xref: /linux/arch/mips/kvm/mips.c (revision 2fe05e1139a555ae91f00a812cb9520e7d3022ab)
1 /*
2  * This file is subject to the terms and conditions of the GNU General Public
3  * License.  See the file "COPYING" in the main directory of this archive
4  * for more details.
5  *
6  * KVM/MIPS: MIPS specific KVM APIs
7  *
8  * Copyright (C) 2012  MIPS Technologies, Inc.  All rights reserved.
9  * Authors: Sanjay Lal <sanjayl@kymasys.com>
10  */
11 
12 #include <linux/bitops.h>
13 #include <linux/errno.h>
14 #include <linux/err.h>
15 #include <linux/kdebug.h>
16 #include <linux/module.h>
17 #include <linux/uaccess.h>
18 #include <linux/vmalloc.h>
19 #include <linux/sched/signal.h>
20 #include <linux/fs.h>
21 #include <linux/bootmem.h>
22 
23 #include <asm/fpu.h>
24 #include <asm/page.h>
25 #include <asm/cacheflush.h>
26 #include <asm/mmu_context.h>
27 #include <asm/pgalloc.h>
28 #include <asm/pgtable.h>
29 
30 #include <linux/kvm_host.h>
31 
32 #include "interrupt.h"
33 #include "commpage.h"
34 
35 #define CREATE_TRACE_POINTS
36 #include "trace.h"
37 
38 #ifndef VECTORSPACING
39 #define VECTORSPACING 0x100	/* for EI/VI mode */
40 #endif
41 
42 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x)
43 struct kvm_stats_debugfs_item debugfs_entries[] = {
44 	{ "wait",	  VCPU_STAT(wait_exits),	 KVM_STAT_VCPU },
45 	{ "cache",	  VCPU_STAT(cache_exits),	 KVM_STAT_VCPU },
46 	{ "signal",	  VCPU_STAT(signal_exits),	 KVM_STAT_VCPU },
47 	{ "interrupt",	  VCPU_STAT(int_exits),		 KVM_STAT_VCPU },
48 	{ "cop_unsuable", VCPU_STAT(cop_unusable_exits), KVM_STAT_VCPU },
49 	{ "tlbmod",	  VCPU_STAT(tlbmod_exits),	 KVM_STAT_VCPU },
50 	{ "tlbmiss_ld",	  VCPU_STAT(tlbmiss_ld_exits),	 KVM_STAT_VCPU },
51 	{ "tlbmiss_st",	  VCPU_STAT(tlbmiss_st_exits),	 KVM_STAT_VCPU },
52 	{ "addrerr_st",	  VCPU_STAT(addrerr_st_exits),	 KVM_STAT_VCPU },
53 	{ "addrerr_ld",	  VCPU_STAT(addrerr_ld_exits),	 KVM_STAT_VCPU },
54 	{ "syscall",	  VCPU_STAT(syscall_exits),	 KVM_STAT_VCPU },
55 	{ "resvd_inst",	  VCPU_STAT(resvd_inst_exits),	 KVM_STAT_VCPU },
56 	{ "break_inst",	  VCPU_STAT(break_inst_exits),	 KVM_STAT_VCPU },
57 	{ "trap_inst",	  VCPU_STAT(trap_inst_exits),	 KVM_STAT_VCPU },
58 	{ "msa_fpe",	  VCPU_STAT(msa_fpe_exits),	 KVM_STAT_VCPU },
59 	{ "fpe",	  VCPU_STAT(fpe_exits),		 KVM_STAT_VCPU },
60 	{ "msa_disabled", VCPU_STAT(msa_disabled_exits), KVM_STAT_VCPU },
61 	{ "flush_dcache", VCPU_STAT(flush_dcache_exits), KVM_STAT_VCPU },
62 #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 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
102 {
103 	return 1;
104 }
105 
106 int kvm_arch_hardware_enable(void)
107 {
108 	return kvm_mips_callbacks->hardware_enable();
109 }
110 
111 void kvm_arch_hardware_disable(void)
112 {
113 	kvm_mips_callbacks->hardware_disable();
114 }
115 
116 int kvm_arch_hardware_setup(void)
117 {
118 	return 0;
119 }
120 
121 void kvm_arch_check_processor_compat(void *rtn)
122 {
123 	*(int *)rtn = 0;
124 }
125 
126 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
127 {
128 	switch (type) {
129 #ifdef CONFIG_KVM_MIPS_VZ
130 	case KVM_VM_MIPS_VZ:
131 #else
132 	case KVM_VM_MIPS_TE:
133 #endif
134 		break;
135 	default:
136 		/* Unsupported KVM type */
137 		return -EINVAL;
138 	};
139 
140 	/* Allocate page table to map GPA -> RPA */
141 	kvm->arch.gpa_mm.pgd = kvm_pgd_alloc();
142 	if (!kvm->arch.gpa_mm.pgd)
143 		return -ENOMEM;
144 
145 	return 0;
146 }
147 
148 bool kvm_arch_has_vcpu_debugfs(void)
149 {
150 	return false;
151 }
152 
153 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
154 {
155 	return 0;
156 }
157 
158 void kvm_mips_free_vcpus(struct kvm *kvm)
159 {
160 	unsigned int i;
161 	struct kvm_vcpu *vcpu;
162 
163 	kvm_for_each_vcpu(i, vcpu, kvm) {
164 		kvm_arch_vcpu_free(vcpu);
165 	}
166 
167 	mutex_lock(&kvm->lock);
168 
169 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
170 		kvm->vcpus[i] = NULL;
171 
172 	atomic_set(&kvm->online_vcpus, 0);
173 
174 	mutex_unlock(&kvm->lock);
175 }
176 
177 static void kvm_mips_free_gpa_pt(struct kvm *kvm)
178 {
179 	/* It should always be safe to remove after flushing the whole range */
180 	WARN_ON(!kvm_mips_flush_gpa_pt(kvm, 0, ~0));
181 	pgd_free(NULL, kvm->arch.gpa_mm.pgd);
182 }
183 
184 void kvm_arch_destroy_vm(struct kvm *kvm)
185 {
186 	kvm_mips_free_vcpus(kvm);
187 	kvm_mips_free_gpa_pt(kvm);
188 }
189 
190 long kvm_arch_dev_ioctl(struct file *filp, unsigned int ioctl,
191 			unsigned long arg)
192 {
193 	return -ENOIOCTLCMD;
194 }
195 
196 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
197 			    unsigned long npages)
198 {
199 	return 0;
200 }
201 
202 void kvm_arch_flush_shadow_all(struct kvm *kvm)
203 {
204 	/* Flush whole GPA */
205 	kvm_mips_flush_gpa_pt(kvm, 0, ~0);
206 
207 	/* Let implementation do the rest */
208 	kvm_mips_callbacks->flush_shadow_all(kvm);
209 }
210 
211 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
212 				   struct kvm_memory_slot *slot)
213 {
214 	/*
215 	 * The slot has been made invalid (ready for moving or deletion), so we
216 	 * need to ensure that it can no longer be accessed by any guest VCPUs.
217 	 */
218 
219 	spin_lock(&kvm->mmu_lock);
220 	/* Flush slot from GPA */
221 	kvm_mips_flush_gpa_pt(kvm, slot->base_gfn,
222 			      slot->base_gfn + slot->npages - 1);
223 	/* Let implementation do the rest */
224 	kvm_mips_callbacks->flush_shadow_memslot(kvm, slot);
225 	spin_unlock(&kvm->mmu_lock);
226 }
227 
228 int kvm_arch_prepare_memory_region(struct kvm *kvm,
229 				   struct kvm_memory_slot *memslot,
230 				   const struct kvm_userspace_memory_region *mem,
231 				   enum kvm_mr_change change)
232 {
233 	return 0;
234 }
235 
236 void kvm_arch_commit_memory_region(struct kvm *kvm,
237 				   const struct kvm_userspace_memory_region *mem,
238 				   const struct kvm_memory_slot *old,
239 				   const struct kvm_memory_slot *new,
240 				   enum kvm_mr_change change)
241 {
242 	int needs_flush;
243 
244 	kvm_debug("%s: kvm: %p slot: %d, GPA: %llx, size: %llx, QVA: %llx\n",
245 		  __func__, kvm, mem->slot, mem->guest_phys_addr,
246 		  mem->memory_size, mem->userspace_addr);
247 
248 	/*
249 	 * If dirty page logging is enabled, write protect all pages in the slot
250 	 * ready for dirty logging.
