xref: /openbmc/linux/arch/powerpc/kvm/book3s_hv.c (revision 63dc02bd)
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 
34 #include <asm/reg.h>
35 #include <asm/cputable.h>
36 #include <asm/cacheflush.h>
37 #include <asm/tlbflush.h>
38 #include <asm/uaccess.h>
39 #include <asm/io.h>
40 #include <asm/kvm_ppc.h>
41 #include <asm/kvm_book3s.h>
42 #include <asm/mmu_context.h>
43 #include <asm/lppaca.h>
44 #include <asm/processor.h>
45 #include <asm/cputhreads.h>
46 #include <asm/page.h>
47 #include <asm/hvcall.h>
48 #include <asm/switch_to.h>
49 #include <linux/gfp.h>
50 #include <linux/vmalloc.h>
51 #include <linux/highmem.h>
52 #include <linux/hugetlb.h>
53 
54 /* #define EXIT_DEBUG */
55 /* #define EXIT_DEBUG_SIMPLE */
56 /* #define EXIT_DEBUG_INT */
57 
58 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
59 static int kvmppc_hv_setup_rma(struct kvm_vcpu *vcpu);
60 
61 void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
62 {
63 	local_paca->kvm_hstate.kvm_vcpu = vcpu;
64 	local_paca->kvm_hstate.kvm_vcore = vcpu->arch.vcore;
65 }
66 
67 void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
68 {
69 }
70 
71 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
72 {
73 	vcpu->arch.shregs.msr = msr;
74 	kvmppc_end_cede(vcpu);
75 }
76 
77 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
78 {
79 	vcpu->arch.pvr = pvr;
80 }
81 
82 void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
83 {
84 	int r;
85 
86 	pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
87 	pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
88 	       vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
89 	for (r = 0; r < 16; ++r)
90 		pr_err("r%2d = %.16lx  r%d = %.16lx\n",
91 		       r, kvmppc_get_gpr(vcpu, r),
92 		       r+16, kvmppc_get_gpr(vcpu, r+16));
93 	pr_err("ctr = %.16lx  lr  = %.16lx\n",
94 	       vcpu->arch.ctr, vcpu->arch.lr);
95 	pr_err("srr0 = %.16llx srr1 = %.16llx\n",
96 	       vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
97 	pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
98 	       vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
99 	pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
100 	       vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
101 	pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
102 	       vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
103 	pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
104 	pr_err("fault dar = %.16lx dsisr = %.8x\n",
105 	       vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
106 	pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
107 	for (r = 0; r < vcpu->arch.slb_max; ++r)
108 		pr_err("  ESID = %.16llx VSID = %.16llx\n",
109 		       vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
110 	pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
111 	       vcpu->kvm->arch.lpcr, vcpu->kvm->arch.sdr1,
112 	       vcpu->arch.last_inst);
113 }
114 
115 struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
116 {
117 	int r;
118 	struct kvm_vcpu *v, *ret = NULL;
119 
120 	mutex_lock(&kvm->lock);
121 	kvm_for_each_vcpu(r, v, kvm) {
122 		if (v->vcpu_id == id) {
123 			ret = v;
124 			break;
125 		}
126 	}
127 	mutex_unlock(&kvm->lock);
128 	return ret;
129 }
130 
131 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
132 {
133 	vpa->shared_proc = 1;
134 	vpa->yield_count = 1;
135 }
136 
137 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
138 				       unsigned long flags,
139 				       unsigned long vcpuid, unsigned long vpa)
140 {
141 	struct kvm *kvm = vcpu->kvm;
142 	unsigned long len, nb;
143 	void *va;
144 	struct kvm_vcpu *tvcpu;
145 	int err = H_PARAMETER;
146 
147 	tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
148 	if (!tvcpu)
149 		return H_PARAMETER;
150 
151 	flags >>= 63 - 18;
152 	flags &= 7;
153 	if (flags == 0 || flags == 4)
154 		return H_PARAMETER;
155 	if (flags < 4) {
156 		if (vpa & 0x7f)
157 			return H_PARAMETER;
158 		if (flags >= 2 && !tvcpu->arch.vpa)
159 			return H_RESOURCE;
160 		/* registering new area; convert logical addr to real */
161 		va = kvmppc_pin_guest_page(kvm, vpa, &nb);
162 		if (va == NULL)
163 			return H_PARAMETER;
164 		if (flags <= 1)
165 			len = *(unsigned short *)(va + 4);
166 		else
167 			len = *(unsigned int *)(va + 4);
168 		if (len > nb)
169 			goto out_unpin;
170 		switch (flags) {
171 		case 1:		/* register VPA */
172 			if (len < 640)
173 				goto out_unpin;
174 			if (tvcpu->arch.vpa)
175 				kvmppc_unpin_guest_page(kvm, vcpu->arch.vpa);
176 			tvcpu->arch.vpa = va;
177 			init_vpa(vcpu, va);
178 			break;
179 		case 2:		/* register DTL */
180 			if (len < 48)
181 				goto out_unpin;
182 			len -= len % 48;
183 			if (tvcpu->arch.dtl)
184 				kvmppc_unpin_guest_page(kvm, vcpu->arch.dtl);
185 			tvcpu->arch.dtl = va;
186 			tvcpu->arch.dtl_end = va + len;
187 			break;
