xref: /openbmc/linux/arch/powerpc/kvm/book3s_hv.c (revision 089a49b6)
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 
56 /* #define EXIT_DEBUG */
57 /* #define EXIT_DEBUG_SIMPLE */
58 /* #define EXIT_DEBUG_INT */
59 
60 /* Used to indicate that a guest page fault needs to be handled */
61 #define RESUME_PAGE_FAULT	(RESUME_GUEST | RESUME_FLAG_ARCH1)
62 
63 /* Used as a "null" value for timebase values */
64 #define TB_NIL	(~(u64)0)
65 
66 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
67 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
68 
69 void kvmppc_fast_vcpu_kick(struct kvm_vcpu *vcpu)
70 {
71 	int me;
72 	int cpu = vcpu->cpu;
73 	wait_queue_head_t *wqp;
74 
75 	wqp = kvm_arch_vcpu_wq(vcpu);
76 	if (waitqueue_active(wqp)) {
77 		wake_up_interruptible(wqp);
78 		++vcpu->stat.halt_wakeup;
79 	}
80 
81 	me = get_cpu();
82 
83 	/* CPU points to the first thread of the core */
84 	if (cpu != me && cpu >= 0 && cpu < nr_cpu_ids) {
85 		int real_cpu = cpu + vcpu->arch.ptid;
86 		if (paca[real_cpu].kvm_hstate.xics_phys)
87 			xics_wake_cpu(real_cpu);
88 		else if (cpu_online(cpu))
89 			smp_send_reschedule(cpu);
90 	}
91 	put_cpu();
92 }
93 
94 /*
95  * We use the vcpu_load/put functions to measure stolen time.
96  * Stolen time is counted as time when either the vcpu is able to
97  * run as part of a virtual core, but the task running the vcore
98  * is preempted or sleeping, or when the vcpu needs something done
99  * in the kernel by the task running the vcpu, but that task is
100  * preempted or sleeping.  Those two things have to be counted
101  * separately, since one of the vcpu tasks will take on the job
102  * of running the core, and the other vcpu tasks in the vcore will
103  * sleep waiting for it to do that, but that sleep shouldn't count
104  * as stolen time.
105  *
106  * Hence we accumulate stolen time when the vcpu can run as part of
107  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
108  * needs its task to do other things in the kernel (for example,
109  * service a page fault) in busy_stolen.  We don't accumulate
110  * stolen time for a vcore when it is inactive, or for a vcpu
111  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
112  * a misnomer; it means that the vcpu task is not executing in
113  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
114  * the kernel.  We don't have any way of dividing up that time
115  * between time that the vcpu is genuinely stopped, time that
116  * the task is actively working on behalf of the vcpu, and time
117  * that the task is preempted, so we don't count any of it as
118  * stolen.
119  *
120  * Updates to busy_stolen are protected by arch.tbacct_lock;
121  * updates to vc->stolen_tb are protected by the arch.tbacct_lock
122  * of the vcpu that has taken responsibility for running the vcore
123  * (i.e. vc->runner).  The stolen times are measured in units of
124  * timebase ticks.  (Note that the != TB_NIL checks below are
125  * purely defensive; they should never fail.)
126  */
127 
128 void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
129 {
130 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
131 
132 	spin_lock(&vcpu->arch.tbacct_lock);
133 	if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE &&
134 	    vc->preempt_tb != TB_NIL) {
135 		vc->stolen_tb += mftb() - vc->preempt_tb;
136 		vc->preempt_tb = TB_NIL;
137 	}
138 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
139 	    vcpu->arch.busy_preempt != TB_NIL) {
140 		vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
141 		vcpu->arch.busy_preempt = TB_NIL;
142 	}
143 	spin_unlock(&vcpu->arch.tbacct_lock);
144 }
145 
146 void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
147 {
148 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
149 
150 	spin_lock(&vcpu->arch.tbacct_lock);
151 	if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE)
152 		vc->preempt_tb = mftb();
153 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
154 		vcpu->arch.busy_preempt = mftb();
155 	spin_unlock(&vcpu->arch.tbacct_lock);
156 }
157 
158 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
159 {
160 	vcpu->arch.shregs.msr = msr;
161 	kvmppc_end_cede(vcpu);
162 }
163 
164 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
165 {
166 	vcpu->arch.pvr = pvr;
167 }
168 
169 void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
170 {
171 	int r;
172 
173 	pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
174 	pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
175 	       vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
176 	for (r = 0; r < 16; ++r)
177 		pr_err("r%2d = %.16lx  r%d = %.16lx\n",
178 		       r, kvmppc_get_gpr(vcpu, r),
179 		       r+16, kvmppc_get_gpr(vcpu, r+16));
180 	pr_err("ctr = %.16lx  lr  = %.16lx\n",
181 	       vcpu->arch.ctr, vcpu->arch.lr);
182 	pr_err("srr0 = %.16llx srr1 = %.16llx\n",
183 	       vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
184 	pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
185 	       vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
186 	pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
187 	       vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
188 	pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
189 	       vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
190 	pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
191 	pr_err("fault dar = %.16lx dsisr = %.8x\n",
192 	       vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
193 	pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
194 	for (r = 0; r < vcpu->arch.slb_max; ++r)
195 		pr_err("  ESID = %.16llx VSID = %.16llx\n",
196 		       vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
197 	pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
198 	       vcpu->kvm->arch.lpcr, vcpu->kvm->arch.sdr1,
199 	       vcpu->arch.last_inst);
200 }
201 
202 struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
203 {
204 	int r;
205 	struct kvm_vcpu *v, *ret = NULL;
206 
207 	mutex_lock(&kvm->lock);
208 	kvm_for_each_vcpu(r, v, kvm) {
209 		if (v->vcpu_id == id) {
210 			ret = v;
211 			break;
212 		}
213 	}
214 	mutex_unlock(&kvm->lock);
215 	return ret;
216 }
217 
218 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
219 {
220 	vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
221 	vpa->yield_count = 1;
222 }
223 
224 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
225 		   unsigned long addr, unsigned long len)
226 {
227 	/* check address is cacheline aligned */
228 	if (addr & (L1_CACHE_BYTES - 1))
229 		return -EINVAL;
230 	spin_lock(&vcpu->arch.vpa_update_lock);
231 	if (v->next_gpa != addr || v->len != len) {
232 		v->next_gpa = addr;
233 		v->len = addr ? len : 0;
234 		v->update_pending = 1;
235 	}
236 	spin_unlock(&vcpu->arch.vpa_update_lock);
237 	return 0;
238 }
239 
240 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
241 struct reg_vpa {
242 	u32 dummy;
243 	union {
244 		u16 hword;
245 		u32 word;
246 	} length;
247 };
248 
249 static int vpa_is_registered(struct kvmppc_vpa *vpap)
250 {
251 	if (vpap->update_pending)
252 		return vpap->next_gpa != 0;
253 	return vpap->pinned_addr != NULL;
254 }
255 
256 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
257 				       unsigned long flags,
258 				       unsigned long vcpuid, unsigned long vpa)
259 {
260 	struct kvm *kvm = vcpu->kvm;
261 	unsigned long len, nb;
262 	void *va;
263 	struct kvm_vcpu *tvcpu;
264 	int err;
265 	int subfunc;
266 	struct kvmppc_vpa *vpap;
267 
268 	tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
269 	if (!tvcpu)
270 		return H_PARAMETER;
271 
272 	subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
273 	if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
274 	    subfunc == H_VPA_REG_SLB) {
275 		/* Registering new area - address must be cache-line aligned */
276 		if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
277 			return H_PARAMETER;
278 
279 		/* convert logical addr to kernel addr and read length */
280 		va = kvmppc_pin_guest_page(kvm, vpa, &nb);
281 		if (va == NULL)
282 			return H_PARAMETER;
283 		if (subfunc == H_VPA_REG_VPA)
284 			len = ((struct reg_vpa *)va)->length.hword;
285 		else
286 			len = ((struct reg_vpa *)va)->length.word;
287 		kvmppc_unpin_guest_page(kvm, va, vpa, false);
288 
289 		/* Check length */
290 		if (len > nb || len < sizeof(struct reg_vpa))
291 			return H_PARAMETER;
292 	} else {
293 		vpa = 0;
294 		len = 0;
295 	}
296 
297 	err = H_PARAMETER;
298 	vpap = NULL;
299 	spin_lock(&tvcpu->arch.vpa_update_lock);
300 
301 	switch (subfunc) {
302 	case H_VPA_REG_VPA:		/* register VPA */
303 		if (len < sizeof(struct lppaca))
304 			break;
305 		vpap = &tvcpu->arch.vpa;
306 		err = 0;
307 		break;
308 
309 	case H_VPA_REG_DTL:		/* register DTL */
310 		if (len < sizeof(struct dtl_entry))
311 			break;
312 		len -= len % sizeof(struct dtl_entry);
313 
314 		/* Check that they have previously registered a VPA */
315 		err = H_RESOURCE;
316 		if (!vpa_is_registered(&tvcpu->arch.vpa))
317 			break;
318 
319 		vpap = &tvcpu->arch.dtl;
320 		err = 0;
321 		break;
322 
323 	case H_VPA_REG_SLB:		/* register SLB shadow buffer */
324 		/* Check that they have previously registered a VPA */
325 		err = H_RESOURCE;
326 		if (!vpa_is_registered(&tvcpu->arch.vpa))
327 			break;
328 
329 		vpap = &tvcpu->arch.slb_shadow;
330 		err = 0;
331 		break;
332 
333 	case H_VPA_DEREG_VPA:		/* deregister VPA */
334 		/* Check they don't still have a DTL or SLB buf registered */
335 		err = H_RESOURCE;
336 		if (vpa_is_registered(&tvcpu->arch.dtl) ||
337 		    vpa_is_registered(&tvcpu->arch.slb_shadow))
338 			break;
339 
340 		vpap = &tvcpu->arch.vpa;
341 		err = 0;
342 		break;
343 
344 	case H_VPA_DEREG_DTL:		/* deregister DTL */
345 		vpap = &tvcpu->arch.dtl;
346 		err = 0;
347 		break;
348 
349 	case H_VPA_DEREG_SLB:		/* deregister SLB shadow buffer */
350 		vpap = &tvcpu->arch.slb_shadow;
351 		err = 0;
352 		break;
353 	}
354 
355 	if (vpap) {
356 		vpap->next_gpa = vpa;
357 		vpap->len = len;
358 		vpap->update_pending = 1;
359 	}
360 
361 	spin_unlock(&tvcpu->arch.vpa_update_lock);
362 
363 	return err;
364 }
365 
366 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
367 {
368 	struct kvm *kvm = vcpu->kvm;
369 	void *va;
370 	unsigned long nb;
371 	unsigned long gpa;
372 
373 	/*
374 	 * We need to pin the page pointed to by vpap->next_gpa,
375 	 * but we can't call kvmppc_pin_guest_page under the lock
376 	 * as it does get_user_pages() and down_read().  So we
377 	 * have to drop the lock, pin the page, then get the lock
378 	 * again and check that a new area didn't get registered
379 	 * in the meantime.
