xref: /openbmc/linux/arch/powerpc/kvm/book3s_hv.c (revision 94cdda6b)
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/cache.h>
40 #include <asm/cacheflush.h>
41 #include <asm/tlbflush.h>
42 #include <asm/uaccess.h>
43 #include <asm/io.h>
44 #include <asm/kvm_ppc.h>
45 #include <asm/kvm_book3s.h>
46 #include <asm/mmu_context.h>
47 #include <asm/lppaca.h>
48 #include <asm/processor.h>
49 #include <asm/cputhreads.h>
50 #include <asm/page.h>
51 #include <asm/hvcall.h>
52 #include <asm/switch_to.h>
53 #include <asm/smp.h>
54 #include <asm/dbell.h>
55 #include <linux/gfp.h>
56 #include <linux/vmalloc.h>
57 #include <linux/highmem.h>
58 #include <linux/hugetlb.h>
59 #include <linux/module.h>
60 
61 #include "book3s.h"
62 
63 #define CREATE_TRACE_POINTS
64 #include "trace_hv.h"
65 
66 /* #define EXIT_DEBUG */
67 /* #define EXIT_DEBUG_SIMPLE */
68 /* #define EXIT_DEBUG_INT */
69 
70 /* Used to indicate that a guest page fault needs to be handled */
71 #define RESUME_PAGE_FAULT	(RESUME_GUEST | RESUME_FLAG_ARCH1)
72 
73 /* Used as a "null" value for timebase values */
74 #define TB_NIL	(~(u64)0)
75 
76 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
77 
78 #if defined(CONFIG_PPC_64K_PAGES)
79 #define MPP_BUFFER_ORDER	0
80 #elif defined(CONFIG_PPC_4K_PAGES)
81 #define MPP_BUFFER_ORDER	3
82 #endif
83 
84 
85 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
86 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
87 
88 static bool kvmppc_ipi_thread(int cpu)
89 {
90 	/* On POWER8 for IPIs to threads in the same core, use msgsnd */
91 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
92 		preempt_disable();
93 		if (cpu_first_thread_sibling(cpu) ==
94 		    cpu_first_thread_sibling(smp_processor_id())) {
95 			unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
96 			msg |= cpu_thread_in_core(cpu);
97 			smp_mb();
98 			__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
99 			preempt_enable();
100 			return true;
101 		}
102 		preempt_enable();
103 	}
104 
105 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
106 	if (cpu >= 0 && cpu < nr_cpu_ids && paca[cpu].kvm_hstate.xics_phys) {
107 		xics_wake_cpu(cpu);
108 		return true;
109 	}
110 #endif
111 
112 	return false;
113 }
114 
115 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
116 {
117 	int cpu = vcpu->cpu;
118 	wait_queue_head_t *wqp;
119 
120 	wqp = kvm_arch_vcpu_wq(vcpu);
121 	if (waitqueue_active(wqp)) {
122 		wake_up_interruptible(wqp);
123 		++vcpu->stat.halt_wakeup;
124 	}
125 
126 	if (kvmppc_ipi_thread(cpu + vcpu->arch.ptid))
127 		return;
128 
129 	/* CPU points to the first thread of the core */
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_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
168 {
169 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
170 	unsigned long flags;
171 
172 	/*
173 	 * We can test vc->runner without taking the vcore lock,
174 	 * because only this task ever sets vc->runner to this
175 	 * vcpu, and once it is set to this vcpu, only this task
176 	 * ever sets it to NULL.
177 	 */
178 	if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE) {
179 		spin_lock_irqsave(&vc->stoltb_lock, flags);
180 		if (vc->preempt_tb != TB_NIL) {
181 			vc->stolen_tb += mftb() - vc->preempt_tb;
182 			vc->preempt_tb = TB_NIL;
183 		}
184 		spin_unlock_irqrestore(&vc->stoltb_lock, flags);
185 	}
186 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
187 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
188 	    vcpu->arch.busy_preempt != TB_NIL) {
189 		vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
190 		vcpu->arch.busy_preempt = TB_NIL;
191 	}
192 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
193 }
194 
195 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
196 {
197 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
198 	unsigned long flags;
199 
200 	if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE) {
201 		spin_lock_irqsave(&vc->stoltb_lock, flags);
202 		vc->preempt_tb = mftb();
203 		spin_unlock_irqrestore(&vc->stoltb_lock, flags);
204 	}
205 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
206 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
207 		vcpu->arch.busy_preempt = mftb();
208 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
209 }
210 
211 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
212 {
213 	vcpu->arch.shregs.msr = msr;
214 	kvmppc_end_cede(vcpu);
215 }
216 
217 void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
218 {
219 	vcpu->arch.pvr = pvr;
220 }
221 
222 int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
223 {
224 	unsigned long pcr = 0;
225 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
226 
227 	if (arch_compat) {
228 		switch (arch_compat) {
229 		case PVR_ARCH_205:
230 			/*
231 			 * If an arch bit is set in PCR, all the defined
232 			 * higher-order arch bits also have to be set.
233 			 */
234 			pcr = PCR_ARCH_206 | PCR_ARCH_205;
235 			break;
236 		case PVR_ARCH_206:
237 		case PVR_ARCH_206p:
238 			pcr = PCR_ARCH_206;
239 			break;
240 		case PVR_ARCH_207:
241 			break;
242 		default:
243 			return -EINVAL;
244 		}
245 
246 		if (!cpu_has_feature(CPU_FTR_ARCH_207S)) {
247 			/* POWER7 can't emulate POWER8 */
248 			if (!(pcr & PCR_ARCH_206))
249 				return -EINVAL;
250 			pcr &= ~PCR_ARCH_206;
251 		}
252 	}
253 
254 	spin_lock(&vc->lock);
255 	vc->arch_compat = arch_compat;
256 	vc->pcr = pcr;
257 	spin_unlock(&vc->lock);
258 
259 	return 0;
260 }
261 
262 void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
263 {
264 	int r;
265 
266 	pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
267 	pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
268 	       vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
269 	for (r = 0; r < 16; ++r)
270 		pr_err("r%2d = %.16lx  r%d = %.16lx\n",
271 		       r, kvmppc_get_gpr(vcpu, r),
272 		       r+16, kvmppc_get_gpr(vcpu, r+16));
273 	pr_err("ctr = %.16lx  lr  = %.16lx\n",
274 	       vcpu->arch.ctr, vcpu->arch.lr);
275 	pr_err("srr0 = %.16llx srr1 = %.16llx\n",
276 	       vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
277 	pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
278 	       vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
279 	pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
280 	       vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
281 	pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
282 	       vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
283 	pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
284 	pr_err("fault dar = %.16lx dsisr = %.8x\n",
285 	       vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
286 	pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
287 	for (r = 0; r < vcpu->arch.slb_max; ++r)
288 		pr_err("  ESID = %.16llx VSID = %.16llx\n",
289 		       vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
290 	pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
291 	       vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
292 	       vcpu->arch.last_inst);
293 }
294 
295 struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
296 {
297 	int r;
298 	struct kvm_vcpu *v, *ret = NULL;
299 
300 	mutex_lock(&kvm->lock);
301 	kvm_for_each_vcpu(r, v, kvm) {
302 		if (v->vcpu_id == id) {
303 			ret = v;
304 			break;
305 		}
306 	}
307 	mutex_unlock(&kvm->lock);
308 	return ret;
309 }
310 
311 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
312 {
313 	vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
314 	vpa->yield_count = cpu_to_be32(1);
315 }
316 
317 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
318 		   unsigned long addr, unsigned long len)
319 {
320 	/* check address is cacheline aligned */
321 	if (addr & (L1_CACHE_BYTES - 1))
322 		return -EINVAL;
323 	spin_lock(&vcpu->arch.vpa_update_lock);
324 	if (v->next_gpa != addr || v->len != len) {
325 		v->next_gpa = addr;
326 		v->len = addr ? len : 0;
327 		v->update_pending = 1;
328 	}
329 	spin_unlock(&vcpu->arch.vpa_update_lock);
330 	return 0;
331 }
332 
333 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
334 struct reg_vpa {
335 	u32 dummy;
336 	union {
337 		__be16 hword;
338 		__be32 word;
339 	} length;
340 };
341 
342 static int vpa_is_registered(struct kvmppc_vpa *vpap)
343 {
344 	if (vpap->update_pending)
345 		return vpap->next_gpa != 0;
346 	return vpap->pinned_addr != NULL;
347 }
348 
349 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
350 				       unsigned long flags,
351 				       unsigned long vcpuid, unsigned long vpa)
352 {
353 	struct kvm *kvm = vcpu->kvm;
354 	unsigned long len, nb;
355 	void *va;
356 	struct kvm_vcpu *tvcpu;
357 	int err;
358 	int subfunc;
359 	struct kvmppc_vpa *vpap;
360 
361 	tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
362 	if (!tvcpu)
363 		return H_PARAMETER;
364 
365 	subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
366 	if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
367 	    subfunc == H_VPA_REG_SLB) {
368 		/* Registering new area - address must be cache-line aligned */
369 		if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
370 			return H_PARAMETER;
371 
372 		/* convert logical addr to kernel addr and read length */
373 		va = kvmppc_pin_guest_page(kvm, vpa, &nb);
374 		if (va == NULL)
375 			return H_PARAMETER;
376 		if (subfunc == H_VPA_REG_VPA)
377 			len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
378 		else
379 			len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
380 		kvmppc_unpin_guest_page(kvm, va, vpa, false);
381 
382 		/* Check length */
383 		if (len > nb || len < sizeof(struct reg_vpa))
384 			return H_PARAMETER;
385 	} else {
386 		vpa = 0;
387 		len = 0;
388 	}
389 
390 	err = H_PARAMETER;
391 	vpap = NULL;
392 	spin_lock(&tvcpu->arch.vpa_update_lock);
393 
394 	switch (subfunc) {
395 	case H_VPA_REG_VPA:		/* register VPA */
396 		if (len < sizeof(struct lppaca))
397 			break;
398 		vpap = &tvcpu->arch.vpa;
399 		err = 0;
400 		break;
401 
402 	case H_VPA_REG_DTL:		/* register DTL */
403 		if (len < sizeof(struct dtl_entry))
404 			break;
405 		len -= len % sizeof(struct dtl_entry);
406 
407 		/* Check that they have previously registered a VPA */
408 		err = H_RESOURCE;
409 		if (!vpa_is_registered(&tvcpu->arch.vpa))
410 			break;
411 
412 		vpap = &tvcpu->arch.dtl;
413 		err = 0;
414 		break;
415 
416 	case H_VPA_REG_SLB:		/* register SLB shadow buffer */
417 		/* Check that they have previously registered a VPA */
418 		err = H_RESOURCE;
419 		if (!vpa_is_registered(&tvcpu->arch.vpa))
420 			break;
421 
422 		vpap = &tvcpu->arch.slb_shadow;
423 		err = 0;
424 		break;
425 
426 	case H_VPA_DEREG_VPA:		/* deregister VPA */
427 		/* Check they don't still have a DTL or SLB buf registered */
428 		err = H_RESOURCE;
429 		if (vpa_is_registered(&tvcpu->arch.dtl) ||
430 		    vpa_is_registered(&tvcpu->arch.slb_shadow))
431 			break;
432 
433 		vpap = &tvcpu->arch.vpa;
434 		err = 0;
435 		break;
436 
437 	case H_VPA_DEREG_DTL:		/* deregister DTL */
438 		vpap = &tvcpu->arch.dtl;
439 		err = 0;
440 		break;
441 
442 	case H_VPA_DEREG_SLB:		/* deregister SLB shadow buffer */
443 		vpap = &tvcpu->arch.slb_shadow;
444 		err = 0;
445 		break;
446 	}
447 
448 	if (vpap) {
449 		vpap->next_gpa = vpa;
450 		vpap->len = len;
451 		vpap->update_pending = 1;
452 	}
453 
454 	spin_unlock(&tvcpu->arch.vpa_update_lock);
455 
456 	return err;
457 }
458 
459 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
460 {
461 	struct kvm *kvm = vcpu->kvm;
462 	void *va;
463 	unsigned long nb;
464 	unsigned long gpa;
465 
466 	/*
467 	 * We need to pin the page pointed to by vpap->next_gpa,
468 	 * but we can't call kvmppc_pin_guest_page under the lock
469 	 * as it does get_user_pages() and down_read().  So we
470 	 * have to drop the lock, pin the page, then get the lock
471 	 * again and check that a new area didn't get registered
472 	 * in the meantime.
