xref: /linux/arch/mips/kvm/mips.c (revision e5c86679d5e864947a52fb31e45a425dea3e7fa9)
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/bootmem.h>
22 
23 #include <asm/fpu.h>
24 #include <asm/page.h>
25 #include <asm/cacheflush.h>
26 #include <asm/mmu_context.h>
27 #include <asm/pgalloc.h>
28 #include <asm/pgtable.h>
29 
30 #include <linux/kvm_host.h>
31 
32 #include "interrupt.h"
33 #include "commpage.h"
34 
35 #define CREATE_TRACE_POINTS
36 #include "trace.h"
37 
38 #ifndef VECTORSPACING
39 #define VECTORSPACING 0x100	/* for EI/VI mode */
40 #endif
41 
42 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x)
43 struct kvm_stats_debugfs_item debugfs_entries[] = {
44 	{ "wait",	  VCPU_STAT(wait_exits),	 KVM_STAT_VCPU },
45 	{ "cache",	  VCPU_STAT(cache_exits),	 KVM_STAT_VCPU },
46 	{ "signal",	  VCPU_STAT(signal_exits),	 KVM_STAT_VCPU },
47 	{ "interrupt",	  VCPU_STAT(int_exits),		 KVM_STAT_VCPU },
48 	{ "cop_unsuable", VCPU_STAT(cop_unusable_exits), KVM_STAT_VCPU },
49 	{ "tlbmod",	  VCPU_STAT(tlbmod_exits),	 KVM_STAT_VCPU },
50 	{ "tlbmiss_ld",	  VCPU_STAT(tlbmiss_ld_exits),	 KVM_STAT_VCPU },
51 	{ "tlbmiss_st",	  VCPU_STAT(tlbmiss_st_exits),	 KVM_STAT_VCPU },
52 	{ "addrerr_st",	  VCPU_STAT(addrerr_st_exits),	 KVM_STAT_VCPU },
53 	{ "addrerr_ld",	  VCPU_STAT(addrerr_ld_exits),	 KVM_STAT_VCPU },
54 	{ "syscall",	  VCPU_STAT(syscall_exits),	 KVM_STAT_VCPU },
55 	{ "resvd_inst",	  VCPU_STAT(resvd_inst_exits),	 KVM_STAT_VCPU },
56 	{ "break_inst",	  VCPU_STAT(break_inst_exits),	 KVM_STAT_VCPU },
57 	{ "trap_inst",	  VCPU_STAT(trap_inst_exits),	 KVM_STAT_VCPU },
58 	{ "msa_fpe",	  VCPU_STAT(msa_fpe_exits),	 KVM_STAT_VCPU },
59 	{ "fpe",	  VCPU_STAT(fpe_exits),		 KVM_STAT_VCPU },
60 	{ "msa_disabled", VCPU_STAT(msa_disabled_exits), KVM_STAT_VCPU },
61 	{ "flush_dcache", VCPU_STAT(flush_dcache_exits), KVM_STAT_VCPU },
62 	{ "halt_successful_poll", VCPU_STAT(halt_successful_poll), KVM_STAT_VCPU },
63 	{ "halt_attempted_poll", VCPU_STAT(halt_attempted_poll), KVM_STAT_VCPU },
64 	{ "halt_poll_invalid", VCPU_STAT(halt_poll_invalid), KVM_STAT_VCPU },
65 	{ "halt_wakeup",  VCPU_STAT(halt_wakeup),	 KVM_STAT_VCPU },
66 	{NULL}
67 };
68 
69 /*
70  * XXXKYMA: We are simulatoring a processor that has the WII bit set in
71  * Config7, so we are "runnable" if interrupts are pending
72  */
73 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
74 {
75 	return !!(vcpu->arch.pending_exceptions);
76 }
77 
78 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
79 {
80 	return 1;
81 }
82 
83 int kvm_arch_hardware_enable(void)
84 {
85 	return 0;
86 }
87 
88 int kvm_arch_hardware_setup(void)
89 {
90 	return 0;
91 }
92 
93 void kvm_arch_check_processor_compat(void *rtn)
94 {
95 	*(int *)rtn = 0;
96 }
97 
98 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
99 {
100 	/* Allocate page table to map GPA -> RPA */
101 	kvm->arch.gpa_mm.pgd = kvm_pgd_alloc();
102 	if (!kvm->arch.gpa_mm.pgd)
103 		return -ENOMEM;
104 
105 	return 0;
106 }
107 
108 bool kvm_arch_has_vcpu_debugfs(void)
109 {
110 	return false;
111 }
112 
113 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
114 {
115 	return 0;
116 }
117 
118 void kvm_mips_free_vcpus(struct kvm *kvm)
119 {
120 	unsigned int i;
121 	struct kvm_vcpu *vcpu;
122 
123 	kvm_for_each_vcpu(i, vcpu, kvm) {
124 		kvm_arch_vcpu_free(vcpu);
125 	}
126 
127 	mutex_lock(&kvm->lock);
128 
129 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
130 		kvm->vcpus[i] = NULL;
131 
132 	atomic_set(&kvm->online_vcpus, 0);
133 
134 	mutex_unlock(&kvm->lock);
135 }
136 
137 static void kvm_mips_free_gpa_pt(struct kvm *kvm)
138 {
139 	/* It should always be safe to remove after flushing the whole range */
140 	WARN_ON(!kvm_mips_flush_gpa_pt(kvm, 0, ~0));
141 	pgd_free(NULL, kvm->arch.gpa_mm.pgd);
142 }
143 
144 void kvm_arch_destroy_vm(struct kvm *kvm)
145 {
146 	kvm_mips_free_vcpus(kvm);
147 	kvm_mips_free_gpa_pt(kvm);
148 }
149 
150 long kvm_arch_dev_ioctl(struct file *filp, unsigned int ioctl,
151 			unsigned long arg)
152 {
153 	return -ENOIOCTLCMD;
154 }
155 
156 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
157 			    unsigned long npages)
158 {
159 	return 0;
160 }
161 
162 void kvm_arch_flush_shadow_all(struct kvm *kvm)
163 {
164 	/* Flush whole GPA */
165 	kvm_mips_flush_gpa_pt(kvm, 0, ~0);
166 
167 	/* Let implementation do the rest */
168 	kvm_mips_callbacks->flush_shadow_all(kvm);
169 }
170 
171 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
172 				   struct kvm_memory_slot *slot)
173 {
174 	/*
175 	 * The slot has been made invalid (ready for moving or deletion), so we
176 	 * need to ensure that it can no longer be accessed by any guest VCPUs.
177 	 */
178 
179 	spin_lock(&kvm->mmu_lock);
180 	/* Flush slot from GPA */
181 	kvm_mips_flush_gpa_pt(kvm, slot->base_gfn,
182 			      slot->base_gfn + slot->npages - 1);
183 	/* Let implementation do the rest */
184 	kvm_mips_callbacks->flush_shadow_memslot(kvm, slot);
185 	spin_unlock(&kvm->mmu_lock);
186 }
187 
188 int kvm_arch_prepare_memory_region(struct kvm *kvm,
189 				   struct kvm_memory_slot *memslot,
190 				   const struct kvm_userspace_memory_region *mem,
191 				   enum kvm_mr_change change)
192 {
193 	return 0;
194 }
195 
196 void kvm_arch_commit_memory_region(struct kvm *kvm,
197 				   const struct kvm_userspace_memory_region *mem,
198 				   const struct kvm_memory_slot *old,
199 				   const struct kvm_memory_slot *new,
200 				   enum kvm_mr_change change)
201 {
202 	int needs_flush;
203 
204 	kvm_debug("%s: kvm: %p slot: %d, GPA: %llx, size: %llx, QVA: %llx\n",
205 		  __func__, kvm, mem->slot, mem->guest_phys_addr,
206 		  mem->memory_size, mem->userspace_addr);
207 
208 	/*
209 	 * If dirty page logging is enabled, write protect all pages in the slot
210 	 * ready for dirty logging.
