xref: /openbmc/linux/arch/mips/kvm/mips.c (revision 87875c10)
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 static_assert(ARRAY_SIZE(kvm_vm_stats_desc) ==
45 		sizeof(struct kvm_vm_stat) / sizeof(u64));
46 
47 const struct kvm_stats_header kvm_vm_stats_header = {
48 	.name_size = KVM_STATS_NAME_SIZE,
49 	.num_desc = ARRAY_SIZE(kvm_vm_stats_desc),
50 	.id_offset = sizeof(struct kvm_stats_header),
51 	.desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
52 	.data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
53 		       sizeof(kvm_vm_stats_desc),
54 };
55 
56 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
57 	KVM_GENERIC_VCPU_STATS(),
58 	STATS_DESC_COUNTER(VCPU, wait_exits),
59 	STATS_DESC_COUNTER(VCPU, cache_exits),
60 	STATS_DESC_COUNTER(VCPU, signal_exits),
61 	STATS_DESC_COUNTER(VCPU, int_exits),
62 	STATS_DESC_COUNTER(VCPU, cop_unusable_exits),
63 	STATS_DESC_COUNTER(VCPU, tlbmod_exits),
64 	STATS_DESC_COUNTER(VCPU, tlbmiss_ld_exits),
65 	STATS_DESC_COUNTER(VCPU, tlbmiss_st_exits),
66 	STATS_DESC_COUNTER(VCPU, addrerr_st_exits),
67 	STATS_DESC_COUNTER(VCPU, addrerr_ld_exits),
68 	STATS_DESC_COUNTER(VCPU, syscall_exits),
69 	STATS_DESC_COUNTER(VCPU, resvd_inst_exits),
70 	STATS_DESC_COUNTER(VCPU, break_inst_exits),
71 	STATS_DESC_COUNTER(VCPU, trap_inst_exits),
72 	STATS_DESC_COUNTER(VCPU, msa_fpe_exits),
73 	STATS_DESC_COUNTER(VCPU, fpe_exits),
74 	STATS_DESC_COUNTER(VCPU, msa_disabled_exits),
75 	STATS_DESC_COUNTER(VCPU, flush_dcache_exits),
76 	STATS_DESC_COUNTER(VCPU, vz_gpsi_exits),
77 	STATS_DESC_COUNTER(VCPU, vz_gsfc_exits),
78 	STATS_DESC_COUNTER(VCPU, vz_hc_exits),
79 	STATS_DESC_COUNTER(VCPU, vz_grr_exits),
80 	STATS_DESC_COUNTER(VCPU, vz_gva_exits),
81 	STATS_DESC_COUNTER(VCPU, vz_ghfc_exits),
82 	STATS_DESC_COUNTER(VCPU, vz_gpa_exits),
83 	STATS_DESC_COUNTER(VCPU, vz_resvd_exits),
84 #ifdef CONFIG_CPU_LOONGSON64
85 	STATS_DESC_COUNTER(VCPU, vz_cpucfg_exits),
86 #endif
87 };
88 static_assert(ARRAY_SIZE(kvm_vcpu_stats_desc) ==
89 		sizeof(struct kvm_vcpu_stat) / sizeof(u64));
90 
91 const struct kvm_stats_header kvm_vcpu_stats_header = {
92 	.name_size = KVM_STATS_NAME_SIZE,
93 	.num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc),
94 	.id_offset = sizeof(struct kvm_stats_header),
95 	.desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
96 	.data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
97 		       sizeof(kvm_vcpu_stats_desc),
98 };
99 
100 bool kvm_trace_guest_mode_change;
101 
102 int kvm_guest_mode_change_trace_reg(void)
103 {
104 	kvm_trace_guest_mode_change = true;
105 	return 0;
106 }
107 
108 void kvm_guest_mode_change_trace_unreg(void)
109 {
110 	kvm_trace_guest_mode_change = false;
111 }
112 
113 /*
114  * XXXKYMA: We are simulatoring a processor that has the WII bit set in
115  * Config7, so we are "runnable" if interrupts are pending
116  */
117 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
118 {
119 	return !!(vcpu->arch.pending_exceptions);
120 }
121 
122 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
123 {
124 	return false;
125 }
126 
127 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
128 {
129 	return 1;
130 }
131 
132 int kvm_arch_hardware_enable(void)
133 {
134 	return kvm_mips_callbacks->hardware_enable();
135 }
136 
137 void kvm_arch_hardware_disable(void)
138 {
139 	kvm_mips_callbacks->hardware_disable();
140 }
141 
142 int kvm_arch_hardware_setup(void *opaque)
143 {
144 	return 0;
145 }
146 
147 int kvm_arch_check_processor_compat(void *opaque)
148 {
149 	return 0;
150 }
151 
152 extern void kvm_init_loongson_ipi(struct kvm *kvm);
153 
154 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
155 {
156 	switch (type) {
157 	case KVM_VM_MIPS_AUTO:
158 		break;
159 	case KVM_VM_MIPS_VZ:
160 		break;
161 	default:
162 		/* Unsupported KVM type */
163 		return -EINVAL;
164 	}
165 
166 	/* Allocate page table to map GPA -> RPA */
167 	kvm->arch.gpa_mm.pgd = kvm_pgd_alloc();
168 	if (!kvm->arch.gpa_mm.pgd)
169 		return -ENOMEM;
170 
171 #ifdef CONFIG_CPU_LOONGSON64
172 	kvm_init_loongson_ipi(kvm);
173 #endif
174 
175 	return 0;
176 }
177 
178 void kvm_mips_free_vcpus(struct kvm *kvm)
179 {
180 	unsigned int i;
181 	struct kvm_vcpu *vcpu;
182 
183 	kvm_for_each_vcpu(i, vcpu, kvm) {
184 		kvm_vcpu_destroy(vcpu);
185 	}
186 
187 	mutex_lock(&kvm->lock);
188 
189 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
190 		kvm->vcpus[i] = NULL;
191 
192 	atomic_set(&kvm->online_vcpus, 0);
193 
194 	mutex_unlock(&kvm->lock);
195 }
196 
197 static void kvm_mips_free_gpa_pt(struct kvm *kvm)
198 {
199 	/* It should always be safe to remove after flushing the whole range */
200 	WARN_ON(!kvm_mips_flush_gpa_pt(kvm, 0, ~0));
201 	pgd_free(NULL, kvm->arch.gpa_mm.pgd);
202 }
203 
204 void kvm_arch_destroy_vm(struct kvm *kvm)
205 {
206 	kvm_mips_free_vcpus(kvm);
207 	kvm_mips_free_gpa_pt(kvm);
208 }
209 
210 long kvm_arch_dev_ioctl(struct file *filp, unsigned int ioctl,
211 			unsigned long arg)
212 {
213 	return -ENOIOCTLCMD;
214 }
215 
216 void kvm_arch_flush_shadow_all(struct kvm *kvm)
217 {
218 	/* Flush whole GPA */
219 	kvm_mips_flush_gpa_pt(kvm, 0, ~0);
220 	kvm_flush_remote_tlbs(kvm);
221 }
222 
223 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
224 				   struct kvm_memory_slot *slot)
225 {
226 	/*
227 	 * The slot has been made invalid (ready for moving or deletion), so we
228 	 * need to ensure that it can no longer be accessed by any guest VCPUs.
229 	 */
230 
231 	spin_lock(&kvm->mmu_lock);
232 	/* Flush slot from GPA */
233 	kvm_mips_flush_gpa_pt(kvm, slot->base_gfn,
234 			      slot->base_gfn + slot->npages - 1);
235 	kvm_arch_flush_remote_tlbs_memslot(kvm, slot);
236 	spin_unlock(&kvm->mmu_lock);
237 }
238 
239 int kvm_arch_prepare_memory_region(struct kvm *kvm,
240 				   struct kvm_memory_slot *memslot,
241 				   const struct kvm_userspace_memory_region *mem,
242 				   enum kvm_mr_change change)
243 {
244 	return 0;
245 }
246 
247 void kvm_arch_commit_memory_region(struct kvm *kvm,
248 				   const struct kvm_userspace_memory_region *mem,
249 				   struct kvm_memory_slot *old,
250 				   const struct kvm_memory_slot *new,
251 				   enum kvm_mr_change change)
252 {
253 	int needs_flush;
254 
255 	kvm_debug("%s: kvm: %p slot: %d, GPA: %llx, size: %llx, QVA: %llx\n",
256 		  __func__, kvm, mem->slot, mem->guest_phys_addr,
257 		  mem->memory_size, mem->userspace_addr);
258 
259 	/*
260 	 * If dirty page logging is enabled, write protect all pages in the slot
261 	 * ready for dirty logging.