380 	 */
381 	for (;;) {
382 		gpa = vpap->next_gpa;
383 		spin_unlock(&vcpu->arch.vpa_update_lock);
384 		va = NULL;
385 		nb = 0;
386 		if (gpa)
387 			va = kvmppc_pin_guest_page(kvm, gpa, &nb);
388 		spin_lock(&vcpu->arch.vpa_update_lock);
389 		if (gpa == vpap->next_gpa)
390 			break;
391 		/* sigh... unpin that one and try again */
392 		if (va)
393 			kvmppc_unpin_guest_page(kvm, va, gpa, false);
394 	}
395 
396 	vpap->update_pending = 0;
397 	if (va && nb < vpap->len) {
398 		/*
399 		 * If it's now too short, it must be that userspace
400 		 * has changed the mappings underlying guest memory,
401 		 * so unregister the region.
402 		 */
403 		kvmppc_unpin_guest_page(kvm, va, gpa, false);
404 		va = NULL;
405 	}
406 	if (vpap->pinned_addr)
407 		kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
408 					vpap->dirty);
409 	vpap->gpa = gpa;
410 	vpap->pinned_addr = va;
411 	vpap->dirty = false;
412 	if (va)
413 		vpap->pinned_end = va + vpap->len;
414 }
415 
416 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
417 {
418 	if (!(vcpu->arch.vpa.update_pending ||
419 	      vcpu->arch.slb_shadow.update_pending ||
420 	      vcpu->arch.dtl.update_pending))
421 		return;
422 
423 	spin_lock(&vcpu->arch.vpa_update_lock);
424 	if (vcpu->arch.vpa.update_pending) {
425 		kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
426 		if (vcpu->arch.vpa.pinned_addr)
427 			init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
428 	}
429 	if (vcpu->arch.dtl.update_pending) {
430 		kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
431 		vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
432 		vcpu->arch.dtl_index = 0;
433 	}
434 	if (vcpu->arch.slb_shadow.update_pending)
435 		kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
436 	spin_unlock(&vcpu->arch.vpa_update_lock);
437 }
438 
439 /*
440  * Return the accumulated stolen time for the vcore up until `now'.
441  * The caller should hold the vcore lock.
442  */
443 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
444 {
445 	u64 p;
446 
447 	/*
448 	 * If we are the task running the vcore, then since we hold
449 	 * the vcore lock, we can't be preempted, so stolen_tb/preempt_tb
450 	 * can't be updated, so we don't need the tbacct_lock.
451 	 * If the vcore is inactive, it can't become active (since we
452 	 * hold the vcore lock), so the vcpu load/put functions won't
453 	 * update stolen_tb/preempt_tb, and we don't need tbacct_lock.
454 	 */
455 	if (vc->vcore_state != VCORE_INACTIVE &&
456 	    vc->runner->arch.run_task != current) {
457 		spin_lock(&vc->runner->arch.tbacct_lock);
458 		p = vc->stolen_tb;
459 		if (vc->preempt_tb != TB_NIL)
460 			p += now - vc->preempt_tb;
461 		spin_unlock(&vc->runner->arch.tbacct_lock);
462 	} else {
463 		p = vc->stolen_tb;
464 	}
465 	return p;
466 }
467 
468 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
469 				    struct kvmppc_vcore *vc)
470 {
471 	struct dtl_entry *dt;
472 	struct lppaca *vpa;
473 	unsigned long stolen;
474 	unsigned long core_stolen;
475 	u64 now;
476 
477 	dt = vcpu->arch.dtl_ptr;
478 	vpa = vcpu->arch.vpa.pinned_addr;
479 	now = mftb();
480 	core_stolen = vcore_stolen_time(vc, now);
481 	stolen = core_stolen - vcpu->arch.stolen_logged;
482 	vcpu->arch.stolen_logged = core_stolen;
483 	spin_lock(&vcpu->arch.tbacct_lock);
484 	stolen += vcpu->arch.busy_stolen;
485 	vcpu->arch.busy_stolen = 0;
486 	spin_unlock(&vcpu->arch.tbacct_lock);
487 	if (!dt || !vpa)
488 		return;
489 	memset(dt, 0, sizeof(struct dtl_entry));
490 	dt->dispatch_reason = 7;
491 	dt->processor_id = vc->pcpu + vcpu->arch.ptid;
492 	dt->timebase = now;
493 	dt->enqueue_to_dispatch_time = stolen;
494 	dt->srr0 = kvmppc_get_pc(vcpu);
495 	dt->srr1 = vcpu->arch.shregs.msr;
496 	++dt;
497 	if (dt == vcpu->arch.dtl.pinned_end)
498 		dt = vcpu->arch.dtl.pinned_addr;
499 	vcpu->arch.dtl_ptr = dt;
500 	/* order writing *dt vs. writing vpa->dtl_idx */
501 	smp_wmb();
502 	vpa->dtl_idx = ++vcpu->arch.dtl_index;
503 	vcpu->arch.dtl.dirty = true;
504 }
505 
506 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
507 {
508 	unsigned long req = kvmppc_get_gpr(vcpu, 3);
509 	unsigned long target, ret = H_SUCCESS;
510 	struct kvm_vcpu *tvcpu;
511 	int idx, rc;
512 
513 	switch (req) {
514 	case H_ENTER:
515 		idx = srcu_read_lock(&vcpu->kvm->srcu);
516 		ret = kvmppc_virtmode_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
517 					      kvmppc_get_gpr(vcpu, 5),
518 					      kvmppc_get_gpr(vcpu, 6),
519 					      kvmppc_get_gpr(vcpu, 7));
520 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
521 		break;
522 	case H_CEDE:
523 		break;
524 	case H_PROD:
525 		target = kvmppc_get_gpr(vcpu, 4);
526 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
527 		if (!tvcpu) {
528 			ret = H_PARAMETER;
529 			break;
530 		}
531 		tvcpu->arch.prodded = 1;
532 		smp_mb();
533 		if (vcpu->arch.ceded) {
534 			if (waitqueue_active(&vcpu->wq)) {
535 				wake_up_interruptible(&vcpu->wq);
536 				vcpu->stat.halt_wakeup++;
537 			}
538 		}
539 		break;
540 	case H_CONFER:
541 		break;
542 	case H_REGISTER_VPA:
543 		ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
544 					kvmppc_get_gpr(vcpu, 5),
545 					kvmppc_get_gpr(vcpu, 6));
546 		break;
547 	case H_RTAS:
548 		if (list_empty(&vcpu->kvm->arch.rtas_tokens))
549 			return RESUME_HOST;
550 
551 		rc = kvmppc_rtas_hcall(vcpu);
552 
553 		if (rc == -ENOENT)
554 			return RESUME_HOST;
555 		else if (rc == 0)
556 			break;
557 
558 		/* Send the error out to userspace via KVM_RUN */
559 		return rc;
560 
561 	case H_XIRR:
562 	case H_CPPR:
563 	case H_EOI:
564 	case H_IPI:
565 	case H_IPOLL:
566 	case H_XIRR_X:
567 		if (kvmppc_xics_enabled(vcpu)) {
568 			ret = kvmppc_xics_hcall(vcpu, req);
569 			break;
570 		} /* fallthrough */
571 	default:
572 		return RESUME_HOST;
573 	}
574 	kvmppc_set_gpr(vcpu, 3, ret);
575 	vcpu->arch.hcall_needed = 0;
576 	return RESUME_GUEST;
577 }
578 
579 static int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
580 			      struct task_struct *tsk)
581 {
582 	int r = RESUME_HOST;
583 
584 	vcpu->stat.sum_exits++;
585 
586 	run->exit_reason = KVM_EXIT_UNKNOWN;
587 	run->ready_for_interrupt_injection = 1;
588 	switch (vcpu->arch.trap) {
589 	/* We're good on these - the host merely wanted to get our attention */
590 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
591 		vcpu->stat.dec_exits++;
592 		r = RESUME_GUEST;
593 		break;
594 	case BOOK3S_INTERRUPT_EXTERNAL:
595 		vcpu->stat.ext_intr_exits++;
596 		r = RESUME_GUEST;
597 		break;
598 	case BOOK3S_INTERRUPT_PERFMON:
599 		r = RESUME_GUEST;
600 		break;
601 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
602 		/*
603 		 * Deliver a machine check interrupt to the guest.
