xref: /openbmc/linux/arch/powerpc/kvm/book3s_hv.c (revision 275876e2)
1 /*
2  * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
3  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
4  *
5  * Authors:
6  *    Paul Mackerras <paulus@au1.ibm.com>
7  *    Alexander Graf <agraf@suse.de>
8  *    Kevin Wolf <mail@kevin-wolf.de>
9  *
10  * Description: KVM functions specific to running on Book 3S
11  * processors in hypervisor mode (specifically POWER7 and later).
12  *
13  * This file is derived from arch/powerpc/kvm/book3s.c,
14  * by Alexander Graf <agraf@suse.de>.
15  *
16  * This program is free software; you can redistribute it and/or modify
17  * it under the terms of the GNU General Public License, version 2, as
18  * published by the Free Software Foundation.
19  */
20 
21 #include <linux/kvm_host.h>
22 #include <linux/err.h>
23 #include <linux/slab.h>
24 #include <linux/preempt.h>
25 #include <linux/sched.h>
26 #include <linux/delay.h>
27 #include <linux/export.h>
28 #include <linux/fs.h>
29 #include <linux/anon_inodes.h>
30 #include <linux/cpumask.h>
31 #include <linux/spinlock.h>
32 #include <linux/page-flags.h>
33 #include <linux/srcu.h>
34 #include <linux/miscdevice.h>
35 
36 #include <asm/reg.h>
37 #include <asm/cputable.h>
38 #include <asm/cache.h>
39 #include <asm/cacheflush.h>
40 #include <asm/tlbflush.h>
41 #include <asm/uaccess.h>
42 #include <asm/io.h>
43 #include <asm/kvm_ppc.h>
44 #include <asm/kvm_book3s.h>
45 #include <asm/mmu_context.h>
46 #include <asm/lppaca.h>
47 #include <asm/processor.h>
48 #include <asm/cputhreads.h>
49 #include <asm/page.h>
50 #include <asm/hvcall.h>
51 #include <asm/switch_to.h>
52 #include <asm/smp.h>
53 #include <linux/gfp.h>
54 #include <linux/vmalloc.h>
55 #include <linux/highmem.h>
56 #include <linux/hugetlb.h>
57 #include <linux/module.h>
58 
59 #include "book3s.h"
60 
61 /* #define EXIT_DEBUG */
62 /* #define EXIT_DEBUG_SIMPLE */
63 /* #define EXIT_DEBUG_INT */
64 
65 /* Used to indicate that a guest page fault needs to be handled */
66 #define RESUME_PAGE_FAULT	(RESUME_GUEST | RESUME_FLAG_ARCH1)
67 
68 /* Used as a "null" value for timebase values */
69 #define TB_NIL	(~(u64)0)
70 
71 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
72 
73 #if defined(CONFIG_PPC_64K_PAGES)
74 #define MPP_BUFFER_ORDER	0
75 #elif defined(CONFIG_PPC_4K_PAGES)
76 #define MPP_BUFFER_ORDER	3
77 #endif
78 
79 
80 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
81 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
82 
83 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
84 {
85 	int me;
86 	int cpu = vcpu->cpu;
87 	wait_queue_head_t *wqp;
88 
89 	wqp = kvm_arch_vcpu_wq(vcpu);
90 	if (waitqueue_active(wqp)) {
91 		wake_up_interruptible(wqp);
92 		++vcpu->stat.halt_wakeup;
93 	}
94 
95 	me = get_cpu();
96 
97 	/* CPU points to the first thread of the core */
98 	if (cpu != me && cpu >= 0 && cpu < nr_cpu_ids) {
99 #ifdef CONFIG_PPC_ICP_NATIVE
100 		int real_cpu = cpu + vcpu->arch.ptid;
101 		if (paca[real_cpu].kvm_hstate.xics_phys)
102 			xics_wake_cpu(real_cpu);
103 		else
104 #endif
105 		if (cpu_online(cpu))
106 			smp_send_reschedule(cpu);
107 	}
108 	put_cpu();
109 }
110 
111 /*
112  * We use the vcpu_load/put functions to measure stolen time.
113  * Stolen time is counted as time when either the vcpu is able to
114  * run as part of a virtual core, but the task running the vcore
115  * is preempted or sleeping, or when the vcpu needs something done
116  * in the kernel by the task running the vcpu, but that task is
117  * preempted or sleeping.  Those two things have to be counted
118  * separately, since one of the vcpu tasks will take on the job
119  * of running the core, and the other vcpu tasks in the vcore will
120  * sleep waiting for it to do that, but that sleep shouldn't count
121  * as stolen time.
122  *
123  * Hence we accumulate stolen time when the vcpu can run as part of
124  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
125  * needs its task to do other things in the kernel (for example,
126  * service a page fault) in busy_stolen.  We don't accumulate
127  * stolen time for a vcore when it is inactive, or for a vcpu
128  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
129  * a misnomer; it means that the vcpu task is not executing in
130  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
131  * the kernel.  We don't have any way of dividing up that time
132  * between time that the vcpu is genuinely stopped, time that
133  * the task is actively working on behalf of the vcpu, and time
134  * that the task is preempted, so we don't count any of it as
135  * stolen.
136  *
137  * Updates to busy_stolen are protected by arch.tbacct_lock;
138  * updates to vc->stolen_tb are protected by the arch.tbacct_lock
139  * of the vcpu that has taken responsibility for running the vcore
140  * (i.e. vc->runner).  The stolen times are measured in units of
141  * timebase ticks.  (Note that the != TB_NIL checks below are
142  * purely defensive; they should never fail.)
143  */
144 
145 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
146 {
147 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
148 	unsigned long flags;
149 
150 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
151 	if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE &&
152 	    vc->preempt_tb != TB_NIL) {
153 		vc->stolen_tb += mftb() - vc->preempt_tb;
154 		vc->preempt_tb = TB_NIL;
155 	}
156 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
157 	    vcpu->arch.busy_preempt != TB_NIL) {
158 		vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
159 		vcpu->arch.busy_preempt = TB_NIL;
160 	}
161 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
162 }
163 
164 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
165 {
166 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
167 	unsigned long flags;
168 
169 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
170 	if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE)
171 		vc->preempt_tb = mftb();
172 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
173 		vcpu->arch.busy_preempt = mftb();
174 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
175 }
176 
177 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
178 {
179 	vcpu->arch.shregs.msr = msr;
180 	kvmppc_end_cede(vcpu);
181 }
182 
183 void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
184 {
185 	vcpu->arch.pvr = pvr;
186 }
187 
188 int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
189 {
190 	unsigned long pcr = 0;
191 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
192 
193 	if (arch_compat) {
194 		if (!cpu_has_feature(CPU_FTR_ARCH_206))
195 			return -EINVAL;	/* 970 has no compat mode support */
196 
197 		switch (arch_compat) {
198 		case PVR_ARCH_205:
199 			/*
200 			 * If an arch bit is set in PCR, all the defined
201 			 * higher-order arch bits also have to be set.
202 			 */
203 			pcr = PCR_ARCH_206 | PCR_ARCH_205;
204 			break;
205 		case PVR_ARCH_206:
206 		case PVR_ARCH_206p:
207 			pcr = PCR_ARCH_206;
208 			break;
209 		case PVR_ARCH_207:
210 			break;
211 		default:
212 			return -EINVAL;
213 		}
214 
215 		if (!cpu_has_feature(CPU_FTR_ARCH_207S)) {
216 			/* POWER7 can't emulate POWER8 */
217 			if (!(pcr & PCR_ARCH_206))
218 				return -EINVAL;
219 			pcr &= ~PCR_ARCH_206;
220 		}
221 	}
222 
223 	spin_lock(&vc->lock);
224 	vc->arch_compat = arch_compat;
225 	vc->pcr = pcr;
226 	spin_unlock(&vc->lock);
227 
228 	return 0;
229 }
230 
231 void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
232 {
233 	int r;
234 
235 	pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
236 	pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
237 	       vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
238 	for (r = 0; r < 16; ++r)
239 		pr_err("r%2d = %.16lx  r%d = %.16lx\n",
240 		       r, kvmppc_get_gpr(vcpu, r),
241 		       r+16, kvmppc_get_gpr(vcpu, r+16));
242 	pr_err("ctr = %.16lx  lr  = %.16lx\n",
243 	       vcpu->arch.ctr, vcpu->arch.lr);
244 	pr_err("srr0 = %.16llx srr1 = %.16llx\n",
245 	       vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
246 	pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
247 	       vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
248 	pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
249 	       vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
250 	pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
251 	       vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
252 	pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
253 	pr_err("fault dar = %.16lx dsisr = %.8x\n",
254 	       vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
255 	pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
256 	for (r = 0; r < vcpu->arch.slb_max; ++r)
257 		pr_err("  ESID = %.16llx VSID = %.16llx\n",
258 		       vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
259 	pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
260 	       vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
261 	       vcpu->arch.last_inst);
262 }
263 
264 struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
265 {
266 	int r;
267 	struct kvm_vcpu *v, *ret = NULL;
268 
269 	mutex_lock(&kvm->lock);
270 	kvm_for_each_vcpu(r, v, kvm) {
271 		if (v->vcpu_id == id) {
272 			ret = v;
273 			break;
274 		}
275 	}
276 	mutex_unlock(&kvm->lock);
277 	return ret;
278 }
279 
280 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
281 {
282 	vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
283 	vpa->yield_count = cpu_to_be32(1);
284 }
285 
286 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
287 		   unsigned long addr, unsigned long len)
288 {
289 	/* check address is cacheline aligned */
290 	if (addr & (L1_CACHE_BYTES - 1))
291 		return -EINVAL;
292 	spin_lock(&vcpu->arch.vpa_update_lock);
293 	if (v->next_gpa != addr || v->len != len) {
294 		v->next_gpa = addr;
295 		v->len = addr ? len : 0;
296 		v->update_pending = 1;
297 	}
298 	spin_unlock(&vcpu->arch.vpa_update_lock);
299 	return 0;
300 }
301 
302 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
303 struct reg_vpa {
304 	u32 dummy;
305 	union {
306 		__be16 hword;
307 		__be32 word;
308 	} length;
309 };
310 
311 static int vpa_is_registered(struct kvmppc_vpa *vpap)
312 {
313 	if (vpap->update_pending)
314 		return vpap->next_gpa != 0;
315 	return vpap->pinned_addr != NULL;
316 }
317 
318 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
319 				       unsigned long flags,
320 				       unsigned long vcpuid, unsigned long vpa)
321 {
322 	struct kvm *kvm = vcpu->kvm;
323 	unsigned long len, nb;
324 	void *va;
325 	struct kvm_vcpu *tvcpu;
326 	int err;
327 	int subfunc;
328 	struct kvmppc_vpa *vpap;
329 
330 	tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
331 	if (!tvcpu)
332 		return H_PARAMETER;
333 
334 	subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
335 	if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
336 	    subfunc == H_VPA_REG_SLB) {
337 		/* Registering new area - address must be cache-line aligned */
338 		if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
339 			return H_PARAMETER;
340 
341 		/* convert logical addr to kernel addr and read length */
342 		va = kvmppc_pin_guest_page(kvm, vpa, &nb);
343 		if (va == NULL)
344 			return H_PARAMETER;
345 		if (subfunc == H_VPA_REG_VPA)
346 			len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
347 		else
348 			len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
349 		kvmppc_unpin_guest_page(kvm, va, vpa, false);
350 
351 		/* Check length */
352 		if (len > nb || len < sizeof(struct reg_vpa))
353 			return H_PARAMETER;
354 	} else {
355 		vpa = 0;
356 		len = 0;
357 	}
358 
359 	err = H_PARAMETER;
360 	vpap = NULL;
361 	spin_lock(&tvcpu->arch.vpa_update_lock);
362 
363 	switch (subfunc) {
364 	case H_VPA_REG_VPA:		/* register VPA */
365 		if (len < sizeof(struct lppaca))
366 			break;
367 		vpap = &tvcpu->arch.vpa;
368 		err = 0;
369 		break;
370 
371 	case H_VPA_REG_DTL:		/* register DTL */
372 		if (len < sizeof(struct dtl_entry))
373 			break;
374 		len -= len % sizeof(struct dtl_entry);
375 
376 		/* Check that they have previously registered a VPA */
377 		err = H_RESOURCE;
378 		if (!vpa_is_registered(&tvcpu->arch.vpa))
379 			break;
380 
381 		vpap = &tvcpu->arch.dtl;
382 		err = 0;
383 		break;
384 
385 	case H_VPA_REG_SLB:		/* register SLB shadow buffer */
386 		/* Check that they have previously registered a VPA */
387 		err = H_RESOURCE;
388 		if (!vpa_is_registered(&tvcpu->arch.vpa))
389 			break;
390 
391 		vpap = &tvcpu->arch.slb_shadow;
392 		err = 0;
393 		break;
394 
395 	case H_VPA_DEREG_VPA:		/* deregister VPA */
396 		/* Check they don't still have a DTL or SLB buf registered */
397 		err = H_RESOURCE;
398 		if (vpa_is_registered(&tvcpu->arch.dtl) ||
399 		    vpa_is_registered(&tvcpu->arch.slb_shadow))
400 			break;
401 
402 		vpap = &tvcpu->arch.vpa;
403 		err = 0;
404 		break;
405 
406 	case H_VPA_DEREG_DTL:		/* deregister DTL */
407 		vpap = &tvcpu->arch.dtl;
408 		err = 0;
409 		break;
410 
411 	case H_VPA_DEREG_SLB:		/* deregister SLB shadow buffer */
412 		vpap = &tvcpu->arch.slb_shadow;
413 		err = 0;
414 		break;
415 	}
416 
417 	if (vpap) {
418 		vpap->next_gpa = vpa;
419 		vpap->len = len;
420 		vpap->update_pending = 1;
421 	}
422 
423 	spin_unlock(&tvcpu->arch.vpa_update_lock);
424 
425 	return err;
426 }
427 
428 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
429 {
430 	struct kvm *kvm = vcpu->kvm;
431 	void *va;
432 	unsigned long nb;
433 	unsigned long gpa;
434 
435 	/*
436 	 * We need to pin the page pointed to by vpap->next_gpa,
437 	 * but we can't call kvmppc_pin_guest_page under the lock
438 	 * as it does get_user_pages() and down_read().  So we
439 	 * have to drop the lock, pin the page, then get the lock
440 	 * again and check that a new area didn't get registered
441 	 * in the meantime.
