xref: /openbmc/linux/arch/mips/kvm/mips.c (revision 41e0d491)
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/bitops.h>
13 #include <linux/errno.h>
14 #include <linux/err.h>
15 #include <linux/kdebug.h>
16 #include <linux/module.h>
17 #include <linux/uaccess.h>
18 #include <linux/vmalloc.h>
19 #include <linux/sched/signal.h>
20 #include <linux/fs.h>
21 #include <linux/memblock.h>
22 #include <linux/pgtable.h>
23 
24 #include <asm/fpu.h>
25 #include <asm/page.h>
26 #include <asm/cacheflush.h>
27 #include <asm/mmu_context.h>
28 #include <asm/pgalloc.h>
29 
30 #include <linux/kvm_host.h>
31 
32 #include "interrupt.h"
33 
34 #define CREATE_TRACE_POINTS
35 #include "trace.h"
36 
37 #ifndef VECTORSPACING
38 #define VECTORSPACING 0x100	/* for EI/VI mode */
39 #endif
40 
41 const struct _kvm_stats_desc kvm_vm_stats_desc[] = {
42 	KVM_GENERIC_VM_STATS()
43 };
44 
45 const struct kvm_stats_header kvm_vm_stats_header = {
46 	.name_size = KVM_STATS_NAME_SIZE,
47 	.num_desc = ARRAY_SIZE(kvm_vm_stats_desc),
48 	.id_offset = sizeof(struct kvm_stats_header),
49 	.desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
50 	.data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
51 		       sizeof(kvm_vm_stats_desc),
52 };
53 
54 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
55 	KVM_GENERIC_VCPU_STATS(),
56 	STATS_DESC_COUNTER(VCPU, wait_exits),
57 	STATS_DESC_COUNTER(VCPU, cache_exits),
58 	STATS_DESC_COUNTER(VCPU, signal_exits),
59 	STATS_DESC_COUNTER(VCPU, int_exits),
60 	STATS_DESC_COUNTER(VCPU, cop_unusable_exits),
61 	STATS_DESC_COUNTER(VCPU, tlbmod_exits),
62 	STATS_DESC_COUNTER(VCPU, tlbmiss_ld_exits),
63 	STATS_DESC_COUNTER(VCPU, tlbmiss_st_exits),
64 	STATS_DESC_COUNTER(VCPU, addrerr_st_exits),
65 	STATS_DESC_COUNTER(VCPU, addrerr_ld_exits),
66 	STATS_DESC_COUNTER(VCPU, syscall_exits),
67 	STATS_DESC_COUNTER(VCPU, resvd_inst_exits),
68 	STATS_DESC_COUNTER(VCPU, break_inst_exits),
69 	STATS_DESC_COUNTER(VCPU, trap_inst_exits),
70 	STATS_DESC_COUNTER(VCPU, msa_fpe_exits),
71 	STATS_DESC_COUNTER(VCPU, fpe_exits),
72 	STATS_DESC_COUNTER(VCPU, msa_disabled_exits),
73 	STATS_DESC_COUNTER(VCPU, flush_dcache_exits),
74 	STATS_DESC_COUNTER(VCPU, vz_gpsi_exits),
75 	STATS_DESC_COUNTER(VCPU, vz_gsfc_exits),
76 	STATS_DESC_COUNTER(VCPU, vz_hc_exits),
77 	STATS_DESC_COUNTER(VCPU, vz_grr_exits),
78 	STATS_DESC_COUNTER(VCPU, vz_gva_exits),
79 	STATS_DESC_COUNTER(VCPU, vz_ghfc_exits),
80 	STATS_DESC_COUNTER(VCPU, vz_gpa_exits),
81 	STATS_DESC_COUNTER(VCPU, vz_resvd_exits),
82 #ifdef CONFIG_CPU_LOONGSON64
83 	STATS_DESC_COUNTER(VCPU, vz_cpucfg_exits),
84 #endif
85 };
86 
87 const struct kvm_stats_header kvm_vcpu_stats_header = {
88 	.name_size = KVM_STATS_NAME_SIZE,
89 	.num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc),
90 	.id_offset = sizeof(struct kvm_stats_header),
91 	.desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
92 	.data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
93 		       sizeof(kvm_vcpu_stats_desc),
94 };
95 
96 bool kvm_trace_guest_mode_change;
97 
98 int kvm_guest_mode_change_trace_reg(void)
99 {
100 	kvm_trace_guest_mode_change = true;
101 	return 0;
102 }
103 
104 void kvm_guest_mode_change_trace_unreg(void)
105 {
106 	kvm_trace_guest_mode_change = false;
107 }
108 
109 /*
110  * XXXKYMA: We are simulatoring a processor that has the WII bit set in
111  * Config7, so we are "runnable" if interrupts are pending
112  */
113 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
114 {
115 	return !!(vcpu->arch.pending_exceptions);
116 }
117 
118 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
119 {
120 	return false;
121 }
122 
123 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
124 {
125 	return 1;
126 }
127 
128 int kvm_arch_hardware_enable(void)
129 {
130 	return kvm_mips_callbacks->hardware_enable();
131 }
132 
133 void kvm_arch_hardware_disable(void)
134 {
135 	kvm_mips_callbacks->hardware_disable();
136 }
137 
138 int kvm_arch_hardware_setup(void *opaque)
139 {
140 	return 0;
141 }
142 
143 int kvm_arch_check_processor_compat(void *opaque)
144 {
145 	return 0;
146 }
147 
148 extern void kvm_init_loongson_ipi(struct kvm *kvm);
149 
150 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
151 {
152 	switch (type) {
153 	case KVM_VM_MIPS_AUTO:
154 		break;
155 	case KVM_VM_MIPS_VZ:
156 		break;
157 	default:
158 		/* Unsupported KVM type */
159 		return -EINVAL;
160 	}
161 
162 	/* Allocate page table to map GPA -> RPA */
163 	kvm->arch.gpa_mm.pgd = kvm_pgd_alloc();
164 	if (!kvm->arch.gpa_mm.pgd)
165 		return -ENOMEM;
166 
167 #ifdef CONFIG_CPU_LOONGSON64
168 	kvm_init_loongson_ipi(kvm);
169 #endif
170 
171 	return 0;
172 }
173 
174 static void kvm_mips_free_gpa_pt(struct kvm *kvm)
175 {
176 	/* It should always be safe to remove after flushing the whole range */
177 	WARN_ON(!kvm_mips_flush_gpa_pt(kvm, 0, ~0));
178 	pgd_free(NULL, kvm->arch.gpa_mm.pgd);
179 }
180 
181 void kvm_arch_destroy_vm(struct kvm *kvm)
182 {
183 	kvm_destroy_vcpus(kvm);
184 	kvm_mips_free_gpa_pt(kvm);
185 }
186 
187 long kvm_arch_dev_ioctl(struct file *filp, unsigned int ioctl,
188 			unsigned long arg)
189 {
190 	return -ENOIOCTLCMD;
191 }
192 
193 void kvm_arch_flush_shadow_all(struct kvm *kvm)
194 {
195 	/* Flush whole GPA */
196 	kvm_mips_flush_gpa_pt(kvm, 0, ~0);
197 	kvm_flush_remote_tlbs(kvm);
198 }
199 
200 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
201 				   struct kvm_memory_slot *slot)
202 {
203 	/*
204 	 * The slot has been made invalid (ready for moving or deletion), so we
205 	 * need to ensure that it can no longer be accessed by any guest VCPUs.
206 	 */
207 
208 	spin_lock(&kvm->mmu_lock);
209 	/* Flush slot from GPA */
210 	kvm_mips_flush_gpa_pt(kvm, slot->base_gfn,
211 			      slot->base_gfn + slot->npages - 1);
212 	kvm_arch_flush_remote_tlbs_memslot(kvm, slot);
213 	spin_unlock(&kvm->mmu_lock);
214 }
215 
216 int kvm_arch_prepare_memory_region(struct kvm *kvm,
217 				   const struct kvm_memory_slot *old,
218 				   struct kvm_memory_slot *new,
219 				   enum kvm_mr_change change)
220 {
221 	return 0;
222 }
223 
224 void kvm_arch_commit_memory_region(struct kvm *kvm,
225 				   struct kvm_memory_slot *old,
226 				   const struct kvm_memory_slot *new,
227 				   enum kvm_mr_change change)
228 {
229 	int needs_flush;
230 
231 	/*
232 	 * If dirty page logging is enabled, write protect all pages in the slot
233 	 * ready for dirty logging.