211 	 *
212 	 * There is no need to do this in any of the following cases:
213 	 * CREATE:	No dirty mappings will already exist.
214 	 * MOVE/DELETE:	The old mappings will already have been cleaned up by
215 	 *		kvm_arch_flush_shadow_memslot()
216 	 */
217 	if (change == KVM_MR_FLAGS_ONLY &&
218 	    (!(old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
219 	     new->flags & KVM_MEM_LOG_DIRTY_PAGES)) {
220 		spin_lock(&kvm->mmu_lock);
221 		/* Write protect GPA page table entries */
222 		needs_flush = kvm_mips_mkclean_gpa_pt(kvm, new->base_gfn,
223 					new->base_gfn + new->npages - 1);
224 		/* Let implementation do the rest */
225 		if (needs_flush)
226 			kvm_mips_callbacks->flush_shadow_memslot(kvm, new);
227 		spin_unlock(&kvm->mmu_lock);
228 	}
229 }
230 
231 static inline void dump_handler(const char *symbol, void *start, void *end)
232 {
233 	u32 *p;
234 
235 	pr_debug("LEAF(%s)\n", symbol);
236 
237 	pr_debug("\t.set push\n");
238 	pr_debug("\t.set noreorder\n");
239 
240 	for (p = start; p < (u32 *)end; ++p)
241 		pr_debug("\t.word\t0x%08x\t\t# %p\n", *p, p);
242 
243 	pr_debug("\t.set\tpop\n");
244 
245 	pr_debug("\tEND(%s)\n", symbol);
246 }
247 
248 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
249 {
250 	int err, size;
251 	void *gebase, *p, *handler, *refill_start, *refill_end;
252 	int i;
253 
254 	struct kvm_vcpu *vcpu = kzalloc(sizeof(struct kvm_vcpu), GFP_KERNEL);
255 
256 	if (!vcpu) {
257 		err = -ENOMEM;
258 		goto out;
259 	}
260 
261 	err = kvm_vcpu_init(vcpu, kvm, id);
262 
263 	if (err)
264 		goto out_free_cpu;
265 
266 	kvm_debug("kvm @ %p: create cpu %d at %p\n", kvm, id, vcpu);
267 
268 	/*
269 	 * Allocate space for host mode exception handlers that handle
270 	 * guest mode exits
271 	 */
272 	if (cpu_has_veic || cpu_has_vint)
273 		size = 0x200 + VECTORSPACING * 64;
274 	else
275 		size = 0x4000;
276 
277 	gebase = kzalloc(ALIGN(size, PAGE_SIZE), GFP_KERNEL);
278 
279 	if (!gebase) {
280 		err = -ENOMEM;
281 		goto out_uninit_cpu;
282 	}
283 	kvm_debug("Allocated %d bytes for KVM Exception Handlers @ %p\n",
284 		  ALIGN(size, PAGE_SIZE), gebase);
285 
286 	/*
287 	 * Check new ebase actually fits in CP0_EBase. The lack of a write gate
288 	 * limits us to the low 512MB of physical address space. If the memory
289 	 * we allocate is out of range, just give up now.
290 	 */
291 	if (!cpu_has_ebase_wg && virt_to_phys(gebase) >= 0x20000000) {
292 		kvm_err("CP0_EBase.WG required for guest exception base %pK\n",
293 			gebase);
294 		err = -ENOMEM;
295 		goto out_free_gebase;
296 	}
297 
298 	/* Save new ebase */
299 	vcpu->arch.guest_ebase = gebase;
300 
301 	/* Build guest exception vectors dynamically in unmapped memory */
302 	handler = gebase + 0x2000;
303 
304 	/* TLB refill */
305 	refill_start = gebase;
306 	refill_end = kvm_mips_build_tlb_refill_exception(refill_start, handler);
307 
308 	/* General Exception Entry point */
309 	kvm_mips_build_exception(gebase + 0x180, handler);
310 
311 	/* For vectored interrupts poke the exception code @ all offsets 0-7 */
312 	for (i = 0; i < 8; i++) {
313 		kvm_debug("L1 Vectored handler @ %p\n",
314 			  gebase + 0x200 + (i * VECTORSPACING));
315 		kvm_mips_build_exception(gebase + 0x200 + i * VECTORSPACING,
316 					 handler);
317 	}
318 
319 	/* General exit handler */
320 	p = handler;
321 	p = kvm_mips_build_exit(p);
322 
323 	/* Guest entry routine */
324 	vcpu->arch.vcpu_run = p;
325 	p = kvm_mips_build_vcpu_run(p);
326 
327 	/* Dump the generated code */
328 	pr_debug("#include <asm/asm.h>\n");
329 	pr_debug("#include <asm/regdef.h>\n");
330 	pr_debug("\n");
331 	dump_handler("kvm_vcpu_run", vcpu->arch.vcpu_run, p);
332 	dump_handler("kvm_tlb_refill", refill_start, refill_end);
333 	dump_handler("kvm_gen_exc", gebase + 0x180, gebase + 0x200);
334 	dump_handler("kvm_exit", gebase + 0x2000, vcpu->arch.vcpu_run);
335 
336 	/* Invalidate the icache for these ranges */
337 	flush_icache_range((unsigned long)gebase,
338 			   (unsigned long)gebase + ALIGN(size, PAGE_SIZE));
339 
340 	/*
341 	 * Allocate comm page for guest kernel, a TLB will be reserved for
342 	 * mapping GVA @ 0xFFFF8000 to this page
343 	 */
344 	vcpu->arch.kseg0_commpage = kzalloc(PAGE_SIZE << 1, GFP_KERNEL);
345 
346 	if (!vcpu->arch.kseg0_commpage) {
347 		err = -ENOMEM;
348 		goto out_free_gebase;
349 	}
350 
351 	kvm_debug("Allocated COMM page @ %p\n", vcpu->arch.kseg0_commpage);
352 	kvm_mips_commpage_init(vcpu);
353 
354 	/* Init */
355 	vcpu->arch.last_sched_cpu = -1;
356 
357 	/* Start off the timer */
358 	kvm_mips_init_count(vcpu);
359 
360 	return vcpu;
361 
362 out_free_gebase:
363 	kfree(gebase);
364 
365 out_uninit_cpu:
366 	kvm_vcpu_uninit(vcpu);
367 
368 out_free_cpu:
369 	kfree(vcpu);
370 
371 out:
372 	return ERR_PTR(err);
373 }
374 
375 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
376 {
377 	hrtimer_cancel(&vcpu->arch.comparecount_timer);
378 
379 	kvm_vcpu_uninit(vcpu);
380 
381 	kvm_mips_dump_stats(vcpu);
382 
383 	kvm_mmu_free_memory_caches(vcpu);
384 	kfree(vcpu->arch.guest_ebase);
385 	kfree(vcpu->arch.kseg0_commpage);
386 	kfree(vcpu);
387 }
388 
389 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
390 {
391 	kvm_arch_vcpu_free(vcpu);
392 }
393 
394 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
395 					struct kvm_guest_debug *dbg)
396 {
397 	return -ENOIOCTLCMD;
398 }
399 
400 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
401 {
402 	int r = -EINTR;
403 	sigset_t sigsaved;
404 
405 	if (vcpu->sigset_active)
406 		sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
407 
408 	if (vcpu->mmio_needed) {
409 		if (!vcpu->mmio_is_write)
410 			kvm_mips_complete_mmio_load(vcpu, run);
411 		vcpu->mmio_needed = 0;
412 	}
413 
414 	if (run->immediate_exit)
415 		goto out;
416 
417 	lose_fpu(1);
418 
419 	local_irq_disable();
420 	guest_enter_irqoff();
421 	trace_kvm_enter(vcpu);
422 
423 	/*
424 	 * Make sure the read of VCPU requests in vcpu_run() callback is not
425 	 * reordered ahead of the write to vcpu->mode, or we could miss a TLB
426 	 * flush request while the requester sees the VCPU as outside of guest
427 	 * mode and not needing an IPI.
