xref: /openbmc/linux/arch/mips/kvm/vz.c (revision e792415c)
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 		 * Call to kvm_write_c0_guest_compare() clears Cause.TI in
296 		 * kvm_mips_emulate_CP0(). Explicitly clear irq associated with
297 		 * Cause.IP[IPTI] if GuestCtl2 virtual interrupt register not
298 		 * supported or if not using GuestCtl2 Hardware Clear.
299 		 */
300 		if (cpu_has_guestctl2) {
301 			if (!(read_c0_guestctl2() & (irq << 14)))
302 				clear_c0_guestctl2(irq);
303 		} else {
304 			clear_gc0_cause(irq);
305 		}
306 		break;
307 
308 	case MIPS_EXC_INT_IO_1:
309 	case MIPS_EXC_INT_IO_2:
310 	case MIPS_EXC_INT_IPI_1:
311 	case MIPS_EXC_INT_IPI_2:
312 		/* Clear GuestCtl2.VIP irq if not using Hardware Clear */
313 		if (cpu_has_guestctl2) {
314 			if (!(read_c0_guestctl2() & (irq << 14)))
315 				clear_c0_guestctl2(irq);
316 		} else {
317 			clear_gc0_cause(irq);
318 		}
319 		break;
320 
321 	default:
322 		break;
323 	}
324 
325 	clear_bit(priority, &vcpu->arch.pending_exceptions_clr);
326 	return 1;
327 }
328 
329 /*
330  * VZ guest timer handling.
331  */
332 
333 /**
334  * kvm_vz_should_use_htimer() - Find whether to use the VZ hard guest timer.
335  * @vcpu:	Virtual CPU.
336  *
337  * Returns:	true if the VZ GTOffset & real guest CP0_Count should be used
338  *		instead of software emulation of guest timer.
339  *		false otherwise.
340  */
341 static bool kvm_vz_should_use_htimer(struct kvm_vcpu *vcpu)
342 {
343 	if (kvm_mips_count_disabled(vcpu))
344 		return false;
345 
346 	/* Chosen frequency must match real frequency */
347 	if (mips_hpt_frequency != vcpu->arch.count_hz)
348 		return false;
349 
350 	/* We don't support a CP0_GTOffset with fewer bits than CP0_Count */
351 	if (current_cpu_data.gtoffset_mask != 0xffffffff)
352 		return false;
353 
354 	return true;
355 }
356 
357 /**
358  * _kvm_vz_restore_stimer() - Restore soft timer state.
359  * @vcpu:	Virtual CPU.
360  * @compare:	CP0_Compare register value, restored by caller.
361  * @cause:	CP0_Cause register to restore.
362  *
363  * Restore VZ state relating to the soft timer. The hard timer can be enabled
364  * later.
365  */
366 static void _kvm_vz_restore_stimer(struct kvm_vcpu *vcpu, u32 compare,
367 				   u32 cause)
368 {
369 	/*
370 	 * Avoid spurious counter interrupts by setting Guest CP0_Count to just
371 	 * after Guest CP0_Compare.
372 	 */
373 	write_c0_gtoffset(compare - read_c0_count());
374 
375 	back_to_back_c0_hazard();
376 	write_gc0_cause(cause);
377 }
378 
379 /**
380  * _kvm_vz_restore_htimer() - Restore hard timer state.
381  * @vcpu:	Virtual CPU.
382  * @compare:	CP0_Compare register value, restored by caller.
383  * @cause:	CP0_Cause register to restore.
384  *
385  * Restore hard timer Guest.Count & Guest.Cause taking care to preserve the
386  * value of Guest.CP0_Cause.TI while restoring Guest.CP0_Cause.
387  */
388 static void _kvm_vz_restore_htimer(struct kvm_vcpu *vcpu,
389 				   u32 compare, u32 cause)
390 {
391 	u32 start_count, after_count;
392 	ktime_t freeze_time;
393 	unsigned long flags;
394 
395 	/*
396 	 * Freeze the soft-timer and sync the guest CP0_Count with it. We do
397 	 * this with interrupts disabled to avoid latency.
398 	 */
399 	local_irq_save(flags);
400 	freeze_time = kvm_mips_freeze_hrtimer(vcpu, &start_count);
401 	write_c0_gtoffset(start_count - read_c0_count());
402 	local_irq_restore(flags);
403 
404 	/* restore guest CP0_Cause, as TI may already be set */
405 	back_to_back_c0_hazard();
406 	write_gc0_cause(cause);
407 
408 	/*
409 	 * The above sequence isn't atomic and would result in lost timer
410 	 * interrupts if we're not careful. Detect if a timer interrupt is due
411 	 * and assert it.
412 	 */
413 	back_to_back_c0_hazard();
414 	after_count = read_gc0_count();
415 	if (after_count - start_count > compare - start_count - 1)
416 		kvm_vz_queue_irq(vcpu, MIPS_EXC_INT_TIMER);
417 }
418 
419 /**
420  * kvm_vz_restore_timer() - Restore timer state.
421  * @vcpu:	Virtual CPU.
422  *
423  * Restore soft timer state from saved context.
424  */
425 static void kvm_vz_restore_timer(struct kvm_vcpu *vcpu)
426 {
427 	struct mips_coproc *cop0 = vcpu->arch.cop0;
428 	u32 cause, compare;
429 
430 	compare = kvm_read_sw_gc0_compare(cop0);
431 	cause = kvm_read_sw_gc0_cause(cop0);
432 
433 	write_gc0_compare(compare);
434 	_kvm_vz_restore_stimer(vcpu, compare, cause);
435 }
436 
437 /**
438  * kvm_vz_acquire_htimer() - Switch to hard timer state.
439  * @vcpu:	Virtual CPU.
440  *
441  * Restore hard timer state on top of existing soft timer state if possible.
442  *
443  * Since hard timer won't remain active over preemption, preemption should be
444  * disabled by the caller.
445  */
446 void kvm_vz_acquire_htimer(struct kvm_vcpu *vcpu)
447 {
448 	u32 gctl0;
449 
450 	gctl0 = read_c0_guestctl0();
451 	if (!(gctl0 & MIPS_GCTL0_GT) && kvm_vz_should_use_htimer(vcpu)) {
452 		/* enable guest access to hard timer */
453 		write_c0_guestctl0(gctl0 | MIPS_GCTL0_GT);
454 
455 		_kvm_vz_restore_htimer(vcpu, read_gc0_compare(),
456 				       read_gc0_cause());
457 	}
458 }
459 
460 /**
461  * _kvm_vz_save_htimer() - Switch to software emulation of guest timer.
462  * @vcpu:	Virtual CPU.
463  * @compare:	Pointer to write compare value to.
464  * @cause:	Pointer to write cause value to.
465  *
466  * Save VZ guest timer state and switch to software emulation of guest CP0
467  * timer. The hard timer must already be in use, so preemption should be
468  * disabled.
469  */
470 static void _kvm_vz_save_htimer(struct kvm_vcpu *vcpu,
471 				u32 *out_compare, u32 *out_cause)
472 {
473 	u32 cause, compare, before_count, end_count;
474 	ktime_t before_time;
475 
476 	compare = read_gc0_compare();
477 	*out_compare = compare;
478 
479 	before_time = ktime_get();
480 
481 	/*
482 	 * Record the CP0_Count *prior* to saving CP0_Cause, so we have a time
483 	 * at which no pending timer interrupt is missing.
484 	 */
485 	before_count = read_gc0_count();
486 	back_to_back_c0_hazard();
487 	cause = read_gc0_cause();
488 	*out_cause = cause;
489 
490 	/*
491 	 * Record a final CP0_Count which we will transfer to the soft-timer.
492 	 * This is recorded *after* saving CP0_Cause, so we don't get any timer
493 	 * interrupts from just after the final CP0_Count point.
494 	 */
495 	back_to_back_c0_hazard();
496 	end_count = read_gc0_count();
497 
498 	/*
499 	 * The above sequence isn't atomic, so we could miss a timer interrupt
500 	 * between reading CP0_Cause and end_count. Detect and record any timer
501 	 * interrupt due between before_count and end_count.
502 	 */
503 	if (end_count - before_count > compare - before_count - 1)
504 		kvm_vz_queue_irq(vcpu, MIPS_EXC_INT_TIMER);
505 
506 	/*
507 	 * Restore soft-timer, ignoring a small amount of negative drift due to
508 	 * delay between freeze_hrtimer and setting CP0_GTOffset.
509 	 */
510 	kvm_mips_restore_hrtimer(vcpu, before_time, end_count, -0x10000);
511 }
512 
513 /**
514  * kvm_vz_save_timer() - Save guest timer state.
515  * @vcpu:	Virtual CPU.
516  *
517  * Save VZ guest timer state and switch to soft guest timer if hard timer was in
518  * use.
519  */
520 static void kvm_vz_save_timer(struct kvm_vcpu *vcpu)
521 {
522 	struct mips_coproc *cop0 = vcpu->arch.cop0;
523 	u32 gctl0, compare, cause;
524 
525 	gctl0 = read_c0_guestctl0();
526 	if (gctl0 & MIPS_GCTL0_GT) {
527 		/* disable guest use of hard timer */
528 		write_c0_guestctl0(gctl0 & ~MIPS_GCTL0_GT);
529 
530 		/* save hard timer state */
531 		_kvm_vz_save_htimer(vcpu, &compare, &cause);
532 	} else {
533 		compare = read_gc0_compare();
534 		cause = read_gc0_cause();
535 	}
536 
537 	/* save timer-related state to VCPU context */
538 	kvm_write_sw_gc0_cause(cop0, cause);
539 	kvm_write_sw_gc0_compare(cop0, compare);
540 }
541 
542 /**
543  * kvm_vz_lose_htimer() - Ensure hard guest timer is not in use.
544  * @vcpu:	Virtual CPU.
545  *
546  * Transfers the state of the hard guest timer to the soft guest timer, leaving
547  * guest state intact so it can continue to be used with the soft timer.
548  */
549 void kvm_vz_lose_htimer(struct kvm_vcpu *vcpu)
550 {
551 	u32 gctl0, compare, cause;
552 
553 	preempt_disable();
554 	gctl0 = read_c0_guestctl0();
555 	if (gctl0 & MIPS_GCTL0_GT) {
556 		/* disable guest use of timer */
557 		write_c0_guestctl0(gctl0 & ~MIPS_GCTL0_GT);
558 
559 		/* switch to soft timer */
560 		_kvm_vz_save_htimer(vcpu, &compare, &cause);
561 
562 		/* leave soft timer in usable state */
563 		_kvm_vz_restore_stimer(vcpu, compare, cause);
564 	}
565 	preempt_enable();
566 }
567 
568 /**
569  * is_eva_access() - Find whether an instruction is an EVA memory accessor.
570  * @inst:	32-bit instruction encoding.
571  *
572  * Finds whether @inst encodes an EVA memory access instruction, which would
573  * indicate that emulation of it should access the user mode address space
574  * instead of the kernel mode address space. This matters for MUSUK segments
575  * which are TLB mapped for user mode but unmapped for kernel mode.
576  *
577  * Returns:	Whether @inst encodes an EVA accessor instruction.
578  */
579 static bool is_eva_access(union mips_instruction inst)
580 {
581 	if (inst.spec3_format.opcode != spec3_op)
582 		return false;
583 
584 	switch (inst.spec3_format.func) {
585 	case lwle_op:
586 	case lwre_op:
587 	case cachee_op:
588 	case sbe_op:
589 	case she_op:
590 	case sce_op:
591 	case swe_op:
592 	case swle_op:
593 	case swre_op:
594 	case prefe_op:
595 	case lbue_op:
596 	case lhue_op:
597 	case lbe_op:
598 	case lhe_op:
599 	case lle_op:
600 	case lwe_op:
601 		return true;
602 	default:
603 		return false;
604 	}
605 }
606 
607 /**
608  * is_eva_am_mapped() - Find whether an access mode is mapped.
609  * @vcpu:	KVM VCPU state.
610  * @am:		3-bit encoded access mode.
611  * @eu:		Segment becomes unmapped and uncached when Status.ERL=1.
612  *
613  * Decode @am to find whether it encodes a mapped segment for the current VCPU
614  * state. Where necessary @eu and the actual instruction causing the fault are
615  * taken into account to make the decision.
616  *
617  * Returns:	Whether the VCPU faulted on a TLB mapped address.
618  */
619 static bool is_eva_am_mapped(struct kvm_vcpu *vcpu, unsigned int am, bool eu)
620 {
621 	u32 am_lookup;
622 	int err;
623 
624 	/*
625 	 * Interpret access control mode. We assume address errors will already
626 	 * have been caught by the guest, leaving us with:
627 	 *      AM      UM  SM  KM  31..24 23..16
628 	 * UK    0 000          Unm   0      0
629 	 * MK    1 001          TLB   1
630 	 * MSK   2 010      TLB TLB   1
631 	 * MUSK  3 011  TLB TLB TLB   1
632 	 * MUSUK 4 100  TLB TLB Unm   0      1
633 	 * USK   5 101      Unm Unm   0      0
634 	 * -     6 110                0      0
635 	 * UUSK  7 111  Unm Unm Unm   0      0
636 	 *
637 	 * We shift a magic value by AM across the sign bit to find if always
638 	 * TLB mapped, and if not shift by 8 again to find if it depends on KM.
639 	 */
640 	am_lookup = 0x70080000 << am;
641 	if ((s32)am_lookup < 0) {
642 		/*
643 		 * MK, MSK, MUSK
644 		 * Always TLB mapped, unless SegCtl.EU && ERL
645 		 */
646 		if (!eu || !(read_gc0_status() & ST0_ERL))
647 			return true;
648 	} else {
649 		am_lookup <<= 8;
650 		if ((s32)am_lookup < 0) {
651 			union mips_instruction inst;
652 			unsigned int status;
653 			u32 *opc;
654 
655 			/*
656 			 * MUSUK
657 			 * TLB mapped if not in kernel mode
658 			 */
659 			status = read_gc0_status();
660 			if (!(status & (ST0_EXL | ST0_ERL)) &&
661 			    (status & ST0_KSU))
662 				return true;
663 			/*
664 			 * EVA access instructions in kernel
665 			 * mode access user address space.
666 			 */
667 			opc = (u32 *)vcpu->arch.pc;
668 			if (vcpu->arch.host_cp0_cause & CAUSEF_BD)
669 				opc += 1;
670 			err = kvm_get_badinstr(opc, vcpu, &inst.word);
671 			if (!err && is_eva_access(inst))
672 				return true;
673 		}
674 	}
675 
676 	return false;
677 }
678 
679 /**
680  * kvm_vz_gva_to_gpa() - Convert valid GVA to GPA.
681  * @vcpu:	KVM VCPU state.
682  * @gva:	Guest virtual address to convert.
683  * @gpa:	Output guest physical address.
684  *
685  * Convert a guest virtual address (GVA) which is valid according to the guest
686  * context, to a guest physical address (GPA).
687  *
688  * Returns:	0 on success.
689  *		-errno on failure.
690  */
691 static int kvm_vz_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
692 			     unsigned long *gpa)
693 {
694 	u32 gva32 = gva;
695 	unsigned long segctl;
696 
697 	if ((long)gva == (s32)gva32) {
698 		/* Handle canonical 32-bit virtual address */
699 		if (cpu_guest_has_segments) {
700 			unsigned long mask, pa;
701 
702 			switch (gva32 >> 29) {
703 			case 0:
704 			case 1: /* CFG5 (1GB) */
705 				segctl = read_gc0_segctl2() >> 16;
706 				mask = (unsigned long)0xfc0000000ull;
707 				break;
708 			case 2:
709 			case 3: /* CFG4 (1GB) */
710 				segctl = read_gc0_segctl2();
711 				mask = (unsigned long)0xfc0000000ull;
712 				break;
713 			case 4: /* CFG3 (512MB) */
714 				segctl = read_gc0_segctl1() >> 16;
715 				mask = (unsigned long)0xfe0000000ull;
716 				break;
717 			case 5: /* CFG2 (512MB) */
718 				segctl = read_gc0_segctl1();
719 				mask = (unsigned long)0xfe0000000ull;
720 				break;
721 			case 6: /* CFG1 (512MB) */
722 				segctl = read_gc0_segctl0() >> 16;
723 				mask = (unsigned long)0xfe0000000ull;
724 				break;
725 			case 7: /* CFG0 (512MB) */
726 				segctl = read_gc0_segctl0();
727 				mask = (unsigned long)0xfe0000000ull;
728 				break;
729 			default:
730 				/*
731 				 * GCC 4.9 isn't smart enough to figure out that
732 				 * segctl and mask are always initialised.