234 	 *
235 	 * There is no need to do this in any of the following cases:
236 	 * CREATE:	No dirty mappings will already exist.
237 	 * MOVE/DELETE:	The old mappings will already have been cleaned up by
238 	 *		kvm_arch_flush_shadow_memslot()
239 	 */
240 	if (change == KVM_MR_FLAGS_ONLY &&
241 	    (!(old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
242 	     new->flags & KVM_MEM_LOG_DIRTY_PAGES)) {
243 		spin_lock(&kvm->mmu_lock);
244 		/* Write protect GPA page table entries */
245 		needs_flush = kvm_mips_mkclean_gpa_pt(kvm, new->base_gfn,
246 					new->base_gfn + new->npages - 1);
247 		if (needs_flush)
248 			kvm_arch_flush_remote_tlbs_memslot(kvm, new);
249 		spin_unlock(&kvm->mmu_lock);
250 	}
251 }
252 
253 static inline void dump_handler(const char *symbol, void *start, void *end)
254 {
255 	u32 *p;
256 
257 	pr_debug("LEAF(%s)\n", symbol);
258 
259 	pr_debug("\t.set push\n");
260 	pr_debug("\t.set noreorder\n");
261 
262 	for (p = start; p < (u32 *)end; ++p)
263 		pr_debug("\t.word\t0x%08x\t\t# %p\n", *p, p);
264 
265 	pr_debug("\t.set\tpop\n");
266 
267 	pr_debug("\tEND(%s)\n", symbol);
268 }
269 
270 /* low level hrtimer wake routine */
271 static enum hrtimer_restart kvm_mips_comparecount_wakeup(struct hrtimer *timer)
272 {
273 	struct kvm_vcpu *vcpu;
274 
275 	vcpu = container_of(timer, struct kvm_vcpu, arch.comparecount_timer);
276 
277 	kvm_mips_callbacks->queue_timer_int(vcpu);
278 
279 	vcpu->arch.wait = 0;
280 	rcuwait_wake_up(&vcpu->wait);
281 
282 	return kvm_mips_count_timeout(vcpu);
283 }
284 
285 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
286 {
287 	return 0;
288 }
289 
290 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
291 {
292 	int err, size;
293 	void *gebase, *p, *handler, *refill_start, *refill_end;
294 	int i;
295 
296 	kvm_debug("kvm @ %p: create cpu %d at %p\n",
297 		  vcpu->kvm, vcpu->vcpu_id, vcpu);
298 
299 	err = kvm_mips_callbacks->vcpu_init(vcpu);
300 	if (err)
301 		return err;
302 
303 	hrtimer_init(&vcpu->arch.comparecount_timer, CLOCK_MONOTONIC,
304 		     HRTIMER_MODE_REL);
305 	vcpu->arch.comparecount_timer.function = kvm_mips_comparecount_wakeup;
306 
307 	/*
308 	 * Allocate space for host mode exception handlers that handle
309 	 * guest mode exits
310 	 */
311 	if (cpu_has_veic || cpu_has_vint)
312 		size = 0x200 + VECTORSPACING * 64;
313 	else
314 		size = 0x4000;
315 
316 	gebase = kzalloc(ALIGN(size, PAGE_SIZE), GFP_KERNEL);
317 
318 	if (!gebase) {
319 		err = -ENOMEM;
320 		goto out_uninit_vcpu;
321 	}
322 	kvm_debug("Allocated %d bytes for KVM Exception Handlers @ %p\n",
323 		  ALIGN(size, PAGE_SIZE), gebase);
324 
325 	/*
326 	 * Check new ebase actually fits in CP0_EBase. The lack of a write gate
327 	 * limits us to the low 512MB of physical address space. If the memory
328 	 * we allocate is out of range, just give up now.
329 	 */
330 	if (!cpu_has_ebase_wg && virt_to_phys(gebase) >= 0x20000000) {
331 		kvm_err("CP0_EBase.WG required for guest exception base %pK\n",
332 			gebase);
333 		err = -ENOMEM;
334 		goto out_free_gebase;
335 	}
336 
337 	/* Save new ebase */
338 	vcpu->arch.guest_ebase = gebase;
339 
340 	/* Build guest exception vectors dynamically in unmapped memory */
341 	handler = gebase + 0x2000;
342 
343 	/* TLB refill (or XTLB refill on 64-bit VZ where KX=1) */
344 	refill_start = gebase;
345 	if (IS_ENABLED(CONFIG_64BIT))
346 		refill_start += 0x080;
347 	refill_end = kvm_mips_build_tlb_refill_exception(refill_start, handler);
348 
349 	/* General Exception Entry point */
350 	kvm_mips_build_exception(gebase + 0x180, handler);
351 
352 	/* For vectored interrupts poke the exception code @ all offsets 0-7 */
353 	for (i = 0; i < 8; i++) {
354 		kvm_debug("L1 Vectored handler @ %p\n",
355 			  gebase + 0x200 + (i * VECTORSPACING));
356 		kvm_mips_build_exception(gebase + 0x200 + i * VECTORSPACING,
357 					 handler);
358 	}
359 
360 	/* General exit handler */
361 	p = handler;
362 	p = kvm_mips_build_exit(p);
363 
364 	/* Guest entry routine */
365 	vcpu->arch.vcpu_run = p;
366 	p = kvm_mips_build_vcpu_run(p);
367 
368 	/* Dump the generated code */
369 	pr_debug("#include <asm/asm.h>\n");
370 	pr_debug("#include <asm/regdef.h>\n");
371 	pr_debug("\n");
372 	dump_handler("kvm_vcpu_run", vcpu->arch.vcpu_run, p);
373 	dump_handler("kvm_tlb_refill", refill_start, refill_end);
374 	dump_handler("kvm_gen_exc", gebase + 0x180, gebase + 0x200);
375 	dump_handler("kvm_exit", gebase + 0x2000, vcpu->arch.vcpu_run);
376 
377 	/* Invalidate the icache for these ranges */
378 	flush_icache_range((unsigned long)gebase,
379 			   (unsigned long)gebase + ALIGN(size, PAGE_SIZE));
380 
381 	/* Init */
382 	vcpu->arch.last_sched_cpu = -1;
383 	vcpu->arch.last_exec_cpu = -1;
384 
385 	/* Initial guest state */
386 	err = kvm_mips_callbacks->vcpu_setup(vcpu);
387 	if (err)
388 		goto out_free_gebase;
389 
390 	return 0;
391 
392 out_free_gebase:
393 	kfree(gebase);
394 out_uninit_vcpu:
395 	kvm_mips_callbacks->vcpu_uninit(vcpu);
396 	return err;
397 }
398 
399 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
400 {
401 	hrtimer_cancel(&vcpu->arch.comparecount_timer);
402 
403 	kvm_mips_dump_stats(vcpu);
404 
405 	kvm_mmu_free_memory_caches(vcpu);
406 	kfree(vcpu->arch.guest_ebase);
407 
408 	kvm_mips_callbacks->vcpu_uninit(vcpu);
409 }
410 
411 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
412 					struct kvm_guest_debug *dbg)
413 {
414 	return -ENOIOCTLCMD;
415 }
416 
417 /*
418  * Actually run the vCPU, entering an RCU extended quiescent state (EQS) while
419  * the vCPU is running.
420  *
421  * This must be noinstr as instrumentation may make use of RCU, and this is not
422  * safe during the EQS.
