1/*
2 * Copyright (C) 2008-2011 Freescale Semiconductor, Inc. All rights reserved.
3 *
4 * Author: Yu Liu, <yu.liu@freescale.com>
5 *
6 * Description:
7 * This file is derived from arch/powerpc/kvm/44x.c,
8 * by Hollis Blanchard <hollisb@us.ibm.com>.
9 *
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License, version 2, as
12 * published by the Free Software Foundation.
13 */
14
15#include <linux/kvm_host.h>
16#include <linux/slab.h>
17#include <linux/err.h>
18#include <linux/export.h>
19#include <linux/module.h>
20#include <linux/miscdevice.h>
21
22#include <asm/reg.h>
23#include <asm/cputable.h>
24#include <asm/tlbflush.h>
25#include <asm/kvm_ppc.h>
26
27#include "../mm/mmu_decl.h"
28#include "booke.h"
29#include "e500.h"
30
31struct id {
32	unsigned long val;
33	struct id **pentry;
34};
35
36#define NUM_TIDS 256
37
38/*
39 * This table provide mappings from:
40 * (guestAS,guestTID,guestPR) --> ID of physical cpu
41 * guestAS	[0..1]
42 * guestTID	[0..255]
43 * guestPR	[0..1]
44 * ID		[1..255]
45 * Each vcpu keeps one vcpu_id_table.
46 */
47struct vcpu_id_table {
48	struct id id[2][NUM_TIDS][2];
49};
50
51/*
52 * This table provide reversed mappings of vcpu_id_table:
53 * ID --> address of vcpu_id_table item.
54 * Each physical core has one pcpu_id_table.
55 */
56struct pcpu_id_table {
57	struct id *entry[NUM_TIDS];
58};
59
60static DEFINE_PER_CPU(struct pcpu_id_table, pcpu_sids);
61
62/* This variable keeps last used shadow ID on local core.
63 * The valid range of shadow ID is [1..255] */
64static DEFINE_PER_CPU(unsigned long, pcpu_last_used_sid);
65
66/*
67 * Allocate a free shadow id and setup a valid sid mapping in given entry.
68 * A mapping is only valid when vcpu_id_table and pcpu_id_table are match.
69 *
70 * The caller must have preemption disabled, and keep it that way until
71 * it has finished with the returned shadow id (either written into the
72 * TLB or arch.shadow_pid, or discarded).
73 */
74static inline int local_sid_setup_one(struct id *entry)
75{
76	unsigned long sid;
77	int ret = -1;
78
79	sid = ++(__get_cpu_var(pcpu_last_used_sid));
80	if (sid < NUM_TIDS) {
81		__get_cpu_var(pcpu_sids).entry[sid] = entry;
82		entry->val = sid;
83		entry->pentry = &__get_cpu_var(pcpu_sids).entry[sid];
84		ret = sid;
85	}
86
87	/*
88	 * If sid == NUM_TIDS, we've run out of sids.  We return -1, and
89	 * the caller will invalidate everything and start over.
90	 *
91	 * sid > NUM_TIDS indicates a race, which we disable preemption to
92	 * avoid.
93	 */
94	WARN_ON(sid > NUM_TIDS);
95
96	return ret;
97}
98
99/*
100 * Check if given entry contain a valid shadow id mapping.
101 * An ID mapping is considered valid only if
102 * both vcpu and pcpu know this mapping.
103 *
104 * The caller must have preemption disabled, and keep it that way until
105 * it has finished with the returned shadow id (either written into the
106 * TLB or arch.shadow_pid, or discarded).
