xref: /openbmc/linux/arch/mips/kvm/vz.c (revision f20c7d91)
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: Support for hardware virtualization extensions
7  *
8  * Copyright (C) 2012  MIPS Technologies, Inc.  All rights reserved.
9  * Authors: Yann Le Du <ledu@kymasys.com>
10  */
11 
12 #include <linux/errno.h>
13 #include <linux/err.h>
14 #include <linux/module.h>
15 #include <linux/preempt.h>
16 #include <linux/vmalloc.h>
17 #include <asm/cacheflush.h>
18 #include <asm/cacheops.h>
19 #include <asm/cmpxchg.h>
20 #include <asm/fpu.h>
21 #include <asm/hazards.h>
22 #include <asm/inst.h>
23 #include <asm/mmu_context.h>
24 #include <asm/r4kcache.h>
25 #include <asm/time.h>
26 #include <asm/tlb.h>
27 #include <asm/tlbex.h>
28 
29 #include <linux/kvm_host.h>
30 
31 #include "interrupt.h"
32 #include "loongson_regs.h"
33 
34 #include "trace.h"
35 
36 /* Pointers to last VCPU loaded on each physical CPU */
37 static struct kvm_vcpu *last_vcpu[NR_CPUS];
38 /* Pointers to last VCPU executed on each physical CPU */
39 static struct kvm_vcpu *last_exec_vcpu[NR_CPUS];
40 
41 /*
42  * Number of guest VTLB entries to use, so we can catch inconsistency between
43  * CPUs.
44  */
45 static unsigned int kvm_vz_guest_vtlb_size;
46 
47 static inline long kvm_vz_read_gc0_ebase(void)
48 {
49 	if (sizeof(long) == 8 && cpu_has_ebase_wg)
50 		return read_gc0_ebase_64();
51 	else
52 		return read_gc0_ebase();
53 }
54 
55 static inline void kvm_vz_write_gc0_ebase(long v)
56 {
57 	/*
58 	 * First write with WG=1 to write upper bits, then write again in case
59 	 * WG should be left at 0.
60 	 * write_gc0_ebase_64() is no longer UNDEFINED since R6.
61 	 */
62 	if (sizeof(long) == 8 &&
63 	    (cpu_has_mips64r6 || cpu_has_ebase_wg)) {
64 		write_gc0_ebase_64(v | MIPS_EBASE_WG);
65 		write_gc0_ebase_64(v);
66 	} else {
67 		write_gc0_ebase(v | MIPS_EBASE_WG);
68 		write_gc0_ebase(v);
69 	}
70 }
71 
72 /*
73  * These Config bits may be writable by the guest:
74  * Config:	[K23, KU] (!TLB), K0
75  * Config1:	(none)
76  * Config2:	[TU, SU] (impl)
77  * Config3:	ISAOnExc
78  * Config4:	FTLBPageSize
79  * Config5:	K, CV, MSAEn, UFE, FRE, SBRI, UFR
80  */
81 
82 static inline unsigned int kvm_vz_config_guest_wrmask(struct kvm_vcpu *vcpu)
83 {
84 	return CONF_CM_CMASK;
85 }
86 
87 static inline unsigned int kvm_vz_config1_guest_wrmask(struct kvm_vcpu *vcpu)
88 {
89 	return 0;
90 }
91 
92 static inline unsigned int kvm_vz_config2_guest_wrmask(struct kvm_vcpu *vcpu)
93 {
94 	return 0;
95 }
96 
97 static inline unsigned int kvm_vz_config3_guest_wrmask(struct kvm_vcpu *vcpu)
98 {
99 	return MIPS_CONF3_ISA_OE;
100 }
101 
102 static inline unsigned int kvm_vz_config4_guest_wrmask(struct kvm_vcpu *vcpu)
103 {
104 	/* no need to be exact */
105 	return MIPS_CONF4_VFTLBPAGESIZE;
106 }
107 
108 static inline unsigned int kvm_vz_config5_guest_wrmask(struct kvm_vcpu *vcpu)
109 {
110 	unsigned int mask = MIPS_CONF5_K | MIPS_CONF5_CV | MIPS_CONF5_SBRI;
111 
112 	/* Permit MSAEn changes if MSA supported and enabled */
113 	if (kvm_mips_guest_has_msa(&vcpu->arch))
114 		mask |= MIPS_CONF5_MSAEN;
115 
116 	/*
117 	 * Permit guest FPU mode changes if FPU is enabled and the relevant
118 	 * feature exists according to FIR register.
119 	 */
120 	if (kvm_mips_guest_has_fpu(&vcpu->arch)) {
121 		if (cpu_has_ufr)
122 			mask |= MIPS_CONF5_UFR;
123 		if (cpu_has_fre)
124 			mask |= MIPS_CONF5_FRE | MIPS_CONF5_UFE;
125 	}
126 
127 	return mask;
128 }
129 
130 static inline unsigned int kvm_vz_config6_guest_wrmask(struct kvm_vcpu *vcpu)
131 {
132 	return MIPS_CONF6_LOONGSON_INTIMER | MIPS_CONF6_LOONGSON_EXTIMER;
133 }
134 
135 /*
136  * VZ optionally allows these additional Config bits to be written by root:
137  * Config:	M, [MT]
138  * Config1:	M, [MMUSize-1, C2, MD, PC, WR, CA], FP
139  * Config2:	M
140  * Config3:	M, MSAP, [BPG], ULRI, [DSP2P, DSPP], CTXTC, [ITL, LPA, VEIC,
141  *		VInt, SP, CDMM, MT, SM, TL]
142  * Config4:	M, [VTLBSizeExt, MMUSizeExt]
143  * Config5:	MRP
144  */
145 
146 static inline unsigned int kvm_vz_config_user_wrmask(struct kvm_vcpu *vcpu)
147 {
148 	return kvm_vz_config_guest_wrmask(vcpu) | MIPS_CONF_M;
149 }
150 
151 static inline unsigned int kvm_vz_config1_user_wrmask(struct kvm_vcpu *vcpu)
152 {
153 	unsigned int mask = kvm_vz_config1_guest_wrmask(vcpu) | MIPS_CONF_M;
154 
155 	/* Permit FPU to be present if FPU is supported */
156 	if (kvm_mips_guest_can_have_fpu(&vcpu->arch))
157 		mask |= MIPS_CONF1_FP;
158 
159 	return mask;
160 }
161 
162 static inline unsigned int kvm_vz_config2_user_wrmask(struct kvm_vcpu *vcpu)
163 {
164 	return kvm_vz_config2_guest_wrmask(vcpu) | MIPS_CONF_M;
165 }
166 
167 static inline unsigned int kvm_vz_config3_user_wrmask(struct kvm_vcpu *vcpu)
168 {
169 	unsigned int mask = kvm_vz_config3_guest_wrmask(vcpu) | MIPS_CONF_M |
170 		MIPS_CONF3_ULRI | MIPS_CONF3_CTXTC;
171 
172 	/* Permit MSA to be present if MSA is supported */
173 	if (kvm_mips_guest_can_have_msa(&vcpu->arch))
174 		mask |= MIPS_CONF3_MSA;
175 
176 	return mask;
177 }
178 
179 static inline unsigned int kvm_vz_config4_user_wrmask(struct kvm_vcpu *vcpu)
180 {
181 	return kvm_vz_config4_guest_wrmask(vcpu) | MIPS_CONF_M;
182 }
183 
184 static inline unsigned int kvm_vz_config5_user_wrmask(struct kvm_vcpu *vcpu)
185 {
186 	return kvm_vz_config5_guest_wrmask(vcpu) | MIPS_CONF5_MRP;
187 }
188 
189 static inline unsigned int kvm_vz_config6_user_wrmask(struct kvm_vcpu *vcpu)
190 {
191 	return kvm_vz_config6_guest_wrmask(vcpu) |
192 		MIPS_CONF6_LOONGSON_SFBEN | MIPS_CONF6_LOONGSON_FTLBDIS;
193 }
194 
195 static gpa_t kvm_vz_gva_to_gpa_cb(gva_t gva)
196 {
197 	/* VZ guest has already converted gva to gpa */
198 	return gva;
199 }
200 
201 static void kvm_vz_queue_irq(struct kvm_vcpu *vcpu, unsigned int priority)
202 {
203 	set_bit(priority, &vcpu->arch.pending_exceptions);
204 	clear_bit(priority, &vcpu->arch.pending_exceptions_clr);
205 }
206 
207 static void kvm_vz_dequeue_irq(struct kvm_vcpu *vcpu, unsigned int priority)
208 {
209 	clear_bit(priority, &vcpu->arch.pending_exceptions);
210 	set_bit(priority, &vcpu->arch.pending_exceptions_clr);
211 }
212 
213 static void kvm_vz_queue_timer_int_cb(struct kvm_vcpu *vcpu)
214 {
215 	/*
216 	 * timer expiry is asynchronous to vcpu execution therefore defer guest
217 	 * cp0 accesses
218 	 */
219 	kvm_vz_queue_irq(vcpu, MIPS_EXC_INT_TIMER);
220 }
221 
222 static void kvm_vz_dequeue_timer_int_cb(struct kvm_vcpu *vcpu)
223 {
224 	/*
225 	 * timer expiry is asynchronous to vcpu execution therefore defer guest
226 	 * cp0 accesses
227 	 */
228 	kvm_vz_dequeue_irq(vcpu, MIPS_EXC_INT_TIMER);
229 }
230 
231 static void kvm_vz_queue_io_int_cb(struct kvm_vcpu *vcpu,
232 				   struct kvm_mips_interrupt *irq)
233 {
234 	int intr = (int)irq->irq;
235 
236 	/*
237 	 * interrupts are asynchronous to vcpu execution therefore defer guest
238 	 * cp0 accesses
239 	 */
240 	kvm_vz_queue_irq(vcpu, kvm_irq_to_priority(intr));
241 }
242 
243 static void kvm_vz_dequeue_io_int_cb(struct kvm_vcpu *vcpu,
244 				     struct kvm_mips_interrupt *irq)
245 {
246 	int intr = (int)irq->irq;
247 
248 	/*
249 	 * interrupts are asynchronous to vcpu execution therefore defer guest
250 	 * cp0 accesses
251 	 */
252 	kvm_vz_dequeue_irq(vcpu, kvm_irq_to_priority(-intr));
253 }
254 
255 static int kvm_vz_irq_deliver_cb(struct kvm_vcpu *vcpu, unsigned int priority,
256 				 u32 cause)
257 {
258 	u32 irq = (priority < MIPS_EXC_MAX) ?
259 		kvm_priority_to_irq[priority] : 0;
260 
261 	switch (priority) {
262 	case MIPS_EXC_INT_TIMER:
263 		set_gc0_cause(C_TI);
264 		break;
265 
266 	case MIPS_EXC_INT_IO_1:
267 	case MIPS_EXC_INT_IO_2:
268 	case MIPS_EXC_INT_IPI_1:
269 	case MIPS_EXC_INT_IPI_2:
270 		if (cpu_has_guestctl2)
271 			set_c0_guestctl2(irq);
272 		else
273 			set_gc0_cause(irq);
274 		break;
275 
276 	default:
277 		break;
278 	}
279 
280 	clear_bit(priority, &vcpu->arch.pending_exceptions);
281 	return 1;
282 }
283 
284 static int kvm_vz_irq_clear_cb(struct kvm_vcpu *vcpu, unsigned int priority,
285 			       u32 cause)
286 {
287 	u32 irq = (priority < MIPS_EXC_MAX) ?
288 		kvm_priority_to_irq[priority] : 0;
289 
290 	switch (priority) {
291 	case MIPS_EXC_INT_TIMER:
292 		/*
293 		 * Call to kvm_write_c0_guest_compare() clears Cause.TI in
294 		 * kvm_mips_emulate_CP0(). Explicitly clear irq associated with
295 		 * Cause.IP[IPTI] if GuestCtl2 virtual interrupt register not
296 		 * supported or if not using GuestCtl2 Hardware Clear.
297 		 */
298 		if (cpu_has_guestctl2) {
299 			if (!(read_c0_guestctl2() & (irq << 14)))
300 				clear_c0_guestctl2(irq);
301 		} else {
302 			clear_gc0_cause(irq);
303 		}
304 		break;
305 
306 	case MIPS_EXC_INT_IO_1:
307 	case MIPS_EXC_INT_IO_2:
308 	case MIPS_EXC_INT_IPI_1:
309 	case MIPS_EXC_INT_IPI_2:
310 		/* Clear GuestCtl2.VIP irq if not using Hardware Clear */
311 		if (cpu_has_guestctl2) {
312 			if (!(read_c0_guestctl2() & (irq << 14)))
313 				clear_c0_guestctl2(irq);
314 		} else {
315 			clear_gc0_cause(irq);
316 		}
317 		break;
318 
319 	default:
320 		break;
321 	}
322 
323 	clear_bit(priority, &vcpu->arch.pending_exceptions_clr);
324 	return 1;
325 }
326 
327 /*
328  * VZ guest timer handling.
329  */
330 
331 /**
332  * kvm_vz_should_use_htimer() - Find whether to use the VZ hard guest timer.
333  * @vcpu:	Virtual CPU.
334  *
335  * Returns:	true if the VZ GTOffset & real guest CP0_Count should be used
336  *		instead of software emulation of guest timer.
337  *		false otherwise.
338  */
339 static bool kvm_vz_should_use_htimer(struct kvm_vcpu *vcpu)
340 {
341 	if (kvm_mips_count_disabled(vcpu))
342 		return false;
343 
344 	/* Chosen frequency must match real frequency */
345 	if (mips_hpt_frequency != vcpu->arch.count_hz)
346 		return false;
347 
348 	/* We don't support a CP0_GTOffset with fewer bits than CP0_Count */
349 	if (current_cpu_data.gtoffset_mask != 0xffffffff)
350 		return false;
351 
352 	return true;
353 }
354 
355 /**
356  * _kvm_vz_restore_stimer() - Restore soft timer state.
357  * @vcpu:	Virtual CPU.
358  * @compare:	CP0_Compare register value, restored by caller.
359  * @cause:	CP0_Cause register to restore.
360  *
361  * Restore VZ state relating to the soft timer. The hard timer can be enabled
362  * later.
363  */
364 static void _kvm_vz_restore_stimer(struct kvm_vcpu *vcpu, u32 compare,
365 				   u32 cause)
366 {
367 	/*
368 	 * Avoid spurious counter interrupts by setting Guest CP0_Count to just
369 	 * after Guest CP0_Compare.
370 	 */
371 	write_c0_gtoffset(compare - read_c0_count());
372 
373 	back_to_back_c0_hazard();
374 	write_gc0_cause(cause);
375 }
376 
377 /**
378  * _kvm_vz_restore_htimer() - Restore hard timer state.
379  * @vcpu:	Virtual CPU.
380  * @compare:	CP0_Compare register value, restored by caller.
381  * @cause:	CP0_Cause register to restore.
382  *
383  * Restore hard timer Guest.Count & Guest.Cause taking care to preserve the
384  * value of Guest.CP0_Cause.TI while restoring Guest.CP0_Cause.
385  */
386 static void _kvm_vz_restore_htimer(struct kvm_vcpu *vcpu,
387 				   u32 compare, u32 cause)
388 {
389 	u32 start_count, after_count;
390 	ktime_t freeze_time;
391 	unsigned long flags;
392 
393 	/*
394 	 * Freeze the soft-timer and sync the guest CP0_Count with it. We do
395 	 * this with interrupts disabled to avoid latency.
396 	 */
397 	local_irq_save(flags);
398 	freeze_time = kvm_mips_freeze_hrtimer(vcpu, &start_count);
399 	write_c0_gtoffset(start_count - read_c0_count());
400 	local_irq_restore(flags);
401 
402 	/* restore guest CP0_Cause, as TI may already be set */
403 	back_to_back_c0_hazard();
404 	write_gc0_cause(cause);
405 
406 	/*
407 	 * The above sequence isn't atomic and would result in lost timer
408 	 * interrupts if we're not careful. Detect if a timer interrupt is due
409 	 * and assert it.
410 	 */
411 	back_to_back_c0_hazard();
412 	after_count = read_gc0_count();
413 	if (after_count - start_count > compare - start_count - 1)
414 		kvm_vz_queue_irq(vcpu, MIPS_EXC_INT_TIMER);
415 }
416 
417 /**
418  * kvm_vz_restore_timer() - Restore timer state.
419  * @vcpu:	Virtual CPU.
420  *
421  * Restore soft timer state from saved context.
422  */
423 static void kvm_vz_restore_timer(struct kvm_vcpu *vcpu)
424 {
425 	struct mips_coproc *cop0 = vcpu->arch.cop0;
426 	u32 cause, compare;
427 
428 	compare = kvm_read_sw_gc0_compare(cop0);
429 	cause = kvm_read_sw_gc0_cause(cop0);
430 
431 	write_gc0_compare(compare);
432 	_kvm_vz_restore_stimer(vcpu, compare, cause);
433 }
434 
435 /**
436  * kvm_vz_acquire_htimer() - Switch to hard timer state.
437  * @vcpu:	Virtual CPU.
438  *
439  * Restore hard timer state on top of existing soft timer state if possible.
440  *
441  * Since hard timer won't remain active over preemption, preemption should be
442  * disabled by the caller.
443  */
444 void kvm_vz_acquire_htimer(struct kvm_vcpu *vcpu)
445 {
446 	u32 gctl0;
447 
448 	gctl0 = read_c0_guestctl0();
449 	if (!(gctl0 & MIPS_GCTL0_GT) && kvm_vz_should_use_htimer(vcpu)) {
450 		/* enable guest access to hard timer */
451 		write_c0_guestctl0(gctl0 | MIPS_GCTL0_GT);
452 
453 		_kvm_vz_restore_htimer(vcpu, read_gc0_compare(),
454 				       read_gc0_cause());
455 	}
456 }
457 
458 /**
459  * _kvm_vz_save_htimer() - Switch to software emulation of guest timer.
460  * @vcpu:	Virtual CPU.
461  * @compare:	Pointer to write compare value to.
462  * @cause:	Pointer to write cause value to.
463  *
464  * Save VZ guest timer state and switch to software emulation of guest CP0
465  * timer. The hard timer must already be in use, so preemption should be
466  * disabled.