423  */
424 static int noinstr kvm_mips_vcpu_enter_exit(struct kvm_vcpu *vcpu)
425 {
426 	int ret;
427 
428 	guest_state_enter_irqoff();
429 	ret = kvm_mips_callbacks->vcpu_run(vcpu);
430 	guest_state_exit_irqoff();
431 
432 	return ret;
433 }
434 
435 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
436 {
437 	int r = -EINTR;
438 
439 	vcpu_load(vcpu);
440 
441 	kvm_sigset_activate(vcpu);
442 
443 	if (vcpu->mmio_needed) {
444 		if (!vcpu->mmio_is_write)
445 			kvm_mips_complete_mmio_load(vcpu);
446 		vcpu->mmio_needed = 0;
447 	}
448 
449 	if (vcpu->run->immediate_exit)
450 		goto out;
451 
452 	lose_fpu(1);
453 
454 	local_irq_disable();
455 	guest_timing_enter_irqoff();
456 	trace_kvm_enter(vcpu);
457 
458 	/*
459 	 * Make sure the read of VCPU requests in vcpu_run() callback is not
460 	 * reordered ahead of the write to vcpu->mode, or we could miss a TLB
461 	 * flush request while the requester sees the VCPU as outside of guest
462 	 * mode and not needing an IPI.
463 	 */
464 	smp_store_mb(vcpu->mode, IN_GUEST_MODE);
465 
466 	r = kvm_mips_vcpu_enter_exit(vcpu);
467 
468 	/*
469 	 * We must ensure that any pending interrupts are taken before
470 	 * we exit guest timing so that timer ticks are accounted as
471 	 * guest time. Transiently unmask interrupts so that any
472 	 * pending interrupts are taken.
473 	 *
474 	 * TODO: is there a barrier which ensures that pending interrupts are
475 	 * recognised? Currently this just hopes that the CPU takes any pending
476 	 * interrupts between the enable and disable.
477 	 */
478 	local_irq_enable();
479 	local_irq_disable();
480 
481 	trace_kvm_out(vcpu);
482 	guest_timing_exit_irqoff();
483 	local_irq_enable();
484 
485 out:
486 	kvm_sigset_deactivate(vcpu);
487 
488 	vcpu_put(vcpu);
489 	return r;
490 }
491 
492 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
493 			     struct kvm_mips_interrupt *irq)
494 {
495 	int intr = (int)irq->irq;
496 	struct kvm_vcpu *dvcpu = NULL;
497 
498 	if (intr == kvm_priority_to_irq[MIPS_EXC_INT_IPI_1] ||
499 	    intr == kvm_priority_to_irq[MIPS_EXC_INT_IPI_2] ||
500 	    intr == (-kvm_priority_to_irq[MIPS_EXC_INT_IPI_1]) ||
501 	    intr == (-kvm_priority_to_irq[MIPS_EXC_INT_IPI_2]))
502 		kvm_debug("%s: CPU: %d, INTR: %d\n", __func__, irq->cpu,
503 			  (int)intr);
504 
505 	if (irq->cpu == -1)
506 		dvcpu = vcpu;
507 	else
508 		dvcpu = kvm_get_vcpu(vcpu->kvm, irq->cpu);
509 
510 	if (intr == 2 || intr == 3 || intr == 4 || intr == 6) {
511 		kvm_mips_callbacks->queue_io_int(dvcpu, irq);
512 
513 	} else if (intr == -2 || intr == -3 || intr == -4 || intr == -6) {
514 		kvm_mips_callbacks->dequeue_io_int(dvcpu, irq);
515 	} else {
516 		kvm_err("%s: invalid interrupt ioctl (%d:%d)\n", __func__,
517 			irq->cpu, irq->irq);
518 		return -EINVAL;
519 	}
520 
521 	dvcpu->arch.wait = 0;
522 
523 	rcuwait_wake_up(&dvcpu->wait);
524 
525 	return 0;
526 }
527 
528 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
529 				    struct kvm_mp_state *mp_state)
530 {
531 	return -ENOIOCTLCMD;
532 }
533 
534 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
535 				    struct kvm_mp_state *mp_state)
536 {
537 	return -ENOIOCTLCMD;
538 }
539 
540 static u64 kvm_mips_get_one_regs[] = {
541 	KVM_REG_MIPS_R0,
542 	KVM_REG_MIPS_R1,
543 	KVM_REG_MIPS_R2,
544 	KVM_REG_MIPS_R3,
545 	KVM_REG_MIPS_R4,
546 	KVM_REG_MIPS_R5,
547 	KVM_REG_MIPS_R6,
548 	KVM_REG_MIPS_R7,
549 	KVM_REG_MIPS_R8,
550 	KVM_REG_MIPS_R9,
551 	KVM_REG_MIPS_R10,
552 	KVM_REG_MIPS_R11,
553 	KVM_REG_MIPS_R12,
554 	KVM_REG_MIPS_R13,
555 	KVM_REG_MIPS_R14,
556 	KVM_REG_MIPS_R15,
557 	KVM_REG_MIPS_R16,
558 	KVM_REG_MIPS_R17,
559 	KVM_REG_MIPS_R18,
560 	KVM_REG_MIPS_R19,
561 	KVM_REG_MIPS_R20,
562 	KVM_REG_MIPS_R21,
563 	KVM_REG_MIPS_R22,
564 	KVM_REG_MIPS_R23,
565 	KVM_REG_MIPS_R24,
566 	KVM_REG_MIPS_R25,
567 	KVM_REG_MIPS_R26,
568 	KVM_REG_MIPS_R27,
569 	KVM_REG_MIPS_R28,
570 	KVM_REG_MIPS_R29,
571 	KVM_REG_MIPS_R30,
572 	KVM_REG_MIPS_R31,
573 
574 #ifndef CONFIG_CPU_MIPSR6
575 	KVM_REG_MIPS_HI,
576 	KVM_REG_MIPS_LO,
577 #endif
578 	KVM_REG_MIPS_PC,
579 };
580 
581 static u64 kvm_mips_get_one_regs_fpu[] = {
582 	KVM_REG_MIPS_FCR_IR,
583 	KVM_REG_MIPS_FCR_CSR,
584 };
585 
586 static u64 kvm_mips_get_one_regs_msa[] = {
587 	KVM_REG_MIPS_MSA_IR,
588 	KVM_REG_MIPS_MSA_CSR,
589 };
590 
591 static unsigned long kvm_mips_num_regs(struct kvm_vcpu *vcpu)
592 {
593 	unsigned long ret;
594 
595 	ret = ARRAY_SIZE(kvm_mips_get_one_regs);
596 	if (kvm_mips_guest_can_have_fpu(&vcpu->arch)) {
597 		ret += ARRAY_SIZE(kvm_mips_get_one_regs_fpu) + 48;
598 		/* odd doubles */
599 		if (boot_cpu_data.fpu_id & MIPS_FPIR_F64)
600 			ret += 16;
601 	}
602 	if (kvm_mips_guest_can_have_msa(&vcpu->arch))
603 		ret += ARRAY_SIZE(kvm_mips_get_one_regs_msa) + 32;
604 	ret += kvm_mips_callbacks->num_regs(vcpu);
605 
606 	return ret;
607 }
608 
609 static int kvm_mips_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *indices)
610 {
611 	u64 index;
612 	unsigned int i;
613 
614 	if (copy_to_user(indices, kvm_mips_get_one_regs,
615 			 sizeof(kvm_mips_get_one_regs)))
616 		return -EFAULT;
617 	indices += ARRAY_SIZE(kvm_mips_get_one_regs);
618 
619 	if (kvm_mips_guest_can_have_fpu(&vcpu->arch)) {
620 		if (copy_to_user(indices, kvm_mips_get_one_regs_fpu,
621 				 sizeof(kvm_mips_get_one_regs_fpu)))
622 			return -EFAULT;
623 		indices += ARRAY_SIZE(kvm_mips_get_one_regs_fpu);
624 
625 		for (i = 0; i < 32; ++i) {
626 			index = KVM_REG_MIPS_FPR_32(i);
627 			if (copy_to_user(indices, &index, sizeof(index)))
628 				return -EFAULT;
629 			++indices;
630 
631 			/* skip odd doubles if no F64 */
632 			if (i & 1 && !(boot_cpu_data.fpu_id & MIPS_FPIR_F64))
633 				continue;
634 
635 			index = KVM_REG_MIPS_FPR_64(i);
636 			if (copy_to_user(indices, &index, sizeof(index)))
637 				return -EFAULT;
638 			++indices;
639 		}
640 	}
641 
642 	if (kvm_mips_guest_can_have_msa(&vcpu->arch)) {
643 		if (copy_to_user(indices, kvm_mips_get_one_regs_msa,
644 				 sizeof(kvm_mips_get_one_regs_msa)))
645 			return -EFAULT;
646 		indices += ARRAY_SIZE(kvm_mips_get_one_regs_msa);
647 
648 		for (i = 0; i < 32; ++i) {
649 			index = KVM_REG_MIPS_VEC_128(i);
650 			if (copy_to_user(indices, &index, sizeof(index)))