251 	 *
252 	 * There is no need to do this in any of the following cases:
253 	 * CREATE:	No dirty mappings will already exist.
254 	 * MOVE/DELETE:	The old mappings will already have been cleaned up by
255 	 *		kvm_arch_flush_shadow_memslot()
256 	 */
257 	if (change == KVM_MR_FLAGS_ONLY &&
258 	    (!(old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
259 	     new->flags & KVM_MEM_LOG_DIRTY_PAGES)) {
260 		spin_lock(&kvm->mmu_lock);
261 		/* Write protect GPA page table entries */
262 		needs_flush = kvm_mips_mkclean_gpa_pt(kvm, new->base_gfn,
263 					new->base_gfn + new->npages - 1);
264 		/* Let implementation do the rest */
265 		if (needs_flush)
266 			kvm_mips_callbacks->flush_shadow_memslot(kvm, new);
267 		spin_unlock(&kvm->mmu_lock);
268 	}
269 }
270 
271 static inline void dump_handler(const char *symbol, void *start, void *end)
272 {
273 	u32 *p;
274 
275 	pr_debug("LEAF(%s)\n", symbol);
276 
277 	pr_debug("\t.set push\n");
278 	pr_debug("\t.set noreorder\n");
279 
280 	for (p = start; p < (u32 *)end; ++p)
281 		pr_debug("\t.word\t0x%08x\t\t# %p\n", *p, p);
282 
283 	pr_debug("\t.set\tpop\n");
284 
285 	pr_debug("\tEND(%s)\n", symbol);
286 }
287 
288 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
289 {
290 	int err, size;
291 	void *gebase, *p, *handler, *refill_start, *refill_end;
292 	int i;
293 
294 	struct kvm_vcpu *vcpu = kzalloc(sizeof(struct kvm_vcpu), GFP_KERNEL);
295 
296 	if (!vcpu) {
297 		err = -ENOMEM;
298 		goto out;
299 	}
300 
301 	err = kvm_vcpu_init(vcpu, kvm, id);
302 
303 	if (err)
304 		goto out_free_cpu;
305 
306 	kvm_debug("kvm @ %p: create cpu %d at %p\n", kvm, id, vcpu);
307 
308 	/*
309 	 * Allocate space for host mode exception handlers that handle
310 	 * guest mode exits
311 	 */
312 	if (cpu_has_veic || cpu_has_vint)
313 		size = 0x200 + VECTORSPACING * 64;
314 	else
315 		size = 0x4000;
316 
317 	gebase = kzalloc(ALIGN(size, PAGE_SIZE), GFP_KERNEL);
318 
319 	if (!gebase) {
320 		err = -ENOMEM;
321 		goto out_uninit_cpu;
322 	}
323 	kvm_debug("Allocated %d bytes for KVM Exception Handlers @ %p\n",
324 		  ALIGN(size, PAGE_SIZE), gebase);
325 
326 	/*
327 	 * Check new ebase actually fits in CP0_EBase. The lack of a write gate
328 	 * limits us to the low 512MB of physical address space. If the memory
329 	 * we allocate is out of range, just give up now.
330 	 */
331 	if (!cpu_has_ebase_wg && virt_to_phys(gebase) >= 0x20000000) {
332 		kvm_err("CP0_EBase.WG required for guest exception base %pK\n",
333 			gebase);
334 		err = -ENOMEM;
335 		goto out_free_gebase;
336 	}
337 
338 	/* Save new ebase */
339 	vcpu->arch.guest_ebase = gebase;
340 
341 	/* Build guest exception vectors dynamically in unmapped memory */
342 	handler = gebase + 0x2000;
343 
344 	/* TLB refill (or XTLB refill on 64-bit VZ where KX=1) */
345 	refill_start = gebase;
346 	if (IS_ENABLED(CONFIG_KVM_MIPS_VZ) && IS_ENABLED(CONFIG_64BIT))
347 		refill_start += 0x080;
348 	refill_end = kvm_mips_build_tlb_refill_exception(refill_start, handler);
349 
350 	/* General Exception Entry point */
351 	kvm_mips_build_exception(gebase + 0x180, handler);
352 
353 	/* For vectored interrupts poke the exception code @ all offsets 0-7 */
354 	for (i = 0; i < 8; i++) {
355 		kvm_debug("L1 Vectored handler @ %p\n",
356 			  gebase + 0x200 + (i * VECTORSPACING));
357 		kvm_mips_build_exception(gebase + 0x200 + i * VECTORSPACING,
358 					 handler);
359 	}
360 
361 	/* General exit handler */
362 	p = handler;
363 	p = kvm_mips_build_exit(p);
364 
365 	/* Guest entry routine */
366 	vcpu->arch.vcpu_run = p;
367 	p = kvm_mips_build_vcpu_run(p);
368 
369 	/* Dump the generated code */
370 	pr_debug("#include <asm/asm.h>\n");
371 	pr_debug("#include <asm/regdef.h>\n");
372 	pr_debug("\n");
373 	dump_handler("kvm_vcpu_run", vcpu->arch.vcpu_run, p);
374 	dump_handler("kvm_tlb_refill", refill_start, refill_end);
375 	dump_handler("kvm_gen_exc", gebase + 0x180, gebase + 0x200);
376 	dump_handler("kvm_exit", gebase + 0x2000, vcpu->arch.vcpu_run);
377 
378 	/* Invalidate the icache for these ranges */
379 	flush_icache_range((unsigned long)gebase,
380 			   (unsigned long)gebase + ALIGN(size, PAGE_SIZE));
381 
382 	/*
383 	 * Allocate comm page for guest kernel, a TLB will be reserved for
384 	 * mapping GVA @ 0xFFFF8000 to this page
385 	 */
386 	vcpu->arch.kseg0_commpage = kzalloc(PAGE_SIZE << 1, GFP_KERNEL);
387 
388 	if (!vcpu->arch.kseg0_commpage) {
389 		err = -ENOMEM;
390 		goto out_free_gebase;
391 	}
392 
393 	kvm_debug("Allocated COMM page @ %p\n", vcpu->arch.kseg0_commpage);
394 	kvm_mips_commpage_init(vcpu);
395 
396 	/* Init */
397 	vcpu->arch.last_sched_cpu = -1;
398 	vcpu->arch.last_exec_cpu = -1;
399 
400 	return vcpu;
401 
402 out_free_gebase:
403 	kfree(gebase);
404 
405 out_uninit_cpu:
406 	kvm_vcpu_uninit(vcpu);
407 
408 out_free_cpu:
409 	kfree(vcpu);
410 
411 out:
412 	return ERR_PTR(err);
413 }
414 
415 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
416 {
417 	hrtimer_cancel(&vcpu->arch.comparecount_timer);
418 
419 	kvm_vcpu_uninit(vcpu);
420 
421 	kvm_mips_dump_stats(vcpu);
422 
423 	kvm_mmu_free_memory_caches(vcpu);
424 	kfree(vcpu->arch.guest_ebase);
425 	kfree(vcpu->arch.kseg0_commpage);
426 	kfree(vcpu);
427 }
428 
429 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
430 {
431 	kvm_arch_vcpu_free(vcpu);
432 }
433 
434 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
435 					struct kvm_guest_debug *dbg)
436 {
437 	return -ENOIOCTLCMD;
438 }
439 
440 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
441 {
442 	int r = -EINTR;
443 	sigset_t sigsaved;
444 
445 	if (vcpu->sigset_active)
446 		sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
447 
448 	if (vcpu->mmio_needed) {
449 		if (!vcpu->mmio_is_write)
450 			kvm_mips_complete_mmio_load(vcpu, run);
451 		vcpu->mmio_needed = 0;
452 	}
453 
454 	if (run->immediate_exit)
455 		goto out;
456 
457 	lose_fpu(1);
458 
459 	local_irq_disable();
460 	guest_enter_irqoff();
461 	trace_kvm_enter(vcpu);
462 
463 	/*
464 	 * Make sure the read of VCPU requests in vcpu_run() callback is not
465 	 * reordered ahead of the write to vcpu->mode, or we could miss a TLB
466 	 * flush request while the requester sees the VCPU as outside of guest
467 	 * mode and not needing an IPI.