467  */
468 static void _kvm_vz_save_htimer(struct kvm_vcpu *vcpu,
469 				u32 *out_compare, u32 *out_cause)
470 {
471 	u32 cause, compare, before_count, end_count;
472 	ktime_t before_time;
473 
474 	compare = read_gc0_compare();
475 	*out_compare = compare;
476 
477 	before_time = ktime_get();
478 
479 	/*
480 	 * Record the CP0_Count *prior* to saving CP0_Cause, so we have a time
481 	 * at which no pending timer interrupt is missing.
482 	 */
483 	before_count = read_gc0_count();
484 	back_to_back_c0_hazard();
485 	cause = read_gc0_cause();
486 	*out_cause = cause;
487 
488 	/*
489 	 * Record a final CP0_Count which we will transfer to the soft-timer.
490 	 * This is recorded *after* saving CP0_Cause, so we don't get any timer
491 	 * interrupts from just after the final CP0_Count point.
492 	 */
493 	back_to_back_c0_hazard();
494 	end_count = read_gc0_count();
495 
496 	/*
497 	 * The above sequence isn't atomic, so we could miss a timer interrupt
498 	 * between reading CP0_Cause and end_count. Detect and record any timer
499 	 * interrupt due between before_count and end_count.
500 	 */
501 	if (end_count - before_count > compare - before_count - 1)
502 		kvm_vz_queue_irq(vcpu, MIPS_EXC_INT_TIMER);
503 
504 	/*
505 	 * Restore soft-timer, ignoring a small amount of negative drift due to
506 	 * delay between freeze_hrtimer and setting CP0_GTOffset.
507 	 */
508 	kvm_mips_restore_hrtimer(vcpu, before_time, end_count, -0x10000);
509 }
510 
511 /**
512  * kvm_vz_save_timer() - Save guest timer state.
513  * @vcpu:	Virtual CPU.
514  *
515  * Save VZ guest timer state and switch to soft guest timer if hard timer was in
516  * use.
517  */
518 static void kvm_vz_save_timer(struct kvm_vcpu *vcpu)
519 {
520 	struct mips_coproc *cop0 = vcpu->arch.cop0;
521 	u32 gctl0, compare, cause;
522 
523 	gctl0 = read_c0_guestctl0();
524 	if (gctl0 & MIPS_GCTL0_GT) {
525 		/* disable guest use of hard timer */
526 		write_c0_guestctl0(gctl0 & ~MIPS_GCTL0_GT);
527 
528 		/* save hard timer state */
529 		_kvm_vz_save_htimer(vcpu, &compare, &cause);
530 	} else {
531 		compare = read_gc0_compare();
532 		cause = read_gc0_cause();
533 	}
534 
535 	/* save timer-related state to VCPU context */
536 	kvm_write_sw_gc0_cause(cop0, cause);
537 	kvm_write_sw_gc0_compare(cop0, compare);
538 }
539 
540 /**
541  * kvm_vz_lose_htimer() - Ensure hard guest timer is not in use.
542  * @vcpu:	Virtual CPU.
543  *
544  * Transfers the state of the hard guest timer to the soft guest timer, leaving
545  * guest state intact so it can continue to be used with the soft timer.
546  */
547 void kvm_vz_lose_htimer(struct kvm_vcpu *vcpu)
548 {
549 	u32 gctl0, compare, cause;
550 
551 	preempt_disable();
552 	gctl0 = read_c0_guestctl0();
553 	if (gctl0 & MIPS_GCTL0_GT) {
554 		/* disable guest use of timer */
555 		write_c0_guestctl0(gctl0 & ~MIPS_GCTL0_GT);
556 
557 		/* switch to soft timer */
558 		_kvm_vz_save_htimer(vcpu, &compare, &cause);
559 
560 		/* leave soft timer in usable state */
561 		_kvm_vz_restore_stimer(vcpu, compare, cause);
562 	}
563 	preempt_enable();
564 }
565 
566 /**
567  * is_eva_access() - Find whether an instruction is an EVA memory accessor.
568  * @inst:	32-bit instruction encoding.
569  *
570  * Finds whether @inst encodes an EVA memory access instruction, which would
571  * indicate that emulation of it should access the user mode address space
572  * instead of the kernel mode address space. This matters for MUSUK segments
573  * which are TLB mapped for user mode but unmapped for kernel mode.
574  *
575  * Returns:	Whether @inst encodes an EVA accessor instruction.
576  */
577 static bool is_eva_access(union mips_instruction inst)
578 {
579 	if (inst.spec3_format.opcode != spec3_op)
580 		return false;
581 
582 	switch (inst.spec3_format.func) {
583 	case lwle_op:
584 	case lwre_op:
585 	case cachee_op:
586 	case sbe_op:
587 	case she_op:
588 	case sce_op:
589 	case swe_op:
590 	case swle_op:
591 	case swre_op:
592 	case prefe_op:
593 	case lbue_op:
594 	case lhue_op:
595 	case lbe_op:
596 	case lhe_op:
597 	case lle_op:
598 	case lwe_op:
599 		return true;
600 	default:
601 		return false;
602 	}
603 }
604 
605 /**
606  * is_eva_am_mapped() - Find whether an access mode is mapped.
607  * @vcpu:	KVM VCPU state.
608  * @am:		3-bit encoded access mode.
609  * @eu:		Segment becomes unmapped and uncached when Status.ERL=1.
610  *
611  * Decode @am to find whether it encodes a mapped segment for the current VCPU
612  * state. Where necessary @eu and the actual instruction causing the fault are
613  * taken into account to make the decision.
614  *
615  * Returns:	Whether the VCPU faulted on a TLB mapped address.
616  */
617 static bool is_eva_am_mapped(struct kvm_vcpu *vcpu, unsigned int am, bool eu)
618 {
619 	u32 am_lookup;
620 	int err;
621 
622 	/*
623 	 * Interpret access control mode. We assume address errors will already
624 	 * have been caught by the guest, leaving us with:
625 	 *      AM      UM  SM  KM  31..24 23..16
626 	 * UK    0 000          Unm   0      0
627 	 * MK    1 001          TLB   1
628 	 * MSK   2 010      TLB TLB   1
629 	 * MUSK  3 011  TLB TLB TLB   1
630 	 * MUSUK 4 100  TLB TLB Unm   0      1
631 	 * USK   5 101      Unm Unm   0      0
632 	 * -     6 110                0      0
633 	 * UUSK  7 111  Unm Unm Unm   0      0
634 	 *
635 	 * We shift a magic value by AM across the sign bit to find if always
636 	 * TLB mapped, and if not shift by 8 again to find if it depends on KM.
637 	 */
638 	am_lookup = 0x70080000 << am;
639 	if ((s32)am_lookup < 0) {
640 		/*
641 		 * MK, MSK, MUSK
642 		 * Always TLB mapped, unless SegCtl.EU && ERL
643 		 */
644 		if (!eu || !(read_gc0_status() & ST0_ERL))
645 			return true;
646 	} else {
647 		am_lookup <<= 8;
648 		if ((s32)am_lookup < 0) {
649 			union mips_instruction inst;
650 			unsigned int status;
651 			u32 *opc;
652 
653 			/*
654 			 * MUSUK
655 			 * TLB mapped if not in kernel mode
656 			 */
657 			status = read_gc0_status();
658 			if (!(status & (ST0_EXL | ST0_ERL)) &&
659 			    (status & ST0_KSU))
660 				return true;
661 			/*
662 			 * EVA access instructions in kernel
663 			 * mode access user address space.
664 			 */
665 			opc = (u32 *)vcpu->arch.pc;
666 			if (vcpu->arch.host_cp0_cause & CAUSEF_BD)
667 				opc += 1;
668 			err = kvm_get_badinstr(opc, vcpu, &inst.word);
669 			if (!err && is_eva_access(inst))
670 				return true;
671 		}
672 	}
673 
674 	return false;
675 }
676 
677 /**
678  * kvm_vz_gva_to_gpa() - Convert valid GVA to GPA.
679  * @vcpu:	KVM VCPU state.
680  * @gva:	Guest virtual address to convert.
681  * @gpa:	Output guest physical address.
682  *
683  * Convert a guest virtual address (GVA) which is valid according to the guest
684  * context, to a guest physical address (GPA).
685  *
686  * Returns:	0 on success.
687  *		-errno on failure.
688  */
689 static int kvm_vz_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
690 			     unsigned long *gpa)
691 {
692 	u32 gva32 = gva;
693 	unsigned long segctl;
694 
695 	if ((long)gva == (s32)gva32) {
696 		/* Handle canonical 32-bit virtual address */
697 		if (cpu_guest_has_segments) {
698 			unsigned long mask, pa;
699 
700 			switch (gva32 >> 29) {
701 			case 0:
702 			case 1: /* CFG5 (1GB) */
703 				segctl = read_gc0_segctl2() >> 16;
704 				mask = (unsigned long)0xfc0000000ull;
705 				break;
706 			case 2:
707 			case 3: /* CFG4 (1GB) */
708 				segctl = read_gc0_segctl2();
709 				mask = (unsigned long)0xfc0000000ull;
710 				break;
711 			case 4: /* CFG3 (512MB) */
712 				segctl = read_gc0_segctl1() >> 16;
713 				mask = (unsigned long)0xfe0000000ull;
714 				break;
715 			case 5: /* CFG2 (512MB) */
716 				segctl = read_gc0_segctl1();
717 				mask = (unsigned long)0xfe0000000ull;
718 				break;
719 			case 6: /* CFG1 (512MB) */
720 				segctl = read_gc0_segctl0() >> 16;
721 				mask = (unsigned long)0xfe0000000ull;
722 				break;
723 			case 7: /* CFG0 (512MB) */
724 				segctl = read_gc0_segctl0();
725 				mask = (unsigned long)0xfe0000000ull;
726 				break;
727 			default:
728 				/*
729 				 * GCC 4.9 isn't smart enough to figure out that
730 				 * segctl and mask are always initialised.
731 				 */
732 				unreachable();
733 			}
734 
735 			if (is_eva_am_mapped(vcpu, (segctl >> 4) & 0x7,
736 					     segctl & 0x0008))
737 				goto tlb_mapped;
738 
739 			/* Unmapped, find guest physical address */
740 			pa = (segctl << 20) & mask;
741 			pa |= gva32 & ~mask;
742 			*gpa = pa;
743 			return 0;
744 		} else if ((s32)gva32 < (s32)0xc0000000) {
745 			/* legacy unmapped KSeg0 or KSeg1 */
746 			*gpa = gva32 & 0x1fffffff;
747 			return 0;
748 		}
749 #ifdef CONFIG_64BIT
750 	} else if ((gva & 0xc000000000000000) == 0x8000000000000000) {
751 		/* XKPHYS */
752 		if (cpu_guest_has_segments) {
753 			/*
754 			 * Each of the 8 regions can be overridden by SegCtl2.XR
755 			 * to use SegCtl1.XAM.
756 			 */
757 			segctl = read_gc0_segctl2();
758 			if (segctl & (1ull << (56 + ((gva >> 59) & 0x7)))) {
759 				segctl = read_gc0_segctl1();
760 				if (is_eva_am_mapped(vcpu, (segctl >> 59) & 0x7,
761 						     0))
762 					goto tlb_mapped;
763 			}
764 
765 		}
766 		/*
767 		 * Traditionally fully unmapped.
768 		 * Bits 61:59 specify the CCA, which we can just mask off here.
769 		 * Bits 58:PABITS should be zero, but we shouldn't have got here
770 		 * if it wasn't.
771 		 */
772 		*gpa = gva & 0x07ffffffffffffff;
773 		return 0;
774 #endif
775 	}
776 
777 tlb_mapped:
778 	return kvm_vz_guest_tlb_lookup(vcpu, gva, gpa);
779 }
780 
781 /**
782  * kvm_vz_badvaddr_to_gpa() - Convert GVA BadVAddr from root exception to GPA.
783  * @vcpu:	KVM VCPU state.
784  * @badvaddr:	Root BadVAddr.
785  * @gpa:	Output guest physical address.
786  *
787  * VZ implementations are permitted to report guest virtual addresses (GVA) in
788  * BadVAddr on a root exception during guest execution, instead of the more
789  * convenient guest physical addresses (GPA). When we get a GVA, this function
790  * converts it to a GPA, taking into account guest segmentation and guest TLB
791  * state.
792  *
793  * Returns:	0 on success.
794  *		-errno on failure.
795  */
796 static int kvm_vz_badvaddr_to_gpa(struct kvm_vcpu *vcpu, unsigned long badvaddr,
797 				  unsigned long *gpa)
798 {
799 	unsigned int gexccode = (vcpu->arch.host_cp0_guestctl0 &
800 				 MIPS_GCTL0_GEXC) >> MIPS_GCTL0_GEXC_SHIFT;
801 
802 	/* If BadVAddr is GPA, then all is well in the world */
803 	if (likely(gexccode == MIPS_GCTL0_GEXC_GPA)) {
804 		*gpa = badvaddr;
805 		return 0;
806 	}
807 
808 	/* Otherwise we'd expect it to be GVA ... */
809 	if (WARN(gexccode != MIPS_GCTL0_GEXC_GVA,
810 		 "Unexpected gexccode %#x\n", gexccode))
811 		return -EINVAL;
812 
813 	/* ... and we need to perform the GVA->GPA translation in software */
814 	return kvm_vz_gva_to_gpa(vcpu, badvaddr, gpa);
815 }
816 
817 static int kvm_trap_vz_no_handler(struct kvm_vcpu *vcpu)
818 {
819 	u32 *opc = (u32 *) vcpu->arch.pc;
820 	u32 cause = vcpu->arch.host_cp0_cause;
821 	u32 exccode = (cause & CAUSEF_EXCCODE) >> CAUSEB_EXCCODE;
822 	unsigned long badvaddr = vcpu->arch.host_cp0_badvaddr;
823 	u32 inst = 0;
824 
825 	/*
826 	 *  Fetch the instruction.
827 	 */
828 	if (cause & CAUSEF_BD)
829 		opc += 1;
830 	kvm_get_badinstr(opc, vcpu, &inst);
831 
832 	kvm_err("Exception Code: %d not handled @ PC: %p, inst: 0x%08x BadVaddr: %#lx Status: %#x\n",
833 		exccode, opc, inst, badvaddr,
834 		read_gc0_status());
835 	kvm_arch_vcpu_dump_regs(vcpu);
836 	vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
837 	return RESUME_HOST;
838 }
839 
840 static unsigned long mips_process_maar(unsigned int op, unsigned long val)
841 {
842 	/* Mask off unused bits */
843 	unsigned long mask = 0xfffff000 | MIPS_MAAR_S | MIPS_MAAR_VL;
844 
845 	if (read_gc0_pagegrain() & PG_ELPA)
846 		mask |= 0x00ffffff00000000ull;
847 	if (cpu_guest_has_mvh)
848 		mask |= MIPS_MAAR_VH;
849 
850 	/* Set or clear VH */
851 	if (op == mtc_op) {
852 		/* clear VH */
853 		val &= ~MIPS_MAAR_VH;
854 	} else if (op == dmtc_op) {
855 		/* set VH to match VL */
856 		val &= ~MIPS_MAAR_VH;
857 		if (val & MIPS_MAAR_VL)
858 			val |= MIPS_MAAR_VH;
859 	}
860 
861 	return val & mask;
862 }
863 
864 static void kvm_write_maari(struct kvm_vcpu *vcpu, unsigned long val)
865 {
866 	struct mips_coproc *cop0 = vcpu->arch.cop0;
867 
868 	val &= MIPS_MAARI_INDEX;
869 	if (val == MIPS_MAARI_INDEX)
870 		kvm_write_sw_gc0_maari(cop0, ARRAY_SIZE(vcpu->arch.maar) - 1);
871 	else if (val < ARRAY_SIZE(vcpu->arch.maar))
872 		kvm_write_sw_gc0_maari(cop0, val);
873 }
874 
875 static enum emulation_result kvm_vz_gpsi_cop0(union mips_instruction inst,
876 					      u32 *opc, u32 cause,
877 					      struct kvm_run *run,
878 					      struct kvm_vcpu *vcpu)
879 {
880 	struct mips_coproc *cop0 = vcpu->arch.cop0;
881 	enum emulation_result er = EMULATE_DONE;
882 	u32 rt, rd, sel;
883 	unsigned long curr_pc;
884 	unsigned long val;
885 
886 	/*
887 	 * Update PC and hold onto current PC in case there is
888 	 * an error and we want to rollback the PC
889 	 */
890 	curr_pc = vcpu->arch.pc;
891 	er = update_pc(vcpu, cause);
892 	if (er == EMULATE_FAIL)
893 		return er;
894 
895 	if (inst.co_format.co) {
896 		switch (inst.co_format.func) {
897 		case wait_op:
898 			er = kvm_mips_emul_wait(vcpu);
899 			break;
900 		default:
901 			er = EMULATE_FAIL;
902 		}
903 	} else {
904 		rt = inst.c0r_format.rt;
905 		rd = inst.c0r_format.rd;
906 		sel = inst.c0r_format.sel;
907 
908 		switch (inst.c0r_format.rs) {
909 		case dmfc_op:
910 		case mfc_op:
911 #ifdef CONFIG_KVM_MIPS_DEBUG_COP0_COUNTERS
912 			cop0->stat[rd][sel]++;
913 #endif
914 			if (rd == MIPS_CP0_COUNT &&
915 			    sel == 0) {			/* Count */
916 				val = kvm_mips_read_count(vcpu);
917 			} else if (rd == MIPS_CP0_COMPARE &&
918 				   sel == 0) {		/* Compare */
919 				val = read_gc0_compare();
920 			} else if (rd == MIPS_CP0_LLADDR &&
921 				   sel == 0) {		/* LLAddr */
922 				if (cpu_guest_has_rw_llb)
923 					val = read_gc0_lladdr() &
924 						MIPS_LLADDR_LLB;
925 				else
926 					val = 0;
927 			} else if (rd == MIPS_CP0_LLADDR &&
928 				   sel == 1 &&		/* MAAR */
929 				   cpu_guest_has_maar &&
930 				   !cpu_guest_has_dyn_maar) {
931 				/* MAARI must be in range */
932 				BUG_ON(kvm_read_sw_gc0_maari(cop0) >=
933 						ARRAY_SIZE(vcpu->arch.maar));
934 				val = vcpu->arch.maar[
935 					kvm_read_sw_gc0_maari(cop0)];
936 			} else if ((rd == MIPS_CP0_PRID &&
937 				    (sel == 0 ||	/* PRid */
938 				     sel == 2 ||	/* CDMMBase */
939 				     sel == 3)) ||	/* CMGCRBase */
940 				   (rd == MIPS_CP0_STATUS &&
941 				    (sel == 2 ||	/* SRSCtl */
942 				     sel == 3)) ||	/* SRSMap */
943 				   (rd == MIPS_CP0_CONFIG &&
944 				    (sel == 6 ||	/* Config6 */
945 				     sel == 7)) ||	/* Config7 */
946 				   (rd == MIPS_CP0_LLADDR &&
947 				    (sel == 2) &&	/* MAARI */
948 				    cpu_guest_has_maar &&
949 				    !cpu_guest_has_dyn_maar) ||
950 				   (rd == MIPS_CP0_ERRCTL &&
951 				    (sel == 0))) {	/* ErrCtl */
952 				val = cop0->reg[rd][sel];
953 #ifdef CONFIG_CPU_LOONGSON64
954 			} else if (rd == MIPS_CP0_DIAG &&
955 				   (sel == 0)) {	/* Diag */
956 				val = cop0->reg[rd][sel];
957 #endif
958 			} else {
959 				val = 0;
960 				er = EMULATE_FAIL;
961 			}
962 
963 			if (er != EMULATE_FAIL) {
964 				/* Sign extend */
965 				if (inst.c0r_format.rs == mfc_op)
966 					val = (int)val;
967 				vcpu->arch.gprs[rt] = val;
968 			}
969 
970 			trace_kvm_hwr(vcpu, (inst.c0r_format.rs == mfc_op) ?