188 		case 3:		/* register SLB shadow buffer */
189 			if (len < 16)
190 				goto out_unpin;
191 			if (tvcpu->arch.slb_shadow)
192 				kvmppc_unpin_guest_page(kvm, vcpu->arch.slb_shadow);
193 			tvcpu->arch.slb_shadow = va;
194 			break;
195 		}
196 	} else {
197 		switch (flags) {
198 		case 5:		/* unregister VPA */
199 			if (tvcpu->arch.slb_shadow || tvcpu->arch.dtl)
200 				return H_RESOURCE;
201 			if (!tvcpu->arch.vpa)
202 				break;
203 			kvmppc_unpin_guest_page(kvm, tvcpu->arch.vpa);
204 			tvcpu->arch.vpa = NULL;
205 			break;
206 		case 6:		/* unregister DTL */
207 			if (!tvcpu->arch.dtl)
208 				break;
209 			kvmppc_unpin_guest_page(kvm, tvcpu->arch.dtl);
210 			tvcpu->arch.dtl = NULL;
211 			break;
212 		case 7:		/* unregister SLB shadow buffer */
213 			if (!tvcpu->arch.slb_shadow)
214 				break;
215 			kvmppc_unpin_guest_page(kvm, tvcpu->arch.slb_shadow);
216 			tvcpu->arch.slb_shadow = NULL;
217 			break;
218 		}
219 	}
220 	return H_SUCCESS;
221 
222  out_unpin:
223 	kvmppc_unpin_guest_page(kvm, va);
224 	return err;
225 }
226 
227 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
228 {
229 	unsigned long req = kvmppc_get_gpr(vcpu, 3);
230 	unsigned long target, ret = H_SUCCESS;
231 	struct kvm_vcpu *tvcpu;
232 
233 	switch (req) {
234 	case H_ENTER:
235 		ret = kvmppc_virtmode_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
236 					      kvmppc_get_gpr(vcpu, 5),
237 					      kvmppc_get_gpr(vcpu, 6),
238 					      kvmppc_get_gpr(vcpu, 7));
239 		break;
240 	case H_CEDE:
241 		break;
242 	case H_PROD:
243 		target = kvmppc_get_gpr(vcpu, 4);
244 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
245 		if (!tvcpu) {
246 			ret = H_PARAMETER;
247 			break;
248 		}
249 		tvcpu->arch.prodded = 1;
250 		smp_mb();
251 		if (vcpu->arch.ceded) {
252 			if (waitqueue_active(&vcpu->wq)) {
253 				wake_up_interruptible(&vcpu->wq);
254 				vcpu->stat.halt_wakeup++;
255 			}
256 		}
257 		break;
258 	case H_CONFER:
259 		break;
260 	case H_REGISTER_VPA:
261 		ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
262 					kvmppc_get_gpr(vcpu, 5),
263 					kvmppc_get_gpr(vcpu, 6));
264 		break;
265 	default:
266 		return RESUME_HOST;
267 	}
268 	kvmppc_set_gpr(vcpu, 3, ret);
269 	vcpu->arch.hcall_needed = 0;
270 	return RESUME_GUEST;
271 }
272 
273 static int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
274 			      struct task_struct *tsk)
275 {
276 	int r = RESUME_HOST;
277 
278 	vcpu->stat.sum_exits++;
279 
280 	run->exit_reason = KVM_EXIT_UNKNOWN;
281 	run->ready_for_interrupt_injection = 1;
282 	switch (vcpu->arch.trap) {
283 	/* We're good on these - the host merely wanted to get our attention */
284 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
285 		vcpu->stat.dec_exits++;
286 		r = RESUME_GUEST;
287 		break;
288 	case BOOK3S_INTERRUPT_EXTERNAL:
289 		vcpu->stat.ext_intr_exits++;
290 		r = RESUME_GUEST;
291 		break;
292 	case BOOK3S_INTERRUPT_PERFMON:
293 		r = RESUME_GUEST;
294 		break;
295 	case BOOK3S_INTERRUPT_PROGRAM:
296 	{
297 		ulong flags;
298 		/*
299 		 * Normally program interrupts are delivered directly
300 		 * to the guest by the hardware, but we can get here
301 		 * as a result of a hypervisor emulation interrupt
302 		 * (e40) getting turned into a 700 by BML RTAS.
303 		 */
304 		flags = vcpu->arch.shregs.msr & 0x1f0000ull;
305 		kvmppc_core_queue_program(vcpu, flags);
306 		r = RESUME_GUEST;
307 		break;
308 	}
309 	case BOOK3S_INTERRUPT_SYSCALL:
310 	{
311 		/* hcall - punt to userspace */
312 		int i;
313 
314 		if (vcpu->arch.shregs.msr & MSR_PR) {
315 			/* sc 1 from userspace - reflect to guest syscall */
316 			kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_SYSCALL);
317 			r = RESUME_GUEST;
318 			break;
319 		}
320 		run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
321 		for (i = 0; i < 9; ++i)
322 			run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
323 		run->exit_reason = KVM_EXIT_PAPR_HCALL;
324 		vcpu->arch.hcall_needed = 1;
325 		r = RESUME_HOST;
326 		break;
327 	}
328 	/*
329 	 * We get these next two if the guest accesses a page which it thinks
330 	 * it has mapped but which is not actually present, either because
331 	 * it is for an emulated I/O device or because the corresonding
332 	 * host page has been paged out.  Any other HDSI/HISI interrupts
333 	 * have been handled already.
334 	 */
335 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
336 		r = kvmppc_book3s_hv_page_fault(run, vcpu,
337 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
338 		break;
339 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
340 		r = kvmppc_book3s_hv_page_fault(run, vcpu,
341 				kvmppc_get_pc(vcpu), 0);
342 		break;
343 	/*
344 	 * This occurs if the guest executes an illegal instruction.
345 	 * We just generate a program interrupt to the guest, since
346 	 * we don't emulate any guest instructions at this stage.