473 	 */
474 	for (;;) {
475 		gpa = vpap->next_gpa;
476 		spin_unlock(&vcpu->arch.vpa_update_lock);
477 		va = NULL;
478 		nb = 0;
479 		if (gpa)
480 			va = kvmppc_pin_guest_page(kvm, gpa, &nb);
481 		spin_lock(&vcpu->arch.vpa_update_lock);
482 		if (gpa == vpap->next_gpa)
483 			break;
484 		/* sigh... unpin that one and try again */
485 		if (va)
486 			kvmppc_unpin_guest_page(kvm, va, gpa, false);
487 	}
488 
489 	vpap->update_pending = 0;
490 	if (va && nb < vpap->len) {
491 		/*
492 		 * If it's now too short, it must be that userspace
493 		 * has changed the mappings underlying guest memory,
494 		 * so unregister the region.
495 		 */
496 		kvmppc_unpin_guest_page(kvm, va, gpa, false);
497 		va = NULL;
498 	}
499 	if (vpap->pinned_addr)
500 		kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
501 					vpap->dirty);
502 	vpap->gpa = gpa;
503 	vpap->pinned_addr = va;
504 	vpap->dirty = false;
505 	if (va)
506 		vpap->pinned_end = va + vpap->len;
507 }
508 
509 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
510 {
511 	if (!(vcpu->arch.vpa.update_pending ||
512 	      vcpu->arch.slb_shadow.update_pending ||
513 	      vcpu->arch.dtl.update_pending))
514 		return;
515 
516 	spin_lock(&vcpu->arch.vpa_update_lock);
517 	if (vcpu->arch.vpa.update_pending) {
518 		kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
519 		if (vcpu->arch.vpa.pinned_addr)
520 			init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
521 	}
522 	if (vcpu->arch.dtl.update_pending) {
523 		kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
524 		vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
525 		vcpu->arch.dtl_index = 0;
526 	}
527 	if (vcpu->arch.slb_shadow.update_pending)
528 		kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
529 	spin_unlock(&vcpu->arch.vpa_update_lock);
530 }
531 
532 /*
533  * Return the accumulated stolen time for the vcore up until `now'.
534  * The caller should hold the vcore lock.
535  */
536 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
537 {
538 	u64 p;
539 	unsigned long flags;
540 
541 	spin_lock_irqsave(&vc->stoltb_lock, flags);
542 	p = vc->stolen_tb;
543 	if (vc->vcore_state != VCORE_INACTIVE &&
544 	    vc->preempt_tb != TB_NIL)
545 		p += now - vc->preempt_tb;
546 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
547 	return p;
548 }
549 
550 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
551 				    struct kvmppc_vcore *vc)
552 {
553 	struct dtl_entry *dt;
554 	struct lppaca *vpa;
555 	unsigned long stolen;
556 	unsigned long core_stolen;
557 	u64 now;
558 
559 	dt = vcpu->arch.dtl_ptr;
560 	vpa = vcpu->arch.vpa.pinned_addr;
561 	now = mftb();
562 	core_stolen = vcore_stolen_time(vc, now);
563 	stolen = core_stolen - vcpu->arch.stolen_logged;
564 	vcpu->arch.stolen_logged = core_stolen;
565 	spin_lock_irq(&vcpu->arch.tbacct_lock);
566 	stolen += vcpu->arch.busy_stolen;
567 	vcpu->arch.busy_stolen = 0;
568 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
569 	if (!dt || !vpa)
570 		return;
571 	memset(dt, 0, sizeof(struct dtl_entry));
572 	dt->dispatch_reason = 7;
573 	dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
574 	dt->timebase = cpu_to_be64(now + vc->tb_offset);
575 	dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
576 	dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
577 	dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
578 	++dt;
579 	if (dt == vcpu->arch.dtl.pinned_end)
580 		dt = vcpu->arch.dtl.pinned_addr;
581 	vcpu->arch.dtl_ptr = dt;
582 	/* order writing *dt vs. writing vpa->dtl_idx */
583 	smp_wmb();
584 	vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
585 	vcpu->arch.dtl.dirty = true;
586 }
587 
588 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
589 {
590 	if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
591 		return true;
592 	if ((!vcpu->arch.vcore->arch_compat) &&
593 	    cpu_has_feature(CPU_FTR_ARCH_207S))
594 		return true;
595 	return false;
596 }
597 
598 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
599 			     unsigned long resource, unsigned long value1,
600 			     unsigned long value2)
601 {
602 	switch (resource) {
603 	case H_SET_MODE_RESOURCE_SET_CIABR:
604 		if (!kvmppc_power8_compatible(vcpu))
605 			return H_P2;
606 		if (value2)
607 			return H_P4;
608 		if (mflags)
609 			return H_UNSUPPORTED_FLAG_START;
610 		/* Guests can't breakpoint the hypervisor */
611 		if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
612 			return H_P3;
613 		vcpu->arch.ciabr  = value1;
614 		return H_SUCCESS;
615 	case H_SET_MODE_RESOURCE_SET_DAWR:
616 		if (!kvmppc_power8_compatible(vcpu))
617 			return H_P2;
618 		if (mflags)
619 			return H_UNSUPPORTED_FLAG_START;
620 		if (value2 & DABRX_HYP)
621 			return H_P4;
622 		vcpu->arch.dawr  = value1;
623 		vcpu->arch.dawrx = value2;
624 		return H_SUCCESS;
625 	default:
626 		return H_TOO_HARD;
627 	}
628 }
629 
630 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
631 {
632 	struct kvmppc_vcore *vcore = target->arch.vcore;
633 
634 	/*
635 	 * We expect to have been called by the real mode handler
636 	 * (kvmppc_rm_h_confer()) which would have directly returned
637 	 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
638 	 * have useful work to do and should not confer) so we don't
639 	 * recheck that here.
640 	 */
641 
642 	spin_lock(&vcore->lock);
643 	if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
644 	    vcore->vcore_state != VCORE_INACTIVE)
645 		target = vcore->runner;
646 	spin_unlock(&vcore->lock);
647 
648 	return kvm_vcpu_yield_to(target);
649 }
650 
651 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
652 {
653 	int yield_count = 0;
654 	struct lppaca *lppaca;
655 
656 	spin_lock(&vcpu->arch.vpa_update_lock);
657 	lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
658 	if (lppaca)
659 		yield_count = be32_to_cpu(lppaca->yield_count);
660 	spin_unlock(&vcpu->arch.vpa_update_lock);
661 	return yield_count;
662 }
663 
664 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
665 {
666 	unsigned long req = kvmppc_get_gpr(vcpu, 3);
667 	unsigned long target, ret = H_SUCCESS;
668 	int yield_count;
669 	struct kvm_vcpu *tvcpu;
670 	int idx, rc;
671 
672 	if (req <= MAX_HCALL_OPCODE &&
673 	    !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
674 		return RESUME_HOST;
675 
676 	switch (req) {
677 	case H_CEDE:
678 		break;
679 	case H_PROD:
680 		target = kvmppc_get_gpr(vcpu, 4);
681 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
682 		if (!tvcpu) {
683 			ret = H_PARAMETER;
684 			break;
685 		}
686 		tvcpu->arch.prodded = 1;
687 		smp_mb();
688 		if (vcpu->arch.ceded) {
689 			if (waitqueue_active(&vcpu->wq)) {
690 				wake_up_interruptible(&vcpu->wq);
691 				vcpu->stat.halt_wakeup++;
692 			}
693 		}
694 		break;
695 	case H_CONFER:
696 		target = kvmppc_get_gpr(vcpu, 4);
697 		if (target == -1)
698 			break;
699 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
700 		if (!tvcpu) {
701 			ret = H_PARAMETER;
702 			break;
703 		}
704 		yield_count = kvmppc_get_gpr(vcpu, 5);
705 		if (kvmppc_get_yield_count(tvcpu) != yield_count)
706 			break;
707 		kvm_arch_vcpu_yield_to(tvcpu);
708 		break;
709 	case H_REGISTER_VPA:
710 		ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
711 					kvmppc_get_gpr(vcpu, 5),
712 					kvmppc_get_gpr(vcpu, 6));
713 		break;
714 	case H_RTAS:
715 		if (list_empty(&vcpu->kvm->arch.rtas_tokens))
716 			return RESUME_HOST;
717 
718 		idx = srcu_read_lock(&vcpu->kvm->srcu);
719 		rc = kvmppc_rtas_hcall(vcpu);
720 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
721 
722 		if (rc == -ENOENT)
723 			return RESUME_HOST;
724 		else if (rc == 0)
725 			break;
726 
727 		/* Send the error out to userspace via KVM_RUN */
728 		return rc;
729 	case H_LOGICAL_CI_LOAD:
730 		ret = kvmppc_h_logical_ci_load(vcpu);
731 		if (ret == H_TOO_HARD)
732 			return RESUME_HOST;
733 		break;
734 	case H_LOGICAL_CI_STORE:
735 		ret = kvmppc_h_logical_ci_store(vcpu);
736 		if (ret == H_TOO_HARD)
737 			return RESUME_HOST;
738 		break;
739 	case H_SET_MODE:
740 		ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
741 					kvmppc_get_gpr(vcpu, 5),
742 					kvmppc_get_gpr(vcpu, 6),
743 					kvmppc_get_gpr(vcpu, 7));
744 		if (ret == H_TOO_HARD)
745 			return RESUME_HOST;
746 		break;
747 	case H_XIRR:
748 	case H_CPPR:
749 	case H_EOI:
750 	case H_IPI:
751 	case H_IPOLL:
752 	case H_XIRR_X:
753 		if (kvmppc_xics_enabled(vcpu)) {
754 			ret = kvmppc_xics_hcall(vcpu, req);
755 			break;
756 		} /* fallthrough */
757 	default:
758 		return RESUME_HOST;
759 	}
760 	kvmppc_set_gpr(vcpu, 3, ret);
761 	vcpu->arch.hcall_needed = 0;
762 	return RESUME_GUEST;
763 }
764 
765 static int kvmppc_hcall_impl_hv(unsigned long cmd)
766 {
767 	switch (cmd) {
768 	case H_CEDE:
769 	case H_PROD:
770 	case H_CONFER:
771 	case H_REGISTER_VPA:
772 	case H_SET_MODE:
773 	case H_LOGICAL_CI_LOAD:
774 	case H_LOGICAL_CI_STORE:
775 #ifdef CONFIG_KVM_XICS
776 	case H_XIRR:
777 	case H_CPPR:
778 	case H_EOI:
779 	case H_IPI:
780 	case H_IPOLL:
781 	case H_XIRR_X:
782 #endif
783 		return 1;
784 	}
785 
786 	/* See if it's in the real-mode table */
787 	return kvmppc_hcall_impl_hv_realmode(cmd);
788 }
789 
790 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
791 					struct kvm_vcpu *vcpu)
792 {
793 	u32 last_inst;
794 
795 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
796 					EMULATE_DONE) {
797 		/*
798 		 * Fetch failed, so return to guest and
799 		 * try executing it again.
