xref: /linux/arch/mips/kvm/vz.c (revision 662f11d55ffd02933e1bd275d732b97eddccf870)
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 #ifdef CONFIG_CPU_LOONGSON64
33 #include "loongson_regs.h"
34 #endif
35 
36 #include "trace.h"
37 
38 /* Pointers to last VCPU loaded on each physical CPU */
39 static struct kvm_vcpu *last_vcpu[NR_CPUS];
40 /* Pointers to last VCPU executed on each physical CPU */
41 static struct kvm_vcpu *last_exec_vcpu[NR_CPUS];
42 
43 /*
44  * Number of guest VTLB entries to use, so we can catch inconsistency between
45  * CPUs.
46  */
47 static unsigned int kvm_vz_guest_vtlb_size;
48 
49 static inline long kvm_vz_read_gc0_ebase(void)
50 {
51 	if (sizeof(long) == 8 && cpu_has_ebase_wg)
52 		return read_gc0_ebase_64();
53 	else
54 		return read_gc0_ebase();
55 }
56 
57 static inline void kvm_vz_write_gc0_ebase(long v)
58 {
59 	/*
60 	 * First write with WG=1 to write upper bits, then write again in case
61 	 * WG should be left at 0.
62 	 * write_gc0_ebase_64() is no longer UNDEFINED since R6.
63 	 */
64 	if (sizeof(long) == 8 &&
65 	    (cpu_has_mips64r6 || cpu_has_ebase_wg)) {
66 		write_gc0_ebase_64(v | MIPS_EBASE_WG);
67 		write_gc0_ebase_64(v);
68 	} else {
69 		write_gc0_ebase(v | MIPS_EBASE_WG);
70 		write_gc0_ebase(v);
71 	}
72 }
73 
74 /*
75  * These Config bits may be writable by the guest:
76  * Config:	[K23, KU] (!TLB), K0
77  * Config1:	(none)
78  * Config2:	[TU, SU] (impl)
79  * Config3:	ISAOnExc
80  * Config4:	FTLBPageSize
81  * Config5:	K, CV, MSAEn, UFE, FRE, SBRI, UFR
82  */
83 
84 static inline unsigned int kvm_vz_config_guest_wrmask(struct kvm_vcpu *vcpu)
85 {
86 	return CONF_CM_CMASK;
87 }
88 
89 static inline unsigned int kvm_vz_config1_guest_wrmask(struct kvm_vcpu *vcpu)
90 {
91 	return 0;
92 }
93 
94 static inline unsigned int kvm_vz_config2_guest_wrmask(struct kvm_vcpu *vcpu)
95 {
96 	return 0;
97 }
98 
99 static inline unsigned int kvm_vz_config3_guest_wrmask(struct kvm_vcpu *vcpu)
100 {
101 	return MIPS_CONF3_ISA_OE;
102 }
103 
104 static inline unsigned int kvm_vz_config4_guest_wrmask(struct kvm_vcpu *vcpu)
105 {
106 	/* no need to be exact */
107 	return MIPS_CONF4_VFTLBPAGESIZE;
108 }
109 
110 static inline unsigned int kvm_vz_config5_guest_wrmask(struct kvm_vcpu *vcpu)
111 {
112 	unsigned int mask = MIPS_CONF5_K | MIPS_CONF5_CV | MIPS_CONF5_SBRI;
113 
114 	/* Permit MSAEn changes if MSA supported and enabled */
115 	if (kvm_mips_guest_has_msa(&vcpu->arch))
116 		mask |= MIPS_CONF5_MSAEN;
117 
118 	/*
119 	 * Permit guest FPU mode changes if FPU is enabled and the relevant
120 	 * feature exists according to FIR register.
121 	 */
122 	if (kvm_mips_guest_has_fpu(&vcpu->arch)) {
123 		if (cpu_has_ufr)
124 			mask |= MIPS_CONF5_UFR;
125 		if (cpu_has_fre)
126 			mask |= MIPS_CONF5_FRE | MIPS_CONF5_UFE;
127 	}
128 
129 	return mask;
130 }
131 
132 static inline unsigned int kvm_vz_config6_guest_wrmask(struct kvm_vcpu *vcpu)
133 {
134 	return LOONGSON_CONF6_INTIMER | LOONGSON_CONF6_EXTIMER;
135 }
136 
137 /*
138  * VZ optionally allows these additional Config bits to be written by root:
139  * Config:	M, [MT]
140  * Config1:	M, [MMUSize-1, C2, MD, PC, WR, CA], FP
141  * Config2:	M
142  * Config3:	M, MSAP, [BPG], ULRI, [DSP2P, DSPP], CTXTC, [ITL, LPA, VEIC,
143  *		VInt, SP, CDMM, MT, SM, TL]
144  * Config4:	M, [VTLBSizeExt, MMUSizeExt]
145  * Config5:	MRP
146  */
147 
148 static inline unsigned int kvm_vz_config_user_wrmask(struct kvm_vcpu *vcpu)
149 {
150 	return kvm_vz_config_guest_wrmask(vcpu) | MIPS_CONF_M;
151 }
152 
153 static inline unsigned int kvm_vz_config1_user_wrmask(struct kvm_vcpu *vcpu)
154 {
155 	unsigned int mask = kvm_vz_config1_guest_wrmask(vcpu) | MIPS_CONF_M;
156 
157 	/* Permit FPU to be present if FPU is supported */
158 	if (kvm_mips_guest_can_have_fpu(&vcpu->arch))
159 		mask |= MIPS_CONF1_FP;
160 
161 	return mask;
162 }
163 
164 static inline unsigned int kvm_vz_config2_user_wrmask(struct kvm_vcpu *vcpu)
165 {
166 	return kvm_vz_config2_guest_wrmask(vcpu) | MIPS_CONF_M;
167 }
168 
169 static inline unsigned int kvm_vz_config3_user_wrmask(struct kvm_vcpu *vcpu)
170 {
171 	unsigned int mask = kvm_vz_config3_guest_wrmask(vcpu) | MIPS_CONF_M |
172 		MIPS_CONF3_ULRI | MIPS_CONF3_CTXTC;
173 
174 	/* Permit MSA to be present if MSA is supported */
175 	if (kvm_mips_guest_can_have_msa(&vcpu->arch))
176 		mask |= MIPS_CONF3_MSA;
177 
178 	return mask;
179 }
180 
181 static inline unsigned int kvm_vz_config4_user_wrmask(struct kvm_vcpu *vcpu)
182 {
183 	return kvm_vz_config4_guest_wrmask(vcpu) | MIPS_CONF_M;
184 }
185 
186 static inline unsigned int kvm_vz_config5_user_wrmask(struct kvm_vcpu *vcpu)
187 {
188 	return kvm_vz_config5_guest_wrmask(vcpu) | MIPS_CONF5_MRP;
189 }
190 
191 static inline unsigned int kvm_vz_config6_user_wrmask(struct kvm_vcpu *vcpu)
192 {
193 	return kvm_vz_config6_guest_wrmask(vcpu) |
194 		LOONGSON_CONF6_SFBEN | LOONGSON_CONF6_FTLBDIS;
195 }
196 
197 static gpa_t kvm_vz_gva_to_gpa_cb(gva_t gva)
198 {
199 	/* VZ guest has already converted gva to gpa */
200 	return gva;
201 }
202 
203 static void kvm_vz_queue_irq(struct kvm_vcpu *vcpu, unsigned int priority)
204 {
205 	set_bit(priority, &vcpu->arch.pending_exceptions);
206 	clear_bit(priority, &vcpu->arch.pending_exceptions_clr);
207 }
208 
209 static void kvm_vz_dequeue_irq(struct kvm_vcpu *vcpu, unsigned int priority)
210 {
211 	clear_bit(priority, &vcpu->arch.pending_exceptions);
212 	set_bit(priority, &vcpu->arch.pending_exceptions_clr);
213 }
214 
215 static void kvm_vz_queue_timer_int_cb(struct kvm_vcpu *vcpu)
216 {
217 	/*
218 	 * timer expiry is asynchronous to vcpu execution therefore defer guest
219 	 * cp0 accesses
220 	 */
221 	kvm_vz_queue_irq(vcpu, MIPS_EXC_INT_TIMER);
222 }
223 
224 static void kvm_vz_dequeue_timer_int_cb(struct kvm_vcpu *vcpu)
225 {
226 	/*
227 	 * timer expiry is asynchronous to vcpu execution therefore defer guest
228 	 * cp0 accesses
229 	 */
230 	kvm_vz_dequeue_irq(vcpu, MIPS_EXC_INT_TIMER);
231 }
232 
233 static void kvm_vz_queue_io_int_cb(struct kvm_vcpu *vcpu,
234 				   struct kvm_mips_interrupt *irq)
235 {
236 	int intr = (int)irq->irq;
237 
238 	/*
239 	 * interrupts are asynchronous to vcpu execution therefore defer guest
240 	 * cp0 accesses
241 	 */
242 	kvm_vz_queue_irq(vcpu, kvm_irq_to_priority(intr));
243 }
244 
245 static void kvm_vz_dequeue_io_int_cb(struct kvm_vcpu *vcpu,
246 				     struct kvm_mips_interrupt *irq)
247 {
248 	int intr = (int)irq->irq;
249 
250 	/*
251 	 * interrupts are asynchronous to vcpu execution therefore defer guest
252 	 * cp0 accesses
253 	 */
254 	kvm_vz_dequeue_irq(vcpu, kvm_irq_to_priority(-intr));
255 }
256 
257 static int kvm_vz_irq_deliver_cb(struct kvm_vcpu *vcpu, unsigned int priority,
258 				 u32 cause)
259 {
260 	u32 irq = (priority < MIPS_EXC_MAX) ?
261 		kvm_priority_to_irq[priority] : 0;
262 
263 	switch (priority) {
264 	case MIPS_EXC_INT_TIMER:
265 		set_gc0_cause(C_TI);
266 		break;
267 
268 	case MIPS_EXC_INT_IO_1:
269 	case MIPS_EXC_INT_IO_2:
270 	case MIPS_EXC_INT_IPI_1:
271 	case MIPS_EXC_INT_IPI_2:
272 		if (cpu_has_guestctl2)
273 			set_c0_guestctl2(irq);
274 		else
275 			set_gc0_cause(irq);
276 		break;
277 
278 	default:
279 		break;
280 	}
281 
282 	clear_bit(priority, &vcpu->arch.pending_exceptions);
283 	return 1;
284 }
285 
286 static int kvm_vz_irq_clear_cb(struct kvm_vcpu *vcpu, unsigned int priority,
287 			       u32 cause)
288 {
289 	u32 irq = (priority < MIPS_EXC_MAX) ?
290 		kvm_priority_to_irq[priority] : 0;
291 
292 	switch (priority) {
293 	case MIPS_EXC_INT_TIMER:
294 		/*
295 		 * Explicitly clear irq associated with Cause.IP[IPTI]
296 		 * if GuestCtl2 virtual interrupt register not
297 		 * supported or if not using GuestCtl2 Hardware Clear.
298 		 */
299 		if (cpu_has_guestctl2) {
300 			if (!(read_c0_guestctl2() & (irq << 14)))
301 				clear_c0_guestctl2(irq);
302 		} else {
303 			clear_gc0_cause(irq);
304 		}
305 		break;
306 
307 	case MIPS_EXC_INT_IO_1:
308 	case MIPS_EXC_INT_IO_2:
309 	case MIPS_EXC_INT_IPI_1:
310 	case MIPS_EXC_INT_IPI_2:
311 		/* Clear GuestCtl2.VIP irq if not using Hardware Clear */
312 		if (cpu_has_guestctl2) {
313 			if (!(read_c0_guestctl2() & (irq << 14)))
314 				clear_c0_guestctl2(irq);
315 		} else {
316 			clear_gc0_cause(irq);
317 		}
318 		break;
319 
320 	default:
321 		break;
322 	}
323 
324 	clear_bit(priority, &vcpu->arch.pending_exceptions_clr);
325 	return 1;
326 }
327 
328 /*
329  * VZ guest timer handling.
330  */
331 
332 /**
333  * kvm_vz_should_use_htimer() - Find whether to use the VZ hard guest timer.
334  * @vcpu:	Virtual CPU.
335  *
336  * Returns:	true if the VZ GTOffset & real guest CP0_Count should be used
337  *		instead of software emulation of guest timer.
338  *		false otherwise.
339  */
340 static bool kvm_vz_should_use_htimer(struct kvm_vcpu *vcpu)
341 {
342 	if (kvm_mips_count_disabled(vcpu))
343 		return false;
344 
345 	/* Chosen frequency must match real frequency */
346 	if (mips_hpt_frequency != vcpu->arch.count_hz)
347 		return false;
348 
349 	/* We don't support a CP0_GTOffset with fewer bits than CP0_Count */
350 	if (current_cpu_data.gtoffset_mask != 0xffffffff)
351 		return false;
352 
353 	return true;
354 }
355 
356 /**
357  * _kvm_vz_restore_stimer() - Restore soft timer state.
358  * @vcpu:	Virtual CPU.
359  * @compare:	CP0_Compare register value, restored by caller.
360  * @cause:	CP0_Cause register to restore.
361  *
362  * Restore VZ state relating to the soft timer. The hard timer can be enabled
363  * later.
364  */
365 static void _kvm_vz_restore_stimer(struct kvm_vcpu *vcpu, u32 compare,
366 				   u32 cause)
367 {
368 	/*
369 	 * Avoid spurious counter interrupts by setting Guest CP0_Count to just
370 	 * after Guest CP0_Compare.
371 	 */
372 	write_c0_gtoffset(compare - read_c0_count());
373 
374 	back_to_back_c0_hazard();
375 	write_gc0_cause(cause);
376 }
377 
378 /**
379  * _kvm_vz_restore_htimer() - Restore hard timer state.
380  * @vcpu:	Virtual CPU.
381  * @compare:	CP0_Compare register value, restored by caller.
382  * @cause:	CP0_Cause register to restore.
383  *
384  * Restore hard timer Guest.Count & Guest.Cause taking care to preserve the
385  * value of Guest.CP0_Cause.TI while restoring Guest.CP0_Cause.
