xref: /openbmc/linux/arch/mips/kvm/mips.c (revision 3381df09)
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 
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_unusable", 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 #ifdef CONFIG_KVM_MIPS_VZ
63 	{ "vz_gpsi",	  VCPU_STAT(vz_gpsi_exits),	 KVM_STAT_VCPU },
64 	{ "vz_gsfc",	  VCPU_STAT(vz_gsfc_exits),	 KVM_STAT_VCPU },
65 	{ "vz_hc",	  VCPU_STAT(vz_hc_exits),	 KVM_STAT_VCPU },
66 	{ "vz_grr",	  VCPU_STAT(vz_grr_exits),	 KVM_STAT_VCPU },
67 	{ "vz_gva",	  VCPU_STAT(vz_gva_exits),	 KVM_STAT_VCPU },
68 	{ "vz_ghfc",	  VCPU_STAT(vz_ghfc_exits),	 KVM_STAT_VCPU },
69 	{ "vz_gpa",	  VCPU_STAT(vz_gpa_exits),	 KVM_STAT_VCPU },
70 	{ "vz_resvd",	  VCPU_STAT(vz_resvd_exits),	 KVM_STAT_VCPU },
71 #endif
72 	{ "halt_successful_poll", VCPU_STAT(halt_successful_poll), KVM_STAT_VCPU },
73 	{ "halt_attempted_poll", VCPU_STAT(halt_attempted_poll), KVM_STAT_VCPU },
74 	{ "halt_poll_invalid", VCPU_STAT(halt_poll_invalid), KVM_STAT_VCPU },
75 	{ "halt_wakeup",  VCPU_STAT(halt_wakeup),	 KVM_STAT_VCPU },
76 	{NULL}
77 };
78 
79 bool kvm_trace_guest_mode_change;
80 
81 int kvm_guest_mode_change_trace_reg(void)
82 {
83 	kvm_trace_guest_mode_change = 1;
84 	return 0;
85 }
86 
87 void kvm_guest_mode_change_trace_unreg(void)
88 {
89 	kvm_trace_guest_mode_change = 0;
90 }
91 
92 /*
93  * XXXKYMA: We are simulatoring a processor that has the WII bit set in
94  * Config7, so we are "runnable" if interrupts are pending
95  */
96 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
97 {
98 	return !!(vcpu->arch.pending_exceptions);
99 }
100 
101 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
102 {
103 	return false;
104 }
105 
106 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
107 {
108 	return 1;
109 }
110 
111 int kvm_arch_hardware_enable(void)
112 {
113 	return kvm_mips_callbacks->hardware_enable();
114 }
115 
116 void kvm_arch_hardware_disable(void)
117 {
118 	kvm_mips_callbacks->hardware_disable();
119 }
120 
121 int kvm_arch_hardware_setup(void *opaque)
122 {
123 	return 0;
124 }
125 
126 int kvm_arch_check_processor_compat(void *opaque)
127 {
128 	return 0;
129 }
130 
131 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
132 {
133 	switch (type) {
134 #ifdef CONFIG_KVM_MIPS_VZ
135 	case KVM_VM_MIPS_VZ:
136 #else
137 	case KVM_VM_MIPS_TE:
138 #endif
139 		break;
140 	default:
141 		/* Unsupported KVM type */
142 		return -EINVAL;
143 	};
144 
145 	/* Allocate page table to map GPA -> RPA */
146 	kvm->arch.gpa_mm.pgd = kvm_pgd_alloc();
147 	if (!kvm->arch.gpa_mm.pgd)
148 		return -ENOMEM;
149 
150 	return 0;
151 }
152 
153 void kvm_mips_free_vcpus(struct kvm *kvm)
154 {
155 	unsigned int i;
156 	struct kvm_vcpu *vcpu;
157 
158 	kvm_for_each_vcpu(i, vcpu, kvm) {
159 		kvm_vcpu_destroy(vcpu);
160 	}
161 
162 	mutex_lock(&kvm->lock);
163 
164 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
165 		kvm->vcpus[i] = NULL;
166 
167 	atomic_set(&kvm->online_vcpus, 0);
168 
169 	mutex_unlock(&kvm->lock);
170 }
171 
172 static void kvm_mips_free_gpa_pt(struct kvm *kvm)
173 {
174 	/* It should always be safe to remove after flushing the whole range */
175 	WARN_ON(!kvm_mips_flush_gpa_pt(kvm, 0, ~0));
176 	pgd_free(NULL, kvm->arch.gpa_mm.pgd);
177 }
178 
179 void kvm_arch_destroy_vm(struct kvm *kvm)
180 {
181 	kvm_mips_free_vcpus(kvm);
182 	kvm_mips_free_gpa_pt(kvm);
183 }
184 
185 long kvm_arch_dev_ioctl(struct file *filp, unsigned int ioctl,
186 			unsigned long arg)
187 {
188 	return -ENOIOCTLCMD;
189 }
190 
191 void kvm_arch_flush_shadow_all(struct kvm *kvm)
192 {
193 	/* Flush whole GPA */
194 	kvm_mips_flush_gpa_pt(kvm, 0, ~0);
195 
196 	/* Let implementation do the rest */
197 	kvm_mips_callbacks->flush_shadow_all(kvm);
198 }
199 
200 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
201 				   struct kvm_memory_slot *slot)
202 {
203 	/*
204 	 * The slot has been made invalid (ready for moving or deletion), so we
205 	 * need to ensure that it can no longer be accessed by any guest VCPUs.
206 	 */
207 
208 	spin_lock(&kvm->mmu_lock);
209 	/* Flush slot from GPA */
210 	kvm_mips_flush_gpa_pt(kvm, slot->base_gfn,
211 			      slot->base_gfn + slot->npages - 1);
212 	/* Let implementation do the rest */
213 	kvm_mips_callbacks->flush_shadow_memslot(kvm, slot);
214 	spin_unlock(&kvm->mmu_lock);
215 }
216 
217 int kvm_arch_prepare_memory_region(struct kvm *kvm,
218 				   struct kvm_memory_slot *memslot,
219 				   const struct kvm_userspace_memory_region *mem,
220 				   enum kvm_mr_change change)
221 {
222 	return 0;
223 }
224 
225 void kvm_arch_commit_memory_region(struct kvm *kvm,
226 				   const struct kvm_userspace_memory_region *mem,
227 				   struct kvm_memory_slot *old,
228 				   const struct kvm_memory_slot *new,
229 				   enum kvm_mr_change change)
230 {
231 	int needs_flush;
232 
233 	kvm_debug("%s: kvm: %p slot: %d, GPA: %llx, size: %llx, QVA: %llx\n",
234 		  __func__, kvm, mem->slot, mem->guest_phys_addr,
235 		  mem->memory_size, mem->userspace_addr);
236 
237 	/*
238 	 * If dirty page logging is enabled, write protect all pages in the slot
239 	 * ready for dirty logging.
240 	 *
241 	 * There is no need to do this in any of the following cases:
242 	 * CREATE:	No dirty mappings will already exist.