604 		 * We have to do this, even if the host has handled the
605 		 * machine check, because machine checks use SRR0/1 and
606 		 * the interrupt might have trashed guest state in them.
607 		 */
608 		kvmppc_book3s_queue_irqprio(vcpu,
609 					    BOOK3S_INTERRUPT_MACHINE_CHECK);
610 		r = RESUME_GUEST;
611 		break;
612 	case BOOK3S_INTERRUPT_PROGRAM:
613 	{
614 		ulong flags;
615 		/*
616 		 * Normally program interrupts are delivered directly
617 		 * to the guest by the hardware, but we can get here
618 		 * as a result of a hypervisor emulation interrupt
619 		 * (e40) getting turned into a 700 by BML RTAS.
620 		 */
621 		flags = vcpu->arch.shregs.msr & 0x1f0000ull;
622 		kvmppc_core_queue_program(vcpu, flags);
623 		r = RESUME_GUEST;
624 		break;
625 	}
626 	case BOOK3S_INTERRUPT_SYSCALL:
627 	{
628 		/* hcall - punt to userspace */
629 		int i;
630 
631 		if (vcpu->arch.shregs.msr & MSR_PR) {
632 			/* sc 1 from userspace - reflect to guest syscall */
633 			kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_SYSCALL);
634 			r = RESUME_GUEST;
635 			break;
636 		}
637 		run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
638 		for (i = 0; i < 9; ++i)
639 			run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
640 		run->exit_reason = KVM_EXIT_PAPR_HCALL;
641 		vcpu->arch.hcall_needed = 1;
642 		r = RESUME_HOST;
643 		break;
644 	}
645 	/*
646 	 * We get these next two if the guest accesses a page which it thinks
647 	 * it has mapped but which is not actually present, either because
648 	 * it is for an emulated I/O device or because the corresonding
649 	 * host page has been paged out.  Any other HDSI/HISI interrupts
650 	 * have been handled already.
651 	 */
652 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
653 		r = RESUME_PAGE_FAULT;
654 		break;
655 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
656 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
657 		vcpu->arch.fault_dsisr = 0;
658 		r = RESUME_PAGE_FAULT;
659 		break;
660 	/*
661 	 * This occurs if the guest executes an illegal instruction.
662 	 * We just generate a program interrupt to the guest, since
663 	 * we don't emulate any guest instructions at this stage.
664 	 */
665 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
666 		kvmppc_core_queue_program(vcpu, 0x80000);
667 		r = RESUME_GUEST;
668 		break;
669 	default:
670 		kvmppc_dump_regs(vcpu);
671 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
672 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
673 			vcpu->arch.shregs.msr);
674 		r = RESUME_HOST;
675 		BUG();
676 		break;
677 	}
678 
679 	return r;
680 }
681 
682 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
683 				  struct kvm_sregs *sregs)
684 {
685 	int i;
686 
687 	memset(sregs, 0, sizeof(struct kvm_sregs));
688 	sregs->pvr = vcpu->arch.pvr;
689 	for (i = 0; i < vcpu->arch.slb_max; i++) {
690 		sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
691 		sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
692 	}
693 
694 	return 0;
695 }
696 
697 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
698 				  struct kvm_sregs *sregs)
699 {
700 	int i, j;
701 
702 	kvmppc_set_pvr(vcpu, sregs->pvr);
703 
704 	j = 0;
705 	for (i = 0; i < vcpu->arch.slb_nr; i++) {
706 		if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
707 			vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
708 			vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
709 			++j;
710 		}
711 	}
712 	vcpu->arch.slb_max = j;
713 
714 	return 0;
715 }
716 
717 int kvmppc_get_one_reg(struct kvm_vcpu *vcpu, u64 id, union kvmppc_one_reg *val)
718 {
719 	int r = 0;
720 	long int i;
721 
722 	switch (id) {
723 	case KVM_REG_PPC_HIOR:
724 		*val = get_reg_val(id, 0);
725 		break;
726 	case KVM_REG_PPC_DABR:
727 		*val = get_reg_val(id, vcpu->arch.dabr);
728 		break;
729 	case KVM_REG_PPC_DSCR:
730 		*val = get_reg_val(id, vcpu->arch.dscr);
731 		break;
732 	case KVM_REG_PPC_PURR:
733 		*val = get_reg_val(id, vcpu->arch.purr);
734 		break;
735 	case KVM_REG_PPC_SPURR:
736 		*val = get_reg_val(id, vcpu->arch.spurr);
737 		break;
738 	case KVM_REG_PPC_AMR:
739 		*val = get_reg_val(id, vcpu->arch.amr);
740 		break;
741 	case KVM_REG_PPC_UAMOR:
742 		*val = get_reg_val(id, vcpu->arch.uamor);
743 		break;
744 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRA:
745 		i = id - KVM_REG_PPC_MMCR0;
746 		*val = get_reg_val(id, vcpu->arch.mmcr[i]);
747 		break;
748 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
749 		i = id - KVM_REG_PPC_PMC1;
750 		*val = get_reg_val(id, vcpu->arch.pmc[i]);
751 		break;
752 #ifdef CONFIG_VSX
753 	case KVM_REG_PPC_FPR0 ... KVM_REG_PPC_FPR31:
754 		if (cpu_has_feature(CPU_FTR_VSX)) {
755 			/* VSX => FP reg i is stored in arch.vsr[2*i] */
756 			long int i = id - KVM_REG_PPC_FPR0;
757 			*val = get_reg_val(id, vcpu->arch.vsr[2 * i]);
758 		} else {
759 			/* let generic code handle it */
760 			r = -EINVAL;
761 		}
762 		break;
763 	case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31:
764 		if (cpu_has_feature(CPU_FTR_VSX)) {
765 			long int i = id - KVM_REG_PPC_VSR0;
766 			val->vsxval[0] = vcpu->arch.vsr[2 * i];
767 			val->vsxval[1] = vcpu->arch.vsr[2 * i + 1];
768 		} else {
769 			r = -ENXIO;
770 		}
771 		break;
772 #endif /* CONFIG_VSX */
773 	case KVM_REG_PPC_VPA_ADDR:
774 		spin_lock(&vcpu->arch.vpa_update_lock);
775 		*val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
776 		spin_unlock(&vcpu->arch.vpa_update_lock);
777 		break;
778 	case KVM_REG_PPC_VPA_SLB:
779 		spin_lock(&vcpu->arch.vpa_update_lock);
780 		val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
781 		val->vpaval.length = vcpu->arch.slb_shadow.len;
782 		spin_unlock(&vcpu->arch.vpa_update_lock);
783 		break;
784 	case KVM_REG_PPC_VPA_DTL:
785 		spin_lock(&vcpu->arch.vpa_update_lock);
786 		val->vpaval.addr = vcpu->arch.dtl.next_gpa;
787 		val->vpaval.length = vcpu->arch.dtl.len;
788 		spin_unlock(&vcpu->arch.vpa_update_lock);
789 		break;
790 	default:
791 		r = -EINVAL;
792 		break;
793 	}
794 
795 	return r;
796 }
797 
798 int kvmppc_set_one_reg(struct kvm_vcpu *vcpu, u64 id, union kvmppc_one_reg *val)
799 {
800 	int r = 0;
801 	long int i;
802 	unsigned long addr, len;
803 
804 	switch (id) {
805 	case KVM_REG_PPC_HIOR:
806 		/* Only allow this to be set to zero */
807 		if (set_reg_val(id, *val))
808 			r = -EINVAL;
809 		break;
810 	case KVM_REG_PPC_DABR:
811 		vcpu->arch.dabr = set_reg_val(id, *val);
812 		break;
813 	case KVM_REG_PPC_DSCR:
814 		vcpu->arch.dscr = set_reg_val(id, *val);
815 		break;
816 	case KVM_REG_PPC_PURR:
817 		vcpu->arch.purr = set_reg_val(id, *val);
818 		break;
819 	case KVM_REG_PPC_SPURR:
820 		vcpu->arch.spurr = set_reg_val(id, *val);
821 		break;
822 	case KVM_REG_PPC_AMR:
823 		vcpu->arch.amr = set_reg_val(id, *val);
824 		break;
825 	case KVM_REG_PPC_UAMOR:
826 		vcpu->arch.uamor = set_reg_val(id, *val);
827 		break;
828 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRA:
829 		i = id - KVM_REG_PPC_MMCR0;
830 		vcpu->arch.mmcr[i] = set_reg_val(id, *val);
831 		break;
832 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
833 		i = id - KVM_REG_PPC_PMC1;
834 		vcpu->arch.pmc[i] = set_reg_val(id, *val);
835 		break;
836 #ifdef CONFIG_VSX
837 	case KVM_REG_PPC_FPR0 ... KVM_REG_PPC_FPR31:
838 		if (cpu_has_feature(CPU_FTR_VSX)) {
839 			/* VSX => FP reg i is stored in arch.vsr[2*i] */
840 			long int i = id - KVM_REG_PPC_FPR0;
841 			vcpu->arch.vsr[2 * i] = set_reg_val(id, *val);
842 		} else {
843 			/* let generic code handle it */
844 			r = -EINVAL;
845 		}
846 		break;
847 	case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31:
848 		if (cpu_has_feature(CPU_FTR_VSX)) {
849 			long int i = id - KVM_REG_PPC_VSR0;
850 			vcpu->arch.vsr[2 * i] = val->vsxval[0];
851 			vcpu->arch.vsr[2 * i + 1] = val->vsxval[1];
852 		} else {
853 			r = -ENXIO;
854 		}
855 		break;
856 #endif /* CONFIG_VSX */
857 	case KVM_REG_PPC_VPA_ADDR:
858 		addr = set_reg_val(id, *val);
859 		r = -EINVAL;
860 		if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
861 			      vcpu->arch.dtl.next_gpa))
862 			break;
863 		r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
864 		break;
865 	case KVM_REG_PPC_VPA_SLB:
866 		addr = val->vpaval.addr;
867 		len = val->vpaval.length;
868 		r = -EINVAL;
869 		if (addr && !vcpu->arch.vpa.next_gpa)
870 			break;
871 		r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
872 		break;
873 	case KVM_REG_PPC_VPA_DTL:
874 		addr = val->vpaval.addr;
875 		len = val->vpaval.length;
876 		r = -EINVAL;
877 		if (addr && (len < sizeof(struct dtl_entry) ||
878 			     !vcpu->arch.vpa.next_gpa))
879 			break;
880 		len -= len % sizeof(struct dtl_entry);
881 		r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
882 		break;
883 	default:
884 		r = -EINVAL;
885 		break;
886 	}
887 
888 	return r;
889 }
890 
891 int kvmppc_core_check_processor_compat(void)
892 {
893 	if (cpu_has_feature(CPU_FTR_HVMODE))
894 		return 0;
895 	return -EIO;
896 }
897 
898 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
899 {
900 	struct kvm_vcpu *vcpu;
901 	int err = -EINVAL;
902 	int core;
903 	struct kvmppc_vcore *vcore;
904 
905 	core = id / threads_per_core;
906 	if (core >= KVM_MAX_VCORES)
907 		goto out;
908 
909 	err = -ENOMEM;
910 	vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
911 	if (!vcpu)
912 		goto out;
913 
914 	err = kvm_vcpu_init(vcpu, kvm, id);
915 	if (err)
916 		goto free_vcpu;
917 
918 	vcpu->arch.shared = &vcpu->arch.shregs;
919 	vcpu->arch.mmcr[0] = MMCR0_FC;
920 	vcpu->arch.ctrl = CTRL_RUNLATCH;
921 	/* default to host PVR, since we can't spoof it */
922 	vcpu->arch.pvr = mfspr(SPRN_PVR);
923 	kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
924 	spin_lock_init(&vcpu->arch.vpa_update_lock);
925 	spin_lock_init(&vcpu->arch.tbacct_lock);
926 	vcpu->arch.busy_preempt = TB_NIL;
927 
928 	kvmppc_mmu_book3s_hv_init(vcpu);
929 
930 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
931 
932 	init_waitqueue_head(&vcpu->arch.cpu_run);
933 
934 	mutex_lock(&kvm->lock);
935 	vcore = kvm->arch.vcores[core];
936 	if (!vcore) {
937 		vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
938 		if (vcore) {
939 			INIT_LIST_HEAD(&vcore->runnable_threads);
940 			spin_lock_init(&vcore->lock);
941 			init_waitqueue_head(&vcore->wq);
942 			vcore->preempt_tb = TB_NIL;
943 		}
944 		kvm->arch.vcores[core] = vcore;
945 		kvm->arch.online_vcores++;
946 	}
947 	mutex_unlock(&kvm->lock);
948 
949 	if (!vcore)
950 		goto free_vcpu;
951 
952 	spin_lock(&vcore->lock);
953 	++vcore->num_threads;
954 	spin_unlock(&vcore->lock);
955 	vcpu->arch.vcore = vcore;
956 
957 	vcpu->arch.cpu_type = KVM_CPU_3S_64;
958 	kvmppc_sanity_check(vcpu);
959 
960 	return vcpu;
961 
962 free_vcpu:
963 	kmem_cache_free(kvm_vcpu_cache, vcpu);
964 out:
965 	return ERR_PTR(err);
966 }
967 
968 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
969 {
970 	if (vpa->pinned_addr)
971 		kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
972 					vpa->dirty);
973 }
974 
975 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
976 {
977 	spin_lock(&vcpu->arch.vpa_update_lock);
978 	unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
979 	unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
980 	unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
981 	spin_unlock(&vcpu->arch.vpa_update_lock);
982 	kvm_vcpu_uninit(vcpu);
983 	kmem_cache_free(kvm_vcpu_cache, vcpu);
984 }
985 
986 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
987 {
988 	unsigned long dec_nsec, now;
989 
990 	now = get_tb();
991 	if (now > vcpu->arch.dec_expires) {
992 		/* decrementer has already gone negative */
993 		kvmppc_core_queue_dec(vcpu);
994 		kvmppc_core_prepare_to_enter(vcpu);
995 		return;
996 	}
997 	dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
998 		   / tb_ticks_per_sec;
999 	hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1000 		      HRTIMER_MODE_REL);
1001 	vcpu->arch.timer_running = 1;
1002 }
1003 
1004 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1005 {
1006 	vcpu->arch.ceded = 0;
1007 	if (vcpu->arch.timer_running) {
1008 		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1009 		vcpu->arch.timer_running = 0;
1010 	}
1011 }
1012 
1013 extern int __kvmppc_vcore_entry(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu);
1014 
1015 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1016 				   struct kvm_vcpu *vcpu)
1017 {
1018 	u64 now;
1019 
1020 	if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1021 		return;
1022 	spin_lock(&vcpu->arch.tbacct_lock);
1023 	now = mftb();
1024 	vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1025 		vcpu->arch.stolen_logged;
1026 	vcpu->arch.busy_preempt = now;
1027 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1028 	spin_unlock(&vcpu->arch.tbacct_lock);
1029 	--vc->n_runnable;
1030 	list_del(&vcpu->arch.run_list);
1031 }
1032 
1033 static int kvmppc_grab_hwthread(int cpu)
1034 {
1035 	struct paca_struct *tpaca;
1036 	long timeout = 1000;
1037 
1038 	tpaca = &paca[cpu];
1039 
1040 	/* Ensure the thread won't go into the kernel if it wakes */
1041 	tpaca->kvm_hstate.hwthread_req = 1;
1042 	tpaca->kvm_hstate.kvm_vcpu = NULL;
1043 
1044 	/*
1045 	 * If the thread is already executing in the kernel (e.g. handling
1046 	 * a stray interrupt), wait for it to get back to nap mode.
