xref: /openbmc/linux/arch/mips/kvm/vz.c (revision c0891ac1)
1 /*
2  * This file is subject to the terms and conditions of the GNU General Public
3  * License.  See the file "COPYING" in the main directory of this archive
4  * for more details.
5  *
6  * KVM/MIPS: Support for hardware virtualization extensions
7  *
8  * Copyright (C) 2012  MIPS Technologies, Inc.  All rights reserved.
9  * Authors: Yann Le Du <ledu@kymasys.com>
10  */
11 
12 #include <linux/errno.h>
13 #include <linux/err.h>
14 #include <linux/module.h>
15 #include <linux/preempt.h>
16 #include <linux/vmalloc.h>
17 #include <asm/cacheflush.h>
18 #include <asm/cacheops.h>
19 #include <asm/cmpxchg.h>
20 #include <asm/fpu.h>
21 #include <asm/hazards.h>
22 #include <asm/inst.h>
23 #include <asm/mmu_context.h>
24 #include <asm/r4kcache.h>
25 #include <asm/time.h>
26 #include <asm/tlb.h>
27 #include <asm/tlbex.h>
28 
29 #include <linux/kvm_host.h>
30 
31 #include "interrupt.h"
32 #ifdef CONFIG_CPU_LOONGSON64
33 #include "loongson_regs.h"
34 #endif
35 
36 #include "trace.h"
37 
38 /* Pointers to last VCPU loaded on each physical CPU */
39 static struct kvm_vcpu *last_vcpu[NR_CPUS];
40 /* Pointers to last VCPU executed on each physical CPU */
41 static struct kvm_vcpu *last_exec_vcpu[NR_CPUS];
42 
43 /*
44  * Number of guest VTLB entries to use, so we can catch inconsistency between
45  * CPUs.
46  */
47 static unsigned int kvm_vz_guest_vtlb_size;
48 
49 static inline long kvm_vz_read_gc0_ebase(void)
50 {
51 	if (sizeof(long) == 8 && cpu_has_ebase_wg)
52 		return read_gc0_ebase_64();
53 	else
54 		return read_gc0_ebase();
55 }
56 
57 static inline void kvm_vz_write_gc0_ebase(long v)
58 {
59 	/*
60 	 * First write with WG=1 to write upper bits, then write again in case
61 	 * WG should be left at 0.
62 	 * write_gc0_ebase_64() is no longer UNDEFINED since R6.
63 	 */
64 	if (sizeof(long) == 8 &&
65 	    (cpu_has_mips64r6 || cpu_has_ebase_wg)) {
66 		write_gc0_ebase_64(v | MIPS_EBASE_WG);
67 		write_gc0_ebase_64(v);
68 	} else {
69 		write_gc0_ebase(v | MIPS_EBASE_WG);
70 		write_gc0_ebase(v);
71 	}
72 }
73 
74 /*
75  * These Config bits may be writable by the guest:
76  * Config:	[K23, KU] (!TLB), K0
77  * Config1:	(none)
78  * Config2:	[TU, SU] (impl)
79  * Config3:	ISAOnExc
80  * Config4:	FTLBPageSize
81  * Config5:	K, CV, MSAEn, UFE, FRE, SBRI, UFR
82  */
83 
84 static inline unsigned int kvm_vz_config_guest_wrmask(struct kvm_vcpu *vcpu)
85 {
86 	return CONF_CM_CMASK;
87 }
88 
89 static inline unsigned int kvm_vz_config1_guest_wrmask(struct kvm_vcpu *vcpu)
90 {
91 	return 0;
92 }
93 
94 static inline unsigned int kvm_vz_config2_guest_wrmask(struct kvm_vcpu *vcpu)
95 {
96 	return 0;
97 }
98 
99 static inline unsigned int kvm_vz_config3_guest_wrmask(struct kvm_vcpu *vcpu)
100 {
101 	return MIPS_CONF3_ISA_OE;
102 }
103 
104 static inline unsigned int kvm_vz_config4_guest_wrmask(struct kvm_vcpu *vcpu)
105 {
106 	/* no need to be exact */
107 	return MIPS_CONF4_VFTLBPAGESIZE;
108 }
109 
110 static inline unsigned int kvm_vz_config5_guest_wrmask(struct kvm_vcpu *vcpu)
111 {
112 	unsigned int mask = MIPS_CONF5_K | MIPS_CONF5_CV | MIPS_CONF5_SBRI;
113 
114 	/* Permit MSAEn changes if MSA supported and enabled */
115 	if (kvm_mips_guest_has_msa(&vcpu->arch))
116 		mask |= MIPS_CONF5_MSAEN;
117 
118 	/*
119 	 * Permit guest FPU mode changes if FPU is enabled and the relevant
120 	 * feature exists according to FIR register.
121 	 */
122 	if (kvm_mips_guest_has_fpu(&vcpu->arch)) {
123 		if (cpu_has_ufr)
124 			mask |= MIPS_CONF5_UFR;
125 		if (cpu_has_fre)
126 			mask |= MIPS_CONF5_FRE | MIPS_CONF5_UFE;
127 	}
128 
129 	return mask;
130 }
131 
132 static inline unsigned int kvm_vz_config6_guest_wrmask(struct kvm_vcpu *vcpu)
133 {
134 	return LOONGSON_CONF6_INTIMER | LOONGSON_CONF6_EXTIMER;
135 }
136 
137 /*
138  * VZ optionally allows these additional Config bits to be written by root:
139  * Config:	M, [MT]
140  * Config1:	M, [MMUSize-1, C2, MD, PC, WR, CA], FP
141  * Config2:	M
142  * Config3:	M, MSAP, [BPG], ULRI, [DSP2P, DSPP], CTXTC, [ITL, LPA, VEIC,
143  *		VInt, SP, CDMM, MT, SM, TL]
144  * Config4:	M, [VTLBSizeExt, MMUSizeExt]
145  * Config5:	MRP
146  */
147 
148 static inline unsigned int kvm_vz_config_user_wrmask(struct kvm_vcpu *vcpu)
149 {
150 	return kvm_vz_config_guest_wrmask(vcpu) | MIPS_CONF_M;
151 }
152 
153 static inline unsigned int kvm_vz_config1_user_wrmask(struct kvm_vcpu *vcpu)
154 {
155 	unsigned int mask = kvm_vz_config1_guest_wrmask(vcpu) | MIPS_CONF_M;
156 
157 	/* Permit FPU to be present if FPU is supported */
158 	if (kvm_mips_guest_can_have_fpu(&vcpu->arch))
159 		mask |= MIPS_CONF1_FP;
160 
161 	return mask;
162 }
163 
164 static inline unsigned int kvm_vz_config2_user_wrmask(struct kvm_vcpu *vcpu)
165 {
166 	return kvm_vz_config2_guest_wrmask(vcpu) | MIPS_CONF_M;
167 }
168 
169 static inline unsigned int kvm_vz_config3_user_wrmask(struct kvm_vcpu *vcpu)
170 {
171 	unsigned int mask = kvm_vz_config3_guest_wrmask(vcpu) | MIPS_CONF_M |
172 		MIPS_CONF3_ULRI | MIPS_CONF3_CTXTC;
173 
174 	/* Permit MSA to be present if MSA is supported */
175 	if (kvm_mips_guest_can_have_msa(&vcpu->arch))
176 		mask |= MIPS_CONF3_MSA;
177 
178 	return mask;
179 }
180 
181 static inline unsigned int kvm_vz_config4_user_wrmask(struct kvm_vcpu *vcpu)
182 {
183 	return kvm_vz_config4_guest_wrmask(vcpu) | MIPS_CONF_M;
184 }
185 
186 static inline unsigned int kvm_vz_config5_user_wrmask(struct kvm_vcpu *vcpu)
187 {
188 	return kvm_vz_config5_guest_wrmask(vcpu) | MIPS_CONF5_MRP;
189 }
190 
191 static inline unsigned int kvm_vz_config6_user_wrmask(struct kvm_vcpu *vcpu)
192 {
193 	return kvm_vz_config6_guest_wrmask(vcpu) |
194 		LOONGSON_CONF6_SFBEN | LOONGSON_CONF6_FTLBDIS;
195 }
196 
197 static gpa_t kvm_vz_gva_to_gpa_cb(gva_t gva)
198 {
199 	/* VZ guest has already converted gva to gpa */
200 	return gva;
201 }
202 
203 static void kvm_vz_queue_irq(struct kvm_vcpu *vcpu, unsigned int priority)
204 {
205 	set_bit(priority, &vcpu->arch.pending_exceptions);
206 	clear_bit(priority, &vcpu->arch.pending_exceptions_clr);
207 }
208 
209 static void kvm_vz_dequeue_irq(struct kvm_vcpu *vcpu, unsigned int priority)
210 {
211 	clear_bit(priority, &vcpu->arch.pending_exceptions);
212 	set_bit(priority, &vcpu->arch.pending_exceptions_clr);
213 }
214 
215 static void kvm_vz_queue_timer_int_cb(struct kvm_vcpu *vcpu)
216 {
217 	/*
218 	 * timer expiry is asynchronous to vcpu execution therefore defer guest
219 	 * cp0 accesses
220 	 */
221 	kvm_vz_queue_irq(vcpu, MIPS_EXC_INT_TIMER);
222 }
223 
224 static void kvm_vz_dequeue_timer_int_cb(struct kvm_vcpu *vcpu)
225 {
226 	/*
227 	 * timer expiry is asynchronous to vcpu execution therefore defer guest
228 	 * cp0 accesses
229 	 */
230 	kvm_vz_dequeue_irq(vcpu, MIPS_EXC_INT_TIMER);
231 }
232 
233 static void kvm_vz_queue_io_int_cb(struct kvm_vcpu *vcpu,
234 				   struct kvm_mips_interrupt *irq)
235 {
236 	int intr = (int)irq->irq;
237 
238 	/*
239 	 * interrupts are asynchronous to vcpu execution therefore defer guest
240 	 * cp0 accesses
241 	 */
242 	kvm_vz_queue_irq(vcpu, kvm_irq_to_priority(intr));
243 }
244 
245 static void kvm_vz_dequeue_io_int_cb(struct kvm_vcpu *vcpu,
246 				     struct kvm_mips_interrupt *irq)
247 {
248 	int intr = (int)irq->irq;
249 
250 	/*
251 	 * interrupts are asynchronous to vcpu execution therefore defer guest
252 	 * cp0 accesses
253 	 */
254 	kvm_vz_dequeue_irq(vcpu, kvm_irq_to_priority(-intr));
255 }
256 
257 static int kvm_vz_irq_deliver_cb(struct kvm_vcpu *vcpu, unsigned int priority,
258 				 u32 cause)
259 {
260 	u32 irq = (priority < MIPS_EXC_MAX) ?
261 		kvm_priority_to_irq[priority] : 0;
262 
263 	switch (priority) {
264 	case MIPS_EXC_INT_TIMER:
265 		set_gc0_cause(C_TI);
266 		break;
267 
268 	case MIPS_EXC_INT_IO_1:
269 	case MIPS_EXC_INT_IO_2:
270 	case MIPS_EXC_INT_IPI_1:
271 	case MIPS_EXC_INT_IPI_2:
272 		if (cpu_has_guestctl2)
273 			set_c0_guestctl2(irq);
274 		else
275 			set_gc0_cause(irq);
276 		break;
277 
278 	default:
279 		break;
280 	}
281 
282 	clear_bit(priority, &vcpu->arch.pending_exceptions);
283 	return 1;
284 }
285 
286 static int kvm_vz_irq_clear_cb(struct kvm_vcpu *vcpu, unsigned int priority,
287 			       u32 cause)
288 {
289 	u32 irq = (priority < MIPS_EXC_MAX) ?
290 		kvm_priority_to_irq[priority] : 0;
291 
292 	switch (priority) {
293 	case MIPS_EXC_INT_TIMER:
294 		/*
295 		 * Explicitly clear irq associated with Cause.IP[IPTI]
296 		 * if GuestCtl2 virtual interrupt register not
297 		 * supported or if not using GuestCtl2 Hardware Clear.
298 		 */
299 		if (cpu_has_guestctl2) {
300 			if (!(read_c0_guestctl2() & (irq << 14)))
301 				clear_c0_guestctl2(irq);
302 		} else {
303 			clear_gc0_cause(irq);
304 		}
305 		break;
306 
307 	case MIPS_EXC_INT_IO_1:
308 	case MIPS_EXC_INT_IO_2:
309 	case MIPS_EXC_INT_IPI_1:
310 	case MIPS_EXC_INT_IPI_2:
311 		/* Clear GuestCtl2.VIP irq if not using Hardware Clear */
312 		if (cpu_has_guestctl2) {
313 			if (!(read_c0_guestctl2() & (irq << 14)))
314 				clear_c0_guestctl2(irq);
315 		} else {
316 			clear_gc0_cause(irq);
317 		}
318 		break;
319 
320 	default:
321 		break;
322 	}
323 
324 	clear_bit(priority, &vcpu->arch.pending_exceptions_clr);
325 	return 1;
326 }
327 
328 /*
329  * VZ guest timer handling.
330  */
331 
332 /**
333  * kvm_vz_should_use_htimer() - Find whether to use the VZ hard guest timer.
334  * @vcpu:	Virtual CPU.
335  *
336  * Returns:	true if the VZ GTOffset & real guest CP0_Count should be used
337  *		instead of software emulation of guest timer.
338  *		false otherwise.
339  */
340 static bool kvm_vz_should_use_htimer(struct kvm_vcpu *vcpu)
341 {
342 	if (kvm_mips_count_disabled(vcpu))
343 		return false;
344 
345 	/* Chosen frequency must match real frequency */
346 	if (mips_hpt_frequency != vcpu->arch.count_hz)
347 		return false;
348 
349 	/* We don't support a CP0_GTOffset with fewer bits than CP0_Count */
350 	if (current_cpu_data.gtoffset_mask != 0xffffffff)
351 		return false;
352 
353 	return true;
354 }
355 
356 /**
357  * _kvm_vz_restore_stimer() - Restore soft timer state.
358  * @vcpu:	Virtual CPU.
359  * @compare:	CP0_Compare register value, restored by caller.
360  * @cause:	CP0_Cause register to restore.
361  *
362  * Restore VZ state relating to the soft timer. The hard timer can be enabled
363  * later.
364  */
365 static void _kvm_vz_restore_stimer(struct kvm_vcpu *vcpu, u32 compare,
366 				   u32 cause)
367 {
368 	/*
369 	 * Avoid spurious counter interrupts by setting Guest CP0_Count to just
370 	 * after Guest CP0_Compare.
371 	 */
372 	write_c0_gtoffset(compare - read_c0_count());
373 
374 	back_to_back_c0_hazard();
375 	write_gc0_cause(cause);
376 }
377 
378 /**
379  * _kvm_vz_restore_htimer() - Restore hard timer state.
380  * @vcpu:	Virtual CPU.
381  * @compare:	CP0_Compare register value, restored by caller.
382  * @cause:	CP0_Cause register to restore.
383  *
384  * Restore hard timer Guest.Count & Guest.Cause taking care to preserve the
385  * value of Guest.CP0_Cause.TI while restoring Guest.CP0_Cause.
386  */
387 static void _kvm_vz_restore_htimer(struct kvm_vcpu *vcpu,
388 				   u32 compare, u32 cause)
389 {
390 	u32 start_count, after_count;
391 	ktime_t freeze_time;
392 	unsigned long flags;
393 
394 	/*
395 	 * Freeze the soft-timer and sync the guest CP0_Count with it. We do
396 	 * this with interrupts disabled to avoid latency.
397 	 */
398 	local_irq_save(flags);
399 	freeze_time = kvm_mips_freeze_hrtimer(vcpu, &start_count);
400 	write_c0_gtoffset(start_count - read_c0_count());
401 	local_irq_restore(flags);
402 
403 	/* restore guest CP0_Cause, as TI may already be set */
404 	back_to_back_c0_hazard();
405 	write_gc0_cause(cause);
406 
407 	/*
408 	 * The above sequence isn't atomic and would result in lost timer
409 	 * interrupts if we're not careful. Detect if a timer interrupt is due
410 	 * and assert it.
411 	 */
412 	back_to_back_c0_hazard();
413 	after_count = read_gc0_count();
414 	if (after_count - start_count > compare - start_count - 1)
415 		kvm_vz_queue_irq(vcpu, MIPS_EXC_INT_TIMER);
416 }
417 
418 /**
419  * kvm_vz_restore_timer() - Restore timer state.
420  * @vcpu:	Virtual CPU.
421  *
422  * Restore soft timer state from saved context.
423  */
424 static void kvm_vz_restore_timer(struct kvm_vcpu *vcpu)
425 {
426 	struct mips_coproc *cop0 = vcpu->arch.cop0;
427 	u32 cause, compare;
428 
429 	compare = kvm_read_sw_gc0_compare(cop0);
430 	cause = kvm_read_sw_gc0_cause(cop0);
431 
432 	write_gc0_compare(compare);
433 	_kvm_vz_restore_stimer(vcpu, compare, cause);
434 }
435 
436 /**
437  * kvm_vz_acquire_htimer() - Switch to hard timer state.
438  * @vcpu:	Virtual CPU.
439  *
440  * Restore hard timer state on top of existing soft timer state if possible.
441  *
442  * Since hard timer won't remain active over preemption, preemption should be
443  * disabled by the caller.
444  */
445 void kvm_vz_acquire_htimer(struct kvm_vcpu *vcpu)
446 {
447 	u32 gctl0;
448 
449 	gctl0 = read_c0_guestctl0();
450 	if (!(gctl0 & MIPS_GCTL0_GT) && kvm_vz_should_use_htimer(vcpu)) {
451 		/* enable guest access to hard timer */
452 		write_c0_guestctl0(gctl0 | MIPS_GCTL0_GT);
453 
454 		_kvm_vz_restore_htimer(vcpu, read_gc0_compare(),
455 				       read_gc0_cause());
456 	}
457 }
458 
459 /**
460  * _kvm_vz_save_htimer() - Switch to software emulation of guest timer.
461  * @vcpu:	Virtual CPU.
462  * @compare:	Pointer to write compare value to.
463  * @cause:	Pointer to write cause value to.