243 	 * MOVE/DELETE:	The old mappings will already have been cleaned up by
244 	 *		kvm_arch_flush_shadow_memslot()
245 	 */
246 	if (change == KVM_MR_FLAGS_ONLY &&
247 	    (!(old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
248 	     new->flags & KVM_MEM_LOG_DIRTY_PAGES)) {
249 		spin_lock(&kvm->mmu_lock);
250 		/* Write protect GPA page table entries */
251 		needs_flush = kvm_mips_mkclean_gpa_pt(kvm, new->base_gfn,
252 					new->base_gfn + new->npages - 1);
253 		/* Let implementation do the rest */
254 		if (needs_flush)
255 			kvm_mips_callbacks->flush_shadow_memslot(kvm, new);
256 		spin_unlock(&kvm->mmu_lock);
257 	}
258 }
259 
260 static inline void dump_handler(const char *symbol, void *start, void *end)
261 {
262 	u32 *p;
263 
264 	pr_debug("LEAF(%s)\n", symbol);
265 
266 	pr_debug("\t.set push\n");
267 	pr_debug("\t.set noreorder\n");
268 
269 	for (p = start; p < (u32 *)end; ++p)
270 		pr_debug("\t.word\t0x%08x\t\t# %p\n", *p, p);
271 
272 	pr_debug("\t.set\tpop\n");
273 
274 	pr_debug("\tEND(%s)\n", symbol);
275 }
276 
277 /* low level hrtimer wake routine */
278 static enum hrtimer_restart kvm_mips_comparecount_wakeup(struct hrtimer *timer)
279 {
280 	struct kvm_vcpu *vcpu;
281 
282 	vcpu = container_of(timer, struct kvm_vcpu, arch.comparecount_timer);
283 
284 	kvm_mips_callbacks->queue_timer_int(vcpu);
285 
286 	vcpu->arch.wait = 0;
287 	if (swq_has_sleeper(&vcpu->wq))
288 		swake_up_one(&vcpu->wq);
289 
290 	return kvm_mips_count_timeout(vcpu);
291 }
292 
293 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
294 {
295 	return 0;
296 }
297 
298 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
299 {
300 	int err, size;
301 	void *gebase, *p, *handler, *refill_start, *refill_end;
302 	int i;
303 
304 	kvm_debug("kvm @ %p: create cpu %d at %p\n",
305 		  vcpu->kvm, vcpu->vcpu_id, vcpu);
306 
307 	err = kvm_mips_callbacks->vcpu_init(vcpu);
308 	if (err)
309 		return err;
310 
311 	hrtimer_init(&vcpu->arch.comparecount_timer, CLOCK_MONOTONIC,
312 		     HRTIMER_MODE_REL);
313 	vcpu->arch.comparecount_timer.function = kvm_mips_comparecount_wakeup;
314 
315 	/*
316 	 * Allocate space for host mode exception handlers that handle
317 	 * guest mode exits
318 	 */
319 	if (cpu_has_veic || cpu_has_vint)
320 		size = 0x200 + VECTORSPACING * 64;
321 	else
322 		size = 0x4000;
323 
324 	gebase = kzalloc(ALIGN(size, PAGE_SIZE), GFP_KERNEL);
325 
326 	if (!gebase) {
327 		err = -ENOMEM;
328 		goto out_uninit_vcpu;
329 	}
330 	kvm_debug("Allocated %d bytes for KVM Exception Handlers @ %p\n",
331 		  ALIGN(size, PAGE_SIZE), gebase);
332 
333 	/*
334 	 * Check new ebase actually fits in CP0_EBase. The lack of a write gate
335 	 * limits us to the low 512MB of physical address space. If the memory
336 	 * we allocate is out of range, just give up now.
337 	 */
338 	if (!cpu_has_ebase_wg && virt_to_phys(gebase) >= 0x20000000) {
339 		kvm_err("CP0_EBase.WG required for guest exception base %pK\n",
340 			gebase);
341 		err = -ENOMEM;
342 		goto out_free_gebase;
343 	}
344 
345 	/* Save new ebase */
346 	vcpu->arch.guest_ebase = gebase;
347 
348 	/* Build guest exception vectors dynamically in unmapped memory */
349 	handler = gebase + 0x2000;
350 
351 	/* TLB refill (or XTLB refill on 64-bit VZ where KX=1) */
352 	refill_start = gebase;
353 	if (IS_ENABLED(CONFIG_KVM_MIPS_VZ) && IS_ENABLED(CONFIG_64BIT))
354 		refill_start += 0x080;
355 	refill_end = kvm_mips_build_tlb_refill_exception(refill_start, handler);
356 
357 	/* General Exception Entry point */
358 	kvm_mips_build_exception(gebase + 0x180, handler);
359 
360 	/* For vectored interrupts poke the exception code @ all offsets 0-7 */
361 	for (i = 0; i < 8; i++) {
362 		kvm_debug("L1 Vectored handler @ %p\n",
363 			  gebase + 0x200 + (i * VECTORSPACING));
364 		kvm_mips_build_exception(gebase + 0x200 + i * VECTORSPACING,
365 					 handler);
366 	}
367 
368 	/* General exit handler */
369 	p = handler;
370 	p = kvm_mips_build_exit(p);
371 
372 	/* Guest entry routine */
373 	vcpu->arch.vcpu_run = p;
374 	p = kvm_mips_build_vcpu_run(p);
375 
376 	/* Dump the generated code */
377 	pr_debug("#include <asm/asm.h>\n");
378 	pr_debug("#include <asm/regdef.h>\n");
379 	pr_debug("\n");
380 	dump_handler("kvm_vcpu_run", vcpu->arch.vcpu_run, p);
381 	dump_handler("kvm_tlb_refill", refill_start, refill_end);
382 	dump_handler("kvm_gen_exc", gebase + 0x180, gebase + 0x200);
383 	dump_handler("kvm_exit", gebase + 0x2000, vcpu->arch.vcpu_run);
384 
385 	/* Invalidate the icache for these ranges */
386 	flush_icache_range((unsigned long)gebase,
387 			   (unsigned long)gebase + ALIGN(size, PAGE_SIZE));
388 
389 	/*
390 	 * Allocate comm page for guest kernel, a TLB will be reserved for
391 	 * mapping GVA @ 0xFFFF8000 to this page
392 	 */
393 	vcpu->arch.kseg0_commpage = kzalloc(PAGE_SIZE << 1, GFP_KERNEL);
394 
395 	if (!vcpu->arch.kseg0_commpage) {
396 		err = -ENOMEM;
397 		goto out_free_gebase;
398 	}
399 
400 	kvm_debug("Allocated COMM page @ %p\n", vcpu->arch.kseg0_commpage);
401 	kvm_mips_commpage_init(vcpu);
402 
403 	/* Init */
404 	vcpu->arch.last_sched_cpu = -1;
405 	vcpu->arch.last_exec_cpu = -1;
406 
407 	/* Initial guest state */
408 	err = kvm_mips_callbacks->vcpu_setup(vcpu);
409 	if (err)
410 		goto out_free_commpage;
411 
412 	return 0;
413 
414 out_free_commpage:
415 	kfree(vcpu->arch.kseg0_commpage);
416 out_free_gebase:
417 	kfree(gebase);
418 out_uninit_vcpu:
419 	kvm_mips_callbacks->vcpu_uninit(vcpu);
420 	return err;
421 }
422 
423 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
424 {
425 	hrtimer_cancel(&vcpu->arch.comparecount_timer);
426 
427 	kvm_mips_dump_stats(vcpu);
428 
429 	kvm_mmu_free_memory_caches(vcpu);
430 	kfree(vcpu->arch.guest_ebase);
431 	kfree(vcpu->arch.kseg0_commpage);
432 
433 	kvm_mips_callbacks->vcpu_uninit(vcpu);
434 }
435 
436 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
437 					struct kvm_guest_debug *dbg)
438 {
439 	return -ENOIOCTLCMD;
440 }
441 
442 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
443 {
444 	int r = -EINTR;
445 
446 	vcpu_load(vcpu);
447 
448 	kvm_sigset_activate(vcpu);
449 
450 	if (vcpu->mmio_needed) {
451 		if (!vcpu->mmio_is_write)
452 			kvm_mips_complete_mmio_load(vcpu, run);
453 		vcpu->mmio_needed = 0;
454 	}
455 
456 	if (run->immediate_exit)
457 		goto out;
458 
459 	lose_fpu(1);
460 
461 	local_irq_disable();
462 	guest_enter_irqoff();
463 	trace_kvm_enter(vcpu);
464 
465 	/*
466 	 * Make sure the read of VCPU requests in vcpu_run() callback is not
467 	 * reordered ahead of the write to vcpu->mode, or we could miss a TLB
468 	 * flush request while the requester sees the VCPU as outside of guest
469 	 * mode and not needing an IPI.