468 	 */
469 	smp_store_mb(vcpu->mode, IN_GUEST_MODE);
470 
471 	r = kvm_mips_callbacks->vcpu_run(run, vcpu);
472 
473 	trace_kvm_out(vcpu);
474 	guest_exit_irqoff();
475 	local_irq_enable();
476 
477 out:
478 	if (vcpu->sigset_active)
479 		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
480 
481 	return r;
482 }
483 
484 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
485 			     struct kvm_mips_interrupt *irq)
486 {
487 	int intr = (int)irq->irq;
488 	struct kvm_vcpu *dvcpu = NULL;
489 
490 	if (intr == 3 || intr == -3 || intr == 4 || intr == -4)
491 		kvm_debug("%s: CPU: %d, INTR: %d\n", __func__, irq->cpu,
492 			  (int)intr);
493 
494 	if (irq->cpu == -1)
495 		dvcpu = vcpu;
496 	else
497 		dvcpu = vcpu->kvm->vcpus[irq->cpu];
498 
499 	if (intr == 2 || intr == 3 || intr == 4) {
500 		kvm_mips_callbacks->queue_io_int(dvcpu, irq);
501 
502 	} else if (intr == -2 || intr == -3 || intr == -4) {
503 		kvm_mips_callbacks->dequeue_io_int(dvcpu, irq);
504 	} else {
505 		kvm_err("%s: invalid interrupt ioctl (%d:%d)\n", __func__,
506 			irq->cpu, irq->irq);
507 		return -EINVAL;
508 	}
509 
510 	dvcpu->arch.wait = 0;
511 
512 	if (swait_active(&dvcpu->wq))
513 		swake_up(&dvcpu->wq);
514 
515 	return 0;
516 }
517 
518 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
519 				    struct kvm_mp_state *mp_state)
520 {
521 	return -ENOIOCTLCMD;
522 }
523 
524 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
525 				    struct kvm_mp_state *mp_state)
526 {
527 	return -ENOIOCTLCMD;
528 }
529 
530 static u64 kvm_mips_get_one_regs[] = {
531 	KVM_REG_MIPS_R0,
532 	KVM_REG_MIPS_R1,
533 	KVM_REG_MIPS_R2,
534 	KVM_REG_MIPS_R3,
535 	KVM_REG_MIPS_R4,
536 	KVM_REG_MIPS_R5,
537 	KVM_REG_MIPS_R6,
538 	KVM_REG_MIPS_R7,
539 	KVM_REG_MIPS_R8,
540 	KVM_REG_MIPS_R9,
541 	KVM_REG_MIPS_R10,
542 	KVM_REG_MIPS_R11,
543 	KVM_REG_MIPS_R12,
544 	KVM_REG_MIPS_R13,
545 	KVM_REG_MIPS_R14,
546 	KVM_REG_MIPS_R15,
547 	KVM_REG_MIPS_R16,
548 	KVM_REG_MIPS_R17,
549 	KVM_REG_MIPS_R18,
550 	KVM_REG_MIPS_R19,
551 	KVM_REG_MIPS_R20,
552 	KVM_REG_MIPS_R21,
553 	KVM_REG_MIPS_R22,
554 	KVM_REG_MIPS_R23,
555 	KVM_REG_MIPS_R24,
556 	KVM_REG_MIPS_R25,
557 	KVM_REG_MIPS_R26,
558 	KVM_REG_MIPS_R27,
559 	KVM_REG_MIPS_R28,
560 	KVM_REG_MIPS_R29,
561 	KVM_REG_MIPS_R30,
562 	KVM_REG_MIPS_R31,
563 
564 #ifndef CONFIG_CPU_MIPSR6
565 	KVM_REG_MIPS_HI,
566 	KVM_REG_MIPS_LO,
567 #endif
568 	KVM_REG_MIPS_PC,
569 };
570 
571 static u64 kvm_mips_get_one_regs_fpu[] = {
572 	KVM_REG_MIPS_FCR_IR,
573 	KVM_REG_MIPS_FCR_CSR,
574 };
575 
576 static u64 kvm_mips_get_one_regs_msa[] = {
577 	KVM_REG_MIPS_MSA_IR,
578 	KVM_REG_MIPS_MSA_CSR,
579 };
580 
581 static unsigned long kvm_mips_num_regs(struct kvm_vcpu *vcpu)
582 {
583 	unsigned long ret;
584 
585 	ret = ARRAY_SIZE(kvm_mips_get_one_regs);
586 	if (kvm_mips_guest_can_have_fpu(&vcpu->arch)) {
587 		ret += ARRAY_SIZE(kvm_mips_get_one_regs_fpu) + 48;
588 		/* odd doubles */
589 		if (boot_cpu_data.fpu_id & MIPS_FPIR_F64)
590 			ret += 16;
591 	}
592 	if (kvm_mips_guest_can_have_msa(&vcpu->arch))
593 		ret += ARRAY_SIZE(kvm_mips_get_one_regs_msa) + 32;
594 	ret += kvm_mips_callbacks->num_regs(vcpu);
595 
596 	return ret;
597 }
598 
599 static int kvm_mips_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *indices)
600 {
601 	u64 index;
602 	unsigned int i;
603 
604 	if (copy_to_user(indices, kvm_mips_get_one_regs,
605 			 sizeof(kvm_mips_get_one_regs)))
606 		return -EFAULT;
607 	indices += ARRAY_SIZE(kvm_mips_get_one_regs);
608 
609 	if (kvm_mips_guest_can_have_fpu(&vcpu->arch)) {
610 		if (copy_to_user(indices, kvm_mips_get_one_regs_fpu,
611 				 sizeof(kvm_mips_get_one_regs_fpu)))
612 			return -EFAULT;
613 		indices += ARRAY_SIZE(kvm_mips_get_one_regs_fpu);
614 
615 		for (i = 0; i < 32; ++i) {
616 			index = KVM_REG_MIPS_FPR_32(i);
617 			if (copy_to_user(indices, &index, sizeof(index)))
618 				return -EFAULT;
619 			++indices;
620 
621 			/* skip odd doubles if no F64 */
622 			if (i & 1 && !(boot_cpu_data.fpu_id & MIPS_FPIR_F64))
623 				continue;
624 
625 			index = KVM_REG_MIPS_FPR_64(i);
626 			if (copy_to_user(indices, &index, sizeof(index)))
627 				return -EFAULT;
628 			++indices;
629 		}
630 	}
631 
632 	if (kvm_mips_guest_can_have_msa(&vcpu->arch)) {
633 		if (copy_to_user(indices, kvm_mips_get_one_regs_msa,
634 				 sizeof(kvm_mips_get_one_regs_msa)))
635 			return -EFAULT;
636 		indices += ARRAY_SIZE(kvm_mips_get_one_regs_msa);
637 
638 		for (i = 0; i < 32; ++i) {
639 			index = KVM_REG_MIPS_VEC_128(i);
640 			if (copy_to_user(indices, &index, sizeof(index)))
641 				return -EFAULT;
642 			++indices;
643 		}
644 	}
645 
646 	return kvm_mips_callbacks->copy_reg_indices(vcpu, indices);
647 }
648 
649 static int kvm_mips_get_reg(struct kvm_vcpu *vcpu,
650 			    const struct kvm_one_reg *reg)
651 {
652 	struct mips_coproc *cop0 = vcpu->arch.cop0;
653 	struct mips_fpu_struct *fpu = &vcpu->arch.fpu;
654 	int ret;
655 	s64 v;
656 	s64 vs[2];
657 	unsigned int idx;
658 
659 	switch (reg->id) {
660 	/* General purpose registers */
661 	case KVM_REG_MIPS_R0 ... KVM_REG_MIPS_R31:
662 		v = (long)vcpu->arch.gprs[reg->id - KVM_REG_MIPS_R0];
663 		break;
664 #ifndef CONFIG_CPU_MIPSR6
665 	case KVM_REG_MIPS_HI:
666 		v = (long)vcpu->arch.hi;
667 		break;
668 	case KVM_REG_MIPS_LO:
669 		v = (long)vcpu->arch.lo;
670 		break;
671 #endif
672 	case KVM_REG_MIPS_PC:
673 		v = (long)vcpu->arch.pc;
674 		break;
675 
676 	/* Floating point registers */
677 	case KVM_REG_MIPS_FPR_32(0) ... KVM_REG_MIPS_FPR_32(31):
678 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
679 			return -EINVAL;
680 		idx = reg->id - KVM_REG_MIPS_FPR_32(0);
681 		/* Odd singles in top of even double when FR=0 */
682 		if (kvm_read_c0_guest_status(cop0) & ST0_FR)
683 			v = get_fpr32(&fpu->fpr[idx], 0);
684 		else
685 			v = get_fpr32(&fpu->fpr[idx & ~1], idx & 1);
686 		break;
687 	case KVM_REG_MIPS_FPR_64(0) ... KVM_REG_MIPS_FPR_64(31):
688 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