1047 	 * The smp_mb() is to ensure that our setting of hwthread_req
1048 	 * is visible before we look at hwthread_state, so if this
1049 	 * races with the code at system_reset_pSeries and the thread
1050 	 * misses our setting of hwthread_req, we are sure to see its
1051 	 * setting of hwthread_state, and vice versa.
1052 	 */
1053 	smp_mb();
1054 	while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1055 		if (--timeout <= 0) {
1056 			pr_err("KVM: couldn't grab cpu %d\n", cpu);
1057 			return -EBUSY;
1058 		}
1059 		udelay(1);
1060 	}
1061 	return 0;
1062 }
1063 
1064 static void kvmppc_release_hwthread(int cpu)
1065 {
1066 	struct paca_struct *tpaca;
1067 
1068 	tpaca = &paca[cpu];
1069 	tpaca->kvm_hstate.hwthread_req = 0;
1070 	tpaca->kvm_hstate.kvm_vcpu = NULL;
1071 }
1072 
1073 static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
1074 {
1075 	int cpu;
1076 	struct paca_struct *tpaca;
1077 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
1078 
1079 	if (vcpu->arch.timer_running) {
1080 		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1081 		vcpu->arch.timer_running = 0;
1082 	}
1083 	cpu = vc->pcpu + vcpu->arch.ptid;
1084 	tpaca = &paca[cpu];
1085 	tpaca->kvm_hstate.kvm_vcpu = vcpu;
1086 	tpaca->kvm_hstate.kvm_vcore = vc;
1087 	tpaca->kvm_hstate.napping = 0;
1088 	vcpu->cpu = vc->pcpu;
1089 	smp_wmb();
1090 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
1091 	if (vcpu->arch.ptid) {
1092 		xics_wake_cpu(cpu);
1093 		++vc->n_woken;
1094 	}
1095 #endif
1096 }
1097 
1098 static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc)
1099 {
1100 	int i;
1101 
1102 	HMT_low();
1103 	i = 0;
1104 	while (vc->nap_count < vc->n_woken) {
1105 		if (++i >= 1000000) {
1106 			pr_err("kvmppc_wait_for_nap timeout %d %d\n",
1107 			       vc->nap_count, vc->n_woken);
1108 			break;
1109 		}
1110 		cpu_relax();
1111 	}
1112 	HMT_medium();
1113 }
1114 
1115 /*
1116  * Check that we are on thread 0 and that any other threads in
1117  * this core are off-line.  Then grab the threads so they can't
1118  * enter the kernel.
1119  */
1120 static int on_primary_thread(void)
1121 {
1122 	int cpu = smp_processor_id();
1123 	int thr = cpu_thread_in_core(cpu);
1124 
1125 	if (thr)
1126 		return 0;
1127 	while (++thr < threads_per_core)
1128 		if (cpu_online(cpu + thr))
1129 			return 0;
1130 
1131 	/* Grab all hw threads so they can't go into the kernel */
1132 	for (thr = 1; thr < threads_per_core; ++thr) {
1133 		if (kvmppc_grab_hwthread(cpu + thr)) {
1134 			/* Couldn't grab one; let the others go */
1135 			do {
1136 				kvmppc_release_hwthread(cpu + thr);
1137 			} while (--thr > 0);
1138 			return 0;
1139 		}
1140 	}
1141 	return 1;
1142 }
1143 
1144 /*
1145  * Run a set of guest threads on a physical core.
1146  * Called with vc->lock held.
1147  */
1148 static void kvmppc_run_core(struct kvmppc_vcore *vc)
1149 {
1150 	struct kvm_vcpu *vcpu, *vcpu0, *vnext;
1151 	long ret;
1152 	u64 now;
1153 	int ptid, i, need_vpa_update;
1154 	int srcu_idx;
1155 	struct kvm_vcpu *vcpus_to_update[threads_per_core];
1156 
1157 	/* don't start if any threads have a signal pending */
1158 	need_vpa_update = 0;
1159 	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1160 		if (signal_pending(vcpu->arch.run_task))
1161 			return;
1162 		if (vcpu->arch.vpa.update_pending ||
1163 		    vcpu->arch.slb_shadow.update_pending ||
1164 		    vcpu->arch.dtl.update_pending)
1165 			vcpus_to_update[need_vpa_update++] = vcpu;
1166 	}
1167 
1168 	/*
1169 	 * Initialize *vc, in particular vc->vcore_state, so we can
1170 	 * drop the vcore lock if necessary.
1171 	 */
1172 	vc->n_woken = 0;
1173 	vc->nap_count = 0;
1174 	vc->entry_exit_count = 0;
1175 	vc->vcore_state = VCORE_STARTING;
1176 	vc->in_guest = 0;
1177 	vc->napping_threads = 0;
1178 
1179 	/*
1180 	 * Updating any of the vpas requires calling kvmppc_pin_guest_page,
1181 	 * which can't be called with any spinlocks held.
1182 	 */
1183 	if (need_vpa_update) {
1184 		spin_unlock(&vc->lock);
1185 		for (i = 0; i < need_vpa_update; ++i)
1186 			kvmppc_update_vpas(vcpus_to_update[i]);
1187 		spin_lock(&vc->lock);
1188 	}
1189 
1190 	/*
1191 	 * Assign physical thread IDs, first to non-ceded vcpus
1192 	 * and then to ceded ones.
1193 	 */
1194 	ptid = 0;
1195 	vcpu0 = NULL;
1196 	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1197 		if (!vcpu->arch.ceded) {
1198 			if (!ptid)
1199 				vcpu0 = vcpu;
1200 			vcpu->arch.ptid = ptid++;
1201 		}
1202 	}
1203 	if (!vcpu0)
1204 		goto out;	/* nothing to run; should never happen */
1205 	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1206 		if (vcpu->arch.ceded)
1207 			vcpu->arch.ptid = ptid++;
1208 
1209 	/*
1210 	 * Make sure we are running on thread 0, and that
1211 	 * secondary threads are offline.
1212 	 */
1213 	if (threads_per_core > 1 && !on_primary_thread()) {
1214 		list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1215 			vcpu->arch.ret = -EBUSY;
1216 		goto out;
1217 	}
1218 
1219 	vc->pcpu = smp_processor_id();
1220 	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1221 		kvmppc_start_thread(vcpu);
1222 		kvmppc_create_dtl_entry(vcpu, vc);
1223 	}
1224 
1225 	vc->vcore_state = VCORE_RUNNING;
1226 	preempt_disable();
1227 	spin_unlock(&vc->lock);
1228 
1229 	kvm_guest_enter();
1230 
1231 	srcu_idx = srcu_read_lock(&vcpu0->kvm->srcu);
1232 
1233 	__kvmppc_vcore_entry(NULL, vcpu0);
1234 
1235 	spin_lock(&vc->lock);
1236 	/* disable sending of IPIs on virtual external irqs */
1237 	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1238 		vcpu->cpu = -1;
1239 	/* wait for secondary threads to finish writing their state to memory */
1240 	if (vc->nap_count < vc->n_woken)
1241 		kvmppc_wait_for_nap(vc);
1242 	for (i = 0; i < threads_per_core; ++i)
1243 		kvmppc_release_hwthread(vc->pcpu + i);
1244 	/* prevent other vcpu threads from doing kvmppc_start_thread() now */
1245 	vc->vcore_state = VCORE_EXITING;
1246 	spin_unlock(&vc->lock);
1247 
1248 	srcu_read_unlock(&vcpu0->kvm->srcu, srcu_idx);
1249 
1250 	/* make sure updates to secondary vcpu structs are visible now */
1251 	smp_mb();
1252 	kvm_guest_exit();
1253 
1254 	preempt_enable();
1255 	kvm_resched(vcpu);
1256 
1257 	spin_lock(&vc->lock);
1258 	now = get_tb();
1259 	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1260 		/* cancel pending dec exception if dec is positive */
1261 		if (now < vcpu->arch.dec_expires &&
1262 		    kvmppc_core_pending_dec(vcpu))
1263 			kvmppc_core_dequeue_dec(vcpu);
1264 
1265 		ret = RESUME_GUEST;
1266 		if (vcpu->arch.trap)
1267 			ret = kvmppc_handle_exit(vcpu->arch.kvm_run, vcpu,
1268 						 vcpu->arch.run_task);
1269 
1270 		vcpu->arch.ret = ret;
1271 		vcpu->arch.trap = 0;
1272 
1273 		if (vcpu->arch.ceded) {
1274 			if (ret != RESUME_GUEST)
1275 				kvmppc_end_cede(vcpu);
1276 			else
1277 				kvmppc_set_timer(vcpu);
1278 		}
1279 	}
1280 
1281  out:
1282 	vc->vcore_state = VCORE_INACTIVE;
1283 	list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1284 				 arch.run_list) {
1285 		if (vcpu->arch.ret != RESUME_GUEST) {
1286 			kvmppc_remove_runnable(vc, vcpu);
1287 			wake_up(&vcpu->arch.cpu_run);
1288 		}
1289 	}
1290 }
1291 
1292 /*
1293  * Wait for some other vcpu thread to execute us, and
1294  * wake us up when we need to handle something in the host.