347 	 */
348 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
349 		kvmppc_core_queue_program(vcpu, 0x80000);
350 		r = RESUME_GUEST;
351 		break;
352 	default:
353 		kvmppc_dump_regs(vcpu);
354 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
355 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
356 			vcpu->arch.shregs.msr);
357 		r = RESUME_HOST;
358 		BUG();
359 		break;
360 	}
361 
362 	return r;
363 }
364 
365 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
366                                   struct kvm_sregs *sregs)
367 {
368 	int i;
369 
370 	sregs->pvr = vcpu->arch.pvr;
371 
372 	memset(sregs, 0, sizeof(struct kvm_sregs));
373 	for (i = 0; i < vcpu->arch.slb_max; i++) {
374 		sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
375 		sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
376 	}
377 
378 	return 0;
379 }
380 
381 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
382                                   struct kvm_sregs *sregs)
383 {
384 	int i, j;
385 
386 	kvmppc_set_pvr(vcpu, sregs->pvr);
387 
388 	j = 0;
389 	for (i = 0; i < vcpu->arch.slb_nr; i++) {
390 		if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
391 			vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
392 			vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
393 			++j;
394 		}
395 	}
396 	vcpu->arch.slb_max = j;
397 
398 	return 0;
399 }
400 
401 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
402 {
403 	int r = -EINVAL;
404 
405 	switch (reg->id) {
406 	case KVM_REG_PPC_HIOR:
407 		r = put_user(0, (u64 __user *)reg->addr);
408 		break;
409 	default:
410 		break;
411 	}
412 
413 	return r;
414 }
415 
416 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
417 {
418 	int r = -EINVAL;
419 
420 	switch (reg->id) {
421 	case KVM_REG_PPC_HIOR:
422 	{
423 		u64 hior;
424 		/* Only allow this to be set to zero */
425 		r = get_user(hior, (u64 __user *)reg->addr);
426 		if (!r && (hior != 0))
427 			r = -EINVAL;
428 		break;
429 	}
430 	default:
431 		break;
432 	}
433 
434 	return r;
435 }
436 
437 int kvmppc_core_check_processor_compat(void)
438 {
439 	if (cpu_has_feature(CPU_FTR_HVMODE))
440 		return 0;
441 	return -EIO;
442 }
443 
444 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
445 {
446 	struct kvm_vcpu *vcpu;
447 	int err = -EINVAL;
448 	int core;
449 	struct kvmppc_vcore *vcore;
450 
451 	core = id / threads_per_core;
452 	if (core >= KVM_MAX_VCORES)
453 		goto out;
454 
455 	err = -ENOMEM;
456 	vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
457 	if (!vcpu)
458 		goto out;
459 
460 	err = kvm_vcpu_init(vcpu, kvm, id);
461 	if (err)
462 		goto free_vcpu;
463 
464 	vcpu->arch.shared = &vcpu->arch.shregs;
465 	vcpu->arch.last_cpu = -1;
466 	vcpu->arch.mmcr[0] = MMCR0_FC;
467 	vcpu->arch.ctrl = CTRL_RUNLATCH;
468 	/* default to host PVR, since we can't spoof it */
469 	vcpu->arch.pvr = mfspr(SPRN_PVR);
470 	kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
471 
472 	kvmppc_mmu_book3s_hv_init(vcpu);
473 
474 	/*
475 	 * We consider the vcpu stopped until we see the first run ioctl for it.
476 	 */
477 	vcpu->arch.state = KVMPPC_VCPU_STOPPED;
478 
479 	init_waitqueue_head(&vcpu->arch.cpu_run);
480 
481 	mutex_lock(&kvm->lock);
482 	vcore = kvm->arch.vcores[core];
483 	if (!vcore) {
484 		vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
485 		if (vcore) {
486 			INIT_LIST_HEAD(&vcore->runnable_threads);
487 			spin_lock_init(&vcore->lock);
488 			init_waitqueue_head(&vcore->wq);
489 		}
490 		kvm->arch.vcores[core] = vcore;
491 	}
492 	mutex_unlock(&kvm->lock);
493 
494 	if (!vcore)
495 		goto free_vcpu;
496 
497 	spin_lock(&vcore->lock);
498 	++vcore->num_threads;
499 	spin_unlock(&vcore->lock);
500 	vcpu->arch.vcore = vcore;
501 
502 	vcpu->arch.cpu_type = KVM_CPU_3S_64;
503 	kvmppc_sanity_check(vcpu);
504 
505 	return vcpu;
506 
507 free_vcpu:
508 	kmem_cache_free(kvm_vcpu_cache, vcpu);
509 out:
510 	return ERR_PTR(err);
511 }
512 
513 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
514 {
515 	if (vcpu->arch.dtl)
516 		kvmppc_unpin_guest_page(vcpu->kvm, vcpu->arch.dtl);
517 	if (vcpu->arch.slb_shadow)
518 		kvmppc_unpin_guest_page(vcpu->kvm, vcpu->arch.slb_shadow);
519 	if (vcpu->arch.vpa)
520 		kvmppc_unpin_guest_page(vcpu->kvm, vcpu->arch.vpa);
521 	kvm_vcpu_uninit(vcpu);
522 	kmem_cache_free(kvm_vcpu_cache, vcpu);
523 }
524 
525 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
526 {
527 	unsigned long dec_nsec, now;
528 
529 	now = get_tb();
530 	if (now > vcpu->arch.dec_expires) {
531 		/* decrementer has already gone negative */
532 		kvmppc_core_queue_dec(vcpu);
533 		kvmppc_core_prepare_to_enter(vcpu);
534 		return;
535 	}
536 	dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
537 		   / tb_ticks_per_sec;
538 	hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
539 		      HRTIMER_MODE_REL);
540 	vcpu->arch.timer_running = 1;
541 }
542 
543 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
544 {
545 	vcpu->arch.ceded = 0;
546 	if (vcpu->arch.timer_running) {
547 		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
548 		vcpu->arch.timer_running = 0;
549 	}
550 }
551 
552 extern int __kvmppc_vcore_entry(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu);
553 extern void xics_wake_cpu(int cpu);
554 
555 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
556 				   struct kvm_vcpu *vcpu)
557 {
558 	struct kvm_vcpu *v;
559 
560 	if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
561 		return;
562 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
563 	--vc->n_runnable;
564 	++vc->n_busy;
565 	/* decrement the physical thread id of each following vcpu */
566 	v = vcpu;
567 	list_for_each_entry_continue(v, &vc->runnable_threads, arch.run_list)
568 		--v->arch.ptid;
569 	list_del(&vcpu->arch.run_list);
570 }
571 
572 static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
573 {
574 	int cpu;
575 	struct paca_struct *tpaca;
576 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
577 
578 	if (vcpu->arch.timer_running) {
579 		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
580 		vcpu->arch.timer_running = 0;
581 	}
582 	cpu = vc->pcpu + vcpu->arch.ptid;
583 	tpaca = &paca[cpu];
584 	tpaca->kvm_hstate.kvm_vcpu = vcpu;
585 	tpaca->kvm_hstate.kvm_vcore = vc;
586 	tpaca->kvm_hstate.napping = 0;
587 	vcpu->cpu = vc->pcpu;
588 	smp_wmb();
589 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
590 	if (vcpu->arch.ptid) {
591 		tpaca->cpu_start = 0x80;
592 		wmb();
593 		xics_wake_cpu(cpu);
594 		++vc->n_woken;
595 	}
596 #endif
597 }
598 
599 static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc)
600 {
601 	int i;
602 
603 	HMT_low();
604 	i = 0;
605 	while (vc->nap_count < vc->n_woken) {
606 		if (++i >= 1000000) {
607 			pr_err("kvmppc_wait_for_nap timeout %d %d\n",
608 			       vc->nap_count, vc->n_woken);
609 			break;
610 		}
611 		cpu_relax();
612 	}
613 	HMT_medium();
614 }
615 
616 /*
617  * Check that we are on thread 0 and that any other threads in
618  * this core are off-line.