470 	 */
471 	smp_store_mb(vcpu->mode, IN_GUEST_MODE);
472 
473 	r = kvm_mips_callbacks->vcpu_run(run, vcpu);
474 
475 	trace_kvm_out(vcpu);
476 	guest_exit_irqoff();
477 	local_irq_enable();
478 
479 out:
480 	kvm_sigset_deactivate(vcpu);
481 
482 	vcpu_put(vcpu);
483 	return r;
484 }
485 
486 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
487 			     struct kvm_mips_interrupt *irq)
488 {
489 	int intr = (int)irq->irq;
490 	struct kvm_vcpu *dvcpu = NULL;
491 
492 	if (intr == 3 || intr == -3 || intr == 4 || intr == -4)
493 		kvm_debug("%s: CPU: %d, INTR: %d\n", __func__, irq->cpu,
494 			  (int)intr);
495 
496 	if (irq->cpu == -1)
497 		dvcpu = vcpu;
498 	else
499 		dvcpu = vcpu->kvm->vcpus[irq->cpu];
500 
501 	if (intr == 2 || intr == 3 || intr == 4) {
502 		kvm_mips_callbacks->queue_io_int(dvcpu, irq);
503 
504 	} else if (intr == -2 || intr == -3 || intr == -4) {
505 		kvm_mips_callbacks->dequeue_io_int(dvcpu, irq);
506 	} else {
507 		kvm_err("%s: invalid interrupt ioctl (%d:%d)\n", __func__,
508 			irq->cpu, irq->irq);
509 		return -EINVAL;
510 	}
511 
512 	dvcpu->arch.wait = 0;
513 
514 	if (swq_has_sleeper(&dvcpu->wq))
515 		swake_up_one(&dvcpu->wq);
516 
517 	return 0;
518 }
519 
520 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
521 				    struct kvm_mp_state *mp_state)
522 {
523 	return -ENOIOCTLCMD;
524 }
525 
526 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
527 				    struct kvm_mp_state *mp_state)
528 {
529 	return -ENOIOCTLCMD;
530 }
531 
532 static u64 kvm_mips_get_one_regs[] = {
533 	KVM_REG_MIPS_R0,
534 	KVM_REG_MIPS_R1,
535 	KVM_REG_MIPS_R2,
536 	KVM_REG_MIPS_R3,
537 	KVM_REG_MIPS_R4,
538 	KVM_REG_MIPS_R5,
539 	KVM_REG_MIPS_R6,
540 	KVM_REG_MIPS_R7,
541 	KVM_REG_MIPS_R8,
542 	KVM_REG_MIPS_R9,
543 	KVM_REG_MIPS_R10,
544 	KVM_REG_MIPS_R11,
545 	KVM_REG_MIPS_R12,
546 	KVM_REG_MIPS_R13,
547 	KVM_REG_MIPS_R14,
548 	KVM_REG_MIPS_R15,
549 	KVM_REG_MIPS_R16,
550 	KVM_REG_MIPS_R17,
551 	KVM_REG_MIPS_R18,
552 	KVM_REG_MIPS_R19,
553 	KVM_REG_MIPS_R20,
554 	KVM_REG_MIPS_R21,
555 	KVM_REG_MIPS_R22,
556 	KVM_REG_MIPS_R23,
557 	KVM_REG_MIPS_R24,
558 	KVM_REG_MIPS_R25,
559 	KVM_REG_MIPS_R26,
560 	KVM_REG_MIPS_R27,
561 	KVM_REG_MIPS_R28,
562 	KVM_REG_MIPS_R29,
563 	KVM_REG_MIPS_R30,
564 	KVM_REG_MIPS_R31,
565 
566 #ifndef CONFIG_CPU_MIPSR6
567 	KVM_REG_MIPS_HI,
568 	KVM_REG_MIPS_LO,
569 #endif
570 	KVM_REG_MIPS_PC,
571 };
572 
573 static u64 kvm_mips_get_one_regs_fpu[] = {
574 	KVM_REG_MIPS_FCR_IR,
575 	KVM_REG_MIPS_FCR_CSR,
576 };
577 
578 static u64 kvm_mips_get_one_regs_msa[] = {
579 	KVM_REG_MIPS_MSA_IR,
580 	KVM_REG_MIPS_MSA_CSR,
581 };
582 
583 static unsigned long kvm_mips_num_regs(struct kvm_vcpu *vcpu)
584 {
585 	unsigned long ret;
586 
587 	ret = ARRAY_SIZE(kvm_mips_get_one_regs);
588 	if (kvm_mips_guest_can_have_fpu(&vcpu->arch)) {
589 		ret += ARRAY_SIZE(kvm_mips_get_one_regs_fpu) + 48;
590 		/* odd doubles */
591 		if (boot_cpu_data.fpu_id & MIPS_FPIR_F64)
592 			ret += 16;
593 	}
594 	if (kvm_mips_guest_can_have_msa(&vcpu->arch))
595 		ret += ARRAY_SIZE(kvm_mips_get_one_regs_msa) + 32;
596 	ret += kvm_mips_callbacks->num_regs(vcpu);
597 
598 	return ret;
599 }
600 
601 static int kvm_mips_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *indices)
602 {
603 	u64 index;
604 	unsigned int i;
605 
606 	if (copy_to_user(indices, kvm_mips_get_one_regs,
607 			 sizeof(kvm_mips_get_one_regs)))
608 		return -EFAULT;
609 	indices += ARRAY_SIZE(kvm_mips_get_one_regs);
610 
611 	if (kvm_mips_guest_can_have_fpu(&vcpu->arch)) {
612 		if (copy_to_user(indices, kvm_mips_get_one_regs_fpu,
613 				 sizeof(kvm_mips_get_one_regs_fpu)))
614 			return -EFAULT;
615 		indices += ARRAY_SIZE(kvm_mips_get_one_regs_fpu);
616 
617 		for (i = 0; i < 32; ++i) {
618 			index = KVM_REG_MIPS_FPR_32(i);
619 			if (copy_to_user(indices, &index, sizeof(index)))
620 				return -EFAULT;
621 			++indices;
622 
623 			/* skip odd doubles if no F64 */
624 			if (i & 1 && !(boot_cpu_data.fpu_id & MIPS_FPIR_F64))
625 				continue;
626 
627 			index = KVM_REG_MIPS_FPR_64(i);
628 			if (copy_to_user(indices, &index, sizeof(index)))
629 				return -EFAULT;
630 			++indices;
631 		}
632 	}
633 
634 	if (kvm_mips_guest_can_have_msa(&vcpu->arch)) {
635 		if (copy_to_user(indices, kvm_mips_get_one_regs_msa,
636 				 sizeof(kvm_mips_get_one_regs_msa)))
637 			return -EFAULT;
638 		indices += ARRAY_SIZE(kvm_mips_get_one_regs_msa);
639 
640 		for (i = 0; i < 32; ++i) {
641 			index = KVM_REG_MIPS_VEC_128(i);
642 			if (copy_to_user(indices, &index, sizeof(index)))
643 				return -EFAULT;
644 			++indices;
645 		}
646 	}
647 
648 	return kvm_mips_callbacks->copy_reg_indices(vcpu, indices);
649 }
650 
651 static int kvm_mips_get_reg(struct kvm_vcpu *vcpu,
652 			    const struct kvm_one_reg *reg)
653 {
654 	struct mips_coproc *cop0 = vcpu->arch.cop0;
655 	struct mips_fpu_struct *fpu = &vcpu->arch.fpu;
656 	int ret;
657 	s64 v;
658 	s64 vs[2];
659 	unsigned int idx;
660 
661 	switch (reg->id) {
662 	/* General purpose registers */
663 	case KVM_REG_MIPS_R0 ... KVM_REG_MIPS_R31:
664 		v = (long)vcpu->arch.gprs[reg->id - KVM_REG_MIPS_R0];
665 		break;
666 #ifndef CONFIG_CPU_MIPSR6
667 	case KVM_REG_MIPS_HI:
668 		v = (long)vcpu->arch.hi;
669 		break;
670 	case KVM_REG_MIPS_LO:
671 		v = (long)vcpu->arch.lo;
672 		break;
673 #endif
674 	case KVM_REG_MIPS_PC:
675 		v = (long)vcpu->arch.pc;
676 		break;
677 
678 	/* Floating point registers */
679 	case KVM_REG_MIPS_FPR_32(0) ... KVM_REG_MIPS_FPR_32(31):
680 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
681 			return -EINVAL;
682 		idx = reg->id - KVM_REG_MIPS_FPR_32(0);
683 		/* Odd singles in top of even double when FR=0 */
684 		if (kvm_read_c0_guest_status(cop0) & ST0_FR)