1295  */
1296 static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
1297 {
1298 	DEFINE_WAIT(wait);
1299 
1300 	prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
1301 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
1302 		schedule();
1303 	finish_wait(&vcpu->arch.cpu_run, &wait);
1304 }
1305 
1306 /*
1307  * All the vcpus in this vcore are idle, so wait for a decrementer
1308  * or external interrupt to one of the vcpus.  vc->lock is held.
1309  */
1310 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
1311 {
1312 	DEFINE_WAIT(wait);
1313 
1314 	prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
1315 	vc->vcore_state = VCORE_SLEEPING;
1316 	spin_unlock(&vc->lock);
1317 	schedule();
1318 	finish_wait(&vc->wq, &wait);
1319 	spin_lock(&vc->lock);
1320 	vc->vcore_state = VCORE_INACTIVE;
1321 }
1322 
1323 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1324 {
1325 	int n_ceded;
1326 	struct kvmppc_vcore *vc;
1327 	struct kvm_vcpu *v, *vn;
1328 
1329 	kvm_run->exit_reason = 0;
1330 	vcpu->arch.ret = RESUME_GUEST;
1331 	vcpu->arch.trap = 0;
1332 	kvmppc_update_vpas(vcpu);
1333 
1334 	/*
1335 	 * Synchronize with other threads in this virtual core
1336 	 */
1337 	vc = vcpu->arch.vcore;
1338 	spin_lock(&vc->lock);
1339 	vcpu->arch.ceded = 0;
1340 	vcpu->arch.run_task = current;
1341 	vcpu->arch.kvm_run = kvm_run;
1342 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
1343 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
1344 	vcpu->arch.busy_preempt = TB_NIL;
1345 	list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
1346 	++vc->n_runnable;
1347 
1348 	/*
1349 	 * This happens the first time this is called for a vcpu.
1350 	 * If the vcore is already running, we may be able to start
1351 	 * this thread straight away and have it join in.
1352 	 */
1353 	if (!signal_pending(current)) {
1354 		if (vc->vcore_state == VCORE_RUNNING &&
1355 		    VCORE_EXIT_COUNT(vc) == 0) {
1356 			vcpu->arch.ptid = vc->n_runnable - 1;
1357 			kvmppc_create_dtl_entry(vcpu, vc);
1358 			kvmppc_start_thread(vcpu);
1359 		} else if (vc->vcore_state == VCORE_SLEEPING) {
1360 			wake_up(&vc->wq);
1361 		}
1362 
1363 	}
1364 
1365 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1366 	       !signal_pending(current)) {
1367 		if (vc->vcore_state != VCORE_INACTIVE) {
1368 			spin_unlock(&vc->lock);
1369 			kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
1370 			spin_lock(&vc->lock);
1371 			continue;
1372 		}
1373 		list_for_each_entry_safe(v, vn, &vc->runnable_threads,
1374 					 arch.run_list) {
1375 			kvmppc_core_prepare_to_enter(v);
1376 			if (signal_pending(v->arch.run_task)) {
1377 				kvmppc_remove_runnable(vc, v);
1378 				v->stat.signal_exits++;
1379 				v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
1380 				v->arch.ret = -EINTR;
1381 				wake_up(&v->arch.cpu_run);
1382 			}
1383 		}
1384 		if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1385 			break;
1386 		vc->runner = vcpu;
1387 		n_ceded = 0;
1388 		list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
1389 			if (!v->arch.pending_exceptions)
1390 				n_ceded += v->arch.ceded;
1391 			else
1392 				v->arch.ceded = 0;
1393 		}
1394 		if (n_ceded == vc->n_runnable)
1395 			kvmppc_vcore_blocked(vc);
1396 		else
1397 			kvmppc_run_core(vc);
1398 		vc->runner = NULL;
1399 	}
1400 
1401 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1402 	       (vc->vcore_state == VCORE_RUNNING ||
1403 		vc->vcore_state == VCORE_EXITING)) {
1404 		spin_unlock(&vc->lock);
1405 		kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
1406 		spin_lock(&vc->lock);
1407 	}
1408 
1409 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
1410 		kvmppc_remove_runnable(vc, vcpu);
1411 		vcpu->stat.signal_exits++;
1412 		kvm_run->exit_reason = KVM_EXIT_INTR;
1413 		vcpu->arch.ret = -EINTR;
1414 	}
1415 
1416 	if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
1417 		/* Wake up some vcpu to run the core */
1418 		v = list_first_entry(&vc->runnable_threads,
1419 				     struct kvm_vcpu, arch.run_list);
1420 		wake_up(&v->arch.cpu_run);
1421 	}
1422 
1423 	spin_unlock(&vc->lock);
1424 	return vcpu->arch.ret;
1425 }
1426 
1427 int kvmppc_vcpu_run(struct kvm_run *run, struct kvm_vcpu *vcpu)
1428 {
1429 	int r;
1430 	int srcu_idx;
1431 
1432 	if (!vcpu->arch.sane) {
1433 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1434 		return -EINVAL;
1435 	}
1436 
1437 	kvmppc_core_prepare_to_enter(vcpu);
1438 
1439 	/* No need to go into the guest when all we'll do is come back out */
1440 	if (signal_pending(current)) {
1441 		run->exit_reason = KVM_EXIT_INTR;
1442 		return -EINTR;
1443 	}
1444 
1445 	atomic_inc(&vcpu->kvm->arch.vcpus_running);
1446 	/* Order vcpus_running vs. rma_setup_done, see kvmppc_alloc_reset_hpt */
1447 	smp_mb();
1448 
1449 	/* On the first time here, set up HTAB and VRMA or RMA */
1450 	if (!vcpu->kvm->arch.rma_setup_done) {
1451 		r = kvmppc_hv_setup_htab_rma(vcpu);
1452 		if (r)
1453 			goto out;
1454 	}
1455 
1456 	flush_fp_to_thread(current);
1457 	flush_altivec_to_thread(current);
1458 	flush_vsx_to_thread(current);
1459 	vcpu->arch.wqp = &vcpu->arch.vcore->wq;
1460 	vcpu->arch.pgdir = current->mm->pgd;
1461 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1462 
1463 	do {
1464 		r = kvmppc_run_vcpu(run, vcpu);
1465 
1466 		if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
1467 		    !(vcpu->arch.shregs.msr & MSR_PR)) {
1468 			r = kvmppc_pseries_do_hcall(vcpu);
1469 			kvmppc_core_prepare_to_enter(vcpu);
1470 		} else if (r == RESUME_PAGE_FAULT) {
1471 			srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1472 			r = kvmppc_book3s_hv_page_fault(run, vcpu,
1473 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1474 			srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1475 		}
1476 	} while (r == RESUME_GUEST);
1477 
1478  out:
1479 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1480 	atomic_dec(&vcpu->kvm->arch.vcpus_running);
1481 	return r;
1482 }
1483 
1484 
1485 /* Work out RMLS (real mode limit selector) field value for a given RMA size.
