case KVM_REG_PPC_SDAR:
1293 		vcpu->arch.sdar = set_reg_val(id, *val);
1294 		break;
1295 	case KVM_REG_PPC_SIER:
1296 		vcpu->arch.sier = set_reg_val(id, *val);
1297 		break;
1298 	case KVM_REG_PPC_IAMR:
1299 		vcpu->arch.iamr = set_reg_val(id, *val);
1300 		break;
1301 	case KVM_REG_PPC_PSPB:
1302 		vcpu->arch.pspb = set_reg_val(id, *val);
1303 		break;
1304 	case KVM_REG_PPC_DPDES:
1305 		vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1306 		break;
1307 	case KVM_REG_PPC_DAWR:
1308 		vcpu->arch.dawr = set_reg_val(id, *val);
1309 		break;
1310 	case KVM_REG_PPC_DAWRX:
1311 		vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1312 		break;
1313 	case KVM_REG_PPC_CIABR:
1314 		vcpu->arch.ciabr = set_reg_val(id, *val);
1315 		/* Don't allow setting breakpoints in hypervisor code */
1316 		if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1317 			vcpu->arch.ciabr &= ~CIABR_PRIV;	/* disable */
1318 		break;
1319 	case KVM_REG_PPC_CSIGR:
1320 		vcpu->arch.csigr = set_reg_val(id, *val);
1321 		break;
1322 	case KVM_REG_PPC_TACR:
1323 		vcpu->arch.tacr = set_reg_val(id, *val);
1324 		break;
1325 	case KVM_REG_PPC_TCSCR:
1326 		vcpu->arch.tcscr = set_reg_val(id, *val);
1327 		break;
1328 	case KVM_REG_PPC_PID:
1329 		vcpu->arch.pid = set_reg_val(id, *val);
1330 		break;
1331 	case KVM_REG_PPC_ACOP:
1332 		vcpu->arch.acop = set_reg_val(id, *val);
1333 		break;
1334 	case KVM_REG_PPC_WORT:
1335 		vcpu->arch.wort = set_reg_val(id, *val);
1336 		break;
1337 	case KVM_REG_PPC_VPA_ADDR:
1338 		addr = set_reg_val(id, *val);
1339 		r = -EINVAL;
1340 		if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1341 			      vcpu->arch.dtl.next_gpa))
1342 			break;
1343 		r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1344 		break;
1345 	case KVM_REG_PPC_VPA_SLB:
1346 		addr = val->vpaval.addr;
1347 		len = val->vpaval.length;
1348 		r = -EINVAL;
1349 		if (addr && !vcpu->arch.vpa.next_gpa)
1350 			break;
1351 		r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1352 		break;
1353 	case KVM_REG_PPC_VPA_DTL:
1354 		addr = val->vpaval.addr;
1355 		len = val->vpaval.length;
1356 		r = -EINVAL;
1357 		if (addr && (len < sizeof(struct dtl_entry) ||
1358 			     !vcpu->arch.vpa.next_gpa))
1359 			break;
1360 		len -= len % sizeof(struct dtl_entry);
1361 		r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1362 		break;
1363 	case KVM_REG_PPC_TB_OFFSET:
1364 		/* round up to multiple of 2^24 */
1365 		vcpu->arch.vcore->tb_offset =
1366 			ALIGN(set_reg_val(id, *val), 1UL << 24);
1367 		break;
1368 	case KVM_REG_PPC_LPCR:
1369 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1370 		break;
1371 	case KVM_REG_PPC_LPCR_64:
1372 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1373 		break;
1374 	case KVM_REG_PPC_PPR:
1375 		vcpu->arch.ppr = set_reg_val(id, *val);
1376 		break;
1377 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1378 	case KVM_REG_PPC_TFHAR:
1379 		vcpu->arch.tfhar = set_reg_val(id, *val);
1380 		break;
1381 	case KVM_REG_PPC_TFIAR:
1382 		vcpu->arch.tfiar = set_reg_val(id, *val);
1383 		break;
1384 	case KVM_REG_PPC_TEXASR:
1385 		vcpu->arch.texasr = set_reg_val(id, *val);
1386 		break;
1387 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1388 		i = id - KVM_REG_PPC_TM_GPR0;
1389 		vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1390 		break;
1391 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1392 	{
1393 		int j;
1394 		i = id - KVM_REG_PPC_TM_VSR0;
1395 		if (i < 32)
1396 			for (j = 0; j < TS_FPRWIDTH; j++)
1397 				vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1398 		else
1399 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1400 				vcpu->arch.vr_tm.vr[i-32] = val->vval;
1401 			else
1402 				r = -ENXIO;
1403 		break;
1404 	}
1405 	case KVM_REG_PPC_TM_CR:
1406 		vcpu->arch.cr_tm = set_reg_val(id, *val);
1407 		break;
1408 	case KVM_REG_PPC_TM_LR:
1409 		vcpu->arch.lr_tm = set_reg_val(id, *val);
1410 		break;
1411 	case KVM_REG_PPC_TM_CTR:
1412 		vcpu->arch.ctr_tm = set_reg_val(id, *val);
1413 		break;
1414 	case KVM_REG_PPC_TM_FPSCR:
1415 		vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1416 		break;
1417 	case KVM_REG_PPC_TM_AMR:
1418 		vcpu->arch.amr_tm = set_reg_val(id, *val);
1419 		break;
1420 	case KVM_REG_PPC_TM_PPR:
1421 		vcpu->arch.ppr_tm = set_reg_val(id, *val);
1422 		break;
1423 	case KVM_REG_PPC_TM_VRSAVE:
1424 		vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1425 		break;
1426 	case KVM_REG_PPC_TM_VSCR:
1427 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1428 			vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1429 		else
1430 			r = - ENXIO;
1431 		break;
1432 	case KVM_REG_PPC_TM_DSCR:
1433 		vcpu->arch.dscr_tm = set_reg_val(id, *val);
1434 		break;
1435 	case KVM_REG_PPC_TM_TAR:
1436 		vcpu->arch.tar_tm = set_reg_val(id, *val);
1437 		break;
1438 #endif
1439 	case KVM_REG_PPC_ARCH_COMPAT:
1440 		r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1441 		break;
1442 	default:
1443 		r = -EINVAL;
1444 		break;
1445 	}
1446 
1447 	return r;
1448 }
1449 
1450 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1451 {
1452 	struct kvmppc_vcore *vcore;
1453 
1454 	vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1455 
1456 	if (vcore == NULL)
1457 		return NULL;
1458 
1459 	INIT_LIST_HEAD(&vcore->runnable_threads);
1460 	spin_lock_init(&vcore->lock);
1461 	spin_lock_init(&vcore->stoltb_lock);
1462 	init_waitqueue_head(&vcore->wq);
1463 	vcore->preempt_tb = TB_NIL;
1464 	vcore->lpcr = kvm->arch.lpcr;
1465 	vcore->first_vcpuid = core * threads_per_subcore;
1466 	vcore->kvm = kvm;
1467 	INIT_LIST_HEAD(&vcore->preempt_list);
1468 
1469 	return vcore;
1470 }
1471 
1472 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1473 static struct debugfs_timings_element {
1474 	const char *name;
1475 	size_t offset;
1476 } timings[] = {
1477 	{"rm_entry",	offsetof(struct kvm_vcpu, arch.rm_entry)},
1478 	{"rm_intr",	offsetof(struct kvm_vcpu, arch.rm_intr)},
1479 	{"rm_exit",	offsetof(struct kvm_vcpu, arch.rm_exit)},
1480 	{"guest",	offsetof(struct kvm_vcpu, arch.guest_time)},
1481 	{"cede",	offsetof(struct kvm_vcpu, arch.cede_time)},
1482 };
1483 
1484 #define N_TIMINGS	(sizeof(timings) / sizeof(timings[0]))
1485 
1486 struct debugfs_timings_state {
1487 	struct kvm_vcpu	*vcpu;
1488 	unsigned int	buflen;
1489 	char		buf[N_TIMINGS * 100];
1490 };
1491 
1492 static int debugfs_timings_open(struct inode *inode, struct file *file)
1493 {
1494 	struct kvm_vcpu *vcpu = inode->i_private;
1495 	struct debugfs_timings_state *p;
1496 
1497 	p = kzalloc(sizeof(*p), GFP_KERNEL);
1498 	if (!p)
1499 		return -ENOMEM;
1500 
1501 	kvm_get_kvm(vcpu->kvm);
1502 	p->vcpu = vcpu;
1503 	file->private_data = p;
1504 
1505 	return nonseekable_open(inode, file);
1506 }
1507 
1508 static int debugfs_timings_release(struct inode *inode, struct file *file)
1509 {
1510 	struct debugfs_timings_state *p = file->private_data;
1511 
1512 	kvm_put_kvm(p->vcpu->kvm);
1513 	kfree(p);
1514 	return 0;
1515 }
1516 
1517 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1518 				    size_t len, loff_t *ppos)
1519 {
1520 	struct debugfs_timings_state *p = file->private_data;
1521 	struct kvm_vcpu *vcpu = p->vcpu;
1522 	char *s, *buf_end;
1523 	struct kvmhv_tb_accumulator tb;
1524 	u64 count;
1525 	loff_t pos;
1526 	ssize_t n;
1527 	int i, loops;
1528 	bool ok;
1529 
1530 	if (!p->buflen) {
1531 		s = p->buf;
1532 		buf_end = s + sizeof(p->buf);