733 				 */
734 				unreachable();
735 			}
736 
737 			if (is_eva_am_mapped(vcpu, (segctl >> 4) & 0x7,
738 					     segctl & 0x0008))
739 				goto tlb_mapped;
740 
741 			/* Unmapped, find guest physical address */
742 			pa = (segctl << 20) & mask;
743 			pa |= gva32 & ~mask;
744 			*gpa = pa;
745 			return 0;
746 		} else if ((s32)gva32 < (s32)0xc0000000) {
747 			/* legacy unmapped KSeg0 or KSeg1 */
748 			*gpa = gva32 & 0x1fffffff;
749 			return 0;
750 		}
751 #ifdef CONFIG_64BIT
752 	} else if ((gva & 0xc000000000000000) == 0x8000000000000000) {
753 		/* XKPHYS */
754 		if (cpu_guest_has_segments) {
755 			/*
756 			 * Each of the 8 regions can be overridden by SegCtl2.XR
757 			 * to use SegCtl1.XAM.
758 			 */
759 			segctl = read_gc0_segctl2();
760 			if (segctl & (1ull << (56 + ((gva >> 59) & 0x7)))) {
761 				segctl = read_gc0_segctl1();
762 				if (is_eva_am_mapped(vcpu, (segctl >> 59) & 0x7,
763 						     0))
764 					goto tlb_mapped;
765 			}
766 
767 		}
768 		/*
769 		 * Traditionally fully unmapped.
770 		 * Bits 61:59 specify the CCA, which we can just mask off here.
771 		 * Bits 58:PABITS should be zero, but we shouldn't have got here
772 		 * if it wasn't.
773 		 */
774 		*gpa = gva & 0x07ffffffffffffff;
775 		return 0;
776 #endif
777 	}
778 
779 tlb_mapped:
780 	return kvm_vz_guest_tlb_lookup(vcpu, gva, gpa);
781 }
782 
783 /**
784  * kvm_vz_badvaddr_to_gpa() - Convert GVA BadVAddr from root exception to GPA.
785  * @vcpu:	KVM VCPU state.
786  * @badvaddr:	Root BadVAddr.
787  * @gpa:	Output guest physical address.
788  *
789  * VZ implementations are permitted to report guest virtual addresses (GVA) in
790  * BadVAddr on a root exception during guest execution, instead of the more
791  * convenient guest physical addresses (GPA). When we get a GVA, this function
792  * converts it to a GPA, taking into account guest segmentation and guest TLB
793  * state.
794  *
795  * Returns:	0 on success.
796  *		-errno on failure.
797  */
798 static int kvm_vz_badvaddr_to_gpa(struct kvm_vcpu *vcpu, unsigned long badvaddr,
799 				  unsigned long *gpa)
800 {
801 	unsigned int gexccode = (vcpu->arch.host_cp0_guestctl0 &
802 				 MIPS_GCTL0_GEXC) >> MIPS_GCTL0_GEXC_SHIFT;
803 
804 	/* If BadVAddr is GPA, then all is well in the world */
805 	if (likely(gexccode == MIPS_GCTL0_GEXC_GPA)) {
806 		*gpa = badvaddr;
807 		return 0;
808 	}
809 
810 	/* Otherwise we'd expect it to be GVA ... */
811 	if (WARN(gexccode != MIPS_GCTL0_GEXC_GVA,
812 		 "Unexpected gexccode %#x\n", gexccode))
813 		return -EINVAL;
814 
815 	/* ... and we need to perform the GVA->GPA translation in software */
816 	return kvm_vz_gva_to_gpa(vcpu, badvaddr, gpa);
817 }
818 
819 static int kvm_trap_vz_no_handler(struct kvm_vcpu *vcpu)
820 {
821 	u32 *opc = (u32 *) vcpu->arch.pc;
822 	u32 cause = vcpu->arch.host_cp0_cause;
823 	u32 exccode = (cause & CAUSEF_EXCCODE) >> CAUSEB_EXCCODE;
824 	unsigned long badvaddr = vcpu->arch.host_cp0_badvaddr;
825 	u32 inst = 0;
826 
827 	/*
828 	 *  Fetch the instruction.
829 	 */
830 	if (cause & CAUSEF_BD)
831 		opc += 1;
832 	kvm_get_badinstr(opc, vcpu, &inst);
833 
834 	kvm_err("Exception Code: %d not handled @ PC: %p, inst: 0x%08x BadVaddr: %#lx Status: %#x\n",
835 		exccode, opc, inst, badvaddr,
836 		read_gc0_status());
837 	kvm_arch_vcpu_dump_regs(vcpu);
838 	vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
839 	return RESUME_HOST;
840 }
841 
842 static unsigned long mips_process_maar(unsigned int op, unsigned long val)
843 {
844 	/* Mask off unused bits */
845 	unsigned long mask = 0xfffff000 | MIPS_MAAR_S | MIPS_MAAR_VL;
846 
847 	if (read_gc0_pagegrain() & PG_ELPA)
848 		mask |= 0x00ffffff00000000ull;
849 	if (cpu_guest_has_mvh)
850 		mask |= MIPS_MAAR_VH;
851 
852 	/* Set or clear VH */
853 	if (op == mtc_op) {
854 		/* clear VH */
855 		val &= ~MIPS_MAAR_VH;
856 	} else if (op == dmtc_op) {
857 		/* set VH to match VL */
858 		val &= ~MIPS_MAAR_VH;
859 		if (val & MIPS_MAAR_VL)
860 			val |= MIPS_MAAR_VH;
861 	}
862 
863 	return val & mask;
864 }
865 
866 static void kvm_write_maari(struct kvm_vcpu *vcpu, unsigned long val)
867 {
868 	struct mips_coproc *cop0 = vcpu->arch.cop0;
869 
870 	val &= MIPS_MAARI_INDEX;
871 	if (val == MIPS_MAARI_INDEX)
872 		kvm_write_sw_gc0_maari(cop0, ARRAY_SIZE(vcpu->arch.maar) - 1);
873 	else if (val < ARRAY_SIZE(vcpu->arch.maar))
874 		kvm_write_sw_gc0_maari(cop0, val);
875 }
876 
877 static enum emulation_result kvm_vz_gpsi_cop0(union mips_instruction inst,
878 					      u32 *opc, u32 cause,
879 					      struct kvm_vcpu *vcpu)
880 {
881 	struct mips_coproc *cop0 = vcpu->arch.cop0;
882 	enum emulation_result er = EMULATE_DONE;
883 	u32 rt, rd, sel;
884 	unsigned long curr_pc;
885 	unsigned long val;
886 
887 	/*
888 	 * Update PC and hold onto current PC in case there is
889 	 * an error and we want to rollback the PC
890 	 */
891 	curr_pc = vcpu->arch.pc;
892 	er = update_pc(vcpu, cause);
893 	if (er == EMULATE_FAIL)
894 		return er;
895 
896 	if (inst.co_format.co) {
897 		switch (inst.co_format.func) {
898 		case wait_op:
899 			er = kvm_mips_emul_wait(vcpu);
900 			break;
901 		default:
902 			er = EMULATE_FAIL;
903 		}
904 	} else {
905 		rt = inst.c0r_format.rt;
906 		rd = inst.c0r_format.rd;
907 		sel = inst.c0r_format.sel;
908 
909 		switch (inst.c0r_format.rs) {
910 		case dmfc_op:
911 		case mfc_op:
912 #ifdef CONFIG_KVM_MIPS_DEBUG_COP0_COUNTERS
913 			cop0->stat[rd][sel]++;
914 #endif
915 			if (rd == MIPS_CP0_COUNT &&
916 			    sel == 0) {			/* Count */
917 				val = kvm_mips_read_count(vcpu);
918 			} else if (rd == MIPS_CP0_COMPARE &&
919 				   sel == 0) {		/* Compare */
920 				val = read_gc0_compare();
921 			} else if (rd == MIPS_CP0_LLADDR &&
922 				   sel == 0) {		/* LLAddr */
923 				if (cpu_guest_has_rw_llb)
924 					val = read_gc0_lladdr() &
925 						MIPS_LLADDR_LLB;
926 				else
927 					val = 0;
928 			} else if (rd == MIPS_CP0_LLADDR &&
929 				   sel == 1 &&		/* MAAR */
930 				   cpu_guest_has_maar &&
931 				   !cpu_guest_has_dyn_maar) {
932 				/* MAARI must be in range */
933 				BUG_ON(kvm_read_sw_gc0_maari(cop0) >=
934 						ARRAY_SIZE(vcpu->arch.maar));
935 				val = vcpu->arch.maar[
936 					kvm_read_sw_gc0_maari(cop0)];
937 			} else if ((rd == MIPS_CP0_PRID &&
938 				    (sel == 0 ||	/* PRid */
939 				     sel == 2 ||	/* CDMMBase */
940 				     sel == 3)) ||	/* CMGCRBase */
941 				   (rd == MIPS_CP0_STATUS &&
942 				    (sel == 2 ||	/* SRSCtl */
943 				     sel == 3)) ||	/* SRSMap */
944 				   (rd == MIPS_CP0_CONFIG &&
945 				    (sel == 6 ||	/* Config6 */
946 				     sel == 7)) ||	/* Config7 */
947 				   (rd == MIPS_CP0_LLADDR &&
948 				    (sel == 2) &&	/* MAARI */
949 				    cpu_guest_has_maar &&
950 				    !cpu_guest_has_dyn_maar) ||
951 				   (rd == MIPS_CP0_ERRCTL &&
952 				    (sel == 0))) {	/* ErrCtl */
953 				val = cop0->reg[rd][sel];
954 #ifdef CONFIG_CPU_LOONGSON64
955 			} else if (rd == MIPS_CP0_DIAG &&
956 				   (sel == 0)) {	/* Diag */
957 				val = cop0->reg[rd][sel];
958 #endif
959 			} else {
960 				val = 0;
961 				er = EMULATE_FAIL;
962 			}
963 
964 			if (er != EMULATE_FAIL) {
965 				/* Sign extend */
966 				if (inst.c0r_format.rs == mfc_op)
967 					val = (int)val;
968 				vcpu->arch.gprs[rt] = val;
969 			}
970 
971 			trace_kvm_hwr(vcpu, (inst.c0r_format.rs == mfc_op) ?
972 					KVM_TRACE_MFC0 : KVM_TRACE_DMFC0,
973 				      KVM_TRACE_COP0(rd, sel), val);
974 			break;
975 
976 		case dmtc_op:
977 		case mtc_op:
978 #ifdef CONFIG_KVM_MIPS_DEBUG_COP0_COUNTERS
979 			cop0->stat[rd][sel]++;
980 #endif
981 			val = vcpu->arch.gprs[rt];
982 			trace_kvm_hwr(vcpu, (inst.c0r_format.rs == mtc_op) ?
983 					KVM_TRACE_MTC0 : KVM_TRACE_DMTC0,
984 				      KVM_TRACE_COP0(rd, sel), val);
985 
986 			if (rd == MIPS_CP0_COUNT &&
987 			    sel == 0) {			/* Count */
988 				kvm_vz_lose_htimer(vcpu);
989 				kvm_mips_write_count(vcpu, vcpu->arch.gprs[rt]);
990 			} else if (rd == MIPS_CP0_COMPARE &&
991 				   sel == 0) {		/* Compare */
992 				kvm_mips_write_compare(vcpu,
993 						       vcpu->arch.gprs[rt],
994 						       true);
995 			} else if (rd == MIPS_CP0_LLADDR &&
996 				   sel == 0) {		/* LLAddr */
997 				/*
998 				 * P5600 generates GPSI on guest MTC0 LLAddr.
999 				 * Only allow the guest to clear LLB.
1000 				 */
1001 				if (cpu_guest_has_rw_llb &&
1002 				    !(val & MIPS_LLADDR_LLB))
1003 					write_gc0_lladdr(0);
1004 			} else if (rd == MIPS_CP0_LLADDR &&
1005 				   sel == 1 &&		/* MAAR */
1006 				   cpu_guest_has_maar &&
1007 				   !cpu_guest_has_dyn_maar) {
1008 				val = mips_process_maar(inst.c0r_format.rs,
1009 							val);
1010 
1011 				/* MAARI must be in range */
1012 				BUG_ON(kvm_read_sw_gc0_maari(cop0) >=
1013 						ARRAY_SIZE(vcpu->arch.maar));
1014 				vcpu->arch.maar[kvm_read_sw_gc0_maari(cop0)] =
1015 									val;
1016 			} else if (rd == MIPS_CP0_LLADDR &&
1017 				   (sel == 2) &&	/* MAARI */
1018 				   cpu_guest_has_maar &&
1019 				   !cpu_guest_has_dyn_maar) {
1020 				kvm_write_maari(vcpu, val);
1021 			} else if (rd == MIPS_CP0_CONFIG &&
1022 				   (sel == 6)) {
1023 				cop0->reg[rd][sel] = (int)val;
1024 			} else if (rd == MIPS_CP0_ERRCTL &&
1025 				   (sel == 0)) {	/* ErrCtl */
1026 				/* ignore the written value */
1027 #ifdef CONFIG_CPU_LOONGSON64
1028 			} else if (rd == MIPS_CP0_DIAG &&
1029 				   (sel == 0)) {	/* Diag */
1030 				unsigned long flags;
1031 
1032 				local_irq_save(flags);
1033 				if (val & LOONGSON_DIAG_BTB) {
1034 					/* Flush BTB */
1035 					set_c0_diag(LOONGSON_DIAG_BTB);
1036 				}
1037 				if (val & LOONGSON_DIAG_ITLB) {
1038 					/* Flush ITLB */
1039 					set_c0_diag(LOONGSON_DIAG_ITLB);
1040 				}
1041 				if (val & LOONGSON_DIAG_DTLB) {
1042 					/* Flush DTLB */
1043 					set_c0_diag(LOONGSON_DIAG_DTLB);
1044 				}
1045 				if (val & LOONGSON_DIAG_VTLB) {
1046 					/* Flush VTLB */
1047 					kvm_loongson_clear_guest_vtlb();
1048 				}
1049 				if (val & LOONGSON_DIAG_FTLB) {
1050 					/* Flush FTLB */
1051 					kvm_loongson_clear_guest_ftlb();
1052 				}
1053 				local_irq_restore(flags);
1054 #endif
1055 			} else {
1056 				er = EMULATE_FAIL;
1057 			}
1058 			break;
1059 
1060 		default:
1061 			er = EMULATE_FAIL;
1062 			break;
1063 		}
1064 	}
1065 	/* Rollback PC only if emulation was unsuccessful */
1066 	if (er == EMULATE_FAIL) {
1067 		kvm_err("[%#lx]%s: unsupported cop0 instruction 0x%08x\n",
1068 			curr_pc, __func__, inst.word);
1069 
1070 		vcpu->arch.pc = curr_pc;
1071 	}
1072 
1073 	return er;
1074 }
1075 
1076 static enum emulation_result kvm_vz_gpsi_cache(union mips_instruction inst,
1077 					       u32 *opc, u32 cause,
1078 					       struct kvm_vcpu *vcpu)
1079 {
1080 	enum emulation_result er = EMULATE_DONE;
1081 	u32 cache, op_inst, op, base;
1082 	s16 offset;
1083 	struct kvm_vcpu_arch *arch = &vcpu->arch;
1084 	unsigned long va, curr_pc;
1085 
1086 	/*
1087 	 * Update PC and hold onto current PC in case there is
1088 	 * an error and we want to rollback the PC
1089 	 */
1090 	curr_pc = vcpu->arch.pc;
1091 	er = update_pc(vcpu, cause);
1092 	if (er == EMULATE_FAIL)
1093 		return er;
1094 
1095 	base = inst.i_format.rs;
1096 	op_inst = inst.i_format.rt;
1097 	if (cpu_has_mips_r6)
1098 		offset = inst.spec3_format.simmediate;
1099 	else
1100 		offset = inst.i_format.simmediate;
1101 	cache = op_inst & CacheOp_Cache;
1102 	op = op_inst & CacheOp_Op;
1103 
1104 	va = arch->gprs[base] + offset;
1105 
1106 	kvm_debug("CACHE (cache: %#x, op: %#x, base[%d]: %#lx, offset: %#x\n",
1107 		  cache, op, base, arch->gprs[base], offset);
1108 
1109 	/* Secondary or tirtiary cache ops ignored */
1110 	if (cache != Cache_I && cache != Cache_D)
1111 		return EMULATE_DONE;
1112 
1113 	switch (op_inst) {
1114 	case Index_Invalidate_I:
1115 		flush_icache_line_indexed(va);
1116 		return EMULATE_DONE;
1117 	case Index_Writeback_Inv_D:
1118 		flush_dcache_line_indexed(va);
1119 		return EMULATE_DONE;
1120 	case Hit_Invalidate_I:
1121 	case Hit_Invalidate_D:
1122 	case Hit_Writeback_Inv_D:
1123 		if (boot_cpu_type() == CPU_CAVIUM_OCTEON3) {
1124 			/* We can just flush entire icache */
1125 			local_flush_icache_range(0, 0);
1126 			return EMULATE_DONE;
1127 		}
1128 
1129 		/* So far, other platforms support guest hit cache ops */
1130 		break;
1131 	default:
1132 		break;
1133 	}
1134 
1135 	kvm_err("@ %#lx/%#lx CACHE (cache: %#x, op: %#x, base[%d]: %#lx, offset: %#x\n",
1136 		curr_pc, vcpu->arch.gprs[31], cache, op, base, arch->gprs[base],
1137 		offset);
1138 	/* Rollback PC */
1139 	vcpu->arch.pc = curr_pc;
1140 
1141 	return EMULATE_FAIL;
1142 }
1143 
1144 #ifdef CONFIG_CPU_LOONGSON64
1145 static enum emulation_result kvm_vz_gpsi_lwc2(union mips_instruction inst,
1146 					      u32 *opc, u32 cause,
1147 					      struct kvm_vcpu *vcpu)
1148 {
1149 	unsigned int rs, rd;
1150 	unsigned int hostcfg;
1151 	unsigned long curr_pc;
1152 	enum emulation_result er = EMULATE_DONE;
1153 
1154 	/*
1155 	 * Update PC and hold onto current PC in case there is
1156 	 * an error and we want to rollback the PC
1157 	 */
1158 	curr_pc = vcpu->arch.pc;
1159 	er = update_pc(vcpu, cause);
1160 	if (er == EMULATE_FAIL)
1161 		return er;
1162 
1163 	rs = inst.loongson3_lscsr_format.rs;
1164 	rd = inst.loongson3_lscsr_format.rd;
1165 	switch (inst.loongson3_lscsr_format.fr) {
1166 	case 0x8:  /* Read CPUCFG */
1167 		++vcpu->stat.vz_cpucfg_exits;
1168 		hostcfg = read_cpucfg(vcpu->arch.gprs[rs]);
1169 
1170 		switch (vcpu->arch.gprs[rs]) {
1171 		case LOONGSON_CFG0:
1172 			vcpu->arch.gprs[rd] = 0x14c000;
1173 			break;
1174 		case LOONGSON_CFG1:
1175 			hostcfg &= (LOONGSON_CFG1_FP | LOONGSON_CFG1_MMI |
1176 				    LOONGSON_CFG1_MSA1 | LOONGSON_CFG1_MSA2 |
1177 				    LOONGSON_CFG1_SFBP);
1178 			vcpu->arch.gprs[rd] = hostcfg;
1179 			break;
1180 		case LOONGSON_CFG2:
1181 			hostcfg &= (LOONGSON_CFG2_LEXT1 | LOONGSON_CFG2_LEXT2 |
1182 				    LOONGSON_CFG2_LEXT3 | LOONGSON_CFG2_LSPW);
1183 			vcpu->arch.gprs[rd] = hostcfg;
1184 			break;
1185 		case LOONGSON_CFG3:
1186 			vcpu->arch.gprs[rd] = hostcfg;
1187 			break;
1188 		default:
1189 			/* Don't export any other advanced features to guest */
1190 			vcpu->arch.gprs[rd] = 0;
1191 			break;
1192 		}
1193 		break;
1194 
1195 	default:
1196 		kvm_err("lwc2 emulate not impl %d rs %lx @%lx\n",
1197 			inst.loongson3_lscsr_format.fr, vcpu->arch.gprs[rs], curr_pc);
1198 		er = EMULATE_FAIL;
1199 		break;
1200 	}
1201 
1202 	/* Rollback PC only if emulation was unsuccessful */
1203 	if (er == EMULATE_FAIL) {
1204 		kvm_err("[%#lx]%s: unsupported lwc2 instruction 0x%08x 0x%08x\n",
1205 			curr_pc, __func__, inst.word, inst.loongson3_lscsr_format.fr);
1206 
1207 		vcpu->arch.pc = curr_pc;
1208 	}
1209 
1210 	return er;
1211 }
1212 #endif
1213 
1214 static enum emulation_result kvm_trap_vz_handle_gpsi(u32 cause, u32 *opc,
1215 						     struct kvm_vcpu *vcpu)
1216 {
1217 	enum emulation_result er = EMULATE_DONE;
1218 	struct kvm_vcpu_arch *arch = &vcpu->arch;
1219 	union mips_instruction inst;
1220 	int rd, rt, sel;
1221 	int err;
1222 
1223 	/*
1224 	 *  Fetch the instruction.
