xref: /openbmc/linux/arch/mips/kvm/mips.c (revision 6db6b729)
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 extern void kvm_init_loongson_ipi(struct kvm *kvm);
139 
140 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
141 {
142 	switch (type) {
143 	case KVM_VM_MIPS_AUTO:
144 		break;
145 	case KVM_VM_MIPS_VZ:
146 		break;
147 	default:
148 		/* Unsupported KVM type */
149 		return -EINVAL;
150 	}
151 
152 	/* Allocate page table to map GPA -> RPA */
153 	kvm->arch.gpa_mm.pgd = kvm_pgd_alloc();
154 	if (!kvm->arch.gpa_mm.pgd)
155 		return -ENOMEM;
156 
157 #ifdef CONFIG_CPU_LOONGSON64
158 	kvm_init_loongson_ipi(kvm);
159 #endif
160 
161 	return 0;
162 }
163 
164 static void kvm_mips_free_gpa_pt(struct kvm *kvm)
165 {
166 	/* It should always be safe to remove after flushing the whole range */
167 	WARN_ON(!kvm_mips_flush_gpa_pt(kvm, 0, ~0));
168 	pgd_free(NULL, kvm->arch.gpa_mm.pgd);
169 }
170 
171 void kvm_arch_destroy_vm(struct kvm *kvm)
172 {
173 	kvm_destroy_vcpus(kvm);
174 	kvm_mips_free_gpa_pt(kvm);
175 }
176 
177 long kvm_arch_dev_ioctl(struct file *filp, unsigned int ioctl,
178 			unsigned long arg)
179 {
180 	return -ENOIOCTLCMD;
181 }
182 
183 void kvm_arch_flush_shadow_all(struct kvm *kvm)
184 {
185 	/* Flush whole GPA */
186 	kvm_mips_flush_gpa_pt(kvm, 0, ~0);
187 	kvm_flush_remote_tlbs(kvm);
188 }
189 
190 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
191 				   struct kvm_memory_slot *slot)
192 {
193 	/*
194 	 * The slot has been made invalid (ready for moving or deletion), so we
195 	 * need to ensure that it can no longer be accessed by any guest VCPUs.
196 	 */
197 
198 	spin_lock(&kvm->mmu_lock);
199 	/* Flush slot from GPA */
200 	kvm_mips_flush_gpa_pt(kvm, slot->base_gfn,
201 			      slot->base_gfn + slot->npages - 1);
202 	kvm_flush_remote_tlbs_memslot(kvm, slot);
203 	spin_unlock(&kvm->mmu_lock);
204 }
205 
206 int kvm_arch_prepare_memory_region(struct kvm *kvm,
207 				   const struct kvm_memory_slot *old,
208 				   struct kvm_memory_slot *new,
209 				   enum kvm_mr_change change)
210 {
211 	return 0;
212 }
213 
214 void kvm_arch_commit_memory_region(struct kvm *kvm,
215 				   struct kvm_memory_slot *old,
216 				   const struct kvm_memory_slot *new,
217 				   enum kvm_mr_change change)
218 {
219 	int needs_flush;
220 
221 	/*
222 	 * If dirty page logging is enabled, write protect all pages in the slot
223 	 * ready for dirty logging.
224 	 *
225 	 * There is no need to do this in any of the following cases:
226 	 * CREATE:	No dirty mappings will already exist.
227 	 * MOVE/DELETE:	The old mappings will already have been cleaned up by
228 	 *		kvm_arch_flush_shadow_memslot()
229 	 */
230 	if (change == KVM_MR_FLAGS_ONLY &&
231 	    (!(old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
232 	     new->flags & KVM_MEM_LOG_DIRTY_PAGES)) {
233 		spin_lock(&kvm->mmu_lock);
234 		/* Write protect GPA page table entries */
235 		needs_flush = kvm_mips_mkclean_gpa_pt(kvm, new->base_gfn,
236 					new->base_gfn + new->npages - 1);
237 		if (needs_flush)
238 			kvm_flush_remote_tlbs_memslot(kvm, new);
239 		spin_unlock(&kvm->mmu_lock);
240 	}
241 }
242 
243 static inline void dump_handler(const char *symbol, void *start, void *end)
244 {
245 	u32 *p;
246 
247 	pr_debug("LEAF(%s)\n", symbol);
248 
249 	pr_debug("\t.set push\n");
250 	pr_debug("\t.set noreorder\n");
251 
252 	for (p = start; p < (u32 *)end; ++p)
253 		pr_debug("\t.word\t0x%08x\t\t# %p\n", *p, p);
254 
255 	pr_debug("\t.set\tpop\n");
256 
257 	pr_debug("\tEND(%s)\n", symbol);
258 }
259 
260 /* low level hrtimer wake routine */
261 static enum hrtimer_restart kvm_mips_comparecount_wakeup(struct hrtimer *timer)
262 {
263 	struct kvm_vcpu *vcpu;
264 
265 	vcpu = container_of(timer, struct kvm_vcpu, arch.comparecount_timer);
266 
267 	kvm_mips_callbacks->queue_timer_int(vcpu);
268 
269 	vcpu->arch.wait = 0;
270 	rcuwait_wake_up(&vcpu->wait);
271 
272 	return kvm_mips_count_timeout(vcpu);
273 }
274 
275 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
276 {
277 	return 0;
278 }
279 
280 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
281 {
282 	int err, size;
283 	void *gebase, *p, *handler, *refill_start, *refill_end;
284 	int i;
285 
286 	kvm_debug("kvm @ %p: create cpu %d at %p\n",
287 		  vcpu->kvm, vcpu->vcpu_id, vcpu);
288 
289 	err = kvm_mips_callbacks->vcpu_init(vcpu);
290 	if (err)
291 		return err;
292 
293 	hrtimer_init(&vcpu->arch.comparecount_timer, CLOCK_MONOTONIC,
294 		     HRTIMER_MODE_REL);
295 	vcpu->arch.comparecount_timer.function = kvm_mips_comparecount_wakeup;
296 
297 	/*
298 	 * Allocate space for host mode exception handlers that handle
299 	 * guest mode exits
300 	 */
301 	if (cpu_has_veic || cpu_has_vint)
302 		size = 0x200 + VECTORSPACING * 64;
303 	else
304 		size = 0x4000;
305 
306 	gebase = kzalloc(ALIGN(size, PAGE_SIZE), GFP_KERNEL);
307 
308 	if (!gebase) {
309 		err = -ENOMEM;
310 		goto out_uninit_vcpu;
311 	}
312 	kvm_debug("Allocated %d bytes for KVM Exception Handlers @ %p\n",
313 		  ALIGN(size, PAGE_SIZE), gebase);
314 
315 	/*
316 	 * Check new ebase actually fits in CP0_EBase. The lack of a write gate
317 	 * limits us to the low 512MB of physical address space. If the memory
318 	 * we allocate is out of range, just give up now.