428 	 */
429 	smp_store_mb(vcpu->mode, IN_GUEST_MODE);
430 
431 	r = kvm_mips_callbacks->vcpu_run(run, vcpu);
432 
433 	trace_kvm_out(vcpu);
434 	guest_exit_irqoff();
435 	local_irq_enable();
436 
437 out:
438 	if (vcpu->sigset_active)
439 		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
440 
441 	return r;
442 }
443 
444 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
445 			     struct kvm_mips_interrupt *irq)
446 {
447 	int intr = (int)irq->irq;
448 	struct kvm_vcpu *dvcpu = NULL;
449 
450 	if (intr == 3 || intr == -3 || intr == 4 || intr == -4)
451 		kvm_debug("%s: CPU: %d, INTR: %d\n", __func__, irq->cpu,
452 			  (int)intr);
453 
454 	if (irq->cpu == -1)
455 		dvcpu = vcpu;
456 	else
457 		dvcpu = vcpu->kvm->vcpus[irq->cpu];
458 
459 	if (intr == 2 || intr == 3 || intr == 4) {
460 		kvm_mips_callbacks->queue_io_int(dvcpu, irq);
461 
462 	} else if (intr == -2 || intr == -3 || intr == -4) {
463 		kvm_mips_callbacks->dequeue_io_int(dvcpu, irq);
464 	} else {
465 		kvm_err("%s: invalid interrupt ioctl (%d:%d)\n", __func__,
466 			irq->cpu, irq->irq);
467 		return -EINVAL;
468 	}
469 
470 	dvcpu->arch.wait = 0;
471 
472 	if (swait_active(&dvcpu->wq))
473 		swake_up(&dvcpu->wq);
474 
475 	return 0;
476 }
477 
478 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
479 				    struct kvm_mp_state *mp_state)
480 {
481 	return -ENOIOCTLCMD;
482 }
483 
484 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
485 				    struct kvm_mp_state *mp_state)
486 {
487 	return -ENOIOCTLCMD;
488 }
489 
490 static u64 kvm_mips_get_one_regs[] = {
491 	KVM_REG_MIPS_R0,
492 	KVM_REG_MIPS_R1,
493 	KVM_REG_MIPS_R2,
494 	KVM_REG_MIPS_R3,
495 	KVM_REG_MIPS_R4,
496 	KVM_REG_MIPS_R5,
497 	KVM_REG_MIPS_R6,
498 	KVM_REG_MIPS_R7,
499 	KVM_REG_MIPS_R8,
500 	KVM_REG_MIPS_R9,
501 	KVM_REG_MIPS_R10,
502 	KVM_REG_MIPS_R11,
503 	KVM_REG_MIPS_R12,
504 	KVM_REG_MIPS_R13,
505 	KVM_REG_MIPS_R14,
506 	KVM_REG_MIPS_R15,
507 	KVM_REG_MIPS_R16,
508 	KVM_REG_MIPS_R17,
509 	KVM_REG_MIPS_R18,
510 	KVM_REG_MIPS_R19,
511 	KVM_REG_MIPS_R20,
512 	KVM_REG_MIPS_R21,
513 	KVM_REG_MIPS_R22,
514 	KVM_REG_MIPS_R23,
515 	KVM_REG_MIPS_R24,
516 	KVM_REG_MIPS_R25,
517 	KVM_REG_MIPS_R26,
518 	KVM_REG_MIPS_R27,
519 	KVM_REG_MIPS_R28,
520 	KVM_REG_MIPS_R29,
521 	KVM_REG_MIPS_R30,
522 	KVM_REG_MIPS_R31,
523 
524 #ifndef CONFIG_CPU_MIPSR6
525 	KVM_REG_MIPS_HI,
526 	KVM_REG_MIPS_LO,
527 #endif
528 	KVM_REG_MIPS_PC,
529 };
530 
531 static u64 kvm_mips_get_one_regs_fpu[] = {
532 	KVM_REG_MIPS_FCR_IR,
533 	KVM_REG_MIPS_FCR_CSR,
534 };
535 
536 static u64 kvm_mips_get_one_regs_msa[] = {
537 	KVM_REG_MIPS_MSA_IR,
538 	KVM_REG_MIPS_MSA_CSR,
539 };
540 
541 static unsigned long kvm_mips_num_regs(struct kvm_vcpu *vcpu)
542 {
543 	unsigned long ret;
544 
545 	ret = ARRAY_SIZE(kvm_mips_get_one_regs);
546 	if (kvm_mips_guest_can_have_fpu(&vcpu->arch)) {
547 		ret += ARRAY_SIZE(kvm_mips_get_one_regs_fpu) + 48;
548 		/* odd doubles */
549 		if (boot_cpu_data.fpu_id & MIPS_FPIR_F64)
550 			ret += 16;
551 	}
552 	if (kvm_mips_guest_can_have_msa(&vcpu->arch))
553 		ret += ARRAY_SIZE(kvm_mips_get_one_regs_msa) + 32;
554 	ret += kvm_mips_callbacks->num_regs(vcpu);
555 
556 	return ret;
557 }
558 
559 static int kvm_mips_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *indices)
560 {
561 	u64 index;
562 	unsigned int i;
563 
564 	if (copy_to_user(indices, kvm_mips_get_one_regs,
565 			 sizeof(kvm_mips_get_one_regs)))
566 		return -EFAULT;
567 	indices += ARRAY_SIZE(kvm_mips_get_one_regs);
568 
569 	if (kvm_mips_guest_can_have_fpu(&vcpu->arch)) {
570 		if (copy_to_user(indices, kvm_mips_get_one_regs_fpu,
571 				 sizeof(kvm_mips_get_one_regs_fpu)))
572 			return -EFAULT;
573 		indices += ARRAY_SIZE(kvm_mips_get_one_regs_fpu);
574 
575 		for (i = 0; i < 32; ++i) {
576 			index = KVM_REG_MIPS_FPR_32(i);
577 			if (copy_to_user(indices, &index, sizeof(index)))
578 				return -EFAULT;
579 			++indices;
580 
581 			/* skip odd doubles if no F64 */
582 			if (i & 1 && !(boot_cpu_data.fpu_id & MIPS_FPIR_F64))
583 				continue;
584 
585 			index = KVM_REG_MIPS_FPR_64(i);
586 			if (copy_to_user(indices, &index, sizeof(index)))
587 				return -EFAULT;
588 			++indices;
589 		}
590 	}
591 
592 	if (kvm_mips_guest_can_have_msa(&vcpu->arch)) {
593 		if (copy_to_user(indices, kvm_mips_get_one_regs_msa,
594 				 sizeof(kvm_mips_get_one_regs_msa)))
595 			return -EFAULT;
596 		indices += ARRAY_SIZE(kvm_mips_get_one_regs_msa);
597 
598 		for (i = 0; i < 32; ++i) {
599 			index = KVM_REG_MIPS_VEC_128(i);
600 			if (copy_to_user(indices, &index, sizeof(index)))
601 				return -EFAULT;
602 			++indices;
603 		}
604 	}
605 
606 	return kvm_mips_callbacks->copy_reg_indices(vcpu, indices);
607 }
608 
609 static int kvm_mips_get_reg(struct kvm_vcpu *vcpu,
610 			    const struct kvm_one_reg *reg)
611 {
612 	struct mips_coproc *cop0 = vcpu->arch.cop0;
613 	struct mips_fpu_struct *fpu = &vcpu->arch.fpu;
614 	int ret;
615 	s64 v;
616 	s64 vs[2];
617 	unsigned int idx;
618 
619 	switch (reg->id) {
620 	/* General purpose registers */
621 	case KVM_REG_MIPS_R0 ... KVM_REG_MIPS_R31:
622 		v = (long)vcpu->arch.gprs[reg->id - KVM_REG_MIPS_R0];
623 		break;
624 #ifndef CONFIG_CPU_MIPSR6
625 	case KVM_REG_MIPS_HI:
626 		v = (long)vcpu->arch.hi;
627 		break;
628 	case KVM_REG_MIPS_LO:
629 		v = (long)vcpu->arch.lo;
630 		break;
631 #endif
632 	case KVM_REG_MIPS_PC:
633 		v = (long)vcpu->arch.pc;
634 		break;
635 
636 	/* Floating point registers */
637 	case KVM_REG_MIPS_FPR_32(0) ... KVM_REG_MIPS_FPR_32(31):
638 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
639 			return -EINVAL;
640 		idx = reg->id - KVM_REG_MIPS_FPR_32(0);
641 		/* Odd singles in top of even double when FR=0 */
642 		if (kvm_read_c0_guest_status(cop0) & ST0_FR)
643 			v = get_fpr32(&fpu->fpr[idx], 0);
644 		else
645 			v = get_fpr32(&fpu->fpr[idx & ~1], idx & 1);
646 		break;
647 	case KVM_REG_MIPS_FPR_64(0) ... KVM_REG_MIPS_FPR_64(31):
648 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
649 			return -EINVAL;