651 				return -EFAULT;
652 			++indices;
653 		}
654 	}
655 
656 	return kvm_mips_callbacks->copy_reg_indices(vcpu, indices);
657 }
658 
659 static int kvm_mips_get_reg(struct kvm_vcpu *vcpu,
660 			    const struct kvm_one_reg *reg)
661 {
662 	struct mips_coproc *cop0 = vcpu->arch.cop0;
663 	struct mips_fpu_struct *fpu = &vcpu->arch.fpu;
664 	int ret;
665 	s64 v;
666 	s64 vs[2];
667 	unsigned int idx;
668 
669 	switch (reg->id) {
670 	/* General purpose registers */
671 	case KVM_REG_MIPS_R0 ... KVM_REG_MIPS_R31:
672 		v = (long)vcpu->arch.gprs[reg->id - KVM_REG_MIPS_R0];
673 		break;
674 #ifndef CONFIG_CPU_MIPSR6
675 	case KVM_REG_MIPS_HI:
676 		v = (long)vcpu->arch.hi;
677 		break;
678 	case KVM_REG_MIPS_LO:
679 		v = (long)vcpu->arch.lo;
680 		break;
681 #endif
682 	case KVM_REG_MIPS_PC:
683 		v = (long)vcpu->arch.pc;
684 		break;
685 
686 	/* Floating point registers */
687 	case KVM_REG_MIPS_FPR_32(0) ... KVM_REG_MIPS_FPR_32(31):
688 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
689 			return -EINVAL;
690 		idx = reg->id - KVM_REG_MIPS_FPR_32(0);
691 		/* Odd singles in top of even double when FR=0 */
692 		if (kvm_read_c0_guest_status(cop0) & ST0_FR)
693 			v = get_fpr32(&fpu->fpr[idx], 0);
694 		else
695 			v = get_fpr32(&fpu->fpr[idx & ~1], idx & 1);
696 		break;
697 	case KVM_REG_MIPS_FPR_64(0) ... KVM_REG_MIPS_FPR_64(31):
698 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
699 			return -EINVAL;
700 		idx = reg->id - KVM_REG_MIPS_FPR_64(0);
701 		/* Can't access odd doubles in FR=0 mode */
702 		if (idx & 1 && !(kvm_read_c0_guest_status(cop0) & ST0_FR))
703 			return -EINVAL;
704 		v = get_fpr64(&fpu->fpr[idx], 0);
705 		break;
706 	case KVM_REG_MIPS_FCR_IR:
707 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
708 			return -EINVAL;
709 		v = boot_cpu_data.fpu_id;
710 		break;
711 	case KVM_REG_MIPS_FCR_CSR:
712 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
713 			return -EINVAL;
714 		v = fpu->fcr31;
715 		break;
716 
717 	/* MIPS SIMD Architecture (MSA) registers */
718 	case KVM_REG_MIPS_VEC_128(0) ... KVM_REG_MIPS_VEC_128(31):
719 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
720 			return -EINVAL;
721 		/* Can't access MSA registers in FR=0 mode */
722 		if (!(kvm_read_c0_guest_status(cop0) & ST0_FR))
723 			return -EINVAL;
724 		idx = reg->id - KVM_REG_MIPS_VEC_128(0);
725 #ifdef CONFIG_CPU_LITTLE_ENDIAN
726 		/* least significant byte first */
727 		vs[0] = get_fpr64(&fpu->fpr[idx], 0);
728 		vs[1] = get_fpr64(&fpu->fpr[idx], 1);
729 #else
730 		/* most significant byte first */
731 		vs[0] = get_fpr64(&fpu->fpr[idx], 1);
732 		vs[1] = get_fpr64(&fpu->fpr[idx], 0);
733 #endif
734 		break;
735 	case KVM_REG_MIPS_MSA_IR:
736 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
737 			return -EINVAL;
738 		v = boot_cpu_data.msa_id;
739 		break;
740 	case KVM_REG_MIPS_MSA_CSR:
741 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
742 			return -EINVAL;
743 		v = fpu->msacsr;
744 		break;
745 
746 	/* registers to be handled specially */
747 	default:
748 		ret = kvm_mips_callbacks->get_one_reg(vcpu, reg, &v);
749 		if (ret)
750 			return ret;
751 		break;
752 	}
753 	if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U64) {
754 		u64 __user *uaddr64 = (u64 __user *)(long)reg->addr;
755 
756 		return put_user(v, uaddr64);
757 	} else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U32) {
758 		u32 __user *uaddr32 = (u32 __user *)(long)reg->addr;
759 		u32 v32 = (u32)v;
760 
761 		return put_user(v32, uaddr32);
762 	} else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U128) {
763 		void __user *uaddr = (void __user *)(long)reg->addr;
764 
765 		return copy_to_user(uaddr, vs, 16) ? -EFAULT : 0;
766 	} else {
767 		return -EINVAL;
768 	}
769 }
770 
771 static int kvm_mips_set_reg(struct kvm_vcpu *vcpu,
772 			    const struct kvm_one_reg *reg)
773 {
774 	struct mips_coproc *cop0 = vcpu->arch.cop0;
775 	struct mips_fpu_struct *fpu = &vcpu->arch.fpu;
776 	s64 v;
777 	s64 vs[2];
778 	unsigned int idx;
779 
780 	if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U64) {
781 		u64 __user *uaddr64 = (u64 __user *)(long)reg->addr;
782 
783 		if (get_user(v, uaddr64) != 0)
784 			return -EFAULT;
785 	} else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U32) {
786 		u32 __user *uaddr32 = (u32 __user *)(long)reg->addr;
787 		s32 v32;
788 
789 		if (get_user(v32, uaddr32) != 0)
790 			return -EFAULT;
791 		v = (s64)v32;
792 	} else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U128) {
793 		void __user *uaddr = (void __user *)(long)reg->addr;
794 
795 		return copy_from_user(vs, uaddr, 16) ? -EFAULT : 0;
796 	} else {
797 		return -EINVAL;
798 	}
799 
800 	switch (reg->id) {
801 	/* General purpose registers */
802 	case KVM_REG_MIPS_R0:
803 		/* Silently ignore requests to set $0 */
804 		break;
805 	case KVM_REG_MIPS_R1 ... KVM_REG_MIPS_R31:
806 		vcpu->arch.gprs[reg->id - KVM_REG_MIPS_R0] = v;
807 		break;
808 #ifndef CONFIG_CPU_MIPSR6
809 	case KVM_REG_MIPS_HI:
810 		vcpu->arch.hi = v;
811 		break;
812 	case KVM_REG_MIPS_LO:
813 		vcpu->arch.lo = v;
814 		break;
815 #endif
816 	case KVM_REG_MIPS_PC:
817 		vcpu->arch.pc = v;
818 		break;
819 
820 	/* Floating point registers */
821 	case KVM_REG_MIPS_FPR_32(0) ... KVM_REG_MIPS_FPR_32(31):
822 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
823 			return -EINVAL;
824 		idx = reg->id - KVM_REG_MIPS_FPR_32(0);
825 		/* Odd singles in top of even double when FR=0 */
826 		if (kvm_read_c0_guest_status(cop0) & ST0_FR)
827 			set_fpr32(&fpu->fpr[idx], 0, v);
828 		else
829 			set_fpr32(&fpu->fpr[idx & ~1], idx & 1, v);
830 		break;
831 	case KVM_REG_MIPS_FPR_64(0) ... KVM_REG_MIPS_FPR_64(31):
832 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
833 			return -EINVAL;
834 		idx = reg->id - KVM_REG_MIPS_FPR_64(0);
835 		/* Can't access odd doubles in FR=0 mode */
836 		if (idx & 1 && !(kvm_read_c0_guest_status(cop0) & ST0_FR))
837 			return -EINVAL;
838 		set_fpr64(&fpu->fpr[idx], 0, v);
839 		break;
840 	case KVM_REG_MIPS_FCR_IR:
841 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
842 			return -EINVAL;
843 		/* Read-only */
844 		break;
845 	case KVM_REG_MIPS_FCR_CSR:
846 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