319 	 */
320 	if (!cpu_has_ebase_wg && virt_to_phys(gebase) >= 0x20000000) {
321 		kvm_err("CP0_EBase.WG required for guest exception base %pK\n",
322 			gebase);
323 		err = -ENOMEM;
324 		goto out_free_gebase;
325 	}
326 
327 	/* Save new ebase */
328 	vcpu->arch.guest_ebase = gebase;
329 
330 	/* Build guest exception vectors dynamically in unmapped memory */
331 	handler = gebase + 0x2000;
332 
333 	/* TLB refill (or XTLB refill on 64-bit VZ where KX=1) */
334 	refill_start = gebase;
335 	if (IS_ENABLED(CONFIG_64BIT))
336 		refill_start += 0x080;
337 	refill_end = kvm_mips_build_tlb_refill_exception(refill_start, handler);
338 
339 	/* General Exception Entry point */
340 	kvm_mips_build_exception(gebase + 0x180, handler);
341 
342 	/* For vectored interrupts poke the exception code @ all offsets 0-7 */
343 	for (i = 0; i < 8; i++) {
344 		kvm_debug("L1 Vectored handler @ %p\n",
345 			  gebase + 0x200 + (i * VECTORSPACING));
346 		kvm_mips_build_exception(gebase + 0x200 + i * VECTORSPACING,
347 					 handler);
348 	}
349 
350 	/* General exit handler */
351 	p = handler;
352 	p = kvm_mips_build_exit(p);
353 
354 	/* Guest entry routine */
355 	vcpu->arch.vcpu_run = p;
356 	p = kvm_mips_build_vcpu_run(p);
357 
358 	/* Dump the generated code */
359 	pr_debug("#include <asm/asm.h>\n");
360 	pr_debug("#include <asm/regdef.h>\n");
361 	pr_debug("\n");
362 	dump_handler("kvm_vcpu_run", vcpu->arch.vcpu_run, p);
363 	dump_handler("kvm_tlb_refill", refill_start, refill_end);
364 	dump_handler("kvm_gen_exc", gebase + 0x180, gebase + 0x200);
365 	dump_handler("kvm_exit", gebase + 0x2000, vcpu->arch.vcpu_run);
366 
367 	/* Invalidate the icache for these ranges */
368 	flush_icache_range((unsigned long)gebase,
369 			   (unsigned long)gebase + ALIGN(size, PAGE_SIZE));
370 
371 	/* Init */
372 	vcpu->arch.last_sched_cpu = -1;
373 	vcpu->arch.last_exec_cpu = -1;
374 
375 	/* Initial guest state */
376 	err = kvm_mips_callbacks->vcpu_setup(vcpu);
377 	if (err)
378 		goto out_free_gebase;
379 
380 	return 0;
381 
382 out_free_gebase:
383 	kfree(gebase);
384 out_uninit_vcpu:
385 	kvm_mips_callbacks->vcpu_uninit(vcpu);
386 	return err;
387 }
388 
389 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
390 {
391 	hrtimer_cancel(&vcpu->arch.comparecount_timer);
392 
393 	kvm_mips_dump_stats(vcpu);
394 
395 	kvm_mmu_free_memory_caches(vcpu);
396 	kfree(vcpu->arch.guest_ebase);
397 
398 	kvm_mips_callbacks->vcpu_uninit(vcpu);
399 }
400 
401 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
402 					struct kvm_guest_debug *dbg)
403 {
404 	return -ENOIOCTLCMD;
405 }
406 
407 /*
408  * Actually run the vCPU, entering an RCU extended quiescent state (EQS) while
409  * the vCPU is running.
410  *
411  * This must be noinstr as instrumentation may make use of RCU, and this is not
412  * safe during the EQS.
413  */
414 static int noinstr kvm_mips_vcpu_enter_exit(struct kvm_vcpu *vcpu)
415 {
416 	int ret;
417 
418 	guest_state_enter_irqoff();
419 	ret = kvm_mips_callbacks->vcpu_run(vcpu);
420 	guest_state_exit_irqoff();
421 
422 	return ret;
423 }
424 
425 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
426 {
427 	int r = -EINTR;
428 
429 	vcpu_load(vcpu);
430 
431 	kvm_sigset_activate(vcpu);
432 
433 	if (vcpu->mmio_needed) {
434 		if (!vcpu->mmio_is_write)
435 			kvm_mips_complete_mmio_load(vcpu);
436 		vcpu->mmio_needed = 0;
437 	}
438 
439 	if (vcpu->run->immediate_exit)
440 		goto out;
441 
442 	lose_fpu(1);
443 
444 	local_irq_disable();
445 	guest_timing_enter_irqoff();
446 	trace_kvm_enter(vcpu);
447 
448 	/*
449 	 * Make sure the read of VCPU requests in vcpu_run() callback is not
450 	 * reordered ahead of the write to vcpu->mode, or we could miss a TLB
451 	 * flush request while the requester sees the VCPU as outside of guest
452 	 * mode and not needing an IPI.
453 	 */
454 	smp_store_mb(vcpu->mode, IN_GUEST_MODE);
455 
456 	r = kvm_mips_vcpu_enter_exit(vcpu);
457 
458 	/*
459 	 * We must ensure that any pending interrupts are taken before
460 	 * we exit guest timing so that timer ticks are accounted as
461 	 * guest time. Transiently unmask interrupts so that any
462 	 * pending interrupts are taken.
463 	 *
464 	 * TODO: is there a barrier which ensures that pending interrupts are
465 	 * recognised? Currently this just hopes that the CPU takes any pending
466 	 * interrupts between the enable and disable.
467 	 */
468 	local_irq_enable();
469 	local_irq_disable();
470 
471 	trace_kvm_out(vcpu);
472 	guest_timing_exit_irqoff();
473 	local_irq_enable();
474 
475 out:
476 	kvm_sigset_deactivate(vcpu);
477 
478 	vcpu_put(vcpu);
479 	return r;
480 }
481 
482 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
483 			     struct kvm_mips_interrupt *irq)
484 {
485 	int intr = (int)irq->irq;
486 	struct kvm_vcpu *dvcpu = NULL;
487 
488 	if (intr == kvm_priority_to_irq[MIPS_EXC_INT_IPI_1] ||
489 	    intr == kvm_priority_to_irq[MIPS_EXC_INT_IPI_2] ||
490 	    intr == (-kvm_priority_to_irq[MIPS_EXC_INT_IPI_1]) ||
491 	    intr == (-kvm_priority_to_irq[MIPS_EXC_INT_IPI_2]))
492 		kvm_debug("%s: CPU: %d, INTR: %d\n", __func__, irq->cpu,
493 			  (int)intr);
494 
495 	if (irq->cpu == -1)
496 		dvcpu = vcpu;
497 	else
498 		dvcpu = kvm_get_vcpu(vcpu->kvm, irq->cpu);
499 
500 	if (intr == 2 || intr == 3 || intr == 4 || intr == 6) {
