xref: /openbmc/linux/arch/powerpc/kvm/book3s_hv.c (revision 77d84ff8)
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 
35 #include <asm/reg.h>
36 #include <asm/cputable.h>
37 #include <asm/cacheflush.h>
38 #include <asm/tlbflush.h>
39 #include <asm/uaccess.h>
40 #include <asm/io.h>
41 #include <asm/kvm_ppc.h>
42 #include <asm/kvm_book3s.h>
43 #include <asm/mmu_context.h>
44 #include <asm/lppaca.h>
45 #include <asm/processor.h>
46 #include <asm/cputhreads.h>
47 #include <asm/page.h>
48 #include <asm/hvcall.h>
49 #include <asm/switch_to.h>
50 #include <asm/smp.h>
51 #include <linux/gfp.h>
52 #include <linux/vmalloc.h>
53 #include <linux/highmem.h>
54 #include <linux/hugetlb.h>
55 #include <linux/module.h>
56 
57 #include "book3s.h"
58 
59 /* #define EXIT_DEBUG */
60 /* #define EXIT_DEBUG_SIMPLE */
61 /* #define EXIT_DEBUG_INT */
62 
63 /* Used to indicate that a guest page fault needs to be handled */
64 #define RESUME_PAGE_FAULT	(RESUME_GUEST | RESUME_FLAG_ARCH1)
65 
66 /* Used as a "null" value for timebase values */
67 #define TB_NIL	(~(u64)0)
68 
69 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
70 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
71 
72 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
73 {
74 	int me;
75 	int cpu = vcpu->cpu;
76 	wait_queue_head_t *wqp;
77 
78 	wqp = kvm_arch_vcpu_wq(vcpu);
79 	if (waitqueue_active(wqp)) {
80 		wake_up_interruptible(wqp);
81 		++vcpu->stat.halt_wakeup;
82 	}
83 
84 	me = get_cpu();
85 
86 	/* CPU points to the first thread of the core */
87 	if (cpu != me && cpu >= 0 && cpu < nr_cpu_ids) {
88 		int real_cpu = cpu + vcpu->arch.ptid;
89 		if (paca[real_cpu].kvm_hstate.xics_phys)
90 			xics_wake_cpu(real_cpu);
91 		else if (cpu_online(cpu))
92 			smp_send_reschedule(cpu);
93 	}
94 	put_cpu();
95 }
96 
97 /*
98  * We use the vcpu_load/put functions to measure stolen time.
99  * Stolen time is counted as time when either the vcpu is able to
100  * run as part of a virtual core, but the task running the vcore
101  * is preempted or sleeping, or when the vcpu needs something done
102  * in the kernel by the task running the vcpu, but that task is
103  * preempted or sleeping.  Those two things have to be counted
104  * separately, since one of the vcpu tasks will take on the job
105  * of running the core, and the other vcpu tasks in the vcore will
106  * sleep waiting for it to do that, but that sleep shouldn't count
107  * as stolen time.
108  *
109  * Hence we accumulate stolen time when the vcpu can run as part of
110  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
111  * needs its task to do other things in the kernel (for example,
112  * service a page fault) in busy_stolen.  We don't accumulate
113  * stolen time for a vcore when it is inactive, or for a vcpu
114  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
115  * a misnomer; it means that the vcpu task is not executing in
116  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
117  * the kernel.  We don't have any way of dividing up that time
118  * between time that the vcpu is genuinely stopped, time that
119  * the task is actively working on behalf of the vcpu, and time
120  * that the task is preempted, so we don't count any of it as
121  * stolen.
122  *
123  * Updates to busy_stolen are protected by arch.tbacct_lock;
124  * updates to vc->stolen_tb are protected by the arch.tbacct_lock
125  * of the vcpu that has taken responsibility for running the vcore
126  * (i.e. vc->runner).  The stolen times are measured in units of
127  * timebase ticks.  (Note that the != TB_NIL checks below are
128  * purely defensive; they should never fail.)
129  */
130 
131 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
132 {
133 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
134 
135 	spin_lock(&vcpu->arch.tbacct_lock);
136 	if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE &&
137 	    vc->preempt_tb != TB_NIL) {
138 		vc->stolen_tb += mftb() - vc->preempt_tb;
139 		vc->preempt_tb = TB_NIL;
140 	}
141 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
142 	    vcpu->arch.busy_preempt != TB_NIL) {
143 		vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
144 		vcpu->arch.busy_preempt = TB_NIL;
145 	}
146 	spin_unlock(&vcpu->arch.tbacct_lock);
147 }
148 
149 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
150 {
151 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
152 
153 	spin_lock(&vcpu->arch.tbacct_lock);
154 	if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE)
155 		vc->preempt_tb = mftb();
156 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
157 		vcpu->arch.busy_preempt = mftb();
158 	spin_unlock(&vcpu->arch.tbacct_lock);
159 }
160 
161 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
162 {
163 	vcpu->arch.shregs.msr = msr;
164 	kvmppc_end_cede(vcpu);
165 }
166 
167 void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
168 {
169 	vcpu->arch.pvr = pvr;
170 }
171 
172 int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
173 {
174 	unsigned long pcr = 0;
175 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
176 
177 	if (arch_compat) {
178 		if (!cpu_has_feature(CPU_FTR_ARCH_206))
179 			return -EINVAL;	/* 970 has no compat mode support */
180 
181 		switch (arch_compat) {
182 		case PVR_ARCH_205:
183 			pcr = PCR_ARCH_205;
184 			break;
185 		case PVR_ARCH_206:
186 		case PVR_ARCH_206p:
187 			break;
188 		default:
189 			return -EINVAL;
190 		}
191 	}
192 
193 	spin_lock(&vc->lock);
194 	vc->arch_compat = arch_compat;
195 	vc->pcr = pcr;
196 	spin_unlock(&vc->lock);
197 
198 	return 0;
199 }
200 
201 void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
202 {
203 	int r;
204 
205 	pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
206 	pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
207 	       vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
208 	for (r = 0; r < 16; ++r)
209 		pr_err("r%2d = %.16lx  r%d = %.16lx\n",
210 		       r, kvmppc_get_gpr(vcpu, r),
211 		       r+16, kvmppc_get_gpr(vcpu, r+16));
212 	pr_err("ctr = %.16lx  lr  = %.16lx\n",
213 	       vcpu->arch.ctr, vcpu->arch.lr);
214 	pr_err("srr0 = %.16llx srr1 = %.16llx\n",
215 	       vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
216 	pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
217 	       vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
218 	pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
219 	       vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
220 	pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
221 	       vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
222 	pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
223 	pr_err("fault dar = %.16lx dsisr = %.8x\n",
224 	       vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
225 	pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
226 	for (r = 0; r < vcpu->arch.slb_max; ++r)
227 		pr_err("  ESID = %.16llx VSID = %.16llx\n",
228 		       vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
229 	pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
230 	       vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
231 	       vcpu->arch.last_inst);
232 }
233 
234 struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
235 {
236 	int r;
237 	struct kvm_vcpu *v, *ret = NULL;
238 
239 	mutex_lock(&kvm->lock);
240 	kvm_for_each_vcpu(r, v, kvm) {
241 		if (v->vcpu_id == id) {
242 			ret = v;
243 			break;
244 		}
245 	}
246 	mutex_unlock(&kvm->lock);
247 	return ret;
248 }
249 
250 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
251 {
252 	vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
253 	vpa->yield_count = 1;
254 }
255 
256 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
257 		   unsigned long addr, unsigned long len)
258 {
259 	/* check address is cacheline aligned */
260 	if (addr & (L1_CACHE_BYTES - 1))
261 		return -EINVAL;
262 	spin_lock(&vcpu->arch.vpa_update_lock);
263 	if (v->next_gpa != addr || v->len != len) {
264 		v->next_gpa = addr;
265 		v->len = addr ? len : 0;
266 		v->update_pending = 1;
267 	}
268 	spin_unlock(&vcpu->arch.vpa_update_lock);
269 	return 0;
270 }
271 
272 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
273 struct reg_vpa {
274 	u32 dummy;
275 	union {
276 		u16 hword;
277 		u32 word;
278 	} length;
279 };
280 
281 static int vpa_is_registered(struct kvmppc_vpa *vpap)
282 {
283 	if (vpap->update_pending)
284 		return vpap->next_gpa != 0;
285 	return vpap->pinned_addr != NULL;
286 }
287 
288 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
289 				       unsigned long flags,
290 				       unsigned long vcpuid, unsigned long vpa)
291 {
292 	struct kvm *kvm = vcpu->kvm;
293 	unsigned long len, nb;
294 	void *va;
295 	struct kvm_vcpu *tvcpu;
296 	int err;
297 	int subfunc;
298 	struct kvmppc_vpa *vpap;
299 
300 	tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
301 	if (!tvcpu)
302 		return H_PARAMETER;
303 
304 	subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
305 	if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
306 	    subfunc == H_VPA_REG_SLB) {
307 		/* Registering new area - address must be cache-line aligned */
308 		if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
309 			return H_PARAMETER;
310 
311 		/* convert logical addr to kernel addr and read length */
312 		va = kvmppc_pin_guest_page(kvm, vpa, &nb);
313 		if (va == NULL)
314 			return H_PARAMETER;
315 		if (subfunc == H_VPA_REG_VPA)
316 			len = ((struct reg_vpa *)va)->length.hword;
317 		else
318 			len = ((struct reg_vpa *)va)->length.word;
319 		kvmppc_unpin_guest_page(kvm, va, vpa, false);
320 
321 		/* Check length */
322 		if (len > nb || len < sizeof(struct reg_vpa))
323 			return H_PARAMETER;
324 	} else {
325 		vpa = 0;
326 		len = 0;
327 	}
328 
329 	err = H_PARAMETER;
330 	vpap = NULL;
331 	spin_lock(&tvcpu->arch.vpa_update_lock);
332 
333 	switch (subfunc) {
334 	case H_VPA_REG_VPA:		/* register VPA */
335 		if (len < sizeof(struct lppaca))
336 			break;
337 		vpap = &tvcpu->arch.vpa;
338 		err = 0;
339 		break;
340 
341 	case H_VPA_REG_DTL:		/* register DTL */
342 		if (len < sizeof(struct dtl_entry))
343 			break;
344 		len -= len % sizeof(struct dtl_entry);
345 
346 		/* Check that they have previously registered a VPA */
347 		err = H_RESOURCE;
348 		if (!vpa_is_registered(&tvcpu->arch.vpa))
349 			break;
350 
351 		vpap = &tvcpu->arch.dtl;
352 		err = 0;
353 		break;
354 
355 	case H_VPA_REG_SLB:		/* register SLB shadow buffer */
356 		/* Check that they have previously registered a VPA */
357 		err = H_RESOURCE;
358 		if (!vpa_is_registered(&tvcpu->arch.vpa))
359 			break;
360 
361 		vpap = &tvcpu->arch.slb_shadow;
362 		err = 0;
363 		break;
364 
365 	case H_VPA_DEREG_VPA:		/* deregister VPA */
366 		/* Check they don't still have a DTL or SLB buf registered */
367 		err = H_RESOURCE;
368 		if (vpa_is_registered(&tvcpu->arch.dtl) ||
369 		    vpa_is_registered(&tvcpu->arch.slb_shadow))
370 			break;
371 
372 		vpap = &tvcpu->arch.vpa;
373 		err = 0;
374 		break;
375 
376 	case H_VPA_DEREG_DTL:		/* deregister DTL */
377 		vpap = &tvcpu->arch.dtl;
378 		err = 0;
379 		break;
380 
381 	case H_VPA_DEREG_SLB:		/* deregister SLB shadow buffer */
382 		vpap = &tvcpu->arch.slb_shadow;
383 		err = 0;
384 		break;
385 	}
386 
387 	if (vpap) {
388 		vpap->next_gpa = vpa;
389 		vpap->len = len;
390 		vpap->update_pending = 1;
391 	}
392 
393 	spin_unlock(&tvcpu->arch.vpa_update_lock);
394 
395 	return err;
396 }
397 
398 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
399 {
400 	struct kvm *kvm = vcpu->kvm;
401 	void *va;
402 	unsigned long nb;
403 	unsigned long gpa;
404 
405 	/*
406 	 * We need to pin the page pointed to by vpap->next_gpa,
407 	 * but we can't call kvmppc_pin_guest_page under the lock
408 	 * as it does get_user_pages() and down_read().  So we
409 	 * have to drop the lock, pin the page, then get the lock
410 	 * again and check that a new area didn't get registered
411 	 * in the meantime.