1225 	 */
1226 	if (cause & CAUSEF_BD)
1227 		opc += 1;
1228 	err = kvm_get_badinstr(opc, vcpu, &inst.word);
1229 	if (err)
1230 		return EMULATE_FAIL;
1231 
1232 	switch (inst.r_format.opcode) {
1233 	case cop0_op:
1234 		er = kvm_vz_gpsi_cop0(inst, opc, cause, vcpu);
1235 		break;
1236 #ifndef CONFIG_CPU_MIPSR6
1237 	case cache_op:
1238 		trace_kvm_exit(vcpu, KVM_TRACE_EXIT_CACHE);
1239 		er = kvm_vz_gpsi_cache(inst, opc, cause, vcpu);
1240 		break;
1241 #endif
1242 #ifdef CONFIG_CPU_LOONGSON64
1243 	case lwc2_op:
1244 		er = kvm_vz_gpsi_lwc2(inst, opc, cause, vcpu);
1245 		break;
1246 #endif
1247 	case spec3_op:
1248 		switch (inst.spec3_format.func) {
1249 #ifdef CONFIG_CPU_MIPSR6
1250 		case cache6_op:
1251 			trace_kvm_exit(vcpu, KVM_TRACE_EXIT_CACHE);
1252 			er = kvm_vz_gpsi_cache(inst, opc, cause, vcpu);
1253 			break;
1254 #endif
1255 		case rdhwr_op:
1256 			if (inst.r_format.rs || (inst.r_format.re >> 3))
1257 				goto unknown;
1258 
1259 			rd = inst.r_format.rd;
1260 			rt = inst.r_format.rt;
1261 			sel = inst.r_format.re & 0x7;
1262 
1263 			switch (rd) {
1264 			case MIPS_HWR_CC:	/* Read count register */
1265 				arch->gprs[rt] =
1266 					(long)(int)kvm_mips_read_count(vcpu);
1267 				break;
1268 			default:
1269 				trace_kvm_hwr(vcpu, KVM_TRACE_RDHWR,
1270 					      KVM_TRACE_HWR(rd, sel), 0);
1271 				goto unknown;
1272 			}
1273 
1274 			trace_kvm_hwr(vcpu, KVM_TRACE_RDHWR,
1275 				      KVM_TRACE_HWR(rd, sel), arch->gprs[rt]);
1276 
1277 			er = update_pc(vcpu, cause);
1278 			break;
1279 		default:
1280 			goto unknown;
1281 		}
1282 		break;
1283 unknown:
1284 
1285 	default:
1286 		kvm_err("GPSI exception not supported (%p/%#x)\n",
1287 				opc, inst.word);
1288 		kvm_arch_vcpu_dump_regs(vcpu);
1289 		er = EMULATE_FAIL;
1290 		break;
1291 	}
1292 
1293 	return er;
1294 }
1295 
1296 static enum emulation_result kvm_trap_vz_handle_gsfc(u32 cause, u32 *opc,
1297 						     struct kvm_vcpu *vcpu)
1298 {
1299 	enum emulation_result er = EMULATE_DONE;
1300 	struct kvm_vcpu_arch *arch = &vcpu->arch;
1301 	union mips_instruction inst;
1302 	int err;
1303 
1304 	/*
1305 	 *  Fetch the instruction.
1306 	 */
1307 	if (cause & CAUSEF_BD)
1308 		opc += 1;
1309 	err = kvm_get_badinstr(opc, vcpu, &inst.word);
1310 	if (err)
1311 		return EMULATE_FAIL;
1312 
1313 	/* complete MTC0 on behalf of guest and advance EPC */
1314 	if (inst.c0r_format.opcode == cop0_op &&
1315 	    inst.c0r_format.rs == mtc_op &&
1316 	    inst.c0r_format.z == 0) {
1317 		int rt = inst.c0r_format.rt;
1318 		int rd = inst.c0r_format.rd;
1319 		int sel = inst.c0r_format.sel;
1320 		unsigned int val = arch->gprs[rt];
1321 		unsigned int old_val, change;
1322 
1323 		trace_kvm_hwr(vcpu, KVM_TRACE_MTC0, KVM_TRACE_COP0(rd, sel),
1324 			      val);
1325 
1326 		if ((rd == MIPS_CP0_STATUS) && (sel == 0)) {
1327 			/* FR bit should read as zero if no FPU */
1328 			if (!kvm_mips_guest_has_fpu(&vcpu->arch))
1329 				val &= ~(ST0_CU1 | ST0_FR);
1330 
1331 			/*
1332 			 * Also don't allow FR to be set if host doesn't support
1333 			 * it.
1334 			 */
1335 			if (!(boot_cpu_data.fpu_id & MIPS_FPIR_F64))
1336 				val &= ~ST0_FR;
1337 
1338 			old_val = read_gc0_status();
1339 			change = val ^ old_val;
1340 
1341 			if (change & ST0_FR) {
1342 				/*
1343 				 * FPU and Vector register state is made
1344 				 * UNPREDICTABLE by a change of FR, so don't
1345 				 * even bother saving it.
1346 				 */
1347 				kvm_drop_fpu(vcpu);
1348 			}
1349 
1350 			/*
1351 			 * If MSA state is already live, it is undefined how it
1352 			 * interacts with FR=0 FPU state, and we don't want to
1353 			 * hit reserved instruction exceptions trying to save
1354 			 * the MSA state later when CU=1 && FR=1, so play it
1355 			 * safe and save it first.
1356 			 */
1357 			if (change & ST0_CU1 && !(val & ST0_FR) &&
1358 			    vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA)
1359 				kvm_lose_fpu(vcpu);
1360 
1361 			write_gc0_status(val);
1362 		} else if ((rd == MIPS_CP0_CAUSE) && (sel == 0)) {
1363 			u32 old_cause = read_gc0_cause();
1364 			u32 change = old_cause ^ val;
1365 
1366 			/* DC bit enabling/disabling timer? */
1367 			if (change & CAUSEF_DC) {
1368 				if (val & CAUSEF_DC) {
1369 					kvm_vz_lose_htimer(vcpu);
1370 					kvm_mips_count_disable_cause(vcpu);
1371 				} else {
1372 					kvm_mips_count_enable_cause(vcpu);
1373 				}
1374 			}
1375 
1376 			/* Only certain bits are RW to the guest */
1377 			change &= (CAUSEF_DC | CAUSEF_IV | CAUSEF_WP |
1378 				   CAUSEF_IP0 | CAUSEF_IP1);
1379 
1380 			/* WP can only be cleared */
1381 			change &= ~CAUSEF_WP | old_cause;
1382 
1383 			write_gc0_cause(old_cause ^ change);
1384 		} else if ((rd == MIPS_CP0_STATUS) && (sel == 1)) { /* IntCtl */
1385 			write_gc0_intctl(val);
1386 		} else if ((rd == MIPS_CP0_CONFIG) && (sel == 5)) {
1387 			old_val = read_gc0_config5();
1388 			change = val ^ old_val;
1389 			/* Handle changes in FPU/MSA modes */
1390 			preempt_disable();
1391 
1392 			/*
1393 			 * Propagate FRE changes immediately if the FPU
1394 			 * context is already loaded.
1395 			 */
1396 			if (change & MIPS_CONF5_FRE &&
1397 			    vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU)
1398 				change_c0_config5(MIPS_CONF5_FRE, val);
1399 
1400 			preempt_enable();
1401 
1402 			val = old_val ^
1403 				(change & kvm_vz_config5_guest_wrmask(vcpu));
1404 			write_gc0_config5(val);
1405 		} else {
1406 			kvm_err("Handle GSFC, unsupported field change @ %p: %#x\n",
1407 			    opc, inst.word);
1408 			er = EMULATE_FAIL;
1409 		}
1410 
1411 		if (er != EMULATE_FAIL)
1412 			er = update_pc(vcpu, cause);
1413 	} else {
1414 		kvm_err("Handle GSFC, unrecognized instruction @ %p: %#x\n",
1415 			opc, inst.word);
1416 		er = EMULATE_FAIL;
1417 	}
1418 
1419 	return er;
1420 }
1421 
1422 static enum emulation_result kvm_trap_vz_handle_ghfc(u32 cause, u32 *opc,
1423 						     struct kvm_vcpu *vcpu)
1424 {
1425 	/*
1426 	 * Presumably this is due to MC (guest mode change), so lets trace some
1427 	 * relevant info.
1428 	 */
1429 	trace_kvm_guest_mode_change(vcpu);
1430 
1431 	return EMULATE_DONE;
1432 }
1433 
1434 static enum emulation_result kvm_trap_vz_handle_hc(u32 cause, u32 *opc,
1435 						   struct kvm_vcpu *vcpu)
1436 {
1437 	enum emulation_result er;
1438 	union mips_instruction inst;
1439 	unsigned long curr_pc;
1440 	int err;
1441 
1442 	if (cause & CAUSEF_BD)
1443 		opc += 1;
1444 	err = kvm_get_badinstr(opc, vcpu, &inst.word);
1445 	if (err)
1446 		return EMULATE_FAIL;
1447 
1448 	/*
1449 	 * Update PC and hold onto current PC in case there is
1450 	 * an error and we want to rollback the PC
1451 	 */
1452 	curr_pc = vcpu->arch.pc;
1453 	er = update_pc(vcpu, cause);
1454 	if (er == EMULATE_FAIL)
1455 		return er;
1456 
1457 	er = kvm_mips_emul_hypcall(vcpu, inst);
1458 	if (er == EMULATE_FAIL)
1459 		vcpu->arch.pc = curr_pc;
1460 
1461 	return er;
1462 }
1463 
1464 static enum emulation_result kvm_trap_vz_no_handler_guest_exit(u32 gexccode,
1465 							u32 cause,
1466 							u32 *opc,
1467 							struct kvm_vcpu *vcpu)
1468 {
1469 	u32 inst;
1470 
1471 	/*
1472 	 *  Fetch the instruction.
1473 	 */
1474 	if (cause & CAUSEF_BD)
1475 		opc += 1;
1476 	kvm_get_badinstr(opc, vcpu, &inst);
1477 
1478 	kvm_err("Guest Exception Code: %d not yet handled @ PC: %p, inst: 0x%08x  Status: %#x\n",
1479 		gexccode, opc, inst, read_gc0_status());
1480 
1481 	return EMULATE_FAIL;
1482 }
1483 
1484 static int kvm_trap_vz_handle_guest_exit(struct kvm_vcpu *vcpu)
1485 {
1486 	u32 *opc = (u32 *) vcpu->arch.pc;
1487 	u32 cause = vcpu->arch.host_cp0_cause;
1488 	enum emulation_result er = EMULATE_DONE;
1489 	u32 gexccode = (vcpu->arch.host_cp0_guestctl0 &
1490 			MIPS_GCTL0_GEXC) >> MIPS_GCTL0_GEXC_SHIFT;
1491 	int ret = RESUME_GUEST;
1492 
1493 	trace_kvm_exit(vcpu, KVM_TRACE_EXIT_GEXCCODE_BASE + gexccode);
1494 	switch (gexccode) {
1495 	case MIPS_GCTL0_GEXC_GPSI:
1496 		++vcpu->stat.vz_gpsi_exits;
1497 		er = kvm_trap_vz_handle_gpsi(cause, opc, vcpu);
1498 		break;
1499 	case MIPS_GCTL0_GEXC_GSFC:
1500 		++vcpu->stat.vz_gsfc_exits;
1501 		er = kvm_trap_vz_handle_gsfc(cause, opc, vcpu);
1502 		break;
1503 	case MIPS_GCTL0_GEXC_HC:
1504 		++vcpu->stat.vz_hc_exits;
1505 		er = kvm_trap_vz_handle_hc(cause, opc, vcpu);
1506 		break;
1507 	case MIPS_GCTL0_GEXC_GRR:
1508 		++vcpu->stat.vz_grr_exits;
1509 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1510 						       vcpu);
1511 		break;
1512 	case MIPS_GCTL0_GEXC_GVA:
1513 		++vcpu->stat.vz_gva_exits;
1514 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1515 						       vcpu);
1516 		break;
1517 	case MIPS_GCTL0_GEXC_GHFC:
1518 		++vcpu->stat.vz_ghfc_exits;
1519 		er = kvm_trap_vz_handle_ghfc(cause, opc, vcpu);
1520 		break;
1521 	case MIPS_GCTL0_GEXC_GPA:
1522 		++vcpu->stat.vz_gpa_exits;
1523 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1524 						       vcpu);
1525 		break;
1526 	default:
1527 		++vcpu->stat.vz_resvd_exits;
1528 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1529 						       vcpu);
1530 		break;
1531 
1532 	}
1533 
1534 	if (er == EMULATE_DONE) {
1535 		ret = RESUME_GUEST;
1536 	} else if (er == EMULATE_HYPERCALL) {
1537 		ret = kvm_mips_handle_hypcall(vcpu);
1538 	} else {
1539 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1540 		ret = RESUME_HOST;
1541 	}
1542 	return ret;
1543 }
1544 
1545 /**
1546  * kvm_trap_vz_handle_cop_unusuable() - Guest used unusable coprocessor.