685 			v = get_fpr32(&fpu->fpr[idx], 0);
686 		else
687 			v = get_fpr32(&fpu->fpr[idx & ~1], idx & 1);
688 		break;
689 	case KVM_REG_MIPS_FPR_64(0) ... KVM_REG_MIPS_FPR_64(31):
690 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
691 			return -EINVAL;
692 		idx = reg->id - KVM_REG_MIPS_FPR_64(0);
693 		/* Can't access odd doubles in FR=0 mode */
694 		if (idx & 1 && !(kvm_read_c0_guest_status(cop0) & ST0_FR))
695 			return -EINVAL;
696 		v = get_fpr64(&fpu->fpr[idx], 0);
697 		break;
698 	case KVM_REG_MIPS_FCR_IR:
699 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
700 			return -EINVAL;
701 		v = boot_cpu_data.fpu_id;
702 		break;
703 	case KVM_REG_MIPS_FCR_CSR:
704 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
705 			return -EINVAL;
706 		v = fpu->fcr31;
707 		break;
708 
709 	/* MIPS SIMD Architecture (MSA) registers */
710 	case KVM_REG_MIPS_VEC_128(0) ... KVM_REG_MIPS_VEC_128(31):
711 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
712 			return -EINVAL;
713 		/* Can't access MSA registers in FR=0 mode */
714 		if (!(kvm_read_c0_guest_status(cop0) & ST0_FR))
715 			return -EINVAL;
716 		idx = reg->id - KVM_REG_MIPS_VEC_128(0);
717 #ifdef CONFIG_CPU_LITTLE_ENDIAN
718 		/* least significant byte first */
719 		vs[0] = get_fpr64(&fpu->fpr[idx], 0);
720 		vs[1] = get_fpr64(&fpu->fpr[idx], 1);
721 #else
722 		/* most significant byte first */
723 		vs[0] = get_fpr64(&fpu->fpr[idx], 1);
724 		vs[1] = get_fpr64(&fpu->fpr[idx], 0);
725 #endif
726 		break;
727 	case KVM_REG_MIPS_MSA_IR:
728 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
729 			return -EINVAL;
730 		v = boot_cpu_data.msa_id;
731 		break;
732 	case KVM_REG_MIPS_MSA_CSR:
733 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
734 			return -EINVAL;
735 		v = fpu->msacsr;
736 		break;
737 
738 	/* registers to be handled specially */
739 	default:
740 		ret = kvm_mips_callbacks->get_one_reg(vcpu, reg, &v);
741 		if (ret)
742 			return ret;
743 		break;
744 	}
745 	if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U64) {
746 		u64 __user *uaddr64 = (u64 __user *)(long)reg->addr;
747 
748 		return put_user(v, uaddr64);
749 	} else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U32) {
750 		u32 __user *uaddr32 = (u32 __user *)(long)reg->addr;
751 		u32 v32 = (u32)v;
752 
753 		return put_user(v32, uaddr32);
754 	} else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U128) {
755 		void __user *uaddr = (void __user *)(long)reg->addr;
756 
757 		return copy_to_user(uaddr, vs, 16) ? -EFAULT : 0;
758 	} else {
759 		return -EINVAL;
760 	}
761 }
762 
763 static int kvm_mips_set_reg(struct kvm_vcpu *vcpu,
764 			    const struct kvm_one_reg *reg)
765 {
766 	struct mips_coproc *cop0 = vcpu->arch.cop0;
767 	struct mips_fpu_struct *fpu = &vcpu->arch.fpu;
768 	s64 v;
769 	s64 vs[2];
770 	unsigned int idx;
771 
772 	if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U64) {
773 		u64 __user *uaddr64 = (u64 __user *)(long)reg->addr;
774 
775 		if (get_user(v, uaddr64) != 0)
776 			return -EFAULT;
777 	} else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U32) {
778 		u32 __user *uaddr32 = (u32 __user *)(long)reg->addr;
779 		s32 v32;
780 
781 		if (get_user(v32, uaddr32) != 0)
782 			return -EFAULT;
783 		v = (s64)v32;
784 	} else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U128) {
785 		void __user *uaddr = (void __user *)(long)reg->addr;
786 
787 		return copy_from_user(vs, uaddr, 16) ? -EFAULT : 0;
788 	} else {
789 		return -EINVAL;
790 	}
791 
792 	switch (reg->id) {
793 	/* General purpose registers */
794 	case KVM_REG_MIPS_R0:
795 		/* Silently ignore requests to set $0 */
796 		break;
797 	case KVM_REG_MIPS_R1 ... KVM_REG_MIPS_R31:
798 		vcpu->arch.gprs[reg->id - KVM_REG_MIPS_R0] = v;
799 		break;
800 #ifndef CONFIG_CPU_MIPSR6
801 	case KVM_REG_MIPS_HI:
802 		vcpu->arch.hi = v;
803 		break;
804 	case KVM_REG_MIPS_LO:
805 		vcpu->arch.lo = v;
806 		break;
807 #endif
808 	case KVM_REG_MIPS_PC:
809 		vcpu->arch.pc = v;
810 		break;
811 
812 	/* Floating point registers */
813 	case KVM_REG_MIPS_FPR_32(0) ... KVM_REG_MIPS_FPR_32(31):
814 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
815 			return -EINVAL;
816 		idx = reg->id - KVM_REG_MIPS_FPR_32(0);
817 		/* Odd singles in top of even double when FR=0 */
818 		if (kvm_read_c0_guest_status(cop0) & ST0_FR)
819 			set_fpr32(&fpu->fpr[idx], 0, v);
820 		else
821 			set_fpr32(&fpu->fpr[idx & ~1], idx & 1, v);
822 		break;
823 	case KVM_REG_MIPS_FPR_64(0) ... KVM_REG_MIPS_FPR_64(31):
824 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
825 			return -EINVAL;
826 		idx = reg->id - KVM_REG_MIPS_FPR_64(0);
827 		/* Can't access odd doubles in FR=0 mode */
828 		if (idx & 1 && !(kvm_read_c0_guest_status(cop0) & ST0_FR))
829 			return -EINVAL;
830 		set_fpr64(&fpu->fpr[idx], 0, v);
831 		break;
832 	case KVM_REG_MIPS_FCR_IR:
833 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
834 			return -EINVAL;
835 		/* Read-only */
836 		break;
837 	case KVM_REG_MIPS_FCR_CSR:
838 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
839 			return -EINVAL;
840 		fpu->fcr31 = v;
841 		break;
842 
843 	/* MIPS SIMD Architecture (MSA) registers */
844 	case KVM_REG_MIPS_VEC_128(0) ... KVM_REG_MIPS_VEC_128(31):
845 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
846 			return -EINVAL;
847 		idx = reg->id - KVM_REG_MIPS_VEC_128(0);
848 #ifdef CONFIG_CPU_LITTLE_ENDIAN
849 		/* least significant byte first */
850 		set_fpr64(&fpu->fpr[idx], 0, vs[0]);
851 		set_fpr64(&fpu->fpr[idx], 1, vs[1]);
852 #else
853 		/* most significant byte first */
854 		set_fpr64(&fpu->fpr[idx], 1, vs[0]);
855 		set_fpr64(&fpu->fpr[idx], 0, vs[1]);
856 #endif
857 		break;
858 	case KVM_REG_MIPS_MSA_IR:
859 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
860 			return -EINVAL;
861 		/* Read-only */
862 		break;
863 	case KVM_REG_MIPS_MSA_CSR:
864 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
865 			return -EINVAL;
866 		fpu->msacsr = v;
867 		break;
868 
869 	/* registers to be handled specially */
870 	default:
871 		return kvm_mips_callbacks->set_one_reg(vcpu, reg, v);
872 	}
873 	return 0;
874 }
875 
876 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
877 				     struct kvm_enable_cap *cap)
878 {
879 	int r = 0;
880 
881 	if (!kvm_vm_ioctl_check_extension(vcpu->kvm, cap->cap))
882 		return -EINVAL;
883 	if (cap->flags)
884 		return -EINVAL;
885 	if (cap->args[0])
886 		return -EINVAL;
887 
888 	switch (cap->cap) {
889 	case KVM_CAP_MIPS_FPU:
890 		vcpu->arch.fpu_enabled = true;
891 		break;
892 	case KVM_CAP_MIPS_MSA:
893 		vcpu->arch.msa_enabled = true;
894 		break;
895 	default:
896 		r = -EINVAL;
897 		break;
898 	}
899 
900 	return r;
901 }
902 
903 long kvm_arch_vcpu_async_ioctl(struct file *filp, unsigned int ioctl,
904 			       unsigned long arg)
905 {
906 	struct kvm_vcpu *vcpu = filp->private_data;
907 	void __user *argp = (void __user *)arg;
908 
909 	if (ioctl == KVM_INTERRUPT) {
910 		struct kvm_mips_interrupt irq;
911 
912 		if (copy_from_user(&irq, argp, sizeof(irq)))
913 			return -EFAULT;
914 		kvm_debug("[%d] %s: irq: %d\n", vcpu->vcpu_id, __func__,
915 			  irq.irq);
916 
917 		return kvm_vcpu_ioctl_interrupt(vcpu, &irq);
918 	}
919 
920 	return -ENOIOCTLCMD;
921 }
922 
923 long kvm_arch_vcpu_ioctl(struct file *filp, unsigned int ioctl,
924 			 unsigned long arg)
925 {
926 	struct kvm_vcpu *vcpu = filp->private_data;
927 	void __user *argp = (void __user *)arg;
928 	long r;
929 
930 	vcpu_load(vcpu);
931 
932 	switch (ioctl) {
933 	case KVM_SET_ONE_REG:
934 	case KVM_GET_ONE_REG: {
935 		struct kvm_one_reg reg;
936 
937 		r = -EFAULT;
938 		if (copy_from_user(&reg, argp, sizeof(reg)))
939 			break;
940 		if (ioctl == KVM_SET_ONE_REG)
941 			r = kvm_mips_set_reg(vcpu, &reg);
942 		else
943 			r = kvm_mips_get_reg(vcpu, &reg);
944 		break;
945 	}
946 	case KVM_GET_REG_LIST: {
947 		struct kvm_reg_list __user *user_list = argp;
948 		struct kvm_reg_list reg_list;
949 		unsigned n;
950 
951 		r = -EFAULT;
952 		if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
953 			break;
954 		n = reg_list.n;
955 		reg_list.n = kvm_mips_num_regs(vcpu);
956 		if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
957 			break;
958 		r = -E2BIG;
959 		if (n < reg_list.n)
960 			break;
961 		r = kvm_mips_copy_reg_indices(vcpu, user_list->reg);
962 		break;
963 	}
964 	case KVM_ENABLE_CAP: {
965 		struct kvm_enable_cap cap;
966 
967 		r = -EFAULT;
968 		if (copy_from_user(&cap, argp, sizeof(cap)))
969 			break;
970 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
971 		break;
972 	}
973 	default:
974 		r = -ENOIOCTLCMD;
975 	}
976 
977 	vcpu_put(vcpu);
978 	return r;
979 }
980 
981 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
982 {
983 
984 }
985 
986 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
987 					struct kvm_memory_slot *memslot)
988 {
989 	/* Let implementation handle TLB/GVA invalidation */
990 	kvm_mips_callbacks->flush_shadow_memslot(kvm, memslot);
991 }
992 
993 long kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
994 {
995 	long r;
996 
997 	switch (ioctl) {
998 	default:
999 		r = -ENOIOCTLCMD;
1000 	}
1001 
1002 	return r;
1003 }
1004 
1005 int kvm_arch_init(void *opaque)
1006 {
1007 	if (kvm_mips_callbacks) {
1008 		kvm_err("kvm: module already exists\n");
1009 		return -EEXIST;
1010 	}
1011 
1012 	return kvm_mips_emulation_init(&kvm_mips_callbacks);
1013 }
1014 
1015 void kvm_arch_exit(void)
1016 {
1017 	kvm_mips_callbacks = NULL;
1018 }
1019 
1020 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1021 				  struct kvm_sregs *sregs)
1022 {
1023 	return -ENOIOCTLCMD;
1024 }
1025 
1026 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1027 				  struct kvm_sregs *sregs)
1028 {
1029 	return -ENOIOCTLCMD;
1030 }
1031 
1032 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
1033 {
1034 }
1035 
1036 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1037 {
1038 	return -ENOIOCTLCMD;
1039 }
1040 
1041 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1042 {
1043 	return -ENOIOCTLCMD;
1044 }
1045 
1046 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
1047 {
1048 	return VM_FAULT_SIGBUS;
1049 }
1050 
1051 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
1052 {
1053 	int r;
1054 
1055 	switch (ext) {
1056 	case KVM_CAP_ONE_REG:
1057 	case KVM_CAP_ENABLE_CAP:
1058 	case KVM_CAP_READONLY_MEM:
1059 	case KVM_CAP_SYNC_MMU:
1060 	case KVM_CAP_IMMEDIATE_EXIT:
1061 		r = 1;
1062 		break;
1063 	case KVM_CAP_NR_VCPUS:
1064 		r = num_online_cpus();
1065 		break;
1066 	case KVM_CAP_MAX_VCPUS:
1067 		r = KVM_MAX_VCPUS;
1068 		break;
1069 	case KVM_CAP_MAX_VCPU_ID:
1070 		r = KVM_MAX_VCPU_ID;
1071 		break;
1072 	case KVM_CAP_MIPS_FPU:
1073 		/* We don't handle systems with inconsistent cpu_has_fpu */
1074 		r = !!raw_cpu_has_fpu;
1075 		break;
1076 	case KVM_CAP_MIPS_MSA:
1077 		/*
1078 		 * We don't support MSA vector partitioning yet:
1079 		 * 1) It would require explicit support which can't be tested
1080 		 *    yet due to lack of support in current hardware.
1081 		 * 2) It extends the state that would need to be saved/restored
1082 		 *    by e.g. QEMU for migration.
1083 		 *
1084 		 * When vector partitioning hardware becomes available, support
1085 		 * could be added by requiring a flag when enabling
1086 		 * KVM_CAP_MIPS_MSA capability to indicate that userland knows
1087 		 * to save/restore the appropriate extra state.