386  */
387 static void _kvm_vz_restore_htimer(struct kvm_vcpu *vcpu,
388 				   u32 compare, u32 cause)
389 {
390 	u32 start_count, after_count;
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 	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_vcpu *vcpu)
878 {
879 	struct mips_coproc *cop0 = vcpu->arch.cop0;
880 	enum emulation_result er = EMULATE_DONE;
881 	u32 rt, rd, sel;
882 	unsigned long curr_pc;
883 	unsigned long val;
884 
885 	/*
886 	 * Update PC and hold onto current PC in case there is
887 	 * an error and we want to rollback the PC
888 	 */
889 	curr_pc = vcpu->arch.pc;
890 	er = update_pc(vcpu, cause);
891 	if (er == EMULATE_FAIL)
892 		return er;
893 
894 	if (inst.co_format.co) {
895 		switch (inst.co_format.func) {
896 		case wait_op:
897 			er = kvm_mips_emul_wait(vcpu);
898 			break;
899 		default:
900 			er = EMULATE_FAIL;
901 		}
902 	} else {
903 		rt = inst.c0r_format.rt;
904 		rd = inst.c0r_format.rd;
905 		sel = inst.c0r_format.sel;
906 
907 		switch (inst.c0r_format.rs) {
908 		case dmfc_op:
909 		case mfc_op:
910 #ifdef CONFIG_KVM_MIPS_DEBUG_COP0_COUNTERS
911 			cop0->stat[rd][sel]++;
912 #endif
913 			if (rd == MIPS_CP0_COUNT &&
914 			    sel == 0) {			/* Count */
915 				val = kvm_mips_read_count(vcpu);
916 			} else if (rd == MIPS_CP0_COMPARE &&
917 				   sel == 0) {		/* Compare */
918 				val = read_gc0_compare();
919 			} else if (rd == MIPS_CP0_LLADDR &&
920 				   sel == 0) {		/* LLAddr */
921 				if (cpu_guest_has_rw_llb)
922 					val = read_gc0_lladdr() &
923 						MIPS_LLADDR_LLB;
924 				else
925 					val = 0;
926 			} else if (rd == MIPS_CP0_LLADDR &&
927 				   sel == 1 &&		/* MAAR */
928 				   cpu_guest_has_maar &&
929 				   !cpu_guest_has_dyn_maar) {
930 				/* MAARI must be in range */
931 				BUG_ON(kvm_read_sw_gc0_maari(cop0) >=
932 						ARRAY_SIZE(vcpu->arch.maar));
933 				val = vcpu->arch.maar[
934 					kvm_read_sw_gc0_maari(cop0)];
935 			} else if ((rd == MIPS_CP0_PRID &&
936 				    (sel == 0 ||	/* PRid */
937 				     sel == 2 ||	/* CDMMBase */
938 				     sel == 3)) ||	/* CMGCRBase */
939 				   (rd == MIPS_CP0_STATUS &&
940 				    (sel == 2 ||	/* SRSCtl */
941 				     sel == 3)) ||	/* SRSMap */
942 				   (rd == MIPS_CP0_CONFIG &&
943 				    (sel == 6 ||	/* Config6 */
944 				     sel == 7)) ||	/* Config7 */
945 				   (rd == MIPS_CP0_LLADDR &&
946 				    (sel == 2) &&	/* MAARI */
947 				    cpu_guest_has_maar &&
948 				    !cpu_guest_has_dyn_maar) ||
949 				   (rd == MIPS_CP0_ERRCTL &&
950 				    (sel == 0))) {	/* ErrCtl */
951 				val = cop0->reg[rd][sel];
952 #ifdef CONFIG_CPU_LOONGSON64
953 			} else if (rd == MIPS_CP0_DIAG &&
954 				   (sel == 0)) {	/* Diag */
955 				val = cop0->reg[rd][sel];
956 #endif
957 			} else {
958 				val = 0;
959 				er = EMULATE_FAIL;
960 			}
961 
962 			if (er != EMULATE_FAIL) {
963 				/* Sign extend */
964 				if (inst.c0r_format.rs == mfc_op)
965 					val = (int)val;
966 				vcpu->arch.gprs[rt] = val;
967 			}
968 
969 			trace_kvm_hwr(vcpu, (inst.c0r_format.rs == mfc_op) ?
970 					KVM_TRACE_MFC0 : KVM_TRACE_DMFC0,
971 				      KVM_TRACE_COP0(rd, sel), val);
972 			break;
973 
974 		case dmtc_op:
975 		case mtc_op:
976 #ifdef CONFIG_KVM_MIPS_DEBUG_COP0_COUNTERS
977 			cop0->stat[rd][sel]++;
978 #endif
979 			val = vcpu->arch.gprs[rt];
980 			trace_kvm_hwr(vcpu, (inst.c0r_format.rs == mtc_op) ?
981 					KVM_TRACE_MTC0 : KVM_TRACE_DMTC0,
982 				      KVM_TRACE_COP0(rd, sel), val);
983 
984 			if (rd == MIPS_CP0_COUNT &&
985 			    sel == 0) {			/* Count */
986 				kvm_vz_lose_htimer(vcpu);
987 				kvm_mips_write_count(vcpu, vcpu->arch.gprs[rt]);
988 			} else if (rd == MIPS_CP0_COMPARE &&
989 				   sel == 0) {		/* Compare */
990 				kvm_mips_write_compare(vcpu,
991 						       vcpu->arch.gprs[rt],
992 						       true);
993 			} else if (rd == MIPS_CP0_LLADDR &&
994 				   sel == 0) {		/* LLAddr */
995 				/*
996 				 * P5600 generates GPSI on guest MTC0 LLAddr.
997 				 * Only allow the guest to clear LLB.
998 				 */
999 				if (cpu_guest_has_rw_llb &&
1000 				    !(val & MIPS_LLADDR_LLB))
1001 					write_gc0_lladdr(0);
1002 			} else if (rd == MIPS_CP0_LLADDR &&
1003 				   sel == 1 &&		/* MAAR */
1004 				   cpu_guest_has_maar &&
1005 				   !cpu_guest_has_dyn_maar) {
1006 				val = mips_process_maar(inst.c0r_format.rs,
1007 							val);
1008 
1009 				/* MAARI must be in range */
1010 				BUG_ON(kvm_read_sw_gc0_maari(cop0) >=
1011 						ARRAY_SIZE(vcpu->arch.maar));
1012 				vcpu->arch.maar[kvm_read_sw_gc0_maari(cop0)] =
1013 									val;
1014 			} else if (rd == MIPS_CP0_LLADDR &&
1015 				   (sel == 2) &&	/* MAARI */
1016 				   cpu_guest_has_maar &&
1017 				   !cpu_guest_has_dyn_maar) {
1018 				kvm_write_maari(vcpu, val);
1019 			} else if (rd == MIPS_CP0_CONFIG &&
1020 				   (sel == 6)) {
1021 				cop0->reg[rd][sel] = (int)val;
1022 			} else if (rd == MIPS_CP0_ERRCTL &&
1023 				   (sel == 0)) {	/* ErrCtl */
1024 				/* ignore the written value */
1025 #ifdef CONFIG_CPU_LOONGSON64
1026 			} else if (rd == MIPS_CP0_DIAG &&
1027 				   (sel == 0)) {	/* Diag */
1028 				unsigned long flags;
1029 
1030 				local_irq_save(flags);
1031 				if (val & LOONGSON_DIAG_BTB) {
1032 					/* Flush BTB */
1033 					set_c0_diag(LOONGSON_DIAG_BTB);
1034 				}
1035 				if (val & LOONGSON_DIAG_ITLB) {
1036 					/* Flush ITLB */
1037 					set_c0_diag(LOONGSON_DIAG_ITLB);
1038 				}
1039 				if (val & LOONGSON_DIAG_DTLB) {
1040 					/* Flush DTLB */
1041 					set_c0_diag(LOONGSON_DIAG_DTLB);
1042 				}
1043 				if (val & LOONGSON_DIAG_VTLB) {
1044 					/* Flush VTLB */
1045 					kvm_loongson_clear_guest_vtlb();
1046 				}
1047 				if (val & LOONGSON_DIAG_FTLB) {
1048 					/* Flush FTLB */
1049 					kvm_loongson_clear_guest_ftlb();
1050 				}
1051 				local_irq_restore(flags);
1052 #endif
1053 			} else {
1054 				er = EMULATE_FAIL;
1055 			}
1056 			break;
1057 
1058 		default:
1059 			er = EMULATE_FAIL;
1060 			break;
1061 		}
1062 	}
1063 	/* Rollback PC only if emulation was unsuccessful */
1064 	if (er == EMULATE_FAIL) {
1065 		kvm_err("[%#lx]%s: unsupported cop0 instruction 0x%08x\n",
1066 			curr_pc, __func__, inst.word);
1067 
1068 		vcpu->arch.pc = curr_pc;
1069 	}
1070 
1071 	return er;
1072 }
1073 
1074 static enum emulation_result kvm_vz_gpsi_cache(union mips_instruction inst,
1075 					       u32 *opc, u32 cause,
1076 					       struct kvm_vcpu *vcpu)
1077 {
1078 	enum emulation_result er = EMULATE_DONE;
1079 	u32 cache, op_inst, op, base;
1080 	s16 offset;
1081 	struct kvm_vcpu_arch *arch = &vcpu->arch;
1082 	unsigned long va, curr_pc;
1083 
1084 	/*
1085 	 * Update PC and hold onto current PC in case there is
1086 	 * an error and we want to rollback the PC
1087 	 */
1088 	curr_pc = vcpu->arch.pc;
1089 	er = update_pc(vcpu, cause);
1090 	if (er == EMULATE_FAIL)
1091 		return er;
1092 
1093 	base = inst.i_format.rs;
1094 	op_inst = inst.i_format.rt;
1095 	if (cpu_has_mips_r6)
1096 		offset = inst.spec3_format.simmediate;
1097 	else
1098 		offset = inst.i_format.simmediate;
1099 	cache = op_inst & CacheOp_Cache;
1100 	op = op_inst & CacheOp_Op;
1101 
1102 	va = arch->gprs[base] + offset;
1103 
1104 	kvm_debug("CACHE (cache: %#x, op: %#x, base[%d]: %#lx, offset: %#x\n",
1105 		  cache, op, base, arch->gprs[base], offset);
1106 
1107 	/* Secondary or tirtiary cache ops ignored */
1108 	if (cache != Cache_I && cache != Cache_D)
1109 		return EMULATE_DONE;
1110 
1111 	switch (op_inst) {
1112 	case Index_Invalidate_I:
1113 		flush_icache_line_indexed(va);
1114 		return EMULATE_DONE;
1115 	case Index_Writeback_Inv_D:
1116 		flush_dcache_line_indexed(va);
1117 		return EMULATE_DONE;
1118 	case Hit_Invalidate_I:
1119 	case Hit_Invalidate_D:
1120 	case Hit_Writeback_Inv_D:
1121 		if (boot_cpu_type() == CPU_CAVIUM_OCTEON3) {
1122 			/* We can just flush entire icache */
1123 			local_flush_icache_range(0, 0);
1124 			return EMULATE_DONE;
1125 		}
1126 
1127 		/* So far, other platforms support guest hit cache ops */
1128 		break;
1129 	default:
1130 		break;
1131 	}
1132 
1133 	kvm_err("@ %#lx/%#lx CACHE (cache: %#x, op: %#x, base[%d]: %#lx, offset: %#x\n",
1134 		curr_pc, vcpu->arch.gprs[31], cache, op, base, arch->gprs[base],
1135 		offset);
1136 	/* Rollback PC */
1137 	vcpu->arch.pc = curr_pc;
1138 
1139 	return EMULATE_FAIL;
1140 }
1141 
1142 #ifdef CONFIG_CPU_LOONGSON64
1143 static enum emulation_result kvm_vz_gpsi_lwc2(union mips_instruction inst,
1144 					      u32 *opc, u32 cause,
1145 					      struct kvm_vcpu *vcpu)
1146 {
1147 	unsigned int rs, rd;
1148 	unsigned int hostcfg;
1149 	unsigned long curr_pc;
1150 	enum emulation_result er = EMULATE_DONE;
1151 
1152 	/*
1153 	 * Update PC and hold onto current PC in case there is
1154 	 * an error and we want to rollback the PC
1155 	 */
1156 	curr_pc = vcpu->arch.pc;
1157 	er = update_pc(vcpu, cause);
1158 	if (er == EMULATE_FAIL)
1159 		return er;
1160 
1161 	rs = inst.loongson3_lscsr_format.rs;
1162 	rd = inst.loongson3_lscsr_format.rd;
1163 	switch (inst.loongson3_lscsr_format.fr) {
1164 	case 0x8:  /* Read CPUCFG */
1165 		++vcpu->stat.vz_cpucfg_exits;
1166 		hostcfg = read_cpucfg(vcpu->arch.gprs[rs]);
1167 
1168 		switch (vcpu->arch.gprs[rs]) {
1169 		case LOONGSON_CFG0:
1170 			vcpu->arch.gprs[rd] = 0x14c000;
1171 			break;
1172 		case LOONGSON_CFG1:
1173 			hostcfg &= (LOONGSON_CFG1_FP | LOONGSON_CFG1_MMI |
1174 				    LOONGSON_CFG1_MSA1 | LOONGSON_CFG1_MSA2 |
1175 				    LOONGSON_CFG1_SFBP);
1176 			vcpu->arch.gprs[rd] = hostcfg;
1177 			break;
1178 		case LOONGSON_CFG2:
1179 			hostcfg &= (LOONGSON_CFG2_LEXT1 | LOONGSON_CFG2_LEXT2 |
1180 				    LOONGSON_CFG2_LEXT3 | LOONGSON_CFG2_LSPW);
1181 			vcpu->arch.gprs[rd] = hostcfg;
1182 			break;
1183 		case LOONGSON_CFG3:
1184 			vcpu->arch.gprs[rd] = hostcfg;
1185 			break;
1186 		default:
1187 			/* Don't export any other advanced features to guest */
1188 			vcpu->arch.gprs[rd] = 0;
1189 			break;
1190 		}
1191 		break;
1192 
1193 	default:
1194 		kvm_err("lwc2 emulate not impl %d rs %lx @%lx\n",
1195 			inst.loongson3_lscsr_format.fr, vcpu->arch.gprs[rs], curr_pc);
1196 		er = EMULATE_FAIL;
1197 		break;
1198 	}
1199 
1200 	/* Rollback PC only if emulation was unsuccessful */
1201 	if (er == EMULATE_FAIL) {
1202 		kvm_err("[%#lx]%s: unsupported lwc2 instruction 0x%08x 0x%08x\n",
1203 			curr_pc, __func__, inst.word, inst.loongson3_lscsr_format.fr);
1204 
1205 		vcpu->arch.pc = curr_pc;
1206 	}
1207 
1208 	return er;
1209 }
1210 #endif
1211 
1212 static enum emulation_result kvm_trap_vz_handle_gpsi(u32 cause, u32 *opc,
1213 						     struct kvm_vcpu *vcpu)
1214 {
1215 	enum emulation_result er = EMULATE_DONE;
1216 	struct kvm_vcpu_arch *arch = &vcpu->arch;
1217 	union mips_instruction inst;
1218 	int rd, rt, sel;
1219 	int err;
1220 
1221 	/*
1222 	 *  Fetch the instruction.
1223 	 */
1224 	if (cause & CAUSEF_BD)
1225 		opc += 1;
1226 	err = kvm_get_badinstr(opc, vcpu, &inst.word);
1227 	if (err)
1228 		return EMULATE_FAIL;
1229 
1230 	switch (inst.r_format.opcode) {
1231 	case cop0_op:
1232 		er = kvm_vz_gpsi_cop0(inst, opc, cause, vcpu);
1233 		break;
1234 #ifndef CONFIG_CPU_MIPSR6
1235 	case cache_op:
1236 		trace_kvm_exit(vcpu, KVM_TRACE_EXIT_CACHE);
1237 		er = kvm_vz_gpsi_cache(inst, opc, cause, vcpu);
1238 		break;
1239 #endif
1240 #ifdef CONFIG_CPU_LOONGSON64
1241 	case lwc2_op:
1242 		er = kvm_vz_gpsi_lwc2(inst, opc, cause, vcpu);
1243 		break;
1244 #endif
1245 	case spec3_op:
1246 		switch (inst.spec3_format.func) {
1247 #ifdef CONFIG_CPU_MIPSR6
1248 		case cache6_op:
1249 			trace_kvm_exit(vcpu, KVM_TRACE_EXIT_CACHE);
1250 			er = kvm_vz_gpsi_cache(inst, opc, cause, vcpu);
1251 			break;
1252 #endif
1253 		case rdhwr_op:
1254 			if (inst.r_format.rs || (inst.r_format.re >> 3))
1255 				goto unknown;
1256 
1257 			rd = inst.r_format.rd;
1258 			rt = inst.r_format.rt;
1259 			sel = inst.r_format.re & 0x7;
1260 
1261 			switch (rd) {
1262 			case MIPS_HWR_CC:	/* Read count register */
1263 				arch->gprs[rt] =
1264 					(long)(int)kvm_mips_read_count(vcpu);
1265 				break;
1266 			default:
1267 				trace_kvm_hwr(vcpu, KVM_TRACE_RDHWR,
1268 					      KVM_TRACE_HWR(rd, sel), 0);
1269 				goto unknown;
1270 			}
1271 
1272 			trace_kvm_hwr(vcpu, KVM_TRACE_RDHWR,
1273 				      KVM_TRACE_HWR(rd, sel), arch->gprs[rt]);
1274 
1275 			er = update_pc(vcpu, cause);
1276 			break;
1277 		default:
1278 			goto unknown;
1279 		}
1280 		break;
1281 unknown:
1282 
1283 	default:
1284 		kvm_err("GPSI exception not supported (%p/%#x)\n",
1285 				opc, inst.word);
1286 		kvm_arch_vcpu_dump_regs(vcpu);
1287 		er = EMULATE_FAIL;
1288 		break;
1289 	}
1290 
1291 	return er;
1292 }
1293 
1294 static enum emulation_result kvm_trap_vz_handle_gsfc(u32 cause, u32 *opc,
1295 						     struct kvm_vcpu *vcpu)
1296 {
1297 	enum emulation_result er = EMULATE_DONE;
1298 	struct kvm_vcpu_arch *arch = &vcpu->arch;
1299 	union mips_instruction inst;
1300 	int err;
1301 
1302 	/*
1303 	 *  Fetch the instruction.