971 					KVM_TRACE_MFC0 : KVM_TRACE_DMFC0,
972 				      KVM_TRACE_COP0(rd, sel), val);
973 			break;
974 
975 		case dmtc_op:
976 		case mtc_op:
977 #ifdef CONFIG_KVM_MIPS_DEBUG_COP0_COUNTERS
978 			cop0->stat[rd][sel]++;
979 #endif
980 			val = vcpu->arch.gprs[rt];
981 			trace_kvm_hwr(vcpu, (inst.c0r_format.rs == mtc_op) ?
982 					KVM_TRACE_MTC0 : KVM_TRACE_DMTC0,
983 				      KVM_TRACE_COP0(rd, sel), val);
984 
985 			if (rd == MIPS_CP0_COUNT &&
986 			    sel == 0) {			/* Count */
987 				kvm_vz_lose_htimer(vcpu);
988 				kvm_mips_write_count(vcpu, vcpu->arch.gprs[rt]);
989 			} else if (rd == MIPS_CP0_COMPARE &&
990 				   sel == 0) {		/* Compare */
991 				kvm_mips_write_compare(vcpu,
992 						       vcpu->arch.gprs[rt],
993 						       true);
994 			} else if (rd == MIPS_CP0_LLADDR &&
995 				   sel == 0) {		/* LLAddr */
996 				/*
997 				 * P5600 generates GPSI on guest MTC0 LLAddr.
998 				 * Only allow the guest to clear LLB.
999 				 */
1000 				if (cpu_guest_has_rw_llb &&
1001 				    !(val & MIPS_LLADDR_LLB))
1002 					write_gc0_lladdr(0);
1003 			} else if (rd == MIPS_CP0_LLADDR &&
1004 				   sel == 1 &&		/* MAAR */
1005 				   cpu_guest_has_maar &&
1006 				   !cpu_guest_has_dyn_maar) {
1007 				val = mips_process_maar(inst.c0r_format.rs,
1008 							val);
1009 
1010 				/* MAARI must be in range */
1011 				BUG_ON(kvm_read_sw_gc0_maari(cop0) >=
1012 						ARRAY_SIZE(vcpu->arch.maar));
1013 				vcpu->arch.maar[kvm_read_sw_gc0_maari(cop0)] =
1014 									val;
1015 			} else if (rd == MIPS_CP0_LLADDR &&
1016 				   (sel == 2) &&	/* MAARI */
1017 				   cpu_guest_has_maar &&
1018 				   !cpu_guest_has_dyn_maar) {
1019 				kvm_write_maari(vcpu, val);
1020 			} else if (rd == MIPS_CP0_CONFIG &&
1021 				   (sel == 6)) {
1022 				cop0->reg[rd][sel] = (int)val;
1023 			} else if (rd == MIPS_CP0_ERRCTL &&
1024 				   (sel == 0)) {	/* ErrCtl */
1025 				/* ignore the written value */
1026 #ifdef CONFIG_CPU_LOONGSON64
1027 			} else if (rd == MIPS_CP0_DIAG &&
1028 				   (sel == 0)) {	/* Diag */
1029 				unsigned long flags;
1030 
1031 				local_irq_save(flags);
1032 				if (val & LOONGSON_DIAG_BTB) {
1033 					/* Flush BTB */
1034 					set_c0_diag(LOONGSON_DIAG_BTB);
1035 				}
1036 				if (val & LOONGSON_DIAG_ITLB) {
1037 					/* Flush ITLB */
1038 					set_c0_diag(LOONGSON_DIAG_ITLB);
1039 				}
1040 				if (val & LOONGSON_DIAG_DTLB) {
1041 					/* Flush DTLB */
1042 					set_c0_diag(LOONGSON_DIAG_DTLB);
1043 				}
1044 				if (val & LOONGSON_DIAG_VTLB) {
1045 					/* Flush VTLB */
1046 					kvm_loongson_clear_guest_vtlb();
1047 				}
1048 				if (val & LOONGSON_DIAG_FTLB) {
1049 					/* Flush FTLB */
1050 					kvm_loongson_clear_guest_ftlb();
1051 				}
1052 				local_irq_restore(flags);
1053 #endif
1054 			} else {
1055 				er = EMULATE_FAIL;
1056 			}
1057 			break;
1058 
1059 		default:
1060 			er = EMULATE_FAIL;
1061 			break;
1062 		}
1063 	}
1064 	/* Rollback PC only if emulation was unsuccessful */
1065 	if (er == EMULATE_FAIL) {
1066 		kvm_err("[%#lx]%s: unsupported cop0 instruction 0x%08x\n",
1067 			curr_pc, __func__, inst.word);
1068 
1069 		vcpu->arch.pc = curr_pc;
1070 	}
1071 
1072 	return er;
1073 }
1074 
1075 static enum emulation_result kvm_vz_gpsi_cache(union mips_instruction inst,
1076 					       u32 *opc, u32 cause,
1077 					       struct kvm_run *run,
1078 					       struct kvm_vcpu *vcpu)
1079 {
1080 	enum emulation_result er = EMULATE_DONE;
1081 	u32 cache, op_inst, op, base;
1082 	s16 offset;
1083 	struct kvm_vcpu_arch *arch = &vcpu->arch;
1084 	unsigned long va, curr_pc;
1085 
1086 	/*
1087 	 * Update PC and hold onto current PC in case there is
1088 	 * an error and we want to rollback the PC
1089 	 */
1090 	curr_pc = vcpu->arch.pc;
1091 	er = update_pc(vcpu, cause);
1092 	if (er == EMULATE_FAIL)
1093 		return er;
1094 
1095 	base = inst.i_format.rs;
1096 	op_inst = inst.i_format.rt;
1097 	if (cpu_has_mips_r6)
1098 		offset = inst.spec3_format.simmediate;
1099 	else
1100 		offset = inst.i_format.simmediate;
1101 	cache = op_inst & CacheOp_Cache;
1102 	op = op_inst & CacheOp_Op;
1103 
1104 	va = arch->gprs[base] + offset;
1105 
1106 	kvm_debug("CACHE (cache: %#x, op: %#x, base[%d]: %#lx, offset: %#x\n",
1107 		  cache, op, base, arch->gprs[base], offset);
1108 
1109 	/* Secondary or tirtiary cache ops ignored */
1110 	if (cache != Cache_I && cache != Cache_D)
1111 		return EMULATE_DONE;
1112 
1113 	switch (op_inst) {
1114 	case Index_Invalidate_I:
1115 		flush_icache_line_indexed(va);
1116 		return EMULATE_DONE;
1117 	case Index_Writeback_Inv_D:
1118 		flush_dcache_line_indexed(va);
1119 		return EMULATE_DONE;
1120 	case Hit_Invalidate_I:
1121 	case Hit_Invalidate_D:
1122 	case Hit_Writeback_Inv_D:
1123 		if (boot_cpu_type() == CPU_CAVIUM_OCTEON3) {
1124 			/* We can just flush entire icache */
1125 			local_flush_icache_range(0, 0);
1126 			return EMULATE_DONE;
1127 		}
1128 
1129 		/* So far, other platforms support guest hit cache ops */
1130 		break;
1131 	default:
1132 		break;
1133 	}
1134 
1135 	kvm_err("@ %#lx/%#lx CACHE (cache: %#x, op: %#x, base[%d]: %#lx, offset: %#x\n",
1136 		curr_pc, vcpu->arch.gprs[31], cache, op, base, arch->gprs[base],
1137 		offset);
1138 	/* Rollback PC */
1139 	vcpu->arch.pc = curr_pc;
1140 
1141 	return EMULATE_FAIL;
1142 }
1143 
1144 #ifdef CONFIG_CPU_LOONGSON64
1145 static enum emulation_result kvm_vz_gpsi_lwc2(union mips_instruction inst,
1146 					      u32 *opc, u32 cause,
1147 					      struct kvm_run *run,
1148 					      struct kvm_vcpu *vcpu)
1149 {
1150 	unsigned int rs, rd;
1151 	unsigned int hostcfg;
1152 	unsigned long curr_pc;
1153 	enum emulation_result er = EMULATE_DONE;
1154 
1155 	/*
1156 	 * Update PC and hold onto current PC in case there is
1157 	 * an error and we want to rollback the PC
1158 	 */
1159 	curr_pc = vcpu->arch.pc;
1160 	er = update_pc(vcpu, cause);
1161 	if (er == EMULATE_FAIL)
1162 		return er;
1163 
1164 	rs = inst.loongson3_lscsr_format.rs;
1165 	rd = inst.loongson3_lscsr_format.rd;
1166 	switch (inst.loongson3_lscsr_format.fr) {
1167 	case 0x8:  /* Read CPUCFG */
1168 		++vcpu->stat.vz_cpucfg_exits;
1169 		hostcfg = read_cpucfg(vcpu->arch.gprs[rs]);
1170 
1171 		switch (vcpu->arch.gprs[rs]) {
1172 		case LOONGSON_CFG0:
1173 			vcpu->arch.gprs[rd] = 0x14c000;
1174 			break;
1175 		case LOONGSON_CFG1:
1176 			hostcfg &= (LOONGSON_CFG1_FP | LOONGSON_CFG1_MMI |
1177 				    LOONGSON_CFG1_MSA1 | LOONGSON_CFG1_MSA2 |
1178 				    LOONGSON_CFG1_SFBP);
1179 			vcpu->arch.gprs[rd] = hostcfg;
1180 			break;
1181 		case LOONGSON_CFG2:
1182 			hostcfg &= (LOONGSON_CFG2_LEXT1 | LOONGSON_CFG2_LEXT2 |
1183 				    LOONGSON_CFG2_LEXT3 | LOONGSON_CFG2_LSPW);
1184 			vcpu->arch.gprs[rd] = hostcfg;
1185 			break;
1186 		case LOONGSON_CFG3:
1187 			vcpu->arch.gprs[rd] = hostcfg;
1188 			break;
1189 		default:
1190 			/* Don't export any other advanced features to guest */
1191 			vcpu->arch.gprs[rd] = 0;
1192 			break;
1193 		}
1194 		break;
1195 
1196 	default:
1197 		kvm_err("lwc2 emulate not impl %d rs %lx @%lx\n",
1198 			inst.loongson3_lscsr_format.fr, vcpu->arch.gprs[rs], curr_pc);
1199 		er = EMULATE_FAIL;
1200 		break;
1201 	}
1202 
1203 	/* Rollback PC only if emulation was unsuccessful */
1204 	if (er == EMULATE_FAIL) {
1205 		kvm_err("[%#lx]%s: unsupported lwc2 instruction 0x%08x 0x%08x\n",
1206 			curr_pc, __func__, inst.word, inst.loongson3_lscsr_format.fr);
1207 
1208 		vcpu->arch.pc = curr_pc;
1209 	}
1210 
1211 	return er;
1212 }
1213 #endif
1214 
1215 static enum emulation_result kvm_trap_vz_handle_gpsi(u32 cause, u32 *opc,
1216 						     struct kvm_vcpu *vcpu)
1217 {
1218 	enum emulation_result er = EMULATE_DONE;
1219 	struct kvm_vcpu_arch *arch = &vcpu->arch;
1220 	struct kvm_run *run = vcpu->run;
1221 	union mips_instruction inst;
1222 	int rd, rt, sel;
1223 	int err;
1224 
1225 	/*
1226 	 *  Fetch the instruction.
1227 	 */
1228 	if (cause & CAUSEF_BD)
1229 		opc += 1;
1230 	err = kvm_get_badinstr(opc, vcpu, &inst.word);
1231 	if (err)
1232 		return EMULATE_FAIL;
1233 
1234 	switch (inst.r_format.opcode) {
1235 	case cop0_op:
1236 		er = kvm_vz_gpsi_cop0(inst, opc, cause, run, vcpu);
1237 		break;
1238 #ifndef CONFIG_CPU_MIPSR6
1239 	case cache_op:
1240 		trace_kvm_exit(vcpu, KVM_TRACE_EXIT_CACHE);
1241 		er = kvm_vz_gpsi_cache(inst, opc, cause, run, vcpu);
1242 		break;
1243 #endif
1244 #ifdef CONFIG_CPU_LOONGSON64
1245 	case lwc2_op:
1246 		er = kvm_vz_gpsi_lwc2(inst, opc, cause, run, vcpu);
1247 		break;
1248 #endif
1249 	case spec3_op:
1250 		switch (inst.spec3_format.func) {
1251 #ifdef CONFIG_CPU_MIPSR6
1252 		case cache6_op:
1253 			trace_kvm_exit(vcpu, KVM_TRACE_EXIT_CACHE);
1254 			er = kvm_vz_gpsi_cache(inst, opc, cause, run, vcpu);
1255 			break;
1256 #endif
1257 		case rdhwr_op:
1258 			if (inst.r_format.rs || (inst.r_format.re >> 3))
1259 				goto unknown;
1260 
1261 			rd = inst.r_format.rd;
1262 			rt = inst.r_format.rt;
1263 			sel = inst.r_format.re & 0x7;
1264 
1265 			switch (rd) {
1266 			case MIPS_HWR_CC:	/* Read count register */
1267 				arch->gprs[rt] =
1268 					(long)(int)kvm_mips_read_count(vcpu);
1269 				break;
1270 			default:
1271 				trace_kvm_hwr(vcpu, KVM_TRACE_RDHWR,
1272 					      KVM_TRACE_HWR(rd, sel), 0);
1273 				goto unknown;
1274 			}
1275 
1276 			trace_kvm_hwr(vcpu, KVM_TRACE_RDHWR,
1277 				      KVM_TRACE_HWR(rd, sel), arch->gprs[rt]);
1278 
1279 			er = update_pc(vcpu, cause);
1280 			break;
1281 		default:
1282 			goto unknown;
1283 		}
1284 		break;
1285 unknown:
1286 
1287 	default:
1288 		kvm_err("GPSI exception not supported (%p/%#x)\n",
1289 				opc, inst.word);
1290 		kvm_arch_vcpu_dump_regs(vcpu);
1291 		er = EMULATE_FAIL;
1292 		break;
1293 	}
1294 
1295 	return er;
1296 }
1297 
1298 static enum emulation_result kvm_trap_vz_handle_gsfc(u32 cause, u32 *opc,
1299 						     struct kvm_vcpu *vcpu)
1300 {
1301 	enum emulation_result er = EMULATE_DONE;
1302 	struct kvm_vcpu_arch *arch = &vcpu->arch;
1303 	union mips_instruction inst;
1304 	int err;
1305 
1306 	/*
1307 	 *  Fetch the instruction.
1308 	 */
1309 	if (cause & CAUSEF_BD)
1310 		opc += 1;
1311 	err = kvm_get_badinstr(opc, vcpu, &inst.word);
1312 	if (err)
1313 		return EMULATE_FAIL;
1314 
1315 	/* complete MTC0 on behalf of guest and advance EPC */
1316 	if (inst.c0r_format.opcode == cop0_op &&
1317 	    inst.c0r_format.rs == mtc_op &&
1318 	    inst.c0r_format.z == 0) {
1319 		int rt = inst.c0r_format.rt;
1320 		int rd = inst.c0r_format.rd;
1321 		int sel = inst.c0r_format.sel;
1322 		unsigned int val = arch->gprs[rt];
1323 		unsigned int old_val, change;
1324 
1325 		trace_kvm_hwr(vcpu, KVM_TRACE_MTC0, KVM_TRACE_COP0(rd, sel),
1326 			      val);
1327 
1328 		if ((rd == MIPS_CP0_STATUS) && (sel == 0)) {
1329 			/* FR bit should read as zero if no FPU */
1330 			if (!kvm_mips_guest_has_fpu(&vcpu->arch))
1331 				val &= ~(ST0_CU1 | ST0_FR);
1332 
1333 			/*
1334 			 * Also don't allow FR to be set if host doesn't support
1335 			 * it.
1336 			 */
1337 			if (!(boot_cpu_data.fpu_id & MIPS_FPIR_F64))
1338 				val &= ~ST0_FR;
1339 
1340 			old_val = read_gc0_status();
1341 			change = val ^ old_val;
1342 
1343 			if (change & ST0_FR) {
1344 				/*
1345 				 * FPU and Vector register state is made
1346 				 * UNPREDICTABLE by a change of FR, so don't
1347 				 * even bother saving it.