800 		 */
801 		return RESUME_GUEST;
802 	}
803 
804 	if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
805 		run->exit_reason = KVM_EXIT_DEBUG;
806 		run->debug.arch.address = kvmppc_get_pc(vcpu);
807 		return RESUME_HOST;
808 	} else {
809 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
810 		return RESUME_GUEST;
811 	}
812 }
813 
814 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
815 				 struct task_struct *tsk)
816 {
817 	int r = RESUME_HOST;
818 
819 	vcpu->stat.sum_exits++;
820 
821 	run->exit_reason = KVM_EXIT_UNKNOWN;
822 	run->ready_for_interrupt_injection = 1;
823 	switch (vcpu->arch.trap) {
824 	/* We're good on these - the host merely wanted to get our attention */
825 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
826 		vcpu->stat.dec_exits++;
827 		r = RESUME_GUEST;
828 		break;
829 	case BOOK3S_INTERRUPT_EXTERNAL:
830 	case BOOK3S_INTERRUPT_H_DOORBELL:
831 		vcpu->stat.ext_intr_exits++;
832 		r = RESUME_GUEST;
833 		break;
834 	/* HMI is hypervisor interrupt and host has handled it. Resume guest.*/
835 	case BOOK3S_INTERRUPT_HMI:
836 	case BOOK3S_INTERRUPT_PERFMON:
837 		r = RESUME_GUEST;
838 		break;
839 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
840 		/*
841 		 * Deliver a machine check interrupt to the guest.
842 		 * We have to do this, even if the host has handled the
843 		 * machine check, because machine checks use SRR0/1 and
844 		 * the interrupt might have trashed guest state in them.
845 		 */
846 		kvmppc_book3s_queue_irqprio(vcpu,
847 					    BOOK3S_INTERRUPT_MACHINE_CHECK);
848 		r = RESUME_GUEST;
849 		break;
850 	case BOOK3S_INTERRUPT_PROGRAM:
851 	{
852 		ulong flags;
853 		/*
854 		 * Normally program interrupts are delivered directly
855 		 * to the guest by the hardware, but we can get here
856 		 * as a result of a hypervisor emulation interrupt
857 		 * (e40) getting turned into a 700 by BML RTAS.
858 		 */
859 		flags = vcpu->arch.shregs.msr & 0x1f0000ull;
860 		kvmppc_core_queue_program(vcpu, flags);
861 		r = RESUME_GUEST;
862 		break;
863 	}
864 	case BOOK3S_INTERRUPT_SYSCALL:
865 	{
866 		/* hcall - punt to userspace */
867 		int i;
868 
869 		/* hypercall with MSR_PR has already been handled in rmode,
870 		 * and never reaches here.
871 		 */
872 
873 		run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
874 		for (i = 0; i < 9; ++i)
875 			run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
876 		run->exit_reason = KVM_EXIT_PAPR_HCALL;
877 		vcpu->arch.hcall_needed = 1;
878 		r = RESUME_HOST;
879 		break;
880 	}
881 	/*
882 	 * We get these next two if the guest accesses a page which it thinks
883 	 * it has mapped but which is not actually present, either because
884 	 * it is for an emulated I/O device or because the corresonding
885 	 * host page has been paged out.  Any other HDSI/HISI interrupts
886 	 * have been handled already.
887 	 */
888 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
889 		r = RESUME_PAGE_FAULT;
890 		break;
891 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
892 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
893 		vcpu->arch.fault_dsisr = 0;
894 		r = RESUME_PAGE_FAULT;
895 		break;
896 	/*
897 	 * This occurs if the guest executes an illegal instruction.
898 	 * If the guest debug is disabled, generate a program interrupt
899 	 * to the guest. If guest debug is enabled, we need to check
900 	 * whether the instruction is a software breakpoint instruction.
901 	 * Accordingly return to Guest or Host.
902 	 */
903 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
904 		if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
905 			vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
906 				swab32(vcpu->arch.emul_inst) :
907 				vcpu->arch.emul_inst;
908 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
909 			r = kvmppc_emulate_debug_inst(run, vcpu);
910 		} else {
911 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
912 			r = RESUME_GUEST;
913 		}
914 		break;
915 	/*
916 	 * This occurs if the guest (kernel or userspace), does something that
917 	 * is prohibited by HFSCR.  We just generate a program interrupt to
918 	 * the guest.
919 	 */
920 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
921 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
922 		r = RESUME_GUEST;
923 		break;
924 	default:
925 		kvmppc_dump_regs(vcpu);
926 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
927 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
928 			vcpu->arch.shregs.msr);
929 		run->hw.hardware_exit_reason = vcpu->arch.trap;
930 		r = RESUME_HOST;
931 		break;
932 	}
933 
934 	return r;
935 }
936 
937 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
938 					    struct kvm_sregs *sregs)
939 {
940 	int i;
941 
942 	memset(sregs, 0, sizeof(struct kvm_sregs));
943 	sregs->pvr = vcpu->arch.pvr;
944 	for (i = 0; i < vcpu->arch.slb_max; i++) {
945 		sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
946 		sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
947 	}
948 
949 	return 0;
950 }
951 
952 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
953 					    struct kvm_sregs *sregs)
954 {
955 	int i, j;
956 
957 	/* Only accept the same PVR as the host's, since we can't spoof it */
958 	if (sregs->pvr != vcpu->arch.pvr)
959 		return -EINVAL;
960 
961 	j = 0;
962 	for (i = 0; i < vcpu->arch.slb_nr; i++) {
963 		if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
964 			vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
965 			vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
966 			++j;
967 		}
968 	}
969 	vcpu->arch.slb_max = j;
970 
971 	return 0;
972 }
973 
974 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
975 		bool preserve_top32)
976 {
977 	struct kvm *kvm = vcpu->kvm;
978 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
979 	u64 mask;
980 
981 	mutex_lock(&kvm->lock);
982 	spin_lock(&vc->lock);
983 	/*
984 	 * If ILE (interrupt little-endian) has changed, update the
985 	 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
986 	 */
987 	if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
988 		struct kvm_vcpu *vcpu;
989 		int i;
990 
991 		kvm_for_each_vcpu(i, vcpu, kvm) {
992 			if (vcpu->arch.vcore != vc)
993 				continue;
994 			if (new_lpcr & LPCR_ILE)
995 				vcpu->arch.intr_msr |= MSR_LE;
996 			else
997 				vcpu->arch.intr_msr &= ~MSR_LE;
998 		}
999 	}
1000 
1001 	/*
1002 	 * Userspace can only modify DPFD (default prefetch depth),
1003 	 * ILE (interrupt little-endian) and TC (translation control).
1004 	 * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
1005 	 */
1006 	mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1007 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
1008 		mask |= LPCR_AIL;
1009 
1010 	/* Broken 32-bit version of LPCR must not clear top bits */
1011 	if (preserve_top32)
1012 		mask &= 0xFFFFFFFF;
1013 	vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1014 	spin_unlock(&vc->lock);
1015 	mutex_unlock(&kvm->lock);
1016 }
1017 
1018 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1019 				 union kvmppc_one_reg *val)
1020 {
1021 	int r = 0;
1022 	long int i;
1023 
1024 	switch (id) {
1025 	case KVM_REG_PPC_DEBUG_INST:
1026 		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1027 		break;
1028 	case KVM_REG_PPC_HIOR:
1029 		*val = get_reg_val(id, 0);
1030 		break;
1031 	case KVM_REG_PPC_DABR:
1032 		*val = get_reg_val(id, vcpu->arch.dabr);
1033 		break;
1034 	case KVM_REG_PPC_DABRX:
1035 		*val = get_reg_val(id, vcpu->arch.dabrx);
1036 		break;
1037 	case KVM_REG_PPC_DSCR:
1038 		*val = get_reg_val(id, vcpu->arch.dscr);
1039 		break;
1040 	case KVM_REG_PPC_PURR:
1041 		*val = get_reg_val(id, vcpu->arch.purr);
1042 		break;
1043 	case KVM_REG_PPC_SPURR:
1044 		*val = get_reg_val(id, vcpu->arch.spurr);
1045 		break;
1046 	case KVM_REG_PPC_AMR:
1047 		*val = get_reg_val(id, vcpu->arch.amr);
1048 		break;
1049 	case KVM_REG_PPC_UAMOR:
1050 		*val = get_reg_val(id, vcpu->arch.uamor);
1051 		break;
1052 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1053 		i = id - KVM_REG_PPC_MMCR0;
1054 		*val = get_reg_val(id, vcpu->arch.mmcr[i]);
1055 		break;
1056 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1057 		i = id - KVM_REG_PPC_PMC1;
1058 		*val = get_reg_val(id, vcpu->arch.pmc[i]);
1059 		break;
1060 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1061 		i = id - KVM_REG_PPC_SPMC1;
1062 		*val = get_reg_val(id, vcpu->arch.spmc[i]);
1063 		break;
1064 	case KVM_REG_PPC_SIAR:
1065 		*val = get_reg_val(id, vcpu->arch.siar);
1066 		break;
1067 	case KVM_REG_PPC_SDAR:
1068 		*val = get_reg_val(id, vcpu->arch.sdar);
1069 		break;
1070 	case KVM_REG_PPC_SIER:
1071 		*val = get_reg_val(id, vcpu->arch.sier);
1072 		break;
1073 	case KVM_REG_PPC_IAMR:
1074 		*val = get_reg_val(id, vcpu->arch.iamr);
1075 		break;
1076 	case KVM_REG_PPC_PSPB:
1077 		*val = get_reg_val(id, vcpu->arch.pspb);
1078 		break;
1079 	case KVM_REG_PPC_DPDES:
1080 		*val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1081 		break;
1082 	case KVM_REG_PPC_DAWR:
1083 		*val = get_reg_val(id, vcpu->arch.dawr);
1084 		break;
1085 	case KVM_REG_PPC_DAWRX:
1086 		*val = get_reg_val(id, vcpu->arch.dawrx);
1087 		break;
1088 	case KVM_REG_PPC_CIABR:
1089 		*val = get_reg_val(id, vcpu->arch.ciabr);
1090 		break;
1091 	case KVM_REG_PPC_CSIGR:
1092 		*val = get_reg_val(id, vcpu->arch.csigr);
1093 		break;
1094 	case KVM_REG_PPC_TACR:
1095 		*val = get_reg_val(id, vcpu->arch.tacr);
1096 		break;
1097 	case KVM_REG_PPC_TCSCR:
1098 		*val = get_reg_val(id, vcpu->arch.tcscr);
1099 		break;
1100 	case KVM_REG_PPC_PID:
1101 		*val = get_reg_val(id, vcpu->arch.pid);
1102 		break;
1103 	case KVM_REG_PPC_ACOP:
1104 		*val = get_reg_val(id, vcpu->arch.acop);
1105 		break;
1106 	case KVM_REG_PPC_WORT:
1107 		*val = get_reg_val(id, vcpu->arch.wort);
1108 		break;
1109 	case KVM_REG_PPC_VPA_ADDR:
1110 		spin_lock(&vcpu->arch.vpa_update_lock);
1111 		*val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1112 		spin_unlock(&vcpu->arch.vpa_update_lock);