1304 	 */
1305 	if (cause & CAUSEF_BD)
1306 		opc += 1;
1307 	err = kvm_get_badinstr(opc, vcpu, &inst.word);
1308 	if (err)
1309 		return EMULATE_FAIL;
1310 
1311 	/* complete MTC0 on behalf of guest and advance EPC */
1312 	if (inst.c0r_format.opcode == cop0_op &&
1313 	    inst.c0r_format.rs == mtc_op &&
1314 	    inst.c0r_format.z == 0) {
1315 		int rt = inst.c0r_format.rt;
1316 		int rd = inst.c0r_format.rd;
1317 		int sel = inst.c0r_format.sel;
1318 		unsigned int val = arch->gprs[rt];
1319 		unsigned int old_val, change;
1320 
1321 		trace_kvm_hwr(vcpu, KVM_TRACE_MTC0, KVM_TRACE_COP0(rd, sel),
1322 			      val);
1323 
1324 		if ((rd == MIPS_CP0_STATUS) && (sel == 0)) {
1325 			/* FR bit should read as zero if no FPU */
1326 			if (!kvm_mips_guest_has_fpu(&vcpu->arch))
1327 				val &= ~(ST0_CU1 | ST0_FR);
1328 
1329 			/*
1330 			 * Also don't allow FR to be set if host doesn't support
1331 			 * it.
1332 			 */
1333 			if (!(boot_cpu_data.fpu_id & MIPS_FPIR_F64))
1334 				val &= ~ST0_FR;
1335 
1336 			old_val = read_gc0_status();
1337 			change = val ^ old_val;
1338 
1339 			if (change & ST0_FR) {
1340 				/*
1341 				 * FPU and Vector register state is made
1342 				 * UNPREDICTABLE by a change of FR, so don't
1343 				 * even bother saving it.
1344 				 */
1345 				kvm_drop_fpu(vcpu);
1346 			}
1347 
1348 			/*
1349 			 * If MSA state is already live, it is undefined how it
1350 			 * interacts with FR=0 FPU state, and we don't want to
1351 			 * hit reserved instruction exceptions trying to save
1352 			 * the MSA state later when CU=1 && FR=1, so play it
1353 			 * safe and save it first.
1354 			 */
1355 			if (change & ST0_CU1 && !(val & ST0_FR) &&
1356 			    vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA)
1357 				kvm_lose_fpu(vcpu);
1358 
1359 			write_gc0_status(val);
1360 		} else if ((rd == MIPS_CP0_CAUSE) && (sel == 0)) {
1361 			u32 old_cause = read_gc0_cause();
1362 			u32 change = old_cause ^ val;
1363 
1364 			/* DC bit enabling/disabling timer? */
1365 			if (change & CAUSEF_DC) {
1366 				if (val & CAUSEF_DC) {
1367 					kvm_vz_lose_htimer(vcpu);
1368 					kvm_mips_count_disable_cause(vcpu);
1369 				} else {
1370 					kvm_mips_count_enable_cause(vcpu);
1371 				}
1372 			}
1373 
1374 			/* Only certain bits are RW to the guest */
1375 			change &= (CAUSEF_DC | CAUSEF_IV | CAUSEF_WP |
1376 				   CAUSEF_IP0 | CAUSEF_IP1);
1377 
1378 			/* WP can only be cleared */
1379 			change &= ~CAUSEF_WP | old_cause;
1380 
1381 			write_gc0_cause(old_cause ^ change);
1382 		} else if ((rd == MIPS_CP0_STATUS) && (sel == 1)) { /* IntCtl */
1383 			write_gc0_intctl(val);
1384 		} else if ((rd == MIPS_CP0_CONFIG) && (sel == 5)) {
1385 			old_val = read_gc0_config5();
1386 			change = val ^ old_val;
1387 			/* Handle changes in FPU/MSA modes */
1388 			preempt_disable();
1389 
1390 			/*
1391 			 * Propagate FRE changes immediately if the FPU
1392 			 * context is already loaded.
1393 			 */
1394 			if (change & MIPS_CONF5_FRE &&
1395 			    vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU)
1396 				change_c0_config5(MIPS_CONF5_FRE, val);
1397 
1398 			preempt_enable();
1399 
1400 			val = old_val ^
1401 				(change & kvm_vz_config5_guest_wrmask(vcpu));
1402 			write_gc0_config5(val);
1403 		} else {
1404 			kvm_err("Handle GSFC, unsupported field change @ %p: %#x\n",
1405 			    opc, inst.word);
1406 			er = EMULATE_FAIL;
1407 		}
1408 
1409 		if (er != EMULATE_FAIL)
1410 			er = update_pc(vcpu, cause);
1411 	} else {
1412 		kvm_err("Handle GSFC, unrecognized instruction @ %p: %#x\n",
1413 			opc, inst.word);
1414 		er = EMULATE_FAIL;
1415 	}
1416 
1417 	return er;
1418 }
1419 
1420 static enum emulation_result kvm_trap_vz_handle_ghfc(u32 cause, u32 *opc,
1421 						     struct kvm_vcpu *vcpu)
1422 {
1423 	/*
1424 	 * Presumably this is due to MC (guest mode change), so lets trace some
1425 	 * relevant info.
1426 	 */
1427 	trace_kvm_guest_mode_change(vcpu);
1428 
1429 	return EMULATE_DONE;
1430 }
1431 
1432 static enum emulation_result kvm_trap_vz_handle_hc(u32 cause, u32 *opc,
1433 						   struct kvm_vcpu *vcpu)
1434 {
1435 	enum emulation_result er;
1436 	union mips_instruction inst;
1437 	unsigned long curr_pc;
1438 	int err;
1439 
1440 	if (cause & CAUSEF_BD)
1441 		opc += 1;
1442 	err = kvm_get_badinstr(opc, vcpu, &inst.word);
1443 	if (err)
1444 		return EMULATE_FAIL;
1445 
1446 	/*
1447 	 * Update PC and hold onto current PC in case there is
1448 	 * an error and we want to rollback the PC
1449 	 */
1450 	curr_pc = vcpu->arch.pc;
1451 	er = update_pc(vcpu, cause);
1452 	if (er == EMULATE_FAIL)
1453 		return er;
1454 
1455 	er = kvm_mips_emul_hypcall(vcpu, inst);
1456 	if (er == EMULATE_FAIL)
1457 		vcpu->arch.pc = curr_pc;
1458 
1459 	return er;
1460 }
1461 
1462 static enum emulation_result kvm_trap_vz_no_handler_guest_exit(u32 gexccode,
1463 							u32 cause,
1464 							u32 *opc,
1465 							struct kvm_vcpu *vcpu)
1466 {
1467 	u32 inst;
1468 
1469 	/*
1470 	 *  Fetch the instruction.
1471 	 */
1472 	if (cause & CAUSEF_BD)
1473 		opc += 1;
1474 	kvm_get_badinstr(opc, vcpu, &inst);
1475 
1476 	kvm_err("Guest Exception Code: %d not yet handled @ PC: %p, inst: 0x%08x  Status: %#x\n",
1477 		gexccode, opc, inst, read_gc0_status());
1478 
1479 	return EMULATE_FAIL;
1480 }
1481 
1482 static int kvm_trap_vz_handle_guest_exit(struct kvm_vcpu *vcpu)
1483 {
1484 	u32 *opc = (u32 *) vcpu->arch.pc;
1485 	u32 cause = vcpu->arch.host_cp0_cause;
1486 	enum emulation_result er = EMULATE_DONE;
1487 	u32 gexccode = (vcpu->arch.host_cp0_guestctl0 &
1488 			MIPS_GCTL0_GEXC) >> MIPS_GCTL0_GEXC_SHIFT;
1489 	int ret = RESUME_GUEST;
1490 
1491 	trace_kvm_exit(vcpu, KVM_TRACE_EXIT_GEXCCODE_BASE + gexccode);
1492 	switch (gexccode) {
1493 	case MIPS_GCTL0_GEXC_GPSI:
1494 		++vcpu->stat.vz_gpsi_exits;
1495 		er = kvm_trap_vz_handle_gpsi(cause, opc, vcpu);
1496 		break;
1497 	case MIPS_GCTL0_GEXC_GSFC:
1498 		++vcpu->stat.vz_gsfc_exits;
1499 		er = kvm_trap_vz_handle_gsfc(cause, opc, vcpu);
1500 		break;
1501 	case MIPS_GCTL0_GEXC_HC:
1502 		++vcpu->stat.vz_hc_exits;
1503 		er = kvm_trap_vz_handle_hc(cause, opc, vcpu);
1504 		break;
1505 	case MIPS_GCTL0_GEXC_GRR:
1506 		++vcpu->stat.vz_grr_exits;
1507 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1508 						       vcpu);
1509 		break;
1510 	case MIPS_GCTL0_GEXC_GVA:
1511 		++vcpu->stat.vz_gva_exits;
1512 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1513 						       vcpu);
1514 		break;
1515 	case MIPS_GCTL0_GEXC_GHFC:
1516 		++vcpu->stat.vz_ghfc_exits;
1517 		er = kvm_trap_vz_handle_ghfc(cause, opc, vcpu);
1518 		break;
1519 	case MIPS_GCTL0_GEXC_GPA:
1520 		++vcpu->stat.vz_gpa_exits;
1521 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1522 						       vcpu);
1523 		break;
1524 	default:
1525 		++vcpu->stat.vz_resvd_exits;
1526 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1527 						       vcpu);
1528 		break;
1529 
1530 	}
1531 
1532 	if (er == EMULATE_DONE) {
1533 		ret = RESUME_GUEST;
1534 	} else if (er == EMULATE_HYPERCALL) {
1535 		ret = kvm_mips_handle_hypcall(vcpu);
1536 	} else {
1537 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1538 		ret = RESUME_HOST;
1539 	}
1540 	return ret;
1541 }
1542 
1543 /**
1544  * kvm_trap_vz_handle_cop_unusuable() - Guest used unusable coprocessor.
1545  * @vcpu:	Virtual CPU context.
1546  *
1547  * Handle when the guest attempts to use a coprocessor which hasn't been allowed
1548  * by the root context.
1549  */
1550 static int kvm_trap_vz_handle_cop_unusable(struct kvm_vcpu *vcpu)
1551 {
1552 	u32 cause = vcpu->arch.host_cp0_cause;
1553 	enum emulation_result er = EMULATE_FAIL;
1554 	int ret = RESUME_GUEST;
1555 
1556 	if (((cause & CAUSEF_CE) >> CAUSEB_CE) == 1) {
1557 		/*
1558 		 * If guest FPU not present, the FPU operation should have been
1559 		 * treated as a reserved instruction!
1560 		 * If FPU already in use, we shouldn't get this at all.
1561 		 */
1562 		if (WARN_ON(!kvm_mips_guest_has_fpu(&vcpu->arch) ||
1563 			    vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU)) {
1564 			preempt_enable();
1565 			return EMULATE_FAIL;
1566 		}
1567 
1568 		kvm_own_fpu(vcpu);
1569 		er = EMULATE_DONE;
1570 	}
1571 	/* other coprocessors not handled */
1572 
1573 	switch (er) {
1574 	case EMULATE_DONE:
1575 		ret = RESUME_GUEST;
1576 		break;
1577 
1578 	case EMULATE_FAIL:
1579 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1580 		ret = RESUME_HOST;
1581 		break;
1582 
1583 	default:
1584 		BUG();
1585 	}
1586 	return ret;
1587 }
1588 
1589 /**
1590  * kvm_trap_vz_handle_msa_disabled() - Guest used MSA while disabled in root.
1591  * @vcpu:	Virtual CPU context.
1592  *
1593  * Handle when the guest attempts to use MSA when it is disabled in the root
1594  * context.
1595  */
1596 static int kvm_trap_vz_handle_msa_disabled(struct kvm_vcpu *vcpu)
1597 {
1598 	/*
1599 	 * If MSA not present or not exposed to guest or FR=0, the MSA operation
1600 	 * should have been treated as a reserved instruction!
1601 	 * Same if CU1=1, FR=0.
1602 	 * If MSA already in use, we shouldn't get this at all.