1348 				 */
1349 				kvm_drop_fpu(vcpu);
1350 			}
1351 
1352 			/*
1353 			 * If MSA state is already live, it is undefined how it
1354 			 * interacts with FR=0 FPU state, and we don't want to
1355 			 * hit reserved instruction exceptions trying to save
1356 			 * the MSA state later when CU=1 && FR=1, so play it
1357 			 * safe and save it first.
1358 			 */
1359 			if (change & ST0_CU1 && !(val & ST0_FR) &&
1360 			    vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA)
1361 				kvm_lose_fpu(vcpu);
1362 
1363 			write_gc0_status(val);
1364 		} else if ((rd == MIPS_CP0_CAUSE) && (sel == 0)) {
1365 			u32 old_cause = read_gc0_cause();
1366 			u32 change = old_cause ^ val;
1367 
1368 			/* DC bit enabling/disabling timer? */
1369 			if (change & CAUSEF_DC) {
1370 				if (val & CAUSEF_DC) {
1371 					kvm_vz_lose_htimer(vcpu);
1372 					kvm_mips_count_disable_cause(vcpu);
1373 				} else {
1374 					kvm_mips_count_enable_cause(vcpu);
1375 				}
1376 			}
1377 
1378 			/* Only certain bits are RW to the guest */
1379 			change &= (CAUSEF_DC | CAUSEF_IV | CAUSEF_WP |
1380 				   CAUSEF_IP0 | CAUSEF_IP1);
1381 
1382 			/* WP can only be cleared */
1383 			change &= ~CAUSEF_WP | old_cause;
1384 
1385 			write_gc0_cause(old_cause ^ change);
1386 		} else if ((rd == MIPS_CP0_STATUS) && (sel == 1)) { /* IntCtl */
1387 			write_gc0_intctl(val);
1388 		} else if ((rd == MIPS_CP0_CONFIG) && (sel == 5)) {
1389 			old_val = read_gc0_config5();
1390 			change = val ^ old_val;
1391 			/* Handle changes in FPU/MSA modes */
1392 			preempt_disable();
1393 
1394 			/*
1395 			 * Propagate FRE changes immediately if the FPU
1396 			 * context is already loaded.
1397 			 */
1398 			if (change & MIPS_CONF5_FRE &&
1399 			    vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU)
1400 				change_c0_config5(MIPS_CONF5_FRE, val);
1401 
1402 			preempt_enable();
1403 
1404 			val = old_val ^
1405 				(change & kvm_vz_config5_guest_wrmask(vcpu));
1406 			write_gc0_config5(val);
1407 		} else {
1408 			kvm_err("Handle GSFC, unsupported field change @ %p: %#x\n",
1409 			    opc, inst.word);
1410 			er = EMULATE_FAIL;
1411 		}
1412 
1413 		if (er != EMULATE_FAIL)
1414 			er = update_pc(vcpu, cause);
1415 	} else {
1416 		kvm_err("Handle GSFC, unrecognized instruction @ %p: %#x\n",
1417 			opc, inst.word);
1418 		er = EMULATE_FAIL;
1419 	}
1420 
1421 	return er;
1422 }
1423 
1424 static enum emulation_result kvm_trap_vz_handle_ghfc(u32 cause, u32 *opc,
1425 						     struct kvm_vcpu *vcpu)
1426 {
1427 	/*
1428 	 * Presumably this is due to MC (guest mode change), so lets trace some
1429 	 * relevant info.
1430 	 */
1431 	trace_kvm_guest_mode_change(vcpu);
1432 
1433 	return EMULATE_DONE;
1434 }
1435 
1436 static enum emulation_result kvm_trap_vz_handle_hc(u32 cause, u32 *opc,
1437 						   struct kvm_vcpu *vcpu)
1438 {
1439 	enum emulation_result er;
1440 	union mips_instruction inst;
1441 	unsigned long curr_pc;
1442 	int err;
1443 
1444 	if (cause & CAUSEF_BD)
1445 		opc += 1;
1446 	err = kvm_get_badinstr(opc, vcpu, &inst.word);
1447 	if (err)
1448 		return EMULATE_FAIL;
1449 
1450 	/*
1451 	 * Update PC and hold onto current PC in case there is
1452 	 * an error and we want to rollback the PC
1453 	 */
1454 	curr_pc = vcpu->arch.pc;
1455 	er = update_pc(vcpu, cause);
1456 	if (er == EMULATE_FAIL)
1457 		return er;
1458 
1459 	er = kvm_mips_emul_hypcall(vcpu, inst);
1460 	if (er == EMULATE_FAIL)
1461 		vcpu->arch.pc = curr_pc;
1462 
1463 	return er;
1464 }
1465 
1466 static enum emulation_result kvm_trap_vz_no_handler_guest_exit(u32 gexccode,
1467 							u32 cause,
1468 							u32 *opc,
1469 							struct kvm_vcpu *vcpu)
1470 {
1471 	u32 inst;
1472 
1473 	/*
1474 	 *  Fetch the instruction.
1475 	 */
1476 	if (cause & CAUSEF_BD)
1477 		opc += 1;
1478 	kvm_get_badinstr(opc, vcpu, &inst);
1479 
1480 	kvm_err("Guest Exception Code: %d not yet handled @ PC: %p, inst: 0x%08x  Status: %#x\n",
1481 		gexccode, opc, inst, read_gc0_status());
1482 
1483 	return EMULATE_FAIL;
1484 }
1485 
1486 static int kvm_trap_vz_handle_guest_exit(struct kvm_vcpu *vcpu)
1487 {
1488 	u32 *opc = (u32 *) vcpu->arch.pc;
1489 	u32 cause = vcpu->arch.host_cp0_cause;
1490 	enum emulation_result er = EMULATE_DONE;
1491 	u32 gexccode = (vcpu->arch.host_cp0_guestctl0 &
1492 			MIPS_GCTL0_GEXC) >> MIPS_GCTL0_GEXC_SHIFT;
1493 	int ret = RESUME_GUEST;
1494 
1495 	trace_kvm_exit(vcpu, KVM_TRACE_EXIT_GEXCCODE_BASE + gexccode);
1496 	switch (gexccode) {
1497 	case MIPS_GCTL0_GEXC_GPSI:
1498 		++vcpu->stat.vz_gpsi_exits;
1499 		er = kvm_trap_vz_handle_gpsi(cause, opc, vcpu);
1500 		break;
1501 	case MIPS_GCTL0_GEXC_GSFC:
1502 		++vcpu->stat.vz_gsfc_exits;
1503 		er = kvm_trap_vz_handle_gsfc(cause, opc, vcpu);
1504 		break;
1505 	case MIPS_GCTL0_GEXC_HC:
1506 		++vcpu->stat.vz_hc_exits;
1507 		er = kvm_trap_vz_handle_hc(cause, opc, vcpu);
1508 		break;
1509 	case MIPS_GCTL0_GEXC_GRR:
1510 		++vcpu->stat.vz_grr_exits;
1511 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1512 						       vcpu);
1513 		break;
1514 	case MIPS_GCTL0_GEXC_GVA:
1515 		++vcpu->stat.vz_gva_exits;
1516 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1517 						       vcpu);
1518 		break;
1519 	case MIPS_GCTL0_GEXC_GHFC:
1520 		++vcpu->stat.vz_ghfc_exits;
1521 		er = kvm_trap_vz_handle_ghfc(cause, opc, vcpu);
1522 		break;
1523 	case MIPS_GCTL0_GEXC_GPA:
1524 		++vcpu->stat.vz_gpa_exits;
1525 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1526 						       vcpu);
1527 		break;
1528 	default:
1529 		++vcpu->stat.vz_resvd_exits;
1530 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1531 						       vcpu);
1532 		break;
1533 
1534 	}
1535 
1536 	if (er == EMULATE_DONE) {
1537 		ret = RESUME_GUEST;
1538 	} else if (er == EMULATE_HYPERCALL) {
1539 		ret = kvm_mips_handle_hypcall(vcpu);
1540 	} else {
1541 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1542 		ret = RESUME_HOST;
1543 	}
1544 	return ret;
1545 }
1546 
1547 /**
1548  * kvm_trap_vz_handle_cop_unusuable() - Guest used unusable coprocessor.
1549  * @vcpu:	Virtual CPU context.
1550  *
1551  * Handle when the guest attempts to use a coprocessor which hasn't been allowed
1552  * by the root context.
1553  */
1554 static int kvm_trap_vz_handle_cop_unusable(struct kvm_vcpu *vcpu)
1555 {
1556 	struct kvm_run *run = vcpu->run;
1557 	u32 cause = vcpu->arch.host_cp0_cause;
1558 	enum emulation_result er = EMULATE_FAIL;
1559 	int ret = RESUME_GUEST;
1560 
1561 	if (((cause & CAUSEF_CE) >> CAUSEB_CE) == 1) {
1562 		/*
1563 		 * If guest FPU not present, the FPU operation should have been
1564 		 * treated as a reserved instruction!
1565 		 * If FPU already in use, we shouldn't get this at all.
1566 		 */
1567 		if (WARN_ON(!kvm_mips_guest_has_fpu(&vcpu->arch) ||
1568 			    vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU)) {
1569 			preempt_enable();
1570 			return EMULATE_FAIL;
1571 		}
1572 
1573 		kvm_own_fpu(vcpu);
1574 		er = EMULATE_DONE;
1575 	}
1576 	/* other coprocessors not handled */
1577 
1578 	switch (er) {
1579 	case EMULATE_DONE:
1580 		ret = RESUME_GUEST;
1581 		break;
1582 
1583 	case EMULATE_FAIL:
1584 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1585 		ret = RESUME_HOST;
1586 		break;
1587 
1588 	default:
1589 		BUG();
1590 	}
1591 	return ret;
1592 }
1593 
1594 /**
1595  * kvm_trap_vz_handle_msa_disabled() - Guest used MSA while disabled in root.
1596  * @vcpu:	Virtual CPU context.
1597  *
1598  * Handle when the guest attempts to use MSA when it is disabled in the root
1599  * context.
1600  */
1601 static int kvm_trap_vz_handle_msa_disabled(struct kvm_vcpu *vcpu)
1602 {
1603 	struct kvm_run *run = vcpu->run;
1604 
1605 	/*
1606 	 * If MSA not present or not exposed to guest or FR=0, the MSA operation
1607 	 * should have been treated as a reserved instruction!
1608 	 * Same if CU1=1, FR=0.
1609 	 * If MSA already in use, we shouldn't get this at all.
1610 	 */
1611 	if (!kvm_mips_guest_has_msa(&vcpu->arch) ||
1612 	    (read_gc0_status() & (ST0_CU1 | ST0_FR)) == ST0_CU1 ||
1613 	    !(read_gc0_config5() & MIPS_CONF5_MSAEN) ||
1614 	    vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA) {
1615 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1616 		return RESUME_HOST;
1617 	}
1618 
1619 	kvm_own_msa(vcpu);
1620 
1621 	return RESUME_GUEST;
1622 }
1623 
1624 static int kvm_trap_vz_handle_tlb_ld_miss(struct kvm_vcpu *vcpu)
1625 {
1626 	struct kvm_run *run = vcpu->run;
1627 	u32 *opc = (u32 *) vcpu->arch.pc;
1628 	u32 cause = vcpu->arch.host_cp0_cause;
1629 	ulong badvaddr = vcpu->arch.host_cp0_badvaddr;
1630 	union mips_instruction inst;
1631 	enum emulation_result er = EMULATE_DONE;
1632 	int err, ret = RESUME_GUEST;
1633 
1634 	if (kvm_mips_handle_vz_root_tlb_fault(badvaddr, vcpu, false)) {
1635 		/* A code fetch fault doesn't count as an MMIO */
1636 		if (kvm_is_ifetch_fault(&vcpu->arch)) {
1637 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1638 			return RESUME_HOST;
1639 		}
1640 
1641 		/* Fetch the instruction */
1642 		if (cause & CAUSEF_BD)
1643 			opc += 1;
1644 		err = kvm_get_badinstr(opc, vcpu, &inst.word);
1645 		if (err) {
1646 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1647 			return RESUME_HOST;
1648 		}
1649 
1650 		/* Treat as MMIO */
1651 		er = kvm_mips_emulate_load(inst, cause, run, vcpu);
1652 		if (er == EMULATE_FAIL) {
1653 			kvm_err("Guest Emulate Load from MMIO space failed: PC: %p, BadVaddr: %#lx\n",
1654 				opc, badvaddr);
1655 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1656 		}
1657 	}
1658 
1659 	if (er == EMULATE_DONE) {
1660 		ret = RESUME_GUEST;
1661 	} else if (er == EMULATE_DO_MMIO) {
1662 		run->exit_reason = KVM_EXIT_MMIO;
1663 		ret = RESUME_HOST;
1664 	} else {
1665 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1666 		ret = RESUME_HOST;
1667 	}
1668 	return ret;
1669 }
1670 
1671 static int kvm_trap_vz_handle_tlb_st_miss(struct kvm_vcpu *vcpu)
1672 {
1673 	struct kvm_run *run = vcpu->run;
1674 	u32 *opc = (u32 *) vcpu->arch.pc;
1675 	u32 cause = vcpu->arch.host_cp0_cause;
1676 	ulong badvaddr = vcpu->arch.host_cp0_badvaddr;
1677 	union mips_instruction inst;
1678 	enum emulation_result er = EMULATE_DONE;
1679 	int err;
1680 	int ret = RESUME_GUEST;
1681 
1682 	/* Just try the access again if we couldn't do the translation */
1683 	if (kvm_vz_badvaddr_to_gpa(vcpu, badvaddr, &badvaddr))
1684 		return RESUME_GUEST;
1685 	vcpu->arch.host_cp0_badvaddr = badvaddr;
1686 
1687 	if (kvm_mips_handle_vz_root_tlb_fault(badvaddr, vcpu, true)) {
1688 		/* Fetch the instruction */
1689 		if (cause & CAUSEF_BD)
1690 			opc += 1;
1691 		err = kvm_get_badinstr(opc, vcpu, &inst.word);
1692 		if (err) {
1693 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1694 			return RESUME_HOST;
1695 		}
1696 
1697 		/* Treat as MMIO */
1698 		er = kvm_mips_emulate_store(inst, cause, run, vcpu);
1699 		if (er == EMULATE_FAIL) {
1700 			kvm_err("Guest Emulate Store to MMIO space failed: PC: %p, BadVaddr: %#lx\n",
1701 				opc, badvaddr);
1702 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1703 		}
1704 	}
1705 
1706 	if (er == EMULATE_DONE) {
1707 		ret = RESUME_GUEST;
1708 	} else if (er == EMULATE_DO_MMIO) {
1709 		run->exit_reason = KVM_EXIT_MMIO;
1710 		ret = RESUME_HOST;
1711 	} else {
1712 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1713 		ret = RESUME_HOST;
1714 	}
1715 	return ret;
1716 }
1717 
1718 static u64 kvm_vz_get_one_regs[] = {
1719 	KVM_REG_MIPS_CP0_INDEX,
1720 	KVM_REG_MIPS_CP0_ENTRYLO0,
1721 	KVM_REG_MIPS_CP0_ENTRYLO1,
1722 	KVM_REG_MIPS_CP0_CONTEXT,
1723 	KVM_REG_MIPS_CP0_PAGEMASK,
1724 	KVM_REG_MIPS_CP0_PAGEGRAIN,
1725 	KVM_REG_MIPS_CP0_WIRED,
1726 	KVM_REG_MIPS_CP0_HWRENA,
1727 	KVM_REG_MIPS_CP0_BADVADDR,
1728 	KVM_REG_MIPS_CP0_COUNT,
1729 	KVM_REG_MIPS_CP0_ENTRYHI,
1730 	KVM_REG_MIPS_CP0_COMPARE,
1731 	KVM_REG_MIPS_CP0_STATUS,
1732 	KVM_REG_MIPS_CP0_INTCTL,
1733 	KVM_REG_MIPS_CP0_CAUSE,
1734 	KVM_REG_MIPS_CP0_EPC,
1735 	KVM_REG_MIPS_CP0_PRID,
1736 	KVM_REG_MIPS_CP0_EBASE,
1737 	KVM_REG_MIPS_CP0_CONFIG,
1738 	KVM_REG_MIPS_CP0_CONFIG1,
1739 	KVM_REG_MIPS_CP0_CONFIG2,
1740 	KVM_REG_MIPS_CP0_CONFIG3,
1741 	KVM_REG_MIPS_CP0_CONFIG4,
1742 	KVM_REG_MIPS_CP0_CONFIG5,
1743 	KVM_REG_MIPS_CP0_CONFIG6,
1744 #ifdef CONFIG_64BIT
1745 	KVM_REG_MIPS_CP0_XCONTEXT,
1746 #endif
1747 	KVM_REG_MIPS_CP0_ERROREPC,
1748 
1749 	KVM_REG_MIPS_COUNT_CTL,
1750 	KVM_REG_MIPS_COUNT_RESUME,
1751 	KVM_REG_MIPS_COUNT_HZ,
1752 };
1753 
1754 static u64 kvm_vz_get_one_regs_contextconfig[] = {
1755 	KVM_REG_MIPS_CP0_CONTEXTCONFIG,
1756 #ifdef CONFIG_64BIT
1757 	KVM_REG_MIPS_CP0_XCONTEXTCONFIG,
1758 #endif
1759 };
1760 
1761 static u64 kvm_vz_get_one_regs_segments[] = {
1762 	KVM_REG_MIPS_CP0_SEGCTL0,
1763 	KVM_REG_MIPS_CP0_SEGCTL1,
1764 	KVM_REG_MIPS_CP0_SEGCTL2,
1765 };
1766 
1767 static u64 kvm_vz_get_one_regs_htw[] = {
1768 	KVM_REG_MIPS_CP0_PWBASE,
1769 	KVM_REG_MIPS_CP0_PWFIELD,
1770 	KVM_REG_MIPS_CP0_PWSIZE,
1771 	KVM_REG_MIPS_CP0_PWCTL,
1772 };
1773 
1774 static u64 kvm_vz_get_one_regs_kscratch[] = {
1775 	KVM_REG_MIPS_CP0_KSCRATCH1,
1776 	KVM_REG_MIPS_CP0_KSCRATCH2,
1777 	KVM_REG_MIPS_CP0_KSCRATCH3,
1778 	KVM_REG_MIPS_CP0_KSCRATCH4,
1779 	KVM_REG_MIPS_CP0_KSCRATCH5,
1780 	KVM_REG_MIPS_CP0_KSCRATCH6,
1781 };
1782 
1783 static unsigned long kvm_vz_num_regs(struct kvm_vcpu *vcpu)
1784 {
1785 	unsigned long ret;
1786 
1787 	ret = ARRAY_SIZE(kvm_vz_get_one_regs);
1788 	if (cpu_guest_has_userlocal)
1789 		++ret;
1790 	if (cpu_guest_has_badinstr)
1791 		++ret;
1792 	if (cpu_guest_has_badinstrp)
1793 		++ret;
1794 	if (cpu_guest_has_contextconfig)
1795 		ret += ARRAY_SIZE(kvm_vz_get_one_regs_contextconfig);
1796 	if (cpu_guest_has_segments)
1797 		ret += ARRAY_SIZE(kvm_vz_get_one_regs_segments);
1798 	if (cpu_guest_has_htw || cpu_guest_has_ldpte)
1799 		ret += ARRAY_SIZE(kvm_vz_get_one_regs_htw);
1800 	if (cpu_guest_has_maar && !cpu_guest_has_dyn_maar)
1801 		ret += 1 + ARRAY_SIZE(vcpu->arch.maar);
1802 	ret += __arch_hweight8(cpu_data[0].guest.kscratch_mask);
1803 
1804 	return ret;
1805 }
1806 
1807 static int kvm_vz_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *indices)
1808 {
1809 	u64 index;
1810 	unsigned int i;
1811 
1812 	if (copy_to_user(indices, kvm_vz_get_one_regs,
1813 			 sizeof(kvm_vz_get_one_regs)))
1814 		return -EFAULT;
1815 	indices += ARRAY_SIZE(kvm_vz_get_one_regs);
1816 
1817 	if (cpu_guest_has_userlocal) {
1818 		index = KVM_REG_MIPS_CP0_USERLOCAL;
1819 		if (copy_to_user(indices, &index, sizeof(index)))
1820 			return -EFAULT;
1821 		++indices;
1822 	}
1823 	if (cpu_guest_has_badinstr) {
1824 		index = KVM_REG_MIPS_CP0_BADINSTR;
1825 		if (copy_to_user(indices, &index, sizeof(index)))
1826 			return -EFAULT;
1827 		++indices;
1828 	}
1829 	if (cpu_guest_has_badinstrp) {
1830 		index = KVM_REG_MIPS_CP0_BADINSTRP;
1831 		if (copy_to_user(indices, &index, sizeof(index)))
1832 			return -EFAULT;
1833 		++indices;
1834 	}
1835 	if (cpu_guest_has_contextconfig) {
1836 		if (copy_to_user(indices, kvm_vz_get_one_regs_contextconfig,
1837 				 sizeof(kvm_vz_get_one_regs_contextconfig)))
1838 			return -EFAULT;
1839 		indices += ARRAY_SIZE(kvm_vz_get_one_regs_contextconfig);
1840 	}
1841 	if (cpu_guest_has_segments) {
1842 		if (copy_to_user(indices, kvm_vz_get_one_regs_segments,
1843 				 sizeof(kvm_vz_get_one_regs_segments)))
1844 			return -EFAULT;
1845 		indices += ARRAY_SIZE(kvm_vz_get_one_regs_segments);
1846 	}
1847 	if (cpu_guest_has_htw || cpu_guest_has_ldpte) {
1848 		if (copy_to_user(indices, kvm_vz_get_one_regs_htw,
1849 				 sizeof(kvm_vz_get_one_regs_htw)))
1850 			return -EFAULT;
1851 		indices += ARRAY_SIZE(kvm_vz_get_one_regs_htw);
1852 	}
1853 	if (cpu_guest_has_maar && !cpu_guest_has_dyn_maar) {
1854 		for (i = 0; i < ARRAY_SIZE(vcpu->arch.maar); ++i) {
1855 			index = KVM_REG_MIPS_CP0_MAAR(i);
1856 			if (copy_to_user(indices, &index, sizeof(index)))
1857 				return -EFAULT;
1858 			++indices;
1859 		}
1860 
1861 		index = KVM_REG_MIPS_CP0_MAARI;
1862 		if (copy_to_user(indices, &index, sizeof(index)))
1863 			return -EFAULT;
1864 		++indices;
1865 	}
1866 	for (i = 0; i < 6; ++i) {
1867 		if (!cpu_guest_has_kscr(i + 2))
1868 			continue;
1869 
1870 		if (copy_to_user(indices, &kvm_vz_get_one_regs_kscratch[i],
1871 				 sizeof(kvm_vz_get_one_regs_kscratch[i])))
1872 			return -EFAULT;
1873 		++indices;
1874 	}
1875 
1876 	return 0;
1877 }
1878 
1879 static inline s64 entrylo_kvm_to_user(unsigned long v)
1880 {
1881 	s64 mask, ret = v;
1882 
1883 	if (BITS_PER_LONG == 32) {
1884 		/*
1885 		 * KVM API exposes 64-bit version of the register, so move the
1886 		 * RI/XI bits up into place.