464  *
465  * Save VZ guest timer state and switch to software emulation of guest CP0
466  * timer. The hard timer must already be in use, so preemption should be
467  * disabled.
468  */
469 static void _kvm_vz_save_htimer(struct kvm_vcpu *vcpu,
470 				u32 *out_compare, u32 *out_cause)
471 {
472 	u32 cause, compare, before_count, end_count;
473 	ktime_t before_time;
474 
475 	compare = read_gc0_compare();
476 	*out_compare = compare;
477 
478 	before_time = ktime_get();
479 
480 	/*
481 	 * Record the CP0_Count *prior* to saving CP0_Cause, so we have a time
482 	 * at which no pending timer interrupt is missing.
483 	 */
484 	before_count = read_gc0_count();
485 	back_to_back_c0_hazard();
486 	cause = read_gc0_cause();
487 	*out_cause = cause;
488 
489 	/*
490 	 * Record a final CP0_Count which we will transfer to the soft-timer.
491 	 * This is recorded *after* saving CP0_Cause, so we don't get any timer
492 	 * interrupts from just after the final CP0_Count point.
493 	 */
494 	back_to_back_c0_hazard();
495 	end_count = read_gc0_count();
496 
497 	/*
498 	 * The above sequence isn't atomic, so we could miss a timer interrupt
499 	 * between reading CP0_Cause and end_count. Detect and record any timer
500 	 * interrupt due between before_count and end_count.
501 	 */
502 	if (end_count - before_count > compare - before_count - 1)
503 		kvm_vz_queue_irq(vcpu, MIPS_EXC_INT_TIMER);
504 
505 	/*
506 	 * Restore soft-timer, ignoring a small amount of negative drift due to
507 	 * delay between freeze_hrtimer and setting CP0_GTOffset.
508 	 */
509 	kvm_mips_restore_hrtimer(vcpu, before_time, end_count, -0x10000);
510 }
511 
512 /**
513  * kvm_vz_save_timer() - Save guest timer state.
514  * @vcpu:	Virtual CPU.
515  *
516  * Save VZ guest timer state and switch to soft guest timer if hard timer was in
517  * use.
518  */
519 static void kvm_vz_save_timer(struct kvm_vcpu *vcpu)
520 {
521 	struct mips_coproc *cop0 = vcpu->arch.cop0;
522 	u32 gctl0, compare, cause;
523 
524 	gctl0 = read_c0_guestctl0();
525 	if (gctl0 & MIPS_GCTL0_GT) {
526 		/* disable guest use of hard timer */
527 		write_c0_guestctl0(gctl0 & ~MIPS_GCTL0_GT);
528 
529 		/* save hard timer state */
530 		_kvm_vz_save_htimer(vcpu, &compare, &cause);
531 	} else {
532 		compare = read_gc0_compare();
533 		cause = read_gc0_cause();
534 	}
535 
536 	/* save timer-related state to VCPU context */
537 	kvm_write_sw_gc0_cause(cop0, cause);
538 	kvm_write_sw_gc0_compare(cop0, compare);
539 }
540 
541 /**
542  * kvm_vz_lose_htimer() - Ensure hard guest timer is not in use.
543  * @vcpu:	Virtual CPU.
544  *
545  * Transfers the state of the hard guest timer to the soft guest timer, leaving
546  * guest state intact so it can continue to be used with the soft timer.
547  */
548 void kvm_vz_lose_htimer(struct kvm_vcpu *vcpu)
549 {
550 	u32 gctl0, compare, cause;
551 
552 	preempt_disable();
553 	gctl0 = read_c0_guestctl0();
554 	if (gctl0 & MIPS_GCTL0_GT) {
555 		/* disable guest use of timer */
556 		write_c0_guestctl0(gctl0 & ~MIPS_GCTL0_GT);
557 
558 		/* switch to soft timer */
559 		_kvm_vz_save_htimer(vcpu, &compare, &cause);
560 
561 		/* leave soft timer in usable state */
562 		_kvm_vz_restore_stimer(vcpu, compare, cause);
563 	}
564 	preempt_enable();
565 }
566 
567 /**
568  * is_eva_access() - Find whether an instruction is an EVA memory accessor.
569  * @inst:	32-bit instruction encoding.
570  *
571  * Finds whether @inst encodes an EVA memory access instruction, which would
572  * indicate that emulation of it should access the user mode address space
573  * instead of the kernel mode address space. This matters for MUSUK segments
574  * which are TLB mapped for user mode but unmapped for kernel mode.
575  *
576  * Returns:	Whether @inst encodes an EVA accessor instruction.
577  */
578 static bool is_eva_access(union mips_instruction inst)
579 {
580 	if (inst.spec3_format.opcode != spec3_op)
581 		return false;
582 
583 	switch (inst.spec3_format.func) {
584 	case lwle_op:
585 	case lwre_op:
586 	case cachee_op:
587 	case sbe_op:
588 	case she_op:
589 	case sce_op:
590 	case swe_op:
591 	case swle_op:
592 	case swre_op:
593 	case prefe_op:
594 	case lbue_op:
595 	case lhue_op:
596 	case lbe_op:
597 	case lhe_op:
598 	case lle_op:
599 	case lwe_op:
600 		return true;
601 	default:
602 		return false;
603 	}
604 }
605 
606 /**
607  * is_eva_am_mapped() - Find whether an access mode is mapped.
608  * @vcpu:	KVM VCPU state.
609  * @am:		3-bit encoded access mode.
610  * @eu:		Segment becomes unmapped and uncached when Status.ERL=1.
611  *
612  * Decode @am to find whether it encodes a mapped segment for the current VCPU
613  * state. Where necessary @eu and the actual instruction causing the fault are
614  * taken into account to make the decision.
615  *
616  * Returns:	Whether the VCPU faulted on a TLB mapped address.
617  */
618 static bool is_eva_am_mapped(struct kvm_vcpu *vcpu, unsigned int am, bool eu)
619 {
620 	u32 am_lookup;
621 	int err;
622 
623 	/*
624 	 * Interpret access control mode. We assume address errors will already
625 	 * have been caught by the guest, leaving us with:
626 	 *      AM      UM  SM  KM  31..24 23..16
627 	 * UK    0 000          Unm   0      0
628 	 * MK    1 001          TLB   1
629 	 * MSK   2 010      TLB TLB   1
630 	 * MUSK  3 011  TLB TLB TLB   1
631 	 * MUSUK 4 100  TLB TLB Unm   0      1
632 	 * USK   5 101      Unm Unm   0      0
633 	 * -     6 110                0      0
634 	 * UUSK  7 111  Unm Unm Unm   0      0
635 	 *
636 	 * We shift a magic value by AM across the sign bit to find if always
637 	 * TLB mapped, and if not shift by 8 again to find if it depends on KM.
638 	 */
639 	am_lookup = 0x70080000 << am;
640 	if ((s32)am_lookup < 0) {
641 		/*
642 		 * MK, MSK, MUSK
643 		 * Always TLB mapped, unless SegCtl.EU && ERL
644 		 */
645 		if (!eu || !(read_gc0_status() & ST0_ERL))
646 			return true;
647 	} else {
648 		am_lookup <<= 8;
649 		if ((s32)am_lookup < 0) {
650 			union mips_instruction inst;
651 			unsigned int status;
652 			u32 *opc;
653 
654 			/*
655 			 * MUSUK
656 			 * TLB mapped if not in kernel mode
657 			 */
658 			status = read_gc0_status();
659 			if (!(status & (ST0_EXL | ST0_ERL)) &&
660 			    (status & ST0_KSU))
661 				return true;
662 			/*
663 			 * EVA access instructions in kernel
664 			 * mode access user address space.
665 			 */
666 			opc = (u32 *)vcpu->arch.pc;
667 			if (vcpu->arch.host_cp0_cause & CAUSEF_BD)
668 				opc += 1;
669 			err = kvm_get_badinstr(opc, vcpu, &inst.word);
670 			if (!err && is_eva_access(inst))
671 				return true;
672 		}
673 	}
674 
675 	return false;
676 }
677 
678 /**
679  * kvm_vz_gva_to_gpa() - Convert valid GVA to GPA.
680  * @vcpu:	KVM VCPU state.
681  * @gva:	Guest virtual address to convert.
682  * @gpa:	Output guest physical address.
683  *
684  * Convert a guest virtual address (GVA) which is valid according to the guest
685  * context, to a guest physical address (GPA).
686  *
687  * Returns:	0 on success.
688  *		-errno on failure.
689  */
690 static int kvm_vz_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
691 			     unsigned long *gpa)
692 {
693 	u32 gva32 = gva;
694 	unsigned long segctl;
695 
696 	if ((long)gva == (s32)gva32) {
697 		/* Handle canonical 32-bit virtual address */
698 		if (cpu_guest_has_segments) {
699 			unsigned long mask, pa;
700 
701 			switch (gva32 >> 29) {
702 			case 0:
703 			case 1: /* CFG5 (1GB) */
704 				segctl = read_gc0_segctl2() >> 16;
705 				mask = (unsigned long)0xfc0000000ull;
706 				break;
707 			case 2:
708 			case 3: /* CFG4 (1GB) */
709 				segctl = read_gc0_segctl2();
710 				mask = (unsigned long)0xfc0000000ull;
711 				break;
712 			case 4: /* CFG3 (512MB) */
713 				segctl = read_gc0_segctl1() >> 16;
714 				mask = (unsigned long)0xfe0000000ull;
715 				break;
716 			case 5: /* CFG2 (512MB) */
717 				segctl = read_gc0_segctl1();
718 				mask = (unsigned long)0xfe0000000ull;
719 				break;
720 			case 6: /* CFG1 (512MB) */
721 				segctl = read_gc0_segctl0() >> 16;
722 				mask = (unsigned long)0xfe0000000ull;
723 				break;
724 			case 7: /* CFG0 (512MB) */
725 				segctl = read_gc0_segctl0();
726 				mask = (unsigned long)0xfe0000000ull;
727 				break;
728 			default:
729 				/*
730 				 * GCC 4.9 isn't smart enough to figure out that
731 				 * segctl and mask are always initialised.
732 				 */
733 				unreachable();
734 			}
735 
736 			if (is_eva_am_mapped(vcpu, (segctl >> 4) & 0x7,
737 					     segctl & 0x0008))
738 				goto tlb_mapped;
739 
740 			/* Unmapped, find guest physical address */
741 			pa = (segctl << 20) & mask;
742 			pa |= gva32 & ~mask;
743 			*gpa = pa;
744 			return 0;
745 		} else if ((s32)gva32 < (s32)0xc0000000) {
746 			/* legacy unmapped KSeg0 or KSeg1 */
747 			*gpa = gva32 & 0x1fffffff;
748 			return 0;
749 		}
750 #ifdef CONFIG_64BIT
751 	} else if ((gva & 0xc000000000000000) == 0x8000000000000000) {
752 		/* XKPHYS */
753 		if (cpu_guest_has_segments) {
754 			/*
755 			 * Each of the 8 regions can be overridden by SegCtl2.XR
756 			 * to use SegCtl1.XAM.
757 			 */
758 			segctl = read_gc0_segctl2();
759 			if (segctl & (1ull << (56 + ((gva >> 59) & 0x7)))) {
760 				segctl = read_gc0_segctl1();
761 				if (is_eva_am_mapped(vcpu, (segctl >> 59) & 0x7,
762 						     0))
763 					goto tlb_mapped;
764 			}
765 
766 		}
767 		/*
768 		 * Traditionally fully unmapped.
769 		 * Bits 61:59 specify the CCA, which we can just mask off here.
770 		 * Bits 58:PABITS should be zero, but we shouldn't have got here
771 		 * if it wasn't.
772 		 */
773 		*gpa = gva & 0x07ffffffffffffff;
774 		return 0;
775 #endif
776 	}
777 
778 tlb_mapped:
779 	return kvm_vz_guest_tlb_lookup(vcpu, gva, gpa);
780 }
781 
782 /**
783  * kvm_vz_badvaddr_to_gpa() - Convert GVA BadVAddr from root exception to GPA.
784  * @vcpu:	KVM VCPU state.
785  * @badvaddr:	Root BadVAddr.
786  * @gpa:	Output guest physical address.
787  *
788  * VZ implementations are permitted to report guest virtual addresses (GVA) in
789  * BadVAddr on a root exception during guest execution, instead of the more
790  * convenient guest physical addresses (GPA). When we get a GVA, this function
791  * converts it to a GPA, taking into account guest segmentation and guest TLB
792  * state.
793  *
794  * Returns:	0 on success.
795  *		-errno on failure.
796  */
797 static int kvm_vz_badvaddr_to_gpa(struct kvm_vcpu *vcpu, unsigned long badvaddr,
798 				  unsigned long *gpa)
799 {
800 	unsigned int gexccode = (vcpu->arch.host_cp0_guestctl0 &
801 				 MIPS_GCTL0_GEXC) >> MIPS_GCTL0_GEXC_SHIFT;
802 
803 	/* If BadVAddr is GPA, then all is well in the world */
804 	if (likely(gexccode == MIPS_GCTL0_GEXC_GPA)) {
805 		*gpa = badvaddr;
806 		return 0;
807 	}
808 
809 	/* Otherwise we'd expect it to be GVA ... */
810 	if (WARN(gexccode != MIPS_GCTL0_GEXC_GVA,
811 		 "Unexpected gexccode %#x\n", gexccode))
812 		return -EINVAL;
813 
814 	/* ... and we need to perform the GVA->GPA translation in software */
815 	return kvm_vz_gva_to_gpa(vcpu, badvaddr, gpa);
816 }
817 
818 static int kvm_trap_vz_no_handler(struct kvm_vcpu *vcpu)
819 {
820 	u32 *opc = (u32 *) vcpu->arch.pc;
821 	u32 cause = vcpu->arch.host_cp0_cause;
822 	u32 exccode = (cause & CAUSEF_EXCCODE) >> CAUSEB_EXCCODE;
823 	unsigned long badvaddr = vcpu->arch.host_cp0_badvaddr;
824 	u32 inst = 0;
825 
826 	/*
827 	 *  Fetch the instruction.
828 	 */
829 	if (cause & CAUSEF_BD)
830 		opc += 1;
831 	kvm_get_badinstr(opc, vcpu, &inst);
832 
833 	kvm_err("Exception Code: %d not handled @ PC: %p, inst: 0x%08x BadVaddr: %#lx Status: %#x\n",
834 		exccode, opc, inst, badvaddr,
835 		read_gc0_status());
836 	kvm_arch_vcpu_dump_regs(vcpu);
837 	vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
838 	return RESUME_HOST;
839 }
840 
841 static unsigned long mips_process_maar(unsigned int op, unsigned long val)
842 {
843 	/* Mask off unused bits */
844 	unsigned long mask = 0xfffff000 | MIPS_MAAR_S | MIPS_MAAR_VL;
845 
846 	if (read_gc0_pagegrain() & PG_ELPA)
847 		mask |= 0x00ffffff00000000ull;
848 	if (cpu_guest_has_mvh)
849 		mask |= MIPS_MAAR_VH;
850 
851 	/* Set or clear VH */
852 	if (op == mtc_op) {
853 		/* clear VH */
854 		val &= ~MIPS_MAAR_VH;
855 	} else if (op == dmtc_op) {
856 		/* set VH to match VL */
857 		val &= ~MIPS_MAAR_VH;
858 		if (val & MIPS_MAAR_VL)
859 			val |= MIPS_MAAR_VH;
860 	}
861 
862 	return val & mask;
863 }
864 
865 static void kvm_write_maari(struct kvm_vcpu *vcpu, unsigned long val)
866 {
867 	struct mips_coproc *cop0 = vcpu->arch.cop0;
868 
869 	val &= MIPS_MAARI_INDEX;
870 	if (val == MIPS_MAARI_INDEX)
871 		kvm_write_sw_gc0_maari(cop0, ARRAY_SIZE(vcpu->arch.maar) - 1);
872 	else if (val < ARRAY_SIZE(vcpu->arch.maar))
873 		kvm_write_sw_gc0_maari(cop0, val);
874 }
875 
876 static enum emulation_result kvm_vz_gpsi_cop0(union mips_instruction inst,
877 					      u32 *opc, u32 cause,
878 					      struct kvm_vcpu *vcpu)
879 {
880 	struct mips_coproc *cop0 = vcpu->arch.cop0;
881 	enum emulation_result er = EMULATE_DONE;
882 	u32 rt, rd, sel;
883 	unsigned long curr_pc;
884 	unsigned long val;
885 
886 	/*
887 	 * Update PC and hold onto current PC in case there is
888 	 * an error and we want to rollback the PC
889 	 */
890 	curr_pc = vcpu->arch.pc;
891 	er = update_pc(vcpu, cause);
892 	if (er == EMULATE_FAIL)
893 		return er;
894 
895 	if (inst.co_format.co) {
896 		switch (inst.co_format.func) {
897 		case wait_op:
898 			er = kvm_mips_emul_wait(vcpu);
899 			break;
900 		default:
901 			er = EMULATE_FAIL;
902 		}
903 	} else {
904 		rt = inst.c0r_format.rt;
905 		rd = inst.c0r_format.rd;
906 		sel = inst.c0r_format.sel;
907 
908 		switch (inst.c0r_format.rs) {
909 		case dmfc_op:
910 		case mfc_op:
911 #ifdef CONFIG_KVM_MIPS_DEBUG_COP0_COUNTERS
912 			cop0->stat[rd][sel]++;
913 #endif
914 			if (rd == MIPS_CP0_COUNT &&
915 			    sel == 0) {			/* Count */
916 				val = kvm_mips_read_count(vcpu);
917 			} else if (rd == MIPS_CP0_COMPARE &&
918 				   sel == 0) {		/* Compare */
919 				val = read_gc0_compare();
920 			} else if (rd == MIPS_CP0_LLADDR &&
921 				   sel == 0) {		/* LLAddr */
922 				if (cpu_guest_has_rw_llb)
923 					val = read_gc0_lladdr() &
924 						MIPS_LLADDR_LLB;
925 				else
926 					val = 0;
927 			} else if (rd == MIPS_CP0_LLADDR &&
928 				   sel == 1 &&		/* MAAR */
929 				   cpu_guest_has_maar &&
930 				   !cpu_guest_has_dyn_maar) {
931 				/* MAARI must be in range */
932 				BUG_ON(kvm_read_sw_gc0_maari(cop0) >=
933 						ARRAY_SIZE(vcpu->arch.maar));
934 				val = vcpu->arch.maar[
935 					kvm_read_sw_gc0_maari(cop0)];
936 			} else if ((rd == MIPS_CP0_PRID &&
937 				    (sel == 0 ||	/* PRid */
938 				     sel == 2 ||	/* CDMMBase */
939 				     sel == 3)) ||	/* CMGCRBase */
940 				   (rd == MIPS_CP0_STATUS &&
941 				    (sel == 2 ||	/* SRSCtl */
942 				     sel == 3)) ||	/* SRSMap */
943 				   (rd == MIPS_CP0_CONFIG &&
944 				    (sel == 6 ||	/* Config6 */
945 				     sel == 7)) ||	/* Config7 */
946 				   (rd == MIPS_CP0_LLADDR &&
947 				    (sel == 2) &&	/* MAARI */
948 				    cpu_guest_has_maar &&
949 				    !cpu_guest_has_dyn_maar) ||
950 				   (rd == MIPS_CP0_ERRCTL &&
951 				    (sel == 0))) {	/* ErrCtl */
952 				val = cop0->reg[rd][sel];
953 #ifdef CONFIG_CPU_LOONGSON64
954 			} else if (rd == MIPS_CP0_DIAG &&
955 				   (sel == 0)) {	/* Diag */
956 				val = cop0->reg[rd][sel];
957 #endif
958 			} else {
959 				val = 0;
960 				er = EMULATE_FAIL;
961 			}
962 
963 			if (er != EMULATE_FAIL) {
964 				/* Sign extend */
965 				if (inst.c0r_format.rs == mfc_op)
966 					val = (int)val;
967 				vcpu->arch.gprs[rt] = val;
968 			}
969 
970 			trace_kvm_hwr(vcpu, (inst.c0r_format.rs == mfc_op) ?