650 		idx = reg->id - KVM_REG_MIPS_FPR_64(0);
651 		/* Can't access odd doubles in FR=0 mode */
652 		if (idx & 1 && !(kvm_read_c0_guest_status(cop0) & ST0_FR))
653 			return -EINVAL;
654 		v = get_fpr64(&fpu->fpr[idx], 0);
655 		break;
656 	case KVM_REG_MIPS_FCR_IR:
657 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
658 			return -EINVAL;
659 		v = boot_cpu_data.fpu_id;
660 		break;
661 	case KVM_REG_MIPS_FCR_CSR:
662 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
663 			return -EINVAL;
664 		v = fpu->fcr31;
665 		break;
666 
667 	/* MIPS SIMD Architecture (MSA) registers */
668 	case KVM_REG_MIPS_VEC_128(0) ... KVM_REG_MIPS_VEC_128(31):
669 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
670 			return -EINVAL;
671 		/* Can't access MSA registers in FR=0 mode */
672 		if (!(kvm_read_c0_guest_status(cop0) & ST0_FR))
673 			return -EINVAL;
674 		idx = reg->id - KVM_REG_MIPS_VEC_128(0);
675 #ifdef CONFIG_CPU_LITTLE_ENDIAN
676 		/* least significant byte first */
677 		vs[0] = get_fpr64(&fpu->fpr[idx], 0);
678 		vs[1] = get_fpr64(&fpu->fpr[idx], 1);
679 #else
680 		/* most significant byte first */
681 		vs[0] = get_fpr64(&fpu->fpr[idx], 1);
682 		vs[1] = get_fpr64(&fpu->fpr[idx], 0);
683 #endif
684 		break;
685 	case KVM_REG_MIPS_MSA_IR:
686 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
687 			return -EINVAL;
688 		v = boot_cpu_data.msa_id;
689 		break;
690 	case KVM_REG_MIPS_MSA_CSR:
691 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
692 			return -EINVAL;
693 		v = fpu->msacsr;
694 		break;
695 
696 	/* registers to be handled specially */
697 	default:
698 		ret = kvm_mips_callbacks->get_one_reg(vcpu, reg, &v);
699 		if (ret)
700 			return ret;
701 		break;
702 	}
703 	if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U64) {
704 		u64 __user *uaddr64 = (u64 __user *)(long)reg->addr;
705 
706 		return put_user(v, uaddr64);
707 	} else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U32) {
708 		u32 __user *uaddr32 = (u32 __user *)(long)reg->addr;
709 		u32 v32 = (u32)v;
710 
711 		return put_user(v32, uaddr32);
712 	} else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U128) {
713 		void __user *uaddr = (void __user *)(long)reg->addr;
714 
715 		return copy_to_user(uaddr, vs, 16) ? -EFAULT : 0;
716 	} else {
717 		return -EINVAL;
718 	}
719 }
720 
721 static int kvm_mips_set_reg(struct kvm_vcpu *vcpu,
722 			    const struct kvm_one_reg *reg)
723 {
724 	struct mips_coproc *cop0 = vcpu->arch.cop0;
725 	struct mips_fpu_struct *fpu = &vcpu->arch.fpu;
726 	s64 v;
727 	s64 vs[2];
728 	unsigned int idx;
729 
730 	if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U64) {
731 		u64 __user *uaddr64 = (u64 __user *)(long)reg->addr;
732 
733 		if (get_user(v, uaddr64) != 0)
734 			return -EFAULT;
735 	} else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U32) {
736 		u32 __user *uaddr32 = (u32 __user *)(long)reg->addr;
737 		s32 v32;
738 
739 		if (get_user(v32, uaddr32) != 0)
740 			return -EFAULT;
741 		v = (s64)v32;
742 	} else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U128) {
743 		void __user *uaddr = (void __user *)(long)reg->addr;
744 
745 		return copy_from_user(vs, uaddr, 16) ? -EFAULT : 0;
746 	} else {
747 		return -EINVAL;
748 	}
749 
750 	switch (reg->id) {
751 	/* General purpose registers */
752 	case KVM_REG_MIPS_R0:
753 		/* Silently ignore requests to set $0 */
754 		break;
755 	case KVM_REG_MIPS_R1 ... KVM_REG_MIPS_R31:
756 		vcpu->arch.gprs[reg->id - KVM_REG_MIPS_R0] = v;
757 		break;
758 #ifndef CONFIG_CPU_MIPSR6
759 	case KVM_REG_MIPS_HI:
760 		vcpu->arch.hi = v;
761 		break;
762 	case KVM_REG_MIPS_LO:
763 		vcpu->arch.lo = v;
764 		break;
765 #endif
766 	case KVM_REG_MIPS_PC:
767 		vcpu->arch.pc = v;
768 		break;
769 
770 	/* Floating point registers */
771 	case KVM_REG_MIPS_FPR_32(0) ... KVM_REG_MIPS_FPR_32(31):
772 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
773 			return -EINVAL;
774 		idx = reg->id - KVM_REG_MIPS_FPR_32(0);
775 		/* Odd singles in top of even double when FR=0 */
776 		if (kvm_read_c0_guest_status(cop0) & ST0_FR)
777 			set_fpr32(&fpu->fpr[idx], 0, v);
778 		else
779 			set_fpr32(&fpu->fpr[idx & ~1], idx & 1, v);
780 		break;
781 	case KVM_REG_MIPS_FPR_64(0) ... KVM_REG_MIPS_FPR_64(31):
782 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
783 			return -EINVAL;
784 		idx = reg->id - KVM_REG_MIPS_FPR_64(0);
785 		/* Can't access odd doubles in FR=0 mode */
786 		if (idx & 1 && !(kvm_read_c0_guest_status(cop0) & ST0_FR))
787 			return -EINVAL;
788 		set_fpr64(&fpu->fpr[idx], 0, v);
789 		break;
790 	case KVM_REG_MIPS_FCR_IR:
791 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
792 			return -EINVAL;
793 		/* Read-only */
794 		break;
795 	case KVM_REG_MIPS_FCR_CSR:
796 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
797 			return -EINVAL;
798 		fpu->fcr31 = v;
799 		break;
800 
801 	/* MIPS SIMD Architecture (MSA) registers */
802 	case KVM_REG_MIPS_VEC_128(0) ... KVM_REG_MIPS_VEC_128(31):
803 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
804 			return -EINVAL;
805 		idx = reg->id - KVM_REG_MIPS_VEC_128(0);
806 #ifdef CONFIG_CPU_LITTLE_ENDIAN
807 		/* least significant byte first */
808 		set_fpr64(&fpu->fpr[idx], 0, vs[0]);
809 		set_fpr64(&fpu->fpr[idx], 1, vs[1]);
810 #else
811 		/* most significant byte first */
812 		set_fpr64(&fpu->fpr[idx], 1, vs[0]);
813 		set_fpr64(&fpu->fpr[idx], 0, vs[1]);
814 #endif
815 		break;
816 	case KVM_REG_MIPS_MSA_IR:
817 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
818 			return -EINVAL;
819 		/* Read-only */
820 		break;
821 	case KVM_REG_MIPS_MSA_CSR:
822 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
823 			return -EINVAL;
824 		fpu->msacsr = v;
825 		break;
826 
827 	/* registers to be handled specially */
828 	default:
829 		return kvm_mips_callbacks->set_one_reg(vcpu, reg, v);
830 	}
831 	return 0;
832 }
833 
834 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
835 				     struct kvm_enable_cap *cap)
