xref: /openbmc/linux/arch/mips/kvm/mips.c (revision 884caada)
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)
122 {
123 	return 0;
124 }
125 
126 int kvm_arch_check_processor_compat(void)
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_arch_vcpu_free(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 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
192 			    unsigned long npages)
193 {
194 	return 0;
195 }
196 
197 void kvm_arch_flush_shadow_all(struct kvm *kvm)
198 {
199 	/* Flush whole GPA */
200 	kvm_mips_flush_gpa_pt(kvm, 0, ~0);
201 
202 	/* Let implementation do the rest */
203 	kvm_mips_callbacks->flush_shadow_all(kvm);
204 }
205 
206 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
207 				   struct kvm_memory_slot *slot)
208 {
209 	/*
210 	 * The slot has been made invalid (ready for moving or deletion), so we
211 	 * need to ensure that it can no longer be accessed by any guest VCPUs.
212 	 */
213 
214 	spin_lock(&kvm->mmu_lock);
215 	/* Flush slot from GPA */
216 	kvm_mips_flush_gpa_pt(kvm, slot->base_gfn,
217 			      slot->base_gfn + slot->npages - 1);
218 	/* Let implementation do the rest */
219 	kvm_mips_callbacks->flush_shadow_memslot(kvm, slot);
220 	spin_unlock(&kvm->mmu_lock);
221 }
222 
223 int kvm_arch_prepare_memory_region(struct kvm *kvm,
224 				   struct kvm_memory_slot *memslot,
225 				   const struct kvm_userspace_memory_region *mem,
226 				   enum kvm_mr_change change)
227 {
228 	return 0;
229 }
230 
231 void kvm_arch_commit_memory_region(struct kvm *kvm,
232 				   const struct kvm_userspace_memory_region *mem,
233 				   const struct kvm_memory_slot *old,
234 				   const struct kvm_memory_slot *new,
235 				   enum kvm_mr_change change)
236 {
237 	int needs_flush;
238 
239 	kvm_debug("%s: kvm: %p slot: %d, GPA: %llx, size: %llx, QVA: %llx\n",
240 		  __func__, kvm, mem->slot, mem->guest_phys_addr,
241 		  mem->memory_size, mem->userspace_addr);
242 
243 	/*
244 	 * If dirty page logging is enabled, write protect all pages in the slot
245 	 * ready for dirty logging.
246 	 *
247 	 * There is no need to do this in any of the following cases:
248 	 * CREATE:	No dirty mappings will already exist.
249 	 * MOVE/DELETE:	The old mappings will already have been cleaned up by
250 	 *		kvm_arch_flush_shadow_memslot()
251 	 */
252 	if (change == KVM_MR_FLAGS_ONLY &&
253 	    (!(old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
254 	     new->flags & KVM_MEM_LOG_DIRTY_PAGES)) {
255 		spin_lock(&kvm->mmu_lock);
256 		/* Write protect GPA page table entries */
257 		needs_flush = kvm_mips_mkclean_gpa_pt(kvm, new->base_gfn,
258 					new->base_gfn + new->npages - 1);
259 		/* Let implementation do the rest */
260 		if (needs_flush)
261 			kvm_mips_callbacks->flush_shadow_memslot(kvm, new);
262 		spin_unlock(&kvm->mmu_lock);
263 	}
264 }
265 
266 static inline void dump_handler(const char *symbol, void *start, void *end)
267 {
268 	u32 *p;
269 
270 	pr_debug("LEAF(%s)\n", symbol);
271 
272 	pr_debug("\t.set push\n");
273 	pr_debug("\t.set noreorder\n");
274 
275 	for (p = start; p < (u32 *)end; ++p)
276 		pr_debug("\t.word\t0x%08x\t\t# %p\n", *p, p);
277 
278 	pr_debug("\t.set\tpop\n");
279 
280 	pr_debug("\tEND(%s)\n", symbol);
281 }
282 
283 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
284 {
285 	int err, size;
286 	void *gebase, *p, *handler, *refill_start, *refill_end;
287 	int i;
288 
289 	struct kvm_vcpu *vcpu = kzalloc(sizeof(struct kvm_vcpu), GFP_KERNEL);
290 
291 	if (!vcpu) {
292 		err = -ENOMEM;
293 		goto out;
294 	}
295 
296 	err = kvm_vcpu_init(vcpu, kvm, id);
297 
298 	if (err)
299 		goto out_free_cpu;
300 
301 	kvm_debug("kvm @ %p: create cpu %d at %p\n", kvm, id, vcpu);
302 
303 	/*
304 	 * Allocate space for host mode exception handlers that handle
305 	 * guest mode exits
306 	 */
307 	if (cpu_has_veic || cpu_has_vint)
308 		size = 0x200 + VECTORSPACING * 64;
309 	else
310 		size = 0x4000;
311 
312 	gebase = kzalloc(ALIGN(size, PAGE_SIZE), GFP_KERNEL);
313 
314 	if (!gebase) {
315 		err = -ENOMEM;
316 		goto out_uninit_cpu;
317 	}
318 	kvm_debug("Allocated %d bytes for KVM Exception Handlers @ %p\n",
319 		  ALIGN(size, PAGE_SIZE), gebase);
320 
321 	/*
322 	 * Check new ebase actually fits in CP0_EBase. The lack of a write gate
323 	 * limits us to the low 512MB of physical address space. If the memory
324 	 * we allocate is out of range, just give up now.