689 			return -EINVAL;
690 		idx = reg->id - KVM_REG_MIPS_FPR_64(0);
691 		/* Can't access odd doubles in FR=0 mode */
692 		if (idx & 1 && !(kvm_read_c0_guest_status(cop0) & ST0_FR))
693 			return -EINVAL;
694 		v = get_fpr64(&fpu->fpr[idx], 0);
695 		break;
696 	case KVM_REG_MIPS_FCR_IR:
697 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
698 			return -EINVAL;
699 		v = boot_cpu_data.fpu_id;
700 		break;
701 	case KVM_REG_MIPS_FCR_CSR:
702 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
703 			return -EINVAL;
704 		v = fpu->fcr31;
705 		break;
706 
707 	/* MIPS SIMD Architecture (MSA) registers */
708 	case KVM_REG_MIPS_VEC_128(0) ... KVM_REG_MIPS_VEC_128(31):
709 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
710 			return -EINVAL;
711 		/* Can't access MSA registers in FR=0 mode */
712 		if (!(kvm_read_c0_guest_status(cop0) & ST0_FR))
713 			return -EINVAL;
714 		idx = reg->id - KVM_REG_MIPS_VEC_128(0);
715 #ifdef CONFIG_CPU_LITTLE_ENDIAN
716 		/* least significant byte first */
717 		vs[0] = get_fpr64(&fpu->fpr[idx], 0);
718 		vs[1] = get_fpr64(&fpu->fpr[idx], 1);
719 #else
720 		/* most significant byte first */
721 		vs[0] = get_fpr64(&fpu->fpr[idx], 1);
722 		vs[1] = get_fpr64(&fpu->fpr[idx], 0);
723 #endif
724 		break;
725 	case KVM_REG_MIPS_MSA_IR:
726 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
727 			return -EINVAL;
728 		v = boot_cpu_data.msa_id;
729 		break;
730 	case KVM_REG_MIPS_MSA_CSR:
731 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
732 			return -EINVAL;
733 		v = fpu->msacsr;
734 		break;
735 
736 	/* registers to be handled specially */
737 	default:
738 		ret = kvm_mips_callbacks->get_one_reg(vcpu, reg, &v);
739 		if (ret)
740 			return ret;
741 		break;
742 	}
743 	if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U64) {
744 		u64 __user *uaddr64 = (u64 __user *)(long)reg->addr;
745 
746 		return put_user(v, uaddr64);
747 	} else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U32) {
748 		u32 __user *uaddr32 = (u32 __user *)(long)reg->addr;
749 		u32 v32 = (u32)v;
750 
751 		return put_user(v32, uaddr32);
752 	} else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U128) {
753 		void __user *uaddr = (void __user *)(long)reg->addr;
754 
755 		return copy_to_user(uaddr, vs, 16) ? -EFAULT : 0;
756 	} else {
757 		return -EINVAL;
758 	}
759 }
760 
761 static int kvm_mips_set_reg(struct kvm_vcpu *vcpu,
762 			    const struct kvm_one_reg *reg)
763 {
764 	struct mips_coproc *cop0 = vcpu->arch.cop0;
765 	struct mips_fpu_struct *fpu = &vcpu->arch.fpu;
766 	s64 v;
767 	s64 vs[2];
768 	unsigned int idx;
769 
770 	if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U64) {
771 		u64 __user *uaddr64 = (u64 __user *)(long)reg->addr;
772 
773 		if (get_user(v, uaddr64) != 0)
774 			return -EFAULT;
775 	} else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U32) {
776 		u32 __user *uaddr32 = (u32 __user *)(long)reg->addr;
777 		s32 v32;
778 
779 		if (get_user(v32, uaddr32) != 0)
780 			return -EFAULT;
781 		v = (s64)v32;
782 	} else if ((reg->id & KVM_REG_SIZE_MASK) == KVM_REG_SIZE_U128) {
783 		void __user *uaddr = (void __user *)(long)reg->addr;
784 
785 		return copy_from_user(vs, uaddr, 16) ? -EFAULT : 0;
786 	} else {
787 		return -EINVAL;
788 	}
789 
790 	switch (reg->id) {
791 	/* General purpose registers */
792 	case KVM_REG_MIPS_R0:
793 		/* Silently ignore requests to set $0 */
794 		break;
795 	case KVM_REG_MIPS_R1 ... KVM_REG_MIPS_R31:
796 		vcpu->arch.gprs[reg->id - KVM_REG_MIPS_R0] = v;
797 		break;
798 #ifndef CONFIG_CPU_MIPSR6
799 	case KVM_REG_MIPS_HI:
800 		vcpu->arch.hi = v;
801 		break;
802 	case KVM_REG_MIPS_LO:
803 		vcpu->arch.lo = v;
804 		break;
805 #endif
806 	case KVM_REG_MIPS_PC:
807 		vcpu->arch.pc = v;
808 		break;
809 
810 	/* Floating point registers */
811 	case KVM_REG_MIPS_FPR_32(0) ... KVM_REG_MIPS_FPR_32(31):
812 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
813 			return -EINVAL;
814 		idx = reg->id - KVM_REG_MIPS_FPR_32(0);
815 		/* Odd singles in top of even double when FR=0 */
816 		if (kvm_read_c0_guest_status(cop0) & ST0_FR)
817 			set_fpr32(&fpu->fpr[idx], 0, v);
818 		else
819 			set_fpr32(&fpu->fpr[idx & ~1], idx & 1, v);
820 		break;
821 	case KVM_REG_MIPS_FPR_64(0) ... KVM_REG_MIPS_FPR_64(31):
822 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
823 			return -EINVAL;
824 		idx = reg->id - KVM_REG_MIPS_FPR_64(0);
825 		/* Can't access odd doubles in FR=0 mode */
826 		if (idx & 1 && !(kvm_read_c0_guest_status(cop0) & ST0_FR))
827 			return -EINVAL;
828 		set_fpr64(&fpu->fpr[idx], 0, v);
829 		break;
830 	case KVM_REG_MIPS_FCR_IR:
831 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
832 			return -EINVAL;
833 		/* Read-only */
834 		break;
835 	case KVM_REG_MIPS_FCR_CSR:
836 		if (!kvm_mips_guest_has_fpu(&vcpu->arch))
837 			return -EINVAL;
838 		fpu->fcr31 = v;
839 		break;
840 
841 	/* MIPS SIMD Architecture (MSA) registers */
842 	case KVM_REG_MIPS_VEC_128(0) ... KVM_REG_MIPS_VEC_128(31):
843 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
844 			return -EINVAL;
845 		idx = reg->id - KVM_REG_MIPS_VEC_128(0);
846 #ifdef CONFIG_CPU_LITTLE_ENDIAN
847 		/* least significant byte first */
848 		set_fpr64(&fpu->fpr[idx], 0, vs[0]);
849 		set_fpr64(&fpu->fpr[idx], 1, vs[1]);
850 #else
851 		/* most significant byte first */
852 		set_fpr64(&fpu->fpr[idx], 1, vs[0]);
853 		set_fpr64(&fpu->fpr[idx], 0, vs[1]);
854 #endif
855 		break;
856 	case KVM_REG_MIPS_MSA_IR:
857 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
858 			return -EINVAL;
859 		/* Read-only */
860 		break;
861 	case KVM_REG_MIPS_MSA_CSR:
862 		if (!kvm_mips_guest_has_msa(&vcpu->arch))
863 			return -EINVAL;
864 		fpu->msacsr = v;
865 		break;
866 
867 	/* registers to be handled specially */
868 	default:
869 		return kvm_mips_callbacks->set_one_reg(vcpu, reg, v);
870 	}
871 	return 0;
872 }
873 
874 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
