1/*
2 * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
3 * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
4 *
5 * Authors:
6 *    Paul Mackerras <paulus@au1.ibm.com>
7 *    Alexander Graf <agraf@suse.de>
8 *    Kevin Wolf <mail@kevin-wolf.de>
9 *
10 * Description: KVM functions specific to running on Book 3S
11 * processors in hypervisor mode (specifically POWER7 and later).
12 *
13 * This file is derived from arch/powerpc/kvm/book3s.c,
14 * by Alexander Graf <agraf@suse.de>.
15 *
16 * This program is free software; you can redistribute it and/or modify
17 * it under the terms of the GNU General Public License, version 2, as
18 * published by the Free Software Foundation.
19 */
20
21#include <linux/kvm_host.h>
22#include <linux/err.h>
23#include <linux/slab.h>
24#include <linux/preempt.h>
25#include <linux/sched.h>
26#include <linux/delay.h>
27#include <linux/export.h>
28#include <linux/fs.h>
29#include <linux/anon_inodes.h>
30#include <linux/cpumask.h>
31#include <linux/spinlock.h>
32#include <linux/page-flags.h>
33#include <linux/srcu.h>
34#include <linux/miscdevice.h>
35#include <linux/debugfs.h>
36
37#include <asm/reg.h>
38#include <asm/cputable.h>
39#include <asm/cache.h>
40#include <asm/cacheflush.h>
41#include <asm/tlbflush.h>
42#include <asm/uaccess.h>
43#include <asm/io.h>
44#include <asm/kvm_ppc.h>
45#include <asm/kvm_book3s.h>
46#include <asm/mmu_context.h>
47#include <asm/lppaca.h>
48#include <asm/processor.h>
49#include <asm/cputhreads.h>
50#include <asm/page.h>
51#include <asm/hvcall.h>
52#include <asm/switch_to.h>
53#include <asm/smp.h>
54#include <asm/dbell.h>
55#include <linux/gfp.h>
56#include <linux/vmalloc.h>
57#include <linux/highmem.h>
58#include <linux/hugetlb.h>
59#include <linux/module.h>
60
61#include "book3s.h"
62
63#define CREATE_TRACE_POINTS
64#include "trace_hv.h"
65
66/* #define EXIT_DEBUG */
67/* #define EXIT_DEBUG_SIMPLE */
68/* #define EXIT_DEBUG_INT */
69
70/* Used to indicate that a guest page fault needs to be handled */
71#define RESUME_PAGE_FAULT	(RESUME_GUEST | RESUME_FLAG_ARCH1)
72
73/* Used as a "null" value for timebase values */
74#define TB_NIL	(~(u64)0)
75
76static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
77
78#if defined(CONFIG_PPC_64K_PAGES)
79#define MPP_BUFFER_ORDER	0
80#elif defined(CONFIG_PPC_4K_PAGES)
81#define MPP_BUFFER_ORDER	3
82#endif
83
84
85static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
86static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
87
88static bool kvmppc_ipi_thread(int cpu)
89{
90	/* On POWER8 for IPIs to threads in the same core, use msgsnd */
91	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
92		preempt_disable();
93		if (cpu_first_thread_sibling(cpu) ==
94		    cpu_first_thread_sibling(smp_processor_id())) {
95			unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
96			msg |= cpu_thread_in_core(cpu);
97			smp_mb();
98			__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
99			preempt_enable();
100			return true;
101		}
102		preempt_enable();
103	}
104
105#if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
106	if (cpu >= 0 && cpu < nr_cpu_ids && paca[cpu].kvm_hstate.xics_phys) {
107		xics_wake_cpu(cpu);
108		return true;
109	}
110#endif
111
112	return false;
113}
114
115static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
116{
117	int cpu = vcpu->cpu;
118	wait_queue_head_t *wqp;
119
120	wqp = kvm_arch_vcpu_wq(vcpu);
121	if (waitqueue_active(wqp)) {
122		wake_up_interruptible(wqp);
123		++vcpu->stat.halt_wakeup;
124	}
125
126	if (kvmppc_ipi_thread(cpu + vcpu->arch.ptid))
127		return;
128
129	/* CPU points to the first thread of the core */
130	if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
131		smp_send_reschedule(cpu);
132}
133
134/*
135 * We use the vcpu_load/put functions to measure stolen time.
136 * Stolen time is counted as time when either the vcpu is able to
137 * run as part of a virtual core, but the task running the vcore
138 * is preempted or sleeping, or when the vcpu needs something done
139 * in the kernel by the task running the vcpu, but that task is
140 * preempted or sleeping.  Those two things have to be counted
141 * separately, since one of the vcpu tasks will take on the job
142 * of running the core, and the other vcpu tasks in the vcore will
143 * sleep waiting for it to do that, but that sleep shouldn't count
144 * as stolen time.
145 *
146 * Hence we accumulate stolen time when the vcpu can run as part of
147 * a vcore using vc->stolen_tb, and the stolen time when the vcpu
148 * needs its task to do other things in the kernel (for example,
149 * service a page fault) in busy_stolen.  We don't accumulate
150 * stolen time for a vcore when it is inactive, or for a vcpu
151 * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
152 * a misnomer; it means that the vcpu task is not executing in
153 * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
154 * the kernel.  We don't have any way of dividing up that time
155 * between time that the vcpu is genuinely stopped, time that
156 * the task is actively working on behalf of the vcpu, and time
157 * that the task is preempted, so we don't count any of it as
158 * stolen.
159 *
160 * Updates to busy_stolen are protected by arch.tbacct_lock;
161 * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
162 * lock.  The stolen times are measured in units of timebase ticks.
163 * (Note that the != TB_NIL checks below are purely defensive;
164 * they should never fail.)
165 */
166
167static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
168{
169	struct kvmppc_vcore *vc = vcpu->arch.vcore;
170	unsigned long flags;
171
172	/*
173	 * We can test vc->runner without taking the vcore lock,
174	 * because only this task ever sets vc->runner to this
175	 * vcpu, and once it is set to this vcpu, only this task
176	 * ever sets it to NULL.
177	 */
178	if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE) {
179		spin_lock_irqsave(&vc->stoltb_lock, flags);
180		if (vc->preempt_tb != TB_NIL) {
181			vc->stolen_tb += mftb() - vc->preempt_tb;
182			vc->preempt_tb = TB_NIL;
183		}
184		spin_unlock_irqrestore(&vc->stoltb_lock, flags);
185	}
186	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
187	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
188	    vcpu->arch.busy_preempt != TB_NIL) {
189		vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
190		vcpu->arch.busy_preempt = TB_NIL;
191	}
192	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
193}
194
195static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
196{
197	struct kvmppc_vcore *vc = vcpu->arch.vcore;
198	unsigned long flags;
199
200	if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE) {
201		spin_lock_irqsave(&vc->stoltb_lock, flags);
202		vc->preempt_tb = mftb();
203		spin_unlock_irqrestore(&vc->stoltb_lock, flags);
204	}
205	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
206	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
207		vcpu->arch.busy_preempt = mftb();
208	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
209}
210
211static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
212{
213	/*
214	 * Check for illegal transactional state bit combination
215	 * and if we find it, force the TS field to a safe state.
216	 */
217	if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
218		msr &= ~MSR_TS_MASK;
219	vcpu->arch.shregs.msr = msr;
220	kvmppc_end_cede(vcpu);
221}
222
223void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
224{
225	vcpu->arch.pvr = pvr;
226}
227
228int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
229{
230	unsigned long pcr = 0;
231	struct kvmppc_vcore *vc = vcpu->arch.vcore;
232
233	if (arch_compat) {
234		switch (arch_compat) {
235		case PVR_ARCH_205:
236			/*
237			 * If an arch bit is set in PCR, all the defined
238			 * higher-order arch bits also have to be set.
239			 */
240			pcr = PCR_ARCH_206 | PCR_ARCH_205;
241			break;
242		case PVR_ARCH_206:
243		case PVR_ARCH_206p:
244			pcr = PCR_ARCH_206;
245			break;
246		case PVR_ARCH_207:
247			break;
248		default:
249			return -EINVAL;
250		}
251
252		if (!cpu_has_feature(CPU_FTR_ARCH_207S)) {
253			/* POWER7 can't emulate POWER8 */
254			if (!(pcr & PCR_ARCH_206))
255				return -EINVAL;
256			pcr &= ~PCR_ARCH_206;
257		}
258	}
259
260	spin_lock(&vc->lock);
261	vc->arch_compat = arch_compat;
262	vc->pcr = pcr;
263	spin_unlock(&vc->lock);
264
265	return 0;
266}
267
268void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
269{
270	int r;
271
272	pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
273	pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
274	       vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
275	for (r = 0; r < 16; ++r)
276		pr_err("r%2d = %.16lx  r%d = %.16lx\n",
277		       r, kvmppc_get_gpr(vcpu, r),
278		       r+16, kvmppc_get_gpr(vcpu, r+16));
279	pr_err("ctr = %.16lx  lr  = %.16lx\n",
280	       vcpu->arch.ctr, vcpu->arch.lr);
281	pr_err("srr0 = %.16llx srr1 = %.16llx\n",
282	       vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
283	pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
284	       vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
285	pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
286	       vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
287	pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
288	       vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
289	pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
290	pr_err("fault dar = %.16lx dsisr = %.8x\n",
291	       vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
292	pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
293	for (r = 0; r < vcpu->arch.slb_max; ++r)
294		pr_err("  ESID = %.16llx VSID = %.16llx\n",
295		       vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
296	pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
297	       vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
298	       vcpu->arch.last_inst);
299}
300
301struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
302{
303	int r;
304	struct kvm_vcpu *v, *ret = NULL;
305
306	mutex_lock(&kvm->lock);
307	kvm_for_each_vcpu(r, v, kvm) {
308		if (v->vcpu_id == id) {
309			ret = v;
310			break;
311		}
312	}
313	mutex_unlock(&kvm->lock);
314	return ret;
315}
316
317static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
318{
319	vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
320	vpa->yield_count = cpu_to_be32(1);
321}
322
323static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
324		   unsigned long addr, unsigned long len)
325{
326	/* check address is cacheline aligned */
327	if (addr & (L1_CACHE_BYTES - 1))
328		return -EINVAL;
329	spin_lock(&vcpu->arch.vpa_update_lock);
330	if (v->next_gpa != addr || v->len != len) {
331		v->next_gpa = addr;
332		v->len = addr ? len : 0;
333		v->update_pending = 1;
334	}
335	spin_unlock(&vcpu->arch.vpa_update_lock);
336	return 0;
337}
338
339/* Length for a per-processor buffer is passed in at offset 4 in the buffer */
340struct reg_vpa {
341	u32 dummy;
342	union {
343		__be16 hword;
344		__be32 word;
345	} length;
346};
347
348static int vpa_is_registered(struct kvmppc_vpa *vpap)
349{
350	if (vpap->update_pending)
351		return vpap->next_gpa != 0;
352	return vpap->pinned_addr != NULL;
353}
354
355static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
356				       unsigned long flags,
357				       unsigned long vcpuid, unsigned long vpa)
358{
359	struct kvm *kvm = vcpu->kvm;
360	unsigned long len, nb;
361	void *va;
362	struct kvm_vcpu *tvcpu;
363	int err;
364	int subfunc;
365	struct kvmppc_vpa *vpap;
366
367	tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
368	if (!tvcpu)
369		return H_PARAMETER;
370
371	subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
372	if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
373	    subfunc == H_VPA_REG_SLB) {
374		/* Registering new area - address must be cache-line aligned */
375		if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
376			return H_PARAMETER;
377
378		/* convert logical addr to kernel addr and read length */
379		va = kvmppc_pin_guest_page(kvm, vpa, &nb);
380		if (va == NULL)
381			return H_PARAMETER;
382		if (subfunc == H_VPA_REG_VPA)
383			len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
384		else
385			len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
386		kvmppc_unpin_guest_page(kvm, va, vpa, false);
387
388		/* Check length */
389		if (len > nb || len < sizeof(struct reg_vpa))
390			return H_PARAMETER;
391	} else {
392		vpa = 0;
393		len = 0;
394	}
395
396	err = H_PARAMETER;
397	vpap = NULL;
398	spin_lock(&tvcpu->arch.vpa_update_lock);
399
400	switch (subfunc) {
401	case H_VPA_REG_VPA:		/* register VPA */
402		if (len < sizeof(struct lppaca))
403			break;
404		vpap = &tvcpu->arch.vpa;
405		err = 0;
406		break;
407
408	case H_VPA_REG_DTL:		/* register DTL */
409		if (len < sizeof(struct dtl_entry))
410			break;
411		len -= len % sizeof(struct dtl_entry);
412
413		/* Check that they have previously registered a VPA */
414		err = H_RESOURCE;
415		if (!vpa_is_registered(&tvcpu->arch.vpa))
416			break;
417
418		vpap = &tvcpu->arch.dtl;
419		err = 0;
420		break;
421
422	case H_VPA_REG_SLB:		/* register SLB shadow buffer */
423		/* Check that they have previously registered a VPA */
424		err = H_RESOURCE;
425		if (!vpa_is_registered(&tvcpu->arch.vpa))
426			break;
427
428		vpap = &tvcpu->arch.slb_shadow;
429		err = 0;
430		break;
431
432	case H_VPA_DEREG_VPA:		/* deregister VPA */
433		/* Check they don't still have a DTL or SLB buf registered */
434		err = H_RESOURCE;
435		if (vpa_is_registered(&tvcpu->arch.dtl) ||
436		    vpa_is_registered(&tvcpu->arch.slb_shadow))
437			break;
438
439		vpap = &tvcpu->arch.vpa;
440		err = 0;
441		break;
442
443	case H_VPA_DEREG_DTL:		/* deregister DTL */
444		vpap = &tvcpu->arch.dtl;
445		err = 0;
446		break;
447
448	case H_VPA_DEREG_SLB:		/* deregister SLB shadow buffer */
449		vpap = &tvcpu->arch.slb_shadow;
450		err = 0;
451		break;
452	}
453
454	if (vpap) {
455		vpap->next_gpa = vpa;
456		vpap->len = len;
457		vpap->update_pending = 1;
458	}
459
460	spin_unlock(&tvcpu->arch.vpa_update_lock);
461
462	return err;
463}
464
465static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
466{
467	struct kvm *kvm = vcpu->kvm;
468	void *va;
469	unsigned long nb;
470	unsigned long gpa;
471
472	/*
473	 * We need to pin the page pointed to by vpap->next_gpa,
474	 * but we can't call kvmppc_pin_guest_page under the lock
475	 * as it does get_user_pages() and down_read().  So we
476	 * have to drop the lock, pin the page, then get the lock
477	 * again and check that a new area didn't get registered
478	 * in the meantime.