1533 		for (i = 0; i < N_TIMINGS; ++i) {
1534 			struct kvmhv_tb_accumulator *acc;
1535 
1536 			acc = (struct kvmhv_tb_accumulator *)
1537 				((unsigned long)vcpu + timings[i].offset);
1538 			ok = false;
1539 			for (loops = 0; loops < 1000; ++loops) {
1540 				count = acc->seqcount;
1541 				if (!(count & 1)) {
1542 					smp_rmb();
1543 					tb = *acc;
1544 					smp_rmb();
1545 					if (count == acc->seqcount) {
1546 						ok = true;
1547 						break;
1548 					}
1549 				}
1550 				udelay(1);
1551 			}
1552 			if (!ok)
1553 				snprintf(s, buf_end - s, "%s: stuck\n",
1554 					timings[i].name);
1555 			else
1556 				snprintf(s, buf_end - s,
1557 					"%s: %llu %llu %llu %llu\n",
1558 					timings[i].name, count / 2,
1559 					tb_to_ns(tb.tb_total),
1560 					tb_to_ns(tb.tb_min),
1561 					tb_to_ns(tb.tb_max));
1562 			s += strlen(s);
1563 		}
1564 		p->buflen = s - p->buf;
1565 	}
1566 
1567 	pos = *ppos;
1568 	if (pos >= p->buflen)
1569 		return 0;
1570 	if (len > p->buflen - pos)
1571 		len = p->buflen - pos;
1572 	n = copy_to_user(buf, p->buf + pos, len);
1573 	if (n) {
1574 		if (n == len)
1575 			return -EFAULT;
1576 		len -= n;
1577 	}
1578 	*ppos = pos + len;
1579 	return len;
1580 }
1581 
1582 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1583 				     size_t len, loff_t *ppos)
1584 {
1585 	return -EACCES;
1586 }
1587 
1588 static const struct file_operations debugfs_timings_ops = {
1589 	.owner	 = THIS_MODULE,
1590 	.open	 = debugfs_timings_open,
1591 	.release = debugfs_timings_release,
1592 	.read	 = debugfs_timings_read,
1593 	.write	 = debugfs_timings_write,
1594 	.llseek	 = generic_file_llseek,
1595 };
1596 
1597 /* Create a debugfs directory for the vcpu */
1598 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1599 {
1600 	char buf[16];
1601 	struct kvm *kvm = vcpu->kvm;
1602 
1603 	snprintf(buf, sizeof(buf), "vcpu%u", id);
1604 	if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1605 		return;
1606 	vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1607 	if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1608 		return;
1609 	vcpu->arch.debugfs_timings =
1610 		debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1611 				    vcpu, &debugfs_timings_ops);
1612 }
1613 
1614 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1615 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1616 {
1617 }
1618 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1619 
1620 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1621 						   unsigned int id)
1622 {
1623 	struct kvm_vcpu *vcpu;
1624 	int err = -EINVAL;
1625 	int core;
1626 	struct kvmppc_vcore *vcore;
1627 
1628 	core = id / threads_per_subcore;
1629 	if (core >= KVM_MAX_VCORES)
1630 		goto out;
1631 
1632 	err = -ENOMEM;
1633 	vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1634 	if (!vcpu)
1635 		goto out;
1636 
1637 	err = kvm_vcpu_init(vcpu, kvm, id);
1638 	if (err)
1639 		goto free_vcpu;
1640 
1641 	vcpu->arch.shared = &vcpu->arch.shregs;
1642 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1643 	/*
1644 	 * The shared struct is never shared on HV,
1645 	 * so we can always use host endianness
1646 	 */
1647 #ifdef __BIG_ENDIAN__
1648 	vcpu->arch.shared_big_endian = true;
1649 #else
1650 	vcpu->arch.shared_big_endian = false;
1651 #endif
1652 #endif
1653 	vcpu->arch.mmcr[0] = MMCR0_FC;
1654 	vcpu->arch.ctrl = CTRL_RUNLATCH;
1655 	/* default to host PVR, since we can't spoof it */
1656 	kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1657 	spin_lock_init(&vcpu->arch.vpa_update_lock);
1658 	spin_lock_init(&vcpu->arch.tbacct_lock);
1659 	vcpu->arch.busy_preempt = TB_NIL;
1660 	vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1661 
1662 	kvmppc_mmu_book3s_hv_init(vcpu);
1663 
1664 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1665 
1666 	init_waitqueue_head(&vcpu->arch.cpu_run);
1667 
1668 	mutex_lock(&kvm->lock);
1669 	vcore = kvm->arch.vcores[core];
1670 	if (!vcore) {
1671 		vcore = kvmppc_vcore_create(kvm, core);
1672 		kvm->arch.vcores[core] = vcore;
1673 		kvm->arch.online_vcores++;
1674 	}
1675 	mutex_unlock(&kvm->lock);
1676 
1677 	if (!vcore)
1678 		goto free_vcpu;
1679 
1680 	spin_lock(&vcore->lock);
1681 	++vcore->num_threads;
1682 	spin_unlock(&vcore->lock);
1683 	vcpu->arch.vcore = vcore;
1684 	vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1685 	vcpu->arch.thread_cpu = -1;
1686 
1687 	vcpu->arch.cpu_type = KVM_CPU_3S_64;
1688 	kvmppc_sanity_check(vcpu);
1689 
1690 	debugfs_vcpu_init(vcpu, id);
1691 
1692 	return vcpu;
1693 
1694 free_vcpu:
1695 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1696 out:
1697 	return ERR_PTR(err);
1698 }
1699 
1700 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1701 {
1702 	if (vpa->pinned_addr)
1703 		kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1704 					vpa->dirty);
1705 }
1706 
1707 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
1708 {
1709 	spin_lock(&vcpu->arch.vpa_update_lock);
1710 	unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1711 	unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1712 	unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1713 	spin_unlock(&vcpu->arch.vpa_update_lock);
1714 	kvm_vcpu_uninit(vcpu);
1715 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1716 }
1717 
1718 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1719 {
1720 	/* Indicate we want to get back into the guest */
1721 	return 1;
1722 }
1723 
1724 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1725 {
1726 	unsigned long dec_nsec, now;
1727 
1728 	now = get_tb();
1729 	if (now > vcpu->arch.dec_expires) {
1730 		/* decrementer has already gone negative */
1731 		kvmppc_core_queue_dec(vcpu);
1732 		kvmppc_core_prepare_to_enter(vcpu);
1733 		return;
1734 	}
1735 	dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1736 		   / tb_ticks_per_sec;
1737 	hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1738 		      HRTIMER_MODE_REL);
1739 	vcpu->arch.timer_running = 1;
1740 }
1741 
1742 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1743 {
1744 	vcpu->arch.ceded = 0;
1745 	if (vcpu->arch.timer_running) {
1746 		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1747 		vcpu->arch.timer_running = 0;
1748 	}
1749 }
1750 
1751 extern void __kvmppc_vcore_entry(void);
1752 
1753 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1754 				   struct kvm_vcpu *vcpu)
1755 {
1756 	u64 now;
1757 
1758 	if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1759 		return;
1760 	spin_lock_irq(&vcpu->arch.tbacct_lock);
1761 	now = mftb();
1762 	vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1763 		vcpu->arch.stolen_logged;
1764 	vcpu->arch.busy_preempt = now;
1765 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1766 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
1767 	--vc->n_runnable;
1768 	list_del(&vcpu->arch.run_list);
1769 }
1770 
1771 static int kvmppc_grab_hwthread(int cpu)
1772 {
1773 	struct paca_struct *tpaca;
1774 	long timeout = 10000;
1775 
1776 	tpaca = &paca[cpu];
1777 
1778 	/* Ensure the thread won't go into the kernel if it wakes */
1779 	tpaca->kvm_hstate.kvm_vcpu = NULL;
1780 	tpaca->kvm_hstate.kvm_vcore = NULL;
1781 	tpaca->kvm_hstate.napping = 0;
1782 	smp_wmb();
1783 	tpaca->kvm_hstate.hwthread_req = 1;
1784 
1785 	/*
1786 	 * If the thread is already executing in the kernel (e.g. handling
1787 	 * a stray interrupt), wait for it to get back to nap mode.