1887 		 */
1888 		mask = MIPS_ENTRYLO_RI | MIPS_ENTRYLO_XI;
1889 		ret &= ~mask;
1890 		ret |= ((s64)v & mask) << 32;
1891 	}
1892 	return ret;
1893 }
1894 
1895 static inline unsigned long entrylo_user_to_kvm(s64 v)
1896 {
1897 	unsigned long mask, ret = v;
1898 
1899 	if (BITS_PER_LONG == 32) {
1900 		/*
1901 		 * KVM API exposes 64-bit versiono of the register, so move the
1902 		 * RI/XI bits down into place.
1903 		 */
1904 		mask = MIPS_ENTRYLO_RI | MIPS_ENTRYLO_XI;
1905 		ret &= ~mask;
1906 		ret |= (v >> 32) & mask;
1907 	}
1908 	return ret;
1909 }
1910 
1911 static int kvm_vz_get_one_reg(struct kvm_vcpu *vcpu,
1912 			      const struct kvm_one_reg *reg,
1913 			      s64 *v)
1914 {
1915 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1916 	unsigned int idx;
1917 
1918 	switch (reg->id) {
1919 	case KVM_REG_MIPS_CP0_INDEX:
1920 		*v = (long)read_gc0_index();
1921 		break;
1922 	case KVM_REG_MIPS_CP0_ENTRYLO0:
1923 		*v = entrylo_kvm_to_user(read_gc0_entrylo0());
1924 		break;
1925 	case KVM_REG_MIPS_CP0_ENTRYLO1:
1926 		*v = entrylo_kvm_to_user(read_gc0_entrylo1());
1927 		break;
1928 	case KVM_REG_MIPS_CP0_CONTEXT:
1929 		*v = (long)read_gc0_context();
1930 		break;
1931 	case KVM_REG_MIPS_CP0_CONTEXTCONFIG:
1932 		if (!cpu_guest_has_contextconfig)
1933 			return -EINVAL;
1934 		*v = read_gc0_contextconfig();
1935 		break;
1936 	case KVM_REG_MIPS_CP0_USERLOCAL:
1937 		if (!cpu_guest_has_userlocal)
1938 			return -EINVAL;
1939 		*v = read_gc0_userlocal();
1940 		break;
1941 #ifdef CONFIG_64BIT
1942 	case KVM_REG_MIPS_CP0_XCONTEXTCONFIG:
1943 		if (!cpu_guest_has_contextconfig)
1944 			return -EINVAL;
1945 		*v = read_gc0_xcontextconfig();
1946 		break;
1947 #endif
1948 	case KVM_REG_MIPS_CP0_PAGEMASK:
1949 		*v = (long)read_gc0_pagemask();
1950 		break;
1951 	case KVM_REG_MIPS_CP0_PAGEGRAIN:
1952 		*v = (long)read_gc0_pagegrain();
1953 		break;
1954 	case KVM_REG_MIPS_CP0_SEGCTL0:
1955 		if (!cpu_guest_has_segments)
1956 			return -EINVAL;
1957 		*v = read_gc0_segctl0();
1958 		break;
1959 	case KVM_REG_MIPS_CP0_SEGCTL1:
1960 		if (!cpu_guest_has_segments)
1961 			return -EINVAL;
1962 		*v = read_gc0_segctl1();
1963 		break;
1964 	case KVM_REG_MIPS_CP0_SEGCTL2:
1965 		if (!cpu_guest_has_segments)
1966 			return -EINVAL;
1967 		*v = read_gc0_segctl2();
1968 		break;
1969 	case KVM_REG_MIPS_CP0_PWBASE:
1970 		if (!cpu_guest_has_htw && !cpu_guest_has_ldpte)
1971 			return -EINVAL;
1972 		*v = read_gc0_pwbase();
1973 		break;
1974 	case KVM_REG_MIPS_CP0_PWFIELD:
1975 		if (!cpu_guest_has_htw && !cpu_guest_has_ldpte)
1976 			return -EINVAL;
1977 		*v = read_gc0_pwfield();
1978 		break;
1979 	case KVM_REG_MIPS_CP0_PWSIZE:
1980 		if (!cpu_guest_has_htw && !cpu_guest_has_ldpte)
1981 			return -EINVAL;
1982 		*v = read_gc0_pwsize();
1983 		break;
1984 	case KVM_REG_MIPS_CP0_WIRED:
1985 		*v = (long)read_gc0_wired();
1986 		break;
1987 	case KVM_REG_MIPS_CP0_PWCTL:
1988 		if (!cpu_guest_has_htw && !cpu_guest_has_ldpte)
1989 			return -EINVAL;
1990 		*v = read_gc0_pwctl();
1991 		break;
1992 	case KVM_REG_MIPS_CP0_HWRENA:
1993 		*v = (long)read_gc0_hwrena();
1994 		break;
1995 	case KVM_REG_MIPS_CP0_BADVADDR:
1996 		*v = (long)read_gc0_badvaddr();
1997 		break;
1998 	case KVM_REG_MIPS_CP0_BADINSTR:
1999 		if (!cpu_guest_has_badinstr)
2000 			return -EINVAL;
2001 		*v = read_gc0_badinstr();
2002 		break;
2003 	case KVM_REG_MIPS_CP0_BADINSTRP:
2004 		if (!cpu_guest_has_badinstrp)
2005 			return -EINVAL;
2006 		*v = read_gc0_badinstrp();
2007 		break;
2008 	case KVM_REG_MIPS_CP0_COUNT:
2009 		*v = kvm_mips_read_count(vcpu);
2010 		break;
2011 	case KVM_REG_MIPS_CP0_ENTRYHI:
2012 		*v = (long)read_gc0_entryhi();
2013 		break;
2014 	case KVM_REG_MIPS_CP0_COMPARE:
2015 		*v = (long)read_gc0_compare();
2016 		break;
2017 	case KVM_REG_MIPS_CP0_STATUS:
2018 		*v = (long)read_gc0_status();
2019 		break;
2020 	case KVM_REG_MIPS_CP0_INTCTL:
2021 		*v = read_gc0_intctl();
2022 		break;
2023 	case KVM_REG_MIPS_CP0_CAUSE:
2024 		*v = (long)read_gc0_cause();
2025 		break;
2026 	case KVM_REG_MIPS_CP0_EPC:
2027 		*v = (long)read_gc0_epc();
2028 		break;
2029 	case KVM_REG_MIPS_CP0_PRID:
2030 		switch (boot_cpu_type()) {
2031 		case CPU_CAVIUM_OCTEON3:
2032 			/* Octeon III has a read-only guest.PRid */
2033 			*v = read_gc0_prid();
2034 			break;
2035 		default:
2036 			*v = (long)kvm_read_c0_guest_prid(cop0);
2037 			break;
2038 		}
2039 		break;
2040 	case KVM_REG_MIPS_CP0_EBASE:
2041 		*v = kvm_vz_read_gc0_ebase();
2042 		break;
2043 	case KVM_REG_MIPS_CP0_CONFIG:
2044 		*v = read_gc0_config();
2045 		break;
2046 	case KVM_REG_MIPS_CP0_CONFIG1:
2047 		if (!cpu_guest_has_conf1)
2048 			return -EINVAL;
2049 		*v = read_gc0_config1();
2050 		break;
2051 	case KVM_REG_MIPS_CP0_CONFIG2:
2052 		if (!cpu_guest_has_conf2)
2053 			return -EINVAL;
2054 		*v = read_gc0_config2();
2055 		break;
2056 	case KVM_REG_MIPS_CP0_CONFIG3:
2057 		if (!cpu_guest_has_conf3)
2058 			return -EINVAL;
2059 		*v = read_gc0_config3();
2060 		break;
2061 	case KVM_REG_MIPS_CP0_CONFIG4:
2062 		if (!cpu_guest_has_conf4)
2063 			return -EINVAL;
2064 		*v = read_gc0_config4();
2065 		break;
2066 	case KVM_REG_MIPS_CP0_CONFIG5:
2067 		if (!cpu_guest_has_conf5)
2068 			return -EINVAL;
2069 		*v = read_gc0_config5();
2070 		break;
2071 	case KVM_REG_MIPS_CP0_CONFIG6:
2072 		*v = kvm_read_sw_gc0_config6(cop0);
2073 		break;
2074 	case KVM_REG_MIPS_CP0_MAAR(0) ... KVM_REG_MIPS_CP0_MAAR(0x3f):
2075 		if (!cpu_guest_has_maar || cpu_guest_has_dyn_maar)
2076 			return -EINVAL;
2077 		idx = reg->id - KVM_REG_MIPS_CP0_MAAR(0);
2078 		if (idx >= ARRAY_SIZE(vcpu->arch.maar))
2079 			return -EINVAL;
2080 		*v = vcpu->arch.maar[idx];
2081 		break;
2082 	case KVM_REG_MIPS_CP0_MAARI:
2083 		if (!cpu_guest_has_maar || cpu_guest_has_dyn_maar)
2084 			return -EINVAL;
2085 		*v = kvm_read_sw_gc0_maari(vcpu->arch.cop0);
2086 		break;
2087 #ifdef CONFIG_64BIT
2088 	case KVM_REG_MIPS_CP0_XCONTEXT:
2089 		*v = read_gc0_xcontext();
2090 		break;
2091 #endif
2092 	case KVM_REG_MIPS_CP0_ERROREPC:
2093 		*v = (long)read_gc0_errorepc();
2094 		break;
2095 	case KVM_REG_MIPS_CP0_KSCRATCH1 ... KVM_REG_MIPS_CP0_KSCRATCH6:
2096 		idx = reg->id - KVM_REG_MIPS_CP0_KSCRATCH1 + 2;
2097 		if (!cpu_guest_has_kscr(idx))
2098 			return -EINVAL;
2099 		switch (idx) {
2100 		case 2:
2101 			*v = (long)read_gc0_kscratch1();
2102 			break;
2103 		case 3:
2104 			*v = (long)read_gc0_kscratch2();
2105 			break;
2106 		case 4:
2107 			*v = (long)read_gc0_kscratch3();
2108 			break;
2109 		case 5:
2110 			*v = (long)read_gc0_kscratch4();
2111 			break;
2112 		case 6:
2113 			*v = (long)read_gc0_kscratch5();
2114 			break;
2115 		case 7:
2116 			*v = (long)read_gc0_kscratch6();
2117 			break;
2118 		}
2119 		break;
2120 	case KVM_REG_MIPS_COUNT_CTL:
2121 		*v = vcpu->arch.count_ctl;
2122 		break;
2123 	case KVM_REG_MIPS_COUNT_RESUME:
2124 		*v = ktime_to_ns(vcpu->arch.count_resume);
2125 		break;
2126 	case KVM_REG_MIPS_COUNT_HZ:
2127 		*v = vcpu->arch.count_hz;
2128 		break;
2129 	default:
2130 		return -EINVAL;
2131 	}
2132 	return 0;
2133 }
2134 
2135 static int kvm_vz_set_one_reg(struct kvm_vcpu *vcpu,
2136 			      const struct kvm_one_reg *reg,
2137 			      s64 v)
2138 {
2139 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2140 	unsigned int idx;
2141 	int ret = 0;
2142 	unsigned int cur, change;
2143 
2144 	switch (reg->id) {
2145 	case KVM_REG_MIPS_CP0_INDEX:
2146 		write_gc0_index(v);
2147 		break;
2148 	case KVM_REG_MIPS_CP0_ENTRYLO0:
2149 		write_gc0_entrylo0(entrylo_user_to_kvm(v));
2150 		break;
2151 	case KVM_REG_MIPS_CP0_ENTRYLO1:
2152 		write_gc0_entrylo1(entrylo_user_to_kvm(v));
2153 		break;
2154 	case KVM_REG_MIPS_CP0_CONTEXT:
2155 		write_gc0_context(v);
2156 		break;
2157 	case KVM_REG_MIPS_CP0_CONTEXTCONFIG:
2158 		if (!cpu_guest_has_contextconfig)
2159 			return -EINVAL;
2160 		write_gc0_contextconfig(v);
2161 		break;
2162 	case KVM_REG_MIPS_CP0_USERLOCAL:
2163 		if (!cpu_guest_has_userlocal)
2164 			return -EINVAL;
2165 		write_gc0_userlocal(v);
2166 		break;
2167 #ifdef CONFIG_64BIT
2168 	case KVM_REG_MIPS_CP0_XCONTEXTCONFIG:
2169 		if (!cpu_guest_has_contextconfig)
2170 			return -EINVAL;
2171 		write_gc0_xcontextconfig(v);
2172 		break;
2173 #endif
2174 	case KVM_REG_MIPS_CP0_PAGEMASK:
2175 		write_gc0_pagemask(v);
2176 		break;
2177 	case KVM_REG_MIPS_CP0_PAGEGRAIN:
2178 		write_gc0_pagegrain(v);
2179 		break;
2180 	case KVM_REG_MIPS_CP0_SEGCTL0:
2181 		if (!cpu_guest_has_segments)
2182 			return -EINVAL;
2183 		write_gc0_segctl0(v);
2184 		break;
2185 	case KVM_REG_MIPS_CP0_SEGCTL1:
2186 		if (!cpu_guest_has_segments)
2187 			return -EINVAL;
2188 		write_gc0_segctl1(v);
2189 		break;
2190 	case KVM_REG_MIPS_CP0_SEGCTL2:
2191 		if (!cpu_guest_has_segments)
2192 			return -EINVAL;
2193 		write_gc0_segctl2(v);
2194 		break;
2195 	case KVM_REG_MIPS_CP0_PWBASE:
2196 		if (!cpu_guest_has_htw && !cpu_guest_has_ldpte)
2197 			return -EINVAL;
2198 		write_gc0_pwbase(v);
2199 		break;
2200 	case KVM_REG_MIPS_CP0_PWFIELD:
2201 		if (!cpu_guest_has_htw && !cpu_guest_has_ldpte)
2202 			return -EINVAL;
2203 		write_gc0_pwfield(v);
2204 		break;
2205 	case KVM_REG_MIPS_CP0_PWSIZE:
2206 		if (!cpu_guest_has_htw && !cpu_guest_has_ldpte)
2207 			return -EINVAL;
2208 		write_gc0_pwsize(v);
2209 		break;
2210 	case KVM_REG_MIPS_CP0_WIRED:
2211 		change_gc0_wired(MIPSR6_WIRED_WIRED, v);
2212 		break;
2213 	case KVM_REG_MIPS_CP0_PWCTL:
2214 		if (!cpu_guest_has_htw && !cpu_guest_has_ldpte)
2215 			return -EINVAL;
2216 		write_gc0_pwctl(v);
2217 		break;
2218 	case KVM_REG_MIPS_CP0_HWRENA:
2219 		write_gc0_hwrena(v);
2220 		break;
2221 	case KVM_REG_MIPS_CP0_BADVADDR:
2222 		write_gc0_badvaddr(v);
2223 		break;
2224 	case KVM_REG_MIPS_CP0_BADINSTR:
2225 		if (!cpu_guest_has_badinstr)
2226 			return -EINVAL;
2227 		write_gc0_badinstr(v);
2228 		break;
2229 	case KVM_REG_MIPS_CP0_BADINSTRP:
2230 		if (!cpu_guest_has_badinstrp)
2231 			return -EINVAL;
2232 		write_gc0_badinstrp(v);
2233 		break;
2234 	case KVM_REG_MIPS_CP0_COUNT:
2235 		kvm_mips_write_count(vcpu, v);
2236 		break;
2237 	case KVM_REG_MIPS_CP0_ENTRYHI:
2238 		write_gc0_entryhi(v);
2239 		break;
2240 	case KVM_REG_MIPS_CP0_COMPARE:
2241 		kvm_mips_write_compare(vcpu, v, false);
2242 		break;
2243 	case KVM_REG_MIPS_CP0_STATUS:
2244 		write_gc0_status(v);
2245 		break;
2246 	case KVM_REG_MIPS_CP0_INTCTL:
2247 		write_gc0_intctl(v);
2248 		break;
2249 	case KVM_REG_MIPS_CP0_CAUSE:
2250 		/*
2251 		 * If the timer is stopped or started (DC bit) it must look
2252 		 * atomic with changes to the timer interrupt pending bit (TI).