479	 */
480	for (;;) {
481		gpa = vpap->next_gpa;
482		spin_unlock(&vcpu->arch.vpa_update_lock);
483		va = NULL;
484		nb = 0;
485		if (gpa)
486			va = kvmppc_pin_guest_page(kvm, gpa, &nb);
487		spin_lock(&vcpu->arch.vpa_update_lock);
488		if (gpa == vpap->next_gpa)
489			break;
490		/* sigh... unpin that one and try again */
491		if (va)
492			kvmppc_unpin_guest_page(kvm, va, gpa, false);
493	}
494
495	vpap->update_pending = 0;
496	if (va && nb < vpap->len) {
497		/*
498		 * If it's now too short, it must be that userspace
499		 * has changed the mappings underlying guest memory,
500		 * so unregister the region.
501		 */
502		kvmppc_unpin_guest_page(kvm, va, gpa, false);
503		va = NULL;
504	}
505	if (vpap->pinned_addr)
506		kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
507					vpap->dirty);
508	vpap->gpa = gpa;
509	vpap->pinned_addr = va;
510	vpap->dirty = false;
511	if (va)
512		vpap->pinned_end = va + vpap->len;
513}
514
515static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
516{
517	if (!(vcpu->arch.vpa.update_pending ||
518	      vcpu->arch.slb_shadow.update_pending ||
519	      vcpu->arch.dtl.update_pending))
520		return;
521
522	spin_lock(&vcpu->arch.vpa_update_lock);
523	if (vcpu->arch.vpa.update_pending) {
524		kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
525		if (vcpu->arch.vpa.pinned_addr)
526			init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
527	}
528	if (vcpu->arch.dtl.update_pending) {
529		kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
530		vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
531		vcpu->arch.dtl_index = 0;
532	}
533	if (vcpu->arch.slb_shadow.update_pending)
534		kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
535	spin_unlock(&vcpu->arch.vpa_update_lock);
536}
537
538/*
539 * Return the accumulated stolen time for the vcore up until `now'.
540 * The caller should hold the vcore lock.
541 */
542static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
543{
544	u64 p;
545	unsigned long flags;
546
547	spin_lock_irqsave(&vc->stoltb_lock, flags);
548	p = vc->stolen_tb;
549	if (vc->vcore_state != VCORE_INACTIVE &&
550	    vc->preempt_tb != TB_NIL)
551		p += now - vc->preempt_tb;
552	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
553	return p;
554}
555
556static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
557				    struct kvmppc_vcore *vc)
558{
559	struct dtl_entry *dt;
560	struct lppaca *vpa;
561	unsigned long stolen;
562	unsigned long core_stolen;
563	u64 now;
564
565	dt = vcpu->arch.dtl_ptr;
566	vpa = vcpu->arch.vpa.pinned_addr;
567	now = mftb();
568	core_stolen = vcore_stolen_time(vc, now);
569	stolen = core_stolen - vcpu->arch.stolen_logged;
570	vcpu->arch.stolen_logged = core_stolen;
571	spin_lock_irq(&vcpu->arch.tbacct_lock);
572	stolen += vcpu->arch.busy_stolen;
573	vcpu->arch.busy_stolen = 0;
574	spin_unlock_irq(&vcpu->arch.tbacct_lock);
575	if (!dt || !vpa)
576		return;
577	memset(dt, 0, sizeof(struct dtl_entry));
578	dt->dispatch_reason = 7;
579	dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
580	dt->timebase = cpu_to_be64(now + vc->tb_offset);
581	dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
582	dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
583	dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
584	++dt;
585	if (dt == vcpu->arch.dtl.pinned_end)
586		dt = vcpu->arch.dtl.pinned_addr;
587	vcpu->arch.dtl_ptr = dt;
588	/* order writing *dt vs. writing vpa->dtl_idx */
589	smp_wmb();
590	vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
591	vcpu->arch.dtl.dirty = true;
592}
593
594static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
595{
596	if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
597		return true;
598	if ((!vcpu->arch.vcore->arch_compat) &&
599	    cpu_has_feature(CPU_FTR_ARCH_207S))
600		return true;
601	return false;
602}
603
604static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
605			     unsigned long resource, unsigned long value1,
606			     unsigned long value2)
607{
608	switch (resource) {
609	case H_SET_MODE_RESOURCE_SET_CIABR:
610		if (!kvmppc_power8_compatible(vcpu))
611			return H_P2;
612		if (value2)
613			return H_P4;
614		if (mflags)
615			return H_UNSUPPORTED_FLAG_START;
616		/* Guests can't breakpoint the hypervisor */
617		if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
618			return H_P3;
619		vcpu->arch.ciabr  = value1;
620		return H_SUCCESS;
621	case H_SET_MODE_RESOURCE_SET_DAWR:
622		if (!kvmppc_power8_compatible(vcpu))
623			return H_P2;
624		if (mflags)
625			return H_UNSUPPORTED_FLAG_START;
626		if (value2 & DABRX_HYP)
627			return H_P4;
628		vcpu->arch.dawr  = value1;
629		vcpu->arch.dawrx = value2;
630		return H_SUCCESS;
631	default:
632		return H_TOO_HARD;
633	}
634}
635
636static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
637{
638	struct kvmppc_vcore *vcore = target->arch.vcore;
639
640	/*
641	 * We expect to have been called by the real mode handler
642	 * (kvmppc_rm_h_confer()) which would have directly returned
643	 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
644	 * have useful work to do and should not confer) so we don't
645	 * recheck that here.
646	 */
647
648	spin_lock(&vcore->lock);
649	if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
650	    vcore->vcore_state != VCORE_INACTIVE)
651		target = vcore->runner;
652	spin_unlock(&vcore->lock);
653
654	return kvm_vcpu_yield_to(target);
655}
656
657static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
658{
659	int yield_count = 0;
660	struct lppaca *lppaca;
661
662	spin_lock(&vcpu->arch.vpa_update_lock);
663	lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
664	if (lppaca)
665		yield_count = be32_to_cpu(lppaca->yield_count);
666	spin_unlock(&vcpu->arch.vpa_update_lock);
667	return yield_count;
668}
669
670int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
671{
672	unsigned long req = kvmppc_get_gpr(vcpu, 3);
673	unsigned long target, ret = H_SUCCESS;
674	int yield_count;
675	struct kvm_vcpu *tvcpu;
676	int idx, rc;
677
678	if (req <= MAX_HCALL_OPCODE &&
679	    !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
680		return RESUME_HOST;
681
682	switch (req) {
683	case H_CEDE:
684		break;
685	case H_PROD:
686		target = kvmppc_get_gpr(vcpu, 4);
687		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
688		if (!tvcpu) {
689			ret = H_PARAMETER;
690			break;
691		}
692		tvcpu->arch.prodded = 1;
693		smp_mb();
694		if (vcpu->arch.ceded) {
695			if (waitqueue_active(&vcpu->wq)) {
696				wake_up_interruptible(&vcpu->wq);
697				vcpu->stat.halt_wakeup++;
698			}
699		}
700		break;
701	case H_CONFER:
702		target = kvmppc_get_gpr(vcpu, 4);
703		if (target == -1)
704			break;
705		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
706		if (!tvcpu) {
707			ret = H_PARAMETER;
708			break;
709		}
710		yield_count = kvmppc_get_gpr(vcpu, 5);
711		if (kvmppc_get_yield_count(tvcpu) != yield_count)
712			break;
713		kvm_arch_vcpu_yield_to(tvcpu);
714		break;
715	case H_REGISTER_VPA:
716		ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
717					kvmppc_get_gpr(vcpu, 5),
718					kvmppc_get_gpr(vcpu, 6));
719		break;
720	case H_RTAS:
721		if (list_empty(&vcpu->kvm->arch.rtas_tokens))
722			return RESUME_HOST;
723
724		idx = srcu_read_lock(&vcpu->kvm->srcu);
725		rc = kvmppc_rtas_hcall(vcpu);
726		srcu_read_unlock(&vcpu->kvm->srcu, idx);
727
728		if (rc == -ENOENT)
729			return RESUME_HOST;
730		else if (rc == 0)
731			break;
732
733		/* Send the error out to userspace via KVM_RUN */
734		return rc;
735	case H_LOGICAL_CI_LOAD:
736		ret = kvmppc_h_logical_ci_load(vcpu);
737		if (ret == H_TOO_HARD)
738			return RESUME_HOST;
739		break;
740	case H_LOGICAL_CI_STORE:
741		ret = kvmppc_h_logical_ci_store(vcpu);
742		if (ret == H_TOO_HARD)
743			return RESUME_HOST;
744		break;
745	case H_SET_MODE:
746		ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
747					kvmppc_get_gpr(vcpu, 5),
748					kvmppc_get_gpr(vcpu, 6),
749					kvmppc_get_gpr(vcpu, 7));
750		if (ret == H_TOO_HARD)
751			return RESUME_HOST;
752		break;
753	case H_XIRR:
754	case H_CPPR:
755	case H_EOI:
756	case H_IPI:
757	case H_IPOLL:
758	case H_XIRR_X:
759		if (kvmppc_xics_enabled(vcpu)) {
760			ret = kvmppc_xics_hcall(vcpu, req);
761			break;
762		} /* fallthrough */
763	default:
764		return RESUME_HOST;
765	}
766	kvmppc_set_gpr(vcpu, 3, ret);
767	vcpu->arch.hcall_needed = 0;
768	return RESUME_GUEST;
769}
770
771static int kvmppc_hcall_impl_hv(unsigned long cmd)
772{
773	switch (cmd) {
774	case H_CEDE:
775	case H_PROD:
776	case H_CONFER:
777	case H_REGISTER_VPA:
778	case H_SET_MODE:
779	case H_LOGICAL_CI_LOAD:
780	case H_LOGICAL_CI_STORE:
781#ifdef CONFIG_KVM_XICS
782	case H_XIRR:
783	case H_CPPR:
784	case H_EOI:
785	case H_IPI:
786	case H_IPOLL:
787	case H_XIRR_X:
788#endif
789		return 1;
790	}
791
792	/* See if it's in the real-mode table */
793	return kvmppc_hcall_impl_hv_realmode(cmd);
794}
795
796static int kvmppc_emulate_debug_inst(struct kvm_run *run,
797					struct kvm_vcpu *vcpu)
798{
799	u32 last_inst;
800
801	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
802					EMULATE_DONE) {
803		/*
804		 * Fetch failed, so return to guest and
805		 * try executing it again.