619  */
620 static int on_primary_thread(void)
621 {
622 	int cpu = smp_processor_id();
623 	int thr = cpu_thread_in_core(cpu);
624 
625 	if (thr)
626 		return 0;
627 	while (++thr < threads_per_core)
628 		if (cpu_online(cpu + thr))
629 			return 0;
630 	return 1;
631 }
632 
633 /*
634  * Run a set of guest threads on a physical core.
635  * Called with vc->lock held.
636  */
637 static int kvmppc_run_core(struct kvmppc_vcore *vc)
638 {
639 	struct kvm_vcpu *vcpu, *vcpu0, *vnext;
640 	long ret;
641 	u64 now;
642 	int ptid;
643 
644 	/* don't start if any threads have a signal pending */
645 	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
646 		if (signal_pending(vcpu->arch.run_task))
647 			return 0;
648 
649 	/*
650 	 * Make sure we are running on thread 0, and that
651 	 * secondary threads are offline.
652 	 * XXX we should also block attempts to bring any
653 	 * secondary threads online.
654 	 */
655 	if (threads_per_core > 1 && !on_primary_thread()) {
656 		list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
657 			vcpu->arch.ret = -EBUSY;
658 		goto out;
659 	}
660 
661 	/*
662 	 * Assign physical thread IDs, first to non-ceded vcpus
663 	 * and then to ceded ones.
664 	 */
665 	ptid = 0;
666 	vcpu0 = NULL;
667 	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
668 		if (!vcpu->arch.ceded) {
669 			if (!ptid)
670 				vcpu0 = vcpu;
671 			vcpu->arch.ptid = ptid++;
672 		}
673 	}
674 	if (!vcpu0)
675 		return 0;		/* nothing to run */
676 	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
677 		if (vcpu->arch.ceded)
678 			vcpu->arch.ptid = ptid++;
679 
680 	vc->n_woken = 0;
681 	vc->nap_count = 0;
682 	vc->entry_exit_count = 0;
683 	vc->vcore_state = VCORE_RUNNING;
684 	vc->in_guest = 0;
685 	vc->pcpu = smp_processor_id();
686 	vc->napping_threads = 0;
687 	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
688 		kvmppc_start_thread(vcpu);
689 
690 	preempt_disable();
691 	spin_unlock(&vc->lock);
692 
693 	kvm_guest_enter();
694 	__kvmppc_vcore_entry(NULL, vcpu0);
695 
696 	spin_lock(&vc->lock);
697 	/* disable sending of IPIs on virtual external irqs */
698 	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
699 		vcpu->cpu = -1;
700 	/* wait for secondary threads to finish writing their state to memory */
701 	if (vc->nap_count < vc->n_woken)
702 		kvmppc_wait_for_nap(vc);
703 	/* prevent other vcpu threads from doing kvmppc_start_thread() now */
704 	vc->vcore_state = VCORE_EXITING;
705 	spin_unlock(&vc->lock);
706 
707 	/* make sure updates to secondary vcpu structs are visible now */
708 	smp_mb();
709 	kvm_guest_exit();
710 
711 	preempt_enable();
712 	kvm_resched(vcpu);
713 
714 	now = get_tb();
715 	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
716 		/* cancel pending dec exception if dec is positive */
717 		if (now < vcpu->arch.dec_expires &&
718 		    kvmppc_core_pending_dec(vcpu))
719 			kvmppc_core_dequeue_dec(vcpu);
720 
721 		ret = RESUME_GUEST;
722 		if (vcpu->arch.trap)
723 			ret = kvmppc_handle_exit(vcpu->arch.kvm_run, vcpu,
724 						 vcpu->arch.run_task);
725 
726 		vcpu->arch.ret = ret;
727 		vcpu->arch.trap = 0;
728 
729 		if (vcpu->arch.ceded) {
730 			if (ret != RESUME_GUEST)
731 				kvmppc_end_cede(vcpu);
732 			else
733 				kvmppc_set_timer(vcpu);
734 		}
735 	}
736 
737 	spin_lock(&vc->lock);
738  out:
739 	vc->vcore_state = VCORE_INACTIVE;
740 	list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
741 				 arch.run_list) {
742 		if (vcpu->arch.ret != RESUME_GUEST) {
743 			kvmppc_remove_runnable(vc, vcpu);
744 			wake_up(&vcpu->arch.cpu_run);
745 		}
746 	}
747 
748 	return 1;
749 }
750 
751 /*
752  * Wait for some other vcpu thread to execute us, and
753  * wake us up when we need to handle something in the host.
754  */
755 static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
756 {
757 	DEFINE_WAIT(wait);
758 
759 	prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
760 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
761 		schedule();
762 	finish_wait(&vcpu->arch.cpu_run, &wait);
763 }
764 
765 /*
766  * All the vcpus in this vcore are idle, so wait for a decrementer
767  * or external interrupt to one of the vcpus.  vc->lock is held.
