1/*
2 * This file is subject to the terms and conditions of the GNU General Public
3 * License.  See the file "COPYING" in the main directory of this archive
4 * for more details.
5 *
6 * KVM/MIPS: MIPS specific KVM APIs
7 *
8 * Copyright (C) 2012  MIPS Technologies, Inc.  All rights reserved.
9 * Authors: Sanjay Lal <sanjayl@kymasys.com>
10 */
11
12#include <linux/errno.h>
13#include <linux/err.h>
14#include <linux/kdebug.h>
15#include <linux/module.h>
16#include <linux/vmalloc.h>
17#include <linux/fs.h>
18#include <linux/bootmem.h>
19#include <asm/fpu.h>
20#include <asm/page.h>
21#include <asm/cacheflush.h>
22#include <asm/mmu_context.h>
23#include <asm/pgtable.h>
24
25#include <linux/kvm_host.h>
26
27#include "interrupt.h"
28#include "commpage.h"
29
30#define CREATE_TRACE_POINTS
31#include "trace.h"
32
33#ifndef VECTORSPACING
34#define VECTORSPACING 0x100	/* for EI/VI mode */
35#endif
36
37#define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x)
38struct kvm_stats_debugfs_item debugfs_entries[] = {
39	{ "wait",	  VCPU_STAT(wait_exits),	 KVM_STAT_VCPU },
40	{ "cache",	  VCPU_STAT(cache_exits),	 KVM_STAT_VCPU },
41	{ "signal",	  VCPU_STAT(signal_exits),	 KVM_STAT_VCPU },
42	{ "interrupt",	  VCPU_STAT(int_exits),		 KVM_STAT_VCPU },
43	{ "cop_unsuable", VCPU_STAT(cop_unusable_exits), KVM_STAT_VCPU },
44	{ "tlbmod",	  VCPU_STAT(tlbmod_exits),	 KVM_STAT_VCPU },
45	{ "tlbmiss_ld",	  VCPU_STAT(tlbmiss_ld_exits),	 KVM_STAT_VCPU },
46	{ "tlbmiss_st",	  VCPU_STAT(tlbmiss_st_exits),	 KVM_STAT_VCPU },
47	{ "addrerr_st",	  VCPU_STAT(addrerr_st_exits),	 KVM_STAT_VCPU },
48	{ "addrerr_ld",	  VCPU_STAT(addrerr_ld_exits),	 KVM_STAT_VCPU },
49	{ "syscall",	  VCPU_STAT(syscall_exits),	 KVM_STAT_VCPU },
50	{ "resvd_inst",	  VCPU_STAT(resvd_inst_exits),	 KVM_STAT_VCPU },
51	{ "break_inst",	  VCPU_STAT(break_inst_exits),	 KVM_STAT_VCPU },
52	{ "trap_inst",	  VCPU_STAT(trap_inst_exits),	 KVM_STAT_VCPU },
53	{ "msa_fpe",	  VCPU_STAT(msa_fpe_exits),	 KVM_STAT_VCPU },
54	{ "fpe",	  VCPU_STAT(fpe_exits),		 KVM_STAT_VCPU },
55	{ "msa_disabled", VCPU_STAT(msa_disabled_exits), KVM_STAT_VCPU },
56	{ "flush_dcache", VCPU_STAT(flush_dcache_exits), KVM_STAT_VCPU },
57	{ "halt_successful_poll", VCPU_STAT(halt_successful_poll), KVM_STAT_VCPU },
58	{ "halt_wakeup",  VCPU_STAT(halt_wakeup),	 KVM_STAT_VCPU },
59	{NULL}
60};
61
62static int kvm_mips_reset_vcpu(struct kvm_vcpu *vcpu)
63{
64	int i;
65
66	for_each_possible_cpu(i) {
67		vcpu->arch.guest_kernel_asid[i] = 0;
68		vcpu->arch.guest_user_asid[i] = 0;
69	}
70
71	return 0;
72}
73
74/*
75 * XXXKYMA: We are simulatoring a processor that has the WII bit set in
76 * Config7, so we are "runnable" if interrupts are pending
77 */
78int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
79{
80	return !!(vcpu->arch.pending_exceptions);
81}
82
83int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
84{
85	return 1;
86}
87
88int kvm_arch_hardware_enable(void)
89{
90	return 0;
91}
92
93int kvm_arch_hardware_setup(void)
94{
95	return 0;
96}
97
98void kvm_arch_check_processor_compat(void *rtn)
99{
100	*(int *)rtn = 0;
101}
102
103static void kvm_mips_init_tlbs(struct kvm *kvm)
104{
105	unsigned long wired;
106
107	/*
108	 * Add a wired entry to the TLB, it is used to map the commpage to
109	 * the Guest kernel
110	 */
111	wired = read_c0_wired();
112	write_c0_wired(wired + 1);
113	mtc0_tlbw_hazard();
114	kvm->arch.commpage_tlb = wired;
115
116	kvm_debug("[%d] commpage TLB: %d\n", smp_processor_id(),
117		  kvm->arch.commpage_tlb);
118}
119
120static void kvm_mips_init_vm_percpu(void *arg)
121{
122	struct kvm *kvm = (struct kvm *)arg;
123
124	kvm_mips_init_tlbs(kvm);
125	kvm_mips_callbacks->vm_init(kvm);
126
127}
128
129int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
130{
131	if (atomic_inc_return(&kvm_mips_instance) == 1) {
132		kvm_debug("%s: 1st KVM instance, setup host TLB parameters\n",
133			  __func__);
134		on_each_cpu(kvm_mips_init_vm_percpu, kvm, 1);
135	}
136
137	return 0;
138}
139
140void kvm_mips_free_vcpus(struct kvm *kvm)
141{
142	unsigned int i;
143	struct kvm_vcpu *vcpu;
144
145	/* Put the pages we reserved for the guest pmap */
146	for (i = 0; i < kvm->arch.guest_pmap_npages; i++) {
147		if (kvm->arch.guest_pmap[i] != KVM_INVALID_PAGE)
148			kvm_mips_release_pfn_clean(kvm->arch.guest_pmap[i]);
149	}
150	kfree(kvm->arch.guest_pmap);
151
152	kvm_for_each_vcpu(i, vcpu, kvm) {
153		kvm_arch_vcpu_free(vcpu);
154	}
155
156	mutex_lock(&kvm->lock);
157
158	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
159		kvm->vcpus[i] = NULL;
160
161	atomic_set(&kvm->online_vcpus, 0);
162
163	mutex_unlock(&kvm->lock);
164}
165
166static void kvm_mips_uninit_tlbs(void *arg)
167{
168	/* Restore wired count */
169	write_c0_wired(0);
170	mtc0_tlbw_hazard();
171	/* Clear out all the TLBs */
172	kvm_local_flush_tlb_all();
173}
174
175void kvm_arch_destroy_vm(struct kvm *kvm)
176{
177	kvm_mips_free_vcpus(kvm);
178
179	/* If this is the last instance, restore wired count */
180	if (atomic_dec_return(&kvm_mips_instance) == 0) {
181		kvm_debug("%s: last KVM instance, restoring TLB parameters\n",
182			  __func__);
183		on_each_cpu(kvm_mips_uninit_tlbs, NULL, 1);
184	}
185}
186
187long kvm_arch_dev_ioctl(struct file *filp, unsigned int ioctl,
188			unsigned long arg)
189{
190	return -ENOIOCTLCMD;
191}
192
193int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
194			    unsigned long npages)
195{
196	return 0;
197}
198
199int kvm_arch_prepare_memory_region(struct kvm *kvm,
200				   struct kvm_memory_slot *memslot,
201				   struct kvm_userspace_memory_region *mem,
202				   enum kvm_mr_change change)
203{
204	return 0;
205}
206
207void kvm_arch_commit_memory_region(struct kvm *kvm,
208				   struct kvm_userspace_memory_region *mem,
209				   const struct kvm_memory_slot *old,
210				   enum kvm_mr_change change)
211{
212	unsigned long npages = 0;
213	int i;
214