107 */
108static inline int local_sid_lookup(struct id *entry)
109{
110	if (entry && entry->val != 0 &&
111	    __get_cpu_var(pcpu_sids).entry[entry->val] == entry &&
112	    entry->pentry == &__get_cpu_var(pcpu_sids).entry[entry->val])
113		return entry->val;
114	return -1;
115}
116
117/* Invalidate all id mappings on local core -- call with preempt disabled */
118static inline void local_sid_destroy_all(void)
119{
120	__get_cpu_var(pcpu_last_used_sid) = 0;
121	memset(&__get_cpu_var(pcpu_sids), 0, sizeof(__get_cpu_var(pcpu_sids)));
122}
123
124static void *kvmppc_e500_id_table_alloc(struct kvmppc_vcpu_e500 *vcpu_e500)
125{
126	vcpu_e500->idt = kzalloc(sizeof(struct vcpu_id_table), GFP_KERNEL);
127	return vcpu_e500->idt;
128}
129
130static void kvmppc_e500_id_table_free(struct kvmppc_vcpu_e500 *vcpu_e500)
131{
132	kfree(vcpu_e500->idt);
133	vcpu_e500->idt = NULL;
134}
135
136/* Map guest pid to shadow.
137 * We use PID to keep shadow of current guest non-zero PID,
138 * and use PID1 to keep shadow of guest zero PID.
139 * So that guest tlbe with TID=0 can be accessed at any time */
140static void kvmppc_e500_recalc_shadow_pid(struct kvmppc_vcpu_e500 *vcpu_e500)
141{
142	preempt_disable();
143	vcpu_e500->vcpu.arch.shadow_pid = kvmppc_e500_get_sid(vcpu_e500,
144			get_cur_as(&vcpu_e500->vcpu),
145			get_cur_pid(&vcpu_e500->vcpu),
146			get_cur_pr(&vcpu_e500->vcpu), 1);
147	vcpu_e500->vcpu.arch.shadow_pid1 = kvmppc_e500_get_sid(vcpu_e500,
148			get_cur_as(&vcpu_e500->vcpu), 0,
149			get_cur_pr(&vcpu_e500->vcpu), 1);
150	preempt_enable();
151}
152
153/* Invalidate all mappings on vcpu */
154static void kvmppc_e500_id_table_reset_all(struct kvmppc_vcpu_e500 *vcpu_e500)
155{
156	memset(vcpu_e500->idt, 0, sizeof(struct vcpu_id_table));
157
158	/* Update shadow pid when mappings are changed */
159	kvmppc_e500_recalc_shadow_pid(vcpu_e500);
160}
161
162/* Invalidate one ID mapping on vcpu */
163static inline void kvmppc_e500_id_table_reset_one(
164			       struct kvmppc_vcpu_e500 *vcpu_e500,
165			       int as, int pid, int pr)
166{
167	struct vcpu_id_table *idt = vcpu_e500->idt;
168
169	BUG_ON(as >= 2);
170	BUG_ON(pid >= NUM_TIDS);
171	BUG_ON(pr >= 2);
172
173	idt->id[as][pid][pr].val = 0;
174	idt->id[as][pid][pr].pentry = NULL;
175
176	/* Update shadow pid when mappings are changed */
177	kvmppc_e500_recalc_shadow_pid(vcpu_e500);
178}
179
180/*
181 * Map guest (vcpu,AS,ID,PR) to physical core shadow id.
182 * This function first lookup if a valid mapping exists,
183 * if not, then creates a new one.
184 *
185 * The caller must have preemption disabled, and keep it that way until
186 * it has finished with the returned shadow id (either written into the
187 * TLB or arch.shadow_pid, or discarded).