1788 	 * The smp_mb() is to ensure that our setting of hwthread_req
1789 	 * is visible before we look at hwthread_state, so if this
1790 	 * races with the code at system_reset_pSeries and the thread
1791 	 * misses our setting of hwthread_req, we are sure to see its
1792 	 * setting of hwthread_state, and vice versa.
1793 	 */
1794 	smp_mb();
1795 	while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1796 		if (--timeout <= 0) {
1797 			pr_err("KVM: couldn't grab cpu %d\n", cpu);
1798 			return -EBUSY;
1799 		}
1800 		udelay(1);
1801 	}
1802 	return 0;
1803 }
1804 
1805 static void kvmppc_release_hwthread(int cpu)
1806 {
1807 	struct paca_struct *tpaca;
1808 
1809 	tpaca = &paca[cpu];
1810 	tpaca->kvm_hstate.hwthread_req = 0;
1811 	tpaca->kvm_hstate.kvm_vcpu = NULL;
1812 	tpaca->kvm_hstate.kvm_vcore = NULL;
1813 	tpaca->kvm_hstate.kvm_split_mode = NULL;
1814 }
1815 
1816 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
1817 {
1818 	int cpu;
1819 	struct paca_struct *tpaca;
1820 	struct kvmppc_vcore *mvc = vc->master_vcore;
1821 
1822 	cpu = vc->pcpu;
1823 	if (vcpu) {
1824 		if (vcpu->arch.timer_running) {
1825 			hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1826 			vcpu->arch.timer_running = 0;
1827 		}
1828 		cpu += vcpu->arch.ptid;
1829 		vcpu->cpu = mvc->pcpu;
1830 		vcpu->arch.thread_cpu = cpu;
1831 	}
1832 	tpaca = &paca[cpu];
1833 	tpaca->kvm_hstate.kvm_vcpu = vcpu;
1834 	tpaca->kvm_hstate.ptid = cpu - mvc->pcpu;
1835 	/* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
1836 	smp_wmb();
1837 	tpaca->kvm_hstate.kvm_vcore = mvc;
1838 	if (cpu != smp_processor_id())
1839 		kvmppc_ipi_thread(cpu);
1840 }
1841 
1842 static void kvmppc_wait_for_nap(void)
1843 {
1844 	int cpu = smp_processor_id();
1845 	int i, loops;
1846 
1847 	for (loops = 0; loops < 1000000; ++loops) {
1848 		/*
1849 		 * Check if all threads are finished.
1850 		 * We set the vcore pointer when starting a thread
1851 		 * and the thread clears it when finished, so we look
1852 		 * for any threads that still have a non-NULL vcore ptr.
1853 		 */
1854 		for (i = 1; i < threads_per_subcore; ++i)
1855 			if (paca[cpu + i].kvm_hstate.kvm_vcore)
1856 				break;
1857 		if (i == threads_per_subcore) {
1858 			HMT_medium();
1859 			return;
1860 		}
1861 		HMT_low();
1862 	}
1863 	HMT_medium();
1864 	for (i = 1; i < threads_per_subcore; ++i)
1865 		if (paca[cpu + i].kvm_hstate.kvm_vcore)
1866 			pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
1867 }
1868 
1869 /*
1870  * Check that we are on thread 0 and that any other threads in
1871  * this core are off-line.  Then grab the threads so they can't
1872  * enter the kernel.
1873  */
1874 static int on_primary_thread(void)
1875 {
1876 	int cpu = smp_processor_id();
1877 	int thr;
1878 
1879 	/* Are we on a primary subcore? */
1880 	if (cpu_thread_in_subcore(cpu))
1881 		return 0;
1882 
1883 	thr = 0;
1884 	while (++thr < threads_per_subcore)
1885 		if (cpu_online(cpu + thr))
1886 			return 0;
1887 
1888 	/* Grab all hw threads so they can't go into the kernel */
1889 	for (thr = 1; thr < threads_per_subcore; ++thr) {
1890 		if (kvmppc_grab_hwthread(cpu + thr)) {
1891 			/* Couldn't grab one; let the others go */
1892 			do {
1893 				kvmppc_release_hwthread(cpu + thr);
1894 			} while (--thr > 0);
1895 			return 0;
1896 		}
1897 	}
1898 	return 1;
1899 }
1900 
1901 /*
1902  * A list of virtual cores for each physical CPU.
1903  * These are vcores that could run but their runner VCPU tasks are
1904  * (or may be) preempted.
1905  */
1906 struct preempted_vcore_list {
1907 	struct list_head	list;
1908 	spinlock_t		lock;
1909 };
1910 
1911 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
1912 
1913 static void init_vcore_lists(void)
1914 {
1915 	int cpu;
1916 
1917 	for_each_possible_cpu(cpu) {
1918 		struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
1919 		spin_lock_init(&lp->lock);
1920 		INIT_LIST_HEAD(&lp->list);
1921 	}
1922 }
1923 
1924 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
1925 {
1926 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
1927 
1928 	vc->vcore_state = VCORE_PREEMPT;
1929 	vc->pcpu = smp_processor_id();
1930 	if (vc->num_threads < threads_per_subcore) {
1931 		spin_lock(&lp->lock);
1932 		list_add_tail(&vc->preempt_list, &lp->list);
1933 		spin_unlock(&lp->lock);
1934 	}
1935 
1936 	/* Start accumulating stolen time */
1937 	kvmppc_core_start_stolen(vc);
1938 }
1939 
1940 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
1941 {
1942 	struct preempted_vcore_list *lp;
1943 
1944 	kvmppc_core_end_stolen(vc);
1945 	if (!list_empty(&vc->preempt_list)) {
1946 		lp = &per_cpu(preempted_vcores, vc->pcpu);
1947 		spin_lock(&lp->lock);
1948 		list_del_init(&vc->preempt_list);
1949 		spin_unlock(&lp->lock);
1950 	}
1951 	vc->vcore_state = VCORE_INACTIVE;
1952 }
1953 
1954 /*
1955  * This stores information about the virtual cores currently
1956  * assigned to a physical core.
1957  */
1958 struct core_info {
1959 	int		n_subcores;
1960 	int		max_subcore_threads;
1961 	int		total_threads;
1962 	int		subcore_threads[MAX_SUBCORES];
1963 	struct kvm	*subcore_vm[MAX_SUBCORES];
1964 	struct list_head vcs[MAX_SUBCORES];
1965 };
1966 
1967 /*
1968  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
1969  * respectively in 2-way micro-threading (split-core) mode.
1970  */
1971 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
1972 
1973 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
1974 {
1975 	int sub;
1976 
1977 	memset(cip, 0, sizeof(*cip));
1978 	cip->n_subcores = 1;
1979 	cip->max_subcore_threads = vc->num_threads;
1980 	cip->total_threads = vc->num_threads;
1981 	cip->subcore_threads[0] = vc->num_threads;
1982 	cip->subcore_vm[0] = vc->kvm;
1983 	for (sub = 0; sub < MAX_SUBCORES; ++sub)
1984 		INIT_LIST_HEAD(&cip->vcs[sub]);
1985 	list_add_tail(&vc->preempt_list, &cip->vcs[0]);
1986 }
1987 
1988 static bool subcore_config_ok(int n_subcores, int n_threads)
1989 {
1990 	/* Can only dynamically split if unsplit to begin with */
1991 	if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
1992 		return false;
1993 	if (n_subcores > MAX_SUBCORES)
1994 		return false;
1995 	if (n_subcores > 1) {
1996 		if (!(dynamic_mt_modes & 2))
1997 			n_subcores = 4;
1998 		if (n_subcores > 2 && !(dynamic_mt_modes & 4))
1999 			return false;
2000 	}
2001 
2002 	return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2003 }
2004 
2005 static void init_master_vcore(struct kvmppc_vcore *vc)
2006 {
2007 	vc->master_vcore = vc;
2008 	vc->entry_exit_map = 0;
2009 	vc->in_guest = 0;
2010 	vc->napping_threads = 0;
2011 	vc->conferring_threads = 0;
2012 }
2013 
2014 /*
2015  * See if the existing subcores can be split into 3 (or fewer) subcores
2016  * of at most two threads each, so we can fit in another vcore.  This
2017  * assumes there are at most two subcores and at most 6 threads in total.
2018  */
2019 static bool can_split_piggybacked_subcores(struct core_info *cip)
2020 {
2021 	int sub, new_sub;
2022 	int large_sub = -1;
2023 	int thr;
2024 	int n_subcores = cip->n_subcores;
2025 	struct kvmppc_vcore *vc, *vcnext;
2026 	struct kvmppc_vcore *master_vc = NULL;
2027 
2028 	for (sub = 0; sub < cip->n_subcores; ++sub) {
2029 		if (cip->subcore_threads[sub] <= 2)
2030 			continue;
2031 		if (large_sub >= 0)
2032 			return false;
2033 		large_sub = sub;
2034 		vc = list_first_entry(&cip->vcs[sub], struct kvmppc_vcore,
2035 				      preempt_list);
2036 		if (vc->num_threads > 2)
2037 			return false;
2038 		n_subcores += (cip->subcore_threads[sub] - 1) >> 1;
2039 	}
2040 	if (large_sub < 0 || !subcore_config_ok(n_subcores + 1, 2))
2041 		return false;
2042 
2043 	/*
2044 	 * Seems feasible, so go through and move vcores to new subcores.