768  */
769 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
770 {
771 	DEFINE_WAIT(wait);
772 	struct kvm_vcpu *v;
773 	int all_idle = 1;
774 
775 	prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
776 	vc->vcore_state = VCORE_SLEEPING;
777 	spin_unlock(&vc->lock);
778 	list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
779 		if (!v->arch.ceded || v->arch.pending_exceptions) {
780 			all_idle = 0;
781 			break;
782 		}
783 	}
784 	if (all_idle)
785 		schedule();
786 	finish_wait(&vc->wq, &wait);
787 	spin_lock(&vc->lock);
788 	vc->vcore_state = VCORE_INACTIVE;
789 }
790 
791 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
792 {
793 	int n_ceded;
794 	int prev_state;
795 	struct kvmppc_vcore *vc;
796 	struct kvm_vcpu *v, *vn;
797 
798 	kvm_run->exit_reason = 0;
799 	vcpu->arch.ret = RESUME_GUEST;
800 	vcpu->arch.trap = 0;
801 
802 	/*
803 	 * Synchronize with other threads in this virtual core
804 	 */
805 	vc = vcpu->arch.vcore;
806 	spin_lock(&vc->lock);
807 	vcpu->arch.ceded = 0;
808 	vcpu->arch.run_task = current;
809 	vcpu->arch.kvm_run = kvm_run;
810 	prev_state = vcpu->arch.state;
811 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
812 	list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
813 	++vc->n_runnable;
814 
815 	/*
816 	 * This happens the first time this is called for a vcpu.
817 	 * If the vcore is already running, we may be able to start
818 	 * this thread straight away and have it join in.
819 	 */
820 	if (prev_state == KVMPPC_VCPU_STOPPED) {
821 		if (vc->vcore_state == VCORE_RUNNING &&
822 		    VCORE_EXIT_COUNT(vc) == 0) {
823 			vcpu->arch.ptid = vc->n_runnable - 1;
824 			kvmppc_start_thread(vcpu);
825 		}
826 
827 	} else if (prev_state == KVMPPC_VCPU_BUSY_IN_HOST)
828 		--vc->n_busy;
829 
830 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
831 	       !signal_pending(current)) {
832 		if (vc->n_busy || vc->vcore_state != VCORE_INACTIVE) {
833 			spin_unlock(&vc->lock);
834 			kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
835 			spin_lock(&vc->lock);
836 			continue;
837 		}
838 		n_ceded = 0;
839 		list_for_each_entry(v, &vc->runnable_threads, arch.run_list)
840 			n_ceded += v->arch.ceded;
841 		if (n_ceded == vc->n_runnable)
842 			kvmppc_vcore_blocked(vc);
843 		else
844 			kvmppc_run_core(vc);
845 
846 		list_for_each_entry_safe(v, vn, &vc->runnable_threads,
847 					 arch.run_list) {
848 			kvmppc_core_prepare_to_enter(v);
849 			if (signal_pending(v->arch.run_task)) {
850 				kvmppc_remove_runnable(vc, v);
851 				v->stat.signal_exits++;
852 				v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
853 				v->arch.ret = -EINTR;
854 				wake_up(&v->arch.cpu_run);
855 			}
856 		}
857 	}
858 
859 	if (signal_pending(current)) {
860 		if (vc->vcore_state == VCORE_RUNNING ||
861 		    vc->vcore_state == VCORE_EXITING) {
862 			spin_unlock(&vc->lock);
863 			kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
864 			spin_lock(&vc->lock);
865 		}
866 		if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
867 			kvmppc_remove_runnable(vc, vcpu);
868 			vcpu->stat.signal_exits++;
869 			kvm_run->exit_reason = KVM_EXIT_INTR;
870 			vcpu->arch.ret = -EINTR;
871 		}
872 	}
873 
874 	spin_unlock(&vc->lock);
875 	return vcpu->arch.ret;
876 }
877 
878 int kvmppc_vcpu_run(struct kvm_run *run, struct kvm_vcpu *vcpu)
879 {
880 	int r;
881 
882 	if (!vcpu->arch.sane) {
883 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
884 		return -EINVAL;
885 	}
886 
887 	kvmppc_core_prepare_to_enter(vcpu);
888 
889 	/* No need to go into the guest when all we'll do is come back out */
890 	if (signal_pending(current)) {
891 		run->exit_reason = KVM_EXIT_INTR;
892 		return -EINTR;
893 	}
894 
895 	/* On the first time here, set up VRMA or RMA */
896 	if (!vcpu->kvm->arch.rma_setup_done) {
897 		r = kvmppc_hv_setup_rma(vcpu);
898 		if (r)
899 			return r;
900 	}
901 
902 	flush_fp_to_thread(current);
903 	flush_altivec_to_thread(current);
904 	flush_vsx_to_thread(current);
905 	vcpu->arch.wqp = &vcpu->arch.vcore->wq;
906 	vcpu->arch.pgdir = current->mm->pgd;
907 
908 	do {
909 		r = kvmppc_run_vcpu(run, vcpu);
910 
911 		if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
912 		    !(vcpu->arch.shregs.msr & MSR_PR)) {
913 			r = kvmppc_pseries_do_hcall(vcpu);
914 			kvmppc_core_prepare_to_enter(vcpu);
915 		}
916 	} while (r == RESUME_GUEST);
917 	return r;
918 }
919 
920 static long kvmppc_stt_npages(unsigned long window_size)
921 {
922 	return ALIGN((window_size >> SPAPR_TCE_SHIFT)
923 		     * sizeof(u64), PAGE_SIZE) / PAGE_SIZE;
924 }
925 
926 static void release_spapr_tce_table(struct kvmppc_spapr_tce_table *stt)
927 {
928 	struct kvm *kvm = stt->kvm;
929 	int i;
930 
931 	mutex_lock(&kvm->lock);