188 */
189unsigned int kvmppc_e500_get_sid(struct kvmppc_vcpu_e500 *vcpu_e500,
190				 unsigned int as, unsigned int gid,
191				 unsigned int pr, int avoid_recursion)
192{
193	struct vcpu_id_table *idt = vcpu_e500->idt;
194	int sid;
195
196	BUG_ON(as >= 2);
197	BUG_ON(gid >= NUM_TIDS);
198	BUG_ON(pr >= 2);
199
200	sid = local_sid_lookup(&idt->id[as][gid][pr]);
201
202	while (sid <= 0) {
203		/* No mapping yet */
204		sid = local_sid_setup_one(&idt->id[as][gid][pr]);
205		if (sid <= 0) {
206			_tlbil_all();
207			local_sid_destroy_all();
208		}
209
210		/* Update shadow pid when mappings are changed */
211		if (!avoid_recursion)
212			kvmppc_e500_recalc_shadow_pid(vcpu_e500);
213	}
214
215	return sid;
216}
217
218unsigned int kvmppc_e500_get_tlb_stid(struct kvm_vcpu *vcpu,
219				      struct kvm_book3e_206_tlb_entry *gtlbe)
220{
221	return kvmppc_e500_get_sid(to_e500(vcpu), get_tlb_ts(gtlbe),
222				   get_tlb_tid(gtlbe), get_cur_pr(vcpu), 0);
223}
224
225void kvmppc_set_pid(struct kvm_vcpu *vcpu, u32 pid)
226{
227	struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
228
229	if (vcpu->arch.pid != pid) {
230		vcpu_e500->pid[0] = vcpu->arch.pid = pid;
231		kvmppc_e500_recalc_shadow_pid(vcpu_e500);
232	}
233}
234
235/* gtlbe must not be mapped by more than one host tlbe */
236void kvmppc_e500_tlbil_one(struct kvmppc_vcpu_e500 *vcpu_e500,
237                           struct kvm_book3e_206_tlb_entry *gtlbe)
238{
239	struct vcpu_id_table *idt = vcpu_e500->idt;
240	unsigned int pr, tid, ts, pid;
241	u32 val, eaddr;
242	unsigned long flags;
243
244	ts = get_tlb_ts(gtlbe);
245	tid = get_tlb_tid(gtlbe);
246
247	preempt_disable();
248
249	/* One guest ID may be mapped to two shadow IDs */
250	for (pr = 0; pr < 2; pr++) {
251		/*
252		 * The shadow PID can have a valid mapping on at most one
253		 * host CPU.  In the common case, it will be valid on this
254		 * CPU, in which case we do a local invalidation of the
255		 * specific address.
256		 *
257		 * If the shadow PID is not valid on the current host CPU,
258		 * we invalidate the entire shadow PID.
259		 */
260		pid = local_sid_lookup(&idt->id[ts][tid][pr]);
261		if (pid <= 0) {
262			kvmppc_e500_id_table_reset_one(vcpu_e500, ts, tid, pr);
263			continue;
264		}
265
266		/*
267		 * The guest is invalidating a 4K entry which is in a PID
268		 * that has a valid shadow mapping on this host CPU.  We
269		 * search host TLB to invalidate it's shadow TLB entry,
270		 * similar to __tlbil_va except that we need to look in AS1.
271		 */
272		val = (pid << MAS6_SPID_SHIFT) | MAS6_SAS;
273		eaddr = get_tlb_eaddr(gtlbe);
274
275		local_irq_save(flags);
276
277		mtspr(SPRN_MAS6, val);
278		asm volatile("tlbsx 0, %[eaddr]" : : [eaddr] "r" (eaddr));
279		val = mfspr(SPRN_MAS1);