262 	 *
263 	 * There is no need to do this in any of the following cases:
264 	 * CREATE:	No dirty mappings will already exist.
265 	 * MOVE/DELETE:	The old mappings will already have been cleaned up by
266 	 *		kvm_arch_flush_shadow_memslot()
267 	 */
268 	if (change == KVM_MR_FLAGS_ONLY &&
269 	    (!(old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
270 	     new->flags & KVM_MEM_LOG_DIRTY_PAGES)) {
271 		spin_lock(&kvm->mmu_lock);
272 		/* Write protect GPA page table entries */
273 		needs_flush = kvm_mips_mkclean_gpa_pt(kvm, new->base_gfn,
274 					new->base_gfn + new->npages - 1);
275 		if (needs_flush)
276 			kvm_arch_flush_remote_tlbs_memslot(kvm, new);
277 		spin_unlock(&kvm->mmu_lock);
278 	}
279 }
280 
281 static inline void dump_handler(const char *symbol, void *start, void *end)
282 {
283 	u32 *p;
284 
285 	pr_debug("LEAF(%s)\n", symbol);
286 
287 	pr_debug("\t.set push\n");
288 	pr_debug("\t.set noreorder\n");
289 
290 	for (p = start; p < (u32 *)end; ++p)
291 		pr_debug("\t.word\t0x%08x\t\t# %p\n", *p, p);
292 
293 	pr_debug("\t.set\tpop\n");
294 
295 	pr_debug("\tEND(%s)\n", symbol);
296 }
297 
298 /* low level hrtimer wake routine */
299 static enum hrtimer_restart kvm_mips_comparecount_wakeup(struct hrtimer *timer)
300 {
301 	struct kvm_vcpu *vcpu;
302 
303 	vcpu = container_of(timer, struct kvm_vcpu, arch.comparecount_timer);
304 
305 	kvm_mips_callbacks->queue_timer_int(vcpu);
306 
307 	vcpu->arch.wait = 0;
308 	rcuwait_wake_up(&vcpu->wait);
309 
310 	return kvm_mips_count_timeout(vcpu);
311 }
312 
313 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
314 {
315 	return 0;
316 }
317 
318 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
319 {
320 	int err, size;
321 	void *gebase, *p, *handler, *refill_start, *refill_end;
322 	int i;
323 
324 	kvm_debug("kvm @ %p: create cpu %d at %p\n",
325 		  vcpu->kvm, vcpu->vcpu_id, vcpu);
326 
327 	err = kvm_mips_callbacks->vcpu_init(vcpu);
328 	if (err)
329 		return err;
330 
331 	hrtimer_init(&vcpu->arch.comparecount_timer, CLOCK_MONOTONIC,
332 		     HRTIMER_MODE_REL);
333 	vcpu->arch.comparecount_timer.function = kvm_mips_comparecount_wakeup;
334 
335 	/*
336 	 * Allocate space for host mode exception handlers that handle
337 	 * guest mode exits
338 	 */
339 	if (cpu_has_veic || cpu_has_vint)
340 		size = 0x200 + VECTORSPACING * 64;
341 	else
342 		size = 0x4000;
343 
344 	gebase = kzalloc(ALIGN(size, PAGE_SIZE), GFP_KERNEL);
345 
346 	if (!gebase) {
347 		err = -ENOMEM;
348 		goto out_uninit_vcpu;
349 	}
350 	kvm_debug("Allocated %d bytes for KVM Exception Handlers @ %p\n",
351 		  ALIGN(size, PAGE_SIZE), gebase);
352 
353 	/*
354 	 * Check new ebase actually fits in CP0_EBase. The lack of a write gate
355 	 * limits us to the low 512MB of physical address space. If the memory
356 	 * we allocate is out of range, just give up now.
357 	 */
358 	if (!cpu_has_ebase_wg && virt_to_phys(gebase) >= 0x20000000) {
359 		kvm_err("CP0_EBase.WG required for guest exception base %pK\n",
360 			gebase);
361 		err = -ENOMEM;
362 		goto out_free_gebase;
363 	}
364 
365 	/* Save new ebase */
366 	vcpu->arch.guest_ebase = gebase;
367 
368 	/* Build guest exception vectors dynamically in unmapped memory */
369 	handler = gebase + 0x2000;
370 
371 	/* TLB refill (or XTLB refill on 64-bit VZ where KX=1) */
372 	refill_start = gebase;
373 	if (IS_ENABLED(CONFIG_64BIT))
374 		refill_start += 0x080;
375 	refill_end = kvm_mips_build_tlb_refill_exception(refill_start, handler);
376 
377 	/* General Exception Entry point */
378 	kvm_mips_build_exception(gebase + 0x180, handler);
379 
380 	/* For vectored interrupts poke the exception code @ all offsets 0-7 */
381 	for (i = 0; i < 8; i++) {
382 		kvm_debug("L1 Vectored handler @ %p\n",
383 			  gebase + 0x200 + (i * VECTORSPACING));
384 		kvm_mips_build_exception(gebase + 0x200 + i * VECTORSPACING,
385 					 handler);
386 	}
387 
388 	/* General exit handler */
389 	p = handler;
390 	p = kvm_mips_build_exit(p);
391 
392 	/* Guest entry routine */
393 	vcpu->arch.vcpu_run = p;
394 	p = kvm_mips_build_vcpu_run(p);
395 
396 	/* Dump the generated code */
397 	pr_debug("#include <asm/asm.h>\n");
398 	pr_debug("#include <asm/regdef.h>\n");
399 	pr_debug("\n");
400 	dump_handler("kvm_vcpu_run", vcpu->arch.vcpu_run, p);
401 	dump_handler("kvm_tlb_refill", refill_start, refill_end);
402 	dump_handler("kvm_gen_exc", gebase + 0x180, gebase + 0x200);
403 	dump_handler("kvm_exit", gebase + 0x2000, vcpu->arch.vcpu_run);
404 
405 	/* Invalidate the icache for these ranges */
406 	flush_icache_range((unsigned long)gebase,
407 			   (unsigned long)gebase + ALIGN(size, PAGE_SIZE));
408 
409 	/* Init */
410 	vcpu->arch.last_sched_cpu = -1;
411 	vcpu->arch.last_exec_cpu = -1;
412 
413 	/* Initial guest state */
414 	err = kvm_mips_callbacks->vcpu_setup(vcpu);
415 	if (err)
416 		goto out_free_gebase;
417 
418 	return 0;
419 
420 out_free_gebase:
421 	kfree(gebase);
422 out_uninit_vcpu:
423 	kvm_mips_callbacks->vcpu_uninit(vcpu);
424 	return err;
425 }
426 
427 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
428 {
429 	hrtimer_cancel(&vcpu->arch.comparecount_timer);
430 
431 	kvm_mips_dump_stats(vcpu);
432 
433 	kvm_mmu_free_memory_caches(vcpu);
434 	kfree(vcpu->arch.guest_ebase);
435 
436 	kvm_mips_callbacks->vcpu_uninit(vcpu);
437 }
438 
439 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
440 					struct kvm_guest_debug *dbg)
441 {
442 	return -ENOIOCTLCMD;
443 }
444 
445 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
446 {
447 	int r = -EINTR;
448 
449 	vcpu_load(vcpu);
450 
451 	kvm_sigset_activate(vcpu);
452 
453 	if (vcpu->mmio_needed) {
454 		if (!vcpu->mmio_is_write)
455 			kvm_mips_complete_mmio_load(vcpu);
456 		vcpu->mmio_needed = 0;
457 	}
458 
459 	if (vcpu->run->immediate_exit)
460 		goto out;
461 
462 	lose_fpu(1);
463 
464 	local_irq_disable();
465 	guest_enter_irqoff();
466 	trace_kvm_enter(vcpu);
467 
468 	/*
469 	 * Make sure the read of VCPU requests in vcpu_run() callback is not
470 	 * reordered ahead of the write to vcpu->mode, or we could miss a TLB
471 	 * flush request while the requester sees the VCPU as outside of guest
472 	 * mode and not needing an IPI.