1603 	 */
1604 	if (!kvm_mips_guest_has_msa(&vcpu->arch) ||
1605 	    (read_gc0_status() & (ST0_CU1 | ST0_FR)) == ST0_CU1 ||
1606 	    !(read_gc0_config5() & MIPS_CONF5_MSAEN) ||
1607 	    vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA) {
1608 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1609 		return RESUME_HOST;
1610 	}
1611 
1612 	kvm_own_msa(vcpu);
1613 
1614 	return RESUME_GUEST;
1615 }
1616 
1617 static int kvm_trap_vz_handle_tlb_ld_miss(struct kvm_vcpu *vcpu)
1618 {
1619 	struct kvm_run *run = vcpu->run;
1620 	u32 *opc = (u32 *) vcpu->arch.pc;
1621 	u32 cause = vcpu->arch.host_cp0_cause;
1622 	ulong badvaddr = vcpu->arch.host_cp0_badvaddr;
1623 	union mips_instruction inst;
1624 	enum emulation_result er = EMULATE_DONE;
1625 	int err, ret = RESUME_GUEST;
1626 
1627 	if (kvm_mips_handle_vz_root_tlb_fault(badvaddr, vcpu, false)) {
1628 		/* A code fetch fault doesn't count as an MMIO */
1629 		if (kvm_is_ifetch_fault(&vcpu->arch)) {
1630 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1631 			return RESUME_HOST;
1632 		}
1633 
1634 		/* Fetch the instruction */
1635 		if (cause & CAUSEF_BD)
1636 			opc += 1;
1637 		err = kvm_get_badinstr(opc, vcpu, &inst.word);
1638 		if (err) {
1639 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1640 			return RESUME_HOST;
1641 		}
1642 
1643 		/* Treat as MMIO */
1644 		er = kvm_mips_emulate_load(inst, cause, vcpu);
1645 		if (er == EMULATE_FAIL) {
1646 			kvm_err("Guest Emulate Load from MMIO space failed: PC: %p, BadVaddr: %#lx\n",
1647 				opc, badvaddr);
1648 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1649 		}
1650 	}
1651 
1652 	if (er == EMULATE_DONE) {
1653 		ret = RESUME_GUEST;
1654 	} else if (er == EMULATE_DO_MMIO) {
1655 		run->exit_reason = KVM_EXIT_MMIO;
1656 		ret = RESUME_HOST;
1657 	} else {
1658 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1659 		ret = RESUME_HOST;
1660 	}
1661 	return ret;
1662 }
1663 
1664 static int kvm_trap_vz_handle_tlb_st_miss(struct kvm_vcpu *vcpu)
1665 {
1666 	struct kvm_run *run = vcpu->run;
1667 	u32 *opc = (u32 *) vcpu->arch.pc;
1668 	u32 cause = vcpu->arch.host_cp0_cause;
1669 	ulong badvaddr = vcpu->arch.host_cp0_badvaddr;
1670 	union mips_instruction inst;
1671 	enum emulation_result er = EMULATE_DONE;
1672 	int err;
1673 	int ret = RESUME_GUEST;
1674 
1675 	/* Just try the access again if we couldn't do the translation */
1676 	if (kvm_vz_badvaddr_to_gpa(vcpu, badvaddr, &badvaddr))
1677 		return RESUME_GUEST;
1678 	vcpu->arch.host_cp0_badvaddr = badvaddr;
1679 
1680 	if (kvm_mips_handle_vz_root_tlb_fault(badvaddr, vcpu, true)) {
1681 		/* Fetch the instruction */
1682 		if (cause & CAUSEF_BD)
1683 			opc += 1;
1684 		err = kvm_get_badinstr(opc, vcpu, &inst.word);
1685 		if (err) {
1686 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1687 			return RESUME_HOST;
1688 		}
1689 
1690 		/* Treat as MMIO */
1691 		er = kvm_mips_emulate_store(inst, cause, vcpu);
1692 		if (er == EMULATE_FAIL) {
1693 			kvm_err("Guest Emulate Store to MMIO space failed: PC: %p, BadVaddr: %#lx\n",
1694 				opc, badvaddr);
1695 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1696 		}
1697 	}
1698 
1699 	if (er == EMULATE_DONE) {
1700 		ret = RESUME_GUEST;
1701 	} else if (er == EMULATE_DO_MMIO) {
1702 		run->exit_reason = KVM_EXIT_MMIO;
1703 		ret = RESUME_HOST;
1704 	} else {
1705 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1706 		ret = RESUME_HOST;
1707 	}
1708 	return ret;
1709 }
1710 
1711 static u64 kvm_vz_get_one_regs[] = {
1712 	KVM_REG_MIPS_CP0_INDEX,
1713 	KVM_REG_MIPS_CP0_ENTRYLO0,
1714 	KVM_REG_MIPS_CP0_ENTRYLO1,
1715 	KVM_REG_MIPS_CP0_CONTEXT,
1716 	KVM_REG_MIPS_CP0_PAGEMASK,
1717 	KVM_REG_MIPS_CP0_PAGEGRAIN,
1718 	KVM_REG_MIPS_CP0_WIRED,
1719 	KVM_REG_MIPS_CP0_HWRENA,
1720 	KVM_REG_MIPS_CP0_BADVADDR,
1721 	KVM_REG_MIPS_CP0_COUNT,
1722 	KVM_REG_MIPS_CP0_ENTRYHI,
1723 	KVM_REG_MIPS_CP0_COMPARE,
1724 	KVM_REG_MIPS_CP0_STATUS,
1725 	KVM_REG_MIPS_CP0_INTCTL,
1726 	KVM_REG_MIPS_CP0_CAUSE,
1727 	KVM_REG_MIPS_CP0_EPC,
1728 	KVM_REG_MIPS_CP0_PRID,
1729 	KVM_REG_MIPS_CP0_EBASE,
1730 	KVM_REG_MIPS_CP0_CONFIG,
1731 	KVM_REG_MIPS_CP0_CONFIG1,
1732 	KVM_REG_MIPS_CP0_CONFIG2,
1733 	KVM_REG_MIPS_CP0_CONFIG3,
1734 	KVM_REG_MIPS_CP0_CONFIG4,
1735 	KVM_REG_MIPS_CP0_CONFIG5,
1736 	KVM_REG_MIPS_CP0_CONFIG6,
1737 #ifdef CONFIG_64BIT
1738 	KVM_REG_MIPS_CP0_XCONTEXT,
1739 #endif
1740 	KVM_REG_MIPS_CP0_ERROREPC,
1741 
1742 	KVM_REG_MIPS_COUNT_CTL,
1743 	KVM_REG_MIPS_COUNT_RESUME,
1744 	KVM_REG_MIPS_COUNT_HZ,
1745 };
1746 
1747 static u64 kvm_vz_get_one_regs_contextconfig[] = {
1748 	KVM_REG_MIPS_CP0_CONTEXTCONFIG,
1749 #ifdef CONFIG_64BIT
1750 	KVM_REG_MIPS_CP0_XCONTEXTCONFIG,
1751 #endif
1752 };
1753 
1754 static u64 kvm_vz_get_one_regs_segments[] = {
1755 	KVM_REG_MIPS_CP0_SEGCTL0,
1756 	KVM_REG_MIPS_CP0_SEGCTL1,
1757 	KVM_REG_MIPS_CP0_SEGCTL2,
1758 };
1759 
1760 static u64 kvm_vz_get_one_regs_htw[] = {
1761 	KVM_REG_MIPS_CP0_PWBASE,
1762 	KVM_REG_MIPS_CP0_PWFIELD,
1763 	KVM_REG_MIPS_CP0_PWSIZE,
1764 	KVM_REG_MIPS_CP0_PWCTL,
1765 };
1766 
1767 static u64 kvm_vz_get_one_regs_kscratch[] = {
1768 	KVM_REG_MIPS_CP0_KSCRATCH1,
1769 	KVM_REG_MIPS_CP0_KSCRATCH2,
1770 	KVM_REG_MIPS_CP0_KSCRATCH3,
1771 	KVM_REG_MIPS_CP0_KSCRATCH4,
1772 	KVM_REG_MIPS_CP0_KSCRATCH5,
1773 	KVM_REG_MIPS_CP0_KSCRATCH6,
1774 };
1775 
1776 static unsigned long kvm_vz_num_regs(struct kvm_vcpu *vcpu)
1777 {
1778 	unsigned long ret;
1779 
1780 	ret = ARRAY_SIZE(kvm_vz_get_one_regs);
1781 	if (cpu_guest_has_userlocal)
1782 		++ret;
1783 	if (cpu_guest_has_badinstr)
1784 		++ret;
1785 	if (cpu_guest_has_badinstrp)
1786 		++ret;
1787 	if (cpu_guest_has_contextconfig)
1788 		ret += ARRAY_SIZE(kvm_vz_get_one_regs_contextconfig);
1789 	if (cpu_guest_has_segments)
1790 		ret += ARRAY_SIZE(kvm_vz_get_one_regs_segments);
1791 	if (cpu_guest_has_htw || cpu_guest_has_ldpte)
1792 		ret += ARRAY_SIZE(kvm_vz_get_one_regs_htw);
1793 	if (cpu_guest_has_maar && !cpu_guest_has_dyn_maar)
1794 		ret += 1 + ARRAY_SIZE(vcpu->arch.maar);
1795 	ret += __arch_hweight8(cpu_data[0].guest.kscratch_mask);
1796 
1797 	return ret;
1798 }
1799 
1800 static int kvm_vz_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *indices)
1801 {
1802 	u64 index;
1803 	unsigned int i;
1804 
1805 	if (copy_to_user(indices, kvm_vz_get_one_regs,
1806 			 sizeof(kvm_vz_get_one_regs)))
1807 		return -EFAULT;
1808 	indices += ARRAY_SIZE(kvm_vz_get_one_regs);
1809 
1810 	if (cpu_guest_has_userlocal) {
1811 		index = KVM_REG_MIPS_CP0_USERLOCAL;
1812 		if (copy_to_user(indices, &index, sizeof(index)))
1813 			return -EFAULT;
1814 		++indices;
1815 	}
1816 	if (cpu_guest_has_badinstr) {
1817 		index = KVM_REG_MIPS_CP0_BADINSTR;
1818 		if (copy_to_user(indices, &index, sizeof(index)))
1819 			return -EFAULT;
1820 		++indices;
1821 	}
1822 	if (cpu_guest_has_badinstrp) {
1823 		index = KVM_REG_MIPS_CP0_BADINSTRP;
1824 		if (copy_to_user(indices, &index, sizeof(index)))
1825 			return -EFAULT;
1826 		++indices;
1827 	}
1828 	if (cpu_guest_has_contextconfig) {
1829 		if (copy_to_user(indices, kvm_vz_get_one_regs_contextconfig,
1830 				 sizeof(kvm_vz_get_one_regs_contextconfig)))
1831 			return -EFAULT;
1832 		indices += ARRAY_SIZE(kvm_vz_get_one_regs_contextconfig);
1833 	}
1834 	if (cpu_guest_has_segments) {
1835 		if (copy_to_user(indices, kvm_vz_get_one_regs_segments,
1836 				 sizeof(kvm_vz_get_one_regs_segments)))
1837 			return -EFAULT;
1838 		indices += ARRAY_SIZE(kvm_vz_get_one_regs_segments);
1839 	}
1840 	if (cpu_guest_has_htw || cpu_guest_has_ldpte) {
1841 		if (copy_to_user(indices, kvm_vz_get_one_regs_htw,
1842 				 sizeof(kvm_vz_get_one_regs_htw)))
1843 			return -EFAULT;
1844 		indices += ARRAY_SIZE(kvm_vz_get_one_regs_htw);
1845 	}
1846 	if (cpu_guest_has_maar && !cpu_guest_has_dyn_maar) {
1847 		for (i = 0; i < ARRAY_SIZE(vcpu->arch.maar); ++i) {
1848 			index = KVM_REG_MIPS_CP0_MAAR(i);
1849 			if (copy_to_user(indices, &index, sizeof(index)))
1850 				return -EFAULT;
1851 			++indices;
1852 		}
1853 
1854 		index = KVM_REG_MIPS_CP0_MAARI;
1855 		if (copy_to_user(indices, &index, sizeof(index)))
1856 			return -EFAULT;
1857 		++indices;
1858 	}
1859 	for (i = 0; i < 6; ++i) {
1860 		if (!cpu_guest_has_kscr(i + 2))
1861 			continue;
1862 
1863 		if (copy_to_user(indices, &kvm_vz_get_one_regs_kscratch[i],
1864 				 sizeof(kvm_vz_get_one_regs_kscratch[i])))
1865 			return -EFAULT;
1866 		++indices;
1867 	}
1868 
1869 	return 0;
1870 }
1871 
1872 static inline s64 entrylo_kvm_to_user(unsigned long v)
1873 {
1874 	s64 mask, ret = v;
1875 
1876 	if (BITS_PER_LONG == 32) {
1877 		/*
1878 		 * KVM API exposes 64-bit version of the register, so move the
1879 		 * RI/XI bits up into place.
1880 		 */
1881 		mask = MIPS_ENTRYLO_RI | MIPS_ENTRYLO_XI;
1882 		ret &= ~mask;
1883 		ret |= ((s64)v & mask) << 32;
1884 	}
1885 	return ret;
1886 }
1887 
1888 static inline unsigned long entrylo_user_to_kvm(s64 v)
1889 {
1890 	unsigned long mask, ret = v;
1891 
1892 	if (BITS_PER_LONG == 32) {
1893 		/*
1894 		 * KVM API exposes 64-bit versiono of the register, so move the
1895 		 * RI/XI bits down into place.