806		 */
807		return RESUME_GUEST;
808	}
809
810	if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
811		run->exit_reason = KVM_EXIT_DEBUG;
812		run->debug.arch.address = kvmppc_get_pc(vcpu);
813		return RESUME_HOST;
814	} else {
815		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
816		return RESUME_GUEST;
817	}
818}
819
820static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
821				 struct task_struct *tsk)
822{
823	int r = RESUME_HOST;
824
825	vcpu->stat.sum_exits++;
826
827	run->exit_reason = KVM_EXIT_UNKNOWN;
828	run->ready_for_interrupt_injection = 1;
829	switch (vcpu->arch.trap) {
830	/* We're good on these - the host merely wanted to get our attention */
831	case BOOK3S_INTERRUPT_HV_DECREMENTER:
832		vcpu->stat.dec_exits++;
833		r = RESUME_GUEST;
834		break;
835	case BOOK3S_INTERRUPT_EXTERNAL:
836	case BOOK3S_INTERRUPT_H_DOORBELL:
837		vcpu->stat.ext_intr_exits++;
838		r = RESUME_GUEST;
839		break;
840	/* HMI is hypervisor interrupt and host has handled it. Resume guest.*/
841	case BOOK3S_INTERRUPT_HMI:
842	case BOOK3S_INTERRUPT_PERFMON:
843		r = RESUME_GUEST;
844		break;
845	case BOOK3S_INTERRUPT_MACHINE_CHECK:
846		/*
847		 * Deliver a machine check interrupt to the guest.
848		 * We have to do this, even if the host has handled the
849		 * machine check, because machine checks use SRR0/1 and
850		 * the interrupt might have trashed guest state in them.
851		 */
852		kvmppc_book3s_queue_irqprio(vcpu,
853					    BOOK3S_INTERRUPT_MACHINE_CHECK);
854		r = RESUME_GUEST;
855		break;
856	case BOOK3S_INTERRUPT_PROGRAM:
857	{
858		ulong flags;
859		/*
860		 * Normally program interrupts are delivered directly
861		 * to the guest by the hardware, but we can get here
862		 * as a result of a hypervisor emulation interrupt
863		 * (e40) getting turned into a 700 by BML RTAS.
864		 */
865		flags = vcpu->arch.shregs.msr & 0x1f0000ull;
866		kvmppc_core_queue_program(vcpu, flags);
867		r = RESUME_GUEST;
868		break;
869	}
870	case BOOK3S_INTERRUPT_SYSCALL:
871	{
872		/* hcall - punt to userspace */
873		int i;
874
875		/* hypercall with MSR_PR has already been handled in rmode,
876		 * and never reaches here.
877		 */
878
879		run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
880		for (i = 0; i < 9; ++i)
881			run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
882		run->exit_reason = KVM_EXIT_PAPR_HCALL;
883		vcpu->arch.hcall_needed = 1;
884		r = RESUME_HOST;
885		break;
886	}
887	/*
888	 * We get these next two if the guest accesses a page which it thinks
889	 * it has mapped but which is not actually present, either because
890	 * it is for an emulated I/O device or because the corresonding
891	 * host page has been paged out.  Any other HDSI/HISI interrupts
892	 * have been handled already.
893	 */
894	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
895		r = RESUME_PAGE_FAULT;
896		break;
897	case BOOK3S_INTERRUPT_H_INST_STORAGE:
898		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
899		vcpu->arch.fault_dsisr = 0;
900		r = RESUME_PAGE_FAULT;
901		break;
902	/*
903	 * This occurs if the guest executes an illegal instruction.
904	 * If the guest debug is disabled, generate a program interrupt
905	 * to the guest. If guest debug is enabled, we need to check
906	 * whether the instruction is a software breakpoint instruction.
907	 * Accordingly return to Guest or Host.
908	 */
909	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
910		if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
911			vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
912				swab32(vcpu->arch.emul_inst) :
913				vcpu->arch.emul_inst;
914		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
915			r = kvmppc_emulate_debug_inst(run, vcpu);
916		} else {
917			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
918			r = RESUME_GUEST;
919		}
920		break;
921	/*
922	 * This occurs if the guest (kernel or userspace), does something that
923	 * is prohibited by HFSCR.  We just generate a program interrupt to
924	 * the guest.
925	 */
926	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
927		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
928		r = RESUME_GUEST;
929		break;
930	default:
931		kvmppc_dump_regs(vcpu);
932		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
933			vcpu->arch.trap, kvmppc_get_pc(vcpu),
934			vcpu->arch.shregs.msr);
935		run->hw.hardware_exit_reason = vcpu->arch.trap;
936		r = RESUME_HOST;
937		break;
938	}
939
940	return r;
941}
942
943static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
944					    struct kvm_sregs *sregs)
945{
946	int i;
947
948	memset(sregs, 0, sizeof(struct kvm_sregs));
949	sregs->pvr = vcpu->arch.pvr;
950	for (i = 0; i < vcpu->arch.slb_max; i++) {
951		sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
952		sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
953	}
954
955	return 0;
956}
957
958static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
959					    struct kvm_sregs *sregs)
960{
961	int i, j;
962
963	/* Only accept the same PVR as the host's, since we can't spoof it */
964	if (sregs->pvr != vcpu->arch.pvr)
965		return -EINVAL;
966
967	j = 0;
968	for (i = 0; i < vcpu->arch.slb_nr; i++) {
969		if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
970			vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
971			vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
972			++j;
973		}
974	}
975	vcpu->arch.slb_max = j;
976
977	return 0;
978}
979
980static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
981		bool preserve_top32)
982{
983	struct kvm *kvm = vcpu->kvm;
984	struct kvmppc_vcore *vc = vcpu->arch.vcore;
985	u64 mask;
986
987	mutex_lock(&kvm->lock);
988	spin_lock(&vc->lock);
989	/*
990	 * If ILE (interrupt little-endian) has changed, update the
991	 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
992	 */
993	if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
994		struct kvm_vcpu *vcpu;
995		int i;
996
997		kvm_for_each_vcpu(i, vcpu, kvm) {
998			if (vcpu->arch.vcore != vc)
999				continue;
1000			if (new_lpcr & LPCR_ILE)
1001				vcpu->arch.intr_msr |= MSR_LE;
1002			else
1003				vcpu->arch.intr_msr &= ~MSR_LE;
1004		}
1005	}
1006
1007	/*
1008	 * Userspace can only modify DPFD (default prefetch depth),
1009	 * ILE (interrupt little-endian) and TC (translation control).