836 {
837 	int r = 0;
838 
839 	if (!kvm_vm_ioctl_check_extension(vcpu->kvm, cap->cap))
840 		return -EINVAL;
841 	if (cap->flags)
842 		return -EINVAL;
843 	if (cap->args[0])
844 		return -EINVAL;
845 
846 	switch (cap->cap) {
847 	case KVM_CAP_MIPS_FPU:
848 		vcpu->arch.fpu_enabled = true;
849 		break;
850 	case KVM_CAP_MIPS_MSA:
851 		vcpu->arch.msa_enabled = true;
852 		break;
853 	default:
854 		r = -EINVAL;
855 		break;
856 	}
857 
858 	return r;
859 }
860 
861 long kvm_arch_vcpu_ioctl(struct file *filp, unsigned int ioctl,
862 			 unsigned long arg)
863 {
864 	struct kvm_vcpu *vcpu = filp->private_data;
865 	void __user *argp = (void __user *)arg;
866 	long r;
867 
868 	switch (ioctl) {
869 	case KVM_SET_ONE_REG:
870 	case KVM_GET_ONE_REG: {
871 		struct kvm_one_reg reg;
872 
873 		if (copy_from_user(&reg, argp, sizeof(reg)))
874 			return -EFAULT;
875 		if (ioctl == KVM_SET_ONE_REG)
876 			return kvm_mips_set_reg(vcpu, &reg);
877 		else
878 			return kvm_mips_get_reg(vcpu, &reg);
879 	}
880 	case KVM_GET_REG_LIST: {
881 		struct kvm_reg_list __user *user_list = argp;
882 		struct kvm_reg_list reg_list;
883 		unsigned n;
884 
885 		if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
886 			return -EFAULT;
887 		n = reg_list.n;
888 		reg_list.n = kvm_mips_num_regs(vcpu);
889 		if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
890 			return -EFAULT;
891 		if (n < reg_list.n)
892 			return -E2BIG;
893 		return kvm_mips_copy_reg_indices(vcpu, user_list->reg);
894 	}
895 	case KVM_INTERRUPT:
896 		{
897 			struct kvm_mips_interrupt irq;
898 
899 			if (copy_from_user(&irq, argp, sizeof(irq)))
900 				return -EFAULT;
901 			kvm_debug("[%d] %s: irq: %d\n", vcpu->vcpu_id, __func__,
902 				  irq.irq);
903 
904 			r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
905 			break;
906 		}
907 	case KVM_ENABLE_CAP: {
908 		struct kvm_enable_cap cap;
909 
910 		if (copy_from_user(&cap, argp, sizeof(cap)))
911 			return -EFAULT;
912 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
913 		break;
914 	}
915 	default:
916 		r = -ENOIOCTLCMD;
917 	}
918 	return r;
919 }
920 
921 /**
922  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
923  * @kvm: kvm instance
924  * @log: slot id and address to which we copy the log
925  *
926  * Steps 1-4 below provide general overview of dirty page logging. See
927  * kvm_get_dirty_log_protect() function description for additional details.
928  *
929  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
930  * always flush the TLB (step 4) even if previous step failed  and the dirty
931  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
932  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
933  * writes will be marked dirty for next log read.
934  *
935  *   1. Take a snapshot of the bit and clear it if needed.
936  *   2. Write protect the corresponding page.
937  *   3. Copy the snapshot to the userspace.
938  *   4. Flush TLB's if needed.
939  */
940 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
941 {
942 	struct kvm_memslots *slots;
943 	struct kvm_memory_slot *memslot;
944 	bool is_dirty = false;
945 	int r;
946 
947 	mutex_lock(&kvm->slots_lock);
948 
949 	r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
950 
951 	if (is_dirty) {
952 		slots = kvm_memslots(kvm);
953 		memslot = id_to_memslot(slots, log->slot);
954 
955 		/* Let implementation handle TLB/GVA invalidation */
956 		kvm_mips_callbacks->flush_shadow_memslot(kvm, memslot);
957 	}
958 
959 	mutex_unlock(&kvm->slots_lock);
960 	return r;
961 }
962 
963 long kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
964 {
965 	long r;
966 
967 	switch (ioctl) {
968 	default:
969 		r = -ENOIOCTLCMD;
970 	}
971 
972 	return r;
973 }
974 
975 int kvm_arch_init(void *opaque)
976 {
977 	if (kvm_mips_callbacks) {
978 		kvm_err("kvm: module already exists\n");
979 		return -EEXIST;
980 	}
981 
982 	return kvm_mips_emulation_init(&kvm_mips_callbacks);
983 }
984 
985 void kvm_arch_exit(void)
986 {
987 	kvm_mips_callbacks = NULL;
988 }
989 
990 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
991 				  struct kvm_sregs *sregs)
992 {
993 	return -ENOIOCTLCMD;
994 }
995 
996 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
997 				  struct kvm_sregs *sregs)
998 {
999 	return -ENOIOCTLCMD;
1000 }
1001 
1002 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
1003 {
1004 }
1005 
1006 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1007 {
1008 	return -ENOIOCTLCMD;
1009 }
1010 
1011 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1012 {
1013 	return -ENOIOCTLCMD;
1014 }
1015 
1016 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
1017 {
1018 	return VM_FAULT_SIGBUS;
1019 }
1020 
1021 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
1022 {
1023 	int r;
1024 
1025 	switch (ext) {
1026 	case KVM_CAP_ONE_REG:
1027 	case KVM_CAP_ENABLE_CAP:
1028 	case KVM_CAP_READONLY_MEM:
1029 	case KVM_CAP_SYNC_MMU:
1030 	case KVM_CAP_IMMEDIATE_EXIT:
1031 		r = 1;
1032 		break;
1033 	case KVM_CAP_COALESCED_MMIO:
1034 		r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1035 		break;
1036 	case KVM_CAP_NR_VCPUS:
1037 		r = num_online_cpus();
1038 		break;
1039 	case KVM_CAP_MAX_VCPUS:
1040 		r = KVM_MAX_VCPUS;
1041 		break;
1042 	case KVM_CAP_MIPS_FPU:
1043 		/* We don't handle systems with inconsistent cpu_has_fpu */
1044 		r = !!raw_cpu_has_fpu;
1045 		break;
1046 	case KVM_CAP_MIPS_MSA:
1047 		/*
1048 		 * We don't support MSA vector partitioning yet:
1049 		 * 1) It would require explicit support which can't be tested
1050 		 *    yet due to lack of support in current hardware.
1051 		 * 2) It extends the state that would need to be saved/restored
1052 		 *    by e.g. QEMU for migration.
1053 		 *
1054 		 * When vector partitioning hardware becomes available, support
1055 		 * could be added by requiring a flag when enabling
1056 		 * KVM_CAP_MIPS_MSA capability to indicate that userland knows
1057 		 * to save/restore the appropriate extra state.