847 			return -EINVAL;
848 		fpu->fcr31 = v;
849 		break;
850 
851 	/* MIPS SIMD Architecture (MSA) registers */
852 	case KVM_REG_MIPS_VEC_128(0) ... KVM_REG_MIPS_VEC_128(31):
853 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
854 			return -EINVAL;
855 		idx = reg->id - KVM_REG_MIPS_VEC_128(0);
856 #ifdef CONFIG_CPU_LITTLE_ENDIAN
857 		/* least significant byte first */
858 		set_fpr64(&fpu->fpr[idx], 0, vs[0]);
859 		set_fpr64(&fpu->fpr[idx], 1, vs[1]);
860 #else
861 		/* most significant byte first */
862 		set_fpr64(&fpu->fpr[idx], 1, vs[0]);
863 		set_fpr64(&fpu->fpr[idx], 0, vs[1]);
864 #endif
865 		break;
866 	case KVM_REG_MIPS_MSA_IR:
867 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
868 			return -EINVAL;
869 		/* Read-only */
870 		break;
871 	case KVM_REG_MIPS_MSA_CSR:
872 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
873 			return -EINVAL;
874 		fpu->msacsr = v;
875 		break;
876 
877 	/* registers to be handled specially */
878 	default:
879 		return kvm_mips_callbacks->set_one_reg(vcpu, reg, v);
880 	}
881 	return 0;
882 }
883 
884 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
885 				     struct kvm_enable_cap *cap)
886 {
887 	int r = 0;
888 
889 	if (!kvm_vm_ioctl_check_extension(vcpu->kvm, cap->cap))
890 		return -EINVAL;
891 	if (cap->flags)
892 		return -EINVAL;
893 	if (cap->args[0])
894 		return -EINVAL;
895 
896 	switch (cap->cap) {
897 	case KVM_CAP_MIPS_FPU:
898 		vcpu->arch.fpu_enabled = true;
899 		break;
900 	case KVM_CAP_MIPS_MSA:
901 		vcpu->arch.msa_enabled = true;
902 		break;
903 	default:
904 		r = -EINVAL;
905 		break;
906 	}
907 
908 	return r;
909 }
910 
911 long kvm_arch_vcpu_async_ioctl(struct file *filp, unsigned int ioctl,
912 			       unsigned long arg)
913 {
914 	struct kvm_vcpu *vcpu = filp->private_data;
915 	void __user *argp = (void __user *)arg;
916 
917 	if (ioctl == KVM_INTERRUPT) {
918 		struct kvm_mips_interrupt irq;
919 
920 		if (copy_from_user(&irq, argp, sizeof(irq)))
921 			return -EFAULT;
922 		kvm_debug("[%d] %s: irq: %d\n", vcpu->vcpu_id, __func__,
923 			  irq.irq);
924 
925 		return kvm_vcpu_ioctl_interrupt(vcpu, &irq);
926 	}
927 
928 	return -ENOIOCTLCMD;
929 }
930 
931 long kvm_arch_vcpu_ioctl(struct file *filp, unsigned int ioctl,
932 			 unsigned long arg)
933 {
934 	struct kvm_vcpu *vcpu = filp->private_data;
935 	void __user *argp = (void __user *)arg;
936 	long r;
937 
938 	vcpu_load(vcpu);
939 
940 	switch (ioctl) {
941 	case KVM_SET_ONE_REG:
942 	case KVM_GET_ONE_REG: {
943 		struct kvm_one_reg reg;
944 
945 		r = -EFAULT;
946 		if (copy_from_user(&reg, argp, sizeof(reg)))
947 			break;
948 		if (ioctl == KVM_SET_ONE_REG)
949 			r = kvm_mips_set_reg(vcpu, &reg);
950 		else
951 			r = kvm_mips_get_reg(vcpu, &reg);
952 		break;
953 	}
954 	case KVM_GET_REG_LIST: {
955 		struct kvm_reg_list __user *user_list = argp;
956 		struct kvm_reg_list reg_list;
957 		unsigned n;
958 
959 		r = -EFAULT;
960 		if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
961 			break;
962 		n = reg_list.n;
963 		reg_list.n = kvm_mips_num_regs(vcpu);
964 		if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
965 			break;
966 		r = -E2BIG;
967 		if (n < reg_list.n)
968 			break;
969 		r = kvm_mips_copy_reg_indices(vcpu, user_list->reg);
970 		break;
971 	}
972 	case KVM_ENABLE_CAP: {
973 		struct kvm_enable_cap cap;
974 
975 		r = -EFAULT;
976 		if (copy_from_user(&cap, argp, sizeof(cap)))
977 			break;
978 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
979 		break;
980 	}
981 	default:
982 		r = -ENOIOCTLCMD;
983 	}
984 
985 	vcpu_put(vcpu);
986 	return r;
987 }
988 
989 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
990 {
991 
992 }
993 
994 int kvm_arch_flush_remote_tlb(struct kvm *kvm)
995 {
996 	kvm_mips_callbacks->prepare_flush_shadow(kvm);
997 	return 1;
998 }
999 
1000 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1001 					const struct kvm_memory_slot *memslot)
1002 {
1003 	kvm_flush_remote_tlbs(kvm);
1004 }
1005 
1006 long kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
1007 {
1008 	long r;
1009 
1010 	switch (ioctl) {
1011 	default:
1012 		r = -ENOIOCTLCMD;
1013 	}
1014 
1015 	return r;
1016 }
1017 
1018 int kvm_arch_init(void *opaque)
1019 {
1020 	if (kvm_mips_callbacks) {
1021 		kvm_err("kvm: module already exists\n");
1022 		return -EEXIST;
1023 	}
1024 
1025 	return kvm_mips_emulation_init(&kvm_mips_callbacks);
1026 }
1027 
1028 void kvm_arch_exit(void)
1029 {
1030 	kvm_mips_callbacks = NULL;
1031 }
1032 
1033 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1034 				  struct kvm_sregs *sregs)
1035 {
1036 	return -ENOIOCTLCMD;
1037 }
1038 
1039 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1040 				  struct kvm_sregs *sregs)
1041 {
1042 	return -ENOIOCTLCMD;
1043 }
1044 
1045 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
1046 {
1047 }
1048 
1049 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1050 {
1051 	return -ENOIOCTLCMD;
1052 }
1053 
1054 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1055 {
1056 	return -ENOIOCTLCMD;
1057 }
1058 
1059 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
1060 {
1061 	return VM_FAULT_SIGBUS;
1062 }
1063 
1064 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
1065 {
1066 	int r;
1067 
1068 	switch (ext) {
1069 	case KVM_CAP_ONE_REG:
1070 	case KVM_CAP_ENABLE_CAP:
1071 	case KVM_CAP_READONLY_MEM:
1072 	case KVM_CAP_SYNC_MMU:
1073 	case KVM_CAP_IMMEDIATE_EXIT:
1074 		r = 1;
1075 		break;
1076 	case KVM_CAP_NR_VCPUS:
1077 		r = min_t(unsigned int, num_online_cpus(), KVM_MAX_VCPUS);
1078 		break;
1079 	case KVM_CAP_MAX_VCPUS:
1080 		r = KVM_MAX_VCPUS;
1081 		break;
1082 	case KVM_CAP_MAX_VCPU_ID:
1083 		r = KVM_MAX_VCPU_IDS;
1084 		break;
1085 	case KVM_CAP_MIPS_FPU:
1086 		/* We don't handle systems with inconsistent cpu_has_fpu */
1087 		r = !!raw_cpu_has_fpu;
1088 		break;
1089 	case KVM_CAP_MIPS_MSA:
1090 		/*
1091 		 * We don't support MSA vector partitioning yet:
1092 		 * 1) It would require explicit support which can't be tested
1093 		 *    yet due to lack of support in current hardware.
1094 		 * 2) It extends the state that would need to be saved/restored
1095 		 *    by e.g. QEMU for migration.