501 		kvm_mips_callbacks->queue_io_int(dvcpu, irq);
502 
503 	} else if (intr == -2 || intr == -3 || intr == -4 || intr == -6) {
504 		kvm_mips_callbacks->dequeue_io_int(dvcpu, irq);
505 	} else {
506 		kvm_err("%s: invalid interrupt ioctl (%d:%d)\n", __func__,
507 			irq->cpu, irq->irq);
508 		return -EINVAL;
509 	}
510 
511 	dvcpu->arch.wait = 0;
512 
513 	rcuwait_wake_up(&dvcpu->wait);
514 
515 	return 0;
516 }
517 
518 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
519 				    struct kvm_mp_state *mp_state)
520 {
521 	return -ENOIOCTLCMD;
522 }
523 
524 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
525 				    struct kvm_mp_state *mp_state)
526 {
527 	return -ENOIOCTLCMD;
528 }
529 
530 static u64 kvm_mips_get_one_regs[] = {
531 	KVM_REG_MIPS_R0,
532 	KVM_REG_MIPS_R1,
533 	KVM_REG_MIPS_R2,
534 	KVM_REG_MIPS_R3,
535 	KVM_REG_MIPS_R4,
536 	KVM_REG_MIPS_R5,
537 	KVM_REG_MIPS_R6,
538 	KVM_REG_MIPS_R7,
539 	KVM_REG_MIPS_R8,
540 	KVM_REG_MIPS_R9,
541 	KVM_REG_MIPS_R10,
542 	KVM_REG_MIPS_R11,
543 	KVM_REG_MIPS_R12,
544 	KVM_REG_MIPS_R13,
545 	KVM_REG_MIPS_R14,
546 	KVM_REG_MIPS_R15,
547 	KVM_REG_MIPS_R16,
548 	KVM_REG_MIPS_R17,
549 	KVM_REG_MIPS_R18,
550 	KVM_REG_MIPS_R19,
551 	KVM_REG_MIPS_R20,
552 	KVM_REG_MIPS_R21,
553 	KVM_REG_MIPS_R22,
554 	KVM_REG_MIPS_R23,
555 	KVM_REG_MIPS_R24,
556 	KVM_REG_MIPS_R25,
557 	KVM_REG_MIPS_R26,
558 	KVM_REG_MIPS_R27,
559 	KVM_REG_MIPS_R28,
560 	KVM_REG_MIPS_R29,
561 	KVM_REG_MIPS_R30,
562 	KVM_REG_MIPS_R31,
563 
564 #ifndef CONFIG_CPU_MIPSR6
565 	KVM_REG_MIPS_HI,
566 	KVM_REG_MIPS_LO,
567 #endif
568 	KVM_REG_MIPS_PC,
569 };
570 
571 static u64 kvm_mips_get_one_regs_fpu[] = {
572 	KVM_REG_MIPS_FCR_IR,
573 	KVM_REG_MIPS_FCR_CSR,
574 };
575 
576 static u64 kvm_mips_get_one_regs_msa[] = {
577 	KVM_REG_MIPS_MSA_IR,
578 	KVM_REG_MIPS_MSA_CSR,
579 };
580 
581 static unsigned long kvm_mips_num_regs(struct kvm_vcpu *vcpu)
582 {
583 	unsigned long ret;
584 
585 	ret = ARRAY_SIZE(kvm_mips_get_one_regs);
586 	if (kvm_mips_guest_can_have_fpu(&vcpu->arch)) {
587 		ret += ARRAY_SIZE(kvm_mips_get_one_regs_fpu) + 48;
588 		/* odd doubles */
589 		if (boot_cpu_data.fpu_id & MIPS_FPIR_F64)
590 			ret += 16;
591 	}
592 	if (kvm_mips_guest_can_have_msa(&vcpu->arch))
593 		ret += ARRAY_SIZE(kvm_mips_get_one_regs_msa) + 32;
594 	ret += kvm_mips_callbacks->num_regs(vcpu);
595 
596 	return ret;
597 }
598 
599 static int kvm_mips_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *indices)
600 {
601 	u64 index;
602 	unsigned int i;
603 
604 	if (copy_to_user(indices, kvm_mips_get_one_regs,
605 			 sizeof(kvm_mips_get_one_regs)))
606 		return -EFAULT;
607 	indices += ARRAY_SIZE(kvm_mips_get_one_regs);
608 
609 	if (kvm_mips_guest_can_have_fpu(&vcpu->arch)) {
610 		if (copy_to_user(indices, kvm_mips_get_one_regs_fpu,
611 				 sizeof(kvm_mips_get_one_regs_fpu)))
612 			return -EFAULT;
613 		indices += ARRAY_SIZE(kvm_mips_get_one_regs_fpu);
614 
615 		for (i = 0; i < 32; ++i) {
616 			index = KVM_REG_MIPS_FPR_32(i);
617 			if (copy_to_user(indices, &index, sizeof(index)))
618 				return -EFAULT;
619 			++indices;
620 
621 			/* skip odd doubles if no F64 */
622 			if (i & 1 && !(boot_cpu_data.fpu_id & MIPS_FPIR_F64))
623 				continue;
624 
625 			index = KVM_REG_MIPS_FPR_64(i);
626 			if (copy_to_user(indices, &index, sizeof(index)))
627 				return -EFAULT;
628 			++indices;
629 		}
630 	}
631 
632 	if (kvm_mips_guest_can_have_msa(&vcpu->arch)) {
633 		if (copy_to_user(indices, kvm_mips_get_one_regs_msa,
634 				 sizeof(kvm_mips_get_one_regs_msa)))
635 			return -EFAULT;
636 		indices += ARRAY_SIZE(kvm_mips_get_one_regs_msa);
637 
638 		for (i = 0; i < 32; ++i) {
639 			index = KVM_REG_MIPS_VEC_128(i);
640 			if (copy_to_user(indices, &index, sizeof(index)))
641 				return -EFAULT;
642 			++indices;
643 		}
644 	}
645 
646 	return kvm_mips_callbacks->copy_reg_indices(vcpu, indices);
647 }
648 
649 static int kvm_mips_get_reg(struct kvm_vcpu *vcpu,
650 			    const struct kvm_one_reg *reg)
651 {
652 	struct mips_coproc *cop0 = &vcpu->arch.cop0;
653 	struct mips_fpu_struct *fpu = &vcpu->arch.fpu;
654 	int ret;
655 	s64 v;
656 	s64 vs[2];
657 	unsigned int idx;
658 
659 	switch (reg->id) {
660 	/* General purpose registers */
661 	case KVM_REG_MIPS_R0 ... KVM_REG_MIPS_R31:
662 		v = (long)vcpu->arch.gprs[reg->id - KVM_REG_MIPS_R0];
663 		break;
664 #ifndef CONFIG_CPU_MIPSR6
665 	case KVM_REG_MIPS_HI:
666 		v = (long)vcpu->arch.hi;
667 		break;
668 	case KVM_REG_MIPS_LO:
669 		v = (long)vcpu->arch.lo;
670 		break;
671 #endif
672 	case KVM_REG_MIPS_PC:
673 		v = (long)vcpu->arch.pc;
674 		break;
675 
676 	/* Floating point registers */
677 	case KVM_REG_MIPS_FPR_32(0) ... KVM_REG_MIPS_FPR_32(31):
678 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
679 			return -EINVAL;
680 		idx = reg->id - KVM_REG_MIPS_FPR_32(0);
681 		/* Odd singles in top of even double when FR=0 */
682 		if (kvm_read_c0_guest_status(cop0) & ST0_FR)
683 			v = get_fpr32(&fpu->fpr[idx], 0);
684 		else
685 			v = get_fpr32(&fpu->fpr[idx & ~1], idx & 1);
686 		break;
687 	case KVM_REG_MIPS_FPR_64(0) ... KVM_REG_MIPS_FPR_64(31):
688 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
689 			return -EINVAL;
690 		idx = reg->id - KVM_REG_MIPS_FPR_64(0);
691 		/* Can't access odd doubles in FR=0 mode */
692 		if (idx & 1 && !(kvm_read_c0_guest_status(cop0) & ST0_FR))