473 	 */
474 	smp_store_mb(vcpu->mode, IN_GUEST_MODE);
475 
476 	r = kvm_mips_callbacks->vcpu_run(vcpu);
477 
478 	trace_kvm_out(vcpu);
479 	guest_exit_irqoff();
480 	local_irq_enable();
481 
482 out:
483 	kvm_sigset_deactivate(vcpu);
484 
485 	vcpu_put(vcpu);
486 	return r;
487 }
488 
489 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
490 			     struct kvm_mips_interrupt *irq)
491 {
492 	int intr = (int)irq->irq;
493 	struct kvm_vcpu *dvcpu = NULL;
494 
495 	if (intr == kvm_priority_to_irq[MIPS_EXC_INT_IPI_1] ||
496 	    intr == kvm_priority_to_irq[MIPS_EXC_INT_IPI_2] ||
497 	    intr == (-kvm_priority_to_irq[MIPS_EXC_INT_IPI_1]) ||
498 	    intr == (-kvm_priority_to_irq[MIPS_EXC_INT_IPI_2]))
499 		kvm_debug("%s: CPU: %d, INTR: %d\n", __func__, irq->cpu,
500 			  (int)intr);
501 
502 	if (irq->cpu == -1)
503 		dvcpu = vcpu;
504 	else
505 		dvcpu = vcpu->kvm->vcpus[irq->cpu];
506 
507 	if (intr == 2 || intr == 3 || intr == 4 || intr == 6) {
508 		kvm_mips_callbacks->queue_io_int(dvcpu, irq);
509 
510 	} else if (intr == -2 || intr == -3 || intr == -4 || intr == -6) {
511 		kvm_mips_callbacks->dequeue_io_int(dvcpu, irq);
512 	} else {
513 		kvm_err("%s: invalid interrupt ioctl (%d:%d)\n", __func__,
514 			irq->cpu, irq->irq);
515 		return -EINVAL;
516 	}
517 
518 	dvcpu->arch.wait = 0;
519 
520 	rcuwait_wake_up(&dvcpu->wait);
521 
522 	return 0;
523 }
524 
525 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
526 				    struct kvm_mp_state *mp_state)
527 {
528 	return -ENOIOCTLCMD;
529 }
530 
531 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
532 				    struct kvm_mp_state *mp_state)
533 {
534 	return -ENOIOCTLCMD;
535 }
536 
537 static u64 kvm_mips_get_one_regs[] = {
538 	KVM_REG_MIPS_R0,
539 	KVM_REG_MIPS_R1,
540 	KVM_REG_MIPS_R2,
541 	KVM_REG_MIPS_R3,
542 	KVM_REG_MIPS_R4,
543 	KVM_REG_MIPS_R5,
544 	KVM_REG_MIPS_R6,
545 	KVM_REG_MIPS_R7,
546 	KVM_REG_MIPS_R8,
547 	KVM_REG_MIPS_R9,
548 	KVM_REG_MIPS_R10,
549 	KVM_REG_MIPS_R11,
550 	KVM_REG_MIPS_R12,
551 	KVM_REG_MIPS_R13,
552 	KVM_REG_MIPS_R14,
553 	KVM_REG_MIPS_R15,
554 	KVM_REG_MIPS_R16,
555 	KVM_REG_MIPS_R17,
556 	KVM_REG_MIPS_R18,
557 	KVM_REG_MIPS_R19,
558 	KVM_REG_MIPS_R20,
559 	KVM_REG_MIPS_R21,
560 	KVM_REG_MIPS_R22,
561 	KVM_REG_MIPS_R23,
562 	KVM_REG_MIPS_R24,
563 	KVM_REG_MIPS_R25,
564 	KVM_REG_MIPS_R26,
565 	KVM_REG_MIPS_R27,
566 	KVM_REG_MIPS_R28,
567 	KVM_REG_MIPS_R29,
568 	KVM_REG_MIPS_R30,
569 	KVM_REG_MIPS_R31,
570 
571 #ifndef CONFIG_CPU_MIPSR6
572 	KVM_REG_MIPS_HI,
573 	KVM_REG_MIPS_LO,
574 #endif
575 	KVM_REG_MIPS_PC,
576 };
577 
578 static u64 kvm_mips_get_one_regs_fpu[] = {
579 	KVM_REG_MIPS_FCR_IR,
580 	KVM_REG_MIPS_FCR_CSR,
581 };
582 
583 static u64 kvm_mips_get_one_regs_msa[] = {
584 	KVM_REG_MIPS_MSA_IR,
585 	KVM_REG_MIPS_MSA_CSR,
586 };
587 
588 static unsigned long kvm_mips_num_regs(struct kvm_vcpu *vcpu)
589 {
590 	unsigned long ret;
591 
592 	ret = ARRAY_SIZE(kvm_mips_get_one_regs);
593 	if (kvm_mips_guest_can_have_fpu(&vcpu->arch)) {
594 		ret += ARRAY_SIZE(kvm_mips_get_one_regs_fpu) + 48;
595 		/* odd doubles */
596 		if (boot_cpu_data.fpu_id & MIPS_FPIR_F64)
597 			ret += 16;
598 	}
599 	if (kvm_mips_guest_can_have_msa(&vcpu->arch))
600 		ret += ARRAY_SIZE(kvm_mips_get_one_regs_msa) + 32;
601 	ret += kvm_mips_callbacks->num_regs(vcpu);
602 
603 	return ret;
604 }
605 
606 static int kvm_mips_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *indices)
607 {
608 	u64 index;
609 	unsigned int i;
610 
611 	if (copy_to_user(indices, kvm_mips_get_one_regs,
612 			 sizeof(kvm_mips_get_one_regs)))
613 		return -EFAULT;
614 	indices += ARRAY_SIZE(kvm_mips_get_one_regs);
615 
616 	if (kvm_mips_guest_can_have_fpu(&vcpu->arch)) {
617 		if (copy_to_user(indices, kvm_mips_get_one_regs_fpu,
618 				 sizeof(kvm_mips_get_one_regs_fpu)))
619 			return -EFAULT;
620 		indices += ARRAY_SIZE(kvm_mips_get_one_regs_fpu);
621 
622 		for (i = 0; i < 32; ++i) {
623 			index = KVM_REG_MIPS_FPR_32(i);
624 			if (copy_to_user(indices, &index, sizeof(index)))
625 				return -EFAULT;
626 			++indices;
627 
628 			/* skip odd doubles if no F64 */
629 			if (i & 1 && !(boot_cpu_data.fpu_id & MIPS_FPIR_F64))
630 				continue;
631 
632 			index = KVM_REG_MIPS_FPR_64(i);
633 			if (copy_to_user(indices, &index, sizeof(index)))
634 				return -EFAULT;
635 			++indices;
636 		}
637 	}
638 
639 	if (kvm_mips_guest_can_have_msa(&vcpu->arch)) {
640 		if (copy_to_user(indices, kvm_mips_get_one_regs_msa,
641 				 sizeof(kvm_mips_get_one_regs_msa)))
642 			return -EFAULT;
643 		indices += ARRAY_SIZE(kvm_mips_get_one_regs_msa);
644 
645 		for (i = 0; i < 32; ++i) {
646 			index = KVM_REG_MIPS_VEC_128(i);
647 			if (copy_to_user(indices, &index, sizeof(index)))
648 				return -EFAULT;
649 			++indices;
650 		}
651 	}
652 
653 	return kvm_mips_callbacks->copy_reg_indices(vcpu, indices);
654 }
655 
656 static int kvm_mips_get_reg(struct kvm_vcpu *vcpu,
657 			    const struct kvm_one_reg *reg)
658 {
659 	struct mips_coproc *cop0 = vcpu->arch.cop0;
660 	struct mips_fpu_struct *fpu = &vcpu->arch.fpu;
661 	int ret;
662 	s64 v;
663 	s64 vs[2];
664 	unsigned int idx;
665 
666 	switch (reg->id) {
667 	/* General purpose registers */
668 	case KVM_REG_MIPS_R0 ... KVM_REG_MIPS_R31:
669 		v = (long)vcpu->arch.gprs[reg->id - KVM_REG_MIPS_R0];
670 		break;
671 #ifndef CONFIG_CPU_MIPSR6
672 	case KVM_REG_MIPS_HI:
673 		v = (long)vcpu->arch.hi;
674 		break;
675 	case KVM_REG_MIPS_LO:
676 		v = (long)vcpu->arch.lo;
677 		break;
678 #endif
679 	case KVM_REG_MIPS_PC:
680 		v = (long)vcpu->arch.pc;
681 		break;
682 
683 	/* Floating point registers */
684 	case KVM_REG_MIPS_FPR_32(0) ... KVM_REG_MIPS_FPR_32(31):
685 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
686 			return -EINVAL;
687 		idx = reg->id - KVM_REG_MIPS_FPR_32(0);