1058 		 */
1059 		r = cpu_has_msa && !(boot_cpu_data.msa_id & MSA_IR_WRPF);
1060 		break;
1061 	default:
1062 		r = 0;
1063 		break;
1064 	}
1065 	return r;
1066 }
1067 
1068 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1069 {
1070 	return kvm_mips_pending_timer(vcpu);
1071 }
1072 
1073 int kvm_arch_vcpu_dump_regs(struct kvm_vcpu *vcpu)
1074 {
1075 	int i;
1076 	struct mips_coproc *cop0;
1077 
1078 	if (!vcpu)
1079 		return -1;
1080 
1081 	kvm_debug("VCPU Register Dump:\n");
1082 	kvm_debug("\tpc = 0x%08lx\n", vcpu->arch.pc);
1083 	kvm_debug("\texceptions: %08lx\n", vcpu->arch.pending_exceptions);
1084 
1085 	for (i = 0; i < 32; i += 4) {
1086 		kvm_debug("\tgpr%02d: %08lx %08lx %08lx %08lx\n", i,
1087 		       vcpu->arch.gprs[i],
1088 		       vcpu->arch.gprs[i + 1],
1089 		       vcpu->arch.gprs[i + 2], vcpu->arch.gprs[i + 3]);
1090 	}
1091 	kvm_debug("\thi: 0x%08lx\n", vcpu->arch.hi);
1092 	kvm_debug("\tlo: 0x%08lx\n", vcpu->arch.lo);
1093 
1094 	cop0 = vcpu->arch.cop0;
1095 	kvm_debug("\tStatus: 0x%08lx, Cause: 0x%08lx\n",
1096 		  kvm_read_c0_guest_status(cop0),
1097 		  kvm_read_c0_guest_cause(cop0));
1098 
1099 	kvm_debug("\tEPC: 0x%08lx\n", kvm_read_c0_guest_epc(cop0));
1100 
1101 	return 0;
1102 }
1103 
1104 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1105 {
1106 	int i;
1107 
1108 	for (i = 1; i < ARRAY_SIZE(vcpu->arch.gprs); i++)
1109 		vcpu->arch.gprs[i] = regs->gpr[i];
1110 	vcpu->arch.gprs[0] = 0; /* zero is special, and cannot be set. */
1111 	vcpu->arch.hi = regs->hi;
1112 	vcpu->arch.lo = regs->lo;
1113 	vcpu->arch.pc = regs->pc;
1114 
1115 	return 0;
1116 }
1117 
1118 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1119 {
1120 	int i;
1121 
1122 	for (i = 0; i < ARRAY_SIZE(vcpu->arch.gprs); i++)
1123 		regs->gpr[i] = vcpu->arch.gprs[i];
1124 
1125 	regs->hi = vcpu->arch.hi;
1126 	regs->lo = vcpu->arch.lo;
1127 	regs->pc = vcpu->arch.pc;
1128 
1129 	return 0;
1130 }
1131 
1132 static void kvm_mips_comparecount_func(unsigned long data)
1133 {
1134 	struct kvm_vcpu *vcpu = (struct kvm_vcpu *)data;
1135 
1136 	kvm_mips_callbacks->queue_timer_int(vcpu);
1137 
1138 	vcpu->arch.wait = 0;
1139 	if (swait_active(&vcpu->wq))
1140 		swake_up(&vcpu->wq);
1141 }
1142 
1143 /* low level hrtimer wake routine */
1144 static enum hrtimer_restart kvm_mips_comparecount_wakeup(struct hrtimer *timer)
1145 {
1146 	struct kvm_vcpu *vcpu;
1147 
1148 	vcpu = container_of(timer, struct kvm_vcpu, arch.comparecount_timer);
1149 	kvm_mips_comparecount_func((unsigned long) vcpu);
1150 	return kvm_mips_count_timeout(vcpu);
1151 }
1152 
1153 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
1154 {
1155 	int err;
1156 
1157 	err = kvm_mips_callbacks->vcpu_init(vcpu);
1158 	if (err)
1159 		return err;
1160 
1161 	hrtimer_init(&vcpu->arch.comparecount_timer, CLOCK_MONOTONIC,
1162 		     HRTIMER_MODE_REL);
1163 	vcpu->arch.comparecount_timer.function = kvm_mips_comparecount_wakeup;
1164 	return 0;
1165 }
1166 
1167 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
1168 {
1169 	kvm_mips_callbacks->vcpu_uninit(vcpu);
1170 }
1171 
1172 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1173 				  struct kvm_translation *tr)
1174 {
1175 	return 0;
1176 }
1177 
1178 /* Initial guest state */
1179 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
1180 {
1181 	return kvm_mips_callbacks->vcpu_setup(vcpu);
1182 }
1183 
1184 static void kvm_mips_set_c0_status(void)
1185 {
1186 	u32 status = read_c0_status();
1187 
1188 	if (cpu_has_dsp)
1189 		status |= (ST0_MX);
1190 
1191 	write_c0_status(status);
1192 	ehb();
1193 }
1194 
1195 /*
1196  * Return value is in the form (errcode<<2 | RESUME_FLAG_HOST | RESUME_FLAG_NV)
1197  */
1198 int kvm_mips_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu)
1199 {
1200 	u32 cause = vcpu->arch.host_cp0_cause;
1201 	u32 exccode = (cause >> CAUSEB_EXCCODE) & 0x1f;
1202 	u32 __user *opc = (u32 __user *) vcpu->arch.pc;
1203 	unsigned long badvaddr = vcpu->arch.host_cp0_badvaddr;
1204 	enum emulation_result er = EMULATE_DONE;
1205 	u32 inst;
1206 	int ret = RESUME_GUEST;
1207 
1208 	vcpu->mode = OUTSIDE_GUEST_MODE;
1209 
1210 	/* re-enable HTW before enabling interrupts */
1211 	htw_start();
1212 
1213 	/* Set a default exit reason */
1214 	run->exit_reason = KVM_EXIT_UNKNOWN;
1215 	run->ready_for_interrupt_injection = 1;
1216 
1217 	/*
1218 	 * Set the appropriate status bits based on host CPU features,
1219 	 * before we hit the scheduler
1220 	 */
1221 	kvm_mips_set_c0_status();
1222 
1223 	local_irq_enable();
1224 
1225 	kvm_debug("kvm_mips_handle_exit: cause: %#x, PC: %p, kvm_run: %p, kvm_vcpu: %p\n",
1226 			cause, opc, run, vcpu);
1227 	trace_kvm_exit(vcpu, exccode);
1228 
1229 	/*
1230 	 * Do a privilege check, if in UM most of these exit conditions end up
1231 	 * causing an exception to be delivered to the Guest Kernel
1232 	 */
1233 	er = kvm_mips_check_privilege(cause, opc, run, vcpu);
1234 	if (er == EMULATE_PRIV_FAIL) {
1235 		goto skip_emul;
1236 	} else if (er == EMULATE_FAIL) {
1237 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1238 		ret = RESUME_HOST;
1239 		goto skip_emul;
1240 	}
1241 
1242 	switch (exccode) {
1243 	case EXCCODE_INT:
1244 		kvm_debug("[%d]EXCCODE_INT @ %p\n", vcpu->vcpu_id, opc);
1245 
1246 		++vcpu->stat.int_exits;
1247 
1248 		if (need_resched())
1249 			cond_resched();
1250 
1251 		ret = RESUME_GUEST;
1252 		break;
1253 
1254 	case EXCCODE_CPU:
1255 		kvm_debug("EXCCODE_CPU: @ PC: %p\n", opc);
1256 
1257 		++vcpu->stat.cop_unusable_exits;
1258 		ret = kvm_mips_callbacks->handle_cop_unusable(vcpu);
1259 		/* XXXKYMA: Might need to return to user space */
1260 		if (run->exit_reason == KVM_EXIT_IRQ_WINDOW_OPEN)
1261 			ret = RESUME_HOST;
1262 		break;
1263 
1264 	case EXCCODE_MOD:
1265 		++vcpu->stat.tlbmod_exits;
1266 		ret = kvm_mips_callbacks->handle_tlb_mod(vcpu);
1267 		break;
1268 
1269 	case EXCCODE_TLBS:
1270 		kvm_debug("TLB ST fault:  cause %#x, status %#lx, PC: %p, BadVaddr: %#lx\n",
1271 			  cause, kvm_read_c0_guest_status(vcpu->arch.cop0), opc,