325 	 */
326 	if (!cpu_has_ebase_wg && virt_to_phys(gebase) >= 0x20000000) {
327 		kvm_err("CP0_EBase.WG required for guest exception base %pK\n",
328 			gebase);
329 		err = -ENOMEM;
330 		goto out_free_gebase;
331 	}
332 
333 	/* Save new ebase */
334 	vcpu->arch.guest_ebase = gebase;
335 
336 	/* Build guest exception vectors dynamically in unmapped memory */
337 	handler = gebase + 0x2000;
338 
339 	/* TLB refill (or XTLB refill on 64-bit VZ where KX=1) */
340 	refill_start = gebase;
341 	if (IS_ENABLED(CONFIG_KVM_MIPS_VZ) && IS_ENABLED(CONFIG_64BIT))
342 		refill_start += 0x080;
343 	refill_end = kvm_mips_build_tlb_refill_exception(refill_start, handler);
344 
345 	/* General Exception Entry point */
346 	kvm_mips_build_exception(gebase + 0x180, handler);
347 
348 	/* For vectored interrupts poke the exception code @ all offsets 0-7 */
349 	for (i = 0; i < 8; i++) {
350 		kvm_debug("L1 Vectored handler @ %p\n",
351 			  gebase + 0x200 + (i * VECTORSPACING));
352 		kvm_mips_build_exception(gebase + 0x200 + i * VECTORSPACING,
353 					 handler);
354 	}
355 
356 	/* General exit handler */
357 	p = handler;
358 	p = kvm_mips_build_exit(p);
359 
360 	/* Guest entry routine */
361 	vcpu->arch.vcpu_run = p;
362 	p = kvm_mips_build_vcpu_run(p);
363 
364 	/* Dump the generated code */
365 	pr_debug("#include <asm/asm.h>\n");
366 	pr_debug("#include <asm/regdef.h>\n");
367 	pr_debug("\n");
368 	dump_handler("kvm_vcpu_run", vcpu->arch.vcpu_run, p);
369 	dump_handler("kvm_tlb_refill", refill_start, refill_end);
370 	dump_handler("kvm_gen_exc", gebase + 0x180, gebase + 0x200);
371 	dump_handler("kvm_exit", gebase + 0x2000, vcpu->arch.vcpu_run);
372 
373 	/* Invalidate the icache for these ranges */
374 	flush_icache_range((unsigned long)gebase,
375 			   (unsigned long)gebase + ALIGN(size, PAGE_SIZE));
376 
377 	/*
378 	 * Allocate comm page for guest kernel, a TLB will be reserved for
379 	 * mapping GVA @ 0xFFFF8000 to this page
380 	 */
381 	vcpu->arch.kseg0_commpage = kzalloc(PAGE_SIZE << 1, GFP_KERNEL);
382 
383 	if (!vcpu->arch.kseg0_commpage) {
384 		err = -ENOMEM;
385 		goto out_free_gebase;
386 	}
387 
388 	kvm_debug("Allocated COMM page @ %p\n", vcpu->arch.kseg0_commpage);
389 	kvm_mips_commpage_init(vcpu);
390 
391 	/* Init */
392 	vcpu->arch.last_sched_cpu = -1;
393 	vcpu->arch.last_exec_cpu = -1;
394 
395 	return vcpu;
396 
397 out_free_gebase:
398 	kfree(gebase);
399 
400 out_uninit_cpu:
401 	kvm_vcpu_uninit(vcpu);
402 
403 out_free_cpu:
404 	kfree(vcpu);
405 
406 out:
407 	return ERR_PTR(err);
408 }
409 
410 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
411 {
412 	hrtimer_cancel(&vcpu->arch.comparecount_timer);
413 
414 	kvm_vcpu_uninit(vcpu);
415 
416 	kvm_mips_dump_stats(vcpu);
417 
418 	kvm_mmu_free_memory_caches(vcpu);
419 	kfree(vcpu->arch.guest_ebase);
420 	kfree(vcpu->arch.kseg0_commpage);
421 	kfree(vcpu);
422 }
423 
424 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
425 {
426 	kvm_arch_vcpu_free(vcpu);
427 }
428 
429 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
430 					struct kvm_guest_debug *dbg)
431 {
432 	return -ENOIOCTLCMD;
433 }
434 
435 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
436 {
437 	int r = -EINTR;
438 
439 	vcpu_load(vcpu);
440 
441 	kvm_sigset_activate(vcpu);
442 
443 	if (vcpu->mmio_needed) {
444 		if (!vcpu->mmio_is_write)
445 			kvm_mips_complete_mmio_load(vcpu, run);
446 		vcpu->mmio_needed = 0;
447 	}
448 
449 	if (run->immediate_exit)
450 		goto out;
451 
452 	lose_fpu(1);
453 
454 	local_irq_disable();
455 	guest_enter_irqoff();
456 	trace_kvm_enter(vcpu);
457 
458 	/*
459 	 * Make sure the read of VCPU requests in vcpu_run() callback is not
460 	 * reordered ahead of the write to vcpu->mode, or we could miss a TLB
461 	 * flush request while the requester sees the VCPU as outside of guest
462 	 * mode and not needing an IPI.
463 	 */
464 	smp_store_mb(vcpu->mode, IN_GUEST_MODE);
465 
466 	r = kvm_mips_callbacks->vcpu_run(run, vcpu);
467 
468 	trace_kvm_out(vcpu);
469 	guest_exit_irqoff();
470 	local_irq_enable();
471 
472 out:
473 	kvm_sigset_deactivate(vcpu);
474 
475 	vcpu_put(vcpu);
476 	return r;
477 }
478 
479 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
480 			     struct kvm_mips_interrupt *irq)
481 {
482 	int intr = (int)irq->irq;
483 	struct kvm_vcpu *dvcpu = NULL;
484 
485 	if (intr == 3 || intr == -3 || intr == 4 || intr == -4)
486 		kvm_debug("%s: CPU: %d, INTR: %d\n", __func__, irq->cpu,
487 			  (int)intr);
488 
489 	if (irq->cpu == -1)
490 		dvcpu = vcpu;
491 	else
492 		dvcpu = vcpu->kvm->vcpus[irq->cpu];
493 
494 	if (intr == 2 || intr == 3 || intr == 4) {
495 		kvm_mips_callbacks->queue_io_int(dvcpu, irq);
496 
497 	} else if (intr == -2 || intr == -3 || intr == -4) {
498 		kvm_mips_callbacks->dequeue_io_int(dvcpu, irq);
499 	} else {
500 		kvm_err("%s: invalid interrupt ioctl (%d:%d)\n", __func__,
501 			irq->cpu, irq->irq);
502 		return -EINVAL;
503 	}
504 
505 	dvcpu->arch.wait = 0;
506 
507 	if (swq_has_sleeper(&dvcpu->wq))
508 		swake_up_one(&dvcpu->wq);
509 
510 	return 0;
511 }
512 
513 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
514 				    struct kvm_mp_state *mp_state)
515 {
516 	return -ENOIOCTLCMD;
517 }
518 
519 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
520 				    struct kvm_mp_state *mp_state)
521 {
522 	return -ENOIOCTLCMD;
523 }
524 
525 static u64 kvm_mips_get_one_regs[] = {
526 	KVM_REG_MIPS_R0,
527 	KVM_REG_MIPS_R1,
528 	KVM_REG_MIPS_R2,
529 	KVM_REG_MIPS_R3,
530 	KVM_REG_MIPS_R4,
531 	KVM_REG_MIPS_R5,
532 	KVM_REG_MIPS_R6,
533 	KVM_REG_MIPS_R7,
534 	KVM_REG_MIPS_R8,
535 	KVM_REG_MIPS_R9,
536 	KVM_REG_MIPS_R10,
537 	KVM_REG_MIPS_R11,
538 	KVM_REG_MIPS_R12,
539 	KVM_REG_MIPS_R13,
540 	KVM_REG_MIPS_R14,
541 	KVM_REG_MIPS_R15,
542 	KVM_REG_MIPS_R16,
543 	KVM_REG_MIPS_R17,
544 	KVM_REG_MIPS_R18,
545 	KVM_REG_MIPS_R19,
546 	KVM_REG_MIPS_R20,
547 	KVM_REG_MIPS_R21,
548 	KVM_REG_MIPS_R22,
549 	KVM_REG_MIPS_R23,
550 	KVM_REG_MIPS_R24,
551 	KVM_REG_MIPS_R25,
552 	KVM_REG_MIPS_R26,
553 	KVM_REG_MIPS_R27,
554 	KVM_REG_MIPS_R28,
555 	KVM_REG_MIPS_R29,
556 	KVM_REG_MIPS_R30,
557 	KVM_REG_MIPS_R31,
558 
559 #ifndef CONFIG_CPU_MIPSR6
560 	KVM_REG_MIPS_HI,