971 					KVM_TRACE_MFC0 : KVM_TRACE_DMFC0,
972 				      KVM_TRACE_COP0(rd, sel), val);
973 			break;
974 
975 		case dmtc_op:
976 		case mtc_op:
977 #ifdef CONFIG_KVM_MIPS_DEBUG_COP0_COUNTERS
978 			cop0->stat[rd][sel]++;
979 #endif
980 			val = vcpu->arch.gprs[rt];
981 			trace_kvm_hwr(vcpu, (inst.c0r_format.rs == mtc_op) ?
982 					KVM_TRACE_MTC0 : KVM_TRACE_DMTC0,
983 				      KVM_TRACE_COP0(rd, sel), val);
984 
985 			if (rd == MIPS_CP0_COUNT &&
986 			    sel == 0) {			/* Count */
987 				kvm_vz_lose_htimer(vcpu);
988 				kvm_mips_write_count(vcpu, vcpu->arch.gprs[rt]);
989 			} else if (rd == MIPS_CP0_COMPARE &&
990 				   sel == 0) {		/* Compare */
991 				kvm_mips_write_compare(vcpu,
992 						       vcpu->arch.gprs[rt],
993 						       true);
994 			} else if (rd == MIPS_CP0_LLADDR &&
995 				   sel == 0) {		/* LLAddr */
996 				/*
997 				 * P5600 generates GPSI on guest MTC0 LLAddr.
998 				 * Only allow the guest to clear LLB.
999 				 */
1000 				if (cpu_guest_has_rw_llb &&
1001 				    !(val & MIPS_LLADDR_LLB))
1002 					write_gc0_lladdr(0);
1003 			} else if (rd == MIPS_CP0_LLADDR &&
1004 				   sel == 1 &&		/* MAAR */
1005 				   cpu_guest_has_maar &&
1006 				   !cpu_guest_has_dyn_maar) {
1007 				val = mips_process_maar(inst.c0r_format.rs,
1008 							val);
1009 
1010 				/* MAARI must be in range */
1011 				BUG_ON(kvm_read_sw_gc0_maari(cop0) >=
1012 						ARRAY_SIZE(vcpu->arch.maar));
1013 				vcpu->arch.maar[kvm_read_sw_gc0_maari(cop0)] =
1014 									val;
1015 			} else if (rd == MIPS_CP0_LLADDR &&
1016 				   (sel == 2) &&	/* MAARI */
1017 				   cpu_guest_has_maar &&
1018 				   !cpu_guest_has_dyn_maar) {
1019 				kvm_write_maari(vcpu, val);
1020 			} else if (rd == MIPS_CP0_CONFIG &&
1021 				   (sel == 6)) {
1022 				cop0->reg[rd][sel] = (int)val;
1023 			} else if (rd == MIPS_CP0_ERRCTL &&
1024 				   (sel == 0)) {	/* ErrCtl */
1025 				/* ignore the written value */
1026 #ifdef CONFIG_CPU_LOONGSON64
1027 			} else if (rd == MIPS_CP0_DIAG &&
1028 				   (sel == 0)) {	/* Diag */
1029 				unsigned long flags;
1030 
1031 				local_irq_save(flags);
1032 				if (val & LOONGSON_DIAG_BTB) {
1033 					/* Flush BTB */
1034 					set_c0_diag(LOONGSON_DIAG_BTB);
1035 				}
1036 				if (val & LOONGSON_DIAG_ITLB) {
1037 					/* Flush ITLB */
1038 					set_c0_diag(LOONGSON_DIAG_ITLB);
1039 				}
1040 				if (val & LOONGSON_DIAG_DTLB) {
1041 					/* Flush DTLB */
1042 					set_c0_diag(LOONGSON_DIAG_DTLB);
1043 				}
1044 				if (val & LOONGSON_DIAG_VTLB) {
1045 					/* Flush VTLB */
1046 					kvm_loongson_clear_guest_vtlb();
1047 				}
1048 				if (val & LOONGSON_DIAG_FTLB) {
1049 					/* Flush FTLB */
1050 					kvm_loongson_clear_guest_ftlb();
1051 				}
1052 				local_irq_restore(flags);
1053 #endif
1054 			} else {
1055 				er = EMULATE_FAIL;
1056 			}
1057 			break;
1058 
1059 		default:
1060 			er = EMULATE_FAIL;
1061 			break;
1062 		}
1063 	}
1064 	/* Rollback PC only if emulation was unsuccessful */
1065 	if (er == EMULATE_FAIL) {
1066 		kvm_err("[%#lx]%s: unsupported cop0 instruction 0x%08x\n",
1067 			curr_pc, __func__, inst.word);
1068 
1069 		vcpu->arch.pc = curr_pc;
1070 	}
1071 
1072 	return er;
1073 }
1074 
1075 static enum emulation_result kvm_vz_gpsi_cache(union mips_instruction inst,
1076 					       u32 *opc, u32 cause,
1077 					       struct kvm_vcpu *vcpu)
1078 {
1079 	enum emulation_result er = EMULATE_DONE;
1080 	u32 cache, op_inst, op, base;
1081 	s16 offset;
1082 	struct kvm_vcpu_arch *arch = &vcpu->arch;
1083 	unsigned long va, curr_pc;
1084 
1085 	/*
1086 	 * Update PC and hold onto current PC in case there is
1087 	 * an error and we want to rollback the PC
1088 	 */
1089 	curr_pc = vcpu->arch.pc;
1090 	er = update_pc(vcpu, cause);
1091 	if (er == EMULATE_FAIL)
1092 		return er;
1093 
1094 	base = inst.i_format.rs;
1095 	op_inst = inst.i_format.rt;
1096 	if (cpu_has_mips_r6)
1097 		offset = inst.spec3_format.simmediate;
1098 	else
1099 		offset = inst.i_format.simmediate;
1100 	cache = op_inst & CacheOp_Cache;
1101 	op = op_inst & CacheOp_Op;
1102 
1103 	va = arch->gprs[base] + offset;
1104 
1105 	kvm_debug("CACHE (cache: %#x, op: %#x, base[%d]: %#lx, offset: %#x\n",
1106 		  cache, op, base, arch->gprs[base], offset);
1107 
1108 	/* Secondary or tirtiary cache ops ignored */
1109 	if (cache != Cache_I && cache != Cache_D)
1110 		return EMULATE_DONE;
1111 
1112 	switch (op_inst) {
1113 	case Index_Invalidate_I:
1114 		flush_icache_line_indexed(va);
1115 		return EMULATE_DONE;
1116 	case Index_Writeback_Inv_D:
1117 		flush_dcache_line_indexed(va);
1118 		return EMULATE_DONE;
1119 	case Hit_Invalidate_I:
1120 	case Hit_Invalidate_D:
1121 	case Hit_Writeback_Inv_D:
1122 		if (boot_cpu_type() == CPU_CAVIUM_OCTEON3) {
1123 			/* We can just flush entire icache */
1124 			local_flush_icache_range(0, 0);
1125 			return EMULATE_DONE;
1126 		}
1127 
1128 		/* So far, other platforms support guest hit cache ops */
1129 		break;
1130 	default:
1131 		break;
1132 	}
1133 
1134 	kvm_err("@ %#lx/%#lx CACHE (cache: %#x, op: %#x, base[%d]: %#lx, offset: %#x\n",
1135 		curr_pc, vcpu->arch.gprs[31], cache, op, base, arch->gprs[base],
1136 		offset);
1137 	/* Rollback PC */
1138 	vcpu->arch.pc = curr_pc;
1139 
1140 	return EMULATE_FAIL;
1141 }
1142 
1143 #ifdef CONFIG_CPU_LOONGSON64
1144 static enum emulation_result kvm_vz_gpsi_lwc2(union mips_instruction inst,
1145 					      u32 *opc, u32 cause,
1146 					      struct kvm_vcpu *vcpu)
1147 {
1148 	unsigned int rs, rd;
1149 	unsigned int hostcfg;
1150 	unsigned long curr_pc;
1151 	enum emulation_result er = EMULATE_DONE;
1152 
1153 	/*
1154 	 * Update PC and hold onto current PC in case there is
1155 	 * an error and we want to rollback the PC
1156 	 */
1157 	curr_pc = vcpu->arch.pc;
1158 	er = update_pc(vcpu, cause);
1159 	if (er == EMULATE_FAIL)
1160 		return er;
1161 
1162 	rs = inst.loongson3_lscsr_format.rs;
1163 	rd = inst.loongson3_lscsr_format.rd;
1164 	switch (inst.loongson3_lscsr_format.fr) {
1165 	case 0x8:  /* Read CPUCFG */
1166 		++vcpu->stat.vz_cpucfg_exits;
1167 		hostcfg = read_cpucfg(vcpu->arch.gprs[rs]);
1168 
1169 		switch (vcpu->arch.gprs[rs]) {
1170 		case LOONGSON_CFG0:
1171 			vcpu->arch.gprs[rd] = 0x14c000;
1172 			break;
1173 		case LOONGSON_CFG1:
1174 			hostcfg &= (LOONGSON_CFG1_FP | LOONGSON_CFG1_MMI |
1175 				    LOONGSON_CFG1_MSA1 | LOONGSON_CFG1_MSA2 |
1176 				    LOONGSON_CFG1_SFBP);
1177 			vcpu->arch.gprs[rd] = hostcfg;
1178 			break;
1179 		case LOONGSON_CFG2:
1180 			hostcfg &= (LOONGSON_CFG2_LEXT1 | LOONGSON_CFG2_LEXT2 |
1181 				    LOONGSON_CFG2_LEXT3 | LOONGSON_CFG2_LSPW);
1182 			vcpu->arch.gprs[rd] = hostcfg;
1183 			break;
1184 		case LOONGSON_CFG3:
1185 			vcpu->arch.gprs[rd] = hostcfg;
1186 			break;
1187 		default:
1188 			/* Don't export any other advanced features to guest */
1189 			vcpu->arch.gprs[rd] = 0;
1190 			break;
1191 		}
1192 		break;
1193 
1194 	default:
1195 		kvm_err("lwc2 emulate not impl %d rs %lx @%lx\n",
1196 			inst.loongson3_lscsr_format.fr, vcpu->arch.gprs[rs], curr_pc);
1197 		er = EMULATE_FAIL;
1198 		break;
1199 	}
1200 
1201 	/* Rollback PC only if emulation was unsuccessful */
1202 	if (er == EMULATE_FAIL) {
1203 		kvm_err("[%#lx]%s: unsupported lwc2 instruction 0x%08x 0x%08x\n",
1204 			curr_pc, __func__, inst.word, inst.loongson3_lscsr_format.fr);
1205 
1206 		vcpu->arch.pc = curr_pc;
1207 	}
1208 
1209 	return er;
1210 }
1211 #endif
1212 
1213 static enum emulation_result kvm_trap_vz_handle_gpsi(u32 cause, u32 *opc,
1214 						     struct kvm_vcpu *vcpu)
1215 {
1216 	enum emulation_result er = EMULATE_DONE;
1217 	struct kvm_vcpu_arch *arch = &vcpu->arch;
1218 	union mips_instruction inst;
1219 	int rd, rt, sel;
1220 	int err;
1221 
1222 	/*
1223 	 *  Fetch the instruction.
1224 	 */
1225 	if (cause & CAUSEF_BD)
1226 		opc += 1;
1227 	err = kvm_get_badinstr(opc, vcpu, &inst.word);
1228 	if (err)
1229 		return EMULATE_FAIL;
1230 
1231 	switch (inst.r_format.opcode) {
1232 	case cop0_op:
1233 		er = kvm_vz_gpsi_cop0(inst, opc, cause, vcpu);
1234 		break;
1235 #ifndef CONFIG_CPU_MIPSR6
1236 	case cache_op:
1237 		trace_kvm_exit(vcpu, KVM_TRACE_EXIT_CACHE);
1238 		er = kvm_vz_gpsi_cache(inst, opc, cause, vcpu);
1239 		break;
1240 #endif
1241 #ifdef CONFIG_CPU_LOONGSON64
1242 	case lwc2_op:
1243 		er = kvm_vz_gpsi_lwc2(inst, opc, cause, vcpu);
1244 		break;
1245 #endif
1246 	case spec3_op:
1247 		switch (inst.spec3_format.func) {
1248 #ifdef CONFIG_CPU_MIPSR6
1249 		case cache6_op:
1250 			trace_kvm_exit(vcpu, KVM_TRACE_EXIT_CACHE);
1251 			er = kvm_vz_gpsi_cache(inst, opc, cause, vcpu);
1252 			break;
1253 #endif
1254 		case rdhwr_op:
1255 			if (inst.r_format.rs || (inst.r_format.re >> 3))
1256 				goto unknown;
1257 
1258 			rd = inst.r_format.rd;
1259 			rt = inst.r_format.rt;
1260 			sel = inst.r_format.re & 0x7;
1261 
1262 			switch (rd) {
1263 			case MIPS_HWR_CC:	/* Read count register */
1264 				arch->gprs[rt] =
1265 					(long)(int)kvm_mips_read_count(vcpu);
1266 				break;
1267 			default:
1268 				trace_kvm_hwr(vcpu, KVM_TRACE_RDHWR,
1269 					      KVM_TRACE_HWR(rd, sel), 0);
1270 				goto unknown;
1271 			}
1272 
1273 			trace_kvm_hwr(vcpu, KVM_TRACE_RDHWR,
1274 				      KVM_TRACE_HWR(rd, sel), arch->gprs[rt]);
1275 
1276 			er = update_pc(vcpu, cause);
1277 			break;
1278 		default:
1279 			goto unknown;
1280 		}
1281 		break;
1282 unknown:
1283 
1284 	default:
1285 		kvm_err("GPSI exception not supported (%p/%#x)\n",
1286 				opc, inst.word);
1287 		kvm_arch_vcpu_dump_regs(vcpu);
1288 		er = EMULATE_FAIL;
1289 		break;
1290 	}
1291 
1292 	return er;
1293 }
1294 
1295 static enum emulation_result kvm_trap_vz_handle_gsfc(u32 cause, u32 *opc,
1296 						     struct kvm_vcpu *vcpu)
1297 {
1298 	enum emulation_result er = EMULATE_DONE;
1299 	struct kvm_vcpu_arch *arch = &vcpu->arch;
1300 	union mips_instruction inst;
1301 	int err;
1302 
1303 	/*
1304 	 *  Fetch the instruction.
1305 	 */
1306 	if (cause & CAUSEF_BD)
1307 		opc += 1;
1308 	err = kvm_get_badinstr(opc, vcpu, &inst.word);
1309 	if (err)
1310 		return EMULATE_FAIL;
1311 
1312 	/* complete MTC0 on behalf of guest and advance EPC */
1313 	if (inst.c0r_format.opcode == cop0_op &&
1314 	    inst.c0r_format.rs == mtc_op &&
1315 	    inst.c0r_format.z == 0) {
1316 		int rt = inst.c0r_format.rt;
1317 		int rd = inst.c0r_format.rd;
1318 		int sel = inst.c0r_format.sel;
1319 		unsigned int val = arch->gprs[rt];
1320 		unsigned int old_val, change;
1321 
1322 		trace_kvm_hwr(vcpu, KVM_TRACE_MTC0, KVM_TRACE_COP0(rd, sel),
1323 			      val);
1324 
1325 		if ((rd == MIPS_CP0_STATUS) && (sel == 0)) {
1326 			/* FR bit should read as zero if no FPU */
1327 			if (!kvm_mips_guest_has_fpu(&vcpu->arch))
1328 				val &= ~(ST0_CU1 | ST0_FR);
1329 
1330 			/*
1331 			 * Also don't allow FR to be set if host doesn't support
1332 			 * it.
1333 			 */
1334 			if (!(boot_cpu_data.fpu_id & MIPS_FPIR_F64))
1335 				val &= ~ST0_FR;
1336 
1337 			old_val = read_gc0_status();
1338 			change = val ^ old_val;
1339 
1340 			if (change & ST0_FR) {
1341 				/*
1342 				 * FPU and Vector register state is made
1343 				 * UNPREDICTABLE by a change of FR, so don't
1344 				 * even bother saving it.