215	kvm_debug("%s: kvm: %p slot: %d, GPA: %llx, size: %llx, QVA: %llx\n",
216		  __func__, kvm, mem->slot, mem->guest_phys_addr,
217		  mem->memory_size, mem->userspace_addr);
218
219	/* Setup Guest PMAP table */
220	if (!kvm->arch.guest_pmap) {
221		if (mem->slot == 0)
222			npages = mem->memory_size >> PAGE_SHIFT;
223
224		if (npages) {
225			kvm->arch.guest_pmap_npages = npages;
226			kvm->arch.guest_pmap =
227			    kzalloc(npages * sizeof(unsigned long), GFP_KERNEL);
228
229			if (!kvm->arch.guest_pmap) {
230				kvm_err("Failed to allocate guest PMAP");
231				return;
232			}
233
234			kvm_debug("Allocated space for Guest PMAP Table (%ld pages) @ %p\n",
235				  npages, kvm->arch.guest_pmap);
236
237			/* Now setup the page table */
238			for (i = 0; i < npages; i++)
239				kvm->arch.guest_pmap[i] = KVM_INVALID_PAGE;
240		}
241	}
242}
243
244struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
245{
246	int err, size, offset;
247	void *gebase;
248	int i;
249
250	struct kvm_vcpu *vcpu = kzalloc(sizeof(struct kvm_vcpu), GFP_KERNEL);
251
252	if (!vcpu) {
253		err = -ENOMEM;
254		goto out;
255	}
256
257	err = kvm_vcpu_init(vcpu, kvm, id);
258
259	if (err)
260		goto out_free_cpu;
261
262	kvm_debug("kvm @ %p: create cpu %d at %p\n", kvm, id, vcpu);
263
264	/*
265	 * Allocate space for host mode exception handlers that handle
266	 * guest mode exits
267	 */
268	if (cpu_has_veic || cpu_has_vint)
269		size = 0x200 + VECTORSPACING * 64;
270	else
271		size = 0x4000;
272
273	/* Save Linux EBASE */
274	vcpu->arch.host_ebase = (void *)read_c0_ebase();
275
276	gebase = kzalloc(ALIGN(size, PAGE_SIZE), GFP_KERNEL);
277
278	if (!gebase) {
279		err = -ENOMEM;
280		goto out_uninit_cpu;
281	}
282	kvm_debug("Allocated %d bytes for KVM Exception Handlers @ %p\n",
283		  ALIGN(size, PAGE_SIZE), gebase);
284
285	/* Save new ebase */
286	vcpu->arch.guest_ebase = gebase;
287
288	/* Copy L1 Guest Exception handler to correct offset */
289
290	/* TLB Refill, EXL = 0 */
291	memcpy(gebase, mips32_exception,
292	       mips32_exceptionEnd - mips32_exception);
293
294	/* General Exception Entry point */
295	memcpy(gebase + 0x180, mips32_exception,
296	       mips32_exceptionEnd - mips32_exception);
297
298	/* For vectored interrupts poke the exception code @ all offsets 0-7 */
299	for (i = 0; i < 8; i++) {
300		kvm_debug("L1 Vectored handler @ %p\n",
301			  gebase + 0x200 + (i * VECTORSPACING));
302		memcpy(gebase + 0x200 + (i * VECTORSPACING), mips32_exception,
303		       mips32_exceptionEnd - mips32_exception);
304	}
305
306	/* General handler, relocate to unmapped space for sanity's sake */
307	offset = 0x2000;
308	kvm_debug("Installing KVM Exception handlers @ %p, %#x bytes\n",
309		  gebase + offset,
310		  mips32_GuestExceptionEnd - mips32_GuestException);
311
312	memcpy(gebase + offset, mips32_GuestException,
313	       mips32_GuestExceptionEnd - mips32_GuestException);
314
315	/* Invalidate the icache for these ranges */
316	local_flush_icache_range((unsigned long)gebase,
317				(unsigned long)gebase + ALIGN(size, PAGE_SIZE));
318
319	/*
320	 * Allocate comm page for guest kernel, a TLB will be reserved for
321	 * mapping GVA @ 0xFFFF8000 to this page
322	 */
323	vcpu->arch.kseg0_commpage = kzalloc(PAGE_SIZE << 1, GFP_KERNEL);
324
325	if (!vcpu->arch.kseg0_commpage) {
326		err = -ENOMEM;
327		goto out_free_gebase;
328	}
329
330	kvm_debug("Allocated COMM page @ %p\n", vcpu->arch.kseg0_commpage);
331	kvm_mips_commpage_init(vcpu);
332
333	/* Init */
334	vcpu->arch.last_sched_cpu = -1;
335
336	/* Start off the timer */
337	kvm_mips_init_count(vcpu);
338
339	return vcpu;
340
341out_free_gebase:
342	kfree(gebase);
343
344out_uninit_cpu:
345	kvm_vcpu_uninit(vcpu);
346
347out_free_cpu:
348	kfree(vcpu);
349
350out:
351	return ERR_PTR(err);
352}
353
354void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
355{
356	hrtimer_cancel(&vcpu->arch.comparecount_timer);
357
358	kvm_vcpu_uninit(vcpu);
359
360	kvm_mips_dump_stats(vcpu);
361
362	kfree(vcpu->arch.guest_ebase);
363	kfree(vcpu->arch.kseg0_commpage);
364	kfree(vcpu);
365}
366
367void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
368{
369	kvm_arch_vcpu_free(vcpu);
370}
371
372int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
373					struct kvm_guest_debug *dbg)
374{
375	return -ENOIOCTLCMD;
376}
377
378int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
379{
380	int r = 0;
381	sigset_t sigsaved;
382
383	if (vcpu->sigset_active)
384		sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
385
386	if (vcpu->mmio_needed) {
387		if (!vcpu->mmio_is_write)
388			kvm_mips_complete_mmio_load(vcpu, run);
389		vcpu->mmio_needed = 0;
390	}
391
392	lose_fpu(1);
393
394	local_irq_disable();
395	/* Check if we have any exceptions/interrupts pending */
396	kvm_mips_deliver_interrupts(vcpu,
397				    kvm_read_c0_guest_cause(vcpu->arch.cop0));
398
399	kvm_guest_enter();
400
401	/* Disable hardware page table walking while in guest */
402	htw_stop();
403
404	r = __kvm_mips_vcpu_run(run, vcpu);
405
406	/* Re-enable HTW before enabling interrupts */
407	htw_start();
408
409	kvm_guest_exit();
410	local_irq_enable();
411
412	if (vcpu->sigset_active)
413		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
414
415	return r;
416}
417
418int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
419			     struct kvm_mips_interrupt *irq)
420{
421	int intr = (int)irq->irq;
422	struct kvm_vcpu *dvcpu = NULL;
423
424	if (intr == 3 || intr == -3 || intr == 4 || intr == -4)
425		kvm_debug("%s: CPU: %d, INTR: %d\n", __func__, irq->cpu,
426			  (int)intr);
427
428	if (irq->cpu == -1)
429		dvcpu = vcpu;
430	else
431		dvcpu = vcpu->kvm->vcpus[irq->cpu];
432
433	if (intr == 2 || intr == 3 || intr == 4) {
434		kvm_mips_callbacks->queue_io_int(dvcpu, irq);
435
436	} else if (intr == -2 || intr == -3 || intr == -4) {
437		kvm_mips_callbacks->dequeue_io_int(dvcpu, irq);
438	} else {