561 	KVM_REG_MIPS_LO,
562 #endif
563 	KVM_REG_MIPS_PC,
564 };
565 
566 static u64 kvm_mips_get_one_regs_fpu[] = {
567 	KVM_REG_MIPS_FCR_IR,
568 	KVM_REG_MIPS_FCR_CSR,
569 };
570 
571 static u64 kvm_mips_get_one_regs_msa[] = {
572 	KVM_REG_MIPS_MSA_IR,
573 	KVM_REG_MIPS_MSA_CSR,
574 };
575 
576 static unsigned long kvm_mips_num_regs(struct kvm_vcpu *vcpu)
577 {
578 	unsigned long ret;
579 
580 	ret = ARRAY_SIZE(kvm_mips_get_one_regs);
581 	if (kvm_mips_guest_can_have_fpu(&vcpu->arch)) {
582 		ret += ARRAY_SIZE(kvm_mips_get_one_regs_fpu) + 48;
583 		/* odd doubles */
584 		if (boot_cpu_data.fpu_id & MIPS_FPIR_F64)
585 			ret += 16;
586 	}
587 	if (kvm_mips_guest_can_have_msa(&vcpu->arch))
588 		ret += ARRAY_SIZE(kvm_mips_get_one_regs_msa) + 32;
589 	ret += kvm_mips_callbacks->num_regs(vcpu);
590 
591 	return ret;
592 }
593 
594 static int kvm_mips_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *indices)
595 {
596 	u64 index;
597 	unsigned int i;
598 
599 	if (copy_to_user(indices, kvm_mips_get_one_regs,
600 			 sizeof(kvm_mips_get_one_regs)))
601 		return -EFAULT;
602 	indices += ARRAY_SIZE(kvm_mips_get_one_regs);
603 
604 	if (kvm_mips_guest_can_have_fpu(&vcpu->arch)) {
605 		if (copy_to_user(indices, kvm_mips_get_one_regs_fpu,
606 				 sizeof(kvm_mips_get_one_regs_fpu)))
607 			return -EFAULT;
608 		indices += ARRAY_SIZE(kvm_mips_get_one_regs_fpu);
609 
610 		for (i = 0; i < 32; ++i) {
611 			index = KVM_REG_MIPS_FPR_32(i);
612 			if (copy_to_user(indices, &index, sizeof(index)))
613 				return -EFAULT;
614 			++indices;
615 
616 			/* skip odd doubles if no F64 */
617 			if (i & 1 && !(boot_cpu_data.fpu_id & MIPS_FPIR_F64))
618 				continue;
619 
620 			index = KVM_REG_MIPS_FPR_64(i);
621 			if (copy_to_user(indices, &index, sizeof(index)))
622 				return -EFAULT;
623 			++indices;
624 		}
625 	}
626 
627 	if (kvm_mips_guest_can_have_msa(&vcpu->arch)) {
628 		if (copy_to_user(indices, kvm_mips_get_one_regs_msa,
629 				 sizeof(kvm_mips_get_one_regs_msa)))
630 			return -EFAULT;
631 		indices += ARRAY_SIZE(kvm_mips_get_one_regs_msa);
632 
633 		for (i = 0; i < 32; ++i) {
634 			index = KVM_REG_MIPS_VEC_128(i);
635 			if (copy_to_user(indices, &index, sizeof(index)))
636 				return -EFAULT;
637 			++indices;
638 		}
639 	}
640 
641 	return kvm_mips_callbacks->copy_reg_indices(vcpu, indices);
642 }
643 
644 static int kvm_mips_get_reg(struct kvm_vcpu *vcpu,
645 			    const struct kvm_one_reg *reg)
646 {
647 	struct mips_coproc *cop0 = vcpu->arch.cop0;
648 	struct mips_fpu_struct *fpu = &vcpu->arch.fpu;
649 	int ret;
650 	s64 v;
651 	s64 vs[2];
652 	unsigned int idx;
653 
654 	switch (reg->id) {
655 	/* General purpose registers */
656 	case KVM_REG_MIPS_R0 ... KVM_REG_MIPS_R31:
657 		v = (long)vcpu->arch.gprs[reg->id - KVM_REG_MIPS_R0];
658 		break;
659 #ifndef CONFIG_CPU_MIPSR6
660 	case KVM_REG_MIPS_HI:
661 		v = (long)vcpu->arch.hi;
662 		break;
663 	case KVM_REG_MIPS_LO:
664 		v = (long)vcpu->arch.lo;
665 		break;
666 #endif
667 	case KVM_REG_MIPS_PC:
668 		v = (long)vcpu->arch.pc;
669 		break;
670 
671 	/* Floating point registers */
672 	case KVM_REG_MIPS_FPR_32(0) ... KVM_REG_MIPS_FPR_32(31):
673 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
674 			return -EINVAL;
675 		idx = reg->id - KVM_REG_MIPS_FPR_32(0);
676 		/* Odd singles in top of even double when FR=0 */
677 		if (kvm_read_c0_guest_status(cop0) & ST0_FR)
678 			v = get_fpr32(&fpu->fpr[idx], 0);
679 		else
680 			v = get_fpr32(&fpu->fpr[idx & ~1], idx & 1);
681 		break;
682 	case KVM_REG_MIPS_FPR_64(0) ... KVM_REG_MIPS_FPR_64(31):
683 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
684 			return -EINVAL;
685 		idx = reg->id - KVM_REG_MIPS_FPR_64(0);
686 		/* Can't access odd doubles in FR=0 mode */
687 		if (idx & 1 && !(kvm_read_c0_guest_status(cop0) & ST0_FR))
688 			return -EINVAL;
689 		v = get_fpr64(&fpu->fpr[idx], 0);
690 		break;
691 	case KVM_REG_MIPS_FCR_IR:
692 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
693 			return -EINVAL;
694 		v = boot_cpu_data.fpu_id;
695 		break;
696 	case KVM_REG_MIPS_FCR_CSR:
697 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
698 			return -EINVAL;
699 		v = fpu->fcr31;
700 		break;
701 
702 	/* MIPS SIMD Architecture (MSA) registers */
703 	case KVM_REG_MIPS_VEC_128(0) ... KVM_REG_MIPS_VEC_128(31):
704 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
705 			return -EINVAL;
706 		/* Can't access MSA registers in FR=0 mode */
707 		if (!(kvm_read_c0_guest_status(cop0) & ST0_FR))
708 			return -EINVAL;
709 		idx = reg->id - KVM_REG_MIPS_VEC_128(0);
710 #ifdef CONFIG_CPU_LITTLE_ENDIAN
711 		/* least significant byte first */
712 		vs[0] = get_fpr64(&fpu->fpr[idx], 0);
713 		vs[1] = get_fpr64(&fpu->fpr[idx], 1);
714 #else
715 		/* most significant byte first */
716 		vs[0] = get_fpr64(&fpu->fpr[idx], 1);
717 		vs[1] = get_fpr64(&fpu->fpr[idx], 0);
718 #endif
719 		break;
720 	case KVM_REG_MIPS_MSA_IR:
721 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
722 			return -EINVAL;
723 		v = boot_cpu_data.msa_id;
724 		break;
725 	case KVM_REG_MIPS_MSA_CSR:
726 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
727 			return -EINVAL;
728 		v = fpu->msacsr;
729 		break;
730 
731 	/* registers to be handled specially */
732 	default:
733 		ret = kvm_mips_callbacks->get_one_reg(vcpu, reg, &v);
734 		if (ret)
735 			return ret;
736 		break;
737 	}
738 	if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U64) {
739 		u64 __user *uaddr64 = (u64 __user *)(long)reg->addr;
740 
741 		return put_user(v, uaddr64);
742 	} else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U32) {
743 		u32 __user *uaddr32 = (u32 __user *)(long)reg->addr;
744 		u32 v32 = (u32)v;
745 
746 		return put_user(v32, uaddr32);
747 	} else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U128) {
748 		void __user *uaddr = (void __user *)(long)reg->addr;
749 
750 		return copy_to_user(uaddr, vs, 16) ? -EFAULT : 0;
751 	} else {
752 		return -EINVAL;
753 	}
754 }
755 
756 static int kvm_mips_set_reg(struct kvm_vcpu *vcpu,
757 			    const struct kvm_one_reg *reg)
758 {
759 	struct mips_coproc *cop0 = vcpu->arch.cop0;
760 	struct mips_fpu_struct *fpu = &vcpu->arch.fpu;
761 	s64 v;