1113 		break;
1114 	case KVM_REG_PPC_VPA_SLB:
1115 		spin_lock(&vcpu->arch.vpa_update_lock);
1116 		val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1117 		val->vpaval.length = vcpu->arch.slb_shadow.len;
1118 		spin_unlock(&vcpu->arch.vpa_update_lock);
1119 		break;
1120 	case KVM_REG_PPC_VPA_DTL:
1121 		spin_lock(&vcpu->arch.vpa_update_lock);
1122 		val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1123 		val->vpaval.length = vcpu->arch.dtl.len;
1124 		spin_unlock(&vcpu->arch.vpa_update_lock);
1125 		break;
1126 	case KVM_REG_PPC_TB_OFFSET:
1127 		*val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1128 		break;
1129 	case KVM_REG_PPC_LPCR:
1130 	case KVM_REG_PPC_LPCR_64:
1131 		*val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1132 		break;
1133 	case KVM_REG_PPC_PPR:
1134 		*val = get_reg_val(id, vcpu->arch.ppr);
1135 		break;
1136 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1137 	case KVM_REG_PPC_TFHAR:
1138 		*val = get_reg_val(id, vcpu->arch.tfhar);
1139 		break;
1140 	case KVM_REG_PPC_TFIAR:
1141 		*val = get_reg_val(id, vcpu->arch.tfiar);
1142 		break;
1143 	case KVM_REG_PPC_TEXASR:
1144 		*val = get_reg_val(id, vcpu->arch.texasr);
1145 		break;
1146 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1147 		i = id - KVM_REG_PPC_TM_GPR0;
1148 		*val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1149 		break;
1150 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1151 	{
1152 		int j;
1153 		i = id - KVM_REG_PPC_TM_VSR0;
1154 		if (i < 32)
1155 			for (j = 0; j < TS_FPRWIDTH; j++)
1156 				val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1157 		else {
1158 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1159 				val->vval = vcpu->arch.vr_tm.vr[i-32];
1160 			else
1161 				r = -ENXIO;
1162 		}
1163 		break;
1164 	}
1165 	case KVM_REG_PPC_TM_CR:
1166 		*val = get_reg_val(id, vcpu->arch.cr_tm);
1167 		break;
1168 	case KVM_REG_PPC_TM_LR:
1169 		*val = get_reg_val(id, vcpu->arch.lr_tm);
1170 		break;
1171 	case KVM_REG_PPC_TM_CTR:
1172 		*val = get_reg_val(id, vcpu->arch.ctr_tm);
1173 		break;
1174 	case KVM_REG_PPC_TM_FPSCR:
1175 		*val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1176 		break;
1177 	case KVM_REG_PPC_TM_AMR:
1178 		*val = get_reg_val(id, vcpu->arch.amr_tm);
1179 		break;
1180 	case KVM_REG_PPC_TM_PPR:
1181 		*val = get_reg_val(id, vcpu->arch.ppr_tm);
1182 		break;
1183 	case KVM_REG_PPC_TM_VRSAVE:
1184 		*val = get_reg_val(id, vcpu->arch.vrsave_tm);
1185 		break;
1186 	case KVM_REG_PPC_TM_VSCR:
1187 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1188 			*val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1189 		else
1190 			r = -ENXIO;
1191 		break;
1192 	case KVM_REG_PPC_TM_DSCR:
1193 		*val = get_reg_val(id, vcpu->arch.dscr_tm);
1194 		break;
1195 	case KVM_REG_PPC_TM_TAR:
1196 		*val = get_reg_val(id, vcpu->arch.tar_tm);
1197 		break;
1198 #endif
1199 	case KVM_REG_PPC_ARCH_COMPAT:
1200 		*val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1201 		break;
1202 	default:
1203 		r = -EINVAL;
1204 		break;
1205 	}
1206 
1207 	return r;
1208 }
1209 
1210 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1211 				 union kvmppc_one_reg *val)
1212 {
1213 	int r = 0;
1214 	long int i;
1215 	unsigned long addr, len;
1216 
1217 	switch (id) {
1218 	case KVM_REG_PPC_HIOR:
1219 		/* Only allow this to be set to zero */
1220 		if (set_reg_val(id, *val))
1221 			r = -EINVAL;
1222 		break;
1223 	case KVM_REG_PPC_DABR:
1224 		vcpu->arch.dabr = set_reg_val(id, *val);
1225 		break;
1226 	case KVM_REG_PPC_DABRX:
1227 		vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1228 		break;
1229 	case KVM_REG_PPC_DSCR:
1230 		vcpu->arch.dscr = set_reg_val(id, *val);
1231 		break;
1232 	case KVM_REG_PPC_PURR:
1233 		vcpu->arch.purr = set_reg_val(id, *val);
1234 		break;
1235 	case KVM_REG_PPC_SPURR:
1236 		vcpu->arch.spurr = set_reg_val(id, *val);
1237 		break;
1238 	case KVM_REG_PPC_AMR:
1239 		vcpu->arch.amr = set_reg_val(id, *val);
1240 		break;
1241 	case KVM_REG_PPC_UAMOR:
1242 		vcpu->arch.uamor = set_reg_val(id, *val);
1243 		break;
1244 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1245 		i = id - KVM_REG_PPC_MMCR0;
1246 		vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1247 		break;
1248 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1249 		i = id - KVM_REG_PPC_PMC1;
1250 		vcpu->arch.pmc[i] = set_reg_val(id, *val);
1251 		break;
1252 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1253 		i = id - KVM_REG_PPC_SPMC1;
1254 		vcpu->arch.spmc[i] = set_reg_val(id, *val);
1255 		break;
1256 	case KVM_REG_PPC_SIAR:
1257 		vcpu->arch.siar = set_reg_val(id, *val);
1258 		break;
1259 	case KVM_REG_PPC_SDAR:
1260 		vcpu->arch.sdar = set_reg_val(id, *val);
1261 		break;
1262 	case KVM_REG_PPC_SIER:
1263 		vcpu->arch.sier = set_reg_val(id, *val);
1264 		break;
1265 	case KVM_REG_PPC_IAMR:
1266 		vcpu->arch.iamr = set_reg_val(id, *val);
1267 		break;
1268 	case KVM_REG_PPC_PSPB:
1269 		vcpu->arch.pspb = set_reg_val(id, *val);
1270 		break;
1271 	case KVM_REG_PPC_DPDES:
1272 		vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1273 		break;
1274 	case KVM_REG_PPC_DAWR:
1275 		vcpu->arch.dawr = set_reg_val(id, *val);
1276 		break;
1277 	case KVM_REG_PPC_DAWRX:
1278 		vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1279 		break;
1280 	case KVM_REG_PPC_CIABR:
1281 		vcpu->arch.ciabr = set_reg_val(id, *val);
1282 		/* Don't allow setting breakpoints in hypervisor code */
1283 		if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1284 			vcpu->arch.ciabr &= ~CIABR_PRIV;	/* disable */
1285 		break;
1286 	case KVM_REG_PPC_CSIGR:
1287 		vcpu->arch.csigr = set_reg_val(id, *val);
1288 		break;
1289 	case KVM_REG_PPC_TACR:
1290 		vcpu->arch.tacr = set_reg_val(id, *val);
1291 		break;
1292 	case KVM_REG_PPC_TCSCR:
1293 		vcpu->arch.tcscr = set_reg_val(id, *val);
1294 		break;
1295 	case KVM_REG_PPC_PID:
1296 		vcpu->arch.pid = set_reg_val(id, *val);
1297 		break;
1298 	case KVM_REG_PPC_ACOP:
1299 		vcpu->arch.acop = set_reg_val(id, *val);
1300 		break;
1301 	case KVM_REG_PPC_WORT:
1302 		vcpu->arch.wort = set_reg_val(id, *val);
1303 		break;
1304 	case KVM_REG_PPC_VPA_ADDR:
1305 		addr = set_reg_val(id, *val);
1306 		r = -EINVAL;
1307 		if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1308 			      vcpu->arch.dtl.next_gpa))
1309 			break;
1310 		r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1311 		break;
1312 	case KVM_REG_PPC_VPA_SLB:
1313 		addr = val->vpaval.addr;
1314 		len = val->vpaval.length;
1315 		r = -EINVAL;
1316 		if (addr && !vcpu->arch.vpa.next_gpa)
1317 			break;
1318 		r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1319 		break;
1320 	case KVM_REG_PPC_VPA_DTL:
1321 		addr = val->vpaval.addr;
1322 		len = val->vpaval.length;
1323 		r = -EINVAL;
1324 		if (addr && (len < sizeof(struct dtl_entry) ||
1325 			     !vcpu->arch.vpa.next_gpa))
1326 			break;
1327 		len -= len % sizeof(struct dtl_entry);
1328 		r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1329 		break;
1330 	case KVM_REG_PPC_TB_OFFSET:
1331 		/* round up to multiple of 2^24 */
1332 		vcpu->arch.vcore->tb_offset =
1333 			ALIGN(set_reg_val(id, *val), 1UL << 24);
1334 		break;
1335 	case KVM_REG_PPC_LPCR:
1336 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1337 		break;
1338 	case KVM_REG_PPC_LPCR_64:
1339 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1340 		break;
1341 	case KVM_REG_PPC_PPR:
1342 		vcpu->arch.ppr = set_reg_val(id, *val);
1343 		break;
1344 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1345 	case KVM_REG_PPC_TFHAR:
1346 		vcpu->arch.tfhar = set_reg_val(id, *val);
1347 		break;
1348 	case KVM_REG_PPC_TFIAR:
1349 		vcpu->arch.tfiar = set_reg_val(id, *val);
1350 		break;
1351 	case KVM_REG_PPC_TEXASR:
1352 		vcpu->arch.texasr = set_reg_val(id, *val);
1353 		break;
1354 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1355 		i = id - KVM_REG_PPC_TM_GPR0;
1356 		vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1357 		break;
1358 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1359 	{
1360 		int j;
1361 		i = id - KVM_REG_PPC_TM_VSR0;
1362 		if (i < 32)
1363 			for (j = 0; j < TS_FPRWIDTH; j++)
1364 				vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1365 		else
1366 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1367 				vcpu->arch.vr_tm.vr[i-32] = val->vval;
1368 			else
1369 				r = -ENXIO;
1370 		break;
1371 	}
1372 	case KVM_REG_PPC_TM_CR:
1373 		vcpu->arch.cr_tm = set_reg_val(id, *val);
1374 		break;
1375 	case KVM_REG_PPC_TM_LR:
1376 		vcpu->arch.lr_tm = set_reg_val(id, *val);
1377 		break;
1378 	case KVM_REG_PPC_TM_CTR:
1379 		vcpu->arch.ctr_tm = set_reg_val(id, *val);
1380 		break;
1381 	case KVM_REG_PPC_TM_FPSCR:
1382 		vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1383 		break;
1384 	case KVM_REG_PPC_TM_AMR:
1385 		vcpu->arch.amr_tm = set_reg_val(id, *val);
1386 		break;
1387 	case KVM_REG_PPC_TM_PPR:
1388 		vcpu->arch.ppr_tm = set_reg_val(id, *val);
1389 		break;
1390 	case KVM_REG_PPC_TM_VRSAVE:
1391 		vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1392 		break;
1393 	case KVM_REG_PPC_TM_VSCR:
1394 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1395 			vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1396 		else
1397 			r = - ENXIO;
1398 		break;
1399 	case KVM_REG_PPC_TM_DSCR:
1400 		vcpu->arch.dscr_tm = set_reg_val(id, *val);
1401 		break;
1402 	case KVM_REG_PPC_TM_TAR:
1403 		vcpu->arch.tar_tm = set_reg_val(id, *val);
1404 		break;
1405 #endif
1406 	case KVM_REG_PPC_ARCH_COMPAT:
1407 		r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1408 		break;
1409 	default:
1410 		r = -EINVAL;
1411 		break;
1412 	}
1413 
1414 	return r;
1415 }
1416 
1417 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1418 {
1419 	struct kvmppc_vcore *vcore;
1420 
1421 	vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1422 
1423 	if (vcore == NULL)
1424 		return NULL;
1425 
1426 	INIT_LIST_HEAD(&vcore->runnable_threads);