688 		/* Odd singles in top of even double when FR=0 */
689 		if (kvm_read_c0_guest_status(cop0) & ST0_FR)
690 			v = get_fpr32(&fpu->fpr[idx], 0);
691 		else
692 			v = get_fpr32(&fpu->fpr[idx & ~1], idx & 1);
693 		break;
694 	case KVM_REG_MIPS_FPR_64(0) ... KVM_REG_MIPS_FPR_64(31):
695 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
696 			return -EINVAL;
697 		idx = reg->id - KVM_REG_MIPS_FPR_64(0);
698 		/* Can't access odd doubles in FR=0 mode */
699 		if (idx & 1 && !(kvm_read_c0_guest_status(cop0) & ST0_FR))
700 			return -EINVAL;
701 		v = get_fpr64(&fpu->fpr[idx], 0);
702 		break;
703 	case KVM_REG_MIPS_FCR_IR:
704 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
705 			return -EINVAL;
706 		v = boot_cpu_data.fpu_id;
707 		break;
708 	case KVM_REG_MIPS_FCR_CSR:
709 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
710 			return -EINVAL;
711 		v = fpu->fcr31;
712 		break;
713 
714 	/* MIPS SIMD Architecture (MSA) registers */
715 	case KVM_REG_MIPS_VEC_128(0) ... KVM_REG_MIPS_VEC_128(31):
716 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
717 			return -EINVAL;
718 		/* Can't access MSA registers in FR=0 mode */
719 		if (!(kvm_read_c0_guest_status(cop0) & ST0_FR))
720 			return -EINVAL;
721 		idx = reg->id - KVM_REG_MIPS_VEC_128(0);
722 #ifdef CONFIG_CPU_LITTLE_ENDIAN
723 		/* least significant byte first */
724 		vs[0] = get_fpr64(&fpu->fpr[idx], 0);
725 		vs[1] = get_fpr64(&fpu->fpr[idx], 1);
726 #else
727 		/* most significant byte first */
728 		vs[0] = get_fpr64(&fpu->fpr[idx], 1);
729 		vs[1] = get_fpr64(&fpu->fpr[idx], 0);
730 #endif
731 		break;
732 	case KVM_REG_MIPS_MSA_IR:
733 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
734 			return -EINVAL;
735 		v = boot_cpu_data.msa_id;
736 		break;
737 	case KVM_REG_MIPS_MSA_CSR:
738 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
739 			return -EINVAL;
740 		v = fpu->msacsr;
741 		break;
742 
743 	/* registers to be handled specially */
744 	default:
745 		ret = kvm_mips_callbacks->get_one_reg(vcpu, reg, &v);
746 		if (ret)
747 			return ret;
748 		break;
749 	}
750 	if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U64) {
751 		u64 __user *uaddr64 = (u64 __user *)(long)reg->addr;
752 
753 		return put_user(v, uaddr64);
754 	} else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U32) {
755 		u32 __user *uaddr32 = (u32 __user *)(long)reg->addr;
756 		u32 v32 = (u32)v;
757 
758 		return put_user(v32, uaddr32);
759 	} else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U128) {
760 		void __user *uaddr = (void __user *)(long)reg->addr;
761 
762 		return copy_to_user(uaddr, vs, 16) ? -EFAULT : 0;
763 	} else {
764 		return -EINVAL;
765 	}
766 }
767 
768 static int kvm_mips_set_reg(struct kvm_vcpu *vcpu,
769 			    const struct kvm_one_reg *reg)
770 {
771 	struct mips_coproc *cop0 = vcpu->arch.cop0;
772 	struct mips_fpu_struct *fpu = &vcpu->arch.fpu;
773 	s64 v;
774 	s64 vs[2];
775 	unsigned int idx;
776 
777 	if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U64) {
778 		u64 __user *uaddr64 = (u64 __user *)(long)reg->addr;
779 
780 		if (get_user(v, uaddr64) != 0)
781 			return -EFAULT;
782 	} else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U32) {
783 		u32 __user *uaddr32 = (u32 __user *)(long)reg->addr;
784 		s32 v32;
785 
786 		if (get_user(v32, uaddr32) != 0)
787 			return -EFAULT;
788 		v = (s64)v32;
789 	} else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U128) {
790 		void __user *uaddr = (void __user *)(long)reg->addr;
791 
792 		return copy_from_user(vs, uaddr, 16) ? -EFAULT : 0;
793 	} else {
794 		return -EINVAL;
795 	}
796 
797 	switch (reg->id) {
798 	/* General purpose registers */
799 	case KVM_REG_MIPS_R0:
800 		/* Silently ignore requests to set $0 */
801 		break;
802 	case KVM_REG_MIPS_R1 ... KVM_REG_MIPS_R31:
803 		vcpu->arch.gprs[reg->id - KVM_REG_MIPS_R0] = v;
804 		break;
805 #ifndef CONFIG_CPU_MIPSR6
806 	case KVM_REG_MIPS_HI:
807 		vcpu->arch.hi = v;
808 		break;
809 	case KVM_REG_MIPS_LO:
810 		vcpu->arch.lo = v;
811 		break;
812 #endif
813 	case KVM_REG_MIPS_PC:
814 		vcpu->arch.pc = v;
815 		break;
816 
817 	/* Floating point registers */
818 	case KVM_REG_MIPS_FPR_32(0) ... KVM_REG_MIPS_FPR_32(31):
819 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
820 			return -EINVAL;
821 		idx = reg->id - KVM_REG_MIPS_FPR_32(0);
822 		/* Odd singles in top of even double when FR=0 */
823 		if (kvm_read_c0_guest_status(cop0) & ST0_FR)
824 			set_fpr32(&fpu->fpr[idx], 0, v);
825 		else
826 			set_fpr32(&fpu->fpr[idx & ~1], idx & 1, v);
827 		break;
828 	case KVM_REG_MIPS_FPR_64(0) ... KVM_REG_MIPS_FPR_64(31):
829 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
830 			return -EINVAL;
831 		idx = reg->id - KVM_REG_MIPS_FPR_64(0);
832 		/* Can't access odd doubles in FR=0 mode */
833 		if (idx & 1 && !(kvm_read_c0_guest_status(cop0) & ST0_FR))
834 			return -EINVAL;
835 		set_fpr64(&fpu->fpr[idx], 0, v);
836 		break;
837 	case KVM_REG_MIPS_FCR_IR:
838 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
839 			return -EINVAL;
840 		/* Read-only */
841 		break;
842 	case KVM_REG_MIPS_FCR_CSR:
843 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
844 			return -EINVAL;
845 		fpu->fcr31 = v;
846 		break;
847 
848 	/* MIPS SIMD Architecture (MSA) registers */
849 	case KVM_REG_MIPS_VEC_128(0) ... KVM_REG_MIPS_VEC_128(31):
850 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
851 			return -EINVAL;
852 		idx = reg->id - KVM_REG_MIPS_VEC_128(0);
853 #ifdef CONFIG_CPU_LITTLE_ENDIAN
854 		/* least significant byte first */
855 		set_fpr64(&fpu->fpr[idx], 0, vs[0]);
856 		set_fpr64(&fpu->fpr[idx], 1, vs[1]);
857 #else
858 		/* most significant byte first */
859 		set_fpr64(&fpu->fpr[idx], 1, vs[0]);
860 		set_fpr64(&fpu->fpr[idx], 0, vs[1]);
861 #endif
862 		break;