412 	 */
413 	for (;;) {
414 		gpa = vpap->next_gpa;
415 		spin_unlock(&vcpu->arch.vpa_update_lock);
416 		va = NULL;
417 		nb = 0;
418 		if (gpa)
419 			va = kvmppc_pin_guest_page(kvm, gpa, &nb);
420 		spin_lock(&vcpu->arch.vpa_update_lock);
421 		if (gpa == vpap->next_gpa)
422 			break;
423 		/* sigh... unpin that one and try again */
424 		if (va)
425 			kvmppc_unpin_guest_page(kvm, va, gpa, false);
426 	}
427 
428 	vpap->update_pending = 0;
429 	if (va && nb < vpap->len) {
430 		/*
431 		 * If it's now too short, it must be that userspace
432 		 * has changed the mappings underlying guest memory,
433 		 * so unregister the region.
434 		 */
435 		kvmppc_unpin_guest_page(kvm, va, gpa, false);
436 		va = NULL;
437 	}
438 	if (vpap->pinned_addr)
439 		kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
440 					vpap->dirty);
441 	vpap->gpa = gpa;
442 	vpap->pinned_addr = va;
443 	vpap->dirty = false;
444 	if (va)
445 		vpap->pinned_end = va + vpap->len;
446 }
447 
448 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
449 {
450 	if (!(vcpu->arch.vpa.update_pending ||
451 	      vcpu->arch.slb_shadow.update_pending ||
452 	      vcpu->arch.dtl.update_pending))
453 		return;
454 
455 	spin_lock(&vcpu->arch.vpa_update_lock);
456 	if (vcpu->arch.vpa.update_pending) {
457 		kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
458 		if (vcpu->arch.vpa.pinned_addr)
459 			init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
460 	}
461 	if (vcpu->arch.dtl.update_pending) {
462 		kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
463 		vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
464 		vcpu->arch.dtl_index = 0;
465 	}
466 	if (vcpu->arch.slb_shadow.update_pending)
467 		kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
468 	spin_unlock(&vcpu->arch.vpa_update_lock);
469 }
470 
471 /*
472  * Return the accumulated stolen time for the vcore up until `now'.
473  * The caller should hold the vcore lock.
474  */
475 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
476 {
477 	u64 p;
478 
479 	/*
480 	 * If we are the task running the vcore, then since we hold
481 	 * the vcore lock, we can't be preempted, so stolen_tb/preempt_tb
482 	 * can't be updated, so we don't need the tbacct_lock.
483 	 * If the vcore is inactive, it can't become active (since we
484 	 * hold the vcore lock), so the vcpu load/put functions won't
485 	 * update stolen_tb/preempt_tb, and we don't need tbacct_lock.
486 	 */
487 	if (vc->vcore_state != VCORE_INACTIVE &&
488 	    vc->runner->arch.run_task != current) {
489 		spin_lock(&vc->runner->arch.tbacct_lock);
490 		p = vc->stolen_tb;
491 		if (vc->preempt_tb != TB_NIL)
492 			p += now - vc->preempt_tb;
493 		spin_unlock(&vc->runner->arch.tbacct_lock);
494 	} else {
495 		p = vc->stolen_tb;
496 	}
497 	return p;
498 }
499 
500 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
501 				    struct kvmppc_vcore *vc)
502 {
503 	struct dtl_entry *dt;
504 	struct lppaca *vpa;
505 	unsigned long stolen;
506 	unsigned long core_stolen;
507 	u64 now;
508 
509 	dt = vcpu->arch.dtl_ptr;
510 	vpa = vcpu->arch.vpa.pinned_addr;
511 	now = mftb();
512 	core_stolen = vcore_stolen_time(vc, now);
513 	stolen = core_stolen - vcpu->arch.stolen_logged;
514 	vcpu->arch.stolen_logged = core_stolen;
515 	spin_lock(&vcpu->arch.tbacct_lock);
516 	stolen += vcpu->arch.busy_stolen;
517 	vcpu->arch.busy_stolen = 0;
518 	spin_unlock(&vcpu->arch.tbacct_lock);
519 	if (!dt || !vpa)
520 		return;
521 	memset(dt, 0, sizeof(struct dtl_entry));
522 	dt->dispatch_reason = 7;
523 	dt->processor_id = vc->pcpu + vcpu->arch.ptid;
524 	dt->timebase = now + vc->tb_offset;
525 	dt->enqueue_to_dispatch_time = stolen;
526 	dt->srr0 = kvmppc_get_pc(vcpu);
527 	dt->srr1 = vcpu->arch.shregs.msr;
528 	++dt;
529 	if (dt == vcpu->arch.dtl.pinned_end)
530 		dt = vcpu->arch.dtl.pinned_addr;
531 	vcpu->arch.dtl_ptr = dt;
532 	/* order writing *dt vs. writing vpa->dtl_idx */
533 	smp_wmb();
534 	vpa->dtl_idx = ++vcpu->arch.dtl_index;
535 	vcpu->arch.dtl.dirty = true;
536 }
537 
538 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
539 {
540 	unsigned long req = kvmppc_get_gpr(vcpu, 3);
541 	unsigned long target, ret = H_SUCCESS;
542 	struct kvm_vcpu *tvcpu;
543 	int idx, rc;
544 
545 	switch (req) {
546 	case H_ENTER:
547 		idx = srcu_read_lock(&vcpu->kvm->srcu);
548 		ret = kvmppc_virtmode_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
549 					      kvmppc_get_gpr(vcpu, 5),
550 					      kvmppc_get_gpr(vcpu, 6),
551 					      kvmppc_get_gpr(vcpu, 7));
552 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
553 		break;
554 	case H_CEDE:
555 		break;
556 	case H_PROD:
557 		target = kvmppc_get_gpr(vcpu, 4);
558 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
559 		if (!tvcpu) {
560 			ret = H_PARAMETER;
561 			break;
562 		}
563 		tvcpu->arch.prodded = 1;
564 		smp_mb();
565 		if (vcpu->arch.ceded) {
566 			if (waitqueue_active(&vcpu->wq)) {
567 				wake_up_interruptible(&vcpu->wq);
568 				vcpu->stat.halt_wakeup++;
569 			}
570 		}
571 		break;
572 	case H_CONFER:
573 		target = kvmppc_get_gpr(vcpu, 4);
574 		if (target == -1)
575 			break;
576 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
577 		if (!tvcpu) {
578 			ret = H_PARAMETER;
579 			break;
580 		}
581 		kvm_vcpu_yield_to(tvcpu);
582 		break;
583 	case H_REGISTER_VPA:
584 		ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
585 					kvmppc_get_gpr(vcpu, 5),
586 					kvmppc_get_gpr(vcpu, 6));
587 		break;
588 	case H_RTAS:
589 		if (list_empty(&vcpu->kvm->arch.rtas_tokens))
590 			return RESUME_HOST;
591 
592 		rc = kvmppc_rtas_hcall(vcpu);
593 
594 		if (rc == -ENOENT)
595 			return RESUME_HOST;
596 		else if (rc == 0)
597 			break;
598 
599 		/* Send the error out to userspace via KVM_RUN */
600 		return rc;
601 
602 	case H_XIRR:
603 	case H_CPPR:
604 	case H_EOI:
605 	case H_IPI:
606 	case H_IPOLL:
607 	case H_XIRR_X:
608 		if (kvmppc_xics_enabled(vcpu)) {
609 			ret = kvmppc_xics_hcall(vcpu, req);
610 			break;
611 		} /* fallthrough */
612 	default:
613 		return RESUME_HOST;
614 	}
615 	kvmppc_set_gpr(vcpu, 3, ret);
616 	vcpu->arch.hcall_needed = 0;
617 	return RESUME_GUEST;
618 }
619 
620 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
621 				 struct task_struct *tsk)
622 {
623 	int r = RESUME_HOST;
624 
625 	vcpu->stat.sum_exits++;
626 
627 	run->exit_reason = KVM_EXIT_UNKNOWN;
628 	run->ready_for_interrupt_injection = 1;
629 	switch (vcpu->arch.trap) {
630 	/* We're good on these - the host merely wanted to get our attention */
631 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
632 		vcpu->stat.dec_exits++;
633 		r = RESUME_GUEST;
634 		break;
635 	case BOOK3S_INTERRUPT_EXTERNAL:
636 		vcpu->stat.ext_intr_exits++;
637 		r = RESUME_GUEST;
638 		break;
639 	case BOOK3S_INTERRUPT_PERFMON:
640 		r = RESUME_GUEST;
641 		break;
642 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
643 		/*
644 		 * Deliver a machine check interrupt to the guest.