442 	 */
443 	for (;;) {
444 		gpa = vpap->next_gpa;
445 		spin_unlock(&vcpu->arch.vpa_update_lock);
446 		va = NULL;
447 		nb = 0;
448 		if (gpa)
449 			va = kvmppc_pin_guest_page(kvm, gpa, &nb);
450 		spin_lock(&vcpu->arch.vpa_update_lock);
451 		if (gpa == vpap->next_gpa)
452 			break;
453 		/* sigh... unpin that one and try again */
454 		if (va)
455 			kvmppc_unpin_guest_page(kvm, va, gpa, false);
456 	}
457 
458 	vpap->update_pending = 0;
459 	if (va && nb < vpap->len) {
460 		/*
461 		 * If it's now too short, it must be that userspace
462 		 * has changed the mappings underlying guest memory,
463 		 * so unregister the region.
464 		 */
465 		kvmppc_unpin_guest_page(kvm, va, gpa, false);
466 		va = NULL;
467 	}
468 	if (vpap->pinned_addr)
469 		kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
470 					vpap->dirty);
471 	vpap->gpa = gpa;
472 	vpap->pinned_addr = va;
473 	vpap->dirty = false;
474 	if (va)
475 		vpap->pinned_end = va + vpap->len;
476 }
477 
478 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
479 {
480 	if (!(vcpu->arch.vpa.update_pending ||
481 	      vcpu->arch.slb_shadow.update_pending ||
482 	      vcpu->arch.dtl.update_pending))
483 		return;
484 
485 	spin_lock(&vcpu->arch.vpa_update_lock);
486 	if (vcpu->arch.vpa.update_pending) {
487 		kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
488 		if (vcpu->arch.vpa.pinned_addr)
489 			init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
490 	}
491 	if (vcpu->arch.dtl.update_pending) {
492 		kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
493 		vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
494 		vcpu->arch.dtl_index = 0;
495 	}
496 	if (vcpu->arch.slb_shadow.update_pending)
497 		kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
498 	spin_unlock(&vcpu->arch.vpa_update_lock);
499 }
500 
501 /*
502  * Return the accumulated stolen time for the vcore up until `now'.
503  * The caller should hold the vcore lock.
504  */
505 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
506 {
507 	u64 p;
508 
509 	/*
510 	 * If we are the task running the vcore, then since we hold
511 	 * the vcore lock, we can't be preempted, so stolen_tb/preempt_tb
512 	 * can't be updated, so we don't need the tbacct_lock.
513 	 * If the vcore is inactive, it can't become active (since we
514 	 * hold the vcore lock), so the vcpu load/put functions won't
515 	 * update stolen_tb/preempt_tb, and we don't need tbacct_lock.
516 	 */
517 	if (vc->vcore_state != VCORE_INACTIVE &&
518 	    vc->runner->arch.run_task != current) {
519 		spin_lock_irq(&vc->runner->arch.tbacct_lock);
520 		p = vc->stolen_tb;
521 		if (vc->preempt_tb != TB_NIL)
522 			p += now - vc->preempt_tb;
523 		spin_unlock_irq(&vc->runner->arch.tbacct_lock);
524 	} else {
525 		p = vc->stolen_tb;
526 	}
527 	return p;
528 }
529 
530 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
531 				    struct kvmppc_vcore *vc)
532 {
533 	struct dtl_entry *dt;
534 	struct lppaca *vpa;
535 	unsigned long stolen;
536 	unsigned long core_stolen;
537 	u64 now;
538 
539 	dt = vcpu->arch.dtl_ptr;
540 	vpa = vcpu->arch.vpa.pinned_addr;
541 	now = mftb();
542 	core_stolen = vcore_stolen_time(vc, now);
543 	stolen = core_stolen - vcpu->arch.stolen_logged;
544 	vcpu->arch.stolen_logged = core_stolen;
545 	spin_lock_irq(&vcpu->arch.tbacct_lock);
546 	stolen += vcpu->arch.busy_stolen;
547 	vcpu->arch.busy_stolen = 0;
548 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
549 	if (!dt || !vpa)
550 		return;
551 	memset(dt, 0, sizeof(struct dtl_entry));
552 	dt->dispatch_reason = 7;
553 	dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
554 	dt->timebase = cpu_to_be64(now + vc->tb_offset);
555 	dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
556 	dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
557 	dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
558 	++dt;
559 	if (dt == vcpu->arch.dtl.pinned_end)
560 		dt = vcpu->arch.dtl.pinned_addr;
561 	vcpu->arch.dtl_ptr = dt;
562 	/* order writing *dt vs. writing vpa->dtl_idx */
563 	smp_wmb();
564 	vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
565 	vcpu->arch.dtl.dirty = true;
566 }
567 
568 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
569 {
570 	if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
571 		return true;
572 	if ((!vcpu->arch.vcore->arch_compat) &&
573 	    cpu_has_feature(CPU_FTR_ARCH_207S))
574 		return true;
575 	return false;
576 }
577 
578 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
579 			     unsigned long resource, unsigned long value1,
580 			     unsigned long value2)
581 {
582 	switch (resource) {
583 	case H_SET_MODE_RESOURCE_SET_CIABR:
584 		if (!kvmppc_power8_compatible(vcpu))
585 			return H_P2;
586 		if (value2)
587 			return H_P4;
588 		if (mflags)
589 			return H_UNSUPPORTED_FLAG_START;
590 		/* Guests can't breakpoint the hypervisor */
591 		if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
592 			return H_P3;
593 		vcpu->arch.ciabr  = value1;
594 		return H_SUCCESS;
595 	case H_SET_MODE_RESOURCE_SET_DAWR:
596 		if (!kvmppc_power8_compatible(vcpu))
597 			return H_P2;
598 		if (mflags)
599 			return H_UNSUPPORTED_FLAG_START;
600 		if (value2 & DABRX_HYP)
601 			return H_P4;
602 		vcpu->arch.dawr  = value1;
603 		vcpu->arch.dawrx = value2;
604 		return H_SUCCESS;
605 	default:
606 		return H_TOO_HARD;
607 	}
608 }
609 
610 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
611 {
612 	unsigned long req = kvmppc_get_gpr(vcpu, 3);
613 	unsigned long target, ret = H_SUCCESS;
614 	struct kvm_vcpu *tvcpu;
615 	int idx, rc;
616 
617 	if (req <= MAX_HCALL_OPCODE &&
618 	    !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
619 		return RESUME_HOST;
620 
621 	switch (req) {
622 	case H_ENTER:
623 		idx = srcu_read_lock(&vcpu->kvm->srcu);
624 		ret = kvmppc_virtmode_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4),
625 					      kvmppc_get_gpr(vcpu, 5),
626 					      kvmppc_get_gpr(vcpu, 6),
627 					      kvmppc_get_gpr(vcpu, 7));
628 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
629 		break;
630 	case H_CEDE:
631 		break;
632 	case H_PROD:
633 		target = kvmppc_get_gpr(vcpu, 4);
634 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
635 		if (!tvcpu) {
636 			ret = H_PARAMETER;
637 			break;
638 		}
639 		tvcpu->arch.prodded = 1;
640 		smp_mb();
641 		if (vcpu->arch.ceded) {
642 			if (waitqueue_active(&vcpu->wq)) {
643 				wake_up_interruptible(&vcpu->wq);
644 				vcpu->stat.halt_wakeup++;
645 			}
646 		}
647 		break;
648 	case H_CONFER:
649 		target = kvmppc_get_gpr(vcpu, 4);
650 		if (target == -1)
651 			break;
652 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
653 		if (!tvcpu) {
654 			ret = H_PARAMETER;
655 			break;
656 		}
657 		kvm_vcpu_yield_to(tvcpu);
658 		break;
659 	case H_REGISTER_VPA:
660 		ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
661 					kvmppc_get_gpr(vcpu, 5),
662 					kvmppc_get_gpr(vcpu, 6));
663 		break;
664 	case H_RTAS:
665 		if (list_empty(&vcpu->kvm->arch.rtas_tokens))
666 			return RESUME_HOST;
667 
668 		idx = srcu_read_lock(&vcpu->kvm->srcu);
669 		rc = kvmppc_rtas_hcall(vcpu);
670 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
671 
672 		if (rc == -ENOENT)
673 			return RESUME_HOST;
674 		else if (rc == 0)
675 			break;
676 
677 		/* Send the error out to userspace via KVM_RUN */
678 		return rc;
679 	case H_SET_MODE:
680 		ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
681 					kvmppc_get_gpr(vcpu, 5),
682 					kvmppc_get_gpr(vcpu, 6),
683 					kvmppc_get_gpr(vcpu, 7));
684 		if (ret == H_TOO_HARD)
685 			return RESUME_HOST;
686 		break;
687 	case H_XIRR:
688 	case H_CPPR:
689 	case H_EOI:
690 	case H_IPI:
691 	case H_IPOLL:
692 	case H_XIRR_X:
693 		if (kvmppc_xics_enabled(vcpu)) {
694 			ret = kvmppc_xics_hcall(vcpu, req);
695 			break;
696 		} /* fallthrough */
697 	default:
698 		return RESUME_HOST;
699 	}
700 	kvmppc_set_gpr(vcpu, 3, ret);
701 	vcpu->arch.hcall_needed = 0;
702 	return RESUME_GUEST;
703 }
704 
705 static int kvmppc_hcall_impl_hv(unsigned long cmd)
706 {
707 	switch (cmd) {
708 	case H_CEDE:
709 	case H_PROD:
710 	case H_CONFER:
711 	case H_REGISTER_VPA:
712 	case H_SET_MODE:
713 #ifdef CONFIG_KVM_XICS
714 	case H_XIRR:
715 	case H_CPPR:
716 	case H_EOI:
717 	case H_IPI:
718 	case H_IPOLL:
719 	case H_XIRR_X:
720 #endif
721 		return 1;
722 	}
723 
724 	/* See if it's in the real-mode table */
725 	return kvmppc_hcall_impl_hv_realmode(cmd);
726 }
727 
728 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
729 				 struct task_struct *tsk)
730 {
731 	int r = RESUME_HOST;
732 
733 	vcpu->stat.sum_exits++;
734 
735 	run->exit_reason = KVM_EXIT_UNKNOWN;
736 	run->ready_for_interrupt_injection = 1;
737 	switch (vcpu->arch.trap) {
738 	/* We're good on these - the host merely wanted to get our attention */
739 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
740 		vcpu->stat.dec_exits++;
741 		r = RESUME_GUEST;
742 		break;
743 	case BOOK3S_INTERRUPT_EXTERNAL:
744 	case BOOK3S_INTERRUPT_H_DOORBELL:
745 		vcpu->stat.ext_intr_exits++;
746 		r = RESUME_GUEST;
747 		break;
748 	case BOOK3S_INTERRUPT_PERFMON:
749 		r = RESUME_GUEST;
750 		break;
751 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
752 		/*
753 		 * Deliver a machine check interrupt to the guest.