863 	case KVM_REG_MIPS_MSA_IR:
864 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
865 			return -EINVAL;
866 		/* Read-only */
867 		break;
868 	case KVM_REG_MIPS_MSA_CSR:
869 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
870 			return -EINVAL;
871 		fpu->msacsr = v;
872 		break;
873 
874 	/* registers to be handled specially */
875 	default:
876 		return kvm_mips_callbacks->set_one_reg(vcpu, reg, v);
877 	}
878 	return 0;
879 }
880 
881 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
882 				     struct kvm_enable_cap *cap)
883 {
884 	int r = 0;
885 
886 	if (!kvm_vm_ioctl_check_extension(vcpu->kvm, cap->cap))
887 		return -EINVAL;
888 	if (cap->flags)
889 		return -EINVAL;
890 	if (cap->args[0])
891 		return -EINVAL;
892 
893 	switch (cap->cap) {
894 	case KVM_CAP_MIPS_FPU:
895 		vcpu->arch.fpu_enabled = true;
896 		break;
897 	case KVM_CAP_MIPS_MSA:
898 		vcpu->arch.msa_enabled = true;
899 		break;
900 	default:
901 		r = -EINVAL;
902 		break;
903 	}
904 
905 	return r;
906 }
907 
908 long kvm_arch_vcpu_async_ioctl(struct file *filp, unsigned int ioctl,
909 			       unsigned long arg)
910 {
911 	struct kvm_vcpu *vcpu = filp->private_data;
912 	void __user *argp = (void __user *)arg;
913 
914 	if (ioctl == KVM_INTERRUPT) {
915 		struct kvm_mips_interrupt irq;
916 
917 		if (copy_from_user(&irq, argp, sizeof(irq)))
918 			return -EFAULT;
919 		kvm_debug("[%d] %s: irq: %d\n", vcpu->vcpu_id, __func__,
920 			  irq.irq);
921 
922 		return kvm_vcpu_ioctl_interrupt(vcpu, &irq);
923 	}
924 
925 	return -ENOIOCTLCMD;
926 }
927 
928 long kvm_arch_vcpu_ioctl(struct file *filp, unsigned int ioctl,
929 			 unsigned long arg)
930 {
931 	struct kvm_vcpu *vcpu = filp->private_data;
932 	void __user *argp = (void __user *)arg;
933 	long r;
934 
935 	vcpu_load(vcpu);
936 
937 	switch (ioctl) {
938 	case KVM_SET_ONE_REG:
939 	case KVM_GET_ONE_REG: {
940 		struct kvm_one_reg reg;
941 
942 		r = -EFAULT;
943 		if (copy_from_user(&reg, argp, sizeof(reg)))
944 			break;
945 		if (ioctl == KVM_SET_ONE_REG)
946 			r = kvm_mips_set_reg(vcpu, &reg);
947 		else
948 			r = kvm_mips_get_reg(vcpu, &reg);
949 		break;
950 	}
951 	case KVM_GET_REG_LIST: {
952 		struct kvm_reg_list __user *user_list = argp;
953 		struct kvm_reg_list reg_list;
954 		unsigned n;
955 
956 		r = -EFAULT;
957 		if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
958 			break;
959 		n = reg_list.n;
960 		reg_list.n = kvm_mips_num_regs(vcpu);
961 		if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
962 			break;
963 		r = -E2BIG;
964 		if (n < reg_list.n)
965 			break;
966 		r = kvm_mips_copy_reg_indices(vcpu, user_list->reg);
967 		break;
968 	}
969 	case KVM_ENABLE_CAP: {
970 		struct kvm_enable_cap cap;
971 
972 		r = -EFAULT;
973 		if (copy_from_user(&cap, argp, sizeof(cap)))
974 			break;
975 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
976 		break;
977 	}
978 	default:
979 		r = -ENOIOCTLCMD;
980 	}
981 
982 	vcpu_put(vcpu);
983 	return r;
984 }
985 
986 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
987 {
988 
989 }
990 
991 int kvm_arch_flush_remote_tlb(struct kvm *kvm)
992 {
993 	kvm_mips_callbacks->prepare_flush_shadow(kvm);
994 	return 1;
995 }
996 
997 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
998 					const struct kvm_memory_slot *memslot)
999 {
1000 	kvm_flush_remote_tlbs(kvm);
1001 }
1002 
1003 long kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
1004 {
1005 	long r;
1006 
1007 	switch (ioctl) {
1008 	default:
1009 		r = -ENOIOCTLCMD;
1010 	}
1011 
1012 	return r;
1013 }
1014 
1015 int kvm_arch_init(void *opaque)
1016 {
1017 	if (kvm_mips_callbacks) {
1018 		kvm_err("kvm: module already exists\n");
1019 		return -EEXIST;
1020 	}
1021 
1022 	return kvm_mips_emulation_init(&kvm_mips_callbacks);
1023 }
1024 
1025 void kvm_arch_exit(void)
1026 {
1027 	kvm_mips_callbacks = NULL;
1028 }
1029 
1030 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1031 				  struct kvm_sregs *sregs)
1032 {
1033 	return -ENOIOCTLCMD;
1034 }
1035 
1036 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1037 				  struct kvm_sregs *sregs)
1038 {
1039 	return -ENOIOCTLCMD;
1040 }
1041 
1042 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
1043 {
1044 }
1045 
1046 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1047 {
1048 	return -ENOIOCTLCMD;
1049 }
1050 
1051 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1052 {
1053 	return -ENOIOCTLCMD;
1054 }
1055 
1056 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
1057 {
1058 	return VM_FAULT_SIGBUS;
1059 }
1060 
1061 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
1062 {
1063 	int r;
1064 
1065 	switch (ext) {
1066 	case KVM_CAP_ONE_REG:
1067 	case KVM_CAP_ENABLE_CAP:
1068 	case KVM_CAP_READONLY_MEM:
1069 	case KVM_CAP_SYNC_MMU:
1070 	case KVM_CAP_IMMEDIATE_EXIT:
1071 		r = 1;
1072 		break;
1073 	case KVM_CAP_NR_VCPUS:
1074 		r = num_online_cpus();
1075 		break;
1076 	case KVM_CAP_MAX_VCPUS:
1077 		r = KVM_MAX_VCPUS;
1078 		break;
1079 	case KVM_CAP_MAX_VCPU_ID:
1080 		r = KVM_MAX_VCPU_ID;
1081 		break;
1082 	case KVM_CAP_MIPS_FPU:
1083 		/* We don't handle systems with inconsistent cpu_has_fpu */
1084 		r = !!raw_cpu_has_fpu;
1085 		break;
1086 	case KVM_CAP_MIPS_MSA:
1087 		/*
1088 		 * We don't support MSA vector partitioning yet:
1089 		 * 1) It would require explicit support which can't be tested
1090 		 *    yet due to lack of support in current hardware.
1091 		 * 2) It extends the state that would need to be saved/restored
1092 		 *    by e.g. QEMU for migration.
1093 		 *
1094 		 * When vector partitioning hardware becomes available, support
1095 		 * could be added by requiring a flag when enabling
1096 		 * KVM_CAP_MIPS_MSA capability to indicate that userland knows
1097 		 * to save/restore the appropriate extra state.