645 		 * We have to do this, even if the host has handled the
646 		 * machine check, because machine checks use SRR0/1 and
647 		 * the interrupt might have trashed guest state in them.
648 		 */
649 		kvmppc_book3s_queue_irqprio(vcpu,
650 					    BOOK3S_INTERRUPT_MACHINE_CHECK);
651 		r = RESUME_GUEST;
652 		break;
653 	case BOOK3S_INTERRUPT_PROGRAM:
654 	{
655 		ulong flags;
656 		/*
657 		 * Normally program interrupts are delivered directly
658 		 * to the guest by the hardware, but we can get here
659 		 * as a result of a hypervisor emulation interrupt
660 		 * (e40) getting turned into a 700 by BML RTAS.
661 		 */
662 		flags = vcpu->arch.shregs.msr & 0x1f0000ull;
663 		kvmppc_core_queue_program(vcpu, flags);
664 		r = RESUME_GUEST;
665 		break;
666 	}
667 	case BOOK3S_INTERRUPT_SYSCALL:
668 	{
669 		/* hcall - punt to userspace */
670 		int i;
671 
672 		if (vcpu->arch.shregs.msr & MSR_PR) {
673 			/* sc 1 from userspace - reflect to guest syscall */
674 			kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_SYSCALL);
675 			r = RESUME_GUEST;
676 			break;
677 		}
678 		run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
679 		for (i = 0; i < 9; ++i)
680 			run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
681 		run->exit_reason = KVM_EXIT_PAPR_HCALL;
682 		vcpu->arch.hcall_needed = 1;
683 		r = RESUME_HOST;
684 		break;
685 	}
686 	/*
687 	 * We get these next two if the guest accesses a page which it thinks
688 	 * it has mapped but which is not actually present, either because
689 	 * it is for an emulated I/O device or because the corresonding
690 	 * host page has been paged out.  Any other HDSI/HISI interrupts
691 	 * have been handled already.
692 	 */
693 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
694 		r = RESUME_PAGE_FAULT;
695 		break;
696 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
697 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
698 		vcpu->arch.fault_dsisr = 0;
699 		r = RESUME_PAGE_FAULT;
700 		break;
701 	/*
702 	 * This occurs if the guest executes an illegal instruction.
703 	 * We just generate a program interrupt to the guest, since
704 	 * we don't emulate any guest instructions at this stage.
705 	 */
706 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
707 		kvmppc_core_queue_program(vcpu, 0x80000);
708 		r = RESUME_GUEST;
709 		break;
710 	default:
711 		kvmppc_dump_regs(vcpu);
712 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
713 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
714 			vcpu->arch.shregs.msr);
715 		run->hw.hardware_exit_reason = vcpu->arch.trap;
716 		r = RESUME_HOST;
717 		break;
718 	}
719 
720 	return r;
721 }
722 
723 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
724 					    struct kvm_sregs *sregs)
725 {
726 	int i;
727 
728 	memset(sregs, 0, sizeof(struct kvm_sregs));
729 	sregs->pvr = vcpu->arch.pvr;
730 	for (i = 0; i < vcpu->arch.slb_max; i++) {
731 		sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
732 		sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
733 	}
734 
735 	return 0;
736 }
737 
738 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
739 					    struct kvm_sregs *sregs)
740 {
741 	int i, j;
742 
743 	kvmppc_set_pvr_hv(vcpu, sregs->pvr);
744 
745 	j = 0;
746 	for (i = 0; i < vcpu->arch.slb_nr; i++) {
747 		if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
748 			vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
749 			vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
750 			++j;
751 		}
752 	}
753 	vcpu->arch.slb_max = j;
754 
755 	return 0;
756 }
757 
758 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr)
759 {
760 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
761 	u64 mask;
762 
763 	spin_lock(&vc->lock);
764 	/*
765 	 * Userspace can only modify DPFD (default prefetch depth),
766 	 * ILE (interrupt little-endian) and TC (translation control).
767 	 */
768 	mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
769 	vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
770 	spin_unlock(&vc->lock);
771 }
772 
773 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
774 				 union kvmppc_one_reg *val)
775 {
776 	int r = 0;
777 	long int i;
778 
779 	switch (id) {
780 	case KVM_REG_PPC_HIOR:
781 		*val = get_reg_val(id, 0);
782 		break;
783 	case KVM_REG_PPC_DABR:
784 		*val = get_reg_val(id, vcpu->arch.dabr);
785 		break;
786 	case KVM_REG_PPC_DSCR:
787 		*val = get_reg_val(id, vcpu->arch.dscr);
788 		break;
789 	case KVM_REG_PPC_PURR:
790 		*val = get_reg_val(id, vcpu->arch.purr);
791 		break;
792 	case KVM_REG_PPC_SPURR:
793 		*val = get_reg_val(id, vcpu->arch.spurr);
794 		break;
795 	case KVM_REG_PPC_AMR:
796 		*val = get_reg_val(id, vcpu->arch.amr);
797 		break;
798 	case KVM_REG_PPC_UAMOR:
799 		*val = get_reg_val(id, vcpu->arch.uamor);
800 		break;
801 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRA:
802 		i = id - KVM_REG_PPC_MMCR0;
803 		*val = get_reg_val(id, vcpu->arch.mmcr[i]);
804 		break;
805 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
806 		i = id - KVM_REG_PPC_PMC1;
807 		*val = get_reg_val(id, vcpu->arch.pmc[i]);
808 		break;
809 	case KVM_REG_PPC_SIAR:
810 		*val = get_reg_val(id, vcpu->arch.siar);
811 		break;
812 	case KVM_REG_PPC_SDAR:
813 		*val = get_reg_val(id, vcpu->arch.sdar);
814 		break;
815 #ifdef CONFIG_VSX
816 	case KVM_REG_PPC_FPR0 ... KVM_REG_PPC_FPR31:
817 		if (cpu_has_feature(CPU_FTR_VSX)) {
818 			/* VSX => FP reg i is stored in arch.vsr[2*i] */
819 			long int i = id - KVM_REG_PPC_FPR0;
820 			*val = get_reg_val(id, vcpu->arch.vsr[2 * i]);
821 		} else {
822 			/* let generic code handle it */
823 			r = -EINVAL;
824 		}
825 		break;
826 	case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31:
827 		if (cpu_has_feature(CPU_FTR_VSX)) {
828 			long int i = id - KVM_REG_PPC_VSR0;
829 			val->vsxval[0] = vcpu->arch.vsr[2 * i];
830 			val->vsxval[1] = vcpu->arch.vsr[2 * i + 1];
831 		} else {
832 			r = -ENXIO;
833 		}
834 		break;
835 #endif /* CONFIG_VSX */
836 	case KVM_REG_PPC_VPA_ADDR:
837 		spin_lock(&vcpu->arch.vpa_update_lock);
838 		*val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
839 		spin_unlock(&vcpu->arch.vpa_update_lock);
840 		break;
841 	case KVM_REG_PPC_VPA_SLB:
842 		spin_lock(&vcpu->arch.vpa_update_lock);
843 		val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
844 		val->vpaval.length = vcpu->arch.slb_shadow.len;
845 		spin_unlock(&vcpu->arch.vpa_update_lock);
846 		break;
847 	case KVM_REG_PPC_VPA_DTL:
848 		spin_lock(&vcpu->arch.vpa_update_lock);
849 		val->vpaval.addr = vcpu->arch.dtl.next_gpa;
850 		val->vpaval.length = vcpu->arch.dtl.len;
851 		spin_unlock(&vcpu->arch.vpa_update_lock);
852 		break;
853 	case KVM_REG_PPC_TB_OFFSET:
854 		*val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
855 		break;
856 	case KVM_REG_PPC_LPCR:
857 		*val = get_reg_val(id, vcpu->arch.vcore->lpcr);
858 		break;
859 	case KVM_REG_PPC_PPR:
860 		*val = get_reg_val(id, vcpu->arch.ppr);
861 		break;
862 	case KVM_REG_PPC_ARCH_COMPAT:
863 		*val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
864 		break;
865 	default:
866 		r = -EINVAL;
867 		break;
868 	}
869 
870 	return r;
871 }
872 
873 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
874 				 union kvmppc_one_reg *val)
875 {
876 	int r = 0;
877 	long int i;
878 	unsigned long addr, len;
879 
880 	switch (id) {
881 	case KVM_REG_PPC_HIOR:
882 		/* Only allow this to be set to zero */
883 		if (set_reg_val(id, *val))
884 			r = -EINVAL;
885 		break;
886 	case KVM_REG_PPC_DABR:
887 		vcpu->arch.dabr = set_reg_val(id, *val);
888 		break;
889 	case KVM_REG_PPC_DSCR:
890 		vcpu->arch.dscr = set_reg_val(id, *val);
891 		break;
892 	case KVM_REG_PPC_PURR:
893 		vcpu->arch.purr = set_reg_val(id, *val);
894 		break;
895 	case KVM_REG_PPC_SPURR:
896 		vcpu->arch.spurr = set_reg_val(id, *val);
897 		break;
898 	case KVM_REG_PPC_AMR:
899 		vcpu->arch.amr = set_reg_val(id, *val);
900 		break;
901 	case KVM_REG_PPC_UAMOR:
902 		vcpu->arch.uamor = set_reg_val(id, *val);
903 		break;
904 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRA:
905 		i = id - KVM_REG_PPC_MMCR0;
906 		vcpu->arch.mmcr[i] = set_reg_val(id, *val);
907 		break;
908 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
909 		i = id - KVM_REG_PPC_PMC1;
910 		vcpu->arch.pmc[i] = set_reg_val(id, *val);
911 		break;
912 	case KVM_REG_PPC_SIAR:
913 		vcpu->arch.siar = set_reg_val(id, *val);
914 		break;
915 	case KVM_REG_PPC_SDAR:
916 		vcpu->arch.sdar = set_reg_val(id, *val);
917 		break;
918 #ifdef CONFIG_VSX
919 	case KVM_REG_PPC_FPR0 ... KVM_REG_PPC_FPR31:
920 		if (cpu_has_feature(CPU_FTR_VSX)) {
921 			/* VSX => FP reg i is stored in arch.vsr[2*i] */
922 			long int i = id - KVM_REG_PPC_FPR0;
923 			vcpu->arch.vsr[2 * i] = set_reg_val(id, *val);
924 		} else {
925 			/* let generic code handle it */
926 			r = -EINVAL;
927 		}
928 		break;
929 	case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31:
930 		if (cpu_has_feature(CPU_FTR_VSX)) {
931 			long int i = id - KVM_REG_PPC_VSR0;
932 			vcpu->arch.vsr[2 * i] = val->vsxval[0];
933 			vcpu->arch.vsr[2 * i + 1] = val->vsxval[1];
934 		} else {
935 			r = -ENXIO;
936 		}
937 		break;
938 #endif /* CONFIG_VSX */
939 	case KVM_REG_PPC_VPA_ADDR:
940 		addr = set_reg_val(id, *val);
941 		r = -EINVAL;
942 		if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
943 			      vcpu->arch.dtl.next_gpa))
944 			break;
945 		r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
946 		break;
947 	case KVM_REG_PPC_VPA_SLB:
948 		addr = val->vpaval.addr;
949 		len = val->vpaval.length;
950 		r = -EINVAL;
951 		if (addr && !vcpu->arch.vpa.next_gpa)
952 			break;
953 		r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
954 		break;
955 	case KVM_REG_PPC_VPA_DTL:
956 		addr = val->vpaval.addr;
957 		len = val->vpaval.length;
958 		r = -EINVAL;
959 		if (addr && (len < sizeof(struct dtl_entry) ||
960 			     !vcpu->arch.vpa.next_gpa))
961 			break;
962 		len -= len % sizeof(struct dtl_entry);
963 		r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
964 		break;
965 	case KVM_REG_PPC_TB_OFFSET:
966 		/* round up to multiple of 2^24 */
967 		vcpu->arch.vcore->tb_offset =
968 			ALIGN(set_reg_val(id, *val), 1UL << 24);
969 		break;
970 	case KVM_REG_PPC_LPCR:
971 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val));
972 		break;
973 	case KVM_REG_PPC_PPR:
974 		vcpu->arch.ppr = set_reg_val(id, *val);
975 		break;
976 	case KVM_REG_PPC_ARCH_COMPAT:
977 		r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
978 		break;
979 	default:
980 		r = -EINVAL;
981 		break;
982 	}
983 
984 	return r;
985 }
986 
987 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
988 						   unsigned int id)
989 {
990 	struct kvm_vcpu *vcpu;
991 	int err = -EINVAL;
992 	int core;
993 	struct kvmppc_vcore *vcore;
994 
995 	core = id / threads_per_core;
996 	if (core >= KVM_MAX_VCORES)
997 		goto out;
998 
999 	err = -ENOMEM;
1000 	vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1001 	if (!vcpu)
1002 		goto out;
1003 
1004 	err = kvm_vcpu_init(vcpu, kvm, id);
1005 	if (err)
1006 		goto free_vcpu;
1007 
1008 	vcpu->arch.shared = &vcpu->arch.shregs;
1009 	vcpu->arch.mmcr[0] = MMCR0_FC;
1010 	vcpu->arch.ctrl = CTRL_RUNLATCH;
1011 	/* default to host PVR, since we can't spoof it */
1012 	kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1013 	spin_lock_init(&vcpu->arch.vpa_update_lock);
1014 	spin_lock_init(&vcpu->arch.tbacct_lock);
1015 	vcpu->arch.busy_preempt = TB_NIL;
1016 
1017 	kvmppc_mmu_book3s_hv_init(vcpu);
1018 
1019 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1020 
1021 	init_waitqueue_head(&vcpu->arch.cpu_run);
1022 
1023 	mutex_lock(&kvm->lock);
1024 	vcore = kvm->arch.vcores[core];
1025 	if (!vcore) {
1026 		vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1027 		if (vcore) {
1028 			INIT_LIST_HEAD(&vcore->runnable_threads);
1029 			spin_lock_init(&vcore->lock);
1030 			init_waitqueue_head(&vcore->wq);
1031 			vcore->preempt_tb = TB_NIL;
1032 			vcore->lpcr = kvm->arch.lpcr;
1033 		}
1034 		kvm->arch.vcores[core] = vcore;
1035 		kvm->arch.online_vcores++;
1036 	}
1037 	mutex_unlock(&kvm->lock);
1038 
1039 	if (!vcore)
1040 		goto free_vcpu;
1041 
1042 	spin_lock(&vcore->lock);
1043 	++vcore->num_threads;
1044 	spin_unlock(&vcore->lock);
1045 	vcpu->arch.vcore = vcore;
1046 
1047 	vcpu->arch.cpu_type = KVM_CPU_3S_64;
1048 	kvmppc_sanity_check(vcpu);
1049 
1050 	return vcpu;
1051 
1052 free_vcpu:
1053 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1054 out:
1055 	return ERR_PTR(err);
1056 }
1057 
1058 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1059 {
1060 	if (vpa->pinned_addr)
1061 		kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1062 					vpa->dirty);
1063 }
1064 
1065 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
1066 {
1067 	spin_lock(&vcpu->arch.vpa_update_lock);
1068 	unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1069 	unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1070 	unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1071 	spin_unlock(&vcpu->arch.vpa_update_lock);
1072 	kvm_vcpu_uninit(vcpu);
1073 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1074 }
1075 
1076 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1077 {
1078 	/* Indicate we want to get back into the guest */
1079 	return 1;
1080 }
1081 
1082 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1083 {
1084 	unsigned long dec_nsec, now;
1085 
1086 	now = get_tb();
1087 	if (now > vcpu->arch.dec_expires) {
1088 		/* decrementer has already gone negative */
1089 		kvmppc_core_queue_dec(vcpu);
1090 		kvmppc_core_prepare_to_enter(vcpu);
1091 		return;
1092 	}
1093 	dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1094 		   / tb_ticks_per_sec;
1095 	hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1096 		      HRTIMER_MODE_REL);
1097 	vcpu->arch.timer_running = 1;
1098 }
1099 
1100 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1101 {
1102 	vcpu->arch.ceded = 0;
1103 	if (vcpu->arch.timer_running) {
1104 		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1105 		vcpu->arch.timer_running = 0;
1106 	}
1107 }
1108 
1109 extern int __kvmppc_vcore_entry(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu);
1110 
1111 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1112 				   struct kvm_vcpu *vcpu)
1113 {
1114 	u64 now;
1115 
1116 	if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1117 		return;
1118 	spin_lock(&vcpu->arch.tbacct_lock);
1119 	now = mftb();
1120 	vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1121 		vcpu->arch.stolen_logged;
1122 	vcpu->arch.busy_preempt = now;
1123 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1124 	spin_unlock(&vcpu->arch.tbacct_lock);
1125 	--vc->n_runnable;
1126 	list_del(&vcpu->arch.run_list);
1127 }
1128 
1129 static int kvmppc_grab_hwthread(int cpu)
1130 {
1131 	struct paca_struct *tpaca;
1132 	long timeout = 1000;
1133 
1134 	tpaca = &paca[cpu];
1135 
1136 	/* Ensure the thread won't go into the kernel if it wakes */
1137 	tpaca->kvm_hstate.hwthread_req = 1;
1138 	tpaca->kvm_hstate.kvm_vcpu = NULL;
1139 
1140 	/*
1141 	 * If the thread is already executing in the kernel (e.g. handling
1142 	 * a stray interrupt), wait for it to get back to nap mode.