693 			return -EINVAL;
694 		v = get_fpr64(&fpu->fpr[idx], 0);
695 		break;
696 	case KVM_REG_MIPS_FCR_IR:
697 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
698 			return -EINVAL;
699 		v = boot_cpu_data.fpu_id;
700 		break;
701 	case KVM_REG_MIPS_FCR_CSR:
702 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
703 			return -EINVAL;
704 		v = fpu->fcr31;
705 		break;
706 
707 	/* MIPS SIMD Architecture (MSA) registers */
708 	case KVM_REG_MIPS_VEC_128(0) ... KVM_REG_MIPS_VEC_128(31):
709 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
710 			return -EINVAL;
711 		/* Can't access MSA registers in FR=0 mode */
712 		if (!(kvm_read_c0_guest_status(cop0) & ST0_FR))
713 			return -EINVAL;
714 		idx = reg->id - KVM_REG_MIPS_VEC_128(0);
715 #ifdef CONFIG_CPU_LITTLE_ENDIAN
716 		/* least significant byte first */
717 		vs[0] = get_fpr64(&fpu->fpr[idx], 0);
718 		vs[1] = get_fpr64(&fpu->fpr[idx], 1);
719 #else
720 		/* most significant byte first */
721 		vs[0] = get_fpr64(&fpu->fpr[idx], 1);
722 		vs[1] = get_fpr64(&fpu->fpr[idx], 0);
723 #endif
724 		break;
725 	case KVM_REG_MIPS_MSA_IR:
726 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
727 			return -EINVAL;
728 		v = boot_cpu_data.msa_id;
729 		break;
730 	case KVM_REG_MIPS_MSA_CSR:
731 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
732 			return -EINVAL;
733 		v = fpu->msacsr;
734 		break;
735 
736 	/* registers to be handled specially */
737 	default:
738 		ret = kvm_mips_callbacks->get_one_reg(vcpu, reg, &v);
739 		if (ret)
740 			return ret;
741 		break;
742 	}
743 	if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U64) {
744 		u64 __user *uaddr64 = (u64 __user *)(long)reg->addr;
745 
746 		return put_user(v, uaddr64);
747 	} else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U32) {
748 		u32 __user *uaddr32 = (u32 __user *)(long)reg->addr;
749 		u32 v32 = (u32)v;
750 
751 		return put_user(v32, uaddr32);
752 	} else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U128) {
753 		void __user *uaddr = (void __user *)(long)reg->addr;
754 
755 		return copy_to_user(uaddr, vs, 16) ? -EFAULT : 0;
756 	} else {
757 		return -EINVAL;
758 	}
759 }
760 
761 static int kvm_mips_set_reg(struct kvm_vcpu *vcpu,
762 			    const struct kvm_one_reg *reg)
763 {
764 	struct mips_coproc *cop0 = &vcpu->arch.cop0;
765 	struct mips_fpu_struct *fpu = &vcpu->arch.fpu;
766 	s64 v;
767 	s64 vs[2];
768 	unsigned int idx;
769 
770 	if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U64) {
771 		u64 __user *uaddr64 = (u64 __user *)(long)reg->addr;
772 
773 		if (get_user(v, uaddr64) != 0)
774 			return -EFAULT;
775 	} else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U32) {
776 		u32 __user *uaddr32 = (u32 __user *)(long)reg->addr;
777 		s32 v32;
778 
779 		if (get_user(v32, uaddr32) != 0)
780 			return -EFAULT;
781 		v = (s64)v32;
782 	} else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U128) {
783 		void __user *uaddr = (void __user *)(long)reg->addr;
784 
785 		return copy_from_user(vs, uaddr, 16) ? -EFAULT : 0;
786 	} else {
787 		return -EINVAL;
788 	}
789 
790 	switch (reg->id) {
791 	/* General purpose registers */
792 	case KVM_REG_MIPS_R0:
793 		/* Silently ignore requests to set $0 */
794 		break;
795 	case KVM_REG_MIPS_R1 ... KVM_REG_MIPS_R31:
796 		vcpu->arch.gprs[reg->id - KVM_REG_MIPS_R0] = v;
797 		break;
798 #ifndef CONFIG_CPU_MIPSR6
799 	case KVM_REG_MIPS_HI:
800 		vcpu->arch.hi = v;
801 		break;
802 	case KVM_REG_MIPS_LO:
803 		vcpu->arch.lo = v;
804 		break;
805 #endif
806 	case KVM_REG_MIPS_PC:
807 		vcpu->arch.pc = v;
808 		break;
809 
810 	/* Floating point registers */
811 	case KVM_REG_MIPS_FPR_32(0) ... KVM_REG_MIPS_FPR_32(31):
812 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
813 			return -EINVAL;
814 		idx = reg->id - KVM_REG_MIPS_FPR_32(0);
815 		/* Odd singles in top of even double when FR=0 */
816 		if (kvm_read_c0_guest_status(cop0) & ST0_FR)
817 			set_fpr32(&fpu->fpr[idx], 0, v);
818 		else
819 			set_fpr32(&fpu->fpr[idx & ~1], idx & 1, v);
820 		break;
821 	case KVM_REG_MIPS_FPR_64(0) ... KVM_REG_MIPS_FPR_64(31):
822 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
823 			return -EINVAL;
824 		idx = reg->id - KVM_REG_MIPS_FPR_64(0);
825 		/* Can't access odd doubles in FR=0 mode */
826 		if (idx & 1 && !(kvm_read_c0_guest_status(cop0) & ST0_FR))
827 			return -EINVAL;
828 		set_fpr64(&fpu->fpr[idx], 0, v);
829 		break;
830 	case KVM_REG_MIPS_FCR_IR:
831 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
832 			return -EINVAL;
833 		/* Read-only */
834 		break;
835 	case KVM_REG_MIPS_FCR_CSR:
836 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
837 			return -EINVAL;
838 		fpu->fcr31 = v;
839 		break;
840 
841 	/* MIPS SIMD Architecture (MSA) registers */
842 	case KVM_REG_MIPS_VEC_128(0) ... KVM_REG_MIPS_VEC_128(31):
843 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
844 			return -EINVAL;
845 		idx = reg->id - KVM_REG_MIPS_VEC_128(0);
846 #ifdef CONFIG_CPU_LITTLE_ENDIAN
847 		/* least significant byte first */
848 		set_fpr64(&fpu->fpr[idx], 0, vs[0]);
849 		set_fpr64(&fpu->fpr[idx], 1, vs[1]);
850 #else
851 		/* most significant byte first */
852 		set_fpr64(&fpu->fpr[idx], 1, vs[0]);
853 		set_fpr64(&fpu->fpr[idx], 0, vs[1]);
854 #endif
855 		break;
856 	case KVM_REG_MIPS_MSA_IR:
857 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
858 			return -EINVAL;
859 		/* Read-only */
860 		break;
861 	case KVM_REG_MIPS_MSA_CSR:
862 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
863 			return -EINVAL;
864 		fpu->msacsr = v;
865 		break;
866 
867 	/* registers to be handled specially */
868 	default:
869 		return kvm_mips_callbacks->set_one_reg(vcpu, reg, v);
870 	}
871 	return 0;
872 }
873 
874 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
875 				     struct kvm_enable_cap *cap)
876 {
877 	int r = 0;
878 
879 	if (!kvm_vm_ioctl_check_extension(vcpu->kvm, cap->cap))
880 		return -EINVAL;
881 	if (cap->flags)
882 		return -EINVAL;
883 	if (cap->args[0])
884 		return -EINVAL;
885 
886 	switch (cap->cap) {
887 	case KVM_CAP_MIPS_FPU:
888 		vcpu->arch.fpu_enabled = true;
889 		break;
890 	case KVM_CAP_MIPS_MSA:
891 		vcpu->arch.msa_enabled = true;
892 		break;
893 	default:
894 		r = -EINVAL;
895 		break;
896 	}
897 
898 	return r;
899 }
900 
901 long kvm_arch_vcpu_async_ioctl(struct file *filp, unsigned int ioctl,
902 			       unsigned long arg)
903 {
904 	struct kvm_vcpu *vcpu = filp->private_data;
905 	void __user *argp = (void __user *)arg;
906 
907 	if (ioctl == KVM_INTERRUPT) {
908 		struct kvm_mips_interrupt irq;
909 
910 		if (copy_from_user(&irq, argp, sizeof(irq)))
911 			return -EFAULT;
912 		kvm_debug("[%d] %s: irq: %d\n", vcpu->vcpu_id, __func__,
913 			  irq.irq);
914 
915 		return kvm_vcpu_ioctl_interrupt(vcpu, &irq);
916 	}
917 
918 	return -ENOIOCTLCMD;
919 }
920 
921 long kvm_arch_vcpu_ioctl(struct file *filp, unsigned int ioctl,
922 			 unsigned long arg)
923 {
924 	struct kvm_vcpu *vcpu = filp->private_data;
925 	void __user *argp = (void __user *)arg;
926 	long r;
927 
928 	vcpu_load(vcpu);
929 
930 	switch (ioctl) {
931 	case KVM_SET_ONE_REG:
932 	case KVM_GET_ONE_REG: {
933 		struct kvm_one_reg reg;
934 
935 		r = -EFAULT;
936 		if (copy_from_user(&reg, argp, sizeof(reg)))
937 			break;
938 		if (ioctl == KVM_SET_ONE_REG)
939 			r = kvm_mips_set_reg(vcpu, &reg);
940 		else
941 			r = kvm_mips_get_reg(vcpu, &reg);
942 		break;
943 	}
944 	case KVM_GET_REG_LIST: {
945 		struct kvm_reg_list __user *user_list = argp;
946 		struct kvm_reg_list reg_list;
947 		unsigned n;
948 
949 		r = -EFAULT;
950 		if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
951 			break;
952 		n = reg_list.n;
953 		reg_list.n = kvm_mips_num_regs(vcpu);
954 		if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
955 			break;
956 		r = -E2BIG;
957 		if (n < reg_list.n)
958 			break;
959 		r = kvm_mips_copy_reg_indices(vcpu, user_list->reg);
960 		break;
961 	}
962 	case KVM_ENABLE_CAP: {
963 		struct kvm_enable_cap cap;
964 
965 		r = -EFAULT;
966 		if (copy_from_user(&cap, argp, sizeof(cap)))
967 			break;
968 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
969 		break;
970 	}
971 	default:
972 		r = -ENOIOCTLCMD;
973 	}
974 
975 	vcpu_put(vcpu);
976 	return r;
977 }
978 
979 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
980 {
981 
982 }
983 
984 int kvm_arch_flush_remote_tlbs(struct kvm *kvm)
985 {
986 	kvm_mips_callbacks->prepare_flush_shadow(kvm);
987 	return 1;
988 }
989 
990 int kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
991 {
992 	int r;
993 
994 	switch (ioctl) {
995 	default:
996 		r = -ENOIOCTLCMD;
997 	}
998 
999 	return r;
1000 }
1001 
1002 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1003 				  struct kvm_sregs *sregs)
1004 {
1005 	return -ENOIOCTLCMD;
1006 }
1007 
1008 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1009 				  struct kvm_sregs *sregs)
1010 {
1011 	return -ENOIOCTLCMD;
1012 }
1013 
1014 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
1015 {
1016 }
1017 
1018 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1019 {
1020 	return -ENOIOCTLCMD;
1021 }
1022 
1023 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1024 {
1025 	return -ENOIOCTLCMD;
1026 }
1027 
1028 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
1029 {
1030 	return VM_FAULT_SIGBUS;
1031 }
1032 
1033 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
1034 {
1035 	int r;
1036 
1037 	switch (ext) {
1038 	case KVM_CAP_ONE_REG:
1039 	case KVM_CAP_ENABLE_CAP:
1040 	case KVM_CAP_READONLY_MEM:
1041 	case KVM_CAP_SYNC_MMU:
1042 	case KVM_CAP_IMMEDIATE_EXIT:
1043 		r = 1;
1044 		break;
1045 	case KVM_CAP_NR_VCPUS:
1046 		r = min_t(unsigned int, num_online_cpus(), KVM_MAX_VCPUS);
1047 		break;
1048 	case KVM_CAP_MAX_VCPUS:
1049 		r = KVM_MAX_VCPUS;
1050 		break;
1051 	case KVM_CAP_MAX_VCPU_ID:
1052 		r = KVM_MAX_VCPU_IDS;
1053 		break;
1054 	case KVM_CAP_MIPS_FPU:
1055 		/* We don't handle systems with inconsistent cpu_has_fpu */
1056 		r = !!raw_cpu_has_fpu;
1057 		break;
1058 	case KVM_CAP_MIPS_MSA:
1059 		/*
1060 		 * We don't support MSA vector partitioning yet:
1061 		 * 1) It would require explicit support which can't be tested
1062 		 *    yet due to lack of support in current hardware.
1063 		 * 2) It extends the state that would need to be saved/restored
1064 		 *    by e.g. QEMU for migration.
1065 		 *
1066 		 * When vector partitioning hardware becomes available, support
1067 		 * could be added by requiring a flag when enabling
1068 		 * KVM_CAP_MIPS_MSA capability to indicate that userland knows
1069 		 * to save/restore the appropriate extra state.