1098 		 */
1099 		r = cpu_has_msa && !(boot_cpu_data.msa_id & MSA_IR_WRPF);
1100 		break;
1101 	default:
1102 		r = kvm_mips_callbacks->check_extension(kvm, ext);
1103 		break;
1104 	}
1105 	return r;
1106 }
1107 
1108 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1109 {
1110 	return kvm_mips_pending_timer(vcpu) ||
1111 		kvm_read_c0_guest_cause(vcpu->arch.cop0) & C_TI;
1112 }
1113 
1114 int kvm_arch_vcpu_dump_regs(struct kvm_vcpu *vcpu)
1115 {
1116 	int i;
1117 	struct mips_coproc *cop0;
1118 
1119 	if (!vcpu)
1120 		return -1;
1121 
1122 	kvm_debug("VCPU Register Dump:\n");
1123 	kvm_debug("\tpc = 0x%08lx\n", vcpu->arch.pc);
1124 	kvm_debug("\texceptions: %08lx\n", vcpu->arch.pending_exceptions);
1125 
1126 	for (i = 0; i < 32; i += 4) {
1127 		kvm_debug("\tgpr%02d: %08lx %08lx %08lx %08lx\n", i,
1128 		       vcpu->arch.gprs[i],
1129 		       vcpu->arch.gprs[i + 1],
1130 		       vcpu->arch.gprs[i + 2], vcpu->arch.gprs[i + 3]);
1131 	}
1132 	kvm_debug("\thi: 0x%08lx\n", vcpu->arch.hi);
1133 	kvm_debug("\tlo: 0x%08lx\n", vcpu->arch.lo);
1134 
1135 	cop0 = vcpu->arch.cop0;
1136 	kvm_debug("\tStatus: 0x%08x, Cause: 0x%08x\n",
1137 		  kvm_read_c0_guest_status(cop0),
1138 		  kvm_read_c0_guest_cause(cop0));
1139 
1140 	kvm_debug("\tEPC: 0x%08lx\n", kvm_read_c0_guest_epc(cop0));
1141 
1142 	return 0;
1143 }
1144 
1145 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1146 {
1147 	int i;
1148 
1149 	vcpu_load(vcpu);
1150 
1151 	for (i = 1; i < ARRAY_SIZE(vcpu->arch.gprs); i++)
1152 		vcpu->arch.gprs[i] = regs->gpr[i];
1153 	vcpu->arch.gprs[0] = 0; /* zero is special, and cannot be set. */
1154 	vcpu->arch.hi = regs->hi;
1155 	vcpu->arch.lo = regs->lo;
1156 	vcpu->arch.pc = regs->pc;
1157 
1158 	vcpu_put(vcpu);
1159 	return 0;
1160 }
1161 
1162 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1163 {
1164 	int i;
1165 
1166 	vcpu_load(vcpu);
1167 
1168 	for (i = 0; i < ARRAY_SIZE(vcpu->arch.gprs); i++)
1169 		regs->gpr[i] = vcpu->arch.gprs[i];
1170 
1171 	regs->hi = vcpu->arch.hi;
1172 	regs->lo = vcpu->arch.lo;
1173 	regs->pc = vcpu->arch.pc;
1174 
1175 	vcpu_put(vcpu);
1176 	return 0;
1177 }
1178 
1179 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1180 				  struct kvm_translation *tr)
1181 {
1182 	return 0;
1183 }
1184 
1185 static void kvm_mips_set_c0_status(void)
1186 {
1187 	u32 status = read_c0_status();
1188 
1189 	if (cpu_has_dsp)
1190 		status |= (ST0_MX);
1191 
1192 	write_c0_status(status);
1193 	ehb();
1194 }
1195 
1196 /*
1197  * Return value is in the form (errcode<<2 | RESUME_FLAG_HOST | RESUME_FLAG_NV)
1198  */
1199 int kvm_mips_handle_exit(struct kvm_vcpu *vcpu)
1200 {
1201 	struct kvm_run *run = vcpu->run;
1202 	u32 cause = vcpu->arch.host_cp0_cause;
1203 	u32 exccode = (cause >> CAUSEB_EXCCODE) & 0x1f;
1204 	u32 __user *opc = (u32 __user *) vcpu->arch.pc;
1205 	unsigned long badvaddr = vcpu->arch.host_cp0_badvaddr;
1206 	enum emulation_result er = EMULATE_DONE;
1207 	u32 inst;
1208 	int ret = RESUME_GUEST;
1209 
1210 	vcpu->mode = OUTSIDE_GUEST_MODE;
1211 
1212 	/* Set a default exit reason */
1213 	run->exit_reason = KVM_EXIT_UNKNOWN;
1214 	run->ready_for_interrupt_injection = 1;
1215 
1216 	/*
1217 	 * Set the appropriate status bits based on host CPU features,
1218 	 * before we hit the scheduler
1219 	 */
1220 	kvm_mips_set_c0_status();
1221 
1222 	local_irq_enable();
1223 
1224 	kvm_debug("kvm_mips_handle_exit: cause: %#x, PC: %p, kvm_run: %p, kvm_vcpu: %p\n",
1225 			cause, opc, run, vcpu);
1226 	trace_kvm_exit(vcpu, exccode);
1227 
1228 	switch (exccode) {
1229 	case EXCCODE_INT:
1230 		kvm_debug("[%d]EXCCODE_INT @ %p\n", vcpu->vcpu_id, opc);
1231 
1232 		++vcpu->stat.int_exits;
1233 
1234 		if (need_resched())
1235 			cond_resched();
1236 
1237 		ret = RESUME_GUEST;
1238 		break;
1239 
1240 	case EXCCODE_CPU:
1241 		kvm_debug("EXCCODE_CPU: @ PC: %p\n", opc);
1242 
1243 		++vcpu->stat.cop_unusable_exits;
1244 		ret = kvm_mips_callbacks->handle_cop_unusable(vcpu);
1245 		/* XXXKYMA: Might need to return to user space */
1246 		if (run->exit_reason == KVM_EXIT_IRQ_WINDOW_OPEN)
1247 			ret = RESUME_HOST;
1248 		break;
1249 
1250 	case EXCCODE_MOD:
1251 		++vcpu->stat.tlbmod_exits;
1252 		ret = kvm_mips_callbacks->handle_tlb_mod(vcpu);
1253 		break;
1254 
1255 	case EXCCODE_TLBS:
1256 		kvm_debug("TLB ST fault:  cause %#x, status %#x, PC: %p, BadVaddr: %#lx\n",
1257 			  cause, kvm_read_c0_guest_status(vcpu->arch.cop0), opc,
1258 			  badvaddr);
1259 
1260 		++vcpu->stat.tlbmiss_st_exits;
1261 		ret = kvm_mips_callbacks->handle_tlb_st_miss(vcpu);
1262 		break;
1263 
1264 	case EXCCODE_TLBL:
1265 		kvm_debug("TLB LD fault: cause %#x, PC: %p, BadVaddr: %#lx\n",
1266 			  cause, opc, badvaddr);
1267 
1268 		++vcpu->stat.tlbmiss_ld_exits;
1269 		ret = kvm_mips_callbacks->handle_tlb_ld_miss(vcpu);
1270 		break;
1271 
1272 	case EXCCODE_ADES:
1273 		++vcpu->stat.addrerr_st_exits;
1274 		ret = kvm_mips_callbacks->handle_addr_err_st(vcpu);
1275 		break;
1276 
1277 	case EXCCODE_ADEL:
1278 		++vcpu->stat.addrerr_ld_exits;
1279 		ret = kvm_mips_callbacks->handle_addr_err_ld(vcpu);
1280 		break;
1281 
1282 	case EXCCODE_SYS:
1283 		++vcpu->stat.syscall_exits;
1284 		ret = kvm_mips_callbacks->handle_syscall(vcpu);
1285 		break;
1286 
1287 	case EXCCODE_RI:
1288 		++vcpu->stat.resvd_inst_exits;
1289 		ret = kvm_mips_callbacks->handle_res_inst(vcpu);
1290 		break;
1291 
1292 	case EXCCODE_BP:
1293 		++vcpu->stat.break_inst_exits;
1294 		ret = kvm_mips_callbacks->handle_break(vcpu);
1295 		break;
1296 
1297 	case EXCCODE_TR:
1298 		++vcpu->stat.trap_inst_exits;
1299 		ret = kvm_mips_callbacks->handle_trap(vcpu);
1300 		break;
1301 
1302 	case EXCCODE_MSAFPE:
1303 		++vcpu->stat.msa_fpe_exits;
1304 		ret = kvm_mips_callbacks->handle_msa_fpe(vcpu);
1305 		break;
1306 
1307 	case EXCCODE_FPE:
1308 		++vcpu->stat.fpe_exits;
1309 		ret = kvm_mips_callbacks->handle_fpe(vcpu);
1310 		break;
1311 
1312 	case EXCCODE_MSADIS:
1313 		++vcpu->stat.msa_disabled_exits;
1314 		ret = kvm_mips_callbacks->handle_msa_disabled(vcpu);
1315 		break;
1316 
1317 	case EXCCODE_GE:
1318 		/* defer exit accounting to handler */
1319 		ret = kvm_mips_callbacks->handle_guest_exit(vcpu);
1320 		break;
1321 
1322 	default:
1323 		if (cause & CAUSEF_BD)
1324 			opc += 1;
1325 		inst = 0;
1326 		kvm_get_badinstr(opc, vcpu, &inst);
1327 		kvm_err("Exception Code: %d, not yet handled, @ PC: %p, inst: 0x%08x  BadVaddr: %#lx Status: %#x\n",
1328 			exccode, opc, inst, badvaddr,
1329 			kvm_read_c0_guest_status(vcpu->arch.cop0));
1330 		kvm_arch_vcpu_dump_regs(vcpu);
1331 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1332 		ret = RESUME_HOST;
1333 		break;
1334 
1335 	}
1336 
1337 	local_irq_disable();
1338 
1339 	if (ret == RESUME_GUEST)
1340 		kvm_vz_acquire_htimer(vcpu);
1341 
1342 	if (er == EMULATE_DONE && !(ret & RESUME_HOST))
1343 		kvm_mips_deliver_interrupts(vcpu, cause);
1344 
1345 	if (!(ret & RESUME_HOST)) {
1346 		/* Only check for signals if not already exiting to userspace */
1347 		if (signal_pending(current)) {
1348 			run->exit_reason = KVM_EXIT_INTR;
1349 			ret = (-EINTR << 2) | RESUME_HOST;
1350 			++vcpu->stat.signal_exits;
1351 			trace_kvm_exit(vcpu, KVM_TRACE_EXIT_SIGNAL);
1352 		}
1353 	}
1354 
1355 	if (ret == RESUME_GUEST) {
1356 		trace_kvm_reenter(vcpu);
1357 
1358 		/*
1359 		 * Make sure the read of VCPU requests in vcpu_reenter()
1360 		 * callback is not reordered ahead of the write to vcpu->mode,
1361 		 * or we could miss a TLB flush request while the requester sees
1362 		 * the VCPU as outside of guest mode and not needing an IPI.
1363 		 */
1364 		smp_store_mb(vcpu->mode, IN_GUEST_MODE);
1365 
1366 		kvm_mips_callbacks->vcpu_reenter(vcpu);
1367 
1368 		/*
1369 		 * If FPU / MSA are enabled (i.e. the guest's FPU / MSA context
1370 		 * is live), restore FCR31 / MSACSR.
1371 		 *
1372 		 * This should be before returning to the guest exception
1373 		 * vector, as it may well cause an [MSA] FP exception if there
1374 		 * are pending exception bits unmasked. (see
1375 		 * kvm_mips_csr_die_notifier() for how that is handled).
1376 		 */
1377 		if (kvm_mips_guest_has_fpu(&vcpu->arch) &&
1378 		    read_c0_status() & ST0_CU1)
1379 			__kvm_restore_fcsr(&vcpu->arch);
1380 
1381 		if (kvm_mips_guest_has_msa(&vcpu->arch) &&
1382 		    read_c0_config5() & MIPS_CONF5_MSAEN)
1383 			__kvm_restore_msacsr(&vcpu->arch);
1384 	}
1385 	return ret;
1386 }
1387 
1388 /* Enable FPU for guest and restore context */
1389 void kvm_own_fpu(struct kvm_vcpu *vcpu)
1390 {
1391 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1392 	unsigned int sr, cfg5;
1393 
1394 	preempt_disable();
1395 
1396 	sr = kvm_read_c0_guest_status(cop0);
1397 
1398 	/*
1399 	 * If MSA state is already live, it is undefined how it interacts with
1400 	 * FR=0 FPU state, and we don't want to hit reserved instruction
1401 	 * exceptions trying to save the MSA state later when CU=1 && FR=1, so
1402 	 * play it safe and save it first.
1403 	 */
1404 	if (cpu_has_msa && sr & ST0_CU1 && !(sr & ST0_FR) &&
1405 	    vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA)
1406 		kvm_lose_fpu(vcpu);
1407 
1408 	/*
1409 	 * Enable FPU for guest
1410 	 * We set FR and FRE according to guest context
1411 	 */
1412 	change_c0_status(ST0_CU1 | ST0_FR, sr);
1413 	if (cpu_has_fre) {
1414 		cfg5 = kvm_read_c0_guest_config5(cop0);
1415 		change_c0_config5(MIPS_CONF5_FRE, cfg5);
1416 	}
1417 	enable_fpu_hazard();
1418 
1419 	/* If guest FPU state not active, restore it now */
1420 	if (!(vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU)) {
1421 		__kvm_restore_fpu(&vcpu->arch);
1422 		vcpu->arch.aux_inuse |= KVM_MIPS_AUX_FPU;
1423 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE, KVM_TRACE_AUX_FPU);
1424 	} else {
1425 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_ENABLE, KVM_TRACE_AUX_FPU);
1426 	}
1427 
1428 	preempt_enable();
1429 }
1430 
1431 #ifdef CONFIG_CPU_HAS_MSA
1432 /* Enable MSA for guest and restore context */
1433 void kvm_own_msa(struct kvm_vcpu *vcpu)
1434 {
1435 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1436 	unsigned int sr, cfg5;
1437 
1438 	preempt_disable();
1439 
1440 	/*
1441 	 * Enable FPU if enabled in guest, since we're restoring FPU context
1442 	 * anyway. We set FR and FRE according to guest context.
1443 	 */
1444 	if (kvm_mips_guest_has_fpu(&vcpu->arch)) {
1445 		sr = kvm_read_c0_guest_status(cop0);
1446 
1447 		/*
1448 		 * If FR=0 FPU state is already live, it is undefined how it
1449 		 * interacts with MSA state, so play it safe and save it first.