1096 		 *
1097 		 * When vector partitioning hardware becomes available, support
1098 		 * could be added by requiring a flag when enabling
1099 		 * KVM_CAP_MIPS_MSA capability to indicate that userland knows
1100 		 * to save/restore the appropriate extra state.
1101 		 */
1102 		r = cpu_has_msa && !(boot_cpu_data.msa_id & MSA_IR_WRPF);
1103 		break;
1104 	default:
1105 		r = kvm_mips_callbacks->check_extension(kvm, ext);
1106 		break;
1107 	}
1108 	return r;
1109 }
1110 
1111 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1112 {
1113 	return kvm_mips_pending_timer(vcpu) ||
1114 		kvm_read_c0_guest_cause(vcpu->arch.cop0) & C_TI;
1115 }
1116 
1117 int kvm_arch_vcpu_dump_regs(struct kvm_vcpu *vcpu)
1118 {
1119 	int i;
1120 	struct mips_coproc *cop0;
1121 
1122 	if (!vcpu)
1123 		return -1;
1124 
1125 	kvm_debug("VCPU Register Dump:\n");
1126 	kvm_debug("\tpc = 0x%08lx\n", vcpu->arch.pc);
1127 	kvm_debug("\texceptions: %08lx\n", vcpu->arch.pending_exceptions);
1128 
1129 	for (i = 0; i < 32; i += 4) {
1130 		kvm_debug("\tgpr%02d: %08lx %08lx %08lx %08lx\n", i,
1131 		       vcpu->arch.gprs[i],
1132 		       vcpu->arch.gprs[i + 1],
1133 		       vcpu->arch.gprs[i + 2], vcpu->arch.gprs[i + 3]);
1134 	}
1135 	kvm_debug("\thi: 0x%08lx\n", vcpu->arch.hi);
1136 	kvm_debug("\tlo: 0x%08lx\n", vcpu->arch.lo);
1137 
1138 	cop0 = vcpu->arch.cop0;
1139 	kvm_debug("\tStatus: 0x%08x, Cause: 0x%08x\n",
1140 		  kvm_read_c0_guest_status(cop0),
1141 		  kvm_read_c0_guest_cause(cop0));
1142 
1143 	kvm_debug("\tEPC: 0x%08lx\n", kvm_read_c0_guest_epc(cop0));
1144 
1145 	return 0;
1146 }
1147 
1148 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1149 {
1150 	int i;
1151 
1152 	vcpu_load(vcpu);
1153 
1154 	for (i = 1; i < ARRAY_SIZE(vcpu->arch.gprs); i++)
1155 		vcpu->arch.gprs[i] = regs->gpr[i];
1156 	vcpu->arch.gprs[0] = 0; /* zero is special, and cannot be set. */
1157 	vcpu->arch.hi = regs->hi;
1158 	vcpu->arch.lo = regs->lo;
1159 	vcpu->arch.pc = regs->pc;
1160 
1161 	vcpu_put(vcpu);
1162 	return 0;
1163 }
1164 
1165 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1166 {
1167 	int i;
1168 
1169 	vcpu_load(vcpu);
1170 
1171 	for (i = 0; i < ARRAY_SIZE(vcpu->arch.gprs); i++)
1172 		regs->gpr[i] = vcpu->arch.gprs[i];
1173 
1174 	regs->hi = vcpu->arch.hi;
1175 	regs->lo = vcpu->arch.lo;
1176 	regs->pc = vcpu->arch.pc;
1177 
1178 	vcpu_put(vcpu);
1179 	return 0;
1180 }
1181 
1182 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1183 				  struct kvm_translation *tr)
1184 {
1185 	return 0;
1186 }
1187 
1188 static void kvm_mips_set_c0_status(void)
1189 {
1190 	u32 status = read_c0_status();
1191 
1192 	if (cpu_has_dsp)
1193 		status |= (ST0_MX);
1194 
1195 	write_c0_status(status);
1196 	ehb();
1197 }
1198 
1199 /*
1200  * Return value is in the form (errcode<<2 | RESUME_FLAG_HOST | RESUME_FLAG_NV)
1201  */
1202 static int __kvm_mips_handle_exit(struct kvm_vcpu *vcpu)
1203 {
1204 	struct kvm_run *run = vcpu->run;
1205 	u32 cause = vcpu->arch.host_cp0_cause;
1206 	u32 exccode = (cause >> CAUSEB_EXCCODE) & 0x1f;
1207 	u32 __user *opc = (u32 __user *) vcpu->arch.pc;
1208 	unsigned long badvaddr = vcpu->arch.host_cp0_badvaddr;
1209 	enum emulation_result er = EMULATE_DONE;
1210 	u32 inst;
1211 	int ret = RESUME_GUEST;
1212 
1213 	vcpu->mode = OUTSIDE_GUEST_MODE;
1214 
1215 	/* Set a default exit reason */
1216 	run->exit_reason = KVM_EXIT_UNKNOWN;
1217 	run->ready_for_interrupt_injection = 1;
1218 
1219 	/*
1220 	 * Set the appropriate status bits based on host CPU features,
1221 	 * before we hit the scheduler
1222 	 */
1223 	kvm_mips_set_c0_status();
1224 
1225 	local_irq_enable();
1226 
1227 	kvm_debug("kvm_mips_handle_exit: cause: %#x, PC: %p, kvm_run: %p, kvm_vcpu: %p\n",
1228 			cause, opc, run, vcpu);
1229 	trace_kvm_exit(vcpu, exccode);
1230 
1231 	switch (exccode) {
1232 	case EXCCODE_INT:
1233 		kvm_debug("[%d]EXCCODE_INT @ %p\n", vcpu->vcpu_id, opc);
1234 
1235 		++vcpu->stat.int_exits;
1236 
1237 		if (need_resched())
1238 			cond_resched();
1239 
1240 		ret = RESUME_GUEST;
1241 		break;
1242 
1243 	case EXCCODE_CPU:
1244 		kvm_debug("EXCCODE_CPU: @ PC: %p\n", opc);
1245 
1246 		++vcpu->stat.cop_unusable_exits;
1247 		ret = kvm_mips_callbacks->handle_cop_unusable(vcpu);
1248 		/* XXXKYMA: Might need to return to user space */
1249 		if (run->exit_reason == KVM_EXIT_IRQ_WINDOW_OPEN)
1250 			ret = RESUME_HOST;
1251 		break;
1252 
1253 	case EXCCODE_MOD:
1254 		++vcpu->stat.tlbmod_exits;
1255 		ret = kvm_mips_callbacks->handle_tlb_mod(vcpu);
1256 		break;
1257 
1258 	case EXCCODE_TLBS:
1259 		kvm_debug("TLB ST fault:  cause %#x, status %#x, PC: %p, BadVaddr: %#lx\n",
1260 			  cause, kvm_read_c0_guest_status(vcpu->arch.cop0), opc,
1261 			  badvaddr);
1262 
1263 		++vcpu->stat.tlbmiss_st_exits;
1264 		ret = kvm_mips_callbacks->handle_tlb_st_miss(vcpu);
1265 		break;
1266 
1267 	case EXCCODE_TLBL:
1268 		kvm_debug("TLB LD fault: cause %#x, PC: %p, BadVaddr: %#lx\n",
1269 			  cause, opc, badvaddr);
1270 
1271 		++vcpu->stat.tlbmiss_ld_exits;
1272 		ret = kvm_mips_callbacks->handle_tlb_ld_miss(vcpu);
1273 		break;
1274 
1275 	case EXCCODE_ADES:
1276 		++vcpu->stat.addrerr_st_exits;
1277 		ret = kvm_mips_callbacks->handle_addr_err_st(vcpu);
1278 		break;
1279 
1280 	case EXCCODE_ADEL:
1281 		++vcpu->stat.addrerr_ld_exits;
1282 		ret = kvm_mips_callbacks->handle_addr_err_ld(vcpu);
1283 		break;
1284 
1285 	case EXCCODE_SYS:
1286 		++vcpu->stat.syscall_exits;
1287 		ret = kvm_mips_callbacks->handle_syscall(vcpu);
1288 		break;
1289 
1290 	case EXCCODE_RI:
1291 		++vcpu->stat.resvd_inst_exits;
1292 		ret = kvm_mips_callbacks->handle_res_inst(vcpu);
1293 		break;
1294 
1295 	case EXCCODE_BP:
1296 		++vcpu->stat.break_inst_exits;
1297 		ret = kvm_mips_callbacks->handle_break(vcpu);
1298 		break;
1299 
1300 	case EXCCODE_TR:
1301 		++vcpu->stat.trap_inst_exits;
1302 		ret = kvm_mips_callbacks->handle_trap(vcpu);
1303 		break;
1304 
1305 	case EXCCODE_MSAFPE:
1306 		++vcpu->stat.msa_fpe_exits;
1307 		ret = kvm_mips_callbacks->handle_msa_fpe(vcpu);
1308 		break;
1309 
1310 	case EXCCODE_FPE:
1311 		++vcpu->stat.fpe_exits;
1312 		ret = kvm_mips_callbacks->handle_fpe(vcpu);
1313 		break;
1314 
1315 	case EXCCODE_MSADIS:
1316 		++vcpu->stat.msa_disabled_exits;
1317 		ret = kvm_mips_callbacks->handle_msa_disabled(vcpu);
1318 		break;
1319 
1320 	case EXCCODE_GE:
1321 		/* defer exit accounting to handler */
1322 		ret = kvm_mips_callbacks->handle_guest_exit(vcpu);
1323 		break;
1324 
1325 	default:
1326 		if (cause & CAUSEF_BD)
1327 			opc += 1;
1328 		inst = 0;
1329 		kvm_get_badinstr(opc, vcpu, &inst);
1330 		kvm_err("Exception Code: %d, not yet handled, @ PC: %p, inst: 0x%08x  BadVaddr: %#lx Status: %#x\n",
1331 			exccode, opc, inst, badvaddr,
1332 			kvm_read_c0_guest_status(vcpu->arch.cop0));
1333 		kvm_arch_vcpu_dump_regs(vcpu);
1334 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1335 		ret = RESUME_HOST;
1336 		break;
1337 
1338 	}
1339 
1340 	local_irq_disable();
1341 
1342 	if (ret == RESUME_GUEST)
1343 		kvm_vz_acquire_htimer(vcpu);
1344 
1345 	if (er == EMULATE_DONE && !(ret & RESUME_HOST))
1346 		kvm_mips_deliver_interrupts(vcpu, cause);
1347 
1348 	if (!(ret & RESUME_HOST)) {
1349 		/* Only check for signals if not already exiting to userspace */
1350 		if (signal_pending(current)) {
1351 			run->exit_reason = KVM_EXIT_INTR;
1352 			ret = (-EINTR << 2) | RESUME_HOST;
1353 			++vcpu->stat.signal_exits;
1354 			trace_kvm_exit(vcpu, KVM_TRACE_EXIT_SIGNAL);
1355 		}
1356 	}
1357 
1358 	if (ret == RESUME_GUEST) {
1359 		trace_kvm_reenter(vcpu);
1360 
1361 		/*
1362 		 * Make sure the read of VCPU requests in vcpu_reenter()
1363 		 * callback is not reordered ahead of the write to vcpu->mode,
1364 		 * or we could miss a TLB flush request while the requester sees
1365 		 * the VCPU as outside of guest mode and not needing an IPI.
1366 		 */
1367 		smp_store_mb(vcpu->mode, IN_GUEST_MODE);
1368 
1369 		kvm_mips_callbacks->vcpu_reenter(vcpu);
1370 
1371 		/*
1372 		 * If FPU / MSA are enabled (i.e. the guest's FPU / MSA context
1373 		 * is live), restore FCR31 / MSACSR.
1374 		 *
1375 		 * This should be before returning to the guest exception
1376 		 * vector, as it may well cause an [MSA] FP exception if there
1377 		 * are pending exception bits unmasked. (see
1378 		 * kvm_mips_csr_die_notifier() for how that is handled).
1379 		 */
1380 		if (kvm_mips_guest_has_fpu(&vcpu->arch) &&
1381 		    read_c0_status() & ST0_CU1)
1382 			__kvm_restore_fcsr(&vcpu->arch);
1383 
1384 		if (kvm_mips_guest_has_msa(&vcpu->arch) &&
1385 		    read_c0_config5() & MIPS_CONF5_MSAEN)
1386 			__kvm_restore_msacsr(&vcpu->arch);
1387 	}
1388 	return ret;
1389 }
1390 
1391 int noinstr kvm_mips_handle_exit(struct kvm_vcpu *vcpu)
1392 {
1393 	int ret;
1394 
1395 	guest_state_exit_irqoff();
1396 	ret = __kvm_mips_handle_exit(vcpu);
1397 	guest_state_enter_irqoff();
1398 
1399 	return ret;
1400 }
1401 
1402 /* Enable FPU for guest and restore context */
1403 void kvm_own_fpu(struct kvm_vcpu *vcpu)
1404 {
1405 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1406 	unsigned int sr, cfg5;
1407 
1408 	preempt_disable();
1409 
1410 	sr = kvm_read_c0_guest_status(cop0);
1411 
1412 	/*
1413 	 * If MSA state is already live, it is undefined how it interacts with
1414 	 * FR=0 FPU state, and we don't want to hit reserved instruction
1415 	 * exceptions trying to save the MSA state later when CU=1 && FR=1, so
1416 	 * play it safe and save it first.
1417 	 */
1418 	if (cpu_has_msa && sr & ST0_CU1 && !(sr & ST0_FR) &&
1419 	    vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA)
1420 		kvm_lose_fpu(vcpu);
1421 
1422 	/*
1423 	 * Enable FPU for guest
1424 	 * We set FR and FRE according to guest context
1425 	 */
1426 	change_c0_status(ST0_CU1 | ST0_FR, sr);
1427 	if (cpu_has_fre) {
1428 		cfg5 = kvm_read_c0_guest_config5(cop0);
1429 		change_c0_config5(MIPS_CONF5_FRE, cfg5);
1430 	}
1431 	enable_fpu_hazard();
1432 
1433 	/* If guest FPU state not active, restore it now */
1434 	if (!(vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU)) {
1435 		__kvm_restore_fpu(&vcpu->arch);
1436 		vcpu->arch.aux_inuse |= KVM_MIPS_AUX_FPU;
1437 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE, KVM_TRACE_AUX_FPU);
1438 	} else {
1439 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_ENABLE, KVM_TRACE_AUX_FPU);
1440 	}
1441 
1442 	preempt_enable();
1443 }
1444 
1445 #ifdef CONFIG_CPU_HAS_MSA
1446 /* Enable MSA for guest and restore context */
1447 void kvm_own_msa(struct kvm_vcpu *vcpu)
1448 {
1449 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1450 	unsigned int sr, cfg5;
1451 
1452 	preempt_disable();
1453 
1454 	/*
1455 	 * Enable FPU if enabled in guest, since we're restoring FPU context
1456 	 * anyway. We set FR and FRE according to guest context.
1457 	 */
1458 	if (kvm_mips_guest_has_fpu(&vcpu->arch)) {
1459 		sr = kvm_read_c0_guest_status(cop0);
1460 
1461 		/*
1462 		 * If FR=0 FPU state is already live, it is undefined how it
1463 		 * interacts with MSA state, so play it safe and save it first.