754 		 * We have to do this, even if the host has handled the
755 		 * machine check, because machine checks use SRR0/1 and
756 		 * the interrupt might have trashed guest state in them.
757 		 */
758 		kvmppc_book3s_queue_irqprio(vcpu,
759 					    BOOK3S_INTERRUPT_MACHINE_CHECK);
760 		r = RESUME_GUEST;
761 		break;
762 	case BOOK3S_INTERRUPT_PROGRAM:
763 	{
764 		ulong flags;
765 		/*
766 		 * Normally program interrupts are delivered directly
767 		 * to the guest by the hardware, but we can get here
768 		 * as a result of a hypervisor emulation interrupt
769 		 * (e40) getting turned into a 700 by BML RTAS.
770 		 */
771 		flags = vcpu->arch.shregs.msr & 0x1f0000ull;
772 		kvmppc_core_queue_program(vcpu, flags);
773 		r = RESUME_GUEST;
774 		break;
775 	}
776 	case BOOK3S_INTERRUPT_SYSCALL:
777 	{
778 		/* hcall - punt to userspace */
779 		int i;
780 
781 		/* hypercall with MSR_PR has already been handled in rmode,
782 		 * and never reaches here.
783 		 */
784 
785 		run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
786 		for (i = 0; i < 9; ++i)
787 			run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
788 		run->exit_reason = KVM_EXIT_PAPR_HCALL;
789 		vcpu->arch.hcall_needed = 1;
790 		r = RESUME_HOST;
791 		break;
792 	}
793 	/*
794 	 * We get these next two if the guest accesses a page which it thinks
795 	 * it has mapped but which is not actually present, either because
796 	 * it is for an emulated I/O device or because the corresonding
797 	 * host page has been paged out.  Any other HDSI/HISI interrupts
798 	 * have been handled already.
799 	 */
800 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
801 		r = RESUME_PAGE_FAULT;
802 		break;
803 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
804 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
805 		vcpu->arch.fault_dsisr = 0;
806 		r = RESUME_PAGE_FAULT;
807 		break;
808 	/*
809 	 * This occurs if the guest executes an illegal instruction.
810 	 * We just generate a program interrupt to the guest, since
811 	 * we don't emulate any guest instructions at this stage.
812 	 */
813 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
814 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
815 		r = RESUME_GUEST;
816 		break;
817 	/*
818 	 * This occurs if the guest (kernel or userspace), does something that
819 	 * is prohibited by HFSCR.  We just generate a program interrupt to
820 	 * the guest.
821 	 */
822 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
823 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
824 		r = RESUME_GUEST;
825 		break;
826 	default:
827 		kvmppc_dump_regs(vcpu);
828 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
829 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
830 			vcpu->arch.shregs.msr);
831 		run->hw.hardware_exit_reason = vcpu->arch.trap;
832 		r = RESUME_HOST;
833 		break;
834 	}
835 
836 	return r;
837 }
838 
839 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
840 					    struct kvm_sregs *sregs)
841 {
842 	int i;
843 
844 	memset(sregs, 0, sizeof(struct kvm_sregs));
845 	sregs->pvr = vcpu->arch.pvr;
846 	for (i = 0; i < vcpu->arch.slb_max; i++) {
847 		sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
848 		sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
849 	}
850 
851 	return 0;
852 }
853 
854 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
855 					    struct kvm_sregs *sregs)
856 {
857 	int i, j;
858 
859 	kvmppc_set_pvr_hv(vcpu, sregs->pvr);
860 
861 	j = 0;
862 	for (i = 0; i < vcpu->arch.slb_nr; i++) {
863 		if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
864 			vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
865 			vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
866 			++j;
867 		}
868 	}
869 	vcpu->arch.slb_max = j;
870 
871 	return 0;
872 }
873 
874 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
875 		bool preserve_top32)
876 {
877 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
878 	u64 mask;
879 
880 	spin_lock(&vc->lock);
881 	/*
882 	 * If ILE (interrupt little-endian) has changed, update the
883 	 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
884 	 */
885 	if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
886 		struct kvm *kvm = vcpu->kvm;
887 		struct kvm_vcpu *vcpu;
888 		int i;
889 
890 		mutex_lock(&kvm->lock);
891 		kvm_for_each_vcpu(i, vcpu, kvm) {
892 			if (vcpu->arch.vcore != vc)
893 				continue;
894 			if (new_lpcr & LPCR_ILE)
895 				vcpu->arch.intr_msr |= MSR_LE;
896 			else
897 				vcpu->arch.intr_msr &= ~MSR_LE;
898 		}
899 		mutex_unlock(&kvm->lock);
900 	}
901 
902 	/*
903 	 * Userspace can only modify DPFD (default prefetch depth),
904 	 * ILE (interrupt little-endian) and TC (translation control).
905 	 * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
906 	 */
907 	mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
908 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
909 		mask |= LPCR_AIL;
910 
911 	/* Broken 32-bit version of LPCR must not clear top bits */
912 	if (preserve_top32)
913 		mask &= 0xFFFFFFFF;
914 	vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
915 	spin_unlock(&vc->lock);
916 }
917 
918 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
919 				 union kvmppc_one_reg *val)
920 {
921 	int r = 0;
922 	long int i;
923 
924 	switch (id) {
925 	case KVM_REG_PPC_HIOR:
926 		*val = get_reg_val(id, 0);
927 		break;
928 	case KVM_REG_PPC_DABR:
929 		*val = get_reg_val(id, vcpu->arch.dabr);
930 		break;
931 	case KVM_REG_PPC_DABRX:
932 		*val = get_reg_val(id, vcpu->arch.dabrx);
933 		break;
934 	case KVM_REG_PPC_DSCR:
935 		*val = get_reg_val(id, vcpu->arch.dscr);
936 		break;
937 	case KVM_REG_PPC_PURR:
938 		*val = get_reg_val(id, vcpu->arch.purr);
939 		break;
940 	case KVM_REG_PPC_SPURR:
941 		*val = get_reg_val(id, vcpu->arch.spurr);
942 		break;
943 	case KVM_REG_PPC_AMR:
944 		*val = get_reg_val(id, vcpu->arch.amr);
945 		break;
946 	case KVM_REG_PPC_UAMOR:
947 		*val = get_reg_val(id, vcpu->arch.uamor);
948 		break;
949 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
950 		i = id - KVM_REG_PPC_MMCR0;
951 		*val = get_reg_val(id, vcpu->arch.mmcr[i]);
952 		break;
953 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
954 		i = id - KVM_REG_PPC_PMC1;
955 		*val = get_reg_val(id, vcpu->arch.pmc[i]);
956 		break;
957 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
958 		i = id - KVM_REG_PPC_SPMC1;
959 		*val = get_reg_val(id, vcpu->arch.spmc[i]);
960 		break;
961 	case KVM_REG_PPC_SIAR:
962 		*val = get_reg_val(id, vcpu->arch.siar);
963 		break;
964 	case KVM_REG_PPC_SDAR:
965 		*val = get_reg_val(id, vcpu->arch.sdar);
966 		break;
967 	case KVM_REG_PPC_SIER:
968 		*val = get_reg_val(id, vcpu->arch.sier);
969 		break;
970 	case KVM_REG_PPC_IAMR:
971 		*val = get_reg_val(id, vcpu->arch.iamr);
972 		break;
973 	case KVM_REG_PPC_PSPB:
974 		*val = get_reg_val(id, vcpu->arch.pspb);
975 		break;
976 	case KVM_REG_PPC_DPDES:
977 		*val = get_reg_val(id, vcpu->arch.vcore->dpdes);
978 		break;
979 	case KVM_REG_PPC_DAWR:
980 		*val = get_reg_val(id, vcpu->arch.dawr);
981 		break;
982 	case KVM_REG_PPC_DAWRX:
983 		*val = get_reg_val(id, vcpu->arch.dawrx);
984 		break;
985 	case KVM_REG_PPC_CIABR:
986 		*val = get_reg_val(id, vcpu->arch.ciabr);
987 		break;
988 	case KVM_REG_PPC_CSIGR:
989 		*val = get_reg_val(id, vcpu->arch.csigr);
990 		break;
991 	case KVM_REG_PPC_TACR:
992 		*val = get_reg_val(id, vcpu->arch.tacr);
993 		break;
994 	case KVM_REG_PPC_TCSCR:
995 		*val = get_reg_val(id, vcpu->arch.tcscr);
996 		break;
997 	case KVM_REG_PPC_PID:
998 		*val = get_reg_val(id, vcpu->arch.pid);
999 		break;
1000 	case KVM_REG_PPC_ACOP:
1001 		*val = get_reg_val(id, vcpu->arch.acop);
1002 		break;
1003 	case KVM_REG_PPC_WORT:
1004 		*val = get_reg_val(id, vcpu->arch.wort);
1005 		break;
1006 	case KVM_REG_PPC_VPA_ADDR:
1007 		spin_lock(&vcpu->arch.vpa_update_lock);
1008 		*val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1009 		spin_unlock(&vcpu->arch.vpa_update_lock);
1010 		break;
1011 	case KVM_REG_PPC_VPA_SLB:
1012 		spin_lock(&vcpu->arch.vpa_update_lock);
1013 		val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1014 		val->vpaval.length = vcpu->arch.slb_shadow.len;
1015 		spin_unlock(&vcpu->arch.vpa_update_lock);
1016 		break;
1017 	case KVM_REG_PPC_VPA_DTL:
1018 		spin_lock(&vcpu->arch.vpa_update_lock);
1019 		val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1020 		val->vpaval.length = vcpu->arch.dtl.len;
1021 		spin_unlock(&vcpu->arch.vpa_update_lock);
1022 		break;
1023 	case KVM_REG_PPC_TB_OFFSET:
1024 		*val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1025 		break;
1026 	case KVM_REG_PPC_LPCR:
1027 	case KVM_REG_PPC_LPCR_64:
1028 		*val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1029 		break;
1030 	case KVM_REG_PPC_PPR:
1031 		*val = get_reg_val(id, vcpu->arch.ppr);
1032 		break;
1033 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1034 	case KVM_REG_PPC_TFHAR:
1035 		*val = get_reg_val(id, vcpu->arch.tfhar);
1036 		break;
1037 	case KVM_REG_PPC_TFIAR:
1038 		*val = get_reg_val(id, vcpu->arch.tfiar);
1039 		break;
1040 	case KVM_REG_PPC_TEXASR:
1041 		*val = get_reg_val(id, vcpu->arch.texasr);
1042 		break;
1043 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1044 		i = id - KVM_REG_PPC_TM_GPR0;
1045 		*val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1046 		break;
1047 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1048 	{
1049 		int j;
1050 		i = id - KVM_REG_PPC_TM_VSR0;
1051 		if (i < 32)
1052 			for (j = 0; j < TS_FPRWIDTH; j++)