439		kvm_err("%s: invalid interrupt ioctl (%d:%d)\n", __func__,
440			irq->cpu, irq->irq);
441		return -EINVAL;
442	}
443
444	dvcpu->arch.wait = 0;
445
446	if (waitqueue_active(&dvcpu->wq))
447		wake_up_interruptible(&dvcpu->wq);
448
449	return 0;
450}
451
452int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
453				    struct kvm_mp_state *mp_state)
454{
455	return -ENOIOCTLCMD;
456}
457
458int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
459				    struct kvm_mp_state *mp_state)
460{
461	return -ENOIOCTLCMD;
462}
463
464static u64 kvm_mips_get_one_regs[] = {
465	KVM_REG_MIPS_R0,
466	KVM_REG_MIPS_R1,
467	KVM_REG_MIPS_R2,
468	KVM_REG_MIPS_R3,
469	KVM_REG_MIPS_R4,
470	KVM_REG_MIPS_R5,
471	KVM_REG_MIPS_R6,
472	KVM_REG_MIPS_R7,
473	KVM_REG_MIPS_R8,
474	KVM_REG_MIPS_R9,
475	KVM_REG_MIPS_R10,
476	KVM_REG_MIPS_R11,
477	KVM_REG_MIPS_R12,
478	KVM_REG_MIPS_R13,
479	KVM_REG_MIPS_R14,
480	KVM_REG_MIPS_R15,
481	KVM_REG_MIPS_R16,
482	KVM_REG_MIPS_R17,
483	KVM_REG_MIPS_R18,
484	KVM_REG_MIPS_R19,
485	KVM_REG_MIPS_R20,
486	KVM_REG_MIPS_R21,
487	KVM_REG_MIPS_R22,
488	KVM_REG_MIPS_R23,
489	KVM_REG_MIPS_R24,
490	KVM_REG_MIPS_R25,
491	KVM_REG_MIPS_R26,
492	KVM_REG_MIPS_R27,
493	KVM_REG_MIPS_R28,
494	KVM_REG_MIPS_R29,
495	KVM_REG_MIPS_R30,
496	KVM_REG_MIPS_R31,
497
498	KVM_REG_MIPS_HI,
499	KVM_REG_MIPS_LO,
500	KVM_REG_MIPS_PC,
501
502	KVM_REG_MIPS_CP0_INDEX,
503	KVM_REG_MIPS_CP0_CONTEXT,
504	KVM_REG_MIPS_CP0_USERLOCAL,
505	KVM_REG_MIPS_CP0_PAGEMASK,
506	KVM_REG_MIPS_CP0_WIRED,
507	KVM_REG_MIPS_CP0_HWRENA,
508	KVM_REG_MIPS_CP0_BADVADDR,
509	KVM_REG_MIPS_CP0_COUNT,
510	KVM_REG_MIPS_CP0_ENTRYHI,
511	KVM_REG_MIPS_CP0_COMPARE,
512	KVM_REG_MIPS_CP0_STATUS,
513	KVM_REG_MIPS_CP0_CAUSE,
514	KVM_REG_MIPS_CP0_EPC,
515	KVM_REG_MIPS_CP0_PRID,
516	KVM_REG_MIPS_CP0_CONFIG,
517	KVM_REG_MIPS_CP0_CONFIG1,
518	KVM_REG_MIPS_CP0_CONFIG2,
519	KVM_REG_MIPS_CP0_CONFIG3,
520	KVM_REG_MIPS_CP0_CONFIG4,
521	KVM_REG_MIPS_CP0_CONFIG5,
522	KVM_REG_MIPS_CP0_CONFIG7,
523	KVM_REG_MIPS_CP0_ERROREPC,
524
525	KVM_REG_MIPS_COUNT_CTL,
526	KVM_REG_MIPS_COUNT_RESUME,
527	KVM_REG_MIPS_COUNT_HZ,
528};
529
530static int kvm_mips_get_reg(struct kvm_vcpu *vcpu,
531			    const struct kvm_one_reg *reg)
532{
533	struct mips_coproc *cop0 = vcpu->arch.cop0;
534	struct mips_fpu_struct *fpu = &vcpu->arch.fpu;
535	int ret;
536	s64 v;
537	s64 vs[2];
538	unsigned int idx;
539
540	switch (reg->id) {
541	/* General purpose registers */
542	case KVM_REG_MIPS_R0 ... KVM_REG_MIPS_R31:
543		v = (long)vcpu->arch.gprs[reg->id - KVM_REG_MIPS_R0];
544		break;
545	case KVM_REG_MIPS_HI:
546		v = (long)vcpu->arch.hi;
547		break;
548	case KVM_REG_MIPS_LO:
549		v = (long)vcpu->arch.lo;
550		break;
551	case KVM_REG_MIPS_PC:
552		v = (long)vcpu->arch.pc;
553		break;
554
555	/* Floating point registers */
556	case KVM_REG_MIPS_FPR_32(0) ... KVM_REG_MIPS_FPR_32(31):
557		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
558			return -EINVAL;
559		idx = reg->id - KVM_REG_MIPS_FPR_32(0);
560		/* Odd singles in top of even double when FR=0 */
561		if (kvm_read_c0_guest_status(cop0) & ST0_FR)
562			v = get_fpr32(&fpu->fpr[idx], 0);
563		else
564			v = get_fpr32(&fpu->fpr[idx & ~1], idx & 1);
565		break;
566	case KVM_REG_MIPS_FPR_64(0) ... KVM_REG_MIPS_FPR_64(31):
567		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
568			return -EINVAL;
569		idx = reg->id - KVM_REG_MIPS_FPR_64(0);
570		/* Can't access odd doubles in FR=0 mode */
571		if (idx & 1 && !(kvm_read_c0_guest_status(cop0) & ST0_FR))
572			return -EINVAL;
573		v = get_fpr64(&fpu->fpr[idx], 0);
574		break;
575	case KVM_REG_MIPS_FCR_IR:
576		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
577			return -EINVAL;
578		v = boot_cpu_data.fpu_id;
579		break;
580	case KVM_REG_MIPS_FCR_CSR:
581		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
582			return -EINVAL;
583		v = fpu->fcr31;
584		break;
585
586	/* MIPS SIMD Architecture (MSA) registers */
587	case KVM_REG_MIPS_VEC_128(0) ... KVM_REG_MIPS_VEC_128(31):
588		if (!kvm_mips_guest_has_msa(&vcpu->arch))
589			return -EINVAL;
590		/* Can't access MSA registers in FR=0 mode */
591		if (!(kvm_read_c0_guest_status(cop0) & ST0_FR))
592			return -EINVAL;
593		idx = reg->id - KVM_REG_MIPS_VEC_128(0);
594#ifdef CONFIG_CPU_LITTLE_ENDIAN
595		/* least significant byte first */
596		vs[0] = get_fpr64(&fpu->fpr[idx], 0);
597		vs[1] = get_fpr64(&fpu->fpr[idx], 1);
598#else
599		/* most significant byte first */
600		vs[0] = get_fpr64(&fpu->fpr[idx], 1);
601		vs[1] = get_fpr64(&fpu->fpr[idx], 0);
602#endif
603		break;
604	case KVM_REG_MIPS_MSA_IR:
605		if (!kvm_mips_guest_has_msa(&vcpu->arch))
606			return -EINVAL;
607		v = boot_cpu_data.msa_id;
608		break;
609	case KVM_REG_MIPS_MSA_CSR:
610		if (!kvm_mips_guest_has_msa(&vcpu->arch))
611			return -EINVAL;
612		v = fpu->msacsr;
613		break;
614
615	/* Co-processor 0 registers */
616	case KVM_REG_MIPS_CP0_INDEX:
617		v = (long)kvm_read_c0_guest_index(cop0);
618		break;
619	case KVM_REG_MIPS_CP0_CONTEXT:
620		v = (long)kvm_read_c0_guest_context(cop0);
621		break;
622	case KVM_REG_MIPS_CP0_USERLOCAL:
623		v = (long)kvm_read_c0_guest_userlocal(cop0);
624		break;
625	case KVM_REG_MIPS_CP0_PAGEMASK:
626		v = (long)kvm_read_c0_guest_pagemask(cop0);
627		break;
628	case KVM_REG_MIPS_CP0_WIRED:
629		v = (long)kvm_read_c0_guest_wired(cop0);
630		break;
631	case KVM_REG_MIPS_CP0_HWRENA:
632		v = (long)kvm_read_c0_guest_hwrena(cop0);
633		break;
634	case KVM_REG_MIPS_CP0_BADVADDR:
635		v = (long)kvm_read_c0_guest_badvaddr(cop0);
636		break;
637	case KVM_REG_MIPS_CP0_ENTRYHI:
638		v = (long)kvm_read_c0_guest_entryhi(cop0);
639		break;
640	case KVM_REG_MIPS_CP0_COMPARE:
641		v = (long)kvm_read_c0_guest_compare(cop0);
642		break;
643	case KVM_REG_MIPS_CP0_STATUS:
644		v = (long)kvm_read_c0_guest_status(cop0);
645		break;
646	case KVM_REG_MIPS_CP0_CAUSE:
647		v = (long)kvm_read_c0_guest_cause(cop0);
648		break;
649	case KVM_REG_MIPS_CP0_EPC:
650		v = (long)kvm_read_c0_guest_epc(cop0);
651		break;
652	case KVM_REG_MIPS_CP0_PRID:
653		v = (long)kvm_read_c0_guest_prid(cop0);
654		break;
655	case KVM_REG_MIPS_CP0_CONFIG:
656		v = (long)kvm_read_c0_guest_config(cop0);
657		break;
658	case KVM_REG_MIPS_CP0_CONFIG1:
659		v = (long)kvm_read_c0_guest_config1(cop0);
660		break;
661	case KVM_REG_MIPS_CP0_CONFIG2:
662		v = (long)kvm_read_c0_guest_config2(cop0);
663		break;
664	case KVM_REG_MIPS_CP0_CONFIG3:
665		v = (long)kvm_read_c0_guest_config3(cop0);
666		break;
667	case KVM_REG_MIPS_CP0_CONFIG4:
668		v = (long)kvm_read_c0_guest_config4(cop0);
669		break;
670	case KVM_REG_MIPS_CP0_CONFIG5:
671		v = (long)kvm_read_c0_guest_config5(cop0);
672		break;
673	case KVM_REG_MIPS_CP0_CONFIG7:
674		v = (long)kvm_read_c0_guest_config7(cop0);
675		break;
676	case KVM_REG_MIPS_CP0_ERROREPC:
677		v = (long)kvm_read_c0_guest_errorepc(cop0);
678		break;
679	/* registers to be handled specially */
680	case KVM_REG_MIPS_CP0_COUNT:
681	case KVM_REG_MIPS_COUNT_CTL:
682	case KVM_REG_MIPS_COUNT_RESUME:
683	case KVM_REG_MIPS_COUNT_HZ:
684		ret = kvm_mips_callbacks->get_one_reg(vcpu, reg, &v);
685		if (ret)
686			return ret;
687		break;
688	default:
689		return -EINVAL;
690	}
691	if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U64) {
692		u64 __user *uaddr64 = (u64 __user *)(long)reg->addr;
693
694		return put_user(v, uaddr64);
695	} else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U32) {
696		u32 __user *uaddr32 = (u32 __user *)(long)reg->addr;
697		u32 v32 = (u32)v;
698
699		return put_user(v32, uaddr32);
700	} else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U128) {
701		void __user *uaddr = (void __user *)(long)reg->addr;
702
703		return copy_to_user(uaddr, vs, 16) ? -EFAULT : 0;
704	} else {
705		return -EINVAL;
706	}
707}
708
709static int kvm_mips_set_reg(struct kvm_vcpu *vcpu,
710			    const struct kvm_one_reg *reg)
711{
712	struct mips_coproc *cop0 = vcpu->arch.cop0;
713	struct mips_fpu_struct *fpu = &vcpu->arch.fpu;
714	s64 v;
715	s64 vs[2];
716	unsigned int idx;
717
718	if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U64) {
719		u64 __user *uaddr64 = (u64 __user *)(long)reg->addr;
720
721		if (get_user(v, uaddr64) != 0)
722			return -EFAULT;
723	} else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U32) {
724		u32 __user *uaddr32 = (u32 __user *)(long)reg->addr;
725		s32 v32;
726
727		if (get_user(v32, uaddr32) != 0)
728			return -EFAULT;
729		v = (s64)v32;
730	} else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U128) {
731		void __user *uaddr = (void __user *)(long)reg->addr;
732
733		return copy_from_user(vs, uaddr, 16) ? -EFAULT : 0;
734	} else {
735		return -EINVAL;
736	}
737
738	switch (reg->id) {
739	/* General purpose registers */
740	case KVM_REG_MIPS_R0:
741		/* Silently ignore requests to set $0 */
742		break;
743	case KVM_REG_MIPS_R1 ... KVM_REG_MIPS_R31:
744		vcpu->arch.gprs[reg->id - KVM_REG_MIPS_R0] = v;
745		break;
746	case KVM_REG_MIPS_HI:
747		vcpu->arch.hi = v;
748		break;
749	case KVM_REG_MIPS_LO:
750		vcpu->arch.lo = v;
751		break;
752	case KVM_REG_MIPS_PC:
753		vcpu->arch.pc = v;
754		break;
755
756	/* Floating point registers */
757	case KVM_REG_MIPS_FPR_32(0) ... KVM_REG_MIPS_FPR_32(31):
758		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
759			return -EINVAL;
760		idx = reg->id - KVM_REG_MIPS_FPR_32(0);
761		/* Odd singles in top of even double when FR=0 */
762		if (kvm_read_c0_guest_status(cop0) & ST0_FR)
763			set_fpr32(&fpu->fpr[idx], 0, v);
764		else
765			set_fpr32(&fpu->fpr[idx & ~1], idx & 1, v);
766		break;
767	case KVM_REG_MIPS_FPR_64(0) ... KVM_REG_MIPS_FPR_64(31):
768		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
769			return -EINVAL;
770		idx = reg->id - KVM_REG_MIPS_FPR_64(0);
771		/* Can't access odd doubles in FR=0 mode */
772		if (idx & 1 && !(kvm_read_c0_guest_status(cop0) & ST0_FR))
773			return -EINVAL;
774		set_fpr64(&fpu->fpr[idx], 0, v);
775		break;
776	case KVM_REG_MIPS_FCR_IR:
777		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
778			return -EINVAL;
779		/* Read-only */
780		break;
781	case KVM_REG_MIPS_FCR_CSR:
782		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
783			return -EINVAL;
784		fpu->fcr31 = v;
785		break;
786
787	/* MIPS SIMD Architecture (MSA) registers */
788	case KVM_REG_MIPS_VEC_128(0) ... KVM_REG_MIPS_VEC_128(31):
789		if (!kvm_mips_guest_has_msa(&vcpu->arch))
790			return -EINVAL;
791		idx = reg->id - KVM_REG_MIPS_VEC_128(0);
792#ifdef CONFIG_CPU_LITTLE_ENDIAN
793		/* least significant byte first */
794		set_fpr64(&fpu->fpr[idx], 0, vs[0]);
795		set_fpr64(&fpu->fpr[idx], 1, vs[1]);
796#else
797		/* most significant byte first */
798		set_fpr64(&fpu->fpr[idx], 1, vs[0]);
799		set_fpr64(&fpu->fpr[idx], 0, vs[1]);
800#endif
801		break;
802	case KVM_REG_MIPS_MSA_IR:
803		if (!kvm_mips_guest_has_msa(&vcpu->arch))
804			return -EINVAL;
805		/* Read-only */
806		break;
807	case KVM_REG_MIPS_MSA_CSR:
808		if (!kvm_mips_guest_has_msa(&vcpu->arch))
809			return -EINVAL;
810		fpu->msacsr = v;
811		break;
812
813	/* Co-processor 0 registers */
814	case KVM_REG_MIPS_CP0_INDEX:
815		kvm_write_c0_guest_index(cop0, v);
816		break;
817	case KVM_REG_MIPS_CP0_CONTEXT:
818		kvm_write_c0_guest_context(cop0, v);
819		break;
820	case KVM_REG_MIPS_CP0_USERLOCAL:
821		kvm_write_c0_guest_userlocal(cop0, v);
822		break;
823	case KVM_REG_MIPS_CP0_PAGEMASK:
824		kvm_write_c0_guest_pagemask(cop0, v);
825		break;
826	case KVM_REG_MIPS_CP0_WIRED:
827		kvm_write_c0_guest_wired(cop0, v);
828		break;
829	case KVM_REG_MIPS_CP0_HWRENA:
830		kvm_write_c0_guest_hwrena(cop0, v);
831		break;
832	case KVM_REG_MIPS_CP0_BADVADDR:
833		kvm_write_c0_guest_badvaddr(cop0, v);
834		break;
835	case KVM_REG_MIPS_CP0_ENTRYHI:
836		kvm_write_c0_guest_entryhi(cop0, v);
837		break;
838	case KVM_REG_MIPS_CP0_STATUS:
839		kvm_write_c0_guest_status(cop0, v);