875 				     struct kvm_enable_cap *cap)
876 {
877 	int r = 0;
878 
879 	if (!kvm_vm_ioctl_check_extension(vcpu->kvm, cap->cap))
880 		return -EINVAL;
881 	if (cap->flags)
882 		return -EINVAL;
883 	if (cap->args[0])
884 		return -EINVAL;
885 
886 	switch (cap->cap) {
887 	case KVM_CAP_MIPS_FPU:
888 		vcpu->arch.fpu_enabled = true;
889 		break;
890 	case KVM_CAP_MIPS_MSA:
891 		vcpu->arch.msa_enabled = true;
892 		break;
893 	default:
894 		r = -EINVAL;
895 		break;
896 	}
897 
898 	return r;
899 }
900 
901 long kvm_arch_vcpu_ioctl(struct file *filp, unsigned int ioctl,
902 			 unsigned long arg)
903 {
904 	struct kvm_vcpu *vcpu = filp->private_data;
905 	void __user *argp = (void __user *)arg;
906 	long r;
907 
908 	switch (ioctl) {
909 	case KVM_SET_ONE_REG:
910 	case KVM_GET_ONE_REG: {
911 		struct kvm_one_reg reg;
912 
913 		if (copy_from_user(&reg, argp, sizeof(reg)))
914 			return -EFAULT;
915 		if (ioctl == KVM_SET_ONE_REG)
916 			return kvm_mips_set_reg(vcpu, &reg);
917 		else
918 			return kvm_mips_get_reg(vcpu, &reg);
919 	}
920 	case KVM_GET_REG_LIST: {
921 		struct kvm_reg_list __user *user_list = argp;
922 		struct kvm_reg_list reg_list;
923 		unsigned n;
924 
925 		if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
926 			return -EFAULT;
927 		n = reg_list.n;
928 		reg_list.n = kvm_mips_num_regs(vcpu);
929 		if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
930 			return -EFAULT;
931 		if (n < reg_list.n)
932 			return -E2BIG;
933 		return kvm_mips_copy_reg_indices(vcpu, user_list->reg);
934 	}
935 	case KVM_INTERRUPT:
936 		{
937 			struct kvm_mips_interrupt irq;
938 
939 			if (copy_from_user(&irq, argp, sizeof(irq)))
940 				return -EFAULT;
941 			kvm_debug("[%d] %s: irq: %d\n", vcpu->vcpu_id, __func__,
942 				  irq.irq);
943 
944 			r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
945 			break;
946 		}
947 	case KVM_ENABLE_CAP: {
948 		struct kvm_enable_cap cap;
949 
950 		if (copy_from_user(&cap, argp, sizeof(cap)))
951 			return -EFAULT;
952 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
953 		break;
954 	}
955 	default:
956 		r = -ENOIOCTLCMD;
957 	}
958 	return r;
959 }
960 
961 /**
962  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
963  * @kvm: kvm instance
964  * @log: slot id and address to which we copy the log
965  *
966  * Steps 1-4 below provide general overview of dirty page logging. See
967  * kvm_get_dirty_log_protect() function description for additional details.
968  *
969  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
970  * always flush the TLB (step 4) even if previous step failed  and the dirty
971  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
972  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
973  * writes will be marked dirty for next log read.
974  *
975  *   1. Take a snapshot of the bit and clear it if needed.
976  *   2. Write protect the corresponding page.
977  *   3. Copy the snapshot to the userspace.
978  *   4. Flush TLB's if needed.
979  */
980 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
981 {
982 	struct kvm_memslots *slots;
983 	struct kvm_memory_slot *memslot;
984 	bool is_dirty = false;
985 	int r;
986 
987 	mutex_lock(&kvm->slots_lock);
988 
989 	r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
990 
991 	if (is_dirty) {
992 		slots = kvm_memslots(kvm);
993 		memslot = id_to_memslot(slots, log->slot);
994 
995 		/* Let implementation handle TLB/GVA invalidation */
996 		kvm_mips_callbacks->flush_shadow_memslot(kvm, memslot);
997 	}
998 
999 	mutex_unlock(&kvm->slots_lock);
1000 	return r;
1001 }
1002 
1003 long kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
1004 {
1005 	long r;
1006 
1007 	switch (ioctl) {
1008 	default:
1009 		r = -ENOIOCTLCMD;
1010 	}
1011 
1012 	return r;
1013 }
1014 
1015 int kvm_arch_init(void *opaque)
1016 {
1017 	if (kvm_mips_callbacks) {
1018 		kvm_err("kvm: module already exists\n");
1019 		return -EEXIST;
1020 	}
1021 
1022 	return kvm_mips_emulation_init(&kvm_mips_callbacks);
1023 }
1024 
1025 void kvm_arch_exit(void)
1026 {
1027 	kvm_mips_callbacks = NULL;
1028 }
1029 
1030 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1031 				  struct kvm_sregs *sregs)
1032 {
1033 	return -ENOIOCTLCMD;
1034 }
1035 
1036 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1037 				  struct kvm_sregs *sregs)
1038 {
1039 	return -ENOIOCTLCMD;
1040 }
1041 
1042 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
1043 {
1044 }
1045 
1046 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1047 {
1048 	return -ENOIOCTLCMD;
1049 }
1050 
1051 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1052 {
1053 	return -ENOIOCTLCMD;
1054 }
1055 
1056 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
1057 {
1058 	return VM_FAULT_SIGBUS;
1059 }
1060 
1061 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
1062 {
1063 	int r;
1064 
1065 	switch (ext) {
1066 	case KVM_CAP_ONE_REG:
1067 	case KVM_CAP_ENABLE_CAP:
1068 	case KVM_CAP_READONLY_MEM:
1069 	case KVM_CAP_SYNC_MMU:
1070 	case KVM_CAP_IMMEDIATE_EXIT:
1071 		r = 1;
1072 		break;
1073 	case KVM_CAP_NR_VCPUS:
1074 		r = num_online_cpus();
1075 		break;
1076 	case KVM_CAP_MAX_VCPUS:
1077 		r = KVM_MAX_VCPUS;
1078 		break;
1079 	case KVM_CAP_MIPS_FPU:
1080 		/* We don't handle systems with inconsistent cpu_has_fpu */
1081 		r = !!raw_cpu_has_fpu;
1082 		break;
1083 	case KVM_CAP_MIPS_MSA:
1084 		/*
1085 		 * We don't support MSA vector partitioning yet:
1086 		 * 1) It would require explicit support which can't be tested
1087 		 *    yet due to lack of support in current hardware.
1088 		 * 2) It extends the state that would need to be saved/restored
1089 		 *    by e.g. QEMU for migration.
1090 		 *
1091 		 * When vector partitioning hardware becomes available, support
1092 		 * could be added by requiring a flag when enabling
1093 		 * KVM_CAP_MIPS_MSA capability to indicate that userland knows
1094 		 * to save/restore the appropriate extra state.