932 	list_del(&stt->list);
933 	for (i = 0; i < kvmppc_stt_npages(stt->window_size); i++)
934 		__free_page(stt->pages[i]);
935 	kfree(stt);
936 	mutex_unlock(&kvm->lock);
937 
938 	kvm_put_kvm(kvm);
939 }
940 
941 static int kvm_spapr_tce_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
942 {
943 	struct kvmppc_spapr_tce_table *stt = vma->vm_file->private_data;
944 	struct page *page;
945 
946 	if (vmf->pgoff >= kvmppc_stt_npages(stt->window_size))
947 		return VM_FAULT_SIGBUS;
948 
949 	page = stt->pages[vmf->pgoff];
950 	get_page(page);
951 	vmf->page = page;
952 	return 0;
953 }
954 
955 static const struct vm_operations_struct kvm_spapr_tce_vm_ops = {
956 	.fault = kvm_spapr_tce_fault,
957 };
958 
959 static int kvm_spapr_tce_mmap(struct file *file, struct vm_area_struct *vma)
960 {
961 	vma->vm_ops = &kvm_spapr_tce_vm_ops;
962 	return 0;
963 }
964 
965 static int kvm_spapr_tce_release(struct inode *inode, struct file *filp)
966 {
967 	struct kvmppc_spapr_tce_table *stt = filp->private_data;
968 
969 	release_spapr_tce_table(stt);
970 	return 0;
971 }
972 
973 static struct file_operations kvm_spapr_tce_fops = {
974 	.mmap           = kvm_spapr_tce_mmap,
975 	.release	= kvm_spapr_tce_release,
976 };
977 
978 long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm,
979 				   struct kvm_create_spapr_tce *args)
980 {
981 	struct kvmppc_spapr_tce_table *stt = NULL;
982 	long npages;
983 	int ret = -ENOMEM;
984 	int i;
985 
986 	/* Check this LIOBN hasn't been previously allocated */
987 	list_for_each_entry(stt, &kvm->arch.spapr_tce_tables, list) {
988 		if (stt->liobn == args->liobn)
989 			return -EBUSY;
990 	}
991 
992 	npages = kvmppc_stt_npages(args->window_size);
993 
994 	stt = kzalloc(sizeof(*stt) + npages* sizeof(struct page *),
995 		      GFP_KERNEL);
996 	if (!stt)
997 		goto fail;
998 
999 	stt->liobn = args->liobn;
1000 	stt->window_size = args->window_size;
1001 	stt->kvm = kvm;
1002 
1003 	for (i = 0; i < npages; i++) {
1004 		stt->pages[i] = alloc_page(GFP_KERNEL | __GFP_ZERO);
1005 		if (!stt->pages[i])
1006 			goto fail;
1007 	}
1008 
1009 	kvm_get_kvm(kvm);
1010 
1011 	mutex_lock(&kvm->lock);
1012 	list_add(&stt->list, &kvm->arch.spapr_tce_tables);
1013 
1014 	mutex_unlock(&kvm->lock);
1015 
1016 	return anon_inode_getfd("kvm-spapr-tce", &kvm_spapr_tce_fops,
1017 				stt, O_RDWR);
1018 
1019 fail:
1020 	if (stt) {
1021 		for (i = 0; i < npages; i++)
1022 			if (stt->pages[i])
1023 				__free_page(stt->pages[i]);
1024 
1025 		kfree(stt);
1026 	}
1027 	return ret;
1028 }
1029 
1030 /* Work out RMLS (real mode limit selector) field value for a given RMA size.
1031    Assumes POWER7 or PPC970. */
1032 static inline int lpcr_rmls(unsigned long rma_size)
1033 {
1034 	switch (rma_size) {
1035 	case 32ul << 20:	/* 32 MB */
1036 		if (cpu_has_feature(CPU_FTR_ARCH_206))
1037 			return 8;	/* only supported on POWER7 */
1038 		return -1;
1039 	case 64ul << 20:	/* 64 MB */
1040 		return 3;
1041 	case 128ul << 20:	/* 128 MB */
1042 		return 7;
1043 	case 256ul << 20:	/* 256 MB */
1044 		return 4;
1045 	case 1ul << 30:		/* 1 GB */
1046 		return 2;
1047 	case 16ul << 30:	/* 16 GB */
1048 		return 1;
1049 	case 256ul << 30:	/* 256 GB */
1050 		return 0;
1051 	default:
1052 		return -1;
1053 	}
1054 }
1055 
1056 static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1057 {
1058 	struct kvmppc_linear_info *ri = vma->vm_file->private_data;
1059 	struct page *page;
1060 
1061 	if (vmf->pgoff >= ri->npages)
1062 		return VM_FAULT_SIGBUS;
1063 
1064 	page = pfn_to_page(ri->base_pfn + vmf->pgoff);
1065 	get_page(page);
1066 	vmf->page = page;
1067 	return 0;
1068 }
1069 
1070 static const struct vm_operations_struct kvm_rma_vm_ops = {
1071 	.fault = kvm_rma_fault,
1072 };
1073 
1074 static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma)
1075 {
1076 	vma->vm_flags |= VM_RESERVED;
1077 	vma->vm_ops = &kvm_rma_vm_ops;
1078 	return 0;
1079 }
1080 
1081 static int kvm_rma_release(struct inode *inode, struct file *filp)
1082 {
1083 	struct kvmppc_linear_info *ri = filp->private_data;
1084 
1085 	kvm_release_rma(ri);
1086 	return 0;
1087 }
1088 
1089 static struct file_operations kvm_rma_fops = {
1090 	.mmap           = kvm_rma_mmap,
1091 	.release	= kvm_rma_release,
1092 };
1093 
1094 long kvm_vm_ioctl_allocate_rma(struct kvm *kvm, struct kvm_allocate_rma *ret)
1095 {
1096 	struct kvmppc_linear_info *ri;
1097 	long fd;
1098 
1099 	ri = kvm_alloc_rma();
1100 	if (!ri)
1101 		return -ENOMEM;
1102 
1103 	fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR);
1104 	if (fd < 0)
1105 		kvm_release_rma(ri);
1106 
1107 	ret->rma_size = ri->npages << PAGE_SHIFT;
1108 	return fd;
1109 }
1110 
1111 /*
1112  * Get (and clear) the dirty memory log for a memory slot.