762 	s64 vs[2];
763 	unsigned int idx;
764 
765 	if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U64) {
766 		u64 __user *uaddr64 = (u64 __user *)(long)reg->addr;
767 
768 		if (get_user(v, uaddr64) != 0)
769 			return -EFAULT;
770 	} else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U32) {
771 		u32 __user *uaddr32 = (u32 __user *)(long)reg->addr;
772 		s32 v32;
773 
774 		if (get_user(v32, uaddr32) != 0)
775 			return -EFAULT;
776 		v = (s64)v32;
777 	} else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U128) {
778 		void __user *uaddr = (void __user *)(long)reg->addr;
779 
780 		return copy_from_user(vs, uaddr, 16) ? -EFAULT : 0;
781 	} else {
782 		return -EINVAL;
783 	}
784 
785 	switch (reg->id) {
786 	/* General purpose registers */
787 	case KVM_REG_MIPS_R0:
788 		/* Silently ignore requests to set $0 */
789 		break;
790 	case KVM_REG_MIPS_R1 ... KVM_REG_MIPS_R31:
791 		vcpu->arch.gprs[reg->id - KVM_REG_MIPS_R0] = v;
792 		break;
793 #ifndef CONFIG_CPU_MIPSR6
794 	case KVM_REG_MIPS_HI:
795 		vcpu->arch.hi = v;
796 		break;
797 	case KVM_REG_MIPS_LO:
798 		vcpu->arch.lo = v;
799 		break;
800 #endif
801 	case KVM_REG_MIPS_PC:
802 		vcpu->arch.pc = v;
803 		break;
804 
805 	/* Floating point registers */
806 	case KVM_REG_MIPS_FPR_32(0) ... KVM_REG_MIPS_FPR_32(31):
807 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
808 			return -EINVAL;
809 		idx = reg->id - KVM_REG_MIPS_FPR_32(0);
810 		/* Odd singles in top of even double when FR=0 */
811 		if (kvm_read_c0_guest_status(cop0) & ST0_FR)
812 			set_fpr32(&fpu->fpr[idx], 0, v);
813 		else
814 			set_fpr32(&fpu->fpr[idx & ~1], idx & 1, v);
815 		break;
816 	case KVM_REG_MIPS_FPR_64(0) ... KVM_REG_MIPS_FPR_64(31):
817 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
818 			return -EINVAL;
819 		idx = reg->id - KVM_REG_MIPS_FPR_64(0);
820 		/* Can't access odd doubles in FR=0 mode */
821 		if (idx & 1 && !(kvm_read_c0_guest_status(cop0) & ST0_FR))
822 			return -EINVAL;
823 		set_fpr64(&fpu->fpr[idx], 0, v);
824 		break;
825 	case KVM_REG_MIPS_FCR_IR:
826 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
827 			return -EINVAL;
828 		/* Read-only */
829 		break;
830 	case KVM_REG_MIPS_FCR_CSR:
831 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
832 			return -EINVAL;
833 		fpu->fcr31 = v;
834 		break;
835 
836 	/* MIPS SIMD Architecture (MSA) registers */
837 	case KVM_REG_MIPS_VEC_128(0) ... KVM_REG_MIPS_VEC_128(31):
838 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
839 			return -EINVAL;
840 		idx = reg->id - KVM_REG_MIPS_VEC_128(0);
841 #ifdef CONFIG_CPU_LITTLE_ENDIAN
842 		/* least significant byte first */
843 		set_fpr64(&fpu->fpr[idx], 0, vs[0]);
844 		set_fpr64(&fpu->fpr[idx], 1, vs[1]);
845 #else
846 		/* most significant byte first */
847 		set_fpr64(&fpu->fpr[idx], 1, vs[0]);
848 		set_fpr64(&fpu->fpr[idx], 0, vs[1]);
849 #endif
850 		break;
851 	case KVM_REG_MIPS_MSA_IR:
852 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
853 			return -EINVAL;
854 		/* Read-only */
855 		break;
856 	case KVM_REG_MIPS_MSA_CSR:
857 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
858 			return -EINVAL;
859 		fpu->msacsr = v;
860 		break;
861 
862 	/* registers to be handled specially */
863 	default:
864 		return kvm_mips_callbacks->set_one_reg(vcpu, reg, v);
865 	}
866 	return 0;
867 }
868 
869 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
870 				     struct kvm_enable_cap *cap)
871 {
872 	int r = 0;
873 
874 	if (!kvm_vm_ioctl_check_extension(vcpu->kvm, cap->cap))
875 		return -EINVAL;
876 	if (cap->flags)
877 		return -EINVAL;
878 	if (cap->args[0])
879 		return -EINVAL;
880 
881 	switch (cap->cap) {
882 	case KVM_CAP_MIPS_FPU:
883 		vcpu->arch.fpu_enabled = true;
884 		break;
885 	case KVM_CAP_MIPS_MSA:
886 		vcpu->arch.msa_enabled = true;
887 		break;
888 	default:
889 		r = -EINVAL;
890 		break;
891 	}
892 
893 	return r;
894 }
895 
896 long kvm_arch_vcpu_async_ioctl(struct file *filp, unsigned int ioctl,
897 			       unsigned long arg)
898 {
899 	struct kvm_vcpu *vcpu = filp->private_data;
900 	void __user *argp = (void __user *)arg;
901 
902 	if (ioctl == KVM_INTERRUPT) {
903 		struct kvm_mips_interrupt irq;
904 
905 		if (copy_from_user(&irq, argp, sizeof(irq)))
906 			return -EFAULT;
907 		kvm_debug("[%d] %s: irq: %d\n", vcpu->vcpu_id, __func__,
908 			  irq.irq);
909 
910 		return kvm_vcpu_ioctl_interrupt(vcpu, &irq);
911 	}
912 
913 	return -ENOIOCTLCMD;
914 }
915 
916 long kvm_arch_vcpu_ioctl(struct file *filp, unsigned int ioctl,
917 			 unsigned long arg)
918 {
919 	struct kvm_vcpu *vcpu = filp->private_data;
920 	void __user *argp = (void __user *)arg;
921 	long r;
922 
923 	vcpu_load(vcpu);
924 
925 	switch (ioctl) {
926 	case KVM_SET_ONE_REG:
927 	case KVM_GET_ONE_REG: {
928 		struct kvm_one_reg reg;
929 
930 		r = -EFAULT;
931 		if (copy_from_user(&reg, argp, sizeof(reg)))
932 			break;
933 		if (ioctl == KVM_SET_ONE_REG)
934 			r = kvm_mips_set_reg(vcpu, &reg);
935 		else
936 			r = kvm_mips_get_reg(vcpu, &reg);
937 		break;
938 	}
939 	case KVM_GET_REG_LIST: {
940 		struct kvm_reg_list __user *user_list = argp;
941 		struct kvm_reg_list reg_list;
942 		unsigned n;
943 
944 		r = -EFAULT;
945 		if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
946 			break;
947 		n = reg_list.n;
948 		reg_list.n = kvm_mips_num_regs(vcpu);
949 		if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
950 			break;
951 		r = -E2BIG;
952 		if (n < reg_list.n)
953 			break;
954 		r = kvm_mips_copy_reg_indices(vcpu, user_list->reg);
955 		break;
956 	}
957 	case KVM_ENABLE_CAP: {
958 		struct kvm_enable_cap cap;
959 
960 		r = -EFAULT;
961 		if (copy_from_user(&cap, argp, sizeof(cap)))
962 			break;
963 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
964 		break;
965 	}
966 	default:
967 		r = -ENOIOCTLCMD;
968 	}
969 
970 	vcpu_put(vcpu);
971 	return r;
972 }
973 
974 /**
975  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
976  * @kvm: kvm instance
977  * @log: slot id and address to which we copy the log
978  *
979  * Steps 1-4 below provide general overview of dirty page logging. See
980  * kvm_get_dirty_log_protect() function description for additional details.