1095 		 */
1096 		r = cpu_has_msa && !(boot_cpu_data.msa_id & MSA_IR_WRPF);
1097 		break;
1098 	default:
1099 		r = kvm_mips_callbacks->check_extension(kvm, ext);
1100 		break;
1101 	}
1102 	return r;
1103 }
1104 
1105 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1106 {
1107 	return kvm_mips_pending_timer(vcpu) ||
1108 		kvm_read_c0_guest_cause(vcpu->arch.cop0) & C_TI;
1109 }
1110 
1111 int kvm_arch_vcpu_dump_regs(struct kvm_vcpu *vcpu)
1112 {
1113 	int i;
1114 	struct mips_coproc *cop0;
1115 
1116 	if (!vcpu)
1117 		return -1;
1118 
1119 	kvm_debug("VCPU Register Dump:\n");
1120 	kvm_debug("\tpc = 0x%08lx\n", vcpu->arch.pc);
1121 	kvm_debug("\texceptions: %08lx\n", vcpu->arch.pending_exceptions);
1122 
1123 	for (i = 0; i < 32; i += 4) {
1124 		kvm_debug("\tgpr%02d: %08lx %08lx %08lx %08lx\n", i,
1125 		       vcpu->arch.gprs[i],
1126 		       vcpu->arch.gprs[i + 1],
1127 		       vcpu->arch.gprs[i + 2], vcpu->arch.gprs[i + 3]);
1128 	}
1129 	kvm_debug("\thi: 0x%08lx\n", vcpu->arch.hi);
1130 	kvm_debug("\tlo: 0x%08lx\n", vcpu->arch.lo);
1131 
1132 	cop0 = vcpu->arch.cop0;
1133 	kvm_debug("\tStatus: 0x%08x, Cause: 0x%08x\n",
1134 		  kvm_read_c0_guest_status(cop0),
1135 		  kvm_read_c0_guest_cause(cop0));
1136 
1137 	kvm_debug("\tEPC: 0x%08lx\n", kvm_read_c0_guest_epc(cop0));
1138 
1139 	return 0;
1140 }
1141 
1142 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1143 {
1144 	int i;
1145 
1146 	for (i = 1; i < ARRAY_SIZE(vcpu->arch.gprs); i++)
1147 		vcpu->arch.gprs[i] = regs->gpr[i];
1148 	vcpu->arch.gprs[0] = 0; /* zero is special, and cannot be set. */
1149 	vcpu->arch.hi = regs->hi;
1150 	vcpu->arch.lo = regs->lo;
1151 	vcpu->arch.pc = regs->pc;
1152 
1153 	return 0;
1154 }
1155 
1156 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1157 {
1158 	int i;
1159 
1160 	for (i = 0; i < ARRAY_SIZE(vcpu->arch.gprs); i++)
1161 		regs->gpr[i] = vcpu->arch.gprs[i];
1162 
1163 	regs->hi = vcpu->arch.hi;
1164 	regs->lo = vcpu->arch.lo;
1165 	regs->pc = vcpu->arch.pc;
1166 
1167 	return 0;
1168 }
1169 
1170 static void kvm_mips_comparecount_func(unsigned long data)
1171 {
1172 	struct kvm_vcpu *vcpu = (struct kvm_vcpu *)data;
1173 
1174 	kvm_mips_callbacks->queue_timer_int(vcpu);
1175 
1176 	vcpu->arch.wait = 0;
1177 	if (swait_active(&vcpu->wq))
1178 		swake_up(&vcpu->wq);
1179 }
1180 
1181 /* low level hrtimer wake routine */
1182 static enum hrtimer_restart kvm_mips_comparecount_wakeup(struct hrtimer *timer)
1183 {
1184 	struct kvm_vcpu *vcpu;
1185 
1186 	vcpu = container_of(timer, struct kvm_vcpu, arch.comparecount_timer);
1187 	kvm_mips_comparecount_func((unsigned long) vcpu);
1188 	return kvm_mips_count_timeout(vcpu);
1189 }
1190 
1191 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
1192 {
1193 	int err;
1194 
1195 	err = kvm_mips_callbacks->vcpu_init(vcpu);
1196 	if (err)
1197 		return err;
1198 
1199 	hrtimer_init(&vcpu->arch.comparecount_timer, CLOCK_MONOTONIC,
1200 		     HRTIMER_MODE_REL);
1201 	vcpu->arch.comparecount_timer.function = kvm_mips_comparecount_wakeup;
1202 	return 0;
1203 }
1204 
1205 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
1206 {
1207 	kvm_mips_callbacks->vcpu_uninit(vcpu);
1208 }
1209 
1210 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1211 				  struct kvm_translation *tr)
1212 {
1213 	return 0;
1214 }
1215 
1216 /* Initial guest state */
1217 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
1218 {
1219 	return kvm_mips_callbacks->vcpu_setup(vcpu);
1220 }
1221 
1222 static void kvm_mips_set_c0_status(void)
1223 {
1224 	u32 status = read_c0_status();
1225 
1226 	if (cpu_has_dsp)
1227 		status |= (ST0_MX);
1228 
1229 	write_c0_status(status);
1230 	ehb();
1231 }
1232 
1233 /*
1234  * Return value is in the form (errcode<<2 | RESUME_FLAG_HOST | RESUME_FLAG_NV)
1235  */
1236 int kvm_mips_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu)
1237 {
1238 	u32 cause = vcpu->arch.host_cp0_cause;
1239 	u32 exccode = (cause >> CAUSEB_EXCCODE) & 0x1f;
1240 	u32 __user *opc = (u32 __user *) vcpu->arch.pc;
1241 	unsigned long badvaddr = vcpu->arch.host_cp0_badvaddr;
1242 	enum emulation_result er = EMULATE_DONE;
1243 	u32 inst;
1244 	int ret = RESUME_GUEST;
1245 
1246 	vcpu->mode = OUTSIDE_GUEST_MODE;
1247 
1248 	/* re-enable HTW before enabling interrupts */
1249 	if (!IS_ENABLED(CONFIG_KVM_MIPS_VZ))
1250 		htw_start();
1251 
1252 	/* Set a default exit reason */
1253 	run->exit_reason = KVM_EXIT_UNKNOWN;
1254 	run->ready_for_interrupt_injection = 1;
1255 
1256 	/*
1257 	 * Set the appropriate status bits based on host CPU features,
1258 	 * before we hit the scheduler
1259 	 */
1260 	kvm_mips_set_c0_status();
1261 
1262 	local_irq_enable();
1263 
1264 	kvm_debug("kvm_mips_handle_exit: cause: %#x, PC: %p, kvm_run: %p, kvm_vcpu: %p\n",
1265 			cause, opc, run, vcpu);
1266 	trace_kvm_exit(vcpu, exccode);
1267 
1268 	if (!IS_ENABLED(CONFIG_KVM_MIPS_VZ)) {
1269 		/*
1270 		 * Do a privilege check, if in UM most of these exit conditions
1271 		 * end up causing an exception to be delivered to the Guest
1272 		 * Kernel
1273 		 */
1274 		er = kvm_mips_check_privilege(cause, opc, run, vcpu);
1275 		if (er == EMULATE_PRIV_FAIL) {
1276 			goto skip_emul;
1277 		} else if (er == EMULATE_FAIL) {
1278 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1279 			ret = RESUME_HOST;
1280 			goto skip_emul;
1281 		}
1282 	}
1283 
1284 	switch (exccode) {
1285 	case EXCCODE_INT:
1286 		kvm_debug("[%d]EXCCODE_INT @ %p\n", vcpu->vcpu_id, opc);
1287 
1288 		++vcpu->stat.int_exits;
1289 
1290 		if (need_resched())
1291 			cond_resched();
1292 
1293 		ret = RESUME_GUEST;
1294 		break;
1295 
1296 	case EXCCODE_CPU:
1297 		kvm_debug("EXCCODE_CPU: @ PC: %p\n", opc);
1298 
1299 		++vcpu->stat.cop_unusable_exits;
1300 		ret = kvm_mips_callbacks->handle_cop_unusable(vcpu);
1301 		/* XXXKYMA: Might need to return to user space */
1302 		if (run->exit_reason == KVM_EXIT_IRQ_WINDOW_OPEN)
1303 			ret = RESUME_HOST;
1304 		break;
1305 
1306 	case EXCCODE_MOD:
1307 		++vcpu->stat.tlbmod_exits;
1308 		ret = kvm_mips_callbacks->handle_tlb_mod(vcpu);
1309 		break;
1310 
1311 	case EXCCODE_TLBS:
1312 		kvm_debug("TLB ST fault:  cause %#x, status %#x, PC: %p, BadVaddr: %#lx\n",
1313 			  cause, kvm_read_c0_guest_status(vcpu->arch.cop0), opc,