1070 		 */
1071 		r = cpu_has_msa && !(boot_cpu_data.msa_id & MSA_IR_WRPF);
1072 		break;
1073 	default:
1074 		r = kvm_mips_callbacks->check_extension(kvm, ext);
1075 		break;
1076 	}
1077 	return r;
1078 }
1079 
1080 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1081 {
1082 	return kvm_mips_pending_timer(vcpu) ||
1083 		kvm_read_c0_guest_cause(&vcpu->arch.cop0) & C_TI;
1084 }
1085 
1086 int kvm_arch_vcpu_dump_regs(struct kvm_vcpu *vcpu)
1087 {
1088 	int i;
1089 	struct mips_coproc *cop0;
1090 
1091 	if (!vcpu)
1092 		return -1;
1093 
1094 	kvm_debug("VCPU Register Dump:\n");
1095 	kvm_debug("\tpc = 0x%08lx\n", vcpu->arch.pc);
1096 	kvm_debug("\texceptions: %08lx\n", vcpu->arch.pending_exceptions);
1097 
1098 	for (i = 0; i < 32; i += 4) {
1099 		kvm_debug("\tgpr%02d: %08lx %08lx %08lx %08lx\n", i,
1100 		       vcpu->arch.gprs[i],
1101 		       vcpu->arch.gprs[i + 1],
1102 		       vcpu->arch.gprs[i + 2], vcpu->arch.gprs[i + 3]);
1103 	}
1104 	kvm_debug("\thi: 0x%08lx\n", vcpu->arch.hi);
1105 	kvm_debug("\tlo: 0x%08lx\n", vcpu->arch.lo);
1106 
1107 	cop0 = &vcpu->arch.cop0;
1108 	kvm_debug("\tStatus: 0x%08x, Cause: 0x%08x\n",
1109 		  kvm_read_c0_guest_status(cop0),
1110 		  kvm_read_c0_guest_cause(cop0));
1111 
1112 	kvm_debug("\tEPC: 0x%08lx\n", kvm_read_c0_guest_epc(cop0));
1113 
1114 	return 0;
1115 }
1116 
1117 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1118 {
1119 	int i;
1120 
1121 	vcpu_load(vcpu);
1122 
1123 	for (i = 1; i < ARRAY_SIZE(vcpu->arch.gprs); i++)
1124 		vcpu->arch.gprs[i] = regs->gpr[i];
1125 	vcpu->arch.gprs[0] = 0; /* zero is special, and cannot be set. */
1126 	vcpu->arch.hi = regs->hi;
1127 	vcpu->arch.lo = regs->lo;
1128 	vcpu->arch.pc = regs->pc;
1129 
1130 	vcpu_put(vcpu);
1131 	return 0;
1132 }
1133 
1134 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1135 {
1136 	int i;
1137 
1138 	vcpu_load(vcpu);
1139 
1140 	for (i = 0; i < ARRAY_SIZE(vcpu->arch.gprs); i++)
1141 		regs->gpr[i] = vcpu->arch.gprs[i];
1142 
1143 	regs->hi = vcpu->arch.hi;
1144 	regs->lo = vcpu->arch.lo;
1145 	regs->pc = vcpu->arch.pc;
1146 
1147 	vcpu_put(vcpu);
1148 	return 0;
1149 }
1150 
1151 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1152 				  struct kvm_translation *tr)
1153 {
1154 	return 0;
1155 }
1156 
1157 static void kvm_mips_set_c0_status(void)
1158 {
1159 	u32 status = read_c0_status();
1160 
1161 	if (cpu_has_dsp)
1162 		status |= (ST0_MX);
1163 
1164 	write_c0_status(status);
1165 	ehb();
1166 }
1167 
1168 /*
1169  * Return value is in the form (errcode<<2 | RESUME_FLAG_HOST | RESUME_FLAG_NV)
1170  */
1171 static int __kvm_mips_handle_exit(struct kvm_vcpu *vcpu)
1172 {
1173 	struct kvm_run *run = vcpu->run;
1174 	u32 cause = vcpu->arch.host_cp0_cause;
1175 	u32 exccode = (cause >> CAUSEB_EXCCODE) & 0x1f;
1176 	u32 __user *opc = (u32 __user *) vcpu->arch.pc;
1177 	unsigned long badvaddr = vcpu->arch.host_cp0_badvaddr;
1178 	enum emulation_result er = EMULATE_DONE;
1179 	u32 inst;
1180 	int ret = RESUME_GUEST;
1181 
1182 	vcpu->mode = OUTSIDE_GUEST_MODE;
1183 
1184 	/* Set a default exit reason */
1185 	run->exit_reason = KVM_EXIT_UNKNOWN;
1186 	run->ready_for_interrupt_injection = 1;
1187 
1188 	/*
1189 	 * Set the appropriate status bits based on host CPU features,
1190 	 * before we hit the scheduler
1191 	 */
1192 	kvm_mips_set_c0_status();
1193 
1194 	local_irq_enable();
1195 
1196 	kvm_debug("kvm_mips_handle_exit: cause: %#x, PC: %p, kvm_run: %p, kvm_vcpu: %p\n",
1197 			cause, opc, run, vcpu);
1198 	trace_kvm_exit(vcpu, exccode);
1199 
1200 	switch (exccode) {
1201 	case EXCCODE_INT:
1202 		kvm_debug("[%d]EXCCODE_INT @ %p\n", vcpu->vcpu_id, opc);
1203 
1204 		++vcpu->stat.int_exits;
1205 
1206 		if (need_resched())
1207 			cond_resched();
1208 
1209 		ret = RESUME_GUEST;
1210 		break;
1211 
1212 	case EXCCODE_CPU:
1213 		kvm_debug("EXCCODE_CPU: @ PC: %p\n", opc);
1214 
1215 		++vcpu->stat.cop_unusable_exits;
1216 		ret = kvm_mips_callbacks->handle_cop_unusable(vcpu);
1217 		/* XXXKYMA: Might need to return to user space */
1218 		if (run->exit_reason == KVM_EXIT_IRQ_WINDOW_OPEN)
1219 			ret = RESUME_HOST;
1220 		break;
1221 
1222 	case EXCCODE_MOD:
1223 		++vcpu->stat.tlbmod_exits;
1224 		ret = kvm_mips_callbacks->handle_tlb_mod(vcpu);
1225 		break;
1226 
1227 	case EXCCODE_TLBS:
1228 		kvm_debug("TLB ST fault:  cause %#x, status %#x, PC: %p, BadVaddr: %#lx\n",
1229 			  cause, kvm_read_c0_guest_status(&vcpu->arch.cop0), opc,
1230 			  badvaddr);
1231 
1232 		++vcpu->stat.tlbmiss_st_exits;
1233 		ret = kvm_mips_callbacks->handle_tlb_st_miss(vcpu);
1234 		break;
1235 
1236 	case EXCCODE_TLBL:
1237 		kvm_debug("TLB LD fault: cause %#x, PC: %p, BadVaddr: %#lx\n",
1238 			  cause, opc, badvaddr);
1239 
1240 		++vcpu->stat.tlbmiss_ld_exits;
1241 		ret = kvm_mips_callbacks->handle_tlb_ld_miss(vcpu);
1242 		break;
1243 
1244 	case EXCCODE_ADES:
1245 		++vcpu->stat.addrerr_st_exits;
1246 		ret = kvm_mips_callbacks->handle_addr_err_st(vcpu);
1247 		break;
1248 
1249 	case EXCCODE_ADEL:
1250 		++vcpu->stat.addrerr_ld_exits;
1251 		ret = kvm_mips_callbacks->handle_addr_err_ld(vcpu);
1252 		break;
1253 
1254 	case EXCCODE_SYS:
1255 		++vcpu->stat.syscall_exits;
1256 		ret = kvm_mips_callbacks->handle_syscall(vcpu);
1257 		break;
1258 
1259 	case EXCCODE_RI:
1260 		++vcpu->stat.resvd_inst_exits;
1261 		ret = kvm_mips_callbacks->handle_res_inst(vcpu);
1262 		break;
1263 
1264 	case EXCCODE_BP:
1265 		++vcpu->stat.break_inst_exits;
1266 		ret = kvm_mips_callbacks->handle_break(vcpu);
1267 		break;
1268 
1269 	case EXCCODE_TR:
1270 		++vcpu->stat.trap_inst_exits;
1271 		ret = kvm_mips_callbacks->handle_trap(vcpu);
1272 		break;
1273 
1274 	case EXCCODE_MSAFPE:
1275 		++vcpu->stat.msa_fpe_exits;
1276 		ret = kvm_mips_callbacks->handle_msa_fpe(vcpu);
1277 		break;
1278 
1279 	case EXCCODE_FPE:
1280 		++vcpu->stat.fpe_exits;
1281 		ret = kvm_mips_callbacks->handle_fpe(vcpu);
1282 		break;
1283 
1284 	case EXCCODE_MSADIS:
1285 		++vcpu->stat.msa_disabled_exits;
1286 		ret = kvm_mips_callbacks->handle_msa_disabled(vcpu);
1287 		break;
1288 
1289 	case EXCCODE_GE:
1290 		/* defer exit accounting to handler */
1291 		ret = kvm_mips_callbacks->handle_guest_exit(vcpu);
1292 		break;
1293 
1294 	default:
1295 		if (cause & CAUSEF_BD)
1296 			opc += 1;
1297 		inst = 0;
1298 		kvm_get_badinstr(opc, vcpu, &inst);
1299 		kvm_err("Exception Code: %d, not yet handled, @ PC: %p, inst: 0x%08x  BadVaddr: %#lx Status: %#x\n",
1300 			exccode, opc, inst, badvaddr,
1301 			kvm_read_c0_guest_status(&vcpu->arch.cop0));
1302 		kvm_arch_vcpu_dump_regs(vcpu);
1303 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1304 		ret = RESUME_HOST;
1305 		break;
1306 
1307 	}
1308 
1309 	local_irq_disable();
1310 
1311 	if (ret == RESUME_GUEST)
1312 		kvm_vz_acquire_htimer(vcpu);
1313 
1314 	if (er == EMULATE_DONE && !(ret & RESUME_HOST))
1315 		kvm_mips_deliver_interrupts(vcpu, cause);
1316 
1317 	if (!(ret & RESUME_HOST)) {
1318 		/* Only check for signals if not already exiting to userspace */
1319 		if (signal_pending(current)) {
1320 			run->exit_reason = KVM_EXIT_INTR;
1321 			ret = (-EINTR << 2) | RESUME_HOST;
1322 			++vcpu->stat.signal_exits;
1323 			trace_kvm_exit(vcpu, KVM_TRACE_EXIT_SIGNAL);
1324 		}
1325 	}
1326 
1327 	if (ret == RESUME_GUEST) {
1328 		trace_kvm_reenter(vcpu);
1329 
1330 		/*
1331 		 * Make sure the read of VCPU requests in vcpu_reenter()
1332 		 * callback is not reordered ahead of the write to vcpu->mode,
1333 		 * or we could miss a TLB flush request while the requester sees
1334 		 * the VCPU as outside of guest mode and not needing an IPI.