981  *
982  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
983  * always flush the TLB (step 4) even if previous step failed  and the dirty
984  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
985  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
986  * writes will be marked dirty for next log read.
987  *
988  *   1. Take a snapshot of the bit and clear it if needed.
989  *   2. Write protect the corresponding page.
990  *   3. Copy the snapshot to the userspace.
991  *   4. Flush TLB's if needed.
992  */
993 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
994 {
995 	struct kvm_memslots *slots;
996 	struct kvm_memory_slot *memslot;
997 	bool flush = false;
998 	int r;
999 
1000 	mutex_lock(&kvm->slots_lock);
1001 
1002 	r = kvm_get_dirty_log_protect(kvm, log, &flush);
1003 
1004 	if (flush) {
1005 		slots = kvm_memslots(kvm);
1006 		memslot = id_to_memslot(slots, log->slot);
1007 
1008 		/* Let implementation handle TLB/GVA invalidation */
1009 		kvm_mips_callbacks->flush_shadow_memslot(kvm, memslot);
1010 	}
1011 
1012 	mutex_unlock(&kvm->slots_lock);
1013 	return r;
1014 }
1015 
1016 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
1017 {
1018 	struct kvm_memslots *slots;
1019 	struct kvm_memory_slot *memslot;
1020 	bool flush = false;
1021 	int r;
1022 
1023 	mutex_lock(&kvm->slots_lock);
1024 
1025 	r = kvm_clear_dirty_log_protect(kvm, log, &flush);
1026 
1027 	if (flush) {
1028 		slots = kvm_memslots(kvm);
1029 		memslot = id_to_memslot(slots, log->slot);
1030 
1031 		/* Let implementation handle TLB/GVA invalidation */
1032 		kvm_mips_callbacks->flush_shadow_memslot(kvm, memslot);
1033 	}
1034 
1035 	mutex_unlock(&kvm->slots_lock);
1036 	return r;
1037 }
1038 
1039 long kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
1040 {
1041 	long r;
1042 
1043 	switch (ioctl) {
1044 	default:
1045 		r = -ENOIOCTLCMD;
1046 	}
1047 
1048 	return r;
1049 }
1050 
1051 int kvm_arch_init(void *opaque)
1052 {
1053 	if (kvm_mips_callbacks) {
1054 		kvm_err("kvm: module already exists\n");
1055 		return -EEXIST;
1056 	}
1057 
1058 	return kvm_mips_emulation_init(&kvm_mips_callbacks);
1059 }
1060 
1061 void kvm_arch_exit(void)
1062 {
1063 	kvm_mips_callbacks = NULL;
1064 }
1065 
1066 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1067 				  struct kvm_sregs *sregs)
1068 {
1069 	return -ENOIOCTLCMD;
1070 }
1071 
1072 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1073 				  struct kvm_sregs *sregs)
1074 {
1075 	return -ENOIOCTLCMD;
1076 }
1077 
1078 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
1079 {
1080 }
1081 
1082 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1083 {
1084 	return -ENOIOCTLCMD;
1085 }
1086 
1087 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1088 {
1089 	return -ENOIOCTLCMD;
1090 }
1091 
1092 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
1093 {
1094 	return VM_FAULT_SIGBUS;
1095 }
1096 
1097 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
1098 {
1099 	int r;
1100 
1101 	switch (ext) {
1102 	case KVM_CAP_ONE_REG:
1103 	case KVM_CAP_ENABLE_CAP:
1104 	case KVM_CAP_READONLY_MEM:
1105 	case KVM_CAP_SYNC_MMU:
1106 	case KVM_CAP_IMMEDIATE_EXIT:
1107 		r = 1;
1108 		break;
1109 	case KVM_CAP_NR_VCPUS:
1110 		r = num_online_cpus();
1111 		break;
1112 	case KVM_CAP_MAX_VCPUS:
1113 		r = KVM_MAX_VCPUS;
1114 		break;
1115 	case KVM_CAP_MAX_VCPU_ID:
1116 		r = KVM_MAX_VCPU_ID;
1117 		break;
1118 	case KVM_CAP_MIPS_FPU:
1119 		/* We don't handle systems with inconsistent cpu_has_fpu */
1120 		r = !!raw_cpu_has_fpu;
1121 		break;
1122 	case KVM_CAP_MIPS_MSA:
1123 		/*
1124 		 * We don't support MSA vector partitioning yet:
1125 		 * 1) It would require explicit support which can't be tested
1126 		 *    yet due to lack of support in current hardware.
1127 		 * 2) It extends the state that would need to be saved/restored
1128 		 *    by e.g. QEMU for migration.
1129 		 *
1130 		 * When vector partitioning hardware becomes available, support
1131 		 * could be added by requiring a flag when enabling
1132 		 * KVM_CAP_MIPS_MSA capability to indicate that userland knows
1133 		 * to save/restore the appropriate extra state.
1134 		 */
1135 		r = cpu_has_msa && !(boot_cpu_data.msa_id & MSA_IR_WRPF);
1136 		break;
1137 	default:
1138 		r = kvm_mips_callbacks->check_extension(kvm, ext);
1139 		break;
1140 	}
1141 	return r;
1142 }
1143 
1144 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1145 {
1146 	return kvm_mips_pending_timer(vcpu) ||
1147 		kvm_read_c0_guest_cause(vcpu->arch.cop0) & C_TI;
1148 }
1149 
1150 int kvm_arch_vcpu_dump_regs(struct kvm_vcpu *vcpu)
1151 {
1152 	int i;
1153 	struct mips_coproc *cop0;
1154 
1155 	if (!vcpu)
1156 		return -1;
1157 
1158 	kvm_debug("VCPU Register Dump:\n");
1159 	kvm_debug("\tpc = 0x%08lx\n", vcpu->arch.pc);
1160 	kvm_debug("\texceptions: %08lx\n", vcpu->arch.pending_exceptions);
1161 
1162 	for (i = 0; i < 32; i += 4) {
1163 		kvm_debug("\tgpr%02d: %08lx %08lx %08lx %08lx\n", i,
1164 		       vcpu->arch.gprs[i],
1165 		       vcpu->arch.gprs[i + 1],
1166 		       vcpu->arch.gprs[i + 2], vcpu->arch.gprs[i + 3]);
1167 	}
1168 	kvm_debug("\thi: 0x%08lx\n", vcpu->arch.hi);
1169 	kvm_debug("\tlo: 0x%08lx\n", vcpu->arch.lo);
1170 
1171 	cop0 = vcpu->arch.cop0;
1172 	kvm_debug("\tStatus: 0x%08x, Cause: 0x%08x\n",
1173 		  kvm_read_c0_guest_status(cop0),
1174 		  kvm_read_c0_guest_cause(cop0));
1175 
1176 	kvm_debug("\tEPC: 0x%08lx\n", kvm_read_c0_guest_epc(cop0));
1177 
1178 	return 0;
1179 }
1180 
1181 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1182 {
1183 	int i;
1184 
1185 	vcpu_load(vcpu);
1186 
1187 	for (i = 1; i < ARRAY_SIZE(vcpu->arch.gprs); i++)