1345 				 */
1346 				kvm_drop_fpu(vcpu);
1347 			}
1348 
1349 			/*
1350 			 * If MSA state is already live, it is undefined how it
1351 			 * interacts with FR=0 FPU state, and we don't want to
1352 			 * hit reserved instruction exceptions trying to save
1353 			 * the MSA state later when CU=1 && FR=1, so play it
1354 			 * safe and save it first.
1355 			 */
1356 			if (change & ST0_CU1 && !(val & ST0_FR) &&
1357 			    vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA)
1358 				kvm_lose_fpu(vcpu);
1359 
1360 			write_gc0_status(val);
1361 		} else if ((rd == MIPS_CP0_CAUSE) && (sel == 0)) {
1362 			u32 old_cause = read_gc0_cause();
1363 			u32 change = old_cause ^ val;
1364 
1365 			/* DC bit enabling/disabling timer? */
1366 			if (change & CAUSEF_DC) {
1367 				if (val & CAUSEF_DC) {
1368 					kvm_vz_lose_htimer(vcpu);
1369 					kvm_mips_count_disable_cause(vcpu);
1370 				} else {
1371 					kvm_mips_count_enable_cause(vcpu);
1372 				}
1373 			}
1374 
1375 			/* Only certain bits are RW to the guest */
1376 			change &= (CAUSEF_DC | CAUSEF_IV | CAUSEF_WP |
1377 				   CAUSEF_IP0 | CAUSEF_IP1);
1378 
1379 			/* WP can only be cleared */
1380 			change &= ~CAUSEF_WP | old_cause;
1381 
1382 			write_gc0_cause(old_cause ^ change);
1383 		} else if ((rd == MIPS_CP0_STATUS) && (sel == 1)) { /* IntCtl */
1384 			write_gc0_intctl(val);
1385 		} else if ((rd == MIPS_CP0_CONFIG) && (sel == 5)) {
1386 			old_val = read_gc0_config5();
1387 			change = val ^ old_val;
1388 			/* Handle changes in FPU/MSA modes */
1389 			preempt_disable();
1390 
1391 			/*
1392 			 * Propagate FRE changes immediately if the FPU
1393 			 * context is already loaded.
1394 			 */
1395 			if (change & MIPS_CONF5_FRE &&
1396 			    vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU)
1397 				change_c0_config5(MIPS_CONF5_FRE, val);
1398 
1399 			preempt_enable();
1400 
1401 			val = old_val ^
1402 				(change & kvm_vz_config5_guest_wrmask(vcpu));
1403 			write_gc0_config5(val);
1404 		} else {
1405 			kvm_err("Handle GSFC, unsupported field change @ %p: %#x\n",
1406 			    opc, inst.word);
1407 			er = EMULATE_FAIL;
1408 		}
1409 
1410 		if (er != EMULATE_FAIL)
1411 			er = update_pc(vcpu, cause);
1412 	} else {
1413 		kvm_err("Handle GSFC, unrecognized instruction @ %p: %#x\n",
1414 			opc, inst.word);
1415 		er = EMULATE_FAIL;
1416 	}
1417 
1418 	return er;
1419 }
1420 
1421 static enum emulation_result kvm_trap_vz_handle_ghfc(u32 cause, u32 *opc,
1422 						     struct kvm_vcpu *vcpu)
1423 {
1424 	/*
1425 	 * Presumably this is due to MC (guest mode change), so lets trace some
1426 	 * relevant info.
1427 	 */
1428 	trace_kvm_guest_mode_change(vcpu);
1429 
1430 	return EMULATE_DONE;
1431 }
1432 
1433 static enum emulation_result kvm_trap_vz_handle_hc(u32 cause, u32 *opc,
1434 						   struct kvm_vcpu *vcpu)
1435 {
1436 	enum emulation_result er;
1437 	union mips_instruction inst;
1438 	unsigned long curr_pc;
1439 	int err;
1440 
1441 	if (cause & CAUSEF_BD)
1442 		opc += 1;
1443 	err = kvm_get_badinstr(opc, vcpu, &inst.word);
1444 	if (err)
1445 		return EMULATE_FAIL;
1446 
1447 	/*
1448 	 * Update PC and hold onto current PC in case there is
1449 	 * an error and we want to rollback the PC
1450 	 */
1451 	curr_pc = vcpu->arch.pc;
1452 	er = update_pc(vcpu, cause);
1453 	if (er == EMULATE_FAIL)
1454 		return er;
1455 
1456 	er = kvm_mips_emul_hypcall(vcpu, inst);
1457 	if (er == EMULATE_FAIL)
1458 		vcpu->arch.pc = curr_pc;
1459 
1460 	return er;
1461 }
1462 
1463 static enum emulation_result kvm_trap_vz_no_handler_guest_exit(u32 gexccode,
1464 							u32 cause,
1465 							u32 *opc,
1466 							struct kvm_vcpu *vcpu)
1467 {
1468 	u32 inst;
1469 
1470 	/*
1471 	 *  Fetch the instruction.
1472 	 */
1473 	if (cause & CAUSEF_BD)
1474 		opc += 1;
1475 	kvm_get_badinstr(opc, vcpu, &inst);
1476 
1477 	kvm_err("Guest Exception Code: %d not yet handled @ PC: %p, inst: 0x%08x  Status: %#x\n",
1478 		gexccode, opc, inst, read_gc0_status());
1479 
1480 	return EMULATE_FAIL;
1481 }
1482 
1483 static int kvm_trap_vz_handle_guest_exit(struct kvm_vcpu *vcpu)
1484 {
1485 	u32 *opc = (u32 *) vcpu->arch.pc;
1486 	u32 cause = vcpu->arch.host_cp0_cause;
1487 	enum emulation_result er = EMULATE_DONE;
1488 	u32 gexccode = (vcpu->arch.host_cp0_guestctl0 &
1489 			MIPS_GCTL0_GEXC) >> MIPS_GCTL0_GEXC_SHIFT;
1490 	int ret = RESUME_GUEST;
1491 
1492 	trace_kvm_exit(vcpu, KVM_TRACE_EXIT_GEXCCODE_BASE + gexccode);
1493 	switch (gexccode) {
1494 	case MIPS_GCTL0_GEXC_GPSI:
1495 		++vcpu->stat.vz_gpsi_exits;
1496 		er = kvm_trap_vz_handle_gpsi(cause, opc, vcpu);
1497 		break;
1498 	case MIPS_GCTL0_GEXC_GSFC:
1499 		++vcpu->stat.vz_gsfc_exits;
1500 		er = kvm_trap_vz_handle_gsfc(cause, opc, vcpu);
1501 		break;
1502 	case MIPS_GCTL0_GEXC_HC:
1503 		++vcpu->stat.vz_hc_exits;
1504 		er = kvm_trap_vz_handle_hc(cause, opc, vcpu);
1505 		break;
1506 	case MIPS_GCTL0_GEXC_GRR:
1507 		++vcpu->stat.vz_grr_exits;
1508 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1509 						       vcpu);
1510 		break;
1511 	case MIPS_GCTL0_GEXC_GVA:
1512 		++vcpu->stat.vz_gva_exits;
1513 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1514 						       vcpu);
1515 		break;
1516 	case MIPS_GCTL0_GEXC_GHFC:
1517 		++vcpu->stat.vz_ghfc_exits;
1518 		er = kvm_trap_vz_handle_ghfc(cause, opc, vcpu);
1519 		break;
1520 	case MIPS_GCTL0_GEXC_GPA:
1521 		++vcpu->stat.vz_gpa_exits;
1522 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1523 						       vcpu);
1524 		break;
1525 	default:
1526 		++vcpu->stat.vz_resvd_exits;
1527 		er = kvm_trap_vz_no_handler_guest_exit(gexccode, cause, opc,
1528 						       vcpu);
1529 		break;
1530 
1531 	}
1532 
1533 	if (er == EMULATE_DONE) {
1534 		ret = RESUME_GUEST;
1535 	} else if (er == EMULATE_HYPERCALL) {
1536 		ret = kvm_mips_handle_hypcall(vcpu);
1537 	} else {
1538 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1539 		ret = RESUME_HOST;
1540 	}
1541 	return ret;
1542 }
1543 
1544 /**
1545  * kvm_trap_vz_handle_cop_unusuable() - Guest used unusable coprocessor.
1546  * @vcpu:	Virtual CPU context.
1547  *
1548  * Handle when the guest attempts to use a coprocessor which hasn't been allowed
1549  * by the root context.
1550  */
1551 static int kvm_trap_vz_handle_cop_unusable(struct kvm_vcpu *vcpu)
1552 {
1553 	u32 cause = vcpu->arch.host_cp0_cause;
1554 	enum emulation_result er = EMULATE_FAIL;
1555 	int ret = RESUME_GUEST;
1556 
1557 	if (((cause & CAUSEF_CE) >> CAUSEB_CE) == 1) {
1558 		/*
1559 		 * If guest FPU not present, the FPU operation should have been
1560 		 * treated as a reserved instruction!
1561 		 * If FPU already in use, we shouldn't get this at all.
1562 		 */
1563 		if (WARN_ON(!kvm_mips_guest_has_fpu(&vcpu->arch) ||
1564 			    vcpu->arch.aux_inuse & KVM_MIPS_AUX_FPU)) {
1565 			preempt_enable();
1566 			return EMULATE_FAIL;
1567 		}
1568 
1569 		kvm_own_fpu(vcpu);
1570 		er = EMULATE_DONE;
1571 	}
1572 	/* other coprocessors not handled */
1573 
1574 	switch (er) {
1575 	case EMULATE_DONE:
1576 		ret = RESUME_GUEST;
1577 		break;
1578 
1579 	case EMULATE_FAIL:
1580 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1581 		ret = RESUME_HOST;
1582 		break;
1583 
1584 	default:
1585 		BUG();
1586 	}
1587 	return ret;
1588 }
1589 
1590 /**
1591  * kvm_trap_vz_handle_msa_disabled() - Guest used MSA while disabled in root.
1592  * @vcpu:	Virtual CPU context.
1593  *
1594  * Handle when the guest attempts to use MSA when it is disabled in the root
1595  * context.
1596  */
1597 static int kvm_trap_vz_handle_msa_disabled(struct kvm_vcpu *vcpu)
1598 {
1599 	/*
1600 	 * If MSA not present or not exposed to guest or FR=0, the MSA operation
1601 	 * should have been treated as a reserved instruction!
1602 	 * Same if CU1=1, FR=0.
1603 	 * If MSA already in use, we shouldn't get this at all.
1604 	 */
1605 	if (!kvm_mips_guest_has_msa(&vcpu->arch) ||
1606 	    (read_gc0_status() & (ST0_CU1 | ST0_FR)) == ST0_CU1 ||
1607 	    !(read_gc0_config5() & MIPS_CONF5_MSAEN) ||
1608 	    vcpu->arch.aux_inuse & KVM_MIPS_AUX_MSA) {
1609 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1610 		return RESUME_HOST;
1611 	}
1612 
1613 	kvm_own_msa(vcpu);
1614 
1615 	return RESUME_GUEST;
1616 }
1617 
1618 static int kvm_trap_vz_handle_tlb_ld_miss(struct kvm_vcpu *vcpu)
1619 {
1620 	struct kvm_run *run = vcpu->run;
1621 	u32 *opc = (u32 *) vcpu->arch.pc;
1622 	u32 cause = vcpu->arch.host_cp0_cause;
1623 	ulong badvaddr = vcpu->arch.host_cp0_badvaddr;
1624 	union mips_instruction inst;
1625 	enum emulation_result er = EMULATE_DONE;
1626 	int err, ret = RESUME_GUEST;
1627 
1628 	if (kvm_mips_handle_vz_root_tlb_fault(badvaddr, vcpu, false)) {
1629 		/* A code fetch fault doesn't count as an MMIO */
1630 		if (kvm_is_ifetch_fault(&vcpu->arch)) {
1631 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1632 			return RESUME_HOST;
1633 		}
1634 
1635 		/* Fetch the instruction */
1636 		if (cause & CAUSEF_BD)
1637 			opc += 1;
1638 		err = kvm_get_badinstr(opc, vcpu, &inst.word);
1639 		if (err) {
1640 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1641 			return RESUME_HOST;
1642 		}
1643 
1644 		/* Treat as MMIO */
1645 		er = kvm_mips_emulate_load(inst, cause, vcpu);
1646 		if (er == EMULATE_FAIL) {
1647 			kvm_err("Guest Emulate Load from MMIO space failed: PC: %p, BadVaddr: %#lx\n",
1648 				opc, badvaddr);
1649 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1650 		}
1651 	}
1652 
1653 	if (er == EMULATE_DONE) {
1654 		ret = RESUME_GUEST;
1655 	} else if (er == EMULATE_DO_MMIO) {
1656 		run->exit_reason = KVM_EXIT_MMIO;
1657 		ret = RESUME_HOST;
1658 	} else {
1659 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1660 		ret = RESUME_HOST;
1661 	}
1662 	return ret;
1663 }
1664 
1665 static int kvm_trap_vz_handle_tlb_st_miss(struct kvm_vcpu *vcpu)
1666 {
1667 	struct kvm_run *run = vcpu->run;
1668 	u32 *opc = (u32 *) vcpu->arch.pc;
1669 	u32 cause = vcpu->arch.host_cp0_cause;
1670 	ulong badvaddr = vcpu->arch.host_cp0_badvaddr;
1671 	union mips_instruction inst;
1672 	enum emulation_result er = EMULATE_DONE;
1673 	int err;
1674 	int ret = RESUME_GUEST;
1675 
1676 	/* Just try the access again if we couldn't do the translation */
1677 	if (kvm_vz_badvaddr_to_gpa(vcpu, badvaddr, &badvaddr))
1678 		return RESUME_GUEST;
1679 	vcpu->arch.host_cp0_badvaddr = badvaddr;
1680 
1681 	if (kvm_mips_handle_vz_root_tlb_fault(badvaddr, vcpu, true)) {
1682 		/* Fetch the instruction */
1683 		if (cause & CAUSEF_BD)
1684 			opc += 1;
1685 		err = kvm_get_badinstr(opc, vcpu, &inst.word);
1686 		if (err) {
1687 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1688 			return RESUME_HOST;
1689 		}
1690 
1691 		/* Treat as MMIO */
1692 		er = kvm_mips_emulate_store(inst, cause, vcpu);
1693 		if (er == EMULATE_FAIL) {
1694 			kvm_err("Guest Emulate Store to MMIO space failed: PC: %p, BadVaddr: %#lx\n",
1695 				opc, badvaddr);
1696 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1697 		}
1698 	}
1699 
1700 	if (er == EMULATE_DONE) {
1701 		ret = RESUME_GUEST;
1702 	} else if (er == EMULATE_DO_MMIO) {
1703 		run->exit_reason = KVM_EXIT_MMIO;
1704 		ret = RESUME_HOST;
1705 	} else {
1706 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1707 		ret = RESUME_HOST;
1708 	}
1709 	return ret;
1710 }
1711 
1712 static u64 kvm_vz_get_one_regs[] = {
1713 	KVM_REG_MIPS_CP0_INDEX,
1714 	KVM_REG_MIPS_CP0_ENTRYLO0,
1715 	KVM_REG_MIPS_CP0_ENTRYLO1,
1716 	KVM_REG_MIPS_CP0_CONTEXT,
1717 	KVM_REG_MIPS_CP0_PAGEMASK,
1718 	KVM_REG_MIPS_CP0_PAGEGRAIN,
1719 	KVM_REG_MIPS_CP0_WIRED,
1720 	KVM_REG_MIPS_CP0_HWRENA,
1721 	KVM_REG_MIPS_CP0_BADVADDR,
1722 	KVM_REG_MIPS_CP0_COUNT,
1723 	KVM_REG_MIPS_CP0_ENTRYHI,
1724 	KVM_REG_MIPS_CP0_COMPARE,
1725 	KVM_REG_MIPS_CP0_STATUS,
1726 	KVM_REG_MIPS_CP0_INTCTL,
1727 	KVM_REG_MIPS_CP0_CAUSE,
1728 	KVM_REG_MIPS_CP0_EPC,
1729 	KVM_REG_MIPS_CP0_PRID,
1730 	KVM_REG_MIPS_CP0_EBASE,
1731 	KVM_REG_MIPS_CP0_CONFIG,
1732 	KVM_REG_MIPS_CP0_CONFIG1,
1733 	KVM_REG_MIPS_CP0_CONFIG2,
1734 	KVM_REG_MIPS_CP0_CONFIG3,
1735 	KVM_REG_MIPS_CP0_CONFIG4,
1736 	KVM_REG_MIPS_CP0_CONFIG5,
1737 	KVM_REG_MIPS_CP0_CONFIG6,
1738 #ifdef CONFIG_64BIT
1739 	KVM_REG_MIPS_CP0_XCONTEXT,
1740 #endif
1741 	KVM_REG_MIPS_CP0_ERROREPC,
1742 
1743 	KVM_REG_MIPS_COUNT_CTL,
1744 	KVM_REG_MIPS_COUNT_RESUME,
1745 	KVM_REG_MIPS_COUNT_HZ,
1746 };
1747 
1748 static u64 kvm_vz_get_one_regs_contextconfig[] = {
1749 	KVM_REG_MIPS_CP0_CONTEXTCONFIG,
1750 #ifdef CONFIG_64BIT
1751 	KVM_REG_MIPS_CP0_XCONTEXTCONFIG,
1752 #endif
1753 };
1754 
1755 static u64 kvm_vz_get_one_regs_segments[] = {
1756 	KVM_REG_MIPS_CP0_SEGCTL0,
1757 	KVM_REG_MIPS_CP0_SEGCTL1,
1758 	KVM_REG_MIPS_CP0_SEGCTL2,
1759 };
1760 
1761 static u64 kvm_vz_get_one_regs_htw[] = {
1762 	KVM_REG_MIPS_CP0_PWBASE,
1763 	KVM_REG_MIPS_CP0_PWFIELD,
1764 	KVM_REG_MIPS_CP0_PWSIZE,
1765 	KVM_REG_MIPS_CP0_PWCTL,
1766 };
1767 
1768 static u64 kvm_vz_get_one_regs_kscratch[] = {
1769 	KVM_REG_MIPS_CP0_KSCRATCH1,
1770 	KVM_REG_MIPS_CP0_KSCRATCH2,
1771 	KVM_REG_MIPS_CP0_KSCRATCH3,
1772 	KVM_REG_MIPS_CP0_KSCRATCH4,
1773 	KVM_REG_MIPS_CP0_KSCRATCH5,
1774 	KVM_REG_MIPS_CP0_KSCRATCH6,
1775 };
1776 
1777 static unsigned long kvm_vz_num_regs(struct kvm_vcpu *vcpu)
1778 {
1779 	unsigned long ret;
1780 
1781 	ret = ARRAY_SIZE(kvm_vz_get_one_regs);
1782 	if (cpu_guest_has_userlocal)
1783 		++ret;
1784 	if (cpu_guest_has_badinstr)
1785 		++ret;
1786 	if (cpu_guest_has_badinstrp)
1787 		++ret;
1788 	if (cpu_guest_has_contextconfig)
1789 		ret += ARRAY_SIZE(kvm_vz_get_one_regs_contextconfig);
1790 	if (cpu_guest_has_segments)
1791 		ret += ARRAY_SIZE(kvm_vz_get_one_regs_segments);
1792 	if (cpu_guest_has_htw || cpu_guest_has_ldpte)
1793 		ret += ARRAY_SIZE(kvm_vz_get_one_regs_htw);
1794 	if (cpu_guest_has_maar && !cpu_guest_has_dyn_maar)
1795 		ret += 1 + ARRAY_SIZE(vcpu->arch.maar);
1796 	ret += __arch_hweight8(cpu_data[0].guest.kscratch_mask);
1797 
1798 	return ret;
1799 }
1800 
1801 static int kvm_vz_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *indices)
1802 {
1803 	u64 index;
1804 	unsigned int i;
1805 
1806 	if (copy_to_user(indices, kvm_vz_get_one_regs,
1807 			 sizeof(kvm_vz_get_one_regs)))
1808 		return -EFAULT;
1809 	indices += ARRAY_SIZE(kvm_vz_get_one_regs);
1810 
1811 	if (cpu_guest_has_userlocal) {
1812 		index = KVM_REG_MIPS_CP0_USERLOCAL;
1813 		if (copy_to_user(indices, &index, sizeof(index)))
1814 			return -EFAULT;
1815 		++indices;
1816 	}
1817 	if (cpu_guest_has_badinstr) {
1818 		index = KVM_REG_MIPS_CP0_BADINSTR;
1819 		if (copy_to_user(indices, &index, sizeof(index)))
1820 			return -EFAULT;
1821 		++indices;
1822 	}
1823 	if (cpu_guest_has_badinstrp) {
1824 		index = KVM_REG_MIPS_CP0_BADINSTRP;
1825 		if (copy_to_user(indices, &index, sizeof(index)))
1826 			return -EFAULT;
1827 		++indices;
1828 	}
1829 	if (cpu_guest_has_contextconfig) {
1830 		if (copy_to_user(indices, kvm_vz_get_one_regs_contextconfig,
1831 				 sizeof(kvm_vz_get_one_regs_contextconfig)))
1832 			return -EFAULT;
1833 		indices += ARRAY_SIZE(kvm_vz_get_one_regs_contextconfig);
1834 	}
1835 	if (cpu_guest_has_segments) {
1836 		if (copy_to_user(indices, kvm_vz_get_one_regs_segments,
1837 				 sizeof(kvm_vz_get_one_regs_segments)))
1838 			return -EFAULT;
1839 		indices += ARRAY_SIZE(kvm_vz_get_one_regs_segments);
1840 	}
1841 	if (cpu_guest_has_htw || cpu_guest_has_ldpte) {
1842 		if (copy_to_user(indices, kvm_vz_get_one_regs_htw,
1843 				 sizeof(kvm_vz_get_one_regs_htw)))
1844 			return -EFAULT;
1845 		indices += ARRAY_SIZE(kvm_vz_get_one_regs_htw);
1846 	}
1847 	if (cpu_guest_has_maar && !cpu_guest_has_dyn_maar) {
1848 		for (i = 0; i < ARRAY_SIZE(vcpu->arch.maar); ++i) {
1849 			index = KVM_REG_MIPS_CP0_MAAR(i);
1850 			if (copy_to_user(indices, &index, sizeof(index)))
1851 				return -EFAULT;
1852 			++indices;
1853 		}
1854 
1855 		index = KVM_REG_MIPS_CP0_MAARI;
1856 		if (copy_to_user(indices, &index, sizeof(index)))
1857 			return -EFAULT;
1858 		++indices;
1859 	}
1860 	for (i = 0; i < 6; ++i) {
1861 		if (!cpu_guest_has_kscr(i + 2))
1862 			continue;
1863 
1864 		if (copy_to_user(indices, &kvm_vz_get_one_regs_kscratch[i],
1865 				 sizeof(kvm_vz_get_one_regs_kscratch[i])))
1866 			return -EFAULT;
1867 		++indices;
1868 	}
1869 
1870 	return 0;
1871 }
1872 
1873 static inline s64 entrylo_kvm_to_user(unsigned long v)
1874 {
1875 	s64 mask, ret = v;
1876 
1877 	if (BITS_PER_LONG == 32) {
1878 		/*
1879 		 * KVM API exposes 64-bit version of the register, so move the
1880 		 * RI/XI bits up into place.