1053 				val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1054 		else {
1055 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1056 				val->vval = vcpu->arch.vr_tm.vr[i-32];
1057 			else
1058 				r = -ENXIO;
1059 		}
1060 		break;
1061 	}
1062 	case KVM_REG_PPC_TM_CR:
1063 		*val = get_reg_val(id, vcpu->arch.cr_tm);
1064 		break;
1065 	case KVM_REG_PPC_TM_LR:
1066 		*val = get_reg_val(id, vcpu->arch.lr_tm);
1067 		break;
1068 	case KVM_REG_PPC_TM_CTR:
1069 		*val = get_reg_val(id, vcpu->arch.ctr_tm);
1070 		break;
1071 	case KVM_REG_PPC_TM_FPSCR:
1072 		*val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1073 		break;
1074 	case KVM_REG_PPC_TM_AMR:
1075 		*val = get_reg_val(id, vcpu->arch.amr_tm);
1076 		break;
1077 	case KVM_REG_PPC_TM_PPR:
1078 		*val = get_reg_val(id, vcpu->arch.ppr_tm);
1079 		break;
1080 	case KVM_REG_PPC_TM_VRSAVE:
1081 		*val = get_reg_val(id, vcpu->arch.vrsave_tm);
1082 		break;
1083 	case KVM_REG_PPC_TM_VSCR:
1084 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1085 			*val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1086 		else
1087 			r = -ENXIO;
1088 		break;
1089 	case KVM_REG_PPC_TM_DSCR:
1090 		*val = get_reg_val(id, vcpu->arch.dscr_tm);
1091 		break;
1092 	case KVM_REG_PPC_TM_TAR:
1093 		*val = get_reg_val(id, vcpu->arch.tar_tm);
1094 		break;
1095 #endif
1096 	case KVM_REG_PPC_ARCH_COMPAT:
1097 		*val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1098 		break;
1099 	default:
1100 		r = -EINVAL;
1101 		break;
1102 	}
1103 
1104 	return r;
1105 }
1106 
1107 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1108 				 union kvmppc_one_reg *val)
1109 {
1110 	int r = 0;
1111 	long int i;
1112 	unsigned long addr, len;
1113 
1114 	switch (id) {
1115 	case KVM_REG_PPC_HIOR:
1116 		/* Only allow this to be set to zero */
1117 		if (set_reg_val(id, *val))
1118 			r = -EINVAL;
1119 		break;
1120 	case KVM_REG_PPC_DABR:
1121 		vcpu->arch.dabr = set_reg_val(id, *val);
1122 		break;
1123 	case KVM_REG_PPC_DABRX:
1124 		vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1125 		break;
1126 	case KVM_REG_PPC_DSCR:
1127 		vcpu->arch.dscr = set_reg_val(id, *val);
1128 		break;
1129 	case KVM_REG_PPC_PURR:
1130 		vcpu->arch.purr = set_reg_val(id, *val);
1131 		break;
1132 	case KVM_REG_PPC_SPURR:
1133 		vcpu->arch.spurr = set_reg_val(id, *val);
1134 		break;
1135 	case KVM_REG_PPC_AMR:
1136 		vcpu->arch.amr = set_reg_val(id, *val);
1137 		break;
1138 	case KVM_REG_PPC_UAMOR:
1139 		vcpu->arch.uamor = set_reg_val(id, *val);
1140 		break;
1141 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1142 		i = id - KVM_REG_PPC_MMCR0;
1143 		vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1144 		break;
1145 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1146 		i = id - KVM_REG_PPC_PMC1;
1147 		vcpu->arch.pmc[i] = set_reg_val(id, *val);
1148 		break;
1149 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1150 		i = id - KVM_REG_PPC_SPMC1;
1151 		vcpu->arch.spmc[i] = set_reg_val(id, *val);
1152 		break;
1153 	case KVM_REG_PPC_SIAR:
1154 		vcpu->arch.siar = set_reg_val(id, *val);
1155 		break;
1156 	case KVM_REG_PPC_SDAR:
1157 		vcpu->arch.sdar = set_reg_val(id, *val);
1158 		break;
1159 	case KVM_REG_PPC_SIER:
1160 		vcpu->arch.sier = set_reg_val(id, *val);
1161 		break;
1162 	case KVM_REG_PPC_IAMR:
1163 		vcpu->arch.iamr = set_reg_val(id, *val);
1164 		break;
1165 	case KVM_REG_PPC_PSPB:
1166 		vcpu->arch.pspb = set_reg_val(id, *val);
1167 		break;
1168 	case KVM_REG_PPC_DPDES:
1169 		vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1170 		break;
1171 	case KVM_REG_PPC_DAWR:
1172 		vcpu->arch.dawr = set_reg_val(id, *val);
1173 		break;
1174 	case KVM_REG_PPC_DAWRX:
1175 		vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1176 		break;
1177 	case KVM_REG_PPC_CIABR:
1178 		vcpu->arch.ciabr = set_reg_val(id, *val);
1179 		/* Don't allow setting breakpoints in hypervisor code */
1180 		if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1181 			vcpu->arch.ciabr &= ~CIABR_PRIV;	/* disable */
1182 		break;
1183 	case KVM_REG_PPC_CSIGR:
1184 		vcpu->arch.csigr = set_reg_val(id, *val);
1185 		break;
1186 	case KVM_REG_PPC_TACR:
1187 		vcpu->arch.tacr = set_reg_val(id, *val);
1188 		break;
1189 	case KVM_REG_PPC_TCSCR:
1190 		vcpu->arch.tcscr = set_reg_val(id, *val);
1191 		break;
1192 	case KVM_REG_PPC_PID:
1193 		vcpu->arch.pid = set_reg_val(id, *val);
1194 		break;
1195 	case KVM_REG_PPC_ACOP:
1196 		vcpu->arch.acop = set_reg_val(id, *val);
1197 		break;
1198 	case KVM_REG_PPC_WORT:
1199 		vcpu->arch.wort = set_reg_val(id, *val);
1200 		break;
1201 	case KVM_REG_PPC_VPA_ADDR:
1202 		addr = set_reg_val(id, *val);
1203 		r = -EINVAL;
1204 		if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1205 			      vcpu->arch.dtl.next_gpa))
1206 			break;
1207 		r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1208 		break;
1209 	case KVM_REG_PPC_VPA_SLB:
1210 		addr = val->vpaval.addr;
1211 		len = val->vpaval.length;
1212 		r = -EINVAL;
1213 		if (addr && !vcpu->arch.vpa.next_gpa)
1214 			break;
1215 		r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1216 		break;
1217 	case KVM_REG_PPC_VPA_DTL:
1218 		addr = val->vpaval.addr;
1219 		len = val->vpaval.length;
1220 		r = -EINVAL;
1221 		if (addr && (len < sizeof(struct dtl_entry) ||
1222 			     !vcpu->arch.vpa.next_gpa))
1223 			break;
1224 		len -= len % sizeof(struct dtl_entry);
1225 		r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1226 		break;
1227 	case KVM_REG_PPC_TB_OFFSET:
1228 		/* round up to multiple of 2^24 */
1229 		vcpu->arch.vcore->tb_offset =
1230 			ALIGN(set_reg_val(id, *val), 1UL << 24);
1231 		break;
1232 	case KVM_REG_PPC_LPCR:
1233 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1234 		break;
1235 	case KVM_REG_PPC_LPCR_64:
1236 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1237 		break;
1238 	case KVM_REG_PPC_PPR:
1239 		vcpu->arch.ppr = set_reg_val(id, *val);
1240 		break;
1241 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1242 	case KVM_REG_PPC_TFHAR:
1243 		vcpu->arch.tfhar = set_reg_val(id, *val);
1244 		break;
1245 	case KVM_REG_PPC_TFIAR:
1246 		vcpu->arch.tfiar = set_reg_val(id, *val);
1247 		break;
1248 	case KVM_REG_PPC_TEXASR:
1249 		vcpu->arch.texasr = set_reg_val(id, *val);
1250 		break;
1251 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1252 		i = id - KVM_REG_PPC_TM_GPR0;
1253 		vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1254 		break;
1255 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1256 	{
1257 		int j;
1258 		i = id - KVM_REG_PPC_TM_VSR0;
1259 		if (i < 32)
1260 			for (j = 0; j < TS_FPRWIDTH; j++)
1261 				vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1262 		else
1263 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1264 				vcpu->arch.vr_tm.vr[i-32] = val->vval;
1265 			else
1266 				r = -ENXIO;
1267 		break;
1268 	}
1269 	case KVM_REG_PPC_TM_CR:
1270 		vcpu->arch.cr_tm = set_reg_val(id, *val);
1271 		break;
1272 	case KVM_REG_PPC_TM_LR:
1273 		vcpu->arch.lr_tm = set_reg_val(id, *val);
1274 		break;
1275 	case KVM_REG_PPC_TM_CTR:
1276 		vcpu->arch.ctr_tm = set_reg_val(id, *val);
1277 		break;
1278 	case KVM_REG_PPC_TM_FPSCR:
1279 		vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1280 		break;
1281 	case KVM_REG_PPC_TM_AMR:
1282 		vcpu->arch.amr_tm = set_reg_val(id, *val);
1283 		break;
1284 	case KVM_REG_PPC_TM_PPR:
1285 		vcpu->arch.ppr_tm = set_reg_val(id, *val);
1286 		break;
1287 	case KVM_REG_PPC_TM_VRSAVE:
1288 		vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1289 		break;
1290 	case KVM_REG_PPC_TM_VSCR:
1291 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1292 			vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1293 		else
1294 			r = - ENXIO;
1295 		break;
1296 	case KVM_REG_PPC_TM_DSCR:
1297 		vcpu->arch.dscr_tm = set_reg_val(id, *val);
1298 		break;
1299 	case KVM_REG_PPC_TM_TAR:
1300 		vcpu->arch.tar_tm = set_reg_val(id, *val);
1301 		break;
1302 #endif
1303 	case KVM_REG_PPC_ARCH_COMPAT:
1304 		r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1305 		break;
1306 	default:
1307 		r = -EINVAL;
1308 		break;
1309 	}
1310 
1311 	return r;
1312 }
1313 
1314 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1315 {
1316 	struct kvmppc_vcore *vcore;
1317 
1318 	vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1319 
1320 	if (vcore == NULL)
1321 		return NULL;
1322 
1323 	INIT_LIST_HEAD(&vcore->runnable_threads);
1324 	spin_lock_init(&vcore->lock);
1325 	init_waitqueue_head(&vcore->wq);
1326 	vcore->preempt_tb = TB_NIL;
1327 	vcore->lpcr = kvm->arch.lpcr;
1328 	vcore->first_vcpuid = core * threads_per_subcore;
1329 	vcore->kvm = kvm;
1330 
1331 	vcore->mpp_buffer_is_valid = false;
1332 
1333 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
1334 		vcore->mpp_buffer = (void *)__get_free_pages(
1335 			GFP_KERNEL|__GFP_ZERO,
1336 			MPP_BUFFER_ORDER);
1337 
1338 	return vcore;
1339 }
1340 
1341 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1342 						   unsigned int id)
1343 {
1344 	struct kvm_vcpu *vcpu;
1345 	int err = -EINVAL;
1346 	int core;
1347 	struct kvmppc_vcore *vcore;
1348 
1349 	core = id / threads_per_subcore;
1350 	if (core >= KVM_MAX_VCORES)
1351 		goto out;
1352 
1353 	err = -ENOMEM;
1354 	vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1355 	if (!vcpu)
1356 		goto out;
1357 
1358 	err = kvm_vcpu_init(vcpu, kvm, id);
1359 	if (err)
1360 		goto free_vcpu;
1361 
1362 	vcpu->arch.shared = &vcpu->arch.shregs;
1363 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1364 	/*