2045 	 * Note that when we have two or more vcores in one subcore,
2046 	 * all those vcores must have only one thread each.
2047 	 */
2048 	new_sub = cip->n_subcores;
2049 	thr = 0;
2050 	sub = large_sub;
2051 	list_for_each_entry_safe(vc, vcnext, &cip->vcs[sub], preempt_list) {
2052 		if (thr >= 2) {
2053 			list_del(&vc->preempt_list);
2054 			list_add_tail(&vc->preempt_list, &cip->vcs[new_sub]);
2055 			/* vc->num_threads must be 1 */
2056 			if (++cip->subcore_threads[new_sub] == 1) {
2057 				cip->subcore_vm[new_sub] = vc->kvm;
2058 				init_master_vcore(vc);
2059 				master_vc = vc;
2060 				++cip->n_subcores;
2061 			} else {
2062 				vc->master_vcore = master_vc;
2063 				++new_sub;
2064 			}
2065 		}
2066 		thr += vc->num_threads;
2067 	}
2068 	cip->subcore_threads[large_sub] = 2;
2069 	cip->max_subcore_threads = 2;
2070 
2071 	return true;
2072 }
2073 
2074 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2075 {
2076 	int n_threads = vc->num_threads;
2077 	int sub;
2078 
2079 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2080 		return false;
2081 
2082 	if (n_threads < cip->max_subcore_threads)
2083 		n_threads = cip->max_subcore_threads;
2084 	if (subcore_config_ok(cip->n_subcores + 1, n_threads)) {
2085 		cip->max_subcore_threads = n_threads;
2086 	} else if (cip->n_subcores <= 2 && cip->total_threads <= 6 &&
2087 		   vc->num_threads <= 2) {
2088 		/*
2089 		 * We may be able to fit another subcore in by
2090 		 * splitting an existing subcore with 3 or 4
2091 		 * threads into two 2-thread subcores, or one
2092 		 * with 5 or 6 threads into three subcores.
2093 		 * We can only do this if those subcores have
2094 		 * piggybacked virtual cores.
2095 		 */
2096 		if (!can_split_piggybacked_subcores(cip))
2097 			return false;
2098 	} else {
2099 		return false;
2100 	}
2101 
2102 	sub = cip->n_subcores;
2103 	++cip->n_subcores;
2104 	cip->total_threads += vc->num_threads;
2105 	cip->subcore_threads[sub] = vc->num_threads;
2106 	cip->subcore_vm[sub] = vc->kvm;
2107 	init_master_vcore(vc);
2108 	list_del(&vc->preempt_list);
2109 	list_add_tail(&vc->preempt_list, &cip->vcs[sub]);
2110 
2111 	return true;
2112 }
2113 
2114 static bool can_piggyback_subcore(struct kvmppc_vcore *pvc,
2115 				  struct core_info *cip, int sub)
2116 {
2117 	struct kvmppc_vcore *vc;
2118 	int n_thr;
2119 
2120 	vc = list_first_entry(&cip->vcs[sub], struct kvmppc_vcore,
2121 			      preempt_list);
2122 
2123 	/* require same VM and same per-core reg values */
2124 	if (pvc->kvm != vc->kvm ||
2125 	    pvc->tb_offset != vc->tb_offset ||
2126 	    pvc->pcr != vc->pcr ||
2127 	    pvc->lpcr != vc->lpcr)
2128 		return false;
2129 
2130 	/* P8 guest with > 1 thread per core would see wrong TIR value */
2131 	if (cpu_has_feature(CPU_FTR_ARCH_207S) &&
2132 	    (vc->num_threads > 1 || pvc->num_threads > 1))
2133 		return false;
2134 
2135 	n_thr = cip->subcore_threads[sub] + pvc->num_threads;
2136 	if (n_thr > cip->max_subcore_threads) {
2137 		if (!subcore_config_ok(cip->n_subcores, n_thr))
2138 			return false;
2139 		cip->max_subcore_threads = n_thr;
2140 	}
2141 
2142 	cip->total_threads += pvc->num_threads;
2143 	cip->subcore_threads[sub] = n_thr;
2144 	pvc->master_vcore = vc;
2145 	list_del(&pvc->preempt_list);
2146 	list_add_tail(&pvc->preempt_list, &cip->vcs[sub]);
2147 
2148 	return true;
2149 }
2150 
2151 /*
2152  * Work out whether it is possible to piggyback the execution of
2153  * vcore *pvc onto the execution of the other vcores described in *cip.
2154  */
2155 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2156 			  int target_threads)
2157 {
2158 	int sub;
2159 
2160 	if (cip->total_threads + pvc->num_threads > target_threads)
2161 		return false;
2162 	for (sub = 0; sub < cip->n_subcores; ++sub)
2163 		if (cip->subcore_threads[sub] &&
2164 		    can_piggyback_subcore(pvc, cip, sub))
2165 			return true;
2166 
2167 	if (can_dynamic_split(pvc, cip))
2168 		return true;
2169 
2170 	return false;
2171 }
2172 
2173 static void prepare_threads(struct kvmppc_vcore *vc)
2174 {
2175 	struct kvm_vcpu *vcpu, *vnext;
2176 
2177 	list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2178 				 arch.run_list) {
2179 		if (signal_pending(vcpu->arch.run_task))
2180 			vcpu->arch.ret = -EINTR;
2181 		else if (vcpu->arch.vpa.update_pending ||
2182 			 vcpu->arch.slb_shadow.update_pending ||
2183 			 vcpu->arch.dtl.update_pending)
2184 			vcpu->arch.ret = RESUME_GUEST;
2185 		else
2186 			continue;
2187 		kvmppc_remove_runnable(vc, vcpu);
2188 		wake_up(&vcpu->arch.cpu_run);
2189 	}
2190 }
2191 
2192 static void collect_piggybacks(struct core_info *cip, int target_threads)
2193 {
2194 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2195 	struct kvmppc_vcore *pvc, *vcnext;
2196 
2197 	spin_lock(&lp->lock);
2198 	list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2199 		if (!spin_trylock(&pvc->lock))
2200 			continue;
2201 		prepare_threads(pvc);
2202 		if (!pvc->n_runnable) {
2203 			list_del_init(&pvc->preempt_list);
2204 			if (pvc->runner == NULL) {
2205 				pvc->vcore_state = VCORE_INACTIVE;
2206 				kvmppc_core_end_stolen(pvc);
2207 			}
2208 			spin_unlock(&pvc->lock);
2209 			continue;
2210 		}
2211 		if (!can_piggyback(pvc, cip, target_threads)) {
2212 			spin_unlock(&pvc->lock);
2213 			continue;
2214 		}
2215 		kvmppc_core_end_stolen(pvc);
2216 		pvc->vcore_state = VCORE_PIGGYBACK;
2217 		if (cip->total_threads >= target_threads)
2218 			break;
2219 	}
2220 	spin_unlock(&lp->lock);
2221 }
2222 
2223 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2224 {
2225 	int still_running = 0;
2226 	u64 now;
2227 	long ret;
2228 	struct kvm_vcpu *vcpu, *vnext;
2229 
2230 	spin_lock(&vc->lock);
2231 	now = get_tb();
2232 	list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2233 				 arch.run_list) {
2234 		/* cancel pending dec exception if dec is positive */
2235 		if (now < vcpu->arch.dec_expires &&
2236 		    kvmppc_core_pending_dec(vcpu))
2237 			kvmppc_core_dequeue_dec(vcpu);
2238 
2239 		trace_kvm_guest_exit(vcpu);
2240 
2241 		ret = RESUME_GUEST;
2242 		if (vcpu->arch.trap)
2243 			ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2244 						    vcpu->arch.run_task);
2245 
2246 		vcpu->arch.ret = ret;
2247 		vcpu->arch.trap = 0;
2248 
2249 		if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2250 			if (vcpu->arch.pending_exceptions)
2251 				kvmppc_core_prepare_to_enter(vcpu);
2252 			if (vcpu->arch.ceded)
2253 				kvmppc_set_timer(vcpu);
2254 			else
2255 				++still_running;
2256 		} else {
2257 			kvmppc_remove_runnable(vc, vcpu);
2258 			wake_up(&vcpu->arch.cpu_run);
2259 		}
2260 	}
2261 	list_del_init(&vc->preempt_list);
2262 	if (!is_master) {
2263 		if (still_running > 0) {
2264 			kvmppc_vcore_preempt(vc);
2265 		} else if (vc->runner) {
2266 			vc->vcore_state = VCORE_PREEMPT;
2267 			kvmppc_core_start_stolen(vc);
2268 		} else {
2269 			vc->vcore_state = VCORE_INACTIVE;
2270 		}
2271 		if (vc->n_runnable > 0 && vc->runner == NULL) {
2272 			/* make sure there's a candidate runner awake */
2273 			vcpu = list_first_entry(&vc->runnable_threads,
2274 						struct kvm_vcpu, arch.run_list);
2275 			wake_up(&vcpu->arch.cpu_run);
2276 		}
2277 	}
2278 	spin_unlock(&vc->lock);
2279 }
2280 
2281 /*
2282  * Run a set of guest threads on a physical core.