1272 			  badvaddr);
1273 
1274 		++vcpu->stat.tlbmiss_st_exits;
1275 		ret = kvm_mips_callbacks->handle_tlb_st_miss(vcpu);
1276 		break;
1277 
1278 	case EXCCODE_TLBL:
1279 		kvm_debug("TLB LD fault: cause %#x, PC: %p, BadVaddr: %#lx\n",
1280 			  cause, opc, badvaddr);
1281 
1282 		++vcpu->stat.tlbmiss_ld_exits;
1283 		ret = kvm_mips_callbacks->handle_tlb_ld_miss(vcpu);
1284 		break;
1285 
1286 	case EXCCODE_ADES:
1287 		++vcpu->stat.addrerr_st_exits;
1288 		ret = kvm_mips_callbacks->handle_addr_err_st(vcpu);
1289 		break;
1290 
1291 	case EXCCODE_ADEL:
1292 		++vcpu->stat.addrerr_ld_exits;
1293 		ret = kvm_mips_callbacks->handle_addr_err_ld(vcpu);
1294 		break;
1295 
1296 	case EXCCODE_SYS:
1297 		++vcpu->stat.syscall_exits;
1298 		ret = kvm_mips_callbacks->handle_syscall(vcpu);
1299 		break;
1300 
1301 	case EXCCODE_RI:
1302 		++vcpu->stat.resvd_inst_exits;
1303 		ret = kvm_mips_callbacks->handle_res_inst(vcpu);
1304 		break;
1305 
1306 	case EXCCODE_BP:
1307 		++vcpu->stat.break_inst_exits;
1308 		ret = kvm_mips_callbacks->handle_break(vcpu);
1309 		break;
1310 
1311 	case EXCCODE_TR:
1312 		++vcpu->stat.trap_inst_exits;
1313 		ret = kvm_mips_callbacks->handle_trap(vcpu);
1314 		break;
1315 
1316 	case EXCCODE_MSAFPE:
1317 		++vcpu->stat.msa_fpe_exits;
1318 		ret = kvm_mips_callbacks->handle_msa_fpe(vcpu);
1319 		break;
1320 
1321 	case EXCCODE_FPE:
1322 		++vcpu->stat.fpe_exits;
1323 		ret = kvm_mips_callbacks->handle_fpe(vcpu);
1324 		break;
1325 
1326 	case EXCCODE_MSADIS:
1327 		++vcpu->stat.msa_disabled_exits;
1328 		ret = kvm_mips_callbacks->handle_msa_disabled(vcpu);
1329 		break;
1330 
1331 	default:
1332 		if (cause & CAUSEF_BD)
1333 			opc += 1;
1334 		inst = 0;
1335 		kvm_get_badinstr(opc, vcpu, &inst);
1336 		kvm_err("Exception Code: %d, not yet handled, @ PC: %p, inst: 0x%08x  BadVaddr: %#lx Status: %#lx\n",
1337 			exccode, opc, inst, badvaddr,
1338 			kvm_read_c0_guest_status(vcpu->arch.cop0));
1339 		kvm_arch_vcpu_dump_regs(vcpu);
1340 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1341 		ret = RESUME_HOST;
1342 		break;
1343 
1344 	}
1345 
1346 skip_emul:
1347 	local_irq_disable();
1348 
1349 	if (er == EMULATE_DONE && !(ret & RESUME_HOST))
1350 		kvm_mips_deliver_interrupts(vcpu, cause);
1351 
1352 	if (!(ret & RESUME_HOST)) {
1353 		/* Only check for signals if not already exiting to userspace */
1354 		if (signal_pending(current)) {
1355 			run->exit_reason = KVM_EXIT_INTR;
1356 			ret = (-EINTR << 2) | RESUME_HOST;
1357 			++vcpu->stat.signal_exits;
1358 			trace_kvm_exit(vcpu, KVM_TRACE_EXIT_SIGNAL);
1359 		}
1360 	}
1361 
1362 	if (ret == RESUME_GUEST) {
1363 		trace_kvm_reenter(vcpu);
1364 
1365 		/*
1366 		 * Make sure the read of VCPU requests in vcpu_reenter()
1367 		 * callback is not reordered ahead of the write to vcpu->mode,
1368 		 * or we could miss a TLB flush request while the requester sees
1369 		 * the VCPU as outside of guest mode and not needing an IPI.
1370 		 */
1371 		smp_store_mb(vcpu->mode, IN_GUEST_MODE);
1372 
1373 		kvm_mips_callbacks->vcpu_reenter(run, vcpu);
1374 
1375 		/*
1376 		 * If FPU / MSA are enabled (i.e. the guest's FPU / MSA context
1377 		 * is live), restore FCR31 / MSACSR.
1378 		 *
1379 		 * This should be before returning to the guest exception
1380 		 * vector, as it may well cause an [MSA] FP exception if there
1381 		 * are pending exception bits unmasked. (see
1382 		 * kvm_mips_csr_die_notifier() for how that is handled).
1383 		 */
1384 		if (kvm_mips_guest_has_fpu(&vcpu->arch) &&
1385 		    read_c0_status() & ST0_CU1)
1386 			__kvm_restore_fcsr(&vcpu->arch);
1387 
1388 		if (kvm_mips_guest_has_msa(&vcpu->arch) &&
1389 		    read_c0_config5() & MIPS_CONF5_MSAEN)
1390 			__kvm_restore_msacsr(&vcpu->arch);
1391 	}
1392 
1393 	/* Disable HTW before returning to guest or host */
1394 	htw_stop();
1395 
1396 	return ret;
1397 }
1398 
1399 /* Enable FPU for guest and restore context */
1400 void kvm_own_fpu(struct kvm_vcpu *vcpu)
1401 {
1402 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1403 	unsigned int sr, cfg5;
1404 
1405 	preempt_disable();
1406 
1407 	sr = kvm_read_c0_guest_status(cop0);
1408 
1409 	/*
1410 	 * If MSA state is already live, it is undefined how it interacts with
1411 	 * FR=0 FPU state, and we don't want to hit reserved instruction
1412 	 * exceptions trying to save the MSA state later when CU=1 && FR=1, so
1413 	 * play it safe and save it first.
1414 	 *
1415 	 * In theory we shouldn't ever hit this case since kvm_lose_fpu() should
1416 	 * get called when guest CU1 is set, however we can't trust the guest
1417 	 * not to clobber the status register directly via the commpage.
1418 	 */
1419 	if (cpu_has_msa && sr & ST0_CU1 && !(sr & ST0_FR) &&
1420 	    vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA)
1421 		kvm_lose_fpu(vcpu);
1422 
1423 	/*
1424 	 * Enable FPU for guest
1425 	 * We set FR and FRE according to guest context
1426 	 */
1427 	change_c0_status(ST0_CU1 | ST0_FR, sr);
1428 	if (cpu_has_fre) {
1429 		cfg5 = kvm_read_c0_guest_config5(cop0);
1430 		change_c0_config5(MIPS_CONF5_FRE, cfg5);
1431 	}
1432 	enable_fpu_hazard();
1433 
1434 	/* If guest FPU state not active, restore it now */
1435 	if (!(vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU)) {
1436 		__kvm_restore_fpu(&vcpu->arch);
1437 		vcpu->arch.aux_inuse |= KVM_MIPS_AUX_FPU;
1438 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE, KVM_TRACE_AUX_FPU);
1439 	} else {
1440 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_ENABLE, KVM_TRACE_AUX_FPU);
1441 	}
1442 
1443 	preempt_enable();
1444 }
1445 
1446 #ifdef CONFIG_CPU_HAS_MSA
1447 /* Enable MSA for guest and restore context */
1448 void kvm_own_msa(struct kvm_vcpu *vcpu)
1449 {
1450 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1451 	unsigned int sr, cfg5;
1452 
1453 	preempt_disable();
1454 
1455 	/*
1456 	 * Enable FPU if enabled in guest, since we're restoring FPU context
1457 	 * anyway. We set FR and FRE according to guest context.