1365 	 * The shared struct is never shared on HV,
1366 	 * so we can always use host endianness
1367 	 */
1368 #ifdef __BIG_ENDIAN__
1369 	vcpu->arch.shared_big_endian = true;
1370 #else
1371 	vcpu->arch.shared_big_endian = false;
1372 #endif
1373 #endif
1374 	vcpu->arch.mmcr[0] = MMCR0_FC;
1375 	vcpu->arch.ctrl = CTRL_RUNLATCH;
1376 	/* default to host PVR, since we can't spoof it */
1377 	kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1378 	spin_lock_init(&vcpu->arch.vpa_update_lock);
1379 	spin_lock_init(&vcpu->arch.tbacct_lock);
1380 	vcpu->arch.busy_preempt = TB_NIL;
1381 	vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1382 
1383 	kvmppc_mmu_book3s_hv_init(vcpu);
1384 
1385 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1386 
1387 	init_waitqueue_head(&vcpu->arch.cpu_run);
1388 
1389 	mutex_lock(&kvm->lock);
1390 	vcore = kvm->arch.vcores[core];
1391 	if (!vcore) {
1392 		vcore = kvmppc_vcore_create(kvm, core);
1393 		kvm->arch.vcores[core] = vcore;
1394 		kvm->arch.online_vcores++;
1395 	}
1396 	mutex_unlock(&kvm->lock);
1397 
1398 	if (!vcore)
1399 		goto free_vcpu;
1400 
1401 	spin_lock(&vcore->lock);
1402 	++vcore->num_threads;
1403 	spin_unlock(&vcore->lock);
1404 	vcpu->arch.vcore = vcore;
1405 	vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1406 
1407 	vcpu->arch.cpu_type = KVM_CPU_3S_64;
1408 	kvmppc_sanity_check(vcpu);
1409 
1410 	return vcpu;
1411 
1412 free_vcpu:
1413 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1414 out:
1415 	return ERR_PTR(err);
1416 }
1417 
1418 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1419 {
1420 	if (vpa->pinned_addr)
1421 		kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1422 					vpa->dirty);
1423 }
1424 
1425 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
1426 {
1427 	spin_lock(&vcpu->arch.vpa_update_lock);
1428 	unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1429 	unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1430 	unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1431 	spin_unlock(&vcpu->arch.vpa_update_lock);
1432 	kvm_vcpu_uninit(vcpu);
1433 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1434 }
1435 
1436 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1437 {
1438 	/* Indicate we want to get back into the guest */
1439 	return 1;
1440 }
1441 
1442 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1443 {
1444 	unsigned long dec_nsec, now;
1445 
1446 	now = get_tb();
1447 	if (now > vcpu->arch.dec_expires) {
1448 		/* decrementer has already gone negative */
1449 		kvmppc_core_queue_dec(vcpu);
1450 		kvmppc_core_prepare_to_enter(vcpu);
1451 		return;
1452 	}
1453 	dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1454 		   / tb_ticks_per_sec;
1455 	hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1456 		      HRTIMER_MODE_REL);
1457 	vcpu->arch.timer_running = 1;
1458 }
1459 
1460 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1461 {
1462 	vcpu->arch.ceded = 0;
1463 	if (vcpu->arch.timer_running) {
1464 		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1465 		vcpu->arch.timer_running = 0;
1466 	}
1467 }
1468 
1469 extern void __kvmppc_vcore_entry(void);
1470 
1471 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1472 				   struct kvm_vcpu *vcpu)
1473 {
1474 	u64 now;
1475 
1476 	if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1477 		return;
1478 	spin_lock_irq(&vcpu->arch.tbacct_lock);
1479 	now = mftb();
1480 	vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1481 		vcpu->arch.stolen_logged;
1482 	vcpu->arch.busy_preempt = now;
1483 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1484 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
1485 	--vc->n_runnable;
1486 	list_del(&vcpu->arch.run_list);
1487 }
1488 
1489 static int kvmppc_grab_hwthread(int cpu)
1490 {
1491 	struct paca_struct *tpaca;
1492 	long timeout = 1000;
1493 
1494 	tpaca = &paca[cpu];
1495 
1496 	/* Ensure the thread won't go into the kernel if it wakes */
1497 	tpaca->kvm_hstate.hwthread_req = 1;
1498 	tpaca->kvm_hstate.kvm_vcpu = NULL;
1499 
1500 	/*
1501 	 * If the thread is already executing in the kernel (e.g. handling
1502 	 * a stray interrupt), wait for it to get back to nap mode.
1503 	 * The smp_mb() is to ensure that our setting of hwthread_req
1504 	 * is visible before we look at hwthread_state, so if this
1505 	 * races with the code at system_reset_pSeries and the thread
1506 	 * misses our setting of hwthread_req, we are sure to see its
1507 	 * setting of hwthread_state, and vice versa.
1508 	 */
1509 	smp_mb();
1510 	while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1511 		if (--timeout <= 0) {
1512 			pr_err("KVM: couldn't grab cpu %d\n", cpu);
1513 			return -EBUSY;
1514 		}
1515 		udelay(1);
1516 	}
1517 	return 0;
1518 }
1519 
1520 static void kvmppc_release_hwthread(int cpu)
1521 {
1522 	struct paca_struct *tpaca;
1523 
1524 	tpaca = &paca[cpu];
1525 	tpaca->kvm_hstate.hwthread_req = 0;
1526 	tpaca->kvm_hstate.kvm_vcpu = NULL;
1527 }
1528 
1529 static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
1530 {
1531 	int cpu;
1532 	struct paca_struct *tpaca;
1533 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
1534 
1535 	if (vcpu->arch.timer_running) {
1536 		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1537 		vcpu->arch.timer_running = 0;
1538 	}
1539 	cpu = vc->pcpu + vcpu->arch.ptid;
1540 	tpaca = &paca[cpu];
1541 	tpaca->kvm_hstate.kvm_vcpu = vcpu;
1542 	tpaca->kvm_hstate.kvm_vcore = vc;
1543 	tpaca->kvm_hstate.ptid = vcpu->arch.ptid;
1544 	vcpu->cpu = vc->pcpu;
1545 	smp_wmb();
1546 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
1547 	if (cpu != smp_processor_id()) {
1548 		xics_wake_cpu(cpu);
1549 		if (vcpu->arch.ptid)
1550 			++vc->n_woken;
1551 	}
1552 #endif
1553 }
1554 
1555 static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc)
1556 {
1557 	int i;
1558 
1559 	HMT_low();
1560 	i = 0;
1561 	while (vc->nap_count < vc->n_woken) {
1562 		if (++i >= 1000000) {
1563 			pr_err("kvmppc_wait_for_nap timeout %d %d\n",
1564 			       vc->nap_count, vc->n_woken);
1565 			break;
1566 		}
1567 		cpu_relax();
1568 	}
1569 	HMT_medium();
1570 }
1571 
1572 /*
1573  * Check that we are on thread 0 and that any other threads in
1574  * this core are off-line.  Then grab the threads so they can't
1575  * enter the kernel.
1576  */
1577 static int on_primary_thread(void)
1578 {
1579 	int cpu = smp_processor_id();
1580 	int thr;
1581 
1582 	/* Are we on a primary subcore? */
1583 	if (cpu_thread_in_subcore(cpu))
1584 		return 0;
1585 
1586 	thr = 0;
1587 	while (++thr < threads_per_subcore)
1588 		if (cpu_online(cpu + thr))
1589 			return 0;
1590 
1591 	/* Grab all hw threads so they can't go into the kernel */
1592 	for (thr = 1; thr < threads_per_subcore; ++thr) {
1593 		if (kvmppc_grab_hwthread(cpu + thr)) {
1594 			/* Couldn't grab one; let the others go */
1595 			do {
1596 				kvmppc_release_hwthread(cpu + thr);
1597 			} while (--thr > 0);
1598 			return 0;
1599 		}
1600 	}
1601 	return 1;
1602 }
1603 
1604 static void kvmppc_start_saving_l2_cache(struct kvmppc_vcore *vc)
1605 {
1606 	phys_addr_t phy_addr, mpp_addr;
1607 
1608 	phy_addr = (phys_addr_t)virt_to_phys(vc->mpp_buffer);
1609 	mpp_addr = phy_addr & PPC_MPPE_ADDRESS_MASK;
1610 
1611 	mtspr(SPRN_MPPR, mpp_addr | PPC_MPPR_FETCH_ABORT);
1612 	logmpp(mpp_addr | PPC_LOGMPP_LOG_L2);
1613 
1614 	vc->mpp_buffer_is_valid = true;
1615 }
1616 
1617 static void kvmppc_start_restoring_l2_cache(const struct kvmppc_vcore *vc)
1618 {
1619 	phys_addr_t phy_addr, mpp_addr;
1620 
1621 	phy_addr = virt_to_phys(vc->mpp_buffer);
1622 	mpp_addr = phy_addr & PPC_MPPE_ADDRESS_MASK;
1623 
1624 	/* We must abort any in-progress save operations to ensure
1625 	 * the table is valid so that prefetch engine knows when to
1626 	 * stop prefetching. */
1627 	logmpp(mpp_addr | PPC_LOGMPP_LOG_ABORT);
1628 	mtspr(SPRN_MPPR, mpp_addr | PPC_MPPR_FETCH_WHOLE_TABLE);
1629 }
1630 
1631 /*
1632  * Run a set of guest threads on a physical core.
1633  * Called with vc->lock held.
1634  */
1635 static void kvmppc_run_core(struct kvmppc_vcore *vc)
1636 {
1637 	struct kvm_vcpu *vcpu, *vnext;
1638 	long ret;
1639 	u64 now;
1640 	int i, need_vpa_update;
1641 	int srcu_idx;
1642 	struct kvm_vcpu *vcpus_to_update[threads_per_core];
1643 
1644 	/* don't start if any threads have a signal pending */
1645 	need_vpa_update = 0;
1646 	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1647 		if (signal_pending(vcpu->arch.run_task))
1648 			return;
1649 		if (vcpu->arch.vpa.update_pending ||
1650 		    vcpu->arch.slb_shadow.update_pending ||
1651 		    vcpu->arch.dtl.update_pending)
1652 			vcpus_to_update[need_vpa_update++] = vcpu;
1653 	}
1654 
1655 	/*
1656 	 * Initialize *vc, in particular vc->vcore_state, so we can
1657 	 * drop the vcore lock if necessary.
1658 	 */
1659 	vc->n_woken = 0;
1660 	vc->nap_count = 0;
1661 	vc->entry_exit_count = 0;
1662 	vc->vcore_state = VCORE_STARTING;
1663 	vc->in_guest = 0;
1664 	vc->napping_threads = 0;
1665 
1666 	/*
1667 	 * Updating any of the vpas requires calling kvmppc_pin_guest_page,
1668 	 * which can't be called with any spinlocks held.
1669 	 */
1670 	if (need_vpa_update) {
1671 		spin_unlock(&vc->lock);
1672 		for (i = 0; i < need_vpa_update; ++i)
1673 			kvmppc_update_vpas(vcpus_to_update[i]);
1674 		spin_lock(&vc->lock);
1675 	}
1676 
1677 	/*
1678 	 * Make sure we are running on primary threads, and that secondary
1679 	 * threads are offline.  Also check if the number of threads in this
1680 	 * guest are greater than the current system threads per guest.