2283  * Called with vc->lock held.
2284  */
2285 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2286 {
2287 	struct kvm_vcpu *vcpu, *vnext;
2288 	int i;
2289 	int srcu_idx;
2290 	struct core_info core_info;
2291 	struct kvmppc_vcore *pvc, *vcnext;
2292 	struct kvm_split_mode split_info, *sip;
2293 	int split, subcore_size, active;
2294 	int sub;
2295 	bool thr0_done;
2296 	unsigned long cmd_bit, stat_bit;
2297 	int pcpu, thr;
2298 	int target_threads;
2299 
2300 	/*
2301 	 * Remove from the list any threads that have a signal pending
2302 	 * or need a VPA update done
2303 	 */
2304 	prepare_threads(vc);
2305 
2306 	/* if the runner is no longer runnable, let the caller pick a new one */
2307 	if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2308 		return;
2309 
2310 	/*
2311 	 * Initialize *vc.
2312 	 */
2313 	init_master_vcore(vc);
2314 	vc->preempt_tb = TB_NIL;
2315 
2316 	/*
2317 	 * Make sure we are running on primary threads, and that secondary
2318 	 * threads are offline.  Also check if the number of threads in this
2319 	 * guest are greater than the current system threads per guest.
2320 	 */
2321 	if ((threads_per_core > 1) &&
2322 	    ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
2323 		list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
2324 					 arch.run_list) {
2325 			vcpu->arch.ret = -EBUSY;
2326 			kvmppc_remove_runnable(vc, vcpu);
2327 			wake_up(&vcpu->arch.cpu_run);
2328 		}
2329 		goto out;
2330 	}
2331 
2332 	/*
2333 	 * See if we could run any other vcores on the physical core
2334 	 * along with this one.
2335 	 */
2336 	init_core_info(&core_info, vc);
2337 	pcpu = smp_processor_id();
2338 	target_threads = threads_per_subcore;
2339 	if (target_smt_mode && target_smt_mode < target_threads)
2340 		target_threads = target_smt_mode;
2341 	if (vc->num_threads < target_threads)
2342 		collect_piggybacks(&core_info, target_threads);
2343 
2344 	/* Decide on micro-threading (split-core) mode */
2345 	subcore_size = threads_per_subcore;
2346 	cmd_bit = stat_bit = 0;
2347 	split = core_info.n_subcores;
2348 	sip = NULL;
2349 	if (split > 1) {
2350 		/* threads_per_subcore must be MAX_SMT_THREADS (8) here */
2351 		if (split == 2 && (dynamic_mt_modes & 2)) {
2352 			cmd_bit = HID0_POWER8_1TO2LPAR;
2353 			stat_bit = HID0_POWER8_2LPARMODE;
2354 		} else {
2355 			split = 4;
2356 			cmd_bit = HID0_POWER8_1TO4LPAR;
2357 			stat_bit = HID0_POWER8_4LPARMODE;
2358 		}
2359 		subcore_size = MAX_SMT_THREADS / split;
2360 		sip = &split_info;
2361 		memset(&split_info, 0, sizeof(split_info));
2362 		split_info.rpr = mfspr(SPRN_RPR);
2363 		split_info.pmmar = mfspr(SPRN_PMMAR);
2364 		split_info.ldbar = mfspr(SPRN_LDBAR);
2365 		split_info.subcore_size = subcore_size;
2366 		for (sub = 0; sub < core_info.n_subcores; ++sub)
2367 			split_info.master_vcs[sub] =
2368 				list_first_entry(&core_info.vcs[sub],
2369 					struct kvmppc_vcore, preempt_list);
2370 		/* order writes to split_info before kvm_split_mode pointer */
2371 		smp_wmb();
2372 	}
2373 	pcpu = smp_processor_id();
2374 	for (thr = 0; thr < threads_per_subcore; ++thr)
2375 		paca[pcpu + thr].kvm_hstate.kvm_split_mode = sip;
2376 
2377 	/* Initiate micro-threading (split-core) if required */
2378 	if (cmd_bit) {
2379 		unsigned long hid0 = mfspr(SPRN_HID0);
2380 
2381 		hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2382 		mb();
2383 		mtspr(SPRN_HID0, hid0);
2384 		isync();
2385 		for (;;) {
2386 			hid0 = mfspr(SPRN_HID0);
2387 			if (hid0 & stat_bit)
2388 				break;
2389 			cpu_relax();
2390 		}
2391 	}
2392 
2393 	/* Start all the threads */
2394 	active = 0;
2395 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
2396 		thr = subcore_thread_map[sub];
2397 		thr0_done = false;
2398 		active |= 1 << thr;
2399 		list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list) {
2400 			pvc->pcpu = pcpu + thr;
2401 			list_for_each_entry(vcpu, &pvc->runnable_threads,
2402 					    arch.run_list) {
2403 				kvmppc_start_thread(vcpu, pvc);
2404 				kvmppc_create_dtl_entry(vcpu, pvc);
2405 				trace_kvm_guest_enter(vcpu);
2406 				if (!vcpu->arch.ptid)
2407 					thr0_done = true;
2408 				active |= 1 << (thr + vcpu->arch.ptid);
2409 			}
2410 			/*
2411 			 * We need to start the first thread of each subcore
2412 			 * even if it doesn't have a vcpu.
2413 			 */
2414 			if (pvc->master_vcore == pvc && !thr0_done)
2415 				kvmppc_start_thread(NULL, pvc);
2416 			thr += pvc->num_threads;
2417 		}
2418 	}
2419 
2420 	/*
2421 	 * Ensure that split_info.do_nap is set after setting
2422 	 * the vcore pointer in the PACA of the secondaries.
2423 	 */
2424 	smp_mb();
2425 	if (cmd_bit)
2426 		split_info.do_nap = 1;	/* ask secondaries to nap when done */
2427 
2428 	/*
2429 	 * When doing micro-threading, poke the inactive threads as well.
2430 	 * This gets them to the nap instruction after kvm_do_nap,
2431 	 * which reduces the time taken to unsplit later.
2432 	 */
2433 	if (split > 1)
2434 		for (thr = 1; thr < threads_per_subcore; ++thr)
2435 			if (!(active & (1 << thr)))
2436 				kvmppc_ipi_thread(pcpu + thr);
2437 
2438 	vc->vcore_state = VCORE_RUNNING;
2439 	preempt_disable();
2440 
2441 	trace_kvmppc_run_core(vc, 0);
2442 
2443 	for (sub = 0; sub < core_info.n_subcores; ++sub)
2444 		list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list)
2445 			spin_unlock(&pvc->lock);
2446 
2447 	kvm_guest_enter();
2448 
2449 	srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2450 
2451 	__kvmppc_vcore_entry();
2452 
2453 	srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2454 
2455 	spin_lock(&vc->lock);
2456 	/* prevent other vcpu threads from doing kvmppc_start_thread() now */
2457 	vc->vcore_state = VCORE_EXITING;
2458 
2459 	/* wait for secondary threads to finish writing their state to memory */
2460 	kvmppc_wait_for_nap();
2461 
2462 	/* Return to whole-core mode if we split the core earlier */
2463 	if (split > 1) {
2464 		unsigned long hid0 = mfspr(SPRN_HID0);
2465 		unsigned long loops = 0;
2466 
2467 		hid0 &= ~HID0_POWER8_DYNLPARDIS;
2468 		stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
2469 		mb();
2470 		mtspr(SPRN_HID0, hid0);
2471 		isync();
2472 		for (;;) {
2473 			hid0 = mfspr(SPRN_HID0);
2474 			if (!(hid0 & stat_bit))
2475 				break;
2476 			cpu_relax();
2477 			++loops;
2478 		}
2479 		split_info.do_nap = 0;
2480 	}
2481 
2482 	/* Let secondaries go back to the offline loop */
2483 	for (i = 0; i < threads_per_subcore; ++i) {
2484 		kvmppc_release_hwthread(pcpu + i);
2485 		if (sip && sip->napped[i])
2486 			kvmppc_ipi_thread(pcpu + i);
2487 	}
2488 
2489 	spin_unlock(&vc->lock);
2490 
2491 	/* make sure updates to secondary vcpu structs are visible now */
2492 	smp_mb();
2493 	kvm_guest_exit();
2494 
2495 	for (sub = 0; sub < core_info.n_subcores; ++sub)
2496 		list_for_each_entry_safe(pvc, vcnext, &core_info.vcs[sub],
2497 					 preempt_list)
2498 			post_guest_process(pvc, pvc == vc);
2499 
2500 	spin_lock(&vc->lock);
2501 	preempt_enable();
2502 
2503  out:
2504 	vc->vcore_state = VCORE_INACTIVE;
2505 	trace_kvmppc_run_core(vc, 1);
2506 }
2507 
2508 /*
2509  * Wait for some other vcpu thread to execute us, and
2510  * wake us up when we need to handle something in the host.