1188 		vcpu->arch.gprs[i] = regs->gpr[i];
1189 	vcpu->arch.gprs[0] = 0; /* zero is special, and cannot be set. */
1190 	vcpu->arch.hi = regs->hi;
1191 	vcpu->arch.lo = regs->lo;
1192 	vcpu->arch.pc = regs->pc;
1193 
1194 	vcpu_put(vcpu);
1195 	return 0;
1196 }
1197 
1198 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1199 {
1200 	int i;
1201 
1202 	vcpu_load(vcpu);
1203 
1204 	for (i = 0; i < ARRAY_SIZE(vcpu->arch.gprs); i++)
1205 		regs->gpr[i] = vcpu->arch.gprs[i];
1206 
1207 	regs->hi = vcpu->arch.hi;
1208 	regs->lo = vcpu->arch.lo;
1209 	regs->pc = vcpu->arch.pc;
1210 
1211 	vcpu_put(vcpu);
1212 	return 0;
1213 }
1214 
1215 static void kvm_mips_comparecount_func(unsigned long data)
1216 {
1217 	struct kvm_vcpu *vcpu = (struct kvm_vcpu *)data;
1218 
1219 	kvm_mips_callbacks->queue_timer_int(vcpu);
1220 
1221 	vcpu->arch.wait = 0;
1222 	if (swq_has_sleeper(&vcpu->wq))
1223 		swake_up_one(&vcpu->wq);
1224 }
1225 
1226 /* low level hrtimer wake routine */
1227 static enum hrtimer_restart kvm_mips_comparecount_wakeup(struct hrtimer *timer)
1228 {
1229 	struct kvm_vcpu *vcpu;
1230 
1231 	vcpu = container_of(timer, struct kvm_vcpu, arch.comparecount_timer);
1232 	kvm_mips_comparecount_func((unsigned long) vcpu);
1233 	return kvm_mips_count_timeout(vcpu);
1234 }
1235 
1236 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
1237 {
1238 	int err;
1239 
1240 	err = kvm_mips_callbacks->vcpu_init(vcpu);
1241 	if (err)
1242 		return err;
1243 
1244 	hrtimer_init(&vcpu->arch.comparecount_timer, CLOCK_MONOTONIC,
1245 		     HRTIMER_MODE_REL);
1246 	vcpu->arch.comparecount_timer.function = kvm_mips_comparecount_wakeup;
1247 	return 0;
1248 }
1249 
1250 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
1251 {
1252 	kvm_mips_callbacks->vcpu_uninit(vcpu);
1253 }
1254 
1255 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1256 				  struct kvm_translation *tr)
1257 {
1258 	return 0;
1259 }
1260 
1261 /* Initial guest state */
1262 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
1263 {
1264 	return kvm_mips_callbacks->vcpu_setup(vcpu);
1265 }
1266 
1267 static void kvm_mips_set_c0_status(void)
1268 {
1269 	u32 status = read_c0_status();
1270 
1271 	if (cpu_has_dsp)
1272 		status |= (ST0_MX);
1273 
1274 	write_c0_status(status);
1275 	ehb();
1276 }
1277 
1278 /*
1279  * Return value is in the form (errcode<<2 | RESUME_FLAG_HOST | RESUME_FLAG_NV)
1280  */
1281 int kvm_mips_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu)
1282 {
1283 	u32 cause = vcpu->arch.host_cp0_cause;
1284 	u32 exccode = (cause >> CAUSEB_EXCCODE) & 0x1f;
1285 	u32 __user *opc = (u32 __user *) vcpu->arch.pc;
1286 	unsigned long badvaddr = vcpu->arch.host_cp0_badvaddr;
1287 	enum emulation_result er = EMULATE_DONE;
1288 	u32 inst;
1289 	int ret = RESUME_GUEST;
1290 
1291 	vcpu->mode = OUTSIDE_GUEST_MODE;
1292 
1293 	/* re-enable HTW before enabling interrupts */
1294 	if (!IS_ENABLED(CONFIG_KVM_MIPS_VZ))
1295 		htw_start();
1296 
1297 	/* Set a default exit reason */
1298 	run->exit_reason = KVM_EXIT_UNKNOWN;
1299 	run->ready_for_interrupt_injection = 1;
1300 
1301 	/*
1302 	 * Set the appropriate status bits based on host CPU features,
1303 	 * before we hit the scheduler
1304 	 */
1305 	kvm_mips_set_c0_status();
1306 
1307 	local_irq_enable();
1308 
1309 	kvm_debug("kvm_mips_handle_exit: cause: %#x, PC: %p, kvm_run: %p, kvm_vcpu: %p\n",
1310 			cause, opc, run, vcpu);
1311 	trace_kvm_exit(vcpu, exccode);
1312 
1313 	if (!IS_ENABLED(CONFIG_KVM_MIPS_VZ)) {
1314 		/*
1315 		 * Do a privilege check, if in UM most of these exit conditions
1316 		 * end up causing an exception to be delivered to the Guest
1317 		 * Kernel
1318 		 */
1319 		er = kvm_mips_check_privilege(cause, opc, run, vcpu);
1320 		if (er == EMULATE_PRIV_FAIL) {
1321 			goto skip_emul;
1322 		} else if (er == EMULATE_FAIL) {
1323 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1324 			ret = RESUME_HOST;
1325 			goto skip_emul;
1326 		}
1327 	}
1328 
1329 	switch (exccode) {
1330 	case EXCCODE_INT:
1331 		kvm_debug("[%d]EXCCODE_INT @ %p\n", vcpu->vcpu_id, opc);
1332 
1333 		++vcpu->stat.int_exits;
1334 
1335 		if (need_resched())
1336 			cond_resched();
1337 
1338 		ret = RESUME_GUEST;
1339 		break;
1340 
1341 	case EXCCODE_CPU:
1342 		kvm_debug("EXCCODE_CPU: @ PC: %p\n", opc);
1343 
1344 		++vcpu->stat.cop_unusable_exits;
1345 		ret = kvm_mips_callbacks->handle_cop_unusable(vcpu);
1346 		/* XXXKYMA: Might need to return to user space */
1347 		if (run->exit_reason == KVM_EXIT_IRQ_WINDOW_OPEN)
1348 			ret = RESUME_HOST;
1349 		break;
1350 
1351 	case EXCCODE_MOD:
1352 		++vcpu->stat.tlbmod_exits;
1353 		ret = kvm_mips_callbacks->handle_tlb_mod(vcpu);
1354 		break;
1355 
1356 	case EXCCODE_TLBS:
1357 		kvm_debug("TLB ST fault:  cause %#x, status %#x, PC: %p, BadVaddr: %#lx\n",
1358 			  cause, kvm_read_c0_guest_status(vcpu->arch.cop0), opc,
1359 			  badvaddr);
1360 
1361 		++vcpu->stat.tlbmiss_st_exits;
1362 		ret = kvm_mips_callbacks->handle_tlb_st_miss(vcpu);
1363 		break;
1364 
1365 	case EXCCODE_TLBL:
1366 		kvm_debug("TLB LD fault: cause %#x, PC: %p, BadVaddr: %#lx\n",
1367 			  cause, opc, badvaddr);
1368 
1369 		++vcpu->stat.tlbmiss_ld_exits;
1370 		ret = kvm_mips_callbacks->handle_tlb_ld_miss(vcpu);
1371 		break;
1372 
1373 	case EXCCODE_ADES:
1374 		++vcpu->stat.addrerr_st_exits;
1375 		ret = kvm_mips_callbacks->handle_addr_err_st(vcpu);
1376 		break;
1377 
1378 	case EXCCODE_ADEL:
1379 		++vcpu->stat.addrerr_ld_exits;
1380 		ret = kvm_mips_callbacks->handle_addr_err_ld(vcpu);
1381 		break;
1382 
1383 	case EXCCODE_SYS:
1384 		++vcpu->stat.syscall_exits;
1385 		ret = kvm_mips_callbacks->handle_syscall(vcpu);
1386 		break;
1387 
1388 	case EXCCODE_RI:
1389 		++vcpu->stat.resvd_inst_exits;
1390 		ret = kvm_mips_callbacks->handle_res_inst(vcpu);
1391 		break;
1392 
1393 	case EXCCODE_BP:
1394 		++vcpu->stat.break_inst_exits;
1395 		ret = kvm_mips_callbacks->handle_break(vcpu);
1396 		break;
1397 
1398 	case EXCCODE_TR:
1399 		++vcpu->stat.trap_inst_exits;
1400 		ret = kvm_mips_callbacks->handle_trap(vcpu);
1401 		break;
1402 
1403 	case EXCCODE_MSAFPE:
1404 		++vcpu->stat.msa_fpe_exits;
1405 		ret = kvm_mips_callbacks->handle_msa_fpe(vcpu);
1406 		break;
1407 
1408 	case EXCCODE_FPE:
1409 		++vcpu->stat.fpe_exits;
1410 		ret = kvm_mips_callbacks->handle_fpe(vcpu);
1411 		break;
1412 
1413 	case EXCCODE_MSADIS:
1414 		++vcpu->stat.msa_disabled_exits;
1415 		ret = kvm_mips_callbacks->handle_msa_disabled(vcpu);
1416 		break;
1417 
1418 	case EXCCODE_GE:
1419 		/* defer exit accounting to handler */
1420 		ret = kvm_mips_callbacks->handle_guest_exit(vcpu);
1421 		break;
1422 
1423 	default:
1424 		if (cause & CAUSEF_BD)
1425 			opc += 1;
1426 		inst = 0;
1427 		kvm_get_badinstr(opc, vcpu, &inst);
1428 		kvm_err("Exception Code: %d, not yet handled, @ PC: %p, inst: 0x%08x  BadVaddr: %#lx Status: %#x\n",
1429 			exccode, opc, inst, badvaddr,
1430 			kvm_read_c0_guest_status(vcpu->arch.cop0));
1431 		kvm_arch_vcpu_dump_regs(vcpu);
1432 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1433 		ret = RESUME_HOST;
1434 		break;
1435 
1436 	}
1437 
1438 skip_emul:
1439 	local_irq_disable();
1440 
1441 	if (ret == RESUME_GUEST)
1442 		kvm_vz_acquire_htimer(vcpu);
1443 
1444 	if (er == EMULATE_DONE && !(ret & RESUME_HOST))
1445 		kvm_mips_deliver_interrupts(vcpu, cause);
1446 
1447 	if (!(ret & RESUME_HOST)) {
1448 		/* Only check for signals if not already exiting to userspace */
1449 		if (signal_pending(current)) {
1450 			run->exit_reason = KVM_EXIT_INTR;
1451 			ret = (-EINTR << 2) | RESUME_HOST;
1452 			++vcpu->stat.signal_exits;
1453 			trace_kvm_exit(vcpu, KVM_TRACE_EXIT_SIGNAL);
1454 		}
1455 	}
1456 
1457 	if (ret == RESUME_GUEST) {
1458 		trace_kvm_reenter(vcpu);
1459 
1460 		/*
1461 		 * Make sure the read of VCPU requests in vcpu_reenter()
1462 		 * callback is not reordered ahead of the write to vcpu->mode,
1463 		 * or we could miss a TLB flush request while the requester sees
1464 		 * the VCPU as outside of guest mode and not needing an IPI.
1465 		 */
1466 		smp_store_mb(vcpu->mode, IN_GUEST_MODE);
1467 
1468 		kvm_mips_callbacks->vcpu_reenter(run, vcpu);
1469 
1470 		/*
1471 		 * If FPU / MSA are enabled (i.e. the guest's FPU / MSA context
1472 		 * is live), restore FCR31 / MSACSR.
1473 		 *
1474 		 * This should be before returning to the guest exception
1475 		 * vector, as it may well cause an [MSA] FP exception if there
1476 		 * are pending exception bits unmasked. (see
1477 		 * kvm_mips_csr_die_notifier() for how that is handled).
1478 		 */
1479 		if (kvm_mips_guest_has_fpu(&vcpu->arch) &&
1480 		    read_c0_status() & ST0_CU1)
1481 			__kvm_restore_fcsr(&vcpu->arch);
1482 
1483 		if (kvm_mips_guest_has_msa(&vcpu->arch) &&
1484 		    read_c0_config5() & MIPS_CONF5_MSAEN)
1485 			__kvm_restore_msacsr(&vcpu->arch);
1486 	}
1487 
1488 	/* Disable HTW before returning to guest or host */
1489 	if (!IS_ENABLED(CONFIG_KVM_MIPS_VZ))
1490 		htw_stop();
1491 
1492 	return ret;
1493 }
1494 
1495 /* Enable FPU for guest and restore context */
1496 void kvm_own_fpu(struct kvm_vcpu *vcpu)
1497 {
1498 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1499 	unsigned int sr, cfg5;
1500 
1501 	preempt_disable();
1502 
1503 	sr = kvm_read_c0_guest_status(cop0);
1504 
1505 	/*
1506 	 * If MSA state is already live, it is undefined how it interacts with
1507 	 * FR=0 FPU state, and we don't want to hit reserved instruction
1508 	 * exceptions trying to save the MSA state later when CU=1 && FR=1, so
1509 	 * play it safe and save it first.
1510 	 *
1511 	 * In theory we shouldn't ever hit this case since kvm_lose_fpu() should
1512 	 * get called when guest CU1 is set, however we can't trust the guest
1513 	 * not to clobber the status register directly via the commpage.
1514 	 */
1515 	if (cpu_has_msa && sr & ST0_CU1 && !(sr & ST0_FR) &&
1516 	    vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA)
1517 		kvm_lose_fpu(vcpu);
1518 
1519 	/*
1520 	 * Enable FPU for guest
1521 	 * We set FR and FRE according to guest context
1522 	 */
1523 	change_c0_status(ST0_CU1 | ST0_FR, sr);
1524 	if (cpu_has_fre) {
1525 		cfg5 = kvm_read_c0_guest_config5(cop0);
1526 		change_c0_config5(MIPS_CONF5_FRE, cfg5);
1527 	}
1528 	enable_fpu_hazard();
1529 
1530 	/* If guest FPU state not active, restore it now */
1531 	if (!(vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU)) {
1532 		__kvm_restore_fpu(&vcpu->arch);
1533 		vcpu->arch.aux_inuse |= KVM_MIPS_AUX_FPU;
1534 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE, KVM_TRACE_AUX_FPU);
1535 	} else {
1536 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_ENABLE, KVM_TRACE_AUX_FPU);
1537 	}
1538 
1539 	preempt_enable();
1540 }
1541 
1542 #ifdef CONFIG_CPU_HAS_MSA
1543 /* Enable MSA for guest and restore context */
1544 void kvm_own_msa(struct kvm_vcpu *vcpu)
1545 {
1546 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1547 	unsigned int sr, cfg5;
1548 
1549 	preempt_disable();
1550 
1551 	/*
1552 	 * Enable FPU if enabled in guest, since we're restoring FPU context
1553 	 * anyway. We set FR and FRE according to guest context.
1554 	 */
1555 	if (kvm_mips_guest_has_fpu(&vcpu->arch)) {
1556 		sr = kvm_read_c0_guest_status(cop0);
1557 
1558 		/*
1559 		 * If FR=0 FPU state is already live, it is undefined how it
1560 		 * interacts with MSA state, so play it safe and save it first.