280		if (val & MAS1_VALID) {
281			mtspr(SPRN_MAS1, val & ~MAS1_VALID);
282			asm volatile("tlbwe");
283		}
284
285		local_irq_restore(flags);
286	}
287
288	preempt_enable();
289}
290
291void kvmppc_e500_tlbil_all(struct kvmppc_vcpu_e500 *vcpu_e500)
292{
293	kvmppc_e500_id_table_reset_all(vcpu_e500);
294}
295
296void kvmppc_mmu_msr_notify(struct kvm_vcpu *vcpu, u32 old_msr)
297{
298	/* Recalc shadow pid since MSR changes */
299	kvmppc_e500_recalc_shadow_pid(to_e500(vcpu));
300}
301
302void kvmppc_core_load_host_debugstate(struct kvm_vcpu *vcpu)
303{
304}
305
306void kvmppc_core_load_guest_debugstate(struct kvm_vcpu *vcpu)
307{
308}
309
310static void kvmppc_core_vcpu_load_e500(struct kvm_vcpu *vcpu, int cpu)
311{
312	kvmppc_booke_vcpu_load(vcpu, cpu);
313
314	/* Shadow PID may be expired on local core */
315	kvmppc_e500_recalc_shadow_pid(to_e500(vcpu));
316}
317
318static void kvmppc_core_vcpu_put_e500(struct kvm_vcpu *vcpu)
319{
320#ifdef CONFIG_SPE
321	if (vcpu->arch.shadow_msr & MSR_SPE)
322		kvmppc_vcpu_disable_spe(vcpu);
323#endif
324
325	kvmppc_booke_vcpu_put(vcpu);
326}
327
328int kvmppc_core_check_processor_compat(void)
329{
330	int r;
331
332	if (strcmp(cur_cpu_spec->cpu_name, "e500v2") == 0)
333		r = 0;
334	else
335		r = -ENOTSUPP;
336
337	return r;
338}
339
340static void kvmppc_e500_tlb_setup(struct kvmppc_vcpu_e500 *vcpu_e500)
341{
342	struct kvm_book3e_206_tlb_entry *tlbe;
343
344	/* Insert large initial mapping for guest. */
345	tlbe = get_entry(vcpu_e500, 1, 0);
346	tlbe->mas1 = MAS1_VALID | MAS1_TSIZE(BOOK3E_PAGESZ_256M);
347	tlbe->mas2 = 0;
348	tlbe->mas7_3 = E500_TLB_SUPER_PERM_MASK;
349
350	/* 4K map for serial output. Used by kernel wrapper. */
351	tlbe = get_entry(vcpu_e500, 1, 1);
352	tlbe->mas1 = MAS1_VALID | MAS1_TSIZE(BOOK3E_PAGESZ_4K);
353	tlbe->mas2 = (0xe0004500 & 0xFFFFF000) | MAS2_I | MAS2_G;
354	tlbe->mas7_3 = (0xe0004500 & 0xFFFFF000) | E500_TLB_SUPER_PERM_MASK;
355}
356
357int kvmppc_core_vcpu_setup(struct kvm_vcpu *vcpu)
358{
359	struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
360
361	kvmppc_e500_tlb_setup(vcpu_e500);
362
363	/* Registers init */
364	vcpu->arch.pvr = mfspr(SPRN_PVR);
365	vcpu_e500->svr = mfspr(SPRN_SVR);
366
367	vcpu->arch.cpu_type = KVM_CPU_E500V2;
368
369	return 0;
370}
371
372static int kvmppc_core_get_sregs_e500(struct kvm_vcpu *vcpu,
373				      struct kvm_sregs *sregs)
374{
375	struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
376
377	sregs->u.e.features |= KVM_SREGS_E_ARCH206_MMU | KVM_SREGS_E_SPE |
378	                       KVM_SREGS_E_PM;
379	sregs->u.e.impl_id = KVM_SREGS_E_IMPL_FSL;
380
381	sregs->u.e.impl.fsl.features = 0;
382	sregs->u.e.impl.fsl.svr = vcpu_e500->svr;
383	sregs->u.e.impl.fsl.hid0 = vcpu_e500->hid0;
384	sregs->u.e.impl.fsl.mcar = vcpu_e500->mcar;