1486    Assumes POWER7 or PPC970. */
1487 static inline int lpcr_rmls(unsigned long rma_size)
1488 {
1489 	switch (rma_size) {
1490 	case 32ul << 20:	/* 32 MB */
1491 		if (cpu_has_feature(CPU_FTR_ARCH_206))
1492 			return 8;	/* only supported on POWER7 */
1493 		return -1;
1494 	case 64ul << 20:	/* 64 MB */
1495 		return 3;
1496 	case 128ul << 20:	/* 128 MB */
1497 		return 7;
1498 	case 256ul << 20:	/* 256 MB */
1499 		return 4;
1500 	case 1ul << 30:		/* 1 GB */
1501 		return 2;
1502 	case 16ul << 30:	/* 16 GB */
1503 		return 1;
1504 	case 256ul << 30:	/* 256 GB */
1505 		return 0;
1506 	default:
1507 		return -1;
1508 	}
1509 }
1510 
1511 static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1512 {
1513 	struct page *page;
1514 	struct kvm_rma_info *ri = vma->vm_file->private_data;
1515 
1516 	if (vmf->pgoff >= kvm_rma_pages)
1517 		return VM_FAULT_SIGBUS;
1518 
1519 	page = pfn_to_page(ri->base_pfn + vmf->pgoff);
1520 	get_page(page);
1521 	vmf->page = page;
1522 	return 0;
1523 }
1524 
1525 static const struct vm_operations_struct kvm_rma_vm_ops = {
1526 	.fault = kvm_rma_fault,
1527 };
1528 
1529 static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma)
1530 {
1531 	vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
1532 	vma->vm_ops = &kvm_rma_vm_ops;
1533 	return 0;
1534 }
1535 
1536 static int kvm_rma_release(struct inode *inode, struct file *filp)
1537 {
1538 	struct kvm_rma_info *ri = filp->private_data;
1539 
1540 	kvm_release_rma(ri);
1541 	return 0;
1542 }
1543 
1544 static const struct file_operations kvm_rma_fops = {
1545 	.mmap           = kvm_rma_mmap,
1546 	.release	= kvm_rma_release,
1547 };
1548 
1549 long kvm_vm_ioctl_allocate_rma(struct kvm *kvm, struct kvm_allocate_rma *ret)
1550 {
1551 	long fd;
1552 	struct kvm_rma_info *ri;
1553 	/*
1554 	 * Only do this on PPC970 in HV mode
1555 	 */
1556 	if (!cpu_has_feature(CPU_FTR_HVMODE) ||
1557 	    !cpu_has_feature(CPU_FTR_ARCH_201))
1558 		return -EINVAL;
1559 
1560 	if (!kvm_rma_pages)
1561 		return -EINVAL;
1562 
1563 	ri = kvm_alloc_rma();
1564 	if (!ri)
1565 		return -ENOMEM;
1566 
1567 	fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR | O_CLOEXEC);
1568 	if (fd < 0)
1569 		kvm_release_rma(ri);
1570 
1571 	ret->rma_size = kvm_rma_pages << PAGE_SHIFT;
1572 	return fd;
1573 }
1574 
1575 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
1576 				     int linux_psize)
1577 {
1578 	struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
1579 
1580 	if (!def->shift)
1581 		return;
1582 	(*sps)->page_shift = def->shift;
1583 	(*sps)->slb_enc = def->sllp;
1584 	(*sps)->enc[0].page_shift = def->shift;
1585 	/*
1586 	 * Only return base page encoding. We don't want to return
1587 	 * all the supporting pte_enc, because our H_ENTER doesn't
1588 	 * support MPSS yet. Once they do, we can start passing all
1589 	 * support pte_enc here
1590 	 */
1591 	(*sps)->enc[0].pte_enc = def->penc[linux_psize];
1592 	(*sps)++;
1593 }
1594 
1595 int kvm_vm_ioctl_get_smmu_info(struct kvm *kvm, struct kvm_ppc_smmu_info *info)
1596 {
1597 	struct kvm_ppc_one_seg_page_size *sps;
1598 
1599 	info->flags = KVM_PPC_PAGE_SIZES_REAL;
1600 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1601 		info->flags |= KVM_PPC_1T_SEGMENTS;
1602 	info->slb_size = mmu_slb_size;
1603 
1604 	/* We only support these sizes for now, and no muti-size segments */
1605 	sps = &info->sps[0];
1606 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
1607 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
1608 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
1609 
1610 	return 0;
1611 }
1612 
1613 /*
1614  * Get (and clear) the dirty memory log for a memory slot.
1615  */
1616 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1617 {
1618 	struct kvm_memory_slot *memslot;
1619 	int r;
1620 	unsigned long n;
1621 
1622 	mutex_lock(&kvm->slots_lock);
1623 
1624 	r = -EINVAL;
1625 	if (log->slot >= KVM_USER_MEM_SLOTS)
1626 		goto out;
1627 
1628 	memslot = id_to_memslot(kvm->memslots, log->slot);
1629 	r = -ENOENT;
1630 	if (!memslot->dirty_bitmap)
1631 		goto out;
1632 
1633 	n = kvm_dirty_bitmap_bytes(memslot);
1634 	memset(memslot->dirty_bitmap, 0, n);
1635 
1636 	r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
1637 	if (r)
1638 		goto out;
1639 
1640 	r = -EFAULT;
1641 	if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1642 		goto out;
1643 
1644 	r = 0;
1645 out:
1646 	mutex_unlock(&kvm->slots_lock);
1647 	return r;
1648 }
1649 
1650 static void unpin_slot(struct kvm_memory_slot *memslot)
1651 {
1652 	unsigned long *physp;
1653 	unsigned long j, npages, pfn;
1654 	struct page *page;
1655 
1656 	physp = memslot->arch.slot_phys;
1657 	npages = memslot->npages;
1658 	if (!physp)
1659 		return;
1660 	for (j = 0; j < npages; j++) {
1661 		if (!(physp[j] & KVMPPC_GOT_PAGE))
1662 			continue;
1663 		pfn = physp[j] >> PAGE_SHIFT;
1664 		page = pfn_to_page(pfn);
1665 		SetPageDirty(page);
1666 		put_page(page);
1667 	}
1668 }
1669 
1670 void kvmppc_core_free_memslot(struct kvm_memory_slot *free,
1671 			      struct kvm_memory_slot *dont)
1672 {
1673 	if (!dont || free->arch.rmap != dont->arch.rmap) {
1674 		vfree(free->arch.rmap);
1675 		free->arch.rmap = NULL;
1676 	}
1677 	if (!dont || free->arch.slot_phys != dont->arch.slot_phys) {
1678 		unpin_slot(free);
1679 		vfree(free->arch.slot_phys);
1680 		free->arch.slot_phys = NULL;
1681 	}
1682 }
1683 
1684 int kvmppc_core_create_memslot(struct kvm_memory_slot *slot,
1685 			       unsigned long npages)
1686 {
1687 	slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
1688 	if (!slot->arch.rmap)
1689 		return -ENOMEM;
1690 	slot->arch.slot_phys = NULL;
1691 
1692 	return 0;
1693 }
1694 
1695 int kvmppc_core_prepare_memory_region(struct kvm *kvm,
1696 				      struct kvm_memory_slot *memslot,
1697 				      struct kvm_userspace_memory_region *mem)
1698 {
1699 	unsigned long *phys;
1700 
1701 	/* Allocate a slot_phys array if needed */
1702 	phys = memslot->arch.slot_phys;
1703 	if (!kvm->arch.using_mmu_notifiers && !phys && memslot->npages) {
1704 		phys = vzalloc(memslot->npages * sizeof(unsigned long));
1705 		if (!phys)
1706 			return -ENOMEM;
1707 		memslot->arch.slot_phys = phys;
1708 	}
1709 
1710 	return 0;
1711 }
1712 
1713 void kvmppc_core_commit_memory_region(struct kvm *kvm,
1714 				      struct kvm_userspace_memory_region *mem,
1715 				      const struct kvm_memory_slot *old)
1716 {
1717 	unsigned long npages = mem->memory_size >> PAGE_SHIFT;
1718 	struct kvm_memory_slot *memslot;
1719 
1720 	if (npages && old->npages) {
1721 		/*
1722 		 * If modifying a memslot, reset all the rmap dirty bits.
1723 		 * If this is a new memslot, we don't need to do anything
1724 		 * since the rmap array starts out as all zeroes,
1725 		 * i.e. no pages are dirty.