1881 		 */
1882 		mask = MIPS_ENTRYLO_RI | MIPS_ENTRYLO_XI;
1883 		ret &= ~mask;
1884 		ret |= ((s64)v & mask) << 32;
1885 	}
1886 	return ret;
1887 }
1888 
1889 static inline unsigned long entrylo_user_to_kvm(s64 v)
1890 {
1891 	unsigned long mask, ret = v;
1892 
1893 	if (BITS_PER_LONG == 32) {
1894 		/*
1895 		 * KVM API exposes 64-bit versiono of the register, so move the
1896 		 * RI/XI bits down into place.
1897 		 */
1898 		mask = MIPS_ENTRYLO_RI | MIPS_ENTRYLO_XI;
1899 		ret &= ~mask;
1900 		ret |= (v >> 32) & mask;
1901 	}
1902 	return ret;
1903 }
1904 
1905 static int kvm_vz_get_one_reg(struct kvm_vcpu *vcpu,
1906 			      const struct kvm_one_reg *reg,
1907 			      s64 *v)
1908 {
1909 	struct mips_coproc *cop0 = vcpu->arch.cop0;
1910 	unsigned int idx;
1911 
1912 	switch (reg->id) {
1913 	case KVM_REG_MIPS_CP0_INDEX:
1914 		*v = (long)read_gc0_index();
1915 		break;
1916 	case KVM_REG_MIPS_CP0_ENTRYLO0:
1917 		*v = entrylo_kvm_to_user(read_gc0_entrylo0());
1918 		break;
1919 	case KVM_REG_MIPS_CP0_ENTRYLO1:
1920 		*v = entrylo_kvm_to_user(read_gc0_entrylo1());
1921 		break;
1922 	case KVM_REG_MIPS_CP0_CONTEXT:
1923 		*v = (long)read_gc0_context();
1924 		break;
1925 	case KVM_REG_MIPS_CP0_CONTEXTCONFIG:
1926 		if (!cpu_guest_has_contextconfig)
1927 			return -EINVAL;
1928 		*v = read_gc0_contextconfig();
1929 		break;
1930 	case KVM_REG_MIPS_CP0_USERLOCAL:
1931 		if (!cpu_guest_has_userlocal)
1932 			return -EINVAL;
1933 		*v = read_gc0_userlocal();
1934 		break;
1935 #ifdef CONFIG_64BIT
1936 	case KVM_REG_MIPS_CP0_XCONTEXTCONFIG:
1937 		if (!cpu_guest_has_contextconfig)
1938 			return -EINVAL;
1939 		*v = read_gc0_xcontextconfig();
1940 		break;
1941 #endif
1942 	case KVM_REG_MIPS_CP0_PAGEMASK:
1943 		*v = (long)read_gc0_pagemask();
1944 		break;
1945 	case KVM_REG_MIPS_CP0_PAGEGRAIN:
1946 		*v = (long)read_gc0_pagegrain();
1947 		break;
1948 	case KVM_REG_MIPS_CP0_SEGCTL0:
1949 		if (!cpu_guest_has_segments)
1950 			return -EINVAL;
1951 		*v = read_gc0_segctl0();
1952 		break;
1953 	case KVM_REG_MIPS_CP0_SEGCTL1:
1954 		if (!cpu_guest_has_segments)
1955 			return -EINVAL;
1956 		*v = read_gc0_segctl1();
1957 		break;
1958 	case KVM_REG_MIPS_CP0_SEGCTL2:
1959 		if (!cpu_guest_has_segments)
1960 			return -EINVAL;
1961 		*v = read_gc0_segctl2();
1962 		break;
1963 	case KVM_REG_MIPS_CP0_PWBASE:
1964 		if (!cpu_guest_has_htw && !cpu_guest_has_ldpte)
1965 			return -EINVAL;
1966 		*v = read_gc0_pwbase();
1967 		break;
1968 	case KVM_REG_MIPS_CP0_PWFIELD:
1969 		if (!cpu_guest_has_htw && !cpu_guest_has_ldpte)
1970 			return -EINVAL;
1971 		*v = read_gc0_pwfield();
1972 		break;
1973 	case KVM_REG_MIPS_CP0_PWSIZE:
1974 		if (!cpu_guest_has_htw && !cpu_guest_has_ldpte)
1975 			return -EINVAL;
1976 		*v = read_gc0_pwsize();
1977 		break;
1978 	case KVM_REG_MIPS_CP0_WIRED:
1979 		*v = (long)read_gc0_wired();
1980 		break;
1981 	case KVM_REG_MIPS_CP0_PWCTL:
1982 		if (!cpu_guest_has_htw && !cpu_guest_has_ldpte)
1983 			return -EINVAL;
1984 		*v = read_gc0_pwctl();
1985 		break;
1986 	case KVM_REG_MIPS_CP0_HWRENA:
1987 		*v = (long)read_gc0_hwrena();
1988 		break;
1989 	case KVM_REG_MIPS_CP0_BADVADDR:
1990 		*v = (long)read_gc0_badvaddr();
1991 		break;
1992 	case KVM_REG_MIPS_CP0_BADINSTR:
1993 		if (!cpu_guest_has_badinstr)
1994 			return -EINVAL;
1995 		*v = read_gc0_badinstr();
1996 		break;
1997 	case KVM_REG_MIPS_CP0_BADINSTRP:
1998 		if (!cpu_guest_has_badinstrp)
1999 			return -EINVAL;
2000 		*v = read_gc0_badinstrp();
2001 		break;
2002 	case KVM_REG_MIPS_CP0_COUNT:
2003 		*v = kvm_mips_read_count(vcpu);
2004 		break;
2005 	case KVM_REG_MIPS_CP0_ENTRYHI:
2006 		*v = (long)read_gc0_entryhi();
2007 		break;
2008 	case KVM_REG_MIPS_CP0_COMPARE:
2009 		*v = (long)read_gc0_compare();
2010 		break;
2011 	case KVM_REG_MIPS_CP0_STATUS:
2012 		*v = (long)read_gc0_status();
2013 		break;
2014 	case KVM_REG_MIPS_CP0_INTCTL:
2015 		*v = read_gc0_intctl();
2016 		break;
2017 	case KVM_REG_MIPS_CP0_CAUSE:
2018 		*v = (long)read_gc0_cause();
2019 		break;
2020 	case KVM_REG_MIPS_CP0_EPC:
2021 		*v = (long)read_gc0_epc();
2022 		break;
2023 	case KVM_REG_MIPS_CP0_PRID:
2024 		switch (boot_cpu_type()) {
2025 		case CPU_CAVIUM_OCTEON3:
2026 			/* Octeon III has a read-only guest.PRid */
2027 			*v = read_gc0_prid();
2028 			break;
2029 		default:
2030 			*v = (long)kvm_read_c0_guest_prid(cop0);
2031 			break;
2032 		}
2033 		break;
2034 	case KVM_REG_MIPS_CP0_EBASE:
2035 		*v = kvm_vz_read_gc0_ebase();
2036 		break;
2037 	case KVM_REG_MIPS_CP0_CONFIG:
2038 		*v = read_gc0_config();
2039 		break;
2040 	case KVM_REG_MIPS_CP0_CONFIG1:
2041 		if (!cpu_guest_has_conf1)
2042 			return -EINVAL;
2043 		*v = read_gc0_config1();
2044 		break;
2045 	case KVM_REG_MIPS_CP0_CONFIG2:
2046 		if (!cpu_guest_has_conf2)
2047 			return -EINVAL;
2048 		*v = read_gc0_config2();
2049 		break;
2050 	case KVM_REG_MIPS_CP0_CONFIG3:
2051 		if (!cpu_guest_has_conf3)
2052 			return -EINVAL;
2053 		*v = read_gc0_config3();
2054 		break;
2055 	case KVM_REG_MIPS_CP0_CONFIG4:
2056 		if (!cpu_guest_has_conf4)
2057 			return -EINVAL;
2058 		*v = read_gc0_config4();
2059 		break;
2060 	case KVM_REG_MIPS_CP0_CONFIG5:
2061 		if (!cpu_guest_has_conf5)
2062 			return -EINVAL;
2063 		*v = read_gc0_config5();
2064 		break;
2065 	case KVM_REG_MIPS_CP0_CONFIG6:
2066 		*v = kvm_read_sw_gc0_config6(cop0);
2067 		break;
2068 	case KVM_REG_MIPS_CP0_MAAR(0) ... KVM_REG_MIPS_CP0_MAAR(0x3f):
2069 		if (!cpu_guest_has_maar || cpu_guest_has_dyn_maar)
2070 			return -EINVAL;
2071 		idx = reg->id - KVM_REG_MIPS_CP0_MAAR(0);
2072 		if (idx >= ARRAY_SIZE(vcpu->arch.maar))
2073 			return -EINVAL;
2074 		*v = vcpu->arch.maar[idx];
2075 		break;
2076 	case KVM_REG_MIPS_CP0_MAARI:
2077 		if (!cpu_guest_has_maar || cpu_guest_has_dyn_maar)
2078 			return -EINVAL;
2079 		*v = kvm_read_sw_gc0_maari(vcpu->arch.cop0);
2080 		break;
2081 #ifdef CONFIG_64BIT
2082 	case KVM_REG_MIPS_CP0_XCONTEXT:
2083 		*v = read_gc0_xcontext();
2084 		break;
2085 #endif
2086 	case KVM_REG_MIPS_CP0_ERROREPC:
2087 		*v = (long)read_gc0_errorepc();
2088 		break;
2089 	case KVM_REG_MIPS_CP0_KSCRATCH1 ... KVM_REG_MIPS_CP0_KSCRATCH6:
2090 		idx = reg->id - KVM_REG_MIPS_CP0_KSCRATCH1 + 2;
2091 		if (!cpu_guest_has_kscr(idx))
2092 			return -EINVAL;
2093 		switch (idx) {
2094 		case 2:
2095 			*v = (long)read_gc0_kscratch1();
2096 			break;
2097 		case 3:
2098 			*v = (long)read_gc0_kscratch2();
2099 			break;
2100 		case 4:
2101 			*v = (long)read_gc0_kscratch3();
2102 			break;
2103 		case 5:
2104 			*v = (long)read_gc0_kscratch4();
2105 			break;
2106 		case 6:
2107 			*v = (long)read_gc0_kscratch5();
2108 			break;
2109 		case 7:
2110 			*v = (long)read_gc0_kscratch6();
2111 			break;
2112 		}
2113 		break;
2114 	case KVM_REG_MIPS_COUNT_CTL:
2115 		*v = vcpu->arch.count_ctl;
2116 		break;
2117 	case KVM_REG_MIPS_COUNT_RESUME:
2118 		*v = ktime_to_ns(vcpu->arch.count_resume);
2119 		break;
2120 	case KVM_REG_MIPS_COUNT_HZ:
2121 		*v = vcpu->arch.count_hz;
2122 		break;
2123 	default:
2124 		return -EINVAL;
2125 	}
2126 	return 0;
2127 }
2128 
2129 static int kvm_vz_set_one_reg(struct kvm_vcpu *vcpu,
2130 			      const struct kvm_one_reg *reg,
2131 			      s64 v)
2132 {
2133 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2134 	unsigned int idx;
2135 	int ret = 0;
2136 	unsigned int cur, change;
2137 
2138 	switch (reg->id) {
2139 	case KVM_REG_MIPS_CP0_INDEX:
2140 		write_gc0_index(v);
2141 		break;
2142 	case KVM_REG_MIPS_CP0_ENTRYLO0:
2143 		write_gc0_entrylo0(entrylo_user_to_kvm(v));
2144 		break;
2145 	case KVM_REG_MIPS_CP0_ENTRYLO1:
2146 		write_gc0_entrylo1(entrylo_user_to_kvm(v));
2147 		break;
2148 	case KVM_REG_MIPS_CP0_CONTEXT:
2149 		write_gc0_context(v);
2150 		break;
2151 	case KVM_REG_MIPS_CP0_CONTEXTCONFIG:
2152 		if (!cpu_guest_has_contextconfig)
2153 			return -EINVAL;
2154 		write_gc0_contextconfig(v);
2155 		break;
2156 	case KVM_REG_MIPS_CP0_USERLOCAL:
2157 		if (!cpu_guest_has_userlocal)
2158 			return -EINVAL;
2159 		write_gc0_userlocal(v);
2160 		break;
2161 #ifdef CONFIG_64BIT
2162 	case KVM_REG_MIPS_CP0_XCONTEXTCONFIG:
2163 		if (!cpu_guest_has_contextconfig)
2164 			return -EINVAL;
2165 		write_gc0_xcontextconfig(v);
2166 		break;
2167 #endif
2168 	case KVM_REG_MIPS_CP0_PAGEMASK:
2169 		write_gc0_pagemask(v);
2170 		break;
2171 	case KVM_REG_MIPS_CP0_PAGEGRAIN:
2172 		write_gc0_pagegrain(v);
2173 		break;
2174 	case KVM_REG_MIPS_CP0_SEGCTL0:
2175 		if (!cpu_guest_has_segments)
2176 			return -EINVAL;
2177 		write_gc0_segctl0(v);
2178 		break;
2179 	case KVM_REG_MIPS_CP0_SEGCTL1:
2180 		if (!cpu_guest_has_segments)
2181 			return -EINVAL;
2182 		write_gc0_segctl1(v);
2183 		break;
2184 	case KVM_REG_MIPS_CP0_SEGCTL2:
2185 		if (!cpu_guest_has_segments)
2186 			return -EINVAL;
2187 		write_gc0_segctl2(v);
2188 		break;
2189 	case KVM_REG_MIPS_CP0_PWBASE:
2190 		if (!cpu_guest_has_htw && !cpu_guest_has_ldpte)
2191 			return -EINVAL;
2192 		write_gc0_pwbase(v);
2193 		break;
2194 	case KVM_REG_MIPS_CP0_PWFIELD:
2195 		if (!cpu_guest_has_htw && !cpu_guest_has_ldpte)
2196 			return -EINVAL;
2197 		write_gc0_pwfield(v);
2198 		break;
2199 	case KVM_REG_MIPS_CP0_PWSIZE:
2200 		if (!cpu_guest_has_htw && !cpu_guest_has_ldpte)
2201 			return -EINVAL;
2202 		write_gc0_pwsize(v);
2203 		break;
2204 	case KVM_REG_MIPS_CP0_WIRED:
2205 		change_gc0_wired(MIPSR6_WIRED_WIRED, v);
2206 		break;
2207 	case KVM_REG_MIPS_CP0_PWCTL:
2208 		if (!cpu_guest_has_htw && !cpu_guest_has_ldpte)
2209 			return -EINVAL;
2210 		write_gc0_pwctl(v);
2211 		break;
2212 	case KVM_REG_MIPS_CP0_HWRENA:
2213 		write_gc0_hwrena(v);
2214 		break;
2215 	case KVM_REG_MIPS_CP0_BADVADDR:
2216 		write_gc0_badvaddr(v);
2217 		break;
2218 	case KVM_REG_MIPS_CP0_BADINSTR:
2219 		if (!cpu_guest_has_badinstr)
2220 			return -EINVAL;
2221 		write_gc0_badinstr(v);
2222 		break;
2223 	case KVM_REG_MIPS_CP0_BADINSTRP:
2224 		if (!cpu_guest_has_badinstrp)
2225 			return -EINVAL;
2226 		write_gc0_badinstrp(v);
2227 		break;
2228 	case KVM_REG_MIPS_CP0_COUNT:
2229 		kvm_mips_write_count(vcpu, v);
2230 		break;
2231 	case KVM_REG_MIPS_CP0_ENTRYHI:
2232 		write_gc0_entryhi(v);
2233 		break;
2234 	case KVM_REG_MIPS_CP0_COMPARE:
2235 		kvm_mips_write_compare(vcpu, v, false);
2236 		break;
2237 	case KVM_REG_MIPS_CP0_STATUS:
2238 		write_gc0_status(v);
2239 		break;
2240 	case KVM_REG_MIPS_CP0_INTCTL:
2241 		write_gc0_intctl(v);
2242 		break;
2243 	case KVM_REG_MIPS_CP0_CAUSE:
2244 		/*
2245 		 * If the timer is stopped or started (DC bit) it must look
2246 		 * atomic with changes to the timer interrupt pending bit (TI).