1896 		 */
1897 		mask = MIPS_ENTRYLO_RI | MIPS_ENTRYLO_XI;
1898 		ret &= ~mask;
1899 		ret |= (v >> 32) & mask;
1900 	}
1901 	return ret;
1902 }
1903 
1904 static int kvm_vz_get_one_reg(struct kvm_vcpu *vcpu,
1905 			      const struct kvm_one_reg *reg,
1906 			      s64 *v)
1907 {
1908 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1909 	unsigned int idx;
1910 
1911 	switch (reg->id) {
1912 	case KVM_REG_MIPS_CP0_INDEX:
1913 		*v = (long)read_gc0_index();
1914 		break;
1915 	case KVM_REG_MIPS_CP0_ENTRYLO0:
1916 		*v = entrylo_kvm_to_user(read_gc0_entrylo0());
1917 		break;
1918 	case KVM_REG_MIPS_CP0_ENTRYLO1:
1919 		*v = entrylo_kvm_to_user(read_gc0_entrylo1());
1920 		break;
1921 	case KVM_REG_MIPS_CP0_CONTEXT:
1922 		*v = (long)read_gc0_context();
1923 		break;
1924 	case KVM_REG_MIPS_CP0_CONTEXTCONFIG:
1925 		if (!cpu_guest_has_contextconfig)
1926 			return -EINVAL;
1927 		*v = read_gc0_contextconfig();
1928 		break;
1929 	case KVM_REG_MIPS_CP0_USERLOCAL:
1930 		if (!cpu_guest_has_userlocal)
1931 			return -EINVAL;
1932 		*v = read_gc0_userlocal();
1933 		break;
1934 #ifdef CONFIG_64BIT
1935 	case KVM_REG_MIPS_CP0_XCONTEXTCONFIG:
1936 		if (!cpu_guest_has_contextconfig)
1937 			return -EINVAL;
1938 		*v = read_gc0_xcontextconfig();
1939 		break;
1940 #endif
1941 	case KVM_REG_MIPS_CP0_PAGEMASK:
1942 		*v = (long)read_gc0_pagemask();
1943 		break;
1944 	case KVM_REG_MIPS_CP0_PAGEGRAIN:
1945 		*v = (long)read_gc0_pagegrain();
1946 		break;
1947 	case KVM_REG_MIPS_CP0_SEGCTL0:
1948 		if (!cpu_guest_has_segments)
1949 			return -EINVAL;
1950 		*v = read_gc0_segctl0();
1951 		break;
1952 	case KVM_REG_MIPS_CP0_SEGCTL1:
1953 		if (!cpu_guest_has_segments)
1954 			return -EINVAL;
1955 		*v = read_gc0_segctl1();
1956 		break;
1957 	case KVM_REG_MIPS_CP0_SEGCTL2:
1958 		if (!cpu_guest_has_segments)
1959 			return -EINVAL;
1960 		*v = read_gc0_segctl2();
1961 		break;
1962 	case KVM_REG_MIPS_CP0_PWBASE:
1963 		if (!cpu_guest_has_htw && !cpu_guest_has_ldpte)
1964 			return -EINVAL;
1965 		*v = read_gc0_pwbase();
1966 		break;
1967 	case KVM_REG_MIPS_CP0_PWFIELD:
1968 		if (!cpu_guest_has_htw && !cpu_guest_has_ldpte)
1969 			return -EINVAL;
1970 		*v = read_gc0_pwfield();
1971 		break;
1972 	case KVM_REG_MIPS_CP0_PWSIZE:
1973 		if (!cpu_guest_has_htw && !cpu_guest_has_ldpte)
1974 			return -EINVAL;
1975 		*v = read_gc0_pwsize();
1976 		break;
1977 	case KVM_REG_MIPS_CP0_WIRED:
1978 		*v = (long)read_gc0_wired();
1979 		break;
1980 	case KVM_REG_MIPS_CP0_PWCTL:
1981 		if (!cpu_guest_has_htw && !cpu_guest_has_ldpte)
1982 			return -EINVAL;
1983 		*v = read_gc0_pwctl();
1984 		break;
1985 	case KVM_REG_MIPS_CP0_HWRENA:
1986 		*v = (long)read_gc0_hwrena();
1987 		break;
1988 	case KVM_REG_MIPS_CP0_BADVADDR:
1989 		*v = (long)read_gc0_badvaddr();
1990 		break;
1991 	case KVM_REG_MIPS_CP0_BADINSTR:
1992 		if (!cpu_guest_has_badinstr)
1993 			return -EINVAL;
1994 		*v = read_gc0_badinstr();
1995 		break;
1996 	case KVM_REG_MIPS_CP0_BADINSTRP:
1997 		if (!cpu_guest_has_badinstrp)
1998 			return -EINVAL;
1999 		*v = read_gc0_badinstrp();
2000 		break;
2001 	case KVM_REG_MIPS_CP0_COUNT:
2002 		*v = kvm_mips_read_count(vcpu);
2003 		break;
2004 	case KVM_REG_MIPS_CP0_ENTRYHI:
2005 		*v = (long)read_gc0_entryhi();
2006 		break;
2007 	case KVM_REG_MIPS_CP0_COMPARE:
2008 		*v = (long)read_gc0_compare();
2009 		break;
2010 	case KVM_REG_MIPS_CP0_STATUS:
2011 		*v = (long)read_gc0_status();
2012 		break;
2013 	case KVM_REG_MIPS_CP0_INTCTL:
2014 		*v = read_gc0_intctl();
2015 		break;
2016 	case KVM_REG_MIPS_CP0_CAUSE:
2017 		*v = (long)read_gc0_cause();
2018 		break;
2019 	case KVM_REG_MIPS_CP0_EPC:
2020 		*v = (long)read_gc0_epc();
2021 		break;
2022 	case KVM_REG_MIPS_CP0_PRID:
2023 		switch (boot_cpu_type()) {
2024 		case CPU_CAVIUM_OCTEON3:
2025 			/* Octeon III has a read-only guest.PRid */
2026 			*v = read_gc0_prid();
2027 			break;
2028 		default:
2029 			*v = (long)kvm_read_c0_guest_prid(cop0);
2030 			break;
2031 		}
2032 		break;
2033 	case KVM_REG_MIPS_CP0_EBASE:
2034 		*v = kvm_vz_read_gc0_ebase();
2035 		break;
2036 	case KVM_REG_MIPS_CP0_CONFIG:
2037 		*v = read_gc0_config();
2038 		break;
2039 	case KVM_REG_MIPS_CP0_CONFIG1:
2040 		if (!cpu_guest_has_conf1)
2041 			return -EINVAL;
2042 		*v = read_gc0_config1();
2043 		break;
2044 	case KVM_REG_MIPS_CP0_CONFIG2:
2045 		if (!cpu_guest_has_conf2)
2046 			return -EINVAL;
2047 		*v = read_gc0_config2();
2048 		break;
2049 	case KVM_REG_MIPS_CP0_CONFIG3:
2050 		if (!cpu_guest_has_conf3)
2051 			return -EINVAL;
2052 		*v = read_gc0_config3();
2053 		break;
2054 	case KVM_REG_MIPS_CP0_CONFIG4:
2055 		if (!cpu_guest_has_conf4)
2056 			return -EINVAL;
2057 		*v = read_gc0_config4();
2058 		break;
2059 	case KVM_REG_MIPS_CP0_CONFIG5:
2060 		if (!cpu_guest_has_conf5)
2061 			return -EINVAL;
2062 		*v = read_gc0_config5();
2063 		break;
2064 	case KVM_REG_MIPS_CP0_CONFIG6:
2065 		*v = kvm_read_sw_gc0_config6(cop0);
2066 		break;
2067 	case KVM_REG_MIPS_CP0_MAAR(0) ... KVM_REG_MIPS_CP0_MAAR(0x3f):
2068 		if (!cpu_guest_has_maar || cpu_guest_has_dyn_maar)
2069 			return -EINVAL;
2070 		idx = reg->id - KVM_REG_MIPS_CP0_MAAR(0);
2071 		if (idx >= ARRAY_SIZE(vcpu->arch.maar))
2072 			return -EINVAL;
2073 		*v = vcpu->arch.maar[idx];
2074 		break;
2075 	case KVM_REG_MIPS_CP0_MAARI:
2076 		if (!cpu_guest_has_maar || cpu_guest_has_dyn_maar)
2077 			return -EINVAL;
2078 		*v = kvm_read_sw_gc0_maari(vcpu->arch.cop0);
2079 		break;
2080 #ifdef CONFIG_64BIT
2081 	case KVM_REG_MIPS_CP0_XCONTEXT:
2082 		*v = read_gc0_xcontext();
2083 		break;
2084 #endif
2085 	case KVM_REG_MIPS_CP0_ERROREPC:
2086 		*v = (long)read_gc0_errorepc();
2087 		break;
2088 	case KVM_REG_MIPS_CP0_KSCRATCH1 ... KVM_REG_MIPS_CP0_KSCRATCH6:
2089 		idx = reg->id - KVM_REG_MIPS_CP0_KSCRATCH1 + 2;
2090 		if (!cpu_guest_has_kscr(idx))
2091 			return -EINVAL;
2092 		switch (idx) {
2093 		case 2:
2094 			*v = (long)read_gc0_kscratch1();
2095 			break;
2096 		case 3:
2097 			*v = (long)read_gc0_kscratch2();
2098 			break;
2099 		case 4:
2100 			*v = (long)read_gc0_kscratch3();
2101 			break;
2102 		case 5:
2103 			*v = (long)read_gc0_kscratch4();
2104 			break;
2105 		case 6:
2106 			*v = (long)read_gc0_kscratch5();
2107 			break;
2108 		case 7:
2109 			*v = (long)read_gc0_kscratch6();
2110 			break;
2111 		}
2112 		break;
2113 	case KVM_REG_MIPS_COUNT_CTL:
2114 		*v = vcpu->arch.count_ctl;
2115 		break;
2116 	case KVM_REG_MIPS_COUNT_RESUME:
2117 		*v = ktime_to_ns(vcpu->arch.count_resume);
2118 		break;
2119 	case KVM_REG_MIPS_COUNT_HZ:
2120 		*v = vcpu->arch.count_hz;
2121 		break;
2122 	default:
2123 		return -EINVAL;
2124 	}
2125 	return 0;
2126 }
2127 
2128 static int kvm_vz_set_one_reg(struct kvm_vcpu *vcpu,
2129 			      const struct kvm_one_reg *reg,
2130 			      s64 v)
2131 {
2132 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2133 	unsigned int idx;
2134 	int ret = 0;
2135 	unsigned int cur, change;
2136 
2137 	switch (reg->id) {
2138 	case KVM_REG_MIPS_CP0_INDEX:
2139 		write_gc0_index(v);
2140 		break;
2141 	case KVM_REG_MIPS_CP0_ENTRYLO0:
2142 		write_gc0_entrylo0(entrylo_user_to_kvm(v));
2143 		break;
2144 	case KVM_REG_MIPS_CP0_ENTRYLO1:
2145 		write_gc0_entrylo1(entrylo_user_to_kvm(v));
2146 		break;
2147 	case KVM_REG_MIPS_CP0_CONTEXT:
2148 		write_gc0_context(v);
2149 		break;
2150 	case KVM_REG_MIPS_CP0_CONTEXTCONFIG:
2151 		if (!cpu_guest_has_contextconfig)
2152 			return -EINVAL;
2153 		write_gc0_contextconfig(v);
2154 		break;
2155 	case KVM_REG_MIPS_CP0_USERLOCAL:
2156 		if (!cpu_guest_has_userlocal)
2157 			return -EINVAL;
2158 		write_gc0_userlocal(v);
2159 		break;
2160 #ifdef CONFIG_64BIT
2161 	case KVM_REG_MIPS_CP0_XCONTEXTCONFIG:
2162 		if (!cpu_guest_has_contextconfig)
2163 			return -EINVAL;
2164 		write_gc0_xcontextconfig(v);
2165 		break;
2166 #endif
2167 	case KVM_REG_MIPS_CP0_PAGEMASK:
2168 		write_gc0_pagemask(v);
2169 		break;
2170 	case KVM_REG_MIPS_CP0_PAGEGRAIN:
2171 		write_gc0_pagegrain(v);
2172 		break;
2173 	case KVM_REG_MIPS_CP0_SEGCTL0:
2174 		if (!cpu_guest_has_segments)
2175 			return -EINVAL;
2176 		write_gc0_segctl0(v);
2177 		break;
2178 	case KVM_REG_MIPS_CP0_SEGCTL1:
2179 		if (!cpu_guest_has_segments)
2180 			return -EINVAL;
2181 		write_gc0_segctl1(v);
2182 		break;
2183 	case KVM_REG_MIPS_CP0_SEGCTL2:
2184 		if (!cpu_guest_has_segments)
2185 			return -EINVAL;
2186 		write_gc0_segctl2(v);
2187 		break;
2188 	case KVM_REG_MIPS_CP0_PWBASE:
2189 		if (!cpu_guest_has_htw && !cpu_guest_has_ldpte)
2190 			return -EINVAL;
2191 		write_gc0_pwbase(v);
2192 		break;
2193 	case KVM_REG_MIPS_CP0_PWFIELD:
2194 		if (!cpu_guest_has_htw && !cpu_guest_has_ldpte)
2195 			return -EINVAL;
2196 		write_gc0_pwfield(v);
2197 		break;
2198 	case KVM_REG_MIPS_CP0_PWSIZE:
2199 		if (!cpu_guest_has_htw && !cpu_guest_has_ldpte)
2200 			return -EINVAL;
2201 		write_gc0_pwsize(v);
2202 		break;
2203 	case KVM_REG_MIPS_CP0_WIRED:
2204 		change_gc0_wired(MIPSR6_WIRED_WIRED, v);
2205 		break;
2206 	case KVM_REG_MIPS_CP0_PWCTL:
2207 		if (!cpu_guest_has_htw && !cpu_guest_has_ldpte)
2208 			return -EINVAL;
2209 		write_gc0_pwctl(v);
2210 		break;
2211 	case KVM_REG_MIPS_CP0_HWRENA:
2212 		write_gc0_hwrena(v);
2213 		break;
2214 	case KVM_REG_MIPS_CP0_BADVADDR:
2215 		write_gc0_badvaddr(v);
2216 		break;
2217 	case KVM_REG_MIPS_CP0_BADINSTR:
2218 		if (!cpu_guest_has_badinstr)
2219 			return -EINVAL;
2220 		write_gc0_badinstr(v);
2221 		break;
2222 	case KVM_REG_MIPS_CP0_BADINSTRP:
2223 		if (!cpu_guest_has_badinstrp)
2224 			return -EINVAL;
2225 		write_gc0_badinstrp(v);
2226 		break;
2227 	case KVM_REG_MIPS_CP0_COUNT:
2228 		kvm_mips_write_count(vcpu, v);
2229 		break;
2230 	case KVM_REG_MIPS_CP0_ENTRYHI:
2231 		write_gc0_entryhi(v);
2232 		break;
2233 	case KVM_REG_MIPS_CP0_COMPARE:
2234 		kvm_mips_write_compare(vcpu, v, false);
2235 		break;
2236 	case KVM_REG_MIPS_CP0_STATUS:
2237 		write_gc0_status(v);
2238 		break;
2239 	case KVM_REG_MIPS_CP0_INTCTL:
2240 		write_gc0_intctl(v);
2241 		break;
2242 	case KVM_REG_MIPS_CP0_CAUSE:
2243 		/*
2244 		 * If the timer is stopped or started (DC bit) it must look
2245 		 * atomic with changes to the timer interrupt pending bit (TI).
2246 		 * A timer interrupt should not happen in between.
2247 		 */
2248 		if ((read_gc0_cause() ^ v) & CAUSEF_DC) {
2249 			if (v & CAUSEF_DC) {
2250 				/* disable timer first */
2251 				kvm_mips_count_disable_cause(vcpu);
2252 				change_gc0_cause((u32)~CAUSEF_DC, v);
2253 			} else {
2254 				/* enable timer last */
2255 				change_gc0_cause((u32)~CAUSEF_DC, v);
2256 				kvm_mips_count_enable_cause(vcpu);
2257 			}
2258 		} else {
2259 			write_gc0_cause(v);
2260 		}
2261 		break;
2262 	case KVM_REG_MIPS_CP0_EPC:
2263 		write_gc0_epc(v);
2264 		break;
2265 	case KVM_REG_MIPS_CP0_PRID:
2266 		switch (boot_cpu_type()) {
2267 		case CPU_CAVIUM_OCTEON3:
2268 			/* Octeon III has a guest.PRid, but its read-only */
2269 			break;
2270 		default:
2271 			kvm_write_c0_guest_prid(cop0, v);
2272 			break;
2273 		}
2274 		break;
2275 	case KVM_REG_MIPS_CP0_EBASE:
2276 		kvm_vz_write_gc0_ebase(v);
2277 		break;
2278 	case KVM_REG_MIPS_CP0_CONFIG:
2279 		cur = read_gc0_config();
2280 		change = (cur ^ v) & kvm_vz_config_user_wrmask(vcpu);
2281 		if (change) {
2282 			v = cur ^ change;
2283 			write_gc0_config(v);
2284 		}
2285 		break;
2286 	case KVM_REG_MIPS_CP0_CONFIG1:
2287 		if (!cpu_guest_has_conf1)
2288 			break;
2289 		cur = read_gc0_config1();
2290 		change = (cur ^ v) & kvm_vz_config1_user_wrmask(vcpu);
2291 		if (change) {
2292 			v = cur ^ change;
2293 			write_gc0_config1(v);
2294 		}
2295 		break;
2296 	case KVM_REG_MIPS_CP0_CONFIG2:
2297 		if (!cpu_guest_has_conf2)
2298 			break;
2299 		cur = read_gc0_config2();
2300 		change = (cur ^ v) & kvm_vz_config2_user_wrmask(vcpu);
2301 		if (change) {
2302 			v = cur ^ change;
2303 			write_gc0_config2(v);
2304 		}
2305 		break;
2306 	case KVM_REG_MIPS_CP0_CONFIG3:
2307 		if (!cpu_guest_has_conf3)
2308 			break;
2309 		cur = read_gc0_config3();
2310 		change = (cur ^ v) & kvm_vz_config3_user_wrmask(vcpu);
2311 		if (change) {
2312 			v = cur ^ change;
2313 			write_gc0_config3(v);
2314 		}
2315 		break;
2316 	case KVM_REG_MIPS_CP0_CONFIG4:
2317 		if (!cpu_guest_has_conf4)
2318 			break;
2319 		cur = read_gc0_config4();
2320 		change = (cur ^ v) & kvm_vz_config4_user_wrmask(vcpu);
2321 		if (change) {
2322 			v = cur ^ change;
2323 			write_gc0_config4(v);
2324 		}
2325 		break;
2326 	case KVM_REG_MIPS_CP0_CONFIG5:
2327 		if (!cpu_guest_has_conf5)
2328 			break;
2329 		cur = read_gc0_config5();
2330 		change = (cur ^ v) & kvm_vz_config5_user_wrmask(vcpu);
2331 		if (change) {
2332 			v = cur ^ change;
2333 			write_gc0_config5(v);
2334 		}
2335 		break;
2336 	case KVM_REG_MIPS_CP0_CONFIG6:
2337 		cur = kvm_read_sw_gc0_config6(cop0);
2338 		change = (cur ^ v) & kvm_vz_config6_user_wrmask(vcpu);
2339 		if (change) {
2340 			v = cur ^ change;
2341 			kvm_write_sw_gc0_config6(cop0, (int)v);
2342 		}
2343 		break;
2344 	case KVM_REG_MIPS_CP0_MAAR(0) ... KVM_REG_MIPS_CP0_MAAR(0x3f):
2345 		if (!cpu_guest_has_maar || cpu_guest_has_dyn_maar)
2346 			return -EINVAL;
2347 		idx = reg->id - KVM_REG_MIPS_CP0_MAAR(0);
2348 		if (idx >= ARRAY_SIZE(vcpu->arch.maar))
2349 			return -EINVAL;
2350 		vcpu->arch.maar[idx] = mips_process_maar(dmtc_op, v);
2351 		break;
2352 	case KVM_REG_MIPS_CP0_MAARI:
2353 		if (!cpu_guest_has_maar || cpu_guest_has_dyn_maar)
2354 			return -EINVAL;
2355 		kvm_write_maari(vcpu, v);
2356 		break;
2357 #ifdef CONFIG_64BIT
2358 	case KVM_REG_MIPS_CP0_XCONTEXT:
2359 		write_gc0_xcontext(v);
2360 		break;
2361 #endif
2362 	case KVM_REG_MIPS_CP0_ERROREPC:
2363 		write_gc0_errorepc(v);
2364 		break;
2365 	case KVM_REG_MIPS_CP0_KSCRATCH1 ... KVM_REG_MIPS_CP0_KSCRATCH6:
2366 		idx = reg->id - KVM_REG_MIPS_CP0_KSCRATCH1 + 2;
2367 		if (!cpu_guest_has_kscr(idx))
2368 			return -EINVAL;
2369 		switch (idx) {
2370 		case 2:
2371 			write_gc0_kscratch1(v);
2372 			break;
2373 		case 3:
2374 			write_gc0_kscratch2(v);
2375 			break;
2376 		case 4:
2377 			write_gc0_kscratch3(v);
2378 			break;
2379 		case 5:
2380 			write_gc0_kscratch4(v);
2381 			break;
2382 		case 6:
2383 			write_gc0_kscratch5(v);
2384 			break;
2385 		case 7:
2386 			write_gc0_kscratch6(v);
2387 			break;
2388 		}
2389 		break;
2390 	case KVM_REG_MIPS_COUNT_CTL:
2391 		ret = kvm_mips_set_count_ctl(vcpu, v);
2392 		break;
2393 	case KVM_REG_MIPS_COUNT_RESUME:
2394 		ret = kvm_mips_set_count_resume(vcpu, v);
2395 		break;
2396 	case KVM_REG_MIPS_COUNT_HZ:
2397 		ret = kvm_mips_set_count_hz(vcpu, v);
2398 		break;
2399 	default:
2400 		return -EINVAL;
2401 	}
2402 	return ret;
2403 }
2404 
2405 #define guestid_cache(cpu)	(cpu_data[cpu].guestid_cache)
2406 static void kvm_vz_get_new_guestid(unsigned long cpu, struct kvm_vcpu *vcpu)
2407 {
2408 	unsigned long guestid = guestid_cache(cpu);
2409 
2410 	if (!(++guestid & GUESTID_MASK)) {
2411 		if (cpu_has_vtag_icache)
2412 			flush_icache_all();
2413 
2414 		if (!guestid)		/* fix version if needed */
2415 			guestid = GUESTID_FIRST_VERSION;
2416 
2417 		++guestid;		/* guestid 0 reserved for root */
2418 
2419 		/* start new guestid cycle */
2420 		kvm_vz_local_flush_roottlb_all_guests();
2421 		kvm_vz_local_flush_guesttlb_all();
2422 	}
2423 
2424 	guestid_cache(cpu) = guestid;
2425 }
2426 
2427 /* Returns 1 if the guest TLB may be clobbered */
2428 static int kvm_vz_check_requests(struct kvm_vcpu *vcpu, int cpu)
2429 {
2430 	int ret = 0;
2431 	int i;
2432 
2433 	if (!kvm_request_pending(vcpu))
2434 		return 0;
2435 
2436 	if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
2437 		if (cpu_has_guestid) {
2438 			/* Drop all GuestIDs for this VCPU */
2439 			for_each_possible_cpu(i)
2440 				vcpu->arch.vzguestid[i] = 0;
2441 			/* This will clobber guest TLB contents too */
2442 			ret = 1;
2443 		}
2444 		/*
2445 		 * For Root ASID Dealias (RAD) we don't do anything here, but we
2446 		 * still need the request to ensure we recheck asid_flush_mask.