1547  * @vcpu:	Virtual CPU context.
1548  *
1549  * Handle when the guest attempts to use a coprocessor which hasn't been allowed
1550  * by the root context.
1551  */
1552 static int kvm_trap_vz_handle_cop_unusable(struct kvm_vcpu *vcpu)
1553 {
1554 	u32 cause = vcpu->arch.host_cp0_cause;
1555 	enum emulation_result er = EMULATE_FAIL;
1556 	int ret = RESUME_GUEST;
1557 
1558 	if (((cause & CAUSEF_CE) >> CAUSEB_CE) == 1) {
1559 		/*
1560 		 * If guest FPU not present, the FPU operation should have been
1561 		 * treated as a reserved instruction!
1562 		 * If FPU already in use, we shouldn't get this at all.
1563 		 */
1564 		if (WARN_ON(!kvm_mips_guest_has_fpu(&vcpu->arch) ||
1565 			    vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU)) {
1566 			preempt_enable();
1567 			return EMULATE_FAIL;
1568 		}
1569 
1570 		kvm_own_fpu(vcpu);
1571 		er = EMULATE_DONE;
1572 	}
1573 	/* other coprocessors not handled */
1574 
1575 	switch (er) {
1576 	case EMULATE_DONE:
1577 		ret = RESUME_GUEST;
1578 		break;
1579 
1580 	case EMULATE_FAIL:
1581 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1582 		ret = RESUME_HOST;
1583 		break;
1584 
1585 	default:
1586 		BUG();
1587 	}
1588 	return ret;
1589 }
1590 
1591 /**
1592  * kvm_trap_vz_handle_msa_disabled() - Guest used MSA while disabled in root.
1593  * @vcpu:	Virtual CPU context.
1594  *
1595  * Handle when the guest attempts to use MSA when it is disabled in the root
1596  * context.
1597  */
1598 static int kvm_trap_vz_handle_msa_disabled(struct kvm_vcpu *vcpu)
1599 {
1600 	/*
1601 	 * If MSA not present or not exposed to guest or FR=0, the MSA operation
1602 	 * should have been treated as a reserved instruction!
1603 	 * Same if CU1=1, FR=0.
1604 	 * If MSA already in use, we shouldn't get this at all.
1605 	 */
1606 	if (!kvm_mips_guest_has_msa(&vcpu->arch) ||
1607 	    (read_gc0_status() & (ST0_CU1 | ST0_FR)) == ST0_CU1 ||
1608 	    !(read_gc0_config5() & MIPS_CONF5_MSAEN) ||
1609 	    vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA) {
1610 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1611 		return RESUME_HOST;
1612 	}
1613 
1614 	kvm_own_msa(vcpu);
1615 
1616 	return RESUME_GUEST;
1617 }
1618 
1619 static int kvm_trap_vz_handle_tlb_ld_miss(struct kvm_vcpu *vcpu)
1620 {
1621 	struct kvm_run *run = vcpu->run;
1622 	u32 *opc = (u32 *) vcpu->arch.pc;
1623 	u32 cause = vcpu->arch.host_cp0_cause;
1624 	ulong badvaddr = vcpu->arch.host_cp0_badvaddr;
1625 	union mips_instruction inst;
1626 	enum emulation_result er = EMULATE_DONE;
1627 	int err, ret = RESUME_GUEST;
1628 
1629 	if (kvm_mips_handle_vz_root_tlb_fault(badvaddr, vcpu, false)) {
1630 		/* A code fetch fault doesn't count as an MMIO */
1631 		if (kvm_is_ifetch_fault(&vcpu->arch)) {
1632 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1633 			return RESUME_HOST;
1634 		}
1635 
1636 		/* Fetch the instruction */
1637 		if (cause & CAUSEF_BD)
1638 			opc += 1;
1639 		err = kvm_get_badinstr(opc, vcpu, &inst.word);
1640 		if (err) {
1641 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1642 			return RESUME_HOST;
1643 		}
1644 
1645 		/* Treat as MMIO */
1646 		er = kvm_mips_emulate_load(inst, cause, vcpu);
1647 		if (er == EMULATE_FAIL) {
1648 			kvm_err("Guest Emulate Load from MMIO space failed: PC: %p, BadVaddr: %#lx\n",
1649 				opc, badvaddr);
1650 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1651 		}
1652 	}
1653 
1654 	if (er == EMULATE_DONE) {
1655 		ret = RESUME_GUEST;
1656 	} else if (er == EMULATE_DO_MMIO) {
1657 		run->exit_reason = KVM_EXIT_MMIO;
1658 		ret = RESUME_HOST;
1659 	} else {
1660 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1661 		ret = RESUME_HOST;
1662 	}
1663 	return ret;
1664 }
1665 
1666 static int kvm_trap_vz_handle_tlb_st_miss(struct kvm_vcpu *vcpu)
1667 {
1668 	struct kvm_run *run = vcpu->run;
1669 	u32 *opc = (u32 *) vcpu->arch.pc;
1670 	u32 cause = vcpu->arch.host_cp0_cause;
1671 	ulong badvaddr = vcpu->arch.host_cp0_badvaddr;
1672 	union mips_instruction inst;
1673 	enum emulation_result er = EMULATE_DONE;
1674 	int err;
1675 	int ret = RESUME_GUEST;
1676 
1677 	/* Just try the access again if we couldn't do the translation */
1678 	if (kvm_vz_badvaddr_to_gpa(vcpu, badvaddr, &badvaddr))
1679 		return RESUME_GUEST;
1680 	vcpu->arch.host_cp0_badvaddr = badvaddr;
1681 
1682 	if (kvm_mips_handle_vz_root_tlb_fault(badvaddr, vcpu, true)) {
1683 		/* Fetch the instruction */
1684 		if (cause & CAUSEF_BD)
1685 			opc += 1;
1686 		err = kvm_get_badinstr(opc, vcpu, &inst.word);
1687 		if (err) {
1688 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1689 			return RESUME_HOST;
1690 		}
1691 
1692 		/* Treat as MMIO */
1693 		er = kvm_mips_emulate_store(inst, cause, vcpu);
1694 		if (er == EMULATE_FAIL) {
1695 			kvm_err("Guest Emulate Store to MMIO space failed: PC: %p, BadVaddr: %#lx\n",
1696 				opc, badvaddr);
1697 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1698 		}
1699 	}
1700 
1701 	if (er == EMULATE_DONE) {
1702 		ret = RESUME_GUEST;
1703 	} else if (er == EMULATE_DO_MMIO) {
1704 		run->exit_reason = KVM_EXIT_MMIO;
1705 		ret = RESUME_HOST;
1706 	} else {
1707 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1708 		ret = RESUME_HOST;
1709 	}
1710 	return ret;
1711 }
1712 
1713 static u64 kvm_vz_get_one_regs[] = {
1714 	KVM_REG_MIPS_CP0_INDEX,
1715 	KVM_REG_MIPS_CP0_ENTRYLO0,
1716 	KVM_REG_MIPS_CP0_ENTRYLO1,
1717 	KVM_REG_MIPS_CP0_CONTEXT,
1718 	KVM_REG_MIPS_CP0_PAGEMASK,
1719 	KVM_REG_MIPS_CP0_PAGEGRAIN,
1720 	KVM_REG_MIPS_CP0_WIRED,
1721 	KVM_REG_MIPS_CP0_HWRENA,
1722 	KVM_REG_MIPS_CP0_BADVADDR,
1723 	KVM_REG_MIPS_CP0_COUNT,
1724 	KVM_REG_MIPS_CP0_ENTRYHI,
1725 	KVM_REG_MIPS_CP0_COMPARE,
1726 	KVM_REG_MIPS_CP0_STATUS,
1727 	KVM_REG_MIPS_CP0_INTCTL,
1728 	KVM_REG_MIPS_CP0_CAUSE,
1729 	KVM_REG_MIPS_CP0_EPC,
1730 	KVM_REG_MIPS_CP0_PRID,
1731 	KVM_REG_MIPS_CP0_EBASE,
1732 	KVM_REG_MIPS_CP0_CONFIG,
1733 	KVM_REG_MIPS_CP0_CONFIG1,
1734 	KVM_REG_MIPS_CP0_CONFIG2,
1735 	KVM_REG_MIPS_CP0_CONFIG3,
1736 	KVM_REG_MIPS_CP0_CONFIG4,
1737 	KVM_REG_MIPS_CP0_CONFIG5,
1738 	KVM_REG_MIPS_CP0_CONFIG6,
1739 #ifdef CONFIG_64BIT
1740 	KVM_REG_MIPS_CP0_XCONTEXT,
1741 #endif
1742 	KVM_REG_MIPS_CP0_ERROREPC,
1743 
1744 	KVM_REG_MIPS_COUNT_CTL,
1745 	KVM_REG_MIPS_COUNT_RESUME,
1746 	KVM_REG_MIPS_COUNT_HZ,
1747 };
1748 
1749 static u64 kvm_vz_get_one_regs_contextconfig[] = {
1750 	KVM_REG_MIPS_CP0_CONTEXTCONFIG,
1751 #ifdef CONFIG_64BIT
1752 	KVM_REG_MIPS_CP0_XCONTEXTCONFIG,
1753 #endif
1754 };
1755 
1756 static u64 kvm_vz_get_one_regs_segments[] = {
1757 	KVM_REG_MIPS_CP0_SEGCTL0,
1758 	KVM_REG_MIPS_CP0_SEGCTL1,
1759 	KVM_REG_MIPS_CP0_SEGCTL2,
1760 };
1761 
1762 static u64 kvm_vz_get_one_regs_htw[] = {
1763 	KVM_REG_MIPS_CP0_PWBASE,
1764 	KVM_REG_MIPS_CP0_PWFIELD,
1765 	KVM_REG_MIPS_CP0_PWSIZE,
1766 	KVM_REG_MIPS_CP0_PWCTL,
1767 };
1768 
1769 static u64 kvm_vz_get_one_regs_kscratch[] = {
1770 	KVM_REG_MIPS_CP0_KSCRATCH1,
1771 	KVM_REG_MIPS_CP0_KSCRATCH2,
1772 	KVM_REG_MIPS_CP0_KSCRATCH3,
1773 	KVM_REG_MIPS_CP0_KSCRATCH4,
1774 	KVM_REG_MIPS_CP0_KSCRATCH5,
1775 	KVM_REG_MIPS_CP0_KSCRATCH6,
1776 };
1777 
1778 static unsigned long kvm_vz_num_regs(struct kvm_vcpu *vcpu)
1779 {
1780 	unsigned long ret;
1781 
1782 	ret = ARRAY_SIZE(kvm_vz_get_one_regs);
1783 	if (cpu_guest_has_userlocal)
1784 		++ret;
1785 	if (cpu_guest_has_badinstr)
1786 		++ret;
1787 	if (cpu_guest_has_badinstrp)
1788 		++ret;
1789 	if (cpu_guest_has_contextconfig)
1790 		ret += ARRAY_SIZE(kvm_vz_get_one_regs_contextconfig);
1791 	if (cpu_guest_has_segments)
1792 		ret += ARRAY_SIZE(kvm_vz_get_one_regs_segments);
1793 	if (cpu_guest_has_htw || cpu_guest_has_ldpte)
1794 		ret += ARRAY_SIZE(kvm_vz_get_one_regs_htw);
1795 	if (cpu_guest_has_maar && !cpu_guest_has_dyn_maar)
1796 		ret += 1 + ARRAY_SIZE(vcpu->arch.maar);
1797 	ret += __arch_hweight8(cpu_data[0].guest.kscratch_mask);
1798 
1799 	return ret;
1800 }
1801 
1802 static int kvm_vz_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *indices)
1803 {
1804 	u64 index;
1805 	unsigned int i;
1806 
1807 	if (copy_to_user(indices, kvm_vz_get_one_regs,
1808 			 sizeof(kvm_vz_get_one_regs)))
1809 		return -EFAULT;
1810 	indices += ARRAY_SIZE(kvm_vz_get_one_regs);
1811 
1812 	if (cpu_guest_has_userlocal) {
1813 		index = KVM_REG_MIPS_CP0_USERLOCAL;
1814 		if (copy_to_user(indices, &index, sizeof(index)))
1815 			return -EFAULT;
1816 		++indices;
1817 	}
1818 	if (cpu_guest_has_badinstr) {
1819 		index = KVM_REG_MIPS_CP0_BADINSTR;
1820 		if (copy_to_user(indices, &index, sizeof(index)))
1821 			return -EFAULT;
1822 		++indices;
1823 	}
1824 	if (cpu_guest_has_badinstrp) {
1825 		index = KVM_REG_MIPS_CP0_BADINSTRP;
1826 		if (copy_to_user(indices, &index, sizeof(index)))
1827 			return -EFAULT;
1828 		++indices;
1829 	}
1830 	if (cpu_guest_has_contextconfig) {
1831 		if (copy_to_user(indices, kvm_vz_get_one_regs_contextconfig,
1832 				 sizeof(kvm_vz_get_one_regs_contextconfig)))
1833 			return -EFAULT;
1834 		indices += ARRAY_SIZE(kvm_vz_get_one_regs_contextconfig);
1835 	}
1836 	if (cpu_guest_has_segments) {
1837 		if (copy_to_user(indices, kvm_vz_get_one_regs_segments,
1838 				 sizeof(kvm_vz_get_one_regs_segments)))
1839 			return -EFAULT;
1840 		indices += ARRAY_SIZE(kvm_vz_get_one_regs_segments);
1841 	}
1842 	if (cpu_guest_has_htw || cpu_guest_has_ldpte) {
1843 		if (copy_to_user(indices, kvm_vz_get_one_regs_htw,
1844 				 sizeof(kvm_vz_get_one_regs_htw)))
1845 			return -EFAULT;
1846 		indices += ARRAY_SIZE(kvm_vz_get_one_regs_htw);
1847 	}
1848 	if (cpu_guest_has_maar && !cpu_guest_has_dyn_maar) {
1849 		for (i = 0; i < ARRAY_SIZE(vcpu->arch.maar); ++i) {
1850 			index = KVM_REG_MIPS_CP0_MAAR(i);
1851 			if (copy_to_user(indices, &index, sizeof(index)))
1852 				return -EFAULT;
1853 			++indices;
1854 		}
1855 
1856 		index = KVM_REG_MIPS_CP0_MAARI;
1857 		if (copy_to_user(indices, &index, sizeof(index)))
1858 			return -EFAULT;
1859 		++indices;
1860 	}
1861 	for (i = 0; i < 6; ++i) {
1862 		if (!cpu_guest_has_kscr(i + 2))
1863 			continue;
1864 
1865 		if (copy_to_user(indices, &kvm_vz_get_one_regs_kscratch[i],
1866 				 sizeof(kvm_vz_get_one_regs_kscratch[i])))
1867 			return -EFAULT;
1868 		++indices;
1869 	}
1870 
1871 	return 0;
1872 }
1873 
1874 static inline s64 entrylo_kvm_to_user(unsigned long v)
1875 {
1876 	s64 mask, ret = v;
1877 
1878 	if (BITS_PER_LONG == 32) {
1879 		/*
1880 		 * KVM API exposes 64-bit version of the register, so move the
1881 		 * RI/XI bits up into place.
1882 		 */
1883 		mask = MIPS_ENTRYLO_RI | MIPS_ENTRYLO_XI;
1884 		ret &= ~mask;
1885 		ret |= ((s64)v & mask) << 32;
1886 	}
1887 	return ret;
1888 }
1889 
1890 static inline unsigned long entrylo_user_to_kvm(s64 v)
1891 {
1892 	unsigned long mask, ret = v;
1893 
1894 	if (BITS_PER_LONG == 32) {
1895 		/*
1896 		 * KVM API exposes 64-bit versiono of the register, so move the
1897 		 * RI/XI bits down into place.