2511  */
2512 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
2513 				 struct kvm_vcpu *vcpu, int wait_state)
2514 {
2515 	DEFINE_WAIT(wait);
2516 
2517 	prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
2518 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2519 		spin_unlock(&vc->lock);
2520 		schedule();
2521 		spin_lock(&vc->lock);
2522 	}
2523 	finish_wait(&vcpu->arch.cpu_run, &wait);
2524 }
2525 
2526 /*
2527  * All the vcpus in this vcore are idle, so wait for a decrementer
2528  * or external interrupt to one of the vcpus.  vc->lock is held.
2529  */
2530 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
2531 {
2532 	struct kvm_vcpu *vcpu;
2533 	int do_sleep = 1;
2534 
2535 	DEFINE_WAIT(wait);
2536 
2537 	prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
2538 
2539 	/*
2540 	 * Check one last time for pending exceptions and ceded state after
2541 	 * we put ourselves on the wait queue
2542 	 */
2543 	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
2544 		if (vcpu->arch.pending_exceptions || !vcpu->arch.ceded) {
2545 			do_sleep = 0;
2546 			break;
2547 		}
2548 	}
2549 
2550 	if (!do_sleep) {
2551 		finish_wait(&vc->wq, &wait);
2552 		return;
2553 	}
2554 
2555 	vc->vcore_state = VCORE_SLEEPING;
2556 	trace_kvmppc_vcore_blocked(vc, 0);
2557 	spin_unlock(&vc->lock);
2558 	schedule();
2559 	finish_wait(&vc->wq, &wait);
2560 	spin_lock(&vc->lock);
2561 	vc->vcore_state = VCORE_INACTIVE;
2562 	trace_kvmppc_vcore_blocked(vc, 1);
2563 }
2564 
2565 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2566 {
2567 	int n_ceded;
2568 	struct kvmppc_vcore *vc;
2569 	struct kvm_vcpu *v, *vn;
2570 
2571 	trace_kvmppc_run_vcpu_enter(vcpu);
2572 
2573 	kvm_run->exit_reason = 0;
2574 	vcpu->arch.ret = RESUME_GUEST;
2575 	vcpu->arch.trap = 0;
2576 	kvmppc_update_vpas(vcpu);
2577 
2578 	/*
2579 	 * Synchronize with other threads in this virtual core
2580 	 */
2581 	vc = vcpu->arch.vcore;
2582 	spin_lock(&vc->lock);
2583 	vcpu->arch.ceded = 0;
2584 	vcpu->arch.run_task = current;
2585 	vcpu->arch.kvm_run = kvm_run;
2586 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
2587 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
2588 	vcpu->arch.busy_preempt = TB_NIL;
2589 	list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
2590 	++vc->n_runnable;
2591 
2592 	/*
2593 	 * This happens the first time this is called for a vcpu.
2594 	 * If the vcore is already running, we may be able to start
2595 	 * this thread straight away and have it join in.
2596 	 */
2597 	if (!signal_pending(current)) {
2598 		if (vc->vcore_state == VCORE_PIGGYBACK) {
2599 			struct kvmppc_vcore *mvc = vc->master_vcore;
2600 			if (spin_trylock(&mvc->lock)) {
2601 				if (mvc->vcore_state == VCORE_RUNNING &&
2602 				    !VCORE_IS_EXITING(mvc)) {
2603 					kvmppc_create_dtl_entry(vcpu, vc);
2604 					kvmppc_start_thread(vcpu, vc);
2605 					trace_kvm_guest_enter(vcpu);
2606 				}
2607 				spin_unlock(&mvc->lock);
2608 			}
2609 		} else if (vc->vcore_state == VCORE_RUNNING &&
2610 			   !VCORE_IS_EXITING(vc)) {
2611 			kvmppc_create_dtl_entry(vcpu, vc);
2612 			kvmppc_start_thread(vcpu, vc);
2613 			trace_kvm_guest_enter(vcpu);
2614 		} else if (vc->vcore_state == VCORE_SLEEPING) {
2615 			wake_up(&vc->wq);
2616 		}
2617 
2618 	}
2619 
2620 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2621 	       !signal_pending(current)) {
2622 		if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2623 			kvmppc_vcore_end_preempt(vc);
2624 
2625 		if (vc->vcore_state != VCORE_INACTIVE) {
2626 			kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
2627 			continue;
2628 		}
2629 		list_for_each_entry_safe(v, vn, &vc->runnable_threads,
2630 					 arch.run_list) {
2631 			kvmppc_core_prepare_to_enter(v);
2632 			if (signal_pending(v->arch.run_task)) {
2633 				kvmppc_remove_runnable(vc, v);
2634 				v->stat.signal_exits++;
2635 				v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
2636 				v->arch.ret = -EINTR;
2637 				wake_up(&v->arch.cpu_run);
2638 			}
2639 		}
2640 		if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2641 			break;
2642 		n_ceded = 0;
2643 		list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
2644 			if (!v->arch.pending_exceptions)
2645 				n_ceded += v->arch.ceded;
2646 			else
2647 				v->arch.ceded = 0;
2648 		}
2649 		vc->runner = vcpu;
2650 		if (n_ceded == vc->n_runnable) {
2651 			kvmppc_vcore_blocked(vc);
2652 		} else if (need_resched()) {
2653 			kvmppc_vcore_preempt(vc);
2654 			/* Let something else run */
2655 			cond_resched_lock(&vc->lock);
2656 			if (vc->vcore_state == VCORE_PREEMPT)
2657 				kvmppc_vcore_end_preempt(vc);
2658 		} else {
2659 			kvmppc_run_core(vc);
2660 		}
2661 		vc->runner = NULL;
2662 	}
2663 
2664 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2665 	       (vc->vcore_state == VCORE_RUNNING ||
2666 		vc->vcore_state == VCORE_EXITING ||
2667 		vc->vcore_state == VCORE_PIGGYBACK))
2668 		kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
2669 
2670 	if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2671 		kvmppc_vcore_end_preempt(vc);
2672 
2673 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2674 		kvmppc_remove_runnable(vc, vcpu);
2675 		vcpu->stat.signal_exits++;
2676 		kvm_run->exit_reason = KVM_EXIT_INTR;
2677 		vcpu->arch.ret = -EINTR;
2678 	}
2679 
2680 	if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
2681 		/* Wake up some vcpu to run the core */
2682 		v = list_first_entry(&vc->runnable_threads,
2683 				     struct kvm_vcpu, arch.run_list);
2684 		wake_up(&v->arch.cpu_run);
2685 	}
2686 
2687 	trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
2688 	spin_unlock(&vc->lock);
2689 	return vcpu->arch.ret;
2690 }
2691 
2692 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
2693 {
2694 	int r;
2695 	int srcu_idx;
2696 
2697 	if (!vcpu->arch.sane) {
2698 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2699 		return -EINVAL;
2700 	}
2701 
2702 	kvmppc_core_prepare_to_enter(vcpu);
2703 
2704 	/* No need to go into the guest when all we'll do is come back out */
2705 	if (signal_pending(current)) {
2706 		run->exit_reason = KVM_EXIT_INTR;
2707 		return -EINTR;
2708 	}
2709 
2710 	atomic_inc(&vcpu->kvm->arch.vcpus_running);
2711 	/* Order vcpus_running vs. hpte_setup_done, see kvmppc_alloc_reset_hpt */
2712 	smp_mb();
2713 
2714 	/* On the first time here, set up HTAB and VRMA */
2715 	if (!vcpu->kvm->arch.hpte_setup_done) {
2716 		r = kvmppc_hv_setup_htab_rma(vcpu);
2717 		if (r)
2718 			goto out;
2719 	}
2720 
2721 	flush_all_to_thread(current);
2722 
2723 	vcpu->arch.wqp = &vcpu->arch.vcore->wq;
2724 	vcpu->arch.pgdir = current->mm->pgd;
2725 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2726 
2727 	do {
2728 		r = kvmppc_run_vcpu(run, vcpu);
2729 
2730 		if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
2731 		    !(vcpu->arch.shregs.msr & MSR_PR)) {
2732 			trace_kvm_hcall_enter(vcpu);
2733 			r = kvmppc_pseries_do_hcall(vcpu);
2734 			trace_kvm_hcall_exit(vcpu, r);
2735 			kvmppc_core_prepare_to_enter(vcpu);
2736 		} else if (r == RESUME_PAGE_FAULT) {
2737 			srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2738 			r = kvmppc_book3s_hv_page_fault(run, vcpu,
2739 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
2740 			srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2741 		}
2742 	} while (is_kvmppc_resume_guest(r));
2743 
2744  out:
2745 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2746 	atomic_dec(&vcpu->kvm->arch.vcpus_running);
2747 	return r;
2748 }
2749 
2750 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
2751 				     int linux_psize)
2752 {
2753 	struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
2754 
2755 	if (!def->shift)
2756 		return;
2757 	(*sps)->page_shift = def->shift;
2758 	(*sps)->slb_enc = def->sllp;
2759 	(*sps)->enc[0].page_shift = def->shift;
2760 	(*sps)->enc[0].pte_enc = def->penc[linux_psize];
2761 	/*
2762 	 * Add 16MB MPSS support if host supports it
2763 	 */
2764 	if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
2765 		(*sps)->enc[1].page_shift = 24;
2766 		(*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
2767 	}
2768 	(*sps)++;
2769 }
2770 
2771 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
2772 					 struct kvm_ppc_smmu_info *info)
2773 {
2774 	struct kvm_ppc_one_seg_page_size *sps;
2775 
2776 	info->flags = KVM_PPC_PAGE_SIZES_REAL;
2777 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
2778 		info->flags |= KVM_PPC_1T_SEGMENTS;
2779 	info->slb_size = mmu_slb_size;
2780 
2781 	/* We only support these sizes for now, and no muti-size segments */
2782 	sps = &info->sps[0];
2783 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
2784 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
2785 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
2786 
2787 	return 0;
2788 }
2789 
2790 /*
2791  * Get (and clear) the dirty memory log for a memory slot.