1561 		 */
1562 		if (!(sr & ST0_FR) &&
1563 		    (vcpu->arch.aux_inuse & (KVM_MIPS_AUX_FPU |
1564 				KVM_MIPS_AUX_MSA)) == KVM_MIPS_AUX_FPU)
1565 			kvm_lose_fpu(vcpu);
1566 
1567 		change_c0_status(ST0_CU1 | ST0_FR, sr);
1568 		if (sr & ST0_CU1 && cpu_has_fre) {
1569 			cfg5 = kvm_read_c0_guest_config5(cop0);
1570 			change_c0_config5(MIPS_CONF5_FRE, cfg5);
1571 		}
1572 	}
1573 
1574 	/* Enable MSA for guest */
1575 	set_c0_config5(MIPS_CONF5_MSAEN);
1576 	enable_fpu_hazard();
1577 
1578 	switch (vcpu->arch.aux_inuse & (KVM_MIPS_AUX_FPU | KVM_MIPS_AUX_MSA)) {
1579 	case KVM_MIPS_AUX_FPU:
1580 		/*
1581 		 * Guest FPU state already loaded, only restore upper MSA state
1582 		 */
1583 		__kvm_restore_msa_upper(&vcpu->arch);
1584 		vcpu->arch.aux_inuse |= KVM_MIPS_AUX_MSA;
1585 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE, KVM_TRACE_AUX_MSA);
1586 		break;
1587 	case 0:
1588 		/* Neither FPU or MSA already active, restore full MSA state */
1589 		__kvm_restore_msa(&vcpu->arch);
1590 		vcpu->arch.aux_inuse |= KVM_MIPS_AUX_MSA;
1591 		if (kvm_mips_guest_has_fpu(&vcpu->arch))
1592 			vcpu->arch.aux_inuse |= KVM_MIPS_AUX_FPU;
1593 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE,
1594 			      KVM_TRACE_AUX_FPU_MSA);
1595 		break;
1596 	default:
1597 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_ENABLE, KVM_TRACE_AUX_MSA);
1598 		break;
1599 	}
1600 
1601 	preempt_enable();
1602 }
1603 #endif
1604 
1605 /* Drop FPU & MSA without saving it */
1606 void kvm_drop_fpu(struct kvm_vcpu *vcpu)
1607 {
1608 	preempt_disable();
1609 	if (cpu_has_msa && vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA) {
1610 		disable_msa();
1611 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_DISCARD, KVM_TRACE_AUX_MSA);
1612 		vcpu->arch.aux_inuse &= ~KVM_MIPS_AUX_MSA;
1613 	}
1614 	if (vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU) {
1615 		clear_c0_status(ST0_CU1 | ST0_FR);
1616 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_DISCARD, KVM_TRACE_AUX_FPU);
1617 		vcpu->arch.aux_inuse &= ~KVM_MIPS_AUX_FPU;
1618 	}
1619 	preempt_enable();
1620 }
1621 
1622 /* Save and disable FPU & MSA */
1623 void kvm_lose_fpu(struct kvm_vcpu *vcpu)
1624 {
1625 	/*
1626 	 * With T&E, FPU & MSA get disabled in root context (hardware) when it
1627 	 * is disabled in guest context (software), but the register state in
1628 	 * the hardware may still be in use.
1629 	 * This is why we explicitly re-enable the hardware before saving.
1630 	 */
1631 
1632 	preempt_disable();
1633 	if (cpu_has_msa && vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA) {
1634 		if (!IS_ENABLED(CONFIG_KVM_MIPS_VZ)) {
1635 			set_c0_config5(MIPS_CONF5_MSAEN);
1636 			enable_fpu_hazard();
1637 		}
1638 
1639 		__kvm_save_msa(&vcpu->arch);
1640 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_SAVE, KVM_TRACE_AUX_FPU_MSA);
1641 
1642 		/* Disable MSA & FPU */
1643 		disable_msa();
1644 		if (vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU) {
1645 			clear_c0_status(ST0_CU1 | ST0_FR);
1646 			disable_fpu_hazard();
1647 		}
1648 		vcpu->arch.aux_inuse &= ~(KVM_MIPS_AUX_FPU | KVM_MIPS_AUX_MSA);
1649 	} else if (vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU) {
1650 		if (!IS_ENABLED(CONFIG_KVM_MIPS_VZ)) {
1651 			set_c0_status(ST0_CU1);
1652 			enable_fpu_hazard();
1653 		}
1654 
1655 		__kvm_save_fpu(&vcpu->arch);
1656 		vcpu->arch.aux_inuse &= ~KVM_MIPS_AUX_FPU;
1657 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_SAVE, KVM_TRACE_AUX_FPU);
1658 
1659 		/* Disable FPU */
1660 		clear_c0_status(ST0_CU1 | ST0_FR);
1661 		disable_fpu_hazard();
1662 	}
1663 	preempt_enable();
1664 }
1665 
1666 /*
1667  * Step over a specific ctc1 to FCSR and a specific ctcmsa to MSACSR which are
1668  * used to restore guest FCSR/MSACSR state and may trigger a "harmless" FP/MSAFP
1669  * exception if cause bits are set in the value being written.
1670  */
1671 static int kvm_mips_csr_die_notify(struct notifier_block *self,
1672 				   unsigned long cmd, void *ptr)
1673 {
1674 	struct die_args *args = (struct die_args *)ptr;
1675 	struct pt_regs *regs = args->regs;
1676 	unsigned long pc;
1677 
1678 	/* Only interested in FPE and MSAFPE */
1679 	if (cmd != DIE_FP && cmd != DIE_MSAFP)
1680 		return NOTIFY_DONE;
1681 
1682 	/* Return immediately if guest context isn't active */
1683 	if (!(current->flags & PF_VCPU))
1684 		return NOTIFY_DONE;
1685 
1686 	/* Should never get here from user mode */
1687 	BUG_ON(user_mode(regs));
1688 
1689 	pc = instruction_pointer(regs);
1690 	switch (cmd) {
1691 	case DIE_FP:
1692 		/* match 2nd instruction in __kvm_restore_fcsr */
1693 		if (pc != (unsigned long)&__kvm_restore_fcsr + 4)
1694 			return NOTIFY_DONE;
1695 		break;
1696 	case DIE_MSAFP:
1697 		/* match 2nd/3rd instruction in __kvm_restore_msacsr */
1698 		if (!cpu_has_msa ||
1699 		    pc < (unsigned long)&__kvm_restore_msacsr + 4 ||
1700 		    pc > (unsigned long)&__kvm_restore_msacsr + 8)
1701 			return NOTIFY_DONE;
1702 		break;
1703 	}
1704 
1705 	/* Move PC forward a little and continue executing */
1706 	instruction_pointer(regs) += 4;
1707 
1708 	return NOTIFY_STOP;
1709 }
1710 
1711 static struct notifier_block kvm_mips_csr_die_notifier = {
1712 	.notifier_call = kvm_mips_csr_die_notify,
1713 };
1714 
1715 static int __init kvm_mips_init(void)
1716 {
1717 	int ret;
1718 
1719 	if (cpu_has_mmid) {
1720 		pr_warn("KVM does not yet support MMIDs. KVM Disabled\n");
1721 		return -EOPNOTSUPP;
1722 	}
1723 
1724 	ret = kvm_mips_entry_setup();
1725 	if (ret)
1726 		return ret;
1727 
1728 	ret = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1729 
1730 	if (ret)
1731 		return ret;
1732 
1733 	register_die_notifier(&kvm_mips_csr_die_notifier);
1734 
1735 	return 0;
1736 }
1737 
1738 static void __exit kvm_mips_exit(void)
1739 {
1740 	kvm_exit();
1741 
1742 	unregister_die_notifier(&kvm_mips_csr_die_notifier);
1743 }
1744 
1745 module_init(kvm_mips_init);
1746 module_exit(kvm_mips_exit);
1747 
1748 EXPORT_TRACEPOINT_SYMBOL(kvm_exit);
1749