385
386	sregs->u.e.ivor_high[0] = vcpu->arch.ivor[BOOKE_IRQPRIO_SPE_UNAVAIL];
387	sregs->u.e.ivor_high[1] = vcpu->arch.ivor[BOOKE_IRQPRIO_SPE_FP_DATA];
388	sregs->u.e.ivor_high[2] = vcpu->arch.ivor[BOOKE_IRQPRIO_SPE_FP_ROUND];
389	sregs->u.e.ivor_high[3] =
390		vcpu->arch.ivor[BOOKE_IRQPRIO_PERFORMANCE_MONITOR];
391
392	kvmppc_get_sregs_ivor(vcpu, sregs);
393	kvmppc_get_sregs_e500_tlb(vcpu, sregs);
394	return 0;
395}
396
397static int kvmppc_core_set_sregs_e500(struct kvm_vcpu *vcpu,
398				      struct kvm_sregs *sregs)
399{
400	struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
401	int ret;
402
403	if (sregs->u.e.impl_id == KVM_SREGS_E_IMPL_FSL) {
404		vcpu_e500->svr = sregs->u.e.impl.fsl.svr;
405		vcpu_e500->hid0 = sregs->u.e.impl.fsl.hid0;
406		vcpu_e500->mcar = sregs->u.e.impl.fsl.mcar;
407	}
408
409	ret = kvmppc_set_sregs_e500_tlb(vcpu, sregs);
410	if (ret < 0)
411		return ret;
412
413	if (!(sregs->u.e.features & KVM_SREGS_E_IVOR))
414		return 0;
415
416	if (sregs->u.e.features & KVM_SREGS_E_SPE) {
417		vcpu->arch.ivor[BOOKE_IRQPRIO_SPE_UNAVAIL] =
418			sregs->u.e.ivor_high[0];
419		vcpu->arch.ivor[BOOKE_IRQPRIO_SPE_FP_DATA] =
420			sregs->u.e.ivor_high[1];
421		vcpu->arch.ivor[BOOKE_IRQPRIO_SPE_FP_ROUND] =
422			sregs->u.e.ivor_high[2];
423	}
424
425	if (sregs->u.e.features & KVM_SREGS_E_PM) {
426		vcpu->arch.ivor[BOOKE_IRQPRIO_PERFORMANCE_MONITOR] =
427			sregs->u.e.ivor_high[3];
428	}
429
430	return kvmppc_set_sregs_ivor(vcpu, sregs);
431}
432
433static int kvmppc_get_one_reg_e500(struct kvm_vcpu *vcpu, u64 id,
434				   union kvmppc_one_reg *val)
435{
436	int r = kvmppc_get_one_reg_e500_tlb(vcpu, id, val);
437	return r;
438}
439
440static int kvmppc_set_one_reg_e500(struct kvm_vcpu *vcpu, u64 id,
441				   union kvmppc_one_reg *val)
442{
443	int r = kvmppc_get_one_reg_e500_tlb(vcpu, id, val);
444	return r;
445}
446
447static struct kvm_vcpu *kvmppc_core_vcpu_create_e500(struct kvm *kvm,
448						     unsigned int id)
449{
450	struct kvmppc_vcpu_e500 *vcpu_e500;
451	struct kvm_vcpu *vcpu;
452	int err;
453
454	vcpu_e500 = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
455	if (!vcpu_e500) {
456		err = -ENOMEM;
457		goto out;
458	}
459
460	vcpu = &vcpu_e500->vcpu;
461	err = kvm_vcpu_init(vcpu, kvm, id);
462	if (err)
463		goto free_vcpu;
464
465	if (kvmppc_e500_id_table_alloc(vcpu_e500) == NULL)
466		goto uninit_vcpu;
467
468	err = kvmppc_e500_tlb_init(vcpu_e500);
469	if (err)
470		goto uninit_id;
471
472	vcpu->arch.shared = (void*)__get_free_page(GFP_KERNEL|__GFP_ZERO);
473	if (!vcpu->arch.shared)
474		goto uninit_tlb;
475
476	return vcpu;
477
478uninit_tlb:
479	kvmppc_e500_tlb_uninit(vcpu_e500);
480uninit_id:
481	kvmppc_e500_id_table_free(vcpu_e500);
482uninit_vcpu:
483	kvm_vcpu_uninit(vcpu);