2792  */
2793 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
2794 					 struct kvm_dirty_log *log)
2795 {
2796 	struct kvm_memslots *slots;
2797 	struct kvm_memory_slot *memslot;
2798 	int r;
2799 	unsigned long n;
2800 
2801 	mutex_lock(&kvm->slots_lock);
2802 
2803 	r = -EINVAL;
2804 	if (log->slot >= KVM_USER_MEM_SLOTS)
2805 		goto out;
2806 
2807 	slots = kvm_memslots(kvm);
2808 	memslot = id_to_memslot(slots, log->slot);
2809 	r = -ENOENT;
2810 	if (!memslot->dirty_bitmap)
2811 		goto out;
2812 
2813 	n = kvm_dirty_bitmap_bytes(memslot);
2814 	memset(memslot->dirty_bitmap, 0, n);
2815 
2816 	r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
2817 	if (r)
2818 		goto out;
2819 
2820 	r = -EFAULT;
2821 	if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
2822 		goto out;
2823 
2824 	r = 0;
2825 out:
2826 	mutex_unlock(&kvm->slots_lock);
2827 	return r;
2828 }
2829 
2830 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
2831 					struct kvm_memory_slot *dont)
2832 {
2833 	if (!dont || free->arch.rmap != dont->arch.rmap) {
2834 		vfree(free->arch.rmap);
2835 		free->arch.rmap = NULL;
2836 	}
2837 }
2838 
2839 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
2840 					 unsigned long npages)
2841 {
2842 	slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
2843 	if (!slot->arch.rmap)
2844 		return -ENOMEM;
2845 
2846 	return 0;
2847 }
2848 
2849 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
2850 					struct kvm_memory_slot *memslot,
2851 					const struct kvm_userspace_memory_region *mem)
2852 {
2853 	return 0;
2854 }
2855 
2856 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
2857 				const struct kvm_userspace_memory_region *mem,
2858 				const struct kvm_memory_slot *old,
2859 				const struct kvm_memory_slot *new)
2860 {
2861 	unsigned long npages = mem->memory_size >> PAGE_SHIFT;
2862 	struct kvm_memslots *slots;
2863 	struct kvm_memory_slot *memslot;
2864 
2865 	if (npages && old->npages) {
2866 		/*
2867 		 * If modifying a memslot, reset all the rmap dirty bits.
2868 		 * If this is a new memslot, we don't need to do anything
2869 		 * since the rmap array starts out as all zeroes,
2870 		 * i.e. no pages are dirty.
2871 		 */
2872 		slots = kvm_memslots(kvm);
2873 		memslot = id_to_memslot(slots, mem->slot);
2874 		kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
2875 	}
2876 }
2877 
2878 /*
2879  * Update LPCR values in kvm->arch and in vcores.
2880  * Caller must hold kvm->lock.
2881  */
2882 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
2883 {
2884 	long int i;
2885 	u32 cores_done = 0;
2886 
2887 	if ((kvm->arch.lpcr & mask) == lpcr)
2888 		return;
2889 
2890 	kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
2891 
2892 	for (i = 0; i < KVM_MAX_VCORES; ++i) {
2893 		struct kvmppc_vcore *vc = kvm->arch.vcores[i];
2894 		if (!vc)
2895 			continue;
2896 		spin_lock(&vc->lock);
2897 		vc->lpcr = (vc->lpcr & ~mask) | lpcr;
2898 		spin_unlock(&vc->lock);
2899 		if (++cores_done >= kvm->arch.online_vcores)
2900 			break;
2901 	}
2902 }
2903 
2904 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
2905 {
2906 	return;
2907 }
2908 
2909 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
2910 {
2911 	int err = 0;
2912 	struct kvm *kvm = vcpu->kvm;
2913 	unsigned long hva;
2914 	struct kvm_memory_slot *memslot;
2915 	struct vm_area_struct *vma;
2916 	unsigned long lpcr = 0, senc;
2917 	unsigned long psize, porder;
2918 	int srcu_idx;
2919 
2920 	mutex_lock(&kvm->lock);
2921 	if (kvm->arch.hpte_setup_done)
2922 		goto out;	/* another vcpu beat us to it */
2923 
2924 	/* Allocate hashed page table (if not done already) and reset it */
2925 	if (!kvm->arch.hpt_virt) {
2926 		err = kvmppc_alloc_hpt(kvm, NULL);
2927 		if (err) {
2928 			pr_err("KVM: Couldn't alloc HPT\n");
2929 			goto out;
2930 		}
2931 	}
2932 
2933 	/* Look up the memslot for guest physical address 0 */
2934 	srcu_idx = srcu_read_lock(&kvm->srcu);
2935 	memslot = gfn_to_memslot(kvm, 0);
2936 
2937 	/* We must have some memory at 0 by now */
2938 	err = -EINVAL;
2939 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
2940 		goto out_srcu;
2941 
2942 	/* Look up the VMA for the start of this memory slot */
2943 	hva = memslot->userspace_addr;
2944 	down_read(&current->mm->mmap_sem);
2945 	vma = find_vma(current->mm, hva);
2946 	if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
2947 		goto up_out;
2948 
2949 	psize = vma_kernel_pagesize(vma);
2950 	porder = __ilog2(psize);
2951 
2952 	up_read(&current->mm->mmap_sem);
2953 
2954 	/* We can handle 4k, 64k or 16M pages in the VRMA */
2955 	err = -EINVAL;
2956 	if (!(psize == 0x1000 || psize == 0x10000 ||
2957 	      psize == 0x1000000))
2958 		goto out_srcu;
2959 
2960 	/* Update VRMASD field in the LPCR */
2961 	senc = slb_pgsize_encoding(psize);
2962 	kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
2963 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
2964 	/* the -4 is to account for senc values starting at 0x10 */
2965 	lpcr = senc << (LPCR_VRMASD_SH - 4);
2966 
2967 	/* Create HPTEs in the hash page table for the VRMA */
2968 	kvmppc_map_vrma(vcpu, memslot, porder);
2969 
2970 	kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
2971 
2972 	/* Order updates to kvm->arch.lpcr etc. vs. hpte_setup_done */
2973 	smp_wmb();
2974 	kvm->arch.hpte_setup_done = 1;
2975 	err = 0;
2976  out_srcu:
2977 	srcu_read_unlock(&kvm->srcu, srcu_idx);
2978  out:
2979 	mutex_unlock(&kvm->lock);
2980 	return err;
2981 
2982  up_out:
2983 	up_read(&current->mm->mmap_sem);
2984 	goto out_srcu;
2985 }
2986 
2987 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
2988 {
2989 	unsigned long lpcr, lpid;
2990 	char buf[32];
2991 
2992 	/* Allocate the guest's logical partition ID */
2993 
2994 	lpid = kvmppc_alloc_lpid();
2995 	if ((long)lpid < 0)
2996 		return -ENOMEM;
2997 	kvm->arch.lpid = lpid;
2998 
2999 	/*
3000 	 * Since we don't flush the TLB when tearing down a VM,
3001 	 * and this lpid might have previously been used,
3002 	 * make sure we flush on each core before running the new VM.