2247 		 * A timer interrupt should not happen in between.
2248 		 */
2249 		if ((read_gc0_cause() ^ v) & CAUSEF_DC) {
2250 			if (v & CAUSEF_DC) {
2251 				/* disable timer first */
2252 				kvm_mips_count_disable_cause(vcpu);
2253 				change_gc0_cause((u32)~CAUSEF_DC, v);
2254 			} else {
2255 				/* enable timer last */
2256 				change_gc0_cause((u32)~CAUSEF_DC, v);
2257 				kvm_mips_count_enable_cause(vcpu);
2258 			}
2259 		} else {
2260 			write_gc0_cause(v);
2261 		}
2262 		break;
2263 	case KVM_REG_MIPS_CP0_EPC:
2264 		write_gc0_epc(v);
2265 		break;
2266 	case KVM_REG_MIPS_CP0_PRID:
2267 		switch (boot_cpu_type()) {
2268 		case CPU_CAVIUM_OCTEON3:
2269 			/* Octeon III has a guest.PRid, but its read-only */
2270 			break;
2271 		default:
2272 			kvm_write_c0_guest_prid(cop0, v);
2273 			break;
2274 		}
2275 		break;
2276 	case KVM_REG_MIPS_CP0_EBASE:
2277 		kvm_vz_write_gc0_ebase(v);
2278 		break;
2279 	case KVM_REG_MIPS_CP0_CONFIG:
2280 		cur = read_gc0_config();
2281 		change = (cur ^ v) & kvm_vz_config_user_wrmask(vcpu);
2282 		if (change) {
2283 			v = cur ^ change;
2284 			write_gc0_config(v);
2285 		}
2286 		break;
2287 	case KVM_REG_MIPS_CP0_CONFIG1:
2288 		if (!cpu_guest_has_conf1)
2289 			break;
2290 		cur = read_gc0_config1();
2291 		change = (cur ^ v) & kvm_vz_config1_user_wrmask(vcpu);
2292 		if (change) {
2293 			v = cur ^ change;
2294 			write_gc0_config1(v);
2295 		}
2296 		break;
2297 	case KVM_REG_MIPS_CP0_CONFIG2:
2298 		if (!cpu_guest_has_conf2)
2299 			break;
2300 		cur = read_gc0_config2();
2301 		change = (cur ^ v) & kvm_vz_config2_user_wrmask(vcpu);
2302 		if (change) {
2303 			v = cur ^ change;
2304 			write_gc0_config2(v);
2305 		}
2306 		break;
2307 	case KVM_REG_MIPS_CP0_CONFIG3:
2308 		if (!cpu_guest_has_conf3)
2309 			break;
2310 		cur = read_gc0_config3();
2311 		change = (cur ^ v) & kvm_vz_config3_user_wrmask(vcpu);
2312 		if (change) {
2313 			v = cur ^ change;
2314 			write_gc0_config3(v);
2315 		}
2316 		break;
2317 	case KVM_REG_MIPS_CP0_CONFIG4:
2318 		if (!cpu_guest_has_conf4)
2319 			break;
2320 		cur = read_gc0_config4();
2321 		change = (cur ^ v) & kvm_vz_config4_user_wrmask(vcpu);
2322 		if (change) {
2323 			v = cur ^ change;
2324 			write_gc0_config4(v);
2325 		}
2326 		break;
2327 	case KVM_REG_MIPS_CP0_CONFIG5:
2328 		if (!cpu_guest_has_conf5)
2329 			break;
2330 		cur = read_gc0_config5();
2331 		change = (cur ^ v) & kvm_vz_config5_user_wrmask(vcpu);
2332 		if (change) {
2333 			v = cur ^ change;
2334 			write_gc0_config5(v);
2335 		}
2336 		break;
2337 	case KVM_REG_MIPS_CP0_CONFIG6:
2338 		cur = kvm_read_sw_gc0_config6(cop0);
2339 		change = (cur ^ v) & kvm_vz_config6_user_wrmask(vcpu);
2340 		if (change) {
2341 			v = cur ^ change;
2342 			kvm_write_sw_gc0_config6(cop0, (int)v);
2343 		}
2344 		break;
2345 	case KVM_REG_MIPS_CP0_MAAR(0) ... KVM_REG_MIPS_CP0_MAAR(0x3f):
2346 		if (!cpu_guest_has_maar || cpu_guest_has_dyn_maar)
2347 			return -EINVAL;
2348 		idx = reg->id - KVM_REG_MIPS_CP0_MAAR(0);
2349 		if (idx >= ARRAY_SIZE(vcpu->arch.maar))
2350 			return -EINVAL;
2351 		vcpu->arch.maar[idx] = mips_process_maar(dmtc_op, v);
2352 		break;
2353 	case KVM_REG_MIPS_CP0_MAARI:
2354 		if (!cpu_guest_has_maar || cpu_guest_has_dyn_maar)
2355 			return -EINVAL;
2356 		kvm_write_maari(vcpu, v);
2357 		break;
2358 #ifdef CONFIG_64BIT
2359 	case KVM_REG_MIPS_CP0_XCONTEXT:
2360 		write_gc0_xcontext(v);
2361 		break;
2362 #endif
2363 	case KVM_REG_MIPS_CP0_ERROREPC:
2364 		write_gc0_errorepc(v);
2365 		break;
2366 	case KVM_REG_MIPS_CP0_KSCRATCH1 ... KVM_REG_MIPS_CP0_KSCRATCH6:
2367 		idx = reg->id - KVM_REG_MIPS_CP0_KSCRATCH1 + 2;
2368 		if (!cpu_guest_has_kscr(idx))
2369 			return -EINVAL;
2370 		switch (idx) {
2371 		case 2:
2372 			write_gc0_kscratch1(v);
2373 			break;
2374 		case 3:
2375 			write_gc0_kscratch2(v);
2376 			break;
2377 		case 4:
2378 			write_gc0_kscratch3(v);
2379 			break;
2380 		case 5:
2381 			write_gc0_kscratch4(v);
2382 			break;
2383 		case 6:
2384 			write_gc0_kscratch5(v);
2385 			break;
2386 		case 7:
2387 			write_gc0_kscratch6(v);
2388 			break;
2389 		}
2390 		break;
2391 	case KVM_REG_MIPS_COUNT_CTL:
2392 		ret = kvm_mips_set_count_ctl(vcpu, v);
2393 		break;
2394 	case KVM_REG_MIPS_COUNT_RESUME:
2395 		ret = kvm_mips_set_count_resume(vcpu, v);
2396 		break;
2397 	case KVM_REG_MIPS_COUNT_HZ:
2398 		ret = kvm_mips_set_count_hz(vcpu, v);
2399 		break;
2400 	default:
2401 		return -EINVAL;
2402 	}
2403 	return ret;
2404 }
2405 
2406 #define guestid_cache(cpu)	(cpu_data[cpu].guestid_cache)
2407 static void kvm_vz_get_new_guestid(unsigned long cpu, struct kvm_vcpu *vcpu)
2408 {
2409 	unsigned long guestid = guestid_cache(cpu);
2410 
2411 	if (!(++guestid & GUESTID_MASK)) {
2412 		if (cpu_has_vtag_icache)
2413 			flush_icache_all();
2414 
2415 		if (!guestid)		/* fix version if needed */
2416 			guestid = GUESTID_FIRST_VERSION;
2417 
2418 		++guestid;		/* guestid 0 reserved for root */
2419 
2420 		/* start new guestid cycle */
2421 		kvm_vz_local_flush_roottlb_all_guests();
2422 		kvm_vz_local_flush_guesttlb_all();
2423 	}
2424 
2425 	guestid_cache(cpu) = guestid;
2426 }
2427 
2428 /* Returns 1 if the guest TLB may be clobbered */
2429 static int kvm_vz_check_requests(struct kvm_vcpu *vcpu, int cpu)
2430 {
2431 	int ret = 0;
2432 	int i;
2433 
2434 	if (!kvm_request_pending(vcpu))
2435 		return 0;
2436 
2437 	if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
2438 		if (cpu_has_guestid) {
2439 			/* Drop all GuestIDs for this VCPU */
2440 			for_each_possible_cpu(i)
2441 				vcpu->arch.vzguestid[i] = 0;
2442 			/* This will clobber guest TLB contents too */
2443 			ret = 1;
2444 		}
2445 		/*
2446 		 * For Root ASID Dealias (RAD) we don't do anything here, but we
2447 		 * still need the request to ensure we recheck asid_flush_mask.
2448 		 * We can still return 0 as only the root TLB will be affected
2449 		 * by a root ASID flush.
2450 		 */
2451 	}
2452 
2453 	return ret;
2454 }
2455 
2456 static void kvm_vz_vcpu_save_wired(struct kvm_vcpu *vcpu)
2457 {
2458 	unsigned int wired = read_gc0_wired();
2459 	struct kvm_mips_tlb *tlbs;
2460 	int i;
2461 
2462 	/* Expand the wired TLB array if necessary */
2463 	wired &= MIPSR6_WIRED_WIRED;
2464 	if (wired > vcpu->arch.wired_tlb_limit) {
2465 		tlbs = krealloc(vcpu->arch.wired_tlb, wired *
2466 				sizeof(*vcpu->arch.wired_tlb), GFP_ATOMIC);
2467 		if (WARN_ON(!tlbs)) {
2468 			/* Save whatever we can */
2469 			wired = vcpu->arch.wired_tlb_limit;
2470 		} else {
2471 			vcpu->arch.wired_tlb = tlbs;
2472 			vcpu->arch.wired_tlb_limit = wired;
2473 		}
2474 	}
2475 
2476 	if (wired)
2477 		/* Save wired entries from the guest TLB */
2478 		kvm_vz_save_guesttlb(vcpu->arch.wired_tlb, 0, wired);
2479 	/* Invalidate any dropped entries since last time */
2480 	for (i = wired; i < vcpu->arch.wired_tlb_used; ++i) {
2481 		vcpu->arch.wired_tlb[i].tlb_hi = UNIQUE_GUEST_ENTRYHI(i);
2482 		vcpu->arch.wired_tlb[i].tlb_lo[0] = 0;
2483 		vcpu->arch.wired_tlb[i].tlb_lo[1] = 0;
2484 		vcpu->arch.wired_tlb[i].tlb_mask = 0;
2485 	}
2486 	vcpu->arch.wired_tlb_used = wired;
2487 }
2488 
2489 static void kvm_vz_vcpu_load_wired(struct kvm_vcpu *vcpu)
2490 {
2491 	/* Load wired entries into the guest TLB */
2492 	if (vcpu->arch.wired_tlb)
2493 		kvm_vz_load_guesttlb(vcpu->arch.wired_tlb, 0,
2494 				     vcpu->arch.wired_tlb_used);
2495 }
2496 
2497 static void kvm_vz_vcpu_load_tlb(struct kvm_vcpu *vcpu, int cpu)
2498 {
2499 	struct kvm *kvm = vcpu->kvm;
2500 	struct mm_struct *gpa_mm = &kvm->arch.gpa_mm;
2501 	bool migrated;
2502 
2503 	/*
2504 	 * Are we entering guest context on a different CPU to last time?
2505 	 * If so, the VCPU's guest TLB state on this CPU may be stale.
2506 	 */
2507 	migrated = (vcpu->arch.last_exec_cpu != cpu);
2508 	vcpu->arch.last_exec_cpu = cpu;
2509 
2510 	/*
2511 	 * A vcpu's GuestID is set in GuestCtl1.ID when the vcpu is loaded and
2512 	 * remains set until another vcpu is loaded in.  As a rule GuestRID
2513 	 * remains zeroed when in root context unless the kernel is busy
2514 	 * manipulating guest tlb entries.
2515 	 */
2516 	if (cpu_has_guestid) {
2517 		/*
2518 		 * Check if our GuestID is of an older version and thus invalid.
2519 		 *
2520 		 * We also discard the stored GuestID if we've executed on
2521 		 * another CPU, as the guest mappings may have changed without
2522 		 * hypervisor knowledge.
2523 		 */
2524 		if (migrated ||
2525 		    (vcpu->arch.vzguestid[cpu] ^ guestid_cache(cpu)) &
2526 					GUESTID_VERSION_MASK) {
2527 			kvm_vz_get_new_guestid(cpu, vcpu);
2528 			vcpu->arch.vzguestid[cpu] = guestid_cache(cpu);
2529 			trace_kvm_guestid_change(vcpu,
2530 						 vcpu->arch.vzguestid[cpu]);
2531 		}
2532 
2533 		/* Restore GuestID */
2534 		change_c0_guestctl1(GUESTID_MASK, vcpu->arch.vzguestid[cpu]);
2535 	} else {
2536 		/*
2537 		 * The Guest TLB only stores a single guest's TLB state, so
2538 		 * flush it if another VCPU has executed on this CPU.
2539 		 *
2540 		 * We also flush if we've executed on another CPU, as the guest
2541 		 * mappings may have changed without hypervisor knowledge.
2542 		 */
2543 		if (migrated || last_exec_vcpu[cpu] != vcpu)
2544 			kvm_vz_local_flush_guesttlb_all();
2545 		last_exec_vcpu[cpu] = vcpu;
2546 
2547 		/*
2548 		 * Root ASID dealiases guest GPA mappings in the root TLB.
2549 		 * Allocate new root ASID if needed.
2550 		 */
2551 		if (cpumask_test_and_clear_cpu(cpu, &kvm->arch.asid_flush_mask))
2552 			get_new_mmu_context(gpa_mm);
2553 		else
2554 			check_mmu_context(gpa_mm);
2555 	}
2556 }
2557 
2558 static int kvm_vz_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2559 {
2560 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2561 	bool migrated, all;
2562 
2563 	/*
2564 	 * Have we migrated to a different CPU?
2565 	 * If so, any old guest TLB state may be stale.
2566 	 */
2567 	migrated = (vcpu->arch.last_sched_cpu != cpu);
2568 
2569 	/*
2570 	 * Was this the last VCPU to run on this CPU?
2571 	 * If not, any old guest state from this VCPU will have been clobbered.
2572 	 */
2573 	all = migrated || (last_vcpu[cpu] != vcpu);
2574 	last_vcpu[cpu] = vcpu;
2575 
2576 	/*
2577 	 * Restore CP0_Wired unconditionally as we clear it after use, and
2578 	 * restore wired guest TLB entries (while in guest context).
2579 	 */
2580 	kvm_restore_gc0_wired(cop0);
2581 	if (current->flags & PF_VCPU) {
2582 		tlbw_use_hazard();
2583 		kvm_vz_vcpu_load_tlb(vcpu, cpu);
2584 		kvm_vz_vcpu_load_wired(vcpu);
2585 	}
2586 
2587 	/*
2588 	 * Restore timer state regardless, as e.g. Cause.TI can change over time
2589 	 * if left unmaintained.