1113  */
1114 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1115 {
1116 	struct kvm_memory_slot *memslot;
1117 	int r;
1118 	unsigned long n;
1119 
1120 	mutex_lock(&kvm->slots_lock);
1121 
1122 	r = -EINVAL;
1123 	if (log->slot >= KVM_MEMORY_SLOTS)
1124 		goto out;
1125 
1126 	memslot = id_to_memslot(kvm->memslots, log->slot);
1127 	r = -ENOENT;
1128 	if (!memslot->dirty_bitmap)
1129 		goto out;
1130 
1131 	n = kvm_dirty_bitmap_bytes(memslot);
1132 	memset(memslot->dirty_bitmap, 0, n);
1133 
1134 	r = kvmppc_hv_get_dirty_log(kvm, memslot);
1135 	if (r)
1136 		goto out;
1137 
1138 	r = -EFAULT;
1139 	if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1140 		goto out;
1141 
1142 	r = 0;
1143 out:
1144 	mutex_unlock(&kvm->slots_lock);
1145 	return r;
1146 }
1147 
1148 static unsigned long slb_pgsize_encoding(unsigned long psize)
1149 {
1150 	unsigned long senc = 0;
1151 
1152 	if (psize > 0x1000) {
1153 		senc = SLB_VSID_L;
1154 		if (psize == 0x10000)
1155 			senc |= SLB_VSID_LP_01;
1156 	}
1157 	return senc;
1158 }
1159 
1160 int kvmppc_core_prepare_memory_region(struct kvm *kvm,
1161 				struct kvm_userspace_memory_region *mem)
1162 {
1163 	unsigned long npages;
1164 	unsigned long *phys;
1165 
1166 	/* Allocate a slot_phys array */
1167 	phys = kvm->arch.slot_phys[mem->slot];
1168 	if (!kvm->arch.using_mmu_notifiers && !phys) {
1169 		npages = mem->memory_size >> PAGE_SHIFT;
1170 		phys = vzalloc(npages * sizeof(unsigned long));
1171 		if (!phys)
1172 			return -ENOMEM;
1173 		kvm->arch.slot_phys[mem->slot] = phys;
1174 		kvm->arch.slot_npages[mem->slot] = npages;
1175 	}
1176 
1177 	return 0;
1178 }
1179 
1180 static void unpin_slot(struct kvm *kvm, int slot_id)
1181 {
1182 	unsigned long *physp;
1183 	unsigned long j, npages, pfn;
1184 	struct page *page;
1185 
1186 	physp = kvm->arch.slot_phys[slot_id];
1187 	npages = kvm->arch.slot_npages[slot_id];
1188 	if (physp) {
1189 		spin_lock(&kvm->arch.slot_phys_lock);
1190 		for (j = 0; j < npages; j++) {
1191 			if (!(physp[j] & KVMPPC_GOT_PAGE))
1192 				continue;
1193 			pfn = physp[j] >> PAGE_SHIFT;
1194 			page = pfn_to_page(pfn);
1195 			SetPageDirty(page);
1196 			put_page(page);
1197 		}
1198 		kvm->arch.slot_phys[slot_id] = NULL;
1199 		spin_unlock(&kvm->arch.slot_phys_lock);
1200 		vfree(physp);
1201 	}
1202 }
1203 
1204 void kvmppc_core_commit_memory_region(struct kvm *kvm,
1205 				struct kvm_userspace_memory_region *mem)
1206 {
1207 }
1208 
1209 static int kvmppc_hv_setup_rma(struct kvm_vcpu *vcpu)
1210 {
1211 	int err = 0;
1212 	struct kvm *kvm = vcpu->kvm;
1213 	struct kvmppc_linear_info *ri = NULL;
1214 	unsigned long hva;
1215 	struct kvm_memory_slot *memslot;
1216 	struct vm_area_struct *vma;
1217 	unsigned long lpcr, senc;
1218 	unsigned long psize, porder;
1219 	unsigned long rma_size;
1220 	unsigned long rmls;
1221 	unsigned long *physp;
1222 	unsigned long i, npages;
1223 
1224 	mutex_lock(&kvm->lock);
1225 	if (kvm->arch.rma_setup_done)
1226 		goto out;	/* another vcpu beat us to it */
1227 
1228 	/* Look up the memslot for guest physical address 0 */
1229 	memslot = gfn_to_memslot(kvm, 0);
1230 
1231 	/* We must have some memory at 0 by now */
1232 	err = -EINVAL;
1233 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
1234 		goto out;
1235 
1236 	/* Look up the VMA for the start of this memory slot */
1237 	hva = memslot->userspace_addr;
1238 	down_read(&current->mm->mmap_sem);
1239 	vma = find_vma(current->mm, hva);
1240 	if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
1241 		goto up_out;
1242 
1243 	psize = vma_kernel_pagesize(vma);
1244 	porder = __ilog2(psize);
1245 
1246 	/* Is this one of our preallocated RMAs? */
1247 	if (vma->vm_file && vma->vm_file->f_op == &kvm_rma_fops &&
1248 	    hva == vma->vm_start)
1249 		ri = vma->vm_file->private_data;
1250 
1251 	up_read(&current->mm->mmap_sem);
1252 
1253 	if (!ri) {
1254 		/* On POWER7, use VRMA; on PPC970, give up */
1255 		err = -EPERM;
1256 		if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1257 			pr_err("KVM: CPU requires an RMO\n");
1258 			goto out;
1259 		}
1260 
1261 		/* We can handle 4k, 64k or 16M pages in the VRMA */
1262 		err = -EINVAL;
1263 		if (!(psize == 0x1000 || psize == 0x10000 ||
1264 		      psize == 0x1000000))
1265 			goto out;
1266 
1267 		/* Update VRMASD field in the LPCR */
1268 		senc = slb_pgsize_encoding(psize);
1269 		kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
1270 			(VRMA_VSID << SLB_VSID_SHIFT_1T);
1271 		lpcr = kvm->arch.lpcr & ~LPCR_VRMASD;
1272 		lpcr |= senc << (LPCR_VRMASD_SH - 4);
1273 		kvm->arch.lpcr = lpcr;
1274 
1275 		/* Create HPTEs in the hash page table for the VRMA */
1276 		kvmppc_map_vrma(vcpu, memslot, porder);
1277 
1278 	} else {