1010	 * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
1011	 */
1012	mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1013	if (cpu_has_feature(CPU_FTR_ARCH_207S))
1014		mask |= LPCR_AIL;
1015
1016	/* Broken 32-bit version of LPCR must not clear top bits */
1017	if (preserve_top32)
1018		mask &= 0xFFFFFFFF;
1019	vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1020	spin_unlock(&vc->lock);
1021	mutex_unlock(&kvm->lock);
1022}
1023
1024static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1025				 union kvmppc_one_reg *val)
1026{
1027	int r = 0;
1028	long int i;
1029
1030	switch (id) {
1031	case KVM_REG_PPC_DEBUG_INST:
1032		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1033		break;
1034	case KVM_REG_PPC_HIOR:
1035		*val = get_reg_val(id, 0);
1036		break;
1037	case KVM_REG_PPC_DABR:
1038		*val = get_reg_val(id, vcpu->arch.dabr);
1039		break;
1040	case KVM_REG_PPC_DABRX:
1041		*val = get_reg_val(id, vcpu->arch.dabrx);
1042		break;
1043	case KVM_REG_PPC_DSCR:
1044		*val = get_reg_val(id, vcpu->arch.dscr);
1045		break;
1046	case KVM_REG_PPC_PURR:
1047		*val = get_reg_val(id, vcpu->arch.purr);
1048		break;
1049	case KVM_REG_PPC_SPURR:
1050		*val = get_reg_val(id, vcpu->arch.spurr);
1051		break;
1052	case KVM_REG_PPC_AMR:
1053		*val = get_reg_val(id, vcpu->arch.amr);
1054		break;
1055	case KVM_REG_PPC_UAMOR:
1056		*val = get_reg_val(id, vcpu->arch.uamor);
1057		break;
1058	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1059		i = id - KVM_REG_PPC_MMCR0;
1060		*val = get_reg_val(id, vcpu->arch.mmcr[i]);
1061		break;
1062	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1063		i = id - KVM_REG_PPC_PMC1;
1064		*val = get_reg_val(id, vcpu->arch.pmc[i]);
1065		break;
1066	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1067		i = id - KVM_REG_PPC_SPMC1;
1068		*val = get_reg_val(id, vcpu->arch.spmc[i]);
1069		break;
1070	case KVM_REG_PPC_SIAR:
1071		*val = get_reg_val(id, vcpu->arch.siar);
1072		break;
1073	case KVM_REG_PPC_SDAR:
1074		*val = get_reg_val(id, vcpu->arch.sdar);
1075		break;
1076	case KVM_REG_PPC_SIER:
1077		*val = get_reg_val(id, vcpu->arch.sier);
1078		break;
1079	case KVM_REG_PPC_IAMR:
1080		*val = get_reg_val(id, vcpu->arch.iamr);
1081		break;
1082	case KVM_REG_PPC_PSPB:
1083		*val = get_reg_val(id, vcpu->arch.pspb);
1084		break;
1085	case KVM_REG_PPC_DPDES:
1086		*val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1087		break;
1088	case KVM_REG_PPC_DAWR:
1089		*val = get_reg_val(id, vcpu->arch.dawr);
1090		break;
1091	case KVM_REG_PPC_DAWRX:
1092		*val = get_reg_val(id, vcpu->arch.dawrx);
1093		break;
1094	case KVM_REG_PPC_CIABR:
1095		*val = get_reg_val(id, vcpu->arch.ciabr);
1096		break;
1097	case KVM_REG_PPC_CSIGR:
1098		*val = get_reg_val(id, vcpu->arch.csigr);
1099		break;
1100	case KVM_REG_PPC_TACR:
1101		*val = get_reg_val(id, vcpu->arch.tacr);
1102		break;
1103	case KVM_REG_PPC_TCSCR:
1104		*val = get_reg_val(id, vcpu->arch.tcscr);
1105		break;
1106	case KVM_REG_PPC_PID:
1107		*val = get_reg_val(id, vcpu->arch.pid);
1108		break;
1109	case KVM_REG_PPC_ACOP:
1110		*val = get_reg_val(id, vcpu->arch.acop);
1111		break;
1112	case KVM_REG_PPC_WORT:
1113		*val = get_reg_val(id, vcpu->arch.wort);
1114		break;
1115	case KVM_REG_PPC_VPA_ADDR:
1116		spin_lock(&vcpu->arch.vpa_update_lock);
1117		*val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1118		spin_unlock(&vcpu->arch.vpa_update_lock);
1119		break;
1120	case KVM_REG_PPC_VPA_SLB:
1121		spin_lock(&vcpu->arch.vpa_update_lock);
1122		val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1123		val->vpaval.length = vcpu->arch.slb_shadow.len;
1124		spin_unlock(&vcpu->arch.vpa_update_lock);
1125		break;
1126	case KVM_REG_PPC_VPA_DTL:
1127		spin_lock(&vcpu->arch.vpa_update_lock);
1128		val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1129		val->vpaval.length = vcpu->arch.dtl.len;
1130		spin_unlock(&vcpu->arch.vpa_update_lock);
1131		break;
1132	case KVM_REG_PPC_TB_OFFSET:
1133		*val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1134		break;
1135	case KVM_REG_PPC_LPCR:
1136	case KVM_REG_PPC_LPCR_64:
1137		*val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1138		break;
1139	case KVM_REG_PPC_PPR:
1140		*val = get_reg_val(id, vcpu->arch.ppr);
1141		break;
1142#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1143	case KVM_REG_PPC_TFHAR:
1144		*val = get_reg_val(id, vcpu->arch.tfhar);
1145		break;
1146	case KVM_REG_PPC_TFIAR:
1147		*val = get_reg_val(id, vcpu->arch.tfiar);
1148		break;
1149	case KVM_REG_PPC_TEXASR:
1150		*val = get_reg_val(id, vcpu->arch.texasr);
1151		break;
1152	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1153		i = id - KVM_REG_PPC_TM_GPR0;
1154		*val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1155		break;
1156	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1157	{
1158		int j;
1159		i = id - KVM_REG_PPC_TM_VSR0;
1160		if (i < 32)
1161			for (j = 0; j < TS_FPRWIDTH; j++)
1162				val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1163		else {
1164			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1165				val->vval = vcpu->arch.vr_tm.vr[i-32];
1166			else
1167				r = -ENXIO;
1168		}
1169		break;
1170	}
1171	case KVM_REG_PPC_TM_CR:
1172		*val = get_reg_val(id, vcpu->arch.cr_tm);
1173		break;
1174	case KVM_REG_PPC_TM_LR:
1175		*val = get_reg_val(id, vcpu->arch.lr_tm);
1176		break;
1177	case KVM_REG_PPC_TM_CTR:
1178		*val = get_reg_val(id, vcpu->arch.ctr_tm);
1179		break;
1180	case KVM_REG_PPC_TM_FPSCR:
1181		*val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1182		break;
1183	case KVM_REG_PPC_TM_AMR:
1184		*val = get_reg_val(id, vcpu->arch.amr_tm);
1185		break;
1186	case KVM_REG_PPC_TM_PPR:
1187		*val = get_reg_val(id, vcpu->arch.ppr_tm);
1188		break;
1189	case KVM_REG_PPC_TM_VRSAVE:
1190		*val = get_reg_val(id, vcpu->arch.vrsave_tm);
1191		break;
1192	case KVM_REG_PPC_TM_VSCR:
1193		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1194			*val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1195		else
1196			r = -ENXIO;
1197		break;
1198	case KVM_REG_PPC_TM_DSCR:
1199		*val = get_reg_val(id, vcpu->arch.dscr_tm);
1200		break;
1201	case KVM_REG_PPC_TM_TAR:
1202		*val = get_reg_val(id, vcpu->arch.tar_tm);
1203		break;
1204#endif
1205	case KVM_REG_PPC_ARCH_COMPAT:
1206		*val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1207		break;
1208	default:
1209		r = -EINVAL;
1210		break;
1211	}
1212
1213	return r;
1214}
1215
1216static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1217				 union kvmppc_one_reg *val)
1218{
1219	int r = 0;
1220	long int i;
1221	unsigned long addr, len;
1222
1223	switch (id) {
1224	case KVM_REG_PPC_HIOR:
1225		/* Only allow this to be set to zero */
1226		if (set_reg_val(id, *val))
1227			r = -EINVAL;
1228		break;
1229	case KVM_REG_PPC_DABR:
1230		vcpu->arch.dabr = set_reg_val(id, *val);
1231		break;
1232	case KVM_REG_PPC_DABRX:
1233		vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1234		break;
1235	case KVM_REG_PPC_DSCR:
1236		vcpu->arch.dscr = set_reg_val(id, *val);
1237		break;
1238	case KVM_REG_PPC_PURR:
1239		vcpu->arch.purr = set_reg_val(id, *val);
1240		break;
1241	case KVM_REG_PPC_SPURR:
1242		vcpu->arch.spurr = set_reg_val(id, *val);
1243		break;
1244	case KVM_REG_PPC_AMR:
1245		vcpu->arch.amr = set_reg_val(id, *val);
1246		break;
1247	case KVM_REG_PPC_UAMOR:
1248		vcpu->arch.uamor = set_reg_val(id, *val);
1249		break;
1250	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1251		i = id - KVM_REG_PPC_MMCR0;
1252		vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1253		break;
1254	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1255		i = id - KVM_REG_PPC_PMC1;
1256		vcpu->arch.pmc[i] = set_reg_val(id, *val);
1257		break;
1258	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1259		i = id - KVM_REG_PPC_SPMC1;
1260		vcpu->arch.spmc[i] = set_reg_val(id, *val);
1261		break;
1262	case KVM_REG_PPC_SIAR:
1263		vcpu->arch.siar = set_reg_val(id, *val);
1264		break;
1265	case KVM_REG_PPC_SDAR:
1266		vcpu->arch.sdar = set_reg_val(id, *val);
1267		break;
1268	case KVM_REG_PPC_SIER:
1269		vcpu->arch.sier = set_reg_val(id, *val);
1270		break;
1271	case KVM_REG_PPC_IAMR:
1272		vcpu->arch.iamr = set_reg_val(id, *val);
1273		break;
1274	case KVM_REG_PPC_PSPB:
1275		vcpu->arch.pspb = set_reg_val(id, *val);
1276		break;
1277	case KVM_REG_PPC_DPDES:
1278		vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1279		break;
1280	case KVM_REG_PPC_DAWR:
1281		vcpu->arch.dawr = set_reg_val(id, *val);
1282		break;
1283	case KVM_REG_PPC_DAWRX:
1284		vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1285		break;
1286	case KVM_REG_PPC_CIABR:
1287		vcpu->arch.ciabr = set_reg_val(id, *val);
1288		/* Don't allow setting breakpoints in hypervisor code */
1289		if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1290			vcpu->arch.ciabr &= ~CIABR_PRIV;	/* disable */
1291		break;
1292	case KVM_REG_PPC_CSIGR:
1293		vcpu->arch.csigr = set_reg_val(id, *val);
1294		break;
1295	case KVM_REG_PPC_TACR:
1296		vcpu->arch.tacr = set_reg_val(id, *val);
1297		break;
1298	case KVM_REG_PPC_TCSCR:
1299		vcpu->arch.tcscr = set_reg_val(id, *val);
1300		break;
1301	case KVM_REG_PPC_PID:
1302		vcpu->arch.pid = set_reg_val(id, *val);
1303		break;
1304	case KVM_REG_PPC_ACOP:
1305		vcpu->arch.acop = set_reg_val(id, *val);
1306		break;
1307	case KVM_REG_PPC_WORT:
1308		vcpu->arch.wort = set_reg_val(id, *val);
1309		break;
1310	case KVM_REG_PPC_VPA_ADDR:
1311		addr = set_reg_val(id, *val);
1312		r = -EINVAL;
1313		if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1314			      vcpu->arch.dtl.next_gpa))
1315			break;
1316		r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1317		break;
1318	case KVM_REG_PPC_VPA_SLB:
1319		addr = val->vpaval.addr;
1320		len = val->vpaval.length;
1321		r = -EINVAL;
1322		if (addr && !vcpu->arch.vpa.next_gpa)
1323			break;
1324		r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1325		break;
1326	case KVM_REG_PPC_VPA_DTL:
1327		addr = val->vpaval.addr;
1328		len = val->vpaval.length;
1329		r = -EINVAL;
1330		if (addr && (len < sizeof(struct dtl_entry) ||
1331			     !vcpu->arch.vpa.next_gpa))
1332			break;
1333		len -= len % sizeof(struct dtl_entry);
1334		r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1335		break;
1336	case KVM_REG_PPC_TB_OFFSET:
1337		/* round up to multiple of 2^24 */
1338		vcpu->arch.vcore->tb_offset =
1339			ALIGN(set_reg_val(id, *val), 1UL << 24);
1340		break;
1341	case KVM_REG_PPC_LPCR:
1342		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1343		break;
1344	case KVM_REG_PPC_LPCR_64:
1345		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1346		break;
1347	case KVM_REG_PPC_PPR:
1348		vcpu->arch.ppr = set_reg_val(id, *val);
1349		break;
1350#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1351	case KVM_REG_PPC_TFHAR:
1352		vcpu->arch.tfhar = set_reg_val(id, *val);
1353		break;
1354	case KVM_REG_PPC_TFIAR:
1355		vcpu->arch.tfiar = set_reg_val(id, *val);
1356		break;
1357	case KVM_REG_PPC_TEXASR:
1358		vcpu->arch.texasr = set_reg_val(id, *val);
1359		break;
1360	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1361		i = id - KVM_REG_PPC_TM_GPR0;
1362		vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1363		break;
1364	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1365	{