2253 		 * A timer interrupt should not happen in between.
2254 		 */
2255 		if ((read_gc0_cause() ^ v) & CAUSEF_DC) {
2256 			if (v & CAUSEF_DC) {
2257 				/* disable timer first */
2258 				kvm_mips_count_disable_cause(vcpu);
2259 				change_gc0_cause((u32)~CAUSEF_DC, v);
2260 			} else {
2261 				/* enable timer last */
2262 				change_gc0_cause((u32)~CAUSEF_DC, v);
2263 				kvm_mips_count_enable_cause(vcpu);
2264 			}
2265 		} else {
2266 			write_gc0_cause(v);
2267 		}
2268 		break;
2269 	case KVM_REG_MIPS_CP0_EPC:
2270 		write_gc0_epc(v);
2271 		break;
2272 	case KVM_REG_MIPS_CP0_PRID:
2273 		switch (boot_cpu_type()) {
2274 		case CPU_CAVIUM_OCTEON3:
2275 			/* Octeon III has a guest.PRid, but its read-only */
2276 			break;
2277 		default:
2278 			kvm_write_c0_guest_prid(cop0, v);
2279 			break;
2280 		}
2281 		break;
2282 	case KVM_REG_MIPS_CP0_EBASE:
2283 		kvm_vz_write_gc0_ebase(v);
2284 		break;
2285 	case KVM_REG_MIPS_CP0_CONFIG:
2286 		cur = read_gc0_config();
2287 		change = (cur ^ v) & kvm_vz_config_user_wrmask(vcpu);
2288 		if (change) {
2289 			v = cur ^ change;
2290 			write_gc0_config(v);
2291 		}
2292 		break;
2293 	case KVM_REG_MIPS_CP0_CONFIG1:
2294 		if (!cpu_guest_has_conf1)
2295 			break;
2296 		cur = read_gc0_config1();
2297 		change = (cur ^ v) & kvm_vz_config1_user_wrmask(vcpu);
2298 		if (change) {
2299 			v = cur ^ change;
2300 			write_gc0_config1(v);
2301 		}
2302 		break;
2303 	case KVM_REG_MIPS_CP0_CONFIG2:
2304 		if (!cpu_guest_has_conf2)
2305 			break;
2306 		cur = read_gc0_config2();
2307 		change = (cur ^ v) & kvm_vz_config2_user_wrmask(vcpu);
2308 		if (change) {
2309 			v = cur ^ change;
2310 			write_gc0_config2(v);
2311 		}
2312 		break;
2313 	case KVM_REG_MIPS_CP0_CONFIG3:
2314 		if (!cpu_guest_has_conf3)
2315 			break;
2316 		cur = read_gc0_config3();
2317 		change = (cur ^ v) & kvm_vz_config3_user_wrmask(vcpu);
2318 		if (change) {
2319 			v = cur ^ change;
2320 			write_gc0_config3(v);
2321 		}
2322 		break;
2323 	case KVM_REG_MIPS_CP0_CONFIG4:
2324 		if (!cpu_guest_has_conf4)
2325 			break;
2326 		cur = read_gc0_config4();
2327 		change = (cur ^ v) & kvm_vz_config4_user_wrmask(vcpu);
2328 		if (change) {
2329 			v = cur ^ change;
2330 			write_gc0_config4(v);
2331 		}
2332 		break;
2333 	case KVM_REG_MIPS_CP0_CONFIG5:
2334 		if (!cpu_guest_has_conf5)
2335 			break;
2336 		cur = read_gc0_config5();
2337 		change = (cur ^ v) & kvm_vz_config5_user_wrmask(vcpu);
2338 		if (change) {
2339 			v = cur ^ change;
2340 			write_gc0_config5(v);
2341 		}
2342 		break;
2343 	case KVM_REG_MIPS_CP0_CONFIG6:
2344 		cur = kvm_read_sw_gc0_config6(cop0);
2345 		change = (cur ^ v) & kvm_vz_config6_user_wrmask(vcpu);
2346 		if (change) {
2347 			v = cur ^ change;
2348 			kvm_write_sw_gc0_config6(cop0, (int)v);
2349 		}
2350 		break;
2351 	case KVM_REG_MIPS_CP0_MAAR(0) ... KVM_REG_MIPS_CP0_MAAR(0x3f):
2352 		if (!cpu_guest_has_maar || cpu_guest_has_dyn_maar)
2353 			return -EINVAL;
2354 		idx = reg->id - KVM_REG_MIPS_CP0_MAAR(0);
2355 		if (idx >= ARRAY_SIZE(vcpu->arch.maar))
2356 			return -EINVAL;
2357 		vcpu->arch.maar[idx] = mips_process_maar(dmtc_op, v);
2358 		break;
2359 	case KVM_REG_MIPS_CP0_MAARI:
2360 		if (!cpu_guest_has_maar || cpu_guest_has_dyn_maar)
2361 			return -EINVAL;
2362 		kvm_write_maari(vcpu, v);
2363 		break;
2364 #ifdef CONFIG_64BIT
2365 	case KVM_REG_MIPS_CP0_XCONTEXT:
2366 		write_gc0_xcontext(v);
2367 		break;
2368 #endif
2369 	case KVM_REG_MIPS_CP0_ERROREPC:
2370 		write_gc0_errorepc(v);
2371 		break;
2372 	case KVM_REG_MIPS_CP0_KSCRATCH1 ... KVM_REG_MIPS_CP0_KSCRATCH6:
2373 		idx = reg->id - KVM_REG_MIPS_CP0_KSCRATCH1 + 2;
2374 		if (!cpu_guest_has_kscr(idx))
2375 			return -EINVAL;
2376 		switch (idx) {
2377 		case 2:
2378 			write_gc0_kscratch1(v);
2379 			break;
2380 		case 3:
2381 			write_gc0_kscratch2(v);
2382 			break;
2383 		case 4:
2384 			write_gc0_kscratch3(v);
2385 			break;
2386 		case 5:
2387 			write_gc0_kscratch4(v);
2388 			break;
2389 		case 6:
2390 			write_gc0_kscratch5(v);
2391 			break;
2392 		case 7:
2393 			write_gc0_kscratch6(v);
2394 			break;
2395 		}
2396 		break;
2397 	case KVM_REG_MIPS_COUNT_CTL:
2398 		ret = kvm_mips_set_count_ctl(vcpu, v);
2399 		break;
2400 	case KVM_REG_MIPS_COUNT_RESUME:
2401 		ret = kvm_mips_set_count_resume(vcpu, v);
2402 		break;
2403 	case KVM_REG_MIPS_COUNT_HZ:
2404 		ret = kvm_mips_set_count_hz(vcpu, v);
2405 		break;
2406 	default:
2407 		return -EINVAL;
2408 	}
2409 	return ret;
2410 }
2411 
2412 #define guestid_cache(cpu)	(cpu_data[cpu].guestid_cache)
2413 static void kvm_vz_get_new_guestid(unsigned long cpu, struct kvm_vcpu *vcpu)
2414 {
2415 	unsigned long guestid = guestid_cache(cpu);
2416 
2417 	if (!(++guestid & GUESTID_MASK)) {
2418 		if (cpu_has_vtag_icache)
2419 			flush_icache_all();
2420 
2421 		if (!guestid)		/* fix version if needed */
2422 			guestid = GUESTID_FIRST_VERSION;
2423 
2424 		++guestid;		/* guestid 0 reserved for root */
2425 
2426 		/* start new guestid cycle */
2427 		kvm_vz_local_flush_roottlb_all_guests();
2428 		kvm_vz_local_flush_guesttlb_all();
2429 	}
2430 
2431 	guestid_cache(cpu) = guestid;
2432 }
2433 
2434 /* Returns 1 if the guest TLB may be clobbered */
2435 static int kvm_vz_check_requests(struct kvm_vcpu *vcpu, int cpu)
2436 {
2437 	int ret = 0;
2438 	int i;
2439 
2440 	if (!kvm_request_pending(vcpu))
2441 		return 0;
2442 
2443 	if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
2444 		if (cpu_has_guestid) {
2445 			/* Drop all GuestIDs for this VCPU */
2446 			for_each_possible_cpu(i)
2447 				vcpu->arch.vzguestid[i] = 0;
2448 			/* This will clobber guest TLB contents too */
2449 			ret = 1;
2450 		}
2451 		/*
2452 		 * For Root ASID Dealias (RAD) we don't do anything here, but we
2453 		 * still need the request to ensure we recheck asid_flush_mask.
2454 		 * We can still return 0 as only the root TLB will be affected
2455 		 * by a root ASID flush.
2456 		 */
2457 	}
2458 
2459 	return ret;
2460 }
2461 
2462 static void kvm_vz_vcpu_save_wired(struct kvm_vcpu *vcpu)
2463 {
2464 	unsigned int wired = read_gc0_wired();
2465 	struct kvm_mips_tlb *tlbs;
2466 	int i;
2467 
2468 	/* Expand the wired TLB array if necessary */
2469 	wired &= MIPSR6_WIRED_WIRED;
2470 	if (wired > vcpu->arch.wired_tlb_limit) {
2471 		tlbs = krealloc(vcpu->arch.wired_tlb, wired *
2472 				sizeof(*vcpu->arch.wired_tlb), GFP_ATOMIC);
2473 		if (WARN_ON(!tlbs)) {
2474 			/* Save whatever we can */
2475 			wired = vcpu->arch.wired_tlb_limit;
2476 		} else {
2477 			vcpu->arch.wired_tlb = tlbs;
2478 			vcpu->arch.wired_tlb_limit = wired;
2479 		}
2480 	}
2481 
2482 	if (wired)
2483 		/* Save wired entries from the guest TLB */
2484 		kvm_vz_save_guesttlb(vcpu->arch.wired_tlb, 0, wired);
2485 	/* Invalidate any dropped entries since last time */
2486 	for (i = wired; i < vcpu->arch.wired_tlb_used; ++i) {
2487 		vcpu->arch.wired_tlb[i].tlb_hi = UNIQUE_GUEST_ENTRYHI(i);
2488 		vcpu->arch.wired_tlb[i].tlb_lo[0] = 0;
2489 		vcpu->arch.wired_tlb[i].tlb_lo[1] = 0;
2490 		vcpu->arch.wired_tlb[i].tlb_mask = 0;
2491 	}
2492 	vcpu->arch.wired_tlb_used = wired;
2493 }
2494 
2495 static void kvm_vz_vcpu_load_wired(struct kvm_vcpu *vcpu)
2496 {
2497 	/* Load wired entries into the guest TLB */
2498 	if (vcpu->arch.wired_tlb)
2499 		kvm_vz_load_guesttlb(vcpu->arch.wired_tlb, 0,
2500 				     vcpu->arch.wired_tlb_used);
2501 }
2502 
2503 static void kvm_vz_vcpu_load_tlb(struct kvm_vcpu *vcpu, int cpu)
2504 {
2505 	struct kvm *kvm = vcpu->kvm;
2506 	struct mm_struct *gpa_mm = &kvm->arch.gpa_mm;
2507 	bool migrated;
2508 
2509 	/*
2510 	 * Are we entering guest context on a different CPU to last time?
2511 	 * If so, the VCPU's guest TLB state on this CPU may be stale.
2512 	 */
2513 	migrated = (vcpu->arch.last_exec_cpu != cpu);
2514 	vcpu->arch.last_exec_cpu = cpu;
2515 
2516 	/*
2517 	 * A vcpu's GuestID is set in GuestCtl1.ID when the vcpu is loaded and
2518 	 * remains set until another vcpu is loaded in.  As a rule GuestRID
2519 	 * remains zeroed when in root context unless the kernel is busy
2520 	 * manipulating guest tlb entries.
2521 	 */
2522 	if (cpu_has_guestid) {
2523 		/*
2524 		 * Check if our GuestID is of an older version and thus invalid.
2525 		 *
2526 		 * We also discard the stored GuestID if we've executed on
2527 		 * another CPU, as the guest mappings may have changed without
2528 		 * hypervisor knowledge.
2529 		 */
2530 		if (migrated ||
2531 		    (vcpu->arch.vzguestid[cpu] ^ guestid_cache(cpu)) &
2532 					GUESTID_VERSION_MASK) {
2533 			kvm_vz_get_new_guestid(cpu, vcpu);
2534 			vcpu->arch.vzguestid[cpu] = guestid_cache(cpu);
2535 			trace_kvm_guestid_change(vcpu,
2536 						 vcpu->arch.vzguestid[cpu]);
2537 		}
2538 
2539 		/* Restore GuestID */
2540 		change_c0_guestctl1(GUESTID_MASK, vcpu->arch.vzguestid[cpu]);
2541 	} else {
2542 		/*
2543 		 * The Guest TLB only stores a single guest's TLB state, so
2544 		 * flush it if another VCPU has executed on this CPU.
2545 		 *
2546 		 * We also flush if we've executed on another CPU, as the guest
2547 		 * mappings may have changed without hypervisor knowledge.
2548 		 */
2549 		if (migrated || last_exec_vcpu[cpu] != vcpu)
2550 			kvm_vz_local_flush_guesttlb_all();
2551 		last_exec_vcpu[cpu] = vcpu;
2552 
2553 		/*
2554 		 * Root ASID dealiases guest GPA mappings in the root TLB.
2555 		 * Allocate new root ASID if needed.
2556 		 */
2557 		if (cpumask_test_and_clear_cpu(cpu, &kvm->arch.asid_flush_mask))
2558 			get_new_mmu_context(gpa_mm);
2559 		else
2560 			check_mmu_context(gpa_mm);
2561 	}
2562 }
2563 
2564 static int kvm_vz_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2565 {
2566 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2567 	bool migrated, all;
2568 
2569 	/*
2570 	 * Have we migrated to a different CPU?
2571 	 * If so, any old guest TLB state may be stale.
2572 	 */
2573 	migrated = (vcpu->arch.last_sched_cpu != cpu);
2574 
2575 	/*
2576 	 * Was this the last VCPU to run on this CPU?
2577 	 * If not, any old guest state from this VCPU will have been clobbered.
2578 	 */
2579 	all = migrated || (last_vcpu[cpu] != vcpu);
2580 	last_vcpu[cpu] = vcpu;
2581 
2582 	/*
2583 	 * Restore CP0_Wired unconditionally as we clear it after use, and
2584 	 * restore wired guest TLB entries (while in guest context).
2585 	 */
2586 	kvm_restore_gc0_wired(cop0);
2587 	if (current->flags & PF_VCPU) {
2588 		tlbw_use_hazard();
2589 		kvm_vz_vcpu_load_tlb(vcpu, cpu);
2590 		kvm_vz_vcpu_load_wired(vcpu);
2591 	}
2592 
2593 	/*
2594 	 * Restore timer state regardless, as e.g. Cause.TI can change over time
2595 	 * if left unmaintained.
2596 	 */
2597 	kvm_vz_restore_timer(vcpu);
2598 
2599 	/* Set MC bit if we want to trace guest mode changes */
2600 	if (kvm_trace_guest_mode_change)
2601 		set_c0_guestctl0(MIPS_GCTL0_MC);
2602 	else
2603 		clear_c0_guestctl0(MIPS_GCTL0_MC);
2604 
2605 	/* Don't bother restoring registers multiple times unless necessary */
2606 	if (!all)
2607 		return 0;
2608 
2609 	/*
2610 	 * Restore config registers first, as some implementations restrict
2611 	 * writes to other registers when the corresponding feature bits aren't
2612 	 * set. For example Status.CU1 cannot be set unless Config1.FP is set.