1898 		 */
1899 		mask = MIPS_ENTRYLO_RI | MIPS_ENTRYLO_XI;
1900 		ret &= ~mask;
1901 		ret |= (v >> 32) & mask;
1902 	}
1903 	return ret;
1904 }
1905 
1906 static int kvm_vz_get_one_reg(struct kvm_vcpu *vcpu,
1907 			      const struct kvm_one_reg *reg,
1908 			      s64 *v)
1909 {
1910 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1911 	unsigned int idx;
1912 
1913 	switch (reg->id) {
1914 	case KVM_REG_MIPS_CP0_INDEX:
1915 		*v = (long)read_gc0_index();
1916 		break;
1917 	case KVM_REG_MIPS_CP0_ENTRYLO0:
1918 		*v = entrylo_kvm_to_user(read_gc0_entrylo0());
1919 		break;
1920 	case KVM_REG_MIPS_CP0_ENTRYLO1:
1921 		*v = entrylo_kvm_to_user(read_gc0_entrylo1());
1922 		break;
1923 	case KVM_REG_MIPS_CP0_CONTEXT:
1924 		*v = (long)read_gc0_context();
1925 		break;
1926 	case KVM_REG_MIPS_CP0_CONTEXTCONFIG:
1927 		if (!cpu_guest_has_contextconfig)
1928 			return -EINVAL;
1929 		*v = read_gc0_contextconfig();
1930 		break;
1931 	case KVM_REG_MIPS_CP0_USERLOCAL:
1932 		if (!cpu_guest_has_userlocal)
1933 			return -EINVAL;
1934 		*v = read_gc0_userlocal();
1935 		break;
1936 #ifdef CONFIG_64BIT
1937 	case KVM_REG_MIPS_CP0_XCONTEXTCONFIG:
1938 		if (!cpu_guest_has_contextconfig)
1939 			return -EINVAL;
1940 		*v = read_gc0_xcontextconfig();
1941 		break;
1942 #endif
1943 	case KVM_REG_MIPS_CP0_PAGEMASK:
1944 		*v = (long)read_gc0_pagemask();
1945 		break;
1946 	case KVM_REG_MIPS_CP0_PAGEGRAIN:
1947 		*v = (long)read_gc0_pagegrain();
1948 		break;
1949 	case KVM_REG_MIPS_CP0_SEGCTL0:
1950 		if (!cpu_guest_has_segments)
1951 			return -EINVAL;
1952 		*v = read_gc0_segctl0();
1953 		break;
1954 	case KVM_REG_MIPS_CP0_SEGCTL1:
1955 		if (!cpu_guest_has_segments)
1956 			return -EINVAL;
1957 		*v = read_gc0_segctl1();
1958 		break;
1959 	case KVM_REG_MIPS_CP0_SEGCTL2:
1960 		if (!cpu_guest_has_segments)
1961 			return -EINVAL;
1962 		*v = read_gc0_segctl2();
1963 		break;
1964 	case KVM_REG_MIPS_CP0_PWBASE:
1965 		if (!cpu_guest_has_htw && !cpu_guest_has_ldpte)
1966 			return -EINVAL;
1967 		*v = read_gc0_pwbase();
1968 		break;
1969 	case KVM_REG_MIPS_CP0_PWFIELD:
1970 		if (!cpu_guest_has_htw && !cpu_guest_has_ldpte)
1971 			return -EINVAL;
1972 		*v = read_gc0_pwfield();
1973 		break;
1974 	case KVM_REG_MIPS_CP0_PWSIZE:
1975 		if (!cpu_guest_has_htw && !cpu_guest_has_ldpte)
1976 			return -EINVAL;
1977 		*v = read_gc0_pwsize();
1978 		break;
1979 	case KVM_REG_MIPS_CP0_WIRED:
1980 		*v = (long)read_gc0_wired();
1981 		break;
1982 	case KVM_REG_MIPS_CP0_PWCTL:
1983 		if (!cpu_guest_has_htw && !cpu_guest_has_ldpte)
1984 			return -EINVAL;
1985 		*v = read_gc0_pwctl();
1986 		break;
1987 	case KVM_REG_MIPS_CP0_HWRENA:
1988 		*v = (long)read_gc0_hwrena();
1989 		break;
1990 	case KVM_REG_MIPS_CP0_BADVADDR:
1991 		*v = (long)read_gc0_badvaddr();
1992 		break;
1993 	case KVM_REG_MIPS_CP0_BADINSTR:
1994 		if (!cpu_guest_has_badinstr)
1995 			return -EINVAL;
1996 		*v = read_gc0_badinstr();
1997 		break;
1998 	case KVM_REG_MIPS_CP0_BADINSTRP:
1999 		if (!cpu_guest_has_badinstrp)
2000 			return -EINVAL;
2001 		*v = read_gc0_badinstrp();
2002 		break;
2003 	case KVM_REG_MIPS_CP0_COUNT:
2004 		*v = kvm_mips_read_count(vcpu);
2005 		break;
2006 	case KVM_REG_MIPS_CP0_ENTRYHI:
2007 		*v = (long)read_gc0_entryhi();
2008 		break;
2009 	case KVM_REG_MIPS_CP0_COMPARE:
2010 		*v = (long)read_gc0_compare();
2011 		break;
2012 	case KVM_REG_MIPS_CP0_STATUS:
2013 		*v = (long)read_gc0_status();
2014 		break;
2015 	case KVM_REG_MIPS_CP0_INTCTL:
2016 		*v = read_gc0_intctl();
2017 		break;
2018 	case KVM_REG_MIPS_CP0_CAUSE:
2019 		*v = (long)read_gc0_cause();
2020 		break;
2021 	case KVM_REG_MIPS_CP0_EPC:
2022 		*v = (long)read_gc0_epc();
2023 		break;
2024 	case KVM_REG_MIPS_CP0_PRID:
2025 		switch (boot_cpu_type()) {
2026 		case CPU_CAVIUM_OCTEON3:
2027 			/* Octeon III has a read-only guest.PRid */
2028 			*v = read_gc0_prid();
2029 			break;
2030 		default:
2031 			*v = (long)kvm_read_c0_guest_prid(cop0);
2032 			break;
2033 		}
2034 		break;
2035 	case KVM_REG_MIPS_CP0_EBASE:
2036 		*v = kvm_vz_read_gc0_ebase();
2037 		break;
2038 	case KVM_REG_MIPS_CP0_CONFIG:
2039 		*v = read_gc0_config();
2040 		break;
2041 	case KVM_REG_MIPS_CP0_CONFIG1:
2042 		if (!cpu_guest_has_conf1)
2043 			return -EINVAL;
2044 		*v = read_gc0_config1();
2045 		break;
2046 	case KVM_REG_MIPS_CP0_CONFIG2:
2047 		if (!cpu_guest_has_conf2)
2048 			return -EINVAL;
2049 		*v = read_gc0_config2();
2050 		break;
2051 	case KVM_REG_MIPS_CP0_CONFIG3:
2052 		if (!cpu_guest_has_conf3)
2053 			return -EINVAL;
2054 		*v = read_gc0_config3();
2055 		break;
2056 	case KVM_REG_MIPS_CP0_CONFIG4:
2057 		if (!cpu_guest_has_conf4)
2058 			return -EINVAL;
2059 		*v = read_gc0_config4();
2060 		break;
2061 	case KVM_REG_MIPS_CP0_CONFIG5:
2062 		if (!cpu_guest_has_conf5)
2063 			return -EINVAL;
2064 		*v = read_gc0_config5();
2065 		break;
2066 	case KVM_REG_MIPS_CP0_CONFIG6:
2067 		*v = kvm_read_sw_gc0_config6(cop0);
2068 		break;
2069 	case KVM_REG_MIPS_CP0_MAAR(0) ... KVM_REG_MIPS_CP0_MAAR(0x3f):
2070 		if (!cpu_guest_has_maar || cpu_guest_has_dyn_maar)
2071 			return -EINVAL;
2072 		idx = reg->id - KVM_REG_MIPS_CP0_MAAR(0);
2073 		if (idx >= ARRAY_SIZE(vcpu->arch.maar))
2074 			return -EINVAL;
2075 		*v = vcpu->arch.maar[idx];
2076 		break;
2077 	case KVM_REG_MIPS_CP0_MAARI:
2078 		if (!cpu_guest_has_maar || cpu_guest_has_dyn_maar)
2079 			return -EINVAL;
2080 		*v = kvm_read_sw_gc0_maari(vcpu->arch.cop0);
2081 		break;
2082 #ifdef CONFIG_64BIT
2083 	case KVM_REG_MIPS_CP0_XCONTEXT:
2084 		*v = read_gc0_xcontext();
2085 		break;
2086 #endif
2087 	case KVM_REG_MIPS_CP0_ERROREPC:
2088 		*v = (long)read_gc0_errorepc();
2089 		break;
2090 	case KVM_REG_MIPS_CP0_KSCRATCH1 ... KVM_REG_MIPS_CP0_KSCRATCH6:
2091 		idx = reg->id - KVM_REG_MIPS_CP0_KSCRATCH1 + 2;
2092 		if (!cpu_guest_has_kscr(idx))
2093 			return -EINVAL;
2094 		switch (idx) {
2095 		case 2:
2096 			*v = (long)read_gc0_kscratch1();
2097 			break;
2098 		case 3:
2099 			*v = (long)read_gc0_kscratch2();
2100 			break;
2101 		case 4:
2102 			*v = (long)read_gc0_kscratch3();
2103 			break;
2104 		case 5:
2105 			*v = (long)read_gc0_kscratch4();
2106 			break;
2107 		case 6:
2108 			*v = (long)read_gc0_kscratch5();
2109 			break;
2110 		case 7:
2111 			*v = (long)read_gc0_kscratch6();
2112 			break;
2113 		}
2114 		break;
2115 	case KVM_REG_MIPS_COUNT_CTL:
2116 		*v = vcpu->arch.count_ctl;
2117 		break;
2118 	case KVM_REG_MIPS_COUNT_RESUME:
2119 		*v = ktime_to_ns(vcpu->arch.count_resume);
2120 		break;
2121 	case KVM_REG_MIPS_COUNT_HZ:
2122 		*v = vcpu->arch.count_hz;
2123 		break;
2124 	default:
2125 		return -EINVAL;
2126 	}
2127 	return 0;
2128 }
2129 
2130 static int kvm_vz_set_one_reg(struct kvm_vcpu *vcpu,
2131 			      const struct kvm_one_reg *reg,
2132 			      s64 v)
2133 {
2134 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2135 	unsigned int idx;
2136 	int ret = 0;
2137 	unsigned int cur, change;
2138 
2139 	switch (reg->id) {
2140 	case KVM_REG_MIPS_CP0_INDEX:
2141 		write_gc0_index(v);
2142 		break;
2143 	case KVM_REG_MIPS_CP0_ENTRYLO0:
2144 		write_gc0_entrylo0(entrylo_user_to_kvm(v));
2145 		break;
2146 	case KVM_REG_MIPS_CP0_ENTRYLO1:
2147 		write_gc0_entrylo1(entrylo_user_to_kvm(v));
2148 		break;
2149 	case KVM_REG_MIPS_CP0_CONTEXT:
2150 		write_gc0_context(v);
2151 		break;
2152 	case KVM_REG_MIPS_CP0_CONTEXTCONFIG:
2153 		if (!cpu_guest_has_contextconfig)
2154 			return -EINVAL;
2155 		write_gc0_contextconfig(v);
2156 		break;
2157 	case KVM_REG_MIPS_CP0_USERLOCAL:
2158 		if (!cpu_guest_has_userlocal)
2159 			return -EINVAL;
2160 		write_gc0_userlocal(v);
2161 		break;
2162 #ifdef CONFIG_64BIT
2163 	case KVM_REG_MIPS_CP0_XCONTEXTCONFIG:
2164 		if (!cpu_guest_has_contextconfig)
2165 			return -EINVAL;
2166 		write_gc0_xcontextconfig(v);
2167 		break;
2168 #endif
2169 	case KVM_REG_MIPS_CP0_PAGEMASK:
2170 		write_gc0_pagemask(v);
2171 		break;
2172 	case KVM_REG_MIPS_CP0_PAGEGRAIN:
2173 		write_gc0_pagegrain(v);
2174 		break;
2175 	case KVM_REG_MIPS_CP0_SEGCTL0:
2176 		if (!cpu_guest_has_segments)
2177 			return -EINVAL;
2178 		write_gc0_segctl0(v);
2179 		break;
2180 	case KVM_REG_MIPS_CP0_SEGCTL1:
2181 		if (!cpu_guest_has_segments)
2182 			return -EINVAL;
2183 		write_gc0_segctl1(v);
2184 		break;
2185 	case KVM_REG_MIPS_CP0_SEGCTL2:
2186 		if (!cpu_guest_has_segments)
2187 			return -EINVAL;
2188 		write_gc0_segctl2(v);
2189 		break;
2190 	case KVM_REG_MIPS_CP0_PWBASE:
2191 		if (!cpu_guest_has_htw && !cpu_guest_has_ldpte)
2192 			return -EINVAL;
2193 		write_gc0_pwbase(v);
2194 		break;
2195 	case KVM_REG_MIPS_CP0_PWFIELD:
2196 		if (!cpu_guest_has_htw && !cpu_guest_has_ldpte)
2197 			return -EINVAL;
2198 		write_gc0_pwfield(v);
2199 		break;
2200 	case KVM_REG_MIPS_CP0_PWSIZE:
2201 		if (!cpu_guest_has_htw && !cpu_guest_has_ldpte)
2202 			return -EINVAL;
2203 		write_gc0_pwsize(v);
2204 		break;
2205 	case KVM_REG_MIPS_CP0_WIRED:
2206 		change_gc0_wired(MIPSR6_WIRED_WIRED, v);
2207 		break;
2208 	case KVM_REG_MIPS_CP0_PWCTL:
2209 		if (!cpu_guest_has_htw && !cpu_guest_has_ldpte)
2210 			return -EINVAL;
2211 		write_gc0_pwctl(v);
2212 		break;
2213 	case KVM_REG_MIPS_CP0_HWRENA:
2214 		write_gc0_hwrena(v);
2215 		break;
2216 	case KVM_REG_MIPS_CP0_BADVADDR:
2217 		write_gc0_badvaddr(v);
2218 		break;
2219 	case KVM_REG_MIPS_CP0_BADINSTR:
2220 		if (!cpu_guest_has_badinstr)
2221 			return -EINVAL;
2222 		write_gc0_badinstr(v);
2223 		break;
2224 	case KVM_REG_MIPS_CP0_BADINSTRP:
2225 		if (!cpu_guest_has_badinstrp)
2226 			return -EINVAL;
2227 		write_gc0_badinstrp(v);
2228 		break;
2229 	case KVM_REG_MIPS_CP0_COUNT:
2230 		kvm_mips_write_count(vcpu, v);
2231 		break;
2232 	case KVM_REG_MIPS_CP0_ENTRYHI:
2233 		write_gc0_entryhi(v);
2234 		break;
2235 	case KVM_REG_MIPS_CP0_COMPARE:
2236 		kvm_mips_write_compare(vcpu, v, false);
2237 		break;
2238 	case KVM_REG_MIPS_CP0_STATUS:
2239 		write_gc0_status(v);
2240 		break;
2241 	case KVM_REG_MIPS_CP0_INTCTL:
2242 		write_gc0_intctl(v);
2243 		break;
2244 	case KVM_REG_MIPS_CP0_CAUSE:
2245 		/*
2246 		 * If the timer is stopped or started (DC bit) it must look
2247 		 * atomic with changes to the timer interrupt pending bit (TI).
2248 		 * A timer interrupt should not happen in between.