1458 	 */
1459 	if (kvm_mips_guest_has_fpu(&vcpu->arch)) {
1460 		sr = kvm_read_c0_guest_status(cop0);
1461 
1462 		/*
1463 		 * If FR=0 FPU state is already live, it is undefined how it
1464 		 * interacts with MSA state, so play it safe and save it first.
1465 		 */
1466 		if (!(sr & ST0_FR) &&
1467 		    (vcpu->arch.aux_inuse & (KVM_MIPS_AUX_FPU |
1468 				KVM_MIPS_AUX_MSA)) == KVM_MIPS_AUX_FPU)
1469 			kvm_lose_fpu(vcpu);
1470 
1471 		change_c0_status(ST0_CU1 | ST0_FR, sr);
1472 		if (sr & ST0_CU1 && cpu_has_fre) {
1473 			cfg5 = kvm_read_c0_guest_config5(cop0);
1474 			change_c0_config5(MIPS_CONF5_FRE, cfg5);
1475 		}
1476 	}
1477 
1478 	/* Enable MSA for guest */
1479 	set_c0_config5(MIPS_CONF5_MSAEN);
1480 	enable_fpu_hazard();
1481 
1482 	switch (vcpu->arch.aux_inuse & (KVM_MIPS_AUX_FPU | KVM_MIPS_AUX_MSA)) {
1483 	case KVM_MIPS_AUX_FPU:
1484 		/*
1485 		 * Guest FPU state already loaded, only restore upper MSA state
1486 		 */
1487 		__kvm_restore_msa_upper(&vcpu->arch);
1488 		vcpu->arch.aux_inuse |= KVM_MIPS_AUX_MSA;
1489 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE, KVM_TRACE_AUX_MSA);
1490 		break;
1491 	case 0:
1492 		/* Neither FPU or MSA already active, restore full MSA state */
1493 		__kvm_restore_msa(&vcpu->arch);
1494 		vcpu->arch.aux_inuse |= KVM_MIPS_AUX_MSA;
1495 		if (kvm_mips_guest_has_fpu(&vcpu->arch))
1496 			vcpu->arch.aux_inuse |= KVM_MIPS_AUX_FPU;
1497 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE,
1498 			      KVM_TRACE_AUX_FPU_MSA);
1499 		break;
1500 	default:
1501 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_ENABLE, KVM_TRACE_AUX_MSA);
1502 		break;
1503 	}
1504 
1505 	preempt_enable();
1506 }
1507 #endif
1508 
1509 /* Drop FPU & MSA without saving it */
1510 void kvm_drop_fpu(struct kvm_vcpu *vcpu)
1511 {
1512 	preempt_disable();
1513 	if (cpu_has_msa && vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA) {
1514 		disable_msa();
1515 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_DISCARD, KVM_TRACE_AUX_MSA);
1516 		vcpu->arch.aux_inuse &= ~KVM_MIPS_AUX_MSA;
1517 	}
1518 	if (vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU) {
1519 		clear_c0_status(ST0_CU1 | ST0_FR);
1520 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_DISCARD, KVM_TRACE_AUX_FPU);
1521 		vcpu->arch.aux_inuse &= ~KVM_MIPS_AUX_FPU;
1522 	}
1523 	preempt_enable();
1524 }
1525 
1526 /* Save and disable FPU & MSA */
1527 void kvm_lose_fpu(struct kvm_vcpu *vcpu)
1528 {
1529 	/*
1530 	 * FPU & MSA get disabled in root context (hardware) when it is disabled
1531 	 * in guest context (software), but the register state in the hardware
1532 	 * may still be in use. This is why we explicitly re-enable the hardware
1533 	 * before saving.
1534 	 */
1535 
1536 	preempt_disable();
1537 	if (cpu_has_msa && vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA) {
1538 		set_c0_config5(MIPS_CONF5_MSAEN);
1539 		enable_fpu_hazard();
1540 
1541 		__kvm_save_msa(&vcpu->arch);
1542 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_SAVE, KVM_TRACE_AUX_FPU_MSA);
1543 
1544 		/* Disable MSA & FPU */
1545 		disable_msa();
1546 		if (vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU) {
1547 			clear_c0_status(ST0_CU1 | ST0_FR);
1548 			disable_fpu_hazard();
1549 		}
1550 		vcpu->arch.aux_inuse &= ~(KVM_MIPS_AUX_FPU | KVM_MIPS_AUX_MSA);
1551 	} else if (vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU) {
1552 		set_c0_status(ST0_CU1);
1553 		enable_fpu_hazard();
1554 
1555 		__kvm_save_fpu(&vcpu->arch);
1556 		vcpu->arch.aux_inuse &= ~KVM_MIPS_AUX_FPU;
1557 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_SAVE, KVM_TRACE_AUX_FPU);
1558 
1559 		/* Disable FPU */
1560 		clear_c0_status(ST0_CU1 | ST0_FR);
1561 		disable_fpu_hazard();
1562 	}
1563 	preempt_enable();
1564 }
1565 
1566 /*
1567  * Step over a specific ctc1 to FCSR and a specific ctcmsa to MSACSR which are
1568  * used to restore guest FCSR/MSACSR state and may trigger a "harmless" FP/MSAFP
1569  * exception if cause bits are set in the value being written.
1570  */
1571 static int kvm_mips_csr_die_notify(struct notifier_block *self,
1572 				   unsigned long cmd, void *ptr)
1573 {
1574 	struct die_args *args = (struct die_args *)ptr;
1575 	struct pt_regs *regs = args->regs;
1576 	unsigned long pc;
1577 
1578 	/* Only interested in FPE and MSAFPE */
1579 	if (cmd != DIE_FP && cmd != DIE_MSAFP)
1580 		return NOTIFY_DONE;
1581 
1582 	/* Return immediately if guest context isn't active */
1583 	if (!(current->flags & PF_VCPU))
1584 		return NOTIFY_DONE;
1585 
1586 	/* Should never get here from user mode */
1587 	BUG_ON(user_mode(regs));
1588 
1589 	pc = instruction_pointer(regs);
1590 	switch (cmd) {
1591 	case DIE_FP:
1592 		/* match 2nd instruction in __kvm_restore_fcsr */
1593 		if (pc != (unsigned long)&__kvm_restore_fcsr + 4)
1594 			return NOTIFY_DONE;
1595 		break;
1596 	case DIE_MSAFP:
1597 		/* match 2nd/3rd instruction in __kvm_restore_msacsr */
1598 		if (!cpu_has_msa ||
1599 		    pc < (unsigned long)&__kvm_restore_msacsr + 4 ||
1600 		    pc > (unsigned long)&__kvm_restore_msacsr + 8)
1601 			return NOTIFY_DONE;
1602 		break;
1603 	}
1604 
1605 	/* Move PC forward a little and continue executing */
1606 	instruction_pointer(regs) += 4;
1607 
1608 	return NOTIFY_STOP;
1609 }
1610 
1611 static struct notifier_block kvm_mips_csr_die_notifier = {
1612 	.notifier_call = kvm_mips_csr_die_notify,
1613 };
1614 
1615 static int __init kvm_mips_init(void)
1616 {
1617 	int ret;
1618 
1619 	ret = kvm_mips_entry_setup();
1620 	if (ret)
1621 		return ret;
1622 
1623 	ret = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1624 
1625 	if (ret)
1626 		return ret;
1627 
1628 	register_die_notifier(&kvm_mips_csr_die_notifier);
1629 
1630 	return 0;
1631 }
1632 
1633 static void __exit kvm_mips_exit(void)
1634 {
1635 	kvm_exit();
1636 
1637 	unregister_die_notifier(&kvm_mips_csr_die_notifier);
1638 }
1639 
1640 module_init(kvm_mips_init);
1641 module_exit(kvm_mips_exit);
1642 
1643 EXPORT_TRACEPOINT_SYMBOL(kvm_exit);
1644