1088 		 */
1089 		r = cpu_has_msa && !(boot_cpu_data.msa_id & MSA_IR_WRPF);
1090 		break;
1091 	default:
1092 		r = kvm_mips_callbacks->check_extension(kvm, ext);
1093 		break;
1094 	}
1095 	return r;
1096 }
1097 
1098 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1099 {
1100 	return kvm_mips_pending_timer(vcpu) ||
1101 		kvm_read_c0_guest_cause(vcpu->arch.cop0) & C_TI;
1102 }
1103 
1104 int kvm_arch_vcpu_dump_regs(struct kvm_vcpu *vcpu)
1105 {
1106 	int i;
1107 	struct mips_coproc *cop0;
1108 
1109 	if (!vcpu)
1110 		return -1;
1111 
1112 	kvm_debug("VCPU Register Dump:\n");
1113 	kvm_debug("\tpc = 0x%08lx\n", vcpu->arch.pc);
1114 	kvm_debug("\texceptions: %08lx\n", vcpu->arch.pending_exceptions);
1115 
1116 	for (i = 0; i < 32; i += 4) {
1117 		kvm_debug("\tgpr%02d: %08lx %08lx %08lx %08lx\n", i,
1118 		       vcpu->arch.gprs[i],
1119 		       vcpu->arch.gprs[i + 1],
1120 		       vcpu->arch.gprs[i + 2], vcpu->arch.gprs[i + 3]);
1121 	}
1122 	kvm_debug("\thi: 0x%08lx\n", vcpu->arch.hi);
1123 	kvm_debug("\tlo: 0x%08lx\n", vcpu->arch.lo);
1124 
1125 	cop0 = vcpu->arch.cop0;
1126 	kvm_debug("\tStatus: 0x%08x, Cause: 0x%08x\n",
1127 		  kvm_read_c0_guest_status(cop0),
1128 		  kvm_read_c0_guest_cause(cop0));
1129 
1130 	kvm_debug("\tEPC: 0x%08lx\n", kvm_read_c0_guest_epc(cop0));
1131 
1132 	return 0;
1133 }
1134 
1135 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1136 {
1137 	int i;
1138 
1139 	vcpu_load(vcpu);
1140 
1141 	for (i = 1; i < ARRAY_SIZE(vcpu->arch.gprs); i++)
1142 		vcpu->arch.gprs[i] = regs->gpr[i];
1143 	vcpu->arch.gprs[0] = 0; /* zero is special, and cannot be set. */
1144 	vcpu->arch.hi = regs->hi;
1145 	vcpu->arch.lo = regs->lo;
1146 	vcpu->arch.pc = regs->pc;
1147 
1148 	vcpu_put(vcpu);
1149 	return 0;
1150 }
1151 
1152 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1153 {
1154 	int i;
1155 
1156 	vcpu_load(vcpu);
1157 
1158 	for (i = 0; i < ARRAY_SIZE(vcpu->arch.gprs); i++)
1159 		regs->gpr[i] = vcpu->arch.gprs[i];
1160 
1161 	regs->hi = vcpu->arch.hi;
1162 	regs->lo = vcpu->arch.lo;
1163 	regs->pc = vcpu->arch.pc;
1164 
1165 	vcpu_put(vcpu);
1166 	return 0;
1167 }
1168 
1169 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1170 				  struct kvm_translation *tr)
1171 {
1172 	return 0;
1173 }
1174 
1175 static void kvm_mips_set_c0_status(void)
1176 {
1177 	u32 status = read_c0_status();
1178 
1179 	if (cpu_has_dsp)
1180 		status |= (ST0_MX);
1181 
1182 	write_c0_status(status);
1183 	ehb();
1184 }
1185 
1186 /*
1187  * Return value is in the form (errcode<<2 | RESUME_FLAG_HOST | RESUME_FLAG_NV)
1188  */
1189 int kvm_mips_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu)
1190 {
1191 	u32 cause = vcpu->arch.host_cp0_cause;
1192 	u32 exccode = (cause >> CAUSEB_EXCCODE) & 0x1f;
1193 	u32 __user *opc = (u32 __user *) vcpu->arch.pc;
1194 	unsigned long badvaddr = vcpu->arch.host_cp0_badvaddr;
1195 	enum emulation_result er = EMULATE_DONE;
1196 	u32 inst;
1197 	int ret = RESUME_GUEST;
1198 
1199 	vcpu->mode = OUTSIDE_GUEST_MODE;
1200 
1201 	/* re-enable HTW before enabling interrupts */
1202 	if (!IS_ENABLED(CONFIG_KVM_MIPS_VZ))
1203 		htw_start();
1204 
1205 	/* Set a default exit reason */
1206 	run->exit_reason = KVM_EXIT_UNKNOWN;
1207 	run->ready_for_interrupt_injection = 1;
1208 
1209 	/*
1210 	 * Set the appropriate status bits based on host CPU features,
1211 	 * before we hit the scheduler
1212 	 */
1213 	kvm_mips_set_c0_status();
1214 
1215 	local_irq_enable();
1216 
1217 	kvm_debug("kvm_mips_handle_exit: cause: %#x, PC: %p, kvm_run: %p, kvm_vcpu: %p\n",
1218 			cause, opc, run, vcpu);
1219 	trace_kvm_exit(vcpu, exccode);
1220 
1221 	if (!IS_ENABLED(CONFIG_KVM_MIPS_VZ)) {
1222 		/*
1223 		 * Do a privilege check, if in UM most of these exit conditions
1224 		 * end up causing an exception to be delivered to the Guest
1225 		 * Kernel
1226 		 */
1227 		er = kvm_mips_check_privilege(cause, opc, run, vcpu);
1228 		if (er == EMULATE_PRIV_FAIL) {
1229 			goto skip_emul;
1230 		} else if (er == EMULATE_FAIL) {
1231 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1232 			ret = RESUME_HOST;
1233 			goto skip_emul;
1234 		}
1235 	}
1236 
1237 	switch (exccode) {
1238 	case EXCCODE_INT:
1239 		kvm_debug("[%d]EXCCODE_INT @ %p\n", vcpu->vcpu_id, opc);
1240 
1241 		++vcpu->stat.int_exits;
1242 
1243 		if (need_resched())
1244 			cond_resched();
1245 
1246 		ret = RESUME_GUEST;
1247 		break;
1248 
1249 	case EXCCODE_CPU:
1250 		kvm_debug("EXCCODE_CPU: @ PC: %p\n", opc);
1251 
1252 		++vcpu->stat.cop_unusable_exits;
1253 		ret = kvm_mips_callbacks->handle_cop_unusable(vcpu);
1254 		/* XXXKYMA: Might need to return to user space */
1255 		if (run->exit_reason == KVM_EXIT_IRQ_WINDOW_OPEN)
1256 			ret = RESUME_HOST;
1257 		break;
1258 
1259 	case EXCCODE_MOD:
1260 		++vcpu->stat.tlbmod_exits;
1261 		ret = kvm_mips_callbacks->handle_tlb_mod(vcpu);
1262 		break;
1263 
1264 	case EXCCODE_TLBS:
1265 		kvm_debug("TLB ST fault:  cause %#x, status %#x, PC: %p, BadVaddr: %#lx\n",
1266 			  cause, kvm_read_c0_guest_status(vcpu->arch.cop0), opc,
1267 			  badvaddr);
1268 
1269 		++vcpu->stat.tlbmiss_st_exits;
1270 		ret = kvm_mips_callbacks->handle_tlb_st_miss(vcpu);
1271 		break;
1272 
1273 	case EXCCODE_TLBL:
1274 		kvm_debug("TLB LD fault: cause %#x, PC: %p, BadVaddr: %#lx\n",
1275 			  cause, opc, badvaddr);
1276 
1277 		++vcpu->stat.tlbmiss_ld_exits;
1278 		ret = kvm_mips_callbacks->handle_tlb_ld_miss(vcpu);
1279 		break;
1280 
1281 	case EXCCODE_ADES:
1282 		++vcpu->stat.addrerr_st_exits;
1283 		ret = kvm_mips_callbacks->handle_addr_err_st(vcpu);