2249 		 */
2250 		if ((read_gc0_cause() ^ v) & CAUSEF_DC) {
2251 			if (v & CAUSEF_DC) {
2252 				/* disable timer first */
2253 				kvm_mips_count_disable_cause(vcpu);
2254 				change_gc0_cause((u32)~CAUSEF_DC, v);
2255 			} else {
2256 				/* enable timer last */
2257 				change_gc0_cause((u32)~CAUSEF_DC, v);
2258 				kvm_mips_count_enable_cause(vcpu);
2259 			}
2260 		} else {
2261 			write_gc0_cause(v);
2262 		}
2263 		break;
2264 	case KVM_REG_MIPS_CP0_EPC:
2265 		write_gc0_epc(v);
2266 		break;
2267 	case KVM_REG_MIPS_CP0_PRID:
2268 		switch (boot_cpu_type()) {
2269 		case CPU_CAVIUM_OCTEON3:
2270 			/* Octeon III has a guest.PRid, but its read-only */
2271 			break;
2272 		default:
2273 			kvm_write_c0_guest_prid(cop0, v);
2274 			break;
2275 		}
2276 		break;
2277 	case KVM_REG_MIPS_CP0_EBASE:
2278 		kvm_vz_write_gc0_ebase(v);
2279 		break;
2280 	case KVM_REG_MIPS_CP0_CONFIG:
2281 		cur = read_gc0_config();
2282 		change = (cur ^ v) & kvm_vz_config_user_wrmask(vcpu);
2283 		if (change) {
2284 			v = cur ^ change;
2285 			write_gc0_config(v);
2286 		}
2287 		break;
2288 	case KVM_REG_MIPS_CP0_CONFIG1:
2289 		if (!cpu_guest_has_conf1)
2290 			break;
2291 		cur = read_gc0_config1();
2292 		change = (cur ^ v) & kvm_vz_config1_user_wrmask(vcpu);
2293 		if (change) {
2294 			v = cur ^ change;
2295 			write_gc0_config1(v);
2296 		}
2297 		break;
2298 	case KVM_REG_MIPS_CP0_CONFIG2:
2299 		if (!cpu_guest_has_conf2)
2300 			break;
2301 		cur = read_gc0_config2();
2302 		change = (cur ^ v) & kvm_vz_config2_user_wrmask(vcpu);
2303 		if (change) {
2304 			v = cur ^ change;
2305 			write_gc0_config2(v);
2306 		}
2307 		break;
2308 	case KVM_REG_MIPS_CP0_CONFIG3:
2309 		if (!cpu_guest_has_conf3)
2310 			break;
2311 		cur = read_gc0_config3();
2312 		change = (cur ^ v) & kvm_vz_config3_user_wrmask(vcpu);
2313 		if (change) {
2314 			v = cur ^ change;
2315 			write_gc0_config3(v);
2316 		}
2317 		break;
2318 	case KVM_REG_MIPS_CP0_CONFIG4:
2319 		if (!cpu_guest_has_conf4)
2320 			break;
2321 		cur = read_gc0_config4();
2322 		change = (cur ^ v) & kvm_vz_config4_user_wrmask(vcpu);
2323 		if (change) {
2324 			v = cur ^ change;
2325 			write_gc0_config4(v);
2326 		}
2327 		break;
2328 	case KVM_REG_MIPS_CP0_CONFIG5:
2329 		if (!cpu_guest_has_conf5)
2330 			break;
2331 		cur = read_gc0_config5();
2332 		change = (cur ^ v) & kvm_vz_config5_user_wrmask(vcpu);
2333 		if (change) {
2334 			v = cur ^ change;
2335 			write_gc0_config5(v);
2336 		}
2337 		break;
2338 	case KVM_REG_MIPS_CP0_CONFIG6:
2339 		cur = kvm_read_sw_gc0_config6(cop0);
2340 		change = (cur ^ v) & kvm_vz_config6_user_wrmask(vcpu);
2341 		if (change) {
2342 			v = cur ^ change;
2343 			kvm_write_sw_gc0_config6(cop0, (int)v);
2344 		}
2345 		break;
2346 	case KVM_REG_MIPS_CP0_MAAR(0) ... KVM_REG_MIPS_CP0_MAAR(0x3f):
2347 		if (!cpu_guest_has_maar || cpu_guest_has_dyn_maar)
2348 			return -EINVAL;
2349 		idx = reg->id - KVM_REG_MIPS_CP0_MAAR(0);
2350 		if (idx >= ARRAY_SIZE(vcpu->arch.maar))
2351 			return -EINVAL;
2352 		vcpu->arch.maar[idx] = mips_process_maar(dmtc_op, v);
2353 		break;
2354 	case KVM_REG_MIPS_CP0_MAARI:
2355 		if (!cpu_guest_has_maar || cpu_guest_has_dyn_maar)
2356 			return -EINVAL;
2357 		kvm_write_maari(vcpu, v);
2358 		break;
2359 #ifdef CONFIG_64BIT
2360 	case KVM_REG_MIPS_CP0_XCONTEXT:
2361 		write_gc0_xcontext(v);
2362 		break;
2363 #endif
2364 	case KVM_REG_MIPS_CP0_ERROREPC:
2365 		write_gc0_errorepc(v);
2366 		break;
2367 	case KVM_REG_MIPS_CP0_KSCRATCH1 ... KVM_REG_MIPS_CP0_KSCRATCH6:
2368 		idx = reg->id - KVM_REG_MIPS_CP0_KSCRATCH1 + 2;
2369 		if (!cpu_guest_has_kscr(idx))
2370 			return -EINVAL;
2371 		switch (idx) {
2372 		case 2:
2373 			write_gc0_kscratch1(v);
2374 			break;
2375 		case 3:
2376 			write_gc0_kscratch2(v);
2377 			break;
2378 		case 4:
2379 			write_gc0_kscratch3(v);
2380 			break;
2381 		case 5:
2382 			write_gc0_kscratch4(v);
2383 			break;
2384 		case 6:
2385 			write_gc0_kscratch5(v);
2386 			break;
2387 		case 7:
2388 			write_gc0_kscratch6(v);
2389 			break;
2390 		}
2391 		break;
2392 	case KVM_REG_MIPS_COUNT_CTL:
2393 		ret = kvm_mips_set_count_ctl(vcpu, v);
2394 		break;
2395 	case KVM_REG_MIPS_COUNT_RESUME:
2396 		ret = kvm_mips_set_count_resume(vcpu, v);
2397 		break;
2398 	case KVM_REG_MIPS_COUNT_HZ:
2399 		ret = kvm_mips_set_count_hz(vcpu, v);
2400 		break;
2401 	default:
2402 		return -EINVAL;
2403 	}
2404 	return ret;
2405 }
2406 
2407 #define guestid_cache(cpu)	(cpu_data[cpu].guestid_cache)
2408 static void kvm_vz_get_new_guestid(unsigned long cpu, struct kvm_vcpu *vcpu)
2409 {
2410 	unsigned long guestid = guestid_cache(cpu);
2411 
2412 	if (!(++guestid & GUESTID_MASK)) {
2413 		if (cpu_has_vtag_icache)
2414 			flush_icache_all();
2415 
2416 		if (!guestid)		/* fix version if needed */
2417 			guestid = GUESTID_FIRST_VERSION;
2418 
2419 		++guestid;		/* guestid 0 reserved for root */
2420 
2421 		/* start new guestid cycle */
2422 		kvm_vz_local_flush_roottlb_all_guests();
2423 		kvm_vz_local_flush_guesttlb_all();
2424 	}
2425 
2426 	guestid_cache(cpu) = guestid;
2427 }
2428 
2429 /* Returns 1 if the guest TLB may be clobbered */
2430 static int kvm_vz_check_requests(struct kvm_vcpu *vcpu, int cpu)
2431 {
2432 	int ret = 0;
2433 	int i;
2434 
2435 	if (!kvm_request_pending(vcpu))
2436 		return 0;
2437 
2438 	if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
2439 		if (cpu_has_guestid) {
2440 			/* Drop all GuestIDs for this VCPU */
2441 			for_each_possible_cpu(i)
2442 				vcpu->arch.vzguestid[i] = 0;
2443 			/* This will clobber guest TLB contents too */
2444 			ret = 1;
2445 		}
2446 		/*
2447 		 * For Root ASID Dealias (RAD) we don't do anything here, but we
2448 		 * still need the request to ensure we recheck asid_flush_mask.
2449 		 * We can still return 0 as only the root TLB will be affected
2450 		 * by a root ASID flush.
2451 		 */
2452 	}
2453 
2454 	return ret;
2455 }
2456 
2457 static void kvm_vz_vcpu_save_wired(struct kvm_vcpu *vcpu)
2458 {
2459 	unsigned int wired = read_gc0_wired();
2460 	struct kvm_mips_tlb *tlbs;
2461 	int i;
2462 
2463 	/* Expand the wired TLB array if necessary */
2464 	wired &= MIPSR6_WIRED_WIRED;
2465 	if (wired > vcpu->arch.wired_tlb_limit) {
2466 		tlbs = krealloc(vcpu->arch.wired_tlb, wired *
2467 				sizeof(*vcpu->arch.wired_tlb), GFP_ATOMIC);
2468 		if (WARN_ON(!tlbs)) {
2469 			/* Save whatever we can */
2470 			wired = vcpu->arch.wired_tlb_limit;
2471 		} else {
2472 			vcpu->arch.wired_tlb = tlbs;
2473 			vcpu->arch.wired_tlb_limit = wired;
2474 		}
2475 	}
2476 
2477 	if (wired)
2478 		/* Save wired entries from the guest TLB */
2479 		kvm_vz_save_guesttlb(vcpu->arch.wired_tlb, 0, wired);
2480 	/* Invalidate any dropped entries since last time */
2481 	for (i = wired; i < vcpu->arch.wired_tlb_used; ++i) {
2482 		vcpu->arch.wired_tlb[i].tlb_hi = UNIQUE_GUEST_ENTRYHI(i);
2483 		vcpu->arch.wired_tlb[i].tlb_lo[0] = 0;
2484 		vcpu->arch.wired_tlb[i].tlb_lo[1] = 0;
2485 		vcpu->arch.wired_tlb[i].tlb_mask = 0;
2486 	}
2487 	vcpu->arch.wired_tlb_used = wired;
2488 }
2489 
2490 static void kvm_vz_vcpu_load_wired(struct kvm_vcpu *vcpu)
2491 {
2492 	/* Load wired entries into the guest TLB */
2493 	if (vcpu->arch.wired_tlb)
2494 		kvm_vz_load_guesttlb(vcpu->arch.wired_tlb, 0,
2495 				     vcpu->arch.wired_tlb_used);
2496 }
2497 
2498 static void kvm_vz_vcpu_load_tlb(struct kvm_vcpu *vcpu, int cpu)
2499 {
2500 	struct kvm *kvm = vcpu->kvm;
2501 	struct mm_struct *gpa_mm = &kvm->arch.gpa_mm;
2502 	bool migrated;
2503 
2504 	/*
2505 	 * Are we entering guest context on a different CPU to last time?
2506 	 * If so, the VCPU's guest TLB state on this CPU may be stale.
2507 	 */
2508 	migrated = (vcpu->arch.last_exec_cpu != cpu);
2509 	vcpu->arch.last_exec_cpu = cpu;
2510 
2511 	/*
2512 	 * A vcpu's GuestID is set in GuestCtl1.ID when the vcpu is loaded and
2513 	 * remains set until another vcpu is loaded in.  As a rule GuestRID
2514 	 * remains zeroed when in root context unless the kernel is busy
2515 	 * manipulating guest tlb entries.
2516 	 */
2517 	if (cpu_has_guestid) {
2518 		/*
2519 		 * Check if our GuestID is of an older version and thus invalid.
2520 		 *
2521 		 * We also discard the stored GuestID if we've executed on
2522 		 * another CPU, as the guest mappings may have changed without
2523 		 * hypervisor knowledge.
2524 		 */
2525 		if (migrated ||
2526 		    (vcpu->arch.vzguestid[cpu] ^ guestid_cache(cpu)) &
2527 					GUESTID_VERSION_MASK) {
2528 			kvm_vz_get_new_guestid(cpu, vcpu);
2529 			vcpu->arch.vzguestid[cpu] = guestid_cache(cpu);
2530 			trace_kvm_guestid_change(vcpu,
2531 						 vcpu->arch.vzguestid[cpu]);
2532 		}
2533 
2534 		/* Restore GuestID */
2535 		change_c0_guestctl1(GUESTID_MASK, vcpu->arch.vzguestid[cpu]);
2536 	} else {
2537 		/*
2538 		 * The Guest TLB only stores a single guest's TLB state, so
2539 		 * flush it if another VCPU has executed on this CPU.
2540 		 *
2541 		 * We also flush if we've executed on another CPU, as the guest
2542 		 * mappings may have changed without hypervisor knowledge.
2543 		 */
2544 		if (migrated || last_exec_vcpu[cpu] != vcpu)
2545 			kvm_vz_local_flush_guesttlb_all();
2546 		last_exec_vcpu[cpu] = vcpu;
2547 
2548 		/*
2549 		 * Root ASID dealiases guest GPA mappings in the root TLB.
2550 		 * Allocate new root ASID if needed.
2551 		 */
2552 		if (cpumask_test_and_clear_cpu(cpu, &kvm->arch.asid_flush_mask))
2553 			get_new_mmu_context(gpa_mm);
2554 		else
2555 			check_mmu_context(gpa_mm);
2556 	}
2557 }
2558 
2559 static int kvm_vz_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2560 {
2561 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2562 	bool migrated, all;
2563 
2564 	/*
2565 	 * Have we migrated to a different CPU?
2566 	 * If so, any old guest TLB state may be stale.
2567 	 */
2568 	migrated = (vcpu->arch.last_sched_cpu != cpu);
2569 
2570 	/*
2571 	 * Was this the last VCPU to run on this CPU?
2572 	 * If not, any old guest state from this VCPU will have been clobbered.
2573 	 */
2574 	all = migrated || (last_vcpu[cpu] != vcpu);
2575 	last_vcpu[cpu] = vcpu;
2576 
2577 	/*
2578 	 * Restore CP0_Wired unconditionally as we clear it after use, and
2579 	 * restore wired guest TLB entries (while in guest context).
2580 	 */
2581 	kvm_restore_gc0_wired(cop0);
2582 	if (current->flags & PF_VCPU) {
2583 		tlbw_use_hazard();
2584 		kvm_vz_vcpu_load_tlb(vcpu, cpu);
2585 		kvm_vz_vcpu_load_wired(vcpu);
2586 	}
2587 
2588 	/*
2589 	 * Restore timer state regardless, as e.g. Cause.TI can change over time
2590 	 * if left unmaintained.
2591 	 */
2592 	kvm_vz_restore_timer(vcpu);
2593 
2594 	/* Set MC bit if we want to trace guest mode changes */
2595 	if (kvm_trace_guest_mode_change)
2596 		set_c0_guestctl0(MIPS_GCTL0_MC);
2597 	else
2598 		clear_c0_guestctl0(MIPS_GCTL0_MC);
2599 
2600 	/* Don't bother restoring registers multiple times unless necessary */
2601 	if (!all)
2602 		return 0;
2603 
2604 	/*
2605 	 * Restore config registers first, as some implementations restrict
2606 	 * writes to other registers when the corresponding feature bits aren't
2607 	 * set. For example Status.CU1 cannot be set unless Config1.FP is set.
2608 	 */
2609 	kvm_restore_gc0_config(cop0);
2610 	if (cpu_guest_has_conf1)
2611 		kvm_restore_gc0_config1(cop0);
2612 	if (cpu_guest_has_conf2)
2613 		kvm_restore_gc0_config2(cop0);
2614 	if (cpu_guest_has_conf3)
2615 		kvm_restore_gc0_config3(cop0);
2616 	if (cpu_guest_has_conf4)
2617 		kvm_restore_gc0_config4(cop0);
2618 	if (cpu_guest_has_conf5)
2619 		kvm_restore_gc0_config5(cop0);
2620 	if (cpu_guest_has_conf6)
2621 		kvm_restore_gc0_config6(cop0);
2622 	if (cpu_guest_has_conf7)
2623 		kvm_restore_gc0_config7(cop0);
2624 
2625 	kvm_restore_gc0_index(cop0);
2626 	kvm_restore_gc0_entrylo0(cop0);
2627 	kvm_restore_gc0_entrylo1(cop0);
2628 	kvm_restore_gc0_context(cop0);
2629 	if (cpu_guest_has_contextconfig)
2630 		kvm_restore_gc0_contextconfig(cop0);
2631 #ifdef CONFIG_64BIT
2632 	kvm_restore_gc0_xcontext(cop0);
2633 	if (cpu_guest_has_contextconfig)
2634 		kvm_restore_gc0_xcontextconfig(cop0);
2635 #endif
2636 	kvm_restore_gc0_pagemask(cop0);
2637 	kvm_restore_gc0_pagegrain(cop0);
2638 	kvm_restore_gc0_hwrena(cop0);
2639 	kvm_restore_gc0_badvaddr(cop0);
2640 	kvm_restore_gc0_entryhi(cop0);
2641 	kvm_restore_gc0_status(cop0);
2642 	kvm_restore_gc0_intctl(cop0);
2643 	kvm_restore_gc0_epc(cop0);
2644 	kvm_vz_write_gc0_ebase(kvm_read_sw_gc0_ebase(cop0));
2645 	if (cpu_guest_has_userlocal)
2646 		kvm_restore_gc0_userlocal(cop0);
2647 
2648 	kvm_restore_gc0_errorepc(cop0);
2649 
2650 	/* restore KScratch registers if enabled in guest */
2651 	if (cpu_guest_has_conf4) {
2652 		if (cpu_guest_has_kscr(2))
2653 			kvm_restore_gc0_kscratch1(cop0);
2654 		if (cpu_guest_has_kscr(3))
2655 			kvm_restore_gc0_kscratch2(cop0);
2656 		if (cpu_guest_has_kscr(4))
2657 			kvm_restore_gc0_kscratch3(cop0);
2658 		if (cpu_guest_has_kscr(5))
2659 			kvm_restore_gc0_kscratch4(cop0);
2660 		if (cpu_guest_has_kscr(6))
2661 			kvm_restore_gc0_kscratch5(cop0);
2662 		if (cpu_guest_has_kscr(7))
2663 			kvm_restore_gc0_kscratch6(cop0);
2664 	}
2665 
2666 	if (cpu_guest_has_badinstr)
2667 		kvm_restore_gc0_badinstr(cop0);
2668 	if (cpu_guest_has_badinstrp)
2669 		kvm_restore_gc0_badinstrp(cop0);
2670 
2671 	if (cpu_guest_has_segments) {
2672 		kvm_restore_gc0_segctl0(cop0);
2673 		kvm_restore_gc0_segctl1(cop0);
2674 		kvm_restore_gc0_segctl2(cop0);
2675 	}
2676 
2677 	/* restore HTW registers */
2678 	if (cpu_guest_has_htw || cpu_guest_has_ldpte) {
2679 		kvm_restore_gc0_pwbase(cop0);
2680 		kvm_restore_gc0_pwfield(cop0);
2681 		kvm_restore_gc0_pwsize(cop0);
2682 		kvm_restore_gc0_pwctl(cop0);
2683 	}
2684 
2685 	/* restore Root.GuestCtl2 from unused Guest guestctl2 register */
2686 	if (cpu_has_guestctl2)
2687 		write_c0_guestctl2(
2688 			cop0->reg[MIPS_CP0_GUESTCTL2][MIPS_CP0_GUESTCTL2_SEL]);
2689 
2690 	/*
2691 	 * We should clear linked load bit to break interrupted atomics. This
2692 	 * prevents a SC on the next VCPU from succeeding by matching a LL on
2693 	 * the previous VCPU.