1681 	 */
1682 	if ((threads_per_core > 1) &&
1683 	    ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
1684 		list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1685 			vcpu->arch.ret = -EBUSY;
1686 		goto out;
1687 	}
1688 
1689 
1690 	vc->pcpu = smp_processor_id();
1691 	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1692 		kvmppc_start_thread(vcpu);
1693 		kvmppc_create_dtl_entry(vcpu, vc);
1694 	}
1695 
1696 	/* Set this explicitly in case thread 0 doesn't have a vcpu */
1697 	get_paca()->kvm_hstate.kvm_vcore = vc;
1698 	get_paca()->kvm_hstate.ptid = 0;
1699 
1700 	vc->vcore_state = VCORE_RUNNING;
1701 	preempt_disable();
1702 	spin_unlock(&vc->lock);
1703 
1704 	kvm_guest_enter();
1705 
1706 	srcu_idx = srcu_read_lock(&vc->kvm->srcu);
1707 
1708 	if (vc->mpp_buffer_is_valid)
1709 		kvmppc_start_restoring_l2_cache(vc);
1710 
1711 	__kvmppc_vcore_entry();
1712 
1713 	spin_lock(&vc->lock);
1714 
1715 	if (vc->mpp_buffer)
1716 		kvmppc_start_saving_l2_cache(vc);
1717 
1718 	/* disable sending of IPIs on virtual external irqs */
1719 	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
1720 		vcpu->cpu = -1;
1721 	/* wait for secondary threads to finish writing their state to memory */
1722 	if (vc->nap_count < vc->n_woken)
1723 		kvmppc_wait_for_nap(vc);
1724 	for (i = 0; i < threads_per_subcore; ++i)
1725 		kvmppc_release_hwthread(vc->pcpu + i);
1726 	/* prevent other vcpu threads from doing kvmppc_start_thread() now */
1727 	vc->vcore_state = VCORE_EXITING;
1728 	spin_unlock(&vc->lock);
1729 
1730 	srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
1731 
1732 	/* make sure updates to secondary vcpu structs are visible now */
1733 	smp_mb();
1734 	kvm_guest_exit();
1735 
1736 	preempt_enable();
1737 	cond_resched();
1738 
1739 	spin_lock(&vc->lock);
1740 	now = get_tb();
1741 	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
1742 		/* cancel pending dec exception if dec is positive */
1743 		if (now < vcpu->arch.dec_expires &&
1744 		    kvmppc_core_pending_dec(vcpu))
1745 			kvmppc_core_dequeue_dec(vcpu);
1746 
1747 		ret = RESUME_GUEST;
1748 		if (vcpu->arch.trap)
1749 			ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
1750 						    vcpu->arch.run_task);
1751 
1752 		vcpu->arch.ret = ret;
1753 		vcpu->arch.trap = 0;
1754 
1755 		if (vcpu->arch.ceded) {
1756 			if (!is_kvmppc_resume_guest(ret))
1757 				kvmppc_end_cede(vcpu);
1758 			else
1759 				kvmppc_set_timer(vcpu);
1760 		}
1761 	}
1762 
1763  out:
1764 	vc->vcore_state = VCORE_INACTIVE;
1765 	list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1766 				 arch.run_list) {
1767 		if (!is_kvmppc_resume_guest(vcpu->arch.ret)) {
1768 			kvmppc_remove_runnable(vc, vcpu);
1769 			wake_up(&vcpu->arch.cpu_run);
1770 		}
1771 	}
1772 }
1773 
1774 /*
1775  * Wait for some other vcpu thread to execute us, and
1776  * wake us up when we need to handle something in the host.
1777  */
1778 static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
1779 {
1780 	DEFINE_WAIT(wait);
1781 
1782 	prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
1783 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
1784 		schedule();
1785 	finish_wait(&vcpu->arch.cpu_run, &wait);
1786 }
1787 
1788 /*
1789  * All the vcpus in this vcore are idle, so wait for a decrementer
1790  * or external interrupt to one of the vcpus.  vc->lock is held.
1791  */
1792 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
1793 {
1794 	DEFINE_WAIT(wait);
1795 
1796 	prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
1797 	vc->vcore_state = VCORE_SLEEPING;
1798 	spin_unlock(&vc->lock);
1799 	schedule();
1800 	finish_wait(&vc->wq, &wait);
1801 	spin_lock(&vc->lock);
1802 	vc->vcore_state = VCORE_INACTIVE;
1803 }
1804 
1805 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1806 {
1807 	int n_ceded;
1808 	struct kvmppc_vcore *vc;
1809 	struct kvm_vcpu *v, *vn;
1810 
1811 	kvm_run->exit_reason = 0;
1812 	vcpu->arch.ret = RESUME_GUEST;
1813 	vcpu->arch.trap = 0;
1814 	kvmppc_update_vpas(vcpu);
1815 
1816 	/*
1817 	 * Synchronize with other threads in this virtual core
1818 	 */
1819 	vc = vcpu->arch.vcore;
1820 	spin_lock(&vc->lock);
1821 	vcpu->arch.ceded = 0;
1822 	vcpu->arch.run_task = current;
1823 	vcpu->arch.kvm_run = kvm_run;
1824 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
1825 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
1826 	vcpu->arch.busy_preempt = TB_NIL;
1827 	list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
1828 	++vc->n_runnable;
1829 
1830 	/*
1831 	 * This happens the first time this is called for a vcpu.
1832 	 * If the vcore is already running, we may be able to start
1833 	 * this thread straight away and have it join in.
1834 	 */
1835 	if (!signal_pending(current)) {
1836 		if (vc->vcore_state == VCORE_RUNNING &&
1837 		    VCORE_EXIT_COUNT(vc) == 0) {
1838 			kvmppc_create_dtl_entry(vcpu, vc);
1839 			kvmppc_start_thread(vcpu);
1840 		} else if (vc->vcore_state == VCORE_SLEEPING) {
1841 			wake_up(&vc->wq);
1842 		}
1843 
1844 	}
1845 
1846 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1847 	       !signal_pending(current)) {
1848 		if (vc->vcore_state != VCORE_INACTIVE) {
1849 			spin_unlock(&vc->lock);
1850 			kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
1851 			spin_lock(&vc->lock);
1852 			continue;
1853 		}
1854 		list_for_each_entry_safe(v, vn, &vc->runnable_threads,
1855 					 arch.run_list) {
1856 			kvmppc_core_prepare_to_enter(v);
1857 			if (signal_pending(v->arch.run_task)) {
1858 				kvmppc_remove_runnable(vc, v);
1859 				v->stat.signal_exits++;
1860 				v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
1861 				v->arch.ret = -EINTR;
1862 				wake_up(&v->arch.cpu_run);
1863 			}
1864 		}
1865 		if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1866 			break;
1867 		vc->runner = vcpu;
1868 		n_ceded = 0;
1869 		list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
1870 			if (!v->arch.pending_exceptions)
1871 				n_ceded += v->arch.ceded;
1872 			else
1873 				v->arch.ceded = 0;
1874 		}
1875 		if (n_ceded == vc->n_runnable)
1876 			kvmppc_vcore_blocked(vc);
1877 		else
1878 			kvmppc_run_core(vc);
1879 		vc->runner = NULL;
1880 	}
1881 
1882 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
1883 	       (vc->vcore_state == VCORE_RUNNING ||
1884 		vc->vcore_state == VCORE_EXITING)) {
1885 		spin_unlock(&vc->lock);
1886 		kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
1887 		spin_lock(&vc->lock);
1888 	}
1889 
1890 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
1891 		kvmppc_remove_runnable(vc, vcpu);
1892 		vcpu->stat.signal_exits++;
1893 		kvm_run->exit_reason = KVM_EXIT_INTR;
1894 		vcpu->arch.ret = -EINTR;
1895 	}
1896 
1897 	if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
1898 		/* Wake up some vcpu to run the core */
1899 		v = list_first_entry(&vc->runnable_threads,
1900 				     struct kvm_vcpu, arch.run_list);
1901 		wake_up(&v->arch.cpu_run);
1902 	}
1903 
1904 	spin_unlock(&vc->lock);
1905 	return vcpu->arch.ret;
1906 }
1907 
1908 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
1909 {
1910 	int r;
1911 	int srcu_idx;
1912 
1913 	if (!vcpu->arch.sane) {
1914 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1915 		return -EINVAL;
1916 	}
1917 
1918 	kvmppc_core_prepare_to_enter(vcpu);
1919 
1920 	/* No need to go into the guest when all we'll do is come back out */
1921 	if (signal_pending(current)) {
1922 		run->exit_reason = KVM_EXIT_INTR;
1923 		return -EINTR;
1924 	}
1925 
1926 	atomic_inc(&vcpu->kvm->arch.vcpus_running);
1927 	/* Order vcpus_running vs. rma_setup_done, see kvmppc_alloc_reset_hpt */
1928 	smp_mb();
1929 
1930 	/* On the first time here, set up HTAB and VRMA or RMA */
1931 	if (!vcpu->kvm->arch.rma_setup_done) {
1932 		r = kvmppc_hv_setup_htab_rma(vcpu);
1933 		if (r)
1934 			goto out;
1935 	}
1936 
1937 	flush_fp_to_thread(current);
1938 	flush_altivec_to_thread(current);
1939 	flush_vsx_to_thread(current);
1940 	vcpu->arch.wqp = &vcpu->arch.vcore->wq;
1941 	vcpu->arch.pgdir = current->mm->pgd;
1942 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1943 
1944 	do {
1945 		r = kvmppc_run_vcpu(run, vcpu);
1946 
1947 		if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
1948 		    !(vcpu->arch.shregs.msr & MSR_PR)) {
1949 			r = kvmppc_pseries_do_hcall(vcpu);
1950 			kvmppc_core_prepare_to_enter(vcpu);
1951 		} else if (r == RESUME_PAGE_FAULT) {
1952 			srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1953 			r = kvmppc_book3s_hv_page_fault(run, vcpu,
1954 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
1955 			srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1956 		}
1957 	} while (is_kvmppc_resume_guest(r));
1958 
1959  out:
1960 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1961 	atomic_dec(&vcpu->kvm->arch.vcpus_running);
1962 	return r;
1963 }
1964 
1965 
1966 /* Work out RMLS (real mode limit selector) field value for a given RMA size.