1284 		break;
1285 
1286 	case EXCCODE_ADEL:
1287 		++vcpu->stat.addrerr_ld_exits;
1288 		ret = kvm_mips_callbacks->handle_addr_err_ld(vcpu);
1289 		break;
1290 
1291 	case EXCCODE_SYS:
1292 		++vcpu->stat.syscall_exits;
1293 		ret = kvm_mips_callbacks->handle_syscall(vcpu);
1294 		break;
1295 
1296 	case EXCCODE_RI:
1297 		++vcpu->stat.resvd_inst_exits;
1298 		ret = kvm_mips_callbacks->handle_res_inst(vcpu);
1299 		break;
1300 
1301 	case EXCCODE_BP:
1302 		++vcpu->stat.break_inst_exits;
1303 		ret = kvm_mips_callbacks->handle_break(vcpu);
1304 		break;
1305 
1306 	case EXCCODE_TR:
1307 		++vcpu->stat.trap_inst_exits;
1308 		ret = kvm_mips_callbacks->handle_trap(vcpu);
1309 		break;
1310 
1311 	case EXCCODE_MSAFPE:
1312 		++vcpu->stat.msa_fpe_exits;
1313 		ret = kvm_mips_callbacks->handle_msa_fpe(vcpu);
1314 		break;
1315 
1316 	case EXCCODE_FPE:
1317 		++vcpu->stat.fpe_exits;
1318 		ret = kvm_mips_callbacks->handle_fpe(vcpu);
1319 		break;
1320 
1321 	case EXCCODE_MSADIS:
1322 		++vcpu->stat.msa_disabled_exits;
1323 		ret = kvm_mips_callbacks->handle_msa_disabled(vcpu);
1324 		break;
1325 
1326 	case EXCCODE_GE:
1327 		/* defer exit accounting to handler */
1328 		ret = kvm_mips_callbacks->handle_guest_exit(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: %#x\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 (ret == RESUME_GUEST)
1350 		kvm_vz_acquire_htimer(vcpu);
1351 
1352 	if (er == EMULATE_DONE && !(ret & RESUME_HOST))
1353 		kvm_mips_deliver_interrupts(vcpu, cause);
1354 
1355 	if (!(ret & RESUME_HOST)) {
1356 		/* Only check for signals if not already exiting to userspace */
1357 		if (signal_pending(current)) {
1358 			run->exit_reason = KVM_EXIT_INTR;
1359 			ret = (-EINTR << 2) | RESUME_HOST;
1360 			++vcpu->stat.signal_exits;
1361 			trace_kvm_exit(vcpu, KVM_TRACE_EXIT_SIGNAL);
1362 		}
1363 	}
1364 
1365 	if (ret == RESUME_GUEST) {
1366 		trace_kvm_reenter(vcpu);
1367 
1368 		/*
1369 		 * Make sure the read of VCPU requests in vcpu_reenter()
1370 		 * callback is not reordered ahead of the write to vcpu->mode,
1371 		 * or we could miss a TLB flush request while the requester sees
1372 		 * the VCPU as outside of guest mode and not needing an IPI.
1373 		 */
1374 		smp_store_mb(vcpu->mode, IN_GUEST_MODE);
1375 
1376 		kvm_mips_callbacks->vcpu_reenter(run, vcpu);
1377 
1378 		/*
1379 		 * If FPU / MSA are enabled (i.e. the guest's FPU / MSA context
1380 		 * is live), restore FCR31 / MSACSR.
1381 		 *
1382 		 * This should be before returning to the guest exception
1383 		 * vector, as it may well cause an [MSA] FP exception if there
1384 		 * are pending exception bits unmasked. (see
1385 		 * kvm_mips_csr_die_notifier() for how that is handled).
1386 		 */
1387 		if (kvm_mips_guest_has_fpu(&vcpu->arch) &&
1388 		    read_c0_status() & ST0_CU1)
1389 			__kvm_restore_fcsr(&vcpu->arch);
1390 
1391 		if (kvm_mips_guest_has_msa(&vcpu->arch) &&
1392 		    read_c0_config5() & MIPS_CONF5_MSAEN)
1393 			__kvm_restore_msacsr(&vcpu->arch);
1394 	}
1395 
1396 	/* Disable HTW before returning to guest or host */
1397 	if (!IS_ENABLED(CONFIG_KVM_MIPS_VZ))
1398 		htw_stop();
1399 
1400 	return ret;
1401 }
1402 
1403 /* Enable FPU for guest and restore context */
1404 void kvm_own_fpu(struct kvm_vcpu *vcpu)
1405 {
1406 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1407 	unsigned int sr, cfg5;
1408 
1409 	preempt_disable();
1410 
1411 	sr = kvm_read_c0_guest_status(cop0);
1412 
1413 	/*
1414 	 * If MSA state is already live, it is undefined how it interacts with
1415 	 * FR=0 FPU state, and we don't want to hit reserved instruction
1416 	 * exceptions trying to save the MSA state later when CU=1 && FR=1, so
1417 	 * play it safe and save it first.
1418 	 *
1419 	 * In theory we shouldn't ever hit this case since kvm_lose_fpu() should
1420 	 * get called when guest CU1 is set, however we can't trust the guest
1421 	 * not to clobber the status register directly via the commpage.
1422 	 */
1423 	if (cpu_has_msa && sr & ST0_CU1 && !(sr & ST0_FR) &&
1424 	    vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA)
1425 		kvm_lose_fpu(vcpu);
1426 
1427 	/*
1428 	 * Enable FPU for guest
1429 	 * We set FR and FRE according to guest context
1430 	 */
1431 	change_c0_status(ST0_CU1 | ST0_FR, sr);
1432 	if (cpu_has_fre) {
1433 		cfg5 = kvm_read_c0_guest_config5(cop0);
1434 		change_c0_config5(MIPS_CONF5_FRE, cfg5);
1435 	}
1436 	enable_fpu_hazard();
1437 
1438 	/* If guest FPU state not active, restore it now */
1439 	if (!(vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU)) {
1440 		__kvm_restore_fpu(&vcpu->arch);
1441 		vcpu->arch.aux_inuse |= KVM_MIPS_AUX_FPU;
1442 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE, KVM_TRACE_AUX_FPU);
1443 	} else {
1444 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_ENABLE, KVM_TRACE_AUX_FPU);
1445 	}
1446 
1447 	preempt_enable();
1448 }
1449 
1450 #ifdef CONFIG_CPU_HAS_MSA
1451 /* Enable MSA for guest and restore context */
1452 void kvm_own_msa(struct kvm_vcpu *vcpu)
1453 {
1454 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1455 	unsigned int sr, cfg5;
1456 
1457 	preempt_disable();
1458 
1459 	/*
1460 	 * Enable FPU if enabled in guest, since we're restoring FPU context
1461 	 * anyway. We set FR and FRE according to guest context.
1462 	 */
1463 	if (kvm_mips_guest_has_fpu(&vcpu->arch)) {
1464 		sr = kvm_read_c0_guest_status(cop0);
1465 
1466 		/*
1467 		 * If FR=0 FPU state is already live, it is undefined how it
1468 		 * interacts with MSA state, so play it safe and save it first.