1314 			  badvaddr);
1315 
1316 		++vcpu->stat.tlbmiss_st_exits;
1317 		ret = kvm_mips_callbacks->handle_tlb_st_miss(vcpu);
1318 		break;
1319 
1320 	case EXCCODE_TLBL:
1321 		kvm_debug("TLB LD fault: cause %#x, PC: %p, BadVaddr: %#lx\n",
1322 			  cause, opc, badvaddr);
1323 
1324 		++vcpu->stat.tlbmiss_ld_exits;
1325 		ret = kvm_mips_callbacks->handle_tlb_ld_miss(vcpu);
1326 		break;
1327 
1328 	case EXCCODE_ADES:
1329 		++vcpu->stat.addrerr_st_exits;
1330 		ret = kvm_mips_callbacks->handle_addr_err_st(vcpu);
1331 		break;
1332 
1333 	case EXCCODE_ADEL:
1334 		++vcpu->stat.addrerr_ld_exits;
1335 		ret = kvm_mips_callbacks->handle_addr_err_ld(vcpu);
1336 		break;
1337 
1338 	case EXCCODE_SYS:
1339 		++vcpu->stat.syscall_exits;
1340 		ret = kvm_mips_callbacks->handle_syscall(vcpu);
1341 		break;
1342 
1343 	case EXCCODE_RI:
1344 		++vcpu->stat.resvd_inst_exits;
1345 		ret = kvm_mips_callbacks->handle_res_inst(vcpu);
1346 		break;
1347 
1348 	case EXCCODE_BP:
1349 		++vcpu->stat.break_inst_exits;
1350 		ret = kvm_mips_callbacks->handle_break(vcpu);
1351 		break;
1352 
1353 	case EXCCODE_TR:
1354 		++vcpu->stat.trap_inst_exits;
1355 		ret = kvm_mips_callbacks->handle_trap(vcpu);
1356 		break;
1357 
1358 	case EXCCODE_MSAFPE:
1359 		++vcpu->stat.msa_fpe_exits;
1360 		ret = kvm_mips_callbacks->handle_msa_fpe(vcpu);
1361 		break;
1362 
1363 	case EXCCODE_FPE:
1364 		++vcpu->stat.fpe_exits;
1365 		ret = kvm_mips_callbacks->handle_fpe(vcpu);
1366 		break;
1367 
1368 	case EXCCODE_MSADIS:
1369 		++vcpu->stat.msa_disabled_exits;
1370 		ret = kvm_mips_callbacks->handle_msa_disabled(vcpu);
1371 		break;
1372 
1373 	case EXCCODE_GE:
1374 		/* defer exit accounting to handler */
1375 		ret = kvm_mips_callbacks->handle_guest_exit(vcpu);
1376 		break;
1377 
1378 	default:
1379 		if (cause & CAUSEF_BD)
1380 			opc += 1;
1381 		inst = 0;
1382 		kvm_get_badinstr(opc, vcpu, &inst);
1383 		kvm_err("Exception Code: %d, not yet handled, @ PC: %p, inst: 0x%08x  BadVaddr: %#lx Status: %#x\n",
1384 			exccode, opc, inst, badvaddr,
1385 			kvm_read_c0_guest_status(vcpu->arch.cop0));
1386 		kvm_arch_vcpu_dump_regs(vcpu);
1387 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1388 		ret = RESUME_HOST;
1389 		break;
1390 
1391 	}
1392 
1393 skip_emul:
1394 	local_irq_disable();
1395 
1396 	if (ret == RESUME_GUEST)
1397 		kvm_vz_acquire_htimer(vcpu);
1398 
1399 	if (er == EMULATE_DONE && !(ret & RESUME_HOST))
1400 		kvm_mips_deliver_interrupts(vcpu, cause);
1401 
1402 	if (!(ret & RESUME_HOST)) {
1403 		/* Only check for signals if not already exiting to userspace */
1404 		if (signal_pending(current)) {
1405 			run->exit_reason = KVM_EXIT_INTR;
1406 			ret = (-EINTR << 2) | RESUME_HOST;
1407 			++vcpu->stat.signal_exits;
1408 			trace_kvm_exit(vcpu, KVM_TRACE_EXIT_SIGNAL);
1409 		}
1410 	}
1411 
1412 	if (ret == RESUME_GUEST) {
1413 		trace_kvm_reenter(vcpu);
1414 
1415 		/*
1416 		 * Make sure the read of VCPU requests in vcpu_reenter()
1417 		 * callback is not reordered ahead of the write to vcpu->mode,
1418 		 * or we could miss a TLB flush request while the requester sees
1419 		 * the VCPU as outside of guest mode and not needing an IPI.
1420 		 */
1421 		smp_store_mb(vcpu->mode, IN_GUEST_MODE);
1422 
1423 		kvm_mips_callbacks->vcpu_reenter(run, vcpu);
1424 
1425 		/*
1426 		 * If FPU / MSA are enabled (i.e. the guest's FPU / MSA context
1427 		 * is live), restore FCR31 / MSACSR.
1428 		 *
1429 		 * This should be before returning to the guest exception
1430 		 * vector, as it may well cause an [MSA] FP exception if there
1431 		 * are pending exception bits unmasked. (see
1432 		 * kvm_mips_csr_die_notifier() for how that is handled).
1433 		 */
1434 		if (kvm_mips_guest_has_fpu(&vcpu->arch) &&
1435 		    read_c0_status() & ST0_CU1)
1436 			__kvm_restore_fcsr(&vcpu->arch);
1437 
1438 		if (kvm_mips_guest_has_msa(&vcpu->arch) &&
1439 		    read_c0_config5() & MIPS_CONF5_MSAEN)
1440 			__kvm_restore_msacsr(&vcpu->arch);
1441 	}
1442 
1443 	/* Disable HTW before returning to guest or host */
1444 	if (!IS_ENABLED(CONFIG_KVM_MIPS_VZ))
1445 		htw_stop();
1446 
1447 	return ret;
1448 }
1449 
1450 /* Enable FPU for guest and restore context */
1451 void kvm_own_fpu(struct kvm_vcpu *vcpu)
1452 {
1453 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1454 	unsigned int sr, cfg5;
1455 
1456 	preempt_disable();
1457 
1458 	sr = kvm_read_c0_guest_status(cop0);
1459 
1460 	/*
1461 	 * If MSA state is already live, it is undefined how it interacts with
1462 	 * FR=0 FPU state, and we don't want to hit reserved instruction
1463 	 * exceptions trying to save the MSA state later when CU=1 && FR=1, so
1464 	 * play it safe and save it first.
1465 	 *
1466 	 * In theory we shouldn't ever hit this case since kvm_lose_fpu() should
1467 	 * get called when guest CU1 is set, however we can't trust the guest
1468 	 * not to clobber the status register directly via the commpage.
1469 	 */
1470 	if (cpu_has_msa && sr & ST0_CU1 && !(sr & ST0_FR) &&
1471 	    vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA)
1472 		kvm_lose_fpu(vcpu);
1473 
1474 	/*
1475 	 * Enable FPU for guest
1476 	 * We set FR and FRE according to guest context
1477 	 */
1478 	change_c0_status(ST0_CU1 | ST0_FR, sr);
1479 	if (cpu_has_fre) {
1480 		cfg5 = kvm_read_c0_guest_config5(cop0);
1481 		change_c0_config5(MIPS_CONF5_FRE, cfg5);
1482 	}
1483 	enable_fpu_hazard();
1484 
1485 	/* If guest FPU state not active, restore it now */
1486 	if (!(vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU)) {
1487 		__kvm_restore_fpu(&vcpu->arch);
1488 		vcpu->arch.aux_inuse |= KVM_MIPS_AUX_FPU;
1489 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE, KVM_TRACE_AUX_FPU);
1490 	} else {
1491 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_ENABLE, KVM_TRACE_AUX_FPU);
1492 	}
1493 
1494 	preempt_enable();
1495 }
1496 
1497 #ifdef CONFIG_CPU_HAS_MSA
1498 /* Enable MSA for guest and restore context */
1499 void kvm_own_msa(struct kvm_vcpu *vcpu)
1500 {
1501 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1502 	unsigned int sr, cfg5;
1503 
1504 	preempt_disable();
1505 
1506 	/*
1507 	 * Enable FPU if enabled in guest, since we're restoring FPU context
1508 	 * anyway. We set FR and FRE according to guest context.