1427 	spin_lock_init(&vcore->lock);
1428 	spin_lock_init(&vcore->stoltb_lock);
1429 	init_waitqueue_head(&vcore->wq);
1430 	vcore->preempt_tb = TB_NIL;
1431 	vcore->lpcr = kvm->arch.lpcr;
1432 	vcore->first_vcpuid = core * threads_per_subcore;
1433 	vcore->kvm = kvm;
1434 
1435 	vcore->mpp_buffer_is_valid = false;
1436 
1437 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
1438 		vcore->mpp_buffer = (void *)__get_free_pages(
1439 			GFP_KERNEL|__GFP_ZERO,
1440 			MPP_BUFFER_ORDER);
1441 
1442 	return vcore;
1443 }
1444 
1445 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1446 static struct debugfs_timings_element {
1447 	const char *name;
1448 	size_t offset;
1449 } timings[] = {
1450 	{"rm_entry",	offsetof(struct kvm_vcpu, arch.rm_entry)},
1451 	{"rm_intr",	offsetof(struct kvm_vcpu, arch.rm_intr)},
1452 	{"rm_exit",	offsetof(struct kvm_vcpu, arch.rm_exit)},
1453 	{"guest",	offsetof(struct kvm_vcpu, arch.guest_time)},
1454 	{"cede",	offsetof(struct kvm_vcpu, arch.cede_time)},
1455 };
1456 
1457 #define N_TIMINGS	(sizeof(timings) / sizeof(timings[0]))
1458 
1459 struct debugfs_timings_state {
1460 	struct kvm_vcpu	*vcpu;
1461 	unsigned int	buflen;
1462 	char		buf[N_TIMINGS * 100];
1463 };
1464 
1465 static int debugfs_timings_open(struct inode *inode, struct file *file)
1466 {
1467 	struct kvm_vcpu *vcpu = inode->i_private;
1468 	struct debugfs_timings_state *p;
1469 
1470 	p = kzalloc(sizeof(*p), GFP_KERNEL);
1471 	if (!p)
1472 		return -ENOMEM;
1473 
1474 	kvm_get_kvm(vcpu->kvm);
1475 	p->vcpu = vcpu;
1476 	file->private_data = p;
1477 
1478 	return nonseekable_open(inode, file);
1479 }
1480 
1481 static int debugfs_timings_release(struct inode *inode, struct file *file)
1482 {
1483 	struct debugfs_timings_state *p = file->private_data;
1484 
1485 	kvm_put_kvm(p->vcpu->kvm);
1486 	kfree(p);
1487 	return 0;
1488 }
1489 
1490 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1491 				    size_t len, loff_t *ppos)
1492 {
1493 	struct debugfs_timings_state *p = file->private_data;
1494 	struct kvm_vcpu *vcpu = p->vcpu;
1495 	char *s, *buf_end;
1496 	struct kvmhv_tb_accumulator tb;
1497 	u64 count;
1498 	loff_t pos;
1499 	ssize_t n;
1500 	int i, loops;
1501 	bool ok;
1502 
1503 	if (!p->buflen) {
1504 		s = p->buf;
1505 		buf_end = s + sizeof(p->buf);
1506 		for (i = 0; i < N_TIMINGS; ++i) {
1507 			struct kvmhv_tb_accumulator *acc;
1508 
1509 			acc = (struct kvmhv_tb_accumulator *)
1510 				((unsigned long)vcpu + timings[i].offset);
1511 			ok = false;
1512 			for (loops = 0; loops < 1000; ++loops) {
1513 				count = acc->seqcount;
1514 				if (!(count & 1)) {
1515 					smp_rmb();
1516 					tb = *acc;
1517 					smp_rmb();
1518 					if (count == acc->seqcount) {
1519 						ok = true;
1520 						break;
1521 					}
1522 				}
1523 				udelay(1);
1524 			}
1525 			if (!ok)
1526 				snprintf(s, buf_end - s, "%s: stuck\n",
1527 					timings[i].name);
1528 			else
1529 				snprintf(s, buf_end - s,
1530 					"%s: %llu %llu %llu %llu\n",
1531 					timings[i].name, count / 2,
1532 					tb_to_ns(tb.tb_total),
1533 					tb_to_ns(tb.tb_min),
1534 					tb_to_ns(tb.tb_max));
1535 			s += strlen(s);
1536 		}
1537 		p->buflen = s - p->buf;
1538 	}
1539 
1540 	pos = *ppos;
1541 	if (pos >= p->buflen)
1542 		return 0;
1543 	if (len > p->buflen - pos)
1544 		len = p->buflen - pos;
1545 	n = copy_to_user(buf, p->buf + pos, len);
1546 	if (n) {
1547 		if (n == len)
1548 			return -EFAULT;
1549 		len -= n;
1550 	}
1551 	*ppos = pos + len;
1552 	return len;
1553 }
1554 
1555 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1556 				     size_t len, loff_t *ppos)
1557 {
1558 	return -EACCES;
1559 }
1560 
1561 static const struct file_operations debugfs_timings_ops = {
1562 	.owner	 = THIS_MODULE,
1563 	.open	 = debugfs_timings_open,
1564 	.release = debugfs_timings_release,
1565 	.read	 = debugfs_timings_read,
1566 	.write	 = debugfs_timings_write,
1567 	.llseek	 = generic_file_llseek,
1568 };
1569 
1570 /* Create a debugfs directory for the vcpu */
1571 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1572 {
1573 	char buf[16];
1574 	struct kvm *kvm = vcpu->kvm;
1575 
1576 	snprintf(buf, sizeof(buf), "vcpu%u", id);
1577 	if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1578 		return;
1579 	vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1580 	if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1581 		return;
1582 	vcpu->arch.debugfs_timings =
1583 		debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1584 				    vcpu, &debugfs_timings_ops);
1585 }
1586 
1587 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1588 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1589 {
1590 }
1591 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1592 
1593 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1594 						   unsigned int id)
1595 {
1596 	struct kvm_vcpu *vcpu;
1597 	int err = -EINVAL;
1598 	int core;
1599 	struct kvmppc_vcore *vcore;
1600 
1601 	core = id / threads_per_subcore;
1602 	if (core >= KVM_MAX_VCORES)
1603 		goto out;
1604 
1605 	err = -ENOMEM;
1606 	vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1607 	if (!vcpu)
1608 		goto out;
1609 
1610 	err = kvm_vcpu_init(vcpu, kvm, id);
1611 	if (err)
1612 		goto free_vcpu;
1613 
1614 	vcpu->arch.shared = &vcpu->arch.shregs;
1615 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1616 	/*
1617 	 * The shared struct is never shared on HV,
1618 	 * so we can always use host endianness
1619 	 */
1620 #ifdef __BIG_ENDIAN__
1621 	vcpu->arch.shared_big_endian = true;
1622 #else
1623 	vcpu->arch.shared_big_endian = false;
1624 #endif
1625 #endif
1626 	vcpu->arch.mmcr[0] = MMCR0_FC;
1627 	vcpu->arch.ctrl = CTRL_RUNLATCH;
1628 	/* default to host PVR, since we can't spoof it */
1629 	kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1630 	spin_lock_init(&vcpu->arch.vpa_update_lock);
1631 	spin_lock_init(&vcpu->arch.tbacct_lock);
1632 	vcpu->arch.busy_preempt = TB_NIL;
1633 	vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1634 
1635 	kvmppc_mmu_book3s_hv_init(vcpu);
1636 
1637 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1638 
1639 	init_waitqueue_head(&vcpu->arch.cpu_run);
1640 
1641 	mutex_lock(&kvm->lock);
1642 	vcore = kvm->arch.vcores[core];
1643 	if (!vcore) {
1644 		vcore = kvmppc_vcore_create(kvm, core);
1645 		kvm->arch.vcores[core] = vcore;
1646 		kvm->arch.online_vcores++;
1647 	}
1648 	mutex_unlock(&kvm->lock);
1649 
1650 	if (!vcore)
1651 		goto free_vcpu;
1652 
1653 	spin_lock(&vcore->lock);
1654 	++vcore->num_threads;
1655 	spin_unlock(&vcore->lock);
1656 	vcpu->arch.vcore = vcore;
1657 	vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1658 
1659 	vcpu->arch.cpu_type = KVM_CPU_3S_64;
1660 	kvmppc_sanity_check(vcpu);
1661 
1662 	debugfs_vcpu_init(vcpu, id);
1663 
1664 	return vcpu;
1665 
1666 free_vcpu:
1667 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1668 out:
1669 	return ERR_PTR(err);
1670 }
1671 
1672 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1673 {
1674 	if (vpa->pinned_addr)
1675 		kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1676 					vpa->dirty);
1677 }
1678 
1679 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
1680 {
1681 	spin_lock(&vcpu->arch.vpa_update_lock);
1682 	unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1683 	unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1684 	unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1685 	spin_unlock(&vcpu->arch.vpa_update_lock);
1686 	kvm_vcpu_uninit(vcpu);
1687 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1688 }
1689 
1690 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1691 {
1692 	/* Indicate we want to get back into the guest */
1693 	return 1;
1694 }
1695 
1696 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1697 {
1698 	unsigned long dec_nsec, now;
1699 
1700 	now = get_tb();
1701 	if (now > vcpu->arch.dec_expires) {
1702 		/* decrementer has already gone negative */
1703 		kvmppc_core_queue_dec(vcpu);
1704 		kvmppc_core_prepare_to_enter(vcpu);
1705 		return;
1706 	}
1707 	dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1708 		   / tb_ticks_per_sec;
1709 	hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1710 		      HRTIMER_MODE_REL);
1711 	vcpu->arch.timer_running = 1;
1712 }
1713 
1714 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1715 {
1716 	vcpu->arch.ceded = 0;
1717 	if (vcpu->arch.timer_running) {
1718 		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1719 		vcpu->arch.timer_running = 0;
1720 	}
1721 }
1722 
1723 extern void __kvmppc_vcore_entry(void);
1724 
1725 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1726 				   struct kvm_vcpu *vcpu)
1727 {
1728 	u64 now;
1729 
1730 	if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1731 		return;
1732 	spin_lock_irq(&vcpu->arch.tbacct_lock);
1733 	now = mftb();
1734 	vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1735 		vcpu->arch.stolen_logged;
1736 	vcpu->arch.busy_preempt = now;
1737 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1738 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
1739 	--vc->n_runnable;
1740 	list_del(&vcpu->arch.run_list);
1741 }
1742 
1743 static int kvmppc_grab_hwthread(int cpu)
1744 {
1745 	struct paca_struct *tpaca;
1746 	long timeout = 10000;
1747 
1748 	tpaca = &paca[cpu];
1749 
1750 	/* Ensure the thread won't go into the kernel if it wakes */
1751 	tpaca->kvm_hstate.kvm_vcpu = NULL;
1752 	tpaca->kvm_hstate.napping = 0;
1753 	smp_wmb();
1754 	tpaca->kvm_hstate.hwthread_req = 1;
1755 
1756 	/*
1757 	 * If the thread is already executing in the kernel (e.g. handling
1758 	 * a stray interrupt), wait for it to get back to nap mode.
1759 	 * The smp_mb() is to ensure that our setting of hwthread_req
1760 	 * is visible before we look at hwthread_state, so if this
1761 	 * races with the code at system_reset_pSeries and the thread
1762 	 * misses our setting of hwthread_req, we are sure to see its
1763 	 * setting of hwthread_state, and vice versa.