2447 		 * We can still return 0 as only the root TLB will be affected
2448 		 * by a root ASID flush.
2449 		 */
2450 	}
2451 
2452 	return ret;
2453 }
2454 
2455 static void kvm_vz_vcpu_save_wired(struct kvm_vcpu *vcpu)
2456 {
2457 	unsigned int wired = read_gc0_wired();
2458 	struct kvm_mips_tlb *tlbs;
2459 	int i;
2460 
2461 	/* Expand the wired TLB array if necessary */
2462 	wired &= MIPSR6_WIRED_WIRED;
2463 	if (wired > vcpu->arch.wired_tlb_limit) {
2464 		tlbs = krealloc(vcpu->arch.wired_tlb, wired *
2465 				sizeof(*vcpu->arch.wired_tlb), GFP_ATOMIC);
2466 		if (WARN_ON(!tlbs)) {
2467 			/* Save whatever we can */
2468 			wired = vcpu->arch.wired_tlb_limit;
2469 		} else {
2470 			vcpu->arch.wired_tlb = tlbs;
2471 			vcpu->arch.wired_tlb_limit = wired;
2472 		}
2473 	}
2474 
2475 	if (wired)
2476 		/* Save wired entries from the guest TLB */
2477 		kvm_vz_save_guesttlb(vcpu->arch.wired_tlb, 0, wired);
2478 	/* Invalidate any dropped entries since last time */
2479 	for (i = wired; i < vcpu->arch.wired_tlb_used; ++i) {
2480 		vcpu->arch.wired_tlb[i].tlb_hi = UNIQUE_GUEST_ENTRYHI(i);
2481 		vcpu->arch.wired_tlb[i].tlb_lo[0] = 0;
2482 		vcpu->arch.wired_tlb[i].tlb_lo[1] = 0;
2483 		vcpu->arch.wired_tlb[i].tlb_mask = 0;
2484 	}
2485 	vcpu->arch.wired_tlb_used = wired;
2486 }
2487 
2488 static void kvm_vz_vcpu_load_wired(struct kvm_vcpu *vcpu)
2489 {
2490 	/* Load wired entries into the guest TLB */
2491 	if (vcpu->arch.wired_tlb)
2492 		kvm_vz_load_guesttlb(vcpu->arch.wired_tlb, 0,
2493 				     vcpu->arch.wired_tlb_used);
2494 }
2495 
2496 static void kvm_vz_vcpu_load_tlb(struct kvm_vcpu *vcpu, int cpu)
2497 {
2498 	struct kvm *kvm = vcpu->kvm;
2499 	struct mm_struct *gpa_mm = &kvm->arch.gpa_mm;
2500 	bool migrated;
2501 
2502 	/*
2503 	 * Are we entering guest context on a different CPU to last time?
2504 	 * If so, the VCPU's guest TLB state on this CPU may be stale.
2505 	 */
2506 	migrated = (vcpu->arch.last_exec_cpu != cpu);
2507 	vcpu->arch.last_exec_cpu = cpu;
2508 
2509 	/*
2510 	 * A vcpu's GuestID is set in GuestCtl1.ID when the vcpu is loaded and
2511 	 * remains set until another vcpu is loaded in.  As a rule GuestRID
2512 	 * remains zeroed when in root context unless the kernel is busy
2513 	 * manipulating guest tlb entries.
2514 	 */
2515 	if (cpu_has_guestid) {
2516 		/*
2517 		 * Check if our GuestID is of an older version and thus invalid.
2518 		 *
2519 		 * We also discard the stored GuestID if we've executed on
2520 		 * another CPU, as the guest mappings may have changed without
2521 		 * hypervisor knowledge.
2522 		 */
2523 		if (migrated ||
2524 		    (vcpu->arch.vzguestid[cpu] ^ guestid_cache(cpu)) &
2525 					GUESTID_VERSION_MASK) {
2526 			kvm_vz_get_new_guestid(cpu, vcpu);
2527 			vcpu->arch.vzguestid[cpu] = guestid_cache(cpu);
2528 			trace_kvm_guestid_change(vcpu,
2529 						 vcpu->arch.vzguestid[cpu]);
2530 		}
2531 
2532 		/* Restore GuestID */
2533 		change_c0_guestctl1(GUESTID_MASK, vcpu->arch.vzguestid[cpu]);
2534 	} else {
2535 		/*
2536 		 * The Guest TLB only stores a single guest's TLB state, so
2537 		 * flush it if another VCPU has executed on this CPU.
2538 		 *
2539 		 * We also flush if we've executed on another CPU, as the guest
2540 		 * mappings may have changed without hypervisor knowledge.
2541 		 */
2542 		if (migrated || last_exec_vcpu[cpu] != vcpu)
2543 			kvm_vz_local_flush_guesttlb_all();
2544 		last_exec_vcpu[cpu] = vcpu;
2545 
2546 		/*
2547 		 * Root ASID dealiases guest GPA mappings in the root TLB.
2548 		 * Allocate new root ASID if needed.
2549 		 */
2550 		if (cpumask_test_and_clear_cpu(cpu, &kvm->arch.asid_flush_mask))
2551 			get_new_mmu_context(gpa_mm);
2552 		else
2553 			check_mmu_context(gpa_mm);
2554 	}
2555 }
2556 
2557 static int kvm_vz_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2558 {
2559 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2560 	bool migrated, all;
2561 
2562 	/*
2563 	 * Have we migrated to a different CPU?
2564 	 * If so, any old guest TLB state may be stale.
2565 	 */
2566 	migrated = (vcpu->arch.last_sched_cpu != cpu);
2567 
2568 	/*
2569 	 * Was this the last VCPU to run on this CPU?
2570 	 * If not, any old guest state from this VCPU will have been clobbered.
2571 	 */
2572 	all = migrated || (last_vcpu[cpu] != vcpu);
2573 	last_vcpu[cpu] = vcpu;
2574 
2575 	/*
2576 	 * Restore CP0_Wired unconditionally as we clear it after use, and
2577 	 * restore wired guest TLB entries (while in guest context).
2578 	 */
2579 	kvm_restore_gc0_wired(cop0);
2580 	if (current->flags & PF_VCPU) {
2581 		tlbw_use_hazard();
2582 		kvm_vz_vcpu_load_tlb(vcpu, cpu);
2583 		kvm_vz_vcpu_load_wired(vcpu);
2584 	}
2585 
2586 	/*
2587 	 * Restore timer state regardless, as e.g. Cause.TI can change over time
2588 	 * if left unmaintained.
2589 	 */
2590 	kvm_vz_restore_timer(vcpu);
2591 
2592 	/* Set MC bit if we want to trace guest mode changes */
2593 	if (kvm_trace_guest_mode_change)
2594 		set_c0_guestctl0(MIPS_GCTL0_MC);
2595 	else
2596 		clear_c0_guestctl0(MIPS_GCTL0_MC);
2597 
2598 	/* Don't bother restoring registers multiple times unless necessary */
2599 	if (!all)
2600 		return 0;
2601 
2602 	/*
2603 	 * Restore config registers first, as some implementations restrict
2604 	 * writes to other registers when the corresponding feature bits aren't
2605 	 * set. For example Status.CU1 cannot be set unless Config1.FP is set.
2606 	 */
2607 	kvm_restore_gc0_config(cop0);
2608 	if (cpu_guest_has_conf1)
2609 		kvm_restore_gc0_config1(cop0);
2610 	if (cpu_guest_has_conf2)
2611 		kvm_restore_gc0_config2(cop0);
2612 	if (cpu_guest_has_conf3)
2613 		kvm_restore_gc0_config3(cop0);
2614 	if (cpu_guest_has_conf4)
2615 		kvm_restore_gc0_config4(cop0);
2616 	if (cpu_guest_has_conf5)
2617 		kvm_restore_gc0_config5(cop0);
2618 	if (cpu_guest_has_conf6)
2619 		kvm_restore_gc0_config6(cop0);
2620 	if (cpu_guest_has_conf7)
2621 		kvm_restore_gc0_config7(cop0);
2622 
2623 	kvm_restore_gc0_index(cop0);
2624 	kvm_restore_gc0_entrylo0(cop0);
2625 	kvm_restore_gc0_entrylo1(cop0);
2626 	kvm_restore_gc0_context(cop0);
2627 	if (cpu_guest_has_contextconfig)
2628 		kvm_restore_gc0_contextconfig(cop0);
2629 #ifdef CONFIG_64BIT
2630 	kvm_restore_gc0_xcontext(cop0);
2631 	if (cpu_guest_has_contextconfig)
2632 		kvm_restore_gc0_xcontextconfig(cop0);
2633 #endif
2634 	kvm_restore_gc0_pagemask(cop0);
2635 	kvm_restore_gc0_pagegrain(cop0);
2636 	kvm_restore_gc0_hwrena(cop0);
2637 	kvm_restore_gc0_badvaddr(cop0);
2638 	kvm_restore_gc0_entryhi(cop0);
2639 	kvm_restore_gc0_status(cop0);
2640 	kvm_restore_gc0_intctl(cop0);
2641 	kvm_restore_gc0_epc(cop0);
2642 	kvm_vz_write_gc0_ebase(kvm_read_sw_gc0_ebase(cop0));
2643 	if (cpu_guest_has_userlocal)
2644 		kvm_restore_gc0_userlocal(cop0);
2645 
2646 	kvm_restore_gc0_errorepc(cop0);
2647 
2648 	/* restore KScratch registers if enabled in guest */
2649 	if (cpu_guest_has_conf4) {
2650 		if (cpu_guest_has_kscr(2))
2651 			kvm_restore_gc0_kscratch1(cop0);
2652 		if (cpu_guest_has_kscr(3))
2653 			kvm_restore_gc0_kscratch2(cop0);
2654 		if (cpu_guest_has_kscr(4))
2655 			kvm_restore_gc0_kscratch3(cop0);
2656 		if (cpu_guest_has_kscr(5))
2657 			kvm_restore_gc0_kscratch4(cop0);
2658 		if (cpu_guest_has_kscr(6))
2659 			kvm_restore_gc0_kscratch5(cop0);
2660 		if (cpu_guest_has_kscr(7))
2661 			kvm_restore_gc0_kscratch6(cop0);
2662 	}
2663 
2664 	if (cpu_guest_has_badinstr)
2665 		kvm_restore_gc0_badinstr(cop0);
2666 	if (cpu_guest_has_badinstrp)
2667 		kvm_restore_gc0_badinstrp(cop0);
2668 
2669 	if (cpu_guest_has_segments) {
2670 		kvm_restore_gc0_segctl0(cop0);
2671 		kvm_restore_gc0_segctl1(cop0);
2672 		kvm_restore_gc0_segctl2(cop0);
2673 	}
2674 
2675 	/* restore HTW registers */
2676 	if (cpu_guest_has_htw || cpu_guest_has_ldpte) {
2677 		kvm_restore_gc0_pwbase(cop0);
2678 		kvm_restore_gc0_pwfield(cop0);
2679 		kvm_restore_gc0_pwsize(cop0);
2680 		kvm_restore_gc0_pwctl(cop0);
2681 	}
2682 
2683 	/* restore Root.GuestCtl2 from unused Guest guestctl2 register */
2684 	if (cpu_has_guestctl2)
2685 		write_c0_guestctl2(
2686 			cop0->reg[MIPS_CP0_GUESTCTL2][MIPS_CP0_GUESTCTL2_SEL]);
2687 
2688 	/*
2689 	 * We should clear linked load bit to break interrupted atomics. This
2690 	 * prevents a SC on the next VCPU from succeeding by matching a LL on
2691 	 * the previous VCPU.