1366		int j;
1367		i = id - KVM_REG_PPC_TM_VSR0;
1368		if (i < 32)
1369			for (j = 0; j < TS_FPRWIDTH; j++)
1370				vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1371		else
1372			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1373				vcpu->arch.vr_tm.vr[i-32] = val->vval;
1374			else
1375				r = -ENXIO;
1376		break;
1377	}
1378	case KVM_REG_PPC_TM_CR:
1379		vcpu->arch.cr_tm = set_reg_val(id, *val);
1380		break;
1381	case KVM_REG_PPC_TM_LR:
1382		vcpu->arch.lr_tm = set_reg_val(id, *val);
1383		break;
1384	case KVM_REG_PPC_TM_CTR:
1385		vcpu->arch.ctr_tm = set_reg_val(id, *val);
1386		break;
1387	case KVM_REG_PPC_TM_FPSCR:
1388		vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1389		break;
1390	case KVM_REG_PPC_TM_AMR:
1391		vcpu->arch.amr_tm = set_reg_val(id, *val);
1392		break;
1393	case KVM_REG_PPC_TM_PPR:
1394		vcpu->arch.ppr_tm = set_reg_val(id, *val);
1395		break;
1396	case KVM_REG_PPC_TM_VRSAVE:
1397		vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1398		break;
1399	case KVM_REG_PPC_TM_VSCR:
1400		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1401			vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1402		else
1403			r = - ENXIO;
1404		break;
1405	case KVM_REG_PPC_TM_DSCR:
1406		vcpu->arch.dscr_tm = set_reg_val(id, *val);
1407		break;
1408	case KVM_REG_PPC_TM_TAR:
1409		vcpu->arch.tar_tm = set_reg_val(id, *val);
1410		break;
1411#endif
1412	case KVM_REG_PPC_ARCH_COMPAT:
1413		r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1414		break;
1415	default:
1416		r = -EINVAL;
1417		break;
1418	}
1419
1420	return r;
1421}
1422
1423static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1424{
1425	struct kvmppc_vcore *vcore;
1426
1427	vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1428
1429	if (vcore == NULL)
1430		return NULL;
1431
1432	INIT_LIST_HEAD(&vcore->runnable_threads);
1433	spin_lock_init(&vcore->lock);
1434	spin_lock_init(&vcore->stoltb_lock);
1435	init_waitqueue_head(&vcore->wq);
1436	vcore->preempt_tb = TB_NIL;
1437	vcore->lpcr = kvm->arch.lpcr;
1438	vcore->first_vcpuid = core * threads_per_subcore;
1439	vcore->kvm = kvm;
1440
1441	vcore->mpp_buffer_is_valid = false;
1442
1443	if (cpu_has_feature(CPU_FTR_ARCH_207S))
1444		vcore->mpp_buffer = (void *)__get_free_pages(
1445			GFP_KERNEL|__GFP_ZERO,
1446			MPP_BUFFER_ORDER);
1447
1448	return vcore;
1449}
1450
1451#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1452static struct debugfs_timings_element {
1453	const char *name;
1454	size_t offset;
1455} timings[] = {
1456	{"rm_entry",	offsetof(struct kvm_vcpu, arch.rm_entry)},
1457	{"rm_intr",	offsetof(struct kvm_vcpu, arch.rm_intr)},
1458	{"rm_exit",	offsetof(struct kvm_vcpu, arch.rm_exit)},
1459	{"guest",	offsetof(struct kvm_vcpu, arch.guest_time)},
1460	{"cede",	offsetof(struct kvm_vcpu, arch.cede_time)},
1461};
1462
1463#define N_TIMINGS	(sizeof(timings) / sizeof(timings[0]))
1464
1465struct debugfs_timings_state {
1466	struct kvm_vcpu	*vcpu;
1467	unsigned int	buflen;
1468	char		buf[N_TIMINGS * 100];
1469};
1470
1471static int debugfs_timings_open(struct inode *inode, struct file *file)
1472{
1473	struct kvm_vcpu *vcpu = inode->i_private;
1474	struct debugfs_timings_state *p;
1475
1476	p = kzalloc(sizeof(*p), GFP_KERNEL);
1477	if (!p)
1478		return -ENOMEM;
1479
1480	kvm_get_kvm(vcpu->kvm);
1481	p->vcpu = vcpu;
1482	file->private_data = p;
1483
1484	return nonseekable_open(inode, file);
1485}
1486
1487static int debugfs_timings_release(struct inode *inode, struct file *file)
1488{
1489	struct debugfs_timings_state *p = file->private_data;
1490
1491	kvm_put_kvm(p->vcpu->kvm);
1492	kfree(p);
1493	return 0;
1494}
1495
1496static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1497				    size_t len, loff_t *ppos)
1498{
1499	struct debugfs_timings_state *p = file->private_data;
1500	struct kvm_vcpu *vcpu = p->vcpu;
1501	char *s, *buf_end;
1502	struct kvmhv_tb_accumulator tb;
1503	u64 count;
1504	loff_t pos;
1505	ssize_t n;
1506	int i, loops;
1507	bool ok;
1508
1509	if (!p->buflen) {
1510		s = p->buf;
1511		buf_end = s + sizeof(p->buf);
1512		for (i = 0; i < N_TIMINGS; ++i) {
1513			struct kvmhv_tb_accumulator *acc;
1514
1515			acc = (struct kvmhv_tb_accumulator *)
1516				((unsigned long)vcpu + timings[i].offset);
1517			ok = false;
1518			for (loops = 0; loops < 1000; ++loops) {
1519				count = acc->seqcount;
1520				if (!(count & 1)) {
1521					smp_rmb();
1522					tb = *acc;
1523					smp_rmb();
1524					if (count == acc->seqcount) {
1525						ok = true;
1526						break;
1527					}
1528				}
1529				udelay(1);
1530			}
1531			if (!ok)
1532				snprintf(s, buf_end - s, "%s: stuck\n",
1533					timings[i].name);
1534			else
1535				snprintf(s, buf_end - s,
1536					"%s: %llu %llu %llu %llu\n",
1537					timings[i].name, count / 2,
1538					tb_to_ns(tb.tb_total),
1539					tb_to_ns(tb.tb_min),
1540					tb_to_ns(tb.tb_max));
1541			s += strlen(s);
1542		}
1543		p->buflen = s - p->buf;
1544	}
1545
1546	pos = *ppos;
1547	if (pos >= p->buflen)
1548		return 0;
1549	if (len > p->buflen - pos)
1550		len = p->buflen - pos;
1551	n = copy_to_user(buf, p->buf + pos, len);
1552	if (n) {
1553		if (n == len)
1554			return -EFAULT;
1555		len -= n;
1556	}
1557	*ppos = pos + len;
1558	return len;
1559}
1560
1561static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1562				     size_t len, loff_t *ppos)
1563{
1564	return -EACCES;
1565}
1566
1567static const struct file_operations debugfs_timings_ops = {
1568	.owner	 = THIS_MODULE,
1569	.open	 = debugfs_timings_open,
1570	.release = debugfs_timings_release,
1571	.read	 = debugfs_timings_read,
1572	.write	 = debugfs_timings_write,
1573	.llseek	 = generic_file_llseek,
1574};
1575
1576/* Create a debugfs directory for the vcpu */
1577static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1578{
1579	char buf[16];
1580	struct kvm *kvm = vcpu->kvm;
1581
1582	snprintf(buf, sizeof(buf), "vcpu%u", id);
1583	if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1584		return;
1585	vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1586	if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1587		return;
1588	vcpu->arch.debugfs_timings =
1589		debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1590				    vcpu, &debugfs_timings_ops);
1591}
1592
1593#else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1594static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1595{
1596}
1597#endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1598
1599static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1600						   unsigned int id)
1601{
1602	struct kvm_vcpu *vcpu;
1603	int err = -EINVAL;
1604	int core;
1605	struct kvmppc_vcore *vcore;
1606
1607	core = id / threads_per_subcore;
1608	if (core >= KVM_MAX_VCORES)
1609		goto out;
1610
1611	err = -ENOMEM;
1612	vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1613	if (!vcpu)
1614		goto out;
1615
1616	err = kvm_vcpu_init(vcpu, kvm, id);
1617	if (err)
1618		goto free_vcpu;
1619
1620	vcpu->arch.shared = &vcpu->arch.shregs;
1621#ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1622	/*
1623	 * The shared struct is never shared on HV,
1624	 * so we can always use host endianness
1625	 */
1626#ifdef __BIG_ENDIAN__
1627	vcpu->arch.shared_big_endian = true;
1628#else
1629	vcpu->arch.shared_big_endian = false;
1630#endif
1631#endif
1632	vcpu->arch.mmcr[0] = MMCR0_FC;
1633	vcpu->arch.ctrl = CTRL_RUNLATCH;
1634	/* default to host PVR, since we can't spoof it */
1635	kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1636	spin_lock_init(&vcpu->arch.vpa_update_lock);
1637	spin_lock_init(&vcpu->arch.tbacct_lock);
1638	vcpu->arch.busy_preempt = TB_NIL;
1639	vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1640
1641	kvmppc_mmu_book3s_hv_init(vcpu);
1642
1643	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1644
1645	init_waitqueue_head(&vcpu->arch.cpu_run);
1646
1647	mutex_lock(&kvm->lock);
1648	vcore = kvm->arch.vcores[core];
1649	if (!vcore) {
1650		vcore = kvmppc_vcore_create(kvm, core);
1651		kvm->arch.vcores[core] = vcore;
1652		kvm->arch.online_vcores++;
1653	}
1654	mutex_unlock(&kvm->lock);
1655
1656	if (!vcore)
1657		goto free_vcpu;
1658
1659	spin_lock(&vcore->lock);
1660	++vcore->num_threads;
1661	spin_unlock(&vcore->lock);
1662	vcpu->arch.vcore = vcore;
1663	vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1664
1665	vcpu->arch.cpu_type = KVM_CPU_3S_64;
1666	kvmppc_sanity_check(vcpu);
1667
1668	debugfs_vcpu_init(vcpu, id);
1669
1670	return vcpu;
1671
1672free_vcpu:
1673	kmem_cache_free(kvm_vcpu_cache, vcpu);
1674out:
1675	return ERR_PTR(err);
1676}
1677
1678static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1679{
1680	if (vpa->pinned_addr)
1681		kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1682					vpa->dirty);
1683}
1684
1685static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
1686{
1687	spin_lock(&vcpu->arch.vpa_update_lock);
1688	unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1689	unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1690	unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1691	spin_unlock(&vcpu->arch.vpa_update_lock);
1692	kvm_vcpu_uninit(vcpu);
1693	kmem_cache_free(kvm_vcpu_cache, vcpu);
1694}
1695
1696static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1697{
1698	/* Indicate we want to get back into the guest */
1699	return 1;
1700}
1701
1702static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1703{
1704	unsigned long dec_nsec, now;
1705
1706	now = get_tb();
1707	if (now > vcpu->arch.dec_expires) {
1708		/* decrementer has already gone negative */
1709		kvmppc_core_queue_dec(vcpu);
1710		kvmppc_core_prepare_to_enter(vcpu);
1711		return;
1712	}
1713	dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1714		   / tb_ticks_per_sec;
1715	hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec),
1716		      HRTIMER_MODE_REL);
1717	vcpu->arch.timer_running = 1;
1718}
1719
1720static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1721{
1722	vcpu->arch.ceded = 0;
1723	if (vcpu->arch.timer_running) {
1724		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1725		vcpu->arch.timer_running = 0;
1726	}
1727}
1728
1729extern void __kvmppc_vcore_entry(void);
1730
1731static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1732				   struct kvm_vcpu *vcpu)
1733{
1734	u64 now;
1735
1736	if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1737		return;
1738	spin_lock_irq(&vcpu->arch.tbacct_lock);
1739	now = mftb();
1740	vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1741		vcpu->arch.stolen_logged;
1742	vcpu->arch.busy_preempt = now;
1743	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1744	spin_unlock_irq(&vcpu->arch.tbacct_lock);
1745	--vc->n_runnable;
1746	list_del(&vcpu->arch.run_list);
1747}
1748
1749static int kvmppc_grab_hwthread(int cpu)
1750{
1751	struct paca_struct *tpaca;
1752	long timeout = 10000;
1753
1754	tpaca = &paca[cpu];
1755
1756	/* Ensure the thread won't go into the kernel if it wakes */
1757	tpaca->kvm_hstate.kvm_vcpu = NULL;
1758	tpaca->kvm_hstate.napping = 0;
1759	smp_wmb();
1760	tpaca->kvm_hstate.hwthread_req = 1;
1761
1762	/*
1763	 * If the thread is already executing in the kernel (e.g. handling
1764	 * a stray interrupt), wait for it to get back to nap mode.