1967    Assumes POWER7 or PPC970. */
1968 static inline int lpcr_rmls(unsigned long rma_size)
1969 {
1970 	switch (rma_size) {
1971 	case 32ul << 20:	/* 32 MB */
1972 		if (cpu_has_feature(CPU_FTR_ARCH_206))
1973 			return 8;	/* only supported on POWER7 */
1974 		return -1;
1975 	case 64ul << 20:	/* 64 MB */
1976 		return 3;
1977 	case 128ul << 20:	/* 128 MB */
1978 		return 7;
1979 	case 256ul << 20:	/* 256 MB */
1980 		return 4;
1981 	case 1ul << 30:		/* 1 GB */
1982 		return 2;
1983 	case 16ul << 30:	/* 16 GB */
1984 		return 1;
1985 	case 256ul << 30:	/* 256 GB */
1986 		return 0;
1987 	default:
1988 		return -1;
1989 	}
1990 }
1991 
1992 static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1993 {
1994 	struct page *page;
1995 	struct kvm_rma_info *ri = vma->vm_file->private_data;
1996 
1997 	if (vmf->pgoff >= kvm_rma_pages)
1998 		return VM_FAULT_SIGBUS;
1999 
2000 	page = pfn_to_page(ri->base_pfn + vmf->pgoff);
2001 	get_page(page);
2002 	vmf->page = page;
2003 	return 0;
2004 }
2005 
2006 static const struct vm_operations_struct kvm_rma_vm_ops = {
2007 	.fault = kvm_rma_fault,
2008 };
2009 
2010 static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma)
2011 {
2012 	vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
2013 	vma->vm_ops = &kvm_rma_vm_ops;
2014 	return 0;
2015 }
2016 
2017 static int kvm_rma_release(struct inode *inode, struct file *filp)
2018 {
2019 	struct kvm_rma_info *ri = filp->private_data;
2020 
2021 	kvm_release_rma(ri);
2022 	return 0;
2023 }
2024 
2025 static const struct file_operations kvm_rma_fops = {
2026 	.mmap           = kvm_rma_mmap,
2027 	.release	= kvm_rma_release,
2028 };
2029 
2030 static long kvm_vm_ioctl_allocate_rma(struct kvm *kvm,
2031 				      struct kvm_allocate_rma *ret)
2032 {
2033 	long fd;
2034 	struct kvm_rma_info *ri;
2035 	/*
2036 	 * Only do this on PPC970 in HV mode
2037 	 */
2038 	if (!cpu_has_feature(CPU_FTR_HVMODE) ||
2039 	    !cpu_has_feature(CPU_FTR_ARCH_201))
2040 		return -EINVAL;
2041 
2042 	if (!kvm_rma_pages)
2043 		return -EINVAL;
2044 
2045 	ri = kvm_alloc_rma();
2046 	if (!ri)
2047 		return -ENOMEM;
2048 
2049 	fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR | O_CLOEXEC);
2050 	if (fd < 0)
2051 		kvm_release_rma(ri);
2052 
2053 	ret->rma_size = kvm_rma_pages << PAGE_SHIFT;
2054 	return fd;
2055 }
2056 
2057 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
2058 				     int linux_psize)
2059 {
2060 	struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
2061 
2062 	if (!def->shift)
2063 		return;
2064 	(*sps)->page_shift = def->shift;
2065 	(*sps)->slb_enc = def->sllp;
2066 	(*sps)->enc[0].page_shift = def->shift;
2067 	(*sps)->enc[0].pte_enc = def->penc[linux_psize];
2068 	/*
2069 	 * Add 16MB MPSS support if host supports it
2070 	 */
2071 	if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
2072 		(*sps)->enc[1].page_shift = 24;
2073 		(*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
2074 	}
2075 	(*sps)++;
2076 }
2077 
2078 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
2079 					 struct kvm_ppc_smmu_info *info)
2080 {
2081 	struct kvm_ppc_one_seg_page_size *sps;
2082 
2083 	info->flags = KVM_PPC_PAGE_SIZES_REAL;
2084 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
2085 		info->flags |= KVM_PPC_1T_SEGMENTS;
2086 	info->slb_size = mmu_slb_size;
2087 
2088 	/* We only support these sizes for now, and no muti-size segments */
2089 	sps = &info->sps[0];
2090 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
2091 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
2092 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
2093 
2094 	return 0;
2095 }
2096 
2097 /*
2098  * Get (and clear) the dirty memory log for a memory slot.
2099  */
2100 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
2101 					 struct kvm_dirty_log *log)
2102 {
2103 	struct kvm_memory_slot *memslot;
2104 	int r;
2105 	unsigned long n;
2106 
2107 	mutex_lock(&kvm->slots_lock);
2108 
2109 	r = -EINVAL;
2110 	if (log->slot >= KVM_USER_MEM_SLOTS)
2111 		goto out;
2112 
2113 	memslot = id_to_memslot(kvm->memslots, log->slot);
2114 	r = -ENOENT;
2115 	if (!memslot->dirty_bitmap)
2116 		goto out;
2117 
2118 	n = kvm_dirty_bitmap_bytes(memslot);
2119 	memset(memslot->dirty_bitmap, 0, n);
2120 
2121 	r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
2122 	if (r)
2123 		goto out;
2124 
2125 	r = -EFAULT;
2126 	if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
2127 		goto out;
2128 
2129 	r = 0;
2130 out:
2131 	mutex_unlock(&kvm->slots_lock);
2132 	return r;
2133 }
2134 
2135 static void unpin_slot(struct kvm_memory_slot *memslot)
2136 {
2137 	unsigned long *physp;
2138 	unsigned long j, npages, pfn;
2139 	struct page *page;
2140 
2141 	physp = memslot->arch.slot_phys;
2142 	npages = memslot->npages;
2143 	if (!physp)
2144 		return;
2145 	for (j = 0; j < npages; j++) {
2146 		if (!(physp[j] & KVMPPC_GOT_PAGE))
2147 			continue;
2148 		pfn = physp[j] >> PAGE_SHIFT;
2149 		page = pfn_to_page(pfn);
2150 		SetPageDirty(page);
2151 		put_page(page);
2152 	}
2153 }
2154 
2155 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
2156 					struct kvm_memory_slot *dont)
2157 {
2158 	if (!dont || free->arch.rmap != dont->arch.rmap) {
2159 		vfree(free->arch.rmap);
2160 		free->arch.rmap = NULL;
2161 	}
2162 	if (!dont || free->arch.slot_phys != dont->arch.slot_phys) {
2163 		unpin_slot(free);
2164 		vfree(free->arch.slot_phys);
2165 		free->arch.slot_phys = NULL;
2166 	}
2167 }
2168 
2169 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
2170 					 unsigned long npages)
2171 {
2172 	slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
2173 	if (!slot->arch.rmap)
2174 		return -ENOMEM;
2175 	slot->arch.slot_phys = NULL;
2176 
2177 	return 0;
2178 }
2179 
2180 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
2181 					struct kvm_memory_slot *memslot,
2182 					struct kvm_userspace_memory_region *mem)
2183 {
2184 	unsigned long *phys;
2185 
2186 	/* Allocate a slot_phys array if needed */
2187 	phys = memslot->arch.slot_phys;
2188 	if (!kvm->arch.using_mmu_notifiers && !phys && memslot->npages) {
2189 		phys = vzalloc(memslot->npages * sizeof(unsigned long));
2190 		if (!phys)
2191 			return -ENOMEM;
2192 		memslot->arch.slot_phys = phys;
2193 	}
2194 
2195 	return 0;
2196 }
2197 
2198 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
2199 				struct kvm_userspace_memory_region *mem,
2200 				const struct kvm_memory_slot *old)
2201 {
2202 	unsigned long npages = mem->memory_size >> PAGE_SHIFT;
2203 	struct kvm_memory_slot *memslot;
2204 
2205 	if (npages && old->npages) {
2206 		/*
2207 		 * If modifying a memslot, reset all the rmap dirty bits.
2208 		 * If this is a new memslot, we don't need to do anything
2209 		 * since the rmap array starts out as all zeroes,
2210 		 * i.e. no pages are dirty.
2211 		 */
2212 		memslot = id_to_memslot(kvm->memslots, mem->slot);
2213 		kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
2214 	}
2215 }
2216 
2217 /*
2218  * Update LPCR values in kvm->arch and in vcores.
2219  * Caller must hold kvm->lock.
2220  */
2221 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
2222 {
2223 	long int i;
2224 	u32 cores_done = 0;
2225 
2226 	if ((kvm->arch.lpcr & mask) == lpcr)
2227 		return;
2228 
2229 	kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
2230 
2231 	for (i = 0; i < KVM_MAX_VCORES; ++i) {
2232 		struct kvmppc_vcore *vc = kvm->arch.vcores[i];
2233 		if (!vc)
2234 			continue;
2235 		spin_lock(&vc->lock);
2236 		vc->lpcr = (vc->lpcr & ~mask) | lpcr;
2237 		spin_unlock(&vc->lock);
2238 		if (++cores_done >= kvm->arch.online_vcores)
2239 			break;
2240 	}
2241 }
2242 
2243 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
2244 {
2245 	return;
2246 }
2247 
2248 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
2249 {
2250 	int err = 0;
2251 	struct kvm *kvm = vcpu->kvm;
2252 	struct kvm_rma_info *ri = NULL;
2253 	unsigned long hva;
2254 	struct kvm_memory_slot *memslot;
2255 	struct vm_area_struct *vma;
2256 	unsigned long lpcr = 0, senc;
2257 	unsigned long lpcr_mask = 0;
2258 	unsigned long psize, porder;
2259 	unsigned long rma_size;
2260 	unsigned long rmls;
2261 	unsigned long *physp;
2262 	unsigned long i, npages;
2263 	int srcu_idx;
2264 
2265 	mutex_lock(&kvm->lock);
2266 	if (kvm->arch.rma_setup_done)
2267 		goto out;	/* another vcpu beat us to it */
2268 
2269 	/* Allocate hashed page table (if not done already) and reset it */
2270 	if (!kvm->arch.hpt_virt) {
2271 		err = kvmppc_alloc_hpt(kvm, NULL);
2272 		if (err) {
2273 			pr_err("KVM: Couldn't alloc HPT\n");
2274 			goto out;
2275 		}
2276 	}
2277 
2278 	/* Look up the memslot for guest physical address 0 */
2279 	srcu_idx = srcu_read_lock(&kvm->srcu);
2280 	memslot = gfn_to_memslot(kvm, 0);
2281 
2282 	/* We must have some memory at 0 by now */
2283 	err = -EINVAL;
2284 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
2285 		goto out_srcu;
2286 
2287 	/* Look up the VMA for the start of this memory slot */
2288 	hva = memslot->userspace_addr;
2289 	down_read(&current->mm->mmap_sem);
2290 	vma = find_vma(current->mm, hva);
2291 	if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
2292 		goto up_out;
2293 
2294 	psize = vma_kernel_pagesize(vma);
2295 	porder = __ilog2(psize);
2296 
2297 	/* Is this one of our preallocated RMAs? */
2298 	if (vma->vm_file && vma->vm_file->f_op == &kvm_rma_fops &&
2299 	    hva == vma->vm_start)
2300 		ri = vma->vm_file->private_data;
2301 
2302 	up_read(&current->mm->mmap_sem);
2303 
2304 	if (!ri) {
2305 		/* On POWER7, use VRMA; on PPC970, give up */
2306 		err = -EPERM;
2307 		if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2308 			pr_err("KVM: CPU requires an RMO\n");
2309 			goto out_srcu;
2310 		}
2311 
2312 		/* We can handle 4k, 64k or 16M pages in the VRMA */
2313 		err = -EINVAL;
2314 		if (!(psize == 0x1000 || psize == 0x10000 ||
2315 		      psize == 0x1000000))
2316 			goto out_srcu;
2317 
2318 		/* Update VRMASD field in the LPCR */
2319 		senc = slb_pgsize_encoding(psize);
2320 		kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
2321 			(VRMA_VSID << SLB_VSID_SHIFT_1T);
2322 		lpcr_mask = LPCR_VRMASD;
2323 		/* the -4 is to account for senc values starting at 0x10 */
2324 		lpcr = senc << (LPCR_VRMASD_SH - 4);
2325 
2326 		/* Create HPTEs in the hash page table for the VRMA */
2327 		kvmppc_map_vrma(vcpu, memslot, porder);
2328 
2329 	} else {
2330 		/* Set up to use an RMO region */
2331 		rma_size = kvm_rma_pages;
2332 		if (rma_size > memslot->npages)
2333 			rma_size = memslot->npages;
2334 		rma_size <<= PAGE_SHIFT;
2335 		rmls = lpcr_rmls(rma_size);
2336 		err = -EINVAL;
2337 		if ((long)rmls < 0) {
2338 			pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size);
2339 			goto out_srcu;
2340 		}
2341 		atomic_inc(&ri->use_count);
2342 		kvm->arch.rma = ri;
2343 
2344 		/* Update LPCR and RMOR */
2345 		if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2346 			/* PPC970; insert RMLS value (split field) in HID4 */
2347 			lpcr_mask = (1ul << HID4_RMLS0_SH) |
2348 				(3ul << HID4_RMLS2_SH) | HID4_RMOR;
2349 			lpcr = ((rmls >> 2) << HID4_RMLS0_SH) |
2350 				((rmls & 3) << HID4_RMLS2_SH);
2351 			/* RMOR is also in HID4 */
2352 			lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff)
2353 				<< HID4_RMOR_SH;
2354 		} else {
2355 			/* POWER7 */
2356 			lpcr_mask = LPCR_VPM0 | LPCR_VRMA_L | LPCR_RMLS;
2357 			lpcr = rmls << LPCR_RMLS_SH;
2358 			kvm->arch.rmor = ri->base_pfn << PAGE_SHIFT;
2359 		}
2360 		pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n",
2361 			ri->base_pfn << PAGE_SHIFT, rma_size, lpcr);
2362 
2363 		/* Initialize phys addrs of pages in RMO */
2364 		npages = kvm_rma_pages;
2365 		porder = __ilog2(npages);
2366 		physp = memslot->arch.slot_phys;
2367 		if (physp) {
2368 			if (npages > memslot->npages)
2369 				npages = memslot->npages;
2370 			spin_lock(&kvm->arch.slot_phys_lock);
2371 			for (i = 0; i < npages; ++i)
2372 				physp[i] = ((ri->base_pfn + i) << PAGE_SHIFT) +
2373 					porder;
2374 			spin_unlock(&kvm->arch.slot_phys_lock);
2375 		}
2376 	}
2377 
2378 	kvmppc_update_lpcr(kvm, lpcr, lpcr_mask);
2379 
2380 	/* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */
2381 	smp_wmb();
2382 	kvm->arch.rma_setup_done = 1;
2383 	err = 0;
2384  out_srcu:
2385 	srcu_read_unlock(&kvm->srcu, srcu_idx);
2386  out:
2387 	mutex_unlock(&kvm->lock);
2388 	return err;
2389 
2390  up_out:
2391 	up_read(&current->mm->mmap_sem);
2392 	goto out_srcu;
2393 }
2394 
2395 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
2396 {
2397 	unsigned long lpcr, lpid;
2398 
2399 	/* Allocate the guest's logical partition ID */
2400 
2401 	lpid = kvmppc_alloc_lpid();
2402 	if ((long)lpid < 0)
2403 		return -ENOMEM;
2404 	kvm->arch.lpid = lpid;
2405 
2406 	/*
2407 	 * Since we don't flush the TLB when tearing down a VM,
2408 	 * and this lpid might have previously been used,
2409 	 * make sure we flush on each core before running the new VM.