1764 	 */
1765 	smp_mb();
1766 	while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1767 		if (--timeout <= 0) {
1768 			pr_err("KVM: couldn't grab cpu %d\n", cpu);
1769 			return -EBUSY;
1770 		}
1771 		udelay(1);
1772 	}
1773 	return 0;
1774 }
1775 
1776 static void kvmppc_release_hwthread(int cpu)
1777 {
1778 	struct paca_struct *tpaca;
1779 
1780 	tpaca = &paca[cpu];
1781 	tpaca->kvm_hstate.hwthread_req = 0;
1782 	tpaca->kvm_hstate.kvm_vcpu = NULL;
1783 }
1784 
1785 static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
1786 {
1787 	int cpu;
1788 	struct paca_struct *tpaca;
1789 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
1790 
1791 	if (vcpu->arch.timer_running) {
1792 		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1793 		vcpu->arch.timer_running = 0;
1794 	}
1795 	cpu = vc->pcpu + vcpu->arch.ptid;
1796 	tpaca = &paca[cpu];
1797 	tpaca->kvm_hstate.kvm_vcore = vc;
1798 	tpaca->kvm_hstate.ptid = vcpu->arch.ptid;
1799 	vcpu->cpu = vc->pcpu;
1800 	/* Order stores to hstate.kvm_vcore etc. before store to kvm_vcpu */
1801 	smp_wmb();
1802 	tpaca->kvm_hstate.kvm_vcpu = vcpu;
1803 	if (cpu != smp_processor_id())
1804 		kvmppc_ipi_thread(cpu);
1805 }
1806 
1807 static void kvmppc_wait_for_nap(void)
1808 {
1809 	int cpu = smp_processor_id();
1810 	int i, loops;
1811 
1812 	for (loops = 0; loops < 1000000; ++loops) {
1813 		/*
1814 		 * Check if all threads are finished.
1815 		 * We set the vcpu pointer when starting a thread
1816 		 * and the thread clears it when finished, so we look
1817 		 * for any threads that still have a non-NULL vcpu ptr.
1818 		 */
1819 		for (i = 1; i < threads_per_subcore; ++i)
1820 			if (paca[cpu + i].kvm_hstate.kvm_vcpu)
1821 				break;
1822 		if (i == threads_per_subcore) {
1823 			HMT_medium();
1824 			return;
1825 		}
1826 		HMT_low();
1827 	}
1828 	HMT_medium();
1829 	for (i = 1; i < threads_per_subcore; ++i)
1830 		if (paca[cpu + i].kvm_hstate.kvm_vcpu)
1831 			pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
1832 }
1833 
1834 /*
1835  * Check that we are on thread 0 and that any other threads in
1836  * this core are off-line.  Then grab the threads so they can't
1837  * enter the kernel.
1838  */
1839 static int on_primary_thread(void)
1840 {
1841 	int cpu = smp_processor_id();
1842 	int thr;
1843 
1844 	/* Are we on a primary subcore? */
1845 	if (cpu_thread_in_subcore(cpu))
1846 		return 0;
1847 
1848 	thr = 0;
1849 	while (++thr < threads_per_subcore)
1850 		if (cpu_online(cpu + thr))
1851 			return 0;
1852 
1853 	/* Grab all hw threads so they can't go into the kernel */
1854 	for (thr = 1; thr < threads_per_subcore; ++thr) {
1855 		if (kvmppc_grab_hwthread(cpu + thr)) {
1856 			/* Couldn't grab one; let the others go */
1857 			do {
1858 				kvmppc_release_hwthread(cpu + thr);
1859 			} while (--thr > 0);
1860 			return 0;
1861 		}
1862 	}
1863 	return 1;
1864 }
1865 
1866 static void kvmppc_start_saving_l2_cache(struct kvmppc_vcore *vc)
1867 {
1868 	phys_addr_t phy_addr, mpp_addr;
1869 
1870 	phy_addr = (phys_addr_t)virt_to_phys(vc->mpp_buffer);
1871 	mpp_addr = phy_addr & PPC_MPPE_ADDRESS_MASK;
1872 
1873 	mtspr(SPRN_MPPR, mpp_addr | PPC_MPPR_FETCH_ABORT);
1874 	logmpp(mpp_addr | PPC_LOGMPP_LOG_L2);
1875 
1876 	vc->mpp_buffer_is_valid = true;
1877 }
1878 
1879 static void kvmppc_start_restoring_l2_cache(const struct kvmppc_vcore *vc)
1880 {
1881 	phys_addr_t phy_addr, mpp_addr;
1882 
1883 	phy_addr = virt_to_phys(vc->mpp_buffer);
1884 	mpp_addr = phy_addr & PPC_MPPE_ADDRESS_MASK;
1885 
1886 	/* We must abort any in-progress save operations to ensure
1887 	 * the table is valid so that prefetch engine knows when to
1888 	 * stop prefetching. */
1889 	logmpp(mpp_addr | PPC_LOGMPP_LOG_ABORT);
1890 	mtspr(SPRN_MPPR, mpp_addr | PPC_MPPR_FETCH_WHOLE_TABLE);
1891 }
1892 
1893 static void prepare_threads(struct kvmppc_vcore *vc)
1894 {
1895 	struct kvm_vcpu *vcpu, *vnext;
1896 
1897 	list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1898 				 arch.run_list) {
1899 		if (signal_pending(vcpu->arch.run_task))
1900 			vcpu->arch.ret = -EINTR;
1901 		else if (vcpu->arch.vpa.update_pending ||
1902 			 vcpu->arch.slb_shadow.update_pending ||
1903 			 vcpu->arch.dtl.update_pending)
1904 			vcpu->arch.ret = RESUME_GUEST;
1905 		else
1906 			continue;
1907 		kvmppc_remove_runnable(vc, vcpu);
1908 		wake_up(&vcpu->arch.cpu_run);
1909 	}
1910 }
1911 
1912 static void post_guest_process(struct kvmppc_vcore *vc)
1913 {
1914 	u64 now;
1915 	long ret;
1916 	struct kvm_vcpu *vcpu, *vnext;
1917 
1918 	now = get_tb();
1919 	list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1920 				 arch.run_list) {
1921 		/* cancel pending dec exception if dec is positive */
1922 		if (now < vcpu->arch.dec_expires &&
1923 		    kvmppc_core_pending_dec(vcpu))
1924 			kvmppc_core_dequeue_dec(vcpu);
1925 
1926 		trace_kvm_guest_exit(vcpu);
1927 
1928 		ret = RESUME_GUEST;
1929 		if (vcpu->arch.trap)
1930 			ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
1931 						    vcpu->arch.run_task);
1932 
1933 		vcpu->arch.ret = ret;
1934 		vcpu->arch.trap = 0;
1935 
1936 		if (vcpu->arch.ceded) {
1937 			if (!is_kvmppc_resume_guest(ret))
1938 				kvmppc_end_cede(vcpu);
1939 			else
1940 				kvmppc_set_timer(vcpu);
1941 		}
1942 		if (!is_kvmppc_resume_guest(vcpu->arch.ret)) {
1943 			kvmppc_remove_runnable(vc, vcpu);
1944 			wake_up(&vcpu->arch.cpu_run);
1945 		}
1946 	}
1947 }
1948 
1949 /*
1950  * Run a set of guest threads on a physical core.
1951  * Called with vc->lock held.
1952  */
1953 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
1954 {
1955 	struct kvm_vcpu *vcpu;
1956 	int i;
1957 	int srcu_idx;
1958 
1959 	/*
1960 	 * Remove from the list any threads that have a signal pending
1961 	 * or need a VPA update done
1962 	 */
1963 	prepare_threads(vc);
1964 
1965 	/* if the runner is no longer runnable, let the caller pick a new one */
1966 	if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
1967 		return;
1968 
1969 	/*
1970 	 * Initialize *vc.
1971 	 */
1972 	vc->entry_exit_map = 0;
1973 	vc->preempt_tb = TB_NIL;
1974 	vc->in_guest = 0;
1975 	vc->napping_threads = 0;
1976 	vc->conferring_threads = 0;
1977 
1978 	/*
1979 	 * Make sure we are running on primary threads, and that secondary
1980 	 * threads are offline.  Also check if the number of threads in this
1981 	 * guest are greater than the current system threads per guest.
1982 	 */
1983 	if ((threads_per_core > 1) &&
1984 	    ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
1985 		list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1986 			vcpu->arch.ret = -EBUSY;
1987 			kvmppc_remove_runnable(vc, vcpu);
1988 			wake_up(&vcpu->arch.cpu_run);
1989 		}
1990 		goto out;
1991 	}
1992 
1993 
1994 	vc->pcpu = smp_processor_id();
1995 	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1996 		kvmppc_start_thread(vcpu);
1997 		kvmppc_create_dtl_entry(vcpu, vc);
1998 		trace_kvm_guest_enter(vcpu);
1999 	}
2000 
2001 	/* Set this explicitly in case thread 0 doesn't have a vcpu */
2002 	get_paca()->kvm_hstate.kvm_vcore = vc;
2003 	get_paca()->kvm_hstate.ptid = 0;
2004 
2005 	vc->vcore_state = VCORE_RUNNING;
2006 	preempt_disable();
2007 
2008 	trace_kvmppc_run_core(vc, 0);
2009 
2010 	spin_unlock(&vc->lock);
2011 
2012 	kvm_guest_enter();
2013 
2014 	srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2015 
2016 	if (vc->mpp_buffer_is_valid)
2017 		kvmppc_start_restoring_l2_cache(vc);
2018 
2019 	__kvmppc_vcore_entry();
2020 
2021 	spin_lock(&vc->lock);
2022 
2023 	if (vc->mpp_buffer)
2024 		kvmppc_start_saving_l2_cache(vc);
2025 
2026 	/* disable sending of IPIs on virtual external irqs */
2027 	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
2028 		vcpu->cpu = -1;
2029 	/* wait for secondary threads to finish writing their state to memory */
2030 	kvmppc_wait_for_nap();
2031 	for (i = 0; i < threads_per_subcore; ++i)
2032 		kvmppc_release_hwthread(vc->pcpu + i);
2033 	/* prevent other vcpu threads from doing kvmppc_start_thread() now */
2034 	vc->vcore_state = VCORE_EXITING;
2035 	spin_unlock(&vc->lock);
2036 
2037 	srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2038 
2039 	/* make sure updates to secondary vcpu structs are visible now */
2040 	smp_mb();
2041 	kvm_guest_exit();
2042 
2043 	preempt_enable();
2044 
2045 	spin_lock(&vc->lock);
2046 	post_guest_process(vc);
2047 
2048  out:
2049 	vc->vcore_state = VCORE_INACTIVE;
2050 	trace_kvmppc_run_core(vc, 1);
2051 }
2052 
2053 /*
2054  * Wait for some other vcpu thread to execute us, and
2055  * wake us up when we need to handle something in the host.
2056  */
2057 static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
2058 {
2059 	DEFINE_WAIT(wait);
2060 
2061 	prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
2062 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
2063 		schedule();
2064 	finish_wait(&vcpu->arch.cpu_run, &wait);
2065 }
2066 
2067 /*
2068  * All the vcpus in this vcore are idle, so wait for a decrementer
2069  * or external interrupt to one of the vcpus.  vc->lock is held.