1765	 * The smp_mb() is to ensure that our setting of hwthread_req
1766	 * is visible before we look at hwthread_state, so if this
1767	 * races with the code at system_reset_pSeries and the thread
1768	 * misses our setting of hwthread_req, we are sure to see its
1769	 * setting of hwthread_state, and vice versa.
1770	 */
1771	smp_mb();
1772	while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1773		if (--timeout <= 0) {
1774			pr_err("KVM: couldn't grab cpu %d\n", cpu);
1775			return -EBUSY;
1776		}
1777		udelay(1);
1778	}
1779	return 0;
1780}
1781
1782static void kvmppc_release_hwthread(int cpu)
1783{
1784	struct paca_struct *tpaca;
1785
1786	tpaca = &paca[cpu];
1787	tpaca->kvm_hstate.hwthread_req = 0;
1788	tpaca->kvm_hstate.kvm_vcpu = NULL;
1789}
1790
1791static void kvmppc_start_thread(struct kvm_vcpu *vcpu)
1792{
1793	int cpu;
1794	struct paca_struct *tpaca;
1795	struct kvmppc_vcore *vc = vcpu->arch.vcore;
1796
1797	if (vcpu->arch.timer_running) {
1798		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1799		vcpu->arch.timer_running = 0;
1800	}
1801	cpu = vc->pcpu + vcpu->arch.ptid;
1802	tpaca = &paca[cpu];
1803	tpaca->kvm_hstate.kvm_vcore = vc;
1804	tpaca->kvm_hstate.ptid = vcpu->arch.ptid;
1805	vcpu->cpu = vc->pcpu;
1806	/* Order stores to hstate.kvm_vcore etc. before store to kvm_vcpu */
1807	smp_wmb();
1808	tpaca->kvm_hstate.kvm_vcpu = vcpu;
1809	if (cpu != smp_processor_id())
1810		kvmppc_ipi_thread(cpu);
1811}
1812
1813static void kvmppc_wait_for_nap(void)
1814{
1815	int cpu = smp_processor_id();
1816	int i, loops;
1817
1818	for (loops = 0; loops < 1000000; ++loops) {
1819		/*
1820		 * Check if all threads are finished.
1821		 * We set the vcpu pointer when starting a thread
1822		 * and the thread clears it when finished, so we look
1823		 * for any threads that still have a non-NULL vcpu ptr.
1824		 */
1825		for (i = 1; i < threads_per_subcore; ++i)
1826			if (paca[cpu + i].kvm_hstate.kvm_vcpu)
1827				break;
1828		if (i == threads_per_subcore) {
1829			HMT_medium();
1830			return;
1831		}
1832		HMT_low();
1833	}
1834	HMT_medium();
1835	for (i = 1; i < threads_per_subcore; ++i)
1836		if (paca[cpu + i].kvm_hstate.kvm_vcpu)
1837			pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
1838}
1839
1840/*
1841 * Check that we are on thread 0 and that any other threads in
1842 * this core are off-line.  Then grab the threads so they can't
1843 * enter the kernel.
1844 */
1845static int on_primary_thread(void)
1846{
1847	int cpu = smp_processor_id();
1848	int thr;
1849
1850	/* Are we on a primary subcore? */
1851	if (cpu_thread_in_subcore(cpu))
1852		return 0;
1853
1854	thr = 0;
1855	while (++thr < threads_per_subcore)
1856		if (cpu_online(cpu + thr))
1857			return 0;
1858
1859	/* Grab all hw threads so they can't go into the kernel */
1860	for (thr = 1; thr < threads_per_subcore; ++thr) {
1861		if (kvmppc_grab_hwthread(cpu + thr)) {
1862			/* Couldn't grab one; let the others go */
1863			do {
1864				kvmppc_release_hwthread(cpu + thr);
1865			} while (--thr > 0);
1866			return 0;
1867		}
1868	}
1869	return 1;
1870}
1871
1872static void kvmppc_start_saving_l2_cache(struct kvmppc_vcore *vc)
1873{
1874	phys_addr_t phy_addr, mpp_addr;
1875
1876	phy_addr = (phys_addr_t)virt_to_phys(vc->mpp_buffer);
1877	mpp_addr = phy_addr & PPC_MPPE_ADDRESS_MASK;
1878
1879	mtspr(SPRN_MPPR, mpp_addr | PPC_MPPR_FETCH_ABORT);
1880	logmpp(mpp_addr | PPC_LOGMPP_LOG_L2);
1881
1882	vc->mpp_buffer_is_valid = true;
1883}
1884
1885static void kvmppc_start_restoring_l2_cache(const struct kvmppc_vcore *vc)
1886{
1887	phys_addr_t phy_addr, mpp_addr;
1888
1889	phy_addr = virt_to_phys(vc->mpp_buffer);
1890	mpp_addr = phy_addr & PPC_MPPE_ADDRESS_MASK;
1891
1892	/* We must abort any in-progress save operations to ensure
1893	 * the table is valid so that prefetch engine knows when to
1894	 * stop prefetching. */
1895	logmpp(mpp_addr | PPC_LOGMPP_LOG_ABORT);
1896	mtspr(SPRN_MPPR, mpp_addr | PPC_MPPR_FETCH_WHOLE_TABLE);
1897}
1898
1899static void prepare_threads(struct kvmppc_vcore *vc)
1900{
1901	struct kvm_vcpu *vcpu, *vnext;
1902
1903	list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1904				 arch.run_list) {
1905		if (signal_pending(vcpu->arch.run_task))
1906			vcpu->arch.ret = -EINTR;
1907		else if (vcpu->arch.vpa.update_pending ||
1908			 vcpu->arch.slb_shadow.update_pending ||
1909			 vcpu->arch.dtl.update_pending)
1910			vcpu->arch.ret = RESUME_GUEST;
1911		else
1912			continue;
1913		kvmppc_remove_runnable(vc, vcpu);
1914		wake_up(&vcpu->arch.cpu_run);
1915	}
1916}
1917
1918static void post_guest_process(struct kvmppc_vcore *vc)
1919{
1920	u64 now;
1921	long ret;
1922	struct kvm_vcpu *vcpu, *vnext;
1923
1924	now = get_tb();
1925	list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1926				 arch.run_list) {
1927		/* cancel pending dec exception if dec is positive */
1928		if (now < vcpu->arch.dec_expires &&
1929		    kvmppc_core_pending_dec(vcpu))
1930			kvmppc_core_dequeue_dec(vcpu);
1931
1932		trace_kvm_guest_exit(vcpu);
1933
1934		ret = RESUME_GUEST;
1935		if (vcpu->arch.trap)
1936			ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
1937						    vcpu->arch.run_task);
1938
1939		vcpu->arch.ret = ret;
1940		vcpu->arch.trap = 0;
1941
1942		if (vcpu->arch.ceded) {
1943			if (!is_kvmppc_resume_guest(ret))
1944				kvmppc_end_cede(vcpu);
1945			else
1946				kvmppc_set_timer(vcpu);
1947		}
1948		if (!is_kvmppc_resume_guest(vcpu->arch.ret)) {
1949			kvmppc_remove_runnable(vc, vcpu);
1950			wake_up(&vcpu->arch.cpu_run);
1951		}
1952	}
1953}
1954
1955/*
1956 * Run a set of guest threads on a physical core.
1957 * Called with vc->lock held.
1958 */
1959static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
1960{
1961	struct kvm_vcpu *vcpu, *vnext;
1962	int i;
1963	int srcu_idx;
1964
1965	/*
1966	 * Remove from the list any threads that have a signal pending
1967	 * or need a VPA update done
1968	 */
1969	prepare_threads(vc);
1970
1971	/* if the runner is no longer runnable, let the caller pick a new one */
1972	if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
1973		return;
1974
1975	/*
1976	 * Initialize *vc.
1977	 */
1978	vc->entry_exit_map = 0;
1979	vc->preempt_tb = TB_NIL;
1980	vc->in_guest = 0;
1981	vc->napping_threads = 0;
1982	vc->conferring_threads = 0;
1983
1984	/*
1985	 * Make sure we are running on primary threads, and that secondary
1986	 * threads are offline.  Also check if the number of threads in this
1987	 * guest are greater than the current system threads per guest.
1988	 */
1989	if ((threads_per_core > 1) &&
1990	    ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
1991		list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads,
1992					 arch.run_list) {
1993			vcpu->arch.ret = -EBUSY;
1994			kvmppc_remove_runnable(vc, vcpu);
1995			wake_up(&vcpu->arch.cpu_run);
1996		}
1997		goto out;
1998	}
1999
2000
2001	vc->pcpu = smp_processor_id();
2002	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
2003		kvmppc_start_thread(vcpu);
2004		kvmppc_create_dtl_entry(vcpu, vc);
2005		trace_kvm_guest_enter(vcpu);
2006	}
2007
2008	/* Set this explicitly in case thread 0 doesn't have a vcpu */
2009	get_paca()->kvm_hstate.kvm_vcore = vc;
2010	get_paca()->kvm_hstate.ptid = 0;
2011
2012	vc->vcore_state = VCORE_RUNNING;
2013	preempt_disable();
2014
2015	trace_kvmppc_run_core(vc, 0);
2016
2017	spin_unlock(&vc->lock);
2018
2019	kvm_guest_enter();
2020
2021	srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2022
2023	if (vc->mpp_buffer_is_valid)
2024		kvmppc_start_restoring_l2_cache(vc);
2025
2026	__kvmppc_vcore_entry();
2027
2028	spin_lock(&vc->lock);
2029
2030	if (vc->mpp_buffer)
2031		kvmppc_start_saving_l2_cache(vc);
2032
2033	/* disable sending of IPIs on virtual external irqs */
2034	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list)
2035		vcpu->cpu = -1;
2036	/* wait for secondary threads to finish writing their state to memory */
2037	kvmppc_wait_for_nap();
2038	for (i = 0; i < threads_per_subcore; ++i)
2039		kvmppc_release_hwthread(vc->pcpu + i);
2040	/* prevent other vcpu threads from doing kvmppc_start_thread() now */
2041	vc->vcore_state = VCORE_EXITING;
2042	spin_unlock(&vc->lock);
2043
2044	srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2045
2046	/* make sure updates to secondary vcpu structs are visible now */
2047	smp_mb();
2048	kvm_guest_exit();
2049
2050	preempt_enable();
2051
2052	spin_lock(&vc->lock);
2053	post_guest_process(vc);
2054
2055 out:
2056	vc->vcore_state = VCORE_INACTIVE;
2057	trace_kvmppc_run_core(vc, 1);
2058}
2059
2060/*
2061 * Wait for some other vcpu thread to execute us, and
2062 * wake us up when we need to handle something in the host.
2063 */
2064static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state)
2065{
2066	DEFINE_WAIT(wait);
2067
2068	prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
2069	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE)
2070		schedule();
2071	finish_wait(&vcpu->arch.cpu_run, &wait);
2072}
2073
2074/*
2075 * All the vcpus in this vcore are idle, so wait for a decrementer
2076 * or external interrupt to one of the vcpus.  vc->lock is held.