1279 		/* Set up to use an RMO region */
1280 		rma_size = ri->npages;
1281 		if (rma_size > memslot->npages)
1282 			rma_size = memslot->npages;
1283 		rma_size <<= PAGE_SHIFT;
1284 		rmls = lpcr_rmls(rma_size);
1285 		err = -EINVAL;
1286 		if (rmls < 0) {
1287 			pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size);
1288 			goto out;
1289 		}
1290 		atomic_inc(&ri->use_count);
1291 		kvm->arch.rma = ri;
1292 
1293 		/* Update LPCR and RMOR */
1294 		lpcr = kvm->arch.lpcr;
1295 		if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1296 			/* PPC970; insert RMLS value (split field) in HID4 */
1297 			lpcr &= ~((1ul << HID4_RMLS0_SH) |
1298 				  (3ul << HID4_RMLS2_SH));
1299 			lpcr |= ((rmls >> 2) << HID4_RMLS0_SH) |
1300 				((rmls & 3) << HID4_RMLS2_SH);
1301 			/* RMOR is also in HID4 */
1302 			lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff)
1303 				<< HID4_RMOR_SH;
1304 		} else {
1305 			/* POWER7 */
1306 			lpcr &= ~(LPCR_VPM0 | LPCR_VRMA_L);
1307 			lpcr |= rmls << LPCR_RMLS_SH;
1308 			kvm->arch.rmor = kvm->arch.rma->base_pfn << PAGE_SHIFT;
1309 		}
1310 		kvm->arch.lpcr = lpcr;
1311 		pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n",
1312 			ri->base_pfn << PAGE_SHIFT, rma_size, lpcr);
1313 
1314 		/* Initialize phys addrs of pages in RMO */
1315 		npages = ri->npages;
1316 		porder = __ilog2(npages);
1317 		physp = kvm->arch.slot_phys[memslot->id];
1318 		spin_lock(&kvm->arch.slot_phys_lock);
1319 		for (i = 0; i < npages; ++i)
1320 			physp[i] = ((ri->base_pfn + i) << PAGE_SHIFT) + porder;
1321 		spin_unlock(&kvm->arch.slot_phys_lock);
1322 	}
1323 
1324 	/* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */
1325 	smp_wmb();
1326 	kvm->arch.rma_setup_done = 1;
1327 	err = 0;
1328  out:
1329 	mutex_unlock(&kvm->lock);
1330 	return err;
1331 
1332  up_out:
1333 	up_read(&current->mm->mmap_sem);
1334 	goto out;
1335 }
1336 
1337 int kvmppc_core_init_vm(struct kvm *kvm)
1338 {
1339 	long r;
1340 	unsigned long lpcr;
1341 
1342 	/* Allocate hashed page table */
1343 	r = kvmppc_alloc_hpt(kvm);
1344 	if (r)
1345 		return r;
1346 
1347 	INIT_LIST_HEAD(&kvm->arch.spapr_tce_tables);
1348 
1349 	kvm->arch.rma = NULL;
1350 
1351 	kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
1352 
1353 	if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1354 		/* PPC970; HID4 is effectively the LPCR */
1355 		unsigned long lpid = kvm->arch.lpid;
1356 		kvm->arch.host_lpid = 0;
1357 		kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4);
1358 		lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH));
1359 		lpcr |= ((lpid >> 4) << HID4_LPID1_SH) |
1360 			((lpid & 0xf) << HID4_LPID5_SH);
1361 	} else {
1362 		/* POWER7; init LPCR for virtual RMA mode */
1363 		kvm->arch.host_lpid = mfspr(SPRN_LPID);
1364 		kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
1365 		lpcr &= LPCR_PECE | LPCR_LPES;
1366 		lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
1367 			LPCR_VPM0 | LPCR_VPM1;
1368 		kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
1369 			(VRMA_VSID << SLB_VSID_SHIFT_1T);
1370 	}
1371 	kvm->arch.lpcr = lpcr;
1372 
1373 	kvm->arch.using_mmu_notifiers = !!cpu_has_feature(CPU_FTR_ARCH_206);
1374 	spin_lock_init(&kvm->arch.slot_phys_lock);
1375 	return 0;
1376 }
1377 
1378 void kvmppc_core_destroy_vm(struct kvm *kvm)
1379 {
1380 	unsigned long i;
1381 
1382 	if (!kvm->arch.using_mmu_notifiers)
1383 		for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
1384 			unpin_slot(kvm, i);
1385 
1386 	if (kvm->arch.rma) {
1387 		kvm_release_rma(kvm->arch.rma);
1388 		kvm->arch.rma = NULL;
1389 	}
1390 
1391 	kvmppc_free_hpt(kvm);
1392 	WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
1393 }
1394 
1395 /* These are stubs for now */
1396 void kvmppc_mmu_pte_pflush(struct kvm_vcpu *vcpu, ulong pa_start, ulong pa_end)
1397 {
1398 }
1399 
1400 /* We don't need to emulate any privileged instructions or dcbz */
1401 int kvmppc_core_emulate_op(struct kvm_run *run, struct kvm_vcpu *vcpu,
1402                            unsigned int inst, int *advance)
1403 {
1404 	return EMULATE_FAIL;
1405 }
1406 
1407 int kvmppc_core_emulate_mtspr(struct kvm_vcpu *vcpu, int sprn, int rs)
1408 {
1409 	return EMULATE_FAIL;
1410 }
1411 
1412 int kvmppc_core_emulate_mfspr(struct kvm_vcpu *vcpu, int sprn, int rt)
1413 {
1414 	return EMULATE_FAIL;
1415 }
1416 
1417 static int kvmppc_book3s_hv_init(void)
1418 {
1419 	int r;
1420 
1421 	r = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1422 
1423 	if (r)
1424 		return r;
1425 
1426 	r = kvmppc_mmu_hv_init();
1427 
1428 	return r;
1429 }
1430 
1431 static void kvmppc_book3s_hv_exit(void)
1432 {
1433 	kvm_exit();
1434 }
1435 
1436 module_init(kvmppc_book3s_hv_init);
1437 module_exit(kvmppc_book3s_hv_exit);
1438