2070  */
2071 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
2072 {
2073 	struct kvm_vcpu *vcpu;
2074 	int do_sleep = 1;
2075 
2076 	DEFINE_WAIT(wait);
2077 
2078 	prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
2079 
2080 	/*
2081 	 * Check one last time for pending exceptions and ceded state after
2082 	 * we put ourselves on the wait queue
2083 	 */
2084 	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
2085 		if (vcpu->arch.pending_exceptions || !vcpu->arch.ceded) {
2086 			do_sleep = 0;
2087 			break;
2088 		}
2089 	}
2090 
2091 	if (!do_sleep) {
2092 		finish_wait(&vc->wq, &wait);
2093 		return;
2094 	}
2095 
2096 	vc->vcore_state = VCORE_SLEEPING;
2097 	trace_kvmppc_vcore_blocked(vc, 0);
2098 	spin_unlock(&vc->lock);
2099 	schedule();
2100 	finish_wait(&vc->wq, &wait);
2101 	spin_lock(&vc->lock);
2102 	vc->vcore_state = VCORE_INACTIVE;
2103 	trace_kvmppc_vcore_blocked(vc, 1);
2104 }
2105 
2106 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2107 {
2108 	int n_ceded;
2109 	struct kvmppc_vcore *vc;
2110 	struct kvm_vcpu *v, *vn;
2111 
2112 	trace_kvmppc_run_vcpu_enter(vcpu);
2113 
2114 	kvm_run->exit_reason = 0;
2115 	vcpu->arch.ret = RESUME_GUEST;
2116 	vcpu->arch.trap = 0;
2117 	kvmppc_update_vpas(vcpu);
2118 
2119 	/*
2120 	 * Synchronize with other threads in this virtual core
2121 	 */
2122 	vc = vcpu->arch.vcore;
2123 	spin_lock(&vc->lock);
2124 	vcpu->arch.ceded = 0;
2125 	vcpu->arch.run_task = current;
2126 	vcpu->arch.kvm_run = kvm_run;
2127 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
2128 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
2129 	vcpu->arch.busy_preempt = TB_NIL;
2130 	list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
2131 	++vc->n_runnable;
2132 
2133 	/*
2134 	 * This happens the first time this is called for a vcpu.
2135 	 * If the vcore is already running, we may be able to start
2136 	 * this thread straight away and have it join in.
2137 	 */
2138 	if (!signal_pending(current)) {
2139 		if (vc->vcore_state == VCORE_RUNNING && !VCORE_IS_EXITING(vc)) {
2140 			kvmppc_create_dtl_entry(vcpu, vc);
2141 			kvmppc_start_thread(vcpu);
2142 			trace_kvm_guest_enter(vcpu);
2143 		} else if (vc->vcore_state == VCORE_SLEEPING) {
2144 			wake_up(&vc->wq);
2145 		}
2146 
2147 	}
2148 
2149 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2150 	       !signal_pending(current)) {
2151 		if (vc->vcore_state != VCORE_INACTIVE) {
2152 			spin_unlock(&vc->lock);
2153 			kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
2154 			spin_lock(&vc->lock);
2155 			continue;
2156 		}
2157 		list_for_each_entry_safe(v, vn, &vc->runnable_threads,
2158 					 arch.run_list) {
2159 			kvmppc_core_prepare_to_enter(v);
2160 			if (signal_pending(v->arch.run_task)) {
2161 				kvmppc_remove_runnable(vc, v);
2162 				v->stat.signal_exits++;
2163 				v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
2164 				v->arch.ret = -EINTR;
2165 				wake_up(&v->arch.cpu_run);
2166 			}
2167 		}
2168 		if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2169 			break;
2170 		n_ceded = 0;
2171 		list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
2172 			if (!v->arch.pending_exceptions)
2173 				n_ceded += v->arch.ceded;
2174 			else
2175 				v->arch.ceded = 0;
2176 		}
2177 		vc->runner = vcpu;
2178 		if (n_ceded == vc->n_runnable) {
2179 			kvmppc_vcore_blocked(vc);
2180 		} else if (should_resched()) {
2181 			vc->vcore_state = VCORE_PREEMPT;
2182 			/* Let something else run */
2183 			cond_resched_lock(&vc->lock);
2184 			vc->vcore_state = VCORE_INACTIVE;
2185 		} else {
2186 			kvmppc_run_core(vc);
2187 		}
2188 		vc->runner = NULL;
2189 	}
2190 
2191 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2192 	       (vc->vcore_state == VCORE_RUNNING ||
2193 		vc->vcore_state == VCORE_EXITING)) {
2194 		spin_unlock(&vc->lock);
2195 		kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
2196 		spin_lock(&vc->lock);
2197 	}
2198 
2199 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2200 		kvmppc_remove_runnable(vc, vcpu);
2201 		vcpu->stat.signal_exits++;
2202 		kvm_run->exit_reason = KVM_EXIT_INTR;
2203 		vcpu->arch.ret = -EINTR;
2204 	}
2205 
2206 	if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
2207 		/* Wake up some vcpu to run the core */
2208 		v = list_first_entry(&vc->runnable_threads,
2209 				     struct kvm_vcpu, arch.run_list);
2210 		wake_up(&v->arch.cpu_run);
2211 	}
2212 
2213 	trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
2214 	spin_unlock(&vc->lock);
2215 	return vcpu->arch.ret;
2216 }
2217 
2218 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
2219 {
2220 	int r;
2221 	int srcu_idx;
2222 
2223 	if (!vcpu->arch.sane) {
2224 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2225 		return -EINVAL;
2226 	}
2227 
2228 	kvmppc_core_prepare_to_enter(vcpu);
2229 
2230 	/* No need to go into the guest when all we'll do is come back out */
2231 	if (signal_pending(current)) {
2232 		run->exit_reason = KVM_EXIT_INTR;
2233 		return -EINTR;
2234 	}
2235 
2236 	atomic_inc(&vcpu->kvm->arch.vcpus_running);
2237 	/* Order vcpus_running vs. hpte_setup_done, see kvmppc_alloc_reset_hpt */
2238 	smp_mb();
2239 
2240 	/* On the first time here, set up HTAB and VRMA */
2241 	if (!vcpu->kvm->arch.hpte_setup_done) {
2242 		r = kvmppc_hv_setup_htab_rma(vcpu);
2243 		if (r)
2244 			goto out;
2245 	}
2246 
2247 	flush_fp_to_thread(current);
2248 	flush_altivec_to_thread(current);
2249 	flush_vsx_to_thread(current);
2250 	vcpu->arch.wqp = &vcpu->arch.vcore->wq;
2251 	vcpu->arch.pgdir = current->mm->pgd;
2252 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2253 
2254 	do {
2255 		r = kvmppc_run_vcpu(run, vcpu);
2256 
2257 		if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
2258 		    !(vcpu->arch.shregs.msr & MSR_PR)) {
2259 			trace_kvm_hcall_enter(vcpu);
2260 			r = kvmppc_pseries_do_hcall(vcpu);
2261 			trace_kvm_hcall_exit(vcpu, r);
2262 			kvmppc_core_prepare_to_enter(vcpu);
2263 		} else if (r == RESUME_PAGE_FAULT) {
2264 			srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2265 			r = kvmppc_book3s_hv_page_fault(run, vcpu,
2266 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
2267 			srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2268 		}
2269 	} while (is_kvmppc_resume_guest(r));
2270 
2271  out:
2272 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2273 	atomic_dec(&vcpu->kvm->arch.vcpus_running);
2274 	return r;
2275 }
2276 
2277 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
2278 				     int linux_psize)
2279 {
2280 	struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
2281 
2282 	if (!def->shift)
2283 		return;
2284 	(*sps)->page_shift = def->shift;
2285 	(*sps)->slb_enc = def->sllp;
2286 	(*sps)->enc[0].page_shift = def->shift;
2287 	(*sps)->enc[0].pte_enc = def->penc[linux_psize];
2288 	/*
2289 	 * Add 16MB MPSS support if host supports it
2290 	 */
2291 	if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
2292 		(*sps)->enc[1].page_shift = 24;
2293 		(*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
2294 	}
2295 	(*sps)++;
2296 }
2297 
2298 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
2299 					 struct kvm_ppc_smmu_info *info)
2300 {
2301 	struct kvm_ppc_one_seg_page_size *sps;
2302 
2303 	info->flags = KVM_PPC_PAGE_SIZES_REAL;
2304 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
2305 		info->flags |= KVM_PPC_1T_SEGMENTS;
2306 	info->slb_size = mmu_slb_size;
2307 
2308 	/* We only support these sizes for now, and no muti-size segments */
2309 	sps = &info->sps[0];
2310 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
2311 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
2312 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
2313 
2314 	return 0;
2315 }
2316 
2317 /*
2318  * Get (and clear) the dirty memory log for a memory slot.
2319  */
2320 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
2321 					 struct kvm_dirty_log *log)
2322 {
2323 	struct kvm_memory_slot *memslot;
2324 	int r;
2325 	unsigned long n;
2326 
2327 	mutex_lock(&kvm->slots_lock);
2328 
2329 	r = -EINVAL;
2330 	if (log->slot >= KVM_USER_MEM_SLOTS)
2331 		goto out;
2332 
2333 	memslot = id_to_memslot(kvm->memslots, log->slot);
2334 	r = -ENOENT;
2335 	if (!memslot->dirty_bitmap)
2336 		goto out;
2337 
2338 	n = kvm_dirty_bitmap_bytes(memslot);
2339 	memset(memslot->dirty_bitmap, 0, n);
2340 
2341 	r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
2342 	if (r)
2343 		goto out;
2344 
2345 	r = -EFAULT;
2346 	if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
2347 		goto out;
2348 
2349 	r = 0;
2350 out:
2351 	mutex_unlock(&kvm->slots_lock);
2352 	return r;
2353 }
2354 
2355 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
2356 					struct kvm_memory_slot *dont)
2357 {
2358 	if (!dont || free->arch.rmap != dont->arch.rmap) {
2359 		vfree(free->arch.rmap);
2360 		free->arch.rmap = NULL;
2361 	}
2362 }
2363 
2364 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
2365 					 unsigned long npages)
2366 {
2367 	slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
2368 	if (!slot->arch.rmap)
2369 		return -ENOMEM;
2370 
2371 	return 0;
2372 }
2373 
2374 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
2375 					struct kvm_memory_slot *memslot,
2376 					struct kvm_userspace_memory_region *mem)
2377 {
2378 	return 0;
2379 }
2380 
2381 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
2382 				struct kvm_userspace_memory_region *mem,
2383 				const struct kvm_memory_slot *old)
2384 {
2385 	unsigned long npages = mem->memory_size >> PAGE_SHIFT;
2386 	struct kvm_memory_slot *memslot;
2387 
2388 	if (npages && old->npages) {
2389 		/*
2390 		 * If modifying a memslot, reset all the rmap dirty bits.
2391 		 * If this is a new memslot, we don't need to do anything
2392 		 * since the rmap array starts out as all zeroes,
2393 		 * i.e. no pages are dirty.
2394 		 */
2395 		memslot = id_to_memslot(kvm->memslots, mem->slot);
2396 		kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
2397 	}
2398 }
2399 
2400 /*
2401  * Update LPCR values in kvm->arch and in vcores.
2402  * Caller must hold kvm->lock.