1335 		 */
1336 		smp_store_mb(vcpu->mode, IN_GUEST_MODE);
1337 
1338 		kvm_mips_callbacks->vcpu_reenter(vcpu);
1339 
1340 		/*
1341 		 * If FPU / MSA are enabled (i.e. the guest's FPU / MSA context
1342 		 * is live), restore FCR31 / MSACSR.
1343 		 *
1344 		 * This should be before returning to the guest exception
1345 		 * vector, as it may well cause an [MSA] FP exception if there
1346 		 * are pending exception bits unmasked. (see
1347 		 * kvm_mips_csr_die_notifier() for how that is handled).
1348 		 */
1349 		if (kvm_mips_guest_has_fpu(&vcpu->arch) &&
1350 		    read_c0_status() & ST0_CU1)
1351 			__kvm_restore_fcsr(&vcpu->arch);
1352 
1353 		if (kvm_mips_guest_has_msa(&vcpu->arch) &&
1354 		    read_c0_config5() & MIPS_CONF5_MSAEN)
1355 			__kvm_restore_msacsr(&vcpu->arch);
1356 	}
1357 	return ret;
1358 }
1359 
1360 int noinstr kvm_mips_handle_exit(struct kvm_vcpu *vcpu)
1361 {
1362 	int ret;
1363 
1364 	guest_state_exit_irqoff();
1365 	ret = __kvm_mips_handle_exit(vcpu);
1366 	guest_state_enter_irqoff();
1367 
1368 	return ret;
1369 }
1370 
1371 /* Enable FPU for guest and restore context */
1372 void kvm_own_fpu(struct kvm_vcpu *vcpu)
1373 {
1374 	struct mips_coproc *cop0 = &vcpu->arch.cop0;
1375 	unsigned int sr, cfg5;
1376 
1377 	preempt_disable();
1378 
1379 	sr = kvm_read_c0_guest_status(cop0);
1380 
1381 	/*
1382 	 * If MSA state is already live, it is undefined how it interacts with
1383 	 * FR=0 FPU state, and we don't want to hit reserved instruction
1384 	 * exceptions trying to save the MSA state later when CU=1 && FR=1, so
1385 	 * play it safe and save it first.
1386 	 */
1387 	if (cpu_has_msa && sr & ST0_CU1 && !(sr & ST0_FR) &&
1388 	    vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA)
1389 		kvm_lose_fpu(vcpu);
1390 
1391 	/*
1392 	 * Enable FPU for guest
1393 	 * We set FR and FRE according to guest context
1394 	 */
1395 	change_c0_status(ST0_CU1 | ST0_FR, sr);
1396 	if (cpu_has_fre) {
1397 		cfg5 = kvm_read_c0_guest_config5(cop0);
1398 		change_c0_config5(MIPS_CONF5_FRE, cfg5);
1399 	}
1400 	enable_fpu_hazard();
1401 
1402 	/* If guest FPU state not active, restore it now */
1403 	if (!(vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU)) {
1404 		__kvm_restore_fpu(&vcpu->arch);
1405 		vcpu->arch.aux_inuse |= KVM_MIPS_AUX_FPU;
1406 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE, KVM_TRACE_AUX_FPU);
1407 	} else {
1408 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_ENABLE, KVM_TRACE_AUX_FPU);
1409 	}
1410 
1411 	preempt_enable();
1412 }
1413 
1414 #ifdef CONFIG_CPU_HAS_MSA
1415 /* Enable MSA for guest and restore context */
1416 void kvm_own_msa(struct kvm_vcpu *vcpu)
1417 {
1418 	struct mips_coproc *cop0 = &vcpu->arch.cop0;
1419 	unsigned int sr, cfg5;
1420 
1421 	preempt_disable();
1422 
1423 	/*
1424 	 * Enable FPU if enabled in guest, since we're restoring FPU context
1425 	 * anyway. We set FR and FRE according to guest context.
1426 	 */
1427 	if (kvm_mips_guest_has_fpu(&vcpu->arch)) {
1428 		sr = kvm_read_c0_guest_status(cop0);
1429 
1430 		/*
1431 		 * If FR=0 FPU state is already live, it is undefined how it
1432 		 * interacts with MSA state, so play it safe and save it first.
1433 		 */
1434 		if (!(sr & ST0_FR) &&
1435 		    (vcpu->arch.aux_inuse & (KVM_MIPS_AUX_FPU |
1436 				KVM_MIPS_AUX_MSA)) == KVM_MIPS_AUX_FPU)
1437 			kvm_lose_fpu(vcpu);
1438 
1439 		change_c0_status(ST0_CU1 | ST0_FR, sr);
1440 		if (sr & ST0_CU1 && cpu_has_fre) {
1441 			cfg5 = kvm_read_c0_guest_config5(cop0);
1442 			change_c0_config5(MIPS_CONF5_FRE, cfg5);
1443 		}
1444 	}
1445 
1446 	/* Enable MSA for guest */
1447 	set_c0_config5(MIPS_CONF5_MSAEN);
1448 	enable_fpu_hazard();
1449 
1450 	switch (vcpu->arch.aux_inuse & (KVM_MIPS_AUX_FPU | KVM_MIPS_AUX_MSA)) {
1451 	case KVM_MIPS_AUX_FPU:
1452 		/*
1453 		 * Guest FPU state already loaded, only restore upper MSA state
1454 		 */
1455 		__kvm_restore_msa_upper(&vcpu->arch);
1456 		vcpu->arch.aux_inuse |= KVM_MIPS_AUX_MSA;
1457 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE, KVM_TRACE_AUX_MSA);
1458 		break;
1459 	case 0:
1460 		/* Neither FPU or MSA already active, restore full MSA state */
1461 		__kvm_restore_msa(&vcpu->arch);
1462 		vcpu->arch.aux_inuse |= KVM_MIPS_AUX_MSA;
1463 		if (kvm_mips_guest_has_fpu(&vcpu->arch))
1464 			vcpu->arch.aux_inuse |= KVM_MIPS_AUX_FPU;
1465 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE,
1466 			      KVM_TRACE_AUX_FPU_MSA);
1467 		break;
1468 	default:
1469 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_ENABLE, KVM_TRACE_AUX_MSA);
1470 		break;
1471 	}
1472 
1473 	preempt_enable();
1474 }
1475 #endif
1476 
1477 /* Drop FPU & MSA without saving it */
1478 void kvm_drop_fpu(struct kvm_vcpu *vcpu)
1479 {
1480 	preempt_disable();
1481 	if (cpu_has_msa && vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA) {
1482 		disable_msa();
1483 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_DISCARD, KVM_TRACE_AUX_MSA);
1484 		vcpu->arch.aux_inuse &= ~KVM_MIPS_AUX_MSA;
1485 	}
1486 	if (vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU) {
1487 		clear_c0_status(ST0_CU1 | ST0_FR);
1488 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_DISCARD, KVM_TRACE_AUX_FPU);
1489 		vcpu->arch.aux_inuse &= ~KVM_MIPS_AUX_FPU;
1490 	}
1491 	preempt_enable();
1492 }
1493 
1494 /* Save and disable FPU & MSA */
1495 void kvm_lose_fpu(struct kvm_vcpu *vcpu)
1496 {
1497 	/*
1498 	 * With T&E, FPU & MSA get disabled in root context (hardware) when it
1499 	 * is disabled in guest context (software), but the register state in
1500 	 * the hardware may still be in use.