3003 	 */
3004 	cpumask_setall(&kvm->arch.need_tlb_flush);
3005 
3006 	/* Start out with the default set of hcalls enabled */
3007 	memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
3008 	       sizeof(kvm->arch.enabled_hcalls));
3009 
3010 	kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
3011 
3012 	/* Init LPCR for virtual RMA mode */
3013 	kvm->arch.host_lpid = mfspr(SPRN_LPID);
3014 	kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
3015 	lpcr &= LPCR_PECE | LPCR_LPES;
3016 	lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3017 		LPCR_VPM0 | LPCR_VPM1;
3018 	kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3019 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
3020 	/* On POWER8 turn on online bit to enable PURR/SPURR */
3021 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
3022 		lpcr |= LPCR_ONL;
3023 	kvm->arch.lpcr = lpcr;
3024 
3025 	/*
3026 	 * Track that we now have a HV mode VM active. This blocks secondary
3027 	 * CPU threads from coming online.
3028 	 */
3029 	kvm_hv_vm_activated();
3030 
3031 	/*
3032 	 * Create a debugfs directory for the VM
3033 	 */
3034 	snprintf(buf, sizeof(buf), "vm%d", current->pid);
3035 	kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
3036 	if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
3037 		kvmppc_mmu_debugfs_init(kvm);
3038 
3039 	return 0;
3040 }
3041 
3042 static void kvmppc_free_vcores(struct kvm *kvm)
3043 {
3044 	long int i;
3045 
3046 	for (i = 0; i < KVM_MAX_VCORES; ++i)
3047 		kfree(kvm->arch.vcores[i]);
3048 	kvm->arch.online_vcores = 0;
3049 }
3050 
3051 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
3052 {
3053 	debugfs_remove_recursive(kvm->arch.debugfs_dir);
3054 
3055 	kvm_hv_vm_deactivated();
3056 
3057 	kvmppc_free_vcores(kvm);
3058 
3059 	kvmppc_free_hpt(kvm);
3060 }
3061 
3062 /* We don't need to emulate any privileged instructions or dcbz */
3063 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
3064 				     unsigned int inst, int *advance)
3065 {
3066 	return EMULATE_FAIL;
3067 }
3068 
3069 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
3070 					ulong spr_val)
3071 {
3072 	return EMULATE_FAIL;
3073 }
3074 
3075 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
3076 					ulong *spr_val)
3077 {
3078 	return EMULATE_FAIL;
3079 }
3080 
3081 static int kvmppc_core_check_processor_compat_hv(void)
3082 {
3083 	if (!cpu_has_feature(CPU_FTR_HVMODE) ||
3084 	    !cpu_has_feature(CPU_FTR_ARCH_206))
3085 		return -EIO;
3086 	return 0;
3087 }
3088 
3089 static long kvm_arch_vm_ioctl_hv(struct file *filp,
3090 				 unsigned int ioctl, unsigned long arg)
3091 {
3092 	struct kvm *kvm __maybe_unused = filp->private_data;
3093 	void __user *argp = (void __user *)arg;
3094 	long r;
3095 
3096 	switch (ioctl) {
3097 
3098 	case KVM_PPC_ALLOCATE_HTAB: {
3099 		u32 htab_order;
3100 
3101 		r = -EFAULT;
3102 		if (get_user(htab_order, (u32 __user *)argp))
3103 			break;
3104 		r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
3105 		if (r)
3106 			break;
3107 		r = -EFAULT;
3108 		if (put_user(htab_order, (u32 __user *)argp))
3109 			break;
3110 		r = 0;
3111 		break;
3112 	}
3113 
3114 	case KVM_PPC_GET_HTAB_FD: {
3115 		struct kvm_get_htab_fd ghf;
3116 
3117 		r = -EFAULT;
3118 		if (copy_from_user(&ghf, argp, sizeof(ghf)))
3119 			break;
3120 		r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
3121 		break;
3122 	}
3123 
3124 	default:
3125 		r = -ENOTTY;
3126 	}
3127 
3128 	return r;
3129 }
3130 
3131 /*
3132  * List of hcall numbers to enable by default.
3133  * For compatibility with old userspace, we enable by default
3134  * all hcalls that were implemented before the hcall-enabling
3135  * facility was added.  Note this list should not include H_RTAS.
3136  */
3137 static unsigned int default_hcall_list[] = {
3138 	H_REMOVE,
3139 	H_ENTER,
3140 	H_READ,
3141 	H_PROTECT,
3142 	H_BULK_REMOVE,
3143 	H_GET_TCE,
3144 	H_PUT_TCE,
3145 	H_SET_DABR,
3146 	H_SET_XDABR,
3147 	H_CEDE,
3148 	H_PROD,
3149 	H_CONFER,
3150 	H_REGISTER_VPA,
3151 #ifdef CONFIG_KVM_XICS
3152 	H_EOI,
3153 	H_CPPR,
3154 	H_IPI,
3155 	H_IPOLL,
3156 	H_XIRR,
3157 	H_XIRR_X,
3158 #endif
3159 	0
3160 };
3161 
3162 static void init_default_hcalls(void)
3163 {
3164 	int i;
3165 	unsigned int hcall;
3166 
3167 	for (i = 0; default_hcall_list[i]; ++i) {
3168 		hcall = default_hcall_list[i];
3169 		WARN_ON(!kvmppc_hcall_impl_hv(hcall));
3170 		__set_bit(hcall / 4, default_enabled_hcalls);
3171 	}
3172 }
3173 
3174 static struct kvmppc_ops kvm_ops_hv = {
3175 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
3176 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
3177 	.get_one_reg = kvmppc_get_one_reg_hv,
3178 	.set_one_reg = kvmppc_set_one_reg_hv,
3179 	.vcpu_load   = kvmppc_core_vcpu_load_hv,
3180 	.vcpu_put    = kvmppc_core_vcpu_put_hv,
3181 	.set_msr     = kvmppc_set_msr_hv,
3182 	.vcpu_run    = kvmppc_vcpu_run_hv,
3183 	.vcpu_create = kvmppc_core_vcpu_create_hv,
3184 	.vcpu_free   = kvmppc_core_vcpu_free_hv,
3185 	.check_requests = kvmppc_core_check_requests_hv,
3186 	.get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
3187 	.flush_memslot  = kvmppc_core_flush_memslot_hv,
3188 	.prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
3189 	.commit_memory_region  = kvmppc_core_commit_memory_region_hv,
3190 	.unmap_hva = kvm_unmap_hva_hv,
3191 	.unmap_hva_range = kvm_unmap_hva_range_hv,
3192 	.age_hva  = kvm_age_hva_hv,
3193 	.test_age_hva = kvm_test_age_hva_hv,
3194 	.set_spte_hva = kvm_set_spte_hva_hv,
3195 	.mmu_destroy  = kvmppc_mmu_destroy_hv,
3196 	.free_memslot = kvmppc_core_free_memslot_hv,
3197 	.create_memslot = kvmppc_core_create_memslot_hv,
3198 	.init_vm =  kvmppc_core_init_vm_hv,
3199 	.destroy_vm = kvmppc_core_destroy_vm_hv,
3200 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
3201 	.emulate_op = kvmppc_core_emulate_op_hv,
3202 	.emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
3203 	.emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
3204 	.fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
3205 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
3206 	.hcall_implemented = kvmppc_hcall_impl_hv,
3207 };
3208 
3209 static int kvmppc_book3s_init_hv(void)
3210 {
3211 	int r;
3212 	/*
3213 	 * FIXME!! Do we need to check on all cpus ?
3214 	 */
3215 	r = kvmppc_core_check_processor_compat_hv();
3216 	if (r < 0)
3217 		return -ENODEV;
3218 
3219 	kvm_ops_hv.owner = THIS_MODULE;
3220 	kvmppc_hv_ops = &kvm_ops_hv;
3221 
3222 	init_default_hcalls();
3223 
3224 	init_vcore_lists();
3225 
3226 	r = kvmppc_mmu_hv_init();
3227 	return r;
3228 }
3229 
3230 static void kvmppc_book3s_exit_hv(void)
3231 {
3232 	kvmppc_hv_ops = NULL;
3233 }
3234 
3235 module_init(kvmppc_book3s_init_hv);
3236 module_exit(kvmppc_book3s_exit_hv);
3237 MODULE_LICENSE("GPL");
3238 MODULE_ALIAS_MISCDEV(KVM_MINOR);
3239 MODULE_ALIAS("devname:kvm");
3240