484free_vcpu:
485	kmem_cache_free(kvm_vcpu_cache, vcpu_e500);
486out:
487	return ERR_PTR(err);
488}
489
490static void kvmppc_core_vcpu_free_e500(struct kvm_vcpu *vcpu)
491{
492	struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
493
494	free_page((unsigned long)vcpu->arch.shared);
495	kvmppc_e500_tlb_uninit(vcpu_e500);
496	kvmppc_e500_id_table_free(vcpu_e500);
497	kvm_vcpu_uninit(vcpu);
498	kmem_cache_free(kvm_vcpu_cache, vcpu_e500);
499}
500
501static int kvmppc_core_init_vm_e500(struct kvm *kvm)
502{
503	return 0;
504}
505
506static void kvmppc_core_destroy_vm_e500(struct kvm *kvm)
507{
508}
509
510static struct kvmppc_ops kvm_ops_e500 = {
511	.get_sregs = kvmppc_core_get_sregs_e500,
512	.set_sregs = kvmppc_core_set_sregs_e500,
513	.get_one_reg = kvmppc_get_one_reg_e500,
514	.set_one_reg = kvmppc_set_one_reg_e500,
515	.vcpu_load   = kvmppc_core_vcpu_load_e500,
516	.vcpu_put    = kvmppc_core_vcpu_put_e500,
517	.vcpu_create = kvmppc_core_vcpu_create_e500,
518	.vcpu_free   = kvmppc_core_vcpu_free_e500,
519	.mmu_destroy  = kvmppc_mmu_destroy_e500,
520	.init_vm = kvmppc_core_init_vm_e500,
521	.destroy_vm = kvmppc_core_destroy_vm_e500,
522	.emulate_op = kvmppc_core_emulate_op_e500,
523	.emulate_mtspr = kvmppc_core_emulate_mtspr_e500,
524	.emulate_mfspr = kvmppc_core_emulate_mfspr_e500,
525};
526
527static int __init kvmppc_e500_init(void)
528{
529	int r, i;
530	unsigned long ivor[3];
531	/* Process remaining handlers above the generic first 16 */
532	unsigned long *handler = &kvmppc_booke_handler_addr[16];
533	unsigned long handler_len;
534	unsigned long max_ivor = 0;
535
536	r = kvmppc_core_check_processor_compat();
537	if (r)
538		goto err_out;
539
540	r = kvmppc_booke_init();
541	if (r)
542		goto err_out;
543
544	/* copy extra E500 exception handlers */
545	ivor[0] = mfspr(SPRN_IVOR32);
546	ivor[1] = mfspr(SPRN_IVOR33);
547	ivor[2] = mfspr(SPRN_IVOR34);
548	for (i = 0; i < 3; i++) {
549		if (ivor[i] > ivor[max_ivor])
550			max_ivor = i;
551
552		handler_len = handler[i + 1] - handler[i];
553		memcpy((void *)kvmppc_booke_handlers + ivor[i],
554		       (void *)handler[i], handler_len);
555	}
556	handler_len = handler[max_ivor + 1] - handler[max_ivor];
557	flush_icache_range(kvmppc_booke_handlers, kvmppc_booke_handlers +
558			   ivor[max_ivor] + handler_len);
559
560	r = kvm_init(NULL, sizeof(struct kvmppc_vcpu_e500), 0, THIS_MODULE);
561	if (r)
562		goto err_out;
563	kvm_ops_e500.owner = THIS_MODULE;
564	kvmppc_pr_ops = &kvm_ops_e500;
565
566err_out:
567	return r;
568}
569
570static void __exit kvmppc_e500_exit(void)
571{
572	kvmppc_pr_ops = NULL;
573	kvmppc_booke_exit();
574}
575
576module_init(kvmppc_e500_init);
577module_exit(kvmppc_e500_exit);
578MODULE_ALIAS_MISCDEV(KVM_MINOR);
579MODULE_ALIAS("devname:kvm");
580