1501 	 * This is why we explicitly re-enable the hardware before saving.
1502 	 */
1503 
1504 	preempt_disable();
1505 	if (cpu_has_msa && vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA) {
1506 		__kvm_save_msa(&vcpu->arch);
1507 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_SAVE, KVM_TRACE_AUX_FPU_MSA);
1508 
1509 		/* Disable MSA & FPU */
1510 		disable_msa();
1511 		if (vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU) {
1512 			clear_c0_status(ST0_CU1 | ST0_FR);
1513 			disable_fpu_hazard();
1514 		}
1515 		vcpu->arch.aux_inuse &= ~(KVM_MIPS_AUX_FPU | KVM_MIPS_AUX_MSA);
1516 	} else if (vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU) {
1517 		__kvm_save_fpu(&vcpu->arch);
1518 		vcpu->arch.aux_inuse &= ~KVM_MIPS_AUX_FPU;
1519 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_SAVE, KVM_TRACE_AUX_FPU);
1520 
1521 		/* Disable FPU */
1522 		clear_c0_status(ST0_CU1 | ST0_FR);
1523 		disable_fpu_hazard();
1524 	}
1525 	preempt_enable();
1526 }
1527 
1528 /*
1529  * Step over a specific ctc1 to FCSR and a specific ctcmsa to MSACSR which are
1530  * used to restore guest FCSR/MSACSR state and may trigger a "harmless" FP/MSAFP
1531  * exception if cause bits are set in the value being written.
1532  */
1533 static int kvm_mips_csr_die_notify(struct notifier_block *self,
1534 				   unsigned long cmd, void *ptr)
1535 {
1536 	struct die_args *args = (struct die_args *)ptr;
1537 	struct pt_regs *regs = args->regs;
1538 	unsigned long pc;
1539 
1540 	/* Only interested in FPE and MSAFPE */
1541 	if (cmd != DIE_FP && cmd != DIE_MSAFP)
1542 		return NOTIFY_DONE;
1543 
1544 	/* Return immediately if guest context isn't active */
1545 	if (!(current->flags & PF_VCPU))
1546 		return NOTIFY_DONE;
1547 
1548 	/* Should never get here from user mode */
1549 	BUG_ON(user_mode(regs));
1550 
1551 	pc = instruction_pointer(regs);
1552 	switch (cmd) {
1553 	case DIE_FP:
1554 		/* match 2nd instruction in __kvm_restore_fcsr */
1555 		if (pc != (unsigned long)&__kvm_restore_fcsr + 4)
1556 			return NOTIFY_DONE;
1557 		break;
1558 	case DIE_MSAFP:
1559 		/* match 2nd/3rd instruction in __kvm_restore_msacsr */
1560 		if (!cpu_has_msa ||
1561 		    pc < (unsigned long)&__kvm_restore_msacsr + 4 ||
1562 		    pc > (unsigned long)&__kvm_restore_msacsr + 8)
1563 			return NOTIFY_DONE;
1564 		break;
1565 	}
1566 
1567 	/* Move PC forward a little and continue executing */
1568 	instruction_pointer(regs) += 4;
1569 
1570 	return NOTIFY_STOP;
1571 }
1572 
1573 static struct notifier_block kvm_mips_csr_die_notifier = {
1574 	.notifier_call = kvm_mips_csr_die_notify,
1575 };
1576 
1577 static u32 kvm_default_priority_to_irq[MIPS_EXC_MAX] = {
1578 	[MIPS_EXC_INT_TIMER] = C_IRQ5,
1579 	[MIPS_EXC_INT_IO_1]  = C_IRQ0,
1580 	[MIPS_EXC_INT_IPI_1] = C_IRQ1,
1581 	[MIPS_EXC_INT_IPI_2] = C_IRQ2,
1582 };
1583 
1584 static u32 kvm_loongson3_priority_to_irq[MIPS_EXC_MAX] = {
1585 	[MIPS_EXC_INT_TIMER] = C_IRQ5,
1586 	[MIPS_EXC_INT_IO_1]  = C_IRQ0,
1587 	[MIPS_EXC_INT_IO_2]  = C_IRQ1,
1588 	[MIPS_EXC_INT_IPI_1] = C_IRQ4,
1589 };
1590 
1591 u32 *kvm_priority_to_irq = kvm_default_priority_to_irq;
1592 
1593 u32 kvm_irq_to_priority(u32 irq)
1594 {
1595 	int i;
1596 
1597 	for (i = MIPS_EXC_INT_TIMER; i < MIPS_EXC_MAX; i++) {
1598 		if (kvm_priority_to_irq[i] == (1 << (irq + 8)))
1599 			return i;
1600 	}
1601 
1602 	return MIPS_EXC_MAX;
1603 }
1604 
1605 static int __init kvm_mips_init(void)
1606 {
1607 	int ret;
1608 
1609 	if (cpu_has_mmid) {
1610 		pr_warn("KVM does not yet support MMIDs. KVM Disabled\n");
1611 		return -EOPNOTSUPP;
1612 	}
1613 
1614 	ret = kvm_mips_entry_setup();
1615 	if (ret)
1616 		return ret;
1617 
1618 	ret = kvm_mips_emulation_init();
1619 	if (ret)
1620 		return ret;
1621 
1622 
1623 	if (boot_cpu_type() == CPU_LOONGSON64)
1624 		kvm_priority_to_irq = kvm_loongson3_priority_to_irq;
1625 
1626 	register_die_notifier(&kvm_mips_csr_die_notifier);
1627 
1628 	ret = kvm_init(sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1629 	if (ret) {
1630 		unregister_die_notifier(&kvm_mips_csr_die_notifier);
1631 		return ret;
1632 	}
1633 	return 0;
1634 }
1635 
1636 static void __exit kvm_mips_exit(void)
1637 {
1638 	kvm_exit();
1639 
1640 	unregister_die_notifier(&kvm_mips_csr_die_notifier);
1641 }
1642 
1643 module_init(kvm_mips_init);
1644 module_exit(kvm_mips_exit);
1645 
1646 EXPORT_TRACEPOINT_SYMBOL(kvm_exit);
1647