1143 	 * The smp_mb() is to ensure that our setting of hwthread_req
1144 	 * is visible before we look at hwthread_state, so if this
1145 	 * races with the code at system_reset_pSeries and the thread
1146 	 * misses our setting of hwthread_req, we are sure to see its
1147 	 * setting of hwthread_state, and vice versa.
1148 	 */
1149 	smp_mb();
1150 	while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1151 		if (--timeout <= 0) {
1152 			pr_err("KVM: couldn't grab cpu %d\n", cpu);
1153 			return -EBUSY;
1154 		}
1155 		udelay(1);
1156 	}
1157 	return 0;
1158 }
1159 
1160 static void kvmppc_release_hwthread(int cpu)
1161 {
1162 	struct paca_struct *tpaca;
1163 
1164 	tpaca = &paca[cpu];
1165 	tpaca->kvm_hstate.hwthread_req = 0;
1166 	tpaca->kvm_hstate.kvm_vcpu = NULL;
1167 }
1168 
1169 static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
1170 {
1171 	int cpu;
1172 	struct paca_struct *tpaca;
1173 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
1174 
1175 	if (vcpu->arch.timer_running) {
1176 		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1177 		vcpu->arch.timer_running = 0;
1178 	}
1179 	cpu = vc->pcpu + vcpu->arch.ptid;
1180 	tpaca = &paca[cpu];
1181 	tpaca->kvm_hstate.kvm_vcpu = vcpu;
1182 	tpaca->kvm_hstate.kvm_vcore = vc;
1183 	tpaca->kvm_hstate.napping = 0;
1184 	vcpu->cpu = vc->pcpu;
1185 	smp_wmb();
1186 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
1187 	if (vcpu->arch.ptid) {
1188 		xics_wake_cpu(cpu);
1189 		++vc->n_woken;
1190 	}
1191 #endif
1192 }
1193 
1194 static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc)
1195 {
1196 	int i;
1197 
1198 	HMT_low();
1199 	i = 0;
1200 	while (vc->nap_count < vc->n_woken) {
1201 		if (++i >= 1000000) {
1202 			pr_err("kvmppc_wait_for_nap timeout %d %d\n",
1203 			       vc->nap_count, vc->n_woken);
1204 			break;
1205 		}
1206 		cpu_relax();
1207 	}
1208 	HMT_medium();
1209 }
1210 
1211 /*
1212  * Check that we are on thread 0 and that any other threads in
1213  * this core are off-line.  Then grab the threads so they can't
1214  * enter the kernel.
1215  */
1216 static int on_primary_thread(void)
1217 {
1218 	int cpu = smp_processor_id();
1219 	int thr = cpu_thread_in_core(cpu);
1220 
1221 	if (thr)
1222 		return 0;
1223 	while (++thr < threads_per_core)
1224 		if (cpu_online(cpu + thr))
1225 			return 0;
1226 
1227 	/* Grab all hw threads so they can't go into the kernel */
1228 	for (thr = 1; thr < threads_per_core; ++thr) {
1229 		if (kvmppc_grab_hwthread(cpu + thr)) {
1230 			/* Couldn't grab one; let the others go */
1231 			do {
1232 				kvmppc_release_hwthread(cpu + thr);
1233 			} while (--thr > 0);
1234 			return 0;
1235 		}
1236 	}
1237 	return 1;
1238 }
1239 
1240 /*
1241  * Run a set of guest threads on a physical core.
1242  * Called with vc->lock held.
1243  */
1244 static void kvmppc_run_core(struct kvmppc_vcore *vc)
1245 {
1246 	struct kvm_vcpu *vcpu, *vcpu0, *vnext;
1247 	long ret;
1248 	u64 now;
1249 	int ptid, i, need_vpa_update;
1250 	int srcu_idx;
1251 	struct kvm_vcpu *vcpus_to_update[threads_per_core];
1252 
1253 	/* don't start if any threads have a signal pending */
1254 	need_vpa_update = 0;
1255 	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1256 		if (signal_pending(vcpu->arch.run_task))
1257 			return;
1258 		if (vcpu->arch.vpa.update_pending ||
1259 		    vcpu->arch.slb_shadow.update_pending ||
1260 		    vcpu->arch.dtl.update_pending)
1261 			vcpus_to_update[need_vpa_update++] = vcpu;
1262 	}
1263 
1264 	/*
1265 	 * Initialize *vc, in particular vc->vcore_state, so we can
1266 	 * drop the vcore lock if necessary.
1267 	 */
1268 	vc->n_woken = 0;
1269 	vc->nap_count = 0;
1270 	vc->entry_exit_count = 0;
1271 	vc->vcore_state = VCORE_STARTING;
1272 	vc->in_guest = 0;
1273 	vc->napping_threads = 0;
1274 
1275 	/*
1276 	 * Updating any of the vpas requires calling kvmppc_pin_guest_page,
1277 	 * which can't be called with any spinlocks held.
1278 	 */
1279 	if (need_vpa_update) {
1280 		spin_unlock(&vc->lock);
1281 		for (i = 0; i < need_vpa_update; ++i)
1282 			kvmppc_update_vpas(vcpus_to_update[i]);
1283 		spin_lock(&vc->lock);
1284 	}
1285 
1286 	/*
1287 	 * Assign physical thread IDs, first to non-ceded vcpus
1288 	 * and then to ceded ones.
1289 	 */
1290 	ptid = 0;
1291 	vcpu0 = NULL;
1292 	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1293 		if (!vcpu->arch.ceded) {
1294 			if (!ptid)
1295 				vcpu0 = vcpu;
1296 			vcpu->arch.ptid = ptid++;
1297 		}
1298 	}
1299 	if (!vcpu0)
1300 		goto out;	/* nothing to run; should never happen */
1301 	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1302 		if (vcpu->arch.ceded)
1303 			vcpu->arch.ptid = ptid++;
1304 
1305 	/*
1306 	 * Make sure we are running on thread 0, and that
1307 	 * secondary threads are offline.
1308 	 */
1309 	if (threads_per_core > 1 && !on_primary_thread()) {
1310 		list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1311 			vcpu->arch.ret = -EBUSY;
1312 		goto out;
1313 	}
1314 
1315 	vc->pcpu = smp_processor_id();
1316 	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1317 		kvmppc_start_thread(vcpu);
1318 		kvmppc_create_dtl_entry(vcpu, vc);
1319 	}
1320 
1321 	vc->vcore_state = VCORE_RUNNING;
1322 	preempt_disable();
1323 	spin_unlock(&vc->lock);
1324 
1325 	kvm_guest_enter();
1326 
1327 	srcu_idx = srcu_read_lock(&vcpu0->kvm->srcu);
1328 
1329 	__kvmppc_vcore_entry(NULL, vcpu0);
1330 
1331 	spin_lock(&vc->lock);
1332 	/* disable sending of IPIs on virtual external irqs */
1333 	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1334 		vcpu->cpu = -1;
1335 	/* wait for secondary threads to finish writing their state to memory */
1336 	if (vc->nap_count < vc->n_woken)
1337 		kvmppc_wait_for_nap(vc);
1338 	for (i = 0; i < threads_per_core; ++i)
1339 		kvmppc_release_hwthread(vc->pcpu + i);
1340 	/* prevent other vcpu threads from doing kvmppc_start_thread() now */
1341 	vc->vcore_state = VCORE_EXITING;
1342 	spin_unlock(&vc->lock);
1343 
1344 	srcu_read_unlock(&vcpu0->kvm->srcu, srcu_idx);
1345 
1346 	/* make sure updates to secondary vcpu structs are visible now */
1347 	smp_mb();
1348 	kvm_guest_exit();
1349 
1350 	preempt_enable();
1351 	kvm_resched(vcpu);
1352 
1353 	spin_lock(&vc->lock);
1354 	now = get_tb();
1355 	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1356 		/* cancel pending dec exception if dec is positive */
1357 		if (now < vcpu->arch.dec_expires &&
1358 		    kvmppc_core_pending_dec(vcpu))
1359 			kvmppc_core_dequeue_dec(vcpu);
1360 
1361 		ret = RESUME_GUEST;
1362 		if (vcpu->arch.trap)
1363 			ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
1364 						    vcpu->arch.run_task);
1365 
1366 		vcpu->arch.ret = ret;
1367 		vcpu->arch.trap = 0;
1368 
1369 		if (vcpu->arch.ceded) {
1370 			if (ret != RESUME_GUEST)
1371 				kvmppc_end_cede(vcpu);
1372 			else
1373 				kvmppc_set_timer(vcpu);
1374 		}
1375 	}
1376 
1377  out:
1378 	vc->vcore_state = VCORE_INACTIVE;
1379 	list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1380 				 arch.run_list) {
1381 		if (vcpu->arch.ret != RESUME_GUEST) {
1382 			kvmppc_remove_runnable(vc, vcpu);
1383 			wake_up(&vcpu->arch.cpu_run);
1384 		}
1385 	}
1386 }
1387 
1388 /*
1389  * Wait for some other vcpu thread to execute us, and
1390  * wake us up when we need to handle something in the host.
1391  */
1392 static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
1393 {
1394 	DEFINE_WAIT(wait);
1395 
1396 	prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
1397 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
1398 		schedule();
1399 	finish_wait(&vcpu->arch.cpu_run, &wait);
1400 }
1401 
1402 /*
1403  * All the vcpus in this vcore are idle, so wait for a decrementer
1404  * or external interrupt to one of the vcpus.  vc->lock is held.