2077 */
2078static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
2079{
2080	struct kvm_vcpu *vcpu;
2081	int do_sleep = 1;
2082
2083	DEFINE_WAIT(wait);
2084
2085	prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
2086
2087	/*
2088	 * Check one last time for pending exceptions and ceded state after
2089	 * we put ourselves on the wait queue
2090	 */
2091	list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) {
2092		if (vcpu->arch.pending_exceptions || !vcpu->arch.ceded) {
2093			do_sleep = 0;
2094			break;
2095		}
2096	}
2097
2098	if (!do_sleep) {
2099		finish_wait(&vc->wq, &wait);
2100		return;
2101	}
2102
2103	vc->vcore_state = VCORE_SLEEPING;
2104	trace_kvmppc_vcore_blocked(vc, 0);
2105	spin_unlock(&vc->lock);
2106	schedule();
2107	finish_wait(&vc->wq, &wait);
2108	spin_lock(&vc->lock);
2109	vc->vcore_state = VCORE_INACTIVE;
2110	trace_kvmppc_vcore_blocked(vc, 1);
2111}
2112
2113static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2114{
2115	int n_ceded;
2116	struct kvmppc_vcore *vc;
2117	struct kvm_vcpu *v, *vn;
2118
2119	trace_kvmppc_run_vcpu_enter(vcpu);
2120
2121	kvm_run->exit_reason = 0;
2122	vcpu->arch.ret = RESUME_GUEST;
2123	vcpu->arch.trap = 0;
2124	kvmppc_update_vpas(vcpu);
2125
2126	/*
2127	 * Synchronize with other threads in this virtual core
2128	 */
2129	vc = vcpu->arch.vcore;
2130	spin_lock(&vc->lock);
2131	vcpu->arch.ceded = 0;
2132	vcpu->arch.run_task = current;
2133	vcpu->arch.kvm_run = kvm_run;
2134	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
2135	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
2136	vcpu->arch.busy_preempt = TB_NIL;
2137	list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads);
2138	++vc->n_runnable;
2139
2140	/*
2141	 * This happens the first time this is called for a vcpu.
2142	 * If the vcore is already running, we may be able to start
2143	 * this thread straight away and have it join in.
2144	 */
2145	if (!signal_pending(current)) {
2146		if (vc->vcore_state == VCORE_RUNNING && !VCORE_IS_EXITING(vc)) {
2147			kvmppc_create_dtl_entry(vcpu, vc);
2148			kvmppc_start_thread(vcpu);
2149			trace_kvm_guest_enter(vcpu);
2150		} else if (vc->vcore_state == VCORE_SLEEPING) {
2151			wake_up(&vc->wq);
2152		}
2153
2154	}
2155
2156	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2157	       !signal_pending(current)) {
2158		if (vc->vcore_state != VCORE_INACTIVE) {
2159			spin_unlock(&vc->lock);
2160			kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE);
2161			spin_lock(&vc->lock);
2162			continue;
2163		}
2164		list_for_each_entry_safe(v, vn, &vc->runnable_threads,
2165					 arch.run_list) {
2166			kvmppc_core_prepare_to_enter(v);
2167			if (signal_pending(v->arch.run_task)) {
2168				kvmppc_remove_runnable(vc, v);
2169				v->stat.signal_exits++;
2170				v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
2171				v->arch.ret = -EINTR;
2172				wake_up(&v->arch.cpu_run);
2173			}
2174		}
2175		if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2176			break;
2177		n_ceded = 0;
2178		list_for_each_entry(v, &vc->runnable_threads, arch.run_list) {
2179			if (!v->arch.pending_exceptions)
2180				n_ceded += v->arch.ceded;
2181			else
2182				v->arch.ceded = 0;
2183		}
2184		vc->runner = vcpu;
2185		if (n_ceded == vc->n_runnable) {
2186			kvmppc_vcore_blocked(vc);
2187		} else if (need_resched()) {
2188			vc->vcore_state = VCORE_PREEMPT;
2189			/* Let something else run */
2190			cond_resched_lock(&vc->lock);
2191			vc->vcore_state = VCORE_INACTIVE;
2192		} else {
2193			kvmppc_run_core(vc);
2194		}
2195		vc->runner = NULL;
2196	}
2197
2198	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2199	       (vc->vcore_state == VCORE_RUNNING ||
2200		vc->vcore_state == VCORE_EXITING)) {
2201		spin_unlock(&vc->lock);
2202		kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE);
2203		spin_lock(&vc->lock);
2204	}
2205
2206	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2207		kvmppc_remove_runnable(vc, vcpu);
2208		vcpu->stat.signal_exits++;
2209		kvm_run->exit_reason = KVM_EXIT_INTR;
2210		vcpu->arch.ret = -EINTR;
2211	}
2212
2213	if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
2214		/* Wake up some vcpu to run the core */
2215		v = list_first_entry(&vc->runnable_threads,
2216				     struct kvm_vcpu, arch.run_list);
2217		wake_up(&v->arch.cpu_run);
2218	}
2219
2220	trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
2221	spin_unlock(&vc->lock);
2222	return vcpu->arch.ret;
2223}
2224
2225static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
2226{
2227	int r;
2228	int srcu_idx;
2229
2230	if (!vcpu->arch.sane) {
2231		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2232		return -EINVAL;
2233	}
2234
2235	kvmppc_core_prepare_to_enter(vcpu);
2236
2237	/* No need to go into the guest when all we'll do is come back out */
2238	if (signal_pending(current)) {
2239		run->exit_reason = KVM_EXIT_INTR;
2240		return -EINTR;
2241	}
2242
2243	atomic_inc(&vcpu->kvm->arch.vcpus_running);
2244	/* Order vcpus_running vs. hpte_setup_done, see kvmppc_alloc_reset_hpt */
2245	smp_mb();
2246
2247	/* On the first time here, set up HTAB and VRMA */
2248	if (!vcpu->kvm->arch.hpte_setup_done) {
2249		r = kvmppc_hv_setup_htab_rma(vcpu);
2250		if (r)
2251			goto out;
2252	}
2253
2254	flush_fp_to_thread(current);
2255	flush_altivec_to_thread(current);
2256	flush_vsx_to_thread(current);
2257	vcpu->arch.wqp = &vcpu->arch.vcore->wq;
2258	vcpu->arch.pgdir = current->mm->pgd;
2259	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2260
2261	do {
2262		r = kvmppc_run_vcpu(run, vcpu);
2263
2264		if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
2265		    !(vcpu->arch.shregs.msr & MSR_PR)) {
2266			trace_kvm_hcall_enter(vcpu);
2267			r = kvmppc_pseries_do_hcall(vcpu);
2268			trace_kvm_hcall_exit(vcpu, r);
2269			kvmppc_core_prepare_to_enter(vcpu);
2270		} else if (r == RESUME_PAGE_FAULT) {
2271			srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2272			r = kvmppc_book3s_hv_page_fault(run, vcpu,
2273				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
2274			srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2275		}
2276	} while (is_kvmppc_resume_guest(r));
2277
2278 out:
2279	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2280	atomic_dec(&vcpu->kvm->arch.vcpus_running);
2281	return r;
2282}
2283
2284static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
2285				     int linux_psize)
2286{
2287	struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
2288
2289	if (!def->shift)
2290		return;
2291	(*sps)->page_shift = def->shift;
2292	(*sps)->slb_enc = def->sllp;
2293	(*sps)->enc[0].page_shift = def->shift;
2294	(*sps)->enc[0].pte_enc = def->penc[linux_psize];
2295	/*
2296	 * Add 16MB MPSS support if host supports it
2297	 */
2298	if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
2299		(*sps)->enc[1].page_shift = 24;
2300		(*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
2301	}
2302	(*sps)++;
2303}
2304
2305static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
2306					 struct kvm_ppc_smmu_info *info)
2307{
2308	struct kvm_ppc_one_seg_page_size *sps;
2309
2310	info->flags = KVM_PPC_PAGE_SIZES_REAL;
2311	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
2312		info->flags |= KVM_PPC_1T_SEGMENTS;
2313	info->slb_size = mmu_slb_size;
2314
2315	/* We only support these sizes for now, and no muti-size segments */
2316	sps = &info->sps[0];
2317	kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
2318	kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
2319	kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
2320
2321	return 0;
2322}
2323
2324/*
2325 * Get (and clear) the dirty memory log for a memory slot.
2326 */
2327static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
2328					 struct kvm_dirty_log *log)
2329{
2330	struct kvm_memory_slot *memslot;
2331	int r;
2332	unsigned long n;
2333
2334	mutex_lock(&kvm->slots_lock);
2335
2336	r = -EINVAL;
2337	if (log->slot >= KVM_USER_MEM_SLOTS)
2338		goto out;
2339
2340	memslot = id_to_memslot(kvm->memslots, log->slot);
2341	r = -ENOENT;
2342	if (!memslot->dirty_bitmap)
2343		goto out;
2344
2345	n = kvm_dirty_bitmap_bytes(memslot);
2346	memset(memslot->dirty_bitmap, 0, n);
2347
2348	r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
2349	if (r)
2350		goto out;
2351
2352	r = -EFAULT;
2353	if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
2354		goto out;
2355
2356	r = 0;
2357out:
2358	mutex_unlock(&kvm->slots_lock);
2359	return r;
2360}
2361
2362static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
2363					struct kvm_memory_slot *dont)
2364{
2365	if (!dont || free->arch.rmap != dont->arch.rmap) {
2366		vfree(free->arch.rmap);
2367		free->arch.rmap = NULL;
2368	}
2369}
2370
2371static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
2372					 unsigned long npages)
2373{
2374	slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
2375	if (!slot->arch.rmap)
2376		return -ENOMEM;
2377
2378	return 0;
2379}
2380
2381static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
2382					struct kvm_memory_slot *memslot,
2383					struct kvm_userspace_memory_region *mem)
2384{
2385	return 0;
2386}
2387
2388static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
2389				struct kvm_userspace_memory_region *mem,
2390				const struct kvm_memory_slot *old)
2391{
2392	unsigned long npages = mem->memory_size >> PAGE_SHIFT;
2393	struct kvm_memory_slot *memslot;
2394
2395	if (npages && old->npages) {
2396		/*
2397		 * If modifying a memslot, reset all the rmap dirty bits.
2398		 * If this is a new memslot, we don't need to do anything
2399		 * since the rmap array starts out as all zeroes,
2400		 * i.e. no pages are dirty.
2401		 */
2402		memslot = id_to_memslot(kvm->memslots, mem->slot);
2403		kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
2404	}
2405}
2406
2407/*
2408 * Update LPCR values in kvm->arch and in vcores.
2409 * Caller must hold kvm->lock.