840		break;
841	case KVM_REG_MIPS_CP0_EPC:
842		kvm_write_c0_guest_epc(cop0, v);
843		break;
844	case KVM_REG_MIPS_CP0_PRID:
845		kvm_write_c0_guest_prid(cop0, v);
846		break;
847	case KVM_REG_MIPS_CP0_ERROREPC:
848		kvm_write_c0_guest_errorepc(cop0, v);
849		break;
850	/* registers to be handled specially */
851	case KVM_REG_MIPS_CP0_COUNT:
852	case KVM_REG_MIPS_CP0_COMPARE:
853	case KVM_REG_MIPS_CP0_CAUSE:
854	case KVM_REG_MIPS_CP0_CONFIG:
855	case KVM_REG_MIPS_CP0_CONFIG1:
856	case KVM_REG_MIPS_CP0_CONFIG2:
857	case KVM_REG_MIPS_CP0_CONFIG3:
858	case KVM_REG_MIPS_CP0_CONFIG4:
859	case KVM_REG_MIPS_CP0_CONFIG5:
860	case KVM_REG_MIPS_COUNT_CTL:
861	case KVM_REG_MIPS_COUNT_RESUME:
862	case KVM_REG_MIPS_COUNT_HZ:
863		return kvm_mips_callbacks->set_one_reg(vcpu, reg, v);
864	default:
865		return -EINVAL;
866	}
867	return 0;
868}
869
870static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
871				     struct kvm_enable_cap *cap)
872{
873	int r = 0;
874
875	if (!kvm_vm_ioctl_check_extension(vcpu->kvm, cap->cap))
876		return -EINVAL;
877	if (cap->flags)
878		return -EINVAL;
879	if (cap->args[0])
880		return -EINVAL;
881
882	switch (cap->cap) {
883	case KVM_CAP_MIPS_FPU:
884		vcpu->arch.fpu_enabled = true;
885		break;
886	case KVM_CAP_MIPS_MSA:
887		vcpu->arch.msa_enabled = true;
888		break;
889	default:
890		r = -EINVAL;
891		break;
892	}
893
894	return r;
895}
896
897long kvm_arch_vcpu_ioctl(struct file *filp, unsigned int ioctl,
898			 unsigned long arg)
899{
900	struct kvm_vcpu *vcpu = filp->private_data;
901	void __user *argp = (void __user *)arg;
902	long r;
903
904	switch (ioctl) {
905	case KVM_SET_ONE_REG:
906	case KVM_GET_ONE_REG: {
907		struct kvm_one_reg reg;
908
909		if (copy_from_user(&reg, argp, sizeof(reg)))
910			return -EFAULT;
911		if (ioctl == KVM_SET_ONE_REG)
912			return kvm_mips_set_reg(vcpu, &reg);
913		else
914			return kvm_mips_get_reg(vcpu, &reg);
915	}
916	case KVM_GET_REG_LIST: {
917		struct kvm_reg_list __user *user_list = argp;
918		u64 __user *reg_dest;
919		struct kvm_reg_list reg_list;
920		unsigned n;
921
922		if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
923			return -EFAULT;
924		n = reg_list.n;
925		reg_list.n = ARRAY_SIZE(kvm_mips_get_one_regs);
926		if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
927			return -EFAULT;
928		if (n < reg_list.n)
929			return -E2BIG;
930		reg_dest = user_list->reg;
931		if (copy_to_user(reg_dest, kvm_mips_get_one_regs,
932				 sizeof(kvm_mips_get_one_regs)))
933			return -EFAULT;
934		return 0;
935	}
936	case KVM_NMI:
937		/* Treat the NMI as a CPU reset */
938		r = kvm_mips_reset_vcpu(vcpu);
939		break;
940	case KVM_INTERRUPT:
941		{
942			struct kvm_mips_interrupt irq;
943
944			r = -EFAULT;
945			if (copy_from_user(&irq, argp, sizeof(irq)))
946				goto out;
947
948			kvm_debug("[%d] %s: irq: %d\n", vcpu->vcpu_id, __func__,
949				  irq.irq);
950
951			r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
952			break;
953		}
954	case KVM_ENABLE_CAP: {
955		struct kvm_enable_cap cap;
956
957		r = -EFAULT;
958		if (copy_from_user(&cap, argp, sizeof(cap)))
959			goto out;
960		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
961		break;
962	}
963	default:
964		r = -ENOIOCTLCMD;
965	}
966
967out:
968	return r;
969}
970
971/* Get (and clear) the dirty memory log for a memory slot. */
972int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
973{
974	struct kvm_memory_slot *memslot;
975	unsigned long ga, ga_end;
976	int is_dirty = 0;
977	int r;
978	unsigned long n;
979
980	mutex_lock(&kvm->slots_lock);
981
982	r = kvm_get_dirty_log(kvm, log, &is_dirty);
983	if (r)
984		goto out;
985
986	/* If nothing is dirty, don't bother messing with page tables. */
987	if (is_dirty) {
988		memslot = id_to_memslot(kvm->memslots, log->slot);
989
990		ga = memslot->base_gfn << PAGE_SHIFT;
991		ga_end = ga + (memslot->npages << PAGE_SHIFT);
992
993		kvm_info("%s: dirty, ga: %#lx, ga_end %#lx\n", __func__, ga,
994			 ga_end);
995
996		n = kvm_dirty_bitmap_bytes(memslot);
997		memset(memslot->dirty_bitmap, 0, n);
998	}
999
1000	r = 0;
1001out:
1002	mutex_unlock(&kvm->slots_lock);
1003	return r;
1004
1005}
1006
1007long kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
1008{
1009	long r;
1010
1011	switch (ioctl) {
1012	default:
1013		r = -ENOIOCTLCMD;
1014	}
1015
1016	return r;
1017}
1018
1019int kvm_arch_init(void *opaque)
1020{
1021	if (kvm_mips_callbacks) {
1022		kvm_err("kvm: module already exists\n");
1023		return -EEXIST;
1024	}
1025
1026	return kvm_mips_emulation_init(&kvm_mips_callbacks);
1027}
1028
1029void kvm_arch_exit(void)
1030{
1031	kvm_mips_callbacks = NULL;
1032}
1033
1034int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1035				  struct kvm_sregs *sregs)
1036{
1037	return -ENOIOCTLCMD;
1038}
1039
1040int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1041				  struct kvm_sregs *sregs)
1042{
1043	return -ENOIOCTLCMD;
1044}
1045
1046void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
1047{
1048}
1049
1050int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1051{
1052	return -ENOIOCTLCMD;
1053}
1054
1055int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1056{
1057	return -ENOIOCTLCMD;
1058}
1059
1060int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
1061{
1062	return VM_FAULT_SIGBUS;
1063}
1064
1065int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
1066{
1067	int r;
1068
1069	switch (ext) {
1070	case KVM_CAP_ONE_REG:
1071	case KVM_CAP_ENABLE_CAP:
1072		r = 1;
1073		break;
1074	case KVM_CAP_COALESCED_MMIO:
1075		r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1076		break;
1077	case KVM_CAP_MIPS_FPU:
1078		r = !!cpu_has_fpu;
1079		break;
1080	case KVM_CAP_MIPS_MSA:
1081		/*
1082		 * We don't support MSA vector partitioning yet:
1083		 * 1) It would require explicit support which can't be tested
1084		 *    yet due to lack of support in current hardware.