2613 	 */
2614 	kvm_restore_gc0_config(cop0);
2615 	if (cpu_guest_has_conf1)
2616 		kvm_restore_gc0_config1(cop0);
2617 	if (cpu_guest_has_conf2)
2618 		kvm_restore_gc0_config2(cop0);
2619 	if (cpu_guest_has_conf3)
2620 		kvm_restore_gc0_config3(cop0);
2621 	if (cpu_guest_has_conf4)
2622 		kvm_restore_gc0_config4(cop0);
2623 	if (cpu_guest_has_conf5)
2624 		kvm_restore_gc0_config5(cop0);
2625 	if (cpu_guest_has_conf6)
2626 		kvm_restore_gc0_config6(cop0);
2627 	if (cpu_guest_has_conf7)
2628 		kvm_restore_gc0_config7(cop0);
2629 
2630 	kvm_restore_gc0_index(cop0);
2631 	kvm_restore_gc0_entrylo0(cop0);
2632 	kvm_restore_gc0_entrylo1(cop0);
2633 	kvm_restore_gc0_context(cop0);
2634 	if (cpu_guest_has_contextconfig)
2635 		kvm_restore_gc0_contextconfig(cop0);
2636 #ifdef CONFIG_64BIT
2637 	kvm_restore_gc0_xcontext(cop0);
2638 	if (cpu_guest_has_contextconfig)
2639 		kvm_restore_gc0_xcontextconfig(cop0);
2640 #endif
2641 	kvm_restore_gc0_pagemask(cop0);
2642 	kvm_restore_gc0_pagegrain(cop0);
2643 	kvm_restore_gc0_hwrena(cop0);
2644 	kvm_restore_gc0_badvaddr(cop0);
2645 	kvm_restore_gc0_entryhi(cop0);
2646 	kvm_restore_gc0_status(cop0);
2647 	kvm_restore_gc0_intctl(cop0);
2648 	kvm_restore_gc0_epc(cop0);
2649 	kvm_vz_write_gc0_ebase(kvm_read_sw_gc0_ebase(cop0));
2650 	if (cpu_guest_has_userlocal)
2651 		kvm_restore_gc0_userlocal(cop0);
2652 
2653 	kvm_restore_gc0_errorepc(cop0);
2654 
2655 	/* restore KScratch registers if enabled in guest */
2656 	if (cpu_guest_has_conf4) {
2657 		if (cpu_guest_has_kscr(2))
2658 			kvm_restore_gc0_kscratch1(cop0);
2659 		if (cpu_guest_has_kscr(3))
2660 			kvm_restore_gc0_kscratch2(cop0);
2661 		if (cpu_guest_has_kscr(4))
2662 			kvm_restore_gc0_kscratch3(cop0);
2663 		if (cpu_guest_has_kscr(5))
2664 			kvm_restore_gc0_kscratch4(cop0);
2665 		if (cpu_guest_has_kscr(6))
2666 			kvm_restore_gc0_kscratch5(cop0);
2667 		if (cpu_guest_has_kscr(7))
2668 			kvm_restore_gc0_kscratch6(cop0);
2669 	}
2670 
2671 	if (cpu_guest_has_badinstr)
2672 		kvm_restore_gc0_badinstr(cop0);
2673 	if (cpu_guest_has_badinstrp)
2674 		kvm_restore_gc0_badinstrp(cop0);
2675 
2676 	if (cpu_guest_has_segments) {
2677 		kvm_restore_gc0_segctl0(cop0);
2678 		kvm_restore_gc0_segctl1(cop0);
2679 		kvm_restore_gc0_segctl2(cop0);
2680 	}
2681 
2682 	/* restore HTW registers */
2683 	if (cpu_guest_has_htw || cpu_guest_has_ldpte) {
2684 		kvm_restore_gc0_pwbase(cop0);
2685 		kvm_restore_gc0_pwfield(cop0);
2686 		kvm_restore_gc0_pwsize(cop0);
2687 		kvm_restore_gc0_pwctl(cop0);
2688 	}
2689 
2690 	/* restore Root.GuestCtl2 from unused Guest guestctl2 register */
2691 	if (cpu_has_guestctl2)
2692 		write_c0_guestctl2(
2693 			cop0->reg[MIPS_CP0_GUESTCTL2][MIPS_CP0_GUESTCTL2_SEL]);
2694 
2695 	/*
2696 	 * We should clear linked load bit to break interrupted atomics. This
2697 	 * prevents a SC on the next VCPU from succeeding by matching a LL on
2698 	 * the previous VCPU.
2699 	 */
2700 	if (vcpu->kvm->created_vcpus > 1)
2701 		write_gc0_lladdr(0);
2702 
2703 	return 0;
2704 }
2705 
2706 static int kvm_vz_vcpu_put(struct kvm_vcpu *vcpu, int cpu)
2707 {
2708 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2709 
2710 	if (current->flags & PF_VCPU)
2711 		kvm_vz_vcpu_save_wired(vcpu);
2712 
2713 	kvm_lose_fpu(vcpu);
2714 
2715 	kvm_save_gc0_index(cop0);
2716 	kvm_save_gc0_entrylo0(cop0);
2717 	kvm_save_gc0_entrylo1(cop0);
2718 	kvm_save_gc0_context(cop0);
2719 	if (cpu_guest_has_contextconfig)
2720 		kvm_save_gc0_contextconfig(cop0);
2721 #ifdef CONFIG_64BIT
2722 	kvm_save_gc0_xcontext(cop0);
2723 	if (cpu_guest_has_contextconfig)
2724 		kvm_save_gc0_xcontextconfig(cop0);
2725 #endif
2726 	kvm_save_gc0_pagemask(cop0);
2727 	kvm_save_gc0_pagegrain(cop0);
2728 	kvm_save_gc0_wired(cop0);
2729 	/* allow wired TLB entries to be overwritten */
2730 	clear_gc0_wired(MIPSR6_WIRED_WIRED);
2731 	kvm_save_gc0_hwrena(cop0);
2732 	kvm_save_gc0_badvaddr(cop0);
2733 	kvm_save_gc0_entryhi(cop0);
2734 	kvm_save_gc0_status(cop0);
2735 	kvm_save_gc0_intctl(cop0);
2736 	kvm_save_gc0_epc(cop0);
2737 	kvm_write_sw_gc0_ebase(cop0, kvm_vz_read_gc0_ebase());
2738 	if (cpu_guest_has_userlocal)
2739 		kvm_save_gc0_userlocal(cop0);
2740 
2741 	/* only save implemented config registers */
2742 	kvm_save_gc0_config(cop0);
2743 	if (cpu_guest_has_conf1)
2744 		kvm_save_gc0_config1(cop0);
2745 	if (cpu_guest_has_conf2)
2746 		kvm_save_gc0_config2(cop0);
2747 	if (cpu_guest_has_conf3)
2748 		kvm_save_gc0_config3(cop0);
2749 	if (cpu_guest_has_conf4)
2750 		kvm_save_gc0_config4(cop0);
2751 	if (cpu_guest_has_conf5)
2752 		kvm_save_gc0_config5(cop0);
2753 	if (cpu_guest_has_conf6)
2754 		kvm_save_gc0_config6(cop0);
2755 	if (cpu_guest_has_conf7)
2756 		kvm_save_gc0_config7(cop0);
2757 
2758 	kvm_save_gc0_errorepc(cop0);
2759 
2760 	/* save KScratch registers if enabled in guest */
2761 	if (cpu_guest_has_conf4) {
2762 		if (cpu_guest_has_kscr(2))
2763 			kvm_save_gc0_kscratch1(cop0);
2764 		if (cpu_guest_has_kscr(3))
2765 			kvm_save_gc0_kscratch2(cop0);
2766 		if (cpu_guest_has_kscr(4))
2767 			kvm_save_gc0_kscratch3(cop0);
2768 		if (cpu_guest_has_kscr(5))
2769 			kvm_save_gc0_kscratch4(cop0);
2770 		if (cpu_guest_has_kscr(6))
2771 			kvm_save_gc0_kscratch5(cop0);
2772 		if (cpu_guest_has_kscr(7))
2773 			kvm_save_gc0_kscratch6(cop0);
2774 	}
2775 
2776 	if (cpu_guest_has_badinstr)
2777 		kvm_save_gc0_badinstr(cop0);
2778 	if (cpu_guest_has_badinstrp)
2779 		kvm_save_gc0_badinstrp(cop0);
2780 
2781 	if (cpu_guest_has_segments) {
2782 		kvm_save_gc0_segctl0(cop0);
2783 		kvm_save_gc0_segctl1(cop0);
2784 		kvm_save_gc0_segctl2(cop0);
2785 	}
2786 
2787 	/* save HTW registers if enabled in guest */
2788 	if (cpu_guest_has_ldpte || (cpu_guest_has_htw &&
2789 	    kvm_read_sw_gc0_config3(cop0) & MIPS_CONF3_PW)) {
2790 		kvm_save_gc0_pwbase(cop0);
2791 		kvm_save_gc0_pwfield(cop0);
2792 		kvm_save_gc0_pwsize(cop0);
2793 		kvm_save_gc0_pwctl(cop0);
2794 	}
2795 
2796 	kvm_vz_save_timer(vcpu);
2797 
2798 	/* save Root.GuestCtl2 in unused Guest guestctl2 register */
2799 	if (cpu_has_guestctl2)
2800 		cop0->reg[MIPS_CP0_GUESTCTL2][MIPS_CP0_GUESTCTL2_SEL] =
2801 			read_c0_guestctl2();
2802 
2803 	return 0;
2804 }
2805 
2806 /**
2807  * kvm_vz_resize_guest_vtlb() - Attempt to resize guest VTLB.
2808  * @size:	Number of guest VTLB entries (0 < @size <= root VTLB entries).
2809  *
2810  * Attempt to resize the guest VTLB by writing guest Config registers. This is
2811  * necessary for cores with a shared root/guest TLB to avoid overlap with wired
2812  * entries in the root VTLB.
2813  *
2814  * Returns:	The resulting guest VTLB size.
2815  */
2816 static unsigned int kvm_vz_resize_guest_vtlb(unsigned int size)
2817 {
2818 	unsigned int config4 = 0, ret = 0, limit;
2819 
2820 	/* Write MMUSize - 1 into guest Config registers */
2821 	if (cpu_guest_has_conf1)
2822 		change_gc0_config1(MIPS_CONF1_TLBS,
2823 				   (size - 1) << MIPS_CONF1_TLBS_SHIFT);
2824 	if (cpu_guest_has_conf4) {
2825 		config4 = read_gc0_config4();
2826 		if (cpu_has_mips_r6 || (config4 & MIPS_CONF4_MMUEXTDEF) ==
2827 		    MIPS_CONF4_MMUEXTDEF_VTLBSIZEEXT) {
2828 			config4 &= ~MIPS_CONF4_VTLBSIZEEXT;
2829 			config4 |= ((size - 1) >> MIPS_CONF1_TLBS_SIZE) <<
2830 				MIPS_CONF4_VTLBSIZEEXT_SHIFT;
2831 		} else if ((config4 & MIPS_CONF4_MMUEXTDEF) ==
2832 			   MIPS_CONF4_MMUEXTDEF_MMUSIZEEXT) {
2833 			config4 &= ~MIPS_CONF4_MMUSIZEEXT;
2834 			config4 |= ((size - 1) >> MIPS_CONF1_TLBS_SIZE) <<
2835 				MIPS_CONF4_MMUSIZEEXT_SHIFT;
2836 		}
2837 		write_gc0_config4(config4);
2838 	}
2839 
2840 	/*
2841 	 * Set Guest.Wired.Limit = 0 (no limit up to Guest.MMUSize-1), unless it
2842 	 * would exceed Root.Wired.Limit (clearing Guest.Wired.Wired so write
2843 	 * not dropped)
2844 	 */
2845 	if (cpu_has_mips_r6) {
2846 		limit = (read_c0_wired() & MIPSR6_WIRED_LIMIT) >>
2847 						MIPSR6_WIRED_LIMIT_SHIFT;
2848 		if (size - 1 <= limit)
2849 			limit = 0;
2850 		write_gc0_wired(limit << MIPSR6_WIRED_LIMIT_SHIFT);
2851 	}
2852 
2853 	/* Read back MMUSize - 1 */
2854 	back_to_back_c0_hazard();
2855 	if (cpu_guest_has_conf1)
2856 		ret = (read_gc0_config1() & MIPS_CONF1_TLBS) >>
2857 						MIPS_CONF1_TLBS_SHIFT;
2858 	if (config4) {
2859 		if (cpu_has_mips_r6 || (config4 & MIPS_CONF4_MMUEXTDEF) ==
2860 		    MIPS_CONF4_MMUEXTDEF_VTLBSIZEEXT)
2861 			ret |= ((config4 & MIPS_CONF4_VTLBSIZEEXT) >>
2862 				MIPS_CONF4_VTLBSIZEEXT_SHIFT) <<
2863 				MIPS_CONF1_TLBS_SIZE;
2864 		else if ((config4 & MIPS_CONF4_MMUEXTDEF) ==
2865 			 MIPS_CONF4_MMUEXTDEF_MMUSIZEEXT)
2866 			ret |= ((config4 & MIPS_CONF4_MMUSIZEEXT) >>
2867 				MIPS_CONF4_MMUSIZEEXT_SHIFT) <<
2868 				MIPS_CONF1_TLBS_SIZE;
2869 	}
2870 	return ret + 1;
2871 }
2872 
2873 static int kvm_vz_hardware_enable(void)
2874 {
2875 	unsigned int mmu_size, guest_mmu_size, ftlb_size;
2876 	u64 guest_cvmctl, cvmvmconfig;
2877 
2878 	switch (current_cpu_type()) {
2879 	case CPU_CAVIUM_OCTEON3:
2880 		/* Set up guest timer/perfcount IRQ lines */
2881 		guest_cvmctl = read_gc0_cvmctl();
2882 		guest_cvmctl &= ~CVMCTL_IPTI;
2883 		guest_cvmctl |= 7ull << CVMCTL_IPTI_SHIFT;
2884 		guest_cvmctl &= ~CVMCTL_IPPCI;
2885 		guest_cvmctl |= 6ull << CVMCTL_IPPCI_SHIFT;
2886 		write_gc0_cvmctl(guest_cvmctl);
2887 
2888 		cvmvmconfig = read_c0_cvmvmconfig();
2889 		/* No I/O hole translation. */
2890 		cvmvmconfig |= CVMVMCONF_DGHT;
2891 		/* Halve the root MMU size */
2892 		mmu_size = ((cvmvmconfig & CVMVMCONF_MMUSIZEM1)
2893 			    >> CVMVMCONF_MMUSIZEM1_S) + 1;
2894 		guest_mmu_size = mmu_size / 2;
2895 		mmu_size -= guest_mmu_size;
2896 		cvmvmconfig &= ~CVMVMCONF_RMMUSIZEM1;
2897 		cvmvmconfig |= mmu_size - 1;
2898 		write_c0_cvmvmconfig(cvmvmconfig);
2899 
2900 		/* Update our records */
2901 		current_cpu_data.tlbsize = mmu_size;
2902 		current_cpu_data.tlbsizevtlb = mmu_size;
2903 		current_cpu_data.guest.tlbsize = guest_mmu_size;
2904 
2905 		/* Flush moved entries in new (guest) context */
2906 		kvm_vz_local_flush_guesttlb_all();
2907 		break;
2908 	default:
2909 		/*
2910 		 * ImgTec cores tend to use a shared root/guest TLB. To avoid
2911 		 * overlap of root wired and guest entries, the guest TLB may
2912 		 * need resizing.
2913 		 */
2914 		mmu_size = current_cpu_data.tlbsizevtlb;
2915 		ftlb_size = current_cpu_data.tlbsize - mmu_size;
2916 
2917 		/* Try switching to maximum guest VTLB size for flush */
2918 		guest_mmu_size = kvm_vz_resize_guest_vtlb(mmu_size);
2919 		current_cpu_data.guest.tlbsize = guest_mmu_size + ftlb_size;
2920 		kvm_vz_local_flush_guesttlb_all();
2921 
2922 		/*
2923 		 * Reduce to make space for root wired entries and at least 2
2924 		 * root non-wired entries. This does assume that long-term wired
2925 		 * entries won't be added later.
2926 		 */
2927 		guest_mmu_size = mmu_size - num_wired_entries() - 2;
2928 		guest_mmu_size = kvm_vz_resize_guest_vtlb(guest_mmu_size);
2929 		current_cpu_data.guest.tlbsize = guest_mmu_size + ftlb_size;
2930 
2931 		/*
2932 		 * Write the VTLB size, but if another CPU has already written,
2933 		 * check it matches or we won't provide a consistent view to the
2934 		 * guest. If this ever happens it suggests an asymmetric number
2935 		 * of wired entries.
2936 		 */
2937 		if (cmpxchg(&kvm_vz_guest_vtlb_size, 0, guest_mmu_size) &&
2938 		    WARN(guest_mmu_size != kvm_vz_guest_vtlb_size,
2939 			 "Available guest VTLB size mismatch"))
2940 			return -EINVAL;
2941 		break;
2942 	}
2943 
2944 	/*
2945 	 * Enable virtualization features granting guest direct control of
2946 	 * certain features:
2947 	 * CP0=1:	Guest coprocessor 0 context.
2948 	 * AT=Guest:	Guest MMU.
2949 	 * CG=1:	Hit (virtual address) CACHE operations (optional).
2950 	 * CF=1:	Guest Config registers.
2951 	 * CGI=1:	Indexed flush CACHE operations (optional).