1450 		 */
1451 		if (!(sr & ST0_FR) &&
1452 		    (vcpu->arch.aux_inuse & (KVM_MIPS_AUX_FPU |
1453 				KVM_MIPS_AUX_MSA)) == KVM_MIPS_AUX_FPU)
1454 			kvm_lose_fpu(vcpu);
1455 
1456 		change_c0_status(ST0_CU1 | ST0_FR, sr);
1457 		if (sr & ST0_CU1 && cpu_has_fre) {
1458 			cfg5 = kvm_read_c0_guest_config5(cop0);
1459 			change_c0_config5(MIPS_CONF5_FRE, cfg5);
1460 		}
1461 	}
1462 
1463 	/* Enable MSA for guest */
1464 	set_c0_config5(MIPS_CONF5_MSAEN);
1465 	enable_fpu_hazard();
1466 
1467 	switch (vcpu->arch.aux_inuse & (KVM_MIPS_AUX_FPU | KVM_MIPS_AUX_MSA)) {
1468 	case KVM_MIPS_AUX_FPU:
1469 		/*
1470 		 * Guest FPU state already loaded, only restore upper MSA state
1471 		 */
1472 		__kvm_restore_msa_upper(&vcpu->arch);
1473 		vcpu->arch.aux_inuse |= KVM_MIPS_AUX_MSA;
1474 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE, KVM_TRACE_AUX_MSA);
1475 		break;
1476 	case 0:
1477 		/* Neither FPU or MSA already active, restore full MSA state */
1478 		__kvm_restore_msa(&vcpu->arch);
1479 		vcpu->arch.aux_inuse |= KVM_MIPS_AUX_MSA;
1480 		if (kvm_mips_guest_has_fpu(&vcpu->arch))
1481 			vcpu->arch.aux_inuse |= KVM_MIPS_AUX_FPU;
1482 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE,
1483 			      KVM_TRACE_AUX_FPU_MSA);
1484 		break;
1485 	default:
1486 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_ENABLE, KVM_TRACE_AUX_MSA);
1487 		break;
1488 	}
1489 
1490 	preempt_enable();
1491 }
1492 #endif
1493 
1494 /* Drop FPU & MSA without saving it */
1495 void kvm_drop_fpu(struct kvm_vcpu *vcpu)
1496 {
1497 	preempt_disable();
1498 	if (cpu_has_msa && vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA) {
1499 		disable_msa();
1500 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_DISCARD, KVM_TRACE_AUX_MSA);
1501 		vcpu->arch.aux_inuse &= ~KVM_MIPS_AUX_MSA;
1502 	}
1503 	if (vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU) {
1504 		clear_c0_status(ST0_CU1 | ST0_FR);
1505 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_DISCARD, KVM_TRACE_AUX_FPU);
1506 		vcpu->arch.aux_inuse &= ~KVM_MIPS_AUX_FPU;
1507 	}
1508 	preempt_enable();
1509 }
1510 
1511 /* Save and disable FPU & MSA */
1512 void kvm_lose_fpu(struct kvm_vcpu *vcpu)
1513 {
1514 	/*
1515 	 * With T&E, FPU & MSA get disabled in root context (hardware) when it
1516 	 * is disabled in guest context (software), but the register state in
1517 	 * the hardware may still be in use.
1518 	 * This is why we explicitly re-enable the hardware before saving.
1519 	 */
1520 
1521 	preempt_disable();
1522 	if (cpu_has_msa && vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA) {
1523 		__kvm_save_msa(&vcpu->arch);
1524 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_SAVE, KVM_TRACE_AUX_FPU_MSA);
1525 
1526 		/* Disable MSA & FPU */
1527 		disable_msa();
1528 		if (vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU) {
1529 			clear_c0_status(ST0_CU1 | ST0_FR);
1530 			disable_fpu_hazard();
1531 		}
1532 		vcpu->arch.aux_inuse &= ~(KVM_MIPS_AUX_FPU | KVM_MIPS_AUX_MSA);
1533 	} else if (vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU) {
1534 		__kvm_save_fpu(&vcpu->arch);
1535 		vcpu->arch.aux_inuse &= ~KVM_MIPS_AUX_FPU;
1536 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_SAVE, KVM_TRACE_AUX_FPU);
1537 
1538 		/* Disable FPU */
1539 		clear_c0_status(ST0_CU1 | ST0_FR);
1540 		disable_fpu_hazard();
1541 	}
1542 	preempt_enable();
1543 }
1544 
1545 /*
1546  * Step over a specific ctc1 to FCSR and a specific ctcmsa to MSACSR which are
1547  * used to restore guest FCSR/MSACSR state and may trigger a "harmless" FP/MSAFP
1548  * exception if cause bits are set in the value being written.
1549  */
1550 static int kvm_mips_csr_die_notify(struct notifier_block *self,
1551 				   unsigned long cmd, void *ptr)
1552 {
1553 	struct die_args *args = (struct die_args *)ptr;
1554 	struct pt_regs *regs = args->regs;
1555 	unsigned long pc;
1556 
1557 	/* Only interested in FPE and MSAFPE */
1558 	if (cmd != DIE_FP && cmd != DIE_MSAFP)
1559 		return NOTIFY_DONE;
1560 
1561 	/* Return immediately if guest context isn't active */
1562 	if (!(current->flags & PF_VCPU))
1563 		return NOTIFY_DONE;
1564 
1565 	/* Should never get here from user mode */
1566 	BUG_ON(user_mode(regs));
1567 
1568 	pc = instruction_pointer(regs);
1569 	switch (cmd) {
1570 	case DIE_FP:
1571 		/* match 2nd instruction in __kvm_restore_fcsr */
1572 		if (pc != (unsigned long)&__kvm_restore_fcsr + 4)
1573 			return NOTIFY_DONE;
1574 		break;
1575 	case DIE_MSAFP:
1576 		/* match 2nd/3rd instruction in __kvm_restore_msacsr */
1577 		if (!cpu_has_msa ||
1578 		    pc < (unsigned long)&__kvm_restore_msacsr + 4 ||
1579 		    pc > (unsigned long)&__kvm_restore_msacsr + 8)
1580 			return NOTIFY_DONE;
1581 		break;
1582 	}
1583 
1584 	/* Move PC forward a little and continue executing */
1585 	instruction_pointer(regs) += 4;
1586 
1587 	return NOTIFY_STOP;
1588 }
1589 
1590 static struct notifier_block kvm_mips_csr_die_notifier = {
1591 	.notifier_call = kvm_mips_csr_die_notify,
1592 };
1593 
1594 static u32 kvm_default_priority_to_irq[MIPS_EXC_MAX] = {
1595 	[MIPS_EXC_INT_TIMER] = C_IRQ5,
1596 	[MIPS_EXC_INT_IO_1]  = C_IRQ0,
1597 	[MIPS_EXC_INT_IPI_1] = C_IRQ1,
1598 	[MIPS_EXC_INT_IPI_2] = C_IRQ2,
1599 };
1600 
1601 static u32 kvm_loongson3_priority_to_irq[MIPS_EXC_MAX] = {
1602 	[MIPS_EXC_INT_TIMER] = C_IRQ5,
1603 	[MIPS_EXC_INT_IO_1]  = C_IRQ0,
1604 	[MIPS_EXC_INT_IO_2]  = C_IRQ1,
1605 	[MIPS_EXC_INT_IPI_1] = C_IRQ4,
1606 };
1607 
1608 u32 *kvm_priority_to_irq = kvm_default_priority_to_irq;
1609 
1610 u32 kvm_irq_to_priority(u32 irq)
1611 {
1612 	int i;
1613 
1614 	for (i = MIPS_EXC_INT_TIMER; i < MIPS_EXC_MAX; i++) {
1615 		if (kvm_priority_to_irq[i] == (1 << (irq + 8)))
1616 			return i;
1617 	}
1618 
1619 	return MIPS_EXC_MAX;
1620 }
1621 
1622 static int __init kvm_mips_init(void)
1623 {
1624 	int ret;
1625 
1626 	if (cpu_has_mmid) {
1627 		pr_warn("KVM does not yet support MMIDs. KVM Disabled\n");
1628 		return -EOPNOTSUPP;
1629 	}
1630 
1631 	ret = kvm_mips_entry_setup();
1632 	if (ret)
1633 		return ret;
1634 
1635 	ret = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1636 
1637 	if (ret)
1638 		return ret;
1639 
1640 	if (boot_cpu_type() == CPU_LOONGSON64)
1641 		kvm_priority_to_irq = kvm_loongson3_priority_to_irq;
1642 
1643 	register_die_notifier(&kvm_mips_csr_die_notifier);
1644 
1645 	return 0;
1646 }
1647 
1648 static void __exit kvm_mips_exit(void)
1649 {
1650 	kvm_exit();
1651 
1652 	unregister_die_notifier(&kvm_mips_csr_die_notifier);
1653 }
1654 
1655 module_init(kvm_mips_init);
1656 module_exit(kvm_mips_exit);
1657 
1658 EXPORT_TRACEPOINT_SYMBOL(kvm_exit);
1659