1085		 * 2) It extends the state that would need to be saved/restored
1086		 *    by e.g. QEMU for migration.
1087		 *
1088		 * When vector partitioning hardware becomes available, support
1089		 * could be added by requiring a flag when enabling
1090		 * KVM_CAP_MIPS_MSA capability to indicate that userland knows
1091		 * to save/restore the appropriate extra state.
1092		 */
1093		r = cpu_has_msa && !(boot_cpu_data.msa_id & MSA_IR_WRPF);
1094		break;
1095	default:
1096		r = 0;
1097		break;
1098	}
1099	return r;
1100}
1101
1102int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1103{
1104	return kvm_mips_pending_timer(vcpu);
1105}
1106
1107int kvm_arch_vcpu_dump_regs(struct kvm_vcpu *vcpu)
1108{
1109	int i;
1110	struct mips_coproc *cop0;
1111
1112	if (!vcpu)
1113		return -1;
1114
1115	kvm_debug("VCPU Register Dump:\n");
1116	kvm_debug("\tpc = 0x%08lx\n", vcpu->arch.pc);
1117	kvm_debug("\texceptions: %08lx\n", vcpu->arch.pending_exceptions);
1118
1119	for (i = 0; i < 32; i += 4) {
1120		kvm_debug("\tgpr%02d: %08lx %08lx %08lx %08lx\n", i,
1121		       vcpu->arch.gprs[i],
1122		       vcpu->arch.gprs[i + 1],
1123		       vcpu->arch.gprs[i + 2], vcpu->arch.gprs[i + 3]);
1124	}
1125	kvm_debug("\thi: 0x%08lx\n", vcpu->arch.hi);
1126	kvm_debug("\tlo: 0x%08lx\n", vcpu->arch.lo);
1127
1128	cop0 = vcpu->arch.cop0;
1129	kvm_debug("\tStatus: 0x%08lx, Cause: 0x%08lx\n",
1130		  kvm_read_c0_guest_status(cop0),
1131		  kvm_read_c0_guest_cause(cop0));
1132
1133	kvm_debug("\tEPC: 0x%08lx\n", kvm_read_c0_guest_epc(cop0));
1134
1135	return 0;
1136}
1137
1138int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1139{
1140	int i;
1141
1142	for (i = 1; i < ARRAY_SIZE(vcpu->arch.gprs); i++)
1143		vcpu->arch.gprs[i] = regs->gpr[i];
1144	vcpu->arch.gprs[0] = 0; /* zero is special, and cannot be set. */
1145	vcpu->arch.hi = regs->hi;
1146	vcpu->arch.lo = regs->lo;
1147	vcpu->arch.pc = regs->pc;
1148
1149	return 0;
1150}
1151
1152int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1153{
1154	int i;
1155
1156	for (i = 0; i < ARRAY_SIZE(vcpu->arch.gprs); i++)
1157		regs->gpr[i] = vcpu->arch.gprs[i];
1158
1159	regs->hi = vcpu->arch.hi;
1160	regs->lo = vcpu->arch.lo;
1161	regs->pc = vcpu->arch.pc;
1162
1163	return 0;
1164}
1165
1166static void kvm_mips_comparecount_func(unsigned long data)
1167{
1168	struct kvm_vcpu *vcpu = (struct kvm_vcpu *)data;
1169
1170	kvm_mips_callbacks->queue_timer_int(vcpu);
1171
1172	vcpu->arch.wait = 0;
1173	if (waitqueue_active(&vcpu->wq))
1174		wake_up_interruptible(&vcpu->wq);
1175}
1176
1177/* low level hrtimer wake routine */
1178static enum hrtimer_restart kvm_mips_comparecount_wakeup(struct hrtimer *timer)
1179{
1180	struct kvm_vcpu *vcpu;
1181
1182	vcpu = container_of(timer, struct kvm_vcpu, arch.comparecount_timer);
1183	kvm_mips_comparecount_func((unsigned long) vcpu);
1184	return kvm_mips_count_timeout(vcpu);
1185}
1186
1187int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
1188{
1189	kvm_mips_callbacks->vcpu_init(vcpu);
1190	hrtimer_init(&vcpu->arch.comparecount_timer, CLOCK_MONOTONIC,
1191		     HRTIMER_MODE_REL);
1192	vcpu->arch.comparecount_timer.function = kvm_mips_comparecount_wakeup;
1193	return 0;
1194}
1195
1196int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1197				  struct kvm_translation *tr)
1198{
1199	return 0;
1200}
1201
1202/* Initial guest state */
1203int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
1204{
1205	return kvm_mips_callbacks->vcpu_setup(vcpu);
1206}
1207
1208static void kvm_mips_set_c0_status(void)
1209{
1210	uint32_t status = read_c0_status();
1211
1212	if (cpu_has_dsp)
1213		status |= (ST0_MX);
1214
1215	write_c0_status(status);
1216	ehb();
1217}
1218
1219/*
1220 * Return value is in the form (errcode<<2 | RESUME_FLAG_HOST | RESUME_FLAG_NV)
1221 */
1222int kvm_mips_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu)
1223{
1224	uint32_t cause = vcpu->arch.host_cp0_cause;
1225	uint32_t exccode = (cause >> CAUSEB_EXCCODE) & 0x1f;
1226	uint32_t __user *opc = (uint32_t __user *) vcpu->arch.pc;
1227	unsigned long badvaddr = vcpu->arch.host_cp0_badvaddr;
1228	enum emulation_result er = EMULATE_DONE;
1229	int ret = RESUME_GUEST;
1230
1231	/* re-enable HTW before enabling interrupts */
1232	htw_start();
1233
1234	/* Set a default exit reason */
1235	run->exit_reason = KVM_EXIT_UNKNOWN;
1236	run->ready_for_interrupt_injection = 1;
1237
1238	/*
1239	 * Set the appropriate status bits based on host CPU features,
1240	 * before we hit the scheduler
1241	 */
1242	kvm_mips_set_c0_status();
1243
1244	local_irq_enable();
1245
1246	kvm_debug("kvm_mips_handle_exit: cause: %#x, PC: %p, kvm_run: %p, kvm_vcpu: %p\n",
1247			cause, opc, run, vcpu);
1248
1249	/*
1250	 * Do a privilege check, if in UM most of these exit conditions end up
1251	 * causing an exception to be delivered to the Guest Kernel
1252	 */
1253	er = kvm_mips_check_privilege(cause, opc, run, vcpu);
1254	if (er == EMULATE_PRIV_FAIL) {
1255		goto skip_emul;
1256	} else if (er == EMULATE_FAIL) {
1257		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1258		ret = RESUME_HOST;
1259		goto skip_emul;
1260	}
1261
1262	switch (exccode) {
1263	case T_INT:
1264		kvm_debug("[%d]T_INT @ %p\n", vcpu->vcpu_id, opc);
1265
1266		++vcpu->stat.int_exits;
1267		trace_kvm_exit(vcpu, INT_EXITS);
1268
1269		if (need_resched())
1270			cond_resched();
1271
1272		ret = RESUME_GUEST;
1273		break;
1274
1275	case T_COP_UNUSABLE:
1276		kvm_debug("T_COP_UNUSABLE: @ PC: %p\n", opc);
1277
1278		++vcpu->stat.cop_unusable_exits;
1279		trace_kvm_exit(vcpu, COP_UNUSABLE_EXITS);
1280		ret = kvm_mips_callbacks->handle_cop_unusable(vcpu);
1281		/* XXXKYMA: Might need to return to user space */
1282		if (run->exit_reason == KVM_EXIT_IRQ_WINDOW_OPEN)
1283			ret = RESUME_HOST;
1284		break;
1285
1286	case T_TLB_MOD:
1287		++vcpu->stat.tlbmod_exits;