1469 		 */
1470 		if (!(sr & ST0_FR) &&
1471 		    (vcpu->arch.aux_inuse & (KVM_MIPS_AUX_FPU |
1472 				KVM_MIPS_AUX_MSA)) == KVM_MIPS_AUX_FPU)
1473 			kvm_lose_fpu(vcpu);
1474 
1475 		change_c0_status(ST0_CU1 | ST0_FR, sr);
1476 		if (sr & ST0_CU1 && cpu_has_fre) {
1477 			cfg5 = kvm_read_c0_guest_config5(cop0);
1478 			change_c0_config5(MIPS_CONF5_FRE, cfg5);
1479 		}
1480 	}
1481 
1482 	/* Enable MSA for guest */
1483 	set_c0_config5(MIPS_CONF5_MSAEN);
1484 	enable_fpu_hazard();
1485 
1486 	switch (vcpu->arch.aux_inuse & (KVM_MIPS_AUX_FPU | KVM_MIPS_AUX_MSA)) {
1487 	case KVM_MIPS_AUX_FPU:
1488 		/*
1489 		 * Guest FPU state already loaded, only restore upper MSA state
1490 		 */
1491 		__kvm_restore_msa_upper(&vcpu->arch);
1492 		vcpu->arch.aux_inuse |= KVM_MIPS_AUX_MSA;
1493 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE, KVM_TRACE_AUX_MSA);
1494 		break;
1495 	case 0:
1496 		/* Neither FPU or MSA already active, restore full MSA state */
1497 		__kvm_restore_msa(&vcpu->arch);
1498 		vcpu->arch.aux_inuse |= KVM_MIPS_AUX_MSA;
1499 		if (kvm_mips_guest_has_fpu(&vcpu->arch))
1500 			vcpu->arch.aux_inuse |= KVM_MIPS_AUX_FPU;
1501 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE,
1502 			      KVM_TRACE_AUX_FPU_MSA);
1503 		break;
1504 	default:
1505 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_ENABLE, KVM_TRACE_AUX_MSA);
1506 		break;
1507 	}
1508 
1509 	preempt_enable();
1510 }
1511 #endif
1512 
1513 /* Drop FPU & MSA without saving it */
1514 void kvm_drop_fpu(struct kvm_vcpu *vcpu)
1515 {
1516 	preempt_disable();
1517 	if (cpu_has_msa && vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA) {
1518 		disable_msa();
1519 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_DISCARD, KVM_TRACE_AUX_MSA);
1520 		vcpu->arch.aux_inuse &= ~KVM_MIPS_AUX_MSA;
1521 	}
1522 	if (vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU) {
1523 		clear_c0_status(ST0_CU1 | ST0_FR);
1524 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_DISCARD, KVM_TRACE_AUX_FPU);
1525 		vcpu->arch.aux_inuse &= ~KVM_MIPS_AUX_FPU;
1526 	}
1527 	preempt_enable();
1528 }
1529 
1530 /* Save and disable FPU & MSA */
1531 void kvm_lose_fpu(struct kvm_vcpu *vcpu)
1532 {
1533 	/*
1534 	 * With T&E, FPU & MSA get disabled in root context (hardware) when it
1535 	 * is disabled in guest context (software), but the register state in
1536 	 * the hardware may still be in use.
1537 	 * This is why we explicitly re-enable the hardware before saving.
1538 	 */
1539 
1540 	preempt_disable();
1541 	if (cpu_has_msa && vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA) {
1542 		if (!IS_ENABLED(CONFIG_KVM_MIPS_VZ)) {
1543 			set_c0_config5(MIPS_CONF5_MSAEN);
1544 			enable_fpu_hazard();
1545 		}
1546 
1547 		__kvm_save_msa(&vcpu->arch);
1548 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_SAVE, KVM_TRACE_AUX_FPU_MSA);
1549 
1550 		/* Disable MSA & FPU */
1551 		disable_msa();
1552 		if (vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU) {
1553 			clear_c0_status(ST0_CU1 | ST0_FR);
1554 			disable_fpu_hazard();
1555 		}
1556 		vcpu->arch.aux_inuse &= ~(KVM_MIPS_AUX_FPU | KVM_MIPS_AUX_MSA);
1557 	} else if (vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU) {
1558 		if (!IS_ENABLED(CONFIG_KVM_MIPS_VZ)) {
1559 			set_c0_status(ST0_CU1);
1560 			enable_fpu_hazard();
1561 		}
1562 
1563 		__kvm_save_fpu(&vcpu->arch);
1564 		vcpu->arch.aux_inuse &= ~KVM_MIPS_AUX_FPU;
1565 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_SAVE, KVM_TRACE_AUX_FPU);
1566 
1567 		/* Disable FPU */
1568 		clear_c0_status(ST0_CU1 | ST0_FR);
1569 		disable_fpu_hazard();
1570 	}
1571 	preempt_enable();
1572 }
1573 
1574 /*
1575  * Step over a specific ctc1 to FCSR and a specific ctcmsa to MSACSR which are
1576  * used to restore guest FCSR/MSACSR state and may trigger a "harmless" FP/MSAFP
1577  * exception if cause bits are set in the value being written.
1578  */
1579 static int kvm_mips_csr_die_notify(struct notifier_block *self,
1580 				   unsigned long cmd, void *ptr)
1581 {
1582 	struct die_args *args = (struct die_args *)ptr;
1583 	struct pt_regs *regs = args->regs;
1584 	unsigned long pc;
1585 
1586 	/* Only interested in FPE and MSAFPE */
1587 	if (cmd != DIE_FP && cmd != DIE_MSAFP)
1588 		return NOTIFY_DONE;
1589 
1590 	/* Return immediately if guest context isn't active */
1591 	if (!(current->flags & PF_VCPU))
1592 		return NOTIFY_DONE;
1593 
1594 	/* Should never get here from user mode */
1595 	BUG_ON(user_mode(regs));
1596 
1597 	pc = instruction_pointer(regs);
1598 	switch (cmd) {
1599 	case DIE_FP:
1600 		/* match 2nd instruction in __kvm_restore_fcsr */
1601 		if (pc != (unsigned long)&__kvm_restore_fcsr + 4)
1602 			return NOTIFY_DONE;
1603 		break;
1604 	case DIE_MSAFP:
1605 		/* match 2nd/3rd instruction in __kvm_restore_msacsr */
1606 		if (!cpu_has_msa ||
1607 		    pc < (unsigned long)&__kvm_restore_msacsr + 4 ||
1608 		    pc > (unsigned long)&__kvm_restore_msacsr + 8)
1609 			return NOTIFY_DONE;
1610 		break;
1611 	}
1612 
1613 	/* Move PC forward a little and continue executing */
1614 	instruction_pointer(regs) += 4;
1615 
1616 	return NOTIFY_STOP;
1617 }
1618 
1619 static struct notifier_block kvm_mips_csr_die_notifier = {
1620 	.notifier_call = kvm_mips_csr_die_notify,
1621 };
1622 
1623 static int __init kvm_mips_init(void)
1624 {
1625 	int ret;
1626 
1627 	if (cpu_has_mmid) {
1628 		pr_warn("KVM does not yet support MMIDs. KVM Disabled\n");
1629 		return -EOPNOTSUPP;
1630 	}
1631 
1632 	ret = kvm_mips_entry_setup();
1633 	if (ret)
1634 		return ret;
1635 
1636 	ret = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1637 
1638 	if (ret)
1639 		return ret;
1640 
1641 	register_die_notifier(&kvm_mips_csr_die_notifier);
1642 
1643 	return 0;
1644 }
1645 
1646 static void __exit kvm_mips_exit(void)
1647 {
1648 	kvm_exit();
1649 
1650 	unregister_die_notifier(&kvm_mips_csr_die_notifier);
1651 }
1652 
1653 module_init(kvm_mips_init);
1654 module_exit(kvm_mips_exit);
1655 
1656 EXPORT_TRACEPOINT_SYMBOL(kvm_exit);
1657