2410 	 */
2411 	cpumask_setall(&kvm->arch.need_tlb_flush);
2412 
2413 	/* Start out with the default set of hcalls enabled */
2414 	memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
2415 	       sizeof(kvm->arch.enabled_hcalls));
2416 
2417 	kvm->arch.rma = NULL;
2418 
2419 	kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
2420 
2421 	if (cpu_has_feature(CPU_FTR_ARCH_201)) {
2422 		/* PPC970; HID4 is effectively the LPCR */
2423 		kvm->arch.host_lpid = 0;
2424 		kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4);
2425 		lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH));
2426 		lpcr |= ((lpid >> 4) << HID4_LPID1_SH) |
2427 			((lpid & 0xf) << HID4_LPID5_SH);
2428 	} else {
2429 		/* POWER7; init LPCR for virtual RMA mode */
2430 		kvm->arch.host_lpid = mfspr(SPRN_LPID);
2431 		kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
2432 		lpcr &= LPCR_PECE | LPCR_LPES;
2433 		lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
2434 			LPCR_VPM0 | LPCR_VPM1;
2435 		kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
2436 			(VRMA_VSID << SLB_VSID_SHIFT_1T);
2437 		/* On POWER8 turn on online bit to enable PURR/SPURR */
2438 		if (cpu_has_feature(CPU_FTR_ARCH_207S))
2439 			lpcr |= LPCR_ONL;
2440 	}
2441 	kvm->arch.lpcr = lpcr;
2442 
2443 	kvm->arch.using_mmu_notifiers = !!cpu_has_feature(CPU_FTR_ARCH_206);
2444 	spin_lock_init(&kvm->arch.slot_phys_lock);
2445 
2446 	/*
2447 	 * Track that we now have a HV mode VM active. This blocks secondary
2448 	 * CPU threads from coming online.
2449 	 */
2450 	kvm_hv_vm_activated();
2451 
2452 	return 0;
2453 }
2454 
2455 static void kvmppc_free_vcores(struct kvm *kvm)
2456 {
2457 	long int i;
2458 
2459 	for (i = 0; i < KVM_MAX_VCORES; ++i) {
2460 		if (kvm->arch.vcores[i] && kvm->arch.vcores[i]->mpp_buffer) {
2461 			struct kvmppc_vcore *vc = kvm->arch.vcores[i];
2462 			free_pages((unsigned long)vc->mpp_buffer,
2463 				   MPP_BUFFER_ORDER);
2464 		}
2465 		kfree(kvm->arch.vcores[i]);
2466 	}
2467 	kvm->arch.online_vcores = 0;
2468 }
2469 
2470 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
2471 {
2472 	kvm_hv_vm_deactivated();
2473 
2474 	kvmppc_free_vcores(kvm);
2475 	if (kvm->arch.rma) {
2476 		kvm_release_rma(kvm->arch.rma);
2477 		kvm->arch.rma = NULL;
2478 	}
2479 
2480 	kvmppc_free_hpt(kvm);
2481 }
2482 
2483 /* We don't need to emulate any privileged instructions or dcbz */
2484 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
2485 				     unsigned int inst, int *advance)
2486 {
2487 	return EMULATE_FAIL;
2488 }
2489 
2490 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
2491 					ulong spr_val)
2492 {
2493 	return EMULATE_FAIL;
2494 }
2495 
2496 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
2497 					ulong *spr_val)
2498 {
2499 	return EMULATE_FAIL;
2500 }
2501 
2502 static int kvmppc_core_check_processor_compat_hv(void)
2503 {
2504 	if (!cpu_has_feature(CPU_FTR_HVMODE))
2505 		return -EIO;
2506 	return 0;
2507 }
2508 
2509 static long kvm_arch_vm_ioctl_hv(struct file *filp,
2510 				 unsigned int ioctl, unsigned long arg)
2511 {
2512 	struct kvm *kvm __maybe_unused = filp->private_data;
2513 	void __user *argp = (void __user *)arg;
2514 	long r;
2515 
2516 	switch (ioctl) {
2517 
2518 	case KVM_ALLOCATE_RMA: {
2519 		struct kvm_allocate_rma rma;
2520 		struct kvm *kvm = filp->private_data;
2521 
2522 		r = kvm_vm_ioctl_allocate_rma(kvm, &rma);
2523 		if (r >= 0 && copy_to_user(argp, &rma, sizeof(rma)))
2524 			r = -EFAULT;
2525 		break;
2526 	}
2527 
2528 	case KVM_PPC_ALLOCATE_HTAB: {
2529 		u32 htab_order;
2530 
2531 		r = -EFAULT;
2532 		if (get_user(htab_order, (u32 __user *)argp))
2533 			break;
2534 		r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
2535 		if (r)
2536 			break;
2537 		r = -EFAULT;
2538 		if (put_user(htab_order, (u32 __user *)argp))
2539 			break;
2540 		r = 0;
2541 		break;
2542 	}
2543 
2544 	case KVM_PPC_GET_HTAB_FD: {
2545 		struct kvm_get_htab_fd ghf;
2546 
2547 		r = -EFAULT;
2548 		if (copy_from_user(&ghf, argp, sizeof(ghf)))
2549 			break;
2550 		r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
2551 		break;
2552 	}
2553 
2554 	default:
2555 		r = -ENOTTY;
2556 	}
2557 
2558 	return r;
2559 }
2560 
2561 /*
2562  * List of hcall numbers to enable by default.
2563  * For compatibility with old userspace, we enable by default
2564  * all hcalls that were implemented before the hcall-enabling
2565  * facility was added.  Note this list should not include H_RTAS.
2566  */
2567 static unsigned int default_hcall_list[] = {
2568 	H_REMOVE,
2569 	H_ENTER,
2570 	H_READ,
2571 	H_PROTECT,
2572 	H_BULK_REMOVE,
2573 	H_GET_TCE,
2574 	H_PUT_TCE,
2575 	H_SET_DABR,
2576 	H_SET_XDABR,
2577 	H_CEDE,
2578 	H_PROD,
2579 	H_CONFER,
2580 	H_REGISTER_VPA,
2581 #ifdef CONFIG_KVM_XICS
2582 	H_EOI,
2583 	H_CPPR,
2584 	H_IPI,
2585 	H_IPOLL,
2586 	H_XIRR,
2587 	H_XIRR_X,
2588 #endif
2589 	0
2590 };
2591 
2592 static void init_default_hcalls(void)
2593 {
2594 	int i;
2595 	unsigned int hcall;
2596 
2597 	for (i = 0; default_hcall_list[i]; ++i) {
2598 		hcall = default_hcall_list[i];
2599 		WARN_ON(!kvmppc_hcall_impl_hv(hcall));
2600 		__set_bit(hcall / 4, default_enabled_hcalls);
2601 	}
2602 }
2603 
2604 static struct kvmppc_ops kvm_ops_hv = {
2605 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
2606 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
2607 	.get_one_reg = kvmppc_get_one_reg_hv,
2608 	.set_one_reg = kvmppc_set_one_reg_hv,
2609 	.vcpu_load   = kvmppc_core_vcpu_load_hv,
2610 	.vcpu_put    = kvmppc_core_vcpu_put_hv,
2611 	.set_msr     = kvmppc_set_msr_hv,
2612 	.vcpu_run    = kvmppc_vcpu_run_hv,
2613 	.vcpu_create = kvmppc_core_vcpu_create_hv,
2614 	.vcpu_free   = kvmppc_core_vcpu_free_hv,
2615 	.check_requests = kvmppc_core_check_requests_hv,
2616 	.get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
2617 	.flush_memslot  = kvmppc_core_flush_memslot_hv,
2618 	.prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
2619 	.commit_memory_region  = kvmppc_core_commit_memory_region_hv,
2620 	.unmap_hva = kvm_unmap_hva_hv,
2621 	.unmap_hva_range = kvm_unmap_hva_range_hv,
2622 	.age_hva  = kvm_age_hva_hv,
2623 	.test_age_hva = kvm_test_age_hva_hv,
2624 	.set_spte_hva = kvm_set_spte_hva_hv,
2625 	.mmu_destroy  = kvmppc_mmu_destroy_hv,
2626 	.free_memslot = kvmppc_core_free_memslot_hv,
2627 	.create_memslot = kvmppc_core_create_memslot_hv,
2628 	.init_vm =  kvmppc_core_init_vm_hv,
2629 	.destroy_vm = kvmppc_core_destroy_vm_hv,
2630 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
2631 	.emulate_op = kvmppc_core_emulate_op_hv,
2632 	.emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
2633 	.emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
2634 	.fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
2635 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
2636 	.hcall_implemented = kvmppc_hcall_impl_hv,
2637 };
2638 
2639 static int kvmppc_book3s_init_hv(void)
2640 {
2641 	int r;
2642 	/*
2643 	 * FIXME!! Do we need to check on all cpus ?
2644 	 */
2645 	r = kvmppc_core_check_processor_compat_hv();
2646 	if (r < 0)
2647 		return -ENODEV;
2648 
2649 	kvm_ops_hv.owner = THIS_MODULE;
2650 	kvmppc_hv_ops = &kvm_ops_hv;
2651 
2652 	init_default_hcalls();
2653 
2654 	r = kvmppc_mmu_hv_init();
2655 	return r;
2656 }
2657 
2658 static void kvmppc_book3s_exit_hv(void)
2659 {
2660 	kvmppc_hv_ops = NULL;
2661 }
2662 
2663 module_init(kvmppc_book3s_init_hv);
2664 module_exit(kvmppc_book3s_exit_hv);
2665 MODULE_LICENSE("GPL");
2666 MODULE_ALIAS_MISCDEV(KVM_MINOR);
2667 MODULE_ALIAS("devname:kvm");
2668