1288		trace_kvm_exit(vcpu, TLBMOD_EXITS);
1289		ret = kvm_mips_callbacks->handle_tlb_mod(vcpu);
1290		break;
1291
1292	case T_TLB_ST_MISS:
1293		kvm_debug("TLB ST fault:  cause %#x, status %#lx, PC: %p, BadVaddr: %#lx\n",
1294			  cause, kvm_read_c0_guest_status(vcpu->arch.cop0), opc,
1295			  badvaddr);
1296
1297		++vcpu->stat.tlbmiss_st_exits;
1298		trace_kvm_exit(vcpu, TLBMISS_ST_EXITS);
1299		ret = kvm_mips_callbacks->handle_tlb_st_miss(vcpu);
1300		break;
1301
1302	case T_TLB_LD_MISS:
1303		kvm_debug("TLB LD fault: cause %#x, PC: %p, BadVaddr: %#lx\n",
1304			  cause, opc, badvaddr);
1305
1306		++vcpu->stat.tlbmiss_ld_exits;
1307		trace_kvm_exit(vcpu, TLBMISS_LD_EXITS);
1308		ret = kvm_mips_callbacks->handle_tlb_ld_miss(vcpu);
1309		break;
1310
1311	case T_ADDR_ERR_ST:
1312		++vcpu->stat.addrerr_st_exits;
1313		trace_kvm_exit(vcpu, ADDRERR_ST_EXITS);
1314		ret = kvm_mips_callbacks->handle_addr_err_st(vcpu);
1315		break;
1316
1317	case T_ADDR_ERR_LD:
1318		++vcpu->stat.addrerr_ld_exits;
1319		trace_kvm_exit(vcpu, ADDRERR_LD_EXITS);
1320		ret = kvm_mips_callbacks->handle_addr_err_ld(vcpu);
1321		break;
1322
1323	case T_SYSCALL:
1324		++vcpu->stat.syscall_exits;
1325		trace_kvm_exit(vcpu, SYSCALL_EXITS);
1326		ret = kvm_mips_callbacks->handle_syscall(vcpu);
1327		break;
1328
1329	case T_RES_INST:
1330		++vcpu->stat.resvd_inst_exits;
1331		trace_kvm_exit(vcpu, RESVD_INST_EXITS);
1332		ret = kvm_mips_callbacks->handle_res_inst(vcpu);
1333		break;
1334
1335	case T_BREAK:
1336		++vcpu->stat.break_inst_exits;
1337		trace_kvm_exit(vcpu, BREAK_INST_EXITS);
1338		ret = kvm_mips_callbacks->handle_break(vcpu);
1339		break;
1340
1341	case T_TRAP:
1342		++vcpu->stat.trap_inst_exits;
1343		trace_kvm_exit(vcpu, TRAP_INST_EXITS);
1344		ret = kvm_mips_callbacks->handle_trap(vcpu);
1345		break;
1346
1347	case T_MSAFPE:
1348		++vcpu->stat.msa_fpe_exits;
1349		trace_kvm_exit(vcpu, MSA_FPE_EXITS);
1350		ret = kvm_mips_callbacks->handle_msa_fpe(vcpu);
1351		break;
1352
1353	case T_FPE:
1354		++vcpu->stat.fpe_exits;
1355		trace_kvm_exit(vcpu, FPE_EXITS);
1356		ret = kvm_mips_callbacks->handle_fpe(vcpu);
1357		break;
1358
1359	case T_MSADIS:
1360		++vcpu->stat.msa_disabled_exits;
1361		trace_kvm_exit(vcpu, MSA_DISABLED_EXITS);
1362		ret = kvm_mips_callbacks->handle_msa_disabled(vcpu);
1363		break;
1364
1365	default:
1366		kvm_err("Exception Code: %d, not yet handled, @ PC: %p, inst: 0x%08x  BadVaddr: %#lx Status: %#lx\n",
1367			exccode, opc, kvm_get_inst(opc, vcpu), badvaddr,
1368			kvm_read_c0_guest_status(vcpu->arch.cop0));
1369		kvm_arch_vcpu_dump_regs(vcpu);
1370		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1371		ret = RESUME_HOST;
1372		break;
1373
1374	}
1375
1376skip_emul:
1377	local_irq_disable();
1378
1379	if (er == EMULATE_DONE && !(ret & RESUME_HOST))
1380		kvm_mips_deliver_interrupts(vcpu, cause);
1381
1382	if (!(ret & RESUME_HOST)) {
1383		/* Only check for signals if not already exiting to userspace */
1384		if (signal_pending(current)) {
1385			run->exit_reason = KVM_EXIT_INTR;
1386			ret = (-EINTR << 2) | RESUME_HOST;
1387			++vcpu->stat.signal_exits;
1388			trace_kvm_exit(vcpu, SIGNAL_EXITS);
1389		}
1390	}
1391
1392	if (ret == RESUME_GUEST) {
1393		/*
1394		 * If FPU / MSA are enabled (i.e. the guest's FPU / MSA context
1395		 * is live), restore FCR31 / MSACSR.
1396		 *
1397		 * This should be before returning to the guest exception
1398		 * vector, as it may well cause an [MSA] FP exception if there
1399		 * are pending exception bits unmasked. (see
1400		 * kvm_mips_csr_die_notifier() for how that is handled).
1401		 */
1402		if (kvm_mips_guest_has_fpu(&vcpu->arch) &&
1403		    read_c0_status() & ST0_CU1)
1404			__kvm_restore_fcsr(&vcpu->arch);
1405
1406		if (kvm_mips_guest_has_msa(&vcpu->arch) &&
1407		    read_c0_config5() & MIPS_CONF5_MSAEN)
1408			__kvm_restore_msacsr(&vcpu->arch);
1409	}
1410
1411	/* Disable HTW before returning to guest or host */
1412	htw_stop();
1413
1414	return ret;
1415}
1416
1417/* Enable FPU for guest and restore context */
1418void kvm_own_fpu(struct kvm_vcpu *vcpu)
1419{
1420	struct mips_coproc *cop0 = vcpu->arch.cop0;
1421	unsigned int sr, cfg5;
1422
1423	preempt_disable();
1424
1425	sr = kvm_read_c0_guest_status(cop0);
1426
1427	/*
1428	 * If MSA state is already live, it is undefined how it interacts with
1429	 * FR=0 FPU state, and we don't want to hit reserved instruction
1430	 * exceptions trying to save the MSA state later when CU=1 && FR=1, so
1431	 * play it safe and save it first.
1432	 *
1433	 * In theory we shouldn't ever hit this case since kvm_lose_fpu() should
1434	 * get called when guest CU1 is set, however we can't trust the guest
1435	 * not to clobber the status register directly via the commpage.
1436	 */
1437	if (cpu_has_msa && sr & ST0_CU1 && !(sr & ST0_FR) &&
1438	    vcpu->arch.fpu_inuse & KVM_MIPS_FPU_MSA)
1439		kvm_lose_fpu(vcpu);
1440
1441	/*
1442	 * Enable FPU for guest
1443	 * We set FR and FRE according to guest context
1444	 */
1445	change_c0_status(ST0_CU1 | ST0_FR, sr);
1446	if (cpu_has_fre) {
1447		cfg5 = kvm_read_c0_guest_config5(cop0);
1448		change_c0_config5(MIPS_CONF5_FRE, cfg5);
1449	}
1450	enable_fpu_hazard();
1451
1452	/* If guest FPU state not active, restore it now */
1453	if (!(vcpu->arch.fpu_inuse & KVM_MIPS_FPU_FPU)) {
1454		__kvm_restore_fpu(&vcpu->arch);
1455		vcpu->arch.fpu_inuse |= KVM_MIPS_FPU_FPU;
1456	}
1457
1458	preempt_enable();
1459}
1460
1461#ifdef CONFIG_CPU_HAS_MSA
1462/* Enable MSA for guest and restore context */
1463void kvm_own_msa(struct kvm_vcpu *vcpu)
1464{
1465	struct mips_coproc *cop0 = vcpu->arch.cop0;
1466	unsigned int sr, cfg5;
1467
1468	preempt_disable();
1469
1470	/*
1471	 * Enable FPU if enabled in guest, since we're restoring FPU context
1472	 * anyway. We set FR and FRE according to guest context.