2590 	 */
2591 	kvm_vz_restore_timer(vcpu);
2592 
2593 	/* Set MC bit if we want to trace guest mode changes */
2594 	if (kvm_trace_guest_mode_change)
2595 		set_c0_guestctl0(MIPS_GCTL0_MC);
2596 	else
2597 		clear_c0_guestctl0(MIPS_GCTL0_MC);
2598 
2599 	/* Don't bother restoring registers multiple times unless necessary */
2600 	if (!all)
2601 		return 0;
2602 
2603 	/*
2604 	 * Restore config registers first, as some implementations restrict
2605 	 * writes to other registers when the corresponding feature bits aren't
2606 	 * set. For example Status.CU1 cannot be set unless Config1.FP is set.
2607 	 */
2608 	kvm_restore_gc0_config(cop0);
2609 	if (cpu_guest_has_conf1)
2610 		kvm_restore_gc0_config1(cop0);
2611 	if (cpu_guest_has_conf2)
2612 		kvm_restore_gc0_config2(cop0);
2613 	if (cpu_guest_has_conf3)
2614 		kvm_restore_gc0_config3(cop0);
2615 	if (cpu_guest_has_conf4)
2616 		kvm_restore_gc0_config4(cop0);
2617 	if (cpu_guest_has_conf5)
2618 		kvm_restore_gc0_config5(cop0);
2619 	if (cpu_guest_has_conf6)
2620 		kvm_restore_gc0_config6(cop0);
2621 	if (cpu_guest_has_conf7)
2622 		kvm_restore_gc0_config7(cop0);
2623 
2624 	kvm_restore_gc0_index(cop0);
2625 	kvm_restore_gc0_entrylo0(cop0);
2626 	kvm_restore_gc0_entrylo1(cop0);
2627 	kvm_restore_gc0_context(cop0);
2628 	if (cpu_guest_has_contextconfig)
2629 		kvm_restore_gc0_contextconfig(cop0);
2630 #ifdef CONFIG_64BIT
2631 	kvm_restore_gc0_xcontext(cop0);
2632 	if (cpu_guest_has_contextconfig)
2633 		kvm_restore_gc0_xcontextconfig(cop0);
2634 #endif
2635 	kvm_restore_gc0_pagemask(cop0);
2636 	kvm_restore_gc0_pagegrain(cop0);
2637 	kvm_restore_gc0_hwrena(cop0);
2638 	kvm_restore_gc0_badvaddr(cop0);
2639 	kvm_restore_gc0_entryhi(cop0);
2640 	kvm_restore_gc0_status(cop0);
2641 	kvm_restore_gc0_intctl(cop0);
2642 	kvm_restore_gc0_epc(cop0);
2643 	kvm_vz_write_gc0_ebase(kvm_read_sw_gc0_ebase(cop0));
2644 	if (cpu_guest_has_userlocal)
2645 		kvm_restore_gc0_userlocal(cop0);
2646 
2647 	kvm_restore_gc0_errorepc(cop0);
2648 
2649 	/* restore KScratch registers if enabled in guest */
2650 	if (cpu_guest_has_conf4) {
2651 		if (cpu_guest_has_kscr(2))
2652 			kvm_restore_gc0_kscratch1(cop0);
2653 		if (cpu_guest_has_kscr(3))
2654 			kvm_restore_gc0_kscratch2(cop0);
2655 		if (cpu_guest_has_kscr(4))
2656 			kvm_restore_gc0_kscratch3(cop0);
2657 		if (cpu_guest_has_kscr(5))
2658 			kvm_restore_gc0_kscratch4(cop0);
2659 		if (cpu_guest_has_kscr(6))
2660 			kvm_restore_gc0_kscratch5(cop0);
2661 		if (cpu_guest_has_kscr(7))
2662 			kvm_restore_gc0_kscratch6(cop0);
2663 	}
2664 
2665 	if (cpu_guest_has_badinstr)
2666 		kvm_restore_gc0_badinstr(cop0);
2667 	if (cpu_guest_has_badinstrp)
2668 		kvm_restore_gc0_badinstrp(cop0);
2669 
2670 	if (cpu_guest_has_segments) {
2671 		kvm_restore_gc0_segctl0(cop0);
2672 		kvm_restore_gc0_segctl1(cop0);
2673 		kvm_restore_gc0_segctl2(cop0);
2674 	}
2675 
2676 	/* restore HTW registers */
2677 	if (cpu_guest_has_htw || cpu_guest_has_ldpte) {
2678 		kvm_restore_gc0_pwbase(cop0);
2679 		kvm_restore_gc0_pwfield(cop0);
2680 		kvm_restore_gc0_pwsize(cop0);
2681 		kvm_restore_gc0_pwctl(cop0);
2682 	}
2683 
2684 	/* restore Root.GuestCtl2 from unused Guest guestctl2 register */
2685 	if (cpu_has_guestctl2)
2686 		write_c0_guestctl2(
2687 			cop0->reg[MIPS_CP0_GUESTCTL2][MIPS_CP0_GUESTCTL2_SEL]);
2688 
2689 	/*
2690 	 * We should clear linked load bit to break interrupted atomics. This
2691 	 * prevents a SC on the next VCPU from succeeding by matching a LL on
2692 	 * the previous VCPU.
2693 	 */
2694 	if (vcpu->kvm->created_vcpus > 1)
2695 		write_gc0_lladdr(0);
2696 
2697 	return 0;
2698 }
2699 
2700 static int kvm_vz_vcpu_put(struct kvm_vcpu *vcpu, int cpu)
2701 {
2702 	struct mips_coproc *cop0 = vcpu->arch.cop0;
2703 
2704 	if (current->flags & PF_VCPU)
2705 		kvm_vz_vcpu_save_wired(vcpu);
2706 
2707 	kvm_lose_fpu(vcpu);
2708 
2709 	kvm_save_gc0_index(cop0);
2710 	kvm_save_gc0_entrylo0(cop0);
2711 	kvm_save_gc0_entrylo1(cop0);
2712 	kvm_save_gc0_context(cop0);
2713 	if (cpu_guest_has_contextconfig)
2714 		kvm_save_gc0_contextconfig(cop0);
2715 #ifdef CONFIG_64BIT
2716 	kvm_save_gc0_xcontext(cop0);
2717 	if (cpu_guest_has_contextconfig)
2718 		kvm_save_gc0_xcontextconfig(cop0);
2719 #endif
2720 	kvm_save_gc0_pagemask(cop0);
2721 	kvm_save_gc0_pagegrain(cop0);
2722 	kvm_save_gc0_wired(cop0);
2723 	/* allow wired TLB entries to be overwritten */
2724 	clear_gc0_wired(MIPSR6_WIRED_WIRED);
2725 	kvm_save_gc0_hwrena(cop0);
2726 	kvm_save_gc0_badvaddr(cop0);
2727 	kvm_save_gc0_entryhi(cop0);
2728 	kvm_save_gc0_status(cop0);
2729 	kvm_save_gc0_intctl(cop0);
2730 	kvm_save_gc0_epc(cop0);
2731 	kvm_write_sw_gc0_ebase(cop0, kvm_vz_read_gc0_ebase());
2732 	if (cpu_guest_has_userlocal)
2733 		kvm_save_gc0_userlocal(cop0);
2734 
2735 	/* only save implemented config registers */
2736 	kvm_save_gc0_config(cop0);
2737 	if (cpu_guest_has_conf1)
2738 		kvm_save_gc0_config1(cop0);
2739 	if (cpu_guest_has_conf2)
2740 		kvm_save_gc0_config2(cop0);
2741 	if (cpu_guest_has_conf3)
2742 		kvm_save_gc0_config3(cop0);
2743 	if (cpu_guest_has_conf4)
2744 		kvm_save_gc0_config4(cop0);
2745 	if (cpu_guest_has_conf5)
2746 		kvm_save_gc0_config5(cop0);
2747 	if (cpu_guest_has_conf6)
2748 		kvm_save_gc0_config6(cop0);
2749 	if (cpu_guest_has_conf7)
2750 		kvm_save_gc0_config7(cop0);
2751 
2752 	kvm_save_gc0_errorepc(cop0);
2753 
2754 	/* save KScratch registers if enabled in guest */
2755 	if (cpu_guest_has_conf4) {
2756 		if (cpu_guest_has_kscr(2))
2757 			kvm_save_gc0_kscratch1(cop0);
2758 		if (cpu_guest_has_kscr(3))
2759 			kvm_save_gc0_kscratch2(cop0);
2760 		if (cpu_guest_has_kscr(4))
2761 			kvm_save_gc0_kscratch3(cop0);
2762 		if (cpu_guest_has_kscr(5))
2763 			kvm_save_gc0_kscratch4(cop0);
2764 		if (cpu_guest_has_kscr(6))
2765 			kvm_save_gc0_kscratch5(cop0);
2766 		if (cpu_guest_has_kscr(7))
2767 			kvm_save_gc0_kscratch6(cop0);
2768 	}
2769 
2770 	if (cpu_guest_has_badinstr)
2771 		kvm_save_gc0_badinstr(cop0);
2772 	if (cpu_guest_has_badinstrp)
2773 		kvm_save_gc0_badinstrp(cop0);
2774 
2775 	if (cpu_guest_has_segments) {
2776 		kvm_save_gc0_segctl0(cop0);
2777 		kvm_save_gc0_segctl1(cop0);
2778 		kvm_save_gc0_segctl2(cop0);
2779 	}
2780 
2781 	/* save HTW registers if enabled in guest */
2782 	if (cpu_guest_has_ldpte || (cpu_guest_has_htw &&
2783 	    kvm_read_sw_gc0_config3(cop0) & MIPS_CONF3_PW)) {
2784 		kvm_save_gc0_pwbase(cop0);
2785 		kvm_save_gc0_pwfield(cop0);
2786 		kvm_save_gc0_pwsize(cop0);
2787 		kvm_save_gc0_pwctl(cop0);
2788 	}
2789 
2790 	kvm_vz_save_timer(vcpu);
2791 
2792 	/* save Root.GuestCtl2 in unused Guest guestctl2 register */
2793 	if (cpu_has_guestctl2)
2794 		cop0->reg[MIPS_CP0_GUESTCTL2][MIPS_CP0_GUESTCTL2_SEL] =
2795 			read_c0_guestctl2();
2796 
2797 	return 0;
2798 }
2799 
2800 /**
2801  * kvm_vz_resize_guest_vtlb() - Attempt to resize guest VTLB.
2802  * @size:	Number of guest VTLB entries (0 < @size <= root VTLB entries).
2803  *
2804  * Attempt to resize the guest VTLB by writing guest Config registers. This is
2805  * necessary for cores with a shared root/guest TLB to avoid overlap with wired
2806  * entries in the root VTLB.
2807  *
2808  * Returns:	The resulting guest VTLB size.
2809  */
2810 static unsigned int kvm_vz_resize_guest_vtlb(unsigned int size)
2811 {
2812 	unsigned int config4 = 0, ret = 0, limit;
2813 
2814 	/* Write MMUSize - 1 into guest Config registers */
2815 	if (cpu_guest_has_conf1)
2816 		change_gc0_config1(MIPS_CONF1_TLBS,
2817 				   (size - 1) << MIPS_CONF1_TLBS_SHIFT);
2818 	if (cpu_guest_has_conf4) {
2819 		config4 = read_gc0_config4();
2820 		if (cpu_has_mips_r6 || (config4 & MIPS_CONF4_MMUEXTDEF) ==
2821 		    MIPS_CONF4_MMUEXTDEF_VTLBSIZEEXT) {
2822 			config4 &= ~MIPS_CONF4_VTLBSIZEEXT;
2823 			config4 |= ((size - 1) >> MIPS_CONF1_TLBS_SIZE) <<
2824 				MIPS_CONF4_VTLBSIZEEXT_SHIFT;
2825 		} else if ((config4 & MIPS_CONF4_MMUEXTDEF) ==
2826 			   MIPS_CONF4_MMUEXTDEF_MMUSIZEEXT) {
2827 			config4 &= ~MIPS_CONF4_MMUSIZEEXT;
2828 			config4 |= ((size - 1) >> MIPS_CONF1_TLBS_SIZE) <<
2829 				MIPS_CONF4_MMUSIZEEXT_SHIFT;
2830 		}
2831 		write_gc0_config4(config4);
2832 	}
2833 
2834 	/*
2835 	 * Set Guest.Wired.Limit = 0 (no limit up to Guest.MMUSize-1), unless it
2836 	 * would exceed Root.Wired.Limit (clearing Guest.Wired.Wired so write
2837 	 * not dropped)
2838 	 */
2839 	if (cpu_has_mips_r6) {
2840 		limit = (read_c0_wired() & MIPSR6_WIRED_LIMIT) >>
2841 						MIPSR6_WIRED_LIMIT_SHIFT;
2842 		if (size - 1 <= limit)
2843 			limit = 0;
2844 		write_gc0_wired(limit << MIPSR6_WIRED_LIMIT_SHIFT);
2845 	}
2846 
2847 	/* Read back MMUSize - 1 */
2848 	back_to_back_c0_hazard();
2849 	if (cpu_guest_has_conf1)
2850 		ret = (read_gc0_config1() & MIPS_CONF1_TLBS) >>
2851 						MIPS_CONF1_TLBS_SHIFT;
2852 	if (config4) {
2853 		if (cpu_has_mips_r6 || (config4 & MIPS_CONF4_MMUEXTDEF) ==
2854 		    MIPS_CONF4_MMUEXTDEF_VTLBSIZEEXT)
2855 			ret |= ((config4 & MIPS_CONF4_VTLBSIZEEXT) >>
2856 				MIPS_CONF4_VTLBSIZEEXT_SHIFT) <<
2857 				MIPS_CONF1_TLBS_SIZE;
2858 		else if ((config4 & MIPS_CONF4_MMUEXTDEF) ==
2859 			 MIPS_CONF4_MMUEXTDEF_MMUSIZEEXT)
2860 			ret |= ((config4 & MIPS_CONF4_MMUSIZEEXT) >>
2861 				MIPS_CONF4_MMUSIZEEXT_SHIFT) <<
2862 				MIPS_CONF1_TLBS_SIZE;
2863 	}
2864 	return ret + 1;
2865 }
2866 
2867 static int kvm_vz_hardware_enable(void)
2868 {
2869 	unsigned int mmu_size, guest_mmu_size, ftlb_size;
2870 	u64 guest_cvmctl, cvmvmconfig;
2871 
2872 	switch (current_cpu_type()) {
2873 	case CPU_CAVIUM_OCTEON3:
2874 		/* Set up guest timer/perfcount IRQ lines */
2875 		guest_cvmctl = read_gc0_cvmctl();
2876 		guest_cvmctl &= ~CVMCTL_IPTI;
2877 		guest_cvmctl |= 7ull << CVMCTL_IPTI_SHIFT;
2878 		guest_cvmctl &= ~CVMCTL_IPPCI;
2879 		guest_cvmctl |= 6ull << CVMCTL_IPPCI_SHIFT;
2880 		write_gc0_cvmctl(guest_cvmctl);
2881 
2882 		cvmvmconfig = read_c0_cvmvmconfig();
2883 		/* No I/O hole translation. */
2884 		cvmvmconfig |= CVMVMCONF_DGHT;
2885 		/* Halve the root MMU size */
2886 		mmu_size = ((cvmvmconfig & CVMVMCONF_MMUSIZEM1)
2887 			    >> CVMVMCONF_MMUSIZEM1_S) + 1;
2888 		guest_mmu_size = mmu_size / 2;
2889 		mmu_size -= guest_mmu_size;
2890 		cvmvmconfig &= ~CVMVMCONF_RMMUSIZEM1;
2891 		cvmvmconfig |= mmu_size - 1;
2892 		write_c0_cvmvmconfig(cvmvmconfig);
2893 
2894 		/* Update our records */
2895 		current_cpu_data.tlbsize = mmu_size;
2896 		current_cpu_data.tlbsizevtlb = mmu_size;
2897 		current_cpu_data.guest.tlbsize = guest_mmu_size;
2898 
2899 		/* Flush moved entries in new (guest) context */
2900 		kvm_vz_local_flush_guesttlb_all();
2901 		break;
2902 	default:
2903 		/*
2904 		 * ImgTec cores tend to use a shared root/guest TLB. To avoid
2905 		 * overlap of root wired and guest entries, the guest TLB may
2906 		 * need resizing.
2907 		 */
2908 		mmu_size = current_cpu_data.tlbsizevtlb;
2909 		ftlb_size = current_cpu_data.tlbsize - mmu_size;
2910 
2911 		/* Try switching to maximum guest VTLB size for flush */
2912 		guest_mmu_size = kvm_vz_resize_guest_vtlb(mmu_size);
2913 		current_cpu_data.guest.tlbsize = guest_mmu_size + ftlb_size;
2914 		kvm_vz_local_flush_guesttlb_all();
2915 
2916 		/*
2917 		 * Reduce to make space for root wired entries and at least 2
2918 		 * root non-wired entries. This does assume that long-term wired
2919 		 * entries won't be added later.
2920 		 */
2921 		guest_mmu_size = mmu_size - num_wired_entries() - 2;
2922 		guest_mmu_size = kvm_vz_resize_guest_vtlb(guest_mmu_size);
2923 		current_cpu_data.guest.tlbsize = guest_mmu_size + ftlb_size;
2924 
2925 		/*
2926 		 * Write the VTLB size, but if another CPU has already written,
2927 		 * check it matches or we won't provide a consistent view to the
2928 		 * guest. If this ever happens it suggests an asymmetric number
2929 		 * of wired entries.
2930 		 */
2931 		if (cmpxchg(&kvm_vz_guest_vtlb_size, 0, guest_mmu_size) &&
2932 		    WARN(guest_mmu_size != kvm_vz_guest_vtlb_size,
2933 			 "Available guest VTLB size mismatch"))
2934 			return -EINVAL;
2935 		break;
2936 	}
2937 
2938 	/*
2939 	 * Enable virtualization features granting guest direct control of
2940 	 * certain features:
2941 	 * CP0=1:	Guest coprocessor 0 context.
2942 	 * AT=Guest:	Guest MMU.
2943 	 * CG=1:	Hit (virtual address) CACHE operations (optional).
2944 	 * CF=1:	Guest Config registers.