2952 	 */
2953 	write_c0_guestctl0(MIPS_GCTL0_CP0 |
2954 			   (MIPS_GCTL0_AT_GUEST << MIPS_GCTL0_AT_SHIFT) |
2955 			   MIPS_GCTL0_CG | MIPS_GCTL0_CF);
2956 	if (cpu_has_guestctl0ext) {
2957 		if (current_cpu_type() != CPU_LOONGSON64)
2958 			set_c0_guestctl0ext(MIPS_GCTL0EXT_CGI);
2959 		else
2960 			clear_c0_guestctl0ext(MIPS_GCTL0EXT_CGI);
2961 	}
2962 
2963 	if (cpu_has_guestid) {
2964 		write_c0_guestctl1(0);
2965 		kvm_vz_local_flush_roottlb_all_guests();
2966 
2967 		GUESTID_MASK = current_cpu_data.guestid_mask;
2968 		GUESTID_FIRST_VERSION = GUESTID_MASK + 1;
2969 		GUESTID_VERSION_MASK = ~GUESTID_MASK;
2970 
2971 		current_cpu_data.guestid_cache = GUESTID_FIRST_VERSION;
2972 	}
2973 
2974 	/* clear any pending injected virtual guest interrupts */
2975 	if (cpu_has_guestctl2)
2976 		clear_c0_guestctl2(0x3f << 10);
2977 
2978 #ifdef CONFIG_CPU_LOONGSON64
2979 	/* Control guest CCA attribute */
2980 	if (cpu_has_csr())
2981 		csr_writel(csr_readl(0xffffffec) | 0x1, 0xffffffec);
2982 #endif
2983 
2984 	return 0;
2985 }
2986 
2987 static void kvm_vz_hardware_disable(void)
2988 {
2989 	u64 cvmvmconfig;
2990 	unsigned int mmu_size;
2991 
2992 	/* Flush any remaining guest TLB entries */
2993 	kvm_vz_local_flush_guesttlb_all();
2994 
2995 	switch (current_cpu_type()) {
2996 	case CPU_CAVIUM_OCTEON3:
2997 		/*
2998 		 * Allocate whole TLB for root. Existing guest TLB entries will
2999 		 * change ownership to the root TLB. We should be safe though as
3000 		 * they've already been flushed above while in guest TLB.
3001 		 */
3002 		cvmvmconfig = read_c0_cvmvmconfig();
3003 		mmu_size = ((cvmvmconfig & CVMVMCONF_MMUSIZEM1)
3004 			    >> CVMVMCONF_MMUSIZEM1_S) + 1;
3005 		cvmvmconfig &= ~CVMVMCONF_RMMUSIZEM1;
3006 		cvmvmconfig |= mmu_size - 1;
3007 		write_c0_cvmvmconfig(cvmvmconfig);
3008 
3009 		/* Update our records */
3010 		current_cpu_data.tlbsize = mmu_size;
3011 		current_cpu_data.tlbsizevtlb = mmu_size;
3012 		current_cpu_data.guest.tlbsize = 0;
3013 
3014 		/* Flush moved entries in new (root) context */
3015 		local_flush_tlb_all();
3016 		break;
3017 	}
3018 
3019 	if (cpu_has_guestid) {
3020 		write_c0_guestctl1(0);
3021 		kvm_vz_local_flush_roottlb_all_guests();
3022 	}
3023 }
3024 
3025 static int kvm_vz_check_extension(struct kvm *kvm, long ext)
3026 {
3027 	int r;
3028 
3029 	switch (ext) {
3030 	case KVM_CAP_MIPS_VZ:
3031 		/* we wouldn't be here unless cpu_has_vz */
3032 		r = 1;
3033 		break;
3034 #ifdef CONFIG_64BIT
3035 	case KVM_CAP_MIPS_64BIT:
3036 		/* We support 64-bit registers/operations and addresses */
3037 		r = 2;
3038 		break;
3039 #endif
3040 	case KVM_CAP_IOEVENTFD:
3041 		r = 1;
3042 		break;
3043 	default:
3044 		r = 0;
3045 		break;
3046 	}
3047 
3048 	return r;
3049 }
3050 
3051 static int kvm_vz_vcpu_init(struct kvm_vcpu *vcpu)
3052 {
3053 	int i;
3054 
3055 	for_each_possible_cpu(i)
3056 		vcpu->arch.vzguestid[i] = 0;
3057 
3058 	return 0;
3059 }
3060 
3061 static void kvm_vz_vcpu_uninit(struct kvm_vcpu *vcpu)
3062 {
3063 	int cpu;
3064 
3065 	/*
3066 	 * If the VCPU is freed and reused as another VCPU, we don't want the
3067 	 * matching pointer wrongly hanging around in last_vcpu[] or
3068 	 * last_exec_vcpu[].
3069 	 */
3070 	for_each_possible_cpu(cpu) {
3071 		if (last_vcpu[cpu] == vcpu)
3072 			last_vcpu[cpu] = NULL;
3073 		if (last_exec_vcpu[cpu] == vcpu)
3074 			last_exec_vcpu[cpu] = NULL;
3075 	}
3076 }
3077 
3078 static int kvm_vz_vcpu_setup(struct kvm_vcpu *vcpu)
3079 {
3080 	struct mips_coproc *cop0 = vcpu->arch.cop0;
3081 	unsigned long count_hz = 100*1000*1000; /* default to 100 MHz */
3082 
3083 	/*
3084 	 * Start off the timer at the same frequency as the host timer, but the
3085 	 * soft timer doesn't handle frequencies greater than 1GHz yet.
3086 	 */
3087 	if (mips_hpt_frequency && mips_hpt_frequency <= NSEC_PER_SEC)
3088 		count_hz = mips_hpt_frequency;
3089 	kvm_mips_init_count(vcpu, count_hz);
3090 
3091 	/*
3092 	 * Initialize guest register state to valid architectural reset state.
3093 	 */
3094 
3095 	/* PageGrain */
3096 	if (cpu_has_mips_r5 || cpu_has_mips_r6)
3097 		kvm_write_sw_gc0_pagegrain(cop0, PG_RIE | PG_XIE | PG_IEC);
3098 	/* Wired */
3099 	if (cpu_has_mips_r6)
3100 		kvm_write_sw_gc0_wired(cop0,
3101 				       read_gc0_wired() & MIPSR6_WIRED_LIMIT);
3102 	/* Status */
3103 	kvm_write_sw_gc0_status(cop0, ST0_BEV | ST0_ERL);
3104 	if (cpu_has_mips_r5 || cpu_has_mips_r6)
3105 		kvm_change_sw_gc0_status(cop0, ST0_FR, read_gc0_status());
3106 	/* IntCtl */
3107 	kvm_write_sw_gc0_intctl(cop0, read_gc0_intctl() &
3108 				(INTCTLF_IPFDC | INTCTLF_IPPCI | INTCTLF_IPTI));
3109 	/* PRId */
3110 	kvm_write_sw_gc0_prid(cop0, boot_cpu_data.processor_id);
3111 	/* EBase */
3112 	kvm_write_sw_gc0_ebase(cop0, (s32)0x80000000 | vcpu->vcpu_id);
3113 	/* Config */
3114 	kvm_save_gc0_config(cop0);
3115 	/* architecturally writable (e.g. from guest) */
3116 	kvm_change_sw_gc0_config(cop0, CONF_CM_CMASK,
3117 				 _page_cachable_default >> _CACHE_SHIFT);
3118 	/* architecturally read only, but maybe writable from root */
3119 	kvm_change_sw_gc0_config(cop0, MIPS_CONF_MT, read_c0_config());
3120 	if (cpu_guest_has_conf1) {
3121 		kvm_set_sw_gc0_config(cop0, MIPS_CONF_M);
3122 		/* Config1 */
3123 		kvm_save_gc0_config1(cop0);
3124 		/* architecturally read only, but maybe writable from root */
3125 		kvm_clear_sw_gc0_config1(cop0, MIPS_CONF1_C2	|
3126 					       MIPS_CONF1_MD	|
3127 					       MIPS_CONF1_PC	|
3128 					       MIPS_CONF1_WR	|
3129 					       MIPS_CONF1_CA	|
3130 					       MIPS_CONF1_FP);
3131 	}
3132 	if (cpu_guest_has_conf2) {
3133 		kvm_set_sw_gc0_config1(cop0, MIPS_CONF_M);
3134 		/* Config2 */
3135 		kvm_save_gc0_config2(cop0);
3136 	}
3137 	if (cpu_guest_has_conf3) {
3138 		kvm_set_sw_gc0_config2(cop0, MIPS_CONF_M);
3139 		/* Config3 */
3140 		kvm_save_gc0_config3(cop0);
3141 		/* architecturally writable (e.g. from guest) */
3142 		kvm_clear_sw_gc0_config3(cop0, MIPS_CONF3_ISA_OE);
3143 		/* architecturally read only, but maybe writable from root */
3144 		kvm_clear_sw_gc0_config3(cop0, MIPS_CONF3_MSA	|
3145 					       MIPS_CONF3_BPG	|
3146 					       MIPS_CONF3_ULRI	|
3147 					       MIPS_CONF3_DSP	|
3148 					       MIPS_CONF3_CTXTC	|
3149 					       MIPS_CONF3_ITL	|
3150 					       MIPS_CONF3_LPA	|
3151 					       MIPS_CONF3_VEIC	|
3152 					       MIPS_CONF3_VINT	|
3153 					       MIPS_CONF3_SP	|
3154 					       MIPS_CONF3_CDMM	|
3155 					       MIPS_CONF3_MT	|
3156 					       MIPS_CONF3_SM	|
3157 					       MIPS_CONF3_TL);
3158 	}
3159 	if (cpu_guest_has_conf4) {
3160 		kvm_set_sw_gc0_config3(cop0, MIPS_CONF_M);
3161 		/* Config4 */
3162 		kvm_save_gc0_config4(cop0);
3163 	}
3164 	if (cpu_guest_has_conf5) {
3165 		kvm_set_sw_gc0_config4(cop0, MIPS_CONF_M);
3166 		/* Config5 */
3167 		kvm_save_gc0_config5(cop0);
3168 		/* architecturally writable (e.g. from guest) */
3169 		kvm_clear_sw_gc0_config5(cop0, MIPS_CONF5_K	|
3170 					       MIPS_CONF5_CV	|
3171 					       MIPS_CONF5_MSAEN	|
3172 					       MIPS_CONF5_UFE	|
3173 					       MIPS_CONF5_FRE	|
3174 					       MIPS_CONF5_SBRI	|
3175 					       MIPS_CONF5_UFR);
3176 		/* architecturally read only, but maybe writable from root */
3177 		kvm_clear_sw_gc0_config5(cop0, MIPS_CONF5_MRP);
3178 	}
3179 
3180 	if (cpu_guest_has_contextconfig) {
3181 		/* ContextConfig */
3182 		kvm_write_sw_gc0_contextconfig(cop0, 0x007ffff0);
3183 #ifdef CONFIG_64BIT
3184 		/* XContextConfig */
3185 		/* bits SEGBITS-13+3:4 set */
3186 		kvm_write_sw_gc0_xcontextconfig(cop0,
3187 					((1ull << (cpu_vmbits - 13)) - 1) << 4);
3188 #endif
3189 	}
3190 
3191 	/* Implementation dependent, use the legacy layout */
3192 	if (cpu_guest_has_segments) {
3193 		/* SegCtl0, SegCtl1, SegCtl2 */
3194 		kvm_write_sw_gc0_segctl0(cop0, 0x00200010);
3195 		kvm_write_sw_gc0_segctl1(cop0, 0x00000002 |
3196 				(_page_cachable_default >> _CACHE_SHIFT) <<
3197 						(16 + MIPS_SEGCFG_C_SHIFT));
3198 		kvm_write_sw_gc0_segctl2(cop0, 0x00380438);
3199 	}
3200 
3201 	/* reset HTW registers */
3202 	if (cpu_guest_has_htw && (cpu_has_mips_r5 || cpu_has_mips_r6)) {
3203 		/* PWField */
3204 		kvm_write_sw_gc0_pwfield(cop0, 0x0c30c302);
3205 		/* PWSize */
3206 		kvm_write_sw_gc0_pwsize(cop0, 1 << MIPS_PWSIZE_PTW_SHIFT);
3207 	}
3208 
3209 	/* start with no pending virtual guest interrupts */
3210 	if (cpu_has_guestctl2)
3211 		cop0->reg[MIPS_CP0_GUESTCTL2][MIPS_CP0_GUESTCTL2_SEL] = 0;
3212 
3213 	/* Put PC at reset vector */
3214 	vcpu->arch.pc = CKSEG1ADDR(0x1fc00000);
3215 
3216 	return 0;
3217 }
3218 
3219 static void kvm_vz_flush_shadow_all(struct kvm *kvm)
3220 {
3221 	if (cpu_has_guestid) {
3222 		/* Flush GuestID for each VCPU individually */
3223 		kvm_flush_remote_tlbs(kvm);
3224 	} else {
3225 		/*
3226 		 * For each CPU there is a single GPA ASID used by all VCPUs in
3227 		 * the VM, so it doesn't make sense for the VCPUs to handle
3228 		 * invalidation of these ASIDs individually.
3229 		 *
3230 		 * Instead mark all CPUs as needing ASID invalidation in
3231 		 * asid_flush_mask, and just use kvm_flush_remote_tlbs(kvm) to
3232 		 * kick any running VCPUs so they check asid_flush_mask.
3233 		 */
3234 		cpumask_setall(&kvm->arch.asid_flush_mask);
3235 		kvm_flush_remote_tlbs(kvm);
3236 	}
3237 }
3238 
3239 static void kvm_vz_flush_shadow_memslot(struct kvm *kvm,
3240 					const struct kvm_memory_slot *slot)
3241 {
3242 	kvm_vz_flush_shadow_all(kvm);
3243 }
3244 
3245 static void kvm_vz_vcpu_reenter(struct kvm_run *run, struct kvm_vcpu *vcpu)
3246 {
3247 	int cpu = smp_processor_id();
3248 	int preserve_guest_tlb;
3249 
3250 	preserve_guest_tlb = kvm_vz_check_requests(vcpu, cpu);
3251 
3252 	if (preserve_guest_tlb)
3253 		kvm_vz_vcpu_save_wired(vcpu);
3254 
3255 	kvm_vz_vcpu_load_tlb(vcpu, cpu);
3256 
3257 	if (preserve_guest_tlb)
3258 		kvm_vz_vcpu_load_wired(vcpu);
3259 }
3260 
3261 static int kvm_vz_vcpu_run(struct kvm_run *run, struct kvm_vcpu *vcpu)
3262 {
3263 	int cpu = smp_processor_id();
3264 	int r;
3265 
3266 	kvm_vz_acquire_htimer(vcpu);
3267 	/* Check if we have any exceptions/interrupts pending */
3268 	kvm_mips_deliver_interrupts(vcpu, read_gc0_cause());
3269 
3270 	kvm_vz_check_requests(vcpu, cpu);
3271 	kvm_vz_vcpu_load_tlb(vcpu, cpu);
3272 	kvm_vz_vcpu_load_wired(vcpu);
3273 
3274 	r = vcpu->arch.vcpu_run(run, vcpu);
3275 
3276 	kvm_vz_vcpu_save_wired(vcpu);
3277 
3278 	return r;
3279 }
3280 
3281 static struct kvm_mips_callbacks kvm_vz_callbacks = {
3282 	.handle_cop_unusable = kvm_trap_vz_handle_cop_unusable,
3283 	.handle_tlb_mod = kvm_trap_vz_handle_tlb_st_miss,
3284 	.handle_tlb_ld_miss = kvm_trap_vz_handle_tlb_ld_miss,
3285 	.handle_tlb_st_miss = kvm_trap_vz_handle_tlb_st_miss,
3286 	.handle_addr_err_st = kvm_trap_vz_no_handler,
3287 	.handle_addr_err_ld = kvm_trap_vz_no_handler,
3288 	.handle_syscall = kvm_trap_vz_no_handler,
3289 	.handle_res_inst = kvm_trap_vz_no_handler,
3290 	.handle_break = kvm_trap_vz_no_handler,
3291 	.handle_msa_disabled = kvm_trap_vz_handle_msa_disabled,
3292 	.handle_guest_exit = kvm_trap_vz_handle_guest_exit,
3293 
3294 	.hardware_enable = kvm_vz_hardware_enable,
3295 	.hardware_disable = kvm_vz_hardware_disable,
3296 	.check_extension = kvm_vz_check_extension,
3297 	.vcpu_init = kvm_vz_vcpu_init,
3298 	.vcpu_uninit = kvm_vz_vcpu_uninit,
3299 	.vcpu_setup = kvm_vz_vcpu_setup,
3300 	.flush_shadow_all = kvm_vz_flush_shadow_all,
3301 	.flush_shadow_memslot = kvm_vz_flush_shadow_memslot,
3302 	.gva_to_gpa = kvm_vz_gva_to_gpa_cb,
3303 	.queue_timer_int = kvm_vz_queue_timer_int_cb,
3304 	.dequeue_timer_int = kvm_vz_dequeue_timer_int_cb,
3305 	.queue_io_int = kvm_vz_queue_io_int_cb,
3306 	.dequeue_io_int = kvm_vz_dequeue_io_int_cb,
3307 	.irq_deliver = kvm_vz_irq_deliver_cb,
3308 	.irq_clear = kvm_vz_irq_clear_cb,
3309 	.num_regs = kvm_vz_num_regs,
3310 	.copy_reg_indices = kvm_vz_copy_reg_indices,
3311 	.get_one_reg = kvm_vz_get_one_reg,
3312 	.set_one_reg = kvm_vz_set_one_reg,
3313 	.vcpu_load = kvm_vz_vcpu_load,
3314 	.vcpu_put = kvm_vz_vcpu_put,
3315 	.vcpu_run = kvm_vz_vcpu_run,
3316 	.vcpu_reenter = kvm_vz_vcpu_reenter,
3317 };
3318 
3319 int kvm_mips_emulation_init(struct kvm_mips_callbacks **install_callbacks)
3320 {
3321 	if (!cpu_has_vz)
3322 		return -ENODEV;
3323 
3324 	/*
3325 	 * VZ requires at least 2 KScratch registers, so it should have been
3326 	 * possible to allocate pgd_reg.
3327 	 */
3328 	if (WARN(pgd_reg == -1,
3329 		 "pgd_reg not allocated even though cpu_has_vz\n"))
3330 		return -ENODEV;
3331 
3332 	pr_info("Starting KVM with MIPS VZ extensions\n");
3333 
3334 	*install_callbacks = &kvm_vz_callbacks;
3335 	return 0;
3336 }
3337