2692 	 */
2693 	if (vcpu->kvm->created_vcpus > 1)
2694 		write_gc0_lladdr(0);
2695 
2696 	return 0;
2697 }
2698 
2699 static int kvm_vz_vcpu_put(struct kvm_vcpu *vcpu, int cpu)
2700 {
2701 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2702 
2703 	if (current->flags & PF_VCPU)
2704 		kvm_vz_vcpu_save_wired(vcpu);
2705 
2706 	kvm_lose_fpu(vcpu);
2707 
2708 	kvm_save_gc0_index(cop0);
2709 	kvm_save_gc0_entrylo0(cop0);
2710 	kvm_save_gc0_entrylo1(cop0);
2711 	kvm_save_gc0_context(cop0);
2712 	if (cpu_guest_has_contextconfig)
2713 		kvm_save_gc0_contextconfig(cop0);
2714 #ifdef CONFIG_64BIT
2715 	kvm_save_gc0_xcontext(cop0);
2716 	if (cpu_guest_has_contextconfig)
2717 		kvm_save_gc0_xcontextconfig(cop0);
2718 #endif
2719 	kvm_save_gc0_pagemask(cop0);
2720 	kvm_save_gc0_pagegrain(cop0);
2721 	kvm_save_gc0_wired(cop0);
2722 	/* allow wired TLB entries to be overwritten */
2723 	clear_gc0_wired(MIPSR6_WIRED_WIRED);
2724 	kvm_save_gc0_hwrena(cop0);
2725 	kvm_save_gc0_badvaddr(cop0);
2726 	kvm_save_gc0_entryhi(cop0);
2727 	kvm_save_gc0_status(cop0);
2728 	kvm_save_gc0_intctl(cop0);
2729 	kvm_save_gc0_epc(cop0);
2730 	kvm_write_sw_gc0_ebase(cop0, kvm_vz_read_gc0_ebase());
2731 	if (cpu_guest_has_userlocal)
2732 		kvm_save_gc0_userlocal(cop0);
2733 
2734 	/* only save implemented config registers */
2735 	kvm_save_gc0_config(cop0);
2736 	if (cpu_guest_has_conf1)
2737 		kvm_save_gc0_config1(cop0);
2738 	if (cpu_guest_has_conf2)
2739 		kvm_save_gc0_config2(cop0);
2740 	if (cpu_guest_has_conf3)
2741 		kvm_save_gc0_config3(cop0);
2742 	if (cpu_guest_has_conf4)
2743 		kvm_save_gc0_config4(cop0);
2744 	if (cpu_guest_has_conf5)
2745 		kvm_save_gc0_config5(cop0);
2746 	if (cpu_guest_has_conf6)
2747 		kvm_save_gc0_config6(cop0);
2748 	if (cpu_guest_has_conf7)
2749 		kvm_save_gc0_config7(cop0);
2750 
2751 	kvm_save_gc0_errorepc(cop0);
2752 
2753 	/* save KScratch registers if enabled in guest */
2754 	if (cpu_guest_has_conf4) {
2755 		if (cpu_guest_has_kscr(2))
2756 			kvm_save_gc0_kscratch1(cop0);
2757 		if (cpu_guest_has_kscr(3))
2758 			kvm_save_gc0_kscratch2(cop0);
2759 		if (cpu_guest_has_kscr(4))
2760 			kvm_save_gc0_kscratch3(cop0);
2761 		if (cpu_guest_has_kscr(5))
2762 			kvm_save_gc0_kscratch4(cop0);
2763 		if (cpu_guest_has_kscr(6))
2764 			kvm_save_gc0_kscratch5(cop0);
2765 		if (cpu_guest_has_kscr(7))
2766 			kvm_save_gc0_kscratch6(cop0);
2767 	}
2768 
2769 	if (cpu_guest_has_badinstr)
2770 		kvm_save_gc0_badinstr(cop0);
2771 	if (cpu_guest_has_badinstrp)
2772 		kvm_save_gc0_badinstrp(cop0);
2773 
2774 	if (cpu_guest_has_segments) {
2775 		kvm_save_gc0_segctl0(cop0);
2776 		kvm_save_gc0_segctl1(cop0);
2777 		kvm_save_gc0_segctl2(cop0);
2778 	}
2779 
2780 	/* save HTW registers if enabled in guest */
2781 	if (cpu_guest_has_ldpte || (cpu_guest_has_htw &&
2782 	    kvm_read_sw_gc0_config3(cop0) & MIPS_CONF3_PW)) {
2783 		kvm_save_gc0_pwbase(cop0);
2784 		kvm_save_gc0_pwfield(cop0);
2785 		kvm_save_gc0_pwsize(cop0);
2786 		kvm_save_gc0_pwctl(cop0);
2787 	}
2788 
2789 	kvm_vz_save_timer(vcpu);
2790 
2791 	/* save Root.GuestCtl2 in unused Guest guestctl2 register */
2792 	if (cpu_has_guestctl2)
2793 		cop0->reg[MIPS_CP0_GUESTCTL2][MIPS_CP0_GUESTCTL2_SEL] =
2794 			read_c0_guestctl2();
2795 
2796 	return 0;
2797 }
2798 
2799 /**
2800  * kvm_vz_resize_guest_vtlb() - Attempt to resize guest VTLB.
2801  * @size:	Number of guest VTLB entries (0 < @size <= root VTLB entries).
2802  *
2803  * Attempt to resize the guest VTLB by writing guest Config registers. This is
2804  * necessary for cores with a shared root/guest TLB to avoid overlap with wired
2805  * entries in the root VTLB.
2806  *
2807  * Returns:	The resulting guest VTLB size.
2808  */
2809 static unsigned int kvm_vz_resize_guest_vtlb(unsigned int size)
2810 {
2811 	unsigned int config4 = 0, ret = 0, limit;
2812 
2813 	/* Write MMUSize - 1 into guest Config registers */
2814 	if (cpu_guest_has_conf1)
2815 		change_gc0_config1(MIPS_CONF1_TLBS,
2816 				   (size - 1) << MIPS_CONF1_TLBS_SHIFT);
2817 	if (cpu_guest_has_conf4) {
2818 		config4 = read_gc0_config4();
2819 		if (cpu_has_mips_r6 || (config4 & MIPS_CONF4_MMUEXTDEF) ==
2820 		    MIPS_CONF4_MMUEXTDEF_VTLBSIZEEXT) {
2821 			config4 &= ~MIPS_CONF4_VTLBSIZEEXT;
2822 			config4 |= ((size - 1) >> MIPS_CONF1_TLBS_SIZE) <<
2823 				MIPS_CONF4_VTLBSIZEEXT_SHIFT;
2824 		} else if ((config4 & MIPS_CONF4_MMUEXTDEF) ==
2825 			   MIPS_CONF4_MMUEXTDEF_MMUSIZEEXT) {
2826 			config4 &= ~MIPS_CONF4_MMUSIZEEXT;
2827 			config4 |= ((size - 1) >> MIPS_CONF1_TLBS_SIZE) <<
2828 				MIPS_CONF4_MMUSIZEEXT_SHIFT;
2829 		}
2830 		write_gc0_config4(config4);
2831 	}
2832 
2833 	/*
2834 	 * Set Guest.Wired.Limit = 0 (no limit up to Guest.MMUSize-1), unless it
2835 	 * would exceed Root.Wired.Limit (clearing Guest.Wired.Wired so write
2836 	 * not dropped)
2837 	 */
2838 	if (cpu_has_mips_r6) {
2839 		limit = (read_c0_wired() & MIPSR6_WIRED_LIMIT) >>
2840 						MIPSR6_WIRED_LIMIT_SHIFT;
2841 		if (size - 1 <= limit)
2842 			limit = 0;
2843 		write_gc0_wired(limit << MIPSR6_WIRED_LIMIT_SHIFT);
2844 	}
2845 
2846 	/* Read back MMUSize - 1 */
2847 	back_to_back_c0_hazard();
2848 	if (cpu_guest_has_conf1)
2849 		ret = (read_gc0_config1() & MIPS_CONF1_TLBS) >>
2850 						MIPS_CONF1_TLBS_SHIFT;
2851 	if (config4) {
2852 		if (cpu_has_mips_r6 || (config4 & MIPS_CONF4_MMUEXTDEF) ==
2853 		    MIPS_CONF4_MMUEXTDEF_VTLBSIZEEXT)
2854 			ret |= ((config4 & MIPS_CONF4_VTLBSIZEEXT) >>
2855 				MIPS_CONF4_VTLBSIZEEXT_SHIFT) <<
2856 				MIPS_CONF1_TLBS_SIZE;
2857 		else if ((config4 & MIPS_CONF4_MMUEXTDEF) ==
2858 			 MIPS_CONF4_MMUEXTDEF_MMUSIZEEXT)
2859 			ret |= ((config4 & MIPS_CONF4_MMUSIZEEXT) >>
2860 				MIPS_CONF4_MMUSIZEEXT_SHIFT) <<
2861 				MIPS_CONF1_TLBS_SIZE;
2862 	}
2863 	return ret + 1;
2864 }
2865 
2866 static int kvm_vz_hardware_enable(void)
2867 {
2868 	unsigned int mmu_size, guest_mmu_size, ftlb_size;
2869 	u64 guest_cvmctl, cvmvmconfig;
2870 
2871 	switch (current_cpu_type()) {
2872 	case CPU_CAVIUM_OCTEON3:
2873 		/* Set up guest timer/perfcount IRQ lines */
2874 		guest_cvmctl = read_gc0_cvmctl();
2875 		guest_cvmctl &= ~CVMCTL_IPTI;
2876 		guest_cvmctl |= 7ull << CVMCTL_IPTI_SHIFT;
2877 		guest_cvmctl &= ~CVMCTL_IPPCI;
2878 		guest_cvmctl |= 6ull << CVMCTL_IPPCI_SHIFT;
2879 		write_gc0_cvmctl(guest_cvmctl);
2880 
2881 		cvmvmconfig = read_c0_cvmvmconfig();
2882 		/* No I/O hole translation. */
2883 		cvmvmconfig |= CVMVMCONF_DGHT;
2884 		/* Halve the root MMU size */
2885 		mmu_size = ((cvmvmconfig & CVMVMCONF_MMUSIZEM1)
2886 			    >> CVMVMCONF_MMUSIZEM1_S) + 1;
2887 		guest_mmu_size = mmu_size / 2;
2888 		mmu_size -= guest_mmu_size;
2889 		cvmvmconfig &= ~CVMVMCONF_RMMUSIZEM1;
2890 		cvmvmconfig |= mmu_size - 1;
2891 		write_c0_cvmvmconfig(cvmvmconfig);
2892 
2893 		/* Update our records */
2894 		current_cpu_data.tlbsize = mmu_size;
2895 		current_cpu_data.tlbsizevtlb = mmu_size;
2896 		current_cpu_data.guest.tlbsize = guest_mmu_size;
2897 
2898 		/* Flush moved entries in new (guest) context */
2899 		kvm_vz_local_flush_guesttlb_all();
2900 		break;
2901 	default:
2902 		/*
2903 		 * ImgTec cores tend to use a shared root/guest TLB. To avoid
2904 		 * overlap of root wired and guest entries, the guest TLB may
2905 		 * need resizing.
2906 		 */
2907 		mmu_size = current_cpu_data.tlbsizevtlb;
2908 		ftlb_size = current_cpu_data.tlbsize - mmu_size;
2909 
2910 		/* Try switching to maximum guest VTLB size for flush */
2911 		guest_mmu_size = kvm_vz_resize_guest_vtlb(mmu_size);
2912 		current_cpu_data.guest.tlbsize = guest_mmu_size + ftlb_size;
2913 		kvm_vz_local_flush_guesttlb_all();
2914 
2915 		/*
2916 		 * Reduce to make space for root wired entries and at least 2
2917 		 * root non-wired entries. This does assume that long-term wired
2918 		 * entries won't be added later.
2919 		 */
2920 		guest_mmu_size = mmu_size - num_wired_entries() - 2;
2921 		guest_mmu_size = kvm_vz_resize_guest_vtlb(guest_mmu_size);
2922 		current_cpu_data.guest.tlbsize = guest_mmu_size + ftlb_size;
2923 
2924 		/*
2925 		 * Write the VTLB size, but if another CPU has already written,
2926 		 * check it matches or we won't provide a consistent view to the
2927 		 * guest. If this ever happens it suggests an asymmetric number
2928 		 * of wired entries.
2929 		 */
2930 		if (cmpxchg(&kvm_vz_guest_vtlb_size, 0, guest_mmu_size) &&
2931 		    WARN(guest_mmu_size != kvm_vz_guest_vtlb_size,
2932 			 "Available guest VTLB size mismatch"))
2933 			return -EINVAL;
2934 		break;
2935 	}
2936 
2937 	/*
2938 	 * Enable virtualization features granting guest direct control of
2939 	 * certain features:
2940 	 * CP0=1:	Guest coprocessor 0 context.
2941 	 * AT=Guest:	Guest MMU.
2942 	 * CG=1:	Hit (virtual address) CACHE operations (optional).
2943 	 * CF=1:	Guest Config registers.
2944 	 * CGI=1:	Indexed flush CACHE operations (optional).
2945 	 */
2946 	write_c0_guestctl0(MIPS_GCTL0_CP0 |
2947 			   (MIPS_GCTL0_AT_GUEST << MIPS_GCTL0_AT_SHIFT) |
2948 			   MIPS_GCTL0_CG | MIPS_GCTL0_CF);
2949 	if (cpu_has_guestctl0ext) {
2950 		if (current_cpu_type() != CPU_LOONGSON64)
2951 			set_c0_guestctl0ext(MIPS_GCTL0EXT_CGI);
2952 		else
2953 			clear_c0_guestctl0ext(MIPS_GCTL0EXT_CGI);
2954 	}
2955 
2956 	if (cpu_has_guestid) {
2957 		write_c0_guestctl1(0);
2958 		kvm_vz_local_flush_roottlb_all_guests();
2959 
2960 		GUESTID_MASK = current_cpu_data.guestid_mask;
2961 		GUESTID_FIRST_VERSION = GUESTID_MASK + 1;
2962 		GUESTID_VERSION_MASK = ~GUESTID_MASK;
2963 
2964 		current_cpu_data.guestid_cache = GUESTID_FIRST_VERSION;
2965 	}
2966 
2967 	/* clear any pending injected virtual guest interrupts */
2968 	if (cpu_has_guestctl2)
2969 		clear_c0_guestctl2(0x3f << 10);
2970 
2971 #ifdef CONFIG_CPU_LOONGSON64
2972 	/* Control guest CCA attribute */
2973 	if (cpu_has_csr())
2974 		csr_writel(csr_readl(0xffffffec) | 0x1, 0xffffffec);
2975 #endif
2976 
2977 	return 0;
2978 }
2979 
2980 static void kvm_vz_hardware_disable(void)
2981 {
2982 	u64 cvmvmconfig;
2983 	unsigned int mmu_size;
2984 
2985 	/* Flush any remaining guest TLB entries */
2986 	kvm_vz_local_flush_guesttlb_all();
2987 
2988 	switch (current_cpu_type()) {
2989 	case CPU_CAVIUM_OCTEON3:
2990 		/*
2991 		 * Allocate whole TLB for root. Existing guest TLB entries will
2992 		 * change ownership to the root TLB. We should be safe though as
2993 		 * they've already been flushed above while in guest TLB.