2694 	 */
2695 	if (vcpu->kvm->created_vcpus > 1)
2696 		write_gc0_lladdr(0);
2697 
2698 	return 0;
2699 }
2700 
2701 static int kvm_vz_vcpu_put(struct kvm_vcpu *vcpu, int cpu)
2702 {
2703 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2704 
2705 	if (current->flags & PF_VCPU)
2706 		kvm_vz_vcpu_save_wired(vcpu);
2707 
2708 	kvm_lose_fpu(vcpu);
2709 
2710 	kvm_save_gc0_index(cop0);
2711 	kvm_save_gc0_entrylo0(cop0);
2712 	kvm_save_gc0_entrylo1(cop0);
2713 	kvm_save_gc0_context(cop0);
2714 	if (cpu_guest_has_contextconfig)
2715 		kvm_save_gc0_contextconfig(cop0);
2716 #ifdef CONFIG_64BIT
2717 	kvm_save_gc0_xcontext(cop0);
2718 	if (cpu_guest_has_contextconfig)
2719 		kvm_save_gc0_xcontextconfig(cop0);
2720 #endif
2721 	kvm_save_gc0_pagemask(cop0);
2722 	kvm_save_gc0_pagegrain(cop0);
2723 	kvm_save_gc0_wired(cop0);
2724 	/* allow wired TLB entries to be overwritten */
2725 	clear_gc0_wired(MIPSR6_WIRED_WIRED);
2726 	kvm_save_gc0_hwrena(cop0);
2727 	kvm_save_gc0_badvaddr(cop0);
2728 	kvm_save_gc0_entryhi(cop0);
2729 	kvm_save_gc0_status(cop0);
2730 	kvm_save_gc0_intctl(cop0);
2731 	kvm_save_gc0_epc(cop0);
2732 	kvm_write_sw_gc0_ebase(cop0, kvm_vz_read_gc0_ebase());
2733 	if (cpu_guest_has_userlocal)
2734 		kvm_save_gc0_userlocal(cop0);
2735 
2736 	/* only save implemented config registers */
2737 	kvm_save_gc0_config(cop0);
2738 	if (cpu_guest_has_conf1)
2739 		kvm_save_gc0_config1(cop0);
2740 	if (cpu_guest_has_conf2)
2741 		kvm_save_gc0_config2(cop0);
2742 	if (cpu_guest_has_conf3)
2743 		kvm_save_gc0_config3(cop0);
2744 	if (cpu_guest_has_conf4)
2745 		kvm_save_gc0_config4(cop0);
2746 	if (cpu_guest_has_conf5)
2747 		kvm_save_gc0_config5(cop0);
2748 	if (cpu_guest_has_conf6)
2749 		kvm_save_gc0_config6(cop0);
2750 	if (cpu_guest_has_conf7)
2751 		kvm_save_gc0_config7(cop0);
2752 
2753 	kvm_save_gc0_errorepc(cop0);
2754 
2755 	/* save KScratch registers if enabled in guest */
2756 	if (cpu_guest_has_conf4) {
2757 		if (cpu_guest_has_kscr(2))
2758 			kvm_save_gc0_kscratch1(cop0);
2759 		if (cpu_guest_has_kscr(3))
2760 			kvm_save_gc0_kscratch2(cop0);
2761 		if (cpu_guest_has_kscr(4))
2762 			kvm_save_gc0_kscratch3(cop0);
2763 		if (cpu_guest_has_kscr(5))
2764 			kvm_save_gc0_kscratch4(cop0);
2765 		if (cpu_guest_has_kscr(6))
2766 			kvm_save_gc0_kscratch5(cop0);
2767 		if (cpu_guest_has_kscr(7))
2768 			kvm_save_gc0_kscratch6(cop0);
2769 	}
2770 
2771 	if (cpu_guest_has_badinstr)
2772 		kvm_save_gc0_badinstr(cop0);
2773 	if (cpu_guest_has_badinstrp)
2774 		kvm_save_gc0_badinstrp(cop0);
2775 
2776 	if (cpu_guest_has_segments) {
2777 		kvm_save_gc0_segctl0(cop0);
2778 		kvm_save_gc0_segctl1(cop0);
2779 		kvm_save_gc0_segctl2(cop0);
2780 	}
2781 
2782 	/* save HTW registers if enabled in guest */
2783 	if (cpu_guest_has_ldpte || (cpu_guest_has_htw &&
2784 	    kvm_read_sw_gc0_config3(cop0) & MIPS_CONF3_PW)) {
2785 		kvm_save_gc0_pwbase(cop0);
2786 		kvm_save_gc0_pwfield(cop0);
2787 		kvm_save_gc0_pwsize(cop0);
2788 		kvm_save_gc0_pwctl(cop0);
2789 	}
2790 
2791 	kvm_vz_save_timer(vcpu);
2792 
2793 	/* save Root.GuestCtl2 in unused Guest guestctl2 register */
2794 	if (cpu_has_guestctl2)
2795 		cop0->reg[MIPS_CP0_GUESTCTL2][MIPS_CP0_GUESTCTL2_SEL] =
2796 			read_c0_guestctl2();
2797 
2798 	return 0;
2799 }
2800 
2801 /**
2802  * kvm_vz_resize_guest_vtlb() - Attempt to resize guest VTLB.
2803  * @size:	Number of guest VTLB entries (0 < @size <= root VTLB entries).
2804  *
2805  * Attempt to resize the guest VTLB by writing guest Config registers. This is
2806  * necessary for cores with a shared root/guest TLB to avoid overlap with wired
2807  * entries in the root VTLB.
2808  *
2809  * Returns:	The resulting guest VTLB size.
2810  */
2811 static unsigned int kvm_vz_resize_guest_vtlb(unsigned int size)
2812 {
2813 	unsigned int config4 = 0, ret = 0, limit;
2814 
2815 	/* Write MMUSize - 1 into guest Config registers */
2816 	if (cpu_guest_has_conf1)
2817 		change_gc0_config1(MIPS_CONF1_TLBS,
2818 				   (size - 1) << MIPS_CONF1_TLBS_SHIFT);
2819 	if (cpu_guest_has_conf4) {
2820 		config4 = read_gc0_config4();
2821 		if (cpu_has_mips_r6 || (config4 & MIPS_CONF4_MMUEXTDEF) ==
2822 		    MIPS_CONF4_MMUEXTDEF_VTLBSIZEEXT) {
2823 			config4 &= ~MIPS_CONF4_VTLBSIZEEXT;
2824 			config4 |= ((size - 1) >> MIPS_CONF1_TLBS_SIZE) <<
2825 				MIPS_CONF4_VTLBSIZEEXT_SHIFT;
2826 		} else if ((config4 & MIPS_CONF4_MMUEXTDEF) ==
2827 			   MIPS_CONF4_MMUEXTDEF_MMUSIZEEXT) {
2828 			config4 &= ~MIPS_CONF4_MMUSIZEEXT;
2829 			config4 |= ((size - 1) >> MIPS_CONF1_TLBS_SIZE) <<
2830 				MIPS_CONF4_MMUSIZEEXT_SHIFT;
2831 		}
2832 		write_gc0_config4(config4);
2833 	}
2834 
2835 	/*
2836 	 * Set Guest.Wired.Limit = 0 (no limit up to Guest.MMUSize-1), unless it
2837 	 * would exceed Root.Wired.Limit (clearing Guest.Wired.Wired so write
2838 	 * not dropped)
2839 	 */
2840 	if (cpu_has_mips_r6) {
2841 		limit = (read_c0_wired() & MIPSR6_WIRED_LIMIT) >>
2842 						MIPSR6_WIRED_LIMIT_SHIFT;
2843 		if (size - 1 <= limit)
2844 			limit = 0;
2845 		write_gc0_wired(limit << MIPSR6_WIRED_LIMIT_SHIFT);
2846 	}
2847 
2848 	/* Read back MMUSize - 1 */
2849 	back_to_back_c0_hazard();
2850 	if (cpu_guest_has_conf1)
2851 		ret = (read_gc0_config1() & MIPS_CONF1_TLBS) >>
2852 						MIPS_CONF1_TLBS_SHIFT;
2853 	if (config4) {
2854 		if (cpu_has_mips_r6 || (config4 & MIPS_CONF4_MMUEXTDEF) ==
2855 		    MIPS_CONF4_MMUEXTDEF_VTLBSIZEEXT)
2856 			ret |= ((config4 & MIPS_CONF4_VTLBSIZEEXT) >>
2857 				MIPS_CONF4_VTLBSIZEEXT_SHIFT) <<
2858 				MIPS_CONF1_TLBS_SIZE;
2859 		else if ((config4 & MIPS_CONF4_MMUEXTDEF) ==
2860 			 MIPS_CONF4_MMUEXTDEF_MMUSIZEEXT)
2861 			ret |= ((config4 & MIPS_CONF4_MMUSIZEEXT) >>
2862 				MIPS_CONF4_MMUSIZEEXT_SHIFT) <<
2863 				MIPS_CONF1_TLBS_SIZE;
2864 	}
2865 	return ret + 1;
2866 }
2867 
2868 static int kvm_vz_hardware_enable(void)
2869 {
2870 	unsigned int mmu_size, guest_mmu_size, ftlb_size;
2871 	u64 guest_cvmctl, cvmvmconfig;
2872 
2873 	switch (current_cpu_type()) {
2874 	case CPU_CAVIUM_OCTEON3:
2875 		/* Set up guest timer/perfcount IRQ lines */
2876 		guest_cvmctl = read_gc0_cvmctl();
2877 		guest_cvmctl &= ~CVMCTL_IPTI;
2878 		guest_cvmctl |= 7ull << CVMCTL_IPTI_SHIFT;
2879 		guest_cvmctl &= ~CVMCTL_IPPCI;
2880 		guest_cvmctl |= 6ull << CVMCTL_IPPCI_SHIFT;
2881 		write_gc0_cvmctl(guest_cvmctl);
2882 
2883 		cvmvmconfig = read_c0_cvmvmconfig();
2884 		/* No I/O hole translation. */
2885 		cvmvmconfig |= CVMVMCONF_DGHT;
2886 		/* Halve the root MMU size */
2887 		mmu_size = ((cvmvmconfig & CVMVMCONF_MMUSIZEM1)
2888 			    >> CVMVMCONF_MMUSIZEM1_S) + 1;
2889 		guest_mmu_size = mmu_size / 2;
2890 		mmu_size -= guest_mmu_size;
2891 		cvmvmconfig &= ~CVMVMCONF_RMMUSIZEM1;
2892 		cvmvmconfig |= mmu_size - 1;
2893 		write_c0_cvmvmconfig(cvmvmconfig);
2894 
2895 		/* Update our records */
2896 		current_cpu_data.tlbsize = mmu_size;
2897 		current_cpu_data.tlbsizevtlb = mmu_size;
2898 		current_cpu_data.guest.tlbsize = guest_mmu_size;
2899 
2900 		/* Flush moved entries in new (guest) context */
2901 		kvm_vz_local_flush_guesttlb_all();
2902 		break;
2903 	default:
2904 		/*
2905 		 * ImgTec cores tend to use a shared root/guest TLB. To avoid
2906 		 * overlap of root wired and guest entries, the guest TLB may
2907 		 * need resizing.
2908 		 */
2909 		mmu_size = current_cpu_data.tlbsizevtlb;
2910 		ftlb_size = current_cpu_data.tlbsize - mmu_size;
2911 
2912 		/* Try switching to maximum guest VTLB size for flush */
2913 		guest_mmu_size = kvm_vz_resize_guest_vtlb(mmu_size);
2914 		current_cpu_data.guest.tlbsize = guest_mmu_size + ftlb_size;
2915 		kvm_vz_local_flush_guesttlb_all();
2916 
2917 		/*
2918 		 * Reduce to make space for root wired entries and at least 2
2919 		 * root non-wired entries. This does assume that long-term wired
2920 		 * entries won't be added later.
2921 		 */
2922 		guest_mmu_size = mmu_size - num_wired_entries() - 2;
2923 		guest_mmu_size = kvm_vz_resize_guest_vtlb(guest_mmu_size);
2924 		current_cpu_data.guest.tlbsize = guest_mmu_size + ftlb_size;
2925 
2926 		/*
2927 		 * Write the VTLB size, but if another CPU has already written,
2928 		 * check it matches or we won't provide a consistent view to the
2929 		 * guest. If this ever happens it suggests an asymmetric number
2930 		 * of wired entries.
2931 		 */
2932 		if (cmpxchg(&kvm_vz_guest_vtlb_size, 0, guest_mmu_size) &&
2933 		    WARN(guest_mmu_size != kvm_vz_guest_vtlb_size,
2934 			 "Available guest VTLB size mismatch"))
2935 			return -EINVAL;
2936 		break;
2937 	}
2938 
2939 	/*
2940 	 * Enable virtualization features granting guest direct control of
2941 	 * certain features:
2942 	 * CP0=1:	Guest coprocessor 0 context.
2943 	 * AT=Guest:	Guest MMU.
2944 	 * CG=1:	Hit (virtual address) CACHE operations (optional).
2945 	 * CF=1:	Guest Config registers.
2946 	 * CGI=1:	Indexed flush CACHE operations (optional).
2947 	 */
2948 	write_c0_guestctl0(MIPS_GCTL0_CP0 |
2949 			   (MIPS_GCTL0_AT_GUEST << MIPS_GCTL0_AT_SHIFT) |
2950 			   MIPS_GCTL0_CG | MIPS_GCTL0_CF);
2951 	if (cpu_has_guestctl0ext) {
2952 		if (current_cpu_type() != CPU_LOONGSON64)
2953 			set_c0_guestctl0ext(MIPS_GCTL0EXT_CGI);
2954 		else
2955 			clear_c0_guestctl0ext(MIPS_GCTL0EXT_CGI);
2956 	}
2957 
2958 	if (cpu_has_guestid) {
2959 		write_c0_guestctl1(0);
2960 		kvm_vz_local_flush_roottlb_all_guests();
2961 
2962 		GUESTID_MASK = current_cpu_data.guestid_mask;
2963 		GUESTID_FIRST_VERSION = GUESTID_MASK + 1;
2964 		GUESTID_VERSION_MASK = ~GUESTID_MASK;
2965 
2966 		current_cpu_data.guestid_cache = GUESTID_FIRST_VERSION;
2967 	}
2968 
2969 	/* clear any pending injected virtual guest interrupts */
2970 	if (cpu_has_guestctl2)
2971 		clear_c0_guestctl2(0x3f << 10);
2972 
2973 #ifdef CONFIG_CPU_LOONGSON64
2974 	/* Control guest CCA attribute */
2975 	if (cpu_has_csr())
2976 		csr_writel(csr_readl(0xffffffec) | 0x1, 0xffffffec);
2977 #endif
2978 
2979 	return 0;
2980 }
2981 
2982 static void kvm_vz_hardware_disable(void)
2983 {
2984 	u64 cvmvmconfig;
2985 	unsigned int mmu_size;
2986 
2987 	/* Flush any remaining guest TLB entries */
2988 	kvm_vz_local_flush_guesttlb_all();
2989 
2990 	switch (current_cpu_type()) {
2991 	case CPU_CAVIUM_OCTEON3:
2992 		/*
2993 		 * Allocate whole TLB for root. Existing guest TLB entries will
2994 		 * change ownership to the root TLB. We should be safe though as
2995 		 * they've already been flushed above while in guest TLB.