1405  */
1406 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
1407 {
1408 	DEFINE_WAIT(wait);
1409 
1410 	prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
1411 	vc->vcore_state = VCORE_SLEEPING;
1412 	spin_unlock(&vc->lock);
1413 	schedule();
1414 	finish_wait(&vc->wq, &wait);
1415 	spin_lock(&vc->lock);
1416 	vc->vcore_state = VCORE_INACTIVE;
1417 }
1418 
1419 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1420 {
1421 	int n_ceded;
1422 	struct kvmppc_vcore *vc;
1423 	struct kvm_vcpu *v, *vn;
1424 
1425 	kvm_run->exit_reason = 0;
1426 	vcpu->arch.ret = RESUME_GUEST;
1427 	vcpu->arch.trap = 0;
1428 	kvmppc_update_vpas(vcpu);
1429 
1430 	/*
1431 	 * Synchronize with other threads in this virtual core
1432 	 */
1433 	vc = vcpu->arch.vcore;
1434 	spin_lock(&vc->lock);
1435 	vcpu->arch.ceded = 0;
1436 	vcpu->arch.run_task = current;
1437 	vcpu->arch.kvm_run = kvm_run;
1438 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
1439 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
1440 	vcpu->arch.busy_preempt = TB_NIL;
1441 	list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
1442 	++vc->n_runnable;
1443 
1444 	/*
1445 	 * This happens the first time this is called for a vcpu.
1446 	 * If the vcore is already running, we may be able to start
1447 	 * this thread straight away and have it join in.
1448 	 */
1449 	if (!signal_pending(current)) {
1450 		if (vc->vcore_state == VCORE_RUNNING &&
1451 		    VCORE_EXIT_COUNT(vc) == 0) {
1452 			vcpu->arch.ptid = vc->n_runnable - 1;
1453 			kvmppc_create_dtl_entry(vcpu, vc);
1454 			kvmppc_start_thread(vcpu);
1455 		} else if (vc->vcore_state == VCORE_SLEEPING) {
1456 			wake_up(&vc->wq);
1457 		}
1458 
1459 	}
1460 
1461 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1462 	       !signal_pending(current)) {
1463 		if (vc->vcore_state != VCORE_INACTIVE) {
1464 			spin_unlock(&vc->lock);
1465 			kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
1466 			spin_lock(&vc->lock);
1467 			continue;
1468 		}
1469 		list_for_each_entry_safe(v, vn, &vc->runnable_threads,
1470 					 arch.run_list) {
1471 			kvmppc_core_prepare_to_enter(v);
1472 			if (signal_pending(v->arch.run_task)) {
1473 				kvmppc_remove_runnable(vc, v);
1474 				v->stat.signal_exits++;
1475 				v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
1476 				v->arch.ret = -EINTR;
1477 				wake_up(&v->arch.cpu_run);
1478 			}
1479 		}
1480 		if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1481 			break;
1482 		vc->runner = vcpu;
1483 		n_ceded = 0;
1484 		list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
1485 			if (!v->arch.pending_exceptions)
1486 				n_ceded += v->arch.ceded;
1487 			else
1488 				v->arch.ceded = 0;
1489 		}
1490 		if (n_ceded == vc->n_runnable)
1491 			kvmppc_vcore_blocked(vc);
1492 		else
1493 			kvmppc_run_core(vc);
1494 		vc->runner = NULL;
1495 	}
1496 
1497 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1498 	       (vc->vcore_state == VCORE_RUNNING ||
1499 		vc->vcore_state == VCORE_EXITING)) {
1500 		spin_unlock(&vc->lock);
1501 		kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
1502 		spin_lock(&vc->lock);
1503 	}
1504 
1505 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
1506 		kvmppc_remove_runnable(vc, vcpu);
1507 		vcpu->stat.signal_exits++;
1508 		kvm_run->exit_reason = KVM_EXIT_INTR;
1509 		vcpu->arch.ret = -EINTR;
1510 	}
1511 
1512 	if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
1513 		/* Wake up some vcpu to run the core */
1514 		v = list_first_entry(&vc->runnable_threads,
1515 				     struct kvm_vcpu, arch.run_list);
1516 		wake_up(&v->arch.cpu_run);
1517 	}
1518 
1519 	spin_unlock(&vc->lock);
1520 	return vcpu->arch.ret;
1521 }
1522 
1523 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
1524 {
1525 	int r;
1526 	int srcu_idx;
1527 
1528 	if (!vcpu->arch.sane) {
1529 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1530 		return -EINVAL;
1531 	}
1532 
1533 	kvmppc_core_prepare_to_enter(vcpu);
1534 
1535 	/* No need to go into the guest when all we'll do is come back out */
1536 	if (signal_pending(current)) {
1537 		run->exit_reason = KVM_EXIT_INTR;
1538 		return -EINTR;
1539 	}
1540 
1541 	atomic_inc(&vcpu->kvm->arch.vcpus_running);
1542 	/* Order vcpus_running vs. rma_setup_done, see kvmppc_alloc_reset_hpt */
1543 	smp_mb();
1544 
1545 	/* On the first time here, set up HTAB and VRMA or RMA */
1546 	if (!vcpu->kvm->arch.rma_setup_done) {
1547 		r = kvmppc_hv_setup_htab_rma(vcpu);
1548 		if (r)
1549 			goto out;
1550 	}
1551 
1552 	flush_fp_to_thread(current);
1553 	flush_altivec_to_thread(current);
1554 	flush_vsx_to_thread(current);
1555 	vcpu->arch.wqp = &vcpu->arch.vcore->wq;
1556 	vcpu->arch.pgdir = current->mm->pgd;
1557 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1558 
1559 	do {
1560 		r = kvmppc_run_vcpu(run, vcpu);
1561 
1562 		if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
1563 		    !(vcpu->arch.shregs.msr & MSR_PR)) {
1564 			r = kvmppc_pseries_do_hcall(vcpu);
1565 			kvmppc_core_prepare_to_enter(vcpu);
1566 		} else if (r == RESUME_PAGE_FAULT) {
1567 			srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1568 			r = kvmppc_book3s_hv_page_fault(run, vcpu,
1569 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1570 			srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1571 		}
1572 	} while (r == RESUME_GUEST);
1573 
1574  out:
1575 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1576 	atomic_dec(&vcpu->kvm->arch.vcpus_running);
1577 	return r;
1578 }
1579 
1580 
1581 /* Work out RMLS (real mode limit selector) field value for a given RMA size.
1582    Assumes POWER7 or PPC970. */
1583 static inline int lpcr_rmls(unsigned long rma_size)
1584 {
1585 	switch (rma_size) {
1586 	case 32ul << 20:	/* 32 MB */
1587 		if (cpu_has_feature(CPU_FTR_ARCH_206))
1588 			return 8;	/* only supported on POWER7 */
1589 		return -1;
1590 	case 64ul << 20:	/* 64 MB */
1591 		return 3;
1592 	case 128ul << 20:	/* 128 MB */
1593 		return 7;
1594 	case 256ul << 20:	/* 256 MB */
1595 		return 4;
1596 	case 1ul << 30:		/* 1 GB */
1597 		return 2;
1598 	case 16ul << 30:	/* 16 GB */
1599 		return 1;
1600 	case 256ul << 30:	/* 256 GB */
1601 		return 0;
1602 	default:
1603 		return -1;
1604 	}
1605 }
1606 
1607 static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1608 {
1609 	struct page *page;
1610 	struct kvm_rma_info *ri = vma->vm_file->private_data;
1611 
1612 	if (vmf->pgoff >= kvm_rma_pages)
1613 		return VM_FAULT_SIGBUS;
1614 
1615 	page = pfn_to_page(ri->base_pfn + vmf->pgoff);
1616 	get_page(page);
1617 	vmf->page = page;
1618 	return 0;
1619 }
1620 
1621 static const struct vm_operations_struct kvm_rma_vm_ops = {
1622 	.fault = kvm_rma_fault,
1623 };
1624 
1625 static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma)
1626 {
1627 	vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
1628 	vma->vm_ops = &kvm_rma_vm_ops;
1629 	return 0;
1630 }
1631 
1632 static int kvm_rma_release(struct inode *inode, struct file *filp)
1633 {
1634 	struct kvm_rma_info *ri = filp->private_data;
1635 
1636 	kvm_release_rma(ri);
1637 	return 0;
1638 }
1639 
1640 static const struct file_operations kvm_rma_fops = {
1641 	.mmap           = kvm_rma_mmap,
1642 	.release	= kvm_rma_release,
1643 };
1644 
1645 static long kvm_vm_ioctl_allocate_rma(struct kvm *kvm,
1646 				      struct kvm_allocate_rma *ret)
1647 {
1648 	long fd;
1649 	struct kvm_rma_info *ri;
1650 	/*
1651 	 * Only do this on PPC970 in HV mode
1652 	 */
1653 	if (!cpu_has_feature(CPU_FTR_HVMODE) ||
1654 	    !cpu_has_feature(CPU_FTR_ARCH_201))
1655 		return -EINVAL;
1656 
1657 	if (!kvm_rma_pages)
1658 		return -EINVAL;
1659 
1660 	ri = kvm_alloc_rma();
1661 	if (!ri)
1662 		return -ENOMEM;
1663 
1664 	fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR | O_CLOEXEC);
1665 	if (fd < 0)
1666 		kvm_release_rma(ri);
1667 
1668 	ret->rma_size = kvm_rma_pages << PAGE_SHIFT;
1669 	return fd;
1670 }
1671 
1672 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
1673 				     int linux_psize)
1674 {
1675 	struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
1676 
1677 	if (!def->shift)
1678 		return;
1679 	(*sps)->page_shift = def->shift;
1680 	(*sps)->slb_enc = def->sllp;
1681 	(*sps)->enc[0].page_shift = def->shift;
1682 	/*
1683 	 * Only return base page encoding. We don't want to return
1684 	 * all the supporting pte_enc, because our H_ENTER doesn't
1685 	 * support MPSS yet. Once they do, we can start passing all
1686 	 * support pte_enc here
1687 	 */
1688 	(*sps)->enc[0].pte_enc = def->penc[linux_psize];
1689 	(*sps)++;
1690 }
1691 
1692 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
1693 					 struct kvm_ppc_smmu_info *info)
1694 {
1695 	struct kvm_ppc_one_seg_page_size *sps;
1696 
1697 	info->flags = KVM_PPC_PAGE_SIZES_REAL;
1698 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1699 		info->flags |= KVM_PPC_1T_SEGMENTS;
1700 	info->slb_size = mmu_slb_size;
1701 
1702 	/* We only support these sizes for now, and no muti-size segments */
1703 	sps = &info->sps[0];
1704 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
1705 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
1706 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
1707 
1708 	return 0;
1709 }
1710 
1711 /*
1712  * Get (and clear) the dirty memory log for a memory slot.