2994 		 */
2995 		cvmvmconfig = read_c0_cvmvmconfig();
2996 		mmu_size = ((cvmvmconfig & CVMVMCONF_MMUSIZEM1)
2997 			    >> CVMVMCONF_MMUSIZEM1_S) + 1;
2998 		cvmvmconfig &= ~CVMVMCONF_RMMUSIZEM1;
2999 		cvmvmconfig |= mmu_size - 1;
3000 		write_c0_cvmvmconfig(cvmvmconfig);
3001 
3002 		/* Update our records */
3003 		current_cpu_data.tlbsize = mmu_size;
3004 		current_cpu_data.tlbsizevtlb = mmu_size;
3005 		current_cpu_data.guest.tlbsize = 0;
3006 
3007 		/* Flush moved entries in new (root) context */
3008 		local_flush_tlb_all();
3009 		break;
3010 	}
3011 
3012 	if (cpu_has_guestid) {
3013 		write_c0_guestctl1(0);
3014 		kvm_vz_local_flush_roottlb_all_guests();
3015 	}
3016 }
3017 
3018 static int kvm_vz_check_extension(struct kvm *kvm, long ext)
3019 {
3020 	int r;
3021 
3022 	switch (ext) {
3023 	case KVM_CAP_MIPS_VZ:
3024 		/* we wouldn't be here unless cpu_has_vz */
3025 		r = 1;
3026 		break;
3027 #ifdef CONFIG_64BIT
3028 	case KVM_CAP_MIPS_64BIT:
3029 		/* We support 64-bit registers/operations and addresses */
3030 		r = 2;
3031 		break;
3032 #endif
3033 	case KVM_CAP_IOEVENTFD:
3034 		r = 1;
3035 		break;
3036 	default:
3037 		r = 0;
3038 		break;
3039 	}
3040 
3041 	return r;
3042 }
3043 
3044 static int kvm_vz_vcpu_init(struct kvm_vcpu *vcpu)
3045 {
3046 	int i;
3047 
3048 	for_each_possible_cpu(i)
3049 		vcpu->arch.vzguestid[i] = 0;
3050 
3051 	return 0;
3052 }
3053 
3054 static void kvm_vz_vcpu_uninit(struct kvm_vcpu *vcpu)
3055 {
3056 	int cpu;
3057 
3058 	/*
3059 	 * If the VCPU is freed and reused as another VCPU, we don't want the
3060 	 * matching pointer wrongly hanging around in last_vcpu[] or
3061 	 * last_exec_vcpu[].
3062 	 */
3063 	for_each_possible_cpu(cpu) {
3064 		if (last_vcpu[cpu] == vcpu)
3065 			last_vcpu[cpu] = NULL;
3066 		if (last_exec_vcpu[cpu] == vcpu)
3067 			last_exec_vcpu[cpu] = NULL;
3068 	}
3069 }
3070 
3071 static int kvm_vz_vcpu_setup(struct kvm_vcpu *vcpu)
3072 {
3073 	struct mips_coproc *cop0 = vcpu->arch.cop0;
3074 	unsigned long count_hz = 100*1000*1000; /* default to 100 MHz */
3075 
3076 	/*
3077 	 * Start off the timer at the same frequency as the host timer, but the
3078 	 * soft timer doesn't handle frequencies greater than 1GHz yet.
3079 	 */
3080 	if (mips_hpt_frequency && mips_hpt_frequency <= NSEC_PER_SEC)
3081 		count_hz = mips_hpt_frequency;
3082 	kvm_mips_init_count(vcpu, count_hz);
3083 
3084 	/*
3085 	 * Initialize guest register state to valid architectural reset state.
3086 	 */
3087 
3088 	/* PageGrain */
3089 	if (cpu_has_mips_r5 || cpu_has_mips_r6)
3090 		kvm_write_sw_gc0_pagegrain(cop0, PG_RIE | PG_XIE | PG_IEC);
3091 	/* Wired */
3092 	if (cpu_has_mips_r6)
3093 		kvm_write_sw_gc0_wired(cop0,
3094 				       read_gc0_wired() & MIPSR6_WIRED_LIMIT);
3095 	/* Status */
3096 	kvm_write_sw_gc0_status(cop0, ST0_BEV | ST0_ERL);
3097 	if (cpu_has_mips_r5 || cpu_has_mips_r6)
3098 		kvm_change_sw_gc0_status(cop0, ST0_FR, read_gc0_status());
3099 	/* IntCtl */
3100 	kvm_write_sw_gc0_intctl(cop0, read_gc0_intctl() &
3101 				(INTCTLF_IPFDC | INTCTLF_IPPCI | INTCTLF_IPTI));
3102 	/* PRId */
3103 	kvm_write_sw_gc0_prid(cop0, boot_cpu_data.processor_id);
3104 	/* EBase */
3105 	kvm_write_sw_gc0_ebase(cop0, (s32)0x80000000 | vcpu->vcpu_id);
3106 	/* Config */
3107 	kvm_save_gc0_config(cop0);
3108 	/* architecturally writable (e.g. from guest) */
3109 	kvm_change_sw_gc0_config(cop0, CONF_CM_CMASK,
3110 				 _page_cachable_default >> _CACHE_SHIFT);
3111 	/* architecturally read only, but maybe writable from root */
3112 	kvm_change_sw_gc0_config(cop0, MIPS_CONF_MT, read_c0_config());
3113 	if (cpu_guest_has_conf1) {
3114 		kvm_set_sw_gc0_config(cop0, MIPS_CONF_M);
3115 		/* Config1 */
3116 		kvm_save_gc0_config1(cop0);
3117 		/* architecturally read only, but maybe writable from root */
3118 		kvm_clear_sw_gc0_config1(cop0, MIPS_CONF1_C2	|
3119 					       MIPS_CONF1_MD	|
3120 					       MIPS_CONF1_PC	|
3121 					       MIPS_CONF1_WR	|
3122 					       MIPS_CONF1_CA	|
3123 					       MIPS_CONF1_FP);
3124 	}
3125 	if (cpu_guest_has_conf2) {
3126 		kvm_set_sw_gc0_config1(cop0, MIPS_CONF_M);
3127 		/* Config2 */
3128 		kvm_save_gc0_config2(cop0);
3129 	}
3130 	if (cpu_guest_has_conf3) {
3131 		kvm_set_sw_gc0_config2(cop0, MIPS_CONF_M);
3132 		/* Config3 */
3133 		kvm_save_gc0_config3(cop0);
3134 		/* architecturally writable (e.g. from guest) */
3135 		kvm_clear_sw_gc0_config3(cop0, MIPS_CONF3_ISA_OE);
3136 		/* architecturally read only, but maybe writable from root */
3137 		kvm_clear_sw_gc0_config3(cop0, MIPS_CONF3_MSA	|
3138 					       MIPS_CONF3_BPG	|
3139 					       MIPS_CONF3_ULRI	|
3140 					       MIPS_CONF3_DSP	|
3141 					       MIPS_CONF3_CTXTC	|
3142 					       MIPS_CONF3_ITL	|
3143 					       MIPS_CONF3_LPA	|
3144 					       MIPS_CONF3_VEIC	|
3145 					       MIPS_CONF3_VINT	|
3146 					       MIPS_CONF3_SP	|
3147 					       MIPS_CONF3_CDMM	|
3148 					       MIPS_CONF3_MT	|
3149 					       MIPS_CONF3_SM	|
3150 					       MIPS_CONF3_TL);
3151 	}
3152 	if (cpu_guest_has_conf4) {
3153 		kvm_set_sw_gc0_config3(cop0, MIPS_CONF_M);
3154 		/* Config4 */
3155 		kvm_save_gc0_config4(cop0);
3156 	}
3157 	if (cpu_guest_has_conf5) {
3158 		kvm_set_sw_gc0_config4(cop0, MIPS_CONF_M);
3159 		/* Config5 */
3160 		kvm_save_gc0_config5(cop0);
3161 		/* architecturally writable (e.g. from guest) */
3162 		kvm_clear_sw_gc0_config5(cop0, MIPS_CONF5_K	|
3163 					       MIPS_CONF5_CV	|
3164 					       MIPS_CONF5_MSAEN	|
3165 					       MIPS_CONF5_UFE	|
3166 					       MIPS_CONF5_FRE	|
3167 					       MIPS_CONF5_SBRI	|
3168 					       MIPS_CONF5_UFR);
3169 		/* architecturally read only, but maybe writable from root */
3170 		kvm_clear_sw_gc0_config5(cop0, MIPS_CONF5_MRP);
3171 	}
3172 
3173 	if (cpu_guest_has_contextconfig) {
3174 		/* ContextConfig */
3175 		kvm_write_sw_gc0_contextconfig(cop0, 0x007ffff0);
3176 #ifdef CONFIG_64BIT
3177 		/* XContextConfig */
3178 		/* bits SEGBITS-13+3:4 set */
3179 		kvm_write_sw_gc0_xcontextconfig(cop0,
3180 					((1ull << (cpu_vmbits - 13)) - 1) << 4);
3181 #endif
3182 	}
3183 
3184 	/* Implementation dependent, use the legacy layout */
3185 	if (cpu_guest_has_segments) {
3186 		/* SegCtl0, SegCtl1, SegCtl2 */
3187 		kvm_write_sw_gc0_segctl0(cop0, 0x00200010);
3188 		kvm_write_sw_gc0_segctl1(cop0, 0x00000002 |
3189 				(_page_cachable_default >> _CACHE_SHIFT) <<
3190 						(16 + MIPS_SEGCFG_C_SHIFT));
3191 		kvm_write_sw_gc0_segctl2(cop0, 0x00380438);
3192 	}
3193 
3194 	/* reset HTW registers */
3195 	if (cpu_guest_has_htw && (cpu_has_mips_r5 || cpu_has_mips_r6)) {
3196 		/* PWField */
3197 		kvm_write_sw_gc0_pwfield(cop0, 0x0c30c302);
3198 		/* PWSize */
3199 		kvm_write_sw_gc0_pwsize(cop0, 1 << MIPS_PWSIZE_PTW_SHIFT);
3200 	}
3201 
3202 	/* start with no pending virtual guest interrupts */
3203 	if (cpu_has_guestctl2)
3204 		cop0->reg[MIPS_CP0_GUESTCTL2][MIPS_CP0_GUESTCTL2_SEL] = 0;
3205 
3206 	/* Put PC at reset vector */
3207 	vcpu->arch.pc = CKSEG1ADDR(0x1fc00000);
3208 
3209 	return 0;
3210 }
3211 
3212 static void kvm_vz_prepare_flush_shadow(struct kvm *kvm)
3213 {
3214 	if (!cpu_has_guestid) {
3215 		/*
3216 		 * For each CPU there is a single GPA ASID used by all VCPUs in
3217 		 * the VM, so it doesn't make sense for the VCPUs to handle
3218 		 * invalidation of these ASIDs individually.
3219 		 *
3220 		 * Instead mark all CPUs as needing ASID invalidation in
3221 		 * asid_flush_mask, and kvm_flush_remote_tlbs(kvm) will
3222 		 * kick any running VCPUs so they check asid_flush_mask.
3223 		 */
3224 		cpumask_setall(&kvm->arch.asid_flush_mask);
3225 	}
3226 }
3227 
3228 static void kvm_vz_vcpu_reenter(struct kvm_vcpu *vcpu)
3229 {
3230 	int cpu = smp_processor_id();
3231 	int preserve_guest_tlb;
3232 
3233 	preserve_guest_tlb = kvm_vz_check_requests(vcpu, cpu);
3234 
3235 	if (preserve_guest_tlb)
3236 		kvm_vz_vcpu_save_wired(vcpu);
3237 
3238 	kvm_vz_vcpu_load_tlb(vcpu, cpu);
3239 
3240 	if (preserve_guest_tlb)
3241 		kvm_vz_vcpu_load_wired(vcpu);
3242 }
3243 
3244 static int kvm_vz_vcpu_run(struct kvm_vcpu *vcpu)
3245 {
3246 	int cpu = smp_processor_id();
3247 	int r;
3248 
3249 	kvm_vz_acquire_htimer(vcpu);
3250 	/* Check if we have any exceptions/interrupts pending */
3251 	kvm_mips_deliver_interrupts(vcpu, read_gc0_cause());
3252 
3253 	kvm_vz_check_requests(vcpu, cpu);
3254 	kvm_vz_vcpu_load_tlb(vcpu, cpu);
3255 	kvm_vz_vcpu_load_wired(vcpu);
3256 
3257 	r = vcpu->arch.vcpu_run(vcpu);
3258 
3259 	kvm_vz_vcpu_save_wired(vcpu);
3260 
3261 	return r;
3262 }
3263 
3264 static struct kvm_mips_callbacks kvm_vz_callbacks = {
3265 	.handle_cop_unusable = kvm_trap_vz_handle_cop_unusable,
3266 	.handle_tlb_mod = kvm_trap_vz_handle_tlb_st_miss,
3267 	.handle_tlb_ld_miss = kvm_trap_vz_handle_tlb_ld_miss,
3268 	.handle_tlb_st_miss = kvm_trap_vz_handle_tlb_st_miss,
3269 	.handle_addr_err_st = kvm_trap_vz_no_handler,
3270 	.handle_addr_err_ld = kvm_trap_vz_no_handler,
3271 	.handle_syscall = kvm_trap_vz_no_handler,
3272 	.handle_res_inst = kvm_trap_vz_no_handler,
3273 	.handle_break = kvm_trap_vz_no_handler,
3274 	.handle_msa_disabled = kvm_trap_vz_handle_msa_disabled,
3275 	.handle_guest_exit = kvm_trap_vz_handle_guest_exit,
3276 
3277 	.hardware_enable = kvm_vz_hardware_enable,
3278 	.hardware_disable = kvm_vz_hardware_disable,
3279 	.check_extension = kvm_vz_check_extension,
3280 	.vcpu_init = kvm_vz_vcpu_init,
3281 	.vcpu_uninit = kvm_vz_vcpu_uninit,
3282 	.vcpu_setup = kvm_vz_vcpu_setup,
3283 	.prepare_flush_shadow = kvm_vz_prepare_flush_shadow,
3284 	.gva_to_gpa = kvm_vz_gva_to_gpa_cb,
3285 	.queue_timer_int = kvm_vz_queue_timer_int_cb,
3286 	.dequeue_timer_int = kvm_vz_dequeue_timer_int_cb,
3287 	.queue_io_int = kvm_vz_queue_io_int_cb,
3288 	.dequeue_io_int = kvm_vz_dequeue_io_int_cb,
3289 	.irq_deliver = kvm_vz_irq_deliver_cb,
3290 	.irq_clear = kvm_vz_irq_clear_cb,
3291 	.num_regs = kvm_vz_num_regs,
3292 	.copy_reg_indices = kvm_vz_copy_reg_indices,
3293 	.get_one_reg = kvm_vz_get_one_reg,
3294 	.set_one_reg = kvm_vz_set_one_reg,
3295 	.vcpu_load = kvm_vz_vcpu_load,
3296 	.vcpu_put = kvm_vz_vcpu_put,
3297 	.vcpu_run = kvm_vz_vcpu_run,
3298 	.vcpu_reenter = kvm_vz_vcpu_reenter,
3299 };
3300 
3301 int kvm_mips_emulation_init(struct kvm_mips_callbacks **install_callbacks)
3302 {
3303 	if (!cpu_has_vz)
3304 		return -ENODEV;
3305 
3306 	/*
3307 	 * VZ requires at least 2 KScratch registers, so it should have been
3308 	 * possible to allocate pgd_reg.
3309 	 */
3310 	if (WARN(pgd_reg == -1,
3311 		 "pgd_reg not allocated even though cpu_has_vz\n"))
3312 		return -ENODEV;
3313 
3314 	pr_info("Starting KVM with MIPS VZ extensions\n");
3315 
3316 	*install_callbacks = &kvm_vz_callbacks;
3317 	return 0;
3318 }
3319