2996 		 */
2997 		cvmvmconfig = read_c0_cvmvmconfig();
2998 		mmu_size = ((cvmvmconfig & CVMVMCONF_MMUSIZEM1)
2999 			    >> CVMVMCONF_MMUSIZEM1_S) + 1;
3000 		cvmvmconfig &= ~CVMVMCONF_RMMUSIZEM1;
3001 		cvmvmconfig |= mmu_size - 1;
3002 		write_c0_cvmvmconfig(cvmvmconfig);
3003 
3004 		/* Update our records */
3005 		current_cpu_data.tlbsize = mmu_size;
3006 		current_cpu_data.tlbsizevtlb = mmu_size;
3007 		current_cpu_data.guest.tlbsize = 0;
3008 
3009 		/* Flush moved entries in new (root) context */
3010 		local_flush_tlb_all();
3011 		break;
3012 	}
3013 
3014 	if (cpu_has_guestid) {
3015 		write_c0_guestctl1(0);
3016 		kvm_vz_local_flush_roottlb_all_guests();
3017 	}
3018 }
3019 
3020 static int kvm_vz_check_extension(struct kvm *kvm, long ext)
3021 {
3022 	int r;
3023 
3024 	switch (ext) {
3025 	case KVM_CAP_MIPS_VZ:
3026 		/* we wouldn't be here unless cpu_has_vz */
3027 		r = 1;
3028 		break;
3029 #ifdef CONFIG_64BIT
3030 	case KVM_CAP_MIPS_64BIT:
3031 		/* We support 64-bit registers/operations and addresses */
3032 		r = 2;
3033 		break;
3034 #endif
3035 	case KVM_CAP_IOEVENTFD:
3036 		r = 1;
3037 		break;
3038 	default:
3039 		r = 0;
3040 		break;
3041 	}
3042 
3043 	return r;
3044 }
3045 
3046 static int kvm_vz_vcpu_init(struct kvm_vcpu *vcpu)
3047 {
3048 	int i;
3049 
3050 	for_each_possible_cpu(i)
3051 		vcpu->arch.vzguestid[i] = 0;
3052 
3053 	return 0;
3054 }
3055 
3056 static void kvm_vz_vcpu_uninit(struct kvm_vcpu *vcpu)
3057 {
3058 	int cpu;
3059 
3060 	/*
3061 	 * If the VCPU is freed and reused as another VCPU, we don't want the
3062 	 * matching pointer wrongly hanging around in last_vcpu[] or
3063 	 * last_exec_vcpu[].
3064 	 */
3065 	for_each_possible_cpu(cpu) {
3066 		if (last_vcpu[cpu] == vcpu)
3067 			last_vcpu[cpu] = NULL;
3068 		if (last_exec_vcpu[cpu] == vcpu)
3069 			last_exec_vcpu[cpu] = NULL;
3070 	}
3071 }
3072 
3073 static int kvm_vz_vcpu_setup(struct kvm_vcpu *vcpu)
3074 {
3075 	struct mips_coproc *cop0 = vcpu->arch.cop0;
3076 	unsigned long count_hz = 100*1000*1000; /* default to 100 MHz */
3077 
3078 	/*
3079 	 * Start off the timer at the same frequency as the host timer, but the
3080 	 * soft timer doesn't handle frequencies greater than 1GHz yet.
3081 	 */
3082 	if (mips_hpt_frequency && mips_hpt_frequency <= NSEC_PER_SEC)
3083 		count_hz = mips_hpt_frequency;
3084 	kvm_mips_init_count(vcpu, count_hz);
3085 
3086 	/*
3087 	 * Initialize guest register state to valid architectural reset state.
3088 	 */
3089 
3090 	/* PageGrain */
3091 	if (cpu_has_mips_r5 || cpu_has_mips_r6)
3092 		kvm_write_sw_gc0_pagegrain(cop0, PG_RIE | PG_XIE | PG_IEC);
3093 	/* Wired */
3094 	if (cpu_has_mips_r6)
3095 		kvm_write_sw_gc0_wired(cop0,
3096 				       read_gc0_wired() & MIPSR6_WIRED_LIMIT);
3097 	/* Status */
3098 	kvm_write_sw_gc0_status(cop0, ST0_BEV | ST0_ERL);
3099 	if (cpu_has_mips_r5 || cpu_has_mips_r6)
3100 		kvm_change_sw_gc0_status(cop0, ST0_FR, read_gc0_status());
3101 	/* IntCtl */
3102 	kvm_write_sw_gc0_intctl(cop0, read_gc0_intctl() &
3103 				(INTCTLF_IPFDC | INTCTLF_IPPCI | INTCTLF_IPTI));
3104 	/* PRId */
3105 	kvm_write_sw_gc0_prid(cop0, boot_cpu_data.processor_id);
3106 	/* EBase */
3107 	kvm_write_sw_gc0_ebase(cop0, (s32)0x80000000 | vcpu->vcpu_id);
3108 	/* Config */
3109 	kvm_save_gc0_config(cop0);
3110 	/* architecturally writable (e.g. from guest) */
3111 	kvm_change_sw_gc0_config(cop0, CONF_CM_CMASK,
3112 				 _page_cachable_default >> _CACHE_SHIFT);
3113 	/* architecturally read only, but maybe writable from root */
3114 	kvm_change_sw_gc0_config(cop0, MIPS_CONF_MT, read_c0_config());
3115 	if (cpu_guest_has_conf1) {
3116 		kvm_set_sw_gc0_config(cop0, MIPS_CONF_M);
3117 		/* Config1 */
3118 		kvm_save_gc0_config1(cop0);
3119 		/* architecturally read only, but maybe writable from root */
3120 		kvm_clear_sw_gc0_config1(cop0, MIPS_CONF1_C2	|
3121 					       MIPS_CONF1_MD	|
3122 					       MIPS_CONF1_PC	|
3123 					       MIPS_CONF1_WR	|
3124 					       MIPS_CONF1_CA	|
3125 					       MIPS_CONF1_FP);
3126 	}
3127 	if (cpu_guest_has_conf2) {
3128 		kvm_set_sw_gc0_config1(cop0, MIPS_CONF_M);
3129 		/* Config2 */
3130 		kvm_save_gc0_config2(cop0);
3131 	}
3132 	if (cpu_guest_has_conf3) {
3133 		kvm_set_sw_gc0_config2(cop0, MIPS_CONF_M);
3134 		/* Config3 */
3135 		kvm_save_gc0_config3(cop0);
3136 		/* architecturally writable (e.g. from guest) */
3137 		kvm_clear_sw_gc0_config3(cop0, MIPS_CONF3_ISA_OE);
3138 		/* architecturally read only, but maybe writable from root */
3139 		kvm_clear_sw_gc0_config3(cop0, MIPS_CONF3_MSA	|
3140 					       MIPS_CONF3_BPG	|
3141 					       MIPS_CONF3_ULRI	|
3142 					       MIPS_CONF3_DSP	|
3143 					       MIPS_CONF3_CTXTC	|
3144 					       MIPS_CONF3_ITL	|
3145 					       MIPS_CONF3_LPA	|
3146 					       MIPS_CONF3_VEIC	|
3147 					       MIPS_CONF3_VINT	|
3148 					       MIPS_CONF3_SP	|
3149 					       MIPS_CONF3_CDMM	|
3150 					       MIPS_CONF3_MT	|
3151 					       MIPS_CONF3_SM	|
3152 					       MIPS_CONF3_TL);
3153 	}
3154 	if (cpu_guest_has_conf4) {
3155 		kvm_set_sw_gc0_config3(cop0, MIPS_CONF_M);
3156 		/* Config4 */
3157 		kvm_save_gc0_config4(cop0);
3158 	}
3159 	if (cpu_guest_has_conf5) {
3160 		kvm_set_sw_gc0_config4(cop0, MIPS_CONF_M);
3161 		/* Config5 */
3162 		kvm_save_gc0_config5(cop0);
3163 		/* architecturally writable (e.g. from guest) */
3164 		kvm_clear_sw_gc0_config5(cop0, MIPS_CONF5_K	|
3165 					       MIPS_CONF5_CV	|
3166 					       MIPS_CONF5_MSAEN	|
3167 					       MIPS_CONF5_UFE	|
3168 					       MIPS_CONF5_FRE	|
3169 					       MIPS_CONF5_SBRI	|
3170 					       MIPS_CONF5_UFR);
3171 		/* architecturally read only, but maybe writable from root */
3172 		kvm_clear_sw_gc0_config5(cop0, MIPS_CONF5_MRP);
3173 	}
3174 
3175 	if (cpu_guest_has_contextconfig) {
3176 		/* ContextConfig */
3177 		kvm_write_sw_gc0_contextconfig(cop0, 0x007ffff0);
3178 #ifdef CONFIG_64BIT
3179 		/* XContextConfig */
3180 		/* bits SEGBITS-13+3:4 set */
3181 		kvm_write_sw_gc0_xcontextconfig(cop0,
3182 					((1ull << (cpu_vmbits - 13)) - 1) << 4);
3183 #endif
3184 	}
3185 
3186 	/* Implementation dependent, use the legacy layout */
3187 	if (cpu_guest_has_segments) {
3188 		/* SegCtl0, SegCtl1, SegCtl2 */
3189 		kvm_write_sw_gc0_segctl0(cop0, 0x00200010);
3190 		kvm_write_sw_gc0_segctl1(cop0, 0x00000002 |
3191 				(_page_cachable_default >> _CACHE_SHIFT) <<
3192 						(16 + MIPS_SEGCFG_C_SHIFT));
3193 		kvm_write_sw_gc0_segctl2(cop0, 0x00380438);
3194 	}
3195 
3196 	/* reset HTW registers */
3197 	if (cpu_guest_has_htw && (cpu_has_mips_r5 || cpu_has_mips_r6)) {
3198 		/* PWField */
3199 		kvm_write_sw_gc0_pwfield(cop0, 0x0c30c302);
3200 		/* PWSize */
3201 		kvm_write_sw_gc0_pwsize(cop0, 1 << MIPS_PWSIZE_PTW_SHIFT);
3202 	}
3203 
3204 	/* start with no pending virtual guest interrupts */
3205 	if (cpu_has_guestctl2)
3206 		cop0->reg[MIPS_CP0_GUESTCTL2][MIPS_CP0_GUESTCTL2_SEL] = 0;
3207 
3208 	/* Put PC at reset vector */
3209 	vcpu->arch.pc = CKSEG1ADDR(0x1fc00000);
3210 
3211 	return 0;
3212 }
3213 
3214 static void kvm_vz_flush_shadow_all(struct kvm *kvm)
3215 {
3216 	if (cpu_has_guestid) {
3217 		/* Flush GuestID for each VCPU individually */
3218 		kvm_flush_remote_tlbs(kvm);
3219 	} else {
3220 		/*
3221 		 * For each CPU there is a single GPA ASID used by all VCPUs in
3222 		 * the VM, so it doesn't make sense for the VCPUs to handle
3223 		 * invalidation of these ASIDs individually.
3224 		 *
3225 		 * Instead mark all CPUs as needing ASID invalidation in
3226 		 * asid_flush_mask, and just use kvm_flush_remote_tlbs(kvm) to
3227 		 * kick any running VCPUs so they check asid_flush_mask.
3228 		 */
3229 		cpumask_setall(&kvm->arch.asid_flush_mask);
3230 		kvm_flush_remote_tlbs(kvm);
3231 	}
3232 }
3233 
3234 static void kvm_vz_flush_shadow_memslot(struct kvm *kvm,
3235 					const struct kvm_memory_slot *slot)
3236 {
3237 	kvm_vz_flush_shadow_all(kvm);
3238 }
3239 
3240 static void kvm_vz_vcpu_reenter(struct kvm_vcpu *vcpu)
3241 {
3242 	int cpu = smp_processor_id();
3243 	int preserve_guest_tlb;
3244 
3245 	preserve_guest_tlb = kvm_vz_check_requests(vcpu, cpu);
3246 
3247 	if (preserve_guest_tlb)
3248 		kvm_vz_vcpu_save_wired(vcpu);
3249 
3250 	kvm_vz_vcpu_load_tlb(vcpu, cpu);
3251 
3252 	if (preserve_guest_tlb)
3253 		kvm_vz_vcpu_load_wired(vcpu);
3254 }
3255 
3256 static int kvm_vz_vcpu_run(struct kvm_vcpu *vcpu)
3257 {
3258 	int cpu = smp_processor_id();
3259 	int r;
3260 
3261 	kvm_vz_acquire_htimer(vcpu);
3262 	/* Check if we have any exceptions/interrupts pending */
3263 	kvm_mips_deliver_interrupts(vcpu, read_gc0_cause());
3264 
3265 	kvm_vz_check_requests(vcpu, cpu);
3266 	kvm_vz_vcpu_load_tlb(vcpu, cpu);
3267 	kvm_vz_vcpu_load_wired(vcpu);
3268 
3269 	r = vcpu->arch.vcpu_run(vcpu->run, vcpu);
3270 
3271 	kvm_vz_vcpu_save_wired(vcpu);
3272 
3273 	return r;
3274 }
3275 
3276 static struct kvm_mips_callbacks kvm_vz_callbacks = {
3277 	.handle_cop_unusable = kvm_trap_vz_handle_cop_unusable,
3278 	.handle_tlb_mod = kvm_trap_vz_handle_tlb_st_miss,
3279 	.handle_tlb_ld_miss = kvm_trap_vz_handle_tlb_ld_miss,
3280 	.handle_tlb_st_miss = kvm_trap_vz_handle_tlb_st_miss,
3281 	.handle_addr_err_st = kvm_trap_vz_no_handler,
3282 	.handle_addr_err_ld = kvm_trap_vz_no_handler,
3283 	.handle_syscall = kvm_trap_vz_no_handler,
3284 	.handle_res_inst = kvm_trap_vz_no_handler,
3285 	.handle_break = kvm_trap_vz_no_handler,
3286 	.handle_msa_disabled = kvm_trap_vz_handle_msa_disabled,
3287 	.handle_guest_exit = kvm_trap_vz_handle_guest_exit,
3288 
3289 	.hardware_enable = kvm_vz_hardware_enable,
3290 	.hardware_disable = kvm_vz_hardware_disable,
3291 	.check_extension = kvm_vz_check_extension,
3292 	.vcpu_init = kvm_vz_vcpu_init,
3293 	.vcpu_uninit = kvm_vz_vcpu_uninit,
3294 	.vcpu_setup = kvm_vz_vcpu_setup,
3295 	.flush_shadow_all = kvm_vz_flush_shadow_all,
3296 	.flush_shadow_memslot = kvm_vz_flush_shadow_memslot,
3297 	.gva_to_gpa = kvm_vz_gva_to_gpa_cb,
3298 	.queue_timer_int = kvm_vz_queue_timer_int_cb,
3299 	.dequeue_timer_int = kvm_vz_dequeue_timer_int_cb,
3300 	.queue_io_int = kvm_vz_queue_io_int_cb,
3301 	.dequeue_io_int = kvm_vz_dequeue_io_int_cb,
3302 	.irq_deliver = kvm_vz_irq_deliver_cb,
3303 	.irq_clear = kvm_vz_irq_clear_cb,
3304 	.num_regs = kvm_vz_num_regs,
3305 	.copy_reg_indices = kvm_vz_copy_reg_indices,
3306 	.get_one_reg = kvm_vz_get_one_reg,
3307 	.set_one_reg = kvm_vz_set_one_reg,
3308 	.vcpu_load = kvm_vz_vcpu_load,
3309 	.vcpu_put = kvm_vz_vcpu_put,
3310 	.vcpu_run = kvm_vz_vcpu_run,
3311 	.vcpu_reenter = kvm_vz_vcpu_reenter,
3312 };
3313 
3314 int kvm_mips_emulation_init(struct kvm_mips_callbacks **install_callbacks)
3315 {
3316 	if (!cpu_has_vz)
3317 		return -ENODEV;
3318 
3319 	/*
3320 	 * VZ requires at least 2 KScratch registers, so it should have been
3321 	 * possible to allocate pgd_reg.
3322 	 */
3323 	if (WARN(pgd_reg == -1,
3324 		 "pgd_reg not allocated even though cpu_has_vz\n"))
3325 		return -ENODEV;
3326 
3327 	pr_info("Starting KVM with MIPS VZ extensions\n");
3328 
3329 	*install_callbacks = &kvm_vz_callbacks;
3330 	return 0;
3331 }
3332