1713  */
1714 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
1715 					 struct kvm_dirty_log *log)
1716 {
1717 	struct kvm_memory_slot *memslot;
1718 	int r;
1719 	unsigned long n;
1720 
1721 	mutex_lock(&kvm->slots_lock);
1722 
1723 	r = -EINVAL;
1724 	if (log->slot >= KVM_USER_MEM_SLOTS)
1725 		goto out;
1726 
1727 	memslot = id_to_memslot(kvm->memslots, log->slot);
1728 	r = -ENOENT;
1729 	if (!memslot->dirty_bitmap)
1730 		goto out;
1731 
1732 	n = kvm_dirty_bitmap_bytes(memslot);
1733 	memset(memslot->dirty_bitmap, 0, n);
1734 
1735 	r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
1736 	if (r)
1737 		goto out;
1738 
1739 	r = -EFAULT;
1740 	if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1741 		goto out;
1742 
1743 	r = 0;
1744 out:
1745 	mutex_unlock(&kvm->slots_lock);
1746 	return r;
1747 }
1748 
1749 static void unpin_slot(struct kvm_memory_slot *memslot)
1750 {
1751 	unsigned long *physp;
1752 	unsigned long j, npages, pfn;
1753 	struct page *page;
1754 
1755 	physp = memslot->arch.slot_phys;
1756 	npages = memslot->npages;
1757 	if (!physp)
1758 		return;
1759 	for (j = 0; j < npages; j++) {
1760 		if (!(physp[j] & KVMPPC_GOT_PAGE))
1761 			continue;
1762 		pfn = physp[j] >> PAGE_SHIFT;
1763 		page = pfn_to_page(pfn);
1764 		SetPageDirty(page);
1765 		put_page(page);
1766 	}
1767 }
1768 
1769 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
1770 					struct kvm_memory_slot *dont)
1771 {
1772 	if (!dont || free->arch.rmap != dont->arch.rmap) {
1773 		vfree(free->arch.rmap);
1774 		free->arch.rmap = NULL;
1775 	}
1776 	if (!dont || free->arch.slot_phys != dont->arch.slot_phys) {
1777 		unpin_slot(free);
1778 		vfree(free->arch.slot_phys);
1779 		free->arch.slot_phys = NULL;
1780 	}
1781 }
1782 
1783 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
1784 					 unsigned long npages)
1785 {
1786 	slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
1787 	if (!slot->arch.rmap)
1788 		return -ENOMEM;
1789 	slot->arch.slot_phys = NULL;
1790 
1791 	return 0;
1792 }
1793 
1794 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
1795 					struct kvm_memory_slot *memslot,
1796 					struct kvm_userspace_memory_region *mem)
1797 {
1798 	unsigned long *phys;
1799 
1800 	/* Allocate a slot_phys array if needed */
1801 	phys = memslot->arch.slot_phys;
1802 	if (!kvm->arch.using_mmu_notifiers && !phys && memslot->npages) {
1803 		phys = vzalloc(memslot->npages * sizeof(unsigned long));
1804 		if (!phys)
1805 			return -ENOMEM;
1806 		memslot->arch.slot_phys = phys;
1807 	}
1808 
1809 	return 0;
1810 }
1811 
1812 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
1813 				struct kvm_userspace_memory_region *mem,
1814 				const struct kvm_memory_slot *old)
1815 {
1816 	unsigned long npages = mem->memory_size >> PAGE_SHIFT;
1817 	struct kvm_memory_slot *memslot;
1818 
1819 	if (npages && old->npages) {
1820 		/*
1821 		 * If modifying a memslot, reset all the rmap dirty bits.
1822 		 * If this is a new memslot, we don't need to do anything
1823 		 * since the rmap array starts out as all zeroes,
1824 		 * i.e. no pages are dirty.
1825 		 */
1826 		memslot = id_to_memslot(kvm->memslots, mem->slot);
1827 		kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
1828 	}
1829 }
1830 
1831 /*
1832  * Update LPCR values in kvm->arch and in vcores.
1833  * Caller must hold kvm->lock.
1834  */
1835 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
1836 {
1837 	long int i;
1838 	u32 cores_done = 0;
1839 
1840 	if ((kvm->arch.lpcr & mask) == lpcr)
1841 		return;
1842 
1843 	kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
1844 
1845 	for (i = 0; i < KVM_MAX_VCORES; ++i) {
1846 		struct kvmppc_vcore *vc = kvm->arch.vcores[i];
1847 		if (!vc)
1848 			continue;
1849 		spin_lock(&vc->lock);
1850 		vc->lpcr = (vc->lpcr & ~mask) | lpcr;
1851 		spin_unlock(&vc->lock);
1852 		if (++cores_done >= kvm->arch.online_vcores)
1853 			break;
1854 	}
1855 }
1856 
1857 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
1858 {
1859 	return;
1860 }
1861 
1862 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
1863 {
1864 	int err = 0;
1865 	struct kvm *kvm = vcpu->kvm;
1866 	struct kvm_rma_info *ri = NULL;
1867 	unsigned long hva;
1868 	struct kvm_memory_slot *memslot;
1869 	struct vm_area_struct *vma;
1870 	unsigned long lpcr = 0, senc;
1871 	unsigned long lpcr_mask = 0;
1872 	unsigned long psize, porder;
1873 	unsigned long rma_size;
1874 	unsigned long rmls;
1875 	unsigned long *physp;
1876 	unsigned long i, npages;
1877 	int srcu_idx;
1878 
1879 	mutex_lock(&kvm->lock);
1880 	if (kvm->arch.rma_setup_done)
1881 		goto out;	/* another vcpu beat us to it */
1882 
1883 	/* Allocate hashed page table (if not done already) and reset it */
1884 	if (!kvm->arch.hpt_virt) {
1885 		err = kvmppc_alloc_hpt(kvm, NULL);
1886 		if (err) {
1887 			pr_err("KVM: Couldn't alloc HPT\n");
1888 			goto out;
1889 		}
1890 	}
1891 
1892 	/* Look up the memslot for guest physical address 0 */
1893 	srcu_idx = srcu_read_lock(&kvm->srcu);
1894 	memslot = gfn_to_memslot(kvm, 0);
1895 
1896 	/* We must have some memory at 0 by now */
1897 	err = -EINVAL;
1898 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
1899 		goto out_srcu;
1900 
1901 	/* Look up the VMA for the start of this memory slot */
1902 	hva = memslot->userspace_addr;
1903 	down_read(&current->mm->mmap_sem);
1904 	vma = find_vma(current->mm, hva);
1905 	if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
1906 		goto up_out;
1907 
1908 	psize = vma_kernel_pagesize(vma);
1909 	porder = __ilog2(psize);
1910 
1911 	/* Is this one of our preallocated RMAs? */
1912 	if (vma->vm_file && vma->vm_file->f_op == &kvm_rma_fops &&
1913 	    hva == vma->vm_start)
1914 		ri = vma->vm_file->private_data;
1915 
1916 	up_read(&current->mm->mmap_sem);
1917 
1918 	if (!ri) {
1919 		/* On POWER7, use VRMA; on PPC970, give up */
1920 		err = -EPERM;
1921 		if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1922 			pr_err("KVM: CPU requires an RMO\n");
1923 			goto out_srcu;
1924 		}
1925 
1926 		/* We can handle 4k, 64k or 16M pages in the VRMA */
1927 		err = -EINVAL;
1928 		if (!(psize == 0x1000 || psize == 0x10000 ||
1929 		      psize == 0x1000000))
1930 			goto out_srcu;
1931 
1932 		/* Update VRMASD field in the LPCR */
1933 		senc = slb_pgsize_encoding(psize);
1934 		kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
1935 			(VRMA_VSID << SLB_VSID_SHIFT_1T);
1936 		lpcr_mask = LPCR_VRMASD;
1937 		/* the -4 is to account for senc values starting at 0x10 */
1938 		lpcr = senc << (LPCR_VRMASD_SH - 4);
1939 
1940 		/* Create HPTEs in the hash page table for the VRMA */
1941 		kvmppc_map_vrma(vcpu, memslot, porder);
1942 
1943 	} else {
1944 		/* Set up to use an RMO region */
1945 		rma_size = kvm_rma_pages;
1946 		if (rma_size > memslot->npages)
1947 			rma_size = memslot->npages;
1948 		rma_size <<= PAGE_SHIFT;
1949 		rmls = lpcr_rmls(rma_size);
1950 		err = -EINVAL;
1951 		if ((long)rmls < 0) {
1952 			pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size);
1953 			goto out_srcu;
1954 		}
1955 		atomic_inc(&ri->use_count);
1956 		kvm->arch.rma = ri;
1957 
1958 		/* Update LPCR and RMOR */
1959 		if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1960 			/* PPC970; insert RMLS value (split field) in HID4 */
1961 			lpcr_mask = (1ul << HID4_RMLS0_SH) |
1962 				(3ul << HID4_RMLS2_SH) | HID4_RMOR;
1963 			lpcr = ((rmls >> 2) << HID4_RMLS0_SH) |
1964 				((rmls & 3) << HID4_RMLS2_SH);
1965 			/* RMOR is also in HID4 */
1966 			lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff)
1967 				<< HID4_RMOR_SH;
1968 		} else {
1969 			/* POWER7 */
1970 			lpcr_mask = LPCR_VPM0 | LPCR_VRMA_L | LPCR_RMLS;
1971 			lpcr = rmls << LPCR_RMLS_SH;
1972 			kvm->arch.rmor = ri->base_pfn << PAGE_SHIFT;
1973 		}
1974 		pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n",
1975 			ri->base_pfn << PAGE_SHIFT, rma_size, lpcr);
1976 
1977 		/* Initialize phys addrs of pages in RMO */
1978 		npages = kvm_rma_pages;
1979 		porder = __ilog2(npages);
1980 		physp = memslot->arch.slot_phys;
1981 		if (physp) {
1982 			if (npages > memslot->npages)
1983 				npages = memslot->npages;
1984 			spin_lock(&kvm->arch.slot_phys_lock);
1985 			for (i = 0; i < npages; ++i)
1986 				physp[i] = ((ri->base_pfn + i) << PAGE_SHIFT) +
1987 					porder;
1988 			spin_unlock(&kvm->arch.slot_phys_lock);
1989 		}
1990 	}
1991 
1992 	kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
1993 
1994 	/* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */
1995 	smp_wmb();
1996 	kvm->arch.rma_setup_done = 1;
1997 	err = 0;
1998  out_srcu:
1999 	srcu_read_unlock(&kvm->srcu, srcu_idx);
2000  out:
2001 	mutex_unlock(&kvm->lock);
2002 	return err;
2003 
2004  up_out:
2005 	up_read(&current->mm->mmap_sem);
2006 	goto out_srcu;
2007 }
2008 
2009 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
2010 {
2011 	unsigned long lpcr, lpid;
2012 
2013 	/* Allocate the guest's logical partition ID */
2014 
2015 	lpid = kvmppc_alloc_lpid();
2016 	if ((long)lpid < 0)
2017 		return -ENOMEM;
2018 	kvm->arch.lpid = lpid;
2019 
2020 	/*
2021 	 * Since we don't flush the TLB when tearing down a VM,
2022 	 * and this lpid might have previously been used,
2023 	 * make sure we flush on each core before running the new VM.