2945 	 * CGI=1:	Indexed flush CACHE operations (optional).
2946 	 */
2947 	write_c0_guestctl0(MIPS_GCTL0_CP0 |
2948 			   (MIPS_GCTL0_AT_GUEST << MIPS_GCTL0_AT_SHIFT) |
2949 			   MIPS_GCTL0_CG | MIPS_GCTL0_CF);
2950 	if (cpu_has_guestctl0ext) {
2951 		if (current_cpu_type() != CPU_LOONGSON64)
2952 			set_c0_guestctl0ext(MIPS_GCTL0EXT_CGI);
2953 		else
2954 			clear_c0_guestctl0ext(MIPS_GCTL0EXT_CGI);
2955 	}
2956 
2957 	if (cpu_has_guestid) {
2958 		write_c0_guestctl1(0);
2959 		kvm_vz_local_flush_roottlb_all_guests();
2960 
2961 		GUESTID_MASK = current_cpu_data.guestid_mask;
2962 		GUESTID_FIRST_VERSION = GUESTID_MASK + 1;
2963 		GUESTID_VERSION_MASK = ~GUESTID_MASK;
2964 
2965 		current_cpu_data.guestid_cache = GUESTID_FIRST_VERSION;
2966 	}
2967 
2968 	/* clear any pending injected virtual guest interrupts */
2969 	if (cpu_has_guestctl2)
2970 		clear_c0_guestctl2(0x3f << 10);
2971 
2972 #ifdef CONFIG_CPU_LOONGSON64
2973 	/* Control guest CCA attribute */
2974 	if (cpu_has_csr())
2975 		csr_writel(csr_readl(0xffffffec) | 0x1, 0xffffffec);
2976 #endif
2977 
2978 	return 0;
2979 }
2980 
2981 static void kvm_vz_hardware_disable(void)
2982 {
2983 	u64 cvmvmconfig;
2984 	unsigned int mmu_size;
2985 
2986 	/* Flush any remaining guest TLB entries */
2987 	kvm_vz_local_flush_guesttlb_all();
2988 
2989 	switch (current_cpu_type()) {
2990 	case CPU_CAVIUM_OCTEON3:
2991 		/*
2992 		 * Allocate whole TLB for root. Existing guest TLB entries will
2993 		 * change ownership to the root TLB. We should be safe though as
2994 		 * they've already been flushed above while in guest TLB.
2995 		 */
2996 		cvmvmconfig = read_c0_cvmvmconfig();
2997 		mmu_size = ((cvmvmconfig & CVMVMCONF_MMUSIZEM1)
2998 			    >> CVMVMCONF_MMUSIZEM1_S) + 1;
2999 		cvmvmconfig &= ~CVMVMCONF_RMMUSIZEM1;
3000 		cvmvmconfig |= mmu_size - 1;
3001 		write_c0_cvmvmconfig(cvmvmconfig);
3002 
3003 		/* Update our records */
3004 		current_cpu_data.tlbsize = mmu_size;
3005 		current_cpu_data.tlbsizevtlb = mmu_size;
3006 		current_cpu_data.guest.tlbsize = 0;
3007 
3008 		/* Flush moved entries in new (root) context */
3009 		local_flush_tlb_all();
3010 		break;
3011 	}
3012 
3013 	if (cpu_has_guestid) {
3014 		write_c0_guestctl1(0);
3015 		kvm_vz_local_flush_roottlb_all_guests();
3016 	}
3017 }
3018 
3019 static int kvm_vz_check_extension(struct kvm *kvm, long ext)
3020 {
3021 	int r;
3022 
3023 	switch (ext) {
3024 	case KVM_CAP_MIPS_VZ:
3025 		/* we wouldn't be here unless cpu_has_vz */
3026 		r = 1;
3027 		break;
3028 #ifdef CONFIG_64BIT
3029 	case KVM_CAP_MIPS_64BIT:
3030 		/* We support 64-bit registers/operations and addresses */
3031 		r = 2;
3032 		break;
3033 #endif
3034 	case KVM_CAP_IOEVENTFD:
3035 		r = 1;
3036 		break;
3037 	default:
3038 		r = 0;
3039 		break;
3040 	}
3041 
3042 	return r;
3043 }
3044 
3045 static int kvm_vz_vcpu_init(struct kvm_vcpu *vcpu)
3046 {
3047 	int i;
3048 
3049 	for_each_possible_cpu(i)
3050 		vcpu->arch.vzguestid[i] = 0;
3051 
3052 	return 0;
3053 }
3054 
3055 static void kvm_vz_vcpu_uninit(struct kvm_vcpu *vcpu)
3056 {
3057 	int cpu;
3058 
3059 	/*
3060 	 * If the VCPU is freed and reused as another VCPU, we don't want the
3061 	 * matching pointer wrongly hanging around in last_vcpu[] or
3062 	 * last_exec_vcpu[].
3063 	 */
3064 	for_each_possible_cpu(cpu) {
3065 		if (last_vcpu[cpu] == vcpu)
3066 			last_vcpu[cpu] = NULL;
3067 		if (last_exec_vcpu[cpu] == vcpu)
3068 			last_exec_vcpu[cpu] = NULL;
3069 	}
3070 }
3071 
3072 static int kvm_vz_vcpu_setup(struct kvm_vcpu *vcpu)
3073 {
3074 	struct mips_coproc *cop0 = vcpu->arch.cop0;
3075 	unsigned long count_hz = 100*1000*1000; /* default to 100 MHz */
3076 
3077 	/*
3078 	 * Start off the timer at the same frequency as the host timer, but the
3079 	 * soft timer doesn't handle frequencies greater than 1GHz yet.
3080 	 */
3081 	if (mips_hpt_frequency && mips_hpt_frequency <= NSEC_PER_SEC)
3082 		count_hz = mips_hpt_frequency;
3083 	kvm_mips_init_count(vcpu, count_hz);
3084 
3085 	/*
3086 	 * Initialize guest register state to valid architectural reset state.
3087 	 */
3088 
3089 	/* PageGrain */
3090 	if (cpu_has_mips_r5 || cpu_has_mips_r6)
3091 		kvm_write_sw_gc0_pagegrain(cop0, PG_RIE | PG_XIE | PG_IEC);
3092 	/* Wired */
3093 	if (cpu_has_mips_r6)
3094 		kvm_write_sw_gc0_wired(cop0,
3095 				       read_gc0_wired() & MIPSR6_WIRED_LIMIT);
3096 	/* Status */
3097 	kvm_write_sw_gc0_status(cop0, ST0_BEV | ST0_ERL);
3098 	if (cpu_has_mips_r5 || cpu_has_mips_r6)
3099 		kvm_change_sw_gc0_status(cop0, ST0_FR, read_gc0_status());
3100 	/* IntCtl */
3101 	kvm_write_sw_gc0_intctl(cop0, read_gc0_intctl() &
3102 				(INTCTLF_IPFDC | INTCTLF_IPPCI | INTCTLF_IPTI));
3103 	/* PRId */
3104 	kvm_write_sw_gc0_prid(cop0, boot_cpu_data.processor_id);
3105 	/* EBase */
3106 	kvm_write_sw_gc0_ebase(cop0, (s32)0x80000000 | vcpu->vcpu_id);
3107 	/* Config */
3108 	kvm_save_gc0_config(cop0);
3109 	/* architecturally writable (e.g. from guest) */
3110 	kvm_change_sw_gc0_config(cop0, CONF_CM_CMASK,
3111 				 _page_cachable_default >> _CACHE_SHIFT);
3112 	/* architecturally read only, but maybe writable from root */
3113 	kvm_change_sw_gc0_config(cop0, MIPS_CONF_MT, read_c0_config());
3114 	if (cpu_guest_has_conf1) {
3115 		kvm_set_sw_gc0_config(cop0, MIPS_CONF_M);
3116 		/* Config1 */
3117 		kvm_save_gc0_config1(cop0);
3118 		/* architecturally read only, but maybe writable from root */
3119 		kvm_clear_sw_gc0_config1(cop0, MIPS_CONF1_C2	|
3120 					       MIPS_CONF1_MD	|
3121 					       MIPS_CONF1_PC	|
3122 					       MIPS_CONF1_WR	|
3123 					       MIPS_CONF1_CA	|
3124 					       MIPS_CONF1_FP);
3125 	}
3126 	if (cpu_guest_has_conf2) {
3127 		kvm_set_sw_gc0_config1(cop0, MIPS_CONF_M);
3128 		/* Config2 */
3129 		kvm_save_gc0_config2(cop0);
3130 	}
3131 	if (cpu_guest_has_conf3) {
3132 		kvm_set_sw_gc0_config2(cop0, MIPS_CONF_M);
3133 		/* Config3 */
3134 		kvm_save_gc0_config3(cop0);
3135 		/* architecturally writable (e.g. from guest) */
3136 		kvm_clear_sw_gc0_config3(cop0, MIPS_CONF3_ISA_OE);
3137 		/* architecturally read only, but maybe writable from root */
3138 		kvm_clear_sw_gc0_config3(cop0, MIPS_CONF3_MSA	|
3139 					       MIPS_CONF3_BPG	|
3140 					       MIPS_CONF3_ULRI	|
3141 					       MIPS_CONF3_DSP	|
3142 					       MIPS_CONF3_CTXTC	|
3143 					       MIPS_CONF3_ITL	|
3144 					       MIPS_CONF3_LPA	|
3145 					       MIPS_CONF3_VEIC	|
3146 					       MIPS_CONF3_VINT	|
3147 					       MIPS_CONF3_SP	|
3148 					       MIPS_CONF3_CDMM	|
3149 					       MIPS_CONF3_MT	|
3150 					       MIPS_CONF3_SM	|
3151 					       MIPS_CONF3_TL);
3152 	}
3153 	if (cpu_guest_has_conf4) {
3154 		kvm_set_sw_gc0_config3(cop0, MIPS_CONF_M);
3155 		/* Config4 */
3156 		kvm_save_gc0_config4(cop0);
3157 	}
3158 	if (cpu_guest_has_conf5) {
3159 		kvm_set_sw_gc0_config4(cop0, MIPS_CONF_M);
3160 		/* Config5 */
3161 		kvm_save_gc0_config5(cop0);
3162 		/* architecturally writable (e.g. from guest) */
3163 		kvm_clear_sw_gc0_config5(cop0, MIPS_CONF5_K	|
3164 					       MIPS_CONF5_CV	|
3165 					       MIPS_CONF5_MSAEN	|
3166 					       MIPS_CONF5_UFE	|
3167 					       MIPS_CONF5_FRE	|
3168 					       MIPS_CONF5_SBRI	|
3169 					       MIPS_CONF5_UFR);
3170 		/* architecturally read only, but maybe writable from root */
3171 		kvm_clear_sw_gc0_config5(cop0, MIPS_CONF5_MRP);
3172 	}
3173 
3174 	if (cpu_guest_has_contextconfig) {
3175 		/* ContextConfig */
3176 		kvm_write_sw_gc0_contextconfig(cop0, 0x007ffff0);
3177 #ifdef CONFIG_64BIT
3178 		/* XContextConfig */
3179 		/* bits SEGBITS-13+3:4 set */
3180 		kvm_write_sw_gc0_xcontextconfig(cop0,
3181 					((1ull << (cpu_vmbits - 13)) - 1) << 4);
3182 #endif
3183 	}
3184 
3185 	/* Implementation dependent, use the legacy layout */
3186 	if (cpu_guest_has_segments) {
3187 		/* SegCtl0, SegCtl1, SegCtl2 */
3188 		kvm_write_sw_gc0_segctl0(cop0, 0x00200010);
3189 		kvm_write_sw_gc0_segctl1(cop0, 0x00000002 |
3190 				(_page_cachable_default >> _CACHE_SHIFT) <<
3191 						(16 + MIPS_SEGCFG_C_SHIFT));
3192 		kvm_write_sw_gc0_segctl2(cop0, 0x00380438);
3193 	}
3194 
3195 	/* reset HTW registers */
3196 	if (cpu_guest_has_htw && (cpu_has_mips_r5 || cpu_has_mips_r6)) {
3197 		/* PWField */
3198 		kvm_write_sw_gc0_pwfield(cop0, 0x0c30c302);
3199 		/* PWSize */
3200 		kvm_write_sw_gc0_pwsize(cop0, 1 << MIPS_PWSIZE_PTW_SHIFT);
3201 	}
3202 
3203 	/* start with no pending virtual guest interrupts */
3204 	if (cpu_has_guestctl2)
3205 		cop0->reg[MIPS_CP0_GUESTCTL2][MIPS_CP0_GUESTCTL2_SEL] = 0;
3206 
3207 	/* Put PC at reset vector */
3208 	vcpu->arch.pc = CKSEG1ADDR(0x1fc00000);
3209 
3210 	return 0;
3211 }
3212 
3213 static void kvm_vz_prepare_flush_shadow(struct kvm *kvm)
3214 {
3215 	if (!cpu_has_guestid) {
3216 		/*
3217 		 * For each CPU there is a single GPA ASID used by all VCPUs in
3218 		 * the VM, so it doesn't make sense for the VCPUs to handle
3219 		 * invalidation of these ASIDs individually.
3220 		 *
3221 		 * Instead mark all CPUs as needing ASID invalidation in
3222 		 * asid_flush_mask, and kvm_flush_remote_tlbs(kvm) will
3223 		 * kick any running VCPUs so they check asid_flush_mask.
3224 		 */
3225 		cpumask_setall(&kvm->arch.asid_flush_mask);
3226 	}
3227 }
3228 
3229 static void kvm_vz_vcpu_reenter(struct kvm_vcpu *vcpu)
3230 {
3231 	int cpu = smp_processor_id();
3232 	int preserve_guest_tlb;
3233 
3234 	preserve_guest_tlb = kvm_vz_check_requests(vcpu, cpu);
3235 
3236 	if (preserve_guest_tlb)
3237 		kvm_vz_vcpu_save_wired(vcpu);
3238 
3239 	kvm_vz_vcpu_load_tlb(vcpu, cpu);
3240 
3241 	if (preserve_guest_tlb)
3242 		kvm_vz_vcpu_load_wired(vcpu);
3243 }
3244 
3245 static int kvm_vz_vcpu_run(struct kvm_vcpu *vcpu)
3246 {
3247 	int cpu = smp_processor_id();
3248 	int r;
3249 
3250 	kvm_vz_acquire_htimer(vcpu);
3251 	/* Check if we have any exceptions/interrupts pending */
3252 	kvm_mips_deliver_interrupts(vcpu, read_gc0_cause());
3253 
3254 	kvm_vz_check_requests(vcpu, cpu);
3255 	kvm_vz_vcpu_load_tlb(vcpu, cpu);
3256 	kvm_vz_vcpu_load_wired(vcpu);
3257 
3258 	r = vcpu->arch.vcpu_run(vcpu);
3259 
3260 	kvm_vz_vcpu_save_wired(vcpu);
3261 
3262 	return r;
3263 }
3264 
3265 static struct kvm_mips_callbacks kvm_vz_callbacks = {
3266 	.handle_cop_unusable = kvm_trap_vz_handle_cop_unusable,
3267 	.handle_tlb_mod = kvm_trap_vz_handle_tlb_st_miss,
3268 	.handle_tlb_ld_miss = kvm_trap_vz_handle_tlb_ld_miss,
3269 	.handle_tlb_st_miss = kvm_trap_vz_handle_tlb_st_miss,
3270 	.handle_addr_err_st = kvm_trap_vz_no_handler,
3271 	.handle_addr_err_ld = kvm_trap_vz_no_handler,
3272 	.handle_syscall = kvm_trap_vz_no_handler,
3273 	.handle_res_inst = kvm_trap_vz_no_handler,
3274 	.handle_break = kvm_trap_vz_no_handler,
3275 	.handle_msa_disabled = kvm_trap_vz_handle_msa_disabled,
3276 	.handle_guest_exit = kvm_trap_vz_handle_guest_exit,
3277 
3278 	.hardware_enable = kvm_vz_hardware_enable,
3279 	.hardware_disable = kvm_vz_hardware_disable,
3280 	.check_extension = kvm_vz_check_extension,
3281 	.vcpu_init = kvm_vz_vcpu_init,
3282 	.vcpu_uninit = kvm_vz_vcpu_uninit,
3283 	.vcpu_setup = kvm_vz_vcpu_setup,
3284 	.prepare_flush_shadow = kvm_vz_prepare_flush_shadow,
3285 	.gva_to_gpa = kvm_vz_gva_to_gpa_cb,
3286 	.queue_timer_int = kvm_vz_queue_timer_int_cb,
3287 	.dequeue_timer_int = kvm_vz_dequeue_timer_int_cb,
3288 	.queue_io_int = kvm_vz_queue_io_int_cb,
3289 	.dequeue_io_int = kvm_vz_dequeue_io_int_cb,
3290 	.irq_deliver = kvm_vz_irq_deliver_cb,
3291 	.irq_clear = kvm_vz_irq_clear_cb,
3292 	.num_regs = kvm_vz_num_regs,
3293 	.copy_reg_indices = kvm_vz_copy_reg_indices,
3294 	.get_one_reg = kvm_vz_get_one_reg,
3295 	.set_one_reg = kvm_vz_set_one_reg,
3296 	.vcpu_load = kvm_vz_vcpu_load,
3297 	.vcpu_put = kvm_vz_vcpu_put,
3298 	.vcpu_run = kvm_vz_vcpu_run,
3299 	.vcpu_reenter = kvm_vz_vcpu_reenter,
3300 };
3301 
3302 int kvm_mips_emulation_init(struct kvm_mips_callbacks **install_callbacks)
3303 {
3304 	if (!cpu_has_vz)
3305 		return -ENODEV;
3306 
3307 	/*
3308 	 * VZ requires at least 2 KScratch registers, so it should have been
3309 	 * possible to allocate pgd_reg.
3310 	 */
3311 	if (WARN(pgd_reg == -1,
3312 		 "pgd_reg not allocated even though cpu_has_vz\n"))
3313 		return -ENODEV;
3314 
3315 	pr_info("Starting KVM with MIPS VZ extensions\n");
3316 
3317 	*install_callbacks = &kvm_vz_callbacks;
3318 	return 0;
3319 }
3320