2403  */
2404 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
2405 {
2406 	long int i;
2407 	u32 cores_done = 0;
2408 
2409 	if ((kvm->arch.lpcr & mask) == lpcr)
2410 		return;
2411 
2412 	kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
2413 
2414 	for (i = 0; i < KVM_MAX_VCORES; ++i) {
2415 		struct kvmppc_vcore *vc = kvm->arch.vcores[i];
2416 		if (!vc)
2417 			continue;
2418 		spin_lock(&vc->lock);
2419 		vc->lpcr = (vc->lpcr & ~mask) | lpcr;
2420 		spin_unlock(&vc->lock);
2421 		if (++cores_done >= kvm->arch.online_vcores)
2422 			break;
2423 	}
2424 }
2425 
2426 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
2427 {
2428 	return;
2429 }
2430 
2431 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
2432 {
2433 	int err = 0;
2434 	struct kvm *kvm = vcpu->kvm;
2435 	unsigned long hva;
2436 	struct kvm_memory_slot *memslot;
2437 	struct vm_area_struct *vma;
2438 	unsigned long lpcr = 0, senc;
2439 	unsigned long psize, porder;
2440 	int srcu_idx;
2441 
2442 	mutex_lock(&kvm->lock);
2443 	if (kvm->arch.hpte_setup_done)
2444 		goto out;	/* another vcpu beat us to it */
2445 
2446 	/* Allocate hashed page table (if not done already) and reset it */
2447 	if (!kvm->arch.hpt_virt) {
2448 		err = kvmppc_alloc_hpt(kvm, NULL);
2449 		if (err) {
2450 			pr_err("KVM: Couldn't alloc HPT\n");
2451 			goto out;
2452 		}
2453 	}
2454 
2455 	/* Look up the memslot for guest physical address 0 */
2456 	srcu_idx = srcu_read_lock(&kvm->srcu);
2457 	memslot = gfn_to_memslot(kvm, 0);
2458 
2459 	/* We must have some memory at 0 by now */
2460 	err = -EINVAL;
2461 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
2462 		goto out_srcu;
2463 
2464 	/* Look up the VMA for the start of this memory slot */
2465 	hva = memslot->userspace_addr;
2466 	down_read(&current->mm->mmap_sem);
2467 	vma = find_vma(current->mm, hva);
2468 	if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
2469 		goto up_out;
2470 
2471 	psize = vma_kernel_pagesize(vma);
2472 	porder = __ilog2(psize);
2473 
2474 	up_read(&current->mm->mmap_sem);
2475 
2476 	/* We can handle 4k, 64k or 16M pages in the VRMA */
2477 	err = -EINVAL;
2478 	if (!(psize == 0x1000 || psize == 0x10000 ||
2479 	      psize == 0x1000000))
2480 		goto out_srcu;
2481 
2482 	/* Update VRMASD field in the LPCR */
2483 	senc = slb_pgsize_encoding(psize);
2484 	kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
2485 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
2486 	/* the -4 is to account for senc values starting at 0x10 */
2487 	lpcr = senc << (LPCR_VRMASD_SH - 4);
2488 
2489 	/* Create HPTEs in the hash page table for the VRMA */
2490 	kvmppc_map_vrma(vcpu, memslot, porder);
2491 
2492 	kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
2493 
2494 	/* Order updates to kvm->arch.lpcr etc. vs. hpte_setup_done */
2495 	smp_wmb();
2496 	kvm->arch.hpte_setup_done = 1;
2497 	err = 0;
2498  out_srcu:
2499 	srcu_read_unlock(&kvm->srcu, srcu_idx);
2500  out:
2501 	mutex_unlock(&kvm->lock);
2502 	return err;
2503 
2504  up_out:
2505 	up_read(&current->mm->mmap_sem);
2506 	goto out_srcu;
2507 }
2508 
2509 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
2510 {
2511 	unsigned long lpcr, lpid;
2512 	char buf[32];
2513 
2514 	/* Allocate the guest's logical partition ID */
2515 
2516 	lpid = kvmppc_alloc_lpid();
2517 	if ((long)lpid < 0)
2518 		return -ENOMEM;
2519 	kvm->arch.lpid = lpid;
2520 
2521 	/*
2522 	 * Since we don't flush the TLB when tearing down a VM,
2523 	 * and this lpid might have previously been used,
2524 	 * make sure we flush on each core before running the new VM.
2525 	 */
2526 	cpumask_setall(&kvm->arch.need_tlb_flush);
2527 
2528 	/* Start out with the default set of hcalls enabled */
2529 	memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
2530 	       sizeof(kvm->arch.enabled_hcalls));
2531 
2532 	kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
2533 
2534 	/* Init LPCR for virtual RMA mode */
2535 	kvm->arch.host_lpid = mfspr(SPRN_LPID);
2536 	kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
2537 	lpcr &= LPCR_PECE | LPCR_LPES;
2538 	lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
2539 		LPCR_VPM0 | LPCR_VPM1;
2540 	kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
2541 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
2542 	/* On POWER8 turn on online bit to enable PURR/SPURR */
2543 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
2544 		lpcr |= LPCR_ONL;
2545 	kvm->arch.lpcr = lpcr;
2546 
2547 	/*
2548 	 * Track that we now have a HV mode VM active. This blocks secondary
2549 	 * CPU threads from coming online.
2550 	 */
2551 	kvm_hv_vm_activated();
2552 
2553 	/*
2554 	 * Create a debugfs directory for the VM
2555 	 */
2556 	snprintf(buf, sizeof(buf), "vm%d", current->pid);
2557 	kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
2558 	if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
2559 		kvmppc_mmu_debugfs_init(kvm);
2560 
2561 	return 0;
2562 }
2563 
2564 static void kvmppc_free_vcores(struct kvm *kvm)
2565 {
2566 	long int i;
2567 
2568 	for (i = 0; i < KVM_MAX_VCORES; ++i) {
2569 		if (kvm->arch.vcores[i] && kvm->arch.vcores[i]->mpp_buffer) {
2570 			struct kvmppc_vcore *vc = kvm->arch.vcores[i];
2571 			free_pages((unsigned long)vc->mpp_buffer,
2572 				   MPP_BUFFER_ORDER);
2573 		}
2574 		kfree(kvm->arch.vcores[i]);
2575 	}
2576 	kvm->arch.online_vcores = 0;
2577 }
2578 
2579 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
2580 {
2581 	debugfs_remove_recursive(kvm->arch.debugfs_dir);
2582 
2583 	kvm_hv_vm_deactivated();
2584 
2585 	kvmppc_free_vcores(kvm);
2586 
2587 	kvmppc_free_hpt(kvm);
2588 }
2589 
2590 /* We don't need to emulate any privileged instructions or dcbz */
2591 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
2592 				     unsigned int inst, int *advance)
2593 {
2594 	return EMULATE_FAIL;
2595 }
2596 
2597 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
2598 					ulong spr_val)
2599 {
2600 	return EMULATE_FAIL;
2601 }
2602 
2603 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
2604 					ulong *spr_val)
2605 {
2606 	return EMULATE_FAIL;
2607 }
2608 
2609 static int kvmppc_core_check_processor_compat_hv(void)
2610 {
2611 	if (!cpu_has_feature(CPU_FTR_HVMODE) ||
2612 	    !cpu_has_feature(CPU_FTR_ARCH_206))
2613 		return -EIO;
2614 	return 0;
2615 }
2616 
2617 static long kvm_arch_vm_ioctl_hv(struct file *filp,
2618 				 unsigned int ioctl, unsigned long arg)
2619 {
2620 	struct kvm *kvm __maybe_unused = filp->private_data;
2621 	void __user *argp = (void __user *)arg;
2622 	long r;
2623 
2624 	switch (ioctl) {
2625 
2626 	case KVM_PPC_ALLOCATE_HTAB: {
2627 		u32 htab_order;
2628 
2629 		r = -EFAULT;
2630 		if (get_user(htab_order, (u32 __user *)argp))
2631 			break;
2632 		r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
2633 		if (r)
2634 			break;
2635 		r = -EFAULT;
2636 		if (put_user(htab_order, (u32 __user *)argp))
2637 			break;
2638 		r = 0;
2639 		break;
2640 	}
2641 
2642 	case KVM_PPC_GET_HTAB_FD: {
2643 		struct kvm_get_htab_fd ghf;
2644 
2645 		r = -EFAULT;
2646 		if (copy_from_user(&ghf, argp, sizeof(ghf)))
2647 			break;
2648 		r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
2649 		break;
2650 	}
2651 
2652 	default:
2653 		r = -ENOTTY;
2654 	}
2655 
2656 	return r;
2657 }
2658 
2659 /*
2660  * List of hcall numbers to enable by default.
2661  * For compatibility with old userspace, we enable by default
2662  * all hcalls that were implemented before the hcall-enabling
2663  * facility was added.  Note this list should not include H_RTAS.
2664  */
2665 static unsigned int default_hcall_list[] = {
2666 	H_REMOVE,
2667 	H_ENTER,
2668 	H_READ,
2669 	H_PROTECT,
2670 	H_BULK_REMOVE,
2671 	H_GET_TCE,
2672 	H_PUT_TCE,
2673 	H_SET_DABR,
2674 	H_SET_XDABR,
2675 	H_CEDE,
2676 	H_PROD,
2677 	H_CONFER,
2678 	H_REGISTER_VPA,
2679 #ifdef CONFIG_KVM_XICS
2680 	H_EOI,
2681 	H_CPPR,
2682 	H_IPI,
2683 	H_IPOLL,
2684 	H_XIRR,
2685 	H_XIRR_X,
2686 #endif
2687 	0
2688 };
2689 
2690 static void init_default_hcalls(void)
2691 {
2692 	int i;
2693 	unsigned int hcall;
2694 
2695 	for (i = 0; default_hcall_list[i]; ++i) {
2696 		hcall = default_hcall_list[i];
2697 		WARN_ON(!kvmppc_hcall_impl_hv(hcall));
2698 		__set_bit(hcall / 4, default_enabled_hcalls);
2699 	}
2700 }
2701 
2702 static struct kvmppc_ops kvm_ops_hv = {
2703 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
2704 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
2705 	.get_one_reg = kvmppc_get_one_reg_hv,
2706 	.set_one_reg = kvmppc_set_one_reg_hv,
2707 	.vcpu_load   = kvmppc_core_vcpu_load_hv,
2708 	.vcpu_put    = kvmppc_core_vcpu_put_hv,
2709 	.set_msr     = kvmppc_set_msr_hv,
2710 	.vcpu_run    = kvmppc_vcpu_run_hv,
2711 	.vcpu_create = kvmppc_core_vcpu_create_hv,
2712 	.vcpu_free   = kvmppc_core_vcpu_free_hv,
2713 	.check_requests = kvmppc_core_check_requests_hv,
2714 	.get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
2715 	.flush_memslot  = kvmppc_core_flush_memslot_hv,
2716 	.prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
2717 	.commit_memory_region  = kvmppc_core_commit_memory_region_hv,
2718 	.unmap_hva = kvm_unmap_hva_hv,
2719 	.unmap_hva_range = kvm_unmap_hva_range_hv,
2720 	.age_hva  = kvm_age_hva_hv,
2721 	.test_age_hva = kvm_test_age_hva_hv,
2722 	.set_spte_hva = kvm_set_spte_hva_hv,
2723 	.mmu_destroy  = kvmppc_mmu_destroy_hv,
2724 	.free_memslot = kvmppc_core_free_memslot_hv,
2725 	.create_memslot = kvmppc_core_create_memslot_hv,
2726 	.init_vm =  kvmppc_core_init_vm_hv,
2727 	.destroy_vm = kvmppc_core_destroy_vm_hv,
2728 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
2729 	.emulate_op = kvmppc_core_emulate_op_hv,
2730 	.emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
2731 	.emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
2732 	.fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
2733 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
2734 	.hcall_implemented = kvmppc_hcall_impl_hv,
2735 };
2736 
2737 static int kvmppc_book3s_init_hv(void)
2738 {
2739 	int r;
2740 	/*
2741 	 * FIXME!! Do we need to check on all cpus ?
2742 	 */
2743 	r = kvmppc_core_check_processor_compat_hv();
2744 	if (r < 0)
2745 		return -ENODEV;
2746 
2747 	kvm_ops_hv.owner = THIS_MODULE;
2748 	kvmppc_hv_ops = &kvm_ops_hv;
2749 
2750 	init_default_hcalls();
2751 
2752 	r = kvmppc_mmu_hv_init();
2753 	return r;
2754 }
2755 
2756 static void kvmppc_book3s_exit_hv(void)
2757 {
2758 	kvmppc_hv_ops = NULL;
2759 }
2760 
2761 module_init(kvmppc_book3s_init_hv);
2762 module_exit(kvmppc_book3s_exit_hv);
2763 MODULE_LICENSE("GPL");
2764 MODULE_ALIAS_MISCDEV(KVM_MINOR);
2765 MODULE_ALIAS("devname:kvm");
2766