1726 		 */
1727 		memslot = id_to_memslot(kvm->memslots, mem->slot);
1728 		kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
1729 	}
1730 }
1731 
1732 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
1733 {
1734 	int err = 0;
1735 	struct kvm *kvm = vcpu->kvm;
1736 	struct kvm_rma_info *ri = NULL;
1737 	unsigned long hva;
1738 	struct kvm_memory_slot *memslot;
1739 	struct vm_area_struct *vma;
1740 	unsigned long lpcr, senc;
1741 	unsigned long psize, porder;
1742 	unsigned long rma_size;
1743 	unsigned long rmls;
1744 	unsigned long *physp;
1745 	unsigned long i, npages;
1746 	int srcu_idx;
1747 
1748 	mutex_lock(&kvm->lock);
1749 	if (kvm->arch.rma_setup_done)
1750 		goto out;	/* another vcpu beat us to it */
1751 
1752 	/* Allocate hashed page table (if not done already) and reset it */
1753 	if (!kvm->arch.hpt_virt) {
1754 		err = kvmppc_alloc_hpt(kvm, NULL);
1755 		if (err) {
1756 			pr_err("KVM: Couldn't alloc HPT\n");
1757 			goto out;
1758 		}
1759 	}
1760 
1761 	/* Look up the memslot for guest physical address 0 */
1762 	srcu_idx = srcu_read_lock(&kvm->srcu);
1763 	memslot = gfn_to_memslot(kvm, 0);
1764 
1765 	/* We must have some memory at 0 by now */
1766 	err = -EINVAL;
1767 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
1768 		goto out_srcu;
1769 
1770 	/* Look up the VMA for the start of this memory slot */
1771 	hva = memslot->userspace_addr;
1772 	down_read(&current->mm->mmap_sem);
1773 	vma = find_vma(current->mm, hva);
1774 	if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
1775 		goto up_out;
1776 
1777 	psize = vma_kernel_pagesize(vma);
1778 	porder = __ilog2(psize);
1779 
1780 	/* Is this one of our preallocated RMAs? */
1781 	if (vma->vm_file && vma->vm_file->f_op == &kvm_rma_fops &&
1782 	    hva == vma->vm_start)
1783 		ri = vma->vm_file->private_data;
1784 
1785 	up_read(&current->mm->mmap_sem);
1786 
1787 	if (!ri) {
1788 		/* On POWER7, use VRMA; on PPC970, give up */
1789 		err = -EPERM;
1790 		if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1791 			pr_err("KVM: CPU requires an RMO\n");
1792 			goto out_srcu;
1793 		}
1794 
1795 		/* We can handle 4k, 64k or 16M pages in the VRMA */
1796 		err = -EINVAL;
1797 		if (!(psize == 0x1000 || psize == 0x10000 ||
1798 		      psize == 0x1000000))
1799 			goto out_srcu;
1800 
1801 		/* Update VRMASD field in the LPCR */
1802 		senc = slb_pgsize_encoding(psize);
1803 		kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
1804 			(VRMA_VSID << SLB_VSID_SHIFT_1T);
1805 		lpcr = kvm->arch.lpcr & ~LPCR_VRMASD;
1806 		lpcr |= senc << (LPCR_VRMASD_SH - 4);
1807 		kvm->arch.lpcr = lpcr;
1808 
1809 		/* Create HPTEs in the hash page table for the VRMA */
1810 		kvmppc_map_vrma(vcpu, memslot, porder);
1811 
1812 	} else {
1813 		/* Set up to use an RMO region */
1814 		rma_size = kvm_rma_pages;
1815 		if (rma_size > memslot->npages)
1816 			rma_size = memslot->npages;
1817 		rma_size <<= PAGE_SHIFT;
1818 		rmls = lpcr_rmls(rma_size);
1819 		err = -EINVAL;
1820 		if ((long)rmls < 0) {
1821 			pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size);
1822 			goto out_srcu;
1823 		}
1824 		atomic_inc(&ri->use_count);
1825 		kvm->arch.rma = ri;
1826 
1827 		/* Update LPCR and RMOR */
1828 		lpcr = kvm->arch.lpcr;
1829 		if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1830 			/* PPC970; insert RMLS value (split field) in HID4 */
1831 			lpcr &= ~((1ul << HID4_RMLS0_SH) |
1832 				  (3ul << HID4_RMLS2_SH));
1833 			lpcr |= ((rmls >> 2) << HID4_RMLS0_SH) |
1834 				((rmls & 3) << HID4_RMLS2_SH);
1835 			/* RMOR is also in HID4 */
1836 			lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff)
1837 				<< HID4_RMOR_SH;
1838 		} else {
1839 			/* POWER7 */
1840 			lpcr &= ~(LPCR_VPM0 | LPCR_VRMA_L);
1841 			lpcr |= rmls << LPCR_RMLS_SH;
1842 			kvm->arch.rmor = ri->base_pfn << PAGE_SHIFT;
1843 		}
1844 		kvm->arch.lpcr = lpcr;
1845 		pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n",
1846 			ri->base_pfn << PAGE_SHIFT, rma_size, lpcr);
1847 
1848 		/* Initialize phys addrs of pages in RMO */
1849 		npages = kvm_rma_pages;
1850 		porder = __ilog2(npages);
1851 		physp = memslot->arch.slot_phys;
1852 		if (physp) {
1853 			if (npages > memslot->npages)
1854 				npages = memslot->npages;
1855 			spin_lock(&kvm->arch.slot_phys_lock);
1856 			for (i = 0; i < npages; ++i)
1857 				physp[i] = ((ri->base_pfn + i) << PAGE_SHIFT) +
1858 					porder;
1859 			spin_unlock(&kvm->arch.slot_phys_lock);
1860 		}
1861 	}
1862 
1863 	/* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */
1864 	smp_wmb();
1865 	kvm->arch.rma_setup_done = 1;
1866 	err = 0;
1867  out_srcu:
1868 	srcu_read_unlock(&kvm->srcu, srcu_idx);
1869  out:
1870 	mutex_unlock(&kvm->lock);
1871 	return err;
1872 
1873  up_out:
1874 	up_read(&current->mm->mmap_sem);
1875 	goto out_srcu;
1876 }
1877 
1878 int kvmppc_core_init_vm(struct kvm *kvm)
1879 {
1880 	unsigned long lpcr, lpid;
1881 
1882 	/* Allocate the guest's logical partition ID */
1883 
1884 	lpid = kvmppc_alloc_lpid();
1885 	if ((long)lpid < 0)
1886 		return -ENOMEM;
1887 	kvm->arch.lpid = lpid;
1888 
1889 	/*
1890 	 * Since we don't flush the TLB when tearing down a VM,
1891 	 * and this lpid might have previously been used,
1892 	 * make sure we flush on each core before running the new VM.
1893 	 */
1894 	cpumask_setall(&kvm->arch.need_tlb_flush);
1895 
1896 	INIT_LIST_HEAD(&kvm->arch.spapr_tce_tables);
1897 	INIT_LIST_HEAD(&kvm->arch.rtas_tokens);
1898 
1899 	kvm->arch.rma = NULL;
1900 
1901 	kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
1902 
1903 	if (cpu_has_feature(CPU_FTR_ARCH_201)) {
1904 		/* PPC970; HID4 is effectively the LPCR */
1905 		kvm->arch.host_lpid = 0;
1906 		kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4);
1907 		lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH));
1908 		lpcr |= ((lpid >> 4) << HID4_LPID1_SH) |
1909 			((lpid & 0xf) << HID4_LPID5_SH);
1910 	} else {
1911 		/* POWER7; init LPCR for virtual RMA mode */
1912 		kvm->arch.host_lpid = mfspr(SPRN_LPID);
1913 		kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
1914 		lpcr &= LPCR_PECE | LPCR_LPES;
1915 		lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
1916 			LPCR_VPM0 | LPCR_VPM1;
1917 		kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
1918 			(VRMA_VSID << SLB_VSID_SHIFT_1T);
1919 	}
1920 	kvm->arch.lpcr = lpcr;
1921 
1922 	kvm->arch.using_mmu_notifiers = !!cpu_has_feature(CPU_FTR_ARCH_206);
1923 	spin_lock_init(&kvm->arch.slot_phys_lock);
1924 
1925 	/*
1926 	 * Don't allow secondary CPU threads to come online
1927 	 * while any KVM VMs exist.
1928 	 */
1929 	inhibit_secondary_onlining();
1930 
1931 	return 0;
1932 }
1933 
1934 void kvmppc_core_destroy_vm(struct kvm *kvm)
1935 {
1936 	uninhibit_secondary_onlining();
1937 
1938 	if (kvm->arch.rma) {
1939 		kvm_release_rma(kvm->arch.rma);
1940 		kvm->arch.rma = NULL;
1941 	}
1942 
1943 	kvmppc_rtas_tokens_free(kvm);
1944 
1945 	kvmppc_free_hpt(kvm);
1946 	WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
1947 }
1948 
1949 /* These are stubs for now */
1950 void kvmppc_mmu_pte_pflush(struct kvm_vcpu *vcpu, ulong pa_start, ulong pa_end)
1951 {
1952 }
1953 
1954 /* We don't need to emulate any privileged instructions or dcbz */
1955 int kvmppc_core_emulate_op(struct kvm_run *run, struct kvm_vcpu *vcpu,
1956                            unsigned int inst, int *advance)
1957 {
1958 	return EMULATE_FAIL;
1959 }
1960 
1961 int kvmppc_core_emulate_mtspr(struct kvm_vcpu *vcpu, int sprn, ulong spr_val)
1962 {
1963 	return EMULATE_FAIL;
1964 }
1965 
1966 int kvmppc_core_emulate_mfspr(struct kvm_vcpu *vcpu, int sprn, ulong *spr_val)
1967 {
1968 	return EMULATE_FAIL;
1969 }
1970 
1971 static int kvmppc_book3s_hv_init(void)
1972 {
1973 	int r;
1974 
1975 	r = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1976 
1977 	if (r)
1978 		return r;
1979 
1980 	r = kvmppc_mmu_hv_init();
1981 
1982 	return r;
1983 }
1984 
1985 static void kvmppc_book3s_hv_exit(void)
1986 {
1987 	kvm_exit();
1988 }
1989 
1990 module_init(kvmppc_book3s_hv_init);
1991 module_exit(kvmppc_book3s_hv_exit);
1992