1464 		 */
1465 		if (!(sr & ST0_FR) &&
1466 		    (vcpu->arch.aux_inuse & (KVM_MIPS_AUX_FPU |
1467 				KVM_MIPS_AUX_MSA)) == KVM_MIPS_AUX_FPU)
1468 			kvm_lose_fpu(vcpu);
1469 
1470 		change_c0_status(ST0_CU1 | ST0_FR, sr);
1471 		if (sr & ST0_CU1 && cpu_has_fre) {
1472 			cfg5 = kvm_read_c0_guest_config5(cop0);
1473 			change_c0_config5(MIPS_CONF5_FRE, cfg5);
1474 		}
1475 	}
1476 
1477 	/* Enable MSA for guest */
1478 	set_c0_config5(MIPS_CONF5_MSAEN);
1479 	enable_fpu_hazard();
1480 
1481 	switch (vcpu->arch.aux_inuse & (KVM_MIPS_AUX_FPU | KVM_MIPS_AUX_MSA)) {
1482 	case KVM_MIPS_AUX_FPU:
1483 		/*
1484 		 * Guest FPU state already loaded, only restore upper MSA state
1485 		 */
1486 		__kvm_restore_msa_upper(&vcpu->arch);
1487 		vcpu->arch.aux_inuse |= KVM_MIPS_AUX_MSA;
1488 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE, KVM_TRACE_AUX_MSA);
1489 		break;
1490 	case 0:
1491 		/* Neither FPU or MSA already active, restore full MSA state */
1492 		__kvm_restore_msa(&vcpu->arch);
1493 		vcpu->arch.aux_inuse |= KVM_MIPS_AUX_MSA;
1494 		if (kvm_mips_guest_has_fpu(&vcpu->arch))
1495 			vcpu->arch.aux_inuse |= KVM_MIPS_AUX_FPU;
1496 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE,
1497 			      KVM_TRACE_AUX_FPU_MSA);
1498 		break;
1499 	default:
1500 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_ENABLE, KVM_TRACE_AUX_MSA);
1501 		break;
1502 	}
1503 
1504 	preempt_enable();
1505 }
1506 #endif
1507 
1508 /* Drop FPU & MSA without saving it */
1509 void kvm_drop_fpu(struct kvm_vcpu *vcpu)
1510 {
1511 	preempt_disable();
1512 	if (cpu_has_msa && vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA) {
1513 		disable_msa();
1514 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_DISCARD, KVM_TRACE_AUX_MSA);
1515 		vcpu->arch.aux_inuse &= ~KVM_MIPS_AUX_MSA;
1516 	}
1517 	if (vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU) {
1518 		clear_c0_status(ST0_CU1 | ST0_FR);
1519 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_DISCARD, KVM_TRACE_AUX_FPU);
1520 		vcpu->arch.aux_inuse &= ~KVM_MIPS_AUX_FPU;
1521 	}
1522 	preempt_enable();
1523 }
1524 
1525 /* Save and disable FPU & MSA */
1526 void kvm_lose_fpu(struct kvm_vcpu *vcpu)
1527 {
1528 	/*
1529 	 * With T&E, FPU & MSA get disabled in root context (hardware) when it
1530 	 * is disabled in guest context (software), but the register state in
1531 	 * the hardware may still be in use.
1532 	 * This is why we explicitly re-enable the hardware before saving.
1533 	 */
1534 
1535 	preempt_disable();
1536 	if (cpu_has_msa && vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA) {
1537 		__kvm_save_msa(&vcpu->arch);
1538 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_SAVE, KVM_TRACE_AUX_FPU_MSA);
1539 
1540 		/* Disable MSA & FPU */
1541 		disable_msa();
1542 		if (vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU) {
1543 			clear_c0_status(ST0_CU1 | ST0_FR);
1544 			disable_fpu_hazard();
1545 		}
1546 		vcpu->arch.aux_inuse &= ~(KVM_MIPS_AUX_FPU | KVM_MIPS_AUX_MSA);
1547 	} else if (vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU) {
1548 		__kvm_save_fpu(&vcpu->arch);
1549 		vcpu->arch.aux_inuse &= ~KVM_MIPS_AUX_FPU;
1550 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_SAVE, KVM_TRACE_AUX_FPU);
1551 
1552 		/* Disable FPU */
1553 		clear_c0_status(ST0_CU1 | ST0_FR);
1554 		disable_fpu_hazard();
1555 	}
1556 	preempt_enable();
1557 }
1558 
1559 /*
1560  * Step over a specific ctc1 to FCSR and a specific ctcmsa to MSACSR which are
1561  * used to restore guest FCSR/MSACSR state and may trigger a "harmless" FP/MSAFP
1562  * exception if cause bits are set in the value being written.
1563  */
1564 static int kvm_mips_csr_die_notify(struct notifier_block *self,
1565 				   unsigned long cmd, void *ptr)
1566 {
1567 	struct die_args *args = (struct die_args *)ptr;
1568 	struct pt_regs *regs = args->regs;
1569 	unsigned long pc;
1570 
1571 	/* Only interested in FPE and MSAFPE */
1572 	if (cmd != DIE_FP && cmd != DIE_MSAFP)
1573 		return NOTIFY_DONE;
1574 
1575 	/* Return immediately if guest context isn't active */
1576 	if (!(current->flags & PF_VCPU))
1577 		return NOTIFY_DONE;
1578 
1579 	/* Should never get here from user mode */
1580 	BUG_ON(user_mode(regs));
1581 
1582 	pc = instruction_pointer(regs);
1583 	switch (cmd) {
1584 	case DIE_FP:
1585 		/* match 2nd instruction in __kvm_restore_fcsr */
1586 		if (pc != (unsigned long)&__kvm_restore_fcsr + 4)
1587 			return NOTIFY_DONE;
1588 		break;
1589 	case DIE_MSAFP:
1590 		/* match 2nd/3rd instruction in __kvm_restore_msacsr */
1591 		if (!cpu_has_msa ||
1592 		    pc < (unsigned long)&__kvm_restore_msacsr + 4 ||
1593 		    pc > (unsigned long)&__kvm_restore_msacsr + 8)
1594 			return NOTIFY_DONE;
1595 		break;
1596 	}
1597 
1598 	/* Move PC forward a little and continue executing */
1599 	instruction_pointer(regs) += 4;
1600 
1601 	return NOTIFY_STOP;
1602 }
1603 
1604 static struct notifier_block kvm_mips_csr_die_notifier = {
1605 	.notifier_call = kvm_mips_csr_die_notify,
1606 };
1607 
1608 static u32 kvm_default_priority_to_irq[MIPS_EXC_MAX] = {
1609 	[MIPS_EXC_INT_TIMER] = C_IRQ5,
1610 	[MIPS_EXC_INT_IO_1]  = C_IRQ0,
1611 	[MIPS_EXC_INT_IPI_1] = C_IRQ1,
1612 	[MIPS_EXC_INT_IPI_2] = C_IRQ2,
1613 };
1614 
1615 static u32 kvm_loongson3_priority_to_irq[MIPS_EXC_MAX] = {
1616 	[MIPS_EXC_INT_TIMER] = C_IRQ5,
1617 	[MIPS_EXC_INT_IO_1]  = C_IRQ0,
1618 	[MIPS_EXC_INT_IO_2]  = C_IRQ1,
1619 	[MIPS_EXC_INT_IPI_1] = C_IRQ4,
1620 };
1621 
1622 u32 *kvm_priority_to_irq = kvm_default_priority_to_irq;
1623 
1624 u32 kvm_irq_to_priority(u32 irq)
1625 {
1626 	int i;
1627 
1628 	for (i = MIPS_EXC_INT_TIMER; i < MIPS_EXC_MAX; i++) {
1629 		if (kvm_priority_to_irq[i] == (1 << (irq + 8)))
1630 			return i;
1631 	}
1632 
1633 	return MIPS_EXC_MAX;
1634 }
1635 
1636 static int __init kvm_mips_init(void)
1637 {
1638 	int ret;
1639 
1640 	if (cpu_has_mmid) {
1641 		pr_warn("KVM does not yet support MMIDs. KVM Disabled\n");
1642 		return -EOPNOTSUPP;
1643 	}
1644 
1645 	ret = kvm_mips_entry_setup();
1646 	if (ret)
1647 		return ret;
1648 
1649 	ret = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1650 
1651 	if (ret)
1652 		return ret;
1653 
1654 	if (boot_cpu_type() == CPU_LOONGSON64)
1655 		kvm_priority_to_irq = kvm_loongson3_priority_to_irq;
1656 
1657 	register_die_notifier(&kvm_mips_csr_die_notifier);
1658 
1659 	return 0;
1660 }
1661 
1662 static void __exit kvm_mips_exit(void)
1663 {
1664 	kvm_exit();
1665 
1666 	unregister_die_notifier(&kvm_mips_csr_die_notifier);
1667 }
1668 
1669 module_init(kvm_mips_init);
1670 module_exit(kvm_mips_exit);
1671 
1672 EXPORT_TRACEPOINT_SYMBOL(kvm_exit);
1673