1509 	 */
1510 	if (kvm_mips_guest_has_fpu(&vcpu->arch)) {
1511 		sr = kvm_read_c0_guest_status(cop0);
1512 
1513 		/*
1514 		 * If FR=0 FPU state is already live, it is undefined how it
1515 		 * interacts with MSA state, so play it safe and save it first.
1516 		 */
1517 		if (!(sr & ST0_FR) &&
1518 		    (vcpu->arch.aux_inuse & (KVM_MIPS_AUX_FPU |
1519 				KVM_MIPS_AUX_MSA)) == KVM_MIPS_AUX_FPU)
1520 			kvm_lose_fpu(vcpu);
1521 
1522 		change_c0_status(ST0_CU1 | ST0_FR, sr);
1523 		if (sr & ST0_CU1 && cpu_has_fre) {
1524 			cfg5 = kvm_read_c0_guest_config5(cop0);
1525 			change_c0_config5(MIPS_CONF5_FRE, cfg5);
1526 		}
1527 	}
1528 
1529 	/* Enable MSA for guest */
1530 	set_c0_config5(MIPS_CONF5_MSAEN);
1531 	enable_fpu_hazard();
1532 
1533 	switch (vcpu->arch.aux_inuse & (KVM_MIPS_AUX_FPU | KVM_MIPS_AUX_MSA)) {
1534 	case KVM_MIPS_AUX_FPU:
1535 		/*
1536 		 * Guest FPU state already loaded, only restore upper MSA state
1537 		 */
1538 		__kvm_restore_msa_upper(&vcpu->arch);
1539 		vcpu->arch.aux_inuse |= KVM_MIPS_AUX_MSA;
1540 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE, KVM_TRACE_AUX_MSA);
1541 		break;
1542 	case 0:
1543 		/* Neither FPU or MSA already active, restore full MSA state */
1544 		__kvm_restore_msa(&vcpu->arch);
1545 		vcpu->arch.aux_inuse |= KVM_MIPS_AUX_MSA;
1546 		if (kvm_mips_guest_has_fpu(&vcpu->arch))
1547 			vcpu->arch.aux_inuse |= KVM_MIPS_AUX_FPU;
1548 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE,
1549 			      KVM_TRACE_AUX_FPU_MSA);
1550 		break;
1551 	default:
1552 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_ENABLE, KVM_TRACE_AUX_MSA);
1553 		break;
1554 	}
1555 
1556 	preempt_enable();
1557 }
1558 #endif
1559 
1560 /* Drop FPU & MSA without saving it */
1561 void kvm_drop_fpu(struct kvm_vcpu *vcpu)
1562 {
1563 	preempt_disable();
1564 	if (cpu_has_msa && vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA) {
1565 		disable_msa();
1566 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_DISCARD, KVM_TRACE_AUX_MSA);
1567 		vcpu->arch.aux_inuse &= ~KVM_MIPS_AUX_MSA;
1568 	}
1569 	if (vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU) {
1570 		clear_c0_status(ST0_CU1 | ST0_FR);
1571 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_DISCARD, KVM_TRACE_AUX_FPU);
1572 		vcpu->arch.aux_inuse &= ~KVM_MIPS_AUX_FPU;
1573 	}
1574 	preempt_enable();
1575 }
1576 
1577 /* Save and disable FPU & MSA */
1578 void kvm_lose_fpu(struct kvm_vcpu *vcpu)
1579 {
1580 	/*
1581 	 * With T&E, FPU & MSA get disabled in root context (hardware) when it
1582 	 * is disabled in guest context (software), but the register state in
1583 	 * the hardware may still be in use.
1584 	 * This is why we explicitly re-enable the hardware before saving.
1585 	 */
1586 
1587 	preempt_disable();
1588 	if (cpu_has_msa && vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA) {
1589 		if (!IS_ENABLED(CONFIG_KVM_MIPS_VZ)) {
1590 			set_c0_config5(MIPS_CONF5_MSAEN);
1591 			enable_fpu_hazard();
1592 		}
1593 
1594 		__kvm_save_msa(&vcpu->arch);
1595 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_SAVE, KVM_TRACE_AUX_FPU_MSA);
1596 
1597 		/* Disable MSA & FPU */
1598 		disable_msa();
1599 		if (vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU) {
1600 			clear_c0_status(ST0_CU1 | ST0_FR);
1601 			disable_fpu_hazard();
1602 		}
1603 		vcpu->arch.aux_inuse &= ~(KVM_MIPS_AUX_FPU | KVM_MIPS_AUX_MSA);
1604 	} else if (vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU) {
1605 		if (!IS_ENABLED(CONFIG_KVM_MIPS_VZ)) {
1606 			set_c0_status(ST0_CU1);
1607 			enable_fpu_hazard();
1608 		}
1609 
1610 		__kvm_save_fpu(&vcpu->arch);
1611 		vcpu->arch.aux_inuse &= ~KVM_MIPS_AUX_FPU;
1612 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_SAVE, KVM_TRACE_AUX_FPU);
1613 
1614 		/* Disable FPU */
1615 		clear_c0_status(ST0_CU1 | ST0_FR);
1616 		disable_fpu_hazard();
1617 	}
1618 	preempt_enable();
1619 }
1620 
1621 /*
1622  * Step over a specific ctc1 to FCSR and a specific ctcmsa to MSACSR which are
1623  * used to restore guest FCSR/MSACSR state and may trigger a "harmless" FP/MSAFP
1624  * exception if cause bits are set in the value being written.
1625  */
1626 static int kvm_mips_csr_die_notify(struct notifier_block *self,
1627 				   unsigned long cmd, void *ptr)
1628 {
1629 	struct die_args *args = (struct die_args *)ptr;
1630 	struct pt_regs *regs = args->regs;
1631 	unsigned long pc;
1632 
1633 	/* Only interested in FPE and MSAFPE */
1634 	if (cmd != DIE_FP && cmd != DIE_MSAFP)
1635 		return NOTIFY_DONE;
1636 
1637 	/* Return immediately if guest context isn't active */
1638 	if (!(current->flags & PF_VCPU))
1639 		return NOTIFY_DONE;
1640 
1641 	/* Should never get here from user mode */
1642 	BUG_ON(user_mode(regs));
1643 
1644 	pc = instruction_pointer(regs);
1645 	switch (cmd) {
1646 	case DIE_FP:
1647 		/* match 2nd instruction in __kvm_restore_fcsr */
1648 		if (pc != (unsigned long)&__kvm_restore_fcsr + 4)
1649 			return NOTIFY_DONE;
1650 		break;
1651 	case DIE_MSAFP:
1652 		/* match 2nd/3rd instruction in __kvm_restore_msacsr */
1653 		if (!cpu_has_msa ||
1654 		    pc < (unsigned long)&__kvm_restore_msacsr + 4 ||
1655 		    pc > (unsigned long)&__kvm_restore_msacsr + 8)
1656 			return NOTIFY_DONE;
1657 		break;
1658 	}
1659 
1660 	/* Move PC forward a little and continue executing */
1661 	instruction_pointer(regs) += 4;
1662 
1663 	return NOTIFY_STOP;
1664 }
1665 
1666 static struct notifier_block kvm_mips_csr_die_notifier = {
1667 	.notifier_call = kvm_mips_csr_die_notify,
1668 };
1669 
1670 static int __init kvm_mips_init(void)
1671 {
1672 	int ret;
1673 
1674 	ret = kvm_mips_entry_setup();
1675 	if (ret)
1676 		return ret;
1677 
1678 	ret = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1679 
1680 	if (ret)
1681 		return ret;
1682 
1683 	register_die_notifier(&kvm_mips_csr_die_notifier);
1684 
1685 	return 0;
1686 }
1687 
1688 static void __exit kvm_mips_exit(void)
1689 {
1690 	kvm_exit();
1691 
1692 	unregister_die_notifier(&kvm_mips_csr_die_notifier);
1693 }
1694 
1695 module_init(kvm_mips_init);
1696 module_exit(kvm_mips_exit);
1697 
1698 EXPORT_TRACEPOINT_SYMBOL(kvm_exit);
1699