1473	 */
1474	if (kvm_mips_guest_has_fpu(&vcpu->arch)) {
1475		sr = kvm_read_c0_guest_status(cop0);
1476
1477		/*
1478		 * If FR=0 FPU state is already live, it is undefined how it
1479		 * interacts with MSA state, so play it safe and save it first.
1480		 */
1481		if (!(sr & ST0_FR) &&
1482		    (vcpu->arch.fpu_inuse & (KVM_MIPS_FPU_FPU |
1483				KVM_MIPS_FPU_MSA)) == KVM_MIPS_FPU_FPU)
1484			kvm_lose_fpu(vcpu);
1485
1486		change_c0_status(ST0_CU1 | ST0_FR, sr);
1487		if (sr & ST0_CU1 && cpu_has_fre) {
1488			cfg5 = kvm_read_c0_guest_config5(cop0);
1489			change_c0_config5(MIPS_CONF5_FRE, cfg5);
1490		}
1491	}
1492
1493	/* Enable MSA for guest */
1494	set_c0_config5(MIPS_CONF5_MSAEN);
1495	enable_fpu_hazard();
1496
1497	switch (vcpu->arch.fpu_inuse & (KVM_MIPS_FPU_FPU | KVM_MIPS_FPU_MSA)) {
1498	case KVM_MIPS_FPU_FPU:
1499		/*
1500		 * Guest FPU state already loaded, only restore upper MSA state
1501		 */
1502		__kvm_restore_msa_upper(&vcpu->arch);
1503		vcpu->arch.fpu_inuse |= KVM_MIPS_FPU_MSA;
1504		break;
1505	case 0:
1506		/* Neither FPU or MSA already active, restore full MSA state */
1507		__kvm_restore_msa(&vcpu->arch);
1508		vcpu->arch.fpu_inuse |= KVM_MIPS_FPU_MSA;
1509		if (kvm_mips_guest_has_fpu(&vcpu->arch))
1510			vcpu->arch.fpu_inuse |= KVM_MIPS_FPU_FPU;
1511		break;
1512	default:
1513		break;
1514	}
1515
1516	preempt_enable();
1517}
1518#endif
1519
1520/* Drop FPU & MSA without saving it */
1521void kvm_drop_fpu(struct kvm_vcpu *vcpu)
1522{
1523	preempt_disable();
1524	if (cpu_has_msa && vcpu->arch.fpu_inuse & KVM_MIPS_FPU_MSA) {
1525		disable_msa();
1526		vcpu->arch.fpu_inuse &= ~KVM_MIPS_FPU_MSA;
1527	}
1528	if (vcpu->arch.fpu_inuse & KVM_MIPS_FPU_FPU) {
1529		clear_c0_status(ST0_CU1 | ST0_FR);
1530		vcpu->arch.fpu_inuse &= ~KVM_MIPS_FPU_FPU;
1531	}
1532	preempt_enable();
1533}
1534
1535/* Save and disable FPU & MSA */
1536void kvm_lose_fpu(struct kvm_vcpu *vcpu)
1537{
1538	/*
1539	 * FPU & MSA get disabled in root context (hardware) when it is disabled
1540	 * in guest context (software), but the register state in the hardware
1541	 * may still be in use. This is why we explicitly re-enable the hardware
1542	 * before saving.
1543	 */
1544
1545	preempt_disable();
1546	if (cpu_has_msa && vcpu->arch.fpu_inuse & KVM_MIPS_FPU_MSA) {
1547		set_c0_config5(MIPS_CONF5_MSAEN);
1548		enable_fpu_hazard();
1549
1550		__kvm_save_msa(&vcpu->arch);
1551
1552		/* Disable MSA & FPU */
1553		disable_msa();
1554		if (vcpu->arch.fpu_inuse & KVM_MIPS_FPU_FPU)
1555			clear_c0_status(ST0_CU1 | ST0_FR);
1556		vcpu->arch.fpu_inuse &= ~(KVM_MIPS_FPU_FPU | KVM_MIPS_FPU_MSA);
1557	} else if (vcpu->arch.fpu_inuse & KVM_MIPS_FPU_FPU) {
1558		set_c0_status(ST0_CU1);
1559		enable_fpu_hazard();
1560
1561		__kvm_save_fpu(&vcpu->arch);
1562		vcpu->arch.fpu_inuse &= ~KVM_MIPS_FPU_FPU;
1563
1564		/* Disable FPU */
1565		clear_c0_status(ST0_CU1 | ST0_FR);
1566	}
1567	preempt_enable();
1568}
1569
1570/*
1571 * Step over a specific ctc1 to FCSR and a specific ctcmsa to MSACSR which are
1572 * used to restore guest FCSR/MSACSR state and may trigger a "harmless" FP/MSAFP
1573 * exception if cause bits are set in the value being written.
1574 */
1575static int kvm_mips_csr_die_notify(struct notifier_block *self,
1576				   unsigned long cmd, void *ptr)
1577{
1578	struct die_args *args = (struct die_args *)ptr;
1579	struct pt_regs *regs = args->regs;
1580	unsigned long pc;
1581
1582	/* Only interested in FPE and MSAFPE */
1583	if (cmd != DIE_FP && cmd != DIE_MSAFP)
1584		return NOTIFY_DONE;
1585
1586	/* Return immediately if guest context isn't active */
1587	if (!(current->flags & PF_VCPU))
1588		return NOTIFY_DONE;
1589
1590	/* Should never get here from user mode */
1591	BUG_ON(user_mode(regs));
1592
1593	pc = instruction_pointer(regs);
1594	switch (cmd) {
1595	case DIE_FP:
1596		/* match 2nd instruction in __kvm_restore_fcsr */
1597		if (pc != (unsigned long)&__kvm_restore_fcsr + 4)
1598			return NOTIFY_DONE;
1599		break;
1600	case DIE_MSAFP:
1601		/* match 2nd/3rd instruction in __kvm_restore_msacsr */
1602		if (!cpu_has_msa ||
1603		    pc < (unsigned long)&__kvm_restore_msacsr + 4 ||
1604		    pc > (unsigned long)&__kvm_restore_msacsr + 8)
1605			return NOTIFY_DONE;
1606		break;
1607	}
1608
1609	/* Move PC forward a little and continue executing */
1610	instruction_pointer(regs) += 4;
1611
1612	return NOTIFY_STOP;
1613}
1614
1615static struct notifier_block kvm_mips_csr_die_notifier = {
1616	.notifier_call = kvm_mips_csr_die_notify,
1617};
1618
1619int __init kvm_mips_init(void)
1620{
1621	int ret;
1622
1623	ret = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1624
1625	if (ret)
1626		return ret;
1627
1628	register_die_notifier(&kvm_mips_csr_die_notifier);
1629
1630	/*
1631	 * On MIPS, kernel modules are executed from "mapped space", which
1632	 * requires TLBs. The TLB handling code is statically linked with
1633	 * the rest of the kernel (tlb.c) to avoid the possibility of
1634	 * double faulting. The issue is that the TLB code references
1635	 * routines that are part of the the KVM module, which are only
1636	 * available once the module is loaded.
1637	 */
1638	kvm_mips_gfn_to_pfn = gfn_to_pfn;
1639	kvm_mips_release_pfn_clean = kvm_release_pfn_clean;
1640	kvm_mips_is_error_pfn = is_error_pfn;
1641
1642	return 0;
1643}
1644
1645void __exit kvm_mips_exit(void)
1646{
1647	kvm_exit();
1648
1649	kvm_mips_gfn_to_pfn = NULL;
1650	kvm_mips_release_pfn_clean = NULL;
1651	kvm_mips_is_error_pfn = NULL;
1652
1653	unregister_die_notifier(&kvm_mips_csr_die_notifier);
1654}
1655
1656module_init(kvm_mips_init);
1657module_exit(kvm_mips_exit);
1658
1659EXPORT_TRACEPOINT_SYMBOL(kvm_exit);
1660