2024 	 */
2025 	cpumask_setall(&kvm->arch.need_tlb_flush);
2026 
2027 	kvm->arch.rma = NULL;
2028 
2029 	kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
2030 
2031 	if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2032 		/* PPC970; HID4 is effectively the LPCR */
2033 		kvm->arch.host_lpid = 0;
2034 		kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4);
2035 		lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH));
2036 		lpcr |= ((lpid >> 4) << HID4_LPID1_SH) |
2037 			((lpid & 0xf) << HID4_LPID5_SH);
2038 	} else {
2039 		/* POWER7; init LPCR for virtual RMA mode */
2040 		kvm->arch.host_lpid = mfspr(SPRN_LPID);
2041 		kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
2042 		lpcr &= LPCR_PECE | LPCR_LPES;
2043 		lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
2044 			LPCR_VPM0 | LPCR_VPM1;
2045 		kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
2046 			(VRMA_VSID << SLB_VSID_SHIFT_1T);
2047 	}
2048 	kvm->arch.lpcr = lpcr;
2049 
2050 	kvm->arch.using_mmu_notifiers = !!cpu_has_feature(CPU_FTR_ARCH_206);
2051 	spin_lock_init(&kvm->arch.slot_phys_lock);
2052 
2053 	/*
2054 	 * Don't allow secondary CPU threads to come online
2055 	 * while any KVM VMs exist.
2056 	 */
2057 	inhibit_secondary_onlining();
2058 
2059 	return 0;
2060 }
2061 
2062 static void kvmppc_free_vcores(struct kvm *kvm)
2063 {
2064 	long int i;
2065 
2066 	for (i = 0; i < KVM_MAX_VCORES; ++i)
2067 		kfree(kvm->arch.vcores[i]);
2068 	kvm->arch.online_vcores = 0;
2069 }
2070 
2071 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
2072 {
2073 	uninhibit_secondary_onlining();
2074 
2075 	kvmppc_free_vcores(kvm);
2076 	if (kvm->arch.rma) {
2077 		kvm_release_rma(kvm->arch.rma);
2078 		kvm->arch.rma = NULL;
2079 	}
2080 
2081 	kvmppc_free_hpt(kvm);
2082 }
2083 
2084 /* We don't need to emulate any privileged instructions or dcbz */
2085 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
2086 				     unsigned int inst, int *advance)
2087 {
2088 	return EMULATE_FAIL;
2089 }
2090 
2091 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
2092 					ulong spr_val)
2093 {
2094 	return EMULATE_FAIL;
2095 }
2096 
2097 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
2098 					ulong *spr_val)
2099 {
2100 	return EMULATE_FAIL;
2101 }
2102 
2103 static int kvmppc_core_check_processor_compat_hv(void)
2104 {
2105 	if (!cpu_has_feature(CPU_FTR_HVMODE))
2106 		return -EIO;
2107 	return 0;
2108 }
2109 
2110 static long kvm_arch_vm_ioctl_hv(struct file *filp,
2111 				 unsigned int ioctl, unsigned long arg)
2112 {
2113 	struct kvm *kvm __maybe_unused = filp->private_data;
2114 	void __user *argp = (void __user *)arg;
2115 	long r;
2116 
2117 	switch (ioctl) {
2118 
2119 	case KVM_ALLOCATE_RMA: {
2120 		struct kvm_allocate_rma rma;
2121 		struct kvm *kvm = filp->private_data;
2122 
2123 		r = kvm_vm_ioctl_allocate_rma(kvm, &rma);
2124 		if (r >= 0 && copy_to_user(argp, &rma, sizeof(rma)))
2125 			r = -EFAULT;
2126 		break;
2127 	}
2128 
2129 	case KVM_PPC_ALLOCATE_HTAB: {
2130 		u32 htab_order;
2131 
2132 		r = -EFAULT;
2133 		if (get_user(htab_order, (u32 __user *)argp))
2134 			break;
2135 		r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
2136 		if (r)
2137 			break;
2138 		r = -EFAULT;
2139 		if (put_user(htab_order, (u32 __user *)argp))
2140 			break;
2141 		r = 0;
2142 		break;
2143 	}
2144 
2145 	case KVM_PPC_GET_HTAB_FD: {
2146 		struct kvm_get_htab_fd ghf;
2147 
2148 		r = -EFAULT;
2149 		if (copy_from_user(&ghf, argp, sizeof(ghf)))
2150 			break;
2151 		r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
2152 		break;
2153 	}
2154 
2155 	default:
2156 		r = -ENOTTY;
2157 	}
2158 
2159 	return r;
2160 }
2161 
2162 static struct kvmppc_ops kvm_ops_hv = {
2163 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
2164 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
2165 	.get_one_reg = kvmppc_get_one_reg_hv,
2166 	.set_one_reg = kvmppc_set_one_reg_hv,
2167 	.vcpu_load   = kvmppc_core_vcpu_load_hv,
2168 	.vcpu_put    = kvmppc_core_vcpu_put_hv,
2169 	.set_msr     = kvmppc_set_msr_hv,
2170 	.vcpu_run    = kvmppc_vcpu_run_hv,
2171 	.vcpu_create = kvmppc_core_vcpu_create_hv,
2172 	.vcpu_free   = kvmppc_core_vcpu_free_hv,
2173 	.check_requests = kvmppc_core_check_requests_hv,
2174 	.get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
2175 	.flush_memslot  = kvmppc_core_flush_memslot_hv,
2176 	.prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
2177 	.commit_memory_region  = kvmppc_core_commit_memory_region_hv,
2178 	.unmap_hva = kvm_unmap_hva_hv,
2179 	.unmap_hva_range = kvm_unmap_hva_range_hv,
2180 	.age_hva  = kvm_age_hva_hv,
2181 	.test_age_hva = kvm_test_age_hva_hv,
2182 	.set_spte_hva = kvm_set_spte_hva_hv,
2183 	.mmu_destroy  = kvmppc_mmu_destroy_hv,
2184 	.free_memslot = kvmppc_core_free_memslot_hv,
2185 	.create_memslot = kvmppc_core_create_memslot_hv,
2186 	.init_vm =  kvmppc_core_init_vm_hv,
2187 	.destroy_vm = kvmppc_core_destroy_vm_hv,
2188 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
2189 	.emulate_op = kvmppc_core_emulate_op_hv,
2190 	.emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
2191 	.emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
2192 	.fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
2193 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
2194 };
2195 
2196 static int kvmppc_book3s_init_hv(void)
2197 {
2198 	int r;
2199 	/*
2200 	 * FIXME!! Do we need to check on all cpus ?
2201 	 */
2202 	r = kvmppc_core_check_processor_compat_hv();
2203 	if (r < 0)
2204 		return r;
2205 
2206 	kvm_ops_hv.owner = THIS_MODULE;
2207 	kvmppc_hv_ops = &kvm_ops_hv;
2208 
2209 	r = kvmppc_mmu_hv_init();
2210 	return r;
2211 }
2212 
2213 static void kvmppc_book3s_exit_hv(void)
2214 {
2215 	kvmppc_hv_ops = NULL;
2216 }
2217 
2218 module_init(kvmppc_book3s_init_hv);
2219 module_exit(kvmppc_book3s_exit_hv);
2220 MODULE_LICENSE("GPL");
2221