2410 */
2411void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
2412{
2413	long int i;
2414	u32 cores_done = 0;
2415
2416	if ((kvm->arch.lpcr & mask) == lpcr)
2417		return;
2418
2419	kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
2420
2421	for (i = 0; i < KVM_MAX_VCORES; ++i) {
2422		struct kvmppc_vcore *vc = kvm->arch.vcores[i];
2423		if (!vc)
2424			continue;
2425		spin_lock(&vc->lock);
2426		vc->lpcr = (vc->lpcr & ~mask) | lpcr;
2427		spin_unlock(&vc->lock);
2428		if (++cores_done >= kvm->arch.online_vcores)
2429			break;
2430	}
2431}
2432
2433static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
2434{
2435	return;
2436}
2437
2438static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
2439{
2440	int err = 0;
2441	struct kvm *kvm = vcpu->kvm;
2442	unsigned long hva;
2443	struct kvm_memory_slot *memslot;
2444	struct vm_area_struct *vma;
2445	unsigned long lpcr = 0, senc;
2446	unsigned long psize, porder;
2447	int srcu_idx;
2448
2449	mutex_lock(&kvm->lock);
2450	if (kvm->arch.hpte_setup_done)
2451		goto out;	/* another vcpu beat us to it */
2452
2453	/* Allocate hashed page table (if not done already) and reset it */
2454	if (!kvm->arch.hpt_virt) {
2455		err = kvmppc_alloc_hpt(kvm, NULL);
2456		if (err) {
2457			pr_err("KVM: Couldn't alloc HPT\n");
2458			goto out;
2459		}
2460	}
2461
2462	/* Look up the memslot for guest physical address 0 */
2463	srcu_idx = srcu_read_lock(&kvm->srcu);
2464	memslot = gfn_to_memslot(kvm, 0);
2465
2466	/* We must have some memory at 0 by now */
2467	err = -EINVAL;
2468	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
2469		goto out_srcu;
2470
2471	/* Look up the VMA for the start of this memory slot */
2472	hva = memslot->userspace_addr;
2473	down_read(&current->mm->mmap_sem);
2474	vma = find_vma(current->mm, hva);
2475	if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
2476		goto up_out;
2477
2478	psize = vma_kernel_pagesize(vma);
2479	porder = __ilog2(psize);
2480
2481	up_read(&current->mm->mmap_sem);
2482
2483	/* We can handle 4k, 64k or 16M pages in the VRMA */
2484	err = -EINVAL;
2485	if (!(psize == 0x1000 || psize == 0x10000 ||
2486	      psize == 0x1000000))
2487		goto out_srcu;
2488
2489	/* Update VRMASD field in the LPCR */
2490	senc = slb_pgsize_encoding(psize);
2491	kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
2492		(VRMA_VSID << SLB_VSID_SHIFT_1T);
2493	/* the -4 is to account for senc values starting at 0x10 */
2494	lpcr = senc << (LPCR_VRMASD_SH - 4);
2495
2496	/* Create HPTEs in the hash page table for the VRMA */
2497	kvmppc_map_vrma(vcpu, memslot, porder);
2498
2499	kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
2500
2501	/* Order updates to kvm->arch.lpcr etc. vs. hpte_setup_done */
2502	smp_wmb();
2503	kvm->arch.hpte_setup_done = 1;
2504	err = 0;
2505 out_srcu:
2506	srcu_read_unlock(&kvm->srcu, srcu_idx);
2507 out:
2508	mutex_unlock(&kvm->lock);
2509	return err;
2510
2511 up_out:
2512	up_read(&current->mm->mmap_sem);
2513	goto out_srcu;
2514}
2515
2516static int kvmppc_core_init_vm_hv(struct kvm *kvm)
2517{
2518	unsigned long lpcr, lpid;
2519	char buf[32];
2520
2521	/* Allocate the guest's logical partition ID */
2522
2523	lpid = kvmppc_alloc_lpid();
2524	if ((long)lpid < 0)
2525		return -ENOMEM;
2526	kvm->arch.lpid = lpid;
2527
2528	/*
2529	 * Since we don't flush the TLB when tearing down a VM,
2530	 * and this lpid might have previously been used,
2531	 * make sure we flush on each core before running the new VM.
2532	 */
2533	cpumask_setall(&kvm->arch.need_tlb_flush);
2534
2535	/* Start out with the default set of hcalls enabled */
2536	memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
2537	       sizeof(kvm->arch.enabled_hcalls));
2538
2539	kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
2540
2541	/* Init LPCR for virtual RMA mode */
2542	kvm->arch.host_lpid = mfspr(SPRN_LPID);
2543	kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
2544	lpcr &= LPCR_PECE | LPCR_LPES;
2545	lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
2546		LPCR_VPM0 | LPCR_VPM1;
2547	kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
2548		(VRMA_VSID << SLB_VSID_SHIFT_1T);
2549	/* On POWER8 turn on online bit to enable PURR/SPURR */
2550	if (cpu_has_feature(CPU_FTR_ARCH_207S))
2551		lpcr |= LPCR_ONL;
2552	kvm->arch.lpcr = lpcr;
2553
2554	/*
2555	 * Track that we now have a HV mode VM active. This blocks secondary
2556	 * CPU threads from coming online.
2557	 */
2558	kvm_hv_vm_activated();
2559
2560	/*
2561	 * Create a debugfs directory for the VM
2562	 */
2563	snprintf(buf, sizeof(buf), "vm%d", current->pid);
2564	kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
2565	if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
2566		kvmppc_mmu_debugfs_init(kvm);
2567
2568	return 0;
2569}
2570
2571static void kvmppc_free_vcores(struct kvm *kvm)
2572{
2573	long int i;
2574
2575	for (i = 0; i < KVM_MAX_VCORES; ++i) {
2576		if (kvm->arch.vcores[i] && kvm->arch.vcores[i]->mpp_buffer) {
2577			struct kvmppc_vcore *vc = kvm->arch.vcores[i];
2578			free_pages((unsigned long)vc->mpp_buffer,
2579				   MPP_BUFFER_ORDER);
2580		}
2581		kfree(kvm->arch.vcores[i]);
2582	}
2583	kvm->arch.online_vcores = 0;
2584}
2585
2586static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
2587{
2588	debugfs_remove_recursive(kvm->arch.debugfs_dir);
2589
2590	kvm_hv_vm_deactivated();
2591
2592	kvmppc_free_vcores(kvm);
2593
2594	kvmppc_free_hpt(kvm);
2595}
2596
2597/* We don't need to emulate any privileged instructions or dcbz */
2598static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
2599				     unsigned int inst, int *advance)
2600{
2601	return EMULATE_FAIL;
2602}
2603
2604static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
2605					ulong spr_val)
2606{
2607	return EMULATE_FAIL;
2608}
2609
2610static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
2611					ulong *spr_val)
2612{
2613	return EMULATE_FAIL;
2614}
2615
2616static int kvmppc_core_check_processor_compat_hv(void)
2617{
2618	if (!cpu_has_feature(CPU_FTR_HVMODE) ||
2619	    !cpu_has_feature(CPU_FTR_ARCH_206))
2620		return -EIO;
2621	return 0;
2622}
2623
2624static long kvm_arch_vm_ioctl_hv(struct file *filp,
2625				 unsigned int ioctl, unsigned long arg)
2626{
2627	struct kvm *kvm __maybe_unused = filp->private_data;
2628	void __user *argp = (void __user *)arg;
2629	long r;
2630
2631	switch (ioctl) {
2632
2633	case KVM_PPC_ALLOCATE_HTAB: {
2634		u32 htab_order;
2635
2636		r = -EFAULT;
2637		if (get_user(htab_order, (u32 __user *)argp))
2638			break;
2639		r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
2640		if (r)
2641			break;
2642		r = -EFAULT;
2643		if (put_user(htab_order, (u32 __user *)argp))
2644			break;
2645		r = 0;
2646		break;
2647	}
2648
2649	case KVM_PPC_GET_HTAB_FD: {
2650		struct kvm_get_htab_fd ghf;
2651
2652		r = -EFAULT;
2653		if (copy_from_user(&ghf, argp, sizeof(ghf)))
2654			break;
2655		r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
2656		break;
2657	}
2658
2659	default:
2660		r = -ENOTTY;
2661	}
2662
2663	return r;
2664}
2665
2666/*
2667 * List of hcall numbers to enable by default.
2668 * For compatibility with old userspace, we enable by default
2669 * all hcalls that were implemented before the hcall-enabling
2670 * facility was added.  Note this list should not include H_RTAS.
2671 */
2672static unsigned int default_hcall_list[] = {
2673	H_REMOVE,
2674	H_ENTER,
2675	H_READ,
2676	H_PROTECT,
2677	H_BULK_REMOVE,
2678	H_GET_TCE,
2679	H_PUT_TCE,
2680	H_SET_DABR,
2681	H_SET_XDABR,
2682	H_CEDE,
2683	H_PROD,
2684	H_CONFER,
2685	H_REGISTER_VPA,
2686#ifdef CONFIG_KVM_XICS
2687	H_EOI,
2688	H_CPPR,
2689	H_IPI,
2690	H_IPOLL,
2691	H_XIRR,
2692	H_XIRR_X,
2693#endif
2694	0
2695};
2696
2697static void init_default_hcalls(void)
2698{
2699	int i;
2700	unsigned int hcall;
2701
2702	for (i = 0; default_hcall_list[i]; ++i) {
2703		hcall = default_hcall_list[i];
2704		WARN_ON(!kvmppc_hcall_impl_hv(hcall));
2705		__set_bit(hcall / 4, default_enabled_hcalls);
2706	}
2707}
2708
2709static struct kvmppc_ops kvm_ops_hv = {
2710	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
2711	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
2712	.get_one_reg = kvmppc_get_one_reg_hv,
2713	.set_one_reg = kvmppc_set_one_reg_hv,
2714	.vcpu_load   = kvmppc_core_vcpu_load_hv,
2715	.vcpu_put    = kvmppc_core_vcpu_put_hv,
2716	.set_msr     = kvmppc_set_msr_hv,
2717	.vcpu_run    = kvmppc_vcpu_run_hv,
2718	.vcpu_create = kvmppc_core_vcpu_create_hv,
2719	.vcpu_free   = kvmppc_core_vcpu_free_hv,
2720	.check_requests = kvmppc_core_check_requests_hv,
2721	.get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
2722	.flush_memslot  = kvmppc_core_flush_memslot_hv,
2723	.prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
2724	.commit_memory_region  = kvmppc_core_commit_memory_region_hv,
2725	.unmap_hva = kvm_unmap_hva_hv,
2726	.unmap_hva_range = kvm_unmap_hva_range_hv,
2727	.age_hva  = kvm_age_hva_hv,
2728	.test_age_hva = kvm_test_age_hva_hv,
2729	.set_spte_hva = kvm_set_spte_hva_hv,
2730	.mmu_destroy  = kvmppc_mmu_destroy_hv,
2731	.free_memslot = kvmppc_core_free_memslot_hv,
2732	.create_memslot = kvmppc_core_create_memslot_hv,
2733	.init_vm =  kvmppc_core_init_vm_hv,
2734	.destroy_vm = kvmppc_core_destroy_vm_hv,
2735	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
2736	.emulate_op = kvmppc_core_emulate_op_hv,
2737	.emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
2738	.emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
2739	.fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
2740	.arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
2741	.hcall_implemented = kvmppc_hcall_impl_hv,
2742};
2743
2744static int kvmppc_book3s_init_hv(void)
2745{
2746	int r;
2747	/*
2748	 * FIXME!! Do we need to check on all cpus ?
2749	 */
2750	r = kvmppc_core_check_processor_compat_hv();
2751	if (r < 0)
2752		return -ENODEV;
2753
2754	kvm_ops_hv.owner = THIS_MODULE;
2755	kvmppc_hv_ops = &kvm_ops_hv;
2756
2757	init_default_hcalls();
2758
2759	r = kvmppc_mmu_hv_init();
2760	return r;
2761}
2762
2763static void kvmppc_book3s_exit_hv(void)
2764{
2765	kvmppc_hv_ops = NULL;
2766}
2767
2768module_init(kvmppc_book3s_init_hv);
2769module_exit(kvmppc_book3s_exit_hv);
2770MODULE_LICENSE("GPL");
2771MODULE_ALIAS_MISCDEV(KVM_MINOR);
2772MODULE_ALIAS("devname:kvm");
2773